Fundamental of Transport Phenomena

Fundamentals
of Fluid Mechanics
and Transport Phenomena
Jean-Laurent Peube
This page intentionally left blank
Fundamentals of Fluid Mechanics and Transport Phenomena
Fundamentals
of Fluid Mechanics
and Transport Phenomena
Jean-Laurent Peube
First published in France in 2006 by Hermes Science/Lavoisier entitled Physique des écoulements et des
transferts in volumes 1 and 2 © LAVOISIER, 2006
First published in Great Britain and the United States in 2009 by ISTE Ltd and John Wiley & Sons, Inc.
Apart from any fair dealing for the purposes of research or private study, or criticism or review, as
permitted under the Copyright, Designs and Patents Act 1988, this publication may only be reproduced,
stored or transmitted, in any form or by any means, with the prior permission in writing of the publishers,
or in the case of reprographic reproduction in accordance with the terms and licenses issued by the CLA.
Enquiries concerning reproduction outside these terms should be sent to the publishers at the
undermentioned address:
ISTE Ltd John Wiley & Sons, Inc.

27-37 St George’s Road 111 River Street
London SW19 4EU Hoboken, NJ 07030
UK USA
www.iste.co.uk www.wiley.com
© ISTE Ltd, 2009
The rights of Jean-Laurent Peube to be identified as the author of this work have been asserted by him in
accordance with the Copyright, Designs and Patents Act 1988.
Library of Congress Cataloging-in-Publication Data
Peube, J. L.
[Physique des écoulements et des transferts. English]
Fundamentals of fluid mechanics and transport phenomena / Jean-Laurent Peube.
p. cm.
Includes bibliographical references and index.
ISBN 978-1-84821-065-3
1. Fluid mechanics. 2. Transport theory. I. Title.
TA357.P49813 2008
620.1'06--dc22
2008036553
British Library Cataloguing-in-Publication Data

A CIP record for this book is available from the British Library
ISBN: 978-1-84821-065-3
Printed and bound in Great Britain by CPI Antony Rowe Ltd, Chippenham, Wiltshire.
Table of Contents
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi
Chapter 1. Thermodynamics of Discrete Systems . . . . . . . . . . . . . . . . 1

1.1. The representational bases of a material system . . . . . . . . . . . . . . 1
1.1.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.2. Systems analysis and thermodynamics . . . . . . . . . . . . . . . . . 8
1.1.3. The notion of state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.1.4. Processes and systems. . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.2. Axioms of thermostatics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.2.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.2.2. Extensive quantities . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.2.3. Energy, work and heat . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.3. Consequences of the axioms of thermostatics . . . . . . . . . . . . . . . 21
1.3.1. Intensive variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.3.2. Thermodynamic potentials . . . . . . . . . . . . . . . . . . . . . . . . 23
1.4. Out-of-equilibrium states. . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.4.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
1.4.2. Discontinuous systems . . . . . . . . . . . . . . . . . . . . . . . . . . 30
1.4.3. Application to heat engines . . . . . . . . . . . . . . . . . . . . . . . 45
Chapter 2. Thermodynamics of Continuous Media . . . . . . . . . . . . . . . 47

2.1. Thermostatics of continuous media. . . . . . . . . . . . . . . . . . . . . . 47
2.1.1. Reduced extensive quantities . . . . . . . . . . . . . . . . . . . . . . 47
2.1.2. Local thermodynamic equilibrium . . . . . . . . . . . . . . . . . . . 48
2.1.3. Flux of extensive quantities . . . . . . . . . . . . . . . . . . . . . . . 50
2.1.4. Balance equations in continuous media . . . . . . . . . . . . . . . . 54
2.1.5. Phenomenological laws . . . . . . . . . . . . . . . . . . . . . . . . . . 57
2.2. Fluid statics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.2.1. General equations of fluid statics . . . . . . . . . . . . . . . . . . . . 63
vi Fundamentals of Fluid Mechanics and Transport Phenomena
2.2.2. Pressure forces on solid boundaries. . . . . . . . . . . . . . . . . . . 68

2.3. Heat conduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
2.3.1. The heat equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
2.3.2. Thermal boundary conditions . . . . . . . . . . . . . . . . . . . . . . 72
2.4. Diffusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
2.4.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
2.4.2. Molar and mass fluxes . . . . . . . . . . . . . . . . . . . . . . . . . . 77
2.4.3. Choice of reference frame . . . . . . . . . . . . . . . . . . . . . . . . 80
2.4.4. Binary isothermal mixture . . . . . . . . . . . . . . . . . . . . . . . . 85
2.4.5. Coupled phenomena with diffusion. . . . . . . . . . . . . . . . . . . 97
2.4.6. Boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
Chapter 3. Physics of Energetic Systems in Flow. . . . . . . . . . . . . . . . . 101

3.1. Dynamics of a material point . . . . . . . . . . . . . . . . . . . . . . . . . 101
3.1.1. Galilean reference frames in traditional mechanics . . . . . . . . . 101
3.1.2. Isolated mechanical system and momentum. . . . . . . . . . . . . . 102
3.1.3. Momentum and velocity . . . . . . . . . . . . . . . . . . . . . . . . . 103
3.1.4. Definition of force . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
3.1.5. The fundamental law of dynamics (closed systems) . . . . . . . . . 106
3.1.6. Kinetic energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
3.2. Mechanical material system . . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.2.1. Dynamic properties of a material system . . . . . . . . . . . . . . . 107
3.2.2. Kinetic energy of a material system . . . . . . . . . . . . . . . . . . 109
3.2.3. Mechanical system in thermodynamic equilibrium:
the rigid solid. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
3.2.4. The open mechanical system. . . . . . . . . . . . . . . . . . . . . . . 112
3.2.5. Thermodynamics of a system in motion . . . . . . . . . . . . . . . . 116
3.3. Kinematics of continuous media . . . . . . . . . . . . . . . . . . . . . . . 119
3.3.1. Lagrangian and Eulerian variables . . . . . . . . . . . . . . . . . . . 119
3.3.2. Trajectories, streamlines, streaklines . . . . . . . . . . . . . . . . . . 121
3.3.3. Material (or Lagrangian) derivative. . . . . . . . . . . . . . . . . . . 122
3.3.4. Deformation rate tensors . . . . . . . . . . . . . . . . . . . . . . . . . 129
3.4. Phenomenological laws of viscosity . . . . . . . . . . . . . . . . . . . . . 132
3.4.1. Definition of a fluid . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
3.4.2. Viscometric flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
3.4.3. The Newtonian fluid. . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
Chapter 4. Fluid Dynamics Equations . . . . . . . . . . . . . . . . . . . . . . . 151

4.1. Local balance equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
4.1.1. Balance of an extensive quantity G . . . . . . . . . . . . . . . . . . . 151
4.1.2. Interpretation of an equation in terms of the balance equation . . . 153
4.2. Mass balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
4.2.1. Conservation of mass and its consequences . . . . . . . . . . . . . . 154
4.2.2. Volume conservation . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
Table of Contents vii
4.3. Balance of mechanical and thermodynamic quantities . . . . . . . . . . 160

4.3.1. Momentum balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
4.3.2. Kinetic energy theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 164
4.3.3. The vorticity equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
4.3.4. The energy equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
4.3.5. Balance of chemical species . . . . . . . . . . . . . . . . . . . . . . . 177
4.4. Boundary conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
4.4.1. General considerations . . . . . . . . . . . . . . . . . . . . . . . . . . 178
4.4.2. Geometric boundary conditions . . . . . . . . . . . . . . . . . . . . . 179
4.4.3. Initial conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
4.5. Global form of the balance equations . . . . . . . . . . . . . . . . . . . . 182
4.5.1. The interest of the global form of a balance . . . . . . . . . . . . . . 182
4.5.2. Equation of mass conservation . . . . . . . . . . . . . . . . . . . . . 184
4.5.3. Volume balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
4.5.4. The momentum flux theorem . . . . . . . . . . . . . . . . . . . . . . 184
4.5.5. Kinetic energy theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 186
4.5.6. The energy equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
4.5.7. The balance equation for chemical species . . . . . . . . . . . . . . 188
4.6. Similarity and non-dimensional parameters . . . . . . . . . . . . . . . . . 189
4.6.1. Principles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
Chapter 5. Transport and Propagation. . . . . . . . . . . . . . . . . . . . . . . 199

5.1. General considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
5.1.1. Differential equations . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
5.1.2. The Cauchy problem for differential equations . . . . . . . . . . . . 202
5.2. First order quasi-linear partial differential equations . . . . . . . . . . . 203
5.2.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
5.2.2. Geometric interpretation of the solutions . . . . . . . . . . . . . . . 204
5.2.3. Comments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
5.2.4. The Cauchy problem for partial differential equations . . . . . . . 206
5.3. Systems of first order partial differential equations . . . . . . . . . . . . 207
5.3.1. The Cauchy problem for n unknowns and two variables . . . . . . 207
5.3.2. Applications in fluid mechanics . . . . . . . . . . . . . . . . . . . . . 210
5.3.3. Cauchy problem with n unknowns and p variables. . . . . . . . . . 216
5.3.4. Partial differential equations of order n . . . . . . . . . . . . . . . . 218
5.3.5. Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
5.3.6. Physical interpretation of propagation . . . . . . . . . . . . . . . . . 223
5.4. Second order partial differential equations . . . . . . . . . . . . . . . . . 225
5.4.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
5.4.2. Characteristic curves of hyperbolic equations. . . . . . . . . . . . . 226
5.4.3. Reduced form of the second order quasi-linear partial
differential equation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
5.4.4. Second order partial differential equations in a finite domain . . . 232
viii Fundamentals of Fluid Mechanics and Transport Phenomena
5.4.5. Second order partial differential equations and their boundary

conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
5.5. Discontinuities: shock waves . . . . . . . . . . . . . . . . . . . . . . . . . 239
5.5.1. General considerations . . . . . . . . . . . . . . . . . . . . . . . . . . 239
5.5.2. Unsteady 1D flow of an inviscid compressible fluid. . . . . . . . . 239
5.5.3. Plane steady supersonic flow . . . . . . . . . . . . . . . . . . . . . . 244
5.5.4. Flow in a nozzle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
5.5.5. Separated shock wave . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
5.5.6. Other discontinuity categories . . . . . . . . . . . . . . . . . . . . . . 248
5.5.7. Balance equations across a discontinuity . . . . . . . . . . . . . . . 249
5.6. Some comments on methods of numerical solution . . . . . . . . . . . . 250
5.6.1. Characteristic curves and numerical discretization schemes . . . . 250
5.6.2. A complex example . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
5.6.3. Boundary conditions of flow problems. . . . . . . . . . . . . . . . . 255
Chapter 6. General Properties of Flows . . . . . . . . . . . . . . . . . . . . . . 257

6.1. Dynamics of vorticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
6.1.1. Kinematic properties of the rotation vector . . . . . . . . . . . . . . 257
6.1.2. Equation and properties of the rotation vector . . . . . . . . . . . . 261
6.2. Potential flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
6.2.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
6.2.2. Bernoulli’s second theorem . . . . . . . . . . . . . . . . . . . . . . . 269
6.2.3. Flow of compressible inviscid fluid. . . . . . . . . . . . . . . . . . . 270
6.2.4. Nature of equations in inviscid flows. . . . . . . . . . . . . . . . . . 271
6.2.5. Elementary solutions in irrotational flows . . . . . . . . . . . . . . . 273
6.2.6. Surface waves in shallow water . . . . . . . . . . . . . . . . . . . . . 284
6.3. Orders of magnitude . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
6.3.1. Introduction and discussion of a simple example. . . . . . . . . . . 288
6.3.2. Obtaining approximate values of a solution . . . . . . . . . . . . . . 291
6.4. Small parameters and perturbation phenomena. . . . . . . . . . . . . . . 296
6.4.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
6.4.2. Regular perturbation. . . . . . . . . . . . . . . . . . . . . . . . . . . . 296
6.4.3. Singular perturbations. . . . . . . . . . . . . . . . . . . . . . . . . . . 305
6.5. Quasi-1D flows . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
6.5.1. General properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
6.5.2. Flows in pipes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
6.5.3. The boundary layer in steady flow . . . . . . . . . . . . . . . . . . . 319
6.6. Unsteady flows and steady flows . . . . . . . . . . . . . . . . . . . . . . . 327
6.6.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
6.6.2. The existence of steady flows . . . . . . . . . . . . . . . . . . . . . . 328
6.6.3. Transitional regime and permanent solution. . . . . . . . . . . . . . 330
6.6.4. Non-existence of a steady solution . . . . . . . . . . . . . . . . . . . 334
Table of Contents ix
Chapter 7. Measurement, Representation and Analysis of

Temporal Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
7.1. Introduction and position of the problem . . . . . . . . . . . . . . . . . . 339
7.2. Measurement and experimental data in flows. . . . . . . . . . . . . . . . 340
7.2.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340
7.2.2. Measurement of pressure . . . . . . . . . . . . . . . . . . . . . . . . . 341
7.2.3. Anemometric measurements . . . . . . . . . . . . . . . . . . . . . . . 342
7.2.4. Temperature measurements . . . . . . . . . . . . . . . . . . . . . . . 346
7.2.5. Measurements of concentration . . . . . . . . . . . . . . . . . . . . . 347
7.2.6. Fields of quantities and global measurements. . . . . . . . . . . . . 347
7.2.7. Errors and uncertainties of measurements . . . . . . . . . . . . . . . 351
7.3. Representation of signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357
7.3.1. Objectives of continuous signal representation . . . . . . . . . . . . 357
7.3.2. Analytical representation . . . . . . . . . . . . . . . . . . . . . . . . . 360
7.3.3. Signal decomposition on the basis of functions; series and
elementary solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 361
7.3.4. Integral transforms. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363
7.3.5. Time-frequency (or timescale) representations . . . . . . . . . . . . 374
7.3.6. Discretized signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 381
7.3.7. Data compression . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385
7.4. Choice of representation and obtaining pertinent information . . . . . . 389
7.4.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
7.4.2. An example: analysis of sound . . . . . . . . . . . . . . . . . . . . . 390
7.4.3. Analysis of musical signals . . . . . . . . . . . . . . . . . . . . . . . 393
7.4.4. Signal analysis in aero-energetics . . . . . . . . . . . . . . . . . . . . 402
Chapter 8. Thermal Systems and Models . . . . . . . . . . . . . . . . . . . . . 405

8.1. Overview of models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 405
8.1.1. Introduction and definitions . . . . . . . . . . . . . . . . . . . . . . . 405
8.1.2. Modeling by state representation and choice of variables . . . . . . 408
8.1.3. External representation . . . . . . . . . . . . . . . . . . . . . . . . . . 410
8.1.4. Command models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411
8.2. Thermodynamics and state representation. . . . . . . . . . . . . . . . . . 412
8.2.1. General principles of modeling . . . . . . . . . . . . . . . . . . . . . 412
8.2.2. Linear time-invariant system (LTIS) . . . . . . . . . . . . . . . . . . 420
8.3. Modeling linear invariant thermal systems . . . . . . . . . . . . . . . . . 422
8.3.1. Modeling discrete systems . . . . . . . . . . . . . . . . . . . . . . . . 422
8.3.2. Thermal models in continuous media. . . . . . . . . . . . . . . . . . 431
8.4. External representation of linear invariant systems . . . . . . . . . . . . 446
8.4.1. Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 446
8.4.2. External description of linear invariant systems . . . . . . . . . . . 446
8.5. Parametric models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 451
8.5.1. Definition of model parameters . . . . . . . . . . . . . . . . . . . . . 451
8.5.2. Established regimes of linear invariant systems . . . . . . . . . . . 453
x Fundamentals of Fluid Mechanics and Transport Phenomena
8.5.3. Established regimes in continuous media . . . . . . . . . . . . . . . 458

8.6. Model reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465
8.6.1. Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 465
8.6.2. Model reduction of discrete systems . . . . . . . . . . . . . . . . . . 466
8.7. Application in fluid mechanics and transfer in flows . . . . . . . . . . . 474
Appendix 1. Laplace Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . 477

A1.1. Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 477
A1.2. Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 477
A1.3. Some Laplace transforms . . . . . . . . . . . . . . . . . . . . . . . . . . 478
A1.4. Application to the solution of constant coefficient
differential equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 479
Appendix 2. Hilbert Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . 481
Appendix 3. Cepstral Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 483

A3.1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 483
A3.2. Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 483
A3.3. Example of echo suppression . . . . . . . . . . . . . . . . . . . . . . . . 484
A3.4. General case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 485
Appendix 4. Eigenfunctions of an Operator . . . . . . . . . . . . . . . . . . . . 487

A4.1. Eigenfunctions of an operator . . . . . . . . . . . . . . . . . . . . . . . . 487
A4.2. Self-adjoint operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 487
A4.2.1. Eigenfunctions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 487
A4.2.2. Expression of a function of f using an eigenfunction basis-set. . 488
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 489
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 497
Preface
The study of fluid mechanics and transfer phenomena in flows involves the
association of difficulties which are encountered in different disciplines:
thermodynamics, mechanics, thermal conduction, diffusion, chemical reactions, etc.
This book is not intended to be an encyclopaedia, and we will thus not endeavour to
cover all of the aforementioned disciplines in a detailed fashion. The main objective
of the text is to present the study of the movement of fluids and the main
consequences in terms of the transfer of mass and heat. The book is the result of
many years of teaching and research, both theoretical and applied, in scientific
domains which are often considered separately. In effect, the development of new
disciplines which are at the same time specialized and universal was very much a
characteristic of science in the 20th century. Thus, signal processing, system
analysis, numerical analysis, etc. are all autonomous disciplines and indispensable
means for students, engineers or researchers working in the domain of fluid
mechanics and energetics. In the same way, various domains such as the design of
chemical reactors, the study of the stars and meteorology require a solid knowledge
of fluid mechanics in addition to that of their specific topics.
This book is primarily aimed at students, engineers and researchers in fluid

mechanics and energetics. However, we feel that it can be useful for people working
in other disciplines, even if the reading of some of the more theoretical and
specialized chapters may be dispensable in this case. The science and technology of
the first half of the 20th century was heavily rooted in classical mechanics, with
concepts and methods which relied on algebra and differential and integral calculus,
these terms being taken into account in the sense they were used at that time.
Furthermore, scientific thought was fundamentally deterministic during this period,
even if the existence of games of chance using mechanical devices (dice, roulette,
etc.) seemed far from the philosophy of science or Cauchy’s theorem. Each time has
xii Fundamentals of Fluid Mechanics and Transport Phenomena
its concepts, which are based on the current state of knowledge, and the science of
fluid mechanics was reduced for the most part to semi-empirical engineering
formulae and to particular analytical solutions. Between the 1920s and the 1950s,
our ideas on boundary layers and hydrodynamic stability were progressively
elucidated. Studies of turbulence, which began in the 1920s from a conceptual
statistical point of view, have really only made further progress in the 1970s, with
the writing of the balance equations using turbulence models with a physical basis.
This progress remains quite modest, however, considering the immensity of the task
which remains.
It should be noted that certain disciplines have seen a spectacular renewal since
the 1970s for two main reasons: on the one hand, the development of information
technology has provided formidable computation and experimental methods, and on
the other hand, multidisciplinary problems have arisen from industrial necessities.
Acoustics is a typical example: many problems of propagation had been solved in
the 1950s-1960s and those which were not made only very slow progress. Physics
focused on other fundamental, more promising sectors (semiconductors, properties
of matter, etc.). However, in the face of a need to provide practical solutions to
industrial problems (sound generated by fluid flow, the development of ultra-sound
equipment, etc.), acoustics became an engineering science in the 1970s. Acoustics is
indeed a domain of compressible fluid mechanics and it will constitute an integral
part of our treatment of the subject.
Parallel to this, systems became an object of study in themselves (automatic

control) and the possibilities of study and understanding of the complexity
progressed (signal processing, modeling of systems with large numbers of variables,
etc.). Determinism itself is now seen in a more modest light: it suffices to remember
the variable level of our ambitions with regard to meteorological prediction in the
last 30 years to see that we have not yet arrived at a point where we have a definite
set of concepts. Meteorological phenomena are largely governed by fluid
mechanics.
The conception of this book results from the preceding observations. The author
refuses to get into the argument which consists of saying that the time of analytical
solutions has passed and that numerical simulation will solve all our problems. The
reality is clearly more subtle than this: analytical solution in the broad sense, that is,
the obtaining of results derived from reasoning and mathematical concepts, is the
basis of physical concepts. Computations performed by computers by themselves
cannot provide any more insight than an experiment, although both must be
performed with great care. The state of knowledge and of understanding of
mechanisms varies depending on the domain studied. In particular, the science of
turbulence is still at a somewhat embryonic stage, and the mystery of turbulent
solutions of the Navier-Stokes equations is far from being thoroughly cleared up.
Preface xiii
We are still at the stage of Galileo who attempted to understand mechanics without
the ideas of differential calculus. Nobody can today say precisely what are the
difficulties to be solved, and the time which will be required for their resolution (10
years, a century or 10 centuries). We will therefore present the state of our
knowledge in the current scientific context by also considering some of the
accompanying disciplines (thermodynamics, ideas related to partial differential
equations, signal processing, system analysis) which are directly useful to the
concepts, modeling, experiments and applications in fluid mechanics and energetics
of flows. We will not cover specific combustion phenomena, limiting ourselves to a
few simplified cases of physico-chemical reactions.
This book covers the necessary fundamentals for the study and understanding of
the specific concepts and general properties of flows: the establishment and
discussion of the balance equations of extensive quantities in fluid motions, the
transport of these quantities by convection, wave-propagation or diffusion. These
physical concepts are issued from the comprehension of theoretical notions
associated with equations, such as characteristic curves or surfaces, perturbation
methods, modal developments (Fourier series, etc.) and integral transforms, model
reduction, etc. These mathematical aspects are either consequences of properties of
partial differential equations or derived from other disciplines such as signal
processing and system analysis, whose impact is important in every scientific or
technological domain. They are discussed and illustrated by some elementary
problems of fluid mechanics and thermal conduction, including measurement
methods and experimental data processing This book is an introduction to the study
of more specialized topics of fluid flow and transfer phenomena encountered in
different domains of application: incompressible or compressible flow, dynamic and
thermal boundary layers, natural or mixed convection, 3D boundary layers, physico-
chemical reactions in flows, acoustics in flows, aerodynamic sound,
thermoacoustics, etc.
Chapter 1 is devoted to a synthetic presentation of thermodynamics. After

recalling the basics of the representation of material systems, thermostatics is
covered in an axiomatic fashion which avoids the use of differential formulations
and which allows for a simplified presentation of classical results. Taking entropy
dynamics as a starting point, the thermodynamics of non-equilibrium states is then
discussed using simple examples with phenomenological laws of linear
thermodynamics.
The continuous medium at rest is obtained by taking the limit of discrete systems
in Chapter 2. The exchange of extensive quantities is modeled by means of flux
densities, and irreversible thermodynamics leads to the diffusion equations. Some
reminders of fluid statics are given. We then discuss the difficulties specific to the
diffusion of matter.
xiv Fundamentals of Fluid Mechanics and Transport Phenomena
The association of mechanical phenomena with thermodynamics is briefly

developed in Chapter 3 along with the formalism used for the description of the
motion of continuous media. The elementary properties of viscosity are then
discussed.
Chapter 4 is dedicated to the writing of the general equations of the dynamics of

fluid and transfer. The integration of local equations in a domain enables the
separation of sources and fluxes of extensive quantities, these fluxes being transfer
phenomena involving definition of input-output mechanisms for that domain,
considered as a system. The energy equation explicitly expresses the interactions
between thermodynamics and the movement of matter. The main usual boundary
conditions and similarity and its consequences are then discussed.
Chapter 5 discusses the classification of partial differential equations in fluid

mechanics. The mathematical aspects at the basis of physical concepts are well
understood, but unfortunately rarely taught. These are very important, both for the
numerical solution of equations and for the understanding of physical phenomena.
We will present them here without providing any thorough demonstrations. The
reader who struggles with this chapter should nonetheless try to assimilate its
content while leaving aside the details of certain calculations.
Chapter 6 is dedicated in the main to the influence of diffusion in the convection

of linear or angular momentum. It firstly covers vortex dynamics, the transposition
to continuous media of concepts used in solid body rotation. Vorticity often results
from transitional processes which may be more or less viscous, but its transport is
very often governed by the equations for an inviscid fluid. Lagrange’s theorem
introduces the idea of conservation of circulation of velocity which allows the
rotation to be treated as a frozen material field. Elementary solutions of the 2D
incompressible potential flows are quickly discussed. We then look at the quasi-1D
approximation, which is particularly important in fluid mechanics, either for pipes
or for flows in the vicinity of walls when a non-dimensional quantity becomes large.
This last circumstance corresponds to a singular perturbation problem in the form of
a boundary layer, which corresponds to the effects of viscous diffusion from the
walls. The discussion of the boundary-layer equations reveals the separation
mechanisms which are associated with the non-linear terms in steady flow
equations.
The measurement of flow and transfer phenomena presents difficulties which are
outlined in Chapter 7. The recent evolution of techniques based on the digitization
of measurements, signal processing, analysis and reduction of models are naturally
suited to applications in fluid mechanics and energetics. These methods have led to
a renewal of progress in disciplines where unsteady phenomena are encountered,
and in particular in the study of acoustic phenomena and turbulent flows.
Preface xv
Improvements in computing have of course also led to considerable progress in the

modeling of phenomena. The use of these methods requires specialized techniques
whose treatment is beyond the scope of this book. The elements of signal processing
and system analysis which we provide are only intended to alert the reader to the
possibilities and utility of these methods, but also to show their limits. The idea that
computers will allow the resolution of all our problems remains too ubiquitous.
Computers only provide a tool to help us find the solutions we seek. These recent
methods, signal processing or system analysis, are also useful for the identification
of physical concepts associated with phenomena and the representation of solutions.
In Chapter 7, we also indicate in a synthetic manner the essential ideas necessary

for measurement and signal processing procedures which are most useful in the
domains studied. The possibility of large computations in modeling and
experimental data processing leads us to evoke the idea of conditioning of linear
systems, which is a generalization of elementary calculations of errors and
uncertainties.
Chapter 8 is dedicated to modeling which provides a general context for the

study of the evolution of physical systems. However, automatic control is reasoning
in a general way on models without taking account of the laws of thermodynamics.
These are essential for the disciplines studied in this book. We will present a few
points of view and methods developed in automatic control, directly applied to the
balance equations of basic problems of thermal conduction. The approximation
procedures for the balance equations are far from being equivalent depending on the
way in which we proceed. In order to simplify the presentation and to clearly
separate the difficulties, we will mainly limit ourselves here to the state
representation which is derived from thermodynamic modeling, leaving aside
models derived from the approximation of solutions which do not exactly satisfy the
balance equations.
NOTE.
We have chosen to respect the usual notation of physical quantities in each
discussed scientific domain, while trying to have consistent notations whenever
possible.
At the same time, the notations for derivatives are different, depending on the
domain covered (thermodynamics, mechanics or more mathematical developments)
and the size of equations. They all are usual and well known:
For functions y (x) of one variable, they are marked y' (x), y'' (x), y''' (x), y''''
(x),..., y(n)(x).
When discussing mechanical questions, the two first temporal derivatives of
x(t) are written with dots: x (t ) and x(t ) .
xvi Fundamentals of Fluid Mechanics and Transport Phenomena
d
The symbol is used only for material (Lagrangian) derivatives, which are
dt
indeed derivatives with respect to time of compound functions in Euler variables;
D
this is equivalent to the other usual notation .
Dt
For functions f (x, y) of several variables, the two following notations are used
wf wf w 2 f w 3 f
according circumstances: either with symbol w ( , , , ,...) or with
wx wy wxwy wx 2 wy
indices marking the variables with respect of which derivations are performed: fx ,
fy, fxy , fxxy.
Integrals are always indicated by a simple integration sign, as the nature of this
(single, double, triple, etc.) should be clear from the integration domain indicated
and the differential element.
When tensor notation is used, vectors or matrices are denoted using upper case
letters, their components being written in lower case letters. The convention of
summation over repeated indices (Einstein’s convention) will systematically be
used.
Chapter 1
Thermodynamics of Discrete Systems
The general objective of thermodynamics is to describe the properties of matter.

After recalling the representational bases of material systems, thermostatics is dealt
with by postulating the existence of a general equation of state which relates the
extensive quantities. In this way we can forgo the need to delve into principles
related to differential forms, and thereby simplify the presentation of traditional
results. Then the thermodynamics of out of equilibrium systems are considered in
terms of entropy dynamics, and discussed using simple examples. Finally, the
phenomenological laws of linear thermodynamics are then considered.
1.1. The representational bases of a material system
1.1.1. Introduction
1.1.1.1. Geometric Euclidean space and physical quantities

The object of the physical sciences is the study of matter, for which the
formulation of physical laws is necessary. However prior to the formulation of any
such laws it is clearly necessary to characterize matter in terms of the various
physical quantities which we can directly or indirectly measure. Matter is present all
around us, and in a first instance we will limit ourselves to considering it in a static
way, at a given instant which we can identify (this supposes a minimal definition of
time); we perform geometric measurements in a 3D Cartesian coordinates system in
order to identify the position and/or dimension of material elements. Measuring
length presents no particular difficulty, excepting the choice of units. We will
observe material elements in a geometric Euclidean space.
2 Fundamentals of Fluid Mechanics and Transport Phenomena
The geometric description of space is independent of the presence of matter; in

other words the metric tensor does not depend on any physical quantity. This is not
true for certain astrophysical phenomena which require us to place ourselves in the
context of general relativity where geometric properties of space are no longer
independent of the presence of matter. Simplistically put, the length of a meter
depends on the mass found in its vicinity, which considerably complicates matters.
In the following we exclude such phenomena, as they only become important at
scales which greatly exceed those of our terrestrial physics.
We thus postulate (Axiom 1) the existence of a geometric space whose structure

is independent of the properties of matter and the associated physical phenomena
(gravitation, force fields, etc.).
We also admit (Axiom 2) that this space is homogenous and isotropic, which
leads us to a traditional geometric Euclidean description of space R3 with its
associated notions of length, surface and volume, whose scalar values are
independent of the particular geometric frame of reference we choose to consider.
This property of homogenity and isotropy will have important consequences for the
expression of physical laws, which must not favor any given point or physical
spatial direction. In particular, physical laws should neither favor any particular
point in the universe, nor change as a result of a change in reference frame.
Finally, we suppose (Axiom 3) that matter can be characterized by physical

quantities which are measurable at each instant in time, and not by mathematical
entities (wavefunctions etc.) which allow, via mathematical operations, access to
information of a probabilistic kind with regard to a physical quantity. This
hypothesis of the possibility of directly measuring physical quantities supposes that
the measure does not change the physical quantities of the material element
considered. We therefore exclude microscopic phenomena relevant to quantum
mechanics from our field of study, and we suppose the smallest material elements
studied to contain a number of atoms or molecules sufficient for the neglect of
statistical microscopic fluctuations to be justified.
1.1.1.2. The existence of isolated systems and the definition of time

The study of physical phenomena presupposes their reproducibility; the same
effects should be observed under identical conditions. The establishment of physical
laws thus supposes the definition of a time with the property of homogenity: in
particular, quantifiable and reproducible observations of the evolution of a given
material system must be possible.
The definition of time should thus be appropriately chosen. Previously associated

with the length of the day, the definition of time has varied considerably between
different individuals and epochs. For example, during the Roman period the lengths
Thermodynamics of Discrete Systems 3
of the day and the night were respectively divided into seven and four parts, the
Babylonians 2,000 years beforehand divided the day and the night each into 12
hours, which were clearly of unequal duration and varied according to the seasons.
The Chinese and the Japanese divided each of the two cycles, from dawn to dusk
and from dusk to dawn, into six periods. Japan only adopted the occidental system in
1873, but this did not prevent Japanese clockmakers from making mechanical clocks
as early as the 17th century, these having quite complex mechanisms in order to
accommodate the variable length of their hour.
The definition and measurement of time are thus not automatic operations for
human beings. The relatively old notion of regular time (homogenous in the physical
sense) is related to the use of indefinitely reproducible phenomena; this notion dates
from the end of antiquity, the early Middle Ages and the invention of the clock
(clepsydras, mechanical clocks, hourglass).
We will thus postulate (Axiom 4) that physical phenomena are reproducible,

regardless of when an experiment is performed. Any evolutionary phenomenon
which is considered reproducible will allow a time unit to be defined. A temporal
dimension can be constructed simply by virtue of the reproducibility of a
phenomenon, which amounts to admitting that time is homogenous, i.e. no instant in
the universe is given any special privilege. This homogenity of time does not really
exist in cosmological problems, and in particular during the time of the initial big
bang. We exclude these kinds of problem.
Having long been attached to the average duration of a solar day, the definition
of time is now effected using the vibration frequency of an atom of caesium 123
under the most stable conditions possible (at very low temperature).
1.1.1.3. Causality and irreversibility

We now dispose of a space-time coordinates system comprising three space
dimensions and one time dimension. However, in contrast with geometric space,
time is not isotropic. In effect, the definition of entropy (section 1.2.2.4) shows that
an irreversible evolution exists in the universe with which we can associate a time
variable (or one related to the age of the universe) in an attempt to characterize it.
This irreversibility is explained by statistical mechanics whereby matter always
tends to states in which it is maximally mixed: gas molecules in a volume will
always be evenly dispersed over the volume. This is the most probable state in
which the molecules will be found; while the probability of finding all of the
molecules confined to the left half of the volume is not strictly zero, this situation is
never observed.
The age of the universe is thus associated with a measure of its entropy on a very
large scale (the universe or at least the earth). However, a time characterized by this
scale has no guarantee of being homogenous. This “age” of the universe does not
give us a useful indication of what time to use, and we will content ourselves with
the time previously defined from the notion of reproducibility. The notion of entropy
(or of the ageing of the universe) shows that time has a considerable anisotropy,
manifest in the distinction between the past, the present and the future. The
equations translating the physical laws and their consequences should not violate
this anisotropy, the effect of which can be immediately seen if we change the
direction of time by letting t ' t .
Let us consider an isolated mechanical oscillator with friction, which can be

described by the equation:
mx(t ) fx (t ) kx(t ) 0 [1.1]
whose oscillatory solution takes the form A exp f t / 2m cosYt M .
By multiplying equation [1.1] by x (t ) and integrating with respect to time

between 0 and T, the total variation of mechanical energy 'Em between these
instants is:
T
ª1 2 1 2º T
'E m « mx kx » ³0 f x 2 dt
¬2 2 ¼0
The absolute value of this variation 'Em is always negative and increasing for a
T
positive value of the friction coefficient. The quantity ³0 f x 2 dt is known as the
dissipation function of the system.
Changing the direction of time would be equivalent to changing the term f x (t )

to f x (t ) , which implies a negative friction coefficient f leading to the solution
A exp f t ' / 2m cosYt M and to an increase in mechanical energy as a time
function. This is impossible with an isolated oscillator and could only be made
possible by the intervention of an exterior energy source. The preceding equation is
clearly unstable in the sense that its solutions diverge analogously to the instabilities
encountered in the local study of equilibrium.
Let us take as an example three equations representative of constant coefficient,

second order partial differential equations (see Chapter 5):
Laplace equation: w f2 w 2f
2 2
0;
wx wt
wave equation: w 2f w 2f
2 2
0;
wx wt
w f wf
2
heat equation: 0.
wx 2 wt
The general solution of Laplace’s equation (which is elliptic) at a point requires

that conditions be known at all points lying on a curve surrounding this point
(Dirichlet condition). All points at the frontier of the domain exert an influence on
the solution at a point (x,t). The result is that no physical phenomenon can be
represented by Laplace's equation if time is chosen as a variable, since the solution
in t would depend on smaller (earlier) and larger (later) values of the time variable.
The wave equation (which is hyperbolic) on the contrary is compatible with the
definition of time. Its general solution:
f x, t I (x t) \ (x t)
represents two waves which propagate along the x-axis with velocities +1 and –1.
The value at a point x and instant t depends on what happens to each of the said
waves to the left and the right of x, and before their arrival at time t. The wave
equation is thus compatible with the non-influence of the future on the present.
The heat equation (which is parabolic) is also compatible with the non-influence
of the future on the present, as we will see for heat conduction problems, since the
initial conditions (or values from the past) suffice for a determination of the solution
at any later time.
Another remark can be made here regarding the inversion of the direction of
time. By replacing t with t', we see that the wave equation remains unchanged,
while the heat equation becomes:
w2 f wf
0
2 wt '
wx
We will see similar behavior for the complete solution of the heat equation in a
wall (Chapter 8) in which the inversion of the direction of time results in a change of
a sum of temporally decaying exponential terms to a sum of temporally increasing
exponential terms. Changing the direction of time in the heat equation leads to a
physically inadmissible equation.
The preceding phenomena can be interpreted in a number of ways:

1) In terms of energy dissipation and of the creation of entropy

The wave equation represents a frictionless mechanical phenomenon, there is no
creation of entropy over time; we have a reversible phenomenon and so an inversion
of the direction of time is not incompatible with the laws of the universe. We should
note however that the wave equation is only valid for relatively short times, for
which the inevitable friction is not to have an influence. Acoustic waves are finally
damped by diverse frictional forces after they have covered a very large distance;
light waves are finally absorbed by matter in an irreversible process (the Joule
effect) etc. Energy transfer creates entropy and is therefore compatible with the
evolution of the universe.
2) In terms of information loss
The wave equation was earlier interpreted as a transmission of a signal by pure
propagation. There is no loss of information during the transmission. The
introduction of dissipation (creation of entropy) leads to the telegrapher's equation,
which is no longer invariant under a change in the direction of time, and thus
involves an attenuation of the signals during transmission, and then a subsequent
loss of information.
The heat equation translates a smoothing of temperature distributions, which

may initially be complex, to a more uniform field. The final state is often a constant
temperature which has no memory of its initial distribution. We note again that an
inversion of the time direction in heat diffusion problems does not allow for a
retrieval of the information which has been lost. The same goes for an oscillator
with friction, whose final state of rest precludes any knowledge of the initial
conditions.
The notions of past and future, with respect to an event, introduce a fundamental
asymmetry; the present does not depend on the future. This has certain
consequences, both in the application of certain mathematical transformations
(Fourier for example) on temporal signals, and in flow problems where the
distinction between upstream and downstream is of the same nature as that between
the past and the future.
1.1.1.4. Causality and determinism

The question of cause and effect is a very old philosophical problem (Aristotle,
the scholastic philosophers of the middle-ages, Descartes, Leibniz, Spinoza, Hume,
Kant, Schopenhauer, Bernard, etc.). We will not go into the complex philosophical
distinctions related to causes (adequate, inadequate, efficient, final, formal, material,
primary, secondary etc.). An effect is the result of and is produced by an efficient
cause.
Kant upholds that the causality relation is “absolutely general and even
necessary”. The general principle of causality is even more clear in determinism,
which holds that all events can be rationally predicted, with a desired degree of
precision, provided that past events and all of the laws of nature are known with
sufficient precision. Such absolute and universal determinism is associated with a
conception of a universe dominated by laws of celestial mechanics (Laplace). In
other words, the same causes produce the same effects, and so our capacity to
predict depends only on our scientific knowledge. Of course, quantum mechanics
has brought this vision of things into question, but not on the scale of the phenomena
studied here.
However, the question of determinism is not as simple as it might seem, in

particular in situations where unstable phenomena intervene, or where chance plays
a central role (chaos). Examples of such situations are usual in mechanical devices
used for games of chance (dice, roulette, etc.) or in fluid mechanics whose equations
have unstable solutions going through unpredictable evolutions in which flows are
fluctuating in a chaotic way. This is the phenomenon of turbulence encountered in
most practical flows; for example, atmospheric flows are results of such instabilities
and then weather prediction is fundamentally impossible beyond a few days.
Nevertheless, a statistical treatment of these turbulent flows leads to a more global
kind of determinism ([LES 98]).
It is useful to note at this point that the conditions for prediction can be defined
mathematically via theorems which treat of the existence of unique solutions for
differential equations given a suitable set of initial conditions. The Cauchy-Lipschitz
theorem is the best known, and deals with differential equations with real variables
(x,y) of the form:
dy dx f ( x, y )
The function f(x,y) is only required to verify a Lipschitz condition1. This theorem
establishes the existence of a unique solution y = M(x) which verifies the initial
condition y0 = M(x0) . This solution is continuous over the interval ( x 0 , x 0 h) ,
where h is characterized by the interval of definition for x and an upper bound of f
in the rectangle considered. This theorem can be extended to systems of differential
equations with the same kinds of conditions.
A similar theorem (Cauchy-Kovalevskaïa), but with stricter analycity conditions

of the function f(x,y) in the neighborhood of the point ( x 0 , y 0 ) (functions which can
1 I.e.: f ( x, y ) f ( x, y ' ) A y y ' , condition in which (x,y) and (x,y') are arbitrarily
chosen in a rectangle where f(x,y) is supposed to be continuous, A being a positive constant.
be developed in power series), leads to a unique analytic solution y M (x ) in the

neighborhood of the point ( x 0 , y 0 ) with the initial condition y 0 M ( x 0 ) . These
results can be extended to systems of differential equations, linear partial differential
equations, etc.
Cauchy’s theorem thus translates a form of determinism, since given a cause (the
initial condition y 0 M ( x 0 ) ), a unique solution y M (x) exits. However, we see
that there are certain limitations, in particular with the Cauchy-Kovalevskaïa
theorem which imposes analyticity conditions, the physical realization of which has
no reason to be assured for the function f(x,y) or any other perturbation which we
may add in order to test the stability of the system
In all causal situations, the preceding Cauchy theorems lead to results of a local
nature, that is to say over a short period of time, considering the variable x to
represent time. In the middle to long term, numerous “mathematical accidents” may
occur. The uniqueness of a local solution is not in contradiction with the
impossibility of prediction of the evolution of this solution on a long enough period
of time due to a chaotic behavior ([BER 84], [ORS 77]).
In conclusion, the notions of determinism and causality are far from being
universally applicable in the domains which we will cover.
1.1.2. Systems analysis and thermodynamics
1.1.2.1. Introduction
The analysis of systems is a discipline which consists of constructing a model or
a representation of a system characterized by observations and measurements, with a
view to predicting the behavior of this system at a later stage, under conditions
which may be different from those first encountered. We also attempt to contrive
means of manipulating the system in order to cause it to evolve in a manner which
we specify a priori. We thus enter into the domain of command and control, since it
is now necessary to verify that the results are those sought, and if not, to perform the
necessary corrections in order to obtain the desired results.
The characterization of a material device can thus vary depending on the

objective which we seek to achieve. The device may be static and we may only be
interested in its “state”; it may be dynamic, in which case it evolves as a time
function.
In general, the objective of a system is to transform some input quantities u(t),

known thanks to some measurement (which provides the input variables), into some
output quantities y(t) which are also obtained via a measurement (output variables).
For example, the input variables of a heating system are the available heating power,
the desired temperature, and the output (controlled) variables are the power
consumed and the temperature observed in the space to be heated. We also dispose
of a command variable for the heating system. The input variables are thus the given
conditions, while the output variables are the quantities obtained. Observations can
be made for the time evolution of the various quantities in a continuous or sampled
manner.
1.1.2.2. External description (black box)

The description of a system may be external, that is we satisfy ourselves to
simply measuring the inputs and outputs of the system, the system itself remaining a
“black box”. We thereby ignore what goes on inside the system. As the system
operates we measure y(t) which depends on the input u(t) and time t. Often, the
system “has a history”, and the output y(t) cannot be represented as a function of the
only two variables u(t) and t.
The external description of a state is thus generally not sufficient. The difference
between a raw egg and a hard-boiled egg is not visible to external measurements
(size, mass, color, etc.); it is a result of internal variables (chemical composition)
which cannot be measured directly, but which can be known indirectly (the
rotational movement of a mass of solid and a mass of liquid are not the same), or by
virtue of some previous known history (the egg was boiled).
From a mathematical point of view, the black box description corresponds to a

direct relationship between the inputs and the outputs, in other words to calculations
defined a priori on the input quantities. As long as the dynamic system is invariant
in time, the formalism of transfer functions (or of impulse responses) is largely used.
It is nonetheless necessary to pay close attention to questions of causality when
using such approaches (see Chapter 7; for more detailed information, the reader is
referred to works which deal with signal processing and automatic control theory).
1.1.2.3. Internal description (state variable approach)

In place of a “black box” description, we substitute a description of the internal
state of the system using a number of state variables X(t) (state vector). These
characterize the “state” of the system, and when combined with a knowledge of the
system inputs, knowledge of the system outputs can be obtained at every instant by
means of evolution equations (ordinary or partial differential equations) which
describe the conditions on geometric boundary of the system and the initial state.
The simplest dynamic systems are represented by constant-coefficient linear

differential equations; these are known as invariant linear systems as their response
does not depend on the initial instant chosen for the study of their evolution:
dX
AX BU (t ); t t 0 : X (t 0 ) X0
dt
In order to identify a state representation, we can use purely mathematical

considerations which are essentially based on the nature of the response of the
system (system outputs) to a specific excitation (Dirac impulse or step function). If it
is possible to identify the existence of different time constants, for example W1 and
W2, then the behavior of the system can be considered to be second order, which
implies the need for a description based on two state variables. State variables
identified via an empirical modeling approach will not necessarily lend themselves
to a clear physical interpretation. They are merely indicators which are linked in
some way to the dominant physical quantities of the system. We will come back to
this point when we discuss model reduction methods (Chapter 8).
'T
x
e e
Figure 1.1. Temperature pulse function unrealizable from imposed conditions

on the walls (identical temperature on the two walls)
Finally, we may wish to manipulate certain system variables in order to achieve

a given desired state. From a mathematical point of view, boundary conditions must
of course exist which allow a solution of the local equations (partial differential
equations) corresponding to the evolution of the physical system towards such a
final state. This condition is not always satisfied, as shown in Figure 1.1. In this
example, the physical system considered is not controllable.
1.1.2.4. Thermodynamics and mechanics

This chapter and Chapter 2 are dedicated to a presentation of those basic physical
laws which are valid regardless of the particular properties of the material elements
considered. These basic laws constitute thermodynamics and mechanics; they need
to be completed by means of other particular laws which may play a role in the
behavior of the material elements, associated for example with physics (state
equations of compressible fluids), chemistry, electricity, magnetism,
electromagnetism, or any combination of these disciplines (laser-matter interactions,
plasmas, chemical reactions or electrolysis in flows, etc.).
The laws of thermodynamics derive from the laws of mechanics applied to

ensembles comprising a very large number n of molecules (statistical mechanics).
The properties resulting from interactions between these n molecules cannot be
exactly established for a variety of reasons (residual quantum effects, computations
rendered impossible for very large numbers of particles, etc.). We therefore need to
complete our microscopic mechanical models (kinetic theory of gas, molecular
theory of liquids) by means of additional statistical axioms.
Thermostatics provides interpretations of physical quantities using the notion of

balance via the intermediary of extensive quantities. This is the equivalent of
imposing conservation principles for certain quantities, whose creation,
disappearance or variation is not spontaneous, but which is associated with a clear
cause that results in the transformation or displacement of the quantity considered.
This static study of the properties of material systems is firstly made in a reference
frame in which the material does not move, or at least under conditions such that the
effects of movement have no effect on this material.
When considering balances, a knowledge of time only serves to localize various

instants, while its definition is not important due to the infinitely slow nature of
thermostatic transformations. On the contrary, the definition of time in
thermodynamics is of great importance for the study and the prediction of the
velocity of a system’s temporal evolution. On the other hand, the equations of
thermodynamics and its related disciplines must be associated with boundary and
initial conditions which allow solutions that are actually observed in reality.
1.1.3. The notion of state
In thermodynamics, a state is a set of material elements which have well-defined

properties. In order to characterize the state (a) of this ensemble, physical quantities
Gi must be defined which can be measured (measurements gi) and which allow us to
distinguish between these and other material elements, or the same elements at
another instant, after a transformation. From a mathematical point of view, a state is
thus constituted by an ensemble of variables gi which characterize the material
contained in some entity or geometric domain. States thus defined obey the usual
rules of the set theory ([GIL 64], [BOC 92]). We often refer to this material as being
in state (a). It is clear that once defined as being in a given single state, the notion of
a system does not supply any additional information with respect to the notion of
state. The state of a system may be more or less complex and its description may
require a more or less large number of variables, depending on the case considered.
As an example, let us consider 2n contiguously arranged plates of a homogenous

material (Figure 1.2a) distributed in three separate blocks by two thin thermally
insulated layers P1 and P2. Suppose that the notion of temperature is known (for this
example); half of these plates are at a temperature Ti which is greater than the
temperature of the other half (Figure 1.2a). The description of this initial state thus
requires that 2n temperatures be given. Let us now cause this state to evolve, under
the constraints imposed by the thermally insulated lateral faces. These 2n variables
are not necessarily required; the walls P1 and P2 play the role of a strong thermal
resistance, the blocks of plates have an approximately uniform temperature at each
instant (Figure 1.2b); these three temperatures suffice for a description of the state of
the system and its subsequent temporal evolution. After a sufficiently long time, the
1 2n
state is at a uniform temperature T f ¦ Ti . This final state, which is described
2n i 1
by a single variable is clearly in a state of equilibrium.
T T
P1 P2 P1 P2
Tf Tf
(a) (b)
Figure 1.2. (a) System with 2n variables; (b) system with 3 variables
The general problem of describing a state comes down to finding the necessary
variables. From the preceding example we see that the number of necessary
variables depends largely on the physical situation we wish to describe. The more
complex the system considered, the greater the number of variables required. We
will frequently come back to this point, emphasizing it with respect to the specific
objectives.
A state of equilibrium is in fact a succession of states for which all of the

variables that constitute it conserve a constant value, physical exchanges with the
exterior having ceased.
1.1.4. Processes and systems

1.1.4.1. Definition of a process
Certain authors define a process (a,b) as a pair of states: initial (a) and final (b).
They are thereby led to distinguish between states which are possible and those
which are not. Insofar as we limit ourselves to only consider processes which are
truly observed (physical processes), the discussion of an axiomatization concerning
impossible processes, being ill-defined, is beyond the scope and objectives of this
book.
From a physical perspective, this means that a process can only be defined if the
initial and boundary conditions are entirely determined during the process. It makes
no sense to speak of a process which allows us to pass from a state (a) to state (b)
unless the external conditions which constrain that process are specified. This is no
longer a mathematical question, but rather a problem related to a determinism which
amounts to admitting that an initial state (a), well-defined and always subjected to
the same constraints, will always lead to the same final state (b). It will always be
possible to relate two given states, under the condition that, on the one hand,
exchanges with the exterior furnish the necessary physical quantities, and on the
other, that the internal system processes which redistribute these quantities allow the
desired distribution of these system quantities to be achieved. For example, it is not
possible to realize a state consisting of a given mass whose temperature distribution
comprises a central peak (Figure 1.1) by means of an action at the exterior walls.
The necessary energy must be directly supplied to the central zone, which must be
insulated from the adjacent regions.
By definition, a process is a series of states. This mathematical definition only

has physical relevance for processes representative of real evolutions. While not
precluding a choice of states with no link (a rabbit, a carrot, etc.), the obtention of
physical evolution laws for matter implies a “certain continuity of content” for this
ensemble of states. The same goes for all practical problems. A process is therefore
a succession of states which must be uniquely defined. Apart from some exceptions
(shocks), we will only consider processes comprising a continuous series of states,
described by variables which must be continuous functions of time. We will
however allow situations with discontinuities (shocks, shockwaves, deflagration)
which momentarily violate this continuity condition.
1.1.4.2. The notion of a system

The notion of a system is a relatively vague one; it is in fact included in that of a
process: a system is an entity which we consider during a process. As our
considerations often take a differential form, the system is the principal part (zero
order) on which we perform differential balances. The notion of a system is not
clearly identified if we do not entirely state the conditions of the process under
study.
Take the following examples:

1) The matter studied remains enclosed in a fixed volume (a cylindrical
calorimetric bomb, for example). The observation domains D and D' at times t and t'
are contained in the interior of the cavity C (Figure 1.3a). The presence of the barrier
constituted by the rigid wall leads to the matter inside C being constrained to remain
within the system (Figure 1.3a). This is a closed system.
D’=D(t’)
D(t) D0
V V
C=D=D’
(a) (b) (c)
Figure 1.3. (a) and (b): closed systems; (c) open system
2) Now consider the case where matter is caused to move at a velocity V with
respect to the used reference. The observation and description of this matter in
movement can be performed:
– either, by following the matter in its movement, in which case the observation
domain D(t) is displaced in time (Figure 1.3b); the matter contained within the
domain D(t) constitutes a closed system in the sense just defined;
– or, by considering a fixed domain D0 in which the matter is continually
renewed; the ensemble of states contained within the fixed geometric domain is
qualified as an open system, in other words a system which exchanges matter with
the exterior (Figure 1.3c).
In Chapter 3 we will encounter these different ways of describing the movement

of matter in the form of substantial (Lagrangian variables) and spatial (Eulerian
variables) descriptions.
By definition, we will say that a process describing the evolution of material

elements which are identified, and remain unchanged, operates on a closed system,
in other words a system which does not exchange (does not provide or receive)
matter with the exterior of the system. We can also use the denomination material
system.
On the other hand, a process (a series of states) during which some matter passes
from the inside of the domain to its outside corresponds to an open system.
1.1.4.3. Types of processes and states

We normally define the terminology of processes and states in the following
way:
– a natural process is undergone by an isolated system which is not subjected to
any external action;
– a reversible process is a process for which the direction of time can be
changed. It is of course a process in which no entropy is created. Such processes
occur over infinitely slow transformations;
– a quasi-static process is a succession of close infinitely equilibrium states;
– a possible process is a process where the constraints placed on entropy are
obeyed. In the opposite case we speak of an impossible process;
– a state of equilibrium is in fact the result of a succession of states whose
variables remain constant, any exchanges of its physical properties with the exterior
having ceased.
1.1.4.4. Enclosures and walls

A diathermic wall is a wall which is permeable to heat and to external sources of
entropy.
An adiabatic wall is impermeable to heat and does not allow the passage of
entropy. It thermally insulates the system from the exterior.
1.2. Axioms of thermostatics
1.2.1. Introduction
The traditional presentation of thermodynamics usually begins with a direct

definition of the various quantities (force, pressure, etc.) which are then used in the
subsequent definition of elementary work and heat. The first and second principles
(conservation of energy and entropy respectively) are then stated, to which further
laws are then added (conservation of mass, chemical species, etc.). This all leads to
the differential form of energy being written as the differential of a function E,
energy. The result is a lack of coherence well known by students of
thermodynamics. This situation can be avoided by means of a more structured
presentation concerning the extensive quantities, among which energy plays a
particular role in physics, whereas entropy is the basis of all considerations
pertaining to irreversible thermodynamics. We will also more clearly outline the fact
that the more complex and irreversible the evolution considered, the greater the
number of variables required.
1.2.2. Extensive quantities
1.2.2.1. Definition of extensive quantities

Among the quantities used for the description of a state, we postulate (basic
principle) that for every system S there exists an ensemble of n extensive quantities
whose measure is proportional to the extension of the system, and which are always
defined regardless of the state considered.
S
b
a
Figure 1.4. Disjoint states
Consider two disjoint states, i.e. states which have no matter in common; the
extensive quantity Xi associated with the ensemble of the two states (a) and (b) is
equal to the sum of the quantities corresponding to each state:
X i (a b) X i (a ) X i (b)
If the sub-ensembles corresponding to (a) and (b) are not disjoint, we clearly
have:
X i ( a b) X i (a) X i (b) X i (a b)
This definition only concerns the description of a state (a collection of matter) at

a given instant. Under no circumstances does it imply the same property for two
separated sub-systems (a) and (b) which we bring together in an externally applied
field (force field, electromagnetic field, etc.), as this would constitute a process
(series of states).
The index i takes on small integer values as these quantities are generally small
in number for a state. The following quantities are extensive: mass, volume, number
of moles, energy, entropy, force, etc.
1.2.2.2. Natural processes

The usual (historic) presentation of thermodynamics begins with a statement of
the first law of thermodynamics, which sees energy play a central role compared
with the other extensive quantities. This is in fact only one of the conservation
principles among the axioms of thermostatics, which affirm the existence of
conservation laws all possessing the same structure. Consider a material system (i.e.
a collection of defined material elements) which evolves according to a natural
process (without exchange with the exterior). The system is described in a Galilean
(or inertial) reference frame in relation to which it is fixed (no dynamic effects in the
sense of mechanics are authorized), and only geometric changes of reference frame
are permitted.
The following principles can be established:

– The entropy S of the material system is non-decaying during the natural
evolution of a process.
– The other extensive quantities Xi of a material system remain constant during
the natural evolution of a process.
The preceding axioms actually express the existence of physical conservation

laws, except for the entropy which can only increase during the spontaneous
evolution of an isolated system. The axiom of extensive quantities also implies that
the concept of variation in the extensive quantities of a system (a) is meaningful: it
is possible to conceive of the quantity 'Xi of which (a) has gained an amount X.
S (b)
X ao b X bo a
(a)
Figure 1.5. Action and reaction in a system
1.2.2.3. Action and reaction

The preceding definition implies a principle of action and reaction: let us divide
the considered system into two sub-systems (a) and (b), separated by a surface S
(Figure 1.5). The extensive quantity X b o a gained by (a) to the detriment of (b) is
opposed to the extensive quantity X a o b gained by (b) to the detriment of (a). This
amounts to saying that the action exercised by (a) on (b), the source of the transfer
between (a) and (b), is opposed to the action of (b) on (a):
X a ob X b o a 0 [1.2]
To be more general, we can consider a process during which the sub-systems (a)
and (b) receive in addition the quantities Xa ext and Xb ext from the exterior of the
system S:
Xa X a ext X b o a ; Xb X b ext X a o b
The conservation of the quantity X in the system implies that the total amount of
the quantity X is equal to the amount of this quantity which is received from the
exterior:
Xa Xb X a ext X b ext
Equality [1.2] is once again the result.
The preceding reasoning is not applicable to entropy, for which we do not have
conservation, but only non-decay. This results in the principle of action and reaction,
which, as applied to the entropy, is written:
S a ob Sb o a t 0 [1.3]
This inequality implies that one of the sub-systems has gained more entropy than
the other has lost.
The actions leading to a gain in the quantity X in system (a) can be classed into
two categories:
– volume actions related to fields which can exert their influence from a distance:
an electric field produces a displacement of electric charges, electromagnetic
radiation can lead to a the creation of heat by absorption in the material medium,
etc.;
– contact actions (on a surface) between (a) and (b) (Figure 1.5) essentially
results from the action between the molecules in the immediate vicinity of the
surface S (at distances of the order of the intermolecular distances or of the mean
free path in a gas). As this volume is based on the surface S and is of small
thickness, it can be modeled by the surface S, on which we can concentrate the
interaction between (a) and (b). The usual contact actions are pressure, frictional
force, thermal conduction, molecular diffusion, etc.
1.2.2.4. States of equilibrium and extensive variables

Let us consider a material system in a given initial state. Suppose that excepting
its entropy, the values of its extensive quantities are known. We thus have a partial
characterization which does not allow a full description of the state of the system. It
is clear that a knowledge of the global values of mass, energy, etc., does not provide
any understanding of how these quantities are distributed within the system.
Let us now isolate the system and leave it to evolve according to a natural
process; the extensive quantities which it contains will naturally distribute
themselves in a balanced way throughout the system. It is easy to understand that the
final equilibrium state of the system is unique, as confirmed by experience. In other
words, the extensive quantities of a material system allow a description of the
equilibrium state achieved by a system at the end of all natural processes (provided
the system remains isolated). Following the axiom pertaining to the entropy (section
1.2.2.2), the entropy associated with such a material system increases in order to
attain a maximum value corresponding to this final state of equilibrium. We thereby
deduce that this entropy of the equilibrium state is a well-determined function of the
other extensive quantities for the system considered:
S S( X i ) [1.4]
Relation [1.4] constitutes the entropic representation of the system in

equilibrium. It is in fact a state equation (in the thermodynamic sense) for
equilibrium states which are only comprised of the extensive variables. In order not
to confuse it with the usual state equations, we will refer to it as the general
equation of state of a system. This relation is unique for a given material system,
although numerous state equations, more or less dependent on it, can be derived
from it. The variance of a system is the number of independent extensive quantities
necessary for its representation.
The existence of this general equation of state leads to the properties of

thermostatics.
1.2.2.5. Homogenity
We have said that the extensive quantities of a material system are proportional
to its extension without stating exactly what this extension is. Let us suppose that
these extensive quantities Xi and the entropy S are all associated with a given
volume of ordinary 3D space. Taking two systems SO and S with material contents
which are homothetic (or which are geometrically similar) with a ratio of O and
which are constituted from the same matter, under the same physical conditions, at
homologous points, we can say that the extensive quantities Xi are proportional to a
cube with a reference dimension:
X Oi O 3 Xi SO O 3S
Denoting our two states by (a) and p (a), with p = O3, state relation [1.4] can be
written:
Sp S ( pX i ) pS ( X i ) [1.5]
The function S X i is a first degree homogenous function.
1.2.2.6. Note
This property no longer holds if the extensive quantities are associated with
spaces of different dimensions. As an example consider a drop of water whose total
energy is the sum of two terms, on the one hand an energy associated with its
weight, Eg, proportional to its volume, and on the other hand an energy associated
with its surface tension, ET, proportional to its surface:
E E g ET
The total energy of the system SO becomes, considering a homothetic ratio O
EO O 3 E g O 2 ET
Other extensive quantities Xi (for example, mass or volume) of this drop remain
proportional to the homothetic ratio O3 and respect relation [1.4]. It is therefore clear
that relation [1.5] is no longer respected. We will not consider such cases in what
follows.
1.2.3. Energy, work and heat
1.2.3.1. Energetic representation of a system

Among the extensive quantities, the energy, which we denote E, is often
assigned a particular role. Depending on whether it is explicitly solved in S, or in E
(energetic representation), the general equation of state [1.4] of a system can be
written:
S S ( E, X i ) or E E (S , X i ) [1.6]
In practical applications at usual temperatures, energy and entropy only appear in

terms of their variations. In these conditions, it is not necessary to define them in an
absolute sense and it is sufficient to evaluate them to the nearest constant.
For a system at rest in a Galilean reference frame, the energy of a system is

essentially comprised of its internal energy and the different forms of potential energy
of the system elements. Changing the Galilean reference frame would amount to adding
a constant amount of kinetic energy, which clearly does not change our description.
This will no longer be the case if elements of the system have different movements, in
which case their respective kinetic energies would need to be accounted for.
1.2.3.2. Work and heat

By calculating the differential of relation [1.6]:
wE wE § ·
dE dS ¦ dX i
1 ¨ dS ¦ wS dX i ¸
wS wX i wS ¨ ¸
i © i wX i ¹
wE
we can define the elementary work dW and the elementary heat dQ by the relations:
wE 1 wE 1 wS
dQ dS dS dW ¦ dX i ¦ dX i
wS wS i wX i wS i wX i [1.7]
wE wE
such that we obtain the classic definition dE dW dQ , in which the differential

forms dW and dQ are not differentials of a function, i.e. they are not exact differentials.
1.3. Consequences of the axioms of thermostatics

1.3.1. Intensive variables
1.3.1.1. Definition and properties
We will only study equilibrium states entirely characterized by the values of the
extensive variables and the relation of state [1.4] or [1.6] between them.
We define entropic intensive variables using the relations:
wS 1 wS
Zi [1.8]
wX i T wE
Zi (resp. 1/T) being called the conjugated variable of Xi (resp. E) with respect to S.
The energetic intensive variables are similarly defined:
wE wE
Yi T [1.9]
wX i wS
Yi (resp. T) is the conjugated variable of Xi (resp. S) with respect to E.
The differentials dE of the energy and dS of the entropy can be written:
dE TdS ¦ Yi dX i [1.10]
i
1
dS dE ¦ Z i dX i [1.11]
T i
From this, the relation between the energetic and entropic intensive variables,
respectively Yi and Zi can be obtained:
Yi TZi [1.12]
Consider a fluid defined by the extensive variables (volume V, number of moles

N, energy E and entropy S), the energetic intensive variables are the pressure p (to
the nearest sign), the temperature T and the chemical potential P
dE pdV TdS PdN

§ wE · § wE · § wE ·
with : p ¨¨ ¸¸ ; T ¨¨ ¸¸ ; P ¨¨ ¸¸
© wV ¹ N , S © wS ¹V , N © wN ¹V , S
1.3.1.2. Consequences of homogenity

Differentiating relation [1.5] with respect to p, and letting p = 1, we obtain:
wS wS E
S E ¦ Xi ¦ Zi X i [1.13]
wE i wX i T i
and:
E TS ¦ Yi X i [1.14]
i
We note that the general relation [1.13] or [1.14] does not allow the
characterization of a system. It constitutes a differential equation which is satisfied
by any homogenous function S of degree 1. For example, for a fluid, the following
relation can be written:
E TS pV PN
Differentiating relation [1.13] or [1.14], and comparing with expression [1.10] or

[1.11], we obtain the relations:
SdT+¦ X i dYi =0 (Gibbs-Duhem relation) [1.15]

i
§1·
Ed ¨¨ ¸¸ ¦ X i dZ i 0 [1.16]
©T ¹ i
1.3.1.3. The equations of state

An expression relating the extensive and the intensive quantities which
characterize a system is habitually called an equation of state.
The intensive variables, being by definition partial derivatives of a function

E ( S , X ,...) , are not independent. In practice we rarely use the general equation of
state [1.6], which only relates the extensive quantities, using instead the more usual
equations of state which relate extensive and intensive variables, which can be more
or less dependent.
For example, for a perfect gas, we know that there are two equations of state:
pV NRT dE mC v dT
These are not independent as the first equation of state implies that the specific
heat Cv can only be a function of the temperature (Joule’s law; see elementary works
on thermostatics).
1.3.2. Thermodynamic potentials
While the preceding presentation assigns a particular role to the extensive
variables, using them to represent a given system does not lead to the most useful
means of studying that system. The temperature or pressure of a fluid are of a more
direct interest than its volume or its energy when it comes to isobaric or isothermal
processes. We often prefer a combination of intensive and extensive variables for
studying a given system. From a mathematical point of view, such changing of
variables can be relatively complex (e.g. contact or Legendre transformations [IGO

89], [COU 89], [BYU 02]).
Let us consider a simplified situation comprising a function f ( x, y ) in two

variables. Instead of taking the couple (x,y), we take the new variables ( x, wf / wy ):
we thus replace the variable y with the derivative wf / wy , the function f not being
known a priori. We cannot therefore hope to obtain the explicit properties, only
those of the differentials being accessible. We thus perform a change of function,
replacing the differential of the extensive variable with the differential of the
conjugate intensive variable. The simplest example is the introduction of the
enthalpy H:
H(p,S,N) E pV
For a fluid, the differential dH can be written:
dH d E pV Vdp TdS PdN
whence we see the intensive variable p appears in differential form. This is

particularly convenient for the study of isobaric transformations or shaft work.
1.3.2.2. Definition of thermodynamic potentials

The preceding method is a general one. Suppose for instance that we want to
choose the collection of independent variables ( X i , Y j )2: we define the function

E X i ,Y j :

E X i ,Y j E ¦ X jY j
j

Using relation [1.10]3 the differential of the function E X i , Y j can be written
as:
dE
dE ¦ X j dY j Y j dX j ¦ Yi dX i ¦ X j dY j [1.17]
j i j
The function E is a thermodynamic potential whose partial derivatives:
2 The subscripts i and j correspond respectively to the different couples of extensive and
intensive variables.
3 Here, S is included in extensive variables X and is not explicitly written.
§ wE ·
Xk ¨ ¸
¨ wY ¸X ,Y
© k ¹ i j z k
with respect to the intensive variables Yk, are the corresponding extensive variables4.
Using this expression in the definition of E , we obtain the Gibbs-Helmholtz

relation (a partial differential equation for E as a function of E):
wE
E¦ Yj E [1.18]
j wY j
Remember that we frequently use:

– the Helmholtz function F V , T , N (free energy);
F V , T , N
E TS
hence: dF pdV SdT μdN
sF sF sF
with: p S μ
sV sT sN
– the Gibbs function G p, T , N (free enthalpy):
G p, T , N
E pV TS
hence: dG Vdp SdT μdN
sG sG sG
with: V S μ
sp sT sN
The thermodynamic potentials are first order homogenous functions which

satisfy Gibbs-Duhem relations (section 1.3.1.2). They are useful for the study of
systems where certain intensive variables remain constant.
4 Notation w wYk X , Y specifies that the chosen independent variables are Xi and Yj.
i jzk
For example, during a constant-pressure (pa) evolution of a system whose

volume increases by an amount 'V, the system receives an amount of work
p a .'V from the exterior. If W denotes the work (aside from that due to the
pressure pa), the variation of enthalpy 'H of the system is (section 1.3.2.1):
'H 'E p a .V 'E p a .'V W Q [1.19]
Generally speaking, the source of an extensive quantity Xj, for which the
intensive variable Yj is constant, providing the system with 'Xj by means of a quasi-
static process, leads to a variation of its thermodynamic potential equal to:
'E 'E Y j 'X j W Q
(W and Q being respectively the work and heat received from means other than the
source, with Yj being constant).
1.3.2.3. Thermostatics and variables change

The traditional practical problem of thermostatics consists of passing from a
representation of a system with n variables to another representation with n variables
related to the first by means of conjugation properties in the entropic and energetic
representations. The number of possible combinations of n independent variables
among the 2n extensive or intensive variables is clearly large. It is in fact the
principal practical difficulty of thermodynamics. This multiplicity of possible
independent variables leads to the numerous Maxwell relations between the
coefficients of the differential forms of energy, entropy, enthalpy and the different
thermodynamic potentials; the general form can be written by means of the
differential d E [1.17]:
wX i wY j wYi wYi ' wX j wX j'

wX j wYi wX i ' wX i wY j ' wYi
We will leave it to the reader to work out examples involving the explicit
functions given below.
Remember the rules for changing variables when using partial derivatives: if the
three variables (x,y,z) are related, it is straightforward to show that:
§ wx · § wy · § wz · § wx · § wy ·
¨ ¸ ¨¨ ¸¸ ¨¨ ¸¸ 1 ¨ ¸ ¨¨ ¸¸ 1 [1.20]
¨ w y ¸ © wz ¹ © wx ¹ ¨ wy ¸ © wx ¹
© ¹z x y © ¹z z
1.3.2.4. “Thermodynamic” coefficients

For quasi-static closed process (N=const) we define the coefficients using units
of mass or of volume. The specific heats at constant volume Cv and at constant
pressure Cp, and the calorific coefficients h and l are:
dQ TdS NMC v dT ldV NMC p dT hdp
from which we can derive the relations:
T § wS · T w2 F
Cv ¨¨ ¸¸
NM © wT ¹ V , N NM wT 2
[1.21]
T § wS · T w2G
Cp ¨¨ ¸¸
NM © wT ¹ p, N NM wT 2
§ wS · w 2F § wp ·
l T ¨¨ ¸¸ T T ¨¨ ¸¸ [1.22]
© wV ¹T , N wVwT © wT ¹V , N
§ wS · w 2G § wV ·
h T ¨¨ ¸¸ T T ¨¨ ¸¸ [1.23]
© wP ¹ T , N wPwT © wT ¹ P, N
In general, depending on the thermodynamic transformations studied, we may

choose diverse independent variables. For example, we define the calorific
coefficients Oand P:
dQ TdS O dp P dv
and the isothermal and adiabatic compressibility coefficients, respectively FT and FS:
1 § wV · 1 § wV ·
FT ¨ ¸ F ¨ ¸ [1.24]
V ¨© wp ¸ V ¨© wp ¸
S
¹T ¹S
Recall the Reech relation (not derived here):
FT Cp
J
FS CV
The speed of sound is defined by the relation (U being specific mass):
§ wp · 1
c2 ¨ ¸
¨ wU ¸ [1.25]
© ¹ SN UF S
The coefficients of dilatation are defined by the relations:
1 § wV · 1 w 2G 1 § wp · 1 w2F
D ¨¨ ¸¸ ; E ¨¨ ¸¸ [1.26]
V © wT ¹ P V wTwp p © wT ¹V p wTwV
We recall the usual relations, which the reader can verify using identities [1.20]:
T § wV · § wp · § wC v · T §¨ w 2 p ·¸
C p Cv ¨¨ ¸¸ ¨¨ ¸¸ ; ¨¨ ¸¸ ;
NM © wT ¹ p © wT ¹V © wV ¹T NM ¨© wT 2 ¸¹
V
§ wC p · T §¨ w 2V ·
¸ ; § wT · § wT ·
¨ ¸ O NMC v ¨¨ ¸ ; P NMC p ¨¨ ¸¸ ;
¨ wp ¸ NM ¨© wT 2 ¸ ¸
© ¹T ¹p © wp ¹V © wV ¹p
E J p § ws · § ws · § ws · C p § wT · Cp
c2 ; ¨ ¸
¨ wU ¸ ¨¨ ¸¸ c 2 ; ¨ ¸
¨ wU ¸ ¨ ¸
D U © ¹p © wp ¹ U © ¹p T ¨© wU ¸¹ p D UT
We deduce from this that the differential of the specific entropy (per unit mass) s
as a function of the dp and dU variables can be written in one of the following forms:
ds
§ ws · § ws ·
¨ ¸ dp ¨
¨ wp ¸
¸
¨ wU ¸ dU
§ ws ·
¨ wp ¸
¨ ¸ dp c 2 dU
© ¹U © ¹p © ¹U
[1.27]
Cv
E pT
dp
Cp
D UT
dU
E pT
Cv
dp c dU
2
1.3.2.5. Perfect gas

Consider a perfect gas with constant specific heats C v and C p and of molar mass
M. By reasoning using the unit of mass, state relation [1.6] between the extensive
quantities can be written:
e ¬ s ¬
s Cv Ln 1 const e 1 exp const
® Cv ®
with: m V NM V ; s S m; e E m hH m
From the intensive variable definitions, we can easily derive the usual state
equations:
p rU T e CvT const with: R r0 . joule/mole
We can thereby obtain expressions for the extensive variables as functions of

other variables, (e0, h0, s0) being some constants:
p Jp
e C v T e0 e0 ; h C p T h0 h0 ;
J 1U J 1U
§ T · § T · § p ·
s C v Ln¨ ¸ s0 C p Ln¨ ¸s C v Ln¨ ¸ s0
¨ U J 1 ¸ ¨ p J 1 J ¸ 0 ¨ UJ ¸
© ¹ © ¹ © ¹
in addition to expressions for the usual coefficients given below:
1 1 J p
l p; h ; D E ; c2 J rT U FS ;
U T U [1.28]
C p Cv r; C p Cv J.
1.4. Out-of-equilibrium states

1.4.1. Introduction
The reasoning of section 1.2.2.4 shows how a system which is not in equilibrium
does not have a general state relation outside of equilibrium conditions. The
extensive quantities remain defined, but they are no longer sufficient to characterize
the state of the system, as the entropy can no longer be defined as a function of these
parameters alone. As seen in section 1.2.2.4, the distribution of extensive variables
is uniquely defined for a system in equilibrium, whereas this is no longer the case
for a system which is not in equilibrium: a more detailed description of the structure
of the system is thus necessary. This implies that more parameters will be necessary
a priori for a description of an out-of-equilibrium system than for one which is in a
state of equilibrium.
The general method for describing an out-of-equilibrium system involves

considering the system as a collection of sub-systems, each of which is in a state of
equilibrium and which can thus be described by means of their extensive variables.
We suppose (postulate) that such a procedure is always possible.
As the system is no longer in equilibrium, exchanges occur between the

extensive variables of the various sub-systems, and the intensity of these exchanges
must be characterized. It is now essential that time be homogenous, in other words,
for an out-of-equilibrium system whose characteristics do not change, the amounts
of quantities exchanged must be proportional to the duration of the transfers. At this
point, it is sufficient to consider continuous matter at rest, i.e. in a fixed reference
frame.
1.4.2. Discontinuous systems

1.4.2.1. General principles
A real system nearly always comprises a continuous variation of its physical
properties. We therefore represent the latter using piecewise constant functions
defined on a partition of the system in P sub-systems, each of which is
approximately in a state of equilibrium, and to which we can therefore apply the
properties of systems in equilibrium. Let k be the number of independent extensive
variables required for a description of each sub-system (number of moles, volume,
energy, entropy, etc.) in terms of an energetic or entropic representation.
A knowledge of the state of the system S requires a complete description of the P

sub-systems, i.e. a total of kP variables. As seen earlier, certain extensive variables
can be replaced by their corresponding intensive variables, which are defined for
each sub-system as a result of the hypothesis that these sub-systems are in a state of
equilibrium. This of course does not change the total number of independent
variables, kP. For each sub-system p, we have the entropic form of the general
equation of state (the energy of each sub-system, which is not individualized here, is
included in the variables Xi):
Sp
S p X ip p 1,..., P; i 1,..., k
where Xip designates the extensive quantity Xi contained in the sub-system p.
The extensive quantities of the complete system can be obtained by adding the
corresponding extensive quantities of the sub-systems:
¦ S p X ip ( p 1,..., P)
P P
Xi ¦ X ip ; S
p 1 p 1
Note that in general, it is obviously not possible to obtain a relation between the
extensive quantities Xi of a system and its entropy S, since the kP variables Xip must
be eliminated from the k + 1 preceding equations. This fact bears witness to the
absence of a general equation of state for an out-of-equilibrium system.
The extensive quantities Xi are generally functions of time. The flux Mip of the
quantity Xi received by the component p is defined by:
dX ip
M ip
dt
The origin of the quantities dXip and the flux Mip can be considered individually
for each of the sub-systems. Let dXip,q (resp. dXiq,p) be the amount of extensive
quantity Xi received by the sub-system p (resp. q) from the sub-system q (resp. p) in
time dt. The amount of entropy dSp,q (resp. dSq,p) received by each of the systems
and associated with the aforementioned transfer is evaluated by means of the
entropic representation differential of each of the sub-systems:
k k
dS p , q ¦ Z ip dX ip, q dS q, p ¦ Z iq dX iq, p
i 1 i 1
The intensive variables Zip and Ziq of two neighboring sub-systems will have
different values if the sub-systems are not in equilibrium, and so the entropy
variations dSp,q and dSq,p will have different absolute values. Indeed, there can only
be an exact balance between the two sub-systems if the intensive variables Zip and
Ziq are equal.
With the exception of entropy, “exchanged” quantities obey the principle of

action and reaction which results from the conservation principles (section 1.2.2.3):
dX ip , q dX iq , p dS p, q dS q , p t 0 [1.29]
During an irreversible transfer of an extensive quantity, the entropy gained by a

body is greater than the entropy lost by the body from which the extensive quantity
is transferred (this statement is algebraically true).
The distribution of intensive variables is thus a constant function of space within

the spatial bounds of a given sub-system, discontinuities at the frontier of each sub-
system existing in proportion to the degree of thermodynamic imbalance within the
system. Such imbalance leads to an exchange of the quantities dXip,q between the
sub-systems. It is thus necessary to introduce relations between these causes and
their effects. Before dealing with the general problem of these relations, we will
illustrate the methodology using the following elementary example, comprising two
sub-systems which only exchange heat between themselves, or with the exterior.
1.4.2.2. An insulated thermal system

1.4.2.2.1. Entropy variation
The simplest example is an ensemble which is insulated from the exterior
(Figure 1.6a), comprising two conducting blocks separated by a diathermic wall
which creates a resistance to heat transfer (thermal resistance). Each of the subs-
systems is, by assumption, characterized by its calorific energy content Q1 (resp. Q2)
or its temperature T1 (resp. T2), assumed to be uniform at every instant. These
quantities vary slowly with time on account of the thermal resistance of the system.
Ti Ti
MT1,2 MT2,1 M1e MT1,2 MT2,1 M2e

T1 O T2 T1 O T2
x
x
A (a) A A (b) A
Figure 1.6. Heat transfer in an (a) insulated or (b) uninsulated system
The ensemble is not in equilibrium and is thus subject to a natural evolution

during which the entropy increases, while the internal energy remains constant. We
will calculate the variation in entropy between the state thus defined and the
equilibrium state.
The entropy of a solid mass m of constant specific heat capacity C is

S mC LnT const; the entropy of the ensemble of the two blocks, which are
assumed to be identical, is thus:
S S1 S 2 mC LnT1T2 const
As the ensemble is insulated from the exterior, the total amount of heat remains
constant:
mCT1 mCT2 const

The natural evolution of the system S from initial temperatures T10 and T20 leads
T10 T20
to the final temperature T f , which is identical for the two blocks, and
2
to the final entropy S of the system:
S 2mC LnTf const
The variation in entropy 'S between the final and initial instants is thus:
T10 T20 2 § T T20 2 ·

'S mcLn mcLn¨1 10 ¸
4T10 T20 ¨ 4T10 T 20 ¸
© ¹
It is always positive and independent of the intermediate evolution between the

initial and final instants.
1.4.2.2.2. Entropy sources

Let dQp,q be the quantity of heat received in time dt by the component p
(p 1,2 ) from the other component ( q 1,2 z p ):
dQ1,2 dQ 2,1 0
The heat flow MTp,q received by the sub-system p is the quantity of heat
dQ p, q dt received per unit time by the component p ( p 1,2 ) from the other
component. Thus, we have:
dS 1 dQ1,2 1 dQ2,1 dQ1, 2 § 1 ·

¨ 1 ¸ [1.30]
dt T1 dt T2 dt dt ¨© T1 T 2 ¸
¹
The rate of entropy creation dS dt is always positive on account of the fact that
the heat transfer naturally occurs from the hot body to the cold body:
T1 > T2 dQ1,2 < 0 and T1 < T2 dQ1,2 > 0
Let us consider the first situation (T1 > T2). The quantity dQ2,1 is positive; the
entropy (positive) dS2 gained by sub-system 2 is greater than the entropy -dS1 lost by
sub-system 1. This is in accordance with the fact that for irreversible heat transfer,
the entropy gained by a system is greater than the entropy lost by the body which
has provided the heat.
The higher the degree of the imbalance, the greater the thermal flux, and the
greater the rate of entropy creation.
1.4.2.3. The insulated system and entropy creation

Let us consider two bodies, each in equilibrium, and which exchange a quantity
X across a wall permeable for this quantity. We define the intensive entropic
parameters:
wS1 wS 2
Z1 or Z 2
wX 1 wX 2
During a quasi-static transformation where the quantity dX1,2 = -dX21 is

exchanged in time dt, we obtain:
dS dX 1, 2 dX 2,1 dX 1,2
Z1 Z2 Z 1 Z 2 [1.31]
dt dt dt dt
As the entropy can only increase, the value of X increases (resp. decreases) in the
sub-system for which the value of Z is greatest (smallest). The transfer of the
quantity X occurs spontaneously from the sub-system with the smallest value of Z to
the sub-system with the greatest value of Z.
As in the preceding section, we see that during an irreversible transfer of an

extensive quantity, the entropy gained by a system is greater than the entropy lost by
the body which has lost this extensive quantity.
Expression [1.31] can be immediately generalized to a system constituted of P

sub-systems characterized by I independent extensive quantities:
dS I P P ª dX ip , q º
¦ ¦ ¦ « Z i, p », q z p.
dt i 1 p 1 q 1 «¬ dt »¼
or, after grouping the opposing fluxes (p < q, so as to only count this component
once):
ª dX ip, q º
¦ ¦ ¦ «Z i, p Z i, q
dS I P P
» pq
dt i 1 p 1 q 1 ¬« dt ¼»
1.4.2.4. Systems with external exchanges and entropy source

Let us first consider the same thermal systems as above (section 1.4.2.2), but
where each of these receives a quantity of heat dQpe (p = 1,2) in time dt from the
exterior. For the heat exchange we have (Figure 1.6b):
dQ1 dQ1,2 dQ1e ; dQ2 dQ2,1 dQ2e ; with : dQ1,2 dQ2,1 0
The increase in system entropy is:
dQ1e dQ2 e dQ1, 2 § 1 ·

dS ¨ 1 ¸ [1.32]
T1 T2 T2 ¨© T1 T 2 ¸
¹
1 dQ1e 1 dQ2e
The external entropy supply terms and can now take any
T1 dt T2 dt
sign. The term in [1.32] which corresponds to irreversible internal heat exchange,
which is always positive, constitutes an entropy source. We thus have:
dQ pe
¦ d dS [1.33]
p 1,2 Tp
Suppose, in addition, that the external sources of heat are in thermostatic

equilibrium, such that their temperatures T1S and T2S are defined. These transfers are
natural internal evolutions for any ensemble constituted of a sub-system and a
corresponding external source. They are, as before, entropy generators; we can
therefore write:
dQ pe dQ pe dQ pe §¨ 1 1 ·
¸t0
¦ ¦ ¦ ¨

¸ [1.34]
p 1, 2 Tp p 1, 2 T pS p 1,2 T p © T p T pS ¹
By then evaluating the entropy lost by the external sources we obtain the
Clausius inequality:
dQ pe dQ pe
¦ d ¦ d dS [1.35]
p 1,2 T pS p 1,2 Tp
Similarly, for internal and external exchanges of an extensive quantity X, we

obtain, using the same notations as above:
dS dX 1e dX 2 e dX 1,2
Z1 Z2 Z1 Z 2
dt dt dt dt
positive term
dX 1,2
The term Z1 Z 2 , which is always positive, is the entropy source
dt
associated with internal exchanges. The rate of entropy creation is proportional to
dX 1, 2
the intensity of the internal exchanges and to the imbalance Z1 Z 2
dt
between the two sub-systems.
In general, the internal production of entropy dS int dt (entropy source) is

associated with the evolution of all of the extensive quantities in this system, which
is made up of two sub-systems:
dS int I dX i1, 2
¦ Z i1 Z i 2 i 1, 2,....I [1.36]
dt i 1 dt
Finally, let us recall that the entropic intensive quantities Zi are related to the
energetic intensive quantities ([1.12]: Yi TZ i ) and that it is possible to express
the entropy source as a function of any other quantities.
dS int
in a
The reasoning used above leads to an expression for entropy sources
dt
system made up of P sub-systems characterized by I independent extensive
quantities:
I P P dxip , q ¯
dsint
¡(Zi, p Zi, p ) ° pq [1.37]
dt ¡
i1 p 1 q 1 ¢ dt °±
Expression [1.37] only applies to internal entropy creation in the system

considered. Entropy production is also associated with the transfer of extensive
quantities from sources external to the system; this entropy production can be
evaluated in the same way. For thermal transfers between external temperature
sources TpS (p = 1,…,P) and the system, we have, as above, the Clausius inequality:
P dQ pe P dQ pe
¦ d ¦ d dS
p 1 T pS p 1 Tp
1.4.2.5. The average intensive quantity

1.4.2.5.1. Definition
An out-of-equilibrium system is characterized by a collection of intensive
quantities whose values differ according to the sub-systems considered. It may be
useful to characterize the system by a global intensive variable, which is an “average
value” of the intensive variables of the sub-systems. In order to define this average
value, we will refer to an “equivalent” equilibrium state of the system.
Consider an out-of-equilibrium system S made up of P sub-systems Sp each of

which is in instantaneous equilibrium (quasi-static transformations). For each of
these, we can define the intensive entropic quantities Zip associated with their N
extensive quantities Xip. The total amount of extensive quantity Xi contained in the
P
system S is the sum ¦ X ip of the extensive quantities of each sub-system.
p 1
It is clear that the system S cannot be described by any intensive quantity

associated with Xi. We can however associate system S with an average intensive
quantity Ym or Zm at any given instant t, defined as the intensive quantity which the
system S would attain following a natural evolution during which values Xi should
be constant (without any external contribution). Let us consider as an example the
variables Zm.
Suppose that during the transformations undergone by the system, certain

extensive quantities X ip (i 1,..., I ; p) of the sub-systems are exchanged, while
the other N I extensive quantities remain constant in each of the sub-systems (for
example mass, number of moles, volume, etc.). All intensive quantities of all sub-
systems vary during the exchange of extensive quantities. In the final state of the
previously defined system, intensive quantities Zip corresponding to exchanged
extensive quantities have the uniform value Zim for all sub-systems:
In each sub-system p, every intensive quantity Zip corresponding to the extensive

exchanged quantities Xip can be expressed as a function of the N extensive quantities
(I extensive quantities Xip exchanged with other sub-systems and N – I other
extensive quantities Xjp which are constant for the sub-system).
Z ip
Z ip X ip , X jp (i 1,..., I ; j 1,..., N I ; p 1,..., P)
By solving the preceding system of I equations for each sub-system p, with

respect to the extensive quantities Xip, we obtain:
X ip
X ip Z ip , X jp (i 1,..., I ; j 1,..., N I ; p 1,..., P )
The quantities Xi for the system S can be immediately obtained:
¦ X ip Z ip , X jp (i 1,..., I ; j 1,..., N I ; )
P
Xi
p 1
The equilibrium conditions for the system can be written Zip Zim (i 1,...,I; p) Z
pI.
Considering the quantities Xi to be constant, we obtain a system of I equations

from which we can evaluate the I average intensive quantities Zim which
characterize the out-of-equilibrium system:
P
Xi Xip Zim , X jp
(i 1,...I ; j 1,..., N 1) [1.38]
p 1
We laid down that certain extensive variables were not exchanged between the
sub-systems. The problem can be easily discussed in the same manner with different
conditions: for example, by fixing the uniform value to certain intensive quantities
in the whole system (section 1.4.2.5.2, example 2), or by choosing different
conditions according to certain ensembles of sub-systems.
1.4.2.5.2. Examples
We will consider two examples in order to illustrate the preceding procedure,
which we will encounter again in the study of fluid mechanics:
1) A thermal system – consider a system S1 (Figure 1.7a) whose P sub-systems
exchange heat via a constant-volume process, such that the temperature Tp is the
only variable intensive quantity of any of the sub-systems during the process. The
energy Ep of each sub-system can be expressed as a function of its temperature and
its specific heat capacity *p, which is assumed to be constant for the sake of
simplicity:
Ep * pT p
The energy E of the system and its average temperature Tm can be obtained:
P § P · P
E ¦ * pT p ¨ ¦ * ¸T *T m ; with : * ¦ *p
p
¨p 1 ¸ m
p 1 © ¹ p 1
The average temperature of the out-of-equilibrium system can thus be written:
1
Tm ¦ * pT p
* p
In this particular case, the average temperature is the average of the temperatures
weighted by the specific heat capacities.
T1
T2
p1 p2 pp
Tp T1 T2 Tp
(a) System S1 (b) System S2
Figure 1.7. Examples of out-of-equilibrium system:

(a) incompressible thermal system S1; (b) compressible thermal system S2
2) A thermo-compressible system – consider now a system S2 (Figure 1.7b),

which is comprised of sub-systems of variable volumes Vi, containing a perfect
gas and susceptible to exchange heat between each other. The volumes Vi are
separated by pistons which may be subject to fluid friction. The equations of
state for a perfect gas (section 1.3.2.5) give the expression sought for the
extensive quantities:
np RTp
Vp E p * vp Tp (* vp : heat capacity at constant volume)
pp
The global volume and internal energy balances for a natural transformation
allow the average pressure and temperature to be defined:
P P np RTp nRTm P
V Vp ; E * vp Tp * v Tm
p 1 p 1 pp pm p 1
P P
with: n np and: * v * vp
p 1 p 1
We obtain the following expressions for the average temperature and pressure:
1 P 1 1 P n pT p
Tm ¦ *vp T p ¦
*v p 1 pm nTm p 1 pp
The average values of the intensive quantities are no longer simply weighted
arithmetic averages.
1.4.2.5.3. Some comments

1) The “equivalent” system used to define the average intensive quantities is in
thermostatic equilibrium. It thus behaves according to a general state equation and
may therefore constitute a reduced representation of the out-of-equilibrium system,
which is clearly incomplete for a detailed description of the sub-systems. With the
exception of entropy, this is coherent with the laws of thermostatics (we can also say
that this is a “consistent” representation).
2) The average intensive quantities are only weighted arithmetic averages if the
expressions for the extensive quantities Xip are linear functions of Xjp and Yip. The
reader can verify that the state equation for the equivalent complete system
comprises the same linear properties.
3) In the examples of the last section we have considered two systems
constituted of sub-systems with identical properties. The study of more complex
systems, comprising combinations of sub-systems with different structures, can be
effected in the same way.
4) The procedure for definition of average intensive quantities can also be
applied to systems whose intensive quantities may be subjected to certain
constraints. Let us reconsider the example of Figure 1.7a, in which the sub-systems
are all at constant pressure equal to that of the atmosphere, rather than being at
constant volume. We suppose nonetheless that the external walls of the system,
which are in contact with the atmosphere, are adiabatic. The conservation of internal
energy can be written by taking into account the work done by the atmospheric
pressure on the system. The reader can verify that this amounts to writing the
conservation of enthalpy of the sub-systems.
1.4.2.6. Phenomenological laws

1.4.2.6.1. Introduction
In general, all causes of the same tensor nature act on all of the corresponding
effects: we thus have a coupling of irreversible phenomena; the example of a
thermocouple is well known. Phenomena of different tensorial order do not mutually
interact (Curie’s principle). The interested reader should refer to the specialized
literature ([DEG 62], [BYU 02], [PRI 68]).
Consider two sub-systems p and q between which scalar quantities X can be

exchanged. This exchange between the two sub-systems is assumed to be
independent of other sub-systems, which is the case when exchanges only occur via
direct contact, action at a distance not being possible. The irreversible evolution
which occurs takes the form of a flux, dX ip, q dt between the two sub-systems, and
is caused by the existence of an imbalance characterized by the different values Zjp
and Zjq of all the intensive entropic quantities of the two sub-systems (j = 1,..., I).
We have a relation between the causes Zjp and Ziq and the effects (the fluxes)
whose general form can be written as:
dX ip , q

Fi Z jp , Z jq i, j 1,..., I
dt
The principle of action and reaction and the condition of zero flux at thermostatic
equilibrium ( Z jp Z jq Z je ) can be represented by the relations:

Fi Z jp , Z jq
Fi Z jq , Z jp ;
Fi Z je , Z je 0
From this we can deduce a property of the derivatives of Fi which will be used in
the following section:
wFi
Z jp , Z jq
wFi
Z jq , Z jp
wZ jp wZ jq
1.4.2.6.2. Linear thermodynamics

Provided the degree of thermodynamic imbalance is reasonably small, it is
possible to perform a limited Taylor expansion of the functions Fi Z jp , Z jq in the
vicinity of the point Zie , (ie 1,..., I ) , of thermodynamic equilibrium. Taking into
account the above equality we have:

Fi Z jp , Z jq Z jp Z jq wFi Z ie , Z ie oZ jp
Z jq , i, j 1,..., I
wZ jp
wFi
In the preceding relations, Lij Z i 0 Z ie , Z ie is a square matrix of
wZ jp
order I, the properties of which cannot be obtained via thermodynamic
considerations. Only an experiment or a suitably adapted theory can provide these
laws, which we here qualify as phenomenological laws.
Statistical reasoning shows that this matrix is often symmetric (Onsager

relations). This symmetry is no longer maintained if other intensive variables are
used in place of intensive entropic variables.
As long as the preceding approximation is valid, we say that we are dealing with
linear thermodynamics of irreversible phenomena. We thus have, for each
thermodynamic flux of an extensive quantity:
dX ip , q

Lij Z ie Z jp Z jq [1.39]
dt
Internal entropy source [1.36], which exists at the interface of the two sub-
systems considered, becomes:
¦ ¦ Lij Z e Z i1 Z i 2 Z j1 Z j 2
dSint [1.40]
i, j 1,2,....n
dt i j
This internal entropy source is thus a positive quadratic form whose principal
diagonal elements are all positive ( Lii Z e ! 0 ).
We can also perform a change of variables which involves expressing the

intensive entropic variables as a function of other thermodynamic variables, for
example the energetic intensive variables. Consider two sub-systems 1 and 2. The
differences Z i1 Z i 2 are generally linear functions (to second order excepted) of
the differences between the new variables which are chosen.
For example, with the intensive entropic variable 1 T which is associated with
energy, we obtain:
1 1 T2 T1
[1.41]
T1 T2 T1T 2
and, generally, by using intensive energy variables Yi TZ i in the place of

intensive entropic variables Zi:
Yi1 Y i2 T1 Yi1
Z i1 Z i2 Yi1 Yi2 T1 T2 [1.42]
T1 T2 T1T 2 T1T 2
that is, to second order excepted, by linearizing in the vicinity of ( T0 , Yi 0 ):
1 1 T1 T 2 1 Yi0
; Z i1 Z i 2 Yi1 Yi2 T1 T2 [1.43]
T1 T2 T 02 T0 T02
By substituting expression [1.43] into expression [1.39] for the thermodynamic

flux in which the temperature appears, we obtain a linearized expression for the flux
as a function of the intensive energy variables (including the temperature):
dX i1 int

Lijc T0 , Yi 0 Y j1 Y j 2
dt
in which the matrix Lijc can be easily deduced from the matrix Lij. It is easy to see
that the matrix Lijc is not symmetric, as noted earlier.
In addition, the terms of the matrix diagonal Lijc are negative: the flux of the
extensive quantities is in the same direction as the decaying intensive energy
quantities.
dQ1, 2
Letting i = j = 1, the heat flux and the temperature T, we define the
dt c of the first diagonal term
thermal resistance RT which is the inverse of the term L11
of the matrix Lijc :
dQ1, 2 T2 T1
dt RT
The entropy source at the interface of the two sub-systems can be written using
the energy variables (accurate to second order excepted):
¦ ¦ L' ij Y j Yi1 Yi 2 Y j1 Y j 2
dS int
i, j 1,2,....I
dt i j
1.4.2.6.3. Conditions for the application of irreversible linear thermodynamics

We have already noted that it is possible to linearize the phenomenological law
which gives the flux of extensive variables if the imbalances are “small”. This idea
can be stated in the following usual way: if the imbalance corresponds to a weak
variation of a “regular” process, then the linearization may be satisfactory. This
leads us to invoke irreversible extensive quantity transfer mechanisms. Let us
examine the case of thermal transfers, which are primarily due to two mechanisms.
1) Intermolecular action within a material
Molecular agitation, the intensity of which increases with temperature, results in
the transmission of extensive properties via collisions between molecules, ions, etc.
This is a statistical mechanism, which tends to cause a uniform distribution of the
properties of a body. For example, the mechanical energy of molecules in hot
regions is transmitted to molecules in cold regions via collisions between the
molecules (gases), and/or by the action of intermolecular forces (liquids, solids).
Within the context of kinetic theory in traditional mechanics, molecules are
animated with a velocity in the order of the speed of sound; they cover a distance
called the mean free path between successive collisions. Under ordinary conditions
of pressure and temperature, this distance is in the order of 10-7 meters. Thermal
energy is due to kinetic and potential energy of molecules. Let us take an example to
evaluate the imbalance due to a temperature gradient. If we admit that statistically
molecules lose one-thousandth of their energy with each collision, we can conclude
that about 100 collisions are necessary in order for gas molecules to lose one-tenth
of their energy. This loss corresponds to a temperature drop of about 30 Kelvin
which will be produced over a distance of the order of 10-5 meters (10 Pm). This
corresponds to a considerable thermal imbalance. However, these collisions, which
correspond to a tiny mean energy loss of 1/1,000, are clearly very small processes in
comparison to two microscopic fluxes of mechanical energy due to the molecules
going through any plane in one direction and also in the opposite direction. These
opposite fluxes have nearly the same absolute value. The macroscopic mechanism
for irreversible transfer of extensive quantities by molecular collision is thus
statistically a tiny perturbation amongst the mechanisms of thermal agitation, and it
is thus not surprising that the macroscopic processes are linear. Our experience
verifies the arguments proposed by this rather simplistic reasoning.
However, we must realize that the collision properties can vary with temperature,
even if the mean free path is not very temperature sensitive. This implies that the
properties of thermal resistance can depend on the temperature chosen T0 in relation
[1.43], in order to evaluate the thermal resistance (the inverse of the first diagonal
term of the matrix L'ij ). In fact, for large temperature differences, it is necessary to
consider thermal resistance as a succession of isothermic thermal resistances in
series and to perform an integration which takes the temperature variation into
account:
dQ1, 2 T2 dT
³T
dt 1 R T T
As a first approximation, we can often take the temperature T0 as being equal to

T1 T2
the mean temperature .
2
2) Thermal radiation
Thermal radiation is in fact an electromagnetic radiation, whose emitted power
qT per unit surface (thermal flux density) of a blackbody can be represented by
Stefan’s law:
qT T 4 (ı: Stefan constant;T Kelvin)
Consider two parallel planes, face to face and respectively heated to

temperatures of T1 and T2 (T1 < T2). The net heat flux density qT received by plane 1
is equal to:
qT
V T24 T14
The preceding relation will quickly deviate from a linear heat transfer law valid
for small temperature differences. The reader can verify that in this case the
preceding law can be represented by a thermal resistance RT equal to 1 3 V T 3 ,
which varies strongly with temperature.
1.4.3. Application to heat engines
A heat engine is a device in which a fluid is made to evolve according to a cycle

C: the material follows an evolution parameterized by the time t, after which the
final state is identical to the initial state. Diverse extensive quantities are exchanged
by this material with external sources of work and of heat. The objective of this
machine is to produce certain desired quantities (work for a heat engine, heat for a
heat pump, etc.) from external sources of other quantities.
The fluid which circulates in the engine passes through successive devices which
either furnish or extract work and heat. Consider a mass m of this fluid whose
quantities and in particular the temperature T(t) evolve in a quasi-static manner as a
time function t. Let GQ be an amount of heat which it receives between the instants t
and t+Gt. During the cycle C, the variations of the fluid’s entropy and of its internal
energy are zero:
³C dE ³C G W G Q W Q 0; ³C dS 0 [1.44]
Thus, the evolution process of the fluid over a cycle, as for all processes, creates
entropy (see section 1.4.2.4). As the entropy variation of the fluid is zero over the
cycle, a generalized version of relation [1.33] can be written, T being the
temperature of the fluid:
dQ
³C d0
T
On the other hand, the heat transfer from the sources at temperature TS(t) is also
accompanied by a creation of entropy, the reasoning of relation shows that we have
the Clausius inequalities:
dQ dQ
³C d ³C d ³C dS 0
TS T
The preceding reasoning has the advantage of highlighting the entropy sources
associated with the Clausius inequality; it also shows that the difference between the
inequality terms is greater in proportion to the level of irreversibility.
We will not get into a detailed discussion of such cycles and the efficiency of
heat engines, all of which can be deduced from the aforementioned inequalities. The
reader will find such discussions in texts on applied thermodynamics.
Chapter 2
Thermodynamics of Continuous Media
The properties of continuous media can be obtained by a limiting process on

variables of discrete systems. The exchange of extensive quantities is modeled by
means of flux densities. Irreversible thermodynamics can be transposed in the same
way, being represented by diffusion equations expressed in terms of intensive
quantities (heat and diffusion equations of chemical species). The principal results of
fluid statics are presented. The diffusion of material leads to the existence of several
macroscopic reference frames, which brings with it specific difficulties which are
discussed.
2.1. Thermostatics of continuous media
2.1.1. Reduced extensive quantities
The continuous medium is defined according to the usual method of passing

from the discrete to the continuous by letting the elementary sub-systems tend
towards zero, their number thereby increasing indefinitely. If the geometric
dimensions of a system tend toward zero, the extensive quantities of that system also
tend toward zero, whereas the intensive quantities do not change. A limiting process
is therefore necessary for the study of continuous media.
We define reduced extensive quantities, i.e. extensive quantities per unit of mass,
volume, or number of moles:
G
– a quantity per unit mass g ,
m
G
– a volume quantity g ,
V
G
– a molar quantity g~ .
N
The amount of G which is contained in the domain D is:
~
G ³D U gdv ³D gdv ³D ngdv
The concept of a local quantity corresponds to an average over a volume which

is large on the microscopic scale, but small on the macroscopic scale. If we imagine
that it is possible to measure an average quantity for a set of particles (molecules, for
example) of a continuous medium, contained within a sphere of radius r and
centered on a point M, the values obtained will only tend towards the value of the
reduced extensive quantity if r is sufficiently large. For small values of r, noticeable
fluctuations would be observed. Figure 2.1 shows the result of a measurement of the
average density which would be obtained for a sphere whose radius r is of the order
of the inter-molecular distances. On a larger scale we would of course observe the
gradient (macroscopic) of the reduced extensive quantity.
r
M
U m M , r
U M
r
Figure 2.1. Local value of the average specific mass in a

sphere of radius r and fluctuations at a molecular scale
2.1.2. Local thermodynamic equilibrium
The description of a given continuous medium can be performed by means of a

field of reduced extensive quantities. We use the same methodology used for finite
discrete systems, as the reduced extensive quantities chosen only allow for a
description of the continuous medium if a hypothesis of local thermodynamic
equilibrium is made: any very small volume obeys the general equation of state for a
system in equilibrium, with all its consequences. We obtain the thermostatic
Thermodynamics of Continuous Media 49
relations for continuous media by a limiting process when the characteristic

dimension of the discrete sub-systems tends to zero.
General equilibrium equation [1.6] for a discrete system is first degree

homogenous; it can be applied immediately for reduced extensive quantities. By
performing a trivial transposition of the reduced notation (Xox, Eoe, Sos, etc.),
the notation S which here designates the mathematical function of formula [1.6], we
have:
s S (x, e) or: s S (x , e ) or: s S (x , e ) [2.1]
The consequences of the hypothesis of local thermostatic equilibrium have

already been described in Chapter 1; the diverse relations obtained being first degree
homogenous relations, they are entirely transposable via a replacement of the
extensive quantities by the corresponding reduced extensive quantities. We note
however that the variance has decreased by one unit. We will often consider a
divariant fluid (two independent state variables) which will often be a perfect gas
with constant Cv, for which the equations of state become:
p
rT e C v T const
U
The local thermodynamic equilibrium hypothesis translates the existence of two

rapid dynamic processes which are opposed, such that locally there is a balanced
exchange between them: in effect, the energy mechanism and momentum
transmission on the molecular scale involves molecular collisions during which the
exchange of extensive quantities occurs in both directions. The macroscopic transfer
is no more than the residuum of these exchanges in opposing directions. Molecular
displacements occur with a velocity in the order of the molecular agitation velocity
(in the order of the speed of sound). The corresponding fluxes are individually very
high, but they are in opposing directions and it is their very weak net outcome that
we observe at the macroscopic scale for irreversible phenomena. These amount to a
very weak perturbation of the local equilibrium of the continuous medium, which
obeys thermostatic relations.
The local thermodynamic equilibrium hypothesis is based on the fact that the
time required for the local gas equilibrium to be achieved (relaxation time) is small
compared with the times associated with the macroscopic gas evolution. For the
molecules of mean free path A and molecular velocity c, the relaxation time is in the
order of A/c, if we consider that each molecular collision is efficient in the transfer of
extensive molecular quantities (quantum conditions).
The molecular interactions which are at the heart of the aforementioned

processes have a very short radius of action, whether we consider the intermolecular
forces or the collisions characterized by their mean free paths. These actions are in
fact volumetric at the scale of the given distances, and at the macroscopic scale they
appear as contacts, i.e. flux densities of extensive quantities, which are determined
by local conditions, i.e. surface forces.
The condition of local thermodynamic equilibrium is not always satisfied in the

case of certain gases undergoing rapid changes (for example, tri-atomic gases during
the passage of a series of shock waves, or in supersonic nozzles) and plasmas. We
are therefore led to separate different populations (ions and electrons) or different
forms of molecular energy (translation and vibration), for which we must introduce
a further extensive quantity and supplementary hypotheses (relaxation law, etc.).
Such separations into sub-systems are not only spatial. The interested reader can
refer to [BAS 98] or texts dealing with ultrasound.
This condition of local thermodynamic equilibrium does not allow the entire
medium considered to be in global thermodynamic equilibrium. It can be shown
using statistical methods taken from kinetic gas theory, for example, that the net
balance of the extensive molecular quantities gives, at first order, a distribution of
Maxwell-Boltzmann velocities which correspond to a local statistic mechanical
equilibrium ([CHA 91], [HIR 64]). At second order we have phenomena associated
with the irreversible transfer of extensive quantities ([BIR 02]); in section 3.4.1.3 we
will discuss the mechanism of this irreversible transfer in the case of momentum
transfer. The statistical equations describing turbulence are identical to the preceding
molecular statistical transfer equations, in which turbulent fluctuations play the role
of molecular fluctuations for the transfer of these extensive quantities.
Unfortunately, turbulent fluctuations do not verify general statistical laws and the
preceding analogy is only formal ([COU 89], [MAT 00], [TEN 72]).
2.1.3. Flux of extensive quantities
2.3.1.1. Flux density

An extensive quantity G can be transferred in a continuous medium by different
processes, the nature of which does not interest us here. The definition of a flux
M GS of the quantity G across the surface S has already been given for the surface of
finite systems (the amount of the quantity G which crosses the surface S per unit
time).
We will show that in a continuous medium, the flux of an extensive scalar

quantity G is characterized by a flux density vector.
Consider the tetrahedron OA1A2A3 whose corners are the origin O and three
points lying on the axes (Figure 2.2). Suppose that the surface dimensions of this
tetrahedron are small, of order H. Let q G1 ds1, q G 2 ds2, qG3 ds3 be, respectively,
the fluxes of quantity G across the surfaces OA2A3, OA1A3 and OA1A2 (of
respective surface areas ds1, ds2 and ds3, of order H2) in the positive direction along
the coordinate axes.
Let dM G be the flux leaving the face A1A2A3 of area ds of the tetrahedron.
Suppose that the (algebraic) sources of the quantity G are volumetric. The amount of
the quantity created in the tetrahedron is O(H3). The balance of the extensive
quantity G leaving the tetrahedron can be written:
dM G q G1ds1 q G 2 ds 2 q G 3 ds 3 O (H 3 )
x3 A3
n
A2 x2
A1
x11
Figure 2.2. Quantities balance on an elementary tetrahedron
We obtain:
ds i D i ds (i 1,2,3)
G
where Di designates the directional cosines of the normal n oriented towards the
exterior of the tetrahedron.
By replacing the surfaces dsi with their previous expressions and by letting H
G
towards 0, the terms of order H2 should cancel out. By defining vector qG or the flux
density of the quantity G, with components q G1 , q G 2 , q G 3 , we obtain the
elementary flux of the quantity G across ds (the quantity of G crossing ds per unit
G
time in the direction normal n ):
G G
dM G q GiD i ds q G .n ds [2.2]
The flux M G6 of the quantity G across the surface S can be written:
G G
M G6 ³S q G .n ds [2.3]
By a similar argument, it can be shown that the transfer of a vector quantity Gi

would be characterized by a tensor q Gij whose flux across an elementary surface ds
would give the vector quantity Gi which crossed ds per unit time:
G
dM Gi q Gij n j ds q G .n ds [2.4]
where the tensor q G is defined, as before, using the vector fluxes across the co-
ordinate planes.
2.1.3.2. Examples of flux of extensive quantities

The preceding property did not require any hypothesis regarding the transfer
mechanisms of the quantity G (action by contact or action at a distance): it results
solely from the notion of balance for an extensive quantity. As an example, we can
consider the flux of extensive quantities due either to propagative phenomena
(example 1) or phenomena involving action by direct contact (examples 2 and 4):
1) Energy transfer by electromagnetic radiation (energy-flux density vector
in lighting, infra-red heating, etc.) or by acoustic propagation (acoustic intensity
vector).
2) Heat transfer in a material (molecular agitation energy at the molecular
scale) is represented by the thermal flux density vector (section 2.1.5.3.1).
3) The diffusion of a chemical species in a material medium is represented
by the molar flux density vector of the species considered (n: the number of
moles/volume) (section 2.4.2).
4) For example, if the extensive quantity is a force density (or momentum
density at the microscopic scale), the corresponding flux density is the stress
tensor V ij . However, the convention for the orientation of the normal is here
reversed. The elementary force is the force exerted on the surface ds, by the
material situated on the side of the normal (side 1 in Figure 2.3a), towards the
other side:
df i V ij n j ds [2.5]
exterior V33
73
n 1 V12
V32
V31
ds V23 V
2 V13 72 V
interior 71
G G x3 V12 V22
df 1o 2 Tds V21
V
V11
O
(a) x2 (b) (c)
x1
Figure 2.3. Definition and symmetry of the stress tensor
The force density at the point M on the surface ds is the stress T :
df G
T (M , t ) V .n
ds
Figure 2.3b shows the stresses on the faces of a parallelepiped whose corners are
parallel to the axes. The reader can easily verify that in the absence of volume sources of
torques, the equality of the moments about the axis Ox3 of the stresses exerted on the
four faces parallel to the axis Ox3 (see Figure 2.3c) leads to V 12 V 21 . The same goes
for the other components of the stress tensor ( V ij V ji ).
2.1.3.3. Volume source equivalent to the fluxes

The balances in a material domain are clearly always effected on closed surfaces
which comprise the boundary. By applying the Ostrogradski theorem the flux M G6 of
the quantity G leaving a closed surface 6 can be written:
G G § wq Gi ·
M G6 ³6 q G .n ds ³D div q G dv ¨ ³ dv ¸ [2.6]
¨ D wx ¸
© i ¹
This flux can thus be written in the form of a volume integral, which implies
that the total flux of the quantity G on a closed surface is equivalent to the
action of the volume source div q G of the quantity G.
In the case of stresses in a continuous medium, a force volume source

generated by a stress is equal to the vector div V :
G
³6 T (M , t ) ds ³6 V .n ds ³D divV dv
Reciprocally, all volume sources which are mathematically expressed by means of a
G
divergence operator ( div q for instance) can be interpreted as a transfer by the flux
G
across a closed surface of vector flux density q .
2.1.4. Balance equations in continuous media
2.1.4.1. Balance equation of an extensive quantity
G
x3 n
D
O
x2 6
x1
Figure 2.4. Balance of an extensive quantity
The balance equation of a volumetric extensive quantity g consists of

writing that the variation of this quantity in a material domain D is due to
contributions from outside, and which have thus crossed the closed surface 6
which constitutes the boundary of D (the normal is directed outwards), and to
volume sources of density VG in D:
wg
³D dv ³D V G dv ³¦ q Gj n j ds [2.7]
wt
Supposing this relation to be true regardless of the domain D, we can deduce the
local equation:
wg wqGj
VG [2.8]
wt wx j
For a fixed medium, which can thus not be deformed, mass1 must be strictly
conserved, and we have:
wU
0
wt
Balance equation [2.8] for the quantity G can also be written, using the mass
quantity g:
wg wq Gj
U VG [2.9]
wt wx j
For a fixed continuous medium, the principal extensive quantities are internal
energy (in its thermal form, and in the absence of possible physicochemical
reactions), volume and the number of moles of a chemical species. Electric
quantities may also be manifested; however, we will not deal with such questions in
this work.
For a material of constant specific heat C, in the absence of chemical reactions,

the variation of volumetric internal energy de is equal to UCdT, and the balance
equation can be written:
wT wqTj
UC VT [2.10]
wt wx j
Heat-source volume VT is in fact a residuum of another form of energy which is

present in the mass and which is degraded in the form of heat (Joule effect caused by
the passage of an electric current (Ohm’s law), absorption of electromagnetic
radiation, etc.). This result could be obtained in a general way, but such a discussion
is beyond the scope of this work. We will come back to this question in Chapters 3
and 4 when it comes to dealing with mechanical energy.
For a chemical species comprising a number of moles ni per unit volume (molar
concentration), the balance equation can be written:
wni wq n ij
U V ni [2.11]
wt wx j
1 Excepting diffusion processes, for which we only give volumic balances in this chapter
(section 2.4.4.2.2).
The molar flux densities qni and the sources Vni of chemical species are due,
respectively, to diffusion phenomena and chemical reactions.
2.1.4.2. Entropy source

During the natural evolution of a system, entropy is not conserved and it can only
increase. We have already described the entropy-creation mechanism in the case of two
discrete sub-systems that are in contact and have different temperatures (section 1.4.2.3)
and (section 1.4.2.4). The zone where entropy is created was localized in the contact
zone between the two sub-systems. Here the distribution of entropic variables is
continuous, and entropy creation will be diffused and associated with the existence of
gradients with entropic variables.
On account of the local equilibrium hypothesis, relation [1.13] can be written, for the
per mass unit quantities:
ws U we w gi
U U Zi [2.12]
wt T wt wt
wg i
By replacing the derivatives U in equation [2.12] by expressions obtained using
wt
balance equations [2.9], the following relation can be obtained:
ws U we § wq ·
U Z k ¨ V Gk Gki ¸
wt T wt ¨ wxi ¸¹
© [2.13]
U we w>Z k q Gki @ wZ k
Z k V Gk q Gki
T wt wxi wx i
in which the surface flux terms associated with the divergence operator have been
separated from those associated with the volume source as defined in section 2.1.3.3.
The entropy is associated with the extensive quantities and the corresponding entropy
addition are represented by two terms:
– Z k V Gk V S external entropy source associated with the source V Gk of the
extensive quantity Gk;
– Z k q Gki Z S entropy flux associated with the flux of extensive quantities.
The thermodynamic imbalance is characterized by the gradients grad Z i , which

wZ k
we designate under the label thermodynamic forces. The additional term q Gki
wx i
of equation [2.13] is the entropy source associated with the local macroscopic
imbalance. It is always positive. Its expression is analogous to the expression for the
dX 1,2
creation of entropy Z1 Z 2 from formula [1.31].
dt
2.1.5. Phenomenological laws
G
An irreversible evolution is characterized by flux density fields of the quantity q Gk
in a continuous medium. These thermodynamic fluxes are associated with gradients of
intensive entropic quantities Zi. The transfer of a quantity Gi can occur via relatively
different kinds of processes:
1) Molecular agitation leads to an exchange of extensive quantities between the
particles involved. These intermolecular actions occur from place to place, because it is
these microscopic particles themselves which transport the extensive quantity considered
(mass, momentum, energy, chemical species, etc.). The interaction zone between two
material domains D1 and D2 separated by the surface S (Figure 2.5) is limited to a
thickness in the order of 2d (d: intermolecular distance in liquids or mean free path in
gases). This extremely thin zone is modeled on the macroscopic scale by the surface S,
on which we can consider contact actions.
S
D2
D1
2d
Figure 2.5. Interaction zone between material domains D1 and D2
2) On the macroscopic scale, and for turbulent flows, we observe chaotic

velocity fluctuations which lead to fluxes of extensive quantities convected by the
fluid. The effective interaction zone between two material domains is no longer
limited to a surface as before. The flux depends on the structure of the whole
turbulent zone. Considerable difficulties result from this situation in which the cause
of the flux of an extensive quantity can no longer be modeled with a general local
phenomenological law ([COU 89], [MAT 00], [TEN 72]).
3) In other situations, the flux of an extensive quantity (essentially energy) is due

to the presence of an electromagnetic field. This is the case for energy transfer by
thermal radiation in semi-transparent media, which both emit and absorb at all
points and whose local state results from the emission balance in a macroscopic
volume surrounding the point considered. Here again, we can no longer localize the
cause of the extensive quantity flux on a single surface.
2.1.5.2. Contact actions and thermodynamic forces

The interaction zone between two material domains D1 and D2 is modeled by the
surface S. The thermodynamic forces, represented by grad Z i , are the cause of
thermodynamic fluxes.
As with discrete systems (section 1.4.2.6) all causes of the same tensorial nature
act on all the corresponding effects and we have a coupling of irreversible
phenomena: for example, a temperature gradient leads to a material flux (thermal
diffusion). Phenomena of different tensorial orders do not interact.
A rudimentary explanation of these facts can be provided from context of kinetic

gas theory. A gas is a set of molecules which are subjected to a thermal agitation.
Irreversible phenomena are the macroscopic result of this spontaneous action.
Molecules with different properties (mass, type, kinetic energy, etc.) do not respond
in the same way to non-symmetries in the mean properties of the medium. A
molecular concentration of a given species will be progressively diluted in the rest
of the gas; a temperature gradient (gradient of the molecular kinetic energy) will not
act in the same way on different species of molecules and so may create a
concentration gradient. For example, at equal energy, we notice that smaller, and
therefore faster, molecules can slip in a gas comprising larger molecules, hence the
phenomenon of thermal diffusion. On the other hand, it is difficult to see how the
static scalar properties of a gas which is macroscopically at rest can spontaneously
generate a vector macroscopic momentum (i.e. a bulk movement) in the absence of
an external influence.
There thus exists a relation between thermodynamic forces and thermodynamic

fluxes of the same tensorial rank. Since in the absence of thermodynamic forces, the
thermodynamic fluxes are zero, the general form of this relation can be written as:
G JJG JJJJJG JJG G G G
q Gk F k gradZl
(k, l 1,..., K ) with: F i 0, 0,...
0 [2.14]
On account of the principle of action and reaction, the function F k is odd

( F k grad Z l
F k grad Z l ). Relation [2.14] must verify properties of
homogeneity and of spatial and material isotropy. In what follows, we will limit our
discussion to cases where the difference from thermodynamic equilibrium is
relatively small, so that we can justify a first order Taylor expansion of the functions
G

Fk grad Z l :
Fk grad Z
l L kl grad Z l
For a set of K extensive quantities, Lkl is a square matrix of order K whose

terms are functions of the values of the K intensive quantities Zk at the point Zke,
about which linearization is effected. When the preceding approximation is valid,
we consider that we are dealing with linear thermodynamics of irreversible
phenomena.
If the medium is considered to be isotropic, the matrix is reduced to a matrix of

dimension K (see section 2.1.5.3.1).
Thermodynamics does not provide access to any properties associated with the
matrix Lkl . However, by means of statistical reasoning the matrix can be shown to
be symmetric (Onsager’s relations). This symmetry is only verified if the entropic
variables Zk are used to define this matrix.
In practice, the local imbalance is characterized using simpler variables than the
intensive entropic variables Zk. For example, we use temperature in place of the
entropic intensive variable 1 T (section 1.3.1.1). Expressions for linearized
thermodynamic force are equivalent to first order, but the matrix coefficients Lij are
modified, such that the symmetric properties generally disappear. An analogous
situation has been observed in the case of discontinuous media (section 1.4.2.6.2).
2.1.5.3. Some simplifying laws for irreversible transfer

2.1.5.3.1. Fourier’s law and thermal conduction
Consider firstly a case where the medium comprises a pure body, such that
thermal transfer occurs alone, without any coupling with diffusion or electrical
conduction. The relation between the thermal flux density and the thermal gradient
can be written in the context of linear thermodynamics (Fourier’s law) as:
G JJJJJG sT
q T Ȝ . gradT or qTi Ȝij
sxj
The principal axes of the 3*3 tensor Oij are the symmetric axes of the medium.
If the medium is homogenous (fluids, non-crystalline solids, etc.), the three
eigenvalues of the tensor Oij are equal and this one is spherical ( O ij OG ij ); it can
thus be expressed solely as a function of the thermal conductivity O of the medium.
Fourier’s law can be written as:
wT
qTi O qT O gradT [2.15]
wx i
The thermal conductivity of the medium O is expressed in Watts/meter.K.
Heat is the macroscopic form of mechanical energy related to thermal agitation

of a medium. Its transfer in the medium is due to interactions between the
microscopic entities (molecules, atoms, ions, electrons). The values of the
coefficient O depends on the nature of the medium.
Metallic media are excellent thermal conductors, thermal conduction being

principally assured by electrons which have a high mobility. The values of the
coefficient O lie in a range spanning from a few tens to a few hundred W/m.K.
Solid crystalline media are generally good conductors: as the crystalline structure
presents a reasonably strong coherence on account of its organization, energy can be
transmitted in a vibrational form (phonons). The coefficient values O are in the order
of a few W/m.K.
Solid amorphous media or composites have weaker conductivity: well under 1

W/m.K for fibrous materials.
Liquids, comprising a looser structure, are poorer conductors of heat than solids
(of the order of 0.1 to 0.2 W/m.K): intermolecular forces here assure conduction.
Energy exchange in gases only occurs by means of molecular collisions in the

gaseous medium; the corresponding values of the coefficient O are of the order of
0.01 to 0.02 W/m.K. Insulating materials are constituted of a matrix (fibers, wools,
foams, etc.) which is as light as possible, thermal insulation being assured by the
gaseous interstices.
The thermal flux across a surface S can thus be written as:
G G G wT
MT ³S qT .nds ³S O gradT .nds ³S O .ds
wn
We can also define the thermal diffusivity a:

O
a
UC
The reader can easily verify that this quantity a can be expressed in m2/sec; we
will see the important role that this quantity plays in the heat equation (section
2.3.1).
The thermal diffusivity takes on the following values at 20ºC:

– air: 0.19 × 10-4 m2. s-1;
– water: 1.4 × 10-7 m2. s-1;
– metals: ~ 10-4 m2. s-1.
2.1.5.3.2. Fick’s law and the diffusion of chemical species

As the extensive variable is here the number of molecules of a chemical species,
P
the corresponding intensive entropic variable is (where μ is the chemical
T
potential (section 1.3.1.1)). In ideal solutions or perfect gases, we replace the
intensive entropic variables with a molar or mass concentration variable. This
simplified approach will suffice here for an exposition and discussion of the
phenomena which we will examine. It will allow us to outline the principal
difficulties associated with fluid movement. The additional complications which
arise when we take account of more complex thermodynamic and chemical
properties are beyond the scope of this work.
The presence of a concentration gradient leads to the existence of a molecular

flux towards regions of weaker concentration. Taking the molecular concentration
n1 (the number of moles per unit volume) of a component, Fick’s law, which
characterizes the diffusion of the species considered, can be written:
qn D grad n1 [2.16]
1
where D is the diffusion coefficient.
Diffusion is a complex phenomenon, as it implies that the material is not

immobile, and in such cases it is not possible to effect a generalized reasoning
using a single constituent. We will come back to this point at a later stage
(section 2.4), and we will specify the conditions under which law [2.16] is valid.
2.1.5.3.3. Electrical conduction and Ohm’s law

Electric conduction presents certain analogies with thermal conduction and
diffusion; however, the underlying physics is a little different. Here we are dealing
with an electric field derived from an electric potential Vel which exerts a force on
mobile electric charge carriers. Ohm’s law has the same form as the preceding laws:
j V el grad V el
G
where j is the electric current density and V el is the electrical conductivity.
Thermal molecular agitation, which grows with temperature, slows down the
movement of electric charge-carriers because of collisions; thus, electrical
conductivity is a decreasing function of temperature, whereas for other irreversible
phenomena, thermal agitation is the driving factor of thermal and diffusional fluxes.
2.1.5.3.4. General case: coupled transfer between diffusion and thermal

conduction
In general, the flux of a scalar quantity (energy, chemical species) depends on all
JJJJJG JJJJJG JJJJJG
the local thermodynamic forces grad T , grad ni , grad p, etc., associated with these
scalar quantities.
We can schematically write under certain conditions the following relations for
the diffusion of a constituent of binary mixing in the presence of heat transfer:
qT O gradT K grad n1
gradT
q n1 D1T D grad n1
T
The coefficient D1T characterizes the thermal diffusion (Soret effect), i.e. the
existence of a flux of chemical species which is associated with a temperature
gradient. The symmetric effect of a heat flux caused by a concentration gradient
(Dufour effect) and characterized by the coefficient K is generally much weaker.
From a physical point of view, the problem is more complex than this, as each
chemical species introduces its own thermal energy (enthalpy at constant pressure,
etc.). Definitions of the preceding coefficients may vary, and we will not provide a
complete discussion of these phenomena, as this would require lengthy discussions
concerning chemical thermodynamics ([BIR 01], [DEG 62], [PRI 68]).
2.2. Fluid statics
2.2.1. General equations of fluid statics
2.2.1.1. Stresses in a fluid at rest

We designate as a fluid any body which is in either a liquid or a gaseous state. In
contrast to solids, fluids do not have any intrinsic shape. They adapt to the shape of
the container in which they are found. On the microscopic scale, there is no longer a
crystalline structure, and the molecules, ions, etc., interact by means of
intermolecular forces and collisions. On the macroscopic scale, fluids generally have
isotropic properties; the forces in a fluid at rest are thus represented at each point by
a spherical stress tensor pGij which depends on a scalar quantity p, designating
pressure. The force dfi exerted on an elementary surface ds by the material situated
on the same side as the positive normal is (Figure 2.6a):
G G
df i pG ij n j ds pn i ds df p n ds
2.2.1.2. Conditions for the existence of equilibrium in a fluid

A fluid in equilibrium is only subjected to pressure forces and to external forces
of mass density gi. By considering the net force balance on a domain D, we obtain:
G G
³6 p.n ds ³D U gdv 0 [2.17]
G
n
n p2
ds
x3 D
C'
-p.n ds C
O p1
6 R'
x2 R
x1
(a) (b)
Figure 2.6. (a) Pressure force on a surface; (b) Laplace’s law
G
Using the identity ³6 f .n ds ³D grad f .dv , we obtain:
wp
U gi 0 [2.18]
wx i
Equation [2.18] can only be solved if the vector Ugi can be derived from a
potential, in other words if the following condition is satisfied:
G G G
rot U g U rot g grad U g 0
We can see that for a fluid of constant density, an equilibrium situation can only
G
exist if the force field g grad U derives from a potential.
Supposing that the force field gi can be derived from a potential U

G
(g grad U ) the preceding condition can be written:
grad U grad U 0
This leads to the fact that surfaces of U=const. and U=const. are identical.
Equation [2.18] then becomes:
grad p U gradU 0 [2.19]
This also leads to the identity of isobaric and equipotential surfaces that are
identical. If the fluid is divariant, with an equation of state in the form p pU , T ,
the said surfaces are also isothermal.
For example, in the case of gravitational forces, equilibrium is only possible if

the horizontal surfaces are isothermal. If this condition is not satisfied, natural
convection will occur.
The resultant pressure force on the exterior of the closed surface 6 can be
immediately deduced from relation [2.17]:
G G
³6 p.nds ³D U gdv
G
This force is the opposite of the sum of the forces exerted by the force field g
on the fluid contained in D.
Let us now consider the case of gravity. As with equation [2.17], the balance
moment of forces on domain D – where 3 is the center of gravity of this domain D
filled with the fluid – can be written:
G G G
³6 3 M p.n ds ³D 3M U gdv g ³D U 3 M dv 0
which is zero from the definition of the center of gravity.
So, the collection of the pressure forces is equivalent to a single force, equal to
the weight of fluid displaced and applied at the center of gravity 3of the domain
supposed to be filled with the fluid. The point 3 is called the center of buoyancy.
This result constitutes Archimedes’ theorem.
Now, considering a vertical cylinder with identical horizontal bases, we see that
the difference in pressure forces between the lower and the upper bases is equal to
the weight of fluid contained in the cylinder.
2.2.1.3. Solution to the general hydrostatic equation

If the preceding conditions are satisfied, the pressure satisfies equation [2.19].
The conditions for the existence of the solution (stratification of the space into
identical surfaces U=const, p=const, U=const, etc.) for a fluid whose equation of
state is p pU , T leads, under equilibrium conditions, to the relation U f p
between p and U. We often say that such a fluid is barotropic (which is not strictly
true, as this is not a property of the fluid, but of the configuration studied).
dp
We can thus define the function h( p) ³ such that:
U
dp § grad p ·¸
dh ¨ grad h
U ¨ U ¸¹
©
If the specific fluid entropy is constant over the entire domain studied, it is said
to be homoentropic, and the function h is therefore the specific fluid enthalpy.
General equation [2.19] can be written as:
grad h U 0
and its solution is: h U const . The pressure p can thus be obtained by inverting
the function h(p).
In the case of gravity, the function U is equal to gz, where z designates an

increasing vertical coordinate.
2.2.1.4. Equilibrium of an incompressible fluid

2.2.1.4.1. Hydrostatic equations
Such a fluid is generally either a liquid or a gaseous domain of limited vertical
extent, such that its density can be considered constant. We thus consider density to
be independent of pressure, which may eventually be a function of the coordinate z.
If the fluid is considered to be of uniform density U0, we have:
p U 0U pg const (U gz for gravity potential)
So, it is this quantity p U 0U (in the case of gravity p U 0 gz ) and not the
pressure alone which is constant in a motionless fluid. Hence, variations of this
quantity must be the causes of movement. For this reason, we will call it “driving
pressure” and we will note it as p g p U 0 gz in the case of gravity.
In fact, for such an incompressible homogenous fluid, we will find (see Chapter
4) that, under certain conditions, movement is associated with variations of this
driving pressure alone, without explicit intervention from pressure.
EXERCISES –
1) Equilibrium of a homogenous liquid in a rotating reference frame of axis Oz.
Give the expression for pressure as a function of the coordinates x and y and z. Show
that in a reference frame which rotates with angular velocity Z about the vertical
axis Oz, the isobar surfaces are paraboloids of revolution about the axis Oz.
1
(Answer: p U Z 2 ( x 2 y 2 ) U gz const)
2
2) Equilibrium of a homogenous liquid in a reference frame which is undergoing
a constant horizontal acceleration J. Derive an expression for pressure as a function
of the abscissa x parallel to the acceleration and the altitude z. Show that in this
reference frame isobar surfaces are inclined planes.
(Answer: p U gz J x const)
2.2.1.4.2. Conditions at fluid-fluid interfaces

Continuity of the pressure between two immiscible fluids is only guaranteed if
the interfacial surface is a plane. If this is not the case, the interface behaves in many
respects like a membrane; its constant surface tension is characterized by the set of
the two fluids, if they are pure, and generally decreases with temperature. The
pressure discontinuity p1 p 2 across the interface is given by Laplace’s law:
§1 1·
p1 p 2 V ¨¨ ¸¸
©R R' ¹
R and R' are the radii of curvature of two interfacial surface sections by
perpendicular planes which contain the normal to this interface at the considered
point of this one. The pressure p1 is greater on the concave side of the surface
(Figure 2.6b).
On the other hand, the contact angle between an interface and wall is a physical
characteristic between these three domains [BEN 06]. A liquid wetting the wall
perfectly is joining tangentially to another liquid (water on a clean wall made of
glass, for instance). Surface active substances generally ensure a perfect wetting of
the surface, but they modify the value of the surface tension.
2.2.1.5. Equilibrium of the atmosphere

By limiting ourselves to the troposphere (lower limit of the stratosphere), the
atmosphere can be considered to comprise a homogenous mixture of nitrogen and
oxygen, and it is characterized by a rapidly decaying temperature with increasing
altitude. It is not in thermodynamic equilibrium, because as oxygen is more dense
than nitrogen, it should be more concentrated in the lower layers. In fact, the
atmosphere is constantly undergoing movements whose characteristic time is small
compared with the time required for the temperature to become uniform. The
atmosphere is in fact perfectly mixed: in the movement of large air masses, the
atmosphere undergoes rapid compression and rarefaction, which occur
isentropically, and so during their movement air masses do not have the time to
exchange heat by thermal conduction. This mixing leads to an atmosphere at
constant entropy. Experiments confirm that the Earth’s atmosphere is in a state
which is very nearly homoentropic. This condition defines the standard atmosphere,
whose differences from reality, in the order of a few percent, are due to
meteorological phenomena.
EXERCISES –
Calculate the pressure p as a function of the altitude z:
§ g z z 0 ·
1) in an isothermal atmosphere of temperature T0: ( p p 0 exp¨ ¸ );
¨ rT0 ¸¹
©
2) in a homoentropic atmosphere, where the temperature at the ground is equal to
T0.
2.2.1.6. Stability of equilibrium solutions in stratified fluids

A stratified fluid with horizontal surfaces of constant density is in an
equilibrium situation in the gravitational field. The stability of this fluid can be
tested in an elementary way by displacing a fluid particle and observing whether
or not it undergoes a restoring force. Let us first of all consider an
incompressible fluid whose density, which depends only on temperature,
increases with increasing altitude. A fluid particle which is displaced in the
upward direction finds itself in a region of heavier fluid; it thus undergoes a
greater thrust, and its movement upward is amplified.
The same reasoning applied to a compressible fluid (the atmosphere for

example) leads to the following situation: as the displacement occurs
isentropically, it is the specific entropy of the particle which must be compared
with that of the local environment into which the fluid particle is moved. If the
local entropy of ambient fluid falls as the fluid particle rises, the temperature of
the particle is greater than that of this ambient fluid: the Archimedes force thus
increases and the fluid particle comes up.
In conclusion, we have equilibrium stability of a fluid if its temperature (or

its entropy) increases with increasing altitude. In the opposite case, the fluid
system is unstable.
2.2.2. Pressure forces on solid boundaries
2.2.2.1. Constant pressure

A constant pressure situation occurs in numerous applications: pipe-systems
and compressed-air reservoirs, water-distribution networks, etc., where the
effects of an external force field is often negligible as a result of the high
pressure levels which occur within the system.
G
Consider a surface S (Figure 2.7a) with normal n on which uniform pressure
p is exerted, the pressure force on this surface S is:
F ³S p n ds
Letting ds' be the projection of the elementary surface ds on the plane

perpendicular to the direction Ox, the component Fx on the axis Ox can be obtained
from:
Fx ³S p x..n ds ³S ' pds' p.S '

The component of the uniform pressure force in the direction Ox is equal to the
pressure force on the surface S', which is a projection of S on the plane
perpendicular to Ox.
P n
n
x
ds' x
S T
ds ds
OF 3
x
S’ S
(a) (b) y z
Figure 2.7. Pressure forces: (a) arbitrary surface with constant pressure;
(b) forces due to a liquid on a plane surface
In practice, this surface is an element of a pipe, a reservoir, etc. Let us take the
example of a cylindrical pipe with radius R. The stress T on the wall, due to the
overpressure 'p in the pipe, must balance the pressure forces on a half-cylinder. It is
thus equal to 'pR per unit length: the stress on the wall increases with the radius R.
For a tube of radius 1 cm transporting compressed air pressurized to 100
atmospheres, this is equal to 105 N/meter.
Pipeline and reservoir dimensions (diameters) ought to be as small as possible, in

order to avoid technical difficulties related to the wall resistances.
2.2.2.2. Hydrostatics
The equilibrium of a homogenous liquid (often water) in a gravitational field is
of considerable practical importance, particularly in cases such as water reservoir
walls, sluice gates and boat hulls.
The pressure forces on a plane surface, one side of which is covered by a liquid,
are parallel to the wall normal: they are thus equivalent to a single vector applied at
a given point on the wall, the center of pressure 3. Consider a plane surface S
identified by the coordinates Oxy in its plane (Figure 2.7b). Suppose that the free
surface of the water is at z = 0. The hydrostatic pressure is equal to 'p = Ug z.
Denoting the center of inertia of the surface S as G, the thrust P (the resultant
pressure force) is equal to:
P ³S Ugzds Ug cos T ³S yds Ug cos T . y G S U gz G S 'p G S

The thrust P is equal to the product of the pressure at the center of inertia G of
the homogenous surface and its surface S. The center of pressure, which can be
easily calculated by taking the moment of the pressure forces with respect to Oz, is
beneath the inertia centers.
The forces exerted by water on the walls of large reservoirs and dams are often
considerable; the construction of dams thus involves massive elevations of earth
whose weight blocks the mass of water (weight dams), or large concrete
constructions which resist by returning the thrust on the rocky walls (mountain arch
dams).
The reader can easily verify that the horizontal component (along Ox) of the
hydrostatic pressure force on a curved surface is equal to the hydrostatic pressure
force on the surface S', which is the projection of S on a plane perpendicular to Ox.
This result clearly makes no sense for a vertical component.
EXERCISES –
1) Calculate the coordinates of the pressure center 3 on the surface S of Figure
2.7b. (Answer: x 3 x G , y 3 y Gy I G y / S , IGy giving the inertia moment of
the surface taking with respect to a straight line parallel to Oy and passing through
G.)
2) Calculate the load-force generated on a vertical rectangular wall, 4 m in width
and containing a mass of water 3 m in height. Determine the position of the center of
pressure. (Answer: 18.0 ×104 newtons, 1 m above the bottom.)
3) Answer the same questions for a dam in the form of a 50 m high equilateral
triangle.
2.2.2.3. Floaters
A floater is a body placed on the surface of a liquid, and whose weight is less
than that of an equivalent volume of the same liquid. In this kind of situation there is
always an equilibrium position, such that the force exerted by the fluid on the body
balances its weight. However, the stability of this position is not guaranteed. We will
not study hull-stability problems ([BAR 01]) which requires geometrical knowledge
related to surface curvature and normal fields ([KRE 91]).
We will only discuss a simple example in order to illustrate the origin of such
problems. Consider a homogenous, elliptical cylinder whose density is half that of
water, such that at equilibrium, the center of the ellipse is always in the plane free
surface of the water. While in equilibrium, the upward thrust which the cylinder
experiences, P, applied at the center of buoyancy 3 opposes that of its weight, P,
applied at the center of gravity of the ellipse, G, which lies in a vertical section of
the meridian plane of the cylinder. There are two equilibrium positions ((a) and (c)
in Figure 2.8) corresponding to the vertical positions for each axes of the ellipse.
The equilibrium stability is studied in the usual way: a small clockwise rotation of
the cylinder is performed (positions (b) and (d) in Figure 2.8).
G G G G
3 3 3 3
(a) (b) (c) (d)
Figure 2.8. Stability study of a floater whose cross-section is elliptic
We assume that the submerged volume (beneath the flotation plane) and the
thrust P remain constant during the displacement; however as the distribution of
the volume submerged has changed, the center of buoyancy 3 is moved to the
side where the submerged volume has increased (with respect to the vertical
passing through G), thus creating either a restoring moment (situation (b)) or an
amplifying moment (situation (d)), depending on the situation: position (a) is
stable, whereas position (c) is unstable.
In fact for a real hull we must also take account of the way it is loaded, and
possible movements of this load. For the preceding elliptical floater, we see that
we can add a weight to the upper surface of the cylinder: the system will be
stable as long as the restoring moment is less than the moment generated by the
extra weight. On the other hand, a circular cylinder is in a state of neutral
equilibrium if it is not loaded; it obviously becomes unstable if any weight is
added to the upper side. Escorting floating logs during their floatation down
rivers is not a straightforward use of fluid statics!
2.3. Heat conduction
2.3.1. The heat equation
By inserting expression [2.15] into relation [2.10] we obtain a thermal balance

equation:
wT w § wT ·
UC Ö ¸ VT
wt wx i ¨© wxi ¸¹
which, if the thermal conductivity of the medium O is constant, becomes:
wT w 2T
UC O 'T V T with '
wt wx i wx i
In the absence of thermal power sources (VT = 0), we obtain the heat equation:
wT
a'T
wt
The thermal diffusivity a O U C can be expressed in meter2/second.
2.3.2. Thermal boundary conditions
Here we will deal with the principal problems of heat conduction, which we
will later encounter in a similar form when we deal with viscous fluid
mechanics. Thermal boundary conditions can generally be classed in one of the
following categories:
1) Dirichlet condition: the temperature distribution T(M) is known at all
points on the boundary. This is often the case for recipient walls which contain
an agitated fluid, walls which are in contact with highly conductive solids
(metals, etc.); the corresponding practical problems relate to heating, thermal
insulation, etc.
2) Neumann condition: the heat-flux density on the boundary qTp ( M ) is
known at all points M on the boundary. These conditions are found in the
presence of heat sources resulting from the dissipation of other forms of energy,
such as heating by Joule effect, heat produced by nuclear reactors, radiation
absorption, etc.
3) Mixed condition: where a linear relationship exists between the wall flux-
density and the temperature difference between the wall temperature Tp(M) and
a given temperature T0:
qTp ( M )
§ wT ·
O ¨¨ ¸¸
h T p T0
© wn ¹ p
The coefficient h of the above relation is called the heat-transfer coefficient

of the wall. It can represent diverse phenomena, such as an external wall, heat
transfer across the boundary layer of a flowing fluid, etc.
Furthermore, as we have already said, it is necessary to fix the initial

temperature distribution in the domain being studied.
In addition to a detailed knowledge of temperature fields, other important

unknown quantities are:
– either the wall heat-flux-density and the thermal power issuing from the
domain studied when the wall temperature is given;
– or the wall temperature in the case where the heat-flux wall is given.
Finally, in the case where chemical reactions occur at the boundary

(evaporation, fusion, catalytic reactions, oxidation, electrolytic reactions, etc.),
we have more complex conditions. So, it is often at the boundary that heat is
released (positively or negatively) with a power proportional to the speed of a
reaction. As the reaction is often limited with respect to the transfer towards the wall
of certain reactants present in the fluid, we have a coupling between a thermal
problem and a problem related to the diffusion of the fluid constituents in which the
presence of flows plays an essential role. For example, in evaporation of a wet wall
in an air flow (temperature measured by a wet-bulb thermometer), the quantity of
liquid evaporated and heat absorbed in the wall surface are increasing functions of
the flow velocity, but the wall temperature is quite independent of the flow velocity
([BIR 01]).
2.4. Diffusion
2.4.1. Introduction
2.4.1.1. Definition
When a fluid F1 is carefully introduced to a recipient which contains a fluid F2
which is miscible with F1, after a certain duration we notice that movements related
to mixing have ceased, and that hydrostatic equilibrium is attained; however, an

inhomogeneity can persist in the concentration (this can be observed by differences
in color for example). This indicates that the phenomenon is an extremely slow
function of time. This migration of the components with respect to one another
constitutes a phenomenon known as material diffusion which is characterized by the
fact all of the components of a mixture do not have exactly the same speed, and
these differences lead to variations in the composition of the mixture.
The molecular velocity of a body of given chemical species which makes up a

mixture is thus engendered by two causes: the diffusion and the bulk movement of
the mixture (convection). The separation of these two causes must be very carefully
considered, as the bulk movement results from the behavior of all of the
constituents. Interactions between the two kinds of phenomena (convection and
diffusion) are encountered in all mixing processes with or without chemical
reactions (for example, sugar, which dissolves in coffee).
2.4.1.2. Microscopic interpretation of diffusion

This difference in the behavior of the two components is obviously related to
molecular agitation, which leads to a macroscopic displacement and is different for
the two components. The molecular agitation can be directly visible under certain
conditions in the presence of very small particles (in the order of a few micrometers)
in suspension in a fluid. Examined under a microscope these particles present erratic
behavior: this is known as Brownian movement, which is a macroscopic
manifestation of molecular agitation. These movements obey diffusion equations.
Let us consider two molecular species (black and white molecules in Figure
2.9a). By supposing that the two categories of molecules have the same energy (in
other words the same temperature), each molecular species moves as a result of
collisions. These collisions have a random character; for a resting gas made up of
one or many species of molecule, uniformly distributed, we see that the molecules
are statistically stationary, because each species only permutes between themselves.
During collisions and molecular interactions in the volume shown in Figure 2.9a,
there are more black (or respectively white) molecules at the bottom (or at the top),
but they will end up uniformly distributed in the container considered. The statistical
aspects of the second principle of thermodynamics leads to a complete mixing of the
two species, whose concentrations become uniform.
However, in this container, where the pressure is supposed constant, the total
molar concentration remains constant at all points, with the species whose
concentration decreases being replaced by the other species whose concentration
increases. This pressure constraint leads to a relation between migrations of the two
species.
x n2a x
A
ns O
(a) (b)
Figure 2.9. (a) Gaseous diffusion in a volume; (b) evaporation of a liquid
Evaporation towards the atmosphere of a liquid contained in a tube (Figure 2.9b)

is another example of gaseous diffusion where we have a vapor flux from the liquid
towards the exterior across air, which is at rest in the tube. Note that the vapor
crosses the free surface of the liquid (at the lower wall where evaporation takes
place) and diffuses towards the exterior air, whereas the air, which does not dissolve
in the liquid, does not cross the free surface.
For the sake of simplicity, we will consider the case of mixing between two
perfect gases: gravity here often plays a negligible role at the laboratory scale.
Kinetic gas theory allows the modeling of phenomena at the molecular scale. We
will limit the discussion to cases where the temperature of the systems studied is
uniform; if this was not the case, it would be necessary to introduce a further
phenomenon: thermal diffusion. The interested reader should refer to classic texts on
the subject (see section 2.4.5).
It should nevertheless be noted that, depending on the problem studied, the

influence of external forces or accelerations on diffusive phenomena should be taken
into account (centrifugation for example).
2.4.1.3. Extensive variables of a mixture

The composition of a continuous medium where k chemical species are present
can be characterized by the local reduced extensive concentration variables which
can be either:
the number of moles ni of each species per unit volume (molar

concentration);
the partial density Ui of the species i, which is the mass per unit volume
( U i ni M i ) of the molecules of species i, of molar mass Mi (mass
concentration);
the partial pressure pi associated with the molecules of species i supposed
only to occupy an elementary local volume of the mixture.
As the preceding quantities are additive, we can define the total number of moles
n (or molecules) per unit volume, and the density U of the mixture:
k k k
n ¦ ni ; U ¦ Ui ¦ ni M i [2.20]
i 1 i 1 i 1
If the system comprises a perfect gas, which we assume in what follows, the
total pressure is equal to the sum of the partial pressures of the different
constituents.
We also consider the concentrations of the species i by means of relative values:

– molar fraction J i ;
– mass fraction Z i :
Ji ni / n ; Z i Ui / U [2.21]
According to [2.20], these variables respect the relations:
k k
¦J i 1; ¦Zi 1
i 1 i 1
While the density U or the number of moles n per unit volume of a mixture
appear in the global mixing equations, the composition of the mixture is in fact
characterized by the given concentrations of only k 1 components.
In a liquid medium, the molar concentration Ci (also called molarity) is

expressed as the number of moles of a species contained in a volume equal to 1 liter.
It often happens that the mixture contains a constituent 1, which is dominant, and
a weak proportion of a species i:
ni n1 or: pi p1 or: i 1 ....
All relative concentrations of species 1 then have the same first order value. For
example, it is thus possible to confuse the molar fraction Ji = ni / n with the quantity
Ji = ni / n1, the errors thus committed being in the order of (ni / n1)2. The same goes
for the mass fraction Zi = Ui / U and the quantity Ui / U1, the partial pressure pi / p
and pi / p1. All these variables of relative concentration are proportional. We will
note as ci the species concentration i present in small quantities in a mixture.
2.4.2. Molar and mass fluxes
2.4.2.1. Flux of a component

We have seen (see section 2.1.3.1) that for any scalar extensive quantity G, we
G
can identify a corresponding vector flux density q G , whose flux across the surface
(S) (equation [2.3]) represents the quantity of G which crosses S per unit time. The
extensive quantities of a pure body (volume, energy, etc.) are associated with the
matter which we suppose fixed in the reference frame. In the presence of diffusion,
not only does this reference not exist, but the flux densities of the components, as a
result of their movement, depend on the particular reference which is chosen. We
will return to this choice at a later stage...
G It can easily be seen that the matter which crosses a surface

G ds at a velocity
V over a Gperiod of length Gt, comprises a cylinder of length VG t whose volume is
G
equal to V.n į t.
G
n
S G
V
ds
G
V.G t
Figure 2.10. Balance in an elementary displacement across ds

Letting g be the volume density of the quantity G (see section 2.1.1), we can
G
derive expressions for the vector flux density q g of G2 and the flux M GS of the
quantity G across the surface S:
G G G G GG
qg gV M GS ³S q G .n ds ³S g V .nds
By considering g to be the number of moles per unit volume, we obtain

G
respectively for the molar flux density q ni and the molar flux M ni across (S) of the
G
constituent i with velocity Vi :
G G G G G
q ni niVi ; M ni ³S q ni .nds ³S ni Vi .nds [2.22]
If we now choose mass to be the extensive quantity, we obtain, respectively for

G
the mass flux density q mi and the mass flux M mi of the component i across (S):
G G G G G G
q mi ȡ i Vi M mi ³S q mi .n ds ³S U iVi .n ds. [2.23]
The molar flux M ni and the mass flux M mi represent, respectively, the number
of moles and the component mass i crossing S per unit time.
2.4.2.2. Balance equations

The balance equation of each constituent i (i =1, 2,…, k) can be obtained by
applying equation [2.8] to the volumic number of moles ni:
wni wq nij § wni G ·

V ni ¨¨ V ni divq ni ¸¸
wt wx j © wt ¹
The volume source Vni is the (volumetric) volumic number of moles in

component i created by homogenous chemical reactions. By replacing the molar flux
density with expression [2.22]:
wni

w
ni uij V ni i 1,2,..k . [2.24]
wt wx j
2 We shall study convective fluxes in more detail later (section 3.3.3.2).

Balance equations can be written for the density Ui , which are equivalent to
[2.24], by multiplying each of these by the molar mass Mi of the corresponding
component:
w Ui

w
U i uij M iV ni i 1,2,..k . [2.25]
wt wx j
2.4.2.3. Global fluxes and mean velocities

For the set of components of a mixture, we define:
G
– the total density q n of the molar flux and the total molar flux M n :
G k G k G G G
qn ¦ q ni ¦ niVi Mn ³S q n .n ds [2.26]
i 1 i 1
G
– the total density q m of the mass flux and the total mass flux M m :
G k G k G G G
qm ¦ q mi ¦ U iVi Mm ³S q m .n ds [2.27]
i 1 i 1
The global molar and mass balances for the mixture allow average mixing
velocities to be defined according to the methodology described in section 1.4.2.5
for the intensive quantities3. We thus define:
G
– the molar average velocity V * (which is independent of the molar mass) using
the molar balance [2.26]:
G 1 k G
V* ¦ niVi [2.28]
ni 1
G
– the mass average velocity V (velocity of the inertia center of a fluid particle4),
using mass balance [2.27]:
G k G G
1 qm
V ¦ U i Vi [2.29]
U i 1 U
where U i ni M i is the partial density of the species i of molar mass Mi.
3 The velocity is an intensive quantity (see section 1.2.1).

4 See section 3.2.1 and following sections.
G G
The velocities V and V * are the same only if the continuous medium comprises
a homogenous mixture of constant composition or of identical molecules (a pure
substance).
The molar fluxes M n and M m represent, respectively, the number of moles and
the total mass of a mixture which are crossing a surface S per unit time:
G *
G G
Mn ³S q n .ds ³S nV .ds Mm ³S q m .ds ³S UV .ds
By summing equations [2.24], term by term, and taking account of definition

[2.28], we obtain the balance equation for the total volumic number of moles n:
wn
wt

G
div nV *
k
¦ V ni [2.30]
i 1
k
The quantity ¦ V ni is the number of moles created by chemical reactions.
i 1
k
We do the same for equations [2.25] and [2.29]; however, the sum ¦ M i V ni is
i 1
zero as there is no mass source. We obtain the conservation equation for the total
mass:
wU

G
div U V 0 [2.31]
wt
We will encounter the preceding equation again in the next chapter, during our
study of fluid flows, where it plays a fundamental role.
The reader will note that the k equation [2.24] (or [2.25]) is equivalent to the
system of k-1 equation [2.24] (or [2.25]) and equation [2.30] (or [2.31]). In other
words, it suffices to describe k-1 component balance equations in addition to the
global balance equation for the mixture.
2.4.3. Choice of reference frame
In this chapter, we have already studied the transfer of diverse quantities G
(energy, entropy, etc.) with respect to matter which was supposed stationary; in
Chapter 3 we will examine the transport of a quantity G associated with convection

of matter, in other words associated with a bulk movement of the continuous
medium.
G
In an inhomogenous mixture, each component i has its own velocity Vi . There is
therefore no preferred reference frame in which all of the matter is stationary. It is
for this reason that we did not specify the reference frame used in the preceding
definitions of flux.
As diffusion is a differential phenomenon between the components, it is

necessary to identify a global condition which defines the bulk movement of the
matter, in other words it is necessary to fix the local mean velocities field for the
matter, as diffusion is a phenomenon which must be considered relative to a mean
velocity. The reference frame chosen for a study of diffusion is thus determined by
the choice of reference frame for the mean velocity. It depends on the kind of
problem studied, as the following examples demonstrate.
The quantities which characterize diffusion (velocities, flux densities), which

depend on the reference frame R chosen for their study, will be represented between
brackets > @* with a superscript (*,G or 1) indicating the reference frame.
2.4.3.2. Diffusion in a fluid at mechanical rest
Let us now consider an isothermal fluid mixture containing k components.
Diffusion phenomena are here characterized by very slow velocities associated with
negligible accelerations. The local composition is characterized by the number of
moles ni per unit volume for each species. Suppose that the total number of moles n
per unit volume is constant at all points in the mixture (for example, in an ideal
isothermal gas). In the absence of chemical reactions, equation [2.30] can thus be
written:
G
divV * 0 [2.32]
We will limit our discussion to a simple case involving a configuration in stable

equilibrium (section 2.2.1.6), in which the concentration is constant in horizontal
planes, the diffusion velocities being vertical and parallel to the axis Ox. The field
G
V * x, t , which thus respects the condition dV * dx 0 , derived from [2.32], is
G
uniform at each instant and the molar average velocity V * t is only a function of
time.
The preceding situation is observed for isothermal gaseous mixtures whose

pressure is uniform and constant. In a gravitational field, diffusion only occurs
without natural convection for fluids in situations of 1D equilibrium. In other cases,

the inevitable density gradients created by variations in the component
concentrations must be taken into account: these induce complex movements due to
natural convection in the fluid.
2.4.3.3. Diffusion in a closed container

Let us now consider a perfect, isothermal, gaseous mixture, contained in a fixed
container (Figure 2.9a), whose total volumic number of moles n is constant. We will
suppose that the walls are impermeable; this results in a molar average velocity
which is zero in the reference frame of the container, which can thus be considered a
preferred reference frame:
G k G * G * G G
nV * ni ¡¢Vi ¯°± 0, with: ¡¢Vi ¯°± Vi V *
i1
This results in the molar flux densities playing an important role in this reference
frame:
G G G *
q ¯ * n V ¯ * , with: ¡ q ni ¯° 0
¢¡ ni ±° i ¢¡ i ±° ¢ ±
i
2.4.3.4. Diffusion in steady evaporation

Let us now examine the 1D problem of steady evaporation of a liquid contained
in a vertical tube as shown in Figure 2.9b. As before, the pressure is constant in the
tube. We suppose that the temperature is constant and the composition is
independent of time, but Gnot uniform in space. The total volumic number of moles n
and the molar velocity V * (section 2.4.3.2) are steady and uniform in space. The
molar flux densities of air (species 1) and vapor (species 2) are steady (section
2.4.3.2).
As the free surface of the liquid at the bottom of the tube is impermeable to air,
the flux density and velocity of the air are both zero at all points within the tube: the
G 1
air is at rest in the reference frame of the tube ( V1 > @
0 ) and so diffusion velocity
of vapor in air can be written:
>VG2 @1 G G
V2 V1
Since we have a steady evaporation regime, we can conclude that the flux of
liquid evaporated >n2V2 @1 across a cross-section of the tube is independent of the
vertical position of this section (otherwise, we should have some accumulation of

vapor between two sections with different fluxes, which would increase with time).
2.4.3.5. Diffusion in a moving medium

By virtue of the laws of dynamics, the characteristics of the inertia center of a
material system can be obtained through knowledge of the external forces acting on
the system. The same goes for fluid particles of a continuous medium if the external
forces acting on the fluid particle are known.
Diffusive phenomena amount to differences in velocity between different

components of a fluid. The separation of bulk movement and diffusive phenomena
can be achieved by identifying the diffusive velocities in a reference frame
associated with the inertia center G of fluid particles (mass or barycentric reference
G
frame) which move with velocity VG (mass velocity) and we will denote simply as
G
V , as obtained from the equations of motion which will be covered in Chapter 3.
From definition [2.29] we have, for the local inertia center:
G k G
UV ¦ U iVi [2.33]
i 1
Quantities measured in a reference frame RG associated with this inertia center

are denoted > @G (between brackets and with the superscript G). Relation [2.33] can
be written:
k G G G G G G
i ¡¢Vi ¯°± 0, with: ¡Vi ¯° Vi V
¢ ±
[2.34]
i1
This relation expresses a particular importance of the mass flux densities:
G G k
G
q ¯ G V ¯ G ,
¡¢ mi °± i ¡¢ i °± with: <qmi >G 0 [2.35]
i1
which describe the diffusive fluxes in this reference frame.
2.4.3.6. Diffusion of components in weak concentration

The preceding difficulties disappear when a mixture is almost entirely comprised
of one of the components (for example species 1). Supposing that the number of
moles of the other components takes on a small value, of order H, we have seen in
section 2.4.1.3 that all concentration definitions were equivalent for the first order. It
can be shown in the same way from expressions [2.28] and [2.29] that the
G G G
velocities V , V * and V1 are equal to order H, so that the predominant component
constitutes the natural reference frame.
In this case, the diffusion velocity of a component is its velocity with respect to
the predominant species, and any extensive quantities can be used to define a flux
density, as the reference frame is no longer described by a balance equation.
2.4.3.7. General methodology: diffusion in an arbitrary reference frame

The preceding examples show that the velocity of a component is not the only
characteristic of its diffusion and we always have a global condition between the
ensemble of diffusion velocities which characterize the particular problem
considered. Diffusion itself is characterized by diffusion velocities and flux densities
in the reference frame best adapted to the problem considered. The flux density of
the extensive quantity associated with this velocity also depends on the choice of
reference frame.
The diffusion velocities and corresponding flux densities (molar, barycentric or

otherwise) of each component are summarized in Table 2.1 for different reference
frames, which are denoted by indices corresponding to the preceding quantities:
the superscript
G* * in the reference frame R* corresponding to the molar average
velocity V ;
the superscript G in the reference frame RG corresponding to the mass average
G
velocity5 V (G being the inertia center);
the superscript 1 in the reference frame R1 of fluid 1.
G
5 The massG speed
G V is the local speed of fluid medium intervening in the mechanical
equations ( V G V ).
Reference frame Extensive Diffusion Diffusion flux densities

quantity velocity
R*
molar average
ni
>VGi @* G G
Vi V * >qG ni @* > @*
G
n i Vi G G
n i Vi V *
G k G
velocity V * ¦ >q ni @
*
0
i 1
RG
mass average
Ui
>VGi @G G G
Vi V >qGmi @G > G @G
U i Vi G
U i Vi V
G

G k G
¦ >q mi @
velocity V G
0
i 1
R1 ni
>VGi @1 G G
Vi V1
G
> qmi @1
G 1
Ui ª¬Vi º¼ , i z 1
component 1 Ui
G
velocity V1 >VG1 @1 0 G
> qni @1
G 1
ni ª¬Vi º¼ , i z 1
G G
> qm1 @1 > qn1 @1 0
weak concentrations
of components other
ci i z 1
>VGi @1 >VGi @G >qGci @1 > @1
G
c i Vi G G
c i Vi V 1
than 1 >VGi @* i z 1 i z1
R * #R G # R 1 >VG1 @1 0
Table 2.1. Variables characterizing diffusion for different reference frames
The situations described in Table 2.1 correspond to those most frequently

encountered. Other more complex situations may arise (unsteady evaporation,
macroscopic mixing of numerous components with walls which are permeable to
only some of these, etc.). In all cases, a global analysis is necessary for an
appropriate choice of representation for the diffusion equations.
2.4.4. Binary isothermal mixture
2.4.4.1. Expressions for diffusion velocities and flux densities

We have reviewed the different definitions which are possible for extensive
quantities in different reference frames. For n flux-density vectors, diffusion
phenomena are characterized by n-1 independent equations for the flux-densities or
the diffusion velocities, and by an equation which describes a global condition. The
n-1 differences in diffusion velocities, two by two are obviously independent of the
reference frame used, and they can be linked to causes of imbalance, such as
concentration gradients, by phenomenological relations. We here limit our treatment

of the problem to a binary mixture; the reader should consult texts treating the
problem of irreversible thermodynamics for developments associated with diffusion
in mixtures with more than two components.
For a binary mixture, the only quantity which is independent of the reference
G G
frame and diffusion characteristic is the velocity difference V1 V 2 , of species 1
and 2 with respect to an arbitrary reference frame. Diffusion velocities in the
G G
reference frames R* or RG can then be easily expressed as a function of V1 V 2 .
G
In effect, in the reference frame corresponding to the mean molar velocity V * ,
we obtain relation [2.28]:
G * G * G G G*
n1 ¢¡ V1 ±°¯ n2 ¢¡ V 2 ±°¯ 0 with: ¯*
¢¡V i ±° V i V
which can be written as:
>VG1 @*
>
V2 @
G *

V1 V 2
G G

n1 n 2 G G
V1 V 2 [2.36]
1 1 1 1 n

n1 n2 n1 n2
or:
>qG n1 @* > @
G *
n1 V1
G
>q n 2 @* > @
G *
n 2 V 2
n

n1 n 2 G G
V1 V 2 [2.37]
In the reference frame R* associated with the molar velocity, the diffusion
velocities and molar flux densities of each component can be written as a function of
G G

the velocity difference V1 V2 .
Similarly, using relation [2.34] between the mass flux densities in reference
frame RG we obtain:
>qG m1 @G > G @G
U1 V1
G
>q m 2 @G U 2 V2> @
G G
U

U1 U 2 G G
V1 V 2 [2.38]
Comparing [2.36] and [2.38] gives:

>VG1 @G U2 n G
> @* n1 n 2 M 2 >VG1 @* > @
M2 G *
[2.39]
V1 V1
U n2 n1 M 1 n 2 M 2 M
>VG2 @G M1 G
M
V2> @* [2.40]
n1 M 1 n 2 M 2
where M is the local average molar mass of the mixture.
n1 n 2
In summary, the diffusion velocity in a given reference frame can be expressed
as a function of the velocity difference between the two components by means of the
law which defines the reference frame. It is thus velocity differences which
characterize binary diffusion phenomena independently of the reference frame
which is chosen.
2.4.4.2. Isothermal diffusion

2.4.4.2.1. Fick’s law
Isothermal diffusion is caused by the existence of concentration gradients which
we will characterize by a volumic number of moles.
In order to define a diffusion coefficient we will use the relation:
n
grad 1
G G n2
V1 V 2 D [2.41]
n1
n2
n1
the ratio being dimensionless. It is equal to the concentrations ratio of two
n2
species, expressed using a proportional definition for the two species6:
G G grad J 1 J 2 grad U1 U 2
V1 V 2 D D
J1 J 2 U1 U 2
[2.42]
grad Z 1 Z 2 grad p1 p 2
D D
Z1 Z 2 p1 p 2
6 Using partial pressures is only correct for perfect gases and ideal solutions.
Relation [2.41] shows that the diffusion coefficient can be expressed in m2.s-1.
For a gas at atmospheric pressure, it is more or less independent of concentration.
For mixtures with air, it usually lies between 1.0 × 10-5 and 2.5 × 10-5 m2.s-1, the
lowest values corresponding to heavy molecules (carbon dioxide, ethanol, benzene,
etc.). Very light molecules (hydrogen, helium) give larger values (up to 13.2 × 10-5
m2.s-1 for hydrogen-helium mixtures).
In liquid mixtures, the diffusion coefficient is in the order of 10-9 m2.s-1.

Diffusion in solids results from different mechanisms, depending on whether we are
dealing with diffusion of impurities which move from a free position in one
crystalline structure to another, or with particles (atoms, etc.) capable of moving
around the structural grid. The diffusion coefficient in solids varies from 10-12 to
10-14 m2.s-1.
Expression [2.41] shows that the quantity D grad has the dimension of velocity
and that this gives an order of magnitude of D A for the diffusion velocities, where
A is the distance over which the concentration gradient is extended. Taking for
example A = 0.1 meter, we can see that the diffusion velocity is in the order of 10-4
m.s-1 in gases and 10-8 m.s-1 in liquids. These velocities increase considerably if the
distance A is significantly diminished; as for momentum transfer (section 6.5.3),
convection effect can reduce this quantity to values comparable with the thickness of
a boundary layer, leading to a significant increase in diffusion velocity ([BIR 01]).
We define the Lewis number Le as the dimensionless ratio between the diffusion
coefficient and thermal diffusivity a:
D
Le
a
Excluding instances of extremely strong force fields or accelerations, the total

number of moles n per unit volume is often nearly constant (n1+ n2 # constant)
under standard conditions (in particular for ideal gases). Thus, we have:
grad n1 grad n 2 0 [2.43]
Substituting [2.43] into Fick’s law [2.41], the velocity difference can be written
as a function of the concentration n1 alone:
G G n n
V1 V 2 D grad n1 D grad n1 [2.44]
n1 n 2 n1 n n1
2.4.4.2.2. Diffusion in the local reference frame of molar velocity

We use the local reference frame R* of molar velocity for diffusion in a closed
fixed volume. From [2.41] and [2.44] we obtain an expression for the molar flux in
the reference frame R*:
>VG1 @* D
grad n1
n1
>VG2 @* D
grad n 2
n2
[2.45]
From this we can deduce the molar flux densities of the two species:
>qG n1 @* n1 V1> @
G * G
>q n 2 @* > @
G *
n 2 V2 D grad n1 D grad n 2 [2.46]
Instead of representing the concentrations by n1 and n2, diffusion velocities in

the fixed reference frame can be expressed as a function of quantities of the two
gases which are proportional to these concentrations, such as the partial specific
masses U1 and U2, the partial pressures p1 and p2, and the molar fractions J1 and J2:
>VG1 @* D
grad U1
U1
D
grad p1
p1
D
grad J 1
J1
[2.47]
> @
G *
V2 D
grad U 2
U2
D
grad p 2
p2
D
grad J 2
J2
The mass fractions cannot be directly used as concentration variables, because

the specific mass of the mixture varies within the volume.
NOTE – Expression [2.47], using the molar fractions as variables, is exact, even if
the volumic number of moles n has a non-zero gradient. This results from the
definition:
JJJJJG JJJJJG
1 2 1 and: grad1 grad2 0
and from expression [2.42] of Fick’s law using molar fractions which can be written:
G G D
V1 V 2 gradJ 1 [2.48]
J 1J 2
From [2.47], we obtain the densities of molar flux:
>qG n1 @* n D grad J 1 >qG n 2 @* n D grad J 2 [2.49]
2.4.4.2.3. Application to diffusion in a medium at macroscopic rest

Here we consider a medium at macroscopic rest, in other words a medium
whose very slight movements are related to diffusion phenomena. We assume
that the medium satisfies the equations of fluid statics and that differences of
specific mass do not lead to movements associated with natural convection.
In the case of a closed impermeable container in which a fluid is at rest, the

molar average velocity is assumed to be zero at all points of the medium in a
cartesian reference frame associated with the container.
The diffusion equation of the components can be obtained by applying [2.8]

for the number of moles n1 and n2,
wn1 wq n1 j
VG
wt wx j
which, on accounting for expression [2.46], gives:
wn1
wt
w §¨ wn1
D
wx j ¨© wx j
·
¸ V
¸ G
§ wn1
¨¨
© wt
·
div D grad n1 V G ¸¸
¹
[2.50]
¹
The volume source VG of species 1 expresses the number of moles per unit volume
of this species created by a homogenous chemical reaction inside the container. It is
fixed by chemical kinetics. It is zero in the absence of any chemical reaction.
In the common case where the diffusion coefficient is constant and where there
is no chemical reaction, equation [2.50] reduces to the heat equation:
wn1
D 'n1 [2.51]
wt
The boundary conditions for impermeable walls can be represented by a zero

G
normal component grad n1 at the walls ( nw : normal to the wall):
G § wn1 ·
n w .grad n1 ¨¨ ¸¸ 0
© wn ¹ w
2.4.4.2.4. Diffusion with respect to a fixed component

The discussion of the steady evaporation problem given in section 2.4.3.4
(Figure 2.9b) showed us that gas 1 is fixed. In this case it suffices to directly apply
Fick’s law in the form [2.44]:
G n
V2 D grad n 2
n1 n 2
The flux density of species 2 can thus be written as:

G G grad n 2
q n2 n 2V 2 D n
n n2
The conservation equation of species 2 can be written:
§ grad n 2 ·¸
div¨ D n 0 [2.52]
¨ n n 2 ¸¹
©
For the 1D evaporation problem of Figure 2.9b, we obtain, by integrating [2.52]:
Dn dn2
qn2 constant [2.53]
n n2 dx
We consider the following boundary conditions:

at the surface of the liquid ( x 0 ), the number of moles per unit volume nS, is
that of the saturated vapor at the experimental temperature;
at the tube extremities ( x A ), we assume that the very slight movements of
atmospheric air suffice to eliminate all vapor, such that we can write that the vapor
concentration n2 is equal to the water vapor concentration n2a of the atmospheric air.
The distribution of the water vapor concentration can be obtained by

integrating equation [2.53]:
§ x q n2 ·
n n2 n n S exp¨¨ ¸
¸
© Dn ¹
Expressing, at the abscissa A, the concentration condition for the ambient air
we obtain a relation giving the flux density of the vapor:
q n2 A § n n 2a ·
ln¨ ¸
Dn ¨ nn ¸
© S ¹
For relatively weak concentrations we have:
Dn S n 2a
q n2 [2.54]
A
Numerical application
Take a tube of height A = 0.10 m. We base our reasoning on a density which is
proportional to the volumic number of moles for water vapor at 20ºC, assuming the
external air to be perfectly dry (n2a = 0) and using the following data: D=2.6 × 10-5
m2.s-1: Us = 0.018 kg.m-3. We find, for the flux density evaporated:
2.6 u 10 5 u 0.018
q m2 4.68 u 10 6 kg.m 2 .s 1
0.1
This corresponds to a reduction in the level of the free surface of about 0.4 mm
per day. For acetone, whose saturated vapor pressure is 10 times greater, the
evaporation is 10 times more rapid.
Evaporation rate in static conditions is very weak. Diffusion in a stationary

medium is a very slow process, which only becomes efficient in the presence of
convection (section 2.4.4.2.1).
2.4.4.2.5. Diffusion in a reference frame related to the local inertia center

Equation [2.38] gives an expression for the mass flux in a reference frame
associated with the local inertia center:
>qG m1 @G > G @G
U1 V1
G
>q m 2 @G > @
G G
U 2 V2
U 1U
U
2
VG1 VG2 [2.55]
The expression of Fick’s law [2.42] which is best adapted to this reference frame
uses mass fractions for the concentration variables:
Z1
grad
G G Z2 ª gradZ gradZ 2 º
1
V1 V2 D D« »
Z1 «¬ Z1 Z 2 »¼
Z2
From the definition of mass fractions, we obtain:
Z1 Z 2 1 and gradZ 1 gradZ 2 0
and thus:
G G D
V1 V 2 gradZ1 [2.56]
Z1Z 2
Substituting [2.56] into [2.55] gives:
>qG m1 @G U D gradZ 1 >qG m 2 @G U D gradZ 2 [2.57]
There are almost no static diffusion problems which lead to a motionless local
inertia center. Expression [2.57] will be used in the diffusion equation in the
presence of convection in Chapter 4.
NOTE – Using relations [2.39] and [2.40], diffusion velocities in the reference
frame RG can be expressed, as before, as a function of the variables n1 and n2, as
well as U1 with U2, p1 and p2 or J1 and J2:
>VG1 @G
DM 2 grad U 1
M U1

DM 2 grad p1
M p1

DM 2 grad J 1
M J1
[2.58]
>VG2 @G
DM 1 grad U 2
M U2

DM 1 grad p 2
M p2

DM 1 grad J 2
M J2
[2.59]
With these concentration variables, diffusion velocities can no longer be

expressed by means of the same diffusion coefficient and so we have two different
diffusion coefficients, in which the molar average mass M is variable:
DM 2 DM 1
DG1 DG 2
M M
2.4.4.2.6. Unsteady evaporation

From a practical point of view, the consequences of the developments
outlined in the preceding sections are that the exact formulation of equations for
diffusion problems should only be performed after an explicit choice of
reference frame when possible; this allows us to obtain precise expressions for
JJG JJG
[V 1 ] and [V 2 ]1 . For example, in certain problems (steady evaporation, see
sections 2.4.1.2 and 2.4.4.2.4) one of the gases, say G1, is stationary, and only
gas G2 moves during diffusion. Taking a reference frame R1 linked with G1, the
JJG
velocity [V 2 ]1 of gas G2 follows immediately from [2.41].
On the other hand, if the evaporation is no longer steady, then the air (species 1)
is no longer stationary. Consider the case where the tube in Figure 2.9b is filled only
with air at the initial instant; the total pressure remains constant; the rise of vapor
(species 2) in the tube moves the molar quantity corresponding to the air which is
now in movement and is thus no longer the favored reference frame.
Let us reconsider the evaporation problem for the unsteady 1D case, in an

atmosphere at constant temperature and pressure (Figure 2.9b). The total
volumetric number of moles n is constant. This leads to a situation where the
molar average velocity V* and the total molar flux q n nV * are independent of
the abscissa x. However, a priori they do depend on time. All that remains is to
write the balance equation for the vapor in the reference frame of the tube:
wn 2 w
n 2V2 0 [2.60]
wt wx
The flux density of the vapor n 2V 2 is now a function of the two variables x and t.
G G *
It can be obtained by decomposing the velocity, V 2 ¡V 2 ¯° V * , and by replacing
¢ ±
JJG
[V 2 ] by its expression in [2.45]:
wn 2 n2
q n2 n 2V2 D qn [2.61]
wx n
By substituting into equation [2.60], we obtain the diffusion equation:
wn 2 w § qn wn ·
¨¨ n2 D 2 ¸¸ 0 [2.62]
wt wx © n wx ¹
As the velocity V1 of the air is zero at the abscissa x = 0 (impermeability of the

liquid surface), the total molar flux q n nV * is equal to the molar flux >n 2V 2 @x 0
of the vapor at the origin. This can be obtained by means of expression [2.44] of
Fick’s law and the condition n 2 (0, t ) ns :
Dn wn 2
qn nV * >n 2V2 @x 0 [2.63]
n n S wx 0
The molar concentration n2 satisfies equation [2.62] and condition [2.63] with
which we can associate the conditions used in section 2.4.4.2.4 for positions 0 and A:
n 2 0, t ns n 2 A, t n 2a
in addition to the concentration distribution n2(x,0) at the initial time.
We will go no further with these calculations. Let us just recover the steady
regime studied earlier (section 2.4.4.2.4) by observing that in this case, gas 1 is
stationary; we therefore have, for all x:
qn nV * n 2V 2
Substituting this expression into [2.61] gives us:
n wn 2
n 2V 2 D
n n 2 wx
Equation [2.62] is thus identical to [2.53]:
w § n ª wn 2 º ·
¨ « D »¸ 0
wx ¨© n n 2 ¬ wx ¼ ¸¹
If evaporation occurs at constant volume in a container in which initially no

vapor is present, the pressure will increase during evaporation, which also produces
a displacement of the mixture. We will leave aside such static problems where the
equations of motion are to be solved.
2.4.4.2.7. Case of weak concentrations

When the mixture contains a weak proportion of species 2:
n 2 n1 or; p 2 p1 or; U 2 U1 ....
it is then possible to confuse the different units of concentration (section 2.4.1.3).
By comparing relations [2.28] and [2.29], we can see that the three reference
frames R*, RG and R1 become identical to the first order. We can note that
G *
> @
expressions [2.47], or [2.58] and [2.59], show that the diffusion velocities V1 or
> @
G G
V1 of component 1, which is excessively large, are negligible compared with the
G *
> @
diffusion velocity V 2 or V 2
G G
> @
of component 2.
It results in the preceding approximations that:

G
– the molecular diffusion velocity V 2 of species 2 is thus the same in any one of
the three reference frames R*, RG and R1;
any variable amongst those defined in section 2.4.1.3 (number of moles per
unit volume n2, molar or mass concentration c2 or ǹ2, partial pressure p2, molar or
mass fraction ǫ2 or ȁ2, etc.) can be used to represent the concentration c2 of
component G2 to the second order excepted.
Then we have, in the three reference frames previously chosen to characterize

the diffusion the relation:
G grad c 2
V2 D [2.64]
c2
In these conditions, we can define the flux density of species 2 by means of the
variable c2:
G G
qc2 c 2V 2 D grad c 2 [2.65]
The flux M c 2 of the variable c2 across the surface (S) related to any one of the
three mentioned reference frames is:
G
M c2 ³S q c 2 .ds ³S D grad c 2 .ds [2.66]
This approximation of law [2.41] in balance equation [2.8] allows the diffusion
equation to be obtained:
wc 2
wt
w §¨ wc 2
D
wx j ¨© wx j
·
¸ V
¸ 2
§ wc 2
¨¨
© wt
·
div D grad c 2 V 2 ¸¸
¹
[2.67]
¹
The volume source V2 of species 2 expresses, with the units of concentration c2,
the quantity per unit of time of this species created by a homogenous chemical
reaction.
For weak concentrations, we can usually assume that the coefficient of diffusion
D is constant; the diffusion equation is then reduced to the heat equation:
wc 2 w 2 c2 § wc 2 ·
D V2 ¨¨ D'c 2 V 2 ¸¸ [2.68]
wt wx j wx j © wt ¹
This approximation for weak concentrations is often used, either on account of

experimental conditions or because it allows for a simplified approximate treatment
of the problem, particularly in situations where it is associated and coupled with a
heat release and with chemical reactions. On the other hand, diffusion often occurs
in flows which present difficulties, and thus the preceding approximation is
necessary.
2.4.5. Coupled phenomena with diffusion
2.4.5.1. Binary non-isothermal mixtures

The elementary discussion of the last section is not entirely rigorous, even in the
absence of movement in the gas, because all phenomena characterized by scalar
quantities are coupled (sections 2.1.5.2 and 2.1.5.3.4): the diffusion of a species in
another species is accompanied by a thermal flux and a temperature gradient
(Dufour effect). However, this is very often negligible in the absence of thermal
conditions leading to a temperature gradient. Similarly, a temperature gradient
(differences in the energy of molecular agitation of diluted gases) will induce a
concentration gradient (i.e. a difference in the ways two different species of
molecule will move in a gaseous mixture, some moving to the cold regions, others
moving in the opposite direction).
The action of a mass force field (gravity, inertial forces, etc.) on the two species
is manifest in a static equilibrium comprising a pressure gradient (or of the volumic
number of moles), as we saw in section 2.2.1.5 (atmospheric equilibrium). In such a
situation, the lighter molecules tend to rise. The thermodynamic force responsible
for this effect is the pressure gradient. In practice, the diffusion of pressure is
entirely negligible in the atmosphere, where the quantity grad n is nearly always
negligible compared with grad n1 or grad n 2 ; diffusion therefore occurs as if there
was no force field. The diffusion of pressure must be accounted for in mixtures
subjected to high accelerations, as in centrifuges.
Another cause for the molecular migration of a chemical species with respect to
molecules of another species is a difference in the forces exerted between the two,
for example between ions with different charges placed in an electric field. This
thermodynamic force is a vector field, in other words of the same rank as the
gradient of a scalar quantity. These phenomena are encountered, for example, during
the electrolysis of solutions or in an ionized gas in the presence of an electric field.
Kinetic gas theory allows us to construct the theory of diffusive phenomena. The
principle involves seeking distribution functions for neighboring velocities of the
Maxwell-Boltzman distribution (Chapman-Enskog method). We find that first order
deviations from this equilibrium distribution lead to the irreversible processes
previously discussed. This theory also shows a decoupling between imbalances of
different tensor orders such as thermal, mass or viscous fluxes. We will not develop
the corresponding equations. For velocity differences in a gas ([PEN 55]) we obtain:
DnM 1 M 2
>VG1 @ >VG2 @ n
D 2 grad
n1
n1
n2

pU
grad p
[2.69]
n 2 D U1 U 2 G n2
G
F1 F2 DT
grad T
n1 n 2 pU n1 n 2 T
DT is the thermal diffusion coefficient; we usually define the thermal diffusion

ratio kT by the relation:
DT kT D
The thermal diffusion ratio kT is more or less independent of temperature, but

varies strongly with concentration: in particular, it tends to zero with each of the
concentrations. Its maximal value, attained when the concentrations of components
are in the same order of magnitude, is of the order of 0.1-0.2.
The preceding expression shows that a temperature gradient leads to

additional diffusion phenomena (thermal diffusion) which can be used to
separate two components of a gaseous mixture when chemical or other physical
methods are not practicable (for example, in gaseous mixtures comprising two
isotopes).
2.4.5.2. Mixtures with several components

Once a mixture is comprised of many components, the number of reduced
extensive variables increases, each component requiring a concentration variable.
There thus exist as many mass or molar flux densities as there are species present.
For each species, we have a balance equation [2.24].
Following the principle outlined in sections 1.4.2.6 and 2.1.5.2, all diffusive fluxes
depend on all of the thermodynamic forces of the same tensor rank. We thus have a
matrix of diffusion coefficients. The general discussion concerning the choice of
reference frames which characterize the diffusion processes is identical to that outlined
earlier.
As we have already said (section 2.4.2.3), the k balance equations for each
component leads to a global mixing equation that describes the conditions under
which the mixture will evolve (in movement, at rest in a fixed container, during
evaporation). The k equations are generally replaced by this global equation and k –
1 equations characterizing the components of the mixture.
The interested reader should consult textbooks covering problems of

irreversible thermodynamics ([BIR 02], [BOC 92], [CHA 91], [DEG 62], [DOU
01], [EU 92], [GER 94], [LEV 62], [PRI 68]).
2.4.6. Boundary conditions
In the absence of chemical reactions, the boundary conditions can be identical to

those of the thermal problem (section 2.3.2). The existence of a heterogenous
reaction on the wall P leads to the production or absorption of the components.
Chemical kinetics provides the law for the reaction speed for the components
concerned. The flux density of a component at the wall must be equal to that
produced or absorbed by the chemical reaction, for example for a reaction of order
m:
§ wc ·
D ¨¨ 1 ¸¸ kc1mP [2.70]
© wn ¹ P
The coefficient k is often an expression of the form k k 0 exp U RT , where

U is the activation energy of the reaction, T designating the absolute temperature.
The form of this relation shows a strong coupling between the temperature and the
reaction speed. The heat release due to heterogenous chemical reactions must be
taken into account in the boundary conditions of the energy equation ([BOR 00],
[PEN 55], [WIL 65]).
Chapter 3
Physics of Energetic Systems in Flow
In the preceding chapters, we examined the physical and mechanical properties

of matter independent of the dynamic effects induced by motion. Before dealing in
the next chapter with the general equations of fluid dynamics and the transfer of
quantities in flows, we will first recall the basic laws of mechanics and their role in
thermodynamics; we will then outline the formalism used to describe the motion of
continuous media and finally we will examine the mechanical properties of moving
fluids.
3.1. Dynamics of a material point
3.1.1. Galilean reference frames in traditional mechanics
As geometric space is homogenous and isotropic, the translational motion of an

isolated material point is necessarily rectilinear and uniform. In effect, for any other
kind of trajectory, a favored direction could be defined and any non-uniform
movement of an isolated material point would imply an inhomogenous time. We
thus postulate the existence of Galilean reference frames in which the distance
traveled by an isolated material particle is a linear function of time. The laws of
physics should be the same in all Galilean the reference frames. We must now
change the reference frame to where the transformation matrix is a function of time
and where all uniform translational movement are required to have the same
properties in the new reference frame. For Cartesian reference frames this results in
new coordinate systems which are in uniform rectilinear translation with respect to
one another, and which form a group.
In traditional mechanics, changes in the reference frame which conserve

distances and time belong to the Galilean group. They have the form:
x' i x i Vi t (i 1,2,3), t' t
The presence of the time variable in these reference frame changes leads to
specific properties of temporal derivatives. As time is the same in all Galilean
reference frames, the values of extensive scalar quantities are independent of the
Galilean reference frame used. On the other hand, components of vector or tensor
quantities vary in reference frame changes according to the usual formulae
(covariant or contravariant according to the case considered). We say that these
quantities are invariant for (geometric) changes of the Galilean reference frame (see
texts on linear or tensor algebra). However, certain vectors (position or velocity of a
particle) are defined with respect to a given Galilean reference frame; the evaluation
of their temporal derivatives depends on the reference frame chosen for this
definition. For example, the components in reference frame R’ of the velocity of a
point defined in reference frame R are not equal to the components in reference
frame R’ of the velocity of the point defined in reference frame R’. In a general
manner, so-called cinematic operations (such as calculations of temporal derivatives
velocity, acceleration) lead to formulae of changing reference frames dependent on
their relative motion. We will assume that these ideas are known to the reader.
Let us recall the following elementary formulae, which will subsequently prove
useful:
G G G G G G G G
Va Ve V r , Ja J e J r 2Z V r
in which the indices a and r indicate that derivatives with respect to time are
calculated in a Galilean reference frame (a) or in some other reference frame (r)
which is in motion relative to the former. In the expression of acceleration the first
term e represents the drag term (centripetal acceleration in the case of a rotating frame)
and the third term is the Coriolis acceleration.
In the following, where not indicated otherwise, momentum, velocity and

acceleration will be calculated in a Galilean reference frame.
3.1.2. Isolated mechanical system and momentum
We have seen in Chapter 2 that we characterize motionless matter by convenient

extensive properties and that the assumption of thermodynamic equilibrium leads to
the existence of relations between thermodynamic properties.
Physics of Energetic Systems in Flow 103
As in Chapter 1, we will begin by considering the simplest discrete system

having mechanical properties: the material particle or the material point. This choice
is made on account of the fact that a geometric point suffices to define its position.
A material system, like any thermodynamic system, is made up of many simple
systems, each comprised of an ensemble of material particles. These are described
by systems whose dimensions are sufficiently small.
The extensive quantity associated with the motion of a material particle is

momentum. The momentum of a material system is the sum of the momentum of its
components. This vector quantity is proportional to the extension of the particle.
The momentum of an isolated mechanical system remains constant.
3.1.3. Momentum and velocity
As space is homogenous, the momentum of an isolated system is a function

neither of its coordinates nor of time. Its expression must be identical in all Galilean
reference frames; this property will impose an expression for the momentum.
We must note firstly that the idea of entropy does not exist in the mechanics of a
particle. It only appears for systems in which the thermodynamic properties are
defined and in which mechanical energy is transformed into heat. We will come
back to this point a little later (section 3.2.5).
The intensive quantity associated with the momentum, for an elementary system,
G
is the velocity V , with components ui, which represents the intensity of motion in a
Galilean reference frame. Velocity depends on the reference frame used.
In a Galilean reference frame, momentum can only be a function of velocity in

that reference frame, and not of position or time. The relationship between
momentum and velocity can be derived from the preceding principles.
In effect, consider an isolated system composed of two identical particles. These

G G
particles, of constant initial velocities V1 and V 2 in a Galilean reference frame R,
interact through a collision (whose details are not important) such that after the
G
collision they have the same constant velocity V ' (a soft impact) in this same

G G
reference frame. If p V is the momentum of each particle, the momentum
conservation of the two particle system can be written in R as:

G G G G
p V1 p V 2
G G
2p V'
This equality must be true in all Galilean reference frames moving with constant
G G
velocity V ( V having any value) with respect to the reference frame R. The
G
following identity is derived, true for all V :
G G
G G G

G
p V1 V p V 2 V
G G G
2 p V 'V G
V
G G
G
This equality implies that the function p V is a linear function (take V
G G
successively equal to V1 and V ' , and then derive the identity

G G G G
p X pY
G G G

p X Y , characteristic of a linear relationship).
The momentum of a particle can thus be written in the form:
G G
pi mij u j p m.V
where m is a constant tensor which characterizes the material particle. As space is

isotropic, this tensor is necessarily spherical ( mij mG ij ).
A material particle is thus characterized by a scalar m (mass) which leads to the

relation:
G G
pi mG ij u j mi u i (p mV ) [3.1]
Similar reasoning using a Lorentz group for changing reference frames leads to a
different expression of mass in special relativity.
3.1.4. Definition of force
As momentum is an extensive quantity it must satisfy a balance equation. When

a material particle is subjected to an external action, its momentum changes. We call
G
this external action force f ; it constitutes a source of momentum for this particle,
G
dp
whose rate of change is:
dt
G G
dp
f [3.2]
dt
Forces are independent of the Galilean reference frame chosen for their
description (this is no longer true in relativistic mechanics).
The effect of a force during a time interval (t', t) is characterized by a vector

Gt
equal to the momentum variation p t ' between these two instants. This is, by
G
definition, the impulse I received by the particle:
G t G Gt
I ³t ' fdt p t' [3.3]
The idea of an impulse is useful in the study of collisions, which are often very
rapid events and the detailed modeling of which is replaced by more or less global
G
assumptions regarding their nature (soft impact, elastic, etc.). The external force f
G
dI
acting on a material particle is the impulse quantity received by the particle per
dt
unit time.
As for the other thermodynamic quantities, this definition is only of interest in so

far as we are able to explain the expression of this force as a function of external
parameters. As with the other thermodynamic quantities, we can distinguish:
– volume forces representing actions at a distance, which are due to force fields:
gravity, electromagnetic forces, etc.;
– contact forces, due to interactions on the microscopic scale between molecules,
ions, atoms, etc. As for other contact actions, these are modeled by surface forces:
stresses, pressure, viscous or dry friction, etc. We will come back to this point in the
sections covering continuous media.
NOTES –
1) A force is associated with the material element on which it acts: forces are
localized vectors (bi-points in mathematics) which can only be composed for each
application point considered separately, except in the case of a rigid solid.
2) Forces are actions which obviously exist independent of motion.
3.1.5. The fundamental law of dynamics (closed systems)
Consider a closed material particle, i.e. one that does not exchange matter with
the exterior; its mass is thus constant. Substituting [3.1] into [3.2] we obtain:
G
f

d mV
m
dV
[3.4]
dt dt
This equation, called the fundamental law of dynamics (or Newton’s second law
G G G
of motion), involves the acceleration Ȗ dV / dt ; the quantity mJ is called the
quantity of acceleration.
The quantity of acceleration is only the momentum’s time derivative for systems
of constant mass. A rocket which ejects a certain momentum in order to propel itself
is not a closed system. In fluid mechanics, we will often reason in terms of open
systems which exchange momentum with the exterior.
3.1.6. Kinetic energy
Every extensive quantity has a corresponding form of energy. The term for the
G
energy differential (kinetic energy Ec) associated with the extensive quantity p for
the preceding system’s motion is:
dE c u i dp i [3.5]
which, taking account of expression [3.1] for a particle of constant mass m, is:
§ mV 2 ·
dE c mu i du i d¨ ¸ [3.6]
¨ 2 ¸
© ¹
Kinetic energy can thus be expressed by the relation:
mV 2
Ec
2
The kinetic energy theorem can be obtained G by taking the scalar product of the
two sides of equation [3.2] with the velocity V :
dE c G dpG G G
V. f .V [3.7]
dt dt
G G G
The quantity f .V is the power of the force f in the reference frame considered.
As with the kinetic energy, it depends on the reference frame used for its evaluation.
3.2. Mechanical material system
3.2.1. Dynamic properties of a material system
A mechanical material system will be constructed, as for any thermodynamic

system, by decomposition into n sub-systems which are points (or nearly points) in
separate equilibriums (here at uniform velocity). The sub-systems interact amongst
themselves according to the principle of action and reaction which results from the
extensive nature of momentum. This quantity and the kinetic energy associated with
the movement for the whole system of points are additive:
G n H n m i Vi 2
p ¦ pi Ec ¦ [3.8]
i 1 i 1 2
The conservative nature of extensive quantities entails that total momentum

remains constant in an isolated system. The velocity, which is the corresponding
intensive variable, is of course not defined for the complete system if this one is not
in a uniform state.
G
Quantity p is called the total linear momentum or momentum of the system. We
G
also define the total angular momentum or angular momentum V O at a point O by
the relation:
G n G
VO ¦ OM i p i
i 1
Introducing the inertia center of the material system, and using formula [3.1],
relation [3.8] can be written:
G G
P mVG
Internal actions between components of the material system balance one another
two by two like all internal actions. Let us separate the forces acting on the particle
G G
Mi into external forces Fext i and internal forces Fint i representing interactions
between Mi and the other material particles of the system. Applying equation [3.2]
gives:
G G G
dp i
Fext i Fint i i 1,2,.....n [3.9]
dt
The n vector equations [3.9] enable a detailed study of the mechanical system.
By taking their sum, we obtain a vector equation which eliminates the internal forces
and describes the effect of external action on the sum of the quantities of particle
G G
acceleration mi J i dp i dt in the system, which is called dynamic
G G
resultant ¦ mi J i dP dt
i
G
dP G n G
mJ G ¦ Fext i [3.10]
dt i 1
G
Equations [3.10] state that the dynamic resultant dP dt is equal to the sum of
the external forces. Internal forces do not participate and the momentum of an
isolated system remains constant.
We proceed in similar fashion for the moments of the internal forces, which
G
cancel one another out, two by two; the moment C at a given point in the ensemble
of forces, naturally reduces to the moment of the external forces. The sum of the
G
moments of the quantities of acceleration d p i dt at O, also called dynamic moment
G G
G O at a point O, is equal to the sum M ext of the moments of the external forces
acting on the points Mi and considered at the same point.
We know that in taking moments, either at a fixed point O, or at the inertia

G
center of the system, the dynamic moment G O is equal to the time derivative of the
G
total angular moment V O , this last moment being the sum of the momentum
moments:
G G G
dV O n
GO ¦ M ext i [3.11]
dt i 1
The conservation of linear momentum for an isolated system is thus a

consequence of the properties of internal forces.
G G
Two particular vectors P and V O define an entity which replaces the detailed
distribution of localized momentums only in the case of rigid solids. The reader
should refer to texts on mechanics for a further study [AME 58].
3.2.2. Kinetic energy of a material system
3.2.2.1. Kinetic energy theorem

Kinetic energy theorem [3.7] is applied for each particle of the system. For the
material system it can be obtained by taking the sum of all the kinetic energy
equations corresponding to the particles Mi (obtained by taking the scalar product of
G
each equation [3.9] with Vi ):
dE c d ª n m i Vi2 º
«¦ » Pf int Pmec [3.12]
dt dt ¬«i 1 2 ¼»
¦ Vi .Fint i , the power of the internal forces;

n G G
with: Pf int
i 1
, the power of the external forces (mechanical power

n G G
Pmec ¦ Vi .Fext i
i 1
provided to the mechanical system from the exterior).
The power of the internal forces is not generally zero.
3.2.2.2. The power of internal forces

The power of internal forces is independent of the reference frame used for its
evaluation.
In order to show this, it suffices to regroup the internal forces by couples of

G G
opposed forces Fij and F ji interacting between the points Mi and Mj. The work Wij
of these forces is:
Wij
K K K K
Fij M j F ji M i
K K
K
Fij M j M i K
Fij .M i M j
The two vectors of the scalar product Wij are independent of the reference frame
chosen (Galilean or otherwise) and the scalar Wij takes on the same value in all
reference frames; the same is true for the total work done by the internal forces. As
time is identical in all reference frames, the power of the internal forces is thus
independent of the reference frame used.
Certain internal forces derive from a potential, such as Newtonian forces or

elastic forces; the same is not true for frictional forces.
3.2.2.3. The power of external forces

Let us consider a case where the external forces and certain internal forces
G
derive from a potential U, the other internal forces being designated by F f int i .
Equation [3.12] can be written:
d ª n §¨ miVi2 ·º n G G
«¦ U Mi ¸» ¦ Vi .F f int i [3.13]
dt «¬i 1¨© 2 ¸»
¹¼ i 1
We define the mechanical energy Em of the system by the relation:
n § m i Vi2 ·
Em ¦ ¨¨ U M i ¸
i 1© 2
¸
¹
dE m
Pf int [3.14]
dt
The total mechanical energy of the isolated material system is not conserved
because the internal forces provide a non-zero power, except in certain particular
cases (rigid, solid, etc.).
Consider the following example: two particles of the same mass m are elastically
linked with friction; they constitute an oscillator where each of their opposed
elongations are centered around the inertia center G.
G
m G V
m
f x
x1 x2
A
Figure 3.1. Isolated oscillator
We assume that there are no external forces. The kinetic energy of the whole is
the kinetic energy of the inertia center animated by the constant velocity VG, which
is increased by the kinetic energy of the oscillator in a reference frame related to the
inertia center G. The total mechanical energy can be obtained by adding the
potential energy of the oscillator. This decreases on account of the friction and is
transformed into heat, thus increasing the internal energy by heat addition. On the
other hand, the momentum of the ensemble has not changed.
The power of the internal forces depends only on the relative particle velocity
and not on the chosen reference frame. On account of the internal friction of the
system, it decays as well as the total mechanical energy of the system.
3.2.3. Mechanical system in thermodynamic equilibrium: the rigid solid
We will now examine how motion intervenes in thermostatics and

thermodynamics. Consider a material system, assumed to be in movement and
isolated, in other words not surrounded by any other material system which exerts an
action on it; it will finish by being “rigidified”: the relative movements of its
different constituents disappear under the effect of internal dissipative forces
(viscous damping or internal friction, exchange of momentum and collisions
between the particles comprising the system, etc.). In other terms, the relative
distances between the different material particles making up the system remain
constant. The mechanical system in thermodynamic equilibrium has become a rigid
system in which there are no longer any entropy sources associated with friction.
The system is in a state of thermostatic equilibrium if its other extensive quantities,
defining its internal state, are no longer susceptible to change.
We obtain the same final state for a material system placed in a force field
derived from a potential. The fluids contained in such a material system are thus in
hydrostatic equilibrium. We have already discussed, in section 2.2.1.2, the restrictive
conditions necessary for the existence of such a situation. However, if the external
force field does not derive from a potential, the exterior will always be able to
communicate an energy to compensate the friction to the fluid (by an

electromagnetic action in a conducting fluid, for example).
The rigid solid thus plays an important role from the point of view of
thermostatics, since it is the ultimate state (thermostatic equilibrium) towards which
all isolated mechanical systems will tend. It is furthermore what is observed for a
gaseous or liquid mass enclosed in a container which is not subject to any external
action; the same is true for all collections of solids (a pile of sand, etc.), including
solar bodies: the moon no longer rotates, always showing the same side to the Earth;
the daily rotation of the Earth slows continually due to diverse dissipative effects
(mostly tidal movements of the oceans, which lead to a dissipation of energy on
account of a gravitational Earth-sun and Earth-moon interactions).
We know that the momenta of a rigid solid are characterized by linear

momentum and angular momentum, which are constant for an isolated rigid solid.
Recall that its motion is not characterized by an instantaneous constant velocity
vector of rotation; (this motion is a Poinsot motion around its inertia center (see texts
on mechanics [AME 58])).
3.2.4. The open mechanical system

We can also perform a balance for an open material system whose momentum
varies because of an input of matter (for example a water bomber collecting water)
or an output of matter (the propulsion of a rocket). In fluid mechanics, the balances
of extensive quantities, including momentum, are performed for open systems.
However, the physical conservation laws for extensive quantities are applied to a
given material quantity, i.e. to closed systems. The balance equations for open
systems must obviously take account of the change of definition in the system where
the balance is performed (section 1.1.4.2), i.e. take into account the input or output
of extensive quantities associated with the input or loss of matter.
3.2.4.2. Momentum of a rocket

As an example, consider the vertical motion of a rocket in a gravitational field.
dm
Its mass m(t) decreases with the ejection of a constant mass flow of gas q m ,
dt
at a velocity U with respect to the rocket.
Let m(t) be the rocket mass at instant t. We consider a fixed reference frame Oz
and perform an elementary momentum balance between time t and t + dt on the
mass m comprised by the rocket at time t.
Figure 3.2. Momentum balance: (a) closed system; rocket + gas,

(b) open system; rocket alone
At the instant t + dt the rocket mass has changed from m to m + dm, whereas the
G
velocity in the fixed reference frame has increased from V to V dV , and the mass
–dm of gas has been ejected. The mass ejected, initially moving at a velocity V has
G G
now changed to V U at time t + dt in the fixed reference frame. Between these
two instants, the momentum variation of the mass m can be calculated as:
m dm V dV dmV U mV
G G G G G G G G
dp m dV U dm [3.15]
Equality [3.15] highlights:

G
the change in momentum dp f
G

d mV of the rocket (of variable mass m):
m dm V dV mV
G G G G G
dp f d mV
G
the momentum d p g of the mass –dm ejected from the rocket:
G G
dp g dm V U
We have, per unit time:
G
dp dm VG UG
G
d mV
G
dp f

G
dp g
m
G
dV G dm
U [3.16]
dt dt dt dt dt dt dt
G
The time derivative of the momentum p f of the rocket is not equal to the
G
dV
quantity of acceleration m .
dt
G
The quantity dp g dt is associated with a matter which is constantly being
generated at the exit of the rocket exhaust: it is a momentum flux.
In order to simplify matters, we will consider that the only external force is
constituted by gravity; taking account of [3.16], equation [3.10] for the momentum
at constant mass m can be written:
G G
dp dV G dm G
m U mg
dt dt dt
or:
G
dV G G
m U q m mg [3.17]
dt
Equation [3.17] shows that we can apply the fundamental dynamic relation to a
system of variable mass by considering the quantity of acceleration for that system,
the ejection of burned gases generating a force on the rocket opposed to the
G
momentum flux Uq m (relative to the rocket) generated by the propulsion system.
We can verify that the opposite of this force, acting for a time dt, corresponds to
G
the impulsion U q m dt necessary to give to the mass q m dt ejected in time dt a
G
variation in velocity equal to U . This force is the rocket thrust .
The preceding balance can also be written in the non-Galilean reference frame of
dV
the rocket (whose acceleration is ). In this reference frame, the rocket ejects
dt
G
momentum Uq m per unit time (momentum flux) which balances the inertial force
G
dV G
m and the gravitational force mg .
dt
A momentum flux is associated with the existence of a force. This observation is

the basis of many concepts in fluid mechanics. We will see it again when we study
the momentum flux theorem.
The preceding results can be extended to moments (of momentum, or of

quantities of acceleration) taken about a fixed point O or about the inertia center of
the system.
3.2.4.3. Kinetic energy balance

The variation of kinetic energy dEc of a system comprising mass m can be
calculated in a similar manner to that used to calculate the momentum. We have:
G
d § V 2 ·¸ ª1 G G V2º d §¨ V 2 ·¸ ªG G U2º
dEc
m ¨
dt ¨© 2 ¸¹
qm « V U
2
» m
dt ¨© 2 ¸¹
q m «V .U »
dt «¬ 2 2 »¼ «¬ 2 »¼
G
By taking the scalar product of equation [3.17] with the velocity V we find the
kinetic energy equation of the rocket:
d §¨ V 2 ·¸ G G G G
m U .V q m mg .V
dt ¨© 2 ¸¹
This gives, for the kinetic energy variation dE c :
dE c U2 G G
qm mg .V [3.18]
dt 2
G G
The term mg .V is the power of the gravitational force. The power Pfint of the
U2
internal forces is here equal to q m which represents the quantity of kinetic
2
energy per unit time created by combustion in the fixed reference frame of the
rocket where chemical energy (reaction enthalpy) is transformed into mechanical
energy.
This preceding reasoning will be used in a more general manner in Chapter 4,

where the balance equations in a moving fluid will be established. We will use
Eulerian variables in order to clearly explain the phenomena associated with the
concept of convection.
3.2.5. Thermodynamics of a system in motion

3.2.5.1. Energy equation
Consider a material system, described by means of the variables of a certain
number p of sub-systems. The internal energy of each component is the energy of its
matter in a reference frame in which it is at rest. The differential [3.5] dE of the
internal energy can be expressed as a function of the extensive quantities Xi and the
intensive quantities Yi previously defined:
p
dE ¦ Ti dS i Yi dX i
i 1
Now suppose that the system components are in motion. The total energy
contained in the system must take account of all energy forms, including its kinetic
energy Ec. In other words, the mechanical properties constitute just one particular
aspect of the physical phenomena to be considered. We denote as Et this “total”
energy in order to distinguish it from the internal energy, denoted E:
Et E Ec
The conservation principle is thus applied, not to the internal energy alone, but to
the total energy, whose variation 'E t is equal to the quantities of heat Q and of
work W received from the exterior, including the work Wmec received by the system
and coming from the external forces:
'E t 'E 'E c W Q
The extensive and intensive scalar variables, and in particular the internal
energy, do not depend on the reference frame chosen: for example, the tension of an
elastic string and its stretching, the pressure of a gas and its volume, the number of
moles of a chemical species and its chemical potential, etc. Interpretations of
internal energy in diverse physical theories are often made using forces whose
details are of little importance, but as these are internal forces, their resultant
moments are zero; we have seen (section 3.2.2.2) that the power Pfint of the internal
forces of a system (or the corresponding work) is independent of the reference frame
chosen for its evaluation. This power of the internal forces is in fact already
described in the differential dE of the internal energy (work done by pressure forces,
etc.).
The kinetic energy and the total energy of a system depends on the reference
frame chosen for their evaluation.
For a system which receives from exterior sources of a purely mechanical kind,
we thus have:
d
( E + Ec ) Pmec + Pth [3.19]
dt
Pmec and Pth being the mechanical and thermal powers provided to the system from
the exterior.
Consider first a mechanical system comprised of a material “particle” which is

animated, for the sake of simplicity, by a macroscopic translational motion of
velocity V. Now suppose that this particle receives, from the exterior, the
G G
mechanical power Pmec V .F and the thermal power Pth. The balance equation for
the (total) energy of the material system can be written:
d §¨ mV 2 ·¸
E Pth Pmec
dt ¨© 2 ¸¹
G
Now consider a system of n material particles of velocities Vk . Letting E denote
the internal energy of the system (which is not in equilibrium), the balance equation
can be written, taking account of motion, as:
d §¨ n mk V 2
k
·
¸
E ¦ Pth Pmec [3.20]
dt ¨ 2 ¸
© k 1 ¹
Subtracting the kinetic energy equation [3.12] from [3.20], we obtain the energy
equation (E, the internal energy) of the material system:
dE
Pth Pf int [3.21]
dt
The rate of internal energy variation of a system is equal to the difference

between the thermal power which it has received and the power of its internal forces.
3.2.5.2. The entropy form of the energy equation

It is useful to explain the preceding phenomena using an example. Consider the
system already discussed in section 3.2.2.3 (Figure 3.1), comprised of two masses
linked by elastic under tension W A and equipped with a viscous friction system.
The tension force W derives from a potential.
G
The masses are subjected to external forces F j (j=1,2); we assume that gravity
does not directly intervene. The viscous friction system is assumed to have no mass
and the heat released by the friction heats the two masses which are assumed to be at
the same temperature. The dynamic equation of motion for the masses in the
reference frame of the inertia center can be written:
G G G G G G G G
mJ 1 Fext1 F f W mJ 2 Fext2 F f W
in which the tension W A of the elastic link and the viscous friction force F f ,
aligned with the axis Ox, are, by definition, the forces applied on each mass:
d x 2 x1
Ff f f A
dt
By adding the terms of the two equations after multiplication of each by its
G
corresponding velocity V k k 1,2 , we obtain the kinetic energy equation. The
power of the internal forces can be immediately calculated:
W F f V1 V2
G G
Pf int W A f A 2 [3.22]
Energy equation [3.21] can be written:
dE
dt
W A f A P
2
th [3.23]
By specifying the differential of the internal energy dE TdS W dA for the

system, we finally obtain the energy balance equation in its entropic form:
dS
T f A 2 Pth [3.24]
dt
We see that in this entropic form of the energy equation the mechanical terms
have been eliminated and all that remains is the irreversible part of the motion
which is transformed into heat. The dissipation function fx 2 is a heat source which
is always positive. It appears in equation [3.24] as an internal entropy source.
The preceding developments can be repeated for a system of n particles at the

same temperature, between which there exist elastic forces and friction forces. We
thus obtain an entropic form of the energy equation, analogous to [3.24], with a
dissipation function for all the internal frictions of the system. The reader will note
that this reasoning is only valid if the entropy of the system studied is a function of
the other extensive quantities, i.e. if the system is in a state of thermostatic

equilibrium at every instant.
We will see these developments (sections 3.1 and 3.2) once again when we study
flowing continuous media, albeit with a formalism which is naturally more complex.
3.3. Kinematics of continuous media
3.3.1. Lagrangian and Eulerian variables
A problem of flowing matter involves the study of the influences of external

conditions on the fluid medium, specifically the many different ways forces exerted
on the medium (presence of fixed boundaries with respect to the observer, moving
boundaries such as propellers blades, turbine blades, etc., pressure differences
between two reservoirs, external forces such as gravity and electrical volume forces
due to some external device, etc.). The properties of a flow thus result from the
action of external causes which modify the mechanical and physical properties of
the matter (velocity and acceleration, internal stresses, pressure, temperature,
chemical composition, etc.) and inversely the flowing medium exerts stresses on
walls or modifies the boundary properties (stresses induced, temperature, chemical
properties, etc.).
The questions posed and the results expected from a study, an experiment or the
operation of a device or system can vary considerably. Here are some examples:
– in a water treatment station or in a chemical reactor, we are naturally
concerned with the product being treated; the practical problem is thus to design
walls, materials and diverse processes so as to obtain the desired result concerning
the product which is treated;
– in meteorological applications, the objective is to predict the weather at a given
place and time, i.e. to predict the motion, and the physical and chemical properties
of large air masses (velocity, temperature, humidity, presence of pollutants, etc.);
– for a boat, a plane or a vehicle in motion, the essential properties are the
external forces exerted on the solid boundaries in contact with the flow when the
vehicle is displaced at a given velocity. The same is true for the flow of a river or
fluids in an industrial pipe network for which suitable constructions are necessary
(dams, turning vanes, pipes, open channels, pumps, turbines, etc.).
The study of a physical problem begins with a definition of its variables. The
description of moving matter involves the localization of a material particle which is
identifiable in a given reference frame by means of coordinates. The physical
properties of this particle are associated with it. In particle mechanics, we give
numbers to the different particles studied and to the physical quantities which
characterize them. We thus define the trajectory of a particle as an ensemble of the
successive positions which it occupies over the course of time.
When studying the physics and mechanics of continuous media, this means of
description is no longer suitably adapted, since the set of the material particles of a
continuous medium is not denumerable. We thus need to identify them with respect
to a continuous ensemble which is simply the ensemble of positions xi0 which this
material occupies at a given reference time. The variables g associated with the
matter are thus functions of this position xi0 occupied at an initial instant and of time:
g g ( xi 0 , t )
This is called material representation by means of Lagrangian variables.

Problems for moving continuous media where the matter must be clearly
individualized are generally treated using Lagrangian variables; this description is
often used in the mechanics and physics of solids.
The disadvantage of Lagrangian variables is that such a description does not

have any particular interest for the observer in many situations: the fisherman at the
water’s edge sees the passage of water particles which he will never again see, and
the same is true for the plumber examining a radiator, etc. As the fluid matter passes
by continuously, the identification of fluid particles is not particularly useful. This is
not the case however when we are interested in identifying the local physico-
chemical properties of the matter. Examples include the progress of a chemical
reaction in a closed reactor, the presence of clouds and the knowledge of humid air
masses in weather applications or the presence and movement of pollutants
downstream of a pollution source, etc.
Frequently, in fluid mechanics we consider a preferred observer reference frame,

which is most often associated with solid boundaries between which or in the
vicinity of which the fluid matter passes (pipe networks, solid obstacles, the wings
of an aircraft, etc.). We thus represent the matter spatially by means of a field where
quantities associated with a material particle are functions of the coordinates xi
which the particles occupy at the observation time t. This kind of representation
involves Eulerian variables.
In physics, the study of particles contained in and subject to a force field

generally involves the combination of these two representations, the particle being
described by its coordinates as a function of time (Lagrangian variables) whereas the
external field (gravity, etc.) is defined using its spatial properties.
Fluids in motion are generally described using Eulerian variables, the observer
reference frame being associated with the solid boundaries. A considerable difficulty
thus appears, as the formulation of the laws of thermodynamics and mechanics
requires the balance of extensive quantities associated with matter. In the following
sections we will write these balances by “following matter” over material domains
represented using Eulerian variables.
3.3.2. Trajectories, streamlines, streaklines
In Eulerian variables the kinematics of a flow is characterized by the velocity

field, which is a function of time:
G G G
ui i, j
ui x j ,t 1, 2,3 or V V M ,t .
In a velocity field, we can define the following curve families:
the streamlines at time t are lines whose tangents are the velocity vectors at
this instant.
These give a vision of the flow at instant t and their differential equations are:
dx1 dx 2 dx3
[3.25]
u1 ( xi , t ) u 2 ( xi , t ) u 3 ( xi , t )
a trajectory is the locus of a fluid particle; trajectories are characterized by

their spatial coordinates which are a function of time (Lagrangian description); their
differential equations (here t is a variable) are:
dx1 dx 2 dx3
dt [3.26]
u1 ( x i , t ) u 2 ( xi , t ) u 3 ( xi , t )
a streakline at time t is the locus of particles which have passed by a given

point earlier than t, such as the cloud of smoke issuing from a chimney or the
colored line obtained in a moving fluid by injection of a colored liquid or a trickle of
smoke. The process of flow visualization involves the generation of visible emission
lines (Figure 3.3) by injection of markers (dyes, smoke, small particles).
fluctuating wind
trajectory
streakline
Figure 3.3. Trajectory of a smoke particle and

streakline from a chimney when wind is blowing
When a velocity field does not explicitly depend on the variable t,

u i x j , t
u i x j and we have a steady flow. The streamlines, which no longer
depend on time, are fixed in space: the appearance of the flow does not change. The
preceding definitions and equations show that streamlines, trajectories and
streaklines are all the same in steady flows.
EXERCISE – Find the equations for the streamlines, the trajectories and the
streaklines of the following 2D velocity field: u = constant, v a cos Zt t. Interpret
the results.
3.3.3. Material (or Lagrangian) derivative
3.3.3.1. Material (or Lagrangian) derivative of a quantity g

Let g x i , t be the field of a continuous scalar quantity associated with a

medium of fluid particles moving with a velocity field u i x j , t defined in a given
reference frame. The material derivative (also called the total or Lagrangian or
substantial derivative) is defined as the time derivative of the quantity g associated

with the material particle of coordinates x j t and velocity u i x j , t . It can be
obtained by calculating the time derivative of g in which the coordinates are
functions of time and describe the trajectory:
dg wg wg dg wg
uj or V .grad g
dt wt wx j dt wt
d
We will denote the Lagrangian derivative , which is by its very definition the
dt
D
derivative of a function of time only for a fluid particle. It is also often denoted
Dt
in fluid mechanics.
In particular, for components of the acceleration, we have:

G G
dui wui wui G dV wV G G
Ji uj or J §¨ grad V ·¸ V [3.27]
dt wt wx j dt wt © ¹
Acceleration [3.27] can also be written in the form:

G G
G dV wV V2 G
J grad 2Z V [3.28]
dt wt 2
G 1 G G
where Z ro t V is the rotation vector. The vorticity vector is twice the rotation
2
vector.
dg
NOTE – The material derivative , defined in a Galilean reference frame, is a
dt
physical quantity which does not depend on the reference frame chosen for its
wg
evaluation; on the contrary, the temporal derivative of the spatial representation
wt
corresponds to an observation of the quantity g at a fixed point and it does depend
on the reference frame chosen to represent the field. Consider two Cartesian
reference frames of coordinates x j , t and ~ ~
xj,t :
~ ~
xj x j U j t; t t.
The reader can easily verify the relation:
wg wg~ wg~
~ U j ~
wt wt wx j
g~ ~
xi , t .
~
where we let g xi , t
The time derivative calculated in a reference frame depends on the choice of

reference frame. Its value is notably associated with spatial inhomogenities of the
quantity considered and which is displaced with respect to this reference frame.
For example, the reader can verify that the field of the quantity g, steady in the
reference frame x1 , t and with spatial period O
§ 2S x1 ·
g ( x1 , t ) A cos¨¨ ¸¸
© O ¹
is seen by a fixed observer in the reference frame ~ >

~
x1 x1 Ut ; t t as an @
~
oscillating field with temporal period T O U . We will come back to this
phenomenon, which is known as the Doppler-Fizeau effect in acoustics.
3.3.3.2. Flux of the quantity G

We saw (section 2.1.3.1) that the transfer of an extensive quantity G is
G
characterized by a flux density vector qG whose flux across a surface is equal to the
amount of the quantity G which crosses this surface per unit time. The
demonstration of the existence of this vector is independent of the transfer
mechanisms considered. In particular, the movement of matter with respect to a
surface leads to a flux of the quantities associated with the matter across this surface,
which we often refer to as a convective flux of the quantity G.
G
n
G
V
ds
G
VGt
Figure 3.4. Balance in a flow through ds

for an elementary displacement
When a quantity G is transported by moving matter, the quantity G which

G
crosses Ga surface ds of unit normal n in time Gt occupies an oblique cylinder of
length VG t (Figure 3.4). We thus derive an expression for the flux density vector of
G:
G G
q g gV (g: quantity of G per volume unit) [3.29]
The convective flux of the quantity G crossing a surface S per unit time is thus
equal to:
MG
G G
³S q g .n ds
GG
³S gV .n ds M G ³6 gu j n j ds [3.30]
The volume and mass transported by the Gmatter’s motion are respectively
characterized
G by the volume flux density vector V and the mass flux density vector
U V which correspond, respectively, to g 1 and g U .
The elementary volume flux dqv and the elementary mass flux dqm are1:
GG GG
dq v V .n ds and dq m U V .n ds
The volume flux qv (or volume flow rate) and the mass flux qm (or mass flow
rate) across a surface S can be written:
GG GG
qv ³S V .n ds and qm ³S U V .n ds
G G
When the quantity G is a vector (for example the momentum m V ), the
preceding results can simply be applied to each of the components g i . We thus
G
define a flux density tensor q G of the quantity G by the relation:
G G
qGij giu j or qG g V [3.31]
G G
The flux of the quantity G crossing the surface S is thus the vector M GS :
G G
M GS ³S qG nds
3.3.3.3. Material derivative of a volume integral

Let g ( xi , t ) Ug ( xi , t ) be the volume density of the quantity G (g designates
the corresponding mass density). The amount of the quantity G contained in a
domain D of fluid inside the closed surface 6 is:
G ³D g ( xi , t ) dv
The laws of physics imply the balance of the extensive quantities (mass,
momentum, etc.) associated with the matter which, in a flow, is in motion. It is
therefore necessary to calculate the variation of the quantity G associatedG with a
material domain D in motion using the usual Eulerian representation. n designates
the unit normal to the surface 6 directed towards the outside of the fluid domain D.
1 We shall use the usual scalar notation qv and qm for volume or mass fluxes which evidently
are fluxes and not flux densities; confusion is not possible, flux densities being at least
vectors.
The extensive quantities are associated with the matter and the balance of the
quantity G must be performed on a material domain. Consider (Figure 3.5a) the
domain Dt occupied by the matter at instant t, of external surface 6, and then the
domain Dt G t occupied by the matter at time t G t . Between these two
instants, the material domain is displaced such that it leaves 6 over the section 6
whereas it is displaced towards the interior of Dt over the surface 6 (Figure
3.5b); let D0 be the volume common to Dt and Dt G t . Let G D1 and G D2 be
the additional parts of D0 in D(t) and Dt G t generated by the surfaces 61 and 62
G
and the vector displacement VG t . The (positive) amounts of the quantity G
GG
contained in each of these domains are, respectively, ³6 gV .nGtds and
1
GG
³6 gV .n Gtds .
2
G GD1 G
n nG
V
x3 D ds
GD2 6 1
O 6 6 2
D0
x2
x1 G
n
(a) (b)
Figure 3.5. Balance of an extensive quantity in a flow:

(a) domain at one instant; (b) displacement of the material domain
The variation GG of the quantity G contained in D between the instants t and

t G t is thus:
GG G (t G t ) G (t )
GG
GG
³D0 >g (t G t ) g (t )@dv ³61 gV .nGtds ³6 2 gV .nGtds
By regrouping the surface integrals, taking the limit as G t o 0 , and applying
Ostrogradski’s theorem, we obtain:
dG
³Dq
wg
wt
GG
dv ³6 gV .n ds
§ wg
³Dq ¨¨
G·
div gV ¸¸dv [3.32]
dt © wt ¹
or, using notation with indices:
dG wg § wg w gu i ·¸
³Dq dv ³6 gu i ni ds ³Dq ¨¨ dv [3.33]
dt wt © wt wxi ¸¹
dG
The notation in equation [3.32] or [3.33] indicates a Lagrangian derivative.
dt
Formula [3.33] shows that the flux of the quantity G across the surface 6 can be
interpreted as due to the integration over the volume of the local volume source
G
div g V .
Material derivatives can also be expressed as a function of the mass density g:
dG wg w
³Dq dv ³6 gdq v ³Dq U g dv ³6 gdq m
dt wt wt
Let us take the elementary case where g 1 . The quantity G is the volume of
the domain D. For a small domain of volume G v , we have:
d G v G
divV .G v
dt
G
The quantity divV is thus the local velocity of the volume expansion
1 d G v
.
Gv dt
G G
If the quantity G is a vector (for example the volume momentum U V ), the
preceding steps can simply be applied to each of the components gi. We thereby
G
define a tensor flux of the quantity G by the relation:
G G
qGij giu j or qG g V , (with : Cartesian product).
NOTE – A more mathematical demonstration of expression [3.33] could be

performed by using the fact that the domain Dt G t can be derived from the
domain Dt by means of a one-to-one geometric transformation which is precisely
defined by the displacement of fluid particles between these instants ([ZIL 06]).
3.3.3.4. Material derivative of a flux integral

GG G
Consider now the integral M S ³S B.nds , in which the vector B and the surface
S are attached to the moving matter. The surface S(t) moves to position S t G t ,
which we will call S1 at time t G t , having thus swept the surface SA (Figure 3.6).
The flux variation M S between these two instants is:
G G G G
GM S M S (t G t ) M S (t ) ³S1 B(t G t ).nds ³S B(t ).nds [3.34]
G
Consider the domain D bounded by the surface 6 S1 S S A ; dW is the
elementary arc of the curve C which encloses the surface S; we take the normal in
G G G G G
the direction V dW , on the lateral surface element VG t dW swept out by dW
during time G t .
G G G
VG t n B
G
ds V
C S S1
G
dĲ SA
Figure 3.6. Material derivative of a flux integral
G
The flux of B leaving the closed surface 6 at time t can be written:
³S1 Bt .nds ³S Bt .nds ³C G tB.V dW

G G G G G G G G G G
³6 Bt .nds ³D divBt dv
Taking account of this expression, flux variation [3.34] can be written:

G G G G
GM S ³S1 B(t G t ).nds ³S1 B(t ).nds
G G G G
³D divBt dv ³C G tB. V dW
Applying Ostrogradski’s theorem, and letting G t tend to zero, we obtain the
material derivative of the flux M S crossing S:
G
dM S wB G G G
G GG
G
³S .nds ³C B. V dW ³S V .n divBds
dt wt
which is, after the application of Stokes’ theorem:

G
§w B
d MS
¨
³S ¨
G
ro t BG VG VGdivBG ·¸¸ . nG ds [3.35]
dt ©wt ¹
G G
If the field B is conservative ( divB 0 ), the total derivative of the flux M S can
be written:
G
§w B
d MS
³S ¨¨
G
ro t BG VG ·¸¸ . nG ds [3.36]
dt ©wt ¹
3.3.3.5. Unsteady and quasi-steady flows

In Eulerian variables, a flow (or the transfer of a quantity G) is considered
unsteady if the time t appears as a variable in the description of the flow:
sui / st z 0 or sg / s t z 0
In Lagrangian variables, the idea of a flow’s unsteadiness does not have any
direct meaning.
If the term wu i / w t (respectively wg / w t ) is very small and negligible compared

with the other terms involving the corresponding material derivative, the flow
(respectively the transfer of quantity G) is quasi-steady and the differential character
with respect to time no longer features in the balances. The time t thus becomes a
parameter, and the unsteady properties of the problem studied are those of a steady
flow (or of the transfer of the quantity G). The physical causes of variations of the
quantity G in the matter are thus solely balanced by convective fluxes.
Following the note made in section 3.3.3.1, the flow can only be steady in a
given, particular reference frame.
3.3.4. Deformation rate tensors
The determination of the stresses undergone by the matter in a flow is related to

our understanding of the manner in which the matter is deformed. Any deformation
of a solid requires the action of a stress. A fluid which does not have any particular
shape or form can be deformed easily without stresses, on condition that the
corresponding action occurs slowly.
In a fluid stresses are thus associated with the structure of the velocity field.
However, velocity fields corresponding to changes of reference frame, or to
displacements, will not be associated with stresses.
The study of the local structure of a velocity field at any time is made by
G G G
considering two neighboring points M and M dM at the respective coordinates xi

and x i dxi (Figure 3.7) whose velocities are u i x j and u i x j dx j .
x3 G G
M dM
G G G
V M dM
M
G G G
V M
x1 x2
Figure 3.7. Velocity field at two neighboring points
The study of the velocities at these two points will allow the nature of the motion
of the matter to be determined. We perform a Taylor expansion about the point
G
M, at coordinates xj:
wu i
ui ( x j h j ) ui ( x j ) h j o( h)
wx j
We can separate the symmetric and anti-symmetric components of the

tensor wu i wx j :
sui 1 su suj ¬ su ¬
; %ij 1 sui j
:ij %ij with: :ij i [3.37]
sxj 2 sxj sxi ® 2 sxj sxi ®
The anti-symmetric part : ij h j of the velocity field dui can be written:
§ 0 : 21 :13 ·§ h1 · § Z 2 h3 Z 3 h2 ·
¨ ¸¨ ¸ ¨ ¸
: ij h j ¨ : 21 0 : 32 ¸¨ h2 ¸ ¨ Z 3 h1 Z1 h3 ¸
¨ :
© 13 : 32 0 ¸¹¨© h3 ¸¹ ¨Z h Z h ¸
© 1 2 2 1¹
with:
Z1 : 32 Z2 :13 Z3 : 21
G G
The preceding expression : ij h j , written in vector form Z h , expresses the
K
velocity field due to the instantaneous rotational velocity vector Z :
G 1 G
Z rotV [3.38]
2
The remaining symmetric tensor H ij characterizes the deformation rates (strain

rates), as the displacement velocities have already been accounted for.
We can easily verify that the diagonal terms of the tensor H ij are expansion (or
compression) rates, the other terms corresponding to shear (or angular) expansion
rates. By considering for example the velocity field >u1 x 2 , u 2 u 3 0@ we see
that the deformation generated comprises a transverse motion of the segment MM'
(Figure 3.8a) with respect to the velocity.
If the segment MM' is parallel to the velocity ( u1 x1 , u 2 u3 0 ), the

segment MM' is expanded or compressed according to the sign of du1 (Figure
3.8b).
x2
x2+dx2 M' u1+du1
x2 u1 M u1 M' u1+du1
M x1
x1 x1+dx1 x1
(a): wu wx z 0 (b): wu1 wx1 z 0

1 2
Figure 3.8. Deformation rates: (a) shear su1 / sx2 z 0;

(b) expansion wu1 wx1 z 0
G
The volume expansion rate divV (section 3.3.3.3) is furthermore always the
trace of the matrix Hij, i.e. the sum of the linear expansion rates following the three
axes. We will leave it to the reader to verify this.
3.4. Phenomenological laws of viscosity
3.4.1. Definition of a fluid

We saw in Chapter 2 that the stresses in a continuous medium can be represented
by a tensor V ij . A fluid particle undergoing a bulk movement (translation and
rotation) is at rest in a moving Cartesian reference frame. It is not subject to stresses
other than those induced by pressure forces. On the other hand, this is no longer the
case in the presence of strain. The internal stresses in this continuous medium do not
depend on the relative position of the fluid particles, but on their relative velocities:
an infinitely slow fluid movement will not generate any stresses, contrary to what
occurs in solid bodies. This distinction between fluids and solids does not always
exist: certain bodies can have the properties of elastic solids for motions which
occur at the scale of seconds or fractions of seconds, and their shape may be
changed by a flow at the scale of many hours or many days (e.g. viscoelasticity,
creep in solids).
3.4.1.2. Viscous fluids

A viscous fluid is one in which the stresses at a given instant are a function only
of the deformation rates at that instant. We will separate pressure stresses and the
viscous stress tensor Wij using the relation:
ij pij ij (Kronecker symbol : ij 0 si i v j, ii 1) [3.39]
This definition corresponds with the idea of a fluid as defined in fluid statics
(section 2.2.1.1). In keeping with what has already been said, the viscous stress
tensor depends only on the strain rates, and not the deformations. There exist “visco-
elastic” bodies in which these two kinds of stress generation can co-exist. We will
not cover the more complex cases ([FRE 64], [GER 94], [TAN 00], [COI 97]),
limiting our attention to the study of viscous fluids.
As in other domains of physics, the “relationship” between viscous stresses and

strain rates become complex when the structure is complex at the molecular level. In
particular, the flow may provoke changes in the “molecular cohesion”, or preferred
orientations in the presence of macromolecules. The establishment of a flow in a
fluid (gas or liquid) containing solid particles which are more or less dispersed may
lead to their becoming suspended. For example, the wind can carry sand in the
desert or on beaches, in many industrial applications powders are often transported
using fluidized beds (airflows which are sufficiently energetic to raise and transport
solid particles, these particles being deposed when the fluid velocity drops
sufficiently), etc. A snow avalanche is a heavy fluid which flows, whereas the snow
comprised a solid structure beforehand. The polymerization of a liquid progressively

increases its viscosity until it becomes solid.
The properties of fluids may thus depend on their history, or on the way we
excite them (this is the case for quicksand, yoghurt stirred with a spoon, soils
liquefaction, etc.), and the chemical transformations which they undergo (cooking of
food, polymerization during the generation of plastics, for example). Such behaviors
have significant practical importance, as they are often encountered in the chemical
industry, in the food industry and in many other natural phenomena. The complete
representation of these properties is a difficult problem which is beyond the scope of
this book; we will here limit ourselves to the presentation of just some of the laws
governing the complex viscous behavior of simple 1D flows.
The general form of the relation between the viscous stress tensor and the strain-
rate tensor must be invariant in changing reference frames (in particular, geometric
space is homogenous and isotropic). Moreover, a fluid is often a body with isotropic
physical properties (with the exception of liquid crystals, nematic and smectic
liquids, etc.), so the relationship between the viscous stresses characterized by the
tensor Wij and the strain rates characterized by the tensor Hij should not have any
favored direction. This means that the two tensors should have the same principal
axes. The general expression of the laws governing non-linear fluid behavior, which
obeys the necessary invariance in changing reference frames, is beyond the scope of
this book and we will limit ourselves in a first instance to the study of 1D flows, and
then to a more general study of Newtonian fluids which correspond to the linear
approximation of irreversible thermodynamics. For other cases the discussion
becomes quite complex and the reader should refer to texts concerning the rheology
of non-Newtonian fluids ([FRE 64], [GER 94], [TAN 00], [VER 97]).
Finally, recall that the action of the stress tensor on matter is equivalent to the
existence of the volume source divV (see section 2.1.3.2) which can be written:
divV grad p divW [3.40]
3.4.1.3. Physical origin of viscosity

A fluid is comprised of matter which does not have any particular structure: the
relative positions of the particles which make up the matter are not fixed with
respect to any reference structure. These particles can move around freely, rather
like a person moving around in a crowd which is confined to a closed space: this
person can have a relatively autonomous motion in the crowd, but he or she must
follow the general motion.
Macroscopic forces in a fluid are either forces induced by a force field at the
microscopic level (intermolecular forces), or the result (or manifestation) of
momentum transfers at the molecular level (thermal agitation). For an ordinary fluid
at macroscopic rest, the effects of molecular forces and thermal agitation reduce to
pressure forces.
A simple physical rationale can be used to explain the origin of viscous

phenomena. Consider a gas in which the molecules are sufficiently sparsely
distributed for their interactions to be entirely localized in the vicinity of the
intermolecular collisions. Consider schematically two layers P and Pƍ in the vicinity
of molecules, of average velocities u and u + du (Figure 3.9). The forces exerted
directly (from a distance) by each of the layers on the other are zero, on account of
the preceding assumption. In addition to their mean velocities, all of the molecules
are subjected to a thermal agitation whose action does not have any privileged
direction. In particular, the molecules of the lower layer move through the upper
layer and vice versa. While a molecule moves from one layer to the other due to a
transverse motion it conserves its longitudinal momentum. Thus, the molecules of
the lower layer Pƍ, of mean velocity u, are slower than those of the upper layer P into
which they arrive. This layer, P, will exert a force on these slower molecules in
order to increase their velocity to u+du. The molecules thus accelerated will
obviously have exerted an opposing force on the layer P which is opposite to the
velocity. An analogous effect is produced in the reverse direction for molecules
which descend into the plane Pƍ. This constant exchange means that the lower layer
slows the upper layer, while the upper layer transports the lower layer.
u+du
M2 P
c
A c
u
M1 P'
Figure 3.9. Viscous friction and momentum transfer
Let us propose a simple model, supposing that the molecules are exchanged
between two layers separated by a distance equal to the mean free path O. Let c be
the mean transfer velocity of molecules towards the plane P (mean quadratic
velocity). A molecule of mass m moving from the lower layer to the upper layer
arrives with a momentum deficit equal to mdu . Now, over the distance O, the
variation of the velocity is equal to O du dy . The mass flux of the molecules, per
unit surface, arriving from the lower layer into P is equal to nMc 2 U c 2 (where
n is the volume number of moles), where the factor 1/2 comes from the simplifying
assumption that half of the molecules move from the lower to the upper layer. This
results in a force per unit surface, exerted by these molecules on the plane P, equal
1 du
to U c O . We note that the friction force increases with the temperature in
2 dy
the same way as the mean velocity does.
In other words, the viscous stress W is proportional to the velocity gradient, and
the order of magnitude of the proportionality coefficient (the dynamic viscosity P
defined in section 3.4.3) is:
1
P| U cO [3.41]
2
The preceding rationale can be rigorously applied in the context of kinetic gas
theory. It cannot be directly applied to liquids in which intermolecular forces act
directly between the fluid layers. The effect of viscous friction is thus directly
related to these forces and to the geometry of the molecules, particularly when these
are complex (long macromolecules which may be oriented, stretched or broken
under the shearing action of a velocity gradient). Furthermore, these mechanisms
show that an increase in temperature, on account of the increased thermal agitation,
will decrease the effect of these intermolecular forces and thence the level of viscous
friction.
The origin of viscosity is thus seen to be a microscopic momentum-mixing

phenomenon. At the macroscopic scale, the same phenomenon is encountered in
turbulent flows where considerable fluctuations occur. However, the scale of these
fluctuations is no longer that of the mean free path or the intermolecular distance,
but that of the flow. This leads to considerable difficulties which completely change
the nature of the problem ([MAT 00], [TEN 72]).
3.4.2. Viscometric flows
We consider here the flow of fluids of constant specific mass in which the
acceleration is everywhere equal to zero. The trajectories and streamlines are thus
straight lines traveled over at constant speed by fluid particles which possess a
uniform motion. The conservation of the mass requires that these straight lines be
parallel, all convergence or divergence of these lines obviously implying an
acceleration or deceleration of particles in the medium which is supposed
continuous. The corresponding dynamic equations are obtained by writing, as in
fluid statics, that the sum of the external forces acting on the fluid is zero. In the
presence of a mass force of density gi and using expression [3.39], we obtain:
wp wW ij
U gi 0 or U g grad p div W 0 [3.42]
wxi wx j
If the force g derives from a potential U, we can, in such flows, replace the
pressure with the driving pressure p g p U 0 gz (z, vertical upwards coordinate;
see section 2.2.1.4.1):
wp g wW ij
0 or grad p g div W 0 [3.43]
wxi wx j
As the axis Ox is taken parallel to and in the same direction as the velocity u(y,z)
of the flow, the strain-rate tensor in Cartesian coordinates is reduced to two
components:
1 wu 1 wu
H xy H yx H xz H zx
2 wy 2 wz
Let us assume that for a fixed flow geometry, the shear velocities do not lead to
normal and transverse stresses (no effect analogous to a Poisson coefficient in
elasticity). All that remains therefore are the viscous shear stresses, Wxy and Wxz.
Using these assumptions with the equations for viscous fluids, we see that there is
no variation of the driving pressure in the directions Oy and Oz. There remains only
a single equation in the direction Ox for the force balance:
wp g wW xy wW xz
0 [3.44]
wx wy wz
We can immediately see that as the viscous stresses are only functions of the
coordinates (y,z), the driving pressure is necessarily a linear function of the
direction x. The gradient dp dx is equal to the ratio 'p g / L , where 'pg is the
pressure drop observed over the distance L (for example, in a tube (Figure 3.10)).
Supposing that we know the driving pressure, then equation [3.44] contains two
unknown viscous stresses which can only be calculated if the physical law which
relates them to the velocity gradient is given.
Flows of this type will often involve plane or cylindrical boundaries aligned with
Ox, and whose viscous properties are independent of physical factors which are a
function of the x-direction. These flows are fully immersed, i.e. without free
surfaces.
3.4.2.2. 2D viscometric flows

Axisymmetric or 2D plane problems can be easily solved, as there is only a single
unknown viscous stress W xy , which we will denote simply as W in order to simplify
notations. Equation [3.44] can thus be integrated and the distribution of viscous
stresses is independent of the phenomenological laws governing the viscous stresses.
For a 1D flow with plane symmetry we obtain:
dp g wW
0
dx wy
By integrating we can show that the distribution of viscous stresses is a linear

function of the y-direction:
dp g
W y const [3.45]
dx
We obtain an analogous result for an axisymmetric problem of axis Ox, either

writing equation [3.44] with cylindrical coordinates or making a balance of forces
exerted from the exterior on the cylinder whose radius is r (Figure 3.10b):
dp
0 S r 2 2S rWr
dx
By integrating with respect to the radial coordinate r, we obtain the distribution
of viscous stress:
dp r
W r const [3.46]
dx 2
Under the assumption that there is no variation of the driving pressure in the medium,
the viscous stress W is constant. The flow can thus only be produced by the viscous
entrainment of the fluid due to friction on some cylindrical walls which are moving at a
different velocity. In the 2D plane, a Couette flow is produced between the two plane
surfaces (Figure 3.10a); such a flow is also generated by two circular cylinders
undergoing axial displacement, but these configurations are of little practical interest.
The flow between two fixed planes or in a cylindrical tube requires a driving
pressure gradient.
In the preceding cases, the viscous stress distribution is independent of the laws
of viscosity. In particular, these results will be valid for turbulent flows for which
there is no local physical law for the stresses ([COU 89], [MAT 00], [TEN 72]).
When a fluid is purely viscous, the phenomenological law takes the form:
W W u ' y or W W u ' r [3.47]
Substituting this relation into the expression for the viscous stress distribution
[3.45] or [3.46], we obtain a differential relation which permits the calculation of the
longitudinal velocity distribution u and of the mass flux across a cross-section. The
volume flow rate in a circular cylinder can therefore be obtained by integration over
the cross-section; for example, for the problem of revolution in the circular cylinder
of radius R:
R
qv
³ 2Srudr
0
3.4.2.3. The Couette flow

Assuming that there is no variation of the driving pressure in the medium, it is
straightforward to verify that with the preceding assumptions the viscous stress is
constant. For the plane geometry, we have a viscous flow generated by entrainment
of the fluid by friction, which occurs between a fixed plane y=0 and a mobile plane
y=e which is moving at velocity V (Figure 3.10a). The velocity field u(y) is parallel
to the axis Ox and the viscous stress tensor reduces to the components
W xy W yx W which cause the fluid situated above the plane to exert a shear stress
on the fluid situated beneath the plane. This tension is constant in the thickness of
fluid contained between the two planes.
G
y V V=ZR
e
u(y) W
O
Z
O
x
(a) (b) (c)
Figure 3.10. Couette flow: (a) principle, (b) Couette rheometer,

(c) rheometer with rotating cone
The practical realization of such a flow involves two co-axial cylinders, one of
radius R, which is fixed, and a second, of radius R+e, which is animated by a
rotational motion of angular velocity Z about its axis (Figure 3.10b). The moment of
the torque measured on the fixed cylinder gives the value of the viscous stress.
Another realization is the plane-cone rheometer which is comprised of a cone

turning over a plane (Figure 3.10c); a fluid placed between these is then studied.
Locally this gives a Couette flow. As the thickness e between the walls and the
velocity of the cone wall are proportional to the distance from the axis, the shear
velocity is constant at all points.
These two devices are particularly well-adapted to studying the

phenomenological law of a fluid whose viscous stress is only a function of the shear
velocity gradient uƍ(y). The result of this is that the shear velocity is also constant
and the velocity distribution is linear regardless of the phenomenological laws
governing the fluid studied (see Figure 3.10), y being the distance of a point situated
between the two cylinders from the inner cylinder:
y
u y V
e
The Couette flow allows us to obtain in a direct and simple manner the viscous
law of a fluid for viscometric flows. As the general law for the viscous behavior of
fluids is of a tensorial nature, the preceding relationship is a simplified law, which is
uniquely valid for this kind of flow.
3.4.2.4. Principal physical laws for viscous behavior of a fluid

We have just discussed a means of simultaneously measuring the velocity
gradient u ' y and the viscous stress W, in other words, means of establishing the
phenomenological law for the viscous behavior of a fluid. Fluids comprised of
molecules of a sufficiently small size generally have a linear behavior, viscous stress
being proportional to shear velocity (Newton’s law). Fluids containing
macromolecules, particles which are more or less solid in suspension, that is to say
pasty substances, may have relatively varied viscous laws. In general, the non-
linearity of the viscous law is closely related to the complexity of the molecular
structure. The viscous behavior may be a function of time, in other words of the
mechanical excitations and motions which the fluid has most recently undergone.
Due to the tensorial nature of the viscous stresses and strain-rates, there may
exist transverse effects associated with a 1D effect (normal stresses for a shear
velocity). This is comparable to the properties of solids (Poisson coefficient in
elasticity). This type of effect is mainly manifest in liquids with complex structures.
Furthermore, elastic effects may occur for relatively weak stresses in viscoelastic
bodies; these may or may not be time dependent: the body may present the
properties of a solid for a stress whose modulus is less than some level W, beyond
which it becomes liquid (Figure 3.11b and Figure 3.11c). In what follows we will
only discuss some simple cases from the vast domain of rheology.
3.4.2.4.2. Time-dependent fluids

We will distinguish:
the thixotropic (Figure 3.11a) situation in which the excitations and the
imposed stresses reduce the viscosity (ketchup, yoghurt, gels, drilling mud,
quicksand, etc.);
rheopexy, which is less frequently encountered (Figure 3.11b), which
corresponds to an opposite effect, where the viscosity increases with agitation
(suspensions of Indian corn starch in water).
The time dependence effect of the viscous behavior leads to hysteresis

phenomena (Figure 3.11c).
W W W
u'(y) u'(y) u'(y)

(a) (b) (c)
Figure 3.11. Hysteresis cycles: thixotropic fluid (a) without or

(b) with elastic properties; (c) rheopexic fluid with elastic properties
These effects are obviously associated with local modification for the particles or
macromolecules which are more or less in contact at the microscopic level. The
explanation for thixotropy is schematically as follows. In suspensions of solid grains
in liquid, the grains, after a sufficiently long duration of immobility, finish by
coming into contact with one another; this leads to solid friction. After a certain
level of agitation (vibration, flow) and the resultant stresses imposed on the
ensemble, all solid-solid contact has been undone, and a lubrication occurs such that
the grains now slide over one another when a velocity gradient is imposed. This
corresponds to the phenomenon of “quicksand”. Further to these purely mechanical

effects, physico-chemical interactions may occur between the grains, leading to a
structural macroscopic organization. This is progressively destroyed under the
effects of a velocity gradient. The possible complexity of these phenomena is such
that explanations of the same nature may, depending on the circumstances, often
lead to inverse effects (rheopexy).
3.4.2.4.3. Purely viscous fluids

Their viscous behavior is characterized by a single physical law which is
independent of time for a fluid of a given structure and composition. The most
commonly encountered types of fluid are:
7he Newtonian fluid which obeys a linear law (curve 1 of Figure 3.12):
W Pu ' ( y ) (P: the coefficient of dynamic viscosity) [3.48]
The Ostwald-de Waele fluid. Behavior is no longer linear for fluids containing
molecules with complex structures or particles in suspension and of strong
concentration:
n 1
W m u' ( y) u ' ( y) [3.49]
When the viscous stress increases faster than the velocity gradient (n>1) the fluid
is said to be dilatant or rheothickening (curve 3 of Figure 3.12); in the opposite case
(n<1), it is known as pseudoplastic or rheofluidifying (curve 4 of Figure 3.12). The
first case occurs in concentrated suspensions of solid particles which are in contact
with one another, when the velocity gradients are steep. The second situation is
observed in flows of polymers of high molecular mass, whose linear molecules are
more or less intertwined and trap a certain quantity of water. Under agitation, the
velocity shear aligns the molecules along their axes, thus freeing the trapped water.
W
(2)
(3)
(1)
W
(4)
u'(r)
0
Figure 3.12. (1) Newtonian fluid, (2) Bingham,

(3) rheothickening, (4) rheofluidifying
The Bingham fluid. This kind of a fluid is characterized by a cohesion such

that if the modulus of the stress W is less than a given threshold W no strain rate
exists in the fluid, which therefore remains rigid; the sliding of liquid layers over
one another only occurs when the stress exceeds the threshold value, and the
relationship between the stress and the velocity gradients then becomes linear (curve
2 of Figure 3.12):
W d W0 u' y 0
[3.50]
W ! W0 W Pu ' y B W 0
This law is used in particular for the flow of pastes and certain types of mud.
As discussed above, fluids containing solid particles in suspension have a

complex behavior. However, when the solid particles are in weak volumic
concentration c, the fluid thus constituted is Newtonian: its viscosity P is modified
with respect to that of the pure fluid according to the (Einstein) relation, valid for
any 3D flow:
§ 5 ·
P P 0 ¨¨1 c ¸¸
© 2 ¹
3.4.2.5. Poiseuille flow

3.4.2.5.1. Flow in a circular cylinder
The flow is here due entirely to a pressure gradient and the conduit is assumed
sufficiently long for the flow to become fully established, in other words such that
phenomena associated with the tube entrance are not present. Furthermore, the fluid
is assumed to be incompressible such that pressure variations do not lead to changes
in the fluid density.
We will consider the flow of a fluid in a circular tube of radius R (using the
notation of Figure 3.13). The flow, directed in the positive direction along the axis
Ox, leads to a constant negative axial pressure gradient, corresponding to a drop in
the driving pressure equal to 'p g for a length L of the pipe.
r
R
SWrL
x
pg(0)Sr 2 O u(r) 2
pg(L)Sr
pg(0) L pg(L)=pg(0)-'pg
Figure 3.13. Established flow in a circular tube
The velocity distribution can be calculated using equation [3.46], along with the
phenomenological law for the viscous fluid. Using physical law [3.47] in relation
[3.46], we obtain an expression for the derivative u’(y) as a function of the radius r
and the pressure gradient; integrating and taking the velocity equal to zero at the
wall ( r R ), we obtain a parabolic velocity distribution. The mass flow qv can be
obtained as a function of the pressure gradient by integrating the velocity over a
cross-section.
We define the mean mass flux velocity uq using the mass flux qv:
qv
uq
S
This leads to the following results:

Newtonian fluid with constant viscosity: P=cte. The velocity distribution is
parabolic (Figure 3.14a):
u

dp r 2 R 2 [3.51]
dx 4P
We can deduce the volume flow rate qv in the tube and the flow velocity uq as a
function of the pressure gradient (Poiseuille’s law):
dp SR 4 'p SR 4 qv 'p R 2
qv uq [3.52]
dx 8P L 8P SR 2 L 8P
– Bingham fluid. Using the relationship between the stresses and the velocity
gradient for the fluid ([3.50]), the velocity profile can be calculated as for Newtonian
fluids. We find a velocity distribution comprising a uniform central section
(corresponding to a bulk displacement) and two half-parabolas (Figure 3.14b).
solid core
(a) Newtonian fluid (b) Bingham fluid
Figure 3.14. Velocity profile for (a) a Newtonian fluid and (b) a Bingham fluid
– Ostwald-de Waele model. The physical law is no longer linear for fluids
containing strong concentrations of particles in suspension:
n 1
du du
W m
dr dr
The velocity distribution thus calculated looks like a parabolic distribution, but
pointier if n is greater than 1, and flatter if n is less than 1.
3.4.2.5.2. The Rabinowitsch-Mooney relation

The law for the drop in driving pressure as a function of the mass flow in a
Poiseuille flow depends in a complex way on the law governing the viscous
behavior of the fluid, as this law must be integrated over the radius. The
measurement of pressure-drop in capillary tubes does not directly give the physical
law governing viscous behavior; such measurements must be specially treated in
order to determine the physical laws. We will here outline the principle of the
method (Rabinowitsch-Mooney) for obtaining the relationship between the viscous
stress W and the velocity gradient uƍ(r).
We note first of all that the pressure gradient measured can be expressed using
[3.46] as a function of the wall friction Wp (which is defined as the stress exerted by
the fluid on the wall, WR ), from which we obtain:
r dp R
with: p R
[3.53]
p R dx 2
The measurement gives a relation between the flow rate qv and the pressure
gradient or, according to [3.53], between the viscous wall stress Wp and the flow rate
qv which can be written:
qv
qv W p [3.54]
We obtain a relation between the flow rate and the velocity gradient using
integration by parts on the defining relation for the flow rate:
³
R R R
³ ³
du
qv 2Srudr S ud r 2 S r2 dr
0 0 0 dr
Formula [3.53] allows us to use the viscous stress W as a variable in place of the
radius r in the expression for the flow rate, which can be written (noting that WR
=W(R)):
WR
q v W 3R
³
du
W2 dW [3.55]
3
SR 0 dr

Then, from [3.54], the relation q v W p is known; we thus differentiate the two
sides of [3.55] with respect to WR:

du 1
d q v W 3R
dr R SR 3 W 2R dW R
which finally gives:
3q v W p dq v
u ' r R [3.56]
SR 3
SR 3 dW p
Relation [3.56] between the velocity gradient u' R and the stress Wp is the
relation sought between the velocity gradient and the viscous stress.
3.4.2.5.3. Diverse remarks

1) The flow between two fixed parallel planes separated by a distance 2e can be
treated in a similar fashion. We find, for a Newtonian fluid of dynamic viscosity P,
the following velocity distribution u(y), flow rate qv and mass-flow velocity uq:
dp y 2 e 2 dp 2e 3 qv dp e 2
u ; qv ; uq
dx 2P dx 3P 2e dx 3P
Flows of Bingham or Ostwald-de Waele fluids produce results analogous to

those obtained in the case of the circular cylinder.
The preceding results remain valid when the thickness e varies slowly, in other
words when the planes are weakly non-parallel. These results are the basic laws used
for the theory of dynamic lubrication flows in a thin fluid film (oil or gas) of slightly
variable thickness between a moving wall and a fixed wall; such a flow generates
pressures high enough to support heavy rotating devices without solid contacts in a
bearing ([GUY 01]).
2) It is necessary to note that experimental results agree with the preceding
results provided the mass-flow velocity is not too great. In fact, the condition which
must be satisfied involves the Reynolds number Re which, for a circular tube of
diameter D equal to 2R, is written:
U uq D
Re
P
The value of the Reynolds number Re must be less than about 1,850 for the
preceding theoretical results to agree with experiments. Above this value the flow
becomes unstable and turbulent, and this entirely changes the momentum transfers,
as stated earlier in section 3.4.1.3 ([MAT 00], [SCH 99], [TEN 72])
3.4.3. The Newtonian fluid
3.4.3.1. Definition of viscosity

We will place ourselves in the context of linear thermodynamics where the
relationship between the causes (the strain-rate tensor Hij) and effects (the viscous
stress tensor Wij) is linear. Furthermore, as the fluid matter considered has isotropic
properties, and as geometric space is homogenous and isotropic, this relationship

cannot involve any preferred direction. As a consequence of this, the two preceding
tensors must have the same principal axes. Hence, they must be proportional,
excluding the addition of an isotropic tensor proportional to the trace divV of the
strain tensor. The linear relationship sought between the two preceding tensors
(Newton’s law) can thus be written:
§ 2 · wu § wu wu j ·¸ § 2 · wu
W ij 2PH ij ¨¨K P ¸¸ k G ij P¨ i
¨¨K P ¸¸ k G ij [3.57]
© 3 ¹ wx k ¨ wx ¸
wxi ¹ © 3 ¹ wx k
© j
Linear relation [3.57] introduces dynamic viscosity P and bulk viscosity K. The
latter corresponds to the friction introduced by a purely spherical expansion, for
which the diagonal terms of the strain-rate tensor are equal, and the off-diagonal
terms are zero:
wu1 wu 2 wu 3 1
divV
wx1 wx 2 wx 3 3
Thus:
G
W 11 W 22 W 33 KdivV
This relation shows that the viscosity K is indeed associated with a expansion
velocity, and that this coefficient must be positive in order to represent friction. In
what follows we will neglect the effects of bulk viscosity, which is only important
for phenomena comprising very strong expansion rates, such as shock waves, very
high frequency ultrasound, etc.
3.4.3.2. Properties of viscosity

Dynamic viscosity P is associated with friction in the sliding of fluid layers over
one another. It is expressed in Pascal.second (Poiseuille or decapoise).
The values of dynamic viscosity under normal conditions are, for air and water:
air: P= 17.08 × 10-6 Pa.s (same order of magnitude for gases);
water: P= 1.793 ×10-3 Pa.s.
The variability in the order of magnitude of the viscosity of liquids compared to

the viscosity of water is a little greater for usual liquids, with the exception of oils
and glycerine, which have very high viscosity (from 10-2 to 1 Pa.s).
In section 3.4.1.3 we have given a rough estimation of the viscosity of a gas. As

thermal agitation increases with temperature, changes occur in momentum exchange
at the molecular level. The viscosity increases with increases in the absolute
temperature of gases in accordance with the following (Sutherland) formula, in
which C is a constant which depends on the gas (C=142K for air):
The viscosity decreases quite rapidly with temperature for liquids, because the
intermolecular forces responsible for viscosity in condensed media have a less
vigorous action on account of increased thermal agitation (section 3.4.1.3).
P
The quantity Q , called the kinematic viscosity, is expressed in m2.s-1.
U
EXERCISE –
Calculate the components of the viscous stress tensor Wij:
at all points of the flow defined by the velocity field u1=kx2, u2=u3=0;
in the vicinity of a solid boundary.
3.4.3.3. Expression of viscous volume forces

We have seen (section 3.4.1.2) that the action of viscous stresses in matter is
equivalent to a volume source divW (or wW ij wx j in tensor notation). Assuming a
variable viscosity, we obtain, from [3.57]:
wW ij w ª §¨ wu i wu j ·º
¸» w ª«§¨K 2 P ·¸ wu k º»
«P
wx j wx j « ¨© wx j wx i ¸» wx «¨© ¸
3 ¹ wx k ¼»
¬ ¹¼ i ¬
At constant viscosity this simplifies to:
wW ij w 2ui § 1 · w ª wu k º
P ¨¨K P ¸¸ « »
wx j wx j wx j © 3 ¹ wx i ¬« wx k ¼»
The first term of wW ij wx j is equal to P 'u i , where 'u i are the components of
the Laplacian vector 'V :

'V G G
rot rotV grad divV
Replacing 'V with the above expression gives, in vector notation:
div W
§
P 'V ¨¨K
©
P·
G
¸¸ grad divV
3¹
§
P rot rotV ¨¨K
©
4P ·
G

¸¸ grad divV [3.58]
3 ¹
Usually expansion velocities are weak compared with shear velocities and only

the first term P rot rotV of divW remains:
div W P rot rotV [3.59]
With the preceding approximation, formula [3.59] can be written in Cartesian

coordinates:
wW ij w 2ui
P [3.60]
wx j wx j wx j
Chapter 4
Fluid Dynamics Equations
This chapter is devoted to general equations describing the dynamics of fluid

flows and of the associated transfers. Writing the balance equations for extensive
quantities leads to the equations of fluid dynamics with heat or mass transfers which
take on either local or global forms, and which allow the separation of input and
output mechanisms and sources of extensive quantities. The energy equation then
allows us to discern the interactions between thermodynamics and the movement of
the fluid matter. The main boundary conditions which must be associated with the
partial differential equations are then discussed. Because of the invariance of
physical laws with respect to unit systems, similarity relations allow us to
characterize dynamic and energy problems by means of non-dimensional
parameters.
4.1. Local balance equations
4.1.1. Balance of an extensive quantity G
4.1.1.1. The global balance equation

In matter, the physical phenomena associated with the scalar quantity G are
governed by the volume sources VG and fluxes characterized by the flux density
G
vector q G . We perform a balance for G over the domain D associated with matter in
motion, in other words on a closed system in the thermodynamic sense (no exchange
of matter with the exterior). This balance can be written in the same way as in a
fixed medium, the material derivative (section 3.3.3.3) replacing the temporal
derivative (section 2.1.4.1):
d
³D g dv ³D dv ³6 gu j n j ds ³D VG dv ³¦ qGj n j ds
wg [4.1]
dt wt
wg
The first term ³D dv of the left-hand side of equation [4.1] represents the
wt
variation rate in the amount of the quantity G contained in D, expressing an
accumulation of G in D.
The second term ³6 gu j n j ds is the flux (convective flux) of the quantity G

leaving the surface 6.
The first term of the right-hand side ³D V G dv is a volume source, in other words
a creation of the quantity G, most often by means of the transformation of another
extensive quantity.
The second term ³6 q Gj n j ds is a convective flux of the quantity G leaving the

surface 6 characterized by physical quantities localized on the surface 6.
4.1.1.2. Local balance equation

We can now transform the surface integrals of equation [4.1] into volume
integrals:
§ wg w gu j ·¸ § wq Gj ·
³Dq ¨¨ ¸
dv ³D ¨¨V G ¸ dv
¸
[4.2]
© wt wx j ¹ © wx j ¹
As balance integral [4.2] is true in any domain D, we can derive a local relation
under the usual regularity conditions of the functions considered:
wg
w gu j wq Gj
VG [4.3]
wt wx j wx j
G
When the volume quantity is a vector, g (or gi , i 1, 2,3) , we have a balance
equation for each of its components, which can be written in the vector form:
G

wg i w gi u j wqGij wg JJJG G K JJJG
wt

wx j
V Gi
wx j
or
wt

div g V K
V G div qG
Fluid Dynamics Equations 153
4.1.2. Interpretation of an equation in terms of the balance equation
The terms in equation [4.3] each have a specific form corresponding to physical
mechanisms which we have discussed earlier (section 2.1.3.3). Consider a partial
differential equation which can be written in the form [4.3]:
wf wq j
V [4.4]
wt wx j
We can integrate equation [4.3] over a geometric domain D to give:
wf
³D dv ³D V dv ³6 q j n j ds [4.5]
wt
We will interpret f as the volume density of a quantity F, the amount of which
wf
present in D is equal to ¨ f dv . The first term ³D dv of equation [4.4] represents
D wt
the variation rate in the amount of the quantity F contained in D.
On the right-hand side of equation [4.4], we have separated the terms which can
be written in the form of a divergent vector qi expressed using data or functions of
the problem, and those which cannot be written in this form. In integrating over the
domain D, the terms written in the form of a divergence have been transformed into
flux integrals of the vector qi on the external surface 6 of the domain D: they can
thus be interpreted as transfers of the quantity F by the vector qi. These transport
terms are thus the input-output of the quantity F in D, in other words quantities
gained by D from (or lost to) the exterior through the surface 6.
On the other hand, if the vector qi can be written in the form q i f vi , we

wq j
interpret the term as representing the convective transport of the quantity F by
wx j
the velocity field vi , whose physical interpretation is not important here. In fact, the
vector qi will most often be the sum of a variety of terms with different physical
interpretations. In general, a flux term will only transport a quantity without creating
any; in balance equations which are integrated over a large domain bounded by a
surface 6, fluxes which only have local effects will not appear.
Terms which cannot be written in the form of the divergence of a vector having a
physical meaning consistent with the problem considered cannot be interpreted as
flux terms. Furthermore, take in this case a balance equation by integration on a
domain D which encloses a zone containing the phenomena to be studied. As the

volume integral of non-divergent terms cannot be written as a surface integral, it is
not possible to characterize as an input or an output the detailed repartition of
quantity of F injected in this domain by the source ǻ. This distinction between
these two kinds of terms is very often used in turbulence theory ([COU 89], [MAT
00], [TEN 72]).
We will have occasion to use this interpretation.
4.2. Mass balance
4.2.1. Conservation of mass and its consequences
4.2.1.1. The equation of mass conservation

Mass is strictly conserved (no sources or fluxes of mass in a barycentric
reference frame); we let g U . We obtain from [4.3] the equation (known as the
continuity equation):
wU wUu i
0 [4.6]
wt wxi
or alternatively:
dU wu i
U 0 [4.7]
dt wxi
1 dU
which shows that the variation rate of the density of a fluid particle is
U dt
wu i
compensated by the volume expansion rate .
wx i
The flow of an incompressible fluid (liquid or gas moving at low velocity) of

variable density satisfies dU dt 0 : matter is displaced with its specific mass on a
trajectory (obvious a priori).
4.2.1.2. Consequences of the mass conservation equation

4.2.1.2.1. Writing the balance of a mass quantity
Let g be a mass quantity ( g U g ); by means of a simple calculation the
following two useful relations can be immediately verified:
d
³
dt D
Ug dv ³D U
dg
dt
dv [4.8]
wg w gu i w Ug w Ugu i dg
U [4.9]
wt wxi wt wxi dt
The balance equation of the quantity G can be written using the mass quantity g:
dg § wg wg ·¸ w Ug w Ugu i wq Gj
U U¨ ui VG [4.10]
dt ¨ wt wx i ¸¹ wt wxi wx j
©
4.2.1.2.2. Stream functions

In the particular case of problems with two variables, the solution of equation
[4.6] or [4.7] can immediately be written by means of a stream function \:
a) For a 1D compressible flow:
wU w Uu w\ w\
0 U Uu [4.11]
wt wx wx wt
b) For a steady plane 2D flow:
w U u w U v w\ w\
0 Uu U0 Uv U 0 [4.12]
wx wy wy wx
It is easy to verify that, in this case, the lines \ = const are streamlines of the
G
flow (we have UV.grad \ 0 ).
The mass flow between two streamlines passing through the points A and B is
G
equal to the difference U 0 \ B \ A . The arc element d A dx, dy of the curve
G
joining A and B allows the elementary vector ndA dy,dx of the normal of this
curve to be expressed, from which we can derive the mass flow:
GG § w\ w\ ·
qm ³AB UV .ndA ³AB U 0 ¨¨ dx dy ¸¸ ³AB U 0 d\ U 0 \ B \ A
© wx wy ¹
c) For a steady axisymmetric flow (where Oz is the axis of symmetry), the mass
conservation equation in a steady regime is equivalent to the following definition for
the stream function \:
w Uru w Urw U 0 w\ U 0 w\
0 Uu , Uw
wr wz r wz r wr
d) For flows with constant density U, the densities disappear from the definition
of the stream function; in a 2D plane flow, we have:
w\ w\
u1 u2
wx 2 wx1
4.2.1.2.3. Velocity fields of a viscous fluid near a solid boundary

The property which we will discuss here only depends on the mass balance and
is of course independent of the reference frame which is chosen (Galilean or
otherwise). We will consider a solid boundary P and study the velocity field in the
vicinity of some point O on this boundary. Taking rectangular axes such that the
plane Oxz is tangential to the surface (whose normal is Oy) at the point O. Assuming
G
the fluid to be viscous, the velocity V with components (u, v, w) is zero at the
G
surface (y = 0). The u and w components of the velocity V , parallel to the wall, can
be expanded about y in the vicinity of this point:
u ( x, y , z , t ) y u1 ( x, z , t ) y 2 u 2 ( x, z , t ) ! ;
w( x, y, z , t ) y w1 ( x, z , t ) y 2 w2 ( x, z , t ) !
O C x
Figure 4.1. Flow near a rigid wall
Substituting into the mass conservation equation [4.7] we obtain:
wv 1 dU § wu ww ·
y ¨¨ 1 1 ¸¸ O ( y 2 )
wy U dt © wx wz ¹
y1 dU y 2 § wu1 ww1 · 3
or, by integrating: v ³0 dy ¨¨ ¸¸ O( y )
U dt 2 © wx wz ¹
When the expansion rate is negligible or equal to zero (incompressible fluid

(section 4.2.2)), the first non-zero term of the expansion of the v component is O(y2).
The normal component v tends to zero more rapidly in y than the components
tangent to the wall: the velocity in the vicinity of the wall is parallel to the wall. The
wall streamline C is the streamline limit when y tends to zero (Figure 4.1).
4.2.1.2.4. The strain-rate tensor near a solid boundary

Replacing the preceding expressions for the velocity components into the
expression for strain rate tensor [3.37], we obtain:
§ 0 u1 2 0 ·
¨ ¸
H ¨ u1 2 dU dt U w1 2 ¸ O( y )
¨ 0 w1 2 0 ¸¹
©
We note the existence of shear velocities associated with a velocity field which is
quasi-parallel to the wall, and a expansion velocity which is essentially normal to
the wall.
4.2.1.2.5. Acceleration near a solid boundary

From the preceding expressions for the velocity we can derive the expressions of
the fluid acceleration components (in the wall reference frame). Let us first consider
du wu wu wu wu
the component J x u v w . In the general case where
dt wt wx wy wz
dU dt is not equal to zero, by substituting the velocity components with their
preceding expressions, we obtain:
wu wu ª wu wu º
u w y 2 «u1 1 w1 1 » O( y 3 )
wx wz ¬ wx wz ¼
wu wu wu1 y 1 dU
v y u1 ³0 dy O( y 2 )
wt wy wt U dt
Fluid expansion and unsteady flow variations involve the existence of terms of
order y near the wall.
In the same way, for the second tangential component

dw ww ww ww ww
Jz u v w we obtain:
dt wt wx wy wz
ww ww
ª ww ww º
u w y 2 «u1 1 w1 1 » O( y 3 )
wx wz ¬ wx wz ¼
ww ww ww1 y 1 dU
v y w1 ³0 dy O( y 2 )
wt wy wt U dt
We finally derive the following expressions for the acceleration component

dv wv wv wv wv
Jy u v w normal to the wall:
dt wt wx wy wz
wv § 1 dU w · ª y 1 dU º
wv
v ¨ ¸«³ dy » O( y 2 )
wt wy ¨© U dt wt ¸¹ ¬ 0 U dt ¼
wv wv § w w · y 1 dU
u w y¨¨ u1 w1 ¸¸ ³0 dy O( y 3 )
wx wz © wx wz ¹ U dt
The acceleration is a linear function of the distance from the wall.
For an unsteady flow of an incompressible fluid, the expansion rate is equal to

zero so that we have dU dt 0 (section 4.2.2). The preceding expressions for
acceleration component Jy normal to the wall is O(y2), the tangential components
remaining O(y).
For a steady flow of an incompressible fluid, we have:
Jx
§ u wu
© 2 wx
u ww1
y 2 ¨¨ 1 1 1
2 wz
wu ·
w1 1 ¸¸ O y 3
wz ¹

y 3 ª wu1 ww1 § w w ·º§ wu ww ·
Jy « ¨¨ u1 w1 ¸¸»¨¨ 1 1 ¸¸ O( y 4 )
2 ¬ wx wz © wx wz ¹¼© wx wz ¹
Jz
§ ww
© wx
w wu
y 2 ¨¨ u1 1 1 1 1
2 wx
w ww1 ·
2 wz ¹

¸¸ O y 3
In the absence of a expansion rate (incompressible fluid), the acceleration

normal to the wall is of higher order than the tangential acceleration components.
4.2.1.2.6. Velocity and acceleration near a fixed surface of path lines

Consider a fixed surface which is the locus of trajectories which may be variable.
As we will later see, the practical realization of such surfaces frequently involves
walls or flow separation surfaces on which the flow of a non-viscous fluid slides.
We proceed as above, performing an expansion following the surface normal, for an
incompressible fluid:
u ( x, y , z , t ) u 0 ( x, y, z, t ) y u1 ( x, y, z, t ) O( y 2 )....
w( x, y, z, t ) w0 ( x, y, z, t ) y w1 ( x, y, z , t ) O( y 2 )
wv § wu ww0 ·
¨¨ 0 ¸¸ O( y )
wy © wx wz ¹
The normal velocity component and the acceleration components can be directly
calculated:
§ wu ww0 ·
v y ¨¨ 0 ¸¸ O( y 2 )
© wx wz ¹
du wu 0 wu wu
Jx u 0 0 w0 0 O( y 2 )
dt wt wx wz
dv ª wu 0 ww0 § w w w ·º§ wu ww0 ·
Jy y« ¨¨ u 0 w0 ¸¸»¨¨ 0 ¸¸ O( y 2 )
dt ¬ wx wz © wt wx wz ¹¼© wx wz ¹
dw ww0 ww0 ww0
Jz u0 w0 O( y)
dt wt wx wz
The expansion is accounted for as before.
4.2.1.2.7. Application to the study of quasi-1D flows

We often encounter flows with one or more preferred directions because these
are guided by walls presenting small angles of divergence or convergence (pipes) or
by a wall with a weak curvature (boundary layer). The same can be true in certain
zones of flows generated from almost parallel flows (jets, mixing-layers, etc.).
The preceding developments show that the velocity components and

accelerations normal to these privileged directions are small compared with the
other components. It is thus possible to neglect these components under certain
conditions: for example, the momentum or acceleration is essentially longitudinal in
a 1D flow and there can obviously be no transverse component except in the
presence of an external action. Thus, a fluid with a trajectory associated with a
curved wall will have a centripetal acceleration which is compensated by a pressure

gradient.
On the other hand, in a steady flow, the transverse component v ensures the
conservation of mass in any evolution which is not strictly parallel. Examination of
mass conservation equation [4.6] shows that the term wv wy is necessary for the
existence of a longitudinal variation wu wx of the u component.
The existence of a weak transverse velocity component leads to a transverse

displacement of the longitudinal momentum associated with and transported by the
fluid matter (sections 4.5.4 and 6.5).
4.2.2. Volume conservation
Volume conservation is not a physical law of matter; it results from particular

physical conditions. The flow of a liquid in ordinary conditions is a transformation
which occurs at practically constant volume, except in the case of natural
convection. We will see that the same goes for gas flows at a small Mach number.
The conservation of volume is expressed by a zero expansion velocity:
wu k G
0 (or div V 0)
wx k
Taking the previous relation into account, mass balance equation [4.7] may be
written for an incompressible fluid:
dU
0
dt
4.3. Balance of mechanical and thermodynamic quantities
4.3.1. Momentum balance
4.3.1.1. Dynamic equations

G G
The quantity considered ( p mV ) is a vector quantity: ( g i U u i or g i u i ).
Hence we obtain the dynamic equations (i, j = 1, 2, 3):
d
³
dt D
Uu i dv ³D U
du i
dt
dv ³D U f i dv ³6 V ij n j ds
with: fi the force mass density (for example gravity gi) and Vij the stress tensor
(section 2.1.3.2) Transforming the surface integral of the stress tensor using
Ostrogradsky’s theorem (section 2.1.3.3), and assuming continuous and differentiable
properties, we obtain the local dynamic equations:
G
du i wV ij dV G
U U fi (i, j 1,2,3) or U U f div V [4.13]
dt wx j dt
Replacing the stress tensor with expression [3.39], we obtain the Navier-Stokes
equations:
du i wp wW ij
U U fi (i, j 1,2,3)
dt wx i wx j
G [4.14]
dV G
or U grad p U f div W
dt
In a gas flow of sufficient velocity (at least a few meters per second), the effects
of gravity are negligible and the corresponding term can be ignored. On the
contrary, pressure gradient is negligible for natural convection ([LAN 89], [SCH
99], [YIH 77]).
Case of an external force deriving from a potential U
The force fi may derive from a potential U ( f i wU i wx i ), essentially for
gravitation, electrostatic forces and a few simple cases of inertial entrainment forces
(when the equations are written in a non-Galilean reference frame). Navier-Stokes
equations [4.14] can then be written:
du i wp wU wW ij
U U
dt wx i wx i wx j
G [4.15]
dV
or U grad p U gradU div W
dt
In many cases of practical interest, the external force field is gravity which
derives from the potential gz, where z is a vertically ascending direction; we have:
G
du i wp wz wW ij dV
U Ug or U grad p U grad gz div W
dt wx i wx i wx j dt
When the fluid has constant density U (often incorrectly referred to as an

“incompressible fluid”), the dynamic equations can be written with the driving
pressure p g p U gz , as previously defined (section 2.2.1.4.1):
du i wp g wW ij
U (i, j 1,2,3)
dt wx i wx j
The form of the equations shows that the cause of movement (the local resultant
of external forces) is the gradient of the driving pressure. This elimination of gravity
from the dynamic equations is only of interest if the boundary conditions can also be
expressed as a function of the driving pressure alone (section 4.4.2.4).
NOTE – The force fi does not derive from a potential in many very important
configurations, such that:
– the electromagnetic Laplace force which results from the interaction of an
G G
electric current of density j with a magnetic field B : this produces the volume
G G
force j B ;
G
– the Coriolis force in a reference
G G frame moving with an angular velocity : ,
which leads to a mass force 2: V .
We note that these two forces have a similar structure; thus, this produces some
analogous properties between flows in a rotating frame and flows of a fluid
conducting electricity in a magnetic fluid..
4.3.1.2. Other expressions of the Navier-Stokes equations

Using vector expression [3.28] for acceleration, we have:
G G
dV wV V2 G grad p 1
grad rot V V gradU div W [4.16]
dt wt 2 U U
For a divariant fluid, we can express the pressure as a function of the specific
enthalpy h and the specific entropy s:
dp § grad p ·
dh Tds ¨ or : grad h T grad s ¸ [4.17]
U ¨ U ¸
© ¹
This gives:
G
wV V2 G 1
grad rot V V grad h T grad s gradU div W [4.18]
wt 2 U
or:
G
wV §V 2 · G 1
grad ¨ U h¸ rot V V T grad s div W [4.19]
wt ¨ 2 ¸ U
© ¹
We note that despite the presence of enthalpy and entropy in equation [4.19], the
origin of this equation is purely mechanical (we have so far not considered any
energy balance).
4.3.1.3. The case of a Newtonian fluid
The term div W , which represents the viscous stresses, can be expressed using
one of the forms discussed in section 3.4.3.3:
wW ij w ª §¨ wu i wu j ·¸º w ª§ 2 · wu º
«P » «¨¨K P ¸¸ k »
wx j wx j « ¨© wx j wx i ¸¹» wx i ¬«© 3 ¹ wx k ¼»
¬ ¼
In the case where the viscosity coefficients are constant, we have expression
[3.58] and the Navier-Stokes equation is written as:
G
U
dV
dt
§
grad p U gradU P rot rotV ¨¨K
©

2P · G
¸¸ grad divV
3 ¹
[4.20]
The expansion velocity is usually small and we have, from [3.60]:
du i wp wU w 2ui
U U P [4.21]
dt wx i wx i wx j wx j
4.3.1.4. Inviscid fluids – the Euler equations

When viscous terms are negligible in [4.21], we obtain Euler equations:
G
du i wp § dV G·
U U fi ; Ü grad p U f ¸ [4.22]
dt wx i ¨ dt ¸
© ¹
For a compressible fluid, we can express the pressure as a function of enthalpy

and entropy ([4.17]):
G
sV JJJJJG V 2 ¬ JJJG G JJG JJJJJG
grad U h rot V V T grad s (gravity: U g z ) [4.23]
st 2 ®
Assuming a compressible inviscid fluid we have a situation where the fluid

particles undergo isentropic transformations.
If the flow is homoentropic (s is constant in all the fluid), we have:

G
sV JJJJJG V 2 ¬ JJJG G JJG
grad U h rot V V (gravity: U gz ) [4.24]
st 2 ®
G G G
Note the presence of the rotation vector Z (or vorticity rotV 2Z ) on the
right-hand side of the preceding equations; we will consider the consequences later.
If the fluid flow has constant and uniform density, the Euler equations can be
written with the driving pressure p g p U gz :
G
du i wp g § dV ·
U Ü grad p g ¸ [4.25]
dt wx i ¨ dt ¸
© ¹
This form of dynamic equation shows that the cause of movement (local result of
external forces) is indeed the driving pressure gradient.
4.3.2. Kinetic energy theorem
4.3.2.1. Local equation

The kinetic energy theorem is a consequence of the Navier-Stokes equations; the
local form of the kinetic energy equation is obtained as before (section 3.2.2.1) by
taking the scalar product of dynamic equations [4.13] and the velocity vector:
du i d §¨ V 2 ·¸ wV ij
Uu i U U f i ui ui [4.26]
dt ¨
dt © 2 ¹ ¸ wx j
Kinetic energy is not an extensive quantity for which a conservation principle

applies; we can nevertheless consider [4.26] as a balance equation (section 3.2.4.3).
Replacing the stress tensor with expression [3.39] and assuming that the external
force fi derives from a potential U, we obtain, after grouping similar terms:
d §¨ V 2 ·¸ wp wU wW ij
U u i U ui ui
dt ¨© 2 ¸¹ wxi wx i wx j
or:
w §¨ V 2 ·¸ w §¨ V 2 · wp wW ij
U U ui U¸ u i ui [4.27]
wt ¨© 2 ¸¹ wx i ¨© 2 ¸
¹ wx i wx j
4.3.2.2. Enthalpic form of the kinetic energy equation

For a divariant fluid we have the thermodynamic relation:
dp 1 sp sh ss ¬
dh Tds or:
sx sx T sx ®
i i i
Substituting the preceding expressions into equation [4.27], we obtain:
w §¨ V 2 ·¸ w §¨ V 2 · ws wW ij
U Uu i U h¸ U Tu i ui [4.28]
wt ¨© 2 ¸¹ wx i ¨© 2 ¸
¹ wx i wx j
The reader will note that equation [4.28] is an equation of essentially mechanical
origin, even if it involves thermodynamic functions.
4.3.2.3. Applications
4.3.2.3.1. The constant density fluid
When the density of a fluid is uniform, we can introduce the driving pressure pg
previously defined in equation [4.27]. We then derive the kinetic energy equation:
w §¨ V 2 ·¸ w §¨ V 2 · wW ij
U ui U p g ¸ ui (i, j 1,2,3) [4.29]
wt ¨© 2 ¸¹ wxi ¨© 2 ¸
¹ wx j
Note the absence of any coupling between the mechanical and the
thermodynamic phenomena for a constant density fluid.
Equation [4.29] leads to the definition of total pressure pt and total driving
pressure pgt:
V2 V2
p gt pg U pt pU
2 2
If we assume that the flow is steady and the viscous stresses are zero, the first
term on the left-hand side and the right-hand side of equation [4.29] can be removed,
giving:
w §¨ V2 · d §¨ V2 ·
uj pg U 0 or pg U 0 [4.30]
w x j ¨© 2 ¹ dt ¨© 2 ¹
Bernoulli’s first theorem: in steady inviscid flow, the total driving pressure
V2
p gt pg U is constant on a trajectory or here on a streamline.
2
Bernoulli’s first theorem can also be written:
p gt p g
V 2 [4.31]
U
We thus have, in differential (Lagrangian) form: dp g U VdV 0.
This expression shows that the driving pressure is a decaying function of the
velocity modulus.
4.3.2.3.2. The flow of compressible fluids

With the exception of flows generated by natural convection, which are due to a
density gradient, the effects of gravity are negligible in gas flows. Kinetic energy can
be written:
w §¨ V 2 ·¸ w §V 2
¨
· ws wW ij
U U ui h ¸ U Tu i ui [4.32]
¨
wt © 2 ¹ ¸ wx i ¨ 2 ¸ wxi wx j
© ¹
For steady flows of inviscid fluid undergoing isentropic transformations, this

relation can be written as:
wht d §¨ V2 ·
uj 0 or : h ¸ 0
w xj dt ¨© 2 ¸
¹
This is Bernoulli’s first theorem: total enthalpy ht h V 2 2 is constant on

streamlines in steady inviscid flow. It is equal to the enthalpy generation h0 which
corresponds to zero velocity. We have:
V2
ht h0 h or : V 2ht h [4.33]
2
In differential form, following a streamline, we obtain:
dp
dht VdV 0 [4.34]
U
Here again, pressure is a decaying function of the velocity modulus along a

streamline.
4.3.2.3.3. Case of a perfect gas: the Saint-Venant relation

Consider a divariant fluid evolving according to an isentropic transformation
along its trajectory; the quantities U and c (sound velocity) are then functions of one
variable, p for example. Relation [4.34] is a differential equation whose integration
gives the pressure-velocity relationship. Consider the case of a perfect gas for which
we have:
p p0
J
U U 0J
Substituting this expression into [4.34] gives:
1J
dp p0 dp
VdV VdV 0
U U 0 p1 J
Integrating and taking as reference (p0, U0, T0) the initial or generation
conditions, where the velocity is zero, we obtain the Saint-Venant relation:
ª J 1 º
2J p 0 « § p J »
·
V «1 ¨¨ ¸ » [4.35]
J 1 U 0 « © p 0 ¸¹ »
«¬ »¼
As the pressure and thermodynamic temperature cannot be negative, the Saint-

Venant relation implies the existence of a maximum velocity Vmax:
2J p0 2
V max c0 .
J 1 U0 J 1
The local sound speed c is given by relation [1.25] c 2 wp wU S . We obtain:
J p
c J rT [4.36]
U
Let us denote by c* the velocity V in conditions (called critical conditions) where

the fluid velocity is equal to the local sound speed: V c c* . Considering a
perfect gas of specific enthalpy h C p T , and substituting the preceding definitions
into [4.33] expressing the conservation of total enthalpy, we obtain:
2
V2 c2 V2 V max J 1 2
C p T0 C pT c*
2 J 1 2 2 2J 1
The velocities c* and Vmax are defined by generation conditions.
p
p
0
p*
J 1
V max c*
J 1
c* V
O
Figure 4.2. Saint-Venant relation

Solving the Saint-Venant relation with respect to pressure, c c ( J p 0 U 0 )1 / 2

being sound velocity in generation conditions, we obtain:
J J
p ª J 1 U 0 2 º J 1 ª J 1 V 2 º J 1
«1 V » «1 »
p0 ¬« 2J p 0 »¼ «¬ 2J c 02 »¼
The development of pressure to fourth order with respect to Mach number

M 0 V c 0 is:
V2 1 V4 V2 1V2
p0 p 0 0 2 ... 0 (1 ...)
2 8 c0 2 4 c02
For values of p close to p0, i.e. for small enough Mach numbers, the Saint-
Venant relation reduces to formula [4.31] for an incompressible fluid, whose
application to compressible fluid induces a relative error amounting only to M 02 4 .
Then by using Bernoulli’s theorem to air flowing at 70 m.s-1 ( M 0 # 0.2 ), error is
only 1%. For pressure calculations, Bernoulli’s incompressible relation can be used
for many industrial or domestic problems of gas flowing such as ventilation, wind
effects, etc.
We will look at some other consequences of the Saint-Venant relation in section

5.5.
4.3.2.3.4. Variation of the mass flux density and the Hugoniot relation
The mass flux density (mass flow per unit section of a stream tube) is equal to
U V. Replacing dp in equation [4.34] by its expression dp c 2 dU , we obtain:
V
dU U dV [4.37]
c2
or, introducing the Mach number M = V / c:
d U V
U 1 M 2 dV [4.38]
We see that the mass flux density U V increases with the Mach number in
subsonic flows while it decreases in supersonic flows; it is at maximum for
UV
V c c* . Variations of as a function of the velocity V are shown in
U * c*
Figure 4.3 (U* is the critical density). Note that two velocity values correspond to the
preceding ratio, one subsonic, the other supersonic.
UV
1
U * c*
O c* Vmax V
Figure 4.3. Variation of the mass flux density (with reference

to critical conditions) as a function of the flow velocity
Consider a stream tube of variable cross-section S, small enough for the

properties p, U , V of the fluid to be uniform in any cross-section (this is the
approximation of the flow by slices).
For an incompressible fluid (dU the mass conservation SV = const is

manifest in the fact that the cross-section and the velocity are inversely proportional.
In the stream tube shown in Figure 4.4a, the velocity increases to a maximum which
occurs at the throat of the minimal section Sc, then decays. According to [4.34], the
driving pressure is minimal at the throat.
VG
x
S
Sc
(a) (b)
Figure 4.4. (a) Flow in a stream tube; (b) nozzle
Consider now the case of a divariant compressible fluid which undergoes

isentropic transformations along a stream tube. The thermodynamic and mechanical
quantities of the fluid now only depend on one variable (pressure, velocity, cross-
section, etc.). We can write the conservation of mass flow USV in differential form
in this stream tube ( d U VS 0 ); taking account of [4.38], we obtain Hugoniot’s
differential relation:
d U VS § V2
¨1
· dV dS
¸ 0 [4.39]
U VS ¨ c2 ¸ V S
© ¹
Equation [4.39] cannot be solved with respect to the derivative dV/dS in a

velocity interval in which we have the velocity V = c (critical conditions), this value
corresponding to a singular point of this differential equation.
On the other hand, we can always solve equation [4.39] with respect to dS/dV,
showing that the cross-section S of the tube is a decreasing function of the velocity
in subsonic flow (V<c), but increasing for supersonic flows (V>c).
The minimum cross-section corresponds to:

– either a velocity maximum in a subsonic flow, as in the incompressible case;
– or a velocity equal to the local sound speed.
As the mass flow rate USV is constant, the area S of the cross-section varies
inversely with the product U V studied earlier (Figure 4.3). The maximum mass
flux density U V occurs in the throat of the nozzle, where the cross-section is
smallest. However, this quantity U V can only take values smaller than the critical
value U* c*. So, the mass flow rate cannot exceed the critical value U* Sc c* calculated
in the throat cross-section. The mass flow rate may be obviously less than this
maximum value, the velocity in cross-section Sc then being smaller than the sound
velocity.
This configuration is characteristic of nozzles (Figure 4.4b). Note that for a given
value of the cross-section S or the quantity ȡSV, there are two corresponding
velocities, which explains the difficulty encountered when we try to solve equation
[4.39] for the velocity V. We will return to look at the consequences of this situation
in section 5.5.4.
4.3.3. The vorticity equation
We have seen that for a mechanical system, we can consider a dynamic moment
(section 3.2.1) taken about the inertia center of the system in a reference frame
which is parallel to a Galilean reference frame. The instantaneous rotational
G G
movements in a fluid are characterized by the vorticity vector 2Z rot V ([3.38]).
The vorticity equation can be obtained by taking the curl of the Navier-Stokes
equations [4.14]. Assuming body forces to derive from a potential, and using the
identity:

G
rot a. A
G G
a.rot A grad a A [4.40]
we obtain, for any fluid:
rot rot VG V
G
w rot V 1 §1 ·
grad U grad p rot ¨¨ div W ¸¸ [4.41]
©U
wt 2
U ¹
For a divariant fluid, we have, applying formulae [4.19] and [4.40]:
rot rot VG V
G
w rot V §1 ·
grad T grad s rot ¨¨ div W ¸¸ [4.42]
wt ©U ¹
Assuming a Newtonian fluid with constant viscosity and density, the vorticity
equation can be written, with [4.21]:
G
wZ
wt
G
rot Z V X 'Z [4.43]
G
For a 2D velocity field, the vorticity vector Z has only one component. This fact
leads to very important specific properties, in particular in the study of turbulence
([COU 89], [MAT 00], [TEN 72])
The reader will note that the vorticity equations in inviscid flows are identical for
homoentropic and incompressible fluids.
4.3.4. The energy equation
4.3.4.1. The total energy equation

4.3.4.1.1. General expression as a function of the stress tensor
We will here apply the same procedure which we applied for an elementary
material system (section 3.2.5.1).
The quantity g is here the total specific energy e V 2 2 whose balance is

obtained by writing that its variation in a domain D is due to the external addition of
energy:
d §¨ §
ë V
2 · ·
¸ dv ¸
³ U Pth Pmec [4.44]
dt ¨© D ¨© 2 ¸ ¸
¹ ¹
The external additions of energy are:

– mechanical power Pmec provided from the exterior, which includes:
- the specific mechanical power fiui due to the specific force fi in the domain D;
- the surface power Vij nj uj due to external stresses Vij nj on the outer surface
6of the domain D;
Pmec ³D U f i u i dv ³6 V ij n j u i ds ³D U f i u i dv ³D
w
V ij u i dv
wx j
– thermal power Pth received from the exterior, which includes:

- volume power VT generated by volume source in the domain D;
- thermal fluxes across the outer surface 6 due to the thermal flux vector
G
density qT :
wqTj
Pth ³D V T dv ³¦ qTj n j ds ³D V T dv ³D dv
wx j
Replacing the powers Pth and Pmec in [4.44] with their values, we obtain:
d §¨ §
ë V
2 · ·
¸ dv ¸
wqTj
³ U Pth Pmec ³D V T dv ³D dv
dt ¨© D ¨© 2 ¸ ¸
¹ ¹ wx j
[4.45]

w V ij u i
³D U f i u i dv ³D dv
wx j
Using Ostrogradsky’s theorem to transform the surface integrals, and relation

[4.45], we get the local equation:
U
d §¨
e
V 2 ·¸
U fi ui

w V ij ui V
wqTj
[4.46]
T
dt ¨© 2 ¸¹ wx j wx j
or by introducing the viscous stress tensor [3.39]:

d §¨ V2 ·
¸ w pu i
w W ij u i wqTj
U e U f i ui VT [4.47]
dt ¨© 2 ¸
¹ wx i wx j wx j
4.3.4.1.2. Expression using the total enthalpy

Taking the divergent form (see [4.10]) of the left-hand side of [4.47], we have:
w ª §¨ V2 ·º ª §
¸» w « U ¨ e V
2 ·
¸u pu »
º
«U e
wt «¬ ¨© 2 ¸» wx « ¨ 2 ¸ j j
»
¹¼ j ¬ © ¹ ¼
w W ij u i
wqTj
U f i ui VT
wx j wx j
Using the definition of enthalpy ( h e p U ) then gives:
w ª §¨ V2 ·º ª § 2 · º
«U e ¸» w « U ¨ h V ¸u »
wt «¬ ¨© 2 ¸» wx j « ¨ 2 ¸ j»
¹¼ ¬ © ¹ ¼ [4.48]
w W ij u i wqTj
U f i ui VT
wx j wx j
4.3.4.1.3. The internal energy equation

Subtracting equation [4.26] from equation [4.46] gives:
de w U e
w U eu j wu i wqTj
U V ij VT [4.49]
dt wt wx j wx j wx j
Expanding the stress tensor, we obtain:
de wu i wu i wqTj
U p W ij VT [4.50]
dt wxi wx j wx j
4.3.4.1.4. Enthalpic and entropic forms of the energy equation

Equation [4.47] can also be written as a function the other extensive variables:
the internal energy e, the enthalpy h or the entropy s, defined by the thermodynamic
relations:
§1· § p·
de pd ¨¨ ¸¸ Tds dh d ¨¨ ¸¸
©U¹ ©U¹
or, taking account of relation [4.7]:
de d §1· ds p dU ds p wu i ds dh d § p·
p ¨ ¸T T T ¨ ¸
¨ ¸
dt © U ¹ U wxi dt ¨© U ¸¹
dt dt U 2 dt dt dt dt
The energy equation can thus be written in one of the following forms:
de wu i ½
U p °
dt wxi °
dh dp ° wu i wqTj
U ¾ W ij VT [4.51]
dt dt ° wx j wx j
ds °
UT °
dt ¿
The third equation of [4.51] allows us to define the dissipation function ):
wu i
) W ij [4.52]
wx j
which represents the thermal power released locally per unit volume by viscous
friction.
4.3.4.1.5. Entropy balance

The entropic form of energy equation [4.51] is of some interest, as it in fact
constitutes an entropy balance, which is not surprising since on account of the
assumption of local equilibrium, the entropy is a state function related to the internal
energy. This equation contains the flux terms w qTj w x j (divergence terms), and
volume source terms associated with W ij wu i wx j and V T . The entropy balance
equation can thus be written:
ds W ij wu i VT 1 wqTj
U [4.53]
dt T wx j T T wx j
4.3.4.1.6. Other forms of the energy equation

The left-hand side of the energy equation can also be written using other
thermodynamic variables. For example, formula [1.27] allows us to write (E is
expansion coefficient [1.26] at constant volume):
ds C v § dp dU ·
¨¨ c2 ¸¸ [4.54]
dt E pT © dt dt ¹
wu i
Finally, equation [4.6] for the conservation of mass allows us to replace
wxi
1 dU
with and to obtain other forms for the left-hand side of equation [4.51]:
U dt
de wu i p dU ½
de
U p U °
dt wxi dt U dt °
dh dp °
U °
dt dt ° wu i wqTj
ds U C v § dp dU ·¾ W ij VT [4.55]
UT ¨¨ c2 ¸¸° wx j wx j
dt E p © dt dt ¹°
°
U C v §¨ dp wu j ·
¸ °
c2U
E p ¨© dt wx j ¸¹ °
¿
The heat convection equation can be obtained from an expression of equation

[4.51] for a transformation:
at constant pressure (dp/dt = 0) using specific enthalpy h;
at constant volume ( wu j wx j 0 ) using specific internal energy e.
For a perfect gas or an incompressible liquid, these quantities can be expressed

as a temperature function ( dh C p dT and de C v dT ) and [4.51] can be written
in the following form, where, depending on the case considered, we take either Cp or
Cv as the specific heat C:
dT wu i wqTj wu i w §¨ wT ·
¸
UC W ij VT W ij VT O
dt wx j wx j wx j wx j ¨© wx j ¸
¹
4.3.4.1.7. Expression of the dissipation function for a Newtonian fluid
Substituting G[3.57] for the viscous tensor of a Newtonian fluid into [4.52], we
obtain ( T divV ):
2
wu i wu i §¨ wu i wu j · §
¸ ¨K 2 P ·¸§¨ wu i ·
) W ij P ¸
wx j wx j ¨© wx j wx i ¸ ¨© ¸
3 ¹¨© wx i ¸
¹ ¹
2
4 P H 12 2
H 23 2
H 31 2

2 P H 11 2
H 22 2
H 33
§ 2 ·
¨¨K P ¸¸T 2
© 3 ¹

Very often, the shear velocities are much greater than the expansion velocities
and the expression for the dissipation function is simplified:
) 2
4P H 12 2
H 23 2
H 31
ª§ 2
§ wu ·
2
§ wu
2º
wu1 wu 2 · wu u · [4.56]
P «¨¨ ¸ ¨ 2 3 ¸ ¨ 3 w 1¸ »
« wx 2 wx1 ¸ ¨ wx ¸ ¨ wx ¸ »
¬© ¹ © 3 wx 2 ¹ © 1 wx 3 ¹ ¼
4.3.5. Balance of chemical species
Equation [2.25] for the mass balance of a chemical species can be written using
G
the partial density Ui and the local velocity Vi of the component i:
wU i G

div U i Vi V mi [4.57]
wt
The volume source of mass Vmi of species i is a homogenous chemical reaction.
G
Introducing the local mass average velocity V [2.29] gives:
wUi
wt
G
div Ui V > G
V mi div Ui Vi V
G
@ [4.58]
The mass flux density qmi Ui Vi V

G G
is associated with diffusive
phenomena considered here in the reference frame of the inertia center (section
2.4.3.5). For the case of a binary mixture, this can be expressed using formula [2.69]
for non-isothermal mixtures.
In the case of isothermal diffusion, the mass flux density of the species i is given
by relation [2.57]. We thus obtain the diffusion equation of the flow:
ª § U i ·º
wU i G
div U i V V mi div « U D grad ¨¨ ¸»
¸ i 1,2 [4.59]
wt ¬« © U ¹¼»
In the case where species 1 is in weak concentration, a discussion analogous to

that of section 2.4.4.2.7 leads to the following equation:
wc1
wt
G
div c1 V V c1 D'c1 [4.60]
Recall that for a binary mixture, diffusion is represented by the balance equation
of one species, associated with the mass conservation equation for the mixture. In
simple cases, we can consider that the creation of species 1 is due to a homogenous
reaction of order m:
V c1 kc1m [4.61]
The coefficient k for the reaction kinetics varies according to a law of the form:
§ U ·
k k 0 exp¨¨ ¸¸
© RT ¹
The quantity U is the activation energy; T designates the absolute temperature.

The form of the preceding relation shows a strong coupling between the temperature
and the reaction speed.
4.4. Boundary conditions
4.4.1. General considerations
The partial differential equations satisfied by the preceding quantities are not
sufficient for the definition of a particular problem. We must also specify boundary
conditions of different kinds. The definition of the domain studied D constitutes the
first step. It is defined by the surface which bounds D and on which we will impose
conditions for the unknowns of the problem (boundary conditions). Because of the
particular nature of the time variable, we must also specify the initial conditions.
In fluid mechanics the boundary conditions must be carefully considered, as we

rarely encounter physical problems in entirely closed domains, since the flow must
be generated by some appropriate device (a pump, a fan, a moving vehicle, a
meteorological situation, an acoustic cavity, etc.). The domain studied is generally
limited by solid walls and zones which are connected to other zones being studied.
4.4.2. Geometric boundary conditions
4.4.2.1. Solid walls

The usual solid walls are relatively easy to treat, since the fluid touches the wall:
the velocities of the solid wall and the fluid are equal where they are in contact. The
reason for this fact is related to the roughness of the walls on the molecular level,
and to the thermal excitation by which mean momentum is transferred from one
medium to another (it is incidentally the same as the interpretation of viscosity and
contact action). This interpretation is such that the physico-chemical nature of the
wall does not influence the adherence condition.
This adherence condition of the fluid at the wall is always very well satisfied in
ordinary conditions where the mean free path of the molecules is small compared
with the roughness. The same is not true in the study of rarefied gases, where we
must take account of the properties of the wall and introduce a slipping coefficient.
In certain cases, the walls are permeable, in other words they let some matter
pass through them. This is the case when we suck or blow through a porous
medium. In these conditions, the difference between the tangential fluid velocities
and the wall are zero at the wall. The normal fluid velocity with respect to the wall
depends on the fluid injection process. An analogous condition is encountered in the
presence of phase changes at the wall: evaporation, fusion, and other heterogenous
chemical reactions which consume or produce fluid.
The boundary conditions for thermal and diffusion problems were discussed in
sections 2.3.2 and 2.4.6, which the reader can refer to.
4.4.2.2. Flow entry and exit zones

In addition to the walls which guide the flow, we generally need to specify the
conditions at the entrance or exit of the domain studied. Real flows are always
generated by machines (solid surface in movement) or by differences in conditions
between upstream and downstream reservoirs. In practice, we know how to impose
wall velocities, injection or extraction flow rates, unsteady forces on a wall (by
means of electromagnetic devices). However, it should be noted that we do not
know how to impose a given pressure or velocity distribution throughout a fluid.
In fact, the entrance of a flow into a domain is a rather particular zone, as we

have an initial condition which is largely analogous to that of the temporal variable
(it is in fact an initial condition in Lagrangian variables). In practice, such a zone is
found at the exit of another fluid domain and we need to specify a velocity or
pressure distribution which is compatible with the equations of motion. This is
straightforward when we can take a uniform flow or zone at rest with a constant
driving pressure in a section perpendicular to the velocity relatively far upstream of

the walls which are to guide the flow. However, in general, this is not the case, and
the choice of an incompatible velocity profile leads to difficulties (“numerical
shocks”) in the subsequent numerical solutions. The same difficulties are
encountered in thermal and acoustic studies, where initial conditions for the entropy
or other material quantities are required. We will come back to this point in section
5.6.2 when we consider the classification of partial differential equations.
Further to the preceding difficulties, we will see in Chapter 5 that certain

material properties can also be transported by waves which propagate in different
directions, including in the upstream direction. This means that the very idea of the
flow entrance (“upstream” and “downstream”) may depend on what it is we wish to
study. Remember also that the idea of an extensive quantity leads to conservation
properties which must be satisfied in the global balances, and it is not always
obvious that this condition is respected.
Finally, for flows which possess unstable zones, it is not sufficient to specify
velocity or pressure distributions: the perturbations which enter the domain must be
defined so as to fully determine the problem to be solved.
In conclusion, it is clear that we are a long way from understanding how to

proceed in all cases in order to obtain a well-posed problem.1
4.4.2.3. Free surfaces

The free surface of a flow2 (or the interface between two immiscible fluids) is an
unknown boundary, on which the two following properties need to be ensured:
1) The free surface (or the interface) is a material boundary between two
immiscible fluids: it comprises the locus of fluid particle trajectories. Let z be the
altitude of this free surface3:
z [ ( x, y , t )
The w component of the velocity in the Oz direction satisfies the relation (known
as the kinematic relation):
1 We will here define a well-posed theoretical problem (to be solved analytically or

numerically) or an experiment whose solution is reproducible, even if we change a calculation
method or an experimental process.
2 For instance, water flows in open channels, rivers, etc.
3 The axis Oz is the opposite of the direction of external forces field, usually gravity field.
d[ w[ w[ w[
w u v
dt wt wx wy
2) The physical nature of this surface must provide the conditions necessary for
the stress tensor. The pressure discontinuity p1 p 2 across the surface is given by
Laplace’s law:
§1 1·
p1 p 2 V ¨¨ ¸¸
©R R' ¹
where the quantities V, R and R’ are, respectively, the surface tension and the radii
of curvature of the free surface at the point considered (section 2.2.1.4.2).
For nearly all industrial flows of water and aqueous solutions, surface tension
does not play an important role: we can assume pressure continuity at the free
surface.
The discontinuity of the tangential stresses is related to the physico-chemical and

flow conditions of the free surface; in particular, if the free surface contains a
surface active substance (a mono-molecular thickness layer can suffice) and is not
regenerated, it behaves like a solid surface. In many cases of water flow, the free
surface is regenerated and we have continuity of the tangential stresses between the
two media. In hydraulics, we can consider that the viscous stress is nearly equal to
zero at the free surface of a flow.
4.4.2.4. Fully immersed flows

Fully immersed flows are flows without free surface whose upstream and
downstream conditions are hydrostatic. We have seen that in the Navier-Stokes
equations, gravity can be eliminated by using a driving pressure variable pg = p +
Ugz. This change of variables is only of interest so long as it does not result in
gravity reappearing in the boundary conditions, which should be expressed in terms
of the driving pressure and not pressure itself. This excludes flows with free surfaces
which are not horizontal (sea swells, rivers, surface runoff, etc.) for which we have,
for example, a constant pressure condition.
4.4.3. Initial conditions
No physical problem is timeless. There is always a beginning to an experiment

and therefore to its modeling. From this point of view, flow problems can be
difficult, because even if the boundary conditions are steady (independent of time),
the solution may present diverse and complex characteristics. Some examples will
demonstrate the degree of these difficulties:
– a steady solution may be established after a transition period, but this solution
may depend on the initial conditions (such properties are used in fluidics when we
manipulate the eventual hysteresis of separated flow zones);
– no steady solution may exist, but a more or less complexly established
unsteady flow regime may occur, which is determined and predictable;
– the flow may become more or less chaotic while maintaining a more or less
organized unsteadiness;
– an established turbulence may be present in certain domains of the established
flow: in other words the flow, while unsteady, may have stable statistical
characteristics;
weak perturbations, which are difficult to characterize, may be present in the
initial or boundary conditions, and these may be of considerable importance for the
evolution of the flow.
We will come back to some of these points in section 6.6.
4.5. Global form of the balance equations
4.5.1. The interest of the global form of a balance
Even the most rudimentary model must satisfy conservation laws for extensive
quantities. The interest in a global balance is that it allows us to observe a system
from the exterior. Balance equation [4.1] for a quantity G in the domain D can be
written by replacing the material derivative by its expression (section 3.3.3.3); it
can be written in one of the following two forms, as a function of the volume
quantity g or of the mass quantity g:
wg
³Dq dv ³6 gu i ni ds ³D V G dv ³¦ q Gj n j ds [4.62]
wt
w Ug
³Dq dv ³6 gdq m ³D V G dv ³¦ q Gj n j ds [4.63]
wt
Note that it is equivalent to writing a Eulerian balance in a domain D which is

assumed fixed. As we saw in section 1.1.4.2, this balance corresponds to an open
system.
wg
³D wt dv ³D V G dv ³¦ qGj n j ds ³6 gui ni ds
“physical” flux convective flux
The left-hand side term corresponds to the possible accumulation of the quantity
G in D. This variation of the quantity G contained in this domain is the sum of:
the production of the sources VG;
the fluxes of the quantity G due to transfers through matter;

the fluxes of quantity G associated in the matter which enters D by crossing the
boundary 6 (convective fluxes).
In the preceding equations, we can interpret the surface and volume integrals in
different ways:
– the surface integrals over 6 comprise inputs and outputs of the material system
contained in D, in other words we can evaluate them without knowing the internal
structure of the flow within the system;
– the volume integrals in the domain D clearly imply the internal variables of the
system for which a model is necessary.
IMPORTANT NOTES –
1) The global balance equations do not require the assumption of a continuous
medium. They can be applied to any domain D constituted by matter, continuous or
discontinuous: diverse media containing many different phases (liquid flows with
bubbles, fluidized beds for transport of pulverized media, etc.), avalanches of solid
substances (stones, snow, mud, etc.) and/or liquids (mud, etc.). The balance equation
can be directly written without assumptions concerning the differentiable properties
of the continuous medium. Only the assumption of integrability of the fluxes across
the surface 6 is necessary. Even the time differentiation assumption is not necessary:
it suffices to effect a balance of the amount of G contained in D between two
instants t1 and t2:
quantity of G
come out from 6 come out from 6
G D (t 2 ) G D (t1 ) created in D
by convection by other process
between t1 and t 2
When the quantity G is a vector (momentum), the same balance can be

performed on its moments.
2) In all modeling we must have a global extensive variable G(t) internal to the
wg
domain D, defined from the amount of G contained in D, such that ³D dv is
q wt
dG
equal to ; equation [4.63] can thus be written in the form:
dt
dG source of ½ inputs of ½
® ¾® ¾
dt ¯G inside D ¿ ¯G into D ¿
4.5.2. Equation of mass conservation
The expression of mass conservation for a fluid of density U inside the domain
D during its movement, can be written from [4.62], with g U :
wU
³D dv q m6 0
wt
where q m6 ³6 dq m is the mass flux leaving the surface 6.
4.5.3. Volume balance

GG
The elementary volume flux dqv crossing the surface ds is: dq v V .n ds . The
volume flux leaving the domain D can be written:
G wu i
³6 dq v ³D div V dv ³D dv
wx i
4.5.4. The momentum flux theorem

G
Applying formulae [4.8] and [3.33] to the quantity g U V , we obtain the linear
momentum theorem:
w U u i
³D dv ³¦ U u i u j n j ds ¦ Fei
wt
or, expressing the elementary mass flux: dqm= Uuj nj ds:

w U ui
³D wt
dv ³ ui dqm
¦
¦F ext i
G [4.64]
dv
w UV G G
or ³
D wt ³
¦
Vdqm ¦ Fext
Note that in a steady flow the balance depends only on the input-output of the
domain. In unsteady flows, the first term corresponds to an eventual accumulation of
momentum in D.
Thus, as discussed in section 3.2.1, we can also perform a balance using the
moments (angular momentum). This gives:
dv OM VG dq
G
w UV G
¦ M ext
³D OM ³¦ m [4.65]
wt
The momentum flux theorem is particularly useful in applications. Let us

consider as an example a propulsion device whose role is to increase the velocity of
the fluid which crosses it (a propeller, a turbine, a pump which produces a jet, etc.).
The reference frame is associated with the propulsion device, which is considered to
operate in a steady regime. We will ignore the details of the system and we will
consider the pressure to be constant on a surface 6 external to the device (Figure
4.5).
G
V1
G
V1 6 D G
V2
G
V1
Figure 4.5. Propeller thrust
Let qm be the mass flow rate passing through the propulsion device. The
G
momentum flow rate parallel to the upstream velocity V1 and leaving the surface 6
is equal to:
G
³¦ V dq m G G
q m V 2 V1
The momentum flux is balanced by the thrust it generates, equal to

G
G
P q m V 2 V1 .
Let us note that conservation of mass flow rate demands the existence of an
incoming mass flow through the lateral sides of surface 6
Very many practical problems can be treated in the same way, the changes in
momentum corresponding to the forces which can be calculated from a knowledge
of the mass flows and the velocities ([EVE 89], [GAR 06], [GUY 01], [SAG 66],
[SPU 97]).
4.5.5. Kinetic energy theorem
The global form can be obtained by integrating the local form [4.27] over the
domain D. Assuming a potential U which is independent of time, we obtain:
d §¨ V 2 · d ª §V2 · º
³D U U ¸ dv «³ U¨ U ¸ dv »
dt ¨© 2 ¸
¹ dt «¬ D ¨ 2
©
¸ »
¹ ¼
w §¨ U V 2 · § 2 ·
¸dv U ¨ V U ¸u n ds wp wW ij
³D ¨ ³ ³D u i dv ³D u i dv
wt © 2 ¸ 6 ¨ 2 ¸ j j wx i wx j
¹ © ¹
We transform the terms of the right-hand side by introducing the divergence of a

vector, the corresponding integral being thus transformed into a surface integral 6;
we obtain, after moving a pressure term to the left-hand side:
w §¨ UV 2 · ª §V2 ·º
³D dv ³6 « p U ¨ U » u j n j ds
wt ¨© 2 ¹ «¬ ¨ 2
© ¹ »¼
wu i wu
³D p dv ³6 W ij u j n j ds ³D W ij i dv
wx i wx j
The first term describes the accumulation of kinetic energy in D in the

transitional regime. The sum of this term and the following terms of mechanical
energy flux (kinetic and pressure) across 6 is equal to the sum of the power P Q6 of
the external viscous stresses exerted on 6 of the power of pressure forces in D, and
of the power –P Q D dissipated in D by viscosity (dissipation function [4.52]):
w §¨ UV 2 · ª §V2 ·º
« U ¨ » u j n j ds
³D ¨ dv ³6 p
¨ 2
U
wt © 2 ¹ «¬ © ¹»¼ [4.66]
wu i
³D p dv Pv6 PvD
wx i
with: PQ6 ³6 W ij u j n j ds the external power provided by the viscosity on 6;

wu i
PQD ³D W ij dv the power dissipated in D by viscous friction.
wx j
In incompressible flows, we obtain:
w §¨ U V 2 ·¸ § 2·
¨ p U gh U V ¸u n ds
³D ¨ dv ³6 ¨ j j Pv6 PvD [4.67]
wt © 2 ¸¹ © 2 ¸¹
In incompressible inviscid flows, the sum of the temporal variation in the kinetic
energy of D and the flux of mechanical energy (kinetic and pressure) leaving 6 is
zero.
4.5.6. The energy equation
The integration of the energy equation over the domain D should be performed
using a form of the equation which allows flux integrals to be written. For a
compressible fluid, the form [4.48] with the total enthalpy is best suited. We obtain,
neglecting gravity:
w ª §¨ V2 ·º §
¸» dv U ¨ h V
2 ·
¸u n ds
³D «U e ³
wt «¬ ¨© 2 ¸» 6 ¨
2 ¸ j j [4.68]
¹¼ © ¹
³6 W ij u i n j ds ³D V T dv ³6 qTj n j ds
This equation has the inconvenience of not including the dissipation function ).
Using one of the forms [4.51] is preferable for thermodynamic balances, but internal
balances remain in the form of volume integrals:
w Uh § wp wp ·¸
³Dq dv ³6 hdq m ³D ¨ uj dv ³D ) dv ³D V G dv ³¦ qTj n j ds
wt q ¨ wt wx ¸
© j ¹
However, these forms become perfectly adapted for the study of thermal transfer
at constant pressure or constant volume. In such conditions, we immediately obtain,
from [4.51]:
w U g
³Dq dv ³6 gdq m ³D ) dv ³D V T dv ³¦ qTj n j ds [4.69]
wt
where for g we must use either the specific internal energy e or the specific enthalpy
h. For the usual case of a fluid of constant specific heat, the preceding equation
becomes:
w ( U CT )
³Dq dv ³6 U CTdq m ³D ) dv ³D V T dv ³¦ qTj n j ds [4.70]
wt
where, depending on the case, we used either Cp or Cv for the specific heat C.
4.5.7. The balance equation for chemical species
Equation [4.58] can be immediately integrated to give:

wU i GG G G
³Dq dv ³6 U i V .n ds ³D V mi dv ³¦ q mi n ds [4.71]
wt
For a binary isothermal mixture, using [4.59] we obtain:
wU i GG G § Ui ·
³Dq dv ³6 U i V .n ds ³D V mi dv ³¦ U D n.grad ¨¨ ¸ds
¸ [4.72]
wt © U ¹
In the case of weak concentrations, balance equation [4.60] gives:
wci GG G
³Dq dv ³6 c i V .n ds ³D V ci dv ³¦ D n.grad ci ds [4.73]
wt
Homogenous or heterogeneous chemical reactions must be specified

accordingly, in the volume source term ³D V ci dv or in the flux term
G
³¦ D n.grad ci ds at the walls. In general, chemical reactions also lead to heat
release, which must be accounted for in a source term of the energy equation
([BIR 01]).
4.6. Similarity and non-dimensional parameters
4.6.1. Principles
4.6.1.1. Invariance of physical laws
The fundamental physical laws are described by relations between real numbers
obtained as measures of physical quantities. A measurement is a comparison
between the quantity studied and a similar quantity considered as a unit. Most units
are “derived”, as they depend on physical laws: for example, the surface unit is a
square whose sides measure one unit of length. With such choices, physical laws are
relations which comprise either dimensional coefficients which have a physical
interpretation (the speed of sound for example) and which can be expressed using
the units of the system, or non-dimensional coefficients which are independent of
the system (for example, S for the surface of a circle). There are four fundamental
units which can be arbitrarily taken for mass, length, time and temperature, and
whose choice determines a coherent system of units. No specific physical
phenomenon is used to govern the particular quantities chosen as units, and for
practical reasons we can arbitrarily choose the four fundamental units of the
international metric system (the meter, the kilogram, the second, the Kelvin).
With these choices, the mathematical relations representing physical phenomena

are true regardless of the fundamental units which are used, which means that they
possess an invariance with respect to changes in the units which are used,
transformations which form a group. A given physical problem thus has an infinity
of equivalent numerical representations. Similarly, a numerical problem can
represent many different physical problems obtained using different systems of
units.
4.6.1.2. Similar problems

A given problem must be repeatable, meaning that its definition must always
lead to identical results, within a certain margin of error. The problem is only
defined if analysis of the phenomena involved has been correct and complete, in
other words if we know the partial differential equations, and the boundary and
initial conditions which define the problem.
Two problems are similar if two systems of units exist, such that the
measurements of all the quantities of one of the problems, using a given system of
units, are equal to the measurements of the corresponding quantities of the other
problem using another system of units.
The conditions for similarity may be obtained by searching for conditions in

which the two problems will obey the same ensemble of equations and boundary
conditions after an appropriate change of the system of units. Because we can
choose the system of units, it is easier to write the equations and conditions of a
problem with a system of units which corresponds to the problem; the equations
obtained are non-dimensional, in other words they are independent of the units
chosen for the physical measurements: calculating the area of a circle using its
radius as a unit gives a result equal to S.
4.6.1.3. Non-dimensional study of a problem

4.6.1.3.1. Dynamic and thermal problem
We will now consider a flow problem involving a perfect gas with heat transfer
around a circular obstacle of diameter L. We will assume that the specific heat Cp,
the viscosity P, and the thermal conductivity O are constant. The problem is posed in
the following manner:
– mass conservation:
wU wUu i
0 [4.74]
wt wxi
– Navier-Stokes equations:
du i wp wz w 2ui
U Ug P (i, j 1,2,3) [4.75]
dt wx i wx i wx j wx j
– energy equation (in enthalpic form) and the equation of state:
w 2T
UCp
dT

dp

) H ij O
dt dt wx j wx j
[4.76]
p
rT
U
– dynamic boundary conditions:

- on the body surface, velocity equal to zero: u i 0;
G G
- at infinity, velocity V equal to U ;
- eventually a free surface condition on a horizontal surface;
– thermal boundary conditions:
- constant temperature T equal to Tw on the surface;
- constant temperature equal to TO at infinity;
– initial conditions:
- the fluid is assumed at rest at time t=0.
The results of the problem are the velocity field, the pressure field and the
temperature field, in addition to the surface densities (friction stresses, pressure,
thermal flux density, etc.) or the global quantities (drag, lift, heat flux, etc.). These
are functions of the coordinates and the n data of the problem (U, L, U0, T0, Tw, P,
O, g, Cp).
4.6.1.3.2. Non-dimensional equations

For the fundamental units we will use the data of the problem such as the
velocity U, the length L and the reference density U0, and we will define the
following non-dimensional variables:
xi ui p ~ Ut U ~ T
~
xi u~i ~
p 2
t U~ T [4.77]
L U U 0U L U0 T0
We have:
w 1 w d U d w2 1 w2 1 ~
; ~; ' '
wxi L w~
xi dt L dt wx j wx j L2 w~
xi w~
xi L2
Performing variable change [4.77] in equations [4.74] to [4.76] and dividing by

the dimensional coefficients of the left-hand side in each equation leads to the
following reduced equations:
wU~ w U~ u~i
~ 0
wt w~xi
du~ w~
p gL w~z P w 2 u~ij
U~ ~i ~ U~ ~ ~ ~ (i, j 1,2,3)
dt wx i U 2 wx i U 0UL wx j wx j
[4.78]
~ ~
U 2 d~ PU O w 2T
~ dT
U ~
p
~
~

) H~ij
dt C p T0 d t U C p T0 L U 0 C p UL w~x j w~x j
~
p rT0 ~
~ T
U U2

where the dissipation function ) H ij (which characterizes kinetic heating) is given
PU 2 ~ ~

by [4.56]: ) H ij ) H ij .
L2
The preceding equations contain the following (non-dimensional) similarity

parameters:
Reynolds number: Re 0UL μ

Froude number: Fr U gA
Mach number: M0 U c0
Prandtl number: Pr μ C p
Eckert number: Ec U 2 C p T0 1
M 02
which allows us to write [4.78] in the form:
wU~ w U~ u~i
~ 0
wt w~xi
2~
du~ w~p ~ 1 w~ z 1 w u ij
U~ ~i ~ U ~ ~ ~ (i, j 1,2,3)
dt wx i Fr 2 wx i Re wx j wx j
[4.79]
~ w 2 u~ij
d~
U~ ~
dT
Ec ~
p Ec ~ ~
) H ij 1
dt dt Re Re.Pr w~
x j w~xj
~
p 1 ~
T
U~ J M 02
The Eckert number and the Mach number are related by the perfect gas relation
between Cp and r ( rJ J 1C p ). We also define the Péclet number Pe = Re. Pr.
These similarity parameters characterize the importance of the non-dimensional

terms by reference to the terms of the left-hand side of the non-dimensional
equations. For example, we can write:
U 0UL U 0U 2 L U 0 dui dt acceleration term

Re
P PU L
|
2

P . w u i wx j wx j
|
viscosity term
U U 0U 2 dynamic pressure
Fr |
gL U 0 gL hydrostatic pressure difference
2 2
U U0 U kinetic energy
Ec |
C pT0 U 0 C pT0 enthalpy
The Eckert number (or the Mach number) represents the influence of
compressibility on the properties of the flow whose kinetic energy is due to enthalpy
variations in the steady flow (conservation of the total enthalpy [4.33]).
The Reynolds number usually takes on very large values (from 103 to 108),
which signifies the globally weak effects of viscosity and thermal conduction in
most flows. We will see in Chapter 6 that this property is not true in zones close to
solid boundaries (boundary layers).
Note that if the viscosity or other physical properties are not constant we must
introduce further non-dimensional functions.
4.6.1.3.3. Non-dimensional boundary conditions

The treatment of boundary conditions is immediate: the conditions U and L are
transformed into a single condition equal to 1. The thermal problem requires, for
example, the temperature Tp, given at the wall to be transformed into a non-
dimensional parameter T p T0 . If other velocity conditions or geometric dimensions
are given, they provide supplementary similarity parameters (velocity ratio, shape
factor, etc.).
The boundary conditions determine the order of magnitude of the similarity

parameters and the associated phenomena. Simplifications can result from this. For
example:
– if the Mach number is small (less than 0.3 in practice), it has to be shown (see
section 4.3.2.3.3) that the density variations can be neglected for the study of fluid
motion and that incompressible fluid relations can be used. Thermal transfer thus
occurs at constant pressure and the dissipation function is generally negligible. The
only unknown of the problem is therefore the temperature T T0 , and we take as a
~ T T0
reduced temperature T ;
T p T0
– for fully immersed flows of a liquid (section 4.4.2.4), we have seen that gravity
no longer has a direct effect on the flow; we therefore take the driving pressure
pg p U g z to be the only variable, gravity disappearing from the equations
and from the boundary conditions: the Froude number is no longer a parameter.
4.6.1.3.4. Non-dimensional expression of results

The non-dimensional quantities of a problem are functions of non-dimensional
coordinates and similarity parameters. They allow the calculation of dimensional
results which take on a specific form. For example, the pressure p can be written:
p 0U 2 p (x i , t, Re, M, ...)
Note that in place of n dimensional data, we now have only n – 4 similarity

parameters (Re, Pr, Ec, Fr, Tp/T0). For a purely dynamic problem (no thermodynamic
equation), we would have only n – 3 similarity parameters. These results, which can
be obtained from a dimensional analysis (without writing the specific equations of
the problem), constitute the Vaschy-Buckingham theorem. Note that the
dimensional analysis does not allow us to see if certain similarity parameters
(Froude number) can be eliminated.
In practice, in place of the preceding non-dimensional parameters, we use other

quantities defined by custom. For example, we define the local pressure coefficient
p ~
Cp which is equal to p 2 . The local viscous friction stress at a solid
U0U 2 2
wu P U wu~
boundary, equal to W p P can be expressed by means of the
wn n 0 L wn~ n~ 0
Wp
friction coefficient Cf , which can also be written:
U 0U 2 2
2 wu~
Cf
Re wn~ n~ 0
wT
The local thermal flux density at the wall qTp O can be written in
wn n 0
non-dimensional form of the local Nusselt number Nu:
~ qTp L
wT
Nu ~
wn n~ 0

O T p T0
The global quantities (components of the force exerted on an obstacle, thermal

flux, etc.) can be obtained by integration of the local quantities over the wall surface.
We here define the non-dimensional coefficients, which can be immediately
deduced from the local coefficients. For example, the force X exerted on an obstacle
in the x-direction can be calculated by integration over the surface S of the obstacle:
§ wu · § 2 wu~ ·
X ³S ¨¨ P cos - p sin T ¸ds U U 2 L2 ³S~ ¨ cos - ~
p sin T ¸ds
¸ ¨ Re wn~ ¸
© wn n 0 ¹ © n~ 0 ¹
Instead of the non-dimensional integral in the above formula, we can use the
drag coefficient Cx in the x-direction, often defined by means of the surface Sc,
projected frontal area (projection of S on to the plane perpendicular to the x-axis):
U0 U 2
X Cx Sc
2
4.6.1.3.5. Case of unsteady flows

The procedure is identical to that outlined above; the temporal conditions
introduce further similarity parameters, for example the Strouhal number
St = NL / U , where N is a characteristic frequency which appears among the data of
the problem.
However, we will see in section 6.6 that a problem posed with steady boundary
conditions will often have an unsteady or turbulent solution: the non-dimensional
~
solution is therefore a function of the variable t . For example, if the wake of the
cylinder discussed above comprises a preferred frequency N, the non-dimensional
frequency St = NL/U will be a function of the similarity parameters St(Re,M,...).
4.6.1.3.6. Validity of the similarity approach and comparison with experiments

The posing of a problem in non-dimensional form constitutes a mathematical
model of the phenomena studied, which is based on a choice of equations supposed
to be representative. The experimental verification of the model involves ensuring
that the non-dimensional results are correct, in other words that the non-dimensional
coefficients are only functions of similarity parameters defined with the model. For
example, in the fully immersed flow of an incompressible Newtonian fluid, the
results should depend on the Reynolds number alone. The problem has been badly
posed if this is not the case.
We can take an historic example: at the beginning of the 20th century, L. Prandtl
and G. Eiffel independently measured the drag of a sphere in a wind tunnel. They
found different drag coefficients for the same Reynolds number. This raised
questions over the validity of the similarity approach. An understanding of this

phenomenon was obtained 20 years later: one of the experiments comprised a
laminar flow in the zone of the boundary layer where the wake was generated,
whereas the turbulence had already been triggered in the other experiment. The
solution of the Navier-Stokes equations was not the same in the two experiments
because of instabilities generated by the residual present turbulence that is different
in the two wind tunnels.
Questioning the similarity methodology amounts to denying the validity of the

basic laws of classical physics, which seems, at the very least, both erroneous and
presumptuous. It is clearly the specific model used which must be questioned when
agreement is not obtained by means of the non-dimensional representation of the
results (for example, the friction coefficient as a function of the Reynolds number).
4.6.1.3.7. Similarity in diffusion problems

We will limit ourselves to the case of weak concentrations [4.60] by considering
the steady flow of a fluid of initial concentration c0 which reacts with a reactive wall
according to a law of the form [2.70]. The concentration c satisfies equation:
w
c ui D 'c
wxi
wc
with the condition at the wall P: D kc P m .
wn n 0
This can be written in non-dimensional form by letting c~ c c0 . The non-

dimensional problem can be written:
w ~~ D ~~
c ui 'c
w~
x i UL
with the absorption condition for a constituent, due to heterogeneous reaction at the
wall:
wc~ kc0m 1 L ~ m 1
~ cP [4.80]
wn n~ 0 D
The similarity parameters of the diffusion problem are:

UL
– the Péclet diffusion number: Pe D Re.Sc;
D
Q
– the Schmidt number: Sc (Q kinematic viscosity);
D
kc 0m 1 L
– the Damköhler number: which characterizes the speed of the
D
chemical reaction.
kc 0m 1 L
Condition [4.80] shows that if the number is large compared to 1, the
D
wall concentration is zero, and the flux of the corresponding constituent is limited by
kc0m 1L
diffusion. If, on the other hand, we have 1 , we have c P c 0 and the
D
reaction speed is equal to kc 0 m : diffusion is no longer important.
Chapter 5
Transport and Propagation
The objective of this chapter is to present the general properties of the equations
which describe the flow of matter. The substantial derivative describes the physical
idea of the properties associated with matter which is in motion. It results in a
particular structure of the equations of fluid mechanics and the properties associated
with the displacement of material quantities. This leads to a specific means of posing
and numerically solving fluid mechanics problems. The corresponding mathematical
techniques which underlie the physical concepts are well known, but unfortunately
they are not widely taught; we will here recall them without providing detailed
demonstrations.
5.1. General considerations
5.1.1. Differential equations
We will first quickly recall the general properties of a differential equation which
is satisfied by a scalar function:
– the order n of the equation: the greater the maximum order n of the derivatives,
the greater the complexity of the solution (n time constants correspond to n
independent solutions for a linear differential equation with constant coefficients).
We know that, in general, the solutions of a differential equation form a family of
functions which depend on the n parameters. This means that the number of
boundary conditions which must be specified in order to determine a unique solution
is generally equal to the order n;
– the general structure of the differential equation: the solutions of a linear

differential equation belong to a vector space, which means that we can express a
solution by means of suitable linear combinations of other solutions. The dimension
of the vector space of the solutions is equal to the order of the differential equation;
– the properties of the coefficients of the differential equation, and in particular
their eventual singular properties: a coefficient which cancels itself out at some point
very often leads to particular properties of the solutions at that point. Such is the
case for solutions to problems of revolution about an axis for the point which
corresponds to the axis (generally denoted r = 0). Other particular points can play an
important role, as we will see with regard to stability problems;
– the nature of the boundary conditions imposed determines the kind of
differential problem which is posed:
- a Cauchy problem is determined by n boundary conditions given at a point
for the function and its n – 1 successive derivatives; such problems generally have a
unique solution1 in the vicinity of this point. This kind of problem is often
encountered in mechanics for initial values of a motion. Regular behavior is only
ensured in the vicinity of the point, and the regular behavior can eventually extend
to the entire domain considered; however, in numerous cases we encounter accidents
in the behavior of the system far from the said point (divergence of the solution,
instabilities and random behavior, etc.),
- the n boundary conditions required may be specified at the two points of the
extremities of an interval. This case is common in physics, for field problems of
physical quantities (electromagnetic fields, velocity and displacement fields, etc.) for
which the solid boundaries impose particular conditions. The existence and
uniqueness of the mathematical solution can be obtained if the problems have been
well posed in physical terms for the entire domain.2
In the preceding particular case where we impose zero conditions at two points,
the differential equation generally has a zero solution. However, there may exist
particular coefficient values for which a non-zero solution exists, and which
therefore depends on some parameter. These particular values correspond to the
eigenvalues of the differential operator. We will treat these problems in more detail
in Appendix 4.
The case of a system of coupled differential equations can amount to the study of
a higher order differential equation for one of the unknown functions after
elimination of the other unknown functions. For example, the system of two second
1 I.e. when we have mathematical properties of regularity, for the rest more or less satisfied in
application conditions of physics and mechanics.
2 We wish to say that any approximation with notable physical consequences has been made
when writing the equations for the problem.
Transport and Propagation 201
order equations with two unknown functions y(x) and z(x) can immediately be
transformed:
y' ' f ( z ' , z , y ' , y, x) z' ' g ( z ' , z, y ' , y, x)
Deriving z'' leads to the appearance in the expression for z''' the derivatives y'' and
z'' which we replace with their earlier expression f and g, giving for z''' an expression
of the form h( z ' , z , y ' , y, x) . Doing the same for z'''', we get a similar expression
m( z ' , z , y ' , y, x) . The elimination of y, y' from the three equations:
z' ' g ( z ' , z , y ' , y, x), z' ' ' h ( z ' , z , y ' , y, x), z' ' ' ' m ( z ' , z, y ' , y, x)
gives a fourth order differential equation for the function z(x).
Conversely the differential equation z '''' f ( z ''', z '', z, x) for the unknown
function z(x) is immediately transformed into a system of four first-order differential
equations by letting z ' ' ' u , z ' ' v, z ' w :
u' f u , v, w, z , x ; v ' u; w' v; z '. w
More generally, the system of first-order differential equations:
x i' t
fi x j , t i, j 1,..., n
can be transformed into a differential equation of order n for one variable, for
example x1; in effect, by differentiating x1' t , the two other derivatives x i' t
appear, which we replace as a function of the quantities xi (t); we thus obtain:
x1' ' t x 'j t

wf i
x j , t
g2 x j ,t
wx j
Doing the same for the subsequent derivatives up to order n, we obtain n

equations which express the first n derivatives of the function x1 as a function of all
the functions xi:
x1(i ) t
gi x j , t i, j 1,2,..., n
Eliminating the n – 1 functions x i t (i = 2, 3,…, n) between the n preceding

equations leads to the nth order differential equation satisfied by the function x1 t .
x1( n ) t
F x1 , x1' , x1' ' ,.., x1( n 1) , t :
The initial conditions x1k 0 k 0,..., n 1 can be obtained in the same way

as a function of the initial values xi 0 .
The preceding procedure is only appropriate if the equations of the system are
suitably coupled. Otherwise, new variables Xi must be chosen, such that the
equations are coupled. We will leave it to the reader to verify this for the elementary
linear system (let: X1 = x1. + x2 and X2.= x1 – x2):
§ x' · § a 0 ·§ x1 ·
¨ 1¸ ¨¨ ¸¸¨¨ ¸¸
¨ x' ¸ © 0 b ¹© x 2 ¹
© 2¹
5.1.2. The Cauchy problem for differential equations
Consider the differential equation:
f ' x g x, f [5.1]
Solving the Cauchy problem involves calculating the value of the solution in the
vicinity of the value x0 of the variable x, at which the value f0 = f (x0) is given. It is
equivalent to saying that given the point (x0, f0) of the plane (x,f), we must find the
variations dx and df which it is possible to calculate from equation [5.1]:
f ' x dx df [5.2]
It is clear that the system of equations [5.1] and [5.2] with the unknown f '(x)
should be of rank 1, hence the condition:
1 g
0
dx df
The curve of the plane x, f on which the value of f extends (or is transmitted)
is therefore given by the relation:
dx df
1 g x, f
The preceding presentation may seem unnecessarily formal, but its interest is
that it can be extended to the study of partial differential equations.
5.2. First order quasi-linear partial differential equations
5.2.1. Introduction
In expressing the laws of physics in a local form, we obtain partial differential
equations, the functions representing the physical quantities depending on space and
time variables. These equations have properties which are quite different from
differential equations, because their general solution no longer comprises a family of
functions with a finite number of parameters, but a family of functions which we can
choose arbitrarily. Let us take the simple example of the equation:
w2 f
0
wxwy
whose solution depends on the two arbitrary functions M and \:
f ( x, y ) M ( x) \ ( y )
This solution is only a general form of the function dependence with respect to
the variables. The choice of these functions will depend on the boundary conditions
which are specified.
As with the differential equation, we can define a Cauchy problem: for example,
for a partial differential equation with two variables x, y , the unknown function f
is given on the curve C0 of the plane (x,y) and we seek to evaluate this function in
the vicinity of C0.
In what follows, we will limit ourselves to a relatively elementary approach

which only consists in verifying the existence of a series expansion in the place of a
solution. We thus identify the essential properties of the solutions and the basic
concepts which govern the general physics of the phenomena studied.
We will not demonstrate the uniqueness of the solution, as this requires advanced
mathematical knowledge and precise assumptions which do not necessarily have
physical reality, for example:
– do analytical functions (indefinitely differentiable) exist?;
– what is the nature of the “relation” between distributions and physical
problems?;
– how can we verify the physical reality of a Lipschitz application (section
1.1.1.4)?
DEFINITION – A quasi-linear partial differential equation is an equation which is

linear with respect to the partial derivatives of highest order, whose coefficients are
functions of variables and derivatives of lower order.
The equations which we will encounter in mechanics and energy of continuous

media will be quasi-linear.
5.2.2. Geometric interpretation of the solutions
We have previously seen that the substantial derivative in the equations of fluid
mechanics describes transport of a quantity associated with matter. We will recover
this interpretation of the substantial derivative by purely mathematical
considerations.
Consider the partial differential equation which represents transport of the

quantity F, in which the right-hand side is equal to g(f,x,y,t):
wf wf wf
u v g ( f , x, y , t ) [5.3]
wt wx wy
The coefficients u and v are given functions of (f, x, y, t). Equation [5.3]
describes the balance of the scalar quantity F (mass, entropy, number of moles of a
chemical species, etc.) associated with the corresponding specific quantity f (section
4.2.1.2.1). The function g can be considered a source of quantity F, as it only
depends on the coordinates of the material particle and the associated value of the
function f. There is however no interaction with the neighboring particles, and
equation [5.3] does not contain any diffusive flux term for the quantity F. Balance
equation [4.3] of the volume quantity f can thus be expressed in the form [5.3].
Let us ignore for the moment the physical interpretation of equation [5.3], of
which we will here give a geometric interpretation. In the 4D space (t, x, y, z),
consider the surface S described by the equation:
) (t , x, y, z ) z f x, y , t 0

We know that the vector grad ) of components f t , f x , f y ,1 is normal to
S. The equation of the plane tangent to the surface S at the point (t, x, y, z) can be
written (designating the usual coordinates (T, X, Y, Z)):
wf wf wf
T t X x Y y Z z 0 [5.4]
wt wx wy
Comparing equations [5.3] and [5.4] shows that the vector (1, u, v, g) is located
in the plane tangent to the surface S. The solutions to the partial differential
equation are thus represented by surfaces tangential to the vector field 1, u , v, g in
the said 4D space. Now the curves tangent to the vector field (1, u, v, g) can be
obtained by integration of the system of three differential equations:
dx dy df
dt [5.5]
u v g
The curves defined by system [5.5] are called characteristic curves. We note
immediately that the first two differential equations define the trajectories of fluid
particles (section 3.3.2). The third differential equation allows the unknown function
f to be calculated on these curves. We thus obtain a Lagrangian (substantial)
representation of the balance of the quantity f associated with the fluid particles.
f G C f
GC
V V D
M
O y O y
M0
S S0
x x
(a) (b)
Figure 5.1. C: characteristic curve; S: solution surface;

S0: surface with initial conditions (Cauchy problem)
Note that the solution of equation [5.3] amounted to the solution of a system of
differential equations with initial conditions. In the preceding 4D space, any surface
S constituted of characteristic curves (characteristic surface) is tangent to the vector
G
field V (1, u, v, g); it is thus a solution of the partial differential equation (Figure
5.1a). If a trajectory point C belongs to a surface S, then the curve C lies entirely on
S.
In summary, any quasi-linear first order partial differential equation can be

interpreted as a transport equation for a quantity on the trajectories associated with
that equation.
This property allows the definition of the ideas of input (the surface S0 on which
the initial conditions are given) and output (any surface derived from S0 by
“translation” following the characteristic curves). This introduces a dissymmetry
between the input and output. Depending on the physical context of a problem, the
ideas of input and output may correspond to upstream and downstream, or to initial
and final conditions.
5.2.3. Comments
1) The preceding reasoning can be applied to any number of variables.

2) The preceding mathematical interpretation actually amounts to writing the
balance of the quantity F in Lagrangian variables. For any material particle M
which is displaced on this solution surface S of the 4D space, the associated volume
quantity satisfies:
wf wf wf
dt dx dy df 0 with : dx udt , dy vdt. [5.6]
wt wx wy
df
The balance equation g for each fluid particle describes the compatibility
dt
between relation [5.6] and partial differential equation [5.3]. The preceding
considerations show the equivalence between the Lagrangian balance formulation
in the form of differential equations [5.5], and the balance equation in Eulerian
variables, expressed in the form of partial differential equation [5.3].
3) In the presence of diffusion of the quantity F, the right-hand side of the
balance equations is not of the form g f , x, y, t , rather it contains the second order
transverse derivatives with respect to the characteristic curves: these derivatives
express an interaction between neighboring characteristic curves due to diffusion of
the quantity F. The introduction of higher order partial derivatives modifies the
properties found earlier. However, these partial derivatives are associated with a
coefficient which is often very small, and which leads to a reduction in the order of
the equation, except in singular zones (sections 6.4.3 and 6.5.3): nearly everywhere,
F is transported on the trajectories, with the source g taken into account.
5.2.4. The Cauchy problem for partial differential equations
We will reconsider the preceding problem in the mathematical form, which

consists of solving the Cauchy problem where the value of the unknown function is
given on the surface S0, and where we seek to calculate its value in the
neighborhood of S0 by means of the partial differential equation. This problem can
be solved by means of geometric interpretation (Figure 5.1b), provided that the

surface S0 does not comprise characteristic curves (characteristic surface).
In effect, we have seen that the calculation of the function f can be performed by
integration over the characteristic curves (trajectories): in order for this calculation
to be possible, these characteristic curves must obviously cross the surface S0 on
which the initial values of the function f are given.
Note that the solution of equation [5.3] is only defined in the domain D of the
space containing the characteristic curves which cross the surface S0: D is the
influence domain of the initial conditions given on S0. It is rigorously delimited.
This property is specific to all quasi-linear first order partial differential equations.
In summary, the Cauchy problem for a first order partial differential equation
was reduced to an ensemble of independent Cauchy problems for a system of
ordinary differential equations on each of the characteristic curves (trajectories). In a
transverse direction with respect to these curves, the partial differential equation
gives no information regarding the function f: the solution on characteristic curve C
has no influence on its neighboring points, except on C. The properties of the
solution in the neighborhood of a characteristic curve are uniquely fixed by the
initial conditions corresponding to this neighborhood. The solution space is thus
found to be as a bundle of fibers.
We will later return to the Cauchy problem (section 5.3.5.1) in a more local
manner, and one which is closer to the practical methods used for numerical
calculations.
5.3. Systems of first order partial differential equations
5.3.1. The Cauchy problem for n unknowns and two variables
Consider now a quasi-linear system of first order partial differential equations

with n unknown functions and which we will here limit to two independent variables
(x,t). The simplest equations of fluid mechanics involve at least three
thermodynamic quantities (for example, the density U, a velocity component u and
the pressure p) which we will represent by the vector F with three components
f1 , f 2 , f 3 . Consider the system of three equations described in vector form (the
case of n unknown functions can be treated in an identical manner):
[5.7]
We will adopt the position of the Cauchy problem: we assume that the value of F
is known on a curve C0 of equation M x, t 0 in the plane (x,t), and that from a
point (x0, t0) of this curve, we seek to calculate the value of the function F in its
wF
neighborhood: this is possible if we know the value of the partial derivatives
wx
wF
and at the point (x0, t0). The calculation of these is possible using data on the
wt
curve C0 and equation [5.7]. Designating the elemental arc G t , G x of the curve C0
and the growth G F of F on this arc, we have:
wF wF
Gt Gx GF [5.8]
wt wx
wF
Eliminating between [5.7] and [5.8] we have the system:
wt
§ G x · wF GF
¨B A¸ GA [5.9]
¨ G t ¸¹ wx Gt
©
Gx
If the determinant Q B A of the system [5.9] is non-zero, the unknown
Gt
wF
has a unique value. We can therefore obtain the value of F in the neighborhood
wx
of the considered point and the Cauchy problem has a unique solution. The
preceding determinant is called the characteristic determinant of system [5.7]. Its
value depends on the ratio G x G t O , in other words on the choice of the curve
C0.
Suppose now that the determinant Q of the system [5.9] is zero, O G x G t is a

root of characteristic equation [5.10]:
Gx
Q B A 0 [5.10]
Gt
The corresponding curve C0 is the characteristic curve associated with the root
O. When all the roots of the equation Q 0 are real, system [5.7] is called
hyperbolic (or totally hyperbolic).
When the characteristic determinant is zero, the system rank [5.9] has diminished
by one unit and a non-zero solution exists for system [5.9], if we have a
GF
compatibility relation between the components of the vector G A of the
Gt
right-hand side.
This relationship can be obtained, for example, from the non-zero left solutions L
(eigenfunctions) of the characteristic equation:
L B O A 0 [5.11]
Multiplying on the left side [5.9] by L and taking account of [5.11], we then
obtain the relationship sought between the components of GF on the characteristic
curve concerned:
§ G F ·¸
L¨ G A 0 [5.12]
¨
© G t ¸¹
In summary, characteristic curves of the plane (x,t), on which the characteristic

determinant cancels out, are such that:
– the given values of F do not allow the Cauchy problem to be solved
(calculation of F in their neighborhood);
– the unknown vector function F satisfies particular differential relations.
As in the case of solutions to quasi-linear partial differential equations, the last

property can be used to study the solutions. We will come back to this point a little
later.
NOTE – The relationship between the components of GF is of “Lagrangian” type in

the sense that the value of this variation is calculated for corresponding values of Gt
and of G x O d t : the evaluation point of G F is displaced at “velocity” O.
We will examine two particular cases which show the physical interest of these
results.
5.3.2. Applications in fluid mechanics
5.3.2.1. Unsteady 1D flow of a compressible inviscid fluid

The equations of a compressible perfect fluid in unsteady inviscid flow can be
written ([4.6], [4.22] and [4.55]):
wU wU wu
u U 0
wt wx wx
wu wu wp
U Uu 0 [5.13]
wt wx wx
wp wp wu
u U c2 0
wt wx wx
§U· §u U 0 · § 0·
¨ ¸ ¨ ¸ ¨ ¸
Letting: F ¨ u ¸; A I; B ¨ 0 u 1 U ¸; G ¨ 0 ¸ , they can be written
¨ p¸ ¨0 U c2 u ¸¹ ¨ 0¸
© ¹ © © ¹
in vector form [5.7]:
wF wF
A B G
wt wx
Equation [5.9] can thus be written:
wF GF
(B O I ) [5.14]
wx Gt
We can thus derive characteristic equation [5.10]:
uO U 0
Q 0 uO 1 U u O u O c u O c 0 [5.15]
0 Uc 2 u O
The three roots u,r c u of equation [5.15] correspond, respectively, to a

displacement of the fluid matter on the trajectory G x uG t , and to propagation
at the speed of sound c with respect to the matter which moves with velocity u
G x r c u G t .
The system for the left eigenfunctions L L1 L2 L3 of [5.11] can here be

written:
L1 u O
t
§ ·

LBOI ¨
L
¨ 1 U L 2 u O L3 U c 2¸
¸ 0 [5.16]
¨ L U L u O ¸
© 2 3 ¹
The vector L corresponding to each of the eigenvalues can be easily calculated.

Relation [5.12], which provides the relation between the variations GF of the
components of the vector F, can thus be written:
L.G F 0 [5.17]
1) For the root O = u, we obtain: L c 2

0 1 and:
§G U ·
LG F c 2 ¨

0 1 ¨G u
¸
¸ G p c 2G U 0 [5.18]
¨G p ¸
© ¹
Here we recover the relationship which characterizes isentropic transformations

(formula [1.27]): the entropy remains constant during matter displacement. The
reader should note that the identification of this simple property required some
mathematical developments.
2) For the eigenfunctions O r c u solution of system [5.16] gives:
¬
t
B L1c

L B I L1 B L2 c L3 c 0 i.e. L 0 o c 1

2

L2 B L3c ®
We thus obtain:
§G U · §G U ·
¨ ¸ ¨ ¸
L¨ G u ¸ 0 r U c 1¨ G u ¸ G p r U cG u 0 [5.19]
¨G p ¸ ¨G p ¸
© ¹ © ¹
These relationships correspond to acoustic waves which propagate in two

directions at the speed of sound c with respect to the fluid matter which moves with
velocity u.
NOTES –
1) Relation [5.16] for each of the roots Ocan be written in Lagrangian variables.
By means of linear combinations using initial system [5.13] of the equations of
motion describing the flow of a compressible perfect fluid, we can easily obtain the
system of equations:
dp dU d w w
c2 0 u
dt dt dt wt wx
Dp Du D w w d w
Uc 0 with : u c c [5.20]
Dt Dt Dt wt wx dt wx
~ ~ ~
Dp Du D w w d w
~ U c ~ 0 with : ~ u c c
Dt Dt Dt wt wx dt wx
This system is the expression in Eulerian variables of relations [5.18] and [5.19].
We furthermore obtain the directions of characteristic curves from the preceding
expression of derivatives in relations [5.20] [YIH 77, p.211].
2) The case of an incompressible fluid wu wx 0 can be obtained by letting the
speed of sound tend to infinity. The eigenvalues of the acoustic propagation
disappear, and only the eigenvalue G x G t u exists with the characteristic variable
GU G t 0 . The reader can verify that this result can be obtained from [5.13].
5.3.2.2. Steady 2D flow of an incompressible perfect fluid

The equations for this kind of flow can be written ([4.6] and [4.22]):
wu wv wu wu wp wv wv wp
0; U u Uv 0; U u Uv 0 [5.21]
wx wy wx wy wx wx wy wy
wF wF
and can be put in the form [5.7] ( A B G ) with:
wx wy
§ 1 0 0· §u· § 0 1 0· § 0·
¨ ¸ ¨ ¸ ¨ ¸ ¨ ¸
A Üu 0 1¸ ; F ¨ v ¸; B Ü v 0 0 ¸; G ¨ 0¸ [5.22]
¨ 0 U u 0 ¸¹ ¨ p¸ ¨ 0 U v 1 ¸¹ ¨ 0¸
© © ¹ © © ¹
Following the steps outlined in section 5.3.1 leads to the vector equation:
§ G y · wF GF
¨B A ¸¸ A [5.23]
¨ G Gx
© x ¹ wx
From this we can calculate the characteristic determinant Q BG x AG y and

the roots O G y G x . Equation [5.10] can here be written as:
O 1 0
Q U v O u 0 O
U uO v O 2 1 0 [5.24]
0 U v O u 1
Equation [5.24] only has one real root O G y G x v u which corresponds to

the differential equation of the trajectories. For this root, system [5.11] can be
written:
t
§ v u 1 0 · § L1v u ·
§ v· ¨ ¸ ¨ ¸
L¨¨ B A ¸¸ L1 L2 L3 ¨ 0 0 v u¸ ¨ L1 ¸ 0
© u¹ ¨ 0 ¨ L v u L ¸
© 0 1 ¸¹ © 2 3¹
hence: L L1 L2 L3 0 1 v / u .
Substituting into relation [5.12] gives:
§ 1 0 0 ·§ G u ·
¨ ¸¨ ¸
LAG F 0 1 v u ¨ U u 0 1 ¸¨ G v ¸ U uG u U vG v G p 0 [5.25]
¨ 0
© U u 0 ¸¹¨© G p ¸¹
The characteristic curves G y G x v u are the trajectories, and the variation of

the quantity U uG u U vG v G p is zero on these. We can recognize Bernoulli’s
first theorem for a perfect incompressible fluid (see section 4.3.2.3.1, relation [4.30])
which was obtained previously.
5.3.2.3. Steady 2D flow of a inviscid compressible fluid

The equations for a steady 2D plane of perfect compressible fluid can be written
([4.6], [4.22] and [4.55]):
wU wU wu wv wu wu wp
u v U U 0; Uu Uv 0;
wx wy wx wy wx wy wx
[5.26]
wv wv wp wp wp wu wv
Uu Uv 0. u v U c2 U c2 0
wx wy wy wx wy wx wy
or, in matrix form [5.7]:
§ w w ·
Ä B ¸F 0
¨ wx wy ¸¹
©
¬ u 0 0 ¬ v 0 0¬

u 0 0 1 0 0 0
; B
u v
with: F ; A .
v 0 0 u 0 0 0 v 1

p 0 c 2 2
® 0 u® 0 0 c v®
Characteristic equation [5.10] can be written:
v Ou UO U 0
Gy 0 U v O u 0 O
Q B A 0
Gx 0 0 U v O u 1
0 O U c2 U c2 v Ou
or:
>
U 2 v O u 2 O 2 u 2 c 2 2O uv v 2 c 2 @ 0 [5.27]
It has the same real root O v u as for the incompressible fluid, but here it is a
double root.
The two other roots satisfy a second degree equation whose discriminant can be
written:
'' >
c2 u2 v2 c2 t 0 @ if V2 u2 v2 t c2 [5.28]
These are real only if the flow is supersonic (V > c). If the flow is subsonic, the
roots and the characteristic curves are imaginary, as in the case of the
incompressible fluid. The values of the slopes of the characteristic curves are:
G y uv r c 2 [(u 2 v 2 ) c 2 ]
O [5.29]
Gx u2 c2
G
The velocity V is the bisectorGof the characteristic curves; in effect, taking the
axis Ox parallel to the velocity V (i.e. v = 0), slopes [5.29] of the characteristic
curves in the plane (x,y) are opposed:
Gy c2 1
O r r
Gx u2 c2 M 2 1
where M V c designates the local Mach number defined using the local sound
speed c.
The reader can easily verify that the characteristic curves lie at an angle T with
respect to the velocity vector (or the trajectory) defined by: sin T 1 M (we have
O tan T ); these curves are thus called Mach lines.
Identification of the relationship between the components of the differential GF

on each of the characteristic curves can be effected as previously by finding a vector
L which is a left solution of equation [5.11] and by substituting it into equation
[5.12]. We will here only show the calculation for the value O v u :
§ v ·
¨0 U U 0 ¸
¨ u ¸
§ v· ¨ v¸
L¨¨ B A ¸¸ L1 L2 L3 L4 ¨ 0 0 0 ¸ 0
¨ u¸
© u¹
¨0 0 0 1 ¸
¨0 v U c2 U c2 0 ¸¸
¨
© u ¹
or:
v v v
U L1 L4 U c 2 0; U L1 L4 U c 2 0; L2 L3 0.
u u u
From this we can calculate L3 and L4 as a function of the two arbitrary values of
L1 and L2:
v
L1 L4 c 2 0 L2 L3 0
u
Substituting into condition [5.12] gives:
§ Gp ·
L1u ¨ G U ¸ L U u G u U vG v G p 0
¨ ¸ 2
© U c2 ¹
As this equation is satisfied regardless of the values of L1 and L2, we have the
two relations:
§ ·
¨G U G p ¸ 0; U u G u U vG v G p 0
¨ U c2 ¸
© ¹
which describe the Lagrangian conservation of entropy and mechanical energy on

the trajectories (Bernoulli’s first theorem).
5.3.3. Cauchy problem with n unknowns and p variables
Now consider an n dimensional vector function F with components fi

i, 1,..., n functions of p variables xj j 1,..., p . The function F satisfies the n
first order quasi-linear partial differential equations:
wf A (i, A 1,..., n ; § wF ·
Ai jA gi 0 ¨¨ A G 0 ¸¸ [5.30]
wx j j 1,...; p) © wx ¹
We will now solve the Cauchy problem, in other words we will calculate the
variation G F of the function F in the neighborhood of a point M of a hypersurface
S0 of dimension p – 1 on which the values of F are given. This results in our
knowing p – 1 independent derivatives of the function F calculated following the
surface directions. The derivative of the vector function F following a transverse
direction S0 is not given, but it can be calculated from system [5.30].
In order to individualize this direction, we will perform a change of coordinates

xj x j [ k j , k 1,..., p .
wx j
The Wronskian (functional determinant) is assumed non-zero such that
w[ k
we can also write:
[k
[k x j j, k 1,..., p
This gives the following relationships between the partial derivatives of F with
respect to the two sets of variables:
wf A wf A w[ k (A 1,..., n ; § wF wF w[ ·
¨ . ¸ [5.31]
wx j w[ k wx j j , k 1,... p ) ¨ wx w[ wx ¸¹
©
System [5.30] can thus be written:
wf A w[ k i, A 1,..., n ; § wF w[ ·
gi Ä 0 ¸¸
AijA
w[ k wx j
0
j , k 1,... p ¨ w[ . wx G [5.32]
© ¹
We will choose the coordinates [k (k 2,3,..., p ) such that they describe the
surface (curvilinear coordinates on S0 (Figure 5.2)). The curve of coordinate [1
crosses the surface S0 whose equation is [1 [1 [ 2 ,...[ p
0 . The derivatives
sfA sk are known from the functions fA on the surface S0.
x 3
[1 S
[3 0
[
O
x 2
x 1
Figure 5.2. Initial surface of the Cauchy problem
All that remains is to determine the n unknown derivatives wf A w[1 in the

direction [1 by means of system [5.30] in which all the other terms in
wf A w[ k k z 1 are now known; system [5.30] can be written:
s1 sfA sf ¬ i, A 1,..., n ;

AijA #i A ,.. gi 0
sxj s1 sk ® j 1,...p; k 2,...p
[5.33]
¬
or: A sF < G 0 with: A A s1

sxj ®
A
s1
ij

where the function \ i is known from the derivatives which are already given
wf A w[ k k z 1 .
~
If the determinant of the matrix A is non-zero, system [5.33] with unknowns
wf A w[1 is of rank n and the n derivatives wf A w[1 can be uniquely determined.
The Cauchy problem thus possesses a unique solution.
~
If, on the other hand, we have A 0 , system [5.33] is of rank n – 1 and the
determination of the n unknown derivatives wf A w[1 is no longer possible.
~
Equation A
0 is a partial differential equation for the scalar function [1 x j
of p variables:
~ w[1
A AijA 0 [5.34]
wx j
We thus have a problem such as that described in section 5.2.4. The vector
w[1 wx j associated with the point M and satisfying equation [5.34] belongs to a
hypersurface (the cone C with summit M) of dimension p – 1 of the geometric space
(x1,…, xp). At the point M, this vector is the direction of the surface normal S0 of the
equation [1= 0; the tangent plane is normal to it, and the surface S0 is surrounded by
tangent planes whose normals belong to the cone C. This surface S0 is itself a cone
of summit M to which the characteristic curves passing through M are tangent.
~
As before, in the case A 0 , there exists a non-zero vector L such that
w[1
Li AijA 0 . Equation [5.33] thus results in a linear relation between the partial
wx j
derivatives wf i w[ k k z 1 :
ª § wf · º
Li «<i ¨ A ,..¸ g i » 0 i, A 1,..., n ; k 2,... p
«¬ ¨© w[ k ¸¹ »¼
5.3.4. Partial differential equations of order n
The same reasoning can be applied for an nth order quasi-linear partial
differential equation with p variables, or for a system of such equations. For
example, consider the quasi-linear second order equation whose coefficients and the
right-hand side term g depend on the function f, on its first derivatives and on the
variables xi:
w2 f
Aij g ( i,j 1,....,p ) [5.35]
wx i wx j
Suppose now that the values of the function f and of its first derivatives are given
at all points M of a hypersurface S0 of dimension p – 1. The values of the second
derivatives taken on this surface are known. We can perform the change of
coordinates described in section 5.3.3, such that the second derivatives of the
function are known on the surface S0 of equation [1 [1 [ 2 ,...[ p
0 in the
reference frame [ i . The change of variables [5.31] allows the terms of equation
[5.35] to be written in the form:
s 2f s 2 f sk sm
...(i, j 1,..., p )
sxi sxj sk sm sxj sxi
The calculation is performed as previously, and the equation thus transformed

w2 f
should allow the calculation of the unknown value of at the point where we
w[12
seek a solution of the Cauchy problem. We obtain:
w[1 w[1 w 2 f § wf w2 f · (i, j 1,..., p )

Aij \¨ f , , ,.. ¸ g 0 [5.36]
wxi wx j w[1 2 ¨ w[ w[ w[ ¸ (k 2,..., p )
© i i k ¹
where the function \ is a function of f, of its first derivatives and of its second
w2 f
derivatives k z 1 , which are known from the initial data given at all
w[ i w[ k
points M of the hypersurface S0. The coefficient of w 2 f w[12 must obviously be

non-zero for the calculation to be possible. If it is zero, we find a direction [1
corresponding to the characteristic surface [1 [ 2 ,...[ p
0 . These characteristic
w[1
directions D i satisfy the equation:
wxi
w[1 w[1
Aij Aij D i D j 0; (i, j 1,..., p ) [5.37]
wx i wx j
The left-hand side of equation [5.37] is a quadratic form which represents the
equation of a second degree cone comprising the normals to the characteristic
surfaces. If it is imaginary, we have an elliptic equation. If on the other hand it is
real, the equation is hyperbolic. The preceding equation, which is easy to write,
allows us to reduce our study of the type of a differential equation to the
identification of the nature of a quadratic form. Instead of searching for the principal
directions and performing an orthogonal change of reference frame, we can follow
the simpler procedure of directly identifying a reduced form for this quadratic form.
We will look at some applications in the next section.
The compatibility relation of equation [5.36] can be written for the solutions of
[5.37]:
§ wf w2 f · (i, j 1,..., p )
\¨ f, , ,.. ¸ g 0
¨ w[ i w[ i w[ k ¸¹ (k 2,..., p )
©
5.3.5. Applications
5.3.5.1. Quasi-linear partial differential equation

We will apply the preceding results to the flow of an inviscid fluid, limiting
ourselves to the identification of characteristic surfaces for the two-variable
examples already studied.
We immediately recover the results of section 5.2, by applying the preceding

results to equation [5.3]. Characteristic equation [5.37] can be written:
w[1 w[1 w[1

u v 0
wt wx wy
Its solutions [1 [1 t , x, y 0 are equations of surfaces comprising the

trajectories (the equation shows in effect that the surface normals are normal to the
vector (1, u, v) which is thus situated in the tangent plane of the surface).
5.3.5.2. System of partial differential equations with two independent variables

The characteristic curves obtained in section 5.3.1 can also be found using
characteristic equation [5.34]. This can be written as:
w[1 w[1
AiA BiA 0 i, A 1,..., n [5.38]
wt wx
In two dimensions, the characteristic “surfaces” are the curves [1 t , x 0 on

w[1 w[1 w[1 w[1
which we have G [1 Gt G x 0 . Eliminating and between
wt wx wt wx
this relation and [5.38], we recover equation [5.10] for the characteristic curves:
Gx
B iA AiA 0
Gt
5.3.5.3. Unsteady 2D flow of a perfect compressible fluid

Let us reconsider the study of section 5.4.2.3. with the two spatial coordinates
(x,y). The equations for an inviscid compressible fluid can be written ([4.6], [4.22]
and [4.54]):
d su sv du sp
0; 0
dt sx sy dt sx
dv sp d dp
0; c 2 0 [5.39]
dt sy dt dt
d s s s
with: u v
dt st sx sy
or in matrix form:
§d w w ·
¨ U U 0 ¸
¨ dt wx wy ¸
¨ d w ¸ §¨ U ·¸
¨0 U 0 ¸
¨ dt wx ¸ ¨ u ¸ 0
¨0 d w ¸¨v ¸
0 U ¨ ¸
¨ dt wy ¸ ¨© p ¸¹
¨ ¸
¨ 0 2 w 2 w d ¸
Uc Uc
¨
© wx wy dt ¸¹
The unknown function fi which has four components (U, u, v, p) is assumed to be

given on the surface S0 defined in the preceding section and characterized by the
coordinates [ k k z 1 . Setting x1 ct , x 2 x, x 3 y , we can perform the
change of variables [5.31]:
wf i wf i w[1 wf w[ k
i i 1,..,4; k 2,3
wx j w[1 wx j w[ k wx j
wf i
Separating the unknown derivatives we have:
w[1
§ d[1 w[1 w[1 ·

¨ U U 0 ¸
¨ dt wx wy ¸
¨ d[1 w[1 ¸ §U· §U·
¨ 0 U 0 ¸ ¨ ¸ terms ¨ ¸
¨ dt wx ¸ w ¨ u ¸ similar to w ü¸
0 k 2,3
¨ 0 d[1 w ¸ w[ ¨ v ¸ w[ k
¨v¸
0 U 1¨ ¸ ¨ ¸
¨ dt wy ¸ ¨ p ¸ (known) ¨ p¸
¨ ¸ © ¹ © ¹
¨ 0 2 w[1 2 w[1 d[1 ¸
¨ Uc Uc
© wx wy dt ¸¹
wf i
This system of equations does not have a unique solution if its determinant
w[1
is zero:
d[1 w[1 w[1

U U 0
dt wx wy
d[1 w[1
0 U 0 2ª 2 2 2º
§ d[1 · § ·
dt wx U 2 ¨¨ ¸¸ «§¨ d[1 ·¸ c 2 ¨ w[1 ¸ c 2 §¨ w[1 ·¸ » 0
d[1 w ¨
«© dt ¹ ¸ ¨ ¸ ¨ ¸
0 0 U © dt ¹ © wy ¹ © wx ¹ »
¬ ¼
dt wy
2 w[1 2 w[1 d[1
0 Uc Uc
wx wy dt
The characteristic equation can be decomposed into two first order partial
differential equations:
d[1
1) A quasi-linear equation 0 whose solution we have already studied in
dt
section 5.2.2. We have seen that the characteristic curves are the trajectories, the
corresponding characteristic surfaces being defined from these.
2) The partial differential equation:
2 2 2
§ 1 d[ 1 · § w[ · § w[ ·
¨¨ ¸¸ ¨¨ 1 ¸¸ ¨¨ 1 ¸¸ 0
© c dt ¹ © wy ¹ © wx ¹
which can be easily interpreted by taking a reference frame for which velocity (u, v)
of a point M is zero at time t (this reference frame has the speed of the matter of
point M at the considered instant). In this last frame the previous substantial
d[1 w[1
derivative is equal to , so that the previous equation relation
dt M , t wt M , t
between derivatives of function [1 is written:
2 2 2
§ w[ · § · § ·
1 ¨ 1 ¸ ¨ w[1 ¸ ¨ w[1 ¸ 0
¨¨ wy ¸¸
c ( M , t ) ¨© wt
2 ¸
M ,t ¹
¨ wx
©
¸
M ,t ¹ © M ,t ¹
ª w[ w[ w[ º
showing that the vector « 1 , 1 , 1 » is located on the revolving
« wt M , t wx M , t wy »
¬ M ,t ¼
cone of summit M, of axis Ot and whose equation is:
1
2
t t M 2 x x M 2 y y M 2 0
c (M , t )
The tangent planes at point M to the characteristic surfaces are normal to the
ª w[ w[ w[ º
vector « 1 , 1 , 1 » , which makes an angle Arc tan[1 c( M , t )]
« wt M , t wx M , t wy »
¬ M , t ¼
with the axis Ot. They envelope the complementary cone of revolution whose angle
with axis Ot is equal to Arc tan[c (M , t)] and whose equation is:
c 2 ( M , t )t t M 2 x x M 2 y y M 2 0
The characteristic curves comprised by the cone satisfy the relations

dr c 2 dt 2 0 which describe radial propagation at the speed of sound c with
2
respect to the matter.
5.3.6. Physical interpretation of propagation
The partial differential equations of fluid dynamics and transfer are balance
equations; we have now outlined the essential ideas which govern the manner in
which material quantities are displaced on the characteristic curves either by
convection (transport of material quantities by fluid particles) or by propagation.
The latter mode results from the exchange of extensive quantities between fluid
particles, from one to the next and so on. It can be simply interpreted.
Consider a string of coupled oscillators comprising the masses m and springs of

stiffness k and length 'x. Let xn be the coordinate where the nth mass is at rest, and
let [n be its displacement with respect to this rest position (Figure 5.3).
m k m k m k m k m
x
xn-1 [n-1 xn [n xn+1 [
n+1
Figure 5.3. Propagation on a line of mass-spring oscillators
The equation of motion for the nth mass can be written:
m[n k 2[ n [ n 1 [ n 1 0 [5.40]
Consider this string to be a model for a continuous medium with spatial

discretization x n 1 x n x n x n 1 ... 'x .
The second spatial derivative can be approximated by:
w 2[ [ n 1 [ n 1 2[ n
2
wx 'x 2
such that by letting c 'x k m 3, equation [5.40] becomes a wave equation:
1 w 2[ w 2[
0
c 2 wt 2 wx 2
The propagation of waves results from interactions between the mass of the
medium and its compressibility (or what is equivalent, its elasticity). In the
continuous compressible medium, the mass and the elasticity are uniformly
distributed.
3 The reader can verify that stiffness per unit length is k'x and mass per unit length is m/'x,
so that by identifying the mass and the stiffness with corresponding properties of gas pressure
and specific mass), we obtain the value of the sound velocity c wp wU S .
5.4. Second order partial differential equations

5.4.1. Introduction
We have just shown that a system which flows generally presents at least one
family of characteristic curves (its trajectories) which correspond to the transport of
matter and on which there exists a balance relation for the extensive quantities
(entropy, mechanical energy, etc.). In many practical cases, the flowing fluid may
possess properties of homogenity, either dynamic (absence of vorticity) or physical
(constant entropy). If we have strict conservation of this quantity everywhere in the
flow, then the corresponding partial differential equation can be immediately
integrated. For example the steady 1D flow studied in section 5.3.2.1 is
homoentropic, which leads to the existence of the relation U U p .
We can thus often obtain a quasi-linear second order partial differential equation
(i.e. linear with respect to the second derivatives) for one of the quantities of the
problem by using a system of first order partial differential equations.
For example, let us assume for the sake of simplicity that the density and
velocity variations are small enough for the linearization of equations [5.13] to be
possible in a constant entropy medium:
wU wu wu wp
U 0; U 0
wt wx wt wx
Using the definition of the speed of sound c 2 wp wU S , here assumed to be

constant on account of the linearization, we obtain the wave equation:
w2 p 1 w2 p
0
wx 2 c 2 wt 2
We will now reconsider quasi-linear second order partial differential equations

with two variables, in a form largely used in practice when a velocity potential exists
(the homentropic flow of a compressible fluid, waves on the surface of liquids, etc.;
see Chapter 6). Furthermore, as their characteristic equation is of second degree, it is
locally of a well-defined type, elliptic or hyperbolic depending on whether the roots
are imaginary or real. This facilitates a discussion of a problem’s boundary
conditions.
Consider the quasi-linear second order partial differential equation with two
variables:
Ar 2 B s C t D [5.41]
with the usual notation:4

fx p fy q f xx r f xy s f yy t
The coefficients A, B, C and D are functions of the unknown function f and of its
first derivatives p and q:
A A f , p, q, x, y ; B B f , p, q, x, y ;
C C f , p, q, x, y ; D D f , p, q, x, y .
5.4.2. Characteristic curves of hyperbolic equations
Because of the practical applications of this formulation, we will reconsider the

Cauchy problem, which consists here of determining the solution using data for f
and its derivatives (p,q) on a curve C0 in the plane (x,y). In order to know the
function f at all points in the neighborhood of the point (x,y) in C0, it suffices to
know the second derivatives (r, s, t) at that point (we can then calculate the higher
order derivatives in a similar fashion by successive differentiations of equation
[5.41]). The functions (r, s, t) satisfy the relations:
Ar 2 Bs Ct D; rG xsG y G p; sG xt G y Gq [5.42]
or, in matrix form:
§ A 2 B C ·§ r · § D ·
¨ ¸¨ ¸ ¨ ¸
¨ G x G y 0 ¸¨ s ¸ ¨G p ¸ [5.43]
¨ 0 G x G y ¸¨ t ¸ ¨G q ¸
© ¹© ¹ © ¹
As a function of the known variations G p, G q of p, q on the curve C0

between the points x G x, y G y and (x, y) on this one, we can generally
calculate the quantities r , s, t at (x, y), except if the characteristic determinant of
system [5.43] is zero:
A 2B C
x y 0 0 or: C x 2 2B x y A y 2 0 [5.44]
0 x y
4 In order to simplify the writing in section 4.6 and in any other case when it will be useful,
partial derivatives w/wx, w/wz will be written fx, fz,, a notation that does not allow any mistake
in mathematical calculations.
Gy B r B 2 AC
The directions , solutions of the preceding equation, are
Gx A
the characteristic directions tangent at each point to the characteristic curves:
– if B 2 AC ! 0 the roots of the characteristic equation are real and the

equation is of a hyperbolic kind;
– if B 2 AC 0 the roots of the characteristic equation are imaginary and the

equation is of an elliptic kind; the Cauchy problem always possesses a unique
solution in the neighborhood of any curve on which the values of the functions (f, p,
q) are given;
– if B 2 AC 0 the characteristic equation possesses a double root, and the
equation is of a parabolic kind.
In the hyperbolic case, the determination of (r, s, t), which is non-unique on the
characteristic curves, is only possible if the system of equations [5.42] is of rank 2.
There then exists a relation between D, Gp and Gq which can be obtained, as
outlined in the preceding sections, by searching for a vector L which is a left
solution of the system LM 0 , where M is the system matrix [5.43] without the
right-hand side. We find:
L G xG y, AG y,CG x
By multiplying the left-hand side of system [5.43] by L we obtain the following
relation for each of the solutions G y G x i of characteristic equation [5.44]:
§G y · §G q · § G y · §G p ·
D¨ ¸ C¨ ¸ A¨ ¸ ¨ ¸ 0 i 1,2. [5.45]
¨G x¸ ¨Gx¸ ¨G x ¸ ¨Gx ¸
© ¹i © ¹i © ¹i © ¹i
Relations [5.45] allow the function f and its derivatives (p, q) to be calculated
from place to place in the following manner. Consider the subdivision ABDEF of an
arc of the curve C0 and trace at each of the points the two families of characteristics
*1 and *2 of slope G y G x i , (i=1, 2). The different families of characteristics
obtained from A and B intersect at G (Figure 5.4). We obtain relation [5.45] on each
of the arcs AG and BG, which allows the values of pG and qG to be calculated as a
function of their values on the curve C0 at A and B. Assuming the arcs to be
sufficiently small, these relationships can be written:
§G y · §G y ·
D¨ ¸ xG x A C qG q A A ¨ ¸ pG p A 0
¨Gx¸ ¨Gx¸
© ¹1 © ¹1
§G y · §G y ·
D¨ ¸ xG x B C qG q B A ¨ ¸ pG p B 0
¨G x¸ ¨G x¸
© ¹2 © ¹2
§G y · §G y ·
the elementary arcs AG and BG having respective slopes ¨ ¸ and ¨ ¸ .
¨G x¸ ¨G x¸
© ¹1 © ¹2
Characteristics *1
M
F J
y E I
D H
B Characteristics *2
x G
A
C0 (a)
Figure 5.4. Domain of influence of the arc AF of the initial curve C0
The preceding relations uniquely determine pG and qG as a function of the values

of f, p and q at A and B. We thus see that the initial values on the curve C0
propagate partially on each of the characteristics by virtue of relation [5.45]. The
value of f at G is determined by the mean of the finite variations formula between
the points A (or B) and G:
G f fG f A p A x G x A q A y G y A .
The values f, p and q of the solution of the partial differential equation can thus
be determined from place to place by the preceding procedure at any point within
the curvilinear triangle AFM, which is delimited by the arc AF and the arcs AM and
FM of the characteristic curves: in effect, it suffices for this point to be attained by
progressing along the two families of characteristic curves starting from the initial
arc AF. The inner domain of the curvilinear triangle AFM is called the influence
domain of the arc AF of the initial curve C0.
This method is not applicable to parabolic equations: by considering the case B

= C = 0 in equation [5.41], relation [5.44] gives G y= 0, and system [5.45] gives no
new relation on the characteristic curves comprising the axis Ox. We will return to
discuss the parabolic equation in section 5.4.5.4.
In summary, the solution of the Cauchy problem is always possible for an elliptic
equation regardless of the choice of initial data. On the other hand, the existence of
real characteristic curves or surfaces implies a propagation of function values along
these curves.
5.4.3. Reduced form of the second order quasi-linear partial differential equation
We will demonstrate that a second order quasi-linear partial differential equation

can be locally changed, at all points, to a standard reduced form.
Recall firstly that a quadratic form Am 2 2 Bmn Cn 2 can be written in a

reduced form by means of an appropriate change of basis. In effect, by considering
the new variables (P, Q):
m cP dQ n eP fQ
the preceding quadratic form can be written as a function of these:
Am 2 2 Bmn Cn 2 DP 2 2 EPQ JQ 2 [5.46]
By appropriately choosing the coefficients (c, d, e, f) of basis change, we can

eliminate E and make equal the absolute values of D and J. Let us apply this
procedure to equation [5.41], which can be associated with the quadratic form
Am 2 2 Bmn Cn 2 whose coefficients are functions of the quantities (f, p, q, x,
y).
The change of coordinates

[ M ( x, y ) K \ ( x, y ) F [ , K f x, y
transforms equation [5.41] into another equation of the same kind. Showing
explicitly only those terms containing second order partial derivatives, we have:
fx F[ M x FK\ x ; fy F[ M x FK\ y
f xx F[[ M x2 2 F[K M x\ x FKK\ x2 ...
f xy
F[[ M xM y F[K M x\ y M y\ x FKK\ x\ y .....
f yy F[[ M y2 2 F[K M y\ y FKK\ 2y .....
The left-hand side of equation [5.41] can be written:

A fxx 2B fxy C fyy F 2 F F ....
with: !x A#x B#y
!y B#x C#y
[5.47]
A!x 2 2B!x !y C!y 2
A#x 2 2B#x #y C#y 2
We can immediately verify the relation:
DJ E 2 AC B M \
2
x y
M y\ x 2
which shows that the discriminant of the quadratic form retains the same sign after
the coordinates change.
The reduction to the normal form can be obtained by letting D J and E = 0.

This last condition is satisfied by letting:
Mx
M . B\ x C\ y My
M . A\ x B\ y [5.48]
By replacing in [5.47] the derivatives of M with the preceding expressions, we

can show that the coefficient D can be written:
D
M 2 AC B 2 A\ x2 2 B\ x\ y C\ y2
M 2 AC B 2 J [5.49]
Depending on the value of AC B 2 , we can distinguish the following cases:

– Elliptic case: AC B 2 ! 0 .
Letting M 2 AC B 2
1 , we have D J . The characteristic equation
[5.44] does not have a real solution and equation [5.41] is elliptic; as the coefficient
D is non-zero, the second derivatives can be regrouped in the form of a Laplacian:
Ar 2 B s C t Af xx 2 Bf xy Cf yy D F[[ FKK .... D
– Hyperbolic case: AC B 2 0 .

Letting M 2 AC B 2
1 , we have D J . The reduced form of the
equation [5.41] can be written in the form of a wave equation:
Ar 2B s C t Af xx 2 Bf xy Cf yy D F[[ FKK .... D.

Another reduced form of the hyperbolic equation can be obtained by

alternatively choosing the functions M and \ to represent the two families of
characteristic curves [ I ( x, y ) const and K \ ( x, y ) const; in the place of
relations [5.48], we let the tangent slopes of these curves equal the roots of
characteristic equation [5.44] (the characteristic curves are taken as local coordinate
curves):
§G y · Mx §G y · \x
¨ ¸ ; ¨ ¸ .
¨G x¸ My ¨G x¸ \y
© ¹1 © ¹2
It now follows from equation [5.44]:
D J C M y2 2 BM xM y A M x2 A\ x2 2 B\ x\ y C\ y2 0
By substituting the product and the sum of the ratios M x / M y and \ x / \ y by

their expressions obtained from the characteristic equation [5.44], we obtain the
non-zero coefficient E defined by [5.47]:
§ B 2 ·¸
E
AM x\ x B M x\ y M y\ x CM y\ y 2M y\ y ¨ C
¨
© A ¸¹
Equation [5.41], now considering the characteristic curves, can be written:
F[K D'
where D' is a function of coordinates, and of the values of F, and of its first
derivatives.
–Parabolic case: AC B 2 0 .
We take M such that M x O1M y 0 to obtain D = 0; it follows that E = 0; and
the normal form of the parabolic equation is then:
F ... D
We will return to the properties of the parabolic equation when we consider it in

terms of constant coefficients (section 5.4.5.4).
NOTE – As pointed out (section 5.4.1), we have verified that second order partial
differential equations no longer have characteristic curves which represent
trajectories, even though they do represent flows.
5.4.4. Second order partial differential equations in a finite domain
5.4.4.1. The significance of the Cauchy problem

In systems of quasi-linear partial differential equations, we can have mixed
situations: for example, the flow of the incompressible fluid discussed in section
5.3.2.2 only represents a single family of characteristic curves, on which only
mechanical energy is transported, but no other quantity propagates. On the other
hand, the quasi-linear second order partial differential equations lead to two
principal kinds of local situation:
– second order elliptic equations always possess a solution to the Cauchy
problem, which implies that the initial data have a significant influence on the
solution in their neighborhood;
– hyperbolic equations lead to a double structure associated with two families of
characteristic curves on which the initial information is transmitted.
While elliptic equations distribute information in all directions, hyperbolic

equations transmit it along the “fibers” of two bundles of curves. However, as the
elliptic or hyperbolic character is a local property, an equation can be hyperbolic in
one region of space and elliptic in another.
Our discussion of the Cauchy problem shows us that the simultaneous presence
in a flow of subsonic and supersonic zones leads to very different modes of
transmitting information and to certain contradictions; this results in important
difficulties regarding the boundary conditions which must be imposed, which are
different in the two cases (section 5.4.5). This situation often leads to the presence of
shock waves. We will consider a simple example by studying the flow of a
compressible fluid in a nozzle (section 5.5.4).
The understanding of these situations and of these properties is particularly

important, not only for the discussion of physical phenomena, but also for numerical
calculations whose algorithms must be chosen such that numerical information is
transmitted in a manner which is compatible with the general properties which we
have just outlined.
5.4.4.2. Constant coefficient second order partial differential equations

When the coefficients A, B and C are constants, the nature of the partial
differential equation is identical in all parts of the domain studied.
Constant coefficients elliptic equations with no right-hand side can be expressed

as a Laplace equation:
w2 f w2 f
0
wx 2 wy 2
This equation is encountered in problems of potential flow of incompressible

fluids, and in problems of thermal conduction or mass diffusion in steady flows. The
presence of a right-hand side (Poisson equation) implies the existence of sources for
the quantities studied.
Constant coefficients hyperbolic equations with no right-hand side take the form
of the wave equation:
w2 f w2 f
0
wx 2 wy 2
in which the y variable is often the time. The characteristic curves are the straight
lines [ y x constant and K y x constant By performing the change of
variables x, y o [ , K , the equation can be written:
w2 f
0
w[ wK
Its general solution can be written using two arbitrary functions ) and <:
f [ , K ) [ < K )x y < x y
The reduced form of the parabolic equation is the heat equation:
w2 f wf
0
2 wy
wx
in which the y variable is very often the time.
5.4.5. Second order partial differential equations and their boundary conditions
As physical problems do not usually occur in the form of a Cauchy problem, on
account of the presence of boundaries (walls or other surfaces) surrounding more or
less completely the fluid domain under study, and on which we must impose specific
conditions. The properties discussed in the preceding sections regarding the

characteristic curves or surfaces show that these depend on the nature of the system
of partial differential equations. We will first examine the case of a quasi-linear
second order equation with constant coefficients.
5.4.5.2. Elliptic equations

The Cauchy problem, which involves specifying the function values and its first
derivatives on a curve C0, generally allows the unique determination of the solution
in the neighborhood of this curve C0. However, physical problems are not given in
this way for elliptic equations, and we generally have a problem posed in a domain
on the boundary of which are given either:
– the unknown function (Dirichlet problem);
– the function’s normal derivative at the boundary (Neumann problem);
– or a mixed condition in the form of a linear relation between the function and
its normal derivative (mixed problem).
Problems involving heat conduction in a steady flow regime have already been
discussed in section 2.3.2. Steady subsonic flow of an inviscid fluid about an
obstacle is a Neumann problem (Chapter 6).
Elliptic problems cannot deal with propagative phenomena; they can generally
only represent steady phenomena or unsteady situations where propagation plays a
negligible role (domains which are small compared with a wavelength for example).
We will come back to discuss this point on numerous occasions.
In the case of the Neumann problem, a compatibility condition exists as a result

of Otrogradski’s theorem. For example, the Poisson equation of a steady 2D heat
conduction problem in a domain D of the plane (x,y):

div O gradT VT
imposes on the curve C, which bounds D, the integral condition:
³C O
wT
wn
dA
G
³C O gradT .n dA
³D div O gradT ds ³D V T ds
This expresses the global conservation of energy, where the thermal flux
crossing the curve C must be equal to the thermal power generated in D under
steady conditions.
Under usual conditions, the solution of an appropriately posed problem, of one

of the preceding kinds (Dirichlet, Neumann or mixed), is unique and elliptic
equations do not lead to difficulties, provided of course that they correctly represent
the physical problem studied.
5.4.5.3. Hyperbolic equations

For equations of this kind, the discussion of section 5.4.2 outlined the ideas of
propagation and influence domain. The concept of the Cauchy problem is here
degenerated to a certain extent, due to the structure of the solutions. As already
discussed in section 5.2.4, the initial data have no influence in transverse directions
with respect to the characteristic curves.
In addition to the initial conditions, we often impose other boundary conditions.

For example, the presence of an obstacle in a flow leads to the conditions
wf wn 0 on its wall, associated with the equation for the velocity potential f. This
condition leads in particular to a reflection of waves on the considered wall, an idea
which we will encounter for diverse phenomena (acoustic, supersonic flows, etc.).
In practice, the Cauchy problem is rather academic, as physical problems are

always posed in limited domains at the boundaries of which the boundary conditions
are generally defined by external data. However, its study (section 5.4.2) has
allowed us to define the characteristic curves and to show that these strictly limit the
influence domain of conditions given on a bounding curve. Conversely, the value of
an unknown function at a point M cannot depend on data outside the influence
domain formed by the characteristic curves which arrive at this point (Figure 5.5).
The notion of information transmission is closely associated with these
considerations: the characteristic curves constitute the means of information transfer
in the medium considered.
t y C1
M B
M
D
D C2
x A x
A B
(a) (b)
Figure 5.5. Domain D influencing the value of the unknown function f in M
The boundary conditions associated with the wave equation can often be
considered as initial conditions. However, the direction of travel of the characteristic
curves is not important for a wave equation which is invariant if the time direction is
changed (section 1.1.1.3). Figure 5.5a shows an example of initial data for an
acoustic problem (given at time t = 0), while Figure 5.5b corresponds to velocity
data given for a supersonic steady flow in the plane (x,y) (which can be reduced
under certain particular conditions to a wave equation).
5.4.5.4. Parabolic equations

We have seen in section 5.4.2 that a double characteristic curve exists on which
there is no longer a propagation relation. Taking it as a coordinate curve, we obtain a
heat equation [5.50] (see section 5.4.3) whose behavior we will study (we let t be the
parabolic variable y which we will interpret as time):
w2 f wf
[5.50]
2 wt
wx
We have already seen, in section 1.1.1.4, the consequences of asymmetry

between the variables x and y; in particular, the variable y must evolve towards
increasing values (from past to future, or from upstream to downstream, etc.).
We will now treat the problem of a thermal shock5 in a semi-infinite medium. We

will let f be the temperature in equation [5.50] and consider the thermal problem
which corresponds to a unit temperature step function imposed at the initial instant
t 0, at the origin x 0 of a semi-infinite medium ( x t 0) , initially at temperature
f = 0. We impose therefore the boundary conditions:
xz0 t 0 f x ,0 0
[5.51]
x 0 t!0 f (0, t ) 1
The scales X and Y corresponding to the variation F of f in [5.50] are:
The preceding estimate leads to the definition of the variable K x 2 t and

we look for a solution of the form6 f x, t M K . Substituting this function into
equation [5.50], we have:
5 The word “shock” is relative here to an initial condition and not to a discontinuity appearing
in the solution.
6 We have here a self-similar solution, i.e. which is invariant with respect to a group of affine
transformations. Flow equations, which are exact or approximate for high Reynolds numbers,
can have such solutions ([SCH 99], [YIH 77]).
f ' '2K f ' 0
from which we obtain:
f ' K A exp K 2
Taking account of boundary conditions [5.51] leads to:
2 d 2
f (x, t)

¨ x 2 t
exp (-u ) du 1 erf ( x 2 t ) [5.52]
2 x 2
with: erf x

¨ 0
exp (-u ) du (erfd 1) .
Figure 5.6 shows the temperature diffusion imposed at the origin, whose
influence can be felt in a zone, whose width, in the order of X, is proportional to t1/2
(curves 1 to 4 for different increasing times). The dimensional presentation of the
problem will be treated in section 8.3.2.2.2.
1 2 3 4
0 x
0
Figure 5.6. Influence zone of a unit step function (of temperature) imposed at the origin
We thus see that the behavior of the parabolic differential equation lies
somewhere between that of the hyperbolic and elliptic equations:
– the double characteristic curve is a preferred axis for the transmission of
information;
– the form of solution [5.52] shows an instantaneous action at all points of the
axis Ox: the notion of propagation has completely disappeared;
– we have “diffusion” (transverse spreading) of the initial data about the
characteristic curve.
The same initial boundary conditions applied to the wave equation would lead to
an acceptable solution for negative values of y, which is not the case here: the heat
equation describes an irreversible phenomenon, in other words the impossibility of
reversing the time direction (section 1.1.1.3).
5.4.5.5. Mixed equations

In section 5.4.2 we saw that, depending on the quantity AC B 2 , quasi-linear
partial differential equation [5.43]:
w2 f w2 f w2 f
A 2B C D
wx 2 wxwy wy 2
can be either elliptic, hyperbolic or parabolic. The nature of the solution depends
therefore on the local coefficients value. The simplest example of a mixed equation
is the Tricomi equation:
w2 f w2 f
x 0
wx 2 wy 2
For negative values of x this equation is elliptic, while for positive values, it is
hyperbolic. It is a summary model of the situation described in section 5.3.2.3,
where, in a subsonic flow, the problem is elliptic, while a supersonic flow has a
hyperbolic character. The passage from a subsonic flow to a supersonic flow
happens in a transonic range, which may be locally represented by a Tricomi
equation (the passage from the reduced form of the elliptic equation to the reduced
form of the hyperbolic equation assumes that a coefficient changes sign when going
to zero):
y
elliptic hyperbolic
zone zone
O x
C0
Figure 5.7 Mixed problem (Tricomi equation)
Consider a flow such that it is subsonic for negative values of x and supersonic
for positive values. We assumed that the speed of sound is attained by the flow
velocity at x = 0. Determination of the solution in the elliptic zone is possible using

the boundary conditions on the curve C0 (Figure 5.7) and the condition
AC B 2 0 here corresponds to sonic velocity on the boundary x = 0 between the
subsonic zone and the hyperbolic zone. In the supersonic zone, the flow is
determined by the upstream conditions. So, as any real subsonic flow is generated
by an ensemble of conditions upstream and downstream, the presence of a
supersonic zone leads to a contradiction corresponding to the presence of shock
waves, which we will now examine.
Note that the coefficients of partial differential equation [5.42] are not only a
function of the coordinates, but also of the unknown function and its first
derivatives. Contrary to the Tricomi equation, the nature of the equation and the
(associated) boundary conditions that must be associated are not easily predictable
in general, as they depend on the solution values.
5.5. Discontinuities: shock waves
5.5.1. General considerations
The presence of shock waves may feature in the varied conditions of supersonic
flows of compressible fluids. We will show, using examples, that the existence of
characteristic curves or surfaces and the nature of the solutions to the hyperbolic
equations lead to the possibility of discontinuities.
We will first of all study two examples (1D flow of an inviscid compressible
fluid and steady supersonic flow) for which a discussion of the characteristic curves
leads to the impossibility of a continuous flow. We will then consider a steady 1D
flow of an inviscid fluid in a nozzle which is governed by a differential equation
which again shows the necessity of shocks.
5.5.2. Unsteady 1D flow of an inviscid compressible fluid
Let us take the example of a one-dimensional flow of a compressible fluid

verifying equations [5.13] (section 5.3.2.1). The characteristic transport curve
G x G t u ensures constant entropy at all points, which leaves only two
independent variables p and u. This condition leads to the relations:
J 1 J J 1 2J
p ( x, t ) p0 § p · § p ·
const; c J p U J ¨ ¸ c0 ¨ ¸
U J ( x, t ) U 0 ¨© p 0 ¸¹ ¨p ¸
© 0¹
in which (p0 , U0) are the initial conditions of the fluid for a zero velocity.
Let us now apply the method of characteristics on a grid discretized as defined in

section 5.4.2. Starting from two neighboring points R and S whose physical
quantities (pR, uR) and (pS, uS) are known, and which are not located on the same
characteristic curve (Figure 5.8a), we trace out two characteristic curve segments
with respective slopes r c u which intersect in plane (x, t) at the point T whose
the values of the unknowns (pT, uT) are calculated using the relations
G p r U cG u 0 (with notations of section 5.4.2) taken on the previous curves
segments. These are written:
pT p R U R c R uT u R 0; pT p S U S c S u T u S 0
These relations can be used for calculation of the values (pT, uT) for pressure and
velocity at point T. At first order, we can take U R c R U S c S U c , so that we have:
1 1 ¡ 1 ¯
pT pR pS c uS uR
¯± uT uR uS pS pR
°
2¢ 2¢ ¡ c °
±
We can then deduce from thermodynamic relations with constant entropy, the
specific mass UT and the sound velocity cT at point T.
Let us consider the case where pressure and velocity take uniform values on the
characteristic curve issued from point R, which having then a constant slope c + u, is
a straight line. If R' is a neighboring point of R on this characteristic curve, at point
T and at the intersection T' of characteristic curves issued from R' and S, we have:
pT p R U cu T u R 0;
pT ' p R ' U c u T ' u R ' 0.
Taking out the first relation from the second relation, we obtain the relation on
characteristic curve of slope c + u: between points T and T':

pT ' pT U c u T ' u T 0
So, between points T and T' we obtain the relation on characteristic curve of
slope c + u:

pT ' p T U c u T ' u T 0
We then obtain the relation:

pT ' pT
U c uT ' uT 0
which shows that pressure, velocity and sound velocity values are the same at points
T and T'. Going on from place to place on characteristic curve of slope cS + uS
issued from S, we see that the flow properties are identical along this curve, which is
also a straight line. The corresponding flow (a simple progressive wave) is then
characterized by the propagation along axis Ox with velocity cS + uS of pressure 'p
and velocity 'u variations between the two characteristic curves going through R
and S. These variations verify the relation:
'p U c 'u
The previous calculation can be continued on the next characteristic curves of

the same family which will possess the same property of simple progressive waves.
The reader can easily show that another family of simple progressive waves exists
associated with the other family of characteristic curves of slope c + u. For these
waves, propagating towards the negative part of axis Ox, we have the relation
'p U c 'u .
x x = c0t + OA uF
t
F R'
E R T' uC
D T
S t
C I
B H M
K uO
A G
N t t
O P Q
Characteristic curves
(a) (b)
Figure 5.8. (a) Application of the characteristics method and evolution of the
characteristic straight lines for a decreasing pressure (or a negative velocity);
(b) evolution of the velocity at the points O, C and F
Now consider the domain formed by the first quadrant of the plane (x, t) with the
following boundary conditions:
t 0:x ! 0 p x ,0 p0 u x ,0 0
t t 0:x 0 p0, t Pt u 0, t U t
The domain thus defined is the influence domain of the given conditions on the
positive sides of the axes Ox and Ot. The positive side of the axis Ox is at rest at the
initial instant and we apply pressure and velocity variations at the origin x = 0, as a
function of time, which will then propagate on the Ox axis.
As previously, we can apply the method of characteristics by starting from the

points O, A, B, C, D, E,…,P, Q, etc., of the Ox and Ot axis (Figure 5.8a), and
calculating the values of the unknowns p and u using the relations G p r U cG u 0
taken on appropriate segments of the characteristic curves of slopes o c + u .
However, the initial state, being uniform and at rest, can be considered as a
regime of simple progressive waves. Indeed, by applying previous relations on
characteristic curves issued from points A, B, C, D, E, etc., we see that the initial
rest data are transmitted to points H, I, etc., on the straight lines issued from
previous points on all parts of the plane above the straight line x = ct + OA.
Previous results show that propagation can only proceed with simple waves for
increasing times. It can be verified that the values (pO, uO) at point O are also
obtained from place to place for all points of the characteristic curve x =cO t issued
from O, on which the fluid state is uniform. Calculation is then carried out at points
P, Q, etc., of axis Ot, and it can be seen that we have a flow with simple progressive
waves, propagating on the characteristic straight lines of a positive slope.
The difference of pressure and velocity values between two neighboring

characteristic curves verify the relation ' p U c' u of simple progressive waves.
This condition must be verified at points P, Q, etc., of axis Ot, so as not to create an
inverse propagation towards negative times on characteristic curves of negative
slopes: this should be not acceptable according to the physical aspects of problem.
This compatibility condition is a consequence of the definition of the initial data
curve itself, since there exist characteristic lines of a negative slope issued from axis
Ox intersecting axis Ot, which involves the relation ' p U c' u .
The simplest practical realization of the preceding flow involves imposing a

suitable velocity on the fluid matter by means of a moving piston. We will leave it to
the reader to determine the domain thus defined in the plane (x,t) for this Lagrangian
condition (we consider the trajectory of the piston).
When the pressure and velocity variations are negative and decreasing
(expansion), the slope of the characteristic lines is a decreasing function of time
(Figure 5.8a). These divergent straight lines thus form the shape of a fan, which
involves a broadening of the wave front as it propagates; Figure 5.8b shows the form
of the temporal velocity variation at the origin O, and then at points C and F.
Figure 5.9. (a) Application of the characteristics method and evolution of

the characteristic straight lines for a positive and growing variation of pressure
(or velocity); (b) evolution of the velocity t the points O, C and F
On the other hand, when the pressure and velocity variations are positive and
increasing (compression), the slope of the characteristic lines is an increasing
function of time (Figure 5.9a). They thus form a beam of straight lines which tighten
and finally intersect. This situation leads to a compression of the wave front as it
propagates, which eventually leads to a discontinuity. Indeed, we note in this last
case that it is impossible (equation [5.44]) to have more than one characteristic curve
of each family at a given point. The result of this is that a continuous solution cannot
exist everywhere in the influence domain of the boundary conditions which are
specified. A discontinuity thus appears (a shock wave) downstream of which the
calculation of the solution can only be achieved using the conditions which result
from the shock wave. Figure 5.9b shows the form of temporal variation at the origin
O, and then at points C and F.
The formation of a shock wave can be physically explained in the following

elementary manner: the increasing pressure can be decomposed into a succession of
elementary (acoustic) waves which propagate at the speed of sound. Each of these
elementary waves corresponds to an isentropic compression which increases the
temperature, and thence the speed of sound. Each elementary wave will thus travel
slightly faster than its predecessor, which it will finally catch. These waves are thus
concentrated at a point where they form a discontinuity. It is clear that when the
pressure decreases, an inverse process occurs: the elementary waves are spread out.
5.5.3. Plane steady supersonic flow
We have already studied this flow in section 5.3.2.3, and so we will limit
ourselves here to a qualitative discussion. Consider a uniform homentropic flow next
to a wall in a semi-infinite medium (Figure 5.10).
Figure 5.10. Plane supersonic flow around (a) a convex wall

and (b) a concave wall
Bernoulli’s first theorem is valid everywhere in the flow, as are the Saint-Venant
and Hugoniot relations (section 4.3.2.3). The characteristic curves form an angle D
with the streamlines, defined by sin D r 1 M .
If the wall is convex (Figure 5.10a), the streamlines spread and the density
G
decreases while the velocity V increases (supersonic expansion, section 4.3.2.3.4).
The velocity (direction and modulus) propagates from the wall along the
characteristic of a positive slope, and which is the only characteristic of consequence
here (straight line CA, CB,… … from A, B, etc.); the Mach number thus increases
and the angle D decreases in the downstream direction. Similar to the previous
example, we see divergent characteristics straight lines.
In the presence of a concave wall, an inverse situation occurs: the streamlines

tighten up, leading to a reduction in the velocity associated with a compression; the
Mach number decreases, and the characteristic lines intersect. Thus, a shock wave is
formed (Figure 5.10b).
5.5.4. Flow in a nozzle
A nozzle is a truncated conduit comprised of a convergent section followed by a

divergent section (Figure 5.11). When it separates two independent gaseous spaces,
it supports a flow between an upstream region at pressure pA and a downstream

region at pressure pE (pE < pA). We will assume that viscous friction effects at the
nozzle walls are small enough to be negligible up to the exit. The flow exiting from
the nozzle is thus in the form of a jet; the main viscous dissipation corresponds to
the energy loss due to the pressure difference; this dissipation occurs in the jet
downstream of the nozzle exit. Experience shows that we can consider that the
pressure in the exit plane is equal to pE as long as the jet is subsonic.
Assuming that the quantities associated with the gas are constant in a normal
section, the Saint-Venant relation (section 4.3.2.3.3) provides an expression for the
velocity as a pressure function using the generation conditions (initial conditions at
zero velocity in the upstream domain) and in particular the velocity VE in the exit
section SE. The isentropic transformation relation ( p p const ) determines the
y
density UE in the exit plane. From this we can deduce the mass flow q m U EVE S E
in the nozzle.
However, we have shown (section 4.3.2.3.4) that a stream tube resulting from a
given set of generation conditions has a maximum flow rate q m max U cVc S c
which occurs when the speed of sound c is attained in the smallest cross-section. We
are thus faced with the following alternative:
– either the flow rate qm evaluated at the exit plane is less than or equal to qm max
and we can calculate the continuous flow in the nozzle;
– or the flow rate qm evaluated at the exit plane is greater than qm max and the
problem thus posed does not have a solution.
In the first case, the Hugoniot relation [4.39] in its differential form (section
4.3.2.3.4) shows that the velocity increases in the convergent part of the nozzle up to
a value which is at most equal to the speed of sound c* at the throat, and which then
decreases such that its value at the exit plane VE is that previously predicted. The
flow is then everywhere subsonic (V < c). Figure 5.11 shows the pressure variations
(contrary to the velocity variations) in the nozzle for regimes 1, 2 and 3 which are
entirely subsonic. For pressures pE1 and pE2, the velocity at the throat Vmax is less
than the speed of sound, while for the pressure at the exit pE3 the throat velocity is
equal to the speed of sound c*.
upstream downstream:
p = pA p = pE
Vmax or c* SE VE
jet
Vc
p
pE
pA pE1
nozzle throat 1 pE2
pE3 subsonic
2 pE4
3 pE5 turbulent jet
p* 4
5 pE6
6
pE7
supersonic jet
shock waves 7 pE sup
x
Figure 5.11. Flow regimes in a nozzle
In the second case, the flow in the convergent part of the nozzle is subsonic, then
it becomes supersonic after passing through the throat where the velocity magnitude
is equal to the speed of sound c* and the critical conditions (p* and U*) are attained.
However the continuous supersonic solution, calculated using the Saint-Venant
relation in the exit plane using the generation conditions, is unique. It provides the
value pEsup for the exit pressure, which is not equal to the exit pressure imposed pE.
However, the supersonic flow in the nozzle must match the exit conditions. This
adaptation is achieved by means of a shock wave. So long as the shock wave
remains within the nozzle, it is plane (from pE3 to pE7 in Figure 5.11). For lower
pressures (pE sup < pE3 < pE7) we have a more or less complex system of shock
waves in the jet (under-expanded jet). For pE less that pE sup, the adaptation is
achieved by means of expansion waves (over-expanded jet).
The idea of characteristic curves and of propagation do not hold for the
differential equation of the 1D model nozzle. We note only the non-existence of a
continuous solution which verifies the boundary conditions at the exit. In fact, the
flow in the nozzle is governed by the models outlined in section 5.3.2.3:
– for the subsonic part of the flow, the system of partial differential equations is
elliptic and the flow is determined by the boundary conditions on all boundaries. Its
solution assumes Neumann conditions which are here known in the upstream region,
on the walls and in the exit section of the subsonic flow: depending on the regime,
this is either at the nozzle exit or at the throat;
– from the throat of nozzle in the sonic regime, and up to the shock, the subsonic
flow is governed by the 2D plane mode discussed in section 5.3.2.3, the fluid being
assumed to be isentropic. The supersonic zone of flow belongs to the influence
domain of the “upstream” conditions (section 5.4.5.3) which are here situated at the
sonic throat. An exit condition cannot influence the supersonic flow, as the
information cannot move upstream due to the characteristic curves which all have
slopes r ( M 2 1) 1 2 . In these conditions no continuous supersonic solution can
account for the conditions at the exit.
The specific characteristics of the plane 2D supersonic model also generally

correspond to the 1D model whose behavior is appropriate but without explaining
the difficulties: we observe that data given at two conditions, one at the throat and
the other at the exit, leads to an impossibility because of the existence of a singular
point for M = 1 in [4.38] and [4.39]. Such difficulties are often encountered in fluid
mechanics, where a global model can lead to contradictions (or to “paradoxes”) that
only a more refined model can explain.
The shock wave is a boundary between two spaces which cannot communicate
completely, the upstream space not being able to receive information regarding the
pressure at the exit. However, matter crosses the shock wave and the balance
equations for the extensive quantities must be satisfied through the shock.
In conclusion, a continuous solution of the 1D equations does not usually exist in

isentropic compressible fluid, for a nozzle whose throat velocity is sonic (critical
velocity). From a physical point of view, we could also consider that the shock wave
comprises an accumulation of pressure waves which travel from the downstream
and which stop when they can no longer do so.
The shock wave is a dissipative structure which leads to an increase in entropy

([LAN 89], [YIH 77]), viscosity playing an important role at the scale of the mean
free molecular path, for which a continuous viscous model is appropriate for the
shock wave.
Let us finish by highlighting a particularly useful application for nozzles

operating in the supersonic regime whose mass flow is fixed and depends only on
the upstream conditions. Such a nozzle, placed upstream of an installation, perfectly
regulates the flow if the upstream generation conditions are fixed, which is often the
case in laboratory situations: downstream perturbations can have no influence on the
mass flow of the device. The pressure loss of such a nozzle is relatively small (| 104
pascal).
5.5.5. Separated shock wave
A problem of the same kind is posed by a supersonic flow around an obstacle

which imposes boundary conditions which cannot travel upstream in the supersonic
flow. In particular, upstream of the obstacle there exists the stagnation point A,
where the velocity is zero, and therefore a subsonic zone of flow in the region near
the surface.
shock wave
supersonic zone subsonic zone
Figure 5.12. Detached shock wave in front of a body in a supersonic flow
Similar to the case of the nozzle, the adaptation of the supersonic flow to
downstream conditions occurs by means of the shock wave.
5.5.6. Other discontinuity categories
Combustion phenomena in gaseous flows obey the fluid dynamics equations

which we have already discussed in Chapter 4. We must however introduce the
properties of chemical reactions using the methods of chemical thermodynamics
which must be applied to the moving matter. Without containing new physical
phenomena, the formalism obtained combines the difficulties of the two domains.
We cannot address these questions in detail in this book (see [BOR 00], [KIR 67],
[OPP 06], [WIL 65], [WIL 85]). Schematically, the possibility of a chemical
reaction amounts to the introduction of a heat source associated with a local increase
in temperature: we thus notice that a shock wave can trigger a chemical reaction
which can augment its effects considerably, transforming the shock wave into a
detonation wave.
Other domains of fluid mechanics also involve hyperbolic equations. Such is the
case for flows including a free surface (see section 6.2.6) in which we observe the
propagation of surface waves (swell in the sea) and flows of stratified fluids (vertical
distribution of density) which present similar properties. We encounter analogous
phenomena to those discussed for a gas, and in particular the possibility of

discontinuities called hydraulic jumps [YIH 77] or tidal bores in estuaries which
propagates against flowing water.
5.5.7. Balance equations across a discontinuity
A velocity discontinuity undergone by a material body in movement assumes

infinite external forces and (finite) inertial effects associated with non-Galilean
reference frames are therefore neglected. The equations for the shocks and collisions
can be written in any reference frame, both in particle or solid body mechanics and
in the dynamics of continuous media.
We will consider a discontinuous surface which is crossed by matter in

movement and we will designate by the indices 1 and 2 the upstream and
downstream quantities of the discontinuity. We apply the balance equations in
global form in a thin volume D comprised of two parallel surfaces at the
discontinuity S, of area ds and with normals oriented towards the exterior (Figure
5.13). The indices n and t designate the velocity components normal and tangent to
the surface of the discontinuity.
p1,U1,T1 1 2 p2,U2,T2
G G
n1
V2
ds G
G V1
D n2
S
Figure 5.13. Balance on a discontinuity surface (shock)
We now write the balance equations for the extensive quantities in global form
(section 4.5) in the domain D for the following quantities:
– mass (section 4.5.2):
U1V1n U 2V 2 n [5.53]
– momentum (section 4.5.4):
p1 U1V12n p 2 U 2V 22n
G G [5.54]
U1V1t V1n U 2V 2t V 2 n
– total energy (enthalpic form) (section 4.5.6):
§ V2 · § V2 ·
U1V1n ¨ h1 1 ¸ U 2V 2 n ¨ h 2 2 ¸ [5.55]
¨
© 2 ¸¹ ¨
© 2 ¸¹
– chemical species (i =1,2 …):
U i1Vi1n U i 2V i 2 n [5.56]
We can immediately deduce the continuity of the tangential velocity components

across the shock:
G G
V1t V 2t [5.57]
NOTE – In the presence of a chemical reaction on the surface S (detonation or

deflagration wave), balances [5.55] and [5.56] take the following form:
V12 V22
h1 Qr h2 U i1Vi1n Qmi U i 2Vi 2n [5.58]
2 2
where, Qmi and Qr denote the mass of species i and thermal surface power released
by the chemical reaction for the mass flux U1V1n crossing the shock S, the
enthalpies hi being taken as equal to Cpi Ti for perfect gases (“sensible enthalpy”).
5.6. Some comments on methods of numerical solution
5.6.1. Characteristic curves and numerical discretization schemes
The numerical resolution of a system of first order differential equations with

initial conditions (Cauchy) given at a point can be achieved from place to place: by
discretizing the first derivatives, we calculate the value of the unknown vector
function at a point using the values at the previous point. However, we often have
conditions on either extremity of an interval instead of Cauchy conditions, in
particular when the system is associated with a flow between solid boundaries. We
therefore often use a shooting method: the missing initial values are determined by
successive approximations so as to obtain suitable values at the other extremity of
the interval.
Such an iterative procedure presents the advantage of not involving the inversion
of a large matrix.
A numerical solution consists of replacing equations of a differential kind with

finite difference algebraic equations obtained by means of a discretization of the
derivatives which can be performed in many different ways which we will quickly
evoke further. By way of a simple example, consider the transport equation in which
the velocity u is constant:
wf wf
u 0 [5.59]
wt wx
The axes Ox and Ot are discretized with step-sizes 'x and 't such that 'x = u.'t.
We will here approximate the derivatives by formulae using the values at two points.
We calculate the partial temporal derivative using the formula:
wf 1
xn , t n > f xn , tn f xn , tn 1 @ [5.60]
wt 't
wf
We approximate the spatial derivative at the point xn by its value at the
wx
wf
instant tn-1, x n , t n 1 such that we have a simple, explicit scheme for the
wx
wf
discussion. Take as an approximation of x n , t n 1 one of the following
wx
schemes, which are apparently locally equivalent:
wf 1
– upwind scheme: x n , t n 1 > f x n , t n 1 f x n 1 , t n 1 @ ;
wx 'x
wf 1
– downwind scheme: xn , tn 1 > f xn1, tn 1 f xn , tn1 @ ;
wx 'x
wf 1
– centered scheme: xn , tn 1 > f xn1, tn 1 f xn1, tn 1 @ .
wx 2'x
Substituting these expressions into [5.59] we can calculate the value of the
function f x n , t n as a function of its values at the instant tn-1 at the points next to
the axis Ox. We obtain for the different schemes:
– upwind scheme: f x n , t n f x n 1 , t n 1 ;
– downwind scheme: f x n , t n 2 f x n , t n 1 f x n 1 , t n 1 ;
– centered scheme:
1 1
f x n , t n f x n , t n 1 f x n 1 , t n 1 f x n 1 , t n 1 .
2 2
Figure 5.14 schematically shows the transmission of numerical information

between the points of the discretized network for the three numerical schemes.
t t t
tn tn tn
tn-1 tn-1 tn-1
x x x
xn-1 xn xn+1 xn-1 xn xn+1 xn-1 xn xn+1
(a): upwind scheme (b): downwind scheme (c): centered scheme
Figure 5.14. Transmission of information in the numerical resolution

following different discretization schemes
Now, equation [5.59] is hyperbolic and its characteristics are the straight lines
x ut const, which are trajectories of the uniform velocity field u. To simplify
matters, the discretization ('x = u.'t) was chosen such that these straight
characteristics pass through the points of the computation. Let us examine a
particular case of the problem, defined by the boundary conditions which
corresponds to transport at velocity u of a unit step function from the origin in a field
with zero initial values:
t 0, x 0: f (x, 0) 0
[5.61]
t p 0, x 0: f (0, t) 1
Table 5.1 indicates, on each line, the numerical values obtained by means of the
three numerical schemes for the six points on the axis Ox at the six first instants (0,
't, 2't, 3't, etc. from the bottom of the table).
1 1 1 1 1 1 1 0 0 0 0 0 1.0 1.312 1.875 1.156 0.312 0.031

1 1 1 1 1 0 1 0 0 0 0 0 1.0 1.500 1.375 0.500 0.062 0
1 1 1 1 0 0 1 0 0 0 0 0 1.0 1.375 0.750 0.125 0 0
1 1 1 0 0 0 1 0 0 0 0 0 1.0 1.000 0.250 0 0 0
1 1 0 0 0 0 1 0 0 0 0 0 1.0 0.500 0 0 0 0
1 0 0 0 0 0 1 0 0 0 0 0 1.0 0 0 0 0 0
(a) upwind scheme (b) downwind scheme (c) centered scheme
Table 5.1. Calculated evolution of the function f x, t for three numerical schemes
The downwind scheme cannot transmit the numerical values in the direction of
the flow. It is contrary to the physical nature of the problem studied. Regardless of
the values specified on the straight line x = 0, we will have a discontinuity between
these and the values of the function f on the neighboring points.
The centered scheme transmits boundary condition information at x 0 in a

partial and deformed manner: if the upstream and downstream values are very
different, this scheme will favor the larger, and this may lead to numerical
oscillations.
These examples, albeit rather rudimentary, show clearly that it is not simply by
increasing the order of the numerical schemes that we can hope to improve the
results. The material balance can only be assured by ensuring the transfer of
information along the trajectories u – ct = const.
Regardless of the nature of a flow, the characteristic curves constitute privileged

lines of information transmission and any numerical scheme which does not
completely respect this constraint will lead to divergence and instability of the
computation. As the trajectories and the characteristic curves of wave propagation
are generally unknown, the respect of this condition introduces notable
complications.
5.6.2. A complex example
We will now consider the more complex practical case of an unsteady 1D

inviscid compressible flow governed by the equations of section 5.3.2.1. Let us
indicate the experimental conditions of the problem modeled. The supply pipes of a
thermal engine are dimensioned such that the quantity of air supplied is maximum at
a suitable operating condition. The amplitudes of the velocity pulsations may be
large on account of acoustic resonances which are contrived in order to ensure the
maximum air supply. The air entropy is not uniform as it issues from zones which
may be more or less heated, but thermodynamic transformations are isentropic.
We have instantaneous measurements of the pressure px, t and the velocities

u x, t at the extremities x 0 and x = A of the pipe system studied. We wish to
calculate the distribution of physical quantities in the pipe (pressure, velocity,
entropy, etc.) from the said measurements. The calculation domain was discretized
and different high precision calculation methods were employed. As a general rule,
with the exception of method of characteristics, these all led to the production of a
numerical shock for the entropy values.
The three families of characteristics defined in section 5.3.2.1 for this system are:
– characteristics C1: G x uG t 0 with: G p c 2 G U 0;
– characteristics C2: G x c u G t 0 with: G p U cG u 0;
– characteristics C3: G x c u G t 0 with: G p U cG u 0.
These are shown in Figure 5.15 with their domain of influence: the families C1
and C2 transport their associated characteristic variables from the axis x = 0, whereas
the family C3 leads to the propagation of its characteristic variable G p U cG u 0
at the speed – c + u from the straight line x = A. The characteristic variables are, by
their nature, “input” variables in the region where their associated characteristic
curves enter the domain. Their associated “exit” values at the other extremity of an
interval cannot be given conditions without being in contradiction with the
mathematical structure of the system of equations.
t
C3 C2 B
C1
O x
A
characteristic curve C1: G x uG t 0
characteristic curve C2: G x c u G t 0
characteristic curve C3: G x c u G t 0
Figure 5.15. Characteristic curves of an unsteady, subsonic, 1D flow

The boundary conditions concerning p and u at the extremities of the interval

[0,A] are thus the sum of the “entry” data and the “exit” values: the values of p and u
at the point A are initial conditions to be given on the characteristics C1 and C2, and
the result of propagation on the characteristic C3. The situation is similar at point B,
but with a single datum on C3. Laying down three conditions on p and u at x = 0
and x = A amounts to an implicit specification of the “entry” conditions on the three
families of characteristics C1, C2 and C3. In order to correctly lay down the
problem, it is thus necessary to specify the suitable information (as a function of
time), i.e., preferentially, two boundary conditions on the left x 0 and one
boundary condition on the right x A . These boundary conditions are
combinations of the “to be given” characteristic variables (entry variables) and
unknowns (exit variables).
Numerical information concerning any physical quantity at a point is issued from

three different progresses, each bringing a partial contribution to the value of this
quantity. Now, only an upwind discretization scheme is suitable for information to
progress in one direction (section 5.6.1). It follows that any discretization of the
physical quantity is necessarily inconsistent with at least one of the three
progressions of the information. Using a numerical scheme, be it of high precision
or otherwise, which does not take into account the preceding physical (or
mathematical) reality can only lead to difficulties in the calculation and to the
appearance of numerical oscillations and discontinuities which are incompatible
with the desired solution. The solution can only be obtained by using a
characteristics method with a suitable discretization scheme; we note that the
solution of this discretized system cannot be obtained by a computation from place
to place [SAN 97].
5.6.3. Boundary conditions of flow problems
We saw earlier how the values of the variables of a problem move along the
characteristic curves. The preceding example shows the difficulties which can be
encountered when we try to correctly write the boundary conditions of a flow
problem which is often posed in an open domain. The physical quantities of the fluid
entering the domain must be given. Even if we assume that there is no propagation
in directions opposed to the trajectories (incompressible or supersonic flow), we still
have to deal with three principle difficulties:
in the region where the fluid enters the domain, the velocity and pressure fields
must satisfy the dynamic equations which are used: this condition, which is satisfied
by a uniform flow, is often difficult to meet for other kinds of flow, even if they are
steady;
furthermore, no characteristic curves must go out through the initial curves or

surfaces unless suitable compatibility conditions are verified, both for compressible
flows and incompressible flows; for instance, in certain near-wall regions
(separation around the downstream part of an obstacle or in a divergent conduit
(section 6.5.3.7)) the flow of a fluid can be in the opposite direction to the main
flow: the corresponding entry zones, situated downstream, depend therefore on the
structure of the solution of the problem;
finally, in the case of real, measured, fluid values used as domain boundary
conditions, measurement errors may lead to computational difficulties in so far as
they may correspond to entry (or exit) conditions which are incompatible with the
problem which is posed;
The considerations developed in this chapter concern flows of inviscid fluids

which are represented by either elliptic or hyperbolic equations. The presence of
viscous structures along certain trajectories in high Reynolds number flows (section
6.5.3) often leads to parabolic equations along these trajectories, so creating an
enlargement of the influence domain of the initial conditions along these (see section
5.4.5.4).
In summary, with the exception of some simple situations, the specification of

boundary conditions for a flow problem is often a delicate operation, and it is
extremely difficult accomplish in a rigorous fashion.
Chapter 6
General Properties of Flows
In this chapter we will study some general physical properties of flows which
result from the structure of the balance equations. Transport and propagation
phenomena are always present, even in systems with uniform initial conditions. The
dynamics of fluids and transfers imply coupled phenomena with multiple
interactions. In the simplest cases, non-dimensional parameters can be identified
which characterize the ratio of orders of magnitude between the terms corresponding
to two phenomena, and this ratio is generally small or large with respect to 1. The
dynamics of fluids and transfer is thus the domain of perturbation phenomena which
lead to singular structures.
After examining the vortex properties, we will discuss flow properties associated
with uniform initial conditions which lead to a relative simplification due to the
existence of a potential. The third part of this chapter will deal with the study of
orders of magnitude and perturbation problems. Quasi-1D approximations in pipes
and boundary layers are then discussed. The last part will be dedicated to a short
presentation of unsteady phenomena in flows.
6.1. Dynamics of vorticity
6.1.1. Kinematic properties of the rotation vector
6.1.1.1. Definitions
G 1 G G
These are associated with definition [3.38] of the rotation vector Z ro t V
2
G
given in section 3.3.4. Let us recall that vorticity vector 2Z (section 4.3.3) has
G
obviously the same properties as Z . At a given instant, we call:
– surface of rotation or vortex surface, a surface which at each of its points is

G
tangent to the vector Z ;
G
– line of rotation or vortex line, a line which is tangent to the vector Z at each of
its points;
– tube of rotation or vortex tube, a surface generated by the lines of rotation
relying on a closed contour.
The divergence of the rotation vector is clearly zero, and the flux of the vortex
vector across any closed surface is also zero.
G
6.1.1.2. Circulation of the velocity vector V
Let *AB be the circulation of the velocity vector along an arc AB:
G G
* AB ³AB V .dl
In the case where the curve C is closed, the circulation *C is equal to the flux of
the vector rot V across a surface S relying on the contour C (Stokes’ theorem):
G G G G
*C ³C V .dl ³S 2 Z . n ds [6.1]
The circulations īC1 and īC 2 along the two closed curves C1 and C2 situated on
the same tube of rotation which they move around the same number of times are
equal.
In effect, let D be the inner domain of a tube of rotation which is bounded by two
surfaces S1 and S2 whose contours on the rotating tube are, respectively, the closed
curves C1 and C2. The flux of the vortex across the lateral surface of the tube and
across the surface 6which bounds the fluid domain D is zero; the result of this is
G
that by orientating, continuously along the tube, the normals n across the surfaces S1
G
and S2, we have equality of the fluxes of the vector Z across the two surfaces, and
consequently across all sections of the vortex tube. This results in the equality of the
circulations īC1 and īC 2 .
The intensity of a vortex tube is defined by the circulation value of the velocity
vector along a closed curve encircling the tube once counter-clockwise.
The circulation *C can be calculated by following the matter. Using result [3.36]
of section 3.3.3.4 concerning the material derivative of the integral of the flux of the
General Properties of Flows 259
G G
conservative vector field B ( divB 0 ) across the material surface S of fluid in
movement, we can express the material derivative of the circulation *C in the form:
G
d *C
d t
2
d
dt
G G
³S Z . n ds 2 ³S
§w Z
¨
¨wt
G
ro t ZG VG ·¸¸ . nG ds [6.2]
© ¹
This expression will be useful for the demonstration of Lagrange’s theorem.
6.1.1.3. The Biot and Savart formula

Any vector field can be decomposed into a field with zero divergence and an
irrotational field. The latter is a gradient field derived from a potential. The zero-
divergence field is a rotational field, defined to a near gradient. Knowledge of the
G G
rotation vector Z allows the computation of the rotational part V r of the velocity
G G G
field V . Letting V r rot A and taking into account divA = 0 1 gives:
G
2Z rot V r G
rot rot A
G G
grad divA ' A
G
' A
G
The components of the vector A in Cartesian coordinates satisfy the Poisson
equation 'Ai 2Z i whose solution is:
1 dv'
Zi x'j
avec : R 2 ¦ x
3
x 'j
2
R x , x
Ai x j ³D ' j
2S j j i 1
Taking the curl of this expression gives:
G G
1 G r
Vr ³D Z dv
2S r3
which is the formula of Biot and Savart for a magnetic field, where the current
density is equivalent to the vortex, and the velocity corresponds to the magnetic
field. Note that this result, which is purely kinematic, is not related to any
assumption regarding the nature of the fluid or the flow.
G G
1 The vector A is notGdefined in a unique way; we can add any gradient G at A without
changing the values of V r . Thus, we choose this gradient vector so that divA 0 .
6.1.1.4. The velocity field induced by a vortex

At a given instant, consider the tube of a cylindrical vortex, of radius r0, and
G
suppose that Z Z is uniform over a cross-section S0 of the tube and zero outside
the tube.
Z
r0 V (r )
r
V
r0 r
(a) ( b)
Figure 6.1. Velocity around an uniform vortex tube
Consider the circular surface S in the plane of a cross-section of the tube,

centered on the axis of the tube, and of radius r (Figure 6.1a). The circulation
*C 2S Vr of the velocity along the circle C of radius r is equal to the vector flux
G
rot V across the surface of the circle of radius r. This flux can be expressed in two
different ways, depending on the relative values of r and r0. We can write:
Z r02
r d r0 V Z r; r t r0 V
r
The velocity induced outside the vortex tube decays as 1/r, whereas it grows
linearly with r on the inside of the vortex tube (Figure 6.1b).
6.1.1.5. Material derivative of the rotation vector

Let us introduce the material derivative of the rotation vector:
G G § dZ
dZ wZ G G wZ i wZ i ·¸
.§¨ grad Z ·¸ .V ¨ i uj
dt wt © ¹ ¨ dt wt wx j ¸¹
©
G G
We can write the following identities for the arbitrary vectors A and B [HAR
98]:

G G
rot A B ©
G
¹
G
© ¹
G G
§¨ grad A ·¸ . BG A . divBG §¨ grad BG ·¸ . A BG . divA
G
or, in this case ( divZ 0 ):

G G
rot Z V §¨ grad ZG ·¸ .VG ZK . divVG §¨ grad VG ·¸ . ZG
© ¹ © ¹

G G §

div Z V ¨ grad V ¸ . Z
©
G· G
¹
G
From expression [3.37] for grad V as a function of the tensor H of the strain-
rates and of the anti-symmetric tensor : , we can derive the relation:
G G G
grad V . Z H . Z ; hence:

G G
rot Z V §¨ grad ZG ·¸ .VG ZK . divVG H . ZG
© ¹

G G
G
div Z V H . Z
Finally we obtain the expression:

G K G
wZ
wt
G
rot Z V
G
dZ
dt
K G G
Z . divV H . Z
wZ
wt
G G
G
div Z V H . Z [6.3]
G
G Z
We can associate the volume quantity Z with the mass quantity ; taking
U
account of the mass conservation equation (see section 4.2.1.2.1 and formula [4.9])
leads to (Helmholtz):
G G
wZ
G G
div Z V U
d §Z ·
¨ ¸ [6.4]
wt dt ¨© U ¸¹
6.1.2. Equation and properties of the rotation vector
6.1.2.1. The vorticity equation in the form of a balance equation

G
Equation [4.41] (section 4.3.3), which is satisfied by the rotation vector Z for
any given fluid, can be written by taking account of expressions [6.3] and [6.4] in
one of the two forms:
G
wZ
wt
G
G½
div Z V ° § 1 ·
° G 1
G ¾ H .Z grad U grad p rot ¨¨ div W ¸¸ [6.5]
d §Z · ° 2U 2
© 2U ¹
U ¨¨ ¸¸
dt © U ¹ °¿
Equation [6.5] has the form of a balance equation of a volume vector quantity
(equation [4.3], section 4.1.1.2). It can be interpreted as a balance of the rotation
G
vector Z , considered as a volume density of an extensive quantity (with which we
G
can associate the mass quantity Z U ).
For a divariant compressible, by taking rotational of grad h (relation [4.17]), we

have:
1
grad U grad p grad T grad s
U2
Vorticity equation [6.5] is thus a transport equation comprising:

– two volume source terms associated respectively:
G
- with the usual properties of the kinetic effects of the rotation ( H . Z );
- with the movement associated with mechanical or thermodynamic imbalance;
– a viscous diffusion term.
If the mechanical equilibrium condition ( grad U grad p 0 ) of a fluid in a

force field is not satisfied, a rotational movement will result. For example, a
horizontal pressure gradient in a fluid with a vertical density gradient, which is
initially at rest, will create a horizontal acceleration inversely proportional to U and
therefore a horizontal velocity gradient.
As for the viscous stresses, their role is essential in diffusing the rotation, as we
will see in an example (section 6.1.2.4.1). We should furthermore note that the
creation of a viscous flow (Poiseuille flow, boundary layer, etc.) from a non-viscous
flow is accompanied by the creation of vorticity as a result of the adherence
condition at the wall, which creates a shear flow in the vicinity of the wall (section
3.4.2.5), which is necessarily rotational. For the sake of simplicity, we will not give
detailed expressions of the viscous stresses here.
G
For the case of an incompressible Newtonian fluid ( divV 0 ) with constant
viscosity and specific mass, we have div W
P rot rotV (section 3.4.3.3) and we
obtain the following form of equation:
JG JG JJJ
s $ JJJG JG JJG JJJG d$ JG JGG
rot ( $ V ) v+$ or: %.$ v+$ [6.6]
st dt
For an inviscid fluid which is homentropic or with constant specific mass, the
preceding equations are simplified:
G

°
° G wt
wZ

G
rot Z V 0
® wZ G [6.7]
° div
G G
Z V U
d §Z ·
¨ ¸
G
H .Z
°¯ w t dt ¨© U ¸¹
6.1.2.2. Interactions between vorticity and strain rates

G
We will characterize the effects of the volume source term H . Z ( Z j .H ij ) which
translates an interaction between vorticity and the strain-rate tensor. In order to
simplify matters let us consider equation [6.7] for a constant specific mass:
G
d$ G d$ i
% .$ ... or: .%ij $j ...
dt dt
Let us first of all examine the effect of the first component Z1 of the rotation
vector on itself. The corresponding equation can be written:
d$1
$1.%11 ...
dt
This demonstrates the creation of Z1 if the source term Z1 .H 11 is positive, for

example if Z1 and H 11 are positive; thus, H 11 is the rate of increase of the
component u1 of the velocity in the x1 direction (section 3.3.4), and it leads to a
stretching of the matter along this axis: a stretching of the matter along a given
direction increases the corresponding component of the rotation vector (Figure
6.2a). This result is in fact analogous to the intensity conservation of a vortex tube,
the stretching rate being inversely proportional to the variation velocity of the tube
cross-section.
Now consider the source term Z1H 21 in the equation for the component Z2:
d$ 2
$1.% 21 ...
dt
wu1
The strain-rate H 21 amounts to a shearing of the type (section 3.3.4). We
wx 2
therefore see here the creation of the component Z 2 , along the axis Ox2, by the
component Z1 (Figure 6.2b), in other words a tilting of the rotation vector (a

gyroscopic effect).
x2 Z
Z1 u1
Z1 Z2
x1
Z1
u1 x1 u1
(a) (b)
Figure 6.2. Deformation of a vortex: (a) stretching in a lengthening velocity;

(b) tilting in a shear velocity
These properties are true for the three spatial directions; their effect is the
creation and maintenance of the 3D character of rotational flows.
If the fluid is compressible, the second term of material derivative [6.3] contains
G G G
the term Z .divV which translates a reduction of Z proportional to the
G
expansion divV , leaving the angular velocity constant in a material volume. This
term is contained in the first term of vorticity equation [6.5] written with the mass
G
quantity Z U .
6.1.2.3. The 2D plane flow

The vorticity source term Z j .H ij presents a marked 3D character but disappears
in a 2D plane flow, because the rotation vector has only a component Z
perpendicular to the velocity plane, and so we have:
§ H 11 H 12 0 ·§ 0 ·
¨ ¸¨ ¸
Z j .H ij ¨ H 21 H 22 0 ¸¨ 0 ¸. 0
¨ 0
© 0 0 ¸¹¨© Z ¸¹
In the 2D case, the rotation vector (or the vortex) satisfies the scalar equation:
d Z
Q 'Z [6.8]
d t
In a 2D plane flow, the vorticity equation is a convection-diffusion equation

which ensures the conservation of the rotation in all space (see definitions of
diffusion terms in Chapter 2).
For an inviscid fluid, scalar equation [6.8] for the vorticity is an equation
describing the transport of vorticity by the matter:
dZ
0 [6.9]
dt
6.1.2.4. Diffusion of the vorticity in a viscous fluid

By way of an example, consider equation [6.8] of an incompressible Newtonian
fluid in plane two-dimensional flow. It has the form of a heat convection equation
(section 4.3.4.1.6) or of the equation for the diffusion of chemical species (in weak
concentration) with source terms and diffusion terms: vorticity is diffused by viscous
action.
Let us examine the case of a vortex system of revolution about the axis Oz and
whose velocity field, parallel to the plane Oxy and of zero radial component, has a
tangential component equal to VT (r, t ). The vortex vector field Z r, t , parallel to
Oz, is a function of r and t. Equation [6.8] can be written:
wZ
Q 'Z [6.10]
wt
or, in plane polar coordinates:
wZ Q w § w Z·
¨r ¸ [6.11]
wt r w r ¨© w r ¸¹
The reader can easily verify that equation [6.11] has a class of solutions2 in the
form t n f K where we have introduced the new variable K r 2 4 Q t . The
function f K satisfies a differential equation which depends on the parameter n.
The circulation *r, t of the velocity vector along the circle centered on the
origin and of radius r is equal to the flux of the vector 2Z across this circle:
2 These solutions are “self-similar” like solution [5.52] (see footnote 5 in section 5.4.5.4).
r § u ·¸
2
r 2 4Q t K
*r , t 2t n ³0 f ¨ 2S rdr 8S Q t n 1 ³0 f v dv 8S Q t n 1 ³0 f v dv
¨ 4Q t ¸
© ¹
Suppose that after the initial instant t = 0, there is no source of vorticity

anywhere in space; the circulation on the circle whose radius tends to infinity
remains constant, and this leads to the choice n 1 . Substituting this expression
for Z into [6.11] we obtain the differential equation satisfied by the function f K :
K f 'K f '' 0
A first integration gives:
const
f f '

Integrating a second time, taking the constant of integration to be zero such that
the circulation remains constant for infinite K, the vorticity tending therefore to zero.
This immediately gives the desired solution:
A A § r2 ·
Z r , t exp K exp¨ A const
t t ¨ 4Q t
© ¹
The circulation *r, t of the velocity vector along a circle of radius r can be

expressed as:
ª § r 2 ·º
*r , t 8 S A Q «1 exp ¨ ¸»
« ¨ 4 Q t ¸»
¬ © ¹¼
Its value is constant and equal to 8 S A Q for sufficiently large r. The velocity
VT(r,t) can be calculated from the circulation:
*( r , t ) 4Q A ª § r 2 ·º
VT (r , t ) «1 exp ¨ ¸»
2Sr r «¬ ¨ 4 Q t ¸»
© ¹¼
This solution represents the diffusive spreading of a Dirac impulse of vorticity

placed at the origin (Figure 6.3). The velocity distribution decays as 1/r outside the
viscous zone (section 6.1.1.4) which contains the vorticity. The velocity gradient
diminishes under the effect of viscosity which has no further real effect for a radius
a little greater than that at which the maximum is located.
Z V Z
t t
G
V
r r r
(a) (b) (c)
Figure 6.3. (a) Vorticity zone, (b) diffusion of the velocity and (c) of the vorticity
For a fixed radius r, when t increases, the circulation and the velocity decay so as
to tend to zero as time tends to infinity: the rotation initially concentrated on the axis
diffuses over time across the entire fluid under the action of viscosity. Defining the
radius Rv t of the viscous core using the condition that this contains 99% of the
circulation of the velocity vector (exp(Ș) = 0.01 or K = 4.605), we have:
Rv 4.29 t .
6.1.2.5. Lagrange’s theorem

Consider an inviscid fluid whose entropy is uniform if it is compressible. Taking
account of vorticity equation [6.8], material derivative [6.2] of the circulation of the
velocity on a closed curve C enclosing the surface S can be written:
G
d *C
2³S
§w Z
¨
¨wt
G
ro t ZG VG ·¸¸ . nG ds 0
d t © ¹
Lagrange’s theorem: the circulation of the velocity vector on a material curve,

or the flux of a vorticity vector across a material surface, is conserved during
movement. In particular, if the flux of the vorticity vector across a material surface S
is zero at an instant t, it will remain zero thereafter.
As we are dealing with a material derivative, the surface S is constituted of fluid

particles. It should not contain any singularity leading to the creation of vorticity (for
example, a vortex whose intensity varies with time). We can deduce the following
very important consequences:
– if, at a given instant, the flux M S is zero in a domain D of a flow regardless of

the surface S, the flow is irrotational in the domain D (it suffices to take three
G
elementary orthogonal surfaces to verify that the vector Z is necessarily zero).
From Lagrange’s theorem we thus see that the flow remains irrotational afterwards
in the material domain D. This situation is encountered when a flow issues from a
fluid region at rest or of uniform velocity;
– a vortex surface (or rotation surface) is a surface to which the vorticity vector
is tangent at an instant t; it moves whilst remaining a vortex surface (the flux of the
vortex remains zero on all elementary surface of rotation). In particular, a vortex
tube remains a vortex tube during any displacement of the matter of which it is
constituted. Considering the circulation of the velocity along a curve situated on the
tube and which encircles it once, we see that the intensity of a vortex tube remains
constant during its displacement: the vortex tube transports its circulation; this can
be easily seen in a rotational smoke ring (a closed rotation tube) in which smoke
makes the motion of the matter and its rotation visible;
– a vortex line (or rotation line) at instant t can be considered as an intersection
of two vortex surfaces: it therefore remains a vortex line. This results in vortex lines
being displaced with the matter.
NOTES –
1) The notion of circulation on a closed material curve C is essential: in effect, it
deforms during its displacement with the matter. It can eventually be divided into
two curves C1 and C2 when passing an obstacle (Figure 6.4), but it cannot be
transformed into a third curve C3 (Figure 6.4). The sum of the circulations
īC1 and īC 2 over the curves C1 and C2 is equal to the circulation *C over the
curve C, whereas the circulation *C 3 over the curve C3 can take on any other value.
We will use an elementary example (section 6.2.5.2.2) for which the circulations
īC, īC1 and īC 2 are zero, whereas *C 3 is non-zero. This property is related to the
structure of the surface S interior to C3 which is not simply connected (to put it
simply, it contains a “hole”) and the vector field is not continuously differentiable on
the interior of C3 .
2) Lagrange’s theorem does not express the transport of the vorticity vector by
the matter; it only expresses a more global property. For the reasons given above, we
G
say that the vector field Z which satisfies equation [6.7] is “frozen in the moving
medium”.
C3
C C2
C1
Figure 6.4. Evolution of closed material curves in the flow around an obstacle
3) The reader will note that once again we recover here a property of flow-
information transfer over characteristic curves associated with convection.
6.2. Potential flows
6.2.1. Introduction
Lagrange’s theorem expresses the property of transport of circulation of the

velocity over any curve C in the flow of an inviscid, homentropic fluid if the fluid is
compressible. When the circulation is zero, we can perform a partial integration of
the equations of fluid mechanics over the family of characteristic curves constituted
by the trajectories: the flow is therefore irrotational and the velocity field derives
from a potential:
ui
wM
wx i
or : VG gradM [6.12]
6.2.2. Bernoulli’s second theorem
Dynamic equation [4.19] can be written in the form:

G
wV §V 2 ·
grad ¨ U h¸ 0
wt ¨ 2 ¸
© ¹
with: h the specific enthalpy (for a perfect gas: h C p T ) and U the potential of the
gravitational forces ( U gz , height z being taken on a vertically ascending axis).
The existence of a velocity potential M allows the immediate integration of the

above equation with respect to the space variables, giving Bernoulli’s second
theorem:
s! V 2
gz h const [6.13]
st 2
We note that equation [6.13] is valid everywhere in the domain of study and for
an unsteady flow, contrary to Bernoulli’s first theorem, which can only be applied in
a steady flow over a streamline. For an incompressible fluid, it can be written:
s! V 2 ¬
gz p const [6.14]
st
2 ®
6.2.3. Flow of compressible inviscid fluid
The partial differential equation satisfied by the velocity potential M is a quasi-

linear second order equation and is derived from the Euler equations. As integration
has already been performed on the trajectories, the equation has only two families of
real or imaginary characteristic curves or surfaces which we have already seen
(section 5.4.2).
The mass conservation equation in the form [4.7] can be written:

1 dU wu i
0 [6.15]
U dt wx i
For a divariant fluid in homentropic flow, we have dp c 2 dU and dp U dh .

Replacing dU as a function of dh in equation [6.15] gives:
1 dh wu i § 1 dh G ·
0 ¨¨ divV 0 ¸¸ [6.16]
2
c dt wx i © c 2 dt ¹
The potential equation can therefore be obtained by replacing the enthalpy h in

equation [6.16] by its expression as obtained from equation [6.13]:
§ 2 ·
1 d ¨ wM grad M
'M ¨ gz 0 [6.17]
c 2 dt ¨ wt 2
© ¹
For the flow of a compressible fluid, we can in general neglect the gravitational
term and potential equation [6.17] becomes:
§ 2 ·
1 d ¨ wM grad M
'M ¨ 0 [6.18]
c 2 dt ¨ wt 2
© ¹
or, by developing:
1 §¨ w 2M w 2M wM · 1 wM wM w 2M w 2M
2 0 [6.19]
c 2 ¨© wt 2 wtwx j wx j ¹ c 2 wx i wx j wx i wx j wx i wxi
where, by introducing the usual notation for the velocity components u , v, w along
the axes x, y, z :
§ 2· § 2· § 2·
¨1 u ¸M ¨1 v ¸M ¨1 w ¸M 2 uv M 2 vw M
¨ xx ¨ yy ¨ zz xy yz
© c 2 ¸¹ © c 2 ¸¹ © c 2 ¸¹ c2 c2 [6.20]
wu 1 u v w
2 2 M zx 2 Mtt 2 2 Mtx 2 2 Mty 2 2 Mtz 0
c c c c c
The potential equation for the flow of an incompressible fluid which can be
obtained by letting the velocity sound tend to infinity in [6.17] reduces to Laplace’s
equation:
w 2M
'M 0 [6.21]
wxi wx i
6.2.4. Nature of equations in inviscid flows
Equation [6.19] or [6.20] is of the type studied in section 5.4. Writing explicitly
the velocity components ui wM wxi and the Mach number M V c , equation
[6.19] can be written:
1 §¨ w 2M w 2M ·¸ u i u j w 2M
2u j 'M 0
c 2 ¨© wt 2 wtwx j ¸¹ c 2 wx i wx j
The characteristic directions are given by equation [5.37] which corresponds to

the preceding equation. Letting D i w[1 wxi and D t w[ 1 wt (section 5.3.4), we
obtain:

1
D 2
t
2D j D t u j
ui u j
D j D i D iD i 0
c2 c2
This characteristic equation can be written in the reduced form:

1
2
D t D j u j 2 D iD i 0
c
This form is hyperbolic3 and the potential equation of a compressible fluid is in

general of hyperbolic character.
In the situation involving acoustic perturbations in a medium at rest, which is

obtained when the Mach number approaches zero in equation [6.19] or [6.20], the
wave equation is obtained:
1 w 2M
'M 0 [6.22]
c 2 wt 2
Considering now only steady solutions, the potential equation can be written:
w 2M 1 wM wM w 2M
0 [6.23]
wx i wxi c 2 wx i wx j wx i wx j
Its characteristic equation can be obtained as before:

2
§ uj ·
¨D j ¸ D iD i 0
¨ c ¸¹
©
or, writing the indices explicitly:

2
§ u1 u u ·
¨¨ D 1 2 D 2 3 D 3 ¸¸ D 12 D 22 D 32 0 [6.24]
© c c c ¹
3 Letting DT D t D j u j c and D iD i r 2 , the previous equation can be written
D T2 r 2 0 ; it represents a cone of revolution around axis OD T in a 4D space.

The nature of the quadratic form can be easily obtained by a geometric

interpretation. Consider the vector OA D 1 , D 2 , D 3 . Equation [6.24] can be
written in the form:
G 2
§ ·
¨ OA. V ¸ OA
2
0 [6.25]
¨ c¹¸
©
G
This equality shows that the projection of V c on OA must be equal to OA. The
existence of the non-zero vectors OA is possible only if M V c is greater than 1,
in other words if the flow is supersonic. We recover the result already obtained in
section 5.3.2.3. We deduce from [6.25] the value of the angle E between the normal
to the characteristic surface and the velocity direction:
cos E cV 1M
The velocity thus makes the complementary angle T with the characteristic
surface; and so the result of section 5.3.2.3 is recovered ( sin - 1 M ).
6.2.5. Elementary solutions in irrotational flows
We will now examine some elementary solutions in simple examples of potential
equations. We will first consider the case of an incompressible fluid. The velocity
potential satisfies Laplace’s equation. Subsonic flows verifying an elliptic equation
have similar properties, but are modified by the compressibility of the fluid
([YIH 77]).
The second case studied is the acoustic wave equation, which can be obtained
via linearization and a suitable referential change in equation [6.18] and which
represents the local properties of all second order hyperbolic equations.
6.2.5.2. Irrotational 2D plane flow of an incompressible fluid

The problem comes down to the solution of Laplace’s equation with free-slip
conditions imposed at the solid boundaries. The best adapted means for the study of
these flows involves the use of complex variables. In effect, relations [6.26] defining
the velocity potential M and the stream function \:
w\ wM w\ wM
u v [6.26]
wy wx wx wy
are Cauchy relations between the derivatives of the real and imaginary parts, M and
\, of an analytic function F z of the complex variable z x jy :
F z Mx, y j\ x, y [6.27]
The function F z is the complex potential of the flow considered. Its derivative
F ' z with respect to z is the complex velocity of the expression:
F ' z u jv [6.28]
If the function F z is analytic, the function jF z is also. The velocity

potential and the stream function of jF z are, respectively, \ and M. The flows
associated with the two potentials F z and jF z are known as conjugated flows.
Any analytical function of complex variables thus provides two solutions to the
Laplace equation corresponding to two conjugated flows where the curves of
potential lines of one flow are streamlines of the other.
Consider the integral ³C F ' z dz taken once counter-clockwise on a closed path

of the complex plane (x, y). It can be written as a function of the velocity circulation
* and of the volume flow rate q v ³ d\ (section 4.2.1.2.2):
³C F ' z dz ³C udx vdy j udy vdx ³C dM jd\ * jq v [6.29]
If the closed path C does not surround any poles of the function F ' z , then the
function F z is uniform: it takes on the same value after any excursion of the
variable z on the contour C. The flow across C and the circulation of the velocity on
C are zero. If C contains a pole of F ' z , then the value of the function F z
increases by * jq v with each excursion around C (see an example of the vortex in
section 6.2.5.2.2).
As the velocity field is determined by the Laplace equation, the pressure is given
by Bernoulli’s second theorem [6.14].
The simplest example of an irrotational flow is a uniform velocity field

corresponding to the complex potential F z U z B, where U and B are complex
constants. The velocity Cartesian components u Re(U ) and v Im(U ) can be

immediately obtained from the complex velocity F ' z U.
NOTE – Time is not a variable in Laplace’s equation, but it can be a parameter

present in the boundary conditions and the coefficients of the solution. The result of
this is that initial conditions do not have any meaning for this equation.
6.2.5.2.2. Source and vortex

A source and a vortex centered at the origin of the coordinate system are
conjugated flows corresponding to the complex potential F z and to the complex
velocity F ' z expressed in plane polar coordinates (with z re jT ):
A A A
F z ln z ( ln r jT ) F ' z [6.30]
2S 2S 2S z
The application of formula [6.29] to a closed contour C1 (Figure 6.5a) which

does not contain the origin leads to zero volume flow rate and circulation, since after
one path counter clockwise, the variation of the polar angle T is zero. On the other
hand, if the closed contour C2 (Figure 6.5a) contains the origin, the variation of the
angle T is equal to 2S and the integral ³C F ' z dz is equal to j A:
³C F ' z dz * jq v jA [6.31]
C1 M y y
ur
C2 T ut
O x x x
(a) (b) source (or sink) (c) vortex

streamlines
potential lines
Figure 6.5. (a) Integral of complex velocity on a closed path,

(b) flow of a source and (c) of a vortex
When the quantity A is real, application of formula [6.31] over a closed contour
C2 containing the origin leads to a volume flow rate q v A , while the circulation *
of the velocity is zero on all closed curves. The function F z represents the radial
flow caused by a source (positive A) or a sink (negative A) of volume flow rate qv.
The potential F z and the complex velocity F ' z can be written:
qv qv qv
F z ln z ( ln r jT ) F ' z [6.32]
2S 2S 2S z
We deduce from this the velocity potential M, the stream function \ and the
radial and tangential components, ur and uT, of the velocity vector:
qv q vT wM qv 1 wM
M ln r \ ur uT 0 [6.33]
2S 2S wr 2S r r wT
The potential lines are circles centered on the origin O and the streamlines
(# const) are straight lines lying on radii from the origin O (Figure 6.5b).
If the constant A is imaginary, application of formula [6.31] to a closed contour

C2 containing the origin leads to the circulation * jA , or, A j * . The volume
flow rate is zero across any closed surface C. The function F z therefore
represents the flow of an irrotational point vortex centered on the origin (Figure
6.5c). The potential F z and the complex velocity F ' z can be written:
j* * j*
F z ln z (T j ln r ) F ' z [6.34]
2S 2S 2S z
We can deduce from this the velocity potential M, the stream function \ and the
radial and tangential components, ur and uT, of the velocity vector:
* * wM 1 wM *
M T; \ ln r ; ur 0; uT . [6.35]
2S 2S wr r wT 2S r
The potential lines are straight radii coming from the origin and the streamlines
are circles centered on the origin.
NOTES –
1) The values of the volume flow rate and circulation can be easily found from
the components of the velocity by direct calculation; we will leave it to the reader to
verify this.
G
2) The vorticity Z of the irrotational point vortex is zero at all points, except at
the origin where it takes on the value of a Dirac impulse, a multiplying factor
excepted. The circulation * can be alternatively written as the flux of the rotation
vector across the surface S enclosed by the curve C. The flux is only non-zero for
surfaces containing the vorticity impulse.
3) When the source, the sink or the vortex are placed at z and not at the origin,
the variable z in the functions F z and F ' z is replaced by z z 0 .
6.2.5.2.3. Superposed flows

Any linear combination of harmonic functions or of analytic functions of
complex variables is also a harmonic or an analytic function. We can therefore
construct new solutions from known solutions. While the velocity fields can be
superposed, the same is not true for the pressure fields, as Bernoulli’s theorem is not
linear. Let us consider some common simple examples.
The potential and the complex velocity of the superposition of a source and a
sink with the same flow rate or of two vortices of opposite circulation positioned at
the points A and A’ of coordinates (0, ± a) (Figure 6.6a) can be obtained from
equation [6.30]:
F z
K
ln
za K
ln rA
r A ' j T T ' ; F ' z
K
.
a
[6.36]
2S za 2S S z a2 2
y y
M
r' r
T
A' A x A' T' O A x
(a) (b)
Figure 6.6. Flow of a vortex (a), of a source (b)

placed near a solid wall
Similarly, the potential and complex velocity of the superposition of two

identical sources or of two vortices of the same circulation positioned at A and A’
can be written:
F z
K
2S

ln z 2 a 2 K
2S
ln r r ' j T T '; F ' z
K
.
S z a2 2
z
.
with (Figure 6.6b): r AM ; r ' A' M ; T Ax, AM ; T ' Ax, A' M .

The constant A takes on the value qv (respectively j*) for the sources or sinks
(respectively two vortices of identical or opposite circulation). The other quantities
(M, \ and the velocity components) of these flows can also be obtained by taking the
difference or sum of the corresponding values of the base flows.
When the axis Oy is a streamline which can be “solidified” (solid boundary with
free-slip condition) we have a representation of the flow associated with a vortex or
a source in the presence of a plane wall. These interesting specific cases (Figure 6.6)
are obtained respectively with:
two vortices of opposite circulation whose stream functions derived from
* r
[6.35] are equal to \ ln and for which we obtain the axis Oy for r r ' ;
2S r'
two sources of equal flow rate whose stream function (derived from [6.33]) is:
qv
\ T T ' and for which we obtain the axis Oy for T T ' S .
2S
A doublet is a combination of a source and a sink of the same strength, in terms

of their absolute value, of which the distance 2a tends to zero such that the quantity
2aqv is equal to C (moment of the doublet). A series development in a z in formula
[6.36] leads immediately to an expression for the complex potential of the doublet:
F ( z ) C 2S z .
The velocity potential M, the stream function \ and the radial and tangential
components, ur and uT of the velocity vector can be derived:
C cos T C sin T C cos T 1 wM C sin T

M \ ur ; uT
2S r 2S r 2S r 2 r wT 2S r 2
The potential lines (respectively the streamlines) are circles centered on the axis
Ox (respectively Oy) and tangent to the axis Oy (respectively Ox).
6.2.5.2.4. Flow around a circular cylinder

Consider a straight circular cylinder of radius R, of unit extent, placed in a flow
G
(Figure 6.7) whose velocity at infinity is equal to U x (U is a constant or a function
of time). Let us use polar coordinates and consider the complex potential, a
superposition of a uniform flow and a doublet at the origin:
R 2 ¬
F (z ) U 0 z with: z re i [6.37]
z ®
The circulation of the velocity vector on a curve surrounding the cylinder is

zero. The velocity potential M, the stream function \ and the radial ur and tangential
uT components of the velocity vector in polar coordinates can be derived from
[6.37]:
§ R 2 ·¸ § R 2 ·¸
M U¨r cos T \ U¨r sin T
¨ r ¸¹ ¨ r ¸¹
© ©
wM R2 1 wM R2
ur U (1 ) cos T uT U (1 ) sin T
wr r2 r wT r2
We can verify that for r R we have \ 0 cte and u r 0 : the circle of

radius R is a streamline.
Some particular values of the velocity components allow us to outline the form
of the streamlines. In particular, on the circle r R , we have: uT R 2U sin T
and on the axis Ox (T = 0 or S), ur is positive (negative) for r > R ( r R ): there
exist two points A and A’ of zero velocity stagnation points, (section 6.2.5.2.5) of
the flow on the cylinder (Figure 6.7a).
When z tends to infinity, F (z) tends to U z, the complex potential of a uniform

flow. The velocity field presents a symmetry with respect to the axes Ox and Oy
(between upstream and downstream).
We obtain an irrotational flow with circulation * around the cylinder of radius R

by superposing the preceding flow and a point vortex, whose streamlines are circles
centered on the origin. The complex potential and the complex velocity of this flow
are:
R2 j* R2 j*
F ( z) U (z ) ln z; F ' ( z) U (1 ) [6.38]
z 2S z 2 2S z
y y
U U
M
r
T
A O A' x A O A' x
(a) (b)
y
y
A
A
O x O x
U U
(c) (d)
Figure 6.7. Flow of an inviscid fluid around a circle:

(a) circulation * 0; (b) * 4RU ; (c) * 4RU ; (d) * 4RU
The expression for the velocity uT R 2U sin T * 2S R on the circle of

radius R indicates that two stagnation points A and A’ are found on the circle if
* 4S RU (Figure 6.7b). These are joined for * 4S RU (Figure 6.7c).
For * ! 4S RU , the points of zero velocity are the solutions of the equation
F ' z 0 ; letting z j y a , we find for ya two roots, only one of which is external
to the circle of radius R (Figure 6.7d).
Bernoulli’s second theorem allows us to calculate the pressure pC on the cylinder

from the velocity distribution on this one. We here limit ourselves to the case of a
steady flow, as unsteadiness introduces secondary effects due to the added mass
([YIH 77]). We have:
pc U 2U 0 sin T * 2S R 2 2 constant [6.39]

G
The force F exerted by the fluid on the cylinder (per unit extent) can be
G G G
decomposed into the drag D and the lift L ( F T x P y ). These can be calculated
from expression [6.39] for the pressure. We find immediately:
2S 2S
D ³0 p c cos T .RdT 0; L ³0 p c sin T .RdT U *U
Regardless of the form of the obstacle, these results are true for the drag
(d’Alembert’s paradox) and for the lift (Kutta-Joukowski theorem [YIH 77], [PAR
98]).
Comparisons with experiment
In the case where * = 0, this theoretical pressure distribution on the cylinder
(Figure 6.8, curve a) can be compared with experimental results. In Figure 6.8, we
p pf
have shown pressure variations C p (difference between the wall
UU 2 2
pressure p and the pressure p f in the uniform flow normalized by the dynamic
pressure U U 2 2 ) as a function of the angle T (defined in Figure 6.7a).
Cp
1
S S 0 O S S T
c
– b -1
a
-2
-3
upstream face downstream face upstream face
Figure 6.8. Distribution of the pressure coefficient on a circular cylinder (* = 0):

(a) irrotational flow; (b) laminar separation; (c) turbulent separation
The values of the preceding calculation are relatively close to those measured on
the upstream side of the cylinder, up to the angular position where a separation of
the flow from the cylinder occurs and a wake is formed (section 6.5.3.7). The
pressure measurements at locations situated on the wall where the flow has
separated illustrate a strong flow dissymmetry between the upstream and
downstream faces, which leads to a non-zero value for the drag (whence
d’Alembert’s paradox given to the “theoretical” result). The difference, which is
quite small, between the calculation and the measurement on the upstream face of
the cylinder comes from the fact that an inviscid fluid flow is not produced about the
cylinder, rather it is produced about the ensemble constituted by both the cylinder
and its wake.
The non-zero drag force D exerted by the fluid on the cylinder can be explained
by the presence of the wake on the downstream face, on which a pressure force is
exerted which is greater than that exerted on the upstream face. This force, known as
pressure drag, is obviously proportional to the dynamic pressure U U 2 2 . The value
of the separation angle D (and therefore of the drag) is different ([SCH 99], [YIH
77]), depending on where the boundary layer (section 6.5.3) is laminar (subcritical
flow, Figure 6.8, curve b) or has become turbulent (supercritical flow, Figure 6.8,
curve c).
The lift due to the circulation (Kutta-Joukowski theorem) is indeed observed for
wing profiles and for cylinders in rotation. However, the question as to the
mechanism by which the circulation has been created has not been discussed. The
latter is created by the beginning of the fluid movement about the airfoil as a result
of viscous stresses on the wall (see section 6.6.4.1). However, the Kutta-Joukowski
theorem is satisfied, and the effect of the lift is a curved trajectory for bodies being
in rotation (the Magnus effect); this phenomenon is used in games with balloons and
balls (the balls are “cut”).
The lift of a stationary circular cylinder can also result from actions which
generate dissymmetries of the wake by modification of viscous effects in the vicinity
of the wall (dissymmetric sucking of the boundary layer).4
6.2.5.2.5. Kzn potential flows

Consider the plane polar coordinate system (r,T) and flows whose potentials and
complex velocities are given by:
F z Kz n K r n cos nT jK r n sin nT
F ' z nKz n 1 nKr n 1e j n 1T
The straight lines T S n are streamlines terminating at, or issuing from, the
zero velocity point z 0 (for negative n). The case n 2 corresponds to the usual
stagnation point of a flow (points A and A’ of Figures 6.7a, Figure 6.7b and point A
in Figure 6.7d). The case n 3 corresponds to a higher order stagnation point
(point A in Figure 6.7c). Taking viscosity into account in these flows is possible
with self-similar solutions of the boundary layer where n can take on any value
([SCH 99], [YIH 77]).
4 A ship with “sails”, “l’Alcyon”, has been built using this principle by Y. Cousteau and L.
Malavard.
6.2.5.3. The wave equation

While the use of characteristics allows in principle the solution step by step of
the acoustic wave equation, writing a complete solution in this way is generally
difficult. We can immediately verify that a progressive plane wave is an elementary
solution of [6.22]:
M x i , t
f OM .n ct f x i ni ct [6.40]
G G
(ni: direction cosines of the unit vector n normal to the plane wave ( n 1 )).
The velocity and the pressure fluctuation p' can be found from solution [6.40]
and from a linearized version of Bernoulli’s second theorem [6.13]:
wM
ui ni f ' x i ni ct ; V ui ui f ' ; p' U U cf ' xi ni ct U cV
wt
G
Taking the axis OX parallel to the normal n , the quantity x i ni of the problem of
the function f is equal to X, and the function f can be written M x i , t f X ct .
This form reveals the transmission without signal deformation (velocity potential,
velocity or pressure). We note that if f X ct is a solution of [6.40], the same is
true of g X ct which propagates in the opposite direction.
The superposition of certain waves can eventually lead to the disappearance of

the propagative character, and we obtain stationary waves in which all points in
space are in phase, as shown by the following simple example:
cosZ t kX cosZ t kX 2 cos Z t cos kX
Conversely, any suitable superposition of harmonic stationary waves can lead to

one progressive harmonic wave:
The ensemble of progressive plane waves is thus equivalent to the ensemble of

stationary plane waves.
The equation for spherical waves M r, t can be immediately obtained (the

expression for the divergence can be obtained by applying Ostrogradski’s theorem
between two spheres of radius r and r + dr). It is written:
1 w 2M 1 §¨ w 2 rM 1 w 2 rM ·
'M 0 [6.41]
c 2 wt 2 r ¨© wr 2 c 2 wt 2 ¹
Its solution is analogous to that of the plane wave equation:
1 1
M r , t f r ct g r ct
r r
In the case where g { 0 , the expressions for the pressure and the velocity are:
wM Uc wM 1 1
p U f ' r ct ; u f ' r ct f r ct
wt r wr r r2
The expression for the velocity of spherical waves differs from the expression
1
for plane waves by the term f r ct , which dominates in the vicinity of the
r2
origin r 0 (nearfield term). We have, for the volume flow rate qv:
wM
q v r , t 4S r 2 4S >rf ' r ct f r ct @
wr
We see that the flow rate at the origin Qt is equal to

Qt q v 0, t 4S f ct .
The velocity potential M and the pressure p can be thus be written:
1 § r· Uc § r ·
M Q¨¨ t ; p Q ' ¨¨ t
4S r © c¹ 4S r © c¹
6.2.6. Surface waves in shallow water
6.2.6.1. 2D equation for potential

We will now consider a problem governed by an elliptic partial differential
equation in space x, y , z , but whose boundary conditions induce propagative
phenomena in the plane x, y .
Consider a horizontal plane Oxy, on which a layer of liquid is subjected to a

vertical gravitational action. Let ] ] x, y be the height of the free surface of the
liquid. If the movement of the liquid is considered irrotational, the velocity potential
Mx, y, z , t satisfies Bernoulli’s equation:
2
wM V p
gz co nst [6.42]
wt 2 U
v
w ]
u
x
Figure 6.9. Flow with free surface in shallow water
As the liquid is inviscid, the velocity is horizontal at the bottom z = 0 where w is

zero (slip condition). We will consider the case of shallow water: the flow is locally
uniform (the horizontal components u and v are independent of z), and the thickness
e is small compared with the horizontal distance L characteristic of variations of the
G
velocity V . The volume conservation equation:
wu wv ww
0 [6.43]
wx wv wz
allows the magnitude of the w component to be determined: the first two terms being
of order V/L, we obtain w | Ve L . The vertical component of the velocity w is of
G
second order with respect to the velocity V . The component w can be calculated by
integration following z of equation [6.43]:
§ wu wv ·
w z ¨¨ ¸¸ [6.44]
© wx wv ¹
The velocity potential M is thus the sum of two terms:

– a function I(x,y,t) whose gradient comprises the u and v components of the
velocity;
z2
– a small term of order 'I associated with the w component which can be
2
written using [6.44] by expressing u and v as a function of I:
w z 'I [6.45]
The free surface is a material surface; its vertical velocity w z ] is the material
derivative of the height ] of the fluid particle being on this surface (kinematic
condition):
d] w] w] w]
wz ] u v
dt wt wx wy
Substituting into [6.45] gives:
d]
] 'M [6.46]
dt
As this w component is small compared with u and v, Bernoulli’s theorem leads

to a z-distribution of the hydrostatic pressure, over a vertical section, which depends
on the height ] x, y, t of the free surface on which the atmospheric pressure pa
acts:
p U gz p a U g] [6.47]
Relation [6.42] can thus be written:
s* u 2 v 2
g const [6.48]
st 2
We replace ] in [6.46] with its expression taken from [6.48] in order to obtain
the potential I:
1 d §¨ wI V 2 ·
'I 0 [6.49]
g] dt ¨© wt 2 ¹
6.2.6.2. Analogy with a compressible fluid
Letting c 2 g] , we see that equation [6.49] can be written in a form identical

to equation [6.18] for the potential of a compressible fluid with two spatial
dimensions. This equation, which is hyperbolic, represents the propagation of

surface waves of velocity amplitude V and height ]
Comparison between the corresponding equations [6.15] and [6.46] of these two
problems shows that the height ] is analogous to the density U. However, the
expression for the velocity of sound c 2 J p U kU J 1 of the compressible fluid
and that c 2 g] of the free surface indicates an exponent J equal to 2 for the
equivalent compressible fluid.
In physical terms, compressing a gas or elevating the free surface of a fluid

creates a reactive force in the form of a pressure increase (section 5.3.6) or of driving
pressure corresponding to the “elastic” energy of an oscillator. The hyperbolic
character of equation [6.49] leads to the existence of shock waves in the form of a
hydraulic jump ([YIH 77]).
6.2.6.3. Influence of surface tension

We have previously assumed (section 6.2.6.1) the continuity of pressure across
the free surface. However, waves of small wavelength require surface tension V to
be taken into account. The pressure difference G pV due to surface tension is given
by Laplace’s law (section 2.2.1.4.2) which can be written by expressing the average
curvature, accurate to second order:
§1 1· § w 2] w 2] ·
G pV p pa V ¨¨ ¸¸ V ¨ ¸ V ']
©R R' ¹ ¨ wx 2 wy 2 ¸
© ¹
Condition [6.47] is replaced by:
p U gz p a U g] G pV p a U g ] V ']
Substituting into Bernoulli’s equation [6.42] gives the relation:
s! V 2 p s! V 2
gz g ' const
st 2 st 2
which, associated with [6.46], gives a complex system which we will not study here
(see [YIH 77]).
6.3. Orders of magnitude
6.3.1. Introduction and discussion of a simple example

The mathematical variable properties of a continuous medium are relatively
regular. The equations governing a continuous medium assume the continuity and
the derivability to at least second order in the physical quantities, with the exception
of regions where shocks or discontinuities occur. The validity of physical models
(axioms of the continuous medium or models obtained from kinetic gas theory)
implies that the physical quantities observed are solutions of ordinary or partial
differential equations whose behavior is locally regular.
It is thus reasonable to admit that a quantity f undergoing variations in the order

of 'f on an interval of a time or space variable of length L, possesses temporal or
spatial derivatives of the order of 'f L and that their second derivatives under the
same conditions are in order 'f L2 . Such a hypothesis should be subsequently
verified in discussing the results which can thence be obtained. The scale L
corresponds to the interval over which the function f varies. For example, for an
exponential function L is the characteristic dimension of the exponential variation
(space or time constant).
The preceding considerations result from the fact that a derivative is the ratio
limit of finite increases of the function and the variable, when the latter tends to
zero. It is clear that to within a factor of at most a few units, this derivative is equal
to the ratio of the finite increases in the region considered.
A partial differential equation (or an ordinary differential equation) is a

numerical balance between a certain number of terms containing derivatives. If this
equation only contains two terms, the absolute values of these are equal. On the
other hand, if the equation contains a sufficiently large number of terms, certain of
these are dominant in a given part of the domain, other terms being more important
in other regions. Each zone of the domain can thus be characterized by the locally
dominant physical phenomena.
Take the elementary example of the mass-spring oscillator with one degree of
freedom:
[6.50]
Suppose that we do not know the solution. We search first of all if a

characteristic time W exists for the phenomena described by this equation and
corresponding to a movement of amplitude xm.
The term kx of given order of magnitude kxm is in the same order as at least one
of the other two terms of the equation which are, respectively: O mx m W 2 and
O f x m W .
Let us first suppose that f x is small compared to kxm which is then of the same
order as mx :
mx m
mx | k x | k xm W | mk
W2
The characteristic time W necessary for the amplitude to vary from zero to xm is
of order m k and we also have f mk .
To first approximation, equation [6.50] can be written:
mx kx 0 with: x0 x 0 ; x 0 0 [6.51]
If, on the other hand, mx is small, then it is the term fx which balances kx ; we
thus have:
f xm
f x | kx | k xm W | f k
W
The characteristic time W necessary for the amplitude to vary from zero to xm is of
order x f / k, with f mk ; equation [6.50] thus reduces to:
[6.52]
The preceding order of magnitude analysis allows us to define the characteristic

time W for the phenomena described by the initial equation and to obtain an
approximate equation for this particular time scale. Simple general considerations
allow us to show that equation [6.51] represents an oscillatory movement, at least at
the scale W. For example, it is useful to discuss the equation in the plane (x, x ), often
referred to as the phase-plane. The trajectory [ xt , x t ] is there a curve
parameterized by the time and a simple qualitative discussion using equation [6.51]
allows us to easily see the form of this (Figure 6.10a), based on the fact that x is the
derivative of x , and its sign indicates the direction of the variation x .
Equation [6.52] corresponds to a damped aperiodic movement. The trajectory in

the phase-plane is therefore a straight line with slope k f . We will later discuss
the effect of the second initial condition x 0 0 which obviously had to be
abandoned for the first order differential equation.
If we have equality in the orders of magnitude f | mk , the three terms of the

equation must be conserved, but we will pass gradually from the form of Figure
6.10a to that of Figure 6.10c.
x x x
x O x O x
(a) (b) (c)
Figure 6.10. Evolution of the oscillator in the plane (x, x ): (a) oscillator with a small
damping; (b) damped oscillator; (c) aperiodic motion (strong damping)
It then remains to study the influence of the small term neglected in each case:
this is a perturbation problem which we will discuss a little later (section 6.4). In the
first oscillatory case, we study the influence of friction with the balance equation for
the mechanical energy derived from [6.50] by multiplying by x and integrating
between 0 and t. We have, taking account of the initial conditions:
mx 2 t kx 2
kx 02
³0 f x 2 dt [6.53]
2 2 2
Equation [6.53] immediately shows the following properties:

1) kinetic energy is localized in the vicinity of the origin x=0;
2) potential energy is localized in the vicinity of the extrema of the stretching
motion;
3) kinetic energy can be transformed into potential energy such that the total
mechanical energy decays with time;
t
4) dissipated energy ³0 f x 2 dt is an increasing function of time; the result of this
is that the velocity must tend to zero if the integral is to remain finite; we can write
equation [6.53] in the form:
t
§ mx 2 kx 2 ·
t
³0 f x 2 dt ¨ [6.54]
¨ 2 2 ¹
© 0
Let Wa be the damping time of the oscillations; let x m denote the maximum of
the velocity at the beginning of the movement which is considered only lightly
damped; this leads to the following orders of magnitude:
2
t 2 2 mx m
³0 f x dt | W a f x m |
2
m
From this we can derive the order of magnitude of the damping time: W a | .
f
It should be noted that the preceding considerations are concerned with orders of
magnitude which do not require an exact expression of equation [6.50]; they are in
fact valid for any equation whose terms have the orders of magnitudes posed above.
Therefore, an approximate knowledge of results do not require complex
mathematical procedures above and beyond the numerical discussion about
monomes (“rule of three”). By this procedure, which needs to be completed by a
discussion of perturbation problems, we can identify the important terms in a system
of equations, in other words, the dominant physical phenomena in each of the zones
of the problem domain. The order of magnitude of the unknown quantities can also
be deduced from this analysis.
6.3.2. Obtaining approximate values of a solution

6.3.2.1. Principles
Most particular functions (called “elementary” or “special”) are solutions of
linear equations with simple algebraic coefficients. For linear partial differential
equations (Laplace equation, wave equation, heat equation, Maxwell’s equations,
etc.), these functions are often useful in the search for solutions with particular
boundary conditions. As the equations of fluid mechanics are not linear, the
elementary or special functions are rarely directly useful.
Having simplified the equations and qualitatively discussed the various

phenomena, we can obtain approximate values in a simple manner by searching for
a global solution in each zone where the equations can be simplified. The method
which we will use here can be generally applied, insofar as we suppose that we
ignore the exact solution of the equations, and that the particularities of the equation
linearity and the coefficient constancy are not used. The method consists in
searching for a global condition which can be obtained by integrating the equations
over each interval of the study. We then represent the form of the solution by a
plausible function in which the main unknown value is a parameter which we can
deduce from the preceding global condition. We will first apply this technique to the
preceding example of the oscillator.
6.3.2.2. Global solution for the linear oscillator

Let us take for W the first instant where the abscissa x is zero (Figure 6.11a), the
velocity being then approximately maximum; this value W is more or less equal to a
quarter of the period T of the oscillatory movement. Let us consider equation [6.51]
with the conditions:
x0 x0 ; x 0 0; xW 0 [6.55]
x (t ) x t
x0 x m
O W t O W W t
(a) (b)
Figure 6.11. (a) Law of motion during the first quarter of a period;
(b) variation of the velocity during half a period
Taking these conditions into account, the integration of equation [6.51] between
0 and W gives the global condition:
W
m x W k ³0 xt dt 0 [6.56]
First write the function x(t) which in the form xt x 0 M K of a non-

dimensional function of the variable K t W satisfying [6.51] and conditions [6.55]:
2
k
! ! 0 with ! 0
1, ! 1
0 and ! 0
0 [6.57]
m
Integrating [6.57] between 0 and 1, we obtain:
W 2k
M 1 ³0 M K dK
1
0 [6.58]
m
Equation [6.58] allows us to determine the unknown W if the function M is known.

It remains to choose the function M K , whose general shape we know in the phase
plane from the discussion of section 6.3.1.1. The simplest algebraic function
satisfying conditions [6.57] is:
M 1K 2 [6.59]
Substituting into [6.58] we obtain:
W 2k
1 0
3m
whose positive root 3m k only is acceptable. Thus, for the period T 4 3m k

we obtain the value 4 3 6.928 which is slightly larger than the exact value
2ʌ 6.28 (error of 10%). The reader will note that the error level is quite small,
given the crudeness of the computation.
However, we can hope to improve the result relatively easily by imposing

supplementary conditions resulting from equation [6.57] on the second derivative
MK at instants 0 and 1:
W 2k
M0 , M1 0 [6.60]
m
Consider firstly second conditions [6.60]: M1 0 . The simplest polynomial

satisfying this and conditions [6.55] is of third order:
3K 2 K3
M K 1 [6.61]
2 2
Substituting into equation [6.58] gives:
5 W 2k
3 0
4 m
We therefore obtain T 4W 4 12m 5k ; the value 4 12 5 6.197 only

differs from the exact value by 1.4%.
The simplest polynomial satisfying the conditions [6.55] and [6.60] is of fourth
order:
W 2 k §¨ K 2 5K 3 K 4 ·¸
M K 1 2K 3 K 4
m ¨© 2 6 3 ¸¹
hence:
W 2k W 2k 7
M 1 2 and ³0 M K dK
1

6m 40m 10
As the form of the curve representing the function M depends on the unknown
parameter W 2 k m , we obtain, by substituting into [6.58] a second order equation:
2
1 §¨ W 2 k ·¸ 13 W 2 k
2 0
40 ¨© m ¸¹ 15 m
which has roots 2.497 and 32.16. It is easy to see that the second value is not
suitable, M having to be positive on the interval [0,1]. With the first value, the
calculation of the period gives:
m m
W 1.58 T 4W 6.32
k k
The use of boundary conditions taken from the equation for the second
derivative thus leads to an improvement (error less than 1%). It should be noted
however that it is not possible to further improve the results of such a method, which
only uses local data at the extremities of the interval considered.
6.3.2.3. Damping of the oscillations

In an oscillatory regime, the essential movement corresponds to an exchange
between the kinetic and potential energy terms; only a small amount of the
mechanical energy is dissipated in each period. The interest of the kinetic energy
theorem [6.54] is that it gives an expression for the dissipated energy as a function
of the variation of total mechanical energy: this is equal to the variation of maximum
2
kinetic energy mx m 2 , x m (t ) being the maximum velocity value at each stretching
value of the mass m equal to zero. During a half-period 2W (notation as in the
preceding section) contained between two instants t and t + 2W where the elongation
is zero ( x x m ), equation [6.54] translates this mechanical energy variation:
§ mx 2 · 2 p 1W
1 t 2W
G ¨ ¸ mx 2 ³t f x 2 dt [6.62]
¨ 2 ¸ 2
© ¹ 2 p 1W
Suppose as before that the amplitude of the velocity x m varies only slightly and
that it can be considered as a constant during the half-period, while the velocity
variation is symmetric with respect to the instant W (Figure 6.11b). The law for the
velocity xt x 0 M K results from the choice of the preceding function M K :

! t
x t
x0 ! t
x m with: x m x0 ! 1
! 1
Taking the time origin at the beginning of the half-period, at the instant where
x 0 , and designating the velocity amplitude by x m , gives, after substitution into
[6.62]:

2
m t 2W 2 f x m W
³0 M K dK
1 2
G x m2 ³t f x 2 dt [6.63]
2 M 1
2
Relation [6.63] is a finite difference equation with a step 2W for the amplitude
x m , which we can replace with the differential equation:
1 2

2
m d xm
2
f xm with:
¨0 !
d
2 dt ! 2 1
or:
m d x m
x m 0 [6.64]
D f dt
The differential equation [6.64] can be solved as before using a global method.
We will here simply note that it represents a first order damped system with time
constant W a m D f .
It remains to calculate the constant D. From the parabolic law of motion [6.59]
we have:
The third order law of motion [6.61] gives:
2m
The exact solution for the damping is W a . The error is a little greater than
f
for the calculation of the period: this is because a given approximation is always
better for a function M than for its derivative M .
6.4. Small parameters and perturbation phenomena
6.4.1. Introduction
The equations governing a physical phenomenon involve various non-

dimensional parameters. Very often, some of these are small. The associated terms
appear as a perturbation of the equations which are obtained when these terms are
zero. Mathematical phenomena associated with these perturbation terms can be
complex and their study should be carefully effected in order to understand their
precise role. We will limit ourselves in this section to the usual elementary cases in
fluid mechanics.
6.4.2. Regular perturbation
6.4.2.1. Elementary example

A perturbation is called regular if the effects resulting from the perturbations
terms are everywhere in the same order of magnitude as the parameter which
characterizes them. A good example involves a first order damped system governed
by the following equation with the small parameter H:
x x 1 H x0 0
whose solution is (Figure 6.12a): x 1 H (1 e t ) .

This idea of a regular perturbation is associated with the mathematical idea of
uniform convergence which we will recall here briefly: a family of functions f (t , H)
converges uniformly towards the function f (t ,0) if the difference between the two
functions is independent of the value of t on a closed interval, and tends to zero with
H. The limit f (t ,0) of the family of functions is thus continuous. This condition is
visibly satisfied for the preceding example.
(1+ H)[1-exp(-t)]
x x
1+H
H
Ho 1
1
1-exp(-t / H)
1-exp(-t) Ho
O 1 (a) t O H (b) 1 t
Figure 6.12. (a) Regular perturbation; (b) singular perturbation
6.4.2.2. Regular perturbation of linear differential equations

Consider the linear differential problem which depends on the parameter H,
whose solution xt , H converges uniformly towards xt ,0 when the small
parameter H tends to zero; we can often search for xt , H in the form of a power
series expansion of the parameter H, if this parameter has been suitably chosen:
xt , H x 0 t Hx1 t H 2 x 2 t ... H i x i t ... [6.65]
Substituting the preceding expression into the differential equation, we obtain a

series in increasing powers of H, the identification of whose coefficients in the two
sides provides successive differential equations. Let us first consider a simple
example of a linear differential equation in order to illustrate the computation5
mechanism. Consider the differential equation representing the movement of a point
subjected to a weak repulsive force, t varying on [0,1]:
5 We assume that the domain of study corresponds to an time interval bounded a priori in
order to ensure uniform convergence.
xt H xt a x0 0 x 0 1 [6.66]
Substituting development [6.65] into differential equation [6.66], we obtain the

successive differential equations:
x0 t a x 0 0 0 x 0 0 1
x1 t x 0 t x1 0 0 x1 0 0
x2 t x1 t x 2 0 0 x 2 0 0
..............
x p t x p 1 t x p 0 0 x p 0 0
..............
The first differential equation is the non-perturbed equation which corresponds

to H 0 ; the successive differential equations only depend on the functions already
calculated, of lower rank in the development. The boundary conditions can be
carried back into the first equation, provided they do not contain the parameter H.
The solution of the preceding system can be calculated immediately from place
to place and we thus obtain a series development of the solution:
t2 t4 t3 t6 t5
x0 a t, x1 a , x2 a ,
2 4! 3! 6! 5!
[6.67]
t 2 p 1 t 2 p 1
...., xp a , ........
>2 p 1@! 2 p 1!
However, a development including many terms is not of much interest,
particularly if we consider the computational methods used by computers.
Furthermore, if the parameter Htakes on values which require many terms of the
development, the principal properties of the unperturbed equation are significantly
modified: in other words the unperturbed equation is no longer a sufficiently
representative model for the mathematical and physical properties of the solution for
these values of H.
For example, we easily recognize that [6.67] is the series development of the
solution which is here easy to calculate directly:
sh t H
x
a
H
>cht H 1@
H
When H is no longer small, the preceding series development is of limited

practical interest, despite an infinite radius of convergence, as the behavior of the
perturbed solution is too far from that of the unperturbed solution.
6.4.2.3. Regular perturbation of non-linear differential equations

Consider a given first order differential equation:
x f t , x, H x0 a
Calculating the development of the function f t , x, H in increasing powers of

the small parameter H becomes quickly complicated, and in general we are satisfied
by a limited development. We will here only outline the principle of the method and
the beginning of the calculation.
As before, we seek a solution of the form [6.65]:
xt , H x 0 t Hx1 t H 2 x 2 t ... H i x i t ...
We have:
H2
f t , x, H f t , x,0 HfH t , x,0 fHH t , x,0
2

f t , x0 Hx1 H 2 x2 ..,0 HfH t , x0 Hx1 H 2 x2 ..,0 H2
2
fHH t , x0 Hx1 ..,0
f t , x0 ,0 H > fH t , x0 ,0 x1 f x t , x0 ,0@
ª 1 x2 º
H 2 « x2 f x t , x0 ,0 fHH t , x0 ,0 x1 fHx t , x0 ,0 1 f xx t , x0 ,0 » H 3....
¬« 2 2 ¼»
Substituting the preceding development into the differential equation and

identifying the following increasing powers of the small parameter H, we obtain:
x 0 f t , x 0 ,0 0 x 0 0 a
x1 x1 f x t , x 0 ,0 f H t , x 0 ,0 x1 0 0
1 x12
x 2 x 2 f x t , x 0 ,0 f HH t , x 0 ,0 x1 f Hx t , x 0 ,0 f xx t , x 0 ,0 x 2 0 0
2 2
The first differential equation corresponds to the zero perturbation for H. We note
that the successive differential equations are linear for the corresponding unknown
function, with a right hand side which only depends on the previous solutions. The
equation linearity facilitates their numerical resolution. However, the expressions for
the differential equations can quickly become complex.
The interest of regular perturbation methods is particularly evident for problems

governed by partial differential equations, if particular solutions can be found which
have a simple mathematical structure (for example, one which approaches
differential equations), in which case the equations resulting from the application of
the regular perturbation method have a structure analogous to the initial solution.
We will see different examples of these methods applied to slightly unsteady flows,
to thermal systems which do not vary too quickly, and to the problems where inertia
terms due to geometric variations must be accounted for.
6.4.2.4. Choice of a perturbation parameter

The perturbation parameter chosen for a study is of the utmost importance.
Suppose that a model leads to the following equation:
§ ·
x ¨ ¦ a n H n ¸ xt , H bt x0 A [6.68]
¨ ¸
© ¹
It is clearly possible to seek a solution in the form of a development of the

solution in powers of H
d
x(t, %) x0 (t) %i xi (t)
i1
f
However, it is more interesting to take D H ¦ a i H i as the perturbation
i 1
parameter and to seek a solution of the differential equation:
x a 0 D H xt , H bt x0 A
in the form x (t, ) x0 (t) x (t).

i1
i
i
We find for the successive differential equations:
i
x i ao xi (t)=- aj xij (t) xj (0) 0
j 1
x i ao x i (t)=-x i1 x i (0) 0

The system obtained is far simpler for ~ x i than for xi. The preceding example is
apparently rudimentary. However, it translates the fact that the model has attributed
an important role to the parameter D, rather than to H which was chosen in order to
establish the model. It is clear that we can have a better development with the
parameter D than with H. This problem of parameter choice is often encountered in
order to best represent the range of solutions, for example, for solutions of boundary
layer equations ([SCH 99], [YIH 77]). The term b(t) of [6.68] can depend on
Hwhich must therefore express as a function of D.
The method can be applied to partial differential equations. We will later see
some examples of this (section 6.4.2.6). Suitable variable changes also allows the
modification or simplification of the differential problem (section 8.5.3.2).
The practical limits of the preceding method are determined by the convergence
of the entire series, but even more by the speed of convergence of the series
obtained. Methods for accelerating the convergence can be used here ([ABR 65]
p. 16, [BRE 91]).
6.4.2.5. Regular perturbations and orders of magnitude

In the domain of studies where the orders of magnitude of the terms are fixed,
knowledge of a solution (exact or approximate) of an unperturbed problem allows
the calculation of correctional terms for the solutions in the neighborhood of the
base solution.
The successive differential equations obtained are linear equations which all
have the same linear operator, the right hand sides being known at each stage from
the preceding solutions. Numerical solution is thus simplified. The computation of
higher order terms of the solution by means of analytical developments, is generally
difficult in practice, on account of its complexity.
6.4.2.6. Applications in fluid mechanics

With the exception of viscous stresses, taking account of other phenomena in
fluid mechanics, when these are relatively weak, very often leads to regular
perturbations: unsteady effects in established flow (in other words a flow which is
independent of its initial conditions), effects of compressibility in steady flow, weak
geometric changes, etc.
Consider the established flow of an incompressible fluid with constant viscosity

in a rectilinear pipe of arbitrary cross-section (Figure 6.13), and let us suppose that
we have a Poiseuille flow, with a driving pressure gradient wp wx , parallel to the
velocity in the direction Ox, which is a given function of time; the velocity satisfies
the following equation (from [4.21]) and boundary conditions:
[6.69]
We will consider the following non-dimensional parameters and variables:
t u § y z· U D2 D 2 wp
f t
~ ~
t u~ ~y , ~z ¨¨ , ¸¸ H
T U ©D D¹ PT P U wx
where D and T are, respectively, a reference length for the cross-section and a unit of
time to be defined.
wu~ §~ w2 w2 ·
f t ' u~;
~ ~ ¨' ¸
H ~ [6.70]
wt ¨ w~
y 2 w~z 2 ¸
© ¹
z
u
y
x
C
Figure 6.13. Established flow in a cylindrical tube
Assume Hto be small and consider that equation [6.70] is the result of a regular
perturbation of this equation for H = 0. We seek the solution of [6.70] in the form
f
u = İ i ui . Substituting into equation [6.70] and identifying terms following the
i =0
increasing powers of H, we obtain the system:
su 0 su1 su i

0 f t
' u 0 ; ' u1 ; ' u 2 ; ...; ' u i1 ; ...
st st st
with the boundary conditions: u~i y, z 0 ~y , ~z C~ i 0,1, 2,..., n,...

It is immediately verifiable from place to place that the functions u~i are variables
of separated functions and that they are proportional to the successive derivatives of
f(t):
f
¦ g i ( ~y , ~z ) f (i ) t
~ ~
u~ ( ~
y, ~
z, t ) [6.71]
i 0
The functions g i ( ~
y , ~z ) are solution of the system:
' g0 1; ' g1 g0 ; ' g2 g1 ; ... ' gn gn1 , ...
with the boundary conditions: gi (y , z ) 0 (y , z ) C ; i 0,1,..., n...
The preceding equations can be solved successively with increasing i, and the
first equation corresponds to the Poiseuille flow (section 3.4.2.5). Let H = 1, which
amounts to taking T D 2 Q as a unit of time. The expression for the velocity
becomes:
f
¦ f (i ) t g i ( ~y , ~z )
~ ~
u y, z , t Uu~ ( ~
y, ~
z, t ) U
i 0
The volume flow rate can thus be easily expressed as a function of the successive
derivatives of the pressure gradient:
f
¦ f (i ) t ³S g i ( ~y , ~z )d~s
~ ~
qv UD 2 ³S u~ ( ~ z , t )d~
y, ~ s UD 2
i 0
~ ~ ~
Defining the dimensionless coefficients Gi ³S g i ( y , z )ds , it can be written in
the form:
f
UD 2 ¦ Gi f (i ) t
~
qv
i 0
Let 'p(t) be the driving pressure loss for the length L; the pressure gradient can
wp 'pt
be written in the form: and we have:
wx L
i
D 2 §¨ D 2 · w i § 'p ·
f (i )
~t ¸ ¨ ¸
¸ wt i ¨© L ¸¹
P U ¨© Q ¹
From this we can derive an expression for the volume flow rate as a function of
the pressure loss and its temporal derivatives:
i
D4 f § D2 · w i § 'p ·
qv ¦ Gi ¨¨ ¸ ¨ ¸
¸ wt i ¨© L ¸¹
P i 0 © Q ¹
~ 1 w §~ w ·
For example, in the case of a circular cross-section C ( ' ~ ¨¨ r ¸¸ ) of
r w~
r © w~ r¹
radius r 1 , we easily find:
1 1 3 r 4 ¬
g0 r
1 r 2
; g1 r
r ;
2
4 16 4 4 ®
1 19 r 6 ¬
g2 r
3r r ; ...
2 4
256 9 9 ®
The physical interpretation of the series is not so simple, as the physical

significance of the higher order derivatives implies a long-term property. It suffices
to consider a function f t and a polynomial approximation of this over an interval
of time. The higher order derivatives of these functions are very different, and so,
consequently, are the preceding corresponding series. In fact, we can easily verify
the convergence of the series for exponential or sinusoidal functions whose time or
frequency constants are relatively small, but the preceding representation is of
limited use for these particular cases.
For the example considered, we can show that we have convergence of the series
if the series of the general term f n t 4 n converges. Recall, however, that, from
~
a mathematical point of view, the solutions established for the partial differential
equation does not require the existence of derivatives for the given function f t .
The preceding results can equally be obtained by means of a Laplace transform
(Appendix 1) in the asymptotic approximation for large periods of time.
The regular perturbation of the solution to a non-linear partial differential

equation is more difficult (section 6.4.2.3), and in general we can only obtain the
first terms of the development. Examples of this kind of development (Blasius
series, Görtler series, etc.) are found for boundary layer equations ([SCH 99], [YIH
77]).
The method of regular perturbations can also be applied in acoustics and

energetics. Consider the geometry of Figure 6.13 in which the cylinder of base C is a
solid, with constant physical properties, subjected to a heat release of volume power
P(t) and whose external wall is maintained at constant temperature T = 0.
The thermal conduction equation for this solid and the associated boundary
conditions:
wT
UC P(t ) O 'T ; with: (y, z ) C : T (y, z , t) 0
wt
can be immediately reduced to equation and boundary conditions given in [6.69].
6.4.3. Singular perturbations
In section 6.4.2 we assumed that the differential equation could be solved with
respect to the first derivative such that the unperturbed equation (H=0) possesses a
differentiable solution. This is not the case if the coefficient of the derivative tends
to zero with H faster than the other coefficients. Let us examine the elementary
example of the following differential equation in which the parameter H is now the
coefficient of the derivative x (t ) :
H x' t x(t ) 1 x0 0 [6.72]
The solution is: xt 1 exp t H .
For H to tend to zero, it is easy to follow the evolution of the solution towards its
discontinuous limit (Figure 6.12b) made up of the half-line (x = 1, t > 0) and the
origin O. Letting H equal to zero in equation [6.72], we obtain the solution x 1
which is valid nearly everywhere, and we can no longer satisfy the initial condition
posed at the origin for this part of the solution (called the external solution). We note
that the family of functions xt , H does not converge uniformly towards its limit for
H = 0.
It is therefore necessary to perform a particular study over a small interval in the

neighborhood of the origin where the solution undergoes significant variations over
a small distance (internal solution) in order to recover the boundary condition which
is not satisfied.
The external solution can also undergo a regular perturbation due to the small
parameter H. For example, the solution of the following differential equation:
x x 1 H x0 0
does only differs from the preceding equation by the coordinate 1 + H of the
horizontal asymptote.
6.4.3.2. Methods for studying singular perturbations

The existence of a singular perturbation is related to the presence of a singularity
for the limit solution of a system of differential equations. This phenomenon is
encountered in diverse circumstances and notably in the case of the lowering of the
order of a differential system in which we impose H as zero, or when the limit
solution or one of the derivatives involves a discontinuity on the interval of the
study. We will limit ourselves here to the study of a very simple case in order to
illustrate the general method by which we discuss such problems.
The direct numerical study of a problem involving singular perturbations often

presents difficulties on account of the very large disparity in the characteristic scales
found in different regions of the study. This point can obviously involve difficulties
for numerical methods in fluid mechanics. It is therefore particularly interesting to
directly obtain information independent of the exact solution (often very difficult) of
the differential problem which is posed. We will illustrate this using a very simple
case; the general analysis method involves a number of stages:
1) searching for the external solution: a study of the order of magnitude of the
solution which is assumed to vary regularly over the interval under study allows us
to obtain equations which are satisfied by the external solution;
2) we then try to identify a contradiction with respect to the data of the problem
in order to identify the singularity to study;
3) a new examination of the order of magnitudes of the terms of the equation
provides information pertaining to the scales of the singularity to be studied;
4) it is then possible to identify the equations of the differential problem in the
singular zone (internal solution), these resulting from simplifications of the
complete problem;
5) we then discuss the boundary conditions to be associated with the two
preceding problems for matching their solutions.
6.4.3.3. An elementary example

The example of section 6.4.3.1 is too simple because the internal solution and the
exact solution are identical. Consider the following differential problem:
H x t x(t ) 1D t x0 0 [6.73]

Ignoring the results of the last example, let us apply the methodology described
above:
1) We first look for the external solution by supposing that the unknown function
has variations of the order of one on the intervals of the same order of magnitude.
To O(H), differential equation [6.73] reduces to:
x(t ) 1D t
The equation obtained is here an algebraic relation (a zero order differential

equation) which explicitly gives the external solution which is valid nearly
everywhere.
2) The condition that x(t) be zero at the origin is not fulfilled. We must therefore
complete the preceding solution by an internal solution on an interval with a
different scale to the interval where the external solution is valid, otherwise we will
only recover the latter. Let G be the order of magnitude of the length scale of the
interval over which the neglected term H x must be taken into account.
The interval sought of scale G must be situated at the origin. In effect, if this was
not the case, we would have between this interval o(1) and the origin a finite interval
on which the results of the external zone would be applicable; the function x would
be equal to 1+Dt and would not therefore be zero at the origin, which would imply
that nothing had been solved. The only reasonable possibility is therefore to place
our small interval G in the neighborhood of the origin.
3) The order of magnitude of G can be obtained by supposing that the term H x' is
of the same order as that of the variations of x, i.e. of the order of 1:
H x t x(t ) 1D t
H
| |1 | 1 DG
G
We can deduce from this that G H , and we need to study the behavior of the
solution in an interval of amplitude H close to the origin.
4) We perform the change of variable:
t
K xt X K
H
Neglecting the terms of order H, equation [6.73] can be written:
X K X (K ) 1 X 0 0
Its solution satisfying the condition at the origin can be written:
X K
A 1 e K
It tends to A when K tends to infinity.
5) We now have two pieces of the solution (Figure 6.14b) which we need to
match. This procedure appears relatively empirical and we will consider only small
values of H. We see that as H tends to zero, the value 1 DH of the external solution
for x = H tends to 1. We therefore adopt the following simple matching rule.
The limit of the internal solution F(K) for K becoming infinite is equal to the
limit of the external solution x(t) when t tends to 0:
lim X K lim x t
Kof t o0
asymptote: x = 1+ Dt external solution:

y y x = 1+ Dt
1 1
exact solution
exact solution internal solution:
H 1-exp(-t/H)
O (a) x O H (b) x
Figure 6.14. (a) Exact solution of differential problem [6.73];

(b) matching of external and internal solutions
This rule provides the boundary conditions which were missing for the internal
solution and leads to a value of 1 being assigned to the constant A. We thus obtain
the internal solution:
X ( ) 1 e with: t / %
In fluid mechanics, the internal solution is called a boundary layer. Such

singularities are encountered in all domains of physics (skin effect in
electromagnetism, penetration depth of an evanescent electromagnetic or acoustic
wave, etc.). We will see other examples in problems of heat transfer or chemical
reactions with or without flow.
The singularity can also consist of a discontinuity of the first or second

derivatives.
Perturbation problems can also present other characteristics. Consider for

example, the following lightly damped oscillator (section 6.3.1.1):
x H x x 0
The friction term H x is a perturbation of the undamped equation. The

modification of the oscillatory solution due to a damping term e H t 2 has an effect
of order H over a duration O(1), but the final result is the suppression of the
movement. The perturbation effects due to friction are cumulative.
More complete developments on singular perturbations can be found in

specialized textbooks ([COL 68], [HIN 91], [NAY 81], [VAN 75]).
6.5. Quasi-1D flows
6.5.1. General properties

6.5.1.1. Assumptions
Many flows can present a quasi-1D character, in other words the evolution scales
of phenomena differ depending on whether we follow the principal direction of the
flow or the transverse directions.
y G
v V
M u
d
z O x
Figure 6.15. Quasi-1D flows
The basic assumption of a viscous quasi-1D flow is that the cross-section and
direction of a stream tube vary quite slowly (Figure 6.15). The thickness d of the
domain of study is small compared to its length L (d << L). We consider the
coordinate on a curve oriented along Ox, more or less parallel to the stream tubes
(the axis of a pipe, jet or wall, etc.) and the orthogonal coordinates in cross-sections
orthogonal to Ox. The local coordinates defined by the curvilinear abscissa x and the
orthogonal coordinates y, z in cross-sections constitute a coordinate system which
is locally Cartesian. These assumptions allow us to write the balance equations in
Cartesian form in this zone.
We furthermore assume that the variations Gf of the quantities f studied are more
or less of the same order of magnitude over the distances d and L.
6.5.1.2. Approximations
The following properties result from the preceding assumptions:
1) Partial derivatives of a quantity in the directions y or z are an order of
magnitude greater than the corresponding axial derivative in the direction Ox. In
effect wf wx | Gf L and wf wy | Gf d lead to:
wf d wf w2 f d 2 w2 f
| and |
wx L wy wx 2 L2 wy 2
This results in vectors of the form k grad f (for example, the thermal or mass
G
flux densities q G , etc.) are perpendicular to the axis Ox and that we only take into
G
account the two components k wf wy and k wf wz of the flux density q G .
G
2) The diffusion term div q G of the balance equation of a scalar quantity G is
therefore reduced to the sum of the transverse derivatives w q Gy w y and
w q Gz w z .
3) The transverse components v and w of the velocity are small compared with
the axial component u ( v u and w u ). In effect, assuming the flow to be
incompressible, the volume balance can be written:
wu wv ww
0
wx wy wz
If variations of the cross-section exist, the transverse velocity is not zero, but it is
very small, in order to ensure the inclination of the velocity with respect to the axis
Ox, of order d/L. Assuming the 2D problem (w = 0) in order to simplify matters, the
two derivatives wu wx and wv wy have the same value (sign excepted). An order of
magnitude analysis shows that:
wu U wv v
| | |
wx L wy d
i.e., sign excepted, v | Ud L .
No simplification is possible in the mass conservation equation.
The result of this evaluation of the velocity components is that the dominant
terms of the viscous stress tensor are the tangential components P wu wy and
P wu wz (friction forces in the Ox direction).
4) The material derivative of a scalar quantity conserves its general expression
in the quasi-1D approximation.
dg wg wg wg
Consider the material derivative u v of the quantity g. The
dt wt wx wy
two convection terms are of the same order:
wg Gg wg d Gg Gg
u |U v |U U
wx L wy L d L
5) On the other hand, an analogous order of magnitude calculation shows that the
small value of the transverse components of the velocity leads to the transverse
accelerations dv/dt and dw/dt being negligible with respect to the axial acceleration
du/dt.
6.5.1.3. Local balance equations

Taking account of the preceding approximations, balance equation [4.3] for the
volume quantity g can be written:
wq Gy
wg G
div Vg U
dg
VG
wq Gz
[6.74]
wt dt wy wz
By way of example, the dynamic equation following Ox [4.21], for a Newtonian

fluid, can be written:
ª wu wu wu wu º wp w § wu · w § wu ·
U« u v w » U fx ¨P ¸ ¨¨ P ¸¸ [6.75]
«¬ wt wx wy wz »¼ wx wy ¨© wy ¸¹ wz © wz ¹
In a plane 2D flow, we obtain for a fluid with constant physical properties

(where pg is the driving pressure):
du ª wu wu wu º wp g w 2u
U U« u v » P [6.76]
dt «¬ wt wx wy »¼ wx wy 2
Finally, as the transverse accelerations dv/dt and dw/dt are negligible with
respect to the axial acceleration du/dt, the viscous stress terms of the transverse
equations are equally negligible with respect to the corresponding term in the axial
dynamic equation. The transverse equations reduce to:
wp g wp g
0
wy wz
The pressure (or driving pressure) distribution in a cross-section is hydrostatic,

but it depends on x and t.
Energy equation [4.51] can thus be written in the quasi-1D approximation:
de wu i ½
U p °
dt wx i ° 2 2
§ wu · wu w § wT · w § wT ·
¸ P §¨¨ ·¸¸ V T
dh dp °
U ¾ P ¨¨ ¸
Ö ¸ ¨¨ O ¸¸ [6.77]
dt dt ° © wy ¹ © wz ¹ wy ¨© wy ¸¹ wz © wz ¹
ds °
UT °
dt ¿
Dissipation function [4.56] can be simplified by only considering the dominant

strain rates wu wy and wu wz . We find immediately:
2 2
§ wu · wu
) P ¨¨ ¸ P §¨¨ ·¸¸ [6.78]
¸
© wy ¹ © wz ¹
6.5.1.4. Parabolic character of the quasi-1D equations

Consider first convection equation [6.74] in which the velocity field is assumed
to be given. This presents a structure analogous to the heat equation, i.e. it has a
dissymmetry in the order of the derivations with respect to the variables: second
derivatives with respect to y and z are encountered, the derivatives with respect to x
and t being only first order. This results in a parabolic behavior with respect to the
variables x and t: the temperature distribution in a cross-section of coordinate x at
the instant t depends only on data from the past or the upstream region. This
explains the fact that the state of the matter in the cross-section depends on its
anterior state on the trajectory; the diffusion term of quantity G only provides a
limited action in this section, the suppression of the conduction term wq Gx wx
amounting to the suppression of all flux in the upstream direction.
The velocity components u and v are the unknowns of dynamic equation [6.76].
The demonstration of the parabolic character can be effected by introducing the
stream function \ such that the mass conservation is satisfied:
u w\ wy v w\ wx
Equation [6.76] can thus be written:
du ª w 2\ w\ w 2\ w\ w 2\ º wp g w 3\
U U« » P [6.79]
dt «¬ wtwy wy wxwy wx wy 2 »¼ wx wy 3
In equation [6.79] the derivations with respect to x and t are of order 1, whereas
the derivation with respect to y is of order 3, which indicates the parabolic character
with respect to the variables x and t.
In applications, the hypotheses of section 6.5.1.1 are very often encountered.

Quasi-1D flows can be produced:
when the geometric boundary conditions impose such an evolution: in flows in
pipes this kind of approximation exists for most macroscopic physical phenomena
(electric, electromagnetic, thermal, etc.);
when diffusion phenomena in flows lead to weak fluxes of extensive quantities
in the axial direction Ox compared with the convection fluxes of these. In inviscid
fluids transport or propagation phenomena governed by the characteristics are, in
fact, perturbed by contact actions (viscosity, thermal conduction, diffusion, etc.), and
this leads to a transverse migration of the extensive quantities. The balance
equations contain high-order derivatives which “perturb” the convective transport
terms. In these situations, there exist non-dimensional parameters (Reynolds, Peclet
numbers, etc.) which take on high values.
We will successively examine two categories of problem in which the data are
different.
6.5.2. Flows in pipes
6.5.2.1. Nature of the problem

A pipe is a stream tube 6 materialized by a wall. The dominant velocity
component u in a cross-section is directed along its axis. The definition of the
problem to solve can be obtained as usual by combining the local balance equations
(mass, axial momentum, energy, etc.) with the initial and usual boundary conditions
on the wall 6L.
u(x,t) Gx dA x
GD
G6 L
S(x)
C 6L
Figure 6.16. Balance in a pipe
The geometric elements of the pipe are given, such that integration of the
dynamic equation leads to a relation between the volume (or mass) flow and a
pressure difference between two cross-sections. Calculation of the velocity
distribution (or of a quantity G) in a cross-section is an internal problem posed in
the interior domain of the stream tube 6 (Figure 6.16).
6.5.2.2. Global balance equation for an extensive quantity in a pipe

The normal of a cross-section S(x) is here oriented parallel to Ox (orientation by
continuity). This convention requires a change in sign at the time of application of
Osstrogradski’s theorem to a closed surface 6 containing cross-sections. By
assumption, the lateral wall 6L is supposed impermeable to the flow, i.e.
G
³6 gV .n ds 0 . Consider the small domain G D of the pipe comprised between the
L
sections of abscissas x and x + Gx of the lateral surface G6L (Figure 6.16).
Formula [4.62] for the global balance of a volume quantity g can be written:
wg
³Dq dv ³6 gu i ni ds ³D V G dv ³¦ q Gj n j ds [6.80]
wt
Taking account of the preceding assumptions, we have:
x Gx
³6 gu i ni ds ³S guds x ; ³D V G dv G x.³S V G ds
G
The density flux q G is essentially normal to the lateral wall (section 6.5.1.2) for
irreversible changes (viscosity, conduction, diffusion) and the integral ³6 q Gj n j ds is
equal to G x.³C q Gw dA taken over the contour C of the cross-section (qGw is the flux
density of G, normal to the wall).
Substituting these expressions into [6.80] and dividing by Gx gives:
³S
wg
wt
ds
w
wx
³S ugds ³S V G ds ³C q Gw dA [6.81]
In the place of the volume quantity g U g , let us take the massive quantity g;
we obtain:
w U g
³S
wt
ds
w
wx
³S U ugds ³S V G ds ³C q Gw dA [6.82]
The quantity M GS ³S U ugds ³S ugds is the flow of quantity G across the

cross-section S.
6.5.2.3. Applications
Taking g 1 , we have the volume flow rate q vS ³S uds across S.
Taking g 1 , we have the equation for the mass balance:
wU wq m
³S ds 0 [6.83]
wt wx
with: q m ³S U uds , the mass flow in the rate cross-section S.
The momentum balance along Ox can be obtained by applying formula [6.82] of

the quantity g = u. Supposing (section 6.5.1.3) that the driving pressure gradient
wp g wx (here, the term VG) is constant over the cross-section S, and designating by
Ww the viscous stress exerted by the fluid on the lateral wall, we have:
w U u
³S
wt
ds
w
wx
³ S
U u 2 ds S
wp g
wx
³C W w dA [6.84]
Proceeding as before, kinetic energy equation [4.66] can immediately be written,

by noting that the power of the viscous forces on the external surface of GD is zero
because of the assumptions which have been made (zero velocity on the wall, and
quasi-1D approximation on S), we obtain:
w §¨ U u 2 ·¸ w ª §¨ U u 2 ·¸ º
³S ds « ³ pg uds » PvS [6.85]
wt ¨© 2 ¸¹ wx «¬ S ¨© 2 ¸¹ »¼
with: PvS ³S )ds , the power dissipated by viscosity per unit length of the pipe.
The first term of [6.85] gives the accumulation of kinetic energy in D in

transitional regime and the second expresses the flow rate of mechanical energy
across S.
The different forms of the energy equation seen in section 4.3.4 can be integrated
over the surface S (or in the domain GD), but it is not possible to write down an
energy flow in the cross-section, except if we use initial formula [4.68] of the
balance equation which contains the flow of total enthalpy
*ht = ¨ [h (V 2 / 2)]dqm across the section S.
s

w W ij u i
Furthermore, the integral over GD of the term can be transformed into a
wx j
surface integral, which is zero, as we have already said.
Proceeding as before for the thermal flux density, we obtain:
w ª §¨ u 2 ·º w
³S « U e » ds Mht ³S U f x uds ³S V T ds ³C qTwdA [6.86]
wt «¬ ¨© 2 ¹»¼ wx
Recall that the power of an external force field is often negligible for a perfect
gas. The expressions for the internal specific energy e and the specific enthalpy h for
a perfect gas can be written:
Cv p Cp J p
Ue U CvT p ; Uh U C pT p
r J 1 r J 1
The energy equation for an incompressible fluid of constant specific heat C

(section 4.3.4.1.6 and equation [4.70]) can be written:
w U CT
³S
wt
ds
w
wx
³S U uCTds PQS ³S V T ds ³C qTw dA
[6.87]
where, depending on the case, we used either Cp or Cv for the specific heat C.
6.5.2.4. Average values of intensive quantities

As we have already said in section 1.4.2.5, to be consistent, the definition of
mean intensive values is effected such that the balance of the corresponding
extensive quantities is verified for the system studied. The application of this general
principle is expressed here by writing that values of the fluxes of extensive
quantities (mass, momentum, energy, etc.) in the cross-section S are identical either
by integration of local values or by using these mean values for balances. Then, we
take the following definitions:
1
the mean density Um: Um ³ U ds
S S
1
the mean velocity uq: uq ³ U u ds
UmS S
1
the mean temperature Tm (C = const): Tm ³ U uT ds
U muq S S
and in general the mean quantity gm: g m 1

³ U ug ds
Umuq S S
These quantities are often called the average mixing values (or mean mixing
values), as they correspond to the intensive value represented by the variable g under
the assumption that the flow of G across the section S would directly fill a volume
where it should be mixed without any external input. The preceding definition of Tm
supposes that the specific heat is independent of the temperature.
In cases where the same quantity contributes differently to several mean values,
we introduce a suitable coefficient, for example:
1
the momentum coefficient E: E u q2 2
³ U u ds
UmS S
3 1 3
the kinetic energy coefficient D: D u m ³ U u ds
UmS S
These mean values and the preceding coefficients allow the equations for the
quasi-1D model of flow in a pipe to be written very simply in a form analogous to
the 1D slice approximation with uniform properties in the cross-section (see section
4.3.2.3.4). However, the system of differential equations obtained only determines a
solution if the preceding coefficients constitute data, which must be chosen more or
less empirically from assumptions derived from the velocity, temperature or
concentration profiles.
We will leave it to the reader to verify that in a laminar flow we have the
following values:
uniform flow: D = E = 1;
Poiseuille flow in a circular tube: D =2; E = 4/3.
In industrial pipe systems, the values of D and E are often of the order of 1.1 to
1.3 ([ASH 89], [IDE 99]). If the differences between the local velocity u and the
mean velocity uq are small, the reader can verify that we have approximately6
E 1 3D 1 .
The mechanical energy balance [6.85] (generalized Bernoulli’s theorem) in a

pipe for an incompressible fluid can be written with the definition of D:
w §¨ U uq ·¸ w ª§¨ U uq2 ·¸ º
2
E « pg D qv » PvS [6.88]
wt ¨© 2 ¸¹ wx «¨© 2 ¸¹ »
¬ ¼
We can thus define the total mean driving pressure:
U u q2
p tm p U gh D .
2
The quantity q v p tm represents the flow rate of mechanical energy across a

cross-section S. Equation [6.88] becomes, on neglecting the power of the viscous
forces on 6 (approximation 2 of section 6.5.1.2):
6 Let u u q u ' and neglect the term in u ' 3 in the calculation of E
w §¨ Uuq ·¸
2
wp
E qv tm PvS
¨
wt © 2 ¹¸ wx
Integrating along the axis in the domain D included between two cross-sections
S1 and S2 gives:
w §¨ U uq ·¸
2
³D E dv qv ptm 2 ptm1 PvD [6.89]
wt ¨© 2 ¸¹
This equation is a model which reveals the inflows and outflows for the studied
domain of the pipe; it is particularly useful in steady flows for evaluating the
mechanical energy dissipated, using measurements of velocity distributions in the
sections S1 and S2 [IDE 99].
6.5.2.5. Local equations

The local equations in the cross-section S of the stream tube are identical to the
local equations of the boundary layer which are developed in the following section.
6.5.3. The boundary layer in steady flow

We will limit our discussion in this section to the case of a steady flow of an
incompressible fluid of constant viscosity. The Navier-Stokes equations ([4.74] and
[4.75]) can be written with non-dimensional variables (section 4.6.1.3), ~ p being
here the non-dimensional driving pressure:
wu~ w~
p 1 ~~ wu~ j
u~ j ~ i ~ 'u i ; 0; i, j 1,2,3 [6.90]
wx j wx i Re w~x j
Under the usual conditions, the Reynolds number is very large compared to 1,
~
and the term 'u~i is weighted by a coefficient which is very small compared to 1.
This therefore appears to be a perturbation quantity whose nature we will study
using the procedure outlined in section 6.4.
6.5.3.2. External solutions and the Euler equations

Assuming that the solution of the preceding equations and its derivatives vary at
the scale of 1, all the non-dimensional derivatives are in the order of 1, and the term
1 ~~
'u i is therefore very small compared to 1.
Re
The dynamic equations can be reduced to the Euler equations:
wu~ w~
p
u~ j ~ i ~ (i, j 1,2,3)
wx j wx i
These are one order less, and require weaker boundary conditions than the
Navier-Stokes equations. It is clear from a physical point of view that we must
abandon the adherence condition, since the viscosity no longer exists, and the fluid
can therefore slide over the walls. We thus find ourselves in the singular
perturbation situation described in section 6.4.3.
6.5.3.3. Finding a singular perturbation zone

Following the preceding reasoning, this zone cannot concern a zone of scale 1 in
all three dimensions. At least one of the dimensions of this zone must be small in
order for the value of a derivative to be sufficiently large to compensate the
coefficient 1/Re. Where can such a zone be found? We note firstly that on account of
the transport of fluid and its properties, it is difficult for such a zone to
spontaneously appear in the heart of the flow. An exterior intervention is then
necessary in order to create a viscous phenomenon sufficiently large which then
develops. This can only happen when the flow of an inviscid fluid encounters an
obstacle on the singular streamline which contains the stagnation point A of zero
velocity (Figure 6.17).
The subsequent velocity evolution on the wall streamlines of the inviscid fluid
lead to a non-zero sliding velocity which increases downstream of the point A. It is
then in the neighborhood of the wall that the viscosity must necessarily act.
The length of this zone is in the order of obstacle dimension L and its thickness
G is necessarily o(L), otherwise we are back in the preceding situation. We must
therefore study a thin zone in the vicinity of the walls where we can make the
approximation of a quasi-1D flow. We will here consider a plane 2D flow over an
obstacle placed in a flow of uniform velocity U (Figure 6.17), and we will allow the
radii of curvature of the walls to be large compared with the thickness G.
6.5.3.4. Boundary layer equations

Let us consider an obstacle inside a uniform flow of an inviscid fluid at speed U.
Let us take a locally Cartesian coordinate system (x, y) defined in the following way
(Figure 6.17): x is the curvilinear abscissa evaluated algebraically on the wall
downstream of the stagnation point A on the obstacle, the coordinate y being
G
evaluated along the wall normal n . The velocity components are designated by (u,v)
in the coordinate system fixed to the wall and in its immediate vicinity.
U y y
boundary layer ue(x)

L n
flow of
inviscid fluid G x G
y x
A
(a) L (b)
Figure 6.17. Boundary layer on the wall of an obstacle in uniform flow:

(a) figure on the scale of L; (b) figure on the scale of G
The discussion is as per section 6.5.1.3. The longitudinal velocity and

acceleration components are nearly parallel to the wall; the normal acceleration
component is negligible compared to the longitudinal component: the pressure,
constant across the thickness of the boundary layer is here only a function of the
abscissa x (section 6.5.1.3). However, in the dynamic equation, it is not possible to
neglect the v component in the material derivative which ensures a part of the
momentum transport (section 6.5.1.2). The dimensional equations of the 2D
boundary layer can thus be written:
§ wu wu · dp w 2u wu wv
U ¨¨ u v ¸ P ; 0 [6.91]
© wx wy ¸¹ dx wy 2 wx wy
The order of magnitude G of the boundary layer thickness can be obtained by

writing that the material derivative and the viscous stress term are of the same order
of magnitude (section 6.5.1.3):
§ wu wu · U U 2 w 2u P U
U ¨¨ u v ¸| | P |
© wx wy ¸¹ L wy 2 G2
where G | P L U U , or by defining the Reynolds number Re L U UL P with

the length L:
1 2
G L | P U UL Re L [6.92]
Introducing the stream function \ ( u w\ wy ; v w\ wx ), equations [6.91]

can be reduced to the equation:
§ w\ w 2\ w\ w 2\ · dp w 3\
U¨ . . ¸ P [6.93]
¨ wy wxwy wx wy 2 ¸ dx wy 3
© ¹
The order of the derivatives with respect to x is less than the order of the
derivatives with respect to y in equation [6.93] which is parabolic: the distribution of
the velocity at a given abscissa x0 only depends on the upstream conditions of the
external velocity ue(x), corresponding to values of x less than x0.
6.5.3.5. Boundary conditions

The boundary layer equations are clearly simpler than the Navier-Stokes
equations. We have already seen that the suppression of the transverse dynamic
equation leads to pressure being a function of the x direction only. We must now
express the adherence condition of the fluid at the wall, as this was our objective in
the introduction to the boundary layer.
y 0 u x ,0 vx,0 0
We must now match the boundary layer and the external inviscid fluid flow. We
proceed in a first approximation as per section 6.4.3.3 by writing that the velocity at
the outer edge of the boundary layer is equal to the velocity u e x of the inviscid
fluid on the wall in the absence of a boundary layer:
y
of u x, y o u e x [6.94]
G
The velocity u e x and pe (x) satisfy Bernoulli’s theorem on the wall for an
inviscid fluid:
u 2 x U2
p e x U e pf U [6.95]
2 2
where pf designates the static pressure in the external uniform flow of velocity U.
Finally, we must fix the initial upstream condition for the velocity distribution at
a given abscissa often taken as the origin x = 0. As we have explained in section
6.5.3.3, the stagnation point A of the flow of a inviscid fluid over the obstacle
(Figure 6.17) is the departure point of the boundary layer that can be easily
calculated in this zone where its thickness is quite small ([SCH 99], [YIH 77]).
If the leading edge of the obstacle is of negligible thickness (plane or wedge-

shaped wall), the thickness of the boundary layer is here taken to be zero.
6.5.3.6. General properties of boundary layers

6.5.3.6.1. Physical interpretation of the boundary layer
We can immediately note that the condition of zero velocity on the wall also
leads to a condition of zero acceleration. Conversely, the viscous stress becomes
zero at the outer edge of the boundary layer. This is therefore a zone where the
pressure gradient, which is constant, sees its action balanced by the acceleration in
the outer region and by the viscous friction at the wall. This situation is shown
schematically in Table 6.1 which indicates the dominant terms following the height
in the boundary layer.
The solution previously obtained in section 5.4.5.4 for the heat equation is of the
same kind as that of the boundary layer, with time replaced by the x coordinate. The
zero-velocity condition imposed on the wall leads to viscous diffusion of the viscous
stress and of the vorticity. This results in an augmentation of the boundary layer
thickness as an abscissa function, as indicated by the order of magnitude
1 2
G L | P U UL Re L found earlier [6.92].
height in the acceleration pressure viscous stress

boundary layer du dp w 2u
quantity U gradient P
dt dx wy 2
outside (inviscid X X 0
fluid)
inside boundary x X x
layer
wall 0 X X
Table 6.1. Balance of “forces” in the boundary layer

This diffusion can be counterbalanced by an aspiration of fluid across the wall.

Consider the simple case of a plane wall placed in a uniform flow of velocity ue and
realize a suction of fluid across the wall at constant velocity v0 (Figure 6.18). We
see immediately that the solution:
§ v ·
u u e exp¨¨ 0 y ¸¸ [6.96]
© Q ¹
satisfies equations [6.91] of the boundary layer and corresponds to an established

solution where the thickness G Q v 0 is constant. Figure 6.18 represents this flow.
y ue ue
v0
G | Q v0
x
v0
Figure 6.18. Established boundary layer with suction

through a wall in an uniform flow
We note that in the case where we have blowing rather than suction ( v0
positive), the solution is not acceptable: the boundary layer no longer exists if the
blowing is sufficiently strong ([SCH 99], [YIH 77]).
6.5.3.6.2. Non-dimensional equations

Let us write equations [6.91] in non-dimensional form with the following change
of variables:
u~ uU v~ vRe1 2 U ~
x x/L ~
y yRe1 2 L u~e u e x L U
This gives:
wu~ wu~ du~ w 2 u~ wu~ wv~

u~ ~ v~ ~ u~e ~e ; 0 [6.97]
wx wy dx w~
y2 w~x w~y
Equations [6.97] do not contain the parameter Re U UL P . This results in the non-
dimensional solution being independent of the Reynolds number. So, all bodies which
have the same form have identical velocity distributions u~e ~
x at the wall; they therefore
have the same boundary layer development. We note however that the Reynolds number
comes into play as soon as we recast the problem using dimensional data for the
coordinate y and notably for the thickness of the boundary layer (formula [6.92]).
6.5.3.7. Separation of steady flows

The preceding discussion shows that, in the external part of the boundary layer,
the flow is close to the flow of an inviscid fluid where the pressure gradient is
determined by Bernoulli’s theorem: the modulus of the velocity decreases if the
pressure increases.
Close to the wall, the weak kinetic energy no longer plays an appreciable role
and the variations of the viscous stresses are opposed to the pressure gradient in the
boundary layer equations. The flow which is nearly purely viscous close to the wall
results from two causes:
1) The pressure gradient leads to a tendency of the flow towards decaying
pressures as in a Poiseuille flow (section 3.4.2.5). The velocity curvature profile at
the wall is equal, according to [6.91] and [6.95] to:
w 2u 1 dp e U du
ue e
wy 2 P dx Q dx
y 0
2) The external velocity u e x imposed creates a viscous entrainment in the

boundary layer in a manner analogous to the corresponding phenomena in a Couette
flow (section 3.4.2.3).
When the flow external to the boundary layer u e x increases with the abscissa
x, the longitudinal pressure gradient is negative and the velocity curvature profile
has a constant sign (Figure 6.19a). In the opposite case (Figure 6.19b), the sign of
the curvature of the profile changes, and this can lead, close to the wall, to a flow
which is reversed with respect to the external flow u e x .
ue(x) increasing ue(x) decreasing
x x
(a) dp/dx < 0 (b) dp/dx > 0
Figure 6.19. Velocity profile near the wall with a

(a) negative or (b) positive pressure gradient
The appearance of a flow from downstream to upstream is in contradiction with

the parabolic properties of the boundary layer equations: the velocity distribution in
the zone of reversed flow no longer depends only on the upstream conditions the
external velocity ue(x) considered, but also on the downstream conditions. The
boundary layer equations are thus no longer applicable in such a zone.
wake
S
R rotational
(a) (b)
Figure 6.20. (a) Separation with reattachment in the boundary layer (profiled obstacle);
(b) separation and creation of an open wake (unprofiled obstacle)
There therefore exists a back stagnation point S in a viscous flow where the
friction stresses on the wall are zero (Figure 6.20a). In fact, the flow, which is
reversed with respect to the main flow, comes from a rotational zone which can be
manifested in two forms:
– either a relatively thin rotational zone is generated within the boundary layer,
but reattaches at point R (Figure 6.20a); following what was said above, this can
only happen if the effect of the pressure gradient is sufficiently weak for the viscous
entrainment to constitute the dominant effect. This is the case for profiled obstacles
whose reducing section in the downstream direction is very gradual;
– or a completely open wake can be generated downstream of an obstacle which
is not profiled (Figure 6.20b).
Note that the position of the separation point is independent of the Reynolds
number, provided the velocity distribution of the inviscid fluid at the wall does not
depend on the Reynolds number either. This is the case for the irrotational inviscid
fluid around a profiled obstacle which is not modified (to second order) and
obviously does not depend on the viscosity. For the unprofiled obstacle, the real
velocity field is different from the field calculated in an inviscid fluid on account of
the presence of a rotational wake which is fed from downstream; however, if we
assume that the structure of this wake is independent of the Reynolds number, we
see that it is consistent to assume that the position of the separation point is also
independent of this parameter: this is confirmed by experiments, provided the

boundary layer is not turbulent ([SCH 99], [YIH 77]).
6.6. Unsteady flows and steady flows
6.6.1. Introduction
The temporal evolution of the properties of matter is fundamentally based on the

balance laws of the associated extensive quantities. We have already discussed in
Chapter 2 the difficulties of representing the continuous medium which we
encounter depending on whether we choose to use a Lagrangian (substantial)
description of the fluid particles or a Eulerian (spatial) representation of the flow.
We must now return to the fundamental difficulties which arise when we use
Eulerian variables.
The fields to which matter is subjected are furthermore always due to actions at a
distance performed by other material elements: a gravitational field is caused by the
presence of mass, an electric field results from the presence of charges, an
electromagnetic field is due to electric charges in movement at either the
macroscopic or the microscopic scale. A field is described by functions of space-
time variables in a reference frame (known as the laboratory reference frame)
associated with a flow device or an object moving with respect to a fluid (vehicle,
plane, etc.). There are numerous situations for this observer in which the velocity
fields and the material quantities are not functions of time, but only of space. The
corresponding phenomena are therefore steady. This terminology only has meaning
in reference to this privileged reference frame, the quantities attached to the
material particles being always functions of time (Lagrangian representation).
However, these steady phenomena, when they exist, always arise as a result of
the evolution of a transitional regime. Thus, in many situations, the transitional
regimes do not lead to steady flows and we observe complex phenomena which we
will describe very briefly here.
In order to simplify the discussion, we will consider in what follows an inviscid

or Newtonian fluid of constant density, unless otherwise stated. The variations of the
physical properties, if they are not too great, do not significantly modify the
structure of the phenomena which we will discuss.
We will leave aside questions related to the existence and to the uniqueness of
solutions of the Navier-Stokes equations, the understanding of which requires a
more advanced course in mathematical analysis. In this domain many questions
remain open, and the physical aspects of the phenomena which we describe in this
textbook demonstrate the formidable complexity of such an eventual theory.
6.6.2. The existence of steady flows
For a flow and the associated transfers to be steady, it is necessary for the
boundary conditions describing the corresponding problem to be steady; in
particular, actions on the flow by fixed elements in this reference frame should be
independent of time. For example, the flow between upstream and downstream
infinite reservoirs at constant pressure and connected by a nozzle (section 5.5.4) can
be independent of time if the system is described in the reference frame of the
nozzle. However, before being observed at a given flow rate, the flow was created
from a zero pressure difference and it followed an evolution through the following
states: a subsonic regime with increasing velocity, then a sonic regime and finally a
supersonic regime with the progressive appearance of a shock wave which descends
in the divergent part of the pipe until the pressure gradient is stabilized.
The solution of the problem defined by the steady boundary conditions is not
always unique. The nature of the boundary conditions to be used is often a source of
considerable difficulty (section 5.6.3). Consider for example the flow between two
cross-sections S1 and S2 (S1 > S2) in a divergent pipe (Figure 6.21). Under the
assumption of an inviscid incompressible fluid and with the approximation of the
flow by slices, Bernoulli’s theorem can be written:
U12 U 22
p1 U p2 U
2 2
The flow rate q v U VS in this section of the pipe can be immediately

obtained:
p1 p2 1 1
qv r with: B 2.
BU
2
2S2 2S1
There exist two opposite flow rate values for this steady flow. We will see later
that only one of these is really acceptable in the context of the preceding
assumptions.
Let us now perform an experiment with a plane pipe whose divergent has a
sufficiently large angle (of the order to 10 degrees or so). Here we observe that the
steady flow follows one of the walls while it separates from the other. The flow
chooses the wall to follow as a result of particular circumstances of the transitional
regime. For example, in the configuration in Figure 6.21a, we have included in the
walls two suction orifices A1 and A2 which are not activated in the steady regime
and which we can use in order to create a small additional transitional depression in
order to “suck” the flow towards a chosen wall. This suction does not need to be
constant: once the flow has attached to one of the walls, it will remain so after the
suction has been stopped. Such flows with two stable positions can be used to
control flows in fluid circuits (fluidics command and control).
p2 suction orifice A 2 p1
S2 suction orifice A1 S1
(a) (b)
Figure 6.21. Steady flows with two configurations
For the two preceding flows, the pressure is nearly constant in each of the
sections S1 and S2. Writing the boundary conditions for the steady flow of a viscous
fluid leads to considerable difficulties, because there no longer exists a velocity
potential and we know that we cannot independently specify the entry and exit
conditions of a flow since the transport properties on the trajectories are dominant
(section 5.6).
Figure 6.21b shows another configuration where two flow structures in a pipe are
possible depending on whether the flow reattaches or not on an obstacle placed on
its wall. The existence of reattachment is also related to the transitional regime
which leads to the fully established flow. Examples of the same kind exist for sheets
of water over spillways, which may flow above cavities either ventilated (i.e. filled
with air) or not ([CHA 04] p. 399, [JAI 01] p. 264).
The preceding problems are examples of systems presenting hysteresis (the state
of a system depends on its history). From a physical point of view, we can note that
we have here a “retroaction” in the upstream direction which leads to the existence
of a memory for the flow.
In summary, a flow problem with steady boundary conditions does not

necessarily have a unique steady solution. We could also say that a steady problem
is not necessarily a well-posed problem.
6.6.3. Transitional regime and permanent solution
6.6.3.1. Relation between pressure and flow rate in a fixed stream tube
Depending on the imposed conditions, pressure can be the cause or the
consequence of movement of a fluid. Aside from cases where the viscosity plays a
dominant role (low Reynolds number flows), the pressure (or the driving pressure
depending on the case) balances the acceleration, and for steady flow of an inviscid
fluid, Bernoulli’s first theorem treats the pressure (or the total enthalpy) as a
component of the total mechanical energy which is conserved for a fluid particle.
This local property no longer exists in unsteady flows.
Consider the flow of an inviscid fluid of constant density whose fluid trajectories
are fixed; they are thus coincident with the streamlines (and the emission lines). A
stream tube is therefore a surface on which the inflow and outflow of fluid occurs
over the cross-sections S1 and S2 with velocities V1 and V2 (Figure 6.22). The
assumption of incompressibility leads to a volume flow q v t SV circulating in
the stream tube which is independent of the cross-section used to evaluate it; we will
treat it thus as a variable. Let us apply kinetic energy theorem [6.85] to the interior
domain D of the stream-tube element limited by the surfaces SL, S1 and S2 (Figure
6.22); we have:
w §¨ V 2 · §
¸dv ¨ p U V
2 · §
¸V .nds ¨ p U V
2 ·
¸V .nds
³D ¨ U ¸ ³S 2 ¨ ¸ ³S1 ¨ ¸
0
wt © 2 ¹ © 2 ¹ © 2 ¹
where p designates the driving pressure by way of simplification.

qv qv
S1 B>0
G S1
S2 V2 G S2
G SL SL
V1
V1 A
L A L
p1 G
S S V2
p2 p2
p1
(a) (b)
Figure 6.22. Flow by slices inside a stream tube:

(a) convergent tube; (b) divergent tube
In the slice approximation of the flow, the quantities are uniform in all cross-
sections, and after simplification by qv, we immediately obtain:
dq v L dA § U 22 U 12 ·
U U ¨ ¸ p p
³ ¨ 2 ¸ 2 1 0
dt 0 S © 2 ¹
dA 1 §¨ 1 1 ·
L
¸
Let: A ³0 and B
¸
US 2 U ¨© S 2
2
S12 ¹
The relation between the flow rate qv and the pressure difference p1 p2 can be
written:
dq v
A B q v2 p1 p 2 [6.98]
dt
This equation shows that the pressure is balanced by two acceleration terms, one
dq
unsteady, A v , and the other, B q v2 , corresponding to the steady flow,
dt
proportional to the square of the flow rate and independent of the direction of the
flow. This separation is related to the Eulerian description of the phenomena, and
from a physical point of view there are not two kinds of acceleration for the fluid
particles. As we have already said (section 3.3.3.5), this description is always
associated with a favored reference frame resulting from the existence of boundary
conditions on the particular surfaces which are here the sides of the stream tube.
We should note that the coefficient A of equation [6.98] has an order of

magnitude proportional to the length, contrary to the coefficient B which only
depends on the values of the inflow and outflow sections. Furthermore, the 1/S
dependence of the integral A shows that a severe intermediate narrowing will not
modify the values S1 and S2 but will lead to a considerable increase of the coefficient
A, in other words to the pressure difference necessary for the transitional
acceleration in the Eulerian representation. The reason for this is the existence of a
strong acceleration in sections of small dimension where the velocity takes on high
values in order to conserve the flow rate.
6.6.3.2. Properties of the solutions

Let us take B to be positive ( S1 ! S 2 ), which does not restrict the generality of
the reasoning. Equation [6.98] can only therefore possess steady solutions if the
difference p1 p2 is positive. In this case, we have the two steady solutions
qv r p1 p 2 B , corresponding to an established flow, either in the positive

direction, or in the negative direction of the axis A (section 6.6.2). However, the
possibility of realizing such a solution depends on the existence of a transitional
regime which can lead to this kind of situation.
In the unsteady regime, the pressure difference p1 p2 can be negative. However,

an examination of equation [6.98] shows that if it is always negative, the same goes
dq v
for the derivative ; we can thus see that the flow rate qv decays indefinitely,
dt
which is physically unacceptable.
Suppose now that the difference p1 p2 is positive and let

Qt p1 p 2 B ; Q(t) is the positive value of the flow rate under the
assumption that the unsteady term is negligible. Equation [6.98] can be written:
A
dq v
dt

B Q 2t q v2 [6.99]
We see that:
dq v
– if the instantaneous flow rate qv is greater than Q(t), the derivative is
dt
negative: the flow rate qv decays and approaches the value Q(t);
– if qv lies between – Q and + Q, it therefore increases, and approaches the value

Q(t) once again;
– if the instantaneous flow rate qv is less than Q(t), the derivative dq v dt is
negative: the flow rate qv decays and moves away from the value Q(t).
This property leads to the flow rate being bounded if the positive quantity Q is
bounded. If Q(t) tends to a limit for infinite t, the same goes for the instantaneous
flow rate qv if the latter remains always greater than – Q. Under the assumption of
constant pressure difference p1 p2, we see that the transitional regime does not
allow the solution q v Qt p1 p 2 B to be attained.
The inviscid fluid model does not apply for negative flow rates, and the viscosity
must be introduced. The study of viscous fluid flow between two infinite divergent
planes can be performed (the Jeffery-Hamel exact solution of the Navier-Stokes
equations). We find that a separation occurs for a Reynolds number which tends to
infinity, even if the angle of the planes tends to zero ([SCH 99], p. 104). In these
conditions, the uniqueness of the solution is not assured.
6.6.3.3. Comparison with experiments and consequences

In practice, flow is possible in a divergent channel, but it is not possible to model
this using an inviscid fluid, even for very large Reynolds number values. A section
of convergent pipe transforms the mechanical power q v p1 p 2 provided by the
3
pressure difference into an increase in the kinetic energy flow D U Su m 2 between
the inflow and outflow sections (section 6.5.2.4). A divergent section leads to the
opposite transformation: kinetic energy is partially recovered in the pressure.
We define the efficiency of these transformations using the kinetic energy flow
3
D U Su m 2 between the inflow and outflow sections (section 6.5.2.4) and the
mechanical power q v p1 p 2 provided or recovered by the pressure difference.
Whereas a pipe with a convergent section in the flow direction transforms pressure
into kinetic energy with good efficiency, the inverse effect in a divergent pipe occurs
with a non-negligible dissipation. For a given ratio S1 / S2 of the sections S1 and S2,
the efficiency is close to 1 in a converging flow for a short length A whereas in a
divergent pipe, the efficiency goes through a maximum of the order of 0.7-0.8 for an
angle of about 7º. A compromise must be found between a small angle and a large
length over which there exists a notable viscous dissipation, and a larger angle
leading to the formation of a separated flow at the wall, whose kinetic energy is
nearly entirely lost.
The approximation of the flow by slices (uniform velocity distribution in the

stream tube constituted by the pipe) is always a good approximation for steady flow
in a convergent pipe. On the other hand, this approximation is a poor model in a
divergent element, because numerous phenomena can occur which contradict the
assumption of uniform velocity and lead to the loss of mechanical energy (tendency
for the flow to separate close to the wall in the boundary layer (section 6.6.3.7),
instabilities close to the wall leading to the generation of turbulence, etc.).
The experiment always amounts to an observable process, and thus to a well-

posed problem provided a suitable analysis is performed. As we have just seen, the
same is not true for theoretical models for which the initial conditions must be
ascertained by experiment. In particular, methods of numerical solution of the
Navier-Stokes equations often relies on the computation of a velocity field from a
pressure field, which allows, at the next iteration, the computation of a new pressure
field, hence a velocity field and so on. These intermediate fields are not solutions of
the equations of motion, and the intermediate pressure fields can thus correspond to
initial conditions which lead to a rapid divergence of the velocity field at the
following approximation. This is an important and unfortunately common difficulty.
6.6.3.4. Separation in unsteady flow

We have seen (section 6.5.3.7) that separation is due to the quadratic term in the
Bernoulli’s theorem for an inviscid fluid. The argument thus invoked is no longer
valid in a strongly unsteady flow. Consider an irrotational unsteady flow for which
we have Bernoulli’s second theorem ([6.13]).
When the dynamic pressure V 2 2 is smaller than the term wM wt , the

phenomenon of flow separation at the wall no longer occurs. If the flow is
oscillatory, the velocity profile may present changes of sign, but there no longer
generally exists a rear stagnation point S at the origin of a streamline towards the
heart of the fluid (Figure 6.20a), which is characteristic of a 2D separation.
6.6.4. Non-existence of a steady solution
Even if we impose steady conditions at the fluid domain boundaries, we do not

encounter steady flows in very many situations. The wakes observed behind
obstacles belong to this category. Let us reconsider the example of a circular
cylinder of diameter D (without circulation) in a velocity field which is assumed to
be uniform and steady at infinity. The structure of the flow depends on the Reynolds
number Re UD Q .
(a) (b)
(c) (d)
Figure 6.23. Real flow around a circular cylinder placed in a uniform steady velocity field
Figure 6.23 shows the evolution of this flow as the Reynolds number increases.
For values of the Reynolds number small compared to 1, the flow is symmetric
(Figure 6.23a); the dissymmetry between upstream and downstream increases with
the Reynolds number; for larger values (of the order of 0.2-20) we observe the
appearance of a fixed zone of reversed flow just downstream of the cylinder
(attached wake (Figure 6.23b)) comprising two symmetric vortices). For values
greater than 20 (Figure 6.23c), we observe the emission of alternate vortices at a
well-defined frequency. Beyond values of Re of the order of 1,000, the periodicity
of the vortices is attenuated (Figure 6.23c) and random fluctuations appear.
6.6.4.1. The creation of the circulation

We have studied the flow around a circular cylinder with circulation (section
6.2.5.2.4). The origin of this circulation cannot be explained by a rotation of the
cylinder since the fluid is inviscid. Experiment furthermore shows that an important
difference exists between the circulation calculated in the fluid and the circulation
calculated on the cylinder using the velocity at the wall. This difference shows that
the circulation is not created directly by viscous diffusion from the wall, which is
not significant in an inviscid fluid.
The circulation is in fact created during the transitional regime of the flow. It is
due to the viscosity of the fluid. Consider a fluid initially at rest about a profiled
body at incidence. In the initial acceleration regime of the flow around the body, a
dissymmetry appears between the intrados (the under-side) and the extrados (the
upper-side) of the profile which generates at the trailing edge (downstream of the
boundary layers) a sheared viscous flow at the origin of a vortex, which grows
(Figure 6.24a and Figure 6.24b) until it separates from the obstacle and is carried
away by the flow (Figure 6.24c). The viscosity is only important for a small zone
and Lagrange’s theorem can be applied in the inviscid fluid external to the ensemble
of the obstacle and the vortex. As the circulation is initially zero, it remains so: the
vortex carries away the circulation *, and the opposite circulation – * is established
about the obstacle.
U(t) U(t) U(t)

-* *
(a) (b) (c)
Figure 6.24. Creation of circulation by initial emission of a vortex around a profiled body
The circulation about a non-profiled obstacle is created by the same transitional

mechanism, to which is added a turbulent wake analogous to that described in
section 6.6.3.4, and which does not create and additional mean circulation.
(a) (b)
T1
(c) (d)
Figure 6.25. Creation of circulation by initial emission of a vortex

around a non-profiled body: (a) and (b) development of the initial vortex;
(c) and (d) the circulation is installed after the emission of the initial vortex T1
6.6.4.2. Instabilities and turbulence

The phenomena described in the preceding sections result from flow instabilities,
in other words to phenomena associated with the amplification of perturbations of
varying amplitudes. Situations of instability are numerous in fluid mechanics and the
viscosity is not necessarily a damping factor for these phenomena. On the other
hand, in so far as the most unstable structures correspond to high Reynolds numbers,
the characteristic time constant of the amplification is much lower than the damping
time.
Instabilities evolve in general towards a variety of more of less complex

situations; for high Reynolds numbers corresponding to the flows encountered in
practice, we see the appearance of random fluctuations the details of which are
unpredictable (chaos). The velocity fluctuations of fluid particles also lead to the
fluctuations of other extensive quantities, and the transfer mechanisms of
irreversible phenomena are significantly changed ([SCH 99], [YIH 77]).
Chapter 7
Measurement, Representation and

Analysis of Temporal Signals
The measurement of flows and transfer mechanisms presents specific difficulties

related to fluid movement. We will review the principles of the main measurement
methods. A synthesis of the different signal processing procedures commonly used
in the domains studied, and the various numerical techniques used in order to
implement these, will be presented.
7.1. Introduction and position of the problem
The modeling of macroscopic phenomena can be more or less detailed. The

representation of the continuous medium by means of fields of continuous quantities
was obtained by extending systems of finite dimensions in thermodynamic
equilibrium to their limits. Writing an infinite number of values of a continuous
field is obviously impossible and constitutes the infinite limit of a practical
realization; this can only be achieved if we dispose of a procedure for computing the
value at every point, which can be trivial and consist of the simple condition of a
constant value. Furthermore, discrete systems in thermodynamic equilibrium
constitute a more or less good approximation to a real system. The information
necessary for the description of a continuous system always therefore comprises an
ensemble, necessarily finite, of numerical values and a procedure for the
computation of quantities at all points, which amounts to an interpolation.
A material ensemble is described by means of quantities characterized by

measurements which permit the representation of its evolution. The equations are
exact equalities between real (or complex) numbers, which represent the physical
quantities studied. Now, these are never known exactly and this exact knowledge is
furthermore neither practically obtainable nor necessary. We work with
approximations or mathematical representations (transformed signals) which are
more convenient to use. The values themselves can be synthesized by a
transformation procedure allowing their number to be reduced. The representation
of physical quantities is thus a basic problem associated with our knowledge of the
phenomena of interest. These considerations can apply both to space and time
variant quantities. In what follows we will reason in terms of a temporal
representation of a signal, but the problems discussed are analogous regardless of
the nature of the representation and the number of variables.
Time is a particular variable, since the cause of a phenomenon observed at time

t0 can only be a function of the state variables from earlier times: only the past
influences the future. In reality, this property is also encountered for spatial
variables in flows, since the extensive properties remain attached to the matter in
movement. The notion of causality, largely developed in signal processing can in
fact by applied in space-time over characteristic curves or the succession of
characteristic surfaces. While causality appears obvious for the time-direction
considered in isolation, the problem is far less simple when we perform a Fourier
transform. We will not cover problems of causality which appear in integral
transformations: the reader should refer to the specialized literature ([AND 99],
[BEE 03], [DEB 06], [PRE 91]).
The idea of continuous phenomena has only really been used by some modern
calculation methods. The first analog calculators produced a physical representation
of the equations, the adjustment of the variables and parameters being achieved by
modification of physical quantities (for example, voltage for electronic calculators
or pressure for hydraulic and pneumatic systems). With numerical calculation, we
operate using numerical values, which requires equations to be discretized.
7.2. Measurement and experimental data in flows
7.2.1. Introduction
Knowledge of a physical system requires the measurement of its quantities. In

general, the local intensive quantities are easier to measure than the extensive
quantities which are most often measured indirectly using equations of state,
phenomenological laws or relationships obtained theoretically (for example,
Bernoulli’s theorem). As we cannot here provide detailed description of
measurement methods, we will simply indicate the principles on which these are
Measurement, Representation and Analysis of Temporal Signals 341
based. The quantities measured are primarily the pressure, the velocity, the
temperature and the concentration of a constituent.
7.2.2. Measurement of pressure
As pressure is a force normal to a surface, its direct measurement can be related

to the measurement of a force; as the force is the energetic intensive quantity
associated with a displacement in a fixed discrete system, we are dealing with a
measurement of displacement which is in fact achieved by the intermediary of
associate electrical or electromagnetic phenomena. Let us consider a few examples
of such measurement devices: piezoresistive probes where deformation of the
silicon membrane produces variations in electrical resistance, capacitive probes
(Figure 7.1b) which comprise a fixed electrode opposite the membrane, the
movement of the latter is accompanied by a variation of the capacitance, or
piezoelectric probes (Figure 7.1a) that comprise a piezoelectric material on which
the action of a force creates a difference of potential.
For steady flow, a measurement which is less and less frequently employed
consists of using liquid manometers, based on hydrostatics laws, to find the
difference in height 'h between the free surfaces in two tubes implying a pressure
difference equal to Ug'h, or between two free surfaces if the two tubes
communicate directly, or, between the pressure taps in the liquid. We measure here
a difference of level (a length), but the real difficulty comes from the accurate
identification of the position of the free surface and the effects of surface tension
(section 2.2.1.4.2).
f(t) f(t) p p+ 'p

'h
+ + + + + + + +
e(t) 'h
e(t) pg
pg +'pg
f(t) f(t)
(a) (b) (c) (d)
Figure 7.1. Probes for pressure (a), (b) and (c) and driving pressure (d) measurements
The pressure measured by a probe is the surface force on the sensitive member
of the probe. Pressure measurement at a point in a fluid requires in principle that the
probe is positioned such that it does not itself create a perturbation: it must therefore
be carried with the matter in movement. This is clearly unrealistic, unless some
optical means of pressure measurement could be made – and this is generally not
possible. In practice, it is only possible to use some instrument, which is integrated
into a solid surface. It is thus necessary to extract the pressure and transmit it to the
sensitive surface of the probe by means of a pneumatic or hydraulic link.
The extraction of pressure can be performed by means of an orifice at the tube

extremity used for its transmission. The tube extremity comprises an obstacle in a
flow, and this must be considered. We can assume that the pressure in an orifice
pierced in a wall is the pressure in the fluid at that point. We can in fact consider
that the flow is here parallel to the wall, without any velocity component normal to
the wall; the balance of transverse forces is zero on any fluid element, hence the
local uniformity of pressure along the wall normal. However, the pressure tube must
be contained in a suitable obstacle, such that it does not modify the flow and
therefore the pressure value. We will come back to this point a little later when we
deal with velocity measurements.
Furthermore, the transmission of pressure fluctuations by a tube is a delicate

problem when the length of the tube is more than a few centimeters, as it is
necessary to avoid problems of acoustic resonance [GAB 98]. However, this
problem is beyond the scope of this work.
7.2.3. Anemometric measurements
On a wall, the velocity of a fluid in motion is equal to that of the wall. It is
obviously only of interest to measure the velocity at points at the heart of the flow.
When making direct measurements of velocity, it is necessary to consider individual
fluid particles, of a given size, and to obtain an electrical or optical signal which
results from a phenomenon associated with the velocity of this fluid particle.
Velocity measurement devices (anemometers) which allow direct measurement are
mainly laser Doppler anemometers, cup anemometers, etc.). Indirect measurement
of velocity can be obtained by means of pressure measurements by application of
the laws of fluid mechanics (Pitot tubes, total pressure probes, etc.) or from
calibrated thermal measurements (hotwire anemometers). The same is true for
measurements of the flow rate, which are not generally made by directly measuring
the volume or mass which flowed during some given time.
As for the pressure, it is necessary to measure the local property of a material

element (Lagrangian measurement) or to introduce a probe which is sensitive to the
local velocity but which will necessarily disturb the flow (Eulerian measurement).
7.2.3.2. Lagrangian anemometry

In general we want to know the velocity at an observation point which we define
with respect to the laboratory reference frame (Euler variables). Fluid particles are
not individually visible and it is necessary in all cases to introduce solid elements,
which are entrained by the fluid. The problem of solid particle entrainment by a
fluid is therefore essential.
In the case of cup anemometers (Figure 7.2a), we use the dissymmetry of a

rotating device, which presents a resistance to the fluid movement in a given
direction (cup C1) and very low resistance in the other direction (cup C2). The result
is that the cup situated at position C1 tends to follow the flow, whereas the other, C2
induces a very low torque on the moveable system. Alternatively a propeller can be
used (Figure 7.2b). As the elements of these devices are not entrained by the fluid
with the fluid velocity, it is necessary to calibrate them.
C1
V V
C2
(a) (b)
Figure 7.2. (a) Cup anemometer or (b) propeller anemometer
Laser Doppler Anemometry (LDA) involves the creation of a system of

interference fringes by means of a laser whose beam is split into two beams focused
by a lens (Figure 7.3a). Very tiny particles crossing the fringes are alternatively
illuminated as they cross the fringes. It is possible to measure the scattered light by
means of a photomultiplier. The frequency value n of the light signal measured
allows a calculation of the particle velocity, equal to i.n, where i is the inter-fringe
spacing, easily obtained from the optical characteristics of the device. The
denomination Doppler comes from the fact that the procedure can also be explained
by considering the variation of the optical frequency which occurs when the particle
scatters the light provided by each of the coherent light beams.
L S
f
Z
V
(a) (b)
Figure 7.3. (a) Principle of anemometric measurement by ALD (L, laser; S, beam-splitter and
focusing lens; Z, fringes in the measurement zone; P, photodetector centered on the zone Z);
(b) bias in ALD measurement in a non-homogenous flow seeding
The principle of anemometric measurement assumes the presence of small

particles, which are in suspension in the fluid [DUR 81]. These nearly always exist
in water, but it is necessary to “seed” gaseous flows by means of droplets obtained
by the pulverization of some suitable oil, or by very light smoke obtained by
combustion of some suitable substance (incense for example) so as to obtain
particles whose dimensions are in the order of 1 Pm The introduction of particles
needs to be effected so as to produce as uniform a distribution as possible in the
flow. Figure 7.3b shows a region of flow between a stream of velocity V and a
region of flow at rest; the frontier f between the two oscillates unpredictably; only
fluid particles coming from the flow containing seeding will provide a measurement
signal: we thus see a systematic bias in the measurement of the velocity which will
“favor” the same kinds of flow structures.
The LDA procedure provides a local and instantaneous measurement in most

conditions. It is well suited to the laboratory study of small flow structures and it
even allows us to obtain acoustic velocities (from 1 mm/s to 1 cm/s) [PEU 92]. Cup
anemometers only allow measurement in a volume in the order of some
considerable fraction of a dm3, and therefore these are only suitable for quite global
measurements (wind velocity in meteorology or air-conditioning units).
Let us finally recall anemometric measurements at the atmospheric scale, which

are obtained by releasing balloons that move at constant altitude and whose position
can be tracked by satellite. This method provides trajectories (Lagrangian
viewpoint) and not streamlines (Eulerian representation of meteorological charts).
7.2.3.3. Eulerian anemometry

We here use a device, which is fixed in space, on which the flow acts by creating
a phenomenon whose intensity (overpressure, temperature, etc.) depends on the
velocity of the flow.
In the steady flow of an ideal incompressible fluid for which Bernoulli’s theorem
is valid, the difference in driving pressure between two points whose velocities are
respectively equal to V and 0 is equal to:
V2
'p g U
2
The measurement of velocity is thus realized by means of a measurement of a

pressure difference. A zero velocity is obtained when a flow is stopped by an
obstacle, on the upstream edge of which a stagnation point A occurs (Figure 7.4a); it
suffices to measure the total driving pressure (or the total pressure for a gas) at this
point A by connecting the orifice to a pressure probe. This orifice needs to be of
very small size in order to obtain a well-defined pressure value. However, if the
flow is not correctly aligned with respect to the obstacle, the orifice will not be at
the stagnation point and will not therefore measure the total pressure (Figure 7.4b).
The effect of the angle of incidence can be reduced by using a tube with a divergent
opening (Prandtl tube, Figure 7.4c).
V V V
L
A A
A A
L
(a) (b) (c) (d)

Prandtl tube Pitot tube
pg+UV2/2 zpg+UV2/2 pg+UV2/2 UV2/2
Figure 7.4. Measurement of stagnation pressure in a flow
By placing the total pressure tube inside a streamlined cylindrical tube (Figure
2.4d), the flow quickly returns to a uniform state (after 5 or 6 external diameters);
lateral orifices L placed on the external wall will measure the static driving pressure
pg (or static pressure in a gas) and the difference in pressure between the two tubes
provides a direct estimate of the dynamic pressure UV 2 2 . This device, known as a
Pitot tube, constitutes a somewhat cumbersome obstacle (it is at least a few

centimeters in length); it assumes that the flow velocity is defined and uniform at
this scale.
Bernoulli’s theorem is also applicable for a compressible fluid in subsonic flow,

provided of course that the compressible form of Bernoulli’s equation is used. The
problem is more difficult in supersonic flows because of the existence of a shock
wave upstream of the tube, across which specific laws need to be applied.
Hot wire anemometry is based on a very different principle: the thermal power
evacuated by a heated body is an increasing function of velocity. If the body is
electrically heated, this power can be known by means of a measurement of
intensity. It can be used in a flow in which velocity fluctuations exist with very
small hotwire diameters (1 to 5 Pm) whose thermal inertia is compensated by a
suitable supply (constant temperature anemometer). It is thus possible to measure
velocity fluctuations whose frequencies may be as large as 1,000 to 2,000 Hz. With
two or three differently inclined hotwires it is possible to measure the instantaneous
velocity vector. We therefore have in the hotwire anemometer a powerful means of
knowing the velocity fluctuations provided these remain smaller than the mean
velocity, as it is clearly not possible to distinguish the velocity direction from a
thermal power measurement alone.
7.2.4. Temperature measurements
The local measurement of temperature at the heart of a continuous medium can

be achieved by optical methods (measurements at a distance) or by means of probes
introduced into the flow. Optical procedures are based on emission phenomena,
which are associated with absorption phenomena, the medium considered being thus
semi-transparent for the wavelengths used. This leads to calculation methods which
are often complex and we direct the reader to specialized texts for more detail (see
for example ([BRU 95], [JOH 98], [MAR 99], [MCG 88]).
In a manner analogous to the introduction of pressure or velocity measurement

probes, the introduction of temperature measurement probes provokes a
modification of the flow and associated thermal phenomena only of importance at
high velocities. As with the pressure, the temperature of a fluid will depend on the
position of the measurement element on the wall of the probe; these measurements
are generally performed either at a stagnation point, or on a wall which is parallel to
the flow. For an ideal gas in adiabatic flow, we have conservation of total enthalpy
(section 4.3.2.3.2), or, with Ta the stagnation temperature:
V2
C pT C p Ta
2
However, the measurement of temperature presents additional problems, as the

temperature of the sensible part of the probe is the result of a thermal balance
between the heat flux transmitted by the flow, the heat flux in the metallic electric
wires and in the support structure, and the thermal radiation caused by the
surrounding walls. As the last two quantities are independent of the flow, their
influence is less important on the temperature measurements at higher velocities.
We will note that for high enough velocities, thermal dissipation in the vicinity of
the obstacle, which is constituted by the temperature probe, may be a non-negligible
factor ([SCH 99]). Temperature measurements in flows are always difficult and
should not be attempted without a careful discussion of the different thermal
phenomena which may be present.
7.2.5. Measurements of concentration
The measurement of the concentration of a mixture is quite difficult. Very often

a fluid sample is taken via local suction by a tubular probe. The probe geometry is
not important here, but the suction velocity needs to be suitably chosen so as not to
alter the fluid trajectories in the extraction zone (isokinetic sampling) ([BRU 95],
[JOH 98], [CHE 88], [LIP 05]). An analysis of the fluid sample is then performed
using physico-chemical methods adapted to the mixture under study.
In certain cases, it is possible to perform measurements from a distance by

means of optical procedures associated with radiation emission. It is also possible to
use physico-chemical reactions on a surface placed in the flow, but the specific
procedure employed is always dependent on the particular mixture studied, and
often requires modeling of the heat and mass transfer phenomena involved. For
example, we have already said that evaporation phenomena induce a variation of
temperature: a wet thermometer does not measure temperature of air (section 2.3.2).
In a conducting liquid medium, the use of electrodes allows the measurement of
concentration in certain circumstances (diffusion phenomena have to be negligible).
7.2.6. Fields of quantities and global measurements
7.2.6.1. Introduction to field quantities

In practice, local quantities are of limited interest, unless they are representative
of the device being studied. The dimensioning of a device or the control of its
operation requires knowledge of global quantities (mass flux, thermal power, etc.)
or of certain local quantities, which may be associated with possible damage

(temperature at a hot spot in thermal systems). These two kinds of data require
knowledge of the associated field quantities, either for integration (computation of a
flow rate for example) or for identifying the point where critical values may be
attained. In many situations, knowledge of field quantities can only be obtained by
exploring the domain by means of local measurement probes. This is clearly only of
interest if the concerned experiment is reproducible, which can only be the case if
the flow is perfectly defined. It is then possible to obtain velocity, pressure,
temperature fields, etc. This situation does not include poorly defined flows
regardless of their origin (instabilities, turbulence, etc.).
Steady flows do not present particular difficulties, since the instant of

measurement is unimportant. For unsteady flows, this is not the case however, as the
fields of a quantity g(xi,t) now depend on time. The installation of an ensemble of
probes, which instantaneously and simultaneously measure the quantity g over the
entire domain considered, is not generally realistic. Only methods which allow the
obtaining of full instantaneous fields by optical means are possible; such techniques
have seen significant progress in recent years.
7.2.6.2. Direct obtaining of field quantities of a flow

7.2.6.2.1. Visualizations
The visualization of a flow can consist of “photographing” visible material
elements which have been placed into the flow without disturbing it at fixed
locations with respect to the observer. For instance, it is possible to place pieces of
light wires on a grid inside a flow; these will then be oriented depending on the
direction of the local velocity. The same procedure can be used on a wall in order to
obtain the direction of streamlines and to visualize separated zones. The visible
effect is here a Eulerian representation (observation of streamlines).
It is also possible to introduce streams of smoke into the air or to inject colored
dyes in a liquid medium. This kind of injection requires certain precautions (suitable
injection velocity, density of colored dye equal to that of the liquid medium, etc.).
Diverse particles can also be introduced and entrained by the flow (which has not at
some point stopped to observe the flow structures visible in a river, a gutter, etc.).
However, it is important to remember that this kind of visualization is Lagrangian,
and it shows streaklines which may be very different from the streamlines and
trajectories (section 3.3.2).
Finally, Eulerian visualizations can be performed on walls by means of physical

processes (entrainment of a coating comprising particles for example) or physico-
chemical processes producing a differential deposit on wall streamlines (evaporation
of the coating solvent, or electrochemical deposit under particular conditions). For a

more complete and practical information, see ([HAN 05], [MER 87], [YAN 01]).
7.2.6.2.2. Obtaining instantaneous fields

The preceding visualization methods are rather qualitative. We can make these
quantitative by introducing tracer particles and photographing these with suitable
lighting, either over a period short enough for visualization of the distance covered
during the lighting-time or by means of two successive light pulses. It then remains
to measure the segment covered by each particle in order to ascertain the velocity. It
is obviously necessary to know the position of the particles in space, and this is done
by generating a light sheet of small thickness, particles external to this light sheet
not being visible. This method requires a seeding of particles with very small
diameter (1 to 3 Pm) which is not biased (section 7.2.3.2). Such approaches are
known as particle image velocimetry (PIV). Modern techniques for the generation
of two close light pulses of short duration (5 to 10 ns) by a suitable laser and the
processing of images have made this technique less fastidious than in the past ([RAF
98], [STA 00], [STA 00]). However, in turbulent flows, the obtaining of a mean
velocity field requires statistics to be gathered from a large number of identical
experiments (sections 7.2.6.3.2 and 7.2.6.3.3).
Optical methods of temperature and concentration measurement provide field

quantities; as mentioned earlier, the semi-transparent aspect of the medium
considered (which is both emissive and absorbent) leads to rather global
measurements which depend on thickness, and a deconvolution of signals acquired
in a band of wavelengths is necessary in order to reconstruct the value of the
quantity in each slice. These optical procedures are used in observations by
meteorological satellites.
7.2.6.3. Application in unsteady flows

7.2.6.3.1. Unsteady repeatable flows
We here assume that the flow can be reproduced at will, in other words there
exists an initial instant with identical initial and boundary conditions. This is true for
transitional regimes, which precede the established regimes, and also that of
periodic flows.
We repeat the flow regime which is to be studied as many times as necessary,

and measure the value of the quantity g(x, t) each time (Figure 7.5a) at different
points (x1, x2,…, xp,…). We then represent the fields (Figure 7.5b) at successive
instants (tx1, t2,…, tn,…) which we have chosen. These operations, which are in
principle quite simple, clearly require the use of numerical calculations ([PEU 79],
[PEU 89], [PEU 91]).
g (x1,t )
g (x2,t )
g (xi t 1) , g (xi,t 2)
t
g (xp,t)
t t t g (xi,t n)
1 2 n
(a) time representation of g(x,t) (b) space representation of g(x,t)
Figure 7.5. Obtaining instantaneous fields of quantity g from

point measurements as a function of time
7.2.6.3.2. “Steady” turbulence flows

A turbulent flow presents random fluctuations. We will here call a “steady”
turbulent flow any flow whose mean value of the measured quantity, over a large
enough time, does not depend on the initial instant chosen to begin the
measurement. We see immediately that an analogous procedure to that described in
the last section allows the obtention of quantities derived from the curves g(x, t) by
means of appropriate operations (mean values, fluctuations, variance of the quantity
g, etc.) used in the study of turbulent flows ([SCH 99], [YIH 77]).
7.2.6.3.3. Unsteady “reproducible” turbulent flows

This category of flow is obtained by imposing fixed initial and boundary
conditions that are defined functions of time. As before, we repeat successive
measurements at different points by reproducing the experiment exactly. As the
flow is maintained in a turbulent state, in other words it fluctuates from one
experiment to the next, it is necessary to perform the measurement a sufficient
number of times at each point, for the associated statistical quantities to be
calculated; these statistical quantities are the only ones which are meaningful for this
ensemble of experiments.
7.2.6.4. Measurement of global quantities

Global quantities such as mass or volume flux, internal energy or enthalpy flux
are very important in industry. Precise knowledge of these can be obtained by
integrating local measurements performed on a grid that covers the region of
interest. This method of integration is used for establishing norms for precise
measurement. However, these methods are time-consuming, expensive and often
cumbersome. It is thus necessary to implement more global procedures based on
laws of fluid mechanics and transfer. The corresponding devices require calibration
in laboratory conditions. Each of these measurement procedures is based on a
particular physical phenomenon, for example, the flow rate in a duct can be obtained
from the mean spatial velocity deduced from the frequency of vortices emission
behind a cylinder, after a calibration of the used device.
7.2.7. Errors and uncertainties of measurements
Errors, which occur in the measurement of physical quantities, arise as a result
of diverse factors. It is first of all necessary to appreciate the experimental
conditions, which lead to an experiment always being of an approximate nature. The
control of experimental conditions (temperature and velocity, etc.) are factors which
can significantly diminish these errors, although without eliminating them. In other
words, the data of a problem are only known within the bounds of some uncertainty.
Measurements are always accompanied by errors of various kinds, depending on the
kind of methods that are used:
- with analog devices errors come from the sensitivity of the components which
are used, parasite phenomena (influence of temperature, etc.), dry friction in devices
with moving parts, aging of components, etc.;
- the numerical treatment of an electric signal introduces no error (with the
exception of rounding errors); it does however introduce errors in the signal
acquisition to be treated (sampling and digitization mainly: for example, the
digitization on 8 bits leads to an additional uncertainty equal to ½8 in absolute
value).
Because they retain many decimal places, numerical devices are too often
considered to be capable of a precision that they do not necessarily provide. They
constitute most often only the last visible phase of a complex process of
measurement, the precision of which is not representative: the precision of the
numerical presentation is not directly related to the uncertainty on the quantity
measured and the error in the measurement of a voltage by means of a measurement
system having a numerical display is not simply the digit of the last decimal place.
7.2.7.2. General properties

A measurement error is not always exactly known, otherwise it would be
possible to eliminate it. It is systematic when due to a particular flaw in the
measurement device; this flaw is generally not known, otherwise a correction would
be possible. It is random, when it is associated with parasite phenomena (noise from
electronic circuits, diverse fluctuations, etc.). Regardless of their origin, errors will
always exist, and can never be exactly evaluated. The only quantity that is really
available is the uncertainty of a measurement or numerical value, in other words an
upper bound on the absolute value of the error. Uncertainties moreover verify rules
that correspond to properties of distances as they are defined in mathematics
(triangular inequality, etc.).
The evaluation of uncertainty of a measurement should always be made after

error calculation, which is performed in accordance with the usual rules for
calculating small variations in the variables of the problem. For example, we
calculate a quantity s (a, b) using measurements of signals a and b. The error G can
immediately be obtained as a function of the errors Ga andGb:
Gs s a' G a s b' G b
The formulae obtained in the calculation of errors allow us to obtain the

uncertainty of the quantity considered by bound estimation for the errors. For the
quantity s, it is possible to obtain the uncertainty 's as a function of the uncertainties
'a and 'b:
's s a' ' a s b' ' b [7.1]
We draw the reader’s attention to this last point, which is very important in
practice. It is indeed regrettable that these elementary ideas are so often completely
forgotten in the beginning of higher level education.
Let us note that depending on the content of the mathematical expressions, the
uncertainties may be partially compensated or notably worsened, as we will see in
the following example, where we evaluate a quantity s (a, b) from measurement of
signals a and b each of which has the same uncertainty 'a . We will leave it to the
reader to calculate the relative uncertainty which results in s in the two following
cases:
ab 2b' a 2a' b

1/ s (answer: ' s )
ab a b 2
a2 § 2a · a2
2/ s a (answer: ' s ¨¨1 ¸¸ ' a 'b )
b © b ¹ b2
In the second case, we see that if b is equal to 2a, the measurement of a does not
cause any uncertainty in s (to second order). The function s thus presents an
extremum with respect to the variable a, making it stationary. The situation is thus
quite optimal for measuring the quantity s. In general, expression [7.1] shows that
the lower the sensitivity of the quantity measured to the variations of the
measurement parameters, the better the precision of the measurement. Note that
systems that are close to instability are particularly sensitive to perturbations and
that the measurement of associated quantities is thus particularly difficult.
These elementary ideas are sufficient in simple cases. However, they must be
completed in the case of complex measurements requiring a large number of
unknowns and of quantities to be measured.
7.2.7.3. Errors and conditioning of a linear system

These ideas of errors and uncertainty have been broadly studied for the
numerical solution of linear systems of various sizes. The development of computer
technology has led to the possibility of measuring large numbers of quantities
simultaneously and of deducing interesting physical quantities by solving the
equations that characterize the measurement process. The problems posed here are
identical to those encountered in the numerical study of systems of equations
obtained from modeling.
Let us assume that the evaluation of a vector quantity X is performed via the
measurement of another vector quantity B of the same size. The vector of unknowns
X and the vector of given data B are assumed to be related by a linear system in
which the square matrix A models the measurement process:
AX B [7.2]
As we are here interested in discussing errors that are assumed small, the linear
system can always be obtained by linearizing the equations about the experimental
conditions and the matrix A is constant. It is invertible (otherwise the measurement
is not meaningful). The precision which can be obtained for the solution X A 1 B
depends on the sensitivity of the system: if a small variation in the right-hand side
GB leads to a large variation of the solution, we must expect poor precision in
obtaining the solution. We quantify the idea of sensitivity by means defining the
condition number N p A of the matrix A by the relation:
N p A A p A 1
p
The symbol . p
designates a matrix norm defined most often by means of a
vector norm. For example, the norm A 1 of the matrix A is the maximum value
taken as the sum of absolute element values of each column of A. A norm . p
satisfies the conditions:
AB p
d A p
B p
; OA p
O. A p
; A.B p
d A p. B p
Using this condition, the reader can easily verify the following inequalities:
GX p
GB p
1 d A p . A 1 ; d A p . A 1
p X p p B
p
This inequality shows that the condition number N p A constitutes an upper

bound of the relative amplification error between the data B and the unknown X in
solution of system [7.2]. The greater this value, the greater the sensitivity of the
solution to variations in B. The precise condition number value depends on the
choice of the norm. A well-conditioned system ( N p A of the order of 1) allows the
obtention of a good accuracy in the solution.
In order to illustrate the influence of the condition number we will consider a

rudimentary example where X has two components. Consider the matrix A with
which we associate the norm A 1 (defined above):
§1 D·
A ¨¨ ¸¸ [7.3]
©0 1 ¹
The reader can certify that we obtain:

§1 D ·
A 1 ¨¨ ¸¸; A1 A 1 1 D; N 1 A 1 D 2
©0 1 ¹ 1
hence:
§ 1 D ·§ b1 · § b1 D b2 ·
X A 1 B ¨¨ ¸¸¨¨ ¸¸ ¨¨ ¸¸ [7.4]
© 0 1 ¹© b2 ¹ © b2 ¹
When D is equal to 0, the system is perfectly conditioned ( N1 A 1 ). In this

case the solution is the identity ( X B ), or: ( x1 b1 , x 2 b2 ). We thus directly
measure the unknowns, which is ideal.
On the other hand, for D 100 , we have an ill-conditioned system and small
variations of measurement B can lead to large variations in solution X. The reader
can easily verify that, for the preceding example, changing the vector B from:
§100 · §100 · §0· §100 ·
¨¨ ¸¸ to ¨¨ ¸¸ will cause solution X to change from ¨¨ ¸¸ to ¨¨ ¸¸ : a variation of
© 1 ¹ © 0 ¹ ©1¹ © 0 ¹
1% in the modulus of the inputs data B leads to a variation of a factor of 100 in the
vector X, which has also rotated by 90°. They can also calculate the relative
variation GX 1 X 1 of the solution and that of the right-hand side, in addition to
the amplification factor of the relative error between B and X.
These considerations show that the conditioning of a linear system of equations

deteriorates as its matrix becomes filled and its coefficients are large. A full matrix
of large size is thus very poorly conditioned.
Poor conditioning can be improved by a suitable change in the way the equation
is written. The system of the last example [7.3] is optimal. The system of the last
example [7.3] can be written in the solved form [7.4] whose conditioning is optimal.
The accuracy problem has not been resolved however, as we see that it is the
quantity b 1 Db 2 which must be measured directly with good precision. Changing
the way the system of equations is written without changing the measurement
method clearly improves nothing.
In conclusion, this idea of conditioning is essential in order to appreciate the

quality of an experiment, a model or a numerical processing of information.
Considerable theoretical and practical progress has been made in this domain. We
refer the reader to [DEM 97], [PRE 07] and to manuals of computation tools
(MATLAB for example; instructions “cond” and “condest”).
7.2.7.4. Consequences for measurement techniques

In general, the evaluation of a quantity is effected by measuring other quantities
that are related to the first by relations that characterize the measurement procedure.
The matrix A of the corresponding system of equations must be well conditioned in
order that the uncertainty is as small as possible. In cases of poor conditioning of the
matrix A, the very principle of the method of measuring the unknown quantities is
questionable and needs to be modified. In simple terms, the preceding example
amounts to saying that if two measurements are determined for the most part by
means of a single quantity, it will not be possible to obtain an indication of the value
of another quantity with good accuracy and it will be necessary to change the
measurement method.
For example, uncertainties always accumulate for independent measurements;

we should always avoid obtaining the value of a quantity g by means of the
difference in measurements between two much larger quantities. Thus, the value for
a small air velocity is obtained using Bernoulli’s theorem from the pressure
difference 'p between two cross-sections that are quite close together, but of
different sections. It is clear that 'p should be directly measured by means of a
differential manometer which is directly sensitive to this pressure difference and not
from the independent measurement of two pressures.
In general, the quantities measured should be in the same order of magnitude as

the unknown. Let us take another example, the propagation velocity c of sound
signals with respect to matter is large (about 340 m/s in air or 1,500 m/s in water). If
the matter is in movement at velocity U (a few m/s), a signal emitted at A (Figure
7.6) will arrive at B (AB parallel to the velocity; AB d ) with a delay equal to
d c U , whereas a signal emitted at B will arrive at A with delay of d c U .
The separation W between the propagation times of the two signals is thus equal to

2Ud c 2 U 2 . In principle, the measurement of W allows the measurement of the
flow velocity U.
c+U c-U
A B
U
W
O t
Figure 7.6. Direct measurement of the delay between the arrival

of two synchronous signals in A and B
However, it is important not to measure the propagation times of the two signals,
but rather the time separation W; for this the two signals should be emitted
simultaneously, and the time difference can be measured for example by starting a
stopwatch when the first signal arrives. Proceeding in this way, errors of
discretization and digitization which are only concerned with a reduced value W
rather than much larger values will be avoided (in accordance with the ratio c/U). If
the measurement were performed by means of analogous measurement devices, the
conclusion would be the same, as these have a relative accuracy, which is associated
with the scale of the measurement.
In general, methods which involve ensuring that the quantity measured is zero in
conditions very close to those in which we operate are known as zero methods. The
electric-bridge methods used (Wheatstone bridge, etc.) to measure impedance are
well-known zero methods.
7.3. Representation of signals
7.3.1. Objectives of continuous signal representation
We have seen in the preceding chapters how a physical system can be
represented by a continuous medium or modeled by associating components in a
state of thermodynamic equilibrium. We have limited ourselves to thermodynamic
aspects without really discussing the “quantity of information” necessary for
knowledge of a system. This idea is quite difficult to define, as it depends in fact on
the complexity of the system, the degree of approximation which can be tolerated in
this knowledge, the sensitivity of the system towards perturbations and the way the
information is structured. We have only indicated that the number of variables
necessary is greater as the thermodynamic imbalances are more pronounced.
The continuous medium described in Chapter 2 is an indispensable mathematical

limit wherein the equations describe the observable macroscopic properties,
provided that the associated physical quantities have been included. In order to aid
the discussion, we will assume that the physical system and its appropriate model
(for example, the Navier-Stokes equations) are equivalent models: results obtained
from each should be identical (Figure 7.7). It is clearly an optimistic departure point,
but one which can afterwards be tempered by taking account of the errors associated
with the choice of models, the measurement procedures, the numerical methods of
calculation, etc.
The solutions of the theoretical model and the physical quantities of the
corresponding experiments possess the same properties of continuity and
differentiation. The representation of these solutions and quantities is the first
practical problem to be solved. In what follows, we will limit ourselves to a
function of one variable, which is for instance a temporal signal.
Data D of Physical problem Results S of

physical problem physical problem
(Inputs) (Outputs)
Continuum model
Boundaries conditions equations Hf=D
Solution f
of model (Navier-Stokes, for
example)
Figure 7.7. Physical system and its mode
In a procedure adopted over many years, any solution to the equations of

continuous media, mechanical or electrical, are presented in the form of an
analytical representation from which it is possible to calculate the value of a
measurable global quantity, for example the flow rate in a pipe as a function of the
pressure difference applied, the value of an electrical resistance of a conductor of a
given form, the thermal flux resulting from a temperature difference imposed
between two external surfaces of a medium, etc.
Thus, experiments have for a long time consisted of measuring certain local or
global quantities which allow the validation of the model of the system studied. The
appearance of computer techniques has not really changed this manner of
proceeding, but it has allowed us to increase to an extraordinary extent the power of
the means by which we compute quantities in the model. Progress in analog and
numerical electronics has also allowed instantaneous measurements that were
previously unimaginable. We thus find ourselves today confronted with an
enormous quantity of information contained in the results of numerical
computations or measurements, which need to be processed in as rational a manner
as possible. These new technologies have radically changed both the way in which
results are represented and used. Beyond these practical aspects, the comprehension
of physical phenomena has been profoundly modified. An analogous mutation was
produced at the beginning of the 19th century with the introduction of the Fourier
series.
Finally, the problem representing a function arises both from the point of view
of analysis and storage of measurement data, and that of the analytical or numerical
resolution of modeling problems. The numerical calculation of solutions of
differential equations was obtained either by performing finite difference
calculations or by means of tables, these having been calculated for discrete values
of the variables. In practice, these procedures required important and careful effort;
only the use of numerical tables was familiar to the physicist or engineer who was
mostly happy to calculate a numerical value by interpolation between two values of
a table.
In this chapter, we will limit our discussion to general procedures for

representing signals; the problems of representation associated with modeling will
be discussed in Chapter 8.
7.3.1.2. Objectives of signal representation

In these conditions, the first question to be considered concerns the practical
utility of a field of a continuous quantity and the reasons for using it. The model of
Figure 7.7 is a knowledge model, which allows us to know everything about the
problem that is posed. It contains variables which are continuous functions of time
and which represent quantities of the associated physical system.
A variety of different reasons lead to the representation of a temporal signal:

– we can firstly analyze it, in other words obtain information which allows us to
understand the physical process and eventually to modify it;
– the signal can also be recorded in its totality with a view to later use;
we can also extract more or less condensed information from this signal that
we will keep for a later, significant, signal restoration;
– in certain cases, we would like later to construct physical synthetic signals:
music synthesizers, speech or image synthesizers, modeling of real devices by
numerical simulation on computers, active control of phenomena, etc.;
– finally, we can perform a signal transformation by applying to it an algorithm
with the objective of computing another signal.
Depending on the objective in mind, the nature and the quantity of the
“information” which we intend to conserve and use, we proceed differently. In fact, we
often find ourselves confronted by a double problem of storage and/or of interpreting
information contained in a physical signal. Conserving billions of numerical values
without understanding what they represent is not very helpful; this operation has a
certain associated cost, even if this is becoming less and less. In certain difficult and
expensive experimental cases, it is of interest to conserve large masses of data, which
is possible using modern equipment, in the hope that we will know how to extract
pertinent information which we do not know how to extract at the moment of the
experiment; but the difficulty is storing the data in a suitable form.
With the exception of measurements performed with a specific objective, any

representation of a function of time s(t) should allow the signal to be recalculated. The
practical needs associated with the storage and reproduction of information concerns
all domains; beyond science and technology, the conservation of music, speech and
image (for historians, etc.) has a social interest. Signal processing is a complex
discipline that we will not consider in detail; the reader should refer to specialized texts
([BAH 01], [CAS 06], [JAC 91], [MEA 91], [PRI 91]). We will here only give certain
general indications, which should allow the reader to appreciate the information
processing problems that arise in the treatment of physical phenomena encountered in
acoustics, fluid mechanics and thermodynamics.
7.3.2. Analytical representation
A signal s(t) can be represented by a simple “analytical” function, in other words

a compact expression which defines a process at each instant of an interval of study,
either by means of predefined functions such as circular functions, polynomials,
Bessel functions, etc. or by means of formulae which imply one or many known
methods (integration, differentiation, convolution, etc.). The “analytical” term is not
here to be taken in the strict mathematical sense, despite the fact that the function
used can satisfy the mathematical definition of analyticity.
This analytical representation, exact or approximate, can be obtained in different

ways:
– an exact explicit solution of a system of equations that constitute a model,
although in practice this is rarely possible for continuous media in flow;
– an approximate global solution of the same system of equations by a procedure
which consists of satisfying the averaged equations (weak solution). Different ways
of proceeding exist; for example, we can replace the equations with integral
conditions which constitute a simpler system of equations containing fewer
variables and to which it is possible to find an analytical solution (see elementary
examples discussed in sections 6.2.6, 6.3.1.2 and 6.5.2.2) or a solution of a form
which is given a priori and for which certain coefficients can be obtained by least
square methods (error minimization, etc.);
– interpolation functions (polynomial or other function) obtained from punctual
measurement data, graphical recordings, etc.
The analytical representation of a signal s(t) thus consists of defining the class of
functions used and the parameters that characterize this particular function. In
general, predefined elementary or special functions allow a particular synthetic
knowledge, which a numerical representation does not provide. Knowledge of their
properties often allows interpretations of the solution thus obtained and reasoning
regarding the relations of cause and effect by means of known analytical properties.
It is thus possible to derive particular properties or other analytical forms without
any numerical computation in the context of the theories used. These analysis
possibilities only exist if the analytical representation comprises only a handful of
coefficients: the properties of a full series are too general to be useful, except if they
represent known functions or if they are defined by laws of recurrence.
Finally, certain exact solutions of the equations of fluid mechanics, of

thermodynamics and acoustics, can be brought to the solution of linear differential
equations if the partial differential equation is linear, or even non-linear in the most
cases. We have already seen examples (sections 5.4.5.4 and 6.1.1.2.4), which most
of the time correspond to a well-defined physical evolution.
In summary, a curve that is measured or numerically calculated or a table of n

numerical values gives raw, unstructured information that is apt to represent many
kinds of signals. On the other hand, an analytical formula, which characterizes the
curve, provides a structured information. An analytical function with few numerical
coefficients often allows clear physical concepts to be associated with the
information; obviously this analytical formula can only represent very specific kinds
of phenomena. The development of computer technology has unfortunately led to
analytical results being neglected, which frequently allow analysis and sometimes
predictions based on physical arguments. Numerical calculations are of course not
to be neglected; on the contrary, it must not be forgotten that they provide
knowledge of the same kind as an experiment.
7.3.3. Signal decomposition on the basis of functions; series and elementary

solutions
7.3.3.1. Representation in the form of a series of functions

Exact mathematical representations can also be obtained in the form of series or
integrals, but in so far as general procedures are concerned, their physical interest is
often limited. Thus, a function of a real variable can be decomposed in terms of a set
of basis functions, of which there are many kinds. For example, a signal can be
represented by a power series development (Taylor) around an instant t0:
f t t 0 n
s (t ) ¦ s ( n) t 0
0 n!
This series is often unusable in physics, as the property of infinite

differentiability does not exist; furthermore, the convergence of the series of an
analytical function is often limited to a finite interval. We will nonetheless note the
following particularity of power series developments: all information concerning the
function of time to be represented can be found concentrated at a given instant t0.
This assumes that the future of the function is completely determined from that
instant: the fact of indicating data in the long term creates problems of accuracy
which are manifest in the numerical value of the higher order derivatives. Such a
representation, which is theoretically possible for large values of time (the power
series cos Zt converges regardless of t), is nonetheless unusable in practice.
A function s (t) can also be developed as a Fourier series on a finite interval

[0,T] of length T:
f § 2Snt 2Snt ·
s (t ) ¦ ¨¨ a n cos bn sin ¸¸
0 © T T ¹
The coefficients an and bn being given by the classical formulae:
1 T /2 2 T /2
a0 ³ s t . dt an ³ s t . cos2Snt / T dt
T T / 2 T T / 2
2 T /2
bn ³ s t . sin 2Snt / T dt
T T / 2
The Fourier series represents a periodic function of period T; it is badly adapted

to the representation of very rapid variations and discontinuities. As the harmonic
functions of period T/n are orthogonal (where n is an integer), we have:
1 T 2
³ s t dt
T 0
a 02
1f

¦ a n2 bn2
2 1

This property can be interpreted as a conservation of energy: the energy of the
signal during time T is equal to the sum of the energies of the harmonics (Parseval’s
theorem).
There exist many other basis functions sets on a finite interval: Bessel functions,
Legendre and Tchehychev polynomials, etc. (see mathematical texts). The interest
in these is often related to the nature of the considered problem. Certain functions
such as real decaying exponentials are particularly useful for the study of damped
systems, but they do not form an orthogonal basis set. Finally, basis functions are
not necessarily continuous, as we will see in a later example (section 8.3.2.3).
7.3.3.2. Representation by combinations of elementary solutions

The series evoked above are often chosen as a function of simple and universal
mathematical properties (Taylor series, etc.) without any prior consideration of the
physical properties of the system studied. Another manner of constructing solution
representations consists of combining elementary solutions of the equations studied.
For example, for problems associated with Laplace’s equation (electrostatics, steady
conduction of heat or electricity, irrotational fluid flow, etc.), we can seek an exact
or approximate solution that satisfies the boundary conditions in the form of a
linear combination of monopoles, dipoles and vortices. In general, such methods are
possible for linear problems in different forms (singularities and multipole
distribution methods, Green’s functions, boundary element methods, etc.). These

methods often use a mathematical formalism which can be quite involved and which
we cannot cover in this text.
We will encounter this kind of methodology in the synthesis of musical sounds;

we can however already note that a musical score constitutes a combination of
elementary solutions (the notes of the instruments and indications regarding their
interpretation) which allow the representation of a musical sequence in a manner
equivalent to the numerical values recorded on a CD.
7.3.3.3. Signal reconstruction

The reconstruction of a signal s(t) represented in the form of a series or as a
combination of functions is an intricate exercise, which can sometimes be quite
difficult. Finding a simple analytical representation of a series is an almost
impossible task, except in very specific cases; only a numerical reconstruction of the
signal is possible.
Let us recall that a power series is associated with a function independent of the
value of its radius of convergence and it is characterized by the infinite sequence of
coefficients. Because of this, calculating the value of a function thus represented is a
normal mathematical operation, even if the series diverges. The simplest way to
obtain the values of the developed function s(t) is to numerically calculate the sum
of the series when possible. In the case of series divergence (or poor convergence),
there exist procedures which allow the divergence problem to be contoured or the
speed of convergence to be improved in order that its sum be efficiently calculated
(analytical prolongation, Euler algorithms, Cesaro or Féjer sums for Fourier series,
etc. The interested reader should refer to mathematical texts ([ABR 65] p. 16, [BRE
91], [PRE 07])).
In summary, a series development of a function s(t) replaces the function by a

denumerable sequence of numerical values (series coefficients) and knowledge of
the basis functions. However, such developments are only of interest if the function
can be calculated with a small number of terms thus obtained, which is often the
case when the basis functions are solutions of the equations which correspond to the
model of the physical problem (Appendix 4).
7.3.4. Integral transforms

An integral transform is a generalization of the idea of a series, which amounts
to developing a function on a continuous infinity of basis functions g Q , t that
depend on the parameter Q. Instead of obtaining a denumerable sequence of
coefficients, the result of the integral transform T of the function s(t) is the function
Ts(Q) of the variable Q; as the information provided by the function is greater than
that contained in a denumerable sequence of coefficients, we see that the conditions
for applying integral transforms are far broader than those of series developments.
We can often associate an integral transform with a series development. The integral
transforms can comprise complex values.
An integral transform can often be inversed, in other words there exits another
integral transform T –1 whose result T –1(T (s(t))) is equal to the original function
s(t). The exact or approximate reconstruction of the initial signal is the obvious
condition for using an integral transform to store a signal. This condition is not
necessary for its analysis.
The correlation coefficient C fg of two temporal functions f(t) and g(t) is

defined by the relation:
f,g ! 1 T
C fg with : f , g ! ³0 f (t ) g (t )dt
f 2
! g 2
! T
In the usual definition, we consider the infinite interval T by taking the limit. In
practice, this coefficient is always calculated on as large a finite interval as possible.
An integral transform is thus a factor excepted no more than the ensemble of
correlations between a function s(t) and the family of basis functions g Q , t .
An integral transform only provides interesting information and simplifications

if a large correlation is obtained for a small number of functions g Q , t . In such
cases significant interpretations may be possible. The use of an integral transform is
only of interest if a reduced set of basis functions represents a notable part of the
properties of the signal studied. There exist many integral transforms. We will give
some examples of commonly used transforms with only some of their basic
properties, chosen on account of their physical interest. The reader will find a more
exhaustive discussion in specialized books ([AND 99], [BEE 03], [DEB 06], [GUP
83], [JER 92], [WOL 79]).
7.3.4.2. Fourier transforms

7.3.4.2.1. Definition and properties
We will leave aside the rigorous definition and conditions for existence of the
Fourier transform. Consider a real-valued temporal signal x(t); its Fourier transform
Fx(Q) is a complex function of the frequency Q defined by the relation:
f 2S jQ t
Fx (Q ) ³ f x(t ) e dt
The real and imaginary parts of this transform Fx (Q) are:
Re>Fx (Q )@ Im>Fx (Q )@
f f
³ f x(t ) cos 2SQ t dt; ³ f x(t ) sin 2S Q t dt
The modulus Fx (Q ) and the phase M >Fx (Q )@ of Fx(Q), respectively known as

the amplitude spectrum and the phase spectrum, can be written:
§ Im>F x (Q )@ ·
F x (Q ) Re>F x (Q )@2 Im>F x (Q )@2 M >Fx (Q )@ Arctg ¨ ¸
¨ Re>F (Q )@ ¸
© x ¹
The temporal signal is reconstructed by the inverse Fourier transform:
f 2S jQ t
x(t ) ³ f Fx (Q ) e dQ
Fx(Q) and x(t) are two different and equivalent representations of the same
quantity, respectively in frequency space and in the temporal domain.
The main properties of the Fourier transform are:

1) the property of evenness:
x(t) real Re[Fx(Q)] even, Im[Fx(Q)] uneven

x(t) real even Fx(Q) real even
x(t) real odd Fx(Q) imaginary odd
x(-t) Fx(-Q) = Fx(Q)*
2) a stretching of the timescale of the function x(t) which leads to a contraction

of the frequency scale for the transform Fx (Q) and vice versa (similarity):
1 Q
Fx ( at ) (Q ) F x (t ) ( )
a a
3) a translation W of the timescale of the function x(t) which leads to a phase

rotation equal to –2SWQ for the transform Fx (Q) and vice versa:
Fx (t W ) (Q ) e 2S jW Q Fx (t ) (Q )
4) the Dirac distribution G(Q) is defined by the relation:
f
³ f G (t )dt 1 with: t v 0, (t) 0
It satisfies:
f 0 .
f
³ f G (t ) f (t )dt
We can easily show that its Fourier transform is equal to one:
f 2S jQ t
FG (t ) (Q ) ³ f G (t )e dt 1
The Fourier transform of a Dirac distribution centered at the origin is a constant

function of the variable Q, equal to one on the interval [-f, +f];
5) the Dirac distribution G (t W ) centered at the instant W verifies:
f W
f
³ f G (t W ) f (t )dt
From the relation of translation of the timescale, we obtain its Fourier transform:
f 2S jQ t
FG (t W ) (Q ) ³ f G (t W )e dt e 2S jW Q
6) the inverse Fourier transform of the Dirac distribution G (Q N ) in frequency

space is:
f 2S jQ t
FG(1Q N ) (t ) ³ f G (Q N ) e dQ e 2S jN t
f
FG(1Q ) ( ³ f G (Q ) e 2S jQ t dQ 1
7) the Fourier transform of the function e 2S jN t is then the Dirac distribution

centered on frequency +N
Fexp(2S jN t ) G (Q N )
8) the Fourier transforms of the functions cos 2SN t and sin 2S1 t are the half-
sum and the half-difference of two Dirac distributions centered on the frequencies
+D and -D.
The function A cos 2 S 1 t M , continuous in temporal space, is thus
represented by only three numerical values A, 1 and M in frequency-space;
9) the Fourier transform of the product of two functions x1(t) and x2 (t) is equal
to a convolution product Fx1 (Q)*Fx2 (Q) of the Fourier transforms Fx1 (Q) and Fx2
(Q), and vice versa;
10) Parseval’s theorem expresses that the energy of a signal s(t) is conserved by
the Fourier transform:
2 2
³ f st dt ³ f Fs Q dQ
f f
Es
We see completely different distributions of information in temporal and

frequency space: the non-zero temporal signal and the infinite duration of the cosine
function is found concentrated in four non-zero values (frequencies rQ, amplitude
and phase or complex amplitude), whereas the information of the Dirac
concentrated at the time axis origin is found spread over the entire frequency
domain.
7.3.4.2.2. Interpretation of the Fourier transform

Using the Fourier transform amounts to seeking the correlation between a signal
x(t) and the harmonic signals of frequency Q, which can be expressed in the form
ejSQt. This interpretation is valuable because it can be shown that the better this
correlation, the closer the signal is to a harmonic signal. For a signal mainly
comprising harmonic discrete signals, the correlation will be high for corresponding
frequencies (spectrum of lines) and the signal will be characterized by the values of
these lines. We thus obtain a small number of numerical values in the place of the
values obtained by temporal discretization of the signal. The Fourier transform
consists of representing the functions x(t) using a basis-set of harmonic functions.
The application of the Fourier transform on a temporal signal of infinite length

poses two problems that are difficult to reconcile a priori with the idea of
irreversible time associated with the second law of thermodynamics:
– on the one hand, we can only know the transform after a very long period of
time;
– on the other hand, it assumes that the beginning of the signal is situated a long
time ago in the past.
The first inconvenience makes it difficult to follow the evolution of a

phenomenon without a long enough delay; this can be remedied by means of
observation windows of limited duration. The second difficulty is more serious, as it
implies that the signals obtained by inverse Fourier transform are not possible to
realize if they occupy the time interval [-f, +f]. Only a non-zero temporal signal
which starts from a given instant is realizable: we refer to this as a causal signal,
which defines a possible action in time.
7.3.4.3. Laplace transform

The Laplace transform possesses properties similar to those of the Fourier
transform, and is defined by the relation:
f pt
L f ( p) ³0 e f (t )dt
Laplace transformation is not fundamentally different from Fourier

transformation, as it consists of taking in this Fourier transform imaginary values for
the variable in complex plane; it has the same properties, except for inverse
transformation.
This transform is of considerable interest, notably for the study of damped

systems in automatic control, the family of functions for comparison with the signal
s(t) being precisely the ensemble of aperiodic damped modes of linear invariant
systems of first order. The reader can refer to texts on the dynamics and control of
systems ([BEE 03], [DEB 06], [GUP 83], [WOL 79]). We will see in Chapter 8 the
application of the Laplace transform for the solution of linear systems with constant
coefficients (Appendix 1).
7.3.4.4. The Hilbert transform

The objective of the Hilbert transform is to define the amplitude and the
instantaneous frequency of a movement which is close to a harmonic movement, by
with variable characteristics. We here have a generalization of the definition of the
complex exponential e j 2Snt (Appendix 2). We define the Hilbert transform H t
of the signal H t by the relation1:
1 1 du
H t * xt VP ³ f xu .
f
[7.5]
St S t u
1 The notation VP means that the integral has to be understood in terms of Cauchy principal
value; notation * indicates the convolution product of two functions.
The Hilbert transform of cos 2SQt is equal to sin 2SQt ; in general, the effect
of the Hilbert transform is to introduce a phase lag of S / 2 in the initial harmonic
function. For any real signal xt , we can associate the complex analytical signal
D x t :
D x t x(t ) jH (t ) [7.6]
The modulus at and the argument Mt of the analytical signal D x t can be

defined as the amplitude and the instantaneous phase of the signal x (t):
xt ReD x t >
Re a (t )e jMt @
The instantaneous frequency Q i is defined as the derivative of the instantaneous
1 dM
phase Q i . . The idea of instantaneous frequency is only meaningful for
2S dt
functions that are close to harmonic functions, in other words for relatively
narrowband signals. We will note that these signals possess two very different
timescales, one rapid timescale corresponding to a “carrier” and another much
slower scale that characterizes a modulation. Figure 7.8 shows an example of a
signal modulated in amplitude (a) or in frequency (b).
Most musical signals are characterized by a fixed frequency, which is modulated

in amplitude, and in phase. They constitute the basis for the synthesis of sounds in
musical synthesizers. We will come back to this point a little later in section 7.4.3.4.
x (t) ax(t) x (t)
t t
(a) (b)
Figure 7.8. Signal modulated in amplitude (a) or in phase (b)
7.3.4.5. Cepstrum
The principle of cepstrum consists of taking the logarithm of a spectral density,
and then performing an inverse transform. Ordinary products and convolution
products are respectively transformed into sums and products; a harmonic
modulation of the spectrum is transformed into a Dirac in the inverse transform.

These properties are often used for signal processing applications such as:
– the suppression of echoes in acoustic signals;
– the characterization of vibrations in rotating machinery and more particularly
in obscuring the operation of machines (for example, in looking for defects in
rotating parts which result from the wear of bearings and which leads to the
appearance of lines in the spectrum or by modulation frequencies which are difficult
to see in a simple spectrum).
We will see examples of applications of the cepstrum in Appendix 3. Depending

on the problem, different definitions of the cepstrum are used ([DES 00], [JUR 08],
[NOR 03], [WAI 90]).
7.3.4.6. Short time Fourier transforms

The Fourier transform is defined on an infinite interval. For diverse reasons, we
can only record a signal over a limited duration, and this modifies the Fourier
transform. It is thus necessary to find a compromise between the volume of
information required and the accuracy of the results obtained. Recording a signal for
a finite duration T, consists of multiplying it by a non-zero function over this
interval and by zero outside of this interval. Following the usual terminology, we
will say that this function, called a gate function 3T(t), is a particular case
(rectangular function) of a window function. The gate function 3T(t) centered at the
origin can be written:
1 T 2 t T 2
37(t ) ®
¯0 t T 2 T 2 t
The signal transform thus truncated is written:
f 2S jQ t
Fx 3T (Q ) ³ f 3T (t ) x(t ) e dt Fx (Q ) * F3 T (Q )
The Fourier transform of the truncated signal x(t).3T(t) is equal to the

convolution product of the Fourier transforms of the signal x (t) on the infinite
interval and the gate T(t). The Fourier transform F3 T (Q ) of this one is a cardinal
sine function:
sin S Q T
F3 T (Q ) T
SQ T
Now consider the effect of recording the signal cos 2Snt using a gate of
duration T. The Fourier transform of a cosine of infinite duration is composed of
two Diracs (Figure 7.9a): the spectrum of the cosine convolved with the transform
of the gate of duration T is comprised of two cardinal2 sinusoids centered on the
frequencies r n of the two preceding Dirac distributions (Figure 7.9b). Considering
that the width of the central peak of the cardinal sinusoid is characterized by its first
zero, the widening of the Dirac function peak is equal to 2/T, which is, for a window
which records 10 periods, a widening in frequency of 0.1n on either side of the
frequency n.
Fs3
2 /T T
-n +n Q -n +n Q -n +n Q
(a) (b) (c)
Figure 7.9. Fourier transform of: (a) cos 2 nt; (b) 3T (t ). cos 2S nt with
1/T << 2n; (c) 3T (t ). cos 2S nt with 1/T = 2n
We see that the shorter the window, the wider the frequency band obtained: an
insufficient observation will reduce the accuracy of the Fourier transform by
“clouding” the signal. We note that as a recording which does not disturb the
spectrum should be applied with a window which is the inverse transform of the
Dirac distribution, in other words the unrealizable infinite width time window,
which we wanted to avoid.
According to the usual Rayleigh criterion, we consider that the frequency peaks
become unidentifiable if the first zero of the cardinal sinusoid centered at the
frequency +n is found at the frequency –n (Figure 7.9c), i.e.:
1/ T d 2 n
The perception of a frequency n requires thus an observation horizon of duration

T greater than 1/2n: the recording duration T should include at least a half period of
the lowest visible frequency of the short-time Fourier transform. Another result of
these considerations is that we can only distinguish two frequencies Q and Q +'Q if
they are separated by at least the value 1/2T. From a physical point of view, an
insufficient recording of information can only give bad results (see formula [7.8]
and Shannon’s sampling theorem (section 7.3.6.4)). We will note that the spectrum
obtained by such a transform has not only lost the details concerning the peak; but it
also contains low frequencies that do not exist physically.
³ f f W G t W dW f t .
f
2 We obtain:
The Fourier transform is thus only a correlation performed between a signal and
a family of harmonic reference signals. This observation allows a simple physical
interpretation of the widening of peaks, which results from the use of a finite
window. In effect, the correlation between two harmonic functions is zero over an
infinite time, unless their frequencies are equal. However, the same correlation will
be increased, as the window size become progressively smaller and as the
frequencies are nearer. We will leave it to the reader to verify these observations.
In reality, the widening of the Dirac spectrum by the cardinal sinusoid function
is not limited to the central peak of this function and the smaller but non-negligible
amplitudes of the lateral lobes can also lead to a net increase in the width of the
spectrum obtained. This last inconvenience is a problem, in particular for analyses
of acoustic signals on account of the sensitivity of the ear (the logarithmic decibel
scale clearly leads to a smaller scale of the amplitude variations). As the energy of
the secondary lobes is quite weak, we try to re-center it on the main lobe, even if
this means widening it slightly. This can be achieved by replacing the gate function
3T(t) by a window function ) T (t ) , which leads to much smaller amplitudes of the
lateral spectrum peaks than those obtained with a rectangular window:
f 2S jQ t
Fx)T (Q ) ³ f ) T (t ) x(t ) e dt Fx (Q ) * F)T (Q ) [7.7]
The recording of a raw signal over a limited duration therefore leads to

deformations of the Fourier transform consisting of two kinds of distributive
modification of spectral energy: the widening of the central peak and the appearance
of secondary lobes. This widening of the signal spectrum can be studied and
characterized for each window by taking the “moment of the signal energy”
2 2 2 2
³ t x(t ) dt and of its transform ³ Q Fx (Q ) dQ . General considerations ([BLA
98], [FLA 98], [HIG 93], [STR 96]) allow the demonstration of the Heisenberg-
Gabor inequality:
1
't.'Q t [7.8]
4S
In this inequality 't and 'Q are respectively the duration of the energy content of
the temporal signal and the width of the frequency band in which the energy is
contained; 't is of the order of T/2 and 'Q is analogous to the quantity 2/T defined
above for the rectangular gate. The equality is obtained for a Gauss window, which
corresponds thus to an optimum of the preceding minimization criterion. The
limitation of the preceding principle is related to the basic uncertainty of quantum
mechanics, but the physical analogy is far from complete, the interpretation of
quantities being very different in the two domains.
In practice, the effective choice of a window results from the ensemble of

various considerations. For a presentation and comparison of commonly used
windows, the reader should refer to ([BAH 01], [MAD 98], [NOR 03], [STR 96]
[MAX 96]).
7.3.4.7. Continuous wavelet transforms
The Fourier transform with a window function can also be interpreted as an
integral transform whose kernel is the product ) T (t )e j 2SQt of the complex
exponential e j 2SQt by the window ) T t . In other words, we no longer perform a
Fourier transform over a limited duration; rather, we perform a transform with a
function g t ,Q ) T (t )e j 2SQt that is different from the complex exponential. Such
a function is often called a wavelet.
This point of view can be generalized by considering functions of time that

contain additional parameters such as window width, which can depend on the
frequency. An example is the Gabor transform in which we decompose the signal
studied according to a basis of functions comprising the product of the complex
exponential e j 2SQt and Gauss windows of width D, centered here at the origin:
1
Gx D , Q e t / 4D e j 2SQt xt dt
f 2
³f
2 SD
In the Gabor transform, we thus have as an additional parameter the width D of

the window; we obtain, after transformation, a function with two variables which
provide redundant information for representation of the signal x (t). The frequency
representation studied in section 7.3.4.6 can be obtained by fixing the value of
window width; by varying this, we can obtain complementary information. Consider
for example a signal comprising one or two periods of a harmonic function of a
given frequency that is centered at the origin. Let us apply the Gabor transform, we
obtain a new transform which is not so different from the previous one, so long as
the analysis window has a width D of the order of that of the preceding signal, but if
this window is widened, the value of the transform will be reduced. A Gabor
transform with variable width allows the length of a signal to be identified.
7.3.4.7.2. Timescale transforms

We thus arrive at the idea that by using window width as a variable of the
analysis wavelet, we can obtain information regarding the timescales of the signal.
We thus use an analysis wavelet whose form is conserved when we dilate the
window width. It is clear that by proceeding in this way, the physical idea of
frequency is no longer so clear. If the window comprises little oscillation, then it is
its size that will be the pertinent variable for the transform.
t t t
(a) (b) (c)
Figure 7.10. Wavelets with variable width ((a) and (b))

or variable frequency ((b) and (c))
We define the transform Ts (a) of the signal s(t) by the wavelet h(t/a):
1 / 2 f
Ts a a ³ f s(t )h t a dt
in which the transform variable is the scale a.
A simple wavelet is the “Mexican hat” wavelet, which is the second derivative
of the Gauss function:
1 § 2 · t 2 / 2a 2
ht , a ¨1 t ¸e
2S a ¨© a2 ¸
¹
Figure 7.10 shows the window variations for a timescale analysis (windows a
and b) and for a time-frequency analysis (windows b and c).
Diverse problems arise in the choice of wavelets and in particular for the
reconstruction of a signal from its transform ([ALL 04], [FLA 77], [MIS 07], [STR
96]).
7.3.5. Time-frequency (or timescale) representations
7.3.5.1. General principles

The representation of a signal x(t) by means of an integral transform is useful so
long as this is not too different from the transform kernel: very short signals
centered at the origin will be well represented by wavelets, whereas the Fourier
representation is better suited to periodic (or nearly periodic) signals. For other
signals the preceding representations are numerically ill-conditioned.
Let us take the example of a centered signal of length 100 and which comprises
two sequences of oscillation (Figure 7.11) separated by a sequence in which the
signal is zero. Its representation by means of a Fourier series on the interval [-50,50]
(after change of time origin) is theoretically possible, but it will require a series with
so many coefficients that if will not be very useful.
x(t)
t
–50 0 50
Figure 7.11. Intermittent signal comprising two different sequences
The use of a Fourier transform in the same window will hardly be more
satisfactory. The use of a Fourier transform with window functions would be even
worse on account of signal attenuation towards the window edges. Furthermore, the
(complex) transforms will not indicate in a simple manner the fact that the signal
comprises two sequences of different frequencies at distinct instants, rather than a
uniformly distributed frequency content. We can also think of performing two
Fourier transforms on the separate intervals, but such a procedure is directly related
to the signal structure and so cannot be easily generalized.
The most direct representation thus consists of recording the signal by means of
a window function which is centered on the instant W, and then performing a short-
time Fourier transform. We are thus led to perform sliding transforms, which allow
the identification of the spectral content of signals contained in the window ) (t - W).
This sliding Fourier transform is expressed:
f 2S jQ t
F x) (Q , W ) ³ f )(t W ) x(t ) e dt
The transform Fx) (Q , W ) is a function of two variables W and Q which we will

qualify as a time-frequency representation. The spectral distributions become
functions of time. However, such a continuous representation contains much
redundant information concerning the signal x(t) since knowledge of a single
transform on the time interval of the signal suffices for a reconstruction of the initial
signal.
7.3.5.2. Practical applications

A natural solution for reducing the size of the time-frequency representation thus
consists of performing a partition of the total study interval into n temporal
segments each of a length 't in the order of the window width, and only conserving
the Fourier transforms performed at the center of the segments. We thus obtain a
succession of spectra, each of which can be attributed to a given instant.
Figure 7.12. Spectrogram
The representation of these spectra in the form of a succession of spectra as a

function of time is known as a spectrogram (Figure 7.12). However, a spectrogram
can quickly become quite complex and difficult to read for long signals. We often
prefer a representation, known as a sonogram, where the values are shown in the
time-frequency plane where the use of color or grayscale allows the indication of the
signal level. In this way the information contained in the signal can be visualized in
terms of the evolution of the frequencies and amplitudes.
Figure 7.13. Sonogram of the word “shah” pronounced by a person

Sonograms (often known as spectrograms) are very often used for the analysis of
complex signals; Figure 7.13 shows the sonogram of an acoustic signal
corresponding to the word “shah”. We can note that the ensemble of letters “sh”
corresponds to higher frequencies than the ensemble “ah”.
The considerations of section 7.3.4.6, relating to the time-frequency uncertainty,

here applies and the Heisenberg-Gabor inequality indicated earlier remains valid: a
precise temporal localization and an accurate analysis are incompatible. This
phenomenon is illustrated by Figure 7.14. which presents sonograms of a signal
composed of a silence of 0.05 second duration, followed by a harmonic oscillation
of frequency 4 kHz during 0.05 seconds, and then, once again a silence. The
analyses performed by means of the software COOL EDIT with windows of width
AF respectively equal to 0.031 and 0.002 seconds show that the first value allows a
suitable identification of the frequency, whereas the second allows a better temporal
localization of the portions of the signal. The Blackmann-Harris window used gave
better results than the other windows that were available; however, the reader
should remember that the choice of window depends on empirical considerations
([ALL 04], [JUR 08], [MAD 98], [STR 96], [WAI 90]).
Figure 7.14. Influence of the width AF of the Fourier window

on the sonogram (4,000 Hz)
The explanations which have just been provided for the time-frequency analysis
can be transposed to timescale analysis. This domain is more common in image
treatment; the interested reader should see specialized texts on this subject. The
principle of time-frequency representation can be applied in two dimensions, the
frequency or scale spectra being thus functions of two variables.
7.3.5.3. Regarding multiple timescales

The fact that spatial or temporal variations of a phenomenon take place at two
completely different scales has important but variable consequences depending on
the mathematical form of the variations. If these are relatively uniform and non-
oscillatory, we have a singular perturbation problem (see section 6.4.3). For
oscillations of a linear system, the preceding time-frequency analysis is only
meaningful if the notion of frequency can be defined, in order to allow separate
identification of the rapid component (short oscillation period) and the slow
component (characteristic time of the amplitude and phase variation, etc. (section
7.3.4.4)).
These notions of fast and slow scales are not however so easy to define. Let us
consider the example of a signal s(t), which is the sum of two harmonic signals of
close frequencies N 0 GN and N 0 GN :
1
s (t ) cos 2SN 0 GN t cos 2S N 0 GN t cos 2SN 0 t. cos 2S GN t
2
This expression shows that s(t) can be considered as a harmonic signal

cos 2SN 0 t with fast variations, whose amplitude is slowly modulated by the
function cos 2S GN t . The two timescales are here clearly 1/N and 1/GN. The
amplitude modulation can be any broadband slow signal a(t) as in the transmission
of radio waves.
Frequency modulation is also a system with two timescales, with a low-

frequency carrier signal N0 and a slow variation of frequency GN:
s (t ) cos 2SN 0 GN t t.
The spectrum of this signal modulated with frequency GN ((t ) GN 0 cos Zt is

not very different from the amplitude modulated spectrum: in addition to the two
close frequencies N 0 G N 0 and N 0 G N 0 , it contains only a small number of
weak peaks.
The amplitude spectra of amplitude and frequency modulated signals are nearly
similar, but the temporal signals and the phase spectra have very different structures.
This comparison illustrates the difficulty of reconstructing, with its detailed
characteristics, the time signal which corresponds to an amplitude or energy
spectrum: theoretically only the temporal and complex spectral representations are
completely equivalent. More generally, a signal similar to a harmonic signal of
frequency N0 can be studied with Hilbert transform, amplitude and frequency

modulations being defined from its analytical signal (section 7.3.4.4).
In practice, signals used in telecommunications are modulated in amplitude or in

frequency by a low frequency signal, which contains information to be transmitted
and decoded. In mechanical or energetic systems, these modulations are the result,
either of the presence of two close frequencies, or of a slow oscillatory variation of
the physical characteristics of an oscillator (stiffness or length of a vibrating system,
for example, or interaction between structures of different frequencies in fluid
mechanics or acoustics).
We will re-encounter the problem of multiple time (and space) scales in

turbulent fluids, at the heart of which are fluctuations of central importance. An
everyday example can be found for flow in the atmosphere: the wind presents
instantaneous fluctuations with very short periods (a couple of seconds for gusts of
wind), and longer periods (from many hours to a day for perturbations (in the
meteorological sense)) which are relatively individual structures at the atmospheric
scale. These two categories of scale are generally distinct and we can thus separate
meteorological and turbulence phenomena by performing statistics for the mean
values at the scale of the turbulent fluctuations. Meteorological predictions and data
provide the mean velocity, the effect of fluctuations due to gusts being smoothed at
timescales of a fraction of an hour.
On the other hand, turbulence frequencies have a continuous spectrum and the
temporal separation of phenomena is hazardous in proportion to the degree of non-
linearity of the turbulence mechanisms. Furthermore, the corresponding flows can
nonetheless be subject to temporal variations at a slower scale due to influences
other than that of the turbulence.
7.3.5.4. Study of intermittency

We call intermittency that property of a signal that comprises structures of the
same nature, which occur at intervals that may be more or less regular or random.
The signal is comprised of “packets” with a specific structure different from the
characteristics of the rest of the signal (Figure 7.15).
x(t)
0 t
Figure 7.15. Example of intermittent signal

Intermittency is encountered in many flows; for example, Figure 7.16 shows that
a fixed observer at the horizontal level N will perform measurements characterized
by intermittency for liquid packets in a water jet (a), the external turbulent zones of
a boundary layer (b), or ascending currents in clouds which are carried by the wind
(c).
U U
N N N
F
(a) (b) (c)
Figure 7.16. Intermittency phenomena in: (a) a liquid jet; (b) in the frontier zone of a
turbulent boundary layer; (c) in atmospheric flow with cumulus
Figure 7.17 shows the sonogram of a sound recording of a water flow that issues
from a floater tap in which a rapid liquid jet bursts in an air pocket which is trapped
at the top of a vertical pipe before filling a reservoir. The impact of the liquid
packets on the walls is particularly loud and is translated by dark regions on the
sonogram. When the air pocket is purged, all that remains is the sound produced by
the turbulence in the water which is much less noisy, and the sonogram is thus a
relatively uniform shade of gray.
Figure 7.17. Sonogram of the noise of a flow of water with

air pockets downstream a tap
In general, the emission of sound in a flow is always associated with unsteady

structures. This is a frequent nuisance, easily identified by an attentive ear, that we
seek to eliminate. The analysis of flow noise with the signal processing techniques
discussed above, and in particular time-frequency analysis, often allows the
identification and improvement of noisy regions of a flow. These procedures merit
broader use, but they require relatively complete knowledge in the domains of fluid
mechanics and acoustics.
7.3.6. Discretized signals
7.3.6.1. Evolution of techniques

While physical quantities are represented by real or complex numbers with
continuous values, the results of experiments are truncated decimal approximations,
which are subject to the uncertainties of the experiment. Since the earliest scientific
developments, it is the experimenter himself who discretizes the values of the
measurements he performs. In the same way numerical applications of the
equations, which result from exact or empirical theories, can only be used with
discretized, truncated numerical values. The introduction of computer technology
has completely modified both experimental techniques and the practice of system
modeling.
The evolution of experimental and measurement methods has been marked by

two types of devices, depending on the nature of the electronic treatment:
– analog devices in which the input signal (voltage or electric current) is a
continuous function of time which is then transformed by electronic circuits into an
electrical output signal, which is itself a continuous function of time; in these
devices, mathematical discontinuities don’t really exist. These measurement devices
generally have a relative accuracy associated with the scale chosen for the
measurement;
– digital devices, which use information technology, and operate on signals
constituted of a finite sequence of truncated numerical values (which are encoded in
a number of “bits” or “bytes”). Here, once again, digitization errors (quantification)
are in direct relation with the maximum value chosen for the representation of
numbers (12 or 16 bits for example).
The interest of digital techniques lies in the ease of adaptation and the near
infinite possibilities: any modification in the treatment of a signal is performed
through the modification of a computer program, on the contrary, the modification
of an electric analog circuit can only be achieved through modifying the physical
properties of the components, some of which may be variable by construction
anyway.
The techniques used in practice are often mixed for different reasons:
– as the initial signals are often continuous, it is necessary to construct a table of
numerical values which we wish to retain by sampling (this is the job of the analog-
to-digital converter);
– the signals which result from a treatment are often destined to be used to act on
analog materials which require a certain power (power amplifiers, actuators,
loudspeakers, etc.); the discontinuities of the numerical values must therefore be
interpolated by means of a digital-to-analog converter which delivers a continuous
electrical signal.
7.3.6.2. Sampling of continuous signals

The representation of continuous signals by tables of numerical values is only
useful if the initial signal can be reconstructed exactly without any loss of
information. We replace the continuous signal by a sequence of values that are
generally recorded at regular intervals. It is clear that the signal should not have
varied too much between two successive values, such that its reconstruction in a
continuous form can be performed with suitable precision.
The temporal representation of a signal s(t) amounts to its decomposition into a

basis of Dirac functions according to the following property:
³f f W G t W dW
f
f (t )
Sampling a signal s(t) comes down to multiplying it by the sampling function

_ _
I I I T (t ) (also known as Dirac comb or impulse train) defined by the sum of Dirac
impulses positioned at the points kT (k integer):
k f
I_I_I T (kT ) ¦ G t-kT , k integer
k f
We thus have:
k d d
s(kT ) ¨
k d
d
tCkT
s(t)dt, k integer
The sampling of temporal signals is a complex operation ([BAH 01], [BEL 00],
[CAS 06], [HIG 93], [MAD 98], [MAX 96]).
7.3.6.3. Fourier transform of a discretized signal

The sampling function is a periodic function of period T. Its development as a
Fourier series can be immediately written by calculating the Fourier coefficients
(section 7.3.3.1) of the Dirac function on interval [-T/2, T/2]:
t
k f 1 k f 2S j T k
I_I_I T (t ) ¦ G t-kT ¦ e k integer
k f T k f
Using the previous second expression of function I_I_I T (t ) and results from
section 7.3.4.2.1, it is easy to obtain:
1 k f 1
F I_I_I Q ¦ G (Q kf 0 ), k integer, f0
T f 0 k f T
For a discretized signal of finite duration, which can be written as:
k N f
s (kT ) ¦ ³ f G t-kT s(t )dt , k integer
k N
the Fourier transform is equal to:
t
k f 2Sj k 1 k f
Fs. I_I_I ¦ Fs * F ( e T ) ¦ Fs Q * G Q - kf 0
T k f f0 k f
1 k f
¦ Fs Q - kf 0 k integer
f0 k f
The Fourier transform of the sampled signal is the transform of the signal s(t)
which is completed by a periodization on the frequency axis, with a period equal to
f0 = 1/T.
The other integral transformations studied earlier can also be applied to

discretized signals. The essential physical notions already discussed for these apply
equally to the discretized signals; of course it is necessary to take account of the
numerical aspects associated with the discretization ([BEE 03], [BEL 02], [FLA 98],
[JER 92], [PRI 91]).
7.3.6.4. Errors associated with digital techniques

The digitization of analog signals can be performed by means of devices
comprising a fixed number of digits (8, 12, 16, etc., binary digits). A quantification
(or rounding) error results, which is all the greater as the number of significant
digits is small. For example, the representation on 12 bits allows the representation
of 212 (=4,096) numerical values; thus a relative precision far greater than 1/1,000
cannot be hoped for, and this is on condition that the 12 bits are nearly all used. For
example, if the 12 bits allow the representation of numerical values between 0 and
100, the absolute quantification error of a given number is in the order of 100.2-12,
i.e. about 0.03; only 5 and 6 bits will be used to represent the number 1.378,
corresponding to a relative accuracy of the order of 2%. The preceding error is
similar to the reading error of measurements on analog apparatus, for which it is
necessary that the quantity to be measured is near of the full scale used.
The errors associated with discretization are greater in proportion to the

evolutionary rapidity of the phenomena as in impulse signals. In practice, an
impulse is applied over a certain duration; if this duration is in the order of a very
small number of temporal points, the measurement error may be very great. This
loss moreover corresponds to irreversible information loss.
In section 7.3.6.3, we saw that signal discretization leads to the periodization of

the Fourier transform of continuous signal. In order to reconstruct this initial signal,
it is thus necessary to conserve one period of the spectrum, and this assumes that the
Fourier transforms of each period do not overlap: if N is the sampling frequency, the
signal spectrum should not contain frequencies greater than N/2. This result
constitutes Shannon’s theorem.3 In these conditions, by taking the inverse Fourier
transform of the central period of the spectrum, we exactly reconstruct the initial
discretized signal without any information loss.
The result of this, given that any signal will contain parasite noise of various
origins and different characteristics, is that it is necessary to suppress the higher
frequencies vis-à-vis Shannon’s theorem. This is the role of low-pass filters
positioned before the analog-to-digital converter; these filters are called anti-aliasing
filters (as they prevent energy at frequencies higher than the Shannon limit from
appearing in the lower frequencies, this being due to the partial overlapping of two
successive periods of periodized spectrum). We cannot discuss these problems in
greater detail; the interested reader should refer to texts on signal processing.
3 A usual and equivalent statement of Shannon’s theorem is that a signal comprising only
frequencies up to p Hz must be sampled at 2p Hz at least, so as not lose any information.
7.3.6.5. Discrete transformations of discretized signals

The transforms described above operate on signals with continuous or discrete
values, but they provide continuous values. These need to be discretized in order
that they can be stored or transformed numerically. In most practical cases,
numerical calculations of integral transforms are performed by means of algorithms
applied to discretized signals that directly provide discrete values.
As a result of binary representation of numerical values, transform algorithms

can be greatly simplified for ensembles of values whose number is a power of two.
The fast Fourier transform is performed by means of a fast algorithm proposed by
Tuckey. Any signal s(t), which is discretized into 2n values, will have a discrete
Fourier transform that contains as many values in the frequency domain as the
discretized temporal signal contained. There has thus not been any data reduction
compared with the initial signal. However, for a real signal, on account of the
properties of evenness, the number of significant values is equal to 2n-1.
Furthermore, the discretization of an unknown signal can create difficulties.

Consider a temporal signal composed of p narrow impulses. Its analysis and
representation are clearly quite simple in the continuous temporal domain. However,
the discretization obtained by temporal sampling will constitute a rather poor
representation of the signal s(t) if the impulses do not occur at the measurement
instants. On the other hand, the Fourier transform of the discretized signal is a sum
of complex exponentials the properties of which are not immediately obvious.
This difficulty is also encountered in the frequency domain for the identification
of one or several isolated peaks (harmonic oscillations) if the central frequencies of
the peaks are not equal to one of the frequencies of the sampled spectrum obtained.
It is for this reason that if we want an accurate measure of a spectrum’s peak
amplitude (a sound spectrum for example), it is important to choose window
functions which broaden the peak, while at the same time suppressing the secondary
lobes of the Fourier transform of the window ([BEL 98], [HIG 93], [MAD 98]).
7.3.7. Data compression
Files containing raw numerical values can be extremely large, even after
transformation. Images for example, in comparison with sound and text, consume an
enormous quantity of data when digitized. Experimental modern measurements also
produce large quantities of numerical data. It is thus necessary to reduce the size of
the files for storage by exploiting the power of processors, and it would not be
appropriate to discuss signal representation without recalling certain specific
procedures of data compression. This is achieved by exploiting particular

properties of data that result from arithmetic or numerical properties of the signal
studied or of one of its transforms.
Data compression procedures may or may not be accompanied by data loss.

Their efficiency can be evaluated by means of a degree of compression (volume of
the compressed file/volume of initial file). These are classed according to two main
categories depending on the arithmetic or analytic nature of the characteristics that
are exploited.
7.3.7.2. Arithmetic methods of data compression

A sequence of numerical values often presents particular combinational
properties: for example, real numbers have a periodic decimal development. Any
data file comprises a sequence of numerical values, which is not the result of pure
chance, as they always represent some specific information. This implies that the
file belongs to a certain class of files and it therefore possesses certain arithmetic
particularities, which can be demonstrated. For example, an image is not made up of
numerical values chosen at random – it includes certain structures (contours, color-
ranges, etc.) which correspond to the images it represents. An image constituted of
random pixels will most likely not represent anything at all. Similarly, a sequence of
letters chosen randomly has a very small probability of representing a text. To
clarify this, consider the following: the number of permutations, without repetition,
of the 26 letters of the alphabet (equal to 26!) is of the order of 4 × 1026. If we admit
repetition, the number of combinations is much greater. It is clear that this number is
far greater than the number of meaningful sentences comprising only 26 letters that
it is possible to write in a given language. Text files thus form a particular class of
files.
The combinatory particularities of a class of files can be used in order to define

appropriately adapted representation conventions. It is therefore possible to estimate
that such an encoding of a file will allow the reduction of its size. Of course, the
reverse decoding operation must be possible without any ambiguity.
Arithmetic methods of compression involve searching for numerical structures in

a sequence of values of a file and exploiting the multiplicity of their occurrence. Let
us consider three examples:
– the method of repetition involves “factorizing” the sequences which are
repeated one after another; for example, a sequence of 30 identical values (96 for
example) for consecutive pixels of an image will be denoted 96*30 instead of (96,
96, 96, etc.); a suitable convention for writing the file is obviously necessary;
– the dictionary method involves recognizing the structure of values which are
repeated and are thus indexed; the name “dictionary” derives from the fact that a
text is not a random sequence of letters, but rather a sequence of words which we
sort and store in a dictionary, indexing them by order of their appearance in the
dictionary. The compressed file is thus the dictionary, and the text is encoded with
the order numbers of the words used. While the words of a text are easy to read,
numerical structures (sequences of similar bytes) of an image file must be sought
with a suitable algorithm. It is also possible to establish a partial dictionary by not
encoding isolated values, which are not recognized as part of a repeated structure.
This method is applicable to all types of files.
– the Huffman method, which is entirely statistical, is based on the fact that in
language, all the letters are not used with the same frequency. In French, for
example, the probability of encountering the vowel “a” is 17.3%, whereas that of
encountering the consonant “w” is 0.05%. Now, letters are encoded on 8 bits
(ASCII characters). In general, a byte file contains variable occurrences for the
different bytes which are possible, while the Huffman method consists of encoding
the bytes encountered in a source file with variable binary lengths such that the most
frequent data are encoded on a very short binary length, rare bytes being represented
by a binary length which is greater than the average. The few bits lost on the rare
bytes are quickly recovered for the more frequent bytes (“a” is 346 times more
frequent than “w”). As the number of bits encoded is now variable, it is necessary to
establish a criterion that allows us to distinguish between successive encoded
elements. The encoded file will finally comprise the used source code file and the
encoded message. Its establishment requires the implementation of a suitable
algorithm; data reading, in other words the reconstruction of the initial file, is
performed by means of a decoding algorithm (decompression).
The Huffman method is applicable to all kinds of file (text, image, music, etc.)
since it can establish a table of byte frequencies when the file is read. Despite its age
(it dates from 1952), this method remains competitive, as research has improved its
capacity to compress data.
All of the above methods of data compression are no-loss methods, as it is

possible to completely reconstruct the initial file. They do not use any underlying
“physical” property of the file structure, the algorithms detecting the structure of
repetitions in a purely logical manner. A suitable data compression code leads to a
reduction of the file volume. Its efficiency is related to the degree of repetition of
the file entities (bits, bytes, structures, etc.).
7.3.7.3. Analytical methods

Let us note first of all that the representation of a signal by an analytical formula
can be considered as signal encoding, its decoding being performed by numerical
calculation with formulae used for analytical representation. However, in most
situations, the signal is constituted by a sequence of numerical values that cannot be

represented by an analytical formula.
Any kind of understanding of the physical structure of the phenomena

represented can allow the identification of numerical particularities.
Transformations, which are adapted to the physical processes encountered, will
reduce the amount of data significant for that phenomenon: for example, the Fourier
transform of a harmonic function only gives non-zero values for the amplitude and
phase for one frequency. The broadening due to the finite length of the observation
window will increase the number of values in the vicinity of the signal frequency. In
these conditions it suffices to retain the data with sufficiently high values, by
adopting the convention that frequencies which are not retained have zero
amplitude. This manner of writing the results of the Fourier transform constitutes
data compression.
MP3 (Mpeg Audio Layer 3) encoding for sound files is based on a

psychoacoustic model and uses 10 to 12 times less data than a standard sound file.
Recall that one second of stereo sound on a CD comprises 2*44,200 values (the
sensitivity of the ear is 0-20 kHz, the sampling frequency which must be respected
(Shannon) is greater than 40 kHz). This encoding is a little destructive, but this loss
is nearly imperceptible to the human ear:
– we eliminate the sounds of a sequence which will not be perceived by the ear,
the frequencies being close to those of the dominant sound whose energy is much
greater (“masking effect”);
– we also eliminate, by means of Fletscher and Munsen curves which determine
the perceptual limits of the ear, all those sounds for which the level and frequencies
are weaker than the values of these curves: thus for an average level, the sensitivity
of the ear is maximum between 1,000 and 5,000 Hz; it decays strongly below 300-
400 Hz, and it also decays from 8,000 Hz. Sounds outside a certain range are thus
eliminated;
– in the case of stereo recording, we compress the data in mono in the low
frequencies, the difference in phase of the low frequency sound between the two
ears being so small as to be imperceptible to the listener.
– finally we use Huffman compression, without loss, which associates an
encoding which is shorter in proportion to the frequency of data sequences. The
single use of this method provides 20-25% of the compression.
the source signal is also decomposed into sub-bands during the Fourier
transform. The psychoacoustic model is used in these sub-bands which are
quantified by thresholds. The assembly of the sub-bands is then realized.
Data compression is thus based on quite general properties [SAY 00]) either due
to purely arithmetic statements, or to some general structural property, such as the
oscillatory phenomena for music, speech data, or the specific occurrence of letters in
a text. Obviously, a Fourier transformation will not have any value for compression
of text file data.
7.4. Choice of representation and obtaining pertinent information
7.4.1. Introduction
Knowledge is structured by successive levels of systems which are more or less

interlinked, and to each of which there corresponds a category of properties, and
whose properties are attributed at each level in order that they can be synthesized at
the next level. Epistemologically speaking, these are elaborated by means of a
“system-analysis” methodology, which is discussed in Chapter 8. It begins with
formal logic and extends as far as the science of living systems. The idea of
pertinent information is thus difficult to define in an absolute sense, as it depends on
the context and the objective of the analysis performed. We will position ourselves
here at the level of the physics of systems of continuous media in flows, acoustic
and transfer phenomena included.
The representation of physical phenomena, arising from simulation or

experimentation, now consists of tables of numerical values that contain all the
corresponding information. Let us assume for example that we have a recording of
the instantaneous velocity modulus of a turbulent flow at 20 points and with 400
measurements per second for 1 minute. We obtain an ensemble of numerical values
containing 40 x 20 x 60 = 48,000 values. The reading of these values is of little
interest to the human mind. At most, we might note that the velocity varies over
certain intervals, and that these values seem to be associated here or there as broad
groups. The graphical representation of the velocity values as a function of time
provides curves whose reading may provide some further indications to the trained
eye, but the information content thus recognized will appear rather small. For at
least a century, we have observed turbulent fluctuations, and despite this, the
science of turbulence remains incomplete.
The choice of representation obviously depends on the objective that is in view.

The detailed reproduction of a signal can be realized from a numerical recording,
eventually using a lossless compression of the data. We will tolerate some small
losses for the approximate reproduction (MP3 procedures for music, JPEG for
images, etc.). We may eventually choose a variable step size in the discretization so
as to suitably represent the rapid variations of phenomena in different zones (in
certain compression procedures, or in finite element calculation).
A signal may similarly be studied with a view of its analysis in order to

recognize certain global properties that are represented by synthetic information. An
integral transform does not generally lead to an economy in the numerical
representation of the initial signal; this is only possible by means of specific
transforms which allow an adapted representation which simplifies the presentation
of information (Fourier transforms, etc.). An integral transform is only of interest in
cases where the physical properties of the signal are analogous to its kernel (section
7.3.4). An adapted representation of a signal thus results from a knowledge of its
properties and, as this is often achieved thanks to a suitable form of representation,
we find ourselves confronted by certain difficulties. Hence, we have to find the
answer to a question which is always phrased in a manner corresponding to our
known concepts. If the answer needs other concepts unknown to us, it will be very
difficult to find in what direction the solution is: the human mind is progressing
from a known place to a nearly known place. In other words, we should never forget
that we will only find what we set out to find, and we only ever seek with ideas
which we know! Could Galileo imagine or ever understand the laws of mechanics in
the 17th century without the knowledge of the derivative?
We will study signal-analysis problems and information processing in the case

of audible sound signals, which are less complex than turbulence, at least in
appearance.
7.4.2. An example: analysis of sound
Sound amounts to pressure fluctuations that propagate in a fluid medium. Their
origin may be due to internal fluid mechanisms (aerodynamic sound for example),
or to the vibration of solid surfaces which are in contact with the fluid, or to the
interaction of fluid with solid surfaces. The structure of a sound field depends both
on the production mechanism and the conditions of propagation. Sound signals are
complex and very often contain important information whose scientific analysis can
be difficult. We will here limit our discussion to audible sounds whose spectral
content is contained between 20 Hz and 20 kHz.
The complexity of an audible sound signal results from Shannon’s theorem,

which requires at least 40,000 values per second in order for the sound to be
characterized. The reader can calculate the (very large) number of sound sequences
that are possible in one second by assuming that each numerical value is encoded on
8 bits.
Audible sounds have the particularity of sending the human brain a signal with
which a sensation is associated; this is then compared to memories and eventually
interpreted. It is clear that these mechanisms of recognition are the consequence of

training with respect to the usual sounds of our environment, which constitutes only
a very small part of the ensemble of possible sound sequences. It is precisely
because we already have an intuitive understanding of these sounds that we will
take them as an example for discussing signal analysis and synthesis methodologies.
The interpretation of audible sounds by the brain concerns the domain of

psychoacoustics. We will not discuss the structure of the human hearing system. For
questions associated with acoustic signals, it is sufficient to note that the response of
the ear is non-linear, sensations being moreover recorded on a logarithmic scale.
Now any action of a linear filter on a linear combination of harmonic signals leads
to another linear combination of harmonic signals of the same frequencies. On the
other hand, any non-linear operation on a linear combination of harmonic signals
of discrete frequencies will create new frequency components (harmonics and sub-
harmonics); for example, consider the signal s(t) which is the sum of harmonic
signals of frequencies 2n and 3n ( Z 2S n ):
s (t ) acos 2Z t cos 3Z t
Its square s2 can be written:
ª 1 º
s 2 t a 2 «1 cos Z t cos 4Z t cos 6Z t cos 5Z t»
¬ 2 ¼
The signal thus obtained does not contain the initial frequencies 2n and 3n which
are replaced by the harmonics 4n, 5n, 6n, and the sub-harmonics of the frequency n.
However, the spectral composition in relative value is here independent of the
amplitude a. In general this is not the case, and the global amplitude variation of a
sound signal modifies the spectral content as soon as the non-linear operation is not
simple. The reader can easily verify that the spectral composition of the signal
s 2 D s depends on the constant value D.
This leads to the human ear hearing frequencies that do not have any physical
existence: this is the question of the “missing fundamental”, the ear hearing the
previous sub-harmonics ([BER 90], [KIN 82]). Navier-Stokes equations being non-
linear, turbulence evolution is a strongly non-linear dynamic process in which
frequencies associated with turbulent fluctuations are increasing (turbulent energy
cascade [MAT 00]).
7.4.2.2. Hearing and time-frequency analysis

As the ear has a spectral response that is rather logarithmic for sound levels, it is
not a good instrument for evaluating the spectral content of a sound. The spectral
content of sounds perceived by the human ear does not vary significantly if we
modify the sound level; this can be easily verified by listening to music on a high-
quality system (except at very low sound level). However, the elementary
calculations of the last section show how the ear can hear sounds whose frequencies
do not exist in the sound spectrum. This phenomenon has been known for a long
time in music, where suitable harmonic combinations can lead to the belief in the
presence of a fundamental frequency that does not exist. The interested reader can
see texts treating psychoacoustics and musical acoustics. Fourier analysis of sound
differs from the perception that the ear can have of sound; in what follows we will
leave aside the problem of sound sensation, and we will limit our results to the
physical analysis performed using time-frequency techniques.
7.4.2.3. “Natural” sounds

We will designate by the term “natural” sounds all those sounds that are
generated mechanically in our environment. These are produced by vibration of
solid bodies (plates, shells, membranes, etc.), by oscillations in fluid velocity
(musical wind instruments, speech, the wind, etc.) or by interactions between solids
and fluids (wavemakers, vibrating walls under the influence of a flow, etc.). These
sounds result from the properties of the movement of fluids and solids. There is
incidentally no physical difference between musical sounds (which are in principle
agreeable to the ear) and industrial sounds that are often a nuisance. The two
categories of sound are produced by means of impacts (percussion instruments, a
hammer, etc.) friction (violin, squeaking of brakes, etc.), by airflows (flute, pipes
which blow, speech, etc.). These properties of natural or forced vibration are quite
well known in many relatively un-complicated instances.
“Natural” sounds are thus particular categories of sound signals whose

function or utility is quite varied:
– familiar sounds are part of the environment and the context of normal life; any
modification of these is immediately perceived as new information; the absence of
any sound can quickly become oppressive, as we can experience by spending time
in an anechoic chamber;
– suitable musical sounds have a relaxing and agreeable effect, which may vary
depending on the individual;
– sounds emitted by a sound source allow the identification of the position and
nature (at least partially) of its source; an anomaly in the content of a sound can
serve to identify an anomaly in the functioning of the source (for example, listening
to the sound of the engine of a car or of an industrial process): we have here a
diagnostic function which is beginning to be used in certain software of preventive
maintenance ([BOU 98], [WAN 06]);
– finally speech allows the transmission of information between individuals in

the form of language.
The scientific problem which is posed is that of defining a methodology for

obtaining the numerical characteristics corresponding to sound signals. The
understanding of mechanisms leading to these characteristics (phase of analysis)
will allow the subsequent implementation of devices that permit the synthesis and
control of the natural signals (music, speech, etc.). We will examine here the case of
musical sounds, which are more “standardized” and better understood than
industrial sounds, before describing the kinds of field quantities encountered in
flows.
7.4.3. Analysis of musical signals
Consider the example of a piece of music one minute long; we know that the
complete reproduction of the sound signal perceived by the ear of a listener needs to
be sampled at least 40 kHz, in order that no information be lost (for high-definition
listening for example). A minute of music is represented by at least 40,000 x 60 =
240,000 numerical values. It is this ensemble of values that we record on to an
ordinary audio CD. In fact, we record two channels in stereo (one for each ear) and
we sample at 44.2 kHz instead of the 40 kHz necessary according to Shannon’s
theorem. From the scientific point of view, such a numerical table can be considered
as a complete musical score.
The listener or the musician does not perceive this table as such, but he feels the
impressions, which do not really correspond to such a quantity of information. The
reaction time of the brain is much less than the sampling time of a piece of music, as
a consequence of which a certain number of modifications to the numerical table of
values can be made without the listener noticing. In fact, the listening apparatus is a
natural receiver which is adapted to the reception of ambient sounds: acoustic
vibrations are perceived by the organs of the inner ear, which transforms these into
electrical signals that are transmitted to the brain, via an analog measurement
“device” (or rather “evaluation” device). The brain performs an analysis of the
sound signal and deduces information (origin and causes of the sound, etc.); this is
an expertise which is based on pre-training (memories of similar sounds already
encountered) and the brain tries then to identify the global sound structures in a
complex ensemble and to compare these to “known” sounds. We thus recognize the
music of a piano, the firing of a cannon, the sound of a train, etc. We will now
describe the physical mechanisms of musical sounds and the analysis methods
which allow us to characterize these.
7.4.3.2. Instruments and sound structures

In fact, each sound structure comes from an individual mechanical system,
which we call an instrument in the domain of music. The production of “natural”
sounds is a complex phenomenon, which assumes the creation of a vibrational
energy and its transformation into sounds that propagate through the atmosphere. It
is interesting to study these in order to better understand the links that exist between
the physical mechanisms to be analyzed and the analysis tools that need to be
implemented.
Since the origin of humanity, sounds have been emitted by mechanical

vibrations produced by bodies in motion. Musical instruments, the human voice,
natural sounds due to the wind or the flow of water, etc., each constitute what can be
termed a mechanical musical instrument in which a form of a more or less
continuous mechanical energy is transformed into sound energy.
Sound is thus a “by-product” of a mechanical system in which occurs a

transformation of mechanical energy into vibrations, often highly complex, and
which are localized in a region of restricted dimensions that we might designate as a
primary acoustic source. It is for example the contact zone between a solid and a
body which strikes it, the flow region behind an obstacle where vortices are
generated, the contact zone between a wheel and the road or a rail, the contact zone
between a bow and the string of a violin (or between a brake pad and disc), etc. The
musician acts essentially in this zone by producing an impulse (percussion
instruments), a continuous movement or a continuous airflow which produces more
or less periodic vibrations (emission of vortices, relaxation oscillations in bowed
string instruments, etc.). This primary source often has a highly non-linear behavior
which varies in time. It may also be periodic (imbalance in wheel rotation or purr of
a transformer for industrial noise, etc.) The oscillatory mechanical energy created is
essentially localized here and only a small part of this is transformed into acoustic
energy.
Let us now take the example of traditional classical music. The primary acoustic
source excites the rest of the musical instrument, which is generally larger, and whose
role is to “filter” the excitation, in other words to transform it without creating
additional vibrational energy. The resonant parts of the instrument are the apparent
acoustic source for the listener, which can be referred to as secondary source (Figure
7.18a). These allow the localized oscillatory energy to be transformed into acoustic
energy that propagates through the air (in fact we are dealing here with an impedance
adaptation mechanism (see horns in [KIN 82]). Furthermore, we know the importance
of certain construction details of a musical instrument for the quality of the sound that
is obtained. The essential role of the instrument is to provide a very weakly damped
filter which supports oscillations of very small amplitude and the equations for which
are thus linear with coefficients which are independent of time, at least as long as the
instrument geometry is not changed.
From a mathematical point of view (Chapter 5), the main mechanical energy of a
wind instrument is contained in the part of the solution associated with the
convective characteristic curve, the sound corresponding to the propagative
characteristic curves.
Then, the listener listens to these sound vibrations in a given environment, which
also possesses particular reverberation properties. Finally, the vibrational
characteristics of the energy-creation zone are far from being the same as those of
the sounds which we hear: the final sound signal which results from these
successive operations, between which there may be retroaction, at least so far as the
primary and secondary sources are concerned. More exceptionally, this retroaction
may exist between the surrounding environment and the primary source, as seen, for
example, in the Larsen effect between a loudspeaker and a microphone connected to
the same sound system.
listening room nasal

passage
Lips buccal passage
pharynx
microphone
larynx and
vocal chords
primary acoustic source Secondary
(Acoustic energy creation) acoustic sources
(a) (b)
Figure 7.18. Acoustic mechanisms: (a) emission of the sound of a violin

in a room; (b) speech emission
The human voice is a musical wind instrument (Figure 7.18b). The initial
mechanical energy of the airflow coming from the lungs produces an oscillatory
energy in the larynx in the vicinity of the vocal cords (the primary acoustic source)
which comprise vibrating obstacles. The secondary source in contact with the
outside environment is an “acoustic filter” comprising the nasal and buccal passages
and the pharynx. Important geometric modulations with slow variations are
produced by movement of the lips in particular, and of the tongue, and this
constitutes a considerable complication compared with the instrument that produces
fixed sounds. The acoustic filter thus realized is not representable by a linear
constant-coefficient differential equation system.
To summarize, musical signals are comprised of combinations of weakly damped

sinusoidal functions and by modulations, which are more or less variable at a
timescale which is greater than the period of the emitted harmonic signals. They
thus constitute a particular class of intermittent signals which correspond to the
eigenfrequencies of the acoustic system.
7.4.3.3. The importance of harmonic signals

Musical instruments produce periodic oscillations which are more or less
variable (amplitude, phase and frequency variations, etc.). As a periodic function is
decomposable in a Fourier series, sinusoidal functions play an essential role in
temporal representation. These are also found in different domains of physics
(electricity, electromagnetism, optics, etc.). The analysis of periodic and harmonic
signals has been used for over 150 years in the domain of physics. However, the
study of speech has only become possible with the use of devices which allow a
graphical representation of the amplitude of sound vibrations as a function of
frequency and of time (the first sonograms appeared before the 1950s).
Let us return to our musical signal, represented by 44,200 numerical values per
second. We need a means of easily recognizing functions that are a priori not so
different from harmonic functions, which are in fact wave-packets. Time-frequency
analysis (section 7.3.5) is thus well adapted to this interpretation. As each note of a
musical instrument is quite well characterized by its pitch (fundamental frequency)
and its harmonic content, it will appear on the sonogram as an ensemble of parallel
bands with respect to the time axis. Figure 7.13 shows this, the human voice being a
musical instrument, which is slightly complicated by its variable geometry. This
representation shows the separation of two phonemes “sh” and “ah”, which is not
truly visible in a temporal representation of the corresponding sound.
A musical signal is thus illustrated in the time-frequency domain by geometric

structures whose forms are associated with physical characteristics and whose levels
are given by a grayscale or a system of colors. In general, these forms vary very
little as the note is played (fundamental frequency) or the sound level is changed.
We can thus say that this global structure is characteristic of a musical instrument.
7.4.3.4. Applications to the synthesis of musical signals

7.4.3.4.1. General principles
The information which characterizes musical signals is encoded in the occidental
world by means of musical scores: it is classically characterized by the tempo
(duration of a black note), the notes to be played, the names of the instruments used
by the musicians, the indicated sound level (forte, piano, etc.) and some indications
of temporal variations (crescendo, etc.) and interpretation (glissando, rubato, etc.).
The data of the score are used by composers in order to write the music and for the
musicians to interpret it; the listeners hardly recognize any more than this. The
number of numerical values that can be encoded in the musical score of an orchestra
for a second is quite restricted (at most a few dozen).
On the other hand, the preceding time-frequency analysis allows the

quantification in a musical sound of:
– the characteristics and the general structure of a musical instrument
characterized by its harmonic content (its timbre);
– the note played, characterized by pitch and duration, this being noted explicitly
on the musical partition;
– the interpretation of the musician (sound intensity, eventual variations of
frequency) indicated more or less completely on the score.
We will now examine how these ideas can be used in order to reconstruct pieces of
synthesized music, which amounts to the reconstruction of a musical signal that has
been stored in a compressed form. The procedure is here very different from data
compression techniques described earlier, as we directly use the structure of the
musical sounds, by reproducing in a digitized form the interpretation of a musical
score with synthetic instruments (a “virtual” realization). The score and its
interpretation are realized in a MIDI file, the instruments being, as in reality, realized
independently of the score in distinct analog or digital modules (synthesizers).
7.4.3.4.2. The MIDI system

The MIDI (“musical instrument digital interface”) system is essentially a tool for
the control and management of information allowing the control of musical
instruments such as samplers or synthesizers, which are integrated into computers in
the form of cards. This amounts to the transcription into computer language of an
ensemble of data which is characteristic of a musical score with its interpretation
and diverse additional parameters. The MIDI system also allows the control of
mixing consoles, effect processors and recording systems. An international norm
defines the MIDI system with certain extensions and variants ([ROT 95]). We will
here limit ourselves to a description of the principles.
The MIDI system is a serial control interface that was originally dedicated to
musical systems. However, it rapidly went beyond its original vocation and can
today be found in the control of a broad range of material that is not only audio, but
also dedicated to theatrical lighting. The MIDI protocol integrates numerous control
parameters, most of which can be freely manipulated by the user depending on his
needs. It is possible to control polyphonic musical instruments in pseudo-real time:
transmission times are imperceptible in most cases. It is also possible to address
numerous devices using the same MIDI data.
The basic data of the system are messages, which comprise:

– a status (engagement and release of a note, “sustain” pedal, modulation,
polyphonic pressure, continuous control, change of program, weak variation of pitch
(“pitchbend”), etc.);
– data characterizing the action indicated in the status (note played, amplitude of
variation in level of pitch applied by the rowel of pitch variation, speed of key
engagement, etc.).
MIDI data are most often elaborated using an electromechanical interface for the
inputs, this being a keyboard which looks like that of a musical instrument (piano,
accordion, saxophone, etc.), and on which the user plays as if it was a real
instrument. The rapidity of the transmission of digitized data allows the control of
16 instruments quasi-simultaneously; we have in fact a single line crossing all of the
instruments played, on which the MIDI messages serve as indicators of which
device they are to be addressed to; each of these transmits all of the messages and
only takes account of the messages with which it is concerned (MIDI channel). The
temporal delay, which results from this in-series configuration, is small enough not
to be perceived by the listener.
A schematic of the MIDI system is shown in Figure 7.19; the data of a MIDI
partition can be written either by playing keyboards 1 or 2 or by writing the MIDI
files directly from a computer. In the same manner sound restitution can be obtained
directly from the keyboards, or off-line from a MIDI file (.mid file extension mid).
The MIDI instructions, which define the sound signals, are then transformed into
continuous electrical signals by sound generators (synthesizers).
The reader will very likely find a MIDI file reader in the accessories of his
computer (Windows Media Player, etc.), associated with a numerical sound
synthesizer. It is easy to verify that the volume of such files is quite small (e.g. 5
kbytes for one minute of music, whereas the same musical piece sampled at 44.2
kHz contains over 200 Mbytes).
sound signal
sound
generator 2
MIDI keyboard 2
MIDI score
or recording sound
generator 1
MIDI keyboard 1
Figure 7.19. Diagram of links of the MIDI system
MIDI constitutes an economical means of controlling and storing sound

information. We take advantage of this information link for the transmission of
other kinds of data (message systems). However, the greatest advantage of the MIDI
system remains the ease of correcting and editing MIDI scores, any parameter being
individually and quickly modified.
7.4.3.4.3. Synthesizing musical instruments

It now remains to study the analysis and synthesis of musical instruments, which
can be analogous or digital. We have seen previously that a musical instrument is
characterized by its timbre (harmonic content). We must add to this a temporal
variation of sound, which takes account of the evolution of sound amplitude
emitted, which depends on the nature of the instrument and its mode of excitation
(Figure 7.20): the sound of a wind instrument can last a very long time, whereas the
sound emitted by a chord which is struck (piano) or plucked (guitar) is essentially
transitional.
p(t) p(t) p(t)
organ t piano t guitar t
Figure 7.20. Time evolution (amplitude envelope) of the amplitude

of an organ, piano and guitar sound
In fact, synthesis, which is based on the realization of a fundamental and its

harmonics, is not sufficient for the ear to have the impression of a real instrument.
Even if the sound obtained seems similar enough to that produced by an instrument,
it is not identical to a note played from the real instrument. The reality is more
complex than the approach outlined earlier from study of the sonogram. The sound
emission is associated with the resonance of the instrument body, but also to certain
more complex transitional properties of the sound which depend on certain details
of the excitation: in the sound emission of a flute we first hear the sound of blowing.
Finally, weak modulations of frequency or amplitude can be produced for diverse
reasons related to the sound amplitude, the way of playing, etc. These can be
deliberate on the part of the musician (player or composer).
Let us quickly describe the parameters of the usual synthesizer. We define the
amplitude envelope by a small number of values indicating the durations T1 and T2
of the attack sound, the duration Tm of the sound established if it exists, and two
parameters T3 and T4 for the duration of the extinction of the sound; we associate
these values with an amplitude curve, which is piece-wise linear in practice.
a(t): amplitude
a(t),'1(t) envelope
'N(t): frequency
variations envelope
T1 T2 T3 Tm T4
t
Figure 7.21. Examples of amplitude and frequency variation envelopes
These parameters allow the representation of characteristics of two principal

kinds of sound which can be excited permanently (wind instruments, string
instruments: flute, organ, trumpet, violin, etc.) or by impulse (percussion
instruments or instruments with plucked strings: drum, piano, guitar, xylophone,
etc.). These also allow the characterization of transitional regimes.
It then remains to realize the spectral sound content sought. For the established
regime, the harmonic content can be obtained by different means: additive synthesis
using harmonics: superposition of signals which are rich in harmonics (triangle,
etc.) which we may eventually filter in order to only conserve the first harmonics,
subtractive or multiplicative synthesis, frequency modulation, etc. We can introduce
temporal variations of the harmonic or amplitude content (modulation) associated
with the envelope parameters and diverse effects. We will note that these are distinct
from variations related to the preceding envelope curves. We can also add recorded
sounds that are difficult to synthesize.
Another technique consists of recording the sound of an instrument and

eventually modifying it by filtering in order to improve it. However, the variation in
pitch of the note by the MIDI command can give imperfect results, as the sound of
the real instrument can vary during the duration of the note.
Finally, we obtain reasonable results for instruments of fixed geometry. The

result is less satisfactory for instruments where it is up to the musician to generate
the note (violin, for example), as the sound of the violin is harmonious only thanks
to small variations in frequency due to the musician, and that depend on the piece
played. This effect cannot be accounted for in the sound generator; it can
theoretically be included in the MIDI file through the introduction of a suitable
controller, but this risks notably increasing the volume of data.
7.4.3.4.4. Regarding the musical sound structures

Synthesizers and sound generators can also create new sounds by manipulating
the system parameters. We will note that variety music contains many synthesized
sounds. However, this synthesized music remains in the context of music, which is
measured and based on the equal temperament.
The preceding technologies do not allow the exploration of the vast domain of
electronic music that is different from the preceding context, being based on sound
structures which are not related to the harmonics of a fundamental frequency or the
modal frequencies. From this perspective, the first stage of research is to define
agreeable sounds, which can allow us to make the distinction between music and
noise. The reader can easily imagine the difficulties which may be encountered by
the composers of electronic music whose objective is to generate new sound
structures which do not result from the resonant properties of vibrating strings or
resonant cavities. Regardless of whether their sound is agreeable or not, it should be
possible to characterize them using a small number of parameters, in order to define
a musical notation of a new kind which would allow their representation by means
of a musical score. We are here dealing with a relatively unexplored domain, which
is well beyond the scope of this book. It is worth noting, however, that a first
difficulty is to succeed in specifying the domain to be explored and to find a concept
that can replace the Pythagorean basis of tonal music ([LIC 02]).
In conclusion, the time-frequency analysis concept has allowed significant

progress in the analysis of sound signals, in particular musical sounds, and also
speech analysis, recognition and synthesis, topics we could not discuss here.
7.4.4. Signal analysis in aero-energetics
We have examined (section 7.2) the different ways in which signals can be
represented and analyzed, these two notions being related as we have seen. An
exhaustive treatment of this subject is beyond the scope of this book. Many other
analysis and extraction procedures also exist: in particular, filtering procedures
consist of the extraction or suppression of certain components of a signal. For
example, we can extract a coherent signal from a random background noise; these
procedures are largely used, but it is always necessary to characterize what it is we
are looking for. The preceding example of acoustic signals has shown the links that
exist between physical analysis and signal processing techniques. However, the
local measurement of quantities is the result of modifications of the fluid medium on
the ensemble of its characteristic curves or surfaces (Chapter 5) which move with
the matter, or propagate “acoustically”.
7.4.4.2. Effects of turbulence

The solutions of the equations of fluid mechanics are very often unstable and
they present a chaotic aspect with random fluctuations (known as “turbulent”)
whose properties determine the flow properties, heat and mass transfer ([SCH 99],
[YIH 77]). A detailed understanding of turbulence is far from complete. On the
other hand, all linear measurements allow the user to obtain mean values of flow
quantities (velocity, pressure, etc.). The measurements using devices with a non-
linear response need to be treated prior to the application of statistical analysis or
integral transforms. The presence of turbulence may considerably hinder
instantaneous measurements, as we will see in the following example.
7.4.4.3. Separation of causes of phenomena in measurements

Classifying the phenomena encountered in flows and discussing the possible
interactions between the various physical mechanisms is not easy (sections 5.3 to
5.6). Obtaining relations or properties depends on the specific dominant phenomena.
An important particular case concerns low Mach numbers (velocities less than 100
m.s-1 in air) which correspond to air or water flow conditions in many industrial,
domestic or environmental problems. All fluids are compressible and we have seen
in Chapter 5 that modifications of properties are transferred either by convection or
by means of acoustic waves. The orders of magnitude of these two kinds of
phenomena are here generally very different: the acoustic component is very often
weak compared to the dynamic or thermal effects which result from the boundary
conditions. For example, acoustic variations of velocity are less than 1 mm/s,
whereas the velocities of the matter are very often between 10 and 100 m/s1. The
variations of temperature that result from dynamic effects are of the order of 1°C (at
about 100 m/s-1), whereas the smallest problem of heat transfer generally involves
fluctuations of some about 10°C. For example, a local, instantaneous measurement
sensor of velocity or of temperature characterizes phenomena associated with the
flow and the heat transfer depending on the circumstances (section 7.2.4).
On the other hand, the respective contributions G p a of the acoustic modes and
G p c of the convective modes to pressure variations may be of the same order. In
effect, we have seen (section 5.3.2) that we have:
– on the acoustic characteristic curve:
G p a | U cG u a
– on the convective characteristic curve:
G p c U VG V 0
The ratio of these pressure fluctuations is thus:
G pa c G ua 1 G ua
|
G pc V GV M GV
As the Mach number is small at low velocity (M ~ 0.05 to 0.15 in ventilation

problems for example), the factor 1 M is quite large and it compensates the small
acoustic velocity fluctuations compared to the turbulent velocity fluctuations. This
results in pressure fluctuations due to local turbulence effects, which are of the same
order of magnitude as acoustic pressure fluctuations.
A pressure probe (microphone) measures the local pressure independently of its

origin and without knowing on which characteristic curve it is transmitted: it is no
longer possible to distinguish acoustic fluctuations from those due to local velocity
fluctuations. This problem is well known to those who measure sound: a
microphone placed in a flow does not only measure noise. Listening to the pressure
signal acquired by a microphone placed in the wind we are immediately aware of a
parasite “sound” due to the wind. This “sound” is not of an acoustic nature, but
rather a signal that corresponds to local pressure fluctuations associated with the
flow created around the microphone. Measurement of sound in the presence of wind
requires suitable precautions in order to reduce the impact of parasitic phenomena
(for example, “anti-wind” foam covers on microphones).
The separation of turbulence and acoustic pressure components is an important

and extremely difficult problem, since it corresponds to Navier-Stokes solutions for
which the theory of characteristics curves is in practice not applicable. This
separation between acoustic and turbulence signatures can only be made possible if
additional information is available: for example the periodic sound due to a
propeller can be extracted from a signal containing random turbulence fluctuations.
7.4.4.4. The study of unsteady flows

The velocity, pressure or temperature fields associated with unsteady flows are
now accessible by means of experimental methods described in section 7.2.6.3; on
the other hand, the different methods of time-frequency analysis allows the temporal
and spectral characteristics of flows to be identified. These modern methods allow
us to consider the vast domain of experimental unsteady fluid mechanics, provided
that experiments can be performed in a reproducible manner (which means that
different realizations obtained are identical (section 7.2.6.3)). The identification of
unsteady flow structures and, in particular, the dynamics of vortex interactions
constitutes a domain of study which is relatively unexplored.
We will see that turbulence, being by its nature unsteady and broadly 3D, is
largely the result of interactions between inviscid fluid structures (turbulent energy
cascade [MAT 00]). The origin of turbulence is furthermore associated with the
development of flow instabilities issued from pre-existing perturbations. The
elementary mechanisms of these highly unsteady interactions are poorly understood.
For these reasons, we have focused on signal-processing and experimental methods
that can be used in a broad range of applications where unsteady phenomena are
encountered in possibly rotational flows These methods allowed for example the
acquisition of experimental data for the problem discussed in section 5.6.2.
Chapter 8
Thermal Systems and Models
This chapter is dedicated to the modeling of systems. The perspective developed

in automatic control and the corresponding methods are outlined. In addition to the
mathematical properties of thermodynamic balance equations and the measurement
and signal processing problems, these provide a general framework for the study of
the representation and evolution of physical systems. Using some basic heat
conduction problems in media at rest, the methods can be easily proved on account
of the linear properties that these present.
8.1. Overview of models
8.1.1. Introduction and definitions
In Chapters 1 to 4 we studied how to describe a physical system in the context of

thermodynamics. The equations for continuous media are hardly useable except for
very simple problems and any more complex system can only be represented by
making approximations, which it must be possible to evaluate. We will limit
ourselves in this chapter to state representation in which a given thermodynamic
system is represented as a group of sub-systems in instantaneous equilibrium. This
makes it possible to define the state variables of each sub-system. The partial
differential equations which represent the continuous medium are thus replaced by
partial differential equations describing the evolution of a restricted number of
variables. The general study of systems is derived from automatic control and signal
processing which provide, as in thermodynamics, general principles which cannot be
ignored. We will recall here the ideas that are necessary for the establishment of
models in fluid dynamics and heat and mass transfer, by limiting ourselves to the
study of thermal systems, whose damped character simplifies the discussion.
A system is a material ensemble that exchanges quantities with the exterior. We

act on the system(s) by means of inputs, which are imposed variables (a
temperature, a mass flow, pressures, etc.); these inputs lead to modifications of the
system which are manifest in observed and measured effects, which constitute the
outputs of the system. The inputs are therefore causes and the outputs are
consequences.
Inputs System Outputs
Initial conditions
Figure 8.1. Scheme of a system
A system is said to be causal (with respect to a variable) if the conditions which

determine its state and its outputs for a value of the variables x0 result from data for
values of the variable x which are less than or equal to x0. This idea of causality
simply translates the mathematical properties associated with the parabolic or
hyperbolic character of the system towards certain variables or groups of variables
of the model that is used. While it is obvious for the time dimension (section
1.1.1.3), this idea, which is used in automatic control and signal processing, can also
be applied to transport and propagation phenomena (sections 5.3.2 and 5.4.5.4).
The description and knowledge of a system can be achieved by means of a

model, which can only be established if the system is observable, in other words if
we can measure the data necessary for its establishment. This condition obviously
depends on the nature of the system and the objective of the modeling.
The control of a system is an action that is often necessary for diverse reasons:
control of temperature in a zone, of a flow structure, a sound level, etc. A system is
said to be controllable if, from any initial state of the system, it is possible to act on
certain inputs in order to bring the system to a desired state.
A system is said to be instantaneous if its state and its outputs at a given instant
depend only on the inputs at that instant. This definition can be applied to any
system whose sub-systems are in mutual thermodynamic equilibrium. A system is
said to be dynamic if it is not instantaneous.
Thermal Systems and Models 407
Systems exhibiting hysteresis present the particularity that the stationary solution
corresponding to fixed inputs depends on their history (state variables at the initial
state and the past evolution of the inputs). Let us take the example of an oscillator
comprising a mass P placed between two springs and subjected to a friction force on
a fixed wall. If the friction is caused by a fluid, the equilibrium position of the point
P is unique, the tension in the springs being therefore equal. In return the existence
of a solid source of friction in P means that the final position of the point P also
depends on the movement history (Figure 8.2a).
Flow systems with hysteresis are encountered not only with fluids whose
behavior laws involve a solid friction (Bingham fluid), but also for Newtonian
fluids. This phenomenon is thus associated with the existence of two possible steady
solutions for a flow with fixed conditions. This circumstance may be due to diverse
phenomena such as the feeding of a siphon (Figures 8.2b and 8.2c), reattachment of
a flow on a curved surface (Figure 8.2d and Figure 8.2e), the flow of water over a
weir depending on whether the sheet of water is aerated (or ventilated) or not
(Figure 8.2f and Figure 8.2g), etc. The observed solution can vary depending on the
manner we seek to achieve it by.
A B
P
(a) (b) (c)
water air water
(d) (e) (f) (g)
Figure 8.2. Systems with hysteresis – (a) springs with dry friction; set of 2 tanks with feuded
(b) or not (c) siphon; flows in a divergent pipe (d) and (e); weir with aerated water sheet (f)
or non-aerated water sheet (g)
In addition, unknown perturbations (noise) are always present to varying degrees

in these inputs and can have diverse effects (“instabilities”). The idea of a stable
system is difficult to define, because it depends on the nature of the applied
perturbations and the properties of the response of the system. Stability is a complex
problem and of extreme importance in fluid mechanics ([SCH 99], [YIH 77]).
A system is said to be stationary or time-invariant if the outputs corresponding to

a given set of inputs do not depend on the instant at which the inputs are applied.
The equations of a time-invariant system are not influenced by a change of the time
origin. Systems whose structure and physical properties only depend on the state
variables are time-invariant. The same is not true of systems whose parameters1
depend on external conditions which vary with time (for example, the flow of a
liquid in the process of polymerization, the action of an obstacle with an incidence
angle D(t) which is imposed in accordance with some external law, etc.).
A linear time-invariant system (LTIS), which is also known as a linear

stationary system (LSS), presents in addition the property of linearity: effects
proportional to the causes which produce them. These systems are described by
constant-coefficient linear models. Apart from some very specific cases, a system
with a non-uniform flow is not linear. The heat transfer of mass by diffusion is often
linear, even in flows, provided the temperature or concentration does not cause the
physical properties of the fluid to change. Coupling between different phenomena
and in particular chemical reactions in flows suppresses this linearity property. The
reader can verify these properties for the general equations in Chapters 1 to 4.
8.1.2. Modeling by state representation and choice of variables
In automatic control, state representation of a discrete system is a model of the

form [8.1] which represents its evolution:
dX
A. X B.U Y D. X E.U [8.1]
dt
The state vector X is a set of n variables which characterize the state of the
system. The state representation is also called the internal representation. This
definition is no different in principle to the definition which was given in Chapter 1
where extensive and intensive variables were used as state variables. However, in
automatic control the components of the vector X can be any variables which
characterize the state of the system. The matrix A is a square matrix of rank n. It
characterizes the make-up of the system and of the sub-systems which correspond to
the choice of the state vector X.
The vector U is the input vector; it is of dimension p and corresponds to different

actions on the system. It does not in general have the same dimension as the vector
X, and the matrix B is a rectangular matrix of dimension n * p.
1 I.e. the coefficients of operators in the equation; the inputs are not parameters of the system
here.
The vector Y is the output vector; it is of dimension q and it comprises the

different results desired for the problem posed. It does not in general have the same
dimension as the vectors X and U. In what follows, we will take E = 0 (the system is
thus called proper).
The representation of the state of a system is not unique as the variables can be
chosen in different ways or can be modified by changes of variable. Furthermore,
the components of the state vector can be related when we are limited to particular
operation regimes as we will see on numerous occasions in the remainder of this
chapter (section 8.3, section 8.5 and section 8.6). However, the systems studied in
this work are constituted of a matter, which obeys the laws of thermodynamics. The
form of differential system [8.1] is analogous to that of the balance equations, but
the coherence with the laws of thermodynamics is only effective because of a
suitable choice of state vector components (section 1.2.1).
We also saw in section 5.1.1 that a system of first order differential equations
can be written in the form of a scalar differential equation of order n for one scalar
variable. This equation can also be written in the form of differential system [8.1],
the state vector X (t) being replaced by a state vector having the same number of
components, but constituted of a variable x1 (t) and of its n – 1 first derivatives. By
reconsidering the calculation of section 5.1.1 the reader will see that the inputs
obtained for this differential system imply the initial inputs and their n – 1 first
derivatives. This form of state-representation is known as the form of observability
([DOU 95], [GUP 83], [KUO 02]). We will use this for a particular case in section
8.4.2.2.
For a non-linear system the matrices A, B and D can be functions of the state
vector X of the system.
For a system involving an instantaneous response, the derivative d dt takes on

small values compared with second order terms; the inertia of this system is weak
and the extensive quantities which result from the inputs are instantaneously
transferred to the sub-systems. The differential character and the idea of the initial
state of system [8.1] have disappeared and the representation is reduced to an
algebraic system:
A. X B.U 0, Y D. X
We note that in the case of an impulse input, the response and the outputs of an
instantaneous-response system involve a discontinuity. As nature does not contain
discontinuities in such situations, we have in reality a continuous transitional regime
involving a singular perturbation (section 6.4.3).
In the case of a time-invariant system, the coefficients matrices A, B and D are

independent of time, but they may be functions of the state variables: the structure of
the system remains unchanged over the course of time. If the system is furthermore
linear, they are constant and we have a time-invariant linear system.
For a system made up of continuous media, the system of differential equations

[8.1] is replaced by partial differential equations. The inputs are either boundary
conditions or volume heat source terms. When the physical properties depend on the
temperature, these equations are non-linear. On the other hand, if the properties are
only a function of the coordinates, we have a time-invariant linear system. In the
presence of a flow, the temporal derivative w wt becomes a material derivative
d dt in the case of a Eulerian representation.
8.1.3. External representation
The external representation of a system consists of considering it as a black box

that links inputs and outputs. However, the conservation laws of extensive quantities
must be considered between the inputs and outputs of a system. The permanent
production of mechanical or thermal energy is only possible if energy is provided to
the system; the same goes for mass, chemical species in the absence of chemical
reactions, etc. The consequences of the second principle of thermodynamics are
directly manifest in the constraints upon the entropy of the system studied. In other
words, the system can only evolve within the bounds of possible processes. In fact,
the coherence between automatic control and thermodynamics is only ensured if the
global thermodynamic balances are satisfied for state representation [8.1] of the
system.
In practice, a system is often characterized using incomplete data which are

obtained by means of indicial experiments: we modify one variable by quickly
increasing it by a small amount (Heaviside unit step) and we record the evolution of
the outputs. If the response of these is not instantaneous, the system undergoes an
internal evolution which must be characterized. For example, the progressive
response of a system to an instantaneous increase of the mass flow rate or of the
input temperature indicates the capacity of the system to accumulate matter or mass
in its interior. We must therefore introduce an internal state variable to the system
(volume or pressure for a compressible fluid, calorific capacity, etc.) or,
equivalently, we must introduce terms involving temporal derivatives in the external
representation (section 8.4.2.2).
The formulae used by engineers for over two centuries are external
representations which correspond to operating conditions of the system, which are
more or less well (or badly) defined. Models which use continuous-medium
equations are state representations in which the inputs and outputs take the form of
simple formulae corresponding to the system studied under particular conditions.
These simplified external representations are then often used in more general
conditions in which their degree of approximation is difficult to evaluate. For
example, in a flow system comprising a pipe of length L and of diameter D, in which
the mass flow q m is given (as an input), an interesting output is the pressure
difference 'p in the flow between the input and output sections. This will depend,
for example, on the mass flux, on the viscosity, on the density of the fluid and on the
dimension k of the roughness of the pipe. We will therefore have, in non-
dimensional form for the pressure 'p :
2
qm L § k·
'p Ue /¨¨ Re, ¸¸ [8.2]
2 D © D¹
2S
where S is the area of a reference cross-section. The non-dimensional coefficient /

of the driving pressure loss is a function of the dimensionless data (Reynolds
number Re and roughness parameter k/D) (section 4.6.1.3.4), this function having
been determined by semi-empirical considerations and experiments.
Approximate external representations can be used in the study of more complex

systems modeled by means of state representations. For example, in models of large
dimension in continuous media studied by means of numerical solution of the
Navier-Stokes equations by finite difference or finite volume methods, boundary
layers in the vicinity of walls are too thin to be discretized (section 6.5.3.3). The
friction on the walls is therefore taken as a boundary condition of certain elements
and it is expressed by a global relation analogous to [8.2] as a function of the
characteristics of the velocity field next to these elements.
8.1.4. Command models
The objective of a knowledge model is to describe quantitatively an evolving

system and its internal and external transfers, with a precise enough description of
the physics of processes both sufficient and pertinent.
A command model defines actions whose objective is to cause the system to

evolve towards a given state. We seek only to represent the evolution of the system
in the vicinity of a given state of operation; a linear relation between input and
output variations is often sufficient. We must however note that this rather blind
manner of proceeding assumes that the internal structure of the system remains
unchanged.
The pragmatic approach, from the point of view of control and command of a
system, often consists of considering only the observable outputs such as the
temperature, the pressure, certain concentrations of chemical species, fluxes, etc., at
points in an industrial system where measurement is possible. The objective of
control is to fix the values of certain output variables (temperature, pressure, etc.) or
certain state variables. The commands act on the input variables which may be of
the same nature as the output variables. In practice, we content ourselves with the
examination of the response of the system to perturbations such that we can estimate
the corrections which need to be made. The study of a response curve allows us to
define the global order of a system.
8.2. Thermodynamics and state representation
8.2.1. General principles of modeling

System [8.1] is of a differential nature with respect to time. We must add an
initial condition for the state variable X (0) . The procedure in automatic control
consists of only considering the state-representation form and “forgetting” the origin
or the physical interpretation of this. In practice, the real problem of establishing a
model lies in the choice of variables for the state vector X, which must be performed
in a manner which conforms to the laws of thermodynamics.
Our study of thermodynamics in Chapters 1 to 4 showed the following general

mathematical structure:
– extensive quantities satisfying the conservation laws (balance equations)
defined for each sub-system regardless of its state of imbalance;
– intensive quantities which are only defined for discrete systems in equilibrium,
or for continuous media with an assumption of local equilibrium;
– fluxes of extensive quantities associated with thermodynamic imbalance
characterized by differences or gradients of the intensive quantities;
– relations between intensive and extensive variables (equations of state) and
phenomenological laws for the thermodynamic fluxes of the extensive quantities.
Modeling a sub-system leads us to define and choose the variables necessary for
its dynamic description, and in particular stocks and fluxes of extensive quantities at
each instant. The discrete equations result from the application of the balance
equations to each of the sub-systems. The extensive quantities are additive and they
are always defined for any given sub-system.
8.2.1.2. Intensive variables and thermodynamic fluxes

We have seen that if the extensive variables of a system or a sub-system are
always defined, the same is not necessarily true of the intensive variables that only
have meaning if the system is in equilibrium. The complete thermodynamic
representation consists of representing the system with sufficient details, such that
each of the elementary sub-systems is close to a state of instantaneous equilibrium.
This condition is fulfilled if the size of the elements is not too great. For a finite sub-
system which is never rigorously in equilibrium, we have discussed the way to
define the mean intensive quantities (section 1.4.2.5 and section 6.5.2.4).
T T
T2
T2
M1 M2 M1 M2
T1 T1
x x
A1 A2 A1 A2
(a) (b)
Figure 8.3. Thermal flux between two neighboring elements
The behavior laws necessary to close the system of equations of the system
create a particular difficulty, as the thermodynamics fluxes which cross a surface are
related to the local gradients of the intensive quantities of the local continuous
medium. The problem is thus to express these by means of the intensive quantities
of the neighboring sub-systems. There is no general solution to this problem. Let us
take as an example the thermal resistance between two elements. Consider a fixed
1D continuous medium which is modeled as two blocks E1 and E2 which are
homogenous and of width A1 and A2 (Figure 8.3). For a fixed homogenous medium,
the mean temperatures T1 and T2 of each element are defined here by taking the
mean of the temperatures. Attributing the mean temperature to the central points of
the elements may appear natural. However, if the temperature gradient at the
interface is equal to the slope of the segment M1M2 in the temperature distribution
of Figure 8.3a, the same is not necessarily so in the case of Figure 8.3b.
The segmentation of the continuous medium is acceptable for the thermal

imbalance of the first case, whereas it is too coarse for that of the second case. We
encounter here a usual interpolation problem.
The modeling of thermal fluxes between the two sub-systems requires in

addition a suitable discretization of the intensive quantities between two neighboring
sub-systems. Let us consider two examples:
1) Consider a 1D solid (Figure 8.4) subjected to a conductive thermal flux and
discretized into n elements Ei of uniform thickness e and temperature Ti.
(i=1,2,…,n). The element Ei is characterized either by its energy Qi or by its mean
temperature Ti:
x
Qi ³xnn 1 U CdT U CeTi
T T1 T2 Ti-1 Ti Ti+1 Tn-1 Tn

Ei
Ei+1
Ei-1
E1
E2
En
En-1
e e e x
Figure 8.4. Model of a 1D thermal transfer
We must now express the thermal fluxes between two neighboring elements.
Assuming that the middle of each element is at temperature Ti, we can write the
thermal flux qTi,i+1 received by Ei from Ei+1:
Ti Ti 1
qTi , i 1 O
e
We note that this expression satisfies the principal of action and reaction
( qTi1, i qTi1, i ) associated with the conservation of energy. If there is no lateral
thermal flux, the balance equation can be written for each element Ei:
dTi Ti Ti 1 Ti Ti 1 2Ti Ti 1 Ti 1
U Ce O O O [8.3]
dt e e e
2) Consider the established flow of a fluid in a cylindrical pipe. We assume, in

order to simplify matters, that the flux and the physical properties of the fluid are
constant.
We have seen (section 6.5.2.2) that the balance equation of a quantity g in a pipe
can be written by means of surface integrals on a cross-section and an integral over
the contour of this section ([6.82]). We thus consider a cutting out of the pipe into n
surfaces Si with an inter-surface spacing of e (Figure 8.4). Let u (t , M ) and T (t , M )
be the velocity and temperature distributions in the cross-section; we define below
the mean temperature Ti and the mean mixing-temperature Tmi for the section Si:
1
Ti t ³Si Tds; Tmi t ³Si u Tds
qv
The energy balance in the pipe can be written ([6.87]):
w U CT
³S
wt
ds
w
wx S

³ U uCTds ³S V T ds ³C qTp dA
T 1 T2 Ti-1 Ti Ti+1 Tn-1 Tn

T
S1 S Si-1 Si Si+1 Sn-1 S
2 n
u(t,M)
Ei+1
Ei
Ei-1
E1
E2
En
En-1
e e e S x
Figure 8.5. Model of a pipe flow with thermal transfer
The calculation of the convective thermal flux between two successive surfaces
must be realized from upstream to downstream of the flow (section 5.6.1) and the
derivative w wx should be discretized between the sections Si-1 and Si. Let Tpi be the
temperature of the wall associated with the surface Si; the preceding equation [6.87]
can thus be written:
U CSe
dTi

U Cq v Tmi 1 Tmi 1
eF h T pi Tmi [8.4]
dt
where we denote by F the perimeter of the cross-section. The exchange coefficient h

between the fluid and the wall is defined as a function of the mean mixing-
temperature Tmi; its value depending on the considered problem is obtained from
experiments or theoretical evaluation in neighboring problems ([SCH 99], [YIH
77]).
8.2.1.3. Balance equations for the sub-systems

The writing of balance equations for a system is not necessarily associated with
the cutting out which is chosen in order to model the n disconnected sub-systems.
Each finite sub-system is characterized by the extensive quantity G which it contains
and the flux MSG which it receives from the neighboring sub-systems. The balance
equations for the extensive mass quantity g in a domain S can be written ([2.7]):
w

³ U gdv
wt S
M SG ³S V G dv
The extensive quantity of a system is obviously the sum of the corresponding

extensive quantities of its sub-systems only if these are disjoint. For two such sub-
systems, such as S1 and S2, we have:
³S1 S 2 U gdv ³S1 U gdv ³S 2 U gdv
S1 S2 S3
M 12
M 23
balance equation S 1 balance equation S 1 balance equation S 1

balance equation S 2 balance equation (S 2 S 3) balance equation (S 2* S 3)
*
balance equation S 3 balance equation (S 3 * S 1)
Figure 8.6. Balance equations for three material adjacent sub-systems
The balances can be obtained for sub-systems which are partially over-lapped.
Figure 3.6 shows the case of three contiguous sub-systems S1, S2 and S3: the sub-
system S2 S3 overlaps S2, but it is possible to write the balance equations for the 3
disjoint sub-systems or for the three ensembles of sub-systems different two by two.
This manner of proceeding is nonetheless limited by the fact that the system of
equations obtained must be suitably enough conditioned (section 7.2.7.3), each
equation providing specific and sufficient information. The sum of the internal
fluxes exchanged between the sub-systems is zero (principle of action and reaction).
In the construction of models and numerical schemes we must be careful that this
property is ensured (conservative scheme).
8.2.1.4. Coupling coefficients between two sub-systems

Consider a system comprising n sub-systems Ei characterized by their extensive
quantities gi and their associated intensive energy quantities, or an equivalent (for
example, quantity of heat and temperature). Let us assume that two elements i and j
exchange a flux of the quantity g under the effect of a difference the intensive

quantity y i y j . The flux N ij y j y i of the quantity x received from the element

j by the element i is opposed to the flux N ji y i y j of the quantity x received from
the element i by the element j (action and reaction) and the positive coefficients Nij
and Nji are equal. The balance equations of the quantity g for these two sub-systems
can be written:
dg j
dg i

! N ij y j y i ! ;
! N ji y i y j !
dt dt
or, by classing the terms:
dg i dg j
! N ij y j ! ; ! N ji y i !
dt dt
Denoting by U the group of sources of the quantity gi and defining the state
vectors G and Y of the extensive quantities gi and the corresponding intensive
quantities yi, the balance equations can be written in the general form:
dG
KY JU [8.5]
dt
The matrix J translates the importance in each element of the external

contributions. As the coupling coefficients Nij and Nji between the elements i and j
are equal, the matrix K is symmetric. This feature is not a general constant, since the
coefficients of Nij are functions of the state variables of the elements i and j.
So, the extensive variable gi and the intensive variable yi of a sub-system are
related by a state equation. Passage to the form [8.1] of the state representation of
the system can often be realized in thermal systems by expressing in each of the sub-
systems the extensive variable gi as a function of the only corresponding intensive
variable yi, g i g i y i . By defining the diagonal matrix ' of positive elements
d ij G ij dg i dy i , we have in this case:
dG dY
' [8.6]
dt dt
Letting Y = X, system [8.5] can be written in the form [8.1]:
dX
A. X BU with : A '1 K B '1 J . [8.7]
dt
It is easy to verify that the matrix A '1 .K of equation [8.1] is not symmetric,
except if the value of each element dij is independent of the index i (identical sub-
systems). The reader can verify these considerations by means of the example in
section 8.3.1.3 by comparing equations [8.23] and [8.24] (or [8.28] and [8.29]). The
importance of this property for the linear case is discussed in section 8.2.2.2.
8.2.1.5. Equivalence of inputs and initial conditions

Inhomogenous system [8.8] with impulse inputs, but with zero initial conditions,
can be reduced to a homogenous system with non-zero initial conditions. Let us
verify this point in an elementary manner using second order differential equation
[8.8] with zero initial conditions and having a Dirac distribution cG t at instant 0 as
the input:
x(t ) a1 x t xt cG t x0 x 0 0 [8.8]
By integrating between 0 and a small time H and assuming that the functions
xt and x t remain bounded, we obtain:
x H c x(H ) oH
Problem [8.8] is thus reduced to a homogenous problem with an initial condition

at the instant H:
x(t ) a1 x t xt 0; xH oH , x H c [8.9]
Letting H tend to zero, we obtain the equivalent problem [8.10]:
x(t ) a1 x t xt 0; x0 0, x 0 c [8.10]
The mechanical interpretation of this result (sometimes known as Schwartz’s

theorem) is clear, given an initial velocity c is equivalent to the application of an
impulse which produces this velocity at the initial instant.
The reader can verify in the same way that non-linear problem [8.11] is
equivalent to problem [8.12]:
dX
A X . X (t ) B.U t X 0G t X 0 0 [8.11]
dt
dX
A X . X (t ) B.U t X 0 X0 [8.12]
dt
We can replace the initial conditions with these additional impulse inputs which
act on the state variables. In certain cases, these can be realized by acting on the
existing inputs U, except if the number of state variables n is greater than the
number of independent inputs. In summary, it is often not possible to make clear
conceptual distinctions between inputs and initial conditions (or boundary
conditions) such that they appear in the mathematical representation of a model.
8.2.1.6. Modeling and numerical solution of equations for continuous media

We have reasoned so far in the context of “thermodynamic” modeling of systems
by means of an ensemble of discrete sub-systems which are each in a state of
thermodynamic quasi-equilibrium. Other approaches are possible for obtaining
discretized equations from the balance equations for continuous media:
– finite difference methods consist of discretizing the domain under study and
replacing the differential operators at the points of the grid thus obtained with finite
difference operators;
– finite element methods represent the local solution by simple algebraic
functions in very small sub-domains; the balance equations are integrated in the
domain after multiplication by the weighting functions, n being the number of
unknown parameters defining the ensemble of local solutions. Integrations by parts
often allow the reduction of the order of the necessary derivatives, and the use of
finite elements of standard form allows the realization of very general calculatory
procedures. We thus obtain an integral formulation comprising n integral conditions
for the determination of n unknowns;
– finite volume methods consist of the definition of finite elements in which the
balance equations are integrated; the approximations of the solution in each element
are simple algebraic functions.
In all cases, we obtain a system of equations which is analogous to the state

representation, the preceding numerical schemes belong to two broad categories
depending on whether they are conservative or not (a numerical scheme is said to be
conservative if the discretized equations exactly satisfy the balance of an extensive
quantity with the variables which are used to represent the system). A conservative
scheme avoids the amplification of numerical errors which are integrated into the
balance of physical quantities. It cannot of course correct the unstable character of a
physical system. The finite element method is not conservative, contrary to the finite
volume method. Finite difference schemes may or may not be conservative. Finally,
it is important to remember that, regardless of the method used for discretizing the
equations, numerical schemes and algorithms used should respect the rules of
transmission of information from one point or element to another which results from
the character elliptic, hyperbolic, parabolic or mixed of the equations (section 5.6.1).
8.2.2. Linear time-invariant system (LTIS)

Differential system [8.7] for A constant and with initial conditions can be
written:
dX
A. X B.U , X 0 X0 [8.13]
dt
The linearity of equation [8.13] makes it possible to use the usual methods of
solving linear differential equations. The solution is the sum:
– of a particular solution X e (t ) of the complete system [8.14]:
dX e
A. X e B.U [8.14]
dt
– and a solution Xˆ (t ) of the homogenous system [8.15] adjusted such that the
solution satisfies the initial conditions:
dXˆ
A. Xˆ , Xˆ 0 X e 0 X0 [8.15]
dt
The solutions of the homogenous system [8.15] are real decaying exponentials or
damped sinusoidal functions if the system is stable. They represent transitional
regimes.
A particular solution X e (t ) is often chosen so as not to contain transitional

terms and represents an established regime of the problem treated (assumed stable),
independent of the initial conditions. This established regime plays a very important
role in numerous cases, in particular when the inputs are either simple algebraic
functions of time or harmonic excitations. This particular solution can be obtained in

general by the variation of constants method (see mathematical texts).
The separation between transitional and established solutions is no longer

theoretically so clear for non-linear systems; it can however be useful in numerous
cases, as dissipation phenomena are often associated with a progressive elimination
of the initial conditions, except of course in the case of a multiplicity of established
solutions (section 6.6.2).
The above result can be immediately transposed to partial differential equations.

In this case the particular solution X e t of the inhomogenous equation should
satisfy all the zero and non-zero boundary conditions, the solutions of the
homogenous equation satisfying zero boundary conditions. The solution obtained for
the homogenous equation is chosen such that the complete solution satisfies the
initial conditions.
8.2.2.2. Recall on transitional regimes

The general solution of homogenous equation [8.15] is a linear combination of
terms of the form 4e /t , / being a solution of the eigenvalue problem:
A /I
4 0 I unit matrix
[8.16]
Equation [8.16] only has non-zero solutions for particular values /i, which are
known as eigenvalues, of which there are n, including their eventual multiplicity.
For each value /i the solutions of equation [8.16] are of the form ci 4 i , where 4i is
an eigenvector associated with the eigenvalue /i (ci is some constant). An
eigenvalue and its associated eigenvectors are known collectively as a mode. The
eigenvalues are real and negative for aperiodic modes (in thermal systems in
particular) or complex with a real part which may be negative for oscillatory damped
modes or zero for unamped modes (in acoustics for example). By limiting ourselves
here to thermal systems, the eigenvalues Xˆ i t can be written:
Xˆ i t 4 i exp/ i t
The complete solution of a given problem is thus written:

n
X t X e t Xˆ t with : Xˆ t ¦ ci 4 i exp/ i t [8.17]
i 1
The scalar coefficients ci are calculated from the initial condition X 0 which is
assumed to be known:
n
X 0 X e 0 ¦ ci 4 i [8.18]
i 1
The transitional regimes are solutions comprising the return to rest of the
homogenous system (or to the established regime X e (t ) of the system with a right-
hand side) from non-zero initial conditions (or eventually zero for an established
regime X e (t ) ). Having determined the eigenvectors 4i, the coefficients ci are
solutions of linear system [8.18], the matrix of which is full; for a large number n of
equations this system is ill-conditioned (section 7.2.7.3). However, if a scalar
product exists U V ! for which the matrix A is self-adjoint
( AU V ! U A V ! ), the eigenvalues 4i are orthogonal (taking account of
[8.16], if / i z / j , we have 4 i 4 j ! /i 1 A4 i 4 j ! /j1 4 i A4 j ! 0 ).
For the scalar product U V U t 'V V t 'U of the vectors U and V , defined
with the diagonal matrix ' (relation [8.6]), K being symmetric, this leads to:
AU V ! U t A t 'V U t K t '1 'V U t KV U t ''1 KV U AV !
We can deduce from this, in a manner analogous to that outlined in section

A.4.2.2 of Appendix 4:
ci X 0 X e 0t '. X i X it '. X i
NOTE – Expression [8.17] is the development at instant t of the function X̂ t on

the basis of eigenfunctions 4i with instantaneous coefficients ci exp /i t . After a
time lapse t, which is large compared to 1//J, the mode j is no longer present in the
transitional solution; its coefficient there being equal to zero, this gives:
Xˆ t t '.4 j X t X e t t '.4 j 0 [8.19]
8.3. Modeling linear invariant thermal systems
8.3.1. Modeling discrete systems

The fundamental problem is the reasoned choice of the number of sub-systems
or the variables necessary for representation of a system, the number of variables
being greater as the thermodynamic imbalance is more pronounced. We will limit

our discussion here to linear systems which only possess modal solutions allowing
an analytical treatment. The methodologies for writing balance equations are
independent of linearity properties. The variation of physical properties as a function
of the state variables often leads to weak non-linearities, which do not change the
general properties and the orders of magnitude obtained. We will first study three
examples of simple discrete thermal systems, then two problems of continuous
media (thermal walls).
8.3.1.2. Models with two sub-systems at constant temperature

Consider a time-invariant linear system which is composed of two blocks E1 and
E2 of width A, of specific heat C and separated by a thermal resistance R. Let us then
consider a representation of this system by means of two sub-systems of uniform
temperatures T1 t and T2 t (Figure 8.7). Each of these two sub-systems is
separated from the exterior by another thermal resistance R’.
A thermal resistance is an element of small thickness which transmits heat in a

quasi-instantaneous manner; the thermal flux M which crosses this element is
proportional to the temperature difference between its external faces. The fluxes
M1,2, M1w and M2w received by each sub-system (Figure 8.7) are:
T1 T2 T1ex T1 T2ex T2
M1, 2 ; M1w ; M 2w . [8.20]
R R' R'
The initial temperatures T1 (0) and T2 (0) and the external temperatures T1ex(t)
and T2ex(t) are given.
m m
T1ex T1 T2 T2ex
M1w M1,2 M2w
R' E1 R E2 R'
Figure 8.7. Thermal conducting system with two equilibrium sub-systems

Taking account of [8.20], the energy balance of each sub-system allows the
obtaining of a differential system for the temperatures T1 t and T2 t :
dT1 T1 T2 T1 T1ex dT2 T2 T1 T2 T2ex

mC ; mC .
dt R R' dt R R'
or in matrix form:
§ §1 1· 1 ·
¨ ¨¨ ¸¸ ¸
d §T · ¨ © R R' ¹ R ¸§¨ T1 ·¸ 1 §¨ T1ex ·¸
mC ¨¨ 1 ¸¸ ¨ [8.21]
dt © T2 ¹ ¨ 1 § 1 1 · ¸¨ T ¸ T ¸
¨¨ ¨¨ ¸¸ ¸¸© 2 ¹ R ' © 2ex ¹
© R © R R' ¹ ¹
Searching for solutions of the form 4 e rt for the homogenous system associated
with [8.21] gives the eigenvalue equation (characteristic equation):
§1 1· 1
¨¨ ¸¸ mCr 2
© R R' ¹ R §1 1 · 1
¨¨ mCr ¸¸ 0
1 §1 1· © R R' ¹ R2
¨¨ ¸¸ mCr
R © R R' ¹
This has two roots (eigenvalues):
1 1 §2 1·
r1 ; r2 ¨¨ ¸¸ [8.22]
mCR ' mC © R R ' ¹
From these we can find the components (4i1, 4i2) of the eigenvectors 4i (i = 1,
2), which are solutions to the system of equations:
§ §1 1· 1 ·
¨ ¨¨ ¸¸ ¸
§4 · ¨ © R R' ¹ R ¸§¨ 4 i1 ·¸
mCr ¨¨ i1 ¸¸
© 4i2 ¹
¨ 1 §1 1 · ¸¨ 4 ¸
¨¨ ¨¨ ¸¸ ¸¸© i 2 ¹
© R © R R' ¹ ¹
We obtain:
– mode 1 (symmetric, i = 1, with 411 = 412 = 1):
1
r1 ; T11 T12 exp t / mCR '
mCR'
– mode 2 (antisymmetric, i = 2, with 421 = 422 = 1):
1 §2 1· § t §2 1 ··
r2 ¨¨ ¸¸; T21 T22 exp¨¨ ¨¨ ¸¸ ¸¸
mC © R R ' ¹ © mC © R R' ¹ ¹
The general solution of the homogenous system can be written:
T1 c1 expr1t c 2 expr2 t , T2 c1 expr1t c 2 expr2 t .
Let T1e(t) and T2e(t) be a particular solution of [8.21] (“established” solution);

we obtain as a general solution to [8.21]:
T1 (t ) c1 expr1t c 2 expr2 t T1e t

T2 (t ) c1 expr1t c 2 expr2 t T2e t
The integration constants c1 and c2 can be calculated with initial conditions;
c1 >T1 0 T2 0 T1e 0 T2e 0@ 2

c2 >T1 0 T2 0 T1e 0 T2e 0@ 2
The decomposition of the solution into two modes can be written:
T1 t >T1 t T2 t @ 2 c1 expr1t >T1e t T2e t @ 2

T 2 t >T1 t T2 t @ / 2 c 2 expr2 t >T1e t T2e t @ 2
The fast mode T2(t) corresponds here to the establishment of internal equilibrium
of the system, while the slow mode T1(t) represents the establishment of equilibrium
between the system and the exterior. For a system isolated from the exterior (infinite
R’), we obtain:
2t ·
T1 (t ) T2 (t ) T1 0 T2 0, T1 (t ) T2 (t ) T1 0 T2 0 exp§¨¨ ¸¸
© mCR ¹
8.3.1.3. System with three components in series

Consider now the system shown in Figure 8.8 consisting of three sub-systems
E1, E2 and E3. The component E3, isolated from the exterior, is in contact with
E1and E2 via the thermal resistances of the same value R/2. As before, the sub-
systems E1 and E2 are each in contact via a thermal resistance of the same value R’,
with the external medium at temperature Ti ex(t) (i = 1,2). Sub-system E3 is isolated
from the exterior.
The energy balance equations can be written:
dT1 §2 1 · 2T T
mC T1 ¨¨ ¸¸ 3 1ex ;
dt © R R' ¹ R R'
dT2 §2 1 · 2T T
mC T2 ¨¨ ¸¸ 3 2ex ; [8.23]
dt © R R' ¹ R R'
dT3 2T1 2T2 4T3
m' C .
dt R R R
m m'=Hm m
T1ex R' T1 R/2 T3 R/2 T2 R' T2ex
M1p M1,3 M3,2 M2p
E1 E3 E2
Figure 8.8. System with three sub-systems (first example)
m'
The system of equations can be written in matrix form (with H ):
m
§ R ·
¨ 2 0 2 ¸
§ T1 · ¨ R' ¸§ T1 · § T1ex ·
d ¨ ¸ 1 ¨ R ¸¨ ¸ 1 ¨ ¸
¨ T2 ¸ ¨ 0 2 2 ¸¨ T2 ¸ ¨ T2ex ¸ [8.24]
dt ¨ ¸ mCR ¨ R' ¨ ¸ mCR ' ¨ 0 ¸
© T3 ¹ ¨ 2H 2H 4 H ¸¸© T3 ¹ © ¹
¨ ¸
© ¹
We will limit ourselves here to the discussion of solutions to the homogenous

R
system associated with [8.24] in the form 4i e rt . By letting a 2 and
R'
/ mCR r , we obtain the non-dimensional eigenvalue equation (characteristic

equation):
a/ 0 2
0 a/ 2 >
a / H /2 /4 H a 4a 8 @ 0 [8.25]
2 2 4H /
The roots of equation [8.25] are:
4 H a r 16 8H 4 a H 2 a 2
/1 a; / i , i 2,3 [8.26]
2H
As the sum of the roots /2 and / 3 of the equation are negative and their product
is positive, the two roots are negative (a diffusive system is aperiodic).
The symmetry of the system allows us to immediately find the modes. A

symmetric mode is characterized by 4 i1 4 i 2 . Substituting this relation into the
homogenous system derived from [8.24], we have:
/ a 4 i1 24 i3 ; H / 44 i3 44 i1
Eliminating 4i3 between these relations, we find that the eigenvalues /i

corresponding to these modes satisfy the second order trinomial of characteristic
equation [8.25]. The values of 4i1 and 4i3 are of the same sign or of opposite sign
depending on whether / a is positive or negative. Substituting –a as the value of
/ in the preceding trinomial, we find that this takes on a negative value, which
shows that the value – a is situated between the roots of the trinomial. We can easily
derive from this that the quantity / a is positive for the largest root /3, whereas
it is negative for the other root /2. We can verify immediately that the root
/1 a corresponds to the anti-symmetric mode 411 412 and 413 0 .
(1 (3(2 (1 (3(2 (1 (3(2
+ + + + 0 – + – +
/2 /1 /3
Figure 8.9. Structure of normal modes classed with increasing modulus of the eigenvalues
r2 r1 r3 (decreasing time constants)
The structure of the modes is shown in Figure 8.9. The time constant of a mode
is equal to 1 r . The preceding discussion shows that we have r2 r1 r3 . The
slowest mode corresponds to the root r2 for which the three sub-systems at the same
temperature constitute a thermodynamic system in equilibrium, a structure for which
the thermal inertia is greatest towards the thermal resistances R’. On the other hand,
the most “agitated” mode corresponds to the smallest time constant 1 r3 .
Suppose now that the mass m’ of the sub-system E3 is small compared to the
mass m of the sub-systems E1 and E2 (i.e. H << 1). Performing a series development
of the roots [8.26], we obtain:
4
/1 a; /3 O1 /2 2 a OH
H
Returning to the dimensional values ri / i mcR , we obtain:
1 §2 1· 4 1
r1 ¨¨ ¸¸ r3 r2
mC © R R ' ¹ H mCR mCR '
The values r1 and r2 are those already found ([8.22]) for the system studied in
section 8.3.1. As the value of r3 is large compared to r1 and r2, mode 3 is very
quickly damped; after this damping, we have the relation [8.19] for i =3, which can
here be written Xˆ t t '.4 0 . Replacing X̂ t , ' (a matrix which allows us to
3
pass from [8.23] to [8.24]) and 43 by their values at small H:
§ T1 · § mC 0 0 · § 1 ·
¨ ¸ ¨ ¸ ¨ ¸
Xˆ t ¨ T2 ¸; ' ¨ 0 mC 0 ¸; 43 ¨ 1 ¸
¨T ¸ ¨ 0 0 H mC ¸¹ ¨ 2 / H ¸
© 3¹ © © ¹
we obtain:
Xˆ t t '.4 3 mC T1 T2 2T3 0 [8.27]
Relation [8.27] reduces to the balance equation of the sub-system E3 (third

equation [8.23]) in which the mass is negligible. We verify that in replacing T3 in
the first two equations of [8.23] with its value taken from [8.27] we recover
equations [8.21]: the three sub-systems model has been reduced to a model with two
sub-systems by removal of the third sub-system of small mass.
8.3.1.4. Thermal systems with 3 components in the form of a star

Now consider the system shown in Figure 8.10, which comprises three sub-
systems E1, E2 and E3 in contact two by two across the same thermal resistances R.
Each of these is in contact across the same thermal resistance R’ with the external
temperature medium Ti ex(t) (i = 1, 2, 3). We will only discuss the structure of the
modes here. The energy balance equations can be written:
dT1 2T1 T2 T3 T T1ex

mC 1 ;
dt R R'
dT2 T1 2T2 T3 T2 T2ex
mC [8.28]
dt R R'
dT3 T1 T2 2T3 T3 T3ex
m' C
dt R R'
§ R· m'
or, in matrix form, letting: a ¨¨ 2 ¸¸ and H :
© R' ¹ m
§ T1 · § a 1 1 ·§ T1 · § T1ex ·
d ¨ ¸ 1 ¨ ¸¨ ¸ 1 ¨ ¸
¨ T2 ¸ ¨ 1 a 1 ¸¨ T2 ¸ ¨ T2ex ¸ [8.29]
dt ¨ ¸ mCR ¨ mCR ' ¨
© T3 ¹ ©1 H 1 H a H ¸¹¨© T3 ¸¹ ¸
© T3ex H ¹
T1ex m m T2ex
R' T1 R T2 R'
M1w M1,2 M2w
E1 R R E2
M3,1 T3 m' M3,2 E3
T3ex R' M3w
Figure 8.10. System with three sub-systems (second example)
Searching for solutions of the form 4i e rt for the homogenous system associated
with [8.29] gives the non-dimensional eigenvalue equation, where we have let
/ mCR r :
a/ 1 1
1 a/ 1 0
1 1 a H /
or:

H / 3 / 2 a1 2H / 2 H 1 a 2 a 2a 12 0 [8.30]
Taking account of the system symmetry, we see immediately that the anti-
symmetric mode verifies the relations 411 412 and 413 0 . Substituting
these relations into the equations of the homogenous system, we find that the
corresponding eigenvalue is equal to /1 (a 1) . The characteristic equation
[8.30] can be written:
/ 1 a
[% / 2 (%(a 1) a)/ (a 1)(a 2)] 0 [8.31]
Its two other roots /L are (i = 2.3):
/i

a H a 1 r a 2 2H a 2 a 4 H 2 1 a 2
2H
The structure of the modes can be obtained by simple reasoning. Taking account
of the system symmetry, the other eigenfunctions should present the symmetry
4 i1 4 i 2 . Substituting this relation into the homogenous system derived from
[8.29], we obtain:
/ i a 14 i1 4 i3 ; H / i a 4 i3 24 i1
We immediately find that the values /L which satisfy the preceding relations are
the roots of the second order trinomial of characteristic equation [8.31]. The values
of 4L1 and 4L3 have the same sign or opposite sign depending on whether H / i a
is positive or negative. Substituting H a as the value of /L in the preceding
trinomial, we find that this takes on a negative value, showing that this value is
between the roots of the trinomial. The result is that the quantity H / i a is
positive for the largest root /3, whereas it is negative for the other root /2. Figure
8.11 shows the structure of the three modes.
E1 E2 E1 E2 E1 E2
+ + + – + +
E3 + E3 0 E3 –
/2 /1 = - a - 1 /3
Figure 8.11. Structure of normal modes
For small values of the calorific capacity m'C of sub-system E3, a development
in H gives the values:
a 2
/3 O1 /2 a 1 OH
H a
The corresponding values of ri / i mCR are:
1 § 3 1· 1 §2 1· 1 § R 3R ' ·
r1 ¨¨ ¸¸ r3 ¨¨ ¸¸ r2 ¨¨ ¸¸
mC © R R ' ¹ mCH © R R ' ¹ mCR ' © R 2 R ' ¹
As in the case of three components in series, r3 is much larger than r1 and r2, and
mode 3 is thus rapidly damped. The third equation of system [8.28] can thus be
written:
2T3 T1 T2 T3 T3ex
0
R R'
The reader can verify that the preceding relation is identical to condition [8.19].
Contrary to the case of three components in series, the limit of this system for m' = 0
is not the preceding system of two sub-systems (section 8.3.1.2).
8.3.2. Thermal models in continuous media
8.3.2.1. Overview
The idea of a continuous medium amounts to replacing an integer valued index
(number of components of the state vector X) by a spatial variable of continuous
values. The temporal and spatial behaviors of the system are thus continuous
functions or piecewise continuous. However, we have seen in Chapter 5 that the
theory of characteristics allows us to identify different behaviors for these variables.
If the time is by nature irreversible, the spatial properties or spatio-temporal
properties are different, depending on the elliptic, parabolic or hyperbolic nature of

the system.
We will here consider problems of thermal conduction in homogenous media.

The method can also be applied analogously in heterogenous media with physical
properties which depend on the spatial variables.
8.3.2.2. Wall problem

8.3.2.2.1. Equations and solution of homogenous systems
Consider a wall of thickness 2A comprising a homogenous material whose faces
are supposed, for example, to be at a given, equal temperature T1(t). The thermal
diffusivity a of the material is supposed constant. The distribution of temperature on
the interval [–A,A] satisfies the heat equation:
wT w 2T
a A d x d A [8.32]
wt wx 2
The given data are:

– the initial conditions: T 0, x T0 x ;
– the boundary conditions (fixed temperature): T A, t T A, t T1 t .
As equation [8.32] is linear, the methodology is the same as before: the complete
solution can be obtained by superposition of the particular solution Te x, t of
equation [8.32] satisfying the non-zero boundary conditions and a general solution
of the heat equation with zero boundary conditions for the temperature. As in section
8.2.2.2, the coefficients of this solution are calculated such that the complete
solution satisfies the initial conditions.
In order to simplify the notation of the general solution, let us take the non-
dimensional variables:
~ ~
x x A; t a t A2 [8.33]
Equation [8.32] becomes:
wT w 2T
~ 1 d ~x d 1 [8.34]
wt w~
x2
We seek a group of solutions with separated variables of the form:

~ ~
T (~
x , t ) f (t ) g (~
x ) . Substituting this expression into [8.34], we obtain2:
~
f ' (t ) g" ( ~
x)
~ ~ cte / 2 [8.35]
f (t ) g(x )
We derive from this the function f:
~ ~
f (t ) exp(/ 2 t ) [8.36]
and the eigenvalue problem for the function g:
g" ( ~
x ) / 2 g (~
x) 0; g r 1 0 [8.37]
The integration of equation [8.37] shows that the eigenfunctions are of the form
x ) A cos/ ~
g (~ x B . The expression of boundary conditions [8.37] at ~x r1
gives the values of the constants / and B. We thus obtain:
S S
/ B pS ; / B p' S ( p, p' : integers)
2 2
or:
S S S
/ p p ' ; B p p '
2 2 2
In the preceding expressions p – p' and p + p' have the same parity:
– if p – p'=2k is even, we have / 2k kS and the corresponding eigenfunctions
are odd: g 2k sin>kS ~ x @;
S
– if p – p' = 2k + 1 is odd, we have / 2k 2k 1 and the eigenvalues are
2
ª S º
even: g 2k 1 cos «2k 1 ~
x» .
¬ 2 ¼
The first eigenfunctions of the wall problem at zero wall temperature are
represented in Figure 8.12.
2 The system being damped, the constant of equation [8.36] is necessarily negative.
gi gi
-1 1 -1 1
~
x ~
x
: g1 : g3 : g5 : g7 : g2 : g4 : g6 : g8
Figure 8.12. First eigenfunctions of the homogenous problem

inside a wall with constant temperature faces
The preceding eigenfunctions form a representation basis of functions which are

zero at the extremities of the interval [-1,1] and which possess sufficient regularity
properties. Taking account of the condition g r 1 0 , the eigenfunctions of
problem [8.37] are orthogonal. In effect, an integration by parts immediately gives:
1 '' ~ 1 '
³1 g 'p g q' d~
1
³1 g p g q dx ³1 g p dg q x
1
g p gq'' gq g p'' ¯ dx 0.
hence: ¨ 1 ¢¡ ±°
The operator of problem [8.37] is self-adjoint (Appendix 4), showing the

orthogonality property. Te being the established solution (section 8.2.2.2), we thus
express the solution in the form:
f
Te~
x , t ¦ ci exp( /2i t ) g i ( ~
~ ~ ~
T (~
x, t ) x) [8.38]
0
~
The coefficients ci are calculated at the instant t 0:
f
T (~
x ,0) T0 ~
x Te ~
x ,0 ¦ c i g i ( ~
x)
0
hence (Appendix 4):

³1 >T0 ( x ) Te x ,0@g i ( x )dx

1 ~ ~ ~ ~
ci [8.39]
1 2 ~ ~
³1 g i ( x )dx
8.3.2.2.2. Thermal shocks on the walls

Let us now consider the simple example of the formation of thermal shocks on
the two faces of a wall whose initial temperature T0 x is uniform and whose
temperatures T1 t on the wall faces are constant and equal to T1 for positive t .
~ ~
The established solution Te ~x , t is thus constant and equal to T1. The coefficients
~
c2k corresponding to the odd eigenfunctions are zero. Formula [8.39] gives, for the
coefficients of the even eigenfunctions:
1k 4
c 2k 1 T0 T1
2k 1S
From this we can derive the solution of the thermal shock problem T1 – T0
applied on the two faces of a wall of thickness 2A:
~ S2 ~
T (~ 1k 4 2k 1
2
x , t ) T0 f t ª S º
1 ¦ e 4 . cos «2k 1 ~
x» [8.40]
T1 T0 k 0 2k 1S ¬ 2 ¼
1 1 ~ ~ ~
The mean3 temperature Tm ³ T ( x , t )dx can be expressed:
2 1
~ S2 ~
2 k 1
2
Tm ( t ) T0 f 8 t
1 ¦ e 4 [8.41]
T1 T0 k 0 2k 12 S 2
This expression allows us to know the contribution of each mode (Table 8.1) to
the thermal energy supplied by conduction, this being proportional to the relative
d
8 2k 1
2
amplitude of the mode (we have 2 1):
k0
3 Defined as the space mean temperature for homogenous media.

Mode number 0 1 2 3 4 5 6
Mode amplitude 0.8106 0.0901 0.0324 0.0165 0.0100 0.0067 0.0048
Table 8.1. Distribution of thermal energy in the seven first modes
The wall behaves in a manner similar to a first order system, since only 20% of
the heat is exchanged more rapidly in modes higher than the first mode. The density
of the thermal flux qTp A, t O wT wx x A on the face of the wall at -A can be
derived from [8.40]. Infinite at time t = 0, it decays very rapidly in the first instants
and it can immediately be written in non-dimensional form:
S 2 at
AqTp A, t 2 k 1
2
f
2¦ e 4 A2 . [8.42]
O T1 T0 0
T
T1
f
5
4 3
2
1
–A T0 x
0 A
Figure 8.13. Evolution of the temperature distribution during

a thermal shock on the two faces of a wall
Figure 8.13 shows the evolution of the temperature profiles with time (curves
from 0 to 5). The curve 0 is the distribution at the initial instant where we impose the
temperature T1 on the walls ( x rA ). The diffusion of this condition occurs
progressively from the walls: the curves 1 and 2 represent the thermal shocks in a
quasi-infinite medium from the walls. The thermal diffusion zones are then rejoined
on the other curves (3 to 5). On curves 4 and 5, observed after a non-dimensional
time in the order of 1//1, only the first mode g1 remains, and it damps until its
amplitude falls to zero.
The temperature distribution of the thermal shock in a semi-infinite domain from

a wall has already been obtained (section 5.4.5.4 and formula [5.52]). It can be
written here for the thermal boundary layer on the wall x A , and by taking non-
dimensional variables [8.33]:
T T0 § x A· §~x 1· 2 ~x 1 / 2 ~
t u2
1 erf ¨¨ ¸
¸ 1 erf ¨ ¸ 1 ³0 e du
T1 T0 ¨ ~ ¸ S
© 2 at ¹ © 2 t ¹
The complete solution which represents the two boundary layers can be written:
T T0 §~x 1 ·¸ §~x 1 ·¸
2 erf ¨ erf ¨ [8.43]
T1 T0 ¨ ~¸ ¨ ~ ¸
©2 t ¹ © 2 t ¹
The thermal flux density at the wall ~

x 1 can be derived from [8.43]:
§ wT · O
qTp O ¨¨ ¸¸ T1 T0
© wx ¹ x A Sat
It can be written in non-dimensional form (section 4.6.1.3.4), with the reduced

~
time t [8.33]:
A qTp A, t
~
1
[8.44]
O T1 T0 S ~t
The mean temperature of the wall in the thermal boundary layer regime can be
obtained by performing the energy balance in the wall between instants 0 and t:
O du
U Tm T0 CA ³0 qTp u du
t
T1 T0 ³0t T1 T0 2 O t
S au Sa
or:
Tm t T0 2 ~
t [8.45]
T1 T0 S
8.3.2.2.3. Composite representation by matched asymptotic expansions

The series expansion of eigenfunctions of the thermal shock problem on the
faces of a wall is a poorly adapted representation in the first instants of the thermal
shock, whereas the representation by thermal boundary layers captures the physics
of the problem more satisfactorily. For small time values, the modal representation
of the mean temperature is nearly acceptable with very few modes (taking account
of a discontinuity at t = 0); on the other hand, modal expression [8.42] for the
thermal flux density is unusable and we thus have recourse to expression [8.44]
which is equivalent to the sum of the series [8.42] in these conditions.
We are thus led to search for a formula which contains the two different
asymptotic expressions f1(t) and f2(t) of the function f(t), which are valid for small
and large values of t respectively following the time value. This can be obtained by
means of a matching formula or a weight between the two temporal domains which
gives exact values for the function f and its temporal derivative at the origin, and
which respects the asymptotic behavior at infinity. A simple means consists of
weighting the two formulae by a suitable auxiliary function M(t) close to 1 for small t
and tending quite quickly to zero for t equal to infinity. The expression:
f t M t f1 t 1 M t f 2 t
satisfies these conditions if the function M(t) at least satisfies the relations
M 0 1, M ' 0 M f 0 and if M'(t) tends at infinity faster to zero than f 2 t ; we
have:
f 0 f1 0 f ' 0 f1' 0 f f f 2 f f ' f f 2' f
~2
The simplest weighting function is the Gaussian M t
~
e D t .
It remains to write a matching condition which can be defined at a point where

the two approximations differ very little and where we require that the value of the
function M(t) is equal to 0.5 (here, the function f(t) is the mean of the values f1(t) and
f2(t)).
Let us apply this procedure to obtain a quite simple expression of the solution
valid all over the interval [-A,+A]. We take as our asymptotic expression at infinity
the modal solution limited to the first mode, and for small t, the boundary layer
solution. We will choose the mean temperature in order to determine the matching
condition. A simple numerical calculation shows that the difference between the
values of expression [8.41] limited to the first mode and formula [8.45] is minimal
~
in the vicinity of t 0.20 . Taking as a weighting function the Gaussian
~2
M t e D t and taking M(0.2) equal to 0.5, we find D = 7.5.
~
The mixed representation thus obtained for the first mode of the temperature Tm
can be written:
Tm t
T0 2 t ¯
2
e 7.5. t 2 t 1 e7.5. t 2 ¡1 8 .e 4 °

[8.46]
T1 T0 ¡ 2 °
¡¢ °±
The corresponding expression for the wall’s thermal flux density is thus:
AqTp A, t
1 2

2.e
t
7.5. t 2 7.5. t 2
e 1 e 4
[8.47]
T1 T0
t
The preceding formulae represent the exact solution to 1% accuracy.
The spatio-temporal temperature distribution can also be written in the same

manner from expressions [8.40] and [8.43]:
T T0 ¬ ¬¯
2
e7.5. t ¡ 2 erf x 1 erf x 1°
¡
T1 T0 ¢¡
2 t ® 2 t ®°±°
4 2 t x ¯°

1 e7.5. t ¡¡1 e 4 cos
2

2 °±°
¢¡
The interest in simple analytic expressions is clear; furthermore, the precision of
the approximations effected can be improved as much as we desire by conserving
additional terms of the modal solution in the preceding composite solution.
8.3.2.3. Thermal systems with continuous components

8.3.2.3.1. Conduction in two walls separated by a thermal resistance
The separation of variables method can be applied to continuous media by
individual segments which comprise discontinuities. We will reconsider the linear
invariant system with two identical components discussed in section 8.3.1.2 with a
continuous medium model of constant thermal diffusivity a; the sub-systems are
separated by the thermal resistance R per unit cross-section, located at the origin
(Figure 8.14). We will assume that the thermal flux densities are given on the
external faces, the external thermal resistance R’ being taken as zero.
With equation [8.32] on the intervals [–A,0–] and [0+,A], we must associate:
– the matching condition involving the thermal resistance between the two sub-
systems:
§ wT · § wT · T (0 ) T ( 0 )
O ¨¨ ¸¸ O ¨¨ ¸¸ [8.48]
© wx ¹0 © wx ¹0 R
– the boundary conditions of the system: thermal flux densities given (or
eventually the temperature or thermal resistance) at the points x rA ;
– the initial temperature distribution T x,0 .
T1ex T(0-) T2ex

T(0+)
T(x)
M1w M2w
M1,2
–A A x
E1 E2
O
Figure 8.14. Thermal conduction in a system of two continuous media

separated by a thermal resistance
As before, the complete solution can be obtained by superposition of a solution

which satisfies the non-zero boundary conditions and a general solution of the heat
equation with homogenous boundary conditions. The coefficients of this general
solution are calculated such that the complete solution satisfies the initial conditions.
8.3.2.3.2. General solution of the homogenous problem

As an example, we will treat the problem where a thermal flux is imposed at the
interval extremities [–A,+A]. The condition at x = ± A of the homogenous problem is
thus a zero thermal flux. As above, we write the homogenous problem with non-
dimensional variables [8.33]. We add to equation [8.34] the boundary conditions and
thermal resistance matching condition [8.48], which can be written in non-
dimensional form as:
§ wT · § wT · § wT · A
¨¨ ~ ¸¸ 0; ¨¨ ~ ¸¸ ¨¨ ~ ¸¸ >T (0 ) T (0 )@ [8.49]
© wx ¹ r1 © wx ¹ 0 © wx ¹ 0 OR
The problem depends on the parameter P A O R , which represents the

importance of the thermal resistance A O of a component with respect to the thermal
resistance R.
The solutions to equation [8.34] where the variables are separated as

T ~
x , t f t g ~
~ ~
x again satisfy relations [8.35] and [8.36] for the function
~ 2~
f ( t ) exp(/ t ) .
The eigenvalue problem for the eigenfunction g can now be written:
g" ( ~
x ) / 2 g (~
x) 0;
A [8.50]
g ' r 1 0; g ' (0 ) g ' (0 ) >g (0 ) g (0 )@
OR
The integration of equation [8.50] leads to eigenfunctions of the form

g ( x ) A cos / x B , the constants (A, B) taking on, respectively, the values (A,
B) and (A+, B+) on the intervals (-1,0) and (0,1):
1 d ~
x 0: g (~
x) A cos/ ~
x B
~
0 x d1: ~
g(x ) A cos/ x B

~

The expression of the boundary conditions at ~

x r1 gives:
A ./ sin / B A ./ sin / B 0
or:
/ B p ' S B pS p, p' : integers [8.51]
Matching condition [8.49] for the thermal flux densities at x = 0 can be written:
/ A sin B / A sin B P>A cos B A cos B @ [8.52]
Equations [8.51] and [8.52] possess two groups of solution4:

1) Even eigenfunctions. We find immediately the solutions to equations [8.51]
and [8.52]:
B B 0; A A ; / 2k kS (k integer) [8.53]
The evenness of the eigenfunctions g 2k ~ x coskS ~x of the two-wall

ensemble leads to the absence of thermal transfer between the two sub-systems
which behave as a single symmetric block. The first four eigenfunctions (k = 0, 1, 2,
3) are represented in Figure 8.15a (we have taken A A 1 ).
4 We will number the functions with even numbers (respectively odd) for even eigenvalues
(respectively odd) with increasing order of the eigenvalues.
gi
+1
-1 +1
~
x /
/ / / /
-1
0 S S S S S S S
: g0 : g2 : g4 : g6 : 2P// : tan/
(a) (b)
Figure 8.15. (a) Even eigenfunctions of two walls separate by a thermal resistance;
(b) graphical solution of the eigenvalue equation in the case of odd eigenfunctions
2) Odd eigenfunctions. Substituting the values [8.51] of the pairs (B–, B+) into
first relation [8.52], we obtain a second group of solutions (B v 0, B v 0):
B B ; A A or B B S ; A A
We derive the relation:

A . cos B A . cos B
Substituting this expression into [8.52], and taking account of [8.51], we obtain
the eigenvalue equation:
/ tan / 2P [8.54]
whose graphical solution is shown in Figure 8.15b. The eigenfunctions g 2k 1 ~ x

which correspond to these eigenvalues are here odd and by taking A A 1
they present a discontinuity amplitude equal to 2 cos / 2k 1 at the origin. They can
be written:
0 x d 1: g 2k 1 ( ~
x) A2k 1 cos / 2k 1 ~
x 1
These discontinuous eigenfunctions depend on the dimensionless parameter P.

We have represented, in Figure 8.16b, the first four odd eigenfunctions
corresponding to P 0.05 (eigenvalues /1=0.3111, /3=3.1731, /5=6.2991,
/7=9.43538).
g1 gi
+1 +1
-1 +1 -1 +1
~x ~x
–1 –1
P = 0.05 P = 0.5 : g1 : g3
P=3 P=f : g5 : g7
(a) Variations of the eigenfunction (b) Discontinuous eigenvalues for
g 1 for different values of P. P = 0.05.
Figure 8.16. Odd eigenvalues for two walls separated by a thermal resistance
8.3.2.3.3. Evolution of odd eigenfunctions

The value of thermal resistance placed between the two walls (in other words the
value of P) has a strong influence on the discontinuity amplitude. Figure 8.16a
shows the evolution of the first odd eigenfunction as a function of P. The small
values of P (high thermal resistance R) lead to eigenvalues of the odd functions
which are thus quite close to kS, which are the eigenvalues of the even functions
(Figure 8.15b). The discontinuity at the origin 2 cos / 2k 1 of the odd eigenfunctions
is strong, and the thermal flux / 2k 1 sin / 2k 1 between the two blocks is all the
smaller as k is large. The absolute values of the odd and even eigenfunctions are thus
very close in each of the two continuous media. A homogenization of the
temperature occurs in each block, the thermal flux / 2k 1 sin / 2k 1 between them
being essentially limited to the first odd mode.
If the thermal resistance R tends to zero, P becomes large and the eigenvalues
S
/ 2k 1 tend to the odd eigenvalues kS of the total isolated system. We thus
2
2k 1 S x
find the odd eigenfunctions sin of the conduction problem for an
2
insulated wall (and of course, the temperature continuity at the origin).
In expression [8.17] of the modal development of the temperature, the

dimensional time constant Wk of each mode is directly related (formulae [8.33] and
[8.36]) to the corresponding eigenvalue /k:
Wk A 2 a/2k
For small values of P, the first term of the development of /1 as a function of P

is equal to 2 P (equation [8.54]). Substituting this value into the above expression
and replacing P by A O R , we find for the time constant W1:
A2 A2 U CAR
W1
a /21 2aP 2
We find the time constant of the thermodynamic model of section 8.3.1.2 in the
particular case where we have R' = 0, with m U A .
8.3.2.3.4. Expression of the solution

The complete solution of the problem can be obtained from formula [8.18] as the
development of the difference T ( x,0) Te ( x,0) as a series of eigenfunctions of the
problem; these are orthogonal if the operator of the problem is self-adjoint
(Appendix 4). The verification of this property can be checked easily by separately
considering the intervals [-1,0–] and [0+,1]; and taking account of the condition
g ' 1 0 , we have, after integration by parts, for p not equal to q:
~ g p 0 g q' 0 ³1 g 'p g q' d~

0 '' 0 ' 0
³1 g p g q dx ³1 g p dg q x
Proceeding in a similar manner for the other interval, we obtain:
1
> '' '' ~
³1 g p g q g q g p dx @
g p 0 g q' 0 g q 0 g 'p 0 g p 0 g q' 0 g q 0 g 'p 0
Taking account of conditions [8.48] for the thermal resistance at x = 0 and after
'' ''
replacing g p and g q by their expressions taken from equation [8.50], we see that
1
g p gq'' gq g p'' ¯ dx is zero if p is not equal to q:
the integral ¨
1
¡¢ °±
1 1
g p gq'' gq g p'' ¯ dx / 2p / q2
¨ ¡
1 ¢ ±° ¨1 g p gq dx 0
The eigenfunctions gp and gq are orthogonal for the scalar product
f , g ! ³1 fgd~
1
x.
The complete solution [8.18] can be written:
f 2 2~ ~
~ ~
T (~ x , t ) ¦ §¨ c 2 k e k S t cos(kS ~
Te ( ~ x ) ·¸ [8.55]
x ) c 2 k 1 e / 2 k 1 t g 2 k 1 ( ~
2
x, t )
k 0 © ¹
1
with c2 k 1
³ >T( x, 0 ) T ( x, 0 )@ g
1 e 2 k 1 ( x )dx
k 0 ,1, 2 ,...,n,...
1
³ g ( x )dx
2
1 2 k 1
1 1 1
c0
2 ³1
>T( x, 0 ) Te ( x, 0 )@ dx;
c2 k ³ >T( x, 0 ) T ( x, 0 )@ cos( kS x )dx k
1 e 1, 2,...,n,...
8.3.2.4. Modal representation of systems and number of parameters

The modal representation of a discrete system consists of replacing the n state
variables xi t , functions of time, with n coefficients ci of the expression of the
solution on the basis of the eigenfunctions whose time dependence is known. We
thus see that, assuming the eigenfunctions to be known, the number of numerical
values (series development coefficients) which characterize the solution decreases
with time.
The same is true for the preceding models for continuous media, for which we
replace the temperature distribution T(x,t) in two continuous variables by a
denumerable sequence of series development coefficients ci [8.55] of eigenfunctions
associated with the physical model used. In a manner analogous to the development
of a periodic function in a Fourier series, the information necessary to characterize
the solution has been considerably reduced through the use of basis functions
adapted to the problem.
Furthermore, these developments offer the advantage that they provide an

organization of information. We saw in Chapter 7 that the frequencies associated
with the terms of a Fourier series are greater as the order of these terms is high. This
thus results in criteria regarding: the nature of the approximation which has been
made where only a limited number of terms are retained, and regarding the sampling
of the corresponding temporal signals (Shannon’s theorem). The modal
representation of thermal systems indicates that the wealth of information decreases
as time increases. For continuous media, this information diffuses from the wall and
becomes progressively poorer. The results are that the representation using space-
time numerical data (x,t) is very variable. During the first instants, we need to
discretize the time and near-wall zones very densely, while the central zone of the
wall is not subjected to any phenomenon. For larger times, the entire wall needs to
be discretized, but less densely, and the same goes for the time variable. We will
note that, as the modes vanish successively, the quantity of information contained in
the solution decays with time, the effect of the second law of thermodynamics
(increasing entropy) being to homogenize the temperature.
8.4. External representation of linear invariant systems
8.4.1. Overview
Transitional solutions of differential system [8.13] of n linear first order

differential equations with n unknowns form a vector space of dimension n. They
allow the characterization of all possible states described by [8.13]. In practice, it
often happens that the initial data of a category of problems is of reduced number p,
the system starting for example from a given standard state (rest for example). The
transitional solutions which correspond to these initial conditions belong to a family
which depends only on these p parameters. The state representation of a system of n
state variables thus constitutes an overly detailed model. Rather than establishing a
state representation, we can be satisfied with the study of system responses by
means of an external representation by limiting ourselves to the effective inputs and
initial conditions.
Consider for example a linear invariant system for which the inputs are zero.
This is the modeling of an isolated homogenous system which evolves towards
equilibrium and which is characterized by a zero value for the state vector and the
outputs. Suppose that we impose as initial conditions given values (step functions
for example) for p components (p < n) of the state vector, the other components
being zero. This initial vector must be decomposed on eigenvectors of the
homogenous system, and apart from some particular cases, none of these
components is zero. The responses to these excitations comprise the n eigenmodes,
but they belong to a family of only p parameters which form a vector sub-space of
dimension p of the output space, which we can study directly.
8.4.2. External description of linear invariant systems

8.4.2.1. Impulse response
The solution corresponding to diverse inputs and initial conditions is the sum of
solutions for each of the inputs and initial conditions taken separately, the others
being considered zero. This superposition property for solutions of time-invariant
linear systems allows us to separately study the response to different inputs and
initial conditions.
Consider firstly a system with a single input u1 to which we apply a Dirac

distribution at the initial instant, the system being initially at rest; the output vector
hY1 t represents the evolution of the system outputs to this impulse on the first
input. Now, any given input u1 t can be considered as the superposition of Dirac
distributions:
u1 t ³0 u1 W G t W dW
f
The response Y1 t of the system outputs to the excitation u1 t can be written:
Y1 t ³0 hY 1 t W u1 W .dW
f
[8.56]
The impulse response of the output vector therefore makes the characterization
of the response of a stationary linear system to any input excitation possible.
Proceeding similarly for the group of p inputs, we obtain a rectangular matrix

hY1 t , hY 2 t ,..., hYp t of dimension q * p formed with the p corresponding output
vectors, the entire group of which constitutes the matrix H Y (t ) of the impulse
response output vectors of the system. This allows us to calculate the output
corresponding to any given input vector U (t ) :
Y t ³0 H Y t W .U W .dW
f
[8.57]
The impulse response matrix H Y (t ) constitutes an external description of the

system taken to be initially at rest. Easily realizable experimentally, it can be
obtained using only impulse excitations, without any knowledge of the internal
variables in the initial reference state. It also allows us to describe the evolutions
corresponding to the given non-zero initial states of the system which are obtainable
by using suitable impulses on the inputs (section 8.2.1.5).
As an input vector is usually of smaller dimension than the state vector, the
impulse response matrix H Y (t ) is a reduced model of the system adapted to its
operating conditions and which is not equivalent to the state representation of
section 8.1.2.
This matrix contains p*q time functions whose useful duration is equal to a
number of times the largest damping time constant. The quantity of information is
here much greater than in the case of the state representation: this is the result of the
absence of a model. For the linear time-invariant5 thermal system, each of the
5 Responses of stable systems usually studied in fluid mechanics and acoustics present an
oscillatory damped character.
functions is a sum of decaying exponentials which we can seek to identify

approximately, at least for the first of these. In the absence of a state representation,
any interpretation of results is difficult and the obtaining of the structure of the
eigenmodes of the system is problematic.
The result of formula [8.57] is that at a given instant, the state of a linear
invariant system depends only on the inputs belonging to a past which is at most
equal to the duration of the impulse response.
NOTE – The preceding procedure can also be applied to a state vector X(t). The
application of a Dirac impulse as an input of index i provides the state vector h Xi t .
Operating in a similar fashion on the ensemble of inputs, we obtain the matrix
H X (t ) of impulse responses of the state vector, of dimension n * p, which allows us
to obtain the response of the state vector to any given input U(t):
X t ³0 H X t W .U W .dW
f
However, the direct measurement of state variables is not easy in general, and
the impulse response matrix H X (t ) of the state vector is only of limited interest.
8.4.2.2. Inputs-outputs analytic representation

The complete external representation of a system can be obtained from its state
representation, cast in the form of a scalar differential equation of order n (section
5.1.1 and section 8.1.2) in a scalar state variable x (t). Let us take the simple
example (n = 3) of the state representation with a single output y(t):
a 0 x a1 x' a 2 x' ' a 3 x' ' ' u; y d 0 x d1 x' d 2 x' ' [8.58]
Calculating the linear combination d 0u d1u 'd 2u '' and eliminating x(t) and its
derivatives with [8.58], we obtain a differential equation for the output y (t):
a0 y a1 y 'a2 y '' a3 y ''' d 0u d1u 'd 2u '' [8.59]
The method can be generalized to a differential equation of arbitrary order:
n p
¦ ai x i u; yj ¦ d jk x k [8.60]
i 0 k 0
We obtain:
n n p p
i
¦ ai y j ¦ ¦ ai d jk x i k ¦ d jk u k [8.61]
i 0 i 0k 0 k 0
The inverse passage of the preceding external representation to a state

representation can be achieved easily, the coefficients of relations [8.60] and [8.61]
(or [8.58] and [8.59]) being identical.
The representation of a dynamic system of n variables, effected by means of a

differential equation of order n, can be replaced by a relation of a differential
nature between the inputs and the outputs. However, obtaining the preceding
external representations by means of measurements, without any knowledge of the
structure of the system, is impractical for complex systems: the evaluation of the
temporal derivatives obtained from differences between measurements lead to errors
which will be greater as the order of the differentiation is increased.
8.4.2.3. Example of a thermal system with two variables

As an example, consider the state representation of the two-component system of
section 8.3.1.2:
T1' a11T1 a12T2 u1 ; T2' a 21T1 a 22T2 u 2 [8.62]
where the quantities ui and aij are defined in equation [8.21]:
1 §1 1· 1 Tiex
a11 a 22 ¨¨ ¸¸; a12 a 21 ; ui i 1,2 .
mC © R R ' ¹ mCR mCR '
Differentiating the first equation [8.62] and eliminating T2' by means of second
relation [8.62], we obtain a second order differential equation satisfied by T1:
a11a 22 a12 a 21 T1 a11 a 22 T1' T1' ' a 22 u1 u1' a12 u 2 [8.63]
Taking the output variable in the form [8.58]: y d 0 T1 d1T1' , the external
representation can be obtained from [8.59]:
a11a 22 a12 a 21 y a11 a 22 y ' y ' ' d 0 u d1u ' [8.64]
with: u a 22 u1 u1' a12 u 2 .

8.4.2.4. Laplace transforms and operational transfer matrices

We have already seen in Chapter 7 the value of Fourier transforms for the study
of oscillatory signals. The Laplace transform (Appendix 1) presents an analogous
interest for the study of aperiodic signals and damped systems. The Laplace
transform Lf of a function f(t) is defined by the relation:
L f p ³0 f t e dt
f pt
[8.65]
It possesses diverse properties which are analogous to those of the Fourier

transform (section 7.3.4.2); in particular, the Laplace transform of the derivative
function f is written:
L f ' p pL f p f 0
This property allows us to transform the system of differential equations and the
initial conditions into a system of algebraic equations between the transforms of the
variables. For example, the differential equation transform:
mx f x kx u t
can be written:
m p 2

f p k L x px0 x' 0 Lu
or:
Lu p px0 x 0
Lx p [8.66]
m p2 f p k
We thus obtain the transform of the complete solution written as a function of

the initial conditions and the transform Lu(p) of the right-hand side (the input). All
that remains is to perform the inverse transform (Appendix 1).
The transform linearity allows it to be applied to a vector or to a matrix of

functions. For the initial conditions X 0 , the Laplace transform of the state
representation [8.1] of a linear invariant dynamic system can be written:
pL X X 0 AL X B.LU ; LY D.L X
We can thus deduce the transforms of the solution X t and of the output vector
Y t :
LX p
pI A
¢ BLU p
X 0
¯± : LY D.LX .
-1
[3.67]
As the inverse of a matrix M is equal to the ratio between the transpose of its
comatrix6 and its determinant ([HAR 98]), the elements of the matrix pI A of
1
rank n are rational fractions whose denominator is equal to pI A , the numerator

being a polynomial in p whose order is at least equal to n. This rational fraction can
be decomposed into simple elements associated with the poles (zeros of the
denominator) which are the eigenvalues of the system studied. Each of these
elements is the Laplace transform of the real or complex exponential functions
which correspond to the eigenmodes of the system studied.
The matrix H p D. pI A1 B is the operational transfer matrix of the

system:
LY p D. pI A1 B.LU p H p .LU p [3.68]
As the Laplace transform of a Dirac distribution at instant t = 0 is equal to 1, the

transfer matrix H p is the Laplace transform of the impulse response matrix
H Y (t ) .
The Laplace transform is particularly useful in the domain of system controls

which are essentially beyond the scope of this textbook, despite applications in the
domain of flow and transfer phenomena.
8.5. Parametric models
8.5.1. Definition of model parameters
A parametric model can be defined as an exact or approximate solution whose

state vector X(t) can be written in the form of a combination (linear or otherwise) of
some state vectors Xj(t) independent of the inputs; the coefficients Pj of this
combination are time functions which are deduced from the input vector U (t ) (the
coefficients Pj are functionals of U t ). For a linear combination, we have:
6 I.e. the matrix of cofactors of each element.

X t ¦ X j P j U t [8.69]
j
Expression [8.69] is quite analogous to a series development of eigenfunctions.

Like these, the vectors Xj are universal, as they do not depend on any particular
inputs of the problem. The coefficients Pj are parameters characterizing, in a simple
manner, the solution represented at each instant as a basis composed of state vectors
Xj which are often chosen in order to best represent the solution for a class of given
inputs. Their non-dimensional expressions are often called form parameters, as they
characterize the relative value distribution of the state vector components (see
example in section 8.5.2). We have adopted the terminology parametric model in
reference to the idea of form parameters often used in fluid mechanics.
The output vector Y D. X can be written in an analogous form [8.69] with the
universal vectors D Xj and the parameters Pj:
Y t D X t ¦ D X j P j U t [8.70]
j
We can note that the solution of the system obtained by solving equations [8.14]
and [8.15] of the state representation using the variation of constants method is not
in general of the form [8.69] or [8.70].
Solutions of parametric type can also be obtained in an approximate manner by

means of global methods. Some examples of parametric solutions of linear systems
will be covered in the following section. Such exact or approximate solutions are
also used in fluid mechanics for the study of the boundary layer (Blasius and
Howarth series, Görtler series, Karman-Polhausen global method, etc. ([SCH 99],
[YIH 77])).
The value of the parametric methods is quite significant:

– with the objective of understanding and interpreting physical phenomena: the
values of the parameters Pj allow in effect the classing of excitation situations by the
function U(t) by characterizing these situations using simple analytical forms of the
solution. We will see examples of the application;
– for writing models of complex systems: expression [8.70] is an external explicit
representation of a system in which the inputs and outputs are variables which
characterize the interactions between neighboring systems; limited to a small
number of terms, it constitutes an approximate representation which is particularly
useful for modeling of interacting systems.
8.5.2. Established regimes of linear invariant systems
8.5.2.1. Systems described by a state representation

In section 8.2.2.1 we defined an established regime as a solution of a differential
system whose expression does not contain transitional terms associated with the
initial conditions. The solutions of discrete time-linear linear systems can be cast in
a simple parametric form, at least when the inputs do not vary too rapidly. The
guiding idea of the method has already been outlined in section 6.4.2.6; it consists of
considering the unsteady term of state equation [8.1] as a perturbation [SAD 93]. By
introducing the parameter H which we will take to be equal to one in the following,
the state equation [8.1] can be written:
dX
H A. X B.U [8.71]
dt
We seek the solution of equation [8.71] in the form:
f
X t ¦ H i X i t [8.72]
0
By substituting expression [8.72] into equation [8.71], and by identification

according to the increasing powers of the parameter H, we obtain the system:
A. X 0 t B.U t 0; ! ! !! !
A. X 1 t X 0' t ; A. X i t X i' 1 t ; [8.73]
A. X 2 t X 1' t ; ! ! !! !
Assuming the matrix A to be invertible, we deduce:
X 0 t A 1 B.U t ; ! ! !! !
X 1 t A 2 B.U ' t ; X i t A i 1 B.U (i ) t ;
X 2 t A 3 B.U ' ' t ; ! ! !! !
Substituting and taking H to be formally equal to 1, we finally obtain:

f
X t A 1 B.U t ¦ A i 1 B.U (i ) t
i 1
[8.74]
Y t C. X D.U > @
ª f º
CA B. D .U t C.« ¦ A i 1 B.U (i ) t »
1
¬ 1 ¼
Series [8.74] is a general solution to the established regime of system [8.1].

Independent of the initial conditions, it converges if the successive derivatives of the
input vector U do not vary too quickly, for example if the quantities u i t are
bounded by k/i1 (/1: smallest eigenvalue of the matrix A, k constant).
The idea of form parameters can be introduced by means of the simple example
of a system with a scalar input u t . Solution [8.74] can thus be written:
ª f u (i ) t º
X t u t « A 1 B. ¦ A i 1 B. » [8.75]
«¬ i 1 u t »¼
i
The parameters u t u t are form parameters of the problem and determine
the structure of the solution.
The Laplace transform allows the preceding results to be retrieved by giving an

-1
explicit expression of the term pI A
BLU p
of the established regime of
equation [8.67]. We obtain:
pI A1
A 1 I pA 1
1
>
A 1 I pA 1 p 2 A 2 .. p n A n .@
or:
LX A pI 1 B.LU >
A 1 I pA 1 p 2 A 2 ... p n A n .... B.LU @
This expression is indeed the Laplace transform of the first equation [8.74].
NOTES –
1) Temporal series of the form [8.74] (or [8.76]) which describes in parametric
form how the established regimes can be derived from responses obtained by the
convolution integral [8.57] in which we perform a Taylor series development. For
example, for the state vector, taking account of the commutativity of the convolution
product, we have:
d 2 d ¯
X t
¨ U t
.H
d ¨¡U t
U ' t
U ' t
...°H
d
0 0 ¡ 2 °
¢ ±
d d d 2
U t
¨ H
d U ' t
.¨ H
d U '' t
¨ H
d ...
0 0 0 2
2) The perturbation method used can also be applied to non-linear equations.
8.5.2.2. Case of representation by differential equations of order n

The preceding method can be applied to write the established system and the
solution to the differential equation [8.60] of order n, representing system [8.1], with
a single state variable x (t):
n n 1
¦ a i x i u t ; yj ¦ d jk x k
i 1 k 0
We obtain, for the established regime:
§ n · ª n 1 kº
¨ ¦ ai p i ¸L x p Lu p ; Ly j « ¦ d jk p » L x p
¨ ¸
©i 1 ¹ ¬k 0 ¼
or:
n 1
¦ d jk p k
Lu p
Lx p ; Ly j k 0
Lu p
n n
i
¦ ai p ¦ ai p i
i 1 i 1
The established regime for small values of p can be obtained by calculating the
above fraction series development following the increasing powers of p, for
example, by means of a division following increasing powers.
By limiting ourselves to the third term, we obtain:
Lu p
Lx p
§ a ·
a 0 ¨1 i p i ¸
¨ a0 ¸
© ¹
1 §¨ a1 § a12 a 2 · 2 § a 3 2a 2 a1 a13 · 3 ·
1 p ¨ ¸p ¨ ¸ p ...¸ L p
a 0 ¨¨ ä ¸¸ u
a0 ¨ a 2 a0 ¸
© 0 ¹ © 0 a 02 a 03 ¸¹
© ¹
whence the parametric expression for the state variable x(t):
xt A0 .u t A1u 't A2u"t A3u ''' t ... [8.76]
1 a1 a12 a2 a3 2a2 a1 a13

with A0 ; A1 ; A2 ; A3 3 4 ;....
a0 a02 a03 a02 a02 a0 a0
8.5.2.3. External representations

The same procedure can be employed for the input-output representation [8.59]:
a 0 y a1 y ' a 2 y ' ' a3 y ' ' ' d 0 u d1u ' d 2 u ' '
for which the Laplace transform can be written:
a 0
a1 p a 2 p 2 a 3 p 3 L y p d 0
d 1 p d 2 p 2 Lu p
or:
d 0 d1 p d 2 p 2
L y p Lu p
a 0 a1 p a 2 p 2 a 3 p 3
By dividing the fraction in p following increasing powers, and then returning to

the space-time domain, we obtain the differential representation:
d0 §d d a · § d a2 d a d a d ·
y t u t ¨ 1 0 1 ¸u 't ¨ 0 1 0 2 1 1 2 ¸u"t ...
a0 ¨ a0
© a 02 ¸¹ ¨ a3
© 0 a 02 a 02 a 0 ¸¹
8.5.2.4. Harmonic established regimes

A system can also be characterized in spectral space. We thus proceed in a
manner analogous to the case of impulse excitation, but with harmonic excitations in
complex form e jZt . The established harmonic regimes of the linear invariant
systems can be studied by means of the complex amplitudes method by searching
jZ t
for solutions in the form X Xˆ e , where we denote by M̂ the constant
complex matrix associated with the matrix function M t M ˆ e , the temporal jZt
derivation having been replaced by a multiplication by jZ. System [8.1] can be

written:
jZ Xˆ A. Xˆ BUˆ [8.77]
Solving linear algebraic system [8.77] makes it possible to calculate complex

amplitudes of the state vector Xˆ (Z ) and the output vector Yˆ (Z ) for all values of
the pulsation Z. We obtain complex algebraic equations analogous to those obtained
by Laplace transform for the established regimes, where the variable p is replaced by
jZ. The complex vector amplitudes Xˆ (Z ) and Yˆ (Z ) can be written:
Xˆ ($ ) j$ I A
B.Uˆ ($ );
1
Yˆ ($ ) Hˆ ($ ).Uˆ ($ ) with: H $
D. j$ I A
B
1
The matrix H (Z ) is the transfer matrix of the system between the inputs and the
outputs. It is the Fourier transform of the impulse response matrix.
The harmonic regimes of external representations obtained above can be

obtained in the same way. For example, from relation [8.59] we can derive the
complex amplitude of the output ŷ as a function of the input uˆ($ ):
d 0 jd1Z d 2Z 2
yˆ
u
ˆ
a0 a 2Z 2 jZ a1 a3Z 2
The parametric representations obtained earlier can be written in the harmonic

regime, by performing a series development in increasing powers of jZ. For
example, with the discrete system of equations [8.1], we have, from [8.74]:
f § f ·
Xˆ ¦ jZ n A n 1 B.Uˆ .; Yˆ Z ¨ D C ¦ jZ n A n 1 B ¸.Uˆ
¨ ¸
n 0 © n 0 ¹
The preceding series developments in Z are a valid approximation for the low
frequencies. We will obtain in section 8.5.3 an estimation of the radius of
convergence of the preceding series in Z in an example.
In the study of certain vibration problems (mechanical, electrical and

electromechanical, acoustics), we define impedances which are functions of the
transfer between the intensive quantities (forces, electric potentials, acoustic
pressure) and the flux of associated extensive7 quantities (velocity, electrical
intensity, acoustic flux) evaluated at the same point. These impedances are in fact a
7 These extensive quantities can be reduced, for instance in the case of mechanical
impedances.
particular kind of transfer function. We will assume that these ideas are known to
the reader at least in the domain of electricity.
NOTE – The different external representations discussed above are equivalent. In

practice, we use the form which is best adapted to the problem under study.
8.5.3. Established regimes in continuous media
8.5.3.1. Systems described by a continuous state representation

Consider the cylinder of axis Ox, of cross-section limited by the arbitrary curve
C (Figure 6.13 of section 6.4.2.6) and on the wall of which we impose the uniform
temperature Tw(t). The 2D temperature distribution T y, z , t satisfies the equations
and conditions:
wT
a 'T ; with : y, z C : T y, z , t Tw t [8.78]
wt
Let us take the dimensionless variables:
~ ~z ~
y y A; z A; t a t A2
Equation [8.78] becomes:
wT
with : ~y , ~z C : T ~
y , ~z , t Tw t
~ ~ ~
~ 'T [8.79]
wt
Let us look for a solution in the form [PEU 84]:
f
¦ g i ( ~y , ~z )Tw(i ) t
~ ~
T (~
y , ~z , t ) [8.80]
i 0
Substituting into [8.79] and identifying the terms corresponding to the

derivatives Tw(i ) t , we obtain:
~
' g0 0; ' g1 g0 ; ' g2 g1 ; ... ' gn gn1 ; ...
with the boundary conditions: y , z

C : g0 y , z
1; gi y , z
0 i 1,..., n,...
The functions g i ~
y , ~z can thus be calculated from place to place. The reader
can immediately verify that these are identical to the corresponding functions of the
unsteady flow problem of a viscous fluid in a cylindrical tube of contour C (section
6.4.2.6), a difference of unity excepted for the value of index i.
The method used can be applied to continuous systems containing discrete

elements. Let us take the system constituted of two walls separated by a thermal
resistance (section 8.3.2.3.1). Consider here the case where one of the faces is
maintained at constant temperature T A 0 , the other face at abscissa +A being
subject to the variable temperature Tw(t). Taking the dimensionless variables [8.33]
~
(~x x A and t a t A 2 ), we obtain the heat equation on the intervals [-1,0] and
[0,1]:
sT s 2 T
; with:-1 b x 0, 0 x b 1 [8.81]
st sx 2
As the thermal resistance R is characterized by relation [8.49] (with P A O R ),

we have the boundary conditions (section 8.3.2.3.1):
§ wT · § wT ·
Tw t ; P >T (0 ) T (0 )@ [8.82]
~
T 1 0; T 1 ¨¨ ~ ¸¸ ¨¨ ~ ¸¸
© wx ¹ 0 © wx ¹ 0
We can easily verify that the equation and the preceding boundary conditions
possess a solution analogous to [8.80]:
f
¦ g i ~x Tw(i ) t
~ ~
T (~
x, t ) [8.83]
i 0
The functions g i ~
x satisfy the following successive relations and conditions:
g 0' ' ~
x 0; g1' ' ~
x g 0 ~
x ; g 2' ' ~
x g 1 ~
x ; !! g n' ' ~
x g n 1 ~ x!
A
i 1, ! , f : g i 1 g i 1 0; g i' 0 g i' 0 >g i 0 g i 0 @
OR
g 0 1 0; g 0 1 1;
We can easily calculate the first two functions:

P ~
x 1 P ~
x 1 ª ~ 8P 6 º
1 d ~ «x 1
2
x 0 g0 ; g1 »
1 2P 61 2 P ¬ 1 2P ¼
P ~
x 1
0 x d1 g0 1; [8.84]
1 2P
Px 1 ª ~ 8P 6 º ~x2 1
«x 1
2
g1 »
61 2 P ¬ 1 2P ¼ 2
The first function g 0 ~x is the temperature distribution in the steady regime.

The reader can easily certify that this corresponds to the thermal resistances in series
associated with two continuous media and to the resistance R. The thermal flux
density on the walls is derived from the following expressions:
§ wT · O §¨ P T p t P4 P 3 A 2 ·
qTp A O ¨¨ ¸¸ T p' t !¸
© wx ¹A A ¨© 1 2 P 31 2 P 2 a ¸
¹ [8.85]
§ wT · O P T p t ª
§ P4 P 3 º A 2 ·
qTp A O ¨¨ ¸¸ ¨ «1 » T p' t ! ¸
© wx ¹A A ¨©1 2 P «¬ 31 2 P 2 »¼ a ¸
¹
More generally, M being an operator which only contains spatial derivatives, let
us consider equation [8.86]:
wf
M f [8.86]
wt
which possesses parametric solutions of the form:
f
T ( x, y , z , t ) ¦ g i x, y, z Tw(i ) t
i 0
The first function g 0 x, y, z is the spatial distribution of the function f in the

steady regime corresponding to the steady conditions imposed which define the
reference temperature Tw. The universal functions gi satisfy the successive
equations:
Mg0 x, y, z 0; Mg1 x, y, z g0 x, y, z ; ! Mg n x, y, z g n 1 x, y, z ; !
For example, the thermal conduction equation for a cylindrical problem in a

heterogenous medium can be written:
wT 1 ªw § wT · w § wT ·º
« ¨¨ O y, z ¸ wy ¨ O y, z wy
¸ ¨ ¸»
¸
wt U y, z c y, z «¬ wy © wy ¹ © ¹»¼
s sT ¬
It corresponds to the operator M a y, z
y, z
:
sy sy ®
1 ªw § w · w § w ·º
M « ¨¨ O y, z ¸¸ ¨ O y , z ¸»
U y, z c y, z ¬« wy © wy ¹ wy ¨© wy ¸¹¼»
The case where the cylinder is the seat of a volume heat source VT(t) can be
treated in an analogous manner to that of the problem of viscous flow in a
cylindrical tube, as studied in section 6.4.2.6:
sT
T t
a 'T ; with: y, z
C : T y, z , t
0
st
The equations of the universal functions g have been established in section

6.4.2.6. The parameters Pj of formula [8.69] which characterize the input are here
i
the successive derivatives V T t .
8.5.3.2. Established boundary layer regimes

The regular perturbation method used for parametric solutions in a finite domain
can be extended to the case of boundary layers, on the condition that the problem be
cast as a regular problem. Let us reconsider the example of a thermal shock in a
semi-infinite medium as described in section 5.4.5.4 (and applied at the end of
section 8.3.2.2.2) and verifying the heat conduction equation [8.78] but with the
variable wall temperature Tw(t). In the plane (x, t) we have a singular perturbation in
the vicinity of x = 0 where the solution obtained presents very different scales: for
small time values the spatial variations of temperature are rapid, whereas for large
time values they are slow. The convergence of solutions close to this base solution
towards this solution cannot therefore be uniform. The variable adapted to the
boundary layer K x 2 at is defined as in section 5.4.5.4. We first of all perform
a change of coordinates x, t o K , W , in order that the difference between the two
solutions can be characterized in a uniform manner:
K x 2 at W t; T x, t T K ,W
Transformed heat equation [8.32] can be written:

1 w 2T K wT wT
W [8.87]
4 wK 2 2 wK wW
This equation is of a form analogous to [8.86] where the differentiation operators

are separated into two parts, the time derivative constitutes the perturbation term.
The calculation is performed as before and we seek a series development of
functions in separated variables:
f
T K , H W ¦ g i W f i K [8.88]
i 0
1 d2 K d
Letting M , equation [8.87] can be written:
4 dK 2 2 dK
wT
MT W
wW
We obtain the recurrence relations:
M f 0 K 0 g0 Tw W
M f1 K f 0 K g1 W W g 0' W
M f 2 K f1 K , g 2 W W g1' W [8.89]
!!!!
M f i K f i 1 K g i W W g i' 1 W
!!!!
which are associated with the boundary conditions:

f i f 0 i 0,1,2, ! , f ; f 0 0 1; f i 0 0 i 1,2, ! , f
Taking account of the conditions at infinity, the function f0(K) is (see section
5.4.5.4):
f 0 K 1 erfK
The analytical calculation of the functions fi is difficult: even if the solution to

the homogenous equation M f i K 0 can be expressed easily as a function of a
multiplicative constant (section 5.4.5.4), the variation of constants method is
impractical on account of the complexity of calculations. A numerical solution,
which is preferable, makes the easy calculation of the first functions fi possible. The
series of functions fi is alternated and tends asymptotically to two functions equal to
r f f K which can be easily determined from system [8.89]:
M fd fd with: fd 0
fd d
0
The reader can verify that we have: fd

c exp( 2 ) (c: constant)
However, the recurrence relation g i W W g i 1 W does not provide a simple

'
expression for the functions gi. Furthermore, we note that the perturbation term
W wT wW of equation [8.87], of order W Tw' t Tw t in relative value, must remain
quite small: the temporal derivative of the imposed temperature must decay quickly
enough as time increases. From a physical point of view, it seems natural that in
order to remain as a small perturbation, the imposed temperature variation decreases
with time as the temperature distribution of the base solution T0 spreads (Figure
8.6). In order to render the perturbation uniform in time, we compress the time scale
by performing the temporal variable change:

t ln W
Gi tˆ g i W ,
which simplifies recurrence relation [8.89] of the functions gi, which can be written:
G0 tˆ Tw tˆ G1 tˆ Tw' tˆ G 2' tˆ
G1' tˆ
Tw' ' tˆ
[8.90]
!!!!
Gi tˆ
Gi' 1 tˆ !
Tw(i ) tˆ !!!!
The parametric expression of the solutions for the thermal shock is thus:
f
T x, t ¦ Tw(i ) tˆ f i K [8.91]
i 0
The derivatives of the function Tw(t) being taken with respect to the variable
tˆ ln t .
The thermal flux density at the wall can be written:
wT O f
¦ Tw(i ) tˆ f i' 0
1
qTw O
wx x 0 at i 02
The values of the derivative f i' 0 are shown in Table 8.2.
i 0 1 2 3 4
f i' (0) / 2 –0.5642 –0.7821 0.3859 –0.3203 0.2985
i 5 6 7 8 9
f i' (0) / 2 –0.2897 0.2857 –0.2838 0.2829 –0.2826
Table 8.2. Values of the derivative f i' 0
The preceding method can also be applied to the expression of solutions in the
vicinity of the base solutions of the form t n f K . In particular, it is easy to verify
that the value n = 1/2 corresponds to the constant thermal flux which is given at the
wall. The solutions where the thermal flux varies gradually can be obtained as
before.
8.5.3.3. Harmonic solutions of the equations for continuous media

The method described in section 8.5.2.4 can be applied to continuous media by
eliminating the temporal variable of the partial differential equation of the problem.
Let us take the example of temperature oscillation applied to the surface of abscissa
x = 0 of a semi-infinite wall. We encounter this problem of oscillatory thermal
penetration into rocks or ground which is subjected to daily or annual temperature
variations. Assuming a homogenous medium, the temperature T x, t satisfies heat
equation [8.32] with the following boundary conditions:
wT w 2T
a ; T 0, t T0 4 cos Z t ; T f, t T0 [8.92]
wt wx 2
We seek solutions of the form:
T T0 4e jZ t f x
The heat equation becomes a differential equation:
jZ f af ' ' x
whose characteristic equation ( a r 2 jZ 0 ) has roots: r r1 j / 2a Z .

The real solutions sought for the temperature distribution can be written:
T T0 4 exp x G cosZ t x G
The depth G 2 a Z is a space constant for the thermal damping oscillation.

The solution represents a temperature oscillation which becomes damped with
increasing depth. A numerical application shows that this damping is fast: taking a
value of 10-6 m2.s-1 for the thermal diffusivity of the ground (corresponding to an
average rock), we find that G is respectively equal to 0.17 meters and to 3.15 meters
for daily and annual oscillations of temperature. The oscillation phase is opposite at
a depth of G /2 where amplitude has been reduced by a factor equal to 1.65.
The complex amplitudes method can be applied equally well to problems defined
in finite domains (walls, cylinders, etc.).
8.6. Model reduction
8.6.1. Overview
The objective of a knowledge model is to capture the evolution of a system, the

sub-system interactions of which we do not know. It comprises a large number of
variables in order to represent all the possible properties. The results involve either a
large quantity of numerical results from a computer solution or analytical
representations which may be more or less complex. The use of a knowledge model
amounts to performing a numerical experiment, which is often less expensive than a
physical experiment. A model is often too complex for a simple description of a
particular category of evolutions. Reduction of this model consists of replacing it
with a reduced model having a smaller number of variables, and whose objective is
to represent the principal phenomena with regard to the objective which is defined
(physical understanding of the model mechanisms, elaboration of a simulation
program or engineering formulae, control of processes). We are therefore interested
in decomposing the system into a small number of components with relatively
simple properties and whose interactions can provide a representative description of
the system in fixed conditions. Of course, the reduced model is not adapted to the
study of other operating conditions in the system.
Model reductions can be performed using various methods. However, the value
of the results of a model or of an experiment always depends on the pertinence of the
original hypotheses, on the reasoned use of numerical solutions or physical
measurement techniques and on a suitable physical analysis of the phenomena.
Certain methods consist of the performance of a numerical solution of a system by
means of a knowledge model, and the coupling of inputs and outputs with
representations of an empirical form which can be obtained by least mean square

methods; this practice has been used for a long time in order to establish engineering
formulae for certain applications. It has evolved into more sophisticated forms
through the use of more sophisticated mathematical methods. The use of these
methods is only justified as a method of exploiting the calculations resulting from a
physical analysis, which evaluates the nature of the approximations that are made
and the relative importance of the components in the functioning of a system. We
will prefer methods which increase our physical knowledge of the system studied,
and we will assume that the variables of the knowledge model have a physical
meaning and are not simply numerical values collected from experiments.
8.6.2. Model reduction of discrete systems

8.6.2.1. Principles of reduction of the state representation
A knowledge model, in the form of a state representation of a system with n state
variables or of an equivalent representation (an nth order differential equation in one
variable, etc.) makes the detailed description of all possible evolutions of the system.
The complexity associated with a large number n of variables leads to it being
replaced by a model with s state variables (s < n), called a reduced model, the
objective of which is to represent a particular evolution of the given physical system
(including its inputs and outputs) with good accuracy, or more general evolutions
with reduced precision. We have already seen some examples in section 8.3.2.2.3.
By definition, a differential system of order s cannot be equivalent to a

differential system of order n (n > s); however, these systems may have an identical
behavior for a family of solutions which depend on at most s parameters. The
reduction in the number of variables appearing in the differential equations can only
come from a regrouping of the equations, allowing the replacement of several
variables by a single variable or simplification of the equations, which lose their
differential character. Consider knowledge model [8.93]:
dX
A. X B.U Y D. X [8.93]
dt
Consider the reduced state vector Xr with s components (s < n) derived from X
by a passage matrix R from X to Xr (Xr = R X); the state vector Xr satisfies the
reduced state representation:
dX r
Ar . X r B r .U Y Dr X r [8.94]
dt
with conditions [8.95] obtained by multiplying the left-hand side of equation [8.93]
by R and by comparing with [8.94]:
Ar R RA; Br RB; X r 0 RX 0 Dr R D. [8.95]
If conditions [8.95] are exactly satisfied, representation [8.94] is an exact

reduced model for the ensemble of solutions X of [8.93] such that RX is non-zero.
However, in general, with the exception of cases where these solutions are known
explicitly, the realization of an exact reduced model is impractical.
Assuming conditions [8.95] are not satisfied, reduced representation [8.94] is an

approximate reduced model. The reduced vector Xr = RX of the solution X of [8.93],
does not exactly satisfy equation [8.94]; if q is the residuum, the error of equation
[8.94]:
dX r
Ar . X r B r .U q [8.96]
dt
Let X̂ r be a solution of [8.94]:
dXˆ r
Ar . Xˆ r B r .U [8.97]
dt
We define the error e X r Xˆ r with respect to the exact solution X̂ r assumed

to be verified by equation [8.94]; it satisfies the differential equation obtained by
subtracting term by term [8.96] and [8.97]:
de dX r
Ar e q Ar e Ar . X r B r .U
dt dt
We must therefore determine the matrices R, Ar and Br, such that the error of
equation q is minimized. This minimization procedure can be performed on an
ensemble of known solutions. We will leave aside the details of this kind of
procedure.
It is easy to impose that the reduction be exact for the steady regime. In this case,
equations [8.93] and [8.94] can be solved and we obtain:
Xr Ar1 Br .U RX RA 1 BU
and consequently the relation: Br Ar RA 1 B .
In the unsteady regime, the preceding elimination is no longer possible and the
vector RX obtained with the solution of [8.93] is no longer equal to the solution X̂ r
of [8.97]. The matrix Ar can be determined by minimizing the error of equation q.

This minimization can be performed for chosen inputs and for criteria which need to
be defined for the weighting to be applied at different instants [PET 91]. This
procedure comes down to performing a numerical interpolation on the knowledge
model by means of a reduced model.
8.6.2.2. Physical aspects of the model reduction

The considerations of the preceding section are essentially of a mathematical
nature and they leave a wide choice for the variables (in particular the matrix R), the
class of solutions and the objectives of error minimization. They can be applied to a
perfect knowledge model from the perspective of thermodynamics (sub-systems in
quasi-equilibrium), this needing to be verified for solutions which vary quite slowly
in space-time. The reduction in a knowledge model can only have a limited interest
if the discretization is too rough in certain domains. The partitioning of a system into
sub-systems is the most important operation and we will assume that it is suitable. In
general, the reduction in the number of variables is associated with a regrouping of
components, which leads to a choice of the reduced variables to be retained; this
should be done such that the definition of the mean intensive variables of the sub-
systems of the reduced model have a reasonable physical meaning in the balance
equations (section 1.4.2.5 and section 6.5.2.4).
The grouping of sub-systems which are in neighboring or identical states leads to

the replacement of many differential equations by a single equation: for example, a
material domain which has been segmented into three components and whose
temperatures are close can be represented by a single component at a suitable
temperature whose thermal energy is the sum of the energies of the components.
The sub-systems whose extensive quantity contents are small can often be
eliminated or assimilated into neighboring components, leading to the suppression
of the corresponding variables. The components whose extensive quantities are
constant or vary little can become sources of established fluxes: they thus play a role
of a simple resistance for the transfer of extensive quantities and their modeling
loses its differential character. We have seen two examples in sections 8.3.1.3 and
8.3.1.4.
In general, the number and the size of the sub-systems (or the grid of the system
domain) must be adapted to the categories of the solutions studied. Let us take the
example of the discretization on a continuous medium of a wall whose surfaces are
subjected to a thermal shock (section 8.3.2.2.2). Let us begin by segmenting the wall
into 50 equal elements in series (a system analogous to that of section 8.3.1.3).
Figure 8.13 shows that the temperature variations in the first instances are rapid in
the vicinity of the wall surface, whereas later, they become quite regular. The
discretization in equal elements is thus not the best solution: a finer discretization is
needed in the vicinity of the wall surfaces if we want to have a suitable precision in
the first instants, whereas such a fine discretization is not necessary later in the
central part of the wall. A better result with 50 elements is obtained when these are
distributed broadly at the center of the wall, and extremely close to the edges of the
wall.
8.6.2.3. The problem of local constraints

However, many systems present local constraints (in the mathematical sense) on
certain intensive variables related to the possible modification of material elements:
deteriorations due to excessive temperature or to large8 instantaneous stresses. In
this case, we obviously cannot eliminate from the model the small component which
contains, for example, a small thermal energy, but which possesses a critical
temperature value (temperature of a thermal measurement probe, a fragile element
whose temperature must be limited, etc.). The intensive variable concerned (often
the local temperature) is necessary for the regulation of a controller whose job it is
to modify or stop the system functioning. Model reduction in such zones is
obviously delicate.
8.6.2.4. Modal reduction of time-invariant linear systems

Modal reduction is essentially concerned with time-invariant linear systems
whose transitional solution is the sum of eigenmodes of the system of equations (see
the examples of discrete or continuous systems in section 8.3). The solution of a
linear system can be expressed in an explicit modal form using initial conditions and
an established solution (section 8.2.2.2); the reduction here amounts to a
simplification operation, but the question regarding the pertinence of the different
modes arises for discretized systems. We are then interested in writing a reduced
state representation which will allow the numerical simulation of the system in
complex conditions.
8.6.2.4.2. Modal reduction of continuous solution of continuous media

The modal solution of the heat equation involves replacing the continuous
variable by a denumerable set of mode coefficients in the explicit expression of
solution [8.38] in the form of a series. The reduction problem therefore consists of
simplifying and/or truncating this series. A consequence of this truncation is the
introduction of a discontinuity at the time origin, which amounts to assuming that
8 For example, cavitation (vaporization and chiefly sudden condensation) when local pressure
in a liquid flow is less than the saturation vapor pressure, high temperature on a wall in high
supersonic or hypersonic air flow, production of pollutants inside engines or chemical
reactors due to local bad conditions of temperature or concentrations, etc.
fast energy transfers in the higher order modes occur instantaneously (section
8.2.2.2). Various authors avoid this discontinuity by adding a fictitious mode or by
attributing to the last mode retained the total remaining amplitude. Let us illustrate
this problem by reconsidering the example of thermal shock on the wall edges
(section 8.3.2.2.2).
Tm(t)
T1
3 2
1 4 ~
T0 t
0 1 2
Figure 8.17. Mean temperature on a wall during a thermal shock on its two faces
(section 8.3.2.2.2). The draughts are indicative
Let us consider the mean temperature Tm (formula [8.41]), which translates the
thermal energy evolution over the course of the heating processes caused by the
thermal shock (curve 1, Figure 8.17). We have seen (Table 8.1) that the first mode
contains 81% of the process energy; in only conserving this first mode, we obtain a
representation for the temperature Tm (curve 2, Figure 8.17), which contains a
discontinuity at the origin equal to 0.19 T1 T0 ; considering the first two modes of
the series representing Tm (curve 3, Figure 8.17) this discontinuity is reduced to
0.10 T1 T0 ).
We can remove this discontinuity by including the total remaining amplitude

(0.19 in relative units) in the highest mode retained (curve 4, Figure 8.17) in
accordance with the formula:
S2 9S 2
Tm ( t ) T1 T0 T1 [ 0 .81e
t t
4
0 .19e 4 ]
However, the second mode is now too large; we can therefore look to determine
a time constant of the second mode in order to achieve a better representation of the
function Tm (t), for example by a mean square error minimization method. It is also
possible to add a fictitious mode. It is clear that a good solution is not achieved and
the values obtained by empirical adjustments are not physically pertinent.
In effect, in the first instants of the evolution, only zones in the vicinity of edges
are concerned with the heat transfer. Further from these, the sum of the series terms
[8.41] is zero: the modal representation is not adapted to the representation of a
thermal boundary layer problem (section 8.3.2.2.2). We can obviously note that the
thermal boundary layer is independent of the wall thickness A that can take any
value: defining modes by means of an arbitrary length is indeed an irrational method
which cannot lead to a judicious mathematical representation.
A good reduced model of thermal conduction in a wall must also take into
account the modal aspect as the evolution of the unsteady boundary layer. There is
no other (or nearly no other) way to obtain such a reduced model than a composite
representation matching the modal representation and the thermal shock solution:
we have presented this method in section 8.3.2.2.3, where a good precision was
obtained for the mean temperature [8.46], using only the first mode. This method
also has the advantage of giving precise values for the thermal flux density at the
edges [8.47] at any instant.
8.6.2.4.3. Modal reduction of discrete models

In section 8.6.2.2 we considered the model with 50 elements of a continuous
wall subjected to a thermal shock, leading to a linear system with 50 variables, and
which thus comprises 50 eigenvalues and eigenmodes. We will consider that a half
period of a sinusoid requires at least ten intervals in order to be represented by a
constant function in each element. The interval under study cannot therefore
comprise more than five arches: we can only represent the first three even modes
and the first two odd modes (see Figure 8.12). The 45 other modes are increasingly
noisy as the order is increased (the 50th mode corresponds to a change of sign of the
eigenfunction between each of the 50 intervals). Their physical existence is
increasingly problematic and it is not useful to consider them despite the fact that
they constitute exact solutions of the model.
A discretization into sub-systems should comprise a sufficiently large number of

elements, but only a few modes are actually useful. The modal solution is obviously
the most interesting because it provides a structured knowledge which highlights the
system properties. However, the discretization of a linear system proceeding from
the calculation of its modes requires more elements than a discretization, taking into
account physical aspects and particularly the level of unbalance between two
neighboring sub-systems: if we consider the preceding example of the wall on the
interval [-1,+1], it is necessary to calculate the modes to be retained over the entire
interval, whose form (Figure 8.12) requires discretization of the interval [-1,+1] in
equal segments, as opposed to a numerical resolution for which a discretization,
narrower near the wall faces and wider in the central part, is more fitted to the form
of the solution (Figure 8.13). This more rational method of discretization has the
advantage of reducing the number of variables.
Once a discretization is chosen for our knowledge model of a time-invariant

linear system, we have to deal with a lot of modes that have no physical
significance, as we have seen at the beginning of this section. The best reduction
method thus consists of using the properties of the modal solution, where only first
modes, corresponding to the actual evolution of the system, are retained. Returning
to the definition of an exact reduced state representation (section 8.6.2.1), let us
assume that the reduced state vector Xr = R X with s components (s < n), derived
from X by means of the reduction matrix R, satisfies reduced state representation
[8.98]:
dX r
Ar . X r Br .U [8.98]
dt
With condition [8.99] obtained by multiplying the left-hand side of equation

[8.93] by R and comparing with [8.98]:
Ar R RA; Br RB; X r 0 RX 0. [8.99]
we verify that an eigenmode of [8.94] is also an eigenmode of complete state

representation [8.93]:
A / i I 4 i 0 RA / i RI 4 i Ar / i I R4 i 0
The s eigenvalues /i of the reduced model are thus eigenvalues of the complete
model, the corresponding eigenvectors being equal to R4i9, the choice of which
modes to retain in the reduced model results from the definition of matrix R. The
reduction matrix R is thus such that the reduced state vector is constituted of the s
retained modes. Taking eigenvalues as a basis (modal basis), the matrix A is the
diagonal eigenvalue matrix /i. We assume that the s eigenvalues to be retained have
been placed in the first s elements of the diagonal. In the modal basis, the reduction
matrix R, of dimension s*n contains 1 on the diagonal, in correspondence with the
first diagonal element. The reader can verify that the reduced square matrix Ar
contains the s eigenvalues which were retained.
The same result can in principle be obtained in another manner. We have seen in
section 8.2.2.2 that the suppression of a mode amounts to requiring that its
coefficient be zero in the modal development at each instant: this results in an
instantaneous linear relationship between the state vector components (formula
9 Eigenvectors which have not been retained belong to the kernel of R.

[8.19]). The suppression of n-s modes of the model comes down to writing n-s
relations between the variables, which allows us to eliminate n-s state variables. We
thus obtain a representation by s state variables of the system which is reduced. This
method does not require us to calculate the modal basis. While it is in principal
arbitrary in the preceding methods, the choice of physical variables which are not
eliminated should nonetheless be performed so as to be conducive to physical
interpretation. The reader will also note that the principle of modal reduction itself
leads to a situation where discontinuous data leads to a discontinuity of the
variables of the reduced representation.
8.6.2.5. Reduction of input-output representations

Simple input-output representations are useful:
– for obtaining a global knowledge of the behavior of a system in view of the
implementation of control;
– for the dimensioning of components during the realization of a system;
– for the representation of sub-systems in models of complex systems.
The properties of these reduced representations differ according to their

utilization: command or engineering formulae.
8.6.2.5.2. Command models

The essential characteristics to know are thus relatively global and it is sufficient
to know one or several “response times” corresponding to the inputs on which we
act. The output variables allow us to define values of actions to be effected. We will
here consider a model of the form [8.59] of low order: for example, for a single
input single output system, we often consider a second order differential equation
which is satisfied by the output variable and whose input is the right-hand side. We
will here only consider an open-loop system (without any retroaction by the
command); the system operation in close-loop depends on the command structure
and is not the object of this work (see [OBI 00]).
Second order differential equations represent accumulation (of mass or heat for
example), damping and stiffness mechanisms, which are themselves represented,
respectively, by second order terms, first order terms and zero order terms (the
function). It is thus possible to represent aperiodic or stable oscillatory systems. The
coefficients of such differential equations can be determined by imposing an
impulse excitation as an input, and then by “best” identifying the response of the
system to an expression of the form ae D t cosZ t M . This is equivalent to
performing a spectral study of the system (section 8.5.2.4).
8.6.2.5.3. Engineering formulae

We will designate under this category the input-output relations derived from
knowledge models and which are simple enough to be immediately useful for
example for quick dimensional assessment of components of a system in view of a
realization or for the representation of a sub-system in a more complex system.
These formulae are the result of reduced state representations whose pertinence has
been previously verified. They can also be obtained by simplification of an
analytical solution when one exists, for example by truncating a series (example of
section 8.6.2.4.2) or when using composite representations (section 8.3.2.2.3).
8.6.2.5.4. Established regimes

The parametric expression of established regimes, obtained in section 8.5.2 in
the form of a series, can be simplified by truncation of this. Let us recall that the
representation in the form of a series is only of interest if it is possible to limit this to
a small number of terms. If this is not the case, this indicates that thermal boundary
layers exist in the domain, and it is therefore preferable to find a direct expression
which represents the boundary layer and to describe a composite solution by
matched asymptotic expansion (see section 8.3.2.2.3).
8.7. Application in fluid mechanics and transfer in flows
The evolution equations of a discrete or continuous system as a time function are

parabolic (or irreversible). In the presence of flow, the evolution variable is the time
only when Lagrangian variables are used. In Eulerian variables, the evolution speed
of a quantity g is no longer represented by its temporal derivative wg wt but by the
material derivative dg dt . This representation does not change the parabolic
character of the balance equations for extensive quantities with Euler variables
expressed along curvilinear abscissa of trajectories (or characteristic curves), which
is a parabolic variable equivalent to the time with Lagrange variables (see
interpretation of section 5.2.1) upstream then becoming the equivalent of the past. In
steady flow, the time variable disappears and the evolution variable becomes the
coordinate of the particle trajectories: systems studied thus appear as dynamic
systems along the trajectories. The same is true of boundary layer equations, or more
generally of the evolution of fluid properties along trajectories. These preceding
ideas are thus applicable to problems encountered in flows.
The methods evoked in this chapter are encountered when dealing with the
solution of flow and transfer problems in boundary layers ([SCH 99], [YIH 77]).
However, these are generally non-linear and computations cannot be effected in as
complete a manner as in this chapter. Writing balance equations, ultimately with
approximations which may be more or less global, leads to state representations
where time is replaced by a boundary layer coordinate. The inputs are therefore
often evolution laws for the velocity of a perfect fluid as a function of the spatial
coordinate, the initial conditions having been provided in the initial section of a
boundary layer or a pipe. It is thus possible to obtain approximate external
representations in forms which are more or less explicit.
In principle, the procedure is of the same nature as for linear problems: an

elementary analytical solution makes it possible to study situations which are more
or less similar, and to identify the parameters associated with the category of
solutions studied. However, non-linearity does not enable the easy use of series
developments of eigenfunctions or of integral transformations. On the other hand,
singular or regular perturbation methods remain useful for writing the equations as a
cascade of successive approximations (section 6.4.2.3). The differential equations
obtained must often be solved, either by numerical means or by some global
methods (section 6.3.1.2). In other words, the general methodology of the present
chapter is applicable to problems of flow and associated transfer by means of an
adaptation of the calculation methods. The preceding discussions of linear dynamic
systems often allow us to organize knowledge gained from these non-linear systems
in a more rational way.
Appendix 1
Laplace Transform
A1.1. Definition
The Laplace transform of a real variable function f(t), considered for t t 0

( f (t ) 0 for t < 0), is defined by the relation
f
L f p ³ f t e
pt
dt
0
where p is a complex valued variable. It is particularly well adapted to causal

signals.
A1.2. Properties
– The Laplace transform is a linear application on integrable functions.

The Laplace transform of the derivative of a function is written:
L f ' p ³0 f ' t e dt p ³0 f t e pt dt f 0 pL f p f 0
f pt f
The derivative is here taken in the sense of distributions: any discontinuity of the
function leads to a Dirac distribution at that point for the derivative.
– The Laplace transform of a primitive is written:

1
g t ³0 f u du
t
L g ( p) L f ( p)
p
– The Laplace transform of a delayed temporal signal can be written:
if: g (t ) f t W t t W, g (t ) 0 tW
px W
Lg p ³0 g t e dt ³W f t e dt ³0 f x e
f f f
pt pt
dx L f ( p)e pW
Conversely, if: g (t ) f (t )e a t
p a t
Lg p ³0 f t e e dt ³0 f t e
f a t pt f
dt L f ( p a)
– The Laplace transform of a convolution product of two functions f and g is

equal to the product of the Laplace transforms of these functions:
L f * g ( p) L f ( p) L g ( p)
A1.3. Some Laplace transforms
– The Laplace transform of a Dirac function is equal to 1:
LG p ³0 G t e dt ³ f G t e
f pt f pt
dt 1
G being zero for negative values of t.

– The unit step H(t), equal to 1 for t positive, is the primitive of the Dirac
f
distribution:
³ Gt dt ; its transform is: L
0
H p 1 p .
– The unit ramp t is the primitive ³0 H u du of the unit step H(f); its transform
t
is: Lt ( p ) 1 p2 .
In general, we obtain: Lt n p 1 p n 1 .
– f 0 lim f t lim pL f p
t o0 p of
Appendices 479
– f f lim f t lim pL f p
t of po 0
– Consider the limited development Pm(t) to order m for small values of time t.
We obtain:
m m a
Pm t ¦ a i t i L Pm p ¦ i
i 1
i 0 i 0p
The lowest order terms in p-1 represent the Laplace transform which corresponds
to small values of time t.
For a limited development, and for large values of t, (t o f) we obtain:

n n
ai
Qm t ¦ ai t i LQm p ¦
i nm i nm p i 1
The highest order terms in p-1 represent the Laplace transform which corresponds
to large values of t.
– The Laplace transforms of exponential and harmonic functions are:
1 1
Lexpat p Lexp jZt p
f ( a p )t
³0 e dt ;
pa p jZ
p Z
LcosZt p ; Lsin Zt p
2 2
p Z p Z2
2
A1.4. Application to the solution of constant coefficient differential equations
dX
Consider the system: A. X BU X 0 X0 .
dt
The Laplace transform LX(p) of vector X(t) satisfies the equation:
pI AL X p L BU p X 0
From this we deduce:
L X ( p) pI A-1 >L BU ( p) X 0 @
The matrix pI A is comprised of elements which are rational fractions of the

-1
frequency p which can be developed in a complete series in p. We can write this in

the form:
f
pI A- 1 ¦ p i Di
i 0
The transformed linear system can thus be written:
f f
L X ( p) ¦ p i Di L BU ( p) pI A-1 X 0 ¦ Di L BU i ( p) pI A-1 X 0
i 0 i 0
The second term represents the transient response to the initial conditions. The
steady response is thus:
f
X ¦ BDiU (i ) t
i 0
Appendix 2
Hilbert Transform
The amplitude a and the frequency Q of an oscillatory harmonic motion appear

naturally if the signal is written as the real part of a complex number in the
trigonometric form ae jMt , in which the time derivative of the instantaneous phase
M t 2S Q t I is equal to 2SQ. In the same way, we can associate a complex-
valued function D x t (called an analytical signal) with any real signal x(t) by
adding to it an imaginary-valued function of time jy(t). This complex signal can be
written in the trigonometric form:
D x (t ) x(t ) jy (t ) at .e jM (t ) with : x(t ) Re[a(t )e jM (t ) ]
The modulus at and the argument Mt of this complex number can be defined
as the amplitude and the instantaneous phase of the signal x(t); we thus define the
instantaneous frequency as the derivative M ' t 2S .
These definitions are only meaningful if the point whose affix is the complex
function turns nearly regularly in the positive direction around the origin, in a way
quite similar to a complex exponential function. This is the case for amplitude- or
phase-modulated harmonic signals, or for the sounds of musical instruments. A
narrow-band signal x(t) appears on an oscilloscope as a carrier of the neighbouring
frequency Qo, whose amplitude [a(t)] varies slowly and whose phase [ M (t)] is
fluctuating.
Given the real part of a complex number, it cannot be determined uniquely,

because we only know the product a cos M t . As the signals to be represented are
quite close to harmonic signals, it is natural to apply the Fourier transform and to
generalize the properties of harmonic signals. However, we note that the definition
of an analytic signal e j 2SQ t for a sinusoidal function cos 2S Q t consists of
eliminating the negative frequencies of the Fourier transform of this analytical
signal, by folding over the negative frequencies onto the positive frequency axis.
This operation can be applied to any signal x(t) and it allows the analytical signal
Dx(t) to be defined. Let Fx Q be the Fourier transform of the signal x(t). The
Fourier transform of the analytical signal D x t associated with x(t) is defined by:
FDx Q { 0 for Q 0
®
¯ FDx Q { 2 Fx Q for Q ! 0
¦£x 0 Sign(x) 1
i.e. Fax (v ) Fx (v ) Sign(v ).Fx (v ) with: ¦¤ [A2.1]
¦¦¥x 0 Sign(x) 1
As the inverse Fourier transform F 1 SignQ . of the sign function is equal to

j S t ([BEE 03], [PRI 91]), the analytical signal D x t is thus obtained by
performing the inverse Fourier transform of expression [A2.1]
D x t F 1 >FDx Q @ xt jH x t
where Hx(t) designates the Hilbert transform of the signal x(t):
1 1 du
H t jF 1 SignQ .F x Q * xt VP ³ f xu .
f
St S t u
The Hilbert transform is limited to positive time, contrary to the Fourier

transform. The Hilbert “filter” is a linear filter whose impulse response is 1/St; it is
thus not causal. The transfer function of this filter is
H H Q F1 S t Q j.signQ . The Hilbert filter transforms a cosine to a sine; it
is a perfect quadrature filter. It is unfortunately not achievable as it is non-causal;
we can only create quadratures which function in a limited band of frequencies.
Appendix 3
Cepstral Analysis
A3.1. Introduction
The cepstrum is an integral transform that can be calculated from a spectrum and
contributes to its analysis. For example, in a vibration problem, it makes it possible
to separate the impulse response from the excitation forces. The analysis of vibration
signals from rotating machinery is important. The different operating conditions of a
machine, as well as defects, can be observed in the spectral domain ([STR 96]).
It is also used in speech processing, the vocal signal coming from the
convolution of the excitation (source) and the impulse response of the vocal passage.
Another application is the extraction of incident sound from a signal containing

both the incident sound and its reflection (echo suppression). We will quickly cover
this application, which enables the demonstration of the use of logarithms and thus
of the cepstrum.
A3.2. Definitions
Let x(t) be a time signal and Fx(Q) its Fourier transform; by definition the
complex cepstrum Cx(t) is the inverse Fourier transform, denoted F-1, of the
logarithm of function Fx (Q ) ([BOU 98], [NOR 03 ], [STR 96])
C x (W ) F 1 (ln^Fx (Q )`)
if Fx (Q ) F x (Q ) e jM (Q ) , then: ln^Fx (Q )` ln^Fx (Q ) ` j M (Q )

The power cepstrum may be defined in several ways, for example, as the square
of the modulus of the complex cepstrum of the signal:
2
C px F 1 (ln^Fx (Q )`) .
or by taking the inverse Fourier transform of logarithm of the modulus square:
C px
©
^
F 1 §¨ ln Fx (Q )
2
`·¸¹
A3.3. Example of echo suppression
Let x(t) be a sound signal comprising the superposition of an original sound s(t)
and an unwanted echo sr(t):
x(t ) s (t ) s r (t )
The reflected sound is attenuated and dephased compared with the original signal
s r (t ) a s (t t 0 )
The Fourier transform of x(t) can be written:
Fx (Q )
Fs (Q ) 1 a e 2S j Q t0

and the square of the modulus of this transform is:
F x (Q )
2 2

Fs (Q ) 1 a 2 2a cos 2S Q t 0
If we take the logarithm of this quantity, the echo phenomenon is seen in
frequency space by the addition of a periodic term, of period 1/t0:
^
ln Fx (Q )
2
` ln^F (Q ) ` ln1 a
s
2 2
2a cos 2S Q t 0
We now take the inverse Fourier transform of this expression ([ALL 04], ([MAD
98], [NOR 03]): the first two terms will be, respectively, the power cepstrums x(t)
and x(t); the third term will be the inverse Fourier transform of the periodic function
shown above, which will be comprised of Dirac functions of different amplitude,
separated by t0
Appendices 485
C px (W ) ^
C ps (W ) F 1 ln 1 a 2 2a cos 2S Q t 0 `
We thus find ourselves in a pseudo-temporal space, whose variable is known as
“quefrency”, in which we perform a “liftering” of the Dirac signals (the terms
“cepstrum”, “quefrency” and “liftering” are respectively anagrams of the words
spectrum, frequency and filtering). The echo signal is thus eliminated by the
suppression of the Dirac signals in this space. The initial signal is thus reconstructed
without its echo by an inverse process ([NOR 03]); however, we note that phase
information has been lost with this procedure of cepstrum power.
A3.4. General case
We consider a source signal x(t) going through a passive linear system whose
impulse response is h(t). The output signal y(t) is the convolution product of x by h:
y (t ) h(t ) * x(t )
The convolution product becomes a simple product after Fourier transform:
F y Q H Q Fx Q
The cepstrum transforms the product to a sum:
C y (W ) C h (W ) C x (W )
In order to perform the deconvolution, in other words the separation of the

source from that of the medium, some assumptions must be made regarding the class
of functions which comprise either the source or the medium:
– in the case of speech, we assume that the vocal excitation is comprised of a
periodic impulse t0 (this property is of course only applied to the stable parts of
sounds) and that the cepstral contribution of the vocal passage is found in the low
quefrencies; we can thus obtain a smoothed spectrum which only contains
information on the vocal passage. The inverse Fourier transform of the spectrum
gives the estimated impulse response of the passage and is used for voice synthesis
([JUR 08], [MAD 98]).
– in mechanics, free oscillatory periods of a medium are generally small
compared to the periods of the excitations: the cepstrum of the impulse response of
the structure is in the left part, while that of the excitation is in the right part of the
cepstrum graph of structure response to excitation ([DES 00], [NOR 03], [ALL
04]).
Appendix 4
Eigenfunctions of an Operator
A4.1. Eigenfunctions of an operator
Consider a linear operator L applied to a function f x which satisfies

homogenous boundary conditions at the extremities of the interval defined by [a,b].
The eigenvalues Oi of this operator and the associated eigenfunctions fi satisfy the
relation:
L fi O fi
A4.2. Self-adjoint operator
A4.2.1. Eigenfunctions
Let us define the scalar product f g ! , for the foregoing functions f, by the
relation:
³a f x .g x dx
b
f g !
The operator L is the self-adjoint if it satisfies the relation
³a Lf x .g x dx ³a f x .Lg x dx
b b
f Lg ! L f g !
Under some general conditions:

the eigenvalues of the self-adjoint operator L are positive and form an infinite
denumerable set Oi i 1, 2,..., n,... . With each eigenvalue Oi we can associate an
eigenfunction fi;
the eigenfunctions fi form a basis for the class of functions f.
A4.2.2. Expression of a function f using an eigenfunction basis-set
Consider the following series development of f on the eigenfunction-basis:
f
f x ¦ c j f j x
j 1
Let us calculate the scalar product f f i ! :
f
f x f i x ! ¦ ci f j x f i x ! [A4.1]
j 1
Now, the eigenfunctions are orthogonal for the scalar product. In effect, with the
assumption of a self-adjoint operator, we obtain for O i z O j :
f i Lf j ! L f i f j ! O j O i fi f j ! 0
We also obtain:
2
f i Lf i ! O i f i f i ! O i f i
Substituting previous results into [A4.1], we obtain the value of the coefficient
ci:
f
f x f i x ! ¦ c i f j x f i x ! c i f i x
2
j 1
f x
fi x

or:
ci 2
fi x
Bibliography
[ABR 65] ABRAMOWITZ M.S., and STEGUN I.A., Handbook of Mathematical Functions,
Dover, New York, 1965.
[ALL 04] ALLEN R.L., MILLS D., Signal Analysis: Time, Frequency, Scale and Structure,
Inc NetLibrary, Wiley-IEEE, 2004.
[AME 58] AMES J.S., MURNAGHAN F.D., Theoretical Mechanics, An Introduction to
Theoretical Physics, Dover, 1958.
[AND 99] ANDREWS L. C., SHIVAMOGGI B. K., Integral Transforms for Engineers,
Society of Photo-optical Instrumentation Engineers, SPIE Press, 1999.
[ASH 05] American Society of Heating, Refrigerating and Air-Conditioning Engineers,
ASHRAE Handbook of Fundamentals, 2005.
[BAH 01] BAHER H., Analog & Digital Signal Processing, John Wiley & Sons, 2001.
[BAL 77] BALIAN R., PEUBE J.-L., Fluid Dynamics, Theoretical physics summer school
course, Les Houches 1973, Gordon and Breach Science Publishers, 1977.
[BAR 01] BARRASS B., YOUNG P., BARRASS C.V., Ship Stability, Notes and Examples,
Butterworth-Heinemann, 3rd edition, 2001.
[BAS 98] BASS H.E., SHIELDS F.D., “Ultrasonic relaxation processes” in Handbook of
Acoustics, Malcolm J. Crocker (ed.), John Wiley & Sons, Inc., 1998.
[BEE 03] BEERENDS R. J.,TER MORSCHE H. G.,VAN DEN BERG J. C., VAN DE VRIE
E. M,. Fourier and Laplace Transforms, Cambridge University Press, 2003.
[BEL 00] BELLANGER M., Digital Processing of Signals: Theory and Practice, Wiley,
2000.
[BEN 74] BENNETT C.O., MYERS J.E, Momemtum, Heat and mass Transfer, McGraw-
Hill, 2nd edition, 1974.
[BEN 99] BENDER C.M., ORSZAG S.A., Advanced Mathematical Methods for Scientists
and Engineers, Springer-Verlag, New York, 1999.
[BEN 06] BENESON W., HARRIS J.W. STOCKER H., LUTZ H., Handbook of Physics,
Springer, 2006.
[BER 84] BERGE P., POMEAU Y., VIDAL C., Order Within Chaos; Towards a
Deterministic Approach to Turbulence, John Wiley & Sons, 1984.
[BER 90] BERANEK L.L., Acoustics, American Institute of Physics, 1990.
[BIR 01] BIRD R.B., STEWART W.E., LIGHTFOOT E.N., Transport Phenomena, John
Wiley & Sons, 2nd edition, 2001.
[BLA 98] BLATTER C., Wavelets: A Primer, A K Peters Ltd, 1998.
[BOC 92] BOCCARA N., Les principes de la thermodynamique classique, Presses
Universitaires de France, 1992.
[BOR 00] BORGHI R., CHAMPION M., Modélisation et théorie des flammes, Editions
Technip, 2000.
[BOU 98] BOULENGER A., PACHAUD C., Diagnostic vibratoire et maintenance
préventive, Dunod, 1998.
[BRE 91] BREZINSKI C. and M. REDIVO-ZAGLIA M., Extrapolation Methods. Theory
and Practice, North-Holland, 1991.
[BRU 95] BRUUN H. H. Hot-wire Anemometry Principles and Signal Analysis, 1995,
Oxford University Press.
[BYU 02] BYUNG CHAN EU, Generalized Thermodynamics: Thermodynamics of
Irreversible Processes and Generalized Hydrodynamics, Springer, 1st edition, 2002.
[CAS 06] CASTANIÉ F., Spectral Analysis, ISTE Ltd, 2006.
[CHA 91] CHAPMAN S, COWLING T. G., The Mathematical Theory of Non-uniform
Gases: An Account of the Kinetic Theory of viscosity, Thermal Conduction and Diffusion in
Gases, Cambridge University Press, 3rd edition, 1991
[CAN 01] CANDEL S., Mécanique des fluides: Cours, Dunod, 2001.
[CAR 59] CARSLAW H.S., JAEGER J.C., Conduction of Heat in Solids, Clarendon Press,
(2nd edition), 1959.
[CHA 70] CHANG P. K., Separation of Flow, Pergamon Press, 1970.
[COI 97] COIRIER J., Mécanique des milieux continus, Dunod, 1997.
[COL 68] COLE J.D., Perturbation Methods in Applied Mathematics, Blaisdell Pub. Co.,
1968.
[COM 02] COMOLET R., Mécanique expérimentale des fluides, volume I, Dunod, 2002.
[COM 94] COMOLET R., Mécanique expérimentale des fluides, volume II, Masson, 1994
[COU 89] COURANT R., HILBERT D., Methods of Mathematical Physics, vol. II, Partial
Differential Equations, Wiley Interscience, 1989.
[COU 89] COUSTEIX J., Aérodynamique, Turbulence et couche limite, Cepadues-éditions,
1989.
Bibliography 491
[CRO 05] CROWS C.T., ELGER D.F., ROBERSON J.A., Engineering Fluid Mechanics,
Wiley, 2005.
[DEB 06] DEBNATH L., BHATTA D., Integral Transforms and their Applications, CRC
Press, 2006.
[DEG 62] DE GROOT S.R., MAZUR P., Non-equilibrium Thermodynamics, North-Holland,
Amsterdam, 1962: Dover Editions, 1984.
[DEM 97] DEMMEL J. W., Applied Numerical Linear Algebra, SIAM, 1997.
[DES 00] DE SILVA W., DE CLARENCE W., Vibration Fundamentals and Practice, CRC
Press, 2000.
[DOU 95] DOUGHERTY T., Systems & Control, An Introduction to Linear, Sampled &
Non-linear Systems, World Scientific, 1995.
[DUR 81] DURST F., MELLING A., WHITELAW J.H., Principles and Practice of laser-
Doppler Anemometry, Academic Press, 1981 (2nd edition)
[EU 92] EU B. C., Kinetic Theory and Irreversible Thermodynamics, John Wiley, 1992.
[EVA 98] EVANS L. C., Partial Differential Equations, American Mathematical Society,
Providence, 1998.
[EVE 89] EVETT J.B., LIU C., 2,500 Solved Problems in Fluid Mechanics and Hydraulics,
McGraw-Hill, 1989
[PRE 64] FREDRICKSON A.O., Principles and Applications of Rheology, Prenctice Hall,
1964.
[GAB 98] GABARD S., PEUBE J., “The influence of a narrow tube on the measurement of
pressure fluctuations”, ACTA ACUSTICA, 1998, vol. 84(6), p.1151.
[GAR 06] GARDE R J., Fluid Mechanics Through Problems, New Age Publishers, 2006.
[GER 94] GERARD D,. KUIKEN C., Thermodynamics of Irreversible Processes:
Applications to Diffusion and Rheology, John Wiley, 1994
[GIL 64] GILES R., Mathematical Foundations of Thermodynamics, Pergamon, Oxford,
1964.
[GRI 71] GRIMSON J., Advanced Fluid Dynamics and Heat Transfer, McGraw-Hill, 1971.
[GUP 83] GUPTA S., HASDORFF L., Fundamentals of Automatic Control, Krieger Pub Co,
1983.
[GUY 01] GUYON E., HULIN J.P., PETIT L., MITESCU C.D., Physical Hydrodynamics,
Oxford University Press, 2001.
[HAN 05] HANSEN C. D., JOHNSON C. R., The Visualization Handbook, Academic Press,
2005.
[HAR 98] HARRIS J., STOCKER H., HARRIS J.W., Handbook of Mathematics and
Computational Science, Springer, 1998.
[HIG 93] HIGGINS G.R., Sampling Theory in Fourier and Signal Analysis Foundations,
Oxford University Press, 1996.
[HIN 91] HINCH E.J., Perturbation Methods, Cambridge, 1991.

[HIR 64] HIRSCHFELDER J.O., CURTISS C.F., BIRD R.B., The Molecular Theory of
Gases and Liquids, Wiley-Interscience 1964.
[IDE 99] IDEL'CIK I.E., Coefficients de pertes de charge singulières et pertes de charge par
frottement, Eyrolles, 1999.
[IGO 89] IGOREVICH A.V.. Mathematical Methods of Classical Mechanics, Springer, 2nd
edition, 1989.
[JAC 91] JACKSON L. B., Signals, Systems and Transforms, Addison-Wesley, 1991
[JAI 01] JAIN S.C., Open-Channel Flow, John Wiley & Sons, 2001.
[JOH 98] JOHNSON R. W., The Handbook of Fluid Dynamics, CRC Press, 1998.
[JER 92] JERRI A.J., Integral and Discrete Transforms with Applications and Error
Analysis, CRC Press, 1992.
[JOU 01] JOU D., CASAS-VASQUEZ J., LEBON G., Extended Irreversible
Thermodynamics, Springer, 3rd edition, 2002.
[JUR 08] JURAFSKY D., MARTIN J.H., Speech And Language Processing: An
Introduction to Natural Language, Pearson/Prentice Hall, 2008.
[KIN 82] KINSLER L.E., FREY A.R., COPPENS A.B and SANDERS J.V., Fundamentals of
Acoustics, John Wiley & Sons, 3rd edition, 1982.
[KIR 67] KIRKWOOD J.G., Shock and Detonation Waves, Gordon and Breach, 1967.
[KRE 91] KREYSZIG E., Differential Geometry, Dover Publications, 1991.
[KUO 02] KUO B, GOLNARAGHI F., Automatic Control Systems, John Wiley & Sons,
2003.
[LAN 89] LANDAU L., LIFCHITZ E., Mécanique des fluides, Editions MIR, 2nd edition,
1989.
[LAT 67] LATTES, R., Quelques méthodes de résolution de problèmes aux limites de la
physique mathématique, Gordon and Breach, Dunod, 1967.
[LES 98] LESIEUR M., HUNT J., Turbulence et déterminisme, Presses universitaires de
Grenoble, 1998.
[LEV 62] LEVICH V.G., Physicochemical Hydrodynamics, Prenctice Hall, 1962.
[LEW 01] LEWY S., Acoustique industrielle et aéroacoustique, Hermes, 2001.
[LIC 02] LICATA T., RISSET J.C., Electroacoustic Music: Analytical Perspectives,
Greenwood Publishing Group, 2002.
[LIP 05] LIPTAK B.G., Instrument Engineers’ Handbook, Inc NetLibrary, CRC Press, 2005.
[MAD 98] MADISETTI D.V., WILLIAMS D.B., The Digital Signal Processing Handbook,
CRC Press, 1998.
[MAL 94] MALDAQUE, MALDAGUE X., Infrared Methodology and Technology, CRC
Press, 1994.
Bibliography 493
[MAR 99] MARTINET J., Thermocinétique approfondie, Tec et Doc, Paris, 1999
[MAT 64] MATHEWS J.,WALKER R.L., Mathematical Methods of Physics, W.A.
Benjamin, Inc., 1964.
[MAT 00] MATHIEU J. and SCOTT J., An Introduction to Turbulent Flow, Cambridge
University Press, 2000.
[MCG 88] MCGEE T, Principles and Methods of Temperature Measurement, Wiley-IEEE,
1988.
[MEA 91] MEADE M. L., DILLON C. R., Signals and Systems: Models and Behaviour,
Springer, 1991.
[MER 87] MERZKIRCH W., Flow Visualization, Academic Press, 1987.
[MIL 61] MILES E.C.R., Supersonics Aerodynamics, Dover Publications, 1961 (reissue of
the first edition published in 1950).
[MIL 66] MILNE-THOMSON L.M., Theoretical Aerodynamics, MacMillan, 4th edition,
1966.
[MIS 07] MISITI M., MISITI Y., OPPENHEIM G. and POGGI J.M. (eds.), Wavelets and
their Applications, ISTE Ltd, 2007.
[NAY 81] NAYFEH A.H., Introduction to Perturbation Techniques, John Wiley & Sons,
New York, 1981.
[NOR 03] NORTON M.P., KARCZUB D.G., Fundamentals of Noise and Vibration Analysis
for Engineers, Cambridge University Press, 2003.
[OBI 00] OBINATA G., ANDERSON B.D.O., Model Reduction for Control System Design,
Springer, 2000.
[OPP 06] OPPENHEIM A.K., Dynamics of Combustion Systems, Springer Verlag, 2006.
[PAR 98] PARASCHIVOIU I., Aérodynamique subsonique, Editions de l'Ecole
Polytechnique de Montréal, 1998.
[PEN 55] PENNER S.S., Introduction to the Study of Chemical Reactions in Flow Systems,
Buttersworths Scientific Publications, 1955.
[PET 91] PETIT D., Réduction des modèles de connaissance et identification de modèles
d'ordre réduit, Thesis, Aix- Marseille University, 1991
[PEU 62] PEUBE J.L., Sur certains régimes variables de l'écoulement laminaire d'un fluide
incompressible entre deux plans parallèles, CRAS Paris, pp 2358-2360, 1962.
[PEU 79] PEUBE J.L., “Experimental study of velocities field at the exit of a resonant pipe
excited at strong amplitudes”, 8th International Symposium on Nonlinear Acoustics, Journal
de Physique,. Colloques 40 (1979) C8 p. 346-352.
[PEU 84] PEUBE J.L., Une technique de modélisation thermique des régimes instationnaires,
Modélisation et simulation en thermique, Poitiers, 1984.
[PEU 89] PEUBE J., PEUBE J. L., VIGNER A. and BARRE C., “Experimental study of
three dimensional flow in admission collector pipe of a thermal engine”, Entropie, 1989,
148.
[PEU 91] PEUBE Janine, “Realisation and characterisation of a reproducible vortex for study
of unsteady flows”, La Recherche Aérospatiale, 1991, 4, p. 23-26.
[PEU 92] PEUBE J, GARRAUD E., “Measurements of acoustic velocity in water by laser-
Doppler anemometry”, European Conference on Underwater Acoustics, Luxembourg, Ed. M
Weydert, Elsevier Applied Science, 1992, p. 161-164.
[PET 55] PETROWSKI I.G., Vorlesungen über partielle Differentialgleichungen,
B.G.Teubner Verlagsgesellschaft, Leipzig, 1955.
[POL 04] POLYANIN A.D., ZAITSEV V.F., Handbook of Non-linear Partial Differential
Equations, Chapman Hall/CRC Press, Boca Raton, 2004.
[PRE 07] PRESS W. H., Numerical Recipes: The Art of Scientific Computing, Cambridge
University Press, 2007.
[PRI 68] PRIGOGINE I., Introduction à la thermodynamique des processus irréversibles,
Monographies Dunod, 1968.
[PRI 68] PRIGOGINE I., Introduction to Thermodynamics of Irreversible Processes, Wiley-
Interscience, 3rd edition, 1968.
[PRI 91] PRIEMER R., Introductory Signal Processing, Advances Series in Electrical and
Computer Engineering, vol. 6, World Scientific, 1991.
[RAF 98] RAFFEL M. R., WILLERT C, KOMPENHANS J., WERELEY J., Particle Image
Velocimetry, A Practical Guide, Springer, 1998.
[ROT 95] ROTHSTEIN J., MIDI: A Comprehensive Introduction, A-R Editions, Inc., 1995.
[SAD 93] SADAT H., PEUBE J-L., “Une méthode de résolution de l'équation de la chaleur
en régime graduellement varié”, International Journal of Heat and Mass Transfer, vol. 36,
Issue 18, pp. 4455-4460, December 1993.
[SAG 66] SAGAN H., Boundary and Eigenvalue Problems in Mathematical Physics, John
Wiley & Sons, 3rd edition, 1966.
[SAN 97] SANAKNAKI H., Contribution à la modélisation des écoulements pulsés à
l'admission d'un moteur thermique à explosions, Thesis, University of Poitiers, 1997.
[SAY 00] SAYOOD K., Introduction to Data Compression, Morgan Kaufmann, 2nd edition,
2000.
[SCH 99] SCHLICHTING H.,GERSTEN K., Boundary Layer Theory, Springer, 8th edition,
1999.
[SPU 97] SPURK J.H., Fluid Mechanics: Problems and Solutions, 1997, Springer.
[STA 00] STANISLAS M., KOMPENHANS J., WESTERWEEL J., Particle Image
Velocimetry: Progress Towards Industrial Application, Kluwer Academic Publishers, 2000.
Bibliography 495
[STA 04] STANISLAS M. WESTERWEEL M. S. J, KOMPENHANS J., Particle Image

Velocimetry: Recent Improvements, Proceedings of the EUROPIV 2 Workshop, Springer,
2004.
[STR 75] STREETER V.L., WYLIE E. B., Fluid Mechanics, McGraw-Hill, 6th edition,
1975.
[STR 96] STRANG G., NGUYEN T, Wavelets and Filter Banks, Wellesley Cambridge Press,
SIAM, 1996.
[TAN 00] TANNER R.I., Engineering Rheology, Oxford University Press, 2000.
[TEN 72] TENNEKES H. and LUMLEY J.L., A First Course in Turbulence, MIT Press,
1972.
[THW 60] THWAITES B., Incompressible Aerodynamics, Clarendon Press, 1960
[VAN 75] VAN DYKE M., Perturbation Methods in Fluid Mechanics, Parabolic Press, 1975.
[VER 97] VERHAS J., VERHAS J., VERHS J., Thermodynamics and Rheology, Kluwer
Academic Publishers,1997.
[WAI 90] WAIBEL A., LEE K-F., Readings in Speech Recognition, Morgan Kaufmann,
1990.
[WAN 06] WANG L., GAO R. X., Condition Monitoring and Control for Intelligent
Manufacturing, Springer, 2006.
[WEB 98] WEBSTER A.G., Partial Differential Equations of Mathematical Physics, Dover
Publications, 2nd edition, 1998.
[WIG 94] WIGHT G. D., Fundamentals of Air Sampling, CRC Press, 1994.
[WIL 65] WILLIAMS F.A., Combustion Theory, Addison-Wesley Publishing Co., 1965.
[WIL 85] WILLIAMS F.A., Combustion Theory, Benjamin Cumings Pub, 1985.
[WOL 79] WOLF K. B., Integral Transforms in Science and Engineering, Plenum Press,
1979.
[YAN 01] YANG W.J., Handbook of Flow Visualization, Taylor & Francis, 2001.
[YIH 77] YIH C. S., Fluid Mechanics, West River Press, 1977.
[ZIL 06] ZILL D.G., CULLEN M.R., Advanced Engineering Mathematics, Jones & Bartlett,
3rd edition, 2006.
Index
A C
Actions Calorific coefficients, 27
by contact, 18 Cauchy problem, 200, 202, 206,
volume, 18 207, 216, 235
Anemometry Causal signal, 368
Eulerian, 345 Causality, 3, 6, 406
Lagrangian, 342 Center of buoyancy, 65, 71
laser Doppler, 343 Cepstrum, 369
Archimedes’ theorem, 65 Characteristic curves, 205, 207, 209,
212, 214, 226, 240, 254
B Characteristic determinant, 208, 213
Characteristic equation, 208, 222,
Balance equations, 54; 227
extensive quantity, 151 Circulation
for sub-systems, 416 creation of, 335
global, 151, 182 of the velocity, 258
global for chemical species, 188 Command model, 411, 473
local, 151 Complex potential, 274
local for chemical species, 177 Compressibility coefficients, 27
Barotropic.fluid, 65 Compression of data, 385
Bernoulli theorem analytical methods, 387
first, 166, 167 arithmetic methods, 386
second, 269 Concentration measurements, 347
Biot and Savart’s formula, 259 Condition number, 353
Boundary conditions, 178, 255, 322 Conditioning of a linear system, 353
Boundary layer, 319, 323 Conjugated flows, 274
equations, 321 Conservation of volume, 160
thermal, 437 Contact actions, 57, 58
Couette flow, 137
D Energy equation, 116, 172, 187

entropy form of, 117
D’Alembert’s paradox, 281 Enthalpy, 24, 25, 26, 41, 164, 168,
Damköhler number, 197 174, 176, 188
Damping of oscillations, 294 generation, 167
Deformation rate tensors, 129 total, 167, 168, 174, 187, 193
Determinism, 6, 13 Entropy, 162, 164, 174, 211
Differential equations, 199 balance equation, 175
first order quasi-linear, 203 source, 33, 43, 56, 118
of order n, 218 Equations
quasi-linear partial, 220 elliptic, 227, 232, 234
reduced form, 229 equation of state, 190
second order partial, 225, 232 hyperbolic, 227, 231, 235, 272
system of partial, 220 mixed, 238
Diffusion, 73 parabolic, 227, 231, 312, 322
coefficient, 61, 88, 93 Equilibrium
equation, 97 local thermodynamic, 48
in a closed container, 82 of the atmosphere, 67
in a fixed component, 91 stability of, 68
in a fluid at rest, 81, 90 Errors
in a moving medium, 83 in measurements, 351
in steady evaporation, 82 in numerical techniques, 384
isothermal, 87 Established regimes, 453, 474
thermal, 58 for boundary layers, 461
velocities, 84, 89, 93 harmonic, 456
Dirac distribution, 418, 446 in continuous media, 458
Discretization, 414, 472 Euler equations, 163
schemes, 250 Evaporation
Discretized signals, 381 steady, 82, 91
Dissipation function, 175, 176 unsteady, 94
Divariant fluid, 49, 162, 165, 167, Extensive quantities, 16
170
Drag, 280
F
coefficient, 195
Driving pressure, 66 Fick’s law, 61, 87
Dynamic Floaters, 70
moment, 108 Flow
resultant, 108 1D unsteady, 210, 239, 253
2D steady, 212, 213
E 2D unsteady, 221
around a circular cylinder, 279
Eckert number, 192 compressible inviscid fluid, 270
Energy in a nozzle, 244
internal, 21, 32, 40 in a pipe, 314
mechanical, 44 quasi-1D, 309
Index 499
steady, 327 Heat-transfer coefficient, 73

unsteady, 327 Heisenberg-Gabor inequality, 372
Flux Hilbert transform, 368
mass, 125 Homoentropic fluid, 65, 67, 164
thermal, 412 Homogenity, 19
volume, 125 Hugoniot relation, 169
Flux density vector, 77, 124 Hydrostatics, 69
diffusion, 84 center of buoyancy, 71
extensive quantities, 50 center of pressure, 69
mass, 78 equations, 66
molar, 52, 78 thrust, 69
thermal, 52 Hysteresis, 407
Force
definition of, 104 I
deriving from a potential, 161
external, 108 Impulse, 105
internal, 109 input, 418
power of, 107 response, 447
power of external, 110 Initial conditions, 181, 418
power of internal, 109 Integral transforms, 363
viscous volume, 148 Intermittency, 379
Fourier Internal energy equation, 174
discrete transform of discretized Inviscid fluid, 163, 166, 172, 187
signal, 385 Irreversibility, 3
fast Fourier transform, 385 Irreversible phenomena, 49, 58
series, 362 Irrotational flows, 273
short time transform, 370 Isolated systems, 2
transform, 364
transform of a discretized signal, K, L
383 Kinetic energy, 106
Fourier’s law, 59 balance, 115
Free surfaces, 180 coefficient, 318
Friction coefficient, 194 enthalpic form of theorem, 165
Froude number, 192 theorem, 107, 109, 164, 186
Fully immersed flows, 181 Lagrange’s theorem, 267
Laplace transform, 368, 450
G, H Laplace’s law, 66, 181, 287
Gabor transform, 373 Laws for viscous fluids, 139
Galilean reference frame, 101, 123 Bingham, 142
Gibbs function, 25 Newtonian, 141
Gibbs-Helmholtz relation, 25 Ostwald-de Waele, 141
Heat convection equation, 176 purely viscous, 138, 141
Heat engines, 45 time-dependent, 140
Heat equation, 72
Lewis number, 88 N, O
Lift, 280
Navier-Stokes equations, 161, 190
Newtonian fluid, 163, 176, 407
M
Non-dimensional parameters, 189
Mach lines, 215 Non-isothermal mixtures, 97
Mach number, 160, 169, 192, 215 Numerical solutions, 250
Mass Nusselt number, 194
average velocity, 79 Ohm’s law, 61
concentration, 76 Onsager relations, 42, 59
conservation, 154, 160, 170, 184,
190 P, Q
enthalpy, 65
fraction, 76 Parametric models, 451
Matched asymptotic expansions, 437 Parseval’s theorem, 362
Material derivative, 122 Particle Image Velocimetry, 349
of a flux integral, 128 Péclet diffusion number, 192, 197
of a volume integral, 125 Perfect gas, 23, 28, 39
Modal reduction, 465 Perturbation, 296
continuous media, 469 parameter, 300
discrete models, 471 regular, 296
discrete systems, 466 singular, 305
Model reduction Poiseuille flow, 142
input-output, 473 Potential
Modulation equation, 271
amplitude, 378 flows, 269, 282
frequency, 378 lines, 276
Molar Power
average velocity, 79 mechanical, 173
concentration, 76 thermal, 173
fraction, 76 Prandtl number, 192
molarity, 76 Pressure
momentum, 102, 103 center of, 70
angular, 107 driving, 162, 170, 180, 181, 194
balance, 113, 160 forces, 63, 68
coefficient, 317 mean driving, 318
conservation of, 103 measurement of, 341
flux, 114 partial, 76
flux theorem, 184 total, 166
linear, 107 total driving, 166
MP3 encoding, 388 Process, 13
natural, 15, 16
possible, 15
quasi-static, 15
reversible, 15
Index 501
Progressive wave, 241 State, 11

Propeller thrust, 185 equation, 19, 23
general equation, 19, 20
Q, R internal, 9
of equilibrium, 12, 15, 19
Quantity of acceleration representation, 412, 453
Rabinowitsch-Mooney relation, 144 variables, 9
Rayleigh criterion, 371 vector, 408
Reconstruction of a signal, 363 Statics of fluids, 63
Reduced extensive quantities, 47 Stefan’s law, 45
Representation Streakline, 121
analytical, 448 Stream function, 155, 273
by differential equations, 455 Streamline, 121, 155, 166, 276
external, 410, 446, 456 Strouhal number, 195
internal, 408 Superposed flows, 277
mixed, 438 Supersonic flow, 244
modal, 445, 471 Surface
state, 408 tension, 287
Representation of signals, 357 waves in shallow water, 284
analytical, 360 Synthesis of musical signals, 397
on basis of functions, 361 MIDI system, 397
Reynolds number, 146, 193, 195 musical instruments, 399
Rotation vector, 123, 257 System, 13
material derivative, 260 closed, 14
insulated thermal, 32
S isolated mechanical, 102
Saint-Venant relation, 167 open, 14, 112
Sampling of signals, 382 out-of-equilibrium, 30, 37
Schmidt number, 197 rigid, 111
Separation
steady flow, 325 T
unsteady flow, 334 Temperature
Shannon’s theorem, 384 mean mixing, 415
Shock waves, 239, 247, 248 measurements, 346
Sound analysis, 390 Thermal
musical signals, 393 conduction, 59, 62, 72
Source, 275 conductivity, 60
Sources of entropy, 35 diffusion, 62
Specific heats, 27 diffusion coefficient 98
Spectrogram diffusivity, 60, 72, 88
sonagram, 376 radiation, 45
Speed of sound, 28 Thermal system
Spring-mass oscillator, 288 2 components, 423, 449
3 components (series), 425
3 components (star), 429 Viscous

n components, 414 fluid, 156
Thermodynamic stress, 163, 173, 186
fluxes, 57 Visualizations, 348
forces, 57, 58 Volume
potentials, 23, 25 balance, 184
Thrust of a rocket, 114 source, 53
Time-frequency representation, 374 Vortex, 275
Time-invariant system, 408 line, 258
linear, 408, 410, 420 stretching, 263
Time-scale transform, 373 tube, 258
Trajectory, 121 velocity field, 260
Transitional regimes, 421 Vorticity
Tricomi equation, 238 diffusion of the, 265
Turbulence equation, 171, 261
analysis of signals, 402 vector, 123, 257
Turbulent and acoustic pressure
fluctuations, 403 W
Two wall thermal system, 439
Wall
adiabatic, 15
V
diathermic, 15
Variables Wall thermal problem
Eulerian, 14, 119, 412 constant temperature faces, 432
extensive, 18 shocks on the faces, 435
intensive, 21, 25, 37, 412 Wave equation, 2
Lagrangian, 14, 119 Wavelet transform, 373
Vaschy-Buckingham theorem, 194 Weak concentration, 83, 96
Velocity potential, 273
Viscometric flow, 135
Viscosity, 132 (see also Laws for
viscous fluids)
bulk, 147
definition of, 146
dynamic, 147
kinematic, 148
physical origin, 133

Fundamental of Transport Phenomena

Uploaded by

Copyright:

Available Formats

Fundamental of Transport Phenomena

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Fundamental of Transport Phenomena

Uploaded by

Copyright:

Available Formats

Fundamentals

ISTE Ltd John Wiley & Sons, Inc.

© ISTE Ltd, 2009

Library of Congress Cataloging-in-Publication Data

British Library Cataloguing-in-Publication Data

Chapter 1. Thermodynamics of Discrete Systems . . . . . . . . . . . . . . . . 1

Chapter 2. Thermodynamics of Continuous Media . . . . . . . . . . . . . . . 47

2.2.2. Pressure forces on solid boundaries. . . . . . . . . . . . . . . . . . . 68

Chapter 3. Physics of Energetic Systems in Flow. . . . . . . . . . . . . . . . . 101

Chapter 4. Fluid Dynamics Equations . . . . . . . . . . . . . . . . . . . . . . . 151

4.3. Balance of mechanical and thermodynamic quantities . . . . . . . . . . 160

Chapter 5. Transport and Propagation. . . . . . . . . . . . . . . . . . . . . . . 199

5.4.5. Second order partial differential equations and their boundary

Chapter 6. General Properties of Flows . . . . . . . . . . . . . . . . . . . . . . 257

Chapter 7. Measurement, Representation and Analysis of

Chapter 8. Thermal Systems and Models . . . . . . . . . . . . . . . . . . . . . 405

8.5.3. Established regimes in continuous media . . . . . . . . . . . . . . . 458

Appendix 1. Laplace Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . 477

Appendix 2. Hilbert Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . 481

Appendix 3. Cepstral Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 483

Appendix 4. Eigenfunctions of an Operator . . . . . . . . . . . . . . . . . . . . 487

This book is primarily aimed at students, engineers and researchers in fluid

Parallel to this, systems became an object of study in themselves (automatic

Chapter 1 is devoted to a synthetic presentation of thermodynamics. After

The association of mechanical phenomena with thermodynamics is briefly

Chapter 4 is dedicated to the writing of the general equations of the dynamics of

Chapter 5 discusses the classification of partial differential equations in fluid

Chapter 6 is dedicated in the main to the influence of diffusion in the convection

Improvements in computing have of course also led to considerable progress in the

In Chapter 7, we also indicate in a synthetic manner the essential ideas necessary

Chapter 8 is dedicated to modeling which provides a general context for the

Thermodynamics of Discrete Systems

The general objective of thermodynamics is to describe the properties of matter.

1.1. The representational bases of a material system

1.1.1.1. Geometric Euclidean space and physical quantities

The geometric description of space is independent of the presence of matter; in

We thus postulate (Axiom 1) the existence of a geometric space whose structure

Finally, we suppose (Axiom 3) that matter can be characterized by physical

1.1.1.2. The existence of isolated systems and the definition of time

The definition of time should thus be appropriately chosen. Previously associated

We will thus postulate (Axiom 4) that physical phenomena are reproducible,

1.1.1.3. Causality and irreversibility

Let us consider an isolated mechanical oscillator with friction, which can be

mx(t )  fx (t )  kx(t ) 0 [1.1]

whose oscillatory solution takes the form A exp  f t / 2m cos Yt  M .

By multiplying equation [1.1] by x (t ) and integrating with respect to time

Changing the direction of time would be equivalent to changing the term f x (t )

Let us take as an example three equations representative of constant coefficient,

The general solution of Laplace’s equation (which is elliptic) at a point requires

The preceding phenomena can be interpreted in a number of ways:

1) In terms of energy dissipation and of the creation of entropy

The heat equation translates a smoothing of temperature distributions, which

1.1.1.4. Causality and determinism

However, the question of determinism is not as simple as it might seem, in

A similar theorem (Cauchy-Kovalevskaïa), but with stricter analycity conditions

be developed in power series), leads to a unique analytic solution y M (x ) in the

1.1.2. Systems analysis and thermodynamics

The characterization of a material device can thus vary depending on the

In general, the objective of a system is to transform some input quantities u(t),

1.1.2.2. External description (black box)

mx(t ) fx (t ) kx(t ) 0 [1.1]

whose oscillatory solution takes the form A exp f t / 2m cosYt M .

By multiplying equation [1.1] by x (t ) and integrating with respect to time

Changing the direction of time would be equivalent to changing the term f x (t )

such that we obtain the classic definition dE dW dQ , in which the differential

dE pdV TdS PdN

dH d E pV Vdp TdS PdN