tessarotto2009

Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

Physica A 388 (2009) 3737–3744

Contents lists available at ScienceDirect

Physica A
journal homepage: www.elsevier.com/locate/physa

Phase-space Lagrangian dynamics of incompressible thermofluids


Marco Tessarotto a,b,c , Claudio Cremaschini d,e , Massimo Tessarotto c,f,∗
a
Department of Electronics, Electrotechnics and Informatics, University of Trieste, Italy
b
Civil Protection Agency, Regione Friuli Venezia-Giulia, Palmanova (Udine), Italy
c
Department of Mathematics and Informatics, University of Trieste, Trieste, Italy
d
International School for Advanced Studies (SISSA), Trieste, Italy
e
National Institute of Nuclear Physics (INFN), Trieste, Italy
f
Consortium for Magnetofluid Dynamics, Trieste, Italy

article info abstract


Article history: Phase-space Lagrangian dynamics in ideal fluids (i.e., continua) is usually related to the so-
Received 3 March 2009 called ideal tracer particles. The latter, which can in principle be permitted to have arbitrary
Received in revised form 31 May 2009 initial velocities, are understood as particles of infinitesimal size which do not produce
Available online 6 June 2009
significant perturbations of the fluid and do not interact among themselves. An unsolved
theoretical problem is the correct definition of their dynamics in ideal fluids. The issue
PACS:
is relevant in order to exhibit the connection between fluid dynamics and the classical
05.20Jj
05.20.Dd
dynamical system, underlying a prescribed fluid system, which uniquely generates its time-
05.70.-a evolution.
The goal of this paper is to show that the tracer-particle dynamics can be exactly
Keywords: established for an arbitrary incompressible fluid uniquely based on the construction of an
Incompressible Navier–Stokes–Fourier inverse kinetic theory (IKT) [M. Tessarotto, M. Ellero, Bull. Am. Phys. Soc. 45 (9) (2000) 40;
equations
M. Tessarotto, M. Ellero, AIP Conf. Proc. 762 (2005) 108. RGD24, Italy, July 10–16, 2004;
Kinetic theory
Lagrangian dynamics
M. Ellero, M. Tessarotto, Physica A 355 (2005) 233; M. Tessarotto, M. Ellero, Physica A 373
(2007) 142, arXiv: physics/0602140; M. Tessarotto, M. Ellero, in: M.S. Ivanov, A.K. Rebrov
(Eds.), Proc. 25th RGD, International Symposium on Rarefied gas Dynamics, St. Petersburg,
Russia, July 21–28, 2006, Novosibirsk Publ. House of the Siberian Branch of the Russian
Academy of Sciences, 2007, p. 1001, arXiv:physics/0611113; M. Tessarotto, C. Cremaschini,
Strong solutions of the incompressible Navier–Stokes equations in external domains: Local
existence and uniqueness, arXiv:0809.5164v1 [math-ph], 2008]. As an example, the case
of an incompressible Newtonian thermofluid is considered here.
© 2009 Elsevier B.V. All rights reserved.

1. Introduction

The basic aspect of fluid dynamics is related to the definition of the Lagrangian dynamics which characterizes both
compressible and incompressible fluids. The customary approach to the Lagrangian formulation is based typically on a
configuration-space description, i.e., on the introduction of the (configuration-space) Lagrangian path, r(t ), spanning the
configuration-space (fluid domain) Ω . Here r(t ) denotes the solution of the initial-value problem Dr Dt
= V(r, t ), with

r(to ) = ro . Here D
Dt
≡ ∂t
+ V · ∇ denotes the so-called ‘‘fluid’’ convective derivative, ro an arbitrary vector belonging
to Ω and V(r, t ) the velocity fluid field, to be assumed continuous in Ω (closure of Ω ) and suitably smooth in Ω . However,

∗ Corresponding author at: Department of Mathematics and Informatics, University of Trieste, Trieste, Italy. Tel.: +39 040 558 2643; fax: +39 040 558
2666.
E-mail address: massimo.tessarotto@cmfd.univ.trieste.it (M. Tessarotto).

0378-4371/$ – see front matter © 2009 Elsevier B.V. All rights reserved.
doi:10.1016/j.physa.2009.06.008
3738 M. Tessarotto et al. / Physica A 388 (2009) 3737–3744

in turbulence theory the statistical formulation for the associated joint probability density for velocity increments requires
the introduction of a phase-space representation of suitable type [1], in which usually the phase-space is identified with
Γ = Ω × R3 (with closure Γ = Ω × R3 ), here denoted as restricted phase-space. Therefore it is natural to seek possible
phase-space representations of this type for fluid systems. The goal of this investigation is concerned with the formulation
of restricted phase-space Lagrangian dynamics in such a way that the phase-space Γ coincides with the direct product space
Γ = Ω × V , Ω being the fluid domain and V (velocity space) the set R3 . In particular in this paper, extending the formulation
previously developed (Cremaschini et al. [2] and Tessarotto et al. [3,4]), we shall adopt for this purpose a so-called phase-
space inverse kinetic theory (IKT)(see also Tessarotto et al. [5–10]). Basic feature (of such an approach) it that is relies on first
principles, i.e., classical statistical mechanics and a prescribed complete set of fluid equation. This permits us to advance in
time the relevant fluid fields by means of phase-space Lagrangian equations defined by the vector field X(x, t ), namely

 dx
= X(x, t ),
dt (1)
x(t ) = x ,
o o

where xo is an arbitrary initial state of Γ (closure of the phase-space Γ ). The result appears relevant in particular for the
following reasons: (1) the Lagrangian dynamics here determined permits to advance in time self-consistently the fluid fields,
i.e., in such a way that they satisfy identically the required set of fluid equations. For isothermal fluids, this conclusion is
consistent with the results indicated previously [7]; (2) the Lagrangian dynamics takes into account the specific form of the
phase-space distribution function which advances in time the fluid fields; (3) the theory permits an exact description of the
motion of those particles immersed in the fluid which follow the Lagrangian dynamics (classical molecules).
Phase-space Lagrangian dynamics and particle dynamics in ideal fluids are closely related issues. In fact both must be
uniquely described via a suitable complete set of fluid fields {Z (r, t )} which define the fluid state. This refers, in particular, to
the so-called ideal tracer particles, for which both self-interaction produced by the perturbations of the fluid fields generated
by the same particles and binary collisions among them are negligible (in this sense they can therefore be intended also
as ‘‘collisionless’’). It is well known, however, that in customary approaches (see for example Maxey and Riley [11]) the
equations of motion for ideal tracer particle are only known in some approximate sense and therefore do not reproduce
exactly the correct fluid dynamics.
The purpose of this paper is to show that an exact solution can be reached for phase-space Lagrangian dynamics, and
in particular for the conventional tracer dynamics, based on the formulation of a suitable IKT. By definition, an IKT must
provide the complete set of fluid equations describing the fluid, by means of velocity moments of an appropriate phase-
space probability density function (pdf). We intend to show that such a theory can be uniquely determined in the framework
of classical statistical mechanics by invoking suitable statistical assumptions on the IKT. In particular, we present here a
theory which applies to incompressible Newtonian fluids, including both isothermal and non-isothermal fluids [2–4]. In
the following we intend to show that customary tracer-dynamics equations due to several authors – including Tchen [12],
Corrsin and Lumley [13], Buevich [14] and Riley [15] and Maxey and Riley [11] – are incompatible with the exact phase-space
Lagrangian formulation here obtained.
In detail the plan of the paper is as follows. First, in Section 2 previous approaches to ideal tracer-particle dynamics
are summarized. Second in Section 3 a comparison between the Eulerian and the Lagrangian phase-space approaches is
provided. Furthermore, in Section 4 the IKT for incompressible thermofluids is presented. This permits us to determine
the appropriate form of the vector field X(x, t ). Next, in Section 5 the Lagrangian formulation of IKT is discussed in detail.
The new set of phase-space Lagrangian equations are shown to advance uniquely in time the relevant fluid fields of an
incompressible thermofluid. As a basic consequence, in Section 6 we will derive the exact dynamics of ideal tracer particles
(see below for definition), comparing it with previous results.

2. Previous approaches to tracer-particle dynamics

The motion of small particles (such as solid particles or droplets, commonly found in natural phenomena and industrial
applications) which can be injected in a fluid with arbitrary initial velocity, in practice, may be very different from that of
the fluid. The accurate description of particle dynamics, as they are pushed along erratic trajectories by binary collisions (in
real fluids) and by fluctuations of the fluid fields (in ideal fluids), is fundamental to transport and mixing in turbulence [1].
It is essential, for example, in combustion processes [16], in the industrial production of nanoparticles [17] as well as in
atmospheric transport, cloud formation and air-quality monitoring of the atmosphere [18,19]. The Lagrangian approach –
denoted as the Lagrangian turbulence (LT) – has been fruitful in advancing the understanding of the anomalous statistical
properties of turbulent flows [20]. In particular, the dynamics of particle trajectories has been used successfully to describe
mixing and transport in turbulence [16,21]. Nevertheless, issues of fundamental importance remain unresolved (see for
example Refs. [22,23] for recent results regarding the Lagrangian view of passive scalar turbulence). In the past, the treatment
of the Lagrangian dynamics in turbulence was based on stochastic models of various nature, pioneered by the meteorologist
Richardson [24] (see also Refs. [22,23]). These models, which are based on tools borrowed from the study of random
dynamical systems, typically rely – however – on experimental verification rather than on first principles. However, in
most cases there remains a lack of experimental data to verify the reliability of such models [23]. Verification can be based,
M. Tessarotto et al. / Physica A 388 (2009) 3737–3744 3739

in particular, on the measurement of fluid particle trajectories, obtained by seeding a turbulent flow with a small number
of tracer particles and following their motions with an imaging system. On the other hand, the accurate evaluation of the
Lagrangian velocity in laboratory turbulence experiments requires measurements of positions of tracer particle by using a
suitable tracking system able to resolve very short time (and spatial) scales. In practice this can be a very challenging task
since particle motions must be measured on very short time scales.
As for the theory itself, rigorous results have been scanty, probably because of the subject complexity. In the case of ideal
or dilute real fluids, however, particle motion is necessarily collisionless (in the sense specified above) while the dynamics
of ideal tracer particles is controlled by the force produced on them only by the unperturbed fluid fields. Several authors
have tried in the past to derive, based on phenomenological arguments, an approximate equation for the ideal tracer-particle
dynamics, describing the motion of a particle suspended in a non-uniform flow. Since the original Basset–Boussinesq–Oseen
(BBO) equation [25–27], formulated in the case of a uniform flow, several papers have appeared proposing modifications or
corrections of the same equation for non-uniform flows (for a review see Ref. [11]). The first attempt at a generalization of
this type is due to Tchen [12], who considered the motion of a rigid sphere in an incompressible isothermal Navier–Stokes
(NS) fluid. Tchen [12] derived an approximate equation of motion for a finite-size spherical particle of radius a and mass mP
describing its dynamics in terms of the Newtonian state of its center of mass {r(t ), v(t )}. His equation was later modified by
Corrsin and Lumley [13], to take into account contributions due to pressure gradients previously ignored, and by Buevich [14]
in order to consider also the effect of viscous stress. The version of the equation currently adopted by some authors (see for
example Gui et al. [28] where it was used to investigate modifications of turbulence) is, however, the one later developed
by Maxey and Riley [11] in which also the buoyancy contribution produced by the volume displaced by the particle was
taken into account. In the approximation in which perturbations of the fluid fields produced by the particle are negligible
the equation of motion developed by Maxey and Riley reduces to:

d DV(x, t ) 1 d
mP v(t ) = mF − mF {v(t ) − V(x(t ), t )} − (mP − mF ) g (2)
dt Dt x=r(t ) 2 dt x=r(t )

(M–R equation), where the last contribution on the r.h.s. denotes the so-called buoyancy effect. Here the notation is standard.
Thus, {V(r, t ), p(r, t )} are respectively the fluid velocity and pressure, f the volume force density and finally ρ0 , ν > 0 the
constant mass density and kinematic viscosity (with ν related to the dynamic viscosity µ by the identity ν =µ/ρ0 ). In
particular, f can be written f = −∇φ + fR , with ρ0 g = −∇φ , where g is the constant local gravity acceleration, φ = ρ0 gz
the gravitational potential (hydrostatic pressure) and fR a possible non-potential force density. Moreover, mF is the mass of
the fluid displaced by the sphere, dt d
≡ ∂∂t + v(t ) · ∇ the ‘‘particle’’ convective derivative and V(r(t ), t ) and v(t ) − V(r(t ), t )
are respectively the fluid and particle relative velocities evaluated at the position of the particle center of mass. It should
be noted, however, that also this equation is still unsatisfactory. In fact, it is obtained by requiring that the particle velocity
remains always suitably close to the fluid one, so that contributions due to the relative velocity – i.e., proportional to the
particle relative velocity u(t ) ≡ v(t ) − V(r(t ), t ) – are actually ignored in Eq. (2), requiring that for all {r(t ), v(t )} there
results:
|u(t )|  |V(r(t ), t )| . (3)
The limitation appears serious because the actual dynamics (i.e., both the velocity and the acceleration) of ideal tracer
particles may be in principle very different from that of the fluid elements. In addition, the accurate description of ideal
particle dynamics is essential both in LR and in environmental fluid dynamics (dynamics of anthropogenic pollutants in the
atmosphere, diffusion of dusty particles, droplets, aerosol particles, etc.). This involves, in fact, the ability to simulate tracer
dynamics in a variety of different physical conditions and in fluid flows characterized by a turbulent behavior. Therefore, an
open issue remains the very definition of the dynamics of tracer particles which may be injected in an ideal fluid with arbitrary
initial velocities. Clearly, such a formulation – if achievable at all – should rely exclusively on first principles, i.e., in particular
the exact validity of the fluid equations. This problem is closely related to the formulation of phase-space approaches for
fluid systems, based on the introduction of suitable phase-space representations of classical fluid dynamics in terms of
an appropriate phase-space pdf. By construction, the dynamics of a fluid is completely described by the time-evolution of
its fluid fields (which, in turn, are assumed as classical solutions of a well-posed initial-boundary value problem defined
by a complete set of fluid equations). Therefore, unless suitable restrictions are posed, phase-space dynamics remains
intrinsically non-unique. This is true, in particular, due to the arbitrariness in the choice of the possible phase-space and
the definition of the evolution equation for the pdf.

3. Eulerian and Lagrangian phase-space approaches

As it is well known, phase-space descriptions of fluids can be achieved in principle choosing either an Eulerian or a
Lagrangian point of view. Based on the IKT approach for incompressible fluids earlier developed [5–9] such a connection
can be uniquely established. IKT is based on the identification of the complete set of fluid fields (which describe the fluid)
with velocity moments of a suitably-defined kinetic distribution function f (x, t ). The pdf is assumed to satisfy the basic
principles of classical statistical mechanics, which include in particular:
(1) the principle of conservation of probability;
3740 M. Tessarotto et al. / Physica A 388 (2009) 3737–3744

(2) a suitable correspondence principle (Ellero et al. [7]) whereby appropriate (velocity) moments of the pdf can be identified
with the relevant fluid fields;
(3) the principle of entropy maximization (PEM, Jaynes, [29]);
(4) and an entropic principle assuring that the statistical entropy cannot decrease in time (Tessarotto [3]).
The first axiom implies that the pdf must satisfy a Liouville equation, i.e., a Vlasov-type inverse kinetic equation (IKE, Ellero
et al. [7]). As it is well known, this type of kinetic equation is, in fact, appropriate for the statistical description of particles
subject solely to mean-field interactions. This is consistent with the assumption of an ideal fluid, i.e., a continuum in which
the fluid elements are, by definition, subject only to mean-field interactions. In such a case, the time-evolution of the pdf is
determined by the Eulerian IKE
Lf (x, t ) = 0. (4)
∂ ∂
Here f (x, t ) denotes the Eulerian representation of the pdf, L is the streaming operator Lf ≡ +
∂t
f · {X(x, t )f },
∂x
X(x, t ) ≡ {v, F(x, t )} a suitably smooth vector field, while v and F(x, t ) denote respectively the velocity and an appropriate
‘‘mean-field’’ acceleration vector field. The implications of the principle of maximum entropy [29] and of the entropic
principle, both involving the assumption that the Boltzmann–Shannon (B–S) entropy functional
Z
S (t ) ≡ S (f (x, t )) = − dxf (x, t ) ln f (x, t ) (5)
Γ

exists, have been discussed elsewhere [2,3] (see also Tessarotto et al. [9]). In particular this provides a well-defined initial
condition for the pdf, f (x, to ) = fM (x, to ) [see below Eq. (14)] and also a (generally non-unique) representation for the
streaming operator L (Tessarotto et al. [8]). As a main consequence the same approach can in principle be used to determine
in a rigorous way the Lagrangian formulation for arbitrary complex fluids. Although the choice of the phase-space Γ is in
principle arbitrary, in the case of incompressible isothermal fluids, it is found [6] that it can always be reduced to the direct-
product space Γ = Ω × V (restricted phase-space), where Ω , V ⊆ R3 , Ω is an open set denoted as configuration space of
the fluid (fluid domain) and V is the velocity space.

4. The case of an incompressible thermofluid

Let us consider for definiteness an incompressible, viscous and generally non-isentropic thermofluid (which comprises
as a particular case also the treatment of incompressible isothermal fluids earlier developed in Ref. [7]). This is described
by the fluid fields {Z } ≡ {ρ ≥ 0, V, p ≥ 0, T > 0, ST } , to be identified respectively with the mass density, the fluid
velocity, pressure, temperature and thermodynamic entropy. In the open set Ω they are assumed to satisfy the so-called
incompressible Navier–Stokes–Fourier equations (INSFE), i.e.,
ρ = ρo > 0, (6)
∇ · V = 0, (7)
D 1
V = FH − [∇ p − f] + ν∇ 2 V, (8)
Dt ρ0
ν ∂ Vi ∂ Vk
 2
D 1
T = χ∇ 2 T + + + J ≡ K, (9)
Dt 2cp ∂ xk ∂ xi ρ0 cp

ST ≥ 0, (10)
∂t
where ρo is a constant,

1
FH ≡ − [∇ p − f] + ν∇ 2 V (11)
ρ0
D
is the fluid acceleration and Dt = ∂∂t +V·∇ the convective derivative. These equations are assumed to satisfy a suitable initial-
boundary value problem (INSFE problem) so that a smooth (strong) solution exists for the fluid fields {Z } . Here the notation
is standard [3]. Thus, Eqs. (6)–(8) together are denoted incompressible Navier–Stokes equations (INSE) [with Eqs. (6) and
(7) representing respectively the so-called incompressibility and isochoricity conditions and Eq. (8) the forced Navier–Stokes
equation forced Navier–Stokes equation written in the Boussinesq approximation], while Eq. (9) is the Fourier equation. Finally
(1) denotes the entropy inequality whichdefines the second principle of thermodynamics. As a consequence, in such a case the
force density f reads f = ρ0 g 1 − kρ0 T + f1 , where the first term represents the (temperature-dependent) gravitational
force density, while the second one (f1 ) the action of a possible non-gravitational externally-produced force. Hence f can be
written also as f = −∇φ + fR ,where φ = ρ0 gz and fR = −ρ0 gkρ0 T + f1 denote respectively the gravitational potential
(hydrostatic pressure) and the non-potential force density. Moreover, in Eq. (9) K and J are the quantities of heat generated
per unit volume and unit time by all sources and, respectively, only by the external sources. In particular, the inequality (10)
M. Tessarotto et al. / Physica A 388 (2009) 3737–3744 3741

defines the so-called second principle for the thermodynamic entropy ST . For its validity in the following we shall assume
that there results everywhere in Ω × I
Z  
1
dr χ∇ 2 T + J ≥ 0, (12)
Ω ρ0 cp
which defines a so-called externally heated thermofluid. In these equations g, kρ0 , ν, χ and cp are all real constants which
denote respectively the local acceleration of gravity, the density thermal-dilatation coefficient, the kinematic viscosity, the
thermometric conductivity and the specific heat at constant pressure. Thus, by taking the divergence of the N–S equation
(8), there it follows the Poisson equation for the fluid pressure p, namely ∇ 2 p = −ρ0 ∇ · (V · ∇ V) + ∇ · f, with p to be
assumed non-negative and bounded in Ω × I.

4.1. Functional uniqueness of IKT

Let us now assume that f (x, t ) is a solution of the Eulerian kinetic equation (4) defined in a suitable extended phase-space
Γ × I, where I ⊆ R is a suitable time interval. In such a case, we intend to show that f (x, t ) [to be assumed R strictly positive]
can be defined in such a way that the fluid fields V, p1 and ST can be identified with its velocity moments R3 dvG(x, t )f (x, t ),
where respectively G(x, t ) = v, ρo u2 /3, −lnf (x, t ), u ≡ v−V(r, t ) is the relative velocity and p1 the kinetic pressure defined
as:
ρ0
p1 = p0 (t ) + p − φ + T. (13)
mP
Here p0 (t ) (to be denotes as pseudo-pressure [7]) is an arbitrary strictly positive and suitably smooth function defined in I.
Moreover, mP > 0 is a constant mass, whose value remains in principle arbitrary. In particular it can be identified with the
average mass of the molecules forming the fluid. Finally, the thermodynamic entropy ST can be identified with the Shannon
statistical entropy functional S (f (x, t )), provided the function p0 (t ) is a suitably prescribed function and f (x, t ) is strictly
positive in the whole set Γ × I. To reach the proof, let us first show that, by a suitable definition of the vector field F(x, t ), a
particular solution of the IKE (4) is delivered by the Maxwellian distribution function:

u2
 
1
fM (x, t ) = exp − , (14)
π 2 vTh vTh

where vTh = 2p1 (r, t )/ρ0 is the thermal velocity driven by the kinetic pressure p1 (r, t ). Based on the results earlier obtained
for ideal isothermal and incompressible fluids [7–9] and incompressible thermofluids [3] the following theorem is reached:
Theorem — IKT formulation for INSFE
Let us assume that the INSFE problem admits a smooth strong solution in Γ × I, such that the inequality (12) is fulfilled
and the fluid fields {Z } belong to the ‘‘minimal functional setting’’ (see Ref. [7]). Moreover, let us assume that in Ω × I :
2
(1) the pdf f (x, t ) is suitably smooth, strictly positive and admits the velocity moments G(x, t ) = 1, v, ρo 3u , ρo u u3 , ρo uu;
2

2
thus, we denote in particular Q = ρo d3 v u u3 f and Π = ρo d3 v uuf ;
R R

(2) the B–S entropy integral (5) exists in the time interval I ⊆ R;
(3) the inequality (12) is assumed to hold;
(4) there results identically (correspondence principle):
Z
d3 vf (x, t ) = 1, (15)
Z
d3 vvf (x, t ) = V(r, t ), (16)
Z
u2
ρo d3 v f (x, t ) = p1 (r, t ), (17)
3
ST (t ) = S (f (x, t )). (18)

Then it follows that:


(T1 ) the local Maxwellian distribution fM (x, t ), defined by Eq. (14), is a particular solution of the inverse kinetic equation
(4);
(T2 ) the mean-field acceleration vector field F reads
F(x, t ; f ) = F0 + F1 . (19)
3742 M. Tessarotto et al. / Physica A 388 (2009) 3737–3744

The functional form of the vector fields F0 , F1 is determined uniquely by requiring that they depend only on the velocity
moments indicated above. They read respectively:
1 h i
F0 (x, t ; f ) = ∇·Π − ∇ p1 + fR + D(x, t ) + ν∇ 2 V, (20)
ρ0
vth
2
u2
   
1 1 1 1 h i 3
F1 (x, t ; f ) = u A+ ∇ ·Q − ∇·Π ·Q + ∇·Π − (21)
2 p1 p1 p21 2p1 vth
2
2
where
1
D(x, t ) ={∇ V · u + u · ∇ V} , (22)
2
∂ ρ0 K
 
D 1 D
A≡ (p0 + p) − V · V− fR − ν∇ 2 V + ≡ p1 ; (23)
∂t Dt ρ0 mP Dt

(T3 ) for an arbitrary pdf f (x, t ) fulfilling assumptions 1–3, Eqs. (15)–(22) are fulfilled identically in Ω × I .
Proof. Let us, first, prove proposition T1 . For this purpose, let us assume that a strong solution of the INSFE problem exists
which in the set Ω × I satisfies identically Eqs. (6)–(9). In such a case it is immediate to prove that fM (x, t ) is a particular
solution of the inverse kinetic equation (4). This can be proved either: (a) by direct substitution of f ≡ fM (x, t ) in Eq. (4)
(Proposition A); (b) by direct evaluation of the velocity moments of the same equation for G(x, t ) = 1, v, u2 /3 (Proposition
B). Regarding Proposition B, we notice that the first two moment equations coincide respectively with the isochoricity and
Navier–Stokes equations [Eqs. (6) and (8)]. Therefore, the third moment equation delivers the Fourier equation [Eq. (9)]. The
same proof (Proposition B) is straightforward also if f 6= fM (x, t ). This is reached again imposing the same constraint Eq. (15)
on first velocity moment of the distribution function f . The proof of T2 and T3 follows in the same way by direct evaluation
2
of the velocity moments of the same equation for G(x, t ) = 1, v, ρo 3u , ρo u u3 , ρo uu.
2


5. Lagrangian formulation of IKT

The previous results permit us to formulate in a straightforward way also the equivalent Lagrangian form of IKE [see Eq.
(4)]. The Lagrangian formulation is achieved in two steps: (a) by identifying a suitable dynamical system, which determines
uniquely the time-evolution of the kinetic probability density prescribed by IKT. Its flow defines a family of phase-space
trajectories, here denoted as phase-space Lagrangian paths (phase-space LPs); (b) by proper parametrization in terms of
these curves of the pdf and the inverse kinetic equation, the explicit solution of the initial-value problem defined by the
inverse kinetic equation (4) is determined. First, we notice that — in view of the previous theorem it is obvious that the
phase-space LPs must be identified with the phase-space trajectories x(t ) of a classical dynamical system

xo → x(t ) = Tt ,to xo ≡ χ (xo , to , t ) (24)

(here denoted as INSFE dynamical system), with Tt ,to the corresponding evolution operator generated by the vector field
X(x, t ), to be prescribed according to the previous theorem. Hence, the initial-value problem (1) is realized by the equations
d
 r(t ) = v(t ),
 dt


d

v(t ) = F(r(t ), t ; f ), (25)
 dt
r(to ) = ro ,


v(to ) = vo ,

where the vector field F(r(t ), t ; f ) is defined by Eq. (19). Here, by construction v(t ) and F(r(t ), t ; f ) are respectively the
Lagrangian velocity and acceleration, both spanning the vector space R3 . In particular, F(r(t ), t ; f ), which is defined by Eqs.
(20)–(21), and depends functionally on the kinetic probability density f (x, t ), is the Lagrangian acceleration which corresponds
to an arbitrary kinetic probability density f (x, t ). From the theorem it follows that in the Lagrangian representation the kinetic
equation (Lagrangian IKE) can be written in the form

J (x(t ), t )f (x(t ), t ) = f (xo , to ) ≡ fo (xo ) (26)


∂ x(t )
where fo (xo ) is a suitably smooth initial pdf and J (x(t ), t ) is the Jacobian J (x(t ), t ) = ∂ xo
of the map xo → x(t ) which is
generated by Eq. (25). It follows that the Lagrangian equation (26) is uniquely specified by the proper definition of a suitable
family of phase-space LPs. Eq. (26) also provides the connection between the Lagrangian and Eulerian viewpoints. In fact
the Eulerian pdf, f (x, t ), is simply obtained from Eq. (26) by letting x = x(t ) in the same equation. As a result, the Eulerian
and Lagrangian formulations of IKT, and hence of the underlying moment (i.e., fluid) equations, are manifestly equivalent.
M. Tessarotto et al. / Physica A 388 (2009) 3737–3744 3743

6. The exact dynamics of ideal tracer particles

Let us now analyze in detail the equations of motion for ideal tracer particles immersed in an incompressible thermofluid
described by INSFE [Eqs. (6)–(10)]. First, it must be remarked that two types of forces can in principle be present: (a) a volume
force, acted by the fluid (the same one which is responsible of the phase-space Lagrangian dynamics); (b) particle-localized
forces, such as the gravitational pull, acting directly on the tracer particle. In particular, regarding the first one, its specific
form depends on the assumed pdf to be associated to the fluid. As discussed elsewhere [30] its choice depends closely on
the type of fluid to be considered, i.e., deterministic or stochastic (as appropriate to describe turbulent flows). In particular,
the position f ≡ fM (x, t ), with fM (x, t ) defined by Eq. (14) is suitable for the description of deterministic, i.e., non-turbulent
flows. In such a case the appropriate form of the equation is obtained from Eqs. (19)–(23). Instead, the general case, in which
one allows f 6= fM (x, t ), is provided by Eq. (19) with (20) and (21).
Let us now assume that mP 6= mF , mF denoting the mass of the displaced fluid. In view of the IKT approach the
equations of motion depend necessarily on the form of the pdf f (x, t ), and hence describe in this sense the conditional
phase-space dynamics. To construct the equation of motion for an ideal tracer particle of arbitrary mass, let us now assume,
for definiteness, that the sole particle-localized force acting directly on the tracer particle is produced by the gravitational
pull. The equation of motion for an ideal tracer particle of mass mP reads simply in such a case:
 
d
mP v(t ) − g = mF [F(x, t ; f ) − g] (27)
dt
which describes the conditional dynamics of an ideal tracer particle immersed in an incompressible thermofluid described
by a pdf f (x, t ). We stress that the form of the pdf depends on the specific assumptions made on the fluid [30]. The
interpretation of this equation is as follows. The terms on the l.h.s. represent the ‘‘inertial’’ and gravitational forces acting
on the tracer particle. Instead, all the terms on the r.h.s. represent the volume force acting responsible for the phase-space
Lagrangian motion. The physical interpretation of the various contributions appearing in the volume force [see Eqs. (19)–
(21)] is made transparent by representing them in terms of the vector fields mF F0 (x, t ; f ) and mF F1 (x, t ; f ), to be interpreted
as mean-field forces. One obtains in fact, in particular:

vth
2
 
mF F0 (x, t ; f ) − g − ∇ p1 ≡ mF FH + mF D(x, t ). (28)
2p1
It is immediate to prove that the terms in the first equation take into account the fluid and convective forces mF FH and
mF D(x, t ), with FH and D(x, t ) denoting respectively the fluid acceleration (11) and the convective term defined above [see
Eq. (22)]. In a similar way one can show that, in case mP 6= mF , Eq. (27) recovers also the buoyancy force density (mP − mF ) g
pointed out by Maxey and Riley [11]. Finally, for the sake of comparison, let us consider the case of an isothermal fluid and
require – consistent with Eq. (3) – that locally in the extended phase-space Γ × I condition (3) holds at time t = to . The
validity of previous tracer-particle equations [i.e., in particular the MR (Maxey–Riley) Eq. (2)] requires that the inequality (3)
holds for all times (t ∈ I). In particular, one can prove that in such a limit our Eq. (27) agrees with the MR equation (2). In
fact, it yields

d DV(x, t )
mP v(t ) ∼
= mF + mF F1 − (mP − mF ) g, (29)
dt Dt x=r(t )

where, in validity of (3), it follows that the pressure mean-field force can be approximated as mF F1 ∼
= − 12 mF
d
dt
{v(t ) − V(x(t ), t )} x=r(t ) . On the other hand, a serious objection to all of the previous tracer-dynamics equations is provided
by the possible violation of the asymptotic condition (3), which may not be uniformly fulfilled in the set I. This occurs, manifestly,
if an initial condition of the type |u(to )| ∼ |V(r(to ), to )| is imposed on a tracer particle (this requirement is not physically
unreasonable since, in principle, tracer particles might be injected in a fluid with arbitrary initial velocities). However, even
if initially (at t = to ) one requires the validity of (3) in general it may well be also that |u(t )| ∼ |V(r(t ), t )| at some later
time (t > to ). This can be achieved, for example, even imposing the initial condition |u(to )| = 0. The result is a consequence
of Eq. (27). Indeed one can prove that, even imposing the initial condition |u(to )| = 0, generally |u(t )| 6= 0, with |u(t )| not
satisfying (3). In other words, ideal tracer particles having initially the same local velocity of the fluid may develop in time
a finite relative velocity. This implies that, generally, the full exact tracer-dynamics equation, i.e., Eq. (27), should be used,
instead of the asymptotic approximation indicated above [see Eq. (29)].

7. Concluding remarks

In this paper the phase-space Lagrangian dynamics has been determined as appropriate for an incompressible
thermofluid described by a suitable set of fluid equations [INSFE, see Eqs. (6)–(10)]. We have shown that, based on the
formulation of a restricted phase-space inverse kinetic theory, the phase-space Lagrangian dynamics can be uniquely
established. The governing equations which determine the phase-space Lagrangian trajectories (LPs) are found to depend
functionally on the pdf [f (x, t )], to be uniquely associated to the fluid by means of the IKT here adopted. In particular, the
3744 M. Tessarotto et al. / Physica A 388 (2009) 3737–3744

theory permits us to advance uniquely in time f (x, t ) and in terms of the same pdf also the complete set of fluid fields which
describe the fluid. This feature is of fundamental importance in turbulence theory (see also related discussion in Ref. [30]).
As a further consequence, the dynamics of ideal tracer particles (i.e., for which the perturbations of the fluid fields
produced by the same particles are negligible) is established. Remarkably, this result overcomes limitations of customary
ideal tracer-dynamics equations [see in particular the Maxey and Riley equation [11] given by Eq. (2)]. All of these
equations are, actually, in disagreement with the present theory for finite particle relative velocities. The basic new result is
represented by Eq. (27) which describes, for arbitrary initial velocity, the conditional dynamics of an ideal tracer particle in
an incompressible thermofluid described by a suitable pdf f .

Acknowledgements

Work developed in cooperation with the CMFD Team, Consortium for Magnetofluiddynamics (Trieste University, Trieste,
Italy). Research partially performed in the framework of the COST Action P17 (EPM, Electromagnetic Processing of Materials),
the GDRE (Groupe de Recherche Européen) GAMAS and the MIUR (Italian Ministry of University and Research) PRIN
Programme: Modelli della teoria cinetica matematica nello studio dei sistemi complessi nelle scienze applicate.

References

[1] S.B. Pope, Turbulent Flows, Cambridge University Press, 2000, p. 463.
[2] C. Cremaschini, M. Tessarotto, AIP Conf. Proc. 1084 (2008) 188, arXiv:0806.4546, 2008.
[3] Marco Tessarotto, Claudio Cremaschini, Piero Nicolini, Massimo Tessarotto, AIP Conf. Proc. 1084 (2008) 182, arXiv:0806.4546, 2008.
[4] Marco Tessarotto, Massimo Tessarotto, AIP Conf. Proc. 1084 (2008) 483, arXiv:0807.0274v1 [physics.flu-dyn], 2008.
[5] M. Tessarotto, M. Ellero, Bull. Am. Phys. Soc. 45 (9) (2000) 40.
[6] M. Tessarotto, M. Ellero, AIP Conf. Proc. 762 (2005) 108. RGD24, Italy, July 10-16, 2004.
[7] M. Ellero, M. Tessarotto, Physica A 355 (2005) 233.
[8] M. Tessarotto, M. Ellero, Physica A 373 (2007) 142, arXiv: physics/0602140.
[9] M. Tessarotto, M. Ellero, in: M.S. Ivanov, A.K. Rebrov (Eds.), Proc. 25th RGD, International Symposium on Rarefied gas Dynamics, St. Petersburg, Russia,
July 21–28, 2006, Novosibirsk Publ. House of the Siberian Branch of the Russian Academy of Sciences, 2007, p. 1001, arXiv:physics/0611113.
[10] M. Tessarotto, C. Cremaschini, Strong solutions of the incompressible Navier–Stokes equations in external domains: Local existence and uniqueness,
arXiv:0809.5164v1 [math-ph], 2008.
[11] M.R. Maxey, J.J. Riley, Phys. Fluids 26 (1983) 883.
[12] C.M. Tchen, Ph.D. thesis, Delft, Martinus Nijhoff, The Hague, 1947.
[13] S. Corrsin, J. Lumley, Appl. Sci. Res. A 6 (1956) 114.
[14] Y.A. Buevich, Fluid Dynam. 1 (1966) 119.
[15] J.J. Riley, Ph.D. Thesis, The John Hopkins University, Baltimore, Maryland, 1971.
[16] S.B. Pope, Annu. Rev. Fluid Mech. 26 (1994) 23–63.
[17] S. Pratsinis, V. Srinivas, Powder Technol. 88 (1996) 267–273.
[18] P.A. Vaillancourt, M.K. Yau, B. Am. Meteorol. Soc. 8 (1) (2000) 285–298.
[19] J.C. Weil, R.I. Sykes, A. Venkatram, J. Appl. Meteorol. 31 (1992) 1121–1145.
[20] G. Falkovich, K. Gawedzki, M. Vergassola, Rev. Modern Phys. 73 (2001).
[21] B.I. Shraiman, E.D. Siggia, Nature 405 (2000) 639–646.
[22] K.R. Sreenivasan, J. Schumacher, Lagrangian views on turbulent mixing of passive scalars, arXiv:0904.0709v1 [physics.flu-dyn], 2009.
[23] C. Beck, PRL 98 (2007) 064502.
[24] L.F. Richardson, Proc. R. Soc. London A 110 (1926) 709.
[25] A.M. Basset, Treatise of Hydrodynamics, Deightob Bell, London 1888, vol. 2, pp. 285–297 (Chapter 22).
[26] J. Boussinesq, Theorie Analytique de la Chaleur, L’Ecole Polytecnique, Paris, 1903, Vol. 2 p. 224.
[27] C.W. Oseen, Hydrodynamik, Leipzig, 1927, p. 132.
[28] N. Gui, J. Fan, K. Cen, Phys. Rev. E 78 (2008) 046307.
[29] E.T. Jaynes, Phys. Rev. 106 (1957) 620.
[30] M. Tessarotto, IKT-approach to MHD turbulence, Magnetohydrodynamics Journal (2009) (in press), arXiv:0901.2454v1 [physics.flu-dyn] (2009).

You might also like