Geometric Algebra Approach to Fluid Dynamics

Carsten Cibura and Dietmar Hildenbrand

Abstract In this work we will use geometric algebra to prove a number of well
known theorems central to the field of fluid dynamics, such as Kelvin’s Circulation
Theorem and Helmholtz’ Theorem, showing that it is accessible by geometric alge-
bra methods and that these methods facilitate the representation of and calculation
with fluid dynamics concepts. Then we propose a generalization of the stream func-
tion to arbitrary dimensions, extending its explanatory power to higher-dimensional
flows. We will show how this extended stream function behaves on streamlines and
we will relate it to the curl of the flow’s vector field, i.e. its vorticity.

1 Introduction

While geometric algebra methods have been successfully applied to a variety of

problems in the recent past, including computer vision [10], robotics [3] [8] and
relativistic electro-magnetic field theory [1] [5], its applications to the field of fluid
dynamics have been restricted to isolated subjects or specialized problem descrip-
tions, e.g. [4] [6] [9]. The systematic development of a comprehensive fluid dynam-
ics calculus using geometric algebra methods is still an open task.
In this work we will use geometric algebra to prove a number of well known the-
orems central to the field of fluid dynamics, showing that some conclusions follow
neatly from geometric properties (e.g. of vector fields) encoded in the calculational
rules of the geometric product. Then we propose a generalization of the stream

Carsten Cibura
Technische Universität Darmstadt, currently at
Universiteit van Amsterdam, Faculteit der Natuurwetenschappen, Wiskunde en Informatica,
Kruislaan 404, 1098 SM Amsterdam, e-mail: c.cibura@uva.nl
Dietmar Hildenbrand
Technische Universität Darmstadt,
Fraunhoferstr. 5, 64283 Darmstadt, e-mail: dietmar.hildenbrand@gris.informatik.tu-darmstadt.de

1
2 Carsten Cibura and Dietmar Hildenbrand

function to arbitrary dimensions. Classically, the stream function is defined for two-
dimensional flows, where it can be used to calculate stream lines or the (complex)
potential of a flow. Generalizing the stream function to arbitrary dimensions offers
extended possibilities of analyzing a fluid flow.
We will proceed as follows. In section 2 we will give a short introduction into
what we consider advanced concepts of geometric algebra, like vector differentia-
tion and integration. Section 3 will introduce the reader to some pecularities of fluid
dynamics on a rather algebraic level, to an extent that is needed to understand this
work. For a more extended and figurative introduction, see [2]. In section 4 we use
geometric algebra to prove a number of theorems central to fluid dynamics, such as
Kelvin’s Circulation Theorem and Helmholtz’ Theorem. Also, we will show how to
define the stream function, which is recognized as useful for the description of 2D
flows, in arbitrary dimensions. Finally, in section 5, we will summarize our findings.

2 Geometric Algebra

We assume that the reader is familiar with the basics of geometric algebra, i.e. the
inner, outer and geometric product, the way these apply to vectors, how they extend
to higher grade objects, the role of the pseudoscalar etc.
In the following we will give a brief introduction into vector differentiation and
directed integration. For a comprehensive and more axiomatic introduction we refer
the reader to [7]. [5] gives an introduction that might be considered more accessible.

2.1 Vector Differentiation

The vector derivative ∇ unites the algebraic properties of a vector with those of an
operator. This becomes clear, if one expands it in terms of a basis. Given a coordinate
frame {ek } and its reciprocal frame {ek }, one can write

∂
∇ = ∑ ek ek · ∇ = ∑ ek , (1)
k k ∂ xk

which shows that ∇ inherits the properties of the partial derivatives – i.e. chain rule
and product rule apply – while at the same time it plays the role of a vector in any
product. The application of the vector derivative to scalar and vector-valued func-
tions or fields, respectively, may serve as an illustration of its calculational power.
Let p(x) be a scalar-valued field, i.e. a function depending on n-dimensional
position x. Then
∂ p(x)
∇p(x) = ∑ ek , (2)
k ∂ xk
is the vector-valued gradient of the scalar field p(x).
Geometric Algebra Approach to Fluid Dynamics 3

Things get more interesting when we apply ∇ to a vector-valued function, e.g.

the vector field u(x). The full vector derivative yields

∇u(x) = ∇ · u(x) + ∇ ∧ u(x) , (3)

a general multivector consisting of two terms of grade zero and two, respectively.
The scalar term ∇ · u(x), which is the divergence of u(x) and a pure bivector ∇ ∧
u(x), which is the vector field’s curl. Note that neither contains an abuse of notation,
but both seamlessly integrate into the geometric algebra calculus.
Moreover, it is possible to extend the vector derivative to general multivectors
and take the derivative ∇F(x) of any multivector-valued function F(x) with respect
to a vector argument.
As a vector the position of ∇ in any geometric algebra expression is not arbitrary,
since the geometric product is not commutative. As a consequence it no longer suf-
fices to determine the target of the operator ∇. The following rule of notation may
help to disambiguate things.
• In the absence of brackets, ∇ acts on the quantity to its immediate right.
• When ∇ is followed by brackets, it acts on all of the terms inside the brackets.
• When ∇ acts on a multivector, which is not on its right, the scope will be signified
by overdots.
• If ∇ has a subscript, then that indicates the variable with respect to which the
derivative has to be taken.
Consider the derivative of a nested function. Let u(v(x)) be such a function, with
u, v and x vectors. Then, by the chain rule

∂u i
∇x u = ∑ e
∂ xi
∂u ∂vj
= ∑ j i ei
∂v ∂x
=∇˙ x (v̇(x) · ∇
´ v )ú(v(x)) . (4)

With that it is possible to define the directional derivative of any multivector field
in the direction of a vector a in terms of the geometric product as
1 ˙ Ḟ(x) .


(a · ∇)F(x) = a∇F(x) + ∇a (5)
2

2.2 Directed Integration

Consider a curve C in n-dimensional space. Furthermore, assume that C has the

parametric representation x(s) with endpoints a = x(α) and b = x(β ). Even though
there exist multiple alternative definitions, for the purpose of this work let us define
the directed line integral in terms of the scalar integral by
4 Carsten Cibura and Dietmar Hildenbrand

dx
Z Z β
F(x)dx = F(x) ds . (6)
C α ds
The term
dx
ds = e(x)ds
dx = (7)
ds
is called directed line element. The argument x of e often is suppressed.
Three things are notable about this definition. First, the vector e(x) is the tangent
vector of the curve C in point x, spanning the tangent space there. Secondly, the
measure dx preserves a sense of direction, since it is defined as a vector-valued in-
finitesimal quantity. And finally, the product between the integrand and the measure
is to be read as a geometric product. That is, for example, for any vector field u(x)
the integral can be split into
Z Z Z
u(x)dx = u(x) · dx + u(x) ∧ dx . (8)
C C C

In analogy to (7) one determines the directed line elements for coordinate curves
in higher dimensions by
∂x
dxk = k dsk = ek dsk . (9)
∂s
From these the directed area element of a surface S = {x(s1 , s2 )} can be found
to be
d2 x = dx1 ∧ dx2 = e1 ∧ e2 ds1 ds2 , (10)
and the directed surface integral defined in terms of an iterated scalar integral by
Z Z Z β1 Z β2 (s1 )
Fd2 x = Fdx1 ∧ dx2 = ds1 ds2 Fe1 ∧ e2 . (11)
S S α1 α2 (s1 )

In n dimensions the grade n directed hyper-volume element dn x is denoted by dX

and the grade (n − 1) hyper-surface element dn−1 x by dS.
Because on a closed curve C, directed line elements occur pairwise with opposite
signs, it is intuitively clear that I
dx = 0 . (12)
C
One of the most powerful applications of this is a generalization of Stokes’ the-
orem, which relates differentiation and integration on manifolds.
Theorem 1 (Generalized Stokes’ Theorem). Let V be a vector manifold with
boundary ∂V , L(A) a multivector-valued function of a multivector argument A. Then
Z I
˙
L̇(∇dX) = L(dS) . (13)
V ∂V

We omit the proof here and refer the reader to [2] or [5]. Let us point out, though,
that the orientation of the manifold as well as its boundary is taken care of by the
definition of the directed measures dX and dS, respectively.
Geometric Algebra Approach to Fluid Dynamics 5

One important consequence of Stokes’ theorem is the fact that the path integral
along a closed curve in any gradient field is zero. In order to see this, consider the
function L(A) = f · A with f and A vectors in a two-dimensional manifold V and f the
gradient of some scalar-valued function w, i.e. f = ∇w = ∇ ∧ w. Then
I Z
f · dx = ˙ 2 x)
ḟ · (∇d
∂V V
Z
= ˙ · d2 x
(ḟ ∧ ∇)
V
Z
=− (∇ ∧ ∇ ∧ w) · d2 x = 0 . (14)
V

Here, d2 x has grade two, serving as a pseudoscalar for the tangent space of V . By
basic rewriting rules available in geometric algebra, it can be used to swap the inner
and outer product, which we did in line two, above.

3 Fluid Dynamics

Consider a fluid as a mass of particles moving in (n-dimensional) space. As a particle

is moved around by external forces and by bouncing off other particles, its position
x(x0 ,t) = (x1 (x01 , . . . , x0n ,t), . . . , xn (x01 , . . . , x0n ,t)) at a certain time t depends on its
initial position x0 = (x01 , . . . , x0n ).
An equivalent description of the fluid flow is to represent it as a vector field as it
exists at a given time t.

∂ x(x0 ,t)
u(x,t) = (15)
∂t
= (u1 (x,t), . . . , un (x,t)) , (16)

where we obtain (16) from (15) by solving for x0 .

However, the particle description has its use and we will let ϕt (x) = ϕ(x,t) de-
note the position of a particle – initially at position x – after time t. This implies
ϕ(x, 0) = x. ϕ is called the fluid flow map.
As a physical entity the flow not only obeys the rules that apply to vector fields
in general, but also has to yield to a number of physical principles.
The change of any component of u with time implicitly depends on all the other
components, as well as explicitly on time. Thus

d ˙ u̇ + ∂ u ,
u(x,t) = (u · ∇) (17)
dt ∂t
D ∂
as can be verified by applying the chain rule. We write Dt = (u · ∇) + ∂t , an operator
which is called the material derivative. It takes into acount that the fluid is moving
and the particle position changes with time.
6 Carsten Cibura and Dietmar Hildenbrand

The law of conservation of mass holds, stating that

∂ρ
+ ∇ · (ρu) = 0 , (18)
∂t
where ρ = ρ(x,t), the scalar-valued mass density of the fluid.
There is a position- and time-dependent scalar-valued function p(x,t), pressure,
which describes a force acting on any arbitrary surface inside the fluid, in the direc-
tion of the surface normal.
While pressure is a force acting locally, most real world fluids are subject to
global forces such as gravity, acting on the whole volume of fluid. These forces are
called body forces and denoted b.
The above properties hold for all fluid flows, but there is a number of restrictions
that can be placed on a flow to create interesting and important special cases.
An ideal fluid is a fluid in which no internal friction occurs. Therefore it does not
have viscosity. Note that in nature ideal fluids do not exist. However, they permit a
simplification of certain concepts that can be extended later on.
In an ideal fluid, the law of balance of momentum holds, stating that
Du
ρ = −∇p + ρb . (19)
Dt
A fluid flow is called incompressible, if its mass density is constant following the
flow, i.e. Dρ
Dt = 0. Equivalently, ∇ · u = 0 (see [2] for proof).
A fluid flow is called homogeneous, if its mass density is constant in space.
A fluid flow is called isentropic, if there is a scalar-valued function w(x,t), called
enthalpy, with
1
∇w = ∇p . (20)
ρ
Figuratively speaking, w is a measure of the heat content of a fluid. It consists of the
internal energy and pressure work. Internal energy is made up by molecular rotation
and oscillation, potential energy of chemical bonds etc. and roughly proportional to
the fluid’s temperature. Pressure work is the work that is done to establish the fluid’s
volume agains pressure’s influence trying to compress the fluid.

4 Application of Geometric Algebra to Fluid Dynamics

Definition 1 (Circulation). Given a closed contour C at time t0 = 0, denote with

Ct = ϕ(C,t) the contour transported by the flow at time t. Then the circulation
around Ct is I
ΓCt = u · dx . (21)
Ct
Geometric Algebra Approach to Fluid Dynamics 7

Theorem 2 (Circulation Theorem). In an isentropic flow not subject to any body

forces, ΓCt is constant in time.

Proof. Suppose x(s), 0 ≤ s ≤ 1 is a parametrization of C. Then ϕ(x(s),t), 0 ≤ s ≤ 1

is a paramtetrization of Ct . Recalling the form of the directed line element we write
Z 1
d d dx
Z
u · dx = u· ds
dt Ct dt 0 ds
Z 1
d d
= u(ϕt (x(s)),t) ·
ϕt (x(s))ds
0 dt ds
Z 1
d d
+ u(ϕt (x(s)),t) · ϕt (x(s))ds
0 dt ds
Z 1
Du(ϕt (x(s),t))
= · dx
0 Dt
Z 1
d
+ u(ϕt (x),t) · u(ϕt (x),t)ds . (22)
0 ds

But the final term is equal to 12 01 ds

R d  2 
u (ϕt (x),t) ds, because vectors commute. On
R d R
the other hand C ds f (s)ds = C df = 0, if C is closed. So the final term vanishes and
we arrive at
d Du
Z Z
u · dx = · dx . (23)
dt Ct Ct Dt

Finally, for isentropic flows ρ1 ∇p = ∇w. Balance of momentum states that ρ Du

Dt =
Du
−∇p + ρb. Excluding body forces (b = 0), one concludes Dt = −∇w and

d
Z
ΓC = − ∇w · dx = 0 , (24)
dt t Ct

because the path integral along a closed curve in a gradient field is zero, as was
shown in (14). tu

Definition 2 (Vortex Sheets, Lines and Tubes). A vortex sheet (resp. vortex line)
is a surface S (resp. a curve C), which – in each of its points – is orthogonal to the
vorticity plane ξ ≡ ∇ ∧ u of a fluid flow u.
A vortex tube consists of a surface, nowhere locally a vortex sheet, with vortex
lines drawn through each point of its bounding curve and extending as far as possible
in each direction.

Theorem 3 (Helmholtz’ Theorems). In an isentropic flow without body forces,

i). if a surface (a curve) moving with the flow is a vortex sheet (a vortex line) at time
t = 0, then it remains so for all time.
ii). if C1 and C2 are two closed curves encircling a common vortex tube, then
Z Z
u · dx = u · dx. (25)
C1 C2
 8 Carsten Cibura and Dietmar Hildenbrand

This common value is called the strength of the vortex tube.

iii). the strength of a vortex tube is constant in time as the tube moves with the flow.

Proof. Let S(x) be a vortex sheet parametrized by s1 and s2 . Then dS = ∂∂sS1 ds1 ∧
∂S
∂ s2
ds2 = e1 (x)∧e2 (x)ds1 ds2 is the directed area element of S at x. The orthogonality
property between S and ξ can be written as (∇ ∧ u(x)) · (e1 (x) ∧ e2 (x)) = 0, for all x.
Integrating over a vortex sheet and taking the derivative with respect to time yields
d d
Z I
(∇ ∧ u) · dS = u · dx = 0 , (26)
dt S dt ∂S

according to (13), (21) and (24), which proves i) for vortex sheets. The claim for
vortex lines is proven by the fact that every vortex line is the intersection of two
vortex sheets, locally.
Consider a vortex tube. Let S1 and S2 be two arbitrary cross sections of the tube
with bounding curves C1 and C2 , respectively. Let S denote the surface between C1
and C2 , i.e. the tube’s side. Then S1 , S2 and S enclose a region V and by (13)
Z
0= (∇ ∧ (∇ ∧ u)) · dX
IV
= ξ · dS
S1 ∪S2 ∪S
I I I
= ξ · dS + ξ · dS + ξ · dS
S1 S2 S
I I
= ξ · dS + ξ · dS
S1 S2
Z Z
= u · dx + u · dx , (27)
C1 C2

because S is a vortex sheet. And since S1 and S2 are oriented oppositely, with sur-
face normals pointing “outward”, this proves ii). iii) follows from the Circulation
Theorem. t u

After having shown that geometric algebra can simplify the representation and
proof of a number of fluid flow properties, we now present the stream function ψ in
mulitple dimensions as a graded function, with the grade varying with the dimension
of the respective flow. This is possible because of the fact that geometric algebra is
a graded algebra and that the grade of an algebraic object can be seen as a variable
on a scale, rather than as an immutable, constituting property.
In two dimensions, this function is uniquely determined up to an additive con-
stant. The behaviour of the stream function on streamlines allows for an adjustment
of this additive constant, which, in turn, determines the stream function and through
it the two-dimensional flow. The lack of such a simple determination is one of the
main problems in describing a three-dimensional flow. Finding and exploring consti-
tuting properties of the multivector-valued stream function ψ in higher dimensions
would greatly facilitate the understanding of general fluid flows.
Geometric Algebra Approach to Fluid Dynamics 9

Employing geometric algebra methods it is easy to define a stream function for

n-dimensional, incompressible flows. Assume such a flow with velocity field u, con-
tained in some region D that is simply connected. There exists a function ψ(x,t) of
(pure) grade n − 2, which fulfills

∇ ∧ ψ = uIn−1 , (28)

where In denotes the unit pseudoscalar in n dimensions. This can be seen by

0 = ∇ ∧ (∇ ∧ ψ)
= ∇ ∧ (uIn−1 )
= (∇ · u)In−1 , (29)

which is zero if and only if (∇ · u) is zero, i.e. the flow is incompressible.

Note that (28) does not suffice to determine ψ. For example, adding the curl of
any grade (n − 3) function g(x) to ψ, such that ψ 0 = ψ + ∇ ∧ g, does not change u.

u0 = (∇ ∧ ψ 0 )In
= (∇ ∧ ψ + ∇ ∧ ∇ ∧ g)In
= (∇ ∧ ψ)In
= u. (30)

But, if we fix the divergence of ψ, i.e. ∇ · ψ = m, then

∇ · ψ 0 = ∇ · (ψ + ∇ ∧ g)
= m + ∇ · (∇ ∧ g) 6= m . (31)

Specifying the divergence of ψ is called choosing a gauge. Together with (28),

choosing a gauge determines ψ. Note that in n = 2 dimensions, ψ is a scalar quantity
and therefore ∇ · ψ is zero.
A relationship between vorticity ξ and the stream function ψ can be established
as follows.

ξ = ∇∧u
= ∇ ∧ ((∇ ∧ ψ)In )
= [∇ ∧ ((∇ ∧ ψ)In )] In In−1
= [∇ · ((∇ ∧ ψ)In In )] In−1
1
= (−1) 2 n(n−1) (∇ · (∇ ∧ ψ))In−1 . (32)

Of special interest is the behaviour of ψ on streamlines.

Definition 3 (Streamlines). At a fixed time, a streamline is an integral curve of the
velocity field u(x,t) of a given fluid flow. If x(s) is a streamline parametrized by s,
then x(s) satisfies
10 Carsten Cibura and Dietmar Hildenbrand

dx
= u(x(s),t), t fixed . (33)
ds
Let x(s) be a streamline parametrized by s. Then by (4)

d
ψ(x(s),t) = ∇s ψ(x(s),t)
ds
=∇˙ s (ẋ(s) · ∇
´ x )ψ́(x(s),t)
 
= u(x(s),t) · ∇ ´ ψ́(x(s),t)

´ ψ́.
= (u · ∇) (34)

5 Conclusion

In the present work we have shown that the highly complex field of fluid dynamics is
accessible by the methods of geometric algebra, which facilitates the understanding
of many concepts by – literally – adding new dimensions of consideration.
Also, we have proposed a definition for a stream function in arbitrary-dimensional
fluid flows as a graded function. It can be seen to be identical to the known stream
function in the special case of two-dimensional flows. We showed how this stream
function relates to the geometric algebra formulation of the flow field’s curl, i.e. the
fluid flow’s vorticity.

References

1. Baylis, W. E.: Electrodynamics – A Modern Geometric Approach. Birkhäuser Boston (2004)

2. Cibura, C.: Derivation of Fluid Dynamics Basics Using Geometric Algebra. Diplomarbeit,
Technische Universität Darmstadt (2007)
3. Sommer, G.: Geometric Computing with Clifford Algebras – Theoretical Foundations and
Applications in Computer Vision and Robotics. Springer-Verlag (2001)
4. Sprößig, W.: Fluid Flow Equations with Variable Viscosity in Quaternionic Setting. In: Ad-
vances in Applied Clifford Algebras, volume 17. Birkhäuser (2007).
5. Doran, C., Lasenby, A.: Geometric Algebra for Physicists. Cambridge University Press, Cam-
bridge (1984)
6. Gürlebeck, K., Sprößig, W.: Quaternionic Analysis and Elliptic Boundary Value Problems.
Birkhäuser (1990)
7. Hestenes, D., Sobczyk, G.: Clifford Algebra to Geometric Calculus. Kluwer, Dordrecht
(1984)
8. Hildenbrand, D., Zamora, J., Bayro-Corrochano, E.: Inverse Kinematics Computation in
Computer Graphics and Robotics Using Conformal Geometric Algebra. (2005)
9. Scheuermann, G., Hagen,H.: Clifford Algebra and Flows. In: T. Lyche and L. L. Schumaker
(eds.), Mathematical Methods for Curves and Surfaces, pp. 173181. Vanderbilt University
Press (2001)
10. Wareham, R., Cameron, J., Lasenby, J.: Applications of Conformal Geometric Algebra in
Computer Vision and Graphics 2005. In: Computer Algebra and Geometric Algebra with
Applications, volume 3519 of Lecture Notes in Computer Science. Springer Verlag (2004)