Elements of Hydrodynamics: Nicolas Borghini
Elements of Hydrodynamics: Nicolas Borghini
Elements of Hydrodynamics: Nicolas Borghini
Nicolas Borghini
The following pages were originally not designed to fall under your eyes. They grew up from
handwritten notes for myself, listing the important points which I should not forget in the lecture
room. As time went by, more and more remarks or developments were added, which is why I
started to replace the growingly dirty sheets of paper by an electronic version—that could then also
be easily uploaded on the web page of my lecture, for the benefit(?) of the students.
Again, additional results, calculations, comments, paragraphs or even whole chapters accumu-
lated, leading to the temporary outcome which you are reading now: a not necessarily optimal
overall outline; at times, unfinished sentences; not fully detailed proofs or calculations—because the
missing steps are obvious to me—; insufficient discussions of the physics of some results—which
I hopefully provide in the classroom—; not-so-good-looking figures; incomplete bibliography; etc.
You may also expect a few solecisms, inconsistent notations, and the usual, unavoidable typos.(∗)
Eventually, you will have to cope with the many idiosyncrasies in my writing, as for instance my
immoderate use of footnotes, dashes or parentheses, quotation marks, which are not considered as
“good practice”.
In short: the following chapters may barely be called “lecture notes”; they cannot replace a
textbook(†) and the active participation in a course and in the corresponding tutorial/exercise
sessions.
(∗)
Comments and corrections are welcome!
(†)
... which is one of several good reasons why you should think at least twice before printing a hard copy!
Contents
Introduction• • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • 1
Appendix to Chapter II • • • • • • • • • • • • • • • • • • • • • • • • • • • • 24
II.A Deformations in a continuous medium 24
Appendices • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • 159
Bibliography • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • 173
viii
Introduction
General references
(in alphabetical order)
• Landau & Lifshitz, Course of theoretical physics. Vol. 6: Fluid mechanics [3]
= Landau & Lifschitz, Lehrbuch der theoretischen Physik. Band VI: Hydrodynamik [4]
• Sommerfeld, Lectures on theoretical physics. Vol. II: Mechanics of deformable bodies [5]
= Vorlesungen über theoretische Physik. Band II: Mechanik der deformierbaren Medien [6]
C HAPTER I
A system of many microscopic degrees of freedom is often more conveniently described as a material
body that fills some region of space continuously, rather than as a collection of discrete points
(Sec. I.1). This theoretical approach, which is especially suited to represent systems whose internal
deformations are relevant, is an instance of physical modeling, originally motivated by the agreement
of its predictions with experimental observations. Like every model, that of a continuous medium
is valid only in some range of physical conditions, in particular on macroscopic scales.
Mathematically, a classical continuous medium at a given instant is described as a volume—or
more generally a manifold—in usual Euclidean space. The infinitesimal elements of this volume
constitute the elementary “material points”, which are entirely characterized by their position.
To describe the time evolution of the physical system modeled as a continuous medium, two
equivalent approaches are available. The first one consists in following the trajectories of the material
points as time progresses (Sec. I.2). The physical picture of continuousness is then enforced by
requesting that the mapping between the position of a given point at some reference initial time
and its position at any later instant is continuous.
The second point of view, which will mostly be adopted in the remainder of these notes, focuses
on the change in the various physical quantities at a fixed position as time elapses (Sec. I.3).
The reference for the medium evolution between successive instants t and t + dt is the “current”
configuration of the material points, i.e. at time t, instead of their positions in the (far) past. In that
description, the spatial variables are no longer dynamical, but only labels for the position at which
some observable is considered. Accordingly, the dynamical quantities in the system are now time-
dependent fields; the desired continuousness of the medium translates into continuity conditions on
those fields.
Eventually, the mathematical object that models internal forces in a continuous medium, i.e.
the influence from neighboring material points on each other, is shortly introduced (Sec. I.4). This
allows the classification of deformable continuous media into two traditional large classes, and in
particular the definition of fluids.
2.7 × 1025 m−3 , so that one cubic millimeter still contains 2.7 × 1016 atoms. Similarly, even though
the number density in the interstellar medium might be as low as about 102 m−3 , any volume
relevant for astrophysics, i.e. with at least a kilometer-long linear size, involves a large number of
atoms.
Additionally, these atoms are in constant chaotic motion, with individual velocities of order
102 –103 m·s−1 for a system at thermal equilibrium at temperature T ' 300 K. Given a mean free
path(i) of order 10−7 m in a gas under normal conditions, each atom undergoes 109 –1010 times per
second, i.e. its trajectory changes direction constantly when viewed with a macroscopic viewpoint.
As is seen in Statistical Mechanics, it is in general unnecessary to know the details of the motion
of each atom in a macroscopic system: as a matter of fact, there emerge global characteristics,
defined as averages, which can be predicted to a high degree of accuracy thanks to the large number
of degrees of freedom involved in their determination, despite the chaoticity of the individual atomic
behaviors. The macroscopic properties of systems at (global) thermodynamic equilibrium are thus
entirely determined by a handful of collective variables, either extensive—like entropy, internal
energy, volume, particle number, total momentum. . . —, or intensive—as e.g. the respective densities
of the various extensive variables, temperature, pressure, chemical potential, average velocity. . . —,
where the latter take the same value throughout the system.
When thermodynamic equilibrium does not hold globally in a system, there is still the possi-
bility that one may consider that it is valid locally, “at each point” in space. In that situation—
whose underlying assumptions will be specified in greater detail in Sec. I.1.3—the intensive thermo-
dynamic variables characterizing the system macroscopically become fields, which can vary from
point to point. More generally, experience shows that it is fruitful to describe a large amount of
characteristics—not only thermodynamic, but also of mechanical nature, like forces and the dis-
placements or deformations they induce—of macroscopic bodies as fields. A “continuous medium”
is then intuitively a system described by such fields, which should satisfy some (mathematical)
continuity property with respect to the spatial variables that parameterize the representation of the
physical system as a geometrical quantity, as will be better specified in Secs. I.2 and I.3.
Assuming the relevance of the model of a medium whose properties are described by continuous
fields is often referred to as continuum hypothesis (ii) .
The reader should keep in mind that the modeling of a given macroscopic system as a continuous
medium does not invalidate the existence of its underlying discrete atomic structure. Specific
phenomena will still directly probe the latter, as e.g. X-ray scattering experiments. That is, the
model has limitations to its validity, especially at small wavelengths or high frequencies, where
“small” or “high” implies a comparison to some microscopic physical scale characteristic of the
system under consideration. Turning the reasoning the other way around, the continuous-medium
picture is often referred to as a long-wavelength, low-frequency approximation to a more microscopic
description—from which it can actually be shown to emerge in the corresponding limits.
It is important to realize that the model itself is blind to its own limitations, i.e. there is
no a priori criterion within the mathematical continuous-medium description that signals the
breakdown of the relevance of the picture to actual physics. In practice, there might be hints
that the equations of the continuous model are being applied in a regime where they should not,
as for instance if they yield negative values for a quantity which should be positive, but such
occurrences are not the general rule.
Remarks:
∗ The model of a continuous model is not only applicable—and applied—to obvious cases like
gases, liquids or (deformable) solids, it may also be used to describe the behaviors of large crowds,
fish schools, car traffic. . . provided the number of “elementary” constituents is large and the system
is studied on a large enough scale.
(i) (ii)
mittlere freie Weglänge Kontinuumshypothese
4 Basic notions on continuous media
∗ Even if the continuous description is valid on “long wavelengths”, it remains obvious that any
physical system, viewed on a scale much larger than its spatial extent, is to first approximation best
described as pointlike.
Consider for instance a molecular cloud of interstellar medium with a 10 parsec radius and about
1010 H2 molecules per cubic meter. For a star forming at its core, it behaves a continuous medium;
1 kpc away, however, the inner degrees of the cloud are most likely already irrelevant and it is best
described as a mere point.
Remark: The volume V —with the topology inherited fromE 3 —need not be simply connected. For
instance, one may want to describe the flow of a river around a bridge pier: the latter represents a
physical region which water cannot penetrate, which is modeled as a hole throughout the volume
V occupied by fluid particles.
To characterize the position of a given material point, as well as some of the observables relative
to the physical system Σ, one still needs to specify the reference frame in which the system is
studied, corresponding to the point of view of a given observer, and to choose a coordinate system
in that reference frame. This choice allows one to define vectors—like position vectors, velocities,
or forces—and tensors.
The basis vectors of the coordinate system will be designated as ~e1 , ~e2 , ~e3 , while the components
of a given vector will be denoted with upper (“contravariant”) indices, as e.g. ~c = ci ~ei , where the
Einstein summation convention over repeated upper and lower indices was used.
Once the reference frame and coordinate system are determined, the macroscopic state of the
physical system at time t is mapped onto a corresponding configuration (vi) κt of the medium,
consisting of the continuous set of the position vectors ~r = xi ~ei of its constituting material points.
Since the volume occupied by the latter may also depends on time, it will also be labeled by t: V t .
To be able to formalize the necessary continuity conditions in the following Sections, one also
introduces a reference time t0 —conveniently taken as the origin of the time axis, t0 = 0—and the
corresponding reference configuration κ0 of the medium, which occupies a volume V 0 . The generic
position vector of a material point in this reference configuration will be denoted as R ~ = X i ~ei .
Remark: In so-called “classical” continuous media, as have been introduced here, the material points
are entirely characterized by their position vector. In particular, they have no intrinsic angular
momentum.
(iii) (iv) (v) (vi)
Materielle Punkte Mediumteilchen Fluidteilchen Konfiguration
I.1 Continuous medium: a model for many-body systems 5
(i) each individual cell can meaningfully be treated as a thermodynamic system, i.e. it must be
large enough that the relative fluctuation of the usual extensive thermodynamic quantities
computed for the content of the cell are negligible;
(ii) the thermodynamic properties vary little over the cell scale, i.e. cells cannot be too large, so
that (approximate) homogeneity is ensured.
The rationale behind these two requirements is illustrated by Fig. I.1, which represents schematically
how the value of a local macroscopic quantity, e.g. a density, depends on the resolution of the
apparatus with which it is measured, i.e. equivalently on the length scale on which it is defined. If the
apparatus probes too small a length scale, so that the discrete degrees of freedom become relevant,
the measured value strongly fluctuates from one observation to the next one, as hinted at by the
displayed envelope of possible results of measurements: this is the issue addressed by condition (i).
Simultaneously, a small change in the measurement resolution, even with the apparatus still centered
on the same point in the system, can lead to a large variation in the measured value of the observable,
corresponding to the erratic behavior of the curve at small scales shown in Fig. I.1. This fluctuating
pattern decreases with increasing size of the observation scale, since this increase leads to a growth
in the number of atoms inside the probed volume, and thus a drop in the size of relative fluctuations.
At the other end of the curve, one reaches a regime where the low resolution of the observation
leads to encompassing domains with enough atoms to be rid of fluctuations, yet with inhomogeneous
macroscopic properties, in a single probed region—in violation of condition (ii). As a result, the
measured value of the density under consideration slowly evolves with the observation scale.
In between these two domains of strong statistical fluctuations and slow macroscopic variations
lies a regime where the value measured for an observable barely depends on the scale over which it is
measured local quantity
observation scale
Figure I.1 – Typical variation of the measured value for a “local” macroscopic observable as
a function of the size scale over which it is determined.
6 Basic notions on continuous media
determined. This represents the appropriate regime for meaningfully defining—and measuring—a
local density, and more general local quantities.
It is important to note that this intermediate “mesoscopic” interval may not always exist. There
are physical systems in which strong macroscopic variations are already present in a range of scales
where microscopic fluctuations are still sizable. For such systems, one cannot find scale-independent
local variables. That is, the proper definition of local quantities implicitly relies on the existence
of a clear separation of scales in the physical system under consideration, which is what will be
assumed in the remainder of these notes.
Remark: The smallest volume over which meaningful local quantities can be defined is sometimes
called representative volume element (vii) (RVE), or representative elementary volume.
When conditions (i) and (ii) hold, one may in particular define local thermodynamic variables,
corresponding to the values taken in each intermediate-size cell—labeled by its position ~r—by the
usual extensive parameters: internal energy, number of atoms. . . Since the separation between
cells is immaterial, nothing prevents energy or matter from being transported from a cell to its
neighbors, even if the global system is isolated. Accordingly, the local extensive variables in any
given cell are actually time-dependent in the general case. In addition, it becomes important to
add linear momentum—with respect to some reference frame—to the set of local extensive variables
characterizing the content of a cell.
The size of each cell is physically irrelevant, as long as it satisfies the two key requirements; there
is thus no meaningful local variable corresponding to volume. Similarly, the values of the extensive
variables in a given cell, which are by definition proportional to the cell size, are as arbitrary as
the latter. They are thus conveniently replaced by the respective local densities: internal energy
density e(t,~r), number density n (t,~r), linear momentum density ρ(t,~r)~v(t,~r), where ρ denotes the
mass density, entropy density s(t,~r). . .
Remark: Rather than considering the densities of extensive quantities, some authors—in particular
Landau & Lifshitz [3, 4]—prefer to work with specific quantities, i.e. their respective amounts per
unit mass, instead of per unit volume. The relation between densities and specific quantities is
trivial: denoting by x j resp. x j,m a generic local density resp. specific amount for the same physical
quantity, one has the identity
x j (t,~r) = ρ(t,~r) x j,m (t,~r) (I.1)
in every cell—labeled by ~r—and at every time t.
Once the local extensive variables have been meaningfully defined, one can develop the usual
formalism of thermodynamics in each cell. In particular, one introduces the conjugate intensive
variables, as e.g. local temperature T (t,~r) and pressure P (t,~r). The underlying, important hypoth-
esis is the assumption of a local thermodynamic equilibrium. According to the latter, the equation(s)
of state of the system inside the small cell, expressed with local thermodynamic quantities, is the
same as for a macroscopic system in the actual thermodynamic limit of infinitely large volume and
particle number.
Consider for instance a non-relativistic classical ideal gas: its (mechanical) equation of state
reads PV = N kB T , with N the number of atoms, which occupy a volume V at uniform pressure P
and temperature T , while kB is the Boltzmann constant. This is trivially recast as P = n kB T , with
n the number density of atoms. The local thermodynamic equilibrium assumption then states that
under non-uniform conditions of temperature and pressure, the equation of state in a local cell at
position ~r is given by
P (t,~r) = n (t,~r)kB T (t,~r) (I.2)
at every time t.
(vii)
Repräsentatives Volumen-Element
I.2 Lagrangian description 7
The last step towards the continuous-medium model is to promote ~r, which till now was simply
the discrete label attached to a given cell, to be a continuous variable taking its values in R3 —
or rather, in the volume V t attached to the system at the corresponding instant t. Accordingly,
taking into account the time-dependence of physical quantities, the local variables, in particular the
thermodynamic parameters, become fields on R × R3 .
The replacement of the fine-resolution description, in which atoms are the relevant degrees of
freedom, by the lower-resolution model which assimilates small finite volumes of the former to
structureless points is called coarse graining (viii) .
This is a quite generic procedure in theoretical physics, whereby the finer degrees of freedom of a
more fundamental description are smoothed away—technically, this is often done by performing
averages or integrals, so that these degrees of freedom are “integrated out”—and replaced by
novel, effective variables in a theory with a more limited range of applicability, but which is
more tractable for “long-range” phenomena.
Coming back to condition (ii), we already stated that it implicitly involves the existence of at
least one large length scale L, over which the macroscopic physical properties of the system may
vary. This scale can be a characteristic dimension of the system under consideration, as e.g. the
diameter of the tube in which a liquid is flowing. In the case of periodic waves propagating in
the continuous medium, L also corresponds to their wavelength. More generally, if G denotes a
macroscopic physical quantity, one may consider
" #−1
∇
~ G (t,~r)
L∼= , (I.3)
|G (t,~r)|
where ∇~ denotes the (spatial) gradient.
Condition (i) in particular implies that the typical size of the cells which are later coarse grained
should be significantly larger than the mean free path `mfp of atoms, so that thermodynamic equi-
librium holds in the local cells. Since on the other hand this same typical size should be significantly
smaller than the scale L of macroscopic variations, one deduces the condition
`mfp
Kn ≡ 1 (I.4)
L
Since the variable R~ is independent of t, one could actually also write ~v (t, R)
~ = d~r(t, R)/dt.
~
In turn, the acceleration of the material point in R is given at time t by
~
~ = ∂~v (t, R) .
~a(t, R) (I.9)
∂t
Remark: The trajectory (or pathline (ix) ) of a material point can be visualized, by tagging the point
~ for instance with a fluorescent or radioactive marker, and then imaging
at time t0 at its position R,
the positions at later times t > t0 .
If on the other hand one regularly—say for every instant t0 ≤ t0 ≤ t—injects some marker at a
fixed geometrical point P , the resulting tagged curve at time t is the locus of the geometrical points
occupied by medium particles which passed through P in the past. This locus is referred to as
streakline.(x) Denoting by ~rP the position vector of point P , the streakline is the set of geometrical
points with position vectors
~ 0 ,~rP ) for t0 ≤ t0 ≤ t.
~r = ~r t, R(t (I.10)
The more usual description is the so-called Eulerian (c) perspective, in which the evolution between
instants t and t + dt takes the system configuration at time t as a reference.
variables take their values differ: constant (V 0 ) in the Lagrangian specification, time-dependent (V t )
in the case of the Eulerian quantities. Accordingly the latter will be denoted with a subscript t.
~ r) .
~vt (t,~r) = ~v t, R(t,~ (I.13a)
This identity in particular shows that ~vt automatically inherits the smoothness properties of ~v : if
~ and its inverse are piecewise C 2 (cf. Sec. I.2.2), then ~vt is (at least)
~ 7→ ~r(t, R)
the mapping (t, R)
piecewise C in both its variables.
1
For a generic physical quantity, the transition from the Lagrangian to the Eulerian point of view
similarly reads
~ r) .
Gt (t,~r) = G t, R(t,~ (I.13b)
Reciprocally, given a (well-enough behaved) Eulerian velocity field ~vt on a continuous medium,
one can uniquely obtain the Lagrangian description of the medium motion by solving the initial
value problem
~
∂~r(t, R) =~v t,~r(t, R)
~
t
∂t (I.14a)
~ ~
~r(t0 , R) = R,
where the second line represents the initial condition. That is, one actually reconstructs the pathline
of every material point of the continuous medium. Introducing differential notations, the above
system can also be rewritten as
~ = R.
d~r =~vt (t,~r) dt with ~r(t0 , R) ~ (I.14b)
I.3.3 Streamlines
At a given time t, the streamlines (xi) of the motion are defined as the field lines of ~vt . That is,
these are curves whose tangent is everywhere parallel to the instantaneous velocity field at the same
geometrical point.
Let ~x(λ) denote a streamline, parameterized by λ. The definition can be formulated as
d~x(λ)
= α(λ)~v t, ~x(λ) (I.15a)
dλ
with α(λ) a scalar function. Equivalently, denoting by d~x(λ) a differential line element tangent to
(xi)
Stromlinien
I.3 Eulerian description 11
• If~v(t,~r) only depends on the position, not on time, then the velocity field and the correspond-
ing motion of the continuous medium are said to be steady or equivalently stationary.
In that case, the streamlines coincide with the pathlines and the streaklines.
Indeed, one checks that Eq. (I.14b) for the pathlines, in which the velocity becomes time-
independent, can then be recast (in a point where all vi are non-zero) as
dx1 dx2 dx3
= = ,
v1 (t,~r) v2 (t,~r) v3 (t,~r)
where the variable t plays no role: this is exactly the system (I.15c) defining the streamlines
at time t. The equivalence between pathlines and streaklines is also trivial.
(xii)
Stromröhre
12 Basic notions on continuous media
More generally, one finds by repeating the same derivation that the time derivative of a physical
quantity G attached to a material point or domain, yet expressed in terms of Eulerian fields, is the
~ G ] part, irrespective of the tensorial nature of G .
sum of a local (∂ G /∂t) and a convective [(~v · ∇)
Accordingly, one introduces the operator
D ∂ ~
≡ +~v(t,~r) · ∇ (I.18)
Dt ∂t
called material derivative (xiii) or (between others) substantial derivative,(xiv) derivative following the
motion, hydrodynamic derivative. Relation (I.17) can thus be recast as
D~v(t,~r)
~a(t) = . (I.19)
Dt
Remarks:
∗ Equation (I.17) shows that even in the case of a steady motion, the acceleration of a material
point may be non-vanishing, thanks to the convective part.
∗ The material derivative (I.18) is also often denoted (and referred to) as total derivative d/dt.
∗ One also finds in the literature the denomination convective derivative.(xv) To the eyes and ears
of the author of these lines, that name has the drawback that it does not naturally evoke the local
part, but only. . . the convective one, which comes from the fact that matter is being transported,
“conveyed”, with a non-vanishing velocity field ~v(t,~r).
∗ The two terms in Eq. (I.18) actually “merge” together when considering the motion of a material
point in Galilean space-time R × R3 . As a matter of fact, one easily shows that D/Dt is the (Lie)
derivative along the world-line of the material point
The world-line element corresponding to the motion between t and t+dt goes from (t, x1 , x2 , x3 ) to
(t+dt, x1 +v1 dt, x2 +v2 dt, x3 +v3 dt). The tangent vector to this world-line thus has components
(1, v1 , v2 , v3 ), i.e. the derivative along the direction of this vector is ∂t + v1 ∂1 + v2 ∂2 + v3 ∂3 , with
the usual shorthand notation ∂i ≡ ∂/∂xi .
• Volume or body forces,(xvi) which act in each point of the bulk volume of V.
Examples are weight, long-range electromagnetic forces or, in non-inertial reference frames,
fictitious forces (Coriolis, centrifugal).
For such forces, which tend to be proportional to the volume they act on, it will later be more
convenient to introduce the corresponding volumic force density.
• Surface or contact forces,(xvii) which act on the surface S, like friction, which we now discuss
in further detail.
d2 S ~en
d2 F~s
Figure I.2
Consider an infinitesimally small geometrical surface element d2 S at point P . Let d2 F~s denote
the surface force through d2 S. That is, d2 F~s is the contact force, due to the medium exterior to V,
that a “test” material surface coinciding with d2 S would experience. The vector
d2 F~s
T~s ≡ 2 , (I.20)
d S
representing the surface density of contact forces, is called (mechanical) stress vector (xviii) on d2 S.
The corresponding unit in the SI system is the Pascal, with 1 Pa = 1 N·m−2 .
Purely geometrically, the stress vector T~s on a given surface element d2 S at a given point can
be decomposed into two components, namely
• a vector orthogonal to plane tangent in P to d2 S, the so-called normal stress (xix) ; when it
is directed towards the interior resp. exterior of the medium domain being acted on, it also
referred to as compression (xx) resp. tension (xxi) ;
• a vector in the tangent plane in P , called shear stress (xxii) and often denoted as ~τ .
Despite the short notation adopted in Eq. (I.20), the stress vector depends not only on the
position of the geometrical point P where the infinitesimal surface element d2 S lies, but also on the
(xvi) (xvii) (xviii) (xix)
Volumenkräfte Oberflächenkräfte Mechanischer Spannungsvektor Normalspannung
(xx) (xxi) (xxii)
Druckspannung Zugsspannung Scher-, Tangential- oder Schubspannung
14 Basic notions on continuous media
orientation of the surface. Let ~en denote the normal unit vector to the surface element, directed
towards the exterior of the volume V (cf. Fig. I.2), and let ~r denote the position vector of P in a
given reference frame. The relation between ~en and the stress vector T~s on d2 S is then linear:
T~s = σ(~r) · ~en , (I.21a)
with σ (~r) a symmetric tensor of rank 2, the so-called (Cauchy(d) ) stress tensor .(xxiii)
In a given coordinate system, relation (I.21a) yields
3
X
Tsi = σ ij ejn (I.21b)
j=1
with Tsi resp. ejn the coordinates of the vectors T~s resp. ~en , and σ ij the
1 -components of the stress
1
tensor.
While valid in the case of a three-dimensional position space, equation (I.21a) should actually
be better formulated to become valid in arbitrary dimension. Thus, the unit-length “normal
vector” to a surface element at point P is rather a 1-form acting on the vectors of the tangent
space to the surface at P . As such, it should be represented as the transposed of a vector [(~en )T ],
which multiplies the stress tensor from the left:
Remark: The symmetry property of the Cauchy stress tensor is intimately linked to the assumption
that the material points constituting the continuous medium have no intrinsic angular momentum.
I.4.2 Fluids
With the help of the notion of mechanical stress, we may now introduce the definition of a fluid ,
which is the class of continuous media whose motion is described by hydrodynamics:
A fluid is a continuous medium that deforms itself as long as it is submitted to shear stresses.
(I.22)
Turning this definition around, one sees that in a fluid at rest—or, to be more accurate, studied
in a reference frame with respect to which it is at rest—the mechanical stresses are necessarily
normal. That is, the stress tensor is in each point diagonal.
More precisely, for a locally isotropic fluid—which means
that the material points are isotropic,
2
which is the case throughout these notes—the stress 0 -tensor is everywhere proportional to the
inverse metric tensor:
σ(t,~r) = −P (t,~r) g−1 (t,~r) (I.23)
identity matrix, i.e. σ ij ∝ δ ij , where we have introduced the Kronecker symbol. To obtain the
2 -components σ ik , one has to multiply σ i by the component g jk of the inverse metric tensor,
0 j
summing over k, which precisely gives Eq. (I.23).
Remarks:
∗ Definition (I.22), as well as the two remarks hereafter, rely on an intuitive picture of “deforma-
tions” in a continuous medium. To support this picture with some mathematical background, we
shall introduce in Sec. II.A an appropriate strain tensor, which quantifies these deformations, at
least as long as they remain small.
∗ A deformable solid will also deform itself when submitted to shear stress! However, for a given
fixed amount of tangential stress, the solid will after some time reach a new, deformed equilibrium
position—otherwise, it is not a solid, but a fluid.
∗ The previous remark is actually a simplification, valid on the typical time scale of human beings. Thus,
materials which in our everyday experience are solids—as for instance those forming the mantle of the Earth—
will behave on a longer time scale as fluids—in the previous example, on geological time scales. Whether
a given substance behaves as a fluid or a deformable solid is sometimes characterized by the dimensionless
Deborah number [7], which compares the typical time scale for the response of the substance to a mechanical
stress and the observation time.
∗ Even nicer, the fluid vs. deformable solid behavior may actually depend on the intensity of the
applied shear stress: ketchup!
(1)
The visualization techniques have probably evolved since the 1960s, yet pathlines, streaklines or streamlines are
still defined in the same way.
C HAPTER II
The goal of fluid dynamics is to investigate the motion of fluids under consideration of the forces
at play, as well as to study the mechanical stresses exerted by moving fluids on bodies with which
they are in contact. The description of the motion itself, irrespective of the forces, is the object of
kinematics.
The possibilities for the motion of a deformable continuous medium, in particular of a fluid, are
richer than for a mere point particle or a rigid body: besides translations and global rotations, a
deformable medium may also rotate locally and undergo. . . deformations! The latter term actually
encompasses two different yet non-exclusive possibilities, namely either a change of shape or a
variation of the volume. All these various types of motion are encoded in the local properties of
the velocity field at each instant (Sec. II.1). Generic fluid motions are then classified according to
several criteria, especially taking into account kinematics (Sec. II.2).
For the sake of reference, the characterization of deformations themselves, complementing that
of their rate of change, is briefly presented in Sec. II.A. That formalism is not needed within fluid
dynamics, but rather for the study of deformable solids, like elastic ones.
x2
~v t,~r + δ~`(t)
δ~`(t + δt)
δ~`(t)
~v(t,~r)
~r
x1
x3
Figure II.1 – Positions of a material line element δ~` at successive times t and t + δt.
II.1 Generic motion of a continuous medium 17
of material points distributed along the (straight) line element stretching between two neighboring
geometrical points. positions. Let ~r and ~r + δ~`(t) denote the geometrical endpoints of this material
vector at time t.
Thanks to the continuity of the mappings R ~ 7→ ~r(t,~r) and its inverse ~r 7→ R(t,~
~ r), the material
vector defined at instant t remains a connected set of material points as time evolves, in particular
at t + δt. Assuming that both the initial length |δ~`(t)| as well as δt are small enough, the evolved
set at t + δt remains approximately along a straight line, and constitutes a new material vector,
denoted by δ~`(t+dt). The position vectors of these endpoints simply follow from the initial positions
of the corresponding material points: ~r resp. ~r + δ~`(t), to which should be added the respective
displacement vectors between t and t+δt, namely the product by δt of the initial velocity ~v t,~r
resp. ~v t,~r + δ~`(t) . That is, one finds
Figure II.1 already suggests that the motion of the material vector consists not only of a translation,
but also of a rotation, as well as an “expansion”—the change in length of the vector.
~ vv(t,~r) whose components in the coordinate system used here are the
Introducing the 11 -tensor ∇~
partial derivatives ∂ v (t,~r)/∂xj , the above relation can be recast in the coordinate-independent
i
form
~ vv(t,~r) · δ~r.
δ~v ' ∇~ (II.2b)
~ vv(t,~r) at time t and position ~r can be
Like every rank 2 tensor, the velocity gradient tensor ∇~
decomposed into the sum of the symmetric and an antisymmetric part:
~ vv(t,~r) = D (t,~r) + R (t,~r),
∇~ (II.3a)
Using the tensors D and R we just introduced, whose physical meaning will be discussed at
length in Secs. II.1.2–II.1.3, relation (II.2b) can be recast as
Subtracting δ~`(t) from both sides, dividing by δt and taking the limit δt → 0, one finds for the rate
of change of the material vector, which is here denoted by a dot:
)·
In the following two subsections, we shall investigate the physical content of each of the tensors
R (t,~r) and D (t,~r).
~ =Ω ~ t,~r × V ~ ∀V ~ ∈ R3 .
R t,~r · V
(xxiv)
Wirbeltensor
II.1 Generic motion of a continuous medium 19
with ijk the totally antisymmetric Levi-Civita symbol. Using the antisymmetry of R t,~r , this
equivalently reads
Ω1 (t,~r) ≡ −R
R23 (t,~r), Ω2 (t,~r) ≡ −R
R31 (t,~r), Ω3 (t,~r) ≡ −R
R12 (t,~r). (II.7b)
~ r) = 1 ∇
Ω(t,~ ~ ×~v(t,~r). (II.8)
2
~ r), assuming that D (t,~r)
Let us now rewrite relation (II.6) with the help of the vector Ω(t,~
vanishes so as to isolate the effect of the remaining term. Under this assumption, the rate of change
of the material vector between two neighboring points reads
)·
The term on the right hand side is then exactly the rate of rotation of a vector ~`(t) in the motion
~ r). Accordingly, the pseudovector Ω(t,~
of a rigid body with instantaneous angular velocity Ω(t,~ ~ r) is
referred to as local angular velocity.(xxv) This a posteriori justifies the denomination rotation rate
tensor for the antisymmetric tensor R (t,~r).
Remarks:
~ r), one also defines the vorticity vector (xxvi) as the curl of
∗ Besides the local angular velocity Ω(t,~
the velocity field
~ r) = ∇
~ (t,~r) ≡ 2 Ω(t,~
ω ~ ×~v(t,~r). (II.10)
In fluid mechanics, the vorticity is actually more used than the local angular velocity.
~ (t,~r) = ~0
d~x × ω (II.11a)
or equivalently, in a point where none of the components of the vorticity vector vanishes,
dx1 dx2 dx3
1
= 2
= 3
. (II.11b)
ω (t,~r) ω (t,~r) ω (t,~r)
As we shall now see, the diagonal and off-diagonal components of D (t,~r) actually describe
the rates of change of different kinds of deformation. For simplicity, we assume throughout this
~ r) = ~0.
subsection that Ω(t,~
(xxv) (xxvi) (xxvii) (xxviii)
Wirbelvektor Wirbligkeit Wirbellinien Verzerrungsgeschwindigkeitstensor,
Deformationsgeschwindigkeitstensor
20 Kinematics of a continuous medium
Taylor-expanding the term between square brackets to first order then yields
3
X ∂ vi (t,~r)
δ`i (t + δt) − δ`i (t) ' δ`j (t) δt,
∂xj
j=1
up to terms of higher order in |δ~`(t)| or δt. Since we have assumed that both Ω(t,~
~ r)—or equivalently
the components R ij (t,~r) of the rotation rate tensor—and the off-diagonal D ij (t,~r) with i 6= j vanish,
one checks that the partial derivative ∂ vi (t,~r)/∂xj vanishes for i 6= j. That is, there is only the
term j = i in the sum, so that the equation simplifies to
∂vi (t,~r) i
δ`i (t + δt) − δ`i (t) ' δ` (t) δt = D ii (t,~r) δ`i (t) δt.
∂xi
Thus, the relative elongation of the i-th component—remember that there is no local rotation, so
that the change in δ`i is entirely due to a variation of the length of the material vector—in δt is
given by
δ`i (t + δt) − δ`i (t)
= D ii (t,~r) δt. (II.12)
δ`i (t)
This means that the diagonal component D ii (t,~r) represents the local rate of linear elongation in
direction i.
In the second line, one recognizes the trace of the tensor D (t,~r), which going back to the definition
of the latter is equal to the divergence of the velocity fluid:
∂ v1 (t,~r) ∂ v2 (t,~r) ∂ v3 (t,~r) ~
D 11 (t,~r) + D 22 (t,~r) + D 33 (t,~r) = + + = ∇ · ~v(t,~r).
∂x1 ∂x2 ∂x3
That is, this divergence represents the local and instantaneous volume expansion rate of the conti-
nuous medium. Accordingly, the flow of a fluid is referred to as incompressible in some region when
the velocity field in that region is divergence free:
~ · ~v(t,~r) = 0.
incompressible flow ⇔ ∇ (II.13)
x2 x2
1111111
0000000
(v1 +δv1111111
0000000
1 )δt
0000000
1111111
0000000
1111111
0000000
1111111
0000000
1111111
0000000
1111111
0000000
1111111
0000000
1111111
δα2 1111111
0000000
0000000
1111111
δ`2 0000000
1111111
0000000
1111111
δα1 (v2 +δv2 )δt
0000000
1111111
v2 δt
x1 x1
δ`1 v1 δt
Figure II.2 – Evolution of a material rectangle caught in the motion of a continuous medium
between times t (left) and t + δt (right).
the time evolution of a different object than a material vector, since anything that can affect the
latter—translation, rotation, dilatation—has already been described above.
Accordingly, we now look at the change between successive instants t and t + δt of an elementary
“material rectangle”, as pictured in Fig. II.2. We denote by ~v resp. ~v + δ~v the velocity at time t at
the lower left resp. upper right corner of the rectangle. Taylor expansions give for the components
of the shift δ~v
∂ v1 (t,~r) 2 ∂ v2 (t,~r) 1
δv1 = 2
δ` , δv2 = d` .
∂x ∂x1
Figure II.2 shows that what was at time t a right angle becomes an angle π/2−δα at t+dt, where
δα = δα1 + δα2 . In the limit of small δt, both δα1 and δα2 will be small and thus approximately
equal to their respective tangents. Using the fact that the parallelogram still has the same area—
since the diagonal components of D vanish—the projection of any side of the deformed rectangle at
time t + δt on its original direction at time t keeps approximately the same length, up to corrections
of order δt. One thus finds for the oriented angles
δv2 δt δv1 δt
δα1 ' and δα2 ' − .
δ`1 δ`2
With the Taylor expansions given above, this leads to
∂ v2 (t,~r) ∂ v1 (t,~r)
δα1 ' δt, δα2 ' − δt.
∂x1 ∂x2
Gathering all pieces, one finds
δα ∂ v2 (t,~r) ∂ v1 (t,~r)
' − = 2 D21 (t,~r). (II.14)
δt ∂x1 ∂x2
Going to the limit δt → 0, one sees that the off-diagonal component D 21 (t,~r) represents half the
local velocity of the “angular deformation”—the shear —around direction x3 .
Remark: To separate the two physical effects present in the strain rate tensor, it is sometimes
written as the sum of a diagonal rate-of-expansion tensor proportional to the identity 1 —which is
in fact the 11 -form of the metric tensor g of Cartesian coordinates—and a traceless rate-of-shear
tensor S :
1 ~
D (t,~r) = ∇ · ~v(t,~r) 1 + S (t,~r) (II.15a)
3
with
1 ~
~ vv(t,~r) T − 2 ∇
~ · ~v(t,~r) 1 .
S (t,~r) ≡ ∇~vv(t,~r) + ∇~ (II.15b)
2 3
22 Kinematics of a continuous medium
Summary
::::::::::
Gathering the findings of this Section, the most general motion of a material volume element
inside a continuous medium, in particular in a fluid, can be decomposed in four elements:
• a translation;
~ r) given by Eq. (II.8)—i.e. related to the anti-
• a rotation, with a local angular velocity Ω(t,~
symmetric part R (t,~r) of the velocity gradient—and equal to twice the (local) vorticity vector
ω
~ (t,~r);
• a local dilatation or contraction, in which the geometric form of the material volume element
remains unchanged, whose rate is given by the divergence of the velocity field ∇~ · ~v(t,~r), i.e.
encoded in the diagonal elements of the strain rate tensor D (t,~r);
• a change of shape (“deformation”) of the material volume element at constant volume, con-
trolled by the rate-of-shear tensor S (t,~r) [Eqs. (II.15b),(II.15d)], obtained by taking the trace-
less symmetric part of the velocity gradient.
smaller or larger than 1. Note that the Mach number can a priori be defined, and take different
values Ma(t,~r), at every point in a flow.
When the fluid flows in layers that do not mix with each other, so that the streamlines remain
parallel, the flow is referred to as laminar . In the opposite case the flow is turbulent.
In an analogous manner, one speaks of viscous resp. non-viscous flows to express the fact that the
fluid under consideration is modeled as viscous resp. inviscid—which leads to different equations of
motion—, irrespective of the fact that every fluid has a non-zero viscosity.
Other thermodynamic criteria are also used to characterize possible fluid motions: isothermal
flows—i.e. in which the temperature is uniform and remains constant—, isentropic flows—i.e. with-
out production of entropy—, and so on.
Fundamental equations of
non-relativistic fluid dynamics
Some of the most fundamental laws of physics are conservation equations for various quantities:
energy, momentum, (electric) charge, and so on. When applying them to many-body systems,
in particular to continuous media like moving fluids, care must be taken to consider isolated and
closed systems, to ensure their validity. At the very least, the amount of quantity exchanged with
the exterior of the system—for example the change in momentum per unit time due to external
forces, as given by Newton’s second law, or the change in energy due to the mechanical work of
these forces—must be quantifiable.
When this is the case, it is possible to re-express global conservation laws or more generally
balance equations—given in terms of macroscopic quantities like total mass, total energy, total
momentum, etc.—in a local form involving densities, using the generic recipe provided by Reynolds’
transport theorem (Sec. III.1). In the framework of a non-relativistic theory, in which the mass
or equivalently the particle number of a closed system is conserved, one may thus derive a general
continuity equation, holding at every point of the continuous medium (Sec. III.2).
The same approach may be followed to derive equations expressing the time evolution of mo-
mentum or energy under the influence of external forces acting at every point of the fluid. In either
case, it is however necessary to account for the possibility that several physical phenomena may
contribute to the transport of momentum and energy. Depending on whether or not, and how, every
form of transport is taken into account, one has different fluid models, leading to different equations
for the local expressions of Newton’s second law (Sec. III.3) or of energy balance (Sec. III.4).
boundary of Σ
boundary of
at time t + δt
2+
Σ at time t -
1
2-
streamlines
Figure III.1 – Time evolution of a closed material system transported in the motion of a
continuous medium.
it occupies a new volume in the reference frame. On Fig. III.1, one can distinguish between three
regions in position space:
• (1), which is common to the successive positions of Σ at t ant t + δt;
Remark: These examples illustrate the fact that the tensorial nature—scalar, vector, tensor of higher
rank—of the function associated with quantity G can be arbitrary.
For a material system Σ occupying at time t a volume V bounded by the control surface S,
Eq. III.1 leads to Z Z
G (t) = g m(t,~r) dM = gm(t,~r) ρ(t,~r) d3~r (III.2)
V V
for the value of G of the system, where in the second identity ρ(t,~r) = dM/d3~r is the local mass
density.
Let us now assume that the material system Σ is moving as part of a larger, flowing continuous
medium. To find the substantial derivative DG (t)/Dt of G (t), we shall first compute the variation
III.1 Reynolds transport theorem 27
δ G for the material system Σ between times t and t + δt, where δ is assumed to be small. At the
end of the calculation, we shall take the limit δt → 0.
Going back to the regions (1), (2−), (2+) defined in Fig. III.1, one can write
δ G = G 1 + G 2+ t+δt − G 1 + G 2− t = δ G 1 + δ G 2 ,
where the various indices denote the respective spatial domains and instants, and
δ G 1 ≡ G 1 t+δt − G 1 t , δ G 2 ≡ G 2+ t+δt − G 2− t .
• δ G 1 represents the variation of G inside region (1) due to the non-stationarity of the fluid
flow. In the limit δt → 0, this region (1) coincides with the control volume V: to leading order
in δt, one thus has
dG 1 (t)
Z Z
d 3 ∂
gm(t,~r) ρ(t,~r) d3~r δt,
δG 1 = δt = gm(t,~r) ρ(t,~r) d ~r δt = (III.3)
dt dt V V ∂t
where the first identity is a trivial Taylor expansion, the second one replaces the volume of
region (1) by V, while the last identity follows from the independence of the control volume
from time.
• δ G 2 represents the algebraic amount of G traversing between t and t+δt the control surface S,
either leaving (region 2+) or entering (region 2−) the control volume V, where in the latter case
the amount is counted negatively. This is precisely the flux —in the mathematical acceptation
of the term—through the surface S, oriented towards the exterior, of an appropriate flux
density for quantity G .(2)
All in all, Eqs. (III.3)–(III.4) yield after dividing by δt and taking the limit δt → 0 the so-called
Reynolds transport theorem:(xxxiii)(f)
DG (t)
Z I
∂ ~
gm(t,~r) ρ(t,~r) d3~r + gm(t,~r) ρ(t,~r)~v(t,~r) · d2 S.
= (III.5)
Dt V ∂t S
The first term on the right hand side of this relation represents a local time derivative ∂ G /∂t,
similar to the first term in Eq. I.18. In contrast, the second term is of convective type, i.e. directly
caused by the motion of matter, and represents the transport of G .
(2)
This flux density can be read off Eq. (III.4), namely gm(t,~r) ρ(t,~r)~v(t,~r).
(xxxiii)
Reynolds’scher Transportsatz
(f)
O. Reynolds, 1842–1912
28 Fundamental equations of non-relativistic fluid dynamics
Anticipating on the rest of the Chapter, this theorem will help us as follows. The “usual” laws
of dynamics are valid for closed, material systems Σ, rather than for open ones. Accordingly, these
laws involve time derivatives “following the system in its motion”, that is precisely the material
derivative D/Dt. Reynolds’ transport theorem (III.5) expresses the latter, for extensive quantities
G (t), in terms of local densities attached to fixed spatial positions, i.e. in Eulerian variables.
Remarks:
∗ The medium velocity ~v(t,~r) entering Reynolds transport theorem (III.5) is measured in the
reference frame R in which the control surface S remains motionless.
∗ Since relation (III.5) is traditionally referred to as a theorem, one may wonder what are its
assumptions. Obviously, the derivation of the result relies on the assumption that the specific
density gm(t,~r) and the velocity field ~v(t,~r) are both continuous and differentiable, in agreement
with the generic hypotheses in Sec. I.2.2. Figure III.1 actually also embodies the hidden, but
necessary assumption that the motion is continuous, which leads to the smooth evolution of the
connected system of material points which are together inside the control surface S at time t. Again,
this follows from suitable properties of ~v.
∗ Accordingly, the Reynolds transport theorem (III.5) does not hold if the velocity field, or the
specific density gm, is discontinuous. As was already mentioned in Sec. I.2.2, such discontinuities are
however necessary to account for some phenomena (shock waves, boundary between two immiscible
fluids. . . ). In such cases, it will be necessary to reformulate the transport theorem to take into
account the discontinuities.
That is, the time derivative of the mass contained in V is the negative of the mass flow rate through
S. In agreement with footnote 2, ρ(t,~r)~v(t,~r) is the mass flux density,(xxxiv) while its integral is
the mass flow rate.(xxxv)
Taking now G (t) = N , the associated specific density is gm(t,~r) = N/M . Since the product
of N/M with the mass density ρ(t,~r) is precisely the particle number density n (t,~r), Reynolds’
theorem (III.5) leads to
Z I
DN (t) ∂ 3 ~ = 0,
n (t,~r)~v(t,~r) · d2 S
= n (t,~r) d ~r + (III.8)
Dt ∂t V S
Equation (III.7) resp. (III.8) consitutes the integral formulation of mass resp. particle number
conservation.
Remark: In the case of a steady motion, Eq. (III.7) shows that the net mass flow rate through an
arbitrary closed geometrical surface S vanishes. That is, the entrance of some amount of fluid into
a (control) volume V must be compensated by the simultaneous departure of an equal mass from
the volume.
∂ρ(t,~r) ~
+ ∇ · ρ(t,~r)~v(t,~r) = 0 (III.9)
∂t
resp.
∂ n (t,~r) ~
+ ∇ · n (t,~r)~v(t,~r) = 0. (III.10)
∂t
Equation (III.9) represents the first of five dynamical (partial differential) equations which govern
the evolution of a non-relativistic fluid flow.
Remarks:
∗ The form of the continuity equation (III.9) does not depend on the properties of the flowing
medium, as for instance whether dissipative effects play a significant role or not. This should be
contrasted with the findings of the next two Sections.
~ · ρ(t,~r)~v(t,~r) = 0, i.e.
∗ In the case of a steady fluid flow, Eq. (III.9) gives ∇
~ · ~v(t,~r) +~v(t,~r) · ∇ρ(t,~
ρ(t,~r) ∇ ~ r) = 0.
Thus, the stationary flow of a homogeneous fluid, i.e. for which ρ(t,~r) is position independent, will
~ ·~v(t,~r) = 0, cf. Eq. (II.13)].
be incompressible [∇
elements, do depend on the properties of the fluid. The two most widespread models used for fluids
are that of a perfect fluid, which leads to the Euler equation (Sec. III.3.2), and of a Newtonian fluid,
for which Newton’s second law (III.11) translates into the Navier–Stokes equation (Sec. III.3.3).
Throughout this Section, we use the shorter designation “momentum” instead of the more accurate
“linear momentum”.
DP~ (t)
Z I
∂ ~
= ~v(t,~r) ρ(t,~r) d ~r + ~v(t,~r) ρ(t,~r)~v(t,~r) · d2 S.
3
(III.12)
Dt ∂t V S
Both terms on the right hand side can be transformed, to yield more tractable expressions. On
the one hand, since the control volume V is immobile in the reference frame R, the time derivative
can be taken inside the integral. Its action on ρ(t,~r)~v(t,~r) is then given by the usual product rule.
On the other hand, one can show the identity
I Z
2~ ∂ρ(t,~r) ~
+ ρ(t,~r) ~v(t,~r) · ∇ ~v(t,~r) d3~r.
~v(t,~r) ρ(t,~r)~v(t,~r) · d S = −~v(t,~r) (III.13)
S V ∂t
All in all, one thus obtains
DP~ (t)
Z Z
∂~v(t,~r) ~
3 D~v(t,~r) 3
= ρ(t,~r) + ~v(t,~r) · ∇ ~v(t,~r) d ~r = ρ(t,~r) d ~r. (III.14)
Dt V ∂t V Dt
~ denote the vector defined by the surface integral on the left
Proof of relation (III.13): let J(t)
hand side of that identity. For the i-th component of this vector, Gauss’ integral theorem gives
I Z
J i (t) = ~ = ∇~ · vi (t,~r) ρ(t,~r)~v(t,~r) d3~r.
i
v (t,~r) ρ(t,~r)~v(t,~r) · d2 S
S V
~ · ρ(t,~r)~v(t,~r) + ρ(t,~r)~v(t,~r) · ∇v
~ i (t,~r):
i
The action of the differential operator yields v (t,~r) ∇
the divergence in the first term can be expressed according to the continuity equation (III.9) as
the negative of the time derivative of the mass density, leading to
Remark: The derivation of Eq. (III.14) relies on purely algebraic transformations, either as encoded
in Reynolds’ transport theorem, or when going from relation (III.12) to (III.14). That is, it does not
imply any model—apart from that of a continuous medium—for the fluid properties. In particular,
whether or not dissipative effects are important in the fluid did not play any role here.
where f~V denotes the local density of body forces, while T~s is the mechanical stress vector introduced
in Eq. (I.20). The latter will now allow us to introduce various models of fluids.
The first, simplest model is that of a perfect fluid , or ideal fluid :
A perfect fluid is a fluid in which there are no shear stresses nor heat conduction. (III.16a)
Stated differently, at every point of a perfect fluid the stress vector T~s on a (test) surface element
d2 S moving with the fluid is normal to d2 S, irrespective of whether the fluid is at rest or in motion.
That is, introducing the normal unit vector ~en (~r) to d2 S oriented towards the exterior of the material
region acted upon,(3) one may write
Using relation (III.16b), the total surface forces in Eq. (III.15) can be transformed into a volume
integral:
I I I Z
T~s (t,~r) d S = − P (t,~r)~en (~r) d S = − P (t,~r) d S
2 2 2 ~ =− ∇ ~ P (t,~r) d3~r, (III.17)
S S S V
where the last identity follows from a corollary of the usual divergence theorem.
Remark: Although this might not be intuitive at first, the pressure P (t,~r) entering Eqs. (III.16b)–
(III.16d) is actually the hydrostatic pressure already introduced in the definition of the mechanical
stress in a fluid at rest, see Eq. (I.23). One heuristic justification is that the stresses are defined
as the forces per unit area exerted by a piece of fluid situated on one side of a surface on the fluid
situated on the other side. Even if the fluid is moving, the two fluid elements on both sides of the
surface—as well as the comoving test surface—have the same velocity,(4) i.e. their relative velocity
vanishes, just like in a fluid at rest.
III.3.2 b Euler equation
:::::::::::::::::::::::
Gathering Eqs. (III.11), (III.14), (III.15) and (III.17) yields
Z Z h
∂~v(t,~r) ~
3
i
~ P (t,~r) + f~V (t,~r) d3~r.
ρ(t,~r) + ~v(t,~r) · ∇ ~v(t,~r) d ~r = −∇
V ∂t V
(3)
Cf. the discussion between Eqs. (I.21a)–(I.21c).
(4)
... thanks to the usual continuity assumption: this no longer holds at a discontinuity!
32 Fundamental equations of non-relativistic fluid dynamics
Since this identity must hold irrespective of the control volume V, the integrands on both sides must
be equal. That is, the various fields they involve obey the Euler equation
∂~v(t,~r) ~
~ P (t,~r) + f~V (t,~r).
ρ(t,~r) + ~v(t,~r) · ∇ ~v(t,~r) = −∇ (III.18)
∂t
Remarks:
∗ The term in curly brackets on the left hand side is exactly the acceleration (I.17) of a material
point, as in Newton’s second law.
~ v, the Euler equation is a non-linear partial differential equation.
∗ Due to the convective term (~v ·∇)~
∗ Besides Newton’s second law for linear momentum, one could also think of investigating the
consequence of its analogue for angular momentum. Since we have assumed that the material
points do not have any intrinsic spin, the conservation of angular momentum, apart from leading to
the necessary symmetry of the stress tensor—which is realized in a perfect fluid, see Eq. (III.16c)
or (III.16d), and will also hold in a Newtonian fluid, see Eq. (III.26)—does not bring any new
dynamical equation.
D~v(t,~r) 1 ~
=− ∇P (t,~r) + ~aV (t,~r). (III.19)
Dt ρ(t,~r)
The interpretation of this form is quite straightforward: the acceleration of a material point (left
hand side) is the sum of the acceleration due to the pressure forces and the acceleration due to
volume forces (right hand side).
III.3 Momentum balance: Euler and Navier–Stokes equations 33
Alternatively, one may use the identity (in which the time and position variables have been
omitted for the sake of brevity)
2
~
~ ~v ~ ~v,
~v × ∇ ×~v = ∇ − ~v · ∇
2
which can be proved either starting from the usual formula for the double cross product—with a
small twist when applying the differential operator to a vector squared—or by working component
by component. Recognizing in the rightmost term the convective part of the Euler equation, one
can rewrite the latter, or equivalently Eq. (III.19), as
∂~v(t,~r) ~ ~v(t,~r)2
1 ~
+∇ − ~v(t,~r) × ω
~ (t,~r) = − ∇P (t,~r) + ~aV (t,~r), (III.20)
∂t 2 ρ(t,~r)
where we have made use of the vorticity vector (II.10). Note that the second term on the left hand
side of this equation involves the (gradient of the) kinetic energy per unit mass dK/dM .
In Sec. IV.2.1, we shall see yet another form of the Euler equation [Eq. (IV.8)], involving ther-
modynamic functions other than the pressure.
(5)
... which must be immobile in the reference frame in which the fluid has the velocity ~v entering definition (III.21b).
(xxxvii) (xxxviii)
Impulsdichte Impulsstromdichte
34 Fundamental equations of non-relativistic fluid dynamics
∗ Given its physical meaning, the momentum flux (density) tensor T is obviously related to the
Cauchy stress tensor σ . More precisely, T represents the forces exerted by a material point on
its neighbors, while σ stands for the stresses acting upon the material point due to its neighbors.
Invoking Newton’s third law—which in continuum mechanics is referred to as Cauchy’s fundamental
lemma—, these two tensors are simply opposite to each other.
∗ Building on the previous remark, the absence of shear stress defining a perfect fluid can be
reformulated as a condition of the momentum flux tensor:
A perfect fluid is a fluid at each point of which one can find a local velocity, such
that for an observer moving with that velocity the fluid is locally isotropic. (III.23)
The momentum flux tensor is thus diagonal in the observer’s reference frame.
Using definitions (III.21), one easily checks that the Euler equation (III.18) is equivalent to the
balance equations (for i = 1, 2, 3)
3
∂ Tij (t,~r)
X dT
ρ(t,~r) vi (t,~r) + = fVi (t,~r). (III.24a)
∂t dxj
j=1
with fVi the i-th component of the volume force density and d /dxi the covariant derivatives (see
Appendix C.1), that coincide with the partial derivatives in Cartesian coordinates.
Proof: For the sake of brevity, the (t,~r)-dependence of the various fields will not be specified.
One finds
3 3 3 3
∂(ρvi ) X dT Tij ∂ρ i ∂ vi X ij dP X j
i d(ρv )
X
j dv
i
+ = v + ρ + g + v + ρv
∂t j=1
dxj ∂t ∂t j=1
dxj j=1 dxj j=1
dxj
~ i + dP ,
i
∂ρ ~ ∂v
= vi + ∇ · (ρ~v) + ρ + (~v · ∇)v
∂t ∂t dxi
where we have used j g d/dxj = d/dxi . The first term between square brackets vanishes
P ij
thanks to the continuity equation (III.9). In turn, the second term is precisely the i-th component
of the left member of the Euler equation (III.18), i.e. it equals the i-th component of f~V minus
the third term, which represents the i-th component of ∇ ~ P.
∂
~ · T (t,~r) = f~V (t,~r),
ρ(t,~r)~v(t,~r) + ∇ (III.24b)
∂t
have used the symmetry of the momentum flux tensor T , while the action of the divergence
where we
on a 20 -tensor is defined through its components, which is to be read off Eq. (III.24a).
Since the coefficients in front of these two terms could a priori depend on~v2 , this represents the most
general symmetric tensor of degree 2 which can be constructed with the help of the flow velocity
only.
If the use of terms that depend on the spatial derivatives of the velocity field is also allowed, the
components of the momentum flux-density tensor can be of the following form, where for the sake
of brevity the variables t and ~r are omitted
!
dv i dv j d2v
i
T ij = P g ij + ρvi vj + A +B +O + ··· , (III.25)
dxj dxi dxj dxk
where dissipative effects are encored in the viscous stress tensor (xxxix)
1 ~ −1
~ v(t,~r) g−1 (t,~r)
π (t,~r) ≡ −2η(t,~r) D (t,~r)− ∇·~v(t,~r) g (t,~r) −ζ(t,~r) ∇·~
3 (III.26c)
for a Newtonian fluid
with D (t,~r) the strain rate tensor discussed in Sec. II.1.3. Component-wise
1 ~ ~ ·~v(t,~r) g ij (t,~r).
πij (t,~r) ≡ −2η(t,~r) D ij (t,~r) − ∇ ·~v(t,~r) g ij (t,~r) − ζ(t,~r) ∇
(III.26d)
3
(xxxix)
viskoser Spannungstensor
36 Fundamental equations of non-relativistic fluid dynamics
In terms of the traceless rate-of-shear tensor (II.15b) or of its components (II.15d), one may alter-
natively write
~ ·~v(t,~r) g−1 (t,~r)
π (t,~r) ≡ −2η(t,~r) S (t,~r) − ζ(t,~r) ∇ (III.26e)
• the (dynamical) shear viscosity (xl) η, which multiplies the traceless symmetric part of the
velocity gradient tensor, i.e. the conveniently termed rate-of-shear tensor;
• the bulk viscosity, also called second viscosity,(xli) ζ, which multiplies the volume-expansion
part of the velocity gradient tensor, i.e. the term proportional to ∇~ ·~v(t,~r).
The two corresponding contributions represent a diffusive transport of momentum in the fluid—
representing a third type of transport besides the convective and thermal ones.
Remarks:
∗ In the case of a Newtonian fluid, the viscosity coefficients η and ζ are independent of the flow
velocity. However, they still depend on the temperature and pressure of the fluid, so that they are
not necessarily uniform and constant in a real flowing fluid.
∗ In an incompressible flow, ∇ ~ · ~v(t,~r) = 0, the last contribution to the momentum flux den-
sity (III.26) drops out. Thus, the bulk viscosity ζ only plays a role in compressible fluid motions.(6)
∗ Expression (III.26c) or (III.26d) of the viscous stress tensor assumes implicitly that the fluid is
(locally) isotropic, since the coefficients η, ζ are independent of the directions i, j.
Accordingly, the mechanical stress vector on an infinitesimally small surface element situated at
point ~r with unit normal vector ~en (~r) reads
3
~
X 2 ~
Ts (t,~r) = σ (t,~r) · ~en (~r) = − P (t,~r) + ζ(t,~r) − η(t,~r) ∇ ·~v(t,~r) g ij (t,~r)
3
i,j=1
i
dv (t,~r) dvj (t,~r)
+ η(t,~r) + nj (~r)~ei (t,~r), (III.28)
dxj dxi
with nj (~r) the coordinate of ~en (~r) along direction j. One easily identifies the two components of
(6)
As a consequence, the bulk viscosity is often hard to measure—one has to devise a compressible flow—so that it
is actually not so well known for many substances, even well-studied ones [11].
(xl) (xli)
Scherviskosität Dehnviskosität, Volumenviskosität, zweite Viskosität
III.3 Momentum balance: Euler and Navier–Stokes equations 37
As in § III.3.2 a, the external contact forces acting on a fluid element delimited by a surface S
can easily be computed. Invoking the Stokes theorem yields
I I I
T~s (t,~r) d2 S = − P (t,~r)~en (~r) d2 S − π (t,~r) · ~en (~r) d2 S
S ZS Z S
=− ∇ ~ P (t,~r) d V + ∇
3 ~ · π(t,~r) d3 V
ZV ZV
=− ∇ ~ P (t,~r) d3 V + f~visc (t,~r) d3 V , (III.29a)
V V
with the local viscous friction force density
3 i j (t,~
X d dv (t,~
r) dv r)
f~visc (t,~r) = η(t,~r) + ~ej (t,~r)
dxi dxj dxi
i,j=1
~ 2 ~
+ ∇ ζ(t,~r) − η(t,~r) ∇ ·~v(t,~r) . (III.29b)
3
with 4 = ∇~ 2 the Laplacian. This is a non-linear partial differential equation of second order, while
the Euler equation (III.18) is of first order.
The difference between the order of the equations is not a mere detail: while the Euler equation
looks like the limit η, ζ → 0 of the Navier–Stokes equation, the corresponding is not necessarily
true of their solutions. This is for instance due to the fact that their respective boundary
conditions differ.
In the case of an incompressible flow, the local expansion rate in the Navier–Stokes equa-
tion (III.31) vanishes, leading to the incompressible Navier–Stokes equation
∂~v(t,~r) ~ ~v(t,~r) = − 1 ∇
~ P (t,~r) + ν4~v(t,~r),
+ ~v(t,~r) · ∇ (III.32)
∂t ρ
with ν ≡ η/ρ the kinematic shear viscosity.
Remark: The dimension of the dynamic viscosity coefficients η, ζ is ML−1 T−1 and the corresponding
unit in the SI system is the Poiseuille(i) , abbreviated Pa·s. In contrast, the kinematic viscosity has
dimension L2 T−1 , i.e. depends only on space and time, hence its denomination.
III.3.3 d Boundary conditions
:::::::::::::::::::::::::::::
At the interface between a viscous fluid, in particular a Newtonian one, and another body—be it
an obstacle in the flow, a wall containing the fluid, or even a second viscous fluid which is immiscible
with the first one—the relative velocity between the fluid and the body must vanish. This holds
not only for the normal component of the velocity (“impermeability” condition), as in perfect fluids,
but also for the tangential one, to account for the friction forces. The latter requirement is often
referred to as no-slip condition.
constrain the fluid flow for the geometry under consideration: we shall see several examples in the
next three Chapters.
A second possibility, which will also be illustrated in Chap. IV–VI, is that of a thermodynamic
constraint: isothermal flow, isentropic flow. . . For instance, one sees in thermodynamics that in
an adiabatic process for an ideal gas, the pressure and volume of the latter obey the relation
PV γ = constant, where γ denotes the ratio of the heat capacities at constant pressure (CP ) and
constant volume (CV ). Since V is proportional to 1/ρ, this so-called “adiabatic equation of state”
provides the needed constraint relating pressure and mass density.
Eventually, one may argue that non-relativistic physics automatically implies a further conser-
vation law besides those for mass and linear momentum, namely energy conservation. Thus, using
the reasoning adopted in Secs. (III.2) and (III.3), the rate of change of the total energy—internal,
kinetic and potential—of the matter inside a given volume equals the negative of the flow of energy
through the surface delimiting this volume. In agreement with the first law of thermodynamics, one
must take into account in the energy exchanged with the exterior of the volume not only the con-
vective transport of internal, kinetic and potential energies, but also the exchange of the mechanical
work of contact forces and—for dissipative fluids—of heat.
∂ 1 2
ρ(t,~r)~v(t,~r) + e(t,~r) + ρ(t,~r)Φ(t,~r)
∂t 2 (III.33)
~ · 1 2
+∇ ρ(t,~r)~v(t,~r) + e(t,~r) + P (t,~r) + ρ(t,~r)Φ(t,~r) ~v(t,~r) = 0,
2
where e denotes the local density of internal energy and Φ the potential energy per unit mass of
volume forces—assumed to be conservative—such that the acceleration ~aV present in Eq. (III.19)
~
equals −∇Φ.
Equation (III.33) will not be proven here—we shall see later in Sec. IX.3.3 that it emerges as
low-velocity limit of one of the equations of non-dissipative relativistic fluid dynamics. It is however
clearly of the usual form for a conservation equation, involving
• the total energy density, consisting of the kinetic ( 12 ρ~v2 ), internal (e) and potential (ρΦ)
energy densities; and
• the total energy flux density, which involves the previous three forms of energy, as well as that
exchanged as mechanical work of the pressure forces.(8)
Remarks:
∗ The presence of pressure in the flux density, however not in the density, is reminiscent of the
same property in definitions (III.21).
∗ The assumption that the volume forces are conservative is of course not innocuous. For instance,
it does not hold for Coriolis forces, which means that one must be careful when working in a rotating
reference frame.
(8)
Remember that when a system with pressure P increases its volume by an amount dV , it exerts a mechanical work
P dV , “provided” to its exterior.
40 Fundamental equations of non-relativistic fluid dynamics
∗ The careful reader will have noticed that energy conservation (III.33) constitutes a fifth equation
complementing the continuity and Euler equations (III.9) and (III.18), yet at the cost of introducing
a new scalar field, the energy density, so that now a sixth equation is needed. The latter is provided
by the thermal equation of state of the fluid, which relates its energy density, mass density and
pressure.(9) In contrast to the other equations, this equation of state is not “dynamical”, i.e. for
instance it does not involve time or spatial derivatives, but is purely algebraic.
One can show—again, this will be done in the relativistic case (§ IX.3.2), can also be seen as spe-
cial case of the result obtained for Newtonian fluids in Sec. III.4.3—that in a perfect, non-dissipative
fluid, the relation (III.33) expressing energy conservation locally, together with thermodynamic re-
lations, lead to the local conservation of entropy, expressed as
∂s(t,~r) ~
+ ∇ · s(t,~r)~v(t,~r) = 0, (III.34)
∂t
where s(t,~r) is the entropy density, while s(t,~r)~v(t,~r) represents the entropy flux density. The
motion of a perfect fluid is thus automatically isentropic.
This equation, together with a thermodynamic relation, is sometimes more practical than the
energy conservation equation (III.33), to which it is however totally equivalent.
(9)
This is where the assumption of local thermodynamic equilibrium (§ I.1.3) plays a crucial role.
(xlii) (xliii)
Wärmestromvektor Wärmeleitfähigkeit
(j)
J. B. Fourier, 1768–1830
III.4 Energy conservation, entropy balance 41
If the three transport coefficients η, ζ and κ vanish, this equation simplifies to that for perfect
fluids, Eq. (III.33).
Remark: The energy flux density can be read off Eq. (III.35), since it represents the term between
curly brackets. One can check that it can also be written as
1 2
ρ(t,~r)~v(t,~r) + e(t,~r) + P (t,~r) ~v(t,~r) − 2η(t,~r) S (t,~r) ·~v(t,~r)
2
~ ·~v(t,~r) ~v(t,~r) − κ(t,~r) ∇T
~ (t,~r), (III.36)
− ζ(t,~r) ∇
with S (t,~r) the traceless symmetric rate-of-shear tensor. One recognizes the various physical sources
of energy transport.
where the second identity follows from the Gibbs–Duhem relation dP = s dT + n dµ. On the other
hand, it leads to de = T ds + µ dn , which under consideration of the continuity equation for particle
number yields
∂e ∂s ∂n ∂s ~ · n~v .
=T +µ =T − µ∇ (III.37c)
∂t ∂t ∂t ∂t
With the help of relations (III.37a)–(III.37c), the energy conservation equation (III.35) can be
rewritten as
3 i
X ∂v ~
i ∂s ~ · s~v +~v · ∇
~P =
ρvi + ~v · ∇ v + T + T∇
∂t ∂t
i=1
3 i 3
∂ vj
X
X ∂ ∂v 2 ij ~ ∂ ~
~ · κ∇T
~
η + − g ∇ ·~
v v i + ζ ∇ ·~
v vi + ∇ . (III.37d)
∂xj ∂xj ∂xi 3 ∂xi
i,j=1 i=1
42 Fundamental equations of non-relativistic fluid dynamics
∂s(t,~r) ~
~ (t,~r)
∇T
+ ∇ · s(t,~r)~v(t,~r) − κ(t,~r) =
∂t T (t,~r)
~ (t,~r) 2
( )
1 2
~ ·~v(t,~r) + κ(t,~r) ∇T
2η(t,~r) S (t,~r) : S (t,~r) + ζ(t,~r) ∇
T (t,~r) T (t,~r)
(III.40b)
with S : S ≡ S ij S ij the scalar obtained by doubly contracting the rate-of-shear tensor with itself.
This equation can then be integrated over the V occupied by the fluid:
• When computing the integral of the divergence term on the left hand side with the Stokes
theorem, it vanishes thanks to the boundary conditions imposed at the surface S;
• the remaining term in the left member is simply the time derivative dS/dt of the total entropy
of the closed system;
• if all three transport coefficients η, ζ, κ are positive, then it is also the case of the three terms
on the right hand side.
dS
One thus finds ≥ 0, in agreement with the second law of thermodynamics.
dt
Remarks:
∗ The previous derivation may be seen as a proof that the transport coefficients must be positive
to ensure that the second law of thermodynamics holds.
∗ If all three transport coefficients η, ζ, κ vanish, i.e. in the case of a non-dissipative fluid,
Eq. (III.40) simply reduces to the entropy conservation equation in perfect fluids (III.34).
III.4 Energy conservation, entropy balance 43
• Landau–Lifshitz [3, 4] Chapter I § 1,2 & § 6,7 (perfect fluids) and Chapters II § 15,16 & V § 49
(Newtonian fluids).
In the previous Chapter, we have introduced the coupled dynamical equations that govern the flows
of perfect fluids in the non-relativistic regime, namely the continuity (III.9), Euler (III.18), and
energy conservation (III.33) equations for the mass density ρ(t,~r), fluid velocity ~v(t,~r) and pressure
P (t,~r). The present Chapter discusses solutions of that system of equations, i.e. possible motions of
perfect fluids,(10) obtained when using various assumptions to simplify the problem so as to render
the equations tractable analytically.
In the simplest possible case, there is simply no motion at all in the fluid; yet the volume forces
acting at each point still drive the behavior of the pressure and local mass density throughout the
medium (Sec. IV.1). Steady flows, in which there is by definition no real dynamics, are also easily
dealt with: both the Euler and energy conservation equations yield the Bernoulli equation, which
can be further simplified by kinematic assumptions on the flow (Sec. IV.2).
Section IV.3 deals with the dynamics of vortices, i.e. of the vorticity vector field, in the motion
of a perfect fluid. In such fluids, in case the pressure only depends on the mass density, there exists
a quantity, related to vorticity, that remains conserved if the volume forces at play are conservative.
The latter assumption is also necessary to define potential flows (Sec. IV.4), in which the further
hypothesis of an incompressible motion leads to simplified equations of motion, for which a number
of exact mathematical results are known, especially in the case of two-dimensional flows.
Throughout the Chapter, it is assumed that the body forces in the fluid, whose volume density
was denoted by f~V in Chapter III, are conservative, so that they derive from a potential. More
specifically, anticipating the fact that these volume forces are proportional to the mass they act
upon, we introduce the potential energy per unit mass Φ, such that
f~V (t,~r) = −ρ(t,~r) ∇Φ(t,~
~ r). (IV.1)
Throughout this Section, we adopt a fixed system of Cartesian coordinates (x1 , x2 , x3 ) = (x, y, z),
with the basis vector ~e3 oriented vertically and pointing upwards. In the following examples, we shall
consider the case of fluids in a homogeneous gravity field, leading to Φ(~r) = gz, with g = 9.8 m· s−2 .
(10)
. . . at least in an idealized world. Yet the reader is encouraged to relate the results to observations of her everyday
life—beyond the few illustrative examples provided by the author—, and to wonder how a small set of seemingly
“simple” mathematical equations can describe a wide variety of physical phenomena.
(11)
This is true only in the case of perfect fluids; for dissipative ones, there emerge new possibilities, see Sec. VI.1.1.
IV.1 Hydrostatics of a perfect fluid 45
Remark: If the stationarity condition is relaxed, the continuity equation still leads to ∂ρ/∂t = 0,
i.e. to a time-independent mass density. Whether time derivatives vanish or not makes no change
in the Euler equation when ~v = ~0. Eventually, energy conservation requires that the internal energy
density e—and thereby the pressure—follow the same time evolution as the “external” potential
energy Φ. Thus, there is a non-stationary hydrostatics, but in which the time evolution decouples
from the spatial problem.
Invoking the equation of state, one sees that the molecule number density n (~r) is also exponentially
distributed, in agreement with the Maxwell distribution of statistical mechanics since mgz is the
potential gravitational energy of a molecule at altitude z.
Taking as example air—which is a fictive ideal gas with molar mass(12) NA mair = 29 g · mol−1 —
the ratio kB T /mair g equals 8.8 × 103 m for T = 300 K, i.e. the pressure drops by a factor 2 for
every elevation gain of ca. 6 km. Obviously, however, assuming a constant temperature in the Earth
atmosphere over such a length scale is unrealistic.
(12)
NA denotes the Avogadro number.
(xliv)
barometrische Höhenformel
(k)
B. Pascal, 1623–1662
46 Non-relativistic flows of perfect fluids
Consider now the enthalpy H = U + PV of the fluid, whose change in an infinitesimal process
is the (exact) differential dH = T dS + V dP + µ dN .(13) In this relation, µ denotes the chemical
potential, which will however play no further role as we assume that the number of molecules in the
fluid is constant, leading to dN = 0. Dividing by N thus gives
V
H S
d =Td + dP ,
N N N
where the first term on the right-hand side vanishes since S/N is assumed to be constant. Dividing
now by the mass of a molecule of the fluid, one finds
w 1
d = dP , (IV.4)
ρ ρ
where w denotes the enthalpy density.
This identity relates the change in enthalpy pro unit mass w/ρ to the change in pressure P in
an elementary isentropic process. If one considers a fluid at local thermodynamic equilibrium, in
which w/ρ and P takes different values at different places, the identity relates the difference in w/ρ
to that in P between two (neighboring) points. Dividing by the distance between the two points,
and in the limit where this distance vanishes, one derives an identity similar to (IV.4) with gradients
instead of differentials.
Together with Eq. (IV.2), one thus obtains
~ w(~r) + Φ(~r) = ~0
∇ (IV.5)
ρ(~r)
w(z)
that is + gz = constant.
ρ(z)
Taking as example an ideal diatomic gas, its internal energy is U = 25 N kB T , resulting in the
enthalpy density
5 7
w = e + P = n kB T + n kB T = n kB T.
2 2
w 7 kB T
That is, = , with m the mass of a molecule of gas. Equation (IV.5) then gives
ρ 2 m
dT (z) mg
=−7 .
dz 2 kB
In the case of air, the term on the right hand side equals 9.77 × 10−3 K · m−1 = 9.77 K · km−1 ,
i.e. the temperature drops by ca. 10 degrees for an elevation gain of 1 km. This represents a much
better modeling of the (lower) Earth atmosphere as the isothermal assumption of Sec. IV.1.2.
Remarks:
∗ The International Standard Atmosphere (ISA)(14) model of the Earth atmosphere assumes a
(piecewise) linear dependence of the temperature on the altitude. The adopted value of the tem-
(13)
The reader in need of a short reminder on thermodynamics is referred to Appendix A.
(14)
See e.g. https://en.wikipedia.org/wiki/International_Standard_Atmosphere.
IV.1 Hydrostatics of a perfect fluid 47
perature gradient in the troposphere is smaller than the above, namely 6.5 K · km−1 , to take into
account the possible condensation of water vapor into droplets or even ice.
∗ Coming back to the derivation of relation (IV.5), if we had not assumed s/n constant, we would
have found
1 ~ ~ w(~r) ~ s(~r)
∇P (~r) = ∇ − T (~r) ∇ , (IV.6)
ρ(~r) ρ(~r) ρ(~r)
which we shall use in Sec. IV.2.1.
F~
S solid body
fluid 2 fluid 2
fluid 1 fluid 1
The system Σ is in mechanical equilibrium, i.e. the sum of the gravity forces acting at each point
of the volume V and the pressure forces exerted at each point of S by the fluid Σ0 must vanish:
• The gravity forces at each point result in a single force F~G , applied at the center of mass G
of Σ, whose direction and magnitude are those of the weight of the system Σ.
• According to the equilibrium condition, the resultant of the pressure forces must equal −F~G :
I
~ = −F~G .
P (~r) d2 S
S
If one now replaces the fluid system Σ by a (solid) body B, while keeping the fluids Σ0 outside
S in the same equilibrium state, the mechanical stresses inside Σ0 remain unchanged. Thus, the
resultant of the contact forces exerted by Σ0 on B is still given by F~ = −F~G and still applies at the
center of mass G of the fluid system Σ. This constitutes the celebrated Archimedes principle:
In addition, we have obtained the point of application of the resultant force (“buoyancy” (xlv) ) from
the fluid.
Remark: If the center of mass G of the “removed” fluid system does not coincide with the center of
mass of the body B, the latter will be submitted to a torque, since F~ and its weight are applied at
two different points.
(xlv)
statischer Auftrieb
48 Non-relativistic flows of perfect fluids
In the stationary regime, the first term on the left-hand side disappears(15) —and we now omit the
time variable from the equations.
Let d~`(~r) denote a vector tangential to the streamline at position ~r, i.e. parallel to ~v(~r). When
considering the scalar product of d~`(~r) with Eq. (IV.9), both terms on the right hand side yield zero.
First, the mixed product d~`(~r) · [~v(~r) × ω
~ (~r)] is zero for it involves two collinear vectors. Second,
d~`(~r) · ∇[s(~
~ r)/ρ(~r)] vanishes due to the conservation of s/n in flows of perfect fluids, which together
~ r)/n (~r)] = 0, where n is proportional to ρ.
with the stationarity reads ~v(~r) · ∇[s(~
On the other hand, d`(~r) · ∇ represents the derivative along the direction of d~`, i.e. along the
~ ~
streamline at ~r. Thus, the derivative of the term in squared brackets on the left hand side of
Eq. (IV.9) vanishes along a streamline, i.e. the term remains constant on a streamline:
~v(~r)2 w(~r)
+ + Φ(~r) = constant along a streamline (IV.10)
2 ρ(~r)
where the value of the constant depends on the streamline. Relation (IV.10) is referred as to the
Bernoulli equation.(m)
In the stationary regime, the energy conservation equation (III.33), in which one recognizes the
enthalpy density w(~r) = e(~r) + P (~r) in the flux term, leads to the same relation (IV.10).
The first term in Eq. (III.33) vanishes due to the stationarity condition, leaving (we drop the
variables) 2
~ ~v w
∇· + + Φ ρ~v = 0.
2 ρ
Applying the product rule to the left member, one finds a first term proportional to ∇ ~ · (ρ~v)—
which vanishes thanks to the continuity equation (III.9)—, leaving only the other term, which
is precisely ρ times the derivative along ~v of the left hand side of the Bernoulli equation.
(15)
This yields a relation known as Crocco’s theorem (xlvi)(l)
(xlvi)
Croccos Wirbelsatz
(l) (m)
L. Crocco, 1909–1986 D. Bernoulli, 1700–1782
IV.2 Steady inviscid flows 49
~v(~r)2 P (~r)
In incompressible flows + + Φ(~r) is constant along a streamline. (IV.11)
2 ρ
Can this form be reconciled with the other one (IV.10), which is still what follows from the
energy conservation equation? Subtracting one from the other, one finds that the ratio e(~r)/ρ
is constant along streamlines. That is, since ρ is uniform, the internal energy density is con-
stant along pathlines—which coincide with streamlines in a steady flow. Now, thermodynamics
expresses the differential de through ds and dn : since both entropy and particle number are
conserved along a pathline, so is internal energy, i.e. Eq. (IV.10) is compatible with Eq. (IV.11).
Figure IV.2
vB2 = vA2 + 2gh.
That is, the speed of efflux is the same as that acquired by a body in free fall from the same height
h in the same gravity field.
(16)
The author confesses that he has a better physical intuition of pressure than of enthalpy, hence his parti pris.
(17)
One can show that the pressure in the liquid at point B equals the atmospheric pressure provided the local
streamlines are parallel to each other—that is, the flow is laminar.
(xlvii)
Torricellis Theorem
(n)
E. Torricelli, 1608–1647
50 Non-relativistic flows of perfect fluids
Remark: To be allowed to apply the Bernoulli equation, one should first show that the liquid flows
steadily. If the horizontal cross section of the vessel is much larger than the aperture of the hole
and h large enough, this holds to a good approximation.
IV.2.2 b Venturi effect
:::::::::::::::::::::
Consider now the incompressible flow of a fluid in the geometry illustrated in Fig. IV.3. As we
shall only be interested in the average velocity or pressure of the fluid across a cross section of the
tube, the flow is effectively one-dimensional.
@
v1- v2 -
@
S s
Figure IV.3
The conservation of the mass flow rate in the tube, which represents the integral formulation of
the continuity equation (III.9), leads to ρSv1 = ρ s v2 , i.e. v2 = (S/s )v1 > v1 , with S resp. s the area
of the tube cross section in its broad resp. narrow section.
On the other hand, the Bernoulli equation at constant height, and thus potential energy, gives
v12 P 1 v2 P 2
+ = 2+ .
2 ρ 2 ρ
All in all, the pressure in the narrow section is thus smaller than in the broad section, P 2 < P 1 ,
which constitutes the Venturi effect.(o)
Using mass conservation and the Bernoulli equation, one can express v1 or v2 in terms of the
tube cross section areas and the pressure difference. For instance, the mass flow rate reads
1/2
P 1 −P 2 S 2
ρS 2 − 1 .
ρ s2
A 0
•
•
- •
O• 0 A B
- I•
O• -
-
-
~v -
• Despite the presence of the hole at the end point I, the flow does not penetrate in the inner
tube, so that ~vI = ~0 to a good approximation.
(o) (p)
G. B. Venturi, 1746–1822 H. Pitot, 1695–1771
IV.2 Steady inviscid flows 51
• In the broader tube, there is a hole at a point A, which is far enough from I to ensure that the
fluid flow in the vicinity of A is no longer perturbed by the extremity of the tube: ~vA =~vA0 '~v,
where the second identity holds thanks to the thinness of the tube, which thus perturbs the
flow properties minimally. In addition, the pressure in the broader tube is uniform, so that
P = PB .
If one neglects the height differences—which is a posteriori justified by the numerical values we shall
find—the (incompressible) Bernoulli equation gives first
~v2
PO + ρ = PI
2
along the streamline OI, and
~v2 ~v2 0
PO0 + ρ = PA0 + ρ A
2 2
along the streamline O0 A0 . Using PO0 ' PO , PA0 ' PA and ~vA0 ' ~v, the latter identity leads to
PO ' PA = PB . One thus finds
~v2
PI − PB = ρ
,
2
so that a measurement of PI − PB gives an estimate of |~v|.
For instance, in air (ρ ∼ 1.3 kg · m−3 ) a velocity of 100 m · s−1 results in a pressure difference of
6.5 × 103 Pa = 6.5 × 10−2 atm. With a height difference h of a few centimeters between O and A0 ,
the neglected term ρgh is of order 1 Pa.
Remarks:
∗ The flow of a fluid with velocity ~v around a motionless Pitot tube is equivalent to the motion of
a Pitot tube with velocity −~v in a fluid at rest. Thus Pitot tubes are used to measure the speed of
airplanes.
∗ Is the flow of air really incompressible at velocities of 100 m · s−1 or higher? Not really, since the
Mach number (II.16) becomes larger than 0.3. In practice, one thus rather uses the “compressible”
Bernoulli equation (IV.10), yet the basic principles presented above remain valid.
Intuitively, one can expect that the cylinder will drag the neighboring fluid layers along in its
rotation.(18) In that case, the fluid velocity due to that rotation will add up to resp. be subtracted
from the initial flow velocity in the lower resp. upper region close to the cylinder in Fig. IV.5.
(18)
Strictly speaking, this is not true in perfect fluids, only in real fluids with friction! Nevertheless, the tangential
forces due to viscosity in the latter may be small enough that the Bernoulli equation remains approximately valid,
as is assumed here.
52 Non-relativistic flows of perfect fluids
Invoking now the Bernoulli equation—in which the height difference between both sides of the
cylinder is neglected—, the pressure will be larger above the cylinder than below it. Accordingly,
the cylinder will experience a resulting force directed downwards—more precisely, it is proportional
~ C —, which constitutes the Magnus effect.(q)
to ~v0 × ω
Stated differently, the vorticity field is the flux density of the circulation of the velocity.
This relationship between circulation and vorticity will be further exploited hereafter: we shall
now establish and formulate results at the integral level, namely for the circulation, which will then
be expressed at the differential level, i.e. in terms of the vorticity, in Sec. IV.3.2.
Many results take a simpler form in a so-called barotropic fluid ,(xlviii) in which the pressure can
be expressed as function of only the mass density: P = P (ρ), irrespective of whether the fluid is
otherwise perfect or dissipative. An example of such a result is
Kelvin’s circulation theorem:(r)
DΓ~γ (t)
= 0. (IV.14b)
Dt
Proof: For the sake of brevity, the arguments of the fields are omitted in case it is not necessary
to specify them. Differentiating definition (IV.12) first gives
D 1 ∂~γ (t, λ)
Z 1 2
∂~v X ∂~v ∂γ i
Z
DΓ~γ ∂ ~γ ∂~γ
= ·~v(t, ~γ (t, λ)) dλ = ·~v + · + dλ.
Dt Dt 0 ∂λ 0 ∂λ ∂t ∂λ ∂t i
∂xi ∂t
∂~γ (t, λ)
Since the contour ~γ (t, λ) flows with the fluid, =~v(t, ~γ (t, λ)), which leads to
∂t
Z 1
DΓ~γ ∂~v ∂~γ ∂~v ~ ~v dλ.
= · ~v + · + ~v · ∇
Dt 0 ∂λ ∂λ ∂t
The first term in the curly brackets is clearly the derivative with respect to λ of ~v2/2, so that
its integral along a closed curve vanishes. The second term involves the material derivative of
~ leads to
~v, as given by the Euler equation. Using Eq. (III.19) with ~aV = −∇Φ
Z 1 ~
∇P ~
DΓ~γ ∂~γ
= − − ∇Φ · dλ.
Dt 0 ρ ∂λ
~ around a closed contour vanishes, leaving
Again, the circulation of the gradient ∇Φ
I ~
DΓ~γ (t) ∇P (t,~r) ~
=− · d`, (IV.15)
Dt γ ρ(t,~
~ r)
which constitutes the general case of Kelvin’s circulation theorem for perfect fluids with conser-
vative volume forces.
Transforming the contour integral with Stokes’ theorem yields the surface integral of
~ ~ ~P ~ P × ∇ρ
~ ~ P × ∇ρ
~
~ ∇P ∇×∇ ∇ ∇
∇× = + = . (IV.16)
ρ ρ ρ2 ρ2
~ P and ∇ρ
In a barotropic fluid, the rightmost term of this identity vanishes since ∇ ~ are collinear,
which yields relation (IV.14).
Remark: Using relation (IV.13) and the fact that the area S~γ (t) bounded by the curve ~
γ at time t
defines a material surface, which will be transported in the fluid motion, Kelvin’s theorem (IV.14)
can be restated as
In a perfect barotropic fluid with conservative volume forces, the flux of the
vorticity across a material surface is conserved. (IV.17)
In the flow of a perfect barotropic fluid with conservative volume forces, the
vorticity lines and vorticity tubes move with the fluid. (IV.18)
Similar to the definition of stream tubes in Sec. I.3.3, a vorticity tube is defined as the surface
formed by the vorticity lines tangent to a given closed geometrical curve.
~ = ~0, one has
And in the case of vanishing vorticity ω
Lagrange’s theorem:
Kelvin’s circulation theorem (IV.14) and its corollaries imply that vorticity cannot be created
nor destroyed in the flow of a perfect barotropic fluid with conservative volume forces. However,
(s)
H. von Helmholtz, 1821–1894
54 Non-relativistic flows of perfect fluids
the more general form (IV.15) already show that in a non-barotropic fluid, there is a “source” for
vorticity, which leads to the non-conservation of the circulation of the flow velocity. Similarly, non-
conservative forces—for instance a Coriolis force in a rotating reference frame—may contribute a
non-vanishing term in Eq. (IV.15) leading to a change in the circulation. We shall see in Sec. VI.5
that viscous stresses also affect the transport of vorticity in a fluid.
Since the divergence of the vorticity field ∇ ~ ·ω ~ (t,~r) vanishes, the previous two equations yield
∂~ω (t,~r) ~ P (t,~r) × ∇ρ(t,~
∇ ~ r)
~ ω ~ ~v(t,~r) = − ∇ ~ ·~v(t,~r) ω
+ ~v(t,~r) · ∇ ~ (t,~r) − ω~ (t,~r) · ∇ ~ (t,~r) − .
∂t ρ(t,~r)2
(IV.21)
While it is tempting to introduce the material derivative D~ ω /Dt on the left hand side of this
equation, for the first two terms, we rather define the whole left member to be a new derivative
D~v ω
~ (t,~r) ∂~
ω (t,~r) ~ ω
~ ~v(t,~r)
≡ + ~v(t,~r) · ∇ ~ (t,~r) − ω~ (t,~r) · ∇ (IV.22a)
Dt ∂t
or equivalently
D~v ω
~ (t,~r) D~
ω (t,~r) ~ ~v(t,~r).
≡ − ω~ (t,~r) · ∇ (IV.22b)
Dt Dt
We shall refer to D~v /Dt as the comoving time derivative, for reasons that will be explained at the
end of this Section.
Using this definition, Eq. (IV.21) reads
D~v ω
~ (t,~r) ~ P (t,~r) × ∇ρ(t,~
∇ ~ r)
~ ·~v(t,~r) ω
=− ∇ ~ (t,~r) − 2
. (IV.23)
Dt ρ(t,~r)
In the particular of a barotropic fluid—recall that we also assumed that it is ideal and only has
conservative volume forces—this becomes
D~v ω
~ (t,~r)
~ ·~v(t,~r) ω
=− ∇ ~ (t,~r). (IV.24)
Dt
~r ~v(t,~r) δ t
x1
x3
Figure IV.6 – Positions of a material line element δ~` at successive times t and t + δt.
More generally, the comoving time derivative introduced in Eq. (IV.22a) may be rewritten as
D~v ∂
( · ) ≡ ( · ) + L~v ( · ), (IV.27)
Dt ∂t
where L~v denotes the Lie derivative along the velocity field ~v(~r), whose action on an arbitrary
56 Non-relativistic flows of perfect fluids
vector field ω
~ (~r) is precisely (time plays no role here)
~ ω ~ ~v(~r),
L~v ω
~ (~r) ≡ ~v(~r) · ∇ ~ (~r) − ω~ (~r) · ∇
~
~v(t,~r) = −∇ϕ(t,~
r) (IV.28)
∗ Since Lagrange’s theorem does not hold in a dissipative fluid, in which vorticity can be created
or annihilated (Sec. VI.5), the rationale behind the definition of the velocity potential disappears.
Using the velocity potential (IV.28) and the relation ~aV = −∇Φ~ expressing that the volume
forces are conservative, the Euler equation (III.20) reads
( 2 )
~
∂ ∇ϕ(t,~ ~
r) ~ ∇ϕ(t,~ r) 1 ~
− +∇ + Φ(t,~r) = − ∇P (t,~r).
∂t 2 ρ(t,~r)
Assuming that the flow is also incompressible, and thus ρ constant, this becomes
( 2 )
~
∂ ∇ϕ(t,~ ~
r) ~ ∇ϕ(t,~ r) P (t,~r)
− +∇ + + Φ(t,~r) = ~0. (IV.29)
∂t 2 ρ
(xlix) (l)
Potentialströmung Geschwindigkeitspotential
IV.4 Potential flows 57
or equivalently
2
~
∂ϕ(t,~r) ∇ϕ(t,~
r) P (t,~r)
− + + + Φ(t,~r) = C(t), (IV.30)
∂t 2 ρ
4ϕ(t,~r) = 0 (IV.31)
~
− ~en (t,~r) · ∇ϕ(t,~
r) = ~en (t,~r) · ~vobs. (t,~r) on S(t). (IV.32a)
∗ In potential flows, the dependences on time and space are somewhat separated: The Laplace
equation (IV.31) governs the spatial dependence of ϕ and thus~v; meanwhile, time enters the bound-
ary conditions (IV.32), thus is used to “normalize” the solution of the Laplace equation. In turn,
when ϕ is known, relation (IV.30) gives the pressure field, where the integration “constant” C(t)
will also be fixed by boundary conditions.
On a simply connected domain, the Neumann problem (IV.31)–(IV.32) for the velocity
potential admits a solution ϕ(t,~r), which is unique up to an additive constant. (IV.33)
In turn, the flow velocity field ~v(t,~r) given by relation (IV.28) is unique.
For a flow on a simply connected region, the relation (IV.28) between the flow velocity and its
potential is “easily” invertible: fixing some reference position ~r0 in the domain, one may write
Z
ϕ(t,~r) = ϕ(t,~r0 ) − ~v(t,~r0 ) · d~`(~r0 ) (IV.34)
~γ
where the line integral is taken along any path ~γ on D connecting the positions ~r0 and ~r.
That the line integral only depends on the path extremities ~r0 , ~r, not on the path itself, is
clearly equivalent to Stokes’ theorem stating that the circulation of velocity along any closed
contour in the domain D is zero—it equals the flux of the vorticity, which is everywhere zero,
through a surface delimited by the contour and entirely contained in D.
Thus, ϕ(t,~r) is uniquely defined once the value ϕ(t,~r0 ), which is the arbitrary additive constant
mentioned above, has been fixed.
This reasoning no longer holds in a multiply connected domain, as we now further discuss.
IV.4.2 b Potential flows in doubly connected regions
:::::::::::::::::::::::::::::::::::::::::::::::::::::
As a matter of fact, in a doubly (or a fortiori multiply) connected domain, there are by definition
non-contractible closed paths. Considering for instance the domain D traversed by an infinite
cylinder—which is not part of the domain—of Fig. IV.7, the path going from ~r0 to ~r2 along ~γ0→2
0 (21) cannot be continuously shrunk to a point without leaving D.
then coming back to ~r0 along ~γ0→2
This opens the possibility that the line integral in relation (IV.34) depend on the path connecting
two points.
~r2
•
~γ 0
60→2
~γ0→2
00
~γ0→2
6
6
~γ 0
-0→1
• •~
r1
~r0 -
~γ0→1
Figure IV.7
In a doubly connected domain D, there is only a single “hole” that prevents closed paths from
being homotopic to a point, i.e. contractible. Let Γ(t) denote the circulation at time t of the
velocity around a closed contour, with a given “positive” orientation, circling the hole once. One
easily checks—e.g. invoking Stokes’ theorem—that this circulation has the same value for all closed
(21) 0
More precisely, if ~γ0→2 is parameterized by λ ∈ [0, 1] when going from ~r0 to ~r2 , a path from ~r2 to ~r0 with the
0 0
same geometric support—which is what is meant by “coming back along ~γ0→2 ”—is λ 7→ ~γ0→2 (1 − λ).
IV.4 Potential flows 59
paths with the same orientation going only once around the hole, since they can be continuously
deformed into each other without leaving D. Accordingly, the “universal” circulation Γ(t) is also
referred to as cyclic constant (li) of the flow.
More generally, the circulation at time t of the velocity around a closed curve circling the hole
n times and oriented in the positive resp. negative direction is nΓ(t) resp. −nΓ(t).
Going back to the line integral in Eq. IV.34, its value will generally depend on the path ~γ from
~r0 to ~r—or more precisely, on the class, defined by the number of loops around the hole, of the
path. Illustrating this idea on Fig. IV.7, while the line integral from ~r0 to ~r2 along the path ~γ0→2
0
will have a given value I, the line integral along ~γ0→2 will differ by one (say, positive) unit of Γ(t)
00
and be equal to I +Γ(t). In turn, the integral along ~γ0→2 , which makes one more negatively oriented
loop than ~γ0→2 around the cylinder, takes the value I − Γ(t).
These preliminary discussions suggest that if the Neumann problem (IV.31)–(IV.32) for the
velocity potential on a doubly connected domain admits a solution ϕ(t,~r), the latter will not be
a scalar function in the usual sense, but rather a multivalued function, whose various values at a
given position ~r at a fixed time t differ by an integer factor of the cyclic constant Γ(t).
All in all, the following result holds provided the cyclic constant Γ(t) is known, i.e. if its value
at time t is part of the boundary conditions:
On a doubly connected domain, the Neumann problem (IV.31)–(IV.32) for the velocity
potential with given cyclic constant Γ(t) admits a solution ϕ(t,~r), which is unique (IV.35)
up to an additive constant. The associated flow velocity field ~v(t,~r) is unique.
The above wording does not specify the nature of the solution ϕ(t,~r):
• if Γ(t) = 0, in which case the flow is said to be acyclic, the velocity potential ϕ(t,~r) is a
univalued function;
• if Γ(t) 6= 0, i.e. in a cyclic flow , the velocity potential ϕ(t,~r) is a multivalued function of its
spatial argument. Yet as the difference between the various values at a given ~r is function of
time only, the velocity field (IV.28) remains uniquely defined.
Remarks:
∗ Inspecting Eq. (IV.30), one might fear that the pressure field P (t,~r) be multivalued, reflecting
the term ∂ϕ(t,~r)/∂t. Actually, however, Eq. (IV.30) is a first integral of Eq. (IV.29), in which the
~r-independent multiples of Γ(t) distinguishing the multiple values of ϕ(t,~r) disappear when the
gradient is taken. That is, the term ∂ϕ(t,~r)/∂t is to be taken with a grain of salt, since in fact it
does not contain Γ(t) or its time derivative.
∗ In agreement with the first remark, the reader should remember that the velocity potential ϕ(t,~r)
is just a useful auxiliary mathematical function,(22) yet the physical quantity is the velocity itself.
Thus the possible multivaluedness of ϕ(t,~r) is not a real physical problem.
(22) ~ ..
Like its cousins: gravitational potential ΦNewt. , electrostatic potential ΦCoul. , magnetic vector potential A.
(li)
zyklische Konstante
60 Non-relativistic flows of perfect fluids
∂ψ(x, y) ∂ψ(x, y)
vx (x, y) = − , vy (x, y) = (IV.36)
∂y ∂x
at every point (x, y). Indeed, when the above two relations hold, the incompressibility criterion
~ ·~v(x, y) = 0 is fulfilled automatically.
∇
Remark: As in the case of the relation between the flow velocity field and the corresponding potential,
Eq. (IV.28), the overall sign in the relation between ~v(~r) and ψ(~r) is conventional. Yet if one wishes
to define the complex flow potential as in Eq. (IV.39) below, the relative sign of ϕ(~r) and ψ(~r) is
fixed.
The stream function for a given planar fluid motion is such that the lines along which ψ(~r) is
constant are precisely the streamlines of the flow.
Let d~x(λ) denote a differential line element of a curve ~x(λ) of constant ψ(~r), i.e. a curve along
~ = ~0. Then d~x(λ) · ∇ψ~ ~x(λ) = 0 at every point on the line: using relations (IV.36),
which ∇ψ
one recovers Eq. (I.15b) characterizing a streamline.
Stream functions are also defined in three-dimensional flows, yet in that case two of them are
needed. More precisely, one can find two linearly independent functions ψ1 (~r), ψ2 (~r), such
that the streamlines are the intersections of the surfaces of constant ψ1 and of constant ψ2 .
~ 1 (~r) × ∇ψ
That is, they are such that the flow velocity obeys ~v(~r) ∝ ∇ψ ~ 2 (~r), with an a priori
position-dependent proportionality factor—which can be taken identically equal to unity in an
incompressible flow.
Consider now a potential flow, i.e. which is not only incompressible, but also irrotational. For
such a two-dimensional flow, the condition of vanishing vorticity reads
∂ vy (x, y) ∂ vx (x, y)
ω z (x, y) = − = 0,
∂x ∂y
which under consideration of relations (IV.36) gives
4ψ(x, y) = 0 (IV.37a)
at every point (x, y). That is, the stream function obeys the Laplace equation—just like the velocity
potential ϕ(~r).
A difference with ϕ(~r) arises with respect to the boundary conditions. At an obstacle or walls,
modeled by a “surface” S—in the plane R2 , this surface is rather a curve—, the impermeability
condition implies that the velocity is tangential to S, i.e. S coincides with a streamline:
ψ(x, y) = constant on S (IV.37b)
(lii)
Stromfunktion
IV.4 Potential flows 61
For a flow on an unbounded domain, the velocity is assumed to be uniform at infinity, ~v(x, y) →~v∞
which is the case if
y x
ψ(x, y) ∼ v∞ x − v∞ y (IV.37c)
|~r|→∞
x , vy the components of ~v .
with v∞ ∞ ∞
The boundary conditions (IV.37b)–(IV.37c) on the stream function are thus dissimilar from the
corresponding conditions (IV.32a)–(IV.32b) on the velocity potential. In particular, the condition
at an obstacle involves the stream function itself, instead of its derivative: the Laplace differential
equation (IV.37a) with conditions (IV.37b)–(IV.37c) represents a Dirichlet problem,(v) or boundary
value problem of the first kind, instead of a Neumann problem.
which will automatically be holomorphic on the domain where the flow is defined. The functions ϕ
and ψ are then said to be conjugate to each other. In line with that notion, the curves in the plane
along which one of the functions is constant are the field lines of the other, and reciprocally.
Besides the complex potential φ(z), one also defines the corresponding complex velocity as the
negative of its derivative, namely
dφ(z)
w(z) ≡ − = vx (x, y) − ivy (x, y) (IV.40)
dz
where the second identity follows at once from the definition of φ and the relations between ϕ or ψ
and the flow velocity. Like φ(z), the complex velocity w(z) is an analytic function of z.
IV.4.3 b Elementary two-dimensional potential flows
:::::::::::::::::::::::::::::::::::::::::::::::::::::
As a converse to the above construction of the complex potential, the real and imaginary parts
of any analytic function of a complex variable are harmonic functions, i.e. any analytical function
φ(z) defines a two-dimensional potential flow on its domain of definition. Accordingly, we now
investigate a few “basic” complex potentials and the flows they describe.
Uniform flow
The simplest possibility is that of a linear complex potential:
(v)
P. G. (Lejeune-)Dirichlet, 1805–1859
62 Non-relativistic flows of perfect fluids
y Using for instance Eq. (IV.40), this trivially leads to a uniform
6
3
3
3
3
3
3
3
velocity field making an angle α with the x-direction,
α
3
3
3
3
3
3
3
- ~v(x, y) = cos α~ex + sin α~ey v,
x
3
3
3
3
3
3
3 3
as illustrated in Fig. IV.8, in which a few streamlines are dis-
3
3
3
3
3
3
played, to which the equipotential lines (not shown) of ϕ(x, y)
are perpendicular.
Figure IV.8
y y
6 6
- -
x x
Figure IV.9 – Streamlines (full) and equipotential lines (dashed) for a flow source (IV.42c)
(left) and a pointlike vortex (IV.43b) (right).
(23)
The reader unwilling to take the logarithm of a dimensionful quantity—to which she is entirely entitled—may
divide z − z0 resp. r by a length in the potentials (IV.42a) and (IV.43a) resp. (IV.42d) and (IV.43c), or write the
difference in Eq. (IV.45) as the logarithm of a quotient. She will however quickly convince herself that this does
not affect the velocities (IV.42b) and (IV.43b), nor the potential (IV.44a).
IV.4 Potential flows 63
Pointlike vortex
The “conjugate” flow to the previous one, i.e. that for which ϕ and ψ are exchanged, corresponds
to the complex potential(23)
iΓ
φ(z) = log(z − z0 ) with Γ ∈ R, z0 ∈ C. (IV.43a)
2π
Using as above polar coordinates (r, θ) centered on z0 , the flow velocity is purely tangential,
Γ
~v(r, θ) = ~eθ , (IV.43b)
2πr2
as shown in Fig. IV.9 (right), where the basis vector ~eθ is normalized to r, cf. Eq. (C.6). The
complex potential (IV.43a) thus describes a vortex situated at z0 .
In turn, the velocity potential and stream function read
Γ Γ
ϕ(r, θ) = − θ, ψ(r, θ) = log r, (IV.43c)
2π 2π
to be compared with those for a flow source, Eq. (IV.42d).
Remark: When writing down the complex velocity potentials (IV.42a) or (IV.43a), we left aside the
issue of the (logarithmic!) branch point at z = z0 —and we did not specify which branch of the
logarithm we consider. Now, either potential corresponds to a flow that is actually defined on a
doubly connected region, since the velocity diverges at z = z0 . From the discussion in § IV.4.2 b, on
such domains the potential is a multivalued object, yet this is irrelevant for the physical quantities,
namely the velocity field, which remains uniquely defined at each point. This is precisely what is
illustrated here by the different branches of the logarithm, which differ by a constant multiple of
2iπ that does not affect the derivative.
Flow dipole
A further possible irrotational and incompressible two-dimensional flow is that defined by the
complex potential
µ eiα
φ(z) = with µ ∈ R, α ∈ R, z0 ∈ C (IV.44a)
z − z0
leading to the complex flow velocity
µ eiα
w(z) = . (IV.44b)
(z − z0 )2
y
6
α
-
x
Figure IV.10 – Streamlines for a flow dipole (IV.44a) centered on the origin.
This is clearly fully analogous to an electric dipole potential being the superposition of the potentials
created by electric charges +q and −q—and justifies the denomination “dipole flow”.
One can similarly define higher-order multipoles: flow quadrupoles, octupoles, . . . , for which
the order of the pole of the velocity at z0 increases (order 1 for a source or a sink, order 2 for a
dipole, order 3 for a quadrupole, and so on).
Remarks:
∗ The complex flow potentials considered until now—namely those of uniform flows (IV.41), sources
or sinks (IV.42a), pointlike vortices (IV.43a), and dipoles (IV.44a) or multipoles—and their super-
positions are the only two-dimensional flows valid on an unbounded domain.
As a matter of fact, demanding that the flow velocity ~v(~r) be uniform at infinity and that the
complex velocity w(z) be analytic except at a finite number of singularities—say only one, at z0 ,
to simplify the argumentation—, then w(z) may be expressed as a superposition of integer powers
of 1/(z − z0 ): ∞
X a−p
w(z) = , (IV.46a)
p=0
(z − z0 )p
since any positive power of (z − z0 ) would be unbounded when |z| → ∞. Integrating over z, see
Eq. (IV.40), the allowed complex potentials are of the form
∞
X p a−p−1
φ(z) = −a0 z − a−1 log(z − z0 ) + . (IV.46b)
p=1
(z − z0 )p
∗ Conversely, the reader can check—by computing the integral of w(z) along a contour at infinity—
that the total mass flow rate and circulation of the velocity field for a given flow are respectively
the real and imaginary parts of the residue a−1 in the Laurent series of its complex velocity w(z),
i.e. are entirely governed by the source/sink term (IV.42a) and vortex term (IV.43a) in the complex
potential.
∗ Eventually, the singularities that arise in the flow velocity will in practice not be a problem, since
these points will not be part of the physical flow, as we shall see on an example in § IV.4.3 c.
Figure IV.11 – Streamlines for the flow defined by potential (IV.47a) with from top to bottom
and from left to right n = 3, 32 , 1, 34 , 53 and 12 .
Except in the case n = 1, this potential cannot represent a flow on an unbounded domain, since
one easily checks that the velocity is unbounded as |z| goes to infinity. The interest of this potential
lies rather the behavior in the vicinity of z = z0 .
As a matter, writing down the flow potential and the stream function in a system of polar
coordinates centered on z0 ,
ϕ(r, θ) = A rn cos(nθ − α), ψ(r, θ) = A rn sin(nθ − α) (IV.47b)
shows that they both are (π/n)-periodic functions of the polar angle θ. Thus the flow on the
domain D delimited by the streamlines ψ(r, α) and ψ(r, α + π/n) is isolated from the motion in the
remainder of the complex plane. One may therefore assume that there are walls along these two
streamlines, and that the complex potential (IV.47a) describes a flow between them.
For n = 1, one recovers the uniform flow (IV.41)—in which we are free to put a wall along any
streamline, restricting the domain D to a half plane instead of the whole plane. If n > 1, π/n is
smaller than π and the domain D is comprised between a half-plane; in that case, the fluid motion
is a flow inside a corner. On the other hand, for 21 ≤ n < 1, π/n > π, so that the motion is a flow
past a corner.
The streamlines for the flows obtained with six different values for n are displayed in Fig. IV.11,
namely two flows in corners with angles π/3 and 2π/3, a uniform flow in the upper half plane, two
flows past corners with inner angles 2π/3 and π/3, and a flow past a flat plaque, corresponding
respectively to n = 3, 23 , 1, 43 , 35 and 21 .
IV.4.3 c Two-dimensional flows past a cylinder
:::::::::::::::::::::::::::::::::::::::::::::::
Thanks to the linearity of the Laplace differential equations, one may add “elementary” solutions
of the previous paragraph to obtain new solutions, which describe possible two-dimensional flows.
We now present two examples, which represent flows coming from infinity, where they are uniform,
and falling on a cylinder—either immobile or rotating around its axis.
Acyclic flow
Let us superpose the complex potentials for a uniform flow (IV.41) along the x-direction and a
flow dipole (IV.44a) situated at the origin and making an angle α = π with the vector ~ex :
R2
φ(z) = −v∞ z + , (IV.48a)
z
where the dipole strength µ was written as R2 v∞ . Adopting polar coordinates (r, θ), this ansatz
66 Non-relativistic flows of perfect fluids
y
6
-
x
Figure IV.12 – Streamlines for the acyclic potential flow past a cylinder (IV.48a).
where P ∞ denotes the pressure at infinity. That is, at each point on the surface of the cylinder
1
2 1
−~v(R, θ)2 = P ∞ + ρv∞
2
1 − 4 sin2 θ ,
P (R, θ) = P ∞ + ρ v∞
2 2
where the second identity follows from Eq. (IV.49b). The resulting stress vector on the vector at
a given θ is directed radially towards the cylinder center, T~s (R, θ) = −P (R, θ)~er (R, θ). Integrating
over θ ∈ [0, 2π], the total force on the cylinder due to the flowing fluid simply vanishes—in conflict
with the intuition—, phenomenon which is known as d’Alembert paradox .(w)
The intuition according to which the moving fluid should exert a force on the immobile obstacle
is good. What we find here is a failure of the perfect-fluid model, which is in that case too
idealized, by allowing the fluid to slip without friction along the obstacle.
Cyclic flow
To the flow profile which was just considered, we add a pointlike vortex (IV.43a) situated at the
origin
R2
iΓ z
φ(z) = −v∞ z + + log , (IV.50a)
z 2π R
where we have divided z by R in the logarithm to have a dimensionless argument, although this
plays no role for the velocity. Comparing with the acyclic flow, which models fluid motion around
a motionless cylinder, the complex potential may be seen as a model for the flow past a rotating
cylinder, as in the case of the Magnus effect (§ IV.2.2 d).
Adopting polar coordinates (r, θ), the velocity potential and stream function read
R2 R2
Γ Γ r
ϕ(r, θ) = −v∞ r + cos θ − θ, ψ(r, θ) = −v∞ r − sin θ + log , (IV.50b)
r 2π r 2π R
(w)
J. le Rond d’Alembert, 1717–1783
68 Non-relativistic flows of perfect fluids
Figure IV.13 – Streamlines for the cyclic potential flow past a (rotating) cylinder (IV.50a)
with Γ/(4πR v∞ ) = 0.25 (left) or 1 (right).
Figure IV.14 – Streamlines for the cyclic potential flow past a (rotating) cylinder (IV.50a)
with Γ/(4πR v∞ ) = 4.
Except in the trivial case n = 1, f (z) is singular at z = 0, where f 0 vanishes, so that the mapping
is non-conformal: cutting a half-line ending at z = 0, f maps the complex plane deprived from this
half-line onto an angular sector delimited by half-lines making an angle π/n—as already seen in
§ IV.4.3 b.
Joukowsky transform
A more interesting set of conformally deformed fluid flows consists of those provided by the use
of the Joukowsky transform (x)
R2
Z = f (z) = z + J (IV.52)
z
where RJ ∈ R.
The mapping (IV.52) is obviously holomorphic in the whole complex z-plane deprived of the
origin—which a single pole—, and has 2 points z = ±RJ at which f 0 vanishes. These two singular
points correspond in the Z-plane to algebraic branch points of the reciprocal function z = F (Z) at
Z = ±2RJ . To remove them, one introduces a branch cut along the line segment |X| ≤ 2RJ . On the
open domain U consisting complex Z-plane deprived from that line segment, F is holomorphic and
conformal. One checks that the cut line segment is precisely the image by f of the circle |z| = RJ in
the complex z-plane. Thus, f and F provide a bijective mapping between the exterior of the circle
|z| = RJ in the z-plane and the domain U in the Z-plane.
Another property of the Joukowsky transform is that the singular points z = ±RJ are zeros of
f 0 of order 1, so that angles are locally multiplied by 2. That is, every continuously differentiable
curve going through z = ±RJ is mapped by f on a curve through Z = ±2RJ with an angular point,
i.e. a discontinuous derivative, there.
Consider first the circle C (0, R) in the z-plane of radius R > RJ centered on the origin; it can
be parameterized as
C (0, R) = z = R eiϑ , 0 ≤ ϑ ≤ 2π .
Its image in the Z-plane by the Joukowsky transform (IV.52) is the set of points such that
RJ2 RJ2
Z = R+ cos ϑ + i R − sin ϑ, 0 ≤ ϑ ≤ 2π,
R R
that is, the ellipse centered on the origin Z = 0 with semi-major resp. semi-minor axis R + RJ2 /R
resp. R−RJ2 /R along the X- resp. Y -direction. Accordingly, the flows past a circular cylinder studied
in § IV.4.3 c can be deformed by f into flows past elliptical cylinders, where the angle between the
ellipse major axis and the flow velocity far from the cylinder may be chosen at will.
(x)
N. E. ukovski = N. E. Zhukovsky, 1847–1921
C HAPTER V
A large class of solutions of the equations of motion (III.9), (III.18) and (III.33) is that of waves.
Quite generically, this denomination designates “perturbations” of some “unperturbed” fluid motion,
which will also be referred to as background flow.
In more mathematical terms, the starting point is a set of fields {ρ0 (t,~r),~v0 (t,~r), P 0 (t,~r)} solving
the equations of motion, representing the background flow. The wave then consists of a second set
of fields {δρ(t,~r), δ~v(t,~r), δ P (t,~r)} which are added the background ones, such that the resulting
fields
ρ(t,~r) = ρ0 (t,~r) + δρ(t,~r), (V.1a)
P (t,~r) = P 0 (t,~r) + δ P (t,~r), (V.1b)
~v(t,~r) =~v0 (t,~r) + δ~v(t,~r) (V.1c)
are solutions to the equations of motion.
Different kinds of perturbations—triggered by some source which will not be specified hereafter,
and is thus to be seen as an initial condition—can be considered, leading to different phenomena.
A first distinction, with which the reader is probably already familiar, is that between traveling
waves, which propagate, and standing waves, which do not. Mathematically, in the former case the
propagating quantity does not depend on space and time independently, but rather on a combination
like (in a one-dimensional case) x − cϕ t, some propagation speed. In contrast, in standing waves
the space and time dependence of the “propagating” quantity factorize. Hereafter, we shall mostly
mention traveling waves.
Another difference is that between “small” and “large” perturbations or, in more technical terms,
between linear and nonlinear waves. In the former case, which is that of sound waves (Sec. V.1) or
the simplest gravity-controlled surface waves in liquids (Sec. V.3.1), the partial differential equation
governing the propagation of the wave is linear—which means that nonlinear terms have been
neglected. Quite obviously, nonlinearities of the dynamical equations—as e.g. the Euler equation—
are the main feature of nonlinear waves, as for instance shock waves (V.2) or solitons (Sec. V.3.2).
∂P
c2s ≡ (V.5)
∂ρ S,N
where both sides actually depend on ρ0 , S0 and N0 . One may then express δ P as function of δρ,
~ P (t,~r) by c2s ∇δρ(t,~
and replace ∇δ ~ r) in Eq. (V.4b).
The resulting equations for δρ(t,~r) and δ~v(t,~r) are linear first order partial differential equa-
tions. Thanks to the linearity, their solutions form a vector space—at least as long as no initial
condition has been specified. One may for instance express the solutions as Fourier transforms, i.e.
superpositions of plane waves. Accordingly, we test the ansatz
~
~k) e−iωt+ik·~r , ev(ω, ~k) e−iωt+i~k·~r ,
δρ(t,~r) = δρ(ω,
e δ~v(t,~r) = δ~ (V.6)
with respective amplitudes δρ, ev that a priori depend on ω and ~k and are determined by the initial
e δ~
conditions for the problem. In turn, ω and ~k are not necessarily independent from each other.
With this ansatz, Eqs. (V.4) become
−iω δρ(ω,
e ~k) + iρ0 ~k · δ~
ev(ω, ~k) = 0 (V.7a)
ev(ω, ~k) + ic2 ~k δρ(ω,
−iωρ0 δ~ e ~k) = 0. (V.7b)
s
From the second equation, the amplitude δ~ ev(ω, ~k) is proportional to ~k; in particular, it lies along
the same direction. That is, the inner product ~k · δ~ev simply equals the product of the norms of the
two vectors.
Omitting from now on the (ω, ~k)-dependence of the amplitudes, the inner product of Eq. (V.7b)
with ~k—which does not lead to any loss of information—allows one to recast the system as
! ! !
−ω ρ0 δρ
e 0
= .
c2s~k 2 −ωρ0 ~k · δ~
ev 0
A first, trivial solution to this system is δρ ev = ~0, i.e. the absence of any perturbation. In
e = 0, δ~
order for non-trivial solutions to exist, the determinant (ω 2 − c2s~k 2 )ρ0 of the system should vanish.
This leads at once to the dispersion relation
Denoting by ~e~k the unit vector in the direction of ~k, the perturbations δρ(t,~r) and δ~v(t,~r) defined
by Eq. (V.6), as well as δ P (t,~r) = c2s δρ(t,~r), are all functions of cs t ±~r ·~e~k . These are thus traveling
waves,(liv) , that propagate with the phase velocity ω(~k)/|~k| = cs , which is independent of ~k. That
is, cs is the speed of sound .
For instance, for air at T = 300 K, the speed of sound is cs = 347 m · s−1 .
Air is a diatomic ideal gas, i.e. it has pressure P = N kB T /V and internal energy U = 25 N kB T .
V 2 ∂P V2
∂P N kB T N kB ∂T
This then gives c2s = =− =− − + .
∂ρ S,N mN ∂V S,N mN V2 V ∂V S,N
(liv)
fortschreitende Wellen
V.1 Sound waves 73
The variation of the pressure δ P (t, x) can again be expressed in terms of the variation in the mass
density δρ(t, x) by invoking a Taylor expansion [cf. the paragraph between Eqs. (V.4) and (V.5)].
Since the perturbation of the background “flow” is no longer small, the thermodynamic state around
which this Taylor expansion is performed is not necessarily that corresponding to the unperturbed
fluid, but rather an arbitrary state, so that
where the speed of sound is that in the perturbed flow. When differentiating this identity, the
derivative of δρ(t, x) with respect to x is also the derivative of ρ(t, x), since the unperturbed fluid
state is uniform. Accordingly, one may recast Eqs. (V.10) as
∂ρ(t, x) ∂δv(t, x) ∂ρ(t, x)
+ ρ(t, x) + δv(t, x) = 0, (V.11a)
∂t ∂x ∂x
∂δv(t, x) ∂δv(t, x) ∂ρ(t, x)
ρ(t, x) + δv(t, x) + cs (ρ)2 = 0, (V.11b)
∂t ∂x ∂x
which constitutes a system of two coupled partial differential equations for the two unknown fields
ρ(t, x) and δv(t, x) = v(t, x).
To tackle these equations, one may assume that the mass density and the flow velocity have
parallel dependences on time and space—as suggested by the fact that this property holds in the
linearized case of sound waves, in which both ρ(t,~r) and ~v(t,~r) propagate with the same phase
(cs |~k|t +~k · ~r). Thus, the dependence of v on t and x is replaced with a functional dependence
v ρ(t, x) , with the known value v(ρ0 ) = 0 corresponding to the unperturbed fluid at rest. Accord-
ingly, the partial derivatives of the flow velocity with respect to t resp. x become
∂v(t, x) dv(ρ) ∂ρ(t, x) ∂v(t, x) dv(ρ) ∂ρ(t, x)
= resp. = .
∂t dρ ∂t ∂x dρ ∂x
The latter identities may then be inserted in Eqs. (V.11). If one further multiplies Eq. (V.11a) by
(lvi)
Stoßwelle
76 Waves in non-relativistic perfect fluids
ρ(t, x) dv(ρ)/dρ and then subtracts Eq. (V.11b) from the result, there comes
2
2 dv(ρ) 2 ∂ρ(t, x)
ρ − cs (ρ) = 0,
dρ ∂x
that is, discarding the trivial solution of a uniform mass density,
dv(ρ) cs (ρ)
=± . (V.12)
dρ ρ
Under the simultaneous replacements v → −v, x → −x, cs → −cs , equations (V.11)-(V.12)
remain invariant. Accordingly, one may restrict the discussion of Eq. (V.12) to the case with a
+ sign—the − case amounts to considering a wave propagating in the opposite direction with the
opposite velocity. The flow velocity is then formally given by
cs (ρ0 ) 0
Z ρ
v(ρ) = 0
dρ ,
ρ0 ρ
t0
-
x
t1 > t0
-
x
t2 > t1
-
x
t3 > t2
-
x
t4 > t3
-
x
Figure V.1 – Schematic representation of the evolution in time of the spatial distribution of
dense and rarefied regions leading to a shock wave.
V.2 Shock waves 77
where the momentum flux density tensor has components T ij = P g ij + ρ vi vj [see Eq. (III.21b)],
with g ij = δ ij in the case of Cartesian coordinates.
The continuity of the mass flux density across the surface of discontinuity (V.15a) can be recast
as
(ρv1 )− = (ρv1 )+ ≡ j1 . (V.16)
A first, trivial solution arises if there is no flow of matter across surface Sd , i.e. if (v1 )+ = (v1 )−
= 0.
In that case, Eq. (V.15c) is automatically satisfied. Condition (V.15b) for i = 1 becomes P = 0,
i.e. the pressure is the same on both sides of Sd . Eventually, Eq. (V.15b) with i = 2 or 3 holds
automatically. All in all, there is no condition on the behavior of ρ, v2 or v3 across the surface of
discontinuity—which means that these quantities may be continuous or not, in the latter case with
an arbitrary jump.
If j1 does not vanish, that is if matter does flow across Sd , then
the jump equation for the
component T 21 = ρv2 v1 resp. T 31 = ρv3 v1 leads to v2 = 0 resp. v3 = 0, i.e. the component v2
In turn, rewriting the jump equation for T 11 = P + ρ(v1 )2 with the help of j1 yields
1 1
2 1 1
P − − P + = j1 (v )+ − (v )− = j1 − . (V.18)
ρ+ ρ−
(lvii) (lviii) (lix)
Unstetigkeitsfläche Sprung Sprunggleichungen
78 Waves in non-relativistic perfect fluids
Thus if ρ+ < ρ− , i.e. if the fluid is denser in the (−)-region “behind” the shock front—as is suggested
by Fig. V.1, yet still needs to be proved—, then P − > P + , while relation (V.16) yields (v1 )+ > (v1 )− .
Conversely, ρ+ > ρ− leads to P − < P + and (v1 )+ < (v1 )− . One can show that the former case
actually holds.
Combining Eqs. (V.16) and (V.18) yields
2 j12 P − − P + ρ− ρ+ P − − P + ρ−
(v1 )+
= = =
ρ2+ ρ− − ρ+ ρ2+ ρ− − ρ+ ρ+
and similarly
1 2 P − − P + ρ+
(v )− = .
ρ− − ρ+ ρ−
If the jumps in pressure and mass density are small, one can show that their ratio is approximately
the derivative ∂ P /∂ρ, here at constant entropy and particle number, i.e.
∂P
1 2 ρ− ρ− 2 1 2 ρ+ 2
(v )+ ' = cs , (v )− ' c .
∂ρ S,N ρ+ ρ+ ρ− s
With ρ− > ρ+ comes (v1 )+ > cs resp. (v1 )− < cs in front of resp. behind the shock wave.(26) The
former identity means that an observer comoving with the surface of discontinuity sees in front a
fluid flowing with a supersonic velocity, that is, going temporarily back to a reference frame bound
to the unperturbed fluid, the shock wave moves with a supersonic velocity.
Invoking the continuity across Sd of the product ρ v1 and of the components v2 , v3 parallel to
the surface of discontinuity, the jump equation (V.15c) for the energy flux density simplifies to
1 1 2 e+P j12 1 e+ + P + e− + P −
1
(v ) + = 2 − 2 + − = 0.
2 ρ 2 ρ+ ρ− ρ+ ρ−
(26)
Here we are being a little sloppy: one should consider the right (x1 → 0+ ) and left (x1 → 0− ) derivatives,
corresponding respectively to the (+) and (−)-regions, and thus find the associated speeds of sound (cs )+ and
(cs )− instead of a single cs .
(lx)
dynamische Adiabate
(y) (z)
P. H. Hugoniot, 1851–1887 W. J. M. Rankine, 1820–1872
V.3 Gravity waves 79
∂2 ∂2
+ ϕ(t, x, z) = 0. (V.20b)
∂x2 ∂z 2
To fully specify the problem, boundary conditions are still needed. As in the generic case for
potential flow (Sec. IV.4), these will be Neumann boundary conditions, involving the derivative of
the velocity potential.
• At the bottom of the sea, the water can have no vertical motion, corresponding to the usual
impermeability condition; that is
∂ϕ
vz (z = 0) = − = 0. (V.21a)
∂z z=0
(lxi)
Schwerewellen
80 Waves in non-relativistic perfect fluids
• At the free surface of the sea, the vertical component vz of the flow velocity equals the velocity
of the surface, i.e. it equals the rate of change of the position of the (material!) surface:
∂ϕ(t, x, z) Dδh(t, x)
− = .
∂z
z=h0 +δh(t,x) Dt
D ∂ ∂ ∂ ∂ϕ ∂
Using = + vx = − , this gives
Dt ∂t ∂x ∂t ∂x ∂x
∂ϕ(t, x, z) ∂δh(t, x) ∂δh(t, x) ∂ϕ(t, x, z)
+ − = 0. (V.21b)
∂z ∂t ∂x ∂x z=h0 +δh(t,x)
• At the free surface of the sea, the pressure on the water side—right below the surface—is
directly related to that just above the surface. The latter is assumed to be constant and equal
at some value P 0 , which represents for instance the atmospheric pressure “at sea level”. As
a first approximation—whose physical content will be discussed in the remark at the end of
this paragraph—, the pressure is equals on both sides of the sea surface:
P t, x, z = h0 +δh(t, x) = P 0 . (V.21c)
Expressing the pressure with the help of Eq. (V.20a), this condition may be recast as
" 2 #
~
∂ϕ(t, x, z) ∇ϕ(t, x, z) P0
− + + g δh(t, x) = − − gh0 + constant, (V.21d)
∂t 2 ρ
z=h0 +δh(t,x)
where the whole right hand side of the equation represents a new constant.
Hereafter we look for solutions consisting of a velocity potential ϕ(t, x, z) and a surface profile
δh(t, x), as determined by Eqs. (V.20) with conditions (V.21).
Remark: The assumption of an identical pressure on both sides of an interface—either between two
immiscible liquids, or between a liquid and a gas, as here—is generally not warranted, unless the
interface happens to be flat. If there is the least curvature, the surface tension associated with the
interface will lead to a larger pressure inside the concavity of the interface. Neglecting this effect—
which we shall consider again in Sec. V.3.2—is valid only if the typical radius of curvature of the
interface, which as we shall see below is the wavelength of the surface waves, is “large”, especially
with respect to the deformation scale δh.
V.3.1 b Harmonic wave assumption
::::::::::::::::::::::::::::::::::::
Since the domain on which the wave propagates is unbounded, a natural ansatz for the solution
of the Laplace equation (V.20b) is that of a harmonic wave
ϕ(t, x, z) = f (z) cos(kx − ωt) (V.22)
propagating in the x-direction with a depth-dependent amplitude f (z). Inserting this form in the
Laplace equation yields the linear ordinary differential equation
d2f (z)
− k 2 f (z) = 0,
dz 2
whose obvious solution is f (z) = a1 ekz + a2 e−kz , with a1 and a2 two real constants.
The boundary condition (V.21a) at the sea bottom z = 0 gives a1 = a2 , i.e.
ϕ(t, x, z) = C cosh(kz) cos(kx − ωt), (V.23)
with C a real constant.
To make further progress with the equations of the system, and in particular to determine the
profile of the free surface, further assumptions are needed, so as to obtain simpler equations. We shall
now present a first such simplification, leading to linear waves. In Sec. V.3.2, another simplification—
of a more complicated started point—will be considered, which gives rise to (analytically tractable!)
nonlinear waves.
V.3 Gravity waves 81
and
∂ϕ(t, x, z)
− + g δh(t, x) = constant, (V.25b)
∂t
z=h0
respectively. Together with the Laplace differential equation (V.20b) and the boundary condition
at the sea bottom (V.21a), the two equations (V.25) constitute the basis of the Airy(aa) linear wave
theory.
Combining the latter two equations yields at once the condition
2
∂ ϕ(t, x, z) ∂ϕ(t, x, z)
+g = 0.
∂t2 ∂z z=h0
• For kh0 1, i.e. in the case of a shallow sea with h0 λ, the approximation tanh(kh0 ) ' kh0
leads to the dispersion relation ω 2 = gh0 k 2 , i.e. to phase and group velocities
p
cϕ = cg = gh0 ,
independent from the wavelength λ, signaling the absence of dispersive behavior.
This phase velocity decreases with decreasing water depth h0 . Accordingly, this might lead to
an accumulation, similar to the case of a shock wave in Sec. V.2, whose description however
requires that one take into account the nonlinear terms in the equations, which have been
discarded here. In particular, we have explicitly assumed |δh(t, x)| h0 , in order to linearize
the problem, so that considering the limiting case h0 → 0 is questionable.
In addition, a temptation when investigating the small-depth behavior h0 → 0 is clearly
to describe the breaking of waves as they come to shore. Yet the harmonic ansatz (V.23)
assumes that the Laplace equation is considered on a horizontally unbounded domain, i.e.
far from any coast, so again the dispersion relation (V.26) may actually no longer be valid.
The boundary condition (V.25b) provides us directly with the shape of the free surface of the
sea, namely
1 ∂ϕ(t, x, z) ωC
δh(t, x) = = cosh(kh0 ) sin(kx − ωt) ≡ δh0 sin(kx − ωt),
g ∂t
z=h0 g
with δh0 ≡ (ωC/g) cosh(kh0 ) the amplitude of the wave, which must remain much smaller than h0 .
The profile of the surface waves of Airy’s linear theory—or rather its cross section—is thus a simple
sinusoidal curve.
This shape automatically suggests a generalization, which is a first step towards taking into
account nonlinearities, such that the free surface profile is sum of (a few) harmonics sin(kx−ωt),
sin 2(kx − ωt), sin 3(kx − ωt). . . The approach leading to such a systematically expanded profile,
which relies on a perturbative expansion to deal with the (still small) nonlinearities, is that of
the Stokes waves.
The gradient of the potential (V.23) yields (the components of) the flow velocity
kg cosh(kz)
vx (t, x, z) = δh0 sin(kx − ωt),
ω cosh(kh0 )
kg sinh(kz)
vz (t, x, z) = − δh0 cos(kx − ωt).
ω cosh(kh0 )
Integrating these functions with respect to time leads to the two functions
kg δh0 cosh(kz) δh0 cosh(kz)
x(t) = x0 + cos(kx − ωt) = x0 + cos(kx − ωt),
ω 2 cosh(kh0 ) sinh(kh0 )
kg δh0 sinh(kz) δh0 sinh(kz)
z(t) = z0 + 2
sin(kx − ωt) = z0 + sin(kx − ωt),
ω cosh(kh0 ) sinh(kh0 )
with x0 and z0 two integration constants. Choosing x0 ' x and z0 ' z, if δh0 k −1 , these functions
represent the components of the trajectory (pathline) of a fluid particle that is at time t in the vicinity
of the point with coordinates (x, z), and whose velocity at that time is thus approximately the flow
velocity ~v(t, x, z). Since
2 2
[x(t) − x0 ]2 [z(t) − z0 ]2
kg δh0 δh0
+ = = ,
cosh2 (kz) sinh2 (kz) ω 2 cosh(kh0 ) sinh(kh0 )
this trajectory is an ellipse, whose major and minor axes decrease with increasing depth h0 − z.
In the deep sea case kh0 1, one can use the approximations sinh(kz) ' cosh(kz) ' ekz /2 for
1 kz . kh0 , which shows that the pathlines close to the sea surface are approximately circles.
V.3 Gravity waves 83
Eventually, the pressure distribution in the sea follows from Eq. (V.24) in which one uses the
velocity potential (V.23), resulting in
∂ϕ(t, x, z) cosh(kz)
P (t, x, z) = P 0 + ρg(h0 − z) + ρ = P 0 + ρg h0 − z + δh0 sin(kx − ωt) .
∂t cosh(kh0 )
The contribution P 0 + ρg(h0 − z) is the usual hydrostatic one, corresponding to the unperturbed
sea, while the effect of the surface wave is proportional to its amplitude δh0 and decreases with
increasing depth.
Let us now recast Eqs. (V.27) in a dimensionless form. For that extent, we introduce two
characteristic lengths: Lc for long-wavelength motions along x or z, and δhc for the amplitude of
the surface deformation; for durations, we define a scale tc , which will later be related to Lc with
the help of a typical velocity. With these scales, we can construct dimensionless variables
t x z
t∗ ≡ , x∗ ≡ , z∗ ≡ ,
tc Lc Lc
(28)
This Section follows closely the Appendix A of Ref. [18].
84 Waves in non-relativistic perfect fluids
and fields:
δh vx vz P dyn.
δh∗ ≡ , vx∗ ≡ , vz∗ ≡ , P∗ ≡ .
δhc δhc /tc δhc /tc ρ δhc Lc /t2c
Considering the latter as functions of the reduced variables t∗ , x∗ , z ∗ , one can rewrite the
equations (V.27a)–(V.27e). The incompressibility thus becomes
∂ vx∗ ∂ vz∗
+ = 0, (V.28a)
∂x∗ ∂z ∗
and the Euler equation, projected successively on the x and z directions
∂ vx∗ ∗ ∗ ∂P ∗
∗ ∂ vx ∗ ∂ vx
+ ε vx + vz = − , (V.28b)
∂t∗ ∂x∗ ∂z ∗ ∂x∗
and
∂ vz∗ ∗ ∗ ∂P ∗
∗ ∂ vz ∗ ∂ vz
+ ε v x + vz = − . (V.28c)
∂t∗ ∂x∗ ∂z ∗ ∂z ∗
where we have introduced the dimensionless parameter ε ≡ δhc /Lc . In turn, the various boundary
conditions are
vz∗ = 0 at z ∗ = 0 (V.28d)
at the sea bottom, and at the free surface
∂δh∗ ∗ ∂δh
∗
vz∗ = + εvx at z ∗ = δ + εδh∗ (V.28e)
∂t∗ ∂x∗
with δ ≡ h0 /Lc , and
gt2c ∗
P∗ = δh at z ∗ = δ + εδh∗ .
Lc
Introducing the further dimensionless number
p
Lc /g
Fr ≡
tc
the latter condition becomes
1
P∗ = 2 δh
∗
at z ∗ = δ + εδh∗ (V.28f)
Fr
Inspecting these equations, one sees that the parameter ε controls the size of nonlinearities—cf.
Eqs. (V.28b), (V.28c) and (V.28e)—, while δ measures the depth of the sea in comparison to the
typical wavelength Lc . Both parameters are a priori independent: δ is given by the physical setup
we want to describe, while ε quantifies the amount of nonlinearity we include in the description.
To make progress, we shall from now on focus on gravity waves on shallow water, i.e. assume
δ 1. In addition, we shall only consider small nonlinearities, ε 1. To write down expansions
in a consistent manner, we shall assume that the two small parameters are not of the same order,
but rather that they obey ε ∼ δ 2 . Calculations will be considered up to order O(δ 3 ) or equivalently
O(δε).
For the sake of brevity, we now drop the subscript ∗ from the dimensionless variables and fields.
V.3.2 b Velocity potential
:::::::::::::::::::::::::
If the flow is irrotational, ∂ vx /∂z = ∂ vz /∂x, so that one may transform Eq. (V.28b) into
∂ vx ∂ vx ∂ vz 1 ∂δh
+ ε vx + vz + 2 = 0. (V.29)
∂t ∂x ∂x Fr ∂x
~
In addition, one may introduce a velocity potential ϕ(t, x, z) such that ~v = −∇ϕ. With the latter,
V.3 Gravity waves 85
with unknown functions ϕn (t, x). Substituting this ansatz in the Laplace equation (V.30) gives after
some straightforward algebra
∞ 2
n ∂ ϕn (t, x)
X
z + (n + 1)(n + 2)ϕn+2 (t, x) = 0.
∂x2
n=0
In order for this identity to hold for arbitrary z—at least, for the values relevant for the flow—,
each coefficient should individually vanish, i.e. the ϕn should obey the recursion relation
1 ∂ 2 ϕn (t, x)
ϕn+2 (t, x) = − for n ∈ N. (V.32)
(n + 1)(n + 2) ∂x2
It is thus only necessary to determine ϕ0 and ϕ1 to know the whole series.
The boundary condition (V.28d) at the bottom reads ∂ϕ(t, x, z = 0)/∂z = 0 for all t and x,
which implies ϕ1 (t, x) = 0, so that all ϕ2n+1 identically vanish. As a consequence, ansatz (V.31)
with the recursion relation (V.32) give
z 2 ∂ 2 ϕ0 (t, x) z 4 ∂ 4 ϕ0 (t, x)
ϕ(t, x, z) = ϕ0 (t, x) − + + ...
2 ∂x2 4! ∂x4
~
Differentiating with respect to x or z yields the components of the velocity ~v = −∇ϕ
∂ϕ0 (t, x) z 2 ∂ 3 ϕ0 (t, x) z 4 ∂ 5 ϕ0 (t, x)
vx (t, x, z) = − + − + ...
∂x 2 ∂x3 4! ∂x5
∂ 2 ϕ0 (t, x) z 3 ∂ 4 ϕ0 (t, x)
vz (t, x, z) = z − + ...
∂x2 3! ∂x4
Introducing the notation u(t, x) ≡ −∂ϕ0 (t, x)/∂x and anticipating that the maximal value of z
relevant for the problem is of order δ, these components may be expressed as
z 2 ∂ 2 u(t, x)
vx (t, x, z) = u(t, x) − + o (δ 3 ), (V.33a)
2 ∂x2
∂u(t, x) z 3 ∂ 3 u(t, x)
vz (t, x, z) = −z + + o (δ 3 ), (V.33b)
∂x 3! ∂x3
where the omitted terms are beyond O(δ 3 ).
Linear waves rediscovered
If we momentarily set ε = 0—which amounts to linearizing the equations of motion and boundary
conditions—, consistency requires that we consider equations up to order δ at most. That is, we
keep only the first terms from Eqs. (V.33): at the surface at z ' δ, they become
∂u(t, x)
vx (t, x, z = δ) ' u(t, x), vz (t, x, z = δ) ' −δ , (V.34a)
∂x
while the boundary condition (V.28e) simplifies to
∂δh(t, x) ∂φ(t, x)
vz (t, x, z = δ) = =δ , (V.34b)
∂t ∂t
where we have introduced φ(t, x) ≡ δh(t, x)/δ.
86 Waves in non-relativistic perfect fluids
∂φ(t, x) + ∂u(t, x) = 0,
∂t ∂x
which admits the solution u(t, x) = φ(t, x) under the condition
∂u(t, x) ∂u(t, x)
+ = 0, (V.40)
∂t ∂x
which describes a traveling wave with (dimensionless) velocity 1, u(t, x) = u(x−t). We again recover
the linear sea surface waves which we have already encountered twice.
V.3 Gravity waves 87
Going to next-to-leading order O(δ 2 ), O(ε), we look for solutions in the form
with φ, u(ε) , u(δ) functions that obey condition (V.40) up to terms of order ε or δ 2 . Inserting this
ansatz in Eqs. (V.38)–(V.39) yields the system
(ε) (δ) 2 3
∂φ + ∂φ + ε ∂u + δ 2 ∂u + 2εφ ∂φ − δ ∂ φ = 0
∂t ∂x ∂x ∂x ∂x 6 ∂x3
(ε) (δ) 2 3
∂φ + ∂φ + ε ∂u + δ 2 ∂u + εφ ∂φ − δ ∂ φ = 0,
∂t ∂x ∂t ∂t ∂x 2 ∂x2 ∂t
where for the sake of brevity, the (t, x)-dependence of the functions was not written. Subtracting
both equations and using condition (V.40) to relate the time and space derivatives of φ, u(ε) , and
u(δ) , one finds
(ε) (δ)
1 ∂ 3 φ(t, x)
∂u (t, x) 1 ∂φ(t, x) 2 ∂u (t, x)
ε + φ(t, x) +δ − = 0.
∂x 2 ∂x ∂x 3 ∂x3
Since the two small parameters ε and δ are independent, each term between square brackets in this
identity must identically vanish. Straightforward integrations then yield
1 1 ∂ 2 φ(t, x)
u(ε)(t, x) = − φ(t, x) + C (ε)(t), u(δ)(t, x) = + C (δ)(t),
4 3 ∂x2
with C (ε) , C (δ) two functions of time only.
These functions can then be substituted in the ansatz (V.41). Inserting the latter in Eq. (V.39)
yields an equation involving the unknown function φ only, namely
∂φ(t, x) ∂φ(t, x) 3 ∂φ(t, x) 1 2 ∂ 3 φ(t, x)
+ + εφ(t, x) + δ = 0. (V.42)
∂t ∂x 2 ∂x 6 ∂x3
The first two terms only are those of the linear-wave equation of motion (V.40). Since the ε and
δ nonlinear corrections also obey the same condition, it is fruitful to perform a change of variables
from (t, x) to (τ, ξ) with τ ≡ t, ξ ≡ x − t. Equation (V.42) then becomes
Solitary waves
The Korteweg–de Vries equation admits many different solutions. Among those, there is the class
of solitary waves or solitons, which describe signals that propagate without changing their shape.
(ab) (ac)
D. Korteweg, 1848–1941 G. de Vries, 1866–1934
88 Waves in non-relativistic perfect fluids
A specific subclass of solitons of the KdV equation of special interest in fluid dynamics consists
of those which at each given instant vanish at (spatial) infinity. As solutions of the normalized
equation (V.44), they read
φ0
φ(τ, ξ) = 2
p (V.45a)
cosh φ0 /2 (ξ − 2φ0 τ)
with φ0 the amplitude of the wave. Note that φ0 must be nonnegative, which means that these
solutions describe bumps above the mean sea level—which is the only instance of such solitary wave
observed experimentally. Going back first to the variables (τ, ξ), then to the dimensionless variables
(t∗ , x∗ ), and eventually to the dimensionful variables (t, x) and field δh, the soliton solution reads
δhmax
δh(t, x) = r , (V.45b)
√
2 1 3δhmax δhmax
cosh x − gh0 1+ t
2h0 h0 2h0
with δhmax the maximum amplitude of the solitary wave. This solution, represented in Fig. V.2,
has a few properties that can be read directly off its expression and differ from those of linear sea
surface waves, namely
• the propagation velocity csoliton of the soliton—which is the factor in front of t—is larger than
for linear waves;
• the velocity csoliton increases with the amplitude δhmax of the soliton;
• the width of the soliton decreases with its amplitude.
δh(t, x)
δhmax = 1, t = t0
δhmax = 0.25, t = t0
δhmax = 1, t = t1 > t0
δhmax = 0.25, t = t1
x
Figure V.2 – Profile of the soliton solution (V.45).
The dynamics of Newtonian fluids is entirely governed by a relatively simple set of equations,
namely the continuity equation (III.9), the Navier–Stokes equation (III.31), and—when phenomena
related with temperature gradients become relevant—the energy conservation equation (III.35). As
in the case of perfect fluids, there are a priori more unknown dynamical fields than equations, so
that an additional relation has to be provided, either a kinematic constraint or an equation of
state. In this Chapter and the next two ones, a number of simple solutions of these equations
are presented, together with big classes of phenomena that are accounted in various more or less
simplified situations.
With the exception of the static-fluid case, in which the only novelty with respect to the hydro-
statics of perfect fluids is precisely the possible transport of energy by heat conduction (Sec. VI.1.1),
the motions of interest in the present Chapter are mostly laminar flows in which viscous effects play
an important role while heat transport is negligible. Thus, the role of the no-slip condition at a
boundary of the fluid is illustrated with a few chosen examples of stationary motions within idealized
geometrical setups (Sec. VI.1).
By introducing flow-specific characteristic length and velocity scales, the Navier–Stokes equa-
tion can be rewritten in a form involving only dimensionless variables and fields, together with
parameters—like for instance the Reynolds number. These parameters quantify the relative impor-
tance of the several physical effects likely to play a role in a motion (Sec. VI.2).
According to the value of the dimensionless numbers entering the dynamical equations, the
latter may possibly be simplified. This leads to simpler equations with limited domain of validity,
yet which become more easily tractable, as exemplified by the case of flows in which shear viscous
effects predominate over the influence of inertia (Sec. VI.3). Another simplified set of equations can
be derived to describe the fluid motion in the thin layer close to a boundary of the flow, in which
the influence of this boundary plays a significant role (Sec VI.4).
Eventually, the viscosity-induced modifications to the dynamics of vorticity (Sec. VI.5) and to
the propagation of sound waves (Sec. VI.6) are presented.
The three coupled equations (III.9), (III.31) and (III.35) respectively simplify to
∂ρ(t,~r)
= 0, (VI.1a)
∂t
from where follows the time independence of the mass density ρ(t,~r),
~ P (t,~r) = −ρ(t,~r) ∇Φ(t,~
∇ ~ r), (VI.1b)
similar to the fundamental equation (IV.2) governing the hydrostatics of a perfect fluid, and
∂e(t,~r) ~ ~ (t,~r) ,
= ∇ · κ(t,~r)∇T (VI.1c)
∂t
which describes the transport of energy without macroscopic fluid motion, i.e. non-convectively,
thanks to heat conduction.
~ ~v(~r) = − 1 ∇
~ P (~r) + ν4~v(~r),
~v(~r) · ∇ (VI.2b)
ρ
with ν the kinematic shear viscosity, assumed to be the same throughout the fluid.
The so-called (plane) Couette flow(ad) is, in its idealized version, the motion of a viscous fluid
between two infinitely extended plane plates, as represented in Fig. VI.1, where the lower plate is
at rest, while the upper one moves in its own plane with a constant velocity ~u. It will be assumed
- ~u-
6 -
h - ~v(y)
- y
? 6
-
x
Figure VI.1 – Setup of the plane Couette flow.
∂ P (~r)/∂z = 0—since the problem is independent of z. Along the x direction, one finds
∂ P (~r) d2 v(y)
=η . (VI.4)
∂x dy 2
Since the right member of this equation is independent of x and z, a straightforward integration gives
P (~r) = α(y)x + β(y), where the functions α, β only depend on y. These functions are determined
by the boundary conditions: since P (x = −∞) = P (x = ∞) = P ∞ , then α(y) = 0, β(y) = P ∞ , and
Eq. (VI.4) simplifies to
d2 v(y)
= 0.
dy 2
This yields v(y) = γy + δ, with γ and δ two integration constants, which are again fixed by the
boundary conditions. At each plate, the relative velocity of the fluid with respect to the plate must
vanish:
v(y = 0) = 0, v(y = h) = |~u|,
leading to δ = 0 and γ = |~u|/h. All in all, the velocity thus depends linearly on y
y
~v(~r) = ~u for 0 ≤ y ≤ h.
h
Consider now a surface element d2 S. The contact force d2 F~s exerted on it by the fluid follows
from the Cauchy stress tensor, whose Cartesian components (III.27c) here read
|~u|
−P η 0
∂ v (~r) ∂ vj (~r) ∼ |~u|∞
i
h
σ ij (~r) = −P (~r)δ ij + η + = ηh −P ∞ 0 .
∂xj ∂xi
0 0 −P ∞
The force per unit surface on the motionless plate at y = 0, corresponding to a unit normal vector
~en (~r) = ~ey , is
|~u|
2 ~
d Fs (~r)
X 3 X 3 ηh
~s (~r) = ij ij
= T σ (~
r)~
e i ⊗ ~
e j · ~
e y = σ (~
r) ~
e j · ~
e ~
y ie = −P ∞ .
d2 S
i,j=1 i,j=1 0
Due to the friction exerted by the fluid, the lower plate is dragged by the flow in the (positive) x
direction.
Remark: The tangential stress on the lower plate is η~u/h, proportional to the shear viscosity:
measuring the tangential stress with known |~u| and h yields a measurement of η. In practice,
this measurement rather involves the more realistic cylindrical analog to the above plane flow, the
so-called Couette–Taylor flow .(ae)
--
6 -
P1 h -
- P2 y
? -- 6
-
x
-
L
Figure VI.2 – Flow between two motionless plates for P 1 > P 2 , i.e. ∆P > 0.
v(y = 0) = 0, v(y = h) = 0,
1 ∆P
which leads to δ = 0 and γ = h. The flow velocity thus has the parabolic profile
2η L
1 ∆P
v(y) = y(h − y) for 0 ≤ y ≤ h, (VI.5)
2η L
directed along the direction of the pressure gradient.
Remark: The flow velocity (VI.5) becomes clearly problematic in the limit η → 0! Tracing the
problem back to its source, the equations of motion (VI.3) cannot hold with a finite gradient along
the x direction and a vanishing viscosity. One quickly checks that the only possibility in the case
of a perfect fluid is to drop one of the assumptions, either incompressibility or laminarity.
a
P 1 P 2 -z
-
L
Figure VI.3 – Setup of the Hagen–Poiseuille flow.
p
Using cylindrical coordinates, the ansatz ~v(~r) = v(r)~ez with r = x2 + y 2 satisfies the conti-
nuity equation ∇~ ·~v(~r) = 0 and gives for the incompressible Navier–Stokes equation
∂ P (~r) ∂ P (~r)
∂x = ∂y = 0
~ P (~r) = η4~v(~r) ⇔
∇ 2 (VI.6)
∂ v(r) ∂ 2 v(r))
2
∂ P (~r)
d v(r) 1 dv(r)
=η + = η + .
∂z ∂x2 ∂y 2 dr2 r dr
(af)
G. Hagen, 1797–1884
VI.1 Statics and steady laminar flows of a Newtonian fluid 93
The right member of the equation in the second line is independent of z, implying that the pressure
gradient along the z direction is constant:
∂ P (~r) ∆P
=− ,
∂z L
with ∆P ≡ P 1 − P 2 . The z component of the Navier–Stokes equation (VI.6) thus becomes
d2 v(r) 1 dv ∆P
+ =− . (VI.7)
dr2 r dr ηL
As always, this linear differential equation is solved in two successive steps, starting with the
associated homogeneous equation. To find the general solution of the latter, one may introduce
χ(r) ≡ dv(r)/dr, which satisfies the simpler equation
dχ(r) χ(r)
+ = 0.
dr r
The generic solution is ln χ(r) = − ln r + const., i.e. χ(r) = A/r with A a constant. This then leads
to v(r) = A ln r + B with B an additional constant.
A particular solution of the inhomogeneous equation (VI.7) is v(r) = Cr2 with C = −∆P /4ηL.
The general solution of Eq. (VI.7) is then given by
∆P 2
v(r) = A ln r + B − r ,
4ηL
where the two integration constants still need to be determined.
To have a regular flow velocity at r = 0, the constant A should vanish. In turn, the boundary
condition at the tube wall, v(r = a) = 0, determines the value of the constant B = (∆P /4ηL)a2 .
All in all, the velocity profile thus reads
∆P 2
a − r2 for r ≤ a.
v(r) = (VI.8)
4ηL
This is again parabolic, with ~v in the same direction as the pressure drop.
The mass flow rate across the tube cross section follows from a straightforward integration:
Z a
∆P a 2 ∆P a4 πρa4 ∆P
Z
a r − r3 dr = 2πρ
Q= ρv(r) 2πr dr = 2πρ = . (VI.9)
0 4ηL 0 4ηL 4 8η L
This result is known as Hagen–Poiseuille law (or equation) and means that the mass flow rate is
proportional to the pressure drop per unit length.
Remarks:
∗ The Hagen–Poiseuille law only holds under the assumption that the flow velocity vanishes at the
tube walls. The experimental confirmation of the law—which was actually deduced from experiment
by Hagen (1839) and Poiseuille (1840)—is thus a proof of the validity of the no-slip assumption for
the boundary condition.
∗ The mass flow rate across the tube cross section may be used to define that average flow velocity
as Q = πa2 ρhvi with Z a
1 1
hvi ≡ v(r) 2πr dr = v(r = 0).
πa2 0 2
The Hagen–Poiseuille law then expresses a proportionality between the pressure drop per unit length
and hvi in a laminar flow.
Viewing ∆P /L as the “generalized force” driving the motion, the corresponding “response” hvi of
the fluid is thus linear.
The relation is quite different in the case of a turbulent flow with the same geometry: for instance,
measurements by Reynolds [19] gave ∆P /L ∝ hvi1.722 .
94 Non-relativistic dissipative flows
This number measures the relative importance of inertia and viscous friction forces on a fluid element
or a body immersed in the moving fluid: at large resp. small Re, viscous effects are negligible resp.
predominant.
Remark: As stated above Eq. (VI.10), both Lc and vc are controlled by the geometry and boundary
conditions. The Reynolds number—and every similar dimensionless we shall introduce hereafter—is
thus a characteristic of a given flow, not of the fluid.
Law of similitude(lxii)
:::::::::::::::::
The solutions for the dynamical fields ~v∗ , P ∗ at fixed boundary conditions and geometry—
specified in terms of dimensionless ratios of geometrical lengths—are functions of the independent
variables t∗ , ~r∗ , and of the Reynolds number:
with ~f1 resp. f2 a vector resp. scalar function. Flow velocity and pressure are then given by
(lxii)
Ähnlichkeitsgesetz
VI.2 Dynamical similarity 95
~ vc t ~r vc t ~r
~v(t,~r) = vc f1 , , Re , P (t,~r) = P 0 + ρvc2 f2 , , Re .
Lc Lc Lc Lc
These equations underlie the use of fluid dynamical simulations with experimental models at a
reduced scale, yet possessing the same (rescaled) geometry. Let Lc , vc resp. LM , vM be the charac-
teristic lengths of the real-size flow resp. of the reduced-scale experimental flow; for simplicity, we
assume that the same fluid is used in both cases. If vM /vc = Lc /LM , the Reynolds number for the
experimental model is the same as for the real-size fluid motion: both flows then admit the same
solutions ~v∗ and P ∗ , and are said to be dynamically similar .
Remark: The functional relationships between the “dependent variables” ~v∗ , P ∗ and the “independent
variables” t∗ , ~r∗ and a dimensionless parameter (Re) represent a simple example of the more general
(Vaschy(ag) –)Buckingham(ah) π-theorem [20] in dimensional analysis, see e.g. Refs. [21, 22] Chapter 7
or [23].
(ag)
A. Vaschy, 1857–1899 (ah) E. Buckingham, 1867–1940 (ai)
W. Froude, 1810–1879 (aj)
V. Ekman, 1874–1954
(ak)
C.-G. Rossby, 1898–1957
96 Non-relativistic dissipative flows
• The motion of fluids past microscopic bodies; the small value of the Reynolds number then
reflects the smallness of the length scale Lc ; for instance:
• The slow-velocity motion of geological material: in that case, the small value of vc and the
large shear viscosity compensate the possibly large value of the typical length scale Lc .
For example, the motion of the Earth’s mantle(31) with Lc ≈ 100 km, vc ≈ 10−5 m · s−1 ,
ρ ≈ 5 · 103 kg · m−3 and η ≈ 1022 Pa · s corresponds to a Reynolds number Re ≈ 5 · 10−19 .
Note that the above examples all represent incompressible flows. For the sake of simplicity, we shall
also only consider steady motions.
VI.3.1 a Stokes equation
::::::::::::::::::::::::
Physically, a small Reynolds number means that the influence of inertia is much smaller than
~
that of shear viscosity. That is, the convective term ~v · ∇ ~v in the Navier–Stokes equation is
negligible with respect to the viscous contribution. Assuming further stationarity—which allows us
to drop the time variable—and incompressibility, the Navier–Stokes equation (III.31) simplifies to
the Stokes equation
~ P (~r) = η4~v(~r) + f~V (~r).
∇ (VI.17)
valid for any vector field ~c(~r), the incompressibility condition, and the definition of vorticity, the
Stokes equation can be rewritten as
~ P (~r) = −η ∇
∇ ~ ×ω
~ (~r). (VI.19)
As a result, the pressure satisfies the differential Laplace equation
4P (~r) = 0. (VI.20)
In practice, however, this equation is not the most useful, because the boundary conditions in a
flow are mostly given in terms of the flow velocity, in particular at walls or obstacles, not of the
pressure.
Taking the curl of Eq. (VI.19) and invoking again relation (VI.18) remembering that the vorticity
vector is itself already a curl, one finds
4~ω (~r) = ~0, (VI.21)
i.e. the vorticity also obeys the Laplace equation. We shall see in Sec. VI.5 that the more general
dynamical equation obeyed by vorticity in Newtonian fluids does indeed yield Eq. (VI.21) in the
case of stationary, small Reynolds number flows.
VI.3.1 b Properties of the solutions of the Stokes equation
:::::::::::::::::::::::::::::::::::::::::::::::::::::::::::
Thanks to the linearity of the Stokes equation (VI.17), its solutions possess various properties:(32)
• Additivity of the solutions: if ~v1 and ~v2 are solutions of Eq. (VI.17) with respective boundary
conditions, then the sum λ1~v1 + λ2~v2 with real numbers λ1 , λ2 is also a solution, for a problem
with adequate boundary conditions.
Physically, the multiplying factors should not be too large, to ensure that the Reynolds number
of the new problem remains small. The multiplication of the velocity field ~v(~r) by a constant
λ represents a change in the mass flow rate, while the streamlines (I.15) remain unchanged.
The dimensionless velocity field ~v∗ associated with the two solutions ~v(~r) and λ~v(~r) is the same,
provided the differing characteristic velocities vc resp. λvc are used. In turn, these define different
values of the Reynolds number. For these solutions, ~v∗ as given by Eq. (VI.13) is thus independent
of the parameter Re, and thereby only depends on the variable ~r∗ : ~v = vc f~ ~r/Lc . This also holds
for the corresponding dimensionless pressure P ∗ .
Using dimensional arguments only, the tangential stress is η∂ vi /∂xj ∼ η vc /Lc , so that the
friction force on an object of linear size(33) Lc is proportional to η vc Lc . This result will now be
illustrated on an explicit example [cf. Eq. (VI.26)], for which the computation can be performed
analytically.
~eϕ
~v∞ ~er
~v(~r) = ~v∞ + ~u(~r), with the boundary condition ~u(~r) = ~0 for |~r| → ∞. In addition, the usual
impermeability and no-slip conditions hold at the surface of the sphere, resulting in the requirement
~u(|~r| = R) = −~v∞ .
Using the linearity of the equations of motion for creeping flows, ~u obeys the equations
~ ×~u(~r) = ~0
4 ∇ (VI.22a)
and
~ ·~u(~r) = 0,
∇ (VI.22b)
which comes from the incompressibility condition.
~ (~r). Using
The latter equation is automatically satisfied if ~u(~r) is the curl of some vector field V
dimensional considerations, the latter should depend linearly on the only explicit velocity scale in
the problem, namely ~v∞ . Accordingly, one makes the ansatz(34)
~ (~r) = ∇
~ × f (r)~v∞ = ∇f ~ (r) ×~v∞ ,
V
with f (r) a function of r = |~r|, i.e. f only depends on the distance from the sphere: apart from the
direction of ~v∞ , which is already accounted for in the ansatz, there is no further preferred spatial
direction, so that f should be spherically symmetric.
Relation (VI.18) together with the identity ∇ ~ · [f (r)~v∞ ] = ∇f
~ (r) ·~v∞ then yield
~ ×V ~ (~r) = ∇
~ ∇f~ (r) ·~v∞ − 4f (r)~v∞ .
~u(~r) = ∇ (VI.23)
The first term on the right hand side has a vanishing curl, and thus does not contribute when
inserting ~u(~r) in equation (VI.22a):
~ ×~u(~r) = −∇
~ × 4f (r)~v∞ = −∇
~ 4f (r) ×~v∞ ,
∇
so that
~ 4f (r) ×~v∞ = ~0.
4 ∇
~ 4f (r) only has a com-
Since f (r) does not depend on the azimuthal and polar angles, 4 ∇
with (unit) basis vector ~er ; as thus, it cannot be always parallel
ponent along the radial direction
to ~v∞ . Therefore, 4 ∇ ~ 4f (r) must vanish identically for the above equation to hold. One can
~ 4f (r) = ∇ ~ 4[4f (r)] , so that the
check—for instance using components—the identity 4 ∇
equation obeyed by f (r) becomes
4[4f (r)] = const.
The integration constant must be zero, since it is a fourth derivative of f (r), while the velocity ~u(~r),
which according to Eq. (VI.23) depends on the second derivatives, must vanish as r → ∞. One
thus has
4[4f (r)] = 0.
(34) ~ (r) ×~v∞ are both unsatisfactory: the velocity ~u(~r) is then always
The simpler guesses ~u(~r) = f (r)~v∞ or ~u(~r) = ∇f
parallel resp. orthogonal to ~v∞ , so that ~v(~r) cannot vanish everywhere at the surface of the sphere.
VI.3 Flows at small Reynolds number 99
∗ In the limit η → 0, corresponding to a perfect fluid, the force (VI.26) exerted by the flow on the
sphere vanishes: this is again the d’Alembert paradox encountered in § IV.4.3 c.
∗ The proportionality factor between the sphere velocity and the friction force it experiences is
called the mobility (lxiv) µ. According to Stokes’ law (VI.26), for a sphere in the creeping-flow regime
one has µ = 1/(6πRη).
In his famous article on Brownian motion [26], A. Einstein related this mobility with the diffusion
coefficient D of a suspension of small spheres in a fluid at rest:
kB T
D = µkB T = .
6πRη
This formula (Stokes–Einstein equation) was checked experimentally by J. Perrin, which allowed
him to determine a value of the Avogadro constant and to prove the “discontinuous structure of
matter” [27].
The invariance of the problem geometry under translations in the x- or z-directions justifies
an ansatz ~v(t,~r) = v(t, y)ex —which automatically fulfills the incompressibility condition—, and
similarly for the pressure field. That is, there are no gradients along the x- and z-directions. As a
result, the incompressible Navier–Stokes equation (III.32), projected onto the x-direction, reads
∂ v(t, y) ∂ 2 v(t, y)
=ν . (VI.27a)
∂t ∂y 2
The boundary conditions are on the one hand the no-slip requirement at the moving plane, namely
v(t, y = 0) = u for t > 0; (VI.27b)
on the other hand, the fluid infinitely far from the moving plane remains unperturbed, i.e.
lim v(t, y) = 0 for t > 0. (VI.27c)
y→∞
The equations governing the motion (VI.27) involve only two dimensionful quantities, namely
the plane velocity u and the fluid kinematic viscosity ν. With their help, one can construct a
characteristic time ν/u2 and a characteristic length ν/u in a unique manner, up to numerical factors.
Invoking dimensional arguments, one thus sees that the fluid velocity is necessarily of the form
2
u t uy
v(t, y) = uf 1 , ,
ν ν
with f1 a dimensionless function of dimensionless variables. Since t and y are independent, so are
their reduced versions
√ u2 t/ν and uy/ν. Instead of the latter, one may adopt the equivalent set
2
u t/ν, ξ ≡ y/(2 νt), i.e. write 2
u t y
v(t, y) = uf 2 , √ ,
ν 2 νt
with f2 again a dimensionless function.
The whole problem (VI.27) is clearly linear in u, since the involved differential equations—
continuity equation and Navier–Stokes equation (VI.27a)—are linear; this allows us to exclude any
dependence of f2 on the variable u2 t/ν, so that the solution is actually of the form
y
v(t, y) = uf √ (VI.28)
2 νt
with f dimensionless, i.e. dependent on a single reduced variable.
Inserting the latter ansatz in Eq. (VI.27a) leads after some straightforward algebra to the ordi-
nary differential equation
f 00 (ξ) + 2ξ f 0 (ξ) = 0, (VI.29a)
with f 0 , f 00 the first two derivatives of f. Meanwhile, the boundary conditions (VI.27b)–(VI.27c)
become
f(0) = 1 , lim f(ξ) = 0. (VI.29b)
ξ→∞
For ξ = 2, erf(2) = 0.99532 . . . , i.e. erfc(2) ≈ 0.005. That is, at given t, the magnitude of the
velocity at √
y = δl (t) ≡ 4 νt (VI.34)
is reduced by a factor 200 with respect to its value at the moving plane. This length δl (t) is a
typical measure for the width of the boundary layer over which momentum is transported from the
plane into the fluid, i.e. the region in which the fluid viscosity plays a role.
The width (VI.34) of the boundary layer increases with the square root of time: this is the
typical behavior expected for a diffusive process—which is understandable since Eq. (VI.27a) is
nothing but the classical diffusion equation.
Remark: The above problem is often referred to as first Stokes problem or Rayleigh problem.(al) In
the second Stokes problem, the plane is not set impulsively into motion, it oscillates sinusoidally
in its own plane with a frequency ω. In that case, the amplitude of the induced fluid oscillations
decrease “only” exponentially withp the distance to the plane, and the typical extent of the region
affected by shear viscous effects is ν/ω.
Since the boundary layer is much extended along the tangential direction than along the normal
one, the range of x values is much larger than that of y values. To obtain dimensionless variables
taking their values over a similar interval, one defines
x y
x∗ ≡ , y∗ ≡ (VI.36)
Lc δl
where the typical extent in the normal direction, i.e. the width of the boundary layer
δl Lc (VI.37)
∗ If the local radius of curvature of the boundary is not much larger than the width δl of the bound-
ary layer, one should use curvilinear coordinates x1 (tangential to the boundary) and x2 instead of
Cartesian ones, yet within that alternative coordinate system the remainder of the derivation still
holds.
the prefactor of the viscous term be of order unity. This suggests the condition
L2c 1
= 1. (VI.43)
δ2l Re
Equations (VI.42)–(VI.43) are then easily solved, yielding for the unknown quantities charac-
terizing the flow along the direction normal to the boundary
Lc v∞
δl = √ , u= √ . (VI.44)
Re Re
As in the first or second Stokes problems, see e.g. Eq. (VI.34), the width of the boundary layer is
proportional to the square root of the kinematic viscosity ν.
Substituting the conditions (VI.42)–(VI.43) in the system of equations (VI.41) and keeping only
the leading terms, one eventually obtains
∂ vx∗ ∂ vy∗
+ = 0; (VI.45a)
∂x∗ ∂y ∗
∂ v∗ ∂ v∗ ∂ P ∗ ∂ 2 v∗
vx∗ x∗ + vy∗ x∗ = − ∗ + ∗2x , (VI.45b)
∂x ∂y ∂x ∂y
∂P ∗
= 0. (VI.45c)
∂y ∗
These equations constitute the simplified system, first by written down by Prandtl,(am) that describe
the fluid motion in a laminar boundary layer—where the laminarity assumption is hidden in the use
of the typical length scale Lc imposed by geometry, rather than of a smaller one driven by turbulent
patterns.
The right hand side of this equation simplifies in various cases. In the particular of a barotropic
fluid, the second term vanishes. In an incompressible flow, the first two terms are zero.
As we shall illustrate on an example, the viscous term, proportional to the Laplacian of vorticity,
is of diffusive nature, and tends to spread out the vorticity lines over a larger region.
This results in a velocity field ~v(t,~r) in the (x, y)-plane, in the orthoradial direction. As a conse-
quence the convective derivative in the left hand side of Eq. (VI.47a) vanishes since ω ~ (t, r) has no
~ ~v also vanishes, since the velocity is independent of z.
gradient along eθ . Similarly, the term ω ~ ·∇
Eventually, the term involving ∇ ~ ·~v vanishes thanks to the assumed incompressibility. All in all,
the vorticity thus obeys the diffusion equation
∂ω z (t, r)
2 z
∂ ω (t, r) 1 ∂ω z (t, r)
z
= ν4ω (t, r) = ν + , (VI.50)
∂t ∂r2 r ∂r
with the initial condition (VI.48).
The problem is clearly linear in Γ0 , so that the solution ω z (t, r) should be proportional to Γ0 ,
without any further dependence on Γ0 . This leaves the kinematic viscosity ν as only dimensionful
parameter available in the problem: using a dimensional reasoning similar to that made in the study
of the first Stokes problem (Sec. VI.4.1), there is a single relevant dimensionless variable, namely
ξ = r2 /(νt), combining the time and space variables. The only ansatz respecting the dimensional
requirements is then
Γ0 r2
ω z (t, r) = f(ξ), with ξ ≡ (VI.51)
νt νt
with f a dimensionless function. Inserting this ansatz into Eq. (VI.50) leads to the ordinary differ-
ential equation
f(ξ) + ξ f 0 (ξ) + 4 f 0 (ξ) + ξ f 00 (ξ)] = 0.
(VI.52)
A first integration yields
ξ f(ξ) + 4ξ f 0 (ξ) = const.
In order to satisfy the initial condition, the integration constant should be zero, leaving with the
linear differential equation f(ξ) + 4f 0 (ξ) = 0, which is readily integrated to yield
f(ξ) = C e−ξ/4 ,
that is
Γ0 2
C e−r /(4νt) ,
ω z (t, r) = (VI.53)
νt
with C an integration which still has to be fixed.
106 Non-relativistic dissipative flows
To determine the latter, let us consider the circulation of the velocity at time t around a circle
CR of radius R centered on the axis z = 0:
I Z R Z 2π Z R
Γ(t, R) = ~v(t,~r) · d~` = ω z (t, r) r dr dθ = 2π ω z (t, r) r dr dθ, (VI.54)
CR 0 0 0
where the second identity follows from the Stokes theorem while the third is trivial. Inserting the
solution (VI.53) yields
2
Γ(t, R) = 4πΓ0 C 1 − e−R /(4νt) ,
showing the C should equal 1/4π to yield the proper circulation at t = 0. All in all, the vorticity
field in the problem reads
Γ0 −r2 /(4νt)
ω z (t, r) = e . (VI.55)
4πνt √
That is, the vorticity extends over a region of typical width δ(t) = 4νt,
√ which increases with time:
√
one recognizes the characteristic diffusive behavior, proportional to t—as well as the typical ν
dependence of the size of the region affected by viscous effects, encountered in Sec. VI.4.
The vorticity (VI.55) leads to the circulation around a circle of radius R
2
Γ(t, R) = Γ0 1 − e−R /(4νt) ,
(VI.56)
which at given R decreases with time, in contrast to the perfect-fluid case, in which the circulation
would be conserved.
Eventually, one can also easily compute the velocity field associated with the expanding vortex,
namely
Γ0 2 ~eθ
~v(t,~r) = 1 − e−r /(4νt) , (VI.57)
2πr r
where |~eθ | = r.
For the sake of simplicity, we consider a one-dimensional problem, i.e. perturbations propagating
along the x-direction and independent of y and z—as are the properties of the underlying background
fluid. Under this assumption, the continuity equation (III.9) reads
∂ρ(t, x) ∂δv(t, x) ∂ρ(t, x)
+ ρ(t, x) + δv(t, x) = 0, (VI.59a)
∂t ∂x ∂x
while the Navier–Stokes equation (III.31) becomes
2
∂δ P (t, x)
∂δv(t, x) ∂δv(t, x) 4 ∂ δv(t, x)
ρ(t, x) + δv(t, x) =− + η+ζ . (VI.59b)
∂t ∂x ∂x 3 ∂x2
Substituting the fields (VI.58a) in these equations and linearizing the resulting equations so as to
keep only the leading order in the small perturbations, one finds
∂δρ(t, x) ∂δv(t, x)
+ ρ0 = 0, (VI.60a)
∂t ∂x
2
∂δ P (t, x)
∂δv(t, x) 4 ∂ δv(t, x)
ρ0 =− + η+ζ . (VI.60b)
∂t ∂x 3 ∂x2
In the second equation, the derivative ∂(δ P )/∂x can be replaced by c2s ∂(δρ)/∂x. Let us in addition
introduce the (traditional) notation(37)
1 4
ν̄ ≡ η+ζ , (VI.61)
ρ0 3
so that Eq. (VI.60b) can be rewritten as
∂δv(t, x) ∂δρ(t, x) ∂ 2 δv(t, x)
ρ0 + c2s = ρ0 ν̄ . (VI.62)
∂t ∂x ∂x2
Subtracting c2s times the time derivative of Eq. (VI.60a) from the derivative of Eq. (VI.62) with
respect to x and dividing the result by ρ then yields
∂ 2 δv(t, x) 2
2 ∂ δv(t, x) ∂ 3 δv(t, x)
− cs = ν̄ . (VI.63a)
∂t2 ∂x2 ∂t ∂x2
One easily checks that the mass density variation obeys a similar equation
∂ 2 δρ(t, x) 2
2 ∂ δρ(t, x) ∂ 3 δρ(t, x)
− cs = ν̄ . (VI.63b)
∂t2 ∂x2 ∂t ∂x2
In the perfect-fluid case ν̄ = 0, one recovers the traditional wave equation (V.9a).
Equations (VI.63) are homogeneous linear partial differential equations, whose solutions can be
written as superposition of plane waves. Accordingly, let us substitute the Fourier ansatz
δρ(t,~r) = δρ(ω,
e ~k) e−iωt+i~k·~r (VI.64)
in Eq. (VI.63b). This yields after some straightforward algebra the dispersion relation
which can only hold if both the real and imaginary parts are equal. The identity between the
imaginary parts reads (for ωr 6= 0)
1
ωi = − ν̄k 2 , (VI.66)
2
which is always negative, since ν̄ is non-negative. This term yields in the Fourier ansatz (VI.64)
2
an exponentially decreasing factor e−i(iωi )t = e−ν̄k t/2 which represents the damping or absorption
of the sound wave. The perturbations with larger wave number k, i.e. corresponding to smaller
length scales, are more dampened that those with smaller k. This is quite natural, since a larger k
also means a larger gradient, thus an increased influence of the viscous term in the Navier–Stokes
equation.
In turn, the identity between the real parts of the dispersion relation yields
1
ωr2 = c2s k 2 − ν̄ 2 k 4 . (VI.67)
4
This gives for the phase velocity cϕ ≡ ω/k of the traveling waves
1
c2ϕ = c2s − ν̄ 2 k 2 . (VI.68)
4
That is, the “speed of sound” actually depends on its wave number k, and is smaller for small
wavelength, i.e. high-k, perturbations—which are also those which are more damped.
Relation (VI.68) also shows that the whole linear description adopted below Eqs. (VI.59) requires
that the perturbations have a relatively large wavelength, namely k 2cs /ν̄, so that cϕ remain real-
valued. This is equivalent to stating that the dissipative term ν̄4δv ∼ k 2 ν̄δv in the Navier–Stokes
equation should be much smaller than the local acceleration ∂t δv ∼ ωδv ∼ cs kδv.
Remarks:
∗ Instead of considering “temporal damping” as was done above by assuming k ∈ R but ω ∈ C,
one may investigate “spatial damping”, i.e. assume ω ∈ R and put the whole complex dependence
in the wave number k = kr + iki . For (angular) frequencies ω much smaller than the inverse of the
typical time scale τν ≡ ν̄/c2s , one finds
2 2 2 3 2 2 ω 3 2 2
ω ' cs kr 1 + ω τν ⇔ cϕ ≡ ' cs 1 + ω τν
4 kr 8
i.e. the phase velocity increases with the frequency, and on the other hand
ν̄ω 2
ki ' . (VI.69)
2c3s
The latter relation is known as Stokes’ law of sound attenuation, ki representing the inverse of the
typical distance over which the sound wave amplitude decreases, due to the factor ei(iki )x = e−ki x
in the Fourier ansatz (VI.64). Larger frequencies are thus absorbed on a smaller distance from the
source of the sound wave.
Substituting k = kr + iki = kr (1 + iκ) in the dispersion relation (VI.65) and writing the identity
of the real and imaginary parts, one obtains the system
(
2κ = ωτν (1 − κ 2 )
ω 2 = c2s kr2 (1 + 2ωτν κ − κ 2 )
The first equation is a quadratic equation in κ that admits one positive and one negative solution:
the latter can be rejected, while the former is κ ' ωτν /2 + O (ωτν )2 . Inserting it in the second
equation leads to the wanted results.
An exact solution of the system of equations exists, yes it is neither enlightening mathematically,
nor relevant from the physical point of view in the general case, as discussed in the next remark.
VI.6 Absorption of sound waves 109
One may naturally also analyze the general case in which both ω and k are complex numbers. In
any case, the phase velocity is given by cϕ ≡ ω/kr , although it is more difficult to recognize the
physical content of the mathematical relations.
∗ For air or water, the reduced kinetic viscosity ν̄ is of order 10−6 –10−5 m2 · s−1 . With speeds of
sound cs ' 300–1500 m · s−1 , this yields typical time scales τν of order 10−12 –10−10 s. That is, the
change in the speed of sound (VI.68), or equivalently deviations from the assumption ωτν 1 under-
lying the attenuation coefficient (VI.69), become relevant for sound waves in the gigahertz/terahertz
regime(!). This explains why measuring the bulk viscosity is a non-trivial task.
The wavelengths cs τν corresponding to the above frequencies τν−1 are of order 10−9 –10−7 m.
This is actually not far from the value of the mean free path in classical fluids, so that the whole
description as a continuous medium starts being questionable.
• Landau–Lifshitz [3, 4] Chapter II § 17–20 & 24, Chapter IV § 39 and Chapter VIII § 79;
All examples of flows considered until now in these notes, either of perfect fluids (Chapters IV and V)
or of Newtonian ones (Chapter VI), share a common property, namely they are all laminar. This
assumption—which often translates into a relative simplicity of the flow velocity profile—is however
not the generic case in real flows, which most often are to some more or less large extent turbulent.
The purpose of this Chapter is to provide an introduction to the problematic of turbulence in fluid
motions.
A number of experiments, in particular those conducted by O. Reynolds, have hinted at the
possibility that turbulence occurs when the Reynolds number (VI.12) is large enough in the flow,
i.e. when convective effects predominate over the shear viscous ones that drive the mean fluid motion
over which the instabilities develop. This distinction between mean flow and turbulent fluctuations
can be modeled directly by splitting the dynamical fields into two parts, and one recovers with the
help of dimensional arguments the role of the Reynolds number in separating two regimes, one in
which the mean viscous flow dominates and one in which turbulence takes over (Sec. VII.1).
Despite its appeal, the decomposition into a mean flow and a turbulent motion has the drawback
that it leads to a system of equations of motion which is not closed. A possibility to remedy this
issue is to invoke the notion of a turbulent viscosity, for which various models have been proposed
(Sec. VII.2).
Even when the system of equations of motion is closed, it still involves averages—with an a
priori unknown underlying probability distribution. That is, the description of turbulent part of the
motion necessitates the introduction of a few concepts characterizing the statistics of the velocity
field (Sec. VII.3).
For the sake of simplicity, we shall mostly consider turbulence in the three-dimensional incom-
pressible motion of Newtonian fluids with constant and homogeneous properties (mass density,
viscosity. . . ), in the absence of relevant external bulk forces, and neglecting possible temperature
gradients—and thereby convective heat transport.
Due to the viscous friction forces, part of the kinetic energy of the fluid motion is transformed
into heat. To compensate for these “losses” and keep the flow in the stationary regime, energy has
to be provided to the fluid, namely in the form of the mechanical work of the pressure forces driving
the flow. Thus, the rate of energy dissipation per unit mass is(38)
1 ∆P 8νhvi2
Ė diss. = − hvi = (VII.1)
ρ L a2
with hvi the average flow velocity across the tube cross section,
Q a2 ∆P
hvi = = − .
πa2 ρ 8η L
Thus, in the laminar regime the rate Ė diss. is proportional to the kinematic viscosity ν and to the
square of the average velocity.
According to the Hagen–Poiseuille law (VI.9), at fixed pressure gradient the average velocity hvi
grows quadratically with the tube radius. In practice, the rise is actually slower, reflecting a higher
rate of energy loss in the flow as given by the laminar prediction (VII.1). Thus, the mean rate of
energy dissipation is no longer proportional to hvi2 , but rather to a higher power of hvi. Besides,
the flow velocity profile across the tube cross section is no longer parabolic, but (in average) flatter
around the cylinder axis, with a faster decrease at the tube walls.
The notion of a critical Reynolds number separating the laminar and turbulent regimes is actu-
ally a simplification. In theoretical studies of the stability of the laminar regime against linear
perturbations, such a critical value Rec can be computed for some very simple geometries, yield-
ing e.g. Rec = 5772 for the plane Poiseuille flow. Yet the stability sometimes also depends on
the size of the perturbation: the larger it is, the smaller the associated critical Rec is, which
hints at the role of nonlinear instabilities.
In the following, we shall leave aside the problem of the temporal onset of turbulence—and
thereby of the (in)stability of laminar flows—, and focus on flows which are already turbulent when
we start looking at them.
with ~v resp. ~v0 the “slow” resp. “fast” component. For the pressure, one similarly writes
The fluid motion with velocity ~v and pressure P is then referred to as “mean flow”, that with
the rapidly varying quantities as “fluctuating motion”.
As hinted at by the notation, ~v(t,~r) represents an average, with some underlying probability
distribution.
Theoretically, the Reynolds average · should be an ensemble average, obtained from an infinitely
large number of realizations, namely experiments or computer simulations; in practice, however,
there is only a finite number N of realizations ~v(n)(t,~r). If the turbulent flow is statistical
stationary, one may invoke the ergodic assumption and replace the ensemble average by a time
average:
N
1 t+T /2 0
Z
1 X (n)
~v(~r) ≡ lim ~v (t,~r) ≈ ~v(t ,~r) dt0 ,
N →∞ N
n=1
T t−T /2
with T much larger than the autocorrelation time of the turbulent velocity ~v0(t,~r). If the flow
is not statistically stationary, so that ~v(t,~r) also depends on time, then T must also be much
smaller than the typical time scale of the variations of the mean flow.
Using the same averaging procedure, the fluctuating velocity must obey the condition
Despite this fact, the turbulent velocity ~v0(t,~r) still plays a role in the dynamics, in particular that
of the mean flow, because its two-point, three-point and higher (auto)correlation functions are still
in general non-vanishing. For instance, one can write—assuming that the mass density ρ is constant
and uniform
ρ vi (t,~r) vj (t,~r) = ρ vi (t,~r) vj (t,~r) + ρ v0i (t,~r) v0j (t,~r).
The first term of the right member corresponds to convective part of the momentum-flux density
(lxvi)
Reynolds-Zerlegung
VII.1 Generalities on turbulence in fluids 113
is due to the rapidly fluctuating motion. TR is called turbulent stress or Reynolds stress.(lxvii)
These two equations involve the material derivative “following the mean flow”
D ∂ ~
≡ + ~v · ∇. (VII.10)
Dt ∂t
Using the incompressibility of the fluctuating motion, the rightmost term in Eq. (VII.9a) can
be rewritten as
~ ~v0 = −ρ∇
~ · ~v0 ⊗~v0 = −∇
~ · TR .
−ρ ~v0 · ∇
The Reynolds equation can thus be recast in the equivalent form [cf. Eq. (III.24b)]
∂
~ · T = −∇~ · TR ,
ρ~v + ∇ (VII.11)
∂t
with T the momentum-flux density of the mean flow, given by [cf. Eqs. (III.26b), (III.26e)]
T ≡ P g−1 + ρ~v ⊗ ~v − 2ηS
S (VII.12a)
i.e., component-wise
(lxvii)
Reynolds-Spannung
114 Turbulence in non-relativistic fluids
T ij ≡ P g ij + ρvi vj − 2ηS
Sij , (VII.12b)
with S the rate-of-shear tensor [Eq. (II.15b)] for the mean flow, with components [cf. Eq. (II.15d)]
1 dvi dvj
ij
2 ij ~
S ≡ + − g ∇ · ~v , (VII.12c)
2 dxj dxi 3
where the third term within the brackets actually vanishes due to the incompressibility of the mean
flow, Eq. (VII.7).
The form (VII.11) of the Reynolds equation emphasizes perfectly the role of the Reynolds stress,
i.e. the turbulent component of the flow, as “external” force driving the mean flow. In particular, the
off-diagonal terms of the Reynolds stress describe shear stresses, which will lead to the appearance
of eddies in the flow.
Starting from the Reynolds equation, one can derive the equation governing the evolution of the
2
kinetic energy 12 ρ ~v associated with the mean flow, namely
D ρ~v2
~ · P ~v + TR − 2ηS
= −∇ S · ~v + TR − 2ηS S : S. (VII.13)
Dt 2
2
This equation is conventionally rather written in terms of the kinetic energy per unit mass k ≡ 12 ~v ,
in which case it reads
Dk ~ 1
= −∇ · P ~v + ~v0 ⊗~v0 − 2ν S · ~v + ~v0 ⊗~v0 − 2ν S : S , (VII.14a)
Dt ρ
or component-wise
3 3 X 3
Dk X d 1 j X 0i 0j ij
0i v0j − 2ν S ij S .
=− P v + v v − 2ν S vi + v ij (VII.14b)
Dt dxj ρ
j=1 i=1 i,j=1
In either form, the physical meaning of each term is rather transparent: first comes the convective
transport of energy in the mean flow, given by the divergence of the energy flux density, inclusive
a term from the turbulent motion. The second term represents the energy which is “lost” to the
mean flow, namely either because it is dissipated by the viscous friction forces (term in ν S : S ), or
because it is transferred to the turbulent part of the motion (term involving the Reynolds stress).
To prove Eq. (VII.13), one should first average the inner product with ~v of the Reynolds equa-
~ P and ~v· ~v0 · ∇
~ ~v0 under consideration of the incompressibility
tion (VII.9), and then rewrite ~v· ∇
condition (VII.7).
Remark: While equations (VII.9) or (VII.14) do describe the dynamics of the mean flow, they rely
on the Reynolds stress, which is not yet specified by the equations.
In that situation, the equation (VII.14) describing the evolution of the kinetic energy of the
mean flow becomes
Dk ~ · 1 P ~v + ~v0 ⊗~v0 · ~v + ~v0 ⊗~v0 : S ,
= −∇ (VII.16a)
Dt ρ
or component-wise
3 3 X 3
Dk X d 1 j X 0i 0j
=− Pv + v v vi + v0i v0j Sij . (VII.16b)
Dt dxj ρ
j=1 i=1 i,j=1
That is, the viscosity is no longer a relevant parameter for the dynamics of the mean flow.
As already argued above, the first term in Eq. (VII.16) represents the convective transport of
energy in the mean flow, while the second, “mixed” term models the transfer of energy from the
mean flow into the turbulent motion, and thus corresponds to the energy “dissipated” by the mean
flow. Invoking the first relation in Eq. (VII.15), the rate of energy dissipation in the mean flow is
* 3 +
X v3
Ė diss. = v0i v0j S ij ∼ c . (VII.17)
Lc
i,j=1
This grows like the third power of the typical velocity, i.e. faster than vc2 , as argued at the end of
§ VII.1.1 a for the turbulent regime of the Hagen–Poiseuille flow. In addition, this energy dissipation
rate is actually independent of the properties (mass density, viscosity. . . ) of the flowing fluid:
turbulence is a characteristic of the motion, not of the fluid itself.
Eventually, the middle term in Eq. (VII.17) must be negative, so that the energy really flows
from the mean flow to the turbulent motion, not in the other direction!
Remark: Looking naively at the definition of the Reynolds number, the limit of an infinitely large
Re corresponds to the case of a vanishing shear viscosity, that is, to the model of a perfect fluid.
As was just discussed, this is clearly not the case: with growing Reynolds number, i.e. increasing
influence of the turbulent motion, the number of eddies in the flow also increases, in which energy
is dissipated into heat. In contrast, the kinetic energy is conserved in the flow of a perfect fluid.
The solution to this apparent paradox is simply that with increasing Reynolds number, the velocity
gradients in the flow also increase. In the incompressible Navier–Stokes equation, the growth of 4~v
compensates the decrease of the viscosity ν, so that the corresponding term does not disappear and
the Navier–Stokes equation does not simplify to the Euler equation.
Irrespective of the physical interpretation of each of the terms in this equation, an important issue
is that the evolution of ρv0i v0j involves a contribution from the components ρv0i v0j v0k of a tensor of
degree 3. In turn, the evolution of ρv0i v0j v0k involves the tensor with components ρv0i v0j v0k v0l , and
so on: at each step, the appearance of a tensor of higher degree simply reflects the nonlinearity of
the Navier–Stokes equation.
All in all, the incompressible Navier–Stokes equation (VII.8) is thus equivalent to an infinite
hierarchy of equations relating the successive n-point autocorrelation functions of the fluctuations
of the velocity field. Any subset of this hierarchy is not closed and involves more unknown fields
than equations. A closure prescription, based on some physical assumption, is therefore necessary,
to obtain a description with a finite number of equations governing the (lower-order) autocorrelation
functions. Such an approach is presented in Sec. VII.2.
An alternative possibility is to assume directly some ansatz for the statistical behavior of the
turbulent velocity, especially for its general two-point autocorrelation function, of which the equal-
time and position correlator v0i (t,~r)v0j (t,~r) is only a special case. This avenue will be pursued in
Sec. VII.3.
where the proportionality factor involves the (kinematic) turbulent viscosity or eddy viscosity (40)
νturb , which a priori depends on time and position. In terms of components in a coordinate system,
and replacing the Reynolds stress and its trace by their expressions in terms of the fluctuating
velocity, this reads
1
ρ v0i (t,~r) v0j (t,~r) − ρ [~v0(t,~r)]2 g ij (t,~r) ≡ 2ρνturb (t,~r) S ij (t,~r). (VII.19b)
3
Using the ansatz (VII.19) and invoking the incompressibility of the mean flow, from which follows
~ · S = 1 4~v, the Reynolds equation (VII.9) can be rewritten as
∇ 2
P (t,~r) [~v0(t,~r)]2
∂~v(t,~r) ~
~
+ ~v(t,~r) · ∇ ~v(t,~r) = −∇ + + 2νeff (t,~r)4~v(t,~r), (VII.20)
∂t ρ 3
with the effective viscosity
νeff (t,~r) = ν + νturb (t,~r), (VII.21)
while the term in curly brackets may be seen as an effective pressure.
(40)
turbulente Viskosität, Wirbelviskosität
(an)
J. Boussinesq, 1842–1929
VII.2 Model of the turbulent viscosity 117
Even if the intrinsic fluid properties, in particular its kinematic viscosity ν, are assumed to
be constant and uniform, this does not hold for the turbulent and effective viscosities νturb , νeff ,
because they model not the fluid, but also its flow—which is time and position dependent.
Either starting from Eq. (VII.20) multiplied by ~v, or substituting the Reynolds stress with the
ansatz (VII.19) in Eq. (VII.14), one can derive the equation governing the evolution of the kinetic
energy of the mean flow. In particular, one finds that the dissipative term is
3
X
Ė diss = 2νeff S : S = 2νeff S ij S ij .
i,j=1
Comparing with the rightmost term in Eq. (VII.14) gives for the effective viscosity
X X
− v0i v0j S ij 2ν S ij S ij
i,j i,j
νeff = X X = ν,
2 S ij S ij 2 S ij S ij
i,j i,j
where the inequality holds in the turbulent regime. There thus follows νeff ≈ νturb ν.
It has been argued that in plasmas, the turbulent viscosity νturb could in some regimes be
negative—and of the same magnitude as ν—, leading to an “anomalaous” effective viscosity νeff
much smaller than ν [30, 31].
Remark: To emphasize the distinction with the turbulent viscosity, ν is sometimes referred to as
“molecular” viscosity.
While the ansatz (VII.19) allows the rewriting of the Reynolds equation in the seemingly simpler
form (VII.20)—in which the two terms contributing to the effective pressure are to be seen as
constituting a single field—, it still involves an unknown, flow-dependent quantity, namely the
effective viscosity νeff , which still needs to be specified.
with |S
S| a typical value of the rate-of-shear tensor of the mean flow. In any case, the turbulent
viscosity is determined by local quantities.
The latter point is actually a weakness of the model. For instance, it implies that the turbulent
viscosity (VII.22) vanishes at an extremum of the mean flow velocity—for instance, on the tube axis
in the Hagen–Poiseuille flow—, which is not realistic. In addition, turbulence can be transported
from a region into another one, which also not describe by the ansatz.
(lxix) (lxx)
Mischungswegansatz Mischungsweglänge
118 Turbulence in non-relativistic fluids
Eventually, the mixing-length model actually merely displaces the arbitrariness from the choice
of the turbulent viscosity νturb to that of the mixing length `m , i.e. it is just a change of unknown
parameter.
VII.2.3 k -model
In order to describe the possible transport of turbulence within the mean flow, the so-called
k-model was introduced.
Denoting by k 0 ≡ 12~v02 the average kinetic energy of the turbulent fluctuations, the turbulent
viscosity is postulated to be
1/2
νturb (t,~r) = `m (t,~r)k 0(t,~r) . (VII.23)
An additional relation is needed to describe the transport of k 0 , to close the system of equations.
For simplicity, one the actual relation [see Eq (VII.26) below] is replaced by a similar-looking
equation, in which the material derivative following the main flow of the average turbulent kinetic
energy equals the sum of a transport term—minus the gradient of a flux density, taken to be
proportional to the gradient of k 0 —, a production term—namely the energy extracted from the
mean flow—, and a dissipation term that describes the rate of energy release as heat, and whose
form
3/2
Ė diss. = Ck0 /`m
is motivated by dimensional arguments, with C a constant. Due to the introduction of this extra
phenomenological transport equation for k 0 , which was not present in the mixing-length model, the
k-model is referred to as a one-equation model .(lxxi)
The k-model allows by construction the transport of turbulence. However, the mixing length
`m remains an empirical parameter, and two further ones are introduced in the transport equation
for the average turbulent kinetic energy.
VII.2.4 (k -ε)-model
In the k-model, the dissipation term Ė diss. which stands for the ultimate transformation of turbu-
lent kinetic energy into heat under the influence of viscous friction, and should thus be proportional
to the viscosity ν, is determined by a dimensional argument.
Another possibility is to consider the energy dissipation rate Ė diss (t,~r)—which is usually rather
denoted as ε̄—as a dynamical variable, whose evolution is governed by a transport equation of its
own. This approach yields a two-equation model ,(lxxii) the so-called (k-ε)-model
3/2
A dimensional argument then gives `m ∼ k 0 /Ė diss. , and thus
2
k 0(t,~r)
νturb (t,~r) = C , (VII.24)
Ė diss. (t,~r)
with C an empirical constant.
In this model—or rather, this class of models—, the mixing length is totally fixed by the dy-
namical variables, thus is no longer arbitrary. On the other hand, the two transport equations
introduced for the average turbulent kinetic energy and the dissipation rate involve a handful of
parameters, which have to be determined empirically for each flow.
In addition, the (k-ε)-model, like all descriptions involving a turbulent viscosity, relies on the
assumption that the typical scale of variations of the mean flow velocity is clearly separated from
the turbulent mixing length. This hypothesis is often not satisfied, in that many flows involve
(lxxi) (lxxii)
Eingleichungsmodell Zweigleichungsmodell
VII.3 Statistical description of turbulence 119
turbulent motion over many length scales, in particular with a larger scale comparable with that
of the gradients of the mean flow. In such flows, the notion of turbulent viscosity is not really
meaningful.
1 dv0i dv0j
0 ij 2 ij ~ 0
with S ≡ + − g ∇ ·~v the components of the fluctuating rate-of-shear tensor.
2 dxj dxi 3
• The first term describes a turbulent yet conservative transport—due to pressure, convective
transport by the fluctuating flow itself, or diffusive transport due the viscous friction—, mixing
the various length scales: the kinetic energy is transported without loss from the large scales,
comparable to that of the variations of the mean flow, to the smaller ones. This process is
referred to as energy cascade.
• The second term describes the “creation” of turbulent kinetic energy, which is actually ex-
tracted from the mean flow: it is precisely—up to the sign!—the loss term in the Eq. (VII.16)
describing the transport of kinetic energy in the mean flow.
• Eventually, the rightmost term in Eq. (VII.26) represents the average energy dissipated as
heat by the viscous friction forces, and will hereafter be denoted as Ė diss. .
120 Turbulence in non-relativistic fluids
In a statistically homogeneous and stationary turbulent flow, the amount of energy dissipated
by viscous friction equals that extracted by turbulence from the mean flow, i.e.
3 3
0
S 0 ij S ij .
X X
− v0i v0j S ij = 2ν (VII.27)
i,j=1 i,j=1
If the turbulence is statistically locally isotropic,(41) the tensor κij only depends on the distance
X ≡ |X| ~ between the two points. Such a statistical local isotropy often represents a good assumption
for the structure of the turbulent motion on small scales—again, far from the boundaries of the
flow—and will be assumed hereafter.
Consider two points at ~r and ~r + X.~ Let ~ek denote a unit vector along X, ~ ~e⊥ a unit vector in
a direction orthogonal to ~ek , and ~e⊥ perpendicular to both ~ek and ~e⊥ . The component vk0 of the
0
~
turbulent velocity—at ~r or ~r + X—along ~ek is referred to as “longitudinal”, those along ~e⊥ or ~e⊥0
0 0
(v⊥ , v⊥0 ) as “lateral”.
The autocorrelation function (VII.29) can be expressed with the help of the two-point functions
~ κ⊥ (X) ≡ v0 (t,~r) v0 (t,~r + X),
κk(X) ≡ vk0 (t,~r) vk0 (t,~r + X), ~ and κ0 (X) ≡ v0 (t,~r) v0 0 (t,~r + X)
~ as
⊥ ⊥ ⊥ ⊥ ⊥
3
Xi Xj 0
X ijk Xk
κij (X) = κk(X) − κ⊥ (X) + κ⊥ (X) δij + κ⊥ (X) ,
~2
X X
k=1
with {Xi } the Cartesian components of X,~ where the last term vanishes for statistically space-parity
invariant turbulence, (42) which is assumed to be the case from now on.(43)
(41)
This means that the probability distribution of the velocity fluctuations ~v0 is stationary (time-independent) resp.
homogeneous (position-independent) resp. locally isotropic (the same for all Cartesian components of ~v0).
(42)
Invariance under the space-parity operation is sometimes considered to be part of the isotropy, sometimes not. . .
(43)
In presence of a magnetic field—i.e. in the realm of magnetohydrodynamics—, this last term is indeed present.
VII.3 Statistical description of turbulence 121
• the fluctuations of the velocity at points separated by a large distance X are not correlated
with another, so that κk(X) must vanish: lim f (X) = 0.
X→∞
• In addition, f is assumed to be integrable over R+ , and such that its integral from 0 to +∞
is convergent.
The function f then defines a typical macroscopic length scale, namely that over which f resp. κk
decreases,(44) the integral scale or external scale (lxxiii)
Z ∞
LI ≡ f (X) dX. (VII.30)
0
Empirically, this integral scale is found to be comparable to the scale of the variations of the mean
flow velocity, i.e. characteristic for the production of turbulence in the flow. For example, in a flow
past an obstacle, LI is of the same order of magnitude as the size of the obstacle.
Assuming—as has been done till now—locally isotropic and space-parity invariant turbulence,
the function f (X) is even, so that its Taylor expansion around X = 0 defines a microscopic length
scale:
1 X 2
1
f (X) ' 1 − + O(X 4 ) with `2T ≡ − 00 > 0. (VII.31)
X→0 2 `T f (0)
`T is the Taylor microscale.(lxxiv)(45)
Let xk denote the coordinate along X. ~ One finds
2 . 0 2
`2T = vk0 (t,~r)
dvk(t,~r)/dxk , (VII.32)
i.e. `T is the typical length scale of the gradients of the velocity fluctuations.
Remark: Even if the Taylor microscale emerges naturally from the formalism, it does not represent
the length scale of the smallest eddies in the flow, despite what one could expect.
To find another, physically more relevant microscopic scale, it is necessary to investigate the
behavior of the longitudinal increment
δvk0 (X) ≡ vk0 (t,~r + X)
~ − vk0 (t,~r) (VII.33)
of the velocity fluctuations, which compares the values of the longitudinal component of the latter
at different points. According to the definition of the derivative, dvk0 /dxk is the limit when X → 0
of the ratio δvk0 (X)/X. The microscopic Kolmogorov length scale `K is then defined by
0
[δvk0 (`K )]2 [δvk0 (X)]2 dvk(t,~r) 2
≡ lim = . (VII.34)
`2K X→0 X2 dxk
The role of this length scale will be discussed in the following Section, yet it can already be mentioned
that it is the typical scale of the smallest turbulent eddies, thus the pendant to the integral scale
LI .
Remark: Squaring the longitudinal velocity increment (VII.33) and averaging under consideration
of the statistical homogeneity, one finds when invoking Eq. (VII.31)
[δvk0 (X)]2 1 X 2
∼ .
2[vk0 (X)]2 X→0 2 `T
On the other hand, experiments or numerical simulations show that the left hand side of this
relation equals about 1 when X is larger than the integral scale LI . That is, the latter and the
Taylor microscale can also be recovered from the longitudinal velocity increment.
Here “universality” refers to an independence from the precise process which triggers the
turbulence.
Considering e.g. the longitudinal increment, this hypothesis gives for the second moment of
the probability distribution
q 3 1/4
0 2 (2) X ν
[δvk(X)] = ν Ė diss. Φ for X LI with `K = (VII.35)
`K Ė diss.
q
and Φ(2) a universal function, irrespective of the flow under study. The factor ν Ė diss. and
the form of `K follow from dimensional considerations—the n-point autocorrelation function
n/4
involves another function Φ(n) multiplying a factor ν Ė diss. .
(ao)
A. N. Kolmogorov = A. N. Kolmogorov, 1903–1987
VII.3 Statistical description of turbulence 123
The hypothesis (K41-1) amounts to assuming that the physics of the fluctuating motion, far
from the scale at which turbulence is created, is fully governed by the available energy extracted
from the mean flow—which in the stationary regime equals the average energy dissipated by
viscous friction in the turbulent motion—and by the amount of friction.
The idea here is that viscous friction only plays a role at the microscopic scale, while the rest
of the turbulent energy cascade is conservative.
The assumption holds for the longitudinal increment (VII.35) if and only if Φ(2)(x) ∼ B (2) x2/3
x1
with B (2) a universal constant, i.e. if
2/3
[δvk0 (X)]2 ∼ B (2) Ė diss. X for `K X LI . (VII.36)
The Kolmogorov 2/3-law (VII.36) does not involve any length scale: this reflects the length-
scale “self-similarity” of the conservative energy-cascading process in the inertial range (lxxv)
`K X LI , in which the only relevant parameter is the energy dissipation rate.
The increase of the autocorrelation function [δvk0 (X)]2 as X 2/3 is observed both experimentally
and in numerical simulations.(46)
A further prediction of the K41-theory regards the energy spectrum of the turbulent motion. Let
~v˜ 0(t, ~k) denote the spatial Fourier transform of the fluctuating velocity. Up to a factor involving the
inverse of the (infinite) volume of the flow, the kinetic energy per unit mass of the turbulent motion
component with wave vector equal to ~k up to d3~k is 21 [~v˜ 0(t, ~k)]2 d3~k. In the case of statistically
isotropic turbulence, 12 [~v˜ 0(t, ~k)]2 d3~k = 2πk 2 [~v˜ 0(t, ~k)]2 dk ≡ SE (k) dk with SE (k) the kinetic-energy
spectral density.
From the 2/3-law (VII.36),(47) one can then derive the −5/3-law for the latter, namely
1/4
2/3 ε̄
SE (k) = CK Ė diss. k −5/3 for L−1 I k `−1
K = , (VII.37)
ν3
with CK a universal constant, the Kolmogorov constant, independent from the fluid or the flow
geometry, yet depending—like the −5/3-law itself—on the space dimensionality. Experimentally(46)
one finds CK ≈ 1.45.
As already mentioned, the laws (VII.36) and (VII.37) provide a rather satisfactory description
of the results of experiments or numerical simulations. The K41-theory also predicts that the
higher-order moments of the probability distribution of the velocity increments should be universal
n/3
as well—and the reader can easily determine their scaling behavior [δvk0 (X)]n ∼ B (n) Ė diss. X
in the inertial range using dimensional arguments—, yet this prediction is no longer supported by
experiment: the moments do depend on X as power laws, yet not with the predicted exponents.
A deficiency of Kolmogorov’s theory is that in his energy cascade, only eddies of similar size
interact with each other to transfer the energy from large to small length scales, which is encoded
in the self-similarity assumption. In that picture, the distribution of the eddy sizes is statistically
stationary.
(46)
Examples from experimental results are presented in Ref. [35, Chapter 5].
(47)
... and assuming that SE (k) behaves properly, i.e. decreases quickly enough, at large k.
(lxxv)
Trägheitsbereich
124 Turbulence in non-relativistic fluids
In contrast, turbulent motion itself tends to deform eddies, by stretching vortices into tubes
of smaller cross section, until they become so small that shear viscosity becomes efficient to
counteract this process (see Sec. VI.5). This behavior somewhat clashes with Kolmogorov’s
picture.
• Frisch [35].
The previous two Chapters were devoted to flows dominated by viscosity (Chap. VI) or by convective
motion (Chap. VII). In either case, the energy-conservation equation (III.35), and in particular the
term representing heat conduction, was never taken into account, with the exception of a brief
mention in the study of static Newtonian fluids (Sec. VI.1.1).
The purpose of this Chapter is to shift the focus, and to discuss motions of Newtonian fluids in
which heat is transfered from one region of the fluid to another. A first such type of transfer is heat
conduction, which was already encountered in the static case. Under the generic term “convection”,
or “convective heat transfer”, one encompasses flows in which heat is also transported by the moving
fluid, not only conductively.
Heat transfer will be caused by differences in temperature in a fluid. Going back to the equations
of motion, one can make a few assumptions so as to eliminate or at least suppress other effects,
and emphasize the role of temperature gradients in moving fluids (Sec VIII.1). A specific instance
of fluid motion driven by a temperature difference, yet also controlled by the fluid viscosity, which
allows for a richer phenomenology, is then presented in Sec. VIII.2.
Eventually, straightforward algebra using the continuity equation allows one to rewrite the entropy
balance equation (III.40b) as
D s(t,~r)
~ (t,~r) + 2η(t,~r) S (t,~r) : S (t,~r) + ζ(t,~r) ∇
~ · κ(t,~r)∇T
~ ·~v(t,~r) 2 . (VIII.1c)
ρ(t,~r) =∇
Dt ρ(t,~r) T (t,~r) T (t,~r)
Since we wish to isolate effects directly related with the transfer of heat, or playing a role in it,
we shall make a few assumptions, so as to simplify the above set of equations.
• The transport coefficients η, ζ, κ depend on the local thermodynamic state of the fluid, i.e.
on its local mass density ρ and temperature T , and thereby indirectly on time and position.
Nevertheless, they will be taken as constant and uniform throughout the fluid, and taken out
of the various derivatives in Eqs. (VIII.1b)–(VIII.1c). This is a reasonable assumption as long
as only small variations of the fluid properties are considered, which is consistent with the
next assumption.
Somewhat abusively, we shall in fact even allow ourselves to consider η resp. κ as uniform in
Eq. (VIII.1b) resp. (VIII.1c), later replace them by related (diffusion) coefficients ν = η/ρ
resp. α = κ/ρc P , and then consider the latter as uniform constant quantities.
The whole procedure is only “justified” in that one can check—by comparing calculations
using this assumption with numerical computations performed without the simplifications—
that it does not lead to omitting a physical phenomenon.
• The fluid motions under consideration will be assumed to be “slow”, i.e. to involve a small flow
velocity, in the following sense:
– The incompressibility condition ∇·~ ~ v(t,~r) = 0 will hold on the right hand sides of each of
Eqs. (VIII.1). Accordingly, Eq. (VIII.1a) simplifies to Dρ(t,~r)/Dt = 0 while Eq. (VIII.1b)
becomes the incompressible Navier–Stokes equation
∂~v(t,~r) ~ ~v(t,~r) = − 1 ∇
~ P (t,~r) − ∇Φ(t,~
~
+ ~v(t,~r) · ∇ r) + ν4~v(t,~r), (VIII.2)
∂t ρ(t,~r)
in which the kinematic viscosity ν is taken to be constant.
– The rate of shear is small, so that its square can be neglected in Eq. (VIII.1c). Accord-
ingly, that equation simplifies to
D s(t,~r)
ρ(t,~r) = κ4T (t,~r). (VIII.3)
Dt ρ(t,~r)
The left member of that equation can be further rewritten. Dividing the fundamental relation
of thermodynamics dU = T dS − P dV (at constant particle number) by the mass of the atoms of
the fluid yields the relation
e s 1
d =Td −P d .
ρ ρ ρ
In keeping with the assumed incompressibility of the motion, the rightmost term vanishes, while
the change in specific energy can be related to the variation of temperature as d(e/ρ) = c P dT with
c P the specific heat capacity at constant pressure. In a fluid particle, one may thus write
s
Td = c P dT, (VIII.4)
ρ
which translates into a relation between material derivatives when the fluid particles are followed
in their motion. The left member of Eq. (VIII.3) may then be expressed in terms of the substantial
derivative of the temperature. Introducing the thermal diffusivity (lxxvi)
κ
α≡ , (VIII.5)
ρc P
(lxxvi)
Temperaturleitfähigkeit
VIII.1 Equations of convective heat transfer 127
which will be assumed to be constant and uniform in the fluid, where ρc P is the volumetric heat
capacity at constant pressure, one eventually obtains
which in contrast to the Mach, Reynolds, Froude, Ekman, Rossby. . . numbers encountered in the
previous Chapters is entirely determined by the fluid, independent of any flow characteristics.
where ' is used to allow for small relative variations in the mass density, which is directly related
to the expansion rate [Eq. (VIII.1a)].
Denoting by T0 a typical temperature in the fluid and ρ0 the corresponding mass density (strictly
speaking, at a given pressure), the effect of thermal expansion on the latter reads
with
Θ ≡ T − T0 (VIII.10)
the temperature difference measured with respect to the reference value, and
1 ∂ρ
α(V ) ≡ − (VIII.11)
ρ ∂T P ,N
the thermal expansion coefficient for volume, where the derivative is taken at the thermodynamic
point corresponding to the reference value ρ0 . Strictly speaking, the linear regime (VIII.9) only
holds when α(V ) Θ 1, which will be assumed hereafter.
(48)
Hence its denomination Boussinesq approximation (for buoyancy).
128 Convective heat transfer
Consistent with relation (VIII.9), the pressure term in the incompressible Navier–Stokes equation
can be approximated as
1 ~ ~ P (t,~r)
∇
− ∇P (t,~r) ' − 1 + α(V ) Θ(t,~r) .
ρ(t,~r) ρ0
Introducing an effective pressure P eff which accounts for the leading effect of the potential from
which the volume forces derive,
P eff.(t,~r) ≡ P (t,~r) + ρ0 Φ(t,~r),
one finds
1 ~ ~ P eff.(t,~r)
∇
− ~
∇P (t,~r) − ∇Φ(t,~
r) ' − ~
+ α(V ) Θ(t,~r)∇Φ(t,~
r),
ρ(t,~r) ρ0
where a term of subleading order α(V ) Θ∇~ P eff. has been dropped. To this level of approximation,
the incompressible Navier–Stokes equation (VIII.2) becomes
~
∂~v(t,~r) ~ ~v(t,~r) = − ∇P eff.(t,~r) + α(V ) Θ(t,~r)∇Φ(t,~
~
+ ~v(t,~r) · ∇ r) + ν4~v(t,~r). (VIII.12)
∂t ρ0
This form of the Navier–Stokes equation emphasizes the role of a finite temperature difference Θ
in providing an extra force density which contributes to the buoyancy, supplementing the effective
pressure term.
Eventually, definition (VIII.10) together with the convective heat transfer equation (VIII.6) lead
at once to
∂Θ(t,~r) ~ Θ(t,~r) = α4Θ(t,~r).
+ ~v(t,~r) · ∇ (VIII.13)
∂t
The (Oberbeck (ap) –)Boussinesq equations (VIII.8), (VIII.12), and (VIII.13) represent a closed
system of five coupled scalar equations for the dynamical fields~v, Θ—which in turn yields the whole
variation of the mass density—and P eff. .
with the pressure given by P st.(t,~r) = P eff.,st.(t,~r) − ρ0 gz. Since |z/d| < 12 and α(V ) ∆T 1, one
sees that the main part of the pressure variation due to gravity is already absorbed in the definition
of the effective pressure.
If ∆T = 0, one recognizes the usual linear pressure profile of a static fluid at constant tempera-
ture in a uniform gravity field.
One can check that the fluid state defined by the profile (VIII.14) is stable against small per-
turbations of any of the dynamical fields. To account for that property, that state (for a given
temperature difference ∆T ) will be referred to as “equilibrium state”.
Increasing now the temperature of the lower plate with respect to that of the upper plate, for
small positive temperature differences ∆T nothing happens, and the static solution (VIII.14) still
holds—and is still stable.
When ∆T reaches a critical value ∆Tc , the fluid starts developing a pattern of somewhat regular
cylindrical domains rotating around their longitudinal, horizontal axes, two neighboring regions
rotating in opposite directions. These domains in which warmer and thus less dense fluid rises on
the one side while colder, denser fluid descends on the other side, are called Bénard cells.(aq)
Figure VIII.1 – Schematic representation of Bénard cells between two horizontal plates.
The transition from a situation in which the static fluid is a stable state, to that in which motion
develops—i.e. the static case is no longer stable—, is referred to as (onset of the) Rayleigh–Bénard
instability. Since the motion of the fluid appears spontaneously, without the need to impose any
external pressure gradient, it is an instance of free convection or natural convection—in opposition
to forced convection).
Remarks:
∗ Such convection cells are omnipresent in Nature, as e.g. in the Earth mantle, in the Earth
atmosphere, or in the Sun convective zone.
∗ When ∆T further increases, the structure of the convection pattern becomes more complicated,
eventually becoming chaotic.
In a series of experiments with liquid helium or mercury, A. Libchaber(ar) and his collaborators
observed the following features [37, 38, 39]: Shortly above ∆Tc , the stable fluid state involve
cylindrical convective cells with a constant profile. Above a second threshold, “oscillatory convec-
tion” develops: that is, undulatory waves start to propagate along the “surface” of the convective
cells, at first at a unique (angular) frequency ω1 , then—as ∆T further increases—also at higher
harmonics n1 ω1 , n1 ∈ N. As the temperature difference ∆T reaches a third threshold, a second
undulation frequency ω2 appears, incommensurate with ω1 , later accompanied by the combina-
tions n1 ω1 + n2 ω2 , with n1 , n2 ∈ N. At higher ∆T , the oscillator with frequency ω2 experiences a
shift from its proper frequency to a neighboring submultiple of ω1 —e.g. , ω1 /2 in the experiments
with He—, illustrating the phenomenon of frequency locking. For even higher ∆T , submultiples of
ω1 appear (“frequency demultiplication”), then a low-frequency continuum, and eventually chaos.
(aq) (ar)
H. Bénard, 1874–1939. A. Libchaber, born 1934
130 Convective heat transfer
Each appearance of a new frequency may be seen as a bifurcation. The ratios of the experimentally
measured lengths of consecutive intervals between successive bifurcations provide an estimate of
the (first) Feigenbaum constant (as) in agreement with its theoretical value—thereby providing the
first empirical confirmation of Feigenbaum’s theory.
On the other hand, the fluid particle is slowed in its vertical motion by the downwards oriented
Stokes friction force acting on it, namely, in projection on the z-axis
FStokes = −6πRηv. (VIII.16)
Note that assuming that the velocity v remains constant, with a counteracting Stokes force that is
automatically the “good” one, relies on the picture that viscous effects adapt instantaneously, i.e.
that momentum diffusion is fast. That is, the above reasoning actually assumes that the Prandtl
number (VIII.7) is much larger than 1; yet its result is independent from that assumption.
Comparing Eqs. (VIII.15) and (VIII.16), buoyancy will overcome friction, and thus the Rayleigh–
Bénard instability take place, when
4πC R5 α(V ) ∆T α(V ) ∆T gR4 9
ρ0 gv > 6πRρ0 νv ⇔ > .
3 α d ανd 2C
Note that the velocity v which was invoked in the reasoning actually drops out from this condition.
Taking for instance R = d/2—which maximizes the left member of the inequality—, this becomes
α(V ) ∆T g d3 72
Ra ≡ > = Rac .
να C
Ra is the so-called Rayleigh number and Rac its critical value, above which the static-fluid state is
instable against perturbation and convection takes place. The “value” 72/C found with the above
simple reasoning on force equilibrium is totally irrelevant—both careful experiments and theoretical
calculations agree with Rac = 1708 for a fluid between two very large plates—, the important lesson
is the existence of a threshold.
incompressibility equation, but clearly violates the impermeability conditions—, and a similar one
for δΘ, with γ a constant. Substituting these forms in Eqs. (VIII.17) yield the linear system
γv0 = −k 2 νv0 + α(V ) δΘ0 g ⇔ γ + νk 2 v0 − gα(V ) δΘ0 = 0,
∆T ∆T
γδΘ0 = −k 2 αδΘ0 + v0 − γ + αk 2 δΘ0 = 0
v0 ⇔
d d
for the amplitudes v0 , δΘ0 . This admits a non-trivial solution only if
α(V ) ∆T
γ + νk 2 γ + αk 2 −
g = 0. (VIII.18)
d
This is a straightforward quadratic equation for γ. It always has two real solutions, one of which
is negative—corresponding to a dampened perturbation—since their sum is −(α + ν)k 2 < 0; the
other solution may change sign since their product
α(V ) ∆T
ανk 4 − g
d
is positive for ∆T = 0, yielding a second negative solution, yet changes sign as ∆T increases. The
vanishing of this product thus signals the onset of instability. Taking for instance k = π/d to fix
ideas, this occurs at a critical Rayleigh number
α(V ) ∆T g d3
Rac = = π4,
αν
where the precise value (here π 4 ) is irrelevant.
From Eq. (VIII.18) also follows that the growth rate of the instability is given in the neighborhood
of the threshold by
Ra − Rac αν 2
γ= k ,
Rac α + ν
i.e. it is infinitely slow at Rac . This is reminiscent of a similar behavior in the vicinity of the critical
point associated with a thermodynamic phase transition.
By performing a more rigorous calculation including non-linear effects, one can show that the
velocity amplitude at a given point behaves like
β
Ra − Rac 1
v∝ with β = (VIII.19)
Rac 2
in the vicinity of the critical value, and this prediction is borne out by experiments [41]. Such
a power law behavior is again reminiscent of the thermodynamics of phase transitions, more
specifically here—since v vanishes below Rac and is finite above—of the behavior of the order
parameter in the vicinity of a critical point. Accordingly the notation β used for the exponent
in relation (VIII.19) is the traditional choice for the critical exponent associated with the order
parameter of phase transitions.
Eventually, a last analogy with phase transitions regards the breaking of a symmetry at the threshold
for the Rayleigh–Bénard instability. Below Rac , the system is invariant under translations parallel
to the plates, while above Rac that symmetry is spontaneously broken.
When the energy density becomes large—as may happen for instance in compact astrophysical
objects, in the early Universe, or in high-energy collisions of heavy nuclei—the “atoms” constituting
a fluid can acquire very high kinetic energies, that become comparable to their (rest) mass energy.
A non-relativistic description of the medium is then no longer adapted, and must be replaced by a
relativistic model. Some introductory elements of such a description are presented in this Chapter—
in which the basic laws governing the dynamics of relativistic fluids are formulated and discussed—,
and the following one—which will deal with a few simple analytically tractable solutions of the
equations.
As in the non-relativistic case, the basic equations governing the motion of a fluid in the rel-
ativistic regime are nothing but formulations of the most fundamental laws of physics, namely
conservation laws for “particle number”—in fact, for the conserved quantum numbers carried by
particles—, and for energy and momentum (Sec. IX.1).
Precisely because the equations simply express general conservation laws, they are not very
specific, and contain at first too many degrees of freedom to be tractable. To make progress, it is
necessary to introduce models for the fluid under consideration, leading for instance to distinguishing
between perfect and dissipative fluids. A convenient way to specify the constitutive equations
characteristic of such models is to do so in terms of a fluid four-velocity, which generalizes the
non-relativistic flow velocity, yet in a non-unique way (Sec. IX.2).
Such a fluid four-velocity also automatically singles out a particular reference frame, the local
rest frame, in which the conserved currents describing the physics of the fluid take a simpler form,
whose physical interpretation is clearer. The perfect fluids are thus those whose properties at each
point are spatially isotropic in the corresponding local rest frame, from which there follows that
the conserved currents can only depend on the flow four-velocity, not on its derivatives (Sec. IX.3).
Conversely, when the conserved currents involve (spatial) gradients of the fluid four-velocity, these
derivatives signal (real) fluids with dissipative effects (Sec. IX.4).
Two topics that lie beyond the main stream of this Chapter are given in appendices, namely the
expression of the conserved currents of relativistic fluid dynamics in terms of underlying microscopic
quantities (Sec. IX.A) and a discussion of relativistic kinematics (Sec. IX.B).
Throughout this Chapter and the next one, the fluids occupy domains of the four-dimensional
Minkowski space-time M 4 of Special Relativity. The position of a generic point of M 4 will be
designated by a 4-vector x. Given a reference frame R and a system of coordinates, those of x will
be denoted by {xµ } ≡ (x0 , x1 , x2 , x3 )—where in the case of Minkowski coordinates(50) x0 = ct with
t the time measured by an observer at rest in R.
(50)
We shall call Minkowski coordinates the analog on the space-time M 4 of the Cartesian coordinates on Euclidean
space E 3 , i.e. those corresponding to a set of four mutually orthogonal 4-vectors (e0 , e1 , e2 , e3 ) such that the
metric tensor has components gµν = eµ · eν = diag(−1, +1, +1, +1) for µ, ν = 0, 1, 2, 3. They are also alternatively
referred to as Lorentz coordinates.
134 Fundamental equations of relativistic fluid dynamics
For the metric tensor g on M 4 , we use the “mostly plus” convention, with signature (−, +, +, +),
i.e. in the case of Minkowski coordinates x0 = −x0 while xi = xi for i = 1, 2, 3. Thus, time-like
resp. space-like 4-vectors have a negative resp. positive semi-norm.
dµ N µ (x) = 0, (IX.2a)
where dµ ≡ d /dxµ denote the components of the 4-gradient. Denoting the latter, which is a
one-form, by d, one may write the even shorter “geometric” (i.e. coordinate-invariant) equation
d · N(x) = 0, (IX.2b)
∗ More generally, one associates to each conserved additive quantum number a 4-current J(x) with
components J µ (x), obeying a similar conservation equation d · J(x) = 0 resp. dµ J µ (x) = 0.
∗ If Minkowski coordinates {xµ } are used, the components of the 4-gradient d are simply the partial
derivatives ∂µ ≡ ∂ /∂xµ , so that Eq. (IX.2a) becomes ∂µ N µ (x) = 0.
6
6
6
6
6 6
t6
x 1
-
x2 Σ
Figure IX.1
(51)
A better argument is to introduce the particle number 4-current from a microscopic definition, see App. IX.A.1.
136 Fundamental equations of relativistic fluid dynamics
The total (net) number N of particles in the fluid is the flux of the particle number 4-current
N(x) across Σ Z Z
N = N (x) d σµ = N(x) · d3 σ,
µ 3
(IX.3)
Σ Σ
where the right member vanishes thanks to the local expression (IX.2) of particle number conser-
vation. Splitting ∂Ω into two space-like parts through which particles enter resp. leave Ω in their
motion—the technical criterion is the sign of N(x) · d3 σ—, one finds that there are as many particles
that leave as those that enter, which expresses particle number conservation globally.
All physical quantities are to be measured with respect to the reference frame R.
Remarks:
∗ The similarity of the notations T resp. T for the energy-momentum four-tensor resp. the three-
dimensional momentum flux-density tensor is not accidental! The former is the natural general-
ization to the 4-dimensional relativistic framework of the latter, just like four-momentum p, with
components pµ is the four-vector associated to the three-dimensional momentum p~. That is, the
3-tensor T is the spatial part of the 4-tensor T , just like the momentum p~ is the spatial part of
four-momentum p.
∗ Starting from a microscopic description of the fluid, one can show that the energy-momentum
tensor is symmetric, i.e. T µν (x) = T νµ (x) for all µ, ν = 0, 1, 2, 3.
(52)
This choice is not universal: the alternative convention 0123 = +1 results in 0123 < 0 due to the odd number of
minus signs in the signature of the metric tensor.
(53)
As in the case of the particle number 4-current, the argument showing that T (x) is a Lorentz tensor is to define
it microscopically as a tensor—see App. IX.A.2—and to later interpret the physical meaning of the components.
(lxxviii)
Energieimpulstensor
IX.2 Four-velocity of a fluid flow. Local rest frame 137
In the absence of external force acting on the fluid, the local conservation of the energy-
momentum tensor reads component-wise
dµ T µν (x) = 0 ∀ν = 0, 1, 2, 3, (IX.7a)
which represents four equations: the equation with ν = 0 is the conservation of energy, while the
equations dµ T µj (x) = 0 for j = 1, 2, 3 are the components of the momentum conservation equation.
In geometric formulation, Eq. (IX.7a) becomes
d · T (x) = 0. (IX.7b)
This is exactly the same form as Eq. (IX.2b), just like Eqs. (IX.2a) and (IX.7a) are similar, up to
the difference in the tensorial degree of the conserved quantity.
As in § IX.1.1 b, one associates to the energy-momentum tensor T (x) a 4-vector P by
Z Z
P ≡ T (x) · d σ ⇔ P = T µν (x) d3 σν ,
3 µ
(IX.8)
Σ Σ
with Σ a space-like 3-hypersurface. P represents the total 4-momentum crossing Σ, and invoking
the Gauss theorem, Eq. (IX.7) implies that it is a conserved quantity.
Let ~v(x) denote the instantaneous velocity of (an observer at rest in) the local rest frame LR(x)
with respect to a fixed reference frame R. In the latter, the components of the flow four-velocity
are !
γ(x)c
uµ (x) =
, (IX.11)
R γ(x)~v(x)
p
with γ(x) = 1/ 1 −~v(x)2 /c2 the corresponding Lorentz factor.
The local rest frame represents the reference frame in which the local thermodynamic variables
of the system—particle number density n (x) and energy density (x)—are defined in their usual
sense:
n (x) ≡ n(x) (x) ≡ T 00 (x)
, . (IX.12)
LR(x) LR(x)
For the remaining local thermodynamic variables in the local rest frame, it is assumed that they
are related to n (x) and (x) in the same way, as when the fluid is at thermodynamic equilibrium.
Thus, the pressure P (x) is given by the mechanical equation of state
P (x) = P ((x), n (x)); (IX.13)
LR(x)
the temperature T (x) is given by the thermal equation of state; the entropy density s(x) is defined
by the Gibbs fundamental relation, and so on.
(lxxix)
lokales Ruhesystem
138 Fundamental equations of relativistic fluid dynamics
Remarks:
∗ A slightly more formal approach to define 4-velocity and local rest frame is to turn the reasoning
round. Namely, one introduces the latter first as a reference frame LR(x) in which “physics at point
x is easy”, that is, in which the fluid is locally motionless. Introducing then an instantaneous inertial
reference frame that momentarily coincides with LR(x), one considers an observer O who is at rest
in that inertial frame. The four-velocity of the fluid u(x), with respect to some fixed reference frame
R, is then the four-velocity of O in R—defined as the derivative of O’s space-time trajectory with
respect to his proper time.
The remaining issue is that of the local absence of motion which defines LR(x). In particular,
there must be no energy flow, i.e. T 0j (x) = 0. One thus looks for a time-like eigenvector u(x) of
the energy-momentum tensor T (x):
with − < 0 the corresponding eigenvalue and u(x) normalized to c. Writing that, thanks to the
symmetry of T (x), u(x) is also a left-eigenvector, i.e. uµ (x)T µν(x) = − uν (x), one finds that the
energy flux density vanishes in the reference frame in which the Minkowski components of u(x)
have the simple form (IX.10). This constitutes an appropriate choice of local rest frame, and one
has at the same time the corresponding four-velocity u(x).
∗ The relativistic energy density differs from its at first sight obvious non-relativistic counterpart,
the internal energy density e. The reason is that also contains the contribution from the mass
energy of the particles and antiparticles—mc2 per (anti)particle—, which is conventionally not taken
into account in the non-relativistic internal energy density.
∗ To distinguish between the reference frame dependent quantities, like particle number density
n(x) or energy density T 00 (x), and the corresponding quantities measured in the local rest frame,
namely n (x) or (x), the latter are referred to as comoving.
The comoving quantities can actually be computed easily within any reference frame and coor-
dinate system. Writing thus
where we used that u0 (x) = g00 (x)u0 (x) in the local rest frame, the rightmost term of the above
identity is a ratio of two Lorentz-invariant scalars, thus itself a Lorentz scalar field, independent of
the reference frame in which it is computed:
In the following Sections, we introduce fluid models, defined by the relations between the con-
served currents—particle number 4-current N(x) and energy-momentum tensor T (x)—and the fluid
4-velocity u(x) and comoving thermodynamic quantities.
IX.3 Perfect relativistic fluid 139
and
T 00 (x)
= (x),
LR(x)
T ij (x) = P (x)δ ij ,
∀i, j = 1, 2, 3 (IX.16b)
LR(x)
T i0 (x) = T 0j (x)
= 0, ∀i, j = 1, 2, 3
LR(x) LR(x)
where the definitions (IX.12) were taken into account, while P (x) denotes the pressure. In matrix
form, the energy-momentum tensor (IX.16b) becomes
(x) 0 0 0
0 P (x) 0 0
T µν (x)
= . (IX.16c)
LR(x) 0 0 P (x) 0
0 0 0 P (x)
Remark: The identification of the diagonal spatial components with a “pressure” term is warranted
by the physical interpretation of the T ii (x). Referring to it as “the” pressure anticipates the fact that
it behaves as the thermodynamic quantity that is related to energy density and particle number by
the mechanical equation of state of the fluid.
In an arbitrary reference frame and allowing for the possible use of curvilinear coordinates, the
components of the particle number 4-current and the energy-momentum tensor of a perfect fluid
are
N µ (x) = n (x)uµ (x) (IX.17a)
and
uµ (x) uν (x)
T µν (x) = P (x)g µν (x) + (x) + P (x)
(IX.17b)
c2
Relation (IX.17a) resp. (IX.17b) is an identity between the components of two 4-vectors resp. two
4-tensors, which transform identically under Lorentz transformations—i.e. changes of reference
frame—and coordinate basis changes. Since the components of these 4-vectors resp. 4-tensors
are equal in a given reference frame—the local rest frame—and a given basis—that of Minkowski
coordinates—, they remain equal in any coordinate system in any reference frame.
In geometric formulation, the particle number 4-current and energy-momentum tensor respec-
tively read
N(x) = n (x)u(x) (IX.18a)
and
u(x) ⊗ u(x)
T (x) = P (x)g−1 (x) + (x) + P (x)
. (IX.18b)
c2
The latter is very reminiscent of the 3-dimensional non-relativistic momentum flux density (III.22);
similarly, the reader may also compare the component-wise formulations (III.21b) and (IX.17a).
Remarks:
∗ The energy-momentum tensor is obviously symmetric—which is a non-trivial physical statement.
For instance, the identity T i0 = T 0i means that (1/c times) the energy flux density in direction
i equals (c times) the density of the i-th component of momentum—where one may rightly argue
that the factors of c are historical accidents in the choice of units. This is possible in a relativistic
theory only because the energy density also contains the mass energy.
∗ In Eq. (IX.17b) or (IX.18b), the sum (x) + P (x) is equivalently the enthalpy density w(x).
∗ Equation (IX.17b), (IX.18b) or (IX.19a) below represents the most general symmetric 20 -tensor
that can be constructed using only the metric tensor and the 4-velocity.
The component form (IX.17b) of the energy-momentum tensor can trivially be recast as
uµ (x) uν (x)
T µν (x) = (x) + P (x)∆µν (x) (IX.19a)
c2
with
uµ (x) uν (x)
∆µν (x) ≡ g µν (x) + (IX.19b)
c2
the components of a tensor ∆ which—in its 11 -form—is actually a projector on the 3-dimensional
vector space orthogonal to the 4-velocity u(x), while uµ (x) uν (x)/c2 projects on the time-like direction
of the 4-velocity.
One easily checks the identities ∆µν (x)∆νρ (x) = ∆µρ (x) and ∆µν (x)uν (x) = 0.
From Eq. (IX.19a) follows at once that the comoving pressure P (x) can be found in any reference
frame as
1
P (x) = ∆µν (x)T µν (x). (IX.20)
3
which complements relations (IX.14) and (IX.15) for the particle number density and energy density,
respectively.
Remark: Contracting the energy-momentum tensor T with the metric tensor twice yields a scalar,
the so-called trace of T
T (x) : g(x) = T µν (x)gµν (x) = T µµ (x) = 3P (x) − (x). (IX.21)
IX.3 Perfect relativistic fluid 141
dµ s(x)uµ (x) = 0
(IX.22)
Proof: The relation U = T S − PV + µN N with U resp. µN the internal energy resp. the chemical
potential gives for the local thermodynamic densities = T s−P +µNn . Inserting this expression
of the energy density in Eq. (IX.17b) yields (dropping the x variable for the sake of brevity):
uµ uν uν
T µν = P g µν + (T s + µNn ) 2 = P g µν + T (suµ ) + µN (n uµ ) 2 .
c c
Taking the 4-gradient dµ of this identity gives
dµ uν uµ uν ν
µ u
dµ T µν = dνP + T (suµ )+µN (n uµ ) n µ
n
+ s d µ T + dµ µ N
+ T dµ (su )+µ N
dµ ( u ) .
c2 c2 c2
Invoking the energy-momentum conservation equation (IX.7), the leftmost member of this iden-
tity vanishes. The second term between square brackets on the right hand side can be rewritten
with the help of the Gibbs–Duhem relation as s dµ T + n dµ µN = dµ P . Eventually, the parti-
cle number conservation formulation (IX.7) can be used in the rightmost term. Multiplying
everything by uν yields
uν dµuν uµ uν uν µ uν u
ν
0 = uν dνP + T (suµ ) + µN (n uµ )
+ (d µ P ) + T d µ (su ) .
c2 c2 c2
The constant normalization uν uν = −c2 of the 4-velocity implies uν dµuν = 0 for µ = 0, . . . , 3,
so that the equation becomes
0 = uν dνP − (dµ P )uµ − T dµ (suµ ),
leading to dµ (suµ ) = 0.
Proof: the total time derivative of the entropy per particle reads
d s ∂ s ~ s 1 s
= +~v · ∇ = u·d ,
dt n ∂t n n γ n
where the second identity makes use of Eq. (IX.11), with γ the Lorentz factor. The rightmost
term is then
s 1 s 1
s
u·d = u · ds − 2 u · dn = u · ds − u · dn .
n n n n n
The continuity equation d · (n u) = 0 gives u · dn = −n d · u, implying
d s 1 s 1 1
= u·d = u · ds + s d · u = d · (su) = 0,
dt n γ n γn γn
where the last identity expresses the conservation of entropy.
142 Fundamental equations of relativistic fluid dynamics
Throughout the Section, we shall omit for the sake of brevity the variables x resp. (t, ~r) of the
various fields. In addition, we adopt for simplicity a system of Minkowski coordinates.
IX.3.3 a Particle number conservation
::::::::::::::::::::::::::::::::::::::
The 4-velocity components (IX.25) give for those of the particle number 4-current (IX.17a)
!
µ nc
N ∼ .
|~v|c n~v
2
00 2 2 ~v2 2
T = −P + γ (ρc + e + P ) ∼ ρc + e + ρ~v + O 2 ; (IX.27a)
|~v|c c
vj j
3
0j j0 2 2 j 2 v |~v|
T = T = γ (ρc + e + P ) ∼ ρcv + e + P + ρ~v +O 3 ; (IX.27b)
c |~v|c c c
vi vj ~v2
2
~v
T ij ij 2 2 ij i j
= P g + γ (ρc + e + P ) 2 ∼ P g + ρ v v + O 2 = T + O 2 . ij
(IX.27c)
c |~v|c c c
In the last line, we have introduced the components Tij , defined in Eq. (III.21b), of the three-
dimensional momentum flux-density tensor for a perfect non-relativistic fluid. As emphasized below
Eq. (IX.23), the internal energy density and pressure in the rightmost terms of the first or second
equations are of the same order of magnitude as the term ρ~v2 with which they appear, i.e. they are
always part of the highest-order term.
Momentum conservation
Considering first the components (IX.27b), (IX.27c), the low-velocity limit of the relativistic
momentum-conservation equation ∂µ T µj = 0 for j = 1, 2, 3 reads
3 3
1 ∂(ρcvj ) X ∂T Tij
2
∂(ρvj ) X ∂T Tij
2
~v ~v
0= + i
+O 2 = + i
+O 2 . (IX.28)
c ∂t ∂x c ∂t ∂x c
i=1 i=1
This is precisely the conservation-equation formulation (III.24a) of the Euler equation in absence
of external volume forces.
Energy conservation
Given the physical interpretation of the components T 00 , T i0 with i = 1, 2, 3, the component
ν = 0 of the energy-momentum conservation equation (IX.7), ∂µ T µ0 = 0, should represent the
conservation of energy.
As was mentioned several times, the relativistic energy density and flux density actually also
contain a term from the rest mass of the fluid constituents. Thus, the leading order contribution to
∂µ T µ0 = 0, coming from the first terms in the right members of Eqs. (IX.27a) and (IX.27b), is
3
∂(ρc) X ∂(ρcvi )
2
~v
0= + i
+O 2 ,
∂t ∂x c
i=1
that is, up to a factor c, exactly the continuity equation (III.9), which was already shown to be the
low-velocity limit of the conservation of the particle-number 4-current.
To isolate the internal energy contribution, it is thus necessary to subtract that of mass energy.
In the fluid local rest frame, relation (IX.23) shows that one must subtract ρc2 from . The former
simply equals ρcu0 |LR , while the latter is the component µ = 0 of T µ0 |LR , whose space-like compo-
nents vanish in the local rest frame. To fully subtract the mass energy contribution in any frame
from both the energy density and flux density, one should thus consider the 4-vector T µ0 − ρcuµ .
Accordingly, instead of simply using ∂µ T µ0 = 0, one should start from the equivalent—thanks
to Eq. (IX.2) and the relation ρ = mn —equation ∂µ (T µ0 − ρcuµ ) = 0. With the approximations
2
0 2 2 1 2 ~v
ρcu = γρc = ρc + ρ~v + O 2
2 c
and
1 2 vj
5
j j j |~v|
ρcu = γρcv = ρcv + ρ~v +O 3
2 c c
one finds
X3 j 2
µ0 µ
1∂ 1 2 ∂ 1 2 v ~v
0 = ∂µ T − ρcu = ρ~v + e + j
ρ~v + e + P +O 2 ,
c ∂t 2 ∂x 2 c c
j=1
144 Fundamental equations of relativistic fluid dynamics
that is
∂ 1 2 ~ 1 2
ρ~v + e + ∇ · ρ~v + e + P ~v ≈ 0. (IX.29)
∂t 2 2
This is the non-relativistic local formulation of energy conservation (III.33) for a perfect fluid in
absence of external volume forces. Since that equation had been postulated in Section III.4.1, the
above derivation may be seen as its belated proof.
IX.3.3 c Entropy conservation
::::::::::::::::::::::::::::::
Using the approximate 4-velocity components (IX.25), the entropy conservation equation (IX.22)
becomes in the low-velocity limit
3
µ 1 ∂(sc) X ∂(svi ) ∂s ~
0 = ∂µ (su ) ≈ + i
= + ∇ · (s~v), (IX.30)
c ∂t ∂x ∂t
i=1
i.e. gives the non-relativistic equation (III.34).
with g µν(x) the components of the inverse metric tensor g−1 (x). For the comprehension it is im-
portant to realize that ∆ plays the role of the identity in the 3-space orthogonal to u(x).
IX.4 Dissipative relativistic fluids 145
In analogy with Eqs. (IX.17a), (IX.18), and (IX.19a), one thus writes
or equivalently
N(x) = n (x)u(x) + n(x) (IX.33b)
and
T µν(x) = (x)uµ (x)uν (x) + P (x)∆µν(x) + τ µν(x) (IX.34a)
i.e., in geometric form
T(x) = (x)u(x)⊗ u(x) + P (x)∆
∆(x) + τ(x). (IX.34b)
The precise physical content and mathematical form of the additional terms can now be further
specified.
Tensor algebra
:::::::::::::::
In order for n (x) to represent the (net) comoving particle density, the dissipative 4-vector n(x)
may have no timelike component in the the local rest frame LR(x) defined by the 4-velocity, see
definition (IX.12). Accordingly, the condition
uµ (x)nµ (x)
=0
LR(x)
must hold in the local rest frame. Since the left hand side of this identity is a Lorentz scalar, it
holds in any reference frame or coordinate system:
Equations (IX.33a), (IX.33) thus represent the decomposition of a 4-vector in a component parallel
to the flow 4-velocity and a component orthogonal to it. In keeping, one can write
Physically, n(x) represents a diffusive particle-number 4-current in the local rest frame, which de-
scribes the non-convective transport of particle number.
Similarly, the dissipative energy-momentum current $(x) can have no 00-component in the local
rest frame, to ensure that T 00 (x) in that frame still define the comoving energy density (x). This
means that the components τ µν (x) may not be proportional to the product uµ (x)uν (x). The most
general symmetric tensor of degree 2 which obeys that condition is of the form
with q µ (x) resp. π µν(x) the components of a 4-vector q(x) resp. π (x) such that
and
uµ (x)π µν (x)uν (x) = u(x) · π (x) · u(x) = 0. (IX.36c)
Condition (IX.36b) expresses that q(x) is a 4-vector orthogonal to the 4-velocity u(x), which physi-
cally represents the heat current or energy flux density in the local rest frame.
In turn, the symmetric tensor π (x) can be decomposed into the sum of a traceless tensor $ (x)
with components $µν(x) and a tensor proportional to the projector (IX.19b) orthogonal to the
4-velocity
π µν(x) = $µν(x) + Π(x)∆µν(x). (IX.36d)
146 Fundamental equations of relativistic fluid dynamics
The tensor $ (x) is the shear stress tensor in the local rest frame of the fluid, that describes the
transport of momentum due to shear deformations. Eventually, Π(x) represents a dissipative pressure
term, since it behaves as the thermodynamic pressure P (x) as shown by Eq. (IX.37) below.
All in all, the components of the energy-momentum tensor in a dissipative relativistic fluid may
thus be written as
T µν(x) = (x)uµ (x)uν (x) + P (x) + Π(x) ∆µν(x) + q µ (x)uν(x) + q ν(x)uµ (x) + $µν(x), (IX.37a)
∗ Let aµν denote the (contravariant) components of an arbitrary 20 -tensor. One encounters in the
• A first natural possibility, proposed by Eckart(at) [44], is to take the 4-velocity proportional
to the particle-number 4-current,(55) namely
N µ (x)
uµEckart(x) ≡ p . (IX.39)
Nν (x)N ν (x)
Accordingly, the dissipative particle-number flux n(x) vanishes automatically, so that the
expression of particle-number conservation is simpler with that choice.
The local rest frame associated with the flow 4-velocity (IX.39) is then referred to as Eckart
frame
A drawback of that definition of the fluid 4-velocity is that the net particle number can possibly
vanish in some regions of a given flow, so that uEckart (x) is not defined unambiguously in such
domains.
or equivalently
T µν (x)uνLandau(x)
uµLandau(x) = q . (IX.40b)
uρLandau(x)Tρσ (x)uσLandau(x)
With this choice, which in turn determines the Landau frame, the heat current q(x) vanishes,
so that the dissipative tensor τ (x) satisfies the condition
Eventually, one may of course choose to work with a general 4-velocity u(x), and thus to keep
both the diffusive particle-number current and the heat flux density in the dynamical fields (IX.33)–
(IX.37).
In turn, the conservation of the energy momentum tensor (IX.7), projected perpendicular to resp.
along the 4-velocity, gives
∆ρν (x)dµ T µν (x) = (x) + P (x) uµ (x)dµ uρ (x) + ∇ρ (x)P (x) + ∆ρν (x)dµ π µν (x) = 0
(IX.42b)
resp.
uν (x)dµ T µν (x) = −uµ (x)dµ (x) − (x) + P (x) dµ uµ (x) + uν (x)dµ π µν (x) = 0.
where the second equality follows from condition (IX.40c) with τ µν = π µν (since q = 0). Using the
identity dµ = uµ (u · d) + ∇µ , and again the condition uν π µν = 0, this becomes
uµ (x)dµ (x) + (x) + P (x) dµ uµ (x) + π µν (x)∇µ (x)uν (x) = 0.
(IX.42c)
Equations (IX.42a)–(IX.42c) represent the relations governing the dynamics of a dissipative fluid in
the Landau frame.
Remark: If one adopts Eckart’s choice of velocity, the resulting equations of motion differ from those
given here—for instance, the third term d · n(x) in Eq. (IX.42a) drops out, since n(x) = 0—, yet
they are physically totally equivalent.
Entropy law in a dissipative relativistic fluid
:::::::::::::::::::::::::::::::::::::::::::::
Combining the dynamical equation (IX.42c) with the thermodynamic relations + P = T s+µNn
and d = T ds + µN dn , one finds
T (x)dµ s(x)uµ (x) = −π µν (x)∇µ (x)uν (x) + µN (x)dµ nµ (x)
(56)
This choice of form for u(x) is often announced as “let us work in the Landau frame”, where frame is to be
understood in its sense of framework.
IX.4 Dissipative relativistic fluids 149
Using the symmetry of π µν , one can replace ∇µ uν by its symmetric part 21 (∇µ uν + ∇ν uµ ) in
the first term on the right hand side. With the decompositions π µν = $µν + Π∆µν [Eq. (IX.36d)]
and
1 1 2 1 1
∇µ uν + ∇ν uµ = ∇µ uν + ∇ν uµ − ∆µν ∇ · u + ∆µν ∇ · u ≡ S µν + ∆µν ∇ · u ,
2 2 3 3 3
where the S µν are the components of a traceless tensor(57) —comparing with Eq. (II.15d), this is the
rate-of-shear tensor—, while ∇ · u is the (spatial) 3-divergence of the 4-velocity field, one finds
$µν (x)
µ µN (x) µ Π(x) µ µN (x)
dµ s(x)u (x) − n (x) = − S µν (x) − ∇ (x) · u(x) − n (x)∇µ . (IX.43b)
T (x) T (x) T (x) T (x)
The left member of this equation is the 4-divergence of the entropy 4-current S(x), with components
S µ (x), comprising on the one hand the convective transport of entropy—which is the only contribu-
tion present in the perfect-fluid case, see Eq. (IX.22)—, and on the other hand a contribution from
the dissipative particle-number current.
Remark: When working in the Eckart frame, the dissipative particle-number current no longer
contributes to the entropy 4-current S—which is obvious since n vanishes in that frame!—, but the
heat 4-current q does. In an arbitrary frame—i.e. using a different choice of fluid 4-velocity and
thereby of local rest frame—, both n and q contribute to S and to the right hand side of Eq. (IX.43b).
Let Ω be the 4-volume that represents the space-time trajectory of the fluid between an initial
and a final times. Integrating Eq. (IX.43b) over Ω while using the same reasoning as in § IX.1.1 b,
one sees that the left member will yield the change in the total entropy of the fluid during these two
times. This entropy variation must be positive to ensure that the second law of thermodynamics
holds. Accordingly, one requests that the integrand be positive: dµ S µ (x) ≥ 0. This requirement
can be used to build models for the dissipative currents.
Working like in Sec. IX.4.3 in the Landau frame,(58) in which the heat flux density q(x) vanishes,
the simplest possibility that satisfies all constraints is to require
n (x)T (x) 2 µ
µ µN (x)
n (x) = κ(x) ∇ (x) (IX.44c)
(x)+ P (x) T (x)
for the dissipative particle-number 4-current, with η, ζ, κ three positive numbers—which depend on
the space-time position implicitly, inasmuch as they vary with temperature and chemical potential.
The first two ones are obviously the shear and bulk viscosity coefficients, respectively, as hinted
at by the similarity with the form (III.26f) of the shear stress tensor of a Newtonian fluid in the
non-relativistic case. Accordingly, the equation of motion (IX.42b) in which the dissipative stress
tensor is substituted by π µν = $µν + Π∆µν with the forms (IX.44a), (IX.44b) yields the relativistic
version of the Navier–Stokes equation.
What is less obvious is that κ in Eq. (IX.44c) does correspond to the heat conductivity—which
explains why the coefficient in front of the gradient is written in a rather contrived way.
Inserting the dissipative currents (IX.44) in the entropy law (IX.43b), the latter becomes
Π(x)2 (x)+ P (x) 2 n(x)2
$(x)
$ (x) :$
d · S(x) = + + . (IX.45)
2η(x)T (x) ζ(x)T (x) n (x)T (x) κ(x)T (x)
Since n(x) is space-like, the right hand side of this equation is positive, as it should.
The constitutive equations (IX.44) only involve first order terms in the derivatives of velocity,
temperature, or chemical potential. In keeping, the theory constructed with such Ansätze is referred
to as first order dissipative fluid dynamics—which is the relativistic generalization of the set of laws
valid for Newtonian fluids.
This simple relation to the non-relativistic case, together with the fact that only 3 transport
coefficients are needed—when working in the Landau or Eckart frames, in the more general case,
one needs 4 coefficients—makes first-order dissipative relativistic fluid dynamics attractive. The
theory suffers however from a severe issue, which does not affect its non-relativistic counterpart.
Indeed, it has been shown that many solutions of the relativistic Navier–Stokes(–Fourier) equations
are unstable against small perturbations [46]. Such disturbances will grow exponentially with time,
on a microscopic typical time scale. As a result, the velocity of given modes can quickly exceed
the speed of light, which is of course unacceptable in a relativistic theory. In addition, gradients
also grow quickly, leading to the breakdown of the small-gradient assumption that underlies the
construction of first-order dissipative fluid dynamics. This exponential growth of perturbation is
especially a problem for numerical implementations of the theory, in which rounding errors which
quickly propagate.
Violations of causality actually occur for short-wavelength modes, which from a physical point
of view should not be described by fluid dynamics since they involve length scales on which the
system is not “continuous”. As such, the issue is more mathematical than physical. These modes
(58)
The corresponding formulae for Π, $µν and q µ valid in the Eckart frame, in which n vanishes, can be found e.g.
in Ref. [45, Sec. 2.4].
IX.4 Dissipative relativistic fluids 151
do however play a role in numerical computations, so that there is indeed a problem when one
is not working with an analytical solution.
In second order dissipative relativistic fluid dynamics, the most general form for the additional
4-vector Q(x) contributing to the entropy density is [47, 48, 49]
β0 (x)Π(x)2 + β1 (x)qN (x)2 + β2 (x)$
$(x) :$
$(x) α0 (x) α1 (x)
Q(x) = u(x) − Π(x)qN (x) − $ (x)·qN (x),
2T (x) T (x) T (x)
(IX.48a)
where
(x) + P (x)
qN (x) ≡ q(x) − n(x);
n (x)
component-wise, this reads
β0 (x)Π(x)2 + β1 (x)qN (x)2 + β2 (x)$νρ (x)$νρ(x) µ α0 (x) α1 (x) µ
Qµ (x) = u (x) − Π(x)qNµ (x)− $ (x)qNρ (x).
2T (x) T (x) T (x) ρ
(IX.48b)
The 4-vector Q(x) is now quadratic (“of second order”) in the dissipative currents—in the wider
sense—q(x), n(x), Π(x) and $ (x), and involves 5 additional coefficients depending on temperature
and particle-number density, α0 , α1 , β0 , β1 , and β2 .
Substituting this form of Q(x) in the entropy 4-current (IX.47), the simplest way to ensure
that its 4-divergence be positive is to postulate linear relationships between the dissipative currents
152 Fundamental equations of relativistic fluid dynamics
and the gradients of velocity, chemical potential (or rather of −µN /T ), and temperature (or rather,
1/T ), as was done in Eqs. (IX.44). This recipe yields differential equations for Π(x), $ (x), qN (x),
representing 9 coupled scalar equations of motion. These describe the relaxation—with appropriate
characteristic time scales τΠ , τ$ , τqN respectively proportional to β0 , β2 , β1 , while the involved “time
derivative” is that in the local rest frame, u · d—, of the dissipative currents towards their first-order
expressions (IX.44).
Adding up the new equations to the usual ones (IX.2) and (IX.7), the resulting set of equations,
known as (Müller(aw) –)Israel(ax) –Stewart(ay) theory, is no longer plagued by the issues that affects
the relativistic Navier–Stokes–Fourier equations.
• Landau–Lifshitz [3, 4], Chapter XV, § 133,134 (perfect fluid) and § 136 (dissipative fluid);
• Romatschke [51];
or component-wise
N Z
uµk (τ )δ (4) xν−xνk (τ ) d(cτ ) for µ = 0, 1, 2, 3,
X
N µ (x) ≡
(IX.A.1b)
k=1
where the k-th integral in either sum is along the world-line of particle k. The right hand sides of
these equations clearly define a 4-vector resp. its components. For the latter, some simple algebra
yields the identities
N
1 0 X
δ (3) ~r − ~xk (t) ,
N (t,~r) = (IX.A.2a)
c
k=1
N
X
N i (t,~r) = vki (t)δ (3) ~r − ~xk (t)
(IX.A.2b)
k=1
Inspecting the right hand sides of relations (IX.A.2), they obviously represent the particle num-
ber density and flux density for the system Σ, respectively.
154 Fundamental equations of relativistic fluid dynamics
where the k-th integral in the sum is along the world-line of particle k, as above; component-wise,
this gives
N Z
pµk (τ )uνk (τ )δ (4) xλ−xλ (τ ) d(cτ ) for µ, ν = 0, 1, 2, 3.
X
µν
T (x) ≡ (IX.A.3b)
k=1
2
The members of these equations clearly define a Lorentz tensor of type 0 resp. its components.
Repeating the same derivation as that leading to Eq. (IX.A.2a), one shows that
N
pµk (t)cδ (3) ~r − ~xk (t) .
X
T µ0 (t,~r) =
(IX.A.4a)
k=1
Recognizing in p0k c the energy of particle k, T 00 represents the energy density of the system Σ—
under the assumption that the potential energy associated with the interaction between particles is
much smaller than their mass and kinetic energies—, while T i0 for i = 1, 2, 3 represents c times the
density of the i-th component of momentum. In turn,
N
p0k (t)vkj (t)δ (3) ~r − ~xk (t)
X
0j
T (t,~r) = (IX.A.4b)
k=1
with j ∈ {1, 2, 3} is the 1/c times the j-th of the energy flux density of the collection of particles.
Eventually, for i, j = 1, 2, 3
N
pik (t)vkj (t)δ (3) ~r − ~xk (t)
X
T ij (t,~r) =
(IX.A.4c)
k=1
is clearly the j-th component of the flux density of momentum along the i-th direction.
Remark: Invoking the relation p = mu between the 4-momentum, mass and 4-velocity of a (massive!)
particle shows at once that the energy-momentum tensor (IX.A.3) is symmetric.
(az)
J. D. Bjorken, born 1934
156 Flows of relativistic fluids
Appendices
A PPENDIX A
To be written!
U = T S − PV + µN (A.1)
dU = T dS − P dV + µ dN (A.2)
e + P = T s + µn (A.3)
de = T ds + µ dn . (A.4)
dP = s dT + n dµ (A.5)
Die letztere Gleichung folgt aus
U 1 U
de = d = dU − 2 dV
V V V
T P µ TS P µN S N
= dS − dV + dN − 2 dV + dV − 2 dV = T d + µd ,
V V V V V V V V
wobei die Relation dU = T dS − P dV + µ dN benutzt wurde.
A PPENDIX B
Tensors on a vector space
In this Appendix, we gather mathematical definitions and results pertaining to tensors. The purpose
is mostly to introduce the “modern”, geometrical view on tensors, which defines them by their action
on vectors or one-forms, i.e. in a coordinate-independent way (Sec. B.1), in contrast to the “old”
definition based on their behavior under basis transformations (Sec. B.2).
The reader is assumed to already possess enough knowledge on linear algebra to know what are
vectors, linear (in)dependence, (multi)linearity, matrices. . . Similarly, the notions of group, field,
application/function/mapping. . . are used without further mention.
In the remainder of these lecture notes, we actually consider tensors on real vector spaces, i.e.
for which the underlying base field K of scalars is the set R of real numbers; here we remain more
general. Einstein’s summation convention is used throughout.
B.1.2 One-forms
. . . on a vector space V are the linear applications, hereafter denoted as h, from V into the
base field of scalars K. e
The set of 1-forms on V , equipped with the “natural” addition and scalar multiplication, is itself
a vector space over the field K, denoted by V ∗ and said to be dual to V .
If V is finite-dimensional, so is V ∗ , with dim V ∗ = dim V . Given a basis B = {~ei } in V , one
can then construct its dual basis B ∗ = {j } in V ∗ such that
e
j (~ei ) = δij , (B.2)
e
where δij denotes the usual Kronecker delta symbol.
B.1 Vectors, one-forms and tensors 161
B.1.3 Tensors
argument. The integer m + n is the order (or often, but improperly, rank ) of the tensor.
Already known objects arise as special cases of this definition when either m or n is zero:
• the 00 -tensors are simply the scalars of the base field K;
(61)
This will be introduced 4 lines further down as a condition on the matrix with elements gij , which is equivalent
to stating that for every non-vanishing vector ~a there exists ~b such that g(~a, ~b) 6= 0.
(62)
More precisely, an inner product if g is (positive or negative) definite, a semi-inner product otherwise.
B.1 Vectors, one-forms and tensors 163
for all i, j—and if the square matrix with elements gij is regular. The number g(~a, ~b) is then also
denoted ~a · ~b, which in particularly gives
gij = g ~ei ,~ej = ~ei · ~ej , (B.6)
where {~ei } is the basis dual to {j }.
Since the D × D-matrix with elements gij is regular, it is invertible. Let gij denote the elements
e
of its inverse matrix: gij g jk = δik , g ij gjk = δki . The D2 scalars g ij define a 20 -tensor g ij ~ei ⊗~ej , the
inverse metric tensor , denoted as g−1 .
Using results on symmetric matrices, the square matrix with elements gij is diagonalizable—i.e.
one can find an appropriate basis {~ei } such that g ~ei ,~ej = 0 for i 6= j. Since g is nondegenerate,
the eigenvalues are non-zero: at the cost
of multiplying the basis vectors {~ei } by a numerical factor,
one may demand that every g ~ei ,~ei be either +1 or −1, which yields the canonical form
gij = diag(−1, . . . , −1, 1, . . . , 1) (B.7)
for the matrix representation of the components of the metric tensor.
In that specific basis, the component g ij of g−1 coincides with gij , yet this does not hold in an
arbitrary basis.
Role of g in tensor algebra
::::::::::::::::::::::::::::
In agreement with the remark at the end of § B.1.3 a, for any given vector ~c = ci~ei the object
g(~c, ) maps vectors into the base field K, i.e. it is a one-form c = cj j , such that
e e
cj = c(~ej ) = g(~c,~ej ) = g(ci~ei ,~ej ) = ci gij . (B.8a)
e
That is, a metric tensor g provides a mapping from vectors onto one-forms. Reciprocally its inverse
metric tensor g−1 maps one-forms onto tensors, leading to the relation
ci = g ij cj . (B.8b)
Generalizing, a metric tensor and its inverse thus allow one “to lower
or to raise indices”, which
m∓1
are operations mapping a tensor of type m
n on a tensor of type n±1 , respectively.
Remarks:
∗ Lowering resp. raising an index actually amounts to an outer product with g resp. g−1 followed
by the contraction of two indices. For instance
outer product contraction
~c = ci~ei 7−→ ~c ⊗ g = ci gjk ~ei ⊗ j ⊗ k 7−→ c = ci gik k = ck k
e e e e e
where the first and second arguments of ~c ⊗ g have been contracted.
∗ Generalizing the “dot product” notation for the inner product defined by the metric tensor, the
contraction is often also denoted with a dot product. For example, for a 2-form T and a vector ~c
T · ~c = T ij i ⊗ j · ck~ek = T ij cj i ,
e e e
where we implicitly used Eq. (B.2). Note that for the dot-notation to be unambiguous, it is better
if T is symmetric, so that which of its indices is being contracted plays no role.
Similarly, if T denotes a dyadic tensor and T0 a 2-form
T · T0 = T ij ~ei ⊗ ~ej · T0kl k ⊗ l = T ij T0jl ~ei ⊗ l ,
e e e
which is different from T0 · T if the tensors are not symmetric. The reader may even find in the
literature the notation
T : T0 ≡ T ij T0ji ,
involving two successive contractions.
164 Tensors on a vector space
Remark: Λ is not a tensor, for the two indices of its elements refer to two different bases—which
is emphasized by the use of one primed and one unprimed index—while both components of a
1 -tensor are with respect to the “same” basis.(63)
1
0
Let Λk i denote the elements of the inverse matrix Λ−1 , that is
0 0 0
Λk i Λi j 0 = δjk0 and Λi k0 Λk j = δji .
0
One then easily checks that the numbers Λk i govern the change of basis from B ∗ to B 0∗ , namely
0 0
j = Λj i i . (B.10)
e e
Accordingly, each “vector” component transforms with Λ−1 :
0 0 0 0 0 0
cj = Λj i ci , T j1 ...jm = Λj1i1 · · · Λjmim T i1 ...im . (B.11)
One can thus obtain the coordinates of an arbitrary tensor in any basis by knowing just the
transformation of basis vectors and one-forms.
(63)
Or rather, with respect to a basis and its dual.
(64)
The reader should be aware that some of the mathematical terms used—as translated from the Russian—are
non-standard, e.g. (linear, bilinear) “functional” for form or “conjugate” (space, basis) for dual.
A PPENDIX C
Tensor calculus
Continuum mechanics, and in particular fluid dynamics, is a theory of (classical) fields. The latter
may be scalars, vectors or more generally tensors—mainly of degree at most 2—, whose dynamical
behavior is governed by partial differential equations, which obviously involve various derivatives of
tensorial quantities.
When describing vector or tensor fields by their respective components on appropriate (local)
bases, the basis vectors or tensors may actually vary from point to point. Accordingly, care must
be taken when differentiating with respect to the space coordinates: instead of the usual partial
derivatives, the quantities that behave in the expected manner are rather covariant derivatives
(Sec. C.1), which are the main topic of this Appendix.
To provide the reader with some elementary background on the proper mathematical framework
to discuss vector and tensor fields and their differentiation, some basic ideas of differential geometry
are gathered in Sec. C.2,
(a) In a neighborhood of every point P ∈ M, one can find a system of local coordinates {xi (P )}.
(b) It is possible to define functions on M with sufficient smoothness properties, as e.g. differen-
tiable functions.
(c) At each point P ∈ M, one can attach vectors—and more generally tensors. Let {~ei (P )}
denote a basis of the vectors at P .
From the physicist’s point of view, the above requirements mean that we want to be able to define
scalar, vector or tensor fields at each point [property (c)], that depend smoothly on the position
[property (b)], where the latter can be labeled by local coordinates [property (a)]. Mathematically,
it will be seen in Sec. C.2 that the proper framework is to look at a differentiable manifold and its
tangent bundle.
166 Tensor calculus
Before we go any further, let us emphasize that the results we state hereafter are independent
of the dimension n of the vectors, from 1 to which the indices i, j, k, l. . . run. In addition, we use
Einstein’s summation convention throughout.
Assuming the above requirements are fulfilled, which we now do without further comment, we
in addition assume that the local basis {~ei (P )} at every point is that which is “naturally induced”
by the coordinates {xi (P )},(65) and that for every possible i the mapping P 7→ ~ei (P ) defines a
continuous, and even differentiable vector field on M.(66) The derivative of ~ei at P with respect to
any of the (local) coordinate direction xk is then itself a vector “at P ”, which may thus be expanded
on the basis {~el (P )}: denoting by Γlik (P ) its coordinates
∂~ei (P )
= Γlik (P )~el (P ). (C.1)
∂xk
The numbers Γlik , which are also alternatively denoted as i lk , are called Christoffel symbols (of
dci (P ) ∂ci (P )
= + Γilk (P )cl (P ). (C.3b)
dxk ∂xk
Remark: The covariant derivative dci /dxk is often denoted by ci;k , with a semicolon in front of the
index (or indices) related to the direction(s) along which one differentiates. In contrast, the partial
derivative ∂ci /∂xk is then written as ci,k , with a comma. That is, Eq. (C.3b) is recast as
ci;k (P ) = ci,k (P ) + Γilk (P )cl (P ). (C.3c)
The proof of Eqs. (C.3) is rather straightforward. Differentiating relation (C.2) with the product
rule first gives
∂~c(P ) ∂ci (P ) i ∂~ei (P ) ∂ci (P )
= ~
e i (P ) + c = ~ei (P ) + ci (P ) Γlik (P )~el (P )
∂xk ∂xk ∂xk ∂xk
where we have used the derivative (C.1). In the rightmost term, the dummy indices i and l may
be relabeled as l and i, respectively, yielding ci Γlik ~el = cl Γilk ~ei , i.e.
∂~c(P ) ∂ci (P ) l i dci (P )
= ~
e i (P ) + c (P ) Γlk (P )~
e i (P ) = ~ei (P ).
∂xk ∂xk dxk
One can show that the covariant derivatives dci (P )/dxk are the components of a 11 -tensor field,
the (1-form-)gradient of the vector field ~c, which may be denoted by ∇~c. On the other hand, neither
the partial derivative on the right hand side of Eq. (C.3b) nor the Christoffel
e symbols are tensors.
(65)
This requirement will be made more precise in Sec. C.2.
(66)
This implicitly relies on the fact that the vectors attached to every point P ∈ M all have the same dimension.
C.1 Covariant differentiation of tensor fields 167
The Christoffel symbols can be expressed in terms of the (local) metric tensor g(P ), whose
components are in agreement with relation (B.6) given by(67)
To recover the usual inner product on R2 , the metric tensor g(P ) should have components
Remarks:
∗ The metric tensor in polar coordinates (C.7a) has signature (0, 2)—i.e. 0 negative and 2 positive
eigenvalues—, just like it has in Cartesian coordinates: the signature of the metric (tensor) is
independent of the choice of coordinates if it defines the same inner product.
∗ It is also interesting to note that the Christoffel symbols for polar coordinates (C.8) are not all
zero, while this is the case for the Christoffel symbols in Cartesian coordinates. This shows that
the Christoffel symbols are not the components of a tensor—a tensor which is identically zero in a
basis remains zero in any basis.
Consider now a constant vector field ~c(P ) = ~c(r, θ) = ~ex . Obviously, it is unchanged when going
from any point (r, θ) to any neighboring point, i.e. a meaningful derivative along either the r or θ
direction should identically vanish.
Let us write
sin θ
~c(r, θ) = ~ex = cos θ~er (r, θ) − ~eθ (r, θ) = cr (r, θ)~er (r, θ) + cθ (r, θ)~eθ (r, θ).
r
The partial derivatives ∂cr /∂xθ , ∂cθ /∂xr and ∂cθ /∂xθ are clearly non-vanishing. On the other
hand, all covariant derivatives are identically zero: omitting the variables, one finds
dcr ∂cr dcθ ∂cθ sin θ 1 (− sin θ)
r
= r
= 0, r
= r
+ Γθθr cθ = 2 + = 0,
dx ∂x dx ∂x r r r
i.e. d~c/dxr = ~0, and
dcr ∂cr (− sin θ) dcθ ∂cθ cos θ 1
θ
= θ
+ Γrθθ cθ = − sin θ − r = 0, θ
= + Γθrθ cr = − + cos θ = 0,
dx ∂x r dx ∂xθ r r
θ ~
i.e. d~c/dx = 0. Thus the covariant derivatives give the expected result, while the partial derivatives
with respect to the coordinates do not.
df (P ) ∂f (P )
k
= . (C.9)
dx ∂xk
C.1.3 b One-forms
::::::::::::::::::
one-form field h(P ) = hj (P ) j (P )
e e
dhj (P ) ∂hj (P )
= − Γljk (P )hl (P ). (C.10)
dxk ∂xk
C.1.3 c ::::::::::::::::::::::::::
:::::::
Tensors of arbitrary type
m
n -tensor field T (P )
Tij11···i
dT m
...jn (P ) Tij11···i
∂T m
...jn (P ) i ···i l
2 ···im
= + Γikl1 (P )T
Tli im
Tj11 ...jm−1
j1 ...jn (P ) + · · · + Γkl (P )T (P )
dxk ∂xk n (C.11)
Tilj12···i
− Γlj1 k (P )T m
...jn (P ) − ··· − Tij11···i
Γljn k (P )T m
...jn−1 l (P ).
relating the first partial derivatives of its real and imaginary parts are satisfied.
Equivalently, the relations (D.3) can be recast as
df
=0 (D.4)
dz̄
where z̄ = x − iy.
Corollary: A function f (z = x+iy) is holomorphic on a domain if and only if its real and imaginary
parts are conjugate harmonic functions, i.e. they obey the Cauchy–Riemann equations (D.3) and
the Laplace equations
4P (x, y) = 0 , 4Q(x, y) = 0 (D.5)
on the domain.
D.1.2 b Integration of holomorphic functions
:::::::::::::::::::::::::::::::::::::::::::::
Z Z b
f γ(t) γ 0 (t) dt
f (z) dz = (D.6)
C a
D.2 Multivalued functions 171
I
f (z) dz = 0. (D.7)
C
I
1 f (z)
f (z0 ) = dz. (D.8)
2πi C z − z0
∞
X f (n)(z0 )
f (z) = (z − z0 )n (D.9)
n!
n=0
I
n! f (z)
f (n)(z0 ) = dz, (D.10)
2πi C (z − z0 )n+1
which generalizes the Cauchy integral formula (D.8) to the successive derivatives of f .
zeroes
D.3.2 a Definitions
::::::::::::::::::
isolated singularity
removable singularity
pole of order m
essential singularity
D.3.2 b Laurent series
::::::::::::::::::::::
∞
X
f (z) = an (z − z0 )n (D.11)
n=−∞
with
I
1 f (z)
an = n+1
dz (D.12)
2πi C (z − z0 )
where C denotes
a−1 residue
φ(z) complex potential on z-plane. Then Φ(Z) ≡ φ(F (Z)) potential on Z-plane, with velocity
w(F (Z))F 0 (Z)
[1] T. E. Faber, Fluid dynamics for physicists (University Press, Cambridge, 1995).
[2] E. Guyon, J.-P. Hulin, L. Petit, C. D. Mitescu, Physical hydrodynamics, 2nd ed. (University
Press, Oxford, 2015).
[3] L. Landau, E. Lifshitz, Course of theoretical physics. Vol. VI : Fluid mechanics, 2nd ed. (Perg-
amon, Oxford, 1987).
[4] L. Landau, E. Lifschitz, Lehrbuch der theoretischen Physik. Band VI : Hydrodynamik , 5. ed.
(Harri Deutsch, Frankfurt am Main, 1991).
[5] A. Sommerfeld, Lectures on Theoretical Physics. Vol. II : Mechanics of deformable bodies (As-
sociated Press, New York, NY, 1950).
[6] A. Sommerfeld, Vorlesungen über theoretische Physik. Band II : Mechanik der deformierbaren
Medien, 6. ed. (Harri Deutsch, Frankfurt am Main, 1992).
[7] M. Reiner, The Deborah Number , Phys. Today 17(1) (1964) 62.
[10] L. Sedov, A course in continuum mechanics. Vol. I : Basic equations and analytical techniques
(Wolters–Noordhoff, Groningen, 1971).
[11] R. E. Graves, B. M. Argrow, Bulk viscosity: Past to present, J. Thermophys. Heat Tr. 13
(1999) 337–342.
[12] R. K. Agarwal, K.-Y. Yun, R. Balakrishnan, Beyond Navier–Stokes: Burnett equations for
flows in the continuum–transition regime, Phys. Fluids 13 (2001) 3061–3085.
[13] T. Fließbach, Lehrbuch zur theoretischen Physik I. Mechanik , 4. ed. (Spektrum Akademischer
Verlag, Heidelberg & Berlin, 2003).
[15] V. I. Arnold, Lectures on Partial Differential Equations (Springer, Berlin, Heidelberg, New
York, 2004).
[16] G. F. Carrier, C. E. Pearson, Partial differential equations: Theory and techniques, 2nd ed.
(Academic Press, New York, 1988).
[39] A. Libchaber, C. Laroche, S. Fauve, Period doubling cascade in mercury, a quantitative mea-
surement, J. Phys. (Paris) Lett. 43 (1982) L-211–L-216.
[40] S. Chandrasekhar, Hydrodynamic and hydromagnetic stability (University Press, Oxford,
1961).
[41] J. Wesfreid, Y. Pomeau, M. Dubois, C. Normand, P. Bergé, Critical effects in Rayleigh–Bénard
convection, J. Phys. (Paris) 39 (1978) 725–731.
[42] M. G. Velarde, C. Normand, Convection, Sci. Am. 243 No.1 (1980) 93–108.
[43] C. Normand, Y. Pomeau, M. G. Velarde, Convective instability: A physicist’s approach, Rev.
Mod. Phys. 49 (1977) 581–624.
[44] C. Eckart, The thermodynamics of irreversible processes. 3. Relativistic theory of the simple
fluid , Phys. Rev. 58 (1940) 919–924.
[45] D. H. Rischke, Fluid dynamics for relativistic nuclear collisions, Lect. Notes Phys. 516 (1999)
21–70 [arXiv:nucl-th/9809044].
[46] W. A. Hiscock, L. Lindblom, Generic instabilities in first-order dissipative relativistic fluid
theories, Phys. Rev. D 31 (1985) 725–733.
[47] I. Müller, Zum Paradoxon der Wärmeleitungstheorie, Z. Phys. 198 (1967) 329–344.
[48] W. Israel, Nonstationary irreversible thermodynamics: A causal relativistic theory, Ann. Phys.
(NY) 100 (1976) 310–331.
[49] W. Israel, J. M. Stewart, Transient relativistic thermodynamics and kinetic theory, Ann. Phys.
(NY) 118 (1979) 341–372.
[50] N. Andersson, G. L. Comer, Relativistic fluid dynamics: Physics for many different scales,
Living Rev. Rel. 10 (2005) 1–87 [arXiv:gr-qc/0605010].
[51] P. Romatschke, New developments in relativistic viscous hydrodynamics, Int. J. Mod. Phys. E
19 (2010) 1–53 [arXiv:0902.3663 [hep-ph]].
[52] S. Weinberg, Gravitation and Cosmology (John Wiley & Sons, New York, 1972).
[53] L. Landau, On the multiparticle production in high-energy collisions, Izv. Akad. Nauk Ser. Fiz.
17 (1953) 51–64 [English translation in Collected papers of L.D.Landau (Gordon and Breach,
New York, NY, 1965) pp. 569–585].
[54] S. Belenkij, L. D. Landau, Hydrodynamic theory of multiple production of particles, Nuovo
Cim. Suppl. 3 S1 (1956) 15–31.
[55] J. D. Bjorken, Highly relativistic nucleus-nucleus collisions: The central rapidity region, Phys.
Rev. D 27 (1983) 140–151.
[56] M. Nakahara, Geometry, Topology, and Physics, 2nd ed. (Institute of Physics, Bristol, 2003).
[57] M. Postnikov, Lectures in Geometry. Semester II : Linear Algebra and Differential Geometry
(Mir Publishers, Moscow, 1982).
[58] H. Cartan, Théorie élémentaire des fonctions analytiques d’une ou plusieurs variables com-
plexes, 6th ed. (Herrmann, Paris, 1985).
[59] E. T. Whittaker, G. N. Watson, A course of modern analysis, 4th ed. (University Press,
Cambridge, 1927).
176 Bibliography