I The Continuum Navier-Stokes Equations

(A) Foundations of Macroscopic Hydrodynamics

See also Tennekes & Lumley, Section 2.2

Hamiltonian equations of a simple molecular fluid:

N
X p2n 1 X
H = + (rnm ), rnm = rn rm
2m 2
n=1 n6=m
@H pn
ṙn = = ,
@pn m
@H X
ṗn = = r (rnm ), r (rnm ) = Fnm .
@rn
m6=n

Conservation laws:

mass : M= mN
N
X
momentumn : P = pn
n=1
XN
p2n 1 X
energy : E = + (rnm )
2m 2
n=1 n6=m

Local conservation laws

mass density:

N
X
⇢ˆ(x, t) = m 3 (x rn (t))
n=1

mass current:

N
X
ĝ(x, t) = pn (t) 3 (x rn (t))
n=1

=) @t ⇢ˆ(x, t) + r · ĝ(x, t) = 0

1
 momentum density:

N
X
ĝ(x, t) = pn (t) 3 (x rn (t))
n=1

momentum current(stress tensor):

XN
1
T̂(x, t) = pn (t)pn (t) 3 (x rn (t))
m
n=1
Z 1
1 X
+ Fnm rnm (t) ds 3 (x rn (t) + srnm (t))
2 0
n6=m

=) @t ĝ(x, t) + r · T̂(x, t) = 0

energy density

0 1
N
X
@ 1 X
p2n
ê(x, t) = + (rnm )A 3
(x rn (t))
2m 2
n=1 m6=n

energy current

0 1
N
X
@ 1 X
p2n pn 3
ŝ(x, t) = + (rnm )A
(x rn (t))
2m 2 m
n=1 m6=n
Z 1
1 X 1
+ [(pn + pm ) · Fnm ]rnm ds 3 (x rn + srnm )
4 m 0
n6=m

=) @t ê(x, t) + r · ŝ(x, t) = 0

Hydrodynamic regime: A sufficiently short-range potential =) the typical distance mf p

travelled by a molecule between O(1) interactions is f inite. Here, mf p is the mean-free path.

Let
3 r
G` (r) = ` G( )
`

2
for G smooth, rapidly decreasing in |r|, positive, with unit integral. With

ˆ (x, t) = (ˆ
⇢ ⇢0 , ⇢ˆ1 , ⇢ˆ2 , ⇢ˆ3 , ⇢ˆ4 ) = (ˆ
⇢, ĝ1 , ĝ2 , ĝ3 , ê)

set
Z
ˆ )(x, t) =
⇢` (x, t) = (G` ? ⇢ ˆ (x + r, t)G` (r)
dr ⇢

For example,
N
X
⇢0` (x, t) = ⇢` (x, t) = mG` (x rn (t))
n=1
Thus, ⇢` (x, t) is the conserved density averaged over all the molecules in a region of radius ⇠ `

around the point x. These are called “coarse grained densities00 at the length-scale `.

Figure 1.

Consider situations for which there exist smooth macroscopic densitities ⇢(x, t), with

|⇢|
Lr = max mf p
|r⇢|
|⇢|
Tr = max ⌧mf p = mf p /vth
|⇢t |

with vth the typical molecular speed (a thermal velocity) such that

⇢` (x, t) ⇠
= ⇢(x, t)

3
 Gas Collision times [10 10 sec] Mean free Paths [10 8 m]

He 2.2 27.45

Ar 2.5 9.88

CO2 1.6 6.15

H2 1.0 17.44

N2 2.0 9.29

O2 2.2 9.93

Table 1: Ideal gases at temperature T = 120 C and pressure p = 1 atm.

for all mf p ⌧ ` ⌧ Lr . This is called a situation with separation of scales. The small parameter
mf p
" = Kn = Lr ⌧ 1 is called the Knudsen number. Since maximum macroscopic velocities

are of the order of the sound speed which is ⇠

= vth , ⌧mf p /Tr ⇠
= ".

Typically, on a timescale O(⌧mf p ), the particle distributions of positions and momentum in

the region of radius ` achieve a local equilibrium distribution

1
P (0) / exp{ (x, t)[H` v(x, t) · P` µ(x, t)N` ]}
Z

with N` , P` ,H` ( i.e. N` ⇠

= 43 ⇡`3 ⇢` , etc) the particle number, total momentum and energy of

the particles inside the region of radius ` at x, t.

1
(x, t) = is inverse temperature
kB T (x, t)
v(x, t) is macroscopic velocity

µ(x, t) is chemical potential

The potentials , µ, v uniquely determine the densities ⇢, g, e by equilibrium thermodynamics,

and vice versa. Note in particular that, integrating over positions, the particle momenta have

a very simple Maxwellian (i.i.d. Gaussian) distribution

!
X p2
(0) n
P ({pn |for n s.t. xn (t) 2 B` (x)}) / exp (x, t) .
n
2m

4
 Space-ergodicity of the local-equilibrium statistics ties together the spatial coarse-graining

and the ensemble average over the equilibrium distribution P (0) :

(0)
f¯` (x, t) ! f (0) (x, t) = hf i (x,t),µ(x,t),v(x,t) , ` ⇠, (A)

where ⇠ is a correlation length of P (0) and f! (x, t) is any “local density”, i.e. a function of x,

t and phase point

! = (rn , pn ), n = 1, . . . , N

such that f! (x, t) depends only upon the particles with location nearby x at the t. For ex-

ample, a strictly local density would satisfy f! (x, t) = f!B(x,R) (x, t) where !B(x,R) is the set

{(rn , pn ), rn 2 B(x, R) : n = 1, . . . , N }, so that R is the range of f . We also require that

f a! (x + a, t) = f! (x, t)

where a! = {(rn + a, pn ), n = 1, . . . , N }. Then (A) holds almost surely for every ! with

respect to P (0) . See:

H. Spohn, Large-Scale Dynamics of Interacting Particles (Springer, 1991)

A. Martin-Löf, Statistical Mechanics and the Foundations of Thermodynamics (Springer,

1979).

The mean currents (fluxes) J` (x, t) ⇠

= j(0) (x, t) which satisfy

@t ⇢a (x, t) + r · j(0)
a (x, t) = 0, a = 0, 1, 2, 3, 4.

are also given by equilibrium thermodynamics

(0)
g(0) (x, t) = j0 (x, t) = ⇢(x, t)v(x, t))
(0)
êi · T(0) (x, t) = ji (x, t) = ⇢(x, t)vi (x, t)v(x, t) + P (x, t)êi
(0) 1
s(0) (x, t) = j4 (x, t) = (u(x, t) + P (x, t) + ⇢(x, t)v 2 (x, t))v(x, t)
2
= (e(x, t) + P (x, t))v(x, t)

5
where P = P (⇢, u) is the thermodynamic pressure. Note that the total energy density is the

sum of u(x, t), the internal energy per volume, and the kinetic energy density:

1
e(x, t) = u(x, t) + ⇢(x, t)v 2 (x, t).
2

The equations

@t ⇢ + r · (⇢v) = 0,

@t (⇢v) + r · (⇢vv + P I) = 0,

@t e + r · [(e + P )v] = 0.

are the compressible Euler equations. There are closed equations for the densities ⇢, g, e.

Closure has been achieved because of the local equilibrium distribution of the molecules in the

small regions of radius `.

Navier-Stokes equation of a simple fluid

There is, however, an O(") correction to local equilibruim

J` (x, t) ⇠
= j(0) (x, t) + j(1) (x, t), j(1) (x, t) ⇠ O(")

where?

g(1) ⌘ 0 (by momentum conservation!)


2
T (1)
= ⌘ rv + (rv)> (r · v)I ⇣(r · v)I
3
s(1) = T(1) · v rT.

Here heat current q (1) = rT. The following transport coefficients appear

⌘ = shear viscosity;

⇣ = bulk viscosity;

 = heat conductivity or thermal conductivity.

with ⌘, ⇣,  0.

6
 ? Note: In principle, there is another rotational viscosity ⌘rot and an additional stress term
(1) 1
Trot = ⌘rot [rv (rv)> ⌦] where ⌦ij = 2 ✏ijk !k and ! is the local angular velocity of

the fluid. However, this term can generally be neglected!

Kinetic Theory derivation

For a linear shear profile u(y), consider transport of x momentum along the direction y

Figure 2.

(1)
Txy is the net flux of x momentum in the y direction

(1)
Txy = ⇢u(y mf p )vth ⇢u(y + mf p )vth

⇠ @u
= 2⇢ mf p vth (y)
@y
@u
= ⌘
@y

where ⌘ ⇠
=⇢ mf p vth .

Note that compared with T(0) = ⇢vv + P I

u
|T(1) | ⇠ ⇢ mf p vth L mf p vth
= = = O(")
|T(0) | ⇢u2 L u

The final equations

7
 @t ⇢ + r · (⇢v) = 0,
⇣ ⌘
@t (⇢v) + r · ⇢vv + P I + T(1) = 0,

1
@t e + r · (u + P + ⇢v 2 )v + T(1) · v + q(1) = 0
2

with

(1) 2
T = ⌘ rv + (rv)T (r · v)I ⇣(r · v)I
3
q(1) = rT

are the compressible Navier-Stokes equations, withh ⌘(⇢, T ), ⇣(⇢, T ), (⇢, T ) functions of mass

density ⇢ and temperature T (given by so-called Green-Kubo formulas).

Remarks:

(i) Like the Euler equations, the Navier-Stokes equations are closed hydrodynamic equations

in terms of (⇢, g, e) [or (⇢, v, e)]. Closure is achieved because not only is the statistics of the

molecules given by local equilibrium to O("0 ), but also the O("1 ) corrections are determined
mf p
by the gradients of (⇢, v, e). To get closure, separation of scales and small " = Lr is crucial.

(ii) There are also small random terms to O("1 ) which we have neglected

2
T (1)
= ⌘ rv + (rv)> (r · v)I ⇣(r · v)I + T̃(1)
3
q(1) = rT + q̃(1)

where the stochastic fluxes T̃(1) , q̃(1) are called molecular noise. They represent the chaotic

e↵ects of the unknown molecular positions and momentum. The covariances of the random

terms are given by fluctuation-dissipation relations in terms of transport coefficients ⌘, ⇣, :

(1)
p p (1)
p
0
T̃ij = 2kB ⌘T ⌘ij + 2kB ⇣T ⌘ij , q̃i = 2kB T 2 ⌘i

0 , ⌘ are space-time Gaussian white-noises with zero mean and covariances

where ⌘ij , ⌘ij i
✓ ◆
0 0 1
h⌘ij (x, t)⌘kl (x , t )i = ik jl + il jk ij kl
3
(x x0 ) (t t0 )
3

8
 0 0
h⌘ij (x, t)⌘kl (x0 , t0 )i = ij kl
3
(x x0 ) (t t0 ), h⌘i (x, t)⌘j (x0 , t0 )i = ij
3
(x x0 ) (t t0 ).

Thus, the statistics of these terms are also given, once ⇢, v, e are known. Such equations of

fluctuating hydrodynamics were first proposed by

L. D. Landau and E. M. Lifshitz, Fluid Mechanics, Course of Theoretical Physics,

Vol. 6 (Pergamon Press, 1959)

A fundamental reference is

V. G. Morozov, “On the Langevin formalism for nonlinear and nonequilibrium hy-

drodynamic fluctuations,” Physica A 126 443-460 (1984)

which first derived the correct fluctuation equations for a compressible Navier-Stokes fluid. It

is important to emphasize that, despite appearances, these are not stochastic PDE’s! Instead,

all fields and delta functions are cut o↵ at some high wavenumber ⇤, where Lr1 ⌧ ⇤ ⌧ 1
mf p .

Even somewhat earlier an essentially equivalent formulation by an Onsager-Machlup “e↵ec-

tive action” had been proposed by Robert Graham, see:

R. Graham, “Onset of cooperative behavior in nonequilibrium steady states”, in

Order and Fluctuations in Equilibrium and Nonequilibrium Statistical Mechanics,

G. Nicolis, G. Dewel, and J. W. Turner, eds. (Wiley, New York, 1981).

G. L. Eyink, “Dissipation and large thermodynamic fluctuations,” J. Stat. Phys.

61 533-572 (1990)

We shall see that thermal fluctuations are non-negligible at very small scales in turbulent flows!

In fact, this was first noted in pioneering papers of Robert Betchov:

R. Betchov, “On the fine structure of turbulent flows,” Journal of Fluid Mechanics

3, 205-216 (1957)

R. Betchov, “Thermal Agitation and Turbulence,” in: Rarefied Gas Dynamics, Pro-

ceedings of the 2nd International Symposium, ed. L. Talbot. (Academic Press, New

York, 1961), pp.307-321

9
We shall discuss this subject later after developing a deeper understanding of turbulence.

Energy balance

From

@t (⇢v) + r · [⇢vv + P I + T(1) ] = 0

one easily derives


1 2 1
@t ( ⇢v ) + r · ( ⇢v 2 + P )v + T(1) · v = Q + P (r · v)
2 2

with Q = T(1) : rv = 2⌘|S|2 + ⇣(r · v)2 0 where S = r̊vs = 12 [rv + (rv)> ] 1

3 (r · v)I.

Thus, kinetic energy is not conserved!

However, total energy is conserved


1
@t e + r · (u + P + ⇢v 2 )v + T(1) · v + q(1) = 0
2

1 2
=) internal energy u = e 2 ⇢v satisfies

@t u + r · (uv + q(1) ) = T(1) : rv P (r · v)

Hence, the viscous energy dissipation appears as a sink for kinetic energy and as a source for

internal energy. As the fluid slows down, it also heats up!

Another very important property of the Navier-Stokes system has to do with thermodynamic

entropy s(u, ⇢) which satisfies the equation

h i Q |rT |2
@t s + r sv + q(1) /T = + 0.
T T2

This is a local form of the second law of thermodynamics.

Incompressible Limit

Another important simplification appears if the typical fluid velocities U are much less than

the sound speed c ⇠

= vth . This corresponds to a limit of small Mach number

U
Ma ⌘ .
c

10
Sound waves are compressible waves with periodic variation of the mass density ⇢. When

c U , then di↵erence in density are adjusted very quickly compared with fluid motions. Thus,

⇢⇠
= constant

In that case

@t ⇢ + r · (⇢v) = 0 =) ⇢r · v = 0 =) r · v = 0

The viscous stress becomes

2
T(1) = ⌘[rv + (rv)> (r · v)I] ⇣(r · v)I
3
= 2⌘S

with S = 12 [rv + (rv)> ] the strain(rate) field since r · v = 0 and?

r · T(1) = ⌘[4v + r(r · v)] = ⌘4v

? Weak variations of ⌘ with temperature have been neglected here!

Thus, the momentum equation becomes

@t (⇢v) + r · [⇢vv + P I + T(1) ] = 0

=) ⇢@t v + ⇢[(v · r)v + (r · v)v] + rP ⌘4v = 0

=) @t v + (v · r)v = rp + ⌫4v, r · v = 0

P ⌘
the incompressible Navier-Stokes equation with p = ⇢ the kinetic pressure; ⌫ = ⇢ the kinetic

viscosity. Note that the (kinematic) pressure is determined from the Poisson equation

rp = r · [(v · r)v] = (@i vj )(@j vi )

using r · v = 0

The energy equation also satisfies

@t u + r · (uv + q(1) ) = T(1) : rv P (r · v)

=) (@t + v · r)u 4T = 2⌘|S|2

11
This equation for internal energy can be written as one for temperature, using the definition?

1 @u
cp = ( )p = specific heat at constant pressure (per unit mass)
⇢ @T

=) ⇢cp (@t + v · r)T = 4T + 2⌘|S|2

"
=) (@t + v · r)T = T 4T +
cp

with


T = = thermal di↵usivity
⇢cp
" = 2⌫|S|2 = energy dissipation per unit mass.

? In fact, one gets directly the stated result with instead the specific heat at constant volume

cv = ⇢1 ( @T
@u
)⇢ , but cv = cp for an incompressible fluid. The result as stated is more generally

valid for a so-called Boussinesq fluid, taking into account some compressibility e↵ects.

NOTES & REFERENCES

Nonequilibrium Thermodynamics

deGroot, S. R. & Mazur P., Nonequilibrium Thermodynamics, Dover Publishing, New

York, 1984.

Nonequilibrium Statistical Mechanics

Zubarev, D. N., Nonequilibrium Statistical Thermodynamics, Plenum, New York, 1974.

McLennan, J. A., Nonequilibrium Statistical Mechanics, Prentice Hall, Englewood Cli↵s,

New York, 1989.

Dorfman, J. R. , van Beijeren, H. & Kirkpatrick, T. R., Contemporary Kinetic Theory of

Matter, Cambridge University Press, 2021.

12
In particular, for the microscopic derivation of the Landau-Lifschitz equations, see:

Zubarev, D. N. and Morozov, V. G., “Statistical mechanics of nonlinear hydrodynamic

fluctuations,” Physica A 120 411-467 (1983)

Español, P. “Stochastic di↵erential equations for non-linear hydrodynamics,” Physica A

248 77-96 (1998)

Derivation of Incompressible Navier-Stokes

Klainerman S. & Majda, A.J.. Singular limits of quasilinear hyberbolic systems with large

parameters and incompressible fluids, Comm. Pure. App. Math, 34, 481-524, 1981

gives a rigorous derivation of incompressible Navier-Stokes from compressible, in the limit of

small Mach number. See also:

Majda, A. J. Compressible Fluid Flow & Systems of Conservation Laws in Several Space

Variables, Springer, New York, 1984

B. J. Bayly, C. D. Levermore, and T. Passot, Density variations in weakly compressible

flows, Phys. Fluids A 4 945–954, 1992

The following paper

Quastel, J. & Yan, H.T., Lattice gases, large deviations and the incompressible Navier-

Stokes equations, Ann. Math., 148, 51-108, 1998.

gives a rigorous derivation of the incompressible Navier-Stokes equation from a microscropic

particle model ( a lattice gas) in the limit of small Knudsen and Mach numbers, for any fixed

but arbitrarily large Reynolds number. An important point of their derivation is that if applies

even if the Navier-Stokes equation develops a singularity! In that case, a distributional solution

in the sense of Leray describes the fluid motion. Furthermore, the authors derive not only the

deterministic fluid equation with probability approaching one (law of large numbers) but they

13
also rigorously derive the nonlinear Onsager-Machlup action for the small thermal fluctuations

(large deviations theorem). We shall argue later, however, that the “scaling limit” employed in

this work is physically unrealistic and never achieved for molecular fluids in Nature.

A careful derivation of the low Mach-number fluctuating hydrodynamic equations for an

incompressible fluid, starting from the compressible Landau-Lifschitz equations, is given by

A. Donev et al., “Low Mach number fluctuating hydrodynamics of di↵usively mixing

fluids,” Commun. in Appl. Math. Comp. Sci. 9 47-105 (2014).

Finite Mach number E↵ects

We focus mainly on low Mach number flows ( a wide class). One important finite-Mach e↵ect

is sound generation by turbulence. See:

Lighthill, M. J., On sound generated aerodynamically: I. General Theory, Proc. Roy.

Soc. land. A, 211, 564-578,1952.

Miksis, M. J. & Ting, L., On vortical flow and sound generations, SIAM J. App. Math.,

50, 521-536, 1990.

Boussinesq Equations

Normand, C., Pomeau, Y. & Velarde, M. G. Convective instability: a physicist’s approach.

Rev. Mod. Phys., 49, 581-627, 1977

14