Revised January 29, 2003

Copyright 2000 David A. Randall

Derivation of the Navier-Stokes Equation

David A. Randall

Department of Atmospheric Science Colorado State University, Fort Collins, Colorado 80523

25.1

Analysis of the relative motion near a point

Suppose that the velocity of the uid at position r 0 ( x, y, z ) and time t is V ( x, y, z, t ) , and that the simultaneous velocity at a neighboring position r 0 + r is V + dV . Then the velocity at the neighboring position r 0 + r relative to the velocity of the reference position r 0 is du dv = dw In vector form, this can be written as dV = D r , (25.2) u ----x v ---x w -----x u ----y v ---z w -----y u ----z v ---z w -----z x y . z

(25.1)

where D is called the deformation tensor because it describes how the uid element is being deformed by the non-uniform motion eld. Introducing Cartesian tensor notation, we can alternatively express (25.1) as v dv i = ------i x j , where i = 1, 2, 3 . x j (25.3)

We can decompose v i x j into two parts which are symmetric and antisymmetric in the sufces i and j . They are 1 v v - e ij = -- ------i + ------j , 2 x j x i (25.4)

Selected Nuggets

235

236

Derivation of the Navier-Stokes Equation

and 1 v v - x ij = -- ------i ------j , 2 x j x i respectively. The deformation D can now be expressed by D = e+X, where we dene two parts of D : e xx e xy e xz e e yx e yy e yz e zx e zy e zz 0 x xy x xz , X x yx 0 x yz x zx x zy 0 . (25.6) (25.5)

(25.7)

We show below that e is related to the divergence, and X is related to the vorticity. Correspondingly, dV can be divided into two parts, i.e. dV = dV
(s)

+ dV
s

(a)

where

d Vi

(s)

= e ij X j, d V i

(a)

= x ij X j .

(25.8)

Here the superscripts

and a denote symmetric and antisymmetric, respectively.

We refer to e as the rate of strain tensor (or strain tensor). The diagonal elements u v w of e , i.e. e xx = -----, e yy = ---- , e zz = ------ , are called the normal strains, and represent the x y z rate of volume expansion, as illustrated in Fig. 25.1 a. The sum of the normal strains is the divergence of V , i.e. e xx + e yy + e zz = e ii = V . (25.9)

The off-diagonal elements of e , namely e xy , e yz , and e zx , are called the shearing strains, and express the rate of shearing deformation, as illustrated in Fig. 25.1 b. They can be written as

Selected Notes of David A. Randall

25.1 Analysis of the relative motion near a point

237

v u v ----- + ---x y

Dx

du

du

Dy dv 0 Dx

u v ----- + ---y x
x

Figure 25.1: a) The normal strains, representing the rate of volume expansion. b) The shearing strains, which represent the rate of shearing deformation.

1 u v e xy = e yx = -- ----- + ---- , 2 y x 1 v w - e yz = e zy = -- ---- + ------ , 2 z y 1 w u e zx = e xz = -- ------ + ----- . 2 x z The elements of the anti-symmetric tensor, X , are (25.10)

Selected Notes of David A. Randall

238

Derivation of the Navier-Stokes Equation

1 w v 1 u w v u x zy = x yz = -- ------ ---- , x xz = x zx = -- ----- ------ , x yx = x xy = ---- ----- . (25.11) y z z x x y 2 2 Introducing the notation w x = 2x yz, w y = 2x zx, w z = 2x zx , we see that ( w x, w y, w z ) is the vorticity vector w = V. In tensor notation, (25.12) may be simply written v k w i = e ijk ------- , x j (25.13) (25.12)

where e ijk = 1 when i, j, k are in even permutation, and e ijk = 1 when i, j, k are in odd v 1 v 2 v u permutation. For example, w z = w 3 = e 321 ------- + e 312 ------- = ---- ----- , which gives x 2 x 1 x y twice the angular velocity about the z-axis, as illustrated in Fig. 25.2. Thus -du

Dy

v u ---- ----- = a + b x y a
Dx dv

Figure 25.2: The component of the vorticity in the z direction.

(a) du (a) dv (a) dw or

0 wz wy x = 1 -- w z 0 w x y , 2 wy wx 0 z

(25.14)

Selected Notes of David A. Randall

25.2 Invariants of the strain tensor

239

dV
25.2

(a)

1 = -- ( w r ) . 2

(25.15)

Invariants of the strain tensor

Consider the scalar product r e r = r (e r) = e xx x + e yy y + e zz z + 2 ( e xy xy + e yz yz + e zx zx ) = f ( x, y, z ) . (25.16)

2 2 2

By putting f ( x, y, z ) = constant , we obtain a surface of second order, the strain quadratic, also called the ellipsoid of strain. We refer the strain quadratic to its principal axes ( X 1, X 2, X 3 ) . Then (25.16) takes the form e 1 X 1 + e 2 X 2 + e 3 X 3 = F ( X 1, X 2, X 3 ) = constant .
2 2 2

(25.17)

Here e 1 , e 2 , and e 3 are the principal values of the strain tensor, and are called the principal normal strains. Using (25.16), we see that du 1 f - = -- ---- = e xx x + e xy y + e xz z 2 x (s) 1 f - dv = -- ---- = e yx x + e yy y + e yz z . 2 y (s) 1 f - dw = -- ---- = e zx x + e zy y + e zz z 2 z
(s)

(25.18)

In terms of the principal axes ( X 1, X 2, X 3 ) ,

(s) (s) (s) 1 F - dv 1 = -- -------- = e 1 X 1, dv 2 = e 2 X 2, dv 3 = e 3 X 3 . 2 X 1 (s) (s) (s) (s)

(25.19)

( dv 1 , dv 2 , dv 3 ) are the projections of dv upon the principal axes. Therefore, the symmetric part of dV is composed of three expansions (or contractions) in the mutually orthogonal directions of the principal axes of the strain quadratic.
Selected Notes of David A. Randall

240

Derivation of the Navier-Stokes Equation

We next consider the sum of the diagonal elements of e, i.e. e xx + e yy + e zz = e ii = V . (25.20)

This quantity is an invariant under the rotation of the axes of reference. Consider the transformation by which the surface f ( x, y, z ) = the constant of (25.16) is transformed into F ( X 1, X 2, X 3 ) = the constant of (25.17). We can write the transformation between the two Cartesian coordinates in the schematic, tabular form shown in Table 25.1 x X1 X2 X3
a1 a2 a3

y
b1 b2 b3

z
g1 g2 g3 bi gi

Table 25.1: The transformation between two Cartesian coordinate systems. Here ai, are the direction cosines between the axes.

, ,

Here ai, bi, gi, are the direction cosines between the axes. We nd that = 0 , i=1 i=1 a i a j + b i b j + g i g j = d ij

a i = 1,

ai bi

(25.21)

where d ij is Kroneckers delta ( d ij = 1 when i = j, d ij = 0 when i j ) . Substituting the expression X i = a i x + b i y + g i z where i = 1, 2, 3 into (25.17), we obtain = 0 . i=1 i=1 a i a j + b i b j + g i g j = d ij (25.22)

2 ai

= 1,

ai bi

(25.23)

This expression must be identical with (25.16), so that

Selected Notes of David A. Randall

25.3 Stresses

241

2 2 2 e xx = a 1 e 1 + a 2 e 2 + a 3 e 3, e xy = a 1 b 1 e 1 + a 2 b 2 e 2 + a 3 b 3 e 3 , 2 2 2 e yy = b 1 e 1 + b 2 e 2 + b 3 e 3, e yz = b 1 g 1 e 1 + b 2 g 2 e 2 + b 3 g 3 e 3 , . 2 2 2 e zz = g 1 e 1 + g 2 e 2 + g 3 e 3, e zy = g 1 a 1 e 1 + g 2 a 2 e 2 + g 3 a 3 e 3 ,

(25.24)

It follows that D = e xx + e yy + e zz =

i=1

( ai + bi + g i )ei
2 2 2

= e1 + e2 + e3

(25.25)

i.e., the sum e xx + e yy + e zz = V is independent of the choice of reference. Of course, this quantity is called the divergence, and we knew already that it is independent of the choice of reference. Another invariant of the strain tensor is e xx e yy + e yy e zz + e zz e xx e xy e yz e zx = constant . The proof is left as an exercise.
25.3 Stresses
2 2 2

(25.26)

In a moving viscous uid, forces act not only normal to a surface but also tangential to it. In Fig. 25.3, tx, ty and tz denote respectively the forces per unit area acting upon the surfaces normal to the positive x, y, and z directions, respectively. In general, each force has 3 components, so that we can express them as t x = t xx i + t yx j + t zx k , t y = t xy i + t yy j + t zy k , . t z = t xz i + t yz j + t zz k .

(25.27)

Symbolically we can write an element of the stresses as t ij ( i = 1, 2, 3 ;j = 1, 2, 3 ) . Here t ij is the i-component of the force per unit area exerted across a plane surface element normal to the j-direction. The stresses thus have nine elements, and can be represented by the matrix t xx t xy t xz T = t yx t yy t yz t zx t zy t zz
Selected Notes of David A. Randall

(25.28)

242

Derivation of the Navier-Stokes Equation

z t t z + ------z Dz z t x = t x t y t y + -------Dy y 0 t y = t y y

t x t x + -------Dx x

t z = t z

Figure 25.3: Sketch illustrating the normal and tangential forces acting on an element of fluid.

For an arbitrarily chosen plane surface, whose normal vector is n , the force acting per unit area is t n . See Fig. 25.4. The equilibrium condition is t y = t y t n Ds t x Ds x t y Ds y t z Ds z = 0 , (25.29)

where Ds is the area of the oblique surface, Ds x , Ds y , Ds z , are the areas of the faces lying in the planes x = 0, y = 0 and z = 0 . Denoting the direction cosines of n by a, Ds x Ds y Ds b and g, we have --------- = a , --------- = b , --------z = g . Therefore (25.29) can be written as Ds Ds Ds t n = at x + bt y + g t z , or (25.30)

Selected Notes of David A. Randall

25.3 Stresses

243

z tn C B tx n ( a, b, g ) y

ty 0 tz A x

Figure 25.4: Sketch illustrating the normal force acting on a plane surface.

t xn t yn t zn or

t xx t xy t xz = t yx t yy t yz t zx t zy t zz

a b , g

(25.31)

tn = T n .

(25.32)

Equation (25.32) shows that matrix T transforms a vector n into another vector t n , so that T is a tensor, called the stress tensor. See Fig. 25.5. An important property of T is its symmetry, i.e. t ij = t ji . This symmetry can be inferred from the condition of moment equilibrium. Consider for example a moment about the z-axis due to t yx and t xy : 1 1 1 ( t yx DyDz ) -- Dx ( t xy DxDz ) -- Dy = -- ( t yx t xy )DxDyDz . 2 2 2 (25.33)

Selected Notes of David A. Randall

244

Derivation of the Navier-Stokes Equation

t yx = t yx t xy 0

t xy = t xy Dz

t yx Dy x

Dx

Figure 25.5: Sketch illustrating the components of the stress tensor.

The shear stresses acting on the negative x- and y-surfaces contribute also, so that the total moment about the z-axis is ( t yx t xy )DxDyDz . Eulers equation (for a rotating rigid body) states that dW z I z --------- = ( t yx t xy )DxDyDz , dt
2 2

(25.34)

(25.35)

2 2 r ( Dx ) + ( Dy ) where I z = --------------------------------- = ----- DxDyDz [ ( Dx ) + ( Dy ) ] is the moment of inertia of 12 12 the uid element rDxDyDz , and W z is the angular velocity about the positive z-axis. Then (25.35) reduces to 2 2 dW z r ----- { ( Dx ) + ( Dy ) } --------- = t yx t xy . 12 dt

(25.36)

Selected Notes of David A. Randall

25.4 Invariants of the stress tensor

245

dW z In order that --------- remains nite, we need ( t yx t xy ) 0 when Dx, Dy 0 . It follows dt that t yx = t xy . Similar considerations establish that t xy = t yx, t yz = t zy, t zx = t xz . The three diagonal elements, elements are tangential stresses.
25.4 Invariants of the stress tensor
txx tyy tzz

(25.37)

, are normal stresses. The six off-diagonal

As in the case of the strain tensor, we have a scalar product r T r = t 1 X 1 + t 2 X 2 + t 3 X 3 = constant ,

2 2 2

(25.38)

where t1, t2, t3 are the principal stresses acting on the surface elements normal to the principal axes (X1, X2, X3) of the ellipsoid of stress. It can be shown that r T r is an invariant under the rotation of the axes of reference. The sum of the normal stresses t xx + t yy + t zz = t ii = t 1 + t 2 + t 3 = 3p , (25.39)

is also an invariant. For a moving uid, (25.39) is the denition of pressure p, which reduces to the static pressure when the uid is at rest.
25.5 The resultant forces due to the spatial variation of the stresses

is

The resultant force acting in the x-direction on a volume element of uid DxDyDz t + t xx Dx DyDz t DyDz + t + t xy Dy DxDz ----------------xx xx x xy y t xz t xx t xy t xz t xy DyDz + t xz + ---------Dz DxDy t xz DxDy = --------- + --------- + --------- DxDyDz . x z y z

(25.40)

Therefore the three components of the resultant force (vector) per unit volume are

Selected Notes of David A. Randall

246

Derivation of the Navier-Stokes Equation

t xx t xy t xz t xj X = --------- + --------- + --------- = -------- x y z x j t yx t yy t yz t yj Y = --------- + --------- + --------- = -------- . x y z x j t zx t zy t zz t zj Z = --------- + --------- + -------- = -------- - x y z x j Equation (25.41) can be compactly written as t x t y t DivT = ------- + ------- + ------z x y z t zj t xj t yj = i -------- + j -------- + k -------- . zj y j x j DivT is a tensor divergence and, therefore, is a vector.
25.6 Physical laws connecting the stress and strain tensors

(25.41)

(25.42)

The relation between the stress tensor and the strain tensor may be established by the following assumptions. i) t ij = pd ij + F ij , (25.43)

where F ij is the frictional stress tensor which depends on the space derivatives of the velocity, i.e. the strain tensor. ii) The relationship between F ij and eij is linear. iii) The relation does not depend on the frame of reference. We have already shown that, with reference to the respective principal axes, t1 0 0 T = 0 t2 0 0 0 t3 Because t1, t2,
t3

e1 0 0 , e = 0 e2 0 . 0 0 e3

(25.44)

are normal stresses and e1, e2, e3 are normal strains, it is natural to

Selected Notes of David A. Randall

25.6 Physical laws connecting the stress and strain tensors

247

assume that the principal axes of both tensors are the same. We therefore assume that t 1 = p + l V + 2me 1 t 2 = p + l V + 2me 2 , t 3 = p + l V + 2me 3 where p satises (25.39). Writing direction cosines between the principal axes (X1, X2, X3) and axes of a Cartesian coordinate system (x, y, z) as shown in Table 25.2, x X1 X2 X3
a1 a2 a3

(25.45)

y
b1 b2 b3

z
g1 g2 g3

Table 25.2: Direction cosines between principle axes.

we nd that t xx = a 1 t 1 + a 2 t 2 + a 3 t 3 t xy = a 1 b 1 t 1 + a 2 b 2 t 2 + a 3 b 3 t 3
2 2 2

(25.46)

etc. Substituting (25.45) into (25.46), and using (25.24), we obtain t = [ p + l ( V ) + 2me ]a 2 + [ p + l ( V ) + 2me ]a 2 xx 1 1 2 2 2 + [ p + l ( V ) + 2me 3 ]a 3 p + lD + 2me xx , t xy = [ p + l ( V ) + 2 m e 1 ]a 1 b 1 + [ p + l ( V ) + 2me 2 ]a 2 b 2 + [ p + l ( V ) + 2me 3 ]a 3 b 3 = 2me xy , etc. In general we have t ij = pd ij + l Vd ij + 2me ij . This expression gives (25.48)

(25.47)

Selected Notes of David A. Randall

248

Derivation of the Navier-Stokes Equation

t ii = 3p + l V + 2m V = 3p + ( 3l + 2m ) V . Because we have dened t ij = 3p , we conclude that 3l + 2m + 0 . Thus we have the nal expression 1 t ij = pd ij + 2m e ij -- Vd ij 3 = pd ij + F ij .
25.7 The Navier-Stokes equation

(25.49)

(25.50)

(25.51)

Substitution of (25.51) into (25.42) gives the resultant forces per unit volume, as expressed by the divergence of the stress tensor: DivT = p + DivF The equation of motion in the inertial frame (with subscript a omitted) is DV -------- = f ap + aDivF Dt or, in Cartesian tensor notation, v Dv i v 1 p f -------- = ------i + v j ------i = ------ a ------ + a ------ 2m e ij -- ( V )d ij 3 x j x i x j t Dt x i p 2 v k v v f = ------ a ------ + a ------ m ------i + ------j -- m ------- . x i x j x j x i 3 x k x i This is called the Navier-Stokes equation. to When the uid is incompressible ( V = 0 ) , and m is constant, (25.54) reduces (25.53) (25.52)

(25.54)

Selected Notes of David A. Randall

25.8 A proof that the dissipation is non-negative

249

v i Dv i p f -------- = ----- a ------ + am --------- , x i Dt x x 2

j

(25.55)

or, in vector form,

2 DV -------- = f ap + am V . Dt

(25.56)

Here we see the Laplacian of the vector V . 1 v v - It should be noted that e ij = -- ------i + ------j vanishes when V = W r (rigid 2 x j x i rotation), and that ( W r ) = 0 . This means that we do not have to worry about the distinction between V a = V + W r and V in the expression F = DivF in the equation of relative motion. The frictional force is the same in the rotating frame as in the inertial frame. This should be intuitively obvious.
25.8 A proof that the dissipation is non-negative

The dissipation rate is given by v ( F ) V = F ij ------i d x j u u u = F 11 ----- + F 12 ----- + F 13 ----z x y v v v + F 21 ---- + F 22 ---- + F 23 ---z x y (25.57)

w w w + F 31 ------ + F 32 ------ + F 33 ------ . z x y Since F ij = F ji , we have v v v d = F ij ------i = F ji ------i = F ij ------j , x j x j x i so we can write v v 2d = F ij ------i + ------j . x x j i But
Selected Notes of David A. Randall

(25.58)

(25.59)

250

Derivation of the Navier-Stokes Equation

1 v v 1 v - - F ij = 2m -- ------i + ------j -- ------i d ij 2 x j x i 3 x i 1 v v 1 v v 1 v - - - = 2m -- ------i + ------j ( 1 d ij ) + -- ------i + ------j -- ------i d ij , x x x x 3 x 2 i j i j i 2 or 1 v v 2 v - - F ij = 2m -- ------i + ------j ( 1 d ij ) + -- ------i d ij . 2 x j x i 3 x j Substituting (25.61) into (25.59), we nd that 1 v v 2 v - - d = m -- ------i + ------j ( 1 d ij ) + -- ------i d ij 2 x j x i 3 x j
2 4 = m 2e ij e ij ( 1 d ij ) + -- ( V ) d ij . 3

(25.60)

(25.61)

v- + v- ------i ------j x x
j i

(25.62)

This expression is obviously non-negative, since ( 1 d ij ) 0 and d ij 0 . For a nondivergent ow, the dissipation can be expanded, using Cartesian coordinates, as d = 2me ij e ij 1 u v 2 u w 2 = -- m ----- + ---- + ----- + ------ z x 2 y x v u 2 v w 2 + ---- + ----- + ---- + ------ x y z y w u 2 w v 2 + ------ + ----- + ------ + ---- x z z z u w 2 w v 2 u v 2 = m ----- + ---- + ----- + ------ + ------ + ---- . z x y z y x This shows that d is the sum of squares; again, it cannot be negative.

(25.63)

Selected Notes of David A. Randall