Fluid Dynamics: Theory and Computation

Dan S. Henningson Martin Berggren August 24, 2005

Contents
1 Derivation of the Navier-Stokes equations 1.1 Notation . . . . . . . . . . . . . . . . . . . . 1.2 Kinematics . . . . . . . . . . . . . . . . . . 1.3 Reynolds transport theorem . . . . . . . . . 1.4 Momentum equation . . . . . . . . . . . . . 1.5 Energy equation . . . . . . . . . . . . . . . 1.6 Navier-Stokes equations . . . . . . . . . . . 1.7 Incompressible Navier-Stokes equations . . 1.8 Role of the pressure in incompressible ow . 2 Flow physics 2.1 Exact solutions . . . . . . . . 2.2 Vorticity and streamfunction 2.3 Potential ow . . . . . . . . . 2.4 Boundary layers . . . . . . . 2.5 Turbulent ow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7 7 8 14 15 19 20 22 24 29 29 32 40 48 54 59 59 62 65 66 69 71 71 72 73 74 76 76 80 82 83 85 85 86 87 88 88

3 Finite volume methods for incompressible ow 3.1 Finite Volume method on arbitrary grids . . . . . 3.2 Finite-volume discretizations of 2D NS . . . . . . 3.3 Summary of the equations . . . . . . . . . . . . . 3.4 Time dependent ows . . . . . . . . . . . . . . . 3.5 General iteration methods for steady ows . . . .

4 Finite element methods for incompressible ow 4.1 FEM for an advectiondiusion problem . . . . . . . . . . 4.1.1 Finite element approximation . . . . . . . . . . . . 4.1.2 The algebraic problem. Assembly. . . . . . . . . . 4.1.3 An example . . . . . . . . . . . . . . . . . . . . . . 4.1.4 Matrix properties and solvability . . . . . . . . . . 4.1.5 Stability and accuracy . . . . . . . . . . . . . . . . 4.1.6 Alternative Elements, 3D . . . . . . . . . . . . . . 4.2 FEM for NavierStokes . . . . . . . . . . . . . . . . . . . 4.2.1 A variational form of the NavierStokes equations 4.2.2 Finite-element approximations . . . . . . . . . . . 4.2.3 The algebraic problem in 2D . . . . . . . . . . . . 4.2.4 Stability . . . . . . . . . . . . . . . . . . . . . . . . 4.2.5 The LBB condition . . . . . . . . . . . . . . . . . . 4.2.6 Mass conservation . . . . . . . . . . . . . . . . . . 4.2.7 Choice of nite elements. Accuracy . . . . . . . . . A Background material A.1 Iterative solutions to linear systems . A.2 Cartesian tensor notation . . . . . . A.2.1 Orthogonal transformation . A.2.2 Cartesian Tensors . . . . . . A.2.3 Permutation tensor . . . . . . A.2.4 Inner products, crossproducts A.2.5 Second rank tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . and determinants . . . . . . . . . . . . 3 . . . . . . .

95 . 95 . 98 . 98 . 99 . 100 . 100 . 101

A.2.6 Tensor elds . . . . . . . . . . . . A.2.7 Gauss & Stokes integral theorems A.2.8 Archimedes principle . . . . . . . . A.3 Curvilinear coordinates . . . . . . . . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . .

. . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . .

102 103 104 106 109 109 113 117 120 124 129 132 136 140 146 148 152 155 157 173

B Recitations 5C1214 B.1 Tensors and invariants . . . . . . . . . . . . . . . . . . . . . B.2 Euler and Lagrange coordinates . . . . . . . . . . . . . . . . B.3 Reynolds transport theorem and stress tensor . . . . . . . . B.4 Rankine vortex and dimensionless form . . . . . . . . . . . B.5 Exact solutions to Navier-Stokes . . . . . . . . . . . . . . . B.6 More exact solutions to Navier-Stokes . . . . . . . . . . . . B.7 Axisymmetric ow and irrotational vortices . . . . . . . . . B.8 Vorticity equation, Bernoulli equation and streamfunction . B.9 Flow around a submarine and other potential ow problems B.10 More potential ow . . . . . . . . . . . . . . . . . . . . . . . B.11 Boundary layers . . . . . . . . . . . . . . . . . . . . . . . . B.12 More boundary layers . . . . . . . . . . . . . . . . . . . . . B.13 Introduction to turbulence . . . . . . . . . . . . . . . . . . . B.14 Old Exams . . . . . . . . . . . . . . . . . . . . . . . . . . . C Study questions 5C1212

Preface

These lecture notes has evolved from a CFD course (5C1212) and a Fluid Mechanics course (5C1214) at the department of Mechanics and the department of Numerical Analysis and Computer Science (NADA) A at KTH. Erik St alberg and Ori Levin has typed most of the L TEXformulas and has created the electronic versions of most gures.

Stockholm, August 2004 Dan Henningson Martin Berggren

In the latest version of the lecture notes study questions for the CFD course 5C1212 and recitation material for the Fluid Mechanics course 5C1214 has been added.

Stockholm, August 2005 Dan Henningson

Chapter 1

Derivation of the Navier-Stokes equations

1.1 Notation
u + (u t u 1 p+

The Navier-Stokes equations in vector notation has the following form )u = = 0

2

where the velocity components are dened u = (u, v, w) = (u1 , u2 , u3 ) the nabla operator is dened as = the Laplace operator is written as
2

, , x y z

, , x1 x2 x3

2 2 2 + 2+ 2 2 x y z

and the following denitions are used p kinematic viscosity density pressure

see gure 1.1 for a denition of the coordinate system and the velocity components. The Cartesian tensor form of the equations can be written ui + uj ui t xj ui xi = = 0 1 p 2 ui + xi xj xj

where the summation convention is used. This implies that a repeated index is summed over, from 1 to 3, as follows ui ui = u 1 u1 + u 2 u2 + u 3 u3 Thus the rst component of the vector equation can be written out as 7

y, x2 v, u2 u, u1

PSfrag replacements x, x1 w, u3

z, x3 Figure 1.1: Denition of coordinate system and velocity components PSfrag replacements x, x1 2 x y, x2 P t=0 z, x3 u, u1 xi 0 v, u2 w, u3

ri

xi

t ,
u i xi 0 , t x1

Figure 1.2: Particle path.

i=1

u1 u1 u1 1 p u1 + u1 + u2 + u3 = + t x1 x2 x3 x1 u u u u 1 p +u +v +w = + t x y z x

2 u1 2 u1 2 u1 + + x1 2 x2 2 x3 2

or

2u 2u 2u + 2 + 2 x2 y z
2u

1.2

Kinematics

Lagrangian and Euler coordinates

Kinematics is the description of motion without regard to forces. We begin by considering the motion of a uid particle in Lagrangian coordinates, the coordinates familiar from classical mechanics. Lagrange coordinates: every particle is marked and followed in ow. The independent variables are xi 0 t initial position of uid particle time r i = r i xi 0 , t and the velocity of the particle is the rate of change of the particle position, i.e.

where the particle path of P, see gure 1.2, is

ui =

ri t

Note here that when xi 0 changes we consider new particles. Instead of marking every uid particle it is most of the time more convenient to use Euler coordinates. Euler coordinates: consider xed point in space, uid ows past point. The independent variables are xi t space coordinates time

Thus the uid velocity ui = ui (xi , t) is now considered as a function of the coordinate xi and time t. The relation between Lagrangian and Euler coordinates, i.e. xi 0 , t and (xi , t), is easily found by noting that the particle position is expressed in xed space coordinates xi , i.e. at the time xi = r i xi 0 , t

Material derivative

= t

Although it is usually most convenient to use Euler coordinates, we still need to consider the rate of change of quantities following a uid particle. This leads to the following denition. Material derivative: rate of change in time following uid particle expressed in Euler coordinates. Consider the quantity F following uid particle, where F = FL xi 0 , t = FE (xi , t) = FE ri xi , t , t The rate of change of F following a uid particle can then be written FE F FE ri FE t FE = + = + ui xi t t t t xi t D Based on this expression we dene the material derivative as Dt D = + ui = + (u ) Dt t xi t t In the material or substantial derivative the rst term measures the local rate of change and the second measures the change due to the motion with velocity ui . As an example we consider the acceleration of a uid particle in a steady converging river, see gure 1.3. The acceleration is dened aj = Dui ui ui = + uj Dt t xj

which can be simplied in 1D for stationary case to Du u u = +u Dt t x Note that the acceleration = 0 even if velocity at xed x does not change. This has been experienced by everyone in a raft in a converging river. The raft which is following the uid is accelerating although the ow eld is steady. a=

Description of deformation
Evolution of a line element Consider the two nearby particles in gure 1.4 during time dt. The position of P2 can by Taylor expansion be expressed as P2 : ri dt + dri dt = = ri (0) + rti (0) dt + dri (0) +

( dri ) t 0

dt

= xi 0 + dxi 0 + ui (0) dt + dui (0) dt

PSfrag replacements x, x1 y, x2 z, x3 u, u1 v, u2 w, u3 x1 x2 xi 0 r i xi 0 , t 0 u i xi , t P t=0 x

u1

u2 > u 1

Figure 1.3: Acceleration of uid particles in converging river.

PSfrag replacements x, x1 y, x2 z, x3 u, u1 v, u2 w, u3 x1 x2 xi 0 r i xi 0 , t 0 u i xi , t P t=0 x u1 u2 > u 1

x2 dxi0

ui d

P2

u i dt P1 xi
0

t ri d

x3 Figure 1.4: Relative motion of two nearly particles.

dr i
x1

dt

where we have used

( dri ) = t = and where ui xj 0 dxj 0 dui We can transform the expression line element in Euler coordinates

ri dxj 0 = xj 0 t xj 0 ui dxj 0 = dui xj 0

ri t

dxj 0

change of ui with initial pos. dierence in initial pos. dierence in velocity

( dri ) = dui in Lagrange coordinates to an equation for a material t

D ( dri ) = dui Dt = {expand in Euler coordinates} ui drj = xj

where ui xj drj change in velocity with spatial position dierence in spatial pos. of particles

Relative motion associated with invariant parts We consider the relative motion dui = ui ui drj by dividing in its invariant parts, i.e. xj xj ui = ij + eij + eij xj
eij

where eij is the deformation rate tensor and

ij eij eij

= = =

uj 1 ui anti-symmetric part 2 uj xi 1 ui uj 2 ur + ij traceless part 2 xj xi 3 ur 1 ur ij isotropic part 3 ur

ui , eij , describes the deformation and is considered in detail below, whereas the xj anti-symmetric part can be written in terms of the vorticity k and is associated with solid body rotation, i.e. no deformation. The anti-symmetric part can be written The symmetric part of

dxi0 dui dt r i dt dri dt P1 P2

r2

x2 R2

+ d12

R1 R x3 r3 Figure 1.5: Deformation of a cube.

3

x1 r1

1 [ij ] = 2 1 2

0
u1 x2 u1 x3

1 2

u1 x2

u2 x1 u3 x1

u2 x1

1 2 1 2

u1 x3 u2 x3

u3 x1 u3 x2

1 2

u2 x3

u3 x2

= {use = 0 1 = 2 3 1 2 2 Deformation of a small cube

u , k =
1 2 2 1 1 2

kji

ui } xj

1 2 3 0 1 2 1

Consider the deformation of the small cube in gure 1.5, where we dene Rl k
l rk

= Rkl = Rl + k

component k of side l uk l R dt xj j dul k relative motion of side l

= R kl +

uk dt xl

deformed cube

First, we consider the deformation on side 1, which can be expressed as dR1 = r1 R = The inner product can be expanded as u1 dt x1
2 1 1 r k rk R

1 1 rk rk

1+

u2 dt x1

u3 dt x1

R2

= {drop quadratic terms} = R2 1 + 2 u1 dt x1

We have dropped the quadratic terms since we are assuming that dt is small. dR1 becomes u1 u1 dt R = R 1 + dt + ... R x1 x1

dR1

= R = R

1+2

u1 dt = Re11 dt x1

which implies that dR1 = Re11 dt

ui Thus the deformation rate of side 1 depends on e11 , both traceless and isotropic part of xj . Second, we consider the deformation of the angle between side 1 and side 2. This can be expressed as

cos

+ d12 2

= = = =

r1 r2 1 2 1 1 2 2 = r k rk rm rm rn rn |r1 | |r2 | k1 + u1 x2 u1 x2

1/2 1/2

uk uk u1 dt k2 + dt 1 + 2 dt x1 x2 x1 u2 u1 u2 dt + dt 1 dt 1 dt x1 x1 x2 u2 + dt = 2e12 dt x1

1+2

u2 dt x2

1/2

where we have dropped quadratic terms. We use the trigonometric identity cos + d12 = cos cos d12 sin sin d12 d12 2 2 2

which allow us to obtain the nial expression d12 = 2e12 dt Thus the deformation rate of angle between side 1 and side 2 depends only on traceless part of Third, we consider the deformation of the volume of the cube. This can be expressed as r1 r2 r3 R 3 1+ = R3
u1 x1 dt u2 x1 dt u3 x1 dt ui xj .

dV

u1 x2 dt 2 1 + u2 dt x u3 x2 dt

u1 x3 dt u2 x3 dt 3 + u3 dt x

R3

= R3 1 +

u1 u2 dt 1+ dt x1 x2 u1 u2 u3 = R3 + + dt x1 x2 x3 uk = R3 dt xk

1+

u3 dt R3 x3

where we have again omitted quadratic terms. Thus we have dV = R3 err dt

ui and the deformation rate of volume of cube (or expansion rate) depends on isotropic part of xj . In summary, the motion of a uid particle with velocity ui can be divided into the following invariant parts

R3 R2 R1 r3 r2 r1 + d12 V (t)
S

Figure 1.6: Volume moving with the uid.

i) ii) iii) iv) ij = eij =

1 2

ui ui uj xj xi
1 = 2 kij k

1 2

ui uj 2 ur + ij xj xi 3 ur eij =
1 3

ur ij ur

1.3

Reynolds transport theorem

Volume integral following with the uid

Consider the time derivative of a material volume integral, i.e. a volume integral where the volume is moving with the uid. We obtain the following expressions 1 lim t0 t
t0 t

D Dt
V (t)

Tij dV

lim

1 + Tij (t + t) dV Tij (t) dV t V (t) V (t) 1 Tij lim Tij (t + t) dV + dV t0 t t

V (t+t)V (t) V (t)

The volume in the rst integral on the last line is represented in gure 1.6, where a volume element describing the change in volume between V at time t and t + t can be written as u nt = uk nk t dV = uk nk t dS

V (t+t)

V (t + t) V (t) S (t + t)

(t)

V (t + t)

solid body translation solid body rotation volume constant deformation volume expansion rate

Tij (t + t) dV
V (t)

V (t+t)

Tij (t + t) dV
V (t)

Tij (t) dV

Tij (t + t) dV

This implies that the volume integral can be converted to a surface integral. This surface integral can in turn be changed back to a volume integral by the use of Gauss (or Greens) theorem. We have D Dt
V (t)

Tij dV

t0

lim

Tij (t + t) uk nk dS +
S(t) V (t)

Tij dV t

=
S(t)

Tij uk nk dS +
V (t)

Tij dV = {Gauss/Greens theorem} t

=
V (t)

Tij + (uk Tij ) dV t xk

which is the Reynolds transport theorem.

Conservation of mass
By the substitution Tij and the use of the Reynolds transport theorem above we can derive the equation for the conservation of mass. We have D Dt
V (t)

dV =
V (t)

+ (uk ) = 0 t xk

Since the volume is arbitrary, the following must hold for the integrand 0= uk D uk + (uk ) = + uk + = + t xk t xk xk Dt xk
1 2 3 4

where we have used the denition of the material derivative in order to simplify the expression. The terms in the expression can be given the following interpretations:
1 2 3 4

: : : :

accumulation of mass in xed element net ow rate of mass out of element rate of density change of material element volume expansion rate of material element

By considering the transport of a quantity given per unit mass, i.e. Tij = tij , we can simplify Reynolds transport theorem. The integrand in the theorem can then be written 0 (cont. eq.) 0 ! tij tij Dtij B (tij ) + (uk tij ) = tij + + tij k ) + uk (u = t xk t t x xk Dt k which implies that Reynolds transport theorem becomes D Dt
V (t)

tij dV =
V (t)

Dtij dV Dt

where V (t) again is a material volume.

1.4

Momentum equation

Conservation of momentum
The momentum equation is based on the principle of conservation of momentum, i.e. that the time rate of change of momentum in a material region = sum of the forces on that region. The quantities involved are: Fi Ri ui body forces per unit mass surface forces per unit area momentum per unit volume

V (t + t) V (t) r i dt nIII(k) S (t + t) dri dt u dl P1 n P2 V (t + t) x1 R(nI ) x2 x3 R3 nII nI R2 R1 dS r3 2 R(nII ) r r1 + d12 2 Figure 1.7: (t) V Momentum balance for uid element S (t) V (t + t) V (t) n S (t + t) u n R V (t + t) R Figure 1.8: Surface force and unit normal. We can put the momentum conservation in integral form as follows D Dt
V (t)
s (k)

ui dV =
V (t)

Fi dV +
S(t)

Ri dS

Using Reynolds transport theorem this can be written

V (t)

Dui dV = Dt
V (t)

Fi dV +
S(t)

Ri dS

which is Newtons second law written for a volume of uid: mass acceleration = sum of forces. To proceed Ri must be investigated so that the surface integral can be transformed to a volume integral. In order to do that we have to dene the stress tensor.

The stress tensor

Remove a uid element and replace outside uid by surface forces as in gure 1.7. Here R(n) is the surface force per unit area on surface dS with normal n, see gure 1.8. Momentum conservation for the small uid particle leads to Dui dS dl = Fi dS dl + Ri nI dS + Ri nII dS + j j Dt Ri nIII(k) s(k) dl j
k

Letting dl 0 gives 0 = Ri nI dS + Ri nII dS j j Now nj = nI = nII which leads to j j Ri (nj ) = Ri (nj ) implying that a surface force on one side of a surface is balanced by an equal an opposite surface force at the other side of that surface. Note that it is a general principle that the terms proportional to the volume of a small uid particle approaches zero faster than the terms proportional to the surface area of the particle. Thus the surface forces acting on a small uid particle has to balance, irrespective of volume forces or acceleration terms.

u P1 R(e1 ) = ( T11 , T21 , T31 ) n V (t + t) P2 x1 T21 x2 x1 T11 x3 T31 PSfrag replacementsR3 2 x, x1 R y, x2 R1 z, x3 r3 Figure 1.9: Denition of surface force components on a surface with a normal in the 1-direction. u, u1 r2 x2 v, u2 r1 R(e2 ) = ( T12 , T22 , T32 ) + d w, u3 12 2 V x1 (t) T22 S x2 (t) V (t + t) xi0 (t) V ri Sxi0 + t) (t , t T12 u i xi 0 , t u P t=0 n V (t + t) x T32 u1 u2 > u 1 x1 x2 x3 xi 0 Figure 1.10: Denition u dt of surface force components on a surface with a normal in the 2-direction. i dxi0 dui dt We now divide the surfaceforces into components along the coordinate directions, as in gures 1.9 and r i dt 1.10, with corresponding denitions for the force components on a surface with a normal in the 3-direction. dri d Thus, Tij is the i-component t the surface force on a surface dS with a normal in the j-direction. of P1 Consider a uid particle with surfaces along the coordinate directions cut by a slanted surface, as in P2 gure 1.11. The areas of the surface elements are related by x1 x2 dSj = ej n dS = nj dS x3 where dS is the area of the slanted surface. Momentum balance require the surface forces to balance R3 in the element, we have R2 R1 3 r0 = R1 dS T11 n1 dS T12 n2 dS T13 n3 dS r2 which implies that the total surface force Ri can be written in terms of the components of the stress r1 tensor Tij as + d12 2 V (t) R = T n + T n + T n = T n i i1 1 i2 2 i3 3 ij j S (t) V (t + t) V (t) e2 S (t + t) R u n n V (t + t)

R1 dS1 dS2 dS3 e1

e3 Figure 1.11: Surface force balance on a uid particle with a slanted surface.

P Here the diagonal components are normal stresses o-diagonal components are t = 0 stresses. shear Further consideration of the moment balance around a uid particle show that Tij is ax symmetric tensor, u1 i.e. u2 > u 1 x1 Tij = Tji x2 x3 Momentum equation xi 0 i dt Using the denition of the surface force in terms of the stress tensor, the momentumuequation can now be dxi0 written dui dt r i dt Dui Tij dV = Fi dV + Tij nj dS = Fi + dV Dt xj dri dt V (t) V (t) S(t) V (t) P1 P2 The volume is again arbitrary implying that the integrand must itself equal zero. We have x1 x2 Dui Tij = Fi + x3 Dt xj R3 R2 Pressure and viscous stress tensor R1 r3 For uid at rest only normal stresses present otherwise uid element would deform. We thus divide the r stress tensor into an isotropic part, the hydrodynamic pressure p, and a part depending2 on the motion of r1 the uid. We have + d12 p 2 V (t) Tij = pij + ij S (t) V (t + t) V (t) p - hydrodynamic pressure, directed inward ij - viscous stress tensor, depends on uid motion S (t + t) u n V (t + t) p

Newtonian uid
It is natural to assume that the viscous stresses are functions of deformation rate e ij or strain. Recall that the invariant symmetric parts are eij = 1 2 ui uj 2 ur + ij xj xi 3 xr
eij

1 ur ij 3 xr
eij

where eij is the volume constant deformation rate and eij is the uniform rate of expansion. For an isotropic uid, the viscous stress tensor is a linear function of the invariant parts of e ij , i.e. ij = eij + 2eij where we have dened the two viscosities as (T ): (T ): dynamic viscosity (here T is the temperature) second viscosity, often=0

For a Newtonian uid, we thus have the following relationship between the viscous stress and the strain (deformation rate) ij = 2eij = which leads to the momentum equation Dui p = + Dt xi xj ui uj 2 ur + ij xj xi 3 xr + Fi ui uj 2 ur + ij xj xi 3 xr

1.5

Energy equation

The energy equation is a mathematical statement which is based on the physical law that the rate of change of energy in material particle = rate that energy is received by heat and work transfers by that particle. We have the following denitions
1 e + 2 ui ui dV

energy of particle, with e the internal energy work rate of Fi on particle

force velocity

ui Fi dV

uj Rj dS ni qi dS

work rate of Rj on particle heat loss from surface, with qi the heat ux vector, directed outward

Using Reynolds transport theorem we can put the energy conservation in integral form as D Dt
V (t)

1 e + ui ui dV = 2
V (t)

Fi ui dV +
S(t)

[ni Tij uj ni qi ] dS =
V (t)

Fi ui +

(Tij uj qi ) dV xi

Compare the expression in classical mechanics, where the momentum equation is mu = F and the associated kinetic energy equation is (u u) work rate ( work
m d 2 dt

= = =

Fu force velocity force dist. )

From the integral energy equation we obtain the total energy equation by the observation that the volume is arbitrary and thus that the integrand itself has to be zero. We have 1 qi D e + ui ui = Fi ui (pui ) + (ij uj ) Dt 2 xi xi xi The mechanical energy equation is found by taking the dot product between the momentum equation and u. We obtain D 1 p ij ui ui = Fi ui ui + ui Dt 2 xi xj Thermal energy equation is then found by subtracting the mechanical energy equation from the total energy equation, i.e. ui ui qi De = p + ij Dt xi xj xi The work of the surface forces divides into viscous and pressure work as follows (pui ) xi = p ui xi ui p xi
2

(ij uj ) xi

ij

ui xj

+ uj

ij xi

where following interpretations can be given to the thermal and the mechanical terms
1: 2:

thermal terms ( force deformation ): heat generated by compression and viscous dissipation mechanical terms ( velocity force gradients ): gradients accelerate uid and increase kinetic energy

The heat ux need to be related to the temperature gradients with Fouriers law qi = T xi

where = (T ) is the thermal conductivity. This allows us to write the thermal energy equation as De ui = p ++ Dt xi xi T xi

where the positive denite dissipation function is dened as ui 1 = 2 eij eij xj 3 1 uk ij 3 xk

2

= ij

uk xk

= 2 eij

>0

Alternative form of the thermal energy equation can be derived using the denition of the enthalpy h=e+ We have Dh De 1 Dp p = + 2 Dt Dt Dt D Dt
xi
i u

which gives the nal result Dh Dp = ++ Dt Dt xi T xi

To close the system of equations we need a i) ii) thermodynamic equation equation of state e = e (T, p) p = RT simplest case: e = cv T or h = cp T

where cv and cp are the specic heats at constant volume and temperature, respectively. At this time we also dene the ratio of the specic heats as = cp cv

1.6

Navier-Stokes equations

The derivation is now completed and we are left with the Navier-Stokes equations. They are the equation describing the conservation of mass, the equation describing the conservation of momentum and the equation describing the conservation of energy. We have D uk + =0 Dt xk where Dui p ij = + + Fi Dt xi xj T xi

De ui = p ++ Dt xi xi

= ij

and the thermodynamic relation and the equation of state for a gas e = e (T, p) p = RT We have 7 equations and 7 unknowns and therefore the necessary requirements to obtain a solution of the system of equations. unknown 1 ui 3 p 1 e 1 T 1 7 equations continuity momentum energy thermodyn. gas law 1 3 1 1 1 7

ij

ui xj uj 2 ur ui + ij = xj xi 3 xr

Equations in conservative form

A slightly dierent version of the equations can be found by using the identity to obtain the system t + x (uk ) = 0 k p ij (ui ) + (uk ui ) = + + Fi xk xi xj t 1 1 = Fi ui (pui ) + (ij uj ) + t e + 2 ui ui + xk uk e + 2 ui ui xi xi xi U G(i) + =J t xi where U = (, u1 , u2 , u3 , (e + ui ui /2)) is the vector of unknowns, J = (0, F1 , F2 , F3 , ui Fi ) is the vector of the right hand sides and G(i) =
T T

Dtij = (tij ) + (uk tij ) Dt t xk

T xi

These are all of the form

or

U G G G + + + =J t x y z

ui u1 ui + p1i 1i u2 ui + p2i 2i u3 ui + p3i 3i T (e + ui ui /2) ui + pui + xi uj ij

is the ux of mass, momentum and energy, respectively. This form of the equation is usually termed the conservative form of the Navier-Stokes equations.

1.7

Incompressible Navier-Stokes equations

The conservation of mass and momentum can be written D + ui = 0 Dt xi Dui p ij = + + Fi xi xj Dt = ui + uj 2 uk ij ij xj xi 3 uk

for incompressible ow = constant, which from the conservation of mass equation implies

ui =0 xi implying that a uid particle experiences no change in volume. Thus the conservation of mass and momentum reduce to ui + uj ui = 1 p + 2 ui + Fi t xj xi ui xi = 0 since the components of the viscous stress tensor can now be written xj ui uj 2 uk + ij xj xi 3 xk = 2 ui = xj xj
2

ui

We have also introduced the kinematic viscosity, , above, dened as = / The incompressible version of the conservation of mass and momentum equations are usually referred to as the incompressible Navier-Stokes equations, or just the Navier-Stokes equations in case it is evident that the incompressible limit is assumed. The reason that the energy equation is not included in the incompressible Navier-Stokes equations is that it decouples from the momentum and conservation of mass equations, as we will see below. In addition it can be worth noting that the conservation of mass equation is sometimes referred to as the continuity equation. The incompressible Navier-Stokes equations need boundary and initial conditions in order for a solution to be possible. Boundary conditions (BC) on solid surfaces are ui = 0. They are for obvious reasons usually referred to as the no slip conditions. As initial condition (IC) one needs to specify the velocity eld at the initial time ui (t = 0) = u0 . The pressure eld does not need to be specied as it can be obtained once the i velocity eld is specied, as will be discussed below.

Integral form of the Navier-Stokes eq.

We have derived the dierential form of the Navier-Stokes equations. Sometimes, for example when a nite-volume discretization is derived, it is convenient to use an integral form of the equations. An integral from of the continuity equation is found by integrating the divergence constraint over a xed volume VF and using the Gauss theorem. We have (ui ) dVF = xi ui ni dSF = 0
S

An integral form of the momentum equation is found by taking the time derivative of a xed volume integral of the velocity, substituting the dierential form of the Navier-Stokes equation, using the continuity equation and Gauss theorem. We have t
VF

ui dV =
VF

ui dV = t
VF

1 p 2 ui (uj ui ) + xj xi xj xj dV =
SF

dV =

VF

1 ui uj ui + pij xj xj

p ui ui uj n j + n i nj dS xj

Tw

T0 , U 0

Figure 1.12: Examples of length, velocity and temperature scales. Note that the integral from of the continuity equation is a compatibility condition for BC in incompressible ow.

Dimensionless form
It is often convenient to work with a non-dimensional form of the Navier-Stokes equations. We use the following scales and non-dimensional variables x = i xi L t = tU0 L u = i ui U0 p = p p0 1 = xi L x i U0 = t L t

where signies a non-dimensional variable. The length and velocity scales have to be chosen appropriately from the problem under investigation, so that they represent typical lengths and velocities present. See gure 1.12 for examples. If we introduce the non-dimensional variables in the Navier-Stokes equations we nd 2 U 2 u p0 p U0 2 u U0 ui i i + 0 u = + 2 j L t L xj L x L x x i j j U0 u i =0 L x i ui + uj ui = p0 p + 2 t xj U0 xi U0 L ui =0 xi
2 U0 L U0 /L = U0 /L2 2

2 Now drop * as a sign of non-dimensional variables and divide through with U0 /L, we nd

We have dened the Reynolds number Re as Re =

inertial forces viscous forces

which can be interpreted as a measure of the inertial forces divided by the viscous forces. The Reynolds number is by far the single most important non-dimensional number in uid mechanics. We also dene the pressure scale as
2 p0 = U0

which leaves us with the nal form of the non-dimensional incompressible Navier-Stokes equations as ui + uj ui = p + 1 t xj xi Re ui =0 xi
2

T0 U0

S (t + t) u n V (t + t) p

Tw L

ui ui

Non-dimensional energy equation

We stated above that the energy equation decouple from the rest of the Navier-Stokes equations for incompressible ow. This can be seen from a non-dimensionalization of the energy equation. We use the denition of the enthalpy (h = cp T ) to write the thermal energy equation as DT Dp = + + 2T Dt Dt and use the following non-dimensional forms of the temperature and dissipation function cp T = This leads to DT Dt T T0 Tw T 0 = L2 2 U0

= =

U0 L cp 1 1 Re P r
2

T +

T +

2 U0 cp (Tw T0 ) M a2 cp (Tw0 T
a2 0)

2 U0 Dp + cp (Tw T0 ) Dt U0 L

Dp 1 + Dt R

where a0 speed of sound in free-stream and M a = to obtain

U0 is Mach number. We now drop and let M a 0 a0

DT 1 1 = 2T Dt Re P r c where the Prandtl number P r = p , which takes on a value of 0.7 for air. As the Mach number approaches zero, the uid behaves as an incompressible medium, which can be seen from the denition of the speed of sound. We have a0 = , = 1 p

where is the isothermal compressibility coecient. If the density is constant, not depending on p, we have that 0 and a0 , which implies that M a 0.

1.8

Role of the pressure in incompressible ow

The role of the pressure in incompressible ow is special due to the absence of a time derivative in the continuity equation. We will illustrate this in two dierent ways.

Articial compressibility
First we discuss the articial compressibility version of the equations, used sometimes for solution of the steady equations. We dene a simplied continuity equation and equation of state as ui + =0 and p = t xi This allows us to write the Navier-Stokes equations as ui + (uj ui ) = p + 1 t xj xi Re ui p + = 0 t xi Let p u v u p + u2 uv u= e= f = v uv p + v2 w uw vw

ui

w uw g= vw p + 2

be the vector of unknowns and their uxes. The Navier-Stokes equations can then be written u e f g 1 + + + = t x y z Re or
2

Du
2

u u u u 1 +A +B +C = t x y z R 0 0 0 0 , B = f = u 0 u 0 u 0 0 0 0 v u 0 , D = 0 0 2v 0 0 w v 0

Du

the eigenvalues of A, B, C are the wave speeds of plane waves in the respective coordinate directions, they are and w, w, w w2 + Note that the eective acoustic or pressure wave speed for incompressible ow. u, u, u v, v, v u2 + , v2 +

where the matrices above are dened 0 e 1 2u = A= u 0 v 0 w 0 g 0 C= = u 1 0

0 0 0 v u 1 0 2v 0 0 w 0 1 0 0 0 0 1 0 0 0 0 1

0 0 0 v

Projection on a divergence free space

Second we discuss the projection of the velocity eld on a divergence free space. We begin by the following theorem. Theorem 1. Any wi in can uniquely be decomposed into wi = u i + where ui = 0, xi p xi on .

u i ni = 0

i.e. into a function ui that is divergence free and parallel to the boundary and the gradient of a function, here called p. Proof. We start by showing that ui and ui , p xi =

p is orthogonal in the L2 inner product. xi p ui dV = (ui p) dV = pui ni dS = 0 xi xi

The uniqueness of the decomposition can be seen by assuming that we have two dierent decompositions and showing that they have to be equivalent. Let wi = u i The inner product between ui gives 0=
(1) (1)

p(1) p(2) (2) = ui + xi xi

ui

(2)

p(1) p(2) xi

and ui

(1)

ui

(2)

ui

(1)

ui

(2)

+ ui

(1)

ui

(2)

p(1) p(2) xi

dV =

ui

(1)

ui

(2)

dV

Thus we have

ui

(1)

= ui ,

(2)

p(1) = p(2) + C

r + d12 V (t) S (t) V (t + t) V (t) S (t + t) u n V (t + t) p T0 , U 0 Tw L T0 U0

2

Gradient elds

wi
p xi

ui

Divergence free vectorelds // to boundary Figure 1.13: Projection of a function on a divergence free space. We can nd an equation for a p with above properties by noting that
2

p=

wi xi

in with

p = wi ni on n

has a unique solution. Now, if ui = w i we have 0= ui wi 2p = ; xi xi xi xi

p xi 0 = u i ni = w i ni p xi p n

and

To sum up, to project wi on divergence free space let wi = ui + 1) solve for p : wi = xi p xi

2

p,

p =0 n

2) let ui = wi

This is schematically shown in gure 1.13. Note also that (wi ui ) ui .

Apply to Navier-Stokes equations

Let P be orthogonal projector which maps wi on divergence free part ui , i.e. Pwi = ui . We then have the following relations wi = Pwi + Pui = ui , P p xi p =0 xi

Applying the orthogonal projector to the Navier-Stokes equations, we have P ui p + t xi = P uj ui 1 + xj Re

2

ui

Now, ui is divergence free and parallel to boundary, thus

P However, P

ui ui = t t

ui =

ui

since 2 ui need not be parallel to the boundary. Thus we nd an evolution equation without the pressure, we have ui ui 1 = P uj + t xj Re
wi 2

The pressure term in the Naiver-Stokes equations ensures that the right hand side w i is divergence free. We can nd a Poisson equation for the pressure by taking the divergence of w i , according to the result above. We nd
2

ui

p=

xi

thus the pressure satises elliptic Poisson equation, which links the velocity eld in the whole domain instantaneously. This can be interpreted such that the information in incompressible ow spreads innitely fast, i.e. we have an innite wave speed for pressure waves, something seen in the analysis of the articial compressibility equations.

ui 1 + uj xj Re
wi

uj ui ui = xi xj

Evolution equation for the divergence

It is common in several numerical solution algorithms for the Naiver-Stokes equations to discard the divergence constraint and instead use the pressure Poisson equation derived above as the second equation in the incompressible Naiver-Stokes equations. If this is done we may encounter solutions that are not divergence free. Consider the equation for the evolution of the divergence, found by taking the divergence of the momentum equations and substituting the Laplacian of the pressure with the right hand side of the Poisson equation. We have t ui xi = 1 Re
2

ui xi

Thus the divergence is not automatically zero, but satises a heat equation. For the stationary case the solution obeys the maximum principle. This states that the maximum of a harmonic function, a function satisfying the Laplace equation, has its maximum on the boundary. Thus, the divergence is zero inside the domain, if and only if it is zero everywhere on the boundary. Since it is the pressure term in the Navier-Stokes equations ensures that the velocity is divergence free, this has implications for the boundary conditions for the pressure Poisson equation. A priori we have none specied, but they have to be chosen such that the divergence of the velocity eld is zero on the boundary. This may be a dicult constraint to satisfy in a numerical solution algorithm.

Chapter 2

z, x3 u, u1 v, u2 w, u3 x1 x2 xi 0 r i xi 0 , t 0 u i xi , t P t=0

Flowuxphysics
1

u2 > u 1 x1 x2 x3 2.1 Exact solutions xi 0 u i dt Plane Pouseuille ow - exact solution for channel ow dxi0 dui dt The ow inside the two-dimensional channel, see gure 2.1, is driven by a pressure dierence p 0 p1 between r i dt the inlet and the outlet of the channel. We will assume two-dimensional, steady ow, i.e. dri dt P1 = 0, u3 = 0 P2 t x1 and that the ow is fully developed, meaning that eects of inlet conditions have disappeared. We have x2

x3 u1 = u1 (x2 ) , u2 = u2 (x2 ) R3 2 R The boundary conditions are u1 = u2 = 0 at x2 = h. The continuity equation reads R1 r3 0 0 r2 U u2  U  u1 u3 r1 +  =0  + x x2  3 x + d12 2  1 V (t) where the rst and the last term disappears due to the assumption of two-dimensional ow and that S (t) the ow is fully developed. This implies that u2 = C = 0, where the constant is seen to be zero from the V (t + t) V (t) boundary conditions. S (t + t) The steady momentum equations read u n u u1 + u u1 = p + 2 u V (t + t) 1 2 1 x1 x2 x1 p u u2 + u u2 = p + 2 u g 1 2 2 T0 , U 0 x1 x2 x2 Tw which, by applying the assumptions above, reduce to L T0 U0 x2 L Gradient elds wi p h xi ui ree vectorelds // to boundary p0 > p 1 x1 p

                                                                 
29

Figure 2.1: Plane channel geometry.

ui Divergence free vectorelds // to boundary L h x1 x2 p0 > p 1 p1 dS

u n

un t

Figure 2.2: Volume of uid owing through surface dS in time t

0=

p d2 u 1 + x1 dx2 2 p g 0= x2

The momentum equation in the x2 -direction (or normal direction) implies p = gx2 + P (x1 ) p dP = x1 dx1

showing that the pressure gradient in the x1 -direction (or streamwise direction) is only a function of x1 . The momentum equation in the streamwise direction implies 0= Since P (x1 ) and u1 (x2 ) we have d2 u 1 1 dP = = const. (independent of x1 , x2 ) dx2 dx1 2 We can integrate u1 in the normal direction to obtain u1 = 1 dP 2 x + C 1 x2 + C 2 2 dx1 2 dP d2 u 1 + dx1 dx2 2

The constants can be evaluated from the boundary conditions u1 (h) = 0, giving the parabolic velocity prole u1 = h2 dP x2 1 2 dx1 h
2

Evaluating the maximum velocity at the center of the channel we have umax = h2 dP >0 2 dx1

if ow is in the positive x1 -direction. The velocity becomes u1 x2 =1 umax h Flow rate From gure 2.2 we can evaluate the ow rate as dQ = u n dS. Integrating over the channel we nd
h 2

Q=
S

ui ni dS = {channel} =

u1 dx2 =

2h3 dP 4 = humax 3 dx1 3

L h x1 x2 p0 > p 1 p1 u n dS u nt

u0 p0 x

Wall shear stress The wall shear stress can be evaluated from the velocity gradient at the wall. We have 12 = 21 = implying that wall = Vorticity The vorticity is zero in the streamwise and normal directions, i.e. 1 = 2 = 0 and has the following expression in the x3 -direction (or spanwise direction) 3 = u2 u1 du1 2umax x2 = = x1 x2 dx2 h2
x2 =h

Stokes 1:st problem: instantaneously started plate

Consider the instantaneously starting plate of innite horizontal extent shown in gure 2.3. This is a time dependent problem where plate velocity is set to u = u0 at time t = 0. We assume that the velocity can be written u = u(y, t) which using continuity and the boundary conditions imply that normal velocity v = 0. The momentum equations reduce to u t 1 p 2u + 2 x y 1 p 0 = g y =

The normal momentum equation implies that p = gy + p0 , where p0 is the atmospheric pressure at the plate surface. This gives us the following diusion equation, initial and boundary conditions for u u 2u = 2 t y u(y, 0) = 0 0 y u(0, t) = u0 t > 0 u(, t) = 0 t > 0

We look for a similarity solution, i.e. we introduce a new dependent variable which is a combination of y and t such that the partial dierential equation reduces to an ordinary dierential equation. Let

          
2 umax x2 du1 = dx2 h2 2 umax h =

Figure 2.3: Stokes instantaneously plate.

u = C y tb u0 where C and b are constants to be chosen appropriately. We transform the time and space derivatives according to f () = t y 2 y 2 d d = Cbytb1 t d d d d = = Ctb y d d d2 = C 2 t2b 2 d =

If we substitute this into the equation for u and use the denition of , we nd an equation for f df d2 f = C 2 t2b 2 d d where no explicit dependence on y and t can remain if a similarity solution is to exist. Thus, the coecient of the two -dependent terms have to be proportional to each other, we have bt1 bt1 = C 2 t2b where is a proportionality constant. The exponents of these expressions, as well as the coecients in front of the t-terms have to be equal. This gives b= 1 2 C= 1 2

We choose = 2 and obtain the following f + 2f = 0

y = 2 t

d where = d . The boundary and initial conditions also have to be compatible with the similarity assumption. We have

u(y, 0) u(0, t) = f () = 0 = f (0) = 1 u0 u0 This equation can be integrated twice to obtain

u(, t) = f () = 0 u0

f = C1
0

e d + C2

Using the boundary conditions and the denition of the error-function, we have
2 2 f =1 e d = 1 erf() 0 This solution is shown in gure 2.4, both as a function of the similarity variable and the unscaled normal coordinate y. Note that the time dependent solution is diusing upward. From the velocity we can calculate the spanwise vorticity as a function of the similarity variable and the unscaled normal coordinate y. Here the maximum of the vorticity is also changing in time, from the innite value at t = 0 it is diusing outward in time.

v u u0 2 = e x y t The vorticity is shown in gure 2.5, both as a function of the similarity variable and the unscaled normal coordinate y. Note again that the time dependent solution is diusing upward. 3 = z =

2.2

Vorticity and streamfunction

Vorticity is an important concept in uid dynamics. It is related to the average angular momentum of a uid particle and the swirl present in the ow. However, a ow with circular streamlines may have zero vorticity and a ow with straight streamlines may have a non-zero vorticity.

relds // to boundaryn Divergence free vectorelds // to boundary n dS dS L L u nt u nt hx hx x1 y x1 y 2 xu x y = 2 0 yp=> p 4t u0 t 4t p0 > p 1 0 1 p0 p0 y p1 p1 = 4t u u n n y = 4t dS dS t u u u nt u nt u0 u0 x x Figure 2.4: Velocity above the instantaneously started plate. a) U as a function of the similarity variable y y . b) U (y, t) for various times. u0 u0 p0 p0 y y = 4t = 4t
2

1.8 1.6 1.4 1.2 1

0.8 0.6 0.4 0.2 0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.8 1.6 1.4 1.2 1 0.8 0.6

y = 4t t y z =
()u0 t

u u0

y = 4t ty

u u0

2
0.4 0.2 0

0.1

0.2

0.3

z = = () Figure 2.5: Vorticity above the instantaneously started plate. a) z t/u0 as a function of the similarity variable . b) z (y, t) for various times.

t u0

0.4

0.5

0.6

0.7

0.8

0.9

z t u0

= ()

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

()u0 t

Vorticity and circulation

The vorticity is dened mathematically as the curl of the velocity eld as = In tensor notation this expression becomes i =
ijk

u uk xj

Another concept closely related to the vorticity is the circulation, which is dened as =
C

ui dxi =
C

ui ti ds

In gure 2.6 we show a closed curve C. Using Stokes theorem we can transform that integral to one over the area S as uk = dS = i ni dS ijk ni xj S S This allows us to nd the following relationship between the circulation and vorticity d = i ni dS, which can also be written d = i ni dS Thus, one interpretation is that the vorticity is the circulation per unit area for a surface perpendicular to the vorticity vector. Example 1. The circulation of an ideal vortex. Let the azimuthal velocity be given as u = C/r. We can calculate the circulation of this ow as
2

=
C

u r d = C
0

d = 2C

Thus we have the following relationship for the ideal vortex u = 2r

 u0 y = 4t t y ()u z = t0 t

dxi = ti ds

Figure 2.6: Integral along a closed curve C with enclosed area S. The ideal vortex has its name from the fact that its vorticity is zero everywhere, except for an innite value at the center of the vortex.

Derivation of the Vorticity Equation

We will now derive an equation for the vorticity. We start with the dimensionless momentum equations. We have ui ui p 1 2 + uj = + ui t xj xi Re We slightly modify this equation by introducing the following alternative form of the non-linear term uj ui = xj xi 1 uj uj 2 +
ijk j uk pqi xq )

this can readily be derived using tensor manipulation. We take the curl ( equations and nd t
pqi

of the momentum ui xq

ui xq

pqi

2 xq xi

1 uj uj 2

pqi

( xq

ijk j uk )

pqi

2p 1 + xq xi Re

pqi

The second and the fourth terms are zero, since i, q

pqi

pqi

is anti-symmetric in i, q and xq xi is symmetric in

2 =0 xq xi

The rst term and the last term can be written as derivatives of the vorticity, and we now have the following from of the vorticity equation 1 2 p + pqi ( ijk j uk ) = p t xq Re The second term in this equation can be written as
ipq ijk

(j uk ) = (pj qk pk qj ) (j uk ) = (p uk ) (j up ) = xq xq xk xj uk p uk j up + p up j xk xk xj xj

Due to continuity

j xj

2 uq jpq xj xp

= 0 and we are left with the relation

ipq ijk

p up (j uk ) = uk j xq xk xj

Thus, the vorticity equation becomes p p up 1 + uk = j + t xk xj Re or written in vector-notation D = ( Dt )u + 1 Re

2 2

Components in Cartesian coordinates

In cartesian coordinates the components of the vorticity vector in three-dimensions become ex =
x

ey
y

ez
z

w v y z

ex +

u w z x

ey +

v u x y

ez

For two-dimensional ow it is easy to see that this reduces to a non-zero vorticity in the spanwise direction, but zero in the other two directions. We have x = 0, y = 0, z = v u x y

This simplies the interpretation and use of the concept of vorticity greatly in two-dimensional ow.

Vorticity and viscocity

There exists a subtle relationship between ows with vorticity and ows on which viscous forces play a role. It is possible to show that = 0 viscous forces = 0 This means that without viscous forces there cannot be a vorticity eld in the ow which varies with the coordinate directions. This is expressed by the following relationship ij = xj
2

ui =

ijk

k xj

This can be derived in the following manner. We have

ijk

k xj

un xj xm 2 un = (im jn in jm ) xj xm =
ijk kmn

= =

2 ui xj xj
2

ui

Terms in the two-dimensional vorticity equation

In two dimensions the equation for the only non-zero vorticity component can be written 3 3 3 + u1 + u2 = t x1 x2
2

We will take two examples elucidating the meaning of the terms in the vorticity equation. First the diusion of vorticity and second the advection or transport of vorticity. Example 2. Diusion of vorticity: the Stokes 1st problem revisited. From the solution of Stokes 1st problem above, we nd that the vorticity has the following form 3 = z = (y, t) and that it thus is a solution of the following diusion equation 2 = 2 t y The time evolution of the vorticiy can be seen in gure 2.7, and is given by u0 2 y = e , where = t 2 t where the factor t can be identied as the diusion length scale, where we have used the following dimensional relations L2 [] = , [t] = T [ t] = L T

ree vectorelds // to boundary L h x1 x2 p0 > p 1 p1 u n dS u nt x y u0 p0 y = 4t y = 4t t y ()u0 z = t t

u u0

y = 4t t y y ()u z = t0 t

y = 4t u u0

u0 p0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Figure 2.7: A sketch of the evolution of vorticity in Stokes 1st problem. v, y

=0 x, u Figure 2.8: Stagnation point ow with a viscous layer close to the wall. In this ow the streamlines are straight lines and therefore the non-linear term, or the transport/advection term, is zero. We now turn to an example where the transport of vorticity is important. Example 3. Transport/advection of vorticity: Stagnation point ow. The stagnation point ow is depicted in gure 2.8, and is a ow where a uniform velocity approaches a plate where it is showed down and turned to a ow parallel two the plates in both the positive and negative x-direction. At the origin we have a stagnation point. The inviscid stagnation point ow is given as follows inviscid: u = cx v = cy z = = v u =0 x y

Note that this ow has zero vorticity and that it does not satisfy the no-slip condition (u = v = 0, y = 0). In order to satisfy this condition we introduce viscosity, or equally, vorticity. Thus close to the plate we have to fulll the equation 2 2 u +v = + x y x2 y 2 The two rst terms represent advection or transport of vorticity. We will try a simple modication of the inviscid ow close to the boundary, which is written as u = xf (y) v = f (y) This satises the continuity equation, and will be zero on the boundary and approach the inviscid ow far above the plate if we apply the following boundary conditions u = v = 0, y=0 u cx, v cy, y The vorticity can be written = xf (y). We introduce this and the above denitions of the velocity components into the vorticity equation, and nd f f + f f + f =0

y = 4t t y F,F , F z = ()u0 t t

y 4t u u0

p0

1.5

0.5

F
0 0 1

Figure 2.9: Solution to the non-dimensional equation for stagnation point ow. F is proportional to the normal velocity, F to the streamwise velocity and F to the vorticity. The boundary conditions become f =f =0 y=0 f cy, f c, y An integral to the equation exists, which can be written f
2

f f f

= K = {evaluate y } = c2

This is as far as we can come analytically and to make further numerical treatment simpler we make the equation non-dimensional with and c. Note that they have the dimensions [] = and that we can use the length scale dependent variables as L2 , T [c] = 1 T c to scale the independent and the

/c and the velocity scale y /c f (y) c

F ()

= =

This results in the following non-dimensional equation F with the boundary conditions F = F = 0, = 0 F , In gure 2.9 a solution of the equation is shown. It can be seen that diusion of vorticity from the wall is balanced by transport downward by v and outward by u. The thin layer of vortical ow close to the wall has the thickness /c.
2

FF F

=1

Stretching and tilting of vortex lines

In three-dimensions the vorticity is not restricted to a single non-zero component and the three-dimensional vorticity equation governing the vorticity vector becomes i i ui 1 + uj = j + t xj xj Re
advection of vorticity 2

diusion of vorticity

y = 4t t y ()u z = t0 t

4t u u0

dr

C(t)
Figure 2.10: A closed material curve. The complex motion of the vorticity vector for a full three-dimensional ow can be understood by an analogy to the equation governing the evolution of a material line, discussed in the rst chapter. The equation for a material line is D ui dri = drj Dt xj Note that if we disregard the diusion term, i.e. the last term of the vorticity equation, the two equations are identical. Thus we can draw the following conclusions about the development of the vortex vector i) Stretching of vortex lines (line ) produce i like stretching of dri produces length

ii) Tilting vortex lines produce i in one direction at expense of i in other direction

Helmholz and Kelvins theorems

The analogy with the equation for the material line directly give us a proof of Helmholz theorem. We have Theorem 1. Helmholz theorem for inviscid ow: Vortex lines are material lines in inviscid ow. A second theorem valid in inviscid ow is Kelvins theorem. Theorem 2. Kelvins theorem for inviscid ow: Circulation around a material curve is constant in inviscid ow. Proof. We show this by considering the circulation for a closed material curve, see gure 2.10, m =
C(t)

ui dri

We can evaluate the material time derivative of this expression with the use of Lagrangian coordinates as follows Dm Dt = D ui dri Dt C(t) Lagrange coordinates ui = ui (x0 , t) = ui (x0 (m), t) i i 0 0 ) ri = ri (xi , t) = ri (xi (m), t d ri dm u m i m dt ui ri ri dm + ui m t m m t m

= = =

dm

The last term of this expression will vanish since ui

m

ri t m

dm =
m

ui

ui dm = m

1 ui ui 2

dm = 0

y, u0 v y = 4t t y z = ()u0 t t dx

u = (u, v)

dy

x, u Figure 2.11: A streamline which is tangent to the velocity vector. Thus we nd the following resulting expression Dm Dt =
C(t)

Dui dri Dt
C(t)

=
C(t)

p 1 + xi Re
2

ui dri Re

1 Re

ui dri 0,

Which shows that the circulation for a material curve changes only if the viscous force = 0

Streamfunction
We end this section with the denition and some properties of the streamfunction. To dene the streamfunction we need the streamlines which are tangent to the velocity vector. In two dimensions, see gure 2.11, we have dx u dx dy = = u dy v dx = 0 dy v u v and in three dimensions we can write dx dy dz = = u v w For two-dimensional ow we can nd the streamlines using a streamfunction. We can introduce a streamfunction = (x, y) with the property that the streamfunction is constant along streamlines. We can make the following derivation d = dx + dy x y , v= and check for consistency = assume u = y x = u dy v dx = 0, on streamlines

We can also quickly check that the denition given above satises continuity. We have u v 2 2 + = =0 x y xy yx Another property of the streamfunction is that the volume ux between two streamlines can be found from the dierence by their respective streamfunctions. We use gure 2.12 and gure 2.13 in the following derivations of the volume ux:

z =

t n dS u nt x y u0 p0 y = 4t y = 4t t y ()u z = t0 t dy
u u0

tu

B = B S n A = A Figure 2.12: Integration paths between two streamlines.

ds

nx

ny

|n| = 1

dx Figure 2.13: The normal vector to a line element.

QAB

=
A B

ui ni ds (nx u + ny v) ds
A

= = =
A

1 ds = , dy nx
B

ds 1 = dx ny
B A

(u dy v dx) =

d = B A

2.3

Potential ow

Flows which can be found from the gradient of a scalar function, a so called potential, will be dealt with in this section. Working with the velocity potential, for example, will greatly simplify calculations, but also restrict the validity of the solutions. We start with a denition of the velocity potential, then discuss simplications of the momentum equations when such potentials exist, and nally we deal with two-dimensional potential ows, where analytic function theory can be used.

Velocity potential
Assume that the velocity can be found from a potential. In three dimensions we have ui = xi or u =

with the corresponding relations in two dimensions u = v =

x y

There are several consequences of this denition of the velocity eld.

i) Potential ow have no vorticity, which can be seen by taking the curl of the gradient of the velocity potential, i.e. uk 2 i = ijk = ijk =0 xj xj xk ii) Potential ow is not inuenced by viscous forces, which is seen by the relationship between viscous forces and vorticity, derived earlier. We have ij = xj
2

ui =

ijk

k xj

iii) Potential ow satises Laplace equation, which is a consequence of the divergence constraint applied to the gradient of the velocity potential, i.e. ui 2 = = xi xi xi
2

=0

Bernoullis equation
For potential ow the velocity eld can thus be found by solving Laplace equation for the velocity potential. The momentum equations can then be used to nd the pressure from the so called Bernoullis equation. We start by rewriting the momentum equation ui ui 1 p + uj = + t xj xi using the alternative form of the non-linear terms, i.e. uj and obtain the following equation ui + t xi 1 p uk uk + + gx3 2 =
ijk uj k 2

ui gi3

ui = xj xi

1 uk uk 2

ijk uj k

ijk

k xj

This equation can be integrated in for two dierent cases: potential ow (without vorticity) and stationary inviscid ow with vorticity.
i) Potential ow. Introduce the denition of the velocity potential ui = xi into the momentum equation above, we have 1 p + uk uk + + gx3 = 0 xi t 2

p 1 2 + (u1 + u2 + u2 ) + + gx3 = f (t) 2 3 t 2 ii) Stationary, inviscid ow with vorticity. Integrate the momentum equations along a streamline, see gure 2.14. We have ti xi 1 p uk uk + + gx3 2 ds =
ijk

ui uj k ds = 0 u

1 2 p (u + u2 + u2 ) + + gx3 = constant along streamlines 2 3 2 1 Thus, Bernoullis equation can be used to calculate the pressure when the velocity is known, for two dierent ow situations. Example 4. Ideal vortex: Swirling stationary ow without vorticity. The velocity potential u = polar coordinates can be written ur = r u = 1 r in

h u t x1 x 2 p()uy 0 > p0 1 z = t 1 p u t n t= u u dS u nt Figure 2.14: Integration of the momentum equation along a streamline. x y u0 r p0 y = 4t y = 4t t y z = ()u0 t t Figure 2.15: Streamlines for the ideal vortex. The velocity potential satises Laplace equation, which in polar coordinates becomes
2

u u0

1 r r

1 r r

1 2 =0 r 2

The rst term vanishes since

= ur = 0. Integrate the resulting equation twice and we nd = C + D

The constant D is arbitrary so we can set it to zero. Using the denition of the circulation we nd u = C , r C= 2

We can easily nd the streamfunction for this ow. In polar coordinates the streamfunction is dened ur = 1 r Note that the streamfunction satises continuity, i.e. potential to nd the streamfunction, we have = r 2r u = r u = 0. We use the solution from the velocity = ln r 2

The streamfunction is constant along streamlines resulting in circular streamlines, see gure 2.15 We can now use Bernoullis equation to calculate the pressure distribution for the ideal vortex as 1 2 u + p = p 2 p = p 2 8 2 r2

Note that the solution has a singularity at the origin.

Two-dimensional potential ow using analytic functions

For two-dimensional ows we can use analytic function theory, i.e. complex valued functions, to calculate the velocity potential. In fact, as we will show in the following. Any analytic function represents a twodimensional potential ow.

We start by showing that both the velocity potential and the streamfunciton satisfy Laplace equation. Velocity potential for two-dimensional ow is dened u = x Using the continuity equation gives u v + = x y u = v v = y
2

=0

The streamfunction is dened

y x
2

By noting that potential ow has no vorticity we nd = v y + = x y =0

Thus, both the velocity potential and the streamfunction satisfy Laplace equation. We can utilize analytic function theory by introducing the complex function F (z) = (x, y) + i(x, y) where z = x + iy = rei = r(cos + i sin ) which will be denoted the complex potential. We now recall some useful results from analytic function theory. F (z) exists and is unique F (z) is an analytic function

The fact that a derivative of a complex function should be unique rests on the validity of the Cauchy Riemann equations. We can derive those equations by requiring that a derivative of a complex function should be the same independent of the direction we take the limit in the complex plane, see gure 2.16. We have 1 F (z) = lim [F (z + z) F (z)] z0 z 1 = lim [(x + x, y) (x, y) + i(x + x, y) i(x, y)] x0 x 1 = lim [(x, y + y) (x, y) + i(x, y + y) i(x, y)] iy0 iy = real part x y 1 i = = imaginary part x i y x y Thus we have the CauchyRiemann equations as a necessary requirement for a complex function to be analytic, i.e. = and = x y y x Using the Cauchy-Riemann equations we can show that the real and imaginary part of the complex potential satisfy Laplace equations and thus that they are candidates for identication with the velocity potential and streamfunction of a two-dimensional potential ow. We have 2 2 = = = 2 x2 x y y x y 2 = 2 y y x = x y = 2 x2

y = 4t t y ()u0 z = t t x

y 4t u u0

u0 p0

iy

iy

Figure 2.16: The complex plane and the denitions of analytic functions. In addition to satisfying the Laplace equations, we also have to show that the Cauchy-Riemann equations are consistent with the denition of the velocity potential and the streamfunciton. The CauchyRiemann equations are u= = x y which recovers the denitions of the velocity potential and the streamfunction. Thus any analytic complex function can be identied with a potential ow, where the real part is the velocity potential and the imaginary part is the streamfunction. We can now dene the complex velocity as the complex conjugate of the velocity vector since W (z) = dF = +i = u iv dz x x v = = y x

It will be useful to have a similar relation for polar coordinates, see gure 2.17. In this case we have u = ur cos u sin v = ur sin + u cos W (z) = u iv = [ur (r, ) iu (r, )] ei

cos i sin

We will now take several examples of how potential velocity elds can be found from analytic functions. Example 5. Calculate the velocity eld from the complex potential F = U ei z. The complex velocity becomes W (z) = F = U ei = U (cos i sin ) which gives us the solution for the velocity eld in physical variables, see gure 2.18, as u = U cos v = U sin Example 6. Calculate the velocity eld from the complex potential F = Az n . In polar coordinates the complex potential becomes Arn ein = Arn cos(n) + iAr n sin(n) Thus the velocity potential and the streamfunction can be identied as = Arn cos(n) = Arn sin(n)

u nt P2 x xy 1 x2 u0 x3 p0 y 3 R = 4t u R2 u1 R0 y y = 4t r3 t 2 v r 1 ry d12 2 + ()u z = t0 u V (t) S (t) t u V (t + t) V (t) S (t + t) ur u u n V (t + t) r p T0 , U 0 Tw x L T0 Figure 2.17: The denition of polar coordinates. U0 Gradient elds wi
p xi

ui Divergence free vectorelds // to boundary L h x1 x2 p0 > p 1 p1 u n dS u nt x y u0 p0 y = 4t y = 4t t y ()u z = t0 t Figure 2.18: Constant velocity at an angle .
u u0

xi y =u nt 4t =0 u i dt x t 0 dxiy = 0 dui uy d0 t ()u 0 p r dt 0 i z = t y = 4t dri dt ut u1 P0 n y = 4t P2 =0 t x1 x2 Figure 2.19: Flow towards a corner. xy 3 ()u0 z = t 3 R R2 t =0 R1 r3 r2Figure 2.20: Uniform ow over a at plate. r1 + d12 2 and we can calculate the complex velocity as V (t) S (t) (z) = F = nAz n1 W V (t + t) V (t) = nArn1 ein ei S (t + t) = nArn1 [cos(n) + i sin(n)]ei u n This gives us the solution for the velocity in polar coordinates as V (t + t) ur = nArn1 cos(n) p u = nArn1 sin(n) T0 , U 0 Tw This is a ow which approaches a corner, see gure 2.19. The streamfunction is zero on the walls and on L the stagnation point streamline. By solving for the zero streamlines we can calculate the angle between a T0 wall aligned with the x-axis and the stagnation point streamline. We have U0 Gradient elds = Arn sin(n) = 0 = k wi n
xi Figure 2.20 shows the uniform ow over a at plate which results when n = 1. ui Figure 2.21 shows the stagnation point ow when n = 2. Here the analysis can be done in Cartesian Divergence coordinates free vectorelds // to boundary L F = Az 2 h W (z) = F = 2Az x1 x2 = 2A(x + iy) p0 > p 1 u = 2Ax p1 v = 2Ay u n Figure 2.22 shows the ow towards a wedge when 1 n 2. The velocity along the wedge edge (x axis dS in gure 2.22) is u nt ur = nArn1 cos(n) { = 0} u = ur = nArn1 = Cxn1 x The wedge angle is y 2 n1 u0 = 2 = 2 n n p0 y = 4t p

Figure 2.21: Stagnation point ow when n = 2.

!! !! !! !! !! !! !! !! !! !! !! !! !! !! !! !! !! !! !! !! !! !! !! !! !! !! !! 

y = 4t t y ()u0 z = t t

u u0

                             

L y = 4t h t x1 x2 py > p1 0 ()u p1 z = t0 u t n dS u 2.22: Figurent The ow over a wedge when 1 n 2. x y u0 p0 y = 4t y = 4t t y ()u z = t0 t Figure 2.23: Line source. We now list several other examples of potential ow elds obtained from analytic functions. Example 7. Line source. F (z) = = = m ln z 2 m ln(rei ) 2 m [ln r + i] 2
u u0

u u0

The velocity potential and the streamfunction become = m ln r 2 The complex velocity can be written as = m 2

W (z) = F =

m m i = e 2z 2r

which gives the solution in polar coordinates, see gure 2.23, as ur = m 2r u = 0 We can now calculate the volume ux from the source as
2

Q=
0

m r d = m 2r

Example 8. Line vortex. F (z) = i = ln z 2

[ i ln r] 2

y = 4t t y z = ()u0 t t Figure 2.24: Ideal vortex. The velocity potential and the streamfunction become = 2 = ln r 2 W (z) = F = i i e 2r

y 4t u u0

x y u0 p0

The complex velocity can be written as

which gives the solution in polar coordinates, see gure 2.24, as ur = 0 u = 2r F (z) = m z Example 9. The dipole has the complex potential

Example 10. The potential ow around a cylinder has the complex potential F (z) = U z + U
2 r0 z

Example 11. The potential ow around a cylinder with circulation has the complex potential F (z) = U z + U
2 r0 z i ln z 2 r0

Example 12. Airfoils. The inviscid ow around some airfoils can be calculated with conformal mappings of solutions from cylinder with circulation.

2.4

Boundary layers

For ows with high Reynolds number Re, where the viscous eects are small, most of the ow can be considered inviscid and a simpler set of equations, the Euler equations, can be solved. The Euler equations are obtained if Re in the Navier-Stokes equations. However, the highest derivatives are present in the viscous terms, thus a smaller number of boundary conditions can be satised when solving the Euler equations. In particular, the no ow condition can be satised at a solid wall but not the no slip condition. In order to satisfy the no slip condition we need to add the viscous term. Thus, there will be a layer close to the solid walls where the viscous terms are important even for very high Reynolds number ow. This layer will be very thin and the ow in that region is called boundary layer ow.

z =

} t = 4t 0 y U t L y ()u0 z = t Figure 2.25: Estimatet of the viscous region for an inviscid vorticity free inow in a plane channel.

()u0 t

0 y 4t u u0

L v y, 0 U
}

L x, u

Figure 2.26: Boundary layer prole close to a solid wall. We begin this section by estimating the thickness of the viscous region in channel with inviscid inow (zero vorticity), see gure 2.25. By analogy to the Stokes solution for the instantaneously starting plate, we can estimate the distance the vorticity has diused from the wall toward the channel centerline as ts . Where ts is time of a L uid particle with velocity U has travelled length L, i.e. ts = . Thus we can estimate /L as U 1/2 ts 1 L 1/2 1 0 as Re = = = = L L L U UL Re We will now divide the ow into a central/outer inviscid part and a boundary layer part close to walls, where the no slip condition is fullled. Since we expect this region to be thin for high Reynolds numbers, we expect large velocity gradients near walls. We choose dierent length scales inside boundary layer and in inviscid region as indicated in gure 2.26, as Inviscid : U velocity scale for u, v : L length scale for x, y BL : : : : U L velocity scale for velocity scale for length scale for length scale for u v x y

The pressure scale is assumed to be U 2 for both the inner and the outer region. The unknown velocity scale for the normal velocity is determined using continuity u v + =0 x y
O

U L

which implies that U = L = U L

In this manner both terms in the continuity equation have the same size. Let us use these scalings in the steady momentum equations and disregard small terms. We obtain the following estimates for the size of the terms in the normal momentum equation

y:

v x 1 U U L L u L
2

v y U 1 U L L v L
2

1 p y
U2

L2

2v x2
U L

2v y 2 U 2 L 1 Re

1 Re

where the last line was obtained by dividing all the terms on the second line by U 2 /. When R and /L is small, only pressure term O(1), which implies that p =0 y p = p(x)

Thus, the pressure is constant through the boundary layer, given by the inviscid outer solution. Next we estimate the size of the terms in the streamwise momentum equation as x: u u x + v u y = 1 p x U2 L 1 + 2u x2 + 2u y 2

U2 L 1

U U L 1

U L2 1 Re

U 2 1 Re L
2

The rst three terms are order one and the fourth term negligible when the Reynolds number become large. In order to keep the last term, the only choice that gives a consistent approximation, we have to assume that 1 L Re which is the same estimate as we found for the extent of the boundary layer in the channel inow case. Thus we are left with the parabolic equation in x since the second derivative term in that direction is neglected. We have u u u 1 p 2u +v = + 2 x y x y

In summary, the steady boundary layer equations for two-dimensional ow is 2 u u + v u = 1 p + u x y x y 2 u v + =0 x y

with the initial (IC) and boundary conditions (BC) IC : u (x = x0 , y) = uin (y) BC : u (x, y = 0) v (x, y = 0) u (x, y ) = = =

0 0 ue (x) outer inviscid ow

where the initial condition is at x = x0 and the equations are marched in the downstream direction. There are several things to note about these equations. First, vin cannot be given as its own initial condition. By using continuity one can show that once the u0 is given the momentum equation can be written in the form v + f (y) v = g (y) y

which can be integrated in the y-direction to give the initial normal velocity. Second, the system of equations is a third order system in y and thus only three boundary conditions can be enforced. We have the no ow and the no slip conditions at the wall and the u = ue in the free stream. Thus no boundary condition for v in the free stream can be given.

Figure 2.27: Boundary layer ow over a at plate with zero pressure gradient. Third, ue (x) = uinv (x, y = 0), i.e. the boundary layer solution in the free stream approaches the inviscid solution evaluated at the wall, since the thickness of the boundary layer is zero from the inviscid point of view. Fourth, the pressure gradient term in the momentum equation can be written in terms of the edge velocity ue using the Bernoulli equation. We have 1 p (x) + u2 (x) = const 2 e which implies that 1 dp due = ue dx dx

Blasius ow over at plate

Consider the boundary layer equations for a ow over a at plate with zero pressure gradient, i.e. the outer inviscid ow is ue (x) = u0 . The boundary layer equations become 2 u u + v u = u x y y 2 u v + =0 x y

IC : BC :

u (x = x0 , y) u (x, y = 0) v (x, y = 0) u (x, y )

= = = =

u0 0 0 u0

We introduce the two-dimensional stream function u= , y v= x

which implicitly satises continuity. The boundary layer equations become 2 2 3 = 3 y xy x y 2 y with appropriate initial and boundary conditions. Similarity solution We will now look for a similarity solution. Let u= = u0 f () , y = y (x)

which by integration in the y-direction implies (x, y) = u0 (x) f () We can now evaluate the various terms in the momentum equation. We give two below v= = u0 f + f x x y = u0 f f = u0 (f f )

$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$ "#"#"#"#"#"#"#"#"#"#"#"#"#"#"#"#"#"#"#"#"#"#"#"#"#"#"#"#"#"#"#"#"# "$#"$#"$#"$#"$#"$#"$#"$#"$#"$#"$#"$#"$#"$#"$#"$#"$#"$#"$#"$#"$#"$#"$#"$#"$#"$#"$#"$#"$#"$#"$#"$#"$# #################################"$"$

U y, v x, u L U

u0

u0

L 0 U y, v x, u = yL x/u0 U u0
}

Blasius prole

0.2

0.4

f =

u u0

0.6

0.8

Figure 2.28: Similarity solution giving the Blasius boundary layer prole.

2 = u0 f xy If these and the other corresponding terms are introduced into the momentum equation we have the following equation for f u2 f f 0 which can be simplied to u0 ff = 0 For a similarity solution to be possible, the coecient in front of the f f term has to be constant. This implies f + u0 1 = 2 which implies that (x) = dx = 1 2 dx u0 1 2 1 x = 2 2 u0 f f + u2 (f f ) = u0 2 0

x , where we have chosen the constant of integration such that = 0 u0 when x = 0. We are left with the Blasius equation and boundary conditions f BC : 1 + ff = 0 2

f (0) = f (0) = 0 f () = 1

The numerical solution to the Blasius equation gives the self-similar velocity prole seen in gure 2.28. The boundary layer thickness can be evaluated as the height where the velocity has reached 99% of the free stream value. We have f (99 ) = 0.99 99 = 99
x u0

= 4.9

x.

99 = 4.9

x u0

where it can be seen that the boundary layer thickness grows as Displacement thickness and skin friction coecient

Another measure of the boundary layer thickness is the displacement of the wall needed in order to have the same volume ux of the inviscid ow as for the ow in the boundary layer. With the help of gure 2.29 we can evaluate this as
y0 0 y y

u dy

u0 dy =
0

u0 dy u0

which implies that

z =

Figure 2.29: The displacement thickness = Pressure force

Figure 2.30: Boundary layer separation caused by a pressure force directed upstream, a so called adverse pressure gradient.

If we evaluate this expression for the Blasius boundary layer we have = 1.72

A quantity which is of large importance in practical applications is the skin friction coecient. The skin friction at the wall is w = u y = u0 u2 0.332u2 0 0 f (0) = f (0) = u0 x Rex

where the last term is the expression evaluated for Blasius ow. This can be used to evaluate the skin friction coecient as Cf = w 1 2 2 u0 0.664 = Rex

where the last terms again is the Blasius value.

Boundary layer separation

p A large adverse pressure gradient x > 0 may cause a back ow region close to wall, see gure 2.30. The boundary layer equations cannot be integrated past a point of separation, since they become singular at that point. This can be seen from the following manipulations of the boundary layer equations. If we take the momentum equation

take the normal derivative and use continuity we have u 2u 2u 3u +v 2 = 3 xy y y

If we again take the normal derivative and use continuity we can write this as 1 2 x u y
2

This expression evaluated at the wall, with the use of the boundary conditions, become

02102102102102102102102102102102110210210210210210210210210210210210211021021021021021111111111111111111111102 02 02 02 02 02 02 02 02 02 02 02 02 02 02 02 02 02 02 02 02 02 02 02 02

u f = u0 Blasius prole

x/u0

Separation point

=
0

u u0

dy =

y=0

u u 1 dp 2u +v = + 2 x y dx y

+u

3u u 2 u 3u 4u +v 3 = 4 2 2 xy x y y y

) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) ) (&(&(&(&(&&(&(&(&(&(&(&(&&(&(&(&(&(& &&&&&()&&&&&&&&()&&&&&&()

' ' ' ' ' ' ' ' ' ' ' ' ' ' ' %&%&%&%&&%&%&%&&%&%&%&&%&%&&%&%&%& &&&&%'&&&&%'&&&&%'&&&%'&&&&%'

t L = y x/u0 } u f = u0 0 Blasius prole U y, v x, u L U u0 = y

U u0

x u0

[1 f ()] d
x u0 .

1 2 x

u y

=
y=0

4u y 4

y=0

which can be integrated to yield u y

2

y=0

= 2 x + C
u y

At the point of separation x = xs , i.e. where u y

=0
y=0

this expression is

y=0

= xs x

We can see that the boundary layer equations are not valid beyond point of separation since we have a square root singularity at that point. In addition, the point of separation is the limiting point after which there is back ow close to the wall. This means that there is information propagating upstream, invalidating the downstream parabolic nature of the equations assumed by the downstream integration of the equations. In practical situations it is very important to be able to predict the point of separation, for example when an airfoil stalls and experience a severe loss of lift.

2.5

Turbulent ow

Reynolds average equations

Turbulent ow is inherently time dependent and chaotic. However, in applications one is not usually interested in knowing full the details of this ow, but rather satised with the inuence of the turbulence on the averaged ow. For this purpose we dene an ensemble average as 1 Ui = ui = lim N N
(n) N

ui
n=1

(n)

where each member of the ensemble ui is regarded as an independent realization of the ow. We are now going to derive an equation governing the mean ow Ui . Divide the total ow into an average and a uctuating component ui as ui = ui + ui p=p+p

and introduce into the Navier-Stokes equations. We nd ui ui ui u ui u 1 p 1 p + + uj + uj i + uj + uj i = + t t xj xj xj xj xi xi We take the average of this equation, useing ui = 0 which gives ui ui 1 p + uj = + t xj xi ui xi = 0 ui uj xj u j ui = u i uj = 0
2

ui +

ui

ui

where the average of the continuity equation also has been added. Now, let U i = ui , as above, and drop the . We nd the Reynolds average equations

where ui uj is the Reynolds stress. Thus the eect of the turbulence on the mean is through an additional stress Rij which explicitly written out is u2 [Rij ] = [ui uj ] = uv uw uv v2 vw uw vw w2

Ui 1 P Ui + Uj = + (ij ui uj ) t xj xi xj Ui xi

The diagonal component of this stress is the kinetic energy of the turbulent uctuations k = Rii /2 = ui ui /2 = 1 2 u + v2 + w2 2

Turbulence modelling
One may derive equations for the components of the Reynolds stress tensor. However, they would involve additional new unknowns, and the equations for those would in turn involve even more unknowns. This closure problem implies that if one cannot calculate the complete turbulent ow, but is only interested in the mean, the eect of the Reynolds stress components on in the Reynolds average equations have to be modelled. The simplest consistent model is to introduce a turbulent viscosity as a proportionality constant between the Reynolds stress components and the strain or deformation rate tensor, in analogy with the stress-strain relationship in for Newtonian uids. We have Rij = 2 kij 2T E ij 3

The rst term on the right hand side is introduced to give the correct value of the trace of R ij . Here the deformation rate tensor is calculated based on the mean ow, i.e. E ij = 1 2 Ui Uj 2 Ur ij + xj xi 3 r x

where we have assumed incompressible ow. Introducing this into the Reynolds average equations gives Ui Ui + Uj = t xj xj P 2 1 P + k ij + 2 ( + T ) E ij = + ( + T ) 3 xi
2

Ui

where the last equality assumes that T is constant, an approximation only true for very simple turbulent ow. It is introduced here only to point out the analogy between the molecular viscosity and the turbulent viscosity. The turbulent viscosity T must be modelled and most numerical codes which solved the Navier-Stokes equations for practical applications uses some kind of model for T . For simple shear ows, i.e. velocity elds of the form U (y), the main Reynolds stress component becomes uv = T where, in analogy with kinetic gas theory, T T = uT l Here, uT is a characteristic turbulent velocity scale and l is a characteristic turbulent length scale. In kinetic gas theory l would be the mean free path of the molecules and uT the characteristic velocity of the molecules. U y

u f = u0 Blasius prole x Separation point = Pressure force

x/u0

uT l

Figure 2.31: Prantdls mixing length l is the length for which a uid particle displaced in the normal direction can be expected to retain its horizontal momentum. Zero-equation models The simplest model for uT and l is to model them algebraically. In simple free shear ows like jets, wakes and shear layers we can take T as constant in normal direction or y-direction and to vary in a self-similar manner in x. In wall bounded shear ow we can take l as Prandtls mixing length (1920s). This is a normal distance over which the particle retains its momentum and is depicted in gure 2.31. We can thus evaluate uT and nd T uT = l dU dy T = l 2 dU dy

Dene w as the shear stress at the wall and u = uT = u , as Prandtls estimation of mixing length we get dU uT u = = dy l y

is the von Karman constant, approximately equal to 0.41. Thus the mixing length is proportional to the distant from the wall and the region close to the wall in a turbulent ow has a logarithmic velocity prole, called the log-region. One-equation models Instead of modelling the characteristic turbulent velocity algebraically, the next level of modelling uses the assumption that 1 ui ui 2 and a dierential equation is used to calculate the turbulent kinetic energy. We can derive such an equation by taking the previously derived equation for Ui + ui and subtracting the equation for U , the Reynolds average equation. We nd uT = k= ui ui Ui ui + Uj = uj uj + t xj xj xj xj p ui ij + + u i uj xj

We now take the average of the inner product between the turbulent uctuations u i and the above equation, and nd t 1 ui ui 2 xj 1 ui ui 2 Ui + xj xj puj 1 ui ui uj + ui ui 2 2 xj ui ui xj xj

+ Uj

where is the mean dissipation rate of the turbulent kinetic energy. We have now introduced a number of new unknowns which have to be modelled.

5 5 5 5 5 5 5 35 34434344343443434443 43544435444354443544543
l =y = ui uj

w / as the skin friction velocity. With

U 1 = ln y + C. u

Second, the turbulent dissipation is modelled purely based on dimension using k and l. We have = cD
L2 T .

2 First, we model the Reynolds stress as previously indicated, i.e. ui uj = 3 kij T

Ui xj

Uj xi

3/2

since the dimension of is [ ] = Here cD is a modelling constant. Third, a gradient diusion model is used for the transport terms, i.e. 1 1 T k ui ui uj puj = 2 P rk xj where P rk is the turbulent Prandtl number. For simplicity we write the resulting equation for the turbulent kinetic energy for constant T . We nd Dk i ui uj Uj = ( + T ) 2 k x Dt rate of increase diusion rate generation rate

dissipation rate

Two-equation models The most popular model in use in engineering applications is a two equation model, the so called k- -model. This is based on the above model for k but instead of modelling a partial dierential equation is derived governing the turbulent dissipation. This equation has the form as the equation for k, with a diusion term, a generation term and a dissipation term. Many new model constants need to be introduced in order to close the equation. Once k and is calculated the turbulent viscosity is evaluated as a quotient of the two, such that the dimension is correct. We have T = C where C is a modelling constant. k
2

L T0 U0 Gradient elds wi
p xi

ui Divergence free vectorelds // to boundary L h x1 x2 p0 > p 1 p1 u n dS u nt x 3.1 Finite Volume method on arbitrary grids y u0 Equations with 1st order derivatives: the continuity equation p0 y = 4t In nite volume methods one makes approximations of the dierential equations in integral form. We start u with an equation with only u0 order derivatives, the continuity equation. Recall that the integral from of rst y= this equation can be written 4t t (uk ) dV = uk nk dS = 0 y xk ()u0 V S z = t

Chapter 3

Finite volume methods for incompressible ow

Evaluating the normal vector ni to the curve l for two-dimensional ow, see gure 3.1, we have the t normal with length dl as n dl = ( dy, dx). This implies that L } 0 uk nk dl = (u dy v dx) = 0 U S S y, v We now apply this to an x, u arbitrary two-dimensional grid dened as in gure 3.2. L We make use of the following denitions and approximations U uab ui+ 1 ,j (ui+1,j + ui,j ) /2 = xab = xb xa 2 u0 y =
u f =ab 0= vi+ 1 ,j (vi+1,j + vi,j ) /2 vu 2 Blasius prole y x Separation point = Pressure force x dx y l uT x/u0

yab = yb ya

x dl = ( dx, dy) dy

dx dy n dl = ( dy, dx) Figure 3.1: Normal vector to the curve l. 59

i 1, j + 1 c

i, j + 1

i + 1, j + 1

i 1, j

i, j Aabcd d

i 1, j 1

Figure 3.2: Arbitrary grid. Note that the grid lines need not be parallel to the coordinate directions. with analogous expressions for the quantities on the bc, cd and da faces. We can now evaluate the integral associated with the two-dimensional version of the continuity equation for the surface dened by the points abcd. We have

(u dy v dx)
S

1 ui+ 2 ,j yab vi+ 1 ,j xab + ui,j+ 1 ybc vi,j+ 1 xbc + 2 2 2 1 1 ui 2 ,j ycd vi 1 ,j xcd + ui,j 1 yda vi,j 2 xda 2 2

On a cartesian grid we have the relations xab = xcd = ybc = yda = 0 and ycd = yab , xbc = xda which implies that the approximation of the continuity equation can be written

(u dy v dx)
S

1 ui+ 2 ,j ui 1 ,j yab + vi,j+ 1 + vi,j 1 xda = 2 2 2

1 1 (ui+1,j ui1,j ) yab + (vi,j+1 vi,j1 ) xda 2 2 Note that dividing with yab xda gives a standard central dierence discretization of the continuity equation, i.e. ui+1,j ui1,j vi,j+1 vi,j1 + =0 2xda 2yab

Equations with 2nd derivatives: the Laplace equation

We use the Gauss or Greens theorem to derive the integral from of the two-dimensional Laplace equation. We nd 0=
V

2 dV = xj xj

nj dS = {2D} = xj

If we approximate this integral in the same manner as for the continuity equation we have x x y y x x y y

1 i+ 2 ,j

yab ycd

xab +
1 i+ 2 ,j

xcd +
1 i 2 ,j

1 i 2 ,j

87777777876 6 6 6 6 6 6 6 78787878787876 87777777876 6 6 6 6 6 6 6 78787878787876 87777777876 6 6 6 6 6 6 6 78787878787876 87777777876 6 86 86 86 86 86 86 6 87777777876 6 6 6 6 6 6 6 78787878787876 87777777876 6 86 86 86 86 86 86 6 87777777876 6 6 6 6 6 6 6 78787878787876 87777777876 6 6 6 6 6 6 6 78787878787876 878787878787876 7777777876 6 6 6 6 6 6 6
c b d a i, j 1
S
1 i,j+ 2 1 i,j 2

i + 1, j

i + 1, j 1

dy dx x y

ybc yda

xbc +
1 i,j+ 2

xda
1 i,j 2

First derivatives are evaluated as mean value over adjacent control volumes (areas in the two-dimensional case). We use Gauss theorem over the area dened by the points a b c d , see gure 3.2. For the threedimensional case we have 1 V
V

1 dV = xj V

nj dS
S

which in the two-dimensional case is approximated by x where dy i+1,j ya b + b yb c + i,j yc d + yd a

abcd

,
1 i+ 2 ,j

1 i+ 2 ,j

1 Aa b c d
abcd

( dy, dx)

and the area evaluated as half magnitude of cross products of diagonals, i.e. Aab Aa b c d = and the value a is taken as the average a = 1 (i,j + i+1,j + i,j1 + i+1,j1 ) 4 1 xd b ya c yd b xa c 2

Evaluating the derivatives , along the other cell faces and substituting back we obtain a 9 point x y formula which couples all of the nine points around the point i, j, of the form Ai,j i+1,j+1 Fi,j i,j1 + Bi,j i+1,j + Gi,j i1,j+1 + Ci,j i+1,j1 + Hi,j i1,j + Di,j i,j+1 + Ii,j i1,j1 + Ei,j i,j = 0 +

If we have a total of N = m n interior nodes, we obtain N equations which in matrix form can be written Ax = b Here xT = (11 , 12 , 13 , . . . , mn ) and b contains the boundary conditions. The matrix A is a N N with the terms Aij , Bij , etc. as coecients. Rather than showing the general case in more detail, we instead work out the expression for cartesian grids in two-dimensions. We nd 2 2 + x2 y 2 x x dx dy = dy dx x y y y x
1 i,j 2

y +
1 i 2 ,j

1 i+ 2 ,j+

1 i,j+ 2

We can evaluate the derivatives as x = (i+1,j i,j ) y 1 x y

1 i+ 2 ,j

and so on for the rest of the terms, which implies that dy dx x y (i+1,j 2i,j + i1,j ) y x + (i,j+1 2i,j + i,j1 ) x y

Dividing with the area x y we obtain the usual 5-point Laplace formula, i. e. i+1,j 2i,j + i1,j i,j+1 2i,j + i,j1 + =0 2 x y 2 which of course also can be written in the matrix form shown for the general case above.

a a i 1, j 1 i, j 1 i + 1, j 1 j

ui,j vi,j pi,j

i+1

Figure 3.3: Cartesian nite-volume grid where the velocity components and the pressure are co-located.

3.2

Finite-volume discretizations of 2D NS

Co-located Cartesian grid

We will apply the nite volume discretizations to the Navier-Stokes equations in integral form, which in two-dimensions can be written (u dy v dx) = 0 u dS = t v dS = t u2 dy uv dx + p dy uv dy v 2 dx p dx 1 Re 1 Re u u dy dx x y v v dy dx x y

where we have used the earlier derived result n = ( dy, dx). We start with a co-located grid which means that the nite volume used for both velocity components and the pressure is the same, see gure 3.3. In addition we choose a Cartesian grid in order to simplify the expressions. Recall that the nite volume discretization of the continuity equation becomes 1 1 (ui+1,j ui1,j ) y + (vi,j+1 vi,j1 ) x = 0 2 2 which is equivalent to the nite dierence discretization or ui+1,j ui1,j vi,j+1 vi,j1 + =0 2x 2y

The discretization of the streamwise momentum equation needs both rst and second derivatives. We have seen that the nite volume discretizations of both are equal to standard second order nite dierence approximations. Thus we obtain the following result for the u-component
2 (uv)i,j+1 (uv)i,j1 u2 ui,j pi+1,j pi1,j i+1,j ui1,j + + + t 2x 2y 2x

1 Re

ui+1,j 2ui,j + ui1,j ui,j+1 2ui,j + ui,j1 + x2 y 2

=0

In a similar manner the discretization for the v-component becomes

2 2 (uv)i+1,j (uv)i1,j vi,j+1 vi,j1 pi,j+1 pi,j1 vi,j + + + t 2x 2y 2y

1 Re

vi+1,j 2vi,j + vi1,j vi,j+1 2vi,j + vi,j1 + 2 x y 2

=0

This simple discretization cannot be used in practice without some kind of modication. The reason is that one solution to these equations are in the form of spurious checkerboard modes. Consider the following solution to the discretized equations ui,j = vi,j = (1)i+j g (t) , p = (1)i+j , x = y

ui1/2,j

control volume for divergence condition. control volume for streamwise momentum equation (u) . control volume for normal momentum equation (v) .

Figure 3.4: Cartesian nite-volume grid where the velocity components and the pressure uses staggered grids. which satises the divergence constraint. If it is input into the equation for u and v we nd the expression 8 dg + g=0 dt Re x2 g = e Rex2
8t

Thus we have a spurious solution consisting of checkerboard modes which are damped slowly for velocity and constant in time for the pressure. This mode will corrupt any true numerical solution and yields this discretization unusable as is. These modes are a result of the very weak coupling between nearby nite volumes or grid points, and can be avoided if they are coupled by one sided dierences or if they are damped with some type of articial viscosity.

Staggered cartesian grid

To eliminate the problem with spurious checkerboard modes, we can use a staggered grid as in gure 3.4, where the control volumes for the streamwise, spanwise and pressure are dierent. The control volume for the continuity equation is centered abound the pressure point and the discretization becomes
1 ui+ 2 ,j ui 1 ,j y + vi,j+ 1 vi,j 1 x = 0 2 2 2 1 ui+ 1 ,j ui 2 ,j 2 1 vi,j+ 2 vi,j 1 2

or

The control volume for the streamwise velocity is centered around the streamwise velocity point and the discretization becomes
2 (uv)i+1/2,j+1/2 (uv)i+1/2,j1/2 u2 ui+1/2,j pi+1,j pi,j i+1,j ui,j + + + t x y x

1 ui+3/2,j 2ui+1/2,j + ui1/2,j 1 ui+1/2,j+1 2ui+1/2,j + ui+1/2,j1 =0 2 Re x Re y 2

@9@9@@9@9@9@@ A A A A A 9A@@@9A@@@@9A@9A @9@9@@9@9@9@@ A A A A A 9A@@@9A@@@@9A@9A @9@9@@9@9@9@@ A A A A A 9A@@@9A@@@@9A@9A 9A 9A 9A 9A 9A 9A @@@9A@@@@9A@9A A A A A A A @ G G G FG G 9@9@9@@9@9@9@@ @F@F@F@@@ @@@9A@@@@9A@9A FG@@@@@FG@FG F@ A@@@9A@@@9A@9A@9A FG@@@@@FG@FG A A A A @ G G G FG G 9@9@9@@9@9@@@ @F@F@F@@@ F@ A A A @ G G G FG G 9@9@9@@9@9@@@ @F@F@F@@@ @A@@9A@A@@9A@9A@9A FG@@@@@FG@FG F@ 9A 9A 9 9A 9 9 @ FG FG FG BCFG @@A@9A@@A@A@9A@9A @C@C@C@@@ FG @@B@B@B@@@ BC@BCFG@@@@@BCFG@BCFG BC@BC@@@@@BC@BC BC BC BC BC @@B@B@B@@@ C C C BC BC@BC@@@@@BC@BC @@B@B@B@@@ C C C DEBC BC@@E@E@E@@@ @@D@D@D@@@ DE@DEBC@@@@@DEBC@DEBC @@D@D@D@@@ E E E DE DE@DE@@@@@DE@DE @@D@D@D@@@ E E E DE DE@DE@@@@@DE@DE DE DE DE DE DE @DE@@@@@DE@DE
vi,j+1/2 pi,j vi,j1/2
1 2

QPRQPR QPQPRR QPRQPR RR PP

I H II H HIH

T S TST S ST

i, j a 1 i 1, j a 1 i + 1, j a 1 i, j 1 j i 1, j 1 i + 1, j 1 i i, j 1 j i+ i + 1, j 1 i ui,j j i+1 vi,j i ui,j pi,j i+1 v pi,j i,j ui,j pi,j vi,j+1/2 v pi,j i,j vi,j1/2 pi,j vi,j+1/2 ui1/2,j pi,j vi,j1/2 i 1 vi,j+1/2 2 ui1/2,j vi,j1/2 1 i i 1 i+ 2 ui1/2,j 2 i i+2 1 i 1 i+ 1 3 2 i+ 2 i +1 1 ji + 1 i+ 3 2 2 i+ 2 j +1 1 ji + 3 1 j2 i+ 2 2 j j 1 1 j+1 j2 2 j j 1 j1 2 ence condition. j 1 ence condition. m equation (u) .

j i i+1 ui,j vi,j pi,j

j+

1 2

j j
1 2

j 1

i+

1 2

i+1

i+

3 2

=0

j1 2j j1 2

Inow A

v1, 3 2 1 B 0 0 1 2 i p1,1
3 u 2 ,1

Figure 3.5: Staggered grid near the boundary.

2 where the u2 i+1,j , ui,j , (uv)i+1/2,j+1/2 , (uv)i+1/2,j1/2 terms need to be interpolated from the points where the corresponding velocities are dened. The control volume for the normal velocity is centered around the normal velocity point and the discretization becomes 2 2 (uv)i+1/2,j+1/2 (uv)i1/2,j+1/2 vi,j+1 vi,j vi,j+1/2 pi,j+1 pi,j + + + t x y y

1 vi+1,j+1/2 2vi,j+1/2 + vi1,j+1/2 1 vi,j+3/2 2vi,j+1/2 + vi,j1/2 =0 2 Re x Re y 2

2 where the (uv)i+1/2,j+1/2 , (uv)i1/2,j+1/2 , u2 i,j+1 , ui,j terms need to be interpolated from the points where the corresponding velocities are dened. For this discretization no checkerboard modes possible, since there is no decoupling in the divergence constraint, the pressure or the convective terms.

Boundary conditions
To close the system we need to augment the discretized equation with discrete versions of the boundary conditions. See gure 3.5 for a denition of the points near a boundary. The boundary conditions on the solid wall, BC in gure 3.5, consists of the no ow and the no slip conditions, i.e. u = v = 0. The normal points are located on the boundary so that the no ow condition simply become
1 1 v1, 2 = v2, 2 = = 0

3 3 The evaluation of the u-equation at the point ( 2 , 1), for example, requires u 2 ,0 . We can nd this value as follows

3 0 = u2,1 = 2

The boundary conditions on the inow, AB in gure 3.5, consist given values of u and v, i.e. so called Dirichlet conditions. The streamwise points are located on the boundary so that the boundary condition is simply consist of the known values
1 1 u 2 ,1 , u 2 ,2 , . . .

The evaluation of the v-equation at the point (1, 3 ), for example, requires v0, 3 . We can nd this value 2 2 as follows
1 v2,3 = 2

1 v 3 + v0, 3 2 2 1, 2

V V V V V V V V V V V V V V V V V U&U&&U&U&U&U&U&U&U&&U&U&U&U&U&U&&U&U& &&UV&&&&&&&&UV&&&&&&&UV&&&UV
C 1 3 3 u 2 ,1 + u 2 ,0 2

Solid wall

3 3 u 2 ,0 = u 2 ,1

3 1 v0, 2 = 2v 2 , 3 v1, 3 2 2

v If AB in gure 3.5 is an outow boundary, the boundary conditions may consist of u = x = 0, i.e. x so called Neumann conditions. These conditions should be evaluated at i = 1 . For the streamwise and 2 normal velocity we nd
3 u 1 ,2 = u 2 ,2 , 2 3 v0, 3 = v1, 2 2

Note that no reference is made to the pressure points outside the domain, so that in this formulation no boundary conditions for p is needed so far. However, depending on the particular discretization of the time derivative and other the details of the solution method one may need to augment the the above with boundary conditions for the pressure. If this is the case particular care has to be taken so that those conditions do not upset the divergence free condition.

3.3

Summary of the equations

The discretized equations derived for the staggered grid can be written in the following form ui+ 1 ,j pi+1,j pi,j 2 1 + Ai+ 2 ,j + =0 t x vi,j+ 1 pi,j+1 pi,j 2 1 + Bi,j+ 2 + =0 t y D =0
i,j

where the expression for the Ai+ 1 ,j can be written 2

1 Ai+ 2 ,j =

2 (uv)i+ 1 ,j+ 1 (uv)i+ 1 ,j 1 u2 i+1,j ui,j 2 2 2 2 + x y 1 1 ui+ 1 ,j+1 2ui+ 2 ,j + ui+ 1 ,j1 1 ui+ 3 ,j 2ui+ 1 ,j + ui 1 ,j 2 2 2 2 2 Re x2 Re y 2

= {expand and interpolate using grid values} = 1 1 2 1 1 1 1 v (u 3 ui 2 ,j ) + (v + ui+ 1 ,j i,j+ 1 vi+1,j 1 vi,j 1 ) + 2 2 2 2 2x i+ 2 ,j 4y i+1,j+ 2 Re x2 y 2 1 1 1 1 1 + u 3 ui+ 3 ,j + u 1 ui 2 ,j 2 4x i+ 2 ,j Rex2 4x i 2 ,j Rex2 1 1 1 1 1 + v 1 + v + (v ui+ 1 ,j+1 + (v ui+ 1 ,j1 i,j+ 1 ) i,j 1 ) 2 2 2 2 4y i+1,j+ 2 Rey 2 4y i+1,j 2 Rey 2
1 = a(u, v)i+ 2 ,j ui+ 1 ,j + 2

a(u, v)nb unb

nb

The last line summarizes the expressions, where the sum over nb indicates a sum over the nearby nodes. The expression for Bi,j+ 1 can in a similar way be written 2
1 Bi,j+ 2 =

x 1 vi+1,j+ 1 2vi,j+ 1 2 2 R x2

(uv)i+ 1 ,j+ 1 (uv)i 1 ,j+ 1

2 2 2 2

2 2 vi,j+1 vi,j y + vi1,j+ 1 1 vi,j+ 3 2vi,j+ 1 + vi,j 1 2 2 2 2 R y 2

= {expand and interpolate using grid values} = b(u, v)i,j+ 1 vi,j+ 1 + 2 2

nb

b(u, v)nb vnb

where the expressions for b(u, v)i,j+ 1 and b(u, v)nb are analogous to a(u, v)i+ 1 ,j and a(u, v)nb , and 2 2 therefore not explicitly written out. The expression originating from the continuity equation is Di,j = ui+ 1 ,j ui 1 ,j 2 2 x +
1 vi,j+ 2 vi,j 1 2

Here we we have used the denitions u v p A(u, v) B(u, v) Gx Gy Dx Dy f -

These equations, including the boundary conditions, can be assembled into matrix form. We have u A(u, v) 0 Gx u fu d v + 0 B(u, v) Gy v = fv dt 0 Dx Dy 0 p 0 vector of unknown streamwise velocities vector of unknown normal velocities vector of pressure unknowns non-linear algebraic operator from advective and viscous terms of u-eq. non-linear algebraic operator from advective and viscous terms of v-eq. linear algebraic operator from stremwise pressure gradient linear algebraic operator from normal pressure gradient linear algebraic operator from x-part of divergence constraint linear algebraic operator from y-part of divergence constraint vector of source terms from BC

With obvious changes in notation, we can write the system of equations in an even more compact form as d dt u 0 + N (u) G D 0 u p = f 0

If this is compared to the nite element discretization in the next chapter, it is seen that the form of the discretized matrix equations are the same.

3.4

Time dependent ows

The projection method

For time-dependent ow a common way to solve the discrete equations is the projection or pressure correction method. Sometimes this method and its generalizations are also called splitting methods. We illustrate the method using the system written in the compact notation dened above. We start to discretize the time derivative with a rst order explicit method n+1 un u + N (un ) un + G pn+1 = f (tn ) t Dun+1 = 0

Then we make a prediction of the velocity eld at the next time step u not satisfying continuity, and then a correction involving p such that Dun+1 = 0. We have n u u + N (un ) un = f (tn ) t n+1 u u + G pn+1 = 0 t Dun+1 = 0 Next, apply the divergence operator D to the correction step DG pn+1 = 1 D u t Here DG represents divgrad = Laplacian, i.e. discrete version of pressure Poisson equation. The velocity at the next time level thus becomes

un+1 = u tG pn+1 Note here that un+1 is projection of u on divergence free space. This is a discrete version of the projection of a function on a divergence free space discussed in the section about the role of the pressure in incompressible ow. In order to be able to discuss the boundary condition on the pressure Poisson equation we write the prediction and correction step more explicitly. Using the staggered grid discretization the prediction step can be written n ui+ 1 ,j ui+ 1 ,j 2 2 + An 1 ,j = 0 i+ 2 t n 1 vi,j+ 1 vi,j+ 2 n 2 + Bi,j+ 1 = 0 2 t n+1 ui+ 1 ,j u 1 ,j pn+1 pn+1 i+ 2 i+1,j i,j 2 + =0 t x n+1 n+1 n+1 vi,j+ 1 vi,j+ 1 pi,j+1 pi,j 2 2 + =0 t y
2

and the projection step as

n+1 By solving for un+1,j and vi,j+ 1 above and using this, and the corresponding expressions for un+1,j and i+ 1 i 1 n+1 vi,j 1 , in the expression for the divergence constraint Di,j we nd the following explicit expression for the 2 discrete pressure Poisson equation pn+1 2pn+1 + pn+1 pn+1 2pn+1 + pn+1 Di,j i,j+1 i,j i,j1 i+1,j i,j i1,j + = x2 y 2 t where Di,j is Di,j evaluated using the discrete velocities at the prediction step, u and v . When the discrete pressure Poisson equation is solved we need values of the unknowns which are outside of the domain, i.e. we need additional boundary conditions. If we take the vector equation for the correction or projection step and project it normal to the solid boundary in gure 3.5 we have
2 2

pn+1 pn+1 1 2,1 2,0 = v n+1 v 1 y t 2, 2

where v = v2, 1 is an unknown value of the velocity in the prediction step at the boundary. In general 2 it is important to choose this boundary condition such that discrete velocity after the correction step is divergence at the boundary, see the discussion in the end of the section on the role of the pressure. However, for this particular discretization it turns out that we can choose v arbitrarily since it completely decouples from the rest of the discrete problem. To see this we write the pressure Poisson equation, centered abound the point (2,1), close to the boundary. We nd

pn+1 2pn+1 + pn+1 pn+1 2pn+1 + pn+1 1 3,1 2,1 1,1 2,2 2,1 2,0 + = 2 2 x y t

u ,1 u ,1 5 3
2 2

v2, 3 v
2

if we substitute the value of pn+1 pn+1 from the boundary condition obtained for the pressure into the 2,1 2,0 above equation we can se that v cancels in the right hand side of the pressure Poisson equation. Since the discretization of the time derivative is explicit there is no other place where the values of v or u on the boundaries enter. Thus we nd that we can choose the value of v arbitrarily. In particular, we can choose n+1 v = v2, 1 , in which case we obtain a Neumann boundary condition for the pressure. Note, however, that 2 this is a numerical artifact and does not imply that the real pressure gradient is zero at the boundary. A number of other splitting methods have been devised for schemes with higher order accuracy for the time discretization. In general it is preferred to make the time discretization rst for the full Navier-Stokes equations and afterward split the discrete equation into one predictor and one corrector step. See Peyret and Taylor for details.

Time step restriction

For stability it is necessary that prediction step is stable, numerical evidence indicates that this is sucient. In the predictor case the momentum equations decouple if we linearize them abound the solution U 0 , V0 . It thus suces to consider the equation u u u 1 + U0 + V0 = t x y Re For simplicity we will here consider one-dimensional version u 1 2u u + U0 = t x Re x2 which in discretized form can be written un+1 = un + t U0 j j un u n 1 un 2un + un j+1 j1 j+1 j j1 + 2x Re x2
2

where we have chosen to call the discrete points where the u-velocity is evaluate for x j . We introduce un = Qn eixj j into the discrete equation, together with the denitions U0 t 2t , = , = x x Re x2 This implies that the amplication factor Q can be written = Q = 1 i sin () 2 sin2 We nd that the absolute value of Q satises |Q| = 1 4 2 2 s2 + 4 2 s2 s = 1 4s where s = sin2
2

2 2 s 2 + 1

. This implies that (s) = 2 2 s + 2 0 s : 0 s 1

(s) is linear function of s and thus the the limiting condition on (s) is given by its values at the two end points of the s-interval. We nd (0) = 2 0, which implies that 2 1 (1) = 2 0

For the two dimensional version of the predictor step Peyret and Taylor [8] nd the following stability limits 1 U 2 + V 2 t Re 1 0 4 0 4t 1 Rex2 Note that this is a stronger condition than in the one-dimensional case.

substituting back the denitions of and we nd the following stability conditions 1 U 2 tRe 1 2 0 2t 1 Rex2

3.5

General iteration methods for steady ows

Distributive iteration
For steady ows we can still use the methods introduced for unsteady ows, and iterate the solution in time to obtain a steady state. However, in many cases it is more eective to use other iteration methods. We will here introduce a general distributive iteration method and apply it to the discretized steady Navier-Stokes equations. For the steady case the discretized equations can be written in the form N (u) G D 0 u p = f 0

which is of the form A y = b. Now, let y = B y which implies that AB y = b and use the iteration scheme on this modied equation. We divide the matrix AB in the following manner AB = M T and use this to dene the iteration scheme M B 1 y k+1 = T B 1 y k + b = M B 1 y k + b A y k which can be simplied to y k+1 = y k + BM 1 b A y k
rk

M y = Ty + b

where rk is the residual error in the k:th step of the iteration procedure. If B = I we obtain a regular iteration method where residual r k is used to update y k by inverting M , a simplied part of A, as in the Gauss-Seidell method for example. For this method we have Ax = b A=LU (L U ) x = b L xn+1 = U xn + b

Application to the steady Navier-Stokes equations

When the distributive iteration method is applied to the steady Naiver-Stokes we have A= N (u) G D 0

we can obtain AB in the following block triangular form AB = if the matrix B has the form B= I 0 N (u)1 G I I 0 N (u)1 G I N (u) D 0 DN (u)1 G

where N(u)1 is a simple approximation of the inverse of the discrete Navier-Stokes operator. We now split AB = M T where M= Q D 0 R

and where Q is an approximation of N (u) and R is an approximation of DN 1 G. Note that R is a discrete Laplace like operator.

The SIMPLE (Semi-Implicit Method for Pressure Linked Equations) by Pantankar and Spalding (1972) can be thought of as a distributive iteration method. Wesseling [10] indicates that by choosing N (u)1 = [diagN (uk )]1

Q = N (uk ) R = D[diagN (uk )]1 G we have a method which is essentially the original version of the SIMPLE method. Thus we have the following iteration steps 1. First, we calculate the residual b A yk = f 0 N (uk ) G D 0 uk pk =
k r1 k r2

2. Second, we calculate M 1 rk which implies

k Q = r1 u

k = Q1 r1 u

k R p = r2 D u

3. Third, we have the distribution step BM 1 rk u p =B u p

which becomes = N 1 G p u p

4. Fourth, there is an under relaxation step, where the velocities and the pressure is updated uk+1 = uk + u u pk+1 = pk + p p The under relaxation is needed for the method to converge, thus u and p is usually substantially lower that one.

Chapter 4

Finite element methods for incompressible ow

u=0 on 0 , (4.1b) i i+ 1 2 u =0 on 1 . (4.1c) i+1 n 3 i+ 2 Uj +1 2 The coecient > 0 is the thermal jdiusivity. Incompressibility yields that xj = 0. We assume that j 1 is a solid wall, so that nj Uj = 0 on 1 . This simplies the analysis but is not essential. The Dirichlet j1 2 isothermal condition, whereas the Neumann condition u = 0 boundary condition u = 0 on 0 is called an n j1 on 1 is an adiabatic condition. A Deriving the integral form that is used to form the nite-element discretization requires some vector B calculus. The product rule of dierentiation yields C u3 2 ,1 u v u v1, 3 v = + v 2 u. (4.2) 2 xi xi xi xi p1,1 0 n 1 0 2 i j Inow Solid wall 1

dx dy dx dl = ( dx, dy) n dl = ( dy, dx) i 1, j + 1 i, j + 1 i + 1, j + 1 c c b b i 1, j i + 1, j i, j Aabcd d d a The nite-element method (FEM) approximates an integral form of the governing equations. However, a the integral formcalled the variational formis usually dierent from the one used for the nite-volume i 1, j 1 method. Moreover, dierent integral forms may be used for the same equation depending on circumstances such as boundary conditions. We i, j 1 introduce FEM rst for a scalar advectiondiusion problem before i + 1, j discussing the NavierStokes equations. 1 j i i+1 4.1 FEM for an advectiondiusion problem ui,j vi,j We consider the steady ow of an incompressible uid in a bounded domain , and we assume that its pi,j velocity eld Uj is known (from a previous numerical solution, for instance). We assume that the boundary pi,j of the domain can be decomposed into two portions 0 and 1 , as in gure 4.1 for instance. We are vi,j+1/2 interested to compute the temperature eld u in the domain, modeled by the advectiondiusion problem vi,j1/2 u ui1/2,j in , (4.1a) Uj 1 2 u = f i xj 2

Figure 4.1: A typical domain associated with the advectiondiusion problem (4.1). 71

Recall the divergence theorem

wi d = xi

ni wi d.

(4.3)

Integrating expression (4.2) and applying the divergence theorem on the left side yields v

u d = n

v u d + xi xi

u d.

(4.4)

(Note that u u = ni .) n xi Let v be any smooth function such that v = 0 on 0 . Multiplying equation (4.1a) with v, integrating over , and utilizing expression (4.4) yields u vUj d xj u vUj d + xj v u d xj xj 0  u d, v n

f v d =

u d =

where the last term vanishes on 0 due to the requirements on v, and on 1 due to boundary condition (4.1c) We have thus shown that a smooth (twice continuously dierentiable) solution to problem (4.1) (also called a classical solution) satises the varational form vUj

u d + xj

v u d = xj xj

f v d

(4.5)

for each smooth function v vanishing on 0 We will now forget about the dierential equation (4.1), and directly consider the variational form (4.5). For this, we need to introduce the function space v v V = v: d < +, v = 0 on 0 . xi xi

The space V is a linear space, that is, u, v V u + v V , R. The requirement that the gradient-square of each function in V is bounded should be intepreted as a requirement that the energy is bounded. The variational problem will thus be Find u V such that u v u vUj d + d = xj xj xj

f v d

v V.

(4.6)

A solution to variational problem (4.6) is called a weak solution to advectiondiusion problem (4.1). The derivation leading up to expression (4.5) shows that a classical solution is a weak solution. However, a weak solution may not be a classical solution since functions in V may fail to be twice continuously dierentiable. Note that the boundary conditions are incorporated into the variational problem. The Dirichlet boundary condition on 0 is explicitly included in the denition of V and is therefore referred to as an essential boundary condition. The Neumann condition on 1 is dened implicitly through the variational form and is therefore called a natural boundary condition.

4.1.1

Finite element approximation

Let us triangulate the domain by dividing it up into triangles, as in gure 4.2. Let M be the number of triangles and N the number of nodes in 1 , that is, the nodes on 0 are excluded in the count. Let h be the largest side of any triangle. Let Vh be the space of all functions that are - continuous on ,

p1,1 0 1 2 i j Inow Solid wall n 0 1

3 u 2 ,1 3 v1, 2 p1,1 0 1 2 i j Inow Solid wall n 0 1

Figure 4.2: A triangulation. Here, the number N are Figure 4.3: The tent or hat basis function associated with continuous, piece-wise linear functions. the black nodes. - linear on each triangle, - vanishing on 0 . Note that Vh V . We get a nite-element approximation of our advection-diusion problem by simply replacing V by Vh in variational problem (4.6), that is: Find uh Vh such that vh uh uh v h Uj d + d = xj xj xj

f vh d

vh Vh .

(4.7)

Restricting the function space in a variational form to a subspace is called a Galerkin approximation. The nite-element method is a Galerkin approximation in which the subspace is given by piecewise polynomials.

4.1.2

The algebraic problem. Assembly.

Once the value of a function uh in Vh is known at all node points, it is easy to reconstruct it by linear interpolation. This interpolation can be written as a sum of the tent basis functions (n) (gure 4.3), which are functions in Vh that are zero at each node point except node n, where it is unity. Thus,
N

uh (xi ) =
n=1

u(n) (n) (xi ) ,

(4.8)

where u(n) denotes the value of uh at node n. Note that expansion (4.8) forces uh = 0 on 0 since it excludes the nodes on 0 in the sum. Substituting expansion (4.8) into FEM approximation (4.7) yields
N

u(n)
n=1

v h Uj

(n) d + u(n) xj n=1

vh (n) d = xj xj

f vh d

v h Vh .

(4.9)

Since each basis function is in Vh , we may choose vh = (m) Vh for m = 1, . . . , N in equation (4.9), which then becomes
N

u(n)
n=1

(m) Uj

(n) d + u(n) xj n=1

(m) (n) d = xj xj

f (m) d, m = 1, . . . , N .

(4.10)

This is a linear system system Au = b, where A = C + K and Kmn =

(m) (n) d, xj xj

Cmn =

(m) Uj

(n) d, xj

(4.11)

h Figure 4.4: An example mesh that gives a particular simple stiness matrix K

u(1) . u= . . u(n)

The union of all triangles constitutes the domain and that the triangles do not overlap. Thus, any integral over can be written as a sum of integrals over each triangle, (m) (n) d = xj xj p=1
M

Kmn =

Note that Kmn vanishes as soon as m or n is not a corner node of triangle p. Thus Kmn contributes to the sum only when m and n both are corner nodes of triangle p. Matrix K can thus be computed by looping (p) all M triangles, calculating the element contribution Kmn for m, n being the three nodes of triangle p and adding these 9 contribution to matrix K. Note that the derivatives of (n) are constant on each triangle, (p) so Kmn is easy to compute exactly. Matrix C and vector b can also be computed by element assembly, but numerical quadrature is usually needed, except if Uj and f happen to be particularly easy functions (such as constants).

(p)

4.1.3

An example

As an example, we will compute the elements of K, the stiness matrix (a term borrowed from structural mechanics), for the particular case of the structured mesh depicted in gure 4.4. Let M be the number of internal nodes in the horizontal as well as the vertical direction (the solid nodes in gure 4.4.) The width of each panel is then h = 1/(M + 1), and the area of each triangle is |Tk | = h2 /2. The components of K are Kmn =

Note that Kmn vanishes for most m and n. For instance, Km,m+2 = 0, since basis functions (m) and (m+2) do not overlap (gure 4.4). In fact, Kmn = 0 only when m and n are associated with nodes that are nearest neighbors. To compute matrix element (4.12), we need expressions for the gradients of basis functions (m) and (m+1) . Since the functions are piecewise linear, the gradient is constant on each triangles and may thus easily be computed by nite-dierences in the coordinate direction (gure 4.5). By the expressions in

X`XXX`X`XX XXWXXXWXWX a `a a a WY Y WY Y Y `aXXX`aXXX`aX`a XWYXXXWYXXXWY a`aX`a `XaX`a`a`a `a `a WYWYWY WY WYWYWY WY WY X`XX`aXaXaX`aX`a XWYXYXXWYXYXYXWY XaXXX`X`XX XXWXXXWXWX `aX`XX`aXaXaX`aX`a XWYXYXXWYXYXYXWY XaXXX`X`XX XXWXXXWXWX `aX`XX`aXXX`aX`a XWYXXXWYXXXWY X`XX`aX`X`X`aX`a XWYXWXXWYXWXWXWY aXXXaXaXX XXYXXXYXYX a`a a a WYWYWY Y WYWYWY Y Y `XX`X`XX XXWXXXWXWX a`aXaXXXaXaXX XXYXXXYXYX `X`XX`aX`X`X`aX`a XWYXWXXWYXWXWXWY aXaXXXaXaXX XXYXXXYXYX `X`XXX`X`XX XXWXXXWXWX XXX`a`aXXX`a`aX`a`a XWYWYXXXWYWYXXXWYWY `a`a WY WY
m m+2 1 M h f (1) d

i j Inow Solid wall n 0 1

M2

b=

. . . f (N ) d

Tp

(m) (n) (p) d = Kmn . xj xj p=1

(p)

(m) (n) d xj xj

(4.12)

m 2 m+2 T3 i on T1 , 1 j on T2 , h T4 Inow T2 on T3 , Solid wall (m) on T4 , n T1 T5 on T5 , 0 1 on T6 , T6 on all other triangles. 2 M (m) M Figure 4.5: The gradient of test function for m being the (black) middle node. m m+2 1 1 1 (m) h, h on T1 , (m) 1 h = ,0 on T2 , T2 (m+1) h 0 on all other triangles T1 1 ,0 on T1 , h (m+1) = 1 1 on T2 . h, h 1 h, 0 1 ,1 h h 0, 1 h 1 = h, 0 1 1 , h h 0, 1 h 0 Figure 4.6: The support for basis functions m and m+1 are marked with vertical and horizontal stripes respectively. gure 4.5, we have Kmm =

(m) (m) d = xj xj

k=1T k

(m) x

1 1 1 1 1 1 |T1 | + 2 2 |T2 | + 2 |T3 | + 4 |T4 | + 2 2 |T5 | + 2 |T6 | 2 h h h h h h 1 h2 =8 2 =4 h 2 = The common support for basis functions m and m+1 is only the two triangles marked in gure 4.6; that is, it is only on these two triangles that both functions are nonzero at the same time. Utilizing the gradient expressions given in gure 4.6 yields
2

Km,m+1 =

(m)

(m+1) d =
k=1T
k

(m)

(m+1) d =

and in the same manner, Altogether, K becomes the block-tridiagonal M 2 -by-M 2 matrix T I I T I .. .. .. K = . . . I T I I T Km,m1 = Km,m+M = Km,mM = 1.

where I is the M -byM identity matrix, and T the M -byM matrix 4 1 1 4 1 .. .. .. T = . . . 1 4 1 1 4 Thus, a typical row in the matrixvector product Ku becomes

4um um+1 um1 um+M umM ,

b b f@f cb f b b f b b b b f b b b b b b b bc e@e@@f cccccccccccccccccbb @f@@e@f@f@@f@ ef@e@ef@f@e@e@ef@e@ef d b e@e@@e@ @f@ef@f@ccccccccccccccccccccccccccccdbdbdbdbdbdbdbdbdbdbdbdbdbdbdbe@f@f@ef@f@ef b cb fef@e@@e@e@e@@e@ e@f@ef@@f@f@ef@f@ef d f@e@@f cbdcc e@e@@e@ e@f@@e@f@f@@f@ @e@ef@f@ccccccccccccccccccccccccccccdbdbdbdbdbdbdbdbdbdbdbdbdbdbdc e@e@ef@e@ef b cb @f@@e@f@f@@f@ ef@e@ef@f@e@e@ef@e@ef @e@@f@e@e@@e@ f@@e@f@f@@f@ ef@f@ef@f@ccccccccccccccccccccccccccccdbdbdbdbdbdbdbdbdbdbdbdbdbdbdbe@f@f@ef@f@ef b cb f@e@@e@e@e@@e@ e@@efef@@ccccccccccccccccccccccccccccdbdbdbdbdbdbdbdbdbdbdbdbdbdbdc @@efef@@efef d c b cb fef@ef e@f@@ef cbd ef ef ef @e@ef@@f@f@ef@f@ef @f@@f cc e@e@@e@ ef@e@ef@f@ccccccccccccccccccccccccccccdbdbdbdbdbdbdbdbdbdbdbdbdbdbdbe@f@f@ef@f@ef b cb @f@@e@e@e@@e@ d ef@e@ef@f@f@f@ef@f@ef @f@@e@e@e@@e@ ef@e@ef@@ccccccccccccccccccccccccccccdbdbdbdbdbdbdbdbdbdbdbdbdbdbdc @@ef@@ef b cb @e@ef@e@f@e@ef@f@ef f@@f@e@f@@e@ df e@f c c e@ cbe@f@e@@f@ df d d f d fef@f@@f@dcb e@dbc b e@dbdf@b ef@dbd@b e@ e@e@ef@e@ccccccccccccccccccccccccccccdbdbdbdbdbdbdbc ef d cd d ef@ef@@ef@d ef@d ef@@d ef@ d d d d d @@efef@@c@c@ced fef@c@ced fef cccccd
2

(m) y

1 1 2 h2 |T1 | 2 |T2 | = 2 = 1, 2 h h h 2

. (4.13)

that is, after dividing expression (4.13) with h2 , we recognize the classical ve-point stencil for the negative Laplacian 2 . Thus our nite-element discretization is for this particular mesh equivalent to a nite-volume or nitedierence discretization. However, for a general unstructured mesh, the nite-element discretization will not give rise to any obvious nite-dierence stencil.

4.1.4

Matrix properties and solvability

Matrix A is sparse: the elements of martix A are mostly zeros. At row m, matrix element A mn = 0 only in the columns where n represents a nearest neighbor to node m. The number of nonzero elements on each row does not grow (on average) when the mesh is rened (since the number of nearest neighbors in a mesh does not grow). Thus, the matrices become sparser and sparser as the mesh is rened. A common technique in nite-element codes is therefore to use sparse representation, that is, to store only the nonzero elements and pointers to matrix locations. It is immediate from expression (4.11) that Kmn = Knm , that is, K is a symmetric matrix (K T = K). Integration by parts also shows that matric C is skew symmetric (C T = C). Theorem 1. K is positive denite, that is, v T Kv > 0 for each v = 0. Proof. Let vh Vh . Then vh =
N N N n=1

v (N ) (n) . Denote v = (v (1) , . . . , v (N ) )T .

v T Kv =
m=1 n=1 N

v (n)

(n) (m) d v (m) xi xi v (m) (n) d = xi n=1

N

v (m) (m) m=1 xi

vh xi

vh vh d 0, xi xi

with equality if and only if

= 0, that is, if vh = Const. But since vh |0 = 0, vh 0.

From Theorem 1 also follows that A is positive denite. To see this, note that the skew-symmetry of C yields that T v T Cv = v T Cv = v T C T v = v T Cv that is, v T Cv = 0 for each v. Thus v T Av = v T (K + C) v = v T Kv + v T Cv = v T Kv > 0 for each v = 0, that is, A is positive denite. However, A is not symmetric whenever U j = 0. Recall that a linear system Au = b has a unique solution for each b if the matrix is nonsingular. A positive-denite matrix is a particular example of a nonsingular matrix. We thus conclude that the linear system resulting from the nite-element discretization (4.7) has a unique solution.

4.1.5

Stability and accuracy

We have dened the nite-element approximation and showed that the nite-element approximation yields a linear system Au = b that has a unique solution. Now the question is whether the approximation is any good. We will study the stability and accuracy of the numerical solution. FEM usually uses integral norms for analysis of stability and accuracy. We start by reviewing some mathematical concepts that will be used in the analysis. Basic denitions and inequalities The most basic integral norm is
1/2

L2 ()

v 2 d

For the current problem, a more fundamental quantity is the energy norm
1/2

v and associated inner product

| v|2 d

v v d xi xi

1/2

(u, v)V =

v d =

u v d. xi xi

(4.14)

The CauchySchwarz inequality for vectors a = (a1 , . . . , an ) is |a b| |a||b| = |a a|1/2 |b b|1/2 , whereas for functions it reads
1/2 1/2

f g d

f 2 d

g 2 d

(4.15)

which can be denoted (f, g)L2 () f

1/2 L2 ()

L2 () .

The CauchySchwarz inequality (4.15) implies that, for each nonzero, square-integrable g, f 2 d

f g d
1/2

g 2 d

Choosing g = f yields equality in above expression. Thus

1/2

L2 ()

f 2 d

= max
g=0

f g d
1/2

g 2 d

= max
g=0

f g d
L2 ()

The rightmost expression is a variational characterization of the L2 () norm, that is, it expresses the norm in terms of something that is optimized. By inserting derivatives in the nominator and/or denominator of the variational characterization, various exotic norms can be dened. A norm, dierent from the L 2 () norm and of importance for the NavierStokes equations is p p
exotic

= max
wi

wi xi
2

d
1/2

(4.16)

| wi | d

This norm will be of crucial importance when we analyze the stability of FEM form the NavierStokes equations. The Poincar inequality relates the energy and L2 () norms: e Theorem 2 (Poincar inequality). There is a C > 0 such that e v 2 d C v for each v V . Remark 1. Theorem 2 says that the energy norm is stronger than then L2 () norm. The smallest possible value of C grows with the size of . The Poincar inequality holds only if is bounded in at least one e direction. Also, the inequality does not hold if 0 d = 0. However, the inequality v 2 d C v 2 d +
2 L2 ()

v v d, xi xi
2 V

that is, C v

v v d xi xi

holds also when Stability

d = 0.

Choosing vh = uh in equation (4.7) yields uh U j

uh d + xj

uh uh d = xj xj

f uh d.

(4.17)

Note that uh U j

uh d = uT Cu = 0, xj

m m+2 1 h

u V Vh uh

Figure 4.7: An orthogonal projection on a line. Note that the vector u uh is the shortest of all vectors u vh for vh Vh , that is, u uh u vh vh Vh . since C = C T (using the notation of 4.1.4). Expression (4.17) thus becomes

uh uh d = xi xi
uh
2 V

f uh d

1/2

1/2

(CauchySchwarz) (Poincar) C e

f 2 d
1/2

u2 d h uh uh d xi xi
1/2

f 2 d
L2 ()

C f Dividing with uh
V

uh

shows that uh
V

C f L2() where the constant C does not depend on h. That is, we have shown that uh cannot blow up as the discretization is rened, which is the same as saying that the numerical solution is stable. Error bounds Choosing v = vh Vh V in variational problem (4.6) and subtracting equation (4.7) from(4.6) yields the Galerkin orthogonality,

vh (u uh ) d + xi xi

v h Uj

(u uh ) d = 0 xj

vh Vh .

This is a true orthogonality condition when Uj 0:

vh (u uh ) d = 0 xi xi

vh Vh ;

(4.18)

that is, the error u uh is orthogonal to each vh Vh with respect to the inner product (4.14). Cf. two orthogonal vectors: a b = 0 . Orthogonality relation (4.18) means that u h is an orthogonal projection of u on Vh . As illustrated in gure 4.7, the orthogonal projection on a subspace yields the vector that is closest to the given element in the norm derived from the inner product in which orthogonality is dened. Orthogonality property (4.18) (cf. gure 4.7) implies that the FE approximation is optimal in the energy norm when Ui 0: u uh V u v h V vh Vh . (4.19) The matrix A is symmetric in this case (since C = 0). When Uj = 0, the FEM approximation is no longer optimal in this sense. However, there is a C > 0 such that Ceas lemma holds, u uh
V

C u vh

vh Vh .

(4.20)

Matrix A is nonsymmetric in this case. Note that the approximation may be far from optimal when is small, as the next example illustrates.

M 0 M 0.8 1 m 0.7 m + 2 0.6 i 0.5 j h 0.4 Inow V Solid wall 0.3 Vh n 0.2 u 0 uh 0.1 1 0 0.9 0.92 0.94 0.96 0.98 1 x 2 M M Figure 4.8: Under-resolved standard Galerkin nite-element approximations may yield oscillations for m advection-dominated ows. m+2 v1, 3 1 2 u V Vh u uh v02 P t = 1, 3 h

Figure 4.9: An interpolant h v interpolates v at the nodal points. Example 1. Let us consider the above FEM applied to the 1D problem 1 u +u =1 500 u(0) = u(1) = 0 in (0, 1). (4.21)

The problem is advection dominated, that is, the viscosity coecient in equation (4.21) is quite small: = 1/500. A boundary layer with sharp gradients forms close to x = 1. The solid line in gure 4.8 shows the FE solution h = 1/1000, whereas the dashed line shows the FE approximation for h = 1/200. The result of example 1 is typical: under-resolved standard Galerkin methods may generate oscillations for advection-dominated ows in areas of sharp gradients. An explanation of this phenomenon is that the directions of ow is not accounted for in the standard Galerkin approximation. Galerkin approximations correspond to nite-dierence central schemes. Upwinding is used to prevent oscillations for nite-dierence methods. Various methods exist to obtain similar eects in FEM, for instance streamline diusion and discontinuous Galerkin. Both these methods modify the basic Galerkin idea to account for ow-direction eects. Approximation and mesh quality Estimates (4.19) and (4.20) are independent of the choice of Vh ! We only used that Vh V . Next step in assessing the error is an approximation problem: How well can functions in V h approximate those in V ? This step is independent of the equation in question. Interpolants are used to study approximation properties. The interpolant is the function h v Vh that interpolates the function v V at the node points of the triangulation, as illustrated for a 1D case in gure 4.9. Approximation theory yields that, for suciently smooth v, v h v
L2 () V

v h v

Ch

Ch2

second order error rst order error

(4.22)

The constant C contains integrals of the second derivatives of v. A mesh quality assumption is needed for (4.22) to hold. The essential point is: beware of thin triangles; do not let the shape deteriorate as the mesh is rened. A strategy to preserve mesh quality when rening is to subdivide each triangle into four new triangles by joining the edge midpoints (gure 4.10). Each of the conditions below is sucient for approximation property (4.22) to hold:

1 2 i j Inow Solid wall n 0 1 Figure 4.10: A subdivision that preserves mesh quality. M2 M m (i) (Maximum angle condition). The largest anm+2 gle of any triangle should not approach 180 as 1 h 0. h (ii) (Chunkiness parameter condition). The V quotient between the diameter of the largest Vh circle that can be inscribed and the diameter u of the triangle should not vanish as h 0. uh Condition (ii) implies (i), but not the reverse. Accuracy of the FE solution Error bounds (4.19) and (4.20) together with approximation property (4.22) implies u uh Further analysis (not provided here) yields u uh
L2 () V

M m m+2 1 h V Vh u uh

Ch

Ch2 .

This second-order convergence rate requires a mesh quality assumption and a suciently smooth solution u (since the constant C contains second derivatives of u). The convergence rate is of optimal order : it agrees with approximation property (4.22) and is the best order that in general can be obtained for, in this case, piecewise-linear functions. However, the solution may still be bad for a particular, too-crude mesh; recall example 1. Improving Accuracy Accuracy is improved by rening the mesh (h-method). This may be done adaptively, where it is needed (say, in areas of large gradients), to prevent the matrices to become too large. Automatic methods for adaptation are common. An alternative is to keep the mesh xed and increase the order of the polynomials at each triangle (p-method). For instance, continuous, piecewise quadratics yields a third-order-accurate solution (in the L2 () norm). These strategies may be combined in the hp method, where the mesh is rened in certain regions and the order of approximations is increased in other.

4.1.6

Alternative Elements, 3D

A quadrilateral is a skewed rectangle: four points connected by straight lines to form a closed geometric object. Nonoverlapping quadrilaterals can be used to triangulate a domain, as in gure 4.11. An example of a nite-element space Vh on quadrilaterals is globally continuous functions varying linearly along the edges. As in the triangular case, the functions are dened uniquely by specifying the function values at each node. The functions are dened by interpolation into the interior of each quadrilateral. This space reduces to piecewise bilinear functions for rectangles with edges in the coordinate directions: vh (x, y) = a + b x + c y + d xy Quadrilaterals on a logically rectangular domain (a distorted rectangle, as in gure 4.11) yields a regular structure to the matrix A. This may

p1,1 p1,1 2 M 0 0 M 1 1 m 2 2 m+2 i i j j 1 h Inow Inow V Solid wall Solid wall Vh n n 0 0 u uh 1 1 M 2 Figure 4.11: A logically rectangular domain triangulated with quadrilaterals. M2 M M m m m+2 m+2 1 1 h h V V Vh Vh u u uh uh Figure 4.12: In 3D, triangular and quadrilateral elements generalize to tetrahedral (left) and hexahedral (right) meshes. give high accuracy (particularly if the mesh is aligned with the ow), and allow ecient solution of the linear system (particularly for uniform, rectangular meshes). Compared to triangular mesh, it is harder to generate quadrilateral meshes automatically for complicated geometries. Triangular and quadrilateral meshes generalize in 3D to tetrahedral and hexahedral meshes (gure 4.12). Advantages and limitations with tetrahedral and hexahedral meshes are similar to corresponding meshes in 2D.

Figure 4.14: The black nodes are the degrees of freedom for the velocity components, which are fewer than the number of triangles.

4.2

FEM for NavierStokes

We now turn to the NavierStokes equations for the steady ow of an incompressible uid, and consider a situation with ow in a bounded domain . A simple example domain is depicted in gure 4.13. The mathematical problem in dierential form is uj ui p 1 + xj xi Re
2

The boundary condition on 1 = \ 0 is of no obvious physical signicance, but is useful as an articial outow condition. The NavierStokes equations is a system of nonlinear advectiondiusion equations together with the incompressibility condition (4.23b). This condition introduces new complications not present in the advectiondiusion problem of section 4.1. Naive approximations of the NavierStokes equations are bound to produce disappoing results, as the following example indicates. Example 2 (the counting argument). Consider a case with only Dirichlet boundary conditionsthat is, 1 = and the mesh of gure 4.14. Assume that we use continuous, piecewise-linear approximations u h , vh for the x- and y-components of the velocity eld. There are I number of panels in each direction and therefore 2(I 1)2 degrees of freedom (two times the number of black dots) for the velocity vector eld. The left side of equation (4.23b), that is uh vh + x y is here constant on each triangle. Thus, demanding equation (4.23b) to be satised on each triangle yields 2I 2 equations, which is more than the degrees of freedom for the velocity! Thus u h = vh = 0 is the only FE function satisfying the incompressibility condition (4.23b)!

h gh h h gh h gh h h gh h gh h h gh h gh h h h h gh h h h h h h h h gXXXgXgXXXgXXXgXgXXXgXXXgXgXXXgXXXgXgXXgXgXXXgXgXgXgXgXgXXgXgX XXghXXXXghXXXghXXXXghXXXghXXXXghXXXghXXXghXXXXghXXXXXXXghXXXgh p iXippi pi pi ippi pi ippi pi pi ippi pi ippi pi pi ippi pi ippi pi pi pi pi ippi pi pi pi pi pi pi pi pi XXXpXpXXXpXXXpXpXXXpXXXpXpXXXpXXXpXpXXpXpXXXpXpXpXpXpXpXXpXpX ipXXpiXiXiXXpiXiXXpiXiXiXXpiXiXXpiXiXiXXpiXiXXpiXiXiXpiXiXiXXpiXiXiXiXiXiXiXpiXiXiXip XXipXXXXipXXXipXXXXipXXXipXXXXipXXXipXXXipXXXXipXXXXXXXipXXXip pipXipipip p p ipipip p ipipip p p ipipip p ipipip p p ipipip p ipipip p p p p ipipip p p p p p p p p iXXXiXiXXXiXXXiXiXXXiXXXiXiXXXiXXXiXiXXiXiXXXiXiXiXiXiXiXXiXiX XXipXpXpXXipXpXXipXpXpXXipXpXXipXpXpXXipXpXXipXpXpXipXpXpXXipXpXpXpXpXpXpXipXpXpXip XXXiXiXXXiXXXiXiXXXiXXXiXiXXXiXXXiXiXXiXiXXXiXiXiXiXiXiXXiXiX ipXXipXpXpXXipXpXXipXpXpXXipXpXXipXpXpXXipXpXXipXpXpXipXpXpXXipXpXpXpXpXpXpXipXpXpXip XXXiXiXXXiXXXiXiXXXiXXXiXiXXXiXXXiXiXXiXiXXXiXiXiXiXiXiXXiXiX ipXXipXpXpXXipXXXipXXXXipXXXipXXXXipXXXipXXXipXXXXipXXXXXXXipXXXip ip XXipXiXiXXipXiXXipXiXiXXipXiXXipXiXiXXipXiXXipXiXiXipXiXiXXipXiXiXiXiXiXiXipXiXiXip pXXiXiXXXiXXXiXiXXXiXXXiXiXXXiXXXiXiXXiXiXXXiXiXiXiXiXiXXiXiX iXXpXpXXXpXXXpXpXXXpXXXpXpXXXpXXXpXpXXpXpXXXpXpXpXpXpXpXXpXpX ipip p ipip p p ipipip p ipipip p p ipipip p ipipip p p p p ipipip p p p p p p p p pipXXXpXpXXXpXXXpXpXXXpXXXpXpXXXpXXXpXpXXpXpXXXpXpXpXpXpXpXXpXpX iXXipXiXiXXipXiXXipXiXiXXipXiXXipXiXiXXipXiXXipXiXiXipXiXiXXipXiXiXiXiXiXiXipXiXiXip pXXXpXpXXXpXXXpXpXXXpXXXpXpXXXpXXXpXpXXpXpXXXpXpXpXpXpXpXXpXpX iXXXiXiXXXiXXXiXiXXXiXXXiXiXXXiXXXiXiXXiXiXXXiXiXiXiXiXiXXiXiX XXipipXXXXipipXXXipipXXXXipipXXXipipXXXXipipXXXipipXXXipipXXXXipipXXXXXXXipipXXXipip ipip ipip ipip ip ip ip ip ip
ui = f i ui =0 xi ui = g i 1 ui nj + pni = 0 Re xj 1

1 h V Vh 0 u uh

0 0 1 2 i j Inow Solid wall 0 n 0 1 Figure 4.13: A simple example of a domain. M2 0 M m m+2 1 h V 0 Vh 0 u uh 0 1 0

ww xx yy

uu vv dd

ss tt ee ff

rr qqq

in , in . on 0 on 1 .

(4.23a) (4.23b) (4.23c) (4.23d)

Following strategies can be used to tackle the problem of example 2. (i) Use more degrees of freedom for the velocityfor instance continuous, piecewise quadratics on each triangleto balance the large number of equations associated with the incompressibility condition. (ii) Accept that the discrete velocities only is approximately divergence-free. (iii) Associate the velocity with the midpoints of edges instead of the vertices. This also gives more degrees of freedom for the velocities. (There are 3I 2 edges, but only 2I 2 vertices). Strategies (i) and (ii) are the standard methods, whereas (iii) leads to non-conforming approximations related to nite-dierence and nite-volume approximation with staggered mesh.

4.2.1

A variational form of the NavierStokes equations

To derive the variational form, we need new integration-by-part formulas. The product rule of dierentiation yields ui zi ui zi = + z i 2 ui (4.24) xj xj xj xj and p zj (zj p) = zi +p . xj xi xj (4.25)

Integrating expressions (4.24) and (4.25) and applying the the divergence theorem (4.3) on the left sides yields ui zi ui nj z i d = d + zi 2 ui d (4.26) xj xj xj

and nj zj p d =

zi

p d + xi

zi d, xi

(4.27)

which are the integration-by-parts formulas needed to derive our variational form. Now, let zi be a smooth vector-valued function on that vanishes on 0 . Multiply both sides of equation (4.23a) with zi , integrating over , and utilizing (4.26) and (4.27), we nd that zi fi d =

z i uj

ui d + xj

zi

p 1 d xi Re

zi

ui d nj z i ui d xj p

ui 1 z i uj d + xj Re z i uj 1 ui d + xj Re

zi ui 1 d xj xj Re zi ui d xj xj

zi d + xi

nj zj p d (4.28)

zi d, xi

where, in the last equality, the boundary integrals vanish on 0 since zi = 0 on 0 and on 1 due to boundary condition (4.23d). Moreover, multiplying equation (4.23b) with a smooth function q and integrating yields q

ui d = 0 xi

(4.29)

Thus, we conclude that a classical solution (a twice continuously dierentiable solution) to the Navier Stokes equations satises the variational forms (4.28) and (4.29) for all smooth functions z i and q such that zi vanishes on 0 . Similarly as for the advectiondiusion problem, the variational form will be used to dene weak solutions, without explicit reference to the dierential equations. A dierence with advectiondiusion problem is that we here need dierent function spaces for the velocity components and the pressure. Also note that ui = gi on 0 whereas zi vanishes on 0 . To avoid this last complication and simplify the exposition, we will only consider homogeneous boundary conditions, gi = 0. Thus, the equations will be driven only by the right-hand side fi , which, for instance, could be gravity or magnetic forcing for a electrically conducting uid. The pressure space is H = L2 () = q| q 2 d < +

and the space of velocity components is V = u| u u d < + and u = 0 on 0 xi xi .

For the analysis in section 4.2.4, it will be convenient to work with the space V of the velocity vector u = (u1 , u2 ). Saying that u V is equivalent to saying that that each velocity component u i V . The energy norm for elements of V is denoted u whereas the L2 () norm is
1/2

ui ui d xj xj

1/2

L2 ()

ui ui d

We assume that both boundary conditions really are active. In the case of only Dirichlet boundary condition, that is, that 0 constitute the whole boundary, the pressure is only possible to specify uniquely up to an additive constant. The variational problem dening weak solutions to equation (4.23) is Find ui V and p H such that ui 1 zi ui z i uj d + d xj Re xj xj

zi d = xi

zi fi d

zi V

ui q d = 0 xi

q H

This is a variational problem of mixed type: two dierent spaces are involved. As for the advectiondiusion problem, the boundary conditions are incorporated in the variational formulation. The essential boundary conditions are here that ui = 0 on 0 and is explicitly assigned through the denition of V . The natural boundary conditions are the boundary conditions on 1 , implicitly specied through the variational form. Note that the natural boundary condition is dierent than for the advectiondiusion problem. Remark 2. The mathematical properties of the NavierStokes equations are complicated, and important issues are still unresolved. In fact, there is a $ 1 000 000 prize for the person that can develop a theory that can satisfyingly explain their behavior in three space dimensions! A short summary of what is known is the following. For the unsteady problem on a given, nite time interval (0, T ), and in two space dimensions: a unique weak solution exists for any square-integrable f and for any Re, smooth data yield smooth solutions, so that weak solutions are classical solutions if data is smooth enough; in three space dimensions: a weak solution exists for any square-integrable f and for any Re. It is not known whether the weak solution always is unique (besides during some initial interval (0, T ), where the size of T is unknown). It is not known whether smooth data always yields a smooth solution, or if singularities can evolve from smooth data. The steady problem has at least one weak solution, in 2D as well as 3D. Note that the steady problem may have several solutions, even in 2D. Note that the natural boundary condition here not just is a Neumann condition, but the more complicated condition (4.23d).

4.2.2

Finite-element approximations

To obtain a FEM, we choose subspaces Vh V , Hh H of piecewise polynomials, and replace V with Vh and H with Hh in the variational form. In 2D and in components, we obtain Find uh , vh Vh and ph Hh such that z h uh

uh uh + vh x y

d +

1 Re

zh uh d xj xj

ph

zh d = x

zh f1 d zh f2 d

zh Vh , zh Vh ,

(4.30a) (4.30b) (4.30c)

zh

vh vh uh + vh x y uh vh + x y

1 d + Re

zh vh d xj xj

zh ph d = y

qh

d = 0.

We postpone for a while the question on how to choose the subspaces Vh and Hh .

4.2.3

The algebraic problem in 2D

Np

Expanding ph in basis for Hh , we obtain ph (x, y) =

n=1

p(n) (n) (x, y) .

(4.31)

Likewise, expanding uh , vh in basis for Vh yields

Nu Nu

uh (x, y) =
n=1

u(n) (n) (x, y) ,

vh (x, y) =
n=1

v (n) (n) (x, y) .

(4.32)

Now insert expansions (4.31) and (4.32) into equations (4.30). In (4.30a) and (4.30b), choose z h = (m) for m = 1, . . . , Nu ; in (4.30c), choose qh = (m) for m = 1, . . . , Np . Then, equations (4.30) become
Nu

u(n)
n=1

(m) uh
Np

(n) (n) + vh x y (n) d = x

(m)

d +

1 u (n) u Re n=1

(m) (n) (m) (n) + x x y y

Nu

p(n)
n=1

(m) f1 d

m = 1, . . . , Nu (m) (n) (m) (n) + x x y y

v (n)
n=1

(m) uh
Np

+ vh x y

(n)

(n)

d +

1 u (n) v Re n=1

(4.33)

Nu

p(n)
n=1 (n)

(m) (n) d = y
Nu

(m) f2 d (n) d = 0 y

m = 1, . . . , Nu

u(n)
n=1

(m)

d + v (n) x n=1

(m)

m = 1, . . . , Np

Placing the coecients into column matrices, u = u(1) , . . . , u(N u) p = p(1) , . . . , p(Np )
T T

, ,

v = v (1) , . . . , v (N u)

we can write the system (4.33) in the matrixvector form (x) 1 C(uh , vh ) + Re K 0 G(x) u b 1 0 C(uh , vh ) + Re K G(y) v = b(y) , p 0 G(x)T G(y)T 0

(4.34)

where the matrix and vector components are Kmn =

(m) (n) (m) (n) + x x y y (m) (n) d, x

d,

Cmn (uh , vh ) =

(m) uh

(n) (n) + vh x y b(y) = m

d,

G(x) = mn

G(y) = mn

(m) (n) d, y

b(x) = m

(m) f1 d,

(m) f2 d. (4.35)

Matrix K represents an approximation of the discrete negative Laplacian 2 ; C(uh , vh ) the discrete advection operator u x + v y ; G(x) and G(y) the discrete gradient ; and G(x)T and G(y)T the discrete divergence. That is, the structure of the NavierStokes equations are directly reected in the nonlinear system (4.34). We obtained the same structure of the nonlinear equation for the nite-volume discretization, except that here, it is guaranteed the the last block row of the matrix is the negative transpose of the last block column.

4.2.4

Stability

So far, the approximating spaces Vh and Hh have not been specied. The counting argument of example 2 indicates that not any choice is good. We will limit the discussion to the Stokes equations, whose niteelement formulation reads Find uh Vh and ph Hh such that i
h zi uh i d xj xj h zi d = xi

ph

h zi fi d

h z i Vh ,

(4.36a)

uh qh i d = 0 xi

q h Hh .

(4.36b)

The Stokes equations is a limit case of the NavierStokes equations for very low Reynolds numbers and assume that inertia can be neglected compared to viscous forces. The Stokes equations may be used to model very viscous ow, creeping ow, lubrication, but also ow of common uids (like air and water) at the micro scale. Due to the lack of nonlinear advection, these are linear equations. Mathematically, the Stokes equations are much easier to analyze than the full NavierStokes equations. Indeed, there exists a unique weak solutions for each square-integrable fi both in two and three space dimension. The reason to start the analysis on the Stokes equation is that a stable Stokes discretization is necessary to obtain a stable NavierStokes discretization. As for the NavierStokes equations, we require that Vh V and Hh H. Recall that each function in V and H is required to be bounded in the energy and the L2 () norm, respectively. Thus, in makes sense to require that the velocity and pressure approximations satisfy the same bounds, that is,
1/2 1/2

ph

L2 ()

p2 d h

C
1/2

fi fi d

=C f

L2 () ,

(4.37a) (4.37b)

uh

uh uh i i d xj xj

C f

L2 () ,

where the constant C should not depend on h. This implies that the approximations will not blow up when the mesh is rened. h Choosing zi = uh in equation (4.36a), we obtain i uh uh i i d xj xj ph

uh i d = xi

uh uh i i d = xj xj

uh fi d, i

where the rst equality follows from equation (4.36b). Thus, uh

2 V

uh fi d u i

L2 ()

L2 ()

C u

L2 ()

(4.38)

where the CauchySchwarz inequality yields the rst inequality, and the Poincar inequality the second. e We thus conclude from expression (4.38) that uh
V

C f

L2 () ,

(4.39)

that is, any Galerkin approximation of the velocity in the Stokes equations is stable! For bad choices of subspaces, we will of course obtain inaccurate velocity approximations, but they will not blow up as the mesh is rened, since their energy norm is bounded by the given data fi . Regarding the pressure, the situation is more complicated. Not any combination Hh and V h yield stable pressure approximations in the sense of (4.37a). We will derive a compatibility condition between the velocity and pressure spaces, the LBB condition, necessary to yield stable pressure approximations.

4.2.5

The LBB condition

h zi d = xi h zi uh i d xj xj h h zi zi d xj xj V V

h Rearranging equation (4.36a) implies that for each zi Vh ,

ph

1/2

h zi fi d

(CauchySchwarz)

L2 () V

uh uh i i d xj xj f f
L2 () L2 () L2 () .

1/2

1/2

1/2

h h zi zi d

fi fi d

(by (4.39)) C z h

= zh (Poincar) z h e

uh uh
V

V V

+ zh + zh

f
V

L2 ()

+ zh

(4.40)
V

Dividing expression (4.40) with z h

yields that ph
h zi d (C + 1) f xi L2 ()

1 zh
V

h for each zi Vh , from which it follows that there is a C > 0 such that

0=zi V

max

1 zh
V

ph

h zi d C f xi

L2 () .

(4.41)

We have thus found a bound on the discrete pressure in the exotic norm (4.16). However, we were aiming for the bound (4.37a). Thus, if it happens that ph
L2 ()

max

1 zh
V

0=zi Vh

ph

h zi d xi

(4.42)

for some > 0 independent of h, then it is immediate that ph L2 () C f L2 () for some C > 0. Condition (4.42) says that our exotic norm is stronger than the L2 ()-norm for ph . The LBB-condition is a condition on the pairing of spaces Vh and Hh that requires the exotic norm to be stronger than the L2 ()-norm for each function qh Hh . The derivation above proves thus that the LBB-condition is sucient for the pressure to be stably determined by data: Theorem 3. Assume that the LBB-condition holds for the spaces H h , Vh . That is, assume that there is an > 0 independent of h, such that, for each qh Hh ,
1/2

2 qh d

max

1 zh
V

0=zi Vh

qh

h zi d, xi

(4.43)

then there is a C > 0 such that

1/2 1/2

p2 d h

|f |2 d

holds, independently of h. The LBB condition (4.43) is satised for Vh = V and Hh = H. That is, the pressure (before discretization) is stably determined by data. The expression qh xii d represents the discrete gradient of qh (see 4.2.3). The LBB condition thus means that a zero discrete pressure gradient implies a zero value of the pressure. Conversely, if the LBB condition is not satised, it means that the discrete pressure gradient may be small even when the pressure is not small: The discrete gradient cannot feel that the pressure is large, an eect that usually manifests itself as wild pressure oscillations.
z h

 M2 M m m+2

h V Vh 1 u uh 0 1 1

Figure 4.15: A checkerboard mode for equal-order interpolation on a regular triangular mesh.

4.2.6

Mass conservation

The integral ni ui d yields the total ux of mass(in kg/s) through the boundary. There is no net mass-ux through the boundary for a constant-density ow, so ni ui d = 0. A velocity approximation uh is is globally mass conservative if i

ni uh d = 0, i uh i d = 0, xi

or, equivalently, if

by the divergence theorem (4.3). Recall from equation (4.30c) that qh

uh i d = 0 xi

qh Hh .

A Galerkin approximation is thus globally mass conservative if 1 Hh , that is, if constant functions can be represented exactly in Hh . All reasonable FE approximation satisfy this. A nite-element approximation is locally mass conservative if uh i d = 0, xi

for each element K in the mesh. Not all nite-element approximations are locally mass conservative. If the pressure approximations consist of piecewise constants on each element, the FE approximation will be locally mass conservative. However, it will not be locally mass conservative if the pressure approximations are continuous and piecewise linear.

4.2.7

Choice of nite elements. Accuracy

Now we consider a regular triangulation of a 2D domain , that is, a triangulation satisfying a mesh quality condition of the sort discussed in section 4.1.5. Equal-order interpolation Let us start by trying the approximation space we used for the advectiondiusion problem (section 4.1.1). That is, we use continuous functions that are linear on each triangle both for the functions in H h and Vh . This pair does not satisfy the LBB condition and yields an unstable discretization. In particular, the system matrix (4.34) is singular : the discrete gradient matrix G has linearly dependent columns. Thus, there is at least one nonzero Np -vector pc such that Gpc = 0. Corresponding function pc Hh , whose h nodal values are equal to the elements of pc is called a checkerboard mode. Figure 4.15 depicts the nodal values of a checkerboard mode for equal-order interpolation. To see this, consider the element contribution, associated with a triangle Tn , to the ith row of vector Gpc (notice from the expressions in gure 4.5 that i is constant on each triangle): x
Tn

pc h

i i d = x x

Tn Tn

pc d 0, h

h V Vh u uh 0 1 Figure 4.16: The evaluation points at each triangle for the continuous, piecewise quadratic velocity approximations of the TaylorHood pair. since the mean value of the function pc vanishes on each triangle, as can be seen from gure 4.15. The right h side of the LBB condition (4.43) vanishes for qh = pc , which immediately implies that the LBB cannot h hold for a space Hh that includes pc . This (and other) checkerboard mode will pollute the solution of the h Stokes as well as NavierStokes equations.. Constant pressure, (bi)linear velocities Now we consider a triangulation either of triangles or rectangles. Let the functions in H h be piecewise constant on each element. In case of triangular elements, the space Vh will be continuous functions, piecewise linear on each element, whereas for quadrilateral elements, continuous functions that are piecewise bilinear on each rectangle comprise Vh . Remark 3. The pressure approximations need not to be continuous in order for Hh H. However, if Vh V , the velocity approximations cannot be discontinuous, since v is not a function for discontinuous v. This choice leads to a locally mass-conserving approximation. However, for triangular elements, the counting argument of example 2 applies here and indicates problems with this choice of elements. Indeed, the LBB conditions is not satised for this unstable discretization. Checkerboard modes similar as in gure 4.15 exist. Still, the constant pressurebilinear velocity space on rectangles is in widespread engineering use! Fixes have been developed to take care of pressure oscillations. For instance, the checkerboard mode can be ltered out from the discrete pressures. Moreover, common iterative methods to solve the resulting discrete equations (such as the projection schemes) implies a stabilization of the discretization. The TaylorHood pair Here, the functions in Hh and Vh are globally continuous and, on each triangle, ph Hh is linear, whereas uh , vh Vh is quadratic. Thus, on each triangle, we have uh = a + b x + c y + d xy + e x2 + f y 2 The six coecients af are uniquely determined by uh s values at the triangle vertices and midpoints (gure 4.16). Nu A basis (m) (xi ) m=1 of continuous, piecewise quadratic functions interpolates the nodal values of gure 4.16, so that uh Vh can be expressed as uh (xi ) = Nu ui (m) (xi ). There are two kinds of basis m=1 functions here, illustrated in gures 4.17 and 4.18. The LBB condition is satised for the TaylorHood pair, and the following error estimates holds for a smooth solution to the NavierStokes equation: u uh p ph
L2 () L2 ()

Ch3

Ch2

The TaylorHood pair is globally, but not locally mass conservative. The TaylorHood pair can be generalized to higher order. Then, the pressure and velocity approximations will be continuous functions, consising of piecewise polynomials of degree k for pressure and k + 1 for velocity (k 1) on each triangle.

m+2 m+2 0 d 11d 1 d 1 2h a h a ia V V a i 1, j j 1 i Vh i 1, j j 1 i, jVh 1 Inowu u j1 i, i + 1, j 1 Inow Solid wallh uh u j i + 1, j 1 Solid wall n0 0 j i 0 1 n i 1 i+ 0 1 1 + 1 i1 ui,j u vi,j i,j The basis function (m) (x ) when M 2 4.18: The basis function (m) (xi ) when Figure 4.17: Figure vi,j i pi,j M2 M m corresponds to an edge-midpoint node m corresponds to a corner node p pi,j M i,j m pi,j vi,j+1/2 m+2 v m vi,j1/2 O(h2) m i,j+1/2 +2 1 vi,j1/2 ui1/2,j 1 h 1 ui1/2,j i 2 V ih 1 2 i V Vh 1 i+ 2 i Vh 1 u i+ 2 i+1 u uh 3 i+ i+1 uh 3 0 2 i+ 2 j+1 0 1 2 j+1 j 1 2 j j 1 2 1 j2 j Figure 4.20:1Isoparametric elements use polynomials j Figure 4.19:1Approximating a curved boundary with of the sameA order as used for the FE approximations A a triangle sides leads to an O(h2 ) error of the bound- to approximate the shape of the boundary. B B ary condition. C 3 u 2 ,1 C 3 u 2 ,1 v1, 3 2 v1, 3 Remark 4. 2 Inaccuracies in approximations of the domain boundary may reduce the convergence rate of p1,1 p1,1 higher-order approximations. For instance, a straight-forward triangulation of a domain with a curved 0 boundary will lead to a boundary-condition error at the mid nodes of O(h2 ) (gure 4.19). This is ne 0 1 for piecewise-linear approximations since the error anyway is O(h2 ). However, this inaccuracy at the 1 2 boundary results in a half-order reduction of the convergence rateiof piecewise quadratic approximations. 2 Using curved elements restores convergence rate. The most common strategy for curved elements is known i j j as Isoparametric Elements: piecewise polynomials of the same degree as used in the FE approximation are Inow used toInow approximate the shape of the boundary (gure 4.20). wall Solid Solid wall n n 0 A stable approximation with piecewise-linear velocities and pressures 0 1 Instead of 1 using dierent approximation orders, as in the TaylorHood pair, we here consider dierent meshes for velocity and pressure. Both the velocity and pressure approximations consist of continuous, M2 piecewise-linear functions on triangles, but each pressure triangle is subdivided into four velocity triangles M2 M as in gure 4.21. M m This is a stable approximation satisfying the LBB condition, + 2 the following error estimates holds for m m and m+2 1 1 h h V V Vh Vh u u uh uh 0 0 1 1 Figure 4.22: Subdividing a pressure rectangle into Figure 4.21: Subdividing a pressure triangle into four four velocity rectangles yields a stable approximation velocity triangles yields a stable approximation with with continuous piecewise-bilinear functions for both continuous, piecewise-linear functions for both veloc- velocity and pressure. ity and pressure.

1 h V Vh u uh 0 1

1 h V Vh u uh 0 1 Figure 4.24: Approximations spanned by the basis functions (4.23) yields piecewiselinear functions that are continuous only along at the edge midpoints, in general.

Figure 4.23: The basis functions associated with the velocity approximation for a nonconforming but stable discretization.

a smooth solution to the NavierStokes equation: u uh

L2 () L2 ()

p ph

Ch2 Ch

These approximations use the same number of unknowns and the same nodal locations as for the Taylor Hood pair, but are one order less accurate. Calculating the matrix elements (4.35) is however somewhat easier than for the TaylorHood pair. There is also a version of this discretization for rectangular meshes: Each pressure rectangle is then subdivided into four velocity rectangles, as in gure 4.22. Both the velocity and pressure approximations consist of continuous and piecewise-bilinear functions. A nonconforming method The discussion after example 2 suggested a strategy to address the problem with the overdetermined incompressibility condition, namely to use nodal points on the edge mid points for the velocity. To accomplish this, consider basis functions, associated with the edges, of the type depicted in gure 4.23. Basis function j is linear on each triangle and continuous along edge j (but discontinuous along four edges!). If x i , y i are the coordinates of the mid point of edge i, we have j (xi , y i ) = 1 for i = j, 0 for i = j.

Now dene the space of the velocity components Vh from all possible expansions
N N

uh (x, y) =
j=1

uj j (x, y),

vh (x, y) =
j=1

vj j (x, y),

Here, N is the total number of edges, excluding those on 1 . Note that uh will be continuous only at edge mid points in general. Thus, Vh V since uh is not a square-integrable (in fact, constant) function. Violating a property like Vh V is called a variational crime. However, uh restricted on each element is a square-integrable function. Letting Hh be piecewise constant on each triangle and Vh dened as above, we dene the nonconforming FE approximation: Find uh , vh Vh and ph Hh such that
M

z h uh
n=1T n M

uh d + x n=1 vh d + x n=1
M

zh vh

Tn

uh 1 d + y Re n=1 vh 1 d + y Re n=1 q h Hh ,
M

Tn

zh uh d xj xj n=1 zh vh d xj xj n=1
M

ph

zh d = x

zh f1 d

z h Vh , z h Vh ,

Tn

z h uh
n=1T n M

zh vh

ph

zh d = y

zh f2 d

Tn

Tn

Tn

qh
n=1T n

uh vh + x y

d = 0

where M is the number of triangles.

The fact that Vh V complicates the analysis of this discretization. Nevertheless, the approximation is stable, although not as accurate as the TaylorHood pair, for instance. Other nice properties with this method is that it is locally mass conservative, and quite easy to implement. But maybe the most interesting with this discretization is that it allows for a construction of a divergence free velocity basis. In all types of discretizations discussed so far, including the present, we have used the same basis for the uh and vh components. By a clever combination of basis functions of the type illustrated (j) (j) in gure 4.23, it is possible to construct a new vector-valued basis functions j = (zh , wh ) for the velocity having the property j = Expanding the velocity in this basis,
M

zh wh + x y

(j)

(j)

= 0.

uh = (uh , vh ) =
j=1

uj j ,

(4.44)

yields that uh = 0. Inserting expansion (4.44) into equations (4.30), and choosing (zh , wh ) = i , the continuity equation and the pressure variable vanishes, and equations (4.30) reduce to
M M dNu

uj
j=1

i (uh

) j d +
j=1

uj

j d +
j=1

uj

i (uh

) j d =

i f d, (4.45)

for i = 1, . . . , M . The NavierStokes equations then reduces to a vector-valued, non-linear advection diusion problem, quite similar to equation (4.10). This leads to a substantial simplication when designing an algorithm to solve the discretized equations; in fact, one of the great challenges for any solver of the NavierStokes equations is how to treat the incompressibility condition (4.30c). Unfortunately, the construction of the basis functions j is not straightforward.

Stabilization of unstable discretizations Stability of the choices of elements presented in sections 4.2.74.2.7 relied on manipulations of the local polynomial order, the meshes, or the conformity of the method. These manipulations complicate the implementation. An alternative is to use the same approximations for velocity and pressure (equal-order interpolation) combined with a stabilization method. This circumvents the LBB condition. We will present a very simple idea of this sort for the Stokes equations. First choose the same type of approximations for the pressure and the velocity components, for instance continuous, piecewise-linear functions on a triangular mesh, and replace equations (4.36) with Find uh Vh and ph Hh such that i
h zi uh i d xj xj

ph

h zi d = xi

h zi fi d

h z i Vh ,

(4.46a)

qh ph d + xi xi

uh qh i d = 0 xi

qh Hh .,

(4.46b)

for some small > 0. Equation (4.46b) formally inconsistent by O( ) with corresponding variational form. More advanced stabilization methods avoids this inconsistency by using a term that vanishes as h 0. The main advantage with stabilization methods is the simplied implementation, but the method introduces an extra parameter that may need judicious tuning. Note also that the increased number of pressure unknowns, as compared with the discretization in section 4.2.7, does not increase the accuracy. The added resolution introduced is ltered away by the introduction of the stability term. h To see that approximation (4.46) is stable, choose zi = uh , qh = ph and add equations (4.46a) i

and (4.46b), uh uh i i d + xi xi ph ph d = xi xi uh fi d i
1/2 1/2

(CauchySchwarz) (Poincar) C e C

uh uh d i i uh i

fi fi d
1/2 1/2

uh i

xi xi

fi fi d uh uh i i d xi xi
1/2 1/2

ph ph d + xi xi

fi fi d

Dividing and squaring yields the stability estimate uh uh i i d + xi xi ph ph d C xi xi fi fi d.

Since C > 0 does not depend on h, both the velocity and pressure approximations cannot blow up as h 0, even though the discretization is unstable for = 0.

j j1 2 j1 A B C 3 u 2 ,1 v1, 3 2 p1,1 0 1 2 i j Inow Solid wall A.1 Iterative solutions to linear systems n 0 1 Gauss-Seidel method The Laplace equation on a Cartesian grids can be discretized as M2 M i+1,j 2i,j + i1,j + 2 (i,j+1 2i,j + i,j1 ) = 0 m m+2 where = x/y. We can solve for i,j , 1 i+1,j + i1,j + 2 (i,j+1 + i,j1 ) h i,j = V 2 (1 + 2 ) Vh Gauss-Seidel uses this expression to update i,j (using the already updated values calculated in same u iteration), i.e.u h 0 k+1 k+1 2 k k i+1,j + i1,j + i,j+1 + i,j1 1 k+1 = i,j 2 (1 + 2 )

Appendix A

Background material

i1

i+1

j1

This iteration method has a slow convergence, it is possible to show k m 0 where = 1 O h2 h = x = y

e.i. the error is reduced by O h2 each iteration. Requiring an error reduction to O h2 the number of iterations must satisfy m = O h2 which implies that m=O ln(h) ln() =O ln(h) ln(1 O(h2 )) = O h2 ln(h)

To proceed we note that we have N = n2 unknown interior nodes, see Figure A.1. Thus 95

ite

: old values

ra

tio

: new values

di

re ct

: to be updated

io

j+1

 1 M2 h = x = y = n1 = N 1/2 M n+1 m which implies that m+2 1 m = O (N ln(N )) h and that the total work is V Vh W = O N 2 ln(N ) u uh since the work per iteration is O (N ). 0 1 y : to be updated : new values1 : old values n i1 i i+1 j1 j j+1 iteration direction 1 1 n 1 Figure A.1: Domain for the solution of the Laplace equation. x

Multigrid method
A general way to accelerate the convergence of e.g. the Gauss-Seidel method is provided by the Multigrid method. We note that the Gauss-Seidel provides good smoothing of the local error, but converges slowly since it takes time for boundary information to propagate into the domain. The idea behind the multigrid method is that a slowly converging low-frequency on a ne grid is a fast converging high-frequency on a coarse grid. Let Li,j = i,j+1 2i,j + i,j1 i+1,j 2i,j + i1,j + x2 y 2

and dene the correction to the solution to the true discrete solution in the following manner i,j k i,j i,j This implies that Li,j = Lk Ri,j i,j since Li,j = 0. Thus we are left with an equation implying that the Laplace operator on the correction is equal to the negative of the residual. This residual equation is solved on a coarser grid and the correction is interpolated onto ner grid. In order to dene the algorithm we use the following denitions I transfer operator between grids: 2 An example of a restriction operator I1 is to 2 I1 - restriction operator from grid 1 to grid 2 1 I2 - prolongation operator from grid 2 to grid 1 1 dene it as an injection, i.e. we use every other value in each direction. The prolongation operator I 2 may be dened by linear interpolation for the intermediate values, i.e. an average of right and left values = k + i,j i,j solution at iteration level k correction to k to true discrete solution i,j

: old values i1 i i+1 j1 j j+1 iteration direction x y 1 n

ne grid (1) coarse grid (2)

j x i Figure A.2: Two level grid. and top and bottom values. The values marked with in gure A.2 are then taken as an average of the averages of the intermediate values. Now we can dene the following two level scheme: 1. Li,j iterated k times
2 2. Li,j = I1 Ri,j 1 3. i,j = k + I2 i,j i,j

Lk = Ri,j i,j

iterated k times, 0 = 0 (starting value) (here i,j is denied on grid 2) i,j and goto 1.

at the end of 1. convergence is checked. Wesseling [10] estimated the work of the multigrid method to be W = O (N ln N ), a substantial improvement over the Gauss-Seidel method.

i y = x2 i i+1 j j1 ne grid (1) j j coarse grid (2) +1 iteration direction r x y e1 e1 1 x = x1 n e3 e2 x z = x3 y i Figure A.3: Cartesian coordinate system. j ne grid (1) x2 coarse grid (2) x2 e2 e2 e1 e1 x1 a e1 e3 x3 x1

e3 x3

Figure A.4: Two dierent coordinate systems.

A.2

Cartesian tensor notation

Tensor notation is a compact way to write vector and tensor formulas. Vectors and tensors are objects which are independent of the coordinate system they are represented in. In the Cartesian coordinate system in Figure A.3 we have the unit vectors |ek | = 1 k = 1, 2, 3

The scalar product of two unit vectors is the Kroneckers delta, kl ek el = kl = and the position vector r is
3

1 k=l 0 k=l

r = x 1 e1 + x2 e2 + x3 e3 =
k=1

xk e k = x k e k .

The last expression make use of Einsteins summation convention: when an index occurs twice in the same expression, the expression is implicitly summed over all values for that index. Example kk = ll = 11 + 22 + 33 = 3 ij aj = i1 a1 + i2 a2 + i3 a3 = ai .

A.2.1

Orthogonal transformation

Consider two arbitrary oriented coordinate systems as in Figure A.4. Assume identical origin, a = 0. The unit vectors are orthogonal ek el = kl

n x y i j ne grid (1) coarse grid (2)

e 1 x1

e1 e2

e2 e1 x1 = cos (x2 , x1 ) x1 = c21 x1 Figure A.5: Projection of e1 x1 on e2 . and a position vector is independent of coordinate system, r = x l el = xl el . Component k of r can be written as ek r = xl ek el = xl el ek = xl kl = xk where we have the transformation tensor ck l = ek el = cos(xk , xl ). Since cos(x2 , x1 ) = cos(x1 , x2 ) in general, we have (for k = 1, l = 2) cl k = el ek = cos(xl , xk ) = ck l From ek r = x l ek el = xl ek el Example xk = ckl xl = ck1 x1 + ck2 x2 + ck3 x3 = cos(xk , x1 )x1 + cos(xk , x2 )x2 + cos(xk , x3 )x3 = e k e 1 x1 + e k e 2 x2 + e k e 3 x3 which for component 2 reads x2 = e 2 e 1 x1 + e 2 e 2 x2 + e 2 e 3 x3 x2 is projection of e1 x1 , e2 x2 , e3 x3 on e2 . A visualization of the projection of e1 x1 on e2 can be seen in Figure A.5. The relation between ckl and cml can be studied through xk
km xm

x k = e k e l xl = c k l xl

xk = el ek xl = cl k xl = clk xl

we conclude that rst index refer to primed system!

= ckl xl
cml xm

= ckl cml xm

(ckl cml km ) xm = 0.

In the same manner we have xk

km xm

= clk

xl
clm xm

= clk clm xm

(clk clm km ) xm = 0.

A.2.2

Cartesian Tensors

Def. a Cartesian tensor of rank R is a set of quantities Tklm... which transform as a vector in each of its R indices. Tklm... = ckr cls cmp . . . Trsp...

Example For scalars the rank is R = 0 a b = ak bk = ckl ckm al bm = al bl ,

lm

for vectors R = 1 ak = ckl al and for a tensor of second order we have R = 2 kl = ckm cln mn = ckm clm = kl . kl is isotropic, i.e. identical in all coordinate systems. An outer product generates a tensor Tkl Srs = ckm cln crp csq Tmn Spq and an inner product generates a tensor Tkl Sls = ckm cln clp csq Tmn Spq = ckm csq Tmn Snq
np

A.2.3

Permutation tensor
klm 1 even 1 odd = if klm is permutation of (1, 2, 3) 0 no even permutation odd permutation no permutation 123 = 1 213 = 1 221 = 0

As examples of permutations of 1,2,3 we have (1,2,3), (3,1,2), (2,3,1) (2,1,3), (3,2,1), (1,3,2) at least 2 subscripts equal

A.2.4

Inner products, crossproducts and determinants

a b = ij ai bj = ai bi = a1 b1 + a2 b2 + a3 b3 .

The inner product of the vectors a and b is

The cross product of a and b can be written as ab= e1 a1 b1 e2 a2 b2 e2 a3 b3 = (a2 b3 a3 b2 )e1 + (a3 b1 a1 b3 )e2 + (a1 b2 a2 b1 )e3 = ijk ei aj bk

and the determinant of the tensor aij is |{aij }| = = a11 a21 a31 a12 a22 a32 a13 a23 a33 = ijk a1i a2j a3k

1 1 ijk (ai1 aj2 ak3 + ai3 aj1 ak2 + ai2 aj3 ak1 ai2 aj1 ak3 ai3 aj2 ak1 ai1 aj3 ak2 = ijk pqr aip ajq akr . 6 6 We have the permutation relation klm ksp = ls mp lp ms . If we set s = l in (A.1) we get and if we also use p = m we end up with klm klp = (3 1)mp = 2mp klm klm = 6 The permutation tensor can also be used to derive the following vector relation (a b) (c d) = ijk aj bk ilm cl dm = (jl km jm lk )aj bk cl dm = aj cj bk dk aj dj bk ck = (a c)(b d) (a d)(b c) (A.1)

A.2.5

Second rank tensors

All second rank tensors can be decomposed into symmetric and antisymmetric parts, T kl = T(kl) + T[kl] , where the antisymmetric part is 1 T[kl] = T[lk] = (Tkl Tlk ) 2 and the symmetric part 1 T(kl) = T(lk) = (Tkl + Tlk ) . 2 The symmetric part can be decomposed into isotropic and traceless part, T(kl) = T kl + T kl , the isotropic part being 1 T kl = Tmm kl 3 and the traceless part is 1 T kl = T(kl) Tmm kl . 3 If we project the isotropic, traceless and entisymmetric parts we get S(kl) T[kl] = S kl T kl = 1 1 S(kl) T[kl] + S(lk) T[lk] = S(kl) T[kl] S(lk) T[kl] = 0 2 2 = 1 1 Smm Tkk Tkk ll 3 3 =0

1 1 Smm kl T(kl) Tmm kl 3 3 Skl T[kl] Skl T(kl) Skl T kl Skl T kl

= S[kl] T[kl] = S(kl) T(kl) = S kl T kl = S kl T kl ,

and each part is invariant under transformation T[kl] = ckr cls T[rs] = ckr cls T[sr] = clr cks T[rs] = T[lk] Trr = crp crq Tpq = pq Tpq = Tpp 1 T kl = ckp clq T pq = ckp clq T(pq) Trr pq 3 1 = T(pq) Trr kl . 3

Since we only need three components to completely describe an antisymmetric tensor of rank R = 2 it can be represented by its dual vector di = ijk Tjk = $$$$ + ijk T[jk] ijk T(jk) The rst term on the right hand side is zero since ijk is antisymmetric in any two indecies. Multiply both sides by ilm ilm di = ilm ijk Tjk = (lj mk lk mj ) Tjk = Tlm Tml = 2T[lm] Note: There are 3 independent components in T[lm] and di . In order to sum up, an arbitrary tensor of rank 2 can be divided into the following parts Tkl = T(kl) + T[kl] = T kl + T kl + T[kl] = 1 1 Trr kl + T(kl) Trr kl + 3 3
isotropic traceless

T[lm] =

1 ilm di 2

1 ikl di 2
antisymmetric

Properties of the above parts are invariant under transformation. In the Navier-Stokes equations we have ui the velocity gradient xj . The antisymmetric part of this tensor describes rotation, the isotropic part the volume change and the traceless part describes the deformation of a uid element.

Example We have the second rank tensor

which can be divided into a symmetric part

1 2 3 {Tkl } = 4 5 6 7 8 9 1 3 5 = 3 5 7 5 7 9

T(kl) and an antisymmetric part T[kl]

The symmetric part can then be divided into a isotropic 5 T kl = 0 0 and a traceless part T kl The dual vector of the antisymmetric tensor is

0 1 2 = 1 0 1 . 2 1 0 part 0 0 5 0 0 5

4 3 5 = 3 0 7 . 5 7 4

2 {di } = 4 . 2

A.2.6

Tensor elds

We have the tensor eld Tij (x1 , x2 , x3 ) Tij (xk ) with the following properties partial derivative transforms like a tensor xk = = cpk xp xp xk xk gradient adds one to tensor rank Tij xk where (xk = cpk xp )

divergence decreases tensor rank by one T1l T2l T3l Tkl = + + xk x1 x2 x3 curl uk xj

u)i = ijk

which can be compared with the expression for the determinant. Example Show that

) = 0. ( ) = m 2 klm = klm m = 0 xk xl xk xl is symmetric in kl.

since klm is antisymmetric in kl and

2 xk xl

y 1 n x y i j ne grid (1) coarse grid (2)

dS = n dS Figure A.6: Volume V bounded by the close surface S. Example Show that

f (r) = 1 f (r) where {r}k = xk and r = |r| = r =

xk xk .

{ f (r)}k

df r (xp xp )1/2 f (r) = =f = xk dr xk xk 1 1 1 1 f (xp xp ) 2 (xp xp ) = f (pk xp + xp pk ) = f xk = 2 xk 2r r

1 fr r

Example Show that (

( F)

F) =
i

F) F.

= ijk

klm Fm = ijk klm Fm = kij klm Fm = xj xl xj xl xj xl ( F) F

i

(il jm im jl ) Example

Fm = Fj Fi = Fj Fi = xj xl xj xi xj xj xi xj xj xj u . This means, show that ijk uk = 2i . 2 xj

Show that if u = r then = ijk

uk = {uk = klm l rm } = ijk (klm l rm ) = ijk klm l rm = xj xj xj l rm = (il jm im jl ) l rm = xj xj ri = il } = xl

kij klm

rm ri rm i rm l ri = { independent of x} = i l ={ =1+1+1=3 xm xl xm xl xm 3i l il = 2i

A.2.7

Gauss & Stokes integral theorems

Let n be the outward pointing normal to the closed surface S bounding the simply-connected volume V , see Figure A.6. Gauss integral theorem then states a n dS =
S V

a dV ak dV xk

or ak nk dS =
S V

j ne grid (1) coarse grid (2) S dl

Figure A.7: The open surface S enclosed by the contour C. for a dierentiable vector eld a. This can be generalized to Tklm np dS =
S V

Tklm dV . xp

Example f n dS =
S V

f dV .

a n dS =
S V

a dV

if

Tklm nm = klm al nm .

Stokes theorem states that for a dierentiable vector eld a if an open simply-connected surface S is enclosed by a closed curve C, of which dl is the line element as in Figure A.7, then a dl =
C S

a) n dS am dS xk

or ak dxk =
C S

klm nk

this can be generalized to Tklm dxp =

C S

ijp ni

Tklm dS xj

Example Let Tklm = f. In the generalized Stokes theorem we then get

S

f ijp ni xj dS in right hand side

f dl =
C S

f dS

A.2.8

Archimedes principle

Gauss theorem can be used to calculate the buoyancy on an object immersed in a uid (Archimedes circa 281212 BC). Consider a water cylinder with volume dV and mass dm, Figure A.8. The force from the water cylinder is dm g = dV g = x3 g dA
pressure

where g is the gravity constant and the water density. The pressure at a distant x 3 from the surface in a uid is then p = gx3 + p0 where p0 is the atmospheric pressure. Let be the buoyancy force resulting from the pressure on the object, m the mass of equivalent volume of water and M the mass of the object in Figure A.8. The force on a small element dS of the body resulting from the pressure is dk = gx3 nk dS

i + 1 1 p0 i+ 3 2 1 j updated : to be+ 2 j : new 1 values n j2 : old values j1i1 dV dm A i Bi+1 Cj 1 3 u 2 ,1 j v1, 3 dA 2 j + 1 p1,1 M x3 iteration direction 0 x 1 y 2 1 Figure A.8: An object with mass M immersed in a uid illustrating Archimedes principle. i n j mg x Inow y Solid wall i n j ne 0 (1) grid coarse 1 (2) grid M2 Mg M m Figure A.9: The forces acting on an object immersed in a uid. m is the mass of equivalent volume of m+2 water and M the mass of the body. 1 h where the minus sign stems from the outward pointing normal n. The buoyancy force can then be calculated V through Vh x3 dV = gk3 V. k = gx3u k dS = g n xk uh S V 0 This means that the object loses some weight as the water is displaced. The buoyancy force is 1 3 = gV = g m : to be updated see Figure A.9. : new values The moment of a force F at some arbitrary point with position vector r from the point of application of : old values the force is M = r F. The moment around the center of gravity G of the homogenous body produced by i1 the pressure force d acting on an element dS of the body is dMG = (r rG ) d as in Figure A.10. i This can then be used to calculate the total moment around G form pressure forces on the body i+1 j1 x3 MGk = gklm (xl xGl ) x3 nm dS = gklm (xl x3 ) xGl dV = j xm xm S V j+1 iteration direction gklm [lm x3 + m3 xl m3 xGl ] dV = gkl3 xl dV xGl V = 0 x y V V 1 If a part of the uid is frozen (without changing in density) this body is in equilibrium in the uid: n F = M = 0. x y i j ne grid (1) G coarse grid (2) r rG n x y i j ne grid (1) coarse grid (2) n Figure A.10: Moment around G from pressure forces acting on the object.

i1 i i+1 j1 j j+1 iteration direction x y 1 n x y i j ne grid (1) coarse grid (2)

z P (r, , z)

z y r x Figure A.11: Cylindrical polar coordinates

A.3

Curvilinear coordinates

Cylindrical Polar Coordinates

The cylindrical polar coordinates are (r, , z), where is the azimuthal angle, see gure A.11. The velocity can be written as u = u r er + u e + uz ez , (A.2) where the unit vectors are related to Cartesian coordinates as er = ex cos + ey sin , e = ex sin + ey cos , ez = e z . Non-zero derivatives of unit vectors Gradient of a scalar p p= Laplacian of a scalar p
2

(A.3)

er = e ,

e = er .

(A.4)

p 1 p p e + e + e . r r r z z + 1 2p 2p + 2. r2 2 z

(A.5)

p=

1 p r r r r

(A.6)

Divergence of a vector u u= Advective derivative of a scalar p (u Curl of a vector u u = 1 uz u er + r z ur uz 1 (ru ) ur e + ez . z r r r (A.9) )p = ur p u p p + + uz . r r z (A.8) 1 (rur ) 1 u uz + + . r r r z (A.7)

Incompressible Navier-Stokes equations with no body force ur + (u t u + (u t )ur )u + u2 1 p = + r r

2

ur

ur 2 u 2 , r2 r

(A.10) (A.11) (A.12)

ur u 1 p 2 u u 2 = + u + 2 r 2 , r r r r uz 1 p + (u )uz = + 2 uz . t z

i1 i i+1 j 1 j j +1 ration direction x y 1 n x y i j ne grid (1) coarse grid (2)

z P (r, , ) r

x Figure A.12: Spherical polar coordinates

Spherical Polar Coordinates

The spherical polar coordinates are (r, , ), where is the azimuthal angle, see gure A.12. The velocity can be written as u = u r er + u e + u e , (A.13) where the unit vectors are related to Cartesian coordinates as er = ex sin cos + ey sin sin + ez cos , e = ex cos cos + ey cos sin ez sin , e = ex sin + ey cos . Non-zero derivatives of unit vectors er = e , er = e sin , e = er , e = e cos , (A.15) (A.16) (A.14)

e = er sin e cos . Gradient of a scalar p p= Laplacian of a scalar p

2

p 1 p 1 p e + e + e . r r r r sin

(A.17)

p=

1 p r2 2 r r r

1 p sin 2 sin r

1 2p . r2 sin2 2

(A.18)

Divergence of a vector u u= Advective derivative of a scalar p (u Curl of a vector u u= 1 r sin (u sin ) u (ru ) 1 1 ur er + e r sin r + 1 (ru ) ur e . r r (A.21) )p = ur p u p u p + + . r r r sin (A.20) 1 (r2 ur ) 1 (u sin ) 1 u + + . r2 r r sin r sin (A.19)

Incompressible Navier-Stokes equations with no body force ur + (u t = 1 p + r

2

)ur

u2 + u 2 r

(A.22)

ur

2ur 2 (u sin ) 2 u 2 2 , r2 r sin r sin )u + (A.23)

u + (u t = 1 p + r

u2 cot ur u r r 2 cos u 2 u u 2 u + 2 r 2 2 2 2 , r r sin r sin )u +

u + (u t = 1 p + r sin

ur u u u cot + r r u 2 ur 2 cos u 2 u + 2 + 2 2 2 2 . r sin r sin r sin

(A.24)

Appendix B

Recitations 5C1214
B.1
Scalar p=p Vector ()i = ui u u If u = v then u2 = v w a11 If A = a21 a31 1 if i = j 0 if i = j a12 a22 a32 a13 a23 then A23 = a23 a33

Tensors and invariants

Tensor Notation

Matrix (A)ij = Aij

Kronecker delta ij = (I)ij =

Einstein summation convention

3

ui ui =
i=1

ui ui

Kinetic energy per unit volume

1 u2 2 || 1 1 = 2 (u2 + v 2 + w2 ) = 2 ui ui

Matrix operations a = a1 b1 + a2 b2 + a3 b3 = ai bi = ij ai bj = aj bj b (A i = Aij bj b) (AB)ij = Aik Bkj tr(A) = Aii (A)ij = Aij (AT )ij = Aji Permutation symbol 109

ijk

1 if ijk in cyclic order. ijk = 123, 231 or 312 = 0 if any two indices are equal 1 if ijk in anticyclic order. ijk = 321, 213 or 132 e1 a1 b1 e2 a2 b2 e3 a3 b3 = e1 (a2 b3 a3 b2 ) e2 (a1 b3 a3 b1 ) + e3 (a1 b2 a2 b1 ) = ijk ei aj bk

a= b

( i = ijk aj bk a b) ijk ilm = jl km jm kl ijk ijm = {l = j} = 3km jm kj = 3km mk = 2km ijk ijk = {m = k} = 2 3 = 6 Rewrite without the cross product: ( ( d) = ( i ( d)i = ijk aj bk ilm cl dm = ijk ilm aj bk cl dm = a b) c a b) c (jl km jm kl ) aj bk cl dm = aj cj bk dk aj dj bk ck = ( c)( d) ( d)( c) a b a b

Invariant parts of Tensors

S A Tij = Tij + Tij S Tij = 1 2 1 2

Symmetric and Anti-symmetric parts.

S = Tji

Tij + Tji Tij Tji

Symmetric Anti-symmetric

A Tij =

A = Tji

The symmetric part of the tensor can be divided into two parts: TS = T + T
ij ij ij 1 S Tij = Tij 3 Tkk ij

trace less

Tij = 1 Tkk ij 3 This gives:

isotropic

S A A Tij = Tij + Tij = Tij + Tij + Tij

ui . The anti-symmetric part describes rotation, the isotropic xj part describes the volume change and the trace-less part describes the deformation of a uid element. In the NS equations we have the tensor

Operators
( p)i = p = p xi

2 p xi xi ui xi uk xj

u= (

u)i = ijk

Gauss theorem (general)

Gauss theorem:

F n dS =

F dV Fi dV xi

or, Fi ni dS =
S V

In general we can write, Tijk nl dS =

S V

Tijk dV xl

Example: Put Tijk nl = Tij ul nl Tij ul nl dS =

S V

(ul Tij ) dV xl

or,
S

T( n) dS = u

(
V

u)T + ( u

)T dV

Identities
Derive the identity (F G) = (G (F G)
i

)F + (

G)F (

F )G ( F

)G

= ijk

klm Fl Gm = ijk klm Fl Gm = kij klm Fl G m = xj xj xj

(il jm im jl ) Gj

Fi Gj Fj Gi Fl G m = Fi G j Fj G i = Gj + Fi Gi + Fj = xj xj xj xj xj xj xj )F + ( G)F ( F )G ( F )G]i

Fi Gj Fj Gi + Fi Gi F j = [(G xj xj xj xj

Show that: ( ( ijk Fk = ijk Fk F) = xi xj xi xj Fk = Fk and thus all terms will cancel. xi xj xj xi F) = 0

Remember that ijk = jik and that ijk Fk = 0 xi xj

Example
Show that if u = r then = ijk u . This means, show that ijk uk = 2i . 2 xj

uk = {uk = klm l rm } = ijk (klm l rm ) = ijk klm l rm = xj xj xj l rm = (il jm im jl ) l rm = xj xj ri = il } = xl

kij klm

rm ri rm i rm l ri = { independent of x} = i l ={ =1+1+1=3 xm xl xm xl xm 3i l il = 2i Derive the identity (F G) = ( F) G F ( G)

 (F G) = kij

Fj Gk Fj Gk ijk Fj Gk = ijk Fj Gk = ijk Gk + ijk Fj = kij Gk + Fj ijk = xi xi xi xi xi xi F) G F ( F) = G) F ) F

Fj Gk Gk Fj jik =( xi xi

Show that: ( F)

= ijk

klm Fm = ijk klm Fm = kij klm Fm = xj xl xj xl xj xl ( F ) F

i

(il jm im jl )

Fm = Fj Fi = Fj Fi = xj xl xj xi xj xj xi xj xj xj

B.2

Euler and Lagrange coordinates

Particle path (Lagranges representation)

Following a uid particle as time proceeds the path is described by r = u, t r (t = 0) = x0

Streamline (Eulers representation)

Instantaneously a uid particles displacement d is tangential to its velocity u: x d x u = |d| x || u or if we put ds = |d|/||: x u d x =u ds

Exercise 1
Show that the streamlines for the unsteady ow u = u0 , v = kt, w = 0, u0 , k > 0

is straight lines, while any uid particle follows a parabolic path as time proceeds. Streamline (x t): dx = u = u0 x = u0 s + a ds dy = u = kt y = kts + b ds Then we see that y(x) is a linear function, y= Particle path (x x0 ): kt x + d(t) u0 x = u 0 t + x0 x= 1 2 k t + y0 2

x = u = u0 t y = u = kt t

Then we see that y(x) is a parabola, y= 1 k 2 (x 2x0 x + x2 ) + y0 0 2 u2 0

Exercise 2
a) Separate the shear ow u = (y, 0, 0) into its (i) local translation, (ii) rotation and (iii) pure straining parts.

Lets denote the dierent parts as u = u T + uR + uD In tensor notation it can be written as ui 2 = ui 1 + ij dxj + eij dxj
P P

(i) Translation is simply uT = {ui 1 } = (y, 0, 0)

1 j2 (ii) j1 A B C 3 u 2 ,1 v1, 3 2 p1,1

0 ui = 0 0 xj 0 0

0 0 0

ij =

1 ui uj 2 xj xi

uR = d = x

1 (dy, dx, 0) 2

0 1 = 2 0

0 0

0 0 0

1 (iii) 2 0 0 1 ui uj 1 i + = 0 0 eij = j 2 xj xi 2 0 0 0 Inow 1 Solid wall uD = ed = (dy, dx, 0) x n 2 0 1 M 2 b) Find the principal axes of the rate of strain tensor eij and make a schematic sketch of the decomposition M of the ow. 1 ui uj ui m + symmetric part of . eij = 2 xj xi xj m+2 1 0 0 1 h eij = 0 0 2 V 0 0 0 Vh uEigenvalues: uh /2 0 0 0 = 3 + 2 /4 = (/2)(+/2) = 0. 1 = /2, 2 = /2, 3 = 0. det(eij ij ) = /2 1 0 0 : to be updatedWe get the eigenvectors from eij kj = ki : : new values 1 1 : old values k1 = (1, 1, 0), k2 = (1, 1, 0), k3 = (0, 0, 1) 2 2 i1 iNow sketch the decomposition of the ow i+1 j1 1 1 (dy, 0, 0) = (dy, dx, 0) + (dy, dx, 0) j 2 2 j+1 iteration direction x 1 1 1 y n x y i j ne grid (1) coarse grid (2)
0.5 0.5 = 0.5 +

0
0.5 1 1

0
0.5 1 1

0
0.5 1 1

1 Exercise 3 2 ia) Sketch the streamlines for the ow [u, v, w] = [x, y, 0], where > 0. j Streamlines in parametric form (x(s), y(s), z(s)). We have a 2D ow since w = 0. Inow dx dy Solid wall = x, = y. ds ds n 0 Integrate w.r.t. s, 1 x(s) = a es , y(s) = b es . M 2 Then we see that, x(s) y(s) = ab = c ( constant ). M mWe have, c c m+2 x(s) = , y(s) = 1 y(s) x(s) h Streamlines: V Vh dx dy > 0, <0 x > 0, y > 0 u ds ds uh dx dy 0 x < 0, y > 0 < 0, <0 ds ds 1 dx dy x < 0, y < 0 < 0, >0 ds ds to be updated dx dy : new values x > 0, y < 0 > 0, >0 : old values ds ds i1 dx(s) dy(s) = x and = y. This gives the direction of the ow: iRemember that ds ds i+1 j 1 1 j j +1 ration direction 0.5 x y 0 n x y 0.5 i j ne grid (1) 1 coarse grid (2) 1

0.8

0.6

0.4

0.2

0.2

0.4

0.6

0.8

b) The concentration of a scalar is c(x, y, t) = x2 yet . Does this concentration change for a particular uid particle in time? Use the material time-derivative c c c Dc = +u +v = x2 yet + x2xyet yx2 et = (1 + 2 1)x2 yet = 0. Dt t x y Thus the concentration c(x, y, t) is constant for a uid particle.

c) Using the alternative (Lagrangian) description of the ow, show that u D u = Dt t and write the concentration as c = c(x0 , y0 , t).

Particle path:

x = u = x x = x0 et t y = v = y y = y0 et t u x = = 2 x0 e t = 2 x t t t v y = = 2 y0 et = 2 y t t t Du u u u = +u +v = 2 x Dt t x y Dv v v v = +u +v = 2 y Dt t x y

u D u = Dt t Both terms describe the acceleration of a uid particle. Now rewrite the concentration c(x, y) = x2 yet = x2 e2t y0 et et = x2 y0 = c(X) 0 0
x2 y

So we have

independent of t!

The concentration foes not change as we follow a uid particle.

B.3

Reynolds transport theorem and stress tensor

Exercise 1
a) Consider a xed closed surface S in a uid. Show that conservation of mass implies + (ui ) = 0. t xi The change of mass in the volume is described by the contribution from the mass ux and the change in density and conservation means that this should be zero dV + t ui ni dS = 0
S

Using Gauss Theorem we can rewrite the surface integral ui ni dS =

S V

(ui ) dV. xi

We have
V

+ (ui )} dV = 0. t xi

This must be true for an arbitrary volume and this implies + (ui ) = 0. t xi b) Show that this can be written D ui + = 0, Dt xi ui ui (ui ) = ui + = ui + xi xi xi xi xi This gives D ui ui + ui + = + t xi xi Dt xi D Dt c) Explain what the following means ui =0 xi D =0 Dt

If we are following a uid particle, the density and volume are constant. But in a x position, the density can change = ui t xi Remember the decomposition of
ui xj

in its invariant parts 1 uk 1 ui uj ij + 3 xk 2 xj xi

rotation ij

ui 1 ui uj = + xj 2 xj xi .

1 uk ij 3 xk
volume change eij

deformation eij

Exercise 2
a) Use Reynolds transport theorem to provide an alternative derivation of the conservation of mass equation + (ui ) = 0. t xi Reynolds Transport Theorem: D Dt Put Tijk = , D Dt dV =
V (t) V (t)

V (t)

Tijk dV =

V (t)

Tijk + (u T ) dV t xl l ijk + (ui ) dV t xi

No ow through the surface. Valid for all V (t): + (ui ) = 0 t xi b) Use this result to show D Dt
V (t)

Fijk dV =

V (t)

DFijk dV Dt

Put Tijk = Fijk in Reynolds transport theorem D Dt Fijk dV = (Fijk ) + (u F ) dV = t xl l ijk dV = dV =

V (t)

V (t)

V (t)

Fijk Fijk + Fijk + Fijk (ul ) + ul t t xl xl Fijk Fijk + (ul ) + + ul t xl t xl

=0 =
DF ijk Dt

V (t)

Fijk

V (t)

DFijk dV Dt

Exercise 3
Show that the stress tensor Tij = pij for the ow u = x, where is constant. Tensor notation: ui = ikl k xl The stress tensor: Tij = pij + pij + uj ui + xj xi = =

(ikl k xl ) + ( x ) xj xi jkl k l xl x + jkl l xj xi =

pij + k ikl

pij + k (ikl lj + jkl li ) = pij + k (ikj + jki ) = pij + k (ijk + ijk ) = pij

The stress tensor Tij

The equation for the stress tensor was deduced by Stokes in (1845) from three elementary hypotheses. Tij = pij + 2 eij Writing this on the form, Tij = pij + ij the following statements should be true for the viscous stress tensor ij in a Newtonian uid. ui (i) each ij should be a linear combination of the velocity gradients . xj (ii) each ij should vanish if the ow involves no deformation of uid elements. ui (iii) the relationship between ij and should be isotropic as the physical properties of the uid xj are assumed to show no preferred direction.

Exercise 4
If ij is a linear function of the components of eij then it can be written ij = cijkl ekl It can be shown that the most general fourth order isotropic tensor is of the form cijkl = Aij kl + Bik jl + Cil jk where A, B and C are scalars. a) Use this to show that ij = ekk ij + 2eij where and are scalars. ij = cijkl ekl = (Aij kl + Bik jl + Cil jk )ekl = Aij ekk + Beij + Ceji Say that A = and that B = C = , then since eij = eji we have ij = ekk ij + 2eij 1 b) Show that if we have a Newtonian uid where p = Tii then 3 ui uj 2 uk Tij = p + ij + + 3 xk xj xi We have Tij = pij + ij = pij + ekk ij + 2eij Then This gives = 2 , 3 Tii = 3p + 3eii + 2eii = 3p + (3 + 2)eii and thus 2 Tij = p ekk ij + 2eij 3 2 uk ui uj Tij = p + ij + + 3 xk xj xi

or

h V B.4 Vh

Rankine vortex and dimensionless form

The Rankine vortex

n x y i j ne grid (1) coarse grid (2)

3
0 0 1

0 1

uh A simple model for a vortex is that it is a rigid body rotation within a core, and a decay of angular velocity 0 outside. This can be described by 1 r, r < a, : to be updated u = a2 ur = u z = 0 , r > a, : new values r : old values i 1and is called a Rankine vortex. i i+1 3 j1 j 2 j+1 2 iteration direction x y 1 u

Figure B.1: Velocity and vorticity in a Rankine vortex with = a = 1.

Exercise 1
a) Find the pressure inside and outside of a Rankine vortex: We use Eulers equations for incompressible ow ( neglecting viscous eects ): u D = 1 p + g Dt Eulers equations u=0 u D u = +( u Dt t
=0

) u

We are working with cylindrical coordinates, use the appendix formulas u ( u ) = u = ur u + + uz r r z

u u u u e u u u2 (u e ) = e + u = e er r r r r r
=r e

p= Insert this into the Euler equations:

p 1 p p er + e + ez r r z

u u u2 1 p 1 p p e er = er + e + ez r r r r z
=0

gz e

Look at the dierent components: er : e : 0=

u2 1 p = r r

1 1 p p = p(r, z) only. r 0= 1 p g z

ez : Solve for the pressure when r < a: er : 2 r =

1 p r

p = 2

r2 + f (z) 2

ez : So we have the pressure:

p = g z

f (z) = gz + C1 r<a

p(r, z) = 2 Solve for the pressure when r > a: er :

r2 gz + C1 2

2 a4 1 p = r3 r p = g z

p=

2 a4 + f (z) 2 r2

ez : So we have the pressure:

f (z) = gz + C2

2 a4 gz + C2 r > a 2 r2 Now determine the dierence between the constants by evaluation at r = a p(r, z) = 2 a2 2 a2 gz + C1 = gz + C2 2 2 C2 C1 = 2 a2

b) Determine the pressure dierence p between r = 0 and r p = p p0 = gz + C2 (gz + C1 ) = C2 C1 = 2 a2 c) Now consider a free surface at atmospheric pressure p0 . Find the dierence in z between r = 0 and r r=0: r: p0 =
2 a2 gz0 + C1 2 2 a2 p0 = 2 gz +

C2

z z0 =

C2 C 1 2 a2 = g g

Determine the shape of the free surface: p0 = p0 =

2 r 2 gz + C1 2 2 a4 2r2 gz + C2

r < a z r2 z = r>a z
1 r

C1 p0 2 r2 2g + g 2 4 p = a2 + C2g 0 2gr

Say that z = 0 at r = 0 then C1 = p0 and we get z= and z= So we have z(r) = 2 r2 2g r<a

2 a4 C2 C 1 2 a4 2 a2 2 a2 a2 + = + = 1 2 2 2 2gr g 2gr g g 2r 2 r2 2g ga r<a 1

a2 2r 2

r>a

2 2

r > a.

i i+1 j1 j j+1 iteration direction x y n x y i j ne grid (1) coarse grid (2)

0.1 0.09 0.08 0.07 0.06 0.05 0.04 0.03 0.02 0.01

0 0

0.5

1.5 r

2.5

Figure B.2: The free surface of a Rankine vortex with = a = 1 and g = 9.82. 2 r2 2g ga

r<a 1
a2 2r 2

z(r) =

2 2

r > a.

Exercise 2
u + ( u t with the boundary conditions

Solving the 2D ow around a cylinder involves solving ) = u 1 p+

2

u,

u = 0,

u = 0 on x2 + y 2 = a2

u (U, 0, 0) as x2 + y 2 .

Rewrite this problem in dimensionless form using the dimensionless variables x = x/a u = u/U Note that the scaling x = x/a implies Change to dimensionless variables =a p = p/U 2 t = tU/a. a = . t U t U a2
2

and t = tU/a gives

U 2 u U2 + ( u a t a Divide by U 2 /a u + ( u t

) = u

U 2 a

p +

) = u

p +

U a2 U2 a
1 Re

U2 Inertial forces Ua The Reynolds number is Re = a = = U Viscous forces a2 u + ( u t The continuity condition u=0 U a u =0 u =0 ) = u p + 1 Re
2

The boundary conditions u = 0 on x2 + y 2 = a2 U u = 0 on a2 x + a2 y

2 2

= a2

u = 0 on x + y
2 2

= 1.

u (U, 0, 0) as x2 + y 2

U u (U, 0, 0) as a2 x + a2 y
2 2

u (1, 0, 0) as x + y

The solutions of this problem will depend on x , Re and t only, and thus the solution of this problem is the same for a specic Re independently of the individual values of U , a and .

Exercise 3
Show that the net viscous force per unit volume is proportional to the spatial derivative of vorticity, i.e. ij k = ijk xj xj and discuss its implication for ows with uniform vorticity (as in solid-body rotation). ij = xj xj ijk ui uj + xj xi = 2 ui 2 uj + xj xj xj xi = 2 ui xj xj

k um = ijk klm xj xj xl

= kij klm

2 um 2 um = (il jm im jl ) = xj xl xj xl 2 ui xj xj

2 ui 2 uj xj xi xj xj

Thus

ij k = ijk xj xj

The net viscous force vanishes when the vorticity is uniform, since no deformation exists.

j+

j j 1Exercise 1 APlane Couette ow: BConsider the ow of a viscous uid between two parallel plates at y = 0 and y = h. The upper plane is Cmoving with velocity U . Calculate the ow eld. 3 u 2 ,1 v1, 3 2 Assume the following: p1,1 Steady ow: 0 =0 1 t 2Parallel ow: ui i =0 v = 0, j x InowTwo-dimensional ow: Solid wall w = 0, =0 z n 0 No pressure gradient: p 1 =0 xi M 2 NavierStokes streamwise momentum equation: M u 1 m + ( )u = u p + 2u t m+2 1We have h 2u u =0 = A u = Ay + B V y 2 y Vh Boundary conditions: u u(0) = 0 B = 0, u(h) = U A = U/h uh 0 So we get: Uy 1 u(y) = h : to be updated : new valuesExercise 2 (i) : old valuesPlane Poiseuille ow: ( Channel ow ) i 1Consider the ow of a viscous uid between two solid boundaries at y = h driven by a constant pressure igradient p = [P, 0, 0]. Show that i+1 P 2 j1 u= (h y 2 ), v = w = 0. j 2 j+1 iteration direction Y x y 1 n x X y i j Z ne grid (1) coarse grid (2) Figure B.3: Coordinate system for plane Poiseuille ow NavierStokes equations: u + ( ) = 1 p + u u t u = 0.

1 2

j B.5

Exact solutions to Navier-Stokes

j+

2 1 2

u(y = h) = 0 j j 1 u We are considering a stationary ow and thus = 0. A t The constant pressure gradient implies u = u(y). Changes of u in x, z would require a changing pressure B gradient in x, z. C 3 v u 2 ,1 The continuity equation u reduces to = 0. The boundary condition v(y = h) = 0 then implies v1, 3 y 2 p1,1 v = 0. Lets study the z component of the NavierStokes equations: 0 1 w 2w v = 2 w = c1 y + c2 2 y y i =0 j InowThe boundary conditions w(y = h) = w(y = h) = 0 imply c1 = c2 = 0 and thus w = 0. Solid wall n We can conclude that u = [u(y), 0, 0]. 0 1 Lets study the x component of the NavierStokes equations: P 2u u P P 2 M2 0= + 2 = y + d1 { = } u(y) = y + d1 y + d 2 M y y 2 m m + 2The boundary conditions at y = h give 1 P 2 P 0= h + d1 h + d2 and 0 = h2 d1 h + d2 h 2 2 V Vh P 2 uWe can directly conclude that d1 = 0 and this gives d2 = 2 h and thus uh 0 P 2 u= (h y 2 ), v = w = 0. 1 2 to be updated Exercise 2 (ii) : new values : old valuesPoiseuille ow: ( Pipe ow ) i 1Consider the viscous ow of a uid down a pipe with a cross-section given by r = a under the constant p i pressure gradient P = . Show that i+1 z j 1 P 2 j uz = (a r2 ) ur = u = 0. 4 j +1
R

Boundary conditions:

ration direction x y 1 n x y i j ne grid (1) coarse grid (2)

Figure B.4: Coordinate system for Poiseuille ow Use the NavierStokes equations in cylindrical coordinates ur + ( u t )ur u2 1 p = + r r
2

ur

ur 2 u 2 2 r r

u + ( u t

)u +

ur u 1 p = + r r )uz =

u +
2

2 ur u 2 r2 r

uz + ( u t

1 p + z

uz

1 1 u uz (r ur ) + + = 0. r r r z p p We know that = 0 and = 0 and can directly see that ur = u = 0 satisfy the two rst equations. r From the continuity equation we get uz =0 z uz = uz (r, ) only.

Considering a steady ow we get from the NavierStokes equations ( u But we know that )uz = uz uz z uz uz 1 = P + z
2

uz

uz = 0 from the continuity equation. We get z

2

uz =

1 uz P (r )= r r r

uz P (r )= r r r

Integrate r Integrating again we get uz =

uz P = r 2 + c1 r 2

uz P c1 = r+ r 2 r

P 2 r + c1 ln(r) + c2 4

using the boundary conditions uz (r = 0) < c1 = 0 P a2 and we have 4 P 2 (a r2 ) 4

We also have uz = 0 at r = a and this gives c2 = uz =

Exercise 3
Calculate the asymptotic suction boundary layer, where the boundary layer over a at plate is kept parallel of a steady suction V0 through the plate. Assumptions. Two-dimensional ow:

= 0, z = 0, x

w=0

Parallel ow:

Continuity satised =0 t

Steady ow:

Momentum equations:

u u u 1 p 2u 2u +u +v = + + 2 t x y x x2 y v v v 1 p 2v 2v +u +v = + + 2 t x y y x2 y

Normal momentum equation gives

p =0 y

Boundary conditions: y=0: u = 0, v = V0 y: Continuity gives u v + =0 x y Streamwise momentum equation at y V0 U 1 p 2 U = + y x y 2 Resulting streamwise momentum equation V0 Characteristic equation 2 = V0 1 = 0, 2 = V0 u 2u = 2 y y 2u V u = 0 y 2 y p =0 x v = V0 u U

u(y) = A + BeV0 y/

Exercise 4
Two incompressible viscous uids ow one on top of the other down an inclined plane at an angle . They both have the same density and the viscosities 1 and 2 . The lower uid has depth h1 and the upper h2 . Assuming that viscous forces from the surrounding air is negligible and that the pressure on the free surface is constant, show that 1 g sin() u1 (y) = [(h1 + h2 )y y 2 ] . 2 1 Make the ansatz u1 = [u1 (y), 0, 0] and u2 = [u2 (y), 0, 0]. The continuity equation u v + =0 x y Layer 1: NS ey : 0 = gives v = 0 v = c and the boundary condition at y = 0 gives v = 0. y 1 p1 g cos() y

p1 = g cos()y + f1 (x) f1 (x) = c1

1 d 2 u1 NS ex : 0 = f1 (x) + 1 2 + g sin() dy Layer 2: NS ey : 0 = 1 p2 g cos() y

p2 = g cos()y + f2 (x) f2 (x) = c2

1 d 2 u2 NS ex : 0 = f2 (x) + 2 2 + g sin() dy The pressure at the free surface y = h1 + h2 is p0 : p0 = g cos()(h1 + h2 ) + f2 (x) The pressure is continuous at y = h1 : p0 + gh2 cos() = gh1 cos() + f1 (x) This gives the pressure:

f2 = p0 + g(h1 + h2 ) cos()

f2 = 0

f1 = p0 + g(h1 + h2 ) cos()

f1 = 0

p1 (y) = p2 (y) = p(y) = g cos()y + p0 + g cos()(h1 + h2 )

We now have two momentum equations in x: 0 = 1 0 = 2 And four boundary conditions: BC1: No slip on the plane: BC2: No viscous forces on the free surface: BC3: Force balance at the uid interface: 1 u1 (0) = 0 2 du2 |y=h1 +h2 = 0 dy d 2 u1 + g sin() dy 2 d 2 u2 + g sin() dy 2 (1) (2)

du1 du2 |y=h1 = 2 |y=h1 dy dy g 2 y sin() + c11 y + c12 2 1 g 2 y sin() + c21 y + c22 2 2

BC4: Continous velocity at the interface: u1 |y=h1 = u2 |y=h1 (1) (2) du1 g = y sin() + c11 dy 1 u1 = u2 =

du2 g = y sin() + c21 dy 2

BC1 c12 = 0 g g BC2 2 ( (h1 + h2 ) sin() + c21 ) = 0 c21 = (h1 + h2 ) sin() 2 2 g g 2 g BC3 1 ( y sin()+c11 ) = 2 ( y sin()+c21 ) { = } c11 = c21 = (h1 +h2 ) sin() 1 2 1 1 g 2 g g 2 g BC4 h1 sin() + (h1 + h2 ) sin()h1 = h1 sin() + (h1 + h2 ) sin()h1 + c22 2 1 1 2 2 2 c22 = g sin() This gives us the velocities, u1 (y) = u2 (y) = = g 2 g g sin() 1 y sin() + (h1 + h2 ) sin()y = [(h1 + h2 )y y 2 ] 2 1 1 1 2 1 1 2 1 h2 1 (h1 + h2 )h1 2 1 1 2 1

g sin() 2 g sin() h2 y + (h1 + h2 )y + g sin() 1 (h1 + h2 )h1 2 2 2 2

g sin() 1 h2 ((h1 + h2 )y y 2 ) + g sin() 1 (h1 + h2 )h1 2 2 2

1 1 2 1

The velocity in layer 1 does depend on h2 but not on the viscosity in layer 2. This is due to that the depth is important for the tangential stress boundary condition at the interface but the viscosity is not. There is no acceleration of the upper layer and thus the tangential stress must be equal to the gravitational force on the upper layer which depends on h2 but not on 2 .

B.6

More exact solutions to Navier-Stokes

Exercise 1
Oscillating RayleighStokes ow (or Stokes second problem). a) Show that the velocity eld u = [u(y, t), 0, 0] satises the equation u 2u = 2 t y Consider the NavierStokes equation in the x direction: u u u 1 p 2u 2u 2u u +u +v +w = + + 2 + 2 t x y z x x2 y z With the current velocity eld only terms with y derivatives will remain since there can be no change in the other directions. Further more, the streamwise pressure gradient has to be zero since the streamwise velocity far from the wall is constant, namely zero. u 2u = 2 t y b) Show that the velocity eld is u(y, t) = U eky cos(ky t), where k = 2

Make the ansatz: u = Insert into the equation:

f (y)ei t = f (y) cos( t). if (y)ei t = f (y)eit f (y) 2 i f (y) = 0 with f (y) = ey i = gives 1+i 2

i =0

Introduce k =

, 2 f (y) = Aeyk(1+i) + Beyk(1+i) , f (y) 0 as y A = 0

ui1/2,j 1 i hWe have 2 V i V1 u= Beyk(1+i) eiwt = Beky ei(tky) = Beky cos(t ky) = Beky cos(ky t) i+ h 2 u i+1 uh i + 3 Boundary condition at y = 0 1 2 0 j+2 1 At y = 0 we have u = U cos(t) B = U j 1 j2 : to be updatedSo we have j1 : new values u(y, t) = U eky cos(ky t), where k = A 2 : old values B i1 C 3 i u 2 ,1 5 iv1, 1 +3 j p 1 4.5 2 1,1 j 4 j+1 1 3.5 iteration direction x 3 i y 2.5 j Inow 2 n Solid wall 1.5 n 0 1 i 1 0.5 j ne grid (1) 0.2 0.4 0.6 0.8 M2 0 0 1 coarse grid (2) 1 0.8 0.6 0.4 0.2 x M m c) How thick is the boundary layer thickness? m+2 1 hIf we dene the thickness of the oscillating layer as the position where u/U = 0.01 we get that V 2 Vh ek = 0.01 k 4.6 4.6 6.5 u uh d) Consider instead the oscillating ow U = U cos(t) over a stationary wall. This will simply result in a 0 change of frame of reference to one following the plate instead. If we consider the solution to the previous 1 problem and look at it in this new frame of reference we get : to be updated u(y, t) = U cos(t) U eky cos(ky t), where k = : new values 2 : old values i 1The solution now looks like this i 5 i+1 j 1 4.5 j 4 j+1 3.5 iteration direction x 3 y 2.5 n 2
1.5

1 i j 0.5 ne grid (1) 0 coarse grid (2) 1.5

0.5

0 x

0.5

1.5

Exercise 2
Consider a long hollow cylinder with inner radius r1 and a concentric rod with radiusr0 inside it. The rod is moving axially with velocity U0 . a) Find the velocity eld of a viscous uid occupying the space between the rod and the cylinder. Assumptions Steady ow:

=0 t u = uz (r)ez , = 0, z =0

Parallel ow and symmetry:

No streamwise pressure gradient:

p =0 z We can directly see that ur = u = 0 satisfy the two rst equations. The streamwise momentum equation reduces to (u )uz = 2 uz where (u
2

)uz = ur

uz u uz uz + + uz =0 r r z + 1 2 uz 2 uz 1 uz + = r 2 2 r z 2 r r r =0

uz =

1 uz r r r r

We end up with uz r r r Integrate twice r Boundary conditions uz (r0 ) = U0 uz (r1 ) = 0 = A ln r0 + B = A ln r1 + B B = A ln r1 A= U0 ln r0 r1 uz =A r

uz = A ln r + B

U0 = A(ln r0 ln r1 ) uz (r) =

ln rr1 U0 U0 ln r r0 ln r1 = U0 r0 ln r0 ln r1 ln r1 r1

b) With what force does one have to pull a rod with length L? Neglect end eects. Shear stress rz = Force F = 2r0 Lrz (r0 ) = uz U0 = r0 r r ln r1 2LU0 r ln r0 1

B.7

Axisymmetric ow and irrotational vortices

Axisymmetric ow
Consider incompressible and rotationally symmetric ow with no mass source at the symmetry axis. The velocity component in the direction of the axis of symmetry and the vorticity is given. uz = z and = (r) ez .

1) Compute ur (r, z) and u (r, z) for the two cases when a) (r) = 0 The vorticity is given, = Look at the dierent components: er : e : 1 uz u = er = 0 r z u = u (r) u= 1 r
r ur

and

b) (r) = 0 er er r e ez

/a2

r u

z uz

ur uz = e = 0 ur = ur (r) z r 1 (r u ) ur ez : = ez = (r) r r = 0. From the ez component we get, (r u ) = r (r) r (B.1)

The rotational symmetry implies

We also have the incompressibility condition: u= From this we get, 1 1 u uz + =0 (rur ) + r r r z

=0

(rur ) = r r Equation (2) is valid both for case a) and b). Case a) (r) = 0 (1) (2) (rur ) = r r (r u ) = 0 r u = A r r2 +B 2

(B.2)

integrate the rhs rur = ur =

r B + 2 r There should be no mass source at r = 0. The mass ow is,

2

Q=
0

ur r d = 2 B 2

r2 . 2

When r 0 then Q 2 B which gives B = 0. ur = r 2 for both a) and b)!

The circulation is, =

0

u r d = 2A A =

Case b) (r) = 0 er

/a

, (1)
2 2 (r u ) = r 0 er /a r

Integrate the right hand side: r u = Look at the circulation, (r) =

0

0 a2 r2 /a2 e +C 2
2

u =

0 a2 r2 /a2 C e + 2r r

u r d = 2( 0 = 2(

0 a2 r2 /a2 e + C) 2 C= 0 a2 + 0 2 2

This gives

for r = 0 we get

0 a 2 + C) 2

u =

2 2 0 a2 ++ 0 1 er /a 2r 2r

c) Consider circles C(t) following the ow with a radius R(t) and position Z(t). Compute R(t) and Z(t). Can any of the ow cases be inviscid? R(0) = R0 1 dR = ur (R) = R dt 2 dZ = uz (Z) = Z dt The circulation is then,
2

Z(0) = Z0 R(t) = R0 e 2 t Z(t) = Z0 e t

1

(t) =
0

u (R)R d = 2Ru (R) =

a)

The circulation is constant for a) but not for b). This means that the ow in a) can be inviscid.

0 + 0 a2 1 eR

/a2

b)

Inviscid and irrotational vortices

Consider a circular ow with u = u (r) . Which vortices are inviscid and which vortices are irrotational? e The NavierStokes equation for u u 1 1 u = + r t r r r r For inviscid ow we require, 1 u r r r r Make the ansatz u = rn , 1 1 (rnrn1 ) rn2 = n2 rn1 rn2 r r r We get the inviscid ow, u (r) = The vorticity is, = =0 Conclusion: Irrotational inviscid Inviscid irrotational u = {A.32} = (ru ) = 0 r Ar
solid body rotation

u r2

u =0 r2

n2 1 = 0 B r

n = 1

irrotational

1 (ru ) ez r r u = C r

Exercise 3
Show that the inviscid vorticity equation D = ( Dt reduces to the equation D Dt r in the case of axisymmetric ow: u = ur (r, z, t)r + uz (r, z, t)z e e The vorticity in an axisymmetric ow = u= ur uz z r

) u

=0

e = e

Study the right hand side of the inviscid vorticity equation ( )= r ( ) = u ur (r, z, t)r + uz (r, z, t)z e e r =

ur r uz e z e e r + ur e z + uz = ur e + r r r r r
=0 = e =0 =0

The left hand side of the inviscid vorticity equation gives D = + ( u Dt t ) = { = } = e + ur + uz e t r z

This gives that the inviscid vorticity equation now is + ur + uz = ur t r z r Multiply by 1 r

t r

1 ur + uz 2 ur = 0 r r z r and that uz 1 =0 z r 1 =0 r

Notice that

1 u = ur 2 r r r r

This means we can write t r And thus we have + 1 ur + uz + ur + uz r r z r z 1 ur + uz r r z D = r Dt r

t r

=0

B.8

Vorticity equation, Bernoulli equation and streamfunction

Solid body rotation

Consider the ow in a uniformly rotating bucket with velocity u = (r, 0, 0) a) Use Bernoulli equation to determine the free surface: What is wrong? Bernoulli: p 1 2 + |u| + gz = C along streamlines 2 This gives the surface of constant pressure z= C p0 2 r2 g 2g

The free surface is highest in the center of the bucket, something is wrong. Bernoulli theorem is valid along a streamline in a steady ideal uid. If the ow had been irrotational, it would have been valid everywhere. But now u = (0, 0, 2). b) Use Euler equation to determine the free surface: 1 Du = p+g Dt this gives er : u2 1 p = r r e : ez : 0 = Thus p(r, z) = This gives the free surface where p = p0 z= 1 p g z p = 2 r r 1 p r 1 2 2 r + f (z) 2

0=

p=

p =0 f = gz + p0

f = g z

1 2 2 r gz + p0 2 2 r2 2g

Flow over a hill (Exam 20020527 question 2)

A hill with the height h has the shape of a half circular cylinder as shown in Figure 1. Far from the hill the wind U is blowing parallel to the ground in the x-direction and the atmospheric pressure at the ground is p0 . a) (5) Assume potential ow and show that the stream function in cylindrical coordinates is of the form = f (r) sin , where f (r) is an arbitrary function. Calculate the velocity eld above the hill. b) (3) Derive an equation for the curve with constant vertical wind velocity V . c) (2) Assume that the density and the gravitational acceleration g is constant. Calculate the atmospheric pressure at the top of the hill. a) The stream function satises continuity: ur = The ow is irrotational: u=0 1 , r u = r

2 1 2 +r 2 + = 0 (1) r r r 2

ration direction x y 1 n x y i j ne grid (1) coarse grid (2)

Figure B.5: Streamlines above a hill with h = 100m and U = 5m/s. A paraglider pilot with a sink of 1m/s will nd lift in the area within the dotted line, while soaring along the hill. Introduce the ansatz = f (r) sin into equation (1): 1 f sin + rf sin f sin = 0 r 1 f + rf f = 0 r Make the ansatz f = r n : 1 nrn1 + rn(n 1)rn2 rn = 0 r n + n2 n 1 = 0 So we have = We need two boundary conditions. 1. Free stream: Ar sin = U r sin 2. Streamline on the hill surface: U h + So we have: = U r Now we can calculate the velocity eld above the hill: ur = 1 h2 = U 1 2 r r h2 = U 1 + 2 r r cos B =0 h h2 r B = U h2 sin A = U Ar + B r n = 1

sin

u =

sin

b) Constant vertical wind velocity is described by: V = ur sin + u cos = U 1 h2 r2 cos sin U 1 + r=h 2 h2 r2 sin cos = 2U h2 sin cos r2

U sin cos V

c) Use Bernoulli equation with free stream pressure p0 at the ground: 1 2 1 po + U = p + (2U )2 + gh 2 2 3 2 p = po U gh 2

Stokes stream function:

Consider a 2D incompressible ow u = 0 or Dene the stream function such that: u= This means y v= x u v + = 0. x y

u v 2 2 + = = 0, x y xy yx

so continuity is fullled. Now we can write ex u= And = Irrotational ow

2

z = e

ey
y

ez
z

0 ez
z

, ,0 y x

ex u=
x y

ey
y x

0, 0,

2 2 x y 2

z e

=0

In spherical coordinates for axisymmetrical ow, dene ur = Incompressibility is still valid u=0 Velocity u= Vorticity = Irrotational e r sin 1 sin =0 r2 1 sin u = 1 r sin r

1 2 sin + 2 r sin r2 r 1 sin

2 sin + 2 r2 r

Flow around a sphere

Consider a sphere with = 0 at r = a, compute the irrotational velocity distribution when the velocity of the freestream at innity is U: 1 r U r2 sin2 2 ur U cos u U sin Make the ansatz: = f (r) sin2 For an irrotational ow we get f From innity we get A= On the sphere 1 2 B Ua + =0 2 a = The slip velocity on the sphere is 2 f =0 r2 1 U 2 1 B = U a3 2 sin2 f = Ar 2 + B r

This gives

1 a3 U r2 2 r

3 u = U sin 2

The radial velocity is ur = 0.

B.9 Flow around a submarine and other potential ow problems : to be updated : new valuesExercise 1 : old values i 1The ow around a submarine moving at a velocity V can be described by the ow caused by a source and a sink with strength Q at a distance 2a from each other. i i+1 x V j1 j j+1 Submarine iteration direction x y Q -Q 1 p1,1 z n y y i j ne grid (1) coarse grid (2)
a

Figure B.6: Coordinate system for submarine problem a) If one wants to construct a pressure sensor that will register an approaching submarine at a distance L, what sensitivity is needed for the sensor? Assume an ideal uid and that 2a = 80 m, Q = 915 m3 /s, U = 8 m/s, L = 200 m and = 1000 kg/m3 .

Use a potential ow description u= , u=

, x

v=

The ow is always irrotational due to the denition of the velocity potential = For incompressibility we get u= = ( )i = = 0 xi Q Q + 4 r1 4 r2
source sink

u=

= 0,

curl(grad)=0

The equation is linear and thus superposition can be used. We have freestream plus 3D source plus sink = Uz
freestream

The rst term is in cylindrical coordinates (R, , z) and the two last are in two dierent spherical coordinate systems with origin in za and z+a, respectively. Transform the two second terms to cylindrical coordinates = Uz + Velocity 4 (z + a)2 + R2 Q + Q 4 (z a)2 + R2

eR

1 eR + e + ez = R R z QR QR Q(z + a) Q(z a) +z U + e 4((z + a)2 + R2 )3/2 4((z a)2 + R2 )3/2 4((z + a)2 + R2 )3/2 4((z a)2 + R2 )3/2 u= =

We need to know the distance, b, from the point source to the stagnation point on the submarine nose. Thus we need to know the length of the submarine.

b) How long is the submarine? Compute where uz = 0 for R = 0 U+ Q 4 1 1 (z + a)2 (z a)2 =0

Solving this system gives z = 43.01 m, z = 36.99 m, where the second solution lies inside the submarine. The length is then 2 43.01 86 m and b = 3.01 m. Now we continue to solve a) Use Bernoulli equation to determine the pressure uctuations at z = L b a, R = 0 1 1 p + ||2 = p + U 2 u 2 2 Evaluate u noticing that uR = 0 u(z = L b a, R = 0) = ez U + Q Q + 4(L + b)2 4((L + b + 2a)2

4U (z a)2 (z + a)2 4az = = 2 Q (z a)2 (z + a)2 (z a2 )2

Inserting the given values gives || = 7.99914 m/s and we get u p p = 1 2 (U 2 ||2 ) 6.86 N/m 0.07 mbar u 2

c) How wide is the submarine? To get this we need to compute the shape of the submarine. The stream function is constant along streamlines and is useful for this. In spherical coordinates the stream function is dened as ur = 1 r2 sin

1 r sin r Transformation between cylindrical and spherical coordinates u = ur = uR sin + uz cos R = r sin , z = r cos Our velocity eld gives 1 Q 1 1 = sin r sin 2 + r2 sin 4 (r2 + a2 + 2ar cos )3/2 (r + a2 2ar cos )3/2 cos U + Q 4 Q 4 r cos + a r cos a 2 (r2 + a2 + 2ar cos )3/2 (r + a2 2ar cos )3/2 (r2

= U cos + This is dicult to integrate.

r + a cos r a cos 2 2 + 2ar cos )3/2 +a (r + a2 2ar cos )3/2

p1,1 Simplify to a Rankine body by neglecting the sink and say that a = 0 Q 1 1 1 ur = U cos + = 2 2 4 r r sin 2 i 1 Q j = U r2 sin2 cos + C 2 4 Inow Solid wallDetermine C from the stagnation point n Q 2 0 ur ( = , r0 ) = 0 r0 = 4U 1 Since = 0 on the body we get Q M2 C = . 4 M mThe stream function is then 1 Q m+2 = U R2 sin2 (cos +1) 2 4 1 h source V d Vh The shape is given by = 0. As r , 0 then r sin . This gives 2 u uh 0 1 : to be updated 10 : new values : old values i1 5 i i+1 j1 j 0 j+1 iteration direction x y 5 1 n x y 10 i j ne grid (1) 15 15 10 5 5 10 15 0 coarse grid (2) 20

Figure B.7: Rankine body for submarine problem Q 1 d2 U 2 =0 2 4 4 4Q U Q U

d2 =

d=2

There is a simple way of determining the radius as z directly. The ow from the source must take up an particular area in the ow at innity. Since no uid can cross the streamlines this area must be equal to that of the Rankine body: Q = U d2 2 d=2 Q = 12.07 m U

uh We can use the computed stream function for a point source and displace it to z = a. In cylindrical 0 coordinates 1 Q z+a = 4 (z a)2 + R2 to be updated : new valuesTransform to spherical coordinates : old values Q Q r cos + a r cos + a i1 = = 4 4 2 r2 + a2 + 2ar cos (r cos + a)2 + r2 sin i i+1 j 1The stream function for the submarine is then j r cos a 1 Q r cos + a j +1 + = U r2 sin2 + 2 + a2 + 2ar cos 2 + a2 2ar cos 2 4 r r ration direction x y 1 n x y i j ne grid (1) coarse grid (2)

10

0
10 80 60 40 20

20

40

60

80

Figure B.8: Submarine body for submarine problem At r = R and = /2 we get = For the body = 0 and we get R2 Multiply by R2 R 2 + a2 + R 2 + a2 1 Q a a U R2 + + 2 + a2 2 4 R R 2 + a2 aQ =0 U aQ U

(R2 )3 + (R2 )2 a2 Computing this

a2 Q2 =0 2 U 2

d = 2R = 12.0006 m

m The complex potential m+2 1 hThe lines with constant stream function are the streamlines. They are orthogonal to the lines of constant V velocity potential which are equipotential lines. Since both of them satisfy Laplaces equation we can Vh dene a complex function F (z) = (x, y) + i (x, y) z = x + iy u uh 0 1 : to be updated 18 : new values : old values 16 i1 i 14 i+1 j 1 12 j j + 1 10 iteration direction x 8 y 1 6 n x 4 y i 2 j ne grid (1) 0 45 40 35 30 25 coarse grid (2) 50 Figure B.9: Complex potential for submarine problem, solid: , dotted: This is an analytical function since the CauchyRiemann equation holds = x y The velocity is then and = y x

dF = +i = u iv dz x x This enables the use of complex analysis, in particular conformal mapping that can be used to compute the ow over airfoil shapes. w(z) =

m Example: m+2 1 h Flow past a rotating cylinder centered at z = at an angle of attack V Vh (a + )2 i i F (z) = U (z + )ei + e log(z + ) u (z + ) 2 uh 1 0 2 1 Mapping by z = 1 Z + 1 Z 2 a2 gives an airfoil shape with the potential F(z). A correct ow is not 2 4 to be updatedachieved unless the KuttaJoukovski condition is satised requiring : new values = 4U (a + ) sin : old values i1 i i+1 Z=f(z) j 1 4 j 3 j +1 ration direction 2 2 x 1 y 0 0 n x y i j ne grid (1) coarse grid (2)
1 2 3 4 4 2 1 z=f (Z) 2 5

Figure B.10: Conformal mapping from circle to airfoil shape, (a = 3, = 0.5)

B.10

More potential ow

Half body over a wall

A line source of strength Q is located at (0, a) above a at plate that coincides with the x-axis. A uniform stream with velocity U ows along the x-axis. Calculate the irrotational ow eld. Method of images. Put a line source of equal strength at (0, a) in order to fulll the condition of no ow through the plate. Superposition of a uniform ow and the two line sources gives the complex potential F = Uz + Complex velocity W = dF Q =U+ dz 2 Q 2 1 1 + z ia z + ia Q Q ln(z ia) + ln(z + ia) 2 2

W =U+ W =U+ W =U+

1 1 + x + i(y a) x + i(y + a) x i(y a) x i(y + a) + 2 2 + (y a)2 x x + (y + a)2

Q 2

The velocity eld now becomes

Q x x ya y+a + 2 i 2 + 2 2 + (y a)2 2 2 2 x x + (y + a) x + (y a) x + (y + a)2 Q 2 x x + 2 x2 + (y a)2 x + (y + a)2

u=U+ v=

Q 2

ya y+a + 2 x2 + (y a)2 x + (y + a)2

0 1 Flow past a symmetric airfoil a) Use conformal mapping to calculate the irrotational ow eld around a symmetric airfoil. to be updated : new values Joukowski transformation : old values c2 (z) = z + i1 z i i+1 j 1 j 1 2 j +1 ration direction 1 x 0.5 y 0 0 n x 1 y 0.5 i j 1 2 ne grid (1) 3 2 1 2 1 1 0.5 0.5 0 1 0 coarse grid (2) 1.5 Equation for the circle z = + (a + )ei Equation for the airfoil = + (a + )ei + Complex potential in the z-plane F = U (z + )ei + U Complex velocity in the z-plane W = Complex velocity in the -plane = dF/dz = d/dz U ei U (a + )2 i i e + 2 (z + ) 2(z + ) 1 a2 z2 2 /2 a2 into (a + )2 i i dF = U ei U e + 2 dz (z + ) 2(z + ) a2 + (a + )ei

(a + )2 i i e + ln(z + ) (z + ) 2

The velocity can then be found by introducing the reversed transformation z = /2 + u = Re{(z)}, b) Calculate the Joukowski condition for the airfoil. v = Im{(z)}

The ow eld has a singular point at the trailing edge of the airfoil at = 2a. Resolve the singularity by choosing the circulation so the numerator vanishes at the trailing edge z = a U ei U ei + k = 4(a + )U i =0 2(a + )

ei ei = 4(a + )U sin 2i

B.11

Boundary layers

Exercise 1
Consider the high Reynolds number ow in a conversing channel. Compute the boundary layer over the surface at y = 0. Assume the free stream: U (x) = where Q is the ux. The boundary layer equations read: u u u Q 2u +v = 3 + 2 x y x y u v + = 0 (2) x y Seek a similarity solution where the dimensionless velocity u/U only depend on the dimensionless wall distance y y y y Q = = = = x x x
|U | Q

Q , x

(1)

The stream function satises (2): u= Make the stream function dimensionless: f () = Determine the terms in equation (1): u= f 1 Q = U = U f = U f = f y y x , y v= U x

v=

(U f ) f U = = U U f f = x x x x x Q Q Q Q x f + f 2x f = f x Q x x Q x Q x = Q f Q Q + 2f = f x x x x Q f x = Q Q + 2 f = 2 (f + f ) x x x Q f

u = x x

Q f x

u = y y

Q f Q 1 Q = f = 2 x y x x Q f Q2 = 3 f y x

2u Q = 2 y 2 x Insert into equation (1): Divide by Q2 /x3 : f f f f Boundary conditions: u(0) = 0 u() = 1
2

Q Q f (f + f ) + x x2

Q Q f 2 x x

Q Q2 Q2 f = 3 3 f x x

+ f f = 1 f f
2

+1=0 f (0) = 0 f () = 1

M2 Substitute F () = f (): M F F 2 + 1 = 0 m F (0) = 0 m+2 1 F () = 1 h V Solution: 2 Vh F = 3 tanh2 + 2 3 2 u uh 0 Exercise 2 1 Consider the ow downstream of a 2D streamlined body at high Reynolds number. Study the thin wake to be updateddownstream of the body where variations in y are much more rapid than variations in the downstream : new valuesdirection. Also assume that the wake is so small that we can write u = U + u1 where u1 is negative and : old valuesdescribes the wake. The length scale in x is L and in y it is with << L. U

i1 i i+1 j 1 j j +1 ration direction x y n x y i j ne grid (1) coarse grid (2)

U+u 1

Figure B.11: Coordinate system for wake problem

a) Find the governing (linear) equations: We start with the boundary layer equations since << L, u u 1 p 2u u +v = + 2 x y x y u v + =0 x y In the wake we have no pressure gradient, p = 0. x u1 x u1 << U L

From the continuity equation we get, v u = = U + u1 y x x = v

Inserting u = U + u1 in the boundary layer equations gives, U Neglect the quadratic terms, u1 u1 u2 1 x L and v u1 1 u2 u1 u1 1 y L L u1 u1 u1 2 u1 + u1 +v =0+ x x y y 2

 b) Show that

u1 2 u1 = x y 2

u1 dy = constant:

Consider the relation, Q1 =

u1 dy

which gives the linear contribution from u1 to the momentum ux. d d Q1 = dx dx

u1 dy =

u1 dy = {From the governing equation} = x

2 u1 u1 dy = U y 2 U y

=0

This means that Q1 is constant. c) Find a similarity solution for u1 : Seek a solution on the form, u1 (x, y) = F (x)f () This gives, Q1 = Q1 = constant requires, F (x) = 1/g(x) and This means, u1 (x, y) = Insert this into the equation of motion, U g (x) f () + f () g(x)2 = 1 f () g(x)3 1 y 1 f () = f g(x) g(x) g(x)

where =

y g(x) f () d

F (x)f () dy = F (x)g(x)

f () d = Q1 = constant

U f () g (x)g(x) = =C f () + f () g(x) = This gives the equation for f , f () + Cf () + Cf () = From the symmetry condition
y u(x, 0)

2C

x U

1/2

d f () + Cf () + D d

=0

= 0 we can determine that f (0) = D = 0. This gives, f () + Cf () = 0

Integrating this we get, f () = ae 2 C

1 2

We can determine the constant a from the condition that Q1 is constant,

f () d = Q1

f () =

2 1 C Q1 e 2 C 2

This now gives u1 : u1 (x, y) =

y2 C Q1 1 C g(x)2 e 2 = 2 g(x)

g(x) = d) Relate u1 to the drag FD :

2C

x U

1/2

Q1 = 2

U x

1/2

e 4x

U y2

The drag is FD =

u1 (U + u1 ) dy

Remember that u1 << U and neglect the u2 term, 1 FD = This means that Q1 = FD U

U u1 dy = U Q1 FD 2 Ux
1/2

u1 (x, y) =

e 4x

U y2

Exercise 3
Give an order of magnitude estimate of the Reynolds number for: i. Flow past the wing of a jumbo jet at 150 m/s ( Mach 0.5) ii. A wing prole in salt water with L = 2 cm and U = 5 cm/s iii. A thick layer of golden syrup draining of a spoon. iv. A spermatozoan with tail length of 103 cm swimming at 101 cm/s in water. Estimate the boundary layer thickness in case (i).

Fluid Water Air Syrup

cm2 /s 0.01 0.15 1200

i. Flow past the wing of a jumbo jet at 150 m/s ( Mach 0.5) U = 150 m/s, = 1.5 105 m2 /s, L = 4 m Re = ii. A wing prole in salt water with L = 2 cm and U = 5 cm/s U = 0.05 m/s, = 106 m2 /s, L = 0.02 m Re = iii. A thick layer of golden syrup draining of a spoon. U = 0.04 m/s, = 0.12 m2/s, L = 0.01 m Re = UL 0.003 UL 103 UL 4 107

iv. A spermatozoan with tail length of 103 cm swimming at 102 cm/s in water. U = 104 m/s, = 106 m2 /s, L = 105 m Re = UL 103

The boundary layer thickness in (i) is O(1 mm) 1 =O L Re

m+2 More boundary layers 1B.12 h V Drag on a circular cylinder Vh Use the integral form of the NavierStokes equations, Consider a x volume in a uid with velocity u. The ucontinuity equation becomes, uh d dV = uk nk dS. 0 dt V S 1 The momentum equation, : to be updated d ui dV = uj nj ui + pni ij nj dS + Fi dV : new values dt V S V : old values i 1Use this to compute the drag force on a cylinder: i U n U 0 0 i+1 j1 j j+1 iteration direction x l n d Wake y n n 1 x y i j ne grid (1) coarse grid (2)

n u = 0. Incompressible ow with constant density. The t

l/2

The cylinder is stationary and so is the ow continuity equation gives: uk nk dS = 0

S

U0 l Lv|y=l/2 + Lv|y=l/2 +
l/2

Uw (y) dy = 0
l/2

Due to the symmetry we get, U0 l + 2Lv|y=l/2 + The momentum equation in x:

S

Uw (y) dy = 0
l/2

nj uj ux + nx p nj xj dS = 0
l/2

Far away from the cylinder we can assume p = p0 and xx = xy = 0. This gives,
2 U0 l + 2U0 v|y=l/2 L + 2 Uw (y) dy +

l/2

cyl

[nx p nj xj ] dS
FD

Multiply the continuity equation by U0 ,

l/2 2 U0 l + 2U0 v|y=l/2 L = U0

Uw (y) dy
l/2

Insert into momentum equation:

l/2 l/2

U0
l/2

Uw (y) dy +
l/2 l/2

2 Uw (y) dy + FD = 0 l/2 l/2

FD =
l/2

2 2 U0 Uw (y) Uw (y) dy = U0

Uw (y) U0

Uw (y) U0

dy

Make the variable substitution r = y/d,

2 FD = U0 d l/2d l/2d

Uw (r) U0

Uw (r) U0
2

dr

Now if d << l we get,

2 FD = U0 d

Uw (r) U0

Uw (r) U0

2 dr = U0

The momentum thickness is called and is a measure of the loss of momentum in the ow. Now compute FD the drag coecient CD = 1 2 = 2 d U0 d 2

Glauert wall jet

Compute an ODE for a self-similar wall jet. The boundary layer equations read: u u u 2u +v = 2 x y y (1)

u v + = 0 (2) x y Seek a similarity solution where the dimensionless velocity u/Um(x), where Um (x) is the jet core velocity, only depend on the dimensionless wall distance y = g(x) The stream function satises (2): u= Make the stream function dimensionless: f () = Determine the terms in equation (1): u= v= f 1 = Um g = Um gf = Um f y y g yg g2 = Um gf Um g f + Um g f Um (x)g(x) , y v= x

(Um gf ) = = Um gf Um g f Um gf x x

u g = (Um f ) = Um f Um f x x g u 1 = (Um f ) = Um f y y g 2u = y 2 y Insert into equation (1): Um f (Um f Um Um Um f

2 2 Um

Um f

1 g

= Um f

1 g2

g 1 1 f ) + (Um g f Um gf Um g f )Um f = Um 2 f g g g

g 1 2 g 2 g f f + Um f f Um Um f f Um f f = Um 2 f g g g g f + Um g 2 Um g g +

ff

Um g 2 f

=0

+ f f + f

=0

Boundary conditions:

Similarity requires that and are constant. Assume the x-dependence: Um = axm , Then we get g = bxn

u(0) = 0 v(0) = 0 u() = 0

f (0) = 0 f (0) = 0 f () = 1

1 ab2 = amxm1 b2 x2n = m, =

2n + m = 1

1 1 ab2 amxm1 b2 x2n + axm bnxn1 bxn = (m + n) We can choose the length scale g such that = 1: ab2 1 = m+n So we have: f + f f + f
2

m m+n

= 0 (3)

For what is the boundary conditions fullled? Work on (3): f f Integrate: f + f f + ( 1)g = 0, Multiply with f : f f + ff f f + (f f Integrate: 1 f 2
2 2 2

+ f f + f

=0

+ (f f + f 2 ) + ( 1)f g() =

=0 f 2 d > 0

+ ( 1)gf = 0

+ gf ) + ( 2)gf = 0

+ f g + ( 2)

gf d = 0

Put in = 0 and use the boundary conditions: ( 2) = Solve for m and n:

0

gf d = 0

=2

m =2 m+n 1 m= , 2

3m + 2n = 0 3 4

n=

B.13

Introduction to turbulence

Exercise 1
Compute the far-eld two-dimensional turbulent wake U (x, y) behind a cylinder. Let U1 (x, y) = U0 U (x, y), Assume Us Continuity U0 U1 U0 , u, v Us , x L, yl V Us l L Us (x) = U1 (x, y = 0)

U V + =0 x y

V U1 = (U0 U1 ) = y x x
V l Us L

The turbulent boundary layer momentum equation read: U U U 1 P0 2U +V = + 2 (uv) x y x y y

Disregard the pressure gradient and insert U = U0 U1 : (U0 U1 ) Neglect small terms: U0 Self-similar hypothesis: U1 (x, y) = f () Us (x) uv = g() 2 Us where = y l(x) y = 2l = l , x l l 1 = y l U1 x V U1 2 U1 = (uv) y y 2 y

U1 = (uv) (1) x y

Determine the terms in equation (1): l U1 = (Us f ) = Us f + Us f = Us f Us f x x x l 2 2 21 (uv) = (Us g) = Us g = Us g y y y l Insert into equation (1): l 21 U0 (Us f Us f ) = Us g l l U0 lUs U0 l g = f+ f 2 Us Us Boundary conditions: U1 () = 0 U1 () = 0 f () = 0 f () = 0

Similarity requires: lUs = const 2 Us Assume the x-dependence: Us = Axm , Then we get Bxn Amxm1 = const x2m n + m 1 = 2m l = Bxn m=n1 l = Bxn l = const Us

Us = Axn1 ,

We need one more condition. The momentum loss thickness is constant: =

U U0

U U0 1 U0

dy =

U0 U 1 U0

U0 U 1 U0

dy =

U1 U0

U1 dy U0

U1 dy =

1 CD d = const (Recitation 12) 2

1 U0

U1 dy = Us l

f d = const

const Us l = Axn1 Bxn = const

1

2n 1 = 0 l = Bx 2
1

n=

1 2

Us = Ax 2 ,

For a suciently large Reynolds number, the wake will be turbulent. Rel = A turbulent wake will remain turbulent! To determine the wake-velocity prole, we need to model the turbulent shear stress. E.g. uv = T
2 Us g = T

AB lUs = = const

U y

1 l

U1 y

= T Us f g = T f Us l

g=

T f Us l

T U0 lUs U0 l f = f+ f 2 Us l Us Us U0 l l U0 l 2 Us f f =0 T T Us U0 B 2 (f + f ) = 0 (2) 2T

f +

f +

Equation (2) can be solved together with the boundary conditions f () = 0

B.14

Old Exams i
j

Exam 031023 Inow

Solid 1. Relative motion.wall n Consider the relative motion of two uid particles initially separated by the distance dx 0 . (Here x0 i i 0 and t are the Lagrangian coordinates). 1 a) (5) Show that the separation at time dt is M2 ui M dri (dt) = dx0 + 0 dx0 dt. i m xj j m+2 b) (2) Use this relation to write down the expression for the deformation of the sides of a small cube 1 aligned with the coordinate directions. Thus, show that the components of the side aligned with the h x-direction, Ri1 , is deformed into V Vh ui u (1) dri = R(i1 + 0 dt), uh x1 0 with analogous expressions for the other two sides. 1 c) (4) Show that the deformation rate of the volume of the cube is given by the divergence of the : to be updated velocity eld. Note that x0 = xi at t = 0. i : new values 2. Flow between values and a cylinder. : old a rod 1 Consider a long ihollow cylinder with radius b, a concentric rod with radius a inside the cylinder, and i water occupying the space between the rod and the cylinder, see gure B.14. The rod is rotating i+1 with the constant angular frequency . j1 (10) Find the velocity eld of the water driven by the rotating rod and the gravitational acceleration j g. j+1 iteration direction z x y 1 n x g y i j ne grid (1) a b r coarse grid (2) Figure B.12: Water and a concentric rotating rod inside a cylinder.

j Inow Solid wall n 0 1 M2 M m m+2 1 h V Vh u uh 0 1 : to be updated : new values : old values i1 i i+1 j1 j j+1 iteration direction x y n x y i j ne grid (1) coarse grid (2)

3. Bernoullis equation a) (3) Derive the following formula: uj ui 1 = (uj uj ) + xj 2 xi

ijk j uk .

b) (3) Use this in the momentum equation to derive Bernoullis equation for unsteady potential ow. c) (3) Use the result in (a) to derive Bernoullis equation for steady inviscid ow. Discuss the validity of the derived relation. 4. Laminar wake ow. Consider the laminar wake ow downstream of a two-dimensional streamlined body at high Reynolds number, see gure B.13. Assume that the wake is so weak that we can write u = U + u 1 where u1 is negative and much smaller than the free-stream velocity U . a) (3) Motivate the usage of the equation U u1 2 u1 = . x y 2

b) (2) Show that the mass ux Q1 , in the direction of the wake is constant in x Q1 =

u1 dy. ()

c) (7) A similarity solution can be found by scaling the wake velocity with Us (x) = u1 (x, 0) so that f () = u1 (x, y)/Us (x), where = y/(x) is the similarity variable. Show, using (), that Us = k/(x), where k is a given constant. Find the similarity solution for u1 . U
Y U+u

Figure B.13: Wake downstream of a at plate. 5. Mean heat equation. The heat equation for the instantaneous turbulent ow is T T 2T + uj = . t xj xj xj (8) Derive the heat equation governing the mean ow.

Answers to exam in Fluid mechanics 5C1214, 2003-10-23

1. See Kundu & Cohen p. 57. 2. Flow between a rod and a cylinder. u = 1g 4 a2 b2 a 2 b2 r 2 r ln(r/a) ln(b/a)

uz = 3. See Kundu & Cohen pp. 110-114. 4. Laminar wake ow.

r2 a2 (b2 a2 )

u1 = 5. See Kundu & Cohen pp. 511-512.

Q1 2

U U y2 e 4x x

Exam 040113

0 1 1. (5) Consider the unsteady ow, M2 u = u0 , v = kt, w = 0, M where u0 and k are positive constants. Show that the streamlines are straight lines, and sketch them m at two dierent times.2Also show that any uid particle follows a parabolic path as time proceeds. m+

2. (5) For surface waves1on water with nite depth h the phase speed is given by the relation c 2 = g h k tanh(kh). Compare the group velocity to the phase speed in the limit of long and short waves. V of a wave packet and explain how it is moving. Make a rough 2D sketch Vh 3. Consider a thin layer u liquid in contact with a solid at vertical wall. See gure 1. The wall is of u temperature gradient with increasing temperature in the downward direction. subject to a constant h 0 Due to a varying surface tension, as a result of the temperature gradient, a constant shear stress at the free surface 1 acting upward. At the same time, the gravitational acceleration g is acting is downward. : to be updated a) (7) Assume constant layer thickness h and calculate the velocity eld of the liquid layer. : new values b) (3) For what layer thickness h is the ux zero? : old values i1 z i i+1 j1 j j+1 iteration direction x g y 1 n x y i x h j ne grid (1) coarse grid (2) Figure B.14: Liquid lm on vertical wall driven by gravity and surface tension.

4. a) (3) Show that uj ui 1 = (uj uj ) + xj 2 xi

ijk j uk

b) (4) Derive the vorticity equation starting with the NavierStokes equation for incompressible ow. c) (3) Show that the following relation holds for incompressible ow and discuss its implication for inviscid ow ij = xj 5. Consider the ow over an oscillating plate. a) (3) Motivate that the solution can be written as [u(y, t), 0, 0] and that it thus satises the equation 2u u = 2 t y b) (7) Show that u(y, t) = U eky cos(ky t), where k = . 2
ijk

k xj

6. a) (5) Derive the Reynolds average equation valid for turbulent ow. b) (5) Assume that the Reynolds stress can be modeled as a turbulent viscosity and introduce this into the Reynolds equation and simplify.

j1 A B C 3 u 2 ,1 v1, 3 2 p1,1 1

Solutions to exam in Fluid mechanics 5C1214, 2004-01-13

1. Streamline (x t): dx = u = u0 ds dy = u = kt ds Then we see that y(x) is a linear function, y= x = u0 s + a y = kts + b

i j Inow Solid wall n 0 1 M2 M m m+2 1 h V Vh u uh 0 1 : to be updated : new values : old values i1 i i+1 j1 j j+1 iteration direction x y n x y i j ne grid (1) coarse grid (2)

kt x + d(t) u0

x = u = u 0 x = u 0 t + x0 t y 1 = u = k t x = k t2 + y 0 2 t Then we see that y(x) is a parabola, y= 1 k 2 (x 2x0 x + x2 ) + y0 0 2 u2 0

Particle path (x x0 ):

2. The wave number is dened as k = 2/, where denotes the wave length of the wave. For long waves, go to the limit k 0. For short waves, go to the limit k . Phase speed: c= Short waves:
x

= k lim

g tanh(kh) = k

gh

tanh(kh) kh g k gh

tanh(x) 1 = x x tanh(x) =1 x d c = dk 2

c=

Long waves:
x0

lim

c=

Group velocity: cg = Short waves:

x

1+

2kh sinh(2kh) cg = c 2

lim

x =0 sinh(x) x =1 sinh(x)

Long waves:
x0

lim

cg = c

Longer waves travel faster. 2

1.5

1
0.5

0
0.5 1 1.5

0.5

1.5

2.5

3. a) Assume steady, parallel ow u = w(x)z and solve the z-momentum equation with boundary e conditions: 2w w 2 g = 0, w(0) = 0, (h) = x x Integrate the equation once and introduce the second boundary condition: w g = x + A, x w gh (h) = +A = x A= 1 gh

Integrate the equation once more and introduce the rst boundary condition: w= So we get: w= 1g 2 1 x + 2 gh x 1g 2 x + Ax + B, 2 w(0) = 0 B=0

b) The ux per unit width of the wall can be written

h

Q=
0

w dx =

1g 3 11 x + 6 2

gh x2 3 2 g

=
0

11 2 1g 3 h h 2 3

Zero ux gives: h=

4. a) Begin with the second term on the right hand side:

ijk j uk

ijk jlm

um ui uk (um )uk = (kl im km il )uk = uk uk xl xl xk xi

ijk j uk

uj

ui uj = uj + xj xi

uj

ui 1 = (uj uj ) + xj 2 xi

ijk j uk

b) NavierStokes momentum equation for incompressible ow: ui ui 1 p 2 ui + uj = + t xj xi xj xj Use the expression from 4. a): ui 1 + (uj uj ) + t 2 xi i + t i + t ( xm
ijk j uk

1 p 2 ui + xi xj xj 2 i xj xj

Take the curl of the equation. The curl of a gradient vanish:

imn njk j uk )

nim njk

2 i (j uk ) = xm xj xj

i 2 i + (ij mk ik mj ) (j uk ) = t xm xj xj i 2 i + (i uk ) (j ui ) = t xk xj xj xj Use continuity. We end up with the vorticity equation: i i ui 2 i + uj = j + t xj xj xj xj c) Right hand side:
ijk

k = xj

ijk

xj

klm

um = xl

kij klm

2 um = xj xl
2

(il jm im jl ) Left hand side:

2 um 2 uj 2 ui = = xj xl xj xi xj xj ui uj + = xj xi
ijk 2

ui

ij = xj xj

ui

ij = xj Irrotational ow is not subject to viscous forces.

Thus:

k xj

5. a) Consider the NavierStokes equation in the x direction: u u u u 1 p 2u 2u 2u +u +v +w = + + 2 + 2 t x y z x x2 y z With the current velocity eld only terms with y derivatives will remain since there can be no change in the other directions, u 2u = 2 t y b) Make the ansatz: u = Insert into the equation: if (y)ei t = f (y)eit f (y) 2 Introduce k = , 2 f (y) = Aeyk(1+i) + Beyk(1+i) , We have u= Beyk(1+i) eiwt = Beky ei(tky) = Beky cos(t ky) = Beky cos(ky t) f (y) 0 as y A = 0 i f (y) = 0 with f (y) = ey i = gives 1+i 2 f (y)ei t = f (y) cos( t).

i =0

Boundary condition at y = 0 At y = 0 we have u = U cos(t) B = U So we have, u(y, t) = U eky cos(ky t), where k = 2

6. a) Turbulent ow is inherently time dependent and chaotic. However, in applications one is not usually interested in knowing full the details of this ow, but rather satised with the inuence of the turbulence on the averaged ow. For this purpose we dene an ensemble average as 1 N N
N (n)

Ui = ui = lim where each member of the ensemble ui

(n)

ui
n=1

is regarded as an independent realization of the ow.

We are now going to derive an equation governing the mean ow Ui . Divide the total ow into an average and a uctuating component ui as ui = u i + u i p=p+p

and introduce into the Navier-Stokes equations. We nd ui ui ui u ui u 1 p 1 p + + uj + uj i + uj + uj i = + t t xj xj xj xj xi xi We take the average of this equation, using ui = 0 which gives uj ui = ui uj = 0
2

ui +

ui

where the average of the continuity equation also has been added. Now, let U i = ui , as above, and drop the . We nd the Reynolds average equations Ui + Uj Ui = 1 P + (ij ui uj ) t xj xi xj Ui =0 xi

ui + uj ui = 1 p + t xj xi ui =0 xi

ui

u i uj xj

where ui uj is the Reynolds stress. b) Assume the Reynolds stress:

ui uj =

2 Kij 2T eij 3

The rst term on the right hand side is introduced to give the correct value of the trace of u i uj . Here the deformation rate tensor is calculated based on the mean ow, i.e. eij = 1 2 Ui Uj 2 Ur + ij xj xi 3 xr

where we have assumed incompressible ow. Introducing this into the Reynolds average equations gives Ui Ui + Uj = t xj xj 1 P P 2 + k ij + 2 ( + T ) eij = + ( + T ) 3 xi
2

Ui

where the last equality assumes that T is constant, an approximation only true for very simple turbulent ow. It is introduced here only to point out the analogy between the molecular viscosity and the turbulent viscosity.

0 Exam 041018 1 1. The kinetic energy dT of an innitesimal volume element, with density , which is rotating with : to be updated angular velocity can be expressed : new values 1 1 : old values dT = v v dV = ( x) ( x) dV, 2 2 i1 i where v is the velocity of the volume element and x is the vector from the center of rotation O to i+1 the volume element, see gure B.15. j1 j x3 j+1 dV iteration direction x x y C x 1 x2 n x3 x1 x z y i j O ne grid (1) x2 coarse grid (2)x1 Figure B.15: Coordinate system and denitions a) (5) Show that the total kinetic energy of the body can be expressed as T =
V

dT =

1 Jjl j l , 2

where Jjl is the moment of inertia tensor with respect to the origin of the unprimed system. It is dened as Jjl = (xk xk jl xj xl ) dV.
V

b) (5) Show that you can relate Jjl to the moment of inertia tensor with respect to the center of mass C of the body Jjl in the following way
C Jjl = (zk zk jl zj zl )M + Jjl , C Jjl =

(xk xk jl xj xl ) dV,

where M is the total mass of the body and the primed coordinates represent the coordinate system centered at the center of mass C of the body. Hints: Integrals of the type V xk dV vanishes as a result of the denition of the center of mass. The two coordinate systems are related by the formula x = z + x .

2. (10) Consider the laminar ow of a viscous uid down a vertical cylindrical pipe with radius a. The ow is driven by the gravitational acceleration g. Calculate the velocity eld. 3. Potential ow and conformal mapping a) (2) Show that the Kutta-Joukowski transformation = z + a2 /z maps a circle with radius a to a at plate with length 4a. b) (2) Write down the complex potential of the ow around a circular cylinder with radius a at angle of attack and with clockwise circulation . The uniform far-eld velocity has the magnitude U . c) (4) Show that a necessary condition for nite velocity at the trailing edge of the at plate is (Kutta condition) = 4U a sin . d) (2) How should one modify the circle in order to obtain a nite velocity at the leading edge? 4. Blasius boundary layer a) (2) Write down the boundary layer equations. b) (8) Make appropriate assumptions for similarity and reduce the boundary layer equations to an ordinary dierential equation for the boundary layer over a at plate (i.e there is no pressure gradient in the free stream). Also state the boundary conditions for the similarity function. c) (2) Sketch the streamwise and normal velocity proles. 5. Turbulent ow a) (2) Dene the velocity and length scales appropriate for turbulent ow near a wall, using the wall shear stress w , the viscosity and the density . b) (2) What is the law of the wall and what is the explicit functional dependence of the turbulent mean velocity in the intermediate region between the wall and the outer ow? c) (4) In two-equation turbulence models, partial dierential equations for the turbulent kinetic energy k and the dissipation rate are solved. In those equations, the turbulent viscosity T = uT l, where uT and l are the turbulent velocity and length scales, respectively. Use dimensional analysis to determine uT , l and T in terms of k and .

Solutions to exam in Fluid mechanics 5C1214, 2004-10-18

1. a) The kinetic energy of an innitesimal volume element dT = dT = 1 1 v v dV = ( x) ( x) dV 2 2

1 1 1 vi vi dV = ijk j xk ilm l xm dV = (jl km jm kl )xk xm j l dV = 2 2 2 1 (jl xk xk xj xl )j l dV 2 T =

V

dT =

1 2

dT (jl xk xk xj xl ) dV j l = (jl xk xk xj xl ) dV.

1 Jjl j l , 2

where Jjl =
V

b) The moment of inertia tensor Jjl =

V

(jl xk xk xj xl ) dV =

[jl (zk + xk )(zk + xk ) (zj + xj )(zl + xl )] dV =

[jl (zk zk + 2zk xk + xk xk ) (zj zl + zj xl + zl xj + xj xl )] dV = (zk zk jl zj zl ) 2zk jl

V

dV +
V V

(xk xk jl xj xl ) dV + xj dV.

xk dV + zj

xl dV + zl

The three last terms vanishes due to the denition of the center of mass. So we have
C Jjl = (zk zk jl zj zl )M + Jjl ,

where M=
V

dV

C and Jjl =

(xk xk jl xj xl ) dV.

2. Poiseuille ow ( Pipe ow ) Use the NavierStokes equations (with no volume force) in cylindrical coordinates ur + (u t u + (u t )ur )u + u2 1 p = + r r
2

ur
2

ur 2 u 2 2 r r 2 ur u 2 r2 r

ur u 1 p = + r r )uz =

u +
2

uz + (u t

1 p + z

uz

1 1 u uz (r ur ) + + = 0. r r r z p p We know that = 0 and = 0 and can directly see that ur = u = 0 satisfy the two rst r equations. From the continuity equation we get uz =0 z uz = uz (r, ) only.

Considering a steady ow we get from the NavierStokes equations (u )uz = uz uz z uz uz = z

2

uz g

But we know that

uz = 0 from the continuity equation. We get z

2

uz =

1 uz g (r )= r r r

uz g (r )= r r r

Integrate r Integrating again we get uz =

uz 1g 2 = r + c1 r 2

uz 1g c1 = r+ r 2 r

1g 2 r + c1 ln r + c2 4

using the boundary conditions uz (r = 0) < c1 = 0 ga2 and we have 4

We also have uz = 0 at r = a and this gives c2 = uz = 3. Potential ow and conformal mapping a) Equation for the circle

g 2 (a r2 ) 4

z = aei Equation for the plate =z+ a2 ei + ei = aei + aei = 2a = 2a cos , z 2

which is a real function taking the values from 2a to 2a, thus a at plate with length 4a. b) Complex potential for a circular cylinder a2 i i e + ln z z 2

F = U zei + U c) Complex velocity for the circular cylinder W = Complex velocity in the -plane = dF/dz = d/dz U ei U

dF a2 i = U ei U 2 ei + dz z 2z

a2 i i e + 2 z 2z

a2 z2

The ow eld has singular points at the leading and trailing edges of the at plate. Resolve the singularity at the trailing edge by choosing the circulation so the numerator vanishes at z = a U ei U ei + k = 4U a i =0 2a

ei ei = 4U a sin . 2i

d) Move the point z = a inside of the circle by putting the center at z = . We still want a sharp trailing edge so the circle crosses the real axis at z = a. The radius of such a circle is a + and the circle is described by z = + (a + )ei 4. Blasius boundary layer a) Boundary layer equations u u u dU 2u +v =U + 2 x y dx y u v + =0 x y (B.3) (B.4)

b) Seek a similarity solution where the dimensionless velocity u/U , where U is the constant free-stream velocity, only depends on the dimensionless wall distance = The stream function satises (2): u= Make the stream function dimensionless: f () = Determine the terms in equation (1): u= v= f 1 = U = U f = U f y y y 2 = U f U f (x, y) U (x) = U f , y v= x y (x)

(U f ) = = U f U f x x

u = (U f ) = U f x x u 1 = (U f ) = U f y y 2u = 2 y y U Insert into equation (1): U f (U U 2 1 1 f ) + (U f U f )U f = U 2 f Uf 1 =U 1 f 2

dU =0 dx

1 f f + U 2 f f U 2 f f = U 2 f U f + ff = 0

Similarity requires that is a constant. We can integrate it with respect to x in order to nd U = 1 2 x = 2 U (x) = 2x U

The value of the constant is just a matter of scaling, we can choose it to be 1/2 so (x) = The ordinary dierential equation becomes f We have the boundary conditions u(0) = 0 v(0) = 0 u() = U 1 + f f = 0. 2 f (0) = 0 f (0) = 0 f () = 1. x . U

c) Figure B.16 shows the dimensionless streamwise and normal velocity proles.

2 1 2

i+ i+1 i+ 3 2 j+1 2 j j1 2 j 1 A B C 3 u 2 ,1 v1, 3 2 p1,1 1

5. Turbulent ow a) Velocity scale near the wall: u = Length scale near the wall: l = b) Law of the wall U+ = U = f (y + ), u where y + = y l w u

Functional dependence in the intermediate region U+ = where is the Karmans constant. c) Dimensions for k and : 1 ln y + + B,

i j Inow Solid wall n 0 1 M2 M m m+2 1 h V Vh u uh 0 1 to be updated : new values 7 : old values i1 6 i i+1 j 15 j j +1 4 ration direction x y3 n2 x y i1 j ne grid (1) 0 coarse grid (2) 0

Dimension analysis for uT :

1 k = 2 ui ui [L2 T 2 ] u u = i j [L2 T 3 ] xj xi uT k m n [LT 1 ]

LT 1 = L2m T 2m L2n T 3n 2m + 2n = 1 2m 3n = 1 uT Dimension analysis for l: l k m n k [L]

n=0 m= 1 2

L = L2m T 2m L2n T 3n 2m + 2n = 1 2m 3n = 0

n = 1 3 m= 2

0.2

0.4

0.6

0.8

Figure B.16: Velocity proles of Blasius boundary layer.

l So we have

3 k2

3 2 k T = uT l k k2 T T = C k2 .

or

Appendix C

Study questions 5C1212

1. Dene and use the Rankine-Hugoniot jump condition to compute the shock speed for the following problem ut + uux = 0 < x < , t>0

u(x, 0) =

1 x0 0 otherwise

2. Dene the entropy condition for a scalar conservation law. ut + f (u)x = 0 < x < , t>0

with a convex ux function f (u). The shock is moving with speed s and the state to the left is given by uL and the state to the right by uR . Why do we need an entropy condition ? 3. Write a dierence approximation (in 1D) on conservative form and dene the notation. 4. Solve ut + ( u2 )x = 0, 2 u(x, 0) = 1 x<0 0 x0 (C.1)

by the upwind scheme on the following form, un+1 = un j j b) If not, suggest another way of solving C.1. 5. Dene a total variation decreasing (TVD) method. Why is this a desirable property ? 6. Investigate the one-sided dierence scheme un+1 = un a j j for the advection equation ut + aux = 0 Consider the cases a > 0 and a < 0. a) Prove that the scheme is consistent and nd the order of accuracy. Assume k/h constant. b) Determine the stability requirement for a > 0 and show that it is unstable for a < 0. 7. Apply Lax-Friedrichs scheme to the linear wave equation ut + aux = 0. a) Write down the modied equation. b) What type of equations is this? c) What kind of behavior can we expect from the solution? 173 t n (u un ) j1 x j t n n u (u un ) j1 x j j

a) Will the numerical solution converge to the analytical solution?

8. A the three-point centered scheme applied to ut + aux = 0, yields the approximation un+1 = un + j j a > 0.

at (uj+1 uj1 ) 2x

Show that this approximation is not stable even though the CFL condition is fullled. 9. What does Laxs equivalence theorem state? 10. What is the condition on the n n real matrix A(u) for the system ut + Aux = 0 to be hyperbolic ? 11. The barotropic gas dynamic equations t + ux = 0 1 ut + uux + px = 0 where p = p() = C i and C a constant, can be linearized by considering small perturbations ( , u ) around a motionless gas. a) Let = 0 + and u = u0 + u where u0 = 0. Linearize the system (C.2) and show that this yields the following linear system (the primes has been dropped) t + 0 ux = 0 a2 ut + x = 0 0 where a is the speed of sound. a and 0 are constants. b) Is the system given by (C.3) a hyperbolic system? Motivate your answer. c) Determine the characteristic variables in terms of and u. d) Determine the partial dierential equations the characteristic variables fulll - characteristic formulation. e) Given initial conditions at t = 0 and let < x < (no boundaries) (0, x) = sin(x) u(0, x) = 0 determine the analytical solution to (C.3) for t > 0. Hint: Start from the characteristic formulation. 12. The linearized form of (C.3) is given by u +
t

(C.2)

(C.3)

0 a2 /0
A

0 0

= 0,
x

(C.4)

where a is the speed of sound. a and 0 are constants. a) Draw the domain of dependence of the solution to the system (C.4) in a point P in the x-t plane. b) The system is solved numerically on a grid given by xj = jx, j = 0, 1, 2... and tn = nt, n = 0, 1, 2, ..... using an explicit three-point scheme, see the gure below. Draw the domain of dependence of the numerical solution at P (in the same gure as a)) of the three-point scheme in the case when i) the CFL condition is fullled ii) the CFL condition is NOT fullled. Assume that P is a grid point.

1t n+1 x y i t ne grid (1) n coarse grid (2) 13. To solve Euler equations in 1D

j1

j+1

t + ux + ux = 0 1 ut + uux + px = 0 pt + c2 ux + upx = 0 How many boundary conditions must be added at (motivate your answer) inow boundary when the ow is a) Supersonic b) Subsonic outow boundary when the ow is c) Supersonic d) Subsonic 14. Describe the ideas behind a ux splitting scheme for solving a non-linear hyperbolic system of equations, Ut + F (U)x = 0 15. a) Show that a vector eld wi can be decomposed into wi = u i + p xi

where u is is divergence free and parallel to the boundary. b) Apply this to the Navier-Stokes equations, show that the pressure term disappears and recover an equation for the pressure from the gradient part. 16. From the dierential form of the Navier-Stokes equations obtain a) the Navier-Stokes equations in integral form used in nite-volume discretizations, b) a variational form of the Navier-Stokes equations used in nite-element discretizations. 17. a) Write down the appropriate function spaces for the pressure and velocity used to dene weak solutions to the Navier-Stokes equations. b) Explain the concept of essential and natural boundary conditions. 18. a) How is a nite-element approximation dened? b) Explain how to convert a FEM discretization to an algebraic problem, e.g. that the Navier-Stokes equations yield K + C(u) G GT 0 u p = f 0

19. a) By choosing zh = uh in the nite element approximation of the Stokes problem, show that any Galerkin velocity solution is stable. b) Describe and interpret the LBB condition.

kk ll

g hgh opopop opop opop opop mnm op

jij i

20. Choice of elements. Discrete elements pairs a) constant pressure-bilinear velocities, b) the Taylor-Hood pair, c) a stable choice with piecewise linear velocities. 21. a) Derive the nite volume (FV) discretization on arbitrary grids of the continuity equation (u i /xi = 0), b) derive the FV discretization for Laplace equation on a Cartesian grid, c) show that both are equivalent to a central dierence approximation for Cartesian grids. 22. Derive the nite element (FEM) discretization for Laplace equation on a Cartesian grid and show that it is equivalent to a central dierence approximation. 23. Iterative techniques for linear systems. a) Dene Gauss-Seidel iterations for the Laplace equations, b) Dene the 2-level multigrid method for the Laplace equation, 24. State the diculties associated with the the nite-volume discretizations of the Navier-Stokes equations on a colocated grid? and show the form of the spurious solution which exist. 25. a) Dene an appropriate staggered grid that can be used for the discretization of the Navier-Stokes equations, b) write down the FV discretization of the Navier-Stokes equations on a staggered cartesian grid, c) discuss how to treat noslip and inow/outow boundary conditions. 26. Time dependent ows. a) Dene a simple projection method for the time dependent Navier-Stokes equations d dt u 0 + N(u) G D 0 u p = f g

b) show in detail the equation for the pressure to be solved at each time step and discuss the boundary conditions for the pressure. 27. Time step restriction for Navier-Stokes solutions. a) Motivate the use of an appropriate form of the advection-diusion equation as a model equation for stability analysis, b) derive the time step restrictions for the 1D version of that equation, c) state the 2D equivalent of that restriction.

Bibliography
[1] D.J. Acheson, 1990, Elementary Fluid Dynamics, Oxford University Press. [2] J.D. Anderson, Jr., 1995, Computational Fluid Dynamics, The basics with applications, Mc Graw-Hill. [3] C.A.J. Fletcher, 1991, Computational Techniques for Fluid Dynamics Volume 1, Second Edition, Springer-Verlag. [4] C.A.J. Fletcher, 1991, Computational Techniques for Fluid Dynamics, Volume 2, Second Edition, Springer-Verlag. [5] M. Griebel, T. Dornseifer and T. Neunhoeer, 1998, Numerical Simulation in Fluid Dynamics, A practical Introduction, SIAM Monographs on Mathematical Modeling and Computation. [6] P.K. Kundu and I.M. Cohen, 2002, Fluid Mechanics, Second Edition, Academic Press. [7] R.L. Panton, 1995, Incompressible Flow, Second Edition, Wiley-Interscience. [8] R. Peyret, T.D. Taylor, 1983, Computational Methods for Fluid Flow, Springer-Verlag. [9] J.C. Tannehill, D.A. Anderson and R.H. Pletcher, 1997, Computational Fluid Mechanics and Heat Transfer, Second Edition, Taylor and Francis. [10] P. Wesseling, 2001, Principles of Computational Fluid Dynamics, Springer-Verlag.

177