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 flow .

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

.
.
.
.
.
.
.
.

7
7
8
14
15
19
20
22
24

2 Flow physics
2.1 Exact solutions . . . . . . . .
2.2 Vorticity and streamfunction
2.3 Potential flow . . . . . . . . .
2.4 Boundary layers . . . . . . .
2.5 Turbulent flow . . . . . . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

29
29
32
40
48
54

3 Finite volume methods for incompressible flow
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 flows . . . . . . . . . . . . . . .
3.5 General iteration methods for steady flows . . . .

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

59
59
62
65
66
69

4 Finite element methods for incompressible flow
4.1 FEM for an advection–diffusion 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 Navier–Stokes . . . . . . . . . . . . . . . . . . .
4.2.1 A variational form of the Navier–Stokes 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 finite elements. Accuracy . . . . . . . . .

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

71
71
72
73
74
76
76
80
82
83
85
85
86
87
88
88

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 . . . . .

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

.
.
.
.
.
.
.

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

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

.
.
.
.
.

. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
. . . . . . . . . . .
and determinants .
. . . . . . . . . . .
3

.
.
.
.
.
.
.

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .14 Old Exams . . . . . .13 Introduction to turbulence . . . . . . . . . . . . . . .3 Curvilinear coordinates . . . . . . 109 109 113 117 120 124 129 132 136 140 146 148 152 155 157 C Study questions 5C1212 . . . . . . . . . . . . . . .A. . . . . . . . . . . . . . . . . . . . . . . . . . . B. . . . . . . B. . . . . .12 More boundary layers . . . . . . . . . B. . . . . . . . . . . . . . . . . . . . . . . . . . B. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2. . . . . . . . . . . . . . . B. . . . . . . . . . . . . . . . . . . . . . . . . B. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6 Tensor fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .10 More potential flow . . . . . . . . . . . . . . . Bernoulli equation and streamfunction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5 Exact solutions to Navier-Stokes . . . . . . .4 Rankine vortex and dimensionless form . . . . . . . . B. . . . . . . . . . .2. . . . . . . . . . . . .11 Boundary layers . . B. . . . . . . . . . . . . .2. 102 103 104 106 B Recitations 5C1214 B. . . . . . . . B. . . . . . . . . . . . . . B. . . .7 Axisymmetric flow and irrotational vortices . . . . . . .3 Reynolds transport theorem and stress tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7 Gauss & Stokes integral theorems A. . .9 Flow around a submarine and other potential flow problems B. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A.6 More exact solutions to Navier-Stokes . . . . . . . . . . . A.1 Tensors and invariants . . . . . . B. . . . .8 Archimedes principle . .2 Euler and Lagrange coordinates . . . . . . . . . . . . . .8 Vorticity equation. . . B. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173 . . . . . . .

Stockholm.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) at KTH. Stockholm. Erik St˚ alberg and Ori Levin has typed most of the LATEXformulas and has created the electronic versions of most figures. August 2005 Dan Henningson . 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.

.

The Cartesian tensor form of the equations can be written  ∂ui ∂ui   + uj   ∂t ∂xj     ∂ui ∂xi = − 1 ∂p ∂ 2 ui +ν ρ ∂xi ∂xj ∂xj = 0 where the summation convention is used. v. .1 Notation The Navier-Stokes equations in vector notation has the following form  ∂u 1  + (u · ∇) u = − ∇p + ν∇2 u  ∂t ρ   ∇·u = 0 where the velocity components are defined u = (u. ∂x1 ∂x2 ∂x3 the Laplace operator is written as ∇2 = ∂2 ∂2 ∂2 + 2+ 2 2 ∂x ∂y ∂z and the following definitions are used ν ρ p − kinematic viscosity − density − pressure see figure 1. ∂x ∂y ∂z = ∂ ∂ ∂ . u3 ) the nabla operator is defined as ∇= ∂ ∂ ∂ . This implies that a repeated index is summed over.1 for a definition of the coordinate system and the velocity components.Chapter 1 Derivation of the Navier-Stokes equations 1. from 1 to 3. w) = (u1 . as follows ui ui = u 1 u1 + u 2 u2 + u 3 u3 Thus the first component of the vector equation can be written out as 7 . . u2 .

x2 P tˆ = 0 z. u2 w. Lagrange coordinates: every particle is marked and followed in flow. i. u2 u. u3 0 ri xi . We begin by considering the motion of a fluid particle in Lagrangian coordinates.2. x1x2 y. x3 u. the coordinates familiar from classical mechanics. u1 xi 0 v.e. .1: Definition of coordinate system and velocity components PSfrag replacements x. u3 z. tˆ x1 Figure 1. x3 Figure 1.y.2: Particle path. x2 v. is ri = ri xi 0 . see figure 1. x1 w.2 Kinematics Lagrangian and Euler coordinates Kinematics is the description of motion without regard to forces. ˆt ui xi0 . The independent variables are xi 0 − initial position of fluid particle tˆ − time where the particle path of P. u1 PSfrag replacements x. i=1 ∂u1 ∂u1 ∂u1 1 ∂p ∂u1 + u1 + u2 + u3 =− +ν ∂t ∂x1 ∂x2 ∂x3 ρ ∂x1 or ∂u ∂u ∂u ∂u 1 ∂p +u +v +w =− +ν ∂t ∂x ∂y ∂z ρ ∂x ∂ 2 u1 ∂ 2 u1 ∂ 2 u1 + + ∂x1 2 ∂x2 2 ∂x3 2 ∂2u ∂2u ∂2u + 2 + 2 ∂x2 ∂y ∂z ∇2 u 1. tˆ and the velocity of the particle is the rate of change of the particle position.

fluid flows past point. Euler coordinates: consider fixed point in space. t) is now considered as a function of the coordinate xi and time t.  at the time  xi = ri xi 0 . t). t The rate of change of F following a fluid particle can then be written ∂FE ∂F ∂FE ∂ri ∂FE ∂t ∂FE = · + · = + ui ∂xi ∂ tˆ ∂t ∂ tˆ ∂t ∂xi ∂ tˆ D Based on this expression we define the material derivative as Dt ∂ ∂ ∂ ∂ D ≡ = + ui = + (u · ∇) ˆ Dt ∂t ∂xi ∂t ∂t In the material or substantial derivative the first term measures the local rate of change and the second measures the change due to the motion with velocity ui . tˆ = FE (xi . we still need to consider the rate of change of quantities following a fluid particle. The acceleration is defined aj = Dui ∂ui ∂ui = + uj Dt ∂t ∂xj which can be simplified in 1D for stationary case to Du ∂u ∂u = +u Dt ∂t ∂x Note that the acceleration = 0 even if velocity at fixed x does not change. Instead of marking every fluid particle it is most of the time more convenient to use Euler coordinates.4 during time dtˆ. where F = FL xi 0 . The relation between Lagrangian and Euler coordinates. This leads to the following definition. Consider the quantity F following fluid particle. The independent variables are xi − space coordinates t − time Thus the fluid velocity ui = ui (xi . a= Description of deformation Evolution of a line element Consider the two nearby particles in figure 1. As an example we consider the acceleration of a fluid particle in a steady converging river. tˆ .ui = ∂ri ∂ tˆ Note here that when xi 0 changes we consider new particles.3. see figure 1. is easily found by noting that the particle position is expressed in fixed space coordinates xi . The position of P2 can by Taylor expansion be expressed as P2 : ri dtˆ + dri dtˆ = i = ri (0) + ∂r (0) dtˆ + dri (0) + ∂ tˆ ∂ ( dri ) ∂ tˆ 0 = xi 0 + dxi 0 + ui (0) dtˆ + dui (0) dtˆ dtˆ . This has been experienced by everyone in a raft in a converging river. tˆ and (xi . i. xi 0 .e. Material derivative: rate of change in time following fluid particle expressed in Euler coordinates. t) = FE ri xi . The raft which is following the fluid is accelerating although the flow field is steady. i.e. tˆ  Material derivative t = tˆ Although it is usually most convenient to use Euler coordinates.

u1 v. x1 y. tˆ ui xi0 . u1 v. u3 x1 x2 xi 0 ri xi0 . tˆ P tˆ = 0 ui dtˆ P1 xi 0 ˆt ri d x1 x3 Figure 1. x2 z. x2 P2 ˆt ui d d x u1 u2 > u 1 dxi0 dr i dtˆ PSfrag replacements x.3: Acceleration of fluid particles in converging river. u2 w.PSfrag replacements x. x1 y. tˆ ui xi0 . x2 z. u3 x1 x2 xi 0 ri xi0 . x3 u. u2 w. tˆ P tˆ = 0 u1 u2 > u 1 x Figure 1. . x3 u.4: Relative motion of two nearly particles.

where we have used ∂ ( dri ) = ∂ tˆ = ∂ ∂ri ∂ dxj 0 = ∂xj 0 ∂ tˆ ∂xj 0 ∂ui dxj 0 = dui ∂xj 0 ∂ri ∂ tˆ dxj 0 and where ∂ui ∂xj 0 dxj 0 dui We can transform the expression line element in Euler coordinates − change of ui with initial pos. i. no deformation.e.e. describes the deformation and is considered in detail below. whereas the ∂xj anti-symmetric part can be written in terms of the vorticity ωk and is associated with solid body rotation. The anti-symmetric part can be written The symmetric part of . eij . ∂xj ∂xj ∂ui = ξij + eij + eij ∂xj eij where eij is the deformation rate tensor and ξij = eij = eij = ∂uj 1 ∂ui − anti-symmetric part 2 ∂uj ∂xi 1 ∂ui ∂uj 2 ∂ur + − δij traceless part 2 ∂xj ∂xi 3 ∂ur 1 ∂ur δij isotropic part 3 ∂ur ∂ui . − difference in velocity ∂ ( dri ) = dui in Lagrange coordinates to an equation for a material ∂ tˆ D ( dri ) = dui Dt = {expand in Euler coordinates} ∂ui drj = ∂xj where ∂ui ∂xj drj − change in velocity with spatial position − difference in spatial pos. − difference in initial 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.

ωk =  − 1 2 ω2 − 12 ω1 0 kji − ∂u1 ∂x3 ∂u2 ∂x3 ∂u3 ∂x2 − − ∂u3 ∂x1 ∂u3 ∂x2 0 ∂ ui } ∂xj   Deformation of a small cube Consider the deformation of the small cube in figure 1. we consider the deformation on side 1.dxi0 dui dtˆ ri dtˆ dri dtˆ x2 r2 R2 P1 P2 π 2 + dϕ12 R1 R 3 x1 x3 r1 r3 Figure 1.  1 2 0   [ξij ] =  − 21  − 21 ∂u1 ∂x2 ∂u1 ∂x3 − − ∂u2 ∂x1 ∂u3 ∂x1 ∂u1 ∂x2 − 21 − 21 ω3 0 1 2 ω1 0 =  21 ω3 1 2 ω2 ∂u2 ∂x1 1 2 1 2 0 ∂u2 ∂x3 = {use ω = ∇ × u . which can be expressed as dR1 = r1 − R = rk1 rk1 − R The inner product can be expanded as rk1 rk1 = 1+ ∂u1 dt ∂x1 2 + ∂u2 dt ∂x1 = {drop quadratic terms} = R2 1 + 2 ∂u1 dt ∂x1 2 + ∂u3 dt ∂x1 2 R2      . where we define Rlk = Rδkl rkl = Rlk + component k of side l ∂uk l R dt ∂xj j relative motion of side l dulk = R δkl + ∂uk dt ∂xl deformed cube First.5.5: Deformation of a cube.

. dR1 becomes dR1 = R = R 1+2 ∂u1 ∂u1 dt − R = R 1 + dt + . we consider the deformation of the volume of the cube. − R ∂x1 ∂x1 ∂u1 dt = Re11 dt ∂x1 which implies that dR1 = Re11 dt ∂ui Thus the deformation rate of side 1 depends on e11 . This can be expressed as dV = r1 r2 r3 − R 3 ∂u1 ∂x1 dt ∂u2 ∂x1 dt ∂u3 ∂x1 dt 1+ = R3 ∂u1 ∂x2 dt 2 1 + ∂u ∂x2 dt ∂u3 ∂x2 dt 1 ∂u1 ∂x3 dt ∂u2 ∂x3 dt 3 + ∂u ∂x3 dt ∂u1 ∂u2 dt 1+ dt ∂x1 ∂x2 ∂u1 ∂u2 ∂u3 = R3 + + dt ∂x1 ∂x2 ∂x3 ∂uk = R3 dt ∂xk = R3 1 + 1+ ∂ui ∂xj .. j Second. This can be expressed as cos π + dφ12 2 = = = = r1 · r2 1 1 = rk1 rk2 · rm rm · rn2 rn2 |r1 | |r2 | −1/2 ∂uk ∂uk ∂u1 dt δk2 + dt · 1 + 2 dt ∂x1 ∂x2 ∂x1 ∂u2 ∂u1 ∂u2 dt + dt 1− dt 1− dt ∂x1 ∂x1 ∂x2 ∂u2 + dt = 2e12 dt ∂x1 δk1 + ∂u1 ∂x2 ∂u1 ∂x2 −1/2 1+2 ∂u2 dt ∂x2 −1/2 where we have dropped quadratic terms. − R3 ∂u3 dt − R3 ∂x3 where we have again omitted quadratic terms.We have dropped the quadratic terms since we are assuming that dt is small. Thus we have dV = R3 err dt ∂ui and the deformation rate of volume of cube (or expansion rate) depends on isotropic part of ∂x . j In summary. the motion of a fluid particle with velocity ui can be divided into the following invariant parts . We use the trigonometric identity cos π π π + dϕ12 = cos · cos dϕ12 − sin · sin dϕ12 ≈ − dϕ12 2 2 2 which allow us to obtain the finial expression dϕ12 = −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 angle between side 1 and side 2. both traceless and isotropic part of ∂x .

e.6: Volume moving with the fluid. a volume integral where the volume is moving with the fluid. where a volume element describing the change in volume between V at time t and t + ∆t can be written as u · n∆t = uk nk ∆t ⇒ dV = uk nk ∆t dS . i. We obtain the following expressions D Dt Tij dV = V (t)     1  lim  ∆t→0  ∆t  Tij (t + ∆t) dV − V (t+∆t) =     1  lim  ∆t→0  ∆t  V (t) Tij (t + ∆t) dV − V (t+∆t) V (t)   Tij (t + ∆t) dV     1   + Tij (t + ∆t) dV − Tij (t) dV    ∆t  V (t) V (t)      1  ∂Tij lim Tij (t + ∆t) dV + dV ∆t→0   ∂t  ∆t   =     Tij (t) dV    V (t+∆t)−V (t) V (t) The volume in the first integral on the last line is represented in figure 1.6. i) ui ii) ξij = iii) eij = 1 2 1 2 iv) 1. ✁    ✂✁   ✂✁   ✂✁   ✂✁   ✂✁   ✂✁   ✂✁   ✂✁   ✂✁   ✂✁   ✂   ✂✁  ✁  ✁✂✁   ✂✁✂✁  ✂✁✂✁  ✂✁✂✁  ✂✁✂✁  ✂✁✂✁  ✂✁✂✁  ✂✁✂✁  ✂✁✂✁  ✂✁✂✁  ✂✁✂✁  ✂✂  ✁ ✂  ✁✂✁   ✂✁  ✂✁  ✂✁  ✂✁  ✂✁  ✂✁  ✂✁  ✂✁  ✂✁  ✂✁  ✂   ✁    ✂✁   ✂✁   ✂✁   ✂✁   ✂✁   ✂✁   ✂✁   ✂✁   ✂✁   ✂✁   ✂    ✁✂✁ ✂✁ ✂✁ ✁✂ ✂✁ ✁✂ ✂✁ ✁✂ ✂✁ ✁✂ ✂✁ ✁✂ ✂✁ ✁✂ ✂✁ ✁✂ ✂✁ ✁✂ ✂✁ ✁✂ ✂✁ ✁✂ ✂ ✂  ✁ ✁   ✂  ✁✂✁   ✂✁  ✂✁  ✂✁  ✂✁  ✂✁  ✂✁  ✂✁  ✂✁  ✂✁  ✂✁  ✂   ✁    ✂✁   ✂✁   ✂✁   ✂✁   ✂✁   ✂✁   ✂✁   ✂✁   ✂✁   ✂✁   ✂   ✂✁  ✁  ✁✂✁   ✂✁✂✁  ✂✁✂✁  ✂✁✂✁  ✂✁✂✁  ✂✁✂✁  ✂✁✂✁  ✂✁✂✁  ✂✁✂✁  ✂✁✂✁  ✂✁✂✁  ✂✂  ✁ ✂  ✁    ✂✁   ✂✁   ✂✁   ✂✁   ✂✁   ✂✁   ✂✁   ✂✁   ✂✁   ✂✁   ✂    ✁✂✁ ✂✁ ✂✁  ✂✁  ✂✁  ✂✁  ✂✁  ✂✁  ✂✁  ✂✁  ✂✁  ✂✁  ✂   ✁   ✁ ✂  ✁✂✁   ✂✁✂✁  ✂✁✂✁  ✂✁✂✁  ✂✁✂✁  ✂✁✂✁  ✂✁✂✁  ✂✁✂✁  ✂✁✂✁  ✂✁✂✁  ✂✁✂✁  ✂✂   ✁✂✁   ✂✁  ✂✁  ✂✁  ✂✁  ✂✁  ✂✁  ✂✁  ✂✁  ✂✁  ✂✁  ✂   ✁✂✁   ✂✁  ✂✁  ✂✁  ✂✁  ✂✁  ✂✁  ✂✁  ✂✁  ✂✁  ✂✁  ✂  V (t + ∆t) − V (t) S (t) R3 R2 R1 r3 r2 r1 + dϕ12 V (t) π 2 S (t + ∆t) u n V (t + ∆t) Figure 1.3 solid body translation ∂ui ∂uj − ∂xj ∂xi = − 21 kij ωk solid body rotation ∂ui ∂uj 2 ∂ur + − δij ∂xj ∂xi 3 ∂ur eij = 1 3 volume constant deformation ∂ur δij ∂ur volume expansion rate Reynolds transport theorem Volume integral following with the fluid Consider the time derivative of a material volume integral.

eq. The integrand in the theorem can then be written 0 (cont. i.This implies that the volume integral can be converted to a surface integral.e. Conservation of mass By the substitution Tij → ρ and the use of the Reynolds transport theorem above we can derive the equation for the conservation of mass. The terms in the expression can be given the following interpretations: 1 2 3 4 : : : : accumulation of mass in fixed element net flow 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. 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 ∂ρ ∂ + (uk ρ) = 0 ∂t ∂xk ρ dV = V (t) V (t) Since the volume is arbitrary. that the time rate of change of momentum in a material region = sum of the forces on that region. 1. the following must hold for the integrand 0= ∂ρ ∂ ∂ρ ∂uk Dρ ∂uk ∂ρ + (uk ρ) = + uk +ρ = +ρ ∂t ∂xk ∂t ∂xk ∂xk Dt ∂xk 1 3 2 4 where we have used the definition of the material derivative in order to simplify the expression. Tij = ρtij . We have D Dt Tij dV = lim ∆t→0 S(t) V (t) = ∂Tij dV ∂t Tij (t + ∆t) uk nk dS + V (t) ∂Tij dV = {Gauss/Green’s theorem} ∂t Tij uk nk dS + S(t) V (t) ∂Tij ∂ + (uk Tij ) dV ∂t ∂xk = V (t) which is the Reynolds transport theorem.) ✯ ✟0 ✕ ✁ ∂ ∂ ∂ρ ∂tij ∂ ∂tij Dtij ✟ (ρtij ) + (uk ρ tij ) = tij ✁ + ρ + tij ✟✟ (uk ρ) + ρuk =ρ ∂t ∂xk ∂t ∂xk ∂xk Dt ✁∂t ✟ which implies that Reynolds transport theorem becomes D Dt ρ tij dV = V (t) ρ Dtij dV Dt V (t) where V (t) again is a material volume.4 Momentum equation Conservation of momentum The momentum equation is based on the principle of conservation of momentum. i. we can simplify Reynolds transport theorem. The quantities involved are: Fi Ri ρui - body forces per unit mass surface forces per unit area momentum per unit volume .e.

8: Surface force and unit normal. irrespective of volume forces or acceleration terms.7.8.∆s (k) V (t + ∆t) − V (t) ri dtˆ nIII(k) S (t + ∆t) dri dtˆ u dl P1 n P2 V (t + ∆t) x1 R(nI ) x2 x3 3 R nII nI R2 R1 dS r3 2 R(nII ) r r1 π + dϕ 12 2 Figure 1. The stress tensor Remove a fluid element and replace outside fluid by surface forces as in figure 1.7: Momentum balance for fluid element V (t) S (t) V (t + ∆t) − V (t) n S (t + ∆t) u n R V (t + ∆t) R Figure 1. Thus the surface forces acting on a small fluid particle has to balance. see figure 1. In order to do that we have to define the stress tensor. Momentum conservation for the small fluid particle leads to ρ Dui dS dl = ρFi dS dl + Ri nIj dS + Ri nII j dS + Dt Ri nIII(k) ∆s(k) dl j k Letting dl → 0 gives 0 = Ri nIj dS + Ri nII j dS Now nj = nIj = −nII j which leads to 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. . We can put the momentum conservation in integral form as follows D Dt ρui dV = V (t) ρFi dV + V (t) Ri dS S(t) Using Reynolds transport theorem this can be written ρ Dui dV = Dt ρFi dV + V (t) V (t) Ri dS S(t) which is Newtons second law written for a volume of fluid: mass · acceleration = sum of forces. Note that it is a general principle that the terms proportional to the volume of a small fluid particle approaches zero faster than the terms proportional to the surface area of the particle. Here R(n) is the surface force per unit area on surface dS with normal n. To proceed Ri must be investigated so that the surface integral can be transformed to a volume integral.

u2 r1 π R(e2 ) = ( T12 . i tˆ 0 dxi du tˆ We now divide the surfacei ˆdforces into components along the coordinate directions.9 and r i dt 1. . Momentum balance require the surface forces to balance R in the element. Tij is the i-component P1 surfaces along the coordinate directions cut by a slanted surface.10: Definition uof dsurface force components on a surface with a normal in the 2-direction. t u P tˆ = 0 n V (t + ∆t) x T32 u1 u2 > u 1 x1 x2 x3 xi 0 Figure 1. we have R2 R1 r03 = R1 dS − T11 n1 dS − T12 n2 dS − T13 n3 dS r2 surface force Ri can be written in terms of the components of the stress which implies that the total r1 tensor Tij as π + dϕ 12 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 dS3 dS1 dS2 e1 e3 Figure 1. t ∆t) i + 0 ˆ T12 u i xi . as in Consider a fluid particle with P2 figure 1. x3 r3 Figure 1. The areas of the surface elements are related by x1 x2 dSj = ej · n dS = nj dS x3 where dS is the area of the3 slanted surface. T22 . dri dtˆof the surface force on a surface dS with a normal in the j-direction. u1 r2 x2 v.10. x2 R1 z. T21 . u 3 12 2 V1 (t) x T22 xS2 (t) V (t + ∆t) −xVi0 (t) 0 ˆ ri Sx(t . T31 ) n V (t + ∆t) P2 x1 T21 x2 x1 T11 x3 T31 PSfrag replacementsR3 2 x. T32 ) + dϕ w.11: Surface force balance on a fluid particle with a slanted surface. x1 R y. as in figures 1. Thus.11.9: Definition of surface force components on a surface with a normal in the 1-direction. with corresponding definitions for the force components on a surface with a normal in the 3-direction. u.u P1 R(e1 ) = ( T11 .

Recall that the invariant symmetric parts are eij = 1 2 ∂ui ∂uj 2 ∂ur + − δij ∂xj ∂xi 3 ∂xr eij + 1 ∂ur δij 3 ∂xr eij where eij is the volume constant deformation rate and eij is the uniform rate of expansion. We r2 on the motion of stress tensor into an isotropic part. u2 > u 1 x1 Tij = Tji x2 x3 Momentum equation xi 0 i dtˆ Using the definition of the surface force in terms of the stress tensor. Here the diagonal components are normal stresses off-diagonal components are Further consideration of the moment balance around a fluid particle show that Tij is axsymmetric tensor.P tˆshear = 0 stresses. the momentumuequation can now be dxi0 written dui dtˆ ˆ r Dui ∂Tij i dt ρ dV = ρFi dV + Tij nj dS = ρFi + dV ˆ Dt ∂xj dri dt V (t) V (t) S(t) V (t) P1 P2 have The volume is again arbitrary implying that the integrand must itself equal zero. τij = λeij + 2µeij where we have defined the two viscosities as µ (T ): λ (T ): dynamic viscosity (here T is the temperature) second viscosity. For an isotropic fluid. We x1 x2 Dui ∂Tij = ρFi + ρ x3 Dt ∂xj R3 R2 Pressure and viscous stress tensor R1 r3 thus divide the For fluid at rest only normal stresses present otherwise fluid element would deform. often=0 For a Newtonian fluid. we thus have the following relationship between the viscous stress and the strain (deformation rate) τij = 2µeij = µ ∂ui ∂uj 2 ∂ur + − δij ∂xj ∂xi 3 ∂xr which leads to the momentum equation ρ Dui ∂p ∂ =− + µ Dt ∂xi ∂xj ∂ui ∂uj 2 ∂ur + − δij ∂xj ∂xi 3 ∂xr + ρFi p . u1 i. We have π + dϕ 12 p 2 V (t) Tij = −pδij + τij S (t) V (t + ∆t) − V (t) p .hydrodynamic pressure. and a part depending r1 the fluid. the hydrodynamic pressure p.viscous stress tensor.e. directed inward τij . the viscous stress tensor is a linear function of the invariant parts of e ij . depends on fluid motion S (t + ∆t) u n V (t + ∆t) p Newtonian fluid It is natural to assume that the viscous stresses are functions of deformation rate e ij or strain. i.e.

We have the following definitions ρ e + 21 ui ui dV energy of particle. We have 1 ∂ ∂ ∂qi D e + ui ui = ρFi ui − (pui ) + (τij uj ) − Dt 2 ∂xi ∂xi ∂xi The mechanical energy equation is found by taking the dot product between the momentum equation and u.e.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 obtain ρ D 1 ∂p ∂τij ui ui = ρFi ui − ui + ui Dt 2 ∂xi ∂xj Thermal energy equation is then found by subtracting the mechanical energy equation from the total energy equation. directed outward Using Reynolds transport theorem we can put the energy conservation in integral form as D Dt 1 ρ e + ui ui dV = 2 V (t) [ni Tij uj − ni qi ] dS = ρFi ui dV + V (t) S(t) ρFi ui + ∂ (Tij uj − qi ) dV ∂xi V (t) Compare the expression in classical mechanics. i. ρ ∂ui ∂ui ∂qi De = −p + τij − Dt ∂xi ∂xj ∂xi The work of the surface forces divides into viscous and pressure work as follows ρ − ∂ (pui ) ∂xi = −p ∂ui ∂xi − ui 1 ∂ (τij uj ) ∂xi = τij ∂ui ∂xj ∂p ∂xi 2 + uj ∂τij ∂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 fluid and increase kinetic energy . where the momentum equation is mu˙ = F and the associated kinetic energy equation is m d 2 dt (u · u) work rate ( work = = = F·u force · velocity force · dist. with e the internal energy work rate of Fi on particle ρui Fi dV force · velocity uj Rj dS work rate of Rj on particle ni qi dS heat loss from surface. with qi the heat flux vector. ) 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.

This allows us to write the thermal energy equation as ρ De ∂ui ∂ = −p +Φ+ Dt ∂xi ∂xi κ ∂T ∂xi where the positive definite dissipation function Φ is defined as Φ = τij ∂ui 1 = 2µ eij eij − ∂xj 3 = 2µ eij − ∂uk ∂xk 2 2 1 ∂uk δij 3 ∂xk >0 Alternative form of the thermal energy equation can be derived using the definition of the enthalpy h=e+ p ρ We have Dh De 1 Dp p = + − 2 Dt Dt ρ Dt ρ Dρ Dt ∂u − ρ ∂xi i which gives the final result ρ Dh Dp ∂ = +Φ+ Dt Dt ∂xi κ ∂T ∂xi To close the system of equations we need a i) thermodynamic equation e = e (T.6 cp cv Navier-Stokes equations The derivation is now completed and we are left with the Navier-Stokes equations. At this time we also define the ratio of the specific heats as γ= 1. the equation describing the conservation of momentum and the equation describing the conservation of energy. respectively. p) ii) equation of state p = ρRT simplest case: e = cv T or h = cp T where cv and cp are the specific heats at constant volume and temperature.The heat flux need to be related to the temperature gradients with Fouriers law qi = −κ ∂T ∂xi where κ = κ (T ) is the thermal conductivity. We have Dρ ∂uk +ρ =0 Dt ∂xk ρ ρ where Dui ∂p ∂τij =− + + ρFi Dt ∂xi ∂xj De ∂ui ∂ = −p +Φ+ Dt ∂xi ∂xi κ ∂T ∂xi . They are the equation describing the conservation of mass.

ρu2 . ρu1 . p) p = ρRT We have 7 equations and 7 unknowns and therefore the necessary requirements to obtain a solution of the system of equations. J = (0. momentum and energy. . ρu3 . gas law 1 3 1 1 1 7 Equations in conservative form A slightly different version of the equations can be found by using the identity ρ Dtij ∂ ∂ = (ρtij ) + (uk ρtij ) Dt ∂t ∂xk to obtain the system     ∂ρ ∂    + (uk ρ) = 0   ∂t ∂x k   ∂ ∂ ∂p ∂τij (ρui ) + (ρuk ui ) = − + + ρFi ∂t ∂xk ∂xi ∂xj     ∂ 1 ∂ 1 ∂ ∂ ∂   = ρFi ui − e + ui ui + ρuk e + ui ui (pui ) + (τij uj ) +   2 ∂xk 2 ∂xi ∂xi ∂xi   ∂t κ ∂T ∂xi These are all of the form ∂U ∂G(i) + =J ∂t ∂xi or ∂U ∂G ∂G ∂G + + + =J ∂t ∂x ∂y ∂z where U = (ρ. unknown ρ 1 ui 3 p 1 e 1 T 1 7 equations continuity momentum energy thermodyn. ρF1 . ρui Fi ) is the vector of the right hand sides and  G(i)     =     T ρui ρu1 ui + pδ1i − τ1i ρu2 ui + pδ2i − τ2i ρu3 ui + pδ3i − τ3i ∂T ρ (e + ui ui /2) ui + pui + κ ∂x − uj τij i           is the flux of mass. ρF3 . ρ (e + ui ui /2)) T is the vector of unknowns. This form of the equation is usually termed the conservative form of the Navier-Stokes equations. ρF2 . respectively.    Φ    τij ∂ui ∂xj ∂uj 2 ∂ur ∂ui + − δij = µ ∂xj ∂xi 3 ∂xr = τij and the thermodynamic relation and the equation of state for a gas e = e (T.

using the continuity equation and Gauss theorem. as will be discussed below. We have V ∂ (ui ) dVF = ∂xi ui ni dSF = 0 S An integral form of the momentum equation is found by taking the time derivative of a fixed volume integral of the velocity. They are for obvious reasons usually referred to as the no slip conditions. We have derived the differential form of the Navier-Stokes equations. which from the conservation of mass equation implies ∂ui =0 ∂xi implying that a fluid particle experiences no change in volume. We have ∂ ∂t ∂ui dV = − ∂t ui dV = VF − VF VF VF ∂ 1 ∂ui uj ui + pδij − ν ∂xj ρ ∂xj ∂ 1 ∂p ∂ 2 ui (uj ui ) + −ν ∂xj ρ ∂xi ∂xj ∂xj dV = − SF dV = p ∂ui ui uj n j + n i − ν nj dS ρ ∂xj .7 Incompressible Navier-Stokes equations The conservation of mass and momentum can be written  Dρ ∂ui  +ρ =0    Dt ∂x i   Dui ∂p ∂τij ρ =− + + ρFi Dt ∂xi ∂xj     ∂ui ∂uj 2 ∂uk   τij = µ + − δij ∂xj ∂xi 3 ∂uk for incompressible flow ρ = constant. for example when a finite-volume discretization is derived. Thus the conservation of mass and momentum reduce to  ∂ui ∂ui 1 ∂p   + uj =− + ν∇2 ui + Fi   ∂t ∂xj ρ ∂xi ∂ui =0    ∂x i  since the components of the viscous stress tensor can now be written ∂ ∂xj ∂ui ∂uj 2 ∂uk + − δij ∂xj ∂xi 3 ∂xk = ∂ 2 ui = ∇ 2 ui ∂xj ∂xj We have also introduced the kinematic viscosity. or just the Navier-Stokes equations in case it is evident that the incompressible limit is assumed. Integral form of the Navier-Stokes eq. 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 fixed volume VF and using the Gauss theorem. In addition it can be worth noting that the conservation of mass equation is sometimes referred to as the continuity equation. ν. defined as ν = µ/ρ The incompressible version of the conservation of mass and momentum equations are usually referred to as the incompressible Navier-Stokes equations. Sometimes. as we will see below. The pressure field does not need to be specified as it can be obtained once the velocity field is specified. substituting the differential form of the Navier-Stokes equation.1. Boundary conditions (BC) on solid surfaces are ui = 0. The incompressible Navier-Stokes equations need boundary and initial conditions in order for a solution to be possible. As initial condition (IC) one needs to specify the velocity field at the initial time ui (t = 0) = u0i . 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. above.

velocity and temperature scales. If we introduce the non-dimensional variables in the Navier-Stokes equations we find  2 U0 ∂u∗i U 2 ∂u∗ p0 ∂p∗ νU0 ∂ 2 u∗i   · ∗ + 0 u∗j ∗i = − +  L ∂t L ∂xj ρL ∂x∗i L2 ∂x∗j ∂x∗j ∗ U ∂u  0 i   · =0 L ∂x∗i Now drop * as a sign of non-dimensional variables and divide through with U02 /L. We also define the pressure scale as p0 = ρU02 which leaves us with the final form of the non-dimensional incompressible Navier-Stokes equations as  ∂ui ∂p 1 2 ∂ui  + uj =− + ∇ ui  ∂t ∂xj ∂xi Re   ∂ui = 0 ∂xi . U 0 L L Figure 1. Note that the integral from of the continuity equation is a compatibility condition for BC in incompressible flow. 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 ∗ signifies a non-dimensional variable. The Reynolds number is by far the single most important non-dimensional number in fluid mechanics. we find  ∂ui ∂ui p0 ∂p ν   + uj =− 2 + ∇2 u i ∂t ∂xj ρU0 ∂xi U0 L   ∂ui = 0 ∂xi We have defined the Reynolds number Re as Re = U0 L U02 /L = ν νU0 /L2 “inertial forces” “viscous forces” which can be interpreted as a measure of the inertial forces divided by the viscous forces.12: Examples of length. Dimensionless form It is often convenient to work with a non-dimensional form of the Navier-Stokes equations.12 for examples. The length and velocity scales have to be chosen appropriately from the problem under investigation. so that they represent typical lengths and velocities present.S (t + ∆t) u n V (t + ∆t) p ✄ ✆☎ ✄ ✆☎ ✄ ✆☎ ✄ ✆☎ ✄ ✆☎ ✄ ✆☎ ✄ ✆☎ ✄ ✆☎ ✄ ✆☎ ✄ ✆✄ ✆☎ ✄ ✆☎ ✄ ✆☎ ✄ ✆☎ ✄ ✆☎ ✄ ✆☎ ✄ ✆☎ ✄ ✆☎ ✄ ✆☎ ✄ ✆☎ ✄ ✆✄ ✆☎ Tw T0 Tw U0 ✝ ✞☎ ✝ ✞☎ ✝ ✞☎ ✝ ✞☎ ✝ ✞☎ ✝ ✞☎ ✝ ✞☎ ✝ ✞☎ ✝ ✞☎ ✝ ✞✝ ✞☎ ✝ ✞☎ ✝ ✞☎ ✝ ✞☎ ✝ ✞☎ ✝ ✞☎ ✝ ✞☎ ✝ ✞☎ ✝ ✞☎ ✝ ✞☎ ✝ ✞✝ ✞☎ T0 . See figure 1.

Non-dimensional energy equation We stated above that the energy equation decouple from the rest of the Navier-Stokes equations for incompressible flow. not depending on p. ρα α= 1 ∂ρ ρ ∂p T where α is the isothermal compressibility coefficient. We define a simplified continuity equation and equation of state as ∂ρ ∂ui + =0 and p = βρ ∂t ∂xi This allows us to write the Navier-Stokes equations as  ∂ui ∂ ∂p 1 2   + (uj ui ) = − + ∇ ui ∂t ∂xj ∂xi Re   ∂p + β ∂ui = 0 ∂t ∂xi Let        p βu βv βw  u   p + u2   uv   uw    u= e= f = g=  v   uv   p + v2   vw w uw vw p + β2     . We now drop ∗ and let M a → 0 a0 DT 1 1 = · · ∇2 T Dt Re P r µc where the Prandtl number P r = κp . we have that α → 0 and a0 → ∞. As the Mach number approaches zero.8 Role of the pressure in incompressible flow The role of the pressure in incompressible flow is special due to the absence of a time derivative in the continuity equation. which implies that M a → 0. 1. We will illustrate this in two different ways. Artificial compressibility First we discuss the artificial compressibility version of the equations.7 for air. the fluid behaves as an incompressible medium. which can be seen from the definition of the speed of sound. This can be seen from a non-dimensionalization of the energy equation. If the density is constant. which takes on a value of ≈ 0. We use the definition of the enthalpy (h = cp T ) to write the thermal energy equation as DT Dp = + Φ + κ∇2 T Dt Dt and use the following non-dimensional forms of the temperature and dissipation function ρcp T∗ = This leads to DT ∗ Dt∗ = = T − T0 Tw − T 0 Φ∗ = L2 Φ U02 µ 2 U02 µ κ Dp∗ µ · · ∇∗ T ∗ + + φ∗ ρU0 L µcp cp (Tw − T0 ) Dt∗ ρU0 L 2 1 1 U02 · · ∇∗ T ∗ + Re P r cp (Tw − T0 ) Dp∗ 1 + Φ∗ Dt∗ R a2 M a2 cp (Tw0−T where a0 speed of sound in free-stream and M a = to obtain 0) U0 is Mach number. We have a0 = γ . used sometimes for solution of the steady equations.

be the vector of unknowns and their fluxes. The Navier-Stokes equations can then be written
∂u ∂e
∂f
∂g
1 2
+
+
+
=
∇ Du
∂t
∂x ∂y
∂z
Re
∂u
∂u
∂u
∂u
1
+A
+B
+C
= ∇2 Du
∂t
∂x
∂y
∂z
R

or

where the matrices above are defined

0 β
∂e 
1
2u
=
A=
∂u  0 v
0 w

0
∂g 
0
C=
=
∂u  1
0



0 0

0 0 
 , B = ∂f = 
u 0 
∂u 
0 u


0 0 β
0
 0
v u 0 
, D = 
 0
0 2v 0 
0 w v
0

0 0 β
0 v u
1 0 2v
0 0 w
0
1
0
0


0
0 

0 
1

0
0
1
0


0
0 

0 
v

the eigenvalues of A, B, C are the wave speeds of plane waves in the respective coordinate directions,
they are
u2 + β ,

and
w, w, w ± w2 + β

Note that the effective acoustic or pressure wave speed ∼ β → ∞ for incompressible flow.
u, u, u ±

v2 + β

v, v, v ±

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

where

∂p
∂xi

u i · ni = 0

on

∂Ω.

i.e. into a function ui that is divergence free and parallel to the boundary and the gradient of a function,
here called p.
∂p
is orthogonal in the L2 inner product.
∂xi
∂p

ui
dV =
(ui p) dV = pui ni dS = 0
∂xi
∂xi

Proof. We start by showing that ui and
ui ,

∂p
∂xi

=

∂Ω

The uniqueness of the decomposition can be seen by assuming that we have two different decompositions
and showing that they have to be equivalent. Let
(1)

wi = u i

+

∂p(1)
∂p(2)
(2)
= ui +
∂xi
∂xi

The inner product between
(1)

ui

(2)

− ui

+


p(1) − p(2)
∂xi

(1)

and ui

(2)

− ui

gives
(1)

0=

ui

(2)

− ui

2

(1)

+ ui

(2)

− ui

Thus we have

(1)

ui

(2)

= ui ,


p(1) − p(2)
∂xi

p(1) = p(2) + C

(1)

dV =

ui

(2)

− ui

2

dV

r
+ dϕ12
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
π
2

Gradient fields

wi
∂p
∂xi

ui

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

∂wi
∂xi

has a unique solution. Now, if
ui = w i −
we have
0=

∂p
= wi ni on ∂Ω
∂n

in Ω with

∂ui
∂wi
∂2p
=

;
∂xi
∂xi
∂xi ∂xi

∂p

∂xi
0 = u i ni = w i ni −

and

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

∂wi
= ∇2 p,
∂xi

2) let ui = wi −

∂p
∂n

∂p
∂xi

∂p
=0
∂n

∂p
∂xi

This is schematically shown in figure 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 2
+
∇ ui
∂xj
Re

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

P

∂ui
∂ui
=
∂t
∂t

However,
P∇2 ui = ∇2 ui

since ∇2 ui need not be parallel to the boundary. Thus we find an evolution equation without the
pressure, we have



∂ui
∂ui
1 2 


= P −uj
+
∇ ui 
∂t
∂xj
Re


wi

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


∇2 p =


∂xi


∂ui
1 2 
∂uj ∂ui


+
∇ ui  = −
−uj

∂xj
Re

∂xi ∂xj
wi

thus the pressure satisfies elliptic Poisson equation, which links the velocity field in the whole domain
instantaneously. This can be interpreted such that the information in incompressible flow spreads infinitely
fast, i.e. we have an infinite wave speed for pressure waves, something seen in the analysis of the artificial
compressibility equations.

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 2

Re

∂ui
∂xi

Thus the divergence is not automatically zero, but satisfies 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
specified, but they have to be chosen such that the divergence of the velocity field is zero on the boundary.
This may be a difficult constraint to satisfy in a numerical solution algorithm.

.

tˆ ui xi0 .1. S (t + ∆t) The steady momentum equations read u  n ∂u ∂p ∂u  V (t + ∆t)  ρu1 1 + ρu2 1 = − + µ∇2 u1 ∂x1 ∂x2 ∂x1 p ∂u ∂u ∂p   ρu1 2 + ρu2 2 = − + µ∇2 u2 − ρg T0 . steady flow. u2 w. tˆ P tˆ = 0 Chapter 2 Flowuxphysics 1 u2 > u 1 x1 x2 x3 2.exact solution for channel flow dxi0 ˆ dui dt The flow inside the two-dimensional channel. by applying the assumptions above.1: Plane channel geometry. We will assume two-dimensional. This implies that u2 = C = 0. U 0 ∂x1 ∂x2 ∂x2 Tw which. u3 = 0 P2 ∂t x1 and that the x2 flow is fully developed. reduce to L T0 U0 x2 L Gradient fields wi ∂p h ∂xi ui ree vectorfields // to boundary p0 > p 1 x1 p ✟ ✡✠ ✟ ✡✠ ✟ ✡✠ ✟ ✡✠ ✟ ✡✠ ✟ ✡✠ ✟ ✡✠ ✟ ✡✠ ✟ ✡✠ ✟ ✡✠ ✟ ✡✠ ✟ ✡✠ ✟ ✡✠ ✟ ✡✠ ✟ ✡✠ ✟ ✡✠ ✟ ✡✠ ✟ ✡✠ ✟ ✡✠ ✟ ✡✠ ✟ ✡✟ ✡✠ 1 ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞✠ ☛ ☞☛ ☞☛✠☞✠ Figure 2. i. The continuity equation reads r3 0 0 r2 ✼ ✓ ✼ ✓ ∂u ∂u ∂u 1 2 3 r1 + ✓ =0 ✓ + π ∂x ∂x ∂x + dϕ 1 2 3 12 2 ✓ ✓ V (t) where the first and the last term disappears due to the assumption of two-dimensional flow and that S (t) the flow is fully developed.z. u1 v. x3 u. u2 = u2 (x2 ) R3 2 R The boundary R1 conditions are u1 = u2 = 0 at x2 = ±h.1 Exact xi0 solutions ui dtˆ Plane Pouseuille flow . u3 x1 x2 xi 0 ri xi0 . 29 . see figure 2. is driven by a pressure difference p 0 −p1 between ri dtˆ the inlet and the outlet of the channel. meaning that effects of inlet conditions have disappeared. We have x3 u1 = u1 (x2 ) . where the constant is seen to be zero from the V (t + ∆t) − V (t) boundary conditions.e. dri dtˆ P1 ∂ = 0.

2: Volume of fluid flowing through surface dS in time ∆t ∂p d2 u 1 +µ ∂x1 dx22 ∂p − ρg 0=− ∂x2 0=− The momentum equation in the x2 -direction (or normal direction) implies p = −ρgx2 + P (x1 ) ∂p dP = ∂x1 dx1 ⇒ showing that the pressure gradient in the x1 -direction (or streamwise direction) is only a function of x1 . Integrating over the channel we find h ui ni dS = {channel} = Q= S −h u1 dx2 = − 2h3 dP 4 · = humax 3µ dx1 3 . (independent of x1 . The momentum equation in the streamwise direction implies 0=− dP d2 u 1 +µ dx1 dx22 Since P (x1 ) and u1 (x2 ) we have d2 u 1 1 dP = = const. The velocity becomes u1 x2 =1− umax h 2 Flow rate From figure 2. x2 ) dx22 µ dx1 We can integrate u1 in the normal direction to obtain u1 = 1 dP 2 x + C 1 x2 + C 2 2µ dx1 2 The constants can be evaluated from the boundary conditions u1 (±h) = 0. giving the parabolic velocity profile u1 = h2 dP x2 · 1− 2µ dx1 h 2 Evaluating the maximum velocity at the center of the channel we have umax = − h2 dP · >0 2µ dx1 if flow is in the positive x1 -direction.2 we can evaluate the flow rate as dQ = u · n dS.ui Divergence free vectorfields // to boundary L h x1 x2 p0 > p 1 p1 u n dS u·n ∆t Figure 2.

we introduce a new dependent variable which is a combination of y and t such that the partial differential equation reduces to an ordinary differential equation. This gives us the following diffusion equation.y L h x1 x2 p0 > p 1 p1 u n dS u · n∆t u0 p0 ✌ ✎✍ ✌ ✎✍ ✌ ✎✍ ✌ ✎✍ ✌ ✎✍ ✌ ✎✍ ✌ ✎✍ ✌ ✎✍ ✌ ✎✍ ✌ ✎✍ ✌ ✎✍ ✌ ✎✍ ✌ ✎✌ ✎✍ x Figure 2. We assume that the velocity can be written u = u(y. initial and boundary conditions for u ∂u ∂2u = ν 2 ∂t ∂y u(y. t) = 0 t > 0 We look for a similarity solution. i.3. i. The momentum equations reduce to 1 ∂p ∂2u +ν 2 ρ ∂x ∂y 1 ∂p 0 = − −g ρ ∂y ∂u ∂t = − The normal momentum equation implies that p = −ρgy + p0 . where p0 is the atmospheric pressure at the plate surface. t) = u0 t > 0 u(∞. This is a time dependent problem where plate velocity is set to u = u0 at time t = 0.3: Stokes instantaneously plate. Let . ω1 = ω2 = 0 and has the following expression in the x3 -direction (or spanwise direction) ω3 = ∂u2 ∂u1 du1 2umax · x2 − =− = ∂x1 ∂x2 dx2 h2 Stokes 1:st problem: instantaneously started plate Consider the instantaneously starting plate of infinite horizontal extent shown in figure 2. Wall shear stress The wall shear stress can be evaluated from the velocity gradient at the wall. 0) = 0 0 ≤ y ≤ ∞ u(0. t) which using continuity and the boundary conditions imply that normal velocity v = 0.e. We have τ12 = τ21 = µ 2µ · umax · x2 du1 =− dx2 h2 implying that τwall = τ x2 =−h = 2µ umax h Vorticity The vorticity is zero in the streamwise and normal directions.e.

u η = C · y · tb u0 where C and b are constants to be chosen appropriately. both as a function of the similarity variable η and the unscaled normal coordinate y. The exponents of these expressions.2 Vorticity and streamfunction Vorticity is an important concept in fluid dynamics. We transform the time and space derivatives according to f (η) = ∂ ∂t ∂ ∂y ∂2 ∂y 2 ∂η d d = Cbytb−1 ∂t dη dη ∂η d d = = Ctb ∂y dη dη 2 d = C 2 t2b 2 dη = If we substitute this into the equation for u and use the definition of η. The boundary and initial conditions also have to be compatible with the similarity assumption. This gives b=− 1 2 C=√ We choose κ = −2 and obtain the following 1 −2κν y η= √ 2 νt f + 2ηf = 0 d where = dη . However. It is related to the average angular momentum of a fluid particle and the swirl present in the flow. Thus. Here the maximum of the vorticity is also changing in time. we find an equation for f df d2 f = νC 2 t2b 2 dη dη where no explicit dependence on y and t can remain if a similarity solution is to exist. we have bt−1 η bt−1 = κνC 2 t2b where κ is a proportionality constant. Note again that the time dependent solution is diffusing upward.4. as well as the coefficients in front of the t-terms have to be equal. t) = f (∞) = 0 = f (0) = 1 u0 u0 This equation can be integrated twice to obtain η f = C1 u(∞. From the velocity we can calculate the spanwise vorticity as a function of the similarity variable η and the unscaled normal coordinate y. ω3 = ω z = 2. the coefficient of the two η-dependent terms have to be proportional to each other. t) = f (∞) = 0 u0 2 eξ dξ + C2 0 Using the boundary conditions and the definition of the error-function. Note that the time dependent solution is diffusing upward. a flow with circular streamlines may have zero vorticity and a flow with straight streamlines may have a non-zero vorticity. from the infinite value at t = 0 it is diffusing outward in time. 0) u(0. .5. we have η 2 2 f =1− √ eξ dξ = 1 − erf(η) π 0 This solution is shown in figure 2. both as a function of the similarity variable η and the unscaled normal coordinate y. ∂v ∂u u0 −η2 − =√ e ∂x ∂y πνt The vorticity is shown in figure 2. We have u(y.

8 0.5 0.4 1 0.6 0.9 0 1 √ y = η 4νt t η y ωz = ω ˜ (η)·u0 √ πνt t 0.6 2 0.1 0.5 0.8 0.1 0. We can calculate the circulation of this flow as 2π Γ= uθ r dθ = C C dθ = 2πC 0 Thus we have the following relationship for the ideal vortex uθ = Γ 2πr . t) for various times.4 5 1.4 0.3 0.6 2 0. Vorticity and circulation The vorticity is defined mathematically as the curl of the velocity field as ω =∇×u In tensor notation this expression becomes ωi = ijk ∂uk ∂xj Another concept closely related to the vorticity is the circulation.8 7 1.7 0.5 0. u0 u0 p0 p0 y y η = √4νt η = √4νt 8 2 1.5: Vorticity above the instantaneously started plate.4 1.8 0.4 0.2 0 0 0.2 4 1 0.6 6 1.5 0.6 we show a closed curve C.3 0.3 0. Example 1.7 0. one interpretation is that the vorticity is the circulation per unit area for a surface perpendicular to the vorticity vector.x a) U as a function of the similarity variable Figure y y η. b) U (y.6 0.8 3 0.2 0.4 1 0.6 0. t) for various times.4 √ πνt u0 0.7 0.8 7 6 5 4 t 3 0.2 0 0 0.2 0.7 0.4: Velocity above the instantaneously started plate.9 1 8 2 1.6 0.1 0. The circulation of an ideal vortex. b) ωz (y.3 ωz 0.6 √ y = η 4νt ty η 1.8 0.1 0. Let the azimuthal velocity be given as u θ = C/r.9 1 ω(η)·u ˜ 0 √ πνt ωz = =ω ˜ (η) √ Figure 2.8 u u0 0 u u0 1. a) ωz πνt/u0 as a function of the similarity variable η.4 0. which can also be written dΓ = ω i ni dS Thus.rfields // to boundaryn Divergence free vectorfields // to boundary n dS dS L L u · n∆t u · n∆t hx hx x1 y x1 y √ xy x2u η = √2u 0 0 yp=> η p 4νt t p0 > p14νt 0 1p p0 0 y p1 1 η = √p4νt u u √n n y = η 4νt dS dS t u u u · n∆t u · n∆t u0 u0 x 2.2 0.2 0. which is defined as Γ= ui dxi = ui ti ds C C In figure 2. Using Stokes theorem we can transform that integral to one over the area S as ∂uk Γ= dS = ωi ni dS ijk ni ∂xj S S This allows us to find the following relationship between the circulation and vorticity dΓ = ω i ni dS.2 1 0.9 √ ωz πνt u0 1 =ω ˜ (η) 0 0 0.

We take the curl ( equations and find ∂ ∂t pqi ∂ui ∂xq + pqi ∂2 ∂xq ∂xi 1 uj uj 2 + The second and the fourth terms are zero. except for an infinite value at the center of the vortex. Derivation of the Vorticity Equation We will now derive an equation for the vorticity. the vorticity equation becomes ∂ωp ∂ωp ∂up 1 2 + uk = ωj + ∇ ωp ∂t ∂xk ∂xj Re or written in vector-notation Dω 1 2 = (ω · ∇)u + ∇ ω Dt Re .6: Integral along a closed curve C with enclosed area S. since i. We start with the dimensionless momentum equations.√ u0 y = η 4νt t η y (η)·u0 ωz = ω˜√ πνt dxi = ti ds t S C Figure 2. 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 ωj uk this can readily be derived using tensor manipulation. q pqi pqi ∂ ( ∂xq pqi ijk ωj uk ) =− pqi ∂ pqi ∂xq ) of the momentum ∂2p 1 + ∂xq ∂xi Re pqi ∇ 2 ∂ui ∂xq is anti-symmetric in i. q and ∂xq ∂xi is symmetric in ∂2 =0 ∂xq ∂xi The first 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 ∂ωp + pqi ( ijk ωj uk ) = ∇ ωp ∂t ∂xq Re The second term in this equation can be written as ipq ijk ∂ ∂ ∂ ∂ (ωj uk ) = (δpj δqk − δpk δqj ) (ωj uk ) = (ωp uk ) − (ωj up ) = ∂xq ∂xq ∂xk ∂xj uk Due to continuity ∂ωj ∂xj = ∂ 2 uq jpq ∂xj ∂xp ∂ωp ∂uk ∂ωj ∂up + ωp − up − ωj ∂xk ∂xk ∂xj ∂xj = 0 and we are left with the relation ipq ijk ∂ ∂ωp ∂up (ωj uk ) = uk − ωj ∂xq ∂xk ∂xj Thus. The ideal vortex has its name from the fact that its vorticity is zero everywhere.

ωy = 0. We have ωx = 0. Diffusion of vorticity: the Stokes 1st problem revisited. Example 2. where we have used the following dimensional relations √ L2 [ν] = . and is given by u0 −η2 y ω= √ e . where η = √ πνt 2 νt √ where the factor νt can be identified as the diffusion length scale. This is expressed by the following relationship ∂τij = µ∇2 ui = −µ ∂xj ijk ∂ωk ∂xj This can be derived in the following manner. ∂v ∂u − ∂x ∂y ωz = This simplifies the interpretation and use of the concept of vorticity greatly in two-dimensional flow. From the solution of Stokes 1st problem above. t) and that it thus is a solution of the following diffusion equation ∂ω ∂2ω =ν 2 ∂t ∂y The time evolution of the vorticiy can be seen in figure 2. First the diffusion of vorticity and second the advection or transport of vorticity. but zero in the other two directions. [t] = T ⇒ [ νt] = L T . We have −µ ijk ∂ωk ∂xj ∂ ∂ un ∂xj ∂xm ∂ 2 un = −µ(δim δjn − δin δjm ) ∂xj ∂xm = −µ = µ ijk kmn ∂ 2 ui ∂xj ∂xj = µ∇2 ui Terms in the two-dimensional vorticity equation In two dimensions the equation for the only non-zero vorticity component can be written ∂ω3 ∂ω3 ∂ω3 + u1 + u2 = ν∇2 ω3 ∂t ∂x1 ∂x2 We will take two examples elucidating the meaning of the terms in the vorticity equation.7.Components in Cartesian coordinates In cartesian coordinates the components of the vorticity vector in three-dimensions become ω= ex ey ez ∂ ∂x ∂ ∂y ∂ ∂z u v w ∂w ∂v − ∂y ∂z = ∂u ∂w − ∂z ∂x ex + ey + ∂v ∂u − ∂x ∂y ez For two-dimensional flow it is easy to see that this reduces to a non-zero vorticity in the spanwise direction. we find that the vorticity has the following form ω3 = ωz = ω(y. Vorticity and viscocity There exists a subtle relationship between flows with vorticity and flows on which viscous forces play a role. It is possible to show that ∇ × ω = 0 ⇐⇒ viscous forces = 0 This means that without viscous forces there cannot be a vorticity field in the flow which varies with the coordinate directions.

and will be zero on the boundary and approach the inviscid flow far above the plate if we apply the following boundary conditions u = v = 0. We will try a simple modification of the inviscid flow close to the boundary. and find −f f + f f + νf =0 . or the transport/advection term. y → ∞ The vorticity can be written ω = −xf (y). y=0 u ∝ cx.3 0. v ∝ −cy. u Figure 2. At the origin we have a stagnation point.6 0. The stagnation point flow is depicted in figure 2.7 0. The inviscid stagnation point flow is given as follows inviscid: u = cx v = −cy ⇒ ωz = ω = ∂v ∂u − =0 ∂x ∂y Note that this flow has zero vorticity and that it does not satisfy the no-slip condition (u = v = 0. Transport/advection of vorticity: Stagnation point flow. We now turn to an example where the transport of vorticity is important.7: A sketch of the evolution of vorticity in Stokes 1st problem. We introduce this and the above definitions of the velocity components into the vorticity equation.5 ω 0.8 0.i ree vectorfields // to boundary L h x1 x2 p0 > p 1 p1 u n dS u · n∆t x y u0 p0 y η = √4νt η u0 p0 y = √4νt u √ u0 y = η 4νt t ηy y (η)·u0 ωz = ω˜√ πνt t 8 7 6 5 4 t 3 2 1 0 0 0.1 0. y u u0 √ y = η 4νt t η y ω ˜ (η)·u0 √ ωz = πνt t ω=0 x. or equally. Example 3.4 0. y = 0). In order to satisfy this condition we introduce viscosity. v. Thus close to the plate we have to fulfill the equation ∂ω ∂ω ∂2ω ∂2ω u +v =ν + ∂x ∂y ∂x2 ∂y 2 The two first terms represent advection or transport of vorticity. is zero.9 1 Figure 2. vorticity.8: Stagnation point flow with a viscous layer close to the wall.8. In this flow the streamlines are straight lines and therefore the non-linear term. which is written as u = xf (y) v = −f (y) This satisfies the continuity equation.2 0. and is a flow where a uniform velocity approaches a plate where it is showed down and turned to a flow parallel two the plates in both the positive and negative x-direction.

F to the streamwise velocity and F to the vorticity. T [c] = 1 T ν/c and the velocity scale η = F (η) = √ νc to scale the independent and the y ν/c f (y) √ νc This results in the following non-dimensional equation F 2 − FF − F =1 with the boundary conditions F = F = 0. Note that they have the dimensions [ν] = and that we can use the length scale dependent variables as L2 .9 a solution of the equation is shown.9: Solution to the non-dimensional equation for stagnation point flow. F0 √ ωz =F. η = 0 F ∝ η. F is proportional to the normal velocity. f ∝ c. The thin layer of vortical flow close to the wall has the thickness ∼ ν/c.0 η= p0 1. η→∞ In figure 2.ω(η)·u πνt t F F 1 0. It can be seen that diffusion of vorticity from the wall is balanced by transport downward by v and outward by u. y → ∞ An integral to the equation exists.5 √y 4νt u √ u0 y = η 4νt t η y ˜F . The boundary conditions become f =f =0 y=0 f ∝ cy.5 η F 0 0 1 2 η Figure 2. 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 ∂ωi ∂ωi ∂ui 1 2 + uj = ωj + ∇ ωi ∂t ∂xj ∂xj Re advection of vorticity diffusion of vorticity . which can be written f 2 = K = {evaluate y → ∞} = c2 − f f − νf This is as far as we can come analytically and to make further numerical treatment simpler we make the equation non-dimensional with ν and c.

Thus we can draw the following conclusions about the development of the vortex vector i) Stretching of vortex lines (line ω) produce ωi like stretching of dri produces length ii) Tilting vortex lines produce ωi in one direction at expense of ωi in other direction Helmholz and Kelvin’s theorems The analogy with the equation for the material line directly give us a proof of Helmholz theorem.e. tˆ) d ∂ri dm ui ˆ dt m ∂m ∂ui ∂ri ∂ ∂ri dm + ui ˆ ∂m ∂ tˆ ∂m m ∂t m dm The last term of this expression will vanish since ui m ∂ ∂ri ∂ tˆ ∂m dm = ui m ∂ui dm = ∂m m ∂ ∂m 1 ui ui 2 dm = 0 . We have Theorem 1. i.10: A closed material curve.η= √ 4νt u √ u0 dr y = η 4νt t η y (η)·u0 ωz = ω˜√ πνt t C(t) Figure 2. Γm = ui dri C(t) We can evaluate the material time derivative of this expression with the use of Lagrangian coordinates as follows DΓm Dt = = = = D ui dri Dt C(t)   Lagrange coordinates ui = ui (x0i . The complex motion of the vorticity vector for a full three-dimensional flow can be understood by an analogy to the equation governing the evolution of a material line. discussed in the first chapter. The equation for a material line is D ∂ui dri = drj Dt ∂xj Note that if we disregard the diffusion term. tˆ) = ui (x0i (m). Helmholz theorem for inviscid flow: Vortex lines are material lines in inviscid flow. see figure 2. A second theorem valid in inviscid flow is Kelvin’s theorem. We show this by considering the circulation for a closed material curve. tˆ) = ri (x0i (m). Kelvin’s theorem for inviscid flow: Circulation around a material curve is constant in inviscid flow. the last term of the vorticity equation. the two equations are identical. Theorem 2. tˆ)  ri = ri (x0i . Proof.10.

u Figure 2. Re → ∞ Which shows that the circulation for a material curve changes only if the viscous force = 0 Streamfunction We end this section with the definition and some properties of the streamfunction. In two dimensions.12 and figure 2. see figure 2. 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 flow we can find the streamlines using a streamfunction. We have ∂u ∂v ∂2ψ ∂2ψ + = − =0 ∂x ∂y ∂x∂y ∂y∂x Another property of the streamfunction is that the volume flux between two streamlines can be found from the difference by their respective streamfunctions. We can make the following derivation dψ ∂ψ ∂ψ dx + dy ∂x ∂y ∂ψ ∂ψ . on streamlines We can also quickly check that the definition given above satisfies continuity. To define the streamfunction we need the streamlines which are tangent to the velocity vector. v y = η 4νt t η y ˜ 0 √ ωz = ω(η)·u πνt u = (u. We can introduce a streamfunction ψ = ψ(x. y) with the property that the streamfunction is constant along streamlines.√ u0y.11.13 in the following derivations of the volume flux: . v=− and check for consistency = assume u = ∂y ∂x = u dy − v dx = = 0. Thus we find the following resulting expression DΓm Dt = C(t) = C(t) = 1 Re Dui dri Dt − C(t) ∂p 1 2 + ∇ ui dri ∂xi Re ∇2 ui dri → 0.11: A streamline which is tangent to the velocity vector. v) dy t dx x. We use figure 2.

ωz =


πνtu

nt
dS
u · n∆t
x
y
u0
p0
y
η = √4νt

u
u0

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

y = η 4νt
t
η
y
(η)·u0
ωz = ω˜√
πνt

ds

nx

t dy
ny

|n| = 1

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

B

QAB

=

ui ni ds
A
B

=

(nx u + ny v) ds
A

=

1
ds
=
,
dy
nx

ds
1
=
dx
ny

B

=
A

2.3

B

(u dy − v dx) =

A

dψ = ψB − ψA

Potential flow

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 definition of the velocity potential, then
discuss simplifications of the momentum equations when such potentials exist, and finally we deal with
two-dimensional potential flows, 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 =

∂φ
∂xi

or u = ∇φ

with the corresponding relations in two dimensions




 u =



 v

=

∂φ
∂x
∂φ
∂y

There are several consequences of this definition of the velocity field.

i) Potential flow have no vorticity, which can be seen by taking the curl of the gradient of the velocity
potential, i.e.
∂uk
∂2φ
ωi = ijk
= ijk
=0
∂xj
∂xj ∂xk
ii) Potential flow is not influenced by viscous forces, which is seen by the relationship between viscous
forces and vorticity, derived earlier. We have
∂τij
= µ∇2 ui = −µ
∂xj

ijk

∂ωk
∂xj

iii) Potential flow satisfies Laplace equation, which is a consequence of the divergence constraint applied
to the gradient of the velocity potential, i.e.
∂ui
∂2φ
=
= ∇2 φ = 0
∂xi
∂xi ∂xi

Bernoulli’s equation
For potential flow the velocity field can thus be found by solving Laplace equation for the velocity potential.
The momentum equations can then be used to find the pressure from the so called Bernoulli’s equation.
We start by rewriting the momentum equation
∂ui
∂ui
1 ∂p
+ uj
=−
+ ν∇2 ui − gδi3
∂t
∂xj
ρ ∂xi
using the alternative form of the non-linear terms, i.e.
uj

∂ui

=
∂xj
∂xi

1
uk uk
2

ijk uj ωk

and obtain the following equation
∂ui

+
∂t
∂xi

1
p
uk uk + + gx3
2
ρ

=

ijk uj ωk

ijk

∂ωk
∂xj

This equation can be integrated in for two different cases: potential flow (without vorticity) and stationary inviscid flow with vorticity.
∂φ
into the momentum equation
i) Potential flow. Introduce the definition of the velocity potential ui = ∂x
i
above, we have

∂φ 1
p
+ uk uk + + gx3 = 0 ⇒
∂xi ∂t
2
ρ

p
∂φ 1 2
+ (u1 + u22 + u23 ) + + gx3 = f (t)
∂t
2
ρ
ii) Stationary, inviscid flow with vorticity. Integrate the momentum equations along a streamline, see
figure 2.14. We have
ti


∂xi

1
p
uk uk + + gx3
2
ρ

ds =

ijk

ui
uj ωk ds = 0
u

1 2
p
(u + u22 + u23 ) + + gx3 = constant along streamlines
2 1
ρ
Thus, Bernoulli’s equation can be used to calculate the pressure when the velocity is known, for two
different flow situations.
Example 4. Ideal vortex: Swirling stationary flow without vorticity. The velocity potential u = ∇φ in
polar coordinates can be written

∂φ



 ur = ∂r



 uθ

=

1 ∂φ
r ∂θ

h
u
x1t
xη2
p˜0 > py01

ωz = ω(η)·u
p1
πνt
ut
n
t = uu
dS
u · n∆t
Figure 2.14: Integration
of the momentum equation along a streamline.
x
y
u0
r
p0
y
η = √4νt

u
u0

y = η 4νt
t
η
y
˜
0

ωz = ω(η)·u
πνt
t
Figure 2.15: Streamlines for the ideal vortex.
The velocity potential satisfies Laplace equation, which in polar coordinates becomes
∇2 φ =
The first term vanishes since

∂φ
∂r

1 ∂
r ∂r

1 ∂φ
r ∂r

+

1 ∂2φ
=0
r ∂θ2

= ur = 0. Integrate the resulting equation twice and we find
φ = Cθ + D

The constant D is arbitrary so we can set it to zero. Using the definition of the circulation we find
uθ =

C
,
r

C=

Γ

We can easily find the streamfunction for this flow. In polar coordinates the streamfunction is defined

1 ∂ψ



 ur = r ∂θ



 uθ

= −

∂ψ
∂r

Note that the streamfunction satisfies continuity, i.e. ∇ · u = 0. We use the solution from the velocity
potential to find the streamfunction, we have
∂ψ
Γ
=−
∂r
2πr

ψ=−

Γ
ln r

The streamfunction is constant along streamlines resulting in circular streamlines, see figure 2.15
We can now use Bernoulli’s equation to calculate the pressure distribution for the ideal vortex as
1 2
u + p = p∞
2 θ

p = p∞ −

Γ2
8π 2 r2

Note that the solution has a singularity at the origin.

Two-dimensional potential flow using analytic functions
For two-dimensional flows 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 flow.

e. y) − iψ(x. We have  2 ∂ φ ∂ ∂ψ ∂ ∂ψ ∂2φ   = = = −   ∂x2  ∂x ∂y ∂y ∂x ∂y 2   ∂2ψ ∂   =  2 ∂y ∂y ∂φ ∂x = ∂ ∂x ∂φ ∂y =− ∂ 2ψ ∂x2 .16. ∂φ ∂ψ ∂φ ∂ψ = and =− ∂x ∂y ∂y ∂x Using the Cauchy-Riemann equations we can show that the real φ and imaginary part ψ of the complex potential satisfy Laplace equations and thus that they are candidates for identification with the velocity potential and streamfunction of a two-dimensional potential flow. Velocity potential φ for two-dimensional flow is defined  ∂φ     u = ∂x Using the continuity equation gives     v = ∂φ ∂y ∂u ∂v + = ∇2 φ = 0 ∂x ∂y The streamfunction ψ is defined      u =     v ∂ψ ∂y = − ∂ψ ∂x By noting that potential flow has no vorticity we find −ω = − ∂v ∂y + = ∇2 ψ = 0 ∂x ∂y Thus. both the velocity potential φ and the streamfunction ψ satisfy Laplace equation. 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. We now recall some useful results from analytic function theory. y + ∆y) − iψ(x. ⇐⇒ 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. i. y) + iψ(x. y)] i∆y→0 i∆y  ∂ψ ∂φ   = real part   ∂x ∂y ⇒   ∂ψ 1 ∂φ ∂ψ ∂φ   i = ⇒ =− imaginary part ∂x i ∂y ∂x ∂y Thus we have the Cauchy–Riemann equations as a necessary requirement for a complex function to be analytic.We start by showing that both the velocity potential and the streamfunciton satisfy Laplace equation. see figure 2. y) − φ(x. We have 1 F (z) = lim [F (z + ∆z) − F (z)] ∆z→0 ∆z 1 = lim [φ(x + ∆x. y) + iψ(x. y + ∆y) − φ(x. y)] ∆x→0 ∆x 1 = lim [φ(x. y) + iψ(x + ∆x. y) where z = x + iy = reiθ = r(cos θ + i sin θ) which will be denoted the complex potential. We can utilize analytic function theory by introducing the complex function F (z) = φ(x.

where the real part is the velocity potential and the imaginary part is the streamfunction. In polar coordinates the complex potential becomes Arn einθ = Arn cos(nθ) + iAr n sin(nθ) Thus the velocity potential and the streamfunction can be identified as φ = Arn cos(nθ) ψ = Arn sin(nθ) . we also have to show that the Cauchy-Riemann equations are consistent with the definition of the velocity potential and the streamfunciton.16: The complex plane and the definitions of analytic functions.17.18. Calculate the velocity field from the complex potential F = U e−iα z. In this case we have u = ur cos θ − uθ sin θ v = ur sin θ + uθ cos θ W (z) = u − iv = [ur (r. see figure 2. The complex velocity becomes W (z) = F = U e−iα = U (cos α − i sin α) which gives us the solution for the velocity field in physical variables. as u = U cos α v = U sin α Example 6. Calculate the velocity field from the complex potential F = Az n . The Cauchy–Riemann equations are  ∂φ ∂ψ   u= =   ∂x ∂y   ∂ψ ∂φ   v= =− ∂y ∂x which recovers the definitions of the velocity potential and the streamfunction. In addition to satisfying the Laplace equations. θ)] e−iθ cos θ−i sin θ We will now take several examples of how potential velocity fields can be found from analytic functions. see figure 2. We can now define the complex velocity as the complex conjugate of the velocity vector since W (z) = dF ∂φ ∂ψ = +i = u − iv dz ∂x ∂x It will be useful to have a similar relation for polar coordinates. Thus any analytic complex function can be identified with a potential flow. θ) − iuθ (r. Example 5.η= u0 p0 iy √y 4νt u √ u0 y = η 4νt t η y ω ˜ (η)·u0 √ ωz = πνt t x i∆y ∆x Figure 2.

u · n∆t P2 x xy1 x u20 x p03 y 3 √ R η = 4νt Ru2 u01 √R y y = η 4νt3 r t v rη2 1 r π dϕ12y0 2 + (η)·u ωz = ω˜√ u ˜ Vπνt (t) S (t)t uθ V (t + ∆t) − V (t) S (t + ∆t) ur u u n V (t + ∆t) r p T0 . U0 Gradient fields wi ∂p ∂xi ui Divergence free vectorfields // to boundary L h x1 x2 p0 > p 1 p1 u n dS u · n∆t x y u0 p0 y η = √4νt √ u u0 y = η 4νt t η y (η)·u0 ωz = ω˜√ πνt α t Figure 2. .17: The definition of polar coordinates.18: Constant velocity at an angle α. U 0 θ Tw x L T0 Figure 2.

xy3 ω ˜ (η)·u0 √ ωz = πνt R3 R2t Ψ=0 R1 r3 r2Figure 2. The streamfunction is zero on the walls and on L By solving for the zero streamlines we can calculate the angle between a the stagnation point streamline.21 shows the stagnation point flow when n = 2.21: Stagnation point flow when n = 2. We have U0 Gradient fields ψ = Arn sin(nθ) = 0 ⇒ θ = πk wi n ✒✑✏ ✒✑✏ ✒✑✏ ✒✑✏ ✒✑✏ ✒✑✏ ✒✑✏ ✒✑✏ ✒✑✏ ✒✑✏ ✒✑✏ ✒✑✏ ✒✑✏ ✒✑✏ ✒✑✏ ✒✑✏ ✒✑✏ ✒✑✏ ✒✑✏ ✒✑✏ ✒✑✏ ✒✑✏ ✒✑✏ ✒✑✏ ✒✑✏ ✒✑✏ ✒✑✏ ✒✑✏ ✒✑✏ ✒✑✏ ✒✑✏ ✒✏ ∂p ∂xi Figure 2.22) is u · n∆t ur = nArn−1 cos(nθ) ⇒ {θ = 0} ⇒ u = ur = nArn−1 = Cxn−1 x The wedge angle is y 2π n−1 u0 ϕ = 2π − = 2π n n p0 y η = √4νt √ u u0 y = η 4νt t η y ω ˜ (η)·u0 √ ωz = πνt ✓✔✓✔✕✔✓✕✔✓ ✕✔✓✕✔✓ ✕✔✓✕✔✓ ✕✔✓✕✔✓ ✕✔✓✕✔✓ ✕✔✓✕✔✓ ✕✔✓✕✔✓ ✕✔✓✕✔✓ ✕✔✓✕✔✓ ✕✔✓✕✔✓ ✕✔✓✕✔✓ ✕✔✓✕✔✓ ✕✔✓✕✔✓ ✕✔✓✕✔✓ ✕✔✓✕✔✓ ✕✔✓✕✔✓ ✕✔✓✕✔✓ ✕✔✓✕✔✓ ✕✔✓✕✔✓ ✕✔✓✕✔✓ ✕✔✓✕✔✓ ✕✔✓✕✔✓ ✕✔✓✕✔✓ ✕✔✓✕✔✓ ✕✔✓✕✔✓ ✕✔✓✕✔✓ ✕✓✕✓ t Figure 2.20: Uniform flow over a flat plate. Here the analysis can be done in Cartesian Divergence coordinates free vectorfields // 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. see figure 2. r1 π + dϕ 12 2 and we can calculate the complex V (t) velocity as S (t)W (z) = F = nAz n−1 V (t + ∆t) − V (t) = nArn−1 einθ e−iθ S (t + ∆t) = nArn−1 [cos(nθ) + i sin(nθ)]e−iθ u n velocity in polar coordinates as This gives us the solution for the V (t + ∆t) ur = nArn−1 cos(nθ) p uθ = −nArn−1 sin(nθ) T0 .20 shows the uniform flow over a flat plate which results when n = 1.xi y =uη· n∆t 4νt Ψ=0 ui dxttˆ dxiyη0 Ψ=0 dui udy0tˆ (η)·u dtˆp00 ωz = rω˜i√ y ηdr= √πνt tˆ i d4νt ut π u01 √P n y = η 4νt P2 Ψ=0 x1t η x2 Figure 2.22 shows the flow towards a wedge when 1 ≤ n ≤ 2. The velocity along the wedge edge (x axis dS in figure 2. T0 wall aligned with the x-axis and the stagnation point streamline. U 0 Tw This is a flow which approaches a corner. ui Figure 2.19. .19: Flow towards a corner.

u u0 L y = η 4νt h t x1 η x2 py0 > p1 ϕ (η)·u0 p1 ωz = ω˜√ πνt u t n dS u · n∆t Figure 2. x y u0 p0 y η = √4νt √ √ u u0 y = η 4νt t η y (η)·u0 ωz = ω˜√ πνt t Figure 2. We now list several other examples of potential flow fields obtained from analytic functions. F (z) = −i = Γ ln z 2π Γ [θ − i ln r] 2π .23: Line source. F (z) = = = m ln z 2π m ln(reiθ ) 2π m [ln r + iθ] 2π The velocity potential and the streamfunction become  m    φ = 2π ln r The complex velocity can be written as    ψ= mθ 2π W (z) = F = m m −iθ = e 2πz 2πr which gives the solution in polar coordinates.23. see figure 2. Line source. Example 7. as  m   ur = 2πr   uθ = 0 We can now calculate the volume flux from the source as 2π Q= 0 m r dθ = m 2πr Example 8. Line vortex.22: The flow over a wedge when 1 ≤ n ≤ 2.

most of the flow can be considered inviscid and a simpler set of equations.η= x y u0 p0 √y 4νt u √ u0 y = η 4νt t η y ˜ 0 √ ωz = ω(η)·u πνt t Figure 2. Airfoils. The potential flow around a cylinder has the complex potential F (z) = U z + U r02 z Example 11. where the viscous effects are small. there will be a layer close to the solid walls where the viscous terms are important even for very high Reynolds number flow. the no flow condition can be satisfied at a solid wall but not the no slip condition. The inviscid flow around some airfoils can be calculated with conformal mappings of solutions from cylinder with circulation. . see figure 2. The dipole has the complex potential F (z) = m πz Example 10. 2. In order to satisfy the no slip condition we need to add the viscous term. The velocity potential and the streamfunction become The complex velocity can be written as  Γ    φ = 2π θ    ψ = − Γ ln r 2π W (z) = F = −i Γ −iθ e 2πr which gives the solution in polar coordinates.24: Ideal vortex. the highest derivatives are present in the viscous terms. thus a smaller number of boundary conditions can be satisfied when solving the Euler equations. However.24. as    ur = 0   uθ = Γ 2πr Example 9. The Euler equations are obtained if Re → ∞ in the Navier-Stokes equations. This layer will be very thin and the flow in that region is called boundary layer flow. In particular. The potential flow around a cylinder with circulation has the complex potential F (z) = U z + U r02 Γ z −i ln z 2π r0 Example 12. can be solved. the Euler equations. Thus.4 Boundary layers For flows with high Reynolds number Re.

26: Boundary layer profile close to a solid wall.25: Estimatet of the viscous region for an inviscid vorticity free inflow in a plane channel. We obtain the following estimates for the size of the terms in the normal momentum equation . as Inviscid : U velocity scale for u. Where ts is time of a L fluid particle with velocity U has travelled length L. ts = . we √ can estimate the distance the vorticity has diffused from the wall toward the channel centerline as δ ∼ νts . We choose different length scales inside boundary layer and in inviscid region as indicated in figure 2. we expect large velocity gradients near walls. see figure 2.26. Since we expect this region to be thin for high Reynolds numbers.ωz = ω ˜ (η)·u0η √ πνt = 0 √y 4νt u √ u0 }δ t ω ≈ 0 y = η 4νt U→ t η L y ω(η)·u ˜ 0 ωz = √πνt Figure 2. Ly. Thus we can estimate δ/L as U √ 1/2 δ νts 1 νL ν 1/2 1 → 0 as Re → ∞ ⇒ = = = =√ L L L U UL Re We will now divide the flow into a central/outer inviscid part and a boundary layer part close to walls. By analogy to the Stokes solution for the instantaneously starting plate. where the no slip condition is fulfilled. u L Figure 2. The unknown velocity scale for the normal velocity α is determined using continuity ∂u ∂v + =0 ∂x ∂y  O  U  L O α δ which implies that U α = L δ ⇒ α= δ U L In this manner both terms in the continuity equation have the same size. i.25.e. y BL : : : : U α L δ velocity scale for velocity scale for length scale for length scale for u v x y The pressure scale is assumed to be ρU 2 for both the inner and the outer region. v : L length scale for x. Let us use these scalings in the steady momentum equations and disregard small terms. v δ ω≈0 U→ } U δ x. We begin this section by estimating the thickness of the viscous region in channel with inviscid inflow (zero vorticity).

y = 0) u (x. the only choice that gives a consistent approximation. When R → ∞ and δ/L is small. There are several things to note about these equations. the system of equations is a third order system in y and thus only three boundary conditions can be enforced. y) = uin (y) BC : u (x.↑y: ∂v ∂x 1 δU U· L L u + ∂v ∂y δU 1 δU · · L δ L v 2 δ L δ L = − 1 ∂p ρ ∂y + ν U2 δ ν L2 2 1 Re 1 ∂2v ∂x2 ∂2v ∂y 2 ν δU δ2 L + ν δU L · δ L 2 1 Re where the last line was obtained by dividing all the terms on the second line by U 2 /δ. the pressure is constant through the boundary layer. y → ∞) = = = 0 0 ue (x) outer inviscid flow where the initial condition is at x = x0 and the equations are marched in the downstream direction. In order to keep the last term. Thus we are left with the parabolic equation in x since the second derivative term in that direction is neglected. only pressure term O(1). which implies that ∂p =0 ∂y ⇒ p = p(x) Thus. vin cannot be given as its own initial condition. y = 0) v (x. 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. we have to assume that 1 δ ∼√ L Re which is the same estimate as we found for the extent of the boundary layer in the channel inflow case. Thus no boundary condition for v in the free stream can be given. We have u ∂u ∂u 1 ∂p ∂2u +v =− +ν 2 ∂x ∂y ρ ∂x ∂y In summary. Next we estimate the size of the terms in the streamwise momentum equation as →x: u ∂u ∂x + v ∂u ∂y = − 1 ∂p ρ ∂x + ν ∂2u ∂x2 U2 L δU U · L δ U2 L ν ·U L2 1 1 1 1 Re + ν ∂2u ∂y 2 ν U δ2 1 · Re L δ 2 The first three terms are order one and the fourth term negligible when the Reynolds number become large. Second. We have the no flow and the no slip conditions at the wall and the u = ue in the free stream. the steady boundary layer equations for two-dimensional flow is  ∂u ∂u 1 ∂p ∂ 2u   u +v =− +ν 2 ∂x ∂y ρ ∂x ∂y   ∂u + ∂v = 0 ∂x ∂y with the initial (IC) and boundary conditions (BC) IC : u (x = x0 . given by the inviscid outer solution. . First.

We give two below v=− ∂Ψ ∂η = −u0 δ f + δf ∂x ∂x y = −u0 δ f − f δ δ = u0 (ηf − f ) δ . y) = u0 δ (x) f (η) We can now evaluate the various terms in the momentum equation. y → ∞) = = = 0 0 u0  ∂u ∂2u ∂u   u +v =ν 2 ∂x ∂y ∂y   ∂u + ∂v = 0 ∂x ∂y We introduce the two-dimensional stream function Ψ u= ∂Ψ . since the thickness of the boundary layer is zero from the inviscid point of view. Let u= ∂Ψ = u0 f (η) . The boundary layer equations become IC : u (x = x0 .u0 U→ y. ∂y v=− ∂Ψ ∂x which implicitly satisfies continuity.e. y = 0) u (x. We have 1 p (x) + ρu2e (x) = const 2 which implies that − 1 dp due = ue ρ dx dx Blasius flow over flat plate Consider the boundary layer equations for a flow over a flat plate with zero pressure gradient. the pressure gradient term in the momentum equation can be written in terms of the edge velocity ue using the Bernoulli equation.27: Boundary layer flow over a flat plate with zero pressure gradient. Fourth. ue (x) = uinv (x. Similarity solution We will now look for a similarity solution. The boundary layer equations become ∂Ψ ∂ 2 Ψ ∂Ψ ∂ 2 Ψ ∂3Ψ · − · = ν ∂y ∂x∂y ∂x ∂y 2 ∂y 3 with appropriate initial and boundary conditions.e. i. y) = u0 BC : u (x. ∂y η= y δ (x) which by integration in the y-direction implies Ψ (x. v x. y = 0) v (x. i. the boundary layer solution in the free stream approaches the inviscid solution evaluated at the wall. y = 0). Third. u L U δ u0 ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✗✖✘✗✖ ✘✖✘✖ Figure 2. the outer inviscid flow is ue (x) = u0 .

This implies f u0 δδ 1 = ν 2 ⇒ + δδ dx = 1 2 ν dx u0 ⇒ 1 2 1 νx δ = 2 2 u0 νx .99 η99 = δ99 νx u0 = 4.28. The boundary layer thickness can be evaluated as the height where the velocity has reached 99% of the free stream value.8 u u0 1 Figure 2. ∂2Ψ η = −u0 f δ ∂x∂y δ If these and the other corresponding terms are introduced into the momentum equation we have the following equation for f −u20 f f η f δ f δ + u20 (ηf − f ) = νu0 2 δ δ δ which can be simplified to u0 δδ ff = 0 ν For a similarity solution to be possible.29 we can evaluate this as y0 0 which implies that y u dy ≡ y u0 dy = δ∗ 0 u0 dy − δ ∗ u0 .9 νx u0 x. v x.9 ⇒ where it can be seen that the boundary layer thickness grows as √ δ99 = 4. With the help of figure 2. We are left with the Blasius equation and boundary conditions which implies that δ (x) = 1 + ff = 0 2 f BC : f (0) = f (0) = 0 f (∞) = 1 The numerical solution to the Blasius equation gives the self-similar velocity profile seen in figure 2.28: Similarity solution giving the Blasius boundary layer profile. u η=√ yL νx/u0 U δ u0 } Blasius profile 5 4 3 2 1 0 0 0. Displacement thickness and skin friction coefficient Another measure of the boundary layer thickness is the displacement δ ∗ of the wall needed in order to have the same volume flux of the inviscid flow as for the flow in the boundary layer.4 0.2 0. We have ⇒ f (η99 ) = 0.L δ ω≈0 U→ y.6 f = 0. the coefficient in front of the f f term has to be constant. where we have chosen the constant of integration such that δ = 0 u0 when x = 0.

ωz =


πνt

U
δ
u0

t
L√ y
η=
νx/u0

u
f
ω ≈ 0 = u0
Blasius
U → profile
y, v
x, u
L
U
δ
u0
η=√ y

δ∗

✛✚✙ ✛✚✙ ✛✚✙ ✛✚✙ ✛✚✙ ✛✚✙ ✛✚✙ ✛✚✙ ✛✚✙ ✛✚✙ ✛✚✙ ✛✚✙ ✛✚✙ ✛✚✙ ✛✚✙ ✛✚✙ ✛✚✙ ✛✚✙ ✛✚✙ ✛✙

✢✚✜ ✢✚✜ ✢✚✜ ✢✚✜ ✢✚✜ ✢✚✜ ✢✚✜ ✢✚✜ ✢✚✜ ✢✚✜ ✢✚✜ ✢✚✜ ✢✚✜ ✢✚✜ ✢✚✜ ✢✚✜ ✢✚✜ ✢✚✜ ✢✚✜ ✢✜

Figure 2.29: The displacement thickness δ ∗
⇐= Pressure force

νx/u0

f = uu0
Blasius profile
δ∗

✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✤✣ ✥✣

x

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

δ∗ =

1−

0

u
u0

dy =

νx
u0


0

[1 − f (η)] dη

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

νx
u0 .

A quantity which is of large importance in practical applications is the skin friction coefficient. The
skin friction at the wall is
τw = µ

∂u
∂y

y=0

=

µu0
ρu20
0.332ρu20

f
(0)
=
f (0) =
u0 x
δ
Rex
ν

where the last term is the expression evaluated for Blasius flow. This can be used to evaluate the skin
friction coefficient as
Cf =

τw
1
2
2 ρu0

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 flow region close to wall, see figure 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

u

∂u
∂u
1 dp
∂2u
+v
=−
+ν 2
∂x
∂y
ρ dx
∂y

take the normal derivative and use continuity we have
u

∂2u
∂2u
∂3u
+v 2 =ν 3
∂x∂y
∂y
∂y

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

∂u
∂y

2

+u

∂3u
∂u ∂ 2 u
∂3u
∂4u

+v 3 =ν 4
2
2
∂x∂y
∂x ∂y
∂y
∂y

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

1 ∂
2 ∂x

∂u
∂y

2

∂4u
∂y 4


y=0

y=0

±κ2

which can be integrated to yield
∂u
∂y

2

= ±κ2 x + C

y=0

∂u
∂y

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

y=0

=0

this expression is

y=0


= κ 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 flow 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 flow

Reynolds average equations
Turbulent flow is inherently time dependent and chaotic. However, in applications one is not usually
interested in knowing full the details of this flow, but rather satisfied with the influence of the turbulence
on the averaged flow. For this purpose we define 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 flow.
We are now going to derive an equation governing the mean flow Ui . Divide the total flow into an
average and a fluctuating component ui as
ui = ui + ui

p=p+p

and introduce into the Navier-Stokes equations. We find
∂ui
∂ui
∂ui
∂u
∂ui
∂u
1 ∂p
1 ∂p
+
+ uj
+ uj i + uj
+ uj i = −

+ ν∇2 ui + ν∇2 ui
∂t
∂t
∂xj
∂xj
∂xj
∂xj
ρ ∂xi
ρ ∂xi
We take the average of this equation, useing
ui = 0

u j ui = u i · uj = 0

which gives




∂ui
∂ui
1 ∂p



+ uj
=−
+ ν∇2 ui −
ui uj

∂t
∂xj
ρ ∂xi
∂xj

 ∂ui = 0



 ∂xi

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





∂Ui
∂Ui
1 ∂P




+ Uj
=−
+
(τij − ui uj )
∂t
∂xj
ρ ∂xi
∂xj
∂Ui





 ∂xi

where ui uj is the Reynolds stress. Thus the effect 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

The diagonal component of this stress is the kinetic energy of the turbulent fluctuations
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 flow, but is only interested in
the mean, the effect 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 fluids. We have
Rij =

2
kδij − 2νT E ij
3

The first 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 flow, i.e.
E ij =

1
2

∂Ui
∂Uj
2 ∂U 
 r δij
+

∂xj
∂xi

∂xr

where we have assumed incompressible flow. Introducing this into the Reynolds average equations gives
∂Ui

∂Ui
+ Uj
=

∂t
∂xj
∂xj

P
2
1 ∂P
+ k δij + 2 (ν + νT ) E ij = −
+ (ν + νT ) ∇2 Ui
ρ
3
ρ ∂xi

where the last equality assumes that νT is constant, an approximation only true for very simple turbulent
flow. It is introduced here only to point out the analogy between the molecular viscosity and the turbulent
viscosity.
The turbulent viscosity νT must be modelled and most numerical codes which solved the Navier-Stokes
equations for practical applications uses some kind of model for νT .
For simple shear flows, i.e. velocity fields of the form U (y), the main Reynolds stress component becomes
uv = −νT

∂U
∂y

where, in analogy with kinetic gas theory, νT
νT = 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.

and find ∂ ∂t 1 ui ui 2 + Uj ∂ ∂xj 1 ui ui 2 = −ui uj ∂Ui ∂ + ∂xj ∂xj − puj ν ∂ 1 − ui ui uj + ui ui ρ 2 2 ∂xj −ν ∂ui ∂ui ∂xj ∂xj where is the mean dissipation rate of the turbulent kinetic energy. as Prandtl’s estimation of mixing length we get dU uT uτ = = dy l κy ⇒ U 1 = ln y + C. We can derive such an equation by taking the previously derived equation for Ui + ui and subtracting the equation for U . uτ κ κ is the von Karman constant.31.41. called the log-region. the next level of modelling uses the assumption that 1 ui ui 2 and a differential equation is used to calculate the turbulent kinetic energy. In simple free shear flows like jets. Thus the mixing length is proportional to the distant from the wall and the region close to the wall in a turbulent flow has a logarithmic velocity profile. One-equation models Instead of modelling the characteristic turbulent velocity algebraically. wakes and shear layers we can take νT as constant in normal direction or y-direction and to vary in a self-similar manner in x. . approximately equal to 0. Zero-equation models The simplest model for uT and l is to model them algebraically. We find uT = k= ∂ui ∂ui ∂Ui ∂ui ∂ + Uj = −uj − uj + ∂t ∂xj ∂xj ∂xj ∂xj p ∂ui − δij + ν + u i uj ρ ∂xj We now take the average of the inner product between the turbulent fluctuations u i and the above equation.31: Prantdl’s mixing length l is the length for which a fluid particle displaced in the normal direction can be expected to retain its horizontal momentum. the Reynolds average equation. We have now introduced a number of new unknowns which have to be modelled.νx/u0 f = uu0 Blasius profile δ∗ x Separation point ⇐= Pressure force y uT l ✦✧✦✧ ✦ ★✧ ✦ ★✧ ✦ ★✧ ✦ ★✧ ✦ ★✧ ✦ ★✧ ✦ ★✧ ✦ ★✧ ✦ ★✧ ✦ ★✧ ✦ ★✦ ★ ★✧ x Figure 2. We can thus evaluate uT and find νT uT = l dU dy ⇒ Define τw as the shear stress at the wall and uτ = νT = l 2 dU dy τw /ρ as the skin friction velocity. This is a normal distance over which the particle retains its momentum and is depicted in figure 2. In wall bounded shear flow we can take l as Prandtl’s mixing length (1920’s). With l =κ·y uT = u τ .

First. We have = cD k 3/2 l L2 T . k 2 . Second. We find Dk i ui uj ∂U = (ν + νT ) ∇2 k − ∂xj Dt rate of increase diffusion rate generation rate − dissipation rate Two-equation models The most popular model in use in engineering applications is a two equation model.e. For simplicity we write the resulting equation for the turbulent kinetic energy for constant ν T . i. i. the so called k. Third. This is based on the above model for k but instead of modelling a partial differential equation is derived governing the turbulent dissipation. Once k and is calculated the turbulent viscosity is evaluated as a quotient of the two. a generation term and a dissipation term.-model. since the dimension of is [ ] = Here cD is a modelling constant. 1 1 νT ∂k − ui ui uj − puj = 2 ρ P rk ∂xj where P rk is the turbulent Prandtl number. we model the Reynolds stress as previously indicated. with a diffusion term. the turbulent dissipation is modelled purely based on dimension using k and l. a gradient diffusion model is used for the transport terms. such that the dimension is correct. This equation has the form as the equation for k. Many new model constants need to be introduced in order to close the equation.e. We have ⇒ νT = Cµ · where Cµ is a modelling constant. ui uj = 32 kδij − νT ∂Ui ∂xj + ∂Uj ∂xi .

.

2. v u We now apply this to an x.1 Finite Volumexy method on arbitrary grids u0 Equations with 1st order p0 derivatives: the continuity equation y √ η= In finite volume methods one4νt makes approximations of the differential equations in integral form. the continuity equation. We start u u0 order derivatives.1: Normal vector to the curve l.j ) /2 2 Blasius profile δ∗ y x Separation point ⇐= Pressure force x dx y l uT ∆yab = yb − ya x dl = ( dx. 59 .j ≈ (vi+1. This implies that }δ ω≈0 uk nk dl = (u dy − v dx) = 0 U→ S S y. arbitrary two-dimensional grid defined as in figure 3. dy) dy − dx dy n dl = ( dy. see figure 3.j ≈ (ui+1. − dx) Figure 3. Recall that the integral from of with an equation with only√first y = η 4νt this equation can be written t η ∂ (uk ) dV = uk nk dS = 0 y ∂xk ω ˜ (η)·u0 V S √ ωz = πνt Chapter 3 Finite volume methods for incompressible flow Evaluating the normal vector t ni to the curve l for two-dimensional flow.j + vi. − dx).j ) /2 ∆xab = xb − xa u0 y η=√ νx/u0 f =vabuu0= vi+ 1 . we have the normal with length dl as n dl = L ( dy.L T0 U0 Gradient fields wi ∂p ∂xi ui Divergence free vectorfields // to boundary L h x1 x2 p0 > p 1 p1 u n dS u · n∆t 3. L definitions and approximations We make use of the following U uab δ= ui+ 12 .j + ui.1.

j ∆xcd + i− 21 .j ∆yab − vi+ 12 .j−1 ) ∆xda 2 2 Note that dividing with ∆yab · ∆xda gives a standard central difference discretization of the continuity equation. We find 0= V ∂2Φ dV = ∂xj ∂xj S ∂φ nj dS = {2D} = ∂xj ∂Φ ∂Φ dy − dx ∂x ∂y S If we approximate this integral in the same manner as for the continuity equation we have ∂Φ ∂x ∂Φ ∂x i+ 21 .j ∆xcd + ui.j i− 21 .j ∆yab + vi.j+ 12 ∆ybc − vi.2: Arbitrary grid.j ∆ycd − vi− 12 .j ∆yab − ∂Φ ∂y ∆ycd − ∂Φ ∂y ∆xab + i+ 21 .i. with analogous expressions for the quantities on the bc. We can now evaluate the integral associated with the two-dimensional version of the continuity equation for the surface defined by the points abcd. j + 1 c i − 1.j ) ∆yab + (vi.j− 21 . ∆xbc = −∆xda which implies that the approximation of the continuity equation can be written (u dy − v dx) ≈ ui+ 21 . j − 1 Figure 3. ui+1.✪j ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✫ ✩✪✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✫ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✫ Aabcd ✩✪✩✪ ✩✪✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✫ d i − 1.j+ 12 ∆xbc + S ui− 21 .j+1 − vi.j+ 12 + vi.j − ui−1.j− 12 ∆xda = S 1 1 (ui+1. j i + 1.j− 21 ∆xda On a cartesian grid we have the relations ∆xab = ∆xcd = ∆ybc = ∆yda = 0 and ∆ycd = −∆yab .j ∂Φ ∂x ∂Φ ∂x i. i.e.j+ 21 ∆xda i.j − ui− 12 .j− 12 ∆yda − vi. Note that the grid lines need not be parallel to the coordinate directions. j − 1 i + 1. j a a d i.j ∆xab + ui. j − 1 b i + 1.j− 21 ∆ybc − ∂Φ ∂y ∆yda − ∂Φ ∂y ∆xbc + i. j + 1 ✩✪✩✪ ✫ c ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ b✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✫ ✩✪✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✫ ✩✪✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✫ ✩✪✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✫ ✩✪✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✪ ✫ ✩✫ ✩i. j + 1 i − 1. We have (u dy − v dx) ≈ ui+ 21 .j−1 + =0 2∆xda 2∆yab 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.j+ 21 i.j+1 − vi. cd and da faces.j vi.j − ui−1.

as coefficients. i+ 21 .j− 21 We can evaluate the derivatives as ∂Φ ∂x i+ 21 .j−1 ) ∆x ∆y Dividing with the area ∆x · ∆y we obtain the usual 5-point Laplace formula. Φ13 .j + Φi.First derivatives are evaluated as mean value over adjacent control volumes (areas in the two-dimensional case). etc.j−1 + Bi.j−1 + Φi+1. −Φ dx) abcd where Φ dy ≈ Φi+1.j ≈ 1 Aa b c d (Φ dy.j ∆yc d + Φ∆yd a abcd and the area evaluated as half magnitude of cross products of diagonals.j Φi−1. j.j Φi.j − 2φi.j + Di.j ∆ya b + Φb ∆yb c + Φi. Φi+1. We use Gauss theorem over the area defined by the points a b c d .j ∂Φ ∂Φ dy − dx ∂x ∂y ∂Φ ∂y i. .j+ 21 − ∂Φ ∂y ≈ ∆x i.j + Φi.j Φi.j Φi. The matrix A is a N × N with the terms Aij .j−1 + Hi. we instead work out the expression for cartesian grids in two-dimensions.j−1 ) 4 ∂Φ Evaluating the derivatives ∂Φ ∂x . we obtain N equations which in matrix form can be written Ax = b Here xT = (Φ11 .j + Φi.j−1 + Ei. Φmn ) and b contains the boundary conditions. which implies that ∂Φ ∂Φ dy − dx ∂x ∂y ≈ (Φi+1.j Φi+1.j+1 + Ii. i.j + Gi.j+1 − 2Φi.j − Φi.j ) ∆y ∆x + (Φi. . Rather than showing the general case in more detail.j = (Φi+1.j+1 Fi. Bij .j Φi−1.j = 0 + If we have a total of N = m × n interior nodes. i.j − 2Φi. For the threedimensional case we have 1 V ∂Φ 1 dV = ∂xj V V Φnj dS ⇒ S which in the two-dimensional case is approximated by ∂Φ ∂x ∂Φ ∂y .j+1 + Ci. ∂y along the other cell faces and substituting back we obtain a 9 point formula which couples all of the nine points around the point i.j + Φi−1. We find ∂2Φ ∂2Φ + ∂x2 ∂y 2 ∂Φ ∂x i+ 21 . Φ12 . e.j i+ 21 .j + Φi+1. Aab ≡ Aa b c d = 1 ∆xd b · ∆ya c − ∆yd b · ∆xa c 2 and the value Φa is taken as the average Φa = 1 (Φi.j + Φi−1.e.j Φi+1.j+ − ∂Φ ∂x dx dy = ∆y + i− 21 .j Φi.j+1 − 2φi.j Φi+1. .j Φi−1.j ) ∆y · 1 ∆x · ∆y and so on for the rest of the terms. .j−1 + =0 2 ∆x ∆y 2 which of course also can be written in the matrix form shown for the general case above. of the form Ai.2. see figure 3. .

j−1 + ∆x2 ∆y 2 =0 In a similar manner the discretization for the v-component becomes 2 2 (uv)i+1.j+1 − 2ui.j+1 − pi.j pi.j+1 − (uv)i. We start with a co-located grid which means that the finite volume used for both velocity components and the pressure is the same.j + + + ∂t 2∆x 2∆y 2∆x − 1 Re ui+1. We have seen that the finite volume discretizations of both are equal to standard second order finite difference approximations. 3.j − pi−1.j−1 pi.j−1 + 2 ∆x ∆y 2 =0 This simple discretization cannot be used in practice without some kind of modification.j−1 + =0 2∆x 2∆y The discretization of the streamwise momentum equation needs both first and second derivatives.j+1 − vi.j+1 − vi.j − ui−1.j i+1 i Figure 3.j ) ∆y + (vi. j − 1 i + 1.3.j vi. Recall that the finite volume discretization of the continuity equation becomes 1 1 (ui+1.j + vi.j vi.j + + + ∂t 2∆x 2∆y 2∆y − 1 Re vi+1.j+1 − 2vi.j−1 ∂vi. j − 1 i. Consider the following solution to the discretized equations ui. which in two-dimensions can be written (u dy − v dx) = 0 ∂u dS = − ∂t u2 dy − uv dx + p dy − 1 Re ∂u ∂u dy − dx ∂x ∂y ∂v dS = − ∂t uv dy − v 2 dx − p dx − 1 Re ∂v ∂v dy − dx ∂x ∂y where we have used the earlier derived result n = ( dy.j − u2i−1.j vi. Thus we obtain the following result for the u-component (uv)i.j − 2ui.j+1 − vi. ∆x = ∆y .j−1 ) ∆x = 0 2 2 which is equivalent to the finite difference discretization or ui+1. p = (−1)i+j .j ∂ui.j pi+1.j−1 u2i+1. see figure 3. The reason is that one solution to these equations are in the form of spurious checkerboard modes.3: Cartesian finite-volume grid where the velocity components and the pressure are co-located. − dx).j ui. j − 1 j ui. In addition we choose a Cartesian grid in order to simplify the expressions.a a i − 1.j − ui−1.j + ui−1.j vi.2 Finite-volume discretizations of 2D NS Co-located Cartesian grid We will apply the finite volume discretizations to the Navier-Stokes equations in integral form.j − (uv)i−1.j = (−1)i+j · g (t) .j + vi−1.j + ui.j = vi.j − 2vi.

control volume for streamwise momentum equation (u) .j ✴✳✭✴✭ ✳ ✴✭ ✳ ✴✭ ✳ i.j p i.j−1 − =0 2 Re ∆x Re ∆y 2 .j+1/2 − (uv)i+1/2.j− 12 ∆y =0 The control volume for the streamwise velocity is centered around the streamwise velocity point and the discretization becomes (uv)i+1/2.j ✱ ✲✭ ✱ ✲✭ ✱ ✲✭ ✱ ✲✭ ✱ ✲✭ ✱ ✲✭ ✱ ✲✱ ✲✭ ✱ ✲✭ ✱ ✲✭ ✱ ✲✭ ✱ ✲✭ ✱ ✲✭ ✱ ✲✱ ✲✱✭✲✭ ✱ ✲✭ ✱ ✲✭ ✱ ✲✭ ✱ ✲✭ ✱ ✲✭ ✱ ✲✱ ✲✱✭✲✭ vi. j −a1j ii − + 1. 1.j−1/2 ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✬ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✬ ✮✭ ✶✁✵✶✁✵ ✶✵✶✵ ✷✸✷✸✹ ✷✸✷✸✹ ✷✸✷✸✹ ✷✷ ✹ ✹✹✹✹ ✻✺✁✻✺✁✻✺✻✺ i− 1 2 i+ i 1 2 i+1 i+ j+ 1 2 j j− 1 2 j −1 3 2 control volume for divergence condition.j+ 21 − vi.j+1/2 ui−1/2.j+1/2 ui−1/2.j− 12 ∆x = 0 or ui+ 12 .j+ 12 − vi. If it is input into the equation for u and v we find the expression 8 dg + g=0 dt Re · ∆x2 ⇒ 8t g = e− Re·∆x2 Thus we have a spurious solution consisting of checkerboard modes which are damped slowly for velocity and constant in time for the pressure.j vi.j−1/2 pi.j i+ 1 pvi.j+1/2 vi.j ∆x + vi.j vi.j i.j−1/2 i − 12 v i.j i. control volume for normal momentum equation (v) . j − 1ji +1 i + 1. j −1 ence condition. Figure 3. j i i+1 ui. Staggered cartesian grid To eliminate the problem with spurious checkerboard modes.j−1/2 u2i+1.i.j + ui−1/2.j✴✭ ✳ ✴✭ ✳ ✴✭ ✳ ✴✭ ✳ ✮✭ ✬ ✴✳ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✬ ✴✭ ✳ ✴✭ ✳ ✴✭ ✳ ✴✭ ✳ ✴✭ ✳ ✴✭ ✳ ✴✭ ✳ ✮✭ ✬ ✴✳ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✬ ✴✭ ✳ ✴✭ ✳ ✴✭ ✳ ✴✭ ✳ ✴✭ ✳ ✴✭ ✳ ✴✭ ✳ ✮✭ ✬ ✴✳ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✬ ✴✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✬ ✮✭ vi.j u pi.j − u2i. m equation (u) . which satisfies the divergence constraint.j v i.j pi+1.j + + + − ∂t ∆x ∆y ∆x 1 ui+3/2.j vi.j vi. The control volume for the continuity equation is centered abound the pressure point and the discretization becomes ui+ 21 .j+1/2 ✱ ✱ ✱ ✱ ✱ ✱✰✭ ✲✱✭ ✭ ✲ ✭ ✲ ✭ ✲ ✭ ✲ ✭ ✲ ✯ ✰✭ ✯ ✰✭ ✯ ✰✭ ✯ ✰✭ ✯ ✲✭ ✯ ✲✱✰✯ ✰✯✭✰✭ ✯ ✰✭ ✯ ✰✭ ✯ ✰✭ ✯ ✰✭ ✯ ✰✭ ✯ ✰✯ ✰✯✭✰✭ ✯ ✰✭ ✯ ✰✭ ✯ ✰✭ ✯ ✰✭ ✯ ✰✭ ✯ ✰✯ ✰✯✭✰✭ ✯ ✯ ✯ ✯ ✯ ✯ ✯✴✭ ✭ ✰ ✭ ✰ ✭ ✰ ✭ ✰ ✭ ✰ ✭ ✰ ✳ ✴✭ ✳p ✴✭ ✳ ✴✭ ✳ ✴✭ ✳ ✰✭ ✳ ✮✭ ✬ ✰✯✴✳ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✭ ✬ ✮✬ ui−1/2. j −aa1 ii − + 1. spanwise and pressure are different.j − ui− 12 .j − pi. ji − ui.j ∂ui+1/2. where the control volumes for the streamwise.j−1/2 1i ii − + 122 ui−1/2.j1 i u pi. jj − −1 1 i.jji i+ vi.j − 2ui+1/2. This mode will corrupt any true numerical solution and yields this discretization unusable as is.j+1 − 2ui+1/2. 1.4.j pi.j − ui− 21 . we can use a staggered grid as in figure 3. These modes are a result of the very weak coupling between nearby finite volumes or grid points.j i.j + 1121i iii − + i + 322 + 111i jii + + i + 3222 + 111j jji + i− + 3222 − 11j jjj + − 212 j − 1j j − 12 ence condition.j ∆y + vi.j 1 ui+1/2. jj − −1 1 i.j + ui+1/2.j vi.4: Cartesian finite-volume grid where the velocity components and the pressure uses staggered grids. and can be avoided if they are coupled by one sided differences or if they are damped with some type of artificial viscosity.j pi.

1 The boundary conditions on the inflow.1 + u 23 . 23 = 2v 21 .1 1 B u 23 . AB in figure 3. u = v = 0. BC in figure 3. 32 .j+1 − pi. See figure 3. requires u 23 . consists of the no flow and the no slip conditions.j+1/2 . (uv)i+1/2. 12 = 1 u 23 . 21 = · · · = 0 The evaluation of the u-equation at the point ( 23 . For this discretization no checkerboard modes possible. where the u2i+1.5: Staggered grid near the boundary.5. for example. (uv)i+1/2. .0 . 32 2 1.j − 12 j j−1 Inflow A 2 v1. since there is no decoupling in the divergence constraint.j+1/2 + vi.j+1 . The control volume for the normal velocity is centered around the normal velocity point and the discretization becomes 2 2 (uv)i+1/2. 32 − v1. 1). The evaluation of the v-equation at the point (1.1 . u2i. 21 = v2.0 2 ⇒ u 23 .1 ✽✚✼ ✽✚✼ ✽✚✼ ✽✚✼ ✽✚✼ ✽✚✼ ✽✚✼ ✽✚✼ ✽✚✼ ✽✚✼ ✽✚✼ ✽✚✼ ✽✚✼ ✽✚✼ ✽✚✼ ✽✚✼ ✽✚✼ ✽✚✼ ✽✚✼ ✽✚✼ ✽✼ C Solid wall 0 1 0 i 2 Figure 3. consist given values of u and v. We can find this value as follows v 21 . (uv)i−1/2.0 = −u 23 .j+1/2 − (uv)i−1/2. The normal points are located on the boundary so that the no flow condition simply become v1. We can find this value as follows 0 = u 23 . 32 = 1 v 3 + v0. the pressure or the convective terms. 32 p1. u2i.j+1/2 1 vi.e. for example.j−1/2 − =0 2 Re ∆x Re ∆y 2 where the (uv)i+1/2.j+1/2 vi. The boundary conditions on the solid wall.j .j+1/2 + vi−1. so called Dirichlet conditions.5 for a definition of the points near a boundary. The streamwise points are located on the boundary so that the boundary condition is simply consist of the known values u 21 .5.j+1/2 . 32 ).j terms need to be interpolated from the points where the corresponding velocities are defined.j−1/2 terms need to be interpolated from the points where the corresponding velocities are defined. i.j+1/2 .2 .j ∂vi.j+1 − vi. u 21 .j . u2i. i.j+1/2 − 2vi. Boundary conditions To close the system we need to augment the discretized equation with discrete versions of the boundary conditions.j+1/2 pi. . 32 . 2 ⇒ v0. requires v0.j + + + − ∂t ∆x ∆y ∆y 1 vi+1.e. .j+3/2 − 2vi.

e. so that in this formulation no boundary conditions for p is needed so far. v)nb unb nb The last line summarizes the expressions.j Re∆x2 4∆x i− 2 .j− 12 − R ∆y 2 2 2 + = {expand and interpolate using grid values} = b(u.j+1 + (v ui+ 12 .j+ 2 Re∆y 2 4∆y i+1.j+ 1 − (uv)i− 1 . v)i+ 21 .j   + Ai+ 21 .j+ 12 − 2vi.j 4∆x i+ 2 .j+ 2 Re ∆x2 ∆y 2 1 1 1 1 + u 3 − ui+ 32 .2 .j− 1 u2i+1.j 4∆y i+1.j+1 − 2ui+ 21 .j ∆y + vi−1. 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.∂v If AB in figure 3.j + ui− 12 .j ) + (v + ui+ 12 .j i.j ui+ 12 .j+ 1 2 2 ∆x 1 vi+1. However.j+ 32 − 2vi. 23 Note that no reference is made to the pressure points outside the domain.j−1 1 ui+ 32 .j− 2 Re∆y 2 = = a(u.j = (uv)i+ 1 . If this is the case particular care has to be taken so that those conditions do not upset the divergence free condition.j+ 21 + =0    ∂t ∆y   D =0 i.j+ 12 + b(u.j where the expression for the Ai+ 12 .5 is an outflow boundary.j+1 − vi.j + =0   ∂t ∆x ∂vi.j+ 12 1 vi.j+ 12 ) − i. so called Neumann conditions.j − pi.j − 2ui+ 12 .j + u 1 − ui− 21 .j+ 12 − vi+1. For the streamwise and normal velocity we find u− 12 .j Re∆x2 1 1 1 1 1 + v 1 + v + (v ui+ 12 .j+ 12 pi.j − u2i. v)nb vnb nb .j− 12 ) + 2∆x i+ 2 . 3. 32 = v1.j− 12 − vi.j+ 21 = = (uv)i+ 1 .j−1 i. The expression for Bi.j+ 12 can in a similar way be written Bi.j− 12 ) − 4∆y i+1.j + Bi.3 Summary of the equations The discretized equations derived for the staggered grid can be written in the following form  ∂ui+ 12 .j+ 12 vi.j 2 2 2 2 + ∆x ∆y 1 ui+ 12 . v)i.j+ 1 − (uv)i+ 1 .j+ 12 + vi. These conditions should be evaluated at i = 12 . v0.j + ui+ 12 . where the sum over nb indicates a sum over the nearby nodes.j+1 − pi.j − − Re ∆x2 Re ∆y 2 = = {expand and interpolate using grid values} 1 1 2 1 1 1 − v (u 3 − ui− 21 .j + a(u.j can be written Ai+ 21 .j  pi+1. i.2 = u 23 .j+ 12 − R ∆x2 2 2 vi. the boundary conditions may consist of ∂u ∂x = ∂x = 0.

Sometimes this method and its generalizations are also called splitting methods.j+ 21 − vi. v) Gy   v  =  fv  dt 0 Dx Dy 0 p 0 Here we we have used the definitions u v p A(u.j− 12 ∆y These equations.j = ui+ 12 .4 Time dependent flows The projection method For time-dependent flow a common way to solve the discrete equations is the projection or pressure correction method. We illustrate the method using the system written in the compact notation defined above. v) B(u. including the boundary conditions. can be assembled into matrix form. We start to discretize the time derivative with a first order explicit method  n+1 u − un   + N (un ) un + G pn+1 = f (tn ) ∆t   Dun+1 = 0 Then we make a prediction of the velocity field at the next time step u∗ not satisfying continuity. i. 3.j and a(u.j ∆x + vi.e. it is seen that the form of the discretized matrix equations are the same. apply the divergence operator D to the correction step 1 D u∗ ∆t Here DG represents divgrad = Laplacian. v) 0 Gx u fu d  v + 0 B(u. We have        u A(u. The velocity at the next time level thus becomes DG pn+1 = . The expression originating from the continuity equation is Di. discrete version of pressure Poisson equation. 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 have  ∗ u − un   + N (un ) un = f (tn )   ∆t    un+1 − u∗ + G pn+1 = 0    ∆t     Dun+1 = 0 Next. v)nb .j+ 12 and b(u. non-linear algebraic operator from advective and viscous terms of v-eq. 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 finite element discretization in the next chapter. v)i+ 12 . v)i. and therefore not explicitly written out.where the expressions for b(u. v) Gx Gy Dx Dy f - 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. v)nb are analogous to a(u.j − ui− 12 . and then a correction involving p such that Dun+1 = 0.

We find n+1 n+1 n+1 n+1 pn+1 pn+1 1 3. To see this we write the pressure Poisson equation.j+ 12 − vi.j is Di. we need additional boundary conditions. however.j   2  i+ 2 .j + pi.1 − p2. and the corresponding expressions for u i+ 1 .1 − p2.j + = ∆x2 ∆y 2 ∆t ∗ where Di. centered abound the point (2.j − 2pi. for this particular discretization it turns out that we can choose vΓ∗ arbitrarily since it completely decouples from the rest of the discrete problem.0 from the boundary condition obtained for the pressure into the ∗ above equation we can se that vΓ cancels in the right hand side of the pressure Poisson equation.1 + p1.j  2 2  + =0 ∆t ∆x n+1 ∗ n+1 n+1  pi.1).un+1 = u∗ − ∆tG pn+1 Note here that un+1 is projection of u∗ on divergence free space.j + pi−1. In particular. If we take the vector equation for the correction or projection step and project it normal to the solid boundary in figure 3. However.j + Ani+ 1 . 1 . we can choose n+1 vΓ∗ = v2. . Thus we find that we can choose the value of vΓ∗ arbitrarily.j+1 − pi.5 we have n+1 pn+1 1 2. 1 is an unknown value of the velocity in the prediction step at the boundary.1 2 2 ∆x + ∗ ∗ v2. close to the boundary.j+ 1 − v  i.0 ∗ =− v n+1 1 − vΓ ∆y ∆t 2.j  vi. 3 − vΓ 2 ∆y n+1 if we substitute the value of pn+1 2.j i. 2 ∗ where vΓ∗ = v2.j+ 12   + =0 ∆t ∆y n+1 n+1 and vi.1 2.e.1 − 2p2.1 − u∗3 .1 + p2.j−1 i+1. A number of other splitting methods have been devised for schemes with higher order accuracy for the time discretization.j − pi. When the discrete pressure Poisson equation is solved we need values of the unknowns which are outside of the domain. that 2 this is a numerical artifact and does not imply that the real pressure gradient is zero at the boundary.0 + = 2 2 ∆x ∆y ∆t u∗5 . In order to be able to discuss the boundary condition on the pressure Poisson equation we write the prediction and correction step more explicitly.j = 0 2 ∆t n ∗ vi.j − u∗i+ 1 .j 2 2 2 n+1 vi. u∗ and v ∗ .j+1 − 2pi. 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 flow. see the discussion in the end of the section on the role of the pressure. 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. Using the staggered grid discretization the prediction step can be written  ∗ u 1 − uni+ 1 . See Peyret and Taylor for details.j− 1 .j+ 1 = 0 2 ∆t and the projection step as  n+1 n+1 ui+ 1 . In general it is preferred to make the time discretization first for the full Navier-Stokes equations and afterward split the discrete equation into one predictor and one corrector step.j  pn+1  i+1. in the expression for the divergence constraint D i.2 − 2p2.j i− 1 . Note.j we find the following explicit expression for the 2 discrete pressure Poisson equation n+1 n+1 n+1 n+1 ∗ pn+1 pn+1 Di.j evaluated using the discrete velocities at the prediction step.j+ and By solving for un+1 1 above and using this. in which case we obtain a Neumann boundary condition for the pressure. i. In general 2 it is important to choose this boundary condition such that discrete velocity after the correction step is divergence at the boundary.j+ 21  n 2  + Bi.

Time step restriction For stability it is necessary that prediction step is stable. We introduce unj = Qn · ei·ωxj into the discrete equation. This implies that Φ (s) = − λ2 − α2 s + λ2 − α ≤ 0 ∀s : 0 ≤ s ≤ 1 Φ (s) is linear function of s and thus the the limiting condition on Φ (s) is given by its values at the two end points of the s-interval. numerical evidence indicates that this is sufficient. α= . V0 . . We find Φ (0) = λ2 − α ≤ 0. Φ (1) = α2 − α ≤ 0 which implies that ⇒ λ2 ≤ α ≤ 1 substituting back the definitions of λ and α we find the following stability conditions    1 U02 ∆tRe ≤ 1 2 2∆t   ≤ 1 Re∆x2 For the two dimensional version of the predictor step Peyret and Taylor [8] find the following stability limits    1 U02 + V02 ∆t · Re ≤ 1 4 4∆t   ≤ 1 Re∆x2 Note that this is a stronger condition than in the one-dimensional case. together with the definitions U0 ∆t 2∆t . ξ = ω · ∆x ∆x Re ∆x2 This implies that the amplification factor Q can be written λ= Q = 1 − iλ sin (ξ) − 2α sin2 ξ 2 We find that the absolute value of Q satisfies |Q| = 1 − 4 λ2 − α2 s2 + 4 λ2 − s2 s = 1 − 4s where s = sin2 ξ 2 λ2 − α 2 s − λ 2 + α ≤ 1 . It thus suffices to consider the equation ∂u ∂u ∂u 1 2 + U0 + V0 = ∇ u ∂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 = unj + ∆t −U0 j unj+1 − unj−1 1 unj+1 − 2unj + unj−1 + 2∆x Re ∆x2 where we have chosen to call the discrete points where the u-velocity is evaluate for x j . In the predictor case the momentum equations decouple if we linearize them abound the solution U 0 .

Note that R is a discrete Laplace like operator.3. . For this method we have Ax = b A=L−U (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 = N (u) D 0 −DN (u)−1 G if the matrix B has the form B= I 0 −N (u)−1 G I I 0 ≈ ˜ (u)−1 G −N I ˜ −1 is a simple approximation of the inverse of the discrete Navier-Stokes operator. If B = I we obtain a regular iteration method where residual r k is used to update y k by inverting M . We divide the matrix AB in the following manner AB = M − T ⇒ M yˆ = T yˆ + b and use this to define 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 simplified to y k+1 = y k + BM −1 b − A y k rk where rk is the residual error in the k:th step of the iteration procedure. Now.5 General iteration methods for steady flows Distributive iteration For steady flows we can still use the methods introduced for unsteady flows. We now where N(u) 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. However. a simplified part of A. and iterate the solution in time to obtain a steady state. in many cases it is more effective to use other iteration methods. 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. We will here introduce a general distributive iteration method and apply it to the discretized steady Navier-Stokes equations. as in the Gauss-Seidell method for example. let y = B yˆ which implies that AB yˆ = b and use the iteration scheme on this modified equation.

The SIMPLE (Semi-Implicit Method for Pressure Linked Equations) by Pantankar and Spalding (1972) can be thought of as a distributive iteration method. . where the velocities and the pressure is updated uk+1 = uk + ωu δu pk+1 = pk + ωp δp The under relaxation is needed for the method to converge. Second. Fourth. Thus we have the following iteration steps 1. we have the distribution step BM −1 rk δu δp =B δ˜ u δ p˜ which becomes = ˜ −1 G δ p˜ δ˜ u−N δ p˜ 4. First. Wesseling [10] indicates that by choosing ˜ (u)−1 N = [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. we calculate the residual b − A yk = f 0 N (uk ) G D 0 − uk pk = r1k r2k 2. thus ωu and ωp is usually substantially lower that one. Third. we calculate M −1 rk which implies Q δ˜ u = r1k δ˜ u = Q−1 r1k ⇒ R δ p˜ = r2k − D δ˜ u 3. there is an under relaxation step.

different integral i.j We consider the steady flow of an incompressible fluid in a bounded domain Ω. 71 .j two portions Γ0 and Γ1 . whereas the Neumann condition ∂n =0 j − 1 on Γ1 is an adiabatic condition. The Dirichlet j− ∂u boundary condition u = 0 on Γ0 is called 2an isothermal condition.j vi.dx dy − dx dl = ( dx.j in Ω.1a) Uj 1 − ν∇2 u = f i −∂x 2 j Chapter 4 Finite element methods for incompressible flow u=0 on Γ0 .2) ∂xi ∂xi ∂xi ∂xi p1. so that nj Uj = 0 on Γ11j.1 ∂u ∂v ∂u v1.j+1/2 interested to compute the temperature field u in the domain. j Aabcd d d a The finite-element method (FEM) approximates an integral form of the governing equations. a the integral form—called the variational form—is usually different from the one used for the finite-volume i − 1.1b) i i + 12 ∂u =0 on Γ1 . − dx) i − 1. This simplifies the analysis but is not essential. j i + 1. Incompressibility yields that ∂xjj = 0. Moreover. and we assume that its pi. We are Γ of the domain can be decomposed into vi.1 for instance. A to form the finite-element discretization requires some vector Deriving the integral form that is used B yields calculus.j−1/2 ∂u ui−1/2.forms j − 1 may be used for the same equation depending on circumstances method. However. as in figure 4. (4. dy) n dl = ( dy. (4. j + 1 c c b b i − 1. We introduce FEM first for a scalar advection–diffusion problem before i + 1. We assume that Γ1 is a solid wall.− 1 j i i+1 4. (4.1 0 n 1 Γ0 2 i j Ω Inflow Solid wall Γ1 Figure 4. j + 1 i. The product rule of differentiation C u3 ∂ 2 . j + 1 i + 1. 32 v = + v∇2 u. j i. (4.1c) i+1 ∂n 3 i+ 2 ∂U + 12 The coefficient ν > 0 is the thermal jdiffusivity. for instance). j−1 such as boundary conditions.1). We assume that the boundary velocity field Uj is known (from a previous pi.1 FEM for an advection–diffusion problem ui. j discussing the Navier–Stokes equations. modeled by the advection–diffusion problem vi.1: A typical domain associated with the advection–diffusion problem (4.j numerical solution.

we need to introduce the function space     ∂v ∂v V = v: dΩ < +∞. The Dirichlet boundary condition on Γ0 is explicitly included in the definition of V and is therefore referred to as an essential boundary condition. The derivation leading up to expression (4. Let Vh be the space of all functions that are . the nodes on Γ0 are excluded in the count. as in figure 4. For this.5).) ∂n ∂xi Let v be any smooth function such that v = 0 on Γ0 .6) is called a weak solution to advection–diffusion problem (4. Multiplying equation (4. 4. and on Γ1 due to boundary condition (4.1.4) yields f v dΩ = Ω Ω ∂u vUj dΩ − ν ∂xj 2 v∇ u dΩ = Ω Ω ∂u vUj dΩ + ν ∂xj Ω ∂v ∂u dΩ − ν ∂xj ∂xj Γ   0 ✒   ∂u   v dΓ. that is. a weak solution may not be a classical solution since functions in V may fail to be twice continuously differentiable.Recall the divergence theorem ∂wi dΩ = ∂xi Ω ni wi dΓ.1).1 Finite element approximation Let us triangulate the domain Ω by dividing it up into triangles.5) Ω for each smooth function v vanishing on Γ0 We will now “forget” about the differential equation (4. (4. Let M be the number of triangles and N the number of nodes in Ω ∪ Γ1 . The Neumann condition on Γ1 is defined implicitly through the variational form and is therefore called a natural boundary condition.2. β ∈ R. .  ∂n where the last term vanishes on Γ0 due to the requirements on v.continuous on Ω. and directly consider the variational form (4. (4. v = 0 on Γ0 . u.2) and applying the divergence theorem on the left side yields v ∂u dΓ = ∂n Γ Ω ∂v ∂u dΩ + ∂xi ∂xi v∇2 u dΩ. and utilizing expression (4.1) (also called a classical solution) satisfies the varational form vUj Ω ∂u dΩ + ν ∂xj Ω ∂v ∂u dΩ = ∂xj ∂xj f v dΩ (4. that is.6) Ω A solution to variational problem (4. The variational problem will thus be Find u ∈ V such that ∂u ∂v ∂u vUj dΩ + ν dΩ = ∂xj ∂xj ∂xj Ω Ω f v dΩ ∀v ∈ V.3) Γ Integrating expression (4. (4.4) Ω (Note that ∂u ∂u = ni .5) shows that a classical solution is a weak solution.1). v ∈ V ⇒ αu + βv ∈ V ∀α.1a) with v.1c) We have thus shown that a smooth (twice continuously differentiable) solution to problem (4. However. Let h be the largest side of any triangle. The requirement that the gradient-square of each function in V is bounded should be intepreted as a requirement that the energy is bounded.   ∂xi ∂xi Ω The space V is a linear space. integrating over Ω. Note that the boundary conditions are incorporated into the variational problem.

N u(n) Φ(n) (xi ) .1. Assembly. N . The finite-element method is a Galerkin approximation in which the subspace is given by piecewise polynomials.7) yields N N u(n) n=1 v h Uj Ω ∂Φ(n) dΩ + ν u(n) ∂xj n=1 Ω ∂vh ∂Φ(n) dΩ = ∂xj ∂xj f vh dΩ ∀ v h ∈ Vh . ∂xj (4. . . Note that Vh ⊂ V . .2 The algebraic problem. m = 1. . where A = C + νK and Kmn = Ω ∂Φ(m) ∂Φ(n) dΩ. Substituting expansion (4. Note that expansion (4.6). . .linear on each triangle. Here.2: A triangulation. 4. Once the value of a function uh in Vh is known at all node points. Thus.3). which then becomes N N u(n) n=1 Φ(m) Uj Ω ∂Φ(n) dΩ + ν u(n) ∂xj n=1 Ω ∂Φ(m) ∂Φ(n) dΩ = ∂xj ∂xj f Φ(m) dΩ. the black nodes.11) . uh (xi ) = (4. f vh dΩ Ω Restricting the function space in a variational form to a subspace is called a Galerkin approximation.1 v1.8) forces uh = 0 on Γ0 since it excludes the nodes on Γ0 in the sum. . that is: Find uh ∈ Vh such that ∂vh ∂uh ∂uh v h Uj dΩ + ν dΩ = ∂xj ∂xj ∂xj Ω Ω (4. ∂xj ∂xj Φ(m) Uj Cmn = Ω ∂Φ(n) dΩ.8) n=1 where u(n) denotes the value of uh at node n.1 0 1 2 i j Inflow Solid wall n Γ0 Γ1 Ω u 23 . which are functions in Vh that are zero at each node point except node n.p1. (4.vanishing on Γ0 . This interpolation can be written as a sum of the “tent” basis functions Φ (n) (figure 4. where it is unity. we may choose vh = Φ(m) ∈ Vh for m = 1. .7) ∀vh ∈ Vh . the number N are Figure 4. N in equation (4.3: The “tent or “hat basis function associated with continuous.9) Ω Since each basis function is in Vh .1 0 1 2 i j Inflow Solid wall n Γ0 Γ1 Ω Γ0 Γ1 Figure 4. 23 p1. We get a finite-element approximation of our advection-diffusion problem by simply replacing V by Vh in variational problem (4. . it is easy to reconstruct it by linear interpolation. piece-wise linear functions.8) into FEM approximation (4. .10) Ω This is a linear system system Au = b. (4.9).

12).  u(n)   b=  f Φ(1) dΩ Ω . except if Uj and f happen to be particularly “easy” functions (such as constants). we need expressions for the gradients of basis functions Φ (m) and (m+1) Φ .5). For instance.. Km. 4. (p) so Kmn is easy to compute exactly. we will compute the elements of K.m+2 = 0. To compute matrix element (4. calculating the element contribution Kmn for m. f Φ(N ) dΩ Ω      The union of all triangles constitutes the domain Ω and that the triangles do not overlap. Note that the derivatives of Φ(n) are constant on each triangle.. M Kmn = Ω ∂Φ(m) ∂Φ(n) dΩ = ∂xj ∂xj p=1 M Tp ∂Φ(m) ∂Φ(n) (p) dΩ = Kmn . Matrix K can thus be computed by looping (p) all M triangles. but numerical quadrature is usually needed. Kmn = 0 only when m and n are associated with nodes that are nearest neighbors. Thus.. and the area of each triangle is |Tk | = h2 /2. In fact. any integral over Ω can be written as a sum of integrals over each triangle. the “stiffness matrix” (a term borrowed from structural mechanics).4). the gradient is constant on each triangles and may thus easily be computed by finite-differences in the coordinate direction (figure 4. Matrix C and vector b can also be computed by element assembly. Thus Kmn contributes to the sum only when m and n both are corner nodes of triangle p.4. Let M be the number of internal nodes in the horizontal as well as the vertical direction (the solid nodes in figure 4.12) Note that Kmn vanishes for most m and n. . Since the functions are piecewise linear.4: An example mesh that gives a particular simple stiffness matrix K    u(1)   u =  .) The width of each “panel” is then h = 1/(M + 1). ∂xj ∂xj p=1 (p) (p) Note that Kmn vanishes as soon as m or n is not a corner node of triangle p.1. By the expressions in . The components of K are Kmn = Ω ∂Φ(m) ∂Φ(n) dΩ ∂xj ∂xj (4.3 An example As an example. n being the three nodes of triangle p and adding these 9 contribution to matrix K. since basis functions Φ(m) and Φ(m+2) do not overlap (figure 4.i j Inflow Solid wall n Γ0 Γ1 Ω M2 ❀✿✾❀✿✾ ❀✿✾❀✿✾ ❀✿✾❀✿✾ ❀✿✾❀✿✾ ❀✿✾❀✿✾ ❀✿✾❀✿✾ ❀✿✾❀✿✾ ❀✾❀✾ ❂✿❁❂✿❁ ❂✿❁❂✿❁ ❂✿❁❂✿❁ ❂✿❁❂✿❁ ❂✿❁❂✿❁ ❂✿❁❂✿❁ ❂✿❁❂✿❁ ❂❁❂❁ ❀✾✿❀✾✿❀✿✾❀✿✾ ❀✿✾❀✿✾ ❀✿✾❀✿✾ ❀✿✾❀✿✾ ❀✿✾❀✿✾ ❀✿✾❀✿✾ ❀✾❀✾ ❂✿❁❂✿❁ ❂✿❁❂✿❁ ❂✿❁❂✿❁ ❂✿❁❂✿❁ ❂✿❁❂✿❁ ❂✿❁❂✿❁ ❂✿❁❂✿❁ ❂❁❂❁ ❀✾✿❀✿✾ ❀✿✾❀✿✾ ❀✿✾❀✿✾ ❀✿✾❀✿✾ ❀✿✾❀✿✾ ❀✿✾❀✿✾ ❀✿✾❀✿✾ ❀✾❀✾ ❂✿❁❂✿❁ ❂✿❁❂✿❁ ❂✿❁❂✿❁ ❂✿❁❂✿❁ ❂✿❁❂✿❁ ❂✿❁❂✿❁ ❂✿❁❂✿❁ ❂❁❂❁ ❀✾✿❀✾✿❀✿✾❀✿✾ ❀✿✾❀✿✾ ❀✿✾❀✿✾ ❀✿✾❀✿✾m❀✿✾❀✿✾ ❀✿✾❀✿✾ ❀✾❀✾ ❂✿❁❂✿❁ ❂✿❁❂✿❁ m❂✿❁❂✿❁ ❂✿❁❂✿❁ +❂✿❁❂✿❁ 2❂✿❁❂✿❁ ❂✿❁❂✿❁ ❂❁❂❁ 1 M h h Figure 4. for the particular case of the structured mesh depicted in figure 4.4.

0 on T2 .6 yields 2 Km.   −I T −I −I T where I is the M -byM identity matrix.. . K becomes the block-tridiagonal M 2 -by-M 2 matrix  T −I  −I T −I   . Figure 4.   −1 4 −1 −1 4 Thus. . .m−M = −1. 0    − h1 .       . n T1 T5 on T5 .6: The support for basis functions Φm and Φm+1 are marked with vertical and horizontal stripes respectively. K = . and T the M -byM matrix  4 −1  −1 4 −1   . . it is only on these two triangles that both functions are nonzero at the same time. 2 h h  1  h. Utilizing the gradient expressions given in figure 4. that is.m 2 m+2 T3 i on T1 . 0 on T .m−1 = Km. 2 M (m) M being the (black) middle node. Altogether. −h   1    0.. h T4 Inflow T2 on T3 . Γ0 Γ1 on T6 . T6 Ω on all other triangles. . we have Kmm = Ω ∂Φ(m) ∂Φ(m) dΩ = ∂xj ∂xj 6 k=1T k ∂Φ(m) ∂x 2 ∂Φ(m) ∂y + 2 dΩ 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 Φm and Φm+1 is only the two triangles marked in figure 4.6. 1  h h h ∇Φ(m) = − h1 .5.13) .5: The gradient of test function Φ for m m m + 2  1 1 Φ(m) 1  − .m+M = Km. h1    0 ❆❇ ❅❄❅❄❇✭ ❆❇ ❅❄❅❄❇✭ ❆❇ ❅❄❅❄❇✭ ❆❇ ❅❄❅❄❇✭ ❆❇ ❅❄❅ ❇✭ ❆❇ ❇✭ ❆❆✭ ❆❆✭ ❅❄❅❄❅❄❅❄❅❄❅❄❇✭ ❇✭ ❇❆❆❇ ❇❇✭ ❇❇✭ ❆❇✭ ❆❇✭ ❆❇✭ ❆❇✭ ❆❇✭ ❆❇✭ ❆ ❆ ❇✭ ❇✭ ❇✭ ❇✭ ❇✭ ❇✭ ❇❆❇❆ ❃❄❃❄❅❄❅❄❃❄❅❄❅❄❃❄❅❄❅❄❃❄❅❄❅❄❃❄❅❄❅❄❃❄❅❄❅❄❆✭ ❄❆ ❅❄❅❄❃❄❅❄❅❄❆✭ ❃❆❇✭ ❄❆ ❅❄❅❄❃❄❅❄❅❄❆✭ ❃❆❇✭ ❄❆ ❅❄❅❄❃❄❅❄❅❄❆✭ ❃❆❇✭ ❄❆ ❅❄❅❄❃❄❅❄❅❄❆✭ ❃❆❇✭ ❄❆ ❅❄❅❄❃ ❅❅ ❆✭ ❃❆❇✭ ❆❇✭ ❆❇✭ ❇✭ ❇✭ ❆❇✭ ❆❇✭ ❆❇✭ ❇❆❇❆ ❃❄❃❄❅❄❃❄❅❄❃❄❅❄❃❄❅❄❃❄❅❄❃❄❅❄❇✭ ❄❆ ❅❄❃❄❅❄❇✭ ❃❇✭ ❄❆ ❅❄❃❄❅❄❇✭ ❃❇✭ ❄❆ ❅❄❃❄❅❄❇✭ ❃❇✭ ❄❆ ❅❄❃❄❅❄❇✭ ❃❇✭ ❄❆ ❅❄❃ ❅ ❆❇✭ ❃❇✭ ❆ ❆ ❆ ❆ ❆ ❆ ❆ ❆ ❆❇✭ ❆❇✭ ❆❇✭ ❇✭ ❇✭ ❇❆❇❆ ❃❄❃❄❅❄❃❄❅❄❃❄❅❄❃❄❅❄❃❄❅❄❃❄❅❄❇✭ ❄❇ ❅❄❃❄❅❄❇✭ ❃❆✭ ❄❇ ❅❄❃❄❅❄❇✭ ❃❆✭ ❄❇ ❅❄❃❄❅❄❇✭ ❃❆✭ ❄❇ ❅❄❃❄❅❄❇✭ ❃❆✭ ❄❇ ❅❄❃ ❅ ❇✭ ❃❆✭ ❆ ❆ ❆ ❆ ❆ ❆ ❆ ❆ ✭ ❆ ✭ ❆ ❆❇❇❆ ❇❇✭ ❇ ❇ ❃❄❃❄❅❄❃❄❅❄❃❄❅❄❃❄❅❄❃❄❅❄❃❄❅❄❇✭ ❄❆ ❅❄❃❄❅❄❇✭ ❃❆❇✭ ❄❆ ❅❄❃❄❅❄❇✭ ❃❆❇✭ ❄❆ ❅❄❃❄❅❄❇✭ ❃❆❇✭ ❄❆ ❅❄❃❄❅❄❇✭ ❃❆❇✭ ❄❆ ❅❄❃ ❅ ❆✭ ❃❆❇✭ ❆❇✭ ❆❇✭ ❇✭ ❇✭ ❆❇✭ ❆ ❆ ❇❆❇❆ ❃❄❃❄❅❄❃❄❅❄❃❄❅❄❃❄❅❄❃❄❅❄❃❄❅❄❇✭ ❄❆ ❅❄❃❄❅❄❇✭ ❃❆❇✭ ❄❆ ❅❄❃❄❅❄❇✭ ❃❆❇✭ ❄❆ ❅❄❃❄❅❄❇✭ ❃❆❇✭ ❄❆ ❅❄❃❄❅❄❇✭ ❃❆❇✭ ❄❆ ❅❄❃ ❅ ❆❇✭ ❃❆❇✭ ❆❇✭ ❆❇✭ ❇✭ ❇✭ ❆❇✭ ❆❇✭ ❆❇✭ ❇❆❇❆ ❃❄❃❄❅❄❃❄❅❄❃❄❅❄❃❄❅❄❃❄❅❄❃❄❅❄❇✭ ❄❆ ❅❄❃❄❅❄❇✭ ❃❇✭ ❄❆ ❅❄❃❄❅❄❇✭ ❃❇✭ ❄❆ ❅❄❃❄❅❄❇✭ ❃❇✭ ❄❆ ❅❄❃❄❅❄❇✭ ❃❇✭ ❄❆ ❅❄❃ ❅ ❆❇✭ ❃❇✭ ❆ ❆ ❆ ❆ ❆ ❆ ❆ ❆ ❆❆✭ ❆ ❆ ✭ ❇ ✭ ❇ ❇❆ ❃❄❃❄❃❄❃❄❃❄❃❄❃❄❆✭ ❃❇ ❄❃❄❆✭ ❃❇ ❄❃❄❆✭ ❃❇ ❄❃❄❆✭ ❃❇ ❄❃❄❆✭ ❃❇ ❄❃ ❇✭ ✭ ❆ ✭ ❆ ❃❄❃❄❃❄❃❄❃❄❃❄❃❄❃❄❃❄❃❄❃❄❃❄❃❄❃❄❃❄❃❄❃ ❇ ❇ ❇ ❆❇ Figure 4. Solid wall ∇Φ(m) on T4 . − on T .. h1      0. T = . . T2 Φ(m+1)   0 on all other triangles T1 1 . Km. . 0    1 1   h. a typical row in the matrix–vector product Ku becomes     . .. (4. on T . 1 h ∇Φ(m+1) = 1 1 .   4um − um+1 − um−1 − um+M − um−M .. − h = − h1 . 1 j on T2 .m+1 = Ω ∇Φ(m) · ∇Φ(m+1) dΩ = k=1T ∇Φ(m) · ∇Φ(m+1) dΩ = − 1 1 2 h2 |T1 | − 2 |T2 | = − 2 = −1.. 2 h h h 2 k and in the same manner. figure 4.

1. to store only the nonzero elements and pointers to matrix locations. 4. A common technique in finite-element codes is therefore to use sparse representation.14) . Integration by parts also shows that matric C is skew symmetric (C T = −C). vh ≡ 0. v (N ) )T . . we recognize the classical five-point stencil for the negative Laplacian −∇2 . To see this. ∂Φ(n) ∂Φ(m) dΩ v (m) ∂xi ∂xi N = ∂ v (m) Φ(m) Ω m=1 ∂xi with equality if and only if ∂vh ∂xi N ∂ v (m) Φ(n) dΩ = ∂x i n=1 Ω ∂vh ∂vh dΩ ≥ 0. We start by reviewing some mathematical concepts that will be used in the analysis. the matrices become sparser and sparser as the mesh is refined.13) with h2 . for a general unstructured mesh.1. . However. However. Recall that a linear system Au = b has a unique solution for each b if the matrix is nonsingular. Basic definitions and inequalities The most basic integral norm is 1/2 v L2 (Ω) v 2 dΩ = Ω For the current problem. Let vh ∈ Vh . v T Kv > 0 for each v = 0. that is. .that is. ∂xi ∂xi = 0. K is a symmetric matrix (K T = K). We thus conclude that the linear system resulting from the finite-element discretization (4. if vh = Const. Thus.7) has a unique solution. At row m. that is. K is positive definite. 4. Proof. a more fundamental quantity is the energy norm 1/2 v V = Ω |∇v|2 dΩ Ω ∂v ∂v dΩ ∂xi ∂xi Ω ∂u ∂v dΩ. v)V = Ω ∇u · ∇v dΩ = (4. that is. that is. matrix element A mn = 0 only in the columns where n represents a nearest neighbor to node m. From Theorem 1 also follows that A is positive definite. Denote v = (v (1) . Then vh = N N n=1 N v T Kv = v (n) Ω m=1 n=1 v (N ) Φ(n) . A is not symmetric whenever U j = 0.11) that Kmn = Knm .4 Matrix properties and solvability Matrix A is sparse: the elements of martix A are mostly zeros. Theorem 1. Thus our finite-element discretization is for this particular mesh equivalent to a finite-volume or finitedifference discretization. the finite-element discretization will not give rise to any obvious finite-difference stencil. Thus v T Av = v T (K + C) v = v T Kv + v T Cv = v T Kv > 0 for each v = 0. v T Cv = 0 for each v. FEM usually uses integral norms for analysis of stability and accuracy. But since vh |Γ0 = 0. that is. A positive-definite matrix is a particular example of a nonsingular matrix. The number of nonzero elements on each row does not grow (on average) when the mesh is refined (since the number of nearest neighbors in a mesh does not grow). . 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. ∂xi ∂xi = 1/2 and associated inner product (u. Now the question is whether the approximation is any good. We will study the stability and accuracy of the numerical solution. It is immediate from expression (4.5 Stability and accuracy We have defined the finite-element approximation and showed that the finite-element approximation yields a linear system Au = b that has a unique solution. after dividing expression (4. A is positive definite.

the inequality Ω holds also when Γ0 v 2 dΩ ≤ C v 2 dΩ + Ω Ω ∂v ∂v dΩ ∂xi ∂xi dΓ = 0.15) Ω which can be denoted (f. v 2 L2 (Ω) ≤C v 2 V for each v ∈ V . . There is a C > 0 such that v 2 dΩ ≤ C Ω Ω ∂v ∂v dΩ. By inserting derivatives in the nominator and/or denominator of the variational characterization. A norm. (4. 1/2 f 2 dΩ Ω ≥ Ω Ω f g dΩ .15) implies that. Also. 1/2 g 2 dΩ Choosing g = f yields equality in above expression. an ) is |a · b| ≤ |a||b| = |a · a|1/2 |b · b|1/2 . However. it expresses the norm in terms of something that is optimized. Stability Choosing vh = uh in equation (4.The Cauchy–Schwarz inequality for vectors a = (a1 . various exotic norms can be defined. ∂xi ∂xi that is. (4. The Poincar´e inequality holds only if Ω is bounded in at least one direction. The Cauchy–Schwarz inequality (4.7) yields uh U j Ω ∂uh dΩ + ν ∂xj Note that uh U j Ω Ω ∂uh ∂uh dΩ = ∂xj ∂xj f uh dΩ.16) 2 Ω |∇wi | dΩ This norm will be of crucial importance when we analyze the stability of FEM form the Navier–Stokes equations. ∂xj (4.17) . different from the L 2 (Ω) norm and of importance for the Navier–Stokes equations is p p exotic Ω = max ∂wi ∂xi dΩ 1/2 wi . whereas for functions it reads 1/2 Ω 1/2 f 2 dΩ f g dΩ ≤ g 2 dΩ Ω . Remark 1. The smallest possible value of C grows with the size of Ω. . Theorem 2 says that the energy norm is stronger than then L2 (Ω) norm. the inequality does not hold if Γ0 dΓ = 0. square-integrable g. Thus 1/2 f L2 (Ω) f 2 dΩ = g=0 Ω f g dΩ Ω = max Ω g 2 dΩ 1/2 = max g=0 Ω f g dΩ g L2 (Ω) The rightmost expression is a “variational characterization” of the L2 (Ω) norm. . g)L2 (Ω) ≤ f g L2 (Ω) L2 (Ω) . for each nonzero. The Poincar´e inequality relates the energy and L2 (Ω) norms: Theorem 2 (Poincar´ e inequality). that is. . Ω ∂uh dΩ = uT Cu = 0.

as the next example illustrates. shows that C f L2(Ω) ν where the constant C does not depend on h. we have shown that uh cannot blow up as the discretization is refined.18) means that u h is an orthogonal projection of u on Vh . the error u − uh is orthogonal to each vh ∈ Vh with respect to the inner product (4. As illustrated in figure 4. .20) Matrix A is nonsymmetric in this case.7) implies that the FE approximation is optimal in the energy norm when Ui ≡ 0: u − uh V ≤ u − v h V ∀vh ∈ Vh . (4.18) that is. Cf. u − uh V ≤ C u − vh ν V ∀vh ∈ Vh . since C = −C T (using the notation of § 4. However. Note that the vector u − uh is the shortest of all vectors u − vh for vh ∈ Vh .1. (4.4).7: An orthogonal projection on a line. Orthogonality property (4. ν Ω ∂vh ∂ (u − uh ) dΩ + ∂xi ∂xi v h Uj Ω ∂ (u − uh ) dΩ = 0 ∂xj ∀vh ∈ Vh .u m m+2 1 h V Vh uh Figure 4. figure 4. that is.19) The matrix A is symmetric in this case (since C = 0).18) (cf. This is a “true” orthogonality condition when Uj ≡ 0: ν Ω ∂vh ∂ (u − uh ) dΩ = 0 ∂xi ∂xi ∀vh ∈ Vh . Orthogonality relation (4. the FEM approximation is no longer optimal in this sense.17) thus becomes ν Ω ∂uh ∂uh dΩ = ∂xi ∂xi uh f uh dΩ Ω 2 V 1/2 1/2 f 2 dΩ (Cauchy–Schwarz) ≤ Ω Ω u2h dΩ 1/2 f 2 dΩ (Poincar´e) ≤ C Ω ≤C f Dividing with uh V Ω L2 (Ω) uh V ∂uh ∂uh dΩ ∂xi ∂xi 1/2 . two orthogonal vectors: a · b = 0 .14). (4. u − uh ≤ u − vh ∀vh ∈ Vh . Expression (4. 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 defined.6) and subtracting equation (4. Note that the approximation may be far from optimal when ν is small.7. uh V ≤ Error bounds Choosing v = vh ∈ Vh ⊂ V in variational problem (4.6) yields the “Galerkin orthogonality”. That is. which is the same as saying that the numerical solution is stable.7) from(4. When Uj = 0. there is a C > 0 such that Cea’s lemma holds.

Approximation theory yields that. 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.9 0.94 0. Approximation and mesh quality Estimates (4. 32 1 P tˆ =vh1.7 m m + 2 0. 0 32 V Vh u uh Figure 4.20) are independent of the choice of Vh ! We only used that Vh ⊂ V . The interpolant is the function Πh v ∈ Vh that interpolates the function v ∈ V at the node points of the triangulation. A boundary layer with sharp gradients forms close to x = 1.22) The constant C contains integrals of the second derivatives of v. for instance streamline diffusion and discontinuous Galerkin.3 n u 0.96 0.1 Γ1 0 0. Each of the conditions below is sufficient for approximation property (4.6 i 0. v − Πh v L2 (Ω) v − Πh v V ≤ Ch2 ≤ Ch “second order error” “first order error” (4. The essential point is: beware of “thin” triangles.22) to hold: . that is. whereas the dashed line shows the FE approximation for h = 1/200.21) is quite small: ν = 1/500.9: An interpolant Πh v interpolates v at the nodal points. An explanation of this phenomenon is that the directions of flow is not accounted for in the standard Galerkin approximation. 1).19) and (4. (4. the viscosity coefficient in equation (4.8 shows the FE solution h = 1/1000. Galerkin approximations correspond to finite-difference central schemes.10). Various methods exist to obtain similar effects in FEM. Interpolants are used to study approximation properties.8 1 0.9.2 Γ uh0 0. do not let the shape deteriorate as the mesh is refined. m+2 v1.5 j h 0. for sufficiently smooth v.21) The problem is advection dominated. A strategy to preserve mesh quality when refining is to subdivide each triangle into four new triangles by joining the edge midpoints (figure 4. as illustrated for a 1D case in figure 4. The result of example 1 is typical: under-resolved standard Galerkin methods may generate oscillations for advection-dominated flows in areas of sharp gradients. The solid line in figure 4. Example 1.92 0.4 Inflow V Solid wall Vh 0.u M M0 0. Let us consider the above FEM applied to the 1D problem 1 u +u =1 500 u(0) = u(1) = 0 − in (0. Upwinding is used to prevent oscillations for finite-difference methods. Both these methods modify the basic Galerkin idea to account for flow-direction effects.22) to hold.98 1 Ω x 2 M M Figure 4. A mesh quality assumption is needed for (4.8: Under-resolved standard Galerkin finite-element approximations may yield oscillations for m advection-dominated flows.

in areas of large gradients). piecewise-linear functions. The convergence rate is of optimal order : it agrees with approximation property (4.1. in this case. continuous. where the mesh is refined in certain regions and the order of approximations is increased in other. h (ii) (“Chunkiness parameter condition”). M2 M m (i) (“Maximum angle condition”). Accuracy of the FE solution Error bounds (4. However. This second-order convergence rate requires a mesh quality assumption and a sufficiently smooth solution u (since the constant C contains second derivatives of u). Automatic methods for adaptation are common.11. the solution may still be bad for a particular. These strategies may be combined in the “h–p method”. where it is needed (say. The functions are defined by interpolation into the interior of each quadrilateral. 3D A quadrilateral is a “skewed” rectangle: four points connected by straight lines to form a closed geometric object.M m m+2 1 h V Vh u uh 1 2 i j Inflow Solid wall n Γ0 Γ1 Ω Figure 4.6 Alternative Elements.11) yields a regular structure to the matrix A. uh Condition (ii) implies (i). as in figure 4. 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 → 0.22) and is the best order that in general can be obtained for. to prevent the matrices to become too large.19) and (4. but not the reverse. This space reduces to piecewise bilinear functions for rectangles with edges in the coordinate directions: vh (x. the functions are defined uniquely by specifying the function values at each node. recall example 1. Nonoverlapping quadrilaterals can be used to “triangulate” a domain. This may be done adaptively. This may . piecewise quadratics yields a third-order-accurate solution (in the L2 (Ω) norm). as in figure 4. For instance. As in the triangular case. 4. An alternative is to keep the mesh fixed and increase the order of the polynomials at each triangle (“p-method”). The largest anm+2 gle of any triangle should not approach 180◦ as 1 h → 0.20) together with approximation property (4.10: A subdivision that preserves mesh quality. y) = a + b x + c y + d xy Quadrilaterals on a “logically rectangular domain” (a distorted rectangle. too-crude mesh. Improving Accuracy Accuracy is improved by refining the mesh (“h-method”). An example of a finite-element space Vh on quadrilaterals is globally continuous functions varying linearly along the edges.22) implies u − uh V ≤ Ch Further analysis (not provided here) yields u − uh L2 (Ω) ≤ Ch2 .

1 Ω 2 M 0 0 M 1 1 m 2 2 m+2 i i j j 1 h Inflow Inflow V Solid wall Solid wall Vh n n Γ0 Γ0 u uh Γ1 Γ1 Ω Ω with quadrilaterals.2 1 2 p1. Compared to triangular mesh. Triangular and quadrilateral meshes generalize in 3D to tetrahedral and hexahedral meshes (figure 4.1 p1. triangular and quadrilateral elements generalize to tetrahedral (left) and hexahedral (right) meshes. M 2 Figure 4. and • allow efficient solution of the linear system (particularly for uniform.12). . rectangular meshes). • give high accuracy (particularly if the mesh is aligned with the flow).11: A logically rectangular domain triangulated 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. it is harder to generate quadrilateral meshes automatically for complicated geometries. Advantages and limitations with tetrahedral and hexahedral meshes are similar to corresponding meshes in 2D.

Thus. and consider a situation with flow in a bounded domain Ω.1 h V VhΓ0 u uh 0 ❉✿❈ ❉✿❈ ❉✿❈ ❉✿❈ ❉✿❈ ❉✿❈ ❉✿❈ 1❉✿❈ ❉✿❈ ❉✿❈ ❉✿❈ ❉✿❈ ❉✿❈ ❉✿❈ ❉✿❈ ❉✿❈ ❉✿❈ ❉✿❈ ❉✿❈ Γ❉✿❈ 0 ❉✿❈ ❉✿❈ ❉✿❈ ❉✿❈ ❉✿❈ ❉✿❈ ❉✿❈ ❉✿❈ ❉✿❈ ❉✿❈ ❉✿❈ ❉✿❈ ❉✿❈ ❉✿❈ ❉✿❈ ❉✿❈ ❉✿❈ ❉✿❈ ❉✿❈ ❉✿❈ ❉❈ 2 i j Ω Inflow Solid wall Γ0 n Γ0 Γ1 Ω Figure 4.1.23a) =0 in Ω. as the following example indicates. Assume that we use continuous. (4.23b) = gi on Γ0 (4. (4. M2 Γ0 M m m+2 1 h V Γ0 Γ Vh 0 u uh Γ0 Γ1 Ω Γ0 ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋❊❋❊ ❋❊✿❋❊✿❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋❊❋❊ ❋❊✿❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋❊❋❊ ❋❊✿❋❊✿❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋✿❊❋✿❊ ❋❊❋❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋✿❊ ❋❊ ● ● ●❍ ❍ ❪ ❪❫ ❫ ❬ ❬❭ ❭ ❨ ❨❩ ❩ ■ ■❏ ❏ ❑ ❑▲ ▲ ▼ ▼◆ ◆ ❴ ❴❵ ❵ ❛ ❛❜ ❜ ❖ ❖€ € ❡ ❡❢ ❢ ❝ ❝❞ ❞ ◗ ◗❘ ❘ ❲ ❲❳ ❳ ❯ ❯❱ ❱ ❙ ❙❚ ❚ Γ1 Figure 4. The mathematical problem in differential form is uj ∂ui ∂p 1 2 + − ∇ ui ∂xj ∂xi Re ∂ui ∂xi ui 1 ∂ui nj + pni − Re ∂xj = fi in Ω. Naive approximations of the Navier–Stokes equations are bound to produce disappoing results. but is useful as an artificial outflow condition. 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. A simple example domain is depicted in figure 4.23d) The boundary condition on Γ1 = ∂Ω \ Γ0 is of no obvious physical significance. The left side of equation (4.14.14: The black nodes are the degrees of freedom for the velocity components.and y-components of the velocity field. (4. that is ∂uh ∂vh + ∂x ∂y is here constant on each triangle. This condition introduces new complications not present in the advection–diffusion problem of section 4.13.23b) to be satisfied on each triangle yields 2I 2 equations.23b)! .23b). Consider a case with only Dirichlet boundary conditions—that is.2 FEM for Navier–Stokes We now turn to the Navier–Stokes equations for the steady flow of an incompressible fluid. 4. Γ1 = ∅—and the mesh of figure 4. vh for the x. piecewise-linear approximations u h . demanding equation (4.13: A simple example of a domain. which are fewer than the number of triangles. 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 field. Example 2 (“the counting argument”). The Navier–Stokes equations is a system of nonlinear advection–diffusion equations together with the incompressibility condition (4.23c) =0 on Γ1 .23b).

(i) Use more degrees of freedom for the velocity—for instance continuous. Moreover. gi = 0. This also gives more degrees of freedom for the velocities. The product rule of differentiation yields ∂ ∂ui ∂zi ∂ui zi = + z i ∇2 u i (4. and utilizing (4.23b) with a smooth function q and integrating yields q Ω ∂ui dΩ = 0 ∂xi (4. integrating over Ω. the boundary integrals vanish on Γ0 since zi = 0 on Γ0 and on Γ1 due to boundary condition (4. 4.26) ∂xj ∂xj ∂xj Γ Ω Ω and nj zj p dΓ = Γ zi Ω ∂p dΩ + ∂xi p Ω ∂zi dΩ. could be gravity or magnetic forcing for a electrically conducting fluid.2.24) and (4.1 A variational form of the Navier–Stokes equations To derive the variational form. Strategies (i) and (ii) are the “standard” methods. Multiply both sides of equation (4. let zi be a smooth vector-valued function on Ω that vanishes on Γ0 . for instance. in the last equality.3) on the left sides yields ∂ui ∂zi ∂ui nj z i dΓ = dΩ + zi ∇2 ui dΩ (4. A difference with advection–diffusion problem is that we here need different function spaces for the velocity components and the pressure. the variational form will be used to define weak solutions.26) and (4. but only ∼ 2I 2 vertices). ∂xj ∂xi ∂xj (4. multiplying equation (4. Similarly as for the advection–diffusion problem. Also note that ui = gi on Γ0 whereas zi vanishes on Γ0 . (iii) Associate the velocity with the midpoints of edges instead of the vertices. The pressure space is H = L2 (Ω) = q| q 2 dΩ < +∞ Ω . ∂xi where. piecewise quadratics on each triangle—to balance the large number of equations associated with the incompressibility condition. we find that zi fi dΩ = Ω z i uj Ω = Ω = zi Ω ∂ui 1 z i uj dΩ + ∂xj Re z i uj Ω ∂ui dΩ + ∂xj 1 ∂ui dΩ + ∂xj Re ∂p 1 dΩ − ∂xi Re Ω Ω zi ∇2 ui dΩ Ω ∂zi ∂ui 1 dΩ − ∂xj ∂xj Re ∂zi ∂ui dΩ − ∂xj ∂xj nj z i Γ p Ω ∂ui dΓ − ∂xj p Ω ∂zi dΩ + ∂xi nj zj p dΓ (4. Thus.25) and applying the the divergence theorem (4. (ii) Accept that the discrete velocities only is approximately divergence-free.27) which are the integration-by-parts formulas needed to derive our variational form.23a) with zi . we will only consider homogeneous boundary conditions. we need new integration-by-part formulas. . without explicit reference to the differential equations. the equations will be “driven” only by the right-hand side fi .27). ∂xi (4. To avoid this last complication and simplify the exposition.Following strategies can be used to tackle the problem of example 2. whereas (iii) leads to “non-conforming” approximations related to finite-difference and finite-volume approximation with “staggered mesh”. which.29) Thus.23d).28) and (4.25) Integrating expressions (4. Now.28) Γ ∂zi dΩ. we conclude that a classical solution (a twice continuously differentiable solution) to the Navier– Stokes equations satisfies the variational forms (4. (There are ∼ 3I 2 edges.29) for all smooth functions z i and q such that zi vanishes on Γ0 .24) ∂xj ∂xj ∂xj ∂xj and ∂ ∂p ∂zj (zj p) = zi +p .

so that weak solutions are classical solutions if data is smooth enough. For the unsteady problem on a given. or if singularities can evolve from smooth data. where the size of T ∗ is unknown). Remark 2. whereas the L2 (Ω) norm is 1/2 u L2 (Ω) = ui ui dΩ .23d). u2 ). – It is not known whether smooth data always yields a smooth solution. . that is.2. the boundary conditions are incorporated in the variational formulation. For the analysis in section 4. – smooth data yield smooth solutions. and • in two space dimensions: – a unique weak solution exists for any square-integrable f and for any Re. implicitly specified through the variational form. even in 2D.4. In the case of only Dirichlet boundary condition. 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. – It is not known whether the weak solution always is unique (besides during some initial interval (0. The natural boundary conditions are the boundary conditions on Γ1 . The essential boundary conditions are here that ui = 0 on Γ0 and is explicitly assigned through the definition of V . The energy norm for elements of V is denoted u V = Ω ∂ui ∂ui dΩ ∂xj ∂xj 1/2 . In fact. As for the advection–diffusion problem. finite time interval (0. it will be convenient to work with the space V of the velocity vector u = (u1 . T ∗ ). the pressure is only possible to specify uniquely up to an additive constant. but the more complicated condition (4. The steady problem has at least one weak solution. Ω We assume that both boundary conditions really are active. Saying that u ∈ V is equivalent to saying that that each velocity component u i ∈ V . T ). The variational problem defining weak solutions to equation (4. Note that the steady problem may have several solutions. in 2D as well as 3D. Note that the natural boundary condition here not just is a Neumann condition. • in three space dimensions: – a weak solution exists for any square-integrable f and for any Re. and important issues are still unresolved.23) is Find ui ∈ V and p ∈ H such that ∂ui 1 ∂zi ∂ui z i uj dΩ + dΩ − ∂xj Re ∂xj ∂xj Ω Ω Ω ∂ui q dΩ = 0 ∂xi p Ω ∂zi dΩ = ∂xi zi fi dΩ ∀zi ∈ V Ω ∀q ∈ H This is a variational problem of mixed type: two different spaces are involved.and the space of velocity components is V = u| Ω ∂u ∂u dΩ < +∞ and u = 0 on Γ0 ∂xi ∂xi . Note that the natural boundary condition is different than for the advection–diffusion problem. that Γ0 constitute the whole boundary. The mathematical properties of the Navier–Stokes equations are complicated.

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

dΩ +

1
Re

∂zh ∂uh
dΩ −
∂xj ∂xj

zh

∂vh
∂vh
uh
+ vh
∂x
∂y

qh

∂uh
∂vh
+
∂x
∂y

1
dΩ +
Re

∂zh ∂vh
dΩ −
∂xj ∂xj

ph

∂zh
dΩ =
∂x

zh f1 dΩ

∀zh ∈ Vh ,

(4.30a)

zh f2 dΩ

∀zh ∈ Vh ,

(4.30b)

∂zh
ph
dΩ =
∂y

dΩ = 0.

(4.30c)

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

4.2.3

The algebraic problem in 2D

Expanding ph in basis for Hh , we obtain
Np

p(n) ψ (n) (x, y) .

ph (x, y) =

(4.31)

n=1

Likewise, expanding uh , vh in basis for Vh yields
Nu

Nu

u(n) ϕ(n) (x, y) ,

uh (x, y) =

v (n) ϕ(n) (x, y) .

vh (x, y) =

n=1

(4.32)

n=1

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

u(n)
n=1

ϕ(m) uh

dΩ +


Np

(m)

∂ϕ
ψ (n) dΩ =
∂x

p(n)
n=1

Nu

v (n)
n=1

ϕ(m) uh

n=1

Nu

dΩ +

u(n)

ψ (m)


(n)

dΩ

m = 1, . . . , Nu

N

(n)

∂ϕ(m) (n)
ψ dΩ =
∂y

p(n)

∂ϕ(m) ∂ϕ(n)
∂ϕ(m) ∂ϕ(n)
+
∂x
∂x
∂y
∂y


(n)

∂ϕ
∂ϕ
+ vh
∂x
∂y

Np

1 u (n)
u
Re n=1

ϕ(m) f1 dΩ

n=1

N

∂ϕ(n)
∂ϕ(n)
+ vh
∂x
∂y

1 u (n)
v
Re n=1

ϕ(m) f2 dΩ

∂ϕ(m) ∂ϕ(n)
∂ϕ(m) ∂ϕ(n)
+
∂x
∂x
∂y
∂y

dΩ

(4.33)

m = 1, . . . , Nu


Nu

∂ϕ
dΩ +
v (n)
∂x
n=1

ψ (m)

∂ϕ(n)
dΩ = 0
∂y

m = 1, . . . , Np

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

T
T

,

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

T

,

,

we can write the system (4.33) in the matrix–vector form


  (x) 
1
C(uh , vh ) + Re
K
0
G(x)
u
b
1

K G(y)   v  =  b(y)  ,
0
C(uh , vh ) + Re
p
0
−G(x)T
−G(y)T
0

(4.34)

where the matrix and vector components are
Kmn =

∂ϕ(m) ∂ϕ(n)
∂ϕ(m) ∂ϕ(n)
+
∂x
∂x
∂y
∂y

dΩ,

∂ϕ(m)
ψ (n)
dΩ,
∂x

∂ϕ(m)
ψ (n)
dΩ,
∂y

ϕ(m) uh

Cmn (uh , vh ) =

∂ϕ(n)
∂ϕ(n)
+ vh
∂x
∂y

dΩ,

G(x)
mn = −

G(y)
mn = −

b(x)
m =

ϕ(m) f1 dΩ,

b(y)
m =

ϕ(m) f2 dΩ.

(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 Navier–Stokes equations are directly reflected in the nonlinear
system (4.34). We obtained the same structure of the nonlinear equation for the finite-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 specified. The counting argument of example 2
indicates that not any choice is good. We will limit the discussion to the Stokes equations, whose finiteelement formulation reads
Find uhi ∈ Vh and ph ∈ Hh such that
∂zih ∂uhi
dΩ −
∂xj ∂xj

ph

∂uh
qh i dΩ = 0
∂xi

∂zih
dΩ =
∂xi

zih fi dΩ

∀ zih ∈ Vh ,

(4.36a)

∀ q h ∈ Hh .

(4.36b)

The Stokes equations is a limit case of the Navier–Stokes 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 flow, creeping flow, lubrication, but also flow of common fluids (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 Navier–Stokes 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 Navier–Stokes discretization.
As for the Navier–Stokes 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 (Ω) norm, respectively. Thus, in makes sense
to require that the velocity and pressure approximations satisfy the same bounds, that is,
1/2

ph

L2 (Ω)

=

uh

V

=

p2h dΩ

1/2

≤C

∂uhi ∂uhi
dΩ
∂xj ∂xj

fi fi dΩ

=C f

L2 (Ω) ,

(4.37a)

1/2

≤C f

L2 (Ω) ,

(4.37b)

where the constant C should not depend on h. This implies that the approximations will not blow up when
the mesh is refined.
Choosing zih = uhi in equation (4.36a), we obtain

∂uhi ∂uhi
dΩ −
∂xj ∂xj

ph

∂uhi
dΩ =
∂xi

∂uhi ∂uhi
dΩ =
∂xj ∂xj

uhi fi dΩ,

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

2
V

=

uhi fi dΩ ≤ u

L2 (Ω)

f

L2 (Ω)

≤C u

V

f

L2 (Ω)

(4.38)

where the Cauchy–Schwarz inequality yields the first inequality, and the Poincar´e inequality the second.
We thus conclude from expression (4.38) that
uh

V

≤C f

L2 (Ω) ,

(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 refined, 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

Rearranging equation (4.36a) implies that for each zih ∈ Vh ,
ph

∂zih
dΩ =
∂xi

(Cauchy–Schwarz) ≤

∂zih ∂uhi
dΩ −
∂xj ∂xj

= zh
(Poincar´e) ≤ z h

∂zih ∂zih
dΩ
∂xj ∂xj
uh
uh

V
V

(by (4.39)) ≤ C z h

V

V

L2 (Ω)

L2 (Ω)

f

V

+ zh

1/2

∂uhi ∂uhi
dΩ
∂xj ∂xj

+ zh
+ zh

V

f

zih fi dΩ


1/2

f

1/2

+

zih zih dΩ

1/2

fi fi dΩ

L2 (Ω)

L2 (Ω)

V

f

L2 (Ω) .

(4.40)

Dividing expression (4.40) with z h

yields that

V

1
zh

ph

V

∂zih
dΩ ≤ (C + 1) f
∂xi

L2 (Ω)

for each zih ∈ Vh , from which it follows that there is a C > 0 such that
1

max

0=zi ∈V

zh

V

ph

∂zih
dΩ ≤ C f
∂xi

L2 (Ω) .

(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
α ph

L2 (Ω)

≤ max

0=zi ∈Vh

1
zh

V

∂zih
dΩ
∂xi

ph

(4.42)

for some α > 0 independent of h, then it is immediate that ph L2 (Ω) ≤ C f L2 (Ω) for some C > 0.
Condition (4.42) says that our exotic norm is stronger than the L2 (Ω)-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 (Ω)-norm for each function qh ∈ Hh . The derivation above proves thus that the LBB-condition is
sufficient 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 α > 0 independent of h, such that, for each qh ∈ Hh ,
1/2

α

qh2 dΩ

≤ max

0=zi ∈Vh

1
zh

V

qh

∂zih
dΩ,
∂xi

(4.43)

then there is a C > 0 such that
1/2

p2h dΩ

1/2

≤C

|f |2 dΩ

holds, independently of h.
The LBB condition (4.43) is satisfied for Vh = V and Hh = H. That is, the pressure (before discretiza∂z h

tion) is stably determined by data. The expression Ω qh ∂xii dΩ 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 satisfied, 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 effect that usually manifests itself as wild pressure oscillations.

∂xi for each element K in the mesh.1. equivalently. there is at least one nonzero Np -vector pc such that Gpc = 0. A finite-element approximation is locally mass conservative if K ∂uhi dΩ = 0. All reasonable FE approximation satisfy this. if constant functions can be represented exactly in Hh . .34) is singular : the discrete gradient matrix G has linearly dependent columns. That is. There is no net mass-flux through the boundary for a constant-density flow. Thus. Not all finite-element approximations are locally mass conservative.7 Choice of finite elements. or. so Γ ρ ni ui dΓ = 0. Corresponding function pch ∈ Hh . that is. Accuracy Now we consider a regular triangulation of a 2D domain Ω.30c) that qh Ω ∂uhi dΩ = 0 ∂xi ∀qh ∈ Hh .6 Mass conservation The integral Γ ρ ni ui dΓ yields the total flux of mass(in kg/s) through the boundary.2. However. consider the element contribution. associated with a triangle Tn .1. a triangulation satisfying a mesh quality condition of the sort discussed in section 4. 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.5 that ∂φ ∂x is constant on each triangle): − Tn pch ∂φi ∂φi dΩ = − ∂x ∂x Tn Tn pch dΩ ≡ 0. To see this.5.15: A checkerboard mode for equal-order interpolation on a regular triangular mesh.15 depicts the nodal values of a checkerboard mode for equal-order interpolation. A Galerkin approximation is thus globally mass conservative if 1 ∈ Hh . In particular. 4. it will not be locally mass conservative if the pressure approximations are continuous and piecewise linear. If the pressure approximations consist of piecewise constants on each element. that is. to the ith row of vector Gpc (notice from i the expressions in figure 4. ∂xi by the divergence theorem (4. Recall from equation (4. the FE approximation will be locally mass conservative.3).1). the system matrix (4.Ω M2 M m m+2 1 −1 0 1 0 1 −1 0 0 1 −1 −1 0 1 h V Vh −1 u uh Γ0 Γ1 1 Ω Figure 4.2. A velocity approximation uhi is is globally mass conservative if Γ ni uhi dΓ = 0. “Equal-order interpolation” Let us start by trying the approximation space we used for the advection–diffusion problem (section 4. whose nodal values are equal to the elements of pc is called a checkerboard mode. if Ω ∂uhi dΩ = 0. 4. Figure 4.

16. for triangular elements. piecewise quadratic functions interpolates the nodal values of u (m) (xi ).h V Vh u uh Γ0 Γ1 Ω Figure 4. whereas for quadrilateral elements. vh ∈ Vh is quadratic. the space Vh will be continuous functions. piecewise quadratic velocity approximations of the Taylor–Hood pair. Constant pressure. This choice leads to a locally mass-conserving approximation.16). since ∇v is not a function for discontinuous v. Moreover. • ph ∈ Hh is linear. Thus. the counting argument of example 2 applies here and indicates problems with this choice of elements. Nu A basis φ(m) (xi ) m=1 of continuous. However. The LBB condition is satisfied for the Taylor–Hood pair.43) vanishes for qh = pch . common iterative methods to solve the resulting discrete equations (such as the projection schemes) implies a stabilization of the discretization.15. but not locally mass conservative. There are two kinds of basis figure 4. the constant pressure—bilinear velocity space on rectangles is in widespread engineering use! “Fixes” have been developed to take care of pressure oscillations. the functions in Hh and Vh are globally continuous and. which immediately implies that the LBB cannot hold for a space Hh that includes pch . . we have uh = a + b x + c y + d xy + e x2 + f y 2 The six coefficients a–f are uniquely determined by uh ’s values at the triangle vertices and midpoints (figure 4. The Taylor–Hood pair can be generalized to higher order. Let the functions in H h be piecewise constant on each element. Indeed. This (and other) checkerboard mode will pollute the solution of the Stokes as well as Navier–Stokes equations. if Vh ⊂ V .18. The right side of the LBB condition (4. Still.17 and 4. whereas • uh . Remark 3. (bi)linear velocities Now we consider a triangulation either of triangles or rectangles.15 exist. Then. on each triangle. the pressure and velocity approximations will be continuous functions. illustrated in figures 4.. the checkerboard mode can be filtered out from the discrete pressures. For instance.16: The evaluation points at each triangle for the continuous. on each triangle. since the mean value of the function pch vanishes on each triangle. the LBB conditions is not satisfied for this unstable discretization. piecewise linear on each element. Checkerboard modes similar as in figure 4. The Taylor–Hood pair Here. as can be seen from figure 4. However. so that uh ∈ Vh can be expressed as uh (xi ) = N m=1 ui φ functions here. In case of triangular elements. The pressure approximations need not to be continuous in order for Hh ⊂ H. the velocity approximations cannot be discontinuous. consising of piecewise polynomials of degree k for pressure and k + 1 for velocity (k ≥ 1) on each triangle. continuous functions that are piecewise bilinear on each rectangle comprise Vh . and the following error estimates holds for a smooth solution to the Navier–Stokes equation: u − uh p − ph L2 (Ω) L2 (Ω) ≤ Ch3 ≤ Ch2 The Taylor–Hood pair is globally.

This is fine for piecewise-linear approximations since the error anyway is O(h 1 2 ).1 approximations. C u 23 .j pi.j+1/2 +2 1 vi.velocity and pressure. as in the Taylor–Hood pair. 2 Using curvedi elements restores convergence rate.17: The basis function φ (x ) when Figure v i i. a straight-forward triangulation of a domain with a curved higher-order 0 2 boundary will 0 lead to a boundary-condition error at the mid nodes 1 2 of O(h ) (figure 4.1 v1.19). p1. but each pressure triangleMis subdivided into four velocity triangles as in figureM4. j − i j 1 h j−1 h1 i −V1.j 2vi.j Ω (m) (m) M Figure 4.m+2 m+2 0 d 11d 1 d 1 2h a h a ia V V a i − 1. this inaccuracy at the boundary results in a half-order reduction of the convergence rateiof piecewise quadratic approximations. Both the velocity and pressure M piecewise-linear M 2 functions on triangles. meshes for Ω velocity and pressure.22: Subdividing a pressure rectangle into Figure 4. For instance. jV− j Inflowu ui. 32 v1.20). j − 1 Inflow Solid wall uh j h j−1 i +u1. and m+ 2 the following error estimates holds for m+2 1 1 h h V V Vh Vh u u uh uh Γ0 Γ0 Γ1 Γ1 Ω Ω Figure 4. B ary condition. j − 1 i + 1.j M i. we here consider different 1 Ω 2 approximations consist of continuous. The most common j strategy for curved elements is known j as Isoparametric Elements: piecewise polynomials of the sameInflow degree as used in the FE approximation are used toInflow approximate the shape of the boundary (figure 4. However. i. .20: j − 1Approximating a curved boundary with Figure 4.j+1/2 +2 v m vi.j u 1 Ω ui.j 1 h ui−1/2.1 may reduce the convergence rate of p1.j i − 12 h 1 V i− 2 V Vh 1i i+ 2 Vh 1i u i+ 2 u ui h+ 31 i+ ui h+ 31 Γ0 21 i+ 2 j + Γ0 Γ1 2 j + 12 j Γ1 j jΩ − 12 Ω 1 j−2 j − 1Isoparametric elements use polynomials Figure 4.to approximate a triangle sides B the shape of the boundary.j m pi.j−1/2 O(h2) m i.j−1/2 ui−1/2. m This is amstable approximation satisfying the LBB condition. SolidΓ wall n Γ0 0 j ΓΓ0 i Γ1 n i 1 Γ0 Γi 1+ 1 Ω Γi + 1 Ωi. Solid wall Solid wall n n Γ0 pressures A stable approximation with piecewise-linear velocities and Γ0 Γ1 Instead of Γusing different approximation orders. ity and pressure. 32Inaccuracies in approximations of the domain boundary Remark 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.j4.18: The basis function φ (xi ) when 2 M M m corresponds to an edge-midpoint node m corresponds to a corner node p pi.1 C u 23 . piecewise-linear functions for both veloc.21.j v m i.19: of the sameAorder as used for the FE approximations A leads to an O(h2 ) error of the bound.

vh (x. but are one order less accurate. Calculating the matrix elements (4. we define the nonconforming FE approximation: Find uh . consider basis functions. Violating a property like Vh ⊂ V is called a variational crime. y i are the coordinates of the mid point of edge i. Thus. Both the velocity and pressure approximations consist of continuous and piecewise-bilinear functions. a smooth solution to the Navier–Stokes equation: u − uh L2 (Ω) p − ph L2 (Ω) ≤ Ch2 ≤ Ch These approximations use the same number of unknowns and the same nodal locations as for the Taylor– Hood pair.23. M ∂zh ∂uh dΩ − ∂xj ∂xj n=1 ph ∂zh dΩ = ∂x Tn Tn ∀ z h ∈ Vh . namely to use nodal points on the edge mid points for the velocity. Tn ∀ q h ∈ Hh . zh f2 dΩ ∀ z h ∈ Vh . vh ∈ Vh and ph ∈ Hh such that M M z h uh n=1T n zh vh Tn M n=1T n ∂vh dΩ + ∂x n=1 M zh vh Tn M qh ∂uh ∂vh + ∂x ∂y M ∂uh 1 dΩ + ∂y Re n=1 Tn M z h uh n=1T n ∂uh dΩ + ∂x n=1 ∂vh 1 dΩ + ∂y Re n=1 dΩ = 0 where M is the number of triangles. N is the total number of edges. Vh ⊂ V since ∇uh is not a square-integrable (in fact.24: Approximations spanned by the basis functions (4. To accomplish this. we have φj (xi . However.23) yields piecewiselinear functions that are continuous only along at the edge midpoints. If x i . Now define the space of the velocity components Vh from all possible expansions N uh (x. as in figure 4. constant) function. of the type depicted in figure 4. Figure 4. associated with the edges. Note that uh will be continuous only at edge mid points in general.35) is however somewhat easier than for the Taylor–Hood pair. y) = N uj φj (x.22. j=1 Here.23: The basis functions associated with the velocity approximation for a nonconforming but stable discretization. y) = j=1 vj φj (x. in general. ∇uh restricted on each element is a square-integrable function. y i ) = 1 for i = j. excluding those on Γ1 . Letting Hh be piecewise constant on each triangle and Vh defined as above. Basis function φj is linear on each triangle and continuous along edge j (but discontinuous along four edges!). y). There is also a version of this discretization for rectangular meshes: Each pressure rectangle is then subdivided into four velocity rectangles. A nonconforming method The discussion after example 2 suggested a strategy to address the problem with the overdetermined incompressibility condition. 0 for i = j. y).1 h V Vh u uh Γ0 Γ1 Ω 1 h V Vh u uh Γ0 Γ1 Ω j Figure 4. Ω M ∂zh ∂vh dΩ − ∂xj ∂xj n=1 zh f1 dΩ ph ∂zh dΩ = ∂y Ω .

More advanced stabilization methods avoids this inconsistency by using a term that vanishes as h → 0.46b) formally inconsistent by O( ) with corresponding variational form. . To see that approximation (4. .10).46a) . An alternative is to use the same approximations for velocity and pressure (equal-order interpolation) combined with a stabilization method.30).46) is stable. . (4. the construction of the basis functions η j is not straightforward. it is possible to construct a new vector-valued basis functions η j = (zh .44) into equations (4. M .36) with Find uhi ∈ Vh and ph ∈ Hh such that Ω ∂zih ∂uhi dΩ − ∂xj ∂xj Ω ph Ω ∂qh ∂ph dΩ + ∂xi ∂xi Ω ∂zih dΩ = ∂xi zih fi dΩ ∀ zih ∈ Vh . piecewise-linear functions on a triangular mesh. This circumvents the LBB condition. The main advantage with stabilization methods is the simplified implementation. Nevertheless. Expanding the velocity in this basis. and choosing (zh .7–4. First choose the same type of approximations for the pressure and the velocity components.30) reduce to M j=1 dNu M uj Ω η i · (uh · ∇) η j dΩ + ν uj Ω j=1 ∇η i · ∇η j dΩ + uj Ω j=1 ηi · (uh · ∇) η j dΩ = Ω η i · f dΩ. Note also that the increased number of pressure unknowns. in fact. These manipulations complicate the implementation. Stabilization of unstable discretizations Stability of the choices of elements presented in sections 4. quite similar to equation (4. qh = ph and add equations (4. The added resolution introduced is filtered away by the introduction of the stability term.46b) for some small > 0. . We will present a very simple idea of this sort for the Stokes equations.44) j=1 yields that ∇ · uh = 0. and replace equations (4. (4.23. Inserting expansion (4. we have used the same basis for the uh and vh components. The Navier–Stokes equations then reduces to a vector-valued. But maybe the most interesting with this discretization is that it allows for a construction of a divergence free velocity basis. as compared with the discretization in section 4. non-linear advection– diffusion problem.. although not as accurate as the Taylor–Hood pair. Unfortunately. (4.7 relied on manipulations of the local polynomial order. (4. This leads to a substantial simplification when designing an algorithm to solve the discretized equations. does not increase the accuracy. including the present.The fact that Vh ⊂ V complicates the analysis of this discretization. M uh = (uh . the continuity equation and the pressure variable vanishes. In all types of discretizations discussed so far.2. wh ) = η i . wh ) for the velocity having the property (j) ∇ · ηj = (j) ∂zh ∂wh + ∂x ∂y = 0.46a) Ω ∂uh qh i dΩ = 0 ∂xi ∀ qh ∈ Hh . vh ) = uj η j . the meshes. but the method introduces an extra parameter that may need judicious tuning. choose zih = uhi . and quite easy to implement.7. the approximation is stable.30c). Equation (4. Other nice properties with this method is that it is locally mass conservative. and equations (4.45) for i = 1.2. one of the great challenges for any solver of the Navier–Stokes equations is how to treat the incompressibility condition (4.2. for instance continuous. for instance. or the conformity of the method. By a clever combination of basis functions of the type illustrated (j) (j) in figure 4.

even though the discretization is unstable for = 0. Ω ∂uhi ∂uhi dΩ + ∂xi ∂xi Ω ∂ph ∂ph dΩ = ∂xi ∂xi Ω uhi fi dΩ 1/2 (Cauchy–Schwarz) ≤ Ω (Poincar´e) ≤ C ≤C uhi uhi dΩ ∂uhi Ω ∂uhi ∂xi ∂xi Ω 1/2 fi fi dΩ Ω 1/2 1/2 dΩ ∂ph ∂ph dΩ + ∂xi ∂xi fi fi dΩ Ω Ω ∂uhi ∂uhi dΩ ∂xi ∂xi 1/2 1/2 fi fi dΩ .and (4. . both the velocity and pressure approximations cannot blow up as h → 0. Ω Dividing and squaring yields the stability estimate Ω ∂uhi ∂uhi dΩ + ∂xi ∂xi Ω ∂ph ∂ph dΩ ≤ C ∂xi ∂xi fi fi dΩ. Ω Since C > 0 does not depend on h.46b).

.

j + β 2 (Φi.u h Γ0 2 Φki+1. 1 Φi+1. Requiring an error reduction to O h2 the number of iterations must satisfy ρ m = O h2 which implies that m=O ln(h) ln(ρ) =O ln(h) ln(1 − O(h2 )) = O −h2 ln(h) To proceed we note that we have N = n2 unknown interior nodes.1.j + β 2 (Φi.1 v1.j−1 ) h Φi.j j − 12 j−1 A B C u 23 .j + Φk+1 Φki. it is possible to show Φ − Φk ≤ ρ m Φ − Φ0 where ρ = 1 − O h2 h = ∆x = ∆y e.1 Iterative solutions to linear systems n Γ0 Γ1 method Gauss-Seidel Ω The Laplace equation on a Cartesian grids can be discretized as M2 M Φi+1.j (using the already updated values calculated in same u iteration). i.j + Φi−1.j 2) 2 (1 + β Ω Appendix A Background material io n j+1 ra j tio : new values n di re ct : to be updated ite : old values i−1 i i+1 j−1 This iteration method has a slow convergence.j+1 + Φi.e. 32 p1.1 0 1 2 i j Inflow Solid wall A.j+1 − 2Φi.j + β i.j − 2φi.j .j = V 2 (1 + β 2 ) Vh Gauss-Seidel uses this expression to update Φi.j + Φi. We can solve for Φi.j−1 Γ1 Φk+1 = i. the error is reduced by O h2 each iteration.i. see Figure A.j−1 ) = 0 m +2 where β =m∆x/∆y. Thus 95 .j+1 + Φk+1 i−1.j + Φi−1.

j ≡ −Ri.j = −LΦki.prolongation operator from grid 2 to grid 1 define it as an injection.j to true discrete solution This implies that L∆Φi.j − solution at iteration level k − correction to Φki.j = Φki.j−1 Φi+1.Ω 1 M2 h = ∆x = ∆y = ≈ n−1 = 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 ).g.j + ∆x2 ∆y 2 and define the correction to the solution to the true discrete solution in the following manner Φi.j Φki.restriction operator from grid 1 to grid 2 I21 . The prolongation operator I 21 may be defined by linear interpolation for the intermediate values. the Gauss-Seidel method is provided by the Multigrid method.j+1 − 2Φi.e. The idea behind the multigrid method is that a slowly converging low-frequency on a fine grid is a fast converging high-frequency on a coarse grid.j = 0. an average of right and left values .j since LΦi. but converges slowly since it takes time for boundary information to propagate into the domain. i. This residual equation is solved on a coarser grid and the correction is interpolated onto finer grid. Multigrid method A general way to accelerate the convergence of e.j + Φi.1: Domain for the solution of the Laplace equation.j + ∆Φi.j = Φi. Γ0 Γ1 y Ω : to be updated : new values1 : old values n i−1 i i+1 j−1 j j+1 iteration direction 1 x 1 n 1 Figure A. Let LΦi.j + Φi−1. we use every other value in each direction. In order to define the algorithm we use the following definitions I transfer operator between grids: An example of a restriction operator I12 is to I12 .j − 2Φi.j ∆Φi. i.e. We note that the Gauss-Seidel provides good smoothing of the local error. Thus we are left with an equation implying that the Laplace operator on the correction is equal to the negative of the residual.

The values marked with × in figure A. convergence is checked.j + I21 ∆Φi.j = Φki.j iterated k times.j = Ri.j ⇒ LΦki. L∆Φi.y : old values i−1 i i+1 j−1 j j+1 iteration direction x y 1 n fine grid (1) coarse grid (2) j x i Figure A. LΦi. .j = −I12 Ri.j is definied on grid 2) and goto 1. Now we can define the following two level scheme: 1. ∆Φ0i.j = 0 (starting value) (here ∆Φi.2: Two level grid.j 3. and top and bottom values. Wesseling [10] estimated the work of the multigrid method to be W = O (N ln N ). Φi.2 are then taken as an average of the averages of the intermediate values. a substantial improvement over the Gauss-Seidel method.j iterated k times 2. at the end of 1.

A. j fine grid (1) x2 coarse grid (2) x2 e2 e2 x1 e1 e3 a x3 e1 e3 x1 e1 x3 Figure A. Vectors and tensors are objects which are independent of the coordinate system they are represented in. 2. Example δkk = δll = δ11 + δ22 + δ33 = 3 δij aj = δi1 a1 + δi2 a2 + δi3 a3 = ai .1 Orthogonal transformation Consider two arbitrary oriented coordinate systems as in Figure A. Assume identical origin. A.4.2 Cartesian tensor notation Tensor notation is a compact way to write vector and tensor formulas. In the Cartesian coordinate system in Figure A.4: Two different coordinate systems. k=1 The last expression make use of Einstein’s summation convention: when an index occurs twice in the same expression.2.i y = x2 i i+1 j j−1 fine grid (1) j j coarse + 1 grid (2) iteration direction r x y e1 e1 1 x = x1 n e 3 e2 x z = x3 y i Figure A. The unit vectors are orthogonal ek · el = δkl . 3 The scalar product of two unit vectors is the Kronecker’s delta. a = 0. δkl ek · el = δkl = 1 k=l 0 k=l and the position vector r is 3 r = x 1 e1 + x2 e2 + x3 e3 = xk e k = x k e k .3: Cartesian coordinate system. the expression is implicitly summed over all values for that index.3 we have the unit vectors |ek | = 1 k = 1.

Trsp.. x1 ) x1 = c21 x1 Figure A.2. The relation between ckl and cml can be studied through xk = ckl xl = ckl cml xm ⇒ (ckl cml − δkm ) xm = 0. l = 2) cl k = el · ek = cos(xl . which transform as a vector in each of its R indices. x1 ) = cos(x1 . Since cos(x2 . Component k of r can be written as ek · r = xl ek · el = xl el · ek = xl δkl = xk ⇒ x k = e k · e l xl = c k l xl where we have the transformation tensor ck l = ek · el = cos(xk .5: Projection of e1 x1 on e2 .. = clk clm xm ⇒ (clk clm − δkm ) xm = 0. cml xm δkm xm In the same manner we have xk δkm xm A. xk ) = ck l From ek · r = x l ek · el = xl ek · el ⇒ xk = el · ek xl = cl k xl = clk xl we conclude that first index refer to primed system! Example xk = ckl xl = ck1 x1 + ck2 x2 + ck3 x3 = cos(xk . e3 x3 on e2 .5... . A visualization of the projection of e1 x1 on e2 can be seen in Figure A. 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 . and a position vector is independent of coordinate system. Tklm.2 = clk xl clm xm Cartesian Tensors Def. x1 )x1 + cos(xk . . xl ).. x2 ) in general. a Cartesian tensor of rank R is a set of quantities Tklm.e 1 x1 n x y i j fine grid (1) coarse grid (2) e1 e2 e2 · e1 x1 = cos (x2 .. . e2 x2 . = ckr cls cmp . we have (for k = 1. x2 )x2 + cos(xk . r = x l el = xl el .

2. (3.e.3).1. identical in all coordinate systems. δlm for vectors R = 1 ak = ckl al and for a tensor of second order we have R = 2 δkl = ckm cln δmn = ckm clm = δkl . (1.1. 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 δnp A.1) we get εklm εklp = (3 − 1)δmp = 2δmp and if we also use p = m we end up with εklm εklm = 6 The permutation tensor can also be used to derive the following vector relation (a × b) · (c × d) = εijk aj bk εilm cl dm = (δjl δkm − δjm δlk )aj bk cl dm = aj cj bk dk − aj dj bk ck = (a · c)(b · d) − (a · d)(b · c) (A. 2. (2.2) at least 2 subscripts equal A. 6 6 We have the permutation relation = εklm εksp = δls δmp − δlp δms .2.3. crossproducts and determinants The inner product of the vectors a and b is a · b = δij ai bj = ai bi = a1 b1 + a2 b2 + a3 b3 . (3.Example For scalars the rank is R = 0 a · b = ak bk = ckl ckm al bm = al bl .3 Permutation tensor εklm      1   even  −1 odd = if klm is permutation of (1.2).3. δkl is isotropic.1).1) . The cross product of a and b can be written as e1 a1 b1 a×b= e2 a2 b2 e2 a3 b3 = (a2 b3 − a3 b2 )e1 + (a3 b1 − a1 b3 )e2 + (a1 b2 − a2 b1 )e3 = εijk ei aj bk and the determinant of the tensor aij is |{aij }| = a11 a21 a31 a12 a22 a32 a13 a23 a33 = εijk a1i a2j a3k 1 1 εijk (ai1 aj2 ak3 + ai3 aj1 ak2 + ai2 aj3 ak1 − ai2 aj1 ak3 − ai3 aj2 ak1 − ai1 aj3 ak2 = εijk εpqr aip ajq akr .2.3 we have (1.2.2. If we set s = l in (A.3). 3)     0 no As examples of permutations of 1.1) (2.4 even permutation odd permutation no permutation ε123 = 1 ε213 = −1 ε221 = 0 Inner products. i.

The antisymmetric part of this tensor describes rotation.2. In the Navier-Stokes equations we have ∂ui the velocity gradient ∂x . 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 δkl + T(kl) − Trr δkl + 3 3 isotropic traceless 1 εikl di 2 antisymmetric Properties of the above parts are invariant under transformation. . the isotropic part being 1 T kl = Tmm δkl 3 and the traceless part is 1 T kl = T(kl) − Tmm δkl . where the antisymmetric part is 1 T[kl] = −T[lk] = (Tkl − Tlk ) 2 and the symmetric part 1 T(kl) = T(lk) = (Tkl + Tlk ) . T kl = T(kl) + T[kl] .5 Second rank tensors All second rank tensors can be decomposed into symmetric and antisymmetric parts. 2 The symmetric part can be decomposed into isotropic and traceless part.A. 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 δkl T(kl) − Tmm δkl 3 3 ⇒ Skl T[kl] Skl T(kl) Skl T kl Skl T kl = 1 1 Smm Tkk − Tkk δll 3 3 =0 = S[kl] T[kl] = S(kl) T(kl) = S kl T kl = S kl T kl . T(kl) = T 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 = δpq Tpq = Tpp 1 T kl = ckp clq T pq = ckp clq T(pq) − Trr δpq 3 1 = T(pq) − Trr δkl . the isotropic part the j volume change and the traceless part describes the deformation of a fluid element. In order to sum up. 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 = εijk Tjk = ✘ εijk ✘T✘ (jk) + εijk T[jk] The first term on the right hand side is zero since εijk is antisymmetric in any two indecies. Multiply both sides by εilm εilm di = εilm εijk Tjk = (δlj δmk − δlk δmj ) Tjk = Tlm − Tml = 2T[lm] ⇒ T[lm] = 1 εilm di 2 Note: There are 3 independent components in T[lm] and di .

Example Show that ∇ · (∇ × Ψ) = 0. 2 1 0 T[kl] The symmetric part can then be divided into a isotropic  5 T kl =  0 0 and a traceless part part  0 0 5 0  0 5  T kl The dual vector of the antisymmetric tensor is  −4 3 5 =  3 0 7 .6 Tensor fields We have the tensor field Tij (x1 . . −2 A.2. ∇ · (∇ × Ψ) = since εklm is antisymmetric in kl and ∂ ∂Ψm ∂2 εklm = εklm Ψm = 0 ∂xk ∂xl ∂xk ∂xl ∂2 ∂xk ∂xl is symmetric in kl.Example We have the second rank tensor   1 2 3 {Tkl } =  4 5 6  7 8 9 which can be divided into a symmetric part   1 3 5 = 3 5 7  5 7 9 T(kl) and an antisymmetric part   0 −1 −2 =  1 0 −1  . 5 7 4   −2 {di } =  4  . 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 where (xk = cpk xp ) ∂ Tij ∂xk • divergence decreases tensor rank by one ∂ ∂T1l ∂T2l ∂T3l Tkl = + + ∂xk ∂x1 ∂x2 ∂x3 • curl (∇ × u)i = εijk ∂uk ∂xj which can be compared with the expression for the determinant. x2 .

6: Volume V bounded by the close surface S. see Figure A.6.y 1 n x y i j fine grid (1) coarse grid (2) n S V dS = n dS Figure A.7 Gauss & Stokes integral theorems Let n be the outward pointing normal to the closed surface S bounding the simply-connected volume V . This means. Example √ Show that ∇f (r) = 1r f (r) where {r}k = xk and r = |r| = xk xk . Show that if u ¯=Ω 2 ∂xj εijk ∂ ∂ ∂ uk = {uk = εklm Ωl rm } = εijk (εklm Ωl rm ) = εijk εklm Ωl rm = ∂xj ∂xj ∂xj εkij εklm ∂ ∂ Ωl rm = (δil δjm − δim δjl ) Ωl rm = ∂xj ∂xj ∂ ∂ ∂rm ∂ri ∂rm Ωi rm − Ωl ri = {Ω independent of x} = Ωi − Ωl ={ =1+1+1=3 ∂xm ∂xl ∂xm ∂xl ∂xm ∂ri = δil } = ∂xl 3Ωi − Ωl δil = 2Ωi A. {∇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 (δpk xp + xp δpk ) = f xk = 2 ∂xk 2r r 1 fr r k Example Show that ∇ × (∇ × F) = ∇(∇ · F) − ∆F. ∇ × (∇ × F) i = εijk (δil δjm − δim δjl ) ∂ ∂ ∂ ∂ ∂ ∂ εklm Fm = εijk εklm Fm = εkij εklm Fm = ∂xj ∂xl ∂xj ∂xl ∂xj ∂xl ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ Fm = Fj − Fi = Fj − Fi = ∇(∇ · F) − ∆F ∂xj ∂xl ∂xj ∂xi ∂xj ∂xj ∂xi ∂xj ∂xj ∂xj i Example ¯ × r¯ then Ω = ∇ × u . show that εijk ∂ uk = 2Ωi . Gauss integral theorem then states a · n dS = S ∇ · a dV V or ak nk dS = S V ∂ak dV ∂xk .2.

In the generalized Stokes theorem we then get S n × ∇f dS f dl = C A.7: The open surface S enclosed by the contour C.j fine grid (1) coarse grid (2) n S dl C Figure A. then a · dl = C (∇ × a) · n dS S or ak dxk = C εklm nk S ∂am dS ∂xk this can be generalized to Tklm dxp = C εijp ni S ∂ Tklm dS ∂xj Example Let Tklm = f. f n dS = S a × n dS = S V ∇ × a dV if Tklm nm = εklm al nm .8.8 ∂f εijp ni ∂x dS in right hand side j S Archimedes principle Gauss theorem can be used to calculate the buoyancy on an object immersed in a fluid (Archimedes circa 281–212 BC). This can be generalized to Tklm np dS = S V ∂ Tklm dV . V Stokes theorem states that for a differentiable vector field a if an open simply-connected surface S is enclosed by a closed curve C. m the mass of equivalent volume of water and M the mass of the object in Figure A. for a differentiable vector field a. of which dl is the line element as in Figure A. Let Π be the buoyancy force resulting from the pressure on the object. ∂xp Example ∇f dV .2. The force from the water cylinder is dm g = dV ρg = x3 ρg dA pressure where g is the gravity constant and ρ the water density. The pressure at a distant x 3 from the surface in a fluid is then p = ρgx3 + p0 where p0 is the atmospheric pressure.7. Figure A.8. The force on a small element dS of the body resulting from the pressure is dΠk = −gρx3 nk dS . Consider a water cylinder with volume dV and mass dm.

32 dA j+ 1 p1.1 M x3 iteration direction 0 x 1 y 2 1 Figure A. i This can then be used to calculate the total moment around G form pressure forces on the body i+1 j−1 ∂x3 ∂ MGk = − ρgεklm (xl − xGl )jx3 nm dS = −ρgεklm (xl x3 ) − xGl dV = ∂xm ∂xm S V j+1   iteration direction −ρgεklm [δlm x3 + δxm3 xl − δm3 xGl ] dV = −ρgεkl3  xl dV − xGl V  = 0 y V V 1 If a part of the fluid is frozen (without changing in density) this body is in equilibrium in the fluid: n F = M = 0.10. Πk = − ρgx3unk dS = −ρg ∂x k u h S V Γ0 This means that the object loses some weight as the water is displaced.9: The forces acting on an object m + 2 water and M the mass of the body. m is the mass of equivalent volume of Figure A. 1 h where the minus sign stems from the outward pointing normal n.1 j v1.i + 1 Γ1 p0 i + 32 Ω 1 j +updated 2 : to be j : new 1values n j−2 : old values j−1i−1 dV dm A i Bi+1 Cj − 1 u 23 . The buoyancy force can then be calculated V through Vh ∂ x3 dV = −ρgδk3 V. The buoyancy force is Γ1 Ω Π3 = −ρgV = −g m : to be updated see Figure A.9.8: An object with mass M in a fluid illustrating Archimedes principle. The moment around the center of gravity G of the homogenous body produced by i−1 the pressure force dΠ acting on an element dS of the body is dMG = (r − rG ) × dΠ as in Figure A. x y i j fine grid (1) G coarse grid (2) r − rG n x y i j fine grid (1) coarse grid (2) n Figure A. i immersed n j mg x Inflow y Solid wall i n j 0 fine Γ grid (1) Γ 1 coarse grid (2) Ω M2 Mg M m immersed in a fluid. .10: Moment around G from pressure forces acting on the object. : 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.

r ∂r r ∂θ ∂z (A. ∂r r r ∂θ θ ∂z z Laplacian of a scalar p ∇2 p = 1 ∂ ∂p r r ∂r ∂r + 1 ∂2p ∂2p + 2. ez = e z . ∂t ρ ∂z (A.4) ∂er = eθ . (A. eθ = −ex sin θ + ey cos θ.11. see figure A.3 Curvilinear coordinates Cylindrical Polar Coordinates The cylindrical polar coordinates are (r. ∂t r ρ ∂r r r ∂θ ∂uθ u u 1 ∂p 2 ∂u u + (u · ∇)uθ + r θ = − + ν ∇2 uθ + 2 r − 2θ .10) (A.8) ∂ur ∂uz 1 ∂(ruθ ) ∂ur − eθ + − ez . r2 ∂θ2 ∂z (A. z). ∂t r ρ r ∂θ r ∂θ r ∂uz 1 ∂p + (u · ∇)uz = − + ν∇2 uz . ∂θ Gradient of a scalar p ∇p = ∂p 1 ∂p ∂p e + e + e .11) (A.6) Divergence of a vector u ∇·u= 1 ∂(rur ) 1 ∂uθ ∂uz + + . ∂θ (A. θ.12) .2) where the unit vectors are related to Cartesian coordinates as Non-zero derivatives of unit vectors er = ex cos θ + ey sin θ. (A.11: Cylindrical polar coordinates A. z) z y r θ x Figure A. where θ is the azimuthal angle. θ.3) ∂eθ = −er . The velocity can be written as u = u r er + uθ eθ + uz ez .9) Advective derivative of a scalar p (u · ∇)p = ur Curl of a vector u ∇×u = 1 ∂uz ∂uθ − er + r ∂θ ∂z Incompressible Navier-Stokes equations with no body force ∂ur u2 1 ∂p u 2 ∂u + (u · ∇)ur − θ = − + ν ∇2 ur − 2r − 2 θ .5) (A. ∂r r ∂θ ∂z (A. ∂z ∂r r ∂r ∂θ (A.7) ∂p uθ ∂p ∂p + + uz .i−1 i i+1 j−1 j j+1 iteration direction x y 1 n x y i j fine grid (1) coarse grid (2) z • P (r.

15) (A. r2 ∂r r sin θ ∂θ r sin θ ∂ϕ (A. ϕ) r θ y ϕ x Figure A.i−1 i i+1 j −1 j j +1 ration direction x y 1 n x y i j fine grid (1) coarse grid (2) z • P (r.18) Divergence of a vector u ∇·u= 1 ∂(r2 ur ) 1 ∂(uθ sin θ) 1 ∂uϕ + + . eθ = ex cos θ cos ϕ + ey cos θ sin ϕ − ez sin θ. (A. θ. The velocity can be written as u = u r er + uθ eθ + uϕ eϕ . r ∂r ∂θ (A.13) where the unit vectors are related to Cartesian coordinates as er = ex sin θ cos ϕ + ey sin θ sin ϕ + ez cos θ. ∂θ ∂eθ = eϕ cos θ. see figure A. eϕ = −ex sin ϕ + ey cos ϕ. where ϕ is the azimuthal angle. ϕ).14) Non-zero derivatives of unit vectors ∂er = eθ . θ. r2 sin2 θ ∂ϕ2 (A. ∂ϕ ∂eθ = −er . ∂ϕ ∂eϕ = −er sin θ − eθ cos θ.16) ∂p 1 ∂p 1 ∂p e + e + e .20) Curl of a vector u ∇×u= 1 r sin θ ∂(uϕ sin θ) ∂uθ ∂(ruϕ ) 1 1 ∂ur − er + − eθ ∂θ ∂ϕ r sin θ ∂ϕ ∂r + 1 ∂(ruθ ) ∂ur − eϕ .17) Laplacian of a scalar p ∇2 p = 1 ∂ ∂p r2 2 r ∂r ∂r + 1 ∂ ∂p sin θ 2 r sin θ ∂θ ∂θ + 1 ∂2p . ∂r r r ∂θ θ r sin θ ∂ϕ ϕ (A.12: Spherical polar coordinates Spherical Polar Coordinates The spherical polar coordinates are (r. ∂r r ∂θ r sin θ ∂ϕ (A. ∂ϕ Gradient of a scalar p ∇p = (A.21) .12. (A.19) Advective derivative of a scalar p (u · ∇)p = ur ∂p uθ ∂p uϕ ∂p + + . ∂θ ∂er = eϕ sin θ.

ρr sin θ ∂ϕ r sin θ ∂ϕ ∂ϕ r sin θ r sin θ (A. ρ ∂r r r sin θ ∂θ r sin θ ∂ϕ u2ϕ cot θ ∂uθ u u + (u · ∇)uθ + r θ − ∂t r r 1 ∂p 2 cos θ ∂uϕ 2 ∂u u =− + ν ∇2 u θ + 2 r − 2 θ 2 − 2 2 .23) (A.24) . ρ r ∂θ r ∂θ r sin θ r sin θ ∂ϕ ∂uϕ ur uϕ u uϕ cot θ + (u · ∇)uϕ + + θ ∂t r r uϕ 1 ∂p 2 ∂ur 2 cos θ ∂uθ =− + ν ∇2 u ϕ + 2 + 2 2 − 2 2 .Incompressible Navier-Stokes equations with no body force u2 + u2ϕ ∂ur + (u · ∇)ur − θ ∂t r =− (A.22) 1 ∂p 2u 2 ∂(uθ sin θ) 2 ∂uϕ + ν ∇2 ur − 2r − 2 − 2 .

1 Tensors and invariants Tensor Notation Scalar p=p Vector   u If u ¯ =  v  then u2 = v w (¯ u)i = ui Matrix (A)ij = Aij  a11 If A = a21 a31 a12 a22 a32  a13 a23  then A23 = a23 a33 Kronecker delta δij = (I)ij = 1 if i = j 0 if i = j Einstein summation convention 3 ui ui = ui ui i=1 Kinetic energy per unit volume 1 u|2 2 ρ|¯ = 21 ρ(u2 + v 2 + w2 ) = 21 ρui ui Matrix operations a ¯ · ¯b = a1 b1 + a2 b2 + a3 b3 = ai bi = δij ai bj = aj bj (A¯b)i = Aij bj (AB)ij = Aik Bkj tr(A) = Aii (A)ij = Aij ⇐⇒ (AT )ij = Aji Permutation symbol 109 .Appendix B Recitations 5C1214 B.

The anti-symmetric part describes rotation. 213 or 132 e¯1 a1 b1 a ¯ × ¯b = e¯2 a2 b2 e¯3 a3 b3 = e¯1 (a2 b3 − a3 b2 ) − e¯2 (a1 b3 − a3 b1 ) + e¯3 (a1 b2 − a2 b1 ) = εijk e¯i aj bk (¯ a × ¯b)i = εijk aj bk εijk εilm = δjl δkm − δjm δkl εijk εijm = {l = j} = 3δkm − δjm δkj = 3δkm − δmk = 2δkm εijk εijk = {m = k} = 2 · 3 = 6 Rewrite without the cross product: ¯ = (¯ ¯ i = εijk aj bk εilm cl dm = εijk εilm aj bk cl dm = (¯ a × ¯b) · (¯ c × d) a × ¯b)i (¯ c × d) ¯ − (¯ ¯ ¯b · c¯) (δjl δkm − δjm δkl ) aj bk cl dm = aj cj bk dk − aj dj bk ck = (¯ a · c¯)(¯b · d) a · d)( Invariant parts of Tensors Tij = TijS + TijA Symmetric and Anti-symmetric parts. In the NS equations we have the tensor Operators (∇p)i = ∆p = ∂ p ∂xi ∂2 p ∂xi ∂xi ∇·u ¯= ∂ ui ∂xi (∇ × u ¯)i = εijk ∂ uk ∂xj Gauss theorem (general) Gauss theorem: . ijk = 321. 231 or 312 = 0 if any two indices are equal  −1 if ijk in anticyclic order.εijk   1 if ijk in cyclic order. the isotropic ∂xj part describes the volume change and the trace-less part describes the deformation of a fluid element. ijk = 123. TijS = 1 2 Tij + Tji = TjiS TijA = 1 2 Tij − Tji = −TjiA Symmetric Anti-symmetric The symmetric part of the tensor can be divided into two parts: T S = T¯ + T¯ ij ij ij T¯ij = TijS − 31 Tkk δij T¯ij = 13 Tkk δij trace less isotropic This gives: Tij = TijS + TijA = T¯ij + T¯ij + TijA ∂ui .

show that εijk uk = 2Ωi . ∂xi ∂xj ∂xj ∂xi ∂ ∂ Fk = 0 ∂xi ∂xj Example ¯ ∂ ¯ × r¯ then Ω ¯ = ∇×u Show that if u ¯=Ω .S F¯ · n ¯ dS = V ∇ · F¯ dV V ∂Fi dV ∂xi or. 2 ∂xj εijk ∂ ∂ ∂ uk = {uk = εklm Ωl rm } = εijk (εklm Ωl rm ) = εijk εklm Ωl rm = ∂xj ∂xj ∂xj εkij εklm ∂ ∂ Ωl rm = (δil δjm − δim δjl ) Ωl rm = ∂xj ∂xj ∂ ∂ ∂rm ∂ri ∂rm Ωi rm − Ωl ri = {Ω independent of x} = Ωi − Ωl ={ =1+1+1=3 ∂xm ∂xl ∂xm ∂xl ∂xm 3Ωi − Ωl δil = 2Ωi Derive the identity ¯ = (∇ × F¯ ) · G ¯ − F¯ · (∇ × G) ¯ ∇ · (F¯ × G) ∂ri = δil } = ∂xl . This means. Fi ni dS = S In general we can write. ∂ Tijk dV ∂xl Tijk nl dS = S V Example: Put Tijk nl = Tij ul nl ∂ (ul Tij ) dV ∂xl Tij ul nl dS = S V or. S T(¯ u·n ¯ ) dS = V (∇ · u ¯)T + (¯ u · ∇)T dV Identities Derive the identity ¯ = (G ¯ · ∇)F¯ + (∇ · G) ¯ F¯ − (∇ · F¯ )G ¯ − (F¯ · ∇)G ¯ ∇ × (F¯ × G) ¯ ∇ × (F¯ × G) i = εijk (δil δjm − δim δjl ) Gj ∂ ∂ ∂ εklm Fl Gm = εijk εklm Fl Gm = εkij εklm Fl G m = ∂xj ∂xj ∂xj ∂ ∂ ∂ ∂Fi ∂Gj ∂Fj ∂Gi Fl G m = Fi G j − Fj G i = Gj + Fi − Gi + Fj = ∂xj ∂xj ∂xj ∂xj ∂xj ∂xj ∂xj ∂Fi ∂Gj ∂Fj ∂Gi ¯ · ∇)F¯ + (∇ · G) ¯ F¯ − (∇ · F¯ )G ¯ − (F¯ · ∇)G] ¯i + Fi − Gi − F j = [(G ∂xj ∂xj ∂xj ∂xj Show that: ∇ · (∇ × F¯ ) = 0 ∂ ∂ ∂ ∂ εijk Fk = εijk Fk ∇ · (∇ × F¯ ) = ∂xi ∂xj ∂xi ∂xj Remember that εijk = −εjik and that εijk ∂ ∂ ∂ ∂ Fk = Fk and thus all terms will cancel.

¯ = ∇ · (F¯ × G) εkij ∂ ∂ ∂Fj ∂Gk ∂Fj ∂Gk εijk Fj Gk = εijk Fj Gk = εijk Gk + εijk Fj = εkij Gk + Fj εijk = ∂xi ∂xi ∂xi ∂xi ∂xi ∂xi ∂Fj ∂Gk ¯ − F¯ · (∇ × G) ¯ Gk − Fj εjik = (∇ × F¯ ) · G ∂xi ∂xi Show that: ∇ × (∇ × F¯ ) ∇ × (∇ × F¯ ) = ∇(∇ · F¯ ) − ∆F¯ i = εijk (δil δjm − δim δjl ) ∂ ∂ ∂ ∂ ∂ ∂ εklm Fm = εijk εklm Fm = εkij εklm Fm = ∂xj ∂xl ∂xj ∂xl ∂xj ∂xl ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ Fm = Fj − Fi = Fj − Fi = ∇(∇ · F¯ ) − ∆F¯ ∂xj ∂xl ∂xj ∂xi ∂xj ∂xj ∂xi ∂xj ∂xj ∂xj i .

u0 . Lets denote the different parts as u ¯=u ¯T + u ¯R + u ¯D In tensor notation it can be written as P P ui 2 = ui 1 + ξij dxj + eij dxj P (i) Translation is simply u ¯T = {ui 1 } = (βy. y= 1 k 2 (x − 2x0 x + x20 ) + y0 2 u20 Exercise 2 a) Separate the shear flow u ¯ = (βy. k > 0 is straight lines. while any fluid particle follows a parabolic path as time proceeds. Streamline (fix 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. v = kt.2 Euler and Lagrange coordinates Particle path (Lagrange’s representation) Following a fluid particle as time proceeds the path is described by ∂¯ r =u ¯. 0. 0) into its (i) local translation. y= Particle path (fix x ¯0 ): ∂x = u = u0 ∂ tˆ ∂y = u = k tˆ ∂ tˆ kt x + d(t) u0 ⇒ ⇒ x = u0 tˆ + x0 x= 1 ˆ2 k t + y0 2 Then we see that y(x) is a parabola.B. w = 0. 0) . 0. (ii) rotation and (iii) pure straining parts. ∂ tˆ r¯(tˆ = 0) = x ¯0 Streamline (Euler’s representation) Instantaneously a fluid particles displacement d¯ x is tangential to its velocity u ¯: 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 flow u = u0 .

5 −1 −1 0. 0) n 2 Γ0 Γ1 Ω 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 flow. 0. 0) 2 β 0 0  0 0 0 1 (iii)   2 0 β 0 1 ∂ui ∂uj 1 i + = β 0 0  eij = j 2 ∂xj ∂xi 2 0 0 0 Inflow 1 Solid wall u ¯D = ed¯ x = (βdy. 32 p1. 0). 0) = (βdy.5 0 1 −1 −1 0 −0. 1) 2 2 i−1 iNow sketch the decomposition of the flow i+1 j−1 1 1 (βdy.5 = 0 −0. 0) + (βdy. λ2 = −β/2.1  0 β ∂ui = 0 0 ∂xj 0 0  0 0 0 ξij = 1 ∂ui ∂uj − 2 ∂xj ∂xi u ¯R = ξd¯ x=  0 1 = −β 2 0 1 (βdy. 0). 1 ∂ui ∂uj ∂ui m + symmetric part of . −βdx. βdx. eij = 2 ∂xj ∂xi ∂xj m+2   1 0 β 0 1 h eij = β 0 0 2 V 0 0 0 Vh uEigenvalues: uh −λ β/2 0 Γ0 0 = −λ3 +β 2 λ/4 = −λ(λ−β/2)(λ+β/2) = 0. 1. ⇒ λ1 = β/2.5 0 1 −1 −1 0 1 . k¯2 = √ (−1. 1.5 0 + −0.1 v1. 0.5 0. k¯3 = (0. det(eij −λδij ) = β/2 −λ Γ1 0 0 −λ Ω : to be updatedWe get the eigenvectors from eij kj = λki : : new values 1 1 : old values k¯1 = √ (1. −βdx.j − 21 (ii) j−1 A B C u 23 . βdx. λ3 = 0. 0) j 2 2 j+1 iteration direction x 1 1 1 y n x y i j fine grid (1) coarse grid (2) 0.

2 0 0. Inflow dx dy Solid wall = αx.r. v. M mWe have.8 −0.4 −0. We have a 2D flow since w = 0. j Streamlines in parametric form (x(s). where α > 0. z(s)).4 0. y. <0 x > 0. >0 ds ds to be updated dx dy : new values x > 0. = −αy. Does this concentration change for a particular fluid particle in time? Use the material time-derivative ∂c ∂c ∂c Dc = +u +v = −αβx2 ye−αt + αx2βxye−αt − αyβx2 e−αt = (−1 + 2 − 1)βx2 ye−αt = 0. y > 0 ⇒ u ds ds uh dx dy Γ0 x < 0.1 Exercise 3 2 ia) Sketch the streamlines for the flow [u. t) = βx2 ye−αt . −αy.5 i j fine grid (1) −1 coarse grid (2) −1 −0. y > 0 ⇒ < 0. ds ds n Γ0 Integrate w. y.8 1 b) The concentration of a scalar is c(x. c c m+2 x(s) = . tˆ).6 −0.5 x y 0 n x y −0. y < 0 ⇒ < 0. y(s). <0 ds ds Γ1 dx dy Ω x < 0. c) Using the alternative (Lagrangian) description of the flow. w] = [αx. Dt ∂t ∂x ∂y Thus the concentration c(x. t) is constant for a fluid particle. show that ∂u ¯ D¯ u = ˆ Dt ∂t and write the concentration as c = c(x0 . . Ω M 2 Then we see that. s. x(s) y(s) = ab = c ( constant ). y < 0 ⇒ > 0.t.2 0. >0 : old values ds ds i−1 dx(s) dy(s) = αx and = −αy. This gives the direction of the flow: iRemember that ds ds i+1 j −1 1 j j +1 ration direction 0. y(s) = 1 y(s) x(s) h Streamlines: V Vh dx dy > 0.6 0. y(s) = b e−αs . y0 . Γ1 x(s) = a eαs . 0].

independent of tˆ! . y) = βx2 ye−αt = β x20 e2αt y0 e−αt e−αt = βx20 y0 = c(X) x2 y The concentration foes not change as we follow a fluid particle.Particle path: ∂x ˆ = u = αx ⇒ x = x0 eαt ˆ ∂t ∂y ˆ = v = −αy ⇒ y = y0 e−αt ˆ ∂t ∂u ∂ ∂x ˆ = = α 2 x0 e α t = α 2 x ˆ ˆ ˆ ∂t ∂t ∂t ∂v ∂ ∂y ˆ = = α2 y0 e−αt = α2 y ∂ tˆ ∂ tˆ ∂ tˆ Du ∂u ∂u ∂u = +u +v = α2 x Dt ∂t ∂x ∂y Dv ∂v ∂v ∂v = +u +v = α2 y Dt ∂t ∂x ∂y So we have ∂u ¯ D¯ u = ˆ Dt ∂t Both terms describe the acceleration of a fluid particle. Now rewrite the concentration ˆ ˆ ˆ ˆ ¯ c(x.

Show that conservation of mass implies ∂ρ ∂ + (ρui ) = 0. Dt ∂xi ∂ρ ∂ui ∂ρ ∂ui ∂ (ρui ) = ui + ρ = ui +ρ ∂xi ∂xi ∂xi ∂xi ∂xi This gives ∂ρ ∂ρ Dρ ∂ui ∂ui + ui +ρ = +ρ ∂t ∂xi ∂xi Dt ∂xi Dρ Dt c) Explain what the following means ∂ui =0 ∂xi ⇒ Dρ =0 Dt If we are following a fluid particle. the density can change ∂ρ ∂ρ = −ui ∂t ∂xi Remember the decomposition of ∂ui ∂xj in its invariant parts ∂ui 1 ∂ui ∂uj = + ∂xj 2 ∂xj ∂xi − 1 ∂uk 1 ∂ui ∂uj δij + − 3 ∂xk 2 ∂xj ∂xi deformation e¯ij . rotation ξij + 1 ∂uk δij 3 ∂xk volume change e¯ij . But in a fix position.B. the density and volume are constant. ∂t ∂xi This must be true for an arbitrary volume and this implies ∂ ∂ρ + (ρui ) = 0. ∂t ∂xi b) Show that this can be written Dρ ∂ui +ρ = 0. ∂xi ∂ρ ∂ + (ρui )} dV = 0. ∂t ∂xi The change of mass in the volume is described by the contribution from the mass flux and the change in density and conservation means that this should be zero V ∂ρ dV + ∂t ρui ni dS = 0 S Using Gauss Theorem we can rewrite the surface integral ρui ni dS = S V We have V { ∂ (ρui ) dV.3 Reynolds transport theorem and stress tensor Exercise 1 a) Consider a fixed closed surface S in a fluid.

D Dt ∂ρ ∂ + (ρui ) dV ∂t ∂xi ρ dV = V (t) V (t) No flow through the surface.Exercise 2 a) Use Reynolds transport theorem to provide an alternative derivation of the conservation of mass equation ∂ρ ∂ + (ρui ) = 0. where ω ¯ is constant. Valid for all V (t): ∂ρ ∂ + (ρui ) = 0 ∂t ∂xi b) Use this result to show D Dt V (t) ρFijk dV = ρ V (t) DFijk dV Dt Put Tijk = ρFijk in Reynolds transport theorem D Dt V (t) ρ V (t) V (t) ρFijk dV = V (t) ∂ ∂ (ρFijk ) + (u ρF ) dV = ∂t ∂xl l ijk ∂Fijk ∂Fijk ∂ρ ∂ + Fijk + Fijk (ρul ) + ρul ∂t ∂t ∂xl ∂xl Fijk ∂Fijk ∂Fijk ∂ρ ∂ + (ρul ) + ρ + ul ∂t ∂xl ∂t ∂xl =0 = ρ V (t) dV = DF ijk Dt DFijk dV Dt Exercise 3 Show that the stress tensor Tij = −pδij for the flow u ¯=ω ¯ ×x ¯. ∂t ∂xi Reynolds Transport Theorem: D Dt V (t) Tijk dV = ∂Tijk ∂ + (u T ) dV ∂t ∂xl l ijk V (t) Put Tijk = ρ. Tensor notation: ui = εikl ωk xl The stress tensor: Tij = −pδij + µ −pδij + µ ∂uj ∂ui + ∂xj ∂xi = ∂ ∂ (εikl ωk xl ) + (ε ω x ) ∂xj ∂xi jkl k l −pδij + µωk εikl ∂xl ∂x + εjkl l ∂xj ∂xi = −pδij + µωk (εikl δlj + εjkl δli ) = −pδij + µωk (εikj + εjki ) = −pδij + µωk (−εijk + εijk ) = −pδij dV = = .

∂ui (iii) the relationship between τij and should be isotropic as the physical properties of the fluid ∂xj are assumed to show no preferred direction. Tij = −pδij + 2 µeij Writing this on the form. Tii = −3p + 3λeii + 2µeii = −3p + (3λ + 2µ)eii and thus 2 Tij = −p − µekk δij + 2µeij 3 2 ∂uk ∂ui ∂uj Tij = − p + µ δij + µ + 3 ∂xk ∂xj ∂xi . then since eij = eji we have τij = λekk δij + 2µeij 1 b) Show that if we have a Newtonian fluid where p = − Tii then 3 ∂ui ∂uj 2 ∂uk Tij = − p + µ δij + µ + 3 ∂xk ∂xj ∂xi We have Tij = −pδij + τij = −pδij + λekk δij + 2µeij Then This gives λ = or − 23 µ. τij = cijkl ekl = (Aδij δkl + Bδik δjl + Cδil δjk )ekl = Aδij ekk + Beij + Ceji Say that A = λ and that B = C = µ. ∂xj (ii) each τij should vanish if the flow involves no deformation of fluid elements.The stress tensor Tij The equation for the stress tensor was deduced by Stokes in (1845) from three elementary hypotheses. a) Use this to show that τij = λekk δij + 2µeij where λ and µ are scalars. B and C are scalars. Exercise 4 If τij is a linear function of the components of eij then it can be written τij = cijkl ekl It can be shown that the most general fourth order isotropic tensor is of the form cijkl = Aδij δkl + Bδik δjl + Cδil δjk where A. ∂ui (i) each τij should be a linear combination of the velocity gradients . Tij = −pδij + τij the following statements should be true for the viscous stress tensor τij in a Newtonian fluid.

i i+1 3 j−1 j 2 j+1 2 iteration direction x y 1 n x y i j fine grid (1) coarse grid (2) 1 0 0 1 r 3 2 0 1 0 r 3 2 Figure B. use the appendix formulas u ¯ · ∇ = ur (¯ u · ∇)¯ u= ∂ uθ ∂ ∂ + + uz ∂r r ∂θ ∂z uθ ∂ uθ ∂uθ uθ ∂¯ eθ uθ ∂uθ u2 (uθ e¯θ ) = e¯θ + uθ = e¯θ − θ e¯r r ∂θ r ∂θ r ∂θ r ∂θ r =−¯ er ∇p = ∂p 1 ∂p ∂p e¯r + e¯θ + e¯z ∂r r ∂θ ∂z Insert this into the Euler equations: uθ ∂uθ u2 1 ∂p 1 ∂p ∂p e¯θ − θ e¯r = − e¯r + e¯θ + e¯z r ∂θ r ρ ∂r r ∂θ ∂z =0 − g¯ ez . and a decay of angular velocity Γ0 outside.4 Vh Rankine vortex and dimensionless form The Rankine vortex u θ ω≈0 uh A simple model for a vortex is that it is a rigid body rotation within a core. : to be updated uθ = ωa2 ur = u z = 0  .1: Velocity and vorticity in a Rankine vortex with ω = a = 1. r < a. r > a. This can be described by Γ1  Ω ωr. Exercise 1 a) Find the pressure inside and outside of a Rankine vortex: We use Eulers equations for incompressible flow ( neglecting viscous effects ):  u 1  D¯ = − ∇p + g¯ Dt ρ Eulers equations  ∇·u ¯=0 ∂u ¯ D¯ u = +(¯ u · ∇)¯ u Dt ∂t =0 We are working with cylindrical coordinates.h V B. : new values r : old values i − 1and is called a Rankine vortex.

z) = − ρω 2 a2 ρω 2 a2 − ρgz + C1 = − − ρgz + C2 2 2 C2 − C1 = ρω 2 a2 ⇒ b) Determine the pressure difference ∆p between r = 0 and r → ∞ ∆p = p∞ − p0 = −ρgz + C2 − (−ρgz + C1 ) = C2 − C1 = ρω 2 a2 c) Now consider a free surface at atmospheric pressure p0 . z) = ρω 2 p = ρω 2 ⇒ r2 + f (z) 2 f (z) = −ρgz + C1 r2 − ρgz + C1 2 r<a Solve for the pressure when r > a: e¯r : − ω 2 a4 1 ∂p =− r3 ρ ∂r e¯z : ∂p = −ρg ∂z ⇒ ⇒ p=− ρω 2 a4 + f (z) 2 r2 f (z) = −ρgz + C2 So we have the pressure: ρω 2 a4 − ρgz + C2 r > a 2 r2 Now determine the difference between the constants by evaluation at r = a p(r. z) only. ρ r ∂θ 0=− e¯z : 1 ∂p −g ρ ∂z Solve for the pressure when r < a: −ω 2 r = − e¯r : e¯z : 1 ∂p ρ ∂r ∂p = −ρg ∂z ⇒ So we have the pressure: p(r.Look at the different components: − e¯r : e¯θ : 0=− u2θ 1 ∂p =− r ρ ∂r 1 1 ∂p ⇒ p = p(r. r>a . Find the difference in z between r = 0 and r → ∞ r=0: ρω 2 a2 − ρgz0 + C1 2 ρω 2 a2 p0 = − 2 − ρgz∞ + p0 = r→∞: ⇒ C2 z∞ − z0 = C2 − C 1 ω 2 a2 = ρg g Determine the shape of the free surface: p0 = p0 = ρω 2 r 2 − ρgz + C1 2 ρω 2 a4 − 2r2 − ρgz + C2 C1 −p0 ω2 r2 2g + ρg 2 4 −p0 = − ω2gra2 + C2ρg r < a z ∼ r2 ⇒ z = r>a z∼ 1 r ⇒z Say that z = 0 at r = 0 then C1 = p0 and we get z= and z=− ω 2 r2 2g r<a ω 2 a4 C2 − C 1 ω 2 a4 ω 2 a2 ω 2 a2 a2 + =− + = 1− 2 2 2 2gr ρg 2gr g g 2r So we have z(r) =  ω2 r2  2g 2 2  ω ga r<a 1− a2 2r 2 r > a.

∂ a ∂ Note that the scaling x ¯ = x¯/a implies ∇ = a∇ and t = tU/a gives = . 0. with the boundary conditions u ¯ = 0 on x2 + y 2 = a2 u ¯ → (U.5 2 3 Figure B.01 0 0 0.5 1.5 r 1 2.82.1 0.08 0.2: The free surface of a Rankine vortex with ω = a = 1 and g = 9.05 0. ∂t U ∂t Change to dimensionless variables U 2 ∂u ¯ U2 ρU 2 νU 2 + (¯ u · ∇ )¯ u =− ∇p + 2 ∇ u ¯ a ∂t a ρa a Divide by U 2 /a ∂u ¯ + (¯ u · ∇ )¯ u = −∇ p + ∂t νU a2 U2 a 2 ∇ u ¯ 1 Re U2 Inertial forces Ua The Reynolds number is Re = a = = νU Viscous forces ν a2 ∂u ¯ 1 2 + (¯ u · ∇ )¯ u = −∇ p + ∇ u ¯ ∂t Re The continuity condition ∇·u ¯=0 ⇒ U ∇ ·u ¯ =0 a ⇒ ∇ ·u ¯ =0 .09 0. Solving the 2D flow around a cylinder involves solving 1 ∂u ¯ + (¯ u · ∇)¯ u = − ∇p + ν∇2 u ¯.06 0. 0) as x2 + y 2 → ∞.07 0. ∂t ρ ∇·u ¯ = 0.z i i+1 j−1 j j+1 iteration direction x y n x y i j fine grid (1) coarse grid (2) 0.04 0. z(r) = Exercise 2  ω2 r2  2g 2 2  ω ga r<a 1− a2 2r 2 r > a. Rewrite this problem in dimensionless form using the dimensionless variables x ¯ = x¯/a u ¯ =u ¯/U p = p/ρU 2 t = tU/a.02 0.03 0.

e. Re and t only. ∂τij ∂ =µ ∂xj ∂xj −µεijk ∂ui ∂uj + ∂xj ∂xi ∂ωk ∂ ∂um = −µεijk εklm ∂xj ∂xj ∂xl −µ Thus =µ ∂ 2 ui ∂ 2 uj + ∂xj ∂xj ∂xj ∂xi = −µεkij εklm ∂ 2 ui ∂ 2 uj − ∂xj ∂xi ∂xj ∂xj =µ ∂ 2 ui ∂xj ∂xj ∂ 2 um ∂ 2 um = −µ(δil δjm − δim δjl ) = ∂xj xl ∂xj xl =µ ∂ 2 ui ∂xj ∂xj ∂τij ∂ωk = −µεijk ∂xj ∂xj The net viscous force vanishes when the vorticity is uniform. 0. since no deformation exists. Exercise 3 Show that the net viscous force per unit volume is proportional to the spatial derivative of vorticity. 0) as x + y 2 2 2 = 1. 0. i. and thus the solution of this problem is the same for a specific Re independently of the individual values of U . ∂τij ∂ωk = −µεijk ∂xj ∂xj and discuss its implication for flows with uniform vorticity (as in solid-body rotation). 0) as a2 x + a2 y 2 u ¯ → (1. . 0.The boundary conditions u ¯ = 0 on x2 + y 2 = a2 ⇒ 2 Uu ¯ = 0 on a2 x + a2 y u ¯ → (U. →∞⇒ →∞ The solutions of this problem will depend on x ¯ . a and ν. 0) as x2 + y 2 → ∞ ⇒ 2 = a2 ⇒ 2 u ¯ = 0 on x + y 2 Uu ¯ → (U.

j+ j B. Calculate the flow field.1 v1.3: Coordinate system for plane Poiseuille flow Navier–Stokes equations:  ¯ 1  ∂u + (¯ u · ∇)¯ u = − ∇p + ν∇2 u ¯ ∂t ρ  ∇·u ¯ = 0. The upper plane is Cmoving with velocity U . u 23 . u(h) = U ⇒ A = U/h uh Γ0 So we get: Uy Γ1 u(y) = h Ω : to be updated : new valuesExercise 2 (i) : old valuesPlane Poiseuille flow: ( Channel flow ) i − 1Consider the flow of a viscous fluid between two solid boundaries at y = ±h driven by a constant pressure igradient ∇p = [−P. 32 Assume the following: p1. . j ∂x InflowTwo-dimensional flow: ∂ Solid wall w = 0. Show that i+1 P 2 j−1 u= (h − y 2 ). =0 ∂z n Γ0 No pressure gradient: ∂p Γ1 =0 ∂xi Ω M 2 Navier–Stokes streamwise momentum equation: M ∂u 1 m + (¯ u · ∇)u = − ∇p + ν∇2 u ∂t ρ 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. v = w = 0.1 Steady flow: ∂ 0 =0 1 ∂t 2Parallel flow: ∂ui i =0 v = 0. 0]. 0.5 2 Exact solutions to Navier-Stokes 1 2 j− j − 1Exercise 1 APlane Couette flow: BConsider the flow of a viscous fluid between two parallel plates at y = 0 and y = h. j 2µ j+1 iteration direction Y x y 1 n x X y i j Z fine grid (1) coarse grid (2) Figure B.

z. n• We can conclude that u Γ0 Γ1 Let’s study the x component of the Navier–Stokes equations: Ω P ∂2u ∂u P P 2 M2 0= +ν 2 ⇒ = − y + d1 {µ = ρν} ⇒ u(y) = − y + d1 y + d 2 M ρ ∂y ∂y νρ 2µ 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 2µ 2µ V Vh P 2 uWe can directly conclude that d1 = 0 and this gives d2 = 2 µ h and thus uh Γ0 P 2 u= (h − y 2 ).1 • The continuity equation ∇ · u ¯ reduces to = 0. 0. The boundary condition v(y = ±h) = 0 then implies v1. Changes of u ¯ in x. Γ1 2µ Ω to be updated Exercise 2 (ii) : new values : old valuesPoiseuille flow: ( Pipe flow ) i − 1Consider the viscous flow of a fluid down a pipe with a cross-section given by r = a under the constant ∂p i pressure gradient P = − . Let’s study the z component of the Navier–Stokes equations: 0 1 ∂w ∂2w v =ν 2 ⇒ w = c1 y + c2 2 ∂y ∂y i =0 j InflowThe boundary conditions w(y = −h) = w(y = h) = 0 imply c1 = c2 = 0 and thus w = 0. v = w = 0.j+ 2 j Boundary conditions: u ¯(y = ±h) = 0 j− j −1 ∂u ¯ • We are considering a stationary flow and thus = 0. C ∂v u 23 . z would require a changing pressure B gradient in x. 0]. 32 ∂y p1.4: Coordinate system for Poiseuille flow Use the Navier–Stokes equations in cylindrical coordinates ∂ur u2 1 ∂p ur 2 ∂uθ + (¯ u · ∇)ur − θ = − + ν ∇2 u r − 2 − 2 ∂t r ρ ∂r r r ∂θ . Solid wall ¯ = [u(y).1 v = 0. 4µ j +1 1 2 ration direction x y 1 n x y i j fine grid (1) coarse grid (2) R Z Figure B. Show that i+1 ∂z j −1 P 2 j uz = (a − r2 ) ur = uθ = 0. A ∂t • The constant pressure gradient implies u ¯ = u ¯ (y).

Considering a steady flow we get from the Navier–Stokes equations (¯ u · ∇)uz = uz But we know that ∂uz ∂z ⇒ uz ∂uz 1 = P + ν∇2 uz ∂z ρ ∂uz = 0 from the continuity equation. where the boundary layer over a flat plate is kept parallel of a steady suction V0 through the plate.∂uθ ur uθ 1 ∂p 2 ∂ur uθ + (¯ u · ∇)uθ + =− + ν ∇2 u θ + 2 − 2 ∂t r ρ r ∂θ r ∂θ r ∂uz 1 ∂p + (¯ u · ∇)uz = − + ν∇2 uz ∂t ρ ∂z 1 ∂ 1 ∂uθ ∂uz (r ur ) + + = 0. We get ∂z ∇2 u z = Integrate r 1 ∂ ∂uz P (r )=− r ∂r ∂r µ ∂uz P = − r 2 + c1 ∂r 2µ ⇒ ∂ ∂uz P (r )=− r ∂r ∂r µ ∂uz P c1 =− r+ ∂r 2µ r ⇒ Integrating again we get uz = − P 2 r + c1 ln(r) + c2 4µ using the boundary conditions uz (r = 0) < ∞ ⇒ c1 = 0 We also have uz = 0 at r = a and this gives c2 = uz = P a2 and we have 4µ P 2 (a − r2 ) 4µ Exercise 3 Calculate the asymptotic suction boundary layer. ∂θ ∂r From the continuity equation we get ∂uz =0 ∂z ⇒ uz = uz (r. θ) only. ∂z Parallel flow: ∂ = 0. ∂x Steady flow: Momentum equations: w=0 Continuity satisfied ∂ =0 ∂t ∂u ∂u ∂u 1 ∂p ∂2u ∂2u +u +v =− +ν + 2 ∂t ∂x ∂y ρ ∂x ∂x2 ∂y ∂v ∂v ∂v 1 ∂p ∂2v ∂2v +u +v =− +ν + ∂t ∂x ∂y ρ ∂y ∂x2 ∂y 2 Normal momentum equation gives ∂p =0 ∂y . Assumptions. Two-dimensional flow: ∂ = 0. r ∂r r ∂θ ∂z ∂p ∂p We know that = 0 and = 0 and can directly see that ur = uθ = 0 satisfy the two first equations.

0] and u ¯2 = [u2 (y). 0]. 0. The continuity equation ∂u ∂v + =0 ∂x ∂y gives ∂v = 0 ⇒ v = c and the boundary condition at y = 0 gives v = 0. λ2 = − V0 ν u(y) = A + Be−V0 y/ν Exercise 4 Two incompressible viscous fluids flow one on top of the other down an inclined plane at an angle α. The lower fluid has depth h1 and the upper h2 . 0. 2 ν1 Make the ansatz u ¯1 = [u1 (y). Assuming that viscous forces from the surrounding air is negligible and that the pressure on the free surface is constant.Boundary conditions: y=0: v = −V0 u = 0. y→∞: u → U∞ Continuity gives ∂u ∂v + =0 ∂x ∂y ⇒ v = −V0 Streamwise momentum equation at y → ∞ ∂U∞ 1 ∂p ∂ 2 U∞ =− +ν ∂y ρ ∂x ∂y 2 −V0 ∂p =0 ∂x ⇒ Resulting streamwise momentum equation −V0 ∂u ∂2u =ν 2 ∂y ∂y ∂2u V ∂u =− 0 ∂y 2 ν ∂y ⇒ Characteristic equation λ2 = − V0 λ ν ⇒ λ1 = 0. show that 1 g sin(α) u1 (y) = [(h1 + h2 )y − y 2 ] . ∂y • Layer 1: N–S · e¯y : 0 = − 1 ∂p1 − g cos(α) ρ ∂y ⇒ p1 = −ρg cos(α)y + f1 (x) 1 d 2 u1 N–S · e¯x : 0 = − f1 (x) + ν1 2 + g sin(α) ρ dy • Layer 2: N–S · e¯y : 0 = − 1 ∂p2 − g cos(α) ρ ∂y ⇒ ⇒ f1 (x) = c1 p2 = −ρg cos(α)y + f2 (x) 1 d 2 u2 N–S · e¯x : 0 = − f2 (x) + ν2 2 + g sin(α) ρ dy ⇒ f2 (x) = c2 The pressure at the free surface y = h1 + h2 is p0 : p0 = −ρg cos(α)(h1 + h2 ) + f2 (x) ⇒ f2 = p0 + ρg(h1 + h2 ) cos(α) ⇒ f2 = 0 The pressure is continuous at y = h1 : p0 + ρgh2 cos(α) = −ρgh1 cos(α) + f1 (x) ⇒ f1 = p0 + ρg(h1 + h2 ) cos(α) This gives the pressure: p1 (y) = p2 (y) = p(y) = −ρ g cos(α)y + p0 + ρ g cos(α)(h1 + h2 ) ⇒ f1 = 0 . They both have the same density ρ and the viscosities µ1 and µ2 .

We now have two momentum equations in x: 0 = ν1 d 2 u1 + g sin(α) dy 2 (1) 0 = ν2 d 2 u2 + g sin(α) dy 2 (2) And four boundary conditions: BC1: No slip on the plane: u1 (0) = 0 BC2: No viscous forces on the free surface: BC3: Force balance at the fluid interface: µ1 µ2 du2 |y=h1 +h2 = 0 dy du1 du2 |y=h1 = µ2 |y=h1 dy dy BC4: Continous velocity at the interface: u1 |y=h1 = u2 |y=h1 (1) ⇒ du1 g = − y sin(α) + c11 dy ν1 ⇒ u1 = − g 2 y sin(α) + c11 y + c12 2 ν1 (2) ⇒ du2 g = − y sin(α) + c21 dy ν2 ⇒ u2 = − g 2 y sin(α) + c21 y + c22 2 ν2 BC1 ⇒ c12 = 0 g g BC2 ⇒ µ2 (− (h1 + h2 ) sin(α) + c21 ) = 0 ⇒ c21 = (h1 + h2 ) sin(α) ν2 ν2 g g µ2 g BC3 ⇒ µ1 (− y sin(α)+c11 ) = µ2 (− y sin(α)+c21 ) {µ = νρ} ⇒ c11 = c21 = (h1 +h2 ) sin(α) ν1 ν2 µ1 ν1 g 2 g g 2 g BC4 ⇒ − h1 sin(α) + (h1 + h2 ) sin(α)h1 = − h1 sin(α) + (h1 + h2 ) sin(α)h1 + c22 2 ν1 ν1 2 ν2 ν2 ⇒ c22 = g sin(α) h21 − (h1 + h2 )h1 2 1 1 − ν2 ν1 This gives us the velocities. 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 ν2 . . u1 (y) = − u2 (y) = − = g 2 g g sin(α) 1 y sin(α) + (h1 + h2 ) sin(α)y = [(h1 + h2 )y − y 2 ] 2 ν1 ν1 ν1 2 g sin(α) 2 g sin(α) h2 y + (h1 + h2 )y + g sin(α) 1 − (h1 + h2 )h1 2 ν2 ν2 2 g sin(α) 1 h2 ((h1 + h2 )y − y 2 ) + g sin(α) 1 − (h1 + h2 )h1 ν2 2 2 1 1 − ν2 ν1 1 1 − ν2 ν1 The velocity in layer 1 does depend on h2 but not on the viscosity in layer 2.

∂u ∂2u =ν 2 ∂t ∂y b) Show that the velocity field is u(y. Further more. Make the ansatz: u = where k = ω 2ν f (y)eiω t = f (y) cos(ω t). a) Show that the velocity field u ¯ = [u(y. 0] satisfies the equation ∂u ∂2u =ν 2 ∂t ∂y Consider the Navier–Stokes equation in the x direction: ∂u ∂u ∂u 1 ∂p ∂2u ∂2u ∂2u ∂u +u +v +w =− +ν + 2 + 2 ∂t ∂x ∂y ∂z ρ ∂x ∂x2 ∂y ∂z With the current velocity field only terms with y derivatives will remain since there can be no change in the other directions.6 More exact solutions to Navier-Stokes Exercise 1 Oscillating Rayleigh–Stokes flow (or Stokes second problem). t). 2ν f (y) = Aeyk(1+i) + Be−yk(1+i) . Insert into the equation: iωf (y)eiω t = νf (y)eiωt f (y) − λ2 − Introduce k = iω f (y) = 0 ν iω =0 ν ⇒ with f (y) = eλy gives iω =± ν 1+i √ 2 λ=± ω ν ω . the streamwise pressure gradient has to be zero since the streamwise velocity far from the wall is constant. 0.B. namely zero. f (y) → 0 as y → ∞ → A = 0 . t) = U e−ky cos(ky − ωt).

5 j Inflow 2 n Solid wall 1.y y ui−1/2.5 j 4 j+1 3.5 .6 −0. t) = U cos(ωt) − U e−ky cos(ky − ωt).4 0.5 j p− 1 1.4 −0.5 j Ω fine grid (1) 2 0 0.5 n Γ0 1 Γ1i 0. If we consider the solution to the previous Γ1 problem and look at it in this new frame of reference we get Ω : to be updated ω u(y.8 −0.5 1 i j 0.6 ⇒ δ ≈ 6.5 1 1. where k = A 2ν : old values B i−1 C u 23 .5 iteration direction x 3 y 2.5 iteration direction x 3 yi 2.6 0.1 j 4 j+1 1 3.5 fine grid (1) 0 coarse grid (2) −1.5 0 x 0. This will simply result in a ∞ Γ0 change of frame of reference to one following the plate instead.5 n 2 1. where k = : new values 2ν : old values i − 1The solution now looks like this i 5 i+1 j − 1 4.2 x M m c) How thick is the boundary layer thickness? m+2 1 hIf we define the thickness δ of the oscillating layer as the position where u/U = 0. 32 4.01 ⇒ kδ ≈ 4. t) = U e−ky cos(ky − ωt).01 we get that V 2ν ν Vh e−kδ = 0.1i 5 iv+ 1 1.j 1 We have i−h 2 V i u= Be−yk(1+i) eiwt = Be−ky ei(ωt−ky) = Be−ky cos(ωt − ky) = Be−ky cos(ky − ωt) i +Vh12 u i+1 i +u32h Boundary condition at y = 0 j +Γ210 Γ1 At y = 0 we have u = U cos(ωt) ⇒ B = U j 1 Ω j−2 So we have : to be updated j−1 ω : new values u(y.6 ⇒ δ ≈ 4.5 ω ω u uh d) Consider instead the oscillating flow U = U cos(ωt) over a stationary wall.5 −1 −0.2 0.8 0 1 coarse grid M (2) −1 −0.

∂ =0 ∂θ No streamwise pressure gradient: ∂p =0 ∂z We can directly see that ur = uθ = 0 satisfy the two first equations. Shear stress τrz = µ ∂uz µU0 = ∂r r ln rr10 Force F = 2πr0 Lτrz (r0 ) = 2πLµU0 ln rr01 .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 . ∂z u = uz (r)ez . The streamwise momentum equation reduces to (u · ∇)uz = ν∇2 uz where (u · ∇)uz = ur ∇2 u z = 1 ∂ ∂uz r r ∂r ∂r ∂uz uθ ∂uz ∂uz + + uz =0 ∂r r ∂θ ∂z + 1 ∂ 2 uz ∂ 2 uz 1 ∂ ∂uz + = r 2 2 r ∂θ ∂z 2 r ∂r ∂r We end up with ∂ ∂uz r ∂r ∂r Integrate twice r ∂uz =A ∂r ⇒ =0 uz = A ln r + B Boundary conditions uz (r0 ) = U0 uz (r1 ) = 0 = A ln r0 + B = A ln r1 + B U0 = A(ln r0 − ln r1 ) uz (r) = ⇒ ⇒ B = −A ln r1 A= ⇒ U0 ln rr01 ln rr1 U0 U0 ln r − ln r = U 1 0 ln rr01 ln rr01 ln rr01 b) With what force does one have to pull a rod with length L? Neglect end effects. a) Find the velocity field of a viscous fluid occupying the space between the rod and the cylinder. Assumptions Steady flow: ∂ =0 ∂t Parallel flow and symmetry: ∂ = 0.

∂θ ∂ (r uθ ) = r ω(r) ∂r (B. The vorticity is given. 1) Compute ur (r. ∂ (rur ) = −rγ ∂r Equation (2) is valid both for case a) and b). z) for the two cases when a) ω(r) = 0 b) ω(r) = ω0 e−r and 2 /a2 . ur = − γr 2 for both a) and b)! r2 γ +B 2 . The mass flow is. z) and uθ (r.1) We also have the incompressibility condition: ∇·u ¯= 1 ∂ 1 ∂uθ ∂uz + =0 (rur ) + r ∂r r ∂θ ∂z γ =0 From this we get.7 Axisymmetric flow and irrotational vortices Axisymmetric flow Consider incompressible and rotationally symmetric flow with no mass source at the symmetry axis. From the e¯z component we get.B.2) uθ = A r integrate the rhs rur = − γr B + 2 r There should be no mass source at r = 0. uz = γ z and ω ¯ = ω(r) e¯z . ω ¯ =∇×u ¯= 1 r e¯r ∂ ∂r ur r¯ eθ e¯z ∂ ∂θ r uθ ∂ ∂z uz Look at the different components: 1 ∂uz ∂uθ − =ω ¯ · e¯r = 0 r ∂θ ∂z e¯r : e¯θ : ⇒ uθ = uθ (r) ∂ur ∂uz − =ω ¯ · e¯θ = 0 ⇒ ur = ur (r) ∂z ∂r 1 ∂(r uθ ) ∂ur e¯z : − =ω ¯ · e¯z = ω(r) r ∂r ∂θ The rotational symmetry implies ∂ = 0. ⇒ ur = − 2π Q= 0 ur r dθ = 2π B − 2π γ r2 . Case a) ω(r) = 0 ∂ (r uθ ) = 0 ∂r (1) ⇒ (2) ⇒ ∂ (rur ) = −rγ ∂r ⇒ ⇒ (B. 2 When r → 0 then Q → 2π B which gives B = 0. The velocity component in the direction of the axis of symmetry and the vorticity is given.

Can any of the flow cases be inviscid? R(0) = R0 1 dR = ur (R) = − γR dt 2 dZ = uz (Z) = γZ dt Z(0) = Z0 1 ⇒ R(t) = R0 e− 2 γ t ⇒ Z(t) = Z0 eγ t The circulation is then. 2π Γ(t) = uθ (R)R dθ = 2πRuθ (R) = 0  Γ a) Γ0 + πω0 a2 1 − e−R 2 /a2 b) The circulation is constant for a) but not for b). ⇒ uθ = − 2π Γ(r) = uθ r dθ = 2π(− 0 ⇒ for r = 0 we get This gives uθ = Γ0 = 2π(− ω0 a2 −r2 /a2 C e + 2r r ω0 a2 −r2 /a2 e + C) 2 ω0 a 2 + C) 2 ⇒ C= Γ0 ω a2 + 0 2π 2 2 2 Γ0 ω a2 ++ 0 1 − e−r /a 2πr 2r c) Consider circles C(t) following the flow with a radius R(t) and position Z(t). This means that the flow in a) can be inviscid. Compute R(t) and Z(t).The circulation Γ is. (1) ⇒ 2 2 ∂ (r uθ ) = r ω0 e−r /a ∂r Integrate the right hand side: r uθ = − ω0 a2 −r2 /a2 e +C 2 Look at the circulation. . 2π Γ= 0 Case b) ω(r) = ω0 e−r 2 /a 2 uθ r dθ = 2πA ⇒ A = Γ 2π .

Inviscid and irrotational vortices Consider a circular flow with u ¯ = uθ (r)¯ eθ . 1 ∂ 1 (rnrn−1 ) − rn−2 = n2 rn−1 − rn−2 r ∂r r → We get the inviscid flow. ω ¯ =∇×u ¯ = {A. uθ (r) = Ar solid body rotation The vorticity is. 1 ∂ ∂uθ r r ∂r ∂r − uθ =0 r2 Make the ansatz uθ = rn . Which vortices are inviscid and which vortices are irrotational? The Navier–Stokes equation for uθ ∂uθ 1 1 ∂ ∂uθ =− +ν r ∂t ρr r ∂r ∂r − uθ r2 For inviscid flow we require.32} = ω ¯ =0 ⇒ ∂ (ruθ ) = 0 ∂r n2 − 1 = 0 B r + irrotational 1 ∂ (ruθ ) e¯z r ∂r ⇒ Conclusion: Irrotational ⇒ inviscid Inviscid irrotational uθ = C r ⇒ n = ±1 .

z. t)¯ er + uz (r.Exercise 3 Show that the inviscid vorticity equation Dω ¯ = (¯ ω · ∇)¯ u Dt reduces to the equation D ω Dt r =0 in the case of axisymmetric flow: ¯ = ur (r. t)¯ ez The vorticity in an axisymmetric flow ¯= ω ¯ =∇×u ∂ur ∂uz − ∂z ∂r ¯θ = ω¯ e eθ ω Study the right hand side of the inviscid vorticity equation (¯ ω · ∇) = ω ∂ r ∂θ ⇒ (¯ ω · ∇)¯ u= ω ∂ ur (r. t)¯ u er + uz (r. z. z. t)¯ ez r ∂θ = ω ∂ur ω ∂¯ er ω ∂uz ω ∂¯ ez ω ¯ r + ur ¯ z + uz ¯θ e e = ur e + r ∂θ r ∂θ r ∂θ r ∂θ r =¯ eθ =0 =0 =0 The left hand side of the inviscid vorticity equation gives Dω ¯ ∂ω ¯ = + (¯ u · ∇)¯ ω = {¯ ω = ω¯ eθ } = Dt ∂t ∂ω ∂ ∂ ¯θ + ur + uz ω e ∂t ∂r ∂z This gives that the inviscid vorticity equation now is ∂ ∂ ω ∂ω + ur + uz ω = ur ∂t ∂r ∂z r Multiply by 1 r ∂ ω ∂t r Notice that − + 1 ∂ ∂ ω ur + uz ω − 2 ur = 0 r ∂r ∂z r ω ∂ 1 ur = ωur 2 r ∂r r and that ωuz ∂ 1 =0 ∂z r This means we can write ∂ ω ∂t r And thus we have + 1 ∂ ∂ ∂ ∂ ur + uz ω + ω ur + uz r ∂r ∂z ∂r ∂z ∂ ω ∂t r + 1 ∂ ∂ ur + uz r ∂r ∂z ω D ω = r Dt r 1 =0 r =0 . z.

Calculate the atmospheric pressure at the top of the hill. Bernoulli equation and streamfunction Solid body rotation Consider the flow in a uniformly rotating bucket with velocity u = (ωr. 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 . ¯ = (0. If the flow had been irrotational. where f (r) is an arbitrary function. Bernoulli theorem is valid along a streamline in a steady ideal fluid. But now ∇ × u b) Use Euler equation to determine the free surface: 1 Du = − ∇p + g Dt ρ this gives er : − u2θ 1 ∂p =− r ρ ∂r eθ : ez : 0 = − ⇒ ∂p = ρω 2 r ∂r ⇒ 0=− 1 ∂p ρr ∂θ ∂p =0 ∂θ 1 ∂p −g ρ ∂z ⇒ Thus p(r. 0. c) (2) Assume that the density ρ and the gravitational acceleration g is constant. 0) a) Use Bernoulli equation to determine the free surface: What is wrong? Bernoulli: p 1 2 + |u| + gz = C along streamlines ρ 2 This gives the surface of constant pressure C − p0 ω 2 r2 − ρg 2g z= The free surface is highest in the center of the bucket. 2ω). b) (3) Derive an equation for the curve with constant vertical wind velocity V . z) = ⇒ ∂f = −ρg ∂z ⇒ p= 1 2 2 ρω r + f (z) 2 f = ρgz + p0 1 2 2 ρω r − ρgz + p0 2 This gives the free surface where p = p0 z= ω 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. it would have been valid everywhere.8 Vorticity equation. a) (5) Assume potential flow and show that the stream function in cylindrical coordinates is of the form ψ = f (r) sin θ.B. a) The stream function satisfies continuity: ur = The flow is irrotational: ∇×u ¯=0 ⇒ 1 ∂ψ . 0. something is wrong. r ∂θ uθ = − ∂ψ ∂r ∂ψ ∂2ψ 1 ∂2ψ +r 2 + = 0 (1) ∂r ∂r r ∂θ2 . Calculate the velocity field above the hill.

1. Free stream: ⇒ Ar sin θ = U∞ r sin θ A = U∞ 2.5: Streamlines above a hill with h = 100m and U∞ = 5m/s. Introduce the ansatz ψ = f (r) sin θ into equation (1): 1 f sin θ + rf sin θ − f sin θ = 0 r ⇒ 1 f + rf − f = 0 r Make the ansatz f = r n : 1 nrn−1 + rn(n − 1)rn−2 − rn = 0 r n + n2 − n − 1 = 0 So we have ψ= Ar + ⇒ B r ⇒ n = ±1 sin θ We need two boundary conditions. while soaring along the hill. Streamline on the hill surface: U∞ h + B =0 h B = −U∞ h2 ⇒ So we have: ψ = U∞ r − h2 r sin θ Now we can calculate the velocity field above the hill: ur = 1 ∂ψ h2 = U∞ 1 − 2 r ∂θ r uθ = − cos θ ∂ψ h2 = −U∞ 1 + 2 ∂r r sin θ b) Constant vertical wind velocity is described by: V = ur sin θ + uθ cos θ = U∞ 1 − h2 r2 cos θ sin θ − U∞ 1 + r=h −2 h2 r2 sin θ cos θ = −2U∞ U∞ sin θ cos θ V c) Use Bernoulli equation with free stream pressure p0 at the ground: 1 2 1 po + ρU∞ = p + ρ(2U∞ )2 + ρgh 2 2 3 2 ⇒ p = po − ρU∞ − ρgh 2 h2 sin θ cos θ r2 ⇒ .ration direction x y 1 n x y i j fine grid (1) coarse grid (2) y x Figure B. A paraglider pilot with a sink of 1m/s will find lift in the area within the dotted line.

∂x ∂y ∂x∂y ∂y∂x so continuity is fulfilled. − ∂Ψ ∂2Ψ − 2 ∂x ∂y 2 Irrotational flow ∇2 Ψ = 0 In spherical coordinates for axisymmetrical flow. Now we can write ¯x e u ¯ = ∇ × Ψ¯ ez = And ω ¯ =∇×u ¯= ¯y e ∂ ∂x ∂ ∂y 0 0 ¯x e ¯y e ¯z e ∂ ∂x ∂Ψ ∂y ∂ ∂y − ∂Ψ ∂x ∂ ∂z = 0 ¯z e ∂ ∂z ∂Ψ ∂Ψ . ∂x ∂y Define the stream function Ψ such that: ∂Ψ ∂y u= This means v=− ∂Ψ ∂x ∂u ∂v ∂ 2Ψ ∂2Ψ + = − = 0.Stokes stream function: Consider a 2D incompressible flow ∇·u ¯ = 0 or ∂u ∂v + = 0.− .0 ∂y ∂x = Ψ 0. define Ψ ur = r2 1 ∂Ψ sin θ ∂θ uθ = − 1 ∂Ψ r sin θ ∂r Incompressibility is still valid ∇·u ¯=0 Velocity u ¯=∇× Vorticity ω ¯=− Irrotational Ψ ¯θ e r sin θ 1 ∂ 2 Ψ sin θ ∂ + 2 r sin θ ∂r2 r ∂θ ∂ 2 Ψ sin θ ∂ + 2 ∂r2 r ∂θ 1 ∂Ψ sin θ ∂θ 1 ∂Ψ sin θ ∂θ =0 = −∇2 Ψ¯ ez . 0.

compute the irrotational velocity distribution when the velocity of the freestream at infinity is U: 1 r → ∞ Ψ → U r2 sin2 θ 2 ⇒ ur → U cos θ uθ → −U sin θ Make the ansatz: Ψ = f (r) sin2 θ For an irrotational flow we get f − 2 f =0 r2 ⇒ From infinity we get A= On the sphere 1 2 B Ua + =0 2 a This gives Ψ= The slip velocity on the sphere is The radial velocity is ur = 0.Flow around a sphere Consider a sphere with Ψ = 0 at r = a. f = Ar 2 + B r 1 U 2 ⇒ 1 a3 U r2 − 2 r 1 B = − U a3 2 sin2 θ 3 uθ = − U sin θ 2 .

θ. z) and the two last are in two different spherical coordinate systems with origin in z−a and z+a. ∂x u= v= ∂φ ∂y The flow is always irrotational due to the definition of the velocity potential ω =∇×u ¯ = ∇ × ∇φ = 0.9 Flow around a submarine and other potential flow problems : to be updated : new valuesExercise 1 : old values i − 1The flow around a submarine moving at a velocity V can be described by the flow caused by a source and a sink with strength Q at a distance 2a from each other. respectively.1 z n y y i j fine grid (1) coarse grid (2) a L Figure B. Transform the two second terms to cylindrical coordinates φ = Uz + 4π Velocity + Q 4π (z − a)2 + R2 ∂φ 1 ∂φ ∂φ e¯R + e¯θ + e¯z = ∂R R ∂θ ∂z QR QR Q(z + a) Q(z − a) − +¯ ez U + − 4π((z + a)2 + R2 )3/2 4π((z − a)2 + R2 )3/2 4π((z + a)2 + R2 )3/2 4π((z − a)2 + R2 )3/2 u ¯ = ∇φ = e¯R −Q (z + a)2 + R2 . L = 200 m and ρ = 1000 kg/m3 . ∂φ . Q = 915 m3 /s. We have freestream plus 3D source plus sink φ= Uz freestream + Q −Q + 4π r1 4π r2 source sink The first term is in cylindrical coordinates (R. Use a potential flow description u ¯ = ∇φ. i i+1 x V j−1 j j+1 Submarine iteration direction x y Q -Q 1 p1. what sensitivity is needed for the sensor? Assume an ideal fluid and that 2a = 80 m. curl(grad)=0 For incompressibility we get ∇·u ¯ = ∇ · ∇φ = ∂(∇φ)i = ∆φ = 0 ∂xi The equation is linear and thus superposition can be used.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.ΩB. U = 8 m/s.

The stream function is constant along streamlines and is useful for this.01 m.86 N/m ≈ 0. In spherical coordinates the stream function is defined as ur = 1 ∂ψ r2 sin θ ∂θ 1 ∂ψ r sin θ ∂r Transformation between cylindrical and spherical coordinates uθ = − ur = uR sin θ + uz cos θ R = r sin θ.99914 m/s and we get p − p∞ = 1 2 ρ(U 2 − |¯ u|2 ) ≈ 6. where the second solution lies inside the submarine. R = 0) = e¯z U + Q −Q + 4π(L + b)2 4π((L + b + 2a)2 Inserting the given values gives |¯ u| = 7. b) How long is the submarine? Compute where uz = 0 for R = 0 U+ − Q 4π 1 1 − (z + a)2 (z − a)2 =0 4πU (z − a)2 − (z + a)2 −4az = = 2 Q (z − a)2 (z + a)2 (z − a2 )2 Solving this system gives z = ±43. The length is then 2 ∗ 43. b. Q 4π r cos θ + a r cos θ − a − 2 (r2 + a2 + 2ar cos θ)3/2 (r + a2 − 2ar cos θ)3/2 (r2 r + a cos θ r − a cos θ − 2 2 3/2 + a + 2ar cos θ) (r + a2 − 2ar cos θ)3/2 . z = ±36.99 m.07 mbar 2 c) How wide is the submarine? To get this we need to compute the shape of the submarine. z = r cos θ Our velocity field gives 1 ∂ψ Q 1 1 = sin θ r sin θ − 2 + r2 sin θ ∂θ 4π (r2 + a2 + 2ar cos θ)3/2 (r + a2 − 2ar cos θ)3/2 cos θ U + Q 4π = U cos θ + This is difficult to integrate. R = 0 1 1 p + ρ|¯ u|2 = p∞ + ρU 2 2 2 Evaluate u ¯ noticing that uR = 0 u ¯(z = −L − b − a. from the point source to the stagnation point on the submarine nose.We need to know the distance. Now we continue to solve a) Use Bernoulli equation to determine the pressure fluctuations at z = −L − b − a.01 ≈ 86 m and b = 3. Thus we need to know the length of the submarine.01 m.

1 Simplify to a Rankine body by neglecting the sink and say that a = 0 Q 1 1 ∂ψ 1 ur = U cos θ + = 2 ⇒ 2 4π r r sin θ ∂θ 2 i 1 Q j Ψ = U r2 sin2 θ − cos θ + C 2 4π Inflow Solid wallDetermine C from the stagnation point n Q Γ0 ur (θ = π. As r → ∞.07 m Uπ . Since no fluid can cross the streamlines this area must be equal to that of the Rankine body: Q = Uπ d2 2 ⇒ d=2 Q = 12.7: Rankine body for submarine problem Q 1 d2 U −2 =0 2 4 4π ⇒ d2 = 4Q Uπ ⇒ d=2 Q Uπ There is a simple way of determining the radius as z → ∞ directly. This gives 2 u uh Γ0 Γ1 Ω : to be updated 10 : new values : old values i−1 5 i i+1 j−1 j 0 j+1 iteration direction x y −5 1 n x y −10 i j fine grid (1) −15 −15 −10 −5 5 10 15 0 coarse grid (2) −20 x z Figure B. θ → 0 then r sin θ → . r0 ) = 0 ⇒ r02 = 4πU Γ1 ΩSince Ψ = 0 on the body we get Q M2 C =− . The flow from the source must take up an particular area in the flow at infinity. 4π M mThe stream function is then 1 Q m+2 Ψ = U R2 sin2 θ − (cos θ +1) 2 4π 1 h source V d Vh The shape is given by Ψ = 0.p1.

uh We can use the computed stream function for a point source and displace it to z = −a. In cylindrical Γ0 coordinates Γ1 −Q z+a Ψ= Ω 4π (z − a)2 + R2 to be updated : new valuesTransform to spherical coordinates : old values −Q −Q r cos θ + a r cos θ + a i−1 √ Ψ= = 2 + a2 + 2ar cos θ 4π 4π 2 r (r cos θ + a)2 + r2 sin θ i i+1 j − 1The stream function for the submarine is then j r cos θ − a 1 Q r cos θ + a j +1 −√ +√ Ψ = U r2 sin2 θ + 2 2 2 2 4π r + a + 2ar cos θ r + a2 − 2ar cos θ ration direction x y 1 n x y i j fine grid (1) coarse grid (2) x 10 0 −10 −80 −60 −40 −20 0 20 40 60 80 z Figure B.0006 m .8: Submarine body for submarine problem At r = R and θ = π/2 we get Ψ= 1 Q a −a U R2 + −√ +√ 2 2 2 4π R +a R 2 + a2 For the body Ψ = 0 and we get R2 R 2 + a2 − aQ =0 πU R2 R 2 + a2 + aQ πU Multiply by (R2 )3 + (R2 )2 a2 − ⇒ a2 Q2 =0 π2 U 2 Computing this d = 2R = 12.

in particular conformal mapping that can be used to compute the flow over airfoil shapes. Since both of them satisfy Laplace’s equation we can Vh define a complex function F (z) = φ(x. y) z = x + iy u uh Γ0 Γ1 Ω : to be updated 18 : new values : old values 16 i−1 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 fine grid (1) 0 −45 −40 −35 −30 −25 coarse grid (2) −50 Figure B. They are orthogonal to the lines of constant V velocity potential φ which are equipotential lines. dotted: φ This is an analytical function since the Cauchy–Riemann equation holds ∂φ ∂Ψ = ∂x ∂y The velocity is then and ∂φ ∂Ψ =− ∂y ∂x dF ∂φ ∂Ψ = +i = u − iv dz ∂x ∂x This enables the use of complex analysis.9: Complex potential for submarine problem.m The complex potential m+2 1 hThe lines with constant stream function Ψ are the streamlines. w(z) = . solid: Ψ . y) + i Ψ(x.

λ = 0.10: Conformal mapping from circle to airfoil shape. A correct flow is not 2 4 Ω to be updatedachieved unless the Kutta–Joukovski condition is satisfied requiring : new values Γ = −4πU (a + λ) sin α : old values i−1 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 fine grid (1) coarse grid (2) −1 −2 −2 −5 0 5 −3 −4 −4 −2 0 2 −1 z=f (Z) Figure B.5) .m Example: m+2 1 h Flow past a rotating cylinder centered at z = λ at an angle of attack α V Vh (a + λ)2 iα iΓ F (z) = U (z + λ)e−iα + e − log(z + λ) u (z + λ) 2π uh 1 Γ0 2 Γ1 Mapping by z = 1 Z + 1 Z 2 − a2 gives an airfoil shape with the potential F(z). (a = 3.

A uniform stream with velocity U flows along the x-axis. Put a line source of equal strength at (0. Method of images. Calculate the irrotational flow field. a) above a flat plate that coincides with the x-axis. −a) in order to fulfill the condition of no flow through the plate.B.10 More potential flow Half body over a wall A line source of strength Q is located at (0. Superposition of a uniform flow and the two line sources gives the complex potential F = Uz + Q Q ln(z − ia) + ln(z + ia) 2π 2π Complex velocity W = dF Q =U+ dz 2π W =U+ W =U+ W =U+ Q 2π 1 1 + z − ia z + ia ⇒ 1 1 + x + i(y − a) x + i(y + a) x − i(y − a) x − i(y + a) + 2 2 2 x + (y − a) x + (y + a)2 Q 2π ⇒ ⇒ Q x x y−a y+a + 2 −i 2 + 2 2 2 2 2 2π x + (y − a) x + (y + a) x + (y − a) x + (y + a)2 The velocity field now becomes u=U+ v= Q 2π Q 2π x x + 2 x2 + (y − a)2 x + (y + a)2 y−a y+a + 2 x2 + (y − a)2 x + (y + a)2 .

5 i j −1 −2 fine grid (1) −3 −2 −1 2 1 −1 −0.Γ0 Γ1 Flow past a symmetric airfoil Ωa) Use conformal mapping to calculate the irrotational flow field around a symmetric airfoil. to be updated : new values Joukowski transformation : old values c2 ζ(z) = z + i−1 z i i+1 j −1 j 1 2 j +1 ration direction 1 x 0. b) Calculate the Joukowski condition for the airfoil.5 Equation for the circle z = −λ + (a + λ)eiθ Equation for the airfoil ζ = −λ + (a + λ)eiθ + Complex potential in the z-plane F = U (z + λ)e−iα + U a2 −λ + (a + λ)eiθ (a + λ)2 iα iΓ e + ln(z + λ) (z + λ) 2π Complex velocity in the z-plane W = (a + λ)2 iα iΓ dF = U e−iα − U e + 2 dz (z + λ) 2π(z + λ) Complex velocity in the ζ-plane ω= dF/dz = dζ/dz U e−iα − U (a + λ)2 iα iΓ e + 2 (z + λ) 2π(z + λ) 1− a2 z2 The velocity can then be found by introducing the reversed transformation z = ζ/2 + ζ 2 /2 − a2 into v∗ = −Im{ω(z)} u∗ = Re{ω(z)}. The flow field has a singular point at the trailing edge of the airfoil at ζ = 2a.5 y 0 0 n x −1 y −0. Resolve the singularity by choosing the circulation Γ so the numerator vanishes at the trailing edge z = a U e−iα − U eiα + Γk = 4π(a + λ)U iΓ =0 2π(a + λ) ⇒ eiα − e−iα = 4π(a + λ)U sin α 2i .5 0.5 0 1 0 coarse grid (2) −1.

Assume the free stream: U (x) = − Q .11 Boundary layers Exercise 1 Consider the high Reynolds number flow in a conversing channel. Compute the boundary layer over the surface at y = 0.B. x where Q is the flux. ∂y u= v=− Make the stream function dimensionless: f (η) = ∂ψ ∂x ψ Uδ Determine the terms in equation (1): u= ∂f ∂η 1 Q ∂ψ = Uδ = U δf = U f = − f ∂y ∂η ∂y δ x ∂ψ ∂(U δf ) ∂f ∂η ∂δ ∂U =− = −U δ −U f− δf = ∂x ∂x ∂η ∂x ∂x ∂x √ Q ν η Q ν Q ν Qν x f − + f − 2x f =− ηf x Q x x Q x Q x v=− ∂u ∂ = ∂x ∂x − Q f x =− ∂u ∂ = ∂y ∂y − Q ∂f ∂η Q Q + 2f = − f x ∂η ∂x x x Q f x =− ∂2u Q =− 2 ∂y 2 x − η Q Q + 2 f = 2 (ηf + f ) x x x Q ∂f ∂η Q 1 Q =− f =− 2 x ∂η ∂y x δ x Q f ν Q ∂f ∂η Q2 =− 3 f ν ∂η ∂y x ν Insert into equation (1): − √ Q Q f (ηf + f ) + x x2 Qν Q ηf 2 x x Q Q2 Q2 f =− 3 −ν 3 f ν x x ν Divide by Q2 /x3 : −ηf f − f 2 f + ηf f = −1 − f −f 2 +1=0 Boundary conditions: u(0) = 0 u(∞) = 1 ⇒ f (0) = 0 f (∞) = 1 ⇒ . The boundary layer equations read: u ∂u ∂u Q ∂2u +v =− 3 +ν 2 ∂x ∂y x ∂y (1) ∂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 η= = = = xν ν δ x ν x |U | Q The stream function satisfies (2): ∂ψ .

The length scale in x is L and in y it is δ with δ << L. Also assume that the wake is so small that we can write u = U + u1 where u1 is negative and : old valuesdescribes the wake. u1 ∂u1 u2 ∼ 1 ∂x L and v ∂u1 δ 1 u2 ∼ u1 u1 ∼ 1 ∂y L δ L .M2 Substitute F (η) = f (η):  M 2  F − F + 1 = 0 m F (0) = 0 m+2   1 F (∞) = 1 h V Solution: η 2 Vh F = 3 tanh2 √ + −2 3 2 u uh Γ0 Exercise 2 Γ1 ΩConsider the flow downstream of a 2D streamlined body at high Reynolds number. ∂v ∂u ∂ =− =− U + u1 ∂y ∂x ∂x =− ∂u1 ∂x ⇒ v∼ δ u1 << U L Inserting u = U + u1 in the boundary layer equations gives.11: Coordinate system for wake problem a) Find the governing (linear) equations: We start with the boundary layer equations since δ << L. i−1 i i+1 j −1 j j +1 ration direction x y U Y U+u 1 X n x y i j fine grid (1) coarse grid (2) Figure B. ∂p = 0. U ∂u1 ∂u1 ∂u1 ∂ 2 u1 + u1 +v =0+ν ∂x ∂x ∂y ∂y 2 Neglect the quadratic terms. 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. ∂x From the continuity equation we get. u ∂u 1 ∂p ∂2u ∂u +v =− +ν 2 ∂x ∂y ρ ∂x ∂y ∂u ∂v + =0 ∂x ∂y In the wake we have no pressure gradient.

d d Q1 = dx dx ∞ u1 dy = −∞ ∞ −∞ ∂u1 dy = {From the governing equation} = ∂x ∞ −∞ ν ∂ 2 u1 ν ∂u1 dy = U ∂y 2 U ∂y This means that Q1 is constant. u1 (x. c) Find a similarity solution for u1 : Seek a solution on the form. −U g (x) ηf (η) + f (η) g(x)2 =ν 1 f (η) g(x)3 U f (η) g (x)g(x) = − =C ν ηf (η) + f (η) g(x) = 2C νx U ⇒ ⇒ 1/2 This gives the equation for f . ∞ Q1 = u1 dy −∞ which gives the linear contribution from u1 to the momentum flux. u1 (x. ∞ F (x) = 1/g(x) and f (η) dη = Q1 = constant −∞ This means.⇒ b) Show that ∞ −∞ ∂u1 ∂ 2 u1 =ν ∂x ∂y 2 U u1 dy = constant: Consider the relation. 1 f (η) = ae− 2 Cη 2 We can determine the constant a from the condition that Q1 is constant. This gives. y) = y2 C Q1 − 12 C g(x) 2 e = 2π g(x) ∞ −∞ =0 . f (η) + Cηf (η) = 0 Integrating this we get. Q1 = ∞ where η = F (x)f (η) dy = F (x)g(x) −∞ y g(x) ∞ f (η) dη −∞ Q1 = constant requires. y) = F (x)f (η) This gives. y) = 1 y 1 f (η) = f g(x) g(x) g(x) Insert this into the equation of motion. 0) d f (η) + Cηf (η) + D dη = 0 we can determine that f (0) = −D = 0. ∞ =0 f (η) dη = Q1 −∞ ⇒ f (η) = 2 1 C Q1 e− 2 Cη 2π This now gives u1 : u1 (x. f (η) + Cηf (η) + Cf (η) = From the symmetry condition ∂ ∂y u(x.

L = 0.5 10−5 m2 /s. A wing profile in salt water with L = 2 cm and U = 5 cm/s iii. A spermatozoan with tail length of 10−3 cm swimming at 10−1 cm/s in water.05 m/s. L = 4 m ⇒ Re = UL ≈ 4 × 107 ν ii. ν = 1. U = 10−4 m/s.12 m2/s.g(x) = 2C 1/2 νx U Q1 = √ 2 π U νx 1/2 U y2 e− 4νx d) Relate u1 to the drag FD : The drag is FD = −ρ ∞ u1 (U + u1 ) dy −∞ Remember that u1 << U and neglect the u21 term.01 m ⇒ Re = UL ≈ 0.02 m ⇒ Re = UL ≈ 103 ν iii.003 ν iv. ν = 0. ν = 10−6 m2 /s. U = 0. A thick layer of golden syrup draining of a spoon. Flow past the wing of a jumbo jet at 150 m/s (≈ Mach 0. Fluid Water Air Syrup ν cm2 /s 0.15 1200 i. L = 0. A spermatozoan with tail length of 10−3 cm swimming at 10−2 cm/s in water. iv. A wing profile in salt water with L = 2 cm and U = 5 cm/s U = 0. ν = 10−6 m2 /s.01 0. y) = − FD √ 2µ π ν Ux 1/2 U y2 e− 4νx Exercise 3 Give an order of magnitude estimate of the Reynolds number for: i.5) ii. Flow past the wing of a jumbo jet at 150 m/s (≈ Mach 0. Estimate the boundary layer thickness in case (i). L = 10−5 m ⇒ Re = The boundary layer thickness in (i) is O(1 mm) δ 1 =O √ L Re UL ≈ 10−3 ν .5) U = 150 m/s. A thick layer of golden syrup draining of a spoon.04 m/s. FD = −ρ ∞ −∞ U u1 dy = −ρU Q1 This means that Q1 = − FD ρU ⇒ u1 (x.

l/2 −U0 l + 2Lv|y=l/2 + Uw (y) dy = 0 l/2 The momentum equation in x: S ρnj uj ux + nx p − nj τxj dS = 0 Far away from the cylinder we can assume p = p0 and τxx = τxy = 0. l/2 −ρU02 l + 2ρU0 v|y=l/2 L + −l/2 ρUw2 (y) dy + cyl [nx p − nj τxj ] dS FD Multiply the continuity equation by ρU0 . Consider a fix volume in a fluid with velocity u ¯.m+2 More boundary layers 1B. Ω : to be updated d ρui dV = − ρuj nj ui + pni − τij nj dS + ρFi dV : new values dt V S V : old values i − 1Use this to compute the drag force on a cylinder: i U n ωU≈ 0 0 i+1 j−1 j j+1 iteration direction x l n d Wake y n n 1 x y i j fine grid (1) coarse grid (2) n L The cylinder is stationary and so is the flow ⇒ continuity equation gives: ∂u ¯ = 0. This gives. l/2 −ρU02 l + 2ρU0 v|y=l/2 L = −ρU0 Uw (y) dy −l/2 Insert into momentum equation: l/2 −ρU0 l/2 FD = ρ −l/2 l/2 Uw (y) dy + −l/2 −l/2 ρUw2 (y) dy + FD = 0 U0 Uw (y) − Uw2 (y) dy = ρU02 l/2 −l/2 Uw (y) − U0 Uw (y) U0 2 dy . The ucontinuity equation becomes. The ∂t l/2 ⇒ uk nk dS = 0 S −U0 l − Lv|y=−l/2 + Lv|y=l/2 + Uw (y) dy = 0 l/2 Due to the symmetry we get. Incompressible flow with constant density. uh d ρ dV = − ρuk nk dS. Γ0 dt V S Γ1 The momentum equation.12 h V Drag on a circular cylinder Vh Use the integral form of the Navier–Stokes equations.

Make the variable substitution r = y/d. ∂y v=− ∂ψ ∂x Make the stream function dimensionless: f (η) = ψ Um (x)g(x) Determine the terms in equation (1): u= v=− ∂ψ ∂f ∂η 1 = Um g = Um gf = Um f ∂y ∂η ∂y g ∂ψ ∂(Um gf ) =− = −Um gf − Um g f − Um gf ∂x ∂x − yg g2 = −Um gf − Um g f + Um g f η ∂u ∂ g = (Um f ) = Um f − Um f η ∂x ∂x g ∂u ∂ 1 = (Um f ) = Um f ∂y ∂y g ∂2u ∂ = ∂y 2 ∂y Um f 1 g = Um f 1 g2 Insert into equation (1): g 1 1 ηf ) + (Um g ηf − Um gf − Um g f )Um f = νUm 2 f g g g ⇒ g 1 2 g 2 g ηf f + Um ηf f − Um Um f f − Um f f = νUm 2 f g g g g ⇒ Um f (Um f − Um Um Um f 2 2 − Um f + Um g 2 Um g g + ν ν ff − Um g 2 f ν β α f + αf f + βf 2 =0 2 =0 ⇒ . where Um (x) is the jet core velocity. The boundary layer equations read: u ∂u ∂u ∂2u +v =ν 2 ∂x ∂y ∂y (1) ∂u ∂v + = 0 (2) ∂x ∂y Seek a similarity solution where the dimensionless velocity u/Um(x). only depend on the dimensionless wall distance y η= g(x) The stream function satisfies (2): u= ∂ψ . ∞ FD = ρU02 d Uw (r) − U0 −∞ 2 Uw (r) U0 dr 2 Uw (r) U0 dr = ρU02 θ θ The momentum thickness is called θ and is a measure of the loss of momentum in the flow. l/2d FD = ρU02 d −l/2d Uw (r) − U0 Now if d << l we get. Now compute FD θ the drag coefficient CD = 1 2 = 2 d ρU d 0 2 Glauert wall jet Compute an ODE for a self-similar wall jet.

=0 ∞ g(η) = f 2 dη > 0 η Multiply with f : f f + ff 2 f f + (f f 2 + (β − 1)gf = 0 + gf ) + (β − 2)gf = 0 Integrate: 1 f 2 2 ⇒ ∞ + f g + (β − 2) gf dη = 0 η Put in η = 0 and use the boundary conditions: (β − 2) β=− Solve for m and n: ∞ ⇒ gf dη = 0 0 m =2 m+n 1 m=− . ν ν 2n + m = 1 1 1 ab2 amxm−1 b2 x2n + axm bnxn−1 bxn = (m + n) ν ν ν We can choose the length scale g such that α = 1: α= ab2 1 = ν m+n ⇒ β=− m m+n So we have: f + f f + βf 2 = 0 (3) For what β is the boundary conditions fulfilled? Work on (3): f f + f f + βf 2 =0 ⇒ + (f f + f 2 ) + (β − 1)f 2 Integrate: f + f f + (β − 1)g = 0. Then we get g = bxn 1 ab2 β = − amxm−1 b2 x2n = − m.Boundary conditions:   u(0) = 0 v(0) = 0   u(∞) = 0   f (0) = 0 f (0) = 0   f (∞) = 1 ⇒ Similarity requires that α and β are constant. Assume the x-dependence: Um = axm . 2 ⇒ β=2 ⇒ 3m + 2n = 0 n= 3 4 .

y) = U0 − U (x. y) behind a cylinder. y). u. ∂ ∂V ∂U1 = − (U0 − U1 ) = ∂y ∂x ∂x y∼l ⇒ V ∼ Us ∼ ULs ∼ Vl The turbulent boundary layer momentum equation read: U ∂U ∂U 1 ∂P0 ∂2U ∂ +V =− +ν 2 − (uv) ∂x ∂y ρ ∂x ∂y ∂y ⇒ Disregard the pressure gradient and insert U = U0 − U1 : (U0 − U1 ) − ∂U1 ∂x Neglect small terms: U0 −V ∂U1 ∂ 2 U1 ∂ = −ν − (uv) ∂y ∂y 2 ∂y ∂U1 ∂ = (uv) (1) ∂x ∂y Self-similar hypothesis: U1 (x. x ∼ L. y) = f (η) Us (x) −uv = g(η) Us2 where η= y l(x) ⇒ ∂η y η = − 2l = − l . Let U1 (x. y = 0) Assume Us Continuity U0 ∂U ∂V + =0 ∂x ∂y ⇒ ⇒ U1 U0 . v ∼ Us .13 Introduction to turbulence Exercise 1 Compute the far-field two-dimensional turbulent wake U (x. ∂x l l ∂η 1 = ∂y l Determine the terms in equation (1): ∂ ∂η l ∂U1 = (Us f ) = Us f + Us f = Us f − Us ηf ∂x ∂x ∂x l ∂η 1 ∂ ∂ (uv) = (−Us2 g) = −Us2 g = −Us2 g ∂y ∂y ∂y l Insert into equation (1): l 1 U0 (Us f − Us ηf ) = −Us2 g l l U0 lUs U0 l g =− f+ ηf Us2 Us ⇒ Boundary conditions: U1 (−∞) = 0 U1 (∞) = 0 ⇒ f (−∞) = 0 f (∞) = 0 ⇒ l L . Us (x) = U1 (x.B.

we need to model the turbulent shear stress. E. Rel = AB lUs = = const ν ν ⇒ A turbulent wake will remain turbulent! To determine the wake-velocity profile. The momentum loss thickness is constant: θ= ∞ −∞ U U0 1− U U0 1 U0 ∞ dy = −∞ ∞ −∞ U0 − U 1 U0 U1 dy ≈= ∞ 1 U0 1− U0 − U 1 U0 ∞ dy = 1− −∞ U1 U0 1 CD d = const (Recitation 12) 2 ∞ U1 dy = Us l −∞ ⇒ f dη = const −∞ const Us l = Axn−1 Bxn = const ⇒ 1 Us = Ax− 2 . 2n − 1 = 0 ⇒ n= 1 2 ⇒ 1 l = Bx 2 For a sufficiently large Reynolds number.Similarity requires: lUs = const Us2 l = const Us ⇒ Assume the x-dependence: Us = Axm . ⇒ m=n−1 ⇒ l = Bxn We need one more condition. −uv = νT Us2 g = νT g=− − − νT f Us l ∂U1 ∂y ⇒ ∂U ∂y ⇒ = −νT Us f g =− 1 l νT f Us l νT U0 lUs U0 l f =− f+ ηf Us l Us2 Us f + U0 l l U0 l 2 Us f − f =0 νT ν T Us f + U0 B 2 (f + f ) = 0 (2) 2νT Equation (2) can be solved together with the boundary conditions f (±∞) = 0 ⇒ ⇒ ⇒ ⇒ U1 dy ≈ U0 . l = Bxn Then we get Bxn Amxm−1 = const · x2m ⇒ n + m − 1 = 2m Us = Axn−1 .g. the wake will be turbulent.

j+1 iteration direction z x y θ 1 n x g y i j fine grid (1) a b r coarse grid (2) Figure B. see figure B. n 0 0 Consider the relative Γ0 motion of two fluid particles initially separated by the distance dx i . Thus. be updated velocity Note that x0i = xi at tˆ = 0. Relative motion. Flow between a rod and a cylinder. Rδi1 . Γ1 Ω deformation rate of the volume of the cube is given by the divergence of the c) (4) Show that the : to field. Γ1 a) (5) Show that the Ω separation at time dtˆ is M2 ∂ui M dri (dtˆ) = dx0i + 0 dx0j dtˆ. (Here xi and tˆ are the Lagrangian coordinates). and Consider a long ihollow water occupying thei space between the rod and the cylinder.12: Water and a concentric rotating rod inside a cylinder. j−1 (10) Find the velocity j field of the water driven by the rotating rod and the gravitational acceleration g. uh ∂x1 Γ0 with analogous expressions for the other two sides.14. a concentric rod with radius a inside the cylinder. m ∂xj m+2 b) (2) Use this relation 1 to write down the expression for the deformation of the sides of a small cube aligned with the coordinate directions.B. isV deformed into Vh ∂ui u (1) dri = R(δi1 + 0 dtˆ). The rod is rotating i +angular 1 with the constant frequency ω. show that the components of the side aligned with the h x-direction. . : new values 2. : old values − 1 cylinder with radius b.14 2 Old Exams i j Exam 031023 Inflow Solid wall 1.

∂t ∂xj ∂xj ∂xj (8) Derive the heat equation governing the mean flow. U Y U+u 1 n x y i j fine grid (1) coarse grid (2) X Figure B. Consider the laminar wake flow downstream of a two-dimensional streamlined body at high Reynolds number. Show. Find the similarity solution for u1 . 0) so that f (η) = u1 (x. The heat equation for the instantaneous turbulent flow is ∂ T˜ ∂ T˜ ∂ 2 T˜ +u ˜j =κ . in the direction of the wake is constant in x ∞ Q1 = ρu1 dy.j Inflow Solid wall n Γ0 Γ1 Ω M2 M m m+2 1 h V Vh u uh Γ0 Γ1 Ω : to be updated : new values : old values i−1 i i+1 j−1 j j+1 iteration direction x y 3. 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 . where k is a given constant. 5.13: Wake downstream of a flat plate. Discuss the validity of the derived relation. a) (3) Motivate the usage of the equation U ∂u1 ∂ 2 u1 =ν . using (∗). (∗) −∞ c) (7) A similarity solution can be found by scaling the wake velocity with Us (x) = u1 (x. c) (3) Use the result in (a) to derive Bernoulli’s equation for steady inviscid flow. . Mean heat equation. 4. Laminar wake flow. see figure B. b) (3) Use this in the momentum equation to derive Bernoulli’s equation for unsteady potential flow. ∂x ∂y 2 b) (2) Show that the mass flux Q1 . y)/Us (x).13. Bernoulli’s equation a) (3) Derive the following formula: uj ∂ui 1 ∂ = (uj uj ) + ∂xj 2 ∂xi ijk ωj uk . where η = y/δ(x) is the similarity variable. that Us = k/δ(x).

See Kundu & Cohen pp. 57. uθ = uz = 1g 4ν ωa2 b2 − a 2 b2 − r 2 r r2 − a2 − (b2 − a2 ) 3. Flow between a rod and a cylinder. Laminar wake flow. 511-512. 110-114. 2003-10-23 1. 4. See Kundu & Cohen p. u1 = 5. 2.Answers to exam in Fluid mechanics 5C1214. See Kundu & Cohen pp. Q1 2ρ U − U y2 e 4νx πνx ln(r/a) ln(b/a) .

Γ0 Γ1 Ω flow. a constant shear stress Due to a varying surface τ at the free surface Γis1 acting upward. : new values b) (3) For what thickness h is the flux zero? : old layer values i−1 z i i+1 j−1 j τ j+1 iteration direction x g y 1 n x y i x h j fine grid (1) coarse grid (2) Figure B. the gravitational acceleration g is acting Ω downward. Compare the group velocity to the phase speed in the limit of long and short waves. 1. At the same time. : to be updated a) (7) Assume constant layer thickness h and calculate the velocity field of the liquid layer. Consider a thin layer uof liquid in contact with a solid flat vertical wall. The wall is subject to a constantuhtemperature gradient with increasing temperature in the downward direction. M where u0 and k are positive constants. v = kt. Show that the streamlines are straight lines. w = 0. and sketch them m at two different times. (5) For surface waves1on water with finite depth h the phase speed is given by the relation c 2 = g h k tanh(kh). (5) Consider the unsteady M2 u = u0 . Vh 3.14: Liquid film on vertical wall driven by gravity and surface tension. V Make a rough 2D sketch of a wave packet and explain how it is moving. as a result of the temperature gradient. Exam 040113 2. . m + 2Also show that any fluid particle follows a parabolic path as time proceeds. See figure 1. Γ0 tension.

a) (3) Show that uj ∂ui 1 ∂ = (uj uj ) + ∂xj 2 ∂xi ijk ωj uk b) (4) Derive the vorticity equation starting with the Navier–Stokes equation for incompressible flow. b) (5) Assume that the Reynolds stress can be modeled as a turbulent viscosity and introduce this into the Reynolds equation and simplify. where k = ω . t) = U e−ky cos(ky − ωt). . a) (3) Motivate that the solution can be written as [u(y.4. 0] and that it thus satisfies the equation ∂2u ∂u =ν 2 ∂t ∂y b) (7) Show that u(y. a) (5) Derive the Reynolds average equation valid for turbulent flow. 0. Consider the flow over an oscillating plate. 2ν 6. c) (3) Show that the following relation holds for incompressible flow and discuss its implication for inviscid flow ∂τij = −µ ∂xj ijk ∂ωk ∂xj 5. t).

∂x ν ∂w gh τ (h) = +A = ∂x ν µ ⇒ A= 1 ν τ − gh ρ . w(0) = 0. The wave number is defined as k = 2π/λ. n x y i j fine grid (1) coarse grid (2) 2 1. For long waves.1 Solutions to exam in Fluid mechanics 5C1214. (h) = ∂x ∂x µ Integrate the equation once and introduce the second boundary condition: ∂w g = x + A.5 3 3.5 1 1.5 1 0. go to the limit k → 0.2 j−1 A B C u 23 .5 −1 −1. parallel flow u ¯ = w(x)¯ ez and solve the z-momentum equation with boundary conditions: ∂ 2w ∂w τ ν 2 − g = 0. Streamline (fix t): dx = u = u0 ds dy = u = kt ds Then we see that y(x) is a linear function. 32 p1.5 0 −2 0. Phase speed: c= g tanh(kh) = k ω = k Short waves: lim tanh(x) 1 = x x lim tanh(x) =1 x x→∞ ⇒ gh tanh(kh) kh g k c= Long waves: x→0 ⇒ gh c= Group velocity: cg = dω c = dk 2 lim x =0 sinh(x) ⇒ cg = lim x =1 sinh(x) ⇒ cg = c Short waves: x→∞ Long waves: x→0 1+ 2kh sinh(2kh) c 2 Longer waves travel faster. y= 1 k 2 (x − 2x0 x + x20 ) + y0 2 u20 2. 2004-01-13 1. 1 y= i j Inflow Solid wall n Γ0 Γ1 Ω M2 M m m+2 1 h V Vh u uh Γ0 Γ1 Ω : to be updated : new values : old values i−1 i i+1 j−1 j j+1 iteration direction x y ⇒ x = u0 s + a ⇒ y = kts + b kt x + d(t) u0 Particle path (fix x ¯0 ): ∂x = u = u0 ⇒ x = u0 tˆ + x0 ∂ tˆ ∂y 1 = u = k tˆ ⇒ x = k tˆ2 + y0 ˆ 2 ∂t Then we see that y(x) is a parabola.5 0 −0. a) Assume steady.1 v1. For short waves.5 2 2. go to the limit k → ∞. where λ denotes the wave length of the wave.

We end up with the vorticity equation: ∂ωi ∂ωi ∂ui ∂ 2 ωi + uj = ωj +ν ∂t ∂xj ∂xj ∂xj ∂xj c) Right hand side: −µ ijk ∂ωk = −µ ∂xj −µ(δil δjm − δim δjl ) Left hand side: ijk ∂ ∂xj klm ∂um = −µ ∂xl kij klm ∂ 2 um = ∂xj ∂xl ∂ 2 um ∂ 2 uj ∂ 2 ui = −µ − = µ∇2 ui ∂xj ∂xl ∂xj ∂xi ∂xj ∂xj ∂τij ∂ =µ ∂xj ∂xj ∂ui ∂uj + = µ∇2 ui ∂xj ∂xi Thus: ∂τij = −µ ∂xj Irrotational flow is not subject to viscous forces. a): ∂ui 1 ∂ + (uj uj ) + ∂t 2 ∂xi ijk ωj uk =− 1 ∂p ∂ 2 ui +ν ρ ∂xi ∂xj ∂xj Take the curl of the equation.Integrate the equation once more and introduce the first boundary condition: w= 1g 2 x + Ax + B. The curl of a gradient vanish: ∂ωi + ∂t imn ∂ωi + ∂t nim njk ∂ 2 ωi ∂xj ∂xj ⇒ ∂ ∂ 2 ωi (ωj uk ) = ν ∂xm ∂xj ∂xj ⇒ ∂ ( ∂xm njk ωj uk ) =ν ∂ωi ∂ ∂ 2 ωi + (δij δmk − δik δmj ) (ωj uk ) = ν ∂t ∂xm ∂xj ∂xj ⇒ ∂ωi ∂ ∂ ∂ 2 ωi + (ωi uk ) − (ωj ui ) = ν ∂t ∂xk ∂xj ∂xj ∂xj Use continuity. a) Begin with the second term on the right hand side: ijk ωj uk = uj ijk jlm ∂ ∂um ∂ui ∂uk (um )uk = (δkl δim − δkm δil )uk = uk − uk ∂xl ∂xl ∂xk ∂xi ∂ui ∂uj = uj + ∂xj ∂xi ⇒ ijk ωj uk uj ∂ui 1 ∂ = (uj uj ) + ∂xj 2 ∂xi ijk ωj uk b) Navier–Stokes momentum equation for incompressible flow: ∂ui ∂ui 1 ∂p ∂ 2 ui + uj =− +ν ∂t ∂xj ρ ∂xi ∂xj ∂xj Use the expression from 4. ijk ∂ωk ∂xj ⇒ . 2ν ⇒ w(0) = 0 B=0 So we get: w= 1g 2 1 x + 2ν ν τ − gh x ρ b) The flux per unit width of the wall can be written h Q= 1g 3 11 x + 6ν 2ν w dx = 0 τ − gh x2 ρ Zero flux gives: h= h = 0 11τ 2 1g 3 h − h 2ν ρ 3ν 3 τ 2 ρg 4.

2ν f (y) = Aeyk(1+i) + Be−yk(1+i) . Divide the total flow into an average and a fluctuating component ui as ui = u i + u i p=p+p and introduce into the Navier-Stokes equations. in applications one is not usually interested in knowing full the details of this flow.5. Insert into the equation: iωf (y)eiω t = νf (y)eiωt f (y) − λ2 − Introduce k = iω f (y) = 0 ν iω =0 ν ⇒ with f (y) = eλy gives iω =± ν 1+i √ 2 λ=± ω ν ω . u(y. t) = U e−ky cos(ky − ωt). We are now going to derive an equation governing the mean flow Ui . For this purpose we define an ensemble average as 1 N →∞ N N (n) Ui = ui = lim (n) where each member of the ensemble ui ui n=1 is regarded as an independent realization of the flow. but rather satisfied with the influence of the turbulence on the averaged flow. using ui = 0 which gives uj ui = ui · uj = 0 . However. ∂u ∂2u =ν 2 ∂t ∂y b) Make the ansatz: u = f (y)eiω t = f (y) cos(ω t). f (y) → 0 as y → ∞ → A = 0 We have u= Be−yk(1+i) eiwt = Be−ky ei(ωt−ky) = Be−ky cos(ωt − ky) = Be−ky cos(ky − ωt) Boundary condition at y = 0 At y = 0 we have u = U cos(ωt) ⇒ B = U So we have. We find ∂ui ∂ui ∂ui ∂u ∂ui ∂u 1 ∂p 1 ∂p + + uj + uj i + uj + uj i = − − + ν∇2 ui + ∇2 ui ∂t ∂t ∂xj ∂xj ∂xj ∂xj ρ ∂xi ρ ∂xi We take the average of this equation. where k = ω 2ν 6. a) Consider the Navier–Stokes equation in the x direction: ∂u ∂u ∂u ∂u 1 ∂p ∂2u ∂2u ∂2u +u +v +w =− +ν + 2 + 2 ∂t ∂x ∂y ∂z ρ ∂x ∂x2 ∂y ∂z With the current velocity field only terms with y derivatives will remain since there can be no change in the other directions. a) Turbulent flow is inherently time dependent and chaotic.

and drop the . Here the deformation rate tensor is calculated based on the mean flow. an approximation only true for very simple turbulent flow. i. It is introduced here only to point out the analogy between the molecular viscosity and the turbulent viscosity. b) Assume the Reynolds stress: ui uj = 2 Kδij − 2νT eij 3 The first term on the right hand side is introduced to give the correct value of the trace of u i uj . as above. Introducing this into the Reynolds average equations gives ⇒ ∂Ui ∂ ∂Ui + Uj = − ∂t ∂xj ∂xj 1 ∂P P 2 + k δij + 2 (ν + νT ) eij = − + (ν + νT ) ∇2 Ui ρ 3 ρ ∂xi where the last equality assumes that νT is constant. Now. . eij = 1 2 ∂Ui ∂Uj 2 ∂Ur + − δij ∂xj ∂xi 3 ∂xr where we have assumed incompressible flow. We find the Reynolds average equations  ∂Ui ∂Ui 1 ∂P ∂  + Uj =− + (τij − ui uj )  ∂t ∂xj ρ ∂xi ∂xj   ∂Ui = 0 ∂xi where ui uj is the Reynolds stress. ∂ui ∂ui ∂ 1 ∂p  + uj =− + ν∇2 ui − u i uj  ∂t ∂xj ρ ∂xi ∂xj   ∂ui = 0 ∂xi where the average of the continuity equation also has been added. let U i = ui .e.

C Jjl = V (xk xk δjl − xj xl )ρ dV. The two coordinate systems are related by the formula x = z + x . Hints: Integrals of the type V xk ρ dV vanishes as a result of the definition of the center of mass. 2 where Jjl is the moment of inertia tensor with respect to the origin of the unprimed system. see figure B.Γ0 Exam 041018 Γ1 Ω 1. with density ρ. 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. which is rotating with : to be updated angular velocity Ω can be expressed : new values 1 1 : old values dT = v · vρ dV = (Ω × x) · (Ω × x) ρ dV. j−1 j x3 j+1 dV iteration direction x x y C x 1 x2 n x3 x1 x z y i j O fine grid (1) x2 coarse grid (2)x1 Figure B.15. . 2 2 i−1 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. 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 δjl − zj zl )M + Jjl . It is defined as Jjl = (xk xk δjl − xj xl )ρ dV. The kinetic energy dT of an infinitesimal volume element.15: Coordinate system and definitions a) (5) Show that the total kinetic energy of the body can be expressed as T = dT = V 1 Jjl Ωj Ωl .

b) (8) Make appropriate assumptions for similarity and reduce the boundary layer equations to an ordinary differential equation for the boundary layer over a flat plate (i. . The flow is driven by the gravitational acceleration g. partial differential equations for the turbulent kinetic energy k¯ and the dissipation rate ε are solved. using the wall shear stress τw . Use dimensional analysis to determine uT . c) (2) Sketch the streamwise and normal velocity profiles. 5. Potential flow and conformal mapping a) (2) Show that the Kutta-Joukowski transformation ζ = z + a2 /z maps a circle with radius a to a flat plate with length 4a. Turbulent flow a) (2) Define the velocity and length scales appropriate for turbulent flow near a wall. 3. b) (2) Write down the complex potential of the flow around a circular cylinder with radius a at angle of attack α and with clockwise circulation Γ. Also state the boundary conditions for the similarity function. the turbulent viscosity ν T = uT l. (10) Consider the laminar flow of a viscous fluid down a vertical cylindrical pipe with radius a. d) (2) How should one modify the circle in order to obtain a finite velocity at the leading edge? 4. c) (4) Show that a necessary condition for finite velocity at the trailing edge of the flat plate is (Kutta condition) Γ = 4πU a sin α. The uniform far-field velocity has the magnitude U . In those equations. respectively. l and νT in terms of k¯ and ε. 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 flow? c) (4) In two-equation turbulence models.2. the viscosity ν and the density ρ. Calculate the velocity field. where uT and l are the turbulent velocity and length scales. Blasius boundary layer a) (2) Write down the boundary layer equations.e there is no pressure gradient in the free stream).

b) The moment of inertia tensor Jjl = V (δjl xk xk − xj xl )ρ dV = V V [δjl (zk + xk )(zk + xk ) − (zj + xj )(zl + xl )]ρ dV = [δjl (zk zk + 2zk xk + xk xk ) − (zj zl + zj xl + zl xj + xj xl )]ρ dV = (zk zk δjl − zj zl ) 2zk δjl V ρ dV + V V xk ρ dV + zj V (xk xk δjl − xj xl )ρ dV + xl ρ dV + zl V xj ρ dV. Jjl = (zk zk δjl − zj zl )M + Jjl where M= ρ dV V C and Jjl = V (xk xk δjl − xj xl )ρ dV. 2004-10-18 1. θ) only.Solutions to exam in Fluid mechanics 5C1214. a) The kinetic energy of an infinitesimal volume element 1 1 v · vρ dV = (Ω × x) · (Ω × x) ρ dV ⇒ 2 2 dT = dT = 1 1 1 vi vi ρ dV = εijk Ωj xk εilm Ωl xm ρ dV = (δjl δkm − δjm δkl )xk xm Ωj Ωl ρ dV = 2 2 2 1 (δjl xk xk − xj xl )Ωj Ωl ρ dV ⇒ 2 T = dT = V 1 2 V dT (δjl xk xk − xj xl )ρ dV Ωj Ωl = 1 Jjl Ωj Ωl . 2. Poiseuille flow ( Pipe flow ) Use the Navier–Stokes equations (with no volume force) in cylindrical coordinates u2 1 ∂p ur 2 ∂uθ ∂ur + (u · ∇)ur − θ = − + ν ∇2 u r − 2 − 2 ∂t r ρ ∂r r r ∂θ ∂uθ ur uθ 1 ∂p 2 ∂ur uθ + (u · ∇)uθ + =− + ν ∇2 u θ + 2 − 2 ∂t r ρ r ∂θ r ∂θ r ∂uz 1 ∂p + (u · ∇)uz = − + ν∇2 uz ∂t ρ ∂z 1 ∂ 1 ∂uθ ∂uz (r ur ) + + = 0. Considering a steady flow we get from the Navier–Stokes equations (u · ∇)uz = uz ∂uz ∂z ⇒ uz ∂uz = ν∇2 uz − g ∂z . So we have C . r ∂r r ∂θ ∂z ∂p ∂p We know that = 0 and = 0 and can directly see that ur = uθ = 0 satisfy the two first ∂θ ∂r equations. The three last terms vanishes due to the definition of the center of mass. From the continuity equation we get ∂uz =0 ∂z ⇒ uz = uz (r. 2 where Jjl = V (δjl xk xk − xj xl )ρ dV.

The radius of such a circle is a + λ and the circle is described by z = −λ + (a + λ)eiθ 4.4) . Potential flow and conformal mapping a) Equation for the circle z = aeiθ Equation for the plate ζ =z+ a2 eiθ + e−iθ = aeiθ + ae−iθ = 2a = 2a cos θ. We get ∂z ∇2 u z = Integrate r 1 ∂ ∂uz g (r )= r ∂r ∂r ν ∂uz 1g 2 = r + c1 ∂r 2ν ⇒ ⇒ ∂ ∂uz g (r )= r ∂r ∂r ν ∂uz 1g c1 = r+ ∂r 2ν r Integrating again we get uz = 1g 2 r + c1 ln r + c2 4ν using the boundary conditions uz (r = 0) < ∞ ⇒ c1 = 0 We also have uz = 0 at r = a and this gives c2 = − uz = − ga2 and we have 4ν g 2 (a − r2 ) 4ν 3. We still want a sharp trailing edge so the circle crosses the real axis at z = a. 2i d) Move the point z = −a inside of the circle by putting the center at z = −λ. z 2 which is a real function taking the values from −2a to 2a. Blasius boundary layer a) Boundary layer equations u ∂u ∂u dU ∂2u +v =U +ν 2 ∂x ∂y dx ∂y (B.But we know that ∂uz = 0 from the continuity equation. b) Complex potential for a circular cylinder F = U ze−iα + U a2 iα iΓ e + ln z z 2π c) Complex velocity for the circular cylinder W = dF a2 iΓ = U e−iα − U 2 eiα + dz z 2πz Complex velocity in the ζ-plane ω= dF/dz = dζ/dz U e−iα − U a2 iα iΓ e + 2 z 2πz 1− a2 z2 The flow field has singular points at the leading and trailing edges of the flat plate. thus a flat plate with length 4a. Resolve the singularity at the trailing edge by choosing the circulation Γ so the numerator vanishes at z = a U e−iα − U eiα + Γk = 4πU a iΓ =0 2πa eiα − e−iα = 4πU a sin α.3) ∂u ∂v + =0 ∂x ∂y (B.

16 shows the dimensionless streamwise and normal velocity profiles. . c) Figure B. U The ordinary differential equation becomes f We have the boundary conditions   u(0) = 0 v(0) = 0  u(∞) = U 1 + f f = 0. We can integrate it with respect to x in order to find δ Uδ δ =α ν ⇒ 1 2 ανx δ = 2 U ⇒ δ(x) = 2ανx U The value of the constant α is just a matter of scaling. y) U δ(x) ⇔ ψ = U δf Determine the terms in equation (1): u= v=− ∂ψ ∂f ∂η 1 = Uδ = U δf = U f ∂y ∂η ∂y δ ∂ψ ∂(U δf ) =− = −U δ f − U δf ∂x ∂x − yδ δ2 = U δ ηf − U δ f ∂u ∂ δ = (U f ) = −U ηf ∂x ∂x δ ∂u ∂ 1 = (U f ) = U f ∂y ∂y δ ∂2u ∂ = 2 ∂y ∂y Uf U 1 δ =U 1 f δ2 dU =0 dx Insert into equation (1): U f (−U −U 2 δ 1 1 ηf ) + (U δ ηf − U δ f )U f = νU 2 f δ δ δ δ δ δ 1 ηf f + U 2 ηf f − U 2 f f = νU 2 f δ δ δ δ Uδ δ f + ff = 0 ν ⇒ ⇒ α Similarity requires that α is a constant. where U is the constant free-stream velocity. we can choose it to be 1/2 so δ(x) = νx . 2 ⇒   f (0) = 0 f (0) = 0  f (∞) = 1. ∂y v=− ∂ψ ∂x Make the stream function dimensionless: f (η) = ψ(x. only depends on the dimensionless wall distance η= y δ(x) The stream function satisfies (2): u= ∂ψ .b) Seek a similarity solution where the dimensionless velocity u/U .

c) Dimensions for k¯ and ε: Dimension analysis for uT : 1 ln y + + B.1 1 i j Inflow Solid wall n Γ0 Γ1 Ω M2 M m m+2 1 h V Vh u uh Γ0 Γ1 Ω to be updated : new values 7 : old values i−1 6 i i+1 j − 15 j j +1 4 ration direction η x y3 n2 x y i1 j fine grid (1) 0 coarse grid (2) 0 5.i− 2 i 1 2 i+ i+1 i + 32 j + 12 j j − 12 j −1 A B C u 23 .16: Velocity profiles of Blasius boundary layer.2 0. uτ where y + = y l∗ Functional dependence in the intermediate region U+ = where κ is the Karmans constant. κ  k¯ = 21 ui ui [L2 T −2 ] ∂u ∂u ε = ν i j [L2 T −3 ] ∂xj ∂xi uT ∝ k¯m εn [LT −1 ] ⇒ LT −1 = L2m T −2m L2n T −3n 2m + 2n = 1 −2m − 3n = −1 l ∝ k¯m εn n=0 m = 12 ⇒ ⇒ [L] L = L2m T −2m L2n T −3n 2m + 2n = 1 −2m − 3n = 0 ⇒ k¯ uT ∝ Dimension analysis for l: ⇒ ⇒ ⇒ n = −1 m = 23 ⇒ v u 0.4 0.1 v1. Turbulent flow a) Velocity scale near the wall: τw ρ uτ = Length scale near the wall: l∗ = ν uτ b) Law of the wall U+ = U = f (y + ).8 1 Figure B.6 0. 32 p1. .

ε ⇒ .l∝ So we have 3 k¯ 2 ε 3 k¯ 2 ¯ νT = uT l ∝ k ε 2 ¯ k νT ∝ ε or νT = C µ k¯2 .

1) . Define and use the Rankine-Hugoniot jump condition to compute the shock speed for the following problem ut + uux − ∞ < x < ∞. Write a difference approximation (in 1D) on conservative form and define the notation. Define the entropy condition for a scalar conservation law. a) Prove that the scheme is consistent and find the order of accuracy.Appendix C Study questions 5C1212 1. 4. b) Determine the stability requirement for a > 0 and show that it is unstable for a < 0. ut + f (u)x = 0 t>0 with a convex flux function f (u). Define a total variation decreasing (TVD) method. Solve ut + ( u2 )x = 0. Why is this a desirable property ? 6. Assume k/h constant. 2 u(x. 7. suggest another way of solving C. − ∞ < x < ∞. 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 . 0) = 1 x<0 0 x≥0 by the upwind scheme on the following form. Investigate the one-sided difference scheme un+1 = unj − a j ∆t n (u − unj−1 ) ∆x j for the advection equation ut + aux = 0 Consider the cases a > 0 and a < 0. Apply Lax-Friedrichs scheme to the linear wave equation ut + aux = 0. Why do we need an entropy condition ? 3. un+1 = unj − j ∆t n n u (u − unj−1 ) ∆x j j a) Will the numerical solution converge to the analytical solution? b) If not.1. 0) = 2. b) What type of equations is this? c) What kind of behavior can we expect from the solution? 173 (C. = 0 t>0 1 x≤0 0 otherwise u(x. 5. a) Write down the modified equation.

can be linearized by considering small perturbations (ρ . using an explicit three-point scheme.4) x A where a is the speed of sound. .2) where p = p(ρ) = Cργ i and C a constant. x) = 0 determine the analytical solution to (C. j = 0.4) in a point P in the x-t plane.3) where a is the speed of sound. Hint: Start from the characteristic formulation. and tn = n∆t.8. see the figure below. d) Determine the partial differential equations the characteristic variables fulfill . x) = sin(x) u(0. Draw the domain of dependence of the numerical solution at P (in the same figure as a)) of the three-point scheme in the case when i) the CFL condition is fulfilled ii) the CFL condition is NOT fulfilled. 1. (C. u ) around a motionless gas. a) Draw the domain of dependence of the solution to the system (C.. a) Let ρ = ρ0 + ρ and u = u0 + u where u0 = 0. b) The system is solved numerically on a grid given by xj = j∆x.2) and show that this yields the following linear system (the primes has been dropped) ρ t + ρ 0 ux = 0 a2 ut + ρ x = 0 ρ0 (C. A the three-point centered scheme applied to ut + aux = 0. 2. a and ρ0 are constants.. a∆t (uj+1 − uj−1 ) 2∆x Show that this approximation is not stable even though the CFL condition is fulfilled. The barotropic gas dynamic equations ρt + ρux = 0 1 ut + uux + px = 0 ρ (C. 2. The linearized form of (C. e) Given initial conditions at t = 0 and let −∞ < x < ∞ (no boundaries) ρ(0.. a and ρ0 are constants. n = 0.3) is given by ρ u + t 0 a2 /ρ0 ρ0 0 ρ u = 0. .3) a hyperbolic system? Motivate your answer. Assume that P is a grid point.. Linearize the system (C.characteristic formulation.. 9.3) for t > 0. 1. yields the approximation un+1 = unj + j a > 0. 12. What is the condition on the n × n real matrix A(u) for the system ut + Aux = 0 to be hyperbolic ? 11. c) Determine the characteristic variables in terms of ρ and u. What does Lax’s equivalence theorem state? 10. b) Is the system given by (C..

a) Write down the appropriate function spaces for the pressure and velocity used to define weak solutions to the Navier-Stokes equations.♣♦♣♦ ♥♠♠ 1t n+1 x y i t fine grid (1) n coarse grid (2) ♣♦♣♦ ♣♦♣♦ ♣♦♣♦ ❦❦ ♣♦♣♦ ❤❤❣ ❣ ❥✐❥ ✐ j−1 ❧❧ j+1 j 13. show that any Galerkin velocity solution is stable. that the Navier-Stokes equations yield νK + C(u) G GT 0 u p = f 0 19. a) Show that a vector field wi can be decomposed into wi = u i + ∂p ∂xi where u is is divergence free and parallel to the boundary. show that the pressure term disappears and recover an equation for the pressure from the gradient part. 17. Describe the ideas behind a flux splitting scheme for solving a non-linear hyperbolic system of equations. b) Apply this to the Navier-Stokes equations. Ut + F (U)x = 0 15. 18. From the differential form of the Navier-Stokes equations obtain a) the Navier-Stokes equations in integral form used in finite-volume discretizations. .g. a) By choosing zh = uh in the finite element approximation of the Stokes problem. b) a variational form of the Navier-Stokes equations used in finite-element discretizations. b) Explain the concept of essential and natural boundary conditions. To solve Euler equations in 1D ρt + ρux + uρx = 0 1 ut + uux + px = 0 ρ pt + ρc2 ux + upx = 0 How many boundary conditions must be added at (motivate your answer) inflow boundary when the flow is a) Supersonic b) Subsonic outflow boundary when the flow is c) Supersonic d) Subsonic 14. b) Describe and interpret the LBB condition. e. a) How is a finite-element approximation defined? b) Explain how to convert a FEM discretization to an algebraic problem. 16.

a) Derive the finite volume (FV) discretization on arbitrary grids of the continuity equation (∂u i /∂xi = 0). c) show that both are equivalent to a central difference approximation for Cartesian grids. a) Define Gauss-Seidel iterations for the Laplace equations. b) derive the time step restrictions for the 1D version of that equation. a) Define an appropriate staggered grid that can be used for the discretization of the Navier-Stokes equations. 23. Time step restriction for Navier-Stokes solutions. b) the Taylor-Hood pair.20. a) Motivate the use of an appropriate form of the advection-diffusion equation as a model equation for stability analysis. 22. 25. State the difficulties associated with the the finite-volume discretizations of the Navier-Stokes equations on a colocated grid? and show the form of the spurious solution which exist. 27. Derive the finite element (FEM) discretization for Laplace equation on a Cartesian grid and show that it is equivalent to a central difference approximation. 26. 24. c) state the 2D equivalent of that restriction. b) derive the FV discretization for Laplace equation on a Cartesian grid. Discrete elements pairs a) constant pressure-bilinear velocities. c) discuss how to treat noslip and inflow/outflow boundary conditions. b) write down the FV discretization of the Navier-Stokes equations on a staggered cartesian grid. a) Define 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. 21. . Iterative techniques for linear systems. b) Define the 2-level multigrid method for the Laplace equation. Choice of elements. Time dependent flows. c) a stable choice with piecewise linear velocities.

A. Acheson. Computational Fluid Mechanics and Heat Transfer. [8] R. 1983. Neunhoeffer.. Computational Techniques for Fluid Dynamics. The basics with applications. Second Edition. [10] P. Peyret. Pletcher. Dornseifer and T. Elementary Fluid Dynamics. [2] J. Wesseling. Springer-Verlag. Computational Techniques for Fluid Dynamics Volume 1. Wiley-Interscience. [4] C. T.K. SIAM Monographs on Mathematical Modeling and Computation. Jr. Fletcher. 1997. [3] C. 2002. Anderson. Panton. 2001. Second Edition. Second Edition. Cohen.Bibliography [1] D. Taylor.J. Volume 2. Second Edition. Incompressible Flow. Academic Press. Anderson and R.D. Springer-Verlag. D. Oxford University Press. 1995. A practical Introduction. [7] R.D. Taylor and Francis. Computational Fluid Dynamics.J.A. Springer-Verlag. Numerical Simulation in Fluid Dynamics. Kundu and I.A. Griebel. 177 .M. 1995. [6] P. [9] J. 1991. Computational Methods for Fluid Flow. Tannehill. 1998. Fletcher. 1991. [5] M. T.C. Springer-Verlag.J. Fluid Mechanics. Second Edition.L. 1990.H. Principles of Computational Fluid Dynamics. Mc Graw-Hill.

Sign up to vote on this title
UsefulNot useful