LECTURE NOTES ON

INTERMEDIATE FLUID MECHANICS

Joseph M. Powers

Department of Aerospace and Mechanical Engineering

University of Notre Dame
Notre Dame, Indiana 46556-5637
USA

updated
17 June 2016, 4:44pm
2

CC BY-NC-ND. 17 June 2016, J. M. Powers.

Contents

Preface 11

1 Governing equations 13
1.1 Philosophy of rational continuum mechanics . . . . . . . . . . . . . . . . . . 13
1.1.1 Approaches to fluid mechanics . . . . . . . . . . . . . . . . . . . . . . 13
1.1.2 Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
1.1.3 Continuum mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.1.4 Rational continuum mechanics . . . . . . . . . . . . . . . . . . . . . . 15
1.1.5 Notions from Newtonian continuum mechanics . . . . . . . . . . . . . 16
1.2 Some necessary mathematics . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.2.1 Vectors and Cartesian tensors . . . . . . . . . . . . . . . . . . . . . . 20
1.2.1.1 Gibbs and Cartesian Index notation . . . . . . . . . . . . . 20
1.2.1.2 Rotation of axes . . . . . . . . . . . . . . . . . . . . . . . . 21
1.2.1.3 Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.2.1.4 Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.2.1.4.1 Definition . . . . . . . . . . . . . . . . . . . . . . . 27
1.2.1.4.2 Alternating unit tensor . . . . . . . . . . . . . . . . 28
1.2.1.4.3 Some secondary definitions . . . . . . . . . . . . . 28
1.2.1.4.3.1 Transpose . . . . . . . . . . . . . . . . . . . 28
1.2.1.4.3.2 Symmetric . . . . . . . . . . . . . . . . . . . 29
1.2.1.4.3.3 Antisymmetric . . . . . . . . . . . . . . . . 29
1.2.1.4.3.4 Decomposition . . . . . . . . . . . . . . . . 29
1.2.1.4.4 Tensor inner product . . . . . . . . . . . . . . . . . 30
1.2.1.4.5 Dual vector of a tensor . . . . . . . . . . . . . . . . 30
1.2.1.4.6 Tensor product: two tensors . . . . . . . . . . . . . 31
1.2.1.4.7 Vector product: vector and tensor . . . . . . . . . . 31
1.2.1.4.7.1 Pre-multiplication . . . . . . . . . . . . . . 32
1.2.1.4.7.2 Post-multiplication . . . . . . . . . . . . . . 32
1.2.1.4.8 Dyadic product: two vectors . . . . . . . . . . . . . 32
1.2.1.4.9 Contraction . . . . . . . . . . . . . . . . . . . . . . 32
1.2.1.4.10 Vector cross product . . . . . . . . . . . . . . . . . 32
1.2.1.4.11 Vector associated with a plane . . . . . . . . . . . . 33

3
4 CONTENTS

1.2.2 Eigenvalues and eigenvectors . . . . . . . . . . . . . . . . . . . . . . . 34

1.2.3 Grad, div, curl, etc. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
1.2.3.1 Gradient operator . . . . . . . . . . . . . . . . . . . . . . . 41
1.2.3.2 Divergence operator . . . . . . . . . . . . . . . . . . . . . . 42
1.2.3.3 Curl operator . . . . . . . . . . . . . . . . . . . . . . . . . . 43
1.2.3.4 Laplacian operator . . . . . . . . . . . . . . . . . . . . . . . 43
1.2.3.5 Relevant theorems . . . . . . . . . . . . . . . . . . . . . . . 43
1.2.3.5.1 Fundamental theorem of calculus . . . . . . . . . . 44
1.2.3.5.2 Gausss theorem . . . . . . . . . . . . . . . . . . . 44
1.2.3.5.3 Stokes theorem . . . . . . . . . . . . . . . . . . . . 44
1.2.3.5.4 Kinetic energy divergence identity . . . . . . . . . . 45
1.2.3.5.5 Leibnizs rule . . . . . . . . . . . . . . . . . . . . . 45
1.2.3.5.6 Reynolds transport theorem . . . . . . . . . . . . . 46
1.3 Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
1.3.1 Lagrangian description . . . . . . . . . . . . . . . . . . . . . . . . . . 46
1.3.2 Eulerian description . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
1.3.3 Material derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
1.3.4 Streamlines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
1.3.5 Pathlines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
1.3.6 Streaklines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
1.3.7 Kinematic decomposition of motion . . . . . . . . . . . . . . . . . . . 53
1.3.7.1 Translation . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
1.3.7.2 Solid body rotation and straining . . . . . . . . . . . . . . . 54
1.3.7.2.1 Solid body rotation . . . . . . . . . . . . . . . . . . 56
1.3.7.2.2 Straining . . . . . . . . . . . . . . . . . . . . . . . 56
1.3.7.2.2.1 Extensional straining . . . . . . . . . . . . . 57
1.3.7.2.2.2 Shear straining . . . . . . . . . . . . . . . . 57
1.3.7.2.2.3 Principal axes of strain rate . . . . . . . . . 58
1.3.8 Expansion rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
1.3.9 Invariants of the strain rate tensor . . . . . . . . . . . . . . . . . . . 60
1.3.10 Invariants of the velocity gradient tensor . . . . . . . . . . . . . . . . 60
1.3.11 Two-dimensional kinematics . . . . . . . . . . . . . . . . . . . . . . . 61
1.3.11.1 General two-dimensional flows . . . . . . . . . . . . . . . . . 61
1.3.11.1.1 Rotation . . . . . . . . . . . . . . . . . . . . . . . . 61
1.3.11.1.2 Extension . . . . . . . . . . . . . . . . . . . . . . . 62
1.3.11.1.3 Shear . . . . . . . . . . . . . . . . . . . . . . . . . 62
1.3.11.1.4 Expansion . . . . . . . . . . . . . . . . . . . . . . . 62
1.3.11.2 Relative motion along 1 axis . . . . . . . . . . . . . . . . . . 62
1.3.11.3 Relative motion along 2 axis . . . . . . . . . . . . . . . . . . 63
1.3.11.4 Uniform flow . . . . . . . . . . . . . . . . . . . . . . . . . . 65
1.3.11.5 Pure rigid body rotation . . . . . . . . . . . . . . . . . . . . 66
1.3.11.6 Pure extensional motion (a compressible flow) . . . . . . . . 67

CC BY-NC-ND. 17 June 2016, J. M. Powers.

CONTENTS 5

1.3.11.7 Pure shear straining . . . . . . . . . . . . . . . . . . . . . . 68

1.3.11.8 Couette flow: shear + rotation . . . . . . . . . . . . . . . . 69
1.3.11.9 Ideal point vortex: extension+shear . . . . . . . . . . . . . . 70
1.3.12 Kinematics as a dynamical system . . . . . . . . . . . . . . . . . . . 71
1.4 Conservation axioms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
1.4.1 Mass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
1.4.2 Linear momenta . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
1.4.2.1 Statement of the principle . . . . . . . . . . . . . . . . . . . 80
1.4.2.2 Surface forces . . . . . . . . . . . . . . . . . . . . . . . . . . 81
1.4.2.3 Final form of linear momenta equation . . . . . . . . . . . . 86
1.4.3 Angular momenta . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
1.4.4 Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
1.4.4.1 Total energy term . . . . . . . . . . . . . . . . . . . . . . . 90
1.4.4.2 Work term . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
1.4.4.3 Heat transfer term . . . . . . . . . . . . . . . . . . . . . . . 91
1.4.4.4 Conservative form of the energy equation . . . . . . . . . . 92
1.4.4.5 Secondary forms of the energy equation . . . . . . . . . . . 92
1.4.4.5.1 Mechanical energy equation . . . . . . . . . . . . . 93
1.4.4.5.2 Thermal energy equation . . . . . . . . . . . . . . . 93
1.4.4.5.3 Non-conservative energy equation . . . . . . . . . . 94
1.4.4.5.4 Energy equation in terms of enthalpy . . . . . . . . 94
1.4.4.5.5 Energy equation in terms of entropy . . . . . . . . 95
1.4.5 Entropy inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
1.4.6 Summary of axioms in differential form . . . . . . . . . . . . . . . . . 100
1.4.6.1 Conservative form . . . . . . . . . . . . . . . . . . . . . . . 100
1.4.6.1.1 Cartesian index form . . . . . . . . . . . . . . . . . 100
1.4.6.1.2 Gibbs form . . . . . . . . . . . . . . . . . . . . . . 100
1.4.6.1.3 Non-orthogonal index form . . . . . . . . . . . . . 100
1.4.6.2 Non-conservative form . . . . . . . . . . . . . . . . . . . . . 101
1.4.6.2.1 Cartesian index form . . . . . . . . . . . . . . . . . 101
1.4.6.2.2 Gibbs form . . . . . . . . . . . . . . . . . . . . . . 101
1.4.6.2.3 Non-orthogonal index form . . . . . . . . . . . . . 102
1.4.6.3 Physical interpretations . . . . . . . . . . . . . . . . . . . . 102
1.4.7 Complete system of equations? . . . . . . . . . . . . . . . . . . . . . 103
1.4.8 Integral forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
1.4.8.1 Mass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
1.4.8.1.1 Fixed region . . . . . . . . . . . . . . . . . . . . . . 104
1.4.8.1.2 Material region . . . . . . . . . . . . . . . . . . . . 105
1.4.8.1.3 Moving solid enclosure with holes . . . . . . . . . . 105
1.4.8.2 Linear momenta . . . . . . . . . . . . . . . . . . . . . . . . 106
1.4.8.3 Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
1.4.8.4 General expression . . . . . . . . . . . . . . . . . . . . . . . 107

CC BY-NC-ND. 17 June 2016, J. M. Powers.

6 CONTENTS

1.5 Constitutive equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

1.5.1 Frame and material indifference . . . . . . . . . . . . . . . . . . . . . 108
1.5.2 Second law restrictions and Onsager relations . . . . . . . . . . . . . 108
1.5.2.1 Weak form of the Clausius-Duhem inequality . . . . . . . . 109
1.5.2.1.1 Non-physical motivating example . . . . . . . . . . 109
1.5.2.1.2 Real physical effects. . . . . . . . . . . . . . . . . . 111
1.5.2.2 Strong form of the Clausius-Duhem inequality . . . . . . . . 112
1.5.3 Fouriers law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
1.5.4 Stress-strain rate relation for a Newtonian fluid . . . . . . . . . . . . 119
1.5.4.1 Underlying experiments . . . . . . . . . . . . . . . . . . . . 119
1.5.4.2 Analysis for isotropic Newtonian fluid . . . . . . . . . . . . 121
1.5.4.2.1 Diagonal component . . . . . . . . . . . . . . . . . 131
1.5.4.2.2 Off-diagonal component . . . . . . . . . . . . . . . 131
1.5.4.3 Stokes assumption . . . . . . . . . . . . . . . . . . . . . . . 131
1.5.4.4 Second law restrictions . . . . . . . . . . . . . . . . . . . . . 132
1.5.4.4.1 One dimensional systems . . . . . . . . . . . . . . . 132
1.5.4.4.2 Two dimensional systems . . . . . . . . . . . . . . 133
1.5.4.4.3 Three dimensional systems . . . . . . . . . . . . . . 134
1.5.5 Equations of state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
1.6 Boundary and interface conditions . . . . . . . . . . . . . . . . . . . . . . . . 138
1.7 Complete set of compressible Navier-Stokes equations . . . . . . . . . . . . . 138
1.7.0.1 Conservative form . . . . . . . . . . . . . . . . . . . . . . . 139
1.7.0.1.1 Cartesian index form . . . . . . . . . . . . . . . . . 139
1.7.0.1.2 Gibbs form . . . . . . . . . . . . . . . . . . . . . . 139
1.7.0.2 Non-conservative form . . . . . . . . . . . . . . . . . . . . . 140
1.7.0.2.1 Cartesian index form . . . . . . . . . . . . . . . . . 140
1.7.0.2.2 Gibbs form . . . . . . . . . . . . . . . . . . . . . . 140
1.8 Incompressible Navier-Stokes equations with constant properties . . . . . . . 141
1.8.1 Mass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
1.8.2 Linear momenta . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
1.8.3 Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
1.8.4 Summary of incompressible constant property equations . . . . . . . 143
1.8.5 Limits for one-dimensional diffusion . . . . . . . . . . . . . . . . . . . 143
1.9 Dimensionless compressible Navier-Stokes equations . . . . . . . . . . . . . . 143
1.9.1 Mass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
1.9.2 Linear momenta . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
1.9.3 Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
1.9.4 Thermal state equation . . . . . . . . . . . . . . . . . . . . . . . . . . 149
1.9.5 Caloric state equation . . . . . . . . . . . . . . . . . . . . . . . . . . 149
1.9.6 Upstream conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
1.9.7 Reduction in parameters . . . . . . . . . . . . . . . . . . . . . . . . . 150
1.10 First integrals of linear momentum . . . . . . . . . . . . . . . . . . . . . . . 150

CC BY-NC-ND. 17 June 2016, J. M. Powers.

CONTENTS 7

1.10.1 Bernoullis equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151

1.10.1.1 Irrotational case . . . . . . . . . . . . . . . . . . . . . . . . 152
1.10.1.2 Steady case . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
1.10.1.2.1 Streamline integration . . . . . . . . . . . . . . . . 152
1.10.1.2.2 Lamb surfaces . . . . . . . . . . . . . . . . . . . . . 153
1.10.1.3 Irrotational, steady, incompressible case . . . . . . . . . . . 154
1.10.2 Croccos theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
1.10.2.1 Stagnation enthalpy variation . . . . . . . . . . . . . . . . . 154
1.10.2.2 Extended Croccos theorem . . . . . . . . . . . . . . . . . . 156
1.10.2.3 Traditional Croccos theorem . . . . . . . . . . . . . . . . . 157

2 Vortex dynamics 159

2.1 Transformations to cylindrical coordinates . . . . . . . . . . . . . . . . . . . 159
2.1.1 Centripetal and Coriolis acceleration . . . . . . . . . . . . . . . . . . 160
2.1.2 Grad and div for cylindrical systems . . . . . . . . . . . . . . . . . . 163
2.1.2.1 Grad . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
2.1.2.2 Div . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
2.1.3 Incompressible Navier-Stokes equations in cylindrical coordinates . . 166
2.2 Ideal rotational vortex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
2.3 Ideal irrotational vortex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
2.4 Helmholtz vorticity transport equation . . . . . . . . . . . . . . . . . . . . . 172
2.4.1 General development . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
2.4.2 Incompressible conservative body force limit . . . . . . . . . . . . . . 174
2.4.2.1 Isotropic, Newtonian, constant viscosity . . . . . . . . . . . 174
2.4.2.2 Two-dimensional, isotropic, Newtonian, constant viscosity . 175
2.4.3 Physical interpretations . . . . . . . . . . . . . . . . . . . . . . . . . 175
2.4.3.1 Baroclinic (non-barotropic) effects . . . . . . . . . . . . . . 175
2.4.3.2 Bending and stretching of vortex tubes . . . . . . . . . . . . 175
2.5 Kelvins circulation theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
2.6 Potential flow of ideal point vortices . . . . . . . . . . . . . . . . . . . . . . . 179
2.6.1 Two interacting ideal vortices . . . . . . . . . . . . . . . . . . . . . . 180
2.6.2 Image vortex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
2.6.3 Vortex sheets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
2.6.4 Potential of an ideal vortex . . . . . . . . . . . . . . . . . . . . . . . 184
2.6.5 Interaction of multiple vortices . . . . . . . . . . . . . . . . . . . . . 184
2.6.6 Pressure field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
2.6.6.1 Single stationary vortex . . . . . . . . . . . . . . . . . . . . 187
2.6.6.2 Group of N vortices . . . . . . . . . . . . . . . . . . . . . . 187
2.7 Influence of walls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
2.7.1 Streamlines and vortex lines at walls . . . . . . . . . . . . . . . . . . 188
2.7.2 Generation of vorticity at walls . . . . . . . . . . . . . . . . . . . . . 190

CC BY-NC-ND. 17 June 2016, J. M. Powers.

8 CONTENTS

3 One-dimensional compressible flow 193

3.1 Generalized one-dimensional equations . . . . . . . . . . . . . . . . . . . . . 193
3.1.1 Mass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
3.1.2 Linear momentum . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
3.1.3 Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 197
3.1.4 Summary of equations . . . . . . . . . . . . . . . . . . . . . . . . . . 200
3.1.4.1 Unsteady conservative form . . . . . . . . . . . . . . . . . . 200
3.1.4.2 Unsteady non-conservative form . . . . . . . . . . . . . . . . 200
3.1.4.3 Steady conservative form . . . . . . . . . . . . . . . . . . . 201
3.1.4.4 Steady non-conservative form . . . . . . . . . . . . . . . . . 201
3.1.5 Influence coefficients . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
3.2 Flow with area change . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
3.2.1 Isentropic Mach number relations . . . . . . . . . . . . . . . . . . . . 205
3.2.2 Sonic properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
3.2.3 Effect of area change . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
3.2.4 Choking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
3.3 Normal shock waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
3.3.1 Rankine-Hugoniot equations . . . . . . . . . . . . . . . . . . . . . . . 214
3.3.2 Rayleigh line . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
3.3.3 Hugoniot curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
3.3.4 Solution procedure for general equations of state . . . . . . . . . . . 218
3.3.5 Calorically perfect ideal gas solutions . . . . . . . . . . . . . . . . . . 218
3.3.6 Acoustic limit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
3.4 Flow with area change and normal shocks . . . . . . . . . . . . . . . . . . . 227
3.4.1 Converging nozzle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
3.4.2 Converging-diverging nozzle . . . . . . . . . . . . . . . . . . . . . . . 228
3.5 Rarefactions and the method of characteristics . . . . . . . . . . . . . . . . . 229
3.5.1 Inviscid one-dimensional equations . . . . . . . . . . . . . . . . . . . 230
3.5.2 Homeoentropic flow of an ideal gas . . . . . . . . . . . . . . . . . . . 235
3.5.3 Simple waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
3.5.4 Centered rarefaction . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
3.5.5 Simple compression . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
3.5.6 Two interacting expansions . . . . . . . . . . . . . . . . . . . . . . . 241
3.5.7 Wall interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
3.5.8 Shock tube . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
3.5.9 Final note on method of characteristics . . . . . . . . . . . . . . . . . 245

4 Potential flow 249

4.1 Stream functions and velocity potentials . . . . . . . . . . . . . . . . . . . . 250
4.2 Mathematics of complex variables . . . . . . . . . . . . . . . . . . . . . . . . 252
4.2.1 Eulers formula . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
4.2.2 Polar and Cartesian representations . . . . . . . . . . . . . . . . . . . 253

CC BY-NC-ND. 17 June 2016, J. M. Powers.

CONTENTS 9

4.2.3 Cauchy-Riemann equations . . . . . . . . . . . . . . . . . . . . . . . 255

4.3 Elementary complex potentials . . . . . . . . . . . . . . . . . . . . . . . . . 257
4.3.1 Uniform flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
4.3.2 Sources and sinks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
4.3.3 Point vortices . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
4.3.4 Superposition of sources . . . . . . . . . . . . . . . . . . . . . . . . . 259
4.3.5 Flow in corners . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
4.3.6 Doublets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
4.3.7 Rankine half body . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
4.3.8 Flow over a cylinder . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
4.4 More complex variable theory . . . . . . . . . . . . . . . . . . . . . . . . . . 269
4.4.1 Contour integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
4.4.1.1 Simple pole . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
4.4.1.2 Constant potential . . . . . . . . . . . . . . . . . . . . . . . 270
4.4.1.3 Uniform flow . . . . . . . . . . . . . . . . . . . . . . . . . . 270
4.4.1.4 Quadrupole . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
4.4.2 Laurent series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
4.5 Pressure distribution for steady flow . . . . . . . . . . . . . . . . . . . . . . . 272
4.6 Blasius force theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
4.7 Kutta-Zhukovsky lift theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 275
4.8 Conformal mapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279

5 Viscous incompressible laminar flow 281

5.1 Fully developed, one dimensional solutions . . . . . . . . . . . . . . . . . . . 281
5.1.1 Pressure gradient driven flow in a slot . . . . . . . . . . . . . . . . . . 281
5.1.2 Couette flow with pressure gradient . . . . . . . . . . . . . . . . . . . 292
5.2 Similarity solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
5.2.1 Stokes first problem . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
5.2.2 Blasius boundary layer . . . . . . . . . . . . . . . . . . . . . . . . . . 311

Bibliography 327

CC BY-NC-ND. 17 June 2016, J. M. Powers.

10 CONTENTS

CC BY-NC-ND. 17 June 2016, J. M. Powers.

Preface

These are lecture notes for AME 60635, Intermediate Fluid Mechanics, taught at the Depart-
ment of Aerospace and Mechanical Engineering of the University of Notre Dame. Most of
the students in this course are beginning graduate students and advanced undergraduates in
engineering. The objective of the course is to provide a survey of a wide variety of topics in
fluid mechanics, including a rigorous derivation of the compressible Navier-Stokes equations,
vorticity dynamics, compressible flow, potential flow, and viscous laminar flow.
While there is a good deal of rigor in the development here, it is not absolute. It is
not hard to find gaps in some of the developments; consequently, the student should call on
textbooks and other reference materials for a full description. A great deal of the develop-
ment and notation for the governing equations closely follows Panton1 , who I find gives an
especially clear presentation. The material in the remaining chapters is drawn from a wide
variety of sources. A full list is given in the bibliography, though specific citations are not
given in the text. The notes, along with much information on the course itself, can be found
on the world wide web at http://www.nd.edu/powers/ame.60635. At this stage, anyone
is free to duplicate the notes.
The notes have been transposed from written notes I developed in teaching this and a
related course in the years 1991-94. Many enhancements have been added, and thanks go
to many students and faculty who have pointed out errors. It is likely that there are more
waiting to be discovered; I would be happy to hear from you regarding these or suggestions
for improvement.

Joseph M. Powers
powers@nd.edu
http://www.nd.edu/powers

Notre Dame, Indiana; USA

CC BY:
$ = 17 June 2016
\

The content of this book is licensed under Creative Commons Attribution-Noncommercial-No Derivative Works 3.0.

1
R. L. Panton, Incompressible Flow, third edition, John Wiley, New York, 2005.

11
12 CONTENTS

CC BY-NC-ND. 17 June 2016, J. M. Powers.

Chapter 1

Governing equations

see Panton, Chapters 1-6,

see Yih, Chapters 1-3, Appendix 1-2,
see Aris.

1.1 Philosophy of rational continuum mechanics

1.1.1 Approaches to fluid mechanics
We seek here to present an approach to fluid mechanics founded on the principles of rational
continuum mechanics. There are many paths to understanding fluid mechanics, and good
arguments can be made for each. A typical first undergraduate class will combine a mix of
basic equations, coupled with strong physical motivations, and allows the student to develop
a knowledge which is of great practical value and driven strongly by intuition. Such an
approach works well within the confines of the intuition we develop in everyday life. It often
fails when the engineer moves in to unfamiliar territory. For example, lack of fundamental
understanding of high Mach number flows led to many aircraft and rocket failures in the
1950s. In such cases, a return to the formalism of a careful theory, one which clearly exposes
the strengths and weaknesses of all assumptions, is invaluable in both understanding the true
fluid physics, and applying that knowledge to engineering design.
Probably the most formal of approaches is that of the school of thought advocated most
clearly by Truesdell,1 sometimes known as Rational Continuum Mechanics. Truesdell de-
veloped a broadly based theory which encompassed all materials which could be regarded
as continua, including solids, liquids, and gases, in the limit when averaging volumes were
sufficiently large so that the micro- and nanoscopic structure of these materials was unimpor-
tant. For fluids (both liquid and gas), such length scales are often on the order of microns,
while for solids, it may be somewhat smaller, depending on the type of crystalline structure.
The difficulty of the Truesdellian approach is that it is burdened with a difficult notation
1
Clifford Ambrose Truesdell, III, 1919-2000, American continuum mechanician and natural philosopher.
Taught at Indiana and Johns Hopkins Universities.

13
14 CHAPTER 1. GOVERNING EQUATIONS

and tends to become embroiled in proofs and philosophy, which while ultimately useful, can
preclude learning basic fluid mechanics in the time scale of the human lifetime.
In this course, we will attempt to steer a course between the pragmatism of undergrad-
uate fluid mechanics and the harsh formalism of the Truesdellian school. The material will
pay homage to rational continuum mechanics and will be geared towards a basic under-
standing of fluid behavior. We shall first spend some time carefully developing the governing
equations for a compressible viscous fluid. We shall then study representative solutions of
these equations in a wide variety of physically motivated limits in order to understand how
the basic conservation principles of mass, momenta, and energy, coupled with constitutive
relations influence the behavior of fluids.

1.1.2 Mechanics
Mechanics is the broad superset of the topic matter of this course. Mechanics is the science
which seeks an explanation for the motion of bodies based upon models grounded in well
defined axioms. Axioms, as in geometry, are statements which cannot be proved; they are
useful insofar as they give rise to results which are consistent with our empirical observations.
A hallmark of science has been the struggle to identify the smallest set of axioms which are
sufficient to describe our universe. When we find an axiom to be inconsistent with observa-
tion, it must be modified or eliminated. A familiar example of this is the Michelson-2 Morley3
experiment, which motivated Einstein4 to modify the Newtonian5 axioms of conservation of
mass and energy into a conservation of mass-energy.
In Truesdells exposition on mechanics, he suggests the following hierarchy:

bodies exist,

bodies are assigned to place,

geometry is the theory of place,

change of place in time is the motion of the body,

a description of the motion of a body is kinematics,

2
Albert Abraham Michelson, 1852-1931, Prussian born American physicist, graduate of the U.S. Naval
Academy and faculty member at Case School of Applied Science, Clark University, and University of Chicago.
3
Edward Williams Morley, 1838-1923, New Jersey-born American physical chemist, graduate of Williams
College, professor of chemistry at Western Reserve College.
4
Albert Einstein, 1879-1955, German physicist who developed the theory of relativity and made funda-
mental contributions to quantum mechanics and Brownian motion in fluid mechanics; spent later life in the
United States.
5
Sir Isaac Newton, 1642-1727, English physicist and mathematician and chief figure of the scientific rev-
olution of the seventeenth and eighteenth centuries. Developed calculus, theories of gravitation and motion
of bodies, and optics. Educated at Cambridge University and holder of the Lucasian chair at Cambridge. In
civil service as Warden of the Mint, he became the terror of counterfeiters, sending many to the gallows.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.1. PHILOSOPHY OF RATIONAL CONTINUUM MECHANICS 15

motion is the consequence of forces,

study of forces on a body is dynamics.

There are many subsets of mechanics, e.g. statistical mechanics, quantum mechanics,
continuum mechanics, fluid mechanics, solid mechanics. Auto mechanics, while a legitimate
topic for study, does not generally fall into the class of mechanics we consider here, though
the intersection of the two sets is not the empty set.

1.1.3 Continuum mechanics

Early mechanicians, such as Newton, dealt primarily with point mass and finite collections
of particles. In one sense this is because such systems are the easiest to study, and it makes
more sense to grasp the simple before the complex. External motivation was also present
in the 18th century, which had a martial need to understand the motion of cannonballs
and a theological need to understand the motion of planets. The discipline which considers
systems of this type is often referred to as classical mechanics. Mathematically, such systems
are generally characterized by a finite number of ordinary differential equations, and the
properties of each particle (e.g. position, velocity) are taken to be functions of time only.
Continuum mechanics, generally attributed to Euler,6 considers instead an infinite num-
ber of particles. In continuum mechanics every physical property (e.g. velocity, density,
pressure) is taken to be functions of both time and space. There is an infinitesimal property
variation from point to point in space. While variations are generally continuous, finite num-
bers of surfaces of discontinuous property variation are allowed. This models, for example,
the contact between one continuous body and another. Point discontinuities are not allowed,
however. Finite valued material properties are required. Mathematically, such systems are
characterized by a finite number of partial differential equations in which the properties of
the continuum material are functions of both space and time. It is possible to show that
a partial differential equation can be thought of as an infinite number of ordinary differen-
tial equations, so this is consistent with our model of a continuum as an infinite number of
particles.

1.1.4 Rational continuum mechanics

The modifier rational was first applied by Truesdell to continuum mechanics to distin-
guish the formal approach advocated by his school, from less formal, though mainly not
irrational, approaches to continuum mechanics. Rational continuum mechanics is developed
in a manner similar to that which Euclid7 used for his geometry: formal definitions, axioms,
and theorems, all accompanied by careful language and proofs. This course will generally
6
Leonhard Euler, 1707-1783, Swiss-born mathematician and physicist who served in the court of Cather-
ine I of Russia in St. Petersburg, regarded by many as one of the greatest mechanicians.
7
Euclid, Greek geometer of profound influence who taught in Alexandria, Egypt, during the reign of
Ptolemy I Soter, who ruled 323-283 BC.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

16 CHAPTER 1. GOVERNING EQUATIONS

follow the less formal, albeit still rigorous, approach of Pantons text, including the adoption
of much of Pantons notation.

1.1.5 Notions from Newtonian continuum mechanics

The following are useful notions from Newtonian continuum mechanics. Here we use New-
tonian to distinguish our mechanics from Einsteinian or relativistic mechanics.

Space is three-dimensional and independent of time.

An inertial frame is a reference frame in which the laws of physics are invariant.

A Galilean transformation specifies how to transform from an inertial frame to a frame

of reference moving at constant velocity. If the inertial frame has zero velocity and the
moving frame has constant velocity vo = uoi + vo j + wo k, the Galilean transformation
(x, y, z, t) (x , y , z , t ) is as follows

x = x uo t, (1.1)
y = y vo t, (1.2)
z = z wo t, (1.3)
t = t. (1.4)

Control volumes are useful; we will study three varieties:

Fixed: constant in space,

Material: no flux of mass through boundaries, can deform,
Arbitrary: can move, can deform, can have different fluid contained within.

Control surfaces enclose control volumes; they have the same three varieties:

Fixed,
Material,
Arbitrary.

Density is a material property, not used in classical mechanics, which only considers
point masses. We can define density as
PN
mi
= lim i=1 . (1.5)
V 0 V
Here V is the volume of the space considered, N is the number of particles contained
within the volume, and mi is the mass of the ith particle. We can define a length scale

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.1. PHILOSOPHY OF RATIONAL CONTINUUM MECHANICS 17

L associated with the volume V to be L = V 1/3 . In commonly encountered physical

scenarios, we expect the density to vary with distance on a macroscale, approach a
limiting value at the microscale, and become ill-defined below a cutoff scale below
which molecular effects are important. That is to say, when V becomes too small, such
that only a few molecules are contained within it, we expect wild oscillations in .
We will in fact assume that matter can be modeled as a continuum: the limit in which
discrete changes from molecule to molecule can be ignored and distances and times
over which we are concerned are much larger than those of the molecular scale. This
will enable the use of calculus in our continuum thermodynamics.
Continuum mechanics will treat macroscopic effects only and ignore individual molec-
ular effects. For example molecules bouncing off a wall exchange momentum with the
wall and induce pressure. We could use Newtonian mechanics for each particle collision
to calculate the net force on the wall. Instead our approach amounts to considering
the average over space and time of the net effect of millions of collisions on a wall.
The continuum theory can break down in important applications where the length
and time scales are of comparable magnitude to molecular time scales. Important
applications where the continuum assumption breaks down include

rarefied gas dynamics of the outer atmosphere (relevant for low orbit space vehi-
cles), and
nano-scale heat transfer (relevant in cooling of computer chips).

To get some idea of the scales involved, we note that for air at atmospheric pressure
and temperature that the time and distance between molecular collisions provide the
limits of the continuum. Under these conditions, we find for air

length > 0.1 m, and

time > 0.1 ns,

will be sufficient to admit the continuum assumption. For denser gases, these cutoff
scales are smaller. For lighter gases, these cutoff scales are larger. A sketch of a possible
density variation in a gas near atmospheric pressure is given in Fig. 1.1.
Details of collision theory can be found in advanced texts such as that of Vincenti and
Kruger.8 They show for air that the mean free path is well modeled by the equation:
M
= . (1.6)
2N d2
Here M is the molecular mass, N is Avogadros number, and d is the molecular diam-
eter.
8
W. G. Vincenti and C. H. Kruger, 1965, Introduction to Physical Gas Dynamics, John Wiley, New York,
pp. 12-26.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

18 CHAPTER 1. GOVERNING EQUATIONS

(kg/m3)

100
variation on the
sub-continuum variation on the
molecular scale continuum scale
1

0.01

x (m)
10 -8 10 -6 10 -4 0.01 1

Figure 1.1: Sketch of possible density variation of a gas near atmospheric pressure.

Example 1.1
Find the variation of mean free path with density for air.

We turn to Vincenti and Kruger for numerical parameter values, which are seen to be M =
28.9 kg/kmole, N = 6.02252 1023 molecule/mole, d = 3.7 1010 m. Thus,
 

kg 1 kmole
28.9 kmole 1000 mole
=
2, (1.7)
2 6.02252 1023 molecule
mole (3.7 1010 m)
kg
7.8895 108 molecule m2
= . (1.8)

Note that the unit molecule is not really a dimension, but really is literally a unit, which may
well be thought of as dimensionless. Thus, we can safely say
kg
7.8895 108 m2
= . (1.9)

A plot of the variation of mean free path as a function of is given in Fig. 1.2. Vincenti and
Kruger go on to consider an atmosphere with density of = 1.288 kg/m3 . For this density
kg
7.8895 108 m2
= kg
, (1.10)
1.288 m3
8
= 6.125 10 m, (1.11)
2
= 6.125 10 m. (1.12)
Vincenti and Kruger also show the mean molecular speed under these conditions is roughly
c = 500 m/s, so the mean time between collisions, , is
6.125 108 m
= = 1.225 1010 s. (1.13)
c 500 ms

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.1. PHILOSOPHY OF RATIONAL CONTINUUM MECHANICS 19

(m)
10-4

10-5

10-6

10-7 atm ~ 6.125 x 10-8 m

atm 1.288 kg/m3

10-8
0.001 0.01 0.1 1 10
(kg/m3)

Figure 1.2: Mean free path length, , as a function of density, , for air.

Density is an example of a scalar property. We shall have more to say later about
scalars. For now we say that a scalar property associates a single number with each
point in time and space. We can think of this by writing the usual notation (x, y, z, t),
which indicates has functional variation with position and time.

Other properties are not scalar, but are vector properties. For example the velocity
vector
v(x, y, z, t) = u(x, y, z, t)i + v(x, y, z, t)j + w(x, y, z, t)k, (1.14)
associates three scalars u, v, w with each point in space and time. We will see that a
vector can be characterized as a scalar associated with a particular direction in space.
Here we use a boldfaced notation for a vector. This is known as Gibbs9 notation. We
will soon study an alternate notation, developed by Einstein, and known as Cartesian10
index notation.

Other properties are not scalar or vector, but are what is know as tensorial. The
relevant properties are called tensors. The best known example is the stress tensor.
One can think of a tensor as a quantity which associates a vector with a plane inclined at
a selected angle passing through a given point in space. An example is the viscous stress
9
Josiah Willard Gibbs, 1839-1903, American physicist and chemist with a lifelong association with Yale
University who made fundamental contributions to vector analysis, statistical mechanics, thermodynamics,
and chemistry. Studied in Europe in the 1860s. Probably one of the few great American scientists of the
nineteenth century.
10
Rene Descartes, 1596-1650, French mathematician and philosopher of great influence. A great doubter
of existence who nevertheless concluded, I think, therefore I am. Developed analytic geometry.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

20 CHAPTER 1. GOVERNING EQUATIONS

tensor , which is best expressed as a three by three matrix with nine components:

xx (x, y, z, t) xy (x, y, z, t) xz (x, y, z, t)
(x, y, z, t) = yx (x, y, z, t) yy (x, y, z, t) yz (x, y, z, t) (1.15)
zx (x, y, z, t) zy (x, y, z, t) zz (x, y, z, t)

1.2 Some necessary mathematics

Here we outline some fundamental mathematical principles which are necessary to under-
stand continuum mechanics as it will be presented here.

1.2.1 Vectors and Cartesian tensors

1.2.1.1 Gibbs and Cartesian Index notation
Gibbs notation for vectors and tensors is the most familiar from undergraduate courses.
It typically uses boldface, arrows, underscores, or overbars to denote a vector or a tensor.
Unfortunately, it also hides some of the structures which are actually present in the equations.
Einstein realized this in developing the theory of general relativity and developed a useful
alternate, index notation. In these notes we will focus on what is known as Cartesian
index notation, which is restricted to Cartesian coordinate systems. Einstein also developed
a more general index system for non-Cartesian systems. We will briefly touch on this in
our summaries of our equations later in this chapter but refer the reader to books such as
those by Aris for a full exposition. While it can seem difficult at the outset, in the end
many agree that the use of index notation actually simplifies many common notions in fluid
mechanics. Moreover, its use in the archival literature is widespread, so to be conversant
in fluid mechanics, one must know index notation. The following table summarizes the
correspondences between Gibbs, Cartesian index, and matrix notation.
Quantity Common Gibbs Cartesian Matrix
Parlance Index
zeroth order tensor scalar a a ( a )
a1
a2
first order tensor vector a ai .
..
an
a11 a12 . . . a1n
a21 a22 . . . a2n
second order tensor tensor A aij . .. .. ..
.. . . .
an1 an2 . . . ann
third order tensor tensor A aijk -
fourth order tensor tensor A aijkl -
.. .. .. ..
. . . . -

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.2. SOME NECESSARY MATHEMATICS 21

Here we adopt a convention for the Gibbs notation, which we will find at times conflicts
with other conventions, in which italics font (a) indicates a scalar, bold font (a) indicates a
vector, upper case sans serif (A) indicates a second order tensor, over-lined upper case sans
serif (A) indicates a third order tensor, double over-lined upper case sans serif (A) indicates
a fourth order tensor. In Cartesian index notation, their is no need to use anything except
italics, as all terms are thought of as scalar components of a more expansive structure, with
the structure indicated by the presence of subscripts.
The essence of the Cartesian index notation is as follows. We can represent a three
dimensional vector a as a linear combination of scalars and orthonormal basis vectors:

a = ax i + ay j + az k. (1.16)

We choose now to associate the subscript 1 with the x direction, the subscript 2 with the
y direction, and the subscript 3 with the z direction. Further, we replace the orthonormal
basis vectors i, j, and k, by e1 , e2 , and e3 . Then the vector a is represented by

X3 a1
a = a1 e1 + a2 e2 + a3 e3 = ai ei = ai ei = ai = a2 .
(1.17)
i=1 a3
Following Einstein, we have adopted the convention that a summation is understood to exist
when two indices, known as dummy indices, are repeated, and have further left the explicit
representation of basis vectors out of our final version of the notation. We have also included
a representation of a as a 3 1 column vector. We adopt the standard that all vectors can
be thought of as column vectors. Often in matrix operations, we will need row vectors.
They will be formed by taking the transpose, indicated by a superscript T , of a column
vector. In the interest of clarity, full consistency with notions from matrix algebra, as well
as transparent translation to the conventions of necessarily meticulous (as well as popular)
software tools such as Matlab, we will scrupulously use the transpose notation. This comes
at the expense of a more cluttered set of equations at times. We also note that most authors
do not explicitly use the transpose notation, but its use is implicit.

1.2.1.2 Rotation of axes

The Cartesian index notation is developed to be valid under transformations from one Carte-
sian coordinate system to another Cartesian coordinate system. It is not applicable to either
general orthogonal systems (such as cylindrical or spherical) or non-orthogonal systems. It
is straightforward, but tedious, to develop a more general system to handle generalized co-
ordinate transformations, and Einstein did just that as well. For our purposes however, the
simpler Cartesian index notation will suffice.
We will consider a coordinate transformation which is a simple rotation of axes. This
transformation preserves all angles; hence, right angles in the original Cartesian system will
be right angles in the rotated, but still Cartesian system. We will require, ultimately, that
whatever theory we develop must generate results in which physically relevant quantities such

CC BY-NC-ND. 17 June 2016, J. M. Powers.

22 CHAPTER 1. GOVERNING EQUATIONS

x2
x2

x* 1 cos + x*2 cos

1 = x*

P
x*2

x1
x*1

x*1 x1

Figure 1.3: Sketch of coordinate transformation which is a rotation of axes

as temperature, pressure, density, and velocity magnitude, are independent of the particular
set of coordinates with which we choose to describe the system. To motivate this, let us
consider a two-dimensional rotation from an unprimed system to a primed system. So,
we seek a transformation which maps (x1 , x2 )T (x1 , x2 )T . We will rotate the unprimed
system counterclockwise through an angle to achieve the primed system. The rotation is
sketched in Figure 1.3. Note that it is easy to show that the angle = /2 . Here a
point P is identified by a particular set of coordinates (x1 , x2 ). One of the keys to all of
continuum mechanics is realizing that while the location (or velocity, or stress, ...) of P may
be represented differently in various coordinate systems, ultimately it must represent the
same physical reality. Straightforward geometry shows the following relation between the
primed and unprimed coordinate systems for x1

x1 = x1 cos + x2 cos . (1.18)

More generally, we can say for an arbitrary point that

x1 = x1 cos + x2 cos . (1.19)

We adopt the following notation

(x1 , x1 ) denotes the angle between the x1 and x1 axes,

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.2. SOME NECESSARY MATHEMATICS 23

(x2 , x2 ) denotes the angle between the x2 and x2 axes,

(x3 , x3 ) denotes the angle between the x3 and x3 axes,

(x1 , x2 ) denotes the angle between the x1 and x2 axes,

.
..

Thus, in two-dimensions, we have

x1 = x1 cos(x1 , x1 ) + x2 cos(x2 , x1 ). (1.20)

In three dimensions, this extends to

x1 = x1 cos(x1 , x1 ) + x2 cos(x2 , x1 ) + x3 cos(x3 , x1 ). (1.21)

Extending this analysis to calculate x2 and x3 gives

x2 = x1 cos(x1 , x2 ) + x2 cos(x2 , x2 ) + x3 cos(x3 , x2 ), (1.22)

x3 = x1 cos(x1 , x3 ) + x2 cos(x2 , x3 ) + x3 cos(x3 , x3 ). (1.23)

The above equations can be written in matrix form as

cos(x1 , x1 ) cos(x1 , x2 ) cos(x1 , x3 )
( x1 x2 x3 ) = ( x1 x2 x3 ) cos(x2 , x1 ) cos(x2 , x2 ) cos(x2 , x3 ) (1.24)
cos(x3 , x1 ) cos(x3 , x2 ) cos(x3 , x3 )

If we use the shorthand notation, for example, that 11 = cos(x1 , x1 ), 12 = cos(x1 , x2 ), etc.,
we have
11 12 13
( x1 x2 x3 ) = ( x1 x2 x3 ) 21 22 23 (1.25)
| {z } | {z }
31 32 33
xT xT | {z }
Q

In Gibbs notation, defining the matrix of s to be Q, and recalling that all vectors are taken
to be column vectors, we can alternatively say xT = xT Q. Taking the transpose of both
sides and recalling the useful identities that (a b)T = bT aT and (aT )T = a, we can also
say x = QT x.11 We call Q = ij the matrix of direction cosines and QT = ji the rotation
11
The more commonly used alternate convention of not explicitly using the transpose notation for vectors
would instead have our xT = xT Q written as x = x Q. In fact, our use of the transpose notation
is strictly viable only for Cartesian coordinate systems, while many will allow Gibbs notation to represent
vectors in non-Cartesian coordinates, for which the transpose operation is ill-suited. However, realizing that
these notes will primarily focus on Cartesian systems, and that such operations relying on the transpose
are useful notions from linear algebra, it will be employed in an overly liberal fashion in these notes. The
alternate convention still typically applies, where necessary, the transpose notation for tensors, so it would
also hold that x = QT x.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

24 CHAPTER 1. GOVERNING EQUATIONS

matrix. It can be shown that coordinate systems which satisfy the right hand rule require
further that

det Q = 1. (1.26)

The equation xT = xT Q is really a set of three linear equations. For instance, the first
is
x1 = x1 11 + x2 21 + x3 31 . (1.27)
More generally, we could say that

xj = x1 1j + x2 2j + x3 3j . (1.28)

Here j is a so-called free index, which for three-dimensional space takes on values j = 1, 2, 3.
Some rules of thumb for free indices are
A free index can appear only once in each additive term.
One free index (e.g. k) may replace another (e.g. j) as long as it is replaced in each
additive term.
We can simplify Eq. (1.28) further by writing
3
X
xj = xi ij . (1.29)
i=1

This is commonly written in the following form:

xj = xi ij . (1.30)

We again note that it is to be understood that whenever an index is repeated, as has the
index i above, that a summation from i = 1 to i = 3 is to be performed and that i is the
dummy index. Some rules of thumb for dummy indices are
dummy indices can appear only twice in a given additive term,
a pair of dummy indices, say i, i, can be exchanged for another, say j, j, in a given
additive term with no need to change dummy indices in other additive terms.
We define the Kronecker delta, ij as

0 i 6= j,
ij = (1.31)
1 i = j,
This is effectively the identity matrix I:

1 0 0
ij = I = 0 1 0 . (1.32)
0 0 1

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.2. SOME NECESSARY MATHEMATICS 25

Direct substitution proves that what is effectively the law of cosines can be written as
ij kj = ik . (1.33)

Example 1.2
Show for the two-dimensional system described in Figure 1.3 that ij kj = ik holds.
Expanding for the two-dimensional system, we get
i1 k1 + i2 k2 = ik .
First, take i = 1, k = 1. We get then
11 11 + 12 12 = 11 = 1,
cos cos + cos( + /2) cos( + /2) = 1,
cos cos + ( sin())( sin()) = 1,
cos2 + sin2 = 1.
This is obviously true. Next, take i = 1, k = 2. We get then
11 21 + 12 22 = 12 = 0,
cos cos(/2 ) + cos( + /2) cos() = 0,
cos sin sin cos = 0.
This is obviously true. Next, take i = 2, k = 1. We get then
21 11 + 22 12 = 21 = 0,
cos(/2 ) cos + cos cos(/2 + ) = 0,
sin cos + cos ( sin ) = 0.
This is obviously true. Next, take i = 2, k = 2. We get then
21 21 + 22 22 = 22 = 1,
cos(/2 ) cos(/2 ) + cos cos = 1,
sin sin + cos cos = 1.
Again, this is obviously true.

Using this, we can easily find the inverse transformation back to the unprimed coordinates
via the following operations:
kj xj = kj xi ij , (1.34)
= ij kj xi , (1.35)
= ik xi , (1.36)

kj xj = xk , (1.37)
ij xj = xi , (1.38)
xi = ij xj . (1.39)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

26 CHAPTER 1. GOVERNING EQUATIONS

The Kronecker delta is also known as the substitution tensor as it has the property that
application of it to a vector simply substitutes one index for another:

xk = ki xi . (1.40)

For students familiar with linear algebra, it is easy to show that the matrix of direction
cosines, ij , is an orthogonal matrix; that is, each of its columns is a vector which is orthogonal
to the other column vectors. Additionally, each column vector is itself normal. Such a
vector has a Euclidean norm of unity, and three eigenvalues which have magnitude of unity.
Operation of such matrix on a vector rotates it, but does not stretch it.

1.2.1.3 Vectors
Three scalar quantities vi where i = 1, 2, 3 are scalar components of a vector if they transform
according to the following rule
vj = vi ij (1.41)
under a rotation of axes characterized by direction cosines ij . In Gibbs notation, we would
say vT = vT Q, or alternatively v = QT v.
We can also say that a vector associates a scalar with a chosen direction in space by an
expression that is linear in the direction cosines of the chosen direction.

Example 1.3
Consider the set of scalars which describe the velocity in a two dimensional Cartesian system:
 
vx
vi = ,
vy

where we return to the typical x, y coordinate system. In a rotated coordinate system, using the same
notation of Figure 1.3, we find that

vx = vx cos + vy cos(/2 ) = vx cos + vy sin ,

vy = vx cos(/2 + ) + vy cos = vx sin + vy cos .

This is linear in the direction cosines, and satisfies the definition for a vector.

Example 1.4
Do two arbitrary scalars, say the quotient of pressure and density and the product of specific heat
and temperature, (p/, cv T )T , form a vector? If this quantity is a vector, then we can say
 
p/
vi = .
cv T

This pair of numbers has an obvious physical meaning in our unrotated coordinate system. If the
system were a calorically perfect ideal gas, the first component would represent the difference between

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.2. SOME NECESSARY MATHEMATICS 27

the enthalpy and the internal energy, and the second component would represent the internal energy.
And if we rotate through an angle , we arrive at a transformed quantity of
p
v1 = cos + cv T cos(/2 ).

p
v2 =
cos(/2 + ) + cv T cos().

This quantity does not have any known physical significance, and so it seems that these quantities do
not form a vector.

We have the following vector algebra

Addition
wi = ui + vi (Cartesian index notation)
w = u + v (Gibbs notation)
Dot product (inner product)
ui vi = b (Cartesian index notation)
uT v = b (Gibbs notation)
both notations require u1 v1 + u2 v2 + u3 v3 = b.
While ui and vi have scalar components which change under a rotation of axes, their inner
product (or dot product) is a true scalar and is invariant under a rotation of axes. Note that
here we have in the Gibbs notation explicitly noted that the transpose is part of the inner
product. Most authors in fact assume the inner product of two vectors implies the transpose
and do not write it explicitly, writing the inner product simply as u v uT v.

1.2.1.4 Tensors
1.2.1.4.1 Definition A second order tensor, or a rank two tensor, is nine scalar compo-
nents that under a rotation of axes transform according to the following rule:

Tij = ki lj Tkl . (1.42)

Note we could also write this in an expanded form as

3 X
X 3 3 X
X 3
Tij = ki lj Tkl = Tik Tkl lj . (1.43)
k=1 l=1 k=1 l=1

In the above expressions, i and j are both free indices; while k and l are dummy indices.
The Gibbs notation for the above transformation is easily shown to be

T = QT T Q. (1.44)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

28 CHAPTER 1. GOVERNING EQUATIONS

Analogously to our conclusion for a vector, we say that a tensor associates a vector with
each direction in space by an expression that is linear in the direction cosines of the chosen
direction. For a given tensor Tij , the first subscript is associated with the face of a unit cube
(hence the memory device, first-face); the second subscript is associated with the vector
components for the vector on that face.
Tensors can also be expressed as matrices. Note that all rank two tensors are two-
dimensional matrices, but not all matrices are rank two tensors, as they do not necessarily
satisfy the transformation rules. We can say

T11 T12 T13
Tij = T21 T22 T23 (1.45)
T31 T32 T33
The first row vector, ( T11 T12 T13 ), is the vector associated with the 1 face. The second
row vector, ( T21 T22 T23 ), is the vector associated with the 2 face. The third row vector,
( T31 T32 T33 ), is the vector associated with the 3 face.
We also have the following items associated with tensors.

1.2.1.4.2 Alternating unit tensor The alternating unit tensor, a tensor of rank 3, ijk
will soon be seen to be useful, especially when we introduce the vector cross product. It is
defined as follows
1 if ijk = 123, 231, or 312,
ijk = 0 if any two indices identical, . (1.46)

1 if ijk = 321, 213, or 132
Another way to remember this is to start with the sequence 123, which is positive. A
sequential permutation, say from 123 to 231, retains the positive nature. A trade, say from
123 to 213, gives a negative value.
An identity which will be used extensively
ijk ilm = jl km jm kl , (1.47)
can be proved a number of ways, including the tedious way of direct substitution for all
values of i, j, k, l, m.

1.2.1.4.3 Some secondary definitions

1.2.1.4.3.1 Transpose The transpose of a second rank tensor, denoted by a super-

script T , is found by exchanging elements about the diagonal. In shorthand index notation,
this is simply
(Tij )T = Tji . (1.48)
Written out in full, if
T11 T12 T13

Tij = T21 T22 T23 , (1.49)
T31 T32 T33

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.2. SOME NECESSARY MATHEMATICS 29

then
T11 T21 T31
T
Tij = Tji = T12 T22 T32 , (1.50)
T13 T23 T33

1.2.1.4.3.2 Symmetric A tensor Dij is symmetric iff

Dij = Dji . (1.51)

Note that a symmetric tensor has only six independent scalars. We will see that D is
associated with the deformation of a fluid element.

1.2.1.4.3.3 Antisymmetric A tensor Rij is anti-symmetric iff

Rij = Rji . (1.52)

Note that an anti-symmetric tensor must have zeroes on its diagonal, and only three inde-
pendent scalars on off-diagonal elements. We will see that R is associated with the rotation
of a fluid element.

1.2.1.4.3.4 Decomposition An arbitrary tensor Tij can be separated into a sym-

metric and anti-symmetric pair of tensors:
1 1 1 1
Tij = Tij + Tij + Tji Tji . (1.53)
2 2 2 2
Rearranging, we get
1 1
Tij = (Tij + Tji ) + (Tij Tji ) . (1.54)
|2 {z } |2 {z }
symmetric antisymmetric

The first term must be symmetric, and the second term must be anti-symmetric. This is
easily seen by considering applying this to any matrix of actual numbers. If we define the
symmetric part of the matrix Tij by the following notation

1
T(ij) = (Tij + Tji ) , (1.55)
2
and the anti-symmetric part of the same matrix by the following notation
1
T[ij] = (Tij Tji ) , (1.56)
2
we then have
Tij = T(ij) + T[ij] . (1.57)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

30 CHAPTER 1. GOVERNING EQUATIONS

1.2.1.4.4 Tensor inner product The tensor inner product of two tensors Tij and Sji
is defined as follows
Tij Sji = a, (1.58)
where a is a scalar. In Gibbs notation, we would say
T : S = a. (1.59)
It is easily shown, and will be important in upcoming derivations, that the tensor inner
product of any symmetric tensor D with any anti-symmetric tensor R is the scalar zero:
Dij Rji = 0, (1.60)
D : R = 0. (1.61)
Further, if we decompose a tensor into its symmetric and anti-symmetric parts, Tij =
T(ij) + T[ij] and take T(ij) = Dij = D and T[ij] = Rij = R, so that T = D + R, we note the
following common term can be expressed as a tensor inner product with a dyadic product:
xi Tij xj = xT T x, (1.62)
xi (T(ij) + T[ij] )xj = xT (D + R) x, (1.63)
xi T(ij) xj = xT D x, (1.64)
T(ij) xi xj = D : xxT . (1.65)

1.2.1.4.5 Dual vector of a tensor We define the dual vector, di , of a tensor Tjk as
follows12
1 1 1
di = ijk Tjk = ijk T(jk) + ijk T[jk] . (1.66)
2 2 | {z } 2
=0
The term ijk is anti-symmetric for any fixed i; thus when its tensor inner product is
taken with the symmetric T(jk) , the result must be the scalar zero. Hence, we also have
1
di = ijk T[jk] . (1.67)
2
Lets find the inverse relation for di , Starting with Eq. (1.66), we take the inner product of
di with ilm to get
1
ilm di = ilm ijk Tjk . (1.68)
2
Employing Eq. (1.47) to eliminate the s in favor of s, we get
1
ilm di = (lj mk lk mj ) Tjk , (1.69)
2
1
= (Tlm Tml ), (1.70)
2
= T[lm] . (1.71)
12
There is a lack of uniformity in the literature in this area. First, note this definition differs from that
given by Panton by a factor of 1/2. It is closer, but not identical, to the approach found in Aris, p. 25.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.2. SOME NECESSARY MATHEMATICS 31

Hence,
T[lm] = ilm di . (1.72)

Note that
0 d3 d2
T[lm] = 1lm d1 + 2lm d2 + 3lm d3 = d3 0 d1 . (1.73)
d2 d1 0

And we can write the decomposition of an arbitrary tensor as the sum of its symmetric
part and a factor related to the dual vector associated with its anti-symmetric part:

Tij = T(ij) + kij dk . (1.74)

|{z} |{z} | {z }
arbitrary tensor symmetric part antisymmetric part

1.2.1.4.6 Tensor product: two tensors The tensor product between two arbitrary
tensors yields a third tensor. For second order tensors, we have the tensor product in
Cartesian index notation as
Sij Tjk = Pik . (1.75)

Note that j is a dummy index, i and k are free indices, and that the free indices in each
additive term are the same. In that sense they behave somewhat as dimensional units, which
must be the same for each term. In Gibbs notation, the equivalent tensor product is written
as
S T = P. (1.76)

Note that in contrast to the tensor inner product, which has two pairs of dummy indices
and two dots, the tensor product has one pair of dummy indices and one dot. The tensor
product is equivalent to matrix multiplication in matrix algebra.
An important property of tensors is that, in general, the tensor product does not commute,
S T 6= T S. In the most formal manifestation of Cartesian index notation, one should also
not commute the elements, and the dummy indices should appear next to another in adjacent
terms as above. However, it is of no great consequence to change the order of terms so that
we can write Sij Tjk = Tjk Sij . That is in Cartesian index notation, elements do commute.
But, in Cartesian index notation, the order of the indices is extremely important, and it is
this order that does not commute: Sij Tjk 6= Sji Tjk in general. The version presented above
Sij Tjk , in which the dummy index j is juxtaposed between each term, is slightly preferable
as it maintains the order we find in the Gibbs notation.

1.2.1.4.7 Vector product: vector and tensor The product of a vector and tensor,
again which does not in general commute, comes in two flavors, pre-multiplication and post-
multiplication, both important, and given in Cartesian index and Gibbs notation below:

CC BY-NC-ND. 17 June 2016, J. M. Powers.

32 CHAPTER 1. GOVERNING EQUATIONS

1.2.1.4.7.1 Pre-multiplication
uj = vi Tij = Tij vi , (1.77)
u = vT T 6= T v. (1.78)
In the Cartesian index notation above the first form is preferred as it has a correspondence
with the Gibbs notation, but both are correct representations given our summation conven-
tion.

1.2.1.4.7.2 Post-multiplication
wi = Tij vj = vj Tij , (1.79)
w = T v 6= vT T. (1.80)

1.2.1.4.8 Dyadic product: two vectors As opposed to the inner product between two
vectors, which yields a scalar, we also have the dyadic product, which yields a tensor. In
Cartesian index and Gibbs notation, we have

Tij = ui vj = vj ui , (1.81)
T = uvT 6= vuT . (1.82)
Notice there is no dot in the dyadic product; the dot is reserved for the inner product.

1.2.1.4.9 Contraction We contract a general tensor, which has all of its subscripts
different, by setting one subscript to be the same as the other. A single contraction will
reduce the order of a tensor by two. For example the contraction of the second order tensor
Tij is Tii , which indicates a sum is to be performed:
Tii = T11 + T22 + T33 . (1.83)
So, in this case the contraction yields a scalar. In matrix algebra, this particular contraction
is the trace of the matrix.

1.2.1.4.10 Vector cross product The vector cross product is defined in Cartesian index
and Gibbs notation as

wi = ijk uj vk , (1.84)
w = u v. (1.85)
Expanding for i = 1, 2, 3 gives
w1 = 123 u2 v3 + 132 u3 v2 = u2 v3 u3 v2 , (1.86)
w2 = 231 u3 v1 + 213 u1 v3 = u3 v1 u1 v3 , (1.87)
w3 = 312 u1 v2 + 321 u2 v1 = u1 v2 u2 v1 . (1.88)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.2. SOME NECESSARY MATHEMATICS 33

x3
(3)
t

33

32
31
23 (2)
t

13
22
12 21

x2
11

(1)
t

x1

Figure 1.4: Sample Cartesian element which is aligned with coordinate axes, along with
tensor components and vectors associated with each face.

1.2.1.4.11 Vector associated with a plane We often have to select a vector which
is associated with a particular direction. Now for any direction we choose, there exists an
associated unit vector and normal plane. Recall that our notation has been defined so that
the first index is associated with a face or direction, and the second index corresponds to
the components of the vector associated with that face. If we take ni to be a unit normal
vector associated with a given direction and normal plane, and we have been given a tensor
Tij , the vector tj associated with that plane is given in Cartesian index and Gibbs notation
by

tj = ni Tij , (1.89)
tT = nT T, (1.90)
t = TT n. (1.91)

A sketch of a Cartesian element with the tensor components sketched on the proper face is
shown in 1.4.

Example 1.5
For example, if we want to know the vector associated with the 1 face, t(1) , as shown in Figure 1.4,
we first choose the unit normal associated with the x1 face, which is the vector ni = (1, 0, 0)T . The

CC BY-NC-ND. 17 June 2016, J. M. Powers.

34 CHAPTER 1. GOVERNING EQUATIONS

x3 x
3
(3)
t
t (3)

33
rotate
32
31
23 (2)
t

13
22
12 21
t (1)
x2
11 x
1 t (2
)

(1)
t

x
2
x1

Figure 1.5: Sample Cartesian element which is rotated so that its faces have vectors which
are aligned with the unit normals associated with the faces of the element.

associated vector is found by doing the actual summation

tj = ni Tij = n1 T1j + n2 T2j + n3 T3j . (1.92)

Now n1 = 1, n2 = 0, and n3 = 0, so for this problem, we have

(1)
tj = T1j . (1.93)

1.2.2 Eigenvalues and eigenvectors

For a given tensor Tij , it is possible to select a plane for which the vector from Tij associated
with that plane points in the same direction as the normal associated with the chosen plane.
In fact for a three dimensional element, it is possible to choose three planes for which the
vector associated with the given planes is aligned with the unit normal associated with those
planes. We can think of this as finding a rotation as sketched in 1.5.
Mathematically, we can enforce this condition by requiring that

ni Tij = nj . (1.94)
| {z } |{z}
vector associated with chosen direction scalar multiple of chosen direction

Here is an as of yet unknown scalar. The vector ni could be a unit vector, but does not
have to be. We can rewrite this as

ni Tij = ni ij . (1.95)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.2. SOME NECESSARY MATHEMATICS 35

In Gibbs notation, this becomes nT T = nT I. In mathematics, this is known as a left

eigenvalue problem. Solutions ni which are non-trivial are known as left eigenvectors. We
can also formulate this as a right eigenvalue problem by taking the transpose of both sides
to obtain TT n = I n. Here we have used the fact that IT = I. We note that the left
eigenvectors of T are the right eigenvectors of TT . Eigenvalue problems are quite general
and arise whenever an operator operates on a vector to generate a vector which leaves the
original unchanged except in magnitude.
We can rearrange to form
ni (Tij ij ) = 0. (1.96)
In matrix notation, this can be written as

T11 T12 T13
( n1 n2 n3 ) T21 T22 T23 = ( 0 0 0 ) . (1.97)
T31 T32 T33

A trivial solution to this equation is (n1 , n2 , n3 ) = (0, 0, 0). But this is not interesting. We
get a non-unique, non-trivial solution if we enforce the condition that the determinant of the
coefficient matrix be zero. As we have an unknown parameter , we have sufficient degrees
of freedom to accomplish this. So, we require
 
 T11 T T 
 12 13 
 T21 T22 T23  = 0 (1.98)

 T31 T32 T33 

We know from linear algebra that such an equation for a third order matrix gives rise to a
characteristic polynomial for of the form13
(1) (2) (3)
3 IT 2 + IT IT = 0, (1.99)

(1) (2) (3)

where IT , IT , IT are scalars which are functions of all the scalars Tij . The IT s are known
as the invariants of the tensor Tij . They can be shown to be given by14

(1)
IT = Tii = tr T, (1.100)
(2) 1 1


IT = (Tii Tjj Tij Tji ) = (tr T)2 tr (T T) = (det T) tr T1 , (1.101)
2 2
1

= T(ii) T(jj) + T[ij] T[ij] T(ij) T(ij) , (1.102)
2
(3)
IT = ijk T1i T2j T3k = det T. (1.103)
13
We employ a slightly more common form here than the very similar Eq. (3.10.4) of Panton.
14
Note the obvious error in the third of Pantons Eq. (3.10.5), where the indices j and q appear three
times.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

36 CHAPTER 1. GOVERNING EQUATIONS

Here det denotes the determinant. It can also be shown that if (1) , (2) , (3) are the three
eigenvalues, then the invariants can also be expressed as
(1)
IT = (1) + (2) + (3) , (1.104)
(2)
IT = (1) (2) + (2) (3) + (3) (1) , (1.105)
(3)
IT = (1) (2) (3) . (1.106)

In general these eigenvalues, and consequently, the eigenvectors are complex. Addition-
ally, in general the eigenvectors are non-orthogonal. If, however, the matrix we are consider-
ing is symmetric, which is often the case in fluid mechanics, it can be formally proven that
all the eigenvalues are real and all the eigenvectors are real and orthogonal. If for instance,
our tensor is the stress tensor, we will show that it is symmetric in the absence of external
couples. The eigenvectors of the stress tensor can form the basis for an intrinsic coordinate
system which has its axes aligned with the principal stress on a fluid element. The eigenval-
ues themselves give the value of the principal stress. This is actually a generalization of the
familiar Mohrs circle from solid mechanics.

Example 1.6
Find the principal axes and principal values of stress if the stress tensor is

1 0 0
Tij = 0 1 2 . (1.107)
0 2 1

A sketch of these stresses is shown on the fluid element in Figure 1.6. We take the eigenvalue problem

ni Tij = nj , (1.108)
= ni ij , (1.109)
ni (Tij ij ) = 0. (1.110)

This becomes for our problem

1 0 0
( n1 n2 n3 ) 0 1 2 = (0 0 0). (1.111)
0 2 1

For a non-trivial solution for ni , we must have

 
1 0 0 

 0 1 2  = 0. (1.112)

 0 2 1 

This gives rise to the polynomial equation

(1 ) ((1 )(1 ) 4) = 0. (1.113)

This has three solutions

= 1, = 1, = 3. (1.114)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.2. SOME NECESSARY MATHEMATICS 37

x3

x2

x1

Figure 1.6: Sketch of stresses being applied to a cubical fluid element. The thinner lines
with arrows are the components of the stress tensor; the thicker lines on each face represent
the vector associated with the particular face.

Notice all eigenvalues are real, which we expect since the tensor is symmetric.
Now lets find the eigenvectors (aligned with the principal axes of stress) for this problem First, it
can easily be shown that when the vector product of a vector with a tensor commutes when the tensor
is symmetric. Although this is not a crucial step, we will use it to write the eigenvalue problem in a
slightly more familiar notation:
ni (Tij ij ) = 0 = (Tij ij ) ni = 0, because scalar components commute. (1.115)
Because of symmetry, we can now commute the indices to get
(Tji ji ) ni = 0, because indices commute if symmetric. (1.116)
Expanding into matrix notation, we get

T11 T21 T31 n1 0
T12 T22 T32 n2 = 0 . (1.117)
T13 T23 T33 n3 0
Note, we have taken the transpose of T in the above equation. Substituting for Tji and considering the
eigenvalue = 1, we get
0 0 0 n1 0
0 0 2 n2 = 0 . (1.118)
0 2 0 n3 0
We get two equations 2n2 = 0, and 2n3 = 0, which require that n2 = n3 = 0. We can satisfy all
equations with an arbitrary value of n1 . It is always the case that an eigenvector will have an arbitrary
magnitude and a well-defined direction. Here we will choose to normalize our eigenvector and take
n1 = 1, so that the eigenvector is

1
nj = 0 for = 1. (1.119)
0

CC BY-NC-ND. 17 June 2016, J. M. Powers.

38 CHAPTER 1. GOVERNING EQUATIONS

Note, geometrically this means that the original 1 face already has an associated vector which is aligned
with its normal vector.
Now consider the eigenvector associated with the eigenvalue = 1. Again substituting into the
original equation, we get
2 0 0 n1 0
0 2 2 n2 = 0 . (1.120)
0 2 2 n3 0
This is simply the system of equations
2n1 = 0, (1.121)
2n2 + 2n3 = 0, (1.122)
2n2 + 2n3 = 0. (1.123)
Clearly n1 = 0. We could take n2 = 1 and n3 = 1 for a non-trivial solution. Alternatively, lets
normalize and take
0

2
nj = 2 . (1.124)
2
2
Finally consider the eigenvector associated with the eigenvalue = 3. Again substituting into the
original equation, we get
2 0 0 n1 0
0 2 2 n2 = 0 . (1.125)
0 2 2 n3 0
This is the system of equations
2n1 = 0, (1.126)
2n2 + 2n3 = 0, (1.127)
2n2 2n3 = 0. (1.128)
Clearly again n1 = 0. We could take n2 = 1 and n3 = 1 for a non-trivial solution. Once again, lets
normalize and take
0
2
nj = 2 . (1.129)
2
2
In summary, the three eigenvectors and associated eigenvalues are

1
(1)
nj = 0 for (1) = 1, (1.130)
0

0
(2) 2
nj = 2 for (2) = 1, (1.131)
2
2

0
(3) 2
nj = 2
for (3) = 3. (1.132)
2
2

Note that the eigenvectors are mutually orthogonal, as well as normal. We say they form an orthonormal
set of vectors. Their orthogonality, as well as the fact that all the eigenvalues are real can be shown to
be a direct consequence of the symmetry of the original tensor. A sketch of the principal stresses on
the element rotated so that it is aligned with the principal axes of stress is shown on the fluid element
in Figure 1.7.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.2. SOME NECESSARY MATHEMATICS 39

x3

(3)
=3

(2)
= -1

(1) x2
=1

Figure 1.7: Sketch of fluid element rotated to be aligned with axes of principal stress, along
with magnitude of principal stress. The 1 face projects out of the page.

Example 1.7
For a given stress tensor, which we will take to be symmetric though the theory applies to non-
symmetric tensors as well,
1 2 4
Tij = T = 2 3 1 , (1.133)
4 1 1
(1) (2) (3)
find the three basic tensor invariants of stress IT , IT , and IT , and show they are truly invariant
when the tensor is subjected to a rotation with direction cosine matrix of
1 q
2 1
3
6 6

ij = Q = 1 13 1 (1.134)
3 3
1 0 12
2

Calculation reveals that

det Q = 1, (1.135)
and that Q QT = I, so that Q is a rotation matrix.
The eigenvalues of T, which are the principal values of stress are easily calculated to be

(1) = 5.28675, (2) = 3.67956, (3) = 3.39281. (1.136)

The three invariants of Tij are

1 2 4
(1)
IT = tr T = tr 2 3 1 = 1 + 3 + 1 = 5, (1.137)
4 1 1
(2) 1

IT = (tr T)2 tr (T T)
2

CC BY-NC-ND. 17 June 2016, J. M. Powers.

40 CHAPTER 1. GOVERNING EQUATIONS

2
1 2 4 1 2 4 1 2 4
1
= tr 2 3 1 tr 2 3 1 2 3 1 ,
2
4 1 1 4 1 1 4 1 1

21 4 6
1 2
= 5 tr 4 14 4 ,
2
6 4 18
1
= (25 21 14 18),
2
= 14, (1.138)

1 2 4
(3)
IT = det T = det 2 3 1 = 66. (1.139)
4 1 1
Now when we rotate the tensor T, we get a transformed tensor given by
1 q
1 1 1 2 1

q6 3 2 1 2 4 3
6 6

T = QT T Q = 23 1
3
0 2 3 1 1 1
3 1 ,
3 3
1 1 12 4 1 1 1 0 12
6 3 2

4.10238 2.52239 1.60948
= 2.52239 0.218951 2.91291 . (1.140)
1.60948 2.91291 1.11657

We then seek the tensor invariants of T . Leaving out some of the details, which are the same as those
for calculating the invariants of the T, we find the invariants indeed are invariant:
(1)
IT = 4.10238 0.218951 + 1.11657 = 5, (1.141)
(2) 1 2
IT = (5 53) = 14, (1.142)
2
(3)
IT = 66. (1.143)

Finally, we verify that the stress invariants are indeed related to the principal values (the eigenvalues
of the stress tensor) as follows
(1)
IT = (1) + (2) + (3) = 5.28675 3.67956 + 3.39281 = 5, (1.144)
(2) (1) (2) (2) (3) (3) (1)
IT = + + ,
= (5.28675)(3.67956) + (3.67956)(3.39281) + (3.39281)(5.28675) = 14, (1.145)
(3) (1) (2) (3)
IT = = (5.28675)(3.67956)(3.39281) = 66. (1.146)

1.2.3 Grad, div, curl, etc.

Thus far, we have mainly dealt with the algebra of vectors and tensors. Now let us consider
the calculus. For now, let us consider variables which are a function of the spatial vector xi .
We shall soon allow variation with time t also. We will typically encounter quantities such
as

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.2. SOME NECESSARY MATHEMATICS 41

(xi ) a scalar function of the position vector,

vj (xi ) a vector function of the position vector, or

Tjk (xi ) a tensor function of the position vector.

1.2.3.1 Gradient operator

The gradient operator, sometimes denoted by grad, is motivated as follows. Consider
(xi ), which when written in full is

(xi ) = (x1 , x2 , x3 ). (1.147)

Taking a derivative using the chain rule gives

d = dx1 + dx2 + dx3 . (1.148)
x1 x2 x3
Following Panton, we define a non-traditional, but useful further notation i for the partial
derivative

x1 1

i = e1 + e2 +
e3 = = x2 = 2 , (1.149)
xi x1 x2 x3 3
x3

so that the chain rule is actually

d = 1 dx1 + 2 dx2 + 3 dx3 , (1.150)

which is written using our summation convention as

d = i dxi . (1.151)

After commuting so as to juxtapose the i subscript, we have

d = dxi i . (1.152)

In Gibbs notation, we say

d = dxT = dxT grad . (1.153)
We can also take the transpose of both sides, recalling that the transpose of a scalar is the
scalar itself, to obtain

T
(d)T = dxT , (1.154)
T
d = () dx, (1.155)
d = T dx. (1.156)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

42 CHAPTER 1. GOVERNING EQUATIONS

Here we expand T as T = (1 , 2 , 3 ). When i or operates on a scalar, it is known as

the gradient operator. The gradient operator operating on a scalar function gives rise to a
vector function.
We next describe the gradient operator operating on a vector. For vectors in Cartesian
index and Gibbs notation, we have, following a similar analysis15

dvi = dxj j vi = j vi dxj ,

dvT = dxT vT ,
dv = (vT )T dx,
dv = (grad v)T dx. (1.157)

Here the quantity j vi is the gradient of a vector, which is a tensor. So the gradient operator
operating on a vector raises its order by one. Note that the Gibbs notation with transposes
suggests properly that the gradient of a vector can be expanded as

1 1 v1 1 v2 1 v3
vT = 2 ( v1 v2 v3 ) = 2 v1 2 v2 2 v3 . (1.158)
3 3 v1 3 v2 3 v3

Lastly we consider the gradient operator operating on a tensor. For tensors in Cartesian
index notation, we have, following a similar analysis

dTij = dxk k Tij = k Tij dxk ,

(1.159)

Here the quantity k Tij is a third order tensor. So the gradient operator operating on a
tensor raises its order by one as well. The Gibbs notation is not straightforward as it can
involve something akin to the transpose of a three-dimensional matrix.

1.2.3.2 Divergence operator

The contraction of the gradient operator on either a vector or a tensor is known as the
divergence, sometimes denoted by div. For the divergence of a vector, we have

i vi = 1 v1 + 2 v2 + 3 v3 = T v = div v. (1.160)

The divergence of a vector is a scalar.

For the divergence of a second order tensor, we have

i Tij = 1 T1j + 2 T2j + 3 T3j = T T = div T. (1.161)

15
A more common approach, not using the transpose notation, would be to say here for the Gibbs notation
that dv = dxv. However, this is only works if we consider dv to be a row vector, as dxv is a row vector.
All in all, while at times clumsy, the transpose notation allows a for great deal of clarity and consistency
with matrix algebra.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.2. SOME NECESSARY MATHEMATICS 43

The divergence operator operating on a tensor gives rise to a row vector. We will sometimes
have to transpose this row vector in order to arrive at a column vector, e.g. we will have

T
need for the column vector T T . We note that, as with the vector inner product, most
texts assume the transpose operation is understood and write the divergence of a vector or
tensor simply as v or T.

1.2.3.3 Curl operator

The curl operator is the derivative analog to the cross product. We write it in the following
three ways:

i = ijk j vk ,
= v, (1.162)
= curl v.

Expanding for i = 1, 2, 3 gives

1 = 123 2 v3 + 132 3 v2 = 2 v3 3 v2 ,
2 = 231 3 v1 + 213 1 v3 = 3 v1 1 v3 ,
3 = 312 1 v2 + 321 2 v1 = 1 v2 2 v1 . (1.163)

1.2.3.4 Laplacian operator

The Laplacian16 operator can operate on a scalar, vector, or tensor function. It is a simple
combination of first the gradient followed by the divergence. It yields a function of the same
order as that which it operates on. For its most common operation on a scalar, it is denoted
by as follows
i i = T = 2 = div grad . (1.164)

In viscous fluid flow, we will have occasion to have the Laplacian operate on vector:

T
T
i i vj = T vT = 2 v T = 2 v = div grad v. (1.165)

1.2.3.5 Relevant theorems

We will use several theorems which are developed in vector calculus. Here we give the
simplest of motivations, and simply present them. The reader should consult a standard
mathematics text for detailed derivations.
16
Pierre-Simon Laplace, 1749-1827, Normandy-born French astronomer of humble origin. Educated at
Caen, taught in Paris at Ecole Militaire.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

44 CHAPTER 1. GOVERNING EQUATIONS

1.2.3.5.1 Fundamental theorem of calculus The fundamental theorem of calculus is

as follows Z x=b Z x=b  
d
f (x) dx = dx = (b) (a). (1.166)
x=a x=a dx
It effectively says that to find the integral of a function f (x), which is the area under the
curve, it suffices to find a function , whose derivative is f , and evaluate at each endpoint,
and take the difference to find the area under the curve.

1.2.3.5.2 Gausss theorem Gausss17 theorem is the analog of the fundamental theorem
of calculus extended to volume integrals. It applies to tensor functions of arbitrary order
and is as follows: Z Z
i (Tjk...) dV = ni Tjk... dS (1.167)
R S
Here R is an arbitrary volume, dV is the element of volume, S is the surface that bounds
V , ni is the outward unit normal to S, and Tjk.. is an arbitrary tensor function. The surface
integral is analogous to evaluating the function at the end points in the fundamental theorem
of calculus.
Note if we take Tjk... to be the scalar of unity (whose derivative must be zero), Gausss
theorem reduces to Z
ni dS = 0. (1.168)
S
That is the unit normal to the surface integrated over the surface, cancels to zero when the
entire surface is included.
We will use Gausss theorem extensively. It allows us to convert sometimes difficult
volume integrals into easier interpreted surface integrals. It is often useful to use this theorem
as a means of toggling back and forth from one form to another.

1.2.3.5.3 Stokes theorem Stokes18 theorem is as follows.

Z I
ni ijk j vk dS = i vi ds. (1.169)
S C

Once again S is a bounding surface and ni is its outward unit normal. The integral with the
circle through it denotes a closed contour integral with respect to arc length s, and i is the
unit tangent vector to the bounding curve C.
In Gibbs notation, it is written as
Z I
T
n v dS = T v ds. (1.170)
S C
17
Carl Friedrich Gauss, 1777-1855, Brunswick-born German mathematician, considered the founder of
modern mathematics. Worked in astronomy, physics, crystallography, optics, biostatistics, and mechanics.
Studied and taught at Gottingen.
18
Sir George Gabriel Stokes, 1819-1903, Irish-born British physicist and mathematician, holder of the Lu-
casian chair of Mathematics at Cambridge University, developed, simultaneously with Navier, the governing
equations of fluid motion, in a form which was more robust than that of Navier.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.2. SOME NECESSARY MATHEMATICS 45

1.2.3.5.4 Kinetic energy divergence identity It is easy to show that a useful identity
involving the divergence of specific kinetic energy holds:
 
1
vj j vi = i vj vj ijk vj k , (1.171)
2
 
T

1 T
v v = v v v . (1.172)
2

This is easily proved by considering the right hand side of Eq. (1.171), expanding, and using
Eqs. (1.162) and then (1.47):
 
1
i vj vj ijk vj k = vj i vj ijk vj klm l vm , (1.173)
2 | {z }
=k
= vj i vj kij klm vj l vm , (1.174)
= vj i vj (il jm im jl ) vj l vm , (1.175)
= vj i vj vj i vj +vj j vi , (1.176)
| {z }
=0
= vj j vi , QED. (1.177)

1.2.3.5.5 Leibnizs rule Leibnizs19 rule relates time derivatives of integral quantities
to a form which distinguishes changes which are happening within the boundaries to changes
due to fluxes through boundaries. It is a generalization of the more familiar control volume
approach which uses the Reynolds20 transport theorem. Leibnizs rule applied to an arbitrary
tensorial function is as follows:
Z Z Z
d Tjk...
Tjk...(xi , t) dV = dV + nl wl Tjk... dS. (1.178)
dt R(t) R(t) t S(t)

R(t) arbitrary moving volume,

S(t) bounding surface of the arbitrary moving volume,

wl velocity vector of points on the moving surface,

nl unit normal to moving surface.

19
Gottfried Wilhelm von Leibniz, 1646-1716, Leipzig-born German philosopher and mathematician. In-
vented calculus independent of Newton and employed a superior notation to that of Newton.
20
Osborne Reynolds, 1842-1912, Belfast-born British engineer and physicist, educated in mathematics at
Cambridge, first professor of engineering at Owens College, Manchester, did fundamental experimental work
in fluid mechanics and heat transfer.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

46 CHAPTER 1. GOVERNING EQUATIONS

Say we have the very special case in which Tjk... = 1; then Leibnizs rule reduces to
Z Z Z
d
dV = (1) dV + nk wk (1) dS, (1.179)
dt R(t) R(t) t S(t)
Z
dVR
= nk wk dS. (1.180)
dt S(t)

This simply says the total volume of the region, which we call VR , changes in response to
net motion of the bounding surface.

1.2.3.5.6 Reynolds transport theorem Leibnizs rule reduces to the Reynolds trans-
port theorem if we replace the tensor function Tjk... with a scalar function, say f . Further,
considering one-dimensional cases only, we can then say
Z Z x=b(t)
d x=b(t) f db da
f (x, t) dx = dx + f (b(t), t) f (a(t), t). (1.181)
dt x=a(t) x=a(t) t dt dt
As in the fundamental theorem of calculus, for the one-dimensional case, we do not have to
evaluate a surface integral; instead, we simply must consider the function at its endpoints.
Here db/dt and da/dt are the velocities of the bounding surface and analogous to wk . The
terms f (b(t), t) and f (a(t), t) are equivalent to evaluating Tjk.. on S(t).

1.3 Kinematics
The previous section was in many ways a discussion of geometry or place. Here we will
consider kinematics, the study of motion in space. Here we will pay no regard to what
causes the motion. If we knew the position of every fluid particle as a function of time, then
we could in principle also describe the velocity and acceleration of each particle. We could
also make statements about how groups of particles translate, rotate, and deform. This is
the essence of kinematics.
Fluid motion is generally a highly non-linear event. In this section, we will develop tools,
using a local linear analysis, to break down the most complex fluid flows to a summation of
fundamental motions.

1.3.1 Lagrangian description

A Lagrangian21 description is similar to a classical description of motion in that each fluid
particle is effectively labeled and tracked in terms of its initial position xoj and time t. We
take the position vector of a particle ri to be
ri = ri (xoj , t). (1.182)
21
Joseph-Louis Lagrange (originally Giuseppe Luigi Lagrangia), 1736-1813, Italian born, Italian-French
mathematician. Worked on celestial mechanics and the three body problem. Worked in Berlin and Paris.
Part of the committee which formulated the metric system.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.3. KINEMATICS 47

The velocity vi of a particular particle is the time derivative of its position, holding xoj fixed:

ri 
vi = (1.183)
t xoj

The acceleration ai of a particular particle is the second time derivative of its position,
holding xoj fixed:

2 ri 
ai = (1.184)
t2 xoj

We can also write other variables as functions of time and initial position, for example, we
could have for pressure p(xoj , t).
The Lagrangian description has important pedagogical value, but is only occasionally
used in practice, except maybe where it can be useful to illustrate a particular point. In
solid mechanics, it is often critically important to know the location of each solid element,
and it is the method of choice.

1.3.2 Eulerian description

It is more common in fluid mechanics to use the Eulerian description of fluid motion. In
this description, all variables are taken to be functions of time and local position, rather
than initial position. Here, we will take the local position to be given by the position vector
xi = ri . The transformation from Lagrangian coordinates to Eulerian coordinates is given
by

xi = ri (xoj , t),
t = t. (1.185)

1.3.3 Material derivatives

The material derivative is the derivative following a fluid particle. It is also known as the
substantial derivative or the total derivative. It is trivial in Lagrangian coordinates, since
by definition, a Lagrangian description tracks a fluid particle. It is not as straightforward in
the Eulerian viewpoint.
Consider a fluid property such as temperature or pressure, which we will call F here,
which is function of position and time. We can characterize the position and time in either
an Eulerian or Lagrangian fashion. Let the Lagrangian representation be F = FL (xoj , t) and
the Eulerian representation be F = FE (xi , t). Now both formulations must give the same
result at the same time and position; applying our transformation between the two systems
thus yields
F = FL (xoj , t) = FE (xi = ri (xoj , t), t = t). (1.186)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

48 CHAPTER 1. GOVERNING EQUATIONS

Now from basic calculus we have

 
ri  ri 
dxi = dt + dxo . (1.187)
t xoj xoj t j

From basic calculus, we also have

 
FL  FL 
dFL = dt + dxo , (1.188)
t xoj xoj t j
 
FE  FE 
dFE = dt + dxi . (1.189)
t  xi xi t

Now, we must have dF = dFL = dFE for the same fluid particle, so making substitutions
from above, we get
      !
FL  FL  F E 
 dt + FE 
 r i 
 dt + r i 
 dxo . (1.190)
dt + dxo =
t xoj
xoj t j t xi xi t t xo xoj t j
j

For the variation of F of a particular particle, we hold xoj fixed, so that dxoj = 0. Using also
the fact that t = t, so dt = dt, and dividing by dt, we get
   
FL  FE  FE  ri 
= + , (1.191)
t xo
j
t xi xi t t xo
j

and using the definition of fluid particle velocity, Eq. (1.183), we get
  
FL  FE  FE 
= + vi . (1.192)
t xoj
t xi xi t

Ignoring the operand F , FL , and FE , we can write the derivative following a particle in the
following manner as an operator
    
    T  D d
 =  + vi  =  +v =  + vT grad (1.193)
t xo t xi xi t t x t x Dt dt
j

We will generally use the following shorthand for the derivative following a particle:

d
= o + vi i . (1.194)
dt
Here a second shorthand for the partial derivative with respect to time has been introduced:
o /t|xi .

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.3. KINEMATICS 49

1.3.4 Streamlines
Streamlines are lines which are everywhere instantaneously parallel to velocity vectors. If a
differential vector dxk is parallel to a velocity vector vj , then the cross product of the two
vectors must be zero; hence for a streamline, we must have

ijk vj dxk = 0. (1.195)

In Gibbs notation, we would say
v dx = 0. (1.196)
Recalling that the cross product can be interpreted as a determinant, we get this condition
to reduce to  
 e1 e2 e3 

 v1 v2 v3  = 0. (1.197)

 dx1 dx2 dx3 

Expanding the determinant gives

e1 (v2 dx3 v3 dx2 ) + e2 (v3 dx1 v1 dx3 ) + e3 (v1 dx2 v2 dx1 ) = 0. (1.198)

Since the basis vectors e1 , e2 , and e3 are linearly independent, the coefficient on each of
them must be zero, giving rise to

dx3 dx2
v2 dx3 = v3 dx2 , = , (1.199)
v3 v2
dx1 dx3
v3 dx1 = v1 dx3 , = , (1.200)
v1 v3
dx2 dx1
v1 dx2 = v2 dx1 , = . (1.201)
v2 v1
(1.202)

Combining, we get
dx1 dx2 dx3
= = . (1.203)
v1 v2 v3
At a fixed instant in time, t = to , we set the above terms all equal to an arbitrary differential
parameter d to obtain

dx1 dx2 dx3

= = = d. (1.204)
v1 (x1 , x2 , x3 ; t = to ) v2 (x1 , x2 , x3 ; t = to ) v3 (x1 , x2 , x3 ; t = to )

Here should not be thought of as time, but just as a dummy parameter. Streamlines are
only defined at a fixed time. While they will generally look different at different times, in
the process of actually integrating to obtain them, time does not enter into the calculation.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

50 CHAPTER 1. GOVERNING EQUATIONS

We then divide each equation by d and find the above equations are equivalent to a system
of differential equations of the autonomous form

dx1
= v1 (x1 , x2 , x3 ; t = to ), x1 ( = 0) = x1o , (1.205)
d
dx2
= v2 (x1 , x2 , x3 ; t = to ), x2 ( = 0) = x2o , (1.206)
d
dx3
= v3 (x1 , x2 , x3 ; t = to ), x3 ( = 0) = x3o . (1.207)
d
After integration, which in general must be done numerically, we find

x1 ( ; to , x1o ), (1.208)
x2 ( ; to , x2o ), (1.209)
x3 ( ; to , x3o ), (1.210)

where we let the parameter vary over whatever domain we choose.

1.3.5 Pathlines
The pathlines are the locus of points traversed by a particular fluid particle. For an Eulerian
description of motion where the velocity field is known as a function of space and time
vj (xi , t), we can get the pathlines by integrating the following set of three non-autonomous
ordinary differential equations, with the associated initial conditions:

dx1
= v1 (x1 , x2 , x3 , t), x1 (t = to ) = x1o , (1.211)
dt
dx2
= v2 (x1 , x2 , x3 , t), x2 (t = to ) = x2o , (1.212)
dt
dx3
= v3 (x1 , x2 , x3 , t), x3 (t = to ) = x3o . (1.213)
dt
In general these are non-linear equations, and often require full numerical solution, which
gives us

x1 (t; x1o ), (1.214)

x2 (t; x2o ), (1.215)
x3 (t; x3o ). (1.216)

1.3.6 Streaklines
A streakline is the locus of points that have passed through a particular point at some past
time t = t. Streaklines can be found by integrating a similar set of equations to those for

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.3. KINEMATICS 51

pathlines.
dx1
= v1 (x1 , x2 , x3 , t), x1 (t = t) = x1o , (1.217)
dt
dx2
= v2 (x1 , x2 , x3 , t), x2 (t = t) = x2o , (1.218)
dt
dx3
= v3 (x1 , x2 , x3 , t), x3 (t = t) = x3o . (1.219)
dt
After integration, which is generally done numerically, we get
x1 (t; x1o , t), (1.220)
x2 (t; x2o , t), (1.221)
x3 (t; x3o , t). (1.222)
Then, if we fix time t and the particular point in which we are interested (x1o , x2o , x3o )T , we
get a parametric representation of a streakline
x1 (t), (1.223)
x2 (t), (1.224)
x3 (t). (1.225)

Example 1.8
If v1 = 2x1 + t, v2 = x2 2t, find a) the streamline through the point (1, 1)T at t = 1, b) the
pathline for the fluid particle which is at the point (1, 1)T at t = 1, and c) the streakline through the
point (1, 1)T at t = 1.

a) streamline

For the streamline we have the following set of differential equations,

dx1
= 2x1 + t|t=1 , x1 ( = 0) = 1,
d
dx2
= x2 2t|t=1 , x2 ( = 0) = 1.
d
Here it is inconsequential where the parameter has its origin, as long as some value of corresponds
to a streamline through (1, 1)T , so we have taken the origin for = 0 to be the point (1, 1)T . These
equations at t = 1 are
dx1
= 2x1 + 1, x1 ( = 0) = 1,
d
dx2
= x2 2, x2 ( = 0) = 1.
d
Solving, we get
3 2 1
x1 = e ,
2 2
x2 = e + 2.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

52 CHAPTER 1. GOVERNING EQUATIONS

Solving for , we find

  
1 2 1
= ln x1 + .
2 3 2
So, eliminating and writing x2 (x1 ), we get the streamline to be
s  
2 1
x2 = 2 x1 + .
3 2

b) pathline

For the pathline we have the following equations

dx1
= 2x1 + t, x1 (t = 1) = 1,
dt
dx2
= x2 2t, x2 (t = 1) = 1.
dt
These have solution
7 2(t1) t 1
x1 = e ,
4 2 4
x2 = 3et1 + 2t + 2.
It is algebraically difficult to eliminate t so as to write x2 (x1 ) explicitly. However, the above certainly
gives a parametric representation of the pathline, which can be plotted in x1 , x2 space.

c) streakline

For the streakline we have the following equations

dx1
= 2x1 + t, x1 (t = t) = 1,
dt
dx2
= x2 2t, x2 (t = t) = 1.
dt
These have solution
5 + 2t 2(tt) t 1
x1 = e ,
4 2 4
x2 = (1 + 2t)ett + 2t + 2.
We evaluate the streakline at t = 1 and get
5 + 2t 2(1t) 3
x1 = e ,
4 4
1t
x2 = (1 + 2t)e + 4.
Once again, it is algebraically difficult to eliminate t so as to write x2 (x1 ) explicitly. However, the
above gives a parametric representation of the streakline, which can be plotted in x1 , x2 space.
A plot of the streamline, pathline, and streakline for this problem is shown in Figure 1.8. Note
that at the point (1, 1)T , all three intersect with the same slope. This can also be deduced from the
equations governing streamlines, pathlines, and streaklines.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.3. KINEMATICS 53

x2

1.5

1
streakline

0.5

pathline
streamline

2 4 6 8 x1

Figure 1.8: Streamline, pathlines, and streaklines for unsteady flow of example problem.

1.3.7 Kinematic decomposition of motion

In general the motion of a fluid is non-linear in nearly all respects. Certainly, it is common
for particle pathlines to be far from straight lines; however, this is not actually a hallmark
of non-linearity in that linear theories of fluid motion routinely predict pathlines with finite
curvature. More to the point, we cannot in general use the method of superposition to add
one flow to another to generate a third. One fundamental source of non-linearity is the
non-linear operator vi i , which we will see appears in most of our governing equations.
However, the local behavior of fluids is nearly always dominated by linear effects. By
analyzing only the linear effects induced by small changes in velocity, which we will associate
with the velocity gradient, we will learn a great deal about the richness of fluid motion. In
the linear analysis, we will see that a fluid particles motion can be described as a summation
of a linear translation, rotation as a solid body, and straining of two types: extensional and
shear. Both types of straining can be thought of as deformation rates. We use the word
straining in contrast to strain to distinguish fluid and flexible solid behavior. Generally
it is the rate of change of strain (that is the straining) which has most relevance for a
fluid, while it is the actual strain that has the most relevance for a flexible solid. This is
because the stress in a flexible solid responds to strain, while the stress in a fluid responds to
a strain rate. Nevertheless, while strain itself is associated with equilibrium configurations
of a flexible solid, when its motion is decomposed, strain rate is relevant. In contrast, a rigid
solid can be described by only a sum of linear translation and rotation. A point mass only

CC BY-NC-ND. 17 June 2016, J. M. Powers.

54 CHAPTER 1. GOVERNING EQUATIONS

v i + dvi
P
dxi = ds dv i
i v
i
x
3
P
v
i
x
2
x1

Figure 1.9: Sketch of fluid particle P in motion with velocity vi and nearby neighbor particle
P with velocity vi + dvi .

translates; it cannot rotate or strain.

fluid motion = translation + rotation + extensional straining + shear straining,

| {z }
straining
flexible solid motion = translation + rotation + extensional straining + shear straining,
| {z }
straining
rigid solid motion = translation + rotation,
point mass motion = translation.

Let us consider in detail the configuration shown in Figure 1.9. Here we have a fluid
particle at point P with coordinates xi and velocity vi . A small distance dri = dxi away is
the fluid particle at point P , with coordinates xi + dxi . This particle moves with velocity
vi + dvi . We can describe the difference in location by the product of a unit tangent vector
i and a scalar differential distance magnitude ds: dri = dxi = i ds. Note that i is in
general not aligned with the velocity vector, and the differential distance ds is not associated
with the arc length along a particle path. Later in Sec. 1.3.12, we will select an alignment
with the particle path, and thus choose i = ti and ds = ds, where ti is the unit tangent
to the particle path and ds is the arc length.

1.3.7.1 Translation
We have the motion at P to be vi + dvi . Obviously, the first term vi represents translation.

1.3.7.2 Solid body rotation and straining

What remains is dvi , and we shall see that it is appropriate to characterize this term by both
a solid body rotation combined with straining.
We have from the chain rule that

dvj = dxi i vj , (1.226)

dvT = dxT vT , (1.227)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.3. KINEMATICS 55

T
dv = vT dx, (1.228)
dv = LT dx. (1.229)

Here i vj = vT L is the velocity gradient tensor. We can break i vj into a symmetric

and anti-symmetric part and say then

dvj = dxi (i vj) + dxi [i vj] (1.230)

| {z } | {z }
Shear and extensional straining Rotation

We also will find it useful to decompose the velocity gradient tensor L into a deformation
tensor, D:

D = Dij (i vj) , (1.231)

a rotation tensor R:

R = Rij [i vj] . (1.232)

This yields

L = D + R. (1.233)

Thus,

dvj = dxi Dij + dxi Rij = (i Dij + i Rij ) ds, (1.234)

dvT = dxT D + dxT R = T D + T R ds, (1.235)


dv = D dx + RT dx = D + RT ds. (1.236)

Let
(s)
dvj = dxi (i vj) = i (i vj) ds, (1.237)
(s) T
dv = dxT D = T D ds, (1.238)
dv(s) = D dx = D ds. (1.239)

We will see this is associated with straining, both by shear and extension. We will call the
symmetric tensor (i vj) = D the strain rate or deformation tensor.
Further, let
(r)
dvj = dxi [i vj] = i [i vj] ds, (1.240)
T
dv(r) = dxT R = T R ds, (1.241)
dv(r) = RT dx = RT ds. (1.242)

We will see this is associated with rotation as a solid body, with [i vj] = R as the rotation
tensor.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

56 CHAPTER 1. GOVERNING EQUATIONS

(r)
1.3.7.2.1 Solid body rotation Let us examine dvj . First, we define the vorticity
vector k as the curl of the velocity field
k = kij i vj , (1.243)
= v. (1.244)
Let us now split the velocity gradient i vj into its symmetric and anti-symmetric parts and
recast the vorticity vector as
k = kij (i vj) +kij [i vj] . (1.245)
| {z }
=0
The first term on the right side is zero because it is the tensor inner product of an anti-
symmetric and symmetric tensor. In what remains, we see that half of the vorticity k is
actually the dual vector, k , associated with the anti-symmetric [i vj] .
k = kij [i vj] = v, (1.246)
1 1 1
k = k = kij [i vj] = v. (1.247)
2 2 2
Using Eq. (1.72) to invert Eq. (1.247), we find
1
[i vj] = kij k = kij k . (1.248)
2
Thus we have
(r) 1
dvj = dxi kij k , (1.249)
2 
k
= kij dxi , (1.250)
 2 
k
= jki dxi , (1.251)
2
1
= dr and if = , (1.252)
2 2
= dr
| {z } . (1.253)
Solid body rotation of one point about another

By introducing the above definition for , we see this term takes on the exact form for
the differential velocity due to solid body rotation of P about P from classical rigid body
kinematics. Hence, we give it the same interpretation.

1.3.7.2.2 Straining Next we consider the remaining term, which we will associate with
straining. First, let us further decompose this into what will be seen to be an extensional
(es) straining and a shear straining (ss):
(s) (es) (ss)
dvk = dvk + dvk , (1.254)
| {z } | {z }
extension shear
(s) (es) (ss)
dv = dv + dv . (1.255)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.3. KINEMATICS 57

1.3.7.2.2.1 Extensional straining Let us define the extensional straining to be the

(s)
component of straining in the direction of dxj . To do this, we need to project dvj onto the
unit vector j , then point the result in the direction of that same unit vector;
 
(es) (s)
dvk = j dvj k . (1.256)
| {z }
projection of straining

(s)
Now using the definition of dvj , Eq. (1.237), we get

(es)


dvk = vj) ds
j |i (i{z k , (1.257)
}
(s)
=dvj


= i (i vj) j k ds, (1.258)


dv(es) = T D ds. (1.259)

Now, since i j is symmetric, we can be led to a useful result. Consider the series of
operations involving the velocity gradient, in general asymmetric, and a scalar quantity, :

= T L , (1.260)
= T (D + R) , (1.261)
= T D + |T {zR }, (1.262)
=0
T
= D . (1.263)

Thus, we can recast Eq. (1.259) as

dv(es) = T L ds. (1.264)

1.3.7.2.2.2 Shear straining What straining that is not aligned with the axis con-
necting P and P must then be normal to that axis, and is easily visualized to represent a
shearing between the two points. Hence the shear straining is
(ss) (s) (es)
dvj = dvj dvj , (1.265)


= (j vi) i i (i vk) k j ds, (1.266)


= (j vi) i p (p vk) k ji i ds, (1.267)



= (j vi) p (p vk) k ji i ds, (1.268)



dv(ss) = D T D I ds. (1.269)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

58 CHAPTER 1. GOVERNING EQUATIONS

1.3.7.2.2.3 Principal axes of strain rate We recall from our earlier discussion that
the principal axes of stress are those axes for which the force associated with a given axis
points in the same direction as that axis. We can extend this idea to straining, but develop it
in a slightly different, but ultimately equivalent fashion based on notions from linear algebra.
We first recall that most22 arbitrary asymmetric square matrices L can be decomposed into
a diagonal form as follows:

L = P P1 . (1.270)

Here P is a matrix of the same dimension as L which has in its columns the right eigenvectors
of L. When L is symmetric, it can be shown that its eigenvalues are guaranteed to be real
and its eigenvectors are guaranteed to be orthogonal. Further, since the eigenvectors can
always be scaled by a constant and remain eigenvectors, we can choose to scale them in such
a way that they are all normalized. In such a case in which the matrix P has orthonormal
columns, the matrix is defined as orthogonal (though orthonormal would be a more accurate
nomenclature). When P has been rendered orthogonal, we call it Q. So, when L is symmetric,
such as when L = D, the symmetric part of the velocity gradient, we also have the following
decomposition

D = Q Q1 . (1.271)

Orthogonal matrices can be shown to have the remarkable property that their transpose is
equal to their inverse, and so we also have the even more useful

D = Q QT . (1.272)

Geometrically Q is equivalent to a matrix of direction cosines; as we have seen before, its

transpose QT is a rotation matrix which rotates but does not stretch a vector when it operates
on the vector.
Now let us consider the straining component of the velocity difference; taking the sym-
metric (i vj) = D, which we further assume to be a constant for this analysis, we rewrite
Eq. (1.239) using Gibbs notation as

T
dv(s) = dxT D, (1.273)
(s) T
dv = D dx, (1.274)
(s)
dv = D dx, since D is symmetric. (1.275)
(s) T
dv = Q Q dx. (1.276)
22
Some matrices, which often do not have enough linearly independent eigenvectors, cannot be diagonal-
ized; however, the argument can be extended through use of the singular value decomposition. The singular
value decomposition can also be used to effectively diagonalize asymmetric matrices; however, in that case
it can be shown there is no equivalent interpretation of the principal axes. Consequently, we will quickly
focus the discussion on symmetric matrices.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.3. KINEMATICS 59

Now let us select what amounts to a special axes rotation via matrix multiplication by the
orthogonal matrix QT :

QT dv(s) = QT Q QT dx, (1.277)

QT dv(s) = 1 T
Q Q Q dx, (1.278)
QT dv(s) = T
Q dx, (1.279)



d QT v(s) = d Q xT T
since D and thus Q are assumed constant. (1.280)
| {z } | {z }
=v(s) =x

Now we recall from the definition of vectors that QT v(s) = v (s) and QT x = x . That is
these are the representations of the vectors in a specially rotated coordinate system, so we
have
(s)
dv = dx . (1.281)

Now since is diagonal, we see that a perturbation in x confined to any one of the rotated
coordinate axes induces a change in velocity which lies in the same direction as that coordi-
(s)
nate axis. For instance on the 1 axis, we have dv 1 = 11 dx 1 . That is to say that in this
specially rotated frame, all straining is extensional; there is no shear straining.

1.3.8 Expansion rate

Consider a small material region of fluid, also called a particle of fluid. We define a material
region as a region enclosed by a surface across which there is no flux of mass. We shall
later see by invoking the mass conservation axiom for a non-relativistic system, that the
implication is that the mass of a material region is constant, but we need not yet consider
this. In general the volume containing this particle can increase or decrease. It is useful
to quantify the rate of this increase or decrease. Additionally, this will give a flavor of the
analysis to come for the conservation axioms.
Taking the both MR and R(t) to denote the same time-dependent finite material region
in space, we must have Z
VM R = dV. (1.282)
R(t)

Using Leibnizs rule, Eq. (1.178), we take the time derivative of both sides and obtain
Z Z
dVM R
= (1) dV + ni vi dS, (1.283)
dt R(t) t S(t)
Z
= ni vi dS, (1.284)
S(t)
Z
= i vi dV by Gausss theorem, (1.285)
R(t)
= (i vi ) VM R by the mean value theorem. (1.286)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

60 CHAPTER 1. GOVERNING EQUATIONS

In the analysis above, we note that the velocity of S(t), in general wi , has been set to the
fluid velocity vi since we have a material region. We also recall from calculus the mean value
theorem which R b states that for any integral, a mean value can be defined, denoted by a , as
for example a f (x) dx = f (b a). As we shrink the size of the material volume to zero,
the mean value approaches the local value, so we get
1 dVM R
= (i vi ) , (1.287)
VM R dt
1 dVM R
lim = i vi = T v = div v = tr D. (1.288)
VM R 0 VM R dt
Equation (1.288)describes the relative expansion rate also known as the dilation rate of a
material fluid particle. A fluid particle for which i vi = 0 must have a relative expansion
rate of zero, and satisfies conditions to be an incompressible fluid.

1.3.9 Invariants of the strain rate tensor

The tensor associated with straining (also called the deformation rate tensor or strain rate
(i)
tensor) (i vj) is symmetric. Consequently, it has three real eigenvalues, , and an orienta-
tion for which the strain rate is aligned with the eigenvectors. As with stress, there are also
three principal invariants of strain rate, namely
(1) (1) (2) (3)
I = (i vi) = i vi = + + , (1.289)
(2) 1 (1) (2) (2) (3) (3) (1)
I = ((i vi) (j vj) (i vj) (j vi) ) = + + , (1.290)
2
(3) (1) (2) (3)
I = ijk (1 vi) (2 vj) (3 vk) = . (1.291)
(1)
The physical interpretation for I is obvious in that it is equal to the relative rate of volume
(2) 2 (3)
change for a material element, V1 dVdt
. Aris discusses how I is related to V1 ddtV2 and I is
3
related to V1 ddtV3 .

1.3.10 Invariants of the velocity gradient tensor

For completeness, the invariants of the more general velocity gradient tensor are included.
They are

(1) (1) (2) (3)

Iv = i vi = v + v + v , (1.292)
(2) 1 (1) (2) (2) (3) (3) (1)
Iv = ((i vi )(j vj ) (i vj )(j vi )) = v v + v v + v v , (1.293)
2
1

= (i vi )(j vj ) + [i vj] [i vj] (i vj) (i vj) , (1.294)
2 
1 1
= (i vi )(j vj ) + i i (i vj) (i vj) , (1.295)
2 2

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.3. KINEMATICS 61

(3) (1) (2) (3)

Iv = ijk 1 vi 2 vj 3 vk = v v v . (1.296)

1.3.11 Two-dimensional kinematics

Next, consider some important two dimensional cases, first for general two-dimensional flows,
and then for specific examples.

1.3.11.1 General two-dimensional flows

For two-dimensional motion, we have the velocity vector as (v1 , v2 , v3 = 0), and for the unit
tangent of the vector separating two nearby particles (1 , 2 , 3 = 0).

1.3.11.1.1 Rotation Recalling that dxi = i ds, for rotation, we have

(r)
dvj = [i vj] dxi = i [i vj] ds, (1.297)
(r)

dvj = 1 [1 vj] + 2 [2 vj] ds, (1.298)

(r)
dv1 = 1 [1 v1] +2 [2 v1] ds, (1.299)
| {z }
=0
(r)
dv1 = 2 [2 v1] ds, (1.300)

(r)
dv2 = 1 [1 v2] + 2 [2 v2] ds, (1.301)
| {z }
=0
(r)
dv2 = 1 [1 v2] ds, (1.302)
rewriting in terms of the actual derivatives
(r) 1
dv1 = 2 (2 v1 1 v2 ) ds, (1.303)
2
(r) 1
dv2 = 1 (1 v2 2 v1 ) ds. (1.304)
2

Also for the vorticity vector, we get

k = ijk i vj . (1.305)

The only non-zero component is 3 , which comes to

3 = 311 1 v1 + 312 1 v2 + 321 2 v1 + 322 2 v2 , thus, (1.306)

|{z} |{z} |{z} |{z}
=0 =1 =1 =0
= 1 v2 2 v1 . (1.307)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

62 CHAPTER 1. GOVERNING EQUATIONS

1.3.11.1.2 Extension
(es)
dvk = k i j (i vj) ds, (1.308)


= k 1 1 (1 v1) + 1 2 (1 v2) + 2 1 (2 v1) + 2 2 (2 v2) ds (1.309)


= k 12 1 v1 + 1 2 (1 v2 + 2 v1 ) + 22 2 v2 ds, so (1.310)
(es)

dv1 = 1 12 1 v1 + 1 2 (1 v2 + 2 v1 ) + 22 2 v2 ds, (1.311)
(es)

dv2 = 2 12 1 v1 + 1 2 (1 v2 + 2 v1 ) + 22 2 v2 ds. (1.312)

1.3.11.1.3 Shear
(ss) (s) (es)
dvj = dvj dvj , (1.313)


= i (i vj) j i k (i vk) ds, (1.314)
  
(ss) 2 v1 + 1 v2
dv1 = 1 1 v1 + 2
2



1 12 1 v1 + 1 2 (1 v2 + 2 v1 ) + 22 2 v2 ds,
  
(ss) 1 v2 + 2 v1
dv2 = 2 2 v2 + 1
2



2 12 1 v1 + 1 2 (1 v2 + 2 v1 ) + 22 2 v2 ds.

1.3.11.1.4 Expansion
1 dV
= 1 v1 + 2 v2 . (1.315)
V dt

1.3.11.2 Relative motion along 1 axis

Let us consider in detail the configuration shown in Figure 1.10 in which the particle sepa-
ration is along the 1 axis. Hence 1 = 1, 2 = 0, and 3 = 0.
Rotation

(r)
dv1 = 0, (1.316)
(r) 1 3
dv2 = (1 v2 2 v1 ) ds = ds. (1.317)
2 2

Extension

(es)
dv1 = 1 v1 ds, (1.318)
(es)
dv2 = 0. (1.319)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.3. KINEMATICS 63

x
2

v i (P)
v i (P)

P ds
P

x
1

Figure 1.10: Sketch of fluid particle P in motion with velocity vi (P ) and nearby neighbor
particle P with velocity vi (P ).

Shear
(ss)
dv1 = 0, (1.320)
(ss) 1
dv2 = (1 v2 + 2 v1 ) ds = (1 v2) ds. (1.321)
2

1 dV
Expansion: V dt
= 1 v1 + 2 v2 .

1.3.11.3 Relative motion along 2 axis

Let us consider in detail the configuration shown in Figure 1.11 in which the particle sepa-
ration is aligned with the 2 axis. Hence 1 = 0, 2 = 1, and 3 = 0.

Rotation

(r) 1 3
dv1 = (2 v1 1 v2 ) ds = ds, (1.322)
2 2
(r)
dv2 = 0. (1.323)

Extension
(es)
dv1 = 0, (1.324)
(es)
dv2 = 2 v2 ds. (1.325)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

64 CHAPTER 1. GOVERNING EQUATIONS

x
2
v i (P)

P
ds
P v i (P)

x
1

Figure 1.11: Sketch of fluid particle P in motion with velocity vi (P ) and nearby neighbor
particle P with velocity vi (P ).

Shear
(ss) 1
dv1 = (2 v1 + 1 v2 ) ds = (1 v2) ds, (1.326)
2
(ss)
dv2 = 0. (1.327)

Expansion
1 dV
= 1 v1 + 2 v2 . (1.328)
V dt

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.3. KINEMATICS 65

x
2

v
j

x
1

Figure 1.12: Sketch of uniform flow

1.3.11.4 Uniform flow

Consider the kinematics of a uniform two-dimensional flow in which

v1 = k1 , v2 = k2 , v3 = 0, (1.329)

as sketched in Figure 1.12.

 
dx1 dx2 dx1 dx2 k1
Streamlines: v1
= v2
, k1
= k2
, x1 = k2
x2 + C.

Rotation: 3 = 1 v2 2 v1 = 1 (k1 ) 2 (k2 ) = 0.

Extension

on 1-axis: 1 v1 = 0.
on 2-axis: 2 v2 = 0.
1
Shear for unrotated element: 2
(1 v2 + 2 v1 ) = 0.

Expansion: 1 v1 + 2 v2 = 0.

Acceleration:
dv1
dt
= o v1 + v1 1 v1 + v2 2 v1 = 0 + k1 1 (k1 ) + k2 2 (k1 ) = 0.
dv2
dt
= o v2 + v1 1 v2 + v2 2 v2 = 0 + k1 1 (k2 ) + k2 2 (k2 ) = 0.

For this very simple flow, the streamlines are straight lines, there is no rotation, no
extension, no shear, no expansion, and no acceleration.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

66 CHAPTER 1. GOVERNING EQUATIONS

x
2

x
1

Figure 1.13: Sketch of pure rigid body rotation.

1.3.11.5 Pure rigid body rotation

Consider the kinematics of a two-dimensional flow in which

v1 = kx2 , v2 = kx1 , v3 = 0, (1.330)

as sketched in Figure 1.13.

dx1 dx2 dx1 dx2
Streamlines: v1
= v2
, kx2
= kx1
, x1 dx1 = x2 dx2 , x21 + x22 = C.
Rotation: 3 = 1 v2 2 v1 = 1 (kx1 ) 2 (kx2 ) = 2k.
Extension
on 1-axis: 1 v1 = 0.
on 2-axis: 2 v2 = 0.
1
Shear for unrotated element: 2
(1 (kx1 ) + 2 (kx2 ) = k k = 0.
Expansion: 1 v1 + 2 v2 = 0 + 0 = 0.

Acceleration:
dv1
dt
= o v1 + v1 1 v1 + v2 2 v1 = 0 kx2 1 (kx2 ) + kx1 2 (kx2 ) = k 2 x1 .
dv2
dt
= o v2 + v1 1 v2 + v2 2 v2 = 0 kx2 1 (kx1 ) + kx1 2 (kx1 ) = k 2 x2 .

In this flow, the velocity magnitude grows linearly with distance from the origin. This
is precisely how a rotating rigid body behaves. The streamlines are circles. The rotation is
positive for positive k, hence counterclockwise, there is no deformation in extension or shear,
and there is no expansion. The acceleration is pointed towards the origin.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.3. KINEMATICS 67

x
2

v
j

x
1

Figure 1.14: Sketch of extensional flow (1-D compressible)

1.3.11.6 Pure extensional motion (a compressible flow)

Consider the kinematics of a two-dimensional flow in which
v1 = kx1 , v2 = 0, v3 = 0, (1.331)
as sketched in Figure 1.14.
dx1 dx2
Streamlines: v1
= v2
, v2 dx1 = v1 dx2 , 0 = kx1 dx2 , x2 = C.
Rotation: 3 = 1 v2 2 v1 = 1 (0) 2 (kx1 ) = 0.
Extension
on 1-axis: 1 v1 = k.
on 2-axis: 2 v2 = 0.
1 1
Shear for unrotated element: 2
(1 v2 + 2 v1 ) = 2
(1 (0) + 2 (kx1 )) = 0.
Expansion: 1 v1 + 2 v2 = k.
Acceleration:
dv1
dt
= o v1 + v1 1 v1 + v2 2 v1 = 0 + kx1 1 (kx1 ) + 02 (kx1 ) = k 2 x1 .
dv2
dt
= o v2 + v1 1 v2 + v2 2 v2 = 0 + kx1 1 (0) + 02 (0) = 0.
In this flow, the streamlines are straight lines, there is no fluid rotation, there is exten-
sion (stretching) deformation along the 1-axis, but no shear deformation along this axis.
The relative expansion rate is positive for positive k, indicating a compressible flow. The
acceleration is confined to the x1 direction.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

68 CHAPTER 1. GOVERNING EQUATIONS

x
2

v
j

x
1

Figure 1.15: Sketch of pure shearing flow

1.3.11.7 Pure shear straining

Consider the kinematics of a two-dimensional flow in which
v1 = kx2 , v2 = kx1 , v3 = 0, (1.332)
as sketched in Figure 1.15.
dx1 dx2 dx1 dx2
Streamlines: v1
= v2
, kx2
= kx1
, x1 dx1 = x2 dx2 , x21 = x22 + C.
Rotation: 3 = 1 v2 2 v1 = 1 (kx1 ) 2 (kx2 ) = k k = 0.
Extension
on 1-axis: 1 v1 = 1 (kx2 ) = 0.
on 2-axis: 2 v2 = 2 (kx1 ) = 0.
1 1
Shear for unrotated element: 2
(1 v2 + 2 v1 ) = 2
(1 (kx1 ) + 2 (kx2 )) = k.
Expansion: 1 v1 + 2 v2 = 0.
Acceleration:
dv1
dt
= o v1 + v1 1 v1 + v2 2 v1 = 0 + kx2 1 (kx2 ) + kx1 2 (kx2 ) = k 2 x1 .
dv2
dt
= o v2 + v1 1 v2 + v2 2 v2 = 0 + kx2 1 (kx1 ) + kx1 2 (kx1 ) = k 2 x2 .
In this flow, the streamlines are hyperbolas, there is no rotation, no axial extension along
the coordinate axes, positive shear deformation for an element aligned with the coordinate
axes, and no expansion. So, the pure shear deformation preserves volume. The fluid is
accelerating away from the origin.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.3. KINEMATICS 69

x
2
v
j

x
1

Figure 1.16: Sketch of Couette flow.

1.3.11.8 Couette flow: shear + rotation

Consider the kinematics of a two-dimensional flow in which
v1 = kx2 , v2 = 0, v3 = 0, (1.333)
as sketched in Figure 1.16. This is known as a Couette23 flow.
dx1 dx2 dx1 dx2
Streamlines: v1
= v2
, kx2
= 0
, 0 = kx2 dx2 , x2 = C.
Rotation: 3 = 1 v2 2 v1 = 1 (0) 2 (kx2 ) = k.
Extension
on 1-axis: 1 v1 = 1 (kx2 ) = 0.
on 2-axis: 2 v2 = 2 (0) = 0.
1 1
Shear for unrotated element: 2
(1 v2 + 2 v1 ) = 2
(1 (0) + 2 (kx2 )) = k2 .
Expansion: 1 v1 + 2 v2 = 0.
Acceleration:
dv1
dt
= o v1 + v1 1 v1 + v2 2 v1 = 0 + kx2 1 (kx2 ) + 02 (kx2 ) = 0.
dv2
dt
= o v2 + v1 1 v2 + v2 2 v2 = 0 + kx2 1 (0) + 02 (0) = 0.
Here the streamlines are straight lines, and the flow is rotational (clockwise since < 0 for
k > 0)! The constant volume rotation is combined with a constant volume shear deformation
for the element aligned with the coordinate axes. The fluid is not accelerating.
23
Maurice Marie Alfred Couette, 1858-1943, French fluid mechanician, student of
Joseph Valentin Boussinesq, and faculty member at Catholic University of Angers.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

70 CHAPTER 1. GOVERNING EQUATIONS

x
2

vj

x
1

Figure 1.17: Sketch of ideal irrotational vortex.

1.3.11.9 Ideal point vortex: extension+shear

Consider the kinematics of a two-dimensional flow sketched in Figure 1.17.
x2 x1
v1 = k , v2 = k , v3 = 0, (1.334)
x21 + x22 x21 + x22

dx1 dx2 dx1 dx2

Streamlines: v1
= v2
, k
x2 = k
x1 , dx
x2
1
= dx2
x1
, x21 + x22 = C.
x2 2
1 +x2 x2 2
1 +x2
   
Rotation: 3 = 1 v2 2 v1 = 1 k x2x+x
1
2 2 k x2
x2 +x2
= 0.
1 2 1 2

Extension
 
on 1-axis: 1 v1 = 1 k x2x+x2
2 = 2k (x2x+x
1 x2
2 )2 .
1 2 1 2
 
x1 x1 x2
on 2-axis: 2 v2 = 2 k x2 +x2 = 2k (x2 +x2 )2 .
1 2 1 2

1 x2 x2
Shear for unrotated element: 2
(1 v2 + 2 v1 ) = k (x22+x21)2 .
1 2

Expansion: 1 v1 + 2 v2 = 0.
Acceleration:
dv1 2
dt
= o v1 + v1 1 v1 + v2 2 v1 = (x2k+x
x1
2 )2
1 2
dv2 2
dt
= o v2 + v1 1 v2 + v2 2 v2 = (x2k+x
x2
2 )2 .
1 2

The streamlines are circles and the fluid element does not rotate about its own axis! It does
rotate about the origin. It deforms by extension and shear in such a way that overall the
volume is constant.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.3. KINEMATICS 71

1.3.12 Kinematics as a dynamical system

Let us apply some standard notions from dynamical systems theory to fluid kinematics. Let
us imagine that we are given a time-independent flow field, where the fluid velocity is known
and is a function of position only. Then the motion of an individual fluid particle is governed
by the following autonomous system of non-linear ordinary differential equations:

dx
= v(x(t)), x(0) = X. (1.335)
dt
Here, the initial position of the fluid particle is given by the constant vector X. The solution
of Eq. (1.335) can be expressed in general form

x = x(t; X), (1.336)

a function of time parameterized by the initial condition of the fluid particle. Such a solution
is certainly a pathline, streamline, and streakline. It is also known as a trajectory in the
dynamical systems literature.

Example 1.9
As an example, we could study the following non-linear autonomous system:

dx1
= v1 (x1 , x2 , x3 ) = 1 + x1 x2 x3 , x1 (0) = 1, (1.337)
dt
dx2
= v2 (x1 , x2 , x3 ) = x1 + x22 + x1 x32 , x2 (0) = 1, (1.338)
dt
dx3
= v3 (x1 , x2 , x3 ) = 2 x1 + x2 x3 , x3 (0) = 4. (1.339)
dt

Numerical solution would yield x1 (t), x2 (t), x3 (t), which for this time-independent velocity field are
the particle pathlines, streamlines, and streaklines. We could also apply the complete mathematical
theory of dynamic systems to understand the system better.

Let us analyze Eq. (1.335) in some more detail. From the chain rule, see Eq. (1.229), we
have

dv = (vT )T dx, (1.340)

| {z }
LT
T
dv = L dx, (1.341)
dv dx
= LT , (1.342)
dt dt
= LT v. (1.343)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

72 CHAPTER 1. GOVERNING EQUATIONS

Now, we seek to analyze a particular pathline. Note that the velocity vector is tangent to
the fluid particle trajectory. Let us study a unit vector which happens to be tangent to the
velocity field:
v
t = . (1.344)
|v|
Next, use the chain rule to examine how the unit tangent vector evolves with time:
dt 1 dv v d|v|
= . (1.345)
dt |v| dt |v|2 dt
We can scale Eq. (1.343) by |v| to get (1/|v|)dv/dt = LT v/|v| = LT t . Thus Eq. (1.345)
can be rewritten as
dt v d|v|
= LT t , (1.346)
dt |v|2 dt
1 d|v|
= LT t t . (1.347)
|v| dt
Next consider the following series of operations starting with Eq. (1.343):
dv
= LT v, (1.348)
dt
dv
vT = vT LT v, (1.349)
 dt

d vT v
= vT LT v, (1.350)
dt 2
 2
d |v|
= vT LT v, (1.351)
dt 2
d
|v| (|v|) = vT LT v, (1.352)
dt
1 d vT T v
(|v|) = L , (1.353)
|v| dt |v| |v|
1 d
(|v|) = Tt LT t . (1.354)
|v| dt
Now substitute Eq. (1.354) into Eq. (1.347) to get
dt

= LT t Tt LT t t . (1.355)
dt
As an aside, take the dot product of Eq. (1.355) with t to get
dt

Tt = Tt LT t T LT Tt t , (1.356)
dt | {z }
=1
= Tt T
L t Tt T
L t , (1.357)
= 0. (1.358)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.3. KINEMATICS 73

This must be an identity, because Tt t = 1, and its time derivative gives Tt d/dt = 0.
Now recalling Eq. (1.233), and employing Tt RT t = 0, because of the antisymmetry
of R, and DT = D, because of the symmetry of D, Eq. (1.355) can be rewritten as
dt

= LT t Tt D t t . (1.359)
dt
Let us consider how a volume stretches in a direction aligned with the velocity vector.
We first specialize the general differential arc length to that found along the particle path:
ds = ds. Now, recall from geometry that the square of the differential arc length must be
ds2 = dxT dx, (1.360)
where dx is also confined to the particle path. Consider now how this quantity changes with
time when we move with the particle:
d d

(ds)2 = dxT dx , (1.361)
dt dt
 T
T d d
= dx (dx) + (dx) dx, (1.362)
dt dt
    T
T dx dx
= dx d + d dx, (1.363)
dt dt
= dxT dv + dvT dx, (1.364)
T
= 2dx dv, (1.365)
T T
= 2dx L dx, (1.366)
d
2ds (ds) = 2dxT LT dx, (1.367)
dt
1 d dx T T dx
(ds) = L . (1.368)
ds dt ds ds
Recall now that
v
t = , (1.369)
|v|
dx
dt
= ds
, (1.370)
dt
dx
= . (1.371)
ds
So, Eq. (1.368) can be rewritten as
1 d
(ds) = Tt LT t , (1.372)
ds dt
d
(ln ds) = Tt (D + R)T t , (1.373)
dt
= Tt D t , (1.374)
= D : t Tt . (1.375)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

74 CHAPTER 1. GOVERNING EQUATIONS

Note that this relative tangential stretching rate is closely related to the result of Eq. (1.259)
for extensional strain rate. Specializing Eq. (1.259) for a particle pathline, and combining,
we can say

dv(es) = (Tt D t )t ds, (1.376)

dv(es)
= (Tt D t )t , (1.377)
ds
dv(es)
Tt = (Tt D t ) Tt t , (1.378)
ds | {z }
=1
1 d
= Tt D t = (ds), (1.379)
ds dt 
1 ds
= Tt D t = d , (1.380)
ds dt
d|v|
= Tt D t = , (1.381)
ds
d|v|
= D : t Tt = . (1.382)
ds
Here, we invoked Eq. (1.374) to obtain Eq. (1.380). The quantity Tt D t = D : t Tt
is a measure of how the magnitude of the velocity changes with respect to arc length along
the particle path.
We can gain further insight into how velocity magnitude changes by a diagonal decom-
position of D = Q QT , where Q is an orthogonal rotation matrix with the normalized
eigenvectors of D in its columns, and is the diagonal matrix with the eigenvalues of D in
its diagonal. Thus
d|v|
= Tt Q
|
{z QT} t , (1.383)
ds
D
= (QT t )T (QT t ), (1.384)
(1.385)

The operation QT t s generates a new rotated unit vector s = (s1 , s2 , s3)T . Thus
we can state
d|v| 2 2 2
= s1 1 + s2 2 + s3 3 , (1.386)
ds
2 2 2
1 = s1 + s2 + s3 . (1.387)

The rate of change of the velocity magnitude along a particle pathline can be understood
to be a weighted average of the eigenvalues of the deformation tensor D. In the very special
case in which t is the ith eigenvector of D, we simply get d|v|/ds = i , where i is the
corresponding eigenvalue.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.3. KINEMATICS 75

Note now that if we extend Eq. (1.288) to differential material volumes, we could say the
relative expansion rate is
1 d
(dV ) = tr D, (1.388)
dV dt
d
(ln dV ) = tr D. (1.389)
dt
Now our differential volume can be formed by
dV = dA ds, (1.390)
where dA is the cross-sectional area normal to the flow direction. Thus
ln dV = ln dA + ln ds, (1.391)
ln dA = ln dV ln ds, (1.392)
d d d
(ln dA) = (ln dV ) (ln ds) , (1.393)
dt dt dt
(1.394)
Substitute from Eqs. (1.374,1.389) to get the relative rate of change of the differential area
normal to the flow direction:
d
(ln dA) = tr D Tt D t . (1.395)
dt
Note this relation, while not identical, is similar to the expression for shear strain rate,
Eq. (1.269). We can also use Eq. (1.65) to rewrite Eq. (1.395) as
d
(ln dA) = D : I D : t Tt , (1.396)
dt

= D : I t Tt . (1.397)
Now the matrix I t Tt has some surprising properties. It is singular and has rank two.
Because it is symmetric, it has a set of three orthogonal eigenvectors which can be normalized
to form an orthonormal set. Its three eigenvalues are 1, 1, and 0. Remarkably, the eigenvector
associated with the zero eigenvalue must be parallel to and can be selected as t , the unit
tangent to the curve. Thus the other two eigenvectors can be thought of as unit normals
to the curve, which we label n1 and n2 . Note that these eigenvectors are not unique;
however, a set can always be found. We can summarize the decomposition in the following
steps:
I t Tt = Q QT , (1.398)
. .. ..
.. . . 1 0 0 Tn1

= n1 n2 t 0 1 0 Tn2 , (1.399)
.. .. .. 0 0 0 Tt
. . .
= n1 Tn1 + n2 Tn2 . (1.400)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

76 CHAPTER 1. GOVERNING EQUATIONS

Note that the two unit normals are orthogonal to each other, Tn1 n2 = 0. Thus, we have

d

(ln dA) = D : n1 Tn1 + n2 Tn2 , (1.401)
dt
= D : n1 Tn1 + D : n2 Tn2 , (1.402)
= Tn1 D n1 + Tn2 D n2 . (1.403)

Comparing to Eq. (1.374) which has one mode associated with t available for stretching of
the one-dimensional arc length in the streamwise direction, there are two modes associated
with n1 , n2 available for stretching the two-dimensional area.
The form Tn1 D n1 suggests it determines the relative normal stretching rate in the
direction of n1 ; a similar rate exists for the other normal direction. One might imagine
that there exists a normal direction which yields extreme values for relative normal stretching
rates. It is easily shown this achieved by the following. First, define a rectangular matrix,
b whose columns are populated by n1 and n2 :
Q,
. ..
.. .

b = n1
Q n2 . (1.404)
.. ..
. .

b associated with
Then project the 3 3 matrix D onto this basis to form the 2 2 matrix D
stretching in the directions normal to the motion:

b=Q
D b T D Q.
b (1.405)

b give the maximum and minimum values of the relative normal stretching
The eigenvalues of D
rates, and the eigenvectors give the associated directions of extremal normal stretching.
Looked at another way and motivated by standard results from differential geometry,
we can make special choices, n1 = np , n2 = nb , where np is the so-called principal
normal unit vector and nb is the so-called bi-normal unit vector. The following results
are described in more detail in many sources, e.g. Sen and Powers.24 We have the so-called
Frenet-Serret25 relations:

dt
= np , (1.406)
ds
dnp
= t nb , (1.407)
ds
dnb
= np . (1.408)
ds
24
Sen, M., and Powers, J. M., 2012, Lecture Notes in Mathematical Methods, Chapter 6.
25
Jean Frederic Frenet, 1816-1900, and Joseph Alfred Serret, 1819-1885, French mathematicians.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.4. CONSERVATION AXIOMS 77

Here is the so-called curvature, of the curve and is the so-called torsion of the curve.
One can, with effort show that and are given by
r
 d2 x 2  dx 2  dx T d2 x 2 
 dx d2 x 

 2    dt dt2 
dt dt dt dt2
=  dx 3 =  3 , (1.409)
   dx 
dt dt
 2
T 3
dx dt
ddt2x ddt3x
=      2 (1.410)
 d2 2x 2  dx 2 dx T d2 2x
dt dt dt dt

Note and are expressed here as functions of time. This certainly the case for a particle
moving along a path in time. But just as the intrinsic curvature of a mountain road is
independent of the speed of the vehicle traveling on the road, despite the traveling vehicle
experiencing a time-dependency of curvature, the curvature and torsion can be considered
more fundamentally to be functions of position only, given that the velocity field is known
as a function of position. Analysis reveals in fact that
q
(vT L LT v) (vT v) (vT LT v)2
= (1.411)
(vT v)3/2
One could also develop an expression for torsion which is explicitly dependent on position.
The expression is complicated and requires the use of third order tensors to capture the
higher order spatial variations.
We can also use this intrinsic orthonormal basis to get
d

(ln dA) = D : np Tnp + nb Tnb , (1.412)
dt
= D : np Tnp + D : nb Tnb , (1.413)
= Tnp D np + Tnb D nb . (1.414)

1.4 Conservation axioms

A fundamental goal of this section is to take the verbal notions which embody the basic
axioms of non-relativistic continuum mechanics into usable mathematical expressions. First,
we must list those axioms. The axioms themselves are simply principles which have been
observed to have wide validity as long as the particle velocity is small relative to the speed
of light and length scales are sufficiently large to contain many molecules. Many of these
axioms can be applied to molecules as well. The axioms cannot be proven. They are simply
statements which have been useful in describing the universe.
A summary of the axioms in words is as follows
Mass conservation principle: The time rate of change of mass of a material region is
zero.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

78 CHAPTER 1. GOVERNING EQUATIONS

Linear momenta principle: The time rate of change of the linear momenta of a material
region is equal to the sum of forces acting on the region. This is Eulers generalization
of Newtons second law of motion.

Angular momenta principle: The time rate of change of the angular momenta of a
material region is equal to the sum of the torques acting on the region. This was first
formulated by Euler.

Energy conservation principle: The time rate of change of energy within a material
region is equal to the rate that energy is received by heat and work interactions. This
is the first law of thermodynamics.

Entropy inequality: The time rate of change of entropy within a material region is
greater than or equal to the ratio of the rate of heat transferred to the region and the
absolute temperature of the region. This is the second law of thermodynamics.

Some secondary concepts related to these axioms are as follows

The local stress on one side of a surface is identically opposite that stress on the
opposite side.

Stress can be separated into thermodynamic and viscous stress.

Forces can be separated into surface and body forces.

In the absence of body couples, the angular momenta principle reduces to a nearly
trivial statement.

The energy equation can be separated into mechanical and thermal components.

Next we shall systematically convert these words above into mathematical form.

1.4.1 Mass
The mass conservation axiom is simple to state mathematically. It is

d
mM R(t) = 0. (1.415)
dt
Here MR(t) stands for a material region which can evolve in time, and mM R(t) is the mass
in the material region. A relevant material region is sketched in Figure 1.18. We can define
the mass of the material region based upon the local value of density:
Z
mM R(t) = dV. (1.416)
M R(t)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.4. CONSERVATION AXIOMS 79

dV

dS
w i = vi

ni

Figure 1.18: Sketch of finite material region MR, infinitesimal mass element dV , and
infinitesimal surface element dS with unit normal ni , and general velocity wi equal to fluid
velocity vi .

So, the mass conservation axiom is

Z
d
dV = 0. (1.417)
dt M R(t)
R R R
Recalling Leibnizs rule, Eq. (1.178), dtd R(t) [ ]dV = R(t) t
[ ]dV + S(t) ni wi [ ]dS, we take
the arbitrary velocity wi = vi as we are considering a material region so we get
Z Z Z
d
dV = dV + ni vi dS = 0. (1.418)
dt M R(t) M R(t) t M S(t)
R R
Now we invoke Gausss theorem, Eq. (1.167) R(t) i [ ]dV = S(t) ni [ ]dS, to convert a surface
integral to a volume integral to get the mass conservation axiom to read as
Z Z

dV + i (vi )dV = 0, (1.419)
M R(t) t M R(t)
Z  

+ i (vi ) dV = 0. (1.420)
M R(t) t
Now, in an important step, we realize that the only way for this integral, which has absolutely
arbitrary limits of integration, to always be zero, is for the integrand itself to always be zero.
Hence, we have

+ i (vi ) = 0, (1.421)
t
which we will write in Cartesian index, Gibbs, and full notation in what we call conservative
or divergence form as
o + i (vi ) = 0, (1.422)
o + T (v) = 0, (1.423)
o + 1 (v1 ) + 2 (v2 ) + 3 (v3 ) = 0. (1.424)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

80 CHAPTER 1. GOVERNING EQUATIONS

There are several alternative forms for this axiom. Using the product rule, we can say also

o + vi i +i vi = 0, (1.425)
| {z }
material derivative of density

or, writing in what is called the non-conservative form,

d
+ i vi = 0, (1.426)
dt
d
+ T v = 0, (1.427)
dt
(o + v1 1 + v2 2 + v3 3 ) + (1 v1 + 2 v2 + 3 v3 ) = 0. (1.428)

So, we can also say

1 d
= i vi . (1.429)
dt |{z}
|{z} relative rate of particle volume expansion
relative rate of density increase

Thus the relative rate of density increase of a fluid particle is the negative of its relative rate
of expansion, as expected. So, we also have
1 d 1 dVM R
= , (1.430)
dt VM R dt
dVM R d
+ VM R = 0, (1.431)
dt dt
d
(VM R ) = 0, (1.432)
dt
d
(mM R ) = 0. (1.433)
dt
We note that in a relativistic system, in which mass-energy is conserved, but not mass, that
we can have a material region, that is a region bounded by a surface across which there is no
flux of mass, for which the mass can indeed change, thus violating our non-relativistic mass
conservation axiom.

1.4.2 Linear momenta

1.4.2.1 Statement of the principle
The linear momenta conservation axiom is simple to state mathematically. It is
Z Z Z
d
vi dV = fi dV + ti dS . (1.434)
dt M R(t) M R(t) M S(t)
| {z } | {z } | {z }
rate of change of linear momenta body forces surface forces

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.4. CONSERVATION AXIOMS 81

f i dV

vi dV

dS
w i = vi

t i ni

Figure 1.19: Sketch of finite material region MR, infinitesimal linear momenta element
vi dV , infinitesimal body force element fi dV , and infinitesimal surface element dS with
unit normal ni , surface traction ti and general velocity wi equal to fluid velocity vi .

Again MR(t) stands for a material region which can evolve in time. A relevant material
region is sketched in Figure 1.19. The term fi represents a body force per unit mass. An
example of such a force would be the gravitational force acting on a body, which when scaled
by mass, yields gi . The term ti is a traction, which is a vector representing force per unit
area. A major challenge of this section will be to express the traction in terms of what is
known as the stress tensor.
Consider first the left hand side, LHS, of the linear momenta principle
Z Z
LHS = o (vi )dV + nj vi vj dS, from Leibniz, (1.435)
M R(t) M S(t)
Z
= (o (vi ) + j (vj vi )) dV, from Gauss. (1.436)
M R(t)

So, the linear momenta principle is

Z Z Z
(o (vi ) + j (vj vi )) dV = fi dV + ti dS. (1.437)
M R(t) M R(t) M S(t)

These are all expressed in terms of volume integrals except for the term involving surface
forces.

1.4.2.2 Surface forces

The surface force per unit area is a vector we call the traction tj . It has the units of stress,
but it is not formally a stress, which is a tensor. The traction is a function of both position
xi and surface orientation nk : tj = tj (xi , nk ).

CC BY-NC-ND. 17 June 2016, J. M. Powers.

82 CHAPTER 1. GOVERNING EQUATIONS

II l
ni

t (nII )
i i

III S
n
i

t (nI)
i i
t (nIII)
i i nI
i

Figure 1.20: Sketch of pillbox element for stress analysis.

We intend to demonstrate the following: The traction can be stated in terms of a stress
tensor Tij as written below:

tj = ni Tij ,
tT = nT T,
t = TT n. (1.438)

The following excursions are necessary to show this.

Show force on one side of surface equal and opposite to that on the opposite side
Let us apply the principle of linear momenta to the material region is sketched in Figure
1.20. Here
we indicate the dependency of the traction on orientation by notation such
II
as ti ni . This does not indicate multiplication, nor that i is a dummy index here. In
Figure 1.20, the thin pillbox has width l, circumference s, and a surface area for the
circular region of S. Surface I is a circular region; surface II is the opposite circular
region, and surface III is the cylindrical side.
We apply the mean value theorem to the linear momenta principle for this region and
get

(o (vi ) + j (vj vi )) (S)(l) =

(fi ) (S)(l) + ti (nIi )S + ti (nII III
i )S + ti (ni )s(l). (1.439)

Now we let l 0, holding for now s and S fixed to obtain

0 = ti (nIi ) + ti (nII
i ) S (1.440)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.4. CONSERVATION AXIOMS 83

x t (-n ) S
3 i 1 1

-n1 x
n 2
i

t (-n ) S
i 2 2 -n t (n ) S
3 i i
-n
2

t (-n ) S
i 3 3

x
1

Figure 1.21: Sketch of tetrahedral element for stress analysis on an arbitrary plane.

Now letting S 0, so that the mean value approaches the local value, and taking
nIi = nII
i ni , we get a useful result

ti (ni ) = ti (ni ). (1.441)

At an infinitesimal length scale, the traction on one side of a surface is equal an opposite
that on the other. That is, there is a local force balance. This applies even if there is
velocity and acceleration of the material on a macroscale. On the microscale, surface
forces dominate inertia and body forces. This is a useful general principle to remember.
It the fundamental reason why microorganisms have very different propulsion systems
that macro-organisms: they are fighting different forces.
Study stress on arbitrary plane and relate to stress on coordinate planes
Now let us consider a rectangular parallelepiped aligned with the Cartesian axes which
has been sliced at an oblique angle to form a tetrahedron. We will apply the linear
momenta principle to this geometry and make a statement about the existence of a
stress tensor. The described material region is sketched in Figure 1.21. Let L be a
characteristic length scale of the tetrahedron. Also let four unit normals nj exist, one
for each surface. They will be n1 , n2 , n3 for the surfaces associated with each
coordinate direction. They are negative because the outer normal points opposite to
the direction of the axes. Let ni be the normal associated with the oblique face. Let
S denote the surface area of each face.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

84 CHAPTER 1. GOVERNING EQUATIONS

Now the volume of the tetrahedron must be of order L3 and the surface area of order
L2 . Thus applying the mean value theorem to the linear momenta principle, we obtain
the form

(inertia) (L)3 = (body forces) (L)3 + (surface forces) (L)2 . (1.442)

As before, for small volumes, L 0, the linear momenta principle reduces to

X
surface forces = 0. (1.443)

Applying this to the configuration of Figure 1.21, we get

0 = ti (ni )S + ti (n1 )S1 + ti (n2 )S2 + ti (n3 )S3 . (1.444)

But we know that tj (nj ) = tj (nj ), so

ti (ni )S = ti (n1 )S1 + ti (n2 )S2 + ti (n3 )S3 . (1.445)

Now it is not a difficult geometry problem to show that ni S = Si , so we get

ti (ni )S = n1 ti (n1 )S + n2 ti (n2 )S + n3 ti (n3 )S, (1.446)

ti (ni ) = n1 ti (n1 ) + n2 ti (n2 ) + n3 ti (n3 ). (1.447)

Now we can consider terms like ti is obviously a vector, and the indicator, for example
(n1 ), tells us with which surface the vector is associated. This is precisely what a tensor
does, and in fact we can say

ti (ni ) = n1 T1i + n2 T2i + n3 T3i . (1.448)

In shorthand, we can say the same thing with

ti = nj Tji , or equivalently tj = ni Tij , QED. (1.449)

Here Tij is the component of stress in the j direction associated with the surface whose
normal is in the i direction.

Consider pressure and the viscous stress tensor

Pressure is a familiar concept from thermodynamics and fluid statics. It is often
tempting and sometimes correct to think of the pressure as the force per unit area
normal to a surface and the force tangential to a surface being somehow related to
frictional forces. We shall see that in general, this view is too simplistic.
First recall from thermodynamics that what we will call p, the thermodynamic pressure,
is for a simple compressible substance a function of at most two intensive thermody-
namic variables, say p = f (, e), where e is the specific internal energy. Also recall

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.4. CONSERVATION AXIOMS 85

that the thermodynamic pressure must be a normal stress, as thermodynamics con-

siders formally only materials at rest, and viscous stresses are associated with moving
fluids.
To distinguish between thermodynamic stresses and other stresses, let us define the
viscous stress tensor ij as follows

ij = Tij + pij . (1.450)

Recall that Tij is the total stress tensor. We obviously also have

Tij = pij + ij . (1.451)

Note with this definition that pressure is positive in compression, while Tij and ij are
positive in tension. Let us also define the mechanical pressure, p(m) , as the negative of
the average normal surface stress
1 1
p(m) Tii = (T11 + T22 + T33 ). (1.452)
3 3

The often invoked Stokes assumption, which remains a subject of widespread mis-
understanding 150 years after it was first made, is often adopted for lack of a good
alternative in answer to a question which will be addressed later in this chapter. It
asserts that the thermodynamic pressure is equal to the mechanical pressure:
1
p = p(m) = Tii . (1.453)
3
Presumably a pressure measuring device in a moving flow field would actually measure
the mechanical pressure, and not necessarily the thermodynamic pressure, so it is im-
portant to have this issue clarified for proper reconciliation of theory and measurement.
It will be seen that Stokes assumption gives some minor aesthetic pleasure in certain
limits, but it is not well-established, and is more a convenience than a requirement
for most materials. It is the case that various incarnations of more fundamental ki-
netic theory under the assumption of a dilute gas composed of inert hard spheres give
rise to the conclusion that Stokes assumption is valid. At moderate densities, these
hard sphere kinetic theory models predict that Stokes assumption is invalid. However,
none of the common kinetic theory models is able to predict results from experiments,
which nevertheless also give indication, albeit indirect, that Stokes assumption is in-
valid. Kinetic theories and experiments which consider polyatomic molecules, which
can suffer vibrational and rotational effects as well, show further deviation from Stokes
assumption. It is often plausibly argued that these so-called non-equilibrium effects,
that is molecular vibration and rotation, which are only important in high speed flow
applications in which the flow velocity is on the order of the fluid sound speed, are the
mechanisms which cause Stokes assumption to be violated. Because they only are im-
portant in high speed applications, they are difficult to measure, though measurement

CC BY-NC-ND. 17 June 2016, J. M. Powers.

86 CHAPTER 1. GOVERNING EQUATIONS

of the decay of acoustic waves has provided some data. For liquids, there is little to no
theory, and the limited data indicates that Stokes assumption is invalid.
Now contracting Eq. (1.451), we get

Tii = pii + ii . (1.454)

Using the fact that ii = 3 and inserting Eq. (1.453) in Eq. (1.454), we find for a fluid
that obeys Stokes assumption that
1
Tii = Tii (3) + ii , (1.455)
3
0 = ii . (1.456)

That is to say, the trace of the viscous stress tensor is zero. Moreover, for a fluid which
obeys Stokes assumption we can interpret the viscous stress as the deviation from the
mean stress; that is, the viscous stress is a deviatoric stress:
1
Tij = Tkk ij + ij , (valid only if Stokes assumption holds)
|{z} |3 {z } |{z}
total stress deviatoric stress
mean stress
(1.457)
If Stokes assumption does not hold, then a portion of ij will also contribute to the
mean stress; that is, the viscous stress is not then entirely deviatoric.
Finally, let us note what the traction vector is when the fluid is static. For a static
fluid, there is no viscous stress, so ij = 0, and we have

Tij = pij , static fluid. (1.458)

We get the traction vector on any surface with normal ni by

tj = ni Tij = pni ij = pnj . (1.459)

Changing indices, we see ti = pni , that is the traction vector must be oriented in the
same direction as the surface normal; all stresses are normal to any arbitrarily oriented
surface.

1.4.2.3 Final form of linear momenta equation

We are now prepared to write the linear momenta equation in final form. Substituting
our expression for the traction vector, Eq. (1.449) into the linear momenta expression, Eq.
(1.437), we get
Z Z Z
(o (vi ) + j (vj vi ))dV = fi dV + nj Tji dS. (1.460)
M R(t) M R(t) M S(t)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.4. CONSERVATION AXIOMS 87

Using Gausss theorem, Eq. (1.167), to convert the surface integral into a volume integral,
and combining all under one integral sign, we get
Z
(o (vi ) + j (vj vi ) fi j Tji )dV = 0. (1.461)
M R(t)

Making the same argument as before regarding arbitrary material volumes, this must then
require that the integrand be zero (we actually must require all variables be continuous to
make this work), so we obtain
o (vi ) + j (vj vi ) fi j Tji = 0. (1.462)
Using then Tij = pij + ij , we get in Cartesian index, Gibbs26 , and full notation
o (vi ) + j (vj vi ) = fi i p + j ji , (1.463)

T
T
(v) + T (vvT ) = f p + T , (1.464)
t
o (v1 ) + 1 (v1 v1 ) + 2 (v2 v1 ) + 3 (v3 v1 ) = f1 1 p + 1 11 + 2 21 + 3 31 , (1.465)
o (v2 ) + 1 (v1 v2 ) + 2 (v2 v2 ) + 3 (v3 v2 ) = f2 2 p + 1 12 + 2 22 + 3 32 , (1.466)
o (v3 ) + 1 (v1 v3 ) + 2 (v2 v3 ) + 3 (v3 v3 ) = f3 3 p + 1 13 + 2 23 + 3 33 . (1.467)

The form above is known as the linear momenta principle cast in conservative or divergence
form. It is the first choice of forms for many numerical simulations, as discretizations of this
form of the equation naturally preserve the correct values of global linear momenta, up to
roundoff error.
However, there is a reduced, non-conservative form which makes some analysis and phys-
ical interpretation easier. Let us use the product rule to expand the linear momenta prin-
ciple, then rearrange it, and use mass conservation and the definition of material derivative
to rewrite the expression:
o vi + vi o + vi j (vj ) + vj j vi = fi i p + j ji , (1.468)
(o vi + vj j vi ) + vi ( o + j (vj ) ) = fi i p + j ji , (1.469)
| {z }
=0 by mass conservation
(o vi + vj j vi ) = fi i p + j ji , (1.470)
| {z }
dvi
= dt

dvi
= fi i p + j ji , (1.471)
dt
dv
T
= f p + T , (1.472)
dt
(o v1 + v1 1 v1 + v2 2 v1 + v3 3 v1 ) = f1 1 p + 1 11 + 2 21 + 3 31(, 1.473)
(o v2 + v1 1 v2 + v2 2 v2 + v3 3 v2 ) = f2 2 p + 1 12 + 2 22 + 3 32(, 1.474)
(o v3 + v1 1 v3 + v2 2 v3 + v3 3 v3 ) = f3 3 p + 1 13 + 2 23 + 3 33(. 1.475)
26
Here the transpose notation is particularly cumbersome and unfamiliar, though necessary for full con-

sistency. One will more commonly see this equation written simply as t (v) + (vv) = f p + .

CC BY-NC-ND. 17 June 2016, J. M. Powers.

88 CHAPTER 1. GOVERNING EQUATIONS

Figure 1.22: Sketch of particle of mass m velocity v rotating about an axis centered at
point O, with radial distance vector r.

So, we see that particles accelerate due to body forces and unbalanced surface forces. If the
surface forces are non-zero but uniform, they will have no gradient or divergence, and hence
not contribute to accelerating a particle.

1.4.3 Angular momenta

It is often easy to overlook the angular momenta principle, and its consequence is so simple
that, it is often just asserted without proof. In fact in classical rigid body mechanics, it
is redundant with the linear momenta principle. It is, however, an independent axiom for
continuous deformable media.
Let us first recall some notions from classical rigid body mechanics, while referring to the
sketch of Figure 1.22. We have the angular momenta vector L for the particle of Figure 1.22

L = r (mv). (1.476)

Any force F which acts on m with lever arm r induces a torque T which is

T = r F. (1.477)

Now let us apply these notions for an infinitesimal fluid particle with differential mass dV .

Angular momenta = r ( dV )v = ijk rj vk dV, (1.478)

Torque of body force = r f( dV ) = ijk rj fk dV, (1.479)
Torque of surface force = r t dS = ijk rj tk dS,
= r (nT T) dS = ijk rj np Tpk dS,(1.480)
Angular momenta from surface couples = nT H dS = nk Hki dS. (1.481)

Now the principle, which in words says the time rate of change of angular momenta of a
material region is equal to the sum of external couples (or torques) on the system becomes

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.4. CONSERVATION AXIOMS 89

mathematically,
Z Z Z
d
ijk rj vk dV = ijk rj fk dV + (ijk rj np Tpk + nk Hki ) dS . (1.482)
dt M R(t) M R(t) M S(t)
| {z } | {z }
Apply Leibniz then Gauss apply Gauss

We apply Leibnizs and Gausss theorem to the indicated terms and let the volume of the
material region shrink to zero now. First with Leibniz, we get
Z Z
o ijk rj vk dV + ijk rj vk np vp dS =
M R(t) M S(t)
Z Z
ijk rj fk dV + (ijk rj np Tpk + nk Hki ) dS. (1.483)
M R(t) M S(t)

Next with Gauss we get

Z Z
o ijk rj vk dV + ijk p (rj vk vp )dV =
M R(t) M R(t)
Z Z Z
ijk rj fk dV + ijk p (rj Tpk )dV + k Hki dV. (1.484)
M R(t) M R(t) M R(t)

As the region is arbitrary, the integrand formed by placing all terms under the same integral
must be zero, which yields

ijk (o (rj vk ) + p (rj vp vk ) rj fk p (rj Tpk )) = k Hki . (1.485)

Using the product rule to expand some of the derivatives, we get

ijk
r (v ) + v r +r (v v ) + v v r r f r T T r = k Hki .
|{z}
j o k k o j j p p k p k p j j k j p pk pk p j
|{z} |{z}
=0 pr pr
(1.486)
Applying the simplifications indicated above and rearranging, we get

ijk rj (o (vk ) + p (vp vk ) fk p Tpk ) = k Hki ijk vj vk + ijk Tjk . (1.487)

| {z }
=0 by linear momenta

So, we can say,

k Hki = ijk (vj vk Tjk ) = ijk vj vk T(jk) T[jk] , (1.488)
|{z} | {z } |{z} |{z}
antisym. sym. sym. antisym.

= ijk T[jk] . (1.489)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

90 CHAPTER 1. GOVERNING EQUATIONS

We have utilized the fact that the tensor inner product of any anti-symmetric tensor with
any symmetric tensor must be zero. Now, if we have the case where there are no externally
imposed angular momenta fields, such as could be the case when electromagnetic forces are
important, we have the common condition of Hki = 0, and the angular momenta principle
reduces to the simple statement that
T[ij] = 0. (1.490)
That is, the anti-symmetric part of the stress tensor must be zero. Hence, the stress tensor,
absent any body or surface couples, must be symmetric, and we get in Cartesian index and
Gibbs notation:

Tij = Tji , (1.491)

T = TT . (1.492)

1.4.4 Energy
We recall the first law of thermodynamics, which states the time rate of change of a material
regions internal and kinetic energy is equal to the rate of heat transferred to the material
region less the rate of work done by the material region. Mathematically, this is stated as

dE dQ dW
= . (1.493)
dt dt dt
In this case (though this is not uniformly enforced in these notes), the upper case letters
denote extensive thermodynamic properties. For example, E is total energy, inclusive of
internal and kinetic, with SI units of J. We could have included potential energy in E, but
will instead absorb it into the work term W . Let us consider each term in the first law of
thermodynamics in detail and then write the equation in final form.

1.4.4.1 Total energy term

For a fluid particle, the differential amount of total energy is
 
1
dE = e + vj vj dV, (1.494)
2
 
1
= dV e + vj vj . (1.495)
|{z} 2
mass | {z }
specific internal +kinetic energy

1.4.4.2 Work term

Recall that work is done when a force acts through a distance, and a work rate arises when
a force acts through a distance at a particular rate in time (hence, a velocity is involved).

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.4. CONSERVATION AXIOMS 91

dV

dS
q
i

ni

Figure 1.23: Sketch of finite material region MR, infinitesimal mass element dV , and
infinitesimal surface element dS with unit normal ni , and heat flux vector qi .

Recall also that it is the dot product (inner product) of the force vector with the position or
velocity that gives the true work or work rate. In shorthand, we could say
dW = dxT F, (1.496)
T
dW dx
= F = vT F. (1.497)
dt dt
Here W has the SI units of J, and F has the SI units of N. We contrast this with our
expression for body force per unit mass f, which has SI units of N/kg = m/s2 . Now for the
materials we consider, we must describe work done by two types of forces: 1) body, and 2)
surface.
Work rate done by a body force
Work rate done by force on fluid = (dV )(fi )vi , (1.498)
Work rate done by fluid = vi fi dV. (1.499)

Work rate done by a surface force

Work rate done by force on fluid = (ti dS)vi = [(nj Tji )dS]vi , (1.500)
Work rate done by fluid = nj Tji vi dS. (1.501)

1.4.4.3 Heat transfer term

The only thing confusing about the heat transfer rate is the sign convention. We recall that
heat transfer to a body is associated with an increase in that bodys energy. Now following
the scenario sketched in the material region of Figure 1.23, we define the heat flux vector qi
as a vector which points in the direction of thermal energy flow which has units of energy
per area per time; in SI this would be W/m2 . So, we have
heat transfer rate from body through dS = ni qi dS, (1.502)
heat transfer rate to body through dS = ni qi dS. (1.503)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

92 CHAPTER 1. GOVERNING EQUATIONS

1.4.4.4 Conservative form of the energy equation

Putting the words of the first law into equation form, we get
Z   Z Z Z
d 1
e + vj vj dV = ni qi dS + ni Tij vj dS + fi vi dS. (1.504)
dt M R(t) 2 M S(t) M S(t) M R(t)

Skipping the details of an identical application of Leibnizs and Gausss theorems, and shrink-
ing the volume to approach zero, we obtain the differential equation of energy in conservative
or divergence form (in first Cartesian index then Gibbs notation):
     
1 1
o e + vj vj + i vi e + vj vj =
2 2
| {z } | {z }
rate of change of total energy advection of total energy
i qi + (T v ) + vi fi , (1.505)
|{z} | i {zij j} | {z }
diffusive heat flux surface force work rate body force work rate
     
1 1
e + vT v + T v e + vT v =
t 2 2
T q + T (T v) + vT f. (1.506)

Note that this is a scalar equation as there are no free indices.

We can segregate the work done by the surface forces into that done by pressure forces
and that done by viscous forces by rewriting this in terms of p and ij as follows
     
1 1
o e + vj vj + i vi e + vj vj =
2 2
i qi i (pvi ) + i (ij vj ) + vi fi , (1.507)
     
1 T T 1 T
e+ v v + v e + v v =
t 2 2
T q T (pv) + T ( v) + vT f. (1.508)

1.4.4.5 Secondary forms of the energy equation

While the energy equation just derived is perfectly valid for all continuous materials, it
is common to see other forms. They will be described here. The first, the mechanical
energy equation, actually has no foundation in the first law of thermodynamics; instead, it
is entirely a consequence of the linear momenta principle. It is the type of energy that is
often considered in classical Newtonian particle mechanics, a world in which energy is either
potential or kinetic but not thermal. We include it here because it is closely related to other
forms of energy.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.4. CONSERVATION AXIOMS 93

1.4.4.5.1 Mechanical energy equation The mechanical energy equation, a pure con-
sequence of the linear momenta principle, is obtained by taking the dot product (inner
product) of the velocity vector with the linear momenta principle:

vT linear momenta.

In detail, we get

vj (o vj + vi i vj ) = vj fj vj j p + (i ij )vj , (1.509)
v v  v v 
j j j j
o + vi i = vj fj vj j p + (i ij )vj , (1.510)
2 2
vj vj vj vj vj vj
2
mass : o + i (vi ) = 0. (1.511)
2 2
We add Eqs. (1.510) and (1.511) and use the product rule to get
 vv   vv 
j j j j
o + i vi = vj fj vj j p + (i ij )vj . (1.512)
 T  2  2 
v v vT v

+ T v = vT f vT p + T v. (1.513)
t 2 2

The term vj vj /2 represents the volume averaged kinetic energy, with SI units J/m3 . Note
that the mechanical energy equation, Eq. (1.512), predicts the kinetic energy increases due
to three effects:
fluid motion in the direction of a body force,

fluid motion in the direction of decreasing pressure, or

fluid motion in the direction of increasing viscous stress.

Note that body forces themselves affect mechanical energy, while it is imbalances in surface
forces which affect mechanical energy.

1.4.4.5.2 Thermal energy equation If we take the conservative form of the energy
equation (1.507) and subtract from it the mechanical energy equation (1.512), we get an
equation for the evolution of thermal energy:

o (e) + i (vi e) = i qi pi vi + ij i vj , (1.514)

(e) + T (ve) = T q pT v + : vT . (1.515)
t
Here e is the volume averaged internal energy with SI units J/m3 . Note that the thermal
energy equation (1.514) predicts thermal energy (or internal energy) increases due to
negative gradients in heat flux (more heat enters than leaves),

CC BY-NC-ND. 17 June 2016, J. M. Powers.

94 CHAPTER 1. GOVERNING EQUATIONS

pressure force accompanied by a mean negative volumetric deformation (that is, a

uniform compression; note that i vi is the relative expansion rate), or
viscous force associated with a deformation27 (well worry about the sign later).
Note that in contrast to mechanical energy, thermal energy changes do not require surface
force imbalances; instead they require kinematic deformation. Moreover, body forces have
no influence on thermal energy. The work done by a body force is partitioned entirely to the
mechanical energy of a body.

1.4.4.5.3 Non-conservative energy equation We can obtain the commonly used non-
conservative form of the energy equation, also known as the energy equation following a fluid
particle, by the following operations. First expand the thermal energy equation (1.514):

o e + eo + vi i e + ei (vi ) = i qi pi vi + ij i vj . (1.516)
Then regroup and notice terms common from the mass conservation equation:
(o e + vi i e) +e (o + i (vi )) = i qi pi vi + ij i vj , (1.517)
| {z } | {z }
de =0 by mass
dt

so we get
de
= i qi pi vi + ij i vj , (1.518)
dt
de
= T q pT v + : vT . (1.519)
dt
We can get an equation which is reminiscent of elementary thermodynamics, valid for
small volumes V by multiplying Eq. (1.518) by V and using Eq. (1.288 to replace i vi by its
known value in terms of the relative expansion rate to obtain
de dV
V = V i qi p + V ij i vj . (1.520)
dt dt
The only term not usually found in elementary thermodynamics texts is the third on the
right hand side, which is a viscous work term.

1.4.4.5.4 Energy equation in terms of enthalpy Often the energy equation is cast
in terms of enthalpy. This is generally valid, but especially useful in constant pressure
environments. Recall from elementary thermodynamics the specific enthalpy h is defined as
p
h=e+ . (1.521)

27
For a general fluid, this includes a mean volumetric deformation as well as a deviatoric deformation. If
the fluid satisfies Stokes assumption, it is only the deviatoric deformation that induces a change in internal
energy in the presence of viscous stress.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.4. CONSERVATION AXIOMS 95

Now starting with the energy equation following a particle (1.518), we can use one form
of the mass equation, Eq. (1.429), to eliminate the relative expansion rate i vi in favor of
density derivatives to get
de p d
= i qi + + ij i vj . (1.522)
dt dt
Rearranging, we get  
de p d
= i qi + ij i vj . (1.523)
dt 2 dt
Now differentiating Eq. (1.521), we find
p 1
dh = de 2
d + dp, (1.524)

dh de p d 1 dp
= 2 + , (1.525)
dt dt dt dt
de p d dh 1 dp
2 = , (1.526)
dt dt dt dt
de p d dh dp
= . (1.527)
dt dt dt dt
So, using Eq. (1.527) to eliminate de/dt in Eq. (1.523) in favor of dh/dt, the energy equation
in terms of enthalpy becomes
dh dp
= i qi + ij i vj , (1.528)
dt dt
dh dp
= T q + : vT . (1.529)
dt dt

1.4.4.5.5 Energy equation in terms of entropy By using standard relations from

thermodynamics, we can write the energy equation in terms of entropy. It is important to
note that this is just an algebraic substitution. The physical principle which this equation
will represent is still energy conservation.
Recall the Gibbs equation from thermodynamics, which serves to define entropy s:

T ds = de + pdv. (1.530)

Here T is the absolute temperature, and v is the specific volume, v = V /m = 1/. In terms
of , the Gibbs equation is
p
T ds = de 2 d. (1.531)

Taking the material derivative of Eq. (1.531) , which is operationally equivalent to dividing
by dt, and solving for de/dt,we get
de ds p d
=T + 2 . (1.532)
dt dt dt

CC BY-NC-ND. 17 June 2016, J. M. Powers.

96 CHAPTER 1. GOVERNING EQUATIONS

This is still essentially a thermodynamic definition of s. Now use Eq. (1.532) in the non-
conservative energy equation (1.518) to get an alternate expression for the first law:

ds p d
T + = i qi pi vi + ij i vj . (1.533)
dt dt

Recalling Eq. (1.429), i vi = (1/)(d/dt), we have

ds
T = i qi + ij i vj , (1.534)
dt
ds 1 1
= i qi + ij i vj . (1.535)
dt T T

Using the fact that from the quotient rule we have i (qi /T ) = (1/T )i qi (qi /T 2 )i T , we
can then say

ds q  1 1
i
= i 2 qi i T + ij i vj , (1.536)
dt T T T
ds q 1 1
= T 2 qT T + : vT . (1.537)
dt T T T
From this statement, we can conclude from the first law of thermodynamics that the entropy
of a fluid particle changes due to heat transfer and to deformation in the presence of viscous
stress. We will make a more precise statement about entropy changes after we introduce the
second law of thermodynamics.
The energy equation in terms of entropy can be written in conservative or divergence
form by adding the product of s and the mass equation, so + si (vi ) = 0, to Eq. (1.536)
to obtain
q  1 1
i
o (s) + i (vi s) = i 2 qi i T + ij i vj , (1.538)
T T T
q 1 1
T T
(s) + (vs) = 2 qT T + : vT . (1.539)
t T T T

1.4.5 Entropy inequality

Let us use a non-rigorous method to suggest a form of the entropy inequality which is
consistent with classical thermodynamics. Recall the mathematical statement of the entropy
inequality from classical thermodynamics:

dQ
dS . (1.540)
T
Here S is the extensive entropy, with SI units J/K, and Q is the heat energy into a system
with SI units of J. Notice that entropy can go up or down in a process, depending on the

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.4. CONSERVATION AXIOMS 97

heat transferred. If the process is adiabatic, dQ = 0, and the entropy can either remain fixed
or rise. Now for our continuous material we have
dS = s dV, (1.541)
dQ = qi ni dA dt, (1.542)
(1.543)
Here we have used s for the specific entropy, which has SI units J/kg/K. We have also
changed, for obvious reasons, the notation for our element of surface area, now dA, rather
than the previous dS. Notice we must be careful with our sign convention. When the
heat flux vector is aligned with the outward normal, heat leaves the system. Since we want
positive dQ to represent heat into a system, we need the negative sign.
The second law becomes then
qi
s dV ni dA dt. (1.544)
T
Now integrate over the finite geometry: on the left side this is a volume integral and the
right side this is an area integral.
Z Z 
qi
s dV ni dA dt. (1.545)
M R(t) M S(t) T
(1.546)
Differentiating with respect to time and then applying our typical machinery to the
second law gives rise to
Z Z
d qi
sdV ni dA, (1.547)
dt M R(t) M S(t) T
Z Z Z
qi
o (s)dV + svi ni dA ni dA, (1.548)
M R(t) M S(t) M S(t) T
Z Z q 
i
(o (s) + i (svi )) dV i dV, (1.549)
M R(t) M R(t) T
Z Z q  Z
i
(o (s) + i (svi )) dV = i dV + IdV, (1.550)
M R(t) M R(t) T M R(t)
where irreversibility I 0,
q 
i
o (s) + i (svi ) = i + I, (1.551)
T
ds q 
i
= i + I. (1.552)
dt T
This is the second law. Now if we subtract from this the first law written in terms of entropy,
Eq. (1.536), we get the result
1 1
I= 2
qi i T + ij i vj . (1.553)
T T | {z }

CC BY-NC-ND. 17 June 2016, J. M. Powers.

98 CHAPTER 1. GOVERNING EQUATIONS

As an aside, we have defined the commonly used viscous dissipation function as

= ij i vj . (1.554)
For symmetric stress tensors, we also have = ij (i vj) . Now since I 0, we can view the
entirety of the second law as the following constraint, sometimes called the weak form of the
Clausius-Duhem2829 inequality:
1 1
2
qi i T + ij i vj 0, (1.555)
T T
1 T 1
2 q T + : vT 0, (1.556)
T T
1 T 1
2 q T + : L 0. (1.557)
T T
Recalling that ij is symmetric by the angular momenta principle for no external body
couples, and, consequently, that its tensor inner product with the velocity gradient only has
a contribution from the symmetric part of the velocity gradient (that is, the deformation
rate or strain rate tensor), the entropy inequality reduces slightly to
1 1
2

qi i T + ij (i vj) 0, (1.558)
T T

T !
1 T 1 v + vT
T
2 q T + : 0, (1.559)
T T 2
1 T 1
2
q T + : D 0. (1.560)
T T
We shall see in upcoming sections that we will be able to specify qi and ij in such a fashion
that is both consistent with experiment and satisfies the entropy inequality.
The more restrictive (and in some cases, overly restrictive) strong form of the Clausius-
Duhem inequality requires each term to be greater than or equal to zero. For our system
the strong form, realizing that the absolute temperature T > 0, is
qi i T 0, ij (i vj) 0, (1.561)
| {z }


T !
vT + vT
qT T 0, : 0. (1.562)
2

It is straightforward to show that terms which generate entropy due to viscous work
also dissipate mechanical energy. This can be cleanly demonstrated by considering the
28
Rudolf Clausius, 1822-1888, Prussian-born German mathematical physicist, key figure in making ther-
modynamics a science, author of well-known statement of the second law of thermodynamics, taught at
Zurich Polytechnikum, University of Wurzburg, and University of Bonn.
29
Pierre Maurice Marie Duhem, 1861-1916, French physicist, mathematician, and philosopher, taught at
Lille, Rennes, and the University of Bordeaux.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.4. CONSERVATION AXIOMS 99

mechanisms which cause mechanical energy the change within a finite fixed control volume
V . First consider a restatement of the mechanical energy equation, Eq. (1.510) in terms of
the material derivative of specific kinetic energy:
d  vj vj 
= vj fj vj j (p) + vj i ij . (1.563)
dt 2
Now use the product rule to restate the pressure and viscous work terms so as to achieve
d  vj vj 
= vj fj j (vj p) + pj vj + i (ij vj ) ij i vj . (1.564)
dt 2 | {z }
=0
So, here we see what induces local changes in mechanical energy. We see that body forces,
pressure forces and viscous forces in general can induce the mechanical energy to rise or fall.
However that part of the viscous stresses which is associated with the viscous dissipation,
, is guaranteed to induce a local decrease in mechanical energy.
To study global changes in mechanical energy, we consider the conservative form of the
mechanical energy equation, Eq. (1.512), here written in the same way which takes advantage
of application of the product rule to the pressure and viscous terms:
 vv   vv 
j j j j
o + i vi = vj fj j (vj p) + pj vj + i (ij vj ) ij i vj . (1.565)
2 2
Now integrate over the fixed control volume, so that
Z  vv  Z  vv  Z Z Z
j j j j
o dV + i vi dV = vj fj dV j (vj p) dV + pj vj dV
V 2 V 2 VZ V V
Z
+ i (ij vj ) dV ij i vj dV. (1.566)
V V

Applying Leibnizs rule and Gausss law, we get

Z Z Z Z Z
vj vj vj vj
dV + ni vi dS = vj fj dV nj vj p dS + pj vj dV
t V 2 S 2 VZ S V
Z
+ ni (ij vj ) dS ij i vj dV. (1.567)
S V

Now on the surface of the fixed volume, the velocity is zero, so we get
Z Z Z Z
vj vj
dV = vj fj dV + pj vj dV ij i vj dV. (1.568)
t V 2 V V V | {z }
positive

Now the strong form of the second law requires that ij i vj = ij (i vj) 0. So, we see for
a finite fixed volume of fluid that a body force and pressure force in conjunction with local
volume changes can cause the global mechanical energy to either grow or decay, the viscous
stress always induces a decay of global mechanical energy; in other words it is a dissipative
effect.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

100 CHAPTER 1. GOVERNING EQUATIONS

1.4.6 Summary of axioms in differential form

Here we pause to summarize the mathematical form of our axioms. We give the Cartesian
index, Gibbs, and the full non-orthogonal index notation. All details of development of
the non-orthogonal index notation are omitted, and the reader is referred to Aris for a full
development. We will first present the conservative form and then the non-conservative form.

1.4.6.1 Conservative form

1.4.6.1.1 Cartesian index form

o + i (vi ) = 0, (1.569)
o (vi ) + j (vj vi ) = fi i p + j ji , (1.570)
ij = ji , (1.571)
     
1 1
o e + vj vj + i vi e + vj vj = i qi i (pvi ) + i (ij vj )
2 2
+vi fi , (1.572)
q 
i
o (s) + i (svi ) i . (1.573)
T

1.4.6.1.2 Gibbs form

+ T (v) = 0, (1.574)
t

T
T
(v) + T (vvT ) = f p + T , (1.575)
t
= T, (1.576)
     
1 1
e + vT v + T v e + vT v = T q T (pv)
t 2 2
+T ( v) + vT f , (1.577)
s q
+ T (sv) T . (1.578)
t T

1.4.6.1.3 Non-orthogonal index form Here we introduce, following Aris and many
others, some standard notation from tensor analysis. In this notation, both sub- and su-
perscripts are needed to distinguish between what are known as covariant and contravari-
ant vectors, which are really different mathematical representations of the same quan-
tity, just cast onto different basis vectors. In brief, we have the metric tensor gij =
k k
xi xj
, where k is a Cartesian coordinate and xi is a non-Cartesian coordinate. We
k
also have gij = 12 imn jpq gmpgnq , g = det
xi
. The Christoffel30 symbols are given by
1 mk gjk gki gij
m
ij = 2 g xi
+ xj
xk
. We note also that few texts give a proper exposition of the
30
Elwin Bruno Christoffel, 1829-1900, German mathematician and physicist.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.4. CONSERVATION AXIOMS 101

conservative form of the equations in non-orthogonal coordinates. Here we have extended

the development of Vinokur31 to include the effects of momentum and energy diffusion. This
extension has been guided by general notions found in standard works such as Aris as well
as the book of Liseikin.


( g ) + k g v k = 0, (1.579)
 t
  x 
i i i
g v j j + k g v j v k j = g f j j
t x x x x
 i

jk
k g pg
x xj
 
jk i
+ k g ,
x xj
(1.580)
     
1 1 k

g e + gij v i v j + k g v k e + gij v i v j = k gq
t 2 x 2 x


k g pv k
x


+ k g gij v j ik
x

+ g gij v j f i , (1.581)
 
k

qk
( g s) + k g sv k g (. 1.582)
t x x T

1.4.6.2 Non-conservative form

1.4.6.2.1 Cartesian index form
d
= i vi , (1.583)
dt
dvi
= fi i p + j ji , (1.584)
dt
ij = ji , (1.585)
de
= i qi pi vi + ij i vj , (1.586)
dt
ds q 
i
i . (1.587)
dt T

1.4.6.2.2 Gibbs form

d
= T v, (1.588)
dt
31
Vinokur, M., 1974, Conservation Equations of Gasdynamics, Journal of Computational Physics, 14(2):
105-125.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

102 CHAPTER 1. GOVERNING EQUATIONS

dv
T
= f p + T , (1.589)
dt
= T, (1.590)
de
= T q pT v + : vT , (1.591)
dt
ds q
T
. (1.592)
dt T

1.4.6.2.3 Non-orthogonal index form These have not been checked carefully!

i

+ vi i = gv , (1.593)
t x g xi
 i  i 
v j v i l i ij p 1 ij

+v j
+ jl v = f g j
+ j
g + ijk jk , (1.594)
t x x g x
 
e e 1 i
p i

+ vi i = g q gv
t x g xi g xi
 i 
kj v i l
+gik + jl v , (1.595)
xj
   
s i s 1 qi
+v g . (1.596)
t xi g xi T

1.4.6.3 Physical interpretations

Each term in the governing axioms represents a physical mechanism. This approach is
emphasized in the classical text by Bird, Stewart, and Lightfoot on transport processes.
In general, the equations which are partial differential equations can be represented in the
following form:
local change = advection + diffusion + source. (1.597)
Here we consider advection and diffusion to be types of transport phenomena. If we have a
fixed volume of material, a property of that material, such as its thermal energy, can change
because an outside flow sweeps energy in from outside. That is advection. It can change
because random molecular motions allow slow leakage to the outside or leakage in from the
outside. That is diffusion. Or the material can undergo intrinsic changes inside, such as
viscous work, which converts kinetic energy into thermal energy.
Let us write the Gibbs form of the non-conservative equations of mass, linear momentum,
and energy in a slightly different way to illustrate these mechanisms:

= local change in mass
t
vT advection of mass
+0 diffusion of mass

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.4. CONSERVATION AXIOMS 103

T v, volume expansion source (1.598)

v
= local change in linear momenta
t

vT v advection of linear momenta
T

T
+ , diffusion of linear momenta (1.599)
+f body force source of linear momenta
p pressure force source of linear momenta
e
= local change in thermal energy
t
vT e advection of thermal energy
T
q diffusion of thermal energy
T
p v pressure work thermal energy source
T
+ : v viscous work thermal energy source. (1.600)

Briefly considering the second law, we note that the irreversibility I is solely associated
with diffusion of linear momenta and diffusion of energy. This makes sense in that diffusion
is associated with random molecular motions and thus disorder. Convection is associated
with an ordered motion of matter in that we retain knowledge of the position of the matter.
Pressure volume work is a reversible work and does not contribute to entropy changes. A
portion of the heat transfer can be considered to be reversible. All of the work done by the
viscous forces is irreversible work.

1.4.7 Complete system of equations?

The beauty of these axioms is that they are valid for any material which can be modeled
as a continuum under the influence of the forces we have mentioned. Specifically, they are
valid for both solid and fluid mechanics, which is remarkable.
While the axioms are complete, the equations are not! Note that we have twenty-three
unknowns here (1), vi (3), fi (3), p(1), ij (9), e(1), qi (3), T (1), s(1), and only eight equations
(one mass, three linear momenta, three independent angular momenta, one energy). We
cannot really count the second law as an equation, as it is an inequality. Whatever result we
get must be consistent with it. Whatever the case we are short a number of equations. We will
see in a later section how we use constitutive equations, equations founded in empiricism,
which in some sense model sub-continuum effects that we have ignored, to complete our
system.
Before we go onto this, however, we will in the next section discuss integral control
volume forms of the governing equations.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

104 CHAPTER 1. GOVERNING EQUATIONS

1.4.8 Integral forms

Our governing equations are formulated based upon laws which apply to a material
element.

We are not often interested in an actual material element but in some other fixed of
moving region in space.

Rules for such systems can by formulated with Leibnizs rule in conjunction with the
differential forms of our axioms.

Let us first apply Leibnizs rule (1.178) to an arbitrary function f over a time dependent
arbitrary region AR(t):
Z Z Z
d f
f dV = dV + ni wi f dS. (1.601)
dt AR(t) AR(t) t AS(t)

Recall that wi is the velocity of the arbitrary surface, not necessarily the particle velocity.

1.4.8.1 Mass
Rewriting the mass equation as
o = i (vi ), (1.602)
Now lets use this, and let f = in Leibnizs rule to get
Z Z Z
d
dV = o dV + ni wi dS, (1.603)
dt AR(t) AR(t) AS(t)
Z Z Z
d
dV = (i (vi ))dV + ni wi dS, (1.604)
dt AR(t) AR(t) AS(t)
with Gauss (1.605)
Z Z
d
dV = ni (wi vi )dS. (1.606)
dt AR(t) AS(t)

Now consider three special cases.

1.4.8.1.1 Fixed region We take wi = 0.

Z Z
d
dV = ni vi dS, (1.607)
dt AR(t) AS(t)
Z Z
d
dV + ni vi dS = 0. (1.608)
dt AR(t) AS(t)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.4. CONSERVATION AXIOMS 105

V (t) (material volume)

2

air

a
H

water V1 (fixed)
D1= 1"

v1= 3 ft/s w

h(t)
D2= 3"
v = 2 ft/s
2

A = 2 ft 2

Figure 1.24: Sketch of volume with water and air being filled with water.

1.4.8.1.2 Material region Here we take wi = vi .

Z
d
dV = 0. (1.609)
dt M R(t)

1.4.8.1.3 Moving solid enclosure with holes Say the region considered is a solid
enclosure with holes through with fluid can enter and exit. The our arbitrary surface AS(t)
can be specified as

AS(t) = Ae (t) area of entrances and exits (1.610)

+As (t) solid moving surface with wi = vi (1.611)
+As fixed solid surface with wi = vi = 0. (1.612)

Then we get Z Z
d
dV + ni (vi wi )dS = 0. (1.613)
dt AR(t) Ae (t)

Example 1.10
Consider the volume sketched in Figure 1.24. Water enters a circular hole of diameter D1 = 1 with
velocity v1 = 3 f t/s. Water enters another circular hole of diameter D2 = 3 with velocity v2 = 2 f t/s.
The cross sectional area of the cylindrical tank is A = 2 f t2 . The tank has height H. Water at density

CC BY-NC-ND. 17 June 2016, J. M. Powers.

106 CHAPTER 1. GOVERNING EQUATIONS

w exists in the tank at height h(t). Air at density a fills the remainder of the tank. Find the rate of
rise of the water dh
dt .
Consider two control volumes
V1 : the fixed region enclosing the entire tank, and
V2 (t): the material region attached to the air.
First, let us write mass conservation for the material region 2:

Z
d
a dV = 0, (1.614)
dt V2
Z H
d
a Adz = 0. (1.615)
dt h(t)

Mass conservation for V1 is

Z Z
d
dV + vi ni dS = 0. (1.616)
dt V1 Ae

Now break up V1 and write Ae explicitly

Z h(t) Z H Z Z
d d
w Adz + a Adz = w vi ni dS w vi ni dS, (1.617)
dt 0 dt h(t) A1 A2
| {z }
=0
Z h(t)
d
w Adz = w v1 A1 + 2 v2 A2 , (1.618)
dt 0
w
= (v1 D12 + v2 D22 ), (1.619)
4
Z h(t)
d w
w A dz = (v1 D12 + v2 D22 ), (1.620)
dt 0 4
dh
= (v1 D12 + v2 D22 ) (1.621)
dt 4A !
 2  
1 3
= 2
(3 f t/s) f t + (2 f t/s) f t)2 (1.622)
4(2 f t ) 12 12
ft
= 0.057 . (1.623)
s

1.4.8.2 Linear momenta

Let us perform the same exercise for the linear momenta equation. First, in a strictly
mathematical step, apply Leibnizs rule to linear momenta, vi :
Z Z Z
d
vi dV = o (vi )dV + nj wj vi dS. (1.624)
dt AR(t) AR(t) AS(t)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.5. CONSTITUTIVE EQUATIONS 107

Now invoke the physical linear momenta axiom. Here the axiom gives us an expression for
o (vi ). We will also convert volume integrals to surface integrals via Gausss theorem to
get
Z Z Z
d
vi dV = (nj (vj wj )vi + ni p nj ij )dS + fi dV. (1.625)
dt AR(t) AS(t) AR(t)

Now momentum flux terms only have values at entrances and exits (at solid surfaces we get
vi = wi , so we can say
Z Z Z Z Z
d
vi dV + nj (vj wj )vi dS = ni pdS + nj ij dS + fi dV.
dt AR(t) Ae (t) AS(t) AS(t) AR(t)
(1.626)
Note that the surface forces are evaluated along all surfaces, not just entrances and exits.

1.4.8.3 Energy
Applying the same analysis to the energy equation, we obtain
Z   Z  
d 1 1
e + vj vj dV = ni (vi wi ) e + vj vj dS
dt AR(t) 2 AS(t) 2
Z
ni qi dS
AR(t)
Z
(ni vi p ni ij vj )dS
AS(t)
Z
+ vi fi dV. (1.627)
AR(t)

1.4.8.4 General expression

If we have a governing equation from a physical principle which is of form

o fj + i (vi fj ) = i gj + hj , (1.628)

then we can say for an arbitrary volume that

Z Z Z Z
d
fj dV = ni fj (vi wi )dS + ni gj dS + hj dV . (1.629)
dt AR(t) AS(t) AS(t) AR(t)
| {z } | {z } | {z } | {z }
change of fj flux of fj effect of gj effect of hj

1.5 Constitutive equations

We now return to the problem of completing our set of equations. We recall we have too
many unknowns and not enough equations. Constitutive equations are additional equations

CC BY-NC-ND. 17 June 2016, J. M. Powers.

108 CHAPTER 1. GOVERNING EQUATIONS

which are not as fundamental as the previously developed axioms. They can be rather
ad hoc relations which in some sense model the sub-continuum nano-structure. In some
cases, for example, the sub-continuum kinetic theory of gases, we can formally show that
when the sub-continuum is formally averaged, that we obtain commonly used constitutive
equations. In most cases however, constitutive equations simply represent curve fits to basic
experimental results, which can vary widely from material to material. As is briefly discussed
below, constitutive equations are not completely arbitrary. Whatever is proposed must allow
our final equations to be invariant under Galilean32 transformations and rotations as well as
satisfy the entropy inequality.
For example, we might hope to develop a constitutive equation for the heat flux vector
qi . Being naive, we might in general expect it to be a function of a large number of variables:

qi = qi (, p, T, vi , ij , fi , e, s, . . .). (1.630)

The principles of continuum mechanics will rule out some possibilities, but still allow a broad
range of forms.

1.5.1 Frame and material indifference

Our choice of a constitutive law must be invariant under a Galilean transformation (frame
invariance) a rotation (material indifference). Say for example, we propose that the heat
flux vector is proportional to the velocity vector

qi = avi , trial constitutive relation. (1.631)

If we changed frames such that velocities in the moving frame were ui = vi V , we would
have qi = a(ui + V ). With this constitutive law, we find a physical quantity is dependent
on the frame velocity, which we observe to be non-physical; hence we rule out this trial
constitutive relation.
A commonly used constitutive law for stress in a one-dimensional experiment is

12 = b(1 v2 )a (1 u2 )b , (1.632)

where u2 is the displacement of particle. While this may fit one-dimensional data well, it is
in no way clear how one could simply extend this to write an expression for ij , and many
propositions will fail to satisfy material indifference.

1.5.2 Second law restrictions and Onsager relations

The entropy inequality from the second law of thermodynamics provides additional restric-
tions on the form of constitutive equations. Recall the second law (equivalently, the weak
32
Galileo Galilei, 1564-1642, Pisa-born Italian astronomer, physicist, and developer of experimental meth-
ods, first employed a pendulum to keep time, builder and user of telescopes used to validate the Copernican
view of the universe, developer of the principle of inertia and relative motion.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.5. CONSTITUTIVE EQUATIONS 109

form of the Clausius-Duhem inequality, Eq. (1.558)) tells us that

1 1
2
qi i T + ij (i vj) 0. (1.633)
T T
We would like to find forms of qi and ij which are consistent with the above weak form of
the entropy inequality.

1.5.2.1 Weak form of the Clausius-Duhem inequality

The weak form suggests that we may want to consider both qi and ij to be functions
involving the temperature gradient i T and the deformation tensor (i vj) .

1.5.2.1.1 Non-physical motivating example To see that this is actually too general
of an assumption, it suffices to consider a one-dimensional limit. In the one-dimensional
limit, the weak form of the entropy inequality, Eq. (1.558), reduces to

1 T 1 u
2
q + 0. (1.634)
T x T x
We can write this in a vector form as
 q

( T1 T
x
1 u
u x
) T
u 0. (1.635)
T

Note that a factor of u/u was introduced to the viscous stress term. This allows for a
necessary dimensional consistency in that q/T has the same units as u/T . Let us then
hypothesize a linear relationship exists between the generalized fluxes q/T and u/T and
the generalized driving gradients T1 T
x
and u1 u
x
:
 
q 1 T 1 u
= C11 + C12 , (1.636)
T T x u x
 
u 1 T 1 u
= C21 + C22 , (1.637)
T T x u x
(1.638)

In matrix form this becomes

 q
   
T
C11 C12 T1 T
x
u = 1 u . (1.639)
T
C21 C22 u x

We then substitute this hypothesized relationship into the entropy inequality to obtain
   1 T 
1 T 1 u C11 C12 T x
( T x u x ) 1 u 0. (1.640)
C21 C22 u x

CC BY-NC-ND. 17 June 2016, J. M. Powers.

110 CHAPTER 1. GOVERNING EQUATIONS

We next segregate the matrix Cij into a symmetric and anti-symmetric part to get
 C12 +C21
  C12 C21
  1 T 
1 T 1 u C11 0 T x
( T x u x ) C21 +C12
2 + C21 C12 2
1 u 0. (1.641)
2
C 22 2
0 u x

Distributing the multiplication, we find

 C12 +C21
  1 T 
1 T 1 u C11 T x
( T x u x ) C21 +C12 2
1 u
2
C 22 u x
 C12 C21
  1 T

1 T 1 u 0
+ ( T x u x ) C21 C12 2 T x
1 u 0. (1.642)
2
0 u x
| {z }
=0

The second term is identically zero for all values of temperature and velocity gradients. So
what remains is the inequality involving only a symmetric matrix:
 C12 +C21
  1 T 
1 T 1 u C11 T x
( T x u x ) C21 +C12 2
1 u 0. (1.643)
2
C22 u x

Now in a well known result from linear algebra, a necessary and sufficient condition for
satisfying the above inequality is that the new coefficient matrix be positive semi-definite.
Further, the matrix will be positive semi-definite if it has positive semi-definite eigenvalues.
The eigenvalues of the new coefficient matrix can be shown to be
1 p 
= (C11 + C22 ) (C11 C22 )2 + (C12 + C21 )2 (1.644)
2
Since the terms inside the radical are positive semi-definite, the eigenvalues must be real.
This is a consequence of the parent matrix being symmetric. Now we require two positive
semi-definite eigenvalues. First, if C11 + C22 < 0, we obviously have at least one negative
eigenvalue, so we demand that C11 + C22 0. We then must have
p
C11 + C22 (C11 C22 )2 + (C12 + C21 )2 . (1.645)

This gives rise to

(C11 + C22 )2 (C11 C22 )2 + (C12 + C21 )2 . (1.646)
Expanding and simplifying, one gets
 2
C12 + C21
C11 C22 . (1.647)
2

Now the right side is positive semi-definite, so the left side must be also. Thus

C11 C22 0. (1.648)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.5. CONSTITUTIVE EQUATIONS 111

The only way for the sum and product of C11 and C22 to be positive semi-definite is to
demand that C11 0 and C22 0. Thus we arrive at the final set of conditions to satisfy
the second law:

C11 0, (1.649)
C22 0, (1.650)
 2
C12 + C21
C11 C22 . (1.651)
2

Now an important school of thought, founded by Onsager33 in twentieth century thermo-

dynamics takes an extra step and makes the further assertion that the original matrix Cij
itself must be symmetric. That is C12 = C21 . This remarkable assertion is independent of the
second law, and is, for other scenarios, consistent with experimental results. Consequently,
the second law in combination with Onsagers independent demand, requires that

C11 0, (1.652)
C22 0, (1.653)
p
C12 C11 C22 . (1.654)

All this said, we must dismiss our hypothesis in this specific case on other physical
grounds, namely that such a hypothesis results in an infinite shear stress for a fluid at
rest! Note that in the special case in which T /x = 0, our hypothesis predicts =
C22 (T /u2)(u/x). Obviously this is inconsistent with any observation and so we reject this
hypothesis. Additionally, this assumed form is not frame invariant because of the velocity
dependency. So, why did we go to the trouble to do the above? First, we now have confidence
that we should not expect to find heat flux to depend on deformation. Second, it illustrates
some general techniques in continuum mechanics. Moreover, the techniques we used have
actually been applied to other more complex phenomena which are physical, and of great
practical importance.

1.5.2.1.2 Real physical effects. That such a matrix such as we studied in the previous
section was asserted to be symmetric is a manifestation of what is known as a general
Onsager relation, developed by Onsager in 1931 with a statistical mechanics basis for more
general systems and for which he was awarded a Nobel Prize in chemistry in 1968. These
actually describe a surprising variety of physical phenomena, and are described in detail
many texts, including Fung and Woods. A well-known example is the Peltier34 effect in
which conduction of both heat and electrical charge is influenced by gradients of charge and
temperature. This forms the basis of the operation of a thermocouple. Other relations exist
are the Soret effect in which diffusive mass fluxes are induced by temperature gradients, the
33
Lars Onsager, 1903-1976, Norwegian-born American physical chemist, earned Ph.D. and taught at Yale,
developed a systematic theory for irreversible chemical processes.
34
Jean Charles Athanase Peltier, 1785-1845, French clockmaker, retired at 30 to study science.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

112 CHAPTER 1. GOVERNING EQUATIONS

Dufour effect in which a diffusive energy flux is induced by a species concentration gradient,
the Hall35 effect for coupled electrical and magnetic effects (which explains the operation
of an electric motor), the Seeback36 effect in which electromotive forces are induced by
different conducting elements at different temperatures, the Thomson37 effect in which heat
is transferred when electric current flows in a conductor in which there is a temperature
gradient, and the principle of detailed balance for multi-species chemical reactions.

1.5.2.2 Strong form of the Clausius-Duhem inequality

A less general way to satisfy the second law is to take the sufficient (but not necessary!)
condition that each individual term in the entropy inequality to be greater than or equal to
zero:
1
qi i T 0, and (1.655)
T2
1
ij (i vj) 0. (1.656)
T
Once again, this is called the strong form of the entropy inequality (or the strong form of
the Clausius-Duhem inequality), and is potentially overly restrictive.

1.5.3 Fouriers law

Let us examine the restriction on qi from the strong form of the entropy inequality to infer
the common constitutive relation known as Fouriers law.38 The portion of the strong form
of the entropy inequality with which we are concerned here is
1
qi i T 0. (1.657)
T2
Now one way to guarantee this inequality is satisfied is to specify the constitutive relation
for the heat flux vector as

qi = ki T, with k 0. (1.658)

This is the well known Fouriers Law for an isotropic material, where k is the thermal
conductivity. It has the proper behavior under Galilean transformations and rotations; more
35
Edwin Herbert Hall, 1855-1938, Maine-born American physicist, educated at Johns Hopkins University
where he discovered the Hall effect while working on his dissertation, taught at Harvard.
36
Thomas Johann Seebeck, 1770-1831, German medical doctor who studied at Berlin and Gottingen.
37
William Thomson (Lord Kelvin), 1824-1907, Belfast-born British mathematician and physicist, grad-
uated and taught at Glasgow University, key figure in all of 19th century engineering science including
mathematics, thermodynamics, and electrodynamics.
38
Jean Baptiste Joseph Fourier, 1768-1830, French mathematician and Egyptologist who studied the trans-
fer of heat and the representation of mathematical functions by infinite series summations of other functions.
Son of a tailor.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.5. CONSTITUTIVE EQUATIONS 113

importantly, it is consistent with macro-scale experiments for isotropic materials, and can be
justified from an underlying micro-scale theory. Substitution of Fouriers law for an isotropic
material into the entropy inequality yields
1
k(i T )(i T ) 0, (1.659)
T2
which for k 0 is a true statement. Note the second law allows other forms as well. The
expression qi = k((j T )(j T ))i T is consistent with the second law. It does not match
experiments well for most materials however.
Following Duhamel,39 we can also generalize Fouriers law for an anisotropic material.
Let us only consider anisotropic materials for which the conductivity in any given direction
is a constant. For such materials, the thermal conductivity is a tensor kij , and Fouriers law
generalizes to
qi = kij j T. (1.660)
This effectively states that for a fixed temperature gradient, the heat flux depends on the
orientation. This is characteristic of anisotropic substances such as layered materials. Sub-
stitution of the generalized Fouriers law into the entropy inequality (for ij = 0) gives now
1
kij (j T )(i T ) 0, (1.661)
T2
1
(i T )kij (j T ) 0, (1.662)
T2
1
(T )T K T 0. (1.663)
T2
Now 1/T 2 > 0, so we must have (i T )kij (j T ) 0 for all possible values of T . Now any
possible anti-symmetric portion of kij cannot contribute to the inequality. We can see this
by expanding kij in the entropy inequality to get
 
1 1
i T (kij + kji ) + (kij kji ) j T 0, (1.664)
2 2


i T k(ij) + k[ij] j T 0, (1.665)
(i T )k(ij) (j T ) + (i T )k[ij](j T ) 0, (1.666)
| {z }
=0
(i T )k(ij) (j T ) 0. (1.667)
The anti-symmetric part of kij makes no contribution to the entropy generation because
it involves the tensor inner product of a symmetric tensor with an anti-symmetric tensor,
which is identically zero.
Next, we again use the well-known result from linear algebra that the entropy inequality
is satisfied if k(ij) is a positive semi-definite tensor. This will be the case if all the eigenvalues
39
Jean Marie Constant Duhamel, 1797-1872, highly regarded mathematics teacher at Ecole Polytechnique
in Paris who applied mathematics to problems in heat transfer, mechanics, and acoustics.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

114 CHAPTER 1. GOVERNING EQUATIONS

of k(ij) are non-negative. That this is sufficient to satisfy the entropy inequality is made
plausible if we consider j T to be an eigenvector, so that k(ij) j T = ij j T giving rise to
an entropy inequality of
(i T )ij (j T ) 0, (1.668)
(i T )(i T ) 0. (1.669)
The inequality holds for all i T as long as 0.
Further now, when we consider the contribution of the heat flux vector to the energy
equation, we see any possible anti-symmetric portion of the conductivity tensor will be
inconsequential as well. This is seen by the following analysis, which considers only relevant
terms in the energy equation
de
= i qi + . . . , (1.670)
dt
= i (kij j T ) + . . . , (1.671)
= kij i j T + . . . , (1.672)


= k(ij) + k[ij] i j T + . . . , (1.673)
= k(ij) i j T + k[ij]i j T + . . . , (1.674)
| {z }
=0
= k(ij) i j T + . . . . (1.675)
So, it seems any possible anti-symmetric portion of kij will have no consequence as far
as the first or second laws are concerned. However, an anti-symmetric portion of kij would
induce a heat flux orthogonal to the direction of the temperature gradient. In a remarkable
confirmation of Onsagers principle, experimental measurements on anisotropic crystalline
materials demonstrate that there is no component of heat flux orthogonal to the temperature
gradient, and thus, the conductivity matrix kij in fact has zero anti-symmetric part, and thus
is symmetric, kij = kji . For our particular case with a tensorial conductivity, the competing
effects are the heat fluxes in three directions, caused by temperature gradients in three
directions:
q1 k11 k12 k13 1 T
q2 = k21 k22 k23 2 T . (1.676)
q3 k31 k32 k33 3 T
The symmetry condition, Onsagers principle, requires that k12 = k21 , k13 = k31 , and k23 =
k32 . So, the experimentally verified Onsager principle further holds that the heat flux for an
anisotropic material is given by

q1 k11 k12 k13 1 T
q2 = k12 k22 k23 2 T . (1.677)
q3 k13 k23 k33 3 T
Now it is well known that the conductivity matrix kij will be positive semi-definite if all
its eigenvalues are non-negative. The eigenvalues will be guaranteed real upon adopting

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.5. CONSTITUTIVE EQUATIONS 115

Onsager symmetry. The characteristic polynomial for the eigenvalues is given by

(1) (2) (3)
3 Ik 2 + Ik Ik = 0, (1.678)
where the invariants of the conductivity tensor kij , are given by the standard
(1)
Ik = kii = tr K, (1.679)
(2) 1

Ik = (kii kjj kij kji ) = (det K) tr K1 , (1.680)
2
(3)
Ik = ijk k1j k2j k3j = det K. (1.681)
In a standard result from linear algebra, one can show that if all three invariants are positive
semi-definite, then the eigenvalues are all positive semi-definite, and as a result, the matrix
itself is positive semi-definite. Hence, in order for kij to be positive semi-definite we demand
that
(1)
Ik 0, (1.682)
(2)
Ik 0, (1.683)
(3)
Ik 0, (1.684)
which is equivalent to demanding that
k11 + k22 + k33 0, (1.685)
2 2 2
k11 k22 + k11 k33 + k22 k33 k12 k13 k23 0, (1.686)
k13 (k12 k23 k22 k13 ) + k23 (k12 k13 k11 k23 ) + k33 (k11 k22 k12 k12 ) 0. (1.687)
If det K 6= 0, the conditions reduce to
tr K 0, (1.688)
tr K1 0, (1.689)
det K > 0. (1.690)
Now by considering i T = (1, 0, 0)T , and demanding (i T )kij (j T ) 0, we conclude that
k11 0. Similarly, by considering i T = (0, 1, 0)T and i T = (0, 0, 1)T , we conclude that
k22 0 and k33 0, respectively. Thus tr K 0 is automatically satisfied. In equation
form, we then have
k11 0, (1.691)
k22 0, (1.692)
k33 0, (1.693)
2 2 2
k11 k22 + k11 k33 + k22 k33 k12 k13 k23 0, (1.694)
k13 (k12 k23 k22 k13 ) + k23 (k12 k13 k11 k23 ) + k33 (k11 k22 k12 k12 ) 0. (1.695)
While by no means a proof, numerical experimentation gives strong indication that the
remaining conditions can be satisfied if, loosely stated, k11 , k22 , k33 >> |k12 |, |k23|, |k13 |. That
is, for positive semi-definiteness,

CC BY-NC-ND. 17 June 2016, J. M. Powers.

116 CHAPTER 1. GOVERNING EQUATIONS

each diagonal element must be positive semi-definite,

off-diagonal terms can be positive or negative, and

diagonal terms must have amplitudes which are, loosely speaking, larger than the
amplitudes of off-diagonal terms.

Example 1.11
Let us consider heat conduction in the limit of two dimensions and a constant anisotropic conduc-
tivity tensor, without imposing Onsagers conditions. Let us take then
    
q1 k11 k12 1 T
= . (1.696)
q2 k21 k22 2 T

The second law demands that

  
k11 k12 1 T
( 1 T 2 T ) 0. (1.697)
k21 k22 2 T

This is expanded as
 k12 +k21   k12 k21   
k11 2 0 2 1 T
( 1 T 2 T ) k21 +k12 + k21 k12 0. (1.698)
2 k22 2 0 2 T

As before, the anti-symmetric portion makes no contribution to the left hand side, giving rise to
 k12 +k21   
k11 2 1 T
( 1 T 2 T ) k21 +k12 0. (1.699)
2 k22 2 T

And, demanding that the eigenvalues of the symmetric part of the conductivity tensor be positive gives
rise to the conditions, identical to that of an earlier analysis, that

k11 0, (1.700)
k22 0, (1.701)
 2
k12 + k21
k11 k22 . (1.702)
2

The energy equation becomes

de
= i qi + . . . , (1.703)
dt   
k11 k12 1 T
= ( 1 2 ) + ..., (1.704)
k21 k22 2 T
 
k11 1 T + k12 2 T
= ( 1 2 ) + ..., (1.705)
k21 1 T + k22 2 T
= k11 1 1 T + (k12 + k21 )1 2 T + k22 2 2 T + . . . , (1.706)
2T 2T 2T
= k11 2 + (k12 + k21 ) + k22 2 + . . . . (1.707)
x1 x1 x2 x2

One sees that the energy evolution depends only on the symmetric part of the conductivity tensor.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.5. CONSTITUTIVE EQUATIONS 117

Imposition of Onsagers relations gives simply k12 = k21 , giving rise to second law restrictions of

k11 0, (1.708)
k22 0, (1.709)
2
k11 k22 k12 , (1.710)

and an energy equation of

de 2T 2T 2T
= k11 2 + 2k12 + k22 2 + . . . . (1.711)
dt x1 x1 x2 x2

Example 1.12
Consider the ramifications of a heat flux vector in violation of Onsagers principle: flux in which
the anisotropic conductivity is purely anti-symmetric. For simplicity consider an incompressible solid
with constant specific heat c. For the heat flux, we take
    
q1 0 1 T
= . (1.712)
q2 0 2 T

This holds that heat flux in the 1 direction is induced only by temperature gradients in the 2 direction
and heat flux in the 2 direction is induced only by temperature gradients in the 1 direction.
The second law demands that
  
0 1 T
( 1 T 2 T ) 0, (1.713)
0 2 T
 
2 T
( 1 T 2 T ) 0, (1.714)
1 T
(1 T )(2 T ) + (1 T )(2 T ) 0, (1.715)
0 0. (1.716)

So, the second law holds.

For the incompressible solid with constant heat capacity, the velocity field is zero, and the energy
equation reduces to the simple
T
c = i qi . (1.717)
t
Imposing our unusual expression for heat flux, we get
  
T 0 1 T
c = ( 1 2 ) , (1.718)
t 0 2 T
 
2 T
= ( 1 2 ) , (1.719)
1 T
= 1 2 T + 1 2 T, (1.720)
= 0. (1.721)

So, this unusual heat flux vector is one which induces no change in temperature. In terms of the first law
of thermodynamics, a net energy flux into a control volume in the 1 direction is exactly counterbalanced

CC BY-NC-ND. 17 June 2016, J. M. Powers.

118 CHAPTER 1. GOVERNING EQUATIONS

by an net energy flux out of the same control volume in the 2 direction. Thus the first law holds as
well.
Let us consider a temperature distribution for this unusual material. And let us consider it to apply
to the domain x [0, 1], y [0, 1], t [0, ]. Take

T (x1 , x2 , t) = x2 . (1.722)
T
Obviously this satisfies the first law as t = 0. Let us check the heat flux.

q1 = 2 T = , (1.723)
q2 = 1 T = 0. (1.724)

Now the lower boundary at x2 = 0 has T = 0. The upper boundary has x2 = 1 so T = 1. And this
constant temperature gradient in the 2 direction is inducing a constant heat flux in the 1 direction,
q1 = . The energy flux that enters at x1 = 0 departs at x1 = 1, maintaining energy conservation.
One can consider an equivalent problem in cylindrical coordinates. Taking

x1 = r cos , x2 = r sin , (1.725)

and applying the chain rule,     

r
1 x1 x1 r
= r , (1.726)
2 x2 x2
one finds
    
1 cos sinr
r
= cos . (1.727)
2 sin r

So, transforming q1 = 2 T , and q2 = 1 T gives

   cos
  T 
q1 sin r r
= sin T . (1.728)
q2 cos r

Standard trigonometry gives

    
qr cos sin q1
= . (1.729)
q sin cos q2
| {z }
rotation matrix

Applying the rotation matrix to both sides gives then

     cos
  T 
cos sin q1 cos sin sin r r
= sin T , (1.730)
sin cos q2 sin cos cos r
   1
  T 
qr 0 r r
= T , (1.731)
q 1 0

or simply
T
qr = , (1.732)
r
T
q = . (1.733)
r
Now the steady state temperature distribution in the annular region 1/2 < r < 1, T = r, describes
a domain with an inner boundary held at T = 1/2 and an outer boundary held at T = 1. Such a

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.5. CONSTITUTIVE EQUATIONS 119

temperature distribution would induce a heat flux in the direction only, so that qr = 0 and q = .
That is, the heat goes round and round the domain, but never enters or exits at any boundary.
Now such a flux is counterintuitive precisely because it has never been observed or measured. It is
for this reason that we can adopt Onsagers hypothesis and demand that, independent of the first and
second laws of thermodynamics,
= 0, (1.734)
and the conductivity tenser is purely symmetric.

1.5.4 Stress-strain rate relation for a Newtonian fluid

We now seek to satisfy the second part of the strong form of the entropy inequality, namely
(and recalling that T > 0)
ij (i vj) 0. (1.735)
| {z }

This form suggests that we seek a constitutive equation for the viscous stress tensor ij
which is a function of the deformation tensor (i vj) . Fortunately, such a form exists, which
moreover agrees with macro-scale experiments and micro-scale theories. Here we will focus
on the simplest of such theories, for what is known as a Newtonian fluid, a fluid which
is isotropic and whose viscous stress varies linearly with strain rate. In general, this is a
discipline unto itself known as rheology.

1.5.4.1 Underlying experiments

We can pull a flat plate over a fluid and measure the force necessary to maintain a specified
velocity. This situation and some expected results are sketched in Figure 1.25. We observe
that
At the upper and lower plate surfaces, the fluid has the same velocity of each plate.
This is called the no slip condition.

The faster the velocity V of the upper plate is, the higher the force necessary to pull
the plate is. The increase can be linear or non-linear.

When experiments are carried out with different plate area and different gap width, a
single universal curve results when F/A is plotted against V /h.

The velocity profile is linear with increasing x2 .

In a way similar on a molecular scale to energy diffusion, this experiment is describing
a diffusion of momentum from the pulled plate into the fluid below it. The constitutive
equation we develop for viscous stress, when combined with the governing axioms, will
model momentum diffusion.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

120 CHAPTER 1. GOVERNING EQUATIONS

cross sectional area A

F

v
x
2
h

x1

F h1 F/A
h2
h3

v v/h

Figure 1.25: Sketch of simple Couette flow experiment with measurements of stress versus
strain rate.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.5. CONSTITUTIVE EQUATIONS 121

21

Bingham
Newtonian
plastic

pseudo-plastic

1
dilatant

v1
x2

Figure 1.26: Variation of viscous stress with strain rate for typical fluids.

We can associate F/A with a shear stress: 21 , recalling stress on the 2 face in the 1
direction. We can associate V /h with a velocity gradient, here 2 v1 . We note that considering
the velocity gradient is essentially equivalent to considering the deformation gradient, as far
as the second law is concerned, and so we will be loose here in our use of the term. We define
the coefficient of viscosity for this configuration as
21 viscous stress
= = . (1.736)
2 v1 strain rate
The viscosity is the analog of Youngs40 modulus in solid mechanics, which is the ratio
of stress to strain. In general is a thermodynamic property of a material. It is often
a strong function of temperature, but can vary with pressure as well. A Newtonian fluid
has a viscosity which does not depend on strain rate (but could depend on temperature and
pressure). A non-Newtonian fluid has a viscosity which is strain rate dependent (and possible
temperature and pressure). Some typical behavior is sketched in Figure 1.26. We shall focus
here on fluids whose viscosity is not a function of strain rate. Much of our development will
be valid for temperature and pressure dependent viscosity, while most actual examples will
consider only constant viscosity.

1.5.4.2 Analysis for isotropic Newtonian fluid

Here we shall outline the method described by Whitaker (p. 139-145) to describe the viscous
stress as a function of strain rate for an isotropic fluid with constant viscosity. An isotropic
40
Thomas Young, 1773-1829, English physician and physicist whose experiments in interferometry revived
the wave theory of light, Egyptologist who helped decipher the Rosetta stone, worked on surface tension in
fluids, gave the word energy scientific significance, and developed Youngs modulus in elasticity.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

122 CHAPTER 1. GOVERNING EQUATIONS

fluid has no directional dependencies when subjected to a force. A fluid composed of aligned
long chain polymers is an example of a fluid that is most likely not isotropic. Following
Whitaker, we

postulate that stress is a function of deformation rate (strain rate) only:41

ij = fij ((k vl) ). (1.737)

Written out in more detail, we have postulated a relationship of the form

11 = f11 ((1 v1) , (2 v2) , (3 v3) , (1 v2) , (2 v3) , (3 v1) (2 v1) , (3 v2) , (1 v3) ), (1.738)
12 = f12 ((1 v1) , (2 v2) , (3 v3) , (1 v2) , (2 v3) , (3 v1) (2 v1) , (3 v2) , (1 v3) ), (1.739)
..
. (1.740)
33 = f33 ((1 v1) , (2 v2) , (3 v3) , (1 v2) , (2 v3) , (3 v1) (2 v1) , (3 v2) , (1 v3) ). (1.741)

require that ij = 0 if (i vj) = 0, hence, no strain rate, no stress.

require that stress is linearly related to strain rate:

ij = Cijkl (k vl) . (1.742)

This is the imposition of the assumption of a Newtonian fluid. Here Cijkl is a fourth
order tensor. Thus we have in matrix form

11 C1111 C1122 C1133 C1112 C1123 C1131 C1121 C1132 C1113 (1 v1)
22 C2211 C2222 C2233 C2212 C2223 C2231 C2221 C2232 C2213
(2 v2)
33 C3311 C3322 C3333 C3312 C3323 C3331 C3321 C3332 C3313
(3 v3)
12 C1211 C1222 C1233 C1212 C1223 C1231 C1221 C1232
C1213 (1 v2)

23 = C2311 C2322 C2333 C2312 C2323 C2331 C2321 C2332 C2313
(2 v3)
31 C3111 C3122 C3133 C3112 C3123 C3131 C3121 C3132 C3113
(3 v1)
21 C2111 C2122 C2133 C2112 C2123 C2131 C2121 C2132
C2113 (2 v1)

32 C3211 C3222 C3233 C3212 C3223 C3231 C3221 C3232 C3213 v (3 2)
13 C1311 C1322 C1333 C1312 C1323 C1331 C1321 C1332 C1313 (1 v3)
(1.743)
There are 34 = 81 unknown coefficients Cijkl . We found one of them in our simple
experiment in which we found

21 = 12 = 2 v1 = (2(1 v2) ).

Hence in this special case C1212 = 2.

41
Thus we are not allowing viscous stress to be a function of the rigid body rotation rate. While it seems
intuitive that rigid body rotation should not induce viscous stress, Batchelor mentions that there is no
rigorous proof for this; hence, we describe our statement as a postulate.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.5. CONSTITUTIVE EQUATIONS 123

Now we could do eighty-one separate experiments, or we could take advantage of the

assumption that the fluid has no directional dependency. We will take the following approach.
Observer A conducts an experiment to measure the stress tensor in reference frame A. The
observer begins with the viscosity matrix Cijkl . The experiment is conducted by varying
strain rate and measuring stress. With complete knowledge A feels confident this knowledge
could be used to predict the stress in rotated frame A.
Consider observer A who is oriented to frame A . Oblivious to observer A, A conducts
the same experiment to measure what for her or him is ij The value that A measures must
be the same that A predicts in order for the system to be isotropic. This places restrictions
on the viscosity matrix Cijkl . We intend to show that if the fluid is isotropic only two of the
eighty-one coefficients are distinct and non-zero.
We first use symmetry properties of the stress and strain rate tensor to reduce to thirty-six
unknown coefficients. We note that in actuality there are only six independent components
of stress and six independent components of deformation since both are symmetric tensors.
Consequently, we can write our linear stress-strain rate relation as

11 C1111 C1122 C1133 C1112 + C1121 C1123 + C1132 C1131 + C1113 (1 v1)
22 C2211 C2222 C2233 C2212 + C2221 C2223 + C2232 C2231 + C2213
(2 v2)
33 C3311 C3322 C3333 C3312 + C3321 C3323 + C3332 C3331 + C3313
(3 v3) .
12 = C1211 C1222 C1233 C1212 + C1221 C1223 + C1232
C1231 + C1213 (1 v2)

23 C2311 C2322 C2333 C2312 + C2321 C2323 + C2332 C2331 + C2313 (2 v3)
31 C3111 C3122 C3133 C3112 + C3121 C3123 + C3132 C3131 + C3113 (3 v1)
(1.744)
Now adopting Whitakers notation for simplification, we define the above matrix of Cs as
a new matrix of Cs. Here, now C itself is not a tensor, while C is a tensor. We take
equivalently then

11 C11 C12 C13 C14 C15 C16 (1 v1)
22 C21 C22 C23 C24 C25 C26 (2 v2)

33 C31 C32 C33 C34 C35 C36 (3 v3)

12 = C41 C42 C43 C44 C45 C46 (1 v2) . (1.745)

23 C51 C52 C53 C54 C55 C56 (2 v3)
31 C61 C62 C63 C64 C65 C66 (3 v1)

Next, recalling that for tensorial quantities

ij = ki lj kl , (1.746)
(i vj)

= ki lj (k vl) , (1.747)

let us subject our fluid to a battery of rotations and see what can be concluded by enforcing
material indifference.

180 rotation about x3 axis

For this rotation, sketched in Figure 1.27. we have direction cosines

CC BY-NC-ND. 17 June 2016, J. M. Powers.

124 CHAPTER 1. GOVERNING EQUATIONS

x3
x3

x1

x2 x2

x1

Figure 1.27: Rotation of 180 about x3 axis.

11 = 1 12 = 0 13 = 0
ki = 21 = 0 22 = 1 23 = 0 . (1.748)
31 = 0 32 = 0 33 = 1
Applying the transform rules to each term in the shear stress tensor, we get

11 = k1 l1 kl = (1)2 11 = 11 , (1.749)

22 = k2 l2 kl = (1)2 22 = 22 , (1.750)

33 = k3 l3 kl = (1)2 33 = 33 , (1.751)

12 = k1 l2 kl = (1)2 12 = 12 , (1.752)

23 = k2 l3 kl = (1)(1)23 = 23 , (1.753)

31 = k3 l1 kl = (1)(1)31 = 31 . (1.754)
Likewise we find that

(1 v1) = (1 v1) , (1.755)

(2 v2) = (2 v2) , (1.756)

(3 v3) = (3 v3) , (1.757)

(1 v2) = (1 v2) , (1.758)

(2 v3) = (2 v3) , (1.759)

(3 v1) = (3 v1) . (1.760)

Now our observer A who is in the rotated system would say, for instance that

11 = C11 (1 v1) + C12 (2 v2) + C13 (3 v3) C14 (1 v2) + C15 (2 v3) + C16 (3 v1) , (1.761)

while our observer A who used tensor algebra to predict 11 would say

11 = C11 (1 v1) + C12 (2 v2) + C13 (3 v3) C14 (1 v2) C15 (2 v3) C16 (3 v1) , (1.762)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.5. CONSTITUTIVE EQUATIONS 125

Since we want both predictions to be the same, we must require that

C15 = C16 = 0. (1.763)

In matrix form, our observer A would predict for the rotated frame that

11 C11 C12 C13 C14 C15 C16 (1 v1)
22 C21 C22 C23 C24 C25 C26 (2
v2)
33 C31 C32 C33 C34 C35 C36 (3 v3)

=
C41 C42 C43 C44 C45 C46 v . (1.764)
12 (1 2)
23 C51 C52 C53 C54 C55 C56 (2
v3)

31 C61 C62 C63 C64 C65 C66 (3 v1)

To retain material difference between the predictions of our two observers, we thus
require that C15 = C16 = C25 = C26 = C35 = C36 = C45 = C46 = C51 = C52 = C53 =
C54 = C61 = C62 = C63 = C64 = 0. This eliminates 16 coefficients and gives our
viscosity matrix the form

C11 C12 C13 C14 0 0
C21 C22 C23 C24 0 0

C31 C32 C33 C34 0 0
. (1.765)
C41 C42 C43 C44 0 0

0 0 0 0 C55 C56
0 0 0 0 C65 C66

with only 20 independent coefficients.

180 rotation about x1 axis

This rotation is sketched in Figure 1.28.
Leaving out the details of the previous section, this rotation has a set of direction
cosines of
1 0 0
ij = 0 1 0 . (1.766)
0 0 1
Application of this rotation leads to the conclusion that the viscosity matrix must be
of the form
C11 C12 C13 0 0 0
C21 C22 C23 0 0 0

C31 C32 C33 0 0 0
. (1.767)
0 0 0 C44 0 0

0 0 0 0 C55 0
0 0 0 0 0 C66
with only 12 independent coefficients.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

126 CHAPTER 1. GOVERNING EQUATIONS

x3

x2 x2

x1
x1

x3

Figure 1.28: Rotation of 180 about x1 axis.

x3

x2

x3

x2

x1
x1

Figure 1.29: Rotation of 90 about x1 axis.

180 rotation about x2 axis

One is tempted to perform this rotation as well, but nothing new is learned from it!

90 rotation about x1 axis

This rotation is sketched in Figure 1.29.
This rotation has a set of direction cosines of

1 0 0
ij = 0 0 1 . (1.768)
0 1 0

Application of this rotation leads to the conclusion that the viscosity matrix must be

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.5. CONSTITUTIVE EQUATIONS 127

x3

x3

x1

x2

x2
x1

Figure 1.30: Rotation of 90 about x3 axis.

of the form
C11 C12 C12 0 0 0
C21 C22 C23 0 0 0

C21 C23 C22 0 0 0
. (1.769)
0 0 0 C44 0 0

0 0 0 0 C55 0
0 0 0 0 0 C66
with only 8 independent coefficients.
90 rotation about x3 axis
This rotation is sketched in Figure 1.30.
This rotation has a set of direction cosines of

0 1 0
ij = 1 0 0 (1.770)
0 0 1
Application of this rotation leads to the conclusion that the viscosity matrix must be
of the form
C11 C12 C12 0 0 0
C12 C11 C12 0 0 0

C12 C12 C11 0 0 0
. (1.771)
0 0 0 C44 0 0

0 0 0 0 C44 0
0 0 0 0 0 C44
with only 3 independent coefficients.
90 rotation about x2 axis
We learn nothing from this rotation.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

128 CHAPTER 1. GOVERNING EQUATIONS

x3

x3

x2

45 o

x2

45 o

x1 x1

Figure 1.31: Rotation of 45 about x3 axis.

45 rotation about x3 axis

This rotation is sketched in Figure 1.31.
This rotation has a set of direction cosines of

2/2 2/2 0
ij = 2/2 2/2 0 . (1.772)
0 0 1
After a lot of algebra, application of this rotation lead to the conclusion that the
viscosity matrix must be of the form

C44 + C12 C12 C12 0 0 0
C12 C + C C 0 0 0
44 12 12
C12 C12 C44 + C12 0 0 0
. (1.773)
0 0 0 C44 0 0

0 0 0 0 C44 0
0 0 0 0 0 C44
with only 2 independent coefficients.
Try as we might, we cannot reduce this any further with more rotations. It can be proved
more rigorously, as shown in most books on tensor analysis, that this is the furthest reduction
that can be made. So, for an isotropic Newtonian fluid, we can expect two independent
coefficients to parameterize the relation between strain rate and viscous stress. The relation
between stress and strain rate can be expressed in detail as


11 = C44 (1 v1) + C12 (1 v1) + (2 v2) + (3 v3) , (1.774)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.5. CONSTITUTIVE EQUATIONS 129

22 = C44 (2 v2) + C12 (1 v1) + (2 v2) + (3 v3) , (1.775)


33 = C44 (3 v3) + C12 (1 v1) + (2 v2) + (3 v3) , (1.776)
12 = C44 (1 v2) , (1.777)
23 = C44 (2 v3) , (1.778)
31 = C44 (3 v1) . (1.779)
Using traditional notation, we take
C44 2, where is the first coefficient of viscosity, and
C12 , where is the second coefficient of viscosity.
A similar analysis in solid mechanics leads one to conclude for an isotropic material
in which the stress tensor is linearly related to the strain (rather than the strain rate)
gives rise to two independent coefficients, the elastic modulus and the shear modulus.
In solids, these both can be measured, and they are independent.
In terms of our original fourth order tensor, we can write the linear relationship ij =
Cijkl (i vj) as

11 2 + 0 0 0 0 0 0 (1 v1)
22 2 + 0 0 0 0 0 0
(2 v2)
33 2 + 0 0 0 0 0 0 (3 v3)

12 0 0 0 2 0 0 0 0 0
(1 v2)
23 = 0 0 0 0 2 0 0 0 0
(2 v3) . (1.780)
31 0 0 0 0 0 2 0 0
0 (3 v1)

21 0 0 0 0 0 0 2 0 0 v
(2 1)
32 0 0 0 0 0 0 0 2 0 (3 v2)
13 0 0 0 0 0 0 0 0 2 (1 v3)
We note that because of the symmetry of (i vj) that the above representation is not unique in
that the following, as well as other linear combinations, is an identically equivalent statement:

11 2 + 0 0 0 0 0 0 (1 v1)
22 2 + 0 0 0 0 0 0
(2 v2)
33 2 + 0 0 0 0 0
0 (3 v3)

12 0 0 0 0 0 0 0
(1 v2)
23 = 0 0 0 0 0 0 0
(2 v3) . (1.781)
31 0 0 0 0 0 0 0
(3 v1)

21 0 0 0 0 0 0 0
(2 v1)
32 0 0 0 0 0 0 0 v (3 2)
13 0 0 0 0 0 0 0 (1 v3)
In shorthand Cartesian index and Gibbs notation, the viscous stress tensor is given by
ij = 2(i vj) + k vk ij , (1.782)
 
vT + (vT )T
= 2 + (T v)I. (1.783)
2

CC BY-NC-ND. 17 June 2016, J. M. Powers.

130 CHAPTER 1. GOVERNING EQUATIONS

By performing minor algebraic manipulations, the viscous stress tensor can be cast in
a way which elucidates more of the physics of how strain rate influences stress. It is easily
verified by direct expansion that the viscous stress tensor can be written as

k vk 1
ij = (2 + 3) ij + 2 (i vj) k vk ij , (1.784)
3}
| {z | {z3 }
mean strain rate deviatoric strain rate
| {z } | {z }
mean viscous stress deviatoric viscous stress
 
T v vT + (vT )T 1 T
= (2 + 3) I + 2 vI . (1.785)
3 2 3
Here it is seen that a mean strain rate, really a volumetric change, induces a mean viscous
stress, as long as 6= (2/3). If either = (2/3) or k vk = 0, all viscous stress is
deviatoric. Further, for 6= 0, a deviatoric strain rate induces a deviatoric viscous stress.
We can form the mean viscous stress by contracting the viscous stress tensor:
 
1 2
ii = + k vk . (1.786)
3 3
Note that the mean viscous stress is a scalar, and is thus independent of orientation; it is
directly proportional to the first invariant of the viscous stress tensor. Obviously the mean
viscous stress is zero if = (2/3). Now the total stress tensor is given by
Tij = pij + 2(i vj) + k vk ij , (1.787)
 
vT + (vT )T
T = pI + 2 + (T v)I. (1.788)
2
We notice the stress tensor has three components, 1) a uniform diagonal tensor with the
hydrostatic pressure, 2) a tensor which is directly proportional to the strain rate tensor, and
3) a uniform diagonal tensor which is proportional to the first invariant of the strain rate
(1)
tensor: I = tr ((i vk) ) = k vk . Consequently, the stress tensor can be written as
 
(1)
Tij = p + I ij + 2(i vj) , (1.789)
| {z } | {z }
isotropic linear in strain rate
   
(1) vT + (vT )T
T = p + I I + 2 . (1.790)
2
(1)
Recalling that ij = I as well as I are invariant under a rotation of coordinate axes, we
deduce that the stress is related linearly to the strain rate. Moreover when the axes are
rotated to be aligned with the principal axes of strain rate, the stress is purely normal stress
and takes on its principal value.
Let us next consider two typical elements to aid in interpreting the relation between
viscous stress and strain rate for a general Newtonian fluid.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.5. CONSTITUTIVE EQUATIONS 131

1.5.4.2.1 Diagonal component Consider a typical diagonal component of the viscous

stress tensor, say 11 :

 
1 v1 + 2 v2 + 3 v3 1
11 = (2 + 3) + 2
1 v1 (1 v1 + 2 v2 + 3 v3 )(1.791)
.
3 3
| {z } | {z }
deviatoric strain rate
| mean strain rate
{z } | {z }
mean viscous stress deviatoric viscous stress

Note that if we choose our axes to be the principal axes of the strain-rate tensor, then
these terms will appear on the diagonal of the stress tensor and there will be no off-diagonal
elements. Thus the fundamental physics of the stress-strain relationship are completely
embodied in a natural way in the above expression.

1.5.4.2.2 Off-diagonal component If we are not aligned with the principal axes, then
off-diagonal terms will be non-zero. A typical off-diagonal component of the viscous stress
tensor, say 12 , has the following form:

12 = 2 (1 v2) + k vk 12 , (1.792)
|{z}
=0
= 2(1 v2) , (1.793)
= (1 v2 + 2 v1 ). (1.794)
Note this is associated with shear deformation for elements aligned with the 1 and 2 axes,
and that it is independent of the value of , which is only associated with the mean strain
rate.

1.5.4.3 Stokes assumption

It is a straightforward matter to measure . It is not at all straightforward to measure
. As discussed earlier, Stokes in the mid-nineteenth century suggested to require that
the mechanical pressure (that is the average normal stress) be equal to the thermodynamic
pressure. We have seen that the consequence of this is Eq. (1.456): ii = 0. If we enforce
this on our expression for ij , we get
ii = 0 = 2(i vi) + k vk ii , (1.795)
= 2i vi + 3k vk , (1.796)
= 2i vi + 3i vi , (1.797)
= (2 + 3)i vi . (1.798)
Since in general i vi 6= 0, Stokes assumption implies that
2
= . (1.799)
3
CC BY-NC-ND. 17 June 2016, J. M. Powers.
132 CHAPTER 1. GOVERNING EQUATIONS

So, a Newtonian fluid satisfying Stokes assumption has the following constitutive equation
for viscous stress
 
1
ij = 2 (i vj) k vk ij , (1.800)
3
| {z }
deviatoric strain rate
| {z }
deviatoric viscous stress
 
(vT + (vT )T ) 1 T
= 2 ( v)I . (1.801)
2 3
Note that incompressible flows have i vi = 0; thus, plays no role in determining the viscous
stress in such flows. For the fluid that obeys Stokes assumption, the viscous stress is entirely
deviatoric and is induced only by a deviatoric strain rate.

1.5.4.4 Second law restrictions

Recall that in order that the constitutive equation for viscous stress be consistent with second
law of thermodynamics, that it is sufficient (but perhaps overly restrictive) to require that
1
ij (i vj) 0. (1.802)
T
Invoking our constitutive equation for viscous stress, and realizing that the absolute tem-
perature T > 0, we have then that
= (2(i vj) + k vk ij )((i vj) ) 0. (1.803)
This reduces to the sum of two squares:
= 2(i vj) (i vj) + k vk i vi 0. (1.804)
We then seek restrictions on and such that this is true. Obviously requiring 0 and
0 guarantees satisfaction of the second law. However, Stokes assumption of = 32
does not meet this criterion, and so we are motivated to check more carefully to see if we
actually need to be that restrictive.

1.5.4.4.1 One dimensional systems Let us first check the criterion for a strictly one-
dimensional system. For such a system, our second law restriction reduces to
2(1 v1) (1 v1) + 1 v1 1 v1 0, (1.805)
(2 + )1 v1 1 v1 0, (1.806)
2 + 0, (1.807)
2. (1.808)
Obviously if > 0 and = 23 , the entropy inequality is satisfied. We also could satisfy
the inequality for negative with sufficiently large positive .

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.5. CONSTITUTIVE EQUATIONS 133

1.5.4.4.2 Two dimensional systems Extending this to a two dimensional system is

more complicated. For such systems, expansion of our second law condition gives

2(1 v1) (1 v1) + 2(1 v2) (1 v2) + 2(2 v1) (2 v1) + 2(2 v2) (2 v2)



+ (1 v1) + (2 v2) (1 v1) + (2 v2) 0. (1.809)

Taking advantage of symmetry of the deformation tensor, we can say

2(1 v1) (1 v1) + 4(1 v2) (1 v2) + 2(2 v2) (2 v2) + (1 v1) + (2 v2) (1 v1) + (2 v2) 0.
(1.810)
Expanding the product and regrouping gives

(2 + )(1 v1) (1 v1) + 4(1 v2) (1 v2) + (2 + )(2 v2) (2 v2) + 2(1 v1) (2 v2) 0. (1.811)

In matrix form, we can write this inequality in the form known from linear algebra as a
quadratic form:

(2 + ) 0 (1 v1)
= ( (1 v1) (2 v2) (1 v2) ) (2 + ) 0 (2 v2) 0. (1.812)
0 0 4 (1 v2)
As we have discussed before, the condition that this hold for all values of the deformation
is that the symmetric part of the coefficient matrix have eigenvalues which are greater than
or equal to zero. In fact, here the coefficient matrix is purely symmetric. Let us find the
eigenvalues of the coefficient matrix. The eigenvalues are found by evaluating the following
equation  
 (2 + ) 0 

 (2 + ) 0  = 0. (1.813)

 0 0 4 
We get the characteristic polynomial


(4 ) (2 + )2 2 = 0. (1.814)

This has roots

= 4, (1.815)
= 2, (1.816)
= 2( + ). (1.817)

For the two-dimensional system, we see now formally that we must satisfy both

0, (1.818)
. (1.819)

This is more restrictive than for the one-dimensional system, but we see that a fluid obeying
Stokes assumption = 23 still satisfies this inequality.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

134 CHAPTER 1. GOVERNING EQUATIONS

1.5.4.4.3 Three dimensional systems For a full three dimensional variation, the en-
tropy inequality (2(i vj) + k vk ij )((i vj) ) 0, when expanded, is equivalent to the fol-
lowing quadratic form

+ 2 0 0 0 (1 v1)
+ 2 0 0 0
(2 v2)
= ( (1 v1) (2 v2) (3 v3) (1 v2) (2 v3) (3 v1) )
0

0
+ 2
0
0
4
0
0
0 (3 v3)
0 (1 v2)
0.

0 0 0 0 4 0 (2 v3)
0 0 0 0 0 4 (3 v1)
(1.820)
Again this must hold for arbitrary values of the deformation, so we must require that the
eigenvalues of the interior matrix be greater than or equal to zero to satisfy the entropy
inequality. It is easy to show that the six eigenvalues for the interior matrix are

= 2, (1.821)
= 2, (1.822)
= 4, (1.823)
= 4, (1.824)
= 4, (1.825)
= 3 + 2. (1.826)

Two of the eigenvalues are degenerate, but this is not a particular problem. We need now
that 0, so the entropy inequality requires that

0, (1.827)
2
. (1.828)
3
Obviously a fluid which satisfies Stokes assumption does not violate the entropy inequality,
but it does give rise to a minimum level of satisfaction. This does not mean the fluid is
isentropic! It simply means one of the six eigenvalues is zero.
Now using standard techniques from linear algebra for quadratic forms, the entropy
inequality can, after much effort, be manipulated into the form

2

= ((1 v1) (2 v2) )2 + ((2 v2) (3 v3) )2 + ((3 v3) (1 v1) )2
3  
2
+ + ((1 v1) + (2 v2) + (3 v3) )2
3
+4(((1 v2) )2 + ((2 v3) )2 + ((3 v1) )2 ) 0. (1.829)

Obviously, this is a sum of perfect squares, and holds for all values of the strain rate tensor.
It can be verified by direct expansion that this term is identical to the strong form of the
entropy inequality for viscous stress. It can further be verified by direct expansion that the

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.5. CONSTITUTIVE EQUATIONS 135

entropy inequality can also be written more compactly as

    
1 1 2
= 2 (i vj) k vk ij (i vj) m vm ij + + (i vi )(j vj ) 0.(1.830)
3 3 3 | {z }
| {z }| {z } (mean strain rate)2
deviatoric strain rate deviatoric strain rate

So, we see that for a Newtonian fluid that the increase in entropy due to viscous dissipation is
attributable to two effects: deviatoric strain rate and mean strain rate. The terms involving
both are perfect squares, so as long as 0 and 32 , the second law is not violated
by viscous effects.
We can also write the strong form of the entropy inequality for a Newtonian fluid
(2(i vj) + k vk ij )((i vj) ) 0, in terms of the principal invariants of strain rate. Leaving
out details, which can be verified by direct expansion of all terms, we find the following form
      
2 (1) 2 (2) 2 (1)
2
= 2 I 2I + + I 0. (1.831)
3 3
Because this is in terms of the invariants, we are assured that it is independent of the
orientation of the coordinate system.
It is, however, not obvious that this form is positive semi-definite. We can use the
definitions of the invariants of strain rate to rewrite the inequality as
   
1 2
= 2 (i vj) (j vi) (i vi ) (j vj ) + + (i vi ) (j vj ) 0. (1.832)
3 3
In terms of the eigenvalues of the strain rate tensor, 1 , 2 , and 3 , this becomes
   
1 2
= 2 1 + 2 + 3 (1 + 2 + 3 ) + + (1 + 2 + 3 )2 0.
2 2 2 2
(1.833)
3 3
This then reduces to a positive semi-definite form:
 
2
2
= (1 2 ) + (1 3 ) + (2 3 ) + + (1 + 2 + 3 )2 0.
2 2 2
(1.834)
3 3
Since the eigenvalues are invariant under rotation, this form is invariant.
We summarize by noting relations between mean and deviatoric stress and strain rates
for Newtonian fluids. The influence of each on each has been seen or is easily shown to be
as follows:
A mean strain rate will induce a time rate of change in the mean thermodynamic stress
via traditional thermodynamic relations42 and will induce an additional mean viscous
stress for fluids that do not obey Stokes assumption.

A deviatoric strain rate will not directly induce a mean stress.

42
e.g. for an isothermal ideal gas dp/dt = RT (d/dt) = RT i vi

CC BY-NC-ND. 17 June 2016, J. M. Powers.

136 CHAPTER 1. GOVERNING EQUATIONS

A deviatoric strain rate will directly induce a deviatoric stress.

A mean strain rate will induce entropy production only for a fluid that does not obey
Stokes assumption.
A deviatoric strain rate will always induce entropy production in a viscous fluid.

1.5.5 Equations of state

Thermodynamic equations of state provide algebraic relations between variables such as
pressure, temperature, energy, and entropy. They do not involve velocity. They are formally
valid for materials at rest. As long as the times scales of equilibration of the thermodynamic
variables are much faster than the finest time scales of fluid dynamics, it is a valid assumption
to use an ordinary equations of state. Such assumptions can be violated in very high speed
flows in which vibrational and rotational modes of oscillation become excited. They may
also be invalid in highly rarefied flows such as might occur in the upper atmosphere.
Typically, we will require two types of equations, a thermal equation of state which gives
the pressure as a function of two independent thermodynamic variables, e.g.
p = p(, T ), (1.835)
and a caloric equation of state which gives the internal energy as a function of two indepen-
dent thermodynamic variables, e.g.
e = e(, T ). (1.836)
There are additional conditions regarding internal consistency of the equations of state; that
is, just any stray functional forms will not do.
We outline here a method for generating equations of state with internal consistency based
on satisfying the entropy inequality. First let us define a new thermodynamic variable, a,
the Helmholtz43 free energy:
a = e T s. (1.837)
We can take the material time derivative of Eq. (1.837) to get
da de ds dT
= T s . (1.838)
dt dt dt dt
It is shown in thermodynamics texts that there are a set of natural, canonical, variables
for describing a which are T and . That is, we take a = a(T, ). Taking the time derivative
of this form of a and using the chain rule tells us another form for da/dt:
 
da a  dT a  d
= + . (1.839)
dt T  dt T dt
43
Hermann von Helmholtz, 1821-1894, Potsdam-born German physicist and philosopher, descendant of
William Penn, the founder of Pennsylvania, empiricist and refuter of the notion that scientific conclusions
could be drawn from philosophical ideas, graduated from medical school, wrote convincingly on the science
and physiology of music, developed theories of vortex motion as well as thermodynamics and electrodynamics.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.5. CONSTITUTIVE EQUATIONS 137

Now we also have the energy equation and entropy inequality:

de
= i qi pi vi + ij i vj , (1.840)
dt
ds q 
i
i . (1.841)
dt T
Using Eq. (1.838) to eliminate de/dt in favor of da/dt in the energy equation, Eq. (1.840),
gives a modified energy equation:
 
da ds dT
+T +s = i qi pi vi + ij i vj . (1.842)
dt dt dt
Next, we use Eq. (1.839) to eliminate da/dt in Eq. (1.842) to get
  !
a  dT a  d ds dT
+ +T +s = i qi pi vi + ij i vj . (1.843)
T  dt T dt dt dt

Now in this modified energy equation, we solve for ds/dt to get

 
ds 1 p 1 a  dT a  d s dT
= i qi i vi + ij i vj . (1.844)
dt T T T T T  dt T T dt T dt

Substituting this version of the energy conservation equation into the second law, Eq. (1.841),
gives
  q 
1 p 1 a  dT a  d s dT i
i qi i vi + ij i vj i . (1.845)
T T T T T  dt T T dt T dt T

Rearranging and using the mass conservation relation to eliminate i vi , we get

   
qi p 1 d 1 a  dT a  d s dT
2 i T + ij i vj 0, (1.846)
T T dt T T T  dt T T dt T dt
 
qi p d a  dT a  d dT
i T + + ij i vj   s 0, (1.847)
T dt T dt T dt dt
    !
qi 1 d a  dT a 
i T + ij i vj + p 2  s+  0. (1.848)
T dt 
T dt T 

Now in our discussion of the strong form of the energy inequality, we have already found forms
for qi and ij for which the terms involving these phenomena are positive semi-definite. We
can guarantee the remaining two terms are consistent with the second law, and are associated
with reversible processes by requiring that


2 a 
p = , (1.849)
T

a 
s = . (1.850)
T 

CC BY-NC-ND. 17 June 2016, J. M. Powers.

138 CHAPTER 1. GOVERNING EQUATIONS

For example, if we take the non-obvious, but experimentally defensible choice for a of
   
T
a = cv (T To ) cv T ln + RT ln , (1.851)
To o
then we get for pressure   
a 
2 2 RT
p= = = RT. (1.852)
T
The above equation for pressure a thermal equation of state for an ideal gas, and R is known
as the gas constant. It is the ratio of the universal gas constant and the molecular mass of
the particular gas.
Solving for entropy s, we get
    
a  T
s=  = cv ln R ln . (1.853)
T To o
Then, we get for e
e = a + T s = cv (T To ). (1.854)
We call the above equation for energy a caloric equation of state for calorically perfect gas.
It is calorically perfect because the specific heat at constant volume cv is assumed a true
constant here. In general for ideal gases, it can be shown to be at most a function of
temperature.

1.6 Boundary and interface conditions

At fluid solid interfaces, it is observed in the continuum regime that the fluid sticks to the
solid boundary, so that we can safely take the fluid and solid velocities to be identical at the
interface. This is called the no slip condition. As one approaches the molecular level, this
breaks down.
At the interface of two distinct, immiscible fluids, one requires that stress be continuous
across the interface, that the energy flux be continuous across the interface. Density need
not be continuous in the absence of mass diffusion. Were mass diffusion present, the fluids
would not be immiscible, and density would be a continuous variable. Additionally the effect
of surface tension may need to be accounted for. We shall not consider surface tension in
this course, but many texts give a complete treatment.

1.7 Complete set of compressible Navier-Stokes equa-

tions
Here we pause once more to write a complete set of equations, the compressible Navier 44
-Stokes equations, written here for a fluid which satisfies Stokes assumption, but for which
44
Claude Louis Marie Henri Navier, 1785-1836, Dijon-born French civil engineer and mathematician, stud-
ied under Fourier, taught applied mechanics at Ecole des Ponts et Chaussees, replaced Cauchy as professor

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.7. COMPLETE SET OF COMPRESSIBLE NAVIER-STOKES EQUATIONS 139

the viscosity (as well as thermal conductivity k) may be variable. They are given in a form
similar to that done in an earlier section.

1.7.0.1 Conservative form

1.7.0.1.1 Cartesian index form
o + i (vi ) = 0, (1.855)
o (vi ) + j (vj vi ) = fi i p
  
1
+j 2 (j vi) k vk ji , (1.856)
3
  
1
o e + vj vj
2
  
1
+i vi e + vj vj = vi fi i (pvi ) + i (ki T )
2
   
1
+i 2 (i vj) k vk ij vj , (1.857)
3
p = p(, T ), (1.858)
e = e(, T ), (1.859)
= (, T ), (1.860)
k = k(, T ). (1.861)

1.7.0.1.2 Gibbs form

+ T (v) = 0, (1.862)
t

T
(v) + T (vvT ) = f p
t
   T
T vT + (vT )T 1 T
+ 2 ( v)I ,(1.863)
2 3
  
1
e + vT v
t 2
  
T 1 T
+ v e + v v = vT f T (pv) + T (kT )
2
   
vT + (vT )T 1
+T 2 (T v)I v (1.864)
,
2 3
p = p(, T ), (1.865)
e = e(, T ), (1.866)
= (, T ), (1.867)
k = k(, T ). (1.868)

at Ecole Polytechnique, specialist in road and bridge building, did not fully understand shear stress in a fluid
and used faulty logic in arriving at his equations.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

140 CHAPTER 1. GOVERNING EQUATIONS

1.7.0.2 Non-conservative form

1.7.0.2.1 Cartesian index form

d
= i vi , (1.869)
dt   
dvi 1
= fi i p + j 2 (j vi) k vk ji , (1.870)
dt 3
 
de 1
= pi vi + i (ki T ) + 2 (i vj) k vk ij i vj , (1.871)
dt 3
p = p(, T ), (1.872)
e = e(, T ), (1.873)
= (, T ), (1.874)
k = k(, T ) (1.875)

1.7.0.2.2 Gibbs form

d
= T v, (1.876)
dt
   T
dv T vT + (v)T 1 T
= f p + 2 ( v)I , (1.877)
dt 2 3
 
de vT + (vT )T 1
= pT v + T (kT ) + 2 (T v)I : vT , (1.878)
dt 2 3
p = p(, T ), (1.879)
e = e(, T ), (1.880)
= (, T ), (1.881)
k = k(, T ). (1.882)

We take , and k to be thermodynamic properties of temperature and density. In practice,

both dependencies are often weak, especially the dependency of and k on density. We
also assume we know the form of the external body force per unit mass fi . We also no
longer formally require the angular momentum principle, as it has been absorbed into our
constitutive equation for viscous stress. We also need not write the second law, as we can
guarantee its satisfaction as long as 0, k 0.
In summary, we have nine unknowns, ,vi (3), p, e, T , , and k, and nine equations,
mass, linear momenta (3), energy, thermal state, caloric state, and thermodynamic relations
for viscosity and thermal conductivity. When coupled with initial, interface, and boundary
conditions, all dependent variables can, in principle, be expressed as functions of position xi
and time t, and this knowledge utilized to design devices of practical importance.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.8. INCOMPRESSIBLE NAVIER-STOKES EQUATIONS WITH CONSTANT
PROPERTIES 141
1.8 Incompressible Navier-Stokes equations with con-
stant properties
If we make the assumption, which can be justified in the limit when fluid particle velocities
are small relative to the velocity of sound waves in the fluid, that density changes following
a particle are negligible (that is d
dt
0), the Navier-Stokes equations simplify considerably.
Note that this does not imply the density is constant everywhere in the flow. Our assump-
tion allows for stratified flows, for which the density of individual particles still can remain
constant. We shall also assume viscosity , and thermal conductivity k are constants, though
this is not necessary.
Let us examine the mass, linear momenta, and energy equations in this limit.

1.8.1 Mass
Expanding the mass equation
o + i (vi ) = 0, (1.883)
we get
o + vi i +i vi = 0. (1.884)
| {z }
d
dt
0
We are assuming the first two terms in the above expression, which form d/dt, go to zero;
hence the mass equation becomes i vi = 0. Since > 0, we can say
i vi = 0. (1.885)
So, for an incompressible fluid, the relative expansion rate for a fluid particle is zero.

1.8.2 Linear momenta

Let us first consider the viscous term:

1
j 2 (j vi) k vk ij , (1.886)
3 |{z}
=0



j 2 (j vi) , (1.887)
j ( (i vj + j vi )) , (1.888)
since is constant here (1.889)
(j i vj + j j vi ) , (1.890)

i j vj +j j vi , (1.891)
|{z}
=0
j j vi . (1.892)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

142 CHAPTER 1. GOVERNING EQUATIONS

Everything else in the linear momenta equation is unchanged; hence we get

o vi + vj j vi = fi i p + j j vi . (1.893)

Note that in the incompressible constant viscosity limit, the mass and linear momenta equa-
tions form a complete set of four equations in four unknowns: p,vi . We will see that in this
limit the energy equation is coupled to mass, and linear momenta, but it is only a one-way
coupling.

1.8.3 Energy
Let us also choose our material to be a liquid, for which the specific heat at constant pres-
sure, cp is nearly identical to the specific heat at constant volume cv as long as the ratio
T p2 /T //cp << 1. Here p is the coefficient of isobaric expansion, and T is the coefficient
of isothermal compressibility. As long as the liquid is well away from the vaporization point,
this is a good assumption for most materials. We will thus take for the liquid cp = cv = c.
For an incompressible gas there are some subtleties to this analysis, involving the low Mach
number limit which makes the results not obvious. We will not address that problem in this
course; many texts do, but many also shove the problem under the rug! For a compress-
ible gas there are no such problems. For an incompressible liquid whose specific heat is a
constant, we have e = cT + eo . The compressible energy equation in full generality is
de
= pi vi i qi + ij i vj . (1.894)
dt
Imposing our constitutive equations and assumption of incompressibility onto this, we get

d 1
(cT + eo ) = p i vi i (ki T ) + 2 (i vj) k vk ij i vj , (1.895)
dt |{z} 3 |{z}
=0 =0
dT
c = ki i T + 2(i vj) i vj , (1.896)
dt

= ki i T + 2 (i vj) (i vj) + [i vj] , (1.897)
| {z } | {z } | {z }
sym. sym. antisym.

= ki i T + 2(i vj) (i vj) . (1.898)

| {z }

For incompressible flows with constant properties, the viscous dissipation function reduces
to
= 2(i vj) (i vj) . (1.899)
It is a scalar function and obviously positive for > 0 since it is a tensor inner product of a
tensor with itself.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.9. DIMENSIONLESS COMPRESSIBLE NAVIER-STOKES EQUATIONS 143

1.8.4 Summary of incompressible constant property equations

The incompressible constant property equations for a liquid are summarized below in Gibbs
notation:

T v = 0, (1.900)
dv
= f p + 2 v, (1.901)
dt
dT
c = k2 T + . (1.902)
dt
For an ideal gas, it turns out that we should replace c in the above equation by cp . The
alternative, cv would seem to be the proper choice, but careful analysis in the limit of low
Mach number shows this to be incorrect.

1.8.5 Limits for one-dimensional diffusion

Note for a static fluid (vi = 0), we have d/dt = /t and = 0; hence the energy equation
can be written in a familiar form
T
= 2 T. (1.903)
t
Here = k/(c) is defined as the thermal diffusivity. For one dimensional cases where all
variation is in the x2 direction, we get
T 2T
= 2. (1.904)
t x2
Compare this to the momentum equation for a very specific form of the velocity field, namely,
vi (xi ) = v1 (x2 , t). When we also have no pressure gradient and no body force, the linear
momenta principle reduces to
v1 2 v1
= 2. (1.905)
t x2
Here = / is the momentum diffusivity. This equation has an identical form to that for
one-dimensional energy diffusion. In fact the physical mechanism governing both, random
molecular collisions, is the same.

1.9 Dimensionless compressible Navier-Stokes equations

Here we discuss how to scale the Navier-Stokes equations into a set of dimensionless equa-
tions. Panton gives a general background for scaling. Whites Viscous Flow has a detailed
discussion of the dimensionless form of the Navier-Stokes equations.
Consider the Navier-Stokes equations for a calorically perfect ideal gas which has Newto-
nian behavior, satisfies Stokes assumption, and has constant viscosity, thermal conductivity,
and specific heat:

CC BY-NC-ND. 17 June 2016, J. M. Powers.

144 CHAPTER 1. GOVERNING EQUATIONS

o + i (vi ) = 0, (1.906)
o (vi ) + j (vj vi ) = fi i p
  
1
+j 2 (j vi) k vk ji , (1.907)
3
     
1 1
o e + vj vj + i vi e + vj vj = vi fi i (pvi ) + ki i T
2 2
   
1
+i 2 (i vj) k vk ij vj , (1.908)
3
p = RT, (1.909)
e = cv T + e. (1.910)

Here R is the gas constant for the particular gas we are considering, which is the ratio of the
universal gas constant and the gass molecular mass M: R = /M. Also e is a constant.
Now solutions to the above equations, which may be of the form, for example, of
p(x1 , x2 , x3 , t), are necessarily parameterized by the constants from constitutive laws such as
cv , R, , k, fi , in addition to parameters from initial and boundary conditions. That is our
solutions will really be of the form

p(x1 , x2 , x3 , t; cv , R, , k, fi , . . .). (1.911)

It is desirable for many reasons to reduce the number of parametric dependencies of these
solutions. Some of these reasons include

identification of groups of terms that truly govern the features of the flow,

efficiency of presentation of results, and

efficiency of design of experiments.

The Navier-Stokes equations (and nearly all sets of physically motivated equations) can be
reduced in complexity by considering scaled versions of the same equations.
For a given problem, the proper scales are non-unique, though some choices will be more
helpful than others. One generally uses the following rules of thumb in choosing scales:

reduce variables so that their scaled value is near unity,

demonstrate that certain physical mechanisms may be negligible relative to other phys-
ical mechanisms, and

simplify initial and boundary conditions.

In forming dimensionless equations, one must usually look for

characteristic length scale L, and

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.9. DIMENSIONLESS COMPRESSIBLE NAVIER-STOKES EQUATIONS 145

vo

Po

Figure 1.32: Figure of known flow from infinity approaching body with characteristic length
L.

characteristic time scale tc .

Often an ambient velocity or sound speed exists which can be used to form either a length
or time scale, for example
L
given vo , L tc = vo
,

given vo , tc L = vo tc .
If for example our physical problem involves the flow over a body of length L (and whose
other dimensions are of the same order as L), and free-stream conditions are known to
be p = po , vi = (vo , 0, 0)T , = o , as sketched in Figure 1.32, Knowledge of free-stream
pressure and density fixes all other free-stream thermodynamic variables, e.g. e, T , via the
thermodynamic relations. For this problem, let the subscript represent a dimensionless
variable. Define the following scaled dependent variables:
p vi o R o
= , p = , vi = , T = T, e = e. (1.912)
o po vo po po
Define the following scaled independent variables:
xi vo
xi = , t = t. (1.913)
L L
With these definitions, the operators must also be scaled, that is,
dt vo vo
o = = = = o ,
t dt t L t L
L
o = o .
vo
dxi 1 1
i = = = = i ,
xi dxi xi L xi L
i = Li . (1.914)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

146 CHAPTER 1. GOVERNING EQUATIONS

1.9.1 Mass
Let us make these substitutions into the mass equation:

o + i (vi ) = 0, (1.915)
vo 1
o (o ) + i (o vo vi ) = 0, (1.916)
L L
o vo
(o + i ( vi )) = 0, (1.917)
L
o + i ( vi ) = 0. (1.918)

The mass equation is unchanged in form when we transform to a dimensionless version.

1.9.2 Linear momenta

We have a similar analysis for the linear momenta equation.

o (vi )
+j (vj vi ) = fi i p
  
1
+j 2 (j vi) k vk ji , (1.919)
3
vo
o (o vo vi )
L
1 1
+ j (o vo vj vo vi ) = o fi i (po p )
L L  
2 1
+ j (j vo vi) k vo vk ji , (1.920)
L L 3L
o vo2
o ( vi )
L
o v 2 po
+ o j ( vj vi ) = o fi i (p )
L L 
 
vo 1
+ 2 j 2 (j vi) k vk ji , (1.921)
L 3
fi L po
o ( vi ) + j ( vj vi ) = 2
i (p )
vo o vo2
 
2 1
+ j (j vi) k vk ji . (1.922)
o vo L 3
With this scaling, we have generated three distinct dimensionless groups of terms which
drive the linear momenta equation:
fi L po
, , and . (1.923)
vo2 o vo2 o vo L

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.9. DIMENSIONLESS COMPRESSIBLE NAVIER-STOKES EQUATIONS 147

These groups are closely related to the following groups of terms, which have the associated
interpretations indicated:

Froude number F r:45 With the body force per unit mass fi = ggi , where g > 0 is the
gravitational acceleration magnitude and gi is a unit vector pointing in the direction
of gravitational acceleration,

vo2 flow kinetic energy

F r2 = . (1.924)
gL gravitational potential energy

Mach number Mo :46 With the Mach number Mo defined as the ratio of the ambient
velocity to the ambient sound speed, and recalling that for a calorically perfect ideal
gas that the square of the ambient sound speed, a2o is a2o = poo , where is the ratio of
specific heats = ccpv = (1 + R/cv ), we have

vo2 vo2 o vo2 vo2 flow kinetic energy

Mo2 2
= po = = = . (1.925)
ao o po RTo thermal energy

Here we have taken To = po /o /R.

Reynolds number Re: We have

o vo L o v 2 dynamic pressure
Re = voo = . (1.926)
L viscous stress

With these definitions, we get

1 1 1
o ( vi ) + j ( vj vi ) = 2
gi i (p )
Fr Mo2
 
2 1
+ j (j vi) k vk ji . (1.927)
Re 3

The relative magnitudes of F r, Mo , and Re play a crucial role in determining which physical
mechanisms are most influential in changing the fluids linear momenta.

1.9.3 Energy
The analysis is of the exact same form, but more tedious, for the energy equation.
45
William Froude, 1810-1879, English engineer and naval architect, Oxford educated.
46
Ernst Mach, 1838-1926, Viennese physicist and philosopher who worked in optics, mechanics, and wave
dynamics, received doctorate at University of Vienna and taught mathematics at University of Graz and
physics at Charles University of Prague, developed fundamental ideas of inertia which influenced Einstein.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

148 CHAPTER 1. GOVERNING EQUATIONS

     
1 1
o e + vj vj + i vi e + vj vj = ki i T i (pvi )
2 2
   
1
+i 2 (i vj) k vk ij vj
3
+vi fi , (1.928)
  
vo po 1
o o e + vo2 vj vj
L 2
  o 
1 po 1 2 k po
+ i o vo vi e + vo vj vj = i i T
L o 2 L2 o R
1
i (po p vo vi )
L    
2 1
+ i (i vo vj) k vo vk ij vo vj
L L 3
+o vo vi fi , (1.929)
!!
o vo po 1 vo2
o e + vj vj
L o 2 poo
!!
2
o vo po 1 vo k po 1
+ i vi e + po vj vj = i i T
L o 2 o L2 o R
po vo
i (p vi )
L   
2vo2 1
+ 2 i (i vj) k vk ij vj
L 3
+o vo fi vi , (1.930)
!!
1 vo2
o e + vj vj
2 poo
!!
1 vo2 k
+i vi e + po vj vj = i i T
2 o LRo vo
i (p vi )
  
2vo2 L 1 1
+ 2 i (i vj) k vk ij vj
L o vo poo 3
fi L
+ po vi . (1.931)
o

Now examining the dimensionless groups, we see that

k k cp 1 k cp 1 1
= = = . (1.932)
LRo vo cp R Lo vo cp cp cv o vo L P r 1 Re

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.9. DIMENSIONLESS COMPRESSIBLE NAVIER-STOKES EQUATIONS 149

Here we have a new dimensionless group, the Prandtl47 number, P r, where

cp o momentum diffusivity
Pr = k
= = . (1.933)
k o cp
energy diffusivity

We also see that

fi L gLgi vo2 gL Mo2
po = = g i = gi , (1.934)
o
poo poo vo2 F r2
vo2
= Mo2 , (1.935)
poo
2vo2 L 1 2 vo2 1
2 p = po = 2 Mo2 . (1.936)
L o vo oo
o vo L o Re

So, the dimensionless energy equation becomes

  
1 2
o e + Mo vj vj
2
  
1 2 1 1
+i vi e + Mo vj vj = i i T
2 1 P r Re
i (p vi )
  
Mo2 1
+2 i (i vj) k vk ij vj
Re 3
2
Mo
+ gi vi . (1.937)
F r2

1.9.4 Thermal state equation

 
po
po p = o R T , (1.938)
o R
p = T . (1.939)

1.9.5 Caloric state equation

 
po po
e = cv T + e, (1.940)
o o R
cv o e
e = T + , (1.941)
R po
47
Ludwig Prandtl, 1875-1953, German mechanician and father of aerodynamics, primarily worked at Uni-
versity of Gottingen, discoverer of the boundary layer, pioneer of dirigibles, and advocate of monoplanes.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

150 CHAPTER 1. GOVERNING EQUATIONS

1 o e
e = T + (1.942)
1 po
|{z}
unimportant

For completeness, we retain the term pooe . It actually plays no role in this non-reactive flow
since energy only enters via its derivatives. When flows with chemical reactions are modeled,
this term may be important.

1.9.6 Upstream conditions

Scaling the upstream conditions, we get

p = 1, = 1, vi = (1, 0, 0)T . (1.943)

With this we then get secondary relationships

1 o e
T = 1, e = + . (1.944)
1 po

1.9.7 Reduction in parameters

We lastly note that our original system had the following ten independent parameters:

o , po , cv , R, L, vo , , k, fi , e. (1.945)

Our scaled system however has only six independent parameters:

o e
Re, P r, Mo , F r, , . (1.946)
po
Note we have lost no information, nor made any approximations, and we have a system with
fewer dependencies.

1.10 First integrals of linear momentum

Under special circumstances, we can integrate the linear momentum principle to obtain a
simplified equation. We will consider two cases here, what is known as Bernoullis48 equation
and Croccos49 equation. In a later chapter on rotational flows, we will also consider the
Helmholtz equation and Kelvins theorem, which are also first integrals in special cases.
48
Daniel Bernoulli, 1700-1782, Dutch-born Swiss mathematician of the prolific and mathematical Bernoulli
family, son of Johann Bernoulli, studied at Heidelberg, Strasbourg, and Basel, receiving M.D. degree, served
in St. Petersburg and lectured at the University of Basel, put forth his fluid mechanical principle in the 1738
Hydrodynamica, in competition with his fathers 1738 Hydraulica.
49
Luigi Crocco, 1909-1986, Sicilian-born, Italian applied mathematician and theoretical aerodynamicist
and rocket engineer, taught at University of Rome, Princeton, and Paris.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.10. FIRST INTEGRALS OF LINEAR MOMENTUM 151

1.10.1 Bernoullis equation

What we commonly call Bernoullis equation is really a first integral of the linear momenta
principle. Under different assumptions, we can get different flavors of Bernoullis equation.
A first integral of the linear momenta principle exists under the following conditions:

viscous stresses are negligible relative to other terms, ij 0,

the fluid is barotropic, p = p() or = (p).

body forces are conservative, so we can write fi = i , where is a known potential

function, and

either

the flow is irrotational, k = kij i vj = 0, or

the flow is steady, o = 0.

First consider a version of the general linear momenta equation in non-conservative form,
Eq. (1.470) scaled by :
1 1
o vi + vj j vi = i p + fi + j ji . (1.947)

Now use our vector identity, Eq. (1.171), to rewrite the advective term, and impose our
assumptions above to arrive at
 
1 1
o vi + i vj vj ijk vj k = i p i . (1.948)
2

Now let us define, just for this particular analysis, a new function . We will take to
be a function of pressure p, and thus implicitly, a function of xi and t. For the barotropic
fluid, we define as
Z p(xi ,t)
dp
(p(xi , t)) . (1.949)
po (p)
Note that in the special case of incompressible flow that = p/. Recalling Leibnizs rule,
Z x=b(t) Z x=b(t)
d f db da
f (x, t) dx = dx + f (b(t), t) f (a(t), t), (1.950)
dt x=a(t) x=a(t) t dt dt

we let /xi play the role of d/dt to get

Z p(xi ,t) Z p(xi ,t)  
dp 1 p 1 po 1
= = + dp. (1.951)
xi xi po (p) (p(xi , t)) xi (po ) xi po xi (p)
|{z} | {z }
=0 =0

CC BY-NC-ND. 17 June 2016, J. M. Powers.

152 CHAPTER 1. GOVERNING EQUATIONS

As po is constant, and the integrand has no explicit dependency on xi , we get

1 p
= . (1.952)
xi (p(xi , t)) xi
So, our linear momenta principle reduces to
 
1
o vi + i vj vj ijk vj k = i i . (1.953)
2
Consider now some special cases:

1.10.1.1 Irrotational case

If the fluid is irrotational, we have k = klm l vm = 0. Consequently, we can write the
velocity vector as the gradient of a potential function , known as the velocity potential:

m = vm . (1.954)

Note that if the velocity takes this form, then the vorticity is

k = klm l m . (1.955)

Since klm is anti-symmetric and l m is symmetric, their tensor inner product must be zero;
hence, such a flow is irrotational: k = klm l m = 0. So, the linear momenta principle,
Eq. (1.953), reduces to
 
1
o i + i (j )(j ) = i i , (1.956)
2
 
1
i o + (j )(j ) + + = 0, (1.957)
2
1
o + (j )(j ) + + = f (t). (1.958)
2
Here f (t) is an arbitrary function of time, which can be chosen to match conditions in a
given problem.

1.10.1.2 Steady case

1.10.1.2.1 Streamline integration Here we take o = 0, but k 6= 0. Rearranging the
steady version of the linear momenta equation, Eq. (1.953), we get
 
1
i vj vj + i + i = ijk vj k , (1.959)
2
 
1
i vj vj + + = ijk vj k . (1.960)
2

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.10. FIRST INTEGRALS OF LINEAR MOMENTUM 153

Taking the inner product of both sides with vi , we get

 
1
vi i vj vj + + = vi ijk vj k , (1.961)
2
= ijk vi vj k , (1.962)
| {z }
=0
= 0. (1.963)

The term on the right hand side is zero because it is the tensor inner product of a symmetric
and anti-symmetric tensor.
For a local coordinate system which has component s aligned with the velocity vector vi ,
and the other two directions n, and b, mutually orthogonal, we have vi = (vs , 0, 0)T . Our
linear momenta principle then reduces to

s []
(vs , 0, 0) n [] = 0. (1.964)
b []
Forming this dot product yields
 
1
vs vj vj + + = 0. (1.965)
s 2
For vs 6= 0, we get that
1
vj vj + + = C(n, b). (1.966)
2
On a particular streamline, the function C(n, b) will be a constant.

1.10.1.2.2 Lamb surfaces We can extend the idea of integration along a streamline to
describe what are known as Lamb surfaces50 by again considering the steady, inviscid linear
momentum principle with conservative body forces, Eq. (1.960):
 
1
i vj vj + + = ijk vj k . (1.967)
2
Now taking the quantity B to be
1
B vj vj + + , (1.968)
2
the linear momentum principle, Eq. (1.960), becomes

i B = ijk vj k (1.969)
50
Sir Horace Lamb, 1849-1934, English fluid mechanician, first studied at Owens College Manchester fol-
lowed by mathematics at Cambridge, taught at Adelaide, Australia, then returned to the University of
Manchester, prolific writer of textbooks.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

154 CHAPTER 1. GOVERNING EQUATIONS

Now the vector ijk vj k is orthogonal to both velocity vj and vorticity k because of the
nature of the cross product. Also the vector i B is orthogonal to a surface on which B
is constant. Consequently, the surface on which B is constant must be tangent to both
the velocity and vorticity vectors. Surfaces of constant B thus are composed of families of
streamlines on which the Bernoulli constant has the same value. In addition they contain
families of vortex lines. These are the Lamb surfaces of the flow, named after Sir Horace
Lamb, the British fluid mechanician of the late 19th and early 20th century.

1.10.1.3 Irrotational, steady, incompressible case

In this case, we recover the form most commonly used (and misused) of Bernoullis equation,
namely,
1
vj vj + + = C. (1.970)
2
The constant is truly constant throughout the flow field. With = p/ here and = gz z
(with gz > 0, and rising z corresponding to rising distance from the earths surface, we get
f = = gz k) for a constant gravitational field, and v the magnitude of the velocity
vector, we get
1 2 p
v + + gz z = C. (1.971)
2

1.10.2 Croccos theorem

It is common, especially in texts on compressible flow, to present what is known as Croccos
theorem. The many different versions presented in many standard texts are non-uniform and
often of unclear validity. Its utility is confined mainly to providing an alternative way of
expressing the linear momentum principle which provides some insight into the factors which
influence fluid motion. In special cases, it can be integrated to form a more useful relation-
ship, similar to Bernoullis equation, between fundamental fluid variables. The heredity of
this theorem is not always clear, though, as we shall see it is nothing more than a combina-
tion of the linear momentum principle coupled with some definitions from thermodynamics.
Its derivation is often confined to inviscid flows. Here we will first present a result valid
for general viscous flows for the evolution of stagnation enthalpy, which is closely related to
Croccos theorem. Next we will show how one of the restrictions can be relaxed so as to
obtain what we call the extended Croccos theorem. We then show how this reduces to a
form which is similar to a form presented in many texts.

1.10.2.1 Stagnation enthalpy variation

First again consider the general linear momenta equation, Eq. (1.947):

1 1
o vi + vj j vi = i p + fi + j ji . (1.972)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.10. FIRST INTEGRALS OF LINEAR MOMENTUM 155

Now, as before in the development of Bernoullis equation, use our vector identity, Eq. (1.171),
to rewrite the advective term, but retain the viscous terms to get
 
1 1 1
o vi + i vj vj ijk vj k = i p + fi + j ji . (1.973)
2
Taking the dot product with vi , and rearranging, we get
   
1 1 1 1
o vi vi + vi i vj vj = ijk vi vj k vi i p + vi fi + vi j ji . (1.974)
2 2 | {z }
=0

Again, since ijk is anti-symmetric and vi vj is symmetric, their tensor inner product is zero,
so we get    
1 1 1 1
o vi vi + vi i vj vj = vi i p + vi fi + vi j ji . (1.975)
2 2
Now recall the Gibbs relation from thermodynamics, Eq. (1.531):
p
T ds = de d. (1.976)
2
Also recall the definition of enthalpy h, Eq. (1.521):
p
h=e+ . (1.977)

Differentiating the equation for enthalpy, we recover Eq. (1.524):
1 p
dh = de + dp 2 d. (1.978)

Eliminating de in favor of dh in the Gibbs equation gives
1
T ds = dh dp. (1.979)

If we choose to apply this relation to the motion following a fluid particle, we can say then
that
ds dh 1 dp
T = . (1.980)
dt dt dt
Expanding, we get
1
T (o s + vi i s) = o h + vi i h (o p + vi i p). (1.981)

Rearranging, we get
1 1
T (o s + vi i s) (o h + vi i h) + o p = vi i p. (1.982)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

156 CHAPTER 1. GOVERNING EQUATIONS

We then use the above identity to eliminate the pressure gradient term from the linear
momentum equation in favor of enthalpy, entropy, and unsteady pressure terms:
   
1 1 1 1
o vi vi + vi i vj vj = T (o s + vi i s) (o h + vi i h) + o p + vi fi + vi j ji . (1.983)
2 2
Rearranging slightly, noting that vi vi = vj vj , and assuming the body force is conservative so
that fi = i , we get
   
1 1 1 1
o h + vj vj + + vi i h + vj vj + = T (o s + vi i s) + o p + vi j ji . (1.984)
2 2

Note that here we have made the common assumption that the body force potential is
independent of time, which allows us to absorb it within the time derivative. If we define,
as is common, the total enthalpy ho as
1
ho = h + vj vj + , (1.985)
2
we can then state
1 1
o ho + vi i ho = T (o s + vi i s) + o p + vi j ji , (1.986)

dho ds 1 p 1 T
T
= T + + v T (1.987)
dt dt t
We can use the first law of thermodynamics written in terms of entropy, Eq. (1.535),
(ds/dt) = (1/T )i qi + (1/T )ij i vj , to eliminate the entropy derivative in favor of those
terms which generate entropy to arrive at
dho
= i (ij vj qi ) + o p. (1.988)
dt
Thus, we see that the total enthalpy of a fluid particle is influenced by energy and momentum
diffusion as well as an unsteady pressure field.

1.10.2.2 Extended Croccos theorem

With a slight modification of the preceding analysis, we can arrive at the extended Croccos
theorem. Begin once more with an earlier version of the linear momenta principle:
 
1 1 1
o vi + i vj vj ijk vj k = i p + fi + j ji . (1.989)
2
Now assume we have a functional representation of enthalpy in the form

h = h(s, p). (1.990)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

1.10. FIRST INTEGRALS OF LINEAR MOMENTUM 157

Then we get  
h  h 
dh = ds + dp. (1.991)
s p p s
We also thus deduce from the Gibbs relation dh = T ds + (1/)dp that
 
h  h  1
 = T,  = . (1.992)
s p p s

Now, since we have h = h(s, p), we can take its derivative with respect to each and all of the
coordinate directions to obtain
 
h h  s h  p
= + . (1.993)
xi s p xi p s xi
or  
h  h 
i h = i s + i p. (1.994)
s p p s
Substituting known values for the thermodynamic derivatives, we get
1
i h = T i s + i p. (1.995)

We can use this to eliminate directly the pressure gradient term from the linear momentum
equation to obtain then
 
1 1
o vi + i vj vj ijk vj k = T i s i h + fi + j ji . (1.996)
2

Rearranging slightly, and again assuming the body force is conservative so that fi = i ,
we get the extended Croccos theorem:
 
1 1
o vi + i h + vj vj + = T i s + ijk vj k + j ji . (1.997)
2

Again, employing the total enthalpy, ho = h + 12 vj vj + , we write the extended Croccos

theorem as
1
o vi + i ho = T i s + ijk vj k + j ji . (1.998)

1.10.2.3 Traditional Croccos theorem

For a steady, inviscid flow, the extended Croccos theorem reduces to what is usually called
Croccos theorem:

i ho = T i s + ijk vj k , (1.999)
ho = T s + v . (1.1000)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

158 CHAPTER 1. GOVERNING EQUATIONS

If the flow is further required to be homeoentropic, we get

i ho = ijk vj k . (1.1001)

Similar to Lamb surfaces, we find that surfaces on which ho is constant are parallel to both
the velocity and vorticity vector fields. Taking the dot product with vi , we get

vi i ho = vi ijk vj k , (1.1002)
= ijk vi vj k , (1.1003)
= 0. (1.1004)

Integrating this along a streamline, as for Bernoullis equation, we find

ho = C(n, b), (1.1005)

1
h + vj vj + = C(n, b), (1.1006)
2
so we see that the stagnation enthalpy is constant along a streamline and varies from stream-
line to streamline. If the flow is steady, homeoentropic, and irrotational, the total enthalpy
will be constant throughout the flow-field:
1
h + vj vj + = C. (1.1007)
2

CC BY-NC-ND. 17 June 2016, J. M. Powers.

Chapter 2

Vortex dynamics

see Panton, Chapter 13,

see Yih, Chapter 2.

In this chapter we will consider in detail the kinematics and dynamics of rotating fluids,
sometimes called vortex dynamics. The two most common quantities which are used to
characterize rotating fluids are
the vorticity vector = v, and
H
the circulation = C vT dr.
Both will be important in this chapter.
Although it is entirely possible to use Cartesian index notation to describe a rotating
fluid, some of the ideas are better conveyed in a non-Cartesian system, such as the cylindrical
coordinate system. For that reason, and for the sake of giving the student more experience
with the other common notation, the Gibbs notation will often be used in the chapter.

2.1 Transformations to cylindrical coordinates

The rotation of a fluid about an axis induces an acceleration in that a fluid particles velocity
vector is certainly changing with respect to time. Such a motion is most easily described
with a set of cylindrical coordinates. The transformation and inverse transformation to and
from cylindrical (r, , z) coordinates to Cartesian (x, y, z) is given by the familiar
p
x = r cos , r = x2 + y 2 , (2.1)
y
y = r sin , = tan1 , (2.2)
x
z = z, z = z. (2.3)

Most of the basic distinctions between the two systems can be understood by considering two-
dimensional geometries. The representation of an arbitrary point in both two-dimensional

159
160 CHAPTER 2. VORTEX DYNAMICS

e
r

r
j

i x

Figure 2.1: Representation of a point in Cartesian and cylindrical coordinates along with
unit vectors for both systems.

(x, y) Cartesian and two-dimensional (r, ) cylindrical coordinate systems along with the
unit basis vectors for both systems, i, j, and er , e , is sketched in Figure 2.1,

2.1.1 Centripetal and Coriolis acceleration

The fact that a point in motion is accompanied by changes in the basis vectors with respect
to time in the cylindrical representation, but not for Cartesian basis vectors, accounts for
the most striking differences in the formulations of the governing equations, namely the
appearance of

centripetal acceleration, and

Coriolis1 acceleration

in the cylindrical representation.

Consider the representations of the velocity vector v in both coordinate systems:

v = ui + vj, or (2.4)
v = vr er + v e . (2.5)

Now the unsteady (as opposed to the convective) part of the acceleration vector of a particle
is simply the partial derivative of the velocity vector with respect to time. Now formally, we
must allow for variations of the unit basis vectors as well as the components themselves so
1
Gaspard Gustave de Coriolis, 1792-1843, Paris-born mathematician, taught with Navier, introduced the
terms work and kinetic energy with modern scientific meaning, wrote on the mathematical theory of
billiards.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

2.1. TRANSFORMATIONS TO CYLINDRICAL COORDINATES 161

sin i
e
r
sin j cos j
e

cos i

Figure 2.2: Geometrical representation of cylindrical unit vectors in terms of Cartesian unit
vectors.

that
v u i v j
= i+u + j+v , (2.6)
t t t
|{z} t t
|{z}
=0 =0
v vr er v e
= er + vr + e + v . (2.7)
t t t t t
Now the time derivatives of the Cartesian basis vectors is zero, as they are defined not to
change with the position of the particle. Hence for a Cartesian representation, we have for
the unsteady component of acceleration the familiar:
v u v
= i + j. (2.8)
t t t
However the time derivative of the cylindrical basis vectors does change with time for
particles in motion! To see this, let us first relate er and e to i and j. From the sketch of
Figure 2.2, it is clear that
er = cos i + sin j, (2.9)
e = sin i + cos j. (2.10)
This is a linear system of equations. We can use Cramers rule to invert to find
i = cos er sin e , (2.11)
j = sin er + cos e . (2.12)
Now, examining time derivatives of the unit vectors, we see that
er
= sin i + cos j, (2.13)
t t t

= e , (2.14)
t
and
e
= cos i sin j, (2.15)
t t t

= er . (2.16)
t

CC BY-NC-ND. 17 June 2016, J. M. Powers.

162 CHAPTER 2. VORTEX DYNAMICS

v dt = ds

Figure 2.3: Sketch of relation of differential distance ds to velocity in angular direction v .

so there is a formal variation of the unit vectors with respect to time as long as the angular
velocity
t
6 0. So the acceleration vector is
=

v vr v
= er + vr e + e v er , (2.17)
t t t  t
 t 
vr v
= v er + + vr e . (2.18)
t t t t

Now from basic geometry, as sketched in Figure 2.3, we have

ds = rd, (2.19)
v dt = rd, (2.20)
v
= . (2.21)
r t
Consequently, we can write the unsteady component of acceleration as

v vr v
2
er + v vr v
= t + e . (2.22)
t r
|{z} t r
|{z}
centripetal Coriolis

Two, apparently new, accelerations have appeared as a consequence of the transformation:

v2
centripetal acceleration, r , directed towards the center, and Coriolis acceleration, vrrv ,
directed in the direction of increasing . These terms do not have explicit dependency on
time derivatives of velocity. And yet when the equations are constructed in this coordinate
system, they represent real accelerations, and are consequences of forces. As can be seen
by considering the general theory of non-orthogonal coordinate transformations, terms like
the centripetal and Coriolis acceleration are associated with the Christoffel symbols of the
transformation.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

2.1. TRANSFORMATIONS TO CYLINDRICAL COORDINATES 163

Such terms perhaps contributed to the development of Einsteins theory of general rel-
ativity as well. Refusing to accept that our typical expression of a body force, mg, was
fundamental, Einstein instead postulated that it was a term which was a relic of a coordinate
transformation. He held that we in fact exist in a more complex geometry than classically
considered. He constructed his theory of general relativity such that no gravitational force
exists, but when coordinate transformations are employed to give us a classical view of the
non-relativistic universe, the term mg appears in much the same way as centripetal and
Coriolis accelerations appear when we transform to cylindrical coordinates.

2.1.2 Grad and div for cylindrical systems

We can use the chain rule to develop expressions for grad and div in cylindrical coordinate
systems. Consider the Cartesian

= i+ j + k. (2.23)
x y z

The chain rule gives us

r z
= + + , (2.24)
x x r x x z
r z
= + + , (2.25)
y y r y y z
r z
= + + , (2.26)
z z r z z z
(2.27)

Now, we have
r 2x x
= p = = cos , (2.28)
x 2 x2 + y 2 r
r 2y y
= p = = sin , (2.29)
y 2
2 x +y 2 r
r
= 0, (2.30)
z
and
y r sin sin
= 2 2
= 2 = , (2.31)
x x +y r r
x r cos cos
= 2 2
= 2
= , (2.32)
y x +y r r

= 0, (2.33)
z

CC BY-NC-ND. 17 June 2016, J. M. Powers.

164 CHAPTER 2. VORTEX DYNAMICS

and

z
= 0, (2.34)
x
z
= 0, (2.35)
y
z
= 1, (2.36)
z
so

sin
= cos , (2.37)
x r r
cos
= sin + , (2.38)
y r r

= . (2.39)
z z

2.1.2.1 Grad

So now we are prepared to write an explicit form for in cylindrical coordinates:

 
sin
= cos (cos er sin e )
r r | {z }
| {z } i

x
 
cos
+ sin + (sin er + cos e ) + ez , (2.40)
r r | {z } z
| {z } j

y
   
2

2 sin cos sin cos
= cos + sin + + er
r r r
  2  
sin cos2
+ ( sin cos + sin cos ) + + e
r r r

+ ez , (2.41)
z
1
= er + e + ez . (2.42)
r r z

We can now write a simple expression for the convective component, vT , of the acceleration
vector:
v
vT = vr + + vz . (2.43)
r r z

CC BY-NC-ND. 17 June 2016, J. M. Powers.

2.1. TRANSFORMATIONS TO CYLINDRICAL COORDINATES 165

2.1.2.2 Div
The divergence is straightforward. In Cartesian coordinates we have

u v w
T v = + + . (2.44)
x y z

In cylindrical, we replace derivatives with respect to x, y, z with those with respect to r, , z,

so
u sin u v cos v w
T v = cos + sin + + . (2.45)
r r r r z
Now u, v and w transform in the same way as x, y, and z, so

u = vr cos v sin , (2.46)

v = vr sin + v cos , (2.47)
w = vz . (2.48)

Substituting and taking partials, we find that

vr v sin vr v
T v = cos cos sin cos sin vr sin cos v
r | {z }
| {zr} r | {z}
C
A
B
vr v cos vr v
+ sin sin + cos + sin + cos vr + cos sin v
r | {z }
| {zr} r | {z}
C
A B
vz
+ . (2.49)
z
When expanded, the terms labeled A, B, and C cancel in the above expression. Then using
the trigonometric identity sin2 + cos2 = 1, we arrive at the simple form

vr vr 1 v vz
T v = + + + , (2.50)
r r r z
which is often rewritten as
1 1 v vz
T v = (rvr ) + + . (2.51)
r r r z
Using the same procedure, we can show that the Laplacian operator transforms to
 
2 1 1 2 2
= r + 2 2 + 2. (2.52)
r r r r z

CC BY-NC-ND. 17 June 2016, J. M. Powers.

166 CHAPTER 2. VORTEX DYNAMICS

2.1.3 Incompressible Navier-Stokes equations in cylindrical coor-

dinates
Leaving out some additional details of the transformations, we find that the incompressible
Navier-Stokes equations for a Newtonian fluid with constant viscosity and body force confined
to the z direction are
1 1 v vz
0 = (rvr ) + + , (2.53)
  r r r z 
vr v2 vr v vr vr 1 p 2 vr2 2 v
+ vr + + vz = + vr 2 2 ,(2.54)
t r r r z r r r
   
v vr v v v v v 1 1 p 2 2 vr v
+ + vr + + vz = + v + 2 2 (2.55)
,
t r r r z r r r
vz vz v vz vz 1 p
+ vr + + vz = + 2 vz gz . (2.56)
t r r z z

Note that in the acceleration terms, strictly unsteady terms, convective terms as well as
centripetal and Coriolis terms appear. Also note that the viscous terms have additional
complications that we have not considered in detail but arise because we must transform
2 v, and there are many non-intuitive terms which arise here when expanded in full.

2.2 Ideal rotational vortex

Let us consider the kinematics and dynamics of an ideal rotational vortex, which we define
to be a fluid rotating as a solid body. Let us assume incompressible flow, so T v = 0,
assume a simple velocity field, and ask what forces could have given rise to that velocity
field. We will simply use z for the azimuthal coordinate instead of z here. Take
r
vr = 0, v = , vz = 0. (2.57)
2
The kinematics of this flow are simple and sketched in Figure 2.4, Here is now defined as
a constant. The velocity is zero at the origin and grows in amplitude with linear distance
from the origin. The flow is steady, and the streamlines are circles centered about the origin.
Obviously, as r , the theory of relativity would suggest that such a flow would break
down as the velocity approached the speed of light. In fact, one would find as well that as the
velocities approached the sound speed that compressibility effects would become important
far before relativistic effects.
Whatever the case, does this assumed velocity field satisfy incompressible mass conser-
vation?
1  r 
?
1 z}|{
(r(0)) + + (0) = 0. (2.58)
r r r | {z2 } z
=0

CC BY-NC-ND. 17 June 2016, J. M. Powers.

2.2. IDEAL ROTATIONAL VORTEX 167

Figure 2.4: Sketch of a fluid rotating as a pure solid body.

Obviously it does.
Next let us consider the acceleration of an element of fluid and the forces which could
give rise to that acceleration. First consider the material derivative for this flow
d v v
= + vr + + vz = . (2.59)
dt |{z}
t |{z} r r |{z} z r
=0 =0
=0

But the only non-zero component of velocity, v , has no dependency on , so the material
derivative of velocity dv
dt
= 0.
Consider now the viscous terms for this flow. We recall for an incompressible Newtonian
fluid that

ij = 2(i vj) + k vk ij , (2.60)

|{z}
=0
= (i vj + j vi ) , (2.61)
j ij = (j i vj + j j vi ) , (2.62)

= i j vj +j j vi , (2.63)
|{z}
=0
2
= v (2.64)

We also note that

= ijk j k = ijk j kmn m vn , (2.65)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

168 CHAPTER 2. VORTEX DYNAMICS

= kij kmn j m vn , (2.66)

= (im jn in jm ) j m vn , (2.67)
= j i vj j j vi , (2.68)
= i j vj j j vi , (2.69)
|{z}
=0
= j j vi . (2.70)

Comparing, we see that for this incompressible flow,

T
T T = ( ). (2.71)

Now, using relations that can be developed for the curl in cylindrical coordinates, we have
for this flow that
1 vz v
r = = 0, (2.72)
r z
vr vz
= = 0, (2.73)
z r
1 1 vr
z = (rv ) , (2.74)
r r r
1  r 
= r , (2.75)
r r 2
= . (2.76)

So the flow has a constant rotation rate, . Since it is constant, its curl is zero, and we have

T
for this flow that T T = 0. We could just as well show for this flow that = 0. That
is because the kinematics are those of pure rotation as a solid body with no deformation.
No deformation implies no viscous stress.
Hence, the three linear momenta equations in the cylindrical coordinate system reduce
to the following:

v2 1 p
= , (2.77)
r r
1 1 p
0 = , (2.78)
r
1 p
0 = gz . (2.79)
z

The r momentum equation strikes a balance between centripetal inertia and radial pressure
gradients. The momentum equation shows that as there is no acceleration in this direction,
there can be no net pressure force to induce it. The z momentum equation enforces a balance
between pressure forces and gravitational body forces.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

2.2. IDEAL ROTATIONAL VORTEX 169

If we take p = p(r, , z) and p(ro , , zo ) = po , then

p p p
dp = dr + d + dz, (2.80)
r z
2
v
= dr + 0d gz dz, (2.81)
r
2 r 2
= dr gz dz, (2.82)
4r
2 r
= dr gz dz, (2.83)
4
2 2
p po = (r ro2 ) gz (z zo ), (2.84)
8
2 2
p(r, z) = po + (r ro2 ) gz (z zo ). (2.85)
8
Now on a surface of constant pressure we have p(r, z) = p. So

2 2
p = po + (r ro2 ) gz (z zo ), (2.86)
8
2 2
gz (z zo ) = po p + (r ro2 ), (2.87)
8
po p 2 2
z = zo + + (r ro2 ). (2.88)
gz 8gz
So a surface of constant pressure is a parabola in r with a minimum at r = 0. This is
consistent with what one observes upon spinning a bucket of water.
Now lets rearrange our general equation for the pressure field and eliminate using
v = r
2
and defining vo = r2 o :

1 1 2
p v2 + gz z = po vo + gz zo = C. (2.89)
2 2
This looks very similar to the steady irrotational incompressible Bernoulli equation in which
p + 12 v 2 + gz z = K. But there is a difference in the sign on one of the terms. Now add v2
to both sides of the equation to get
1
p + v2 + gz z = C + v2 . (2.90)
2
Now since v = r2
,vr = 0, we have lines of constant r as streamlines, and v is constant on
those streamlines, so that we get
1
p + v2 + gz z = C , on a streamline. (2.91)
2
Here C varies from streamline to streamline.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

170 CHAPTER 2. VORTEX DYNAMICS

x
2

dr = a d e

x
1

Figure 2.5: Sketch of an ideal irrotational point vortex.

We lastly note that the circulation for this system depends on position. If we choose our
contour integral to be a circle of radius a about the origin we find
I
= vT dr, (2.92)
C
Z 2  
1
= a (ad), (2.93)
0 2
= a2 . (2.94)

2.3 Ideal irrotational vortex

Now let us perform a similar analysis for the following velocity field:
o
vr = 0, v = , vz = 0. (2.95)
2r
The kinematics of this flow are also simple and sketched in Figure 2.5, We see once again
that the streamlines are circles about the origin. But here, as opposed to the ideal rotational
vortex, v 0 as r and v as r 0. The vorticity vector of this flow is
1 vz v
r = = 0, (2.96)
r z
vr vz
= = 0, (2.97)
z r
1 1 vr
z = (rv ) , (2.98)
r r r

CC BY-NC-ND. 17 June 2016, J. M. Powers.

2.3. IDEAL IRROTATIONAL VORTEX 171

 
1 o
= r , (2.99)
r r 2r
 
1 o
= = 0! (2.100)
r r 2

This flow field, which seems the epitome of a rotating flow, is formally irrotational as it
has zero vorticity. What is happening is that a fluid element not at the origin is actually
undergoing severe deformation as it rotates about the origin; however, it does not rotate
about its own center of mass. Therefore, the vorticity vector is zero, except at the origin,
where it is undefined.
The circulation for this flow about a circle of radius a is
I
= vT dr, (2.101)
C
Z 2
= v (ad), (2.102)
0
Z 2
o
= ad, (2.103)
0 2a
= o . (2.104)

So the circulation is independent of the radius of the closed contour. In fact it can be shown
that as long as the closed contour includes the origin in its interior that any closed contour
will have this same circulation. We call o the ideal irrotational vortex strength, in that it
is proportional to the magnitude of the velocity at any radius.
Let us once again consider the forces which could induce the motion of this vortex if the
flow happens to be incompressible with constant properties and in a potential field where the

T
gravitational body force per unit mass is gz k. Recall again that T = ( ),

T
and that since = 0 that T = 0 for this flow. Note also that because there is
deformation here, that itself is not zero, its divergence is. For example, if we consider
one component of viscous stress r and use standard relations which can be derived for
incompressible Newtonian fluids, we find that
   
 v  1 vr o o
r = r + = r = . (2.105)
r r r r 2r 2 r 2

The equations of motion reduce to the same ones as for the ideal rotational vortex:

v2 1 p
= , (2.106)
r r
1 1 p
0 = , (2.107)
r
1 p
0 = gz . (2.108)
z

CC BY-NC-ND. 17 June 2016, J. M. Powers.

172 CHAPTER 2. VORTEX DYNAMICS

Once more we can deduce a pressure field which is consistent with these and the same set
of conditions at r = ro , z = zo , with p = po :
p p p
dp = dr + d + dz, (2.109)
r
|{z} z
=0
v2
= dr gz dz, (2.110)
r
2o dr
= gz dz, (2.111)
4 2 r 3 
2o 1 1
p po = 2 gz (z zo ), (2.112)
8 r 2 ro2
2o 1 2 1
p+ + gz z = po + 2o 2 + gz zo , (2.113)
8 2 r 2 8 ro
1 1 2
p + v2 + gz z = po + vo + gz zo = C (2.114)
2 2
(2.115)

This is once again Bernoullis equation. Here it is for an irrotational flow field that is also
time-independent, so the Bernoulli constant C is truly constant for the entire flow field and
not just along a streamline.
On isobars we have p = p which gives us
 
2o 1 1
p po = 2 gz (z zo ), (2.116)
8 r 2 ro2
 
po p 2o 1 1
z = zo + + 2 (2.117)
gz 8 gz r 2 ro2
Note that the pressure goes to negative infinity at the origin. One can show that actual
forces, obtained by integrating pressure over area, are in fact bounded.

2.4 Helmholtz vorticity transport equation

Here we will take the curl of the linear momenta principle to obtain a relationship, the
Helmholtz vorticity transport equation, which shows how the vorticity field evolves in a
general fluid.

2.4.1 General development

First, we recall some useful vector identities:
 T 
T v v
(v )v = + v, (2.118)
2

CC BY-NC-ND. 17 June 2016, J. M. Powers.

2.4. HELMHOLTZ VORTICITY TRANSPORT EQUATION 173

(a b) = (bT )a (aT )b + a(T b) b(T a), (2.119)

() = 0, (2.120)
( v) = T = 0.
T
(2.121)

The first is Eq. (1.171); the others are easily proved.

We start now with the linear momenta principle for a general fluid:
v 1 1
T
+ (vT )v = f p + T . (2.122)
t
We expand the term (vT )v and then apply the curl operator to both sides to get
  T    
v v v 1 1 T

T
+ + v = f p + . (2.123)
t 2
This becomes, via the linearity of the various operators,
  T     
v v 1 1 T

T
( v) + + v = f p + .
t | {z } 2
| {z }
=0
(2.124)
Using our vector identity for the term with two cross products we get
   
T T T T 1 1 T

T
+ (v ) ( )v+( | {z v} )v(
| {z }) = f p + .
|t {z } =0
= 1 d
= ddt
dt

(2.125)
Rearranging, we have
   
d d T 1 1 T

T
= ( )v + f p + , (2.126)
dt dt
 T     
1 d d 1 1 1 1 1 T

T
2 = v+ f p + , (2.127)
dt dt
   T     
d 1 1 1 1 1
T
= v+ f p + T , (2.128)
dt
 
1
Now consider the term
p . In Einstein notation, we have
   
1 1 1
ijk j k p = ijk j k p 2 (j )(k p) , (2.129)

1 1
= ijk j k p ijk (j )(k p), (2.130)
| {z } 2
=0
1
= 2 p. (2.131)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

174 CHAPTER 2. VORTEX DYNAMICS

Multiplying both sides by , we write the final general form of the vorticity transport equation
as    
d T

1 1 T

T
= v + |{z f} + 2 p + . (2.132)
dt | {z }
B | {z } | {z }
A
C D
Here we see the evolution of the vorticity scaled by the density is affected by four physical
processes, which we describe in greater detail directly, namely
A: bending and stretching of vortex tubes,
B: non-conservative body forces (if f = , then f is conservative, and f =
= 0. For example f = gz k gives = gz z),
C: non-barotropic, also known as baroclinic, effects (if a fluid is barotropic, then
p = p() and p = (dp/d) thus p = (dp/d) = 0.), and
D: viscous effects.

2.4.2 Incompressible conservative body force limit

The Helmholtz vorticity transport equation (2.132) reduces significantly in special limiting
cases involving incompressible flow in the limit of a conservative body force. In this limit
Eq. (2.132) reduces to the following
d 1
= ( T )v + (T )T . (2.133)
dt

2.4.2.1 Isotropic, Newtonian, constant viscosity

Now if we further require that the fluid be isotropic and Newtonian with constant viscosity,
the viscous term can be written as
(T )T = ijk j m (2((m vk) (1/3) l vl mk )), (2.134)
|{z}
=0
= ijk j m (m vk + k vm ), (2.135)
= ijk j (m m vk + m k vm ), (2.136)
= ijk j (m m vk + k m vm ), (2.137)
| {z }
=0
= m m ijk j vk , (2.138)
| {z }

2
= . (2.139)
So we get, recalling that = /,
d
= ( T )v + 2 . (2.140)
dt

CC BY-NC-ND. 17 June 2016, J. M. Powers.

2.4. HELMHOLTZ VORTICITY TRANSPORT EQUATION 175

2.4.2.2 Two-dimensional, isotropic, Newtonian, constant viscosity

If we further require two-dimensionality, then we have = (0, 0, 3(x1 , x2 ))T , and =
(1 , 2 , 0)T , so T = 0. Thus, we get the very simple
 2 
d3 2 3 2 3
= 3 = + . (2.141)
dt x21 x22
If the flow is further inviscid = 0, we get
d3
= 0, (2.142)
dt
and we find that there is no tendency for vorticity to change along a streamline. If we further
have an initially irrotational state, then we get = 0 for all space and time.

2.4.3 Physical interpretations

Let us consider how two of the terms in Eq. (2.132) contribute to the generation of vorticity.

2.4.3.1 Baroclinic (non-barotropic) effects

If a fluid is barotropic then we can write p = p(), or = (p). An isentropic calorically
perfect ideal gas has p/po = (/o ) , where is the ratio of specific heats, and the o subscript
indicates a constant value. Such a gas is barotropic. for such a fluid, we must have by the
chain rule that i p = (dp/d)i . Hence p and are vectors which point in the same
direction. Moreover, isobars (lines of constant pressure) must be parallel to isochores (lines
of constant density). If, as sketched in Figure 2.6. we calculate the resultant vector from the
net pressure force, as well as the center of mass for a finite fluid volume, we would see that
the resultant force had no lever arm with the center of gravity. Hence it would generate no
torque, and no tendency for the fluid element to rotate about its center of mass, hence no
vorticity would be generated by this force.
For a baroclinic fluid, we do not have p = p(); hence, we must expect that p points
in a different direction than . If we examine this scenario, as sketched in Figure 2.7, we
discover that the resultant force from the pressure has a non-zero lever arm with the center
of mass of the fluid element. Hence, it generates a torque, a tendency to rotate the fluid
about G, and vorticity.

2.4.3.2 Bending and stretching of vortex tubes

Now let us consider generation of vorticity by three-dimensional effects. Such effects are
commonly characterized as the bending and stretching of what is known as vortex tubes.
Here we focus on just the following inviscid equation:
d
= ( T )v. (2.143)
dt

CC BY-NC-ND. 17 June 2016, J. M. Powers.

176 CHAPTER 2. VORTEX DYNAMICS

low

plow

high

phigh

F
net pressure

Figure 2.6: Isobars and isochores, center of mass G, and center of pressure for barotropic
fluid.

low plow

high

G phigh

F
net pressure

Figure 2.7: Isobars and isochores, center of mass G, and center of pressure for baroclinic
fluid.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

2.4. HELMHOLTZ VORTICITY TRANSPORT EQUATION 177

b
s
curve everywhere parallel
to vorticity vector

locally orthogonal
n
coordinate system
s,n,b

Figure 2.8: Local orthogonal intrinsic coordinate system oriented with local vorticity field.

If we consider a coordinate system which is oriented with the vorticity field as sketched in
Figure 2.8, we will get many simplifications. We take the following directions
s: the streamwise direction parallel to the vorticity vector,

n: the principal normal direction, pointing towards the center of curvature,

b: the biorthogonal direction, orthogonal to s and n.

With this system, we can say that


s
( T )v = ( s 0 0)
n
v, (2.144)

b
v
= s . (2.145)
s
So for the inviscid flow we have
d v
= s . (2.146)
dt s
We have in terms of components
ds vs
= s , (2.147)
dt s
dn vn
= s , (2.148)
dt s
db vb
= s . (2.149)
dt s
The term vs
s
we know from kinematics represents a local stretching or extension. Just as
a rotating figure skater increases his or her angular velocity by concentrating his or her
mass about a vertical axis, so does a rotating fluid. The first of these expressions says that
the component of rotation aligned with the present increases if there is stretching in that
direction. This is sketched in Figure 2.9,
The second and third terms enforce that if vn or vb are changing in the s direction, when
accompanied by non-zero s , that changes in the non-aligned components of are induced.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

178 CHAPTER 2. VORTEX DYNAMICS

unstretched
vortex tube

stretched
vortex
tube

Figure 2.9: Increase in vorticity due to stretching of vortex tube.

Hence the previously zero components n , b acquired non-zero values, and the lines parallel
to the vorticity vector bend. Hence, we have the term, bending of vortex tubes.
It is generally accepted that the bending and stretching of vortex tubes is an important
mechanism in the transition from laminar to turbulent flow.

2.5 Kelvins circulation theorem

Kelvins circulation theorem describes how the circulation of a material region in a fluid
changes with time. We first recall the definition of circulation :
I
= vT dx, (2.150)
C

where C is a closed contour. We next take the material derivative of to get

I
d d
= vT dx, (2.151)
dt dt
I CT I
dv d
= dx + vT dx, (2.152)
dt dt
IC T IC  
dv T dx
= dx + v d , (2.153)
C dt C dt
I I
dvT
= dx + vT dv, (2.154)
dt
IC T IC  
dv 1 T
= dx + d v v , (2.155)
C dt C 2
| {z }
=0
I  T
dv
= dx. (2.156)
C dt

Here we note that because we have chosen a material region for our closed contour that dx dt
must be the fluid particle velocity. This then allows us to write the second term as a perfect

CC BY-NC-ND. 17 June 2016, J. M. Powers.

2.6. POTENTIAL FLOW OF IDEAL POINT VORTICES 179

differential, which integrates over the closed contour to be zero. We continue now by using
the linear momentum principle to replace the particle acceleration with density-scaled forces
to arrive at
I  
d 1 1 T

T T
= f p + dx. (2.157)
dt C

If now the fluid is inviscid ( = 0), the body force is conservative (f = ), and the fluid
is barotropic ((1/)p = ), then we have
I  T
d
= dx, (2.158)
dt C
 I 
= T + dx, (2.159)
IC  
= d + . (2.160)
| C {z }
=0

The integral on the right hand side is zero because the contour is closed; hence, the integral
is path independent. Consequently, we arrive at the common version of Kelvins circula-
tion theorem which holds that for a fluid which is inviscid, barotropic, and subjected to
conservative body forces, the circulation following a material region does not change with
time:
d
= 0. (2.161)
dt
Note that this is very similar to the Helmholtz equation, which, when we make the
additional stipulation of two-dimensionality and incompressibility, gives d/dt = 0. This is
not surprising as the vorticity is closely linked to the circulation via Stokes theorem, which
states I Z Z
T T
= v dx = ( v) ndA = T ndA. (2.162)
C A A

2.6 Potential flow of ideal point vortices

Consider the fluid motion induced by the simultaneous interaction of a family of ideal ir-
rotational point vortices in an incompressible flow field. Since the flow is irrotational and
incompressible, we have the following useful results:

Since v = 0, we can write the velocity vector as the gradient of a scalar potential
:
v = , if irrotational. (2.163)
We call the velocity potential.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

180 CHAPTER 2. VORTEX DYNAMICS

Since T v = 0, we have
T = 2 = 0, (2.164)
or expanding, we have
2 2 2
+ 2 + 2 = 0. (2.165)
x2 y z

We notice that the equation for is linear; hence the method of superposition is valid
here for the velocity potential. That is, we can add an arbitrary number of velocity
potentials together and get a viable flow field.

The irrotational unsteady Bernoulli equation gives us the time and space dependent
pressure field. This equation is not linear, so we do not expect pressures from elemen-
tary solutions to add to form total pressures.

Recalling that the incompressible, three dimensional constant viscosity Helmholtz equa-
tion can be written as
d
= ( T )v + 2 , (2.166)
dt
we see that a flow which is initially irrotational everywhere in an unbounded fluid will always
be irrotational, as ddt = 0. There is no mechanism to change the vorticity from its uniform
initial value of zero. This even holds for a viscous flow. However, in a bounded medium, the
no-slip boundary condition almost always tends to diffuse vorticity into the flow as we shall
see.
Further from Kelvins circulation theorem, we also note that the circulation has no
tendency to change following a particle; that is convects along particle pathlines.

2.6.1 Two interacting ideal vortices

Let us apply this notion to two ideal counterrotating vortices 1 and 2, with respective
strengths, 1 and 2 , as shown in Figure 2.10, Were it isolated, vortex 1 would have no
tendency to move itself, but would induce a velocity at a distance h away from its center of
1
2h
. This induced velocity in fact convects vortex 2, to satisfy Kelvins circulation theorem.
2
Similarly, vortex 2 induces a velocity of vortex 1 of 2h .
The center of rotation G is the point along the 1-2 axis for which the induced velocity is
zero, as is illustrated in Figure 2.11, To calculate it we equate the induced velocities of each
vortex
1 2
= , (2.167)
2hG 2(h hG )
(h hG )1 = hG 2 , (2.168)
h1 = hG (1 + 2 ), (2.169)
1
hG = h . (2.170)
1 + 2

CC BY-NC-ND. 17 June 2016, J. M. Powers.

2.6. POTENTIAL FLOW OF IDEAL POINT VORTICES 181

1
v2 =
2h
1 2

1 h 2

2
v1 =
2h

Figure 2.10: Sketch of the mutual influence of two ideal point vortices on each other.

G
1 2
hG
h

Figure 2.11: Sketch showing the center of rotation G.

v= v=
2h 2h

Figure 2.12: Sketch showing a pair of counterrotating vortices of equal strength

CC BY-NC-ND. 17 June 2016, J. M. Powers.

182 CHAPTER 2. VORTEX DYNAMICS

image main
vortex vortex

Figure 2.13: Sketch showing a vortex and its image to simulate an inviscid wall.

A pair of equal strength counterrotating vortices is illustrated in Figure 2.12. Such

vortices induce the same velocity in each other, so they will propagate as a pair at a fixed
distance from one another.

2.6.2 Image vortex

If we choose to model the fluid as inviscid, then there is no viscous stress, and we can no
longer enforce the no slip condition at a wall. However at a slip wall, we must require that
the velocity vector be parallel to the wall. We can model the motion of an ideal vortex
separated by a distance h from an inviscid slip wall by placing a so-called image vortex on
the other side of the wall. The image vortex will induce a velocity which when superposed
with the original vortex, renders the resultant velocity to be parallel to the wall. A vortex
and its image vortex, which generates a straight streamline at a wall, is sketched in Figure
2.13,

2.6.3 Vortex sheets

We can model the slip line between two inviscid fluids moving at different velocities by what
is known as a vortex sheet. A vortex sheet is sketched in Figure 2.14. Here we have a
distribution of small vortices, each of strength d, on the x axis. Each of these vortices
induces a small velocity dv at an arbitrary point (x, y). The influence of the point vortex at
(x, 0) is sketched in the figure. It generates a small velocity with magnitude

d d
d|v| = = p (2.171)
2h 2 (x x)2 + y 2

CC BY-NC-ND. 17 June 2016, J. M. Powers.

2.6. POTENTIAL FLOW OF IDEAL POINT VORTICES 183

dv

dv ~ ~
(x, y)
du

1 d
u=
2 dx

x
(x,0)
1 d
u=
2 dx

Figure 2.14: Sketch showing schematic of vortex sheet.

Using basic trigonometry, we can deduce that the influence of the single vortex of differential
strength on each velocity component is

dy d
dx
y
du = = dx, (2.172)
2 ((x x)2 + y 2) 2 ((x x)2 + y 2)
d
d(x x) dx
(x x)
dv = 2 2
= dx. (2.173)
2 ((x x) + y ) 2 ((x x)2 + y 2)

Here d
dx
is a measure of the strength of the vortex sheet. Let us account for the effects of all
of the differential vortices by integrating from x = L to x = L and then letting L .
We obtain then the total velocity components u and v at each point to be

d    
L x L + x
u = lim
dx
arctan + arctan ,
(2.174)
L 2 y y
| {z } | {z }
2 2
1 d
2 dx , if y > 0,
= (2.175)
1 d
2 dx
, if y < 0,
d
dx (L x)2 + y 2
v = lim ln = 0. (2.176)
L 4 (L + x)2 + y 2

So the vortex sheet generates no y component of velocity anywhere in the flow field and two
uniform x components of velocity of opposite sign above and below the x axis.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

184 CHAPTER 2. VORTEX DYNAMICS

2.6.4 Potential of an ideal vortex

Let us calculate the velocity potential function associated with a single ideal vortex. Con-
sider an ideal vortex centered at the origin, and represent the velocity field here in cylindrical
coordinates:

vr = 0, v = , vz = 0. (2.177)
2r
Now in cylindrical coordinates the gradient operating on a scalar function gives

= v, (2.178)
1
er + e + ez = 0er + e + 0ez , (2.179)
r r z 2r

= 0, (2.180)
r
1
= , so = + C(r, z), (2.181)
r 2r 2

= 0. (2.182)
z
But since the partials of with respect to r and z are zero, C(r, z) is at most a constant,
which we can set to zero without losing any information regarding the velocity itself

= . (2.183)
2
In Cartesian coordinates, we have

y
= arctan (2.184)
2 x
Lines of constant potential for the ideal vortex centered at the origin are sketched in Figure
2.15.

2.6.5 Interaction of multiple vortices

Here we will consider the interactions of a large number of vortices by using the method of
superposition for the velocity potentials.
If we have two vortices with strengths 1 and 2 centered at arbitrary locations (x1 , y1 )
and (x2 , y2 ) are sketched in Figure 2.16, the potential for each is given by
 
1 y y1
1 = arctan , (2.185)
2 x x1
 
2 y y2
2 = arctan . (2.186)
2 x x2

CC BY-NC-ND. 17 June 2016, J. M. Powers.

2.6. POTENTIAL FLOW OF IDEAL POINT VORTICES 185

=
4

= =0
2

3
=
4

Figure 2.15: Lines of constant potential for ideal irrotational vortex.

y2
1

y1

x1 x2

Figure 2.16: Two vortices at arbitrary locations.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

186 CHAPTER 2. VORTEX DYNAMICS

Since the equation governing the velocity potential, 2 = 0, is linear we can add the two
potentials and still satisfy the overall equation so that
   
1 y y1 2 y y2
= arctan + arctan , (2.187)
2 x x1 2 x x2
is a legitimate solution. Taking the gradient of ,
     
1 y y1 2 y y2
= i
2 (x x1 )2 + (y y1 )2 2 (x x2 )2 + (y y2 )2
    
1 x x1 2 x x2
+ 2 2
+ j, (2.188)
2 (x x1 ) + (y y1 ) 2 (x x2 )2 + (y y2 )2
so that
   
1 y y1 2 y y2
u(x, y) = 2 2
, (2.189)
2 (x x1 ) + (y y1 ) 2 (x x2 )2 + (y y2 )2
   
1 x x1 2 x x2
v(x, y) = + . (2.190)
2 (x x1 )2 + (y y1 )2 2 (x x2 )2 + (y y2 )2
Extending this to a collection of N vortices located at (xi , yi ) at a given time, we have the
following for the velocity field:
XN  
i y yi
u(x, y) = 2 + (y y )2
, (2.191)
i=1
2 (x xi ) i

XN  
i x xi
v(x, y) = . (2.192)
i=1
2 (x xi )2 + (y yi )2

Now to convect (that is, to move) the kth vortex, we move it with the velocity induced
by the other vortices, since vortices convect with the flow. Recalling that the velocity is the
time derivative of the position uk = dxdtk , vk = dydtk , we then get the following 2N non-linear
ordinary differential equations for the 2N unknowns, the x and y positions of each of the N
vortices:
N
X  
dxk i yk yi
= , xk (0) = xok , k = 1, . . . , N,(2.193)
dt 2 (xk xi )2 + (yk yi )2
i=1,i6=k
N
X  
dyk i xk xi
= 2 + (y y )2
, yk (0) = yko , k = 1, . . . , N. (2.194)
dt i=1,i6=k
2 (xk xi ) k i

This set of equations, except for three or fewer point vortices, must be integrated numeri-
cally. These equations form what is commonly termed a Biot-Savart23 law. These ordinary
2
Jean-Baptiste Biot, 1774-1862, Paris-born applied mathematician
3
Felix Savart, 1791-1841, French mathematician who worked on magnetic fields and acoustics.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

2.6. POTENTIAL FLOW OF IDEAL POINT VORTICES 187

differential equations are highly non-linear and give rise to chaotic motion of the point vor-
tices in general. It is a very similar calculation to the motion of point masses in a Newtonian
gravitational field, except that the essential variation goes as 1/r for vortices and 1/r 2 for
Newtonian gravitational fields. Thus the dynamics are different. Nevertheless just as calcu-
lations for large numbers of celestial bodies can give rise to solar systems, clusters of planets,
and galaxies, similar galaxies of vortices can be predicted with the equations for vortex
dynamics.

2.6.6 Pressure field

We have thus far examined essentially only the kinematics of vortices. We have actually
used dynamics in our incorporation of the Helmholtz equation and Kelvins theorem, but
their simple results really only justify the use of a simple kinematics. Dynamics asks what
are the forces which give rise to the motion. Here, we will assume there is no body force
and that the fluid is inviscid, in which case it must be pressure forces which give rise to the
motion. We have the proper conditions for which Bernoullis equation can be used to give
the pressure field. We consider two cases, a single stationary point vortex, and a group of
N moving point vortices.

2.6.6.1 Single stationary vortex

If we take p = p in the far field and fi = g = 0, this steady flow gives us
1 T p 1 T p
v v+ = v v + , (2.195)
2 2
 2
1 p p
+ = 0+ , (2.196)
2 2r
2 1
p(r) = p 2 2 . (2.197)
8 r
Note that the pressure goes to negative infinity at the origin. This is obviously unphysical.
It can be corrected by including viscous effects, which turn out not to substantially alter our
main conclusions.

2.6.6.2 Group of N vortices

For a collection of N vortices, the flow is certainly not steady, and we must in general retain
the time dependent velocity potential in Bernoullis equation yielding
1 p
+ ()T + = f (t). (2.198)
t 2
Now we require that as r that p p . We also know that as r that 0,
hence 0 as well. Hence as r , we have p = f (t). So our final result is
1
p = p ()T . (2.199)
2 t

CC BY-NC-ND. 17 June 2016, J. M. Powers.

188 CHAPTER 2. VORTEX DYNAMICS

wall vortex line

wall streamline

x
v

wall streamline
wall vortex line

Figure 2.17: Wall streamlines and vortex lines at wall y = 0.

So with a knowledge of the velocity field through , we can determine the pressure field
which must have given rise to that velocity field.

2.7 Influence of walls

The Helmholtz equation considers mechanisms that generate vorticity in the interior of a
flow. It does not, however, include one of the most important mechanisms, namely the
introduction of vorticity due to the no-slip boundary condition at a solid wall. In this
section we shall focus on that mechanism.

2.7.1 Streamlines and vortex lines at walls

It seems odd that a streamline can be defined at a wall where the velocity is formally zero, but
in the neighborhood of the wall, the fluid velocity is small but non-zero. We can extrapolate
the position of streamlines near the wall to the wall to define a wall streamline. We shall
also consider a so-called vortex line, a line everywhere parallel to the vorticity vector, at the
wall.
We consider the geometry sketched in Figure 2.17. Here the xz plane is locally attached
to a wall at y = 0, and the y direction is normal to the wall. Wall streamlines and vortex
lines are sketched in the figure.
Because the flow satisfies no-slip, we have at the wall

u(x, y = 0, z) = 0, v(x, y = 0, z) = 0, w(x, y = 0, z) = 0. (2.200)

Because of this, partial derivatives of all velocities with respect to either x or z will also be

CC BY-NC-ND. 17 June 2016, J. M. Powers.

2.7. INFLUENCE OF WALLS 189

zero at y = 0:
     
u  u  v  v  w  w 
= = = = = =0 (2.201)
x y=0 z y=0 x y=0 z y=0 x y=0 z y=0

Near the wall, the velocity is near zero, so the Mach number is very small, and the flow is well
modeled as incompressible. So here, the mass conservation equation implies that T v = 0,
so applying this at the wall, we get
  
u  v  w 
+ + = 0 so (2.202)
x y=0 y y=0 z y=0
| {z } | {z }
=0 =0

v 
= 0. (2.203)
y y=0

Now let us examine the behavior of u, v, and w, as we leave the wall in the y direction.
Consider a Taylor series of each:
 
u  1 2 u 
u = u|y=0 +  y+ 2  y2 + . . . , (2.204)
| {z } y y=0 2 y y=0
=0
 
v  1 2 v 
v = v|y=0 + y+ y2 + . . . , (2.205)
| {z } y y=0 2 y 2 y=0
=0 | {z }
=0
 
w  1 2 w 
w = w|y=0 + y+ y2 + . . . , (2.206)
| {z } y y=0 2 y 2 y=0
=0
(2.207)

So we get

u 
u = y + ..., (2.208)
y y=0

1 2 v 
v = 2  y2 + . . . , (2.209)
2 y y=0

w 
w = y + .... (2.210)
y y=0

Now for streamlines, we must have

dx dy dz
= = . (2.211)
u v w

CC BY-NC-ND. 17 June 2016, J. M. Powers.

190 CHAPTER 2. VORTEX DYNAMICS

For the streamline near the wall, consider just dx/u = dz/w, and also tag the streamline as
dzs , so that the slope of the wall streamline, which is the tangent of the angle between the
wall streamline and the x axis is

 w 
dzs  w y 
tan =  = lim =  y=0 (2.212)
dx y=0 y0 u u 
y 
y=0

Now consider the vorticity vector evaluated at the wall:

  
w  v  w 
x |y=0 = = , (2.213)
y y=0 z y=0 y y=0
| {z }
=0
 
u  w 
y |y=0 = = 0, (2.214)
z y=0 x y=0
| {z } | {z }
=0 =0
  

v  u  u 
z |y=0 = = . (2.215)
x y=0 y y=0 y y=0
| {z }
=0
So we see that on the wall at y = 0, the vorticity vector has no component in the y direction.
Hence, it must be parallel to the wall itself. Further, we can then define the slope of the
vortex line, dz
dx
v
, at the wall in the same fashion as we define a streamline:

 u 
dzv  z y  1
y=0
 = =  = dzs  (2.216)
dx y=0 x w 
y  dx y=0
y=0

Since the slope of the vortex line is the negative reciprocal of the slope of the streamline,
we have that at a no-slip wall, streamlines are orthogonal to vortex lines. We also note that
streamlines are orthogonal to vortex lines for flow with variation in the x and y directions
only. For general three-dimensional flows away from walls, we do not expect the two lines
to be orthogonal.

2.7.2 Generation of vorticity at walls

Now further restrict the coordinate system of the previous subsection so that the x axis is
aligned with the wall streamline and the z axis is aligned with the wall vortex line. As before
the y axis is normal to the wall. The coordinate system aligned with the wall streamlines
and vortex lines is sketched in Figure 2.18, In the figure we take the direction n to be normal
to the wall. Now for this coordinate system, we have

 w 
dzs  w  y 
=  =  y=0 = 0. (2.217)
dx y=0 u y=0 u 

y
y=0

CC BY-NC-ND. 17 June 2016, J. M. Powers.

2.7. INFLUENCE OF WALLS 191

v
x

Figure 2.18: Coordinate system aligned with wall streamlines and vortex lines.

Hence, we must have 

w 
= 0. (2.218)
y y=0
Now consider the viscous traction vector associated with the wall:
  

u  v 
y y=0 + x y=0 u 
   
yx   y y=0

tj = ni ij = ny yj = yy = y 
v v
+ y  = (2.219)
y=0 y=0
0
yz   
 0
v  w 
z y=0 + y 
y=0

So the viscous force is parallel to the surface, hence it is a tangential or shear force; moreover,
it points in the same direction as the streamline. Now if we examine the vorticity vector at
the surface we find first by our definition of the coordinate systems that x = y = 0 at
y = 0 and that  
v  u 
z |y=0 = . (2.220)
x y=0 y y=0
| {z }
=0

For this case, we can say that the viscous force is t1 = z . In fact in general, we can say

tviscous = n , at a wall. (2.221)

Since the viscous force is orthogonal to both the surface normal and the vorticity vector, it
must always at the wall be aligned with the flow direction.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

192 CHAPTER 2. VORTEX DYNAMICS

CC BY-NC-ND. 17 June 2016, J. M. Powers.

Chapter 3

One-dimensional compressible flow

see Yih, Chapter 6

see Liepmann and Roshko, Chapter 2
see Shapiro, Chapters 4-8

This chapter will focus on one-dimensional flow of a compressible fluid. The following topics
will be covered:

development of generalized one-dimensional flow equations,

isentropic flow with area change,

flow with normal shock waves, and

the method of characteristics for isentropic rarefactions.

We will assume for this chapter:

v 0, w 0, /y 0, /z 0; one-dimensional flow.

Friction and heat transfer will not be modeled rigorously. Instead, they will be modeled
in a fashion which captures the relevant physics and retains analytic tractability. Further,
we will ignore the influences of an external body force, fi = 0.

3.1 Generalized one-dimensional equations

Here we will re-derive, in a rather conventional way, the one-dimensional equations of flow
with area change. Although for the geometry we use, it will appear that we should be using
at least two-dimensional equations, our results will be correct when we interpret them as an
average value at a given x location. Our results will be valid as long as the area changes
slowly relative to how fast the flow can adjust to area changes.

193
194 CHAPTER 3. ONE-DIMENSIONAL COMPRESSIBLE FLOW

We could start directly with our equations from an earlier chapter as well. However,
the ad hoc nature of friction and heat transfer commonly employed makes a re-derivation
useful. The flow we wish to consider, flow with area change, heat transfer, and wall friction,
is illustrated by the following sketch of a control volume, Figure 3.1.

x = x - x1
2
x2
nw
x1 n2

u1
1 n1 2
u2
A1 A2
P1 P2
e1 e2
w

q
w

Perimeter length = L
Figure 3.1: Control volume sketch for one-dimensional compressible flow with area change,
heat transfer, and wall friction.

For this flow, we will adopt the following conventions

surface 1 and 2 are open and allow fluxes of mass, momentum, and energy,

surface w is a closed wall; no mass flux through the wall

external heat flux qw (Energy/Area/Time: W/m2 ) through the wall allowed-qw known
fixed parameter,

diffusive, longitudinal heat transfer ignored, qx = 0,

wall shear w (Force/Area: N/m2 ) allowedw known, fixed parameter,

diffusive viscous stress not allowed xx = 0, and

cross-sectional area a known fixed function: A(x).

CC BY-NC-ND. 17 June 2016, J. M. Powers.

3.1. GENERALIZED ONE-DIMENSIONAL EQUATIONS 195

3.1.1 Mass
Take the over-bar notation to indicate a volume averaged quantity.

The amount of mass in a control volume after a time increment t is equal to the original
amount of mass plus that which came in minus that which left:
 
Axt+t = Axt + 1 A1 (u1 t) 2 A2 (u2 t) . (3.1)

Rearrange and divide by xt:

 
At+t At 2 A2 u2 1 A1 u1
+ = 0. (3.2)
t x
Taking the limit as t 0, x 0, we get

(A) + (Au) = 0. (3.3)
t x
If the flow is steady, then

d
(Au) = 0, (3.4)
dx
d dA du
Au + u + A = 0, (3.5)
dx dx dx
1 d 1 dA 1 du
+ + = 0. (3.6)
dx A dx u dx

Now integrate from x1 to x2 to get

Z x2 Z x2
d
(Au)dx = 0dx, (3.7)
x1 dx x1
Z 2
d (Au) = 0, (3.8)
1
2 u2 A2 1 u1 A1 = 0, (3.9)
2 u2 A2 = 1 u1 A1 m = C1 . (3.10)

3.1.2 Linear momentum

Newtons Second Law says the time rate of change of linear momentum of a body equals the
sum of the forces acting on the body. In the x direction this is roughly as follows:

d
(mu) = Fx . (3.11)
dt

CC BY-NC-ND. 17 June 2016, J. M. Powers.

196 CHAPTER 3. ONE-DIMENSIONAL COMPRESSIBLE FLOW

In discrete form this would be

mu|t+t mu|t
= Fx , (3.12)
t
mu|t+t = mu|t + Fx t. (3.13)

For a control volume containing fluid, we must also account for the momentum which
enters and leaves the control volume. The amount of momentum in a control volume after
a time increment t is equal to the original amount of momentum plus that which came in
minus that which left plus that introduced by the forces acting on the control volume.
Note that

pressure force at surface 1 pushes fluid,

pressure force at surface 2 restrains fluid,

force due to the reaction of the wall to the pressure force pushes fluid if area change
positive, and

force due to the reaction of the wall to the shear force restrains fluid.

We write the linear momentum principle as



Ax ut+t = Ax ut
+ (1 A1 (u1 t)) u1
(2 A2 (u2 t)) u2
+ (p1 A1 ) t (p2 A2 ) t
+ (p (A2 A1 )) t


w Lx t. (3.14)

Rearrange and divide by xt to get

 
Aut+t Aut 2 A2 u22 1 A1 u21
+
t x
p2 A2 p1 A1 A2 A1
= + p w L. (3.15)
x x
In the limit x 0, t 0 we get

A
(Au) + Au2 = (pA) + p w L. (3.16)
t x x x
In steady state we find
d
d dA
Au2 = (pA) + p w L, (3.17)
dx dx dx

CC BY-NC-ND. 17 June 2016, J. M. Powers.

3.1. GENERALIZED ONE-DIMENSIONAL EQUATIONS 197

du d dA dp dA
Au + u (Au) = p A +p w L, (3.18)
dx dx dx dx dx
du dp L
u = w , (3.19)
dx dx A
L
udu + dp = w dx, (3.20)
A
1 L
du + dp = w dx, (3.21)
u m
 2
u L
d + dp = w dx. (3.22)
2 A
Wall shear lowers the combination of pressure and dynamic head.
If there is no wall shear, then we have
 2
u
dp = d . (3.23)
2
An increase in velocity magnitude decreases the pressure.
If there is no area change, dA = 0, and no friction w 0 we have
du dp
u + = 0, (3.24)
dx dx
d
add product of u and mass equation u (u) = 0 u, (3.25)
dx
d

u2 + p = 0, (3.26)
dx
u2 + p = o u2o + po = C2 . (3.27)

3.1.3 Energy
The first law of thermodynamics states that the change of total energy of a body equals the
heat transferred to the body minus the work done by the body:

E2 E1 = Q W, (3.28)
E2 = E1 + Q W. (3.29)
So for our control volume this becomes the following when we also account for the energy
flux in and out of the control volume in addition to the work and heat transfer:
   

u2 
u2 
Ax e + = Ax e +
2 t+t 2 t
   
u21 u22
+1 A1 (u1 t) e1 + 2 A2 (u2 t) e2 +
2 2


+qw Lx t + (p1 A1 ) (u1 t) (p2 A2 ) (u2 t) . (3.30)
Note:

CC BY-NC-ND. 17 June 2016, J. M. Powers.

198 CHAPTER 3. ONE-DIMENSIONAL COMPRESSIBLE FLOW

the mean pressure times area difference does no work because it is acting on a stationary
boundary, and

the work done by the wall shear force is not included.1

Rearrange and divide by tx:
   
u2  u2 
A e + 2  A e + 2 
t+t t

   t 2 
u22 u
2 A2 u2 e2 + 2
+ p22 1 A1 u1 e1 + 21 + p1
1
+ = qw L. (3.31)
x
In differential form as x 0, t 0
     
u2 u2 p
A e + + Au e + + = qw L. (3.32)
t 2 x 2
In steady state:
  
d u2 p
Au e + + = qw L, (3.33)
dx 2
 2
  2

d u p u p d
Au e+ + + e+ + (Au) = qw L, (3.34)
dx 2 2 dx
 
d u2 p qw L
u e+ + = , (3.35)
dx 2 A
 
de du 1 dp p d qw L
u +u + = , (3.36)
dx dx dx 2 dx A
subtract the product of momentum and velocity (3.37)
du dp w Lu
u2 +u = , (3.38)
dx dx A
de pu d qw L w Lu
u = + , (3.39)
dx dx A A
de p d (qw + w u) L
2 = . (3.40)
dx dx m
1
In neglecting work done by the wall shear force, I have taken an approach which is nearly universal, but
fundamentally difficult to defend. At this stage of the development of these notes, I am not ready to enter
into a grand battle with all established authors and probably confuse the student; consequently, results for
flow with friction will be consistent with those of other sources. The argument typically used to justify this
is that the real fluid satisfies no-slip at the boundary; thus, the wall shear actually does no work. However,
one can easily argue that within the context of the one-dimensional model which has been posed that the
shear force behaves as an external force which reduces the fluids mechanical energy. Moreover, it is possible
to show that neglect of this term results in the loss of frame invariance,


a serious defect indeed. To model
the work of the wall shear, one would include the term w Lx (ut) in the energy equation.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

3.1. GENERALIZED ONE-DIMENSIONAL EQUATIONS 199

Since e = e(p, )
 
e  e 
de = d + dp, (3.41)
p p 
 
de e  d e  dp
= + . (3.42)
dx  dx
p p  dx

so the steady energy equation becomes

 
e  d e  dp p d (qw + w u) L
 +  2 = , (3.43)
p dx p dx dx m

e  p

dp p d 2
(qw + w u) L
+  =  . (3.44)
dx e  dx e 
p  m p 

Now let us consider the term in braces in the previous equation. We can put that term
in a more common form by considering the Gibbs equation, Eq. (1.531):
p
T ds = de 2 d, (3.45)

along with a general caloric equation of state e = e(p, ), from which we get
 
e  e 
de = dp + d. (3.46)
p  p
Substituting into the Gibbs equation, we get
 
e  e  p
T ds =  dp +  d 2 d. (3.47)
p p
Taking s to be constant and dividing by d, we get
  
e  p  e  p
0=   +  2. (3.48)
p s p
Rearranging, we get

 e  p
p   2
2 p

c = , (3.49)
s e 

p

so
dp d (qw + w u) L
c2 =  , (3.50)
dx dx e 
m p 

dp d (qw + w u) L
c2 =  . (3.51)
dx dx e 
uA p 

CC BY-NC-ND. 17 June 2016, J. M. Powers.

200 CHAPTER 3. ONE-DIMENSIONAL COMPRESSIBLE FLOW

In the above c is the isentropic sound speed, a thermodynamic property of the material. We
shall see in a later section why it is appropriate to interpret this property as the propagation
speed of small disturbances. At this point, it should simply be thought of as a state property.

Consider now the special case of flow with no heat transfer qw 0. We still allow area
change and wall friction allowed (see earlier footnote):
 
d u2 p
u e+ + = 0, (3.52)
dx 2
u2 p u 2 po
e+ + = eo + o + = C3 , (3.53)
2 2 o
u2 u2
h+ = ho + o = C3 . (3.54)
2 2

3.1.4 Summary of equations

We can summarize the one-dimensional compressible flow equations in various forms here.
In the equations below, we assume A(x), w , qw , and L are all known.

3.1.4.1 Unsteady conservative form

(A) + (Au) = 0, (3.55)
t x

A
(Au) + Au2 + pA = p w L, (3.56)
   t  x
  x
u2 u2 p
A e + + Au e + + = qw L, (3.57)
t 2 x 2
e = e(, p), (3.58)
p = p(, T ). (3.59)

3.1.4.2 Unsteady non-conservative form

d
= (Au), (3.60)
dt A x
du p w L
= + , (3.61)
dt x A
de u qw L w Lu
= p + , (3.62)
dt x A
e = e(, p), (3.63)
p = p(, T ). (3.64)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

3.1. GENERALIZED ONE-DIMENSIONAL EQUATIONS 201

3.1.4.3 Steady conservative form

d
(Au) = 0, (3.65)
dx
d
dA
Au2 + pA = p w L, (3.66)
 dx
  dx
d u2 p
Au e + + = qw L, (3.67)
dx 2
e = e(, p), (3.68)
p = p(, T ). (3.69)

3.1.4.4 Steady non-conservative form

d d
u = (Au), (3.70)
dx A dx
du dp w L
u = + , (3.71)
dx dx A
de du qw L w Lu
u = p + , (3.72)
dx dx A
e = e(, p), (3.73)
p = p(, T ). (3.74)
In whatever form we consider, we have five equations in five unknown dependent variables:
, u, p, e, and T . We can always use the thermal and caloric state equations to eliminate e
and T to give rise to three equations in three unknowns.

Example 3.1
Flow of air with heat addition

Given: Air initially at p1 = 100 kP a, T1 = 300 K, u1 = 10 m s flows in a duct of length 100 m.

The duct has a constant circular cross sectional area of A = 0.02 m2 and is isobarically heated with
a constant heat flux qw along the entire surface of the duct. At the end of the duct the flow has
p2 = 100 kP a, T2 = 500 K

Find: the mass flow rate m, the wall heat flux qw and the entropy change s2 s1 ; check for
satisfaction of the second law.

Assume: Calorically perfect ideal gas, R = 0.287 kJ/(kg K), cp = 1.0035 kJ/(kg K)

Analysis:
Geometry:
A = r2 , (3.75)
r
A
r = (3.76)

p
L = 2r = 2 A = 2 (0.02 m2 ) = 0.501 m. (3.77)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

202 CHAPTER 3. ONE-DIMENSIONAL COMPRESSIBLE FLOW

Now get the mass flux.

p1 = 1 RT1 , (3.78)
p1 100 kP a
1 = =  , (3.79)
RT1 0.287 kgkJK (300 K)
kg
= 1.161 (3.80)
m3
So
 
kg  m 
kg
m = 1 u1 A1 = 1.161 3 10 0.02 m2 = 0.2322 . (3.81)
m s s
Now get the flow variables at state 2:
p2 100 kP a
2 = ,=   , (3.82)
RT2 0.287 kgkJK (500 K)
kg
= 0.6969 , (3.83)
m3
2 u2 A2 = 1 u1 A1 , (3.84)
1 u1 A1 1 u 1
u2 = = , (3.85)
2 A2
  2

kg
1.161 m 3 10 m s m
= kg
= 16.67 . (3.86)
0.6969 m3
s
Now consider the energy equation:
 
d u2 p qw L
u e+ + = , (3.87)
dx 2 A
 
d u2 qw L
h+ = , (3.88)
dx 2 m
Z L  2
 Z L
d u qw L
h+ dx = dx, (3.89)
0 dx 2 0 m
u2 u2 qw LL
h2 + 2 h1 1 = , (3.90)
2 2 m
u22 u21 qw LL
cp (T2 T1 ) + = . (3.91)
2 2 m
Solving for qw , we get
  
m u22 u21
qw = cp (T2 T1 ) + , (3.92)
LL 2 2
!  

!
m 2 m 2
0.2322 kgs J 16.67 s 10 s
qw = 1003.5 (500 K 300 K) + , (3.93)
(100 m) (0.501 m) kg K 2 2
 
kg J m2
qw = 0.004635 2 200700 88.9 2 , (3.94)
m s kg s
 
kg J J
qw = 0.004635 2 200700 88.9 , (3.95)
m s kg kg
W
qw = 930 2 . (3.96)
m

CC BY-NC-ND. 17 June 2016, J. M. Powers.

3.1. GENERALIZED ONE-DIMENSIONAL EQUATIONS 203

The heat flux is positive, which indicates a transfer of thermal energy into the air.

Now find the entropy change.

   
T2 p2
s2 s1 = cp ln R ln , (3.97)
T1 p1
       
J 500 K J 100 kP a
s2 s1 = 1003.5 ln 287 ln , (3.98)
kg K 300 K kg K 100 kP a
J
s2 s1 = 512.6 0 = 512.6 . (3.99)
kg K
Is the second law satisfied? Assume the heat transfer takes place from a reservoir held at 500 K. The
reservoir would have to be at least at 500 K in order to bring the fluid to its final state of 500 K. It
could be greater than 500 K and still satisfy the second law.
Q12
S2 S1 , (3.100)
T
Q12
S2 S1 , (3.101)
T
Q12
m (s2 s1 ) , (3.102)
T
qw Atot
m (s2 s1 ) , (3.103)
T
qw LL
m (s2 s1 ) , (3.104)
T
qw LL
s2 s1 , (3.105)
mT

J 930 s Jm2 (100 m) (0.501 m)
512.6   , (3.106)
kg K 0.2322 kg (500 K)
s
J J
512.6 401.3 . (3.107)
kg K kg K

3.1.5 Influence coefficients

Now, let us uncouple the steady one-dimensional equations. First let us summarize again,
in a slightly different manner than before:
d du u dA
u + = , (3.108)
dx dx A dx
du dp w L
u + = , (3.109)
dx dx A
dp d (qw + w u) L
c2 =  . (3.110)
dx dx e 
uA p 

CC BY-NC-ND. 17 June 2016, J. M. Powers.

204 CHAPTER 3. ONE-DIMENSIONAL COMPRESSIBLE FLOW

In matrix form this is

d u dA
u 0 dx
A dx
0 du w L
u 1 dx = A . (3.111)
dp (qw +w u)L
c2 0 1 dx uA |e
p

Use Cramers Rule to solve for the derivatives. First calculate the determinant of the coef-
ficient matrix:



u ((u)(1) (1)(0)) (0)(1) (c2 )(1) = u2 c2 . (3.112)

Implementing Cramers Rule:

 


(qw +w u)L
u u dA
wAL +
d A dx

e
uA p |
= , (3.113)
dx (u2 c2 )
 
2 u dA

w L

(qw +w u)L
c A dx + u A u uA e
du p |
= 2 2
, (3.114)
dx (u c )
 
2 u dA

2 w L

2 (qw +w u)L
uc A dx c A + u
dp e
uA p |
= . (3.115)
dx (u2 c2 )

Simplify to find
(qw +w u)L
u2 dA + w L +
d 1 dx e
u p |
= , (3.116)
dx A (u2 c2 )
(qw +w u)L
c2 u dA uw L
du 1 dx e
p |
= . (3.117)
dx A (u2 c2 )
(qw +w u)Lu
c2 u2 dA + c2 w L +
dp 1 dx e
p |
= . (3.118)
dx A (u2 c2 )

Note, we have

a system of coupled non-linear ordinary differential equations,

in standard form for dynamic system analysis: du/dx = f(u),

valid for general equations of state, and

singular when velocity sonic u = c.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

3.2. FLOW WITH AREA CHANGE 205

3.2 Flow with area change

This section will consider flow with area change with an emphasis on isentropic flow. Some
problems will involve non-isentropic flow but a detailed discussion of such flows will be
delayed.

3.2.1 Isentropic Mach number relations

Take the special case of

w = 0,

qw = 0,

calorically perfect ideal gas (CPIG).

Then
d
(uA) = 0, (3.119)
dx
d

u2 + p = 0, (3.120)
 dx 2 
d u p
e+ + = 0. (3.121)
dx 2

Integrate the energy equation using Eq. (1.521) h = e + p/ to get

u2 u2
h+ = ho + o . (3.122)
2 2
If we define the o condition to be a condition of rest, then uo 0. This is a stagnation
condition. So
u2
h+ = ho , (3.123)
2
u2
(h ho ) + = 0. (3.124)
2
Since we have a CPIG,

u2
cp (T To ) + = 0, (3.125)
2
u2
T To + = 0, (3.126)
2cp
To u2
1 + = 0. (3.127)
T 2cp T

CC BY-NC-ND. 17 June 2016, J. M. Powers.

206 CHAPTER 3. ONE-DIMENSIONAL COMPRESSIBLE FLOW

Now note that

cp cv cp cp cv R
cp = cp = cp = , (3.128)
cp cv cv c v 1 1
so
To 1 u2
1 + = 0, (3.129)
T 2 RT
To 1 u2
= 1+ . (3.130)
T 2 RT
Recall the sound speed and Mach number for a CPIG:

c2 = RT if p = RT, e = cv T + eo , (3.131)
 u 2
M2 . (3.132)
c
Thus,
To 1 2
= 1+ M , (3.133)
T 2
 1
T 1 2
= 1+ M . (3.134)
To 2
Now if the flow is isentropic, we have
 1   1
T p
= = . (3.135)
To o po
Thus,
  1
1
1 2
= 1+ M , (3.136)
o 2
  1

p 1 2
= 1+ M . (3.137)
po 2
For air = 7/5, so
 1
T 1 2
= 1+ M , (3.138)
To 5
  5
1 2 2
= 1+ M , (3.139)
o 5
  7
p 1 2 2
= 1+ M . (3.140)
po 5

CC BY-NC-ND. 17 June 2016, J. M. Powers.

3.2. FLOW WITH AREA CHANGE 207

Figures 3.2, 3.3 3.4 show the variation of T , and p with M 2 for isentropic flow.

Other thermodynamic properties can be determined from these, e.g. the sound speed:

s r  1/2
c RT T 1 2
= = = 1+ M . (3.141)
co RTo To 2

T(K)
300
calorically perfect
250
ideal gas
200 R = 0.287 kJ/(kg K)
= 7/5
150
stagnation temperature = 300 K
100

50

M2
0 2 4 6 8 10

Figure 3.2: Static temperature versus Mach number squared.

p(bar)
1

0.8 calorically perfect

ideal gas
0.6 R = 0.287 kJ/(kg K)
= 7/5
0.4
stagnation pressure = 1 bar
0.2

M2
0 2 4 6 8 10

Figure 3.3: Static pressure versus Mach number squared.

(kg/m3)

1.2 calorically perfect

1 ideal gas
0.8
R = 0.287 kJ/(kg K)
= 7/5
0.6
stagnation density = 1.16 kg/m3
0.4

0.2

M2
0 2 4 6 8 10

Figure 3.4: Static density versus Mach number squared.

Example 3.2
Airplane problem

CC BY-NC-ND. 17 June 2016, J. M. Powers.

208 CHAPTER 3. ONE-DIMENSIONAL COMPRESSIBLE FLOW

Given: An airplane is flying into still air at u = 200 m/s. The ambient air is at 288 K and
101.3 kP a.
Find: Temperature, pressure, and density at nose of airplane
Assume: Steady isentropic flow of C.P.I.G.
Analysis: In the steady wave frame, the ambient conditions are static while the nose conditions are
stagnation.
u u 200 m/s
M= = = r   = 0.588 (3.142)
c RT 7 J
5 287 kgK 288 K

so
   
1 1
To = T 1 + M 2 = (288 K) 1 + 0.5882 = 307.9 K (3.143)
5 5
 5
1 2 2 101.3 kP a kg
o = 1 + M =  
5 = 1.45 (3.144)
5 0.287 kgkJK (288 K) 1 + 51 0.5882 2 m3
 7   27
1 2 2 1 2
po = p 1 + M = (101.3 kP a) 1 + 0.588 = 128 kP a (3.145)
5 5
Note the temperature, pressure, and density all rise in the isentropic process. In this wave frame, the
kinetic energy of the flow is being converted isentropically to thermal energy.

3.2.2 Sonic properties

Let * denote a property at the sonic state M 2 1

 1
T 1 2 2
= 1+ 1 = , (3.146)
To 2 +1
  1   1
1
1 2 1 2
= 1+ 1 = , (3.147)
o 2 +1
    1

p 1 2 1 2
= 1+ 1 = , (3.148)
po 2 +1
 1/2 r
c 1 2 2
= 1+ 1 = , (3.149)
co 2 +1
r
p 2
u = c = RT = RTo . (3.150)
+1
If the fluid is air, we have = 7/5 and
T
= 0.8333, (3.151)
To

CC BY-NC-ND. 17 June 2016, J. M. Powers.

3.2. FLOW WITH AREA CHANGE 209

= 0.6339, (3.152)
o
p
= 0.5283, (3.153)
po
c
= 0.9123. (3.154)
co

3.2.3 Effect of area change

To understand the effect of area change, the influence of the mass equation must be con-
sidered. So far we have really only looked at energy. In the isentropic limit the mass,
momentum, and energy equation for a C.P.I.G. reduce to
d du dA
+ + = 0, (3.155)
u A
udu + dp = 0, (3.156)
dp d
= . (3.157)
p
Substitute energy, then mass into momentum:
p
udu + d = 0, (3.158)

p 
udu + du dA = 0, (3.159)
u A
 
p 1 1
du + 2 du dA = 0, (3.160)
u uA
 
p/ p dA
du 1 2 = , (3.161)
u uA
 
du p/ p/ dA
1 2 = , (3.162)
u u u2 A
 
du 1 1 dA
1 2 = , (3.163)
u M M2 A
du
dA
M2 1 = , (3.164)
u A
du 1 dA
= 2
. (3.165)
u M 1 A
Figure 3.5 gives show the performance of a fluid in a variable area duct. We note
there is a singularity when M 2 = 1,

if M 2 = 1, we need dA = 0,

area minimum necessary to transition from subsonic to supersonic flow,

CC BY-NC-ND. 17 June 2016, J. M. Powers.

210 CHAPTER 3. ONE-DIMENSIONAL COMPRESSIBLE FLOW

Consider u > 0

Subsonic Subsonic
Diffuser Nozzle

2
dA > 0, M < 1 so
2
du < 0, flow slows down dA < 0, M <1 so
dp > 0 du > 0, flow speeds up
dp < 0

Supersonic Supersonic
Nozzle Diffuser

2 2
dA > 0, M >1 so dA < 0, M > 1 so
du > 0, flow speeds up du < 0, flow slows down
dp < 0 dp > 0

Figure 3.5: Behavior of fluid in sub- and supersonic nozzles and diffusers.

it can be shown an area maximum is not relevant.

Consider A at a sonic state. From the mass equation:
uA = u A , (3.166)
uA = c A , (3.167)

A 1 p 1 RT RT
= c = RT = , (3.168)
A u u RT u
r r
A T 1 o T To 1
= = . (3.169)
A T M o To T M
Substitute from earlier-developed relations and get
   21 1
+1
A 1 2 1 2
= 1+ M . (3.170)
A M +1 2
Figure 3.6 shows the performance of a fluid in a variable area duct.
Note that
A/A has a minimum value of 1 at M = 1,
For each A/A > 1, there exist two values of M, and
A/A as M 0 or M .

CC BY-NC-ND. 17 June 2016, J. M. Powers.

3.2. FLOW WITH AREA CHANGE 211

A/A*
6

4
calorically perfect
ideal gas
3 R = 0.287 kJ/(kg K)
2 = 7/5

M
0 0.5 1 1.5 2 2.5 3

Figure 3.6: Area versus Mach number for a calorically perfect ideal gas.

3.2.4 Choking
Consider mass flow rate variation with pressure difference. We have then

small pressure difference gives small velocity and small mass flow,

as pressure difference grows, velocity and mass flow rate grow,

velocity is limited to sonic at a particular duct location,

this provides fundamental restriction on mass flow rate,

it can be proven rigorously that sonic condition gives maximum mass flow rate.

mmax = u A , (3.171)
 1 r
 1 
2 2
if ideal gas: = = o RTo A , (3.172)
+1 +1
  11  1/2
2 2 p
= o RTo A , (3.173)
+1 +1
  21 1
+1
2 p
= o RTo A . (3.174)
+1

A flow which has a maximum mass flow rate is known as choked flow. Flows will choke
at area minima in a duct.

Example 3.3
2
Isentropic area change problem with choking
2
adopted from White, Fluid Mechanics McGraw-Hill: New York, 1986, p. 529, Ex. 9.5

CC BY-NC-ND. 17 June 2016, J. M. Powers.

212 CHAPTER 3. ONE-DIMENSIONAL COMPRESSIBLE FLOW

Given: Air with stagnation conditions po = 200 kP a To = 500 K flows through a throat to an exit
Mach number of 2.5. The desired mass flow is 3.0 kg/s,

Find: a) throat area, b) exit pressure, c) exit temperature, d) exit velocity, and e) exit area.

Assume: C.P.I.G., isentropic flow, = 7/5

Analysis:
po 200 kP a
o = =   = 1.394 kg/m3 . (3.175)
RTo kJ
0.287 kg K (500 K)

Since it necessarily flows through a sonic throat:

  21 1
+1
2 p
mmax = o RTo A , (3.176)
+1
mmax
A =   21 1
+1 , (3.177)
2
o +1 RTo
3 kg
A =   r s
  = 0.008297 m2 . (3.178)
kg J
1.394 m3 (0.5787) 1.4 287 kg K (500 K)

Since we know Me use isentropic relations to find other exit conditions.

  1
 3.5
1 2 1 2
pe = po 1 + Me = (200 kP a) 1 + 2.5 = 11.71 kP a, (3.179)
2 5
 1  1
1 2 1
Te = To 1 + Me = (500 K) 1 + 2.52 = 222.2 K. (3.180)
2 5

Note
pe 11.71 kP a kg
e = =   = 0.1834 3 . (3.181)
RTe kJ
0.287 kg K (222.2 K) m

Now the exit velocity is simply

s  
p J m
ue = Me ce = Me RTe = 2.5 1.4 287 (222.2 K) = 747.0 . (3.182)
kg K s

Now determine the exit area.

   12 1
+1
A 2 1 2
A = 1+ Me , (3.183)
Me +1 2
  3
0.008297 m2 5 1
= 1 + 2.52 = 0.0219 m2 . (3.184)
2.5 6 5

CC BY-NC-ND. 17 June 2016, J. M. Powers.

3.3. NORMAL SHOCK WAVES 213

3.3 Normal shock waves

This section will develop relations for normal shock waves in fluids with general equations of
state. It will be specialized to calorically perfect ideal gases to illustrate the general features
of the waves.
Assume for this section we have

one-dimensional flow,

steady flow,

no area change,

viscous effects and wall friction do not have time to influence flow, and

heat conduction and wall heat transfer do not have time to influence flow.

We will consider the problem in the context of the piston problem as sketched in Figure 3.7.

vp= v
2 D

v = v2 u = u2 u = -D
v=0
p2 p2 p1
p1
2 2 1
1

Laboratory Frame Steady Wave Frame

x* x

u=v-D v=u+D
x = x* - D t, x* = x + D t

Figure 3.7: Normal shock sketch.

The physical problem is as follows:

Drive a piston with known velocity vp into a fluid at rest (v1 = 0) with known proper-
ties, p1 , 1 in the x laboratory frame,

Determine the disturbance speed D,

Determine the disturbance properties v2 , p2 , 2 ,

in this frame of reference we have an unsteady problem.

Transformed Problem:

CC BY-NC-ND. 17 June 2016, J. M. Powers.

214 CHAPTER 3. ONE-DIMENSIONAL COMPRESSIBLE FLOW

use a Galilean transformation x = x Dt, u = v D to transform to the frame in

which the wave is at rest, therefore rending the problem steady in this frame,

solve as though D is known to get downstream 2 conditions: u2 (D), p2(D), ...,

invert to solve for D as function of u2 , the transformed piston velocity: D(u2),

back transform to get all variables as function of v2 , the laboratory piston velocity:
D(v2 ), p2 (v2 ), 2 (v2 ), ....

3.3.1 Rankine-Hugoniot equations

Under these assumptions the conservation principles in conservative form and equation of
state are in the steady frame as follows:

d
(u) = 0, (3.185)
dx
d

u2 + p = 0, (3.186)
 dx
 
d u2
u h + = 0, (3.187)
dx 2
h = h(p, ). (3.188)

Upstream conditions are = 1 , p = p1 , u = D. With knowledge of the equation of state,

we get h = h1 . In what is a natural, but in fact naive, step we can integrate the equations
from upstream to state 2 to give the correct Rankine-Hugoniot jump equations.34

2 u2 = 1 D, (3.189)
2 u22 + p2 = 1 D 2 + p1 , (3.190)
u2 D2
h2 + 2 = h1 + , (3.191)
2 2
h2 = h(p2 , 2 ). (3.192)

This analysis is straightforward and yields the correct result. In actuality, however, the
analysis should be more nuanced. We are going to solve these algebraic equations to arrive at
discontinuous shock jumps. Thus, we should be concerned about the validity of of differential
equations in the vicinity of a discontinuity.
As described by LeVeque,5 the proper way to arrive at the shock jump equations is to use
a more primitive form of the conservation laws, expressed in terms of integrals of conserved
3
William John Macquorn Rankine, 1820-1872, Scottish engineer and mechanician, pioneer of thermody-
namics and steam engine theory, taught at University of Glasgow, studied fatigue in railway engine axles.
4
Pierre Henri Hugoniot, 1851-1887, French engineer.
5
LeVeque, R. J., 1992, Numerical Methods for Conservation Laws, Birkhauser, Basel.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

3.3. NORMAL SHOCK WAVES 215

quantities balanced by fluxes of those quantities. If q is a set of conserved variables, and

f(q) is the flux of q (e.g. for mass conservation, is a conserved variable and u is the flux),
then the primitive form of the conservation law can be written as
Z
d x2
q(x, t)dx = f(q(x1 , t)) f(q(x2 , t)). (3.193)
dt x1
Here we have considered flow into and out of a one-dimensional box for x [x1 , x2 ]. For our
Euler equations we have

u
2
q= u
, f(q) =  u + p . (3.194)
1 2 1 2 p
e + 2u u e + 2 u +

If we assume there is a discontinuity in the region x [x1 , x2 ] propagating at speed D, we

can break up the integral into the form
Z Z
d x1 +Dt d x2
q(x, t)dx + q(x, t)dx = f(q(x1 , t)) f(q(x2 , t)). (3.195)
dt x1 dt x1 +Dt+
Here x1 + Dt lies just before the discontinuity and x1 + Dt+ lies just past the discontinuity.
Using Leibnizs rule, we get
Z x1 +Dt Z x2
q + q
q(x1 + Dt , t)D + 0 + dx + 0 q(x1 + Dt , t)D + dx (3.196)
x1 t x1 +Dt+ t
= f(q(x1 , t)) f(q(x2 , t)).
Now if we assume that on either side of the discontinuity the volume of integration is suffi-
ciently small so that the time and space variation of q is negligibly small, we get
q(x1 )D q(x2 )D = f(q(x1 )) f(q(x2 )), (3.197)
D (q(x1 ) q(x2 )) = f(q(x1 )) f(q(x2 )). (3.198)
Defining next the notation for a jump as
Jq(x)K q(x2 ) q(x1 ), (3.199)
the jump conditions are rewritten as
D Jq(x)K = Jf(q(x))K . (3.200)
If D = 0, as is the case when we transform to the frame where the wave is at rest, we
simply recover
0 = f(q(x1 )) f(q(x2 )), (3.201)
f(q(x1 )) = f(q(x2 )), (3.202)
Jf(q(x))K = 0. (3.203)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

216 CHAPTER 3. ONE-DIMENSIONAL COMPRESSIBLE FLOW

That is the fluxes on either side of the discontinuity are equal. This is precisely what we
obtained by our naive analysis. We also get a more general result for D 6= 0, which is the
well-known

f(q(x2 )) f(q(x1 )) Jf(q(x))K

D= = . (3.204)
q(x2 ) q(x1 ) Jq(x)K

The general Rankine-Hugoniot equation then for the one-dimensional Euler equations across
a non-stationary jump is given by

2 1 2 u2 1 u1
2
D 2 u
2 1 u1 =  u
2 2 + p2 1 u21 
p1 .

p2 p1
2 1 2 1 2
e2 + 2 u2 1 e1 + 2 u1 2 u2 e2 + 2 u2 + 2 1 u1 e1 + 21 u21 +
1 2
1
(3.205)

3.3.2 Rayleigh line

If we operate on the momentum equation as follows

p2 = p1 + 1 D 2 2 u22 , (3.206)
21 D 2 22 u22
p2 = p1 + . (3.207)
1 2

Since mass gives us 22 u22 = 21 D 2 we get an equation for the Rayleigh Line,6 a line in (p, 1/)
space:
 
1 1
p2 = p1 + 21 D 2 . (3.208)
1 2

Note that the Rayleigh line

passes through ambient state,

has negative slope,

has a slope with magnitude proportional to square of the wave speed, and

is independent of state and energy equations.

6
John William Strutt (Lord Rayleigh), 1842-1919, aristocratic-born English mathematician and physi-
cist, studied at Cambridge, influenced by Stokes, toured the United States rather than the traditional
continent of Europe, described correctly why the sky is blue, appointed Cavendish professor experimental
physics at Cambridge, won the Nobel prize for the discovery of Argon, described traveling waves and solitons.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

3.3. NORMAL SHOCK WAVES 217

3.3.3 Hugoniot curve

Let us now work on the energy equation, using both mass and momentum to eliminate
velocity. First eliminate u2 via the mass equation:
u2 D2
h2 + 2 = h1 + , (3.209)
2 2
 2
1 1 D D2
h2 + = h1 + , (3.210)
2 2 2
  !
2
D2 1
h2 h1 + 1 = 0, (3.211)
2 2
 
D 2 21 22
h2 h1 + = 0, (3.212)
2 22
 
D 2 (1 2 ) (1 + 2 )
h2 h1 + = 0. (3.213)
2 22
Now use the Rayleigh line, Eq. (3.208), to eliminate D 2 :
  1
2 1 1 1
D = (p2 p1 ) , (3.214)
21 1 2
  1
2 1 2 1
D = (p2 p1 ) , (3.215)
21 1 2
  
2 1 1 2
D = (p2 p1 ) . (3.216)
21 2 1
So the energy equation becomes
   
1 1 1 2 (1 2 ) (1 + 2 )
h2 h1 + (p2 p1 ) = 0, (3.217)
2 21 2 1 22
  
1 1 1 + 2
h2 h1 (p2 p1 ) = 0, (3.218)
2 1 2
 
1 1 1
h2 h1 (p2 p1 ) + = 0. (3.219)
2 2 1
Regrouping to see what induces enthalpy changes, we get
  
1 1 1
h2 h1 = (p2 p1 ) + . (3.220)
2 2 1
This equation is the Hugoniot equation. It
holds that enthalpy change equals the product of the pressure difference and mean
volume,
is independent of wave speed D and velocity u2 , and
is independent of the equation of state.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

218 CHAPTER 3. ONE-DIMENSIONAL COMPRESSIBLE FLOW

3.3.4 Solution procedure for general equations of state

The shocked state can be determined by the following procedure:

specify the equation of state h(p, ),

substitute the equation of state into the Hugoniot, Eq. (3.220), to get a second relation
between p2 and 2 ,
use the Rayleigh line, Eq. (3.208), to eliminate p2 in the Hugoniot so that the Hugoniot
is a single equation in 2 ,
solve for 2 as functions of 1 and D,
back substitute to solve for p2 , u2 , h2 , T2 as functions of 1 and D,
invert to find D as function of 1 state and u2 ,
back transform to laboratory frame to get D as function of 1 state and piston velocity
v2 = vp .

3.3.5 Calorically perfect ideal gas solutions

Let us follow this procedure for the special case of a calorically perfect ideal gas.

h = cp (T To ) + ho , (3.221)
p = RT. (3.222)

Thus,
 
p po
h = cp + ho , (3.223)
R Ro
 
cp p p o
h = + ho , (3.224)
R o
 
cp p po
h = + ho , (3.225)
cp cv o
 
p po
h = + ho . (3.226)
1 o
Evaluate at states 1 and 2 and substitute into the Hugoniot equation, Eq. (3.220):
       
p2 po p1 po
+ ho + ho
1 2 o 1 1 o
  
1 1 1
= (p2 p1 ) + ,
2 2 1

CC BY-NC-ND. 17 June 2016, J. M. Powers.

3.3. NORMAL SHOCK WAVES 219

    
p2 p1 1 1 1
(p2 p1 ) + = 0,
1 2 1 2 2 1
   
1 1 1 1 1 1
p2 p1 = 0,
1 2 22 21 1 1 22 21
   
+1 1 1 +1 1 1
p2 p1 = 0,
2 ( 1) 2 21 2 ( 1) 1 22
   
+1 1 1 +1 1 1
p2 p1 = 0,
1 2 1 1 1 2
+1 1
1 1
12
p2 = p1 +1 1 .
1 2
11

a hyperbola in (p, 1/) space,

1/2 ( 1)/( + 1)(1/1 ) causes p2 , note = = 1.4, 2 6 for infinite

pressure, and

as 1/2 , p2 p1 ( 1)/( + 1), note negative pressure, not physical here.

The Rayleigh line and Hugoniot curve are sketched in Figure 3.8.
p (kPa)
500
shocked state
p2

400
excluded zone
slope of Rayleigh line < 0
excluded
300 2
zone, Rayleigh line, slope ~ D
1/ < 1/min from mass and momentum

200
Hugoniot,
initial state from energy
100
p1
excluded zone, 2nd law violation

1/2 1/1 2 3 4 5 6 7 1/ (kg/m3 )

-(-1) p1
+1 excluded zone, negative pressure

1/min = (-1) 1
(+1) 1

Figure 3.8: Rayleigh line and Hugoniot curve for a typical shocked gas.

Note:

intersections of the two curves are solutions to the equations,

CC BY-NC-ND. 17 June 2016, J. M. Powers.

220 CHAPTER 3. ONE-DIMENSIONAL COMPRESSIBLE FLOW

the ambient state 1 is one solution,

the other solution 2 is known as the shock solution,

the shock solution has higher pressure and higher density,

higher wave speed implies higher pressure and higher density,

a minimum wave speed exists, it

occurs when the Rayleigh line is tangent to the Hugoniot curve,

occurs for infinitesimally small pressure changes,
corresponds to a sonic wave speed, and
has disturbances which are acoustic.

if pressure increases, it can be shown that entropy increases, and

if pressure decreases (for wave speeds which are less than sonic), entropy decreases;
this is non-physical.

Substitute the Rayleigh line into the Hugoniot equation to get a single equation for 2 :
  +1 1 1
1 1 1 1
2
p1 + 21 D 2 = p1 +1 1 1
. (3.227)
1 2 1 2
1

1
This equation is quadratic in 2
and factorizable. Use computer algebra to solve and get
two solutions, one ambient 12 = 1
1
and one shocked solution:
 
1 1 1 2 p1
= 1+ 2
. (3.228)
2 1 + 1 ( 1) D 1

The shocked density 2 is plotted against wave speed D for CPIG air in Figure 3.9.

Note

density solution allows allows all wave speeds 0 < D < ,

plot range, however, is c1 < D < ,

Rayleigh line and Hugoniot show D c1 ,

solution for D = D(vp ), to be shown, rigorously shows D c1 ,

strong shock limit: D 2 , 2 ( + 1)/( 1),

acoustic limit: D 2 p1 /1 , 2 1 , and

CC BY-NC-ND. 17 June 2016, J. M. Powers.

3.3. NORMAL SHOCK WAVES 221

strong
shock
2 (kg/m3) limit
7
6 calorically perfect
5 ideal air
4 exact = 7/5
3 solution R = 0.287 kJ/(kg K)
2
1
D (m/s)
500 1000 1500 2000 2500 3000

D = Dmin = c1

Figure 3.9: Shock density versus shock wave speed for calorically perfect ideal air.

non-physical limit: D 2 0, 2 0.

Back substitute into Rayleigh line and mass conservation to solve for the shocked pressure
and the fluid velocity in the shocked wave frame:
2 1
p2 = 1 D 2 p1 , (3.229)
+1 +1
 
1 2 p1
u2 = D 1+ . (3.230)
+1 ( 1) D 2 1

The shocked pressure p2 is plotted against wave speed D for CPIG air in Figure 3.10 including
both the exact solution and the solution in the strong shock limit. Note for these parameters,
the results are indistinguishable. The shocked wave frame fluid particle velocity u2 is plotted
against wave speed D for CPIG air in Figure 3.11. The shocked wave frame fluid particle
Mach number, M22 = 2 u22 /(p2 ), is plotted against wave speed D for CPIG air in Figure
3.12.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

222 CHAPTER 3. ONE-DIMENSIONAL COMPRESSIBLE FLOW

p2 (Pa)

6
8.x 10
calorically perfect
6 ideal air
6.x 10
ambient = = 7/5
6 100000 Pa
4.x 10 R = 0.287 kJ/(kg K)
exact
6 solution and
2.x 10
strong shock limit
D (m/s)
500 1000 1500 2000 2500 3000

D = Dmin = c1

Figure 3.10: Shock pressure versus shock wave speed for calorically perfect ideal air.

u2 (m/s)

D (m/s)
500 1000 1500 2000 2500 3000
-100
strong calorically perfect
-200
shock ideal air
-300 limit = 7/5
R = 0.287 kJ/(kg K)
-400
exact
-500 solution

u1 = - c1
D = Dmin = c1

Figure 3.11: Shock wave frame fluid particle velocity versus shock wave speed for calorically
perfect ideal air.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

3.3. NORMAL SHOCK WAVES 223

M2 2
calorically perfect
1 ideal air
= 7/5
0.8 exact
R = 0.287 kJ/(kg K)
solution
0.6
strong
shock 0.4
limit
0.2

D (m/s)
0 500 1000 1500 2000 2500 3000

D = Dmin = c1
M2 2 = 1

Figure 3.12: Mach number squared of shocked fluid particle versus shock wave speed for
calorically perfect ideal air.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

224 CHAPTER 3. ONE-DIMENSIONAL COMPRESSIBLE FLOW

Note in the steady frame that the Mach number of the

undisturbed flow is (and must be) > 1: supersonic, and

shocked flow is (and must be) < 1: subsonic.

Transform back to the laboratory frame u = v D:

 
1 2 p1
v2 D = D 1+ , (3.231)
+1 ( 1) D 2 1
 
1 2 p1
v2 = D D 1+ . (3.232)
+1 ( 1) D 2 1

Manipulate the above equation and solve the resulting quadratic equation for D and get

s  2
+1 p1 2 +1
D= v2 + v2 . (3.233)
4 1 4

Now if v2 > 0, we expect D > 0 so take positive root, also set the velocity equal to the
piston velocity v2 = vp .
s  2
+1 p1 2
+1
D= vp + + vp . (3.234)
4 1 4

Note:

acoustic limit: as vp 0, D c1 ; the shock speed approaches the sound speed, and

strong shock limit: as vp , D vp ( + 1)/2.

The shock speed D is plotted against piston velocity vp for CPIG air in Figure 3.13. Both
the exact solution and strong shock limit are shown.

calorically perfect
D (m/s)
ideal air
exact
= 7/5
1200 solution
R = 0.287 kJ/(kg K)
1000
800 strong
shock
600
acoustic limit
limit, 400
D c1 200
v (m/s )
200 400 600 800 1000 p

Figure 3.13: Shock speed versus piston velocity for calorically perfect ideal air.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

3.3. NORMAL SHOCK WAVES 225

If we define the Mach number of the shock as

D
Ms , (3.235)
c1
we get
s  2
+ 1 vp vp2 +1
Ms = + 1+ . (3.236)
4 RT1 RT1 4

The shock Mach number Ms is plotted against piston velocity vp for CPIG air in Figure 3.14.
Both the exact solution and strong shock limit are shown.

Ms
exact calorically perfect
3.5 solution ideal air
3 = 7/5
2.5 R = 0.287 kJ/(kg K)
2 strong
1.5
shock
acoustic limit
limit, 1
Ms 1 0.5
vp (m/s)
200 400 600 800 1000

Figure 3.14: Shock Mach number versus piston velocity for calorically perfect ideal air.

3.3.6 Acoustic limit

Consider that state 2 is a small perturbation of state 1 so that

2 = 1 + , (3.237)
u2 = u1 + u1 , (3.238)
p2 = p1 + p. (3.239)

Substituting into the normal shock equations, we get

(1 + ) (u1 + u) = 1 u1 , (3.240)
(1 + ) (u1 + u)2 + (p1 + p) = 1 u1 2 + p1 , (3.241)
p1 + p 1 p1 1 2
+ (u1 + u)2 = + u1 . (3.242)
1 1 + 2 1 1 2
Expanding, we get

1 u1 + u1 () + 1 (u) + () (u) = 1 u1


1 u1 2
+ 21 u1 (u) + u1 () + 1 (u)2 + 2u1 (u) () + () (u)2
2

CC BY-NC-ND. 17 June 2016, J. M. Powers.

226 CHAPTER 3. ONE-DIMENSIONAL COMPRESSIBLE FLOW

+ (p1 + p) = 1 u1 2 + p1
 
p1 1 p1 1

+ P 2 + ... + u1 2 + 2u1 (u) + (u)2
1 1 1 1 2
p1 1 2
= + u1
1 1 2
Subtracting the base state and eliminating products of small quantities yields

u1 () + 1 (u) = 0, (3.243)
21 u1 (u) + u1 2 () + p = 0, (3.244)
 
1 p1
p 2 + u1 (u) = 0. (3.245)
1 1 1
In matrix form this is

u1 1 0 0
u1 2 21 u1 1 u = 0 . (3.246)
p1 1
1 21
u1 1 1 p 0
As the right hand side is zero, the determinant must be zero and there must be a linear
dependency of the solution. First check the determinant:
   
2 u1 2 p1
u1 u1 u1 1 + = 0, (3.247)
1 1 1 1 21
 
u1 2 1 2 p1
(2 ( 1)) u1 + = 0, (3.248)
1 1 1
 
2 2 p1
u1 ( + 1) u1 + = 0, (3.249)
1
p1
u1 2 = = c21 . (3.250)
1
So the velocity is necessarily sonic for a small disturbance.
Take u to be known and solve a resulting 2 2 system:
    
u1 0 1 u
p1 1 = . (3.251)
1 21 1 1 p u1 u
Solving yields
1 u u
= q = 1 (3.252)
p1
1 c1
r
p1
p = 1 u = 1 c1 u. (3.253)
1

CC BY-NC-ND. 17 June 2016, J. M. Powers.

3.4. FLOW WITH AREA CHANGE AND NORMAL SHOCKS 227

3.4 Flow with area change and normal shocks

This section will consider flow from a reservoir with the fluid at stagnation conditions to a
constant pressure environment. The pressure of the environment is commonly known as the
back pressure: pb .

Generic problem: Given A(x), stagnation conditions and pb , find the pressure, tempera-
ture, density at all points in the duct and the mass flow rate.

3.4.1 Converging nozzle

A converging nozzle operating at several different values of pb is sketched in Figure 3.15. The
. .
m/mmax
e d c
pb 1

b
po pe

0 pb/p o
p*/p o 1
p(x)/p o
1
a--subsonic exit
b--subsonic exit
p*/p o c--sonic exit
d--choked, external expansion
e--choked, external expansion

x
xe

Figure 3.15: Converging nozzle sketch.

flow through the duct can be solved using the following procedure:
check if pb p ,
if so, set pe = pb ,
determine Me from isentropic flow relations,
determine A from A/A relation,

at any point in the flow where A is known, compute A/A and then invert A/A relation
to find local M.
Note:

CC BY-NC-ND. 17 June 2016, J. M. Powers.

228 CHAPTER 3. ONE-DIMENSIONAL COMPRESSIBLE FLOW

These flows are subsonic throughout and correspond to points a and b in Figure 3.15.
If pb = p then the flow is sonic at the exit and just choked. This corresponds to point
c in Figure 3.15.
If pb < p , then the flow chokes, is sonic at the exit, and continues to expand outside
of the nozzle. This corresponds to points d and e in Figure 3.15.

3.4.2 Converging-diverging nozzle

A converging-diverging nozzle operating at several different values of pb is sketched in Figure
3.16.

pb

po pt pe

possible
normal
shock

p(x)/p o

1 a--subsonic exit
b--subsonic exit
c--subsonic design
d--shock in duct
p*/p o e-shock at end of duct
sonic f--external compression
throat
g--supersonic design
h--external expansion
x
xt xe
. .
m/mmax
hg f e d c
1

0 1 pb / po
p*/p o

Figure 3.16: Converging-diverging nozzle sketch.

The flow through the duct can be solved using the a very similar following procedure

CC BY-NC-ND. 17 June 2016, J. M. Powers.

3.5. RAREFACTIONS AND THE METHOD OF CHARACTERISTICS 229

set At = A ,

with this assumption, calculate Ae /A ,

determine Mesub , Mesup , both supersonic and subsonic, from A/A relation,

determine pesub , pesup, from Mesub , Mesup ; these are the supersonic and subsonic design
pressures,

if pb > pesub, the flow is subsonic throughout and the throat is not sonic. Use same
procedure as for converging duct: Determine Me by setting pe = pb and using isentropic
relations,

if pesub > pb > pesup, the procedure is complicated.

estimate the pressure with a normal shock at the end of the duct, pesh .
If pb pesh , there is a normal shock inside the duct,
If pb < pesh , the duct flow is shockless, and there may be compression outside the
duct.

if pesup = pb , the flow is at supersonic design conditions and the flow is shockless, and

if pb < pesup , the flow in the duct is isentropic and there is expansion outside the duct.

3.5 Rarefactions and the method of characteristics

Here we discuss how to model expansion waves in a one-dimensional unsteady, inviscid, non-
heat conducting fluid. This analysis is a good deal more rigorous than much of traditional
one-dimensional gas dynamics, and draws upon some of the more difficult mathematical
methods we will encounter.
In assuming no diffusive transport, we have eliminated all mechanisms for entropy gen-
eration; consequently, we will be able to model the process as isentropic. We note that even
without diffusion, shocks can generate entropy. However, the expansion waves are inherently
continuous, and do remain isentropic. We will consider a general equation of state, and
later specialize to a calorically perfect ideal gas. The problem is inherently non-linear and
is modeled by partial differential equations of the type which is known as hyperbolic. Such
problems, in contrast to say Laplaces equation, which requires boundary conditions, require
initial data only, and no boundary data.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

230 CHAPTER 3. ONE-DIMENSIONAL COMPRESSIBLE FLOW

3.5.1 Inviscid one-dimensional equations

The equations to be considered are shown here in non-conservative form
u
+u + = 0, (3.254)
t x x
u u p
+ u + = 0, (3.255)
t x x
s s
+u = 0, (3.256)
t x
p = p(, s). (3.257)

Here we have written the energy equation in terms of entropy. The development of this was
shown in Chapter 1. We have also utilized the general result from thermodynamics that any
intensive property can be written as a function of two other independent thermodynamic
properties. Here we have chosen to write pressure as a function of density and entropy. Thus
we have four equations for the four unknowns, , u, p, s.
Now we note that
 
p  p 
dp = d + ds, so, (3.258)
s s 
    
p  p   p  s 
= + . (3.259)
x t s x t s  x t

Now, let us define thermodynamic properties c2 and as follows

 
2 p  p 
c , . (3.260)
s s 

We will see that will be unimportant, and will be able to ascribe to c2 the physical
significance of the speed of propagation of small disturbances, the so-called sound speed,
which we have already encountered in acoustics. If we know the equation of state, then we
can think of c2 and as known thermodynamic functions of and s. Our definitions give us
p s
= c2 + . (3.261)
x x x
Substituting into our governing equations, we see that pressure can be eliminated to give
three equations in three unknowns:
u
+u + = 0, (3.262)
t x x
u u s
+ u + c2 + = 0, (3.263)
t x x x
s s
+u = 0. (3.264)
t x

CC BY-NC-ND. 17 June 2016, J. M. Powers.

3.5. RAREFACTIONS AND THE METHOD OF CHARACTERISTICS 231

Now if s/t + u(s/x) = 0, we can say that if s = s(x, t),

s s
ds = dt + dx, (3.265)
t x
ds s dx s
= + . (3.266)
dt t dt x
Thus, on curves where dx/dt = u, we have from substituting Eq. (3.266) into the energy
equation (3.264)
ds
= 0. (3.267)
dt
Thus we have converted the partial differential equation into an ordinary differential equa-
tion. This can be integrated to give us
dx
s = C, on a particle pathline, dt
= u. (3.268)

This scenario is sketched on the so-called x t diagram of Figure 3.17.

t
pathlines

s = s1 s = s2
s = s0
s = s3

s = s4

dx
Figure 3.17: xt diagram showing maintenance of entropy s along particle pathlines dt
=u
for isentropic flow.

This result is satisfying, but not complete, as we do not in general know where the
pathlines are. Let us try to apply this technique to the system in general. Consider our
equations in matrix form:

1 0 0 t
u 0 x
0
0 0 u + c2 u u = 0 . (3.269)
t x
s s
0 0 1 t
0 0 u x
0

CC BY-NC-ND. 17 June 2016, J. M. Powers.

232 CHAPTER 3. ONE-DIMENSIONAL COMPRESSIBLE FLOW

These equations are of the form

uj uj
Aij + Bij = Ci . (3.270)
t x

As described by Whitham,7 there is a general technique to analyze such equations. First

pre-multiply both sides of the equation by a yet to be determined vector of variables i :

uj uj
i Aij + i Bij = i C i . (3.271)
t x
Now, this method will work if we can choose i to render the above product to be of the
form similar to /t + u(/x). Let us take
 
uj uj uj uj
i Aij + i Bij = mj + , (3.272)
t x t x
duj dx
= mj on = . (3.273)
dt dt
So comparing terms, we see that

i Aij = mj , i Bij = mj , (3.274)

i Aij = mj , (3.275)

so, we get by eliminating mj that

i (Aij Bij ) = 0. (3.276)

This is a left eigenvalue problem. We set the determinant of Aij Bij to zero for a non-trivial
solution and find  
 u 0 

 c2 ( u)  = 0. (3.277)

 0 0 u
Evaluating, we get


( u) ( u)2 + ( u)(c2 ) = 0, (3.278)


( u) ( u)2 c2 = 0. (3.279)

Solving we get
= u, = u c. (3.280)
7
Gerald Beresford Whitham, 1927-, applied mathematician and developer of theory for non-linear wave
propagation.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

3.5. RAREFACTIONS AND THE METHOD OF CHARACTERISTICS 233

Now the left eigenvectors i give us the actual equations. First for = u, we get

uu 0
( 1 2 3 ) c2 (u u) = ( 0 0 0 ) , (3.281)
0 0 uu

0 0
( 1 2 3 ) c2 0 = ( 0 0 0 ) . (3.282)
0 0 0
Two of the equations require that 1 = 0 and 2 = 0. There is no restriction on 3 . We will
select a normalized solution so that
i = (0, 0, 1). (3.283)
Thus i Aij (uj /t) + i Bij (uj /x) = i Ci gives

1 0 0 t
u 0 x
0
( 0 0 1 ) 0 0 u t
+ ( 0 0 1 ) c2 u u
x
= (0 0 1)0,
s s
0 0 1 t
0 0 u 0
x


t x
( 0 0 1 ) u
t
+ ( 0 0 u ) u = 0,
x
s s
t x
s s
+u = 0. (3.284)
t x
So as before with s = s(x, t), we have ds = (s/t)dt + (s/x)dx, and ds/dt = s/t +
(dx/dt)(s/x). Now if we require dx/dt to be a particle pathline, dx/dt = u, then our
energy equation gives us
ds dx
= 0, on = u. (3.285)
dt dt
The special case in which the pathlines are straight in xt space, corresponding to a uniform
velocity field of u(x, t) = uo , is sketched in the x t diagram of Figure 3.18.
Now let us look at the remaining eigenvalues, = u c.

ucu 0
( 1 2 3 ) c2 (u c u) = (0 0 0), (3.286)
0 0 ucu

c 0
( 1 2 3 ) c2 c = ( 0 0 0 ) . (3.287)
0 0 c
As one of the components of the left eigenvector should be arbitrary, we will take 1 = 1; we
arrive at the following equations then
1
c c2 2 = 0, = 2 = , (3.288)
c

CC BY-NC-ND. 17 June 2016, J. M. Powers.

234 CHAPTER 3. ONE-DIMENSIONAL COMPRESSIBLE FLOW

pathlines

s = s2 uo
s = s1 s = s3
s = s0
s = s4

Figure 3.18: x t diagram showing maintenance of entropy s along particle pathlines

dx/dt = uo for isentropic flow.

1
c2 = 0, = 2 = , (3.289)
c

2 c3 = 0, = 3 = 2 . (3.290)
c
Thus i Aij (uj /t) + i Bij (uj /x) = i Ci gives

1 0 0 t
u 0 x
0
1
( 1 c c2 ) 0 0 u 1 2
+ ( 1 c c2 ) c u u
= ( 1 c c2 ) 0 ,
1
t x
s s
0 0 1 0 0 u 0
t x


t x
( 1 c
c2
) u
t
+ ( u c u
c
c + u
c2
) u
x
= 0,
s s
t x
  
u u  u s u s
+ (u c) + 1 + + = 0, (3.291)
t x c t c x c2 t c2 c x
     
u u s s
+ (u c) + (u c) + 2 + (u c) = 0, (3.292)
t x c t x c t x
     
2 u u s s
c + (u c) c + (u c) + + (u c) = 0. (3.293)
t x t x t x
Now on lines where dx/dt = uc, we get a transformation of the partial differential equations
to ordinary differential equations:
d du ds dx
c2 c + = 0, on = u c. (3.294)
dt dt dt dt

CC BY-NC-ND. 17 June 2016, J. M. Powers.

3.5. RAREFACTIONS AND THE METHOD OF CHARACTERISTICS 235

A sketch of the characteristics, the lines on which the differential equations are obtained,
are sketched in the x t diagram of Figure 3.19.

pathline
characteristic
t acoustic dx
__ = u
characteristic dt
dx
__ = u- c dx
__ = u+ c
dt dt
acoustic
characteristic

Figure 3.19: x t diagram showing characteristics for pathlines dx/dt = u and acoustic
waves dx/dt = u c.

3.5.2 Homeoentropic flow of an ideal gas

The equations developed so far are valid for a general equation of state. Here let us now
consider the flow of a calorically perfect ideal gas, so p = RT and e = cv T + e. Further
let us take the flow to be homeoentropic, that is to say, not only does the entropy remain
constant on pathlines, which is isentropic, but it has the same value on each streamline.
That is the entropy field is a constant. Consequently, we have the standard relations for a
calorically perfect ideal gas:
p
c2 = , (3.295)

p
= A, (3.296)

where A is a constant. Because of homeoentropy, we no longer need consider the energy

equation, and the linear combination of mass and linear momentum equations reduces to

d du dx
c2 c = 0, on = u c. (3.297)
dt dt dt

CC BY-NC-ND. 17 June 2016, J. M. Powers.

236 CHAPTER 3. ONE-DIMENSIONAL COMPRESSIBLE FLOW

Rearranging, we get
du d c dx
= , on = u c. (3.298)
dt dt dt
1
Now c2 = p/ = A1 , and c = A 2 , so
du p 1 d p 2 d  1 
= A 2 1 = A 2 . (3.299)
dt dt 1 dt
Regrouping, we find
 
d p 2 1
u A 2 = 0, (3.300)
dt 1
 
d 2
u c = 0. (3.301)
dt 1
Following notation used by Courant8 and Friedrichs,9 we then integrate each of these equa-
tions, which are homogeneous, along characteristics to obtain algebraic relations
2 dx
u+ c = 2r, on = u + c, C + characteristic, (3.302)
1 dt
2 dx
u c = 2s, on = u c, C characteristic. (3.303)
1 dt
A sketch of the characteristics is given in the x t diagram of Figure 3.20. Now r and
s can take on different values, depending on which characteristic we are on. On a given
characteristic, they remain constant. Let us define additional parameters and to identify
which characteristic we are on. So we have
2 dx
u+ c = 2r(), on = u + c, C + characteristic, (3.304)
1 dt
2 dx
u c = 2s(), on = u c, C characteristic. (3.305)
1 dt
These quantities are known as Riemann invariants.10
8
Richard Courant, 1888-1972, Prussian-born German mathematician, received Ph.D. under David Hilbert
at Gottingen, compiled Hilberts course notes into classic two-volume text of applied mathematics, drafted
into German army in World War I, where half of his unit was killed in action, developed telegraph system
which used the earth as a conductor for use in the trenches of the Western front, expelled from Gottingen
by the Nazis in 1933, fled Germany, and founded the Courant Institute of Mathematical Sciences at New
York University, author of classic mathematical text on supersonic fluid mechanics.
9
Kurt Otto Friedrichs, 1901-1982, German-born mathematician who emigrated to the United States in
1937, student of Richard Courants at Gottingen, taught at Aachen, Braunschweig, and New York University,
worked on partial differential equations of mathematical physics and fluid mechanics.
10
Georg Friedrich Bernhard Riemann, 1826-1866, German mathematician and geometer whose work in
non-Euclidean geometry was critical to Einsteins theory of general relativity, produced the first major study
of shock waves.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

3.5. RAREFACTIONS AND THE METHOD OF CHARACTERISTICS 237

arbitrary region __ = u+ c on C+
dx __ = u - c on C -
dx
dt dt
of interest

+
C +
C

-
C - +
C C
-
C

+
C

Figure 3.20: x t diagram showing C + and C characteristics dx/dt = u c.

3.5.3 Simple waves

Simple waves are defined to exist when either r() or s() are constant everywhere in x t
space and not just on characteristics. For example say s() = so . Then the Riemann
invariant
2
u c = 2so , (3.306)
1
is actually invariant over all of x t space. Now the other Riemann invariant,
2
u+ c = 2r(), (3.307)
1
takes on many values depending on . However, it is easily shown that for the simple wave
that the characteristics have a constant slope in the x t plane as sketched in the x t
diagram of Figure 3.21.
Now consider a rarefaction with a prescribed piston motion u = up (t). A sketch is given
in the x t diagram of Figure 3.22.
2
For this configuration, the Riemann invariant u 1 c = 2so is valid everywhere. Now
when t = 0, we have u = 0, c = co , so
2 2
u c= co . (3.308)
1 1
Consider now a special characteristic C + at t = t. At this time the piston moves with
velocity up , and the fluid velocity at the piston face is
uf ace (t) = up . (3.309)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

238 CHAPTER 3. ONE-DIMENSIONAL COMPRESSIBLE FLOW

arbitrary region
of interest

+
C

+
C

+
C

+
C

Figure 3.21: x t diagram showing C + for a simple wave.

We get cf ace (t) from Eq. (3.308):

2 2
uf ace cf ace = co , (3.310)
| {z } 1 1
up
1
cf ace (t = t) = co + up . (3.311)
2
Also from Eq. (3.308), we have
1
c = co + u, (3.312)
2
which is valid everywhere.
Now on C + , we have
 
2 2
u+ c = uf ace + cf ace , (3.313)
1 1 t=t
   
2 1 2 1
u+ co + u = up + co + up , (3.314)
1 2 1 2
2 2
2u + co = 2up + co , (3.315)
1 1
u = up on C + . (3.316)
So on C + , we have
1
c = co + up . (3.317)
2

CC BY-NC-ND. 17 June 2016, J. M. Powers.

3.5. RAREFACTIONS AND THE METHOD OF CHARACTERISTICS 239

+
C

+ 2 c = 2 r ( )
u p (t) C: u + ______
+1 3

+ 2 c = 2 r ( )
C: u + ______
+1 2

+ 2 c = 2 r ( )
C : u + ______
+1 1

u p (t)

x=0

Figure 3.22: x t diagram showing C + characteristics for isentropic rarefaction problem,

along with piston cylinder arrangement.

So for C + , we get
dx 1 +1
= u + c = up + co + up = up + co . (3.318)
dt 2 2
for a particular characteristic, this slope is a constant, as was earlier suggested.
Now for prescribed motion, up decreases with time and becomes more negative; hence
the slope of our C + characteristic decreases, and they diverge in x t space. The slope of
the leading characteristic is co , the ambient sound speed. The characteristic we consider,
C + , along with a few other is sketched in the x t diagram of Figure 3.23. We can use our
Riemann invariant along with isentropic relations to obtain other flow variables. From Eq.
(3.308), we get
c 1 u
=1+ . (3.319)
co 2 co
1
Since the flow is homeoentropic, we have c/co = (/o ) 2 and p/po = (/o ) , so
  2
p 1 u 1
= 1+ , (3.320)
po 2 co
  2
1 u 1
= 1+ . (3.321)
o 2 co

CC BY-NC-ND. 17 June 2016, J. M. Powers.

240 CHAPTER 3. ONE-DIMENSIONAL COMPRESSIBLE FLOW

+
u p (t) C

up u face( t ) = up
-1
c ( t ) = co + up
face 2

Figure 3.23: x t diagram showing C + for our rarefaction problem.

3.5.4 Centered rarefaction

If the piston is suddenly accelerated to a constant velocity, then a family of characteristics
clusters at the origin on the xt diagram and fans out in a centered rarefaction. This can also
be studied using the similarity transformation = x/t which reduces the partial differential
equations to ordinary differential equations. Relevant sketches comparing centered to non-
centered rarefactions are shown in in the x t diagram of Figure 3.24.

Example 3.4
Analyze a centered rarefaction fan propagating into calorically perfect ideal air for a piston suddenly
accelerated from rest to up = 100 m/s. Take the ambient air to be at po = 100 kP a, To = 300 K.
po 100 kP a 3
The ideal gas law gives o = RT o
= (287 J/kg/K)(300 K) = 1.16 kg/m . Now
s  
p 7 J
co = RTo = 287 (300 K) = 347 m/s. (3.322)
5 kg K

On the final characteristic of the fan, Cf+ : u = up = 100 m

s . So

1  m  7/5 1  m m
c = co + up = 347 + 100 = 327 . (3.323)
2 s 2 s s
Now the final pressure is

  2
! 2(7/5)
7/5 1 100 m
7/51
pf 1 uf 1 s
= 1+ = 1+ = 0.660 (3.324)
po 2 co 2 347 m
s

Hence pf = 66.0 kP a. Since the flow is homeoentropic, we get

  1  
pf kg kg
f = o = 1.16 3 (.660)5/7 = 0.863 3 . (3.325)
po m m

CC BY-NC-ND. 17 June 2016, J. M. Powers.

3.5. RAREFACTIONS AND THE METHOD OF CHARACTERISTICS 241

Centered Rarefaction Non-Centered Rarefaction

t t
up + + +
C C C +
C u p (t) +
C +
+ + C
C C +
+ C
C
+
C

particle path particle path

x x

up
x
p

Figure 3.24: x t diagram centered and non-centered rarefactions, along with pressure and
velocity profiles for centered fans.

And the final temperature is

pf 66.0 103 P a
Tf = = = 266.5 K. (3.326)
f R (0.863 kg/m3 )(287 J/kg/K)
Recall from linear theory that
u u
o , p o co u, T ( 1)To . (3.327)
co co
We compare the results of this problem with the estimates of linear acoustic theory and see
kg kg
exact = 0.298 , linear = 0.334 , (3.328)
m3 m3
pexact = 34 kP a, plinear = 40.3 kP a, (3.329)
Texact = 33.5 K, Tlinear = 34.6 K. (3.330)

3.5.5 Simple compression

We sketch a simple compression in the x t diagram of Figure 3.25.

3.5.6 Two interacting expansions

We sketch two interacting expansion waves in the x t diagram of Figure 3.26.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

242 CHAPTER 3. ONE-DIMENSIONAL COMPRESSIBLE FLOW

+
C
+
C
+
u p (t) C
+
C shock
+ formation
C
+
C +
C

ambient
region

u p (t)

x=0

Figure 3.25: x t diagram for simple compression.

3.5.7 Wall interactions

We sketch an expansion wall interaction in the x t diagram of Figure 3.27.

3.5.8 Shock tube

We sketch the behavior of a shock tube in the diagrams of Figure 3.28.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

3.5. RAREFACTIONS AND THE METHOD OF CHARACTERISTICS 243

uniform
flow

u p (t)
u p (t)
uniform uniform
flow flow

fluid
at rest

x
x=L

u p (t)

x=0 x=L

Figure 3.26: x t diagram for two interacting expansion waves.

t
rest

uniform
u p (t)
flow

flow
at rest

x
x=L

u p (t)

x=0 x=L

Figure 3.27: x t diagram for expansion wall interaction.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

244 CHAPTER 3. ONE-DIMENSIONAL COMPRESSIBLE FLOW

compression
contact
discontinuity
rarefaction

shock
rest rest

uniform
uniform

rest rest
x
x=L

t=0

x
p

t>0

t>0

x

t>0

Figure 3.28: x t and p, , T versus x behavior for a shock tube.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

3.5. RAREFACTIONS AND THE METHOD OF CHARACTERISTICS 245

3.5.9 Final note on method of characteristics

We have described here a common and traditional approach to the method of characteristics
(MOC). Using common notation, we have written what began as partial differential equations
(PDEs) in the form of ordinary differential equations (ODEs), and it is often said that the
method of characteristics is a way to transform PDEs into ODEs. However, the equations
which result are certainly not in a standard form for ODEs; they are burdened with unusual
side conditions.
It is in fact more sound to state that the MOC transforms the PDEs in (x, t) space to
another set of PDEs in a new space (r, s) in which the integration is much easier. Consider
for example a model equation which is hyperbolic, the inviscid Burgers equation:

u u
+u = 0. (3.331)
t x
Now consider a general transformation (x, t) (r, s). Applying the chain rule, we get

u u r u s
= + , (3.332)
t r t s t
u u r u s
= + . (3.333)
x r x s x
In transformed space, the inviscid Burgers equation becomes
 
u r u s u r u s
+ +u + = 0. (3.334)
r t s t r x s x

Now we also have

x x
dx = dr + ds, (3.335)
r s
t t
dt = dr + ds. (3.336)
r s
With the Jacobian11
x t x t
J= , (3.337)
r s s r
we invert to find
 
1 t x
dr = dx dt , (3.338)
J s s
 
1 t x
ds = dx + dt . (3.339)
J r r
11
Carl Gustav Jacob Jacobi, 1804-1851, Prussian born, prolific German mathematician. The Jacobian
determinant was extensively studied by Jacobi, but first identified by Cauchy.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

246 CHAPTER 3. ONE-DIMENSIONAL COMPRESSIBLE FLOW

So it is easy to see that we get the following for the partial derivatives

r 1 x r 1 t
= , = , (3.340)
t J s x J s
s 1 x s 1 t
= , = . (3.341)
t J r x J r
Substituting into the inviscid Burgers equation, we get
  
1 u x u x u t u t
+ +u = 0, (3.342)
J r s s r r s s r
u x u x u t u t
+ +u u = 0. (3.343)
r s s r r s s r
Up to this point we have a perfectly general transformation and a perfectly general inviscid
Burgers equation, now cast in the transformed space. Let us now demand of our transfor-
mation that
x t
=u , t(r, s) = s. (3.344)
s s
The first of these says that on any line on which r is a constant that for a given change in
s, the ratio of the change in x to that in t will be equal to u. This is a generalization of our
more standard statement that on characteristics, dx/dt = u. The second is a convenience,
and we actually need not be as restrictive. With this specification, our inviscid Burgers
equation becomes

u x u x u t u t
+ +u u = 0, (3.345)
r s
|{z} s r r |{z}
s s |{z}
r
t =1 =0
= u s =u
u u x u
u + +u = 0, (3.346)
r s r r
u x
= 0. (3.347)
s r
Now, let us require that x/r 6= 0; hence in this special transformed space, we have that

u
= 0. (3.348)
s
This has solution
u = f (r), (3.349)
where f (r) is an arbitrary function. We now substitute this into x/s = u(t/s) to get

x t
= f (r) , (3.350)
s s

CC BY-NC-ND. 17 June 2016, J. M. Powers.

3.5. RAREFACTIONS AND THE METHOD OF CHARACTERISTICS 247

which can be integrated to get

x = f (r)t + g(r). (3.351)
Now substituting t = s and setting g(r) = r arbitrarily so that our transformation maps x
into r when t = s = 0, we get
x = f (r)s + r. (3.352)
In summary we can write a solution parametrically in terms of our transformed space as

u(r, s) = f (r), (3.353)

x(r, s) = f (r)s + r, (3.354)
t(r, s) = s. (3.355)

So given an initial distribution of u, we can select a domain in (r, s) and parametrically

determine u as a function of x and t. While this formulation maps every (r, s) into (u, x, t),
we cannot be assured that in physical space that the same (x, t) may not map into non-
unique values of u! This multi-valuedness actually indicates that a shock has formed, and
correct insertion of a shock will eliminate the difficulty.

Example 3.5
If we have the inviscid Burgers equation u/t + u(u/x) = 0 with u(x, 0) = sin(x), find u, and
plot u(x) for t = 0, 1, 2.
When t = s = 0, we have x = r, so f (r) = sin(r), and our solution is

u(r, s) = sin(r), (3.356)

x(r, s) = s sin(r) + r, (3.357)
t(r, s) = s. (3.358)

We can use this solution to form parametric plots and effectively form u(x) for various values of t.
These are shown in Figure 3.29. It is clear that as time advances the left side of the wave is flattening
and the right side is steepening. The left side is undergoing what is equivalent to a rarefaction, and the
right side is undergoing what is equivalent to a compression. At t = 3, the wave has steepened enough
so that u is a multivalued function of x. In a physical problem, this would indicate that a shock has
formed.

This procedure can be extended to the Euler equations, though it is somewhat more com-
plicated. For isentropic Euler equations, Courant and Friedrichs give some special solutions
for rarefactions.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

248 CHAPTER 3. ONE-DIMENSIONAL COMPRESSIBLE FLOW

u
1

0.8
t=0
0.6 t=1
0.4 t=2

0.2

x
0.2 0.4 0.6 0.8 1 1.2

Figure 3.29: u(x) for t = 0, 1, 2 for inviscid Burgers equation problem.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

Chapter 4

Potential flow

see Panton, Chapter 18

see Yih, Chapter 4

This chapter will consider potential flow. A good deal of highly developed and beautiful
mathematical theory was generated for potential flows in the nineteenth century. Addition-
ally, these solutions can be applied in highly disparate fields, as the equations governing
potential flow of a fluid are identical in form to those governing some forms of energy and
mass diffusion, as well as electro-magnetics.
Despite its beauty, in some ways it is impractical for many engineering applications,
though not all. As the theory necessarily ignores all vorticity generating mechanisms, it
must ignore viscous effects. Consequently, the theory is incapable of predicting drag forces
on solid bodies. Consequently, those who needed to know the drag, resorted in the nineteenth
century to far more empirically based methods.
In the early twentieth century, Prandtl took steps to reconcile the practical viscous world
of engineering with the more mathematical world of potential flow with his viscous boundary
layer theory. He showed that indeed potential flow solutions could be of great value away
from no-slip walls, and provided a recipe to fix the solutions in the neighborhood of the wall.
In so doing, he opened up a new field of applied mathematics known as matched asymptotic
analysis.
So why study potential flows? The following arguments offer some justification.

portions of real flow fields are well described by this theory, and those that are not can
often be remedied by application of a viscous boundary layer theory,

study of potential flow solutions can give great insight into fluid behavior and aid in
the honing of a more precise intuition,

fundamental solutions are useful as test cases for verification of numerical methods,
and

there is pedantic and historical value in knowing potential flow.

249
250 CHAPTER 4. POTENTIAL FLOW

4.1 Stream functions and velocity potentials

We first consider stream functions and velocity potentials. We have seen velocity potentials
before in study of ideal vortices. In this chapter, we will adopt the same assumption of
irrotationality, and further require that the flow be two-dimensional. Recall if a flow velocity
is confined to the x y plane, then the vorticity vector is confined to the z direction and
takes the form
0
= 0 . (4.1)
v u
x
y
Now if the flow is two-dimensional and irrotational, we have
v u
= 0. (4.2)
x y
Moreover, because of irrotationality, we can express the velocity vector v as the gradient of
a potential , the velocity potential:
v = . (4.3)
Note that with this definition, fluid flows from regions of low velocity potential to regions of
high velocity potential. Thus,

u = , (4.4)
x

v = . (4.5)
y
We see by substitution into the equation for vorticity, that this is true identically:
   
v u
= = 0. (4.6)
x y x y y x
Now for two-dimensional incompressible flows, we have
u v
+ = 0. (4.7)
x y
Substituting for u and v in favor of , we get
   

+ = 0, (4.8)
x x y y
2 = 0. (4.9)

Now if the flow is incompressible, we can also define the stream function as follows:

u= , v= . (4.10)
y x

CC BY-NC-ND. 17 June 2016, J. M. Powers.

4.1. STREAM FUNCTIONS AND VELOCITY POTENTIALS 251

Direct substitution into the mass conservation equation shows that this yields an identity:
   
u v
+ = + = 0. (4.11)
x y x y y x
Now, in an equation which will be critically important soon, we can set our definitions of u
and v in terms of and equal to each other, as they must be:

= , (4.12)
x
|{z} y
|{z}
u u

= . (4.13)
y | x}
{z
|{z}
v v
Now if we differentiate the first equation with respect to y, and the second with respect to
x we see
2 2
= , (4.14)
yx y 2
2 2
= 2, (4.15)
xy x
now subtract the second from the first to get (4.16)
2 2
0 = + 2, (4.17)
y 2 x
2
= 0. (4.18)
Let us know examine lines of constant (equipotential lines) and lines of constant
(which we will see are streamlines). So take = C1 , = C2 . For we get

d = dx + dy = 0, (4.19)
x y
d = udx + vdy = 0, (4.20)


dy  u
 = . (4.21)
dx =C1 v

Now for we similarly get

d = dx + dy = 0, (4.22)
x y
d = vdx + udy = 0, (4.23)

dy  v
 = (4.24)
dx =C2 u

We note two features

CC BY-NC-ND. 17 June 2016, J. M. Powers.

252 CHAPTER 4. POTENTIAL FLOW


dy  1
= ; hence, lines of constant are orthogonal to lines of constant ,
dx =C1 dy
|
dx =C
2
and

on = C2 , we see that dx/u = dy/v; hence, lines of = C2 must be streamlines.

As an aside, we note that the definition of the stream function u = /y, v = /x,
can be rewritten as
dx dy
= , = . (4.25)
y dt x dt

This is a common form from classical dynamics in which we can interpret as the Hamil-
tonian1 of the system. We shall not pursue this path, but note that a significant literature
exists for Hamiltonian systems.
Now the study of and is essentially kinematics. The only incursion of dynamics is that
we must have irrotational flow. Recalling the Helmholtz equation, Eq. (2.132), we realize that
we can only have potential flow when the vorticity generating mechanisms (three-dimensional
effects, non-conservative body forces, baroclinic effects, and viscous effects) are suppressed.
In that case, the dynamics, that is the driving force for the fluid motion, can be understood
in the context of the unsteady Bernoulli equation, Eq. (1.958, taken for incompressible flow
and negligible body force, in which limit, Eq. (1.949) reduces to = p/:

1 p
+ ()T + = f (t). (4.26)
t 2

Note that we do not have to require steady flow to have a potential flow field. It is also easy
to correct for the presence of a conservative body force.
Now solutions to the two key equations of potential flow 2 = 0, 2 = 0, are most effi-
ciently studied using methods involving complex variables. We will delay discussing solutions
until we have reviewed the necessary mathematics.

4.2 Mathematics of complex variables

Here we briefly introduce relevant elements of complex variable theory. Recall that the
imaginary number i is defined such that

i2 = 1, i= 1. (4.27)

1
William Rowan Hamilton, 1805-1865, Anglo-Irish mathematician.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

4.2. MATHEMATICS OF COMPLEX VARIABLES 253

4.2.1 Eulers formula

We can get a very useful formula Eulers formula, by considering the following Taylor2
expansions of common functions about t = 0:
1 2 1 3 1 4 1 5
et = 1 + t + t + t + t + t ..., (4.28)
2! 3! 4! 5!
1 2 1 3 1 4 1 5
sin t = 0 + t + 0 t t + 0 t + t . . . , (4.29)
2! 3! 4! 5!
1 2 1 3 1 4 1
cos t = 1 + 0t t + 0 t + t + 0 t5 . . . (4.30)
2! 3! 4! 5!
With these expansions now consider the following combinations: (cos t + i sin t)t= and
et |t=i :
1 2 1 1 1
cos + i sin = 1 + i i 3 + 4 + i 5 + . . . , (4.31)
2! 3! 4! 5!
1 1 1 1
ei = 1 + i + (i)2 + (i)3 + (i)4 + (i)5 + . . . , (4.32)
2! 3! 4! 5!
1 2 1 3 1 4 1 5
= 1 + i i + + i + . . . (4.33)
2! 3! 4! 5!
As the two series are identical, we have Eulers formula

ei = cos + i sin . (4.34)

4.2.2 Polar and Cartesian representations

Now if we take x and y to be real numbers and define the complex number z to be

z = x + iy, (4.35)
p
we can multiply and divide by x2 + y 2 to obtain
!
p x y
z= x2 + y 2 p + ip . (4.36)
x2 + y 2 x2 + y 2

Noting the similarities between this and the transformation between Cartesian and polar
coordinates suggests we adopt
p x y
r = x2 + y 2 , cos = p , sin = p . (4.37)
x2 + y 2 x2 + y 2
2
Brook Taylor, 1685-1731, English mathematician and artist, Cambridge educated, published on capillary
action, magnetism, and thermometers, adjudicated the dispute between Newton and Leibniz over priority
in developing calculus, contributed to the method of finite differences, invented integration by parts, name
ascribed to Taylor series of which variants were earlier discovered by Gregory, Newton, Leibniz, Johann
Bernoulli, and de Moivre.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

254 CHAPTER 4. POTENTIAL FLOW

iy

y
2
2 +y
x
r=

x x

Figure 4.1: Polar and Cartesian representation of a complex number z.

Thus we have

z = r (cos + i sin ) , (4.38)

z = rei . (4.39)

The polar and Cartesian representation of a complex number z is shown in Figure 4.1.
Now we can define the complex conjugate z as

z = x iy, (4.40)
!
p x y
z = x2 + y 2 p ip , (4.41)
+ x2 y2 x2 + y2
z = r (cos i sin ) , (4.42)
z = r (cos() + i sin()) , (4.43)
z = rei . (4.44)

Note now that

zz = (x + iy)(x iy) = x2 + y 2 = |z|2 , (4.45)

= rei rei = r 2 = |z|2 . (4.46)

We also have

ei ei
sin = , (4.47)
2i
ei + ei
cos = . (4.48)
2

CC BY-NC-ND. 17 June 2016, J. M. Powers.

4.2. MATHEMATICS OF COMPLEX VARIABLES 255

4.2.3 Cauchy-Riemann equations

Now it is possible to define complex functions of complex variables W (z). For example take
a complex function to be defined as
W (z) = z 2 + z, (4.49)
= (x + iy)2 + (x + iy), (4.50)
= x2 + 2xyi y 2 + x + iy, (4.51)


= x2 + x y 2 + i (2xy + y) . (4.52)
In general, we can say
W (z) = (x, y) + i(x, y). (4.53)
Here and are real functions of real variables.
Now W (z) is defined as analytic at zo if dW/dz exists at zo and is independent of the
direction in which it was calculated. That is, using the definition of the derivative

dW  W (zo + z) W (zo )
 = . (4.54)
dz zo z
Now there are many paths that we can choose to evaluate the derivative. Let us consider
two distinct paths, y = C1 and x = C2 . We will get a result which can be shown to be valid
for arbitrary paths. For y = C1 , we have z = x, so
 
dW  W (xo + iyo + x) W (xo + iyo ) W 
= = . (4.55)
dz zo x x y
For x = C2 , we have z = iy, so
  
dW  W (xo + iyo + iy) W (xo + iyo ) 1 W  W 
= = = i . (4.56)
dz zo iy i y x y x
Now for an analytic function, we need
 
W  W 
= i , (4.57)
x y y x
or, expanding, we need
 

+i = i +i , (4.58)
x x y y

= i . (4.59)
y y
For equality, and thus path independence of the derivative, we require

= , = . (4.60)
x y y x

CC BY-NC-ND. 17 June 2016, J. M. Powers.

256 CHAPTER 4. POTENTIAL FLOW

These are the well known Cauchy-Riemann3 equations for analytic functions of com-
plex variables. They are identical to our kinematic equations for incompressible irrotational
fluid mechanics. Consequently, any analytic complex function is guaranteed to be a physical
solution. There are essentially an infinite number of functions to choose from.
Thus we define the complex velocity potential as

W (z) = (x, y) + i(x, y), (4.61)

and taking a derivative of the analytic potential, we have

dW
= +i = u iv. (4.62)
dz x x
We can equivalently say
   
dW
= i +i = i = u iv. (4.63)
dz y y y y

Now most common functions are easily shown to be analytic. For example for the function
W (z) = z 2 + z, which can be expressed as W (z) = (x2 + x y 2) + i(2xy + y), we have

(x, y) = x2 + x y 2 , (x, y) = 2xy + y, (4.64)

= 2x + 1, = 2y, (4.65)
x x

= 2y, = 2x + 1. (4.66)
y y

Note that the Cauchy-Riemann equations are satisfied since /x = /y and /y =

/x. So the derivative is independent of direction, and we can say

dW W 
= = (2x + 1) + i(2y) = 2(x + iy) + 1 = 2z + 1. (4.67)
dz x y

We could get this result by ordinary rules of derivatives for real functions.
For example of a non-analytic function consider W (z) = z. Thus

W (z) = x iy. (4.68)

So = x and = y, /x = 1, /y = 0, and /x = 0, /y = 1. Since

/x 6= /y, the Cauchy-Riemann equations are not satisfied, and the derivative depends
on direction.
3
Augustin-Louis Cauchy, 1789-1857, French mathematician and military engineer, worked in complex
analysis, optics, and theory of elasticity.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

4.3. ELEMENTARY COMPLEX POTENTIALS 257

iy
2



1


0
x

-1

-2
-2 -1 0 1 2

Figure 4.2: Streamlines for uniform flow.

4.3 Elementary complex potentials

Let us examine some simple analytic functions and see the fluid mechanics to which they
correspond.

4.3.1 Uniform flow

Take
W (z) = Az, with A C1 . (4.69)
Then
dW
= A = u iv. (4.70)
dz
Since A is complex, we can say
A = Uo ei = Uo cos iUo sin . (4.71)
Thus we get
u = Uo cos , v = Uo sin . (4.72)
This represents a spatially uniform flow with streamlines inclined at angle to the x axis.
The flow is sketched in Figure 4.2.

4.3.2 Sources and sinks

Take
W (z) = A ln z, with A R1 . (4.73)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

258 CHAPTER 4. POTENTIAL FLOW

With z = rei we have ln z = ln r + i. So

W (z) = A ln r + iA. (4.74)

Consequently, we have for the velocity potential and stream function

= A ln r, = A. (4.75)

Now v = , so
A 1
vr = = , v = = 0. (4.76)
r r r
So the velocity is all radial, and becomes infinite at r = 0. We can show that the volume
flow rate is bounded, and is in fact a constant. The volume flow rate Q through a surface is
Z Z 2 Z 2
T A
Q= v n dA = vr rd = rd = 2A. (4.77)
A 0 0 r

The volume flow rate is a constant. If A > 0, we have a source. If A < 0, we have a sink.
The potential for a source/sink is often written as

Q
W (z) = ln z. (4.78)
2
For a source located at a point zo which is not at the origin, we can say

Q
W (z) = ln(z zo ). (4.79)
2
The flow is sketched in Figure 4.3.

4.3.3 Point vortices

For an ideal point vortex, identical to what we studied in an earlier chapter, we have

W (z) = iB ln z, with B R1 . (4.80)

So
W (z) = iB (ln r + i) = B + iB ln r. (4.81)
Consequently,
= B, = B ln r. (4.82)
We get the velocity field from

1 B
vr = = 0, v = = . (4.83)
r r r

CC BY-NC-ND. 17 June 2016, J. M. Powers.

4.3. ELEMENTARY COMPLEX POTENTIALS 259

iy

Figure 4.3: Velocity vectors and equipotential lines for source flow.

So we see that the streamlines are circles about the origin, and there is no radial component
of velocity. Consider the circulation of this flow
I Z 2
T B
= v dr = rd = 2B. (4.84)
C 0 r

So we often write the complex potential in terms of the ideal vortex strength :
i
W (z) = ln z. (4.85)
2
For an ideal vortex not at z = zo , we say
i
W (z) = ln(z zo ). (4.86)
2
The point vortex flow is sketched in Figure 4.4.

4.3.4 Superposition of sources

Since the equation for velocity potential is linear, we can use the method of superposition
to create new solutions as summations of elementary solutions. Say we want to model the
effect of a wall on a source as sketched in Figure 4.5. At the wall we want u(0, y) = 0. That
is  
dW
= {u iv} = 0, on z = iy. (4.87)
dz
Here denotes the real part of a complex function. Now let us place a source at z = a
and superpose a source at z = a, where a is a real number. So we have for the complex

CC BY-NC-ND. 17 June 2016, J. M. Powers.

260 CHAPTER 4. POTENTIAL FLOW

iy

Figure 4.4: Streamlines, equipotential, and velocity vectors lines for a point vortex.

iy

Q x Q
a a

Figure 4.5: Sketch for source-wall interaction.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

4.3. ELEMENTARY COMPLEX POTENTIALS 261

potential
Q Q
W (z) = ln(z a) + ln(z + a), (4.88)
2
| {z } | 2 {z }
original image
Q
= (ln(z a) + ln(z + a)) , (4.89)
2
Q
= (ln(z a)(z + a)) , (4.90)
2
Q
= ln(z 2 a2 ), (4.91)
2
dW Q 2z
= . (4.92)
dz 2 z 2 a2
Now on z = iy, which is the location of the wall, we have
 
dW Q 2iy
= . (4.93)
dz 2 y 2 a2
The term is purely imaginary; hence, the real part is zero, and we have u = 0 on the wall,
as desired.
On the wall we do have a non-zero y component of velocity. Hence the wall is not a
no-slip wall. On the wall we have then
Q y
v= . (4.94)
y 2 + a2
We find the location on the wall of the maximum v velocity by setting the derivative with
respect to y to be zero,
v Q (y 2 + a2 ) y(2y)
= = 0. (4.95)
y (y 2 + a2 )2
Solving, we find a critical point at y = a, which can be shown to be a maximum.
So on the wall we have
1 2 1 Q2 y2
(u + v 2 ) = . (4.96)
2 2 2 (y 2 + a2 )2
We can use Bernoullis equation to find the pressure field, assuming steady flow and that
p po as r . So Bernoullis equation in this limit
1 p po
()T + = , (4.97)
2
reduces to
1 Q2 y2
p = po 2 . (4.98)
2 (y 2 + a2 )2
Note that the pressure is po at y = 0 and is po as y . By integrating the pressure over
the wall surface, one would find that the source exerted a net force on the wall.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

262 CHAPTER 4. POTENTIAL FLOW

iy iy iy

4
4 4

2 2 2

0
x 0 0 x x

-2 -2 -2

-4 -4 -4

-4 -2 0 2 4 -4 -2 0 2 -4 4 -2 0 2 4

Figure 4.6: Sketch for impingement flow, stagnation flow, and flow in a corner, n = 2.

4.3.5 Flow in corners

Flow in or around a corner can be modeled by the complex potential

W (z) = Az n , with A R1 , (4.99)

n
= A rei , (4.100)
= Ar n ein , (4.101)
= Ar n (cos(n) + i sin(n)). (4.102)

So we have
= Ar n cos n, = Ar n sin n. (4.103)
Now recall that lines on which is constant are streamlines. Examining the stream function,
we obviously have streamlines when = 0 which occurs whenever = 0 or = /n.
For example if n = 2, we model a stream striking a flat wall. For this flow, we have

W (z) = Az 2 , (4.104)
= A(x + iy)2 , (4.105)
= A((x2 y 2) + i(2xy)), (4.106)
= A(x2 y 2 ), = A(2xy). (4.107)

So the streamlines are hyperbolas. For the velocity field, we take

dW
= 2Az = 2A(x + iy) = u iv, (4.108)
dz
u = 2Ax, v = 2Ay. (4.109)

This flow actually represents flow in a corner formed by a right angle or flow striking a flat
plate, or the impingement of two streams. For n = 2, streamlines are sketched in in Figure
4.6.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

4.3. ELEMENTARY COMPLEX POTENTIALS 263

iy

Q -Q
x

source sink
Figure 4.7: Source sink pair.

4.3.6 Doublets
We can form what is known as a doublet flow by considering the superposition of a source
and sink and let the two approach each other. Consider a source and sink of equal and
opposite strength straddling the y axis, each separated from the origin by a distance as
sketched in Figure 4.7. The complex velocity potential is
Q Q
W (z) = ln(z + ) ln(z ), (4.110)
2   2
Q z+
= ln . (4.111)
2 z
It can be shown by synthetic division that as 0, that
z+ 2 2
= 1 + + 2 2 + . . . . (4.112)
z z z
So the potential approaches
 
Q 2 2 2
W (z) ln 1 + + 2 + . . . . (4.113)
2 z z
Now since ln(1 + x) x as x 0, we get for small that
Q 2 Q
W (z) . (4.114)
2 z z
Now if we require that
Q
lim , (4.115)
0
we have
x iy (x iy)
W (z) = = = 2 . (4.116)
z x + iy x iy x + y2
So
x y
(x, y) = , (x, y) = . (4.117)
x2 + y2 x2 + y2

CC BY-NC-ND. 17 June 2016, J. M. Powers.

264 CHAPTER 4. POTENTIAL FLOW

iy

=a =b
4

=a

=b
2

=c
0
x

-2

-4

-4 -2 0 2 4

Figure 4.8: Streamlines and equipotential lines for a doublet.

In polar coordinates, we then say

cos sin
= , = . (4.118)
r r
Streamlines and equipotential lines for a doublet are plotted in Figure 4.8.

4.3.7 Rankine half body

Now consider the superposition of a uniform stream and a source, which we define to be a
Rankine half body:
Q
W (z) = Uz + ln z, with U, Q R1 , (4.119)
2
i Q
= Ure + (ln r + i), (4.120)
2
Q
= Ur(cos + i sin ) + (ln r + i), (4.121)
 2  
Q Q
= Ur cos + ln r + i Ur sin + . (4.122)
2 2
So
Q Q
= Ur cos + ln r, = Ur sin + . (4.123)
2 2

CC BY-NC-ND. 17 June 2016, J. M. Powers.

4.3. ELEMENTARY COMPLEX POTENTIALS 265

iy
4





2

0 x

-2

-4
-4 -2 0 2 4

Figure 4.9: Streamlines for a Rankine half body.

Streamlines for a Rankine half body are plotted in Figure 4.9. Now for the Rankine half
body, it is clear that there is a stagnation point somewhere on the x axis, along = . With
the velocity given by
dW Q
=U+ = u iv, (4.124)
dz 2z
we get
Q 1 i
U+ e = u iv, (4.125)
2 r
Q1
U+ (cos i sin ) = u iv, (4.126)
2 r
Q Q
u=U+ cos , v = sin . (4.127)
2r 2r
When = , we get u = 0 when;
Q
0 = U+ (1), (4.128)
2r
Q
r = . (4.129)
2U

4.3.8 Flow over a cylinder

We can model flow past a cylinder without circulation by superposing a uniform flow with
a doublet. Defining a2 = /U, we write
 
a2
W (z) = Uz + = U z + , (4.130)
z z

CC BY-NC-ND. 17 June 2016, J. M. Powers.

266 CHAPTER 4. POTENTIAL FLOW

iy
2

0 x

-1

-2
-2 -1 0 1 2

Figure 4.10: Streamlines and equipotential lines for flow over a cylinder without circulation.
 
i a2
= U re + i , (4.131)
re
 
a2
= U r(cos + i sin ) + (cos i sin ) , (4.132)
r
 2
  
a a2
= U r cos + cos + i r sin sin , (4.133)
r r
    
a2 a2
= Ur cos 1 + 2 + i sin 1 2 . (4.134)
r r

So    
a2 a2
= Ur cos 1 + 2 , = Ur sin 1 2 . (4.135)
r r
Now on r = a, we have = 0. Since the stream function is constant here, the curve r = a,
a circle, must be a streamline through which no mass can pass.
A sketch of the streamlines and equipotential lines is plotted in Figure 4.10.
For the velocities, we have
   
a2 a2
vr = = U cos 1 + 2 + Ur cos 2 3 , (4.136)
r r r

CC BY-NC-ND. 17 June 2016, J. M. Powers.

4.3. ELEMENTARY COMPLEX POTENTIALS 267

 
a2
= U cos 1 2 , (4.137)
r
 
1 a2
v = = U sin 1 + 2 . (4.138)
r r

So on r = a, we have vr = 0, and v = 2U sin . Thus on the surface, we have

()T = 4U 2 sin2 . (4.139)

Bernoullis equation for a steady flow with p p as r then gives

p 1 p U 2
+ ()T = + , (4.140)
2 2
1
p = p + U 2 (1 4 sin2 ). (4.141)
2
The pressure coefficient Cp , defined below, then is

p p
Cp 1 2
= 1 4 sin2 . (4.142)
2
U

A sketch of the pressure distribution, both predicted and experimentally observed, is

plotted in Figure 4.11. We note that the potential theory predicts the pressure well on
the front surface of the cylinder, but not so well on the back surface. This is because
in most real fluids, a phenomenon known as flow separation manifests itself in regions of
negative pressure gradients. Correct modeling of separation events requires a re-introduction
of viscous stresses. A potential theory cannot predict separation.

Example 4.1
For a cylinder of radius c at rest in an accelerating potential flow field with a far field velocity of
U = a + bt, find the pressure on the stagnation point of the cylinder.
The velocity potential and velocities for this flow are

 
c2
(r, , t) = (a + bt)r cos 1 + 2 , (4.143)
r
 
c2
vr = = (a + bt) cos 1 2 , (4.144)
r r
 
1 c2
v = = (a + bt) sin 1 + 2 , (4.145)
r r
  2  2 !
1 1 c2 2 c2
()T = 2 2
(a + bt) cos 1 2 + sin 1 + 2 , (4.146)
2 2 r r
 
1 c4 2c2

= (a + bt)2 1 + 4 + 2 sin2 cos2 . (4.147)
2 r r

CC BY-NC-ND. 17 June 2016, J. M. Powers.

268 CHAPTER 4. POTENTIAL FLOW

C
p
1

0.5 1 1.5 2 2.5 3

-1 experiment
potential
theory
-2

-3

pressure
distribution
on cylinder
surface from
potential theory

Figure 4.11: Pressure distribution for ideal flow over a cylinder without circulation.

Also, since the flow is unsteady, we will need /t:

 
c2
= br cos 1 + 2 . (4.148)
t r

Now we note in the limit as r that

1 1
br cos , ()T (a + bt)2 . (4.149)
t 2 2
We also note that on the surface of the cylinder

vr (r = c, , t) = 0. (4.150)

Bernoullis equation gives us

1 p
+ ()T + = f (t). (4.151)
t 2

We use the far field behavior to evaluate f (t):

1 p
br cos + (a + bt)2 + = f (t). (4.152)
2
po
Now if we make the non-intuitive choice of f (t) = 12 (a + bt)2 + , we get

1 p 1 po
br cos + (a + bt)2 + = (a + bt)2 + . (4.153)
2 2

CC BY-NC-ND. 17 June 2016, J. M. Powers.

4.4. MORE COMPLEX VARIABLE THEORY 269

So
p = po br cos = po bx. (4.154)

Note that since the flow at infinity is accelerating, there must be a far-field pressure gradient to induce
this acceleration. Consider the x momentum equation in the far field

du p
= , (4.155)
dt x
(b) = (b). (4.156)

So for the pressure field, we have

   
c2 1 c4 2c2 p 1 po
br cos 1 + 2 + (a + bt)2 1 + 4 + 2 (sin2 cos2 ) + = (a + bt)2 + , (4.157)
r 2 r r 2

which reduces to
   4 
c2 1 2 c 2c2 2 2
p(r, , t) = po br cos 1 + 2 (a + bt) + 2 (sin cos ) . (4.158)
r 2 r4 r

For the stagnation point, we evaluate as

1
p(c, , t) = po bc(1) (1 + 1) (a + bt)2 (1 + 2(1)(0 1)) , (4.159)
2
1
= po + (a + bt)2 + 2bc. (4.160)
2

The first two terms would be predicted by a naive extension of the steady Bernoullis equation. The
final term however is not intuitive and is a purely unsteady effect.

4.4 More complex variable theory

There are more basic ways to describe the force on bodies using complex variables directly.
We shall give those methods, but first a discussion of the motivating complex variable theory
is necessary.

4.4.1 Contour integrals

Consider the closed contour integral of a complex function in the complex plane. For such in-
tegrals, we have a useful theory which we will not prove, but will demonstrate here. Consider
contour integrals enclosing the origin with a circle in the complex plane for four functions.
The contour in each is C : z = Rei with 0 2. For such a contour dz = iRei d.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

270 CHAPTER 4. POTENTIAL FLOW

4.4.1.1 Simple pole

We describe a simple pole with the complex potential
a
W (z) = . (4.161)
z
and the contour integral is
I I Z =2
a a
W (z)dz = dz = iRei d, (4.162)
C z Re i
C =0
Z 2
= ai d = 2ia. (4.163)
0

4.4.1.2 Constant potential

We describe a constant with the complex potential

W (z) = b. (4.164)

and the contour integral is

I I Z =2
W (z)dz = bdz = biRei d, (4.165)
C C =0
2
biR i 
= e  = 0. (4.166)
i 
0

since e0i = e2i = 1.

4.4.1.3 Uniform flow

We describe a constant with the complex potential

W (z) = cz. (4.167)

and the contour integral is

I I Z =2
W (z)dz = czdz = cRei iRei d, (4.168)
C C =0
Z 2 2
ic R 2 

= icR2 e2 d = e2i  = 0. (4.169)
0 2i 
0

since e0i = e4i = 1.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

4.4. MORE COMPLEX VARIABLE THEORY 271

4.4.1.4 Quadrupole
A quadrupole potential is described by

k
W (z) = (4.170)
z2
Taking the contour integral, we find
I Z 2
k iRei
dz = k d, (4.171)
C z2 0 R2 e2i
Z 2
ki 2 i ki 1 i 
= e d = e  = 0. (4.172)
R 0 R i 0

So the only non-zero contour integral is for functions of the form W (z) = a/z. We find all
polynomial powers of z have a zero contour integral about the origin for arbitrary contours
except this special one.

4.4.2 Laurent series

Now it can be shown that any function can be expanded, much as for a Taylor series, as a
Laurent series:4

W (z) = . . . + C2 (z zo )2 + C1 (z zo )1 + C0 (z zo )0 + C1 (z zo )1 + C2 (z zo )2 + . . . .
(4.173)
In compact summation notation, we can say
n=
X
W (z) = Cn (z zo )n . (4.174)
n=

Taking the contour integral of both sides we get

I I n=
X
W (z)dz = Cn (z zo )n dz, (4.175)
C C n=
n=
X I
= Cn (z zo )n dz, (4.176)
n= C

this has value 2i only when n = 1, so (4.177)

= C1 2i (4.178)
4
Pierre Alphonse Laurent, 1813-1854, Parisian engineer who worked on port expansion in Le Harve,
submitted his work on Laurent series for a Grand Prize in 1842, with the recommendation of Cauchy, but
was rejected because of a late submission.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

272 CHAPTER 4. POTENTIAL FLOW

Here C1 is known as the residue of the Laurent series. In general we have the Cauchy
integral theorem which holds that if W (z) is analytic within and on a closed curve C except
for a finite number of singular points, then
I X
W (z)dz = 2i residues. (4.179)
C

The constants Cn can be shown to be found by evaluating the contour integral

I
1 W (z)
Cn = dz, (4.180)
2i C (z zo )n+1

where C is any closed contour which has zo in its interior.

4.5 Pressure distribution for steady flow

For steady, irrotational, incompressible flow with no body force present, we have the Bernoulli
equation:
p 1 p 1 2
+ ()T = + U . (4.181)
2 2
We can write this in terms of the complex potential in a simple fashion. First, recall that

()T = u2 + v 2 . (4.182)

We also have dW/dz = u iv, so dW/dz = u + iv. Consequently,

dW dW
= u2 + v 2 = ()T . (4.183)
dz dz
So we get the pressure field from Bernoullis equation to be
 
1 2 dW dW
p = p + U . (4.184)
2 dz dz

The pressure coefficient Cp is

p p 1 dW dW
Cp = 1 2
=1 2 . (4.185)
2
U U dz dz

4.6 Blasius force theorem

For steady flows, we can find the net contribution of a pressure force on an arbitrary shaped
solid body with the Blasius5 force theorem.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

4.6. BLASIUS FORCE THEOREM 273

iy

Sb

Figure 4.12: Potential flow about arbitrarily shaped two-dimensional body with fluid control
volume indicated.

Consider the geometry sketched in Figure 4.12. The surface of the arbitrarily shaped
body is described by Sb , and C is a closed contour containing Sb . First consider the linear
momenta equation for steady flow, no body forces, and no viscous forces,


vT v = p, add mass to get conservative form, (4.186)

T
T (vvT ) = p, integrate over V , (4.187)
Z Z

T
T (vvT ) dV = p dV, use Gauss, (4.188)
V Z ZV
v(vT n)dS = pndS. (4.189)
S S

Now the surface integral here is really a line integral with unit depth b, dS = bds. Moreover
the surface enclosing the fluid has an inner contour Sb and an outer contour C. Now on C,
which we prescribe, we will know x(s) and y(s), where s is arc length. So on C we also get
the unit tangent and unit outward normal n:
 dx   dy 
= dy ,ds n= ds , on C. (4.190)
dx
ds
ds

Moreover, on Sb we have, since it is a solid surface

vT n = 0, on C. (4.191)
5
Paul Richard Heinrich Blasius, 1883-1970, student of Ludwig Prandtl and long time teacher at the tech-
nical college of Hamburg whose 1907 Ph.D. thesis gave mathematical description of similarity solution to
the boundary layer problem.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

274 CHAPTER 4. POTENTIAL FLOW

Now let the force on the body due to fluid pressure be F:

Z
pndS = F. (4.192)
Sb

Now return to our linear momentum equation

Z Z
T
vv ndS = pndS, break this up, (4.193)
I IS IS I
T
| {z n} dS +
v v vvT ndS = pndS pndS, (4.194)
Sb C S C
=0 | {z }
b

I =IF
vvT ndS = F pndS. (4.195)
C C

We can break this into x and y components:

I   I
dy dx dy
u u v bds = p bds Fx , (4.196)
C ds ds C ds
| {z }
vT n
I   I
dy dx dx
v u v bds = p bds Fy , (4.197)
C ds ds C ds
| {z }
vT n
solving for Fx and Fy per unit depth, (4.198)
I
Fx
= pdy u2 dy + uvdx, (4.199)
b
IC
Fy
= pdx + v 2 dx uvdy. (4.200)
b C

Now Bernoulli gives us p = K (1/2)(u2 + v 2 ), where K is some constant. So the x force

per unit depth becomes
I
Fx 1
= Kdy + (u2 + v 2 )dy u2 dy + uvdx, (4.201)
b C 2
since the integral over a closed contour of a constant K is zero, (4.202)
I
1
= (u2 + v 2 )dy + uvdx, (4.203)
2
C I
1
= (u2 + v 2 )dy + 2uvdx. (4.204)
2 C
Similarly for the y direction, we get
I
Fy 1
= Kdx (u2 + v 2 )dy + v 2 dx uvdy, (4.205)
b C I 2
1
= (u2 + v 2 )dx 2uvdx. (4.206)
2 C

CC BY-NC-ND. 17 June 2016, J. M. Powers.

4.7. KUTTA-ZHUKOVSKY LIFT THEOREM 275

iy

Uo

Q
x

Figure 4.13: Potential flow about arbitrarily shaped two-dimensional body with distribution
of sources, sinks, vortices, and dipoles.

Now consider the group of terms Fx iF

b
y
:
I
Fx iFy 1
= (u2 + v 2 )dy + 2uvdx (u2 + v 2 )idx + 2uvidy, (4.207)
b 2 C
I
1
= (i(u2 v 2 ) + 2uv)dx + ((u2 + v 2 ) + 2uvi)dy, (4.208)
2 C
I
1
= (i(u2 v 2 ) + 2uv)dx + (i(u2 v 2 ) + 2uv)idy, (4.209)
2 C
I
1
= (i(u2 v 2 ) + 2uv)(dx + idy), (4.210)
2 C
I
1
= i(u iv)2 (dx + idy), (4.211)
2 C
I  2
1 dW
= i dz. (4.212)
2 C dz

So if we have the complex potential, we can easily get the force on a body.

4.7 Kutta-Zhukovsky lift theorem

Consider the geometry sketched in Figure 4.13. Here we consider a flow with a freestream
constant velocity of Uo . We take an arbitrary body shape to enclose a distribution of cancel-
ing source sink pairs, doublets, point vortices, quadruples, and any other non-mass adding
potential flow term. This combination gives rise to some surface which is a streamline.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

276 CHAPTER 4. POTENTIAL FLOW

Now far from the body surface a contour sees all of these features as effectively concen-
trated at the origin. Then, the potential can be written as

Q Q i
W (z) Uz
|{z} + ln z ln z + ln z + +... (4.213)
uniform flow |2 {z 2 } |2{z } z
|{z}
canceling source sink pair clockwise! vortex doublet

Note that the sign convention for has been violated here, by tradition. Now let us take
D to be the so-called drag force per unit depth and L to be the so-called lift force per unit
depth, so in terms of Fx and Fy , we have

Fx Fy
= D, = L. (4.214)
b b
Now by the Blasius force theorem, we have
I  2
1 dW
D iL = i dz, (4.215)
2 C dz
I  2
1 i
= i U+ + . . . dz, (4.216)
2 C 2z z 2
I   2  
1 2 iU 1
= i U + 2 + 2U + . . . dz. (4.217)
2 C z z 4 2
P
Now the Cauchy integral theorem gives is the contour integral is 2i residues. Here the
residue is iU/. So we get
  
1 iU
D iL = i 2i , (4.218)
2
= iU. (4.219)

So we see that

D = 0, (4.220)
L = U. (4.221)

Note that

is associated with clockwise circulation here. This is something of a tradition in

aerodynamics.

Since for airfoils U, we get the lift force L U 2 ,

For steady inviscid flow, there is no drag. Consideration of either unsteady or viscous
effects would lead to a non-zero x component of force.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

4.7. KUTTA-ZHUKOVSKY LIFT THEOREM 277

Example 4.2
Consider the flow over a cylinder of radius a with clockwise circulation . To do so, we can
superpose a point vortex onto the potential for flow over a cylinder in the following fashion:
 
a2 i  z 
W (z) = U z + + ln . (4.222)
z 2 a
Breaking this up as before into real and complex parts, we get
     
a2 a2 i   r  
W (z) = U r cos 1 + 2 + i U r sin 1 2 + ln + i . (4.223)
r r 2 a
So, we find
  r
a2
= (W (z)) = U r sin 1 2 + ln . (4.224)
r 2 a
On r = a, we find that = 0, so the addition of the circulation in the way we have proposed maintains
the cylinder surface to be a streamline. It is important to note that this is valid for arbitrary . That
is the potential flow solution for flow over a cylinder is non-unique. In aerodynamics, this is used to
advantage to add just enough circulation to enforce the so-called Kutta condition.6 The Kutta condition
is an experimentally observed fact that for a steady flow, the trailing edge of an airfoil is a stagnation
point.
The Kutta-Zhukovsky7 lift theorem tells us whenever we add circulation, that a lift force L = U
is induced. This is consistent with the phenomena observed in baseball that the fastball rises. The
fastball leaves the pitchers hand traveling towards the batter and rotating towards the pitcher. The
induced aerodynamic force is opposite to the force of gravity.
Let us get the lift force the hard way and verify the Kutta-Zhukovsky theorem. We can easily get
the velocity field from the velocity potential:
 
a2
= (W (z)) = U r cos 1 + 2 , (4.225)
r 2
   
2a2 a2
vr = = U r cos 3 + U cos 1 + 2 , (4.226)
r r r
 
2a3 a2
vr |r=a = U cos 3 + 1 + 2 = 0, (4.227)
a a
   
1 1 a2
v = = U r sin 1 + 2 , (4.228)
r r r 2
 
a2
v |r=a = U sin 1 + 2 , (4.229)
a 2a

= 2U sin . (4.230)
2a
We get the pressure on the cylinder surface from Bernoullis equation:
1 1
p = p + U 2 ()T , (4.231)
2 2
6
Martin Wilhelm Kutta, 1867-1944, Silesian-born German mechanician, studied at Breslau, taught
mainly at Stuttgart, co-developer of Runge-Kutta method for integrating ordinary differential equations.
7
Nikolai Egorovich Zhukovsky, 1847-1921, Russian applied mathematician and mechanician, father of
Russian aviation, purchased glider from Lilienthal, developed lift theorem independently of Kutta, organized
Central Aerohydrodynamic Institute in 1918.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

278 CHAPTER 4. POTENTIAL FLOW

ce
for
ure
iy

pre al
ss
loc

Figure 4.14: Pressure force on a differential area element of cylindrical surface.

 2
1 2 1
= p + U 2U sin . (4.232)
2 2 2a
Now for a small element of the cylinder at r = a, the surface area is dA = brd = bad. This is sketched
in Figure 4.14. We also note that the x and y forces depend on the orientation of the element, given
by . Elementary trigonometry shows that the elemental x and y forces per depth are
dFx
= p cos ad, (4.233)
b
dFy
= p sin ad. (4.234)
b
So integrating over the entire cylinder, we obtain,
Z 2  2 !
Fx 1 2 1
= p + U 2U sin cos ad, (4.235)
b 0 2 2 2a
Z 2  2 !
Fy 1 1
= p + U 2 2U sin sin ad. (4.236)
b 0 2 2 2a

Integration via computer algebra gives

Fx
= 0, (4.237)
b
Fy
= U . (4.238)
b
This is identical to the result we expect from the Kutta-Zhukovsky lift theorem.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

4.8. CONFORMAL MAPPING 279

4.8 Conformal mapping

Conformal mapping is a technique by which we can render results obtained for simple flows,
such as those over a cylinder, applicable to flows over more complicated geometries. We
will not consider these in any detail here, but the reader should refer to texts on potential
flow for a full explanation. In short, one relies on a coordinate transformation to map the
complicated geometry in an ordinary space into a simple geometry in a warped geometric
space. In the warped space, on can obtain pressure fields in terms of the warped coordinates,
then transform them back into ordinary space to get the actual pressure field.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

280 CHAPTER 4. POTENTIAL FLOW

CC BY-NC-ND. 17 June 2016, J. M. Powers.

Chapter 5

Viscous incompressible laminar flow

see Panton, Chapter 7, 11

see Yih, Chapter 7

Here we consider a few standard problems in viscous incompressible laminar flow. For this
entire chapter, we will make the following assumptions:

the flow is incompressible,

body forces are negligible, and

the fluid properties, c, and k, are constants.

5.1 Fully developed, one dimensional solutions

The first type of solution we will consider is known as a one-dimensional fully developed
solution. These are commonly considered in first courses in fluid mechanics and heat transfer.
The flows here are essentially one-dimensional, but not absolutely, as they were in the chapter
on one-dimensional compressible flow. In this section, we will further enforce that

the flow is time-independent, o = 0,

the velocity and temperature gradients in the x and z direction are zero, v/x = 0,
v/z = 0, T /x = 0, T /z = 0.

We will see that these assumptions give rise to flows with a non-zero x velocity u which
varies in the y direction, and that other velocities v, and w, will be zero.

5.1.1 Pressure gradient driven flow in a slot

Consider the flow sketched in Figure 5.1. Here we have a large reservoir of fluid with a
long narrow slot located around y = 0. We take the length of the slot in the z direction,

281
282 CHAPTER 5. VISCOUS INCOMPRESSIBLE LAMINAR FLOW

b
= constant y
= constant
z

po x
p1
L

Figure 5.1: Pressure gradient driven flow in a slot.

b, to be long relative to the slot width in the y direction h. Attached to the slot are two
parallel plates, separated by distance in the y direction h. The length of the plates in the x
direction is L. We take L >> h. Because of gravity forces, which we neglect in the slot, the
pressure at the entrance of the slot po is higher than atmospheric. At the end of the slot,
the fluid expels to the atmosphere which is at p1 . Hence, there is a pressure gradient in the
x direction, which drives the flow in the slot. We will see that the flow is resisted by viscous
stresses. An analogous flow in a circular duct is defined as a Hagen-Poiseuille12 flow.
Near x = 0, the flow accelerates in what is known as the entrance length. If L is sufficiently
long, we observe that the fluid particles no longer accelerate after traveling in the slot. It is
at this point where the viscous shear forces exactly balance the pressure forces to give rise
to the fully developed velocity field.
For this flow, let us make the additional assumptions that
there is no imposed pressure gradient in the z direction, and
the walls are held at a constant temperature, To .
Incorporating some of these assumptions, we write the incompressible constant property
Navier-Stokes equations as
i vi = 0, (5.1)
o vi + vj j vi = i p + j j vi , (5.2)
co T + cvj j T = ki i T + 2(i vj) (i vj) . (5.3)
1
Gotthilf Ludwig Hagen, 1797-1884, German engineer who measured velocity of water in small diameter
tubes.
2
Jean Louis Poiseuille, 1799-1869, French physician who repeated experiments of Hagen for simulated
blood flow.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

5.1. FULLY DEVELOPED, ONE DIMENSIONAL SOLUTIONS 283

Here we have five equations in five unknowns, vi , p, and T .

As for all incompressible flows with constant properties, we can get the velocity field by
only considering the mass and momenta equations; velocity is only coupled one way to the
energy equation.
The mass equation, recalling that gradients in x and z are zero, gives us

u+ v+ w = 0. (5.4)
x
|{z} y z
|{z}
=0 =0

So the mass equation gives us

v
= 0. (5.5)
y
Now, from our assumptions of steady and fully developed flow, we know that v cannot be a
function of x, z, or t. So the partial becomes a total derivative, and mass conservation holds
that dv/dy = 0. Integrating, we find that v(y) = C. The constant C must be zero, since we
must satisfy a no-slip boundary condition at either wall that v(y = h/2) = v(y = h/2) = 0.
Hence, mass conservation, coupled with the no slip boundary condition gives us

v = 0. (5.6)

Now consider the x momentum equation:

p 2 2 2
u + u u + |{z}
v u + w u = + 2 u + 2 u + 2 u ,
t
|{z} x
|{z} y z
|{z} x x
|{z} y z
|{z}
=0
=0 =0 =0 =0 =0
(5.7)
2
p u
0 = + 2. (5.8)
x y
We note for this fully developed flow that the acceleration, that is the material derivative
of velocity, is formally zero, and the equation gives rise to a balance of pressure and viscous
surface forces.
For the y momentum equation, we get
p
v +u
|{z} v + |{z}
|{z} v v +w
|{z} v
|{z} = (5.9)
t
|{z} x
|{z} y z
|{z} y
=0 =0 =0 =0 =0
=0 =0 =0

2 2 2
+ 2 v + 2 |{z}
v + 2 v ,
x
|{z} y z
|{z}
=0
=0 =0
p
0 = . (5.10)
y

CC BY-NC-ND. 17 June 2016, J. M. Powers.

284 CHAPTER 5. VISCOUS INCOMPRESSIBLE LAMINAR FLOW

Hence, p = p(x, z), but since we have assumed there is no pressure gradient in the z direction,
we have at most that
p = p(x). (5.11)
For the z momentum equation we get:

p
w + u w + |{z}
v w + w w = (5.12)
t
|{z} x
|{z} y z
|{z} z
|{z}
=0
=0 =0 =0 =0

2 2 2
+ 2 w + 2 w + 2 w ,
x
|{z} y z
|{z}
=0 =0
2w
0 = . (5.13)
y 2

Solution of this partial differential equation gives us

w = f (x, z)y + g(x, z). (5.14)

Now to satisfy no-slip, we must have w = 0 at y = h/2. This leads us to two linear
equations for f and g:  h    
2
1 f (x, z) 0
h = . (5.15)
2 1 g(x, z) 0
Since the determinant of the coefficient matrix, h/2 + h/2 = h, is non-zero, the only solution
is the trivial solution f (x, z) = g(x, z) = 0. Hence,

w = 0. (5.16)

Next consider how the energy equation reduces:

2 2 2
c T + c u T + |{z}
v T + |{z}
w T = k 2 T + 2T + 2 T
t
|{z} x
|{z} y z x y z
=0 =0 |{z} |{z} |{z}
=0 =0 =0 =0 =0
+2(i vj) (i vj) , (5.17)
2T
0 = k 2 + 2(i vj) (i vj) . (5.18)
y

Note that there is no tendency for a particles temperature to increase. There is a balance
between thermal energy generated by viscous dissipation and that conducted away by ther-
mal diffusion. Thus the energy path is 1) viscous work is done to generate thermal energy,
2) thermal energy diffuses throughout the channel and out the boundary. Now consider the

CC BY-NC-ND. 17 June 2016, J. M. Powers.

5.1. FULLY DEVELOPED, ONE DIMENSIONAL SOLUTIONS 285

viscous dissipation term for this flow.

1 v1 1 v2 1 v3
|{z} |{z} |{z} |{z} |{z}
=0 =0
=0 =0 =0 0 0 0
v 2 v2 2 v3
i vj = 2 1 |{z} |{z} = 2 v1 0 0 , (5.19)
=0 =0
v v v 0 0 0
3 1 3 2 3 3
|{z} |{z} |{z} |{z} |{z}

=0 =0 =0
=0 =0
0 1 2 v1 + 1 v2 0
2 |{z} 0 1 u
0

=0 2 y
1 u
(i vj) = = 2 y 0 0 . (5.20)
1
2 2 v1 + 1 v2 0 0 0 0 0
|{z}
=0
0 0 0
Further,  2  2  2
1 u 1 u 1 u
(i vj) (i vj) = + = . (5.21)
2 y 2 y 2 y
So the energy equation becomes finally
 2
2T u
0=k 2 + . (5.22)
y y
At this point we have the x momentum and energy equations as the only two which seem
to have any substance.
p 2u
0 = + 2, (5.23)
x y
2
 2
T u
0 = k 2 + . (5.24)
y y
This looks like two equations in three unknowns. One peculiarity of incompressible equations
is that there is always some side condition, which ultimately hinges on the mass equation,
which really gives a third equation. Without going into details, it involves for general flows
solving a Poisson3 equation for pressure which is of the form 2 p = f (u, v). Note that this
involves second derivatives of pressure. Here we can obtain a simple form of this general
equation by taking the partial derivative with respect to x of the x momentum equation:
2p 2u
0 = + , (5.25)
x2 x y 2
2p 2 u
0 = 2+ 2 . (5.26)
x y |{z}
x
=0
3
Simeon Denis Poisson, 1781-1840, French mathematician taught by Laplace, Lagrange, and Legendre,
studied partial differential equations, potential theory, elasticity, and electrodynamics.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

286 CHAPTER 5. VISCOUS INCOMPRESSIBLE LAMINAR FLOW

The viscous term above is zero because of our assumption of fully developed flow. Moreover,
since p = p(x) only, we then get

d2 p
= 0, p(0) = po , p(L) = p1 , (5.27)
dx2
which has a solution showing the pressure field must be linear in x:
po p1
p(x) = po x, (5.28)
L
dp po p1
= . (5.29)
dx L
Now, since u is at most a function of y, we can convert partial derivatives to ordinary deriva-
tives, and write the x momentum equation and energy equation as two ordinary differential
equations in two unknowns with appropriate boundary conditions at the wall y = h/2:
   
d2 u po p1 h h
= , u = 0, u = 0, (5.30)
dy 2 L 2 2
 2    
d2 T du h h
= , T = To , T = To . (5.31)
dy 2 k dy 2 2

We could solve these equations directly, but instead let us first cast them in dimensionless
form. This will give our results some universality and efficiency. Moreover, it will reveal more
fundamental groups of terms which govern the fluid behavior. Let us select scales such that
dimensionless variables, denoted by a * subscript, are as follows
y T To u
y = , T = , u = . (5.32)
h To uc
We have yet to determine the characteristic velocity uc . Note that the dimensionless tem-
perature has been chosen to render it zero at the boundaries. With these choices, the x
momentum equation becomes

u c d2 u po p1
= , (5.33)
h2 dy2 L
d2 u (po p1 )h2
= , (5.34)
dy2 Luc
uc u (x h = h/2) = uc u (x h = h/2) = 0, (5.35)
u (x = 1/2) = u (x = 1/2) = 0. (5.36)

Let us now choose the characteristic velocity to render the x momentum equation to have a
simple form:
(po p1 )h2
uc . (5.37)
L

CC BY-NC-ND. 17 June 2016, J. M. Powers.

5.1. FULLY DEVELOPED, ONE DIMENSIONAL SOLUTIONS 287

Now scale the energy equation:

 2
To d2 T u2c du
= , (5.38)
h2 dy2 kh2 dy
2 
d2 T u2c
du
= , (5.39)
dy2 kTo
dy
 2
c u2c du
= , (5.40)
k cTo dy
 2
du
= P rEc , (5.41)
dy
   
1 1
T = T = 0. (5.42)
2 2

Here we have grouped terms so that the Prandtl number P r = c/k, explicitly appears.
Further, we have defined the Eckert4 number Ec as
 2
(po p1 )h2
u2c L
Ec = = . (5.43)
cTo cTo

In summary our dimensionless differential equations and boundary conditions are

 
d2 u 1
2
= 1, u = 0, (5.44)
dy 2
 2  
d2 T du 1
2
= P rEc , T = 0. (5.45)
dy dy 2

These boundary conditions are homogeneous; hence, they do not contribute to a non-trivial
solution. The pressure gradient is an inhomogeneous forcing term in the momentum equa-
tion, and the viscous dissipation is a forcing term in the energy equation.
The solution for the velocity field which satisfies the differential equation and boundary
conditions is quadratic in y and is
 2 !
1 1
u = y2 . (5.46)
2 2

Note that the maximum velocity occurs at y = 0 and has value

1
umax = . (5.47)
8
4
Ernst R. G. Eckert, 1904-2004, scholar of convective heat transfer.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

288 CHAPTER 5. VISCOUS INCOMPRESSIBLE LAMINAR FLOW

y = 1/2
*

u
*

y = -1/2
*

Figure 5.2: Velocity profile for pressure gradient driven flow in a slot.

The mean velocity is found through integrating the velocity field to arrive at
Z 1/2
umean = u (y )dy , (5.48)
1/2
Z  2 !
1/2
1
1 2
= y dy , (5.49)
1/2 2
2
 1/2
1 1 1 3 
= y y  , (5.50)
2 4 3 1/2
1
= . (5.51)
12
Note that we could have scaled the velocity field in such a fashion that either the maximum
or the mean velocity was unity. The scaling we chose gave rise to a non-unity value of both.
In dimensional terms we could say
 2   !
u 1 1 y 2
(po p1 )h 2 = . (5.52)
2 2 h
L

The velocity profile is sketched in Figure 5.2.

Now let us get the temperature field.
 2 !!!2
d2 T d 1 1 2
= P rEc y , (5.53)
dy2 dy 2 2
= P rEc (y )2 , (5.54)
= P rEc y2 , (5.55)
dT 1
= P rEc y3 + C1 , (5.56)
dy 3

CC BY-NC-ND. 17 June 2016, J. M. Powers.

5.1. FULLY DEVELOPED, ONE DIMENSIONAL SOLUTIONS 289

y
*

1/2

T
*

-1/2

Figure 5.3: Temperature profile for pressure gradient driven flow in a slot.

1
T = P rEc y4 + C1 y + C2 , (5.57)
12
1 1 1 1
0 = P rEc + C1 + C2 , y = , (5.58)
12 16 2 2
1 1 1 1
0 = P rEc C1 + C2 , y = , (5.59)
12 16 2 2
P rEc
C1 = 0, C2 = . (5.60)
192
Regrouping, we find that !
 4
P rEc 1 4
T = y . (5.61)
12 2
In terms of dimensional quantities, we can say

2 4
 4   !
T To (po p1 ) h 1 y 4
= . (5.62)
To 12L2 kTo 2 h

The temperature profile is sketched in Figure 5.3.

From knowledge of the velocity and temperature field, we can calculate other quantities
of interest. Let us calculate the field of shear stress and heat flux, and then evaluate both
at the wall.
First for the shear stress, recall that in dimensional form we have

ij = 2(i vj) + k vk ij , (5.63)

|{z}
=0
= 2(i vj) . (5.64)

We have already seen the only non-zero components of the symmetric part of the velocity

CC BY-NC-ND. 17 June 2016, J. M. Powers.

290 CHAPTER 5. VISCOUS INCOMPRESSIBLE LAMINAR FLOW

gradient tensor are the 12 and 21 components. Thus the 21 stress component is
v + v
2 1 1 2
|{z}
=0
21 = 2(2 v1) = 2 , (5.65)
2

= 2 v1 . (5.66)

In (x, y) space, we then say here that

du
yx = . (5.67)
dy

Note this is a stress on the y (tangential) face which points in the x direction; hence, it
is certainly a shearing stress. In dimensionless terms, we can define a characteristic shear
stress c , so that the scale shear is = yx /c . Thus, our equation for shear becomes

uc du
c = . (5.68)
h dy

Now take  
uc (po p1 )h2 h
c = = (po p1 ) . (5.69)
h hL L
With this definition, we get
du
= . (5.70)
dy
Evaluating for the velocity profile of the pressure gradient driven flow, we find

= y . (5.71)

The stress is zero at the centerline y = 0 and has maximum magnitude of 1/2 at either
wall, y = 1/2. In dimensional terms, the wall shear stress w is
 
1 h
w = (po p1 ) . (5.72)
2 L

Note that the wall shear stress is governed by the pressure difference and not the viscosity.
However, the viscosity plays a determining role in selecting the maximum fluid velocity. The
shear profile is sketched in Figure 5.4.
Next, let us calculate the heat flux vector. Recall that, for this flow, with no x or z
variation of T , we have the heat flux vector as

T
qy = k . (5.73)
y

CC BY-NC-ND. 17 June 2016, J. M. Powers.

5.1. FULLY DEVELOPED, ONE DIMENSIONAL SOLUTIONS 291

y = 1/2
*

y = -1/2

Figure 5.4: Shear stress profile for pressure gradient driven flow in a slot.

Now define scale the heat flux by a characteristic heat flux qc , to be determined, to obtain
a dimensionless heat flux:
qy
q = . (5.74)
qc
So,

kTo dT
qc q = , (5.75)
h dy
kTo dT
q = . (5.76)
hqc dt

Let qc kTo /h, so

dT
q = , (5.77)
dy
1
q = P rEc y3. (5.78)
3
For our flow, we have a cubic variation of the heat flux vector. There is no heat flux at the
centerline, which corresponds to this being a region of no shear. The magnitude of the heat
flux is maximum at the wall, the region of maximum shear. At the upper wall, we have

1
q |y =1/2 = P rEc. (5.79)
24
The heat flux profile is sketched in Figure 5.5. In dimensional terms we have

qw 1 (po p1 )2 h4
kTo
= , (5.80)
h
24 L2 kTo
1 (po p1 )2 h3
qw = . (5.81)
24 L2

CC BY-NC-ND. 17 June 2016, J. M. Powers.

292 CHAPTER 5. VISCOUS INCOMPRESSIBLE LAMINAR FLOW

y = 1/2
*

q
*

y = -1/2

Figure 5.5: Heat flux profile for pressure gradient driven flow in a slot.
L

y = h, u = U U

p=p p=p
o 1

x y = 0, u = 0

Figure 5.6: Configuration for Couette flow with pressure gradient.

5.1.2 Couette flow with pressure gradient

We next consider Couette flow with a pressure gradient. Couette flow implies that there is a
moving plate at one boundary and a fixed plate at the other. It is a common experimental
configuration, and used often to actually determine a fluids viscosity. Here we will take the
same assumptions as for pressure gradient driven flow in a slot, expect for the boundary
condition at the upper surface, which we will require to have a constant velocity U. We will
also shift the coordinates so that y = 0 matches the lower plate surface and y = h matches
the upper plate surface. The configuration for this flow is shown in Figure 5.6.
Our equations governing this flow are

d2 u po p1
2
= , u(0) = 0, u(h) = U, (5.82)
dy L
 2
d2 T du
= , T (0) = To , T (h) = To . (5.83)
dy 2 k dy

Once again in momentum, there is no acceleration, and viscous stresses balance shear stresses.
In energy, there is no energy increase, and generation of thermal energy due to viscous work
is balanced by diffusion of the thermal energy, ultimately out of the system through the

CC BY-NC-ND. 17 June 2016, J. M. Powers.

5.1. FULLY DEVELOPED, ONE DIMENSIONAL SOLUTIONS 293

boundaries. Here there are inhomogeneities in both the forcing terms and the boundary
conditions. In terms of work, both the pressure gradient and the pulling of the plate induce
work.
Once again let us scale the equations. This time, we have a natural velocity scale, U, the
upper plate velocity. So take
y T To u
y = , T = , u = . (5.84)
h To U
The momentum equation becomes

U d2 u po p1
2 2
= , (5.85)
h dy L
d2 u (po p1 )h2
= , (5.86)
dy2 UL
(po p1 )h2
with dimensionless pressure gradient P , (5.87)
UL
d2 u
= P, (5.88)
dy2
u (0) = 0, u (1) = 1. (5.89)
(5.90)

This has solution

1
u = Py2 + C1 y + C2 . (5.91)
2
Applying the boundary conditions, we get
1
0 = P(0)2 + C1 (0) + C2 , (5.92)
2
0 = C2 , (5.93)
1
1 = P(1)2 + C1 (1), (5.94)
2
1
C1 = 1 + P, (5.95)
2  
1 2 1
u = Py + 1 + P y , (5.96)
2 2
1
u = Py (1 y ) + y . (5.97)
|2 {z } |{z}
Couette effect
pressure effect

We see that the pressure gradient generates a velocity profile that is quadratic in y . This is
distinguished from the Couette effect, that is the effect of the upper plates motion, which
gives a linear profile. Because our governing equation here is linear, it is appropriate to

CC BY-NC-ND. 17 June 2016, J. M. Powers.

294 CHAPTER 5. VISCOUS INCOMPRESSIBLE LAMINAR FLOW

y=1 u=1
* *

-6

-2
0
2
P=

8
- P

P=

P=
P=
8
P

y =0 u=0
* *

Figure 5.7: Velocity profiles for various values of P for Couette flow with pressure gradient.

think of these as superposed solutions. Velocity profiles for various values of P are shown in
Figure 5.7.
Let us now calculate the shear stress profile. With = (du/dy), and taking = /c ,
we get

U du
c = , (5.98)
h dy
U du
= , (5.99)
hc dy
U
taking c , (5.100)
h
du
= , so here, (5.101)
dy
1
= Py + P + 1, and (5.102)
2
1
|y =0 = P + 1, (5.103)
2
1
|y =1 = P + 1. (5.104)
2

The wall shear has a pressure gradient effect and a Couette effect as well. In fact we can
select a pressure gradient to balance the Couette effect at one or the other wall, but not
both.
We can also calculate the dimensionless volume flow rate Q , which for incompressible
flow, is directly proportional to the mass flux. Ignoring how the scaling would be done, we
arrive at
Z 1
Q = u dy , (5.105)
0
Z 1   
1 2 1
= Py + 1 + P y dy , (5.106)
0 2 2

CC BY-NC-ND. 17 June 2016, J. M. Powers.

5.1. FULLY DEVELOPED, ONE DIMENSIONAL SOLUTIONS 295

 1   2 1
1 3 1 y 
= Py + 1 + P , (5.107)
6 0 2 2 0
 
P 1 1
= + 1+ P , (5.108)
6 2 2
P 1
= + . (5.109)
12 2
Again there is a pressure gradient contribution and a Couette contribution, and we could
select P to give no net volume flow rate.
We can summarize some of the special cases as follows

P : u = (1/2)Py (1 y ); = P (1/2 y ), Q = P/12. Here the fluid flows

in the opposite direction as driven by the plate because of the large pressure gradient.

P = 6. Here we get no net mass flow and u = 3y2 2y , = 2y , Q = 0.

P = 2. Here we get no shear at the bottom wall and u = y2 , = 2y , Q = 1/3.

P = 0. Here we have no pressure gradient and u = y , = 1, Q = 1/2.

P = 2. Here we get no shear at the top wall and u = y2 + 2y , = 2y + 2,

Q = 2/3.

P : u = (1/2)Py (1 y ); = P (1/2 y ), Q = P/12. Here the fluid flows

in the same direction as driven by the plate.

We now consider the heat transfer problem. Scaling, we get

 2
To d2 T U 2 du
= , T (0) = T (1) = 0, (5.110)
h2 dy2 kh2 dy
 2
d2 T U 2 du
= , (5.111)
dy2 kTo dy
 2
c U 2 du
= , (5.112)
k cTo dy
 2
du
= P rEc , (5.113)
dy
= P rEc 2 , (5.114)
 2
1
= P rEc Py + P + 1 , (5.115)
2
   2 !
1 1
= P rEc P 2 y2 2P P + 1 y + 1 + P , (5.116)
2 2

CC BY-NC-ND. 17 June 2016, J. M. Powers.

296 CHAPTER 5. VISCOUS INCOMPRESSIBLE LAMINAR FLOW

   2 !
dT P2 3 1 1
= P rEc y P P + 1 y2 + 1 + P y + C1 , (5.117)
dy 3 2 2
   2 !
P2 4 P 1 1 1
T = P rEc y P + 1 y3 + 1 + P y2
12 3 2 2 2
+C1 y + C2 , (5.118)
T (0) = 0 = C2 , (5.119)
   2 !
P2 P 1 1 1
T (1) = 0 = P rEc P +1 + 1+ P + C1 , (5.120)
12 3 2 2 2
 
1 P P2
C1 = P rEc + + , (5.121)
2 6 24
   2 !
P2 4 P 1 1 1
T = P rEc y P + 1 y3 + 1 + P y2
12 3 2 2 2
 
1 P P2
+P rEc + + y . (5.122)
2 6 24

Factoring, we can write the temperature profile as

P rEc
T = y (1 y )(12 + 4P + P 2 8Py 2P 2 y + 2P 2 y2 ). (5.123)
24
For the wall heat transfer, recall qy = k(dT /dy). Scaling, we get

kTo dT
qc q = , (5.124)
h dy
kTo dT
q = , (5.125)
hqc dy
kTo
choosing qc , (5.126)
h
dT
q = . (5.127)
dy

So
   2 !
P2 3 1 1 1 P P 2
q = P rEc y P P + 1 y2 + 1 + P y . (5.128)
3 2 2 2 6 24

At the bottom wall y = 0, we get for the heat transfer vector

 
1 P P2
q |y =0 = P rEc + + . (5.129)
2 6 24

CC BY-NC-ND. 17 June 2016, J. M. Powers.

5.2. SIMILARITY SOLUTIONS 297

Figure 5.8: Schematic for Stokes first problem of a suddenly accelerated plate diffusing
linear momentum into a fluid at rest.

5.2 Similarity solutions

In this section, we will consider problems which can be addressed by what is known as a
similarity transformation. The problems themselves will be fundamental ones which have
variation in either time and one spatial coordinate, or with two spatial coordinates. This is in
contrast with solutions of the previous section which varied only with one spatial coordinate.
Since two coordinates are involved, we must resort to solving partial differential equations.
The similarity transformation actually reveals a hidden symmetry of the partial differential
equations by defining a new independent variable, which is a grouping of the original in-
dependent variables, under which the partial differential equations transform into ordinary
differential equations. We then solve the resulting ordinary differential equations by standard
techniques.

5.2.1 Stokes first problem

The first problem we will consider which uses a similarity transformation is known as Stokes
first problem, as Stokes addressed it in his original work which developed the Navier-Stokes
equations in the mid-nineteenth century.5 The problem is described as follows, and is
sketched in Figure 5.8. Consider a flat plate of infinite extent lying at rest for t < 0 on
the y = 0 plane in x y z space. In the volume described by y > 0 exists a fluid of semi-
infinite extent which is at rest at time t < 0. At t = 0, the flat plate is suddenly accelerated to
a constant velocity of U, entirely in the x direction. Because the no-slip condition is satisfied
for the viscous flow, this induces the fluid at the plate surface to acquire an instantaneous
velocity of u(0) = U. Because of diffusion of linear x momentum via tangential viscous shear
forces, the fluid in the region above the plate begins to acquire a positive velocity in the x
direction as well. We will use the Navier-Stokes equations to quantify this behavior. Let
5
Stokes, G. G., 1851, On the effect of the internal friction of fluids on the motion of pendulums, Trans-
actions of the Cambridge Philosophical Society, 9(2): 8-106.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

298 CHAPTER 5. VISCOUS INCOMPRESSIBLE LAMINAR FLOW

us make identical assumptions as we did in the previous section, except that 1) we will not
neglect time derivatives, as they are an obviously important feature of the flow, and 2) we
will assume all pressure gradients are zero; hence the fluid has a constant pressure.
Under these assumptions, the x momentum equation,

p 2 2 2
u + u u + |{z}
v u + w u = + 2 u + 2 u + 2 u ,
t x
|{z} y z
|{z} x
|{z} x
|{z} y z
|{z}
=0
=0 =0 =0 =0 =0
(5.130)
is the only relevant component of linear momenta, and reduces to
u 2u
= 2 . (5.131)
t
|{z} y
| {z }
(mass)(acceleration) shear force

The energy equation reduces as follows

2 2 2
c T + c u T + |{z}
v T + |{z}
w T = k 2 T + 2T + 2 T
t x
|{z} y z x y z
=0 =0 |{z} |{z} |{z}
=0 =0 =0 =0
+2(i vj) (i vj) , (5.132)
 2
T 2T u
c = k 2 + (5.133)
.
| {zt} y
| {z }
y
| {z }
energy increase thermal diffusion viscous work source

Let us first consider the x momentum equation. Recalling the momentum diffusivity
definition = /, we get the following partial differential equation, initial and boundary
conditions:
u 2u
= 2, (5.134)
t y
u(y, 0) = 0, u(0, t) = U, u(, t) = 0. (5.135)
Now let us scale the equations. Choose
u t y
u = , t = , y = . (5.136)
U tc yc
We have yet to choose characteristic length, (yc ), and time, (tc ), scales. The equations
become
U u U 2 u
= , (5.137)
tc t yc2 y2
u tc 2 u
= . (5.138)
t yc2 y2

CC BY-NC-ND. 17 June 2016, J. M. Powers.

5.2. SIMILARITY SOLUTIONS 299

Wasting no time, we choose

yc = . (5.139)
U U
3 kg m s m3 s
Noting the SI units, we see /(U) has units of length: Nm2s m
kg
s
m
= s2 m2 kg m
= m. With
this choice, we get
tc tc U 2 tc U 2
= = . (5.140)
yc2 2
This suggests we choose

.
tc = (5.141)
U2
With all of these choices the complete system can be written as

u 2 u
= , (5.142)
t y2
u (y , 0) = 0, u (0, t ) = 1, u (, t ) = 0. (5.143)

Now for self-similarity, we seek transformation which reduce this partial differential equation,
as well as its initial and boundary conditions, into an ordinary differential equation with
suitable boundary conditions. If this transformation does not exist, no similarity solution
exists. In this, but not all cases, the transformation does exist.
Let us first consider a general transformation from a y , t coordinate system to a new
, t coordinate system. We assume then a general transformation

= (y , t ), (5.144)
t = t (y , t ). (5.145)

We assume then that a general variable which is a function of y and t also has the same
value at the transformed point , t :

(y , t ) = ( , t ). (5.146)

The chain rule then gives expressions for derivatives:

    
    t 
= + , (5.147)
t y t t y t  t y
    
    t 
= + . (5.148)
y t  t y 
t t 
y  t

Now we will restrict ourselves to the transformation

t = t , (5.149)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

300 CHAPTER 5. VISCOUS INCOMPRESSIBLE LAMINAR FLOW

 
t  t 
so we have t 
= 1 and y 
= 0, so our rules for differentiation reduce to
y t
   
   
= + , (5.150)
t y t t y t 
  
  
= . (5.151)
y  t  t y  t

The next assumption is key for a similarity solution to exist. We restrict ourselves to
transformations for which  = ( ). That is we allow no dependence of on t . Hence

we must require that
t
 = 0. Moreover, partial derivatives of become total derivatives,

giving us a final form of transformations for the derivatives
 
 d 
= , (5.152)
t y d t y
 
 d 
= . (5.153)
y t d y t

In terms of operators we can say

 
  d
= , (5.154)
t y t y d
 
  d
= . (5.155)
y 
t y  d
t

Now returning to Stokes first problem, let us assume that a similarity solution exists of
the form u (y , t ) = u( ). It is not always possible to find a similarity variable . One
of the more robust ways to find a similarity variable, if it exists, comes from group theory,6
6
Group theory has a long history in mathematics and physics. Its complicated origins generally include
attribution to Evariste Galois, 1811-1832, a somewhat romantic figure, as well as Niels Henrick Abel, 1802-
1829, the Norwegian mathematician. Critical developments were formalized by Marius Sophus Lie, 1842-
1899, another Norwegian mathematician, in what today is known as Lie group theory. A modern variant,
known as renormalization group (RNG) theory is an area for active research. The 1982 Nobel prize in
physics went to Kenneth Geddes Wilson, 1936-, of Cornell University and The Ohio State University, for use
of RNG in studying phase transitions, first done in the 1970s. The award citation refers to the possibilities
of using RNG in studying the great unsolved problem of turbulence, a modern area of research in which
Steven Alan Orszag, 1943-2011, made many contributions.
Quoting from the useful Eric Weissteins World of Mathematics, available online at
http://mathworld.wolfram.com/Group.html, A group G is a finite or infinite set of elements together
with a binary operation which together satisfy the four fundamental properties of closure, associativity, the
identity property, and the inverse property. The operation with respect to which a group is defined is often
called the group operation, and a set is said to be a group under this operation. Elements A, B, C, . . .
with binary operations A and B denoted AB form a group if
1. Closure: If A and B are two elements in G, then the product AB is also in G.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

5.2. SIMILARITY SOLUTIONS 301

and is explained in detail in the recent monograph by Cantwell. Group theory, which is too
detailed to explicate in full here, relies on a generalized symmetry of equations to find simpler
forms. In the same sense that a snowflake, subjected to rotations of /3, 2/3, , 4/3,
5/3, or 2, is transformed into a form which is indistinguishable from its original form,
we seek transformations of the variables in our partial differential equation which map the
equation into a form which is indistinguishable from the original. When systems are subject
to such transformations, known as group operators, they are said to exhibit symmetry.
Let us subject our governing partial differential equation along with initial and boundary
conditions to a particularly simple type of transformation, a simple stretching of space, time,
and velocity:
t = ea t , y = eb y , u = ec u. (5.156)
Here the variables are stretched variables, and a, b, and c are constant parameters. The
exponential will be seen to be a convenience, which is not absolutely necessary. Note that
for a (, ), b (, ), c (, ), that ea (0, ), eb (0, ), ec (0, ).
So the stretching does not change the direction of the variable; that is it is not a reflecting
transformation. We note that with this stretching, the domain of the problem remains
unchanged; that is t [0, ) maps into t [0, ); y [0, ) maps into y [0, ).
The range is also unchanged if we allow u [0, ), which maps into u [0, ). Direct
substitution of the transformation shows that in the stretched space, the system becomes

u 2 u
eac = e2bc 2 , (5.157)
t y
ec u(y, 0) = 0, c
e u(0, t) = 1, ec u(, t) = 0. (5.158)

In order that the stretching transformation map the system into a form indistinguishable
from the original, that is for the transformation to exhibit symmetry, we must take

c = 0, a = 2b. (5.159)

So our symmetry transformation is

t = e2b t , y = eb y , u = u , (5.160)
2. Associativity: The defined multiplication is associative, i.e. for all A, B, C G, (AB)C = A(BC).
3. Identity: There is an identity element I (a.k.a. 1, E, or e) such that IA = AI = A for every element
A G.
4. Inverse: There must be an inverse or reciprocal of each element. Therefore, the set must contain an
element B = A1 such that AA1 = A1 A = I for each element of G.
. . ., A map between two groups which preserves the identity and the group operation is called a homomor-
phism. If a homomorphism has an inverse which is also a homomorphism, then it is called an isomorphism
and the two groups are called isomorphic. Two groups which are isomorphic to each other are considered to
be the same when viewed as abstract groups. For example, the group of 90 degree rotations of a square
are isomorphic.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

302 CHAPTER 5. VISCOUS INCOMPRESSIBLE LAMINAR FLOW

giving in transformed space

u 2 u
= , (5.161)
t y 2
u(y, 0) = 0, u(0, t) = 1, u(, t) = 0. (5.162)

Now both the original and transformed systems are the same, and the remaining stretching
parameter b does not enter directly into either formulation, so we cannot expect it in the
solution of either form. That is we expect a solution to be independent of the stretching
parameter b. This can be achieved if we take both u and u to be functions of special
combinations of the independent variables, combinations that are formed such that b does
not appear. Eliminating b via
y
eb = , (5.163)
y
we get
 2
t y
= , (5.164)
t y
or after rearrangement
y y
= . (5.165)
t t

 
We thus expect u = u y / t or equivalently u = u y/ t . This form also allows


u = u y/ t , where is any constant. Let us then define our similarity variable
as
y
= . (5.166)
2 t
Here the factor of 1/2 is simply a convenience adopted so that the solution takes on a
traditional form. We would find that any constant in the similarity transformation would
induce a self-similar result.
Let us rewrite the differential equation, boundary, and initial conditions (u /t =
u /y2 , u(y , 0) = 0, u (0, t ) = 1, u (, t) = 0), in terms of the similarity variable .
2

We first must use the chain rule to get expressions for the derivatives. Applying the general
results just developed, we get
u du 1 y 3/2 du du
= = t = , (5.167)
t t d 22 d 2t d
u du 1 du
= = , (5.168)
y y d 2 t d
   
2 u u 1 du
= = , (5.169)
y2 y y y 2 t d
   
1 du 1 1 d2 u 1 d2 u
= = = . (5.170)
2 t y d 2 t 2 t d2 4t d2

CC BY-NC-ND. 17 June 2016, J. M. Powers.

5.2. SIMILARITY SOLUTIONS 303

Thus, applying these rules to our governing linear momenta equation, we recover
du 1 d2 u
= , (5.171)
2t d 4t d2
d2 u du
2
+ 2 = 0. (5.172)
d d
Note our governing equation has a singularity at t = 0. As it appears on both sides of
the equation, we cancel it on both sides, but we shall see that this point is associated with
special behavior of the similarity solution. The important result is that the reduced equation
has dependency on only. If this did not occur, we could not have a similarity solution.
Now consider the initial and boundary conditions. They transform as follows:
y = 0, = = 0, (5.173)
y , = , (5.174)
t 0, = . (5.175)
Note that the three important points for t and y collapse into two corresponding points in
. This is also necessary for the similarity solution to exist. Consequently, our conditions
in space reduce to
u (0) = 1, no slip, (5.176)
u () = 0, initial and far-field. (5.177)
We solve the second order differential equation by the method of reduction of order, noticing
that it is really two first order equations in disguise:
   
d du du
+ 2 = 0. (5.178)
d d d
2
multiplying by the integrating factor e , (5.179)
   
2 d du 2 du
e + 2 e = 0. (5.180)
d d d
 
d 2 du
e = 0, (5.181)
d d
2 du
e = A, (5.182)
d
du 2
= Ae , (5.183)
d
Z
2
u = B + A es ds. (5.184)
0
Now applying the condition u = 1 at = 0 gives
Z 0
2
1 = B+A es ds, (5.185)
| 0 {z }
=0

CC BY-NC-ND. 17 June 2016, J. M. Powers.

304 CHAPTER 5. VISCOUS INCOMPRESSIBLE LAMINAR FLOW

B = 1. (5.186)

So we have
Z
2
u = 1 + A es ds. (5.187)
0

Now applying the condition u = 0 at , we get

Z
2
0 = 1+A es ds, (5.188)
| 0 {z }

= /2


0 = 1+A , (5.189)
2
2
A = . (5.190)

Though not immediately obvious, it can be shown by a simple variable transformation
to a
polar coordinate system that the above integral from 0 to has the value /2. It is not
surprising that this integral has finite value over the semi-infinite domain as the integrand
is bounded between zero and one, and decays rapidly to zero as s . Consequently, the
velocity profile can be written as
Z
2 2
u ( ) = 1 es ds, (5.191)
0
Z y
2 2 t 2
u (y , t ) = 1 es ds, (5.192)
0
 
y
u (y , t ) = erfc . (5.193)
2 t
In the last form above, we have introduced the so-called error function complement, erfc.
Plots for the velocity profile in terms of both and y , t are given in Figure 5.9. We see
that in similarity space, the curve is a single curve that in which u has a value of unity at
= 0 and has nearly relaxed to zero when = 1. In dimensionless physical space, we see
that at early time, there is a thin momentum layer near the surface. At later time more
momentum is present in the fluid. We can say in fact that momentum is diffusing into the
fluid.
We define the momentum diffusion length as the length for which significant momentum
has diffused into the fluid. This is well estimated by taking = 1. In terms of physical
variables, we have
y
= 1, (5.194)
2 t

y = 2 t , (5.195)

CC BY-NC-ND. 17 June 2016, J. M. Powers.

5.2. SIMILARITY SOLUTIONS 305

y
* *

t*
=
3
1

t =
* 2
t =
* 1

1 1
u* u*

Figure 5.9: Sketch of velocity field solution for Stokes first problem in both similarity
coordinate and primitive coordinates y , t .
s
y t
= 2 , (5.196)
U U2
r
2 U 2 t
y = , (5.197)
U
y = 2 t. (5.198)
We can in fact define this as a boundary layer thickness. That is to say the momentum
boundary layer thickness in Stokes first problem grows at a rate proportional to the square
root of momentum diffusivity and time. This class of result is a hallmark of all diffusion
processes, be it mass, momentum, or energy.
Taking standard properties of air, we find after one minute that its boundary layer
thickness is 0.01 m. For oil after one minute, we get a thickness of 0.002 m.
We next consider the shear stress field. For this problem, the shear stress reduces to
simply
u
= . (5.199)
y
Scaling as before by a characteristic stress c , we get
U u
c = , (5.200)
U
y
U 2 1 u
= . (5.201)
c y
Taking c = U 2 / = U 2 /(/) = U 2 , we get
u 1 du
= = , (5.202)
y 2 t d

CC BY-NC-ND. 17 June 2016, J. M. Powers.

306 CHAPTER 5. VISCOUS INCOMPRESSIBLE LAMINAR FLOW

 
1 2 2
= e , (5.203)
2 t
1 2
= e , (5.204)
t
 2
1 y
= exp . (5.205)
t 2 t

Now at the wall, y = 0, and we get

1
|y =0 = . (5.206)
t

So the shear stress does not have a similarity solution, but is directly related to time variation.
The equation holds that the stress is infinite at t = 0, and decreases as time increases. This
is because the velocity gradient flattens as time progresses. It can also be shown that while
the stress is unbounded at a single point in time, that the impulse over a finite time span
is finite, even when the time span includes t = 0. It can also be shown that the flow
corresponds to a pulse of vorticity being introduced at the wall, which subsequently diffuses
into the fluid.
In dimensional terms, we can say

1
2
= q , (5.207)
U U 2t

2
U
= , (5.208)
U t
q
U
= , (5.209)
t

U
= . (5.210)
t

Now let us consider the heat transfer problem. Recall the governing equation, initial and
boundary conditions are
 2
T 2T u
c = k 2 + , (5.211)
t y y
T (y, 0) = To , T (0, t) = To , T (, t) = To . (5.212)

We will adopt the same time tc an length yc scales as before. Take the dimensionless tem-
perature to be
T To
T = . (5.213)
To

CC BY-NC-ND. 17 June 2016, J. M. Powers.

5.2. SIMILARITY SOLUTIONS 307

So we get
 2
cTo T kTo 2 T U 2 u
= + 2 , (5.214)
tc t yc2 y2 yc y
 2
T kTo tc 2 T U 2 tc u
= + 2 , (5.215)
t yc2 cTo y2 yc cTo y
k tc kU 2 1 k k 1
now 2 = 2 2
= = = , (5.216)
yc c U c c c Pr
T 2 tc U 2 U 2 1 U 2 U2
= = = = Ec. (5.217)
yc2 cTo 2 U 2 cTo
cTo cTo

So we have in dimensionless form

 2
T 1 2 T u
= + Ec , (5.218)
t P r y2 y
T (y , 0) = 0, T (0, t ) = 0, T (, t ) = 0. (5.219)

Notice that the only driving inhomogeneity is the viscous work. Now we know from our
solution of the linear momentum equation that
 
u 1 y2
= exp . (5.220)
y t 4t

So we can rewrite the equation for temperature variation as

 
T 1 2 T Ec y2
= + exp , (5.221)
t P r y2 t 2t
T (y , 0) = 0, T (0, t ) = 0, T (, t ) = 0. (5.222)

Before considering the general solution, let us consider some limiting cases.

Ec 0
In the limit as Ec 0, we get a trivial solution, T (y , t ) = 0.

Pr
Recalling that the Prandtl number is the ratio of momentum diffusivity to thermal
diffusivity, this limit corresponds to materials for which momentum diffusivity is much
greater than thermal diffusivity. For example for SAE 30 oil, the Prandtl number is
around 3500. Naively assuming that we can simply neglect conduction, we write the
energy equation in this limit as
 
T Ec y2
= exp . (5.223)
t t 2t

CC BY-NC-ND. 17 June 2016, J. M. Powers.

308 CHAPTER 5. VISCOUS INCOMPRESSIBLE LAMINAR FLOW

and with T = T ( ) and = y /(2 t ), we get the transformed partial time deriva-
tive to be
T dT
= . (5.224)
t 2t d
So the governing equation reduces to
dT Ec 22
= e , (5.225)
2t d t
dT 2Ec 1 22
= e , (5.226)
d
Z
2Ec 1 2s2
T = e ds. (5.227)
s

We cannot satisfy both boundary conditions; the equation has been solved so as to
satisfy the boundary condition in the far field of T () = 0.
Unfortunately, we notice that we cannot satisfy the boundary condition at = 0.
We simply do not have enough degrees of freedom. In actuality, what we have found
is an outer solution, and to match the boundary condition at 0, we would have to
reintroduce conduction, which has a higher derivative.
First let us see how the outer solution behaves near = 0. Expanding the differential
equation in a Taylor series about = 0 and solving gives
 
dT 2Ec 1 3
= 2 + 2 + . . . , (5.228)
d
 
2Ec 2 1 4
T = ln + + . . . . (5.229)
2

It turns out that solving the inner layer problem and the matching is of about the
same difficulty as solving the full general problem, so we will defer this until later in
this section.

Pr 0
In this limit, we get
2 T
= 0. (5.230)
y2
The solution which satisfies the boundary conditions is

T = 0. (5.231)

In this limit, momentum diffuses slowly relative to energy. So we can interpret the
results as follows. In the boundary layer, momentum is generated in a thin layer.
Viscous dissipation in this layer gives rise to a local change in temperature in the layer

CC BY-NC-ND. 17 June 2016, J. M. Powers.

5.2. SIMILARITY SOLUTIONS 309

which rapidly diffuses throughout the entire flow. The effect of smearing a localized
finite thermal energy input over a semi-infinite domain has a negligible influence on
the temperature of the global domain.
So let us bring back diffusion and study solutions for finite Prandtl number. Our gov-
erning equation in similarity variables then becomes
dT 1 1 d2 T Ec 22
= 2
+ e , (5.232)
2t d P r 4t d t
dT 1 d2 T 4Ec 22
2 = + e , (5.233)
d P r d2
2
d T dT 4 2

2
+ 2P r = EcP r e2 , (5.234)
d d
T (0) = 0, T () = 0. (5.235)
The second order differential equation is really two first order differential equations in dis-
2
guise. There is an integrating factor of eP r . Multiplying by the integrating factor and
operating on the system, we find
2
2 d
T P r 2 dT 4 2
eP r + 2P r e = EcP r e(P r2) , (5.236)
d2 d
 
d P r 2 dT 4 2
e = EcP r e(P r2) , (5.237)
d d
Z
2 dT 4 2
eP r = EcP r e(P r2)s ds + C1 , (5.238)
d 0
Z
dT 4 P r 2 2 2
= EcP r e e(P r2)s ds + C1 eP r(5.239)
,
d 0
Z Z p
4 P r p2 2
T = EcP r e e(P r2)s ds dp

Z 0 0
2
+C1 eP r s ds + C2 . (5.240)
0

The boundary condition T (0) = 0 gives us C2 = 0. The boundary condition at gives us

then
Z Z p Z
4 P r p2 (P r2)s2 2
0 = EcP r e e ds dp + C1 eP r s ds . (5.241)
0 0 0
| {z }
1
p
2 Pr
(5.242)
Therefore, we get
Z Z p
r
4 P r p2 (P r2)s2 C1
EcP r e e ds dp = , (5.243)
0 0 2 Pr

CC BY-NC-ND. 17 June 2016, J. M. Powers.

310 CHAPTER 5. VISCOUS INCOMPRESSIBLE LAMINAR FLOW

Z Z p
8 3/2 P r p2 2
C1 = 3/2
EcP r e e(P r2)s ds dp. (5.244)
0 0

So finally, we have for the temperature profile

Z Z p
4 P r p2 2
T ( ) = EcP r e e(P r2)s ds dp
0
 Z 0 Z p  Z
8 3/2 P r p2 (P r2)s2 s2
+ EcP r e e ds dp eP r ds. (5.245)
3/2 0 0 0

This simplifies somewhat to

Z    Z   
2EcP r P r p2 p2
p e erf 2 P r p dp erf Pr eP r erf 2 P r p dp .
(2 P r) 0 0
(5.246)
This analysis simplifies considerably in the limit of P r = 1, that is when momentum and
energy diffuse at the same rate. This is a close to reality for many gases. In this case, the
temperature profile becomes
Z Z p Z
4 p2 s2 2
T ( ) = Ec e e ds dp + C1 es ds. (5.247)
0 0 0
Rp 2 2
Now if h(p) = 0 es ds, we get dh/dp = ep . Using this, we can rewrite the temperature
profile as
Z Z
4 dh 2
T ( ) = Ec h(p) dp + C1 es ds, (5.248)
0 dp 0
Z  2 Z
4Ec h 2
= d + C1 es ds, (5.249)
0 2 0
  Z 2 Z
4Ec 1 s2 2
= e ds + C1 es ds, (5.250)
2 0 0
 Z  Z
2Ec s2 2
= e ds + C1 es ds. (5.251)
0 0

Now for T () = 0, we get

 Z Z
2Ec s2 2
0 = e ds + C1 es ds, (5.252)
0 0
 
2Ec
0 = + C1 , (5.253)
2 2
Ec
C1 = . (5.254)

So the temperature profile can be expressed as
Z  Z 
Ec s2 2 s2
T ( ) = e ds 1 e ds . (5.255)
0 0

CC BY-NC-ND. 17 June 2016, J. M. Powers.

5.2. SIMILARITY SOLUTIONS 311

*
3

Pr = 1
Ec = 1

0 0.12 T
*

Figure 5.10: Plot of temperature field for Stokes first problem for P r = 1, Ec = 1.

We notice that we can write this directly in terms of the velocity as

Ec
T ( ) = u ( ) (1 u ( )) . (5.256)
2
This is a consequence of what is known as Reynolds analogy which holds for P r = 1 that
the temperature field can be directly related to the velocity field. The temperature field for
Stokes first problem for P r = 1, Ec = 1 is plotted in Figure 5.10.

5.2.2 Blasius boundary layer

We next consider the well known problem of the flow of a viscous fluid over a flat plate. This
problem forms the foundation for a variety of viscous flows over more complicated geometries.
It also illustrates some important features of viscous flow physics, as well as giving the original
motivating problem for the mathematical technique of matched asymptotic expansions. Here
we will consider, as sketched in Figure 5.11, the incompressible flow of viscous fluid of
constant viscosity and thermal conductivity over a flat plate. In the far field, the fluid will
be a uniform stream with constant velocity. At the plate surface, the no-slip condition must
be enforced, which will give rise to a zone of adjustment where the fluids velocity changes
from zero at the plate surface to its freestream value. This zone is called the boundary layer.
Considering first the velocity field, we find, assuming the flow is steady as well, that the
dimensionless two-dimensional Navier-Stokes equations are as follows
u v
+ = 0, (5.257)
x y

CC BY-NC-ND. 17 June 2016, J. M. Powers.

312 CHAPTER 5. VISCOUS INCOMPRESSIBLE LAMINAR FLOW

uo

Figure 5.11: Schematic for flat plate boundary layer problem.

 
u u p 1 2u 2u
u +v = + + , (5.258)
x y x Re x2 y 2
 
v v p 1 2v 2v
u +v = + + . (5.259)
x y y Re x2 y 2
The dimensionless boundary conditions are

u(x, y ) = 1, (5.260)
v(x, y ) = 0, (5.261)
p(x, y ) = 0, (5.262)
u(x, 0) = 0, (5.263)
v(x, 0) = 0. (5.264)

In this section, we are dispensing with the s and assuming all variables are dimensionless.
In fact we have assumed a scaling of the following form, where dim is a subscript denoting
a dimensional variable.
udim vdim xdim ydim pdim po
u= , v= , x= , y= , p= . (5.265)
uo uo L L u2o
Note for our flat plate of semi-infinite extent, we do not have a natural length scale. This
suggests that we may find a similarity solution which removes the effect of L.
Now let us consider that for Re , we have an outer solution of u = 1 to be valid
for most of the flow field sufficiently far away from the plate surface. In fact the solution
u = 1, v = 0, p = 0, satisfies all of the governing equations and boundary conditions except
for the no slip condition at y = 0. Because in the limit as Re , we effectively ignore the
high order derivatives found in the viscous terms, we cannot expect to satisfy all boundary

CC BY-NC-ND. 17 June 2016, J. M. Powers.

5.2. SIMILARITY SOLUTIONS 313

conditions for the full problem. We call this the outer solution, which is also an inviscid
solution to the equations, allowing for a slip condition at the boundary.
Let us rescale our equations near the plate surface y = 0 to
bring back the effect of the viscous terms,

bring back the no-slip condition, and

match our inviscid outer solution to a viscous inner solution.

This is the first example of the use of the method of matched asymptotic expansions as
introduced by Prandtl and his student Blasius in the early twentieth century.
With some difficulty, we could show how to choose the scaling, let us simply adopt a
scaling and show that it indeed achieves our desired end. So let us take a scaled y distance
and velocity, denoted by asuperscript, to be

v = Re v, y = Re y. (5.266)

With this scaling, assuming the Reynolds number is large, when we examine small y or
v, we are examining an order unity y or v. Our equations rescale as

u 1/ Re v
+ = 0, (5.267)
x 1/ Re y
 
u 1/ Re u p 1 2u 2u
u + v = + + Re 2 , (5.268)
x 1/ Re y x Re x2 y

1 v (1/ Re)(1/ Re) v 1 p
u + v =
Re x 1/ Re y 1/ Re y
!
1 1 2 v 1/ Re 2 v
+ + . (5.269)
Re Re x2 1/Re y 2

Simplifying, this reduces to

u v
+ = 0, (5.270)
x y
u u p 1 2u 2u
u + v = + + , (5.271)
x y x Re x2 y 2
v v p 1 2 v 2 v
u + v = Re + + . (5.272)
x y y Re x2 y 2
Now in the limit as Re , the rescaled equations reduce to the well known boundary
layer equations:
u v
+ = 0, (5.273)
x y

CC BY-NC-ND. 17 June 2016, J. M. Powers.

314 CHAPTER 5. VISCOUS INCOMPRESSIBLE LAMINAR FLOW

u u p 2 u
u + v = + , (5.274)
x y x y 2
p
0 = . (5.275)
y
To match the outer solution, we need the boundary conditions which are
u(x, y ) = 1, (5.276)
v(x, y ) = 0, (5.277)
p(x, y ) = 0, (5.278)
u(x, 0) = 0, (5.279)
v(x, 0) = 0. (5.280)
The y momentum equation gives us
p = p(x). (5.281)
In general, we can consider this to be an imposed pressure gradient which is supplied by
the outer inviscid solution. For general flows, that pressure gradient dp/dx will be non-zero.
For the Blasius problem, we will choose to study problems for which there is no pressure
gradient. That is we take
p(x) = 0, for Blasius flat plate boundary layer. (5.282)
So called Falkner-Skan solutions consider flows over curved plates, for which the outer inviscid
solution does not have a constant pressure. This ultimately affects the behavior of the fluid
in the boundary layer, giving results which differ in important features from our Blasius
problem.
With our assumptions, the Blasius problem reduces to
u v
+ = 0, (5.283)
x y
u u 2u
u + v = . (5.284)
x y y 2
The boundary conditions are now only on velocity and are
u(x, y ) = 1, (5.285)
v(x, y ) = 0, (5.286)
u(x, 0) = 0, (5.287)
v(x, 0) = 0. (5.288)
Now to simplify, we invoke the stream function , which allows us to satisfy continuity
automatically and eliminate u and v at the expense of raising the order of the differential
equation. So taking

u= , v = , (5.289)
y x

CC BY-NC-ND. 17 June 2016, J. M. Powers.

5.2. SIMILARITY SOLUTIONS 315

2 2
we find that mass conservation reduces to x y
yx
= 0. The x momentum equation and
associated boundary conditions become

2 2 3
= , (5.290)
y x y x y 2 y 3

(x, y ) = 1, (5.291)
y

(x, y ) = 0, (5.292)
x

(x, 0) = 0, (5.293)
y

(x, 0) = 0. (5.294)
x
Let us try stretching all the variables of this system to see if there are stretching transforma-
tions under which the system exhibits symmetry; that is we seek a stretching transformation
under which the system is invariant. Take

x = ea x, y = eb y, = ec . (5.295)

Under this transformation, the x-momentum equation and boundary conditions transform
to

2 2 3
ea+2b2c ea+2b2c = e3bc , (5.296)
y x y x y 2 y 3

ebc (x, y ) = 1, (5.297)
y

eac (x, y ) = 0, (5.298)
x

ebc (x, 0) = 0, (5.299)
y

eac (x, 0) = 0. (5.300)
x
If we demand b = c and a = 2c, then the transformation is invariant, yielding

2 2 3
= , (5.301)
y x y x y 2 y 3

(x, y ) = 1, (5.302)
y

(x, y ) = 0, (5.303)
x

CC BY-NC-ND. 17 June 2016, J. M. Powers.

316 CHAPTER 5. VISCOUS INCOMPRESSIBLE LAMINAR FLOW

(x, 0) = 0, (5.304)
y

(x, 0) = 0. (5.305)
x
Now our transformation is reduced to

x = e2c x, y = ec y, = ec . (5.306)

Since c does not appear explicitly in either the original equation set nor the transformed
equation set, the solution
p must not depend on this stretching. Eliminating c from the
transformation by ec = x/x we find that
r r
y x x
= , = , (5.307)
y x x
or
y y
= , = . (5.308)
x x x x
Thus motivated, let us seek solutions of the form
 
y
=f . (5.309)
x x

That is taking
y
= , (5.310)
x
we seek

= xf (). (5.311)
Let us check that our similarity variable is independent of L our unknown length scale.
r r
y Re y Re ydim /L uo L ydim L uo ydim
= = = p = = . (5.312)
x x xdim /L L xdim xdim

So indeed, our similarity variable is independent of any arbitrary length scale we happen to
have chosen.
With our similarity transformation, we have

1 1
= yx3/2 = , (5.313)
x 2 2x
1
= . (5.314)
y x

CC BY-NC-ND. 17 June 2016, J. M. Powers.

5.2. SIMILARITY SOLUTIONS 317

Now we need expressions for /x, / y, 2 /x y, 2 / y 2 , and 3 / y 3 . First, con-

sider the partial derivatives of the stream function . Operating on each partial derivative,
we find

df 1 1
= xf () = x + f, (5.315)
x x d x 2 x
   
1 df 1 1 1 df
= x + f= f , (5.316)
2 x d 2 x 2 x d
 
1 df
so v = f . (5.317)
2 x d

df df 1 df
= xf () = x (f ()) = x = x = , (5.318)
y y y d y d x d
df
so u= = . (5.319)
y d
   
2 df d2 f 1 d2 f
= = = 2 = 2. (5.320)
x y x y x d d x 2x d
   
2 df d2 f 1 d2 f
= = = = . (5.321)
y 2 y y y d d 2 y x d 2
     
3 2 1 d2 f 1 d2 f 1 d3 f
= = = = , (5.322)
y 3 y y 2 y x d 2 x y d 2 x d 3 y
1 d3 f
= . (5.323)
x d 3
Now we substitute each of these expressions into the x momentum equation and get
   
df 1 d2 f 1 df 1 d2 f 1 d3 f
2 f = , (5.324)
d 2x d 2 x d x d 2 x d 3
 
df d2 f df d2 f d3 f
f = 2 , (5.325)
d d 2 d d 2 d 3
d2 f d3 f
f 2 = 2 3 , (5.326)
d d
d3 f 1 d2 f
+ f = 0. (5.327)
d 3 2 d 2
This is a third order non-linear ordinary differential equation for f (). We need three
boundary conditions. Now at the surface y = 0, we have = 0. And as y , we have
. To satisfy the no-slip condition on u at the plate surface, we require

df 
= 0. (5.328)
d  =0

For no-slip on v, we require

 
1 df
v(0) = 0 = f , (5.329)
2 x d

CC BY-NC-ND. 17 June 2016, J. M. Powers.

318 CHAPTER 5. VISCOUS INCOMPRESSIBLE LAMINAR FLOW


df 
0 = f (0) 0  , (5.330)
d =0
| {z }
=0
f (0) = 0. (5.331)

And to satisfy the freestream condition on u as , we need


df 
= 1. (5.332)
d 

The most standard way to solve non-linear ordinary differential equations of this type is to
reduce them to systems of first order ordinary differential equations and use some numerical
technique, such as a Runge7 -Kutta integration. We recall that Runge-Kutta techniques, as
well as most other common techniques, require a well-defined set of initial conditions to
predict the final state. To achieve the desired form, we define

df d2 f
g , h 2. (5.333)
d d

Thus the x momentum equation becomes

dh 1
+ f h = 0. (5.334)
d 2

But this is one equation in three unknowns. We need to write our equations as a system of
three first order equations, along with associated initial conditions. They are

df
= g, f (0) = 0, (5.335)
d
dg
= h, g(0) = 0, (5.336)
d
dh 1
= f h, h(0) =?. (5.337)
d 2

Everything is well-defined except we do not have an initial condition on h. We do however

have a far-field condition on g which is g() = 1. One viable option we have for getting a
final solution is to use a numerical trial and error procedure, guessing h(0) until we find that
g() 1. We will use a slightly more efficient method here, which only requires one guess.
To do this, let us first demonstrate the following lemma: If F () is a solution to d3 f /d 3 +
1
2
f (d2 f /d 2) = 0, then aF (a) is also a solution. The proof is as follows. Take w() =
7
Carl David Tolme Runge, 1856-1927, German mathematician and physicist, close friend of Max Planck,
studied spectral line elements of non-Hydrogen molecules, held chairs at Hanover and Gottingen, entertained
grandchildren at age 70 by doing handstands.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

5.2. SIMILARITY SOLUTIONS 319

aF (a). Then we have

w = aF (a), (5.338)
dw dF (a)
= a2 , (5.339)
d d
d2 w d2 F (a)
= a3 , (5.340)
d 2 d 2
d3 w d3 F (a)
= a4 . (5.341)
d 3 d 3

Substituting these expressions into the x momentum equation, we find

d3 F (a) 1 4 d2 F (a)
a4 + a F (a) = 0, (5.342)
d 3 2 d 2
d3 F (a) 1 d2 F (a)
+ F (a) = 0. (5.343)
d 3 2 d 2

But we know this to be true as F (a) is a solution. Hence aF (a) is also a solution.
So to solve our non-linear system, let us first solve the following related system:

dF
= G, F (0) = 0, (5.344)
d
dG
= H, G(0) = 0, (5.345)
d
dH 1
= F H, H(0) = 1. (5.346)
d 2

After one numerical integration, we find that with this guess for H(0) that

G() = 2.08540918... (5.347)

Now our numerical solution also gives us F , and so we know that f = aF (a) is also a
solution. Moreover

df dF (a)
= a2 , that is (5.348)
d d
g() = a2 G(a). (5.349)

Now we want g() = 1, so take 1 = a2 G(), so a2 = 1/G(). So

1
a= p . (5.350)
G()

CC BY-NC-ND. 17 June 2016, J. M. Powers.

320 CHAPTER 5. VISCOUS INCOMPRESSIBLE LAMINAR FLOW

*
5

0 1 u*

Figure 5.12: Velocity profile for Blasius boundary layer.

Now

d2 f 2
3 d F (a)
= a , (5.351)
d 2 d 2
 
d2 f  2
3 d F (a) 

= a , (5.352)
d 2 
=0 d 2  =0
h(0) = a3 H(0), (5.353)
h(0) = a3 (1), (5.354)
h(0) = a3 = G3/2 (), (5.355)
h(0) = (2.08540918...)3/2 = 0.332057335... (5.356)

This is the proper choice for the initial condition on h. Numerically integrating once more,
we get the behavior of f , g, and h as functions of which indeed satisfies the condition at
. A plot of u = df /d as a function of is shown in Figure 5.12. From this plot, we see
that when = 5, the velocity has nearly acquired the freestream value of u = 1. In fact,
examination of the numerical results shows that when = 4.9, that the u component of
velocity has 0.99 of its freestream value. As the velocity only reaches its freestream value at
, we define the boundary layer thickness, 0.99 , as that value of ydim for which the velocity

CC BY-NC-ND. 17 June 2016, J. M. Powers.

5.2. SIMILARITY SOLUTIONS 321

has 0.99 of its freestream value. Recalling that

r
uo ydim
= , (5.357)
xdim

we say that r
uo 0.99
4.9 = . (5.358)
xdim
Rearranging, we get
r
0.99
= 4.9 , (5.359)
xdim uo xdim
= 4.9Re1/2
xdim . (5.360)

Here we have taken a Reynolds number based on local distance to be

uo xdim
Rexdim = . (5.361)

This formula is valid for laminar flows, and has been seen to be valid for Rexdim < 3 106 .
For greater lengths, there can be a transition to turbulent flow. For water flowing a 1 m/s
and a downstream distance of 1 m, we find 0.99 = 0.5 cm. For air under the same conditions,
we find 0.99 = 1.9 cm. We also note that the boundary layer grows with the square root
of distance along the plate. We further note that higher kinematic viscosity leads to thicker
boundary layers, while lower kinematic viscosity lead to thinner boundary layers.
Now let us determine the shear stress at the wall, and the viscous force acting on the
wall. So let us find 
udim 
w = . (5.362)
ydim ydim =0
Consider
u 2 1 d2 f
= = , (5.363)
y y 2 x d 2
 
uudim
o 1 d2 f
q  = p xdim 2 , (5.364)
uo L ydim
L
d
L
r
udim uo 1 d2 f
= uo , (5.365)
ydim xdim d 2
r
udim uo d2 f
= = uo , (5.366)
ydim xdim d 2
r
(0) d2 f
1 = Cf = 2 (0), (5.367)
2
u2o uo xdim d 2

CC BY-NC-ND. 17 June 2016, J. M. Powers.

322 CHAPTER 5. VISCOUS INCOMPRESSIBLE LAMINAR FLOW

2
1/2 d f
Cf = 2(Rexdim ) (0), (5.368)
d 2
0.664...
Cf = p . (5.369)
Rexdim

We notice that at xdim = 0 that the stress is infinite. This seeming problem is seen not to be
one when we consider the actual viscous force on a finite length of plate. Consider a plate
of length L and width b. Then the viscous force acting on the plate is
Z L
F = dA, (5.370)
0
Z L
= (xdim , 0)bdxdim , (5.371)
0
Z L
1
= b f (0)uo uo dxdim , (5.372)
0 xdim
Z L
dxdim
= bf (0)uo uo , (5.373)
0 xdim
L
= bf (0)uo uo (2 xdim )0 , (5.374)

= 2bf (0)uo uo L, (5.375)
r
F 1/2 1/2
1 = CD = 4f (0) = 4f (0)ReL = 1.328ReL . (5.376)
2
u2o Lb uo L

Now let us consider the thermal boundary layer. Here we will take the boundary condi-
tions so that the wall and far field are held at a constant fixed temperature Tdim = To . We
need to do the scaling on the energy equation, so let us start with the steady incompressible
two-dimensional dimensional energy equation:
   2 
Tdim Tdim Tdim 2 Tdim
cp udim + vdim = k + (5.377)
xdim ydim x2dim 2
ydim
 2  2  2 !
udim vdim udim vdim
+ 2 +2 + + .
xdim ydim ydim xdim

Taking as before,

xdim ydim Tdim To udim vdim

x= , y= , T = , u= , v= . (5.378)
L L To uo uo

Making these substitutions, we get

   
cp uo To T T kTo 2 T 2T
u +v = + (5.379)
L x y L2 x2 y 2

CC BY-NC-ND. 17 June 2016, J. M. Powers.

5.2. SIMILARITY SOLUTIONS 323

 2  2  2 !
u2 u v u v
+ 2o 2 +2 + + ,
L x y y x
 2 
T T k T 2T
u +v = + (5.380)
x y cp uo L x2 y 2
 2  2  2 !
uo u v u v
+ 2 +2 + + .
cp LTo x y y x
(5.381)

Now we have
k k 1 1
= = , (5.382)
cp uo L cp uo L P r Re
uo u2o Ec
= = . (5.383)
cp LTo uo L cp To Re

So the dimensionless energy equation with boundary conditions can be written as

 2 
T T 1 T 2T
u +v = + (5.384)
x y P rRe x2 y 2
 2  2  2 !
Ec u v u v
+ 2 +2 + + ,
Re x y y x
T (x, 0) = 0, T (x, ) = 0. (5.385)

Now as Re , we see that T = 0 is a solution that satisfies the energy equation and
all boundary conditions. For finite Reynolds number, non-zero velocity gradients generate
a temperature
field. Once again, we rescale in the boundary layer using v = Re v, and
y = Re y. This gives
 2 
T 1 1 T 1 T 2T
u + v = + Re 2 (5.386)
x Re 1/ Re y P rRe x2 y
 2  2  2 !
Ec u v u 1 v
+ 2 +2 + Re + .
Re x y y Re x
 
T T 1 1 2T 2T
u + v = + (5.387)
x y P r Re x2 y 2
 2  2  2 !
2 u 2 v u 1 v
+Ec + + + .
Re x Re y y Re x
as Re , (5.388)
 2
T T 1 2T u
u + v = + Ec . (5.389)
x y P r y 2 y

CC BY-NC-ND. 17 June 2016, J. M. Powers.

324 CHAPTER 5. VISCOUS INCOMPRESSIBLE LAMINAR FLOW

Now take T = T () with = y/ x as well as u = df /d, v = (1/(2 x)) (f (df /d))
and u/ y = (1/ x)(d2 f /d 2). We also have for derivatives, that
 
T dT dT 1
= = , (5.390)
x d x d 2x
T dT dT 1
= = , (5.391)
y d y d x
   
2T T 1 dT 1 dT 1 d2 T
= = = = . (5.392)
y 2 y y y x d x y d x d 2

The energy equation is then rendered as

      2
df 1 dT 1 df 1 dT 1 1 d2 T Ec d2 f
+ f = + ,(5.393)
d 2 x d 2 x d x d P r x d 2 x d 2
   2 2
1 df dT 1 df dT 1 d2 T df
f = 2
+ Ec , (5.394)
2 d d 2 d d P r d d 2
 2 2
1 dT 1 d2 T df
f = + Ec , (5.395)
2 d P r d 2 d 2
 2 2
d2 T 1 dT df
2
+ Pr f = P rEc , (5.396)
d 2 d d 2
T (0) = 0, T () = 0. (5.397)

Now for Ec 0, we get T = 0 as a solution which satisfies the governing differential

equation and boundary conditions. Let us consider a solution for non-trivial Ec, but for
P r = 1. We could extend this for general values of P r as well. Here, following Reynolds
analogy, when thermal diffusivity equals momentum diffusivity, we expect the temperature
field to be directly related to the velocity field. For P r = 1, the energy equation reduces to
 2 2
d2 T 1 dT df
2
+ f = Ec , (5.398)
d 2 d d 2
T (0) = 0, T () = 0. (5.399)

Here the integrating factor is R 1

e 0 2 f (t)dt . (5.400)
Multiplying the energy equation by the integrating factor gives
 2 2
R 1 d2 T 1 R 1 f (t)dt dT R 1 df
e 0 2 f (t)dt + fe 0 2 = Ec e 0 2 f (t)dt
, (5.401)
d 2 2 d d 2
 R   2 2
d 1
f (t)dt dT
R 1
f (t)dt df
e 0 2 = Ec e 0 2 . (5.402)
d d d 2

CC BY-NC-ND. 17 June 2016, J. M. Powers.

5.2. SIMILARITY SOLUTIONS 325

Now from the x momentum equation, f + 12 f f = 0, we have

f
f = 2 . (5.403)
f
So we can rewrite the integrating factor as
f (0)
 
R 1
R 1 (2)f f ()
f (t)dt dt ln f (0)
e 0 2 =e 0 2 f =e = . (5.404)
f ()
So the energy equation can be written as
     2 2
d f (0) dT f (0) d f
= Ec , (5.405)

d f () d
f () d 2
d2 f
= Ecf (0) 2 , (5.406)
d
Z 2
f (0) dT df

= Ecf (0) 2
ds + C1 , (5.407)
f () d 0 ds
Z
dT d2 f d2 f d2 f
= Ec 2 ds + C 1 , (5.408)
d d 0 ds2 d 2

2
d f df d2 f
= = Ec 2 f (0) + C1 2 , (5.409)
d d | {z } d
=0
2 2
d f df df
= Ec2
+ C1 2 , (5.410)
d d d
 2 !
d 1 df d2 f
= Ec + C1 2 , (5.411)
d 2 d d
 2
Ec df df
T = + C1 + C2 , (5.412)
2 d d
Ec
T (0) = 0 = (f (0))2 + C1 f (0) +C2 , (5.413)
2 | {z } | {z }
=0 =0
C2 = 0, (5.414)
Ec
T () = 0 = (f ())2 + C1 f (), (5.415)
2 | {z } | {z }
=1 =1
Ec
C1 = , (5.416)
2  
Ec df df
T () = 1 , (5.417)
2 d d
Ec
T () = u()(1 u()). (5.418)
2
A plot of the temperature profile for P r = 1 and Ec = 1 is given in Figure 5.13.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

326 CHAPTER 5. VISCOUS INCOMPRESSIBLE LAMINAR FLOW

*
6
Pr = 1
Ec = 1
5

1 T
*

Figure 5.13: Temperature profile for Blasius boundary layer, Ec = 1, P r = 1.

CC BY-NC-ND. 17 June 2016, J. M. Powers.

Bibliography

This bibliography focuses on books which are closely related to the material presented in
this course in classical fluid mechanics, especially with regards to graduate level treatment
of continuum mechanical principles applied to fluids, compressible flow, viscous flow, and
vortex dynamics. It also has some general works of historic importance. It is by no means
a comprehensive survey of works on fluid mechanics. Only a few works are given here
which focus on such important topics as low Reynolds number flows, turbulence, bio-fluids,
computational fluid dynamics, microfluids, molecular dynamics, magneto-hydrodynamics,
geo-physical flows, rheology, astrophysical flows, as well as elementary undergraduate texts.
That said, those which are listed are among the best that exist and would be useful to
examine.

I. H. Abbott, Theory of Wing Sections, Dover, New York, 1980.

D. J. Acheson, Elementary Fluid Dynamics, Oxford University Press, Oxford, 1990.

J. D. Anderson, Modern Compressible Flow with Historical Perspective, 3rd edition, McGraw-
Hill, New York, 2003.

J. D. Anderson, Hypersonic and High Temperature Gas Dynamics, 2nd edition, AIAA,
Reston, Virginia, 2006.

R. Aris, Vectors, Tensors, and the Basic Equations of Fluid Mechanics, Dover, New York,
1962.

H. Ashley and M. Landahl, Aerodynamics of Wings and Bodies, Dover, New York, 1985.

G. I. Barenblatt, Scaling, Self-Similarity, and Intermediate Asymptotics, Cambridge Uni-

versity Press, New York, 1996.

G. Batchelor, An Introduction to Fluid Dynamics, Cambridge University Press, Cambridge,

UK, 1967.

J. Bear, Dynamics of Fluids in Porous Media, Dover, New York, 1988.

R. B. Bird, W. E. Stewart, and E. N. Lightfoot, Transport Phenomena, Revised 2nd edition,

John Wiley, New York, 2007.

327
R. B. Bird, R. C. Armstrong, and O. Hassager, Dynamics of Polymeric Liquids, Vol. 1:
Fluid Mechanics, Vol. 2: Kinetic Theory, John Wiley, New York, 1987.

A. I. Borisenko and I. E. Tarapov, Vector and Tensor Analysis with Applications, Dover,
New York, 1968.

R. S. Brodkey, The Phenomena of Fluid Motions, Dover, New York, 1995.

B. J. Cantwell, Introduction to Symmetry Analysis, Cambridge University Press, New York,

2002.

S. Chandrasekhar, Hydrodynamic and Hydromagnetic Stability, Dover, New York, 1961.

S. Chapman and T. G. Cowling, The Mathematical Theory of Non-Uniform Gases, 3rd

edition, Cambridge University Press, Cambridge, UK, 1991.

R. Chevray and J. Mathieu, Topics in Fluid Mechanics, Cambridge University Press, Cam-
bridge, UK, 1993.

A. J. Chorin and J. E. Marsden, A Mathematical Introduction to Fluid Mechanics, 3rd

edition, Springer, New York, 1993.

R. Courant and K. O. Friedrichs, Supersonic Flow and Shock Waves, Springer, New York,
1976.

W. O. Criminale, T. L. Jackson, and R. D. Joslin, Theory and Computation in Hydrody-

namic Stability, Cambridge University Press, Cambridge, UK, 2003.

I. G. Currie, Fundamental Mechanics of Fluids, 4th edition, CRC Press, Boca Raton,
Florida, 2013.

S. R. de Groot and P. Mazur, Non-Equilibrium Thermodynamics, Dover, New York, 1984.

P. G. Drazin and W. H. Reid, Hydrodynamic Stability, Cambridge University Press, Cam-

bridge, UK, 1981.

P. G. Drazin, Introduction to Hydrodynamic Stability, Cambridge University Press, Cam-

bridge, UK, 2002.

D. A. Drew and S. L. Passman, Theory of Multicomponent Fluids, Springer, New York,

1999.

G. Emanuel, Analytical Fluid Dynamics, 3rd edition, CRC Press, Boca Raton, Florida,
2016.

G. Emanuel, Gasdynamics: Theory and Applications, AIAA, New York, 1986.

A. C. Eringen, Mechanics of Continua, 2nd edition, Krieger, Malabar, Florida, 1989.

328
J. H. Ferziger and M. Peric, Computational Methods for Fluid Dynamics, 3rd revised edition,
Springer, New York, 2002.

R. P. Feynman, Lectures on Physics, Addison-Wesley, New York, 1970.

Y. C. Fung, Foundations of Solid Mechanics, Prentice-Hall, Englewood Cliffs, New Jersey,

1965.

M. Gad-el-Hak, Flow Control: Passive, Active and Reactive Flow Management, Cambridge
University Press, Cambridge, UK, 2000.

H. Glauert, The Elements of Aerofoil and Airscrew Theory, Cambridge University Press,
Cambridge, UK, 1983.

S. Goldstein, ed., Modern Developments in Fluid Dynamics, Vols. I and II, Dover, New
York, 1965.

W. T. Grandy, Entropy and the Time Evolution of Macroscopic Systems, Oxford University
Press, Oxford, 2008.

R. A. Granger, Fluid Mechanics, Dover, New York, 1995.

W. D. Hayes and R. F. Probstein, Hypersonic Flow Theory, Academic Press, New York,
1959.

H. Helmholtz, On the Sensations of Tone, Dover, New York, 1954.

C. Hirsch, Numerical Computation of Internal and External Flows, Vols. 1 and 2, John
Wiley, New York, 1989.

J. O. Hirschfelder, C. F. Curtiss, and R. B. Bird, Molecular Theory of Gases and Liquids,

Wiley, New York, 1954.

R. T. Jones, Wing Theory, Princeton University Press, Princeton, 1990.

D. D. Joseph, Fluid Dynamics of Viscoelastic Liquids, Springer, New York, 1990.

G. E. Karniadakis and A. Beskok, Microflows: Fundamentals and Simulation, Springer,

Berlin, 2001.

G. E. Karniadakis and S. J. Sherwin, Spectral/HP Element Methods for CFD, 2nd edition
Oxford University Press, Oxford, 2005.

E. L. Koschmieder, Benard Cells and Taylor Vortices, Cambridge University Press, Cam-
bridge, UK, 1993.

A. M. Kuethe and C.-Y. Chow, Foundations of Aerodynamics: Bases of Aerodynamic De-

sign, 5th edition, John Wiley, New York, 1998.

329
P. K. Kundu, I. M. Cohen, D. R. Dowling, Fluid Mechanics, 6th edition, Academic Press,
Amsterdam 2015.
H. Lamb, Hydrodynamics, 6th edition, Dover, New York, 1993.

C. B. Laney, Computational Gasdynamics, Cambridge University Press, Cambridge, UK,

1998.
M. T. Landahl, Unsteady Transonic Flow, Cambridge University Press, Cambridge, UK,
1989.

L. D. Landau, and E. M. Lifshitz, Fluid Mechanics, Pergamon Press, Oxford, 1959.

G. L. Leal, Laminar Flow and Convective Transport Processes: Scaling Principles and
Asymptotic Analysis, Butterworth-Heinemann, Stoneham, Massachusetts, 1992.

R. J. LeVeque, Finite Volume Methods for Hyperbolic Problems, Cambridge University

Press, Cambridge, UK, 2002.

H. W. Liepmann, and A. Roshko, Elements of Gasdynamics, Dover, New York, 2002.

V. L. Liseikin, Grid Generation Methods, 2nd edition, Springer, Dordrecht, 2010.

J. Lighthill, Waves in Fluids, Cambridge University Press, Cambridge, UK, 1978.

A. J. McConnell, Applications of Tensor Analysis, Dover, New York, 1957.

R. E. Meyer, Introduction to Mathematical Fluid Dynamics, Dover, New York, 1982.

L. M. Milne-Thompson, Theoretical Aerodynamics, 4th edition, Dover, New York, 1958.

L. M. Milne-Thompson, Theoretical Hydrodynamics, 5th edition, Dover, New York, 1996.

P. M. Morse and H. Feshbach, Methods of Theoretical Physics, Vols. 1 and 2, McGraw-Hill,

New York, 1953.

I. Newton, Principia, Volume I, The Motion of Bodies, University of California Press,

Berkeley, 1934.

J. M. Ottino, The Kinematics of Mixing: Stretching, Chaos, and Transport, Cambridge

University Press, Cambridge, UK, 1989.

S. Paolucci, Continuum Mechanics and Thermodynamics of Matter, Cambridge University

Press, New York, 2016
R. L. Panton, Incompressible Flow, 4th edition, John Wiley, New York, 2013.

D. Pnueli and C. Gutfinger, Fluid Mechanics, Cambridge University Press, Cambridge, UK,
1992.

330
L. Prandtl and O. G. Tietjens, Fundamentals of Hydro and Aeromechanics, Dover, New
York, 1957.

L. Rosenhead, ed., Laminar Boundary Layers, Oxford University Press, Oxford, 1963.

R. H. Sabersky, A. J. Acosta, E. G. Hauptmann, and E. M. Gates, Fluid Flow: A First

Course in Fluid Mechanics, 4th edition, Prentice-Hall, Upper Saddle River, New Jersey,
1999.

M. Samimy, K. S. Breuer, L. G. Leal, and P. H. Steen, A Gallery of Fluid Motion, Cambridge

University Press, Cambridge, UK, 2004.

J. A. Schetz, Boundary Layer Analysis, Prentice-Hall, Englewood Cliffs, New Jersey, 1993.

H. M. Schey, Div, Grad, Curl, and All That, 2nd Ed., W.W. Norton, London, 1992.

H. Schlichting and K. Gersten, Boundary Layer Theory, 8th edition, McGraw-Hill, New
York, 2000.

L. I. Sedov, Similarity and Dimensional Methods in Mechanics, Academic Press, New York,
1959.

L. A. Segel, Mathematics Applied to Continuum Mechanics, Dover, New York, 1987.

A. H. Shapiro, The Dynamics and Thermodynamics of Compressible Fluid Flow, Volume I,

John Wiley, New York, 1953.

A. H. Shapiro, The Dynamics and Thermodynamics of Compressible Fluid Flow, Volume II,
Krieger, Malabar, Florida, 1954.

F. S. Sherman, Viscous Flow, McGraw-Hill, New York, 1990.

J. Smoller, Shock Waves and Reaction-Diffusion Equations, 2nd edition, Springer, New
York, 1994.

J. H. Spurk, Fluid Mechanics: Problems and Solutions, Springer, New York, 1997.

G. G. Stokes, Mathematical and Physical Papers, Johnson Reprint Corporation, New York,
1966.

J. W. Strutt (Lord Rayleigh), The Theory of Sound, Vols. 1 and 2, Dover, New York, 1945.

G. I. Taylor, Scientific Papers, Cambridge University Press, Cambridge, UK, 1958.

J. C. Tannehill, D. A. Anderson, and R. H. Pletcher, Computational Fluid Mechanics and

Heat Transfer, 3rd edition, CRC, Boca Raton, Florida, 2013.

331
W. Thomson (Lord Kelvin), Mathematical and Physical Papers, Cambridge University
Press, Cambridge, UK, 1911.

E. F. Toro, Riemann Solvers and Numerical Methods for Fluid Dynamics, Springer, New
York, 1999.

D. J. Tritton, Physical Fluid Dynamics, 2nd edition, Oxford University Press, Oxford, 1988.

C. A. Truesdell, A First Course in Rational Continuum Mechanics, Vol. 1, 2nd edition,

Academic Press, Boston, 1991.

M. Van Dyke, Perturbation Methods in Fluid Mechanics, Parabolic Press, Stanford, Cali-
fornia, 1975.

M. Van Dyke, An Album of Fluid Motion, Parabolic Press, Stanford, California, 1982.

W. G. Vincenti and C. H. Kruger, Introduction to Physical Gas Dynamics, Wiley, New

York, 1965.

R. von Mises, Theory of Flight, McGraw-Hill, New York, 1945.

R. von Mises, Mathematical Theory of Compressible Flow, Dover, Mineola, New York, 2004.

J. von Neumann, Collected Works, Pergamon, New York, 1961.

S. Whitaker, Introduction to Fluid Mechanics, Krieger, Malabar, Florida, 1968.

F. M. White, Viscous Fluid Flow, 3rd edition, McGraw-Hill, New York, 2006.

G. B. Whitham, Linear and Nonlinear Waves, John Wiley, New York, 1974.

L. C. Woods, The Thermodynamics of Fluid Systems, Clarendon Press, Oxford, 1975.

C.-S. Yih, Fluid Mechanics, West River Press, East Hampton, Connecticut, 1977.

Y. B. Zeldovich and Y. P. Raizer, Physics of Shock Waves and High-Temperature Hydrody-

namic Phenomena, Dover, New York, 2002.

R. Kh. Zeytounian, Theory and Applications of Nonviscous Fluid Flows, Springer, Berlin,
2002.

R. Kh. Zeytounian, Theory and Applications of Viscous Fluid Flows, Springer, Berlin, 2004.

M. J. Zucrow and J. D. Hoffman, Gas Dynamics, Vol. I, John Wiley, New York, 1976.

332