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 last updated August 19, 2003

Contents
1 Governing equations 1.1 Philosophy of rational continuum mechanics . . . . . . . . . . . . . . . . . . 1.1.1 Approaches to uid mechanics . . . . . . . . . . . . . . . . . . . . . . 1.1.2 Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1.3 Continuum mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1.4 Rational continuum mechanics . . . . . . . . . . . . . . . . . . . . . . 1.1.5 Notions from Newtonian continuum mechanics . . . . . . . . . . . . . 1.2 Some necessary mathematics . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.1 Vectors and Cartesian tensors . . . . . . . . . . . . . . . . . . . . . . 1.2.1.1 Gibbs and Cartesian Index notation . . . . . . . . . . . . . 1.2.1.2 Rotation of axes . . . . . . . . . . . . . . . . . . . . . . . . 1.2.1.3 Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.1.4 Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.1.4.1 Denition . . . . . . . . . . . . . . . . . . . . . . . 1.2.1.4.2 Alternating unit tensor . . . . . . . . . . . . . . . . 1.2.1.4.3 Transpose, symmetric, antisymmetric, and decomposition . . . . . . . . . . . . . . . . . . . . . . . . 1.2.1.4.3.1 Transpose . . . . . . . . . . . . . . . . . . . 1.2.1.4.3.2 Symmetric . . . . . . . . . . . . . . . . . . . 1.2.1.4.3.3 Antisymmetric . . . . . . . . . . . . . . . . 1.2.1.4.3.4 Decomposition . . . . . . . . . . . . . . . . 1.2.1.4.4 Tensor inner product . . . . . . . . . . . . . . . . . 1.2.1.4.5 Dual vector of a tensor . . . . . . . . . . . . . . . . 1.2.1.4.6 Tensor product: two tensors . . . . . . . . . . . . . 1.2.1.4.7 Vector product: vector and tensor . . . . . . . . . . 1.2.1.4.7.1 Pre-multiplication . . . . . . . . . . . . . . 1.2.1.4.7.2 Post-multiplication . . . . . . . . . . . . . . 1.2.1.4.8 Dyadic product: two vectors . . . . . . . . . . . . . 1.2.1.4.9 Contraction . . . . . . . . . . . . . . . . . . . . . . 1.2.1.4.10 Vector cross product . . . . . . . . . . . . . . . . . 1.2.1.4.11 Vector associated with a plane . . . . . . . . . . . . 1.2.2 Eigenvalues and eigenvectors . . . . . . . . . . . . . . . . . . . . . . . 1.2.3 Grad, div, curl, etc. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.3.1 Gradient operator . . . . . . . . . . . . . . . . . . . . . . . 3 13 13 13 14 15 15 16 18 18 18 20 24 25 25 26 26 26 27 27 27 27 28 28 29 29 29 29 29 30 30 31 37 37

1.3

1.4

Divergence operator . . . . . . . . . . . . . . Curl operator . . . . . . . . . . . . . . . . . . Laplacian operator . . . . . . . . . . . . . . . Relevant theorems . . . . . . . . . . . . . . . 1.2.3.5.1 Fundamental theorem of calculus . . 1.2.3.5.2 Gausss theorem . . . . . . . . . . . 1.2.3.5.3 Stokes theorem . . . . . . . . . . . . 1.2.3.5.4 Leibnizs theorem . . . . . . . . . . . 1.2.3.5.5 Reynolds transport theorem . . . . . Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.3.1 Lagrangian description . . . . . . . . . . . . . . . . . . 1.3.2 Eulerian description . . . . . . . . . . . . . . . . . . . 1.3.3 Material derivatives . . . . . . . . . . . . . . . . . . . . 1.3.4 Streamlines . . . . . . . . . . . . . . . . . . . . . . . . 1.3.5 Pathlines . . . . . . . . . . . . . . . . . . . . . . . . . 1.3.6 Streaklines . . . . . . . . . . . . . . . . . . . . . . . . . 1.3.7 Kinematic decomposition of motion . . . . . . . . . . . 1.3.7.1 Translation . . . . . . . . . . . . . . . . . . . 1.3.7.2 Solid body rotation and straining . . . . . . . 1.3.7.2.1 Solid body rotation . . . . . . . . . . 1.3.7.2.2 Straining . . . . . . . . . . . . . . . 1.3.7.2.2.1 Extensional straining . . . . . 1.3.7.2.2.2 Shear straining . . . . . . . . 1.3.7.2.2.3 Principal axes of strain rate . 1.3.8 Expansion rate . . . . . . . . . . . . . . . . . . . . . . 1.3.9 Invariants of the strain rate tensor . . . . . . . . . . . 1.3.10 Two-dimensional kinematics . . . . . . . . . . . . . . . 1.3.10.1 General two-dimensional ows . . . . . . . . . 1.3.10.1.1 Rotation . . . . . . . . . . . . . . . . 1.3.10.1.2 Extension . . . . . . . . . . . . . . . 1.3.10.1.3 Shear . . . . . . . . . . . . . . . . . 1.3.10.1.4 Expansion . . . . . . . . . . . . . . . 1.3.10.2 Relative motion along 1 axis . . . . . . . . . . 1.3.10.3 Relative motion along 2 axis . . . . . . . . . . 1.3.10.4 Uniform ow . . . . . . . . . . . . . . . . . . 1.3.10.5 Pure rigid body rotation . . . . . . . . . . . . 1.3.10.6 Pure extensional motion (a compressible ow) 1.3.10.7 Pure shear straining . . . . . . . . . . . . . . 1.3.10.8 Couette ow: shear + rotation . . . . . . . . 1.3.10.9 Ideal point vortex: extension+shear . . . . . . Conservation axioms . . . . . . . . . . . . . . . . . . . . . . . 1.4.1 Mass . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.2 Linear momenta . . . . . . . . . . . . . . . . . . . . . . 1.4.2.1 Statement of the principle . . . . . . . . . . . 1.4.2.2 Surface forces . . . . . . . . . . . . . . . . . . 4

1.2.3.2 1.2.3.3 1.2.3.4 1.2.3.5

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

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

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

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

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

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

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

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

39 39 40 40 40 40 41 41 42 42 42 43 43 44 46 46 49 50 50 50 51 51 51 51 53 54 54 54 54 55 55 55 56 57 58 59 60 61 62 63 64 65 67 67 68

1.5

1.4.2.3 Final form of linear momenta equation . . . . . . . . . . . . Angular momenta . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.4.1 Total energy term . . . . . . . . . . . . . . . . . . . . . . . 1.4.4.2 Work term . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.4.3 Heat transfer term . . . . . . . . . . . . . . . . . . . . . . . 1.4.4.4 Conservative form of the energy equation . . . . . . . . . . 1.4.4.5 Secondary forms of the energy equation . . . . . . . . . . . 1.4.4.5.1 Mechanical energy equation . . . . . . . . . . . . . 1.4.4.5.2 Thermal energy equation . . . . . . . . . . . . . . . 1.4.4.5.3 Non-conservative energy equation . . . . . . . . . . 1.4.4.5.4 Energy equation in terms of enthalpy . . . . . . . . 1.4.4.5.5 Energy equation in terms of entropy (not the second law!) . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.5 Entropy inequality . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.6 Summary of axioms in dierential form . . . . . . . . . . . . . . . . . 1.4.6.1 Conservative form . . . . . . . . . . . . . . . . . . . . . . . 1.4.6.1.1 Cartesian index form . . . . . . . . . . . . . . . . . 1.4.6.1.2 Gibbs form . . . . . . . . . . . . . . . . . . . . . . 1.4.6.1.3 Non-orthogonal index form . . . . . . . . . . . . . 1.4.6.2 Non-conservative form . . . . . . . . . . . . . . . . . . . . . 1.4.6.2.1 Cartesian index form . . . . . . . . . . . . . . . . . 1.4.6.2.2 Gibbs form . . . . . . . . . . . . . . . . . . . . . . 1.4.6.2.3 Non-orthogonal index form . . . . . . . . . . . . . 1.4.6.3 Physical interpretations . . . . . . . . . . . . . . . . . . . . 1.4.7 Complete system of equations? . . . . . . . . . . . . . . . . . . . . . 1.4.8 Integral forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.8.1 Mass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.8.1.1 Fixed region . . . . . . . . . . . . . . . . . . . . . . 1.4.8.1.2 Material region . . . . . . . . . . . . . . . . . . . . 1.4.8.1.3 Moving solid enclosure with holes . . . . . . . . . . 1.4.8.2 Linear momenta . . . . . . . . . . . . . . . . . . . . . . . . 1.4.8.3 Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4.8.4 General expression . . . . . . . . . . . . . . . . . . . . . . . Constitutive equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.5.1 Frame and material indierence . . . . . . . . . . . . . . . . . . . . . 1.5.2 Second law restrictions and Onsager relations . . . . . . . . . . . . . 1.5.2.1 Weak form of the Clausius-Duhem inequality . . . . . . . . 1.5.2.1.1 Non-physical motivating example . . . . . . . . . . 1.5.2.1.2 Real physical eects. . . . . . . . . . . . . . . . . . 1.5.2.2 Strong form of the Clausius-Duhem inequality . . . . . . . . 1.5.3 Fouriers law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.5.4 Stress-strain rate relation for a Newtonian uid . . . . . . . . . . . . 1.5.4.1 Underlying experiments . . . . . . . . . . . . . . . . . . . . 1.5.4.2 Analysis for isotropic Newtonian uid . . . . . . . . . . . . 1.4.3 1.4.4 5

72 73 75 75 76 76 77 77 78 78 79 79 80 81 84 84 84 84 85 85 85 86 86 86 87 87 88 88 88 89 90 91 91 91 92 92 92 92 93 94 94 96 96 98

1.5.4.2.1 Diagonal component . . . . . . . . . . 1.5.4.2.2 O-diagonal component . . . . . . . . 1.5.4.3 Stokes assumption . . . . . . . . . . . . . . . . 1.5.4.4 Second law restrictions . . . . . . . . . . . . . . 1.5.4.4.1 One dimensional systems . . . . . . . . 1.5.4.4.2 Two dimensional systems . . . . . . . 1.5.4.4.3 Three dimensional systems . . . . . . . 1.5.5 Equations of state . . . . . . . . . . . . . . . . . . . . . . 1.6 Boundary and interface conditions . . . . . . . . . . . . . . . . . 1.7 Complete set of compressible Navier-Stokes equations . . . . . . 1.7.0.1 Conservative form . . . . . . . . . . . . . . . . 1.7.0.1.1 Cartesian index form . . . . . . . . . . 1.7.0.1.2 Gibbs form . . . . . . . . . . . . . . . 1.7.0.2 Non-conservative form . . . . . . . . . . . . . . 1.7.0.2.1 Cartesian index form . . . . . . . . . . 1.7.0.2.2 Gibbs form . . . . . . . . . . . . . . . 1.8 Incompressible Navier-Stokes equations with constant properties 1.8.1 Mass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.8.2 Linear momenta . . . . . . . . . . . . . . . . . . . . . . . 1.8.3 Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.8.4 Summary of incompressible constant property equations 1.8.5 Limits for one-dimensional diusion . . . . . . . . . . . . 1.9 Dimensionless compressible Navier-Stokes equations . . . . . . . 1.9.1 Mass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.9.2 Linear momenta . . . . . . . . . . . . . . . . . . . . . . . 1.9.3 Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.9.4 Thermal state equation . . . . . . . . . . . . . . . . . . . 1.9.5 Caloric state equation . . . . . . . . . . . . . . . . . . . 1.9.6 Upstream conditions . . . . . . . . . . . . . . . . . . . . 1.9.7 Reduction in parameters . . . . . . . . . . . . . . . . . . 1.10 First integrals of linear momentum . . . . . . . . . . . . . . . . 1.10.1 Bernoullis equation . . . . . . . . . . . . . . . . . . . . . 1.10.1.1 Irrotational case . . . . . . . . . . . . . . . . . 1.10.1.2 Steady case . . . . . . . . . . . . . . . . . . . . 1.10.1.2.1 Streamline integration . . . . . . . . . 1.10.1.2.2 Lamb surfaces . . . . . . . . . . . . . . 1.10.1.3 Irrotational, steady, incompressible case . . . . 1.10.2 Croccos theorem . . . . . . . . . . . . . . . . . . . . . . 1.10.2.1 Stagnation enthalpy variation . . . . . . . . . . 1.10.2.2 Extended Croccos theorem . . . . . . . . . . . 1.10.2.3 Traditional Croccos theorem . . . . . . . . . . 6

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

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

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

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

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

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

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

106 107 107 108 108 108 109 111 114 114 114 114 115 115 115 115 116 116 117 117 118 118 119 121 121 123 124 124 125 125 125 125 127 127 127 128 128 129 129 131 132

2 Vortex dynamics 133 2.1 Transformations to cylindrical coordinates . . . . . . . . . . . . . . . . . . . 133 2.1.1 Centripetal and Coriolis acceleration . . . . . . . . . . . . . . . . . . 134 2.1.2 Grad and div for cylindrical systems . . . . . . . . . . . . . . . . . . 137 2.1.2.1 Grad . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138 2.1.2.2 Div . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138 2.1.3 Incompressible Navier-Stokes equations in cylindrical coordinates . . 139 2.2 Ideal rotational vortex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139 2.3 Ideal irrotational vortex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143 2.4 Helmholtz vorticity transport equation . . . . . . . . . . . . . . . . . . . . . 145 2.4.1 General development . . . . . . . . . . . . . . . . . . . . . . . . . . . 145 2.4.2 Limiting cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147 2.4.2.1 Incompressible with conservative body force . . . . . . . . . 147 2.4.2.2 Incompressible with conservative body force, isotropic, Newtonian, constant viscosity . . . . . . . . . . . . . . . . . . . 147 2.4.2.3 Two-dimensional, incompressible with conservative body force, isotropic, Newtonian, constant viscosity . . . . . . . . . . . 148 2.4.3 Physical interpretations . . . . . . . . . . . . . . . . . . . . . . . . . 148 2.4.3.1 Baroclinic (non-barotropic) eects . . . . . . . . . . . . . . 148 2.4.3.2 Bending and stretching of vortex tubes: three-dimensional eects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148 2.5 Kelvins circulation theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 151 2.6 Potential ow of ideal point vortices . . . . . . . . . . . . . . . . . . . . . . . 152 2.6.1 Two interacting ideal vortices . . . . . . . . . . . . . . . . . . . . . . 153 2.6.2 Image vortex . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154 2.6.3 Vortex sheets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154 2.6.4 Potential of an ideal vortex . . . . . . . . . . . . . . . . . . . . . . . 156 2.6.5 Interaction of multiple vortices . . . . . . . . . . . . . . . . . . . . . 157 2.6.6 Pressure eld . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159 2.6.6.1 Single stationary vortex . . . . . . . . . . . . . . . . . . . . 159 2.6.6.2 Group of N vortices . . . . . . . . . . . . . . . . . . . . . . 159 2.7 Inuence of walls . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160 2.7.1 Streamlines and vortex lines at walls . . . . . . . . . . . . . . . . . . 160 2.7.2 Generation of vorticity at walls . . . . . . . . . . . . . . . . . . . . . 162 3 One-dimensional compressible ow 3.1 Generalized one-dimensional equations . . . . . 3.1.1 Mass . . . . . . . . . . . . . . . . . . . . 3.1.2 Momentum . . . . . . . . . . . . . . . . 3.1.3 Energy . . . . . . . . . . . . . . . . . . . 3.1.4 Summary of equations . . . . . . . . . . 3.1.4.1 Unsteady conservative form . . 3.1.4.2 Unsteady non-conservative form 3.1.4.3 Steady conservative form . . . 3.1.4.4 Steady non-conservative form . 7 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165 165 166 167 169 172 172 172 172 172

3.2

3.3

3.4

3.5

3.1.5 Inuence coecients . . . . . . . . . . . . Flow with area change . . . . . . . . . . . . . . . 3.2.1 Isentropic Mach number relations . . . . . 3.2.2 Sonic properties . . . . . . . . . . . . . . . 3.2.3 Eect of area change . . . . . . . . . . . . 3.2.4 Choking . . . . . . . . . . . . . . . . . . . Normal shock waves . . . . . . . . . . . . . . . . 3.3.1 Rankine-Hugoniot equations . . . . . . . . 3.3.2 Rayleigh line . . . . . . . . . . . . . . . . 3.3.3 Hugoniot curve . . . . . . . . . . . . . . . 3.3.4 Solution procedure for general equations of 3.3.5 Calorically perfect ideal gas solutions . . . 3.3.6 Acoustic limit . . . . . . . . . . . . . . . . Flow with area change and normal shocks . . . . 3.4.1 Converging nozzle . . . . . . . . . . . . . . 3.4.2 Converging-diverging nozzle . . . . . . . . Rarefactions and the method of characteristics . . 3.5.1 Inviscid one-dimensional equations . . . . 3.5.2 Homeoentropic ow of an ideal gas . . . . 3.5.3 Simple waves . . . . . . . . . . . . . . . . 3.5.4 Prandtl-Meyer rarefaction . . . . . . . . . 3.5.5 Simple compression . . . . . . . . . . . . . 3.5.6 Two interacting expansions . . . . . . . . 3.5.7 Wall interactions . . . . . . . . . . . . . . 3.5.8 Shock tube . . . . . . . . . . . . . . . . . 3.5.9 Final note on method of characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

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

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

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

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

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

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

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

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

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

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

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

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

175 176 176 179 180 182 183 185 186 187 188 188 194 195 195 196 198 198 203 205 208 210 210 210 210 210 217 218 220 220 221 222 224 224 224 225 226 229 230 231 232 236 236 236

4 Potential ow 4.1 Stream functions and velocity potentials . 4.2 Mathematics of complex variables . . . . . 4.2.1 Eulers formula . . . . . . . . . . . 4.2.2 Polar and Cartesian representations 4.2.3 Cauchy-Riemann equations . . . . 4.3 Elementary complex potentials . . . . . . 4.3.1 Uniform ow . . . . . . . . . . . . 4.3.2 Sources and sinks . . . . . . . . . . 4.3.3 Point vortices . . . . . . . . . . . . 4.3.4 Superposition of sources . . . . . . 4.3.5 Flow in corners . . . . . . . . . . . 4.3.6 Doublets . . . . . . . . . . . . . . . 4.3.7 Rankine half body . . . . . . . . . 4.3.8 Flow over a cylinder . . . . . . . . 4.4 More complex variable theory . . . . . . . 4.4.1 Contour integrals . . . . . . . . . . 4.4.1.1 Simple pole . . . . . . . . 8

4.5 4.6 4.7 4.8

4.4.1.2 Constant potential . 4.4.1.3 Uniform ow . . . . 4.4.1.4 Quadrapole . . . . . 4.4.2 Laurent series . . . . . . . . . Pressure distribution for steady ow . Blasius force theorem . . . . . . . . . Kutta-Zhukovsky lift theorem . . . . Conformal mapping . . . . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

236 236 237 237 238 238 241 244 245 245 245 255 259 260 272 289

5 Viscous incompressible laminar ow 5.1 Fully developed, one dimensional solutions . . 5.1.1 Pressure gradient driven ow in a slot . 5.1.2 Couette ow with pressure gradient . . 5.2 Similarity solutions . . . . . . . . . . . . . . . 5.2.1 Stokes rst problem . . . . . . . . . . 5.2.2 Blasius boundary layer . . . . . . . . . Bibliography

10

Preface
These are lecture notes for AME 538 Intermediate Fluid Mechanics, taught at the Department 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 uid mechanics, including a rigorous derivation of the compressible Navier-Stokes equations, vorticity dynamics, compressible ow, potential ow, and viscous laminar ow. While there is a good deal of rigor in the development here, it is not absolute. It is not hard to nd 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 development and notation for the governing equations closely follows Panton 1 , who I nd 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 specic 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.538. 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. Additionally many enhancements to those original written notes have been made as the course was taught in Fall 2000. Thanks go to Prof. Patrick F. Dunn of the University of Notre Dame, who while teaching AME 538 in Fall 2001, discovered several minor errors in an earlier version of these notes. These have been xed. Additional modications were made in the most recent semester in which these notes were used, Fall 2002. 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 August 19, 2003 Copyright c 2003 by Joseph M. Powers. All rights reserved.
1

R. L. Panton, Incompressible Flow, 2nd edition, John Wiley, New York, 1995.

11

12

Chapter 1 Governing equations

see Panton, Chapters 1-6, see Yih, Chapters 1-3, Appendix 1-2, see Aris.

1.1
1.1.1

Philosophy of rational continuum mechanics

Approaches to uid mechanics

We seek here to present an approach to uid mechanics founded on the principles of rational continuum mechanics. There are many paths to understanding uid mechanics, and good arguments can be made for each. A typical rst 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 connes 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 ows 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 uid 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 developed 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 suciently large so that the micro- and nanoscopic structure of these materials was unimportant. For uids (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 diculty of the Truesdellian approach is that it is burdened with a dicult notation and tends to become embroiled in proofs and philosophy, which while ultimately useful, can preclude learning basic uid mechanics in the time scale of the human lifetime.
Cliord Ambrose Truesdell, III, 1919-2000, American continuum mechanician and natural philosopher. Taught at Indiana and and Johns Hopkins Universities.
1

13

In this course, we will attempt to steer a course between the pragmatism of undergraduate uid 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 understanding of uid behavior. We shall rst spend some time carefully developing the governing equations for a compressible viscous uid. 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 inuence the behavior of uids.

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 dened 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 sucient to describe our universe. When we nd an axiom to be inconsistent with observation, it must be modied or eliminated. A familiar example of this is the Michelson2 Morley 3 experiment, which motivated Einstein 4 to modify the Newtonian 5 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, motion is the consequence of forces, study of forces on a body is dynamics.
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 theory of relativity and made fundamental contributions to quantum mechanics and Brownian motion in uid mechanics; spent later life in the United States. 5 Sir Isaac Newton, 1642-1727, English physicist and mathematician and chief gure of the scientic revolution of the seventeenth and eighteenth centuries. Developed calculus, theories of gravitation and motion of bodies, and optics. Educated at Cambridge 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.
2

14

There are many subsets of mechanics, e.g. statistical mechanics, quantum mechanics, continuum mechanics, uid 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 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 nite 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 nite number of ordinary dierential 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 innite number 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 innitesimal property variation from point to point in space. While variations are generally continuous, nite numbers 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 nite number of partial dierential equations in which the properties of the continuum material are functions of both space and time. It is possible to show that a partial dierential equation can be thought of as an innite number of ordinary dierential equations, so this is consistent with our model of a continuum as an innite number of particles.

1.1.4

Rational continuum mechanics

The modier rational was rst applied by Truesdell to continuum mechanics to distinguish 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 Euclid 7 used for his geometry: formal denitions, axioms, and theorems, all accompanied by careful language and proofs. This course will generally follow the less formal, albeit still rigorous, approach of Pantons text, including the adoption of much of Pantons notation.
Leonhard Euler, 1707-1783, Swiss-born mathematician and physicist who served in the court of Catherine I of Russia in St. Petersburg, regarded by many as one of the greatest mechanicians. 7 Euclid, Greek geometer of profound inuence who taught in Alexandria, Egypt, during the reign of Ptolemy I Soter, who ruled 323-283 BC.
6

15

1.1.5

Notions from Newtonian continuum mechanics

The following are useful notions from Newtonian continuum mechanics. Here we use Newtonian 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 species 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 v o = uo i + vo j + wo k, the Galilean transformation (x, y, z, t) (x , y , z , t ) is as follows x y z t = = = = x uo t, y vo t, z wo t, t. (1.1) (1.2) (1.3) (1.4)

Control volumes are useful; we will study three varieties: Fixed-constant in space, Material-no ux of mass through boundaries, can deform, Arbitrary-can move, can deform, can have dierent uid 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 dene density as = lim
N i=1

mi

V 0

(1.5)

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 dene a length scale L associated with the volume V to be L = V 1/3 . In the world in which we live, we expect the density to approach a limiting value, but only until a nite value of V . When V becomes too small, such that only a few molecules are contained within it, we expect wild deviations. 16

0.5 x

1 -0.5

-1
Figure 1.1: Variation of density with position x from sample generic continuum-based axiom. Suppose for example that our axioms of continuum mechanics for a particular problem, gave rise to the model equation for density variation of d2 + k 2 = 0, dx2 (1.6)

where k is some constant physical parameter. Then solution of the elementary dierential equation is found to be = C1 sin kx + C2 cos kx. (1.7)

A possible variation of with x is shown in Figure 1.1. Equation (1.7) predicts a sinusoidal variation of with a wavelength = 2/k . For the continuum assumption to be valid in a gas, we must expect to have many particle collisions to take place within a box with a characteristic length of L = . For gases, the important dividing line between continuous and non-continuous behavior is the mean free path m , that is the mean length which exists between molecular collisions. For gases under typical atmospheric conditions we nd from observation that
m

0.1 m.
8

(1.8)

So for our example the continuum assumption is valid i 2 >> 0.1 m, k or k <<

2 . 0.1 m

(1.9)

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
8

Here i is a common shorthand for if and only if.

17

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, (1.10) 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 Gibbs 9 notation. We will soon study an alternate notation, developed by Einstein, and known as Cartesian 10 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 every point in space. An example is the viscous stress 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) zx (x, y, z, t) zy (x, y, z, t) zz (x, y, z, t)

(1.11)

1.2

Some necessary mathematics

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

1.2.1
1.2.1.1

Vectors and Cartesian tensors

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 briey 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 dicult at the outset, in the end many agree
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 Ren e Descartes, 1596-1650, French mathematician and philosopher of great inuence. A great doubter of existence who nevertheless concluded, I think, therefore I am. Developed analytic geometry.
9

18

that the use of index notation actually simplies many common notions in uid mechanics. Moreover, its use in the archival literature is widespread, so to really be conversant in uid mechanics, one must know the index notation. The following table summarizes the correspondences between Gibbs, Cartesian index, and matrix notation. Quantity zeroth order tensor rst order tensor Common Parlance scalar vector Gibbs a a Cartesian Index a ai

Matrix (a) a1 a2 . . . an a12 . . . a1n a22 . . . a2n . . . . . . . . . an2 . . . ann

a11

second order tensor third order tensor fourth order tensor . . .

tensor tensor tensor . . .

A A A . . .

aij aijk aijkl . . .

a21 . . .

an1

Here we adopt a convention for the Gibbs notation, which we will nd at times conicts with other conventions, in which times font italics indicates a scalar, bold times font indicates a vector, upper case sans serif indicates a second order tensor, overlined upper case sans serif indicates a third order tensor, double overlined upper case sans serif 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. (1.12)

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 a1 a = a 1 e1 + a 2 e2 + a 3 e3 = ai ei = a i ei = a i = a2 . i=1 a3
3

(1.13)

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 nal 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 19

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 Cartesian 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 coordinate transformations, and Einstein did just that as well. For our purposes however, the simpler Cartesian index notation will suce. 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 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 1.2. Note that it is easy to show that the angle = /2 . Here a point P is identied by a particular set of coordinates (x 1 , 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 dierently 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
x 1 = x1 cos + x2 cos .

(1.14)

More generally, we can say for an arbitrary point that x1 = x1 cos + x2 cos . We adopt the following notation (x1 , x1 ) denotes the angle between the x1 and x1 axes, (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, . . . 20 (1.15)

x2

x2 x* 1 = x* 2 cos 1 cos + x*
P

x* 2

x* 1

x1

x* 1

x1

Figure 1.2: Sketch of coordinate transformation which is a rotation of axes Thus, in two-dimensions, we have x1 = x1 cos(x1 , x1 ) + x2 cos(x2 , x1 ). In three dimensions, this extends to x1 = x1 cos(x1 , x1 ) + x2 cos(x2 , x1 ) + x3 cos(x3 , x1 ). Extending this analysis to calculate x2 and x3 gives x2 = x1 cos(x1 , x2 ) + x2 cos(x2 , x2 ) + x3 cos(x3 , x2 ), x3 = x1 cos(x1 , x3 ) + x2 cos(x2 , x3 ) + x3 cos(x3 , x3 ). The above equations can be written in matrix form as ( x1 x2 x3 ) = ( x 1 x2 cos(x1 , x1 ) cos(x1 , x2 ) cos(x1 , x3 ) x3 ) cos(x2 , x1 ) cos(x2 , x2 ) cos(x2 , x3 ) cos(x3 , x1 ) cos(x3 , x2 ) cos(x3 , x3 )
12

(1.16)

(1.17)

(1.18) (1.19)

(1.20)

If we use the shorthand notation, for example, that 11 = cos(x1 , x1 ), we have

11 21 31 12 22 32 L 13 23 33

= cos(x1 , x2 ), etc., (1.21)

( x1

x2

x3 ) = ( x 1

x2

xT

xT 21

x3 )

In Gibbs notation, dening the matrix of s to be L, and recalling that all vectors are taken to be column vectors, we can alternatively say x T = xT L. 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 = LT x. 11 We call L = ij the matrix of direction cosines and LT = ji the rotation matrix. The matrix equation above is really a set of three linear equations. For instance, the rst is x1 = x1 11 + x2 21 + x3 31 . (1.22) More generally, we could say that xj = x 1
1j

+ x2

2j

+ x3

3j .

(1.23)

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 the expression further by writing
3

xj =
i=1

xi

ij .

(1.24)

This is commonly written in the following form: xj = x i

ij .

(1.25)

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 dene the Kronecker delta, ij as ij = This is eectively the identity matrix I: 1 0 0 ij = I = 0 1 0 . 0 0 1

0 1

i = j, i = j,

(1.26)

(1.27)

11 The widely used alternate convention of not explicitly using the transpose notation for vectors would instead have our x T = xT L written as x = x L. The alternate convention still typically applies, where necessary, the transpose notation for tensors, so it would also hold that x = LT x.

22

Direct substitution proves that what is eectively the law of cosines can be written as
ij kj

= ik .

(1.28)

Example 1.1
Show for the two-dimensional system described in Figure 1.2 that Expanding for the two-dimensional system, we get
i1 k1 ij kj

= ik holds.

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, This is obviously true. Next, take i = 1, k = 2. We get then
11 21

cos2 + sin2 = 1.

12 22

= 12

= 0, = 0, = 0.

cos cos(/2 ) + cos( + /2) cos() cos sin sin cos This is obviously true. Next, take i = 2, k = 1. We get then

+ 22 12 = 21 = 0, cos(/2 ) cos + cos cos(/2 + ) = 0,

21 11

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 nd the inverse transformation back to the unprimed coordinates via the following operations:
kj xj

kj xj ij xj

xi

= kj xi ij , = ij kj xi , = ik xi , = xk , = xi , = ij xj . 23

(1.29) (1.30) (1.31) (1.32) (1.33) (1.34)

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 . (1.35)

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 value 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 (1.36) under a rotation of axes characterized by direction cosines ij . In Gibbs notation, we would say v T = vT L, or alternatively v = LT 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 1.2
Consider the set of scalars which describe the velocity in a two dimensional Cartesian system: vi = vx vy ,

where we return to the typical x, y coordinate system. In a rotated coordinate system, using the same notation of Figure 1.1, we nd 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 satises the denition for a vector.

Example 1.3
Do two arbitrary scalars, say the quotient of pressure and density and the product of specic heat and temperature, (p/, cv T )T , form a vector? If this quantity is a vector, then we can say vi = p/ 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 rst component would represent the dierence between

24

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 v1 = v2 = p cos + cv T cos(/2 ). p cos(/2 + ) + cv T cos().

This quantity does not have any known physical signicance, 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) 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 uT v = b (Gibbs notation)

1.2.1.4.1 Denition A second order tensor, or a rank two tensor, is nine scalar components that under a rotation of axes transform according to the following rule: Tij =
ki lj Tkl .

(1.37)

Note we could also write this in an expanded form as

3 3 ki k =1 l=1 lj Tkl = k =1 l=1 3 3 ki Tkl lj .

Tij =

(1.38)

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 = LT T L. 25 (1.39)

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 rst subscript is associated with the face of a unit cube (hence the memory device, rst-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 twodimensional matrices, but not all matrices are rank two tensors, as they do not necessarily satisfy the transformation rules. We can say T11 Tij = T21 T31

T12 T22 T32

T13 T23 T33

(1.40)

The rst 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 dened as follows 1 if ijk = 123, 231, or 312, = 0 if any two indices identical, . (1.41) ijk 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 ,

(1.42)

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 Transpose, symmetric, antisymmetric, and decomposition

1.2.1.4.3.1 Transpose The transpose of a second rank tensor, denoted by a superscript T , is found by exchanging elements about the diagonal. In shorthand index notation, this is simply (Tij )T = Tji . (1.43) Written out in full, if T11 Tij = T21 T31

T12 T22 T32 26

T13 T23 , T33

(1.44)

then

T11 T Tij = Tji = T12 T13

T21 T22 T23

T31 T32 , T33

(1.45)

1.2.1.4.3.2

Symmetric A tensor Qij is symmetric i Qij = Qji . (1.46)

Note that a symmetric tensor has only six independent scalars. 1.2.1.4.3.3 Antisymmetric A tensor Rij is antisymmetric i Rij = Rji . (1.47)

Note that an antisymmetric tensor must have zeroes on its diagonal, and only three independent scalars on o-diagonal elements. 1.2.1.4.3.4 Decomposition An arbitrary tensor Tij can be separated into a symmetric and antisymmetric pair of tensors: 1 1 1 1 Tij = Tij + Tij + Tji Tji . 2 2 2 2 Rearranging, we get Tij = 1 1 (Tij + Tji ) + (Tij Tji ) . 2 2
symmetric antisymmetric

(1.48)

(1.49)

The rst term must be symmetric, and the second term must be antisymmetric. This is easily seen by considering applying this to any matrix of actual numbers. If we dene the symmetric part of the matrix Tij by the following notation 1 (Tij + Tji ) , 2 and the antisymmetric part of the same matrix by the following notation T(ij ) = T[ij ] = we then have Tij = T(ij ) + T[ij ] . (1.52) 1 (Tij Tji ) , 2 (1.50)

(1.51)

1.2.1.4.4 Tensor inner product The tensor inner product of two tensors Tij and Sji is dened as follows Tij Sji = a, (1.53) where a is a scalar. In Gibbs notation, we would say T : S = a. (1.54)

It is easily shown, and will be important in upcoming derivations, that the tensor inner product of any symmetric tensor with any antisymmetric tensor is the scalar zero. 27

1.2.1.4.5 follows

Dual vector of a tensor We dene the dual vector, di , of a tensor Tjk as di =

ijk Tjk

ijk T(jk )

ijk T[jk ] .

(1.55)

The term ijk is antisymmetric for any xed 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 di =
ijk T[jk ] .

(1.56)

Lets nd the inverse relation for di , Starting with Eq. (1.55), we take the inner product of di with ilm to get (1.57) ilm di = ilm ijk Tjk Employing Eq. (1.42) to eliminate the s in favor of s, we get
ilm di

= (lj mk lk mj ) Tjk = Tlm Tml , = 2T[lm] .

(1.58) (1.59) (1.60)

1 (1.61) ilm di . 2 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 antisymmetric part: T[lm] = Tij
arbitrary tensor

Hence,

T(ij )
symmetric part

1 2

kij dk

(1.62)

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 = Rik . (1.63) 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 = R. (1.64) 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 = 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. 28

But, in Cartesian index notation, the order of the indices is extremely important, and it is this order that does not commute: Sij Tjk = 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 nd 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 avors, pre-multiplication and postmultiplication, both important, and given in Cartesian index and Gibbs notation below: 1.2.1.4.7.1 Pre-multiplication uj = vi Tij = Tij vi , u = vT T = T v. (1.65) (1.66)

In the Cartesian index notation above the rst form is preferred as it has a correspondence with the Gibbs notation, but both are correct representations given our summation convention. 1.2.1.4.7.2 Post-multiplication wi = Tij vj = vj Tij , w = T v = vT T. (1.67) (1.68)

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 , T = uvT = vuT .

(1.69) (1.70)

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 dierent, 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 . (1.71)

So in this case the contraction yields a scalar. In matrix algebra, this particular contraction is the trace of the matrix. 29

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

wi = ijk uj vk , w = u v. Expanding for i = 1, 2, 3 gives w1 = w2 = w3 =

123 u2 v3

(1.72) (1.73)

+ 231 u3 v1 + 312 u1 v2 +

132 u3 v2

= u2 v3 u3 v2 , 213 u1 v3 = u3 v1 u1 v3 , 321 u2 v1 = u1 v2 u2 v1 .

(1.74) (1.75) (1.76)

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 dened so that the rst 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 , tT = nT T, t = TT n. (1.77) (1.78) (1.79)

A sketch of a Cartesian element with the tensor components sketched on the proper face is shown in 1.3. Example 1.4
For example, if we want to know the vector associated with the 1 face, t (1) , as shown in Figure 1.3, T we rst choose the unit normal associated with the x1 face, which is the vector ni = (1, 0, 0) . The associated vector is found by doing the actual summation tj = ni Tij = n1 T1j + n2 T2j + n3 T3j . Now n1 = 1, n2 = 0, and n3 = 0, so for this problem, we have tj
(1)

(1.80)

= T1j .

(1.81)

30

x3 t (3) 33 31 32 23 13 12 11 t
(1)

(2)

21

22 x2

x1

Figure 1.3: Sample Cartesian element which is aligned with coordinate axes, along with tensor components and vectors associated with each face.

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 nding a rotation as sketched in 1.4. Mathematically, we can enforce this condition by requiring that ni Tij
vector associated with chosen direction

nj
scalar multiple of chosen direction

(1.82)

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 . (1.83)

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. 31

x3 t (3) 33 31 32 23 13 12 11 t
(1)

(3 )

rotate

(2)

21

22 x2
t x
1 (1 )

(2

x1

Figure 1.4: 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. We can rearrange to form In matrix notation, this can be written as ( n1 n2

ni (Tij ij ) = 0.

(1.84)

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 coecient matrix be zero. As we have an unknown parameter , we have sucient degrees of freedom to accomplish this. So we require T11 T12 T13 T21 T22 T23 =0 T31 T32 T33 (1.86)

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

(1.85)

We know from linear algebra that such an equation for a third order matrix gives rise to a characteristic polynomial for of the form 12 3 I (1) 2 + I (2) I (3) = 0, (1.87)

where I (1) , I (2) , I (3) are scalars which are functions of all the scalars Tij . The I s are known as the invariants of the tensor Tij . They can be shown to be given by 13 I (1) = Tii = tr(T), 1 1 I (2) = (tr(T))2 tr(T T) = det(T)tr T1 , (Tii Tjj Tij Tji ) = 2 2 I (3) = ijk T1i T2j T3k = det (T) .
12 13

(1.88) (1.89) (1.90)

We employ a slightly more common form here than the very similar Eq. (3.10.4) of Panton. Note the obvious error in the third of Pantons Eq. (3.10.5), where the indices j and q appear three times.

32

x3

x2

x1

Figure 1.5: Sketch of stresses being applied to a cubical uid 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. 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 I (1) = (1) + (2) + (3) , I (2) = (1) (2) + (2) (3) + (3) (1) , I (3) = (1) (2) (3) . (1.91) (1.92) (1.93)

In general these eigenvalues, and consequently, the eigenvectors are complex. Additionally, in general the eigenvectors are non-orthogonal. If, however, the matrix we are considering is symmetric, which is often the case in uid 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 uid element. The eigenvalues themselves give the value of the principal stress. This is actually a generalization of the familiar Mohrs circle from solid mechanics. Example 1.5
Find the principal axes and principal values of stress if the stress tensor is 1 0 0 Tij = 0 1 2 . 0 2 1 ni Tij ni (Tij ij ) = nj , = ni ij , = 0.

(1.94)

A sketch of these stresses is shown on the uid element in Figure 1.5. We take the eigenvalue problem (1.95) (1.96) (1.97)

33

This becomes for our problem ( n1 n2 1 n3 ) 0 0 1 0 0 This gives rise to the polynomial equation (1 ) ((1 )(1 ) 4) = 0. This has three solutions = 1, = 1, = 3. (1.101) Notice all eigenvalues are real, which we expect since the tensor is symmetric. Now lets nd 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. (1.102) (1.100) 0 1 2 0 2 = (0 0 0). 1 (1.98)

For a non-trivial solution for ni , we must have 0 1 2

0 2 = 0. 1

(1.99)

Because of symmetry, we can now commute the indices to get (Tji ji ) ni = 0, because indices commute if symmetric. Expanding into matrix notation, we get T11 T21 T12 T22 T13 T23 n1 T31 0 T32 n2 = 0 . n3 T33 0 (1.103)

(1.104)

Note, we have taken the transpose of T in eigenvalue = 1, we get 0 0 0 0 0 2

the above equation. Substituting for T ji and considering the 0 0 n1 2 n2 = 0 . n3 0 0 (1.105)

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-dened direction. Here we will choose to normalize our eigenvector and take n1 = 1, so that the eigenvector is 1 for = 1. (1.106) nj = 0 0

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 n1 2 0 0 0 0 2 2 n2 = 0 . (1.107) 0 2 2 0 n3

34

This is simply the system of equations 2n1 2n2 + 2n3 2n2 + 2n3 = 0, = 0, = 0. (1.108) (1.109) (1.110)

This is the system of equations

Clearly n1 = 0. We could take n2 = 1 and n3 = 1 for a non-trivial solution. Alternatively, lets normalize and take 0 2 . (1.111) nj = 2 22 Finally consider the eigenvector associated with the eigenvalue = 3. Again substituting into the original equation, we get 0 n1 2 0 0 0 2 2 n2 = 0 . (1.112) 0 n3 0 2 2 2n1 2n2 + 2n3 = 0, = 0, = 0. (1.113) (1.114) (1.115)

2n2 2n3

In summary, the three eigenvectors and associated eigenvalues are 1 (1) nj = 0 for (1) = 1, 0 0 2 (2) nj = for (2) = 1, 2 nj
(3)

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 = . (1.116) 2
2 2

(1.117)

(1.118)

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 uid element in Figure 1.6.

2 2

0 2 2 2 2

for

(3) = 3.

(1.119)

Example 1.6
For a given stress tensor, which we will take to be symmetric though the theory applies to nonsymmetric tensors as well, 1 2 4 (1.120) Tij = T = 2 3 1 , 4 1 1

35

x3
(3)

=3 = -1
(2)

=1

(1)

x2

Figure 1.6: Sketch of uid element rotated to be aligned with axes of principal stress, along with magnitude of principal stress. The 1 face projects out of the page.
nd the three basic tensor invariants I (1) , I (2) , and I (3) , and show they are truly invariant when the tensor is subjected to a rotation with direction cosine matrix of 1 1 2
ij

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. (1.122) The three invariants of Tij are 1 2 4 I (1) = tr(T) = tr 2 3 1 = 1 + 3 + 1 = 5, 4 1 1 1 (tr(T))2 tr(T T) I (2) = 2 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 2 4 I (3) = det(T) = det 2 3 1 = 66. 4 1 1 Now when we rotate the tensor T, we get a transformed tensor given by 1 1 1 1 6 3 2 1 2 4 6 1 1 2 3 1 T = LT T L = 2 0 3 3 3 4 1 1 1 1 1 1
6 3 2 2 2 3 1 3 1 6 1 3 1 2

16 =L=
3 1 2

3 1 3

6 1 3 1 2

(1.121)

(1.123)

(1.124) (1.125)

36

4.10238 2.52239 1.60948 2.52239 0.218951 2.91291 . 1.60948 2.91291 1.11657

(1.126)

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 nd the invariants indeed are invariant: I (1) I (2) I (3) = 4.10238 0.218951 + 1.11657 = 5, 1 2 = (5 53) = 14, 2 = 66. (1.127) (1.128) (1.129)

Finally, we verify that the stress invariants are indeed related to the principal values (the eigenvalues of the stress tensor) as follows I (1) I (2) I
(3)

= (1) + (2) + (3) = 5.28675 3.67956 + 3.39281 = 5, = (1) (2) + (2) (3) + (3) (1) = (5.28675)(3.67956) + (3.67956)(3.39281) + (3.39281)(5.28675) = 14, = (1) (2) (3) = (5.28675)(3.67956)(3.39281) = 66.

(1.130) (1.131) (1.132)

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 (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 ). Taking a derivative using the chain rule gives d = dx1 + dx2 + dx3 . x1 x2 x3 37 (1.134) (1.133)

Following Panton, we dene a non-traditional, but useful further notation i for the partial derivative 1 x1 i = e1 + e2 + e3 = = x2 = 2 , (1.135) xi x1 x2 x3 3
x3

so that the chain rule is actually

d = 1 dx1 + 2 dx2 + 3 dx3 , which is written using our summation convention as d = i dxi . After commuting so as to juxtapose the i subscript, we have d = dxi i . In Gibbs notation, we say d = dxT = dxT grad .

(1.136)

(1.137)

(1.138) (1.139)

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

(1.140) (1.141) (1.142)

d = () dx, d = T dx.

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 analysis 14 dvi dvT dv dv = = = = dxj j vi = j vi dxj , dxT vT , (vT )T dx, (grad v)T dx.

(1.143)

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 T v = 2 ( v1 3

v2

A more common approach, not using the transpose notation, would be to say here for the Gibbs notation that dv = dx v. However, this is only works if we consider dv to be a row vector, as dx v 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.

14

1 v1 v3 ) = 2 v1 3 v1

1 v2 2 v2 3 v2

1 v3 2 v3 . 3 v3

(1.144)

38

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 , (1.145) 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. 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. (1.147) (1.146)

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: si = ijk j vk , s = v, s = curl v. Expanding for i = 1, 2, 3 gives s1 = s2 = s3 =
123 2 v3

(1.148)

+ 231 3 v1 + 312 1 v2 +

132 3 v2

= 2 v3 3 v2 , 213 1 v3 = 3 v1 1 v3 , 321 2 v1 = 1 v2 2 v1 . 39

(1.149)

1.2.3.4

Laplacian operator

The Laplacian 15 operator can operate on a scalar, vector, or tensor function. It is a simple combination of rst 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 . (1.150) In viscous uid ow, we will have occasion to have the Laplacian operate on vector: i i v j = T v T 1.2.3.5 Relevant theorems
T

= 2 v T

= 2 v = div grad v.

(1.151)

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. 1.2.3.5.1 as follows Fundamental theorem of calculus The fundamental theorem of calculus is
x=b x=a

f (x) dx =

x=b x=a

d dx

dx = (b) (a).

(1.152)

It eectively says that to nd the integral of a function f (x), which is the area under the curve, it suces to nd a function , whose derivative is f , and evaluate at each endpoint, and take the dierence to nd the area under the curve. 1.2.3.5.2 Gausss theorem Gausss 16 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: i (Tjk...) dV = ni Tjk... dS (1.153)
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 ni dS = 0. (1.154)
S

That is the unit normal to the surface integrated over the surface, cancels to zero when the entire surface is included.
Pierre-Simon Laplace, 1749-1827, Normandy-born French astronomer of humble origin. Educated at Caen, taught in Paris at Ecole Militaire. 16 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 G ottingen.
15

40

We will use Gausss theorem extensively. It allows us to convert sometimes dicult 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 Stokes
S 17

theorem is as follows. =
C

ni

ijk j vk dS

i vi ds.

(1.155)

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
S

nT v dS =

T v ds.

(1.156)

1.2.3.5.4 Leibnizs theorem Leibnizs 18 theorem relates time derivatives of integral quantities to a form which distinguishes changes which are happening within the boundaries to changes due to uxes through boundaries. It is a generalization of the more familiar control volume approach which uses the Reynolds 19 transport theorem. Leibnizs theorem applied to an arbitrary tensorial function is as follows: d dt Tjk...(xi , t) dV = Tjk... dV + t nl wl Tjk... dS. (1.157)

R(t)

R(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. Say we have the very special case in which Tjk... = 1; then Leibnizs theorem reduces to d dt
R(t)

dV

R(t)

(1) dV + t nk wk dS.

S (t)

nk wk (1) dS,

(1.158) (1.159)

dVR = 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.
Sir George Gabriel Stokes, 1819-1903, Irish-born British physicist and mathematician, holder of the Lucasian chair of Mathematics at Cambridge University, developed, simultaneously with Navier, the governing equations of uid motion, in a form which was more robust than that of Navier. 18 Gottfried Wilhelm von Leibniz, 1646-1716, Leipzig-born German philosopher and mathematician. Invented calculus independent of Newton and employed a superior notation to that of Newton. 19 Osborne Reynolds, 1842-1912, Belfast-born British engineer and physicist, educated in mathematics at Cambridge, rst professor of engineering at Owens College, Manchester, did fundamental experimental work in uid mechanics and heat transfer.
17

41

1.2.3.5.5 Reynolds transport theorem Leibnizs theorem reduces to the Reynolds transport theorem if we replace the tensor function Tjk... with a scalar function, say f . Further, considering one-dimensional cases only, we can then say d dt
x=b(t) x=a(t)

f (x, t) dx =

x=b(t) x=a(t)

db da f dx + f (b(t), t) f (a(t), t). t dt dt

(1.160)

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 and da are the velocities of the bounding surface and analogous to wk . The terms dt dt 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 uid 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 uid ows to a summation of fundamental motions.

1.3.1

Lagrangian description

A Lagrangian 20 description is similar to a classical description of motion in that each uid particle is eectively labeled and tracked in terms of its initial position xo j and time t. We take the position vector of a particle ri to be ri = r i (xo j , t). (1.161)

The velocity vi of a particular particle is the time derivative of its position, holding xo j xed: vi = r i t (1.162)
xo j

The acceleration ai of a particular particle is the second time derivative of its position, holding xo j xed: 2r i ai = (1.163) t 2 xo
j

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.

20

42

We can also write other variables as functions of time and initial position, for example, we could have for pressure p(xo j , 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 uid mechanics to use the Eulerian description of uid 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 = r i (xo j , t), t = t

(1.164)

1.3.3

Material derivatives

The material derivative is the derivative following a uid particle. It is also known as the substantial derivative or the total derivative. It is trivial in Lagrangian coordinates, since by denition, a Lagrangian description tracks a uid particle. It is not as straightforward in the Eulerian viewpoint. Consider a uid 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 (xo j , 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 (xo i (xo (1.165) j , t) = FE (xi = r j , t), t = t). Now from basic calculus we have dxi = From basic calculus, we also have dFL = dFE = FL t FE t + dt
xo j

r i t

+ dt
xo j

r i dxo j. xo j t

(1.166)

FL dxo j, xo j t FE dxi . xi t

(1.167) (1.168)

dt +
xi

43

Now, we must have dF = dFL = dFE for the same uid particle, so making substitutions from above, we get FL t + FL dxo dt j = xo j xo t
j

FE t

FE r i dt + xi t t xi

i + r dxo dt j . o x j t xo
j

(1.169)

o For the variation of F of a particular particle, we hold xo j xed, so that dxj = 0. Using also = t, so dt = dt, and dividing by dt , we get the fact that t

FL t

=
xo j

FE t

+
xi

FE xi

r i t

,
xo j

(1.170)

and using the denition of uid particle velocity, Eq. (1.162), we get FL t =
xo j

FE t

+ vi
xi

FE . xi t

(1.171)

Ignoring the operand F , FL , and FE , we can write the derivative following a particle in the following manner as an operator = xo t
j

+ vi
xi

xi

=
t

+ vT =

+ vT grad

d D (1.172) Dt dt

We will generally use the following shorthand for the derivative following a particle: d = o + v i i . dt (1.173)

Here a second shorthand for the partial derivative with respect to time has been introduced: o t .
xi

1.3.4

Streamlines

Streamlines are lines which are everywhere instantaneously parallel to velocity vectors. If a dierential 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.

(1.174) (1.175)

In Gibbs notation, we would say v dx = 0. 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. (1.176) dx1 dx2 dx3 44

Expanding the determinant gives e1 (v2 dx3 v3 dx2 ) + e2 (v3 dx1 v1 dx3 ) + e3 (v1 dx2 v2 dx1 ) = 0. (1.177)

Since the basis vectors e1 , e2 , and e3 are linearly independent, the coecient on each of them must be zero, giving rise to v2 dx3 = v3 dx2 , v3 dx1 = v1 dx3 , v1 dx2 = v2 dx1 , dx3 dx2 = , v3 v2 dx3 dx1 = , v1 v3 dx2 dx1 = . v2 v1 (1.178) (1.179) (1.180) (1.181) Combining, we get dx1 dx2 dx3 = = . v1 v2 v3 (1.182)

At a xed instant in time, t = to , we set the above terms all equal to an arbitrary dierential parameter d to obtain dx2 dx3 dx1 = = = d. v1 (x1 , x2 , x3 ; t = to ) v2 (x1 , x2 , x3 ; t = to ) v3 (x1 , x2 , x3 ; t = to ) (1.183)

Here should not be thought of as time, but just as a dummy parameter. Streamlines are only dened at a xed time. While they will generally look dierent at dierent times, in the process of actually integrating to obtain them, time does not enter into the calculation. We then divide each equation by d and nd the above equations are equivalent to a system of dierential equations of the autonomous form dx1 = v1 (x1 , x2 , x3 ; t = to ), d dx2 = v2 (x1 , x2 , x3 ; t = to ), d dx3 = v3 (x1 , x2 , x3 ; t = to ), d x1 ( = 0) = x1o , x2 ( = 0) = x2o , x3 ( = 0) = x3o . (1.184) (1.185) (1.186)

After integration, which in general must be done numerically, we nd x1 ( ; to , x1o ), x2 ( ; to , x2o ), x3 ( ; to , x3o ), where we let the parameter vary over whatever domain we choose. 45 (1.187) (1.188) (1.189)

1.3.5

Pathlines

The pathlines are the locus of points traversed by a particular uid particle. For an Eulerian description of motion where the velocity eld 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 dierential equations, with the associated initial conditions: dx1 = v1 (x1 , x2 , x3 , t), dt dx2 = v2 (x1 , x2 , x3 , t), dt dx3 = v3 (x1 , x2 , x3 , t), dt x1 (t = 0) = x1o , x2 (t = 0) = x2o , x3 (t = 0) = x3o . (1.190) (1.191) (1.192)

In general these are non-linear equations, and often require full numerical solution, which gives us x1 (t; x1o ), x2 (t; x2o ), x3 (t; x3o ). (1.193) (1.194) (1.195)

1.3.6

Streaklines

A streakline is the locus of points that have passed through a particular point at some past . Streaklines can be found by integrating a similar set of equations to those for time t = t pathlines. dx1 = v1 (x1 , x2 , x3 , t), dt dx2 = v2 (x1 , x2 , x3 , t), dt dx3 = v3 (x1 , x2 , x3 , t), dt ) = x1o , x1 (t = t ) = x2o , x2 (t = t ) = x3o . x3 (t = t (1.196) (1.197) (1.198)

After integration, which is generally done numerically, we get ), x1 (t; x1o , t ), x2 (t; x2o , t ). x3 (t; x3o , t (1.199) (1.200) (1.201)

Then if we x 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 ), x2 ( t ). x3 ( t 46 (1.202) (1.203) (1.204)

Example 1.7
If v1 = 2x1 + t, v2 = x2 t, nd a) the streamline through the point (1, 1)T at t = 1, b) the pathline for the uid 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 dierential equations, dx1 d dx2 d = = 2x1 + t|t=1 , x2 2t|t=1 , x1 ( = 0) = 1, x2 ( = 0) = 1.

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 d dx2 d Solving, we get x1 x2 Solving for , we nd = 1 ln 2 2 3 x1 + 1 2 . 3 2 1 e , 2 2 = e + 2. = = 2x1 + 1, = x2 2, x1 ( = 0) = 1, x2 ( = 0) = 1.

So eliminating and writing x2 (x1 ), we get the streamline to be x2 = 2 b) pathline For the pathline we have the following equations dx1 dt dx2 dt These have solution x1 x2 7 2(t1) t 1 e , 4 2 4 = 3et1 + 2t + 2. = = 2x1 + t, = x2 2t, x1 (t = 1) = 1, x2 (t = 1) = 1. 2 3 x1 + 1 . 2

It is algebraically dicult 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.

47

x2

1.5

1
streakline

0.5
pathline streamline

8 x1

Figure 1.7: Streamline, pathlines, and streaklines for unsteady ow of example problem.
c) streakline For the streakline we have the following equations dx1 dt dx2 dt These have solution x1 x2 = 2(tt t 1 5 + 2t ) e , 4 2 4 )ett = (1 + 2t + 2t + 2. 2(1t 5 + 2t 3 ) e , 4 4 1t = (1 + 2t)e + 4. = = 2x1 + t, = x2 2t, ) = 1, x1 (t = t ) = 1. x2 (t = t

We evaluate the streakline at t = 1 and get x1 x2

so as to write x2 (x1 ) explicitly. However, the Once again, it is algebraically dicult to eliminate t above gives a parametric representation of the streakline, which can be plotted in x 1 , x2 space. A plot of the streamline, pathline, and streakline for this problem is shown in Figure 1.7. 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.

48

P dxi = ds
i

v i + dvi v
i

dv i

P x x1
2

Figure 1.8: Sketch of uid particle P in motion with velocity vi and nearby neighbor particle P with velocity vi + dvi .

1.3.7

Kinematic decomposition of motion

In general the motion of a uid 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 uid motion routinely predict pathlines with nite curvature. More to the point, we cannot in general use the method of superposition to add one ow 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 uids is nearly always dominated by linear eects. By analyzing only the linear eects induced by small changes in velocity, which we will associate with the velocity gradient, we will learn a great deal about the richness of uid motion. In the linear analysis, we will see that a uid 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 uid and exible solid behavior. Generally it is the rate of change of strain (that is the straining) which has most relevance for a uid, while it is the actual strain that has the most relevance for a exible solid. This is because the stress in a exible solid responds to strain, while the stress in a uid responds to a strain rate. Nevertheless, while strain itself is associated with equilibrium congurations of a exible 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 translates; it cannot rotate or strain. uid motion = translation + rotation + extensional straining + shear straining,
straining

exible solid motion = translation + rotation + extensional straining + shear straining ,

straining

rigid solid motion = translation + rotation, point mass motion = translation.

Let us consider in detail the conguration shown in Figure 1.8. Here we have a uid particle at point P with coordinates xi and velocity vi . A small distance dri = dxi away is the uid particle at point P , with coordinates xi + dxi . This particle moves with velocity vi + dvi . We can describe the dierence in location by the product of a unit tangent vector 49

i and a scalar dierential distance magnitude ds: dri = dxi = i ds. 1.3.7.1 Translation

We have the motion at P to be vi + dvi . Obviously, the rst 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 . (1.205)

We can break the tensor i vj into a symmetric and antisymmetric part and say then dvj = Let dxi (i vj )
Shear and extensional straining

+ dxi [i vj ] .
Rotation

(1.206)

dvj = dxi (i vj ) .

(s)

(1.207)

We will see this is associated with straining, both by shear and extension. We will call the symmetric tensor (i vj ) the strain rate tensor. Further, let (r ) (1.208) dvj = dxi [i vj ] . We will see this is associated with rotation as a solid body. 1.3.7.2.1 Solid body rotation Let us examine dvj . First, we dene the vorticity vector k as the curl of the velocity eld k =
kij i vj (r )

= v.

(1.209)

Let us now split the velocity gradient i vj into its symmetric and antisymmetric parts and recast the vorticity vector as k = kij (i vj ) + kij [i vj ] . (1.210)
=0

The rst term on the right side is zero because it is the tensor inner product of an antisymmetric and symmetric tensor. In what remains, we see that vorticity k is actually the dual vector associated with the antisymmetric [i vj ] . k =
kij [i vj ]

= v.

(1.211)

Using Eq. (1.61) to invert the previous equation, we nd [i vj ] = 1 2

ijk k .

(1.212)

50

Thus we have dvj

(r )

= = = =

1 2

ijk k dxi ,

(1.213) (1.214) = , 2 . (1.215) (1.216)

ijk

k dxi , 2 and if dr

1 dr 2

Solid body rotation of one point about another

By introducing the above denition for , we see this term takes on the exact form for the dierential 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): dvk
(s)

dvk

(es)

+ dvk

(ss)

(1.217)

extension

shear

1.3.7.2.2.1 Extensional straining Let us dene 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; dvk
(es)

j dvj

(s)

k .

(1.218)

projection of straining

Now using the denition of dvj , we get dvk

(es)

(s)

j (i vj ) (i ds)

k , (1.219)

= k i j (i vj ) ds.

1.3.7.2.2.2 Shear straining What straining that is not aligned with the axis connecting 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 dvj
(ss)

= dvj dvj , =

(s)

(es)

(1.220) ds (1.221)

i (i vj ) j i k (i vk)

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 dierent, but ultimately equivalent fashion based on notions from linear algebra. 51

We rst recall that most form as follows:

21

arbitrary square matrices A can be decomposed into a diagonal A = S S1 . (1.222)

Here S is is a matrix of the same dimension as A which has in its columns the right eigenvectors of A. When A 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 S has orthonormal columns, the matrix is dened as orthogonal (though orthonormal would be a more accurate nomenclature). When S has been rendered orthogonal, we call it L. So when A is symmetric, we also have the following decomposition A = L L1 . (1.223)

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 A = L LT . (1.224)

Geometrically L is equivalent to a matrix of direction cosines; as we have seen before, its transpose LT 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 dierence; taking the symmetric (i vj ) = A, which we further assume to be a constant for this analysis, we rewrite Eq. (1.207) using Gibbs notation as dv(s)
T

d LT v(s)

dv(s) dv(s) dv(s) LT dv(s) LT dv(s) LT dv(s)

= dxT A, = = = = = =

(1.225) (1.226) (1.227) (1.228) (1.229) (1.230) (1.231)

= d LT x

AT dx, A dx, since A is symmetric. T L L dx, LT L LT dx, L1 L LT dx, LT dx,

since A and thus LT are assumed constant. (1.232)

Now we recall from the denition of vectors that LT v(s) = v (s) and LT x = x . That is these are the representations of the vectors in a rotated coordinate system, so we have dv
21

(s)

Some matrices, which often do not have enough linearly independent eigenvectors, cannot be diagonalized; however, the argument can be extended through use of the singular value decomposition. The singular value decomposition can also be used to eectively 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.

= dx .

(1.233)

52

Now since is diagonal, we see that a perturbation in x conned 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 uid, also called a particle of uid. We dene a material region as a region enclosed by a surface across which there is no ux 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 avor of the analysis to come for the conservation axioms. Taking the both M R and R(t) to denote the same time-dependent nite material region in space, we must have VM R =
R(t)

dV.

(1.234)

Using Leibnizs theorem, we take the time derivative of both sides and obtain dVM R = dt = = (1) dV + t ni vi dS, i vi dV by Gausss theorem, by the mean value theorem. ni vi dS, (1.235) (1.236) (1.237) (1.238)

R(t)

S (t)

S (t)

R(t)

= (i vi ) VM R

In the analysis above, we note that the velocity of S (t), in general wi , has been set to the uid velocity vi since we have a material region. We also recall from calculus the mean value theorem which states that for any integral, a mean value can be dened, denoted by a , as b 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 ) , VM R dt 1 dVM R lim = i vi = T v = div v VM R 0 VM R dt (1.239) (1.240)

The above expression describes the relative expansion rate also known as the dilation rate of a material uid particle. A uid particle for which i vi = 0 must have a relative expansion rate of zero, and satises conditions to be an incompressible uid. This is a common and useful assumption about which much theory has been developed. It is certainly not always valid, even in some very routine ows. 53

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 orientation for which the strain rate is aligned with the eigenvectors. As with stress, there are also three principal invariants of strain rate, namely I
(1) (2)

I I

(3)

= (i vi) = i vi = + + , 1 (1) (2) (2) (3) (3) (1) ((i vi) (j vj ) (i vj ) (j vi) ) = + + , = 2 (1) (2) (3) = ijk (1 vi) (2 vj ) (3 vk) = .
(1)

(1)

(2)

(3)

(1.241) (1.242) (1.243)

The physical interpretation for I is obvious in that it is equal to the relative rate of volume (2) (3) 1 d2 V 1 dV . Aris discusses how I is related to V and I is change for a material element, V dt dt2 1 d3 V related to V . dt3

1.3.10

Two-dimensional kinematics

We have developed relations which are valid for three dimensional kinematics. This is most important, but when learning, can lead to lengthy algebra and results which are hard to visualize. Here we will consider some important two dimensional cases, rst for general two-dimensional ows, and then for specic examples. 1.3.10.1 General two-dimensional ows

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). 1.3.10.1.1 Rotation Recalling that dxi = i ds, for rotation, we have dvj dvj
(r )

= [i vj ] dxi = i [i vj ] ds, = 1 [1 vj ] + 2 [2 vj ] ds,

(1.244) (1.245) (1.246) (1.247) (1.248) (1.249) (1.250) (1.251)

(r )

dv1

(r )

(r ) dv1

= 1 [1 v1] +2 [2 v1] ds,

=0

= 2 [2 v1] ds,

dv2

(r )

(r ) dv2

dv1

(r )

dv2

(r )

= 1 [1 v2] ds, rewriting in terms of the actual derivatives 1 2 (2 v1 1 v2 ) ds, = 2 1 = 1 (1 v2 2 v1 ) ds, 2 54

= 1 [1 v2] + 2 [2 v2] ds,

=0

Also for the vorticity vector, we get k =

ijk i vj .

(1.252)

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

311 =0

1 v1 +

312 =1

1 v2 +

321 =1

2 v1 +

322 =0

2 v2 ,

thus,

(1.253) (1.254)

= 1 v2 2 v1 . 1.3.10.1.2 Extension dvk

(es)

= k i j (i vj ) ds, = k 1 1 (1 v1) + 1 2 (1 v2) + 2 1 (2 v1) + 2 2 (2 v2) ds

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

(1.255) (1.256) (1.257) (1.258) (1.259) so

dv1

(es) (es)

2 2 = 1 1 1 v1 + 1 2 (1 v2 + 2 v1 ) + 2 2 v2 ds, 2 2 = 2 1 1 v1 + 1 2 (1 v2 + 2 v1 ) + 2 2 v2 ds.

dv2 1.3.10.1.3

Shear dvj
(ss)

= dvj dvj , = = 1 1 v 1 + 2

(s)

(es)

(1.260) (1.261) 2 v1 + 1 v2 2 ds,

i (i vj ) j i k (i vk) ds,

dv1

(ss)

2 2 1 1 1 v1 + 1 2 (1 v2 + 2 v1 ) + 2 2 v2

dv2

(ss)

2 2 v 2 + 1

1 v2 + 2 v1 2 ds.

2 2 2 1 1 v1 + 1 2 (1 v2 + 2 v1 ) + 2 2 v2

1.3.10.1.4

Expansion 1 dV = 1 v1 + 2 v2 . V dt (1.262)

55

v i (P)

v i (P)

ds

P x

Figure 1.9: Sketch of uid particle P in motion with velocity vi (P ) and nearby neighbor particle P with velocity vi (P ). 1.3.10.2 Relative motion along 1 axis

Let us consider in detail the conguration shown in Figure 1.9 in which the particle separation is along the 1 axis. Hence 1 = 1, 2 = 0, and 3 = 0. Rotation
(r )

dv1 dv2 Extension

(r )

= 0, 1 3 = (1 v2 2 v1 ) ds = ds. 2 2

(1.263) (1.264)

dv1

(es)

= 1 v1 ds, = 0.

(1.265) (1.266)

(es) dv2

Shear dv1 dv2 Expansion:

1 dV V dt (ss) (ss)

= 0, 1 = (1 v2 + 2 v1 ) ds = (1 v2) ds. 2

(1.267) (1.268)

= 1 v1 + 2 v2 . 56

v i (P) P ds P v i (P) x

Figure 1.10: Sketch of uid particle P in motion with velocity vi (P ) and nearby neighbor particle P with velocity vi (P ). 1.3.10.3 Relative motion along 2 axis

Let us consider in detail the conguration shown in Figure 1.10 in which the particle separation is aligned with the 2 axis. Hence 1 = 0, 2 = 1, and 3 = 0. Rotation dv1 dv2 Extension dv1
(es) (r )

(r )

3 1 (2 v1 1 v2 ) ds = ds, 2 2 = 0. =

(1.269) (1.270)

= 0. = 2 v2 ds.

(1.271) (1.272)

(es) dv2

Shear dv1
(ss) (ss)

dv2 Expansion

1 (2 v1 + 1 v2 ) ds = (1 v2) ds, 2 = 0.

(1.273) (1.274)

1 dV = 1 v1 + 2 v2 . V dt

(1.275)

57

Figure 1.11: Sketch of uniform ow 1.3.10.4 Uniform ow

Consider the kinematics of a uniform two-dimensional ow in which v1 = k 1 , as sketched in Figure 1.11. Streamlines:
dx1 v1

v2 = k 2 ,

v3 = 0,

(1.276)

dx2 , v2

dx1 k1

dx2 , k2

x1 =

k1 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. Shear for unrotated element: Expansion: 1 v1 + 2 v2 = 0. Acceleration:
dv1 dt dv2 dt 1 2

(1 v2 + 2 v1 ) = 0.

= o v1 + v1 1 v1 + v2 2 v1 = 0 + k1 1 (k1 ) + k2 2 (k1 ) = 0. = o v2 + v1 1 v2 + v2 2 v2 = 0 + k1 1 (k2 ) + k2 2 (k2 ) = 0.

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

58

Figure 1.12: Sketch of pure rigid body rotation. 1.3.10.5 Pure rigid body rotation

Consider the kinematics of a two-dimensional ow in which v1 = kx2 , as sketched in Figure 1.12. Streamlines: Extension on 1-axis: 1 v1 = 0. on 2-axis: 2 v2 = 0. Shear for unrotated element: Acceleration:
dv1 dt dv2 dt 1 2 dx1 v1

v2 = kx1 ,

v3 = 0,

(1.277)

dx2 , v2

dx1 kx2

dx2 , kx1

x1 dx1 = x2 dx2 ,

2 x2 1 + x2 = C .

Rotation: 3 = 1 v2 2 v1 = 1 (kx1 ) 2 (kx2 ) = 2k.

(1 (kx1 ) + 2 (kx2 ) = k k = 0.

Expansion: 1 v1 + 2 v2 = 0 + 0 = 0. = o v1 + v1 1 v1 + v2 2 v1 = 0 kx2 1 (kx2 ) + kx1 2 (kx2 ) = k 2 x1 . = o v2 + v1 1 v2 + v2 2 v2 = 0 kx2 1 (kx1 ) + kx1 2 (kx1 ) = k 2 x2 .

In this ow, 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. 59

Figure 1.13: Sketch of extensional ow (1-D compressible) 1.3.10.6 Pure extensional motion (a compressible ow)

Consider the kinematics of a two-dimensional ow in which v1 = kx1 , as sketched in Figure 1.13. Streamlines: Extension on 1-axis: 1 v1 = k . on 2-axis: 2 v2 = 0. Shear for unrotated element: Expansion: 1 v1 + 2 v2 = k . Acceleration:
dv1 dt dv2 dt 1 2 1 2 dx1 v1

v2 = 0,

v3 = 0,

(1.278)

dx2 , v2

v2 dx1 = v1 dx2 ,

0 = kx1 dx2 ,

x2 = C .

Rotation: 3 = 1 v2 2 v1 = 1 (0) 2 (kx1 ) = 0.

(1 v2 + 2 v1 ) =

(1 (0) + 2 (kx1 )) = 0.

= o v1 + v1 1 v1 + v2 2 v1 = 0 + kx1 1 (kx1 ) + 02 (kx1 ) = k 2 x1 . = o v2 + v1 1 v2 + v2 2 v2 = 0 + kx1 1 (0) + 02 (0) = 0.

In this ow, the streamlines are straight lines, there is no uid rotation, there is extension (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 ow. The acceleration is conned to the x1 direction. 60

Figure 1.14: Sketch of pure shearing ow 1.3.10.7 Pure shear straining

Consider the kinematics of a two-dimensional ow in which v1 = kx2 , as sketched in Figure 1.14. Streamlines: Extension on 1-axis: 1 v1 = 1 (kx2 ) = 0. on 2-axis: 2 v2 = 2 (kx1 ) = 0. Shear for unrotated element: Expansion: 1 v1 + 2 v2 = 0. Acceleration:
dv1 dt dv2 dt 1 2 dx1 v1

v2 = kx1 ,

v3 = 0,

(1.279)

dx2 , v2

dx1 kx2

dx2 , kx1

x1 dx1 = x2 dx2 ,

2 x2 1 = x2 + C .

Rotation: 3 = 1 v2 2 v1 = 1 (kx1 ) 2 (kx2 ) = k k = 0.

(1 v2 + 2 v1 ) =

1 2

(1 (kx1 ) + 2 (kx2 )) = k .

= o v1 + v1 1 v1 + v2 2 v1 = 0 + kx2 1 (kx2 ) + kx1 2 (kx2 ) = k 2 x1 . = o v2 + v1 1 v2 + v2 2 v2 = 0 + kx2 1 (kx1 ) + kx1 2 (kx1 ) = k 2 x2 .

In this ow, 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 uid is accelerating away from the origin. 61

x
Figure 1.15: Sketch of Couette ow. 1.3.10.8 Couette ow: shear + rotation

Consider the kinematics of a two-dimensional ow in which v1 = kx2 , v2 = 0, v3 = 0,

22

(1.280)

as sketched in Figure 1.15. This is known as a Couette Streamlines: Extension on 1-axis: 1 v1 = 1 (kx2 ) = 0. on 2-axis: 2 v2 = 2 (0) = 0. Shear for unrotated element: Expansion: 1 v1 + 2 v2 = 0. Acceleration:
dv1 dt dv2 dt 1 2 1 2 dx1 v1

ow. x2 = C .

dx2 , v2

dx1 kx2

dx2 , 0

0 = kx2 dx2 ,

Rotation: 3 = 1 v2 2 v1 = 1 (0) 2 (kx2 ) = k.

(1 v2 + 2 v1 ) =

(1 (0) + 2 (kx2 )) = k . 2

= o v1 + v1 1 v1 + v2 2 v1 = 0 + kx2 1 (kx2 ) + 02 (kx2 ) = 0. = o v2 + v1 1 v2 + v2 2 v2 = 0 + kx2 1 (0) + 02 (0) = 0.

Here the streamlines are straight lines, and the ow 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 uid is not accelerating.
Couette, M., little biographical information is readily available for this researcher, who did publish his work on ows of uids between xed and moving cylinders in the French scientic journals in the 1890s.
22

62

vj

Figure 1.16: Sketch of ideal irrotational vortex. 1.3.10.9 Ideal point vortex: extension+shear

Consider the kinematics of a two-dimensional ow sketched in Figure 1.16. v1 = k x2 1 x2 , + x2 2 v2 = k x2 1 x1 , + x2 2 v 3 = 0, (1.281)

Streamlines:

dx1 v1

dx2 , v2

dx1 k

x2 x2 +x2 1 2

dx2 x1 x2 +x2 1 2

1 = dx x2

dx2 , x1

2 x2 1 + x2 = C .

2 2 Rotation: 3 = 1 v2 2 v1 = 1 k x2x 2 k x2x = 0. +x2 +x2 1 2 1 2

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

on 2-axis: 2 v2 = 2

1 k x2x 2 1 +x2

x1 x2 2k (x2 2 2. 1 +x2 ) x2 x2
1 2

Shear for unrotated element: Expansion: 1 v1 + 2 v2 = 0. Acceleration:

dv1 dt dv2 dt

1 2

(1 v2 + 2 v1 ) = k (x22+x21)2 .

k x1 = o v1 + v1 1 v1 + v2 2 v1 = (x2 +x2 )2
1 2

= o v2 + v 1 1 v2 + v 2 2 v2 =

k 2 x2 (x2 2 2. 1 +x2 )

The streamlines are circles and the uid 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.

63

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 suciently 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. 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 rst 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 rst 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. 64

 dV dS w i = vi ni

Figure 1.17: Sketch of nite material region M R, innitesimal mass element dV , and innitesimal surface element dS with unit normal ni , and general velocity wi equal to uid velocity vi .

1.4.1

Mass
d mM R(t) = 0. dt

The mass conservation axiom is simple to state mathematically. It is (1.282)

Here M R(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 1.17. We can dene the mass of the material region based upon the local value of density: mM R(t) = So the mass conservation axiom is d dt
M R(t) M R(t)

dV.

(1.283)

dV = 0.

(1.284)

d Recalling Leibnizs theorem, Eq. (1.157), dt 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

d dt

M R(t)

dV =

M R(t)

dV + t

M S (t)

ni vi dS = 0.

(1.285)

Now we invoke Gausss theorem, Eq. (1.153) 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 dV + i (vi )dV = 0, t M R(t) + i (vi ) dV = 0. t M R(t) 65 (1.286) (1.287)

M R(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, (1.288) 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, o + T (v) = 0, o + 1 (v1 ) + 2 (v2 ) + 3 (v3 ) = 0. (1.289) (1.290) (1.291)

There are several alternative forms for this axiom. Using the product rule, we can say also o + v i i
material derivative of density

+i vi = 0,

(1.292)

or, writing in what is called the non-conservative form, d + i vi = 0, dt d + T v = 0, dt (o + v1 1 + v2 2 + v3 3 ) + (1 v1 + 2 v2 + 3 v3 ) = 0. So we can also say 1 d dt
relative rate of density increase

(1.293) (1.294) (1.295)

i vi
relative rate of particle volume expansion

(1.296)

Thus the relative rate of density increase of a uid particle is the negative of its relative rate of expansion, as expected. So we also have 1 d dt d + VM R dt d (VM R ) dt d (mM R ) dt = = 0, = 0, = 0. 1 dVM R , VM R dt (1.297) (1.298) (1.299) (1.300)

dVM R 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 ux of mass, for which the mass can indeed change, thus violating our non-relativistic mass conservation axiom. 66

f i dV vi dV dS w i = vi t
i

ni

Figure 1.18: Sketch of nite material region M R, innitesimal linear momenta element vi dV , innitesimal body force element fi dV , and innitesimal surface element dS with unit normal ni , surface traction ti and general velocity wi equal to uid velocity vi .

1.4.2
1.4.2.1

Linear momenta
Statement of the principle

The linear momenta conservation axiom is simple to state mathematically. It is d dt vi dV = fi dV + ti dS . (1.301)

M R(t)

M R(t)

M S (t)

rate of change of linear momenta

body forces

surface forces

Again M R(t) stands for a material region which can evolve in time. A relevant material region is sketched in Figure 1.18. 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 rst the left hand side, LHS , of the linear momenta principle LHS = = o (vi )dV + nj vi vj dS, from Leibniz, (1.302) (1.303)

M R(t)

M S (t)

M R(t)

(o (vi ) + j (vj vi )) dV,

from Gauss.

So the linear momenta principle is (o (vi ) + j (vj vi )) dV = fi dV + ti dS. (1.304)

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. 67

II ni t (nII ) i i n III i

t (nIII) i i

t (nI) i i nI i

Figure 1.19: Sketch of pillbox element for stress analysis. 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 ). 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. 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 1.19. Here we indicate the dependency of the traction on orientation by notation such as ti nII . This does not indicate multiplication, nor that i is a dummy index here. In i Figure 1.19, 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) = I II III (fi ) (S )(l) + t i (ni )S + ti (ni )S + ti (ni )s(l ). Now we let l 0, holding for now s and S xed to obtain
I II 0 = t i (ni ) + ti (ni ) S

(1.305)

(1.306)

(1.307)

68

3 -n1

t (-n ) S i 1 1 x i 2

t (-n ) S i 2 2

-n

-n 2

t (n ) S i i

t (-n ) S i 3 3 x 1
Figure 1.20: 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 II nI i = ni ni , we get a useful result ti (ni ) = ti (ni ). (1.308) At an innitesimal 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 dierent propulsion systems that macro-organisms: they are ghting dierent 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 1.20. 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. 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 69

the form (inertia) (L)3 = (body forces) (L)3 + (surface forces) (L)2 . (1.309)

As before, for small volumes, L 0, the linear momenta principle reduces to surface forces = 0. Applying this to the conguration of Figure 1.20, we get
0 = t i (ni )S + ti (n1 )S1 + ti (n2 )S2 + ti (n3 )S3 .

(1.310)

(1.311)

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

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

(1.312)

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

t i (ni )S = n1 ti (n1 )S + n2 ti (n2 )S + n3 ti (n3 )S, t i (ni ) = n1 ti (n1 ) + n2 ti (n2 ) + n3 ti (n3 ).

(1.313) (1.314)

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 . In shorthand, we can say the same thing with ti = nj Tji , or equivalently tj = ni Tij , QED. (1.316) (1.315)

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 uid 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 thermodynamic variables, say p = f (, e), where e is the specic internal energy. Also recall that the thermodynamic pressure must be a normal stress, as thermodynamics considers formally only materials at rest, and viscous stresses are associated with moving uids. To distinguish between thermodynamic stresses and other stresses, let us dene the viscous stress tensor ij as follows ij = Tij + pij . 70 (1.317)

Recall that Tij is the total stress tensor. We obviously also have Tij = pij + ij . (1.318)

Note with this denition that pressure is positive in compression, while Tij and ij are positive in tension. Let us also dene the mechanical pressure, p(m) , as the negative of the average normal surface stress 1 1 p(m) Tii = (T11 + T22 + T33 ). 3 3 (1.319)

The often invoked Stokes assumption, which remains a subject of widespread misunderstanding 150 years after it was rst 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 . 3 (1.320)

Presumably a pressure measuring device in a moving oweld would actually measure the mechanical pressure, and not necessarily the thermodynamic pressure, so it is important to have this issue claried 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 kinetic 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 the kinetic theory models is able to predict results from experiments, which nevertheless also give indication, albeit indirect, that Stokes assumption is invalid. Kinetic theories and experiments which consider polyatomic molecules, which can suer vibrational and rotational eects as well, show further deviation from Stokes assumption. It is often plausibly argued that these so-called non-equilibrium eects, that is molecular vibration and rotation, which are only important in high speed ow applications in which the ow velocity is on the order of the uid sound speed, are the mechanisms which cause Stokes assumption to be violated. Because they only are important in high speed applications, they are dicult to measure, though measurement 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. (1.318), we get Tii = pii + ii . (1.321)

Using the fact that ii = 3 and inserting Eq. (1.320) in Eq. (1.321), we nd for a uid that obeys Stokes assumption that Tii = 1 Tii (3) + ii , 3 0 = ii . 71 (1.322) (1.323)

That is to say, the trace of the viscous stress tensor is zero. Moreover, for a uid 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 ij = Tkk ij + , (valid only if Stokes assumption holds) 3
total stress

(1.324) 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 uid is static. For a static uid, there is no viscous stress, so ij = 0, and we have We get the traction vector on any surface with normal ni by 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 tj = ni Tij = pni ij = pnj . (1.326) Tij = pij , static uid. (1.325)

mean stress

deviatoric stress

We are now prepared to write the linear momenta equation in nal form. Substituting our expression for the traction vector, Eq. (1.316) into the linear momenta expression, Eq. (1.304), we get
M R(t)

(o (vi ) + j (vj vi ))dV =

M R(t)

fi dV +

M S (t)

nj Tji dS.

(1.327)

Using Gausss theorem, Eq. (1.153), to convert the surface integral into a volume integral, and combining all under one integral sign, we get
M R(t)

(o (vi ) + j (vj vi ) fi j Tji )dV = 0.

(1.328)

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 Using then Tij = pij + ij , we get in Cartesian index, Gibbs o (vi ) + j (vj vi ) fi j Tji = 0. (1.329) , and full notation
(1.330) (1.331) (1.332) (1.333) (1.334)
23

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 + .

23

o (v3 ) + 1 (v1 v3 ) + 2 (v2 v3 ) + 3 (v3 v3 ) = f3 3 p + 1 13 + 2 23 + 3 33 .

o (vi ) + j (vj vi ) = fi i p + j ji , T T (v) + T (vvT ) = f p + T , t o (v1 ) + 1 (v1 v1 ) + 2 (v2 v1 ) + 3 (v3 v1 ) = f1 1 p + 1 11 + 2 21 + 3 31 , o (v2 ) + 1 (v1 v2 ) + 2 (v2 v2 ) + 3 (v3 v2 ) = f2 2 p + 1 12 + 2 22 + 3 32 ,

72

The form above is known as the linear momenta principle cast in conservative or divergence form. It is the rst 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 roundo error. However, there is a reduced, non-conservative form which makes some analysis and physical interpretation easier. Let us use the product rule to expand the linear momenta principle, then rearrange it, and use mass conservation and the denition of material derivative to rewrite the expression: o vi + vi o + vi j (vj ) + vj j vi = fi i p + j ji , (o vi + vj j vi ) +vi ( o + j (vj ) ) = fi i p + j ji ,
dvi dt =0 by mass conservation

(1.335) (1.336)

dvi dt dv dt (o v1 + v1 1 v1 + v2 2 v1 + v3 3 v1 ) (o v2 + v1 1 v2 + v2 2 v2 + v3 3 v2 ) (o v3 + v1 1 v3 + v2 2 v3 + v3 3 v3 )

= fi i p + j ji , = f p + T
T

(1.337) , (1.338)

= f1 1 p + 1 11 + 2 21 + 3 31 , (1.339) = f2 2 p + 1 12 + 2 22 + 3 32 , (1.340) = f3 3 p + 1 13 + 2 23 + 3 33 . (1.341)

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 rst recall some notions from classical rigid body mechanics, while referring to the sketch of Figure 1.21. We have the angular momenta vector L for the particle of Figure 1.21 L = r (mv). which is Any force F which acts on m with lever arm r induces a torque T = r F. T Angular momenta Torque of body force Torque of surface force Angular momenta from surface couples 73 = = = = r (dV )v = ijk rj vk dV, r f (dV ) = ijk rj fk dV, r tdS = ijk rj np Tpk dS, nT dS = nk ki dS (1.343) (1.342)

Now let us apply these notions for an innitesimal uid particle with dierential mass dV . (1.344) (1.345) (1.346) (1.347)

Figure 1.21: Sketch of particle of mass m velocity v rotating about an axis centered at point O , with radial distance vector r. 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 mathematically, d dt
ijk rj vk dV

M R(t)

M R(t)

ijk rj fk dV

M S (t)

ijk rj np Tpk

+ nk ki ) dS .

(1.348)

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
M R(t)

ijk rj vk dV

M S (t)

ijk rj vk np vp dS

= (1.349)

M R(t)

ijk rj fk dV

M S (t)

ijk rj np Tpk

+ nk ki ) dS.

Next with Gauss we get

M R(t)

ijk rj vk dV

M R(t)

ijk p (rj vk vp )dV

= (1.350)

M R(t)

ijk rj fk dV

M R(t)

ijk p (rj Tpk )

M R(t)

k ki dV.

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 ki .

(1.351)

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

ijk

rj o (vk ) + vk o rj +rj p (vp vk ) + vp vk p rj rj fk

=0

pr

rj p Tpk Tpk p rj = k ki . pr

(1.352)

74

Applying the simplications indicated above and rearranging, we get

ijk rj

(o (vk ) + p (vp vk ) fk p Tpk ) = k ki

=0 by linear momenta

ijk vj vk

ijk Tjk .

(1.353)

So we can say, k ki =
ijk (vj vk

Tjk ) =

ijk antisym.

ijk T[jk ] .

vj vk T(jk)
sym. sym.

antisym.

T[jk] ,

(1.354) (1.355)

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 elds, such as could be the case when electromagnetic forces are important, we have the common condition of ki = 0, and the angular momenta principle reduces to the simple statement that T[ij ] = 0. (1.356) That is, the antisymmetric 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 , T = TT . (1.357) (1.358)

1.4.4

Energy

We recall the rst law of thermodynamics, which states the 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 = . (1.359) 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 rst law of thermodynamics in detail and then write the equation in nal form. 1.4.4.1 Total energy term

For a uid particle, the dierential amount of total energy is 1 dE = e + vj vj dV, 2 1 = dV e + vj vj 2

mass specic internal +kinetic energy

(1.360) . (1.361)

75

 dV dS ni

Figure 1.22: Sketch of nite material region M R, innitesimal mass element dV , and innitesimal surface element dS with unit normal ni , and heat ux vector qi . 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). 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, dW dxT = F = vT F, dt dt (1.362) (1.363) (1.364) 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 uid = (dV )(fi )vi , Work rate done by uid = vi fi dV. Work rate done by a surface force Work rate done by force on uid = (ti dS )vi = [(nj Tji )dS ]vi , Work rate done by uid = nj Tji vi dS. 1.4.4.3 Heat transfer term (1.367) (1.368) (1.365) (1.366)

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 1.22, we dene the heat ux vector q i 76

as a vector which points in the direction of thermal energy ow which has units of energy per area per time; in SI this would be W/m2 . So we have heat transferred from body through dS = ni qi dS, heat transferred to body through dS = ni qi dS, 1.4.4.4 Conservative form of the energy equation (1.369) (1.370) (1.371)

Putting the words of the rst law into equation form, we get d dt 1 e + vj vj dV = 2 M R(t)
M S (t)

ni qi dS +

M S (t)

ni Tij vj dS +

M R(t)

fi vi dS. (1.372)

Skipping the details of an identical application of Leibnizs and Gausss theorems, and shrinking the volume to approach zero, we obtain the dierential equation of energy in conservative or divergence form (in rst Cartesian index then Gibbs notation): 1 o e + vj vj 2
rate of change of total energy

1 + i vi e + vj vj 2
convection of total energy

= + vi fi
body force work rate

i qi
diusive heat ux

i (Tij vj )
surface force work rate

(1.373)

1 e + vT v t 2

1 + T v e + vT v = 2 T q + T (T v) + vT f .

(1.374)

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 o e + vj vj 2 1 e + vT v t 2 1 + i vi e + vj vj = 2 i qi + i (pvi ) + i (ij vj ) + vi fi , 1 + T v e + vT v = 2 T q T (pv) + T ( v) + vT f .

(1.375)

(1.376)

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 rst, the mechanical energy equation, actually has no foundation in the rst 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. 77

1.4.4.5.1 Mechanical energy equation The mechanical energy equation, a pure consequence 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 vj vj vj o + vi i 2 2 vj vj vj vj vj vj o + i (vi ) mass : 2 2 2 add to get, vj vj vj vj o + i vi 2 2 T T v v v v + T v t 2 2 = vj fj vj j p + (i ij )vj , = vj fj vj j p + (i ij )vj , = 0, (1.377) (1.378) (1.379) (1.380) = vj fj vj j p + (i ij )vj . = vT f vT p + T v. (1.381) (1.382)

The term vj2vj represents the volume averaged kinetic energy, with SI units J/m3 . Note that the mechanical energy equation, (1.381), predicts the kinetic energy increases due to three eects: uid motion in the direction of a body force, uid motion in the direction of decreasing pressure, or uid motion in the direction of increasing viscous stress. Note that body forces themselves aect mechanical energy, while it is imbalances in surface forces which aect mechanical energy. 1.4.4.5.2 Thermal energy equation If we take the conservative form of the energy equation (1.375) and subtract from it the mechanical energy equation (1.381), we get an equation for the evolution of thermal energy: o (e) + i (vi e) = i qi pi vi + ij i vj , (1.383) (1.384)

(e) + T (ve) = T q pT v + : vT . t

Here e is the volume averaged internal energy with SI units J/m3 . Note that the thermal energy equation (1.383) predicts thermal energy (or internal energy) increases due to negative gradients in heat ux (more heat enters than leaves), pressure force accompanied by a mean negative volumetric deformation (that is, a uniform compression; note that i vi is the relative expansion rate), or 78

 viscous force associated with a deformation

24

(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 inuence 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 nonconservative form of the energy equation, also known as the energy equation following a uid particle, by the following operations. First expand the thermal energy equation (1.383): o e + eo + vi i e + ei (vi ) = i qi pi vi + ij i vj . 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 ,
de dt =0 by mass

(1.385)

(1.386)

so we get de = i qi pi vi + ij i vj , dt de = T q pT v + : vT . dt (1.387) (1.388)

We can get an equation which is reminiscent of elementary thermodynamics, valid for small volumes V by replacing i vi by its known value in terms of the relative expansion rate to obtain dV de = V i qi p + V ij i vj . (1.389) V dt dt The only term not usually found in elementary thermodynamics texts is the third, 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 the specic enthalpy h is dened as p h=e+ . (1.390)

Now starting with the energy equation following a particle (1.387), we can use one form of the mass equation to eliminate the relative expansion rate i vi in favor of density derivatives to get p d de + ij i vj . (1.391) = i qi + dt dt
24 For a general uid, this includes a mean volumetric deformation as well as a deviatoric deformation. If the uid satises Stokes assumption, it is only the deviatoric deformation that induces a change in internal energy in the presence of viscous stress.

79

Rearranging, we get de p d 2 dt dt = i qi + ij i vj . (1.392)

Now from the denition of h, we have 1 p d + dp, 2 de p d 1 dp dh = 2 + , dt dt dt dt de p d dh 1 dp 2 = . dt dt dt dt dh = de So, the energy equation in terms of enthalpy becomes dh dp = i qi + ij i vj , dt dt dp dh = T q + : vT . dt dt (1.396) (1.397) (1.393) (1.394) (1.395)

1.4.4.5.5 Energy equation in terms of entropy (not the second law!) 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 again the Gibbs equation from thermodynamics, which serves to dene entropy s: T ds = de + pdv . (1.398)

Here T is the absolute temperature, and v is the specic volume, v = V /m = 1/. In terms of , the Gibbs equation is p T ds = de 2 d. (1.399) Taking the material derivative, which is operationally equivalent to dividing by dt, and solving for de/dt,we get de ds p d =T + 2 . (1.400) dt dt dt The above is still basically a thermodynamic denition of s. Now use this relation in the non-conservative energy equation (1.387) to get an alternate expression for the rst law: T Recalling that i vi =
1 d , dt

ds p d + = i qi pi vi + ij i vj . dt dt

(1.401)

we have T ds = i qi + ij i vj , dt ds 1 1 = i qi + ij i vj . dt T T 80 (1.402) (1.403)

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 1 1 qi 2 qi i T + ij i vj , = i dt T T T 1 1 q ds = T 2 qT T + : vT . dt T T T (1.404) (1.405)

From this statement, we can conclude from the rst law of thermodynamics that the entropy of a uid 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 the above expression (1.404) to obtain 1 qi 1 2 qi i T + ij i vj , T T T 1 1 q 2 qT T + : vT . (s) + T (vs) = T t T T T o (s) + i (vi s) = i (1.406) (1.407)

1.4.5

Entropy inequality

Recall the mathematical statement of the entropy inequality from classical thermodynamics: dS dQ . T (1.408)

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 heat transferred. If the process is adiabatic, dQ = 0, and the entropy can either remain xed or rise. In terms of a rate equation, we have then 1 dQ dS . dt T dt Now for our continuous material we have dS = sdV, dQ = qi ni dA (1.410) (1.411) (1.409)

Here we have used s for the specic 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 ux vector is aligned with the outward normal, heat leaves the system. Since we want positive dQ to represent heat in to a system, we need the negative sign. 81

Now, applying our typical machinery to the second law gives rise to d dt
M R(t) M R(t)

sdV

o (s)dV +
M R(t)

M S (t)

svi ni dA =

(o (s) + i (svi )) dV (o (s) + i (svi )) dV

M R(t)

qi ni dA, T M S (t) qi ni dA, T M S (t) qi i dV, T M R(t) qi dV + i T M R(t) qi + I, T qi + I. T

(1.412) (1.413) (1.414)

M R(t)

IdV,

(1.415) (1.416) (1.417) (1.418)

where irreversibility I 0, o (s) + i (svi ) = i ds = i dt

This is the second law. Now if we subtract from this the rst law written in terms of entropy, Eq. (1.404), we get the result I= 1 1 qi i T + ij i vj . 2 T T = ij i vj . (1.419)

As an aside, we have dened the commonly used viscous dissipation function as (1.420)

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-Duhem 25 26 inequality: 1 1 (1.421) qi i T + ij i vj 0, 2 T T 1 1 2 qT T + : vT 0 (1.422) T T Recalling that ij is symmetric by the angular momenta principle, 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 qi i T + ij (i vj ) 0, 2 T T
T

(1.423) (1.424)

Rudolf Clausius, 1822-1888, Prussian-born German mathematical physicist, key gure in making thermodynamics a science, author of well-known statement of the second law of thermodynamics, taught at Z urich Polytechnikum, University of W urzburg, and University of Bonn. 26 Pierre Maurice Marie Duhem, 1861-1916, French physicist, mathematician, and philosopher, taught at Lille, Rennes, and the University of Bordeaux.

25

T T 1 T 1 v + v 2 q T + : T T 2

82

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 satises the entropy inequality. The more restrictive (and in some cases, overly restrictive) strong form of the ClausiusDuhem 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, qT T 0,

ij (i vj ) 0,
T

(1.425)

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 mechanisms which cause mechanical energy the change within a nite xed control volume V . First consider a restatement of the mechanical energy equation, Eq. (1.378) in terms of the material derivative of specic kinetic energy: d vj vj dt 2 = vj fj vj j (p) + vj i ij . (1.427)

T T v + v

(1.426)

Now use the product rule to restate the pressure and viscous work terms so as to achieve d vj vj dt 2 = vj fj j (vj p) + pj vj + i (ij vj ) ij i vj . =0 (1.428)

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. (1.381), here written in the same way which takes advantage of application of the product rule to the pressure and viscous terms: o vj vj 2 + i vi vj vj 2 = vj fj j (vj p) + pj vj + i (ij vj ) ij i vj . vj vj 2 (1.429)

Now integrate over the xed control volume, so that

V

vj vj 2

dV +

i vi

dV

vj fj dV
V

j (vj p) dV +
V

pj vj dV (1.430)

+ Applying Leibnizs rule and Gausss law, we get t

V

i (ij vj ) dV

ij i vj dV.

vj vj dV + 2

ni vi

vj vj dS = 2

vj fj dV
S

nj vj p dS +
V

pj vj dV (1.431)

+ 83

ni (ij vj ) dS

ij i vj dV.

Now on the surface of the xed volume, the velocity is zero, so we get t vj vj dV = 2 vj fj dV + pj vj dV ij i vj dV.
positive

(1.432)

Now the strong form of the second law requires that ij i vj = ij (i vj ) 0. So we see for a nite xed volume of uid 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 eect.

1.4.6

Summary of axioms in dierential 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 rst present the conservative form and then the non-conservative form. 1.4.6.1 1.4.6.1.1 Conservative form Cartesian index form o + i (vi ) o (vi ) + j (vj vi ) ij 1 + i vi e + vj vj 2 = 0, = fi i p + j ji , = ji , = i qi i (pvi ) + i (ij vj ) (1.436) (1.437) (1.433) (1.434) (1.435)

1 o e + vj vj 2

+vi fi , qi o (s) + i (svi ) i . T 1.4.6.1.2 Gibbs form

+ T (v) = 0, t (v) + T (vvT ) t 1 e + vT v t 2 1 + T v e + vT v 2
T

(1.438)
T

= f p + T = T,

(1.439) (1.440)

= T q T (pv) (1.441) (1.442)

+T ( v) + vT f , q s . + T (sv) T t T

84

1.4.6.1.3 Non-orthogonal index form 27 These have not been completely veried! g v k ( g ) + k t x i i g v j j + k g v j v k j t x x x = 0, = g f j (1.443) i xj i g pg jk j x jk i , g xj (1.444) k gq

xk + k x

1 g e + gij v i v j t 2

1 k e + g v gij v i v j k x 2

g sv k ( g s) + k t x 1.4.6.2 1.4.6.2.1 Non-conservative form Cartesian index form d dt dvi dt ij de dt ds dt

27

xk k g pv k x + k g gij v j ik x + g gij v j f i , (1.445) qk k . (1.446) g x T =

= i vi , = fi i p + j ji , = ji , = i qi pi vi + ij i vj , i qi . T

(1.447) (1.448) (1.449) (1.450) (1.451)

Here we introduce, following Aris and many others, some standard notation from tensor analysis. In this notation, both sub- and superscripts are needed to distinguish between what are known as covariant and contravariant vectors, which are really dierent mathematical representations of the same quantity, just k k k cast onto dierent basis vectors. In brief, we have the metric tensor gij = xi xj , where is a Cartesian k imn jpq coordinate and xi is a non-Cartesian coordinate. We also have g ij = 1 gmp gnq , g = det 2 xi . The
gki ij jk 1 mk Christoel symbols are given by m ij = 2 g xi + xj xk . We note also that few texts give a proper exposition of the conservative form of the equations in non-orthogonal coordinates. Here we have extended the development of Vinokur, 1974, Conservation Equations of Gasdynamics, Journal of Computational Physics, Vol. 14, No. 2, pp. 105-125, to include the eects of momentum and energy diusion. This extension has been guided by general notions found in standard works such as Aris as well as the recent book of Liseikin. g g

85

1.4.6.2.2

Gibbs form d dt dv dt de dt ds dt = T v, = f p + T = T, = T q pT v + : vT , T q . T These have not been checked carefully! (1.457) (1.458)
T

(1.452) , (1.453) (1.454) (1.455) (1.456)

1.4.6.2.3

Non-orthogonal index form

i gv , + vi i = t x g xi v i v i 1 ij jk i ij p i l = f g g + i + vj + v + jk , jl t xj xj g xj p i e e 1 i gq gv + vi i = i t x g x g xi v i l +gik kj + i , jl v j x s s 1 qi g . + vi i t x g xi T 1.4.6.3 Physical interpretations

(1.459) (1.460)

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 dierential equations can be represented in the following form: local change = convection + diusion + source. (1.461) Here we consider convection and diusion to be types of transport phenomena. If we have a xed volume of material, a property of that material, such as its thermal energy, can change because an outside ow sweeps energy in from outside. That is convection. It can change because random molecular motions allow slow leakage to the outside or leakage in from the outside. That is diusion. 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 dierent way to illustrate these mechanisms: = t local change in mass vT convection of mass 86

v = t

+0 diusion of mass T v, volume expansion source local change in linear momenta vT v

T

(1.462)

convection of linear momenta (1.463)

e = t

diusion of linear momenta + T , +f body force source of linear momenta p pressure force source of linear momenta local change in thermal energy vT e convection of thermal energy T q diusion of thermal energy T p v pressure work thermal energy source T + : v viscous work thermal energy source.

(1.464)

Briey considering the second law, we note that the irreversibility I is solely associated with diusion of linear momenta and diusion of energy. This makes sense in that diusion 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 modelled as a continuum under the inuence of the forces we have mentioned. Specically, they are valid for both solid and uid 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 eects 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.

1.4.8

Integral forms

Our governing equations are formulated based upon laws which apply to a material element. 87

 We are not often interested in an actual material element but in some other xed of moving region in space. Rules for such systems can by formulated with Leibnizs theorem in conjunction with the dierential forms of our axioms. Let us rst apply Leibnizs theorem (1.157) to an arbitrary function f over a time dependent arbitrary region AR(t): d dt f dV = f dV + t ni wi f dS. (1.465)

AR(t)

AR(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 ), Now lets use this, and let f = in Leibnizs theorem to get d dt d dt d dt dV dV = = o dV + ni wi dS,
AS (t)

(1.466)

AR(t)

AR(t)

AS (t)

(1.467) (1.468) (1.469) (1.470)

AR(t)

AR(t)

(i (vi ))dV + ni (wi vi )dS.

ni wi dS,

with Gauss
AR(t)

dV

AS (t)

Now consider three special cases. 1.4.8.1.1 Fixed region We take wi = 0. d dt d dt 1.4.8.1.2
AR(t)

AR(t)

dV

AS (t)

ni vi dS,

(1.471) (1.472)

dV +

AS (t)

ni vi dS = 0.

Material region Here we take wi = vi . d dt dV = 0. (1.473)

M R(t)

88

V (t) (material volume)

2

z air H
a

water D1= 1" v1= 3 ft/s h(t)

w

V1 (fixed)

D2= 3" v = 2 ft/s

2

A = 2 ft 2

Figure 1.23: Sketch of volume with water and air being lled with water. 1.4.8.1.3 Moving solid enclosure with holes Say the region considered is a solid enclosure with holes through with uid can enter and exit. The our arbitrary surface AS (t) can be specied as AS (t) = Ae (t) area of entrances and exits +As (t) solid moving surface with wi = vi +As xed solid surface with wi = vi = 0. Then we get d dt (1.474) (1.475) (1.476)

AR(t)

dV +

Ae (t)

ni (vi wi )dS = 0.

(1.477)

Example 1.8
Consider the volume sketched in Figure 1.23. Water enters a circular hole of diameter D 1 = 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 w exists in the tank at height h(t). Air at density a lls the remainder of the tank. Find the rate of rise of the water dh dt . Consider two control volumes V1 : the xed 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: d dt

a dV
V2

= 0,

(1.478)

89

d dt Mass conservation for V1 is d dt

a Adz
h(t)

= 0.

(1.479)

dV +
V1 Ae

vi ni dS = 0.

(1.480)

Now break up V1 and write Ae explicitly d dt

h(t)

w Adz +
0

d dt

a Adz
h(t) =0 h(t)

A1

w vi ni dS

w vi ni dS,
A2

(1.481)

d dt

w Adz
0

= w v1 A1 + 2 v2 A2 , = w 2 2 (v1 D1 + v 2 D2 ), 4 w 2 2 (v1 D1 + v 2 D2 ), 4 2 2 (v1 D1 + v 2 D2 ) 4A 1 (3 f t/s) ft 2 4(2 f t ) 12 ft . s

(1.482) (1.483) (1.484) (1.485)

2

w A

d dt

h(t)

dz
0

= = =

dh dt

+ (2 f t/s)

3 12

f t)2

(1.486) (1.487)

= 0.057

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 theorem to linear momenta, vi : d dt
AR(t)

vi dV =

AR(t)

o (vi )dV +

AS (t)

nj wj vi dS.

(1.488)

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 d dt
AR(t)

vi dV =

AS (t)

(nj (vj wj )vi + ni p nj ij )dS +

AR(t)

fi dV.

(1.489)

Now momentum ux terms only have values at entrances and exits (at solid surfaces we get vi = wi , so we can say d dt
AR(t)

vi dV +

Ae (t)

nj (vj wj )vi dS =

AS (t)

ni pdS +

AS (t)

nj ij dS +

AR(t)

fi dV.

(1.490) Note that the surface forces are evaluated along all surfaces, not just entrances and exits. 90

1.4.8.3

Energy

Applying the same analysis to the energy equation, we obtain d dt 1 e + vj vj dV 2 AR(t) = + 1.4.8.4 General expression 1 ni (vi wi ) e + vj vj dS 2 AS (t)
AR(t)

ni qi dS

AS (t)

(ni vi p ni ij vj )dS vi fi dV. (1.491)

AR(t)

If we have a governing equation from a physical principle which is of form o fj + i (vi fj ) = i gj + hj , then we can say for an arbitrary volume that d dt
AR(t)

(1.492)

fj dV = fj

AS (t)

ni fj (vi wi )dS +
ux of

AS (t)

ni gj dS + gj

AR(t)

hj dV . hj

(1.493)

change of

fj

eect of

eect of

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 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 subcontinuum is formally averaged, that we obtain commonly used constitutive equations. In most cases however, constitutive equations simply represent curve ts to basic experimental results, which can vary widely from material to material. As is briey discussed below, constitutive equations are not completely arbitrary. Whatever is proposed must allow our nal equations to be invariant under Galilean 28 transformations and rotations as well as satisfy the entropy inequality. For example, we might hope to develop a constitutive equation for the heat ux 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, . . .). (1.494)

The principles of continuum mechanics will rule out some possibilities, but still allow a broad range of forms.
28 Galileo Galilei, 1564-1642, Pisa-born Italian astronomer, physicist, and developer of experimental methods, rst 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.

91

1.5.1

Frame and material indierence

Our choice of a constitutive law must be invariant under a Galilean transformation (frame invariance) a rotation (material indierence). Say for example, we propose that the heat ux vector is proportional to the velocity vector qi = avi , trial constitutive relation (1.495)

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 nd 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 , (1.496)

where u2 is the displacement of particle. While this may t 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 indierence.

1.5.2

Second law restrictions and Onsager relations

The entropy inequality from the second law of thermodynamics provides additional restrictions on the form of constitutive equations. Recall the second law (equivalently, the weak form of the Clausius-Duhem inequality) tells us that 1 1 qi i T + ij (i vj ) 0. 2 T T (1.497)

We would like to nd 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 suces to consider a one-dimensional limit. In the one-dimensional limit, the weak form of the entropy inequality reduces to 1 T 1 u + 0. q 2 T x T x
q T u T

(1.498)

We can write this in a vector form as

1 T ( T x 1 u u x

0.

(1.499)

92

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 uxes q/T and u/T and 1 T 1 u the generalized driving gradients T and u : x x q 1 T = C11 T T x u 1 T = C21 T T x 1 u , u x 1 u + C22 , u x + C12 (1.500) (1.501) (1.502) In matrix form this becomes
q T u T

C11 C21

C12 C22

1 T T x 1 u u x

(1.503)

We then substitute this hypothesized relationship into the entropy inequality to obtain
1 T ( T x 1 u u x

C11 C21

C12 C22

1 T T x 1 u u x

0.

(1.504)

Now in a well known result from linear algebra, a necessary and sucient condition for satisfying the above inequality is that the matrix Cij be positive denite. Further, the matrix will be positive denite if it is symmetric and has positive eigenvalues. Thus we must have C12 = C21 . All this said, we must dismiss our general hypothesis on other physical grounds, namely that such a hypothesis results in an innite shear stress for a uid at rest! Note that in T u the special case in which T = 0, our hypothesis predicts = C22 u Obviously this 2 x . x is not consistent 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 condence that we should not expect to nd heat ux 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 eects. That such a matrix such as we studied in the previous section was required to be symmetric is a manifestation of what is known as an Onsager relation, developed by Onsager 29 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 Peltier 30 eect in which conduction of both heat and electrical charge is inuenced by gradients of charge and temperature. This forms the basis of the operation of a thermocouple. Other relations exist are the Soret
29 Lars Onsager, 1903-1976, Norwegian-born American physical chemist, earned Ph.D. and taught at Yale, developed a systematic theory for irreversible chemical processes. 30 Jean Charles Athanase Peltier, 1785-1845, French clockmaker, retired at 30 to study science.

93

eect in which diusive mass uxes are induced by temperature gradients, the Dufour eect in which a diusive energy ux is induced by a species concentration gradient, the Hall 31 eect for coupled electrical and magnetic eects (which explains the operation of an electric motor), the Seeback 32 eect in which electromotive forces are induced by dierent conducting elements at dierent temperatures, the Thomson 33 eect in which heat is transfered when electric current ows in a conductor in which there is a temperature gradient, and the principle of detailed balance for multispecies 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 the sucient (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, T2 1 ij (i vj ) 0. T and (1.505) (1.506)

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. 34 The portion of the strong form of the entropy inequality with which we are concerned here is 1 qi i T 0. T2 (1.507)

Now one way to guarantee this inequality is satised is to specify the constitutive relation for the heat ux vector as qi = ki T, with k 0. (1.508)

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 importantly, it is consistent with macro-scale experiments for isotropic materials, and can be
Edwin Herbert Hall, 1855-1938, Maine-born American physicist, educated at Johns Hopkins University where he discovered the Hall eect while working on his dissertation, taught at Harvard. 32 Thomas Johann Seebeck, 1770-1831, German medical doctor who studied at Berlin and G ottingen. 33 William Thomson (Lord Kelvin), 1824-1907, Belfast-born British mathematician and physicist, graduated and taught at Glasgow University, key gure in all of 19th century engineering science including mathematics, thermodynamics, and electrodynamics. 34 Jean Baptiste Joseph Fourier, 1768-1830, French mathematician and Egyptologist who studied the transfer of heat and the representation of mathematical functions by innite series summations of other functions. Son of a tailor.
31

94

justied 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, (1.509) 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. We can also generalize Fouriers law for an anisotropic material. For such materials, the thermal conductivity is a tensor kij , and Fouriers law generalizes to qi = kij j T. (1.510) This eectively states that for a xed temperature gradient, the heat ux depends on the orientation. This is characteristic of anisotropic substances such as layered materials. Substitution of the generalized Fouriers law into the entropy inequality (for ij = 0) gives now 1 kij (j T )(i T ) 0, T2 1 (i T )kij (j T ) 0, T2 (1.511) (1.512)

1 (T )T K T 0. (1.513) T2 Now 1/T 2 > 0, so we must have (T )T K T 0 for all possible values of T . Once again we use the well-known result from linear algebra that this is guaranteed if K = kij is a positive denite tensor. This will be the case if 1) K = KT , kij = kji so as to prevent imaginary eigenvalues and possible imaginary values of entropy, and 2) all the eigenvalues of K are positive. The symmetry of K = kij in fact guarantees that its eigenvalues are real, although symmetry alone does not guarantee they are positive. That this is sucient to satisfy the entropy inequality is made plausible if we consider T to be an eigenvector, so that K T = T giving rise to an entropy inequality of 1 (T )T (T ) 0, (1.514) T2 (T )T T 0. (1.515) T2 The inequality holds for all T (and T ) as long as 0. The fact that the conductivity tensor is symmetric for even anisotropic materials may be somewhat surprising, but it has been borne out by careful experiments on crystalline materials, and is eectively demanded by the second law. It is also another example of Onsagers principle. For our particular case with a tensorial conductivity, the competing eects are the heat uxes in three directions, caused by temperature gradients in three directions: 1 T k11 k12 k13 q1 (1.516) q2 = k21 k22 k23 2 T . 3 T k31 k32 k33 q3 The symmetry condition, Onsagers principle, requires that k12 = k21 , k13 = k31 , and k23 = k32 . 95

cross sectional area A

F v x

x1

h1

F/A h2 h3

v/h

Figure 1.24: Sketch of simple Couette ow experiment with measurements of stress versus strain rate.

1.5.4

Stress-strain rate relation for a Newtonian uid

We now seek to satisfy the second part of the strong form of the entropy inequality, namely (and recalling that T > 0) (1.517) ij (i vj ) 0. 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 uid, a uid 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 at plate over a uid and measure the force necessary to maintain a specied velocity. This situation and some expected results are sketched in Figure 1.24. We observe that At the upper and lower plate surfaces, the uid 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. 96

 21
Bingham plastic pseudo-plastic 1 Newtonian

dilatant

v1 x2

Figure 1.25: Variation of viscous stress with strain rate for typical uids. When experiments are carried out with dierent plate area and dierent gap width, a single universal curve results when F/A is plotted against V /h. The velocity prole is linear with increasing x2 . In a way similar on a molecular scale to energy diusion, this experiment is describing a diusion of momentum from the pulled plate into the uid below it. The constitutive equation we develop for viscous stress, when combined with the governing axioms, will model momentum diusion. 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 dene the coecient of viscosity for this conguration as = 21 viscous stress = . 2 v1 strain rate (1.518)

The viscosity is the analog of Youngs 35 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 uid has a viscosity which does not depend on strain rate (but could depend on temperature and pressure). A non-Newtonian uid has a viscosity which is strain rate dependent (and possible temperature and pressure). Some typical behavior is sketched in Figure 1.25. We shall focus here on uids whose viscosity is not a function of strain rate. Much of our development will
35 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 uids, gave the word energy scientic signicance, and developed Youngs modulus in elasticity.

97

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 uid

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 uid with constant viscosity. An isotropic uid has no directional dependencies when subjected to a force. A uid composed of aligned long chain polymers is an example of a uid that is most likely not isotropic. Following Whitaker, we postulate that stress is a function of deformation rate (strain rate) only: ij = fij ((k vl) ). 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) ), 12 = f12 ((1 v1) , (2 v2) , (3 v3) , (1 v2) , (2 v3) , (3 v1) (2 v1) , (3 v2) , (1 v3) ), . . . 33 = f33 ((1 v1) , (2 v2) , (3 v3) , (1 v2) , (2 v3) , (3 v1) (2 v1) , (3 v2) , (1 v3) ). (1.520) (1.521) (1.522) (1.523)
36

(1.519)

require that ij = 0 if (i vj ) = 0, hence, no strain rate, no stress. require that stress is linearly related to strain rate: ijkl (k vl) . ij = C (1.524)

ijkl is a fourth This is the imposition of the assumption of a Newtonian uid. Here C order tensor. Thus we have in matrix form
1111 C 11 C 22 2211 33 C 3311 12 C 1211 23 = C 2311 C 31 3111 21 C 2111 32 C 3211 1311 C 13

1122 C 2222 C 3322 C 1222 C 2322 C 3122 C 2122 C 3222 C 1322 C

1133 C 2233 C 3333 C 1233 C 2333 C 3133 C 2133 C 3233 C 1333 C

1112 C 2212 C 3312 C 1212 C 2312 C 3112 C 2112 C 3212 C 1312 C

1123 C 2223 C 3323 C 1223 C 2323 C 3123 C 2123 C 3223 C 1323 C

1131 C 2231 C 3331 C 1231 C 2331 C 3131 C 2131 C 3231 C 1331 C

1121 C 2221 C 3321 C 1221 C 2321 C 3121 C 2121 C 3221 C 1321 C

1132 C 2232 C 3332 C 1232 C 2332 C 3132 C 2132 C 3232 C 1332 C

1113 (1 v1) C 2213 C (2 v2) 3313 (3 v3) C 1213 (1 v2) C 2313 (2 v3) C 3113 C (3 v1) 2113 (2 v1) C (3 v2) C3213 1313 (1 v3) C (1.525)

ijkl . We found one of them in our simple There are 34 = 81 unknown coecients C experiment in which we found 21 = 12 = 2 v1 = (2(1 v2) ). 1212 = 2. Hence in this special case C
36 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.

98

Now we could do eighty-one separate experiments, or we could take advantage of the assumption that the uid 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 ijkl . The experiment is conducted by varying observer begins with the viscosity matrix C strain rate and measuring stress. With complete knowledge A feels condent 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 ijkl . We intend to show that if the uid is isotropic only two of the on the viscosity matrix C eighty-one coecients are distinct and non-zero. We rst use symmetry properties of the stress and strain rate tensor to reduce to thirty-six unknown coecients. 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
C1111 11 22 C 2211 C 33 3311 = 12 C 1211 2311 23 C 3111 C 31

1122 C C2222 3322 C C1222 2322 C 3122 C

1133 C C2233 3333 C C1233 2333 C 3133 C

1112 + C 1121 C C2212 + C2221 3312 + C 3321 C C1212 + C1221 2312 + C 2321 C 3112 + C 3121 C

1123 + C 1132 C C2223 + C2232 3323 + C 3332 C C1223 + C1232 2323 + C 2332 C 3123 + C 3132 C

s as Now adopting Whitakers notation for simplication, we dene the above matrix of C is a tensor. We take a new matrix of C s. Here, now C itself is not a tensor, while C equivalently then

1131 + C 1113 (1 v1) C 2231 + C 2213 C (2 v2) 3331 + C 3313 (3 v3) C 1231 + C 1213 (1 v2) C 2331 + C 2313 (2 v3) C 3131 + C 3113 (3 v1) C (1.526)

Next, recalling that for tensorial quantities ij = (i vj ) =

ki lj kl , ki lj (k vl) ,

C11 11 C21 22 33 = C31 C 41 12 C51 23 C61 31

C12 C22 C32 C42 C52 C62

C13 C23 C33 C43 C53 C63

C14 C24 C34 C44 C54 C64

C15 C25 C35 C45 C55 C65

(1 v1) C16 C26 (2 v2) (3 v3) C36 C46 (1 v2) C56 (2 v3) (3 v1) C66

(1.527)

(1.528) (1.529)

let us subject our uid to a battery of rotations and see what can be concluded by enforcing material indierence. 180 rotation about x3 axis

For this rotation, sketched in Figure 1.26. we have direction cosines = 1 = 21 = 0 31 = 0

11

ki

=0 = 1 22 32 = 0
12

13 23 33

=0 = 0 . =1

(1.530)

99

x3 x3 x1

x2

x2

x1

Figure 1.26: Rotation of 180 about x3 axis. Applying the transform rules to each term in the shear stress tensor, we get 11 22 33 12 23 31 Likewise we nd that (1 v1) (2 v2) (3 v3) (1 v2) (2 v3) (3 v1) = = = = = = (1 v1) , (2 v2) , (3 v3) , (1 v2) , (2 v3) , (3 v1) . (1.537) (1.538) (1.539) (1.540) (1.541) (1.542) = = = = = =
k 1 l1 kl k 2 l2 kl k 3 l3 kl k 1 l2 kl k 2 l3 kl k 3 l1 kl

= (1)2 11 = 11 , = (1)2 22 = 22 , = (1)2 33 = 33 , = (1)2 12 = 12 , = (1)(1)23 = 23 , = (1)(1)31 = 31 .

(1.531) (1.532) (1.533) (1.534) (1.535) (1.536)

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) , 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) , Since we want both predictions to be the same, we must require that C15 = C16 = 0. 100 (1.545) (1.544) (1.543)

x3

x2

x2

x1

x1 x3

Figure 1.27: Rotation of 180 about x1 axis. In matrix form, our observer A would predict for the rotated frame that

11 C11 22 C21 33 = C31 C 12 41 23 C51 31 C61

C12 C22 C32 C42 C52 C62

C13 C23 C33 C43 C53 C63

C14 C24 C34 C44 C54 C64

C15 C25 C35 C45 C55 C65

C16 (1 v1) C26 (2 v2) (3 v3) C36 C46 (1 v2) C56 (2 v3) C66 (3 v1)

(1.546)

To retain material dierence 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 coecients and gives our viscosity matrix the form

C11 C21 C31 C 41 0 0 180 rotation about x1 axis

C12 C22 C32 C42 0 0

C13 C23 C33 C43 0 0

C14 C24 C34 C44 0 0

0 0 0 0 C55 C65

0 0 0 . 0 C56 C66

(1.547)

with only 20 independent coecients.

This rotation is sketched in Figure 1.27. Leaving out the details of the previous section, this rotation has a set of direction cosines of 1 0 0 0 (1.548) ij = 0 1 0 0 1 101

x3 x2

x3 x2

x1

x1

Figure 1.28: Rotation of 90 about x1 axis. Application of this rotation leads to of the form C11 C12 C21 C22 C31 C32 0 0 0 0 0 0 180 rotation about x2 axis 90 rotation about x1 axis This rotation is sketched in Figure 1.28. This rotation has a set of direction cosines of 1 0 0 = 0 0 1 0 1 0 C12 C23 C22 0 0 0 0 0 0 C44 0 0 0 0 0 0 C55 0 0 0 0 . 0 0 C66

the conclusion that the viscosity matrix must be C13 C23 C33 0 0 0 0 0 0 C44 0 0 0 0 0 0 C55 0 0 0 0 . 0 0 C66

(1.549)

with only 12 independent coecients.

One is tempted to perform this rotation as well, but nothing new is learned from it!

ij

(1.550)

Application of this rotation leads to of the form C11 C12 C21 C22 C21 C23 0 0 0 0 0 0

the conclusion that the viscosity matrix must be

(1.551)

with only 8 independent coecients. 102

x3 x3

x1 x2

x1

x2

Figure 1.29: Rotation of 90 about x3 axis. 90 rotation about x3 axis This rotation is sketched in Figure 1.29. This rotation has a set of direction cosines of 0 1 0 = 1 0 0 0 0 1

ij

(1.552)

Application of this rotation leads to of the form C11 C12 C12 C11 C12 C12 0 0 0 0 0 0 90 rotation about x2 axis

the conclusion that the viscosity matrix must be C12 C12 C11 0 0 0 0 0 0 C44 0 0 0 0 0 0 C44 0 0 0 0 . 0 0 C44

(1.553)

with only 3 independent coecients.

We learn nothing from this rotation. 45 rotation about x3 axis This rotation is sketched in Figure 1.30. This rotation has a set of direction cosines of 2 / 2 2/2 0 2/2 0 ij = 2/2 0 0 1 103

(1.554)

x3 x3

x2 45 o x2 45 o

x1

x1

Figure 1.30: Rotation of 45 about x3 axis. After a lot of algebra, application of this rotation lead to the conclusion that the viscosity matrix must be of the form

with only 2 independent coecients.

C44 + C12 C12 C 12 0 0 0

C12 C44 + C12 C12 0 0 0

C12 C12 C44 + C12 0 0 0

0 0 0 C44 0 0

0 0 0 0 C44 0

0 0 0 . 0 0 C44

(1.555)

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 uid, we can expect two independent coecients 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) , 22 = C44 (2 v2) + C12 (1 v1) + (2 v2) + (3 v3) , 33 = C44 (3 v3) + C12 (1 v1) + (2 v2) + (3 v3) , 12 = C44 (1 v2) , 23 = C44 (2 v3) , 31 = C44 (3 v1) . Using traditional notation, we take C44 2, where is the rst coecient of viscosity, and 104 (1.556) (1.557) (1.558) (1.559) (1.560) (1.561)

 C12 , where is the second coecient 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 coecients, 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 = ijkl (i vj ) as C

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 + 22 33 12 0 23 = 0 0 31 21 0 32 0 13 0

2 + 0 0 0 0 0 0

2 + 0 0 0 0 0 0

0 0 0 2 0 0 0 0 0

0 0 0 0 2 0 0 0 0

0 0 0 0 0 2 0 0 0

0 0 0 0 0 0 2 0 0

0 0 0 0 0 0 0 2 0

0 (1 v1) v 0 (2 2) 0 (3 v3) 0 (1 v2) (2 v3) 0 0 (3 v1) 0 (2 v1) 0 (3 v2) (1 v3) 2

(1.562)

In shorthand Cartesian index and Gibbs notation, the viscous stress tensor is given by ij = 2(i vj ) + k vk ij , vT + (vT )T = 2 2 (1.564) + (T v)I. (1.565)

2 + 11 22 33 12 0 23 = 0 0 31 21 0 32 0 0 13

2 + 0 0 0 0 0 0

2 + 0 0 0 0 0 0

0 0 0 0 0 0 0 0 0 0 0 0 0 0

0 0 0 0 0 0 0

0 0 0 0 0 0 0 0 0 0 0 0 0 0

(1 v1) 0 v 0 (2 2) 0 (3 v3) 0 (1 v2) 0 (2 v3) (3 v1) 0 (2 v1) 0 (3 v2) (1 v3)

(1.563)

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 inuences stress. It is easily veried by direct expansion that the viscous stress tensor can be written as

ij = (2 + 3) = (2 + 3)

k vk 3
mean strain rate

mean viscous stress T

1 ij v , + 2 v k k ij ( i j ) 3
deviatoric strain rate T deviatoric viscous stress (vT )T 1 T

(1.566)

v + v I + 2 3 2 105

vI . 3

(1.567)

Here it is seen that a mean strain rate, really a volumetric change, induces a mean viscous stress, as long as = (2/3). If either = (2/3) or k vk = 0, all viscous stress is deviatoric. Further, for = 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 . 3 3 (1.568)

Note that the mean viscous stress is a scalar, and is thus independent of orientation; it is directly proportional to the rst 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 , vT + (vT )T T = pI + 2 + (T v)I. 2 (1.569) (1.570)

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 rst invariant of the strain rate (1) tensor: I = tr((i vk) ) = k vk . Consequently, the stress tensor can be written as Tij = p + I p + I
(1) (1)

ij +

2(i vj )
linear in strain rate

(1.571)

isotropic

T =

(1)

I + 2

vT + (vT )T 2

(1.572)

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 uid. 1.5.4.2.1 Diagonal component Consider a typical diagonal component of the viscous stress tensor, say 11 :
(2 + 3)

11 =

1 v1 + 2 v2 + 3 v3 3
mean strain rate

mean viscous stress

+ 2 1 v 1

deviatoric viscous stress

1 (1 v1 + 2 v2 + 3 v3 ) (1.573) . 3
deviatoric strain rate

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 o-diagonal elements. Thus the fundamental physics of the stress-strain relationship are completely embodied in a natural way in the above expression. 106

1.5.4.2.2 O-diagonal component If we are not aligned with the principle axes, then o-diagonal terms will be non-zero. A typical o-diagonal component of the viscous stress tensor, say 12 , has the following form: 12 = 2 (1 v2) + k vk 12 ,
=0

(1.574) (1.575) (1.576)

= 2(1 v2) , = (1 v2 + 2 v1 ).

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. (1.323): ii = 0. If we enforce this on our expression for ij , we get ii = 0 = = = = 2(i vi) + k vk ii , 2i vi + 3k vk , 2i vi + 3i vi , (2 + 3)i vi . (1.577) (1.578) (1.579) (1.580)

Since in general i vi = 0, Stokes assumption implies that 2 = . 3 (1.581)

So a Newtonian uid satisfying Stokes assumption has the following constitutive equation for viscous stress 1 ij = 2 (i vj ) k vk ij , 3
deviatoric strain rate deviatoric viscous stress T T T

(1.582)

= 2

(v + (v ) ) 1 T ( v)I . 2 3

(1.583)

Note that incommpressible ows have i vi = 0; thus, plays no role in determining the viscous stress in such ows. For the uid that obeys Stokes assumption, the viscous stress is entirely deviatoric and is induced only by a deviatoric strain rate. 107

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 sucient (but perhaps overly restrictive) to require that 1 ij (i vj ) 0. T (1.584)

Invoking our constitutive equation for viscous stress, and realizing that the absolute temperature T > 0, we have then that = (2(i vj ) + k vk ij )((i vj ) ) 0. This reduces to the sum of two squares: = 2(i vj ) (i vj ) + k vk i vi 0. (1.586) (1.585)

We then seek restrictions on and such that this is true. Obviously requiring 0 and 2 0 guarantees satisfaction of the second law. However, Stokes assumption of = 3 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 rst check the criterion for a strictly onedimensional system. For such a system, our second law restriction reduces to 2(1 v1) (1 v1) + 1 v1 1 v1 (2 + )1 v1 1 v1 2 + 0, 0, 0, 2. (1.587) (1.588) (1.589) (1.590)

2 , the entropy inequality is satised. We also could satisfy Obviously if > 0 and = 3 the inequality for negative with suciently large positive .

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. (1.591)

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) 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. 108 (1.593) (1 v1) + (2 v2) 0. (1.592)

In matrix form, we can write this inequality in the form known from linear algebra as a quadratic form: = ( (1 v1) (2 v2) (2 + ) 0 (1 v1) (1 v2) ) (2 + ) 0 (2 v2) 0. 0 0 4 (1 v2)

(1.594)

As we have discussed before, the conditions that this hold for all values of the deformation is that the coecient matrix be symmetric, which it is, and that it have eigenvalues which are greater than or equal to zero. Let us nd the eigenvalues of the coecient matrix. The eigenvalues are found by evaluating the following equation (2 + ) 0 = 0. (2 + ) 0 0 0 4 We get the characteristic polynomial (4 ) (2 + )2 2 = 0. This has roots = 4, = 2, = 2( + ). For the two-dimensional system, we see now formally that we must satisfy both 0, . (1.600) (1.601) (1.597) (1.598) (1.599) (1.596) (1.595)

This is more restrictive than for the one-dimensional system, but we see that a uid obeying still satises this inequality. Stokes assumption = 2 3 1.5.4.4.3 Three dimensional systems For a full three dimensional variation, the entropy inequality (2(i vj ) + k vk ij )((i vj ) ) 0, when expanded, is equivalent to the following quadratic form
) + 2
0 0 0 + 2 0 0 0 + 2 0 0 0 0 0 0 4 0 0 0 0 0 0 4 0 0 (1 v1) 0 (2 v2) 0 (3 v3) 0 (1 v2) 0 (2 v3) 4 (3 v1)

( (1 v1)

(2 v2)

(3 v3)

(1 v2)

(2 v3)

(3 v1)

0.

(1.602) 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 109

inequality. It is easy to show that the six eigenvalues for the interior matrix are = = = = = = 2, 2, 4, 4, 4, 3 + 2. (1.603) (1.604) (1.605) (1.606) (1.607) (1.608)

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, (1.609) (1.610) 2 . 3

Obviously a uid which satises Stokes assumption does not violate the entropy inequality, but it does give rise to a minimum level of satisfaction. This does not mean the uid 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 eort, 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.

(1.611)

Obviously, this is a sum of perfect squares, and holds for all values of the strain rate tensor. It can be veried by direct expansion that this term is identical to the strong form of the entropy inequality for viscous stress. It can further be veried by direct expansion that the entropy inequality can also be written more compactly as 1 = 2 (i vj ) k vk ij 3
deviatoric strain rate

1 2 (i vj ) m vm ij + + 3 3
deviatoric strain rate

(i vi )(j vj ) 0. (1.612)
(mean strain rate)2

So, we see that for a Newtonian uid that the increase in entropy due to viscous dissipation is attributable to two eects: deviatoric strain rate and mean strain rate. The terms involving 2 both are perfect squares, so as long as 0 and 3 , the second law is not violated by viscous eects. We can also write the strong form of the entropy inequality for a Newtonian uid (2(i vj ) + k vk ij )((i vj ) ) 0, in terms of the principal invariants of strain rate. Leaving out details, which can be veried by direct expansion of all terms, we nd the following form = 2 2 (1) I 3
2

2I

(2)

2 + + 3

(1) 2

0.

(1.613)

110

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-denite. We can use the denitions of the invariants of strain rate to rewrite the inequality as = 2 (i vj ) (j vi) 1 2 (i vi ) (j vj ) + + (i vi ) (j vj ) 0 3 3 (1.614)

In terms of the eigenvalues of the strain rate tensor, 1 , 2 , and 3 , this becomes
2 2 = 2 2 1 + 2 + 3

1 2 (1 + 2 + 3 )2 + + (1 + 2 + 3 )2 0 3 3

(1.615)

This then reduces to a positive semi-denite form: 2 2 = (1 2 )2 + (1 3 )2 + (2 3 )2 + + (1 + 2 + 3 )2 0 3 3 (1.616)

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 uids. The inuence 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 relations 37 and will induce an additional mean viscous stress for uids that do not obey Stokes assumption. A deviatoric strain rate will not directly induce a mean stress. A deviatoric strain rate will directly induce a deviatoric stress. A mean strain rate will induce entropy production only for a uid that does not obey Stokes assumption. A deviatoric strain rate will always induce entropy production in a viscous uid.

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 nest time scales of uid dynamics, it is a valid assumption to use an ordinary equations of state. Such assumptions can be violated in very high speed ows in which vibrational and rotational modes of oscillation become excited. They may also be invalid in highly rareed ows 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 ),
37

(1.617)

e.g. for an isothermal ideal gas

dp dt

= RT

d dt

= RT i vi

111

and a caloric equation of state which gives the internal energy as a function of two independent thermodynamic variables, e.g. e = e(, T ). (1.618) 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 dene a new thermodynamic variable, a, the Helmholtz 38 free energy: a = e T s. (1.619) We can take the material time derivative of the above expression to get de ds dT da = T s . dt dt dt dt (1.620)

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 T dT a + dt d . dt (1.621)

Now we also have the energy equation and entropy inequality: de = i qi pi vi + ij i vj , dt qi ds i . dt T Eliminating
de dt

(1.622) (1.623)

in the energy equation gives a modied energy equation da ds dT +T +s dt dt dt = i qi pi vi + ij i vj . (1.624)

Next, we eliminate a T

da dt

in the above equation to get d ds dT +T + s = i qi pi vi + ij i vj . dt dt dt dT a dt T d s dT . dt T dt

dT a + dt

(1.625)

Now in this modied energy equation, we solve for ds to get dt

38

1 p 1 a ds = i qi i vi + ij i vj dt T T T T T

(1.626)

Hermann von Helmholtz, 1821-1894, Potsdam-born German physicist and philosopher, descendent of William Penn, the founder of Pennsylvania, empiricist and refuter of the notion that scientic 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.

112

Substituting this version of the energy conservation equation into the second law gives us
1 p 1 a i qi i vi + ij i vj T T T T T

dT a dt T

d s dT dt T dt

qi . (1.627) T

Rearranging and using the mass conservation relation to eliminate i vi , we get

qi p i T 2 T T 1 a dT a d s dT ij i vj T T T dt T T dt T dt p d a dT a d dT qi + ij i vj s i T + T dt T dt T dt dt +
T

1 d dt

0, 0, 0.

(1.628) (1.629) (1.630)

1 d a qi p 2 i T + ij i vj + T dt

dT dt

s+

a 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 denite. We can guarantee the remaining two terms are consistent with the second law, and are associated with reversible processes by requiring that p = 2 s = a a T ,
T

(1.631) (1.632)

For example, if we take the non-obvious, but experimentally defensible choice for a of a = cv (T To ) cv T ln then we get for pressure p = 2 a = 2
T

T + RT ln , To o RT

(1.633)

= RT.

(1.634)

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 s= Then, we get for e e = a + T s = cv (T To ). (1.636) We call the above equation for energy a caloric equation of state for calorically perfect gas. It is calorically perfect because the specic 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. 113 a T = cv ln

T R ln . To o

(1.635)

1.6

Boundary and interface conditions

At uid solid interfaces, it is observed in the continuum regime that the uid sticks to the solid boundary, so that we can safely take the uid 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 uids, one requires that stress be continuous across the interface, that the energy ux be continuous across the interface. Density need not be continuous in the absence of mass diusion. Were mass diusion present, the uids would not be immiscible, and density would be a continuous variable. Additionally the eect 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 equations

Here we pause once more to write a complete set of equations, the compressible Navier 39 -Stokes equations, written here for a uid which satises Stokes assumption, but for which 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 1.7.0.1.1 Conservative form Cartesian index form
o + i (vi ) = 0, o (vi ) + j (vj vi ) = fi i p 1 o e + vj vj 2 1 +i vi e + vj vj 2 1 +j 2 (j vi) k vk ji 3 , (1.637)

(1.638)

= vi fi i (pvi ) + i (ki T ) 1 +i 2 (i vj ) k vk ij vj , 3 p = p(, T ), e = e(, T ), = (, T ), k = k (, T ). (1.639) (1.640) (1.641) (1.642) (1.643)

Claude Louis Marie Henri Navier, 1785-1836, Dijon-born French civil engineer and mathematician, studied under Fourier, taught applied mechanics at Ecole des Ponts et Chauss ees, replaced Cauchy as professor at Ecole Polytechnique, specialist in road and bridge building, did not fully understand shear stress in a uid and used faulty logic in arriving at his equations.

39

114

1.7.0.1.2

Gibbs form
+ T (v) = 0, t (1.644)

(v) + T (vvT ) t

= f p + T 2 vT + (vT )T 1 (T v)I 2 3
T

, (1.645)

1 e + vT v t 2 1 +T v e + vT v 2

= vT f T (pv) + T (k T ) +T p = p(, T ), e = e(, T ), = (, T ), k = k (, T ). 2 1 vT + (vT )T (T v)I 2 3 v ( , 1.646) (1.647) (1.648) (1.649) (1.650)

1.7.0.2 1.7.0.2.1

Non-conservative form Cartesian index form d dt dvi dt de dt p e k = i vi , 1 = fi i p + j 2 (j vi) k vk ji , 3 1 = pi vi + i (ki T ) + 2 (i vj ) k vk ij i vj , 3 = p(, T ), = e(, T ), = (, T ), = k (, T ) (1.651) (1.652) (1.653) (1.654) (1.655) (1.656) (1.657)

1.7.0.2.2

Gibbs form
d dt = T v, = f p + 2
T

(1.658) vT + (v)T 1 (T v)I 2 3

T

dv dt de dt

, : vT ,

(1.659) (1.660) (1.661) (1.662) (1.663)

= pT v + T (k T ) + 2

vT + (vT )T 1 (T v)I 2 3

p = p(, T ), e = e(, T ), = (, T ),

115

k = k (, T ).

(1.664)

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.

1.8

Incompressible Navier-Stokes equations with constant properties

If we make the assumption, which can be justied in the limit when uid particle velocities are small relative to the velocity of sound waves in the uid, that density changes following a particle are negligible (that is d 0), the Navier-Stokes equations simplify considerably. dt Note that this does not imply the density is constant everywhere in the ow. Our assumption allows for stratied ows, 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
o + i (vi ) = 0, (1.665)

Expanding the mass equation

we get o + vi i +i vi = 0.
d dt

(1.666)

0
d , dt

We are assuming the rst two terms in the above expression, which form hence the mass equation becomes i vi = 0. Since > 0, we can say i vi = 0.

go to zero;

(1.667)

So for an incompressible uid, the relative expansion rate for a uid particle is zero. 116

1.8.2

Linear momenta
1 j 2 (j vi) k vk ij , 3
=0

Let us rst consider the viscous term:

(1.668) (1.669) (1.670) (1.671) (1.672) (1.673) (1.674) (1.675)

j 2 (j vi)

j ( (i vj + j vi )) , since is constant here (j i vj + j j vi ) , i j vj +j j vi ,

=0

j j vi ,

Everything else in the linear momenta equation is unchanged; hence we get o vi + vj j vi = fi i p + j j vi . (1.676)

Note that in the incompressible constant viscosity limit, the mass and linear momenta equations 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 specic heat at constant pressure, cp is nearly identical to the specic heat at constant volume cv as long as the ratio 2 T p /T //cp << 1. Here p is the coecient of isobaric expansion, and T is the coecient 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 compressible gas there are no such problems. For an incompressible liquid whose specic 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 . dt

(1.677)

Imposing our constitutive equations and assumption of incompressibility onto this, we get 1 d (cT + eo ) = p i vi i (ki T ) + 2 (i vj ) k vk ij i vj , dt 3
=0 =0

(1.678)

117

dT dt

= ki i T + 2(i vj ) i vj ,
= ki i T + 2 (i vj ) (i vj ) + [i vj ] ,
sym. sym. antisym.

(1.679)

(1.680) (1.681) (1.682)

= ki i T + 2(i vj ) (i vj ) ,

For incompressible ows with constant properties, the viscous dissipation function reduces to = 2(i vj ) (i vj ) . (1.683) It is a scalar function and obviously positive for > 0 since it is a tensor inner product of a tensor with itself.

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, (1.684) dv = f p + 2 v, (1.685) dt dT c = k 2 T + . (1.686) 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 diusion

d = t and = 0; hence the energy equation can Note for a static uid (vi = 0), we have dt be written in a familiar form T = 2 T. (1.687) t k Here = c is dened as the thermal diusivity. For one dimensional cases where all variation is in the x2 direction, we get

2T T = 2. t x2

(1.688)

Compare this to the momentum equation for a very specic form of the velocity eld, namely, vi (xi ) = v1 (x2 , t). When we also have no pressure gradient and no body force, the linear momenta principle reduces to 2 v1 v1 = 2. (1.689) t x2 118

Here = is the momentum diusivity. This equation has an identical form to that for one-dimensional energy diusion. 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 equations. 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 Newtonian behavior, satises Stokes assumption, and has constant viscosity, thermal conductivity, and specic heat:
o + i (vi ) = 0, o (vi ) + j (vj vi ) = fi i p +j 1 o e + vj vj 2 1 + i vi e + vj vj 2 1 2 (j vi) k vk ji 3 (1.690)

(1.691)

= vi fi i (pvi ) + ki i T 1 +i 2 (i vj ) k vk ij 3 p = RT, e = cv T + e . vj , (1.692) (1.693) (1.694)

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 , . . .). (1.695)

It is desirable for many reasons to reduce the number of parametric dependencies of these solutions. Some of these reasons include identication of groups of terms that truly govern the features of the ow, eciency of presentation of results, and eciency 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: 119

vo P o
o

Figure 1.31: Figure of known ow from innity approaching body with characteristic length L. reduce variables so that their scaled value is near unity, demonstrate that certain physical mechanisms may be negligible relative to other physical mechanisms, and simplify initial and boundary conditions. In forming dimensionless equations, one must usually look for characteristic length scale L, and 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 given vo , L tc =
L , vo

given vo , tc L = vo tc . If for example our physical problem involves the ow over a body of length L (and whose other dimensions are of the same order as L), and freestream conditions are known to be p = po , vi = (vo , 0, 0)T , = o , as sketched in Figure 1.31, Knowledge of freestream pressure and density xes all other freestream thermodynamic variables, e.g. e, T , via the thermodynamic relations. For this problem, let the subscript represent a dimensionless variable. Dene the following scaled dependent variables: = , o p = p , po vi = vi , vo T = o R T, po e = o e. po (1.696)

Dene the following scaled independent variables: xi = xi , L 120 t = vo t. L (1.697)

With these denitions, the operators must also be scaled, that is, dt vo vo = = = o , t dt t L t L L o = o . vo dxi 1 1 i = = = = i , xi dxi xi L xi L i = Li . o =

(1.698)

1.9.1

Mass

Let us make these substitutions into the mass equation: o + i (vi ) vo 1 o (o ) + i (o vo vi ) L L o vo (o + i ( vi )) L o + i ( vi ) = 0, = 0, = 0, = 0. (1.699) (1.700) (1.701) (1.702)

The mass equation is unchanged in form when we transform to a dimensionless version.

1.9.2

Linear momenta
o (vi ) +j (vj vi ) = fi i p vo o (o vo vi ) L

We have a similar analysis for the linear momenta equation.

1 +j 2 (j vi) k vk ji 3

(1.703)

1 1 + j (o vo vj vo vi ) = o fi i (po p ) L L 1 2 + j (j vo vi) k vo vk ji L L 3L 2 o vo o ( vi ) L o v 2 po + o j ( vj vi ) = o fi i (p ) L L 1 vo + 2 j 2 (j vi) k vk ji , L 3 fi L po o ( vi ) + j ( vj vi ) = (p ) 2 2 i vo o vo 121

(1.704)

(1.705)

1 2 j (j vi) k vk ji . o vo L 3

(1.706)

With this scaling, we have generated three distinct dimensionless groups of terms which drive the linear momenta equation: fi L , 2 vo po , 2 o vo and . o vo L (1.707)

These groups are closely related to the following groups of terms, which have the associated interpretations indicated: Froude number F r : 40 With the body force per unit mass fi = g g i , where g > 0 is the gravitational acceleration magnitude and g i is a unit vector pointing in the direction of gravitational acceleration, F r2
2 ow kinetic energy vo = . gL gravitational potential energy

(1.708)

Mach number Mo : 41 With the Mach number Mo dened as the ratio of the ambient velocity to the ambient sound speed, and recalling that for a calorically perfect ideal po 2 gas that the square of the ambient sound speed, a2 o is ao = o , where is the ratio of cp specic heats = cv = (1 + R/cv ), we have
2 Mo 2 2 2 2 vo vo vo ow kinetic energy o vo = = = . = po 2 ao o po RTo thermal energy

(1.709)

Here we have taken To = po /o /R. Reynolds number Re: We have Re With these denitions, we get o ( vi ) + j ( vj vi ) = 1 1 1 g (p ) i 2 i F r2 Mo 1 2 + j (j vi) k vk ji . Re 3 o v 2 dynamic pressure o vo L = voo = . L viscous stress (1.710)

(1.711)

The relative magnitudes of F r , Mo , and Re play a crucial role in determining which physical mechanisms are most inuential in changing the uids linear momenta.
William Froude, 1810-1879, English engineer and naval architect, Oxford educated. 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 inuenced Einstein.
41 40

122

1.9.3

Energy

The analysis is of the exact same form, but more tedious, for the energy equation. 1 o e + vj vj 2 1 + i vi e + vj vj 2

= ki i T i (pvi ) 1 +i 2 (i vj ) k vk ij vj 3 +vi fi ,

(1.712)

vo 1 2 po e + vo o o vj vj L o 2 1 2 po 1 e + vo vj vj + i o vo vi L o 2

2 1 vo k po 1 o vo po = i vi e + i i T + po vj vj L o 2 o L2 o R po vo i (p vi ) L 2v 2 1 + 2 o i (i vj ) k vk ij vj L 3 +o vo fi vi , (1.714)

2 1 vo o vo po o e + po vj vj L o 2 o

k po T i i L2 o R 1 i (po p vo vi ) L 1 2 + i (i vo vj ) k vo vk ij vo vj L L 3 +o vo vi fi , (1.713)

o e + +i vi e +

1 2 1 2

2 vo po vj vj o

2 vo po vj vj o

k i i T LRo vo i (p vi ) 2v 2 L 1 + 2o i o L o vo p o fi L + po vi .
o

1 (i vj ) k vk ij vj 3 (1.715)

Now examining the dimensionless groups, we see that k cp 1 k cp 1 1 k = = = . LRo vo cp R Lo vo cp cp cv o vo L P r 1 Re 123 (1.716)

Here we have a new dimensionless group, the Prandtl Pr We also see that fi L
po o

42

number, P r , where (1.717)

cp = k

o k o c p

momentum diusivity = . energy diusivity

2 2 Mo gLg i vo gL g , g = = po po 2 i 2 i v F r o o o 2 vo 2 po = Mo , o

(1.718) (1.719) (1.720)

2 2 2vo 2 vo 1 L 1 2 Mo . = po po = 2 2 L o v o o o v o L o Re

So the dimensionless energy equation becomes 1 2 vj vj o e + Mo 2 1 2 +i vi e + Mo vj vj 2

1 1 i i T 1 P r Re i (p vi ) 1 M2 +2 o i (i vj ) k vk ij vj Re 3 2 Mo + g i vi . (1.721) F r2

1.9.4

Thermal state equation

po p = o R p = T . po T , o R (1.722) (1.723)

1.9.5

Caloric state equation

po po e = c v T + e , o o R o e cv T + , e = R po 1 o e e = T + 1 po (1.724) (1.725) (1.726)

unimportant

Ludwig Prandtl, 1875-1953, German mechanician and father of aerodynamics, primarily worked at University of G ottingen, discoverer of the boundary layer, pioneer of dirigibles, and advocate of monoplanes.

42

124

 For completeness, we retain the term pooe . It actually plays no role in this non-reactive ow since energy only enters via its derivatives. When ows with chemical reactions are modelled, this term may be important.

1.9.6

Upstream conditions
vi = (1, 0, 0)T .

Scaling the upstream conditions, we get p = 1, = 1, (1.727)

With this we then get secondary relationships T = 1, e = 1 o e . + 1 po (1.728)

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 . Our scaled system however has only six independent parameters: Re, P r, Mo , F r, , o e . po (1.730) (1.729)

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 simplied equation. We will consider two cases here, what is known as Bernoullis 43 equation and Croccos 44 equation. In a later chapter on rotational ows, we will also consider the Helmholtz equation and Kelvins theorem, which are also rst integrals in special cases.

1.10.1

Bernoullis equation

What we commonly call Bernoullis equation is really a rst integral of the linear momenta principle. Under dierent assumptions, we can get dierent avors of Bernoullis equation. A rst integral of the linear momenta principle exists under the following conditions:
Daniel Bernoulli, 1700-1782, Dutch-born Swiss mathematician of the prolic 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 uid mechanical principle in the 1738 Hydrodynamica, in competition with his fathers 1738 Hydraulica. 44 Luigi Crocco, 1909-1986, Sicilian-born, Italian applied mathematician and theoretical aerodynamicist and rocket engineer, taught at University of Rome, Princeton, and Paris.
43

125

 viscous stresses are negligible relative to other terms, ij 0, the uid is barotropic, p = p() or = (p). , where is a known potential body forces are conservative, so we can write fi = i function, and either the ow is irrotational, k = the ow is steady, o = 0. First consider the general linear momenta equation: 1 1 o vi + vj j vi = i p + fi + j ji (1.731)
kij i vj

= 0, or

Now use our vector identity to rewrite the convective term, and impose our assumptions above to arrive at 1 1 o vi + i vj vj ijk vj k = i p i . (1.732) 2 Now let us dene, 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 uid, we dene as p(xi ,t) dp . (1.733) (p(xi , t)) ( p) po Note that in the special case of incompressible ow that = p/. Recalling Leibnizs rule, d dt we let
xi x=b(t) x=a(t)

f (x, t) dx =
d dt

x=b(t) x=a(t)

f db da dx + f (b(t), t) f (a(t), t), t dt dt

(1.734)

play the role of

p(xi ,t) po

to get
p(xi ,t) po

= xi xi

dp 1 p 1 po + = ( p) (p(xi , t)) xi (po ) xi

=0

xi

1 dp . ( p)
=0

(1.735)

As po is constant, and the integrand has no explicit dependency on xi , we get 1 p = . xi (p(xi , t)) xi So, our linear momenta principle reduces to o vi + i 1 vj vj 2
ijk vj k

(1.736)

= i i .

(1.737)

Consider now some special cases: 126

1.10.1.1

Irrotational case

If the uid 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 = v m . Note that if the velocity takes this form, then the vorticity is k =
klm l m .

(1.738)

(1.739)

Since klm is antisymmetric and l m is symmetric, their tensor inner product must be zero; hence, such a ow is irrotational. So the linear momenta principle reduces to 1 (j )(j ) = i i , 2 1 = 0, o + (j )(j ) + + 2 1 = f (t). o + (j )(j ) + + 2 o i + i (1.740) (1.741) (1.742) (1.743) 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 = 0. Rearranging the steady linear momenta equation, we get i 1 = vj vj + i + i 2
ijk vj k .

(1.744)

Regrouping, and taking the inner product of both sides with vi , we get vi i 1 vj vj + + 2 = vi
ijk vj k ,

(1.745) (1.746) (1.747)

= ijk vi vj k , = 0.

The term on the right hand side is zero because it is the tensor inner product of a symmetric and antisymmetric tensor. For a local coordinate system which has component s aligned with the velocity vector v i , 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. b [] 127

(1.748)

Forming this dot product yields vs For vs = 0, we get that 1 = 0. vj vj + + s 2 (1.749)

1 = C (n, b). vj vj + + 2 On a particular streamline, the function C (n, b) will be a constant.

(1.750)

1.10.1.2.2 Lamb surfaces We can extend the idea of integration along a streamline to describe what are known as Lamb surfaces 45 by again considering the steady, inviscid linear momentum principle with conservative body forces: i Now taking the quantity B to be 1 B vj vj + + , 2 the linear momentum principle becomes i B =
ijk vj k

1 vj vj + + 2

ijk vj k .

(1.751)

(1.752)

(1.753)

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 ow, named after Sir Horace Lamb, the British uid 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 = C. vj vj + + (1.754) 2 = gz z The constant is truly constant throughout the ow eld. With = p/ here and (with gz > 0, and rising z corresponding to rising distance from the earths surface, we get = gz k) for a constant gravitational eld, and v the magnitude of the velocity f = vector, we get 1 2 p v + + gz z = C. (1.755) 2
45 Horace Lamb, 1849-1934, English uid mechanician, rst studied at Owens College Manchester followed by mathematics at Cambridge, taught at Adelaide, Australia, then returned to the University of Manchester, prolic writer of textbooks.

128

1.10.2

Croccos theorem

It is common, especially in texts on compressible ow, to present what is known as Croccos theorem. The many dierent versions presented in many standard texts are non-uniform and often of unclear validity. Its utility is conned mainly to providing an alternative way of expressing the linear momentum principle which provides some insight into the factors which inuence uid motion. In special cases, it can be integrated to form a more useful relationship, similar to Bernoullis equation, between fundamental uid variables. The heredity of this theorem is not always clear, though, as we shall see it is nothing more than a combination of the linear momentum principle coupled with some denitions from thermodynamics. Its derivation is often conned to inviscid ows. Here we will rst present a result valid for general viscous ows 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: 1 1 o vi + vj j vi = i p + fi + j ji (1.756)

Now, as before in the development of Bernoullis equation, use our vector identity to rewrite the convective term, but retain the viscous terms to get o vi + i 1 vj vj 2
ijk vj k

1 1 = i p + fi + j ji . 1 1 vi i p + vi fi + vi j ji .

(1.757)

Taking the dot product with vi , and rearranging, we get o Again, since so we get 1 1 vi vi + v i i vj vj = 2 2
ijk ijk vi vj k

(1.758)

is antisymmetric and vi vj is symmetric, their tensor inner product is zero, o

1 1 1 1 vi vi + v i i vj vj = vi i p + vi fi + vi j ji . (1.759) 2 2 Now recall the Gibbs relation from thermodynamics, which in eect denes the entropy s: p T ds = de 2 d. (1.760) Also recall the denition of enthalpy h: p h=e+ . Dierentiating the equation for enthalpy, we get p 1 dh = de + dp 2 d. 129 (1.762) (1.761)

Eliminating de in favor of dh in the Gibbs equation gives 1 T ds = dh dp. (1.763)

If we choose to apply this relation to the motion following a uid particle, we can say then that ds dh 1 dp T = . (1.764) dt dt dt Expanding, we get 1 T (o s + vi i s) = o h + vi i h (o p + vi i p). Rearranging, we get 1 1 T (o s + vi i s) (o h + vi i h) + o p = vi i p. (1.766) (1.765)

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: o 1 1 1 1 vi vi + v i i vj vj = T (o s + vi i s) (o h + vi i h) + o p + vi fi + vi j ji . (1.767) 2 2

Rearranging slightly, noting that vi vi = vj vj , and assuming the body force is conservative so , we get that fi = i 1 + v i i h + 1 vj vj + = T (o s + vi i s) + 1 o p + 1 vi j ji . o h + vj vj + 2 2 (1.768)

is Note that here we have made the common assumption that the body force potential independent of time, which allows us to absorb it within the time derivative. If we dene, as is common, the total enthalpy ho as 1 ho = h + vj vj + , 2 we can then state 1 1 o ho + vi i ho = T (o s + vi i s) + o p + vi j ji , T dho ds 1 p 1 T = T + + v T dt dt t (1.770) (1.771) (1.769)

We can use the rst law of thermodynamics written in terms of entropy, ds = (1/T )i qi + dt (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. (1.772) dt Thus, we see that the total enthalpy of a uid particle is inuenced by energy and momentum diusion as well as an unsteady pressure eld. 130

1.10.2.2

Extended Croccos theorem

With a slight modication of the preceeding analysis, we can arrive at the extended Croccos theorem. Begin once more with an earlier version of the linear momenta principle: o vi + i 1 vj vj 2
ijk vj k

1 1 = i p + fi + j ji .

(1.773)

Now assume we have a functional representation of enthalpy in the form h = h(s, p). Then we get dh = h h ds + dp. s p p s (1.775) (1.774)

We also thus deduce from the Gibbs relation dh = T ds + (1/)dp that h s = T,

p

h p

1 = .

(1.776)

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 = xi s or i h = s h + xi p p . xi (1.777)

h h i s + i p. s p p s

(1.778)

Substituting known values for the thermodynamic derivatives, we get 1 i h = T i s + i p. (1.779)

We can use this to eliminate directly the pressure gradient term from the linear momentum equation to obtain then o vi + i 1 vj vj 2
ijk vj k

1 = T i s i h + fi + j ji .

(1.780)

, Rearranging slightly, and again assuming the body force is conservative so that f i = i we get the extended Croccos theorem: 1 = T i s + o vi + i h + vj vj + 2
ijk vj k

1 + j ji .

(1.781)

, we write the extended Croccos v v + Again, employing the total enthalpy, ho = h + 1 2 j j theorem as 1 (1.782) o vi + i ho = T i s + ijk vj k + j ji . 131

1.10.2.3

Traditional Croccos theorem

For a steady, inviscid ow, the extended Croccos theorem reduces to what is usually called Croccos theorem: i ho = T i s + ijk vj k , ho = T s + v . If the ow is further requried to be homeoentropic, we get i ho =
ijk vj k .

(1.783) (1.784)

(1.785)

Similar to Lamb surfaces, we nd that surfaces on which ho is constant are parallel to both the velocity and vorticity vector elds. Taking the dot product with vi , we get vi i ho = vi ijk vj k , = ijk vi vj k , = 0. Integrating this along a streamline, as for Bernoullis equation, we nd ho = C (n, b), 1 = C (n, b), h + vj vj + 2 (1.789) (1.790) (1.786) (1.787) (1.788)

so we see that the stagnation enthalpy is constant along a streamline and varies from streamline to streamline. If the ow is steady, homeoentropic, and irrotational, the total enthalpy will be constant throughout the oweld: 1 = C. h + vj vj + 2 (1.791)

132

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 uids, sometimes called vortex dynamics. The two most common quantities which are used to characterize rotating uids are the vorticity vector = v, and 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 uid, 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 uid about an axis induces an acceleration in that a uid 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 x = r cos , x2 + y 2 , y , y = r sin , = tan1 x z = z , z = z. r= (2.1) (2.2) (2.3)

Most of the basic distinctions between the two systems can be understood by considering twodimensional geometries. The representation of an arbitrary point in both two-dimensional (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, 133

e e y
r
r

j i

Figure 2.1: Representation of a point in Cartesian and cylindrical coordinates along with unit vectors for both systems.

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 dierences in the formulations of the governing equations, namely the appearance of centripetal acceleration, and Coriolis
1

acceleration

in the cylindrical representation. Consider the representations of the velocity vector v in both coordinate systems: v = ui + v j, or v = v r er + v e . (2.4) (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 that v u i v j = i+u + j+v , t t t t t
=0 =0

(2.6)

v vr er v e = er + v r + e + v . t t t t t

(2.7)

1 Gaspard Gustave de Coriolis, 1792-1843, Paris-born mathematician, taught with Navier, introduced the terms work and kinetic energy with modern scientic meaning, wrote on the mathematical theory of billiards.

134

sin i
r

sin j

cos i

cos j

Figure 2.2: Geometrical representation of cylindrical unit vectors in terms of Cartesian unit vectors. Now the time derivatives of the Cartesian basis vectors is zero, as they are dened 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. t t t (2.8)

However the time derivative of the cylindrical basis vectors does change with time for particles in motion! To see this, let us rst relate er and e to i and j. From the sketch of Figure 2.2, it is clear that er = cos i + sin j, e = sin i + cos j. This is a linear system of equations. We can use Cramers rule to invert to nd i = cos er sin e , j = sin er + cos e . Now, examining time derivatives of the unit vectors, we see that er t = sin = and e t = cos = i sin j, t t (2.15) (2.16) e , t i + cos j, t t (2.13) (2.14) (2.11) (2.12) (2.9) (2.10)

er . t

so there is a formal variation of the unit vectors with respect to time as long as the angular velocity = 0. So the acceleration vector is t vr v v = er + v r e + e v er , t t t t t vr v = v er + + vr e . t t t t 135 (2.17) (2.18)

v dt = ds

r
d

Figure 2.3: Sketch of relation of dierential distance ds to velocity in angular direction v . Now from basic geometry, as sketched in Figure 2.3, we have ds = rd, v dt = rd, v = . r t Consequently, we can write the unsteady component of acceleration as

(2.19) (2.20) (2.21)

v vr = t t

2 v r centripetal

er

v t

vr v r

Coriolis

e .

(2.22)

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 Christoel symbols of the transformation. Such terms perhaps contributed to the development of Einsteins theory of general relativity 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. 136

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 = The chain rule gives us r z = + + , x x r x x z r z = + + , y y r y y z r z = + + , z z r z z z Now, we have r x 2x = = 2 = cos , x r 2 x + y2 y 2y r = = sin , = 2 2 y r 2 x +y r = 0, z and y r sin sin = 2 = 2 = , 2 x x +y r r x r cos cos = 2 = = , 2 2 y x +y r r = 0, z and z = 0, x z = 0, y z = 1, z so sin = cos , x r r cos = sin + , y r r = . z z 137 (2.37) (2.38) (2.39) (2.34) (2.35) (2.36) (2.31) (2.32) (2.33) (2.28) (2.29) (2.30) (2.24) (2.25) (2.26) (2.27) i+ j + k. x y z (2.23)

2.1.2.1

Grad

So now we are prepared to write an explicit form for in cylindrical coordinates: = cos sin (cos er sin e ) r r i
x

+ sin

cos + (sin er + cos e ) + ez , r r z j

y

(2.40)

sin cos sin cos er + + r r r sin2 cos2 e + + + ( sin cos + sin cos ) r r r + ez , z 1 = er + e + ez . r r z cos2 + sin2

(2.41) (2.42)

We can now write a simple expression for the convective component, v T , of the acceleration vector: v + + vz . (2.43) v T = vr r r z 2.1.2.2 Div

The divergence is straightforward. In Cartesian coordinates we have T v = u v w + + . x y z (2.44)

In cylindrical, we replace derivatives with respect to x, y, z with those with respect to r, , z , so v cos v w u sin u + sin + + . (2.45) T v = cos 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 , v = vr sin + v cos , w = vz . Substituting and taking partials, we nd that T v = cos cos
v vr sin vr v sin cos v cos sin vr sin r r r
A B C

(2.46) (2.47) (2.48)

138

+ sin sin + vz . z

cos v vr v vr + + cos sin v + cos vr + cos sin r r r

A B C

(2.49)

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 T v = which is often rewritten as T v = 1 1 v vz (rvr ) + + . r r r z 1 2 2 + . r 2 2 z 2 (2.51) vr vr 1 v vz + + + , r r r z (2.50)

Using the same procedure, we can show that the Laplacian operator transforms to 2 = 1 r r r r + (2.52)

2.1.3

Incompressible Navier-Stokes equations in cylindrical coordinates

Leaving out some additional details of the transformations, we nd that the incompressible Navier-Stokes equations for a Newtonian uid with constant viscosity and body force conned to the z direction are 0 =
2 vr v vr vr vr v + vr + + vz t r r r z v v vr v v v v + + + vz + vr t r r r z vz vz v vz vz + vr + + vz t r r z

1 v vz 1 (rvr ) + + , r r r z 2 1 p vr 2 v 2 = + vr 2 2 r r r 1 1 p 2 vr v = + 2 v + 2 2 r r r 1 p = + 2 vz gz . z

(2.53) , (2.54) (2.55) , (2.56)

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 dene to be a uid rotating as a solid body. Let us assume incompressible ow, so T v = 0, 139

Figure 2.4: Sketch of a uid rotating as a pure solid body. assume a simple velocity eld, and ask what forces could have given rise to that velocity eld. We will simply use z for the azimuthal coordinate instead of z here. Take vr = 0, v = r , 2 vz = 0. (2.57)

The kinematics of this ow are simple and sketched in Figure 2.4, Here is now dened as a constant. The velocity is zero at the origin and grows in amplitude with linear distance from the origin. The ow is steady, and the streamlines are circles centered about the origin. Obviously, as r , the theory of relativity would suggest that such a ow would break down as the velocity approached the speed of light. In fact, one would nd as well that as the velocities approached the sound speed that compressibility eects would become important far before relativistic eects. Whatever the case, does this assumed velocity eld satisfy incompressible mass conservation? ? 1 1 r + (0) = 0. (r (0)) + (2.58) r r r 2 z
=0

Obviously it does. Next let us consider the acceleration of an element of uid and the forces which could give rise to that acceleration. First consider the material derivative for this ow d v v = + vr + + vz = . dt t r r z r
=0 =0 =0

(2.59)

But the only non-zero component of velocity, v , has no dependency on , so the material v derivative of velocity d = 0. dt 140

Consider now the viscous terms for this ow. We recall for an incompressible Newtonian uid that ij = 2(i vj ) + k vk ij ,
=0

(2.60) (2.61) (2.62) (2.63) (2.64)

j ij

= (i vj + j vi ) , = (j i vj + j j vi ) , = i j vj +j j vi ,
=0

= v

We also note that = ijk j k = ijk j kmn m vn , = kij kmn j m vn , = (im jn in jm ) j m vn , = j i vj j j vi , = i j vj j j vi ,

=0

(2.65) (2.66) (2.67) (2.68) (2.69) (2.70)

= j j vi . Comparing, we see that for this incompressible ow, T T

T

= ( ).

(2.71)

Now, using relations that can be developed for the curl in cylindrical coordinates, we have for this ow that r = = z = = = v 1 vz = 0, r z vr vz = 0, z r 1 1 vr (rv ) , r r r r 1 r , r r 2 . (2.72) (2.73) (2.74) (2.75) (2.76)

So the ow has a constant rotation rate, . Since it is constant, its curl is zero, and we have T for this ow that T T = 0. We could just as well show for this ow 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. 141

Hence, the three linear momenta equations in the cylindrical coordinate system reduce to the following:
2 v 1 p = , r r 1 1 p , 0 = r 1 p 0 = gz . z

(2.77) (2.78) (2.79)

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. If we take p = p(r, , z ) and p(ro , , zo ) = po , then dp = = = = p po = p(r, z ) = p p p dr + d + dz, r z 2 v dr + 0d gz dz, r 2 r 2 dr gz dz, 4r 2 r dr gz dz, 4 2 2 2 (r ro ) gz (z zo ), 8 2 2 2 po + (r ro ) gz (z zo ). 8 (2.80) (2.81) (2.82) (2.83) (2.84) (2.85)

Now on a surface of constant pressure we have p(r, z ) = p . So 2 2 2 (r ro ) gz (z zo ), 8 2 2 2 (r ro ), gz (z zo ) = po p + 8 po p 2 2 2 z = zo + + (r ro ). gz 8gz p = po + (2.86) (2.87) (2.88)

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 eld and eliminate using o and dening vo = r : v = r 2 2 1 2 1 2 p v + gz z = po vo + gz zo = C. 2 2 (2.89)

This looks very similar to the steady irrotational incompressible Bernoulli equation in which 1 2 v 2 + gz z = K . But there is a dierence in the sign on one of the terms. Now add v p+ 2 142

to both sides of the equation to get 1 2 2 + gz z = C + v . p + v 2 (2.90)

Now since v = r ,vr = 0, we have lines of constant r as streamlines, and v is constant on 2 those streamlines, so that we get 1 2 p + v + gz z = C , 2 on a streamline. (2.91)

Here C varies from streamline to streamline. 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 nd = =
C 2 0

vT dr, 1 a (ad ), 2

(2.92) (2.93) (2.94)

= a2 .

2.3

Ideal irrotational vortex

o , 2r

Now let us perform a similar analysis for the following velocity eld: vr = 0, v = vz = 0. (2.95)

The kinematics of this ow 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 ow is r = = z = = = 1 vz v = 0, r z vr vz = 0, z r 1 vr 1 (rv ) , r r r 1 o r , r r 2r 1 o = 0! r r 2 (2.96) (2.97) (2.98) (2.99) (2.100)

This ow eld, which seems the epitome of a rotating ow, is formally irrotational as it has zero vorticity. What is happening is that a uid 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 undened. 143

dr = a d e

Figure 2.5: Sketch of an ideal irrotational point vortex. The circulation for this ow about a circle of radius a is = = =
C 2 0 2 0

vT dr, v (ad ), o ad, 2a

(2.101) (2.102) (2.103) (2.104)

= o .

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 ow happens to be incompressible with constant properties and in a potential eld where the T gravitational body force per unit mass is gz k. Recall again that T = ( ), = 0 for this ow. Note also that because there is and that since = 0 that T 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 uids, we nd that r = r r v 1 vr + r r = r r o 2r 2 = o . r 2
T

(2.105)

The equations of motion reduce to the same ones as for the ideal rotational vortex:
2 1 p v = , r r

(2.106)

144

1 1 p , (2.107) r 1 p 0 = gz . (2.108) z Once more we can deduce a pressure eld 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 0 =
=0

= = p po = p+ 2 o 1 + gz z = 2 8 r 2 1 2 + gz z = p + v 2

dr gz dz, r 2 o dr gz dz, 2 4 r 3 1 1 2 o 2 gz (z zo ), 2 2 8 r ro 2 o 1 + gz zo , po + 2 2 8 ro 1 2 po + vo + gz zo = C 2

2 v

(2.110) (2.111) (2.112) (2.113) (2.114) (2.115)

This is once again Bernoullis equation. Here it is for an irrotational ow eld that is also time-independent, so the Bernoulli constant C is truly constant for the entire ow eld and not just along a streamline. On isobars we have p = p which gives us p po = 2 1 1 o 2 gz (z zo ), 2 2 8 r ro 1 po p 2 1 z = zo + + 2o 2 2 gz 8 gz r ro (2.116) (2.117)

Note that the pressure goes to negative innity 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 eld evolves in a general uid.

2.4.1

General development
(vT )v = vT v + v, 2 145

First, we recall some useful vector identities: (2.118)

 (a b) = (bT )a (aT )b + a(T b) b(T a), () = 0, T ( v) = T = 0. We start now with the linear momenta principle for a general uid: 1 1 v + (vT )v = f p + T t
T

(2.119) (2.120) (2.121)

(2.122)

We expand the term (vT )v and then apply the curl operator to both sides to get v vT v 1 1 + v = f p + T + t 2
T

(2.123)

This becomes, via the linearity of the various operators, vT v ( v) + t 2

=0

+ v = f

1 1 T p +

(2.124)
T

Using our vector identity for the term with two cross products we get 1 1 T + (vT ) ( T )v+ ( T v )v(T ) = f p + t =0 d = 1 dt = d
dt

Rearranging, we have
d d dt dt 1 d d 2 dt dt d dt = ( T )v + f = = 1 p + 1 T
T

(2.125)
,
T

(2.126) , , (2.127) (2.128)

T 1 1 v+ f 1 1 T v+ f
1 p

1 1 p + 1 1 p +

1 T 1 T

Now consider the term

ijk j

. In Einstein notation, we have 1 1 j k p 2 (j )(k p) , 1 1 = ijk j k p 2 ijk (j )(k p), =

ijk =0

1 k p

(2.129) (2.130) (2.131)

1 = 2 p.

Multiplying both sides by , we write the nal general form of the vorticity transport equation as T d 1 1 = T v + f + 2 p + T . (2.132) dt B A C D 146

Here we see the evolution of the vorticity scaled by the density is aected by four physical processes, which we describe in greater detail directly, namely A: bending and stretching of vortex tubes, , then f is conservative, and f = B : non-conservative body forces (if f = = 0. For example f = gz k gives = gz z ), C : non-barotropic, also known as baroclinic, eects (if a uid is barotropic, then dp p = p() and p = dp = 0.), and thus p = d d D : viscous eects.

2.4.2

Limiting cases

The Helmholtz vorticity transport equation (2.132) reduces signicantly in special limiting cases. 2.4.2.1 Incompressible with conservative body force

In the limit in which the ow is incompressible and the body force is conservative, Eq. (2.132) reduces to the following d 1 = ( T )v + (T )T . dt 2.4.2.2 (2.133)

Incompressible with conservative body force, isotropic, Newtonian, constant viscosity

Now if we further require that the uid 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 )),
=0

(2.134) (2.135) (2.136) (2.137) (2.138) (2.139)

= = =

ijk j m (m vk + k vm ), ijk j (m m vk + m k vm ), ijk j (m m vk + k m vm ), =0 ijk j vk ,

= m m = 2 . So we get, recalling that = /,

d = ( T )v + 2 . dt 147

(2.140)

2.4.2.3

Two-dimensional, incompressible with conservative body force, 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 d3 2 3 2 3 2 = 3 = + . dt x2 x2 1 2 (2.141)

If the ow is further inviscid = 0, we get d3 = 0, (2.142) dt and we nd 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) eects

If a uid 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 specic heats, and the o subscript indicates a constant value. Such a gas is barotropic. for such a uid, we must have by the dp i . Hence p and are vectors which point in the same direction. chain rule that i p = d 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 nite uid 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 uid element to rotate about its center of mass, hence no vorticity would be generated by this force. For a baroclinic uid, we do not have p = p(); hence, we must expect that p points in a dierent 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 uid element. Hence, it generates a torque, a tendency to rotate the uid about G, and vorticity. 2.4.3.2 Bending and stretching of vortex tubes: three-dimensional eects

Now let us consider generation of vorticity by three-dimensional eects. Such eects 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 If we consider a coordinate system which is oriented with the vorticity eld as sketched in Figure 2.8, we will get many simplications. We take the following directions 148

low

plow
high

G phigh
high

F net pressure

Figure 2.6: Isobars and isochores, center of mass G, and center of pressure for barotropic uid.

low

plow

phigh

F net pressure

Figure 2.7: Isobars and isochores, center of mass G, and center of pressure for baroclinic uid. 149

 b s

curve everywhere parallel to vorticity vector

locally orthogonal coordinate system s,n,b

Figure 2.8: Local orthogonal intrinsic coordinate system oriented with local vorticity eld. s: the streamwise direction parallel to the vorticity vector, n: the principle normal direction, pointing towards the center of curvature, b: the biorthogonal direction, orthogonal to s and n. With this system, we can say that ( T )v = ( s

s n b

= s

v . s

0 0)

v ,

(2.144) (2.145)

So for the inviscid ow we have

v d = s . dt s ds vs = s , dt s dn vn = s , dt s db vb = s . dt s

(2.146)

We have in terms of components (2.147) (2.148) (2.149)

s The term v we know from kinematics represents a local stretching or extension. Just as s a rotating gure skater increases his or her angular velocity by concentrating his or her mass about a vertical axis, so does a rotating uid. The rst 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. 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 ow.

150

unstretched vortex tube

stretched vortex tube

Figure 2.9: Increase in vorticity due to stretching of vortex tube.

2.5

Kelvins circulation theorem

Kelvins circulation theorem describes how the circulation of a material region in a uid changes with time. We rst recall the denition of circulation : =
C

vT dx,

(2.150)

where C is a closed contour. We next take the material derivative of to get d d = dt dt = = = =

C C

vT dx, d dx, dt C dx , dx + vT d dt C dx + vT dx + dx +
T C

(2.151) (2.152) (2.153) (2.154) (2.155)

dvT dt dvT dt dvT dt dvT dt dv dt

vT dv, d 1 T v v , 2
=0

dx.

(2.156)

x Here we note that because we have chosen a material region for our closed contour that d dt must be the uid particle velocity. This then allows us to write the second term as a perfect dierential, 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 T T d 1 1 = f p + T dx. (2.157) dt C

), and the uid If now the uid is inviscid ( = 0), the body force is conservative (f = is barotropic ((1/)p = ), then we have d = dt
C

151

dx,

(2.158)

= =

C C

+ dx, T + . d
=0

(2.159) (2.160)

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 circulation theorem which holds that for a uid 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 = 0. This is additional stipulation of two-dimensionality and incompressibility, gives ddt not surprising as the vorticity is closely linked to the circulation via Stokes theorem, which states = vT dx = ( v)T ndA = T ndA. (2.162)
C A A

2.6

Potential ow of ideal point vortices

Consider the uid motion induced by the simultaneous interaction of a family of ideal irrotational point vortices in an incompressible ow eld. Since the ow 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. Since T v = 0, we have or expanding, we have 2 2 2 + 2 + 2 = 0. x2 y z (2.165)

T = 2 = 0,

(2.164)

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 ow eld. The irrotational unsteady Bernoulli equation gives us the time and space dependent pressure eld. This equation is not linear, so we do not expect pressures from elementary solutions to add to form total pressures. 152

v2 = 1 1 h 2 2

1 2h

v1 =

2 2h

Figure 2.10: Sketch of the mutual inuence of two ideal point vortices on each other.

1 hG

G h

Figure 2.11: Sketch showing the center of rotation G. Recalling that the incompressible, three dimensional constant viscosity Helmholtz equation can be written as d = ( T )v + 2 , (2.166) dt we see that a ow which is initially irrotational everywhere in an unbounded uid will always = 0. There is no mechanism to change the vorticity from its uniform be irrotational, as ddt initial value of zero. This even holds for a viscous ow. However, in a bounded medium, the no-slip boundary condition almost always tends to diuse vorticity into the ow 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 . This induced velocity in fact convects vortex 2, to satisfy Kelvins circulation theorem. 2h 2 . Similarly, vortex 2 induces a velocity of vortex 1 of 2 h 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 153

v=

2h

v=

2h

Figure 2.12: Sketch showing a pair of counterrotating vortices of equal strength vortex 2 1 = , 2hG 2 (h hG ) (h hG )1 = hG 2 , h1 = hG (1 + 2 ), 1 . hG = h 1 + 2 (2.167) (2.168) (2.169) (2.170)

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 xed distance from one another.

2.6.2

Image vortex

If we choose to model the uid 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 uids moving at dierent 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 inuence of the point vortex at (x, 0) is sketched in the gure. It generates a small velocity with magnitude d|v| = d d = 2h 2 ( x x)2 + y 2 154 (2.171)

h image vortex main vortex

Figure 2.13: Sketch showing a vortex and its image to simulate an inviscid wall.

y dv dv du ~ ~ (x, y)

u=

1 d 2 dx d x

u=

1 d 2 dx

(x,0)

Figure 2.14: Sketch showing schematic of vortex sheet.

155

Using basic trigonometry, we can deduce that the inuence of the single vortex of dierential strength on each velocity component is du =
y d d y dx = dx, 2 2 2 (( x x) + y ) 2 (( x x)2 + y 2 ) d ( x x) d( x x) dx dv = = dx. 2 2 2 (( x x) + y ) 2 (( x x)2 + y 2 )

(2.172) (2.173)

is a measure of the strength of the vortex sheet. Let us account for the eects of all Here d dx of the dierential 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

u =

lim

d dx

v =

1 d 2 dx ,
1 d , 2 dx d lim dx L 4

arctan

Lx L+x + arctan y y y > 0, 2

if

(2.174)

(2.175) if ln y < 0, (2.176) (L x )2 + y 2 = 0. (L + x )2 + y 2

So the vortex sheet generates no y component of velocity anywhere in the ow eld and two uniform x components of velocity of opposite sign above and below the x axis.

2.6.4

Potential of an ideal vortex

Let us calculate the velocity potential function associated with a single ideal vortex. Consider an ideal vortex centered at the origin, and represent the velocity eld here in cylindrical coordinates: , vz = 0. (2.177) vr = 0, v = 2r Now in cylindrical coordinates the gradient operating on a scalar function gives 1 er + e + ez r r z r 1 r z = v, = 0er + = 0, = , 2r so = + C (r, z ), 2 e + 0ez , 2r (2.178) (2.179) (2.180) (2.181) (2.182)

= 0. 156

=0

3 4

Figure 2.15: Lines of constant potential for ideal irrotational vortex. 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 = In Cartesian coordinates, we have = y arctan 2 x (2.184) . 2 (2.183)

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 = 1 arctan 2 2 arctan 2 = 2 y y1 , x x1 y y2 . x x2 (2.185) (2.186)

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 = y y1 y y2 1 2 arctan arctan + , 2 x x1 2 x x2 157 (2.187)

2 y2 1 y1

x1

x2

Figure 2.16: Two vortices at arbitrary locations. is a legitimate solution. Taking the gradient of , = + so that u(x, y ) = v (x, y ) = 1 2 y y1 2 y y2 , 2 2 (x x1 ) + (y y1 ) 2 (x x2 )2 + (y y2 )2 x x1 x x2 2 + . 2 2 (x x1 ) + (y y1 ) 2 (x x2 )2 + (y y2 )2 (2.189) (2.190) 1 2 1 2 y y1 y y2 2 i 2 2 (x x1 ) + (y y1 ) 2 (x x2 )2 + (y y2 )2 x x1 x x2 2 + j, (2.188) 2 2 (x x1 ) + (y y1 ) 2 (x x2 )2 + (y y2 )2

1 2

Extending this to a collection of N vortices located at (xi , yi ) at a given time, we have the following for the velocity eld:
N

u(x, y ) =
i=1 N

i 2

v (x, y ) =
i=1

i 2

x xi . (x xi )2 + (y yi )2

y yi , (x xi )2 + (y yi )2

(2.191) (2.192)

Now to convect (that is, to move) the k th vortex, we move it with the velocity induced by the other vortices, since vortices convect with the ow. Recalling that the velocity is the k k time derivative of the position uk = dx , vk = dy , we then get the following 2N non-linear dt dt ordinary dierential equations for the 2N unknowns, the x and y positions of each of the N vortices: dxk = dt
N i=1,i=k

i 2

yk y i , (xk xi )2 + (yk yi )2 158

xk (0) = xo k,

k = 1, . . . , N, (2.193)

dyk = dt

N i=1,i=k

i 2

xk x i , (xk xi )2 + (yk yi )2

o yk (0) = yk ,

k = 1, . . . , N.

(2.194)

This set of equations, except for three or fewer point vortices, must be integrated numerically. These equations form what is commonly termed a Biot-Savart 2 3 law. These ordinary dierential equations are highly non-linear and give rise to chaotic motion of the point vortices in general. It is a very similar calculation to the motion of point masses in a Newtonian gravitational eld, except that the essential variation goes as 1/r for vortices and 1/r 2 for Newtonian gravitational elds. Thus the dynamics are dierent. Nevertheless just as calculations 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 eld

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 uid 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 eld. We consider two cases, a single stationary point vortex, and a group of N moving point vortices. 2.6.6.1 Single stationary vortex 1 T v v+ 2 1 2 + 2 2r

p 1 T p = v v + , (2.195) 2 p p = 0+ , (2.196) 2 1 (2.197) p(r ) = p 2 2 . 8 r Note that the pressure goes to negative innity at the origin. This is obviously unphysical. It can be corrected by including viscous eects, which turn out not to substantially alter our main conclusions. 2.6.6.2 Group of N vortices

If we take p = p in the far eld and fi = g = 0, this steady ow gives us

For a collection of N vortices, the ow is certainly not steady, and we must in general retain the time dependent velocity potential in Bernoullis equation yielding 1 p + ()T + = f (t). t 2
2 3

(2.198)

Jean-Baptiste Biot, 1774-1862, Paris-born applied mathematician Felix Savart, 1791-1841, French mathematician who worked on magnetic elds and acoustics.

159

wall streamline wall vortex line

wall vortex line x

v wall streamline

Figure 2.17: Wall streamlines and vortex lines at wall y = 0. 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 nal result is 1 p = p ()T . 2 t (2.199)

So with a knowledge of the velocity eld through , we can determine the pressure eld which must have given rise to that velocity eld.

2.7

Inuence of walls

The Helmholtz equation considers mechanisms that generate vorticity in the interior of a ow. 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 dened at a wall where the velocity is formally zero, but in the neighborhood of the wall, the uid velocity is small but non-zero. We can extrapolate the position of streamlines near the wall to the wall to dene 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 x z 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 gure. Because the ow satises no-slip, we have at the wall u(x, y = 0, z ) = 0, v (x, y = 0, z ) = 0, 160 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 zero at y = 0: u x =
y =0

u z

=
y =0

v x

=
y =0

v z

=
y =0

w x

=
y =0

w z

=0
y =0

(2.201)

Near the wall, the velocity is near zero, so the Mach number is very small, and the ow 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 x +
y =0

v y

+
y =0

w z v y

= 0
y =0

so

(2.202)

=0

=0

= 0.
y =0

(2.203)

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 = u|y=0 +
=0

u y v y

y+
y =0

1 2u 2 y 2 1 2v 2 y 2

y2 + . . . ,
y =0

(2.204)

v = v |y=0 +
=0

y+
y =0

y2 + . . . ,
y =0

(2.205)

=0

w = w |y=0 +
=0

w y

y+
y =0

1 2w 2 y 2

y2 + . . . ,
y =0

(2.206) (2.207)

So we get u = v = w = Now for streamlines, we must have dx dy dz = = . u v w (2.211) u y y + ...,

y =0 2 y =0

(2.208) (2.209) (2.210)

1 v 2 y 2 w y

y2 + . . . , y +....

y =0

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 161

wall streamline and the x axis is dzs tan = dx w = lim = y 0 u

w y y =0 u y y =0

(2.212)

y =0

Now consider the vorticity vector evaluated at the wall: x |y=0 = w y u z v x v z w x u y =

y =0

y =0

w y

,
y =0

(2.213)

=0

y |y=0 =

y =0

= 0,
y =0

(2.214)

=0

=0

z |y=0 =

y =0

y =0

u y

.
y =0

(2.215)

=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 dene the slope of the v vortex line, dz , at the wall in the same fashion as we dene a streamline: dx dzv dx z = = x
u y y =0 w y y =0

y =0

1
dzs dx y =0

(2.216)

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 ow with variation in the x and y directions only. For general three-dimensional ows 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 gure we take the direction n to be normal to the wall. Now for this coordinate system, we have dzs dx Hence, we must have w y = 0.
y =0

y =0

w = u

=
y =0

w y y =0 u y y =0

= 0.

(2.217)

(2.218)

162

v x

Figure 2.18: Coordinate system aligned with wall streamlines and vortex lines. Now consider the viscous traction vector associated with the wall:
yx = yy = yz
u y y =0 v y y =0 v z y =0

+ + +

v x y =0 v y y =0 w y y =0

tj = ni ij = ny yj

u y y =0

0 0

(2.219)

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 nd rst by our denition of the coordinate systems that x = y = 0 at y = 0 and that u v . (2.220) z |y=0 = x y=0 y y=0
=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 ow direction.

163

164

Chapter 3 One-dimensional compressible ow

see Yih, Chapter 6 see Liepmann and Roshko, Chapter 2 see Shapiro, Chapters 4-8 This chapter will focus on one-dimensional ow of a compressible uid. The following topics will be covered: development of generalized one-dimensional ow equations, isentropic ow with area change, ow with normal shock waves, and the method of characteristics for isentropic rarefactions. We will assume for this chapter:
0, z 0; one-dimensional ow. v 0, w 0, y

Friction and heat transfer will not be modelled rigorously. Instead, they will be modelled in a fashion which captures the relevant physics and retains analytic tractability. Further, we will ignore the inuences 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 ow 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 ow can adjust to area changes. 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 ow we wish to consider, ow with area change, heat transfer, and wall friction, is illustrated by the following sketch of a control volume, Figure 3.1. 165

x = x - x1
2

x2 n2

x1 1 u1 A1 P1 e1 n1

nw 2 u2 A2 P2 e2

q
w

Perimeter length = L
Figure 3.1: Control volume sketch for one-dimensional compressible ow with area change, heat transfer, and wall friction. For this ow, we will adopt the following conventions surface 1 and 2 are open and allow uxes of mass, momentum, and energy, surface w is a closed wall; no mass ux through the wall external heat ux qw (Energy/Area/Time: parameter,
W ) m2

through the wall allowed-qw known xed

diusive, longitudinal heat transfer ignored, qx = 0, wall shear w (Force/Area:

N ) m2

allowedw known, xed parameter,

diusive viscous stress not allowed xx = 0, and cross-sectional area a known xed function: A(x).

3.1.1

Mass

Take the overbar 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: x A
t+t

x + 1 A1 (u1 t) 2 A2 (u2 t) . = A
t

(3.1)

166

Rearrange and divide by xt: A

t+t

2 A 2 u2 1 A 1 u1 = 0. x

(3.2)

Taking the limit as t 0, x 0, we get (A) + (Au) = 0. t x If the ow is steady, then d (Au) = 0, dx d dA du Au + u + A = 0, dx dx dx 1 d 1 dA 1 du + + = 0. dx A dx u dx Now integrate from x1 to x2 to get
x2 x1

(3.3)

(3.4) (3.5) (3.6)

d (Au)dx = dx
2 1

x2 x1

0dx,

(3.7) (3.8) (3.9) (3.10)

d (Au) = 0,

2 u2 A2 1 u1 A1 = 0, 2 u2 A 2 = 1 u1 A 1 m = C1 .

3.1.2

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 . dt In discrete form this would be mu|t+t mu|t = Fx , t mu|t+t = mu|t + Fx t. (3.12) (3.13) (3.11)

For a control volume containing uid, 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 uid, 167

 pressure force at surface 2 restrains uid, force due to the reaction of the wall to the pressure force pushes uid if area change positive, and force due to the reaction of the wall to the shear force restrains uid. We write the linear momentum principle as x u A
t+t

x u A

+ [1 A1 (u1 t)] u1 [2 A2 (u2 t)] u2 + (p1 A1 ) t (p2 A2 ) t + ( p (A2 A1 )) t x t. w L Rearrange and divide by xt to get u A
2 2 A 2 u2 2 1 A 1 u1 t x A2 A 1 p2 A2 p 1 A1 . +p w L = x x t+t t

(3.14)

u A

(3.15)

In the limit x 0, t 0 we get A (Au) + w L. Au2 = (pA) + p t x x x In steady state we nd d d dA Au2 = (pA) + p w L, dx dx dx d dA dp dA du A +p w L, Au + u (Au) = p dx dx dx dx dx du dp L u = w , dx dx A L udu + dp = w dx, A 1 L du + dp = w dx, u m 2 u L d + dp = w dx. 2 A Wall shear lowers the combination of pressure and dynamic head. If there is no wall shear, then we have dp = d u2 . 2 (3.23) (3.17) (3.18) (3.19) (3.20) (3.21) (3.22) (3.16)

168

An increase in velocity magnitude decreases the pressure. If there is no area change, dA = 0, and no friction w 0 we have u add product of u and mass equation du dp + = 0, dx dx d u (u) = 0 u, dx (3.24) (3.25)

d u2 + p = 0, (3.26) dx u2 + p = o u2 o + po = C2 . (3.27)

3.1.3

Energy

The rst 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, E2 = E1 + Q W.

(3.28) (3.29)

So for our control volume this becomes the following when we also account for the energy ux in and out of the control volume in addition to the work and heat transfer: x A e + u 2 2 =
t+t

x A

e +

u 2 2

+1 A1 (u1 t) e1 +

u2 1 2

2 A2 (u2 t) e2 +

u2 2 2

x t + (p1 A1 ) (u1 t) (p2 A2 ) (u2 t) . (3.30) +qw L Note: the mean pressure times area dierence does no work because it is acting on a stationary boundary, and the work done by the wall shear force is not included.
1

In neglecting work done by the wall shear force, I have taken an approach which is nearly universal, but fundamentally dicult 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 ow with friction will be consistent with those of other sources. The argument typically used to justify this is that the real uid satises 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 uids 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 x ( the work of the wall shear, one would include the term w L ut) in the energy equation.

169

Rearrange and divide by tx: e A + 2 A 2 u2 e 2 +

u2 2 2 u 2 2 t+t

p2 2

1 A 1 u1

t u2 e1 + 21 +

e A +
p1 1

u 2 2

. = qw L

(3.31)

In dierential form as x 0, t 0 u2 A e + t 2 In steady state: u2 p d Au e + + dx 2 2 2 d u u p p d Au e+ + e+ + + (Au) dx 2 2 dx u2 p d + e+ u dx 2 du 1 dp p d de u +u + 2 dx dx dx dx subtract the product of momentum and velocity du dp u2 +u dx dx de pu d u dx dx de p d 2 dx dx Since e = e(p, ) de = de = dx so the steady energy equation becomes e d e + dx p dp + dx

u2 p + Au e + x 2

= qw L.

(3.32)

= qw L, = qw L, qw L , A qw L = , A = w Lu = , A q w L w Lu = + , A A (qw + w u) L = . m

(3.33) (3.34) (3.35) (3.36) (3.37) (3.38) (3.39) (3.40)

e e d + dp, p p e
p

(3.41) (3.42)

e d + dx p

dp . dx

dp p d (qw + w u) L 2 = , dx dx m

(3.43)

e p2 p e p

d (qw + w u) L . = e dx m p

(3.44)

170

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 T ds = de p d, 2 (3.45)

along with a general caloric equation of state e = e(p, ), from which we get de = e e dp + d. p p (3.46)

Substituting into the Gibbs equation, we get T ds = e p e dp + d 2 d. p p (3.47)

Taking s to be constant and dividing by d, we get 0= Rearranging, we get p so c2 =

e p2 p e p

e p

p e + s

p . 2

(3.48)

(3.49)

d dp (qw + w u) L , c2 = e dx dx m p

(3.50) (3.51)

d (qw + w u) L dp . c2 = e dx dx uA p

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 ow with no heat transfer qw 0. We still allow area change and wall friction allowed (see earlier footnote): u d u2 p e+ + = 0, dx 2 u2 p u2 p o e+ + = eo + o + = C3 , 2 2 o u2 u2 = ho + o = C 3 . h+ 2 2 171 (3.52) (3.53) (3.54)

3.1.4

Summary of equations

We can summarize the one-dimensional compressible ow 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) t x (Au) + Au2 + pA t x u2 p + + Au e + x 2 e p = 0, = p A w L, x (3.55) (3.56) (3.57) (3.58) (3.59)

u2 A e + t 2

= qw L, = e(, p), = p(, T ).

3.1.4.2

Unsteady non-conservative form d dt du dt de dt e p = = = = = (Au), A x p w L + , x A u qw L w Lu p + , x A e(, p), p(, T ). (3.60) (3.61) (3.62) (3.63) (3.64)

3.1.4.3

Steady conservative form d (Au) = 0, dx d dA Au2 + pA = p w L, dx dx d u2 p Au e + + = qw L, dx 2 e = e(, p), p = p(, T ). (3.65) (3.66) (3.67) (3.68) (3.69)

3.1.4.4

Steady non-conservative form d d = (Au), dx A dx du dp w L u = + , dx dx A u 172 (3.70) (3.71)

du qw L w Lu de = p + , dx dx A e = e(, p), p = p(, T ).

(3.72) (3.73) (3.74)

In whatever form we consider, we have ve equations in ve 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 3.9
Flow of Air with Heat Addition Given: Air initially at p1 = 100 kP a, T1 = 300 K , u1 = 10 m s ows 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 ux qw along the entire surface of the duct. At the end of the duct the ow has p2 = 100 kP a, T2 = 500 K Find: the mass ow rate m , the wall heat ux qw and the entropy change s2 s1 ; check for satisfaction of the second law. Assume: Calorically perfect ideal gas, R = 0.287 Analysis: Geometry: A r L Now get the mass ux. p1 1 = 1 RT1 , 100 kP a p1 = = , kJ RT1 0.287 kg K (300 K ) = 1.161 So m = 1 u1 A 1 = 1.161 kg m3 10 m s 0.02 m2 = 0.2322 kg . s (3.81) kg m3 (3.78) (3.79) (3.80) = r2 , = A = 2r = 2 A = 2 (0.02 m2 ) = 0.501 m. (3.75) (3.76) (3.77)
kJ kg K ,

cp = 1.0035

kJ kg K

Now get the ow variables at state 2: 2 = p2 100 kP a , ,= kJ RT2 0.287 kg (500 K ) K (3.82) (3.83) (3.84) (3.85)
m s

2 u2 A 2 u2

kg , m3 = 1 u1 A 1 , 1 u1 A 1 1 u1 = = , 2 A2 2 = 0.6969 1.161 =
kg m3

10
kg m3

0.6969

= 16.67

m . s

(3.86)

173

Now consider the energy equation: u d u2 p e+ + dx 2 u2 d h+ dx 2

L

= = = = =

qw L , A qw L , m qw L dx, m 0 qw LL , m qw LL , m
L

(3.87) (3.88) (3.89) (3.90) (3.91) (3.92)

d u2 h+ 2 0 dx u2 h2 + 2 h 1 2 u2 cp (T2 T1 ) + 2 2 Solving for qw , we get qw qw qw qw qw = = m LL cp ( T2 T 1 ) + u2 u2 2 1 2 2 1, 003.5 ,

dx u2 1 2 u2 1 2

(3.93)
m 2 s

0.2322 kg s (100 m) (0.501 m)

16.67 J (500 K 300 K ) + kg K 2 m2 , s2 J , kg

10

m 2 s

, (3.94) (3.95) (3.96) (3.97)

= 0.004635

kg m2 s kg = 0.004635 2 m s W = 930 2 m

J 88.9 kg J 200, 700 88.9 kg 200, 700

The heat ux is positive, which indicates a transfer of thermal energy into the air. Now nd the entropy change. s2 s 1 s2 s 1 s2 s 1 p2 , p1 J 500 K = 1, 003.5 ln kg K 300 K J . = 512.6 0 = 512.6 kg K = cp ln T2 T1 R ln (3.98) 287 J kg K ln 100 kP a 100 kP a , (3.99) (3.100)

Is the second law satised? 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 uid to its nal state of 500 K . It could be greater than 500 K and still satisfy the second law. S2 S 1 2 S 1 S Q12 , T 12 Q , T 12 Q , T qw Atot , T qw LL , T (3.101) (3.102) (3.103) (3.104) (3.105)

m ( s2 s1 ) m ( s2 s1 ) m ( s2 s1 )

174

s2 s 1 512.6 512.6 J kg K J kg K

qw LL , mT 930 s J m2 (100 m) (0.501 m) 0.2322 J . kg K

kg s

(3.106) , (3.107) (3.108)

(500 K )

401.3

3.1.5

Inuence coecients

Now, let us uncouple the steady one-dimensional equations. First let us summarize again, in a slightly dierent manner than before: du u dA d + = , dx dx A dx du dp w L u + = , dx dx A d (qw + w u) L dp c2 = . e dx dx uA p u

(3.109) (3.110) (3.111)

In matrix form this is

u 0 c2

dA d u A dx 0 dx w L u 1 du dx = (qw +A w u)L dp 0 1 e uA | dx

(3.112)

Use Cramers Rule to solve for the derivatives. First calculate the determinant of the coefcient matrix: u [(u)(1) (1)(0)] (0)(1) (c2 )(1) = u2 c2 . Implementing Cramers Rule:
dA L w + u u A dx A (qw +w u)L e uA p |

(3.113)

d = dx du = dx dp = dx

(u2

c2 )

(3.114)

dA L c2 u + u w u A dx A

(u2

c2 )

(qw +w u)L e uA p |

(3.115)

dA L uc2 u c2 w + u2 A dx A

(u2 c2 ) 175

(qw +w u)L e uA p |

(3.116)

Simplify to nd d 1 = dx A 1 du = dx A 1 dp = dx A Note, we have a system of coupled non-linear ordinary dierential equations in standard form for dynamic system analysis: valid for general equations of state singular when velocity sonic u = c
du dx

+ w L + u2 dA dx c2 u dA uw L dx (u2 c2 )

(qw +w u)L e u p |

(3.117)

+ c 2 w L + c2 u2 dA dx (u2 c2 )

(u2 c2 )

(qw +w u)L e p |

(3.118)

(qw +w u)Lu e p |

(3.119)

= f (u)

3.2

Flow with area change

This section will consider ow with area change with an emphasis on isentropic ow. Some problems will involve non-isentropic ow but a detailed discussion of such ows 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, dx d u2 + p dx d u2 p e+ + dx 2 = 0, = 0. (3.120) (3.121) (3.122)

Integrate the energy equation with h = e + p/ to get h+ u2 u2 = ho + o . 2 2 176 (3.123)

If we dene the o condition to be a condition of rest, then uo 0. This is a stagnation condition. So u2 h+ = ho , (3.124) 2 u2 = 0. (3.125) (h ho ) + 2 Since we have a CPIG, cp (T To ) + u2 = 0, 2 u2 T To + = 0, 2cp u2 To + = 0. 1 T 2cp T (3.126) (3.127) (3.128)

Now note that cp = c p so 1 cp cp c v R cp c v = , = cp cp c v cv cv 1 1 (3.129)

T o 1 u2 + = 0, T 2 RT 1 u2 To = 1+ . T 2 RT if p = RT, e = c v T + eo ,

(3.130) (3.131)

Recall the sound speed and Mach number for a CPIG: c2 = RT u 2 M2 , c thus, To 1 2 = 1+ M , T 2 T 1 2 = 1+ M To 2 Now if the ow is isentropic, we have T = To Thus, = o p = po 1 2 1+ M 2 1 2 1+ M 2 177
1 1

(3.132) (3.133)

(3.134)
1

(3.135)

p po

(3.136)

, .

(3.137) (3.138)

For air = 7/5 so T = To = o p = po 1 1 + M2 5 1 1 + M2 5 1 1 + M2 5

1
5 2

, , .

(3.139) (3.140) (3.141)

7 2

Figures 3.2, 3.3 3.4 show the variation of T , and p with M 2 for isentropic ow. Other thermodynamic properties can be determined from these, e.g. the sound speed: c = co
T(K) 300 250 200 150 100 50 0 2 4 6 8 10 M2

RT = RTo

1 2 T = 1+ M To 2

1/2

(3.142)

Calorically Perfect Ideal Gas R = 0.287 kJ/(kg K) = 7/5 Stagnation Temp = 300 K

Figure 3.2: Static temperature versus Mach number squared

P(bar) 1 0.8 0.6 0.4 0.2 4 M2

Calorically Perfect Ideal Gas R = 0.287 kJ/(kg K) = 7/5 Stagnation Pressure = 1 bar

10

Figure 3.3: Static pressure versus Mach number squared

Example 3.10
Airplane problem Given: An airplane is ying 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

178

rho(m3/kg)

1.2 1 0.8 0.6 0.4 0.2 0 2 4 6 8 10 M2

Calorically Perfect Ideal Gas R = 0.287 kJ/(kg K) = 7/5 Stagnation Density = 1.16 kg/m3

Figure 3.4: Static density versus Mach number squared

Assume: Steady isentropic ow of C.P.I.G. Analysis: In the steady wave frame, the ambient conditions are static while the nose conditions are stagnation. M= u u = = c RT 200 m/s
7 5 J 287 kgK 288 K

= 0.588

(3.143)

so To = T 1 o = 1 + M 2 5
5 2

1 1 + M2 5

1 = (288 K ) 1 + 0.5882 5 1 1 + 0.5882 5

5 2

= 307.9 K

(3.144) (3.145) (3.146)

101.3 kP a = kJ 288 K 0.287 kgK

7 2

= 1.45 kg/m3
7 2

1 po = p 1 + M 2 5

1 = (101.3 kP a) 1 + 0.5882 5

= 128 kP a

Note the temperature, pressure, and density all rise in the isentropic process. In this wave frame, the kinetic energy of the ow is being converted isentropically to thermal energy.

3.2.2

Sonic properties

Let * denote a property at the sonic state M 2 1 T = To = o p = po c = co 1 2 1+ 1 2 1 2 1+ 1 2 1 2 1+ 1 2 1 2 1+ 1 2

1

2 , +1 = = 2 +1 2 +1 2 , +1
1 1

(3.147) , , (3.148) (3.149) (3.150) (3.151)

1 1

1/2

u = c =

RT = 179

2 RTo . +1

If the uid is air, we have = 7/5 and T To o p po c co = 0.8333, = 0.6339, = 0.5283, = 0.9123. (3.152) (3.153) (3.154) (3.155)

3.2.3

Eect of area change

To understand the eect of area change, the inuence of the mass equation must be considered. 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, u A udu + dp = 0, d dp = . p Substitute energy, then mass into momentum: p udu + d p udu + du dA u A p 1 1 du + 2 du dA u uA p/ du 1 2 u du p/ 1 2 u u du 1 1 2 u M du M2 1 u du u = 0, = 0, = 0, = = = = = p dA , uA p/ dA , u2 A 1 dA , M2 A dA , A 1 dA . 2 M 1 A (3.159) (3.160) (3.161) (3.162) (3.163) (3.164) (3.165) (3.166) (3.156) (3.157) (3.158)

Figure 3.5 gives show the performance of a uid in a variable area duct. We note there is a singularity when M 2 = 1, if M 2 = 1, we need dA = 0, 180

Consider u > 0 Subsonic Diffuser Subsonic Nozzle

dA >0, M <1 so du < 0, flow slows down dp >0

dA <0, M <1 so du > 0, flow speeds up dp < 0

Supersonic Nozzle

Supersonic Diffuser

dA >0, M >1 so du > 0, flow speeds up dp < 0

dA < 0, M > 1 so du < 0, flow slows down dp >0

Figure 3.5: Behavior of uid in sub- and supersonic nozzles and diusers area minimum necessary to transition from subsonic to supersonic ow, it can be shown an area maximum is not relevant. Consider A at a sonic state. From the mass equation: uA = u A , uA = c A , A 1 RT RT 1 = c = , RT = A u u RT u A = A T 1 o = T M o T To 1 . To T M
1 +1 2 1

(3.167) (3.168) (3.169) (3.170)

Substitute from earlier-developed relations and get A 2 1 2 1 1+ = M A M +1 2 . (3.171)

Figure 3.6 shows the performance of a uid in a variable area duct. Note that
A A

has a minimum value of 1 at M = 1,

A A

For each
A A

> 1, there exist two values of M , and

as M 0 or M . 181

A/A* 6 5 4 3 2 1 0 0.5 1 1.5 2 2.5 3 M

Calorically Perfect Ideal Gas R = 0.287 kJ/(kg K) = 7/5

Figure 3.6: Area versus Mach number for a calorically perfect ideal gas

3.2.4

Choking

Consider mass ow rate variation with pressure dierence. We have then small pressure dierence gives small velocity and small mass ow, as pressure dierence grows, velocity and mass ow rate grow, velocity is limited to sonic at a particular duct location, this provides fundamental restriction on mass ow rate, it can be proven rigorously that sonic condition gives maximum mass ow rate. m max = u A , if ideal gas: = = o = o = o 2 +1 (3.172)
1 1

2 RTo A , +1 2 +1
1/2

(3.173) (3.174) (3.175)

2 +1 2 +1

1 1

RTo A ,

1 +1 2 1

RTo A .

A ow which has a maximum mass ow rate is known as choked ow. Flows will choke at area minima in a duct. Example 3.11
Isentropic area change problem with choking
2

Given: Air with stagnation conditions po = 200 kP a To = 500 K ows through a throat to an exit Mach number of 2.5. The desired mass ow is 3.0 kg/s, Find: a) throat area, b) exit pressure, c) exit temperature, d) exit velocity, and e) exit area.
2

adopted from White, Fluid Mechanics McGraw-Hill: New York, 1986, p. 529, Ex. 9.5

182

Assume: C.P.I.G., isentropic ow, = 7/5 Analysis: o = po 200 kP a = = 1.394 kg/m3 . RTo (0.287 kJ/kg ) (500 K )

(3.176)

Since it necessarily ows through a sonic throat: m max A A = o = o = 1.394

kg m3
2 1 2 +1 m max 1 +1 2 1 2 1 +1

RTo A , , RTo 3 kg/s = 0.008297 m2 .

J kg K

(3.177) (3.178)

+1

(3.179)

(0.5787)

1.4 287

(500 K )

Since we know Me use isentropic relations to nd other exit conditions. pe Te Note e = kg 11.71 kP a pe = 0.1834 3 . = kJ RTe m 0.287 kgK (222.2 K ) (3.182) = po 1 + = To 1 + 1 2 Me 2 1 2 Me 2
1

1 = (200 kP a) 1 + 2.52 5
1

3.5

= 11.71 kP a,

(3.180) (3.181)

1 = (500 K ) 1 + 2.52 5

= 222.2 K.

Now the exit velocity is simply ue = M e c e = M e Now determine the exit area. A = = A 2 Me + 1 1 2 Me 1+ 2 1 1 + 2.52 5
1 +1 2 1

RTe = 2.5 1.4 287

J kg K

(222.2 K ) = 747.0

m . s

(3.183)

,
3

(3.184) (3.185)

0.008297 m2 5 2.5 6

= 0.0219 m2 .

3.3

Normal shock waves

This section will develop relations for normal shock waves in uids 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 183

 one-dimensional ow, steady ow, no area change, viscous eects and wall friction do not have time to inuence ow, and heat conduction and wall heat transfer do not have time to inuence ow. We will consider the problem in the context of the piston problem as sketched in Figure 3.7.

vp= v

D v=0 p1 1 u = u2 p2 2 u = -D p1 1

v = v2 p2 2

Laboratory Frame x* u=v-D x = x* - D t, v=u+D x* = x + D t

Steady Wave Frame x

Figure 3.7: Normal shock sketch The physical problem is as follows: Drive a piston with known velocity vp into a uid at rest (v1 = 0) with known properties, 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: 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 ), .... 184

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, dx d u2 + p = 0, dx d u2 u h + = 0, dx 2 h = h(p, ).

(3.186) (3.187) (3.188) (3.189)

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. 3 4 2 u2 + p2 u2 h2 + 2 2 h2 = 1 D, = 1 D 2 + p1 , D2 = h1 + , 2 = h(p2 , 2 ). (3.190) (3.191) (3.192) (3.193)

2 u2 2

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 dierential equations in the vicinity of a discontinuity. As described by LeVeque, 1992, 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 quantities balanced by uxes of those quantities. If q is a set of conserved variables, and f (q) is the ux of q (e.g. for mass conservation, is a conserved variable and u is the ux), then the primitive form of the conservation law can be written as d dt
x2 x1

q(x, t)dx = f (q(x1 , t)) f (q(x2 , t)).

(3.194)

Here we have considered ow into and out of a one-dimensional box for x [x1 , x2 ]. For our Euler equations we have u q= , 1 2 e + 2u

3 William John Macquorn Rankine, 1820-1872, Scottish engineer and mechanician, pioneer of thermodynamics and steam engine theory, taught at University of Glasgow, studied fatigue in railway engine axles. 4 Pierre Henri Hugoniot, 1851-1887, French engineer.

u u2 + p f (q) = u2 + u e + 1 2

(3.195)

185

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 d dt
x1 +Dt x1

q(x, t)dx +

d dt

x2 x1 +Dt+

q(x, t)dx = f (q(x1 , t)) f (q(x2 , t)).

(3.196)

Here x1 + Dt lies just before the discontinuity and x1 + Dt+ lies just past the discontinuity. Using Leibnitzs rule, we get q(x1 + Dt , t)D + 0 +
x1 +Dt x1 x2 q q dx + 0 q(x1 + Dt+ , t)D + dx t x1 +Dt+ t = f (q(x1 , t)) f (q(x2 , t)).

(3.197)

Now if we assume that on either side of the discontinuity the volume of integration is suciently 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 )), D (q(x1 ) q(x2 )) = f (q(x1 )) f (q(x2 )). (3.198) (3.199)

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 )), f (q(x1 )) = f (q(x2 )). (3.200) (3.201)

That is the uxes 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 = 0, which is the well-known D= f (q(x2 )) f (q(x1 )) . q(x2 ) q(x1 ) (3.202)

3.3.2

Rayleigh line
p 2 = p 1 + 1 D 2 2 u2 2, 2 2 2 2 u D 2 2. p2 = p 1 + 1 1 2 (3.203) (3.204)
5

If we operate on the momentum equation as follows

2 2 2 Since mass gives us 2 2 u2 = 1 D we get an equation for the Rayleigh Line, space: 2 p2 = p 1 + 2 1D

a line in (p, 1 ) (3.205)

1 1 . 1 2

Note that the

John William Strutt (Lord Rayleigh), 1842-1919, aristocratic-born English mathematician and physicist, studied at Cambridge, inuenced 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.
5

186

 Rayleigh line passes through ambient state, Rayleigh line has negative slope, magnitude of the slope is proportional to square of the wave speed, and the Rayleigh line is independent of state and energy equations.

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: h2 + h2 +

1 2

D2 u2 2 = h1 + , 2 2 2 D2 1 D = h1 + , 2 2
2

(3.206) (3.207) (3.208) (3.209) (3.210)

2 D 2 2 1 2 h2 h 1 + 2 2 2 2 D (1 2 ) (1 + 2 ) h2 h 1 + 2 2 2

D 2 1 h2 h 1 + 2 2

1 = 0, = 0, = 0.

Now use the Rayleigh line to eliminate D 2 : D D

2

1 = (p2 p1 ) 2 1 1 = (p2 p1 ) 2 1 1 = (p2 p1 ) 2 1

1 1 1 2

(3.211) (3.212) (3.213)

D2

2 1 , 1 2 1 2 . 2 1

So the energy equation becomes h2 h 1 + 1 1 (p2 p1 ) 2 2 1 (1 2 ) (1 + 2 ) 1 2 2 1 2 2 1 1 + 2 1 h2 h1 (p2 p1 ) 2 1 2 1 1 1 h2 h1 (p2 p1 ) + 2 2 1 = 0, = 0, = 0. (3.214) (3.215) (3.216)

Regrouping to see what induces enthalpy changes, we get h2 h1 = (p2 p1 ) 187 1 2 1 1 + . 2 1 (3.217)

This equation is the Hugoniot equation. It holds that enthalpy change equals pressure dierence times mean volume, is independent of wave speed D and velocity u2 , and is independent of the equation of state.

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 to get a second relation between p 2 and 2 , use the Rayleigh line 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 nd 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 = v p .

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 , p = RT. so po p + ho , R Ro cp p p o h = + ho , R o cp p po h = + ho , cp c v o p po h = + ho . 1 o h = cp 188 (3.220) (3.221) (3.222) (3.223) (3.218) (3.219)

Evaluate at states 1 and 2 and substitute into Hugoniot: p1 po + ho 1 o 1 1 1 = (p2 p1 ) + , 2 2 1 p2 p1 1 1 1 + (p2 p1 ) = 0, 1 2 1 2 2 1 1 1 1 1 1 1 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 p1 = 0, p2 1 2 1 1 1 2 + ho p2 =

1 ) space a hyperbola in (p, +1 1 p1 +1 1 1 1 1 2

p2 po 2 o

p2

1 2 1 1

1 2

1 2

1 1 +1 1

causes p2 , note = = 1.4, 2 6 for innite pressure

as

The Rayleigh line and Hugoniot curves are sketched in Figure 3.8.
P

1 , p2 p1 , note negative pressure, not physical here +1

shocked state

P2

Excluded Zone D (1/ < 1/ minimum)

Excluded Zone A (since slope of Rayleigh line < 0)

P1

Excluded Zone A

initial stae

Hugoniot Curve (from energy equation) independent of D^2

Excluded Zone B (2nd law violated)

( -1) - P1 ------( +1)

1 / 2

1 / 1

1/

Excluded Zone C (negative pressure)

( -1) 1 high pressure ------- --asymptotic limit ( + 1) 1

Rayleigh Line (from momentum and mass equations) |Slope| proportional to D^2

Figure 3.8: Rayleigh line and Hugoniot curve Note: 189

 intersections of the two curves are solutions to the equations, the ambient state 1 is one solution, the other solution 2 is known as the shock solution, shock solution has higher pressure and higher density, higher wave speed implies higher pressure and higher density, a minimum wavespeed exists, occurs when Rayleigh line tangent to Hugoniot, occurs for very small pressure changes, corresponds to a sonic wave speed, disturbances are acoustic. if pressure increases, can be shown entropy increases, if pressure decreases (wave speed less than sonic), entropy decreases; this is nonphysical. Substitute Rayleigh line into Hugoniot to get single equation for 2 p1 +
2 2 1D

1 1 1 2

+1 1 p1 +1 1

1 1 1 2

1 2 1 1

(3.224)

This equation is quadratic in two solutions, one ambient 12 =

1 2 1 1

and factorizable. Use computer algebra to solve and get and one shocked solution: (3.225)

1 2 1 1 p1 1+ . = 2 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
p1 , 2 1 , acoustic limit: D 2 1 +1 , 1

190

rho2 (kg/m^3) 7 6 5 4 3 2 1 500

Strong Shock Limit Calorically perfect ideal air = 7/5 R = 287 kJ/(kg K)

Exact Solution

D (m/s) 1000 1500 2000 2500 3000

D = Dmin = c1

Figure 3.9: Shock density vs. 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 uid velocity in the shocked wave frame: p2 = u2 1 2 1 D 2 p1 , +1 +1 1 p1 2 = D 1+ . 2 +1 ( 1) D 1 (3.226) (3.227)

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.
P2 (Pa) 8. 10 6. 10 4. 10 2. 10 6 6 6 6

Ambient = 100,000 Pa

Exact Solution and Strong Shock Limit

Calorically perfect ideal air = 7/5 R = 287 kJ/(kg K)

500

D (m/s) 1000 1500 2000 2500 3000

D = Dmin = c1

Figure 3.10: Shock pressure vs. shock wave speed for calorically perfect ideal air The shocked wave frame uid particle velocity u2 is plotted against wave speed D for CPIG air in Figure 3.11.
2 The shocked wave frame uid particle velocity M2 = D for CPIG air in Figure 3.12. 2 u 2 2 p2

is plotted against wave speed

191

u2 (m/s) 500 -100 -200 -300 -400 -500 1000 1500 2000 2500 D (m/s) 3000

Strong Shock Limit Exact Solution u1 = - co D = Dmin = c1

Calorically perfect ideal air = 7/5 R = 287 kJ/(kg K)

Figure 3.11: Shock wave frame uid particle velocity vs. shock wave speed for calorically perfect ideal air
M2^2 1 0.8

Strong Shock Limit

0.6 0.4 0.2 0 500

Exact Solution

Calorically perfect ideal air = 7/5 R = 287 kJ/(kg K)

1000

1500

2000

2500

D (m/s) 3000

D = Dmin = c1 M2^2 = 1

Figure 3.12: Mach number squared of shocked uid particle vs. shock wave speed for calorically perfect ideal air Note in the steady frame that The Mach number of the undisturbed ow is (and must be) > 1: supersonic, The Mach number of the shocked ow is (and must be) < 1: subsonic. Transform back to the laboratory frame u = v D : v2 D = D v2 p1 1 2 1+ , 2 +1 ( 1) D 1 2 p1 1 1+ . = DD 2 +1 ( 1) D 1 (3.228) (3.229)

Manipulate the above equation and solve the resulting quadratic equation for D and get 192

D=

+1 v2 4

p1 2 +1 + v2 1 4

(3.230)

Now if v2 > 0, we expect D > 0 so take positive root, also set the velocity equal to the piston velocity v2 = vp . D= Note: acoustic limit: as vp 0, D c1 ; the shock speed approaches the sound speed, and strong shock limit: as vp , D
+1 vp . 2

+1 vp + 4

p1 +1 2 + vp 1 4

(3.231)

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.
D (m/s) 1200 1000 800

Exact Solution

Calorically perfect ideal air = 7/5 R = 287 kJ/(kg K) Strong Shock Limit

Acoustic Limit, D -> c1

600 400 200 200 400 600 800

vp (m/s) 1000

Figure 3.13: Shock speed vs. piston velocity for calorically perfect ideal air If we dene the Mach number of the shock as Ms we get Ms = + 1 vp + 4 RT1 1+
2 vp +1 RT1 4 2

D , c1

(3.232)

(3.233)

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. 193

Ms 3.5 3 2.5 2 1.5 1 0.5 200

Exact Solution Strong Shock Limit

Calorically perfect ideal air = 7/5 R = 287 kJ/(kg K)

Acoustic Limit, Ms -> 1

400

600

800

vp (m/s) 1000

Figure 3.14: Shock Mach number vs. 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 + , u2 = u1 + u1 , p2 = p1 + p. Substituting into the normal shock equations, we get (1 + ) (u1 + u) = 1 u1 , (1 + ) (u1 + u)2 + (p1 + p) = 1 u1 2 + p1 , p1 + p 1 p1 1 2 + (u1 + u)2 = + u1 . 1 1 + 2 1 1 2 Expanding, we get 1 u1 + u1 () + 1 (u) + () (u) = 1 u1 1 u1 2 + 21 u1 (u) + u1 2 () + 1 (u)2 + 2u1 (u) () + () (u)2 + (p1 + p) = 1 u1 2 + p1 1 1 p1 p1 1 + P 2 + ... + u1 2 + 2u1 (u) + (u)2 1 1 1 2 p1 1 2 + u1 = 1 1 2

(3.234) (3.235) (3.236)

(3.237) (3.238) (3.239)

Subtracting the base state and eliminating products of small quantities yields

u1 () + 1 (u) = 0, 21 u1 (u) + u1 2 () + p = 0, 1 p1 p 2 + u1 (u) = 0. 1 1 1 194

(3.240) (3.241) (3.242)

In matrix form this is

u1 u1 2

p1 2 1

1 21 u1 u1

0 1
1 1 1

0 u = 0 . p 0

(3.243)

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: u1 u1 2 2 u1 u 1 1 + 1 1 1 1 2 u1 1 u1 2 + (2 ( 1)) 1 1 p1 2 1 p1 1 p1 u1 2 ( + 1) u1 2 + 1 = 0, = 0, = 0, p1 = c2 1. 1 (3.244) (3.245) (3.246) (3.247)

u1 2 = So the velocity is necessarily sonic for a small disturbance. Take u to be known and solve a resulting 2 2 system:
p1 1 2 1

u1

0
1 1 1

1 u . u1 u

(3.248)

Solving yields = u 1 u = 1 p1 c1 1 p1 u = 1 c1 u. 1 (3.249) (3.250)

p = 1

3.4

Flow with area change and normal shocks

This section will consider ow from a reservoir with the uid 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 , nd the pressure, temperature, density at all points in the duct and the mass ow rate.

3.4.1

Converging nozzle

A converging nozzle operating at several dierent values of pb is sketched in Figure 3.15. The ow through the duct can be solved using the following procedure check if pb p , 195

Pb Po Pe

. . m/mmax e 1

c b a

0 P(x)/Po 1 P*/Po

P*/Po

Pb/Po

a--subsonic exit b--subsonic exit c--sonic exit d--choked, external expansion e--choked, external expansion x

xe

Figure 3.15: Converging nozzle sketch if so, set pe = pb , determine Me from isentropic ow relations, determine A from
A A

relation,
A A

at any point in the ow where A is known, compute nd local M Note:

and then invert

A A

relation to

These ows are subsonic throughout and correspond to points a and b in Figure 3.15. If pb = p then the ow is sonic at the exit and just choked. This corresponds to point c in Figure 3.15. If pb < p , then the ow 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 dierent values of pb is sketched in Figure 3.16. The ow through the duct can be solved using the a very similar following procedure set At = A , with this assumption, calculate
Ae , A

196

Pb

Po

Pt

Pe possible normal shock

P(x)/Po 1 P*/Po
a--subsonic exit b--subsonic exit c--subsonic design d--shock in duct

Sonic Throat xt . . m/mmax hg f e 1 xe

e-shock at end of duct f--external compression g--supersonic design

h--external expansion

c b a

P*/Po

1 Pb/Po

Figure 3.16: Converging-diverging nozzle sketch 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 ow 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 , the duct ow is shockless, and there may be compression outside the duct. 197 If pb pesh , there is a normal shock inside the duct,

 if pesup = pb the ow is at supersonic design conditions and the ow is shockless, if pb < pesup , the ow 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, nonheat conducting uid. This analysis is a good deal more rigorous than much of traditional one-dimensional gas dynamics, and draws upon some of the more dicult mathematical methods we will encounter. In assuming no diusive transport, we have eliminated all mechanisms for entropy generation; consequently, we will be able to model the process as isentropic. We note that even without diusion, 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 dierential 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.

3.5.1

Inviscid one-dimensional equations

u +u + t x x u p u + u + t x x s s +u t x p

The equations to be considered are shown here in non-conservative form = 0, = 0, = 0, = p(, s). (3.251) (3.252) (3.253) (3.254)

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 dp = p x =
t

p p d + ds, s s p
s

so,

(3.255) (3.256)

p + x t s

s . x t

Now, let us dene thermodynamic properties c2 and as follows c2 p , s 198 p . s (3.257)

We will see that will be unimportant, and will be able to ascribe to c2 the physical signicance 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 denitions give us p s = c2 + . x x x (3.258)

Substituting into our governing equations, we see that pressure can be eliminated to give three equations in three unknowns: u +u + = 0, t x x u s u + u + c2 + = 0, t x x x s s +u = 0. t x Now if
s t s + u x = 0, we can say that if s = s(x, t),

(3.259) (3.260) (3.261)

s s dt + dx, t x ds s dx s = + . dt t dt x ds = Thus on curves where (3.261)

dx dt

(3.262) (3.263)

= u, we have from substituting Eq. (3.263) into the energy equation

ds = 0. (3.264) dt Thus we have converted the partial dierential equation into an ordinary dierential equation. This can be integrated to give us s = C, on a particle pathline,
dx dt

= u.

(3.265)

This scenario is sketched on the so-called x t diagram of Figure 3.17. 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 u t 2 u + 0 0 c t s 0 0 1 0 t

0 0 x u u x = 0 s 0 u 0 x

(3.266)

These equations are of the form Aij uj uj + Bij = Ci . t x 199 (3.267)

t pathlines

s = s0

s = s1

s = s2 s = s3 s = s4 x

Figure 3.17: x t diagram showing maintenance of entropy s along particle pathlines for isentropic ow.

dx dt

=u

As described by Whitham, 6 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 :
i Aij

uj uj + i Bij = i Ci . t x
i

(3.268)

Now, this method will work if we can choose + u x . Let us take form similar to t
i Aij

to render the above product to be of the

uj uj + i Bij t x

= mj

uj uj , + t x dx duj on = . = mj dt dt

(3.269) (3.270)

So comparing terms, we see that

i Aij

i Aij so, we get by eliminating mj that

= mj , = mj ,

i Bij

= mj ,

(3.271) (3.272)

(Aij Bij ) = 0.

(3.273)

6 Gerald B. Whitham, Ph.D. 1953, University of Manchester, applied mathematician and developer of theory for non-linear wave propagation, Powell Professor of Applied Mathematics at California Institute of Technology.

200

This is a left eigenvalue problem. We set the determinant of Aij Bij to zero for a non-trivial solution and nd u 0 2 (3.274) c ( u) = 0. 0 0 u Evaluating, we get ( u) ( u)2 + ( u)(c2 ) = 0, ( u) ( u)2 c2 = u c.

(3.275) (3.276)

= 0.

Solving we get = u, Now the left eigenvectors (

i

(3.277)

give us the actual equations. First for = u, we get (3.278)

uu 0 2 (u u) = ( 0 0 0 ) , 3 ) c 0 0 uu 0 0 ( 1 2 3 ) c2 0 = ( 0 0 0 ) . 0 0 0
2

(3.279)
3.

Two of the equations require that 1 = 0 and select a normalized solution so that
i

= 0. There is no restriction on

We will (3.280)

= (0, 0, 1).

Thus

uj i Aij t

+ i Bij

uj x

= i Ci gives

0 0 x u u x = ( 0 0 1 ) 0 , s 0 u 0 x

1 0 0 u t u 2 ( 0 0 1 ) 0 0 t + ( 0 0 1 ) c s 0 0 1 0 t

(0 0 1)

t u t s t

+ (0

0 u)

x u x s x

= 0, (3.281)

s s +u = 0. t x

s s dt + x dx, and ds = s + dx . Now if we So as before with s = s(x, t), we have ds = s t dt t dt x dx dx require dt to be a particle pathline, dt = u, then our energy equation gives us

ds = 0, dt

on

dx = u. dt

(3.282)

The special case in which the pathlines are straight in x t space, corresponding to a uniform velocity eld of u(x, t) = uo , is sketched in the x t diagram of Figure 3.18. 201

t pathlines

1 uo s = s3 s = s4 x

s = s0

s = s1

s = s2

Figure 3.18: x t diagram showing maintenance of entropy s along particle pathlines for isentropic ow. Now let us look at the remaining eigenvalues, = u c. (
1 2

dx dt

= uo

As one of the components of the left eigenvector should be arbitrary, we will take arrive at the following equations then c c2 c Thus (1
uj i Aij t 2 2

ucu 0 2 c (u c u) = (0 0 0), 3 ) 0 0 ucu c 0 ( 1 2 3 ) c2 c = ( 0 0 0 ) . 0 0 c

(3.283)

(3.284) = 1; we

= 0, = = 0, = = 0, =

1 = , c 1 = , c = 2. c

(3.285) (3.286) (3.287)

+ i Bij

uj x

= i Ci gives

c2

1 c

c2

1 0 0 t u 1 ) 0 0 t + ( 1 c s 0 0 1 t

( 1 c

c2

t u t s t

u 2 )c 0

0 x u 1 u x = ( 1 c s 0 u x

c2

+ (u c

u c 202

+ c

u c2

x u x s x

) 0,

0 0

= 0,

c2

u u u s u s + (u c) + 1 + 2 + = 0, t x c t c x c t c2 c x u s u s + (u c) + (u c) + (u c) + 2 = 0, t x c t x c t x u s u s + (u c) + (u c) + (u c) c + = 0. t x t x t x

(3.288) (3.289) (3.290)

= u c, we get a transformation of the partial dierential equations Now on lines where dx dt to ordinary dierential equations: c2 d du ds c + = 0, dt dt dt on dx = u c. dt (3.291)

A sketch of the characteristics, the lines on which the dierential equations are obtained, are sketched in the x t diagram of Figure 3.19.
t
acoustic characteristic dx __ = u- c dt pathline characteristic dx __ = u dt

dx __ = u+ c dt acoustic characteristic

Figure 3.19: x t diagram showing characteristics for pathlines dx = u c. dt

dx dt

= u and acoustic waves

3.5.2

Homeoentropic ow of an ideal gas

The equations developed so far are valid for a general equation of state. Here let us now consider the ow of a calorically perfect ideal gas, so p = RT and e = cv T + e . Further let us take the ow 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 eld is a constant. Consequently, we have the standard relations for a 203

calorically perfect ideal gas: p c2 = , p = A, (3.292) (3.293)

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 c2 Rearranging, we get du d c = , dt dt
p = A 1 , and c = Now c2 =

du d c = 0, dt dt

on

dx = u c. dt

(3.294)

on
1 2

dx = u c. dt

(3.295)

, so (3.296)

1 1 du d 2 d = A 2 1 = A 2 . dt dt 1 dt

Regrouping, we nd d u dt
1 2 2 1 2 d u c dt 1

= 0, = 0.

(3.297) (3.298)

Following notation used by Courant 7 and Friedrichs, 8 we then integrate each of these equations, which are homogeneous, along characteristics to obtain algebraic relations dx 2 c = 2r, on = u + c, C + characteristic, 1 dt 2 dx u c = 2s, on = u c, C characteristic, 1 dt u+ (3.299) (3.300) (3.301)

A sketch of the characteristics is given in the x t diagram of Figure 3.20. Now r and
Richard Courant, 1888-1972, Prussian-born German mathematician, received Ph.D. under David Hilbert at G ottingen, 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 G ottingen by the Nazis in 1933, ed Germany, and founded the Courant Institute of Mathematical Sciences at New York University, author of classic mathematical text on supersonic uid mechanics. 8 Kurt Otto Friedrichs, 1901-1982, German-born mathematician who emigrated to the United States in 1937, student of Richard Courants at G ottingen, taught at Aachen, Braunschweig, and New York University, worked on partial dierential equations of mathematical physics and uid mechanics.
7

204

t
dx __ = u+ c on C+ dt + C C + C + C C

arbitrary region of interest

dx __ = u - c on C dt

+ C

Figure 3.20: x t diagram showing C + and C characteristics

dx dt

= u c.

s can take on dierent values, depending on which characteristic we are on. On a given characteristic, they remain constant. Let us dene additional parameters and to identify which characteristic we are on. So we have dx 2 c = 2r ( ), on = u + c, C + characteristic, 1 dt 2 dx u c = 2s(), on = u c, C characteristic, 1 dt u+ These quantities are known as Riemann invariants.
9

(3.302) (3.303) (3.304)

3.5.3

Simple waves

Simple waves are dened 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 , (3.305) 1 is actually invariant over all of x t space. Now the other Riemann invariant, u+ 2 c = 2r ( ), 1

(3.306)

9 Georg Friedrich Bernhard Riemann, 1826-1866, German mathematician and geometer whose work in non-Euclidian geometry was critical to Einsteins theory of general relativity, produced the rst major study of shock waves.

205

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.
t

arbitrary region of interest + C + C + C

+ C

Figure 3.21: x t diagram showing C + for a simple wave. 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 c = 2so is valid everywhere. Now For this conguration, the Riemann invariant u 1 when t = 0, we have u = 0, c = co , so u 2 2 c= co . 1 1 (3.307)

+ at t = t . At this time the piston moves with Consider now a special characteristic C velocity u p , and the uid velocity at the piston face is ) = u uf ace (t p . ) from Eq. (3.307): We get cf ace (t uf ace u p 2 2 cf ace = co , 1 1 (3.309) (3.308)

) = co + 1 u cf ace (t = t p . 2 Also from Eq. (3.307), we have c = co + 1 u, 2

(3.310)

(3.311)

206

t + C u p (t) + C:
2 c = 2 r ( ) u + ______ 3 +1

+ C:

2 c = 2 r ( ) u + ______ 2 +1

+ C :

2 c = 2 r ( ) 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. which is valid everywhere. + , we have Now on C 2 c 1 2 1 u+ co + u 1 2 2 co 2u + 1 u u+ + , we have So on C c = co + + , we get So for C dx 1 +1 =u+c=u p + co + u p = u p + c o . dt 2 2 (3.317) = 2 cf ace , 1 t=t 1 2 co + = u p + u p , 1 2 2 = 2 up + co , 1 +. = u p on C uf ace + 1 u p . 2 (3.312) (3.313) (3.314) (3.315)

(3.316)

for a particular characteristic, this slope is a constant, as was earlier suggested. Now for prescribed motion, u p 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, 207

u p (t)

+ C

t up u face( t ) = up c
face

( t ) = co +

-1 up 2

+ for our rarefaction problem. Figure 3.23: x t diagram showing C + , along with a few other is sketched in the x t diagram of Figure 3.23. We can use our C Riemann invariant along with isentropic relations to obtain other ow variables. From Eq. (3.307), we get c 1 u =1+ . (3.318) co 2 co Since the ow is homeoentropic, we have p = po = o
c co

1 2

and
2 1

p po

, o

so (3.319) (3.320)

1 u 1+ 2 co 1 u 1+ 2 co

, .

2 1

3.5.4

Prandtl-Meyer rarefaction

If the piston is suddenly accelerated to a constant velocity, then a family of characteristics clusters at the origin on the x t diagram and fan out in a centered rarefaction fan also called a Prandtl-Meyer 10 This can also be studied using the similarity transformation = x/t which reduces the partial dierential equations to ordinary dierential equations. Relevant sketches comparing Prandtl-Meyer fans to non-centered rarefactions are shown in in the x t diagram of Figure 3.24. Example 3.12
Analyze a centered Prandtl-Meyer fan 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 . Theodor Meyer, graduate student of Prandtls at G ottingen who in his 1908 dissertation gave the rst practical development of both two-dimensional expansion waves and oblique shock waves rarefaction.
10

208

up

Prandtl-Meyer Rarefaction t + + + + C C C C + C + C

t u p (t) + C + C + C + C + C

particle path

particle path x

u 0

up p

Figure 3.24: x t diagram centered and uncentered rarefactions, along with pressure and velocity proles for Prandtl-Meyer fans.
The ideal gas law gives o = co =
po RTo

100 kP a (287 J/kg/K )(300 K )

= 1.16 kg/m3 . Now (3.321)

RTo =

7 (287 J/kg/K )(300 K ) = 347 m/s. 5

+ On the nal characteristic of the fan, Cf : u = up = 100 m/s. So

c = co + Now the nal pressure is pf = po

1 7/5 1 up = 347 m/s + (100 m/s) = 327 m/s. 2 2

2 1 2(7/5) 7/51

(3.322)

1 uf 1+ 2 co
1

7/5 1 (100 m/s) 1+ 2 347 m/s

= 0.660

(3.323)

Hence pf = 66.0 kP a. Since the ow is homeoentropic, we get f = o And the nal temperature is Tf = pf 66.0 103 P a = = 266.5 K. f R (0.863 kg/m3 )(287 J/kg/K ) u , co u . co (3.325) Pf po = (1.16 kg/m3 )(.660)5/7 = 0.863 kg/m3 . (3.324)

Recall from linear theory that o exact pexact Texact p o co u, T ( 1)To (3.326)

We compare the results of this problem with the estimates of linear acoustic theory and see = 34 kP a, = 33.5 K, = 0.298 kg/m3 , plinear = 40.3 kP a, linear = 0.334 kg/m3 , (3.327) (3.328) (3.329)

Tlinear = 34.6 K.

209

3.5.5

Simple compression
t + C + C u p (t) + C + C + C + C + C ambient region x shock formation

We sketch a simple compression in the x t diagram of Figure 3.25.

u p (t)

x=0

Figure 3.25: x t diagram for simple compression.

3.5.6

Two interacting expansions

We sketch two interacting expansion waves in the x t diagram of Figure 3.26.

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.

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 dierential equations (PDEs) in the form of ordinary dierential equations (ODEs), and it is often said that the 210

uniform flow u p (t) uniform flow uniform flow

u p (t)

fluid at rest

x=L u p (t)

x=0

x=L

Figure 3.26: x t diagram for two interacting expansion waves. 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. (3.330) t x Now consider a general transformation (x, t) (r, s). Applying the chain rule, we get u u r u s = + , t r t s t u u r u s = + . x r x s x In transformed space, the inviscid Burgers equation becomes u r u s u r u s + +u + r t s t r x s x Now we also have dx = x x dr + ds, r s t t dt = dr + ds. r s 211 (3.334) (3.335) (3.331) (3.332)

= 0.

(3.333)

t rest

u p (t)

uniform flow

flow at rest x=L x

u p (t)

x=0

x=L

Figure 3.27: x t diagram for expansion wall interaction. With the Jacobian we invert to nd dr = x 1 t dx dt , J s s 1 t x ds = dx + dt . J r r (3.337) (3.338)
11

J=

x t x t , r s s r

(3.336)

So it is easy to see that we get the following for the partial derivatives 1 x r = , t J s s 1 x = , t J r r 1 t = , x J s s 1 t = , x J r (3.339) (3.340) (3.341) Substituting into the inviscid Burgers equation, we get u t u t 1 u x u x + +u J r s s r r s s r
11

= 0,

(3.342)

Carl Gustav Jacob Jacobi, 1804-1851, Prussian born, prolic German mathematician. The Jacobian determinant was extensively studied by Jacobi, but rst identied by Cauchy.

212

u t u t u x u x + +u u = 0, r s s r r s s r

(3.343) (3.344)

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 transformation that t x =u , t(x, s) = s. (3.345) s s The rst 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 = u. The second is a convenience, and more standard statement that on characteristics, dx dt we actually need not be as restrictive. With this specication, our inviscid Burgers equation becomes u r x s
t = u s =u

u x u t u t +u u = 0, s r r s s r
=1 =0

(3.346)

u u x u + +u = 0, r s r r u x = 0. s r

(3.347) (3.348)

Now, let us require that

x r

= 0; hence in this special transformed space, we have that u = 0. s (3.349)

This has solution u = f (r ), where f (r ) is an arbitrary function. We now substitute this into x t = f (r ) , s s which can be integrated to get x = f (r )t + g (r ). (3.352) 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. (3.353) In summary we can write a solution parametrically in terms of our transformed space as u(r, s) = f (r ), x(r, s) = f (r )s + r, t(r, s) = s. 213 (3.354) (3.355) (3.356)
x s t = u s to get

(3.350)

(3.351)

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 nonunique values of u! This multivaluedness actually indicates that a shock has formed, and correct insertion of a shock will eliminate the diculty. Example 3.13

u If we have the inviscid Burgers equation u t + u x with u(x, 0) = sin(x), nd 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), x(r, s) t(r, s) = s sin(r) + r, = s.

(3.357) (3.358) (3.359)

We can use this solution to form parametric plots and eectively 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 attening 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 complicated. For isentropic Euler equations, Courant and Friedrichs give some special solutions for rarefactions.

214

t compression contact discontinuity rarefaction shock rest rest

uniform rest

uniform

rest x=L x

p t=0

p t>0

t>0

t>0

Figure 3.28: x t and p, , T vs. x behavior for a shock tube. 215

u 1 0.8 0.6 0.4 0.2 0.2 0.4 0.6 0.8 1 x 1.2

t=0 t=1 t=2

Figure 3.29: u(x) for t = 0, 1, 2 for inviscid Burgers equation problem.

216

Chapter 4 Potential ow
see Panton, Chapter 18 see Yih, Chapter 4 This chapter will consider potential ow. A good deal of highly developed and beautiful mathematical theory was generated for potential ows in the nineteenth century. Additionally, these solutions can be applied in highly disparate elds, as the equations governing potential ow of a uid are identical in form to those governing some forms of energy and mass diusion, as well as electromagnetics. 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 eects. 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 ow with his viscous boundary layer theory. He showed that indeed potential ow solutions could be of great value away from no-slip walls, and provided a recipe to x the solutions in the neighborhood of the wall. In so doing, he opened up a new eld of applied mathematics known as matched asymptotic analysis. So why study potential ows? The following arguments oer some justication. portions of real ow elds 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 ow solutions can give great insight into uid behavior and aid in the honing of a more precise intuition, fundamental solutions are useful as test cases for verication of numerical methods, and there is pedantic and historical value in knowing potential ow. 217

4.1

Stream functions and velocity potentials

We rst 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 ow be two-dimensional. Recall if a ow velocity is conned to the x y plane, then the vorticity vector is conned to the z direction and takes the form 0 0 = (4.1) v u x y Now if the ow is two-dimensional and irrotational, we have v u = 0. x y

(4.2)

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 denition, uid ows from regions of low velocity potential to regions of and v = . We see by substitution into the equation high velocity potential. Thus u = x y for vorticity, that this is true identically: v u = x y x y y x = 0. (4.4)

Now for two-dimensional incompressible ows, we have u v + = 0. x y Substituting for u and v in favor of , we get x + x y = 0, y 2 = 0. (4.6) (4.7) (4.5)

Now if the ow is incompressible, we can also dene the stream function as follows: u= , y v= . x (4.8)

Direct substitution into the mass conservation equation shows that this yields an identity: u v + = x y x y 218 + y x = 0. (4.9)

Now, in an equation which will be critically important soon, we can set our denitions of u and v in terms of and equal to each other, as they must be: x u y v , y u = . x v = (4.10)

(4.11)

Now if we dierentiate the rst equation with respect to y , and the second with respect to x we see 2 2 = , yx y 2 2 2 = 2, xy x now subtract the second from the rst to get 2 2 0 = + 2, y 2 x 2 = 0. (4.12) (4.13) (4.14) (4.15) (4.16)

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 dx + dy = 0, x y d = udx + vdy = 0, u dy = dx =C1 v d = Now for we similarly get dx + dy = 0, x y d = vdx + udy = 0, dy v = dx =C2 u d = We note two features
dy dx =C1

(4.17) (4.18) (4.19)

(4.20) (4.21) (4.22)

1
dy dx =C 2

and

; hence, lines of constant are orthogonal to lines of constant ,

on = C2 , we see that

dx u

dy ; v

hence, lines of = C2 must be streamlines. 219

As an aside, we note that the denition of the stream function u = , v = , can be y x rewritten as dx dy = , = . (4.23) y dt x dt This is a common form from classical dynamics in which we can interpret as the Hamiltonian 1 of the system. We shall not pursue this path, but note that a signicant literature exists for Hamiltonian systems. Now the study of and is essentially kinematics. The only incursion of dynamics is that we must have irrotational ow. Recalling the Helmholtz equation we realize that we can only have potential ow when the vorticity generating mechanisms (three-dimensional eects, non-conservative body forces, baroclinic eects, and viscous eects) are suppressed. In that case, the dynamics, that is the driving force for the uid motion, can be understood in the context of the unsteady Bernoulli equation p 1 + ()T + = f (t). t 2 (4.24)

Note that we do not have to require steady ow to have a potential ow eld. Now solutions to the two key equations of potential ow 2 = 0, 2 = 0, are most eciently 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 briey introduce relevant elements of complex variable theory. Recall that the imaginary number i is dened such that i2 = 1, i = 1. (4.25)

4.2.1

Eulers formula
2

We can get a very useful formula Eulers formula, by considering the following Taylor expansions of common functions about t = 0: 1 2 1 3 1 4 1 5 t + t + t + t ..., 2! 3! 4! 5! 1 2 1 3 1 4 1 5 sin t = 0 + t + 0 t t + 0 t + t . . . , 2! 3! 4! 5! 1 3 1 4 1 1 2 cos t = 1 + 0t t + 0 t + t + 0 t5 . . . , 2! 3! 4! 5! et = 1 + t +
1

(4.26) (4.27) (4.28) (4.29)

William Rowan Hamilton, 1805-1865, Anglo-Irish mathematician. 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 nite dierences, invented integration by parts, name ascribed to Taylor series of which variants were earlier discovered by Gregory, Newton, Leibniz, Johann Bernoulli, and de Moivre.
2

220

With these expansions now consider the following combinations: (cos t + i sin t) t= and et |t=i : cos + i sin = 1 + i ei 1 2 1 1 1 i 3 + 4 + i 5 + . . . , 2! 3! 4! 5! 1 1 1 1 = 1 + i + (i )2 + (i )3 + (i )4 + (i )5 + . . . , 2! 3! 4! 5! 1 3 1 4 1 5 1 2 = 1 + i i + + i + . . . 2! 3! 4! 5! ei = cos + i sin . (4.30) (4.31) (4.32)

As the two series are identical, we have Eulers formula (4.33)

4.2.2

Polar and Cartesian representations

z = x + iy, (4.34)

Now if we take x and y to be real numbers and dene the complex number z to be 2 x + y 2 to obtain x2 + y 2 x y . + i x2 + y 2 x2 + y 2 (4.35)

we can multiply and divide by z=

Noting the similarities between this and the transformation between Cartesian and polar coordinates suggests we adopt x y r = x2 + y 2 , cos = 2 , sin = 2 . (4.36) 2 x +y x + y2 Thus we have z = r (cos + i sin ) , z = rei . (4.37) (4.38)

The polar and Cartesian representation of a complex number z is shown in Figure 4.1. Now we can dene the complex conjugate z as z = x iy, z = (4.39) (4.40) (4.41) (4.42) (4.43) x y i , x2 + y 2 x2 + y 2 z = r (cos i sin ) , z = r (cos( ) + i sin( )) , z = rei . x2 + y 2 Note now that zz = (x + iy )(x iy ) = x2 + y 2 = |z |2 , = rei rei = r 2 = |z |2 . 221 (4.44) (4.45)

iy

y
x
2

r=

+y

Figure 4.1: Polar and Cartesian representation of a complex number z . We also have ei ei , sin = 2i ei + ei . cos = 2 (4.46) (4.47)

4.2.3

Cauchy-Riemann equations

Now it is possible to dene complex functions of complex variables W (z ). For example take a complex function to be dened as W (z ) = z 2 + z, = (x + iy )2 + (x + iy ), = x2 + 2xyi y 2 + x + iy, = In general, we can say W (z ) = (x, y ) + i (x, y ). (4.52) Here and are real functions of real variables. Now W (z ) is dened as analytic at zo if dW exists at zo and is independent of the direction dz in which it was calculated. That is, using the denition of the derivative dW dz =
zo

(4.48) (4.49) (4.50) (4.51)

x2 + x y 2 + i (2xy + y ) .

W (zo + z ) W (zo ) . z

(4.53)

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 dz =
zo

W (xo + iyo + x) W (xo + iyo ) W = x x 222

.
y

(4.54)

For x = C2 , we have z = iy , so dW dz =
zo

W (xo + iyo + iy ) W (xo + iyo ) 1 W = iy i y

= i

W y

.
x

(4.55)

Now for an analytic function, we need W x or, expanding, we need , +i = i +i x x y y = i . y y For equality, and thus path independence of the derivative, we require = , x y = . y x (4.59) (4.57) (4.58) = i W y ,
x

(4.56)

These are the well known Cauchy-Riemann 3 equations for analytic functions of complex variables. They are identical to our kinematic equations for incompressible irrotational uid mechanics. Consequently, any analytic complex function is guaranteed to be a physical solution. There are essentially an innite number of functions to choose from. Thus we dene the complex velocity potential as W (z ) = (x, y ) + i (x, y ), and taking a derivative of the analytic potential, we have dW = +i = u iv. dz x x We can equivalently say dW = i +i dz y y = i y y = u iv. (4.62) (4.61) (4.60)

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, = 2x + 1, = 2y, x x = 2y, = 2x + 1. y y
3

(4.63) (4.64) (4.65)

Augustin-Louis Cauchy, 1789-1857, French mathematician and military engineer, worked in complex analysis, optics, and theory of elasticity.

223

Note that the Cauchy-Riemann equations are satised since derivative is independent of direction, and we can say dW W = dz x

and

= . So the x (4.66)

= (2x + 1) + i(2y ) = 2(x + iy ) + 1 = 2z + 1.

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. (4.67)

= 1, = 0, and = 0, = 1. Since = , the So = x and = y , x y x y x y Cauchy-Riemann equations are not satised, and the derivative depends on direction.

4.3

Elementary complex potentials

Let us examine some simple analytic functions and see the uid mechanics to which they correspond.

4.3.1
Take

Uniform ow
W (z ) = Az, with A complex. (4.68) (4.69)

Then Since A is complex, we can say

dW = A = u iv. dz A = Uo ei = Uo cos iUo sin .

(4.70) (4.71)

Thus we get u = Uo cos , v = Uo sin . This represents a spatially uniform ow with streamlines inclined at angle to the x axis. The ow is sketched in Figure 4.2.

4.3.2
Take

Sources and sinks

W (z ) = A ln z, with A real. (4.72)

With z = rei we have ln z = ln r + i . So W (z ) = A ln r + iA. Consequently, we have for the velocity potential and stream function = A ln r, 224 = A. (4.74) (4.73)

iy
2

-2 -1 0 1 2

-1

-2

Figure 4.2: Streamlines for uniform ow. Now v = , so

A 1 = , v = = 0. (4.75) r r r So the velocity is all radial, and becomes innite at r = 0. We can show that the volume ow rate is bounded, and is in fact a constant. The volume ow rate Q through a surface is vr = Q= vT n dA =
2 0

vr rd =

2 0

A rd = 2A. r

(4.76)

The volume ow 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 W (z ) = Q ln z. 2 (4.77)

For a source located at a point zo which is not at the origin, we can say W (z ) = The ow is sketched in Figure 4.3. Q ln(z zo ). 2 (4.78)

4.3.3

Point vortices
W (z ) = iB ln z, with B real. (4.79) (4.80)

For an ideal point vortex, identical to what we studied in an earlier chapter, we have

So W (z ) = iB (ln r + i ) = B + iB ln r. 225

iy

Figure 4.3: Velocity vectors and equipotential lines for source ow. Consequently, We get the velocity eld from vr = = B, = 0, r = B ln r. 1 B = . r r (4.81)

v =

(4.82)

So we see that the streamlines are circles about the origin, and there is no radial component of velocity. Consider the circulation of this ow =
C

vT dr =

2 0

B rd = 2B. r

(4.83)

So we often write the complex potential in terms of the ideal vortex strength : W (z ) = For an ideal vortex not at z = zo , we say W (z ) = i ln(z zo ). 2 (4.85) i ln z. 2 (4.84)

The point vortex ow 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 eect 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. (4.86) dz 226

iy

Figure 4.4: Streamlines, equipotential, and velocity vectors lines for a point vortex.

iy Q a x a Q

Figure 4.5: Sketch for source-wall interaction.

227

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 potential W (z ) = Q Q ln(z a) + ln(z + a), 2 2
original image

(4.87)

= = = dW dz =

Now on z = iy , which is the location of the wall, we have dW dz = Q 2

Q (ln(z a) + ln(z + a)) , 2 Q (ln(z a)(z + a)) , 2 Q ln(z 2 a2 ), 2 Q 2z . 2 z 2 a2 2iy . y 2 a2

(4.88) (4.89) (4.90) (4.91)

(4.92)

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 v= Q y . y 2 + a2 (4.93)

We nd the location on the wall of the maximum v velocity by setting the derivative with respect to y to be zero, v Q (y + a)2 y (2y ) = = 0. (4.94) y (y 2 + a2 )2 Solving, we nd 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 ) = . (4.95) 2 2 2 (y 2 + a2 )2 We can use Bernoullis equation to nd the pressure eld, assuming steady ow and that p po as r . So Bernoullis equation in this limit p po 1 ()T + = , 2 reduces to 1 Q2 y2 p = po 2 2 2 (y + a2 )2 (4.96)

(4.97)

Note that the pressure is po at y = 0 and is po as y . By integrating the pressure over the wall surface, one would nd that the source exerted a net force on the wall. 228

iy

iy

4

iy

-2 -2

-2

-4

-4 -2 0 2 4

-4

-4 -2 0

-4

2 -4 4 -2 0 2 4

Figure 4.6: Sketch for impingement ow, stagnation ow, and ow in a corner, n = 2.

4.3.5

Flow in corners

Flow in or around a corner can be modelled by the complex potential W (z ) = Az n , = A re

i n n in

with A real, ,

(4.98) (4.99) (4.100) (4.101)

= Ar e , = Ar n (cos(n ) + i sin(n )). So we have = Ar n cos n, = Ar n sin n.

(4.102)

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 at wall. For this ow, we have W (z ) = = = = Az 2 , A(x + iy )2 , A((x2 y 2 ) + i(2xy )), A(x2 y 2 ), = A(2xy ). (4.103) (4.104) (4.105) (4.106)

So the streamlines are hyperbolas. For the velocity eld, we take dW = 2Az = 2A(x + iy ) = u iv, dz u = 2Ax, v = 2Ay. (4.107) (4.108)

This ow actually represents ow in a corner formed by a right angle or ow striking a at plate, or the impingement of two streams. For n = 2, streamlines are sketched in in Figure 4.6. 229

iy Q source
Figure 4.7: Source sink pair.

-Q

sink

4.3.6

Doublets

We can form what is known as a doublet ow 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 ln(z + ) ln(z ), 2 2 Q z+ = . ln 2 z It can be shown by synthetic division that as 0, that W (z ) = z+ 2 =1+ + z z
2

(4.109) (4.110)

2 + .... z2

(4.111)

So the potential approaches

W (z )

2 Q ln 1 + + 2 z

2 +... z2

(4.112)

Now since ln(1 + x) x as x 0, we get for small W (z ) Now if we require that lim we have W (z ) = So (x, y ) = In polar coordinates, we then say
0

that (4.113)

Q 2 Q . 2 z z

Q ,

(4.114) (4.115) (4.116)

x iy (x iy ) = = 2 . z x + iy x iy x + y2 x , x2 + y 2 (x, y ) = y . x2 + y 2

sin cos , = r r Streamlines and equipotential lines for a doublet are plotted in Figure 4.8. = 230

(4.117)

iy

=a 4

=b

=a 2 =b

=c 0

-2

-4

-4

-2

Figure 4.8: Streamlines and equipotential lines for a doublet.

4.3.7

Rankine half body

Now consider the superposition of a uniform stream and a source, which we dene to be a Rankine half body: W (z ) = U z + Q ln z, with U, Q real, 2 Q (ln r + i ), = U rei + 2 Q (ln r + i ), = U r (cos + i sin ) + 2 Q Q = U r cos + ln r + i U r sin + . 2 2 (4.118) (4.119) (4.120) (4.121)

Q Q ln r, = U r sin + . (4.122) 2 2 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, (4.123) dz 2z we get = U r cos + U+ Q 1 i e = u iv, 2 r 231 (4.124)

So

iy
4

-2

-4 -4 -2 0 2 4

Figure 4.9: Streamlines for a Rankine half body. Q1 (cos i sin ) = u iv, 2 r Q Q u=U+ cos , v = sin . 2r 2r U+ When = , we get u = 0 when; 0 = U+ r = Q (1), 2r (4.127) (4.128) (4.125) (4.126)

Q . 2U

4.3.8

Flow over a cylinder

We can model ow past a cylinder without circulation by superposing a uniform ow with a doublet. Dening a2 = U , we write W (z ) = U z + a2 =U z+ , z z a2 i = U re + i , re a2 = U r (cos + i sin ) + (cos i sin ) , r 2 a a2 = U r cos + cos + i r sin sin r r a2 a2 . = U r cos 1 + 2 + i sin 1 2 r r 232 (4.129) (4.130) (4.131) , (4.132) (4.133)

iy
2

-1

-2 -1

-2

Figure 4.10: Streamlines and equipotential lines for ow over a cylinder without circulation. a2 a2 , = U r sin 1 . (4.134) r2 r2 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 = U r cos 1 + vr = a2 a2 = U cos 1 + 2 + U r cos 2 3 , r r r 2 a = U cos 1 2 , r a2 1 = U sin 1 + 2 . = r r ()T = 4U 2 sin2 . (4.135) (4.136) (4.137) So

So on r = a, we have vr = 0, and v = 2U sin . Thus on the surface, we have Bernoullis equation for a steady ow with p p as r then gives p U 2 p 1 + ()T = + , 2 2 1 p = p + U 2 (1 4 sin2 ). 2 233

(4.138)

(4.139) (4.140)

C
1

0.5 -1

1.5

2.5

experiment potential theory

-2

-3

pressure distribution on cylinder surface from potential theory

Figure 4.11: Pressure distribution for ideal ow over a cylinder without circulation. The pressure coecient Cp , dened below, then is p p Cp 1 2 = 1 4 sin2 . U 2

(4.141)

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 uids, a phenomenon known as ow 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 4.14
For a cylinder of radius c at rest in an accelerating potential ow eld with a far eld velocity of U = a + bt, nd the pressure on the stagnation point of the cylinder. The velocity potential and velocities for this ow are (r, , t) vr v 1 ()T 2 = (a + bt)r cos 1 + = = = c2 r2 , (4.142) (4.143) , c2 r2
2

c2 = (a + bt) cos 1 2 , r r 1 c2 = (a + bt) sin 1 + 2 r r 1 (a + bt)2 2 cos2 1 c2 r2

2

(4.144) , (4.145)

+ sin2 1 +

234

2c2 c4 1 (a + bt)2 1 + 4 + 2 sin2 cos2 2 r r

t :

(4.146)

Also, since the ow is unsteady, we will need

c2 = br cos 1 + 2 t r Now we note in the limit as r that br cos , t

(4.147)

1 1 ()T (a + bt)2 . 2 2

(4.148)

We also note that on the surface of the cylinder vr (r = c, , t) = 0. Bernoullis equation gives us p 1 + ()T + = f (t). t 2 We use the far eld behavior to evaluate f (t): p 1 br cos + (a + bt)2 + = f (t). 2
1 Now if we make the non-intuitive choice of f (t) = 2 (a + bt)2 + po ,

(4.149)

(4.150)

(4.151) we get (4.152)

p 1 po 1 br cos + (a + bt)2 + = (a + bt)2 + . 2 2 So p = po br cos = po bx.

(4.153)

Note that since the ow at innity is accelerating, there must be a far-eld pressure gradient to induce this acceleration. Consider the x momentum equation in the far eld du p = , dt x (b) = (b). (4.154) (4.155)

So for the pressure eld, we have br cos 1 + c2 r2 2c2 p 1 po 1 c4 + (a + bt)2 1 + 4 + 2 (sin2 cos2 ) + = (a + bt)2 + , 2 r r 2 c2 r2 c4 2c2 + (sin2 cos2 ) . r4 r2 (4.156)

which reduces to p(r, , t) = po br cos 1 + 1 (a + bt)2 2 (4.157)

For the stagnation point, we evaluate as 1 p(c, , t) = po bc(1) (1 + 1) (a + bt)2 (1 + 2(1)(0 1)) , 2 1 2 = po + (a + bt) + 2bc. 2 (4.158) (4.159)

The rst two terms would be predicted by a naive extension of the steady Bernoullis equation. The nal term however is not intuitive and is a purely unsteady eect.

235

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 rst 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 integrals, 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. i with 0 2 . For such a contour dz = iRe i d . The contour in each is C : z = Re 4.4.1.1 Simple pole

We describe a simple pole with the complex potential a W (z ) = . z and the contour integral is W (z )dz = a dz = z
2 0 =2 =0

(4.160)

a i iRe d, i Re

(4.161) (4.162)

= ai 4.4.1.2 Constant potential

d = 2ia.

We describe a constant with the complex potential W (z ) = b. and the contour integral is W (z )dz = = since e0i = e2i = 1. 4.4.1.3 Uniform ow bdz =
2 =2 =0

(4.163)

i d, biRe

(4.164) (4.165)

biR ei i

= 0.
0

We describe a constant with the complex potential W (z ) = cz. 236 (4.166)

and the contour integral is

C

W (z )dz =

czdz = 2
0 2

=2 =0 2

i iRe i d, cRe
2

(4.167) (4.168)

= icR since e0i = e4i = 1. 4.4.1.4 Quadrapole

2 icR e2i d = 2i

= 0.
0

A quadrapole potential is described by W (z ) = Taking the contour integral, we nd

C

k z2

(4.169)

So the only non-zero contour integral is for functions of the form W (z ) = a . We nd all z polynomial powers of z have a zero contour integral about the origin for arbitrary contours except this special one.

i 2 iRe k d, dz = k 2 e2i z2 0 R ki 2 i ki 1 i = e e d = 0 i R R

(4.170)
2

= 0.
0

(4.171)

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 + . . . . (4.172) In compact summation notation, we can say W (z ) =
n= n= n= C n= n= n=

Cn (z zo )n .

(4.173)

Taking the contour integral of both sides we get

C

W (z )dz = =

Cn (z zo )n dz,
C

(4.174) (4.175) (4.176) (4.177)

Cn

(z zo )n dz,

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.

this has value 2i only when n = 1, so = C1 2i

237

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 nite number of singular points, then
C

W (z )dz = 2i

residues.

(4.178)

The constants Cn can be shown to be found by evaluating the contour integral Cn = 1 2i W (z ) dz, (z zo )n+1 (4.179)

where C is any closed contour which has zo in its interior.

4.5

Pressure distribution for steady ow

For steady, irrotational, incompressible ow with no body force present, we have the Bernoulli equation: p 1 p 1 2 + ()T = + U . (4.180) 2 2 We can write this in terms of the complex potential in a simple fashion. First, recall that ()T = u2 + v 2 . We also have
dW dz

(4.181)

= u iv , so

dW dz

= u + iv . Consequently, (4.182)

dW dW = u2 + v 2 = ()T . dz dz So we get the pressure eld from Bernoullis equation to be 1 dW dW 2 p = p + U 2 dz dz The pressure coecient Cp is Cp = p p 1 dW dW =1 2 . 1 2 U dz dz U 2

(4.183)

(4.184)

4.6

Blasius force theorem

For steady ows, we can nd the net contribution of a pressure force on an arbitrary shaped solid body with the Blasius 5 force theorem. 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 .
Paul Richard Heinrich Blasius, 1883-1970, student of Ludwig Prandtl, wrote thesis that gave mathematical description of similarity solution to boundary layer problem.
5

238

iy n

Sb

Figure 4.12: Potential ow about arbitrarily shaped two-dimensional body with uid control volume indicated. First consider the linear momenta equation for steady ow, no body forces, and no viscous forces, vT v = p, T (vvT )
V T

add mass to get conservative form, integrate over V , use Gauss,

(4.185) (4.186) (4.187) (4.188)

T (vvT )
S

= p, =
V S

dV

p dV, pndS.

v(vT n)dS =

Now the surface integral here is really a line integral with unit depth b, dS = bds. Moreover the surface enclosing the uid 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 ds dy ds

n=

dy ds dx ds

on C.

(4.189)

Moreover, on Sb we have, since it is a solid surface vT n = 0, on C. (4.190)

Now let the force on the body due to uid pressure be F:

Sb

pndS = F.

(4.191)

Now return to our linear momentum equation

S

vvT ndS = 239

pndS,

break this up,

(4.192)

Sb

v vT n dS +
=0

vvT ndS =
T

Sb

pndS

pndS,

(4.193)

=F
C

vv ndS = F

pndS.

(4.194)

We can break this into x and y components: u u dx dy v bds = ds ds dy dx v bds = ds ds

vT n vT n

dy bds Fx , ds

(4.195)

v u

dx bds Fy , ds

(4.196) (4.197) (4.198) (4.199)

solving for Fx and Fy per unit depth, Fx = b Fy = b

C

pdy u2 dy + uvdx, pdx + v 2 dx uvdy.

1 (u2 + v 2 ), where K is some constant. So the x force per Now Bernoulli gives us p = K 2 unit depth becomes

Fx = b

1 Kdy + (u2 + v 2 )dy u2 dy + uvdx, 2 C since the integral over a closed contour of a constant K is zero, 1 = (u2 + v 2 )dy + uvdx, C 2 1 = (u2 + v 2 )dy + 2uvdx. 2 C Fy = b 1 Kdx (u2 + v 2 )dy + v 2 dx uvdy, 2 C 1 = (u2 + v 2 )dx 2uvdx. 2 C
Fx iFy : b

(4.200) (4.201) (4.202) (4.203)

Similarly for the y direction, we get (4.204) (4.205)

Now consider the group of terms Fx iFy b 1 2 1 = 2 1 = 2 1 = 2 =

(u2 + v 2 )dy + 2uvdx (u2 + v 2 )idx + 2uvidy, (i(u2 v 2 ) + 2uv )dx + ((u2 + v 2 ) + 2uvi)dy, (i(u2 v 2 ) + 2uv )dx + (i(u2 v 2 ) + 2uv )idy, (i(u2 v 2 ) + 2uv )(dx + idy ), 240

(4.206) (4.207) (4.208) (4.209)

iy Uo

Q x

Figure 4.13: Potential ow about arbitrarily shaped two-dimensional body with distribution of sources, sinks, vortices, and dipoles. = 1 2 1 i = 2 i(u iv )2 (dx + idy ), dW dz
2

(4.210) (4.211)

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 ow with a freestream constant velocity of Uo . We take an arbitrary body shape to enclose a distribution of canceling source sink pairs, doublets, point vortices, quadrapoles, and any other non-mass adding potential ow term. This combination gives rise to some surface which is a streamline. Now far from the body surface a contour sees all of these features as eectively concentrated at the origin. Then, the potential can be written as W (z ) Uz
uniform ow

Q Q ln z ln z + 2 2
canceling source sink pair

i ln z 2
clockwise! vortex

z
doublet

+...

(4.212)

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 = D, b 241 Fy = L. b (4.213)

Now by the Blasius force theorem, we have 1 i D iL = 2 1 = i 2 1 = i 2 dW dz

2

dz,

(4.214) (4.215) (4.216)

2 i U+ + . . . dz, 2z z 2 iU 1 2 U2 + 2 + 2U + . . . dz. z z 4 2

Now the Cauchy integral theorem gives is the contour integral is 2i residue is iU/ . So we get D iL = So we see that D = 0, L = U . Note that 1 iU i 2i 2 = iU ,

residues. Here the

(4.217) (4.218)

(4.219) (4.220)

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 ow, there is no drag. Consideration of either unsteady or viscous eects would lead to a non-zero x component of force.

Example 4.15
Consider the ow over a cylinder of radius a with clockwise circulation . To do so, we can superpose a point vortex onto the potential for ow over a cylinder in the following fashion: W (z ) = U z+ a2 z + i z . ln 2 a (4.221)

Breaking this up as before into real and complex parts, we get W (z ) = So, we nd = (W (z )) = U r sin 1 a2 r2 + r ln . 2 a (4.223) U r cos 1 + a2 r2 + i U r sin 1 a2 r2 + i r ln + i 2 a (4.222)

On r = a, we nd 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

242

is the potential ow solution for ow 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 ow, the trailing edge of an airfoil is a stagnation point. The Kutta-Zhukovsky 7 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 eld from the velocity potential: = vr vr |r=a v v |r=a = = = = = (W (z )) = U r cos 1 + a2 r2 , 2 , (4.224) (4.225) (4.226) (4.227) (4.228) (4.229)

a2 2a2 = U r cos 3 + U cos 1 + 2 r r r 3 2 2a a U cos 3 + 1 + 2 = 0, a a a2 1 1 U r sin 1 + 2 , = r r r 2 a2 U sin 1 + 2 , a 2a 2U sin . 2a

We get the pressure on the cylinder surface from Bernoullis equation: 1 p = p + U 2 2 1 = p + U 2 2 1 ()T , 2 1 2U sin 2 2a (4.230)
2

(4.231)

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 b dFy b = p cos ad, = p sin ad. (4.232) (4.233)

So integrating over the entire cylinder, we obtain, Fx b Fy b

2

=
0 2

1 1 p + U 2 2U sin 2 2 2a 1 1 p + U 2 2U sin 2 2 2a

cos ad,
2

(4.234) (4.235) (4.236)

=
0

sin ad,

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 dierential 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.

243

iy

loc pre al ss ure

for ce

Figure 4.14: Pressure force on a dierential area element of cylindrical surface.

Integration via computer algebra gives Fx b Fy b = 0, = U . (4.237) (4.238)

This is identical to the result we expect from the Kutta-Zhukovsky lift theorem.

4.8

Conformal mapping

Conformal mapping is a technique by which we can render results obtained for simple ows, such as those over a cylinder, applicable to ows over more complicated geometries. We will not consider these in any detail here, but the reader should refer to texts on potential ow 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 elds in terms of the warped coordinates, then transform them back into ordinary space to get the actual pressure eld.

244

Chapter 5 Viscous incompressible laminar ow

see Panton, Chapter 7, 11 see Yih, Chapter 7 Here we consider a few standard problems in viscous incompressible laminar ow. For this entire chapter, we will make the following assumptions: the ow is incompressible, body forces are negligible, the uid properties, c, and k , are constants,

5.1

Fully developed, one dimensional solutions

The rst type of solution we will consider is known as a one-dimensional fully developed solution. These are commonly considered in rst courses in uid mechanics and heat transfer. The ows here are essentially one-dimensional, but not absolutely, as they were in the chapter on one-dimensional compressible ow. In this section, we will further enforce that the ow is time-independent, o = 0, the velocity and temperature gradients in the x and z direction are zero, v = 0, T = 0, T = 0. z x z
v x

= 0,

We will see that these assumptions give rise to ows 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 ow in a slot

Consider the ow sketched in Figure 5.1. Here we have a large reservoir of uid with a long narrow slot located around y = 0. We take the length of the slot in the z direction, b, to be very 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 245

b = constant y = constant z x L

po

p1

Figure 5.1: Pressure gradient driven ow in a slot. pressure at the entrance of the slot po is higher than atmospheric. At the end of the slot, the uid expels to the atmosphere which is at p1 . Hence, there is a pressure gradient in the x direction, which drives the ow in the slot. We will see that the ow is resisted by viscous stresses. An analogous ow in a circular duct is dened as a Hagen-Poiseuille 1 2 ow. Near x = 0, the ow accelerates in what is known as the entrance length. If L is suciently long, we observe that the uid 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 eld. For this ow, let us make the additional assumptions that there is no imposed pressure gradient in the z direction, 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, o vi + vj j vi = i p + j j vi , co T + cvj j T = ki i T + 2(i vj ) (i vj ) . (5.1) (5.2) (5.3)

Here we have ve equations in ve unknowns, vi , p, and T . As for all incompressible ows with constant properties, we can get the velocity eld by only considering the mass and momenta equations; velocity is only coupled one way to the energy equation.
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 ow.
1

246

The mass equation, recalling that gradients in x and z are zero gives us u+ v+ w = 0. x y z
=0 =0

(5.4)

v = 0. (5.5) y Now, from our assumptions of steady and fully developed ow, we know that v cannot be a function of x, z , or t. So the partial becomes a total derivative, and mass conservation holds dv that dy = 0. Integrating, we nd 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. Now consider the x momentum equation:
2 p 2 2 u + u u+ v u + w u = + u + u + u ,(5.7) x2 2 2 t x y z x y z =0
=0 =0 =0 =0 =0

So the mass equation gives us

(5.6)

0 =

p u + 2. x y

(5.8)

We note for this fully developed ow 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 v +u v + v v v +w t =0 x =0 y =0 z =0 =0
=0 =0 =0

p y

(5.9)

p 0 = . y

2 2 2 v + v + x2 y 2 =0 z 2
=0 =0

v , (5.10)

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+ v w + w w = t x y z z =0
=0 =0 =0 =0

(5.12)

247

2w . 0 = y 2 Solution of this partial dierential equation gives us w = f (x, z )y + g (x, z ).

w + 2 w + 2 w , 2 x y z
=0 =0

(5.13)

(5.14)

Now to satisfy no-slip, we must have w = 0 at y = h . This leads us to two linear equations 2 for f and g : h 1 f (x, z ) 0 2 = . (5.15) h 1 g ( x, z ) 0 2 Since the determinant of the coecient 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. Next consider how the energy equation reduces: c
2 2 2 = k T + c u T + v T T + T T + w T + x x2 2 2 t y z y z =0 =0
=0 =0 =0 =0

(5.16)

=0

+2(i vj ) (i vj ) , 2T 0 = k 2 + 2(i vj ) (i vj ) . y

(5.17) (5.18)

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 thermal diusion. Thus the energy path is 1) viscous work is done to generate thermal energy, 2) thermal energy diuses throughout the channel and out the boundary. Now consider the viscous dissipation term for this ow.

1 v1
=0

1 v2 2 v2
=0 =0

1 v3 2 v3
=0 =0

i vj =

2 v1

3 v1
=0

3 v2
=0 =0

=0

3 v3
=0 =0 1 2

=0

0 = 2 v1 0

0 0 0 0, 0 0

1 u 2 y

(5.19)

(i vj ) =

0
1 2

2 v 1

+ 1 v2
=0

2 v1

+ 1 v2 0
=0

0 0 248

u =1 2 y 0

0 0

0 0. 0

(5.20)

Further, (i vj ) (i vj ) = 1 u 2 y
2

1 u 2 y

1 2

u y

(5.21)

So the energy equation becomes nally 2T u 0=k 2 + y y

2

(5.22)

At this point we have the x momentum and energy equations as the only two which seem to have any substance. 0 = 2u p + 2, x y
2

(5.23) . (5.24)

2T u 0 = k 2 + 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 ows solving a Poisson 3 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 = 2 + , x x y 2 2p 2 u . 0 = 2+ 2 x y x
=0

(5.25) (5.26)

The viscous term above is zero because of our assumption of fully developed ow. Moreover, since p = p(x) only, we then get d2 p = 0, dx2 p(0) = po , p(L) = p1 , (5.27)

which has a solution showing the pressure eld must be linear in x: p(x) = po po p 1 x, L dp po p 1 = . dx L (5.28) (5.29)

Now, since u is at most a function of y , we can convert partial derivatives to ordinary derivatives, and write the x momentum equation and energy equation as two ordinary dierential
Sim eon Denis Poisson, 1781-1840, French mathematician taught by Laplace, Lagrange, and Legendre, studied partial dierential equations, potential theory, elasticity, and electrodynamics.
3

249

equations in two unknowns with appropriate boundary conditions at the wall y = h : 2 d2 u po p 1 = , 2 dy L d2 T dy 2 = k du dy

2

u(h/2) = 0, ,

u(h/2) = 0, T (h/2) = To .

(5.30) (5.31)

T (h/2) = To ,

We could solve these equations directly, but instead let us rst cast them in dimensionless form. This will give our results some universality and eciency. Moreover, it will reveal more fundamental groups of terms which govern the uid behavior. Let us select scales such that dimensionless variables, denoted by a * subscript, are as follows y = y , h T = T To , To u = u . uc (5.32)

We have yet to determine the characteristic velocity uc . Note that the dimensionless temperature has been chosen to render it zero at the boundaries. With these choices, the x momentum equation becomes uc d 2 u 2 h2 dy d 2 u 2 dy uc u (x h = h/2) u (x = 1/2) po p 1 , L (po p1 )h2 = , Luc = uc u (x h = h/2) = 0, = u (x = 1/2) = 0. = (5.33) (5.34) (5.35) (5.36)

Let us now choose the characteristic velocity to render the x momentum equation to have a simple form: (po p1 )h2 . (5.37) uc L Now scale the energy equation: To d 2 T u2 c = 2 h2 dy kh2 d 2 T u2 c = 2 dy kTo = du dy du dy
2

,
2

(5.38) (5.39)
2

, ,
2

c u2 c k cTo

du dy

(5.40) (5.41) (5.42)

du dy T (1/2) = T (1/2) = 0. = P rEc 250

Here we have grouped terms so that the Prandtl number P r = c/k , explicitly appears. Further, we have dened the Eckert 4 number Ec as u2 Ec = c = cTo
(po p1 )h2 2 L

cTo

(5.43)

In summary our dimensionless dierential equations and boundary conditions are d 2 u = 1, 2 dy u(1/2) = 0,
2

(5.44) T (1/2) = 0. (5.45)

du d 2 T = P rEc 2 dy dy

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 equation, and the viscous dissipation is a forcing term in the energy equation. The solution for the velocity eld which satises the dierential equation and boundary conditions is quadratic in y and is 1 u = 2 1 2
2 2 . y

(5.46)

Note that the maximum velocity occurs at y = 0 and has value 1 umax = . 8 The mean velocity is found through integrating the velocity eld to arrive at umean = = =
1/2 1/2 1/2 1/2

(5.47)

u (y )dy , 1 2 1 2
2 2 dy , y 1/2

(5.48) (5.49) (5.50) (5.51)

1 3 1 1 y y 2 4 3 1 = . 12

,
1/2

Note that we could have scaled the velocity eld 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 u
(po p1 L )h2

1 2

1 2

y h

(5.52)

The velocity prole is sketched in Figure 5.2.

4

E. R. G. Eckert, scholar of convective heat transfer at University of Minnesota.

251

y = 1/2 *

y = -1/2 *

Figure 5.2: Velocity prole for pressure gradient driven ow in a slot. Now let us get the temperature eld. d 2 T = 2 dy = = dT = dy T = 0 = 0 = C1 = Regrouping, we nd that P rEc 1 12 2 In terms of dimensional quantities, we can say T = T To (po p1 )2 h4 = To 12L2 kTo d 1 1 2 2 y dy 2 2 2 P rEc (y ) , 2 P rEc y , 1 3 P rEc y + C1 , 3 1 4 + C 1 y + C 2 , P rEc y 12 1 1 1 P rEc + C1 + C2 , 12 16 2 1 1 1 C1 + C2 , P rEc 12 16 2 P rEc 0, C2 = 192 P rEc
4 4 y . 2

(5.53) (5.54) (5.55) (5.56) (5.57)

1 y = , 2 1 y = , 2

(5.58) (5.59) (5.60)

(5.61)

1 2

y h

(5.62)

The temperature prole is sketched in Figure 5.3. From knowledge of the velocity and temperature eld, we can calculate other quantities of interest. Let us calculate the eld of shear stress and heat ux, 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 ,
=0

(5.63) (5.64)

= 2(i vj ) 252

1/2

T *

-1/2

Figure 5.3: Temperature prole for pressure gradient driven ow in a slot. We have already seen the only non-zero components of the symmetric part of the velocity gradient tensor are the 12 and 21 components. Thus the 21 stress component is 21 = 2(2 v1) = = 2 v1 . In x y space, we then say here that
2

2 v1 + 1 v2
=0

(5.65) (5.66)

du . (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 dene a characteristic shear stress c , so that the scale shear is = yx /c . Thus our equation for shear becomes yx = c = Now take c With this denition, we get uc du . h dy (5.68)

h uc (po p1 )h2 = = (po p1 ) . h hL L =

(5.69)

du . dy Evaluating for the velocity prole of the pressure gradient driven ow, we nd = y .

(5.70)

(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 ) 2 L 253 (5.72)

y = 1/2 *

y = -1/2

Figure 5.4: Shear stress prole for pressure gradient driven ow in a slot. Note that the wall shear stress is governed by the pressure dierence and not the viscosity. However, the viscosity plays a determining role in selecting the maximum uid velocity. The shear prole is sketched in Figure 5.4. Next, let us calculate the heat ux vector. Recall that, for this ow, with no x or z variation of T , we have the heat ux vector as qy = k T . y (5.73)

Now dene scale the heat ux by a characteristic heat ux qc , to be determined, to obtain a dimensionless heat ux: qy q = . (5.74) qc So, qc q = q Let qc kTo /h, so q = q = dT , dy (5.77) (5.78) kTo dT , h dy kTo dT . = hqc dt (5.75) (5.76)

1 3 P rEc y . 3

For our ow, we have a cubic variation of the heat ux vector. There is no heat ux at the centerline, which corresponds to this being a region of no shear. The magnitude of the heat ux is maximum at the wall, the region of maximum shear. At the upper wall, we have q |y =1/2 = The heat ux prole is sketched in Figure 5.5. 254 1 P rEc. 24 (5.79)

y = 1/2 *

y = -1/2

Figure 5.5: Heat ux prole for pressure gradient driven ow in a slot.

L y = h, u = U U

p=p y

p=p

y = 0, u = 0

Figure 5.6: Conguration for Couette ow with pressure gradient. In dimensional terms we have qw
kTo h

1 (po p1 )2 h4 , 24 L2 kTo 1 (po p1 )2 h3 , 24 L2

(5.80) (5.81) (5.82)

qw =

5.1.2

Couette ow with pressure gradient

We next consider Couette ow with a pressure gradient. Couette ow implies that there is a moving plate at one boundary and a xed plate at the other. It is a very common experimental conguration, and used often to actually determine a uids viscosity. Here we will take the same assumptions as for pressure gradient driven ow 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 conguration for this ow is shown in Figure 5.6. Our equations governing this ow are po p 1 d2 u = , dy 2 L u(0) = 0, 255 u(h) = U, (5.83)

d2 T dy 2

du dy

T (0) = To ,

T (h) = To .

(5.84)

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 diusion of the thermal energy, ultimately out of the system through the 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 = y , h T = T To , To u = u . U (5.85)

The momentum equation becomes U d 2 u po p 1 = , 2 2 h dy L d 2 u (po p1 )h2 , = 2 dy U L (po p1 )h2 with dimensionless pressure gradient P , U L d 2 u = P , 2 dy u (0) = 0, u (1) = 1. This has solution (5.86) (5.87) (5.88) (5.89) (5.90) (5.91)

1 2 + C 1 y + C 2 . u = P y 2 Applying the boundary conditions, we get 1 0 = P (0)2 + C1 (0) + C2 , 2 0 = C2 , 1 1 = P (1)2 + C1 (1), 2 1 C1 = 1 + P , 2 1 1 2 u = P y + 1 + P y , 2 2 1 u = P y (1 y ) + y 2

pressure eect

(5.92)

(5.93) (5.94) (5.95) (5.96) (5.97) . (5.98)

Couette eect

256

y=1 u=1 * *

-6

-2 P= 0 P= 2

P=

P=

Figure 5.7: Velocity proles for various values of P for Couette ow with pressure gradient. We see that the pressure gradient generates a velocity prole that is quadratic in y . This is distinguished from the Couette eect, that is the eect of the upper plates motion, which gives a linear prole. Because our governing equation here is linear, it is appropriate to think of these as superposed solutions. Velocity proles for various values of P are shown in Figure 5.7. , and taking = /c , we get Let us now calculate the shear stress prole. With = du dy c = taking c |y =0 |y =1 U du , h dy U du , = hc dy U , h du , so here, = dy 1 = P y + P + 1, 2 1 = P + 1, 2 1 = P + 1, 2 (5.99) (5.100) (5.101) (5.102) and (5.103) (5.104) (5.105) (5.106) The wall shear has a pressure gradient eect and a Couette eect as well. In fact we can select a pressure gradient to balance the Couette eect at one or the other wall, but not both. We can also calculate the dimensionless volume ow rate Q , which for incompressible ow, is directly proportional to the mass ux. Ignoring how the scaling would be done, we arrive at Q = =
1 0 1 0

y =0 *

u=0 *

u dy , 1 2 1 P y + 1 + P y dy , 2 2 257

(5.107) (5.108)

1 3 1 y2 1 P y + 1+ P , 6 2 2 0 0 P 1 1 = + 1+ P , 6 2 2 P 1 = + . 12 2 =

(5.109) (5.110) (5.111)

Again there is a pressure gradient contribution and a Couette contribution, and we could select P to give no net volume ow rate. We can summarize some of the special cases as follows
1 P P : u = 2 P y (1 y ); = P 1 y , Q = 12 . Here the uid ows in the 2 opposite direction as driven by the plate because of the large pressure gradient. 2 P = 6. Here we get no net mass ow and u = 3y 2y , = 2y , Q = 0. 1 2 . P = 2. Here we get no shear at the bottom wall and u = y , = 2y , Q 3 1 P = 0. Here we have no pressure gradient and u = y , = 1, Q = 2 . 2 P = 2. Here we get no shear at the top wall and u = y + 2y , = 2y + 2, 2 Q = 3 .

P : u = 1 P y (1 y ); = P 2 same direction as driven by the plate.

1 2

y , Q =

P . 12

Here the uid ows in the

We now consider the heat transfer problem. Scaling, we get To d 2 T U 2 = 2 h2 dy kh2 U 2 d 2 T = 2 dy kTo du dy du dy
2

,
2

T (0) = T (1) = 0,

(5.112) (5.113)

,
2

c U 2 = k cTo

du dy

,
2

(5.114) (5.115) (5.116)

2

dT dy T

1 = P rEc P y + P + 1 , 2 1 2 1 2 P + 1 y + 1 + P , = P rEc P 2 y 2P 2 2 1 2 P2 3 1 2 = P rEc y P P + 1 y + 1 + P y + C 1 , 3 2 2 2 1 1 2 2 P 4 P 1 3 y P + 1 y + 1 + P y = P rEc 12 3 2 2 2 258

du = P rEc dy 2 = P rEc ,

(5.117) (5.118) (5.119)

+C1 y + C2 , T (0) = 0 = C2 , T (1) = 0 = P rEc C1 T

(5.120) (5.121)
2

P2 P 1 1 1 P +1 + 1+ P 12 3 2 2 2 2 1 P P = P rEc , + + 2 6 24 1 1 P2 4 P 1 3 y P + 1 y + = P rEc 1+ P 12 3 2 2 2 2 1 P P y +P rEc + + 2 6 24

+ C1 ,

(5.122) (5.123)

2 y

(5.124)

Factoring, we can write the temperature prole as T = P rEc 2 y (1 y )(12 + 4P + P 2 8P y 2P 2 y + 2P 2 y ). 24 (5.125)

. Scaling, we get For the wall heat transfer, recall qy = k dT dy qc q = q choosing qc q So q 1 P2 3 1 2 y P P + 1 y + 1+ P = P rEc 3 2 2
2

kTo dT , h dy kTo dT , = hqc dy kTo , h dT = . dy

(5.126) (5.127) (5.128) (5.129)

1 P P2 y . (5.130) 2 6 24

At the bottom wall y = 0, we get for the heat transfer vector q |y =0 = P rEc 1 P P2 + + . 2 6 24 (5.131)

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 dierential equations. The similarity transformation actually reveals a hidden symmetry of the partial dierential 259

x U

Figure 5.8: Schematic for Stokes rst problem of a suddenly accelerated plate diusing linear momentum into a uid at rest. equations by dening a new independent variable, which is a grouping of the original independent variables, under which the partial dierential equations transform into ordinary dierential equations. We then solve the resulting ordinary dierential equations by standard techniques.

5.2.1

Stokes rst problem

The rst problem we will consider which uses a similarity transformation is known as Stokes rst problem, as Stokes addressed it in his original work which developed the Navier-Stokes equations in the mid-nineteenth century. The problem is described as follows, and is sketched in Figure 5.8. Consider a at plate of innite 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 uid of semi-innite extent which is at rest at time t < 0. At t = 0, the at plate is suddenly accelerated to a constant velocity of U , entirely in the x direction. Because the no-slip condition is satised for the viscous ow, this induces the uid at the plate surface to acquire an instantaneous velocity of u(0) = U . Because of diusion of linear x momentum via tangential viscous shear forces, the uid 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 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 ow, and 2) we will assume all pressure gradients are zero; hence the uid has a constant pressure. Under these assumptions, the x momentum equation,
2 2 p 2 u + u u + w u+ v u = + u + u + u (5.132) , x2 2 2 t x y z x y z =0
=0 =0 =0 =0 =0

is the only relevant component of linear momenta, and reduces to u t = 2u . y 2 (5.133)

(mass)(acceleration)

shear force

260

The energy equation reduces as follows c

2 2 2 = k T + v T T + T T + c u T + w T + x2 x 2 2 t y z y z =0 =0
=0 =0 =0 =0

+2(i vj ) (i vj ) , k 2T y 2 +

(5.134) (5.135) .

T t

u y

energy increase

thermal diusion

viscous work source

Let us rst consider the x momentum equation. Recalling the momentum diusivity denition = , we get the following partial dierential equation, initial and boundary conditions: 2u u = 2, t y u(0, t) = U, u(, t) = 0. u , U t , tc y . yc (5.136) (5.137)

u(y, 0) = 0,

Now let us scale the equations. Choose u = t = y = (5.138)

We have yet to choose characteristic length, (yc ), and time, (tc ), scales. The equations become U 2 u U u = , 2 y 2 tc t yc u tc 2 u = . 2 y 2 t yc Wasting no time, we choose yc Noting the SI units, we see choice, we get
U

(5.139) (5.140)

= . U U
N s m3 s m2 kg m

(5.141) =
kg m s m3 s s2 m2 kg m

has units of length:

= m. With this (5.142)

tc U 2 tc U 2 tc = = . 2 yc 2 tc =

. U2 With all of these choices the complete system can be written as 2 u u , = 2 t y u (0, t ) = 1, u (, t ) = 0. 261

This suggests we choose

(5.143)

(5.144) (5.145)

u (y , 0) = 0,

Now for self-similarity, we seek transformation which reduce this partial dierential equation, as well as its initial and boundary conditions, into an ordinary dierential 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 rst consider a general transformation from a y , t coordinate system to a new , t coordinate system. We assume then a general transformation = (y , t ), = t (y , t ). t (5.146) (5.147)

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 The chain rule then gives expressions for derivatives: t y =
y

(5.148) t t

t y

+
y

t t

,
y

(5.149) (5.150)

=
t

+
t

t y

.
t

Now we will restrict ourselves to the transformation = t , t so we have

t t y

(5.151)

= 1 and

t y t

= 0, so our rules for dierentiation reduce to =

y

t y

t y

+
y

(5.152) (5.153)

=
t

.
t

The next assumption is key for a similarity solution to exist. We restrict ourselves to . Hence transformations for which = ( ). That is we allow no dependence of on t we must require that t = 0. Moreover, partial derivatives of become total derivatives, giving us a nal form of transformations for the derivatives t y In terms of operators we can say t y =
y

=
y

d d t

,
y

(5.154) (5.155)

d = d y t y 262

.
t

d , d d . d

(5.156) (5.157)

=
t

Now returning to Stokes rst problem, let us assume that a similarity solution exists of the form u (y , t ) = u ( ). It is not always possible to nd a similarity variable . One of the more robust ways to nd a similarity variable, if it exists, comes from group theory, 5 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 nd simpler forms. In the same sense that a snowake, 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 dierential 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 dierential equation along with initial and boundary conditions to a particularly simple type of transformation, a simple stretching of space, time, and velocity: = e a t , t y = e b y , u = e c u . (5.158) 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 reecting transformation. We note that with this stretching, the domain of the problem remains [0, ); y [0, ) maps into y unchanged; that is t [0, ) maps into t [0, ).
Group theory has a long history in mathematics and physics. Its complicated origins generally include attribution to Evariste Galois, 1811-1832, a somewhat romantic gure, as well as Niels Henrick Abel, 18021829, the Norwegian mathematician. Critical developments were formalized by Marius Sophus Lie, 18421899, another Norwegian mathematician, in what today is known as Lie group theory. A modern variant, known as renomralization group (RNG) theory is an area for active research. The 1982 Nobel prize in physics went to Kenneth G. Wilson, 1936-, of Cornell University and The Ohio State University, for use of RNG in studying phase transitions, rst done in the 1970s. The award citation refers to the possibilities of using RNG in studying the great unsolved problem of turbulence, an active area of research in which Steven A. Orszag has 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 nite or innite 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 dened 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. 2. Associativity: The dened 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 homomorphism. 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.
5

263

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 t c ) = 1, e u (0, t e ac u = e2bc 2u , y 2 ) = 0. ec u (, t (5.159) (5.160)

ec u ( y , 0) = 0,

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, So our symmetry transformation is = e2b t , t giving in transformed space 2u u , = y 2 t ) = 0. u (0, t) = 1, u (, t (5.163) (5.164) y = e b y , u = u , (5.162) a = 2b. (5.161)

u ( y , 0) = 0,

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 (5.165) eb = , y we get t = t or after rearrangement . This form also allows We thus expect u = u y / t or equivalently u = u y / t u = u y / t , where is any constant. Let us then dene our similarity variable as y (5.168) = . 2 t Here the factor of 1/2 is simply a convenience adopted so that the solution takes on a traditional form. We would nd that any constant in the similarity transformation would induce a self-similar result. 264 y y = . t t (5.167) y y
2

(5.166)

u Let us rewrite the dierential equation, boundary, and initial conditions ( u = y 2 , t u (y , 0) = 0, u (0, t ) = 1, u (, t ) = 0), in terms of the similarity variable . We rst 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 = , t t d 22 d 2t d u du 1 du = = , y y d 2 t d u 1 du 2 u = = , 2 y y y y 2 t d du 1 d 2 u 1 1 1 d 2 u = = = . 2 2 2 t y d 2 t 2 t d 4t d Thus, applying these rules to our governing linear momenta equation, we recover du 1 d 2 u , = 2 2t d 4t d du d 2 u + 2 = 0. 2 d d

(5.169) (5.170) (5.171) (5.172)

(5.173) (5.174)

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, y , = , t 0, = . (5.175) (5.176) (5.177)

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, u () = 0, no slip, initial and far-eld. (5.178) (5.179)

We solve the second order dierential equation by the method of reduction of order, noticing that it is really two rst order equations in disguise: d d du d + 2 du d = 0.
2

(5.180) (5.181) (5.182)

multiplying by the integrating factor e , du du 2 d 2 e + 2 e = 0. d d d 265

d 2 du e = 0, d d 2 du = A, e d du 2 = Ae , d u = B + A Now applying the condition u = 1 at = 0 gives 1 = B+A B = 1. So we have u = 1 + A

0 0 0 0

(5.183) (5.184) (5.185)

es ds.

(5.186)

es ds,
=0

(5.187) (5.188)

es ds.

(5.189)

Now applying the condition u = 0 at , we get 0 = 1+A

0

es ds,

(5.190)

= /2 0 = 1+A , 2 2 A = .

(5.191) (5.192)

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 nite value over the semi-innite domain as the integrand is bounded between zero and one, and decays rapidly to zero as s . Consequently, the velocity prole can be written as
2 2 es ds, u ( ) = 1 0 y 2 2 2 t u (y , t ) = 1 es ds, 0 y . u (y , t ) = erfc 2 t

(5.193) (5.194) (5.195)

In the last form above, we have introduced the so-called error function complement, erfc. Plots for the velocity prole 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 266

y *

= t*

t = * 2 t = * 1
1

3
1

u*

u*

Figure 5.9: Sketch of velocity eld solution for Stokes rst problem in both similarity coordinate and primitive coordinates y , t . = 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 uid. We can say in fact that momentum is diusing into the uid. We dene the momentum diusion length as the length for which signicant momentum has diused into the uid. This is well estimated by taking = 1. In terms of physical variables, we have y = 1, 2 t y = 2 t , y t = 2 ,
U U2

(5.196) (5.197) (5.198) (5.199) (5.200)

y =

2 U 2 t , U y = 2 t.

We can in fact dene this as a boundary layer thickness. That is to say the momentum boundary layer thickness in Stokes rst problem grows at a rate proportional to the square root of momentum diusivity and time. This class of result is a hallmark of all diusion processes, be it mass, momentum, or energy. Taking standard properties of air, we nd 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 eld. For this problem, the shear stress reduces to simply u (5.201) = . y 267

Scaling as before by a characteristic stress c , we get U u , c = y U U 2 1 u = . c y

(5.202) (5.203)

Taking c = U 2 / = U 2 /(/) = U 2 , we get 1 du u = , = y 2 t d 1 2 2 = e , 2 t 1 2 = e , t y 1 exp = t 2 t Now at the wall, y = 0, and we get

(5.204) (5.205) (5.206)

2

(5.207) (5.208)

1 . (5.209) 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 innite at t = 0, and decreases as time increases. This is because the velocity gradient attens 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 nite time span is nite, even when the time span includes t = 0. It can also be shown that the ow corresponds to a pulse of vorticity being introduced at the wall, which subsequently diuses into the uid. In dimensional terms, we can say 1 , (5.210) = 2 2 U U t |y =0 =

U 2 , U t

(5.211)

U = , (5.212) t U = . (5.213) t Now let us consider the heat transfer problem. Recall the governing equation, initial and boundary conditions are u 2T T = k 2 + , t y y T (y, 0) = To , T (0, t) = To , c 268
2

(5.214) T (, t) = To . (5.215)

We will adopt the same time tc an length yc scales as before. Take the dimensionless temperature to be T To . (5.216) T = To So we get cTo T kTo 2 T U 2 = + 2 2 y 2 tc t yc yc u y
2

,
2

(5.217) (5.218) (5.219) (5.220)

u kTo tc 2 T U 2 tc T , = + 2 2 2 t yc cTo y yc cTo y k tc kU 2 1 k k now 2 = = = = P r, 2 2 yc c U c c c U2 T 2 tc U 2 U 2 1 U 2 = = = = Ec. 2 cT cTo yc 2 U 2 cTo cTo o So we have in dimensionless form 1 2 T u T = + Ec 2 t P r y y T (y , 0) = 0, T (0, t ) = 0,
2

, T (, t ) = 0.

(5.221) (5.222)

Notice that the only driving inhomogeneity is the viscous work. Now we know from our solution of the linear momentum equation that 1 y2 u = exp y t 4t So we can rewrite the equation for temperature variation as
2 T 1 2 T y Ec , = exp + 2 t P r y t 2t T (y , 0) = 0, T (0, t ) = 0, T (, t ) = 0.

(5.223)

(5.224) (5.225)

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 diusivity to thermal diusivity, this limit corresponds to materials for which momentum diusivity is much greater than thermal diusivity. 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 . t t 2t 269 (5.226)

and with T = T ( ) and = be

y , 2 t

we get the transformed partial time derivative to (5.227)

T dT = . t 2t d So the governing equation reduces to Ec 2 dT 2 = e , 2t d t dT 2Ec 1 2 2 = e , d 2Ec 1 2s2 e ds T = s

(5.228) (5.229) (5.230)

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 dierential equation in a Taylor series about = 0 and solving gives dT 2Ec 1 3 = 2 + 2 +... , d 2Ec 1 4 2 +... . T = ln + 2 (5.231) (5.232)

We cannot satisfy both boundary conditions; the equation has been solved so as to satisfy the boundary condition in the far eld of T () = 0.

It turns out that solving the inner layer problem and the matching is of about the same diculty 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 2 y T = 0.

(5.233)

The solution which satises the boundary conditions is (5.234)

In this limit, momentum diuses 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 which rapidly diuses throughout the entire ow. The eect of smearing a localized nite thermal energy input over a semi-innite domain has a negligible inuence on the temperature of the global domain. 270

So let us bring back diusion and study solutions for nite Prandtl number. Our governing equation in similarity variables then becomes dT 2t d dT 2 d 2 d T dT + 2P r 2 d d T (0) Ec 2 1 1 d 2 T 2 + e , 2 P r 4t d t 1 d2 T 4Ec 2 2 + = e , 2 P r d 4 2 = EcP r e2 , = 0, T () = 0. = (5.235) (5.236) (5.237) (5.238)

The second order dierential equation is really two rst order dierential equations in dis2 guise. There is an integrating factor of eP r . Multiplying by the integrating factor and operating on the system, we nd eP r
2d

T 2 dT + 2P r eP r 2 d d d 2 dT eP r d d 2 dT eP r d dT d T

4 2 = EcP r e(P r2) , (5.239) 4 2 (5.240) = EcP r e(P r2) , 4 2 = EcP r e(P r2)s ds + C1 , (5.241) 0 4 2 2 2 , e(P r2)s ds + C1 eP r(5.242) = EcP r eP r 0 p 4 2 2 = EcP r eP r p e(P r2)s ds dp 0 0 +C1
0

eP r

s2

ds + C2 .

(5.243)

The boundary condition T (0) = 0 gives us C2 = 0. The boundary condition at gives us then p 4 2 2 2 eP r s ds e(P r2)s ds dp + C1 eP r p (5.244) 0 = EcP r 0 0 0
1 2 Pr

Therefore, we get 4 EcP r

0

eP r

p2 0

e(P r2)s ds dp = eP r
p2 0 p

C1 2
2

, Pr

(5.245) (5.246)

2 C1 = EcP r 3/2

e(P r2)s ds dp.

So nally, we have for the temperature prole

p 4 2 2 T ( ) = EcP r eP r p e(P r2)s ds dp 0 0 p 2 2 2 eP r p e(P r2)s ds dp + EcP r 3/2 0 0

eP r

s2

ds.

(5.247)

271

This analysis simplies considerably in the limit of P r = 1, that is when momentum and energy diuse at the same rate. This is a close to reality for many gases. In this case, the temperature prole becomes 4 T ( ) = Ec Now if h(p) = as
p s2 ds, 0 e 0

ep
2

p 0

es ds dp + C1

es ds.

(5.248)

we get

dh dp

= ep . Using this, we can rewrite the temperature prole

4 dh 2 T ( ) = Ec h(p) dp + C1 es ds, dp 0 0 2 4Ec h 2 d es ds, + C1 = 0 2 0 2 4Ec 1 s2 2 = e ds + C1 es ds, 2 0 0 2Ec 2 2 es ds. es ds + C1 = 0 0

(5.249) (5.250) (5.251) (5.252)

Now for T () = 0, we get 2Ec s2 e ds + C1 0 2Ec 0 = + C1 , 2 2 Ec C1 = . 0 = So the temperature prole can be expressed as Ec T ( ) =
0 0

es ds,

(5.253) (5.254) (5.255)

es ds

2 1

es ds .

(5.256)

We notice that we can write this directly in terms of the velocity as T ( ) = Ec u ( ) (1 u ( )) . 2 (5.257)

This is a consequence of what is known as Reynolds analogy which holds for P r = 1 that the temperature eld can be directly related to the velocity eld. The temperature eld for Stokes rst 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 ow of a viscous uid over a at plate. This problem forms the foundation for a variety of viscous ows over more complicated geometries. It also illustrates some very important features of viscous ow physics, as well as 272

3 Pr = 1 Ec = 1
2

0.12

T *

Figure 5.10: Plot of temperature eld for Stokes rst problem for P r = 1, Ec = 1.

uo

Figure 5.11: Schematic for at plate boundary layer problem.

273

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 ow of viscous uid of constant viscosity and thermal conductivity over a at plate. In the far eld, the uid 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 uids velocity changes from zero at the plate surface to its freestream value. This zone is called the boundary layer. Considering rst the velocity eld, we nd, assuming the ow is steady as well, that the dimensionless two-dimensional Navier-Stokes equations are as follows u v + = 0, x y u p 1 2u 2u u +v = + + , u x y x Re x2 y 2 v p 1 2v 2v v . +v = + + u x y y Re x2 y 2 The dimensionless boundary conditions are u(x, y ) v (x, y ) p(x, y ) u(x, 0) v (x, 0) = = = = = 1, 0, 0, 0, 0. (5.261) (5.262) (5.263) (5.264) (5.265) (5.258) (5.259) (5.260)

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. u= udim , uo v= vdim , uo x= xdim , L y= ydim , L p= pdim po . u2 o (5.266)

Note for our at plate of semi-innite extent, we do not have a natural length scale. This suggests that we may nd a similarity solution which removes the eect of L. Now let us consider that for Re , we have an outer solution of u = 1 to be valid for most of the ow eld suciently far away from the plate surface. In fact the solution u = 1,v = 0, p = 0, satises all of the governing equations and boundary conditions except for the no slip condition at y = 0. Because in the limit as Re , we eectively ignore the high order derivatives found in the viscous terms, we cannot expect to satisfy all boundary 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 eect of the viscous terms, bring back the no-slip condition, and 274

This is the rst 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 diculty, 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. (5.267) 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, (5.268) x 1/ Re y u 1/ Re u p 1 2u 2u + v = + + Re , (5.269) u x 1/ Re y x Re x2 y 2 1 v (1/ Re)(1/ Re) v 1 p u + v = y Re x 1/ Re 1/ Re y 1 2v 1/ Re 2 v 1 . (5.270) + + 2 Re 1/Re y 2 Re x Simplifying, this reduces to u v + = 0, x y u u p 1 2u 2u u +v = + + 2, x y x Re x2 y 2 v p 1 v v 2v +v = Re + u + 2. 2 x y y Re x y (5.271) (5.272) (5.273)

match our inviscid outer solution to a viscous inner solution.

Now in the limit as Re , the rescaled equations reduce to the well known boundary layer equations: u v + = 0, x y u p 2 u u +v = + 2, u x y x y p 0 = . y To match the outer solution, we need the boundary conditions which are u(x, y ) v (x, y ) p(x, y ) u(x, 0) v (x, 0) 275 = = = = = 1, 0, 0, 0, 0. (5.277) (5.278) (5.279) (5.280) (5.281) (5.274) (5.275) (5.276)

The y momentum equation gives us p = p(x). (5.282) In general, we can consider this to be an imposed pressure gradient which is supplied by the dp outer inviscid solution. For general ows, that pressure gradient 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 at plate boundary layer. (5.283) So called Falkner-Skan solutions consider ows over curved plates, for which the outer inviscid solution does not have a constant pressure. This ultimately aects the behavior of the uid in the boundary layer, giving results which dier in important features from our Blasius problem. With our assumptions, the Blasius problem reduces to u v + = 0, x y u 2u u +v = . u x y y 2 The boundary conditions are now only on velocity and are u(x, y ) v (x, y ) u(x, 0) v (x, 0) = = = = 1, 0, 0, 0. (5.286) (5.287) (5.288) (5.289) (5.284) (5.285)

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 dierential equation. So taking u= , v = , (5.290) y x we nd that mass conservation reduces to associated boundary conditions become
2 x y

2 y x

= 0. The x momentum equation and 3 , y 3

2 2 y x y x y 2 (x, y ) y (x, y ) x (x, 0) y (x, 0) x

(5.291) (5.292) (5.293) (5.294) (5.295) (5.296)

= 1, = 0, = 0, = 0,

276

Let us try stretching all the variables of this system to see if there are stretching transformations 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 . (5.297)

Under this transformation, the x-momentum equation and boundary conditions transform to ea+2b2c 2 2 ea+2b2c y x y x y 2 ebc ( x, y ) y x, y ) eac ( x ebc ( x, 0) y eac ( x, 0) x = e3bc = 1, = 0, = 0, = 0, 3 , y 3 (5.298) (5.299) (5.300) (5.301) (5.302) (5.303) If we demand b = c and a = 2c, then the transformation is invariant, yielding 2 2 y x y x y 2 ( x, y ) y ( x, y ) x ( x, 0) y ( x, 0) x Now our transformation is reduced to x = e2c x, y = ec y , = ec . (5.310) = 3 , y 3 (5.304) (5.305) (5.306) (5.307) (5.308) (5.309)

= 1, = 0, = 0, = 0,

Since c does not appear explicitly in either the original equation set nor the transformed equation set, the solution must not depend on this stretching. Eliminating c from the transformation by ec = x /x we nd that y = y x , x 277 = x , x (5.311)

or y y = , x x

= . x x

(5.312)

Thus motivated, let us seek solutions of the form =f x That is taking y = , x we seek = xf ( ). uo L ydim L = L xdim (5.314) y . x (5.313)

(5.315)

Let us check that our similarity variable is independent of L our unknown length scale. y = = x Re y Re ydim /L = = x xdim /L uo ydim . xdim (5.316)

So indeed, our similarity variable is independent of any arbitrary length scale we happen to have chosen. With our similarity transformation, we have 1 3/2 1 = y x = , x 2 2x 1 = . y x
2 2 3

(5.317) (5.318)

, , , , and . Now we need expressions for x y x y y 2 y 3 Now consider the partial derivatives of the stream function . Operating on each partial derivative, we nd

df 1 1 = xf ( ) = x + f, x x d x 2 x 1 1 1 1 df df + f = f = x , 2 x d 2 x 2 x d 1 df so v = f . 2 x d df df 1 df = xf ( ) = x (f ( )) = x = x = , y y y d y d x d df = . so u= y d 2 df d2 f 1 d2 f = = = 2 = 2. x y x y x d d x 2x d 278

(5.319) (5.320) (5.321) (5.322) (5.323) (5.324)

2 df d2 f 1 d2 f = = = = . y 2 y y y d d 2 y x d 2 1 d2 f 2 1 d2 f 1 d3 f 3 = , = = = y 3 y y 2 y x d 2 x y d 2 x d 3 y 1 d3 f = . x d 3 Now we substitute each of these expressions into the x momentum equation and get df 1 d2 f 2 d 2x d df 1 1 d2 f f 2 x d x d 2 df d2 f df d2 f f d d 2 d d 2 d2 f f 2 d 3 df 1 d2 f + f d 3 2 d 2 1 d3 f , x d 3 d3 f = 2 3, d d3 f = 2 3, d = = 0.

(5.325) (5.326) (5.327)

(5.328) (5.329) (5.330) (5.331)

This is a third order non-linear ordinary dierential 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 d For no-slip on v , we require df 1 , v (0) = 0 = f 2 x d df 0 = f (0) 0 , d =0
=0

= 0.
=0

(5.332)

(5.333) (5.334)

f (0) = 0 And to satisfy the freestream condition on u as , we need df d = 1.

(5.335)

(5.336)

The most standard way to solve non-linear ordinary dierential equations of this type is to reduce them to systems of rst order ordinary dierential equations and use some numerical technique, such as a Runge-Kutta 6 integration. We recall that Runge-Kutta techniques,
6 Carle David Tolm` e Runge, 1856-1927, German mathematician and physicist, close friend of Max Planck, studied spectral line elements of non-Hydrogen molecules, held chairs at Hannover and G ottingen, entertained grandchildren at age 70 by doing handstands.

279

as well as most other common techniques, require a well-dened set of initial conditions to predict the nal state. To achieve the desired form, we dene d2 f df , h 2. (5.337) d d Thus the x momentum equation becomes dh 1 + f h = 0. (5.338) d 2 But this is one equation in three unknowns. We need to write our equations as a system of three rst order equations, along with associated initial conditions. They are df = g, f (0) = 0, (5.339) d dg = h, g (0) = 0, (5.340) d 1 dh = f h, h(0) =?. (5.341) d 2 Everything is well-dened except we do not have an initial condition on h. We do however have a far-eld condition on g which is g () = 1. One viable option we have for getting a nal solution is to use a numerical trial and error procedure, guessing h(0) until we nd that g () 1. We will use a slightly more ecient method here, which only requires one guess. 3f To do this, let us rst demonstrate the following lemma: If F ( ) is a solution to d + d 3 g = 0, then aF (a ) is also a solution. The proof is as follows. Take w ( ) = aF (a ). Then we have dF (a ) dw = a2 , (5.342) d d 2 d2 w 3 d F (a ) = a , (5.343) d 2 d 2 3 d3 w 4 d F (a ) = a . (5.344) d 3 d 3 Substituting these expressions into the x momentum equation, we nd d3 F (a ) 1 4 d2 F (a ) + a F ( a ) = 0, (5.345) d 3 2 d 2 d2 F (a ) d3 F (a ) 1 F ( a ) + = 0. (5.346) 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 rst solve the following related system: dF = G, F (0) = 0, (5.347) d dG = H, G(0) = 0, (5.348) d dH 1 = F H, H (0) = 1. (5.349) d 2 a4 280
1 d2 f f 2 d 2

After one numerical integration, we nd that with this guess for H (0) that G() = 2.08540918... (5.350)

Now our numerical solution also gives us F , and so we know that f = aF (a ) is also a solution. Moreover dF (a ) df = a2 , d d g ( ) = a2 G(a ). Now we want g () = 1, so take 1 = a2 G(), so a2 = a= Now d2 f d 2 d2 f d 2
=0

that is

(5.351) (5.352)

1 . G()

So (5.353)

1 G()

d2 F (a ) , d 2 2 3 d F (a ) = a , d 2 =0 = a3 = = = = a3 H (0), a3 (1), a3 = G3/2 (), (2.08540918...)3/2 = 0.332057335...

(5.354) (5.355) (5.356) (5.357) (5.358) (5.359)

h(0) h(0) h(0) h(0)

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 satises the condition at . df A plot of u = 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 dene the boundary layer thickness, 0.99 , as that value of ydim for which the velocity has 0.99 of its freestream value. Recalling that uo ydim = , (5.360) xdim we say that 4.9 = Rearranging, we get 0.99 = 4.9 , xdim uo xdim
1/2 = 4.9Rex . dim

uo 0.99 . xdim

(5.361)

(5.362) (5.363)

281

u*

Figure 5.12: Velocity prole for Blasius boundary layer. Here we have taken a Reynolds number based on local distance to be uo xdim Rexdim = . (5.364) This formula is valid for laminar ows, and has been seen to be valid for Rexdim < 3 106 . For greater lengths, there can be a transition to turbulent ow. For water owing a 1 m/s and a downstream distance of 1 m, we nd 0.99 = 0.5 cm. For air under the same conditions, we nd 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 nd udim w = . (5.365) ydim ydim =0 Consider 2 1 d2 f u = = , y y 2 x d 2
udim uo uo L ydim L

(5.366) (5.367) (5.368)

1
xdim L

d2 f , d 2

udim uo 1 d2 f = uo , ydim xdim d 2 282

 = (0)
1 u2 o 2

Cf Cf

udim uo d2 f = uo , ydim xdim d 2 d2 f (0), = Cf = 2 uo xdim d 2 d2 f = 2(Rexdim )1/2 2 (0), d 0.664... = . Rexdim

(5.369) (5.370) (5.371) (5.372)

We notice that at xdim = 0 that the stress is innite. This seeming problem is seen not to be one when we consider the actual viscous force on a nite length of plate. Consider a plate of length L and width b. Then the viscous force acting on the plate is F = = = b = = = F
1 u2 o Lb 2 L 0 L 0

dA, (xdim , 0)bdxdim ,

L

(5.373) (5.374) (5.375) (5.376) (5.377) (5.378) (5.379)

1 f (0)uo uo dxdim , xdim 0 L dx dim bf (0)uo uo , xdim 0 L bf (0)uo uo (2 xdim )0 , 2bf (0)uo uo L, 1/2 1/2 = 4f (0)ReL = 1.328ReL . CD = 4f (0) uo L

Now let us consider the thermal boundary layer. Here we will take the boundary conditions so that the wall and far eld are held at a constant xed 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: cp Tdim Tdim udim + vdim xdim ydim 2 Tdim 2 Tdim = k + 2 x2 ydim dim udim + 2 xdim y= ydim , L T = Tdim To , To

(5.380)
2

vdim +2 ydim

udim vdim + + ydim xdim

Taking as before, x= xdim , L

u=

udim , uo

v=

vdim . uo

(5.381)

Making these substitutions, we get cp uo To T T u +v L x y = kTo L2 2T 2T + x2 y 2 283 (5.382)

T T +v x y

u2 u + 2o 2 L x = k cp uo L

v +2 y

u v + + y x

2T 2T + x2 y 2
2

(5.383) v +2 y
2

Now we have

uo u 2 + cp LTo x

u v + + y x

( .5.384)

k k 1 1 = = , cp uo L cp uo L P r Re u2 Ec uo o = = . cp LTo uo L cp To Re So the dimensionless energy equation with boundary conditions can be written as u T T +v x y = 1 P rRe

(5.385) (5.386)

2T 2T + x2 y 2
2

(5.387)
2

T (x, 0) = 0,

Ec u 2 + Re x

v +2 y

u v + + y x

T (x, ) = 0.

(5.388)

Now as Re , we see that T = 0 is a solution that satises the energy equation and all boundary conditions. For nite Reynolds number, non-zero velocity gradients generate Re v , and a temperature eld. Once again, we rescale in the boundary layer using v = y = Re y . This gives u T 1 1 1 T v + = x P rRe Re 1/ Re y

2T 2T + Re x2 y 2
2

(5.389)
2

T T 1 +v = x y Pr

Ec u 2 + Re x 2 +Ec Re

v +2 y

Re

1 v u + y Re x

1 2T 2T + Re x2 y 2 u x
2

(5.390) v y
2

2 + Re
2

1 v u + + y Re x

u T T 1 2T +v = + Ec 2 x y P r y y
df , d

as Re ,

(5.391) (5.392)

.
df 1 v = 2 f d and x u y

y Now take T = T ( ) with = as well as u = x We also have for derivatives, that

1 d2 f . x d 2

T x

dT dT = d x d

1 , 2x 284

(5.393)

dT dT 1 T , = = y d y d x 1 dT T 2T = = 2 y y y y x d The energy equation is then rendered as df 1 dT d 2 x d 1 df + f 2 x d 1 dT x d dT d

(5.394) 1 d2 T 1 dT = . = x y d x d 2 (5.395)

1 1 d2 T Ec = + 2 P r x d x

d2 f d 2
2

, (5.396) , (5.397) (5.398) (5.399) (5.400)

1 df 1 df dT f 2 d d 2 d

d2 f 1 d2 T = + Ec P r d 2 d 2 1 d2 T d2 f = + Ec P r d 2 d 2
2

1 dT f 2 d

d2 T 1 dT d2 f + P r f = P rEc , d 2 2 d d 2 T (0) = 0, T () = 0.

Now for Ec 0, we get T = 0 as a solution which satises the governing dierential 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 diusivity equals momentum diusivity, we expect the temperature eld to be directly related to the velocity eld. For P r = 1, the energy equation reduces to d2 f 1 dT d2 T , + f = Ec d 2 2 d d 2 T (0) = 0, T () = 0. Here the integrating factor is e
1 f (t)dt 0 2

(5.401) (5.402) (5.403)

2

Multiplying the energy equation by the integrating factor gives e

1 f (t)dt 0 2

d2 T 1 + fe 2 d 2 d e d

1 f (t)dt 0 2

dT d

= Ec e = Ec e

1 f (t)dt 0 2

d2 f d 2 d2 f d 2

,
2

(5.404) (5.405)

1 f (t)dt 0 2

dT d

1 f (t)dt 0 2

f f = 0, we have Now from the x momentum equation, f + 1 2 f = 2 So we can rewrite the integrating factor as e
1 f (t)dt 0 2

f . f

(5.406)

=e

1 (2)f f 0 2

dt

=e

ln

f ( ) f (0)

f (0) . f ( )

(5.407)

285

So the energy equation can be written as d d f (0) dT f ( ) d f (0) d2 f , f ( ) d 2 d2 f = Ecf (0) 2 , d d2 f = Ecf (0) ds + C1 , 0 ds2 d2 f d2 f d2 f = Ec 2 ds + C , 1 d 0 ds2 d 2 = Ec d2 f df d2 f f (0) , = = Ec 2 + C 1 d d d 2
=0 2

(5.408) (5.409) (5.410) (5.411) (5.412) (5.413)

f (0) dT f ( ) d dT d

d2 f d2 f df + C1 2 , = Ec 2 d d d d 1 = Ec d 2

df d
2

df df + C1 + C2 , d d Ec T (0) = 0 = (f (0))2 + C1 f (0) +C2 , 2 Ec T = 2

=0 =0

+ C1

d2 f , d 2

(5.414) (5.415) (5.416) (5.417) (5.418) (5.419) (5.420) (5.421)

C2 = 0, Ec T () = 0 = (f ())2 + C1 f (), 2
=1 =1

Ec C1 = , 2 Ec df df T ( ) = 1 , 2 d d Ec T ( ) = u( )(1 u( )). 2 A plot of the temperature prole for P r = 1 and Ec = 1 is given in Figure 5.13.

286

 6 5 4 3 2 1

* Pr = 1 Ec = 1

T *

Figure 5.13: Temperature prole for Blasius boundary layer, Ec = 1, P r = 1.

287

288

Bibliography
This bibliography focuses on books which are closely related to the material presented in this course in classical uid mechanics, especially with regards to graduate level treatment of continuum mechanical principles applied to uids, compressible ow, viscous ow, and vortex dynamics. It also has some general works of historic importance. It is by no means a comprehensive survey of works on uid mechanics. Only a few works are given here which focus on such important topics as low Reynolds number ows, turbulence, bio-uids, computational uid dynamics, microuids, molecular dynamics, magneto-hydrodynamics, geo-physical ows, rheology, astro-physical ows, as well as elementary undergraduate texts. That said, those which are listed are among the best that exist and would be useful to examine. D. J. Acheson, Elementary Fluid Dynamics, Oxford Univ. Press, Oxford, 1990. J. Anderson, Modern Compressible Flow with Historical Perspective, second ed., McGrawHill, New York, 1989. R. Aris, Vectors, Tensors, and the Basic Equations of Fluid Mechanics, Dover, New York, 1962. G. I. Barenblatt, Scaling, Self-Similarity, and Intermediate Asymptotics, Cambridge, New York, 1996. G. Batchelor, An Introduction to Fluid Dynamics, Cambridge University Press, Cambridge, 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, Second Edition, John Wiley, New York, 2001. 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, New York, 2002. S. Chandrasekhar, Hydrodynamic and Hydromagnetic Stability, Dover, New York, 1961. 289

S. Chapman and T. G. Cowling, The Mathematical Theory of Non-Uniform Gases, Cambridge University Press, Cambridge, 1939. R. Chevray and J. Mathieu, Topics in Fluid Mechanics, Cambridge, Cambridge, 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. I. G. Currie, Fundamental Mechanics of Fluids, 2nd edition, McGraw-Hill, New York, 1993. S. R. de Groot and P. Mazur, Non-Equilibrium Thermodynamics, Dover, New York, 1984. P. G. Drazin and W. H. Reid, Hydrodynamic Stability, Cambridge, Cambridge, 1981. G. Emanuel, Analytical Fluid Dynamics, CRC Press, Boca Raton, Florida, 1994. R. P. Feynman, Lectures on Physics, Addison-Wesley, New York, 1970. R. W. Fox and A. T. McDonald, Introduction to Fluid Mechanics, 5th edition, John Wiley, New York, 1998. Y. C. Fung, Foundations of Solid Mechanics, Prentice-Hall, Englewood Clis, New Jersey, 1965. M. Gad-el-Hak, Flow Control: Passive, Active and Reactive Flow Management, Cambridge Univ. Press, Cambridge, 2000. S. Goldstein, ed., Modern Developments in Fluid Dynamics, Vols. I and II, Dover, New York, 1965. 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. J. O. Hirschfelder, C. F. Curtiss, and R. B. Bird, Molecular Theory of Gases and Liquids, Wiley, New York, 1954. D. D. Joseph, Fluid Dynamics of Viscoelastic Liquids, Springer, New York, 1990. G. E. Karniadakis and A. Beskok, Microows: Fundamentals and Simulation, Springer Verlag, Berlin, 2001. E. L. Koschmieder, B` enard Cells and Taylor Vortices, Cambridge, Cambridge, 1993. P. K. Kundu and I. M. Cohen, Fluid Mechanics, Second Edition, Academic Press, San Diego, 2001. H. Lamb, Hydrodynamics, 6th edition, Dover, New York, 1993. M. T. Landahl, Unsteady Transonic Flow, Cambridge, Cambridge, 1989. L. D. Landau, and E. M. Lifshitz, Fluid Mechanics, Pergamon Press, Oxford, 1959. 290

G. L. Leal, Laminar Flow and Convective Transport Processes: Scaling Principles and Asymptotic Analysis, Butterworth-Heinemann, 1992. H. W. Liepmann, and A. Roshko, Elements of Gasdynamics, Dover, New York, 2002. V. L. Liseikin, Grid Generation Methods, Springer, 1999. J. Lighthill, Waves in Fluids, Cambridge University Press, Cambridge, 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. R. L. Panton, Incompressible Flow, 2nd edition, John Wiley, New York, 1995. D. Pnueli and C. Gutnger, Fluid Mechanics, Cambridge, Cambridge, 1992. L. Prandtl and O. G. Tietjens, Fundamentals of Hydro and Aeromechanics, Dover, New York, 1957. L. Rosenhead, ed., Laminar Boundary Layers, Oxford, 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. J. A. Schetz, Boundary Layer Analysis, Prentice-Hall, Englewood Clis, New Jersey, 1993. H. M. Schey, Div, Grad, Curl, and All That, 2nd Ed., W.W. Norton, London, 1992. H. Schlichting, K. Gersten, E. Krause, and C. Mayes, 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. H. Spurk, Fluid Mechanics: Problems and Solutions, Springer, New York, 1997. G. G. Stokes, Mathematical and Physical Papers, Johnson Reprint Corporation, New York, 1966. 291

J. W. Strutt (Lord Rayleigh), The Theory of Sound, Vols. 1 and 2, Dover, New York, 1945. G. I. Taylor, Scientic Papers, Cambridge University Press, Cambridge, 1958. J. C. Tannehill, D. A. Anderson, and R. H. Pletcher, Computational Fluid Mechanics and Heat Transfer, 2nd edition, Taylor and Francis, New York, 1997. W. Thomson (Lord Kelvin), Mathematical and Physical Papers, Cambridge Univ. Press, Cambridge, 1911. D. J. Tritton, Physical Fluid Dynamics, Second Edition, Oxford Univ. Press, Oxford, 1988. C. A. Truesdell, A First Course in Rational Continuum Mechanics, Vol. 1, Second Edition, Academic Press, Boston, 1991. M. Van Dyke, Perturbation Methods in Fluid Mechanics, Parabolic Press, Stanford, CA, 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, Academic Press, New York, 1958. 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, 2nd edition, McGraw-Hill, New York, 1991. 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 Hydrodynamic Phenomena, Dover, New York, 2002. M. J. Zucrow and J. D. Homan, Gas Dynamics, Vol. I, John Wiley, New York, 1976.

292