Fluid Mechanics

Fluid Mechanics
Richard Fitzpatrick
Professor of Physics
The University of Texas at Austin
Contents
1 Overview 7
1.1 Intended Audience . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2 Major Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 To Do List . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2 Mathematical Models of Fluid Motion 9
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 What is a Fluid? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.3 Volume and Surface Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.4 General Properties of Stress Tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.5 Stress Tensor in a Static Fluid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.6 Stress Tensor in a Moving Fluid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.7 Viscosity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.8 Conservation Laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.9 Mass Conservation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.10 Convective Time Derivative . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.11 Momentum Conservation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.12 Navier-Stokes Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.13 Energy Conservation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.14 Equations of Incompressible Fluid Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.15 Equations of Compressible Fluid Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.16 Dimensionless Numbers in Incompressible Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.17 Dimensionless Numbers in Compressible Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.18 Fluid Equations in Cartesian Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
2.19 Fluid Equations in Cylindrical Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.20 Fluid Equations in Spherical Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
2.21 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3 Hydrostatics 31
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.2 Hydrostatic Pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.3 Buoyancy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.4 Equilibrium of Floating Bodies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.5 Vertical Stability of Floating Bodies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.6 Angular Stability of Floating Bodies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.7 Determination of Metacentric Height . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.8 Energy of a Floating Body . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.9 Curve of Buoyancy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.10 Rotational Hydrostatics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2 FLUID MECHANICS
3.11 Equilibrium of a Rotating Liquid Body . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
3.12 Maclaurin Spheroids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.13 Jacobi Ellipsoids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.14 Roche Ellipsoids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.15 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4 Surface Tension 61
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.2 Young-Laplace Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.3 Spherical Interfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.4 Capillary Length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.5 Angle of Contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.6 Jurins Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.7 Capillary Curves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.8 Axisymmetric Soap-Bubbles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.9 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5 Incompressible Inviscid Fluid Dynamics 77
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.2 Streamlines, Stream Tubes, and Stream Filaments . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.3 Bernoullis Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.4 Vortex Lines, Vortex Tubes, and Vortex Filaments . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
5.5 Circulation and Vorticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.6 Kelvin Circulation Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
5.7 Irrotational Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.8 Two-Dimensional Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.9 Two-Dimensional Uniform Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5.10 Two-Dimensional Sources and Sinks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
5.11 Two-Dimensional Vortex Filaments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
5.12 Two-Dimensional Irrotational Flow in Cylindrical Coordinates . . . . . . . . . . . . . . . . . . . . 90
5.13 Inviscid Flow Past a Cylindrical Obstacle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
5.14 Inviscid Flow Past a Semi-Innite Wedge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
5.15 Inviscid Flow Over a Semi-Innite Wedge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
5.16 Velocity Potentials and Stream Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
5.17 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
6 2D Potential Flow 101
6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
6.2 Complex Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
6.3 Cauchy-Riemann Relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
6.4 Complex Velocity Potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
6.5 Complex Velocity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
6.6 Method of Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
6.7 Conformal Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
6.8 Complex Line Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
6.9 Theorem of Blasius . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
6.10 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
7 Incompressible Boundary Layers 121
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
7.2 No Slip Condition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
7.3 Boundary Layer Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
CONTENTS 3
7.4 Self-Similar Boundary Layers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
7.5 Boundary Layer on a Flat Plate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
7.6 Wake Downstream of a Flat Plate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
7.7 Von K arm an Momentum Integral . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
7.8 Boundary Layer Separation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
7.9 Criterion for Boundary Layer Separation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
7.10 Approximate Solutions of Boundary Layer Equations . . . . . . . . . . . . . . . . . . . . . . . . . 142
7.11 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
8 Incompressible Aerodynamics 149
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
8.2 Theorem of Kutta and Zhukovskii . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
8.3 Cylindrical Airfoils . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
8.4 Zhukovskiis Hypothesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
8.5 Vortex Sheets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
8.6 Induced Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
8.7 Three-Dimensional Airfoils . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
8.8 Aerodynamic Forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
8.9 Ellipsoidal Airfoils . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
8.10 Simple Flight Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
8.11 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
9 Incompressible Viscous Flow 171
9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
9.2 Flow Between Parallel Plates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
9.3 Flow Down an Inclined Plane . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
9.4 Poiseuille Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
9.5 Taylor-Couette Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174
9.6 Flow in Slowly-Varying Channels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
9.7 Lubrication Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
9.8 Stokes Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
9.9 Axisymmetric Stokes Flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
9.10 Axisymmetric Stokes Flow Around a Solid Sphere . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
9.11 Axisymmetric Stokes Flow In and Around a Fluid Sphere . . . . . . . . . . . . . . . . . . . . . . . 185
9.12 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 188
10 Waves in Incompressible Fluids 191
10.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
10.2 Gravity Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
10.3 Gravity Waves in Deep Water . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
10.4 Gravity Waves in Shallow Water . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
10.5 Energy of Gravity Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
10.6 Wave Drag on Ships . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 196
10.7 Ship Wakes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
10.8 Gravity Waves in a Flowing Fluid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
10.9 Gravity Waves at an Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
10.10 Steady Flow over a Corrugated Bottom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
10.11 Surface Tension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
10.12 Capillary Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 206
10.13 Capillary Waves at an Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
10.14 Wind Driven Waves in Deep Water . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
10.15 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
4 FLUID MECHANICS
11 Equilibrium of Compressible Fluids 211
11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
11.2 Isothermal Atmosphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
11.3 Adiabatic Atmosphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
11.4 Atmospheric Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
11.5 Eddington Solar Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
11.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 219
A Vectors and Vector Fields 223
A.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
A.2 Scalars and Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
A.3 Vector Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
A.4 Cartesian Components of a Vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
A.5 Coordinate Transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
A.6 Scalar Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
A.7 Vector Area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
A.8 Vector Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
A.9 Rotation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 231
A.10 Scalar Triple Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
A.11 Vector Triple Product . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
A.12 Vector Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
A.13 Line Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
A.14 Vector Line Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
A.15 Surface Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
A.16 Vector Surface Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
A.17 Volume Integrals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
A.18 Gradient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
A.19 Grad Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
A.20 Divergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
A.21 Laplacian Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
A.22 Curl . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
A.23 Useful Vector Identities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
A.24 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
B Cartesian Tensors 253
B.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
B.2 Tensors and Tensor Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
B.3 Tensor Transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
B.4 Tensor Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 257
B.5 Isotropic Tensors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
B.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
C Non-Cartesian Coordinates 265
C.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
C.2 Orthogonal Curvilinear Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
C.3 Cylindrical Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
C.4 Spherical Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
C.5 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
D Calculus of Variations 273
D.1 Euler-Lagrange Equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
D.2 Conditional Variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
CONTENTS 5
D.3 Multi-Function Variation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 276
D.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
E Ellipsoidal Potential Theory 279
6 FLUID MECHANICS
Overview 7
1 Overview
1.1 Intended Audience
This book presents a single semester course on uid mechanics that is intended primarily for advanced undergraduate
students majoring in physics. A thorough understanding of physics at the lower-division level, including a basic
working knowledge of the laws of mechanics, is assumed. It is also taken for granted that students are familiar
with the fundamentals of multi-variate integral and dierential calculus, complex analysis, and ordinary dierential
equations. On the other hand, vector analysis plays such a central role in the study of uid mechanics that a brief,
but fairly comprehensive, review of this subject area is provided in Appendix A. Likewise, those aspects of cartesian
tensor theory, orthogonal curvilinear coordinate systems, and the calculus of variations, that are required in the study
of uid mechanics are outlined in Appendices B, C, and D, respectively.
1.2 Major Sources
The material appearing in Appendix A is largely based on the authors recollections of a vector analysis course given
by Dr. Stephen Gull at the University of Cambridge. Major sources for the material appearing in other chapters and
appendices include:
Statics, Including Hydrostatics and the Elements of the Theory of Elasticity H. Lamb, 3rd Edition (Cambridge Uni-
versity Press, Cambridge UK, 1928).
Hydrodynamics H. Lamb, 6th Edition (Dover, New York NY, 1945).
Theoretical Aerodynamics L.M. Milne-Thomson, 4th Edition, Revised and enlarged (Dover, New York NY, 1958).
Ellipsoidal Figures of Equilibrium S. Chandrasekhar (Yale University Press, New Haven CT, 1969).
Boundary Layer Theory H. Schlichting, 7th Edition (McGraw-Hill, New York NY, 1970).
Mathematical Methods for the Physical Sciences K.F. Riley (Cambridge University Press, Cambridge UK, 1974).
Fluid Mechanics L.D. Landau, and E.M. Lifshitz, 2nd Edition (Butterworth-Heinemann, Oxford UK, 1987).
Physical Fluid Dynamics D.J. Tritton, 2nd Edition (Oxford University Press, Oxford UK, 1988).
Fluid Dynamics for Physicists T.E. Faber, 1st Edition (Cambridge University Press, Cambridge UK, 1995).
Schaums Outline of Fluid Dynamics W. Hughes, and J. Brighton, 3rd Edition (McGraw-Hill, New York NY, 1999).
An Introduction to Fluid Dynamics G.K. Batchelor (Cambridge University Press, Cambridge UK, 2000).
Theoretical Hydrodynamics L.M. Milne-Thomson, 5th Edition (Dover, New York NY, 2011).
1.3 To Do List
1. Add chapter on vortex dynamics.
2. Add chapter on 3D potential ow.
3. Add appendix on group velocity and Fourier transforms.
4. Add chapter on incompressible ow in rotating systems.
5. Add chapter on instabilities.
8 FLUID MECHANICS
6. Add chapter on turbulence.
7. Add chapter on 1D compressible ow.
8. Add chapter on sound waves.
9. Add chapter on compressible boundary layers.
10. Add chapter on supersonic aerodynamics.
11. Add chapter on convection.
Mathematical Models of Fluid Motion 9
2 Mathematical Models of Fluid Motion
2.1 Introduction
In this chapter, we set forth the mathematical models commonly used to describe the equilibrium and dynamics of
uids. Unless stated otherwise, all of the analysis is performed using a standard right-handed Cartesian coordinate
system: x
1
, x
2
, x
3
. Moreover, the Einstein summation convention is employed (so repeated roman subscripts are
assumed to be summed from 1 to 3see Appendix B).
2.2 What is a Fluid?
By denition, a solid material is rigid. Now, although a rigid material tends to shatter when subjected to very large
stresses, it can withstand a moderate shear stress (i.e., a stress that tends to deform the material by changing its shape,
without necessarily changing its volume) for an indenite period. To be more exact, when a shear stress is rst applied
to a rigid material it deforms slightly, but then springs back to its original shape when the stress is relieved.
A plastic material, such as clay, also possess some degree of rigidity. However, the critical shear stress at which it
yields is relatively small, and once this stress is exceeded the material deforms continuously and irreversibly, and does
not recover its original shape when the stress is relieved.
By denition, a uid material possesses no rigidity at all. In other words, a small uid element is unable to
withstand any tendency of an applied shear stress to change its shape. Incidentally, this does not preclude the possibility
that such an element may oer resistance to shear stress. However, any resistance must be incapable of preventing
the change in shape from eventually occurring, which implies that the force of resistance vanishes with the rate of
deformation. An obvious corollary is that the shear stress must be zero everywhere inside a uid that is in mechanical
equilibrium.
Fluids are conventionally classied as either liquids or gases. The most important dierence between these two
types of uid lies in their relative compressibility: i.e., gases can be compressed much more easily than liquids. Con-
sequently, any motion that involves signicant pressure variations is generally accompanied by much larger changes
in mass density in the case of a gas than in the case of a liquid.
Of course, a macroscopic uid ultimately consists of a huge number of individual molecules. However, most
practical applications of uid mechanics are concerned with behavior on length-scales that are far larger than the
typical intermolecular spacing. Under these circumstances, it is reasonable to suppose that the bulk properties of a
given uid are the same as if it were completely continuous in structure. A corollary of this assumption is that when,
in the following, we talk about innitesimal volume elements, we really mean elements which are suciently small
that the bulk uid properties (such as mass density, pressure, and velocity) are approximately constant across them,
but are still suciently large that they contain a very great number of molecules (which implies that we can safely
neglect any statistical variations in the bulk properties). The continuum hypothesis also requires innitesimal volume
elements to be much larger than the molecular mean-free-path between collisions.
In addition to the continuum hypothesis, our study of uid mechanics is premised on three major assumptions:
1. Fluids are isotropic media: i.e., there is no preferred direction in a uid.
2. Fluids are Newtonian: i.e., there is a linear relationship between the local shear stress and the local rate of strain,
as rst postulated by Newton. It is also assumed that there is a linear relationship between the local heat ux
density and the local temperature gradient.
3. Fluids are classical: i.e., the macroscopic motion of ordinary uids is well-described by Newtonian dynamics,
and both quantum and relativistic eects can be safely ignored.
It should be noted that the above assumptions are not valid for all uid types (e.g., certain liquid polymers, which
are non-isotropic; thixotropic uids, such as jelly or paint, which are non-Newtonian; and quantum uids, such as
liquid helium, which exhibit non-classical eects on macroscopic length-scales). However, most practical applications
10 FLUID MECHANICS
of uid mechanics involve the equilibrium and motion of bodies of water or air, extending over macroscopic length-
scales, and situated relatively close to the Earths surface. Such bodies are very well-described as isotropic, Newtonian,
classical uids.
2.3 Volume and Surface Forces
Generally speaking, uids are acted upon by two distinct types of force. The rst type is long-range in nature
i.e., such that it decreases relatively slowly with increasing distance between interacting elementsand is capable
of completely penetrating into the interior of a uid. Gravity is an obvious example of a long-range force. One
consequence of the relatively slow variation of long-range forces with position is that they act equally on all of the
uid contained within a suciently small volume element. In this situation, the net force acting on the element
becomes directly proportional to its volume. For this reason, long-range forces are often called volume forces. In the
following, we shall write the total volume force acting at time t on the uid contained within a small volume element
of magnitude dV, centered on a xed point whose position vector is r, as
F(r, t) dV. (2.1)
The second type of force is short-range in nature, and is most conveniently modeled as momentumtransport within
the uid. Such transport is generally due to a combination of the mutual forces exerted by contiguous molecules, and
momentum uxes caused by relative molecular motion. Suppose that
x
(r, t) is the net ux density of x-directed uid
momentumdue to short-range forces at position r and time t. In other words, suppose that, at position r and time t, as a
direct consequence of short-range forces, x-momentum is owing at the rate of
x
newton-seconds per meter squared
per second in the direction of vector
x
. Consider an innitesimal plane surface element, dS = ndS , located at point
r. Here, dS is the area of the element, and n its unit normal. (See Section A.7.) The uid which lies on that side of the
element toward which n points is said to lie on its positive side, and vice versa. The net ux of x-momentumacross the
element (in the direction of n) is
x
dS newtons, which implies (from Newtons second law of motion) that the uid
on the positive side of the surface element experiences a force
x
dS in the x-direction due to short-range interaction
with the uid on the negative side. According to Newtons third law of motion, the uid on the negative side of the
surface experiences a force
x
dS in the x-direction due to interaction with the uid on the positive side. Short-range
forces are often called surface forces because they are directly proportional to the area of the surface element across
which they act. Let
y
(r, t) and
z
(r, t) be the net ux density of y- and z- momentum, respectively, at position r and
time t. By a straightforward extension of above argument, the net surface force exerted by the uid on the positive side
of some planar surface element, dS, on the uid on its negative side is
f = (
x
dS,
y
dS,
z
dS). (2.2)
In tensor notation (see Appendix B), the above equation can be written
f
i
=
i j
dS
j
, (2.3)
where
11
= (
x
)
x
,
12
= (
x
)
y
,
21
= (
y
)
x
, etc. (Note that, since the subscript j is repeated, it is assumed to be
summed from 1 to 3. Hence,
i j
dS
j
is shorthand for
_
j=1,3
i j
dS
j
.) Here, the
i j
(r, t) are termed the local stresses
in the uid at position r and time t, and have units of force per unit area. Moreover, the
i j
are the components of a
second-order tensor (see Appendix B), known as the stress tensor. [This follows because the f
i
are the components
of a rst-order tensor (since all forces are proper vectors), and the dS
i
are the components of an arbitrary rst-order
tensor (since surface elements are also proper vectorssee Section A.7and (2.3] holds for surface elements whose
normals point in any direction), so application of the quotient rule (see Section B.3) to Equation (2.3) reveals that the
i j
transform under rotation of the coordinate axes as the components of a second-order tensor.] We can interpret
i j
(r, t) as the i-component of the force per unit area exerted, at position r and time t, across a plane surface element
normal to the j-direction. The three diagonal components of
i j
are termed normal stresses, since each of them gives
the normal component of the force per unit area acting across a plane surface element parallel to one of the Cartesian
coordinate planes. The six non-diagonal components are termed shear stresses, since they drive shearing motion in
which parallel layers of uid slide relative to one another.
2.4 General Properties of Stress Tensor
The i-component of the total force acting on a uid element consisting of a xed volume V enclosed by a surface S is
written
f
i
=
_
V
F
i
dV +
_
S
i j
dS
j
, (2.4)
where the rst term on the right-hand side is the integrated volume force acting throughout V, whereas the second
term is the net surface force acting across S . Making use of the tensor divergence theorem(see Section B.4), the above
expression becomes
f
i
=
_
V
F
i
dV +
_
V
i j
x
j
dV. (2.5)
In the limit V 0, it is reasonable to suppose that the F
i
and
i j
/x
j
are approximately constant across the element.
In this situation, both contributions on the right-hand side of the above equation scale as V. Now, according to
Newtonian dynamics, the i-component of the net force acting on the element is equal to the i-component of the rate
of change of its linear momentum. However, in the limit V 0, the linear acceleration and mass density of the uid
are both approximately constant across the element. In this case, the rate of change of the elements linear momentum
also scales as V. In other words, the net volume force, surface force, and rate of change of linear momentum of an
innitesimal uid element all scale as the volume of the element, and consequently remain approximately the same
order of magnitude as the volume shrinks to zero. We conclude that the linear equation of motion of an innitesimal
uid element places no particular restrictions on the stress tensor.
The i-component of the total torque, taken about the origin O of the coordinate system, acting on a uid element
that consists of a xed volume V enclosed by a surface S is written [see Equations (A.46) and (B.6)]
i
=
_
V
i jk
x
j
F
k
dV +
_
S
i jk
x
j

kl
dS
l
, (2.6)
where the rst and second terms on the right-hand side are due to volume and surface forces, respectively. [Here,
i jk
is the third-order permutation tensor. See Equation (B.7).] Making use of the tensor divergence theorem (see
Section B.4), the above expression becomes
i
=
_
V
i jk
x
j
F
k
dV +
_
V
i jk
(x
j
kl
)
x
l
dV, (2.7)
which reduces to
i
=
_
V
i jk
x
j
F
k
dV +
_
V
i jk

k j
dV +
_
V
i jk
x
j
kl
x
l
dV, (2.8)
since x
i
/x
j
=
i j
. [Here,
i j
is the second-order identity tensor. See Equation (B.9).] Assuming that point O lies
within the uid element, and taking the limit V 0 in which the F
i
,
i j
, and
i j
/x
j
are all approximately constant
across the element, we deduce that the rst, second, and third terms on the right-hand side of the above equation scale
as V
4/3
, V, and V
4/3
, respectively (since x V
1/3
). Now, according to Newtonian dynamics, the i-component of the
total torque acting on the uid element is equal to the i-component of the rate of change of its net angular momentum
about O. Assuming that the linear acceleration of the uid is approximately constant across the element, we deduce
that the rate of change of its angular momentum scales as V
4/3
(since the net linear acceleration scales as V, so the
net rate of change of angular momentum scales as x V, and x V
1/3
). Hence, it is clear that the rotational equation of
motion of a uid element, surrounding a general point O, becomes completely dominated by the second term on the
right-hand side of (2.8) in the limit that the volume of the element approaches zero (since this term is a factor V
1/3
larger than the other terms). It follows that the second term must be identically zero (otherwise an innitesimal uid
element would acquire an absurdly large angular velocity). This is only possible, for all choices of the position of
point O, and the shape of the element, if
i jk

k j
= 0 (2.9)
throughout the uid. The above relation shows that the stress tensor must be symmetric: i.e.,
ji
=
i j
. (2.10)
12 FLUID MECHANICS
It immediately follows that the stress tensor only has six independent components (i.e.,
11
,
22
,
33
,
12
,
13
, and
23
).
Now, it is always possible to choose the orientation of a set of Cartesian axes in such a manner that the non-
diagonal components of a given symmetric second-order tensor eld are all set to zero at a given point in space. (See
Exercise B.6.) With reference to such principal axes, the diagonal components of the stress tensor
i j
become so-
called principal stresses
11
,
22
,
33
, say. Of course, in general, the orientation of the principal axes varies with
position. The normal stress
11
acting across a surface element perpendicular to the rst principal axis corresponds
to a tension (or a compression if
11
is negative) in the direction of that axis. Likewise, for
22
and
33
. Thus, the
general state of the uid, at a particular point in space, can be regarded as a superposition of tensions, or compressions,
in three orthogonal directions.
The trace of the stress tensor,
ii
=
11
+
22
+
33
, is a scalar, and, therefore, independent of the orientation of
the coordinate axes. (See Appendix B.) Thus, it follows that, irrespective of the orientation of the principal axes, the
trace of the stress tensor at a given point is always equal to the sum of the principal stresses: i.e.,
ii
=
11
+
22
+
33
. (2.11)
2.5 Stress Tensor in a Static Fluid
Consider the surface forces exerted on some innitesimal cubic volume element of a static uid. Suppose that the
components of the stress tensor are approximately constant across the element. Suppose, further, that the sides of the
cube are aligned parallel to the principal axes of the local stress tensor. This tensor, which now has zero non-diagonal
components, can be regarded as the sum of two tensors: i.e.,
_
_
1
3
ii
0 0
0
1
3
ii
0
0 0
1
3
ii
_
_
, (2.12)
and
_
11

1
3
ii
0 0
0
22

1
3
ii
0
0 0
33

1
3
ii
_
_
. (2.13)
The rst of the above tensors is isotropic (see Section B.5), and corresponds to the same normal force per unit
area acting inward (since the sign of
ii
/3 is invariably negative) on each face of the volume element. This uniform
compression acts to change the elements volume, but not its shape, and can easily be withstood by the uid within the
element.
The second of the above tensors represents the departure of the stress tensor from an isotropic form. The diagonal
components of this tensor have zero sum, in view of (2.11), and thus represent equal and opposite forces per unit
area, acting on opposing faces of the volume element, which are such that the forces on at least one pair of opposing
faces constitute a tension, and the forces on at least one pair constitute a compression. Such forces necessarily tend to
change the shape of the volume element, either elongating or compressing it along one of its symmetry axes. Moreover,
this tendency cannot be oset by any volume force acting on the element, since such forces become arbitrarily small
compared to surface forces in the limit that the elements volume tends to zero (because the ratio of the net volume
force to the net surface force scales as the volume to the surface area of the element, which tends to zero in the limit
that the volume tends to zerosee Section 2.4). Now, we have previously dened a uid as a material that is incapable
of withstanding any tendency of applied forces to change its shape. (See Section 2.2.) It follows that if the diagonal
components of the tensor (2.13) are non-zero anywhere inside the uid then it is impossible for the uid at that point to
be at rest. Hence, we conclude that the principal stresses,
11
,
22
, and
33
, must be equal to one another at all points
in a static uid. This implies that the stress tensor takes the isotropic form (2.12) everywhere in a stationary uid.
Furthermore, this is true irrespective of the orientation of the coordinate axes, since the components of an isotropic
tensor are rotationally invariant. (See Section B.5.)
Fluids at rest are generally in a state of compression, so it is convenient to write the stress tensor of a static uid in
the form
i j
= p
i j
, (2.14)
where p =
ii
/3 is termed the static uid pressure, and is generally a function of r and t. It follows that, in a stationary
uid, the force per unit area exerted across a plane surface element with unit normal n is p n. [See Equation (2.3).]
Moreover, this normal force has the same value for all possible orientations of n. This well-known resultnamely,
that the pressure is the same in all directions at a given point in a static uidis known as Pascals law, and is a direct
consequence of the fact that a uid element cannot withstand shear stresses, or, alternatively, any tendency of applied
forces to change its shape.
2.6 Stress Tensor in a Moving Fluid
We have seen that in a static uid the stress tensor takes the form
i j
= p
i j
, (2.15)
where p =
ii
/3 is the static pressure: i.e., minus the normal stress acting in any direction. Now, the normal stress at
a given point in a moving uid generally varies with direction: i.e., the principal stresses are not equal to one another.
However, we can still dene the mean principal stress as (
11
+
22
+
33
)/3 =
ii
/3. Moreover, given that the principal
stresses are actually normal stresses (in a coordinate frame aligned with the principal axes), we can also regard
ii
/3
as the mean normal stress. It is convenient to dene pressure in a moving uid as minus the mean normal stress: i.e.,
p =
1
3

ii
. (2.16)
Thus, we can write the stress tensor in a moving uid as the sum of an isotropic part, p
i j
, which has the same form
as the stress tensor in a static uid, and a remaining non-isotropic part, d
i j
, which includes any shear stresses, and also
has diagonal components whose sum is zero. In other words,
i j
= p
i j
+ d
i j
, (2.17)
where
d
ii
= 0. (2.18)
Moreover, since
i j
and
i j
are both symmetric tensors, it follows that d
i j
is also symmetric: i.e.,
d
ji
= d
i j
. (2.19)
It is clear that the so-called deviatoric stress tensor, d
i j
, is a consequence of uid motion, since it is zero in a static
uid. Suppose, however, that we were to view a static uid both in its rest frame and in a frame of reference moving
at some constant velocity relative to the rest frame. Now, we would expect the force distribution within the uid to
be the same in both frames of reference, since the uid does not accelerate in either. However, in the rst frame,
the uid appears stationary and the deviatoric stress tensor is therefore zero, whilst in the second it has a spatially
uniform velocity eld and the deviatoric stress tensor is also zero (because it is the same as in the rest frame). We,
thus, conclude that the deviatoric stress tensor is zero both in a stationary uid and in a moving uid possessing no
spatial velocity gradients. This suggests that the deviatoric stress tensor is driven by velocity gradients within the uid.
Moreover, the tensor must vanish as these gradients vanish.
Let the v
i
(r, t) be the Cartesian components of the uid velocity at point r and time t. The various velocity
gradients within the uid then take the form v
i
/x
j
. The simplest possible assumption, which is consistent with the
above discussion, is that the components of the deviatoric stress tensor are linear functions of these velocity gradients:
i.e.,
d
i j
= A
i jkl
v
k
x
l
. (2.20)
14 FLUID MECHANICS
Here, A
i jkl
is a fourth-order tensor (this follows from the quotient rule because d
i j
and v
i
/x
j
are both proper second-
order tensors). Any uid in which the deviatoric stress tensor takes the above form is termed a Newtonian uid, since
Newton was the rst to postulate a linear relationship between shear stresses and velocity gradients.
Now, in an isotropic uidthat is, a uid in which there is no preferred directionwe would expect the fourth-
order tensor A
i jkl
to be isotropicthat is, to have a form in which all physical distinction between dierent directions
is absent. As demonstrated in Section B.5, the most general expression for an isotropic fourth-order tensor is
A
i jkl
=
i j

kl
+
ik

jl
+
il

jk
, (2.21)
where , , and are arbitrary scalars (which can be functions of position and time). Thus, it follows from (2.20) and
(2.21) that
d
i j
=
v
k
x
k
i j
+
v
i
x
j
+
v
j
x
i
. (2.22)
However, according to Equation (2.19), d
i j
is a symmetric tensor, which implies that = , and
d
i j
= e
kk

i j
+ 2 e
i j
, (2.23)
where
e
i j
=
1
2
_
v
i
x
j
+
v
j
x
i
_
(2.24)
is called the rate of strain tensor. Finally, according to Equation (2.18), d
i j
is a traceless tensor, which yields 3 =
2 , and
d
i j
= 2
_
e
i j

1
3
e
kk

i j
_
, (2.25)
where = . We, thus, conclude that the most general expression for the stress tensor in an isotropic Newtonian uid
is
i j
= p
i j
+ 2
_
e
i j

1
3
e
kk

i j
_
, (2.26)
where p(r, t) and (r, t) are arbitrary scalars.
2.7 Viscosity
The signicance of the parameter , appearing in the previous expression for the stress tensor, can be seen from the
form taken by the relation (2.25) in the special case of simple shearing motion. With v
1
/x
2
as the only non-zero
velocity derivative, all of the components of d
i j
are zero apart from the shear stresses
d
12
= d
21
=
v
1
x
2
. (2.27)
Thus, is the constant of proportionality between the rate of shear and the tangential force per unit area when parallel
plane layers of uid slide over one another. This constant of proportionality is generally referred to as viscosity. It is a
matter of experience that the force between layers of uid undergoing relative sliding motion always tends to oppose
the motion, which implies that > 0.
The viscosities of dry air and pure water at 20
C and atmospheric pressure are about 1.8 10

5
kg/(ms) and
1.010
3
kg/(ms), respectively. In neither case does the viscosity exhibit much variation with pressure. However, the
viscosity of air increases by about 0.3 percent, and that of water decreases by about 3 percent, per degree Centigrade
rise in temperature.
2.8 Conservation Laws
Suppose that (r, t) is the density of some bulk uid property (e.g., mass, momentum, energy) at position r and time t.
In other words, suppose that, at time t, an innitesimal uid element of volume dV, located at position r, contains an
amount (r, t) dV of the property in question. Note, incidentally, that can be either a scalar, a component of a vector,
or even a component of a tensor. The total amount of the property contained within some xed volume V is
=
_
V
dV, (2.28)
where the integral is taken over all elements of V. Let dS be an outward directed element of the bounding surface of
V. Suppose that this element is located at point r. The volume of uid that ows per second across the element, and so
out of V, is v(r, t)dS. Thus, the amount of the uid property under consideration that is convected across the element
per second is (r, t) v(r, t) dS. It follows that the net amount of the property that is convected out of volume V by uid
ow across its bounding surface S is
=
_
S
v dS, (2.29)
where the integral is taken over all outward directed elements of S . Suppose, nally, that the property in question is
created within the volume V at the rate S
per second. The conservation equation for the uid property takes the form
d
dt
= S
. (2.30)
In other words, the rate of increase in the amount of the property contained within V is the dierence between the
creation rate of the property inside V, and the rate at which the property is convected out of V by uid ow. The above
conservation law can also be written
d
dt
+
= S
. (2.31)
Here,
is termed the ux of the property out of V, whereas S
is called the net generation rate of the property within

V.
2.9 Mass Conservation
Let (r, t) and v(r, t) be the mass density and velocity of a given uid at point r and time t. Consider a xed volume
V, surrounded by a surface S . The net mass contained within V is
M =
_
V
dV, (2.32)
where dV is an element of V. Furthermore, the mass ux across S , and out of V, is [see Equation (2.29)]
M
=
_
S
v dS, (2.33)
where dS is an outward directed element of S . Mass conservation requires that the rate of increase of the mass
contained within V, plus the net mass ux out of V, should equal zero: i.e.,
dM
dt
+
M
= 0 (2.34)
[cf., Equation (2.31)]. Here, we are assuming that there is no mass generation (or destruction) within V (since individ-
ual molecules are eectively indestructible). It follows that
_
V
t
dV +
_
S
v dS = 0, (2.35)
16 FLUID MECHANICS
since V is non-time-varying. Making use of the divergence theorem (see Section A.20), the above equation becomes
_
V
_
t
+ ( v)
_
dV = 0. (2.36)
However, this result is true irrespective of the size, shape, or location of volume V, which is only possible if
t
+ ( v) = 0 (2.37)
throughout the uid. The above expression is known as the equation of uid continuity, and is a direct consequence of
mass conservation.
2.10 Convective Time Derivative
The quantity (r, t)/t, appearing in Equation (2.37), represents the time derivative of the uid mass density at the
xed point r. Suppose that v(r, t) is the instantaneous uid velocity at the same point. It follows that the time derivative
of the density, as seen in a frame of reference which is instantaneously co-moving with the uid at point r, is
lim
t0
(r + v t, t + t) (r, t)
t
=
t
+ v =
D
Dt
, (2.38)
where we have Taylor expanded (r + v t, t + t) up to rst order in t, and where
D
Dt
=
t
+ v =
t
+ v
i
x
i
. (2.39)
Clearly, the so-called convective time derivative, D/Dt, represents the time derivative seen in the local rest frame of
the uid.
The continuity equation (2.37) can be rewritten in the form
1
D
Dt
=
Dln
Dt
= v, (2.40)
since ( v) = v + v [see (A.174)]. Consider a volume element V that is co-moving with the uid. In general,
as the element is convected by the uid its volume changes. In fact, it is easily seen that
DV
Dt
=
_
S
v dS =
_
S
v
i
dS
i
=
_
V
v
i
x
i
dV =
_
V
v dV, (2.41)
where S is the bounding surface of the element, and use has been made of the divergence theorem. In the limit that
V 0, and v is approximately constant across the element, we obtain
1
V
DV
Dt
=
Dln V
Dt
= v. (2.42)
Hence, we conclude that the divergence of the uid velocity at a given point in space species the fractional rate of
increase in the volume of an innitesimal co-moving uid element at that point.
2.11 Momentum Conservation
Consider a xed volume V surrounded by a surface S . The i-component of the total linear momentumcontained within
V is
P
i
=
_
V
v
i
dV. (2.43)
Moreover, the ux of i-momentum across S , and out of V, is [see Equation (2.29)]
i
=
_
S
v
i
v
j
dS
j
. (2.44)
Finally, the i-component of the net force acting on the uid within V is
f
i
=
_
V
F
i
dV +
_
S
i j
dS
j
, (2.45)
where the rst and second terms on the right-hand side are the contributions from volume and surface forces, respec-
tively.
Momentum conservation requires that the rate of increase of the net i-momentum of the uid contained within
V, plus the ux of i-momentum out of V, is equal to the rate of i-momentum generation within V. Of course, from
Newtons second law of motion, the latter quantity is equal to the i-component of the net force acting on the uid
contained within V. Thus, we obtain [cf., Equation (2.31)]
dP
i
dt
+
i
= f
i
, (2.46)
which can be written
_
V
( v
i
)
t
dV +
_
S
v
i
v
j
dS
j
=
_
V
F
i
dV +
_
S
i j
dS
j
, (2.47)
since the volume V is non-time-varying. Making use of the tensor divergence theorem, this becomes
_
V
_
( v
i
)
t
+
( v
i
v
j
)
x
j
_
dV =
_
V
_
F
i
+
i j
x
j
_
dV. (2.48)
However, the above result is valid irrespective of the size, shape, or location of volume V, which is only possible if
( v
i
)
t
+
( v
i
v
j
)
x
j
= F
i
+
i j
x
j
(2.49)
everywhere inside the uid. Expanding the derivatives, and rearranging, we obtain
_
t
+ v
j
x
j
+
v
j
x
j
_
v
i
+
_
v
i
t
+ v
j
v
i
x
j
_
= F
i
+
i j
x
j
. (2.50)
Now, in tensor notation, the continuity equation (2.37) is written
t
+ v
j
x
j
+
v
j
x
j
= 0. (2.51)
So, combining Equations (2.50) and (2.51), we obtain the following uid equation of motion,
_
v
i
t
+ v
j
v
i
x
j
_
= F
i
+
i j
x
j
. (2.52)
An alternative form of this equation is
Dv
i
Dt
=
F
i
+
1
i j
x
j
. (2.53)
The above equation describes howthe net volume and surface forces per unit mass acting on a co-moving uid element
determine its acceleration.
18 FLUID MECHANICS
2.12 Navier-Stokes Equation
Equations (2.24), (2.26), and (2.53) can be combined to give the equation of motion of an isotropic, Newtonian,
classical uid: i.e.,
Dv
i
Dt
= F
i

p
x
i
+
x
j
_
_
v
i
x
j
+
v
j
x
i
__

x
i
_
2
3

v
j
x
j
_
. (2.54)
This equation is generally known as the Navier-Stokes equation. Now, in situations in which there are no strong
temperature gradients in the uid, it is a good approximation to treat viscosity as a spatially uniformquantity, in which
case the Navier-Stokes equation simplies somewhat to give
Dv
i
Dt
= F
i

p
x
i
+
_

2
v
i
x
j
x
j
+
1
3
2
v
j
x
i
x
j
_
. (2.55)
When expressed in vector form, the above expression becomes
Dv
Dt
=
_
v
t
+ (v) v
_
= F p +
_
2
v +
1
3
(v)
_
, (2.56)
where use has been made of Equation (2.39). Here,
[(a)b]
i
= a
j
b
i
x
j
, (2.57)
(
2
v)
i
=
2
v
i
. (2.58)
Note, however, that the above identities are only valid in Cartesian coordinates. (See Appendix C.)
2.13 Energy Conservation
Consider a xed volume V surrounded by a surface S . The total energy content of the uid contained within V is
E =
_
V
cdV +
_
V
1
2
v
i
v
i
dV, (2.59)
where the rst and second terms on the right-hand side are the net internal and kinetic energies, respectively. Here,
c(r, t) is the internal (i.e., thermal) energy per unit mass of the uid. The energy ux across S , and out of V, is [cf.,
Equation (2.29)]
E
=
_
S
_
c +
1
2
v
i
v
i
_
v
j
dS
j
=
_
V
x
j
_
_
c +
1
2
v
i
v
i
_
v
j
_
dV, (2.60)
where use has been made of the tensor divergence theorem. According to the rst law of thermodynamics, the rate of
increase of the energy contained within V, plus the net energy ux out of V, is equal to the net rate of work done on
the uid within V, minus the net heat ux out of V: i.e.,
dE
dt
+
E
=
.
W
.
Q, (2.61)
where
.
W is the net rate of work, and
.
Q the net heat ux. It can be seen that
.
W
.
Q is the eective energy generation
rate within V [cf., Equation (2.31)].
Now, the net rate at which volume and surface forces do work on the uid within V is
.
W =
_
V
v
i
F
i
dV +
_
S
v
i

i j
dS
j
=
_
V
_
v
i
F
i
+
(v
i
i j
)
x
j
_
dV, (2.62)
where use has been made of the tensor divergence theorem.
Generally speaking, heat ow in uids is driven by temperature gradients. Let the q
i
(r, t) be the Cartesian com-
ponents of the heat ux density at position r and time t. It follows that the heat ux across a surface element dS,
located at point r, is q dS = q
i
dS
i
. Let T(r, t) be the temperature of the uid at position r and time t. Thus, a general
temperature gradient takes the form T/x
i
. Let us assume that there is a linear relationship between the components
of the local heat ux density and the local temperature gradient: i.e.,
q
i
= A
i j
T
x
j
, (2.63)
where the A
i j
are the components of a second-rank tensor (which can be functions of position and time). Now, in an
isotropic uid we would expect A
i j
to be an isotropic tensor. (See Section B.5.) However, the most general second-
order isotropic tensor is simply a multiple of
i j
. Hence, we can write
A
i j
=
i j
, (2.64)
where (r, t) is termed the thermal conductivity of the uid. It follows that the most general expression for the heat
ux density in an isotropic uid is
q
i
=
T
x
i
, (2.65)
or, equivalently,
q = T. (2.66)
Moreover, it is a matter of experience that heat ows down temperature gradients: i.e., > 0. We conclude that the net
heat ux out of volume V is
.
Q =
_
S
T
x
i
dS
i
=
_
V
x
i
_
T
x
i
_
dV, (2.67)
where use has been made of the tensor divergence theorem.
Equations (2.59)(2.62) and (2.67) can be combined to give the following energy conservation equation:
_
V
_

t
_
_
c +
1
2
v
i
v
i
__
+
x
j
_
_
c +
1
2
v
i
v
i
_
v
j
__
dV
=
_
V
_
v
i
F
i
+
x
j
_
v
i

i j
+
T
x
j
__
dV. (2.68)
However, this result is valid irrespective of the size, shape, or location of volume V, which is only possible if
t
_
_
c +
1
2
v
i
v
i
__
+
x
j
_
_
c +
1
2
v
i
v
i
_
v
j
_
= v
i
F
i
+
x
j
_
v
i

i j
+
T
x
j
_
(2.69)
everywhere inside the uid. Expanding some of the derivatives, and rearranging, we obtain
D
Dt
_
c +
1
2
v
i
v
i
_
= v
i
F
i
+
x
j
_
v
i
i j
+
T
x
j
_
, (2.70)
where use has been made of the continuity equation (2.40). Now, the scalar product of v with the uid equation of
motion (2.53) yields
v
i
Dv
i
Dt
=
D
Dt
_
1
2
v
i
v
i
_
= v
i
F
i
+ v
i
i j
x
j
. (2.71)
Combining the previous two equations, we get
Dc
Dt
=
v
i
x
j
i j
+
x
j
_
T
x
j
_
. (2.72)
20 FLUID MECHANICS
Finally, making use of (2.26), we deduce that the energy conservation equation for an isotropic Newtonian uid takes
the general form
Dc
Dt
=
p
v
i
x
i
+
1
_
+
x
j
_
T
x
j
__
. (2.73)
Here,
=
v
i
x
j
d
i j
= 2
_
e
i j
e
i j

1
3
e
ii
e
j j
_
=
_
v
i
x
j
v
i
x
j
+
v
i
x
j
v
j
x
i
2
3
v
i
x
i
v
j
x
j
_
(2.74)
is the rate of heat generation per unit volume due to viscosity. When written in vector form, Equation (2.73) becomes
Dc
Dt
=
p
v +
+
( T)
. (2.75)
According to the above equation, the internal energy per unit mass of a co-moving uid element evolves in time as
a consequence of work done on the element by pressure as its volume changes, viscous heat generation due to ow
shear, and heat conduction.
2.14 Equations of Incompressible Fluid Flow
In most situations of general interest, the owof a conventional liquid, such as water, is incompressible to a high degree
of accuracy. Now, a uid is said to be incompressible when the mass density of a co-moving volume element does not
change appreciably as the element moves through regions of varying pressure. In other words, for an incompressible
uid, the rate of change of following the motion is zero: i.e.,
D
Dt
= 0. (2.76)
In this case, the continuity equation (2.40) reduces to
v = 0. (2.77)
We conclude that, as a consequence of mass conservation, an incompressible uid must have a divergence-free, or
solenoidal, velocity eld. This immediately implies, from Equation (2.42), that the volume of a co-moving uid
element is a constant of the motion. In most practical situations, the initial density distribution in an incompressible
uid is uniform in space. Hence, it follows from (2.76) that the density distribution remains uniform in space and
constant in time. In other words, we can generally treat the density, , as a uniform constant in incompressible uid
ow problems.
Suppose that the volume force acting on the uid is conservative in nature (see Section A.18): i.e.,
F = , (2.78)
where (r, t) is the potential energy per unit mass, and the potential energy per unit volume. Assuming that
the uid viscosity is a spatially uniform quantity, which is generally the case (unless there are strong temperature
variations within the uid), the Navier-Stokes equation for an incompressible uid reduces to
Dv
Dt
=
p
+
2
v, (2.79)
where
=
(2.80)
is termed the kinematic viscosity, and has units of meters squared per second. Roughly speaking, momentum diuses
a distance of order

t meters in t seconds as a consequence of viscosity. The kinematic viscosity of water at 20
C
is about 1.0 10
6
m
2
/s. It follows that viscous momentum diusion in water is a relatively slow process.
The complete set of equations governing incompressible ow is
v = 0, (2.81)
Dv
Dt
=
p
+
2
v. (2.82)
Here, and are regarded as known constants, and (r, t) as a known function. Thus, we have four equations
namely, Equation (2.81), plus the three components of Equation (2.82)for four unknownsnamely, the pressure,
p(r, t), plus the three components of the velocity, v(r, t). Note that an energy conservation equation is redundant in the
case of incompressible uid ow.
2.15 Equations of Compressible Fluid Flow
In many situations of general interest, the ow of gases is compressible: i.e., there are signicant changes in the
mass density as the gas ows from place to place. For the case of compressible ow, the continuity equation (2.40),
and the Navier-Stokes equation (2.56), must be augmented by the energy conservation equation (2.75), as well as
thermodynamic relations that specify the internal energy per unit mass, and the temperature in terms of the density
and pressure. For an ideal gas, these relations take the form
c =
c
V
A
T, (2.83)
T =
A
F
p
, (2.84)
where c
V
is the molar specic heat at constant volume, F = 8.3145 J K
1
mol
1
the molar ideal gas constant, A the
molar mass (i.e., the mass of 1 mole of gas molecules), and T the temperature in degrees Kelvin. Incidentally, 1 mole
corresponds to 6.0221 10
24
molecules. Here, we have assumed, for the sake of simplicity, that c
V
is a uniform
constant. It is also convenient to assume that the thermal conductivity, , is a uniform constant. Making use of these
approximations, Equations (2.40), (2.75), (2.83), and (2.84) can be combined to give
1
1
_
Dp
Dt

p
D
Dt
_
= +
A
F

2
_
p
_
, (2.85)
where
=
c
p
c
V
=
c
V
+ F
c
V
(2.86)
is the ratio of the molar specic heat at constant pressure, c
p
, to that at constant volume, c
V
. (Incidentally, the result
that c
p
= c
V
+ F for an ideal gas is a standard theorem of thermodynamics.) The ratio of specic heats of dry air at
20
C is 1.40.
The complete set of equations governing compressible ideal gas ow are
D
Dt
= v, (2.87)
Dv
Dt
=
p
2
v +
1
3
(v)
_
, (2.88)
1
1
_
Dp
Dt

p
D
Dt
_
= +
A
F

2
_
p
_
, (2.89)
where the dissipation function is specied in terms of and v in Equation (2.74). Here, , , , A, and Fare regarded
as known constants, and (r, t) as a known function. Thus, we have ve equationsnamely, Equations (2.87) and
(2.89), plus the three components of Equation (2.88)for ve unknownsnamely, the density, (r, t), the pressure,
p(r, t), and the three components of the velocity, v(r, t).
22 FLUID MECHANICS
2.16 Dimensionless Numbers in Incompressible Flow
It is helpful to normalize the equations of incompressible uid ow, (2.81)(2.82), in the following manner: = L,
v = v/V
0
, t = (V
0
/L) t, = /(g L), and p = p/( V
2
0
+ g L + V
0
/L). Here, L is a typical spatial variation length-
scale, V
0
a typical uid velocity, and g a typical gravitational acceleration (assuming that represents a gravitational
potential energy per unit mass). All barred quantities are dimensionless, and are designed to be comparable with unity.
The normalized equations of incompressible uid ow take the form
v = 0, (2.90)
Dv
Dt
=
_
1 +
1
Fr
2
+
1
Re
_
p

Fr
2
+
2
v
Re
, (2.91)
where D/Dt = /t + v, and
Re =
LV
0
, (2.92)
Fr =
V
0
(g L)
1/2
. (2.93)
Here, the dimensionless quantities Re and Fr are known as the Reynolds number and the Froude number, respectively.
The Reynolds number is the typical ratio of inertial to viscous forces within the uid, whereas the square of the Froude
number is the typical ratio of inertial to gravitational forces. Thus, viscosity is relatively important compared to inertia
when Re 1, and vice versa. Likewise, gravity is relatively important compared to inertia when Fr 1, and vice
versa. Note that, in principal, Re and Fr are the only quantities in Equations (2.90) and (2.91) that can be signicantly
greater or smaller than unity.
For the case of water at 20
C, located on the surface of the Earth,

Re 1.0 10
6
L(m) V
0
(ms
1
), (2.94)
Fr 3.2 10
1
V
0
(ms
1
)/[L(m)]
1/2
. (2.95)
Thus, if L 1 m and V
0
1 ms
1
, as is often the case for terrestrial water dynamics, then the above expressions sug-
gest that Re 1 and Fr O(1). In this situation, the viscous term on the right-hand side of (2.91) becomes negligible,
and the (unnormalized) incompressible uid ow equations reduce to the following inviscid, incompressible, uid ow
equations,
v = 0, (2.96)
Dv
Dt
=
p
. (2.97)
For the case of lubrication oil at 20
C, located on the surface of the Earth, 1.0 10

4
m
2
s
1
(i.e., oil is about
100 times more viscous than water), and so
Re 1.0 10
4
L(m) V
0
(ms
1
), (2.98)
Fr 3.2 10
1
V
0
(ms
1
)]/[L(m)]
1/2
. (2.99)
Suppose that oil is slowly owing down a narrow lubrication channel such that L 10
3
m and V
0
10
1
ms
1
. It
follows, from the above expressions, that Re 1 and Fr 1. In this situation, the inertial term on the left-hand
side of (2.91) becomes negligible, and the (unnormalized) incompressible uid ow equations reduce to the following
inertia-free, incompressible, uid ow equations,
v = 0, (2.100)
0 =
p
+
2
v. (2.101)
2.17 Dimensionless Numbers in Compressible Flow
It is helpful to normalize the equations of compressible ideal gas ow, (2.87)(2.89), in the following manner: =
L, v = v/V
0
, t = (V
0
/L) t, = /
0
, = /(g L), = (L/V
0
)
2
, and p = (p p
0
)/(
0
V
2
0
+
0
g L +
0
V
0
/L).
Here, L is a typical spatial variation length-scale, V
0
a typical uid velocity,
0
a typical mass density, and g a typical
gravitational acceleration (assuming that represents a gravitational potential energy per unit mass). Furthermore, p
0
corresponds to atmospheric pressure at ground level, and is a uniform constant. It follows that p represents deviations
from atmospheric pressure. All barred quantities are dimensionless, and are designed to be comparable with unity.
The normalized equations of compressible ideal gas ow take the form
D
Dt
= v, (2.102)
Dv
Dt
=
_
1 +
1
Fr
2
+
1
Re
_
p

Fr
2
+
1
Re
_
2
v
1
3
(v)
_
, (2.103)
1
1
_
Dp
Dt

_
p
0
+ p
_
D
Dt
_
=
1 + Re (1 + 1/Fr
2
)
+
1
Re Pr

2
_
p
0
+ p
_
, (2.104)
p
0
=
1
M
2
(1 + 1/Fr
2
+ 1/Re)
, (2.105)
where D/Dt /t + v,
Re =
L V
0
, (2.106)
Fr =
V
0
(g L)
1/2
, (2.107)
Pr =
H
, (2.108)
M =
V
0
_
p
0
/
0
, (2.109)
and
=
0
, (2.110)
H
=
A
F
0
. (2.111)
Here, the dimensionless numbers Re, Fr, Pr, and M are known as the Reynolds number, Froude number, Prandtl
number, and Mach number, respectively. The Reynolds number is the typical ratio of inertial to viscous forces within
the gas, the square of the Froude number the typical ratio of inertial to gravitational forces, the Prandtl number the
typical ratio of the momentum and thermal diusion rates, and the Mach number the typical ratio of gas ow and
sound propagation speeds. Thus, thermal diusion is far faster than momentum diusion when Pr 1, and vice
versa. Moreover, the gas ow is termed subsonic when M 1, supersonic when M 1, and transonic when
M O(1). Note that
_
p
0
/
0
is the speed of sound in the undisturbed gas. The quantity
H
is called the thermal
diusivity of the gas, and has units of meters squared per second. Thus, heat typically diuses through the gas a
distance

H
t meters in t seconds. The thermal diusivity of dry air at atmospheric pressure and 20
C is about
H
= 2.1 10
5
m
2
s
1
. It follows that heat diusion in air is a relatively slow process. The kinematic viscosity of
dry air at atmospheric pressure and 20
C is about = 1.5 10
5
m
2
s
1
. Hence, momentum diusion in air is also a
relatively slow process.
For the case of dry air at atmospheric pressure and 20
C,
Re 6.7 10
4
L(m) V
0
(ms
1
), (2.112)
24 FLUID MECHANICS
Fr 3.2 10
1
V
0
(ms
1
)/[L(m)]
1/2
, (2.113)
Pr 7.2 10
1
, (2.114)
M 2.9 10
3
V
0
(ms
1
). (2.115)
Thus, if L 1 m and V
0
1 ms
1
, as is often the case for subsonic air dynamics close to the Earths surface, the
above expressions suggest that Re 1, M 1, and Fr, Pr O(1). It immediately follows from Equation (2.105)
that p
0
1. However, in this situation, Equation (2.104) is dominated by the second term in square brackets on its
left-hand side. Hence, this equation can only be satised if the term in question is small, which implies that
D
Dt
1. (2.116)
Equation (2.102) then gives
v 1. (2.117)
Thus, it is evident that subsonic (i.e., M 1) gas ow is essentially incompressible. The fact that Re 1 implies
that such ow is also essentially inviscid. In the incompressible inviscid limit (in which v = 0 and Re 1), the
(unnormalized) compressible ideal gas ow equations reduce to the previously derived inviscid, incompressible, uid
ow equations: i.e.,
v = 0, (2.118)
Dv
Dt
=
p
. (2.119)
It follows that the equations which govern subsonic gas dynamics close to the surface of the Earth are essentially the
same as those which govern the ow of water.
Suppose that L 1 m and V
0
300 ms
1
, as is typically the case for transonic air dynamics (e.g., air ow over the
wing of a ghter jet). In this situation, Equations (2.105) and (2.112)(2.115) yield Re, Fr 1 and M, Pr, p
0
O(1).
It follows that the nal two terms on the right-hand sides of Equations (2.103) and (2.104) can be neglected. Thus,
the (unnormalized) compressible ideal gas ow equations reduce to the following set of inviscid, adiabatic, ideal gas,
ow equations,
D
Dt
= v, (2.120)
Dv
Dt
=
p
, (2.121)
D
Dt
_
p
_
= 0. (2.122)
In particular, if the initial distribution of p/
is uniform in space, as is often the case, then Equation (2.122) ensures

that the distribution remains uniform as time progresses. In fact, it can be shown that the entropy per unit mass of an
ideal gas is
S =
c
V
A
ln
_
p
_
. (2.123)
Hence, the assumption that p/
is uniform in space is equivalent to the assumption that the entropy per unit mass of
the gas is a spatial constant. A gas in which this is the case is termed homentropic. Equation (2.122) ensures that the
entropy of a co-moving gas element is a constant of the motion in transonic ow. A gas in which this is the case is
termed isentropic. In the homentropic case, the above compressible gas ow equations simplify somewhat to give
D
Dt
= v, (2.124)
Dv
Dt
=
p
, (2.125)
p
p
0
=
_

0
_
. (2.126)
Here, p
0
is atmospheric pressure, and
0
is the density of air at atmospheric pressure. Equation (2.126) is known as the
adiabatic gas law, and is a consequence of the fact that transonic gas dynamics takes place far too quickly for thermal
heat conduction (which is a relatively slow process) to have any appreciable eect on the temperature distribution
within the gas. Incidentally, a gas in which thermal diusion is negligible is generally termed adiabatic.
2.18 Fluid Equations in Cartesian Coordinates
Let us adopt the conventional Cartesian coordinate system, x, y, z. According to Equation (2.26), the various compo-
nents of the stress tensor are
xx
= p + 2
v
x
x
, (2.127)
yy
= p + 2
v
y
y
, (2.128)
zz
= p + 2
v
z
z
, (2.129)
xy
=
yx
=
_
v
x
y
+
v
y
x
_
, (2.130)
xz
=
zx
=
_
v
x
z
+
v
z
x
_
, (2.131)
yz
=
zy
=
_
v
y
z
+
v
z
y
_
, (2.132)
where v is the velocity, p the pressure, and the viscosity. The equations of compressible uid ow, (2.87)(2.89)
(from which the equations of incompressible uid ow can easily be obtained by setting = 0), become
D
Dt
= , (2.133)
Dv
x
Dt
=
1
p
x

x
+
2
v
x
+
1
3
x
_
, (2.134)
Dv
y
Dt
=
1
p
y

y
+
2
v
y
+
1
3
y
_
, (2.135)
Dv
z
Dt
=
1
p
z

z
+
2
v
z
+
1
3
z
_
, (2.136)
1
1
_
D
Dt

p
D
Dt
_
= +
A
F

2
_
p
_
, (2.137)
where is the mass density, the ratio of specic heats, the heat conductivity, A the molar mass, and F the molar
ideal gas constant. Furthermore,
=
v
x
x
+
v
y
y
+
v
z
z
, (2.138)
D
Dt
=
t
+ v
x
x
+ v
y
y
+ v
z
z
, (2.139)
2
=
2
x
2
+
2
y
2
+
2
z
2
, (2.140)
= 2
_
_
_
v
x
x
_
2
+
_
v
y
y
_
2
+
_
v
z
z
_
2
+
1
2
_
v
x
y
+
v
y
x
_
2
26 FLUID MECHANICS
+
1
2
_
v
x
z
+
v
z
x
_
2
+
1
2
_
v
y
z
+
v
z
y
_
2
_
_
. (2.141)
In the above, , , , and A are treated as uniform constants.
2.19 Fluid Equations in Cylindrical Coordinates
Let us adopt the cylindrical coordinate system, r, , z. Making use of the results quoted in Section C.3, the components
of the stress tensor are
rr
= p + 2
v
r
r
, (2.142)
= p + 2
_
1
r
v
+
v
r
r
_
, (2.143)
zz
= p + 2
v
z
z
, (2.144)
r
=
r
=
_
1
r
v
r
+
v
r

v
r
_
, (2.145)
rz
=
zr
=
_
v
r
z
+
v
z
r
_
, (2.146)
z
=
z
=
_
1
r
v
z
+
v
z
_
, (2.147)
whereas the equations of compressible uid ow become
D
Dt
= , (2.148)
Dv
r
Dt

v
2
r
=
1
p
r

r
+
2
v
r

v
r
r
2

2
r
2
v
+
1
3
r
_
, (2.149)
Dv
Dt
+
v
r
v
r
=
1
r
p
1
r
2
v
+
2
r
2
v
r
r
2
+
1
3r
_
, (2.150)
Dv
z
Dt
=
1
p
z

z
+
2
v
z
+
1
3
z
_
, (2.151)
1
1
_
D
Dt

p
D
Dt
_
= +
A
F

2
_
p
_
, (2.152)
where
=
1
r
(r v
r
)
r
+
1
r
v
+
v
z
z
, (2.153)
D
Dt
=
t
+ v
r
r
+
v
+ v
z
z
, (2.154)
2
=
1
r
r
_
r

r
_
+
1
r
2
2
+
2
z
2
, (2.155)
= 2
_
_
_
v
r
r
_
2
+
_
1
r
v
+
v
r
r
_
2
+
_
v
z
z
_
2
+
1
2
_
1
r
v
r
+
v
r

v
r
_
2
+
1
2
_
v
r
z
+
v
z
r
_
2
+
1
2
_
v
z
+
1
r
v
z
_
2
_
_
. (2.156)
2.20 Fluid Equations in Spherical Coordinates
Let us, nally, adopt the spherical coordinate system, r, , . Making use of the results quoted in Section C.4, the
components of the stress tensor are
rr
= p + 2
v
r
r
, (2.157)
= p + 2
_
1
r
v
+
v
r
r
_
, (2.158)
= p + 2
_
1
r sin
v
+
v
r
r
+
cot v
r
_
, (2.159)
r
=
r
=
_
1
r
v
r
+
v
r

v
r
_
, (2.160)
r
=
r
=
_
1
r sin
v
r
+
v
r

v
r
_
, (2.161)
=
_
1
r sin
v
+
1
r
v
cot v
r
_
, (2.162)
whereas the equations of compressible uid ow become
D
Dt
= , (2.163)
Dv
r
Dt

v
2
+ v
2
r
=
1
p
r

r
+
2
v
r

2v
r
r
2

2
r
2
v
2 cot v
r
2

2
r
2
sin
v
+
1
3
r
_
, (2.164)
Dv
Dt
+
v
r
v
cot v
2
r
=
1
r
p
1
r
2
v
+
2
r
2
v
r
(2.165)
r
2
sin
2
2 cot
r
2
sin
v
+
1
3r
_
, (2.166)
Dv
Dt
+
v
r
v
+ cot v
r
=
1
r sin
p
1
r sin
2
v
r
2
sin
2
+
2
r
2
sin
2
v
r
+
2 cot
r
2
sin
v
+
1
3r sin
_
, (2.167)
1
1
_
D
Dt

p
D
Dt
_
= +
A
F

2
_
p
_
, (2.168)
where
=
1
r
2
(r
2
v
r
)
r
+
1
r sin
(sin v
+
1
r sin
v
, (2.169)
28 FLUID MECHANICS
D
Dt
=
t
+ v
r
r
+
v
+
v
r sin
, (2.170)
2
=
1
r
2
r
_
r
2

r
_
+
1
r
2
sin
_
sin

_
+
1
r
2
sin
2
2
, (2.171)
= 2
_
_
_
v
r
r
_
2
+
_
1
r
v
+
v
r
r
_
2
+
_
1
r sin
v
+
v
r
r
+
cot v
r
_
2
+
1
2
_
1
r
v
r
+
v
r

v
r
_
2
+
1
2
_
1
r sin
v
r
+
v
r

v
r
_
2
+
1
2
_
1
r sin
v
+
1
r
v
cot v
r
_
2
_
_
. (2.172)
2.21 Exercises
2.1. Equations (2.66), (2.75), and (2.87) can be combined to give the following energy conservation equation for a non-ideal
compressible uid:
Dc
Dt

p
D
Dt
= q,
where is the mass density, p the pressure, c the internal energy per unit mass, the viscous energy dissipation rate per unit
volume, and q the heat ux density. We also have
D
Dt
= v,
q = T,
where v is the uid velocity, T the temperature, and the thermal conductivity. Now, according to a standard theorem in
thermodynamics,
T dS = dc
p
2
d,
where S is the entropy per unit mass. Moreover, the entropy ux density at a given point in the uid is
s = Sv +
q
T
,
where the rst term on the right-hand side is due to direct entropy convection by the uid, and the second is the entropy ux
density associated with heat conduction.
Derive an entropy conservation equation of the form
dS
dt
+
S
=
S
,
where S is the net amount of entropy contained in some xed volume V,
S
the entropy ux out of V, and
S
the net rate of
entropy creation within V. Give expressions for S ,
S
, and
S
. Demonstrate that the entropy creation rate per unit volume
is
=
T
+
q q
T
2
.
Finally, show that 0, in accordance with the second law of thermodynamics.
2.2. The Navier-Stokes equation for an incompressible uid of uniform mass density takes the form
Dv
Dt
=
p
+
2
v,
where v is the uid velocity, p the pressure, the potential energy per unit mass, and the (uniform) kinematic viscosity.
The incompressibility constraint requires that
v = 0.
Finally, the quantity
v
is generally referred to as the uid vorticity.
Derive the following vorticity evolution equation from the Navier-Stokes equation:
D
Dt
= ( ) v +
2
.
2.3. Consider two-dimensional incompressible uid ow. Let the velocity eld take the form
v = v
x
(x, y, t) e
x
+ v
y
(x, y, t) e
y
.
Demonstrate that the equations of incompressible uid ow (see Exercise 2.2) can be satised by writing
v
x
=
y
,
v
y
=
x
,
where
t
+
y
=
2
,
and
=
2
.
Here,
2
=
2
/x
2
+
2
/y
2
. Furthermore, the quantity is termed a stream function, since v = 0: i.e., the uid ow is
everywhere parallel to contours of .
2.4. Consider incompressible irrotational ow: i.e., ow that satises
Dv
Dt
=
p
+
2
v,
v = 0,
as well as
v = 0.
Here, v is the uid velocity, the uniform mass density, p the pressure, the potential energy per unit mass, and the
(uniform) kinematic viscosity.
Demonstrate that the above equations can be satised by writing
v = ,
where
2
= 0,
and
t
+
1
2
v
2
+
p
+ = ((t).
Here, ((t) is a spatial constant. This type of ow is known as potential ow, since the velocity eld is derived from a scalar
potential.
2.5. The equations of inviscid adiabatic ideal gas ow are
D
Dt
= v,
Dv
Dt
=
p
,
D
Dt
_
p
_
= 0.
Here, is the mass density, v the ow velocity, p the pressure, the potential energy per unit mass, and the (uniform)
ratio of specic heats. Suppose that the pressure and potential energy are both time independent: i.e., p/t = /t = 0.
Demonstrate that
H =
1
2
v
2
+
1
p
+
is a constant of the motion: i.e., DH/Dt = 0. This result is known as Bernoullis theorem.
30 FLUID MECHANICS
2.6. The equations of inviscid adiabatic non-ideal gas ow are
D
Dt
= v,
Dv
Dt
=
p
,
Dc
Dt

p
2
D
Dt
= 0.
Here, is the mass density, v the ow velocity, p the pressure, the potential energy per unit mass, and c the internal
energy per unit mass. Suppose that the pressure and potential energy are both time independent: i.e., p/t = /t = 0.
Demonstrate that
H =
1
2
v
2
+ c +
p
+
is a constant of the motion: i.e., DH/Dt = 0. This result is a more general form of Bernoullis theorem.
2.7. Demonstrate that Bernoullis theorem for incompressible, inviscid uid ow takes the form DH/Dt = 0, where
H =
1
2
v
2
+
p
+ .
Hydrostatics 31
3 Hydrostatics
3.1 Introduction
This chapter discusses hydrostatics, which is the study of the mechanical equilibrium of incompressible uids.
3.2 Hydrostatic Pressure
Consider a body of water that is stationary in a reference frame that is xed with respect to the Earths surface. In this
chapter, such a frame is treated as approximately inertial. Let z measure vertical height, and suppose that the region
z 0 is occupied by water, and the region z > 0 by air. According to Equation (2.79), the air/water system remains in
mechanical equilibrium (i.e., v = Dv/Dt = 0) provided
0 = p + , (3.1)
where p is the static uid pressure, the mass density, = g z the gravitational potential energy per unit mass, and g
the (approximately uniform) acceleration due to gravity. Now,
(z) =
_
0 z > 0
0
z 0
, (3.2)
where
0
is the (approximately uniform) mass density of water. Here, the comparatively small mass density of air has
been neglected. Since = (z) and = (z), it immediately follows, from (3.1), that p = p(z), where
dp
dz
= g. (3.3)
We conclude that constant pressure surfaces in a stationary body of water take the form of horizontal planes. Making
use of (3.2), the above equation can be integrated to give
p(z) =
_
p
0
z > 0
p
0

0
g z z 0
, (3.4)
where p
0
10
5
Nm
2
is atmospheric pressure at ground level. According to this expression, pressure in stationary
water increases linearly with increasing depth (i.e., with decreasing z, for z < 0). In fact, given that g 9.8 ms
2
and
0
10
3
kg m
3
, we deduce that hydrostatic pressure rises at the rate of 1 atmosphere (i.e., 10
5
Nm
2
) every 10.2 m
increase in depth below the surface.
3.3 Buoyancy
Consider the air/water system described in the previous section. Let V be some volume, bounded by a closed surface
S , that straddles the plane z = 0, and is thus partially occupied by water, and partially by air. The i-component of the
net force acting on the uid (i.e., either water or air) contained within V is written (see Section 2.3)
f
i
=
_
S
i j
dS
j
+
_
V
F
i
dV, (3.5)
where
i j
= p
i j
(3.6)
is the stress tensor for a static uid (see Section 2.5), and
F = g e
z
(3.7)
32 FLUID MECHANICS
the gravitational force density. (Recall that the indices 1, 2, and 3 refer to the x-, y-, and z-axes, respectively. Thus,
f
3
f
z
, etc.) The rst term on the right-hand side of (3.5) represents the net surface force acting across S , whereas the
second term represents the net volume force distributed throughout V. Making use of the tensor divergence theorem
(see Section B.4), Equations (3.5)(3.7) yield the following expression for the net force:
f = B + W, (3.8)
where
B
i
=
_
V
p
x
i
dV, (3.9)
and
W
x
= W
y
= 0, (3.10)
W
z
=
_
V
g dV. (3.11)
Here, B is the net surface force, and W the net volume force.
It follows from Equations (3.4) and (3.9) that
B = M
0
g e
z
, (3.12)
where M
0
=
0
V
0
. Here, V
0
is volume of that part of V which lies below the waterline, and M
0
the total mass of water
contained within V. Moreover, from Equations (3.2), (3.10), and (3.11),
W = M
0
g e
z
. (3.13)
It can be seen that the net surface force, B, is directed vertically upward, and exactly balances the net volume force, W,
which is directed vertically downward. Of course, W is the weight of the water contained within V. On the other hand,
B, which is generally known as the buoyancy force, is the resultant pressure of the water immediately surrounding
V. We conclude that, in equilibrium, the net buoyancy force acting across S exactly balances the weight of the water
inside V, so that the total force acting on the contents of V is zero, as must be the case for a system in mechanical
equilibrium. We can also deduce that the line of action of B (which is vertical) passes through the center of gravity of
the water inside V. Otherwise, a net torque would act on the contents of V, which would contradict our assumption
that the system is in mechanical equilibrium.
3.4 Equilibrium of Floating Bodies
Consider the situation described in the previous section. Suppose that the uid contained within V is replaced by
a partially submerged solid body whose outer surface corresponds to S . Furthermore, suppose that this body is in
mechanical equilibrium with the surrounding uid (i.e., it is stationary, and oating on the surface of the water). It
follows that the pressure distribution in the surrounding uid is unchanged [since the force balance criterion (3.3) can
be integrated to give the pressure distribution (3.4) at all contiguous points in the uid, provided that the uid remains
in mechanical equilibrium]. We conclude that the net surface force acting across S is also unchanged (since this is
directly related to the pressure distribution in the uid immediately surrounding V), which implies that the buoyancy
force acting on the oating body is the same as that acting on the displaced water: i.e., the water that previously
occupied V. In other words, from (3.12),
B = W
0
e
z
, (3.14)
where W
0
= M
0
g and M
0
are the weight and mass of the displaced water, respectively. The fact that the buoyancy
force is unchanged also implies that the vertical line of action of B passes through the center of gravity, H (say), of the
displaced water. Incidentally, H is generally known as the center of buoyancy.
A oating body of weight W is acted upon by two forces: namely, its own weight,
W = W e
z
, (3.15)
Hydrostatics 33
and the buoyancy force, B = W
0
e
z
, due to the pressure of the surrounding water. Of course, the line of action of W
passes through the bodys center of gravity, G (say). Now, to remain in equilibrium, the body must be subject to zero
net force and zero net torque. The requirement of zero net force yields W
0
= W. In other words, in equilibrium, the
weight of the water displaced by a oating body is equal to the weight of the body, or, alternatively, in equilibrium, the
magnitude of the buoyancy force acting on a oating body is equal to the weight of the displaced water. This famous
result is known as Archimedes principle. The requirement of zero net torque implies that, in equilibrium, the center
of gravity, G, and center of buoyancy, H, of a oating body lie on the same vertical straight-line.
Consider a oating body of mass M and volume V. Let = M/V be the bodys mean mass density. Archimedes
principle implies that, in equilibrium,
V
0
V
= s, (3.16)
where
s =
0
(3.17)
is termed the bodys specic gravity. (Recall, that V
0
is the submerged volume, and
0
the mass density of water.)
We conclude, from (3.16), that the volume fraction of a oating body that is submerged is equal to the bodys specic
gravity. Obviously, the specic gravity must be less than unity, since the submerged volume fraction cannot exceed
unity. In fact, if the specic gravity exceeds unity then it is impossible for the buoyancy force to balance the bodys
weight, and the body consequently sinks.
Consider a body of volume V and specic gravity s that oats in equilibrium. It follows, fromEquation (3.16), that
the submerged volume is V
0
= s V. Hence, the volume above the waterline is V
1
= V V
0
= (1s) V. Suppose that the
body is inverted such that its previously submerged part is raised above the waterline, and vice versa: i.e., V
0
V
1
.
According to (3.16), the body can only remain in equilibrium in this conguration if its specic gravity changes to
s
=
V
1
V
=
V V
0
V
= 1 s. (3.18)
We conclude that for every equilibrium conguration of a oating body of specic gravity s there exists an inverted
equilibrium conguration for a body of the same shape having the complementary specic gravity 1 s.
3.5 Vertical Stability of Floating Bodies
Consider a oating body of weight W which, in equilibrium, has a submerged volume V
0
. Thus, the bodys downward
weight is balanced by the upward buoyancy force, B =
0
V
0
g: i.e.,
0
V
0
g = W. Let A
0
be the cross-sectional
area of the body at the waterline (i.e., in the plane z = 0). It is convenient to dene the bodys mean draft (or mean
submerged depth) as
0
= V
0
/A
0
. Suppose that the body is displaced slightly downward, without rotation, such that
its mean draft becomes
0
+
1
, where
1

0
. Assuming that the cross-sectional area in the vicinity of the waterline
is constant, the new submerged volume is V
0
= A
0
(
0
+
1
) = V
0
+ A
0

1
, and the new buoyancy force becomes
B
=
0
V
0
g = W +
0
A
0
g
1
= (1 +
1
/
0
) W. However, the weight of the body is unchanged. Thus, the bodys
perturbed vertical equation of motion is written
W
g
d
2
1
dt
2
= W B
=
W
1
, (3.19)
which reduces to the simple harmonic equation
d
2
1
dt
2
=
g
1
. (3.20)
We conclude that when a oating body of mean draft
0
is subject to a small vertical displacement it oscillates about
its equilibrium position at the characteristic frequency
=
_
g
0
. (3.21)
34 FLUID MECHANICS
STABLE
H H
G
UNSTABLE
M
M
G
H H
Figure 3.1: Stable and unstable congurations for a oating body.

It follows that such a body is unconditionally stable to small vertical displacements. Of course, the above analysis
presupposes that the oscillations take place suciently slowly that the water immediately surrounding the body always
remains in approximate hydrostatic equilibrium.
3.6 Angular Stability of Floating Bodies
Let us now investigate the stability of oating bodies to angular displacements. For the sake of simplicity, we shall
only consider bodies that have two mutually perpendicular planes of symmetry. Suppose that when such a body is in
an equilibriumstate the two symmetry planes are vertical, and correspond to the x = 0 and y = 0 planes. As before, the
z = 0 plane coincides with the surface of the water. It follows, from symmetry, that when the body is in an equilibrium
state its center of gravity, G, and center of buoyancy, H, both lie on the z-axis.
Suppose that the body turns through a small angle about some horizontal axis, lying in the plane z = 0, that goes
through the origin. Let GH be that, originally vertical, straight-line that passes through the bodys center of gravity,
G, and original center of buoyancy, H. Owing to the altered shape of the volume of displaced water, the center of
buoyancy is shifted to some new position H
. Let the vertical straight-line going through H
meet GH at M. See
Figure 3.1. Point M (in the limit 0) is called the metacenter. In the disturbed state, the bodys weight W acts
downward through G, and the buoyancy
0
V
0
g acts upward through M. Let us assume that the submerged volume,
V
0
, is unchanged from the equilibrium state (which excludes vertical oscillations from consideration). It follows that
the weight and the buoyancy force are equal and opposite, so that there is no net force on the body. However, as can be
seen from Figure 3.1, the weight and the buoyancy force generate a net torque of magnitude = W sin . Here, is
the length MG: i.e., the distance between the metacenter and the center of gravity. This distance is generally known as
the metacentric height, and is dened such that it is positive when M lies above G, and vice versa. Moreover, as is also
clear from Figure 3.1, when M lies above G the torque acts to reduce , and vice versa. In the former case, the torque
is known as a righting torque. We conclude that a oating body is stable to small angular displacements about some
horizontal axis lying in the plane z = 0 provided that its metacentric height is positive: i.e., provided that the metacenter
lies above the center of gravity. Since we have already demonstrated that a oating body is unconditionally stable to
small vertical displacements (and since it is also fairly obvious that such a body is neutrally stable to both horizontal
displacements and angular displacements about a vertical axis passing through its center of gravity), it follows that
a necessary and sucient condition for the stability of a oating body to a general small perturbation (made up of
arbitrary linear and angular components) is that its metacentric height be positive for angular displacements about any
Hydrostatics 35
horizontal axis.
3.7 Determination of Metacentric Height
Suppose that the oating body considered in the previous section is in an equilibrium state. Let A
0
be the cross-
sectional area at the waterline: i.e., in the plane z = 0. Since the body is assumed to be symmetric with respect to the
x = 0 and y = 0 planes, we have
_
A
0
x dx dy =
_
A
0
y dx dy =
_
A
0
x y dx dy = 0, (3.22)
where the integrals are taken over the whole cross-section at z = 0. Let (x, y) be the bodys draft: i.e., the vertical
distance between the surface of the water and the bodys lower boundary. It follows, from symmetry, that (x, y) =
(x, y) and (x, y) = (x, y). Moreover, the submerged volume is
V
0
=
_
A
0
_

0
dx dy dz =
_
A
0
(x, y) dx dy. (3.23)
It also follows from symmetry that
_
A
0
x (x, y) dx dy =
_
A
0
y (x, y) dx dy = 0. (3.24)
Now, the depth of the unperturbed center of buoyancy below the surface of the water is
h =
_
A
0
_

0
z dx dy dz
V
0
=
1
2 V
0
_
A
0
2
(x, y) dx dy =
2
0
A
0
2 V
0
, (3.25)
where
0
=
_
_
_
A
0
2
(x, y) dx dy
A
0
_
_
1/2
. (3.26)
Finally, from symmetry, the unperturbed center of buoyancy lies at x = y = 0.
Suppose that the body now turns through a small angle about the x-axis. As is easily demonstrated, the bodys
new draft becomes
(x, y) (x, y) y. Hence, the new submerged volume is

V
0
=
_
A
0
[(x, y) y] dx dy = V
0
+
_
A
0
y dx dy = V
0
, (3.27)
where use has been made of Equations (3.22) and (3.23). Thus, the submerged volume is unchanged, as should be the
case for a purely angular displacement. The new depth of the center of buoyancy is
h
=
_
A
0
_

0
z dx dy dz
V
0
=
1
2 V
0
_
A
0
[
2
(x, y) 2 y (x, y) + O(
2
)] dx dy = h, (3.28)
where use has been made of (3.24) and (3.25). Thus, the depth of the center of buoyancy is also unchanged. Moreover,
from symmetry, it is clear that the center of buoyancy still lies at x = 0. Finally, the new y-coordinate of the center of
buoyancy is
_
A
0
_

0
y dx dy dz
V
0
=
_
A
0
y [(x, y) y] dx dy
V
0
=
2
x
A
0
V
0
, (3.29)
where use has been made of (3.24). Here,
x
=
_
_
_
A
0
y
2
dx dy
A
0
_
_
1/2
, (3.30)
36 FLUID MECHANICS
is the radius of gyration of area A
0
about the x-axis.
It follows, fromthe above analysis, that if the oating body under consideration turns through a small angle about
the x-axis then its center of buoyancy shifts horizontally a distance
2
x
A
0
/V
0
in the plane perpendicular to the axis of
rotation. In other words, the distance HH
in Figure 3.1 is
2
x
A
0
/V
0
. Simple trigonometry reveals that HH
/MH
(assuming that is small). Hence, MH
= HH
/ =
2
x
A
0
/V
0
. Now, MH
is the height of the metacenter relative to the

center of buoyancy. However, the center of buoyancy lies a depth h below the surface of the water (which corresponds
to the plane z = 0). Hence, the z-coordinate of the metacenter is z
M
=
2
x
A
0
/V
0
h. Finally, if z
G
and z
H
= h are the
z-coordinates of the unperturbed centers of gravity and buoyancy, respectively, then
z
M
=
2
x
A
0
V
0
+ z
H
, (3.31)
and the metacentric height, = z
M
z
G
, becomes
=
2
x
A
0
V
0
z
GH
, (3.32)
where z
GH
= z
G
z
H
. Note that, since
2
x
A
0
/V
0
> 0, the metacenter always lies above the center of buoyancy.
A simple extension of the above argument reveals that if the body turns through a small angle about the y-axis
then the metacentric height is
=
2
y
A
0
V
0
z
GH
, (3.33)
where
y
=
_
_
_
A
0
x
2
dx dy
A
0
_
_
1/2
, (3.34)
is radius of gyration of area A
0
about the y-axis. Finally, as is easily demonstrated, if the body rotates about a horizontal
axis which subtends an angle with the x-axis then
=
2
A
0
V
0
z
GH
, (3.35)
where
2
=
2
x
cos
2
+
2
y
sin
2
. (3.36)
Thus, the minimum value of
2
is the lesser of
2
x
and
2
y
. It follows that the equilibrium state in question is uncondi-
tionally stable provided it is stable to small amplitude angular displacements about horizontal axes normal to its two
vertical symmetry planes (i.e., the x = 0 and y = 0 planes).
As an example, consider a uniform rectangular block of specic gravity s oating such that its sides of length a, b,
and c are parallel to the x-, y-, and z-axes, respectively. Such a block can be thought of as a very crude model of a ship.
The volume of the block is V = a b c. Hence, the submerged volume is V
0
= s V = s a b c. The cross-sectional area of
the block at the waterline (z = 0) is A
0
= a b. It is easily demonstrated that (x, y) =
0
= V
0
/A
0
= s c. Thus, the center
of buoyancy lies a depth h =
2
0
A
0
/2 V
0
= s c/2 below the surface of the water [see Equation (3.25)]. Moreover, by
symmetry, the center of gravity is a height c/2 above the bottom surface of the block, which is located a depth s c
below the surface of the water. Hence, z
H
= h = s c/2, z
G
= c/2 s c, and z
GH
= c (1 s)/2. Consider the stability
of the block to small amplitude angular displacements about the x-axis. We have
2
x
=
_
a/2
a/2
_
b/2
b/2
y
2
dx dy
a b
=
b
2
12
. (3.37)
Hence, from (3.32), the metacentric height is
=
b
2
12 s c

c
2
(1 s). (3.38)
Hydrostatics 37
The stability criterion > 0 yields
b
2
6 c
2
s (1 s) > 0. (3.39)
Since the maximum value that s (1 s) can take is 1/4, it follows that the block is stable for all specic gravities when
c < c
0
=
_
2
3
b. (3.40)
On the other hand, if c > c
0
then the block is unstable for intermediate specic gravities such that s
< s < s
+
, where
s
=
1
_
1 c
2
0
/c
2
2
, (3.41)
and is stable otherwise. Assuming that the block is stable, its angular equation of motion is written
I
d
2
dt
2
= W sin W , (3.42)
where
I =
W
g V
_
a/2
a/2
_
b/2
b/2
_
cs c
s c
(y
2
+ z
2
) dx dy dz =
W
12 g
_
b
2
+ 4 [(1 s)
3
+ s
3
] c
2
_
(3.43)
is the moment of inertia of the block about the x-axis. Thus, we obtain the the simple harmonic equation
d
2
dt
2
=
2
, (3.44)
where
2
=
W
I
=
g
s c
c
2
0
4 s (1 s) c
2
c
2
0
+ (8/3) [(1 s)
3
+ s
3
] c
2
. (3.45)
We conclude that the block executes small amplitude angular oscillations about the x-axis at the angular frequency
. However, this result is only accurate in the limit in which the oscillations are suciently slow that the water
surrounding the block always remains in approximate hydrostatic equilibrium. For the case of rotation about the
y-axis, the above analysis is unchanged except that a b.
The metacentric height of a conventional ship whose length greatly exceeds its width is typically much less for
rolling (i.e., rotation about a horizontal axis running along the ships length) than for pitching (i.e., rotation about a
horizontal axis perpendicular to the ships length), since the radius of gyration for pitching greatly exceeds that for
rolling. As is clear from Equation (3.45), a ship with a relatively small metacentric height (for rolling) has a relatively
long roll period, and vice versa. Now, an excessively low metacentric height increases the chances of a ship capsizing
if the weather is rough, or if its cargo/ballast shifts, or if it is damaged and partially ooded. For this reason, maritime
regulatory agencies, such as the International Maritime Organization, specify minimummetacentric heights for various
dierent types of sea-going vessel. A relatively large metacentric height, on the other hand, generally renders a ship
uncomfortable for passengers and crew, because the ship executes short period rolls, resulting in large g-forces. Such
forces also increase the risk that cargo may break loose or shift.
We saw earlier, in Section 3.4, that if a body of specic gravity s oats in vertical equilibrium in a certain position
then a body of the same shape, but of specic gravity 1 s, can oat in vertical equilibrium in the inverted position.
We can now demonstrate that these positions are either both stable, or both unstable, provided the body is of uniform
density. Let V
1
and V
2
be the volumes that are above and below the waterline, respectively, in the rst position. Let H
1
and H
2
be the mean centers of these two volumes, and H that of the whole volume. It follows that H
2
is the center of
buoyancy in the rst position, H
1
the center of buoyancy in the second (inverted) position, and H the center of gravity
in both positions. Moreover,
V
1
H
1
G = V
2
H
2
G =
_
V
1
V
2
V
1
+ V
2
_
H
1
H
2
, (3.46)
38 FLUID MECHANICS
where H
1
G is the distance between points H
1
and G, etc. The metacentric heights in the rst and second positions are
1
=
2
A
V
1
H
1
G =
1
V
1
_
A
2
_
V
1
V
2
V
1
+ V
2
_
H
1
H
2
_
, (3.47)
2
=
2
A
V
2
H
2
G =
1
V
2
_
A
2
_
V
1
V
2
V
1
+ V
2
_
H
1
H
2
_
, (3.48)
respectively, where A and are the area and radius of gyration of the common waterline section, respectively. Thus,
1

2
=
1
V
1
V
2
_
A
2
_
V
1
V
2
V
1
+ V
2
_
H
1
H
2
_
2
0, (3.49)
which implies that
1
0 as
2
0, and vice versa. It follows that the rst and second positions are either both stable,
both marginally stable, or both unstable.
3.8 Energy of a Floating Body
The conditions governing the equilibrium and stability of a oating body can also be deduced from the principle of
energy.
For the sake of simplicity, let us suppose that the water surface area is innite, so that the immersion of the body
does not generate any change in the water level. The potential energy of the body itself is W z
G
, where W is the bodys
weight, and z
G
the height of its center of gravity, G, relative to the surface of the water. If the body displaces a volume
V
0
of water then this eectively means that a weight
0
V
0
of water, whose center of gravity is located at the center
of buoyancy, H, is removed, and then spread as an innitely thin lm over the surface of the water. This involves a
gain of potential energy of
0
V
0
z
H
, where z
H
is the height of H relative to the surface of the water. Now, vertical
force balance requires that W =
0
V
0
. Thus, the potential energy of the system is W z
GH
(modulo an arbitrary additive
constant), where z
GH
= z
G
z
H
is the height of the center of gravity relative to the center of buoyancy.
According to the principles of statics, an equilibrium state corresponds to either a minimum or a maximum of the
potential energy. However, such an equilibrium is only stable when the potential energy is minimized. Thus, it follows
that a stable equilibrium conguration of a oating body is such as to minimize the height of the bodys center of
gravity relative to its center of buoyancy.
3.9 Curve of Buoyancy
Consider a oating body in vertical force balance that is slowly rotated about a horizontal axis normal to one of its
vertical symmetry planes. Let us take the center of gravity, G, which necessarily lies in this plane, as the origin of a
coordinate system that is xed with respect to the body. As illustrated in Figure 3.2, as the body rotates, the locus of its
center of buoyancy, H, as seen in the xed reference frame, appears to traces out a curve, AB, in the plane of symmetry.
This curve is known as the curve of buoyancy. Let r represent the radial distance from the origin, G, to some point,
H, on the curve of buoyancy. Note that the tangent to the curve of buoyancy is always orientated horizontally. This
follows because, as was shown in the previous section, small rotations of a oating body in vertical force balance cause
its center of buoyancy to shift horizontally, rather than vertically, in the plane perpendicular to the axis of rotation.
Thus, the dierence in vertical height, z
GH
, between the center of gravity and the center of buoyancy is equal to the
perpendicular distance, p, between G and the tangent to the curve of buoyancy at H. An equilibrium conguration
therefore corresponds to a maximum or a minimum of p as point H moves along the curve of buoyancy. However,
the equilibrium is only stable if p is minimized. Now, if R is the radius of curvature of the curve of buoyancy then,
according to a standard result in dierential calculus,
R = r
dr
dp
. (3.50)
Writing this result in the form
r =
R
r
p, (3.51)
Hydrostatics 39
r
G
p
H
0
A
H
B
Figure 3.2: Curve of buoyancy for a oating body.
it can be seen that maxima and minima of p, which are the points on the curve of buoyancy where p = 0, correspond
to the points where r = 0, and are, thus, coincident with maxima and minima of r. In other words, an equilibrium
conguration corresponds to a point of maximum or minimum r on the curve of buoyancy: i.e., a point at which GH
meets the curve at right-angles. At such a point, r = p, and the potential energy consequently takes the value W r.
Let H
0
be a point on the curve of buoyancy, and let r
0
, p
0
, and R
0
be the corresponding values of r, p, and R. For
neighboring points on the curve, we can write
r r
0
=
dr
dp
H
0
(p p
0
), (3.52)
or
r r
0
=
R
0
r
0
(p p
0
). (3.53)
It follows that p p
0
has the same sign as r r
0
(since R
0
and r
0
are both positive). [The fact that R
0
is positive
(i.e., dr/dp > 0) follows from the previously established result that the metacenter, which is the center of curvature of
the curve of buoyancy, always lies above the center of buoyancy, implying that the curve of buoyancy is necessarily
concave upwards.] Hence, the minima and maxima of r occur simultaneously with those of p. Consequently, a stable
equilibrium conguration corresponds to a point of minimum r on the curve of buoyancy: i.e., a minimum in the
distance GH between the center of gravity and the center of buoyancy.
We can use the above result to determine the stable equilibrium congurations for a beam of square cross-section,
and uniform specic gravity s, that oats with its length horizontal. In order to achieve this goal, we must calculate
the distance GH for all possible congurations of the beam that are in vertical force balance. However, we need only
consider cases where s < 1/2, since, according to the analysis of Section 3.6, for every stable equilibriumconguration
with s = s
0
< 1/2 there is a corresponding stable inverted conguration with s = 1 s
0
> 1/2, and vice versa.
Let us dene xed rectangular axes, x and y, passing through the center of the middle section of the beam, and
running parallel to its sides. Let us start with the case where the waterline PQ is parallel to a side. See Figure 3.3. If
the length of a side is 2 a then (3.16) yields
AP = BQ = 2a s. (3.54)
40 FLUID MECHANICS
Q
A B
P
P
O
y
x
Q
Figure 3.3: Beam of square cross-section oating with two corners immersed.
Suppose that the beam is turned through an angle > 0 such that the waterline assumes the position P
, in
Figure 3.3, but still intersects two opposite sides. The lengths AP
and BQ
satisfy
BQ
AP
= 2 a tan . (3.55)
Moreover, the area of the trapezium P
ABQ
must match that of the rectangle PABQ in order to ensure that the
submerged volume remains invariant (otherwise, the beam would not remain in vertical force balance): i.e.,
(AP
+ BQ
) a = 4 a
2
s. (3.56)
It follows that
AP
= a (2 s tan ), (3.57)
BQ
= a (2 s + tan ). (3.58)
The constraint that the waterline intersect two opposite sides of the beam implies that AP
> 0, and, hence, that

tan < 2 s. (3.59)
The coordinates of the center of buoyancy, H, which is the mean center of the trapezium P
ABQ
, are
x =
_
a
a
_
a
ah(x)
x dx dy
_
a
a
_
a
ah(x)
dx dy
=
(2/3) a
3
tan
4 a
2
s
=
a
6 s
tan , (3.60)
y =
_
a
a
_
a
ah(x)
y dx dy
_
a
a
_
a
ah(x)
dx dy
=
4 a
3
s (1 s) (1/3) a
3
tan
2
4 a
2
s
= (1 s) a
a
12 s
tan
2
, (3.61)
where
h(x) = 2 a s + x tan . (3.62)
Hydrostatics 41
A B
O
y
x
P
Figure 3.4: Beam of square cross-section oating with one corner immersed.
Thus, if u = r
2
/a
2
= ( x
2
+ y
2
)/a
2
then
u =
t
2
36 s
2
+
_
(1 s)
t
2
12 s
_
2
, (3.63)
where t = tan . Now, a stable equilibriumstate corresponds to a minimum of r with respect to , and, hence, of u with
respect to t. However,
du
dt
=
t
36 s
2
_
t
2
12 s (1 s) + 2
_
, (3.64)
d
2
u
dt
2
=
1
36 s
2
_
3 t
2
12 s (1 s) + 2
_
. (3.65)
The minima and maxima of u occur when du/dt = 0, d
2
u/dt
2
> 0 and du/dt = 0, d
2
u/dt
2
< 0, respectively. It follows
that the symmetrical position, t = 0, in which the sides of the beam are either parallel or perpendicular to the waterline,
is always an equilibrium, but is only stable when
s
2
s +
1
6
> 0 : (3.66)
i.e., when s < 1/2 1/
12 = 0.2113. It is also possible to obtain equilibria in asymmetric positions such that t is the
root of
t
2
= 12s (1 s) 2. (3.67)
Such equilibria only exist for s > 0.2113, and are stable. Finally, in order to satisfy the constraint (3.59), we must have
t < 2 s, which, in combination with the above equation, implies that
8 s
2
6 s + 1 > 0, (3.68)
or s < 0.25.
Suppose that the constraint (3.59) is not satised, so that the immersed portion of the beams cross-section is
triangular. See Figure 3.4. It is clear that
BQ
BP
= tan . (3.69)
42 FLUID MECHANICS
Moreover, the area of the triangle P
BQ
, in Figure 3.4, must match that of the rectangle PABQ, in Figure 3.3, in order
to ensure that the submerged volume remain invariant: i.e.,
1
2
BP
BQ
= 4 a
2
s. (3.70)
It follows that
BP
= (8 s/ tan )
1/2
a, (3.71)
BQ
= (8 s tan )
1/2
a, (3.72)
or, writing z
2
= tan and
2
= (8/9) s,
BP
= 3 z
1
a, (3.73)
BQ
= 3 z a. (3.74)
The coordinates of the center of buoyancy, H, which is the mean center of triangle P
BQ
, are
x = a BP
/3 = a (1 z
1
), (3.75)
y = a BQ
/3 = a (1 z), (3.76)
since the perpendicular distance of the mean center of a triangle from one of its sides is one third of the perpendicular
distance from the side to the opposite vertex. Thus, if u = r
2
/a
2
= ( x
2
+ y
2
)/a
2
then
u = (1 z
1
)
2
+ (1 z)
2
, (3.77)
du
dz
=
2
2
(z
2
1) (z
2

1
z + 1)
z
3
, (3.78)
d
2
u
dz
2
=
2
2
(z
4
2
1
z + 3)
z
4
. (3.79)
Moreover, the constraint (3.59) yields
z >
3
2
. (3.80)
The stable and unstable equilibria correspond to du/dz = 0, d
2
u/dz
2
> 0 and du/dz, d
2
u/dz
2
< 0, respectively. It
follows that the symmetrical position, z = 1, in which the diagonals of the beam are either parallel or perpendicular to
the waterline is an equilibriumprovided < 2/3, or s < 1/2, but is only stable when > 1/2, or s > 9/32 = 0.28125.
It is also possible to obtain equilibria in asymmetric positions such that z is the root of
z
2

1
z + 1 = 0. (3.81)
Such equilibria only exist for

2/3 < < 1/2, or 1/4 < s < 9/32, and are stable.
In summary, the stable equilibrium congurations of a beam of square cross-section, oating with its length hori-
zontal, are such that the sides are either parallel or perpendicular to the waterline for s < 0.2113, such that two corners
are immersed but the sides and diagonals are neither parallel nor perpendicular to the waterline for 0.2113 < s < 0.25,
such that only one corner is immersed but the sides and diagonals are neither parallel nor perpendicular to the wa-
terline for 0.25 < s < 0.28125, and such that the diagonals are either parallel or perpendicular to waterline for
0.28125 < s < 0.5. For s > 0.5, the stable congurations are the same as those for a beam with the complimentary
specic gravity 1 s.
3.10 Rotational Hydrostatics
Consider the equilibrium of an incompressible uid that is uniformly rotating at a xed angular velocity in some
inertial frame of reference. Of course, such a uid appears stationary in a non-inertial co-rotating reference frame.
Hydrostatics 43
Moreover, according to standard Newtonian dynamics, the force balance equation for the uid in the co-rotating frame
takes the form (cf., Section 3.2)
0 = p + + ( r), (3.82)
where p is the static uid pressure, the mass density, the gravitational potential energy per unit mass, and r a
position vector (measured with respect to an origin that lies on the axis of rotation). The nal term on the right-hand
side of the above equation represents the ctitious centrifugal force density. Without loss of generality, we can assume
that = e
z
. It follows that
0 = p + ( +
), (3.83)
where
=
1
2

2
(x
2
+ y
2
) (3.84)
is the so-called centrifugal potential. Recall, incidentally, that is a uniform constant in an incompressible uid.
As an example, consider the equilibrium of a body of water, located on the Earths surface, that is uniformly
rotating about a vertical axis at the xed angular velocity . It is convenient to adopt cylindrical coordinates (see
Section C.3), r, , z, whose symmetry axis coincides with the axis of rotation. Let z increase upward. It follows that
= g z and
= (1/2)
2
r
2
. Assuming that the pressure distribution is axisymmetric, so that p = p(r, z), the force
balance equation, (3.83), reduces to
p
r
+
( +
)
r
= 0, (3.85)
p
z
+
( +
)
z
= 0, (3.86)
or
p
r

2
r = 0, (3.87)
p
z
+ g = 0. (3.88)
The previous two equations can be integrated to give
p(r, z) = p
0
+
_
1
2

2
r
2
g z
_
, (3.89)
where p
0
is a constant. Thus, constant pressure surfaces in a uniformly rotating body of water take the form of
paraboloids of revolution about the rotation axis. Suppose that p
0
represents atmospheric pressure. In this case, the
surface of the water is the locus of p(r, z) = p
0
: i.e., it is the constant pressure surface whose pressure matches that of
the atmosphere. It follows that the surface of the water is the paraboloid of revolution
z =
2
2 g
r
2
, (3.90)
where r is the perpendicular distance from the axis of rotation, and z = 0 the on-axis height of the surface.
Now, from Section 3.3, it is plain that the buoyancy force acting on any co-rotating solid body, which is wholly
or partially immersed in the water, is the same as that which would maintain the mass of water displaced by the body
in relative equilibrium. In the case of a oating body, this mass is limited by the continuation of the waters curved
surface through the body. Let points G and H represent the centers of gravity and buoyancy, respectively, of the body.
Of course, the latter point is simply the center of gravity of the displaced water. Suppose that G and H are located
perpendicular distances r
G
and r
H
from the axis of rotation, respectively. Finally, let M be the mass of the body,
and M
0
the mass of the displaced water. It follows that the buoyancy force has an upward vertical component M
0
g,
and an outward horizontal component M
0

2
r
H
. Thus, according to standard Newtonian dynamics, the equation of
horizontal motion of a general co-rotating body is
M (
..
r
2
r
G
) = M
0

2
r
H
, (3.91)
44 FLUID MECHANICS
where
.
= d/dt. Now, from Archimedes principle, M
0
= M for the case of a oating body that is less dense than
water. However, if the body is of uniform density then r
H
> r
G
, as a consequence of the curvature of the waters
surface. Hence, we obtain
..
r =
2
(r
H
r
G
) < 0. (3.92)
In other words, a oating body drifts radially inward towards the rotation axis. On the other hand, M
0
< M for a fully
submerged body that is more dense than water. However, if the body is of uniform density then its centers of gravity
and buoyancy coincide with one another, so that r
H
= r
G
. Hence, we obtain
..
r = (M M
0
)
2
r
G
> 0. (3.93)
In other words, a fully submerged body drifts radially outward from the rotation axis. The above analysis accounts for
the common observation that objects heavier than water, such as grains of sand, tend to collect on the outer side of a
bend in a fast owing river, whilst oating objects, such as sticks, tend to collect on the inner side.
3.11 Equilibrium of a Rotating Liquid Body
Consider a self-gravitating liquid body in outer space that is rotating uniformly about some xed axis passing through
its center of mass. What is the shape of the bodys bounding surface? This famous theoretical problem had its origins
in investigations of the gure of a rotating planet, such as the Earth, that were undertaken by Newton, Maclaurin,
Jacobi, Meyer, Liouville, Dirichlet, Dedekind, Riemann, and other celebrated scientists, in the 17th, 18th, and 19th
centuries.
1
Incidentally, it is reasonable to treat the Earth as a liquid, for the purpose of this calculation, because the
shear strength of the solid rock out of which the terrestrial crust is composed is nowhere near sucient to allow the
actual shape of the Earth to deviate signicantly from that of a hypothetical liquid Earth.
Now, in a co-rotating reference frame, the shape of a self-gravitating, rotating, liquid planet is determined by a
competition between uid pressure, gravity, and the ctitious centrifugal force. The latter force opposes gravity in the
plane perpendicular to the axis of rotation. Of course, in the absence of rotation, the planet would be spherical. Thus,
we would expect rotation to cause the planet to expand in the plane perpendicular to the rotation axis, and to contract
along the rotation axis (in order to conserve volume).
For the sake of simplicity, we shall restrict our investigation to a rotating planet of uniform density whose outer
boundary is ellipsoidal. Now, an ellipsoid is the three-dimensional generalization of an ellipse. Let us adopt the right-
handed Cartesian coordinate system x
1
, x
2
, x
3
. An ellipse whose principal axes are aligned along the x
1
- and x
2
-axes
satises
x
2
1
a
2
1
+
x
2
2
a
2
2
= 1, (3.94)
where a
1
and a
2
are the corresponding principal radii. Moreover, as is easily demonstrated,
A =
_
dA = a
1
a
2
, (3.95)
_
x
2
i
dA =
1
4
a
2
i
A, (3.96)
_
x
1
x
2
dA = 0, (3.97)
where A is the area, dA an element of A, and the integrals are taken over the whole interior of the ellipse. Likewise, an
ellipsoid whose principal axes are aligned along the x
1
-, x
2
-, and x
3
-axes satises
x
2
1
a
2
1
+
x
2
2
a
2
2
+
x
2
3
a
2
3
= 1, (3.98)
1
See Ellipsoidal Figures of Equilibrium, S. Chandrasekhar (Yale University Press, New Haven CT, 1969).
Hydrostatics 45
where a
1
, a
2
, and a
3
are the corresponding principal radii. Moreover, as is easily demonstrated,
V =
_
dV =
4
3
a
1
a
2
a
3
, (3.99)
_
x
2
i
dV =
1
5
a
2
i
V, (3.100)
_
x
1
x
2
dV =
_
x
2
x
3
dV = 0, (3.101)
where V is the volume, dV an element of V, and the integrals are taken over the whole interior of the ellipsoid.
Suppose that the planet is rotating uniformly about the x
3
-axis at the xed angular velocity . The planets moment
of inertia about this axis is [cf., Equation (3.100)]
I
33
=
1
5
M (a
2
1
+ a
2
2
), (3.102)
where M is its mass. Thus, the planets angular momentum is
L = I
33
=
1
5
M (a
2
1
+ a
2
2
) , (3.103)
and its rotational kinetic energy becomes
K =
1
2
I
33

2
=
1
10
M (a
2
1
+ a
2
2
)
2
. (3.104)
According to Equations (3.83) and (3.84), the uid pressure distribution within the planet takes the form
p = p
0

_

1
2

2
(x
2
1
+ x
2
2
)
_
, (3.105)
where is the gravitational potential (i.e., the gravitational potential energy of a unit test mass) due to the planet,
= M/V the uniform planetary mass density, and p
0
a constant. However, it is demonstrated in Appendix E that the
gravitational potential inside a homogeneous self-gravitating ellipsoidal body can be written
=
3
4
G M
_
i=1,3
i
x
2
i
_
_
, (3.106)
where G is the gravitational constant, and
0
=
_

0
du
, (3.107)
i
=
_

0
du
(a
2
i
+ u)
, (3.108)
= (a
2
1
+ u)
1/2
(a
2
2
+ u)
1/2
(a
2
3
+ u)
1/2
. (3.109)
Thus, we obtain
p = p
0

1
2

__
3
2
G M
1

2
_
x
2
1
+
_
3
2
G M
2

2
_
x
2
2
+
3
2
G M
3
x
2
3
_
, (3.110)
where p
0
is the central uid pressure. Now, the pressure at the planets outer boundary must be zero, otherwise there
would be a force imbalance across the boundary. In other words, we require
1
2

__
3
2
G M
1

2
_
x
2
1
+
_
3
2
G M
2

2
_
x
2
2
+
3
2
G M
3
x
2
3
_
= p
0
, (3.111)
46 FLUID MECHANICS
whenever
x
2
1
a
2
1
+
x
2
2
a
2
2
+
x
2
3
a
2
3
= 1. (3.112)
The previous two equations can only be simultaneously satised if
_
1

2
(3/2) G M
_
a
2
1
=
_
2

2
(3/2) G M
_
a
2
2
=
3
a
2
3
. (3.113)
Rearranging the above expression, we obtain
2
2G
=
a
1
a
3
a
2
(a
2
2
a
2
3
)
_

0
u du
(a
2
2
+ u) (a
2
3
+ u)
, (3.114)
subject to the constraint
(a
2
1
a
2
2
)
_

0
_
_
a
2
1
a
2
2
(a
2
1
+ u) (a
2
2
+ u)

a
2
3
(a
2
3
+ u)
_
_
du
= 0, (3.115)
where use has been made of Equation (3.99).
Finally, according to Section E, the net gravitational potential energy of the planet is
U =
3
10
G M
2
0
. (3.116)
Hence, the bodys total mechanical energy becomes
E = K + U =
1
10
M (a
2
1
+ a
2
2
)
2
3
10
G M
2
0
. (3.117)
3.12 Maclaurin Spheroids
One, fairly obvious, way in which the constraint (3.115) can be satised is if a
2
= a
1
. In other words, if the planet
is rotationally symmetric about its axis of rotation. Now, an ellipsoid that is rotationally symmetric about a principal
axisor, equivalently, an ellipsoid with two equal principal radiiis known as a spheroid. In fact, if a
2
= a
1
then
the cross-section of the planets outer boundary in any plane passing though the x
3
-axis is an ellipse of major radius
a
1
, in the direction perpendicular to the x
3
-axis, and minor radius a
3
, in the direction parallel to the x
3
-axis. Here,
we are assuming that a
1
> a
3
: i.e., that the planet is attened along its axis of rotation. The degree of attening is
conveniently measured by the eccentricity,
e
13
(1 a
2
3
/a
2
1
)
1/2
. (3.118)
Thus, if e
13
= 0 then there is no attening, and the planet is consequently spherical, whereas if e
13
1 then the
attening is complete, and the planet consequently collapses to a disk in the x
1
-x
2
plane.
Let u = a
2
1
and = e
2
13
/z
2
1. Setting a
2
= a
1
in Equation (3.114), we obtain
2
2G
= (1 e
2
13
)
1/2
e
2
13
_

0
d
(1 + )
2
(1 + e
2
13
)
3/2
=
2 (1 e
2
13
)
1/2
e
3
13
__
e
13
0
z
2
dz
(1 z
2
)
1/2
(1 e
2
13
)
_
e
13
0
z
2
dz
(1 z
2
)
3/2
_
. (3.119)
Performing the integrals, which are standard,
2
we nd that
2
2G
=
_
_
3 2 e
2
13
e
3
13
_
_
(1 e
2
13
)
1/2
sin
1
e
13

_
_
3
e
2
13
_
_
(1 e
2
13
). (3.120)
2
See Schaums Mathematical Handbook of Formulas and Tables, 2nd Edition, Murray R. Spiegel, (Mc-Graw Hill, New York NY, 1998).
Hydrostatics 47
e
13

L
E e
13

L
E
0.00 0.00000 0.00000 0.60000 0.60 0.31729 0.18037 0.56233
0.05 0.02582 0.01266 0.59980 0.65 0.34484 0.20286 0.55320
0.10 0.05168 0.02540 0.59919 0.70 0.37239 0.22834 0.54200
0.15 0.07758 0.03830 0.59817 0.75 0.39967 0.25792 0.52800
0.20 0.10357 0.05144 0.59672 0.80 0.42612 0.29345 0.51001
0.25 0.12967 0.06491 0.59479 0.85 0.45046 0.33833 0.48587
0.30 0.15591 0.07882 0.59236 0.90 0.46932 0.39994 0.45107
0.35 0.18231 0.09329 0.58936 0.95 0.47045 0.50074 0.39272
0.40 0.20889 0.10846 0.58572 0.96 0.46472 0.53194 0.37485
0.45 0.23567 0.12450 0.58135 0.97 0.45418 0.57123 0.35273
0.50 0.26267 0.14163 0.57612 0.98 0.43475 0.62486 0.32351
0.55 0.28989 0.16013 0.56986 0.99 0.39389 0.71209 0.27916
Table 3.1: Properties of the Maclaurin spheroids.
This famous result was rst obtained by Colin Maclaurin in 1742. Finally, in order to calculate the potential energy,
(3.116), we need to evaluate
0
=
1
a
1
_

0
d
(1 + ) (1 + e
2
13
)
1/2
=
2
a
1
e
13
_
e
13
0
dz
(1 z
2
)
1/2
=
2
a
1
sin
1
e
13
e
13
. (3.121)
Let e
13
= sin . Thus, = 0 corresponds to no rotational attening, and = /2 to complete attening. Moreover,
a
1
= a
0
(cos )
1/3
and a
3
= a
0
(cos )
2/3
, where a
0
= (a
1
a
2
a
3
)
1/3
= (3 V/4)
1/3
is the mean radius. It is also helpful
to dene = /(2G)
1/2
,

L = L/(G M
3
a
0
)
1/2
, and

E = E/(G M
2
/a
0
). The above analysis leads to the following
set of equations which specify the properties of the so-called Maclaurin spheroids:
2
=
cos
sin
2
_
(1 + 2 cos
2
)

sin
3 cos
_
, (3.122)
L =
6
5
(cos )
1/6
sin
_
(1 + 2 cos
2
)

sin
3 cos
_
1/2
, (3.123)
E =
3
10
(cos )
1/3
sin
2
_
(1 4 cos
2
)

sin
+ 3 cos
_
. (3.124)
These properties are set out in Table 3.1.
In the limit, 0, in which the planet is relatively slowly rotating (i.e., 1), and its degree of attening
consequently slight, Equations (3.122)(3.124) reduce to
e
13

15
2
, (3.125)
6
5
, (3.126)
E
3
5
. (3.127)
In other words, in the limit of relatively slow rotation, when the planet is almost spherical, its eccentricity becomes
directly proportional to its angular velocity. In this case, it is more conventional to parameterize angular velocity in
terms of
m =
2
a
0
g
0
=
3
2

2
, (3.128)
48 FLUID MECHANICS
0
0.1
0.2

2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
e
13
Figure 3.5: Normalized angular velocity squared of a Maclaurin spheroid (solid) and a Jacobi ellipsoid (dashed)
versus the eccentricity e
13
in the x
1
-x
3
plane.
where g
0
= G M
2
/a
2
0
is the mean surface gravitational acceleration. Furthermore, the degree of rotational attening is
more conveniently expressed in terms of the ellipticity,
=
a
1
a
3
a
0
e
2
13
2
. (3.129)
Thus, it follows from (3.125) that

5
4
m. (3.130)
Now, for the case of the Earth ( = 7.27 10
5
rad. s
1
, a
0
= 6.37 10
6
m, g
0
= 9.81 ms
1
), we obtain
m
1
291
. (3.131)
Thus, it follows that, were the Earth homogeneous, its gure would be a spheroid, attened at the poles, of ellipticity

5
4
1
291

1
233
. (3.132)
This result was rst obtained by Newton. Now, the actual ellipticity of the Earth is about 1/294, which is substantially
smaller than Newtons prediction. The discrepancy is due to the fact that the Earth is strongly inhomogeneous, being
much denser at its core than in its outer regions.
Figures 3.5 and 3.6 illustrate the variation of the normalized angular velocity, , and angular momentum,

L,
of a Maclaurin spheroid with its eccentricity, e
13
, as predicted by Equations (3.122)(3.124). It can be seen, from
Figure 3.5, that there is a limit to how large the normalized angular velocity of such a spheroid can become. The
Hydrostatics 49
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
L
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
e
13
Figure 3.6: Normalized angular momentum of a Maclaurin spheroid (solid) and a Jacobi ellipsoid (dashed) versus the
eccentricity e
13
in the x
1
-x
3
plane.
limiting value corresponds to = 0.47399, and occurs when e
13
= 0.92995. For values of lying below0.47399 there
are two possible Maclaurin spheroids, one with an eccentricity less than 0.92995, and one with an eccentricity greater
than 0.92995. Note, however, fromFigure 3.6, that despite the fact that the angular velocity, , of a Maclaurin spheroid
varies in a non-monotonic manner with the eccentricity, e
13
, the angular momentum,
L, increases monotonically with

e
13
, becoming innite in the limit e
13
1. It follows that there is no upper limit to the angular momentum of a
Maclaurin spheroid.
3.13 Jacobi Ellipsoids
If a
2
a
1
(i.e., if the outer boundary of the rotating body is ellipsoidal, rather than spheroidal) then the constraint
(3.115) can only be satised when
_

0
_
_
a
2
1
a
2
2
(a
2
1
+ u) (a
2
2
+ u)

a
2
3
(a
2
3
+ u)
_
_
du
= 0. (3.133)
Without loss of generality, we can assume that a
1
a
2
a
3
. Let
a
2
= a
1
cos , (3.134)
a
3
= a
1
cos , (3.135)
where . It follows that the cross-sections of the planets outer boundary in the x
1
-x
2
and x
1
-x
3
planes are ellipses
of eccentricities
e
12
= (1 a
2
3
/a
2
2
)
1/2
= sin , (3.136)
e
13
= (1 a
2
3
/a
2
1
)
1/2
= sin , (3.137)
50 FLUID MECHANICS
respectively. It is also helpful to dene
= sin
1
(sin / sin ). (3.138)
Let sin
2
= [a
2
1
/(a
2
1
+ u)] sin
2
. Here, u = 0 corresponds to = , and u = to = 0. Equations (3.115) and
(3.114) transform to
3
E(, ) 2 F(, ) +
_
1 + (sin tan cos )
2
cos
2
_
E(, )
sin tan cos (1 + sin

2
)
cos
2
= 0, (3.139)
and
2
= 2
_
F(, ) E(, )
tan sin tan
+
cos E(, )
tan
3
cos
2
cos
2
tan
2
cos
2
_
, (3.140)
respectively, where
E(, ) =
_

0
(1 sin
2
sin
2
)
1/2
d, (3.141)
F(, ) =
_

0
d
(1 sin
2
sin
2
)
1/2
, (3.142)
are special functions known as incomplete elliptic integrals.
4
The integral
0
, dened in (3.107), transforms to
0
=
2 (cos cos )
1/3
a
0
sin
F(, ). (3.143)
Finally, making use of some of the analysis in the previous two sections, the normalized angular momentum, and
normalized mechanical energy, of the planet can be written
L = =
6
10
1 + cos
2
(cos cos )
2/3
, (3.144)
E =
3
5
(cos cos )
1/3
sin
F(, ) +
3
20
1 + cos
2
(cos cos )
2/3

2
, (3.145)
respectively.
Now, the constraint (3.139) is obviously satised in the limit 0, since this implies that 0 and E(, ),
F(, ) . Of course, this limit corresponds to the axisymmetric Maclaurin spheroids discussed in the previous sec-
tion. Jacobi, in 1834, was the rst researcher to obtain the very surprising result that (3.139) also has non-axisymmetric
ellipsoidal solutions characterized by > 0. These solutions are known as the Jacobi ellipsoids in his honor. The
properties of the Jacobi ellipsoids, as determined from Equations (3.139), (3.140), (3.144), and (3.145), are set out in
Table 3.2, and illustrated in Figures 3.5 and 3.6. It can be seen that the sequence of Jacobi ellipsoids bifurcates fromthe
sequence of Maclaurin spheroids when e
13
= 0.81267. Moreover, there are no Jacobi ellipsoids with e
13
< 0.81267.
However, as e
13
increases above this critical value, the eccentricity, e
12
, of the Jacobi ellipsoids in the x
1
-x
2
plane
grows rapidly, approaching unity as e
13
approaches unity. Thus, in the limit e
13
1, in which a Maclaurin spheroid
collapses to a disk in the x
1
-x
2
plane, a Jacobi ellipsoid collapses to a line running along the x
1
-axis. Note, from
Figures 3.5 and 3.6, that, at xed e
13
, the Jacobi ellipsoids have lower angular velocity and angular momentum than
Maclaurin spheroids (with the same mass and volume). Furthermore, as is the case for a Maclaurin spheroid, there is a
maximum angular velocity that a Jacobi ellipsoid can have (i.e., = 0.43257), but no maximum angular momentum.
Figure 3.7 shows the mechanical energy of the Maclaurin spheroids and Jacobi ellipsoids plotted as a function of
their angular momentum. It can be seen that the Jacobi ellipsoid with a given angular momentum has a lower energy
3
See On Jacobis Figure of Equilibrium for a Rotating Mass of Fluid, G.H. Darwin, Proc. Roy. Soc. London 41, 319 (1886).
4
See Handbook of Mathematical Functions, M. Abramowitz, and I.A. Stegun (Dover, New York NY, 1965).
Hydrostatics 51
e
12
e
13

L
E e
12
e
13

L
E
0.00 0.81267 0.43257 0.30375 0.50452 0.60 0.85585 0.42827 0.30984 0.50138
0.05 0.81293 0.43257 0.30375 0.50459 0.65 0.86480 0.42609 0.31296 0.49975
0.10 0.81372 0.43257 0.30375 0.50459 0.70 0.87510 0.42288 0.31760 0.49734
0.15 0.81504 0.43256 0.30377 0.50458 0.75 0.88705 0.41807 0.32462 0.49372
0.20 0.81691 0.43253 0.30380 0.50457 0.80 0.90102 0.41069 0.33562 0.48814
0.25 0.81934 0.43248 0.30388 0.50453 0.85 0.91761 0.39879 0.35390 0.47908
0.30 0.82237 0.43237 0.30402 0.50445 0.90 0.93778 0.37787 0.38783 0.46295
0.35 0.82603 0.43220 0.30427 0.50432 0.95 0.96340 0.33353 0.46860 0.42782
0.40 0.83037 0.43191 0.30468 0.50410 0.96 0.96950 0.31776 0.50078 0.41499
0.45 0.83544 0.43146 0.30532 0.50376 0.97 0.97605 0.29691 0.54672 0.39771
0.50 0.84131 0.43078 0.30628 0.50326 0.98 0.98317 0.26722 0.62003 0.37241
0.55 0.84808 0.42976 0.30772 0.50250 0.99 0.99101 0.21809 0.76872 0.32842
Table 3.2: Properties of the Jacobi ellipsoids.
that the corresponding Maclaurin spheroid (i.e., the spheroid with the same angular momentum, mass, and volume).
This is signicant because, in the presence of a small amount of dissipation (i.e., viscosity), we would generally expect
an isolated uid system to slowly evolve toward the equilibrium state with the lowest energy, subject to any global
constraints on the system. For the case of a weakly viscous, isolated, rotating, liquid planet, the relevant constraints are
that the mass, volume, and net angular momentumof the system cannot spontaneously change. Thus, we expect such a
planet to evolve toward the equilibriumstate with the lowest energy for a given mass, volume, and angular momentum.
This suggests, from Figure 3.7, that at relatively high angular momentum (i.e.,

L > 0.30375, e
13
> 0.81267), when
the Jacobi ellipsoid solutions exist, they are stable equilibrium states (since there is no lower energy state to which
the system can evolve), whereas the Maclaurin spheroids are unstable. On the other hand, at relatively low angular
momentum (i.e.,

L < 0.30375, e
13
< 0.81267), when there are no Jacobi ellipsoid solutions, the Maclaurin spheroids
are stable equilibriumstates (again, because there is no lower energy state to which they can evolve). These predictions
are borne by the results of direct stability analysis performed on the Maclaurin spheroids and Jacobi ellipsoids.
5
In
fact, such stability studies demonstrate that the Maclaurin spheroids are unstable in the presence of weak dissipation
for e
13
> 0.81267, and unconditionally unstable for e
13
> 0.95289. The Jacobi ellipsoids, on the other hand, are
unconditionally stable for e
13
< 0.93858, but are unconditionally unstable for e
13
> 0.93858, evolving toward lower
energy pear shaped equilibria (which are, themselves, unstable in the presence of weak dissipation).
3.14 Roche Ellipsoids
Consider a homogeneous liquid moon of mass M which is in a circular orbit of radius R about a planet of mass
M
. Let C, C
, and C
be the center of the moon, the center of the planet, and the center of mass of the moon-planet
system, respectively. As is easily demonstrated, all three points lie on the same straight-line, and the distances between
them take the constant values CC
= R and CC
= [M
/(M + M
)] R. Moreover, according to standard Newtonian

dynamics, there exists an inertial frame of reference in which C
is stationary, and the line CC
rotates at the xed

angular velocity , where
2
=
G(M + M
)
R
3
. (3.146)
In other words, in the inertial frame, the moon and the planet orbit in a xed plane about their common center of mass
at the angular velocity . It is convenient to transformto a non-inertial reference frame that rotates (with respect to the
inertial frame), about an axis passing through C
, at the angular velocity . It follows that points C, C
, and C
appear
stationary in this frame. It is also convenient to adopt the standard right-handed Cartesian coordinates, x
1
, x
2
, x
3
, and
to choose the coordinate axes such that = e
3
, C = (0, 0, 0), C
= (R, 0, 0), and C
= ([M
/(M + M
)] R, 0, 0).
5
See Ellipsoidal Figures of Equilibrium, S. Chandrasekhar (Yale University Press, New Haven CT, 1969).
52 FLUID MECHANICS
0
0.1
0.2
0.3
0.4
0.5
0.6
E
0 0.2 0.4 0.6 0.8 1 1.2
L
Figure 3.7: Normalized mechanical energy of a Maclaurin spheroid (solid) and a Jacobi ellipsoid (dashed) versus the
normalized angular momentum.
Thus, in the non-inertial reference frame, the orbital rotation axis runs parallel to the x
3
-axis, and the centers of the
moon and the planet both lie on the x
1
-axis.
Suppose that the moon does not rotate (about an axis passing through its center of mass) in the non-inertial refer-
ence frame. This implies that, in the inertial frame, the moon appears to rotate about an axis parallel to the x
3
-axis, and
passing through C, at the same angular velocity as it orbits about C
. This type of rotation is termed synchronous, and

ensures that the same hemisphere of the moon is always directed toward the planet. Such rotation is fairly common in
the Solar System. For instance, the Moon rotates synchronously in such a manner that the same hemisphere is always
visible from the Earth. Synchronous rotation in the Solar System is a consequence of process known as tidal locking.
Since a synchronously rotating moon is completely stationary in the aforementioned non-inertial frame, its internal
pressure, p, is governed by a force balance equation of the form [cf., Equation (3.83)]
0 = p + ( +
), (3.147)
where is the uniform internal mass density, the gravitational potential due to the moon,
the gravitational
potential due to the planet, and
=
1
2

2
_
_
_
x
1

M
M + M
R
_
2
+ x
2
2
_
_
(3.148)
the centrifugal potential due to the fact that the non-inertial frame is rotating (about an axis parallel to the x
3
-axis and
passing through point C
) at the angular velocity [cf., Equation (3.84)]. Suppose that the moon is much less massive
that the planet (i.e., M/M
1). In this limit, the centrifugal potential (3.148) reduces to
_
1
2

x
1
R
+
(1/2) x
2
1
+ (1/2) x
2
2
R
2
_
_
, (3.149)
Hydrostatics 53
where use has been made of (3.146).
Suppose that the planet is spherical. It follows that the potential
is the same as that which would be generated

by a point mass M
located at C
. In other words,
=
G M
R
_
_
1 2
x
1
R
+
x
2
1
+ x
2
2
+ x
2
3
R
2
_
_
1/2

G M
R
_
_
1 +
x
1
R
+
x
2
1
(1/2) x
2
2
(1/2) x
2
3
R
2
+
_
_
, (3.150)
where we have expanded up to second order in x
1
/R, etc.
The previous two equations can be combined to give

_
3
2
x
2
1

1
2
x
2
3
_
, (3.151)
where
=
G M
R
3
, (3.152)
and any constant terms have been neglected. Thus, the net force eld experienced by the moon due to the combined
action of the ctitious centrifugal force and the gravitational force eld of the planet is
(
) = (3 x
1
, 0, x
3
). (3.153)
The above type of force eld is known as a tidal force eld, and clearly acts to elongate the moon along the axis
joining the centers of the moon and planet (i.e., the x
1
-axis), and to compress it along the orbital rotation axis (i.e., the
x
3
-axis). Moreover, the magnitude of the tidal force increases linearly with distance from the center of the moon. The
tidal force eld is a consequence of the dierent spatial variation of the centrifugal force and the planets gravitational
force of attraction. This dierent variation causes these two forces, which balance one another at the center of the
moon, to not balance away from the center. As a result of the tidal force eld, we expect the shape of the moon to
be distorted from a sphere. Of course, the moon also generates a tidal force eld that acts to distort the shape of the
planet. However, we are assuming that the tidal distortion of the planet is much smaller than that of the moon (which
justies our earlier statement that the planet is essentially spherical). As will be demonstrated later, this assumption
is reasonable provided the mass of the moon is much less than that of the planet (assuming that the planet and moon
have similar densities).
Suppose that the bounding surface of the moon is the ellipsoid
x
2
1
a
2
1
+
x
2
2
a
2
2
+
x
2
3
a
2
3
= 1, (3.154)
where a
1
a
2
a
3
. It follows, from Section E, that the gravitational potential of the moon at an interior point can be
written
=
3
4
G M
_
i=1,3
i
x
2
i
_
_
, (3.155)
where the integrals
i
, for i = 0, 3, are dened in Equations (E.30) and (E.31). Hence, from (3.147) and (3.151), the
pressure distribution within the moon is given by
p = p
0

1
2

__
3
2
G M
1
3
_
x
2
1
+
3
2
G M
2
x
2
2
+
_
3
2
G M
3
+
_
x
2
3
_
, (3.156)
where p
0
is the central pressure. Now, the pressure must be zero on the moons bounding surface, otherwise this
surface would not be in equilibrium. Thus, in order to achieve equilibrium, we require
1
2

__
3
2
G M
1
3
_
x
2
1
+
3
2
G M
2
x
2
2
+
_
3
2
G M
3
+
_
x
2
3
_
= p
0
, (3.157)
54 FLUID MECHANICS
0
0.01
0.02
0.03
0.04
0.05
e
1
3
e
1
2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
e
13
Figure 3.8: Properties of the Roche ellipsoids.
whenever
x
2
1
a
2
1
+
x
2
2
a
2
2
+
x
2
3
a
2
3
= 1. (3.158)
The previous two equations can only be simultaneously satised if
_
1

3
(3/2) G M
_
a
2
1
=
2
a
2
2
=
_
3
+
(3/2) G M
_
a
2
3
. (3.159)
Let a
2
= a
1
cos and a
3
= a
1
cos , where . It is also helpful to dene = sin
1
(sin / sin ). With the help
of some of the analysis presented in the previous section, the integrals
i
, for i = 1, 3, can be shown to take the form
1
=
2
a
3
1
sin
3
F(, ) E(, )
sin
2
, (3.160)
2
=
2
a
2
1
sin
3
_
E(, )
sin
2
cos
2
F(, )
sin
2
cos sin
cos
2
cos
_
, (3.161)
3
=
2
a
2
1
sin
3
E(, )
cos
2
+
cos sin
cos
2
cos
_
, (3.162)
where the incomplete elliptic integrals E(, ) and F(, ) are dened in Equations (3.141) and (3.142), respectively.
Thus, (3.159) yields
=
1
sin tan tan
_
F(, ) (1 + cos
2
) E(, )
_
1 +
cos
2
cos
2
_
+
sin sin cos cos
cos
2
_
, (3.163)
Hydrostatics 55
e
12
e
13
e
12
e
13

0.00 0.00000 0.00000 0.52 0.56740 0.33440
0.04 0.04613 0.00213 0.56 0.60632 0.38204
0.08 0.09223 0.00852 0.60 0.64445 0.43094
0.12 0.13809 0.01913 0.64 0.68182 0.48027
0.16 0.18364 0.03392 0.68 0.71848 0.52890
0.20 0.22879 0.05282 0.72 0.75446 0.57532
0.24 0.27346 0.07573 0.76 0.78984 0.61729
0.28 0.31756 0.10253 0.80 0.82472 0.65150
0.32 0.36104 0.13308 0.84 0.85923 0.67265
0.36 0.40383 0.16721 0.88 0.89353 0.67151
0.40 0.44588 0.20470 0.92 0.92793 0.62978
0.44 0.48718 0.24528 0.96 0.96294 0.50135
0.48 0.52769 0.28865 1.00 1.00000 0.00000
Table 3.3: Properties of the Roche ellipsoids.
subject to the constraint
0 = cos
2
_
F(, ) 2 E(, ) + E(, )
cos
2
cos
2
sin sin cos cos

cos
2
_
+(3 + cos
2
)
_
E(, ) + [F(, ) cos
2
2 E(, )]
cos
2
cos
2
+
2 sin sin cos cos
cos
2
_
, (3.164)
where
=
M
M
a
3
0
R
3
, (3.165)
and a
0
= (a
1
a
2
a
3
)
1/3
is the mean radius of the moon. The dimensionless parameter measures the strength of the
tidal distortion eld, generated by the planet, that acts on the moon. There is an analogous parameter,
=
M
M
0
3
R
3
, (3.166)
where a
0
is the mean radius of the planet, which measures the tidal distortion eld, generated by the moon, that acts
on the planet. Now, we previously assumed that the former distortion eld is much stronger than the latter, allowing
us to neglect the tidal distortion of the planet altogether, and so to treat it as a sphere. This assumption is only justied
if
, which implies that

M
M
, (3.167)
where = M/[(4/3) a
3
0
] and
= M
/[(4/3) a
0
3
] are the mean densities of the moon and the planet, respectively.
Assuming that these densities are similar, the above condition reduces to M M
, or, equivalently, a
0
< a
0
. In other
words, neglecting the tidal distortion of the planet, whilst retaining that of the moon, is generally only reasonable when
the mass of the moon is much less than that of the planet, as was previously assumed to be the case.
Equations (3.163) and (3.164), which describe the ellipsoidal equilibria of a synchronously rotating, relatively low
mass, liquid moon due to the tidal force eld of the planet about which it orbits, were rst obtained by Roche in 1850.
The properties of the so-called Roche ellipsoids are set out in Table 3.3, and Figures 3.8 and 3.9.
It can be seen, from Table 3.3 and Figure 3.8, that the eccentricity e
12
= sin of a Roche ellipsoid in the x
1
-x
2
plane is almost equal to its eccentricity e
13
= sin in the x
1
-x
3
plane. In other words, Roche ellipsoids are almost
56 FLUID MECHANICS
0
0.01
0.02
0.03
0.04
0.05
0.06
0.07
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

e
13
Figure 3.9: Properties of the Roche ellipsoids.
spheroidal in shape, being elongated along the x
1
-axis (i.e., the axis joining the centers of the moon and the planet),
and compressed by almost equal amounts along the x
2
- and x
3
-axes. In the limit 1, in which the tidal distortion
eld due to the planet is weak, it is easily shown that
e
2
12
e
2
13

15
2
. (3.168)
For the case of the tidal distortion eld generated by the Earth, and acting on the Moon, which is characterized
by M/M
= 0.01230 and R/a

0
= 221.29, we obtain = 7.50 10
6
. It follows that e
13
= 7.50 10
3
, and
(a
1
a
3
)/a
1
e
2
13
/2 = 2.81 10
5
. In other words, were the Moon a homogeneous liquid body, the elongation
generated by the tidal eld of the Earth would be about 50 m.
It can be seen, from Table 3.3 and Figure 3.9, that the parameter attains a maximum value as the eccentricity
of a Roche ellipsoid varies from 0 to 1. In fact, this maximum value, = 0.06757, occurs when e
12
= 0.8594 and
e
13
= 0.8759. It follows that there is a maximum strength of the tidal distortion eld, generated by a planet, that is
consistent with an ellipsoidal equilibrium of a synchronously rotating, homogeneous, liquid moon in a circular orbit
about the planet. It is plausible that when this maximum strength is exceeded the moon is tidally disrupted by the
planet. The equilibrium condition < 0.06757 is equivalent to
R
a
0
> 2.455
_
_
1/3
, (3.169)
where = M/[(4/3) a
3
0
] and
= M
/[(4/3) a
0
3
] are the mean densities of the moon and the planet, respectively.
According to the above expression, there is a minimum orbital radius of a moon circling a planet. Below this radius,
which is called the Roche radius, the moon is presumably torn apart by tidal eects. The Roche radius for a syn-
chronously rotating, self-gravitating, liquid moon in a circular orbit about a spherical planet is about 2.5 times the
Hydrostatics 57
planets radius (assuming that the moon and the planet are about the same density). Of course, relatively small objects,
such as articial satellites, which are held together by internal tensile strength, rather than gravity, can orbit inside the
Roche radius without being disrupted.
3.15 Exercises
3.1. A hollow vessel oats in a basin. If, as a consequence of a leak, water ows slowly into the vessel, how will the level of the
water in the basin be aected?
3.2. A hollow spherical shell made up of material of specic gravity s > 1 has external and internal radii a and b, respectively.
Demonstrate that the sphere will only oat in water if
b
a
>
_
1
1
s
_
1/3
.
3.3. Show that the equilibrium of a solid of uniform density oating with an edge or corner just emerging from the water is
unstable.
3.4. Prove that if a solid of uniform density oats with a at face just above the waterline then the equilibrium is stable.
3.5. Demonstrate that a uniform solid cylinder oating with its axis horizontal is in a stable equilibrium provided that its length
exceeds the breadth of the waterline section. [Hint: The cylinder is obviously neutrally stable to rotations about its axis,
which means that the corresponding metacentric height is zero.]
3.6. Show that a uniform solid cylinder of radius a and height h can oat in stable equilibrium, with its axis vertical, if h/a <
2.
If the ratio h/a exceeds this value, prove that the equilibrium is only stable when the specic gravity of the cylinder lies
outside the range
1
2
_
_
1
_
1 2
a
2
h
2
_
_
.
3.7. A uniform, thin, hollow cylinder of radius a and height h is open at both ends. Assuming that h > 2 a, prove that the cylinder
cannot oat upright if its specic gravity lies in the range
1
2

_
1
4

a
2
h
2
.
3.8. Show that the cylinder of the preceding exercise can oat with its axis horizontal provided
h
2a
>
3 sin(s ),
where s is the specic gravity of the cylinder.
3.9. Prove that any segment of a uniform sphere, made up of a substance lighter than water, can oat in stable equilibrium with
its plane surface horizontal and immersed.
3.10. A vessel carries a tank of oil, of specic gravity s, running along its length. Assuming that the surface of the oil is at sea
level, show that the eect of the oils uidity on the rolling of the vessel is equivalent to a reduction in the metacentric height
by A
2
s/V, where V is the displacement of the ship, A the surface-area of the tank, and the radius of gyration of this area.
In what ratio is the eect diminished when a longitudinal partition bisects the tank?
3.11. Find the stable equilibrium congurations of a cylinder of elliptic cross-section, with major and minor radii a and b < a,
respectively, made up of material of specic gravity s, which oats with its axis horizontal.
3.12. A cylindrical tank has a circular cross-section of radius a. Let the center of gravity of the tank be located a distance c above
its base. Suppose that the tank is pivoted about a horizontal axis passing through its center of gravity, and is then lled with
uid up to a depth h above its base. Demonstrate that the position in which the tanks axis is upright is unstable for all lling
depths provided
c
2
<
1
2
a
2
.
Show that if c
2
> (1/2) a
2
then the upright position is stable when h lies in the range
c
_
c
2
a
2
/2.
58 FLUID MECHANICS
3.13. A thin cylindrical vessel of cross-sectional area A oats upright, being immersed to a depth h, and contains water to a depth
k. Show that the work required to pump out the water is
0
Ak (h k) g.
3.14. A sphere of radius a is just immersed in water that is contained in a cylindrical vessel of radius R whose axis is vertical.
Prove that if the sphere is raised just clear of the water then the waters loss of potential energy is
W a
_
1
2
3
a
2
R
2
_
,
where W is the weight of the water originally displaced by the sphere.
3.15. A sphere of radius a, weight W, and specic gravity s > 1, rests on the bottom of a cylindrical vessel of radius R whose axis
is vertical, and which contains water to a depth h > 2 a. Show that the work required to lift the sphere out of the vessel is
less than if the water had been absent by an amount
_
h a
2
3
a
3
R
2
_
W
s
.
3.16. A lead weight is immersed in water that is steadily rotating at an angular velocity about a vertical axis, the weight being
suspended from a xed point on this axis by a string of length l. Prove that the position in which the weight hangs vertically
downward is stable or unstable depending on whether l < g/
2
or l > g/
2
, respectively. Also, show that if the vertical
position is unstable then there exists a stable inclined position in which the string is normal to the surface of equal pressure
passing though the weight.
3.17. A thin cylindrical vessel of radius a and height H is orientated such that its axis is vertical. Suppose that the vessel is lled
with liquid of density to some height h < H above the base, spun about its axis at a steady angular velocity , and the
liquid allowed to attain a steady state. Demonstrate that, provided
2
a
2
/g < 4 h and
2
a
2
/g < 4 (H h), the net radial
thrust on the vertical walls of the vessel is
a h
2
g
_
1 +
2
a
2
4 g h
_
2
.
3.18. A thin cylindrical vessel of radius a with a plane horizontal lid is just lled with liquid of density , and the whole rotated
about a vertical axis at a xed angular velocity . Prove that the net upward thrust of the uid on the lid is
1
4
a
4

2
.
3.19. A liquid-lled thin spherical vessel of radius a spins about a vertical diameter at the xed angular velocity . Assuming that
the liquid co-rotates with the vessel, and that
2
> g/a, show that the pressure on the wall of the vessel is greatest a depth
g/
2
below the center. Also prove that the net normal thrusts on the lower and upper hemispheres are
5
4
Mg +
3
16
M
2
a,
and
1
4
Mg
3
16
M
2
a,
respectively, where M is the mass of the liquid.
3.20. A closed cubic vessel lled with water is rotating about a vertical axis passing through the centers of two opposite sides.
Demonstrate that, as a consequence of the rotation, the net thrust on a side is increased by
1
6
a
4

2
,
where a is the length of an edge of the cube, and the angular velocity of rotation.
3.21. A closed vessel lled with water is rotating at constant angular velocity about a horizontal axis. Show that, in the state of
relative equilibrium, the constant pressure surfaces in the water are circular cylinders whose common axis is a height g/
2
above the axis of rotation.
3.22. Consider a homogeneous, rotating, liquid body of mass M, mean radius a
0
, and angular velocity , whose outer boundary
is a Maclaurin spheroid of eccentricity e.
Hydrostatics 59
(a) Demonstrate that
e
_
5
2
(G M/a
3
0
)
1/2
in the low rotation limit, (G M/a
3
0
)
1/2
. Hence, show that e = 0.09262 for the case of a homogeneous body with
the same mass and volume as the Earth, which rotates once every 24 hours.
(b) Show that the critical angular velocity at which the bifurcation to the sequence of Jacobi ellipsoids takes place is
= 0.5298 (G M/a
3
0
)
1/2
,
and occurs when e = 0.81267. Hence, show that, for the case of a homogeneous body with the same mass and volume
as the Earth, the bifurcation would take place at a critical rotation period of 2 h 39 m.
(c) Demonstrate that the maximum angular velocity consistent with a spheroidal shape is
= 0.5805 (G M/a
3
0
)
1/2
,
and occurs when e = 0.92995. Hence, show that, for the case of a homogeneous body with the same mass and volume
as the Earth, this maximum velocity corresponds to a minimum rotation period of 2 h 25 m.
60 FLUID MECHANICS
Surface Tension 61
4 Surface Tension
4.1 Introduction
As is well-known, small drops of water in air, and small bubbles of gas in water, tend to adopt spherical shapes. This
phenomenon, and a host of other natural phenomena, can only be accounted for on the hypothesis that an interface
between two dierent media is associated with a particular form of energy whose magnitude is directly proportional to
the interfacial area. To be more exact, if S is the interfacial area then the contribution of the interface to the Helmholtz
free energy of the system takes the form S , where only depends on the temperature and chemical composition of
the two media on either side of the interface. It follows, from standard thermodynamics, that S is the work that must
be performed on the system in order to create the interface via an isothermal and reversible process. However, this
work is exactly the same as that which we would calculate on the assumption that the interface is in a state of uniform
constant tension per unit length . Thus, can be interpreted as both a free energy per unit area of the interface, and
a surface tension. This tension is such that a force of magnitude per unit length is exerted across any line drawn on
the interface, in a direction normal to the line, and tangential to the interface.
Surface tension originates from intermolecular cohesive forces. The average free energy of a molecule in a given
isotropic medium possessing an interface with a second medium is independent of its position, provided that the
molecule does not lie too close to the interface. However, the free energy is modied when the molecules distance
from the interface becomes less than the range of the cohesive forces (which is typically 10
9
m). Since this range is
so small, the number of molecules in a macroscopic system whose free energies are aected by the presence of an
interface is directly proportional to the interfacial area. Hence, the contribution of the interface to the total free energy
of the system is also proportional to the interfacial area. If only one of the two media in question is a condensed phase
then the parameter is invariably positive (i.e., such that a reduction in the surface area is energetically favorable).
This follows because the molecules of a liquid or a solid are subject to an attractive force from neighboring molecules.
However, molecules that are near to an interface with a gas lack neighbors on one side, and so experience an unbalanced
cohesive force directed toward the interior of the liquid/solid. The existence of this force makes it energetically
favorable for the interface to contract (i.e., > 0). On the other hand, if the interface separates a liquid and a solid,
or a liquid and another liquid, then the sign of cannot be predicted by this argument. In fact, it is possible for both
signs of to occur at liquid/solid and liquid/liquid interfaces.
The surface tension of a water/air interface at 20
C is = 7.28 10
2
Nm
1
. The surface tension at most oil/air
interfaces is much lowertypically, 2 10
2
Nm
1
. On the other hand, interfaces between liquid metals and
air generally have very large surface tensions. For instance, the surface tension of a mercury/air interface at 20
C is
4.87 10
1
Nm
1
.
For some pairs of liquids, such as water and alcohol, an interface cannot generally be observed because it is in
compression (i.e., < 0). Such an interface tends to become as large as possible, leading to complete mixing of the
two liquids. In other words, liquids for which > 0 are immiscible, whereas those for which < 0 are miscible.
Finally, the surface tension at a liquid/gas or a liquid/liquid interface can be aected by the presence of adsorbed
impurities at the interface. For instance, the surface tension at a water/air interface is signicantly deceased in the
presence of adsorbed soap molecules. Impurities that tend to reduce surface tension at interfaces are termed surfac-
tants.
4.2 Young-Laplace Equation
Consider an interface separating two immiscible uids that are in equilibrium with one another. Let these two uids
be denoted 1 and 2. Consider an arbitrary segment S of this interface that is enclosed by some closed curve C. Let
t denote a unit tangent to the curve, and let n denote a unit normal to the interface directed from uid 1 to uid 2.
(Note that C circulates around n in a right-handed manner.) See Figure 4.1. Suppose that p
1
and p
2
are the pressures
of uids 1 and 2, respectively, on either side of S . Finally, let be the (uniform) surface tension at the interface.
62 FLUID MECHANICS
C
t
n
t n
2
1
S
Figure 4.1: Interface between two immiscible uids.
The net force acting on S is
f =
_
S
(p
1
p
2
) ndS +
_
C
t ndr, (4.1)
where dS = ndS is an element of S , and dr = t dr an element of C. Here, the rst term on the right-hand side is the
net normal force due to the pressure dierence across the interface, whereas the second term is the net surface tension
force. Note that body forces play no role in (4.1), because the interface has zero volume. Furthermore, viscous forces
can be neglected, since both uids are static. Now, in equilibrium, the net force acting on S must be zero: i.e.,
_
S
(p
1
p
2
) ndS =
_
C
t ndr. (4.2)
(In fact, the net force would be zero even in the absence of equilibrium, because the interface has zero mass.)
Applying Stokes theorem (see Section A.22) to the curve C, we nd that
_
C
F dr =
_
S
F dS, (4.3)
where F is a general vector eld. This theorem can also be written
_
C
F t dr =
_
S
F ndS. (4.4)
Suppose that F = g b, where b is an arbitrary constant vector. We obtain
_
C
(g b) t dr =
_
S
(g b) ndS. (4.5)
However, the vector identity (A.179) yields
(g b) = ( g) b + (b ) g, (4.6)
since b is a constant vector. Hence, we get
b
_
C
t g dr = b
_
S
[(g) n ( g) n] dS, (4.7)
where b (g) n b
i
(g
j
/x
i
) n
j
. Now, since b is also an arbitrary vector, the above equation gives
_
C
t g dr =
_
S
_
(g) n ( g) n
_
dS. (4.8)
Taking g = n, we nd that
_
C
t ndr =
_
S
[(n) n ( n) n] dS. (4.9)
Surface Tension 63
But, (n) n (1/2) (n
2
) = 0, because n is a unit vector. Thus, we obtain
_
C
t ndr =
_
S
( n) ndS, (4.10)
which can be combined with (4.2) to give
_
S
_
(p
1
p
2
) ( n)
_
ndS = 0. (4.11)
Finally, given that S is arbitrary, the above expression reduces to the pressure balance constraint
p = n, (4.12)
where p = p
1
p
2
. The above relation is generally known as the Young-Laplace equation, and can also be derived
by minimizing the free energy of the interface. (See Section 4.8.) Note that p is the jump in pressure seen when
crossing the interface in the opposite direction to n. Of course, a plane interface is characterized by n = 0. On
the other hand, a curved interface generally has n 0. In fact, n measures the local mean curvature of the
interface. Thus, according to the Young-Laplace equation, there is a pressure jump across a curved interface between
two immiscible uids, the magnitude of the jump being proportional to the surface tension.
4.3 Spherical Interfaces
Generally speaking, the equilibrium shape of an interface between two immiscible uids is determined by solving the
force balance equation (3.1) in each uid, and then applying the Young-Laplace equation to the interface. However,
in situations in which a mass of one uid is completely immersed in a second uide.g., a mist droplet in air, or a
gas bubble in waterthe shape of the interface is fairly obvious. Provided that either the size of the droplet or bubble,
or the dierence in densities on the two sides of the interface, is suciently small, we can safely ignore the eect of
gravity. This implies that the pressure is uniform in each uid, and consequently that the pressure jump p is constant
over the interface. Hence, from (4.12), the mean curvature n of the interface is also constant. Since a sphere is the
only closed surface which possesses a constant mean curvature, we conclude that the interface is spherical. This result
also follows from the argument that a stable equilibrium state is one which minimizes the free energy of the interface,
subject to the constraint that the enclosed volume be constant. In other words, the equilibrium shape of the interface
is that which has the least surface area for a given volume: i.e., a sphere.
Suppose that the interface corresponds to the spherical surface r = R, where r is a spherical coordinate. (See
Section C.4.) It follows that n = e
r
r=R
. (Note, for future reference, that n points away from the center of curvature of
the interface.) Hence, from (C.65),
n =
1
r
2
r
2
r
r=R
=
2
R
. (4.13)
The Young-Laplace equation, (4.12), then gives
p =
2
R
. (4.14)
Thus, given that p is the pressure jump seen crossing the interface in the opposite direction to n, we conclude that the
pressure inside a droplet or bubble exceeds that outside by an amount proportional to the surface tension, and inversely
proportional to the droplet or bubble radius. This explains why small bubbles are louder that large ones when they
burst at a free surface: e.g., champagne zzes louder than beer. Note that soap bubbles in air have two interfaces
dening the inner and outer extents of the soap lm. Consequently, the net pressure dierence is twice that across a
single interface.
4.4 Capillary Length
Consider an interface separating the atmosphere from a liquid of uniform density that is at rest on the surface of the
Earth. Neglecting the density of air compared to that of the liquid, the pressure in the atmosphere can be regarded as
64 FLUID MECHANICS
r
2 1
3
Figure 4.2: Interface between a liquid (1), a gas (2), and a solid (3).
constant. On the other hand, the pressure in the liquid varies as p = p
0
g z (see Chapter 3), where p
0
is the pressure
of the atmosphere, g the acceleration due to gravity, and z measures vertical height (relative to the equilibrium height
of the interface in the absence of surface tension). Note that z increases upward. In this situation, the Young-Laplace
equation (4.12) yields
g z = n, (4.15)
where n is the normal to the interface directed from liquid to air. Now, if R represents the typical radius of curvature
of the interface then the left-hand side of the above equation dominates the right-hand side whenever R l, and vice
versa. Here,
l =
_

g
_
1/2
(4.16)
is known as the capillary length, and takes the value 2.7 10
3
m for pure water at 20
C. We conclude that the eect

of surface tension on the shape of an liquid/air interface is likely to dominate the eect of gravity when the interfaces
radius of curvature is much less than the capillary length, and vice versa.
4.5 Angle of Contact
Suppose that a liquid/air interface is in contact with a solid, as would be the case for water in a glass tube, or a drop of
mercury resting on a table. Figure 4.2 shows a section perpendicular to the edge at which the liquid, 1, the air, 2, and
the solid, 3, meet. Suppose that the free energies per unit area at the liquid/air, liquid/solid, and air/solid interfaces are
12
,
13
, and
23
, respectively. If the boundary between the three media is slightly modied in the neighborhood of
the edge, as indicated by the dotted line in the gure, then the area of contact of the air with the solid is increased by a
small amount r per unit breadth (perpendicular to the gure), whereas that of the liquid with the solid is decreased by
r per unit breadth, and that of the liquid with the air is decreased by r cos per unit breadth. Thus, the net change
in free energy per unit breadth is
23
r
13
r
12
r cos . (4.17)
However, an equilibrium state is one which minimizes the free energy, implying that the above expression is zero for
arbitrary (small) r: i.e.,
cos =
23

13
12
. (4.18)
We conclude that, in equilibrium, the angle of contact, , between the liquid and the solid takes a xed value that
depends on the free energies per unit area at the liquid/air, liquid/solid, and air/solid interfaces. Note that the above
Surface Tension 65
R
a
h
air
liquid
glass tube
z = 0
free surface
liquid
air
Figure 4.3: Elevation of liquid level in a capillary tube.
formula could also be obtained from the requirement that the various surface tension forces acting at the edge balance
one another, assuming that it is really appropriate to interpret
13
and
23
as surface tensions when one of the media
making up the interface is a solid.
As explained in Section 4.1, we would generally expect
12
and
23
to be positive. On the other hand,
13
could be
either positive or negative. Now, since cos 1, Equation (4.18) can only be solved when
13
lies in the range
23
+
12
>
13
>
23

12
. (4.19)
If
13
>
23
+
12
then the angle of contact is 180
, which corresponds to the case where the free energy at the

liquid/solid interface is so large that the liquid does not wet the solid at all, but instead breaks up into beads on its
surface. On the other hand, if
13
<
23

12
then the angle of contact is 0
, which corresponds to the case where the

free energy at the liquid/solid interface is so small that the liquid completely wets the solid, spreading out indenitely
until it either covers the whole surface, or its thickness reaches molecular dimensions.
The angle of contact between water and glass typically lies in the range 25
to 29
, whereas that between mercury

and glass is about 127
.
4.6 Jurins Law
Consider a situation in which a narrow, cylindrical, glass tube of radius a is dipped vertically into a liquid of density
, and the liquid level within the tube rises a height h above the free surface as a consequence of surface tension.
See Figure 4.3. Suppose that the radius of the tube is much less than the capillary length. A tube for which this is
the case is generally known as a capillary tube. According to the discussion in Section 4.4, the shape of the internal
liquid/air interface within a capillary tube is not signicantly aected by gravity. Thus, from Section 4.3, the interface
is a segment of a sphere of radius R (say). If is the angle of contact of interface with the glass then simple geometry
(see Figure 4.3) reveals that
R =
a
cos
. (4.20)
66 FLUID MECHANICS
Hence, from Equation (4.13), the mean curvature of the interface is given by
n =
2
R
=
2 cos
a
, (4.21)
where is the associated surface tension. [The minus sign in the above expression arises from the fact that n points
towards the center of curvature of the interface, whereas the opposite is true for Equation (4.13).] Finally, from (4.15),
application of the Young-Laplace equation to the interface yields
g h =
2 cos
a
, (4.22)
which can be rearranged to give
h
2 cos
g a
. (4.23)
This result, which relates the height, h, to which a liquid rises in a capillary tube of radius a to the liquids surface
tension, , is known as Jurins law. Note that the assumption that the radius of the tube is much less than the capillary
length is equivalent to the assumption that the height of the interface above the free surface of the liquid is much
greater than the radius of the tube. This follows, from (4.16) and (4.23), because
h
a
= 2 cos
l
2
a
2
. (4.24)
Thus, the ordering a l implies that h a.
For the case of water at 20
, assuming a contact angle of 25
, Jurins law yields h(mm) = 13.5/a(mm). Thus,

water rises a height 13.5 mm in a capillary tube of radius 1 mm, but rises 13.5 cm in a capillary tube of radius 0.1 mm.
Note that in the case of a liquid, such a mercury, that has an oblique angle of contact with glass, so that cos < 0, the
liquid level in a capillary tube is depressed below that of the free surface (i.e., h < 0).
4.7 Capillary Curves
Let adopt Cartesian coordinates on the Earths surface such that z increases vertically upward. Suppose that the
interface of a liquid of density and surface tension with the atmosphere corresponds to the surface z = f (x), where
the liquid occupies the region z < f (z). Note that the shape of the interface is y-independent. The unit normal to the
interface (directed from liquid to air) is thus
n =
(z f )
(z f )
=
e
z
f
x
e
x
(1 + f
2
x
)
1/2
, (4.25)
where f
x
= d f /dx. Hence, the mean curvature of the interface is
n =
f
xx
(1 + f
2
x
)
3/2
, (4.26)
where f
xx
= d
2
f /dx
2
. According to (4.15) and (4.18), the shape of the interface is governed by the nonlinear dieren-
tial equation
f =
l
2
f
xx
(1 + f
2
x
)
3/2
. (4.27)
where the vertical height, f , of the interface is measured relative to its equilibrium height in the absence of surface
tension. Multiplying the above equation by f
x
/l
2
, and integrating with respect to x, we obtain
1
(1 + f
2
x
)
1/2
= C
f
2
2 l
2
, (4.28)
where C is a constant. It follows that
C
f
2
2 l
2
1, (4.29)
Surface Tension 67
0
1
2
3
z/l
2 1 0 1 2
x/l
Figure 4.4: Capillary curves for /4 3/4 and (in order from the top to the bottom) k = 0.6, 0.7, 0.8, 0.9, and
0.99.
and
1
f
x
=
C f
2
/2 l
2
[1 (C f
2
/2 l
2
)
2
]
1/2
. (4.30)
Let
C =
2
k
2
1, (4.31)
where 0 < k < 1, and
f =
2 l
k
(1 k
2
sin
2
)
1/2
. (4.32)
Thus, from (4.31) and (4.32),
C
f
2
2 l
2
= cos(2 ), (4.33)
and so the constraint (4.29) implies that /4 3/4. Moreover, Equations (4.30) and (4.33) reduce to
1
f
x
=
dx
d f
=
1
tan(2 )
. (4.34)
It follows from (4.32) and (4.34) that
dx
d
=
dx
d f
d f
d
=
l k cos(2 )
(1 k
2
sin
2
)
1/2
, (4.35)
which can be integrated to give
x
l
=
_
/2
k cos(2 )
(1 k
2
sin
2
)
1/2
d, (4.36)
assuming that x = 0 when = /2. Thus, we get
x
l
=
_
k
2
k
_

F(, k) +
2
k

E(, k), (4.37)
68 FLUID MECHANICS
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
z/l
1 0.5 0 0.5 1
x/l
Figure 4.5: Liquid/air interface for a liquid trapped between two vertical parallel plates located at x = l. The contact
angle of the interface with the plates is = 30
.
where
E(, k) = E(/2, k) E(, k), (4.38)
F(, k) = F(/2, k) F(, k), (4.39)

and
E(, k) =
_

0
(1 k
2
sin
2
)
1/2
, (4.40)
F(, k) =
_

0
(1 k
2
sin
2
)
1/2
, (4.41)
are types of incomplete elliptic integral.
1
In conclusion, the interface shape is determined parametrically by
x
l
=
_
k
2
k
_

F(, k) +
2
k

E(, k), (4.42)
z
l
=
2
k
(1 k
2
sin
2
)
1/2
, (4.43)
where /4 3/4. Here, the parameter k is restricted to lie in the range 0 < k < 1.
Figure 4.4 shows the capillary curves predicted by (4.42) and (4.43) for various dierent values of k. Here, we
have chosen the plus sign in (4.43). However, if the minus sign is chosen then the curves are simply inverted: i.e.,
x x and z z. In can be seen that all of the curves shown in the gure are symmetric about x = 0: i.e., z z as
x x. Consequently, we can use these curves to determine the shape of the liquid/air interface which arises when a
liquid is trapped between two at vertical plates (made of the same material) that are parallel to one another. Suppose
1
Surface Tension 69
that the plates in question lie at x = d. Furthermore, let the angle of contact of the interface with the plates be ,
where < /2. Since the angle of contact is acute, we expect the liquid to be drawn upward between the plates, and
the interface to be concave (from above). This corresponds to the positive sign in (4.43). In order for the interface to
meet the plates at the correct angle, we require f
x
= 1/ tan at x = d and f
x
= 1/ tan at x = +d. However, if one
of these boundary conditions is satised then, by symmetry, the other is automatically satised. From Equation (4.34)
(choosing the positive sign), the latter boundary condition yields tan(2 ) = 1/ tan at x = +d, which is equivalent to
x = +d when = 3/4/2. Substituting this value of into Equation (4.42), we can numerically determine the value
of k for which x = d. The interface shape is then given by Equations (4.42) and (4.43), using the aforementioned value
of k, and in the range /4 +/2 to 3/4 /2. For instance, if d = l and = 30
then k = 0.9406, and the associated

interface is shown in Figure 4.5. Furthermore, if we invert this interface (i.e., x x and z z) then we obtain the
interface which corresponds to the same plate spacing, but an obtuse contact angle of = 180
30
= 150
.
Consider the limit k 1, which is such that the distance between the two plates is much less than the capillary
length. It is easily demonstrated that, at small k,
2
E(, k)

k
2
4
(
+ sin

cos

), (4.44)
F(, k)

+
k
2
4
(
+ sin

cos

), (4.45)
where

= /2 . Thus, Equations (4.42) and (4.43) reduce to
x
l

k
2
sin(2

), (4.46)
z
l

2
k

k
2
[1 cos(2

)]. (4.47)
It follows that the interface is a segment of the curved surface of a cylinder whose axis runs parallel to the y-axis. If
the distance between the plates is 2 d, and the contact angle is , then we require x = d when = 3/4 /2 (which
corresponds to

= /4 + /2). From Equation (4.46), this constraint yields
d
l

k
2
cos . (4.48)
Thus, the height that the liquid rises between the two platesi.e., h = z(x = 0) = z(
= /2) 2 l/kis given by

h
cos
g d
. (4.49)
This result is the form taken by Jurins law, (4.23), for a liquid drawn up between two parallel plates of spacing 2 d.
Consider the case k = C = 1, which is such that the distance between the two plates is innite. Let the leftmost
plate lie at x = 0, and let us completely neglect the rightmost plate, since it lies at innity. Suppose that h = z(x = 0)
is the height of the interface above the free surface of the liquid at the point where the interface meets the leftmost
plate. If is the angle of contact of the interface with the plate then we require f
x
= 1/ tan at x = 0. Since C = 1, it
follows from (4.28) that
h
2
2 l
2
= 1 sin , (4.50)
or
h = 2 l sin(/4 /2). (4.51)
Furthermore, again recalling that C = 1, Equation (4.30) can be integrated to give
x =
_
h
z
d f
f
x
= l
_
h/2l
z/2l
1 2 y
2
y (1 y
2
)
1/2
dy, (4.52)
2
70 FLUID MECHANICS
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
z/l
0 1 2 3 4 5
x/l
Figure 4.6: Liquid/air interface for a liquid in contact with a vertical plate located at x = 0. The contact angle of the
interface with the plate is = 25
.
where we have chosen the minus sign, and y = f /2l. Making the substitution y = sin u, this becomes
x
l
=
_
sin
1
(h/2l)
sin
1
(z/2l)
_
1
sin u
2 sin u
_
du =
_
ln
_
1 + cos u
sin u
_
+ 2 cos u
_
sin
1
(h/2l)
sin
1
(z/2l)
, (4.53)
which reduces to
x
l
= cosh
1
_
2 l
z
_
cosh
1
_
2 l
h
_
+
_
4
h
2
l
2
_
1/2
_
4
z
2
l
2
_
1/2
, (4.54)
since cosh
1
(z) ln[z + (z
2
1)
1/2
]. Thus, Equations (4.51) and (4.54) specify the shape of a liquid/air interface that
meets an isolated vertical plate at x = 0. In particular, (4.51) gives the height that the interface climbs up the plate
(relative to the free surface) due to the action of surface tension. Note that this height is restricted to lie in the range
2 l h 2 l, irrespective of the angle of contact. Figure 4.6 shows an example interface calculated for = 25
.
4.8 Axisymmetric Soap-Bubbles
Consider an axisymmetric soap-bubble whose surface takes the form r = f (z) in cylindrical coordinates. See Sec-
tion C.3. The unit normal to the surface is
n
(r f )
(r f )
=
e
r
f
z
e
z
(1 + f
2
z
)
1/2
, (4.55)
where f
z
d f /dz. Hence, from (C.39), the mean curvature of the surface is given by
n =
1
f f
z
d
dz
_
f
(1 + f
2
z
)
1/2
_
. (4.56)
Surface Tension 71
The Young-Laplace equation, (4.12), then yields
f f
z
a
=
d
dz
_
f
(1 + f
2
z
)
1/2
_
, (4.57)
where
a =
p
0
. (4.58)
Here, is the net surface tension, including the contributions from the internal and external soap/air interfaces. More-
over, p
0
= p is the pressure dierence between the interior and the exterior of the bubble. Equation (4.57) can be
integrated to give
f
(1 + f
2
z
)
1/2
=
f
2
2 a
+ C, (4.59)
where C is a constant.
Suppose that the bubble occupies the region z
1
z z
2
, where z
1
< z
2
, and has a xed radius at its two end-points,
z = z
1
and z = z
2
. This could most easily be achieved by supporting the bubble on two rigid parallel co-axial rings
located at z = z
1
and z = z
2
. The net free energy required to create the bubble can be written
c = S p
0
V, (4.60)
where S is area of the bubble surface, and V the enclosed volume. The rst term on the right-hand side of the
above expression represents the work needed to overcome surface tension, whilst the second term represents the work
required to overcome the pressure dierence, p
0
, between the exterior and the interior of the bubble. Now, from the
general principles of statics, we expect a stable equilibrium state of a mechanical system to be such as to minimize the
net free energy, subject to any dynamical constraints. It follows that the equilibrium shape of the bubble is such as to
minimize
c =
_
z
2
z
1
2 f (1 + f
2
z
)
1/2
dz p
0
_
z
2
z
1
f
2
dz, (4.61)
subject to the constraint that the bubble radius, f , be xed at z = z
1
and z = z
2
. Hence, we need to nd the function
f (z) that minimizes the integral
_
z
2
z
1
[( f , f
z
) dz, (4.62)
where
[( f , f
z
) = 2 f (1 + f
2
z
)
1/2
p
0
f
2
, (4.63)
subject to the constraint that f is xed at the limits. This is a standard problem in the calculus of variations. (See
Appendix D.) In fact, since the functional [( f , f
z
) does not depend explicitly on z, the minimizing function is the
solution of [see Equation (D.14)]
[ f
z
[
f
z
= C
, (4.64)
where C
is an arbitrary constant. Thus, we obtain

2
_
f
(1 + f
2
z
)
1/2

f
2
2 a
_
= C
, (4.65)
which can be rearranged to give Equation (4.59). Hence, we conclude that application of the Young-Laplace equation
does indeed lead to a bubble shape that minimizes the net free energy of the soap/air interfaces.
Consider the case p
0
= 0, in which there is no pressure dierence across the surface of the bubble. In this situation,
writing C = b > 0, Equation (4.59) reduces to
f = b (1 + f
2
z
)
1/2
. (4.66)
72 FLUID MECHANICS
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
r/c
0.6 0.4 0.2 0 0.2 0.4 0.6
z/c
Figure 4.7: Radius versus axial distance for a catenoid soap bubble supported by two parallel co-axial rings of radius
c located at z = 0.65 c.
Moreover, according to the previous discussion, the bubble shape specied by (4.66) is such as to minimize the surface
area of the bubble (since the only contribution to the free energy of the soap/air interfaces is directly proportional to
the bubble area). The above equation can be rearranged to give
f
z
=
_
f
2
b
2
1
_
1/2
, (4.67)
which leads to
z z
0
=
_
r
b
d f
f
z
=
_
r
b
d f
( f
2
/b
2
1)
1/2
= b cosh
1
(r/b), (4.68)
or
r = b cosh(z z
0
/b), (4.69)
where z
0
is a constant. This expression describes an axisymmetric surface known as a catenoid.
Suppose, for instance, that the soap bubble is supported by identical rings of radius c that are located a perpendic-
ular distance 2 d apart. Without loss of generality, we can specify that the rings lie at z = d. It thus follows, from
(4.69), that z
0
= 0, and
r = b cosh(z/b). (4.70)
Here, the parameter b must be chosen so as to satisfy
c = b cosh(d/b). (4.71)
For example, if d = 0.65 c then b = 0.6416 c, and the resulting bubble shape is illustrated in Figure 4.7.
Let d/c = and d/b = u, in which case the above equation becomes
G(u) = u cosh u = 0. (4.72)
Surface Tension 73
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
r
/
2 1 0 1 2
z/
Figure 4.8: Radius versus axial distance for an unduloid soap bubble calculated with k = 0.95.
Now, the function G(u) attains a maximum value
G(u
0
) = u
0

1
tanh u
0
, (4.73)
when u
0
= sinh
1
(1/). Moreover, if G(u
0
) > 0 then Equation (4.72) possesses two roots. It turns out that the root
associated with the smaller value of u minimizes the interface system energy, whereas the other root maximizes the
free energy. Hence, the former root corresponds to a stable equilibrium state, whereas the latter corresponds to an
unstable equilibrium state. On the other hand, if G(u
0
) < 0 then Equation (4.72) possesses no roots, implying the
absence of any equilibrium state. The critical case G(u
0
) = 0 corresponds to u = u
c
and =
c
, where u
c
tanh u
c
= 1
and
c
= 1/ sinh u
c
. It is easily demonstrated that u
c
= 1.1997 and
c
= 0.6627. We conclude that a stable equilibrium
state of a catenoid bubble only exists when
c
, which corresponds to d 0.6627 c. If the relative ring spacing d
exceeds the critical value 0.6627 c then the bubble presumably bursts.
Consider the case p
0
0, in which there is a pressure dierence across the surface of the bubble. In this situation,
writing
2 a = + , (4.74)
2 a C = , (4.75)
Equation (4.59) becomes
( + ) f
(1 + f
2
z
)
1/2
= f
2
+ , (4.76)
which can be rearranged to give
f
z
=
(
2
f
2
)
1/2
( f
2

2
)
1/2
f
2
+
. (4.77)
We can assume, without loss of generality, that > . It follows, from the above expression, that f .
Hence, we can write
f
2
=
2
cos
2
+
2
sin
2
, (4.78)
74 FLUID MECHANICS
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
r
/
0.5 0 0.5
z/
Figure 4.9: Radius versus axial distance for a positive pressure nodoid soap bubble calculated with k = 0.95.
k
2
=
2

2
2
, (4.79)
where 0 /2 and 0 < k 1. It follows that
f = (1 k
2
sin
2
)
1/2
, (4.80)
= sgn() (1 k
2
)
1/2
, (4.81)
and
dz
d
=
1
f
z
d f
d
=
_
f +
f
_
, (4.82)
z =
_
E(, k) + sgn() (1 k
2
)
1/2
F(, k)
_
, (4.83)
where E(, k) and F(, k) are incomplete elliptic integrals [see Equations (4.40) and (4.41)]. Here, we have assumed
that = 0 when z = 0. There are three cases of interest.
In the rst case, > 0 and > 0. It follows that (1 k
2
)
1/2
r/ 1 for /2 0, and 0.5 /(p
0
) < 1 for
1 k > 0, where
r = (1 k
2
sin
2
), (4.84)
z =
_
E(, k) + (1 k
2
)
1/2
F(, k)
_
. (4.85)
The axisymmetric curve parameterized by the above pair of equations is known as an unduloid. Note that an unduloid
bubble always has positive internal pressure (relative to the external pressure): i.e., p
0
> 0. An example unduloid soap
bubble is illustrated in Figure 4.8
In the second case, > 0 and < 0. It follows that (1 k
2
)
1/4
r/ 1 for
0
0, and 0 < /(p
0
) 0.5
for 0 < k 1, where
0
= sin
1
([1 (1 k
2
)
1/2
]
1/2
/k), and
r = (1 k
2
sin
2
), (4.86)
z =
_
E(, k) (1 k
2
)
1/2
F(, k)
_
. (4.87)
Surface Tension 75
0
0.1
0.2
0.3
0.4
0.5
0.6
r
/
|
|
0.1 0.2 0.3 0.4 0.5
z/||
Figure 4.10: Radius versus axial distance for a negative pressure nodoid soap bubble calculated with k = 0.95.
The axisymmetric curve parameterized by the above pair of equations is known as an nodoid. This particular type
of nodoid bubble has positive internal pressure: i.e., p
0
> 0. An example positive pressure nodoid soap bubble is
illustrated in Figure 4.9.
In the third case, < 0 and > 0. It follows that (1 k
2
)
1/2
r/ (1 k
2
)
1/4
for /2
0
(or
/2
0
), and 0 > /(p
0
) 0.5 for 0 < k 1, where
r = (1 k
2
sin
2
), (4.88)
z =
_
E(, k) (1 k
2
)
1/2
F(, k)
_
. (4.89)
The axisymmetric curve parameterized by the above pair of equations is again a nodoid. However, this particular type
of nodoid bubble has negative internal pressure: i.e., p
0
< 0. An example negative pressure nodoid soap bubble is
illustrated in Figure 4.10.
4.9 Exercises
4.1. Show that if N equal spheres of water coalesce so as to form a single spherical drop then the surface energy is decreased by
a factor 1/N
1/3
.
4.2. A circular cylinder of radius a, height h, and specic gravity s oats upright in water. Show that the depth of the base below
the general level of the water surface is
s h +
2
a
cos ,
where is the surface tension at the air/water interface, and the contact angle of the interface with the cylinder.
4.3. A lm of water is held between two parallel plates of glass a small distance 2d apart. Prove that the apparent attraction
between the plates is
2 A cos
d
+ L sin ,
where is the surface tension at the air/water interface, the angle of contact of the interface with glass, A the area of the
lm, and L the circumference of the lm.
76 FLUID MECHANICS
4.4. Show that if the surface of a sheet of water is slightly corrugated then the surface energy is increased by
2
_ _
x
_
2
dx
per unit breadth of the corrugations. Here, x is measured horizontally, perpendicular to the corrugations. Moreover, denotes
the elevation of the surface above the mean level. Finally, is the surface tension at an air/water interface. If the corrugations
are sinusoidal, such that
= a sin(k x),
show that the average increment of the surface energy per unit area is (1/4) a
2
k
2
.
4.5. A mass of liquid, which is held together by surface tension alone, revolves about a xed axis at a small angular velocity ,
so as to assume a slightly spheroidal shape of mean radius a. Prove that the ellipticity of the spheroid is
=

2
a
3
8
,
where is the uniform mass density, and the surface tension. [If r
+
is the maximum radius, and r
the minimum radius,

then a = (r
2
+
r
)
1/3
, and the ellipticity is dened = (r
+
r
)/r
+
.]
4.6. A liquid mass rotates, in the form of a circular ring of radius a and small cross-section, with a constant angular velocity ,
about an axis normal to the plane of the ring, and passing through its center. The mass is held together by surface tension
alone. Show that the section of the ring must be approximately circular. Demonstrate that
=
_
2
a c
2
_
1/2
,
where is the density, the surface tension, and c the radius of the cross-section.
4.7. Two spherical soap bubbles of radii a
1
and a
2
are made to coalesce. Show that when the temperature of the gas in the
resulting bubble has returned to its initial value the radius a of the bubble satises
p
0
a
3
+ 4 a
2
= p
0
(a
3
1
+ a
3
2
) + 4 (a
2
1
+ a
2
2
),
where p
0
is the ambient pressure, and the surface tension of the soap/air interfaces.
4.8. A rigid sphere of radius a rests on a at rigid surface, and a small amount of liquid surrounds the contact point, making a
concave-planar lens whose diameter is small compared to a. The angle of contact of the liquid/air interface with each of the
solid surfaces is zero, and the surface tension of the interface is . Show that there is an adhesive force of magnitude 4 a
acting on the sphere. (It is interesting to note that the force is independent of the volume of liquid.)
4.9. Two small solid bodies are oating on the surface of a liquid. Show that the eect of surface tension is to make the objects
approach one another if the liquid/air interface has either an acute or an obtuse angle of contact with both bodies, and to
make them move away from one another if the interface has an acute angle of contact with one body, and an obtuse angle of
contact with the other.
Incompressible Inviscid Fluid Dynamics 77
5 Incompressible Inviscid Fluid Dynamics
5.1 Introduction
This chapter introduces some of the fundamental concepts that arise in the theory of incompressible, inviscid (or, to
be more exact, high Reynolds number) uid motion.
5.2 Streamlines, Stream Tubes, and Stream Filaments
A line drawn in a uid such that its tangent at each point is parallel to the local uid velocity is called a streamline. The
aggregate of all the streamlines at a given instance in time constitutes the instantaneous ow pattern. The streamlines
drawn through each point of a closed curve constitute a stream tube. Finally, a stream lament is dened as a stream
tube whose cross-section is a curve of innitesimal dimensions.
When the ow is unsteady then the conguration of the stream tubes and laments changes from time to time.
However, when the ow is steady then the stream tubes and laments are stationary. In the latter case, a stream tube
acts like an actual tube through which the uid is owing. This follows because there can be no ow across the walls,
and into the tube, since the ow is, by denition, always tangential to these walls. Moreover, the walls are xed in
space and time, since the motion is steady. Thus, the motion of the uid within the tube would be unchanged were the
walls replaced by a rigid frictionless boundary.
Consider a stream lament of an incompressible uid whose motion is steady. Suppose that the cross-sectional
area of the lament is suciently small that the uid velocity is the same at each point on the cross-section. Moreover,
let the cross-section be everywhere normal to the direction of this common velocity. Suppose that v
1
and v
2
are the
ow speeds at two points on the lament at which the cross-sectional areas are S
1
and S
2
, respectively. Consider the
section of the lament lying between these points. Since the uid is incompressible, the same volume of uid must
ow into one end of the section, in a given time interval, as ows out of the other, which implies that
v
1
S
1
= v
2
S
2
. (5.1)
This is the simplest manifestation of the equation of uid continuity discussed in Section 2.9. The above result is
equivalent to the statement that the product of the speed and cross-sectional area is constant along any stream lament
of an incompressible uid in steady motion. Thus, a stream lament within such a uid cannot terminate unless the
velocity at that point becomes innite. Leaving this case out of consideration, it follows that stream laments in
steadily owing incompressible uids are either closed loops, or terminate at the boundaries of the uid. The same is,
of course, true of streamlines.
5.3 Bernoullis Theorem
In its most general form, Bernoullis theoremwhich was discovered by Daniel Bernoulli (17001783)states that,
in the steady ow of an inviscid uid, the quantity
p
+ 7 (5.2)
is constant along a streamline, where p is the pressure, the density, and 7 the total energy per unit mass.
The proof is straightforward. Consider the body of uid bounded by the cross-sectional areas AB and CD of the
stream lament pictured in Figure 5.1. Let us denote the values of quantities at AB and CD by the suxes 1 and 2,
respectively. Thus, p
1
, v
1
,
1
, S
1
, 7
1
are the pressure, uid speed, density, cross-sectional area, and total energy per
unit mass, respectively, at AB, etc. Suppose that, after a short time interval t, the body of uid has moved such that it
occupies the section of the lament bounded by the cross-sections A
and C
, where AA
= v
1
t and CC
= v
2
t.
Since the motion is steady, the mass m of the uid between AB and A
is the same as that between CD and C
, so
that
m = S
1
v
1
t
1
= S
2
v
2
t
2
. (5.3)
78 FLUID MECHANICS
A
C
C
D
D
B
A
Figure 5.1: Bernoullis theorem.
Let T denote the total energy of the section of the uid lying between A
and CD. Thus, the increase in energy of

the uid body in the time interval t is
(m7
2
+ T) (m7
1
+ T) = m(7
2
7
1
). (5.4)
In the absence of viscous energy dissipation, this energy increase must equal the net work done by the uid pressures
at AB and CD, which is
p
1
S
1
v
1
t p
2
S
2
v
2
t = m
_
p
1
p
2
2
_
. (5.5)
Equating expressions (5.4) and (5.5), we nd that
p
1
1
+ 7
1
=
p
2
2
+ 7
2
, (5.6)
which demonstrates that p/ + 7 has the same value at any two points on a given stream lament, and is therefore
constant along the lament. Note that Bernoullis theorem has only been proved for the case of the steady motion of
an inviscid uid. However, the uid in question may either be compressible or incompressible.
For the particular case of an incompressible uid, moving in a conservative force-eld, the total energy per unit
mass is the sum of the kinetic energy per unit mass, (1/2) v
2
, and the potential energy per unit mass, , and Bernoullis
theorem thus becomes
p
+
1
2
v
2
+ = constant along a streamline. (5.7)
If we focus on a particular streamline, 1 (say), then Bernoullis theorem states that
p
+
1
2
v
2
+ = C
1
, (5.8)
where C
1
is a constant characterizing that streamline. If we consider a second streamline, 2 (say), then
p
+
1
2
v
2
+ = C
2
, (5.9)
where C
2
is another constant. It is not generally the case that C
1
= C
2
. If, however, the uid motion is irrotational
then the constant in Bernoullis theorem is the same for all streamlines (see Section 5.7), so that
p
+
1
2
v
2
+ = C (5.10)
S
2
A
B
S
1
Figure 5.2: A vortex lament.
throughout the uid.
5.4 Vortex Lines, Vortex Tubes, and Vortex Filaments
The curl of the velocity eld of a uid, which is generally termed vorticity, is usually represented by the symbol :
i.e.,
= v. (5.11)
A vortex line is a line whose tangent is everywhere parallel to the local vorticity vector. The vortex lines drawn
through each point of a closed curve constitute the surface of a vortex tube. Finally, a vortex lament is a vortex tube
whose cross-section is of innitesimal dimensions.
Consider a section AB of a vortex lament. The lament is bounded by the curved surface that forms the lament
wall, as well as two plane surfaces, whose vector areas are S
1
and S
2
(say), which form the ends of the section at
points A and B, respectively. See Figure 5.2. Let the plane surfaces have outward pointing normals that are parallel (or
anti-parallel) to the vorticity vectors,
1
and
2
, at points A and B, respectively. Gausss theorem (see Section A.20),
applied to the section, yields
_
dS =
_
dV, (5.12)
where dS is an outward directed surface element, and dV a volume element. However,
= v 0 (5.13)
[see Equation (A.173)], implying that
_
dS = 0. (5.14)
Now, dS = 0 on the curved surface of the lament, since is, by denition, tangential to this surface. Thus, the
only contributions to the surface integral come from the plane areas S
1
and S
2
. It follows that
_
dS = S
2

2
S
1

1
= 0. (5.15)
This result is essentially an equation of continuity for vortex laments. It implies that the product of the magnitude of
the vorticity and the cross-sectional area, which is termed the vortex intensity, is constant along the lament. It follows
that a vortex lament cannot terminate in the interior of the uid. For, if it did, the cross-sectional area, S , would have
80 FLUID MECHANICS
to vanish, and, therefore, the vorticity, , would have to become innite. Thus, a vortex lament must either form a
closed vortex ring, or must terminate at the uid boundary.
Since a vortex tube can be regarded as a bundle of vortex laments whose net intensity is the sum of the intensities
of the constituent laments, we conclude that the intensity of a vortex tube remains constant along the tube.
5.5 Circulation and Vorticity
Consider a closed curve C situated entirely within a moving uid. The vector line integral (see Section A.14)
C
=
_
C
v dr, (5.16)
where dr is an element of C, and the integral is taken around the whole curve, is termed the circulation of the ow
around the curve. The sense of circulation (e.g., either clockwise or counter-clockwise) is arbitrary.
Let S be a surface having the closed curve C for a boundary, and let dS be an element of this surface (see Sec-
tion A.7) with that direction of the normal which is related to the chosen sense of circulation around C by the right-hand
circulation rule (see Section A.8). According to Stokes theorem (see Section A.22),
C
=
_
C
v dr =
_
S
dS. (5.17)
Thus, we conclude that circulation and vorticity are intimately related to one another. In fact, according to the above
expression, the circulation of the uid around loop C is equal to the net sum of the intensities of the vortex laments
passing through the loop and piercing the surface S (with a lament making a positive, or negative, contribution to the
sum depending on whether it pierces the surface in the direction determined by the chosen sense of circulation around
C and the right-hand circulation rule, or in the opposite direction). One important proviso to (5.17) is that the surface
S must lie entirely within the uid.
5.6 Kelvin Circulation Theorem
According to the Kelvin circulation theorem, which is named after Lord Kelvin (18241907), the circulation around
any co-moving loop in an inviscid uid is independent of time. The proof is as follows. The circulation around a given
loop C is dened
C
=
_
C
v dr. (5.18)
However, for a loop that is co-moving with the uid, we have dv = d(dr/dt) = d(dr)/dt. Thus,
d
C
dt
=
_
C
dv
dt
dr +
_
C
v dv. (5.19)
However, by denition, dv/dt = Dv/Dt for a co-moving loop (see Section 2.10). Moreover, the equation of motion of
an incompressible inviscid uid can be written [see Equation (2.79)]
Dv
Dt
=
_
p
+
_
, (5.20)
since is a constant. Hence,
d
C
dt
=
_
C
_
p
1
2
v
2
+
_
dr = 0, (5.21)
since v dv = d(v
2
/2) = (v
2
/2) dr (see Section A.18), and p/ v
2
/2 + is obviously a single-valued function.
One corollary of the Kelvin circulation theorem is that the uid particles that form the walls of a vortex tube at a
given instance in time continue to form the walls of a vortex tube at all subsequent times. To prove this, imagine a
closed loop C that is embedded in the wall of a vortex tube but does not circulate around the interior of the tube. See
C
C
Figure 5.3: A vortex tube.
Figure 5.3. The normal component of the vorticity over the surface enclosed by C is zero, since all vorticity vectors are
tangential to this surface. Thus, from (5.17), the circulation around the loop is zero. By Kelvins circulation theorem,
the circulation around the loop remains zero as the tube is convected by the uid. In other words, although the surface
enclosed by C deforms, as it is convected by the uid, it always remains on the tube wall, since no vortex laments
can pass through it.
Another corollary of the circulation theoremis that the intensity of a vortex tube remains constant as it is convected
by the uid. This can be proved by considering the circulation around the loop C
pictured in Figure 5.3.

5.7 Irrotational Flow
Flow is said to be irrotational when the vorticity has the magnitude zero everywhere. It immediately follows, from
Equation (5.17), that the circulation around any arbitrary loop in an irrotational ow pattern is zero (provided that the
loop can be spanned by a surface that lies entirely within the uid). Hence, from Kelvins circulation theorem, if an
inviscid uid is initially irrotational then it remains irrotational at all subsequent times. This can be seen more directly
from the equation of motion of an inviscid incompressible uid which, according to Equations (2.39) and (2.79), takes
the form
v
t
+ (v ) v =
_
p
+
_
, (5.22)
since is a constant. However, from Equation (A.171),
(v ) v = (v
2
/2) v . (5.23)
Thus, we obtain
v
t
=
_
p
+
1
2
v
2
+
_
+ v . (5.24)
Taking the curl of this equation, and making use of the vector identities 0 [see Equation (A.176)], A 0
[see Equation (A.173)], as well as the identity (A.179), and the fact that v = 0 in an incompressible uid, we obtain
the vorticity evolution equation
D
Dt
= ( ) v. (5.25)
Thus, if = 0, initially, then D/Dt = 0, and, consequently, = 0 at all subsequent times.
82 FLUID MECHANICS
Suppose that O is a xed point, and P an arbitrary movable point, in an irrotational uid. Let O and P be joined by
two dierent paths, OAP and OBP (say). It follows that OAPBO is a closed curve. Now, since the circulation around
such a curve in an irrotational uid is zero, we can write
_
OAP
v dr +
_
PBO
v dr = 0, (5.26)
which implies that
_
OAP
v dr =
_
OBP
v dr =
P
(5.27)
(say). It is clear that
P
is a scalar function whose value depends on the position of P (and the xed point O), but not
on the path taken between O and P. Thus, if O is the origin of our coordinate system, and P an arbitrary point whose
position vector is r, then we have eectively dened a scalar eld (r) =
_
P
O
v dr.
Consider a point Q that is suciently close to P that the velocity v is constant along PQ. Let be the position
vector of Q relative to P. It then follows that (see Section A.18)
=
Q
+
P
=
_
Q
P
v dr v . (5.28)
The above equation becomes exact in the limit that 0. Since Q is arbitrary (provided that it is suciently close
to P), the direction of the vector is also arbitrary, which implies that
v = . (5.29)
We, thus, conclude that if the motion of a uid is irrotational then the associated velocity eld can always be expressed
as minus the gradient of a scalar function of position, (r). This scalar function is called the velocity potential, and
ow which is derived from such a potential is known as potential ow. Note that the velocity potential is undened to
an arbitrary additive constant.
We have demonstrated that a velocity potential necessarily exists in a uid whose velocity eld is irrotational.
Conversely, when a velocity potential exists the ow is necessarily irrotational. This follows because [see Equa-
tion (A.176)]
= v = = 0. (5.30)
Incidentally, the uid velocity at any given point in an irrotational uid is normal to the constant- surface that passes
through that point.
If a ow pattern is both irrotational and incompressible then we have
v = (5.31)
and
v = 0. (5.32)
These two expressions can be combined to give (see Section A.21)
2
= 0. (5.33)
In other words, the velocity potential in an incompressible irrotational uid satises Laplaces equation.
According to Equation (5.24), if the ow pattern in an incompressible inviscid uid is also irrotational, so that
= 0 and v = , then we can write
_
p
+
1
2
v
2
+

t
_
= 0, (5.34)
which implies that
p
+
1
2
v
2
+

t
= C(t), (5.35)
P
C
B
A
Figure 5.4: Two-dimensional ow.
where C(t) is uniformin space, but can vary in time. In fact, the time variation of C(t) can be eliminated by adding the
appropriate function of time (but not of space) to the velocity potential, . Note that such a procedure does not modify
the instantaneous velocity eld v derived from . Thus, the above equation can be rewritten
p
+
1
2
v
2
+

t
= C, (5.36)
where C is constant in both space and time. Expression (5.36) is a generalization of Bernoullis theorem (see Sec-
tion 5.3) that takes non-steady ow into account. However, this generalization is only valid for irrotational ow. For
the special case of steady ow, we get
p
+
1
2
v
2
+ = C, (5.37)
which demonstrates that for steady irrotational ow the constant in Bernoullis theorem is the same on all streamlines.
(See Section 5.3.)
5.8 Two-Dimensional Flow
Fluid motion is said to be two-dimensional when the velocity at every point is parallel to a xed plane, and is the same
everywhere on a given normal to that plane. Thus, in Cartesian coordinates, if the xed plane is the x-y plane then we
can express a general two-dimensional ow pattern in the form
v = v
x
(x, y, t) e
x
+ v
y
(x, y, t) e
y
. (5.38)
Let A be a xed point in the x-y plane, and let ABP and ACP be two curves, also in the x-y plane, that join A to an
arbitrary point P. See Figure 5.4. Suppose that uid is neither created nor destroyed in the region, R (say), bounded by
these curves. Since the uid is incompressible, which essentially means that its density is uniform and constant, uid
continuity requires that the rate at which the uid ows into the region R, from right to left across the curve ABP, is
equal to the rate at which it ows out the of the region, from right to left across the curve ACP. Now, the rate of uid
ow across a surface is generally termed the ux. Thus, the ux (per unit length parallel to the z-axis) from right to
left across ABP is equal to the ux from right to left across ACP. Since ACP is arbitrary, it follows that the ux from
right to left across any curve joining points A and P is equal to the ux from right to left across ABP. In fact, once the
base point A has been chosen, this ux only depends on the position of point P, and the time t. In other words, if we
denote the ux by then it is solely a function of the location of P and the time. Thus, if point A lies at the origin, and
84 FLUID MECHANICS
P
2
A
P
1
Figure 5.5: Two-dimensional ow.
point P has Cartesian coordinates x, y, then we can write
= (x, y, t). (5.39)
The function is known as the stream function. Moreover, the existence of a stream function is a direct consequence
of the assumed incompressible nature of the ow.
Consider two points, P
1
and P
2
, in addition to the xed point A. See Figure 5.5. Let
1
and
2
be the uxes from
right to left across curves AP
1
and AP
2
. Now, using similar arguments to those employed above, the ux across AP
2
is
equal to the ux across AP
1
plus the ux across P
1
P
2
. Thus, the ux across P
1
P
2
, from right to left, is
2

1
. Now,
if P
1
and P
2
both lie on the same streamline then the ux across P
1
P
2
is zero, since the local uid velocity is directed
everywhere parallel to P
1
P
2
. It follows that
1
=
2
. Hence, we conclude that the stream function is constant along a
streamline. The equation of a streamline is thus = c, where c is an arbitrary constant.
Let P
1
P
2
= s be an innitesimal arc of a curve that is suciently short that it can be regarded as a straight-line.
The uid velocity in the vicinity of this arc can be resolved into components parallel and perpendicular to the arc. The
component parallel to s contributes nothing to the ux across the arc from right to left. The component perpendicular
to s contributes v
s to the ux. However, the ux is equal to

2

1
. Hence,
v
2

1
s
. (5.40)
In the limit s 0, the perpendicular velocity from right to left across ds becomes
v
=
d
ds
. (5.41)
Thus, in Cartesian coordinates, by considering innitesimal arcs parallel to the x- and y-axes, we deduce that
v
x
=
y
, (5.42)
v
y
=
x
. (5.43)
These expressions can be combined to give
v = e
z
= z . (5.44)
Note that when the uid velocity is written in this form then it immediately becomes clear that the incompressibility
constraint v = 0 is automatically satised [since (AB) = 0see Equations (A.175) and (A.176)]. It is also
clear that the stream function is undened to an arbitrary additive constant.
The vorticity in two-dimensional ow takes the form
=
z
e
z
, (5.45)
where
z
=
v
y
x

v
x
y
. (5.46)
Thus, it follows from Equations (5.42) and (5.43) that
z
=
x
2
+
y
2
=
2
. (5.47)
Moreover, irrotational two-dimensional ow is characterized by
2
= 0. (5.48)
When expressed in terms of cylindrical coordinates (see Section C.3), Equation (5.44) yields
v = v
r
(r, , t) e
r
+ v
(r, , t) e
, (5.49)
where
v
r
=
1
r
, (5.50)
v
r
. (5.51)
Moreover, the vorticity is =
z
e
z
, where
z
=
1
r
r
_
r

r
_
+
1
r
2
2
. (5.52)
5.9 Two-Dimensional Uniform Flow
Consider a steady two-dimensional ow pattern that is uniform: i.e., such that the uid velocity is the same everywhere
in the x-y plane. For instance, suppose that the common uid velocity is
v = V
0
cos
0
e
x
+ V
0
sin
0
e
y
, (5.53)
which corresponds to ow at the uniform speed V
0
in a xed direction that subtends a (counter-clockwise) angle
0
with the x-axis. It follows, from Equations (5.42) and (5.43), that the stream function for steady uniform ow takes
the form
(x, y) = V
0
(sin
0
x cos
0
y) . (5.54)
When written in terms of cylindrical coordinates, this becomes
(r, ) = V
0
r sin(
0
). (5.55)
Note, from (5.54), that
2
/x
2
=
2
/y
2
= 0. Thus, it follows from Equation (5.47) that uniform ow is
irrotational. Hence, according to Section 5.7, such ow can also be derived from a velocity potential. In fact, it is
easily demonstrated that
(r, ) = V
0
r cos(
0
). (5.56)
86 FLUID MECHANICS
y
S
x
Figure 5.6: Streamlines of the ow generated by a line source coincident with the z-axis.
5.10 Two-Dimensional Sources and Sinks
Consider a uniform line source, coincident with the z-axis, that emits uid isotropically at the steady rate of Q unit
volumes per unit length per unit time. By symmetry, we expect the associated steady ow pattern to be isotropic, and
everywhere directed radially away from the source. See Figure 5.6. In other words, we expect
v = v
r
(r) e
r
. (5.57)
Consider a cylindrical surface S of unit height (in the z-direction) and radius r that is co-axial with the source. In a
steady state, the rate at which uid crosses this surface must be equal to the rate at which the section of the source
enclosed by the surface emits uid. Hence,
_
S
v dS = 2 r v
r
(r) = Q, (5.58)
which implies that
v
r
(r) =
Q
2 r
. (5.59)
According to Equations (5.50) and (5.51), the stream function associated with a line source of strength Q that is
coincident with the z-axis is
(r, ) =
Q
2
. (5.60)
Note that the streamlines, = c, are directed radially away from the z-axis, as illustrated in Figure 5.6. Note, also, that
the stream function associated with a line source is multivalued. However, this does not cause any particular diculty,
since the stream function is continuous, and its gradient single-valued.
Note, from Equation (5.60), that /r =
2
/
2
= 0. Hence, according to (5.52),
z
=
2
= 0. In other
words, the steady ow pattern associated with a uniform line source is irrotational, and can, thus, be derived from a
velocity potential. In fact, it is easily demonstrated that this potential takes the form
(r, ) =
Q
2
ln r. (5.61)
A uniform line sink, coincident with the z-axis, which absorbs uid isotropically at the steady rate of Q unit
volumes per unit length per unit time has an associated steady ow pattern
v =
Q
2 r
e
r
, (5.62)
whose stream function is
(r, ) =
Q
2
. (5.63)
This ow pattern is also irrotational, and can be derived from the velocity potential
(r, ) =
Q
2
ln r. (5.64)
Consider a line source and a line sink of equal strength, which both run parallel to the z-axis, and are located a
small distance apart in the x-y plane. Such an arrangement is known as a dipole or doublet line source. Suppose that
the line source, which is of strength Q, is located at r = d/2 (where r is a position vector in the x-y plane), and that
the line sink, which is also of strength Q, is located at r = d/2. Let the function
Q
(r) =
Q
2
=
Q
2
tan
1
(y/x) (5.65)
be the stream function associated with a line source of strength Q located at r = 0. Thus,
Q
(r r
0
) is the stream
function associated with a line source of strength Q located at r = r
0
. Furthermore, the stream function associated
with a line sink of strength Q located at r = r
0
is
Q
(r r
0
). Now, we expect the ow pattern associated with the
combination of a source and a sink to be the vector sum of the ow patterns generated by the source and sink taken in
isolation. It follows that the overall stream function is the sum of the stream functions generated by the source and the
sink taken in isolation. In other words,
(r) =
Q
(r d/2)
Q
(r + d/2) d
Q
(r), (5.66)
to rst order in d/r. Hence, if d = d (cos
0
e
x
+ sin
0
e
y
) = d [cos(
0
) e
r
sin(
0
) e
], so that the line joining

the sink to the source subtends a (counter-clockwise) angle
0
with the x-axis, then
(r, ) =
D
2
sin(
0
)
r
, (5.67)
where D = Qd is termed the strength of the dipole source. Note that the above stream function is antisymmetric across
the line =
0
joining the source to the sink. It follows that the associated dipole ow pattern,
v
r
(r, ) =
D
2
cos(
0
)
r
2
, (5.68)
v
(r, ) =
D
2
sin(
0
)
r
2
, (5.69)
is symmetric across this line. Figure 5.7 shows the streamlines associated with a dipole ow pattern characterized by
D > 0 and
0
= 0. Note that the ow speed in a dipole pattern falls o like 1/r
2
.
A dipole ow pattern is necessarily irrotational since it is a linear superposition of two irrotational ow patterns.
The associated velocity potential is
(r, ) =
D
2
cos(
0
)
r
. (5.70)
5.11 Two-Dimensional Vortex Filaments
Consider a vortex lament of intensity that is coincident with the z-axis. By symmetry, we expect the associated
ow pattern to circulate isotropically around the lament. See Figure 5.8. In other words, we expect
v = v
(r) e
. (5.71)
88 FLUID MECHANICS
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
y
1 0.5 0 0.5 1
x
Figure 5.7: Streamlines of the ow generated by a dipole line source coincident with the z-axis and aligned along the
x-axis. The ow is outward along the positive x-axis and inward along the negative x-axis.
Now, according to Section 5.5, the circulation,
_
v dr, around any closed curve in the x-y plane is equal to the net
intensity of the vortex laments that pass through the curve. Consider a circular curve of radius r that is concentric
with the origin. It follows that
r
=
_
v dr = 2 r v
(r) = , (5.72)
or
v
(r) =
2 r
. (5.73)
According to Equations (5.50) and (5.51), the stream function associated with a vortex lament of intensity that
is coincident with the z-axis is
(r, ) =
2
ln r. (5.74)
Note that the streamlines, = c, circulate around the z-axis, as illustrated in Figure 5.8.
It can be seen, from Equation (5.74), that (/r)(r /r) = / = 0. Hence, it follows from (5.52) that
z
=
2
= 0. In other words, the ow pattern associated with a straight vortex lament is irrotational. This
is a somewhat surprising result, since there is a net circulation of the ow around the lament, and, according to
Section 5.5, non-zero circulation implies non-zero vorticity. The paradox can be resolved by supposing that the
lament has a small, but nite, radius. In fact, let the lament have the nite radius a, and be such that the vorticity is
uniform inside this radius, and zero outside: i.e.,
z
=
_
/ a
2
r a
0 r > a
. (5.75)
Note that the intensity of the lament (i.e., the product of its vorticity and cross-sectional area) is still . According to
Equation (5.52), and assuming that = (r),
1
r
d
r
_
r
d
dr
_
=
_
/ a
2
r a
0 r > a
. (5.76)
x
y
Figure 5.8: Streamlines of the ow generated by a line vortex coincident with the z-axis.
The solution that is well-behaved at r = 0, and continuous (up to its rst derivative) at r = a, is
(r, ) =
_
(/4) (r
2
/a
2
1) r a
(/2) ln(r/a) r > a
. (5.77)
Note that this expression is equivalent to (5.74) (apart from an unimportant additive constant) outside the lament, but
diers inside. The associated circulation velocity, v
(r) = /r, is
v
(r) =
_
(/2) (r/a
2
) r a
(/2) (1/r) r > a
, (5.78)
whereas the circulation,
r
(r) = 2 r v
(r), is written
r
(r) =
_
(r/a)
2
r a
r > a
. (5.79)
Thus, we conclude that the ow pattern associated with a straight vortex lament is irrotational outside the lament,
but has nite vorticity inside the lament. Moreover, the non-zero internal vorticity generates a constant net circulation
of the ow outside the lament. In the limit in which the radius of the lament tends to zero, the vorticity within the
lament tends to innity (in such a way that the product of the vorticity and the cross-sectional area of the lament
remains constant), and the region of the uid in which the vorticity is non-zero becomes innitesimal in extent.
Let us determined the pressure prole in the vicinity of a vortex lament of nite radius. Assuming, from symme-
try, that p = p(r), Equation (2.149), yields
dp
dr
=
v
2
r
, (5.80)
p = p

_

r
v
2
r
dr, (5.81)
where p
is the pressure at innity. Making use of expression (5.78), we obtain

p(r) =
_
p
(/2) (/2 a)
2
(2 r
2
/a
2
) r a
p
(/2) (/2 a)
2
(a/r)
2
r > a
. (5.82)
90 FLUID MECHANICS
It follows that the minimum pressure occurs at the center of the vortex (r = 0), and takes the value
p
0
= p

_

2 a
_
2
. (5.83)
Now, under normal circumstances, the pressure in a uid must remain positive, which implies that a vortex lament
of intensity , embedded in a uid of density and background pressure p
, has a minimum radius of order

a
min

_

p
_
1/2 _

2
_
. (5.84)
Finally, since the ow pattern outside a straight vortex lament is irrotational, it can be derived from a velocity
potential. In fact, it is easily demonstrated that the appropriate potential takes the form
(r, ) =

2
. (5.85)
Note that the above potential is multivalued. However, this does not cause any particular diculty, since the potential
is continuous, and its gradient single-valued.
5.12 Two-Dimensional Irrotational Flow in Cylindrical Coordinates
As we have seen, in a two-dimensional ow pattern, we can automatically satisfy the incompressibility constraint,
v = 0, by expressing the pattern in terms of a stream function. Suppose, however, that, in addition to being
incompressible, the ow pattern is also irrotational. In this case, Equation (5.47) yields
2
= 0. (5.86)
In cylindrical coordinates, since = (r, , t), this expression implies that (see Section C.3)
1
r
r
_
r

r
_
+
1
r
2
2
= 0. (5.87)
Let us search for a separable steady-state solution of Equation (5.87) of the form
(r, ) = R(r) (). (5.88)
It is easily seen that
r
R
d
dr
_
r
dR
dr
_
=
1
d
2
d
2
, (5.89)
which can only be satised if
r
d
dr
_
r
dR
dr
_
= m
2
R, (5.90)
d
2
d
2
= m
2
, (5.91)
where m
2
is an arbitrary (positive) constant. The general solution of Equation (5.91) is a linear combination of
exp( i m) and exp(i m) factors. However, assuming that the ow extends over all values, the function ()
must be single-valued in , otherwise and, hence, vwould not be be single-valued (which is unphysical). It
follows that m can only take integer values (and that m
2
must be a positive, rather than a negative, constant). Now, the
general solution of Equation (5.90) is a linear combination of r
m
and r
m
factors, except for the special case m = 0,
when it is a linear combination of r
0
and ln r factors. Thus, the general stream function for steady two-dimensional
irrotational ow (that extends over all values of ) takes the form
(r, ) =
0
+
0
ln r +
m>0
(
m
r
m
+
m
r
m
) sin[m(
m
)], (5.92)
where
m
,
m
, and
m
are arbitrary constants. We can recognize the rst few terms on the right-hand side of the above
expression. The constant term
0
has zero gradient, and, therefore, does not give rise to any ow. The term
0
ln r is
the ow pattern generated by a vortex lament of intensity 2
0
, coincident with the z-axis. (See Section 5.11.) The
term
1
r sin(
1
) corresponds to uniformow of speed
1
whose direction subtends a (counter-clockwise) angle
1
with the minus x-axis. (See Section 5.9.) Finally, the term
1
sin(
1
)/r corresponds to a dipole ow pattern. (See
Section 5.10.)
The velocity potential associated with the irrotational stream function (5.92) satises [see Equations (5.29) and
(5.44)]
r
=
1
r
, (5.93)
1
r
r
. (5.94)
It follows that
(r, ) =
0

0
+
m>0
(
m
r
m

m
r
m
) cos[m(
0
)]. (5.95)
5.13 Inviscid Flow Past a Cylindrical Obstacle
Consider the steady ow pattern produced when an impenetrable rigid cylindrical obstacle is placed in a uniformly
owing, incompressible, inviscid uid, with the cylinder orientated such that its axis is normal to the ow. For instance,
suppose that the radius of the cylinder is a, and that its axis corresponds to the line x = y = 0. Furthermore, let the
unperturbed uid velocity be of magnitude V
0
, and be directed parallel to the x-axis. Now, we expect the ow pattern
to remain unperturbed very far away from the cylinder. In other words, we expect v(r, ) V
0
e
x
as r/a , which
corresponds to a boundary condition on the stream function of the form (see Section 5.9)
(r, ) V
0
r sin as r/a . (5.96)
Given that the uid velocity eld a large distance upstream of the cylinder is irrotational (since we have already
seen that the ow pattern associated with uniform ow is irrotationalsee Section 5.9), it follows from the Kelvin
circulation theorem (see Section 5.6) that the velocity eld remains irrotational as it is convected past the cylinder.
Hence, according to Section 5.8, the stream function of the ow satises Laplaces equation,
2
= 0. (5.97)
The appropriate boundary condition at the surface of the cylinder is simply that the normal uid velocity there be
zero, since the uid must stay in contact with the cylinder, but cannot penetrate its surface. Hence, v
r
(a, )
(1/a) /
r=a
= 0, which implies that
(a, ) = 0, (5.98)
since is undetermined to an arbitrary additive constant. It follows that we are searching for the most general solution
of (5.97) that satises the boundary conditions (5.96) and (5.98). Comparison with Equation (5.92) reveals that this
solution takes the form
(r, ) = V
0
a
_
ln
_
r
a
_
_
r
a

a
r
_
sin
_
, (5.99)
where
=

2 a V
0
, (5.100)
and is the circulation of the ow around the cylinder. (Note that the velocity eld can be irrotational, but still possess
nonzero circulation around the cylinder, because a loop that encloses the cylinder cannot be spanned by a surface lying
entirely within the uid. Thus, zero uid vorticity does not necessarily imply zero circulation around such a loop from
Stokes theorem.) Let us assume that 0, for the sake of deniteness.
92 FLUID MECHANICS
5
4
3
2
1
0
1
2
3
4
5
y/a
5 4 3 2 1 0 1 2 3 4 5
x/a
Figure 5.9: Streamlines of the ow generated by a cylindrical obstacle of radius a, whose axis runs along the z-axis,
placed in the uniform ow eld v = V
0
e
x
. The normalized circulation is = 0.
Figure 5.95.11 show streamlines of the ow calculated for various dierent values of the normalized circulation,
. For < 2 there exist a pair of points on the surface of the cylinder at which the ow speed is zero. These are known
as stagnation points, and can be located in Figures 5.9 and 5.10 as the points at which streamlines intersect the surface
of the cylinder at right-angles. Now, the tangential uid velocity at the surface of the cylinder is
v
t
() = v
(a, ) =
r=a
= V
0
( + 2 sin ). (5.101)
The stagnation points correspond to the points at which v
t
= 0 (since the normal velocity is automatically zero at the
surface of the cylinder). Thus, the stagnation points lie at = sin
1
(/2). When > 2 the stagnation points coalesce
and move o the surface of the cylinder, as illustrated in Figure 5.11 (the stagnation point corresponds to the point at
which two streamlines cross at right-angles).
The irrotational form of Bernoullis theorem, (5.37), can be combined with the boundary condition v V
0
as
r/a , as well as the fact that is constant in the present case, to give
p = p
0
+
1
2

_
V
2
0
v
2
_
, (5.102)
where p
0
is the constant static uid pressure a large distance from the cylinder. In particular, the uid pressure on the
surface of the cylinder is
P() = p(a, ) = p
0
+
1
2

_
V
2
0
v
2
t
_
= p
1
+ V
2
0
(cos 2 2 sin ) , (5.103)
where p
1
= p
0
(1/2) V
2
0
(1 +
2
). The net force per unit length exerted on the cylinder by the uid has the Cartesian
components
F
x
=
_
P cos a d, (5.104)
F
y
=
_
P sin a d. (5.105)
5
4
3
2
1
0
1
2
3
4
5
y/a
5 4 3 2 1 0 1 2 3 4 5
x/a
0
e
x
. The normalized circulation is = 1.
Thus, it follows from (5.103) that
F
x
= 0, (5.106)
F
y
= 2 V
2
0
a = V
0
(). (5.107)
Now, the component of the force which a moving uid exerts on an obstacle, placed in its path, in a direction parallel
to that of the unperturbed ow is usually called drag. On the other hand, the component of the force which the uid
exerts in a direction perpendicular to that of the unperturbed ow is usually called lift. Hence, the above equations
imply that if a cylindrical obstacle is placed in a uniformly owing inviscid uid then there is zero drag. On the other
hand, as long as there is net circulation of the ow around the cylinder, the lift is non-zero. Now, lift is generated
because (negative) circulation tends to increase the uid speed directly above, and to decrease it directly below, the
cylinder. Thus, from Bernoullis theorem, the uid pressure is decreased above, and increased below, the cylinder,
giving rise to a net upward force (i.e., a force in the +y-direction).
Suppose that the cylinder is placed in a uid which is initially at rest, and that the uids uniform ow velocity, V
0
,
is then very slowly ramped up (in such a manner that no vorticity is induced in the upstream ow at innity). Since
the ow pattern is initially irrotational, and since the ow pattern well upstream of the cylinder is assumed to remain
irrotational, the Kelvin circulation theoremindicates that the ow pattern around the cylinder also remains irrotational.
Consider the time evolution of the circulation, =
_
C
v dr, around some xed curve C that lies entirely within the
uid, and encloses the cylinder. We have
d
C
dt
=
_
C
v
t
dr =
_
C
_
_
p
+
1
2
v
2
_
+ v
_
dr =
_
C
v dr, (5.108)
where use has been made of (5.24) (with assumed constant). However, =
z
e
z
in two-dimensional ow, and
dr e
z
= dS, where dS is an outward surface element of a unit depth (in the z-direction) surface whose normal lies in
the x-y plane, and that cuts the x-y plane at C. In other words,
d
C
dt
=
_
S
z
v dS. (5.109)
We, thus, conclude that the rate of change of the circulation around C is equal to minus the ux of the vorticity across
S [assuming that vorticity is convected by the ow, which follows from (5.25), the fact that =
z
e
z
, and the fact
94 FLUID MECHANICS
5
4
3
2
1
0
1
2
3
4
5
y/a
5 4 3 2 1 0 1 2 3 4 5
x/a
0
e
x
. The normalized circulation is = 2.5.
that /z = 0 in two-dimensional ow]. However, we have already seen that the ow eld surrounding the cylinder
is irrotational (i.e., such that
z
= 0). It follows that
C
is constant in time. Moreover, since
C
= 0 originally,
because the uid surrounding the cylinder was initially at rest, we deduce that
C
= 0 at all subsequent times. Hence,
we conclude that, in an inviscid uid, if the circulation of the ow around the cylinder is initially zero then it remains
zero. It follows, from the above analysis, that, in such a uid, zero drag force and zero lift force are exerted on the
cylinder as a consequence of the uid ow. This result is known as dAlemberts paradox, after the French scientist
Jean-Baptiste le Rond dAlembert (17171783). DAlemberts result is paradoxical because it would seem, at rst
sight, to be a reasonable approximation to neglect viscosity alltogether in high Reynolds number ow. However, if we
do this then we end up with the nonsensical prediction that a high Reynolds number uid is incapable of exerting any
force on an obstacle placed in its path.
5.14 Inviscid Flow Past a Semi-Innite Wedge
Consider the situation, illustrated in Figure 5.12, in which incompressible irrotational ow is incident on a impene-
trable rigid wedge whose apex subtends an angle . Let the cross-section of the wedge in the x-y plane be both
z-independent and symmetric about the x-axis. Furthermore, let the apex of the wedge lie at x = y = 0. Finally, let the
upstream ow a large distance from the wedge be parallel to the x-axis.
Since the ow is two-dimensional, incompressible, and irrotational, it can be represented in terms of a stream
function that satises Laplaces equation. Moreover, in cylindrical coordinates, this equation takes the form (5.87).
The boundary conditions on the stream function are
(r, /2) = (r, 2 /2) = (r, ) = 0. (5.110)
The rst two boundary conditions ensure that the normal velocity at the surface of the wedge is zero. The third
boundary condition follows from the observation that, by symmetry, the streamline that meets the apex of the wedge
splits in two, and then ows along its top and bottom boundaries, combined with well-known result that is constant
on a streamline. It is easily demonstrated that
(r, ) =
A
1 + m
r
1+m
sin [(1 + m) ( )] (5.111)
y
x
Figure 5.12: Inviscid ow past a wedge.

is a solution of (5.87). Moreover, this solution satises the boundary conditions provided (1 + m) (1 /2) = 1, or
m =
2
. (5.112)
Since, as is well-known, the solutions to Laplaces equation (for problems with well-posed boundary conditions) are
unique, we can be sure that (5.111) is the correct solution to the problem under investigation. According to this
solution, the tangential velocity on the surface of the wedge is given by
v
t
(r) = Ar
m
, (5.113)
where m 0. Note that the tangential velocity is zero at the apex of the wedge. Since the normal velocity is also zero
at this point, we conclude that the apex is a stagnation point of the ow. Figure 5.13 shows the streamlines of the ow
for the case = 1/2.
5.15 Inviscid Flow Over a Semi-Innite Wedge
Consider the situation illustrated in Figure 5.14 in which an incompressible irrotational uid ows over an impene-
trable rigid wedge whose apex subtends an angle . Let the cross-section of the wedge in the x-y plane be both
z-independent and symmetric about the y-axis. Furthermore, let the apex of the wedge lie at x = y = 0. Finally, let the
upstream ow a large distance from the wedge be parallel to the x-axis.
Since the ow is two-dimensional, incompressible, and irrotational, it can be represented in terms of a stream
function that satises Laplaces equation. The boundary conditions on the stream function are
(r, [3 ] /2) = (r, [1 ] /2) = 0. (5.114)
These boundary conditions ensure that the normal velocity at the surface of the wedge is zero. It is easily demonstrated
that
(r, ) =
A
1 m
r
1m
cos [(1 m) ( /2)] (5.115)
is a solution of Laplaces equation, (5.87). Moreover, this solution satises the boundary conditions provided that
(1 m) (1 /2) = 1/2, or
m =
1 +
, (5.116)
96 FLUID MECHANICS
3
2
1
0
1
2
3
y
3 2 1 0 1 2 3
x
Figure 5.13: Streamlines of inviscid incompressible irrotational ow past a 90
wedge.
x
y
Figure 5.14: Inviscid ow over a wedge.

3
2
1
0
1
2
3
y
3 2 1 0 1 2 3
x
Figure 5.15: Streamlines of inviscid incompressible irrotational ow over a 90
wedge.
where
= 1 . Since the solutions to Laplaces equation are unique, we can again be sure that (5.115) is the correct
solution to the problem under investigation. According to this solution, the tangential velocity on the surface of the
wedge is given by
v
t
(r) = Ar
m
, (5.117)
where m 0. Note that the tangential velocity, and hence the ow speed, is innite at the apex of the wedge. However,
this singularity in the ow can be eliminated by slightly rounding the apex. Figure 5.15 shows the streamlines of the
ow for the case = 1/2.
5.16 Velocity Potentials and Stream Functions
As we have seen, a two-dimensional velocity eld in which the ow is everywhere parallel to the x-y plane, and there
is no variation along the z-direction, takes the form
v = v
x
(x, y, t) e
x
+ v
y
(x, y, t) e
y
. (5.118)
Moreover, if the ow is irrotational then v = 0 is automatically satised by writing v = , where (x, y, t) is
termed the velocity potential. (See Section 5.7.) Hence,
v
x
=
x
, (5.119)
v
y
=
y
. (5.120)
On the other hand, if the ow is incompressible then v = 0 is automatically satised by writing v = z , where
(x, y, t) is termed the stream function. (See Section 5.8.) Hence,
v
x
=
y
, (5.121)
v
y
=
x
. (5.122)
98 FLUID MECHANICS
Finally, if the ow is both irrotational and incompressible then Equations (5.119)(5.120) and (5.121)(5.122) hold
simultaneously, which implies that
x
=
y
, (5.123)
x
=
y
. (5.124)
It immediately follows, from the previous two expressions, that
x
2
=
x y
=
y x
=
y
2
, (5.125)
or
x
2
+
y
2
= 0. (5.126)
Likewise, it can also be shown that
x
2
+
y
2
= 0. (5.127)
We conclude that, for two-dimensional, irrotational, and incompressible ow, the velocity potential and the stream
function both satisfy Laplaces equation. Equations (5.123) and (5.124) also imply that
= 0 : (5.128)
i.e., the contours of the velocity potential and the stream function cross at right-angles.
5.17 Exercises
5.1. Liquid is led steadily through a pipeline that passes over a hill of height h into the valley below, the speed at the crest being
v. Show that, by properly adjusting the ratio of the cross-sectional areas of the pipe at the crest and in the valley, the pressure
may be equalized at these two places.
5.2. For the case of the two-dimensional motion of an incompressible uid, determine the condition that the velocity components
v
x
= a x + b y,
v
y
= c x + d y
satisfy the equation of continuity. Show that the magnitude of the vorticity is c b.
5.3. For the case of the two-dimensional motion of an incompressible uid, show that
v
x
= 2 c x y,
v
y
= c (a
2
+ x
2
y
2
)
are the velocity components of a possible ow pattern. Determine the stream function and sketch the streamlines. Prove that
the motion is irrotational, and nd the velocity potential.
5.4. A cylindrical vortex in an incompressible uid is co-axial with the z-axis, and such that
z
takes the constant value for
r a, and is zero for r > a, where r is a cylindrical coordinate. Show that
1
dp
dr
=
2
r
a
4
,
where p(r) is the pressure at radius r inside the vortex, and the circulation of the uid outside the vortex is 2 . Deduce that
p(r) =
2
r
2
2 a
4
+ p
0
,
where p
0
is the pressure at the center of the vortex.
5.5. Consider the cylindrical vortex discussed in Exercise 5.4. If p(r) is the pressure at radius r external to the vortex, demonstrate
that
p(r) =
2 r
2
+ p
,
where p
is the pressure at innity.

5.6. Show that the stream function for the cylindrical vortex discussed in Exercises 5.4 and 5.5 is (r) = (1/2) a
2
ln(r/a) for
r > a, and (r) = (1/4) (r
2
a
2
) for r a.
5.7. Consider a volume V whose boundary is the surface S . Suppose that V contains an incompressible uid whose motion is
irrotational. Let the velocity potential be constant over S . Prove that has the same constant value throughout V. [Hint:
Consider the identity (AA) A A + A
2
A.]
5.8. In Exercise 5.7, suppose that, instead of taking a constant value on the boundary, the normal velocity is everywhere zero
on the boundary. Show that is constant throughout V.
5.9. Prove that in the two-dimensional motion of a liquid the mean tangential uid velocity around any small circle of radius r is
r, where 2 is the value of
v
y
x

v
x
y
at the center of the circle. Neglect terms of order r
3
.
5.10. Show that the equation of continuity for the two-dimensional motion of an incompressible uid can be written
(r v
r
)
r
+
v
= 0,
where r, are cylindrical coordinates. Demonstrate that this equation is satised when v
r
= a k r
n
exp[k (n + 1) ] and
v
= a r
n
exp[k (n + 1) ]. Determine the stream function, and show that the uid speed at any point is
(n + 1)
1 + k
2
/r,
where is the stream function at that point (dened such that = 0 at r = 0).
5.11. Demonstrate that streamlines cross at right-angles at a stagnation point in two-dimensional, incompressible, irrotational ow.
5.12. Consider two-dimensional, incompressible, inviscid ow. Demonstrate that the uid motion is governed by the following
equations:
t
+ [, ] = 0,
2
= ,
2
= +
2
,
where v = e
z
, [A, B] = e
z
A B, and = p/ + (1/2) v
2
+ .
5.13. For irrotational, incompressible, inviscid motion in two-dimensions show that
q q = q
2
q,
where q = v.
100 FLUID MECHANICS
2D Potential Flow 101
6 2D Potential Flow
6.1 Introduction
This chapter discusses the use of complex analysis to simplify calculations in two-dimensional, incompressible, in-
viscid, irrotational, uid dynamics. Incidentally, incompressible, inviscid, irrotational ow is usually referred to a
potential ow, since the velocity eld can be represented in terms of a velocity potential that satises Laplaces equa-
tion. In the following, all ow patterns are assumed to be such that the z-coordinate is ignorable. In other words, the
uid velocity is everywhere parallel to the x-y plane, and /z = 0. It follows that all line sources and vortex laments
run parallel to the z-axis. Moreover, all solid surfaces are of innite extent along the z-axis, and have uniform cross-
sections. Hence, it is only necessary to specify the locations of line sources, vortex laments, and solid surfaces in the
x-y plane.
6.2 Complex Functions
The complex variable is conventionally written
z = x + i y, (6.1)
where i represents the square root of minus one. Here, x and y are both real, and are identied with the corresponding
Cartesian coordinates. (Incidentally, z should not be confused with a z-coordinate: this is a strictly two-dimensional
discussion.) We can also write
z = r e
i
, (6.2)
where r =
_
x
2
+ y
2
and = tan
1
(y/x) are termed the modulus and argument of z, respectively, but can also be
identied with the corresponding plane polar coordinates. Finally, De Moivres theorem,
e
i
= cos + i sin , (6.3)
implies that
x = r cos , (6.4)
y = r sin . (6.5)
We can dene functions of the complex variable, F(z), just like we would dene functions of a real variable. For
instance,
F(z) = z
2
, (6.6)
F(z) =
1
z
. (6.7)
For a given function, F(z), we can substitute z = x + i y and write
F(z) = (x, y) + i (x, y), (6.8)
where (x, y) and (x, y) are real two-dimensional functions. Thus, if
F(z) = z
2
, (6.9)
then
F(x + i y) = (x + i y)
2
= (x
2
y
2
) + 2 i x y, (6.10)
giving
(x, y) = x
2
y
2
, (6.11)
(x, y) = 2 x y. (6.12)
102 FLUID MECHANICS
6.3 Cauchy-Riemann Relations
We can dene the derivative of a complex function in just the same manner that we would dene the derivative of a
real function: i.e.,
dF
dz
=
limz
F(z + z) F(z)
z
. (6.13)
However, we now have a problem. If F(z) is a well-behaved function (i.e., nite, single-valued, and dierentiable)
then it should not matter from which direction in the complex plane we approach the point z when taking the limit in
Equation (6.13). There are, of course, many dierent possible approach directions, but if we look at a regular complex
function, F(z) = z
2
(say), then
dF
dz
= 2 z (6.14)
is perfectly well-dened, and is, therefore, completely independent of the details of how the limit is taken in Equa-
tion (6.13).
The fact that Equation (6.13) has to give the same result, no matter from which direction we approach z, means
that there are some restrictions on the forms of the functions (x, y) and (x, y) in Equation (6.8). Suppose that we
approach z along the real axis, so that z = x. We obtain
dF
dz
=
limx0
(x + x, y) + i (x + x, y) (x, y) i (x, y)
x
=
x
+ i

x
. (6.15)
Suppose that we now approach z along the imaginary axis, so that z = i y. We get
dF
dz
=
limy0
(x, y + y) + i (x, y + y) (x, y) i (x, y)
i y
= i

y
+
y
. (6.16)
But, if F(z) is a well-behaved function then its derivative must be well-dened, which implies that the above two
expressions are equivalent. This requires that
x
=
y
, (6.17)
x
=
y
. (6.18)
These expressions are called the Cauchy-Riemann relations, and are, in fact, sucient to ensure that all possible ways
of taking the limit (6.13) give the same result.
6.4 Complex Velocity Potential
Note that Equations (6.17)(6.18) are identical to Equations (5.123)(5.124). This suggests that the real and imagi-
nary parts of a well-behaved function of the complex variable can be interpreted as the velocity potential and stream
function, respectively, of some two-dimensional, irrotational, incompressible ow pattern. For instance, suppose that
F(z) = V
0
z, (6.19)
where V
0
is real. It follows that
(r, ) = V
0
r cos , (6.20)
(r, ) = V
0
r sin . (6.21)
It can be seen, by comparison with the analysis of Section 5.9, that the complex velocity potential (6.19) corresponds
to uniform ow of speed V
0
directed along the x-axis. Furthermore, as is easily demonstrated, the complex velocity
potential associated with uniform ow of speed V
0
whose direction subtends a (counter-clockwise) angle
0
with the
x-axis is F(z) = V
0
z e
i
0
.
Suppose that
F(z) =
Q
2
ln z, (6.22)
where Q is real. Since ln z = ln r + i , it follows that
(r, ) =
Q
2
ln r, (6.23)
(r, ) =
Q
2
. (6.24)
Thus, according to the analysis of Section 5.10, the complex velocity potential (6.22) corresponds to the ow pattern
of a line source, of strength Q, located at the origin. (See Figure 5.6.) As a simple generalization of this result, the
complex potential of a line source, of strength Q, located at the point (x
0
, y
0
), is F(z) = (Q/2) ln(z z
0
), where
z
0
= x
0
+ i y
0
. Note, from (6.22), that the complex velocity potential of a line source is singular at the location of the
source.
Suppose that
F(z) = i

2
ln z, (6.25)
where is real. It follows that
(r, ) =

2
, (6.26)
(r, ) =
2
ln r. (6.27)
Thus, according to the analysis of Section 5.11, the complex velocity potential (6.25) corresponds to the ow pattern
of a vortex lament of intensity located at the origin. (See Figure 5.8.) Note, from (6.25), that the complex velocity
potential of a vortex lament is singular at the location of the lament.
Suppose, nally, that
F(z) = V
0
_
z +
a
2
z
_
+ i

2
ln
_
z
a
_
, (6.28)
where V
0
, a, and , are real. It follows that
(r, ) = V
0
_
r +
a
2
r
_
cos

2
, (6.29)
(r, ) = V
0
_
r
a
2
r
_
sin +
2
ln
_
r
a
_
. (6.30)
Thus, according to the analysis of Section 5.13, the complex velocity potential (6.28) corresponds to uniform inviscid
ow of unperturbed speed V
0
, running parallel to the x-axis, around an impenetrable cylinder of radius a, centered on
the origin. (See Figures 5.9, 5.10, and 5.11.) Here, is the circulation of the ow about the cylinder. Note that = 0
on the surface of the cylinder (r = a), which ensures that the normal velocity is zero on this surface, as must be the
case if the cylinder is impenetrable.
6.5 Complex Velocity
It follows from Equations (5.119), (5.122), and (6.15) that
dF
dz
=
x
+ i

x
= v
x
+ i v
y
. (6.31)
Consequently, dF/dz is termed the complex velocity. Note that
dF
dz
2
= v
2
x
+ v
2
y
= v
2
, (6.32)
104 FLUID MECHANICS
where v is the ow speed.
A stagnation point is dened as a point in a ow pattern where the ow speed, v, falls to zero. It follows, from the
previous expression, that
dF
dz
= 0 (6.33)
at a stagnation point. For instance, the stagnation points of the ow pattern produced when a cylindrical obstacle
of radius a, centered on the origin, is placed in a uniform ow of speed V
0
, directed parallel to the x-axis, and the
circulation of the ow around is cylinder is , are found by setting the derivative of the complex potential (6.28) to
zero. It follows that the stagnation points satisfy the quadratic equation
dF
dz
= V
0
_
1
a
2
z
2
_
+ i

2 z
= 0. (6.34)
The solutions are
z
a
= i
_
1
2
, (6.35)
where = /(4 V
0
a), with the proviso that z/a > 1, since the region z/a < 1 is occupied by the cylinder. Thus,
if 1 then there are two stagnation points on the surface of the cylinder at x/a =
_
1
2
and y/a = . On the
other hand, if > 1 then there is a single stagnation point below the cylinder at x/a = 0 and y/a =
_
2
1.
Now, according to Section 5.7, Bernoullis theorem in an steady, irrotational, incompressible uid takes the form
p +
1
2
v
2
= p
0
, (6.36)
where p
0
is a uniform constant, and where gravity (or any other body force) has been neglected. Thus, the pressure
distribution in such a uid can be written
p = p
0

1
2

dF
dz
2
. (6.37)
6.6 Method of Images
Let F
1
(z) =
1
(x, y) + i
1
(x, y) and F
2
(z) =
2
(x, y) + i
2
(x, y) be complex velocity potentials corresponding to
distinct, two-dimensional, irrotational, incompressible ow patterns whose stream functions are
1
(x, y) and
2
(x, y),
respectively. It follows that both stream functions satisfy Laplaces equation: i.e.,
2
1
=
2
2
= 0. Suppose that
F
3
(z) = F
1
(z) + F
2
(z). Writing F
3
(z) =
3
(x, y) + i
3
(x, y), it is clear that
3
(x, y) =
1
(x, y) +
2
(x, y). Moreover,
3
=
2
1
+
2
2
= 0, so
3
also satises Laplaces equation. We deduce that two complex velocity potentials,
corresponding to distinct, two-dimensional, irrotational, incompressible ow patterns, can be superposed to produce
a third velocity potential that corresponds to another two-dimensional, irrotational, incompressible ow pattern. As
described below, this idea can be exploited to determine the ow patterns produced by line sources and vortex laments
in the vicinity of rigid boundaries.
As an example, consider a situation in which there are two line sources of strength Q located at the points (0, a).
See Figure 6.1. The complex velocity potential of the resulting ow pattern is the sum of the complex potentials of
each source taken in isolation. Hence, from Section 6.4,
F(z) =
Q
2
ln(z i a)
Q
2
ln(z + i a) =
Q
2
ln(z
2
+ a
2
). (6.38)
Thus, the stream function of the ow pattern (which is the imaginary part of the complex potential) is
(x, y) =
Q
2
tan
1
_
2 x y
x
2
y
2
+ a
2
_
. (6.39)
Note that (x, 0) = 0, which implies that there is zero ow normal to the plane y = 0. Hence, in the region y > 0, we
could interpret the above stream function as that generated by a single line source of strength Q, located at the point
a
x
Q
Q
y
a
Figure 6.1: Two line sources.
(0, a), in the presence of a planar rigid boundary at y = 0. This follows because the stream function satises
2
= 0
everywhere in the region y > 0, has the requisite singularity (corresponding to a line source of strength Q) at (0, a),
and satises the physical boundary condition that the normal velocity be zero at the rigid boundary. Moreover, as is
well-known, the solutions of Poissons equation are unique. The streamlines of the resulting ow pattern are shown in
Figure 6.2. Incidentally, we can think of the two sources in Figure 6.1 as the images of one another in the boundary
plane. Hence, this method of calculation is usually referred to as the method of images.
Now, the complex velocity associated with the complex velocity potential (6.38) is
dF
dz
=
Q
z
z
2
+ a
2
. (6.40)
Hence, the ow speed at the boundary is
v(x, 0) =
dF
dz
(x, 0) =
Q
a
x/a
1 + x
2
/a
2
. (6.41)
It follows from (6.37) (and the fact that the ow speed at innity is zero) that the excess pressure on the boundary, due
to the presence of the source, is
p(x, 0) =
dF
dz
2
y=0
=
2
_
Q
a
_
2
x
2
/a
2
(1 + x
2
/a
2
)
2
. (6.42)
Thus, the excess force per unit length (in the z-direction) acting on the boundary in the y-direction is
F
y
=
_

p(x, 0) dx =
Q
2
2
2
a
_

2
(1 +
2
)
2
d =
Q
2
4 a
. (6.43)
The fact that the force is positive implies that the boundary is attracted to the source, and vice versa.
As a second example, consider the situation, illustrated in Figure 6.3, in which there are two vortex laments of
intensities and situated at (0, a). As before, the complex velocity potential of the resulting ow pattern is the
sum of the complex potentials of each lament taken in isolation. Hence, from Section 6.4,
F(z) = i

2
ln(z i a) i

2
ln(z + i a) = i

2
ln
_
z i a
z + i a
_
. (6.44)
106 FLUID MECHANICS
0
1
2
3
4
5
6
y/a
3 2 1 0 1 2 3
x/a
Figure 6.2: Stream lines of the 2D ow pattern due to a line source at (0, a) in the presence of a rigid boundary at
y = 0.
x
y
a
a
Figure 6.3: Two vortex laments.

0
1
2
3
4
5
6
y/a
3 2 1 0 1 2 3
x/a
Figure 6.4: Stream lines of the 2D ow pattern due to a vortex lament at (0, a) in the presence of a rigid boundary at
y = 0.
Thus, the stream function of the ow pattern is
(x, y) =
2
ln
_
x
2
+ (y a)
2
x
2
+ (y + a)
2
_
. (6.45)
As before, (x, 0) = 0, which implies that there is zero ow normal to the plane y = 0. Hence, in the region y > 0,
we could interpret the above stream function as that generated by a single vortex lament of intensity , located at
the point (0, a), in the presence of a planar rigid boundary at y = 0. The streamlines of the resulting ow pattern are
shown in Figure 6.4. We conclude that a vortex lament reverses its sense of rotation (i.e., ) when reected
in a boundary plane.
As a nal example, consider the situation, illustrated in Figure 6.5, in which there is an impenetrable cylinder of
radius a, centered on the origin, and a line source of strength Q located at (b, 0), where b > a. Consider the so-called
analog problem, also illustrated in Figure 6.5, in which the cylinder is replaced by a source of strength Q, located at (c,
0), where c < a, and a source of strength Q, located at the origin. We can think of these two sources as the images
of the external source in the cylinder. Moreover, given that the solutions of Poissons equation are unique, if the analog
problem can be adjusted in such a manner that r = a is a streamline then the ow in the region r > a will become
identical to that in the actual problem. Now, the complex velocity potential in the analog problem is simply
F(z) =
Q
2
ln(z b)
Q
2
ln(z c) +
Q
2
ln z =
Q
2
ln
_
(z b) (z c)
z
_
. (6.46)
Hence, writing z = r e
i
, the corresponding stream function takes the form
(r, ) =
Q
2
tan
1
_
(r b c/r) sin
(r + b c/r) cos (b + c)
_
. (6.47)
Now, we require the surface of the cylinder, r = a, to be a streamline: i.e., (a, ) = constant. This is easily achieved
by setting c = a
2
/b. Thus, the stream function becomes
(r, ) =
Q
2
tan
1
_
(r/a a/r) sin
(r/a + a/r) cos (b/a + a/b)
_
. (6.48)
The corresponding streamlines in the region external to the cylinder are shown in Figure 6.6.
108 FLUID MECHANICS
a
x
b
Q Q Q
x
y y
Q
c
Figure 6.5: A line source in the presence of an impenetrable cylinder.
4
3
2
1
0
1
2
3
4
y/a
4 3 2 1 0 1 2 3 4
x/a
Figure 6.6: Stream lines of the 2D ow pattern due to a line source at (2a, 0) in the presence of a rigid cylinder of
radius a centered on the origin.
d
x
y
0
d
dz
dz
z
0
Figure 6.7: A conformal map.
6.7 Conformal Maps
Let = + i and z = x + i y, where , , x, and y are real. Suppose that = f (z), where f is a well-behaved (i.e.,
single-valued, non-singular, and dierentiable) function. We can think of = f (z) as a map from the complex z-plane
to the complex -plane. In other words, every point (x, y) in the complex z-plane maps to a corresponding point (, )
in the complex -plane. Moreover, if f (z) is indeed a well-behaved function then this mapping is unique, and also has
a unique inverse. Suppose that the point z = z
0
in the z-plane maps to the point =
0
in the -plane. Let us investigate
how neighboring points map. We have
0
+ d
= f (z
0
+ dz
), (6.49)
0
+ d
= f (z
0
+ dz
). (6.50)
In other words, the points z
0
+ dz
and z
0
+ dz
in the complex z-plane map to the points

0
+ d
and
0
+ d
in the
complex -plane, respectively. Now, if dz
, dz
1 then
d
= f
(z
0
) dz
, (6.51)
d
= f
(z
0
) dz
, (6.52)
where f
(z) = d f /dz. Hence,

d
=
dz
dz
. (6.53)
Thus, it follows that
d
=
dz
dz
, (6.54)
and
arg(d
) arg(d
) = arg(dz
) arg(dz
). (6.55)
Now, we can think of dz
and dz
as innitesimal vectors connecting neighboring points in the complex z-plane to

the point z = z
0
. Likewise, d
and d
are innitesimal vectors connecting the corresponding points in the complex

-plane. It is clear, from the previous two equations, that, in the vicinity of z = z
0
, the mapping from the complex
z-plane to the complex -plane is such that the lengths of dz
and dz
expand or contract by the same factor, and

the angle subtended between these two vectors remains the same. See Figure 6.7. This type of mapping is termed
conformal.
110 FLUID MECHANICS
Suppose that F() = (, ) + i (, ) is a well-behaved function of the complex variable . It follows that
2
=
2
= = 0. Hence, the functions (, ) and (, ) can be interpreted as the velocity potential and stream
function, respectively, of some two-dimensional, inviscid, incompressible ow pattern, where and are Cartesian
coordinates. However, if = f (z), where f (z) is well-behaved, then F() = F[( f (z)] = G(z) =
(x, y) +i
(x, y), where

G(z) is also well-behaved. It follows that
2
=
2
= 0. In other words, the functions

(x, y) and

(x, y)
can be interpreted as the velocity potential and stream function, respectively, of some new, two-dimensional, inviscid,
incompressible ow pattern, where x and y are Cartesian coordinates. In other words, we can use a conformal map
to convert a given two-dimensional, inviscid, incompressible ow pattern into another, quite dierent, pattern. Note,
incidentally, that a conformal map converts a line source into a line source of the same strength, and a vortex lament
into a vortex lament of the same intensity. (See Exercise 6.12.)
As an example, consider the conformal map
= i e
z/a
. (6.56)
Writing = r e
i
, it is easily demonstrated that x = a ln r/ and y = a (/ 1/2). Hence, the positive -axis ( = 0)
maps to the line y = a/2, the negative -axis ( = ) maps to the line y = a/2, and the region > 0 (0 )
maps to the region a/2 < y < a/2. Moreover, the points = (0, 1) map to the points z = a (0,1/2 1/2). See
Figure 6.8. As we saw in Section 6.6, in the region > 0, the velocity potential
F() = i

2
ln
_
i
+ i
_
(6.57)
corresponds to the ow pattern generated by a vortex lament of intensity , located at the point = (0, 1), in the
presence of a rigid plane at = 0. Hence,
G(z) = F(i e
z/a
) = i

2
ln tanh
_
z
2 a
_
, (6.58)
corresponds to the ow pattern generated by a vortex lament of intensity , located at the origin, in the presence of
two rigid planes at y = a/2. This follows because the line = 0 is mapped to the lines y = a/2, the point = (0, 1)
is mapped to the origin, and if the line = 0 corresponds to a streamline then the lines y = a/2 also correspond to
streamlines. The stream function associated with the above complex velocity potential,
(x, y) =
ln
_
cosh( x a) cos( y/a)
cosh( x/a) + cos( y/a)
_
, (6.59)
is shown in Figure 6.9.
As a second example, consider the map
= z
2
. (6.60)
This maps the positive -axis to the positive x-axis, the negative -axis to the positive y-axis, the region > 0 to the
region x > 0, y > 0, and the point = (0, 2 a
2
) to the point z = (a, a). As we saw in Section 6.6, in the region > 0,
the velocity potential
F() =
Q
2
ln(
2
+ 4 a
4
), (6.61)
corresponds to the ow pattern generated by a line source of strength Q, located at the point = (0, 2 a
2
), in the
presence of a rigid plane at = 0. Thus, the complex velocity potential
G(z) = F(z
2
) =
Q
2
ln(z
4
+ 4a
4
), (6.62)
corresponds to the ow pattern generated by a line source of strength Q, located at the point z = (a, a), in the presence
of two orthogonal rigid planes at y = 0 and x = 0. The stream function associated with the above complex potential,
(x, y) =
Q
2
tan
1
_
4 x y (x
2
y
2
)
x
4
6 x
2
y
2
+ y
4
+ 4 a
4
_
, (6.63)
1
y
x
E
C B
C
A
B
D
E
D a
Figure 6.8: The conformal map = i e
z/a
.
1
0.8
0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
y/a
1 0.5 0 0.5 1
x/a
Figure 6.9: Stream lines of the 2D ow pattern due to a vortex lament at the origin in the presence of two rigid planes
at y = a/2.
112 FLUID MECHANICS
0
1
2
3
y/a
0 1 2 3
x/a
Figure 6.10: Stream lines of the 2D ow pattern due to a line source at (a, a) in the presence of two rigid planes at
x = 0 and y = 0.
is shown in Figure 6.10.
As a nal example, consider the map
z = +
l
2
, (6.64)
where l is real and positive. Writing = r e
i
, we nd that
x = 2 l cosh[ln(r/l)] cos , (6.65)
y = 2 l sinh[ln(r/l)] sin . (6.66)
Thus, the map converts the circle
2
+
2
= a
2
in the -plane, where a > l, into the ellipse
_
x
2 l cosh[ln(a/l)]
_
2
+
_
y
2 l sinh[ln(a/l)]
_
2
= 1 (6.67)
in the z-plane. Note that the center of the ellipse lies at the origin, and its major and minor axes run parallel to the x-
and the y-axes, respectively. As we saw in Section 6.4, in the -plane, the complex velocity potential
F = V
0
_
+
a
2
_
, (6.68)
corresponds to uniform ow of unperturbed speed V
0
, running parallel to the -axis, around a circular cylinder of
radius a, centered on the origin. Thus, assuming that a > l, in the z-plane the potential corresponds to uniform ow
of unperturbed speed V
0
, running parallel to the x-axis (which follows because at large z the map (6.64) reduces to
z = , and so the ow at large distances fromthe origin is the same in the complex z- and -planes), around an elliptical
cylinder of major radius 2 l cosh[ln(a/l)] = a + l
2
/a, aligned along the x-axis, and minor radius 2 l sinh[ln(a/l)] =
a l
2
/a, aligned along the y-axis. The corresponding stream function in the z-plane is
(x, y) = V
0
_
r
a
2
r
_
sin , (6.69)
4
3
2
1
0
1
2
3
4
y/l
4 3 2 1 0 1 2 3 4
x/l
Figure 6.11: Stream lines of the 2D ow pattern due to uniform ow parallel to the x-axis around an elliptical cylinder.
where
r = l exp(cosh
1
p), (6.70)
= tan
1
_
y
x
p
[p
2
1]
1/2
_
, (6.71)
p =
_
_
x
2
/l
2
+ y
2
/l
2
+ 4 +
_
[x
2
/l
2
+ y
2
/l
2
+ 4]
2
16 x
2
/l
2
_
1/2
8
_
_
1/2
. (6.72)
Figure 6.11 shows the streamlines of the ow pattern calculated for a = 1.5 l.
6.8 Complex Line Integrals
Consider the line integral of some function F(z) of the complex variable taken (counter-clockwise) around a closed
curve C in the complex plane:
J =
_
C
F(z) dz. (6.73)
Since dz = dx + i dy, and writing F(z) = (x, y) + i (x, y), where (x, y) and (x, y) are real functions, it follows that
J = J
r
+ i J
i
, where
J
r
=
_
C
( dx dy), (6.74)
J
i
=
_
C
(dx + dy). (6.75)
However, we can also write the above expressions in the two-dimensional vector form
J
r
=
_
C
A dr, (6.76)
114 FLUID MECHANICS
J
i
=
_
C
B dr, (6.77)
where dr = (dx, dy), A = (, ), and B = (, ). Now, according to Stokes theorem (see Section A.22),
_
C
A dr =
_
S
( A)
z
dS, (6.78)
_
C
B dr =
_
S
( B)
z
dS, (6.79)
where S is the region of the x-y plane enclosed by C. Hence, we obtain
J
r
=
_
S
_
x
+
y
_
dS, (6.80)
J
i
=
_
S
_
x

y
_
dS. (6.81)
Let
J =
_
C
F(z) dz, (6.82)
J
=
_
C
F(z) dz, (6.83)

where C
is a closed curve in the complex plane that completely surrounds the smaller curve C. Consider
J = J J
. (6.84)
Writing J = J
r
+ i J
i
, a direct generalization of the previous analysis reveals that
J
r
=
_
S
_
x
+
y
_
dS, (6.85)
J
i
=
_
S
_
x

y
_
dS, (6.86)
where S is now the region of the x-y plane lying between the curves C and C
. Suppose that F(z) is well-behaved (i.e.,

nite, single-valued, and dierentiable) throughout S . It immediately follows that its real and imaginary components,
and , respectively, satisfy the Cauchy-Riemann relations, (6.17)(6.18), throughout S . However, if this is the case
then it is apparent, from the previous two expressions, that J
r
= J
i
= 0. In other words, if F(z) is well-behaved
throughout S then J = J
.
The circulation of the ow about some closed curve C in the x-y plane is dened
=
_
C
(v
x
dx + v
y
dy) = Re
_
C
dF
dz
dz, (6.87)
where F(z) is the complex velocity potential of the ow, and use has been made of Equation (6.31). Thus, the
circulation can be evaluated by performing a line integral in the complex z-plane. Moreover, as is clear from the
previous discussion, this integral can be performed around any loop that can be continuously deformed into the loop
C whilst still remaining in the uid, and not passing over a singularity of the complex velocity, dF/dz.
6.9 Theorem of Blasius
Consider some ow pattern in the complex z-plane that is specied by the complex velocity potential F(z). Let C be
some closed curve in the complex z-plane. The uid pressure on this curve is determined from Equation (6.37), which
yields
P = p
0

1
2

dF
dz
2
. (6.88)
y
dy
dx
dl
P dx
P dl
P dy
x
Figure 6.12: Force acting across a short section of a curve.
Let us evaluate the resultant force (per unit length), and the resultant moment (per unit length), acting on the uid
within the curve as a consequence of this pressure distribution.
Consider a small element of the curve C, lying between x, y and x + dx, y + dy, which is suciently short that it
can be approximated as a straight line. Let P be the local uid pressure on the outer (i.e., exterior to the curve) side
of the element. As illustrated in Figure 6.12, the pressure force (per unit length) acting inward (i.e., toward the inside
of the curve) across the element has a component Pdy in the minus x-direction, and a component Pdx in the plus
y-direction. Thus, if X and Y are the components of the resultant force (per unit length) in the x- and y-directions,
respectively, then
dX = Pdy, (6.89)
dY = Pdx. (6.90)
The pressure force (per unit length) acting across the element also contributes to a moment (per unit length), M, acting
about the z-axis, where
dM = x dY y dX = P(x dx + y dy). (6.91)
Thus, the x- and y-components of the resultant force (per unit length) acting on the of the uid within the curve, as
well as the resultant moment (per unit length) about the z-axis, are given by
X =
_
C
Pdy, (6.92)
Y =
_
C
Pdx, (6.93)
M =
_
C
P(x dx + y dy), (6.94)
respectively, where the integrals are taken (counter-clockwise) around the curve C. Finally, given that the pressure
distribution on the curve takes the form (6.88), and that a constant pressure obviously yields zero force and zero
moment, we nd that
X =
1
2

_
C
dF
dz
2
dy, (6.95)
116 FLUID MECHANICS
Y =
1
2

_
C
dF
dz
2
dx, (6.96)
M =
1
2

_
C
dF
dz
2
(x dx + y dy). (6.97)
Now, z = x +i y, and z = x i y, where indicates a complex conjugate. Hence, d z = dx i dy, and i d z = dy +i dx.
It follows that
X i Y =
1
2
i
_
C
dF
dz
2
d z. (6.98)
However,
dF
dz
2
d z =
dF
dz
d

F
d z
d z =
dF
dz
d

F, (6.99)
where dF = d + i d and d

F = d i d. Suppose that the curve C corresponds to a streamline of the ow, in which
case = constant on C. Thus, d = 0 on C, and so d

F = dF. Hence, on C,
dF
dz
d

F =
dF
dz
dF =
_
dF
dz
_
2
dz, (6.100)
which implies that
X i Y =
1
2
i
_
C
_
dF
dz
_
2
dz. (6.101)
This result is known as the Blasius theorem.
Now, x dx + y dy = Re(z d z). Hence,
M = Re
_
1
2

_
C
dF
dz
2
z d z
_
, (6.102)
or, making use of an analogous argument to that employed above,
M = Re
_
1
2

_
C
_
dF
dz
_
2
z dz
_
_
, (6.103)
Note, nally, that Equations (6.101) and (6.103) hold even when is not constant on the curve C, as long as C can
be continuously deformed into a constant- curve without leaving the uid or crossing over a singularity of (dF/dz)
2
.
As an example of the use of the Blasius theorem, consider again the situation, discussed in Section 6.6, in which
a line source of strength Q is located at (0, a), and there is a rigid boundary at y = 0. As we have seen, the complex
velocity in the region y > 0 takes the form
dF
dz
=
Q
z
z
2
+ a
2
. (6.104)
Suppose that we evaluate the Blasius integral, (6.103), about the contour C shown in Figure 6.13. This contour runs
along the boundary, and is completed by a semi-circle in the upper half of the z-plane. As is easily demonstrated, in
the limit in which the radius of the semi-circle tends to innity, the contribution of the curved section of the contour to
the overall integral becomes negligible. In this case, only the straight section of the contour contributes to the integral.
Note that the straight section corresponds to a streamline (since it is coincident with a rigid boundary). In other words,
the contour C corresponds to a streamline at all constituent points that make a nite contribution to the Blasius integral,
which ensures that C is a valid contour for the application of the Blasius theorem. In fact, the Blasius integral species
the net force (per unit length) exerted on the whole uid by the boundary. Note, however, that the contour C can be
deformed into the contour C
, which takes the form of a small circle surrounding the source, without passing over a
singularity of (dF/dz)
2
. See Figure 6.13. Hence, we can evaluate the Blasius integral around C
without changing its

value. Thus,
X i Y =
1
2
i
_
C
_
dF
dz
_
2
dz =
1
2
i
_
Q
_
2
_
C
z
2
(z
2
+ a
2
)
2
dz, (6.105)
x
a
Q
C
C
y
Figure 6.13: Source in the presence of a rigid boundary.
or
X i Y =
1
8
i
_
Q
_
2
_
C
_
1
(z i a)
2
+
2
(z + i a) (z i a)
+
1
(z + i a)
_
dz. (6.106)
Writing z = i a + e
i
, dz = i e
i
d, and taking the limit 0, we nd that
X i Y =
i Q
2
4 a
. (6.107)
In other words, the boundary exerts a force (per unit length) F = ( Q
2
/4 a) e
y
on the uid. Hence, the uid
exerts an equal and opposite force F = ( Q
2
/4 a) e
y
on the boundary. Of course, this result is consistent with
Equation (6.43). Incidentally, it is easily demonstrated from (6.103) that there is zero moment (about the z-axis)
exerted on the boundary by the uid, and vice versa.
Consider a line source of strength Q placed (at the origin) in a uniformly owing uid whose velocity is V =
V
0
(cos
0
, sin
0
). From Section 6.4, the complex velocity potential of the net ow is
F(z) =
Q
2
ln z V
0
z e
i
0
. (6.108)
The net force (per unit length) acting on the source (which is calculated by performing the Blasius integral around
a large loop that follows streamlines, and then shrinking the loop to a small circle centered on the source) is (see
Exercise 6.1)
F = QV. (6.109)
Note that the force acts in the opposite direction to the ow. Thus, an external force F, acting in the same direction
as the ow, must be applied to the source in order for it to remain stationary. In fact, the above result is valid even in
a non-uniformly owing uid, as long as V is interpreted as the uid velocity at the location of the source (excluding
the velocity eld of the source itself).
Finally, consider a vortex lament of intensity placed at the origin in a uniformly owing uid whose velocity
is V = V
0
(cos
0
, sin
0
). From Section 6.4, the complex velocity potential of the net ow is
F(z) = i

2
ln z V
0
z e
i
0
. (6.110)
The net force (per unit length) acting on the lament (which is calculated by performing the Blasius integral around a
small circle centered on the lament) is (see Exercise 6.2)
F = V e
z
. (6.111)
118 FLUID MECHANICS
Note that the force is directed at right-angles to the direction of the ow (in the sense obtained by rotating V through
90
in the opposite direction to the laments direction of rotation). Again, the above result is valid even in a non-
uniformly owing uid, as long as V is interpreted as the uid velocity at the location of the lament (excluding the
velocity eld of the lament itself).
6.10 Exercises
6.1. Demonstrate that a line source of strength Q (running along the z-axis) situated in a uniform ow of (unperturbed) velocity
V (lying in the x-y plane) and density experiences a force per unit length
F = QV.
6.2. Demonstrate that a vortex lament of intensity (running along the z-axis) situated in a uniform ow of (unperturbed)
velocity V (lying in the x-y plane) and density experiences a force per unit length
F = V e
z
.
6.3. Show that two parallel line sources of strengths Q and Q
, located a perpendicular distance r apart, exert a radial force per

unit length Q Q
/(2 r) on one another, the force being attractive if QQ
> 0, and repulsive if Q Q
< 0.
6.4. Show that two parallel vortex laments of intensities and
, located a perpendicular distance r apart, exert a radial force

per unit length
/(2 r) on one another, the force being repulsive if
> 0, and attractive if
< 0.
6.5. A vortex lament of intensity runs parallel to, and lies a perpendicular distance a from, a rigid planar boundary. Demon-
strate that the boundary experiences a net force per unit length
2
/(4 a) directed toward the lament.
6.6. Two rigid planar boundaries meet at right-angles. A line source of strength Q runs parallel to the line of intersection of the
planes, and is situated a perpendicular distance a from each. Demonstrate that the source is subject to a force per unit length
3
2 Q
2
8 a
directed towards the line of intersection of the planes.
6.7. A line source of strength Q is located a distance b from an impenetrable cylinder of radius a < b (the axis of the cylinder
being parallel to the source). Demonstrate that the cylinder experiences a net force per unit length
Q
2
2 b
a
2
(b
2
a
2
)
directed toward the source.
6.8. A dipole line source consists of a line source of strength Q, running parallel to the z-axis, and intersecting the x-y plane at
(d/2) (cos , sin ), and a parallel source of strength Q that intersects the x-y plane at (d/2)(cos , sin ). Show that, in
the limit d 0, and Qd D, the complex velocity potential of the source is
F(z) =
De
i
2 z
.
Here, De
i
is termed the complex dipole strength.
6.9. A dipole line source of complex strength De
i
is placed in a uniformly owing uid of speed V
0
whose direction of motion
subtends a (counter-clockwise) angle
0
with the x-axis. Show that, while no net force acts on the source, it is subject to a
moment (per unit length) M = DV
0
sin(
0
) about the z-axis.
6.10. Consider a dipole line source of complex strength D
1
e
i
1
running along the z-axis, and a second parallel source of complex
strength D
2
e
i
2
that intersects the x-y plane at (x, 0). Demonstrate that the rst source is subject to a moment (per unit
length) about the z-axis of
M =
D
1
D
2
2 x
2
sin(
1
+
2
),
as well as a force (per unit length) whose x- and y-components are
X =
D
1
D
2
x
3
cos(
1
+
2
),
Y =
D
1
D
2
x
3
sin(
1
+
2
),
respectively. Show that the second source is subject to the same moment, but an equal and opposite force.
6.11. A dipole line source of complex strength De
i
runs parallel to, and is located a perpendicular distance a from, a rigid planar
boundary. Show that the boundary experiences a force per unit length
D
2
8 a
3
acting toward the source.
6.12. Demonstrate that a conformal map converts a line source into a line source of the same strength, and a vortex lament into
a vortex lament of the same intensity.
6.13. Consider the conformal map
z = i c cot(/2),
where z = x + i y, = + i , and c is real and positive. Show that
x =
c sinh
cosh cos
,
y =
c sin
cosh cos
.
Demonstrate that =
0
, where 0
0
, maps to a circular arc of center (0, c cot
0
), and radius c cosec
0
, that connects
the points (c, 0), and lies in the region y > 0. Demonstrate that =
0
+ maps to the continuation of this arc in the region
y < 0. In particular, show that = 0 maps to the region x > c on the x-axis, whereas = maps to the region x < c.
Finally, show that =
0
maps to a circle of center (c coth
0
, 0), and radius c cosech
0
.
6.14. Consider the complex velocity potential
F(z) =
2 c i V
0
n
cot(/n),
where
z = i c cot(/2).
Here, V
0
, n, and c are real and positive. Show that
dF
dz
=
4V
0
n
2
sin
2
(/2)
sin
2
(/n)
.
Hence, deduce that the ow at z is uniform, parallel to the x-axis, and of speed V
0
. Show that
(, ) =
2 V
0
c
n
sin(2 /n)
cosh(2 /n) cos(2 /n)
.
Hence, deduce that the streamline = 0 runs along the x-axis for x > c, but along a circular arc connecting the points (c,
0) for x < c. Furthermore, show that if 1 < n < 2 then this arc lies above the x-axis, and is of maximum height
h = c
_
cos( n/2) + 1
sin( n/2)
_
,
but if 2 < n < 3 then the arc lies below the x-axis, and is of maximum depth
d = c
_
cos( n/2) + 1
sin( n/2)
_
.
Hence, deduce that if 1 < n < 2 then the complex velocity potential under investigation corresponds to uniform ow of
speed V
0
, parallel to a planar boundary that possesses a cylindrical bump (whose axis is normal to the ow) of height h and
width 2 c, but if 2 < n < 3 then the potential corresponds to ow parallel to a planar boundary that possesses a cylindrical
depression of depth d and width 2 c. Show, in particular, that if n = 1 then the bump is a half-cylinder, and if n = 3 then the
depression is a half-cylinder. Finally, demonstrate that the ow speed at the top of the bump (in the case 1 < n < 2), or the
bottom of the depression (in the case 2 < n < 3) is
v =
2 V
0
n
2
[1 cos( n/2)] .
120 FLUID MECHANICS
6.15. Show that z = cosh( /a) maps the semi-innite strip 0 a, 0 in the -plane onto the upper half (y 0) of the
z-plane. Hence, show that the stream function due to a line source of strength Q placed at = (0, a/2), in the rectangular
region 0 a, 0 bounded by the rigid planes = 0, = 0, and = a, is
(, ) =
Q sinh( /a) sin( /a)
2 [sinh
2
( /a) + cos
2
( /a)]
.
6.16. Show that the complex velocity potential
F(z) =
a V
0
tanh(a /z)
can be interpreted as that due to uniform ow of speed V
0
over a cylindrical log of radius a lying on the at bed of a deep
stream (the axis of the log being normal to the ow). Demonstrate that the ow speed at the top of the log is (
2
/4) V
0
.
Finally, show that the pressure dierence between the top and bottom of the log is
4
V
2
0
/32.
Incompressible Boundary Layers 121
7 Incompressible Boundary Layers
7.1 Introduction
Previously, in Section 5.13, we saw that a uniformly owing incompressible uid that is modeled as inviscid is inca-
pable of exerting a drag force on a rigid stationary obstacle placed in its path. This result is surprising since, in practice,
a stationary obstacle experiences a signicant drag when situated in such a uid, even in the limit that the Reynolds
number tends to innity (which corresponds to the inviscid limit). In this chapter, we shall attempt to reconcile these
two results by introducing the concept of a boundary layer. This is a comparatively thin layer that covers the surface of
an obstacle placed in a high Reynolds number incompressible uidviscosity is assumed to have a signicant eect
on the ow inside the layer, but a negligible eect on the ow outside. For the sake of simplicity, we shall restrict our
discussion to the two-dimensional boundary layers that form when a high Reynolds number uid ows transversely
around a stationary obstacle of innite length and uniform cross-section.
7.2 No Slip Condition
We saw, in Section 5.13, that when an inviscid uid ows around a rigid stationary obstacle then the normal uid
velocity at the surface of the obstacle is required to be zero. However, in general, the tangential velocity is non-zero.
In fact, if the uid velocity eld is both incompressible and irrotational then it is derivable from a stream function that
satises Laplaces equation. (See Section 5.8.) It is a well-known property of Laplaces equation that we can either
specify the solution itself, or its normal derivative, on a bounding surface, but we cannot specify both these quantities
simultaneously. Now, the constraint of zero normal velocity is equivalent to the requirement that the stream function
take the constant value zero (say) on the surface of the obstacle. Hence, the normal derivative of the stream function,
which determines the tangential velocity, cannot also be specied at this surface, and is, in general, non-zero.
In reality, all physical uids possess nite viscosity. Moreover, when a viscous uid ows around a rigid stationary
obstacle both the normal and the tangential components of the uid velocity are found to be zero at the obstacles
surface. The additional constraint that the tangential uid velocity be zero at a rigid stationary boundary is known as
the no slip condition, and is ultimately justied via experimental observations.
The concept of a boundary layer was rst introduced into uid mechanics by Ludwig Prandtl (18751953) in
order to account for the modication to the ow pattern of a high Reynolds number irrotational uid necessitated by
the imposition of the no slip condition on the surface of an impenetrable stationary obstacle. According to Prandtl, the
boundary layer covers the surface of the obstacle, but is relatively thin in the direction normal to this surface. Outside
the layer, the ow pattern is the same as that of an idealized inviscid uid, and is thus generally irrotational. This
implies that the normal uid velocity is zero on the outer edge of the layer, where it interfaces with the irrotational
ow, but, in general, the tangential velocity is non-zero. However, the no slip condition requires the tangential velocity
to be zero on the inner edge of the layer, where it interfaces with the rigid surface. It follows that there is a very
large normal gradient of the tangential velocity across the layer, which implies the presence of intense internal vortex
laments trapped within the layer. Consequently, the ow within the layer is not irrotational. In the following, we
shall attempt to make the concept of a boundary layer more precise.
7.3 Boundary Layer Equations
Consider a rigid stationary obstacle whose surface is (locally) at, and corresponds to the x-z plane. Let this surface
be in contact with a high Reynolds number uid that occupies the region y > 0. See Figure 7.1. Let be the typical
normal thickness of the boundary layer. The layer thus extends over the region 0 < y
<
. Now, the uid that occupies

the region
<
y < , and thus lies outside the layer, is assumed to be both irrotational and (eectively) inviscid. On
the other hand, viscosity must be included in the equation of motion of the uid within the layer. The uid both inside
and outside the layer is assumed to be incompressible.
122 FLUID MECHANICS
boundary layer
irrotational uid
solid surface
x
y
U(x)
Figure 7.1: A boundary layer.
Suppose that the equations of irrotational ow have already been solved to determine the uid velocity outside the
boundary layer. This velocity must be such that its normal component is zero at the outer edge of the layer (i.e., y ).
On the other hand, the tangential component of the uid velocity at the outer edge of the layer, U(x) (say), is generally
non-zero. Here, we are assuming, for the sake of simplicity, that there is no spatial variation in the z-direction, so that
both the irrotational ow and the boundary layer are eectively two-dimensional. Likewise, we are also assuming that
all ows are steady, so that any time variation can be neglected. Now, the motion of the uid within the boundary layer
is governed by the equations of steady-state, incompressible, two-dimensional, viscous ow, which take the form (see
Section 2.14)
v
x
x
+
v
y
y
= 0, (7.1)
v
x
v
x
x
+ v
y
v
x
y
=
1
p
x
+
_
2
v
x
x
2
+
2
v
x
y
2
_
, (7.2)
v
x
v
y
x
+ v
y
v
y
y
=
1
p
y
+
_
2
v
y
x
2
+
2
v
y
y
2
_
, (7.3)
where is the (constant) density, and the kinematic viscosity. Here, Equation (7.1) is the equation of continuity,
whereas Equations (7.2) and (7.3) are the x- and y-components of the uid equation of motion, respectively. The
boundary conditions at the outer edge of the layer, where it interfaces with the irrotational uid, are
v
x
(x, y) U(x), (7.4)
p(x, y) P(x) (7.5)
as y/ . Here, P(x) is the uid pressure at the outer edge of the layer, and
U
dU
dx
=
1
dP
dx
(7.6)
(since v
y
= 0, and viscosity is negligible, just outside the layer). The boundary conditions at the inner edge of the
layer, where it interfaces with the impenetrable surface, are
v
x
(x, 0) = 0, (7.7)
v
y
(x, 0) = 0. (7.8)
Of course, the rst of these constraints corresponds to the no slip condition.
Let U
0
be a typical value of the external tangential velocity, U(x), and let L be the typical variation length-scale of
this quantity. It is reasonable to suppose that U
0
and L are also the characteristic tangential ow velocity and variation
length-scale in the x-direction, respectively, of the boundary layer. Of course, is the typical variation length-scale
of the layer in the y-direction. Moreover, /L 1, since the layer is assumed to be thin. It is helpful to dene the
normalized variables
X =
x
L
, (7.9)
Y =
y
, (7.10)
V
x
(X, Y) =
v
x
U
0
, (7.11)
V
y
(X, Y) =
v
y
U
1
, (7.12)
P(X, Y) =
p
p
0
, (7.13)
where U
1
and p
0
are constants. All of these variables are designed to be O(1) inside the layer. Equation (7.1) yields
U
0
L
V
x
X
+
U
1
V
y
Y
= 0. (7.14)
In order for the terms in this equation to balance one another, we need
U
1
=
L
U
0
. (7.15)
In other words, within the layer, continuity requires the typical ow velocity in the y-direction, U
1
, to be much smaller
than that in the x-direction, U
0
.
Equation (7.2) gives
U
2
0
L
_
V
x
V
x
X
+ V
y
V
x
Y
_
=
p
0
L
P
X
+
_
U
0
2
_
_
_
L
_
2
2
V
x
X
2
+
2
V
x
Y
2
_
. (7.16)
In order for the pressure term on the right-hand side of the above equation to be of similar magnitude to the advective
terms on the left-hand side, we require that
p
0
= U
2
0
. (7.17)
Furthermore, in order for the viscous term on the right-hand side to balance the other terms, we need
L
=
U
1
U
0
=
1
Re
1/2
, (7.18)
where
Re =
U
0
L
(7.19)
is the Reynolds number of the ow external to the layer. The assumption that /L 1 can be seen to imply that
Re 1. In other words, the normal thickness of the boundary layer separating an irrotational ow pattern from a rigid
surface is only much less than the typical variation length-scale of the pattern when the Reynolds number of the ow
is much greater than unity.
Equation (7.3) yields
1
Re
_
V
x
V
y
X
+ V
y
V
y
Y
_
=
P
Y
+
1
Re
_
1
Re
2
V
y
X
2
+
2
V
y
Y
2
_
. (7.20)
In the limit Re 1, this reduces to
P
Y
= 0. (7.21)
124 FLUID MECHANICS
Hence,

P =

P(X), where
d
P
dX
=
U
d
U
dX
, (7.22)
U(X) = U/U
0
, and use has been made of (7.6). In other words, the pressure is uniform across the layer, in the direction
normal to the surface of the obstacle, and is thus the same as that on the outer edge of the layer.
Retaining only O(1) terms, our nal set of normalized layer equations becomes
V
x
X
+
V
y
Y
= 0, (7.23)
V
x
V
x
X
+ V
y
V
y
Y
=

U
d
U
X
+
2
V
y
Y
2
, (7.24)
subject to the boundary conditions
V
x
(X, ) =

U(X), (7.25)
and
V
x
(X, 0) = 0, (7.26)
V
y
(X, 0) = 0. (7.27)
In unnormalized form, the above set of layer equations are written
v
x
x
+
v
y
y
= 0, (7.28)
v
x
v
x
x
+ v
y
v
x
y
= U
dU
dx
+

2
v
x
y
2
, (7.29)
v
x
(x, ) = U(x) (7.30)
(note that y = really means y/ ), and
v
x
(x, 0) = 0, (7.31)
v
y
(x, 0) = 0. (7.32)
Now, Equation (7.28) can be automatically satised by expressing the ow velocity in terms of a stream function: i.e.,
v
x
=
y
, (7.33)
v
y
=
x
. (7.34)
In this case, Equation (7.29) reduces to
y
3

x
y
2
+
x y
= U
dU
dx
, (7.35)
(x, )
y
= U(x), (7.36)
and
(x, 0) = 0, (7.37)
(x, 0)
y
= 0. (7.38)
To lowest order, the vorticity internal to the layer, = e
z
, is given by
=
y
2
, (7.39)
whereas the x-component of the viscous force per unit area acting on the surface of the obstacle is written (see Sec-
tion 2.18)
xy
y=0
=
v
x
y
y=0
=
y
2
y=0
. (7.40)
7.4 Self-Similar Boundary Layers
The boundary layer equation, (7.35), takes the form of a nonlinear partial dierential equation that is extremely
dicult to solve exactly. However, considerable progress can be made if this equation is converted into an ordi-
nary dierential equation by demanding that its solutions be self-similar. Self-similar solutions are such that, at
a given distance, x, along the layer, the tangential ow prole, v
x
(x, y), is a scaled version of some common pro-
le: i.e., v
x
(x, y) = U(x) F[y/(x)], where (x) is a scale-factor, and F(z) a dimensionless function. It follows that
(x, y) = U(x) (x) f [y/(x)], where f
(z) = F(z).
Let us search for a self-similar solution to Equation (7.35) of the general form
(x, y) =
_
2 U
0
x
m+1
m + 1
_
1/2
f () = U
0
x
m
_
2
(m + 1) U
0
x
m1
_
1/2
f (), (7.41)
where
=
_
(m + 1) U
0
x
m1
2
_
1/2
y. (7.42)
This implies that (x) = [2 /(m + 1) U
0
x
m1
]
1/2
, and U(x) = U
0
x
m
. Here, U
0
and m are constants. Moreover, U
0
x
m
has dimensions of velocity, whereas m, , and f , are dimensionless. Transforming variables from x, y to x, , we nd
that
y
=
+
m 1
2
x
, (7.43)
x
=
_
(m + 1) U
0
x
m1
2
_
1/2
x
. (7.44)
Hence,
x
=
_
U
0
x
m1
2 (m + 1)
_
1/2
[(m+ 1) f + (m 1) f
], (7.45)
y
= U
0
x
m
f
, (7.46)
y
2
=
_
_
(m + 1) U
3
0
x
3m1
2
_
_
1/2
f
, (7.47)
x y
=
U
0
x
m1
2
[2 m f
+ (m 1) f
], (7.48)
y
3
=
(m + 1) U
2
0
x
2m1
2
f
, (7.49)
where

= d/d. Thus, Equation (7.35) becomes
(m + 1) f
+ (m + 1) f f
2m f
2
=
1
U
2
0
x
2m1
dU
2
dx
. (7.50)
126 FLUID MECHANICS
Since the left-hand side of the above equation is a (non-constant) function of , whilst the right-hand side is a function
of x (and since and x are independent variables), the equation can only be satised if its right-hand side takes a
constant value. In fact, if
1
U
2
0
x
2m1
dU
2
dx
= 2m (7.51)
then
U(x) = U
0
x
m
(7.52)
(which is consistent with our initial guess), and
f
+ f f
+ (1 f
2
) = 0, (7.53)
where
=
2 m
m + 1
. (7.54)
Expression (7.53) is known as the Falkner-Skan equation. The solutions to this equation that satisfy the physical
boundary conditions (7.36)(7.38) are such that
f (0) = f
(0) = 0, (7.55)
and
f
() = 1, (7.56)
f
() = 0. (7.57)
(The nal condition corresponds to the requirement that the vorticity tend to zero at the edge of the layer.) Note, from
(7.39), (7.42), (7.47), (7.52), (7.55), and (7.56), that the normally integrated vorticity within the boundary layer is
_

0
dy = U(x). (7.58)
Furthermore, from (7.40), (7.47), and (7.52), the x-component of the viscous force per unit area acting on the surface
of the obstacle is
xy
y=0
=
1
2
U
2
_

U x
_
1/2
(m + 1)
1/2

2 f
(0). (7.59)
It is convenient to parameterize this quantity in terms of a skin friction coecient,
c
f
=
xy
y=0
(1/2) U
2
. (7.60)
It follows that
c
f
(x) =
(m + 1)
1/2

2 f
(0)
[Re(x)]
1/2
, (7.61)
where
Re(x) =
U(x) x
(7.62)
is the eective Reynolds number of the ow on the outer edge of the layer at position x. Hence, c
f
(x) x
(m+1)/2
.
Finally, according to Equation (7.41), the width of the boundary layer is approximately
(x)
x

1
[Re(x)]
1/2
, (7.63)
which implies that (x) x
(m1)/2
.
0
0.5
1
1.5
f
(
0
)
0 1 2
Figure 7.2: f
(0) calculated as a function of for solutions of the Falkner-Skan equation.

Note that if m > 0 then the external tangential velocity prole, U(x) = U
0
x
m
, corresponds to that of irrotational
inviscid ow incident, in a symmetric fashion, on a semi-innite wedge whose apex subtends an angle , where
= 2m/(m + 1). (See Section 5.14, and Figure 5.12.) In this case, U(x) can be interpreted as the tangential velocity
a distance x along the surface of the wedge from its apex (in the direction of the ow). By analogy, if m = 0 then
the external velocity prole corresponds to that of irrotational inviscid ow parallel to a semi-innite at plate (which
can be thought of as a wedge whose apex subtends zero angle). In this case, U(x) can be interpreted as the tangential
velocity a distance x along the surface of the plate fromits leading edge (in the direction of the ow). (See Section 7.5.)
Finally, if m < 0 then the external velocity prole is that of symmetric irrotational inviscid ow over the back surface
of a semi-innite wedge whose apex subtends an angle (1
) , where
= m/(1 + m). (See Section 5.15, and

Figure 5.13.) In this case, U(x) can be interpreted as the tangential velocity a distance x along the surface of the wedge
from its apex (in the direction of the ow).
Unfortunately, the Falkner-Skan equation, (7.53), possesses no general analytic solutions. However, this equation
is relatively straightforward to solve via numerical methods. Figure 7.2 shows f
(0), calculated numerically as a

function of = 2 m/(m+1), for the solutions of (7.53) that satisfy the boundary conditions (7.55)(7.57). In addition,
Figure 7.3 shows f
() versus , calculated numerically for various dierent values of m. Note that, since 2 as
m , solutions of the Falkner-Skan equation with > 2 have no physical signicance. For 0 < < 2 it can be
seen, from Figures 7.2 and 7.3, that there is a single solution branch characterized by f
() > 0 and f
(0) > 0. This

branch is termed the forward ow branch, since it is such that the tangential velocity, v
x
() f
(), is in the same

direction as the external tangential velocity [i.e., v
x
()] across the whole layer (i.e., 0 < < ). The forward ow
branch is characterized by a positive skin friction coecient, c
f
f
(0). It can also be seen that for < 0 there

exists a second solution branch, which is termed the reversed ow branch, since it is such that the tangential velocity
is in the opposite direction to the external tangential velocity in the region of the layer immediately adjacent to the
surface of the obstacle (which corresponds to = 0). The reversed ow branch is characterized by a negative skin
friction coecient. Note that the reversed ow solutions are probably unphysical, since reversed ow close to the wall
is generally associated with a phenomenon known as boundary layer separation (see Section 7.10) which invalidates
the boundary layer orderings. It can be seen that the two solution branches merge together at =
= 0.1989, which
corresponds to m = m
= 0.0905. Moreover, there are no solutions to the Falkner-Skan equation with <
or
m < m
. The disappearance of solutions when m becomes too negative (i.e., when the deceleration of the external ow
becomes too large) is also related to boundary layer separation.
128 FLUID MECHANICS
0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
f
0 1 2 3 4 5 6 7 8 9 10
Figure 7.3: Solutions of the Falkner-Skan equation. In order from the left to the right, the various solid curves
correspond to forward ow solutions calculated with m = 4, 1, 1/3, 1/9, 0, 0.05, and 0.0904, respectively. The
dashed curve shows a reversed ow solution calculated with m = 0.05.
7.5 Boundary Layer on a Flat Plate
Consider a at plate of length L, innite width, and negligible thickness, which lies in the x-z plane, and whose two
edges correspond to x = 0 and x = L. Suppose that the plate is immersed in a low viscosity uid whose unperturbed
velocity eld is v = U
0
e
x
. See Figure 7.4. In the inviscid limit, the appropriate boundary condition at the surface of
the plate, v
y
= 0corresponding to the requirement of zero normal velocityis already satised by the unperturbed
ow. Hence, the original ow is not modied by the presence of the plate. However, when we take the nite viscosity
of the uid into account, an additional boundary condition, v
x
= 0corresponding to the no slip conditionmust be
satised at the plate. The imposition of this additional constraint causes thin boundary layers, of thickness (x) L, to
form above and below the plate. The uid ow outside the boundary layers remains eectively inviscid, whereas that
inside the layers is modied by viscosity. It follows that the ow external to the layers is unaected by the presence
of the plate. Hence, the tangential velocity at the outer edge of the boundary layers is U(x) = U
0
. This corresponds to
the case m = 0 discussed in the previous sectionsee Equation (7.52). (Here, we are assuming that the ow upstream
of the trailing edge of the plate, x = L, is unaected by the edges presence, and, is, therefore, the same as if the plate
were of innite length. Of course, the ow downstream of the edge is modied as a consequence of the nite length
of the plate.)
Making use of the analysis contained in the previous section (with m = 0), as well as the fact that, by symmetry,
the lower boundary layer is the mirror image of the upper one, the tangential velocity prole across the both layers is
written
v
x
(x, y) = U
0
f
(), (7.64)
where
=
_
U
0
2 x
_
1/2
y. (7.65)
Here, f () is the solution of
f
+ f f
= 0 (7.66)
y
x
plate
boundary layer
wake
L
U
0
Figure 7.4: Flow over a at plate.
that satises the boundary conditions
f (0) = f
(0) = 0, (7.67)
and
f
() = 1, (7.68)
f
() = 0. (7.69)
Equation (7.66) is known as the Blasius equation.
It is convenient to dene the so-called displacement thickness of the upper boundary layer,
(x) =
_

0
_
1
v
x
(x, y)
U
0
_
dy, (7.70)
which can be interpreted as the distance through which streamlines just outside the layer are displaced laterally due to
the retardation of the ow within the layer. (Of course, the thickness of the lower boundary layer is the same as that of
the upper layer.) It follows that
(x) =
_
x
U
0
_
1/2
2
_

0
[1 f
()] d. (7.71)
In fact, the numerical solution of (7.66), subject to the boundary conditions (7.67)(7.69), yields
(x) = 1.72
_
x
U
0
_
1/2
. (7.72)
Hence, the thickness of the boundary layer increases as the square root of the distance from the leading edge of the
plate. In particular, the thickness at the trailing edge of the plate is
(L)
L
=
1.72
Re
1/2
, (7.73)
where
Re =
U
0
L
(7.74)
is the appropriate Reynolds number for the interaction of the ow with the plate. Note that if Re 1 then the thickness
of the boundary layer is much less than its length, as was previously assumed.
130 FLUID MECHANICS
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
v
x
/
U
0
5 4 3 2 1 0 1 2 3 4 5
y /
Figure 7.5: Tangential velocity prole across the boundary layers located above and below a at plate of negligible
thickness located at y = 0.
The tangential velocity prole across the both boundary layers, which takes the form
v
x
(x, y) = U
0
f
_
1.22
y
(x)
_
, (7.75)
is plotted in Figure 7.5. In addition, the vorticity prole across the layers, which is written
(x, y) = sgn(y) 1.22
U
0
(x)
f
_
1.22
y
(x)
_
, (7.76)
is shown in Figure 7.6. Note that the vorticity is negative in the upper boundary layer (i.e., y > 0), positive in the lower
boundary layer (i.e., y < 0), and discontinuous across the plate (which is located at y = 0). Finally, the net viscous
drag force per unit width (along the z-axis) acting on the plate in the x-direction is
D = 2
_
L
0
xy
y=0
dx, (7.77)
where the factor of 2 is needed to take into account the presence of boundary layers both above and below the plate. It
follows from Equation (7.59) (with m = 0) that
D = U
2
0
_

U
0
_
1/2
2 f
(0)
_
L
0
x
1/2
dx = U
2
0
_
L
U
0
_
1/2
2
2 f
(0). (7.78)
In fact, the numerical solution of (7.66) yields
D = 1.33
U
2
0
L
Re
1/2
= 1.33 U
0
( U
0
L)
1/2
. (7.79)
The above discussion is premised on the assumption that the ow in the upper (or lower) boundary layer is both
steady and z-independent. It turns out that this assumption becomes invalid when the Reynolds number of the layer,
0.6
0.5
0.4
0.3
0.2
0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
/
(
U
0
/
)
5 4 3 2 1 0 1 2 3 4 5
y /
Figure 7.6: Vorticity prole across the boundary layers located above and below a at plate of negligible thickness
located at y = 0.
U
0
/, exceeds a critical value which is about 600. In this case, small-scale z-dependent disturbances spontaneously
grow to large amplitude, and the layer becomes turbulent. Since x
1/2
, if the criterion for boundary layer turbulence
is not satised at the trailing edge of the plate, x = L, then it is not satised anywhere else in the layer. Thus, the
previous analysis, which neglects turbulence, remains valid provided U
0
(L)/ < 600. According to (7.73), this
implies that the analysis is valid when 1 Re < 1.2 10
5
, where Re = U
0
L/ is the Reynolds number of the external
ow.
Consider, nally, the situation illustrated in Figure 7.7 in which an initially irrotational uid passes between two
at parallel plates. Let d be the perpendicular distance between the plates. As we have seen, the nite viscosity of the
uid causes boundary layers to form on the inner surfaces of the upper and lower plates. The ow within these layers
possesses non-zero vorticity, and is signicantly aected by viscosity. On the other hand, the ow outside the layers
is irrotational and essentially inviscidthis type of ow is usually termed potential ow (since it can be derived from
a velocity potential satisfying Poissons equation). Now, the thickness of the two boundary layers increases like x
1/2
,
where x represents distance, parallel to the ow, measured from the leading edges of the plates. It follows that, as x
increases, the region of potential ow shrinks in size, and eventually disappears. See Figure 7.7. Assuming that, prior
to merging, the two boundary layers do not signicantly aect one another, their thickness, (x), is given by formula
(7.72), where U
0
is the speed of the incident uid. The region of potential ow thus extends from x = 0 (which
corresponds to the leading edge of the plates) to x = l, where
(l) =
d
2
. (7.80)
It follows that
l
d
= 11.8 Re, (7.81)
where
Re =
U
0
d
. (7.82)
Thus, when an irrotational high Reynolds number uid passes between two parallel plates then the region of potential
ow extends a comparatively long distance between the plates, relative to their spacing (i.e., l/d 1). By analogy, if
132 FLUID MECHANICS
U
0
boundary layer
potential ow
plate
l
d
Figure 7.7: Flow between two at parallel plates.
an irrotational high Reynolds number uid passes into a pipe then the uid remains essentially irrotational until it has
travelled a considerable distance along the pipe, compared to its diameter. Obviously, these conclusions are modied
if the ow becomes turbulent.
7.6 Wake Downstream of a Flat Plate
As we saw in the previous section, if a at plate of negligible thickness, and nite length, is placed in the path of a
uniform high Reynolds number ow, directed parallel to the plate, then thin boundary layers form above and below
the plate. Outside the layers, the ow is irrotational, and essentially inviscid. Inside the layers, the ow is modied
by viscosity, and has non-zero vorticity. Downstream of the plate, the boundary layers are convected by the ow, and
merge to form a thin wake. See Figure 7.4. Within the wake, the ow is modied by viscosity, and possesses nite
vorticity. Outside the wake, the downstream ow remains irrotational, and eectively inviscid.
Since there is no solid surface embedded in the wake, acting to retard the ow, we would expect the action of
viscosity to cause the velocity within the wake, a long distance downstream of the plate, to closely match that of the
unperturbed ow. In other words, we expect the uid velocity within the wake to take the form
v
x
(x, y) = U
0
u(x, y), (7.83)
v
y
(x, y) = v(x, y), (7.84)
where
u U
0
. (7.85)
Assuming that, within the wake,
x

1
x
, (7.86)
, (7.87)
where x is the wake thickness, uid continuity requires that
v

x
u. (7.88)
Now, the owexternal to the boundary layers, and the wake, is both uniformand essentially inviscid. Hence, according
to Bernoullis theorem, the pressure in this region is also uniformsee Equation (7.22). However, as we saw in
Section 7.3, there is no y-variation of the pressure across the boundary layers. It follows that the pressure is uniform
within the layers. Thus, it is reasonable to assume that the pressure is also uniform within the wake, since the wake is
formed via the convection of the boundary layers downstream of the plate. We conclude that
p(x, y) p
0
(7.89)
everywhere in the uid, where p
0
is a constant.
The x-component of the uid equation of motion is written
v
x
v
x
x
+ v
y
v
x
y
=
1
p
x
+
_
2
v
x
x
2
+
2
v
x
y
2
_
. (7.90)
Making use of (7.83)(7.89), the above expression reduces to
U
0
u
x

2
u
y
2
. (7.91)
The boundary condition
u(x, ) = 0 (7.92)
ensures that the ow outside the wake remains unperturbed. Note that Equation (7.91) has the same mathematical
form as a conventional diusion equation, with x playing the role of time, and /U
0
playing the role of the diusion
coecient. Hence, by analogy with the standard solution of the diusion equation, we would expect ( x/U
0
)
1/2
.
As can easily be demonstrated, the self-similar solution to (7.91), subject to the boundary condition (7.92), is
u(x, y) =
Q

exp
_
y
2
2
_
, (7.93)
where
(x) = 2
_
x
U
0
_
1/2
, (7.94)
and Q is a constant. It follows that
_

u dy = Q, (7.95)
since, as is well-known,
_

exp(t
2
) dt =
. As expected, the width of the wake scales as x

1/2
.
The tangential velocity prole across the wake, which takes the form
v
x
(x, y)
U
0
= 1
Q
U
0

1
exp(y
2
/
2
), (7.96)
is plotted in Figure 7.8. In addition, the vorticity prole across the wake, which is written
(x, y)
U
0
/
=
Q
U
0

2
exp(y
2
/
2
) (7.97)
is shown in Figure 7.9. It can be seen that the proles pictured in Figures 7.8 and 7.9 are essentially smoothed out
versions of the boundary layer proles shown in Figures 7.5 and 7.6, respectively.
Suppose that the plate and a portion of its trailing wake are enclosed by a cuboid control volume of unit depth (in
the z-direction) that extends from x = l to x = +l and from y = h to y = h. See Figure 7.10. Here, l L and
h (l), where L is the length of the plate, and (x) the width of the wake. Hence, the control volume extends well
upstream and downstream of the plate. Moreover, the volume is much wider than the wake.
Let us apply the integral form of the uid equation of continuity to the control volume. For a steady-state, this
reduces to (see Section 2.9)
_
S
v dS = 0, (7.98)
134 FLUID MECHANICS
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
v
x
/
U
0
3 2 1 0 1 2 3
y /
Figure 7.8: Tangential velocity prole across the wake of a at plate of negligible thickness located at y = 0. The
prole is calculated for Q/(U
0
) = 0.5.
0.5
0.4
0.3
0.2
0.1
0
0.1
0.2
0.3
0.4
0.5
/
(
U
0
/
)
3 2 1 0 1 2 3
y /
Figure 7.9: Vorticity prole across the boundary layers above and below a at plate of negligible thickness located at
y = 0. The prole is calculated for Q/(U
0
) = 0.5.
U
0
u(y)
wake plate
y = h
y = h
U
0
x = l x = l
v(x)
v(x)
y
x
Figure 7.10: Control volume surrounding a at plate and its trailing wake.
where S is the bounding surface of the control volume. The normal uid velocity is U
0
at x = l, U
0
u(y) at x = l,
and v(x) at y = h, as indicated in the gure. Hence, (7.98) yields
_
h
h
U
0
dy +
_
h
h
[U
0
u(y)] dy + 2
_
l
l
v(x) dx = 0, (7.99)
or
_
h
h
u(y) dy = 2
_
l
l
v(x) dx. (7.100)
However, given that u 0 for y , and since h , it is a good approximation to replace the limits of integration
on the left-hand side of the above expression by . Thus, from Equation (7.95),
_
h
h
u(y) dy = 2
_
l
l
v(x) dx Q, (7.101)
where Q is independent of x. Note that the slight retardation of the ow inside the wake, due to the presence of the
plate, which is parameterized by Q, necessitates a small lateral outow, v(x), in the region of the uid external to the
wake.
Let us now apply the integral form of the x-component of the uid equation of motion to the control volume. For
a steady-state, this reduces to (see Section 2.11)
_
S
v
x
v dS = F
x
+
_
S
x j
dS
j
, (7.102)
where F
x
is the net x-directed force exerted on the uid within the control volume by the plate. It follows, from
Newtons third law of motion, that F
x
= D, where D is the viscous drag force per unit width (in the z-direction)
acting on the plate in the x-direction. Now, in an incompressible uid (see Section 2.6),
i j
= p
i j
+
_
v
i
x
j
+
v
j
x
i
_
. (7.103)
136 FLUID MECHANICS
Hence, we obtain
_
h
h
U
2
0
dy +
_
h
h
_
U
0
u(y)
_
2
dy
+2
_
l
l
U
0
v(x) dx = D 2
d
dl
_
h
h
u(y) dy, (7.104)
since the pressure within the uid is essentially uniform, the tangential uid velocity at y = h is U
0
, and v is assumed
to be negligible at x = l. Making use of Equation (7.101), as well as the fact that Q is independent of l, we get
D = U
0
Q. (7.105)
Here, we have neglected any terms that are second-order in the small quantity u. A comparison with Equation (7.79)
reveals that
Q = 1.33 ( U
0
L)
1/2
, (7.106)
or
Q
U
0

= 0.664
_
L
x
_
1/2
. (7.107)
Hence, from (7.96) and (7.97), the velocity and vorticity proles across the layer are
v
x
(x, y)
U
0
= 1 0.375
_
L
x
_
1/2
exp(y
2
/
2
), (7.108)
and
(x, y)
U
0
/
= 0.749
_
L
x
_
1/2
y
exp(y
2
/
2
), (7.109)
where (x) = 2 ( x/U
0
)
1/2
. Finally, since the above analysis is premised on the assumption that 1v
x
/U
0
= u/U
0

1, it is clear that the previous three expressions are only valid when x L (i.e., well downstream of the plate).
The above analysis only holds when the ow within the wake is non-turbulent. Let us assume, by analogy with
the discussion in the previous section, that this is the case as long as the Reynolds number of the wake, U
0
(x)/,
remains less than some critical value that is approximately 600. Since the Reynolds number of the wake can be written
2 Re
1/2
(x/L)
1/2
, where Re = U
0
L/ is the Reynolds number of the external ow, we deduce that the wake becomes
turbulent when x/L
>
9 10
4
/Re. Hence, the wake is always turbulent suciently far downstream of the plate. Our
analysis, which eectively assumes that the wake is non-turbulent in some region, immediately downstream of the
plate, whose extent (in x) is large compared with L, is thus only valid when 1 Re 9 10
4
.
7.7 Von K arm an Momentum Integral
Consider a boundary layer that forms on the surface of a rigid stationary obstacle of arbitrary shape (but innite length
and uniform cross-section) placed in a steady, uniform, transverse, high Reynolds number ow. Let x represent arc
length along the surface, measured (in the direction of the external ow) from the stagnation point that forms at the
front of the obstacle. (See Figure 7.11.) Moreover, let y represent distance across the boundary layer, measured
normal to the surface. Suppose that the boundary layer is suciently thin that it is well approximated as a plane slab
in the immediate vicinity of a general point on the surface. In this case, writing the velocity eld within the layer in
the form v = u(x, y) e
x
+ v(x, y) e
y
, it is reasonable to model this ow using the slab boundary layer equations [see
Equations (7.28) and (7.29)]
u
x
+
v
y
= 0, (7.110)
u
u
x
+ v
u
y
U
dU
dx
=

2
u
y
2
, (7.111)
subject to the standard boundary conditions
u(x, ) = U(x), (7.112)
u(x, 0) = v(x, 0) = 0. (7.113)
Here, U(x) is the external tangential uid velocity at the edge of the layer. Integrating (7.111) across the layer, making
use of the boundary conditions (7.113), leads to
u
y
y=0
=
_

0
_
U
dU
dx
u
u
x
v
u
y
_
dy
=
_

0
_
(U u)
dU
dx
+ u
(U u)
x
+ v
(U u)
y
_
dy
=
_

0
_
(U u)
dU
dx
+ u
(U u)
x
(U u)
v
y
_
dy
=
_

0
_
(U u)
dU
dx
+ u
(U u)
x
+ (U u)
u
x
_
dy
=
dU
dx
_

0
(U u) dy +
d
dx
_

0
u (U u) dy. (7.114)
Here, we have integrated the nal term on the right-hand side by parts, making use of Equations (7.110), (7.112), and
(7.113). Let us dene the displacement thickness of the layer [see Equation (7.70)]
1
(x) =
_

0
_
1
u
U
_
dy, (7.115)
as well as the so-called momentum thickness
2
(x) =
_

0
u
U
_
1
u
U
_
dy. (7.116)
It follows from (7.114) that
u
y
y=0
= U
2
d
2
dx
+ U
dU
dx
(
1
+ 2
2
). (7.117)
This important result is known as the von K arm an momentum integral, and is fundamental to many of the approxima-
tion methods commonly employed to calculate boundary layer thicknesses on the surfaces of general obstacles placed
in high Reynolds number ows. (See Section 7.10.)
7.8 Boundary Layer Separation
As we saw in Section 7.5, when a high Reynolds number uid passes around a streamlined obstacle, such as a slender
plate that is aligned with the ow, a relatively thin boundary layer form on the obstacles surface. Here, by relatively
thin, we mean that the typical transverse (to the ow) thickness of the layer is L/Re
1/2
, where L is the length
of the obstacle (in the direction of the ow), and Re the Reynolds number of the external ow. Suppose, however,
that the obstacle is not streamlined: i.e., the surface of the obstacle is not closely aligned with the streamlines of the
unperturbed ow pattern. In this case, the typically observed behavior is illustrated in Figure 7.11, which shows the
ow pattern of a high Reynolds number irrotational uid around a cylindrical obstacle (whose axis is normal to the
direction of the unperturbed ow). It can be seen that a stagnation point, at which the ow velocity is locally zero,
forms in front of the obstacle. Moreover, a thin boundary layer covers the front side of the obstacle. The thickness
of this layer is smallest at the stagnation point, and increases towards the back side of the obstacle. However, at
some point on the back side, the boundary layer separates from the obstacles surface to form a vortex-lled wake
whose transverse dimensions are similar to those of the obstacle itself. This phenomenon is known as boundary layer
separation.
138 FLUID MECHANICS
stagnation point
potential ow streamlines
boundary layer
wake
separation point
obstacle
Figure 7.11: Boundary layer separation.
Outside the boundary layer, and the wake, the ow pattern is irrotational and essentially inviscid. So, from Sec-
tion 5.13, the tangential ow speed just outside the boundary layer (neglecting any circulation of the external ow
around the cylinder) is
U() = 2 U
0
sin , (7.118)
where U
0
is the unperturbed ow speed, and is a cylindrical coordinate dened such that the stagnation point
corresponds to = 0. Note that the tangential ow accelerates (i.e., increases with increasing arc-length, along the
surface of the obstacle, in the direction of the ow) on the front side of the obstacle (i.e., 0 /2), and decelerates
on the back side. Boundary layer separation is always observed to take place at a point on the surface of an obstacle
where there is deceleration of the external tangential ow. In addition, from Section 5.13, the pressure just outside the
boundary layer (and, hence, on the surface of the obstacle, since the pressure is uniform across the layer) is
P() = p
1
+ U
2
0
cos 2, (7.119)
where p
1
is a constant. Note that the tangential pressure gradient is such as to accelerate the tangential ow on the front
side of the obstaclethis is known as a favorable pressure gradient. On the other hand, the pressure gradient is such
as to decelerate the ow on the back sidethis is known as an adverse pressure gradient. Boundary layer separation
is always observed to take place at a point on the surface of an obstacle where the pressure gradient is adverse.
Boundary layer separation is an important physical phenomenon because it gives rise to a greatly enhanced drag
force acting on a non-streamlined obstacle placed in a high Reynolds number ow. This is the case because the pressure
in the comparatively wide wake that forms behind a non-streamlined obstacle, as a consequence of separation, is
relatively low. To be more exact, in the case of a cylindrical obstacle, Equation (7.119) species the expected pressure
variation over the obstacles surface in the absence of separation. It can be seen that the variation on the front side of
the obstacle mirrors that on the back side: i.e., P() = P(). (See Figure 7.12.) In other words, the resultant pressure
force on the front side of the obstacle is equal and opposite to that on the back side, so that the pressure distribution
gives rise to zero net drag acting on the obstacle. Figure 7.12 illustrates how the pressure distribution is modied as
a consequence of boundary layer separation. In this case, the pressure between the separation points is signicantly
less than that on the front side of the obstacle. Consequently, the resultant pressure force on the front side is greater in
magnitude than the oppositely directed force on the back side, giving rise to a signicant drag acting on the obstacle.
Let D be the drag force per unit width (parallel to the axis of the cylinder) exerted on the obstacle. It is convenient to
P() p
1
separation points
3/2 /2 0
stagnation point
Figure 7.12: Pressure variation over surface of a cylindrical obstacle in a high Reynolds number ow both with
(dashed curve) and without (solid curve) boundary layer separation.
parameterize this force in terms of a dimensionless drag coecient,
C
D
=
D
U
2
0
a
, (7.120)
where is the uid density, and a the typical transverse size of the obstacle (in the present example, the radius of
the cylinder). The drag force that acts on a non-streamlined obstacle placed in a high Reynolds number ow, as a
consequence of boundary layer separation, is generally characterized by a drag coecient of order unity. The exact
value of the coecient depends strongly on the shape of the obstacle, but only relatively weakly on the Reynolds
number of the ow. Consequently, this type of drag is termed form drag, since it depends primarily on the external
shape, or form, of the obstacle. Formdrag scales roughly as the cross-sectional area (per unit width) of the vortex-lled
wake that forms behind the obstacle.
Boundary layer separation is associated with strong adverse pressure gradients, or, equivalently, strong ow decel-
eration, on the back side of an obstacle placed in a high Reynolds number ow. Such gradients can be signicantly
reduced by streamlining the obstacle: i.e., by closely aligning its back surface with the unperturbed streamlines of
the external ow. Indeed, boundary layer separation can be delayed, or even completely prevented, on the surface
of a suciently streamlined obstacle, thereby signicantly decreasing, or even eliminating, the associated form drag
(essentially, by reducing the cross-sectional area of the wake). However, even in the limit that the form drag is reduced
to a negligible level, there is still a residual drag acting on the obstacle due to boundary layer viscosity. This type
of drag is called friction drag. As is clear from a comparison of Equations (7.79) and (7.120), the drag coecient
associated with friction drag is O(Re
1/2
), where Re is the Reynolds number of the ow. Friction drag thus tends to
zero as the Reynolds number tends to innity.
The phenomenon of boundary layer separation allows us to resolve dAlemberts paradox. Recall, from Sec-
tion 5.13, that an idealized uid that is modeled as inviscid and irrotational is incapable of exerting a drag force on
a stationary obstacle, despite the fact that very high Reynolds number, ostensibly irrotational, uids are observed to
exert signicant drag forces on stationary obstacles. The resolution of the paradox lies in the realization that, in such
uids, viscosity can only be neglected (and the ow is consequently only irrotational) in the absence of boundary layer
140 FLUID MECHANICS
separation. In this case, the region of the uid in which viscosity plays a signicant role is localized to a thin boundary
layer on the surface of the obstacle, and the resultant friction drag scales as Re
1/2
, and, therefore, disappears in the
inviscid limit (essentially, because the boundary layer shrinks to zero thickness in this limit). On the other hand, if the
boundary layer separates then viscosity is important both in a thin boundary layer on the front of the obstacle, and in a
wide, low-pressure, vortex-lled, wake that forms behind the obstacle. Moreover, the wake does not disappear in the
inviscid limit. The presence of signicant uid vorticity within the wake invalidates irrotational uid dynamics. Con-
sequently, the pressure on the back side of the obstacle is signicantly smaller than that predicted by irrotational uid
dynamics. Hence, the resultant pressure force on the front side is larger than that on the back side, and a signicant
drag is exerted on the obstacle. The drag coecient associated with this type of drag is generally of order unity, and
does not tend to zero as the Reynolds number tends to innity.
7.9 Criterion for Boundary Layer Separation
As we have seen, the boundary layer equations (7.110)(7.113) generally lead to the conclusion that the tangential
velocity in a thin boundary layer, u, is large compared with the normal velocity, v. Mathematically speaking, this
result holds everywhere except in the immediate vicinity of singular points. But, if v u then it follows that the
uid moves predominately parallel to the surface of the obstacle, and can only move away from this surface to a very
limited extent. This restriction eectively precludes separation of the ow from the surface. Hence, we conclude that
separation can only occur at a point at which the solution of the boundary layer equations is singular.
As we approach a separation point, we expect the ow to deviate from the boundary layer towards the interior of
the uid. In other words, we expect the normal velocity to become comparable with the tangential velocity. However,
we have seen that the ratio v/u is of order Re
1/2
[see Equation (7.18)]. Hence, an increase of v to such a degree that
v u implies an increase by a factor Re
1/2
. For suciently large Reynolds numbers, we may suppose that v eectively
increases by an innite factor. Indeed, if we employ the dimensionless form of the boundary layer equations, (7.23)
(7.27), the situation just described is formally equivalent to an innite value of the dimensionless normal velocity, V
y
,
at the separation point.
Let the separation point lie at x = x
0
, and let x < x
0
correspond to the region of the boundary layer upstream of
this point. According to the above discussion,
v(x
0
, y) = (7.121)
at all y (except, of course, y = 0, where the boundary conditions at the surface of the obstacle require that v = 0). It
follows that the deriviative v/y is also innite at x = x
0
. Hence, the equation of continuity, u/x + v/y = 0,
implies that (u/x)
x=x
0
= , or x/u = 0, if x is regarded as a function of u and y. Let u(x
0
, y) = u
0
(y). Close to the
point of separation, x
0
x and u u
0
are small. Thus, we can expand x
0
x in powers of u u
0
(at xed y). Since
(x/u)
u=u
0
= 0, the rst term in this expansion vanishes identically, and we are left with
x
0
x = f (y) (u u
0
)
2
+ O
_
(u u
0
)
3
_
, (7.122)
or
u(x, y) u
0
(y) + (y)

x
0
x, (7.123)
where = 1/
_
f is some function of y. From the equation of continuity,
v
y
=
u
x

(y)
2

x
0
x
. (7.124)
Upon integration, the above expression yields
v(x, y)
(y)
x
0
x
, (7.125)
where
(y) =
1
2
_
y
(y
) dy
. (7.126)
The equation of tangential motion in the boundary layer, (7.111), is written
u
u
x
+ v
u
y
= U
dU
dx
+
2
u
y
2
. (7.127)
As is clear from Equation (7.123), the derivative
2
u/y
2
does not become innite as x x
0
. The same is true of
the function U dU/dx, which is determined from the ow outside the boundary layer. However, both terms on the
left-hand side of the above expression become innite as x x
0
. Hence, in the immediate vicinity of the separation
point,
u
u
x
+ v
u
y
0. (7.128)
Since u/x = v/y, we can rewrite this equation in the form
u
v
y
+ v
u
y
= u
2

y
_
v
u
_
0. (7.129)
Since u does not, in general, vanish at x = x
0
, we conclude that
y
_
v
u
_
0. (7.130)
In other words, v/u is a function of x only. Now, from (7.123) and (7.125),
v
u
=
(y)
u
0
(y)

x
0
x
+ O(1). (7.131)
Hence, if this ratio is a function of x alone then (y) = (1/2) Au
0
(y), where A is a constant: i.e.,
v(x, y)
Au
0
(y)
2

x
0
x
. (7.132)
Finally, since (7.126) yields = 2 d/dy = Adu
0
/dy, we obtain
u(x, y) u
0
(y) + A
du
0
dy
x
0
x. (7.133)
The previous two expressions specify u and v as functions of x and y near the point of separation. Beyond the point
of separation, that is for x > x
0
, the expressions are physically meaningless, since the square roots become imaginary.
This implies that the solutions of the boundary layer equations cannot sensibly be continued beyond the separation
point.
Now, the standard boundary conditions at the surface of the obstacle require that u = v = 0 at y = 0. It, therefore,
follows from Equations (7.132) and (7.133) that
u
0
(0) = 0, (7.134)
du
0
dy
y=0
= 0. (7.135)
Thus, we obtain the important prediction that both the tangential velocity, u, and its rst derivative, u/y, are zero at
the separation point (i.e., x = x
0
and y = 0). This result was originally obtained by Prandtl, although the argument we
have used to derive it is due to L.D. Landau.
Note that if the constant A in expressions (7.132) and (7.133) happens to be zero then the point x = x
0
and y = 0,
at which the derivative u/y vanishes, has no particular properties, and is not a point of separation. However, there is
no reason, in general, why A should take the special value zero. Thus, in practice, a point on the surface of an obstacle
at which u/y = 0 is always a point of separation.
Incidentally, if there were no separation at the point x = x
0
(i.e., if A = 0) then we would have u/y < 0 for
x > x
0
. In other words, u would become negative as we move away from the surface, y being still small. That is, the
142 FLUID MECHANICS
uid beyond the point x = x
0
would move tangentially, in the region of the boundary layer immediately adjacent to
the surface, in the direction opposite to that of the external ow: i.e., there would be back-ow in this region. In
practice, the ow separates from the surface at x = x
0
, and the back-ow migrates into the wake.
Note that the dimensionless boundary layer equations, (7.23)(7.27), are independent of the Reynolds number of
the external ow (assuming that this number is much greater than unity). Thus, it follows that the point on the surface
of the obstacle at which u/y = 0 is also independent of the Reynolds number. In other words, the location of the
separation point is independent of the Reynolds number (as long as this number is large, and the ow in the boundary
layer is non-turbulent).
At y = 0, the equation of tangential motion in the boundary layer, (7.111), is written
2
u
y
2
y=0
=
1
U
dU
dx
=
1
dP
dx
, (7.136)
where P(x) is the pressure just outside the layer, and use has been made of (7.6). Now, since u is positive, and increases
away from the surface (upstream of the separation point), it follows that (
2
u/y
2
)
y=0
> 0 at the separation point itself,
where (u/y)
y=0
= 0. Hence, according to the above equation,
_
dU
dx
_
x=x
0
< 0, (7.137)
_
dP
dx
_
x=x
0
> 0. (7.138)
In other words, we predict that the external tangential ow is always decelerating at the separation point, whereas the
pressure gradient is always adverse (i.e., such as to decelerate the tangential ow), in agreement with experimental
observations.
7.10 Approximate Solutions of Boundary Layer Equations
The boundary layer equations, (7.110)(7.113), take the form
u
x
+
v
y
= 0, (7.139)
u
u
x
+ v
u
y
U
dU
dx
=

2
u
y
2
, (7.140)
u(x, ) = U(x), (7.141)
u(x, 0) = 0, (7.142)
v(x, 0) = 0. (7.143)
Furthermore, it follows from (7.140), (7.142), and (7.143) that

2
u
y
2
y=0
= U
dU
dx
. (7.144)
The above expression can be thought of as an alternative form of (7.143). As we saw in Section 7.4, the boundary
layer equations can be solved exactly when U(x) takes the special form U
0
x
m
. However, in the general case, we must
resort to approximation methods.
Following Pohlhausen, let us assume that
u(x, y)
U(x)
= f (), (7.145)
where = y/(x), and /x 1/. In particular, suppose that
f () =
_
a + b + c
2
+ d
3
+ e
4
0 1
1 > 1
, (7.146)
where a, b, c, d, and e are constants. This expression automatically satises the boundary condition (7.141). Moreover,
the boundary conditions (7.142) and (7.144) imply that a = 0, and
f
(0) = (x), (7.147)

where

= d/d, and
=
dU
dx
. (7.148)
Finally, let us assume that f , f
, and f
are continuous at = 1: i.e.,

f (1) = 1, (7.149)
f
(1) = 0, (7.150)
f
(1) = 0. (7.151)
These constraints corresponds to the reasonable requirements that the velocity, vorticity, and viscous stress tensor,
respectively, be continuous across the layer. Given that a = 0, Equations (7.146), (7.147), and (7.149)(7.151) yield
f () = F() + G() (7.152)
for 0 1, where
F() = 1 (1 )
3
(1 + ), (7.153)
G() =
1
6
(1 )
3
. (7.154)
Thus, the tangential velocity prole across the layer is a function of a single parameter, , which is termed the
Pohlhausen parameter. The behavior of this prole is illustrated in Figure 7.13. Note that, under normal circum-
stances, the Pohlhausen parameter must lie in the range 12 12. For > 12, the prole is such that f () > 1
for some < 1, which is not possible in a steady-state solution. On the other hand, for < 12, the prole is such
that f
(0) < 0, which implies ow reversal close to the wall. As we have seen, ow reversal is indicative of separation.
Indeed, the separation point, f
(0) = 0, corresponds to = 12. Note that expression (7.152) is only an approxima-

tion, since it satises some, but not all, of the boundary conditions satised by the true velocity prole. For instance,
dierentiation of (7.140) with respect to y reveals that (
3
u/y
3
)
y=0
f
(0) = 0, which is not the case for expression

(7.152).
It follows from Equations (7.115), (7.116), and (7.152)(7.154) that
1
(x) =
_
1
0
(1 f ) d =
_
3
10

120
_
, (7.155)
2
(x) =
_
1
0
f (1 f ) d =
_
37
315

945

2
9072
_
. (7.156)
Furthermore,
u
y
y=0
=
U
(0) =
U
_
2 +
6
_
. (7.157)
Now, the von K arm an momentum integral, (7.117), can be rearranged to give
U
2
d
2
dx
+
2
2
dU
dx
_
2 +
2
_
=
2
U
u
y
y=0
. (7.158)
144 FLUID MECHANICS
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
f
0 0.2 0.4 0.6 0.8 1
Figure 7.13: Pohlhausen velocity proles for = 12 (solid curve) and = 12 (dashed curve).
Dening
(x) =
2
2
dU
dx
, (7.159)
we obtain
U
d
dx
_

dU/dx
_
= 2 [F
2
() 2 + F
1
()] = F(), (7.160)
where
=
_
37
315

945

2
9072
_
2
, (7.161)
F
1
() =
2
=
_
3
10

120
_ __
37
315

945

2
9072
_
, (7.162)
F
2
() =
2
U
u
y
y=0
=
_
2 +
6
_
_
37
315

945

2
9072
_
, (7.163)
F() = 2
_
37
315

945

2
9072
_ _
2
116
315
+
_
2
945
+
1
120
_
2
+
2
9072

3
_
. (7.164)
It is generally necessary to integrate Equation (7.158) fromthe stagnation point at the front of the obstacle, through the
point of maximum tangential velocity, to the separation point on the back side of the obstacle. Now, at the stagnation
point we have U = 0 and dU/dx 0, which implies that F() = 0. Furthermore, at the point of maximum tangential
velocity we have dU/dx = 0 and U 0, which implies that = = 0. Finally, as we have already seen, = 12 at
the separation point, which implies, from (7.161), that = 0.1567.
As was rst pointed out by Walz, and is illustrated in Figure 7.14, it is a fairly good approximation to replace F()
by the linear function 0.47 6 for in the physically relevant range. The approximation is particularly accurate on
the front side of the obstacle (where > 0). Making use of this approximation, Equations (7.159) and (7.160) reduce
0
1
F
0.15 0.1 0.05 0 0.05 0.1
Figure 7.14: The function F() (solid curve) and the linear function 0.47 6 (dashed line).
to the linear dierential equation
d
dx
_
_
U
2
2
_
= 0.47 5
dU
dx
2
2
, (7.165)
2
2
=
0.47
U
6
_
x
0
U
5
(x
) dx
, (7.166)
assuming that the stagnation point corresponds to x = 0. It follows that
=
0.47
U
6
dU
dx
_
x
0
U
5
(x
) dx
. (7.167)
Recall that the separation point corresponds to x = x
s
, where (x
s
)
s
= 0.1567.
Suppose that U(x) = U
0
, which corresponds to uniform ow over a at plate. (See Section 7.5.) It follows from
Equations (7.166) and (7.167) that
2
(x)
x
=
0.69
Re
1/2
, (7.168)
where Re = U
0
x/, and = 0. Moreover, according to Equations (7.148) and (7.162), = 0 and
1
/
2
= 2.55.
Hence, the displacement width of the boundary layer becomes
1
(x)
x
=
1.75
Re
1/2
. (7.169)
This approximate result compares very favorably with the exact result, (7.73).
Suppose that x = a and U() = 2 U
0
sin , which corresponds to uniform transverse ow around a circular
cylinder of radius a. (See Section 7.8.) Equation (7.167) yields
() = 0.47
cos
sin
6
_

0
sin
5
. (7.170)
146 FLUID MECHANICS
0
0.1
0.2
s
0 20 40 60 80 100 120
(
)
Figure 7.15: The function () for ow around a circular cylinder.
x
Figure 7.16: Flow over the back surface of a semi-innite wedge.
Figure 7.15 shows () determined from the above formula. It can be seen that =
s
= 0.1567 when =
s
108
.
In other words, the separation point is located 108
from the stagnation point at the front of the cylinder. This suggests
that the low pressure wake behind the cylinder is almost as wide as the cylinder itself, and that the associated form
drag is comparatively large.
Suppose, nally, that U = U
0
x
m
. If m is negative then, as illustrated in Figure 7.16, this corresponds to uniform
ow over the back surface of a semi-innite wedge whose angle of dip is
=
m
1 + m
2
. (7.171)
(See Section 7.4.) It follows from (7.167) that
=
0.47 m
1 + 5 m
=
0.47
/2 4
. (7.172)
Now, we expect boundary layer separation on the back surface of the wedge when <
s
= 0.1567. This corresponds
to >
s
, where
s
=
2
(
s
)
0.47 + 4 (
s
)
13
. (7.173)
Hence, boundary layer separation can be prevented by making the wedges angle of dip suciently shallow: i.e., by
streamlining the wedge, which has the eect of reducing the deceleration of the ow on the wedges back surface.
Note that the critical value of m (i.e., m
s
= 0.0125) at which separation occurs in our approximate solution is very
similar to the critical value of m (i.e., m
= 0.0905) at which the exact self-similar solutions described in Section 7.4

can no longer be found. This suggests that the absence of self-similar solutions for m < m
is related to boundary layer

separation.
7.11 Exercises
7.1. Fluid ows between two non-parallel plane walls, towards the intersection of the planes, in such a manner that if x is
measured along a wall from the intersection of the planes then U(x) = U
0
/x, where U
0
is a positive constant. Verify that
a solution of the boundary layer equation (7.35) can be found such that is a function of y/x only. Demonstrate that this
solution yields
u(x, y)
U(x)
= F
_
_
U
0
_1/2
y
x
_
,
where u = /y, and
F
F
2
= 1,
subject to the boundary conditions F(0) = 0 and F() = 1. Verify that
F(z) = 3 tanh
2
_
+
z
2
_
2
is a suitable solution of the above dierential equation, where tanh
2
= 2/3.
7.2. A jet of water issues from a straight narrow slit in a wall, and mixes with the surrounding water, which is at rest. On the
assumption that the motion is non-turbulent and two-dimensional, and that the approximations of boundary layer theory
apply, the stream function satises the boundary layer equation
y
3

x
y
2
+
x y
= 0.
Here, the symmetry axis of the jet is assumed to run along the x-direction, whereas the y-direction is perpendicular to this
axis. The velocity of the jet parallel to the symmetry axis is
u(x, y) =
y
,
where u(x, y) = u(x, y), and u(x, y) 0 as y . We expect the momentum ux of the jet parallel to its symmetry axis,
M =
_

u
2
dy,
to be independent of x.
Consider a self-similar stream function of the form
(x, y) =
0
x
p
F(y/x
q
).
Demonstrate that the boundary layer equation requires that p +q = 1, and that M is only independent of x when 2 p q = 0.
Hence, deduce that p = 1/3 and q = 2/3.
Suppose that
(x, y) = 6 x
1/3
F(y/x
2/3
).
Demonstrate that F(z) satises
F
+ 2 F F
+ 2 F
2
= 0,
subject to the constraints that F
(z) = F
(z), and F
(z) 0 as z . Show that

F(z) = tanh(z)
is a suitable solution, and that
M = 48
2
3
.
148 FLUID MECHANICS
7.3. The growth of a boundary layer can be inhibited by sucking some of the uid through a porous wall. Consider conventional
boundary layer theory. As a consequence of suction, the boundary condition on the normal velocity at the wall is modied
to v(x, 0) = v
s
, where v
s
is the (constant) suction velocity. Demonstrate that, in the presence of suction, the von K arm an
velocity integral becomes
u
y
y=0
= U
2
d
2
dx
+ U
dU
dx
(
1
+ 2
2
) + U v
s
.
Suppose that
u(x, y) = U(x)
_
sin(y) 0 y /(2 )
1 y > /(2 )
,
where = (x). Demonstrate that the displacement and momentum widths of the boundary layer are
1
= (/2 1)
1
,
2
= (1 /4)
1
,
respectively. Hence, deduce that
(/2 1)
2
1
= U (1 /4)
d
1
dx
+
dU
dx

1
+ (/2 1) v
s
.
Consider a boundary layer on a at plate, for which U(x) = U
0
. Show that, in the absence of suction,
1
= (/2 1)
_
8
4
_
1/2
_
x
U
0
_
1/2
,
but that in the presence of suction
1
=
(/2 1)
v
s
.
Hence, deduce that, for a plate of length L, suction is capable of signicantly reducing the thickness of the boundary layer
when
v
s
U
0
1
Re
1/2
,
where Re = U
0
L/.
Incompressible Aerodynamics 149
8 Incompressible Aerodynamics
8.1 Introduction
This chapter investigates the forces exerted on a stationary obstacle situated in a uniform, high Reynolds number wind,
on the assumption that the obstacle is suciently streamlined that there is no appreciable separation of the boundary
layer from its back surface. Such an obstacle is termed an airfoil (or aerofoil). Obviously, airfoil theory is fundamental
to the theory of ight. The ow around an airfoil is essentially irrotational and inviscid everywhere apart from a thin
boundary layer localized to its surface, and a thin wake emitted by its trailing edge. (See Sections 7.5 and 7.6.) It
follows that, for the ow external to the boundary layer and wake, we can write
v = , (8.1)
which automatically ensures that the owis irrotational. Assuming that the owis also incompressible, so that v = 0,
the velocity potential, , satises Laplaces equation: i.e.,
2
= 0. (8.2)
The appropriate boundary condition at the surface of the airfoil is that the normal velocity be zero. In other words,
n = 0, where n is a unit vector normal to the surface. In general, the tangential velocity at the airfoil surface,
obtained by solving
2
= 0 in the external region, subject to the boundary condition n = 0 on the surface, is
non-zero. Of course, this is inconsistent with the no slip condition, which demands that the tangential velocity be zero
at the surface. (See Section 7.2.) However, as described in the previous chapter, this inconsistency is resolved by the
boundary layer, across which the tangential velocity is eectively discontinuous, being non-zero on the outer edge
of the layer (where it interfaces with the irrotational ow), and zero on the inner edge (where it interfaces with the
airfoil). The discontinuity in the tangential velocity across the layer implies the presence of bound vortices covering
the surface of the airfoil (see Section 8.7), and also gives rise to a friction drag acting on the airfoil in the direction of
the external ow. However, the magnitude of this drag scales as Re
1/2
, where Re is the Reynolds number of the wind.
(See Section 7.5.) Hence, such drag becomes negligibly small in the high Reynolds number limit. In the following,
we shall assume that any form drag, due to the residual separation of the boundary layer at the back of the airfoil, is
also negligibly small. Moreover, for the sake of simplicity, we shall initially restrict our discussion to two-dimensional
situations in which a high Reynolds number wind ows transversely around a stationary airfoil of innite length (in
the z-direction) and uniform cross-section (parallel to the x-y plane).
8.2 Theorem of Kutta and Zhukovskii
Consider a two-dimensional airfoil that is at rest in a uniform wind of speed V whose direction subtends a (clockwise)
angle with the negative x-axis. It follows that the wind velocity is V = V cos e
x
+V sin e
y
, and the correspond-
ing complex velocity is dF/dz = V e
i
. (See Section 6.4.) Now, the air velocity a great distance from the airfoil must
tend toward this uniform velocity. Thus, for suciently large z, we can write (see Section 6.4)
dF
dz
= V e
i
+
A
z
+
B
z
2
+ . (8.3)
According to Equation (6.87), the circulation, , of air about the airfoil is determined by performing the integral
Re
_
C
dF
dz
dz = (8.4)
around a loop C that lies just above the airfoil surface. However, as discussed in Section 6.8, the value of this integral
is unchanged if it is performed around any loop that can be continuously deformed onto C, whilst not passing through
the airfoil surface, or crossing a singularity of the complex velocity, dF/dz (i.e., a line source or a z-directed vortex
150 FLUID MECHANICS
lament). Since (in the high Reynolds number limit in which the boundary layer and the wake are innitely thin) there
are no line sources or z-directed vortex laments external to the airfoil, we can evaluate the integral around a large
circle of radius R, centered on the origin. It follows that z = Re
i
and dz = i Re
i
d = i z d. Hence,
= Re
_
i
_
_
V Re
i (+)
+ A + O(R
1
)
_
d
_
= 2 Im(A), (8.5)
which implies that
dF
dz
= V e
i
+ i

2 z
+
B
z
2
+ (8.6)
at large z.
As discussed in Section 6.9, the net force (per unit length) acting on the airfoil, L = X e
x
+ Y e
y
, is determined by
performing the Blasius integral,
1
2
i
_
C
_
dF
dz
_
2
dz = X i Y, (8.7)
around a loop C that lies just above the airfoil surface. However, as before, the value of the integral is unchanged if
we perform it instead around a large circle of radius R, centered on the origin. Now, far from the airfoil,
_
dF
dz
_
2
= V
2
e
2 i
+ i
V e
i
z
+
8
2
V Be
i

2
4
2
z
2
+ O(z
3
). (8.8)
So, we obtain
X i Y =
1
2
i
2
_ _
V
2
Re
i (+2 )
+ i
V e
i
+ O(R
1
)
_
d = i e
i
V , (8.9)
or
X + i Y = i e
i
V = e
i (/2)
V . (8.10)
In other words, the resultant force (per unit length) acting on the airfoil is of magnitude V , and has the direction
obtained by rotating the wind vector through a right-angle in the sense opposite to that of the circulation. This type of
force is known as lift, and is responsible for ight. The result (8.10) is known as the theorem of Kutta and Zhukovskii,
after the German scientist M.W. Kutta, and the Russian scientist N.E. Zhukovskii, who discovered it independently.
Note that (at xed circulation) the lift is independent of the shape of the airfoil. Furthermore, according to the Kutta
Zhukovskii theorem, there is zero drag acting on the airfoil (i.e., zero force acting in the direction of the wind). In
reality, there is always a small friction drag due to air viscosity, as well as a (hopefully) small formdrag due to residual
separation of the boundary layer from the back of the airfoil. There is actually a third type of drag, known as induced
drag, which is discussed in Section 8.8.
As discussed in Section 6.9, the net moment per unit length (about the origin), M, acting on the airfoil is determined
by performing the integral
Re
_
1
2

_
C
_
dF
dz
_
2
z dz
_
_
= M (8.11)
around a loop C that lies just above the airfoil surface. As before, we can deform C into a circle of radius R, centered
on the origin, without changing the value of the integral. Hence, we obtain
M = Re
_
1
2
i
_ _
V
2
R
2
e
2 i (+)
+ i
V Re
i (+)
+
8
2
V Be
i

2
4
2
+ O(R
1
)
_
d
_
, (8.12)
or
M = Re
_
2 V Be
i (/2)
_
. (8.13)
8.3 Cylindrical Airfoils
For the moment, let us work in the complex -plane, where = + i . Consider a cylindrical airfoil with a circular
cross-section of radius a, centered on the origin, that is situated in a uniform, high Reynolds number wind of speed
V whose direction subtends a (clockwise) angle with the negative -axis. Let be the circulation of air around the
airfoil. A slight generalization of the analysis of Section 6.4 reveals that the appropriate complex velocity potential is
F() = V
_
e
i
+
a
2
e
i
_
+ i

2
ln
_
a
_
, (8.14)
whereas the associated stream function takes the form
(r, ) = V
_
r
a
2
r
_
sin( + ) +
2
ln
_
r
a
_
, (8.15)
where = r e
i
. It follows that
dF
d
= V e
i
+ i

2
V a
2
e
i
2
. (8.16)
Comparison with Equation (8.6) (with z = ) reveals that
B = V a
2
e
i
. (8.17)
Hence, Equations (8.10) and (8.13) yield
V = V
_
cos e
+ sin e
_
, (8.18)
L = V
_
sin e
+ cos e
_
, (8.19)
M = 0, (8.20)
where V is the wind vector, L the lift vector, and M the moment of the lift vector about the origin. We conclude that,
for a cylindrical airfoil of circular cross-section, the lift vector is normal to the wind vector, and the line of action of
the lift passes through the centroid of the cross-section (since the lift generates zero moment about the origin). See
Figure 8.1.
Of course, a cylindrical airfoil of circular cross-section is completely unrealistic, since its back side (i.e., the side
opposite to that from which the wind is incident) is not suciently streamlined to prevent boundary layer separation.
(See Chapter 7.) However, as described in Section 6.7, we can use the conformal transformation
z = +
l
2
(8.21)
to transform a cylinder of circular cross-section in the -plane to a cylinder of elliptical cross-section in the z-plane.
(Note that both cross-sections have centroids located at the origin.) Moreover, a cylindrical airfoil of elliptical cross-
section that is suciently elongated, and whose major axis subtends a suciently small angle with the incident wind
direction, constitutes a realistic airfoil, since its back side is, for the most part, closely aligned with the external ow.
An elliptical airfoil of width c and thickness < c, as shown in Figure 8.2, is obtained when the parameters a and
l are given the following values:
a =
1
4
(c + ), (8.22)
l =
1
4
(c
2

2
)
1/2
. (8.23)
In this case, the surface of the airfoil satises the parametric equations
x =
c
2
cos , (8.24)
y =
2
sin . (8.25)
152 FLUID MECHANICS
a
V
L
Figure 8.1: A cylindrical airfoil of circular cross-section.
M
y
x
c
L
V
F C
Figure 8.2: A cylindrical airfoil of elliptical cross-section.
In particular, the airfoils leading and trailing edges correspond to = 0 and = , respectively.
According to Equations (8.16) and (8.21), the complex velocity in the z-plane is given by
dF
dz
=
dF
d
d
dz
=
_
V e
i
+ i

2
V a
2
e
i
2
_ _

2
2
l
2
_
. (8.26)
Thus, on the airfoil surface, where = a e
i
, we obtain
dF
dz
= i
_
V sin( + ) +
(c + )
_
(c + )
( cos + i c sin )
. (8.27)
A long way from the airfoil, z l
2
/z, so that Equation (8.26) reduces to
dF
dz
V e
i
+ i

2 z
+
V (l
2
e
i
a
2
e
i
)
z
2
. (8.28)
A comparison with Equation (8.6) reveals that the circulation of air around the airfoil takes the same value, , in both
the complex - and z-planes. In other words, the conformal transformation (8.21) does not modify the circulation.
Note that the transformation also does not modify the external wind speed or direction [since, from (8.16) and (8.28),
dF/d = dF/dz = V e
i
at very large and z]. On the other hand, it is clear that the constant B, which takes the
value zero in the complex -plane, takes the value
B = V (l
2
e
i
a
2
e
i
) (8.29)
in the complex z-plane. Hence, Equations (8.10) and (8.13) reveal that
V = V e
j
, (8.30)
L = V e
, (8.31)
M =
8
V
2
(c
2

2
) sin(2 ), (8.32)
where V is the wind vector, L the lift vector, and M the moment of the lift vector about the origin. Here,
e
j
= cos e
x
+ sin e
y
(8.33)
is a unit vector parallel to the incident wind direction, and
e
= e
j
e
z
= sin e
x
+ cos e
y
(8.34)
is a unit vector perpendicular to the wind direction. We conclude that, for a cylindrical airfoil of elliptic cross-section,
the lift vector is normal to the wind vector, but the line of action of the lift intersects the major axis of the airfoil a
distance
d =
M
L cos
=
1
4
(c )
sin
(8.35)
in front of the cross-sections centroid, C, where

= /[ V (c + )]. See Figure 8.2. Incidentally, the point, F, at
which the line of action of the lift intersects the airfoils major axis is conventionally termed the focus of the airfoil.
8.4 Zhukovskiis Hypothesis
According to the previous analysis, the lift acting on a cylindrical airfoil of elliptic cross-section, situated in a uniform,
high Reynolds number wind, depends on the circulation, , of air about the airfoil. But, how can we determine the
value of this circulation?
Figure 8.3 shows the boundary layer and wake of a streamlined airfoil. The boundary layer, which is localized on
the surface of the airfoil, has a vortex intensity per unit length in the z-direction equal to U, where U is the tangential air
154 FLUID MECHANICS
Airfoil
Trailing Edge
Boundary Layer
Wake
C
dS
V
U
Figure 8.3: The boundary layer and wake of a streamlined airfoil. Only shaded regions posses non-zero vorticity.
speed immediately above the layer. [See Equation (8.64).] Moreover, the wake is emitted by the airfoils trailing edge,
and subsequently convected by the external air ow. (See Section 7.6.) Note that the ow is irrotational everywhere
apart from inside the boundary layer and the wake. Now, according to the analysis of Section 5.13, the rate of change
of the circulation, , around some curve C that encloses the airfoil is equal to minus the ux of z-directed vorticity
across this curve: i.e.,
d
dt
=
_

z
v dS. (8.36)
Here, v is the wind velocity, the wind vorticity, and dS an outward surface element (of unit depth in the z-direction)
lying on C. We expect the vorticity ux to be independent of the size and shape of C, otherwise the circulation of the
ow, , about the airfoil would not have a unique value. In the limit that C becomes very large, v V, where V is
the incident wind velocity. Thus,
d
dt
= V
z
, (8.37)
where V is the wind speed, and
z
the vortex intensity per unit length in the wake (at the point where it crosses the
curve C). Here, we are assuming that the vorticity within the wake is convected by the ow, giving rise to a net ux
of vorticity across C. Since the wake is essentially an extension of the boundary layer, it is reasonable to assume that
its vortex intensity per unit length is proportional to that in the boundary layer at the airfoils trailing edge, where the
wake and boundary layer intersect. In other words,
z
= k U
0
, where U
0
is the tangential velocity immediately above
the trailing edge of the airfoil, and k is a constant. It follows that
d
dt
= k V U
0
. (8.38)
According to Equation (8.27), the tangential velocity just above the surface of the airfoil is
U() = Re
_
c sin i cos
(c
2
sin
2
+
2
cos
2
)
1/2
dF
dz
_
=a
=
_
V sin( + ) +
(c + )
_
(c + )
(c
2
sin
2
+
2
cos
2
)
1/2
. (8.39)
2
1
0
1
2
y/c
2 1 0 1 2
x/c
Figure 8.4: Streamlines around a slender cylindrical airfoil of elliptic cross-section situated in a uniform, high
Reynolds number wind. The parameters for this calculation are /c = 0.1, = /12, and = 0.
Hence, given that the airfoils trailing edge corresponds to = , we obtain
U
0
=
_

(c + )
V sin
_
_
c +
_
. (8.40)
Thus, Equation (8.38) yields
d
d t
=
+ sin , (8.41)
where

= /[ V (c + )], t = t/t
0
, and t
0
= /(k V). Assuming that the circulation of the ow about the airfoil is
initially zero (i.e.,

= 0 at t = 0), the above equation can be solved to give
= sin
_
1 exp( t)
_
. (8.42)
Clearly, as t the normalized circulation

asymptotes to the constant value
= sin . (8.43)
The corresponding constant value of the unnormalized circulation is
= V (c + ) sin . (8.44)
Note that, according to Equation (8.40), when the circulation, , takes the value
the tangential velocity at the

airfoils training edge, U
0
, is zero. In other words, the steady-state circulation set up around the airfoil is such as to
render its trailing edge a stagnation point of the ow. This conclusion is known as Zhukovskiis hypothesis, after its
discoverer N.E. Zhukovskii.
Incidentally, it should be clear, from the above discussion, that the air circulation about the airfoil is only able to
change its value because of the presence of the boundary layer, and the associated wake that trails from the airfoils
156 FLUID MECHANICS
2
1
0
1
2
y/c
2 1 0 1 2
x/c
Figure 8.5: Streamlines around a slender cylindrical airfoil of elliptic cross-section situated in a uniform, high
Reynolds number wind. The parameters for this calculation are /c = 0.1, = /12, and =
.
trailing edge. This follows because the ow is irrotational everywhere except within the boundary layer and the wake.
Moreover, as we have seen, a change in circulation is necessarily associated with a net vorticity ux away from the
airfoil, and such a ux cannot be carried by an irrotational wind. Thus, in the absence of the boundary layer and the
wake, the air circulation about the airfoil would be constrained to remain zero (assuming that it was initially zero),
in accordance with the Kelvin circulation theorem. (See Section 5.13.) This implies, from Equation (8.31), that zero
lift would act on the airfoil, irrespective of its shape, and irrespective of the incident wind speed or direction. In
other words, ight would be impossible. Fortunately, as long as the tangential air velocity at the trailing edge of
the airfoil is non-zero, the wake that trails behind the airfoil carries a net ux of z-directed vorticity, which causes
the airfoil circulation to evolve in time. This process continues until the circulation becomes such that the tangential
velocity at the airfoils trailing edge is zero: i.e., such that the trailing edge is a stagnation point. Thereafter, the
circulation remains constant (assuming that the wind speed and direction remain constant). Figures 8.4 and 8.5 show
the streamlines of the ow around a slender cylindrical airfoil of elliptic cross-section situated in a uniform, high
Reynolds number wind whose direction of incidence is slightly inclined to the airfoils major axis. In the rst gure,
the air circulation about the airfoil is zero. In the second gure, the circulation is such as to make the trailing edge of
the airfoil a stagnation point.
According to Equations (8.31) and (8.44), when the air circulation about the airfoil has attained its steady-state
value,
, the lift acting on the airfoil becomes

L = V
= V
2
(c + ) sin . (8.45)
Note that the lift is positive (i.e., upward) when > 0 (i.e., when the wind is incident on the airfoils bottom surface),
negative (i.e., downward) when < 0 (i.e., when the wind is incident on the airfoils top surface), and zero when = 0
(i.e., when the wind is incident parallel to airfoils major axis). Incidentally, the angle is conventionally termed the
angle of attack. Finally, from Equations (8.35) and (8.43), the focus of the airfoil is located a distance
d =
1
4
(c )
sin
=
1
4
(c ) (8.46)
0
180
)
0 10 20
(
)
Figure 8.6: The angular locations of the boundary layer separation points,
, calculated as a function of the angle of

attack, , for a cylindrical airfoil of elliptic cross-section situated in a uniform, high Reynolds number wind. The solid,
dashed, short-dashdotted, and long-dashdotted curves correspond to airfoils of ellipticity /c = 1.0, 0.5, 0.25, and
0.125, respectively. The trailing edge of the airfoil is located at = 180
.
in front of the centroid of its cross-section. In the limit that the airfoil becomes very thin (i.e., c), this distance
asymptotes to c/4. Thus, we conclude that the focus of a thin airfoil, which is dened as the point of action of the lift,
is located one quarter of the way along the airfoil from its leading edge.
The above analysis is premised on the assumption that there is no appreciable separation of the boundary layer
from the back of the airfoil, which implies the neglect of form drag. We can check that this assumption is reasonable
by calculating the approximate locations of the boundary layer separation points using the analysis of Section 7.10.
Let s represent arc-length along the surface of the airfoil, measured from the front stagnation point. Assuming that,
in accordance with Zhukovskiis hypothesis, the circulation is such that =
, this stagnation point is located at

=
0
, where
0
= 2 . [See Equation (8.49).] It follows that
ds = (dx
2
+ dy
2
)
1/2
=
1
2
h() d, (8.47)
where
h() = (c
2
sin
2
+
2
cos
2
)
1/2
. (8.48)
Moreover, from (8.39) and (8.44), the tangential air speed just above the surface of the airfoil can be written
U() = V (c + )
f ()
h()
, (8.49)
with
f () = sin( + ) + sin . (8.50)
In addition, it can be shown that
d
d
ln
_
f
h
_
= g(), (8.51)
158 FLUID MECHANICS
v
z
(y = 0
)
dz
v
z
(y = 0
+
)
y
z
Figure 8.7: Side view of a vortex sheet.
where
g() =
cos( + )
f ()

(c
2

2
) cos sin
h
2
()
. (8.52)
According to the analysis of Section 7.10, the separation points are located at =
, where (
) = 0.1567, and
() = 0.47
h
4
() g()
f
5
()
_

0
f
5
(
)
h
4
(
)
d
. (8.53)
Here, >
+
>
0
and
0
>
> . Moreover, the x- and y-coordinates of the separation points are x
= (c/2) cos
and y
= (/2) sin
, respectively. Figure 8.6 shows the angular locations of the separation points, calculated as
a function of the angle of attack, for cylindrical airfoils of various dierent ellipticity, /c. (Note that
has been
re-expressed as an angle in the range 0 to 2.) It can be seen that for a blu airfoil (e.g., /c = 1) the angular distance
between the separation points is large, indicating the presence of a wide wake, and a high associated form drag (since
the magnitude of form drag is roughly proportional to the width of the wake). On the other hand, for a slender airfoil
(e.g., /c = 0.125) the angular distance between the separation points is much smaller, indicating the presence of a
narrow wake, and a low associated form drag. Note, however, that, in the latter case, as the angle of attack is gradually
increased from zero, there is an initial gradual increase in the angular distance between the separation points, followed
by an abrupt, and very large, increase. We would expect there to be a similar gradual increase in the form drag,
followed by an abrupt, and very large, increase. The value of the angle of attack at which this abrupt increase occurs
is termed the critical angle of attack. We conclude that the previous analysis, which neglects form drag, is valid only
for slender airfoils whose angles of attack do not exceed the critical value (which is generally only a few degrees).
8.5 Vortex Sheets
A vortex sheet is dened as a planar array of parallel vortex laments. Consider a uniform vortex sheet, lying in the
x-z plane, in which the vortex laments run parallel to the x-axis. See Figure 8.7. The vorticity within the sheet can
be written
=
x
(y) e
x
. (8.54)
Here, =
x
e
x
is the sheets vortex intensity per unit length. Let v
z
(y = 0
+
) and v
z
(y = 0
) be the z-component of
the uid velocity immediately above and below the sheet, respectively. Consider a small rectangular loop in the y-z
plane that straddles the sheet, as shown in the gure. Integration of = v around the loop (making use of the curl
theorem) yields
v
z
v
z
(y = 0
+
) v
z
(y = 0
) =
x
. (8.55)
In other words, a vortex sheet induces a discontinuity in the tangential ow across the sheet. The above expression can
easily be generalized to give
= n v, (8.56)
where is the sheets vortex intensity per unit length, n is a unit vector normal to the sheet, and v is the jump in
tangential velocity across the sheet (traveling in the direction of n). Furthermore, it is reasonable to assume that the
above relation holds locally for non-planar and non-uniformvortex sheets.
8.6 Induced Flow
A vortex lament is necessarily associated with uid ow circulating about the lament. Let us determine the rela-
tionship between the lament vorticity and the ow eld that it induces. This problem is mathematically identical to
determining the magnetic eld generated by a current lament. In the latter case, the Maxwell equation
0
j = B (8.57)
can be inverted to give the well-known Biot-Savart law
B(r) =
1
4
_
j(r
) (r r
)
r r
3
d
3
r
. (8.58)
Here, j is the current density, and B the magnetic eld-strength. By analogy, given that vorticity is related to uid
velocity via the familiar relation
= v, (8.59)
we can write
v(r) =
1
4
_
(r
) (r r
)
r r
3
d
3
r
. (8.60)
This expression allows us to determine the ow eld induced by a given vorticity distribution. In particular, for a
vortex lament of intensity the above expression reduces to
v(r) =
1
4
_
(r
) (r r
)
r r
3
dl
, (8.61)
where dl is an element of length along the lament. Likewise, for a vortex sheet of intensity per unit length , we
obtain
v(r) =
1
4
_
(r
) (r r
)
r r
3
dS
, (8.62)
where dS is an element of area of the sheet.
8.7 Three-Dimensional Airfoils
Let us now take into account the fact that realistic three-dimensional airfoils are of nite size. Consider Figure 8.8,
which shows a top view of a stationary airfoil of nite size, situated in a (predominately) horizontal wind of velocity
V = V e
j
. In the following, we shall sometimes refer to such an airfoil as a wing (although it actually represents a
pair of wings on a standard xed wing aircraft). Let us adopt the coordinate system shown in the gure, which is
such that the x-z plane is horizontal, the wind is incident predominately from the x-direction, and the y-axis points
vertically upward. The wing is assumed to lie in the x-z plane. Let b be the wingspan, and let c(z) and (z) be the
width and thickness, respectively, of the wing cross-section (parallel to the x-y plane). See Figure 8.9. Suppose
that the wing is symmetric about the median plane, z = 0, so that c(z) = c(z) and (z) = (z). It follows that
c(z > b/2) = (z > b/2) = 0: i.e., the wing extends from z = b/2 to z = b/2.
Suppose that air circulation is set up around the wing parallel to the x-y plane in such a manner as to produce an
upward lift. It follows that the average pressure on the lower surface of the wing must exceed that on its upper surface.
Consider Figure 8.9, which shows a back view of the airfoil shown in Figure 8.8. As we go from the median plane
(z = 0) to a wing tip, Y, whether along the upper or the lower surface of the wing, we must arrive at the same pressure
at Y. It follows that there is a drop in pressure as we move outward, away from the median plane, along the wings
bottom surface, and a further drop in pressure as we move inward, toward the median plane, along the upper surface.
Since air is pushed in the direction of decreasing pressure, it follows that the air that impinges on the wings leading
edge, and then passes over its upper surface, deviates sideways toward the median plane. Likewise, the air that passes
over the wings lower surface deviates sideways away from the median plane. See Figure 8.8.
Now, the air that leaves the trailing edge of the wing at some point Q must have impinged on the leading edge at
the dierent points P and P
, depending on whether it travelled over the wings upper or lower surfaces, respectively.
160 FLUID MECHANICS
V
airfoil
z = b/2 z = 0 z = b/2
c(z)
x
z
Figure 8.8: Top view of a three-dimensional airfoil of nite size.
(z)
z
y
b
Y Y
+
Figure 8.9: Back view of a three-dimensional airfoil of nite size, indication the pressure variation over its surface.
z
P
P
Q
Q
x
Figure 8.10: Top view of the airow over the top (left) and bottom (right) surfaces of a three-dimensional airfoil of
nite size.
Moreover, air that travels to Q via the wings upper surface acquires a small sideways velocity directed towards the
median plane, whereas that which travels to Q via the lower surface acquires a small sideways velocity directed away
from the median plane. On the other hand, the air speed at Q must be the same, irrespective of whether the air arrives
fromthe wings upper or lower surface, because the pressure (which, according to Bernoullis theorem, depends on the
air speed) must be continuous at Q. Thus, we conclude that there is a discontinuity in the direction of the air emitted
by the trailing edge of a wing. This implies that the interface, , between the two streams of air that travel over the
upper and lower surfaces of the wing is a vortex sheet. (See Section 8.5.) Of course, this vortex sheet constitutes the
wake that trails behind the airfoil. Moreover, we would generally expect the wake to be convected by the incident
wind. It follows that the vorticity per unit length in the wake can be written
= I(z) e
j
, (8.63)
where I(z) = v
z
, and v
z
is tangential velocity discontinuity across the wake. [See Equation (8.56).]
As we saw previously, the boundary layer that covers the airfoil is such that the tangential velocity U just outside
the layer is sharply reduced to zero at the airfoil surface. Actually, the nature of the substance enclosed by the surface
is irrelevant to our argument, and nothing is changed in our analysis if we suppose that this region contains air at rest.
Thus, we can replace the airfoil by air at rest, and the boundary layer by a vortex sheet, S , with a vortex intensity per
unit length
S
that is determined by the velocity discontinuity U between the air just outside the boundary layer and
that at rest in the region where the airfoil was previously located. In fact, Equation (8.56) yields
S
= n U, (8.64)
where n is an outward unit normal to the airfoil surface.
We conclude that a stationary airfoil situated in a uniform wind of constant velocity is equivalent to a vortex sheet
S , located at the airfoil surface, and a wake that trails behind the airfoil, the airfoil itself being replaced by air at rest.
The vorticity within S is largely parallel to the z-axis [since n and U are both essentially parallel to the x-y planesee
Equation (8.64)], whereas that in is parallel to the incident wind direction. See Figure 8.11. The vortex laments
within S are generally termed bound laments (since they cannot move o the airfoil surface). Conversely, the vortex
laments within are generally termed free laments. The air velocity both inside and outside S can be written
v = V + v
+ v
S
, (8.65)
where V is the external wind velocity, v
the velocity eld induced by the free vortex laments that constitute , and
v
S
the velocity eld induced by the bound laments that constitute S .
Consider some point P that lies on S . Let P
+
and P
be two neighboring points that are equidistant from P, where

P
+
lies just outside S , and P
lies just inside S , and the line P
P
+
is normal to S . We can write
v(P
+
) = V + v
(P
+
) + v
S
(P
+
), (8.66)
v(P
) = V + v
(P
) + v
S
(P
). (8.67)
162 FLUID MECHANICS
V
x
y
S
z
Figure 8.11: Vortex structure around a wing.
However, v(P
+
) = U(P), where U(P) is the tangential air velocity just above point P on the airfoil surface, and
v(P
) = 0, since the air within S is stationary. Moreover, v
(P
+
) = v
(P
) = v
(P), since we expect v
to be
continuous across S . On the other hand, we expect the tangential component of v
S
to be discontinuous across S . Let
us dene
v
S
(P) =
1
2
[v
S
(P
+
) + v
S
(P
)] . (8.68)
This quantity can be identied as the velocity induced at point P by the bound vortices on S , excluding the contribution
from the local bound vortex at P (since this vortex induces equal and opposite velocities at P
+
and P
). Finally, taking
half the sum of Equations (8.66) and (8.67), we obtain
1
2
U(P) = V + v
(P) + v
S
(P). (8.69)
8.8 Aerodynamic Forces
The net aerodynamic force acting on an three-dimensional airfoil of nite size can be written
A =
_
S
p ndS, (8.70)
where the integral is taken over the surface of the airfoil, S . Here, n is an outward unit normal vector on S , dS is
an element of S , and p is the air pressure. From Bernoullis theorem (in an irrotational uid), we can write p =
p
0
(1/2) v
2
, where p
0
is a constant pressure. Since a constant pressure exerts no net force on a closed surface, we
get
A =
1
2

_
S
U
2
ndS, (8.71)
where U is the tangential air velocity just above the surface of the airfoil. Now,
U (n U) = U
2
n (n U) U = U
2
n, (8.72)
since n U = 0 on the surface. Hence,
A =
1
2

_
S
U (n U) dS. (8.73)
Making use of Equations (8.64) and (8.69), the above expression can be written
A =
_
S
(V + v
+ v
S
)
S
dS = L + D + F, (8.74)
where
L = V
_
S
S
dS, (8.75)
D =
_
S
v

S
dS, (8.76)
F =
_
S
v
S

S
dS. (8.77)
Here, V, v
, and v
S
are the incident wind velocity, the velocity induced by the free vortices in the wake, and the
velocity induced by the bound vortices covering the surface of the airfoil, respectively. The forces L and D are called
the lift and the induced drag, respectively. (Note, that L now represents a net force, rather than a force per unit length.)
We shall presently demonstrate that the force F is negligible.
Let us assume that
S

z
e
z
: (8.78)
i.e., that the bound vortices covering the surface of the airfoil run parallel to the z-axis. This assumption is exactly
correct for an airfoil of innite wingspan and constant cross-section. Moreover, it is a good approximation for an
airfoil of nite wingspan, provided the airfoils length greatly exceeds its width (i.e., b c). Now, the incident wind
velocity is written V = V e
j
. Moreover, dS = dl dz, where dl is an element of length that runs parallel to the x-y plane
whilst lying on the airfoil surface. Now, making use of the curl theorem, we can easily show that
_
C
z
dl = (z), (8.79)
where the closed curve C is the intersection of the airfoil surface with the plane z = z, and (z) is the air circulation
about the airfoil in this plane. Thus, it follows from Equation (8.75) that
L = V
_
b/2
b/2
(z) dz e
. (8.80)
This expression is the generalization of Equation (8.31) for a three-dimensional airfoil of nite size. As before, the lift
is at right-angles to the incident wind direction.
Let us make the further assumptionknown as the lifting line approximation (because the lifting action of the
wing is eectively concentrated onto a line)that
v
= w(z) e
(8.81)
throughout S , where w(z) e
is the induced velocity due to the free vortices in , evaluated at the trailing edge of the
airfoil. Here, the velocity w(z) is called the downwash velocity. It follows from Equation (8.76) that
D =
_
b/2
b/2
w(z) (z) dz e
j
. (8.82)
Note that the induced drag is parallel to the incident wind direction. The origin of induced drag is as follows. It
takes energy to constantly resupply free vortices to the wake, as they are swept downstream by the wind (note that a
vortex lament possesses energy by virtue of the kinetic energy of its induced ow pattern), and this energy is supplied
by the work done in opposing the induced drag. The drag acting on a well-designed airfoil (i.e., an airfoil with an
aerodynamic shape that minimizes formdrag) situated in a high Reynolds number wind (which implies that the friction
drag is negligible) is generally dominated by induced drag.
According to Equations (8.62) and (8.77), the force F is written
F =
4
_
S
_
S
[
S
(r
) (r r
)]
S
(r)
r r
3
dS dS
. (8.83)
164 FLUID MECHANICS
z
l
O
z
P
x
r r
Figure 8.12: Semi-innite vortex lament.

We can interchange primed and unprimed variables without changing the value of the integral. Hence,
F =
4
_
S
_
S
[
S
(r) (r
r)]
S
(r
)
r
r
3
dS
dS. (8.84)
Taking the half the sum of the previous two equations, we obtain
F =
8
_
S
_
S
[
S
(r
) (r r
)]
S
(r) + [(r r
)
S
(r)]
S
(r
)
r r
3
dS dS
. (8.85)
However, (a b) c + (b c) a + (c a) b = 0. Thus, the above expression yields
F =
8
_
S
_
S
[
S
(r
)
S
(r)] (r r
)
r r
3
dS dS
. (8.86)
But, the assumption (8.78) implies that
S
(r
)
S
(r) 0. Hence, F is negligible, as was previously stated.
Consider a closed surface covering the small section of the airfoil lying between the parallel planes z = z and
z = z + dz. The ux of vorticity into the surface due to bound vortices at z is (z). The ux of vorticity out of the
surface due to bound vortices at z + dz is (z + dz). Finally, the ux of vorticity out of the surface due to the free
vortices in the part of the wake lying between z and z + dz is I(z) dz. However, the net ux of vorticity out of a closed
surface is zero, since vorticity is divergence free. Hence,
(z) = (z + dz) + I(z) dz, (8.87)
which implies that
I(z) =
d
dz
. (8.88)
Finally, consider a semi-innite straight vortex lament of vortex intensity = e
x
that terminates at the origin,
O, as shown in Figure 8.12. Let us calculate the ow velocity induced by this lament at the point P = (0, 0, z). From
the diagram l = z tan , dl = z sec
2
d, r r
= z sec , and r r
= z e
y
. Hence, from Equation (8.61), the
induced velocity at P is v = v
y
e
y
, where
v
y
=
4 z
_
/2
0
cos d =
4 z
. (8.89)
This result allows us to calculate the downwash velocity, w(z) = v
y
(z), induced at the trailing edge of the airfoil by
the semi-innite free vortices in the wake. The vortex intensity in the small section of the wake lying between z and
z + dz is I(z) dz, so we obtain
w(z) =
1
4
_
b/2
b/2
I(z
) dz
z z
=
1
4
_
d(z
)
z z
, (8.90)
where use has been made of (8.88).
8.9 Ellipsoidal Airfoils
Consider an ellipsoidal airfoil whose outer surface is specied by the parametric equations
x =
c
0
2
sin cos , (8.91)
y =
0
2
sin sin , (8.92)
z =
b
2
cos , (8.93)
where 0 and 0 2. Here, b is the wingspan, c
0
the maximum wing width, and
0
the maximum
wing thickness. Note that the wings cross-section is elliptical both in the x-y and the x-z planes. It is assumed that
b > c
0

0
: i.e., the wingspan is greater than the wing width, which in turn is much greater than the wing thickness.
At xed (i.e., xed z), the width and thickness of the airfoil are c() = c
0
sin and () =
0
sin , respectively.
Assuming that the two-dimensional result (8.44) holds at xed z, we deduce that the air circulation about the wing
satises
(z) = V c(z) sin =
0
sin , (8.94)
where
0
V c
0
. (8.95)
Here, the angle of attack, , is assumed to be small. From Equations (8.90) and (8.94), the downwash velocity in the
region z < b/2 is given by
w() =
0
2 b
_

0
cos
cos
cos
=
0
2 b
_
1 +
cos
_

0
d
cos
cos
_
. (8.96)
Now, the integrand in the integral
_

0
d
cos
cos
(8.97)
is singular when
= . However, we can still obtain a nite value for the integral by taking its principal part: i.e.,
lim
0
__

0
d
cos
cos
+
_

+
d
cos
cos
_
. (8.98)
Physically, this is equivalent to omitting the contribution of the local free vortex at a given point on the airfoils trailing
edge to the downwash velocity induced at that point, which is reasonable because a vortex induces zero velocity at its
center. Hence, we obtain
_

0
d
cos
cos
= lim
0
_
_
_
1
sin
ln
_
sin (1/2) ( +
)
sin (1/2) (
)
__
0
+
_
1
sin
ln
_
sin (1/2) (
+ )
sin (1/2) (
)
__
+
_
= lim
0
_
1
sin
ln
_
sin( /2)
sin( + /2)
__
= 0, (8.99)
which implies that
w() =
0
2 b
. (8.100)
In the region z > b/2, we can write = 2 z/b, so that
w() =
0
2 b
_
1

_

0
d
cos
+
_
=
0
2 b
_
_
1

_
2
1
_
_
. (8.101)
166 FLUID MECHANICS
2
1.6
1.2
0.8
0.4
0
0.4
0.8
1.2
w
/
(
0
/
2
b
)
2 1 0 1 2
z/b)
Figure 8.13: Downwash velocity prole induced at the trailing edge by an ellipsoidal airfoil.
Hence, we conclude that the downwash velocity prole induced by an ellipsoidal airfoil takes the form
w(z) =
0
2 b
_
1 z < b/2
1 z/(z
2
b
2
/4)
1/2
z > b/2
. (8.102)
This prole is shown in Figure 8.13. It can be seen that the downwash velocity is uniform and positive in the region
between the wingtips (i.e., b/a < z < b/2), but negative and decaying in the region outside the wingtips. Hence, we
conclude that as air passes over an airfoil subject to an upward lift it acquires a net downward velocity component,
which, of course, is a consequence of the reaction to the lift. On the other hand, the air immediately behind and to the
sides of the airfoil acquires a net upward velocity component. In other words, the lift acting on the airfoil is associated
with a downwash of air behind, and an upwash behind and to either side of, the airfoil. The existence of upwash
slightly behind and to the side of a ying object allows us to explain the V-formation adopted by wild geesea bird
ying in the upwash of another bird needs to generate less lift in order to stay in the air, and, consequently, experiences
less induced drag.
It follows from Equation (8.93) and (8.94) that
_
b/2
b/2
(z) dz =
0
b
2
_

0
sin
2
d =
4

0
b. (8.103)
Hence, Equation (8.80), (8.82), and (8.100) yield the following expression for the lift and induced drag acting on an
ellipsoidal airfoil,
L =
4
V b
0
, (8.104)
D =
8

2
0
. (8.105)
Now, the surface area of the airfoil in the x-z plane is
S =
4
b c
0
. (8.106)
V
T
W
airfoil
x
y
L
D
Figure 8.14: Side view of a xed wing aircraft in ight.
Moreover, the airfoils aspect-ratio is conventionally dened as the length to width ratio for a rectangle of length b
that has the same area as the airfoil: i.e.,
A =
b
2
S
=
4
b
c
0
. (8.107)
It thus follows from Equation (8.95) that
L = L
0
, (8.108)
D =
2
A
L
0

2
, (8.109)
where
L
0
= V
2
S. (8.110)
8.10 Simple Flight Problems
Figure 8.14 shows a side-view schematic of a xed wing aircraft ying in a straight-line at constant speed through
stationary air. Here, x is a horizontal coordinate, and y a vertical coordinate. The center of mass of the aircraft
is assumed to be moving with some xed velocity V that subtends an angle with the horizontal. Thus, the wind
velocity in the aircrafts rest frame is V. Let the aircrafts wings, which are assumed to be parallel to its fuselage,
be inclined at an angle + to the horizontal. It follows that is the angle of attack. The aircraft is subject to four
forces: the thrust, T, developed by its engine, which is assumed to act parallel to its fuselage; the lift L, which acts
at right-angles to V; the induced drag, D, which acts in the opposite direction to V; and the weight, W, which acts
vertically downward.
Vertical force balance yields
T sin( + ) + L cos = W + D sin , (8.111)
whereas horizontal force balance gives
T cos( + ) = D cos + L sin . (8.112)
168 FLUID MECHANICS
Let us assume that the angles and are both small. According to Equations (8.108) and (8.109), L L
0
and
D (2 L
0
/A)
2
. Thus, L O() and D O(
2
). Moreover, it is clear from (8.111) and (8.112) that T O(
2
) and
W O(). Thus, to lowest order in , Equation (8.111) yields

W
L
0
, (8.113)
whereas Equation (8.112) gives

T
W

2
A
W
L
0
. (8.114)
Expression (8.113) relates the angle of attack to the ratio of the aircrafts weight to its (theoretical) maximum lift (at
a given airspeed). Expression (8.114) relates the aircrafts angle of controlled (i.e., at constant airspeed) ascent to the
thrust developed by its engine. Now an unpowered aircraft, such as a glider, has zero thrust. For such an aircraft,
(8.114) reveals that the angle of controlled decentwhich is usually termed the glide angletakes the value
g =
2
A
W
L
0
. (8.115)
At xed airspeed, V, and wing surface area, S , (which implies that L
0
is xed) this angle can be minimized by making
the wing aspect-ratio, A, as large as possible. This result explains accounts for the fact that gliders (and albatrosses)
have long thin wings, rather than short stubby ones. For a powered aircraft, the critical thrust to weight ratio required
to maintain level ight (i.e., = 0) is
T
W
= g. (8.116)
Hence, this ratio is minimized by minimizing the glide angle, which explains why long-haul aircraft, which generally
need to minimize fuel consumption, tend to have long thin wings. Finally, as we saw in Section 8.4, if the angle of
attack exceeds some (generally small) critical value
c
then boundary layer separation occurs on the back sides of
the wings, giving rise to a greatly increased level of drag acting on the aircraft. In aerodynamics, this phenomenon is
called a stall. As is clear from Equations (8.110) and (8.113), the requirement <
c
is equivalent to
V > V
s
=
_
W
S
c
_
1/2
. (8.117)
In other words, a stall can be avoided by keeping the airspeed above the critical value V
s
, which is known as the stall
speed. Note that the stall speed decreases with decreasing altitude, as the air becomes denser.
8.11 Exercises
8.1. Consider the integral
I
n
() =
_

0
cos(n
) d
cos
cos
,
where n is a non-negative integer. This integral is dened by its principal value
I
n
() = lim
0
__

0
cos(n
) d
cos
cos
+
_

+
cos(n
) d
cos
cos
_
.
As was demonstrated in Section 8.9,
I
0
= 0.
Show that
I
1
= ,
and
I
n+1
+ I
n1
= 2 cos I
n
,
and hence that
I
n
=
sin(n )
sin
.
8.2. Suppose that an airfoil of negligible thickness, and wingspan b, has a width whose z variation is expressed parametrically as
c() =
=1,3,5,
c
sin( ),
for 0 , where
z =
b
2
cos .
Show that the air circulation about the airfoil takes the form
() =
=1,3,5,
sin( ),
where
= V c
. Here, is the angle of attack (which is assumed to be small). Demonstrate that the downwash velocity
at the trailing edge of the airfoil is
w() =
=1,3,5,

2 b
sin( )
sin
.
Hence, show that the lift and induced drag acting on the airfoil take the values
L =
4
V b
1
,
D =
=1,3,5,

2
,
respectively. Demonstrate that the drag to lift ratio can be written
D
L
=
2
A

_
_
1 +
=3,5,7,
c
2
c
2
1
_
_
,
where A is the aspect ratio. Hence, deduce that the airfoil shape (in the x-y) plane that minimizes this ratio (at xed aspect
ratio) is an ellipse (i.e., such that c
= 0 for > 1).

8.3. Consider a plane that ies with a constant angle of attack, and whose thrust is adjusted such that it cancels the induced drag.
The plane is eectively subject to two forces. First, its weight, W = W e
y
, and second its lift L = k v v
y
e
x
+ k v v
x
e
y
.
Here, x and y are horizontal and vertical coordinates, respectively, v is the planes instantaneous velocity, and k is a positive
constant. Note that the lift is directed at right angles to the planes instantaneous direction of motion, and has a magnitude
proportional to the square of its airspeed. Demonstrate that the planes equations of motion can be written
dv
x
dt
=
v v
y
h
,
dv
y
dt
=
v v
x
h
g,
where h = k g/W is a positive constant with the dimensions of length. Show that
1
2
v
2
+ g y = c,
where c is a constant. Suppose that v
x
=
_
g h (1 + u) and v
y
=
_
g h w, where u, w 1. Demonstrate that, to rst order in
perturbed quantities,
du
dt

_
g
h
w,
dw
dt
2
_
g
h
u.
Hence, deduce that if the plane is ying horizontally at some speed v
0
, and is subject to a small perturbation, then its altitude
oscillates sinusoidally at the angular frequency =
2 g/v
0
. This type of oscillation is known as a phugoid oscillation.
170 FLUID MECHANICS
Incompressible Viscous Flow 171
9 Incompressible Viscous Flow
9.1 Introduction
This chapter investigates incompressible ow in which viscosity plays a signicant role throughout the bulk of the
uid. Such ow generally takes place at relatively low Reynolds number. From Section 2.14, the equations governing
incompressible viscous uid motion can be written
v = 0, (9.1)
Dv
Dt
= P +
2
v, (9.2)
where the quantity
P = p + , (9.3)
which is a combination of the actual uid pressure, p, and the gravitational potential energy per unit volume, , is
known as the eective pressure. Here, is the uid density, the uid viscosity, and the gravitational potential.
9.2 Flow Between Parallel Plates
Consider steady, two-dimensional, viscous ow between two parallel plates that are situated a perpendicular distance
d apart. Let x be a longitudinal coordinate measuring distance along the plates, and let y be a transverse coordinate
such that the plates are located at y = 0 and y = d. See Figure 9.1.
Suppose that there is a uniform eective pressure gradient in the x-direction, so that
dP
dx
= G, (9.4)
where G is a constant. Here, the quantity G could represent a gradient in actual uid pressure, a gradient in gravitational
potential energy (due to an inclination of the plates to the horizontal), or some combination of the twoit actually
makes no dierence to the nal result. Suppose that the uid velocity prole between the plates takes the form
v = v
x
(y) e
x
. (9.5)
From Section 2.18, this prole automatically satises the incompressibility constraint v = 0, and is also such that
Dv/Dt = 0. Hence, Equation (9.2) reduces to
2
v =
P
, (9.6)
v
x
(y)
y = d
y = 0
P
x
y
Figure 9.1: Viscous ow between parallel plates.
172 FLUID MECHANICS
x
h
y
Figure 9.2: Viscous ow down an inclined plane.
or. taking the x-component,
d
2
v
x
dy
2
=
G
. (9.7)
If the two plates are stationary then the solution that satises the no slip constraint, v
x
(0) = v
x
(d) = 0, at each plate is
v
x
(y) =
G
2
y (d y). (9.8)
Thus, steady, two-dimensional, viscous ow between two stationary parallel plates is associated with a parabolic
velocity prole that is symmetric about the midplane, y = d/2. The net volume ux (per unit width in the z-direction)
of uid between the plates is
Q =
_
d
0
v
x
dy =
Gd
3
12
. (9.9)
Note that this ux is directly proportional to the eective pressure gradient, inversely proportional to the uid viscosity,
and increases as the cube of the distance between the plates.
Suppose that the upper plate is stationary, but that the lower plate is moving in the x-direction at the constant speed
U. In this case, the no slip boundary condition at the lower plate becomes v
x
(0) = U, and the modied solution to
Equation (9.7) is
v
x
(y) =
G
2
y (d y) + U
_
d y
d
_
. (9.10)
Hence, the modied velocity prole is a combination of parabolic and linear proles. This type of ow is known as
Couette ow. The net volume ux (per unit width) of uid between the plates becomes
Q =
Gd
3
12
+
1
2
U d. (9.11)
9.3 Flow Down an Inclined Plane
Consider steady, two-dimensional, viscous ow down a plane that is inclined at an angle to the horizontal. Let x
measure distance along the plane, and let y be a transverse coordinate such that the surface of the plane corresponds
to y = 0. Suppose that the uid forms a uniform layer of depth h covering this surface. See Figure 9.2.
The generalized pressure gradient within the uid is written
dP
dx
= G = g sin , (9.12)
where g is the acceleration due to gravity. In this case, there is no gradient in the actual pressure in the x-direction,
and the ow down the plane is driven entirely by gravity. As before, we can write
v = v
x
(y) e
x
, (9.13)
and Equation (9.2) again reduces to
d
2
v
x
dy
2
=
G
, (9.14)
where
G = g sin . (9.15)
Application of the no slip condition at the surface of the plane, y = 0, yields the standard boundary condition v
x
(0) = 0.
However, the appropriate physical constraint at the uid/air interface, y = h, is that the normal viscous stress there
be zero (i.e.,
xy
y=h
= 0), since there is nothing above the interface that can exchange momentum with the uid
(assuming that the nite inertia and viscosity of air are both negligible.) Hence, fromSection 2.18, we get the boundary
condition
dv
x
dy
y=h
= 0. (9.16)
The solution to Equation (9.14) that satises the boundary conditions is
v
x
(y) =
G
2
y (2 h y). (9.17)
Thus, the prole is again parabolic. In fact, it is the same as the lower half of the prole obtained when uid ows
between two (stationary) parallel plates situated a perpendicular distance 2 h apart.
The net volume ux (per unit width in the z-direction) of uid down the plane is
Q =
_
h
0
v
x
dy =
Gh
3
3
=
g sin h
3
3
, (9.18)
where use has been made of Equation (9.15). Here, = / is the kinematic viscosity of the uid. Thus, given the
rate Q that uid is poured down the plane, the depth of the layer covering the plane becomes
h =
_
3 Q
g sin
_
1/3
. (9.19)
Suppose that the rate at which uid is poured down the plane is suddenly increased slightly from Q to Q+Q. We
would expect an associated change in depth of the layer covering the plane from h to h + h, where
h =
_
dh
dQ
_
Q. (9.20)
Let the interface between the layers of dierent depth propagate in the x-direction at the constant velocity V. In a
frame of reference that co-moves with this interface, the volume uxes (per unit width) immediately to the right and
to the left of the interface are Q V h and Q + Q V (h + h), respectively. However, in a steady state, these uxes
must equal one another. Hence,
Q = V h, (9.21)
or
V =
dQ
dh
=
Gh
2
=
_
9 Q
2
g sin
_
1/3
. (9.22)
As can easily be veried, this velocity is twice the maximum uid velocity in the layer.
174 FLUID MECHANICS
9.4 Poiseuille Flow
Steady viscous uid ow driven by an eective pressure gradient established between the two ends of a long straight
pipe of uniformcircular cross-section is generally known as Poiseuille ow, since it was rst studied experimentally by
J. L. M. Poiseuille in 1838. Suppose that the pipe is of radius a. Let us adopt cylindrical coordinates whose symmetry
axis coincides with that of the pipe. Thus, z measures distance along the pipe, r = 0 corresponds to the center of the
pipe, and r = a corresponds to the pipe wall. Suppose that
P = Ge
z
(9.23)
is the uniform eective pressure gradient along the pipe, and
v = v
z
(r) e
z
(9.24)
the time independent velocity prole driven by this gradient. It follows fromSection 2.19 that v = 0 and Dv/Dt = 0.
Hence, (9.2) reduces to
2
v =
G
e
z
. (9.25)
Taking the z-component of this equation, we obtain
1
r
d
dr
_
r
dv
z
dr
_
=
G
, (9.26)
where use has been made of (2.155). The most general solution of the above equation is
v
z
(r) =
G
4
r
2
+ A ln r + B, (9.27)
where A and B are arbitrary constants. The physical constraints are that the ow velocity is non-singular at the center
of the pipe (which implies that A = 0), and is zero at the edge of the pipe [i.e., v
z
(a) = 0], in accordance with the no
slip condition. Thus, we obtain
v
z
(r) =
G
4
(a
2
r
2
). (9.28)
The volume ux of uid down the pipe is
Q =
_
a
0
2 r v
z
dr =
Ga
4
8
. (9.29)
According to the above analysis, the quantity Q/a
4
should be directly proportional to the eective pressure gradient
along the pipe. The accuracy with which experimental observations show that this is indeed the case (at relatively low
Reynolds number) is strong evidence in favor of the assumptions that there is no slip at the pipe walls, and that the
ow is non-turbulent. In fact, the result (9.29), which is known as Poiseuilles law, is valid experimentally provided
the Reynolds number of the ow, Re = U a/, remains less than about 6.5 10
3
. Here, U = Q/ a
2
is the mean
ow speed. On the other hand, if the Reynolds number exceeds the critical value 6.5 10
3
then the ow in the pipe
becomes turbulent, and Poiseuilles law breaks down.
9.5 Taylor-Couette Flow
Consider two thin cylindrical shells with the same vertical axis. Let the inner and outer shells be of radius r
1
and r
2
,
respectively. Suppose that the annular region r
1
r r
2
is lled with uid of density and viscosity . Let the inner
and outer cylinders rotate at the constant angular velocities
1
and
2
, respectively. We wish to determine the steady
ow pattern set up within the uid. Incidentally, this type of ow is generally known as Taylor-Couette ow.
It is convenient to adopt cylindrical coordinates, r, , z, whose symmetry axis coincides with the common axis of
the two shells. Thus, the inner and outer shells correspond to r = r
1
and r = r
2
, respectively. Suppose that the ow
velocity within the uid is written
v = v
(r) e
= r (r) e
, (9.30)
where (r) = v
(r)/r is the angular velocity prole. Application of the no slip condition at the two shells leads to the
boundary conditions
(r
1
) =
1
, (9.31)
(r
2
) =
2
. (9.32)
It again follows from Section 2.19 that v = 0 and Dv/Dt = 0. Hence, (9.2) reduces to
2
v =
P
. (9.33)
Assuming that P = 0 within the uid, since any ow is driven by the angular rotation of the two shells, rather than by
pressure gradients or gravity, and again making use of the results quoted in Section 2.19, the above expression yields
1
r
d
dr
_
r
v
dr
_
r
2
= 0, (9.34)
or
1
r
2
d
dr
_
r
3
d
dr
_
= 0. (9.35)
The solution of (9.35) that satises the boundary conditions is
(r) =
1
r
2
_
1

2
r
2
1
r
2
2
_
_
+
_
2
r
2
1

1
r
2
2
r
2
1
r
2
2
_
_
. (9.36)
Note that this angular velocity prole is a combination of the solid body rotation prole = constant, and the
irrotational rotation prole r
2
.
From Section 2.19, the only non-zero component of the viscous stress tensor within the uid is
r
= r
d
dr
_
v
r
_
= r
d
dr
. (9.37)
Thus, the viscous torque (acting in the -direction) per unit height (in the z-direction) exerted on the inner cylinder is
1
= 2 r
2
1

r
(r
1
) = 4
_
1

2
r
2
1
r
2
2
_
_
. (9.38)
Likewise, the torque per unit height exerted on the outer cylinder is
2
= 2 r
2
2

r
(r
2
) = 4
_
1

2
r
2
1
r
2
2
_
_
. (9.39)
As expected, these two torques are equal and opposite, and act to make the two cylinders rotate at the same angular
velocity (in which case, the uid between them rotates as a solid body).
9.6 Flow in Slowly-Varying Channels
According to Section 9.1, the equations governing steady, incompressible, viscous uid ow are
v = 0, (9.40)
(v ) v = P +
2
v. (9.41)
176 FLUID MECHANICS
As we saw in Sections 9.2 and 9.4, for the case of ow along a straight channel of uniform cross-section, v and
(v ) v are both identically zero, and the governing equations consequently reduce to the simple relation
2
v =
P
. (9.42)
Suppose, however, that the cross-section of the channel varies along its length. As we shall demonstrate, provided this
variation is suciently slow, the ow is still approximately described by the above relation.
Consider steady, two-dimensional, viscous ow, that is predominately in the x-direction, between two plates that
are predominately parallel to the y-z plane. Let the spacing between the plates, d(x), vary on some length scale l d.
Suppose that
P = G(x) e
x
, (9.43)
where G(x) also varies on the same length scale. Assuming that /x O(1/l) and /y O(1/d), it follows from
Equation (9.40) that
v
y
v
x
O
_
d
l
_
. (9.44)
Hence,
[(v ) v]
x
O
_
v
2
x
l
_
, (9.45)
(
2
v)
x
=
2
v
x
y
2
_
_
1 + O
_
d
l
_
2
_
_
. (9.46)
The x-component of Equation (9.41) reduces to
2
v
x
y
2
_
_
1 + O
_
v
x
d
2
l
_
+ O
_
d
l
_
2
_
_
=
G
. (9.47)
Thus, if
d
l
1, (9.48)
v
x
d
2
l
1 (9.49)
i.e., if the channel is suciently narrow, and its cross-section varies suciently slowly along its lengththen
Equation (9.47) can be approximated as
2
v
x
y
2

G
. (9.50)
This, of course, is the same as the equation governing steady, two-dimensional, viscous ow between exactly parallel
plates. See Section 9.2. Assuming that the plates are located at y = 0 and y = d(x), and making use of the analysis of
Section 9.2, the appropriate solution to the above equation is
v
x
(x, y) =
G(x)
2
y [d(x) y]. (9.51)
The volume ux (per unit width) of uid between the plates is thus
Q =
_
d
0
v
x
dy =
G(x) d
3
(x)
12
. (9.52)
However, for steady incompressible ow, this ux must be independent of x, which implies that
G(x) =
dP
dx
= 12 Qd
3
(x). (9.53)
Suppose that a constant dierence in eective pressure, P, is established between the xed points x = x
1
and x = x
2
,
where x
2
x
1
= l. Integration of the above equation between these two points yields
P = 12 Ql (d
3
), (9.54)
where ( ) =
_
x
2
x
1
( ) dx/l. Hence, the volume ux (per unit width) of uid between the plates that is driven by the
eective pressure dierence becomes
Q =
P
l
1
12 (d
3
)
. (9.55)
Moreover, the eective pressure gradient at a given point is
G(x) =
dP
dx
=
P
l
1
d
3
(x) (d
3
)
, (9.56)
which allows us to determine the velocity prole at that point from Equation (9.51). Thus, given the average eective
pressure gradient, P/l, and the variable separation, d(x), we can fully specify the ow between the plates.
Using analogous arguments to those employed above, but adapting the analysis of Section 9.4, rather than that of
Section 9.1, we can easily show that steady viscous ow down a straight pipe of circular cross-section, whose radius
a varies slowly with distance, z, along the pipe, is characterized by
v
z
(r, z) =
G(z)
4
_
a
2
(z) r
2
_
, (9.57)
G(z) =
P
l
1
a
4
(z) (a
4
)
, (9.58)
Q =
P
l
8 (a
4
)
. (9.59)
Here, Q is the volume ux of uid down the pipe, P = P(z
2
) P(z
1
), l = z
2
z
1
, and ( ) =
_
z
2
z
1
( ) dz/l. The
approximations used to derive the above results are valid provided
a
l
1, (9.60)
v
z
a
2
l
1. (9.61)
9.7 Lubrication Theory
It is well-known that two solid bodies can slide over one another particularly easily when there is a thin layer of
uid sandwiched between them. Moreover, under certain circumstances, a large positive pressure develops within the
layerthis phenomenon is exploited in hydraulic bearings, whose aim is to substitute uid-solid friction for the much
larger friction that acts between solid bodies that are in direct contact with one another. Once set up, the uid layer in
hydraulic bearings oers great resistance to being squeezed out, and is often capable of supporting a useful load.
Consider the simple two-dimensional case of a solid body with a plane surface (that is almost parallel to the x-z
plane) gliding steadily over another such body, the surface of the gliding body being of nite length l in the direction
of the motion (the x-direction), and of innite width (in the z-direction). See Figure 9.3. Experience shows that the
plane surfaces need to be slightly inclined to one another. Suppose that 1 is the angle of inclination. Let us
transform to a frame of reference in which the upper body is stationary. In this frame, the lower body moves in the
x-direction at some xed speed U. Suppose that the upper body extends from x = 0 to x = l, and that the surface of
the lower body corresponds to y = 0. Let d(x) be the thickness (in the y-direction) of the uid layer trapped between
the bodies, where d(0) = d
1
and d(l) = d
2
. It follows that
d(x) = d
1
x, (9.62)
178 FLUID MECHANICS
U
d
1
d
d
2
l
x
y

Figure 9.3: The lubrication layer between two planes in relative motion.
where
=
d
1
d
2
l
. (9.63)
As discussed in the previous section, provided that
d
l
1, (9.64)
U d
2
l
1, (9.65)
the cross-section of the channel between the two bodies is suciently slowly varying in the x-direction that the channel
can be treated as eectively uniform at each point along its length. Thus, it follows from Equation (9.10) that the
velocity prole within the channel takes the form
v
x
(x, y) =
G(x)
2
y
_
d(x) y
_
+ U
_
d(x) y
d
_
, (9.66)
where
G(x) =
dp
dx
(9.67)
is the pressure gradient. Here, we are neglecting gravitational forces with respect to both pressure and viscous forces.
The volume ux per unit width (in the z-direction) of uid along the channel is thus
Q =
_
d
0
v
x
dy =
G(x) d
3
(x)
12
+
1
2
U d(x). (9.68)
Of course, in a steady state, this ux must be independent of x. Hence,
dp
dx
= G(x) = 6
_
U
d
2
(x)

2 Q
d
3
(x)
_
, (9.69)
where d(x) = d
1
x. Integration of the above equation yields
p(x) p
0
=
6
_
U
_
1
d

1
d
1
_
Q
_
_
1
d
2

1
d
2
1
_
_
_
_
, (9.70)
where p
0
= p(0). Now, assuming that the sliding block is completely immersed in uid of uniform ambient pressure
p
0
, we would expect the pressures at the two ends of the lubricating layer to both equal p
0
, which implies that p(l) = p
0
.
It follows from the above equation that
Q = U
_
d
1
d
2
d
1
+ d
2
_
, (9.71)
and
p(x) p
0
=
6 U
[d
1
d(x)] [d(x) d
2
]
d
2
(x) (d
1
+ d
2
)
. (9.72)
Note that if d
1
> d
2
then the pressure increment p(x) p
0
is positive throughout the layer, and vice versa. In other
words, a lubricating layer sandwiched between two solid bodies in relative motion only generates a positive pressure,
that is capable of supporting a normal load, when the motion is such as to drag (by means of viscous stresses) uid
from the wider to the narrower end of the layer. The pressure increment has a single maximum in the layer, and its
value at this point is of order l U/d
2
, assuming that (d
1
d
2
)/d
1
is of order unity. This suggests that very large
pressures can be set up inside a thin lubricating layer.
The net normal force (per unit width in the z-direction) acting on the lower plane is
f
y
=
_
l
0
[p(x) p
1
] dx =
6 U
2
_
ln
_
d
1
d
2
_
2
_
d
1
d
2
d
1
+ d
2
__
. (9.73)
Moreover, the net tangential force (per unit width) acting on the lower plane is
f
x
=
_
l
0
_
v
x
y
_
y=0
dx =
2 U
_
2 ln
_
d
1
d
2
_
3
_
d
1
d
2
d
1
+ d
2
__
. (9.74)
Of course, equal and opposite forces,
f
x
= f
x
=
U l
2
d
2
0
3
2 k
2
_
ln
_
1 + k
1 k
_
2 k
_
, (9.75)
f
y
= f
y
=
U l
d
0
1
k
_
2 ln
_
1 + k
1 k
_
3 k
_
, (9.76)
act on the upper plane. Here, d
0
= (d
1
+ d
2
)/2 is the mean channel width, and k = (d
1
d
2
)/(d
1
+ d
2
). Note that if
0 < d
2
< d
1
then 0 < k < 1. The eective coecient of friction, C
f
, between the two sliding bodies is conventionally
dened as the ratio of the tangential to the normal force that they exert on one another. Hence,
C
f
=
f
x
f
y
=
f
x
f
y
=
4
3
d
0
l
H(k), (9.77)
where
H(k) = k
_
ln
_
1 + k
1 k
_
3 k
2
__ _
ln
_
1 + k
1 k
_
2 k
_
. (9.78)
The function H(k) is a monotonically decreasing function of k in the range 0 < k < 1. In fact, H(k 0) 3/(4 k),
whereas H(k 1) 1. Thus, if k O(1) [i.e., if (d
1
d
2
)/d
1
O(1)] then C
f
O(d
0
/l) 1. In other words, the
eective coecient of friction between two solid bodies in relative motion that are separated by a thin uid layer is
independent of the uid viscosity, and much less than unity. This result is signicant because the coecient of friction
between two solid bodies in relative motion that are in direct contact with one another is typical of order unity. Hence,
the presence of a thin lubricating layer does indeed lead to a large reduction in the frictional drag acting between the
bodies.
9.8 Stokes Flow
Steady ow in which the viscous force density in the uid greatly exceeds the advective inertia per unit volume is
generally known as Stokes ow. Since, by denition, the Reynolds number of a uid is the typical ratio of the advective
inertia per unit volume to the viscous force density (see Section 2.16), Stokes ow implies Reynolds numbers that are
much less than unity. Now, in the time independent, low Reynolds number limit, Equations (9.1) and (9.2) reduce to
v = 0, (9.79)
0 = P +
2
v. (9.80)
180 FLUID MECHANICS
It follows from these equations that
P =
2
v = ( v) = , (9.81)
where = v, and use has been made of Equation (A.177). Taking the curl of this expression, we obtain
2
= 0, (9.82)
which is the governing equation for Stokes ow. Here, use has been made of Equations (A.173), (A.176), and (A.177).
9.9 Axisymmetric Stokes Flow
Let r, , be standard spherical coordinates. Consider axisymmetric Stokes ow such that
v(r) = v
r
(r, ) e
r
+ v
(r, ) e
. (9.83)
According to Equations (A.175) and (A.176), we can automatically satisfy the incompressibility constraint (9.79) by
writing
v = , (9.84)
where (r, ) is a stream function (i.e., v = 0). It follows that
v
r
(r, ) =
1
r
2
sin
, (9.85)
v
(r, ) =
1
r sin
r
. (9.86)
Moreover, according to Section C.4,
r
=
= 0, and
(r, ) =
1
r
(r v
)
r

1
r
v
r
=
[()
r sin
, (9.87)
where
[ =
2
r
2
+
sin
r
2
1
sin
. (9.88)
Hence, given that = 1/(r sin ), we can write
= v = [() . (9.89)
It follows from Equations (A.176) and (A.178) that
= [[()]. (9.90)
Hence, by analogy with Equations (9.84) and (9.89), and making use of (A.173) and (A.177), we obtain
( ) =
2
= [
2
() . (9.91)
Equation (9.82) implies that
[
2
() = 0, (9.92)
which is the governing equation for axisymmetric Stokes ow. In addition, Equations (9.81) and (9.90) yield
P = [[()]. (9.93)
9.10 Axisymmetric Stokes Flow Around a Solid Sphere
Consider a solid sphere of radius a that is moving under gravity at the constant vertical velocity V e
z
through a station-
ary uid of density and viscosity . Here, gravitational acceleration is assumed to take the form g = g e
z
. Now,
provided the typical Reynolds number,
Re =
2 V a
, (9.94)
is much less than unity, the ow around the sphere is an example of axisymmetric Stokes ow. Let us transform to
a frame of reference in which the sphere is stationary, and centered at the origin. Adopting the standard spherical
coordinates r, , , the surface of the sphere corresponds to r = a, and the surrounding uid occupies the region r > a.
By symmetry, the ow eld outside the sphere is axisymmetric (i.e., / = 0), and has no toroidal component (i.e.,
v
= 0). The physical boundary conditions at the surface of the sphere are
v
r
(a, ) = 0, (9.95)
v
(a, ) = 0 : (9.96)
i.e., the normal and tangential uid velocities are both zero at the surface. A long way from the sphere, we expect the
uid velocity to asymptote to v = V e
z
. In other words,
v
r
(r , ) V cos , (9.97)
v
(r , ) V sin . (9.98)
Let us write
v = , (9.99)
where (r, ) is the stream function. As we saw in the previous section, axisymmetric Stokes ow is characterized by
[
2
() = 0. (9.100)
Here, the dierential operator [ is specied in Equation (9.88). The boundary conditions (9.95)(9.98) reduce to
r=a
= 0, (9.101)
r=a
= 0, (9.102)
(r , )
1
2
V r
2
sin
2
. (9.103)
Equation (9.103) suggests that (r, ) can be written in the separable form
(r, ) = sin
2
f (r). (9.104)
In this case,
v
r
(r, ) =
2 cos f (r)
r
2
, (9.105)
v
(r, ) =
sin
r
d f
dr
, (9.106)
and Equations (9.100)(9.103) reduce to
_
d
2
dr
2

2
r
2
_
2
f = 0, (9.107)
f (a) =
d f
dr
r=a
= 0, (9.108)
f (r )
1
2
V r
2
. (9.109)
182 FLUID MECHANICS
3
2
1
0
1
2
3
z/a
3 2 1 0 1 2 3
x/a
Figure 9.4: Contours of the stream function in the x-z plane for Stokes ow around a solid sphere.
Let us try a test solution to Equation (9.107) of the form f (r) = r
n
. We nd that
[n (n 1) 2] [(n 2) (n 3) 2] = 0, (9.110)
which implies that n = 1, 1, 2, 4. Hence, the most general solution to Equation (9.107) is
f (r) =
A
r
+ Br + C r
2
+ Dr
4
, (9.111)
where A, B, C, D are arbitrary constants. However, the boundary condition (9.109) yields C = (1/2) V and D = 0,
whereas the boundary condition (9.108) gives A = (1/4) V a
3
and B = (3/4) V a. Thus, we conclude that
f (r) =
V (r a)
2
(2 r + a)
4 r
, (9.112)
and the stream function becomes
(r, ) = sin
2
V (r a)
2
(2 r + a)
4 r
. (9.113)
See Figure 9.4. From (9.87), the uid vorticity is
(r, ) =
[()
r sin
=
sin
r
_
d
2
dr
2

2
r
2
_
f =
3 V a sin
2 r
2
. (9.114)
See Figure 9.5. Moreover, from (9.81),
P = = (
r sin ). (9.115)
Hence,
P
r
=
3 V a cos
r
3
, (9.116)
P
=
3 V a sin
2 r
2
, (9.117)
3
2
1
0
1
2
3
z/a
3 2 1 0 1 2 3
x/a
Figure 9.5: Contours of the vorticity,
, in the x-z plane for Stokes ow around a solid sphere. Solid/dashed lines
correspond to opposite signs of
.
which implies that the eective pressure distribution within the uid is
P(r, ) = p
0
+
3 V a cos
2 r
2
, (9.118)
where p
0
is an arbitrary constant. See Figure 9.6. However, P = p + , where = g z = g r cos . Thus, the actual
pressure distribution is
p(r, ) = p
0
g r cos +
3 V a
2 r
2
cos . (9.119)
From Section 2.20, the radial and tangential components of the force per unit area exerted on the sphere by the
uid are
f
r
() =
rr
(a, ) =
_
p + 2
v
r
r
_
r=a
, (9.120)
f
() =
r
(a, ) =
_
1
r
v
r
+
v
r

v
r
_
r=a
. (9.121)
Now, v
r
(a, ) = v
(a, ) = 0. Moreover, since v = 0, it follows from (2.169) that (v

r
/r)
r=a
= 0. Finally,
Equation (9.87) yields (v
/r)
r=a
=
(a, ). Hence,
f
r
() = p(a, ) = p
0
+ g a cos
3 V
2 a
cos , (9.122)
f
() =
(a, ) =
3 V
2 a
sin . (9.123)
Thus, the force density at the surface of the sphere is
f() =
3 V
2 a
e
z
+ (p
0
+ g a cos ) e
r
. (9.124)
184 FLUID MECHANICS
3
2
1
0
1
2
3
z/a
3 2 1 0 1 2 3
x/a
Figure 9.6: Contours of the eective pressure, P p
0
, in the x-z plane for Stokes ow around a solid sphere.
Solid/dashed lines correspond to opposite signs of P p
0
.
It follows that the net vertical force exerted on the sphere by the uid is
F
z
=
_
S
f e
z
dS =
3 V
2 a
4 a
2
+ 2 a
2
_

0
(p
0
+ g a cos ) cos sin d, (9.125)
which reduces to
F
z
= 6 a V +
4
3
a
3
g. (9.126)
By symmetry, the horizontal components of the net force both average to zero. We can recognize the second term on
the right-hand side of the above equation as the buoyancy force due to the weight of the uid displaced by the sphere.
(See Chapter 3.) Moreover, the rst term can be interpreted as the viscous drag acting on the sphere. Note that this
drag acts in the opposite direction to the relative motion of the sphere with respect to the uid, and its magnitude is
directly proportional to the relative velocity.
Vertical force balance requires that
F
z
= Mg, (9.127)
where M is the spheres mass. In other words, in a steady state, the weight of the sphere balances the vertical force
exerted by the surrounding uid. If the sphere is composed of material of mean density then M = (4/3) a
3
.
Hence, in the frame in which the uid a large distance from the sphere is stationary, the steady vertical velocity with
which the sphere moves through the uid is
V =
2
9
a
2
g
_
1

_
, (9.128)
where = / is the uids kinematic viscosity. Obviously, if the sphere is more dense than the uid (i.e., if / > 1)
then it moves downward (i.e., V < 0), and vice versa. Finally, the typical Reynolds number of the uid ow in the
vicinity of the sphere is
Re =
2 V a
=
4
9
a
3
g
. (9.129)
For the case of a grain of sand falling through water at 20
C, we have / 2 and = 1.0 10

6
m
2
/s. Hence,
Re = (a/6 10
5
)
3
, where a is measured in meters. Thus, expression (9.128), which is strictly speaking only valid
when Re 1, but which turns out to be approximately valid for all Reynolds numbers less than unity, only holds for
sand grains whose radii are less than about 60 microns. Such grains fall through water at approximately 8 10
3
m/s.
For the case of a droplet of water falling through air at 20
C and atmospheric pressure, we have / = 780 and

= 1.5 10
5
m
2
/s. Hence, Re = (a/4 10
5
)
3
, where a is measured in meters. Thus, expression (9.128) only holds
for water droplets whose radii are less than about 40 microns. Such droplets fall through air at approximately 0.2 m/s.
At large values of r/a, Equations (9.105), (9.106), and (9.112) yield
v
r
(r, ) = V cos +
3
2
V cos
a
r
+ O
_
a
r
_
2
, (9.130)
v
(r, ) = V sin
3
4
V sin
a
r
+ O
_
a
r
_
2
. (9.131)
It follows that
[ (v ) v]
r
=
_
_
v
r
v
r
r
+
v
r
v
r
v
2
r
_
_

V
2
a
r
2
, (9.132)
and
(
2
v)
r

2
v
r
r
2

V a
r
3
. (9.133)
Hence,
[ (v ) v]
r
(
2
v)
r
V r
Re
r
a
, (9.134)
where Re is the Reynolds number of the ow in the immediate vicinity of the sphere. [See Equation (9.94).] Now, our
analysis is based on the assumption that advective inertia is negligible with respect to viscosity. However, as is clear
from the above expression for the ratio of inertia to viscosity within the uid, even if this ratio is much less than unity
close to the spherein other words, if Re 1it inevitably becomes much greater than unity far from the sphere:
i.e., for r a/Re. In other words, inertia always dominates viscosity, and our Stokes ow solution therefore breaks
down, at suciently large r/a.
9.11 Axisymmetric Stokes Flow In and Around a Fluid Sphere
Suppose that the solid sphere discussed in the previous section is replaced by a spherical uid drop of radius a. Let
the drop move through the surrounding uid at the constant velocity V e
z
. Obviously, the uid from which the drop is
composed must be immiscible with the surrounding uid. Let us transform to a frame of reference in which the drop
is stationary, and centered at the origin. Assuming that the Reynolds numbers immediately outside and inside the drop
are both much less than unity, and making use of the previous analysis, the most general expressions for the stream
function outside and inside the drop are
(r, ) = sin
2
_
A
r
+ Br + C r
2
+ Dr
4
_
, (9.135)
and
(r, ) = sin
2
_
A
r
+ Br + C r
2
+ Dr
4
_
_
, (9.136)
respectively. Here, A, B, C, etc. are arbitrary constants. Likewise, the previous analysis also allows us to deduce that
v
r
(r, ) = 2 cos
_
A
r
3
+
B
r
+ C + Dr
2
_
, (9.137)
v
(r, ) = sin
_
A
r
3
+
B
r
+ 2 C + 4 Dr
2
_
, (9.138)
186 FLUID MECHANICS
2
1
0
1
2
z/a
2 1 0 1 2
x/a
Figure 9.7: Contours of the stream function in the x-z plane for Stokes ow in and around a uid sphere. Here,
/ = 10.
r
(r, ) = sin
_
6 A
r
4
+ 6 Dr
_
, (9.139)
rr
(r, ) = p
0
+ g r cos + cos
_
12 A
r
4
+
6 B
r
2
+ 12 Dr
_
(9.140)
in the region r > a, with analogous expressions in the region r < a. Here, , , p
0
are the viscosity, density, and
ambient pressure of the uid surrounding the drop. Let , , and p
0
be the corresponding quantities for the uid that
makes up the drop.
In the region outside the drop, the uid velocity must asymptote to v = V e
z
at large r/a. This implies that
C = (1/2) V and D = 0. Furthermore, v
r
(a, ) = 0i.e., the normal velocity at the drop boundary must be zero
otherwise, the drop would change shape. This constraint yields A/a
3
+ B/a + (1/2) V = 0.
Inside the drop, the uid velocity must remain nite as r 0. This implies that A = B = 0. Furthermore, we
again require that v
r
(a, ) = 0, which yields C + Da
2
= 0.
Two additional physical constraints that must be satised at the interface between the two uids are, rstly,
continuity of tangential velocityi.e., v
(a
, ) = v
(a
+
, )and, secondly, continuity of tangential stressi.e.,
r
(a
, ) =
r
(a
+
, ). These constraints yield A/a
3
+ B/a V = 2 C + 4 Da
2
and 6 A/a
4
= 6 Da, respec-
tively.
At this stage, we have enough information to determine the values of A, B, C, and D. In fact, the stream functions
outside and inside the drop can be shown to take the form
(r, ) =
1
4
V a
2
sin
2
__

+
_
a
r

_
2 + 4
+
_
r
a
+ 2
_
r
a
_
2
_
, (9.141)
and
(r, ) =
1
4
V a
2
sin
2
_

+
_
_
r
a
_
2
_
1
_
r
a
_
2
_
, (9.142)
2
1
0
1
2
z/a
2 1 0 1 2
x/a
Figure 9.8: Contours of the stream function in the x-z plane for Stokes ow in and around a uid sphere. Here,
/ = 1/10.
respectively. See Figures 9.7 and 9.8.
The discontinuity in the radial stress across the drop boundary is
rr
(a
+
, )
rr
(a
, ) = p
0
p
0
+ ( ) g a cos 3
V
a
_
+ (3/2)
+
_
cos . (9.143)
The nal physical constraint that must be satised at r = a is
rr
(a
+
, )
rr
(a
, ) =
2
a
, (9.144)
where is the surface tension of the interface between the two uids. (See Section 4.3.) Hence, we obtain
p
0
p
0
=
2
a
, (9.145)
and
V =
a
2
g
3
_
1

_ _
+
+ (3/2)
_
, (9.146)
where = / is the kinematic viscosity of the surrounding uid. The fact that we have been able to completely satisfy
all of the physical constraints at the interface between the two uids, as long as the drop moves at the constant vertical
velocity V, proves that our previous assumptions that the interface is spherical, and that the drop moves vertically
through the surrounding uid at a constant speed without changing shape, were correct. In the limit, , in which
the drop is much more viscous than the surrounding uid, we recover Equation (9.128): i.e., the drop acts like a solid
sphere. On the other hand, in the limit , and , which is appropriate to an air bubble rising through a liquid,
we obtain
V =
a
2
g
3
. (9.147)
188 FLUID MECHANICS
9.12 Exercises
9.1. Consider viscous uid ow down a plane that is inclined at an angle to the horizontal. Let x measure distance along
the plane (i.e., along the path of steepest decent), and let y be a transverse coordinate such that the surface of the plane
corresponds to y = 0, and the free surface of the uid to y = h. Show that within the uid (i.e., 0 y h)
v
x
(y) =
g sin
2
y (2 h y),
p(y) = p
0
+ g cos (h y),
where is the kinematic viscosity, the density, and p
0
is atmospheric pressure.
9.2. If a viscous uid ows along a cylindrical pipe of circular cross-section that is inclined at an angle to the horizontal show
that the ow rate is
Q =
a
4
8
(G + g sin ) ,
where a is the pipe radius, the uid viscosity, the uid density, and G the pressure gradient.
9.3. Viscous uid ows steadily, parallel to the axis, in the annular region between two coaxial cylinders of radii a and n a, where
n > 1. Show that the volume ux of uid ow is
Q =
G a
4
8
_
n
4
1
(n
2
1)
2
ln n
_
,
where G is the eective pressure gradient, and the viscosity. Find the mean ow speed.
9.4. Consider viscous ow along a cylindrical pipe of elliptic cross-section. Suppose that the pipe runs parallel to the z-axis, and
that its boundary satises
x
2
a
2
+
y
2
b
2
= 1.
Let
v = v
z
(x, y) e
z
.
Demonstrate that
_

2
x
2
+
2
y
2
_
v
z
=
G
,
where G is the eective pressure gradient, and the uid viscosity. Show that
v
z
(x, y) =
G
2
a
2
b
2
b
2
x
2
a
2
y
2
a
2
+ b
2
is a solution of this equation that satises the no slip condition at the boundary. Demonstrate that the ow rate is
Q =
G
4
a
3
b
3
a
2
+ b
2
.
Finally, show that a pipe with an elliptic cross-section has lower ow rate than an otherwise similar pipe of circular cross-
section that has the same cross-sectional area.
9.5. Consider a velocity eld of the form
v(r) = r
2
(r) sin
2
,
where r, , are spherical coordinates. Demonstrate that this eld satises the equations of steady, incompressible, viscous
uid ow (neglecting advective inertia) with uniform pressure (neglecting gravity) provided that
d
dr
_
r
4
d
dr
_
= 0.
Suppose that a solid sphere of radius a, centered at the origin, is rotating about the z-axis, at the uniform angular velocity
0
, in a viscous uid, of viscosity , that is stationary at innity. Demonstrate that
(r) =
0
a
3
r
3
,
for r a. Show that the torque that the sphere exerts on the uid is
= 8
0
a
3
.
9.6. Consider a solid sphere of radius a moving through a viscous uid of viscosity at the xed velocity V = V e
z
. Let r, ,
be spherical coordinates whose origin coincides with the instantaneous location of the spheres center. Show that, if inertia
and gravity are negligible, the uid velocity, and the radial components of the stress tensor, a long way from the sphere, are
v
r

3
2
V cos
a
r
,
v

3
4
V sin
a
r
,
rr
p
0

9
2
V cos
a
r
2
,
r
0,
respectively. Hence, deduce that the net force exerted on the uid lying inside a large spherical surface of radius r, by the
uid external to the surface, is
F = 6 a V,
independent of the surface radius.
190 FLUID MECHANICS
Waves in Incompressible Fluids 191
10 Waves in Incompressible Fluids
10.1 Introduction
This chapter investigates low amplitude waves propagating through incompressible uids.
10.2 Gravity Waves
Consider a stationary body of water, of uniform depth d, located on the surface of the Earth. This body is assumed
to be suciently small compared to the Earth that its unperturbed surface is approximately planar. Let the Cartesian
coordinate z measure vertical height, with z = 0 corresponding to the aforementioned surface. Suppose that a small
amplitude wave propagates horizontally through the water, and let v(r, t) be the associated velocity eld.
Since water is essentially incompressible, its equations of motion are
v = 0, (10.1)
v
t
+ (v ) v = p g e
z
+
2
v, (10.2)
where is the (uniform) density, the (uniform) viscosity, and g the (uniform) acceleration due to gravity. (See
Section 2.14.) Let us write
p(r, t) = p
0
g z + p
1
(r, t), (10.3)
where p
0
is atmospheric pressure, and p
1
the pressure perturbation due to the wave. Of course, in the absence of the
wave, the water pressure a depth h below the surface is p
0
+ g h. (See Chapter 3.) Substitution into (10.2) yields
v
t
p
1
+
2
v, (10.4)
where we have neglected terms that are second-order in small quantities (i.e., terms of order v
2
).
Let us also neglect viscosity, which is a good approximation provided that the wavelength is not ridiculously small.
[For instance, for gravity waves in water, viscosity is negligible as long as (
2
/g)
1/3
5 10
5
m.] It follows that
v
t
p
1
. (10.5)
Taking the curl of this equation, we obtain
t
0, (10.6)
where = v is the vorticity. We conclude that the velocity eld associated with the wave is irrotational. Conse-
quently, the previous equation is automatically satised by writing
v = , (10.7)
where (r, t) is a velocity potential. (See Section 5.7.) However, from Equation (10.1), the velocity eld is also
divergence-free. It follows that the velocity potential satises Laplaces equation,
2
= 0. (10.8)
Finally, Equations (10.5) and (10.7) yield
p
1
=

t
. (10.9)
We now need to derive the physical constraints that must be satised at the waters upper and lower boundaries. It
is assumed that the water is bounded from below by a solid surface located at z = d. Since the water must always
192 FLUID MECHANICS
remain in contact with this surface, the appropriate physical constraint at the lower boundary is v
z
z=d
= 0 (i.e., the
normal velocity is zero at the lower boundary), or
z=d
= 0. (10.10)
The waters upper boundary is a little more complicated, since it is a free surface. Let represent the vertical displace-
ment of this surface due to the wave. It follows that
t
= v
z
z=0
=
z=0
. (10.11)
The appropriate physical constraint at the upper boundary is that the water pressure there must equal atmospheric
pressure, since there cannot be a pressure discontinuity across a free surface (in the absence of surface tensionsee
Section 10.11). Accordingly, from (10.3), we obtain
p
0
= p
0
g + p
1
z=0
, (10.12)
or
g = p
1
z=0
, (10.13)
which implies that
g
t
= g
z=0
=
p
1
t
z=0
, (10.14)
where use has been made of (10.11). The above expression can be combined with Equation (10.9) to give the boundary
condition
z=0
= g
1

2
t
2
z=0
. (10.15)
Let us search for a wave-like solution of Equation (10.8) of the form
(r, t) = F(z) cos(t k x). (10.16)
This solution actually corresponds to a propagating plane wave of wave vector k = k e
x
, angular frequency , and
amplitude F(z). Substitution into Equation (10.8) yields
d
2
F
dz
2
k
2
F = 0, (10.17)
whose independent solutions are exp(+k z) and exp(k z). Hence, a general solution to (10.8) takes the form
(x, z, t) = Ae
k z
cos(t k x) + Be
k z
cos(t k x), (10.18)
where A and B are arbitrary constants. The boundary condition (10.10) is satised provided that B = A exp(2 k d),
giving
(x, z, t) = A
_
e
k z
+ e
k (z+2 d)
_
cos(t k x), (10.19)
The boundary condition (10.15) then yields
Ak
_
1 e
2 k d
_
cos(t k x) = A

2
g
_
1 + e
2 k d
_
cos(t k x). (10.20)
which reduces to the dispersion relation
2
= g k tanh(k d). (10.21)
The type of wave described in this section is known as a gravity wave.
10.3 Gravity Waves in Deep Water
Consider the so-called deep water limit,
k d 1, (10.22)
in which the depth, d, of the water greatly exceeds the wavelength, = 2/k, of the wave. In this limit, the gravity
wave dispersion relation (10.21) reduces to
= (g k)
1/2
, (10.23)
since tanh(x) 1 as x . It follows that the phase velocity of gravity waves in deep water is
v
p
=
k
=
_
g
k
_
1/2
. (10.24)
Note that this velocity is proportional to the square root of the wavelength. Hence, deep water gravity waves with
long wavelengths propagate faster than those with short wavelengths. Now, the phase velocity, v
p
= /k, is dened
as the propagation velocity of a plane wave with the denite wave number, k [and a frequency given by the dispersion
relation (10.23)]. Such a wave has an innite spatial extent. A more realistic wave of nite spatial extent, with an
approximate wave number k, can be formed as a linear superposition of plane waves having a range of dierent wave
numbers centered on k. Such a construct is known as a wave pulse. As is well-known, wave pulses propagate at the
group velocity,
v
g
=
d
dk
. (10.25)
For the case of gravity waves in deep water, the dispersion relation (10.23) yields
v
g
=
1
2
_
g
k
_
1/2
=
1
2
v
p
. (10.26)
In other words, the group velocity of such waves is half their phase velocity.
Let (r, t) be the displacement of a particle of water, found at position r and time t, due to the passage of a deep
water gravity wave. It follows that
t
= v, (10.27)
where v(r, t) is the perturbed velocity. For a plane wave of wave number k = k e
x
, in the limit k d 1, Equation (10.19)
yields
(x, z, t) = Ae
k z
cos(t k x). (10.28)
Hence, [cf., Equations (10.45)(10.48)]
x
(x, z, t) = a e
k z
cos(t k x), (10.29)
z
(x, z, t) = a e
k z
sin(t k x), (10.30)
v
x
(x, z, t) = a e
k z
sin(t k x), (10.31)
v
z
(x, z, t) = a e
k z
cos(t k x), (10.32)
and
p
1
= g
z
, (10.33)
where use has been made of Equations (10.7), (10.9), and (10.27). Here, a is the amplitude of the vertical oscillation at
the waters surface. According to Equations (10.29)(10.32), the passage of the wave causes a water particle located a
depth h below the surface to execute a circular orbit of radius a e
k h
about its equilibriumposition. Note that the radius
of the orbit decreases exponentially with increasing depth. Furthermore, whenever the particles vertical displacement
attains a maximum value the particle is moving horizontally in the same direction as the wave, and vice versa. See
Figure 10.1.
194 FLUID MECHANICS
x
z
surface
z = 0
Figure 10.1: Motion of water particles associated with a deep water gravity wave propagating in the x-direction.
Finally, if we dene h(x, z, t) =
z
(x, z, t) z as the equilibrium depth of the water particle found at a given point
and time then Equations (10.3) and (10.33) yield
p(x, z, t) = p
0
+ g h(x, z, t). (10.34)
In other words, the pressure at this point and time is the same as the unperturbed pressure calculated at the equilibrium
depth of the water particle.
10.4 Gravity Waves in Shallow Water
Consider the so-called shallow water limit,
k d 1, (10.35)
in which the depth, d, of the water is much less than the wavelength, = 2/k, of the wave. In this limit, the gravity
wave dispersion relation (10.21) reduces to
= (g d)
1/2
k, (10.36)
since tanh(x) x as x 0. It follows that the phase velocities and group velocities of gravity waves in shallow water
all take the xed value
v
p
= v
g
= (g d)
1/2
, (10.37)
irrespective of wave number. We conclude thatunlike deep water wavesshallow water gravity waves are non-
dispersive in nature: i.e., waves pulses and plane waves all propagate at the same speed. Note, also, that the velocity
(10.37) increases with increasing water depth.
For a plane wave of wave number k = k e
x
, in the limit k d 1, Equation (10.19) yields
(x, z, t) = A[1 + k
2
(z + d)
2
/2] cos(t k x). (10.38)
Hence, Equations (10.7) and (10.27) give [cf., Equations (10.45)(10.48)]
x
(x, z, t) = a (k d)
1
cos(t k x), (10.39)
z
(x, z, t) = a (1 + z/d) sin(t k x), (10.40)
v
x
(x, z, t) = a (k d)
1
sin(t k x) (10.41)
v
z
(x, z, t) = a (1 + z/d) cos(t k x). (10.42)
Here, a is again the amplitude of the vertical oscillation at the waters surface. According to the above expressions,
the passage of a shallow water gravity wave causes a water particle located a depth h below the surface to execute an
elliptical orbit, of horizontal radius a/(k d), and vertical radius a (1h/d), about its equilibriumposition. Note that the
orbit is greatly elongated in the horizontal direction. Furthermore, its vertical radius decreases linearly with increasing
depth such that it becomes zero at the bottom (i.e., at h = d). As before, whenever the particles vertical displacement
attains a maximum value the particle is moving horizontally in the same direction as the wave, and vice versa.
10.5 Energy of Gravity Waves
It is easily demonstrated, from the analysis contained in the previous sections, that a gravity wave of arbitrary
wavenumber k, propagating horizontally through water of depth d, has a phase velocity
v
p
= (g d)
1/2
_
tanh(k d)
k d
_
1/2
. (10.43)
Moreover, the ratio of the group to the phase velocity is
v
g
v
p
=
1
2
_
1 +
2 k d
sinh(2 k d)
_
. (10.44)
Note, that neither the phase velocity nor the group velocity of a gravity wave can ever exceed the critical value
(g d)
1/2
. It is also easily demonstrated that the displacement and velocity elds associated with a plane gravity wave
of wavenumber k e
x
, angular frequency , and surface amplitude a, are
x
(x, z, t) = a
cosh[k (z + d)]
sinh(k d)
cos(t k x), (10.45)
z
(x, z, t) = a
sinh[k (z + d)]
sinh(k d)
sin(t k x), (10.46)
v
x
(x, z, t) = a
cosh[k (z + d)]
sinh(k d)
sin(t k x), (10.47)
v
z
(x, z, t) = a
sinh[k (z + d)]
sinh(k d)
cos(t k x). (10.48)
Now, the mean kinetic energy per unit surface area associated with a gravity wave is dened
K = (
_

d
1
2
v
2
dz), (10.49)
where
(x, t) = a sin(t k x) (10.50)
is the vertical displacement at the surface, and
( ) =
_
2
0
( )
d(k x)
2
(10.51)
is an average over a wavelength. Given that (cos
2
(t k x)) = (sin
2
(t k x)) = 1/2, it follows from (10.47) and
(10.48) that, to second order in a,
K =
1
4
a
2
2
_
0
d
cosh[2 k (z + d)]
sinh
2
(k d)
dz =
1
4
g a
2

2
g k tanh(k d)
. (10.52)
Making use of the general dispersion relation (10.21), we obtain
K =
1
4
g a
2
. (10.53)
196 FLUID MECHANICS
The mean potential energy perturbation per unit surface area associated with a gravity wave is dened
U = (
_

d
g z dz) +
1
2
g d
2
, (10.54)
which yields
U = (
1
2
g (
2
d
2
)) +
1
2
g d
2
=
1
2
g (
2
), (10.55)
or
U =
1
4
g a
2
. (10.56)
In other words, the mean potential energy per unit surface area of a gravity wave is equal to its mean kinetic energy
per unit surface area.
Finally, the mean total energy per unit surface area associated with a gravity wave is
E = K + U =
1
2
g a
2
. (10.57)
Note that this energy depends on the wave amplitude at the surface, but is independent of the wavelength, or the water
depth.
10.6 Wave Drag on Ships
Under certain circumstances (see the following section), a ship traveling over a body of water leaves behind it a train of
gravity waves whose wavefronts are transverse to the ships direction of motion. Since these waves possess energy that
is carried away from the ship, and eventually dissipated, this energy must have been produced at the ships expense.
The ship consequently experiences a drag force, D. Suppose that the ship is moving at the constant velocity V. We
would expect the transverse waves making up the train to have a matching phase velocity, so that they maintain a
constant phase relation with respect to the ship. To be more exact, we would generally expect the ships bow to always
correspond to a wave maximum (because of the pile up of water in front of the bow produced by the ships forward
motion). The condition v
p
= V, combined with expression (10.43), yields
tanh(k d)
k d
=
V
2
g d
. (10.58)
Suppose, for the sake of argument, that the wave train is of uniform transverse width w. Consider a xed line drawn
downstream of the ship at right-angles to its path. The rate at which the length of the train is increasing ahead of this
line is V. Therefore, the rate at which the energy of the train is increasing ahead of the line is (1/2) g a
2
wV, where
a is the typical amplitude of the transverse waves in the train. As is well-known, wave energy travels at the group
velocity, rather than the phase velocity. Thus, the energy ux per unit width of a propagating gravity wave is simply
E v
g
. Wave energy consequently crosses our xed line in the direction of the ships motion at the rate (1/2) g a
2
wv
g
.
Finally, the ship does work against the drag force, which goes to increase the energy of the train in the region ahead of
our line, at the rate DV. Energy conservation thus yields
1
2
g a
2
wV =
1
2
g a
2
wv
g
+ DV. (10.59)
However, since V = v
p
, we obtain
D =
1
2
g a
2
w
_
1
v
g
v
p
_
=
1
4
g a
2
w
_
1
2 k d
sinh(2 k d)
_
, (10.60)
where use has been made of Equation (10.44). Here, k d is determined implicitly in terms of the ship speed via
expression (10.58). Note that this expression cannot be satised when the speed exceeds the critical value (g d)
1/2
,
since gravity waves cannot propagate at speeds in excess of this value. In this situation, no transverse wave train can
keep up with the ship, and the drag associated with such waves consequently disappears. In fact, we can see, from
the above formulae, that when V (g d)
1/2
then k d 0, and so D 0. Note, however, that the transverse wave
amplitude, a, generally increases signicantly as the ship speed approaches the critical value. Hence, the drag due to
transverse waves actually peaks strongly at speeds just below the critical speed, before eectively falling to zero as
this speed is exceeded. Consequently, it usually requires a great deal of propulsion power to force a ship to travel at
speeds faster than (g d)
1/2
.
In the deep water limit k d 1, Equation (10.60) reduces to
D =
1
4
g a
2
w. (10.61)
Note that (at xed wave amplitude) this expression is independent of the wavelength of the wave train, and, hence,
independent of the ships speed. This result is actually rather misleading. In fact, (at xed wave amplitude) the drag
acting on a ship traveling through deep water varies signicantly with the ships speed. We can account for this
variation by incorporating the nite length of the ship into our analysis. A real ship moving through water generates
a bow wave from its bow, and a stern wave from its stern. Moreover, the bow wave tends to have a positive vertical
displacement, because water naturally piles up in front of the bow due to the forward motion of the ship, whereas the
stern wave tends tends to have a negative vertical displacement, because water rushes into the void left by the stern.
Very roughly speaking, suppose that the vertical displacement of the water surface caused by the ship is of the form
(x) cos
_
x
l
_
. (10.62)
Here, l is the length of the ship. Moreover, the bow lies (instantaneously) at x = 0 [hence, (0) > 0], and the stern
at x = l [hence, (l) < 0]. For the sake of simplicity, the upward water displacement due to the bow is assumed
to equal the downward displacement due to the stern. At xed bow wave displacement, the amplitude of transverse
gravity waves of wave number k = g/V
2
(chosen so that the phase velocity of the waves matches the ships speed, V)
produced by the ship is
a
1
l
_
l
0
cos
_
x
l
_
cos(k x) dx =
sin( k l)
k l
k l
+ k l
: (10.63)
i.e., the amplitude is proportional to the Fourier coecient of the ships vertical displacement pattern evaluated for a
wave number that matches that of the wave train. Hence, (at xed bow wave displacement) the drag produced by the
transverse waves is
D a
2
_
sin( F
2
)
F
2
1
1 + F
2
_
2
, (10.64)
where the dimensionless parameter
F =
V
(g l)
1/2
(10.65)
is known as the Froude number. See Section 2.16.
Figure 10.2 illustrates the variation of the wave drag with Froude number predicted by Equation (10.64). As we
can see, if the Froude number is much less than unity, which implies that the wavelength of the wave train is much
smaller than the length of the ship, then the drag is comparatively small. This is the case because the ship is extremely
inecient at driving short wavelength gravity waves. It can also be seen that the drag increases as the Froude number
increases, reaching a relatively sharp maximum when Fr = Fr
c
= 1/
= 0.56, and then falls rapidly. Now, Fr = Fr

c
corresponds to the case in which the length of the ship is equal to half the wavelength of the wave train. In this
situation, the bow and stern waves interfere constructively, leading to a particularly large amplitude wave train, and,
hence, to a particularly large wave drag. The smaller peaks visible in the gure correspond to other situations in which
the bow and stern waves interfere constructively. (For instance, when the length of the ship corresponds to one and a
half wavelengths of the wave train.) A heavy ship with large a displacement, and limited propulsion power, generally
cannot overcome the peak in the wave drag that occurs when Fr = Fr
c
. Such a ship is, therefore, limited to Froude
numbers in the range 0 < Fr < Fr
c
, which implies a maximum speed of
V
c
= 0.56 (g l)
1/2
= 1.75 [l(m)]
1/2
m/s = 3.4 [l(m)]
1/2
kts. (10.66)
This characteristic speed is sometimes called the hull speed. Note that the hull speed increases with the length of the
ship: i.e., long ships have higher hull speeds than short ones.
198 FLUID MECHANICS
0
0.1
0.2
0.3
D
(
a
.
u
.
)
0 0.2 0.4 0.6 0.8 1 1.2
Fr
Figure 10.2: Variation of wave drag with Froude number for a ship traveling through deep water.
10.7 Ship Wakes
Let us now make a detailed investigation of the wake pattern generated behind a ship as it travels over a body of water,
taking into account obliquely propagating gravity waves, in addition to transverse waves. For the sake of simplicity,
the nite length of the ship is neglected in the following analysis. In other words, the ship is treated as a point source
of gravity waves. Consider Figure 10.3. This shows a plane gravity wave generated on the surface of the water by a
moving ship. The water surface corresponds to the x-y plane. The ship is traveling along the x-axis, in the negative
x-direction, at the constant speed V. Suppose that the ships bow is initially at point A
, and has moved to point A after

a time interval t. Now, the only type of gravity wave that is continuously excited by the passage of the ship is one that
maintains a constant phase relation with respect to its bow. In fact, as we have already mentioned, the bow should
always correspond to a wave maximum. An oblique wavefront associated with such a wave is shown in the gure.
Here, the wavefront C
, which initially passes through the bow at point A
, has moved to CD after a time interval t,

such that it again passes through the bow at point A. Of course, the wavefront propagates at the phase velocity, v
p
. It
follows that, in the right-angled triangle AA
E, the sides AA
and A
E are of lengths V t and v

p
t, respectively, so that
sin =
v
p
V
. (10.67)
This, therefore, is the condition that must be satised in order for an obliquely propagating gravity wave to maintain a
constant phase relation with respect to the ship.
In shallow water, all gravity waves propagate at the same phase velocity: i.e.,
v
p
= (g d)
1/2
, (10.68)
where d is the water depth. Hence, Equation (10.67) yields
= sin
1
_
(g d)
1/2
V
_
. (10.69)
Note that this equation can only be satised when
V > (g d)
1/2
. (10.70)
E
D D
C C
A A
y
x
V t
v
p
t
Figure 10.3: An oblique plane wave generated on the surface of the water by a moving ship.
D
y
x
C
E
Figure 10.4: A shallow water wake.
In other words, the ship must be traveling faster than the critical speed (g d)
1/2
. Moreover, if this is the case then
there is only one value of that satises Equation (10.69). This implies the scenario illustrated in Figure 10.4. Here,
the ship is instantaneously at A, and the wave maxima that it previously generatedwhich all propagate obliquely,
subtending a xed angle with the x-axishave interfered constructively to produce a single strong wave maximum
DAE. In fact, the wave maxima generated when the ship was at A
have travelled to B
and C
, the wave maxima

generated when the ship was at A
have travelled to B
and C
, etc. We conclude that a ship traveling over shallow

water produces a V-shaped wake whose semi-angle, , is determined by the ships speed. Indeed, as is apparent from
Equation (10.69), the faster the ship travels over the water, the smaller the angle becomes. Shallow water wakes
are especially dangerous to other vessels, and particularly destructive of the coastline, because all of the wave energy
produced by the ship is concentrated into a single large wave maximum. Note, nally, that the wake contains no
transverse waves, since, as we have already mentioned, such waves cannot keep up with a ship traveling faster than
the critical speed ( g)
1/2
.
Let us now discuss the wake generated by a ship traveling over deep water. In this case, the phase velocity of
200 FLUID MECHANICS
D
y
x
A A
v
g
t
V t
B
C
P
Figure 10.5: Formation of an interference maximum in a deep water wake.
gravity waves is v
p
= (g/k)
1/2
. Thus, Equation (10.67) yields
sin =
v
p
V
=
_
g
k V
2
_
1/2
. (10.71)
It follows that in deep water any obliquely propagating gravity wave whose wave number exceeds the critical value
k
0
=
g
V
2
(10.72)
can keep up with the ship, as long as its direction of propagation is such that Equation (10.71) is satised. In other
words, the ship continuously excites gravity waves with a wide range of dierent wave numbers and propagation
directions. The wake is essentially the interference pattern generated by these waves. Now, as is well-known, an inter-
ference maximum generated by the superposition of plane waves with a range of dierent wave numbers propagates
at the group velocity, v
g
. Furthermore, as we have already seen, the group velocity of deep water gravity waves is half
their phase velocity: i.e., v
g
= v
p
/2.
Consider Figure 10.5. The curve APD corresponds to a particular interference maximum in the wake. Here, A
is the ships instantaneous position. Consider a point P on this curve. Let x and y be the coordinates of this point,
relative to the ship. Now, the interference maximum at P is part of the plane wavefront BC emitted some time t earlier,
when the ship was at point A
. Let be the angle subtended between this wavefront and the x-axis. Since interference
maxima propagate at the group velocity, the distance A
P is equal to v
g
t. Of course, the distance AA
is equal to V t.
Simple trigonometry reveals that
x = V t v
g
t sin , (10.73)
y = v
g
t cos . (10.74)
Moreover,
dy
dx
= tan , (10.75)
since BC is the tangent to the curve APDi.e., the curve y(x)at point P. It follows from Equation (10.71), and the
fact that v
g
= v
p
/2, that
x = X
_
1
1
2
sin
2
_
, (10.76)
y =
1
2
X sin cos , (10.77)
0.3
0.2
0.1
0
0.1
0.2
0.3
y
/
X
0
0 0.1 0.2 0.3 0.4 0.5 0.6
x/X
0
A
B
C
D
Figure 10.6: Locus of an interference maximum in a deep water wake.
where X() = V t. The previous three equations can be combined to produce
dy
dx
=
dy/d
dx/d
=
(1/2) dX/d sin cos + (1/2) X (cos
2
sin
2
)
dX/d [1 (1/2) sin
2
] X sin cos
=
sin
cos
, (10.78)
which reduces to
dX
d
=
X
tan
. (10.79)
This expression can be solved to give
X = X
0
sin , (10.80)
where X
0
is a constant. Hence, the locus of our interference maximum is determined parametrically by
x = X
0
sin
_
1
1
2
sin
2
_
, (10.81)
y =
1
2
X
0
sin
2
cos . (10.82)
Here, the angle ranges from /2 to +/2. The curve specied by the above equations is plotted in Figure 10.6. As
usual, A is the instantaneous position of the ship. It can be seen that the interference maximum essentially consists
of the transverse maximum BCD, and the two radial maxima AB and AD. As is easily demonstrated, point C, which
corresponds to = 0, lies at x = X
0
/2, y = 0. Moreover, the two cusps, B and D, which correspond to =
tan
1
(1/
8) = 19.47
, lie at x = (8/27)
1/2
X
0
, y = (1/27)
1/2
X
0
.
The complete interference pattern that constitutes the wake is constructed out of many dierent wave maximum
curves of the form shown in Figure 10.6, corresponding to many dierent values of the parameter X
0
. However, these
X
0
values must be chosen such that the wavelength of the pattern along the x-axis corresponds to the wavelength
0
= 2/k
0
= 2 V
2
/g of transverse (i.e., = 0) gravity waves whose phase velocity matches the speed of the
ship. This implies that X
0
= 2 j
0
, where j is a positive integer. A complete deep water wake pattern is shown in
Figure 10.7. Note that this pattern, which is made up of interlocking transverse and radial wave maxima, lls a wedge-
shaped regionknown as a Kelvin wedgewhose semi-angle takes the value tan
1
(1/
8) = 19.47
. This angle is
202 FLUID MECHANICS
6
5
4
3
2
1
0
1
2
3
4
5
6
y
/
0
1 0 1 2 3 4 5 6 7 8 9 10 11
x/
0
Figure 10.7: A deep water wake.
independent of the ships speed. Finally, our initial assumption that the gravity waves that form the wake are all deep
water waves is valid provided k
0
d 1, which implies that
V (g d)
1/2
. (10.83)
In other words, the ship must travel at a speed that is much less than the critical speed (g d)
1/2
. This explains why the
wake contains transverse wave maxima.
10.8 Gravity Waves in a Flowing Fluid
Consider a gravity wave traveling through a uid that is owing horizontally at the uniform velocity V = V e
x
. Let us
write
v(r, t) = V + v
1
(r, t), (10.84)
p(r, t) = p
0
g z + p
1
(r, t), (10.85)
where v
1
and p
1
are the small velocity and pressure perturbations, respectively, due to the wave. To rst order in small
quantities, the uid equations of motion, (10.1) and (10.2), reduce to
v
1
= 0, (10.86)
_

t
+ V
_
v
1
=
p
1
, (10.87)
respectively. We can also dene the displacement, (r, t), of a uid particle due to the passage of the wave, as seen in
a frame co-moving with the uid, as
_

t
+ V
_
= v
1
. (10.88)
The curl of Equation (10.87) implies that v
1
= 0. Hence, we can write v
1
= , and (10.87) yields
_

t
+ V
_
=
p
1
. (10.89)
z = d
z
z = 0
x
V
V
z = d
Figure 10.8: Gravity waves at an interface between two immiscible uids.
Finally, Equation (10.86) gives
2
= 0. (10.90)
The most general traveling wave solution to (10.90), with wavevector k = k e
x
, and angular frequency , is
(x, z, t) = [A cosh(k z) + B sinh(k z)] cos(t k x). (10.91)
It follows from Equation (10.89) that
p
1
(x, z, t) = k (V c) [A cosh(k z) + B sinh(k z)] sin(t k x), (10.92)
and from Equation (10.88) that
z
(x, z, t, ) = (V c)
1
[A sinh(k z) + B cosh(k z)] sin(t k x). (10.93)
Here, c = /k is the phase velocity of the wave.
10.9 Gravity Waves at an Interface
Consider a layer of uid of density
, depth d
, and uniform horizontal velocity V
, situated on top of a layer of

another uid of density , depth d, and uniform horizontal velocity V. Suppose that the uids are bounded from above
and below by rigid horizontal planes. Let these planes lie at z = d and z = d
, and let the unperturbed interface

between the two uids lie at z = 0. See Figure 10.8.
Consider a gravity wave of angular frequency , and wavenumber k, propagating through both uids in the x-
direction. Let
(x, t) =
0
sin(t k x) (10.94)
be the small vertical displacement of the interface due to the wave. In the lower uid, the perturbed velocity potential
must be of the form (10.91), with the constants A and B chosen such that v
z
z=d
= 0 and
z
(x, 0, t) = (x, t). It follows
that A = (V c)
0
/ tanh(k d) and B = (V c)
0
, so that
(x, z, t) = (V c)
0
cosh[k (z + d)]
sinh(k d)
cos(t k x). (10.95)
In the upper uid, the perturbed velocity potential must again be of the form (10.91), with the constants A and B
chosen such that v
z
z=d
= 0 and
z
(x, 0, t) = (x, t). It follows that A = (V
c)
0
/ tanh(k d
) and B = (V
c)
0
, so
that
(x, z, t) = (V
c)
0
cosh[k (z d
)]
sinh(k d
)
cos(t k x). (10.96)
204 FLUID MECHANICS
Here, c = /k is the phase velocity of the wave. From Equations (10.85) and (10.92), the uid pressure just below the
interface is
p(x, 0
, t) = p
0
g + p
1
(x, 0
, t)
= p
0
g
0
sin(t k x) k (V c) A sin(t k x)
= p
0

_
g
k (V c)
2
tanh(k d)
_
0
sin(t k x). (10.97)
Likewise, the uid pressure just above the interface is
p(x, 0
+
, t) = p
0

g + p
1
(x, 0
+
, t)
= p
0

g
0
sin(t k x)
k (V
c) A sin(t k x)
= p
0

_
g +
k (V
c)
2
tanh(k d
)
_
0
sin(t k x). (10.98)
Now, in the absence of surface tension at the interface, these two pressure must equal one another: i.e.,
_
p
_
z=0
+
z=0
= 0. (10.99)
Hence, we obtain the dispersion relation
(
) g =
k (V c)
2
tanh(k d)
+
k
(V
c)
2
tanh(k d
)
, (10.100)
which takes the form of a quadratic equation for the phase velocity, c, of the wave. We can see that:
i. If
= 0 and V = 0 then the dispersion relation reduces to (10.43) (with v

p
= c).
ii. If the two uids are of innite depth then the dispersion relation simplies to
(
) g = k (V c)
2
+ k
(V
c)
2
. (10.101)
iii. In general, there are two values of c that satisfy the quadratic equation (10.100). These are either both real, or
form a complex conjugate pair.
iv. The condition for stability is that c is real. The alternative is that c is complex, which implies that is also
complex, and, hence, that the perturbation grows or decays exponentially in time. Since the complex roots of
a quadratic equation occur in complex conjugate pairs, one of the roots always corresponds to an exponentially
growing mode: i.e., an instability.
v. If both uids are at rest (i.e., V = V
= 0), and of innite depth, then the dispersion reduces to

c
2
=
g (
)
k ( +
)
. (10.102)
It follows that the conguration is only stable when >
: i.e., when the heavier uid is underneath.

As a particular example, suppose that the lower uid is water, and the upper uid is the atmosphere. Let s =
/ =
1.225 10
3
be the specic density of air at s.t.p. (relative to water). Putting V = V
= 0, d
, and making use of

the fact that s is small, the dispersion relation (10.100) yields
c (g d)
1/2
_
tanh(k d)
k d
_
1/2
_
1
1
2
s [1 + tanh(k d)]
_
. (10.103)
Comparing this with (10.43), we can see that the presence of the atmosphere tends to slightly diminish the phase
velocities of gravity waves propagating over the surface of a body of water.
10.10 Steady Flow over a Corrugated Bottom
Consider a stream of water of mean depth d, and uniform horizontal velocity V = V e
x
, that ows over a corrugated
bottom whose elevation is z = d + a sin(k x), where a is much smaller than d. Let the elevation of the free surface of
the water be z = b sin(k x). We wish to determine the relationship between a and b.
Now, we expect the velocity potential, perturbed pressure, and vertical displacement of the water to be of the form
(10.91), (10.92), and (10.93), respectively, with = c = 0, since we are looking for a stationary (i.e., non-propagating)
perturbation driven by the static corrugations in the bottom. The boundary condition at the bottom is
z
(x, d) = a sin(k x), (10.104)
which yields
V
1
[A sinh(k d) + B cosh(k d)] = a. (10.105)
At the free surface, we have
z
(x, 0) = b sin(k x), (10.106)
which gives
b = V
1
B. (10.107)
In addition, pressure balance across the free surface yields
g b sin(k x) = p
1
(x, 0) = k V A sin(k x), (10.108)
which leads to
g b = k V A. (10.109)
Hence, from (10.105), (10.107), and (10.109),
b =
a
cosh(k d) (g/k V
2
) sinh(k d)
, (10.110)
or
b =
a
cosh(k d) (1 c
2
/V
2
)
, (10.111)
where c = [(g/k) tanh(k d)]
1/2
is the phase velocity of a gravity wave of wave number k. See Equation (10.21). It
follows that the peaks and troughs of the free surface coincide with those of the bottom when V > c, and the troughs
coincide with the peaks, and vice versa, when V < c. If V = c then the ratio b/a becomes innite, implying that
the oscillations driven by the corrugations are not of small amplitude, and, therefore, cannot be described by linear
theory.
10.11 Surface Tension
As described in Chapter 4, there is a positive excess energy per unit area, , associated with an interface between two
immiscible uids. The quantity can also be interpreted as a surface tension. Let us now incorporate surface tension
into our analysis. Suppose that the interface lies at
z = (x, t), (10.112)
where is small. Thus, the unperturbed interface corresponds to the plane z = 0. The unit normal to the interface is
n =
(z )
(z )
. (10.113)
It follows that
n
x

x
, (10.114)
n
z
1. (10.115)
206 FLUID MECHANICS
Now, the Young-Laplace Equation yields
p = n, (10.116)
where p is the jump in pressure seen crossing the interface in the opposite direction to n. See Section 4.2. However,
from (10.114) and (10.115), we have
n
x
2
. (10.117)
Hence, Equation (10.116) gives
[p]
z=0
+
z=0
=

2
x
2
. (10.118)
This expression is the generalization of (10.99) that takes surface tension into account.
Suppose that the interface in question is that between a body of water, of density and depth d, and the atmosphere.
Let the unperturbed water lie between z = d and z = 0, and let the unperturbed atmosphere occupy the region z > 0.
In the limit in which the density of the atmosphere is neglected, the pressure in the atmosphere takes the xed value
p
0
, whereas the pressure just below the surface of the water is p
0
g + p
1
z=0
. Here, p
1
is the pressure perturbation
due to the wave. The relation (10.118) yields
g p
1
z=0
=

2
x
2
, (10.119)
where is the surface tension at an air/water interface. However, /t = (/z)
z=0
, where is the perturbed velocity
potential of the water. Moreover, from (10.9), p
1
= (/t). Hence, the above expression gives
g

z
z=0
+
t
2
z=0
=
z
2
x
z=0
. (10.120)
This relation, which is a generalization of Equation (10.15), is the condition satised at a free surface in the presence
of non-negligible surface tension. Applying this boundary condition to the general solution, (10.19) (which already
satises the boundary condition at the bottom), we obtain the dispersion relation
2
=
_
g k +
k
3
_
tanh(k d), (10.121)
which is a generalization of (10.21) that takes surface tension into account.
10.12 Capillary Waves
In the deep water limit k d 1, the dispersion relation (10.121) simplies to
2
= g k +
k
3
. (10.122)
It is helpful to introduce the capillary length,
l =
_

g
_
1/2
. (10.123)
(See Section 4.4.) The capillary length of an air/water interface at s.t.p. is 2.7 10
3
m. The associated capillary
wavelength is
c
= 2 l = 1.7 10
2
m. Roughly speaking, surface tension is negligible for waves whose wavelengths
are much larger than the capillary wavelength, and vice versa. It is also helpful to introduce the critical phase velocity
v
c
= (2 g l)
1/2
. (10.124)
This critical velocity takes the value 0.23 m/s for an air/water interface at s.t.p. It follows from (10.122) that the phase
velocity, v
p
= /k, of a surface water wave can be written
v
p
v
c
=
_
1
2
_
k l +
1
k l
__
1/2
. (10.125)
Moreover, the ratio of the phase velocity to the group velocity, v
g
= d/dk, becomes
v
g
v
p
=
1
2
_
1 + 3 (k l)
2
1 + (k l)
2
_
. (10.126)
In the long wavelength limit
c
(i.e., k l 1), we obtain
v
p
v
0
1
(2 k l)
1/2
, (10.127)
and
v
g
v
p
1
2
. (10.128)
We can identify this type of wave as the deep water gravity wave discussed in Section 10.3.
In the short wavelength limit
c
(i.e., k l 1), we get
v
p
v
c
_
k l
2
_
1/2
, (10.129)
and
v
g
v
p
3
2
. (10.130)
This corresponds to a completely new type of wave known as a capillary wave. Such waves have wavelengths that are
much less than the capillary wavelength. Moreover, (10.129) can be rewritten
v
p

_
k
_
1/2
, (10.131)
which demonstrates that gravity plays no role in the propagation of a capillary waveits place is taken by surface
tension. Finally, it is easily seen that the phase velocity (10.125) attains the minimum value v
p
= v
c
when =
c
(i.e.,
when k l = 1). Moreover, from (10.126), v
g
= v
p
at this wavelength. It follows that the phase velocity of a surface
wave propagating over a body of water can never be less than the critical value, v
c
.
10.13 Capillary Waves at an Interface
Consider a layer of uid of density
, depth d
, and uniformhorizontal velocity V
, situated on top of a layer of another

uid of density , depth d, and uniform horizontal velocity V. Suppose that the uids are bounded from above and
below by rigid horizontal planes. Let these planes be at z = d and z = d
, and let the unperturbed interface between

the two uids be at z = 0. Suppose that the elevation of the perturbed interface is z = , where =
0
sin(t k x).
Finally, let be the surface tension of the interface. Equations (10.97), (10.98), and (10.118) yield the dispersion
relation
(
) g + k
2
=
k (V c)
2
tanh(k d)
+
k (V
c)
2
tanh(k d
)
, (10.132)
which is a generalization of the dispersion relation (10.100) that takes surface tension into account. Here, c = /k is
the phase velocity of a wave propagating along the interface.
For the case in which both uids are at rest, and of innite depth, the above dispersion relation simplies to give
(
) g + k
2
= ( +
) k c
2
. (10.133)
Suppose that s =
/ is the specic gravity of the upper uid with respect to the lower. In the case in which s < 1
(i.e., the upper uid is lighter than the lower one), it is helpful to dene
l =
_

g (1 s)
_
1/2
(10.134)
c
0
=
_
2 g l
_
1 s
1 + s
__
1/2
. (10.135)
208 FLUID MECHANICS
It follows that
c
2
c
2
0
=
1
2
_
1
k l
+ k l
_
. (10.136)
Thus, we conclude that the phase velocity of a wave propagating along the interface between the two uids achieves its
minimum value, c = c
0
, when k l = 1. Furthermore, waves of all wavelength are able to propagate along the interface
(i.e., c
2
> 0 for all k). In the opposite case, in which s > 1 (i.e., the upper uid is heavier than the lower one), we can
redene the capillary length as
l =
_

g (s 1)
_
1/2
. (10.137)
The dispersion relation (10.133) then becomes
c
2
= g l
_
s 1
s + 1
_ _
k l
1
k l
_
. (10.138)
It is apparent that c
2
< 0 for k l < 1, indicating instability of the interface for waves whose wavelengths exceed the
critical value
c
= 2 l. On the other hand, waves whose wavelengths are less than the critical value are stabilized by
surface tension. This result is exemplied by the experiment in which water is retained by atmospheric pressure in an
inverted glass whose mouth is closed by a gauze of ne mesh (the purpose of which is to put an upper limit on the
wavelengths of waves that can exist at the interface.)
10.14 Wind Driven Waves in Deep Water
Consider the scenario described in the previous section. Suppose that the lower uid is a body of deep water at rest,
and the upper uid is the atmosphere. Let the air above the surface of the water move horizontally at the constant
velocity V
. Suppose that is the density of water, s =
/ the specic gravity of air with respect to water, and the

surface tension at an air/water interface. With V = 0, k d , k d
, the dispersion relation (10.132) reduces to

(1 s) g + k
2
= k c
2
+ s k (V
c)
2
. (10.139)
This expression can be rearranged to give
c
2
_
2 V
s
1 + s
_
+
V
2
s
1 + s
= c
2
1
, (10.140)
which is a quadratic equation for the phase velocity, c, of the wave. Here,
c
2
1
=
g
k
_
1 s
1 + s
_
+
k
(1 + s)
, (10.141)
where c
1
is the phase velocity that the wave would have in the absence of the wind. In fact, we can write
c
2
1
=
1
2
_

c
+
_
c
2
0
(10.142)
where
c
= 2 l is the capillary wavelength, and l and c
0
are dened in Equations (10.134) and (10.135), respectively.
For a given wavelength, , the wave velocity, c, attains its maximum value, c
m
, when dc/dV
= 0. According to
the dispersion relation (10.140), this occurs when
V
= c
m
= (1 + s)
1/2
c
1
. (10.143)
If the wind has any other velocity, greater or less than c
m
, then the wave velocity is less than c
m
.
According to (10.140), the wave velocity, c, becomes complex, indicating an instability, when
V
2
>
(1 + s)
2
s
c
2
1
=
(1 + s)
2
2 s
_

c
+
_
c
2
0
. (10.144)
We conclude that if the wind speed exceeds the critical value
V
c
=
(1 + s)
s
1/2
c
0
= 6.6 m/s = 12.8 kts (10.145)
then waves whose wavelengths fall within a certain range, centered around
c
, are unstable and grow to large ampli-
tude.
The two roots of Equation (10.140) are
c =
V
s
1 + s

_
c
2
1

s V
2
(1 + s)
2
_
1/2
. (10.146)
Moreover, if
V
< (1 + s
1
)
1/2
c
1
(10.147)
then these roots have opposite signs. Hence, the waves can either travel with the wind, or against it, but travel faster
when they are moving with the wind. If V
exceeds the value given above then the waves cannot travel against the
wind. Since c
1
has the minimum value c
0
, it follows that waves traveling against the wind are completely ruled out
when
V
> (1 + s
1
)
1/2
c
0
= 6.6 m/s = 12.8 kts. (10.148)
10.15 Exercises
10.1. Find the velocity potential of a standing gravity wave in deep water for which the associated elevation of the free surface is
z = a cos(t) cos(k x).
Determine the paths of water particles perturbed by the wave.
10.2. Deep water lls a rectangular tank of length l and breadth b. Show that the resonant frequencies of the water in the tank are
(g )
1/2
(n
2
l
2
+ m
2
b
2
)
1/4
,
where n and m are integers. You may neglect surface tension.
10.3. Demonstrate that a sinusoidal gravity wave on deep water with surface elevation
= a cos(t k x)
possesses a mean momentum per unit surface area
1
2
a
2
.
10.4. A seismic wave passes along the bed of an ocean of uniform depth d such that the vertical perturbation of the bed is
a cos[k (x V t)]. Show that the amplitude of the consequent gravity waves at the surface is
a
__
1
c
2
V
2
_
cosh(k d)
_
1
,
where c is the phase velocity of waves of wavenumber k.
10.5. A layer of liquid of density and depth d has a free upper surface, and lies over liquid of innite depth and density > .
Neglecting surface tension, show that two possible types of wave of wavenumber k, with phase velocities
c
2
= k g,
c
2
=
k g ( )
coth(k d) +
,
can propagate along the layer.
210 FLUID MECHANICS
10.6. Show that, taking surface tension into account, a sinusoidal wave of wavenumber k and surface amplitude a has a mean
kinetic energy per unit surface area
K =
1
4
( g + k
2
) a
2
,
and a mean potential energy per unit surface area
U =
1
4
( g + k
2
) a
2
.
10.7. Show that in water of uniform depth d the phase velocity of surface waves can only attain a stationary (i.e., maximum or
minimum) value as a function of wavenumber, k, when
k =
_
sinh(2 k d) 2 k d
sinh(2 k d) + 2 k d
_
1/2
k
c
,
where k
c
= ( g/)
1/2
. Hence, deduce that the phase velocity has just one stationary value (a minimum) for any depth greater
than 3
1/2
k
1
c
4.8 mm, but no stationary values for depths less than that.
10.8. Unlike gravity waves in deep water, whose group velocities are half their phase velocities, the group velocities of capillary
waves are 3/2 times their phase velocities. Adapt the analysis of Section 10.7 to investigate the generation of capillary waves
by a very small object traveling across the surface of the water at the constant speed V. Suppose that the unperturbed surface
corresponds to the x-y plane. Let the object travel in the minus x-direction, such that it is instantaneously found at the origin.
Find the present position of waves that were emitted, traveling at an angle to the objects direction of motion, when it was
located at (X, 0). Show that along a given interference maximum the quantities X and vary in such a manner that X cos
3
takes a constant value, X

1
(say). Deduce that the interference maximum is given parametrically by the equations
x = X
1
sec
_
tan
2

1
2
_
,
y =
3
2
X
1
sec tan .
Sketch this curve, noting that it goes through the points (0.5 X
1
, 0) and (0, 1.3 X
1
), and asymptotes to y = 1.5 X
1/3
1
x
2/3
.
Equilibrium of Compressible Fluids 211
11 Equilibrium of Compressible Fluids
11.1 Introduction
In this chapter, we investigate the equilibria of compressible uids such as gases. As is the case for an incompressible
uid (see Chapter 3), a compressible uid in mechanical equilibrium must satisfy the force balance equation
0 = p + , (11.1)
where p is the static uid pressure, the mass density, and the gravitational potential energy per unit mass. In an
ideal gas, the relationship between p and is determined by the energy conservation equation, (2.89), which can be
written
1
D
Dt
_
p
_
=
A
F

2
_
p
_
. (11.2)
Here, is the ratio of specic heats, the thermal conductivity, Athe molar mass, and F the molar ideal gas constant.
Note that the viscous heat generation term has been omitted from the above equation (since it is zero in a stationary
gas). The limits in which the left- and right-hand sides of Equation (11.2) are dominant are termed the adiabatic and
isothermal limits, respectively. In the isothermal limit, in which thermal transport is comparatively large, so that (11.2)
can only be satised when
2
(p/) 0, the temperature (recall that T p/ in an ideal gas) distribution in the gas
becomes uniform, and the pressure and density are consequently related according to the isothermal gas law,
p
= constant. (11.3)
On the other hand, in the adiabatic limit, in which thermal transport is negligible, so that (11.2) can only be satised
when D/Dt(p/
) 0, the pressure and density are related according to the adiabatic gas law,
p
= constant. (11.4)
11.2 Isothermal Atmosphere
The vertical thickness of the atmosphere is only a few tens of kilometers, and is, therefore, much less than the radius
of the Earth, which is about 6000 km. Consequently, it is a good approximation to treat the atmosphere as a relatively
thin layer, covering the surface of the Earth, in which the pressure and density are only functions of altitude above
ground level, z, and the gravitational potential energy per unit mass takes the form = g z, where g is the acceleration
due to gravity at z = 0. It follows from Equation (11.1) that
dp
dz
= g. (11.5)
Now, in an isothermal atmosphere, in which the temperature, T, is assumed not to vary with height, the ideal gas
equation of state (2.84) yields [cf., Equation (11.3)]
p
=
FT
A
. (11.6)
The previous two equations can be combined to give
dp
dz
=
g A
FT
p. (11.7)
Hence, we obtain
p(z) = p
0
exp(z/H), (11.8)
212 FLUID MECHANICS
where p
0
10
5
Nm
2
is atmospheric pressure at ground level, and
H =
FT
g A
(11.9)
is known as the isothermal scale height of the atmosphere. Using the values T = 273 K (0
C), A = 29, and

g = 9.8 ms
2
, which are typical of the Earths atmosphere (at ground level), as well as F = 8.315 J mol
1
K
1
, we nd
that H = 7.99 km. Equations (11.6) and (11.8) yield
(z) =
0
exp(z/H), (11.10)
where
0
= p
0
/(g H) is the atmospheric mass density at z = 0. According to Equations (11.8) and (11.10), in an
isothermal atmosphere, the pressure and density both decrease exponentially with increasing altitude, falling to 37%
of their values at ground level when z = H, and to only 5% of these values when z = 3 H.
11.3 Adiabatic Atmosphere
In fact, the temperature of the Earths atmosphere is not uniform, but instead decreases steadily with increasing alti-
tude. This eect is largely due to the action of convection currents. When a packet of air ascends, under the inuence
of such currents, the diminished pressure at higher altitudes causes it to expand. Since this expansion generally takes
place far more rapidly than heat can diuse into the packet, the work done against the pressure of the surrounding
gas, as the packet expands, leads to a reduction in its internal energy, and, hence, in its temperature. Assuming that
the atmosphere is in a continually mixed state, whilst remaining in approximate vertical force balance (such a state is
known as a convective equilibrium), and that the eect of heat conduction is negligible (because the mixing takes place
too rapidly for thermal diusion to aect the temperature), we would expect the adiabatic gas law, (11.4), to oer a
much more accurate description of the relationship between atmospheric pressure and density than the isothermal gas
law, (11.3).
Let p = p/p
0
, = /
0
, and

T = T/T
0
, where p
0
,
0
, and T
0
are the pressure, mass density, and temperature of the
atmosphere, respectively, at ground level. The adiabatic gas law, (11.4), can be combined with the ideal gas equation
of state, (11.6), to give
p =
=

T
/(1)
. (11.11)
The isothermal scale height of the atmosphere is conveniently redened as [cf., (11.9)]
H =
FT
0
g A
=
p
0
g
0
. (11.12)
Equations (11.5), (11.11), and (11.12) yield
dp
dz
= p
1/
, (11.13)
wherez = z/H, or, from (11.11),
d
T
dz
=
1
. (11.14)
The above equation can be integrated to give
T(z) = T
0
_
1
1
z
H
_
. (11.15)
It follows that the temperature in an adiabatic atmosphere decreases linearly with increasing altitude at the rate of
[/( 1)] (T
0
/H) degrees per meter. This rate is known as the adiabatic lapse rate of the atmosphere. Using the
values = 1.4, T
0
= 273 K, and H = 7.99 km, which are typical of the Earths atmosphere, we estimate the lapse rate
to be 9.8 Kkm
1
. In reality, the lapse rate only takes this value in dry air. In moist air, the lapse rate is considerably
reduced because of the latent heat released when water vapor condenses.
Equations (11.11) and (11.15) yield
p(z) = p
0
_
1
1
z
H
_
/(1)
, (11.16)
(z) =
0
_
1
1
z
H
_
1/(1)
. (11.17)
Since /( 1) 3.5 and 1/( 1) 2.5, it follows that pressure decreases more rapidly than density in an adiabatic
atmosphere. Moreover, the previous three equations imply that an adiabatic atmosphere has a sharp upper boundary
at z = [/( 1)] H 28 km. At this altitude, the temperature, pressure, and density all fall to zero. Of course, above
this altitude, the temperature, pressure, and density remain zero (since they cannot take negative or imaginary values).
In contrast, an isothermal atmosphere has a diuse upper boundary in which the pressure and density never fall to
zero, even at extreme altitudes. It should be noted that, in reality, the Earths atmosphere does not have a sharp upper
boundary, since the adiabatic gas law does not hold at very high altitudes.
11.4 Atmospheric Stability
Suppose that the atmosphere is static (i.e., non-convecting). Moreover, let p(z) and (z) be the pressure and density,
respectively, as functions of altitude. Consider a packet of air that is in equilibrium with the surrounding air at some
initial altitude z
1
, but subsequently moves to a higher altitude z
2
. Thus, the packets initial pressure and density are
p
1
= p(z
1
) and
1
= (z
1
), respectively. Now, at the higher altitude, the packet must adjust its volume in such a manner
that its pressure matches that of the surrounding air, otherwise there would be a force imbalance across the packet
boundary. It follows that the packet pressure at altitude z
2
is p
2
= p(z
2
). Assuming that the packet moves upward
on a much faster time-scale than that needed for heat to diuse across it (but still a suciently slow time-scale that it
remains in approximate pressure balance with the surrounding air), we would expect its internal pressure and density
to be related according to the adiabatic gas law, (11.4). Thus, the packets density at altitude z
2
is
2
= (p
2
/p
1
)
1/
1
.
Now, if
2
> (z
2
) then the packet is denser than the surrounding air. It follows that the packet weight exceeds the
buoyancy due to the atmosphere, causing the packet to sink back to its original altitude. On the other hand, if
2
< (z
2
)
then the packet is less dense than the surrounding air. It follows that the buoyancy force exceeds the packet weight,
causing it to rise to an even higher altitude. In other words, the atmosphere is unstable to vertical convection when
[p(z
2
)/p(z
1
)]
1/
(z
1
) < (z
2
) for any z
2
> z
1
: i.e., when
p
z
2
<
p
z
1
, (11.18)
for any z
2
> z
1
. It follows that the atmosphere is only stable to vertical convection when p/
is a monotonically
decreasing function of altitude. As is easily demonstrated, this stability criterion can also be written
d ln p
d ln
< , (11.19)
or, making use of the ideal gas equation of state,
d ln T
d ln
< 1. (11.20)
Convection is triggered in regions of the atmosphere where the above stability criterion is violated. However, such
convection acts to relax these regions back to a marginally stable state in which p/
is uniform: i.e., an adiabatic

equilibrium.
11.5 Eddington Solar Model
Let us investigate the internal structure of the Sun, which is basically a self-gravitating sphere of incandescent ionized
gas (consisting mostly of hydrogen). Adopting a spherical coordinate system (see Section C.4), r, , , whose origin
214 FLUID MECHANICS
coincides with the Suns geometric center, and making the simplifying (and highly accurate) assumption that the mass
distribution within the Sun is spherically symmetric, we nd that
dm
dr
= 4 r
2
, (11.21)
where m(r) is the total mass contained within a sphere of radius r, centered on the origin, and (r) the mass density
at radius r. Now, as is well-known, the gravitational acceleration at some radius r in a spherically symmetric mass
distribution is the same as would be obtained were all the mass located within this radius concentrated at the center,
and the remainder of the mass neglected. In other words,
d
dr
=
Gm
r
2
, (11.22)
where (r) is the gravitational potential energy per unit mass, and d/dr the radial gravitational acceleration. The
force balance criterion (11.1) yields
dp
dr
+
d
dr
= 0, (11.23)
where p(r) is the pressure. The previous three equations can be combined to give
1
r
2
d
dr
_
r
2
dp
dr
_
= 4G. (11.24)
In order to make any further progress, we need to determine the relationship between the Suns internal pressure
and density. Unfortunately, this relationship is ultimately controlled by energy transport, which is a very complicated
process in a star. In fact, a stars energy is ultimately derived from nuclear reactions occurring deep within its core,
the details of which are extremely involved. This energy is then transported from the core to the outer boundary via a
combination of convection and radiation. (Conduction plays a much less important role in this process.) Unfortunately,
an exact calculation of radiative transport requires an understanding of the opacity of stellar material, which is an
exceptionally dicult subject. Finally, once the energy reaches the boundary of the star it is radiated away. The
following ingenious model, due to Eddington,
1
is appropriate to a star whose internal energy transport is dominated
by radiation. This turns out to be a fairly good approximation for the Sun. The main advantage of Eddingtons model
is that it does not require us to know anything about stellar nuclear reactions or opacity.
Now, the temperature inside the Sun is suciently large that radiation pressure cannot be completely neglected
with respect to conventional gas pressure. In other words, we must write the solar equation of state in the form
p = p
g
+ p
r
, (11.25)
where
p
g
=
k T
m
p
(11.26)
is the gas pressure (modeling the plasma within the Sun as an ideal gas of free electrons and ions), and
p
r
=
1
3
T
4
(11.27)
the radiation pressure (assuming that the radiation within the Sun is everywhere in local thermodynamic equilibrium
with the plasma). Here, T(r) is the Suns internal temperature, k the Boltzmann constant, m
p
the mass of a proton, and
the relative molecular mass (i.e., the ratio of the mean mass of the free particles making up the solar plasma to that
of a proton). Note that the electron mass has been neglected with respect to that of a proton. Furthermore, = 4/c,
where is the Stefan-Boltzmann constant, and c the velocity of light in a vacuum. Incidentally, in writing (11.26), we
have expressed A/F in the equivalent form m
p
/k.
1
A.S. Eddington, The Internal Constitution of the Stars (Cambridge University Press, Cambridge UK, 1926).
0
1
2
, y
0 1 2 3 4 5 6 7
Figure 11.1: The functions () (solid) and y() (dashed).

Let
p
g
= (1 ) p, (11.28)
p
r
= p, (11.29)
where the parameter is assumed to be uniform. In other words, the ratio of the radiation pressure to the gas pressure
is assumed to be the same everywhere inside the Sun. This fairly drastic assumption turns outperhaps, somewhat
fortuitously
2
to lead to approximately the correct internal pressure-density relation for the Sun. In fact, Equa-
tions (11.26)(11.29) can be combined to give
p = K
4/3
, (11.30)
where
K =
_
_
_
k
m
p
_
4
3
4
(1 )
4
_
_
1/3
. (11.31)
It can be seen, by comparison with Equation (11.4), that the above pressure-density relation takes the form of an
adiabatic gas law with an eective ratio of specic heats = 4/3. Note, however, that the actual ratio of specic heats
for a fully ionized hydrogen plasma, in the absence of radiation, is = 5/3. Hence, the 4/3 exponent, appearing in
(11.30), is entirely due to the non-negligible radiation pressure within the Sun.
Let T
c
= T(0),
c
= (0), and p
c
= p(0), be the Suns central temperature, density, and pressure, respectively. It
follows from (11.30) that
p
c
= K
4/3
c
, (11.32)
and from (11.26) and (11.28) that
T
c
=
p
c
c
m
p
k
(1 ) . (11.33)
2
L. Mestel, Phys. Reports 311, 295 (1999).
216 FLUID MECHANICS
6
7
l
o
g
1
0
(
T
[
K
]
)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
m/M
Figure 11.2: Solar temperature versus mass fraction obtained from the Eddington Solar Model (solid) and the Standard
Solar Model (dashed).
Suppose that
T
T
c
= , (11.34)
where is a dimensionless function. According to Equations (11.26), (11.28), and (11.30),
c
=
3
, (11.35)
p
p
c
=
4
. (11.36)
Moreover, it is clear, from the above expressions, that = 1 at the center of the Sun, r = 0, and = 0 at the edge,
r = R (say), where the temperature, density, and pressure are all assumed to fall to zero. Suppose, nally, that
r = a , (11.37)
where is a dimensionless radial coordinate, and
a =
_
K
G
2/3
c
_
1/2
. (11.38)
Thus, the center of the Sun corresponds to = 0, and the edge to =
1
(say), where (
1
) = 0, and
R =
1
a. (11.39)
Equations (11.35)(11.38) can be used to transformthe equilibriumrelation (11.24) into the non-dimensional form
1
2
d
d
_
2
d
d
_
=
3
. (11.40)
1
2
3
4
5
l
o
g
1
0
(
[
k
g
m
3
]
)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
m/M
Figure 11.3: Solar mass density versus mass fraction obtained from the Eddington Solar Model (solid) and the Stan-
dard Solar Model (dashed).
Moreover, Equation (11.21) can be integrated, with the aid of Equations (11.35), (11.37), and (11.40), and the physical
boundary condition m(0) = 0, to give
m = 4
c
a
3
y, (11.41)
where
y() =
2
d
d
. (11.42)
Equation (11.40) is known as the Lane-Emden equation (of degree 3), and can, unfortunately, only be solved numer-
ically. The appropriate solution takes the form = 1
2
/6 + O(
4
) when 1, and must be integrated to =
1
,
where (
1
) = 0. Figure 11.1 shows (), and the related function y(), obtained via numerical methods. Note that
1
= 6.897, and y
1
= y(
1
) = 2.018.
According to Equation (11.41), the solar mass, M = m(R), can be written
M = 4
c
a
3
y
1
, (11.43)
which reduces, with the aid of Equations (11.31) and (11.38), to
(1 )
4
=
2
, (11.44)
where
=
2
M
M
0
, (11.45)
and
M
0
=
_
_
1
(G)
3
_
k
m
p
_
4
3
_
1/2
4 y
1
= 3.586 10
31
kg. (11.46)
218 FLUID MECHANICS
10
11
12
13
14
15
16
17
l
o
g
1
0
(
p
[
N
m
2
]
)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
m/M
Figure 11.4: Solar pressure versus mass fraction obtained from the Eddington Solar Model (solid) and the Standard
Solar Model (dashed).
Moreover, it is easily demonstrated that
K =
G M
2/3
(4 y
1
)
2/3
. (11.47)
According to Equations (11.44) and (11.45), the ratio, /(1 ), of the radiation pressure to the gas pressure in a
radiative star is a strongly increasing function of the stellar mass, M, and mean molecular weight, . In the case of the
Sun, for which 1, Equation (11.44) can be inverted to give the approximate solution
=
2
1
4

4
+ O(
6
). (11.48)
Using the observed solar mass M = 1.989 10
30
kg, and the value = 0.68 (which represents the best t to the
Standard Solar Model mentioned below), we nd that = 6.58 10
4
. In other words, the radiation pressure inside
the Sun is only a very small fraction of the gas pressure. This immediately implies that radiative energy transport is
much less ecient than convective energy transport. Indeed, in regions of the Sun in which convection occurs we
would expect the convective transport to overwhelm the radiative transport, and so to drive the local pressure-density
relation toward an adiabatic law with an exponent 5/3. Fortunately, convection only takes place in the Suns outer
regions, which contain a minuscule fraction of its mass.
Equations (11.31), (11.32), (11.39), (11.43), and (11.47) yield
T
c
=
1
4 y
1
(1 )
G M m
p
Rk
= 1.34 10
7
K, (11.49)
c
=
3
1
4 y
1
M
R
3
= 7.63 10
4
kg m
3
, (11.50)
p
c
=
4
1
16 y
2
1
G M
2
R
4
= 1.24 10
16
Nm
2
, (11.51)
4
3.5
3
2.5
l
o
g
1
0
[
/
(
1
)
]
0 0.2 0.4 0.6 0.8 1
m/M
Figure 11.5: Ratio of radiation pressure to gas pressure calculated from the Eddington Solar Model (solid) and the
Standard Solar Model (dashed).
where the solar radius R has been given the observed value 6.960 10
8
m. The actual values of the Suns central
temperature, density, and pressure, as determined by the so-called Standard Solar Model (SSM),
3
which incorporates
detailed treatments of nuclear reactions and opacity, are T
c
= 1.58 10
7
K,
c
= 15.6 10
4
kg m
3
, and p
c
=
2.38 10
16
Nm
2
, respectively. It can be seen that the values of T
c
,
c
, and p
c
obtained from the Eddington model
lie within a factor of two of those obtained from the much more accurate SSM. Figures 11.2, 11.3, and 11.4 show
the temperature, density, and pressure proles, respectively, obtained from the SSM
4
and the Eddington model. The
proles are plotted as functions of the mass fraction, m(r)/M = y()/y
1
, where = r/a. It can be seen that there
is fairly good agreement between the proles calculated by the two models. Finally, Figure 11.5 compares the ratio,
/(1 ), of the radiation pressure to the gas pressure obtained from the SSM and the Eddington model. Recall, that
it is a fundamental assumption of the Eddington model that this pressure ratio is uniform throughout the Sun. In fact,
it can be seen that the pressure ratio calculated by the SSM is not spatially uniform. On the other hand, the spatial
variation of the ratio is fairly weak, except close to the edge of the Sun, where convection sets in, and the Eddington
model, thus, becomes invalid. We conclude that, despite its simplicity, the Eddington solar model does a remarkably
good job of accounting for the Suns internal structure.
11.6 Exercises
11.1. Prove that the fraction of the whole mass of an isothermal atmosphere which is included between the ground and a horizontal
plane of height z is
1 e
z/H
.
Evaluate this for z = H, 2 H, 3 H, respectively.
11.2. If the absolute temperature in the atmosphere diminishes upwards according to the law
T
T
0
= 1
z
c
,
3
http://en.wikipedia.org/wiki/Standard\_Solar\_Model
4
The SSM data is obtained from http://www.ap.stmarys.ca/
guenther/evolution/ssm1998.html
220 FLUID MECHANICS
where c is a constant, show that the pressure varies as
p
p
0
=
_
1
z
c
_
c/H
.
11.3. If the absolute temperature in the atmosphere diminishes upward according to the law
T
T
0
=
1
1 + z
,
where is a constant, show that the pressure varies as
p
p
0
= exp
_
z
H

1
2
z
2
H
_
.
11.4. Show that if the absolute temperature, T, in the atmosphere is any given function of the altitude, z, then the vertical distribu-
tion of pressure in the atmosphere is given by
ln
p
p
0
=
T
0
H
_
z
0
dz
T
.
11.5. Show that if the Earth were surrounded by an atmosphere of uniform temperature then the pressure a distance r from the
Earths center would be
p
p
0
= exp
_
a
2
H
_
1
r

1
a
__
,
where a is the Earths radius.
11.6. Show that if the whole of space were occupied by air at the uniform temperature T then the densities at the surfaces of the
various planets would be proportional to the corresponding values of
exp
_
g Aa
FT
_
,
where a is the radius of the planet, and g its surface gravitational acceleration.
11.7. Prove that in an atmosphere arranged in horizontal strata the work (per unit mass) required to interchange two thin strata of
equal mass without disturbance of the remaining strata is
1
1
_
1
2

1
1
_
_
p
1
p
2
2
_
,
where the suxes refer to the initial states of the two strata. Hence, show that for stability the ratio p/
must increase
upwards.
11.8. A spherically symmetric star is such that m(r) is the mass contained within radius r. Show that the stars total gravitational
potential energy can be written in the following three alternative forms:
U =
_
M
0
Gm
r
dm =
1
2
_
M
0
dm = 3
_
R
0
p dV
Here, M is the total mass, R the radius, (r) the gravitational potential per unit mass (dened such that 0 as r ),
p(r) the pressure, and dV = 4 r
2
dr.
11.9. Suppose that the pressure and density inside a spherically symmetric star are related according to the polytropic gas law,
p = K
(1+n)/n
,
where n is termed the polytropic index. Let =
c

n
, where
c
is the central mass density. Demonstrate that satises the
Lane-Emden equation
1
2
d
d
_
2
d
d
_
=
n
,
where r = a , and
=
_
(n + 1) K
4G

(n1)/n
c
_
1/2
.
Show that the physical solution to the Lane-Emden equation, which is such that (0) = 1 and (
1
) = 0, for some
1
> 0, is
= 1

2
6
for n = 0,
=
sin
for n = 1, and
=
1
(1 +
2
/3)
1/2
for n = 5. Determine the ratio of the central density to the mean density in all three cases. Finally, demonstrate that, in the
general case, the total gravitational potential energy can be written
U =
3
5 n
G M
2
R
,
where M is the total mass, and R = a
1
the radius.
11.10. A spherically symmetric star of radius R has a mass density of the form
(r) =
c
(1 r/R).
Show that the central mass density is four times the mean density. Demonstrate that the central pressure is
p
c
=
5
4
G M
2
R
,
where M is the mass of the star. Finally, show that the total gravitational potential energy of the star can be written
U =
26
35
G M
2
R
.
222 FLUID MECHANICS
Vectors and Vector Fields 223
A Vectors and Vector Fields
A.1 Introduction
This Appendix outlines those aspects of vector algebra, vector calculus, and vector eld theory that are helpful in the
study of uid dynamics.
A.2 Scalars and Vectors
Many physical entities (e.g., mass, energy) are entirely dened by a numerical magnitude (expressed in appropriate
units). Such entities, which have no directional element, are known as scalars. Moreover, since scalars can be
represented by real numbers it follows that they obey the laws of ordinary algebra. However, there exits a second
class of physical entities (e.g., velocity, acceleration, force) that are only completely dened when both a numerical
magnitude and a direction in space are specied. Such entities are known as vectors. By denition, a vector obeys
the same algebra as a displacement in space, and may thus be represented geometrically by a straight-line,

PQ (say),
where the arrow indicates the direction of the displacement (i.e., from point P to point Q). See Figure A.1. The
magnitude of the vector is represented by the length of the straight-line.
It is conventional to denote vectors by bold-faced symbols (e.g., a, F) and scalars by non-bold-faced symbols (e.g.,
r, S ). The magnitude of a general vector, a, is denoted a, or just a, and is, by denition, always greater than or equal
to zero. It is convenient to dene a vector with zero magnitudethis is denoted 0, and has no direction. Finally, two
vectors, a and b, are said to be equal when their magnitudes and directions are both identical.
A.3 Vector Algebra
Suppose that the displacements

PQ and

QR represent the vectors a and b, respectively. See Figure A.2. It can be seen
that the result of combining these two displacements is to give the net displacement

PR. Hence, if

PR represents the
vector c then we can write
c = a + b. (A.1)
This denes vector addition. By completing the parallelogram PQRS , we can also see that
PR=
PQ +
QR=
PS +
S R . (A.2)
However,

PS has the same length and direction as

QR, and, thus, represents the same vector, b. Likewise,

PQ and

S R
both represent the vector a. Thus, the above equation is equivalent to
c = a + b = b + a. (A.3)
P
Q
Figure A.1: A vector.
224 FLUID MECHANICS
c = a + b
P
S
R
Q
b
a
a
b
Figure A.2: Vector addition.
c = a b
b
a
c
b
Figure A.3: Vector subtraction.
We conclude that the addition of vectors is commutative. It can also be shown that the associative law holds: i.e.,
a + (b + c) = (a + b) + c. (A.4)
The null vector, 0, is represented by a displacement of zero length and arbitrary direction. Since the result of
combining such a displacement with a nite length displacement is the same as the latter displacement by itself, it
follows that
a + 0 = a, (A.5)
where a is a general vector. The negative of a is dened as that vector which has the same magnitude, but acts in the
opposite direction, and is denoted a. The sum of a and a is thus the null vector: i.e.,
a + (a) = 0. (A.6)
We can also dene the dierence of two vectors, a and b, as
c = a b = a + (b). (A.7)
This denition of vector subtraction is illustrated in Figure A.3.
If n > 0 is a scalar then the expression n a denotes a vector whose direction is the same as a, and whose magnitude
is n times that of a. (This denition becomes obvious when n is an integer.) If n is negative then, since n a = n (a),
it follows that n a is a vector whose magnitude is n times that of a, and whose direction is opposite to a. These
P y
O
x
z
Figure A.4: A right-handed Cartesian coordinate system.
denitions imply that if n and m are two scalars then
n (ma) = n ma = m(n a), (A.8)
(n + m) a = n a + ma, (A.9)
n (a + b) = n a + n b. (A.10)
A.4 Cartesian Components of a Vector
Consider a Cartesian coordinate system Oxyz consisting of an origin, O, and three mutually perpendicular coordinate
axes, Ox, Oy, and Oz. See Figure A.4. Such a system is said to be right-handed if, when looking along the Oz
direction, a 90
clockwise rotation about Oz is required to take Ox into Oy. Otherwise, it is said to be left-handed. It
is conventional to always use a right-handed coordinate system.
It is convenient to dene unit vectors, e
x
, e
y
, and e
z
, parallel to Ox, Oy, and Oz, respectively. Incidentally, a
unit vector is a vector whose magnitude is unity. The position vector, r, of some general point P whose Cartesian
coordinates are (x, y, z) is then given by
r = x e
z
+ y e
y
+ z e
z
. (A.11)
In other words, we can get from O to P by moving a distance x parallel to Ox, then a distance y parallel to Oy, and
then a distance z parallel to Oz. Similarly, if a is an arbitrary vector then
a = a
x
e
x
+ a
y
e
y
+ a
z
e
z
, (A.12)
where a
x
, a
y
, and a
z
are termed the Cartesian components of a. It is coventional to write a (a
x
, a
y
, a
z
). It follows
that e
x
(1, 0, 0), e
y
(0, 1, 0), and e
z
(0, 0, 1). Of course, 0 (0, 0, 0).
According to the three-dimensional generalization of the Pythagorean theorem, the distance OP r = r is given
by
r =
_
x
2
+ y
2
+ z
2
. (A.13)
By analogy, the magnitude of a general vector a takes the form
a =
_
a
2
x
+ a
2
y
+ a
2
z
. (A.14)
If a (a
x
, a
y
, a
z
) and b (b
x
, b
y
, b
z
) then it is easily demonstrated that
a + b (a
x
+ b
x
, a
y
+ b
y
, a
z
+ b
z
). (A.15)
Furthermore, if n is a scalar then it is apparent that
n a (n a
x
, n a
y
, n a
z
). (A.16)
226 FLUID MECHANICS
P
y
y
x
z
O
Figure A.5: Rotation of the coordinate axes about Oz.
A.5 Coordinate Transformations
A Cartesian coordinate system allows position and direction in space to be represented in a very convenient manner.
Unfortunately, such a coordinate systemalso introduces arbitrary elements into our analysis. After all, two independent
observers might well choose Cartesian coordinate systems with dierent origins, and dierent orientations of the
coordinate axes. In general, a given vector a will have dierent sets of components in these two coordinate systems.
However, the direction and magnitude of a are the same in both cases. Hence, the two sets of components must be
related to one another in a very particular fashion. Actually, since vectors are represented by moveable line elements
in space (i.e., in Figure A.2,

PQ and

S R represent the same vector), it follows that the components of a general vector
are not aected by a simple shift in the origin of a Cartesian coordinate system. On the other hand, the components
are modied when the coordinate axes are rotated.
Suppose that we transformto a new coordinate system, Ox
, which has the same origin as Oxyz, and is obtained

by rotating the coordinate axes of Oxyz through an angle about Oz. See Figure A.5. Let the coordinates of a general
point P be (x, y, z) in Oxyz and (x
, y
, z
) in Ox
. According to simple trigonometry, these two sets of coordinates

are related to one another via the transformation
x
= x cos + y sin , (A.17)

y
= x sin + y cos , (A.18)

z
= z. (A.19)
Consider the vector displacement r
OP. Note that this displacement is represented by the same symbol, r, in

both coordinate systems, since the magnitude and direction of r are manifestly independent of the orientation of the
coordinate axes. The coordinates of r do depend on the orientation of the axes: i.e., r (x, y, z) in Oxyz, and
r (x
, y
, z
) in Ox
. However, they must depend in a very specic manner [i.e., Equations (A.17)(A.19)] which
preserves the magnitude and direction of r.
The components of a general vector a transform in an analogous manner to Equations (A.17)(A.19): i.e.,
a
x
= a
x
cos + a
y
sin , (A.20)
a
y
= a
x
sin + a
y
cos , (A.21)
a
z
= a
z
. (A.22)
Moreover, there are similar transformation rules for rotation about Ox and Oy. Equations (A.20)(A.22) eectively
constitute the denition of a vector: i.e., the three quantities (a
x
, a
y
, a
z
) are the components of a vector provided that
they transform under rotation of the coordinate axes about Oz in accordance with Equations (A.20)(A.22). (And also
transform correctly under rotation about Ox and Oy). Conversely, (a
x
, a
y
, a
z
) cannot be the components of a vector
if they do not transform in accordance with Equations (A.20)(A.22). Of course, scalar quantities are invariant under
rotation of the coordinate axes. Thus, the individual components of a vector (a
x
, say) are real numbers, but they are not
scalars. Displacement vectors, and all vectors derived from displacements (e.g., velocity, acceleration), automatically
satisfy Equations (A.20)(A.22). There are, however, other physical quantities that have both magnitude and direction,
but which are not obviously related to displacements. We need to check carefully to see whether these quantities are
really vectors (see Sections A.7 and A.9).
A.6 Scalar Product
A scalar quantity is invariant under all possible rotational transformations. The individual components of a vector
are not scalars because they change under transformation. Can we form a scalar out of some combination of the
components of one, or more, vectors? Suppose that we were to dene the percent product,
a %b a
x
b
z
+ a
y
b
x
+ a
z
b
y
= scalar number, (A.23)
for general vectors a and b. Is a %b invariant under transformation, as must be the case if it is a scalar number?
Let us consider an example. Suppose that a (0, 1, 0) and b (1, 0, 0). It is easily seen that a %b = 1. Let
us now rotate the coordinate axes through 45
about Oz. In the new coordinate system, a (1/
2, 1/
2, 0) and
b (1/
2, 1/
2, 0), giving a %b = 1/2. Clearly, a %b is not invariant under rotational transformation, so the
above denition is a bad one.
Consider, now, the dot product or scalar product:
a b a
x
b
x
+ a
y
b
y
+ a
z
b
z
= scalar number. (A.24)
Let us rotate the coordinate axes though degrees about Oz. According to Equations (A.20)(A.22), a b takes the
form
a b = (a
x
cos + a
y
sin ) (b
x
cos + b
y
sin )
+(a
x
sin + a
y
cos ) (b
x
sin + b
y
cos ) + a
z
b
z
= a
x
b
x
+ a
y
b
y
+ a
z
b
z
(A.25)
in the new coordinate system. Thus, a b is invariant under rotation about Oz. It can easily be shown that it is also
invariant under rotation about Ox and Oy. We conclude that a b is a true scalar, and that the above denition is a
good one. Incidentally, a b is the only simple combination of the components of two vectors which transforms like a
scalar. It is readily shown that the dot product is commutative and distributive: i.e.,
a b = b a,
a (b + c) = a b + a c. (A.26)
The associative property is meaningless for the dot product, because we cannot have (a b) c, since a b is scalar.
We have shown that the dot product a b is coordinate independent. But what is the geometric signicance of this?
Well, in the special case where a = b, we get
a b = a
2
x
+ a
2
y
+ a
2
z
= a
2
= a
2
. (A.27)
So, the invariance of a a is equivalent to the invariance of the magnitude of vector a under transformation.
Let us now investigate the general case. The length squared of AB in the vector triangle shown in Figure A.6 is
(b a) (b a) = a
2
+ b
2
2 a b. (A.28)
However, according to the cosine rule of trigonometry,
(AB)
2
= (OA)
2
+ (OB)
2
2 (OA) (OB) cos , (A.29)
228 FLUID MECHANICS
b a
O
B
A
b
a
Figure A.6: A vector triangle.

where (AB) denotes the length of side AB. It follows that
a b = a b cos . (A.30)
In this case, the invariance of a b under transformation is equivalent to the invariance of the angle subtended between
the two vectors. Note that if a b = 0 then either a = 0, b = 0, or the vectors a and b are mutually perpendicular.
The angle subtended between two vectors can easily be obtained from the dot product: i.e.,
cos =
a b
a b
. (A.31)
The work W performed by a constant force F which moves an object through a displacement r is the product of
the magnitude of F times the displacement in the direction of F. If the angle subtended between F and r is then
W = F (r cos ) = F r. (A.32)
The work dW performed by a non-constant force f which moves an object through an innitesimal displacement
dr in a time interval dt is dW = f dr. Thus, the rate at which the force does work on the object, which is usually
referred to as the power, is P = dW/dt = f dr/dt, or P = f v, where v = dr/dt is the objects instantaneous velocity.
A.7 Vector Area
Suppose that we have planar surface of scalar area S . We can dene a vector area S whose magnitude is S , and whose
direction is perpendicular to the plane, in the sense determined by a right-hand circulation rule (see Section A.8)
applied to the rim, assuming that a direction of circulation around the rim is specied. See Figure A.7. This quantity
clearly possesses both magnitude and direction. But is it a true vector? We knowthat if the normal to the surface makes
an angle
x
with the x-axis then the area seen looking along the x-direction is S cos
x
. This is the x-component of S
(since S
x
= e
x
S = e
x
nS = cos
x
S , where n is the unit normal to the surface). Similarly, if the normal makes an
angle
y
with the y-axis then the area seen looking along the y-direction is S cos
y
. This is the y-component of S. If
we limit ourselves to a surface whose normal is perpendicular to the z-direction then
x
= /2
y
= . It follows that
S = S (cos , sin , 0). If we rotate the basis about the z-axis by degrees, which is equivalent to rotating the normal
to the surface about the z-axis by degrees, so that , then
S
x
= S cos ( ) = S cos cos + S sin sin = S
x
cos + S
y
sin , (A.33)
which is the correct transformation rule for the x-component of a vector. The other components transform correctly
as well. This proves both that a vector area is a true vector and that the components of a vector area are the projected
areas seen looking down the coordinate axes.
According to the vector addition theorem, the projected area of two plane surfaces, joined together at a line, looking
along the x-direction (say) is the x-component of the resultant of the vector areas of the two surfaces. Likewise, for
S
Figure A.7: A vector area.
many joined-up plane areas, the net area seen looking down the x-axis, which is the same as the area of the outer rim
seen looking down the x-axis, is the x-component of the resultant of all the vector areas: i.e.,
S =
i
S
i
. (A.34)
If we approach a limit, by letting the number of plane facets increase, and their areas reduce, then we obtain a contin-
uous surface denoted by the resultant vector area
S =
i
S
i
. (A.35)
It is clear that the area of the rim seen looking down the x-axis is just S
x
. Similarly, for the areas of the rim seen
looking down the other coordinate axes. Note that it is the rim of the surface that determines the vector area rather
than the nature of the surface spanning the rim. So, two dierent surfaces sharing the same rim both possess the same
vector area.
In conclusion, a loop (not all in one plane) has a vector area S which is the resultant of the component vector areas
of any surface ending on the loop. The components of S are the areas of the loop seen looking down the coordinate
axes. As a corollary, a closed surface has S = 0, since it does not possess a rim.
A.8 Vector Product
We have discovered how to construct a scalar from the components of two general vectors, a and b. Can we also
construct a vector which is not just a linear combination of a and b? Consider the following denition:
a b (a
x
b
x
, a
y
b
y
, a
z
b
z
). (A.36)
Is a b a proper vector? Suppose that a = (0, 1, 0), b = (1, 0, 0). In this case, a b = 0. However, if we rotate the
coordinate axes through 45
about Oz then a = (1/
2, 1/
2, 0), b = (1/
2, 1/
2, 0), and a b = (1/2, 1/2, 0).

Thus, a b does not transform like a vector, because its magnitude depends on the choice of axes. So, above denition
is a bad one.
Consider, now, the cross product or vector product:
a b (a
y
b
z
a
z
b
y
, a
z
b
x
a
x
b
z
, a
x
b
y
a
y
b
x
) = c. (A.37)
Does this rather unlikely combination transform like a vector? Let us try rotating the coordinate axes through an angle
about Oz using Equations (A.20)(A.22). In the new coordinate system,
c
x
= (a
x
sin + a
y
cos ) b
z
a
z
(b
x
sin + b
y
cos )
= (a
y
b
z
a
z
b
y
) cos + (a
z
b
x
a
x
b
z
) sin
= c
x
cos + c
y
sin . (A.38)
230 FLUID MECHANICS
b
middle nger
index nger
thumb
a b
a
Figure A.8: The right-hand rule for cross products. Here, is less that 180
.
Thus, the x-component of a b transforms correctly. It can easily be shown that the other components transform
correctly as well, and that all components also transform correctly under rotation about Ox and Oy. Thus, a b is a
proper vector. Incidentally, a b is the only simple combination of the components of two vectors that transforms like
a vector (which is non-coplanar with a and b). The cross product is anticommutative,
a b = b a, (A.39)
distributive,
a (b + c) = a b + a c, (A.40)
but is not associative,
a (b c) (a b) c. (A.41)
The cross product transforms like a vector, which means that it must have a well-dened direction and magnitude.
We can show that a b is perpendicular to both a and b. Consider a a b. If this is zero then the cross product must
be perpendicular to a. Now,
a a b = a
x
(a
y
b
z
a
z
b
y
) + a
y
(a
z
b
x
a
x
b
z
) + a
z
(a
x
b
y
a
y
b
x
)
= 0. (A.42)
Therefore, a b is perpendicular to a. Likewise, it can be demonstrated that a b is perpendicular to b. The vectors
a, b, and a b form a right-handed set, like the unit vectors e
x
, e
y
, and e
z
. In fact, e
x
e
y
= e
z
. This denes a unique
direction for a b, which is obtained from a right-hand rule. See Figure A.8.
Let us now evaluate the magnitude of a b. We have
(a b)
2
= (a
y
b
z
a
z
b
y
)
2
+ (a
z
b
x
a
x
b
z
)
2
+ (a
x
b
y
a
y
b
x
)
2
= (a
2
x
+ a
2
y
+ a
2
z
) (b
2
x
+ b
2
y
+ b
2
z
) (a
x
b
x
+ a
y
b
y
+ a
z
b
z
)
2
= a
2
b
2
(a b)
2
= a
2
b
2
a
2
b
2
cos
2
= a
2
b
2
sin
2
. (A.43)
Thus,
a b = a b sin , (A.44)
where is the angle subtended between a and b. Clearly, a a = 0 for any vector, since is always zero in this case.
Also, if a b = 0 then either a = 0, b = 0, or b is parallel (or antiparallel) to a.
Consider the parallelogram dened by the vectors a and b. See Figure A.9. The scalar area of the parallelogram
is a b sin . By convention, the vector area has the magnitude of the scalar area, and is normal to the plane of the
parallelogram, in the sense obtained from a right-hand circulation rule by rotating a on to b (through an acute angle):
i.e., if the ngers of the right-hand circulate in the direction of rotation then the thumb of the right-hand indicates the
direction of the vector area. So, the vector area is coming out of the page in Figure A.9. It follows that
S = a b, (A.45)
Suppose that a force F is applied at position r. See Figure A.10. The torque about the origin O is the product
of the magnitude of the force and the length of the lever arm OQ. Thus, the magnitude of the torque is F r sin .
The direction of the torque is conventionally dened as the direction of the axis through O about which the force tries
to rotate objects, in the sense determined by a right-hand circulation rule. Hence, the torque is out of the page in
Figure A.10. It follows that the vector torque is given by
= r F. (A.46)
The angular momentum, l, of a particle of linear momentum p and position vector r is simply dened as the
moment of its momentum about the origin: i.e.,
l = r p. (A.47)
A.9 Rotation
Let us try to dene a rotation vector whose magnitude is the angle of the rotation, , and whose direction is parallel
to the axis of rotation, in the sense determined by a right-hand circulation rule. Unfortunately, this is not a good vector.
The problem is that the addition of rotations is not commutative, whereas vector addition is commuative. Figure A.11
shows the eect of applying two successive 90
rotations, one about Ox, and the other about the Oz, to a standard
six-sided die. In the left-hand case, the z-rotation is applied before the x-rotation, and vice versa in the right-hand case.
It can be seen that the die ends up in two completely dierent states. In other words, the z-rotation plus the x-rotation
does not equal the x-rotation plus the z-rotation. This non-commuting algebra cannot be represented by vectors. So,
although rotations have a well-dened magnitude and direction, they are not vector quantities.
But, this is not quite the end of the story. Suppose that we take a general vector a and rotate it about Oz by a small
angle
z
. This is equivalent to rotating the coordinate axes about the Oz by
z
. According to Equations (A.20)
(A.22), we have
a
a +
z
e
z
a, (A.48)
where use has been made of the small angle approximations sin and cos 1. The above equation can easily be
generalized to allow small rotations about Ox and Oy by
x
and
y
, respectively. We nd that
a
a + a, (A.49)
where
=
x
e
x
+
y
e
y
+
z
e
z
. (A.50)
Clearly, we can dene a rotation vector, , but it only works for small angle rotations (i.e., suciently small that the
small angle approximations of sine and cosine are good). According to the above equation, a small z-rotation plus a
S
b
a
b
a
Figure A.9: A vector parallelogram.
232 FLUID MECHANICS
P
Q
F
r sin
r
Figure A.10: A torque.
x
x-axis z-axis
x-axis z-axis
z
y
Figure A.11: Eect of successive rotations about perpendicular axes on a six-sided die.
c
b
a
Figure A.12: A vector parallelepiped.
small x-rotation is (approximately) equal to the two rotations applied in the opposite order. The fact that innitesimal
rotation is a vector implies that angular velocity,
= lim
t0
t
, (A.51)
must be a vector as well. Also, if a
is interpreted as a(t + t) in Equation (A.49) then it follows that the equation of

motion of a vector which precesses about the origin with some angular velocity is
da
dt
= a. (A.52)
A.10 Scalar Triple Product
Consider three vectors a, b, and c. The scalar triple product is dened a b c. Now, b c is the vector area of the
parallelogram dened by b and c. So, a b c is the scalar area of this parallelogram multiplied by the component of
a in the direction of its normal. It follows that a b c is the volume of the parallelepiped dened by vectors a, b, and
c. See Figure A.12. This volume is independent of how the triple product is formed from a, b, and c, except that
a b c = a c b. (A.53)
So, the volume is positive if a, b, and c form a right-handed set (i.e., if a lies above the plane of b and c, in the sense
determined from a right-hand circulation rule by rotating b onto c) and negative if they form a left-handed set. The
triple product is unchanged if the dot and cross product operators are interchanged,
a b c = a b c. (A.54)
The triple product is also invariant under any cyclic permutation of a, b, and c,
a b c = b c a = c a b, (A.55)
but any anti-cyclic permutation causes it to change sign,
a b c = b a c. (A.56)
The scalar triple product is zero if any two of a, b, and c are parallel, or if a, b, and c are coplanar.
If a, b, and c are non-coplanar then any vector r can be written in terms of them: i.e.,
r = a + b + c. (A.57)
Forming the dot product of this equation with b c, we then obtain
r b c = a b c, (A.58)
so
=
r b c
a b c
. (A.59)
Analogous expressions can be written for and . The parameters , , and are uniquely determined provided
a b c 0: i.e., provided that the three vectors are non-coplanar.
234 FLUID MECHANICS
A.11 Vector Triple Product
For three vectors a, b, and c, the vector triple product is dened a (b c). The brackets are important because
a (b c) (a b) c. In fact, it can be demonstrated that
a (b c) (a c) b (a b) c (A.60)
and
(a b) c (a c) b (b c) a. (A.61)
Let us try to prove the rst of the above theorems. The left-hand side and the right-hand side are both proper
vectors, so if we can prove this result in one particular coordinate system then it must be true in general. Let us take
convenient axes such that Ox lies along b, and c lies in the x-y plane. It follows that b (b
x
, 0, 0), c (c
x
, c
y
, 0),
and a (a
x
, a
y
, a
z
). The vector b c is directed along Oz: i.e., b c (0, 0, b
x
c
y
). Hence, a (b c) lies in the
x-y plane: i.e., a (b c) (a
y
b
x
c
y
, a
x
b
x
c
y
, 0). This is the left-hand side of Equation (A.60) in our convenient
coordinate system. To evaluate the right-hand side, we need a c = a
x
c
x
+ a
y
c
y
and a b = a
x
b
x
. It follows that the
right-hand side is
RHS = ( [a
x
c
x
+ a
y
c
y
] b
x
, 0, 0) (a
x
b
x
c
x
, a
x
b
x
c
y
, 0)
= (a
y
c
y
b
x
, a
x
b
x
c
y
, 0) = LHS, (A.62)
which proves the theorem.
A.12 Vector Calculus
Suppose that vector a varies with time, so that a = a(t). The time derivative of the vector is dened
da
dt
= lim
t0
_
a(t + t) a(t)
t
_
. (A.63)
When written out in component form this becomes
da
dt

_
da
x
dt
,
da
y
dt
,
da
z
dt
_
. (A.64)
Suppose that a is, in fact, the product of a scalar (t) and another vector b(t). What now is the time derivative of
a? We have
da
x
dt
=
d
dt
( b
x
) =
d
dt
b
x
+
db
x
dt
, (A.65)
which implies that
da
dt
=
d
dt
b +
db
dt
. (A.66)
Moreover, it is easily demonstrated that
d
dt
(a b) =
da
dt
b + a
db
dt
, (A.67)
and
d
dt
(a b) =
da
dt
b + a
db
dt
. (A.68)
Hence, it can be seen that the laws of vector dierentiation are analogous to those in conventional calculus.
.
O
x
y
P
Q
l
P
f
Q
l
Figure A.13: A line integral.
A.13 Line Integrals
Consider a two-dimensional function f (x, y) which is dened for all x and y. What is meant by the integral of f along
a given curve joining the points P and Q in the x-y plane? Well, we rst draw out f as a function of length l along the
path. See Figure A.13. The integral is then simply given by
_
Q
P
f (x, y) dl = Area under the curve, (A.69)
where dl =
_
dx
2
+ dy
2
.
As an example of this, consider the integral of f (x, y) = x y
2
between P and Q along the two routes indicated in
Figure A.14. Along route 1 we have x = y, so dl =
2 dx. Thus,
_
Q
P
x y
2
dl =
_
1
0
x
3

2 dx =
2
4
. (A.70)
The integration along route 2 gives
_
Q
P
x y
2
dl =
_
1
0
x y
2
dx
y=0
+
_
1
0
x y
2
dy
x=1
= 0 +
_
1
0
y
2
dy =
1
3
. (A.71)
Note that the integral depends on the route taken between the initial and nal points.
The most common type of line integral is that in which the contributions from dx and dy are evaluated separately,
rather that through the path-length element dl: i.e.,
_
Q
P
_
f (x, y) dx + g(x, y) dy
_
. (A.72)
As an example of this, consider the integral
_
Q
P
_
y dx + x
3
dy
_
(A.73)
along the two routes indicated in Figure A.15. Along route 1 we have x = y + 1 and dx = dy, so
_
Q
P
_
y dx + x
3
dy
_
=
_
1
0
_
y dy + (y + 1)
3
dy
_
=
17
4
. (A.74)
236 FLUID MECHANICS
y
P = (0, 0)
Q = (1, 1)
1
2
2
x
Figure A.14: An example line integral.
O
P = (1, 0)
Q = (2, 1)
1
x
y
2
2
Figure A.15: An example line integral.
Along route 2,
_
Q
P
_
y dx + x
3
dy
_
=
_
1
0
x
3
dy
x=1
+
_
2
1
y dx
y=1
=
7
4
. (A.75)
Again, the integral depends on the path of integration.
Suppose that we have a line integral which does not depend on the path of integration. It follows that
_
Q
P
( f dx + g dy) = F(Q) F(P) (A.76)
for some function F. Given F(P) for one point P in the x-y plane, then
F(Q) = F(P) +
_
Q
P
( f dx + g dy) (A.77)
denes F(Q) for all other points in the plane. We can then draw a contour map of F(x, y). The line integral between
points P and Q is simply the change in height in the contour map between these two points:
_
Q
P
( f dx + g dy) =
_
Q
P
dF(x, y) = F(Q) F(P). (A.78)
Thus,
dF(x, y) = f (x, y) dx + g(x, y) dy. (A.79)
For instance, if F = x
3
y then dF = 3 x
2
y dx + x
3
dy and
_
Q
P
_
3 x
2
y dx + x
3
dy
_
=
_
x
3
y
_
Q
P
(A.80)
x
2
1
2
Q
P
y
a
O
Figure A.16: An example vector line integral.
is independent of the path of integration.
It is clear that there are two distinct types of line integralthose which depend only on their endpoints and not on
the path of integration, and those which depend both on their endpoints and the integration path. Later on, we shall
learn how to distinguish between these two types (see Section A.18).
A.14 Vector Line Integrals
A vector eld is dened as a set of vectors associated with each point in space. For instance, the velocity v(r) in a
moving liquid (e.g., a whirlpool) constitutes a vector eld. By analogy, a scalar eld is a set of scalars associated with
each point in space. An example of a scalar eld is the temperature distribution T(r) in a furnace.
Consider a general vector eld A(r). Let dr (dx, dy, dz) be the vector element of line length. Vector line
integrals often arise as
_
Q
P
A dr =
_
Q
P
(A
x
dx + A
y
dy + A
z
dz). (A.81)
For instance, if A is a force-eld then the line integral is the work done in going from P to Q.
As an example, consider the work done by a repulsive inverse-square central eld, F = r/r
3
. The element of
work done is dW = F dr. Take P = (, 0, 0) and Q = (a, 0, 0). The rst route considered is along the x-axis, so
W =
_
a
1
x
2
_
dx =
_
1
x
_
a
=
1
a
. (A.82)
The second route is, rstly, around a large circle (r = constant) to the point (a, , 0), and then parallel to the y-axis.
See Figure A.16. In the rst part, no work is done, since F is perpendicular to dr. In the second part,
W =
_
0
y dy
(a
2
+ y
2
)
3/2
=
_
1
(y
2
+ a
2
)
1/2
_
0
=
1
a
. (A.83)
In this case, the integral is independent of the path. However, not all vector line integrals are path independent.
A.15 Surface Integrals
Let us take a surface S , that is not necessarily co-planar, and divide it up into (scalar) elements S
i
. Then
_ _
S
f (x, y, z) dS = lim
S
i
0
i
f (x, y, z) S
i
(A.84)
238 FLUID MECHANICS
dy
x
2
x
1
y
1
y
2
y
x
Figure A.17: Decomposition of a surface integral.
is a surface integral. For instance, the volume of water in a lake of depth D(x, y) is
V =
_ _
D(x, y) dS. (A.85)
To evaluate this integral we must split the calculation into two ordinary integrals. The volume in the strip shown in
Figure A.17 is
__
x
2
x
1
D(x, y) dx
_
dy. (A.86)
Note that the limits x
1
and x
2
depend on y. The total volume is the sum over all strips: i.e.,
V =
_
y
2
y
1
dy
__
x
2
(y)
x
1
(y)
D(x, y) dx
_

_ _
S
D(x, y) dx dy. (A.87)
Of course, the integral can be evaluated by taking the strips the other way around: i.e.,
V =
_
x
2
x
1
dx
_
y
2
(x)
y
1
(x)
D(x, y) dy. (A.88)
Interchanging the order of integration is a very powerful and useful trick. But great care must be taken when evaluating
the limits.
As an example, consider
_ _
S
x y
2
dx dy, (A.89)
where S is shown in Figure A.18. Suppose that we evaluate the x integral rst:
dy
__
1y
0
x y
2
dx
_
= y
2
dy
_
x
2
2
_
1y
0
=
y
2
2
(1 y)
2
dy. (A.90)
Let us now evaluate the y integral:
_
1
0
_
y
2
2
y
3
+
y
4
2
_
dy =
1
60
. (A.91)
We can also evaluate the integral by interchanging the order of integration:
_
1
0
x dx
_
1x
0
y
2
dy =
_
1
0
x
3
(1 x)
3
dx =
1
60
. (A.92)
1 y = x
x
(1, 0)
(0, 1)
y
(0, 0)
Figure A.18: An example surface integral.
In some cases, a surface integral is just the product of two separate integrals. For instance,
_ _
S
x
2
y dx dy (A.93)
where S is a unit square. This integral can be written
_
1
0
dx
_
1
0
x
2
y dy =
__
1
0
x
2
dx
_ __
1
0
y dy
_
=
1
3
1
2
=
1
6
, (A.94)
since the limits are both independent of the other variable.
A.16 Vector Surface Integrals
Surface integrals often occur during vector analysis. For instance, the rate of ow of a liquid of velocity v through an
innitesimal surface of vector area dS is v dS. The net rate of ow through a surface S made up of lots of innitesimal
surfaces is
_ _
S
v dS = lim
dS 0
_
v cos dS
_
, (A.95)
where is the angle subtended between the normal to the surface and the ow velocity.
Analogously to line integrals, most surface integrals depend both on the surface and the rim. But some (very
important) integrals depend only on the rim, and not on the nature of the surface which spans it. As an example of
this, consider incompressible uid ow between two surfaces S
1
and S
2
which end on the same rim. See Figure A.23.
The volume between the surfaces is constant, so what goes in must come out, and
_ _
S
1
v dS =
_ _
S
2
v dS. (A.96)
It follows that
_ _
v dS (A.97)
depends only on the rim, and not on the form of surfaces S
1
and S
2
.
A.17 Volume Integrals
A volume integral takes the form
_ _ _
V
f (x, y, z) dV, (A.98)
240 FLUID MECHANICS
where V is some volume, and dV = dx dy dz is a small volume element. The volume element is sometimes written
d
3
r, or even d.
As an example of a volume integral, let us evaluate the center of gravity of a solid pyramid. Suppose that the
pyramid has a square base of side a, a height a, and is composed of material of uniform density. Let the centroid of
the base lie at the origin, and let the apex lie at (0, 0, a). By symmetry, the center of mass lies on the line joining the
centroid to the apex. In fact, the height of the center of mass is given by
z =
_ _ _
z dV
_ _ _ _
dV. (A.99)
The bottom integral is just the volume of the pyramid, and can be written
_ _ _
dV =
_
a
0
dz
_
(az)/2
(az)/2
dy
_
(az)/2
(az)/2
dx =
_
a
0
(a z)
2
dz =
_
a
0
(a
2
2 a z + z
2
) dz
=
_
a
2
z a z
2
+ z
3
/3
_
a
0
=
1
3
a
3
. (A.100)
Here, we have evaluated the z-integral last because the limits of the x- and y- integrals are z-dependent. The top integral
takes the form
_ _ _
z dV =
_
a
0
z dz
_
(az)/2
(az)/2
dy
_
(az)/2
(az)/2
dx =
_
a
0
z (a z)
2
dz =
_
a
0
(z a
2
2 a z
2
+ z
3
) dz
=
_
a
2
z
2
/2 2 a z
3
/3 + z
4
/4
_
a
0
=
1
12
a
4
. (A.101)
Thus,
z =
1
12
a
4
_
1
3
a
3
=
1
4
a. (A.102)
In other words, the center of mass of a pyramid lies one quarter of the way between the centroid of the base and the
apex.
A.18 Gradient
A one-dimensional function f (x) has a gradient d f /dx which is dened as the slope of the tangent to the curve at x.
We wish to extend this idea to cover scalar elds in two and three dimensions.
Consider a two-dimensional scalar eld h(x, y), which is (say) height above sea-level in a hilly region. Let dr
(dx, dy) be an element of horizontal distance. Consider dh/dr, where dh is the change in height after moving an
innitesimal distance dr. This quantity is somewhat like the one-dimensional gradient, except that dh depends on the
direction of dr, as well as its magnitude. In the immediate vicinity of some point P, the slope reduces to an inclined
plane. See Figure A.19. The largest value of dh/dr is straight up the slope. It is easily shown that for any other
direction
dh
dr
=
_
dh
dr
_
max
cos , (A.103)
where is the angle shown in Figure A.19. Let us dene a two-dimensional vector, gradh, called the gradient of h,
whose magnitude is (dh/dr)
max
, and whose direction is the direction of steepest ascent. The cos variation exhibited
in the above expression ensures that the component of gradh in any direction is equal to dh/dr for that direction.
The component of dh/dr in the x-direction can be obtained by plotting out the prole of h at constant y, and then
nding the slope of the tangent to the curve at given x. This quantity is known as the partial derivative of h with respect
to x at constant y, and is denoted (h/x)
y
. Likewise, the gradient of the prole at constant x is written (h/y)
x
. Note
that the subscripts denoting constant x and constant y are usually omitted, unless there is any ambiguity. It follows
that in component form
gradh
_
h
x
,
h
y
_
. (A.104)
O
y
contours of h(x, y)
x
P
high
low
direction of steepest ascent
dr
Figure A.19: A two-dimensional gradient.
Now, the equation of the tangent plane at P = (x
0
, y
0
) is
h
T
(x, y) = h(x
0
, y
0
) + (x x
0
) + (y y
0
). (A.105)
This has the same local gradients as h(x, y), so
=
h
x
, =
h
y
, (A.106)
by dierentiation of the above. For small dx = x x
0
and dy = y y
0
, the function h is coincident with the tangent
plane, so
dh =
h
x
dx +
h
y
dy. (A.107)
But, gradh (h/x, h/y) and dr (dx, dy), so
dh = gradh dr. (A.108)
Incidentally, the above equation demonstrates that gradh is a proper vector, since the left-hand side is a scalar, and,
according to the properties of the dot product, the right-hand side is also a scalar provided that dr and gradh are both
proper vectors (dr is an obvious vector, because it is directly derived from displacements).
Consider, now, a three-dimensional temperature distribution T(x, y, z) in (say) a reaction vessel. Let us dene
gradT, as before, as a vector whose magnitude is (dT/dr)
max
, and whose direction is the direction of the maximum
gradient. This vector is written in component form
gradT
_
T
x
,
T
y
,
T
z
_
. (A.109)
Here, T/x (T/x)
y,z
is the gradient of the one-dimensional temperature prole at constant y and z. The change
in T in going from point P to a neighbouring point oset by dr (dx, dy, dz) is
dT =
T
x
dx +
T
y
dy +
T
z
dz. (A.110)
In vector form, this becomes
dT = gradT dr. (A.111)
Suppose that dT = 0 for some dr. It follows that
dT = gradT dr = 0. (A.112)
242 FLUID MECHANICS
dr
isotherms
T = constant
gradT
Figure A.20: Isotherms.
So, dr is perpendicular to gradT. Since dT = 0 along so-called isotherms (i.e., contours of the temperature), we
conclude that the isotherms (contours) are everywhere perpendicular to gradT. See Figure A.20.
It is, of course, possible to integrate dT. For instance, the line integral of dT between points P and Q is written
_
Q
P
dT =
_
Q
P
gradT dr = T(Q) T(P). (A.113)
This integral is clearly independent of the path taken between P and Q, so
_
Q
P
gradT dr must be path independent.
Consider a vector eld A(r). In general, the line integral
_
Q
P
A dr depends on the path taken between the end
points. However, for some special vector elds the integral is path independent. Such elds are called conservative
elds. It can be shown that if A is a conservative eld then A = gradV for some scalar eld V. The proof of this is
straightforward. Keeping P xed, we have
_
Q
P
A dr = V(Q), (A.114)
where V(Q) is a well-dened function, due to the path independent nature of the line integral. Consider moving the
position of the end point by an innitesimal amount dx in the x-direction. We have
V(Q + dx) = V(Q) +
_
Q+dx
Q
A dr = V(Q) + A
x
dx. (A.115)
Hence,
V
x
= A
x
, (A.116)
with analogous relations for the other components of A. It follows that
A = gradV. (A.117)
The force eld due to gravity is a good example of a conservative eld. Now, if A(r) is a force-eld then
_
A dr
is the work done in traversing some path. If A is conservative then
_
A dr = 0, (A.118)
where
_
corresponds to the line integral around a closed loop. The fact that zero net work is done in going around a
closed loop is equivalent to the conservation of energy (which is why conservative elds are called conservative). A
good example of a non-conservative eld is the force eld due to friction. Clearly, a frictional system loses energy in
going around a closed cycle, so
_
A dr 0.
A.19 Grad Operator
It is useful to dene the vector operator

_

x
,
y
,
z
_
, (A.119)
which is usually called the grad or del operator. This operator acts on everything to its right in a expression, until the
end of the expression or a closing bracket is reached. For instance,
grad f = f
_
f
x
,
f
y
,
f
z
_
. (A.120)
For two scalar elds and ,
grad( ) = grad + grad (A.121)
can be written more succinctly as
( ) = + . (A.122)
Suppose that we rotate the coordinate axes through an angle about Oz. By analogy with Equations (A.17)(A.19),
the old coordinates (x, y, z) are related to the new ones (x
, y
, z
) via
x = x
cos y
sin , (A.123)
y = x
sin + y
cos , (A.124)
z = z
. (A.125)
Now,
=
_
x
x
_
y
,z
x
+
_
y
x
_
y
,z
y
+
_
z
x
_
y
,z
z
, (A.126)
giving
= cos
x
+ sin
y
, (A.127)
and
x
= cos
x
+ sin
y
. (A.128)
It can be seen, from Equations (A.20)(A.22), that the dierential operator transforms in an analogous manner to a
vector. This is another proof that f is a good vector.
A.20 Divergence
Let us start with a vector eld A(r). Consider
_
S
A dS over some closed surface S , where dS denotes an outward
pointing surface element. This surface integral is usually called the ux of A out of S . If A represents the velocity of
some uid then
_
S
A dS is the rate of uid ow out of S .
If A is constant in space then it is easily demonstrated that the net ux out of S is zero,
_
A dS = A
_
dS = A S = 0, (A.129)
since the vector area S of a closed surface is zero.
Suppose, now, that A is not uniform in space. Consider a very small rectangular volume over which A hardly
varies. The contribution to
_
A dS from the two faces normal to the x-axis is
A
x
(x + dx) dy dz A
x
(x) dy dz =
A
x
x
dx dy dz =
A
x
x
dV, (A.130)
244 FLUID MECHANICS
y + dy
z
z
x + dx
x
z + dz
y
x
y
Figure A.21: Flux of a vector eld out of a small box.
where dV = dx dy dz is the volume element. See Figure A.21. There are analogous contributions from the sides
normal to the y- and z-axes, so the total of all the contributions is
_
A dS =
_
A
x
x
+
A
y
y
+
A
z
z
_
dV. (A.131)
The divergence of a vector eld is dened
div A = A =
A
x
x
+
A
y
y
+
A
z
z
. (A.132)
Divergence is a good scalar (i.e., it is coordinate independent), since it is the dot product of the vector operator with
A. The formal denition of A is
A = lim
dV0
_
A dS
dV
. (A.133)
This denition is independent of the shape of the innitesimal volume element.
One of the most important results in vector eld theory is the so-called divergence theorem or Gauss theorem.
This states that for any volume V surrounded by a closed surface S ,
_
S
A dS =
_
V
A dV, (A.134)
where dS is an outward pointing volume element. The proof is very straightforward. We divide up the volume into
lots of very small cubes, and sum
_
A dS over all of the surfaces. The contributions from the interior surfaces cancel
out, leaving just the contribution from the outer surface. See Figure A.22. We can use Equation (A.131) for each cube
individually. This tells us that the summation is equivalent to
_
A dV over the whole volume. Thus, the integral
of A dS over the outer surface is equal to the integral of A over the whole volume, which proves the divergence
theorem.
Now, for a vector eld with A = 0,
_
S
A dS = 0 (A.135)
for any closed surface S . So, for two surfaces, S
1
and S
2
, on the same rim,
_
S
1
A dS =
_
S
2
A dS. (A.136)
See Figure A.23. (Note that the direction of the surface elements on S
1
has been reversed relative to those on the
closed surface. Hence, the sign of the associated surface integral is also reversed.) Thus, if A = 0 then the surface
.
exterior contributions survive
S
interior contributions cancel
Figure A.22: The divergence theorem.
S
rim
S
2
S
1
Figure A.23: Two surfaces spanning the same rim (right), and the equivalent closed surface (left).
246 FLUID MECHANICS
2 1
Figure A.24: Divergent lines of force.
integral depends on the rim but not on the nature of the surface which spans it. On the other hand, if A 0 then
the integral depends on both the rim and the surface.
Consider an incompressible uid whose velocity eld is v. It is clear that
_
v dS = 0 for any closed surface, since
what ows into the surface must ow out again. Thus, according to the divergence theorem,
_
v dV = 0 for any
volume. The only way in which this is possible is if v is everywhere zero. Thus, the velocity components of an
incompressible uid satisfy the following dierential relation:
v
x
x
+
v
y
y
+
v
z
z
= 0. (A.137)
It is sometimes helpful to represent a vector eld A by lines of force or eld-lines. The direction of a line of force
at any point is the same as the local direction of A. The density of lines (i.e., the number of lines crossing a unit surface
perpendicular to A) is equal to A. For instance, in Figure A.24, A is larger at point 1 than at point 2. The number of
lines crossing a surface element dS is A dS. So, the net number of lines leaving a closed surface is
_
S
A dS =
_
V
A dV. (A.138)
If A = 0 then there is no net ux of lines out of any surface. Such a eld is called a solenoidal vector eld. The
simplest example of a solenoidal vector eld is one in which the lines of force all form closed loops.
A.21 Laplacian Operator
So far we have encountered
=
_
x
,

y
,

z
_
, (A.139)
which is a vector eld formed from a scalar eld, and
A =
A
x
x
+
A
y
y
+
A
z
z
, (A.140)
which is a scalar eld formed from a vector eld. There are two ways in which we can combine gradient and diver-
gence. We can either formthe vector eld ( A) or the scalar eld (). The former is not particularly interesting,
but the scalar eld () turns up in a great many physical problems, and is, therefore, worthy of discussion.
Let us introduce the heat ow vector h, which is the rate of ow of heat energy per unit area across a surface
perpendicular to the direction of h. In many substances, heat ows directly down the temperature gradient, so that we
can write
h = T, (A.141)
where is the thermal conductivity. The net rate of heat ow
_
S
h dS out of some closed surface S must be equal to
the rate of decrease of heat energy in the volume V enclosed by S . Thus, we have
_
S
h dS =

t
__
c T dV
_
, (A.142)
where c is the specic heat. It follows from the divergence theorem that
h = c
T
t
. (A.143)
Taking the divergence of both sides of Equation (A.141), and making use of Equation (A.143), we obtain
( T) = c
T
t
. (A.144)
If is constant then the above equation can be written
(T) =
c
T
t
. (A.145)
The scalar eld (T) takes the form
(T) =
x
_
T
x
_
+
y
_
T
y
_
+
z
_
T
z
_
=
2
T
x
2
+
2
T
y
2
+
2
T
z
2

2
T. (A.146)
Here, the scalar dierential operator

2
x
2
+
2
y
2
+
2
z
2
(A.147)
is called the Laplacian. The Laplacian is a good scalar operator (i.e., it is coordinate independent) because it is formed
from a combination of divergence (another good scalar operator) and gradient (a good vector operator).
What is the physical signicance of the Laplacian? In one dimension,
2
T reduces to
2
T/x
2
. Now,
2
T/x
2
is positive if T(x) is concave (from above) and negative if it is convex. So, if T is less than the average of T in its
surroundings then
2
T is positive, and vice versa.
In two dimensions,
2
T =
2
T
x
2
+
2
T
y
2
. (A.148)
Consider a local minimum of the temperature. At the minimum, the slope of T increases in all directions, so
2
T is
positive. Likewise,
2
T is negative at a local maximum. Consider, now, a steep-sided valley in T. Suppose that the
bottom of the valley runs parallel to the x-axis. At the bottom of the valley
2
T/y
2
is large and positive, whereas
2
T/x
2
is small and may even be negative. Thus,
2
T is positive, and this is associated with T being less than the
average local value.
Let us now return to the heat conduction problem:
2
T =
c
T
t
. (A.149)
It is clear that if
2
T is positive then T is locally less than the average value, so T/t > 0: i.e., the region heats up.
Likewise, if
2
T is negative then T is locally greater than the average value, and heat ows out of the region: i.e.,
T/t < 0. Thus, the above heat conduction equation makes physical sense.
A.22 Curl
Consider a vector eld A(r), and a loop which lies in one plane. The integral of A around this loop is written
_
A dr,
where dr is a line element of the loop. If A is a conservative eld then A = and
_
A dr = 0 for all loops. In
general, for a non-conservative eld,
_
A dr 0.
For a small loop we expect
_
A dr to be proportional to the area of the loop. Moreover, for a xed-area loop
we expect
_
A dr to depend on the orientation of the loop. One particular orientation will give the maximum value:
248 FLUID MECHANICS
z
z + dz
1
4
3
z
y
2
y + dy
y
Figure A.25: A vector line integral around a small rectangular loop in the y-z plane.
_
A dr = I
max
. If the loop subtends an angle with this optimum orientation then we expect I = I
max
cos . Let us
introduce the vector eld curl A whose magnitude is
curl A = lim
dS 0
_
A dr
dS
(A.150)
for the orientation giving I
max
. Here, dS is the area of the loop. The direction of curl A is perpendicular to the plane
of the loop, when it is in the orientation giving I
max
, with the sense given by a right-hand circulation rule.
Let us now express curl A in terms of the components of A. First, we shall evaluate
_
A dr around a small
rectangle in the y-z plane. See Figure A.25. The contribution from sides 1 and 3 is
A
z
(y + dy) dz A
z
(y) dz =
A
z
y
dy dz. (A.151)
The contribution from sides 2 and 4 is
A
y
(z + dz) dy + A
y
(z) dy =
A
y
y
dy dz. (A.152)
So, the total of all contributions gives
_
A dr =
_
A
z
y
A
y
z
_
dS, (A.153)
where dS = dy dz is the area of the loop.
Consider a non-rectangular (but still small) loop in the y-z plane. We can divide it into rectangular elements, and
form
_
A dr over all the resultant loops. The interior contributions cancel, so we are just left with the contribution
from the outer loop. Also, the area of the outer loop is the sum of all the areas of the inner loops. We conclude that
_
A dr =
_
A
z
y
A
y
z
_
dS
x
(A.154)
is valid for a small loop dS = (dS
x
, 0, 0) of any shape in the y-z plane. Likewise, we can show that if the loop is in
the x-z plane then dS = (0, dS
y
, 0) and
_
A dr =
_
A
x
z

A
z
x
_
dS
y
. (A.155)
Finally, if the loop is in the x-y plane then dS = (0, 0, dS
z
) and
_
A dr =
_
A
y
x

A
x
y
_
dS
z
. (A.156)
dS
y x
3
2 1
z
Figure A.26: Decomposition of a vector area into its Cartesian components.
Imagine an arbitrary loop of vector area dS = (dS
x
, dS
y
, dS
z
). We can construct this out of three vector areas, 1,
2, and 3, directed in the x-, y-, and z-directions, respectively, as indicated in Figure A.26. If we form the line integral
around all three loops then the interior contributions cancel, and we are left with the line integral around the original
loop. Thus,
_
A dr =
_
A dr
1
+
_
A dr
2
+
_
A dr
3
, (A.157)
giving
_
A dr = curl A dS = curl A dS cos , (A.158)
where
curl A =
_
A
z
y
A
y
z
,
A
x
z

A
z
x
,
A
y
x

A
x
y
_
, (A.159)
and is the angle subtended between the directions of curl A and dS. Note that
curl A = A. (A.160)
This demonstrates that A is a good vector eld, since it is the cross product of the operator (a good vector
operator) and the vector eld A.
Consider a solid body rotating about the z-axis. The angular velocity is given by = (0, 0, ), so the rotation
velocity at position r is
v = r (A.161)
[see Equation (A.52)]. Let us evaluate v on the axis of rotation. The x-component is proportional to the integral
_
v dr around a loop in the y-z plane. This is plainly zero. Likewise, the y-component is also zero. The z-component
is
_
v dr/dS around some loop in the x-y plane. Consider a circular loop. We have
_
v dr = 2 r r with dS = r
2
.
Here, r is the perpendicular distance from the rotation axis. It follows that ( v)
z
= 2 , which is independent of r.
So, on the axis, v = (0 , 0 , 2 ). O the axis, at position r
0
, we can write
v = (r r
0
) + r
0
. (A.162)
The rst part has the same curl as the velocity eld on the axis, and the second part has zero curl, since it is constant.
Thus, v = (0, 0, 2 ) everywhere in the body. This allows us to form a physical picture of A. If we imagine
A(r) as the velocity eld of some uid then A at any given point is equal to twice the local angular rotation
velocity: i.e., 2 . Hence, a vector eld with A = 0 everywhere is said to be irrotational.
250 FLUID MECHANICS
Another important result of vector eld theory is the curl theorem or Stokes theorem:
_
C
A dr =
_
S
A dS, (A.163)
for some (non-planar) surface S bounded by a rim C. This theorem can easily be proved by splitting the loop up into
many small rectangular loops, and forming the integral around all of the resultant loops. All of the contributions from
the interior loops cancel, leaving just the contribution from the outer rim. Making use of Equation (A.158) for each of
the small loops, we can see that the contribution from all of the loops is also equal to the integral of A dS across
the whole surface. This proves the theorem.
One immediate consequence of Stokes theorem is that A is incompressible. Consider any two surfaces, S
1
and S
2
, which share the same rim. See Figure A.23. It is clear from Stokes theorem that
_
A dS is the same for
both surfaces. Thus, it follows that
_
A dS = 0 for any closed surface. However, we have from the divergence
theorem that
_
A dS =
_
( A) dV = 0 for any volume. Hence,
( A) 0. (A.164)
So, A is a solenoidal eld.
We have seen that for a conservative eld
_
A dr = 0 for any loop. This is entirely equivalent to A = .
However, the magnitude of A is lim
dS 0
_
A dr/dS for some particular loop. It is clear then that A = 0 for
a conservative eld. In other words,
() 0. (A.165)
Thus, a conservative eld is also an irrotational one.
A.23 Useful Vector Identities
Notation: a, b, c, d are general vectors; , are general scalar elds; A, B are general vector elds; (A )B
(A B
x
, A B
y
, A B
z
) and
2
A = (
2
A
x
,
2
A
y
,
2
A
z
) (but, only in Cartesian coordinatessee Appendix C).
a (b c) = (a c) b (a b) c, (A.166)
(a b) c = (c a) b (c b) a, (A.167)
(a b) (c d) = (a c) (b d) (a d) (b c), (A.168)
(a b) (c d) = (a b d) c (a b c) d, (A.169)
( ) = + , (A.170)
(A B) = A ( B) + B ( A) + (A )B + (B )A, (A.171)
=
2
, (A.172)
A = 0, (A.173)
( A) = A + A , (A.174)
(A B) = B A A B, (A.175)
= 0, (A.176)
( A) = ( A)
2
A, (A.177)
( A) = A + A, (A.178)
(A B) = ( B) A ( A) B + (B )A (A )B. (A.179)
A.24 Exercises
A.1. The position vectors of the four points A, B, C, and D are a, b, 3 a +2 b, and a 3 b, respectively. Express

AC,

DB,

BC, and
CD in terms of a and b.
A.2. Prove the trigonometric law of sines
sin a
A
=
sin b
B
=
sin c
C
using vector methods. Here, a, b, and c are the three angles of a plane triangle, and A, B, and C the lengths of the corre-
sponding opposite sides.
A.3. Demonstrate using vectors that the diagonals of a parallelogram bisect one another. In addition, show that if the diagonals
of a quadrilateral bisect one another then it is a parallelogram.
A.4. From the inequality
a b = a b cos a b
deduce the triangle inequality
a + b a + b.
A.5. Find the scalar product a b and the vector product a b when
(a) a = e
x
+ 3 e
y
e
z
, b = 3 e
x
+ 2 e
y
+ e
z
,
(b) a = e
x
2 e
y
+ e
z
, b = 2 e
x
+ e
y
+ e
z
.
A.6. Which of the following statements regarding the three general vectors a, b, and c are true?
(a) c (a b) = (b a) c.
(b) a (b c) = (a b) c.
(c) a (b c) = (a c) b (a b) c.
(d) d = a + b implies that (a b) d = 0.
(e) a c = b c implies that c a c b = c a b.
(f) (a b) (c b) = [b (c a)] b.
A.7. Prove that the length of the shortest straight-line from point a to the straight-line joining points b and c is
a b + b c + c a
b c
.
A.8. Identify the following surfaces:
(a) r = a,
(b) r n = b,
(c) r n = c r,
(d) r (r n) n = d.
Here, r is the position vector, a, b, c, and d are positive constants, and n is a xed unit vector.
A.9. Let a, b, and c be coplanar vectors related via
a + b + c = 0,
where , , and are not all zero. Show that the condition for the points with position vectors u a, v b, and wc to be colinear
is
u
+
v
+
w
= 0.
A.10. If p, q, and r are any vectors, demonstrate that a = q+r, b = r+ p, and c = p+ q are coplanar provided that = 1,
where , , and are scalars. Show that this condition is satised when a is perpendicular to p, b to q, and c to r.
A.11. The vectors a, b, and c are non-coplanar, and form a non-orthogonal vector base. The vectors A, B, and C, dened by
A =
b c
a b c
,
plus cyclic permutations, are said to be reciprocal vectors. Show that
a = (B C)/(A B C),
plus cyclic permutations.
252 FLUID MECHANICS
A.12. In the notation of the previous question, demonstrate that the plane passing through points a/, b/, and c/ is normal to
the direction of the vector
h = A + B + C.
In addition, show that the perpendicular distance of the plane from the origin is h
1
.
A.13. Evaluate
_
A dr for
A =
x e
x
+ y e
y
_
x
2
+ y
2
around the square whose sides are x = 0, x = a, y = 0, y = a.
A.14. Consider the following vector eld:
A(r) = (8 x
3
+ 3 x
2
y
2
, 2 x
3
y + 6 y, 6).
Is this eld conservative? Is it solenoidal? Is it irrotational? Justify your answers. Calculate
_
C
A dr, where the curve C is
a unit circle in the x-y plane, centered on the origin, and the direction of integration is clockwise looking down the z-axis.
A.15. Consider the following vector eld:
A(r) = (3 x y
2
z
2
y
2
, y
3
z
2
+ x
2
y, 3 x
2
x
2
z).
Is this eld conservative? Is it solenoidal? Is it irrotational? Justify your answers. Calculate the ux of A out of a unit sphere
centered on the origin.
A.16. Find the gradients of the following scalar functions of the position vector r = (x, y, z):
(a) k r,
(b) r
n
,
(c) r k
n
,
(d) cos(k r).
Here, k is a xed vector.
A.17. Find the divergences and curls of the following vector elds:
(a) k r,
(b) r
n
r,
(c) r k
n
(r k),
(d) a cos(k r).
Here, k and a are xed vectors.
A.18. Calculate
2
when = f (r). Find f if
2
= 0.
Cartesian Tensors 253
B Cartesian Tensors
B.1 Introduction
As we saw in Appendix A, many physical entities can be represented mathematically as either scalars or vectors,
depending on their transformation properties under rotation of the coordinate axes. However, it turns out that scalars
and vectors are particular types of a more general class of mathematical constructs called tensors. In fact, a scalar is a
tensor of order zero, and a vector is a tensor of order one. In uid mechanics, certain important physical entities (i.e.,
stress and rate of strain) are represented mathematically by tensors of order greater than one. It is therefore necessary
to supplement our investigation of uid mechanics with a brief discussion of the mathematics of tensors. For the sake
of simplicity, we shall limit this discussion to Cartesian coordinate systems. Tensors expressed in such coordinate
systems are known as Cartesian tensors.
B.2 Tensors and Tensor Notation
Let the Cartesian coordinates x, y, z be written as the x
i
, where i runs from 1 to 3. In other words, x = x
1
, y = x
2
, and
z = x
3
. Incidentally, in the following, any lowercase roman subscript (e.g., i, j, k) is assumed to run from 1 to 3. We
can also write the Cartesian components of a general vector v as the v
i
. In other words, v
x
= v
1
, v
y
= v
2
, and v
z
= v
3
. By
contrast, a scalar is represented as a variable without a subscript: e.g., a, . Thus, a scalarwhich is a tensor of order
zerois represented as a variable with zero subscripts, and a vectorwhich is a tensor of order oneis represented as
a variable with one subscript. It stands to reason, therefore, that a tensor of order two is represented as a variable with
two subscripts: e.g., a
i j
,
i j
. Moreover, an nth-order tensor is represented as a variable with n subscripts: e.g., a
i jk
is a
third-order tensor, and b
i jkl
a fourth-order tensor. Note that a general nth-order tensor has 3
n
independent components.
Now, the components of a second-order tensor are conveniently visualized as a two-dimensional matrix, just as the
components of a vector are sometimes visualized as a one-dimensional matrix. However, it is important to recognize
that an nth-order tensor is not simply another name for an n-dimensional matrix. A matrix is just an ordered set of
numbers. A tensor, on the other hand, is an ordered set of components that have specic transformation properties
under rotation of the coordinate axes. (See Section B.3.)
Consider two vectors a and b that are represented as a
i
and b
i
, respectively, in tensor notation. According to
Section A.6, the scalar product of these two vectors takes the form
a b = a
1
b
1
+ a
2
b
2
+ a
3
b
3
. (B.1)
The above expression can be written more compactly as
a b = a
i
b
i
. (B.2)
Here, we have made use of the Einstein summation convention, according to which, in an expression containing lower
case roman subscripts, any subscript that appears twice (and only twice) in any term of the expression is assumed to
be summed from 1 to 3 (unless stated otherwise). Thus, a
i
b
i
= a
1
b
1
+a
2
b
2
+a
3
b
3
, and a
i j
b
j
= a
i1
b
1
+a
i2
b
2
+a
i3
b
3
.
Note that when an index is summed it becomes a dummy index, and can be written as any (unique) symbol: i.e.,
a
i j
b
j
and a
ip
b
p
are equivalent. Moreover, only non-summed, or free, indices count toward the order of a tensor
expression. Thus, a
ii
is a zeroth-order tensor (because there are no free indices), and a
i j
b
j
is a rst-order tensor
(because there is only one free index). The process of reducing the order of a tensor expression by summing indices
is known as contraction. For example, a
ii
is a zeroth-order contraction of the second-order tensor a
i j
. Incidentally,
when two tensors are multiplied together without contraction the resulting tensor is called an outer product: e.g., the
second-order tensor a
i
b
j
is the outer product of the two rst-order tensors a
i
and b
i
. Likewise, when two tensors are
multiplied together in a manner that involves contraction then the resulting tensor is called an inner product: e.g.,
the rst-order tensor a
i j
b
j
is an inner product of the second-order tensor a
i j
and the rst-order tensor b
i
. Note, from
Equation (B.2), that the scalar product of two vectors is equivalent to the inner product of the corresponding rst-order
tensors.
254 FLUID MECHANICS
According to Section A.8, the vector product of two vectors a and b takes the form
(a b)
1
= a
2
b
3
a
3
b
2
, (B.3)
(a b)
2
= a
3
b
1
a
1
b
3
, (B.4)
(a b)
3
= a
1
b
2
a
2
b
1
(B.5)
in tensor notation. The above expression can be written more compactly as
(a b)
i
=
i jk
a
j
b
k
. (B.6)
Here,
i jk
=
_
_
+1 if i, j, k is an even permutation of 1, 2, 3
1 if i, j, k is an odd permutation of 1, 2, 3
0 otherwise
(B.7)
is known as the third-order permutation tensor (or, sometimes, the third-order Levi-Civita tensor). Note, in particular,
that
i jk
is zero if one of its indices is repeated: e.g.,
113
=
212
= 0. Furthermore, it follows from (B.7) that
i jk
=
jki
=
ki j
=
k ji
=
jik
=
ik j
. (B.8)
It is helpful to dene the second-order identity tensor (also known as the Kroenecker delta tensor),
i j
=
_
1 if i = j
0 otherwise
. (B.9)
It is easily seen that
i j
=
ji
, (B.10)
ii
= 3, (B.11)
ik

k j
=
i j
, (B.12)
i j
a
j
= a
i
, (B.13)
i j
a
i
b
j
= a
i
b
i
, (B.14)
i j
a
ki
b
j
= a
ki
b
i
, (B.15)
etc.
The following is a particularly important tensor identity:
i jk

ilm
=
jl
km

jm

kl
. (B.16)
In order to establish the validity of the above expression, let us consider the various cases that arise. As is easily seen,
the right-hand side of (B.16) takes the values
+1 if j = l and k = m j, (B.17)
1 if j = m and k = l j, (B.18)
0 otherwise. (B.19)
Moreover, in each product on the left-hand side, i has the same value in both factors. Thus, for a non-zero contribu-
tion, none of j, k, l, and m can have the same value as i (because each factor is zero if any of its indices are repeated).
Since a given subscript can only take one of three values (1, 2, or 3), the only possibilities that generate non-zero
contributions are j = l and k = m, or j = m and k = l, excluding j = k = l = m (since each factor would then have
repeated indices, and so be zero). Thus, the left-hand side reproduces (B.19), as well as the conditions on the indices
in (B.17) and (B.18). The left-hand side also reproduces the values in (B.17) and (B.18) since if j = l and k = m then
i jk
=
ilm
and the product
i jk

ilm
(no summation) is equal to +1, whereas if j = m and k = l then
i jk
=
iml
=
ilm
and the product
i jk

ilm
(no summation) is equal to 1. Here, use has been made of Equation (B.8). Hence, the validity
of the identity (B.16) has been established.
In order to illustrate the use of (B.16), consider the vector triple product identity (see Section A.11)
a (b c) = (a c) b (a b) c. (B.20)
In tensor notation, the left-hand side of this identity is written
[a (b c)]
i
=
i jk
a
j
(
klm
b
l
c
m
), (B.21)
where use has been made of Equation (B.6). Employing Equations (B.8) and (B.16), this becomes
[a (b c)]
i
=
ki j

klm
a
j
b
l
c
m
=
_
il

jm

im
jl
_
a
j
b
l
c
m
, (B.22)
which, with the aid of Equations (B.2) and (B.13), reduces to
[a (b c)]
i
= a
j
c
j
b
i
a
j
b
j
c
i
= [(a c) b (a b) c]
i
. (B.23)
Thus, we have established the validity of the vector identity (B.20). Moreover, our proof is much more rigorous than
that given earlier (in Section A.11).
B.3 Tensor Transformation
As we saw in Appendix A, scalars and vectors are dened according to their transformation properties under rotation
of the coordinate axes. In fact, a scalar is invarient under rotation of the coordinate axes. On the other hand, according
to Equations (A.49) and (B.6), the components of a general vector a transform under an innitesimal rotation of the
coordinate axes according to
a
i
= a
i
+
i jk

j
a
k
. (B.24)
Here, the a
i
are the components of the vector in the original coordinate system, the a
i
are the components in the rotated
coordinate system, and the latter system is obtained from the former via a combination of an innitesimal rotation
through an angle
1
about coordinate axis 1, an innitesimal rotation through an angle
2
about axis 2, and an
innitesimal rotation through an angle
3
about axis 3. These three rotations can take place in any order. Incidentally,
a nite rotation can be built up out of a great many innitesimal rotations, so if a vector transforms properly under an
innitesimal rotation of the coordinate axes then it will also transform properly under a nite rotation.
Equation (B.24) can also be written
a
i
= F
i j
a
j
, (B.25)
where
F
i j
=
i j

k

ki j
(B.26)
is a rotation matrix (which is not a tensor, since it is specic to the two coordinate systems it transforms between). To
rst-order in the
i
, Equation (B.25) can be inverted to give
a
i
= F
ji
a
j
. (B.27)
This follows because, to rst-order in the
i
,
F
ik
F
jk
= (
ik

l

lik
) (
jk

m

mjk
) =
ik

jk

l

jk

lik

m

ik

mjk
=
i j

l

li j

l

l ji
=
i j
, (B.28)
where the dummy index m has been relabeled l, and use has been made of Equations (B.8), (B.10), and (B.12).
Likewise, it is easily demonstrated that
F
ki
F
k j
=
i j
. (B.29)
256 FLUID MECHANICS
It can also be shown that, to rst-order in the
i
,
i jk
F
li
F
mj
F
nk
=
lmn
. (B.30)
This follows because
i jk
F
li
F
mj
F
nk
=
i jk
(
li

a

ali
) (
mj

b

bmj
) (
nk

c

cnk
)
=
i jk
_
li

mj
nk

a
_
ali

mj

nk
+
amj

li

nk
+
ank

li

mj
__
=
lmn

a
(
imn

ial
+
inl

iam
+
ilm

ian
)
=
lmn

a
(
ma

nl

ml

na
+
na

lm

nm

la
+
la

mn

ln

ma
)
=
lmn
. (B.31)
Here, there has been much relabeling of dummy indices, and use has been made of Equations (B.10) and (B.16). It
can similarly be shown that
i jk
F
il
F
jm
F
kn
=
lmn
. (B.32)
As a direct generalization of Equation (B.25), a second-order tensor transforms under rotation as
a
i j
= F
ik
F
jl
a
kl
, (B.33)
whereas a third-order tensor transforms as
a
i jk
= F
il
F
jm
F
kn
a
lmn
. (B.34)
The generalization to higher-order tensors is straight-forward. For the case of a scalar, which is a zeroth-order tensor,
the transformation rule is particularly simple: i.e.,
a
= a. (B.35)
By analogy with Equation (B.27), the inverse transform is exemplied by
a
i jk
= F
li
F
mj
F
nk
a
lmn
. (B.36)
Incidentally, since all tensors of the same order transform in the same manner, it immediately follows that two tensors
of the same order whose components are equal in one particular Cartesian coordinate system will have their compo-
nents equal in all coordinate systems that can be obtained from the original system via rotation of the coordinate axes.
In other words, if
a
i j
= b
i j
(B.37)
in one particular Cartesian coordinate system then
a
i j
= b
i j
(B.38)
in all Cartesian coordinate systems (with the same origin and system of units as the original system). Conversely,
it does not make sense to equate tensors of dierent order, since such an equation would only be valid in one par-
ticular coordinate system, and so could not have any physical signicance (since the laws of physics are coordinate
independent).
It can easily be shown that the outer product of two tensors transforms as a tensor of the appropriate order. Thus,
if
c
i jk
= a
i
b
jk
, (B.39)
and
a
i
= F
i j
a
j
, (B.40)
b
i j
= F
ik
F
jl
b
kl
, (B.41)
then
c
i jk
= a
i
b
jk
= F
il
a
l
F
jm
F
kn
b
mn
= F
il
F
jm
F
kn
a
l
b
mn
= F
il
F
jm
F
kn
c
lmn
, (B.42)
which is the correct transformation rule for a third-order tensor.
The tensor transformation rule can be combined with the identity (B.29) to show that the scalar product of two
vectors transforms as a scalar. Thus,
a
i
b
i
= F
i j
a
j
F
ik
b
k
= F
i j
F
ik
a
j
b
k
=
jk
a
j
b
k
= a
j
b
j
= a
i
b
i
, (B.43)
where use has been made of Equation (B.14). Again, the above proof is more rigorous than that given previously (in
Section A.6). The proof also indicates that the inner product of two tensors transforms as a tensor of the appropriate
order.
The result that both the inner and outer products of two tensors transform as tensors of the appropriate order is
known as the product rule. Closely related to this rule is the so-called quotient rule, according to which if (say)
c
i j
= a
ik
b
jk
, (B.44)
where b
jk
is an arbitrary tensor, and c
i j
transforms as a tensor under all rotations of the coordinate axes, then a
ik
which can be thought of as the quotient of c

i j
and b
jk
also transforms as a tensor. The proof is as follows:
a
ik
b
jk
= c
i j
= F
il
F
jm
c
lm
= F
il
F
jm
a
lk
b
mk
= F
il
F
jm
a
lk
F
pm
F
qk
b
pq
= F
il
F
qk
a
lk
b
jq
= F
il
F
km
a
lm
b
jk
, (B.45)
where use has been made of the fact that c
i j
and b
i j
transform as tensors, as well as Equation (B.28). Rearranging, we
obtain
(a
ik
F
il
F
km
a
lm
) b
jk
= 0. (B.46)
However, the b
i j
are arbitrary, so the above equation can only be satised, in general, if
a
ik
= F
il
F
km
a
lm
, (B.47)
which is the correct transformation rule for a tensor. Incidentally, the quotient rule applies to any type of valid tensor
product.
The components of the second-order identity tensor,
i j
, have the special property that they are invariant under
rotation of the coordinate axes. This follows because
i j
= F
ik
F
jl

kl
= F
ik
F
jk
=
i j
, (B.48)
where use has been made of Equation (B.28). The components of the third-order permutation tensor,
i jk
, also have
this special property. This follows because
i jk
= F
il
F
jm
F
ln

lmn
=
i jk
, (B.49)
where use has been made of Equation (B.31). The fact that
i jk
transforms as a proper third-order tensor immediately
implies, from the product rule, that the vector product of two vectors transforms as a proper vector: i.e.,
i jk
a
j
b
k
is a
rst-order tensor provided that a
i
and b
i
are both rst-order tensors. This proof is much more rigorous that that given
earlier (in Section A.8).
B.4 Tensor Fields
We saw in Appendix A that a scalar eld is a set of scalars associated with every point in space: e.g., (x), where
x = (x
1
, x
2
, x
3
) is a position vector. We also saw that a vector eld is a set of vectors associated with every point in
258 FLUID MECHANICS
space: e.g., a
i
(x). It stands to reason, then, that a tensor eld is a set of tensors associated with every point in space:
e.g., a
i j
(x). It immediately follows that a scalar eld is a zeroth-order tensor eld, and a vector eld is a rst-order
tensor eld.
Most tensor eld encountered in physics are smoothly varying and dierentiable. Consider the rst-order tensor
eld a
i
(x). The various partial derivatives of the components of this eld with respect to the Cartesian coordinates x
i
are written
a
i
x
j
. (B.50)
Moreover, this set of derivatives transform as the components of a second-order tensor. In order to demonstrate this,
we need the transformation rule for the x
i
, which is the same as that for a rst-order tensor: i.e.,
x
i
= F
i j
x
j
. (B.51)
Thus,
x
i
x
j
= F
i j
. (B.52)
It is also easily shown that
x
i
x
j
= F
ji
. (B.53)
Now,
a
i
x
j
=
a
i
x
k
x
k
x
j
=
(F
il
a
l
)
x
k
F
jk
= F
il
F
jk
a
l
x
k
, (B.54)
which is the correct transformation rule for a second-order tensor. Here, use has been made of the chain rule, as well
as Equation (B.53). [Note, from Equation (B.26), that the F
i j
are not functions of position.] It follows, from the above
argument, that dierentiating a tensor eld increases its order by one: e.g., a
i j
/x
k
is a third-order tensor. The only
exception to this rule occurs when dierentiation and contraction are combined. Thus, a
i j
/x
j
is a rst-order tensor,
since it only contains a single free index.
The gradient (see Section A.18) of a scalar eld is an example of a rst-order tensor eld (i.e., a vector eld):
()
i
=
x
i
. (B.55)
The divergence (see Section A.20) of a vector eld is a contracted second-order tensor eld that transforms as a scalar:
a =
a
i
x
i
. (B.56)
Finally, the curl (see Section A.22) of a vector eld is a contracted fth-order tensor that transforms as a vector
( a)
i
=
i jk
a
k
x
j
. (B.57)
The above denitions can be used to prove a number of useful results. For instance,
( )
i
=
i jk
x
j
_

x
k
_
=
i jk
x
j
x
k
= 0, (B.58)
which follows from symmetry because
ik j
=
i jk
whereas
2
/x
k
x
j
=
2
/x
j
x
k
. Likewise,
( a) =
x
i
_
i jk
a
k
x
j
_
=
i jk
a
k
x
i
x
j
= 0, (B.59)
which again follows from symmetry. As a nal example,
(a b) =
x
i
_
i jk
a
j
b
k
_
=
i jk
a
j
x
i
b
k
+
i jk
a
j
b
k
x
i
= b
i

i jk
a
k
x
j
a
i

i jk
b
k
x
j
= b ( a) a ( b). (B.60)
According to the divergence theorem (see Section A.20),
_
S
a
i
dS
i
=
_
V
a
i
x
i
dV, (B.61)
where S is a closed surface surrounding the volume V. The above theorem is easily generalized to give, for example,
_
S
a
i j
dS
i
=
_
V
a
i j
x
i
dV, (B.62)
or
_
S
a
i j
dS
j
=
_
V
a
i j
x
j
dV, (B.63)
or even
_
S
a dS
i
=
_
V
a
x
i
dV. (B.64)
B.5 Isotropic Tensors
A tensor which has the special property that its components take the same value in all Cartesian coordinate systems is
called an isotropic tensor. We have already encountered two such tensors: namely, the second-order identity tensor,
i j
,
and the third-order permutation tensor,
i jk
. Of course, all scalars are isotropic. Moreover, as is easily demonstrated,
there are no isotropic vectors (other than the null vector). It turns out that the most general isotropic Cartesian tensors
of second-, third-, and fourth-order are
i j
,
i jk
, and
i j
kl
+
ik

jl
+
il

jk
, respectively, where , , , , and
are scalars. Let us prove these important results.
1
The most general second-order isotropic tensor, a
i j
, is such that
a
i j
= F
ip
F
jq
a
pq
= a
i j
(B.65)
for arbitrary rotations of the coordinate axes. It follows from Equation (B.24) that, to rst-order in the
i
,
m
_
mis
a
s j
+
mjs
a
is
_
= 0. (B.66)
However, the
i
are arbitrary, so we can write
mis
a
s j
+
mjs
a
is
= 0. (B.67)
Let us multiply by
mik
. With the aid of Equation (B.16), we obtain
(
ii

ks

is

ki
) a
s j
+ (
i j
ks

is

k j
) a
is
= 0, (B.68)
which reduces to
2 a
i j
+ a
ji
= a
ss

i j
. (B.69)
Interchanging the labels i and j, and then taking the dierence between the two equations thus obtained, we deduce
that
a
i j
= a
ji
. (B.70)
1
This proof is adapted from P.G. Hodge, Jr., American Mathematical Monthly 68, 793 (1961).
260 FLUID MECHANICS
Hence,
a
i j
=
a
ss
3

i j
, (B.71)
which implies that
a
i j
=
i j
. (B.72)
For the case of an isotropic third-order tensor, Equation (B.67) generalizes to
mis
a
s jk
+
mjs
a
isk
+
mks
a
i js
= 0. (B.73)
Multiplying by
mit
,
mjt
, and
mkt
, and then setting t = i, t = j, and t = k, respectively, we obtain
2 a
i jk
+ a
jik
+ a
k ji
= a
ssk
i j
+ a
s js
ik
, (B.74)
2 a
i jk
+ a
jik
+ a
ik j
= a
ssk
i j
+ a
iss
jk
, (B.75)
2 a
i jk
+ a
k ji
+ a
ik j
= a
s js
ik
+ a
iss
jk
, (B.76)
respectively. However, multiplying the above equations by
jk
,
ik
, and
i j
, and then setting i = i, j = i, and k = i,
respectively, we obtain
2 a
iss
+ a
sis
+ a
ssi
= a
ssi
+ a
sis
, (B.77)
2 a
sis
+ a
iss
+ a
ssi
= a
ssi
+ a
iss
, (B.78)
2 a
ssi
+ a
iss
+ a
sis
= a
sis
+ a
iss
, (B.79)
respectively, which implies that
a
iss
= a
sis
= a
ssi
= 0. (B.80)
Hence, we deduce that
2 a
i jk
+ a
jik
+ a
k ji
= 0, (B.81)
2 a
i jk
+ a
jik
+ a
ik j
= 0, (B.82)
2 a
i jk
+ a
k ji
+ a
ik j
= 0. (B.83)
The solution to the above equation must satisfy
a
ik j
= a
jik
= a
k ji
= a
i jk
. (B.84)
This implies, from Equation (B.8), that
a
i jk
=
i jk
. (B.85)
For the case of an isotropic fourth-order tensor, Equation (B.73) generalizes to
mis
a
s jkl
+
mjs
a
iskl
+
mks
a
i jsl
+
mls
a
i jks
= 0. (B.86)
Multiplying the above by
mit
,
mjt
,
mkt
,
mlt
, and then setting t = i, t = j, t = k, and t = l, respectively, we obtain
2 a
i jkl
+ a
jikl
+ a
k jil
+ a
l jki
= a
sskl
i j
+ a
s jsl
ik
+ a
s jks
il
, (B.87)
2 a
i jkl
+ a
jikl
+ a
ik jl
+ a
il jk
= a
sskl
i j
+ a
isks

jl
+ a
issl
jk
, (B.88)
2 a
i jkl
+ a
k jil
+ a
ik jl
+ a
i jlk
= a
i jss

kl
+ a
s jsl
ik
+ a
issl
jk
, (B.89)
2 a
i jkl
+ a
l jki
+ a
ilk j
+ a
i jlk
= a
i jss

kl
+ a
isks

jl
+ a
s jks
il
, (B.90)
respectively. Now, if a
i jkl
is an isotropic fourth-order tensor then a
sskl
is clearly an isotropic second-order tensor, which
means that is a multiple of
kl
. This, and similar arguments, allows us to deduce that
a
sskl
=
kl
, (B.91)
a
s jsl
=
jl
, (B.92)
a
s jks
=
jk
. (B.93)
Let us assume, for the moment, that
a
i jss
= a
ssi j
, (B.94)
a
isks
= a
sisk
, (B.95)
a
issl
= a
sils
. (B.96)
Thus, we get
2 a
i jkl
+ a
jikl
+ a
k jil
+ a
l jki
=
i j
kl
+
ik

jl
+
il

jk
, (B.97)
2 a
i jkl
+ a
jikl
+ a
ik jl
+ a
ilk j
=
i j
kl
+
ik

jl
+
il

jk
, (B.98)
2 a
i jkl
+ a
k jil
+ a
ik jl
+ a
i jlk
=
i j
kl
+
ik

jl
+
il

jk
, (B.99)
2 a
i jkl
+ a
l jki
+ a
ilk j
+ a
i jlk
=
i j
kl
+
ik

jl
+
il

jk
. (B.100)
Relations of the form
a
i jkl
= a
jilk
= a
kli j
= a
lk ji
(B.101)
can be obtained by subtracting the sum of one pair of Equations (B.97)(B.100) from the sum of the other pair. These
relations justify Equations (B.94)(B.96). Equations (B.97) and (B.101) can be combined to give
2 a
i jkl
+ (a
i jlk
+ a
ik jl
+ a
ilk j
) =
i j

kl
+
ik

jl
+
il

jk
, (B.102)
2 a
ikl j
+ (a
ik jl
+ a
ilk j
+ a
i jlk
) =
ik

jl
+
il

jk
+
i j

kl
, (B.103)
2 a
il jk
+ (a
ilk j
+ a
i jlk
+ a
ik jl
) =
il

jk
+
i j

kl
+
ik

jl
. (B.104)
The latter two equations are obtained from the rst via cyclic permutation of j, k, and l, with i remaining unchanged.
Summing Equations (B.102)(B.104), we get
2 (a
i jkl
+ a
ikl j
+ a
il jk
) + 3 (a
i jlk
+ a
ik jl
+ a
ilk j
) = ( + + ) (
i j

kl
+
ik

jl
+
il

jk
). (B.105)
It follows from symmetry that
a
i jkl
+ a
ikl j
+ a
il jk
= a
i jlk
+ a
ik jl
+ a
ilk j
=
1
5
( + + ) (
i j
kl
+
ik

jl
+
il

jk
). (B.106)
This can be seen by swapping the indices k and l in the above expression. Finally, substitution into Equation (B.102)
yields
a
i jkl
=
i j
kl
+
ik

jl
+
il

jk
, (B.107)
where
= (4 )/10, (B.108)
= (4 )/10, (B.109)
= (4 )/10. (B.110)
B.6 Exercises
B.1. Show that a general second-order tensor a
i j
can be decomposed into three tensors
a
i j
= u
i j
+ v
i j
+ s
i j
,
where u
i j
is symmetric (i.e., u
ji
= u
i j
) and traceless (i.e., u
ii
= 0), v
i j
is isotropic, and s
i j
only has three independent
components.
B.2. Use tensor methods to establish the following vector identities:
262 FLUID MECHANICS
(a) a b c = a b c = b a c.
(b) (a b) c = (a c) b (c b) a.
(c) (a b) (c d) = (a c) (b d) (a d) (b c).
(d) (a b) (c d) = (a b d) c (a b c) d.
(e) (a) = a + a .
(f) (a) = a + a.
(g) (a b) = (b ) a (a ) b + ( b) a ( a) b.
(h) (a a) = 2 a ( a) + 2 (a ) a.
(i) ( a) = ( a)
2
a.
Here, [(b )a]
i
= b
j
a
i
/x
j
, and (
2
a)
i
=
2
a
i
.
B.3. A quadric surface has an equation of the form
a x
2
1
+ b x
2
2
+ c x
2
3
+ 2 f x
1
x
2
+ 2 g x
1
x
3
+ 2 h x
2
x
3
= 1.
Show that the coecients in the above expression transform under rotation of the coordinate axes like the components of a
symmetric second-order tensor. Hence, demonstrate that the equation for the surface can be written in the form
x
i
T
i j
x
j
= 1,
where the T
i j
are the components of the aforementioned tensor.
B.4. The determinant of a second-order tensor A
i j
is dened
det(A) =
i jk
A
i1
A
j2
A
k3
.
(a) Show that
det(A) =
i jk
A
1i
A
2j
A
3k
is an alternative, and entirely equivalent, denition.
(b) Demonstrate that det(A) is invariant under rotation of the coordinate axes.
(c) Suppose that C
i j
= A
ik
B
k j
. Show that
det(C) = det(A) det(B).
B.5. If
A
i j
x
j
= x
i
then and x
j
are said to be eigenvalues and eigenvectors of the second-order tensor A
i j
, respectively. The eigenvalues of A
i j
are calculated by solving the related homogeneous matrix equation
(A
i j

i j
) x
j
= 0.
Now, it is a standard result in linear algebra that an equation of the above form only has a non-trivial solution when
det(A
i j

i j
) = 0.
Demonstrate that the eigenvalues of A
i j
satisfy the cubic polynomial
3
tr(A)
2
+ (A) det(A) = 0,
where tr(A) = A
ii
and (A) = (A
ii
A
j j
A
i j
A
ji
)/2. Hence, deduce that A
i j
possesses three eigenvalues
1
,
2
, and
3
(say).
Moreover, show that
tr(A) =
1
+
2
+
3
,
det(A) =
1

2

3
.
B.6. Suppose that A
i j
is a (real) symmetric second-order tensor: i.e., A
ji
= A
i j
.
(a) Demonstrate that the eigenvalues of A
i j
are all real, and that the eigenvectors can be chosen to be real.
(b) Show that eigenvectors of A
i j
corresponding to dierent eigenvalues are orthogonal to one another. Hence, deduce
that the three eigenvectors of A
i j
are, or can be chosen to be, mutually orthogonal.
(c) Demonstrate that A
i j
takes the diagonal form A
i j
=
i

i j
(no sum) in a Cartesian coordinate system in which the
coordinate axes are each parallel to one of the eigenvectors.
B.7. In an isotropic elastic medium under stress the displacement u
i
satises
i j
x
j
=
2
u
i
t
2
,
i j
= c
i jkl
1
2
_
u
k
x
l
+
u
l
x
k
_
,
where
i j
is the stress tensor (note that
ji
=
i j
), the mass density (which is a uniform constant), and
c
i jkl
= K
i j

kl
+ [
ik

jl
+
il

jk
(2/3)
i j

kl
].
the isotropic stiness tensor. Here, K and are the bulk modulus and shear modulus of the medium, respectively. Show
that the divergence and the curl of u both satisfy wave equations. Furthermore, demonstrate that the characteristic wave
velocities of the divergence and curl waves are [(K + 4/3)/]
1/2
and (/)
1/2
, respectively.
264 FLUID MECHANICS
Non-Cartesian Coordinates 265
C Non-Cartesian Coordinates
C.1 Introduction
In uid mechanics non-Cartesian coordinates are often used to exploit the symmetry of particular uid systems. For
example, it is convenient to employ cylindrical coordinates to describe systems possessing axial symmetry. In this
Appendix, we investigate a particularly useful class of non-Cartesian coordinates known as orthogonal curvilinear
coordinates. The two most commonly occurring examples of this class in uid mechanics are cylindrical and spherical
coordinates. (Note, incidentally, that the Einstein summation convention is not used in this Appendix.)
C.2 Orthogonal Curvilinear Coordinates
Let x
1
, x
2
, x
3
be a set of standard right-handed Cartesian coordinates. Furthermore, let u
1
(x
1
, x
2
, x
3
), u
2
(x
1
, x
2
, x
3
),
u
3
(x
1
, x
2
, x
3
) be three independent functions of these coordinates which are such that each unique triplet of x
1
, x
2
, x
3
values is associated with a unique triplet of u
1
, u
2
, u
3
values. It follows that u
1
, u
2
, u
3
can be used as an alternative set
of coordinates to distinguish dierent points in space. Since the surfaces of constant u
1
, u
2
, and u
3
are not generally
parallel planes, but rather curved surfaces, this type of coordinate system is termed curvilinear.
Let h
1
= u
1
1
, h
2
= u
2
1
, and h
3
= u
3
1
. It follows that e
1
= h
1
u
1
, e
2
= h
2
u
2
, and e
3
= h
3
u
3
are a set
of unit basis vectors which are normal to surfaces of constant u
1
, u
2
, and u
3
, respectively, at all points in space. Note,
however, that the direction of these basis vectors is generally a function of position. Suppose that the e
i
, where i runs
from 1 to 3, are mutually orthogonal at all points in space: i.e.,
e
i
e
j
=
i j
. (C.1)
In this case, u
1
, u
2
, u
3
are said to constitute an orthogonal coordinate system. Suppose, further, that
e
1
e
2
e
3
= 1 (C.2)
at all points in space, so that u
1
, u
2
, u
3
also constitute a right-handed coordinate system. It follows that
e
i
e
j
e
k
=
i jk
. (C.3)
Finally, a general vector A, associated with a particular point in space, can be written
A =
i
A
i
e
i
, (C.4)
where the e
i
are the local basis vectors of the u
1
, u
2
, u
3
system, and A
i
= e
i
A is termed the ith component of A in this
system.
Consider two neighboring points in space whose coordinates in the u
1
, u
2
, u
3
system are u
1
, u
2
, u
3
and u
1
+ du
1
,
u
2
+ du
2
, u
3
+du
3
. It is easily shown that the vector directed from the rst to the second of these points takes the form
dx =
du
1
u
1
e
1
+
du
2
u
2
e
2
+
du
3
u
3
e
3
=
i
h
i
du
i
e
i
. (C.5)
Hence, from (C.1), an element of length (squared) in the u
1
, u
2
, u
3
coordinate system is written
dx dx =
i
h
2
i
du
2
i
. (C.6)
Here, the h
i
, which are generally functions of position, are known as the scale factors of the system. Elements of area
that are normal to e
1
, e
2
, and e
3
, at a given point in space, take the form dS
1
= h
2
h
3
du
2
du
3
, dS
2
= h
1
h
3
du
1
du
3
,
266 FLUID MECHANICS
and dS
3
= h
1
h
2
du
1
du
2
, respectively. Finally, an element of volume, at a given point in space, is written dV =
h du
1
du
2
du
3
, where
h = h
1
h
2
h
3
. (C.7)
Note that [see Equation (A.176)]
u
i
= 0, (C.8)
and

_
_
h
2
i
h
u
i
_
_
= 0. (C.9)
The latter result follows from Equations (A.175) and (A.176) because (h
2
1
/h) u
1
= u
2
u
3
, etc. Finally, it is easily
demonstrated from (C.1) and (C.3) that
u
i
u
j
= h
2
i

i j
, (C.10)
u
i
u
j
u
k
= h
1
i jk
. (C.11)
Consider a scalar eld (u
1
, u
2
, u
3
). It follows from the chain rule, and the relation e
i
= h
i
u
i
, that
=
u
i
u
i
=
i
1
h
i
u
i
e
i
. (C.12)
Hence, the components of in the u
1
, u
2
, u
3
coordinate system are
()
i
=
1
h
i
u
i
. (C.13)
Consider a vector eld A(u
1
, u
2
, u
3
). We can write
A =
i
(A
i
e
i
) =
i
(h
i
A
i
u
i
) =
_
h
h
i
A
i
h
2
i
h
u
i
_
_
=
i
h
2
i
h
u
i

_
h
h
i
A
i
_
=
i
1
h
u
i
_
h
h
i
A
i
_
, (C.14)
where use has been made of Equations (A.174), (C.9), and (C.10). Thus, the divergence of Ain the u
1
, u
2
, u
3
coordinate
system takes the form
A =
i
1
h
u
i
_
h
h
i
A
i
_
. (C.15)
We can write
A =
k
(A
k
e
k
) =
k
(h
k
A
k
u
k
) =
k
(h
k
u
k
) u
k
=
j,k
(h
k
A
k
)
u
j
u
j
u
k
, (C.16)
where use has been made of Equations (A.178), (C.8), and (C.12). It follows from (C.11) that
( A)
i
= e
i
A =
j,k
h
i
(h
k
A
k
)
u
j
u
i
u
j
u
k
=
j,k
i jk
h
i
h
(h
k
A
k
)
u
j
. (C.17)
Hence, the components of A in the u
1
, u
2
, u
3
coordinate system are
( A)
i
=
j,k
i jk
h
i
h
(h
k
A
k
)
u
j
. (C.18)
Now,
2
= [see (A.172)], so Equations (C.12) and (C.15) yield the following expression for
2
in the u
1
,
u
2
, u
3
coordinate system:
2
=
i
1
h
u
i
_
_
h
h
2
i
u
i
_
_
. (C.19)
The vector identities (A.171) and (A.179) can be combined to give the following expression for (A )B that is
valid in a general coordinate system:
(A )B =
1
2
[(A B) (A B) ( A) B + ( B) A
A ( B) B ( A)] . (C.20)
Making use of Equations (C.13), (C.15), and (C.18), as well as the easily demonstrated results
A B =
i
A
i
B
i
, (C.21)
A B =
j,k
i jk
A
j
B
k
, (C.22)
and the tensor identity (B.16), Equation (C.20) reduces (after a great deal of tedious algebra) to the following expres-
sion for the components of (A )B in the u
1
, u
2
, u
3
coordinate system:
[(A ) B]
i
=
j
_
A
j
h
j
B
i
u
j
A
j
B
j
h
i
h
j
h
j
u
i
+
A
i
B
j
h
i
h
j
h
i
u
j
_
. (C.23)
Note, incidentally, that the commonly quoted result [(A )B]
i
= A B
i
is only valid in Cartesian coordinate systems
(for which h
1
= h
2
= h
3
= 1).
Let us dene the gradient A of a vector eld A as the tensor whose components in a Cartesian coordinate system
take the form
(A)
i j
=
A
i
x
j
. (C.24)
In an orthogonal curvilinear coordinate system, the above expression generalizes to
(A)
i j
= [(e
j
) A]
i
. (C.25)
It thus follows from (C.23), and the relation (e
i
)
j
= e
i
e
j
=
i j
, that
(A)
i j
=
1
h
j
A
i
u
j
A
j
h
i
h
j
h
j
u
i
+
i j
k
A
k
h
i
h
k
h
i
u
k
. (C.26)
The vector identity (A.177) yields the following expression for
2
A that is valid in a general coordinate system:
2
A = ( A) ( A). (C.27)
Making use of Equations (C.15), (C.18), and (C.19), as well as (C.21) and (C.22), and the tensor identity (B.16), the
above equation reduces (after a great deal of tedious algebra) to the following expression for the components of
2
A
in the u
1
, u
2
, u
3
coordinate system:
(
2
A)
i
=
2
A
i
+
j
_
2
h
i
h
j
_
1
h
i
h
i
u
j
u
i
1
h
j
h
j
u
i
u
j
_
A
j
+
h
h
i
h
2
j
_
_
A
j
h
2
i
h
j
u
i
u
j
_
_
h
2
i
h
_
A
i
h
2
j
h
i
u
j
u
j
_
_
h
2
j
h
_
_
_
_
+
A
j
h
i
h
h
3
j
_
_
1
h
j
h
j
u
i
u
j
_
_
h
2
j
h
_
_
+
h
h
2
j
u
i
_
_
h
2
j
h
_
u
j
_
_
h
2
j
h
_

2
u
i
u
j
_
_
h
2
j
h
_
_
_
A
i
h
i
h
2
j
_
_
2
h
i
_
h
i
u
j
_
2

2
h
i
u
2
j
_
_
_
_
. (C.28)
268 FLUID MECHANICS
O
y
x
e
r
e
r
z
Figure C.1: Cylindrical coordinates.
Note, again, that the commonly quoted result (
2
A)
i
=
2
A
i
is only valid in Cartesian coordinate systems (for which
h
1
= h
2
= h
3
= h = 1).
C.3 Cylindrical Coordinates
In the cylindrical coordinate system, u
1
= r, u
2
= , and u
3
= z, where r =
_
x
2
+ y
2
, = tan
1
(y/x), and x, y, z
are standard Cartesian coordinates. Thus, r is the perpendicular distance from the z-axis, and the angle subtended
between the projection of the radius vector (i.e., the vector connecting the origin to a general point in space) onto the
x-y plane and the x-axis. See Figure C.1.
A general vector A is written
A = A
r
e
r
+ A
+ A
z
e
z
, (C.29)
where e
r
= r/r, e
= /, and e
z
= z/z. See Figure C.1. Of course, the unit basis vectors e
r
, e
, and e
z
are
mutually orthogonal, so A
r
= A e
r
, etc.
As is easily demonstrated, an element of length (squared) in the cylindrical coordinate system takes the form
dx dx = dr
2
+ r
2
d
2
+ dz
2
. (C.30)
Hence, comparison with Equation (C.6) reveals that the scale factors for this system are
h
r
= 1, (C.31)
h
= r, (C.32)
h
z
= 1. (C.33)
Thus, surface elements normal to e
r
, e
, and e
z
are written
dS
r
= r d dz, (C.34)
dS
= dr dz, (C.35)
dS
z
= r dr d, (C.36)
respectively, whereas a volume element takes the form
dV = r dr d dz. (C.37)
According to Equations (C.13), (C.15), and (C.18), gradient, divergence, and curl in the cylindrical coordinate
system are written
=
r
e
r
+
1
r
z
e
z
, (C.38)
A =
1
r
r
(r A
r
) +
1
r
A
+
A
z
z
, (C.39)
A =
_
1
r
A
z
z
_
e
r
+
_
A
r
z

A
z
r
_
e
+
_
1
r
r
(r A
)
1
r
A
r
_
e
z
,
(C.40)
respectively. Here, (r) is a general scalar eld, and A(r) a general vector eld.
According to Equation (C.19), when expressed in cylindrical coordinates, the Laplacian of a scalar eld becomes
2
=
1
r
r
_
r

r
_
+
1
r
2
2
+
z
2
. (C.41)
Moreover, from Equation (C.23), the components of (A )A in the cylindrical coordinate system are
[(A )A]
r
= A A
r

A
2
r
, (C.42)
[(A )A]
= A A
+
A
r
A
r
, (C.43)
[(A )A]
z
= A A
z
. (C.44)
Let us dene the symmetric gradient tensor
A =
1
2
_
A + (A)
T
_
. (C.45)
Here, the superscript T denotes a transpose. Thus, if the i j element of some second-order tensor S is S
i j
then the
corresponding element of S
T
is S
ji
. According to Equation (C.26), the components of

A in the cylindrical coordinate
system are
(
A)
rr
=
A
r
r
, (C.46)
(
A)
=
1
r
A
+
A
r
r
, (C.47)
(
A)
zz
=
A
z
z
, (C.48)
(
A)
r
= (
A)
r
=
1
2
_
1
r
A
r
+
A
r

A
r
_
, (C.49)
(
A)
rz
= (
A)
zr
=
1
2
_
A
r
z
+
A
z
r
_
, (C.50)
(
A)
z
= (
A)
z
=
1
2
_
A
z
+
1
r
A
z
_
. (C.51)
Finally, from Equation (C.28), the components of
2
A in the cylindrical coordinate system are
(
2
A)
r
=
2
A
r

A
r
r
2

2
r
2
A
, (C.52)
(
2
A)
=
2
A
+
2
r
2
A
r
r
2
, (C.53)
(
2
A)
z
=
2
A
z
. (C.54)
270 FLUID MECHANICS
r
O
x
z
y
Figure C.2: Spherical coordinates.
C.4 Spherical Coordinates
In the spherical coordinate system, u
1
= r, u
2
= , and u
3
= , where r =
_
x
2
+ y
2
+ z
2
, = cos
1
(z/r), =
tan
1
(y/x), and x, y, z are standard Cartesian coordinates. Thus, r is the length of the radius vector, the angle
subtended between the radius vector and the z-axis, and the angle subtended between the projection of the radius
vector onto the x-y plane and the x-axis. See Figure C.2.
A general vector A is written
A = A
r
e
r
+ A
+ A
, (C.55)
where e
r
= r/r, e
= /, and e
= /. See Figure C.2. Of course, the unit vectors e

r
, e
, and e
are
mutually orthogonal, so A
r
= A e
r
, etc.
As is easily demonstrated, an element of length (squared) in the spherical coordinate system takes the form
dx dx = dr
2
+ r
2
d
2
+ r
2
sin
2
d
2
. (C.56)
Hence, comparison with Equation (C.6) reveals that the scale factors for this system are
h
r
= 1, (C.57)
h
= r, (C.58)
h
= r sin . (C.59)
Thus, surface elements normal to e
r
, e
, and e
are written
dS
r
= r
2
sin d d, (C.60)
dS
= r sin dr d, (C.61)
dS
= r dr d, (C.62)
respectively, whereas a volume element takes the form
dV = r
2
sin dr d d. (C.63)
According to Equations (C.13), (C.15), and (C.18), gradient, divergence, and curl in the spherical coordinate
system are written
=
r
e
r
+
1
r
+
1
r sin
, (C.64)
A =
1
r
2
r
(r
2
A
r
) +
1
r sin
(sin A
) +
1
r sin
A
, (C.65)
A =
_
1
r sin
(sin A
)
1
r sin
A
_
e
r
+
_
1
r sin
A
r
1
r
r
(r A
)
_
e
+
_
1
r
r
(r A
)
1
r
A
r
_
e
, (C.66)
respectively. Here, (r) is a general scalar eld, and A(r) a general vector eld.
According to Equation (C.19), when expressed in spherical coordinates, the Laplacian of a scalar eld becomes
2
=
1
r
2
r
_
r
2

r
_
+
1
r
2
sin
_
sin

_
+
1
r
2
sin
2
2
. (C.67)
Moreover, from Equation (C.23), the components of (A )A in the spherical coordinate system are
[(A )A]
r
= A A
r

A
2
+ A
2
r
, (C.68)
[(A )A]
= A A
+
A
r
A
cot A
2
r
, (C.69)
[(A )A]
= A A
+
A
r
A
+ cot A
r
. (C.70)
Now, according to Equation (C.26), the components of

A in the spherical coordinate system are
(
A)
rr
=
A
r
r
, (C.71)
(
A)
=
1
r
A
+
A
r
r
, (C.72)
(
A)
=
1
r sin
A
+
A
r
r
+
cot A
r
, (C.73)
(
A)
r
= (
A)
r
=
1
2
_
1
r
A
r
+
A
r

A
r
_
, (C.74)
(
A)
r
= (
A)
r
=
1
2
_
1
r sin
A
r
+
A
r

A
r
_
, (C.75)
(
A)
= (
A)
=
1
2
_
1
r sin
A
+
1
r
A
cot A
r
_
. (C.76)
Finally, from Equation (C.28), the components of
2
A in the spherical coordinate system are
(
2
A)
r
=
2
A
r

2A
r
r
2

2
r
2
A
2 cot A
r
2

2
r
2
sin
A
, (C.77)
(
2
A)
=
2
A
+
2
r
2
A
r
r
2
sin
2
2
r
2
sin
A
, (C.78)
272 FLUID MECHANICS
(
2
A)
=
2
A
r
2
sin
2
+
2
r
2
sin
2
A
r
+
2 cot
r
2
sin
A
. (C.79)
C.5 Exercises
C.1. Find the Cartesian components of the basis vectors e
r
, e
, and e
z
of the cylindrical coordinate system. Verify that the vectors
are mutually orthogonal. Do the same for the basis vectors e
r
, e
, and e
of the spherical coordinate system.

C.2. Use cylindrical coordinates to prove that the volume of a right cylinder of radius a and length l is a
2
l. Demonstrate that
the moment of inertia of a uniform cylinder of mass M and radius a about its symmetry axis is (1/2) Ma
2
.
C.3. Use spherical coordinates to prove that the volume of a sphere of radius a is (4/3) a
3
. Demonstrate that the moment of
inertia of a uniform sphere of mass M and radius a about an axis passing through its center is (2/5) Ma
2
.
C.4. For what value(s) of n is (r
n
e
r
) = 0, where r is a spherical coordinate?
C.5. For what value(s) of n is (r
n
e
r
) = 0, where r is a spherical coordinate?
C.6. (a) Find a vector eld F = F
r
(r) e
r
satisfying F = r
m
for m 0. Here, r is a spherical coordinate.
(b) Use the divergence theorem to show that
_
V
r
m
dV =
1
m+ 3
_
S
r
m+1
e
r
dS,
where V is a volume enclosed by a surface S .
(c) Use the above result (for m = 0) to demonstrate that the volume of a right cone is one third the volume of the right
cylinder having the same base and height.
C.7. The electric eld generated by a z-directed electric dipole of moment p, located at the origin, is
E(r) =
1
4
0
_
3 (e
r
p) e
r
p
r
3
_
,
where p = p e
z
, and r is a spherical coordinate. Find the components of E(r) in the spherical coordinate system. Calculate
E and E.
C.8. Show that the parabolic cylindrical coordinates u, v, z, dened by the equations x = (u
2
v
2
)/2, y = u v, z = z, where x, y, z
are Cartesian coordinates, are orthogonal. Find the scale factors h
u
, h
v
, h
z
. What shapes are the u = constant and v = constant
surfaces? Write an expression for
2
f in parabolic cylindrical coordinates.
C.9. Show that the elliptic cylindrical coordinates , , z, dened by the equations x = cosh cos , y = sinh sin , z = z, where
x, y, z are Cartesian coordinates, and 0 , < , are orthogonal. Find the scale factors h
, h
, h
z
. What shapes
are the = constant and = constant surfaces? Write an expression for f in elliptical cylindrical coordinates.
Calculus of Variations 273
D Calculus of Variations
D.1 Euler-Lagrange Equation
It is a well-known fact, rst enunciated by Archimedes, that the shortest distance between two points in a plane is
a straight-line. However, suppose that we wish to demonstrate this result from rst principles. Let us consider the
length, l, of various curves, y(x), which run between two xed points, A and B, in a plane, as illustrated in Figure D.1.
Now, l takes the form
l =
_
B
A
[dx
2
+ dy
2
]
1/2
=
_
b
a
[1 + y
2
(x)]
1/2
dx, (D.1)
where y
dy/dx. Note that l is a function of the function y(x). In mathematics, a function of a function is termed a
functional.
b
y
x
A
B
a
Figure D.1: Dierent paths between points A and B.
Now, in order to nd the shortest path between points A and B, we need to minimize the functional l with respect
to small variations in the function y(x), subject to the constraint that the end points, A and B, remain xed. In other
words, we need to solve
l = 0. (D.2)
The meaning of the above equation is that if y(x) y(x) + y(x), where y(x) is small, then the rst-order variation
in l, denoted l, vanishes. In other words, l l + O(y
2
). The particular function y(x) for which l = 0 obviously
yields an extremum of l (i.e., either a maximum or a minimum). Hopefully, in the case under consideration, it yields a
minimum of l.
Consider a general functional of the form
I =
_
b
a
F(y, y
, x) dx, (D.3)
where the end points of the integration are xed. Suppose that y(x) y(x) + y(x). The rst-order variation in I is
written
I =
_
b
a
_
F
y
y +
F
y
_
dx, (D.4)
274 FLUID MECHANICS
where y
= d(y)/dx. Setting I to zero, we obtain

_
b
a
_
F
y
y +
F
y
_
dx = 0. (D.5)
This equation must be satised for all possible small perturbations y(x).
Integrating the second term in the integrand of the above equation by parts, we get
_
b
a
_
F
y
d
dx
_
F
y
__
y dx +
_
F
y
y
_
b
a
= 0. (D.6)
Now, if the end points are xed then y = 0 at x = a and x = b. Hence, the last term on the left-hand side of the above
equation is zero. Thus, we obtain
_
b
a
_
F
y
d
dx
_
F
y
__
y dx = 0. (D.7)
The above equation must be satised for all small perturbations y(x). The only way in which this is possible is for the
expression enclosed in square brackets in the integral to be zero. Hence, the functional I attains an extremum value
whenever
d
dx
_
F
y
F
y
= 0. (D.8)
This condition is known as the Euler-Lagrange equation.
Let us consider some special cases. Suppose that F does not explicitly depend on y. It follows that F/y = 0.
Hence, the Euler-Lagrange equation (D.8) simplies to
F
y
= const. (D.9)
Next, suppose that F does not depend explicitly on x. Multiplying Equation (D.8) by y
, we obtain
y
d
dx
_
F
y
_
y
F
y
= 0. (D.10)
However,
d
dx
_
y
F
y
_
= y
d
dx
_
F
y
_
+ y
F
y
. (D.11)
Thus, we get
d
dx
_
y
F
y
_
= y
F
y
+ y
F
y
. (D.12)
Now, if F is not an explicit function of x then the right-hand side of the above equation is the total derivative of F,
namely dF/dx. Hence, we obtain
d
dx
_
y
F
y
_
=
dF
dx
, (D.13)
which yields
y
F
y
F = const. (D.14)
Returning to the case under consideration, we have F =
_
1 + y
2
, according to Equation (D.1) and (D.3). Hence,
F is not an explicit function of y, so Equation (D.9) yields
F
y
=
y
_
1 + y
2
= c, (D.15)
where c is a constant. So,
y
=
c
1 c
2
= const. (D.16)
Of course, y
= constant is the equation of a straight-line. Thus, the shortest distance between two xed points in a
plane is indeed a straight-line.
D.2 Conditional Variation
Suppose that we wish to nd the function y(x) which maximizes or minimizes the functional
I =
_
b
a
F(y, y
, x) dx, (D.17)
subject to the constraint that the value of
J =
_
b
a
G(y, y
, x) dx (D.18)
remains constant. We can achieve our goal by nding an extremum of the new functional K = I + J, where (x) is
an undetermined function. We know that J = 0, since the value of J is xed, so if K = 0 then I = 0 as well. In
other words, nding an extremum of K is equivalent to nding an extremum of I. Application of the Euler-Lagrange
equation yields
d
dx
_
F
y
F
y
+
_
d
dx
_
[G]
y
[G]
y
_
= 0. (D.19)
In principle, the above equation, together with the constraint (D.18), yields the functions (x) and y(x). Incidentally,
is generally termed a Lagrange multiplier. If F and G have no explicit x-dependence then is usually a constant.
As an example, consider the following famous problem. Suppose that a uniformchain of xed length l is suspended
by its ends from two equal-height xed points which are a distance a apart, where a < l. What is the equilibrium
conguration of the chain?
Suppose that the chain has the uniform density per unit length . Let the x- and y-axes be horizontal and vertical,
respectively, and let the two ends of the chain lie at (a/2, 0). The equilibrium conguration of the chain is specied
by the function y(x), for a/2 x +a/2, where y(x) is the vertical distance of the chain below its end points at
horizontal position x. Of course, y(a/2) = y(+a/2) = 0.
According to standard Newtonian dynamics, the stable equilibriumstate of a conservative dynamical system is one
which minimizes the systems potential energy. Now, the potential energy of the chain is written
U = g
_
y ds = g
_
a/2
a/2
y [1 + y
2
]
1/2
dx, (D.20)
where ds =
_
dx
2
+ dy
2
is an element of length along the chain, and g is the acceleration due to gravity. Hence,
we need to minimize U with respect to small variations in y(x). However, the variations in y(x) must be such as to
conserve the xed length of the chain. Hence, our minimization procedure is subject to the constraint that
l =
_
ds =
_
a/2
a/2
[1 + y
2
]
1/2
dx (D.21)
remains constant.
It follows, from the above discussion, that we need to minimize the functional
K = U + l =
_
a/2
a/2
( g y + ) [1 + y
2
]
1/2
dx, (D.22)
where is an, as yet, undetermined constant. Since the integrand in the functional does not depend explicitly on x, we
have from Equation (D.14) that
y
2
( g y + ) [1 + y
2
]
1/2
( g y + ) [1 + y
2
]
1/2
= k, (D.23)
where k is a constant. This expression reduces to
y
2
=
_
+
y
h
_
2
1, (D.24)
276 FLUID MECHANICS
where
= /k, and h = k/ g.
Let
+
y
h
= cosh z. (D.25)
Making this substitution, Equation (D.24) yields
dz
dx
= h
1
. (D.26)
Hence,
z =
x
h
+ c, (D.27)
where c is a constant. It follows from Equation (D.25) that
y(x) = h [
+ cosh(x/h + c)]. (D.28)

The above solution contains three undetermined constants, h,
, and c. We can eliminate two of these constants by

application of the boundary conditions y(a/2) = 0. This yields
+ cosh(a/2 h + c) = 0. (D.29)
Hence, c = 0, and
= cosh(a/2 h). It follows that

y(x) = h [cosh(a/2 h) cosh(x/h)]. (D.30)
The nal unknown constant, h, is determined via the application of the constraint (D.21). Thus,
l =
_
a/2
a/2
[1 + y
2
]
1/2
dx =
_
a/2
a/2
cosh(x/h) dx = 2 h sinh(a/2 h). (D.31)
Hence, the equilibrium conguration of the chain is given by the curve (D.30), which is known as a catenary (from
the Latin for chain), where the parameter h satises
l
2 h
= sinh
_
a
2 h
_
. (D.32)
D.3 Multi-Function Variation
Suppose that we wish to maximize or minimize the functional
I =
_
b
a
F(y
1
, y
2
, , y
/
, y
1
, y
2
, , y
/
, x) dx. (D.33)
Here, the integrand F is now a functional of the / independent functions y
i
(x), for i = 1, /. A fairly straightforward
extension of the analysis in Section D.1 yields / separate Euler-Lagrange equations,
d
dx
_
F
y
i
_
F
y
i
= 0, (D.34)
for i = 1, /, which determine the / functions y
i
(x). If F does not explicitly depend on the function y
k
then the kth
Euler-Lagrange equation simplies to
F
y
k
= const. (D.35)
Likewise, if F does not explicitly depend on x then all / Euler-Lagrange equations simplify to
y
i
F
y
i
F = const, (D.36)
for i = 1, /.
D.4 Exercises
D.1. Find the extremal curves y = y(x) of the following constrained optimization problems, using the method of Lagrange
multipliers:
(a)
_
1
0
_
y
2
+ x
2
_
dx, such that
_
1
0
y
2
dx = 2.
(b)
_

0
y
2
dx, such that y(0) = y() = 0, and
_

0
y
2
dx = 2.
(c)
_
1
0
y dx, such that y(0) = y(1) = 1, and
_ _
1 + y
2
dx = 2/3.
D.2. Suppose P and Q are two points lying in the x-y plane, which is orientated vertically such that P is above Q. Imagine there is
a thin, exible wire connecting the two points and lying entirely in the x-y plane. A frictionless bead travels down the wire,
impelled by gravity alone. Show that the shape of the wire that results in the bead reaching the point Q in the least amount
of time is a cycloid, which takes the parametric form
x() = k ( sin ) ,
y() = k (1 cos ) ,
where k is a constant.
D.3. Find the curve y(x), in the interval 0 x p, which is of length and maximizes
_
p
0
y dx.
278 FLUID MECHANICS
Ellipsoidal Potential Theory 279
E Ellipsoidal Potential Theory
Let us adopt the right-handed Cartesian coordinate system x
1
, x
2
, x
3
. Consider a homogeneous ellipsoidal body whose
outer boundary satises
x
2
1
a
2
1
+
x
2
2
a
2
2
+
x
2
3
a
2
3
= 1. (E.1)
Let us calculate the gravitational potential (i.e., the potential energy of a unit test mass) at some point P (x
1
, x
2
, x
3
)
lying within this body.
Consider the contribution to the potential at P from the mass contained within a double cone, whose apex is P,
and which is terminated in both directions at the bodys outer boundary. See Figure E.1. If the cone subtends a solid
angle d then a volume element is written dV = r
2
dr d, where r measures displacement from P along the axis of
the cone. Thus, from standard classical gravitational theory, the contribution to the potential takes the form
d =
_
r
0
G
r
dV
_
0
r
G
(r)
dV, (E.2)
where r
= PQ, r
= PR, and is the constant mass density of the ellipsoid. Hence, we obtain
d = G
__
r
0
r dr +
_
r
0
r dr
_
d =
1
2
G (r
2
+ r
2
) d. (E.3)
The net potential at P is obtained by integrating over all solid angle, and dividing the result by two to adjust for double
counting. This yields
=
1
4
G
_
(r
2
+ r
2
) d. (E.4)
From Figure E.1, the position vector of point Q, relative to the origin, O, is
x
= x + r
n, (E.5)
where x = (x
1
, x
2
, x
3
) is the position vector of point P, and n a unit vector pointing from P to Q. Likewise, the
position vector of point R is
x
= x + r
n. (E.6)
However, Q and R both lie on the bodys outer boundary. It follows, from (E.1), that r
and r
are the two roots of
i=1,3
_
x
i
+ r n
i
a
i
_
2
= 1, (E.7)
which reduces to the quadratic
Ar
2
+ Br + C = 0, (E.8)
where
A =
i=1,3
n
2
i
a
2
i
, (E.9)
B = 2
i=1,3
x
i
n
i
a
2
i
, (E.10)
C =
i=1,3
x
2
i
a
2
i
1. (E.11)
280 FLUID MECHANICS
P O
x
r
Q
R
d
Figure E.1: Calculation of ellipsoidal gravitational potential.
Now, according to standard polynomial equation theory, r
+ r
= B/A, and r
= C/A. Thus,
r
2
+ r
2
= (r
+ r
)
2
2 r
=
B
2
A
2
2
C
A
, (E.12)
and (E.4) becomes
=
1
2
G
_
_
_
2
_
_
i=1,3
x
i
n
i
/a
2
i
_
2
_
_
i=1,3
n
2
i
/a
2
i
_
2
+
1
_
i=1,3
x
2
i
/a
2
i
_
i=1,3
n
2
i
/a
2
i
_
_
d. (E.13)
The above expression can also be written
=
1
2
G
_
_
_
2
_
i, j=1,3
x
i
x
j
n
i
n
j
/(a
2
i
a
2
j
)
_
_
i=1,3
n
2
i
/a
2
i
_
2
+
1
_
i=1,3
x
2
i
/a
2
i
_
i=1,3
n
2
i
/a
2
i
_
_
d. (E.14)
However, the cross terms (i.e., i j) integrate to zero by symmetry, and we are left with
=
1
2
G
_
_
_
2
_
i=1,3
x
2
i
n
2
i
/a
4
i
_
_
i=1,3
n
2
i
/a
2
i
_
2
+
1
_
i=1,3
x
2
i
/a
2
i
_
i=1,3
n
2
i
/a
2
i
_
_
d. (E.15)
Let
J =
_
d
_
i=1,3
n
2
i
/a
2
i
. (E.16)
It follows that
1
a
i
J
a
i
=
_
2 n
2
i
/a
4
i
_
_
i=1,3
n
2
i
/a
2
i
_
2
d. (E.17)
Thus, (E.15) can be written
=
1
2
G
_
_
J
i=1,3
A
i
x
2
i
_
_
, (E.18)
where
A
i
=
J
a
2
i
1
a
i
J
a
i
. (E.19)
Ellipsoidal Potential Theory 281
At this stage, it is convenient to adopt the spherical angular coordinates, and (see Section C.4), in terms of
which
n = (sin cos , sin sin , cos ), (E.20)
and d = sin d d. We nd, from (E.16), that
J = 8
_
/2
0
sin d
_
/2
0
_
_
sin
2
cos
2
a
2
1
+
sin
2
sin
2
a
2
2
+
cos
2
a
2
3
_
_
1
d. (E.21)
Let t = tan . It follows that
J = 8
_
/2
0
sin d
_

0
dt
a + b t
2
= 4
_
/2
0
sin d
(a b)
1/2
, (E.22)
where
a =
sin
2
a
2
1
+
cos
2
a
2
3
, (E.23)
b =
sin
2
a
2
2
+
cos
2
a
2
3
. (E.24)
Hence, we obtain
J = 4 a
1
a
2
a
2
3
_
/2
0
sin sec
2
d
(a
2
1
+ a
2
3
tan
2
)
1/2
(a
2
2
+ a
2
3
tan
2
)
1/2
. (E.25)
Let u = a
2
3
tan
2
. It follows that
J = 2 a
1
a
2
a
3
_

0
du
, (E.26)
where
= (a
2
1
+ u)
1/2
(a
2
2
+ u)
1/2
(a
2
3
+ u)
1/2
. (E.27)
Now, from (E.19), (E.26), and (E.27),
A
i
=
2 a
1
a
2
a
3
a
2
i
_

0
du
1
a
i
a
i
_
2 a
1
a
2
a
3
_

0
du
_
=
2 a
1
a
2
a
3
a
i
_

0
a
i
_
1
_
du = 2 a
1
a
2
a
3
_

0
du
(a
2
i
+ u)
. (E.28)
Thus, from (E.18), (E.26), and (E.28),
=
3
4
G M
_
i=1,3
i
x
2
i
_
_
, (E.29)
where
0
=
_

0
du
, (E.30)
i
=
_

0
du
(a
2
i
+ u)
. (E.31)
Here, M = V and V = (4/3) a
1
a
2
a
3
are the bodys mass and volume, respectively.
The total gravitational potential energy of the body is written
U =
1
2
_
dV, (E.32)
282 FLUID MECHANICS
where the integral is taken over all interior points. It follows from (E.29) that
U =
3
8
G M
2
_
0

1
5
i=1,3
i
a
2
i
_
_
. (E.33)
In writing the above, use has been made of the easily demonstrated result
_
x
2
i
dV = (1/5) a
2
i
V. Now,
2
d
du
_
u
_
=
1
i=1,3
a
2
i
(a
2
i
+ u)
, (E.34)
so
i=1,3
i
a
2
i
=
_

0
i=1,3
a
2
i
du
(a
2
i
+ u)
=
_

0
_
2
d
du
_
u
_
+
1
_
du =
0
. (E.35)
Hence, we obtain
U =
3
10
G M
2
0
. (E.36)

Fluid Mechanics

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Fluid Mechanics

Uploaded by

Copyright:

Available Formats

Fluid Mechanics

C and atmospheric pressure are about 1.8 10

is termed the ux of the property out of V, whereas S

is called the net generation rate of the property within

C, located on the surface of the Earth,

C, located on the surface of the Earth, 1.0 10

is uniform in space, as is often the case, then Equation (2.122) ensures

Figure 3.1: Stable and unstable congurations for a oating body.

. Let the vertical straight-line going through H

(x, y) (x, y) y. Hence, the new submerged volume is

(assuming that is small). Hence, MH

is the height of the metacenter relative to the

> 0, and, hence, that

L, increases monotonically with

sin tan cos (1 + sin

)] R. Moreover, according to standard Newtonian

is stationary, and the line CC

rotates at the xed

, at the angular velocity . It follows that points C, C

= (R, 0, 0), and C

. This type of rotation is termed synchronous, and

1). In this limit, the centrifugal potential (3.148) reduces to

is the same as that which would be generated

sin sin cos cos

, which implies that

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

= 0.01230 and R/a

C. We conclude that the eect

, which corresponds to the case where the free energy at the

, which corresponds to the case where the

, whereas that between mercury

, assuming a contact angle of 25

, Jurins law yields h(mm) = 13.5/a(mm). Thus,

E(, k) = E(/2, k) E(, k), (4.38)

F(, k) = F(/2, k) F(, k), (4.39)

then k = 0.9406, and the associated

= /2) 2 l/kis given by

is an arbitrary constant. Thus, we obtain

the minimum radius,

is the same as that between CD and C

and CD. Thus, the increase in energy of

pictured in Figure 5.3.

s to the ux. However, the ux is equal to

], so that the line joining

is the pressure at innity. Making use of expression (5.78), we obtain

, has a minimum radius of order

Figure 5.12: Inviscid ow past a wedge.

Figure 5.14: Inviscid ow over a wedge.

is the pressure at innity.

Figure 6.3: Two vortex laments.

in the complex z-plane map to the points

(z) = d f /dz. Hence,

as innitesimal vectors connecting neighboring points in the complex z-plane to

are innitesimal vectors connecting the corresponding points in the complex

expand or contract by the same factor, and

(x, y), where

= 0. In other words, the functions

F(z) dz, (6.83)

. Suppose that F(z) is well-behaved (i.e.,

without changing its

, located a perpendicular distance r apart, exert a radial force per

/(2 r) on one another, the force being attractive if QQ

> 0, and repulsive if Q Q

, located a perpendicular distance r apart, exert a radial force