CFD Open Series

Revision 2.25

Essentials of CFD
Edited by :
Ideen Sadrehaghighi, Ph.D.

Pressure
countours over
NACA 0012
wing section on
transonic flow

High Lifting
Surfaces

Aerodynamics

ANNAPOLIS, MD
 1

Contents
List of Figures .................................................................................................................................................................... 15

1 Introduction ................................................................................................................................ 22
1.1 Fluid & Aerodynamic Flow Regions As Characterized By Mach Number.................................... 23
1.1.1 Subsonic Flow (M < 1 everywhere)................................................................................................. 24
1.1.2 Transonic Flow (Mixed Regions M < 1 & M > 1) ....................................................................... 24
1.1.3 Supersonic Flow (M > 1 Everywhere) ............................................................................................ 25
1.1.4 Hypersonic Flow (Very High Supersonic Speeds) ..................................................................... 26
1.2 Motivation .............................................................................................................................................................. 26
1.3 CFD Processes....................................................................................................................................................... 28
1.3.1 Partial Differential Equations (PDEs) & Their Methods of Solution .................................. 28
1.3.2 Boundary Conditions............................................................................................................................. 28
1.3.3 Math Modeling and Simulation of Physical Processes............................................................. 29
1.3.4 Numerical Algorithms Evolution ...................................................................................................... 30
1.3.4.1 References ........................................................................................................................................... 32
1.3.5 Verification, Validation & Calibration............................................................................................. 32
1.4 History and Success Stories ............................................................................................................................ 33
1.5 Advantages of using CFD .................................................................................................................................. 37
1.6 Vision for Further................................................................................................................................................ 37
1.7 CFD in Aerospace Industry.............................................................................................................................. 39
1.8 CFD vs. Wind Tunnel.......................................................................................................................................... 40

2 Linear PDE’s and Exact Solution of Some Model Equations ....................................... 42

2.1 Mathematical Character of Basic Equations ............................................................................................ 42
2.2 Behavior of the Different Classes of PDE ................................................................................................... 43
2.2.1 Hyperbolic Equations............................................................................................................................ 43
2.2.1.1 Steady Inviscid Supersonic Flow ................................................................................................ 43
2.2.1.2 Unsteady Inviscid Compressible Flow ..................................................................................... 44
2.2.2 Parabolic Equations ............................................................................................................................... 45
2.2.3 Elliptic Equations .................................................................................................................................... 45
2.2.4 Some Observation................................................................................................................................... 46
2.2.5 The 'Par-Elliptic' Problem ................................................................................................................... 46
2.3 Nonsingular Transformation.......................................................................................................................... 47
2.4 Exact (Closed Form) Solution to Linear Model Equations ................................................................. 47
2.4.1 Linear Wave Equation (1st Order).................................................................................................... 48
2.4.2 Inviscid Burgers Equation ................................................................................................................... 48
2.4.3 Diffusion (Heat) Equation ................................................................................................................... 49
2.4.4 Viscous Burgers Equation ................................................................................................................... 49
2.4.5 Tricomi Equation .................................................................................................................................... 49
2.4.6 2D Laplace Equation .............................................................................................................................. 49
2.4.6.1 Boundary Conditions ...................................................................................................................... 50
2.4.7 Poisson’s Equation ................................................................................................................................. 50
Viscous Burgers Equation ................................................................................................................... 50
2.4.8 The Advection-Diffusion Equation .................................................................................................. 51
2.4.9 The Korteweg-De Vries Equation..................................................................................................... 51
2.4.10 Helmholtz Equation ............................................................................................................................... 51
2.4.11 Summary List of Exact Solution Methods ..................................................................................... 51
 2

3 Solving System of Linear Algebraic Equation.................................................................. 53

3.1 Interpolation (Morphing) ................................................................................................................................ 53
3.2 Numerical Integration ....................................................................................................................................... 54
3.3 Algebraic Eigenvalues & Eigenvectors Problems .................................................................................. 55
3.3.1 Eigenvalues Calculation ....................................................................................................................... 56
3.3.1.1 Eigenvector Calculation ................................................................................................................. 56
3.3.1.2 Right and Left Eigenvectors ......................................................................................................... 56
3.3.1.3 Diagonalization of a Matrix........................................................................................................... 57
3.3.1.4 Case Study 1 - Eigenvalues and Eigenvectors of 3D Euler Equation ........................... 57
3.3.1.5 Governing Equations ....................................................................................................................... 57
3.3.1.6 Transformation Matrix ................................................................................................................... 57
3.3.2 Case Study 2 – Non-Reflecting Boundary Condition in 2D ................................................... 59
3.3.2.1 General Fourier Analysis and Eigenvectors........................................................................... 59
3.4 Solving Systems of Linear Algebraic Equations...................................................................................... 60
3.4.1 Direct Methods ......................................................................................................................................... 61
3.4.1.1 Cramer’s Rule ..................................................................................................................................... 61
3.4.1.2 Gaussian Elimination ................................................................................................................ 61
3.4.1.3 Thomas Algorithm............................................................................................................................ 61
3.4.1.4 Advanced-Direct Methods............................................................................................................. 61
3.4.2 Iterative Methods.................................................................................................................................... 61
3.4.2.1 Stationary Iterative Methods ....................................................................................................... 62
3.4.2.1.1 Jacobi Method ........................................................................................................................ 62
3.4.2.1.2 Gauss-Seidel Iteration......................................................................................................... 62
3.4.2.1.3 Successive Over-Relaxation (SOR) ................................................................................ 63
3.4.2.1.4 SOR by Line ............................................................................................................................. 63
3.4.2.1.5 Block-Iterative Methods .................................................................................................... 64
3.4.2.1.6 Strongly–Implicit Methods ............................................................................................... 64
3.4.2.1.7 ADI Methods ........................................................................................................................... 65
3.4.2.2 Non-Stationary Iterative Methods ............................................................................................. 65
3.4.2.2.1 Conjugate Gradient (CG) .................................................................................................... 65
3.4.2.2.2 Minimum Residual (MINRES) and Symmetric LQ (SYMMLQ) ........................... 65
3.4.2.2.3 Generalized Minimal Residual (GMRES)..................................................................... 66
3.4.2.2.4 Bi-Conjugate Gradient (BiCG).......................................................................................... 66
3.4.2.2.5 Quasi-Minimal Residual (QMR) ...................................................................................... 66
3.4.2.2.6 Conjugate Gradient Squared (CGS) ............................................................................... 66
3.4.2.2.7 Chebyshev Iteration ............................................................................................................ 66
3.4.3 Classical (Stationary) Iterative Schemes vs. Krylov Subspace Methods .......................... 66
3.5 Converting Conservative to Primitive Variables for 2D Euler Equations .................................... 67
3.5.1 Symmetrizing Matrix Variables ........................................................................................................ 68
3.6 Programming Languages & Paradigms for CFD ..................................................................................... 69
3.6.1 Difference Between Object Oriented (OO) vs. Procedural Programming ....................... 69
3.6.1.1 Object Oriented Programming (OOP) ...................................................................................... 69
3.6.1.2 Procedural ........................................................................................................................................... 69
3.6.2 Pillars of Object Oriented Programming (OOP) ......................................................................... 69
3.6.2.1 Abstraction .......................................................................................................................................... 69
3.6.2.2 Encapsulation ..................................................................................................................................... 69
3.6.2.3 Inheritance .......................................................................................................................................... 70
3.6.2.4 Polymorphism .................................................................................................................................... 70
3.6.3 Main Language Styles ............................................................................................................................ 70
3.6.4 Comparison Between MATLAB, C/C++, Python and FORTRAN .......................................... 70
 3

3.6.4.1 FORTRAN ............................................................................................................................................. 71

3.6.4.2 C/C++ ..................................................................................................................................................... 71
3.6.4.3 Python ................................................................................................................................................... 71
3.6.4.4 MATLAB ................................................................................................................................................ 72
3.6.4.5 Graphics and GUI programming in MATLAB ........................................................................ 72

4 CFD Basics .................................................................................................................................... 74

4.1 Marching vs. Equilibrium Problems ............................................................................................................ 75
4.2 Non-Linear CFD Equations .............................................................................................................................. 75
4.3 Coupled vs Segregated Algorithms .............................................................................................................. 76
4.4 Explicit vs. Implicit Methods .......................................................................................................................... 76
4.5 Model Finite Difference Schemes ................................................................................................................. 77
4.6 General Discretization Scheme (Beam-Warming) ................................................................................ 78
4.7 Linearization of Equations .............................................................................................................................. 79
4.7.1 Frozen (Lagging) Coefficient Method ............................................................................................. 79
4.7.2 Simple Iterative Method....................................................................................................................... 79
4.7.3 Newton Raphson Linearization Method........................................................................................ 80
4.7.3.1 Newton Linearization with Coupling........................................................................................ 80
4.7.4 Extrapolating the Coefficients ........................................................................................................... 80
4.7.5 Case Study – Linearization of 1D Unsteady Euler Equation.................................................. 80
4.7.6 Multi-Dimensional Problem and Approximate Factorization (AF) Scheme................... 81
4.7.7 Recommendation on Linearization ................................................................................................. 83
4.8 Convection & Diffusion Terms Discretization ......................................................................................... 83
4.8.1 Upwind Differencing For Convection ............................................................................................. 84
4.8.1.1 Flux Vector Splitting Schemes ..................................................................................................... 85
4.8.1.2 Flux Difference Splitting (Godunov) Schemes ...................................................................... 86
4.8.2 Diffusion Term Discretization ........................................................................................................... 86
4.9 Pressure Based (Incompressible) Schemes ............................................................................................. 88
4.9.1 Methods Based on the Vorticity Equation .................................................................................... 89
4.9.2 Methods Based on Artificial (Pseudo) Compressibility .......................................................... 89
4.9.3 Methods Based on Pressure Iterations or Pressure Correction .......................................... 90
4.9.3.1 PISO Algorithm .................................................................................................................................. 92
4.9.3.2 Case Study - Second Order Non-Iterative PISO-Algorithm vs Iterative PISO........... 92
4.9.3.2.1 Introduction............................................................................................................................ 93
4.9.3.2.2 Preliminary Results ............................................................................................................. 93
4.9.4 Fast Fluid Dynamic (FFD).................................................................................................................... 94
4.10 Density Based (Compressible) Schemes ................................................................................................... 96
4.10.1 Case Study 1 - Density Based Solver for Turbulent Compressible Flows ........................ 97
4.10.1.1 Result for Subsonic and Transonic Flow over a Bump ............................................... 98
4.10.2 Case Study 2 - Transonic Flow Through a Turbine Cascade ................................................. 98
4.10.3 Case Study 3 - Numerical Study of Compressible Lid Driven Cavity Flow ...................... 99
4.10.4 Case Study 4 - 3D Structured/Unstructured Hybrid Navier-Stokes Method for
Turbine Blade Rows .............................................................................................................................................. 100
4.10.4.1 Introduction ............................................................................................................................... 101
4.10.4.2 Governing Equation Unstructured Solver ..................................................................... 102
4.10.4.3 Inviscid Flux Spatial Discretization .................................................................................. 102
4.10.4.4 Viscous Flux Spatial Discretization .................................................................................. 103
4.10.4.5 Time Integration ...................................................................................................................... 103
4.10.4.6 Spatial Discretization ............................................................................................................. 104
 4

4.10.4.7 Hybrid Coupling ....................................................................................................................... 104

4.10.4.8 Results and Discussion .......................................................................................................... 105
4.10.4.8.1 Transonic Wing .................................................................................................................. 105
4.10.4.8.2 3D Viscous Flow Through Turbine Blades-Hybrid Procedure ....................... 106
4.11 Some Observations on Usage of Pressure Based vs. Density Based Schemes ....................... 107
4.12 Residual Implication ....................................................................................................................................... 108
4.12.1 Explicit Schemes ................................................................................................................................... 109
4.12.2 Implicit Schemes .................................................................................................................................. 109
4.12.2.1 Convergence Rate and Storage Requirement .............................................................. 110
4.12.2.2 Reflections on the Evolution of Implicit Navier-Stokes Algorithms ................... 112
4.12.2.2.1 The Noniterative Time-Linearized Block-Coupled ADI Scheme .................... 112
4.12.2.3 Use of Implicitly Coupled, Characteristic-Compatible Boundary Conditions . 112
4.12.2.3.1 Early Work at NASA Ames Research Center .......................................................... 113
4.12.2.3.2 Characteristic-Based Upwind Schemes .................................................................... 113
4.12.2.3.3 Reducing Factorization Error in Time-Linearized Schemes ........................... 114
4.12.2.3.4 Iterative Time-Linearized Schemes for Structured and Unstructured Grids
114
4.12.2.3.5 Iterative Newton-Linearized Unsteady Schemes ................................................. 114
4.12.2.3.6 Scalable Parallel Algorithms for Structured and Unstructured Grids ......... 115
4.13 Listing to CFD Numerical Methods ........................................................................................................... 116
4.14 What Constitute Solver in CFD ................................................................................................................... 119
4.14.1 Linear vs. Non-linear CFD Solver .................................................................................................. 119
4.15 Major Integration Schemes for CFD ......................................................................................................... 119

5 Finite Differencing (FD) & Spatial Discretization ....................................................... 121

5.1 Finite Difference Method (FD) .................................................................................................................... 121
5.1.1 Some Issues Related Finite Differencing .................................................................................... 121
5.1.1.1 Truncation Error ............................................................................................................................ 121
5.1.1.2 Consistency ...................................................................................................................................... 122
5.1.1.3 Stability and Effect of CFL Number ........................................................................................ 123
5.1.1.4 Convergence .................................................................................................................................... 124
5.1.1.4.1 Numerical Convergence .................................................................................................. 125
5.2 Spatial Discretization Schemes................................................................................................................... 125
5.2.1 1st order Upwind (UD) ....................................................................................................................... 126
5.2.2 Central Differencing (CD) ................................................................................................................. 126
5.2.3 2nd order Upwind Differencing (UD)............................................................................................ 126
5.2.4 Power Law Scheme ............................................................................................................................. 126
5.2.5 Quadratic Upwind Interpolation for Convection Kinetics (QUICK) ................................ 127
5.2.6 Monotone Upstream-Centered Schemes for Conservation Laws (MUSCL) ................. 127
5.2.7 Total Variation Diminishing (TVD) .............................................................................................. 128
5.2.7.1 Total Variation Diminishing Cubic Interpolated Propagation (TVDCIP) ............... 128
5.2.8 Monotone Advection and Recons traction Scheme (MARS) .............................................. 128
5.2.8.1 Reconstruction................................................................................................................................ 128
5.2.8.2 Advection .......................................................................................................................................... 128
5.2.9 Blended Differencing .......................................................................................................................... 128
5.2.9.1 Higher Order Upwind Schemes ............................................................................................... 129
5.2.9.2 Other High Resolution Differencing Scheme For Arbitrarily Unstructured Meshes
129
5.2.9.3 Other High-Order Differencing Schemes : ENO and WENO Schemes ...................... 131
 5

5.2.9.4 Case Study 1 - Adaptive Central-Upwind Weighted Compact Non-linear Scheme

with Increasing Order of Accuracy ........................................................................................................... 132
5.2.9.4.1 Abstract ................................................................................................................................. 132
5.2.9.4.2 Introduction......................................................................................................................... 132
5.2.9.4.3 References ............................................................................................................................ 133
5.2.9.5 Case Study 2 – Uniform Flow Preserving Property of High Order Upwind Finite
Difference Schemes on Generalized Coordinate System ................................................................. 135
5.2.9.5.1 Introduction......................................................................................................................... 135
5.2.9.5.2 WENO and WCNS .............................................................................................................. 136
5.2.9.5.3 Metric Computation for Each Case ............................................................................. 136
5.2.9.5.4 Uniform Flow and Vortex Preservation Tests ....................................................... 137
5.2.9.5.5 Conclusions .......................................................................................................................... 138
5.2.9.5.6 References ............................................................................................................................ 138
5.2.9.6 Case Study 3 - Increasing Order of Accuracy of Weighted Compact Non-Linear
Scheme 139
5.2.9.6.1 Order of WCNS.................................................................................................................... 139
5.2.9.6.2 2D Numerical Tests .......................................................................................................... 139
5.2.9.6.3 References ............................................................................................................................ 139
5.2.9.7 Case Study 4 - A Novel Improvement 0f Compact Nonlinear Scheme For
Simulating Compressible Flows ................................................................................................................. 141
5.2.9.8 Introduction ..................................................................................................................................... 141
5.2.9.9 Test Case : Rayleigh-Taylor Instability ................................................................................. 142
5.2.9.10 Conclusions ................................................................................................................................ 142
5.2.9.11 References .................................................................................................................................. 143
5.3 Implementation of Boundary Conditions............................................................................................... 145

6 Finite Volume Method (FVM) ............................................................................................. 147

6.1 Discretization of Control Volume .............................................................................................................. 147
6.2 Cell Centered Finite Volume (CCFV) Discretization........................................................................... 148
6.3 Finite Volume Method using a 1D Example .......................................................................................... 149
6.4 Unsteady 2D Poisson Equation .................................................................................................................. 150
6.5 General Steady 2D Transport Equation in Finite Volume ............................................................... 150
6.5.1 Approximation of Surface and Volume Integrals ................................................................... 152
6.5.2 Discretization of Convective Fluxes ............................................................................................. 153
6.5.3 Central Differences .............................................................................................................................. 153
6.5.4 Upwind Techniques ............................................................................................................................ 154
6.5.5 Flux-Blending Technique.................................................................................................................. 155
6.5.6 Discretization of Diffusive Fluxes.................................................................................................. 156
6.5.7 Non-Cartesian Grids............................................................................................................................ 157
6.5.8 Discrete Transport Equation........................................................................................................... 159
6.5.9 Treatment of Boundary Conditions.............................................................................................. 160
6.5.10 Case Study - Numerical Analysis of Turbulent Flow Around 2D Bodies....................... 162
6.5.10.1 Introduction and Literature Survey................................................................................. 162
6.5.10.2 Governing Equations .............................................................................................................. 163
6.5.10.3 Boundary Conditions ............................................................................................................. 165
6.5.10.4 Finite Volume Discretization of Governing Equation ............................................... 166
6.5.10.5 The Wall Functions ................................................................................................................. 168
6.5.10.6 SIMPLE Algorithm ................................................................................................................... 168
6.5.10.7 Results and Discussion for NACA 0012 Hydrofoil ..................................................... 168
 6

6.5.10.8 Conclusions ................................................................................................................................ 170

6.6 Some Abnormalities Associated with Control Volume (CV) .......................................................... 170
6.6.1 Numerical Diffusion Effects ............................................................................................................. 171
6.6.2 Balancing the Diffusion vs. Convection using Peclet Number (Pe) ................................. 172
6.6.2.1 Case Study – 1D Convection / Diffusion Problem ............................................................ 173
6.7 Some Attributes to Spatial Discretization .............................................................................................. 174
6.7.1 Cell-Centered (CC) vs. Node-Centered (NC) ............................................................................. 174
6.7.2 Staggered vs. Collocated Grids ....................................................................................................... 175
6.7.3 Desired Properties in Discretization Schemes ........................................................................ 176
6.8 Wall Distance Estimation .............................................................................................................................. 176
6.8.1 Case Study - Influence of Hybrid Grid and Turbulence Model .......................................... 177
6.8.1.1 Gridding ............................................................................................................................................. 177
6.8.1.2 Boundary Conditions ................................................................................................................... 178
6.8.1.3 Governing Equations and Solver Settings ........................................................................... 178
6.8.1.4 Drag findings ................................................................................................................................... 179
6.8.1.5 Impact of Turbulence Model ..................................................................................................... 179
6.8.1.6 Concluding Remarks..................................................................................................................... 180
6.9 Gradient Smoothing Method (GSM) ......................................................................................................... 180
6.9.1 Gradient Smooth Operation............................................................................................................. 181
6.9.2 Case Study 1 - Implicit Gradient Method for Cell-Centered Finite-Volume Solver on
Unstructured Meshes............................................................................................................................................ 182
6.9.2.1 Background and Introduction .................................................................................................. 182
6.9.2.2 Finite-Volume Discretization .................................................................................................... 183
6.9.3 Case Study 2 - Transonic Flow Over a Joukowsky Airfoil at M = 0.8 ............................ 184
6.9.4 Concluding Remarks ........................................................................................................................... 185

7 Boundary Types...................................................................................................................... 187

7.1 Introduction ....................................................................................................................................................... 187
7.2 Naming Convention for Different Types of Boundaries ................................................................... 187
7.2.1 Dirichlet Boundary Condition......................................................................................................... 187
7.2.2 Von Neumann Boundary Condition ............................................................................................. 187
7.2.3 Mixed or Combination of Dirichlet and von Neumann Boundary Condition .............. 187
7.2.4 Robin Boundary Condition .............................................................................................................. 188
7.2.5 Cauchy Boundary Condition............................................................................................................ 188
7.2.6 Periodic (Cyclic Symmetry) Boundary Condition .................................................................. 188
7.2.7 Generic Boundary Conditions ......................................................................................................... 188
7.3 Wall Boundary Conditions ........................................................................................................................... 189
7.3.1 Velocity Field ......................................................................................................................................... 189
7.3.2 Pressure ................................................................................................................................................... 189
7.3.3 Scalars/Temperature ......................................................................................................................... 190
7.3.3.1 Common Inputs for Wall Boundary Condition .................................................................. 190
7.4 Symmetry Planes .............................................................................................................................................. 190
7.5 Inflow Boundaries ........................................................................................................................................... 191
7.5.1 Velocity Inlet .......................................................................................................................................... 191
7.5.2 Pressure Inlet ........................................................................................................................................ 191
7.5.3 Mass Flow Inlet ..................................................................................................................................... 191
7.5.4 Inlet Vent ................................................................................................................................................. 192
7.6 Outflow Boundaries ........................................................................................................................................ 192
7.6.1 Pressure Outlet ..................................................................................................................................... 192
 7

7.6.2 Pressure Far-Field ............................................................................................................................... 193

7.6.3 Outflow ..................................................................................................................................................... 193
7.6.4 Outlet Vent .............................................................................................................................................. 193
7.6.5 Exhaust Fan ............................................................................................................................................ 193
7.7 Free Surface Boundaries ............................................................................................................................... 194
7.7.1 Velocity Field and Pressure ............................................................................................................. 194
7.7.2 Scalars/Temperature ......................................................................................................................... 194
7.8 Pole (Axis) Boundaries .................................................................................................................................. 194
7.9 Periodic Flow Boundaries ............................................................................................................................ 194
7.9.1 Case Study – Boundary Condition on A 2-D HPT Blade Subject to Strong Pressure
Gradient and Curvature Effects ........................................................................................................................ 195
7.10 Non-Reflecting Boundary Conditions (NRBCs) ................................................................................... 196
7.10.1 Case Study 1 - Turbomachinery Application of 2D Subsonic Cascade ........................... 197
7.10.2 Case Study 2 - CAA Application of Airfoil Turbulence Interaction Noise Simulation
198
7.11 Turbulence Intensity Boundaries .............................................................................................................. 198
7.11.1 Turbulence Intensity .......................................................................................................................... 198
7.12 Free Surface Boundary .................................................................................................................................. 199
7.12.1 The Kinematic Boundary Condition ............................................................................................. 199
7.12.2 The Dynamic Boundary Condition................................................................................................ 199
7.13 Other Boundary Conditions ......................................................................................................................... 200
7.14 Further Remarks .............................................................................................................................................. 200

8 Temporal Discretization and Time Marching .............................................................. 202

8.1 Transient 1D Diffusion with Source Term ............................................................................................ 203
8.1.1 Euler Implicit ......................................................................................................................................... 203
8.1.2 Crank-Nicolson ..................................................................................................................................... 203
8.1.3 Simple Explicit....................................................................................................................................... 205
Case Study – Numerical Solution of 1D Heat Model Equation .............................. 205
8.1.4 Runga-Kutta Method .......................................................................................................................... 205
8.1.4.1 Explicit Treatment......................................................................................................................... 205
8.1.4.2 Implicit Runge-Kutta Methods ................................................................................................. 206
8.1.4.3 Case Study - Time Integration Schemes for the Unsteady NS Equations ............... 207
8.1.5 Piecewise Analytical Method ......................................................................................................... 207
8.2 Transient Simulation ...................................................................................................................................... 207
8.2.1 Time Step Size ....................................................................................................................................... 208
8.2.2 Steady-State Vs. Transient Run ...................................................................................................... 208
8.2.2.1 Case Study 1 - Vortex Shedding Behind a Cylinder - Ran as Steady State (Wrong!)
209
8.2.2.2 Case study 2 - Double Sided Membrane - Ran as Steady State (Right!) .................. 209
8.2.2.3 Case Study 3 – Evolution of the Wake of 3 Inline Square Prisms ............................. 209
8.2.3 Concept of Dual Time Stepping ...................................................................................................... 210
8.2.3.1 Backward Differencing Formula (BDF) ................................................................................ 211
8.3 Essence of Local Preconditioning in CFD ............................................................................................... 211
8.3.1 Pseudo (Artificial) Compressibility .............................................................................................. 212
8.3.2 Case Study 1 – Preconditioned Methods for Solving 2D Incompressible Flow.......... 213
8.3.3 Case Study 2 – Preconditioned Methods for 3D Unsteady Low Speed Compressible N-
S 215
8.3.3.1 Low Speed Preconditioning Formulation ............................................................................ 215
 8

8.3.3.2 Numerical Integration ................................................................................................................. 216

8.3.4 Case Study 3 - Numerical Analysis of a Dual-Time Pseudo-Compressibility Method
216
8.3.4.1 Introduction and Background .................................................................................................. 217
8.3.4.2 Governing Equations .................................................................................................................... 218
8.3.4.3 Time Integration ............................................................................................................................ 219
8.3.4.4 Six-Degree-of-Freedom Rigid Body Motion ........................................................................ 220
8.3.4.5 3D Flow Evolution of a Dam Break......................................................................................... 221
8.3.4.6 Free-Surface Wave Around Surface-Piercing NACA Foils............................................. 223
8.3.4.7 Computational Model and on a NACA 66 Hydrofoil ........................................................ 224
8.3.4.8 Water Entry of Free-Falling Rigid Objects........................................................................... 225
8.3.4.8.1 Hemisphere.......................................................................................................................... 225
8.3.4.8.2 Sphere .................................................................................................................................... 226

9 Panel Methods ......................................................................................................................... 228

9.1 Preliminaries...................................................................................................................................................... 228
9.2 Gridding (Paneling) ......................................................................................................................................... 229
9.3 Method of Solving the Potential Equation ............................................................................................. 229
9.3.1 Conservative Forms ............................................................................................................................ 230
9.4 Method for Transonic Flow.......................................................................................................................... 231
9.4.1 Case Study - Panel Methods for Solving the Laplace’s Equation ...................................... 232
9.5 Vortex Lattice Method (VLM)...................................................................................................................... 232
9.5.1 Case Study - Transonic and Viscous Potential Flow Method Applied to Flexible Wing
Transport Aircraft .................................................................................................................................................. 233
9.5.1.1 Abstract.............................................................................................................................................. 233
9.5.1.2 Introduction ..................................................................................................................................... 234
9.5.1.3 Aero-Structural Modeling .......................................................................................................... 236
9.5.1.3.1 Vortex Lattice Model ........................................................................................................ 236
9.5.1.3.2 Transonic Small Disturbance Model / Integral Boundary Layer Model ..... 236
9.5.1.3.3 Euler-IBL Model (MSES) ................................................................................................. 238
9.5.1.3.4 Static Aeroelastic Modeling Framework.................................................................. 239
9.5.1.3.5 Transonic and Viscous Corrections for the VLM .................................................. 240
9.5.1.4 Results ................................................................................................................................................ 242
9.5.1.4.1 Rigid Clean Wing GTM Results ..................................................................................... 242
9.5.1.4.2 Comparison of Viscous Aerodynamic Solutions ................................................... 242
9.5.1.4.3 Aeroelastic Clean Wing GTM Results ........................................................................ 243
9.5.1.4.4 Rigid GTM Wing with VCCTEF Deployed................................................................. 244
9.5.1.4.5 Wall Clock Time Comparison ....................................................................................... 246
9.5.1.5 Conclusion ........................................................................................................................................ 247
9.5.1.6 References ........................................................................................................................................ 248
9.6 VLM vs DLM ........................................................................................................................................................ 249
9.7 Vortex Panel Method (VPM) vs. DLM....................................................................................................... 250

10 Solution Methods for Inviscid (Euler) Equations........................................................ 251

10.1 Background......................................................................................................................................................... 251
10.2 Method of Characteristics ............................................................................................................................. 251
10.2.1 Linear Systems ...................................................................................................................................... 251
10.2.2 Non-Linear Systems ............................................................................................................................ 253
10.3 Shock Capturing Method ............................................................................................................................... 255
 9

10.3.1 Explicit McCormack Method ........................................................................................................... 255

10.3.2 1D Upwind Flux-Splitting Scheme (Steger-Warming) ......................................................... 256
10.3.3 Total Variation Diminishing (TVD) as other Upwind Schemes ........................................ 257
10.3.3.1 The Upwind Connection to Explicit Artificial Dissipation ...................................... 257
10.3.4 3D Unsteady Euler Equation Solutions Using Flux Vector Splitting ............................... 258
10.3.4.1 Flux Splitting.............................................................................................................................. 260
10.4 The Riemann Problem ................................................................................................................................... 260
10.4.1 Roe Approximate Riemann Solver ................................................................................................ 261
10.4.2 Case Study – Further Discussion Regarding the Shock Capturing Techniques to
Evaluating the Discourteous Solution............................................................................................................ 262
10.4.3 References............................................................................................................................................... 264
10.5 Euler Equation With 2-Species ................................................................................................................... 265
10.5.1 Numerical Methods ............................................................................................................................. 266
10.5.2 References............................................................................................................................................... 266

11 Solution Methods for Viscous (Navier-Stokes) Equations ....................................... 268

11.1 General Transformation ................................................................................................................................ 269
11.2 Solution Methods for Compressible (Density Based) N-S Equation ........................................... 270
11.2.1 Explicit Scheme (Mac Cormack) .................................................................................................... 270
11.2.2 Case Study – Dual Block Applied to Navier-Stokes Equations in 2D ............................. 272
11.2.2.1 Governing Equations .............................................................................................................. 272
11.2.2.2 Numerical Method................................................................................................................... 273
11.2.2.3 Block Interface .......................................................................................................................... 273
11.2.2.4 Stability Consideration .......................................................................................................... 274
11.2.2.4.1 Test 1 - Flow Over a Backward-Facing Step ........................................................... 274
11.2.2.4.2 Test 2 - Flow Over a Curved Ramp ............................................................................. 276
11.2.2.5 Conclusions ................................................................................................................................ 277
11.2.3 Other Explicit Schemes ...................................................................................................................... 277
11.2.4 Implicit Schemes .................................................................................................................................. 277
11.2.4.1 Beam-Warming ........................................................................................................................ 277
11.2.4.2 Mac Cormack ............................................................................................................................. 278
11.3 Solution Methods for Incompressible N-S Equations (Pressure Based) ................................... 279
11.3.1 Explicit Schemes ................................................................................................................................... 279
11.3.1.1 Vorticity-Stream Function Approach (2D).................................................................... 279
11.3.1.2 Primitive-Variable Approach .............................................................................................. 280
11.3.2 Implicit Schemes .................................................................................................................................. 281
11.3.2.1 SIMPLE (Semi-Implicit Method for Pressure Linked Equation) for 2D N-S
Equation 281
11.3.3 Projection Methods ............................................................................................................................. 282
11.4 Approximation (Thin-Layer) of N-S Equation...................................................................................... 282
11.5 Parabolized N-S Equation ............................................................................................................................. 283
11.5.1 3D Compressible Implicit Numerical Solution of PNS Equation ...................................... 284
11.5.2 Case Study - New-Generation Hyperbolic Navier-Stokes Schemes: Speed-Up and
Accurate Viscous/Heat Fluxes ......................................................................................................................... 287
11.5.2.1 Introduction ............................................................................................................................... 287
11.5.2.2 Simplified Discretization ...................................................................................................... 287
11.5.2.3 Speed-up...................................................................................................................................... 288
11.5.2.4 Accurate Viscous/Heat Fluxes ........................................................................................... 288
11.5.2.5 First-Order 2D Hyperbolic Navier-Stokes System ..................................................... 288
 10

11.5.2.6 Discretization ............................................................................................................................ 292

11.5.2.7 Results .......................................................................................................................................... 293
11.5.2.8 Concluding Remarks and Future Developments ........................................................ 295
11.6 Multigrid Methods ........................................................................................................................................... 297
11.6.1 Multigrid Cycle & Definitions .......................................................................................................... 297
11.6.2 Classification .......................................................................................................................................... 298
11.6.3 Unstructured Mesh Cycling ............................................................................................................. 299
11.6.4 Viscous Flow Consideration ............................................................................................................ 300
11.6.5 Case Study - Development and Application of Agglomerated Multigrid Methods for
Complex Geometries ............................................................................................................................................. 301
11.6.5.1 Introduction ............................................................................................................................... 301
11.6.5.2 Discretization ............................................................................................................................ 302
11.6.5.3 Agglomeration Scheme ......................................................................................................... 302
11.6.5.4 Single-Grid Iterations............................................................................................................. 308
11.6.5.5 Multigrid...................................................................................................................................... 308
11.6.5.5.1 Multigrid V-Cycle ............................................................................................................... 308
11.6.5.5.2 Inter-Grid Operators ........................................................................................................ 309
11.6.5.5.3 Coarse-Grid Discretization ............................................................................................ 309
11.6.5.5.4 Relaxations ........................................................................................................................... 310
11.6.5.5.5 Cost of Multigrid V-Cycle ................................................................................................ 311
11.6.5.6 Results for Complex Geometries ....................................................................................... 312
11.6.5.6.1 Model Diffusion Equation .............................................................................................. 312
11.6.5.6.2 Inviscid Flows ..................................................................................................................... 313
11.6.5.6.3 Turbulent Flows (RANS) ................................................................................................ 314
11.6.5.7 Concluding Remarks .............................................................................................................. 314
11.6.5.8 References .................................................................................................................................. 314

12 Finite Element Method (FE)................................................................................................ 316

12.1 Continuous & Discontinuous Galerkin Finite Element Methods .................................................. 316
12.2 Steps in the Finite Element Discretization ............................................................................................ 317
12.3 Concept of Shape (Basis) Function ........................................................................................................... 318
12.3.1 Lagrangian Elements .......................................................................................................................... 318
12.4 Simple 1D Discretization Example using Shape (Basis) Function ............................................... 319
12.5 Mathematics Behind Finite Element Method ....................................................................................... 319
12.6 Summary of Math Procedure for Finite Element Analysis .............................................................. 320
12.7 Example with Heat Source ........................................................................................................................... 320
12.7.1 Steady State Heat Sink; Weak Formulation; Basis Functions & Test Functions ........ 321
12.7.2 Test Function ......................................................................................................................................... 322
12.7.3 Weak Formulation ............................................................................................................................... 322
12.8 Effect of Basis Function ................................................................................................................................. 324
12.9 Different Elements ........................................................................................................................................... 324
12.10 Error Estimation ............................................................................................................................................... 325
12.11 Additional Finite Element Formulations ................................................................................................ 325
12.12 Case Study - A High Order Discontinuous Galerkin – Fourier Incompressible 3D Navier–
Stokes Solver with Rotating Sliding Meshes ..................................................................................................... 326
12.12.1 Introduction ..................................................................................................................................... 326
12.12.2 Methodology .................................................................................................................................... 329
12.12.2.1 Arbitrary Lagrangian–Eulerian Temporal Discretization ...................................... 330
12.12.2.2 3D Fourier Extension ............................................................................................................. 330
 11

12.12.2.3 Rotating Cases ........................................................................................................................... 332

12.12.3 Case Study - Three Bladed Cross-Flow Turbine Case Study ........................................ 332
12.12.4 3D Flows ............................................................................................................................................ 334
12.12.5 Conclusions ...................................................................................................................................... 334
12.13 Conceptual Differences Between 4 Most used Prediction Methods............................................ 335
12.13.1 Finite Differencing Method (FDM) .................................................................................... 336
12.13.2 Finite Volume Method (FVM) ................................................................................................... 336
12.13.3 Finite Element Method (FEM) .................................................................................................. 336
12.13.4 Discontinuous Galerkin Methods (DG) ................................................................................. 337
12.13.5 Comparing FV and FE for Spatial Convergence of 3D Benchmark Cases ............... 338
12.13.5.1 Case Study - Subsonic Flow Over 3D Swept Bump Configuration ...................... 338
12.13.5.2 Results and Discussion .......................................................................................................... 339

13 Spectral and Other High Degree Methods ..................................................................... 341

13.1 Spectral Difference Method (SD) ............................................................................................................... 341
13.2 Spectral Volume Method (SV) ..................................................................................................................... 343
13.2.1 Basic Formulation................................................................................................................................ 343
13.3 Fast Fourier Transform Methods (FFT).................................................................................................. 344
13.3.1 Case Study - Flow and Fast Fourier Transform (FFT) Analyses for Tip Clearance
Effect in an Operating Kaplan Turbine .......................................................................................................... 345
13.3.1.1 Background and Literature Survey .................................................................................. 345
13.3.1.2 Governing Equations .............................................................................................................. 349
13.3.1.3 Validation of Numerical Results ........................................................................................ 350
13.3.1.3.1 Performance Characteristics ........................................................................................ 350
13.3.1.3.2 Effect of Tip Clearance..................................................................................................... 351
13.3.1.4 Conclusions ................................................................................................................................ 352
13.4 Finite Differences vs. Spectral Methods.................................................................................................. 353
13.5 Weighted Residual Method (WRM) .......................................................................................................... 353
13.5.1 General Formulation .......................................................................................................................... 354
13.6 Boundary Element Method (BEM) ............................................................................................................ 354
13.6.1 Comparison to Other Methods ....................................................................................................... 355
13.7 Dispersion-Relation-Preserving (DRP) Algorithms for 3D Problems ....................................... 355
13.7.1 Effects of AeroAcoustics .................................................................................................................... 356
13.8 Lattice Boltzmann Method (LBM) ............................................................................................................. 357
13.8.1 Approaches ............................................................................................................................................. 357
13.8.2 Dilute Gas Regimes .............................................................................................................................. 358
13.8.3 Middle of the Scale............................................................................................................................... 359
13.8.4 Maxwell Distribution Function ...................................................................................................... 359
13.8.5 Boltzmann Transport Equation ..................................................................................................... 360
13.8.6 The BGKW Approximation ............................................................................................................... 361
13.8.7 Lattices & DnQm Classification ...................................................................................................... 362
13.8.8 Lattice Arrangements......................................................................................................................... 363
13.8.8.1 1D Lattice Boltzmann Method (D1O2) ........................................................................... 363
13.8.8.2 2D Lattice Boltzmann Method (D2Q9) ........................................................................... 364
13.8.9 Case Study 1 – Solving 2D Conduction with Heat Flux Boundary Using The Lattice
Boltzmann Method (LBM) .................................................................................................................................. 366
13.8.9.1 Formulation and Kinetic Equation ................................................................................... 366
13.8.9.2 Results and Discussion .......................................................................................................... 368
13.8.9.2.1 Case Study 1 - Four Boundaries are at Known Temperatures ....................... 368
 12

13.8.9.2.2 Case Study 2 - Effects of Heat Generation and the Four Boundaries are at
Specified Temperatures .......................................................................................................................... 369
13.8.9.2.3 Case Study 3 - The Bottom and Top Boundaries are at Prescribed Fluxes and
Remaining Two Boundaries at Known Temperatures .............................................................. 370
13.8.9.3 Conclusions ................................................................................................................................ 370
13.8.9.3.1 Case Study 4 - Lid-Driven Cavity Flow...................................................................... 370
13.9 Immersed Boundary Method (IBM) ......................................................................................................... 371
13.9.1 Case Study - Immersed Boundary Methods (IBM)................................................................. 371
13.9.1.1 Introduction ............................................................................................................................... 371
13.9.1.2 General Considerations ......................................................................................................... 372
13.9.2 Categories of IB Methods .................................................................................................................. 372
13.9.3 Continuous Forcing Approach With Boundary Imposition ................................................ 373
13.9.3.1 Elastic Boundaries................................................................................................................... 373
13.9.3.2 Rigid Boundaries ..................................................................................................................... 375
13.9.3.3 General Considerations ......................................................................................................... 376
13.9.4 Direct Forcing Approach (Discrete Approach) ........................................................................ 376
13.9.4.1 General Idea ............................................................................................................................... 377
13.9.4.2 Direct Forcing............................................................................................................................ 377
13.9.4.3 Improvements........................................................................................................................... 378
13.9.4.4 Fulfilling Conservation Laws .............................................................................................. 380
13.9.4.5 Conservation of Mass ............................................................................................................. 381
13.9.5 Cut-Cell Finite-Volume Approach ................................................................................................. 382
13.9.5.1 Basic Formulation ................................................................................................................... 382
13.9.5.2 Improvements........................................................................................................................... 384
13.9.5.3 LS-STAG ....................................................................................................................................... 386
13.9.5.4 General Considerations ......................................................................................................... 386
13.9.6 Case Study - Flow Past a Sphere ................................................................................................... 387
13.9.7 References............................................................................................................................................... 387

14 Components of CFD Simulation ......................................................................................... 392

14.1 Simulation vs. Modeling ................................................................................................................................ 393
14.1.1 Part I .......................................................................................................................................................... 393
14.1.2 Part II ........................................................................................................................................................ 393
14.1.3 Part III ....................................................................................................................................................... 393
14.1.4 Part IV ....................................................................................................................................................... 393
14.1.5 Part V......................................................................................................................................................... 393
14.2 Simulation Components as Envisioned by NASA .............................................................................. 394
14.2.1 Formulate the Flow Problem .......................................................................................................... 394
14.2.2 Model the Geometry and Flow Domain ...................................................................................... 395
14.2.3 Establish the Boundary and Initial Conditions ........................................................................ 395
14.2.4 Generate the Grid ................................................................................................................................. 395
14.2.5 Establish the Simulation Strategy ................................................................................................. 395
14.2.6 Establish the Input Parameters and Files .................................................................................. 395
14.2.7 Perform the Simulation ..................................................................................................................... 395
14.2.8 Monitor the Simulation for Completion ..................................................................................... 396
14.2.9 Post-Process the Simulation to get the Results ....................................................................... 396
14.2.10 Make Comparisons of the Results ........................................................................................... 396
14.2.11 Repeat the Process to Examine Sensitivities ...................................................................... 396
14.2.12 Document .......................................................................................................................................... 396
 13

14.3 Comparison of Some Commercially Available CFD Software ........................................................ 396

14.3.1 AcuSolve© (www.altair.com)......................................................................................................... 397
14.3.2 OpenFOAM© (www.openfoam.com) ............................................................................................ 397
14.3.3 OpenFlower (sourceforge.net/projects/openflower) .......................................................... 397
14.3.4 FLASH (flash.uchicago.edu) ............................................................................................................. 398
14.3.5 GADGET© (www.mpa-garching.mpg.de/~volker/gadget)............................................... 398
14.3.6 HYDRA© (hydra.mcmaster.ca/hydra) ......................................................................................... 398
14.3.7 ZEUS-MP (lca.ucsd.edu/portal/software) ................................................................................. 398
14.3.8 ANSYS CFX© (www.ansys.com/Products/Simulation+Technology/Fluid+Dynamics)
398
14.3.9 ANSYS ICEM CFD© (www.ansys.com/Products/Other+Products/ANSYS+ICEM+CFD)
398
14.3.10 FLUENT© (www.fluent.com)..................................................................................................... 399
14.3.11 COMSOL Multi-physics© (www.comsol.com/products/multiphysics) ................... 399
14.3.12 CFDRC© (www.cfdrc.com) ......................................................................................................... 399
14.3.13 STAR-CD/STAR-CCM© (www.cd-adapco.com) ................................................................. 399
14.3.14 FLOW3D© (www.flow3d.com) ................................................................................................. 399
14.3.15 TACOMA© (www.ge.com) .......................................................................................................... 399
14.3.16 CONVERGE™ ( www.convergecfd.com) ............................................................................... 400
14.4 NASA Developed CFD Software .................................................................................................................. 400
14.4.1 FUN3D ...................................................................................................................................................... 400
14.4.2 USM3D ...................................................................................................................................................... 401
14.4.3 CFL3D........................................................................................................................................................ 401
14.4.4 Case Study - Grid Convergence for 3D Benchmark Turbulent Flows............................. 402
14.4.4.1 Subsonic Flow Around a Hemisphere Cylinder .......................................................... 402
14.4.4.2 Geometry, Flow Parameters, and Boundary Conditions ......................................... 402
14.4.4.3 Results for Hemisphere Cylinder ...................................................................................... 403
14.4.4.4 Forces and Pitching Moment .............................................................................................. 404
14.4.4.5 Fine Grid Surface Pressure, Skin Friction, and Off-Body Variation ................... 406
14.4.4.6 Effect of Grid Refinement on Surface Pressure and Skin Friction ....................... 407
14.4.4.7 Transonic Flow Around an M6 Wing............................................................................... 407
14.4.4.8 Geometry, Flow Parameters and Boundary Conditions .......................................... 408
14.4.4.9 Grids for M6 Wing ................................................................................................................... 408
14.4.4.10 Results for M6 Wing ............................................................................................................... 409
14.4.4.11 Concluding Remarks .............................................................................................................. 412
14.5 Symbolic Math Packages ............................................................................................................................... 413
14.5.1 Maxima (maxima.sourceforge.net) ............................................................................................ 413
14.5.2 Mathematica (www.wolfram.com) ............................................................................................ 413
14.5.3 Maple (www.maplesoft.com) ....................................................................................................... 413
14.5.4 MatLab (www.mathworks.com) ................................................................................................ 413
14.5.5 MATHGL (http://mathgl.sourceforge.net/doc_en/index.html)..................................... 413

15 Best Guidelines for Optimal CFD Simulation ................................................................ 414

15.1 Element of Uncertainty .................................................................................................................................. 415
15.1.1 Geometry & Grid Generation........................................................................................................... 415
15.1.2 Convergence........................................................................................................................................... 415
15.1.3 Numerical ................................................................................................................................................ 416
15.1.4 Reaction ................................................................................................................................................... 416
15.1.5 Multiphase Flows ................................................................................................................................. 416
 14

15.1.5.1 Not Knowing the Most Important Physical Mechanisms ........................................ 416
15.1.5.2 Closure Models ......................................................................................................................... 416
15.1.5.3 Time-Scale and Length-Scale Separation....................................................................... 416
15.1.5.4 Choice of Model and Governing Equations ................................................................... 417
15.1.5.5 Numerical Errors ..................................................................................................................... 417
15.1.5.6 Avoiding Risks in Multiphase Flows ................................................................................ 417
15.2 Verification, Validation, Calibration and Certification ...................................................................... 417
15.2.1 Validation for an Intended Purpose. ............................................................................................ 418
15.3 Simulation for Everyone ............................................................................................................................... 418
15.4 CFD on a Deadline ............................................................................................................................................ 419
15.4.1 Optimizing your Modeling Workflow .......................................................................................... 419
15.5 Myths of Computational Fluid Dynamics ............................................................................................... 420
15.5.1.1 CFD is Difficult and Take Too Long to be Used in the Design Process ............... 420
15.5.1.2 Accuracy has to be sacrificed to use CFD during the Design Process ................ 421
15.5.1.3 Experts are Needed to Get Accurate CFD Simulation Results ............................... 421

List of Tables
Table 1.1 Classification of Mach Number ................................................................................................................ 23
Table 1.2 Solution Methods for PDEs ........................................................................................................................ 29
Table 3.1 Comparison of Iterative Schemes ........................................................................................................... 63
Table 3.2 Differences between OOP vs Procedural ............................................................................................. 69
Table 4.1 Implicit Discretization Methods of Model Equations ..................................................................... 78
Table 4.2 Explicit Discretization Methods of Model Equations...................................................................... 78
Table 4.3 Choice of Parameters ................................................................................................................................... 79
Table 5.1 Solution Error (rms) Reduction with Grid Refinement – Courtesy of Fletcher............... 125
Table 5.2 Results of uniform flow preserving test. is number of the grid points) ............................. 137
Table 6.1 Approximations for Surface Integrals Over the Face Se - Courtesy of [Schafer]............. 152
Table 9.1 Automated Mesh Refinement Sequence for MSES........................................................................ 238
Table 9.2 Comparison of 3D Inviscid Lift Curve and Drag Polar Metrics of Interest for the Rigid
GTM .......................................................................................................................................................................................... 242
Table 9.3 Comparison of 3D Viscous Lift Curve and Drag Polar Metrics of Interest for the Rigid
GTM .......................................................................................................................................................................................... 242
Table 9.4 Drag Reduction (counts) for the Top Five GTM VCCTEF Configurations ........................... 245
Table 10.1 Classification of the Euler Equation on Different Regimes..................................................... 251
Table 11.1 Comparation of the Results For Flow Over a Backward Facing Step ................................. 275
Table 11.2 Admissible Agglomerations ................................................................................................................ 303
Table 11.3 Summary of discretization used to define the residual, ^R ................................................... 310
Table 11.4 Summary of costs and typical numbers of linear-sweeps. The multigrid cycle is a 5-
level V (2; 1) with a typical coarsening ratio 8. The numbers in parenthesis denote the number of
point and line sweeps, respectively, and the second set for RANS denotes the number of point and
line sweeps of the turbulent equation. ...................................................................................................................... 311
Table 11.5 Summary of Jacobians, ∂^R*/∂U........................................................................................................ 311
Table 11.6 Cost of V-cycle relative to a single-grid iteration and speed-up factor. The expected
speed-up factors have been computed with the actual coarsening ratio.................................................... 312
Table 11.7 Summary of grid sizes and parameters for the inviscid cases .............................................. 313
Table 12.1 Summarizing the Discontinuous Galerkin Finite Element Method..................................... 337
Table 12.2 Tetrahedral Grids for Finite-Volume and Finite-Element Solutions .................................. 339
Table 13.1 Main Design Parameters of the prototype Kaplan Turbine ................................................... 348
Table 13.2 Mesh Quality of the Kaplan Turbine ................................................................................................ 349
 15

Table 14.1 An Overview of some commercially available CFD software ................................................ 397
Table 14.2 Statistics of four finest grids for hemisphere cylinder grid families. ................................. 402
Table 14.3 Hemisphere Cylinder: Variation of Aerodynamic Coefficients on L1 Grids ................... 406
Table 14.4 Statistics of Grids for OM6 Wing Grid Families ........................................................................... 408
Table 14.5 Variations of Aerodynamic Coefficients ......................................................................................... 410

List of Figures
Figure 1.1 CFD Application in Various Industries ............................................................................................... 22
Figure 1.2 An F/A-18 Hornet Creating a Vapor Cone at Transonic Speed ................................................ 23
Figure 1.3 Block Diagram Categorizing the Types of ......................................................................................... 24
Figure 1.4 Different Regimes of Flow (Courtesy of John D. Anderson) ...................................................... 25
Figure 1.5 Flow over a Supersonic Blunt Body ..................................................................................................... 27
Figure 1.6 Relationship Between Physics Fidelity & Numerical Algorithm .............................................. 30
Figure 1.7 A range of V & V techniques and their interaction with each other ....................................... 33
Figure 1.8 Simulation of the Hyper-X Scramjet Vehicle in Operation by NASA ...................................... 35
Figure 1.9 Required fine-scale geometric details space shuttle external tank ........................................ 36
Figure 1.10 CFD simulation of space shuttle launching configuration ....................................................... 37
Figure 1.11 Road Map for CFD Studies ..................................................................................................................... 38
Figure 1.12 Impact of CFD at Boeing. Green areas have strong CFD penetration; Purple areas have
some penetration; Red areas present future opportunities ................................................................................ 39
Figure 1.13 Typical Design Project ............................................................................................................................ 41
Figure 2.1 Domain and Boundaries for the Solution of Hyperbolic Equations (Steady) ..................... 44
Figure 2.2 Domain and Boundaries for the Solution of Hyperbolic Equations (Unsteady) ............... 44
Figure 2.3 Domain and Boundaries of Elliptic Equations in Two Dimensions ........................................ 45
Figure 2.4 Domain and Boundaries for the Solution of Parabolic Equations in Two Dimensions .. 45
Figure 2.5 Two-way interchange of information between Parabolic and Elliptic flows ..................... 47
Figure 2.6 Solution of linear Wave equation.......................................................................................................... 48
Figure 2.7 Rate of Decay of solution to diffusion equation .............................................................................. 48
Figure 2.8 Formulation of discontinuities in non-linear Burgers (wave) equation .............................. 49
Figure 2.9 Solution to Laplace equation .................................................................................................................. 50
Figure 2.10 Solution to Poisson's Equation ............................................................................................................ 51
Figure 3.1 Bi-linear (middle) and Bi-Cubic (far right) Interpolation........................................................... 53
Figure 3.2 Depiction of Image Morphing Using a single Parameter α ........................................................ 54
Figure 3.3 Iteration Methods Path ............................................................................................................................. 61
Figure 3.4 SOR by line ..................................................................................................................................................... 63
Figure 3.5 ADI Stencil for Implicit Method ............................................................................................................. 65
Figure 3.6 3D Representation of a Sine Function using MATLAB ................................................................. 72
Figure 4.1 Contributions from other disciplines to CFD ................................................................................... 74
Figure 4.2 Linkage between CFD, Experimental & Analytical approach .................................................... 75
Figure 4.3 Segregated Solution .................................................................................................................................... 76
Figure 4.4 Coupled Solutions........................................................................................................................................ 77
Figure 4.5 Schematic physical representation of a propagating wave in the positive x direction .. 84
Figure 4.6 An Arbitrary Polyhedral Control Volume .......................................................................................... 87
Figure 4.7 Geometric Interpretation of the Diffusion Term Approximation –[H. Jasak] ..................... 88
Figure 4.8 Accessible Pressure Solvers in Fluent................................................................................................. 89
Figure 4.9 PISO algorithm flow chart (Courtesy of Giannopapa, and G. Papadakis ) ........................... 92
Figure 4.10 Flow Around Circular Cylinder in a Channel ................................................................................. 93
Figure 4.11 Lift Force as a Function of Time, using Different PISO-Algorithms for the Flow Around
Fixed Cylinder in a Channel (Courtesy of Tukovic et al.) ..................................................................................... 94
 16

Figure 4.12 Solution Method for FFD........................................................................................................................ 94

Figure 4.13 Mach number distribution for flow over a Bump ........................................................................ 98
Figure 4.14 Streamlines of the Base Flow in the Symmetry Plane z = 0.5 and Re = 1000 .................. 99
Figure 4.15 2D Mach Number Distribution in the Test Turbine Cascade (left) - 3D Pressure
Distribution at the Blades (right).................................................................................................................................... 99
Figure 4.16 Distribution of Velocities (u, v) along Centerline Horizontal Distance for Re = 100 100
Figure 4.17 Structured-Unstructured Mesh Overlapping ............................................................................ 105
Figure 4.18 Pressure contour lines across the structured-unstructured zonal interface – Courtesy
of [Tsung et al.] .................................................................................................................................................................... 106
Figure 4.19 Pressure Contour Across the Structured-Unstructured Interface – Courtesy of [Tsung
et al.]......................................................................................................................................................................................... 107
Figure 4.20 Scope of Incompressible Vs. Compressible Flow Solvers...................................................... 107
Figure 4.21 Comparing the Convergence Rates of the ILU-Preconditioned Newton-Krylov Method
.................................................................................................................................................................................................... 110
Figure 5.1 Correlation between truncation error and order of accuracy ............................................... 121
Figure 5.2 Free Jet Flow profile for different order of accuracy ................................................................. 122
Figure 5.3 Conceptual Relationship Between Consistency, Stability and Convergence – Courtesy of
Fletcher ................................................................................................................................................................................... 123
Figure 5.4 1D Stability Analysis Based on CFL Number; (a) CFL < 1 Unstable; (b) CFL ≥ 1 Stable
.................................................................................................................................................................................................... 124
Figure 5.5 Different Spatial Scheme ....................................................................................................................... 127
Figure 5.6 Viscous & Inviscid Stencil for a 2nd order accurate MUSCL ..................................................... 127
Figure 5.7 Deciding the Upwind Direction in Arbitrary Unstructured Mesh ........................................ 130
Figure 5.8 Testing Grid................................................................................................................................................. 138
Figure 5.9 Results of vortex preserving test on the wavy grid ................................................................... 137
Figure 5.10 Density of Mach 3 forwarding step problem with higher order WCNS; (a)WCNS5E
with SHUS, (b)WCNS5E.................................................................................................................................................... 140
Figure 5.11 Rayleigh-Taylor instability problem: 30 density contour lines ranging from 0.9 to 2.2.
.................................................................................................................................................................................................... 143
Figure 5.12 Grid Points at a Boundary ................................................................................................................... 145
Figure 6.1 Control-Volume Partitioning For Finite-Volume Discretization ........................................... 147
Figure 6.2 Definition of CVs and Nodes for Triangular Grids with Donald Polygons ........................ 151
Figure 6.3 Finite Volume in a 2D Hexahedra Cell – Courtesy of [Schafer]............................................. 151
Figure 6.4 Cartesian Control Volume with Notations in Formulas Analogous to Finite-Difference
Methods – Courtesy of [Schafer] ................................................................................................................................. 153
Figure 6.5 Approximation of φe with CDS Method – Courtesy of [Schafer] .......................................... 154
Figure 6.6 Mass flux Dependent Approximation of φe with UDS Method .............................................. 154
Figure 6.7 Mass flux dependent approximation of φe with QUICK method – Courtesy of [Schafer]
.................................................................................................................................................................................................... 155
Figure 6.8 Central Differencing Formula for Approximation of 1st Derivative at CV face – Courtesy
of [Schafer] ........................................................................................................................................................................... 156
Figure 6.9 Central difference approximation ..................................................................................................... 157
Figure 6.10 Approximation of diffusive fluxes for non-Cartesian control volumes – Courtesy of
[Schafer] ................................................................................................................................................................................. 158
Figure 6.11 Interpolation of values in CV edges for discretization of diffusive fluxes for non-
Cartesian CV – Courtesy of [Schafer] ......................................................................................................................... 159
Figure 6.12 Cartesian boundary CV at west boundary ................................................................................... 160
Figure 6.13 Coordinate System and C-type Control Volume ........................................................................ 165
Figure 6.14 Collocated Grid Arrangement ........................................................................................................... 166
Figure 6.15 Cp Contours for NACA 0012 Hydrofoil at 6 Degree Incidence ............................................ 169
 17

Figure 6.16 Potent Numerical Errors in CFD ...................................................................................................... 171

Figure 6.17 Effects a) exact b) 1ST order (Dissipation) c) 2nd order (Dispersion) . 171
Figure 6.18 Progression of Shock Wave on Supersonic Flow over a Backward Step ........................ 172
Figure 6.19 Effect of retaining TE terms ............................................................................................................... 172
Figure 6.20 Correlation Between CD and UD Schemes for Different (Pe) Number ............................ 173
Figure 6.21 Ratio of Convection Vs Diffusion - Effect of (Pe) number ..................................................... 173
Figure 6.22 Spatial effects of (Pe) number on 1D flow ................................................................................... 174
Figure 6.23 Cell Centered and Node Centered ................................................................................................... 175
Figure 6.24 Control Volume for spatial discretization of staggered grid vs collocated .................... 175
Figure 6.25 Wall Function Relation in Boundary Layer ................................................................................. 177
Figure 6.26 Near Wall Velocity Profile .................................................................................................................. 177
Figure 6.27 Grid 1 - Tetrahedral Cell Growth(Left) vs, Grid 2 – Hybrid Prism Layer Growth (Right)
.................................................................................................................................................................................................... 178
Figure 6.28 Drag Force (N) on the Pilot and Stoker using Different Grids ............................................. 178
Figure 6.29 CP for Different Grids ............................................................................................................................ 179
Figure 6.30 Drag/ Lift and Lateral Forces (N) Acting on the Pilot and the Stoker .............................. 180
Figure 6.31 Stencil for Cell-Centered Finite-Volume Discretization – (Courtesy of Nishikawa) . 183
Figure 6.32 Inviscid Flow over a Joukowsky Airfoil at M1 = 0.85 using Unstructured Mesh -
(Courtesy of Nishikawa) ................................................................................................................................................. 184
Figure 7.1 Mixed Boundary Conditions................................................................................................................. 187
Figure 7.2 Symmetry Plane to Model one Quarter of a 3D Duct ................................................................ 190
Figure 7.3 Pole (Axis) Boundary .............................................................................................................................. 194
Figure 7.4 Periodic Boundary ................................................................................................................................... 195
Figure 7.5 Schematic for HPT case setup. The computational grid is showing every fifteenth ..... 195
Figure 7.6 Overview of the Flow Field................................................................................................................... 196
Figure 7.7 Pressure contours plot for 2nd order spatial discretization scheme ................................... 197
Figure 7.8 Aero-Acoustics Application for NRBC’ ............................................................................................. 198
Figure 7.9 Sketch Exemplifying the conditions at a Free Surface Formed by the Interface Between
Two Fluids ............................................................................................................................................................................. 199
Figure 8.1 Numerical Stability of 1D Heat Equation using Explicit & Implicit Algorithms
(Wikipedia) ........................................................................................................................................................................... 204
Figure 8.2 Transient Solution Sequences ............................................................................................................. 207
Figure 8.3 Vortex shedding behind a cylinder run wrongly as a Steady State...................................... 208
Figure 8.4 Steady-State Force and Monitoring for a flow over a double sided membrane case ... 209
Figure 8.5 Contours of instantaneous vorticity field showing vortex shedding from the prisms 210
Figure 8.6 Flowchart of the Strong coupling of NS/VOF Flow Solver and 6DOF Model (Courtesy of
Nguyen & Park) ................................................................................................................................................................... 221
Figure 8.7 Snapshots of 3D dam-break flow: (a) experimental image and (b) simulation result at
0.5 s (Courtesy of Nguyen & Park) .............................................................................................................................. 221
Figure 8.8 Image sequences of the 3D evolution of a dam break in the case of closed downstream
walls, where the color of the free surface denotes the water level (red, z = 0.6 m; green, z = 0)
(Courtesy of Nguyen & Park) ......................................................................................................................................... 222
Figure 8.9 Visual Comparison of Simulated and Experimental Free-Surface Wave Profiles .......... 223
Figure 8.10 Wave profile Around a Surface-Piercing NACA 0024 Hydrofoil at Fr = 0.37obtained
from Grid Refinement ....................................................................................................................................................... 224
Figure 8.11 Grid and Wave Breaking by a NACA 66 Hydrofoil (Courtesy of Nguyen & Park) ....... 224
Figure 8.12 Impact Pressure Distributions over the Surface of a Water Entry Hemisphere over
time (Courtesy of V.T. Nguyen, W.-G. Park) ............................................................................................................ 225
Figure 8.13 Chimera Grids for Water Entry of a Sphere: (a) Curvilinear Background Grid (b)
Body-Fitted Grid (Courtesy of Nguyen & Park) ..................................................................................................... 226
 18

Figure 8.14 Center-depth variation for a free-falling sphere with time at various entry velocities
(Courtesy of Nguyen & Park) ......................................................................................................................................... 226
Figure 9.1 Physical domain for Laplace's equation .......................................................................................... 228
Figure 9.2 Illustration of a panel geometry; any three dimensional shape can be constructed;
shown here is a surface of what could be a three dimensional object such as an entire airplane ... 229
Figure 9.3 Cp Contours for a Rotor using Panel Method............................................................................... 232
Figure 9.4 Simulation of an Airplane Based in the VLM ................................................................................. 233
Figure 9.5 VCCTEF Configuration on a Generic Transport Model (GTM) ............................................... 234
Figure 9.6 VCCTEF Section with 3 Camber Segments (Green) Compared to a Traditional Flap
System (Blue) ....................................................................................................................................................................... 235
Figure 9.7 Illustration of the GTM Aircraft Equipped with the VCCTEF.................................................. 235
Figure 9.8 GTM Aircraft Wing-Body Model in Vorview.................................................................................. 236
Figure 9.9 TSFOIL Mesh Sensitivity Study; Lift (a) and Drag (b) .............................................................. 237
Figure 9.10 MSES Mesh Sensitivity Study ; Lift (a) and Drag (b) ............................................................... 238
Figure 9.11 Static Aeroelastic Modeling Framework for Fixed Angle of Attack (α) and Fixed Lift
Coefficient (b)....................................................................................................................................................................... 239
Figure 9.12 Streamwise Slices of the GTM wing for 2D Aerodynamic Analysis .................................... 240
Figure 9.13 Transonic and Viscous Correction Method Flow Chart ......................................................... 241
Figure 9.14 Lift Curve (Left) and Drag Polar (Right) for the Rigid GTM Using Inviscid Aero Solvers
.................................................................................................................................................................................................... 242
Figure 9.15 Lift Curve (Left) and Drag Polar (Right) for the Rigid GTM Using Viscous Aero Solvers
.................................................................................................................................................................................................... 243
Figure 9.16 Lift Curve (a) and Drag Polar (b) for the Aeroelastic GTM with Transonic and Viscous
Flow VLM Models ............................................................................................................................................................... 243
Figure 9.17 Aeroelastic Effect on Lift Distribution using TSD/IBL (a) and MSES (b) at CL = 0.497
.................................................................................................................................................................................................... 244
Figure 9.18 Contour Plots of Pressure Coefficient Obtained with the Transonic and Viscous
Potential Flow Models for the Flexible GTM at ...................................................................................................... 244
Figure 9.19 Lift Distribution with the Best Performing VCCTEF Candidate Calculated with
TSD/IBL (a) and MSES (b) .............................................................................................................................................. 245
Figure 9.20 Change in Cp due to the Best Performing VCCTEF Candidate Calculated with the
TSD/IBL (a) and MSES (b) Models .............................................................................................................................. 246
Figure 9.21 Wall Clock Comparison Between Multiple Aerodynamic Codes for a Single Fixed Lift
Aeroelastic Solution........................................................................................................................................................... 247
Figure 10.1 Characteristics of Linear Equation ................................................................................................. 252
Figure 10.2 Characteristics of Nonlinear solution point (exaggerated) .................................................. 254
Figure 10.3 Supersonic Flow Over Circular Arc with Inlet M=1.4 - Courtesy of [Mahdi and Al-
Kwarizmi] .............................................................................................................................................................................. 255
Figure 10.4 Coefficient of Pressure for a Shock ................................................................................................. 258
Figure 10.5 Comparison of dispersion error of compact-difference scheme with others methods
.................................................................................................................................................................................................... 263
Figure 11.1 Velocity Contours for Flow Over a Backward Facing Step ................................................... 275
Figure 11.2 Flow Field Over a Curved Ramp ...................................................................................................... 276
Figure 11.3 Vorticity-Stream Function Approach For 2D Incompressible N-S Equations .............. 279
Figure 11.4 SIMPLE Procedure as applied to 2D Equation ........................................................................... 282
Figure 11.5 Dual control volume for node-centered finite-volume schemes with unit normal
associated with an edge {j , k}. ...................................................................................................................................... 292
Figure 11.6 Irregular Triangular Grid for the Viscous Shock-Structure.................................................. 293
Figure 11.7 Comparison of Convergence for the 2D Problem. Stars: the Traditional Scheme.
Circles: the Hyperbolic Scheme .................................................................................................................................... 294
 19

Figure 11.8 Error Convergence Results for the Viscous Stresses and the Heat Fluxes in the 2D
problem .................................................................................................................................................................................. 295
Figure 11.9 Multi-Gridding Cycle............................................................................................................................. 298
Figure 11.10 Sequence of gridding in unstructured multigrid scheme ................................................. 299
Figure 11.11 Illustration of a node-centered median-dual control volume .......................................... 302
Figure 11.12 Typical implicit line-agglomeration showing a curved solid body surface on the left
and a symmetry plane on the right. The projection of the line-agglomerations can be seen on the
symmetry plane................................................................................................................................................................... 304
Figure 11.13 Trailing-edge area of a 3D wing agglomerated by the hierarchical scheme. Primal
grid is shown by thin lines; agglomerated grid is shown by thick lines. ..................................................... 304
Figure 11.14 Grids and convergence of the model diffusion equation for the F6 wing-body
combination .......................................................................................................................................................................... 305
Figure 11.15 Grids and Convergence of the Model Diffusion Equation for the DPW-W2 case ...... 306
Figure 11.16 Grids and Convergence for the wing-ap inviscid case. ........................................................ 307
Figure 11.17 Residual versus CPU time for the F6 wing-body case (RANS) ......................................... 313
Figure 12.1 Continuous vs. Discontinuous Galerkin Finite Element Method ....................................... 317
Figure 12.2 Illustration of Different Shape, Basis, and Blending Functions .......................................... 318
Figure 12.3 Mathematical Model of a Steady State Heat Sink...................................................................... 322
Figure 12.4 Basis functions sharing two triangular elements in 2D ......................................................... 324
Figure 12.5 Finite element approximation of the temperature field in the heat sink ....................... 325
Figure 12.6 Mixed Triangular–Quadrilateral Mesh for a Symmetric Airfoil with Curved
Boundaries and a Circular Sliding Mesh. Static and Rotating Subdomains are Distinguished........... 328
Figure 12.7 Snapshots of the Rotating NACA0015 for Polynomial Order k = 5 and Rotational
Speed Lx/U = 0.05 for (a) AOA = 17.2 deg. and (b) AOA = 28.6 deg ............................................................ 332
Figure 12.8 Solution Snapshots of Flow Through a 3D Cross-Flow Turbine: (a) 11 pressure
contours [_1.6:0.8] and (b) iso-surfaces of z-velocity w = ±1 x10-6 ............................................................... 333
Figure 12.9 Solution Snapshots of a 3 Bladed Cross-Flow Turbine for Polynomial Order k = 3 for
(a) Unconstrained and (b) Ducted Cases. ................................................................................................................. 333
Figure 12.10 Solution Snapshots of a 3D Circular Cylinder Shadowed by a Rotating NACA Blade
(a) Pressure Contours (b) Iso-surfaces of z velocity w = ±1.5 x 10-6............................................................. 334
Figure 12.11 Conceptual Differences Between Three Most used Prediction Methods ..................... 336
Figure 12.12 Propagation Mean Flow Residuals on Different Mesh Density (Courtesy of Pandya, et
al.) ............................................................................................................................................................................................. 338
Figure 12.13 Convergence of Cp along symmetry plane (y = 0) using Tetrahedral cells ................. 340
Figure 13.1 Placement of Unknown ( Red Dot) and Flux (Blue ) Points for a Triangular Element;
From left to right - Case (a): First Order; Case (b): Second Order; Case (c): Third Order.................... 342
Figure 13.2 Partitions of a Triangular SV . Linear, Quadratic and Cubic Reconstructions Publicized
in Case (a), Case(b) and Case(c) respectively ......................................................................................................... 343
Figure 13.3 Runner Vane of the Kaplan Turbine............................................................................................... 345
Figure 13.4 3D Geometry of the prototype Kaplan Turbine......................................................................... 346
Figure 13.5 The Block Diagram ................................................................................................................................ 347
Figure 13.6 Tip Cearance Gap of the Kaplan Turbine ..................................................................................... 347
Figure 13.7 Mesh Independence Test of the Kaplan Turbine ...................................................................... 348
Figure 13.8 Turbine Domain for Computational Analysis ............................................................................ 350
Figure 13.9 Comparison Between Computed and Experimental Results as a Function of Runner
Vane.......................................................................................................................................................................................... 350
Figure 13.10 Tip Clearance (a) Velocity and (b) Pressure Profiles Between the Runner Blade and
the ............................................................................................................................................................................................. 351
Figure 13.11 Blade-to-Blade Velocity Contours ................................................................................................ 351
Figure 13.12 Distribution of the Turbulence Kinetic Energy at Draft Tube .......................................... 352
 20

Figure 13.13 Partial Sums of the Fourier Series of a Square Wave ........................................................... 353
Figure 13.14 Typical Cross Sectional View of a Square Channel Showing Wall Bisectors and
Corner Bisectors (Courtesy of Sriramkrishnan).................................................................................................... 356
Figure 13.15 Flow Regimes for Diluted Gas ........................................................................................................ 358
Figure 13.16 Simulations Spectrum of Rarefied Gas ....................................................................................... 359
Figure 13.17 Position and velocity vector for a particle after and before applying a force, F ....... 361
Figure 13.18 Real Molecules versus LB particles.............................................................................................. 363
Figure 13.19 Lattice Arrangements for Velocity Vectors for Typical 1D, 2D and 3D Discretization
.................................................................................................................................................................................................... 364
Figure 13.20 Schematics of Solving 2D Lattice Boltzmann Model ............................................................. 365
Figure 13.21 Schematic Diagram of the D2Q9 Lattice .................................................................................... 367
Figure 13.22 Centerline(x/X=0.5) Temperature Evolution for Different Instants (Case 1) ........... 368
Figure 13.23 Comparison of Centerline (x/X=0.5) Temperature in the Presence/Absence of Heat
Generation (Case 2) - Courtesy of [Chaabane et al.] ........................................................................................... 369
Figure 13.24 Centerline(x/X=0.5) Temperature Evolution for Different Instants (Case3) -
Courtesy of [Chaabane et al.] ........................................................................................................................................ 370
Figure 13.25 Color plot of the norm of the velocity: Re = 1000, ν = 1/18, τ = 2/3, 256 x 256 lattice
.................................................................................................................................................................................................... 370
Figure 13.26 Immersed Boundaries ........................................................................................................................ 371
Figure 13.27 Immersed Boundary illustration; Eulerian Mesh (¯x) and Lagrangian Mesh ( ¯ xk) [53]
.................................................................................................................................................................................................... 372
Figure 13.28 Hierarchy of IB Methods .................................................................................................................. 373
Figure 13.29 Effect of Explicitly Adding a Forcing Function, [42]. ............................................................ 374
Figure 13.30 (a) Schematic showing a generic body past which flow is to be simulated. The body
occupies the volume Ωb with boundary Гb. The body has a characteristic length scale L, and a
boundary layer of thickness δ develops over the body. (b) Schematic of body immersed in a
Cartesian grid on which the governing equations are discretized................................................................. 375
Figure 13.31 Schematic Interpolation of Mohd-Yusof ethod, [70]. ........................................................... 378
Figure 13.32 Treatment of the Interface Cells; Forcing is Applied on the Filled Circles. [1]. ......... 379
Figure 13.33 Schematic interpolation of Zhang method, which is the same as Saiki et al. [50], [70].
.................................................................................................................................................................................................... 380
Figure 13.34 Implementation of the zero gradient pressure condition on a collocated grid [37].
.................................................................................................................................................................................................... 381
Figure 13.35 Different schemes for defining control volumes for mass conservation near the
immersed boundary. [27] ............................................................................................................................................... 381
Figure 13.36 Schematic of computational domain with immersed boundaries. [69]. ...................... 383
Figure 13.37 Schematic of interpolation for cell face values and derivative at boundary cell;
Points ....................................................................................................................................................................................... 384
Figure 13.38 A master and slave cell are shown for the u component of velocity. ............................. 385
Figure 13.39 Difference between Chung and Ye [5]. ....................................................................................... 386
Figure 13.40 Flow structures in the near wake behind a sphere: (a) Re = 104 (IB simulation); (b)
Re = 1.5 × 104 (experiment). .......................................................................................................................................... 387
Figure 14.1 A typical CFD Simulation .................................................................................................................... 392
Figure 14.2 Streamlined CFD Simulation Process ............................................................................................ 394
Figure 14.3 Projectile Simulation (BMI Corporation) ..................................................................................... 400
Figure 14.4 Global view of hemisphere cylinder geometry and boundary conditions ..................... 403
Figure 14.5 Global View of Hemisphere Cylinder Pressure Contours using L1 grid at surfaces y = 0
(left) and x = 6 (right) ....................................................................................................................................................... 404
Figure 14.6 Grid Convergence of Aerodynamic Forces for Hemisphere Cylinder .............................. 405
 21

Figure 14.7 Global View of Surface Pressure and Skin Friction at symmetry plane (y = 0) for
Hemisphere Cylinder ........................................................................................................................................................ 407
Figure 14.8 M6 wing: pressure contours computed by USM3D on family 4 prism/hex L1 grid... 408
Figure 14.9 M6 Grid Convergence of Aerodynamic Forces CL, CD .............................................................. 409
Figure 14.10 M6 Grid Convergence of Pitching Moment ............................................................................... 410
Figure 14.11 M6 section (η=x/c=0.2) Global View of leeside Pressure Grid Refinement ................ 411
Figure 15.1 Component of Uncertainty within a Simulation........................................................................ 414
 22

1 Introduction
Computational fluid dynamics, usually abbreviated as CFD, is a branch of fluid mechanics that uses
numerical analysis and algorithms to solve and analyze problems that involve fluid flows. Computers
are used to perform the calculations required to simulate the interaction of liquids and gases with
surfaces defined by boundary conditions. With high-speed supercomputers, better solutions can be
achieved. Ongoing research yields software that improves the accuracy and speed of complex
simulation scenarios such as transonic or turbulent flows. Initial experimental validation of such
software is performed using a wind tunnel with the final validation coming in full-scale testing, e.g.
flight tests1. In design and development, CFD programs are now considered to be standard numerical
tools which predict not only fluid flow behavior, but also the transfer of heat, mass (such as in
perspiration or dissolution), phase change (such as in freezing, melting or boiling), chemical reaction
(such as combustion or rusting), mechanical movement (such as an impeller turning, pistons, fans or
rudders) and stress or deformation of related solid structures (such as a mast bending in the wind).
Furthermore, CFD has been applied to deal with problems in environment and architecture. Figure
1.1 summarizes the scope of CFD application2.

Industrial Environmental Physiological

applications applications applications
•Aerospace •Atmospheric pollution •Cadiovascular flows
•Architecture •Climate calculations (heart, major vessels)
•Automotive •Fire in buildings •Flow in lungs and
•Biomedical •Oceanic flows breathing passages
•Chemical Process •Pollution of natural
•Combustion waters
•Electronics and •Safety
computers
•Food Processing
•Glass manufacturing
•HVAC (heat, ventilation
and cooling)
•Petroleum
•Power
•Marine
•Mechanical
•Metallurgical
•Nuclear
•Train design
•Turbo Machinery
•Water Treatment

Figure 1.1 CFD Application in Various Industries

1From Wikipedia.
2Bin Xia, Da-Wen Sun, “Applications of computational fluid dynamics (CFD) in the food industry: a review”,
Computers and Electronics in Agriculture 34 (2002) 5–24.
 23

1.1 Fluid & Aerodynamic Flow Regions As Characterized By Mach Number

In fluid dynamics, the Mach number (M or Ma) is a dimensionless quantity representing the ratio of
flow velocity past a boundary to the local speed of sound 3-4,

V
M=
a
Eq. 1.1
Where M is the Mach number, V is the local flow velocity with respect to the boundaries (either
internal, such as an object immersed in the flow, or
external, like a channel), and a is the speed of sound in
the medium. The local speed of sound, and thereby the
Mach number, depends on the condition of the
surrounding medium, in particular the temperature and
pressure. Figure 1.2 shows an F/A-18 creating a vapor
cone at transonic speed just before reaching Mach 1 (By
Ensign John Gay, U.S. Navy). The Mach number is
primarily used to determine the approximation with
which a flow can be treated as an incompressible flow.
The medium can be a gas or a liquid.
While the terms "subsonic" and "supersonic," in the
purest sense, refer to speeds below and above the local
speed of sound respectively, aerodynamicists often use
the same terms to talk about particular ranges of Mach Figure 1.2 An F/A-18 Hornet Creating
values. This occurs because of the presence of a a Vapor Cone at Transonic Speed
"transonic regime" around M = 1 where approximations
of the Navier-Stokes equations used for subsonic design actually no longer apply; the simplest
explanation is that the flow locally begins to exceed M = 1 even though the freestream Mach number
is below this value. Meanwhile, the "supersonic regime" is usually used to talk about the set of Mach
numbers for which linearized theory may be used, where for example the (air) flow is not chemically
reacting, and where heat-transfer between air and vehicle may be reasonably neglected in
calculations. In the following table (see Table 1.1), the "regimes" or "ranges of Mach values" are
referred to, and not the "pure" meanings of the words "subsonic" and "supersonic". Generally, NASA
defines "high" hypersonic as any Mach number from 10 to 25, and re-entry speeds as anything

Flow Regime Mach knots mph Km/h m/s

Subsonic < 0.8 < 530 < 609 < 980 < 273
Transonic 0.8 - 1.2 530-794 609-914 980-1470 273-409
Supersonic 1.2 - 5.0 794-3308 915-3806 1470-6126 410-1702
Hypersonic 5.0 - 10.0 3308-6615 3806-7680 6125-12251 1702-3403
High-Hypersonic 10.0 -24.0 6615-16537 7680-19031 12251-30626 3403-8508
Re-Entry Speeds > 24.0 > 16537 > 19031 > 30626 > 8508

Table 1.1 Classification of Mach Number

3 Young, Donald F.; Bruce R. Munson; Theodore H. Okiishi; Wade W. Huebsch (2010). A Brief Introduction to
Fluid Mechanics (5 Ed.). John Wiley & Sons. p. 95.
4 Graebel, W.P. (2001). Engineering Fluid Mechanics. Taylor & Francis. p. 16.
 24

greater than Mach 24.

Aircraft operating in this
regime include the Space
Shuttle and various space
planes in development.
Further details regarding
the Mach number regimes
can be obtained from
[Anderson]5.
The local speed of sound,
and thereby the Mach
number, depends on the
condition of the
surrounding medium, in
particular the temperature
and pressure.
In summary, we attempt to
organize our study of
aerodynamic flows
according to one or more of
the various categories
discussed. The block Figure 1.3 Block Diagram Categorizing the Types of
diagram in Figure 1.3 is Aerodynamic Flows (Courtesy of John D. Anderson)
presented to help
emphasize these categories
and to show how they are related. Looking at the whole field simultaneously, four different speed
regimes can be identified using Mach number as the criterion:
1.1.1 Subsonic Flow (M < 1 everywhere)
A flow field is defined as subsonic if the Mach number is less than 1 at every point. Subsonic flows
are characterized by smooth streamlines (no discontinuity in slope), as sketched in Figure 1.4(a).
Moreover, since the flow velocity is everywhere less than the speed of sound, disturbances in the flow
(say, the sudden deflection of the trailing edge of the airfoil in Figure 1.4(a)) propagate both
upstream and downstream, and are felt throughout the entire flow field. Note that a freestream Mach
number M∞ less than 1 does not guarantee a totally subsonic flow over the body. In expanding over
an aerodynamic shape, the flow velocity increases above the freestream value, and if M ∞ is close
enough to 1, the local Mach number may become supersonic in certain regions of the flow. This gives
rise to a rule of thumb that M∞ < 0.8 for subsonic flow over slender bodies. For blunt bodies, M∞ must
be even lower to ensure totally subsonic flow. (Again, emphasis is made that the above is just a loose
rule of thumb and should not be taken as a precise quantitative definition). Also, we will show later
that incompressible flow is a special limiting case of subsonic flow where M → 0.
1.1.2 Transonic Flow (Mixed Regions M < 1 & M > 1)
As stated above, if M∞ is subsonic but is near unity, the flow can become locally supersonic (M > 1).
This is sketched in Figure 1.4(b), which shows pockets of supersonic flow over both the top and
bottom surfaces of the airfoil, terminated by weak shock waves behind which the flow becomes
subsonic again. Moreover, if M∞ is increased slightly above unity, a bow shock wave is formed in front
of the body; behind this shock wave the flow is locally subsonic, as shown in Figure 1.4(c). This
subsonic flow subsequently expands to a low supersonic value over the airfoil. Weak shock waves

5 John D. Anderson, “Fundamentals of Aerodynamics”, McGraw Hill Inc. pp.37-39, 1976.

 25

are usually generated at the trailing edge, sometimes in a “fishtail” pattern as shown in Figure
1.4(c). The flow fields shown in Figure 1.4(b) and (c) are characterized by mixed subsonic-
supersonic flows and are dominated by the physics of both types of flow. Hence, such flow fields are
called transonic flows. Again, as a rule of thumb for slender bodies, transonic flows occur for
freestream Mach numbers in the range 0.8 < M∞ < 1.2.

Figure 1.4 Different Regimes of Flow (Courtesy of John D. Anderson)

1.1.3 Supersonic Flow (M > 1 Everywhere)

A flow field is defined as supersonic if the Mach number is greater than 1 at every point. Supersonic
flows are frequently characterized by the presence of shock waves across which the flow properties
 26

and streamlines change discontinuously. This is illustrated in Figure 1.4(d) for supersonic flow over
a sharp-nosed wedge; the flow remains supersonic behind the oblique shock wave from the tip. Also
shown are distinct expansion waves, which are common in supersonic flow. (Again, the listing of M∞
> 1.2 is strictly a rule of thumb. For example, in Figure 1.4(d), if θ is made large enough, the oblique
shock wave will detach from the tip of the wedge and will form a strong, curved bow shock ahead of
the wedge with a substantial region of subsonic flow behind the wave. Hence, the totally supersonic
flow sketched in Figure 1.4(d) is destroyed if θ is too large for a given M∞. This shock detachment
phenomenon can occur at any value of M∞ > 1, but the value of θ at which it occurs increases as M∞
increases. In turn, if θ is made infinitesimally small, the flow field in Figure 1.4(d) holds for M∞ ≥
1.0.
The above discussion clearly shows that the listing of M∞ > 1.2 in Figure 1.4(d) is a very tenuous
rule of thumb and should not be taken literally.) In a supersonic flow, because the local flow velocity
is greater than the speed of sound, disturbances created at some point in the flow cannot work their
way upstream (in contrast to subsonic flow). This property is one of the most significant physical
differences between subsonic and supersonic flows. It is the basic reason why shock waves occur in
supersonic flows, but do not occur in steady subsonic flow6.
1.1.4 Hypersonic Flow (Very High Supersonic Speeds)
Refer again to the wedge in Figure 1.4(d). Assume θ is a given, fixed value. As M∞ increases above
the shock wave moves closer to the body surface. Also, the strength of the shock wave increases,
leading to higher temperatures in the region between the shock and the body (the shock layer). If
M∞ is sufficiently large, the shock layer becomes very thin, and interactions between the shock
wave and the viscous boundary layer on the surface occur. Also, the shock layer temperature
becomes high enough that chemical reactions occur in the air. The O2 and N2 molecules are torn apart;
that is, the gas molecules dissociate. When M∞ becomes large enough such that viscous interaction
and/or chemically reacting effects begin to dominate the flow (Figure 1.4(e)), the flow field is called
hypersonic. (Again, a somewhat arbitrary but frequently used rule of thumb for hypersonic flow is
M∞ > 5). Hypersonic aerodynamics received a great deal of attention during the period 1955 –1970
because atmospheric entry vehicles encounter the atmosphere at Mach numbers between 25
(ICBMs) and 36 (the Apollo lunar return vehicle). Again during the period 1985–1995, hypersonic
flight received a great deal of attention with the concept of air-breathing supersonic-combustion
ramjet-powered trans atmospheric vehicles to provide single-stage-to-orbit capability. Today,
hypersonic aerodynamics is just part of the whole spectrum of realistic flight speeds7.

1.2 Motivation
Imagine that you are an aeronautical engineer in the later 1950s. You have been given the task of
designing an atmospheric entry vehicle in those days it would have been an intercontinental ballistic
missile8. You are well aware of the fact that such vehicles will enter the earth’s atmosphere at very
high velocities, about 7.9 km/s for entry from earth orbit and about 11.2 km/s for entry after
returning from a lunar mission. At these extreme hypersonic speeds, aerodynamic heating of the
entry vehicle becomes very severe, and is the dominant concern in the design of such vehicles.
Therefore, you know that your task involves the design of a blunt body for hypersonic speed.
Moreover, you know from supersonic wind tunnel experiments that the flow field over the blunt body
is qualitatively like that sketched in Figure 1.5. You know that a strong curved bow shock wave sits
in front of the blunt nose, detached from the nose by the distance δ, called the shock detachment

6 John D. Anderson,Jr, Professor Emeritus University of Maryland, “Fundamentals of Aerodynamics”, 5th edition,
The McGraw-Hill Companies, Inc., 1221 Avenue of the Americas, New York, NY 10020, 2011.
7 See Previous.
8 John D. Anderson Jr., Joris Degroote, G´erard Degrez, Erik Dick, Roger Grundmann and Jan Vierendeels,

“Computational Fluid Dynamics - An Introduction”, 3rd Edition, ISBN: 978-3-540-85055-7, 2009.

 27

distance. You know that the gas temperatures between the shock and the body can be as high as
7000K for an ICBM, and 11000K for entry from a lunar mission.
And you know that you must understand some of the details of this flow field in order to intelligently
design the entry vehicle. So, your first logical step is to perform an analysis of the aerodynamic flow
over a blunt body in order to provide detailed information on the pressure and heat transfer
distributions over the body surface, and to examine the properties of the high-temperature shock
layer between the bow shock wave and the body. You ask such questions as: what is the shape of the
bow shock wave; what is the detachment distance δ; what are the velocity, temperature and pressure
distributions throughout the shock layer, etc.? However, much to your dismay, you find that no
reliable, accurate aerodynamic theory exists to answer your questions. You quickly discover that an
accurate and practical analysis of supersonic blunt body flows is beyond your current state-of-the-
art. As a result, you ultimately resort to empirical information along with some simplified but
approximate theories (such as Newtonian theory) in order to carry out your designated task of
designing the entry vehicle.
The above paragraph illustrates one of the most important, yet perplexing, aerodynamic problems of
the 1950s and early 1960s. The application of blunt bodies had become extremely important due to
the advent of ICBMs, and later
the manned space program. Yet,
no aerodynamic theory existed
to properly calculate the flow
over such bodies. Indeed, entire
sessions of technical meetings
(such as meetings of the Institute
for Aeronautical Sciences in the
USA, later to become the
American Institute for
Aeronautics and Astronautics)
were devoted exclusively to
research on the supersonic blunt
body problem. Moreover, some
of the best aerodynamicists of
that day spent their time on this
problem, funded and strongly
encouraged by the NACA (later Figure 1.5 Flow over a Supersonic Blunt Body
NASA), the US Air Force and
others. What was causing the
difficulty? Why was the flow field over a body moving at supersonic and hypersonic speeds so hard
to calculate? The answer rests basically in the sketch shown in Figure 1.5, which illustrates the
steady flow over a supersonic blunt body. The region of steady flow near the nose region behind the
shock is locally subsonic, and hence is governed by elliptic partial differential equations. In contrast,
the flow further downstream of the nose becomes supersonic, and this locally
steady supersonic flow is governed by hyperbolic partial differential equations. (What is meant by
‘elliptic’ and ‘hyperbolic’ equations, and the mathematical distinction between them, will be
discussed them later). The dividing line between the subsonic and supersonic regions is called the
sonic line, as sketched in Figure 1.5.
The change in the mathematical behavior of the governing equations from elliptic in the subsonic
region to hyperbolic in the supersonic region made a consistent mathematical analysis, which
included both regions, virtually impossible to obtain. Techniques were developed for just the
 28

subsonic portion, and other techniques (such as the standard ‘method of characteristics’) were
developed for the supersonic region. Unfortunately, the proper patching of these different techniques
through the transonic region around the sonic line was extremely difficult. Hence, as late as the mid-
1960s, no uniformly valid aerodynamic technique existed to treat the entire flow field over the blunt
body. This situation was clearly noted in the classic textbook by [Liepmann and Roshko]9 published
in 1957, where in a discussion of blunt body they state: The shock shape and detachment distance
cannot, at present, be theoretically predicted. The purpose of this lengthy discussion on the status
of the blunt body problem in the late 1950s is to set the background for the following important point.
In 1966, a breakthrough occurred in the blunt body problem. Using the developing power of
Computational Fluid Dynamics (CFD) at that time, and employing the concept of a ‘time-dependent’
approach to the steady state, [Moretti and Abbett]10 developed a numerical, finite-difference solution
to the supersonic blunt body problem which constituted the first practical, straightforward
engineering solution for this flow. After 1966, the blunt body problem was no longer a real problem.
Industry and government laboratories quickly adopted this computational technique for their blunt
body analyses. Perhaps the most striking aspect of this comparison is that the supersonic blunt body
problem, which was one of the most serious, most difficult, and most researched theoretical
aerodynamic problems of the 1950s and 1960s, is today assigned as a homework problem in a
computational fluid dynamics graduate course at the University of Maryland. Therein lies an example
of the power of Computational Fluid Dynamics (CFD). The purpose of these notes is to provide an
introduction to computational fluid dynamics. The above example concerning blunt body flows
serves to illustrate the importance of computational fluid dynamics to modern aerodynamic
applications. Here is an important problem which was impossible to solve in a practical fashion
before the advent of computational fluid dynamics (CFD), but which is now tractable and
straightforward using the modern techniques of CFD. Indeed, this is but one example out of many
where CFD is revolutionizing the world of aerodynamics. The purpose of the present author writing
these notes, and your reading these notes and attending the VKI short course, is to introduce you to
this revolution.

1.3 CFD Processes

1.3.1 Partial Differential Equations (PDEs) & Their Methods of Solution
PDEs are mathematical models of continuous physical phenomenon in which a dependent variable,
say u, is a function of more than one sovereign variable, say t (time), and x (spatial position). PDEs
derived by applying a physical principle such as conservation of mass, momentum or energy. These
equations, governing the kinematic and mechanical behavior of general bodies are referred to as
Conservation Laws. These laws can be written in either the strong of differential form or an
integral form. Numerical methods for partial differential equations is the branch of numerical
analysis that studies the numerical solution of partial differential equations (PDEs). Table 1.2 shows
the different numeric methods of solving the PDE’s. In subsequent chapters we will acquainted with
most of their terminology.
1.3.2 Boundary Conditions
Conditions on the (finite) boundary of the domain and/or initial conditions (for transient problems)
are required to obtain for a well posed problem. The Well Posed Problems indicate that conditions
on the (finite) boundary of the domain and/or initial conditions (for transient problems) are required
to obtain a well posed problem. The properties of a well posed problem are:

9Liepmann, H.W. and Roshko, A., “Elements of Gas dynamics”, Wiley, New York, 1957.
10Moretti, G. and Abbett, M., ‘A Time-Dependent Computational Method for Blunt Body Flows,’ AIAA Journal,
Vol. 4, No. 12, December 1966, pp. 2136–2141.
 29

1. Solution exists
2. Solution is unique
3. Solution depends continuously on the data

Numerical PDEs

Finite Difference Finite Volume Finite Element Meshless/Meshfree

Elliptic and
Parabolic Hyperbolic Godunov hp-FEM Smoothed-particle
Others
hydrodynamics (SPH)
Forward- Lax– Alternating High-
time Friedrichs direction- Extended
resolution Moving particle
central- implicit (ADI) (XFEM)
semi-implicit
space method (MPS)
Lax– Monotonic
(FTCS) Wendroff Finite- upstream- Discontinuous
difference centered Galerkin (DG) Material point
Crank– time-domain
MacCormack (MUSCL) method (MPM)
Nicolson (FDTD)
Spectral
Advection element (SEM) Particle-in-
upstream- cell (PIC)
Upwind
splitting
Mortar
(AUSM)
Method of
characteristics Gradient discretisation
Riemann solver
(GDM)

essentially non-
Loubignac
oscillatory
iteration
(ENO)

Smoothed (S-FEM)
weighted
essentially non-
oscillatory
(WENO)

Table 1.2 Solution Methods for PDEs

1.3.3 Math Modeling and Simulation of Physical Processes

Altogether, the CFD process can be lumped together as:
1. Define the physical problem
2. Create a mathematical (PDE) model
3. Systems of PDEs, ODEs, algebraic equations
4. Define Initial and or boundary conditions to get a well-posed problem
5. Create a Discrete (Numerical) Model
6. Discretize the domain, generate the grid and obtain discrete model
7. Solve the discrete system
8. Analyze Errors in the discrete system
9. Consistency, stability and convergence analysis
 30

1.3.4 Numerical Algorithms Evolution

Following is a direct replica from paper by [J.S. Shang]11. The numerical algorithm is the heart of
computational fluid dynamic, because it is the necessary translator between numerical analysis for
fluid dynamics via computers. The history of CFD is also ultimately related to the development of
programmable digital computers: In 1833 Charles Baggage originated the idea of a programmable
computer, but the first patent for the ENIAC computer (electronic numerical integrator and
computer) was recorded in 1947 for the truly programmable computer using transistors.

Figure 1.6 Relationship Between Physics Fidelity & Numerical Algorithm

The interrelation between numerical algorithms and computational results is depicted by a graphic
presentation in Figure 1.6. The illustrated algorithm is the diminishing residue return (DRR)
scheme; the right-hand-side of the conservative law represents the physics to be simulated. The left-
hand-side of the equation is the numerical process and its sole purpose is keeping a stable
computation. In fact, the illustration also implies the equivalent principle held for which if a stable
numerical algorithm leading to a converged asymptote, the numerical result is ensured to be the
unique solution.
There are two entirely different concepts for CFD formulations, and the most widely adopted
approach is the Eulerian frame of reference. In this formulation the fluid dynamics is analyzed in a
control volume fixed in space. Whereas, the Lagrangian approach is analyzing fluid dynamics by
following a group moving gas particles in an enclosed control volume. The well-known direct
simulation Monte Carlo (DSMC) method is built on the Largangian formulation, together with the
particle-in-cell (PIC) method by Harlow [10]. For the PIC method, the fluid dynamics is represented
by Lagrangian mass particles within a control volume. At each time step, the calculated internal

11Joseph J. S. Shang, “Landmarks and new frontiers of computational fluid dynamics”, Shang Advances in
Aerodynamics (2019) 1:5, https://doi.org/10.1186/s42774-019-0003-x
 31

energy and velocity are obtained and the conservation properties are checked by the sum of these
final values before the process advances to the next time level. The DSMC and PIC methods have
demonstrated to be well suited to study the time dependent and multi-spices fluid medium, and had
been widely used for simulating rarefied gas dynamics and plasma dynamics [11].
The most predominant CFD algorithm pioneers are led by Richardson who introduced point iterative
scheme to solve the elliptic partial differential equation as far back as 1910 [12]. Then Courant,
Friedrichs and Lewy initiated the rigorous investigation procedure for examining the stability of a
numerical algorithm by Fourier analysis in 1928. They also addressed the uniqueness and existence
of the numerical results for partial differential equations [13]. It was Southwell who introduced a
relaxation scheme to solve both the fluid dynamic and structure problem to become an accepted
procedure for engineering application in 1940 [14]. Lax [15] and Godunov [16] addressed the most
challenging and difficult issues in numerical analyses for resolving discontinuous fluid phenomena
in a discrete space the approximate Riemann problem. As it will be seen later, it remains to be the
most studied problem in CFD.
In the early 1960s, the dominated numerical algorithms are mostly explicit schemes, such as the Lax-
Wendroff, leap-frog, and fractional step methods for multi-dimensional problems [17]. When CFD
ventures into increasingly complex fluid phenomena, the more efficient and stable implicit schemes
are required. Especially, the ADI method [18, 19] has been effectively applied to all type of partial
equations, except when applying to the time-dependent, three-dimensional hyperbolic system for
which some forms of artificial dissipative terms must be appended to maintain computational
stability. This shortcoming is removed by finite-volume formulation with an iterative solving scheme.
In the subsequent developments, the basic ADI scheme has evolved into the strong implicit scheme
(SIP) and the diminishing residual return (DRR) formulations. The matrix inversion procedures for
ADI have also expanded to include the lower-upper (LU) decomposition technique, as well as, the
line, Jacobi, Gauss-Seidel, conjugate relaxation procedures. In short, the ADI scheme still remains as
the mainstay for most multi-dimensional solving procedure even to date [20].
The increasing demand of greater numerical efficiency and maintaining computational stability was
met by the multi-grid approach [21]. The convergence rate of an iterative scheme has a close tie to
the spectral radius of eigenvalues of the equation system and the residue error reduction process
from the initial estimate. For the multi-grid approach, the data is processing through a sequence of
different grid-point spacings in the computational domain to filter out the undesirable high
frequency, unstable Fourier components from the solution. Another popular approach is adopting
the unstructured grid technique to cluster cells where the high grid density is needed; the Delaunay
Scheme [22] was introduced first for generating two-dimensional triangular and three-dimensional
tetrahedron meshes.
From the analytic geometry viewpoint, the pyramidal control volume topology is natural to uniquely
define the control volume. The numerical results of an unstructured grid however can be inaccurate
by introducing numerous artificial slip streams in evaluating the viscous-inviscid interaction from
multiple intersecting shock waves. This issue and the high-order method development for the
unstructured grid method remain as the mainstay for CFD research into the future.
A major pacing item for CFD adopting shock capturing techniques is evaluating discontinuous
solution generated by shock waves and slip streams from shock interceptions. A breakthrough by
Godunov demonstrates a multi-dimensional flow field that contains shock waves and contact
surfaces can still be analyzed [16]. The discontinuities of the hyperbolic differential systems are
treated as a piecewise continuous data distribution within a control volume and to be solved across
the singular point as the Riemann problem. The underpinning principle is the monotonicity
preserving property of the hyperbolic difference equation; namely, temporal increment/decrement
of dependent variable is monotonic. Based on this property, Harten originates the total variation
diminishing (TVD) scheme and spans off a huge amount of research on TVD schemes and a variety of
flux limiters for analyzing piecewise discontinuous solutions for CFD [23].
 32

From the physics viewpoint, the treatment of shock jump by flux splitting can be easily understood
through the concept of zone of dependence for supersonic flows. By solving a set of Riemann
problems over the entire computational domain according to their distinctive characteristics, this
approach actually honors the physics of domain of influence. The directional wave propagation is
constructed according to the phase velocity from the permissible database. In an outstanding work
by Steger and Warming, it has shown a systematic relationship of the real eigenvalue and eigenvector
for the split flux formulation. They also demonstrate the Euler equations, together with the equation
of state for gas, possessing the homogeneous function of degree one property [24].
1.3.4.1 References
10. Harlow FH (1964) The particle-in-cell computing method for fluid dynamics, method in
computational physics, vol 3, p 319
11. Shang JS, Surzhikov ST (2018) Plasma dynamics for aerospace engineering. Cambridge
University Press, Cambridge, New York
12. Richardson LF (1910) The approximate arithmetical solution by definite differences of physical
problems involving differential equations, with an application to the stresses in a masonry dam.
Philos Trans Res Soc London, Series A 210: 307–357
13. Courant R, Friederichs KO, Lewy H (1928) Uber die partiellen differenzengleichungen der
mathematischen physik. Mathenmatishe Annalen 100:32–74
14. Southwell RV (1940) Relaxation method in engineering science. Oxford University Press, London
15. Lax PD (1954) Weak solution of nonlinear hyperbolic equations and their numerical computation.
Commun Pure Appl Math 7:159–163
16. Godunov SK (1959) Finite-difference method for numerical computational of discontinuous
solution of the equations of fluid dynamics. Mat Sb 47:271–306
17. Anderson DA, Tannehill JC, Pletcher RH (1984) Computational fluid mechanics and heat transfer,
2nd edn. Taylor & Francis, Bristol
18. Peaceman DW, Rachford HH (1955) The numerical solution of parabolic and elliptic differential
equations. J Soc Ind Appl Mat 3:28–41
19. Richtmyer RD, Morton KW (1967) Differential methods for initial-value problem, 2nd Ed. Inter
science Publishers, Wiley, New York
20. Shang JS (2009) Computational fluid dynamics application to aerospace science. Aeronaut J 113
(1148)
21. Brandt A (1973) Multi-level adaptive technique (MALT) for fast numerical solution to boundary
value problem, lecture notes in physics, vol 18. Springer-Verlag, Berlin, pp 82–89
22. Delaunay, B., Sur la Sphere Vide, Bull. Acad. Science, USSR, VII, Class. Sci. Mat. Nat. 793–800, 1934
23. Harten A (1983) High-resolution schemes for hyperbolic conservation Laws. J Comp Phys
49:375–385
24. Steger JL, Warming RF (1981) Flux vector splitting of the inviscid Gas dynamics equations with
application to finite difference methods. [ J Comp. Phys 40 . 263–293].
1.3.5 Verification, Validation & Calibration
Verification is a procedure to ensure that the program solves the equations correctly. Validation is
done to test how accurately the model represents reality, and Calibration is often used to adjust the
simulation to known experimental data in order to study parameter sensitivity in the design process.
The distinction between Verification and Validation is rather important. While validation is
concerned with checking that the system will meet the customer’s actual needs, verification is
concerned with whether the system is well-engineered, error-free, and so on. In other word,
Verification will help to determine whether the software is of high quality, but it will not ensure that
the system is useful. Verification includes all the activities associated with the producing high quality
software namely, testing, inspection, design analysis, specification analysis, and so on. It is a relatively
objective process, in that if the various products and documents are expressed precisely enough, no
 33

subjective judgements should be needed in order to verify software. In contrast, validation is an

extremely subjective process. It involves making subjective assessments of how well the (proposed)
system addresses a real-world need. Validation includes activities such as requirements modelling,
prototyping and user evaluation. Having thus carefully distinguished the two terms, the advice to
V&V practitioners was then to forget about the distinction, and think instead about V & V as a toolbox,
which provides a wide range of tools for asking different kinds of questions about software. And to
master the use of each tool and figure out when and how to use it. Figure 1.7 shows an attempts to
visualize such a toolbox12.

Figure 1.7 A range of V & V techniques and their interaction with each other

1.4 History and Success Stories

The fundamental basis of almost all CFD problems are the Navier–Stokes equations, which define
many single-phase (gas or liquid, but not both) fluid flows. These equations can be simplified by
removing terms describing viscous actions to yield the Euler equations. Further simplification, by
removing terms describing vorticity yields the full potential equations. Finally, for small
perturbations in subsonic and supersonic flows (not transonic or hypersonic) these equations can be
linearized to yield the linearized potential equations. Historically, methods were first developed to
solve the linearized potential equations. Two-dimensional (2D) methods, using conformal
transformations of the flow about a cylinder to the flow about an airfoil were developed in the
1930s13. One of the earliest type of calculations resembling modern CFD are those by Lewis Fry
Richardson, in the sense that these calculations used finite differences and divided the physical space
in cells. Although they failed dramatically, these calculations, together with Richardson's book
"Weather prediction by numerical process", set the basis for modern CFD and numerical
meteorology14.
The computer power available paced development of 3D. Probably the first work using computers to

12 Serendipity 2017®.
13 Milne-Thomson, L.M. (1973), “Theoretical Aerodynamics”, Dover Publications. ISBN 0-486-61980-X.
14 Hunt (1998). "Lewis Fry Richardson and his contributions to mathematics, meteorology, and models of conflict".

Annual Review of Fluid Mechanics.

 34

model fluid flow, as governed by the Navier-Stokes equations, was performed at Los Alamos National
Lab, in the T3 group15-16. This group was led by Francis H. Harlow, who is widely considered as one
of the pioneers of CFD. From 1957 to late 1960s, this group developed a variety of numerical methods
to simulate transient 2D fluid flows, such as Particle-in-cell method, Fluid-in-cell method17, Vorticity
stream function method, and Marker-and-cell method. Fromm's vorticity-stream-function method
for 2D, transient, incompressible flow was the first treatment of strongly contorting incompressible
flows in the world. The first paper with three-dimensional model was published by John Hess and
A.M.O. Smith of Douglas Aircraft in 1967. This method discretized the surface of the geometry with
panels, giving rise to this class of programs being called Panel Methods. Their method itself was
simplified, in that it did not include lifting flows and hence was mainly applied to ship hulls and
aircraft fuselages. The first lifting Panel Code was described in a paper written by [Paul Rubbert and
Gary Saaris] of Boeing Aircraft18. In time, more advanced three-dimensional Panel Codes were
developed at Boeing (PANAIR, A502), Lockheed (Quadpan), Douglas (HESS), McDonnell Aircraft
(MACAERO), NASA (PMARC) and Analytical Methods (WBAERO, USAERO and VSAERO). Some
(PANAIR, HESS and MACAERO) were higher order codes, using higher order distributions of surface
singularities, while others (Quadpan, PMARC, USAERO and VSAERO) used single singularities on each
surface panel. The advantage of the lower order codes was that they ran much faster on the
computers of the time. Today, VSAERO has grown to be a multi-order code and is the most widely
used program of this class. It has been used in the development of many submarines, surface ships,
automobiles, helicopters, aircraft, and more recently wind turbines. Its sister code, USAERO is an
unsteady panel method that has also been used for modeling such things as high speed trains and
racing yachts. The NASA PMARC code from an early version of VSAERO and a derivative of PMARC,
named CMARC, is also commercially available. Figure 1.8 pictures the simulation of the Hyper-X
scramjet vehicle in operation at Mach 7 developed by NASA.
In the two-dimensional realm, a number of Panel Codes have been developed for airfoil analysis and
design. The codes typically have a boundary layer analysis included, so that viscous effects can be
modeled. Professor Richard Eppler of the University of Stuttgart developed the PROFILE code, partly
with NASA funding, which became available in the early 1980s. This was soon followed by MIT
Professor Mark Drela's XFOIL code. Both PROFILE and XFOIL incorporate two-dimensional panel
codes, with coupled boundary layer codes for airfoil analysis work. PROFILE uses a conformal
transformation method for inverse airfoil design, while XFOIL has both a conformal transformation
and an inverse panel method for airfoil design.
An intermediate step between Panel Codes and Full Potential codes were codes that used the
Transonic Small Disturbance equations. In particular, the three-dimensional WIBCO code, developed
by Charlie Boppe of Grumman Aircraft in the early 1980s has seen heavy use. Developers turned to
Full Potential codes, as panel methods could not calculate the non-linear flow present at transonic
speeds. The first description of a means of using the Full Potential equations was published by Earll
Murman and Julian Cole of Boeing in 1970. Antony Jameson, originally at Grumman Aircraft and the
Courant Institute of NYU, worked with David Caughey to develop the important three-dimensional
Full Potential code FLO22 in 1975. Many Full Potential codes emerged after this, culminating in
Boeing's Tranair (A633) code, which still sees heavy use. The next step was the Euler equations,

15 "The Legacy of Group T-3". Retrieved March 13, 2013.

16 Harlow, F. H. (2004). "Fluid dynamics in Group T-3 Los Alamos National Laboratory:(LA-UR-03-3852)".
Journal of Computational Physics (Elsevier) 195 (2): 414–433.
17 Gentry, R. A., Martin, R. E., Daly, J. B. (1966). "An Eulerian differencing method for unsteady compressible

flow problems". Journal of Computational Physics 1 (1): 87–118.

18 Rubbert, Paul and Saaris, Gary, "Review and Evaluation of a Three-Dimensional Lifting Potential Flow Analysis

Method for Arbitrary Configurations," AIAA paper 72-188, presented at the AIAA 10th Aerospace Sciences
Meeting, San Diego California, January 1972.
 35

which promised to provide more accurate solutions of transonic flows. The methodology used by
Jameson in his 3D code (1981) was used by others to produce such programs as Lockheed's TEAM
program and Analytical Methods' MGAERO program. MGAERO is unique in being a structured
Cartesian mesh code, while most other such codes use structured body-fitted grids (with the
exception of NASA's highly successful CART3D code, Lockheed's SPLITFLOW code and Georgia Tech's
NASCART-GT).

Figure 1.8 Simulation of the Hyper-X Scramjet Vehicle in Operation by NASA

Antony Jameson also developed the 3D AIRPLANE code which made use of unstructured tetrahedral
grids. In the two-dimensional realm, Mark Drela and Michael Giles, then graduate students at MIT,
developed the ISES Euler program (actually a suite of programs) for airfoil design and analysis. This
code first became available in 1986 and has been further developed to design, analyze and optimize
single or multi-element airfoils, as the MSES program. MSES sees wide use throughout the world. A
derivative of MSES, for the design and analysis of airfoils in a cascade, is MISES, developed by Harold
"Guppy" Youngren while he was a graduate student at MIT. The Navier–Stokes equations were the
ultimate target of development. Two-dimensional codes, such as NASA Ames' ARC2D code first
emerged. A number of 3D codes were developed (ARC3D, OVERFLOW, CFL3D are three successful
NASA contributions), leading to numerous commercial packages.
The CFD applications in the 1990s’ were dominated by NASP and Space Shuttle operations19. The
NASP Program initiated from 1986 through to 1994 was the principal motivator for progresses in
CFD. Nearly all sectors of the US national laboratories, NASA Centers, aerospace industry, and more
than fourteen major universities are actively participated in this program20. The NASP design was
and still is a quantum leap from the traditional approaches to aircraft and space vehicle design. The
demonstration aircraft, X-30, was to operate through the atmosphere from subsonic to orbital

19 Joseph J. S. Shang, “Landmarks and new frontiers of computational fluid dynamics”, Advances in Aerodynamics.
20 Barthelemy RR (1989) The National Aero-Space Plane program. AIAA:1989–5053
 36

velocities at a Mach number exceeding 25. The challenge in developing the high-temperature
material for vehicle fabrication was one of the major issues. The design and analysis of X-30 must
integrate multiple engineering disciplines consisting of aerodynamics, propulsion, structure, and
flight control. The capability for airframe design and flight control/stability was not an overreach,
but the required propulsion systems based on subsonic and supersonic ramjet combustion faced a
formidable challenge.
The CFD application to the Space Shuttle accelerated after the Challenger (STS51-L) accident in 1986.
A higher accuracy requirement was imposed to predict the orbiter wing root shear to within 5% of
the maximum structural capability of the wing. The multiple-zonal grid consists 111 grid blocks with
a 16 M grid point system, and the surface resolution on average was about 10 cm21-22.
Figure 1.9 depicting all the possible perturbations to flow field by fine-scale structures in the grid
generation process on the external
tank. The full-scale, high fidelity grid
generation of the Space Shuttle
launch vehicle with all fine-scale
details was using the ICEM/CFD
CAD and gridding software. The
earlier numerical solutions for the
vehicle were obtained using the
chimera domain decomposition
technique, and then logically
transitioned into the unstructured
grid method to describe the
complex and multi-component
configurations. The simulated Space
Shuttle configuration consists of the
orbiter, external tanks, plus two
solid rocket boosters, and the
complex multiple shock-on-shock
interactions are capturing by the
solving procedure. The CFD
simulation for space shuttle in Figure 1.9 Required fine-scale geometric details space
launching configuration represents shuttle external tank
the state-of-the-art at that period
and is displaying by Figure 1.1023.

21 Buning PG, Parks SJ, Chan WM, Renze KJ (1991) Application of the chimera overlapped grid scheme to
simulation of space shuttle accent flows, proceedings of the 4th international symposium on CFD, Davis
22 Sotnick JP, Kandula M (1994–1860) Buning P (1994) Navier-Stokes simulation of the space shuttle launch

vehicle flight transonic flow field using a large scale chimera grid system. AIAA.
23 Gomez RJ (2011) 20+ years of CFD for space shuttle, NASA Johnson Space Center.
 37

1.5 Advantages of using CFD

CFD has grown from a
mathematical curiosity
to become an essential
tool in almost every
branch of fluid
dynamics. It allows for
a deep analysis of the
fluid mechanics and
local effects in a lot of
equipment. Most of the
CFD results will give an
improved
performance, better
reliability, more
confident scale-up,
improved product
consistency, and
higher plant
productivity. Some
design engineers
actually use CFD to
analyze new systems
before deciding which
and how many
validation tests need to
be performed. The
advantages of CFD can Figure 1.10 CFD simulation of space shuttle launching configuration
be categorized as
(Wanot, 1996):
• It provides a detailed understanding of flow distribution, weight losses, mass and heat
transfer, particulate separation, etc. Consequently, all these will give plant managers a much
better and deeper understanding of what is happening in a particular process or system.
• It makes it possible to evaluate geometric changes with much less time and cost than would
be involved in laboratory testing.
• It can answer many ‘what if’ questions in a short time.
• It is able to reduce scale-up problems because the models are based on fundamental physics
and are scale independent.
• It is particularly useful in simulating conditions where it is not possible to take detailed
measurements such as high temperature or dangerous environment in an oven.
• Since it is a pro-active analysis and design tool, it can highlight the root cause not just the
effect when evaluating plant problems.

1.6 Vision for Further

But of course with advances comes shortcoming like the complaint by [Dimitri Mavriplis]24 which
seems valid. Chief among them are Lack of investment in new fundamental Algorithm Development,

24Dimitri Mavriplis, “Exascale Opportunities for Aerospace Engineering”, Department of Mechanical

Engineering University of Wyoming and the Vision CFD2030 Team.
 38

lack of reliable Turbulence modeling, better collaboration between NASA, government agencies and
private companies, and Poorly positioned to exploit coming Exascale revolution in HPC. It criticizes
while in recent years the design cycle enjoyed considerable advances in isolated development such
as airplane wings, it lacks on the other arenas such as full flight envelope in airplane design. This
might be contributed to either deficiency of robust algorithm and turbulence, or both. It also
contributed this to lack of Holy grail of product development, namely Certification. Borrowing
concepts from John D. Anderson’s book, we try to established a road map to help you keep track of
our building blocks. Figure 1.11 shows such a road map. Another point of view, expressed by
[Charles Hirsch], indicates that there are key issues and major challenges for industrial CFD analysis
and design25, namely:
• To Create Automatic Structured Grid Generation Tools For Families of Topologies, For
Instance for Turbomachinery Passages.
• Efficient full automatic grid generation systems and flow solvers are to be developed further,
particularly for very complex geometries.
• The necessity for improvements in physical modeling, in particular turbulence and
combustion models.
• Fast, full parallel, CFD algorithms are required to reduce design cycle times (provided by
Exascale HPC).
• The development of robust design methodologies taking into account the presence of
uncertainties.
• Next generation of industrial software systems requires high levels of integration of pre and
post-processors, with CFD/CHT/FSI solvers within a global optimization environment, with
highly effective GUI’s to minimize the engineering time associated to simulations and design.

Navier Stokes
and Family
Viscous
Boundary
Layer
Steady State Incompressible Turbulent
Euler
Flow Field
Transient Compressible Laminar Velocity
Potentials

Invscid Laplace

Bernouli

Elementry
Flows

Figure 1.11 Road Map for CFD Studies

25Charles Hirsch, Prof. Em. Vrije Universiteit Brussel and President, NUMECA int. “The Challenges Of Present
And Future Industrial CFD”, AIAA-Scitech 2015.
 39

1.7 CFD in Aerospace Industry

Here we concern with the perspective of (external) aerodynamics, although many of the issues
identified carry over to the other application areas [Spalart and Venkatakrishnan]26. Turbomachinery
is covered elsewhere in this special issue of the journal. It is fair to write that CFD has had a
tremendous although gradual impact on both commercial and military aircraft. The use of CFD in
commercial aircraft is well documented, with particular success in the design of the high-speed wing
(cruise shape) and its close integration with the engine, dating back to the Boeing 737 Classic of the
late 1980s. The extensive use of CFD in the latest aircraft from Boeing is illustrated in the ‘walk-
around’ chart depicted in Figure 1.12; Airbus has presented a similar chart. It is seen that in certain
areas the use of CFD is at a mature state but there are also emerging areas where CFD is expected to
affect aircraft design in significant ways only in the future.

Figure 1.12 Impact of CFD at Boeing. Green areas have strong CFD penetration; Purple areas have some
penetration; Red areas present future opportunities

CFD is increasingly being used in multi-disciplinary design and analysis of aerospace products.
Examples of these include high-speed aerodynamic design taking into account the flexibility of wings
(aero-elastics), icing models, far-field noise propagation models and conjugate heat transfer. An
attractive development is the use of a range of CFD tools to calculate the benefits of formation flight
for large aircrafts. Increasingly, CFD results are compared directly with flight test, rather than wind
tunnel, and the status of the two sources of information in the engineering process and company
culture is slowly shifting, with enlightened organizations drawing on both to good effect. It is
important to transition from wind tunnel to CFD for the right reasons, such as wall effects or Reynolds
number and aero-elastics, whereas doing so only for speed and cost advantages has its dangers. We
believe there is a tendency towards overconfidence in CFD in some circles, even to the extent of
ignoring well-known sources of error, which creates a risk of backlash, were CFD to be blamed for
costly mistakes.

26P. R. Spalart, V. Venkatakrishnan, “On the role and challenges of CFD in the aerospace industry”, Boeing
Commercial Airplanes, Seattle, USA.
 40

CFD still faces several challenges that need to be addressed. The turnaround time associated with
CFD is one of the factors limiting the use of CFD in the design and creation of databases and also in
multi-disciplinary applications. Another limiting factor is the level of skills required of the user of
CFD. CFD practiced in industry is vastly different from the CFD theory taught in universities,
especially in the late 20th century. A long lead time can be required for a user of CFD to become
proficient in all the various phases of CFD (geometry preparation, gridding, solution set-up, post-
processing). Other limitations include various uncertainties in CFD related to numeric, physical
modelling (especially transition and turbulence), and the time involved in preparing geometries for
carrying out grid generation and aerodynamic analyses. The latter two tasks are still highly manual
and in many instances dominate in terms of effort, compared to the solution of the fluid-dynamic
equations.

1.8 CFD vs. Wind Tunnel

In addressing CFD, it is essential to compare CFD with Wind tunnels as a validation tool, where
possible. This has indeed been the case, supported by the relentless development in computational
technology and modelling techniques of the modern era. However, CFD has not yet advanced enough
to replace Wind Tunnels entirely, as was initially forecasted years ago. The first thing to understand
is that despite the CFD vs Wind Tunnel debate, the two tools actually measure very different
aerodynamic performance parameters. This was highlighted by Professor Jochen Wiedemann, who
suggested that the primary strength of Wind Tunnel testing is to establish the integral values such as
drag coefficients and lift coefficients. Whereas CFD is more focused on understanding the local and
internal flow field properties that would be almost impossible to define in a full scale Wind Tunnel,
due to the practical constraints. For example, areas such as brake cooling are best investigated
through the use of CFD. The detailed behavior in this specific area would be extremely difficult to
obtain reliably during a Wind Tunnel test, unless the test went on for months, which no company,
motorsport or automotive, has the time or budget to do. However, in areas which are not yet fully
understood such as the complex downstream effects of tire wake, this cannot be accurately measured
in the wind tunnel or precisely simulated by CFD. In fact the head of Aerodynamics Thermal
Management and Acceleration Performance at Honda R&D suggested that only in 2044 will
computing power be fast enough to fully predict such airflows and match the accuracies of wind
tunnels. It is also highlights how no two wind tunnels are the same, and their different designs results
in a wide array of inaccuracies. Whenever you measure something, whether it be by software
algorithms or with a physical instrument there is always some degree of error. For Wind tunnels that
error can be due to either the effects of blockage, the interference between the freestream air with
the walls or the boundary layer strategy. None of which are present in CFD. With each wind tunnel
differing in design, these inaccuracies vary for each individual wind tunnel, making any direct
comparisons unreliable and inaccurate, or as Professor Wiedemann said ‘comparing apples and
oranges’. Therefore, not only do you need to correlate the CFD results with those obtained from the
Wind Tunnel, but you also need to establish a correlation factor between Wind Tunnels. So, CFD is
accurate in some of the areas that wind tunnels are not, while wind tunnels allow investigation into
areas that cannot be done in CFD and both are unreliable in some areas. This is why wind tunnels
have to be used hand-in-hand with CFD. Other researchers such as [John C. Chien] express similar
results below:
In testing, there are two environments to consider, one is the almost real test environment, and
the other is the simulated test environment. The hardware can be either the real hardware or
the scaled model. So, in wind tunnel testing, the issue is in the simulation of the real environment.
Sometimes it is easy, sometimes it is difficult, and sometimes it is impossible. In the CFD
approach, the specification of the flow condition is in general not a problem. The computer
simply has no feeling about the inlet or free stream conditions, whether it is hot or cold. In a way,
the simulation of the real environment in the CFD approach is not an issue. At this point, the CFD
 41

approach is slightly better than the wind

tunnel testing approach, in simulating
the real environment. Beyond this point,
it is really hard to say which approach
will produce more reliable results. For
simple problems, CFD approach can
generate highly accurate results easily
on computer. But for more complex
geometry, the demand for a lot of mesh
points, and the lack of reliable turbulence
models always create difficulties in
getting a converged solution. These
problems in principle can be resolved in Figure 1.13 Typical Design Project
the future, for now, these are big
problems. The only advantage of using
CFD approach now is, the cost effectiveness. Simpler problems can now be simulated on
workstations or PC workstations. The cost of wind tunnel testing can be improved in the rate of
data handling, that is high volume of high quality data can be obtained in shorter time. Thus the
cost of testing can be reduced. I think, we will be talking about the optimum use of both the wind
tunnel testing and CFD simulation, instead of which one can better simulate the free stream
conditions. What I am saying is " We have long past the era of Wind Tunnel vs. CFD ". Since there
is no short term solutions to the shortage of wind tunnel testing and CFD simulation, we are
forced to combined these two approaches in the design loops.
Figure 1.13 illustrates today’s characteristics of CFD and the wind tunnel27, for a typical design
process, in terms of relative cost and flow time. The optimal usage of the two is concentrated at
opposite corners of the spectrum. Keep in mind, One complete airplane development requires about
2.5 million aerodynamic simulations.

27Edward N. Tinoco, “The Changing Role of Computational Fluid Dynamics in Aircraft Development”, AIAA-98-
2512.
 42

2 Linear PDE’s and Exact Solution of Some Model Equations

2.1 Mathematical Character of Basic Equations
The general theory concerning the character of PDE has been developed based on following relation
where A, B, C, D, E, F are assumed to be function of (x, y) only for the time being; making Eq. 2.1
linear in nature

∂2 φ ∂2 φ ∂2 φ ∂φ ∂φ
A 2 +B +C 2 +D +E + F(φ) = G(x , y)
∂x ∂x ∂y ∂y ∂x ∂y
Eq. 2.1
It is found that character of Eq. 2.1 depends upon the sign of determinate function B2 - 4AC as the
flow dependencies for each case shown by solid line. In summary,
 2  2  2   2  2
+ =0 − =0 − =0
x 2 y 2 x 2 y x 2 y 2

B2- 4AC < 0 B2- 4AC = 0 B2- 4AC > 0

Elliptic Parabolic Hyperbolic

Boundary Value mixed initial and
initial value problem
problem Boundary value problem
complete contour boundary conditions must be specifying boundary
closed at one end but remain conditions at one end but
boundary specification open at other remain open at others

Laplace equation Heat conduction equation Wave equation

No real charateristics One real charateristics Two real charateristics

In reality, the viscous flow equations are simply too complicated to fit into a single mode. They can
be elliptic, parabolic, and hyperbolic or mixture of all three, depending to specific flow, geometry
and time dependencies. Some examples of these model equations will be dealt extensively later. For
example, the unsteady compressible N-S equations are a mixed set of hyperbolic-parabolic equations,
while, for 2D unsteady incompressible N-S for x-momentum is mixed set of elliptic-parabolic-
hyperbolic equations as depicted below:

Parabolic
⏞
∂u ∂u ∂u 1 ∂P μ ∂2 u ∂2 u
+u +v = + ( + )
∂t
⏟ ∂x ∂y ρ ∂x ρ ⏟∂x 2 ∂y 2
Hyperbolic Elliptic
Eq. 2.2
Consequently, different numerical techniques must be used in to solve the N-S equations in
compressible and incompressible flow regions. Similarly the Euler equations governing an in-viscid,
non-heat conducting gas have a different character in different flow regions. If the time dependent
 43

terms are retained, the resulting unsteady equations are hyperbolic and solutions can be obtained by
marching procedures. The situation is different when a steady flow is assumed. In that case, the Euler
equations are elliptic when flow is subsonic and hyperbolic when it is supersonic. It could be said
that Euler’s equation is hyperbolic in temporal domain and elliptic in special domain. Therefore,
different flow regions means different characteristics and demands different solving procedure. A
major difference between subsonic and supersonic flows is that flow disturbances propagate
everywhere throughout a subsonic flow; whereas they cannot propagate upstream in supersonic
flow.

2.2 Behavior of the Different Classes of PDE

In this section we simply discuss, without proof, some of the behavior of hyperbolic, parabolic and
elliptic partial differential equations, and relate this behavior to the solution of problems in fluid
dynamics28. For more details on the characteristics of partial differential equations, see any good text
on advanced mathematics, such as [Hildebrand]29.
2.2.1 Hyperbolic Equations
For hyperbolic equations, information at a given point P influences only those regions between the
advancing characteristics. For example, examine Figure 2.2-(B), which is sketched for a two-
dimensional problem with two independent space variables. Point P is located at a given (x, y).
Consider the left and right-running characteristics through point P, as shown. Information at point P
influences only the shaded region the region labelled I between the two advancing characteristics
through P as demonstrated in Figure 2.2-(B). This has a direct effect on boundary conditions for
hyperbolic equations. Assume that the x-axis is a given boundary condition for the problem, i.e. the
dependent variables u and v are known along the x-axis. Then the solution can be obtained by
‘marching forward’ in the distance y, starting from the given boundary. However, the solution for u
and v at point P will depend only on that part of the boundary between a and b, as shown. Information
at point c, which is outside the interval ab, is propagated along characteristics through c, and
influences only region II in Figure 2.2-(B). Point P is outside region II, and hence does not feel the
information from point c. For this reason, point P depends on only that part of the boundary which is
intercepted by and included between the two retreating characteristic lines through point P, i.e.
interval ab. As a general rule, in fluid dynamics, the following types of flows are governed by
hyperbolic partial differential equations, and hence exhibit the behavior described above:
2.2.1.1 Steady Inviscid Supersonic Flow
If the flow is 3D, there are characteristic surfaces in xyz space, as sketched in Figure 2.1-(A).
Consider point P at a given (x, y, z) location. Information at P influences the shaded volume within the
advancing characteristic surface. In addition, if the x−y plane is a boundary surface, then only that
portion of the boundary shown as the cross-hatched area in the x−y plane, intercepted by the
retreating characteristic surface, has any effect on P. In Figure 2.1-(A) , the dependent variables are
solved by starting with data given in the xy-plane, and ‘marching’ in the z-direction. For an inviscid
supersonic flow problem, the general flow direction would also be in the z-direction.

28 John D. Anderson Jr., Joris Degroote, G´erard Degrez, Erik Dick, Roger Grundmann and Jan Vierendeels,
“Computational Fluid Dynamics - An Introduction”, 3rd Edition, ISBN: 978-3-540-85055-7, 2009.
29 Hildebrand, F.B., “Advanced Calculus for Applications”, Prentice-Hall, New Jersey, 1976.
 44

(A) 3D Steady Flow (B) 2D Steady Flow

Figure 2.1 Domain and Boundaries for the Solution of Hyperbolic Equations (Steady)

2.2.1.2 Unsteady Inviscid Compressible Flow

For unsteady one and two-dimensional inviscid flows, the governing equations are hyperbolic, no
matter whether the flow is locally subsonic or supersonic. Here, time is the marching direction. For
one dimensional unsteady flow, consider a point P in the (x, t) plane shown in Figure 2.2-(B). Once
again, the region influenced by P is the shaded area between the two advancing characteristics
through P, and the interval ab is the only portion of the boundary along the x-axis upon which the
solution at P depends. For two-dimensional unsteady flow, consider a point P in the (x, y, t) space as
shown Figure 2.2-(B). The region influenced by P, and the portion of the boundary in the xy-plane
upon which the solution at P depends, are shown in this figure. Starting with known initial data in
the xy-plane, the solution ‘marches’ forward in time. Standard spacial 3D hyperbolic equations
include the wave equation is (see Eq. 2.3 and Eq. 2.4),

(A) 2D Unsteady Flow (B) 1D Unsteedy Flow

Figure 2.2 Domain and Boundaries for the Solution of Hyperbolic Equations (Unsteady)
 45

∂2 u
2
= ∇2 u
∂t
Eq. 2.3
and the advection (convection or transport) equation:

∂u ∂u
=
∂t ∂x
Eq. 2.4
These equations are time-dependent; they model the transient movement of signals. The wave
equation models acoustic and electromagnetic fields, the advection equation models the translation
of waves such as water waves.
2.2.2 Parabolic Equations
For parabolic equations, information at point P in the xy-plane influences the entire region of the
plane to one side of P. This is sketched in Figure 2.4 where the single characteristic line through
point P is drawn. Assume the x- and y-axes are
boundaries; the solution at P depends on the
boundary conditions along the entire y axis, as
well as on that portion of the x-axis from a to
b. Solutions to parabolic equations are also
‘marching’ solutions; starting with boundary
conditions along both the x- and y-axes, the
flow-field solution is obtained by ‘marching’ in
the general x-direction. In fluid dynamics,
there are reduced forms of the Navier–Stokes
equations which exhibit parabolic-type
behavior. If the viscous stress terms involving
derivatives with respect to x are ignored in Figure 2.4 Domain and Boundaries for the
these equations, we obtain the ‘parabolized’ Solution of Parabolic Equations in Two Dimensions
Navier Stokes equations, which allows a
solution to march downstream in the x-direction, starting with some prescribed data along the x- and
y-axes. A further reduction of the Navier–Stokes equations for the case of high Reynolds number
leads to the well-known boundary layer equations.
These boundary layer equations exhibit the
parabolic behavior shown in Figure 2.4. The
prototypical parabolic partial differential equation is
the diffusion equation. As the term suggests, such
equations diffusion phenomena in physics, such as
the spreading of heat in a conducting material (see
Eq. 2.5).
∂u
= ∇2 u
∂t
Eq. 2.5
2.2.3 Elliptic Equations
For elliptic equations, information at point P in the
xy-plane influences all other regions of the domain.
Figure 2.3 Domain and Boundaries of
This is sketched in Figure 2.3, which shows a
Elliptic Equations in Two Dimensions
rectangular domain. Here, the domain is fully closed,
 46

surrounded by the closed boundary. This is in contrast to the open domains for parabolic and
hyperbolic equations discussed earlier. For elliptic equations, because point P influences all points
in the domain, then in turn the solution at point P is influenced by the entire closed boundary.
Therefore, the solution at point P must be carried out simultaneously with the solution at all other
points in the domain. This is in stark contrast to the ‘marching’ solutions germane to parabolic and
hyperbolic equations. For this reason, problems involving elliptic equations are frequently called
‘equilibrium’, or ‘jury’ problems, because the solution within the domain depends on the total
boundary around the domain30. In fluid dynamics steady, subsonic, inviscid flow is governed by
elliptic equations. As a sub-case, this also includes incompressible flow (which theoretically implies
that the Mach number is zero). Hence, for such flows, physical boundary conditions must be applied
over a closed boundary that totally surrounds the flow, and the flow-field solution at all points in the
flow must be obtained simultaneously because the solution at one point influences the solution at all
other points. Elliptic equations (such as Laplace or Poisson’s equations) are usually used to model
steady state phenomena. When the solution to either hyperbolic or parabolic equations are assumed
to be invariant with time then they reduce to elliptic equations. (See
Eq. 2.6).
∇2 u = 0 or ∇2 u = f
Eq. 2.6
2.2.4 Some Observation
At this stage it is instructive to return to our discussion of the inviscid flow over a supersonic blunt
body in particular to Figure 1.5. There we pointed out that the locally subsonic steady flow is
governed by elliptic partial differential equations, and that the locally supersonic steady flow is
governed by hyperbolic partial differential equations. Now we have a better understanding of what
this means mathematically; and because of the totally different mathematical behavior of elliptic and
hyperbolic equations, we have a new appreciation for the difficulties that were encountered by early
researchers in trying to solve the blunt body problem. The sudden change in the nature of the
governing equations across the sonic line virtually excluded any practical solution of the steady flow
blunt body problem involving a uniform treatment of both the subsonic and supersonic regions.
However, recall that unsteady inviscid flow is governed by hyperbolic equations no matter whether
the flow is locally subsonic or supersonic. This provides the following opportunity. Starting with
rather arbitrary initial conditions for the flow field in the xy-plane in Figure 2.2-(B), solve the
unsteady, two-dimensional inviscid flow equations, marching forward in time as sketched in Figure
2.2-(A). At large times, the solution approaches a steady state, where the time derivatives of the
flow variables approach zero. This steady state is the desired result, and what you have when you
approach this steady state is a solution for the entire flow field including both the subsonic and
supersonic regions. Moreover, this solution is obtained with the same, uniform method throughout
the entire flow. The above discussion gives the elementary philosophy of the time-dependent
technique for the solution of flow problems. Its practical numerical implementation by [Moretti and
Abbett]31 in 1966 constituted the major scientific breakthrough for the solution of the supersonic
blunt body problem. We will examine the actual, closed-form solution to some linear partial
differential equations of the elliptic, parabolic and hyperbolic types.
2.2.5 The 'Par-Elliptic' Problem
An important practical use of the parabolic solution procedure is to refine an elliptic flow solution
of the region outside the boundary layer. In this case it is from the elliptic flow solution that the

30 Anderson, D.A., Tannehill, John C. and Pletcher, Richard

H., Computational Fluid Mechanics and Heat Transfer,
McGraw-Hill, New York, 1984.
31 Moretti, G. and Abbett, M., ‘A Time-Dependent Computational Method for Blunt Body Flows,’ AIAA Journal, Vol.

4, No. 12, December 1966, pp. 2136–2141.

 47

pressure must be extracted: pressures for the

nearest-to-surface cells of the elliptic grid are
transferred to all the cells in the parabolic
grid, at the same z location. In general,
because the cells of the parabolic and elliptic
grids are likely to be of different sizes, as
shown below, interpolation is needed. The
above sketch contains a reminder that a two-
way interchange of information may take
place between the elliptic and parabolic
calculations (see Figure 2.5). Thus the elliptic
calculation may take place at first, with the
assumption that friction at the solid surfaces
is absent. Its predicted pressure distribution
is for that reason not quite correct. However Figure 2.5 Two-way interchange of information
the ensuing parabolic calculation takes between Parabolic and Elliptic flows
detailed account of friction and can report the
so-called 'displacement thickness' of the boundary layer. If this is transmitted back to the elliptic
solver, that can repeat its calculation on the assumption that the effective size of the solid object is
larger than it first supposed. Its second flow prediction will be correspondingly more accurate32.

2.3 Nonsingular Transformation

Suppose a we would like to transform (x, y) ⇾ (ξ, η), which means x and y are transformed into new
independent variables ξ and η. We also require that this transformation be nonsingular which
provides that a one to one relationship exists between (x, y) and (ξ, η)33. We are also assumed of a
nonsingular mapping provided that the Jacobian of Transformation is nonzero.

∂ξ ∂ξ
∂(ξ, η) ∂x ∂y
J= = || |=ξ η −ξ η ≠0
∂(x,y) ∂η ∂η| x y y x

∂x ∂y
Eq. 2.7
Therefore, any real nonsingular transformation does not change the type of PDE34.

2.4 Exact (Closed Form) Solution to Linear Model Equations

Since the exact solutions of Naver-Stokes , Boundary Layer, or Euler methods is not available yet, we
resort to model equation with reduced order to find a closed form solution. For example the viscous
Burger equation can be modelled as a reduced NS equation. The discussion here has centered on the
2nd order equation given by Wave, Heat, and Laplace Equations. In addition, system of 1st order
equations were examined. A number of other very important equations should be mentioned since
they govern common physical phenomena or they are used as simple models for more complex
problems. In many cases, exact analytical solutions for these equations exist.

32 Parabolic Flows by “PHOENICS”.

33 Anderson, Dale A; Tannehill, John C; Plecher Richard H; 1984:”Computational Fluid Mechanics and Heat
Transfer”, Hemisphere Publishing Corporation.
34 Taylor, A., E., “Advanced Calculus”, Ginn and Company, Boston, 1955.
 48

2.4.1 Linear Wave Equation (1st Order)

The first and widely used is the 1st order linear u u
+a =0
wave equation which governs the propagation of a t x
wave moving to the right at a constant speed a.
This is also called the advection equation (see
Figure 2.6). A classic example of hyperbolic
equation also requires an initial condition, u(x, 0)
= u0(x). The question of what boundary conditions
are appropriate for this equation can be more
easily be answered after considering its solution.
It is easy to verify that the solution is given by u(x,
t) = u0(x − at). This describes the propagation of
the quantity u(x, t) moving with speed “a” in the x-
direction. The solution is constant along the
characteristic line x − at = c with u(x, t) = u0(c). Figure 2.6 Solution of linear Wave equation
From the knowledge of the solution, we can
appreciate that for a > 0 a boundary condition
should be prescribed at x = 0, (e.g., u (0) = α0) where information is being fed into the solution domain.
The value of the solution at x = 1 is determined by the initial conditions or the boundary condition at
x = 0 and cannot, therefore, be prescribed. This simple argument shows that, in a hyperbolic problem,
the selection of appropriate conditions at a boundary point depends on the solution at that point. If
the velocity is negative, the previous treatment of the boundary conditions is reversed.
2.4.2 Inviscid Burgers Equation
This is also called non-linear 1st order
wave equation and governs the u  2u
propagation of nonlinear wave for 1-D −a 2 =0
t x
case. An analogous analysis to that used
for the advection equation shows that
u(x, t) is constant if we are moving with
a local velocity also given by u(x, t). This
means that some regions of the solution
advance faster than other regions
leading to the formation of sharp
gradients. This is illustrated in Figure
2.8. The initial velocity is represented
by a triangular “zig-zag” wave. Peaks
and troughs in the solution will advance,
in opposite directions, with maximum
speed. This will eventually lead to an
overlap as depicted by the dotted line.
This also results in a non-uniqueness of Figure 2.7 Rate of Decay of solution to diffusion equation
the solution which is obviously non-
physical and to resolve this problem we must allow for the formation and propagation of
discontinuities when two characteristics intersect.
 49

2.4.3 Diffusion (Heat) Equation

Which is parabolic in nature and in addition to appropriate boundary conditions of the form used for
elliptic equations, we also require an initial
condition at t = 0 of the form u(x, 0) = u0(x) u u
where u0 is a given function. If b is constant, this +u =0
t x
equation admits solutions of the form u(x, t) =
Aeβt sin kx if β + k2a = 0. A notable feature of the u ( x, t ) =  Am e  mt sin k m x
m
solution is that it decays when b is positive as
the exponent β < 0. The rate of decay is a
function of a. The more diffusive the equation
(i.e., larger a) the faster the decay of the solution
is. In general the solution can be made up of
many sine waves of different frequencies, i.e., a
Fourier expansion of the form

u ( x, t ) =  Ame  mt sin km x Eq. 2.8

m

Where Am and km represent the amplitude and

the frequency of a Fourier mode, respectively.
The decay of the solution depends on the Figure 2.8 Formulation of discontinuities in non-
Fourier contents of the initial data since βm = linear Burgers (wave) equation
−k2m b. High frequencies decay at a faster rate
than the low frequencies which physically means that the solution is being smoothed. This is
illustrated in Figure 2.7 which shows the time evolution of u(x, t) for an initial condition u0(x) = 20x
for 0 ≤ x ≤ 1/2 and u0(x) = 20(1 − x) for 1/2 ≤ x ≤ 1. The solution shows a rapid smoothing of the slope
discontinuity of the initial condition at x = 1/2. The presence of a positive diffusion (a > 0) physically
results in a smoothing of the solution which stabilizes it. On the other hand, negative diffusion (a < 0)
is de-stabilizing but most physical problems have positive diffusion.
2.4.4 Viscous Burgers Equation
This is the nonlinear equation with diffusion added. This particular form is very similar to the
equations of governing fluid flow and can be used as a simple nonlinear model for numeric.

u u  2u
+u =υ 2
t x x
Eq. 2.9
1 + (2a0 − 1)exp((1− a 0 )/ )
with solution u = where ξ = x − a 0 t − x 0
1 + exp((1 − a 0 )/ )
2.4.5 Tricomi Equation
This equation governs problems of the mixed type such as inviscid transonic flows. The properties of
Tricomi equations include a change of form from elliptic to hyperbolic character depending upon sign
of y.
 2u  2u
y 2 + 2 =0 Eq. 2.10
x y
2.4.6 2D Laplace Equation
The linear Elliptic 2D Laplace equation (see Figure 2.9) has following solutions
 50

 2  2
+ =0 have a solution :
 2 x y 2
2y Eq. 2.11
 = xy ,  = x 2 − y 2 ,  = ,  = e kx sin(kx) k = constant ,   [0,1]
(1 + x) 2 + y 2
sinh(x) sin(y) + sinh(y) sin(x)
=
sinh( )
2.4.6.1 Boundary Conditions
The Dirichlet problem for Laplace's equation consists of finding a solution φ on some domain D such
that φ on the boundary of D is equal to some given function. Since the Laplace operator appears in
the heat equation, one physical interpretation of this problem is as follows: fix the temperature on
the boundary of the domain according to the given specification of the boundary condition. Allow
heat to flow until a stationary state is reached in
which the temperature at each point on the domain  2  2
doesn't change anymore. The temperature + 2 =0
distribution in the interior will then be given by the  2
x y
solution to the corresponding Dirichlet problem . 35

The Neumann boundary conditions for Laplace's

equation specify not the function φ itself on the
boundary of D, but its normal derivative. Physically,
this corresponds to the construction of a potential
for a vector field whose effect is known at the
boundary of D alone. Solutions of Laplace's equation
are called harmonic functions; they are all analytic
within the domain where the equation is satisfied.
If any two functions are solutions to Laplace's
equation (or any linear homogeneous differential
equation), their sum (or any linear combination) Figure 2.9 Solution to Laplace equation
is also a solution. This property, called the
principle of superposition, is very useful, e.g., solutions to complex problems can be constructed
by summing simple solutions36-37.

2.4.7 Poisson’s Equation

This elliptic equation governs the temperature distribution in the solid with heat source described
by f(x, y). Poisson’s equation (see Figure 2.10) also determines the electric field in a region
containing a charge density f(x, y). The f(x, y) in this case is sin(πx)Sin(πy) and the B.C. is u (x, y)= 0
subject to 0 ≤ x , 1 ≤ y, ux(0,y) =-sin(yπ)/2π. The solution is obtained in bottom of Figure 2.10.
Viscous Burgers Equation
This is the nonlinear equation with diffusion added. This particular form is very similar to the
equations of governing fluid flow and can be used as a simple nonlinear model for numeric.

35 From Wikipedia, the free encyclopedia.

36 Hazewinkel, Michiel, ed. (2001), "Laplace equation", Encyclopedia of Mathematics, Springer, ISBN 978-1-
55608-010-4.
37 Example initial-boundary value problems using Laplace's equation from exampleproblems.com.
 51

u u  2u
+u =υ 2
t x x
Eq. 2.12
1 + (2a0 − 1)exp((1− a 0 )/ )
with solution u = where ξ = x − a 0 t − x 0
1 + exp((1 − a 0 )/ )
2.4.8 The Advection-Diffusion Equation
This particular expression represents the advection of a quantity ξ in a region with velocity u. The
quantity υ is a diffusion or viscosity coefficient and a is a constant > 0.

   2
+a = 2
t x x
 (x, t) = exp( −kxt) sin(kx − at) where k = constant and  (x) = sin(kx)   [0,1]

Eq. 2.13
 2u  2u
2.4.9 The Korteweg-De Vries Equation + = f(x, y) = sin(π x) sin(π y)
In 1895, the Korteweg-De Vries (KDV) equation was x 2 y 2
created as a means to model water waves. Since the sin( x ) sin( y )
equation doesn’t introduce dissipation, the waves u ( x, y ) =
− 2 2
travel seemingly forever. These waves are now
called solitons, which are seen as single “humps”
that can travel over long distances without altering
their shape or speed38. The motion of nonlinear
dispersive wave is governed by this example.

u u  3 u
+u + =0 Eq. 2.14
t x x 3

Today, engineers use the KDV equation to

understand light waves. As a result, one of the main
modern applications of solitons is in optical fibers.
Figure 2.10 Solution to Poisson's
2.4.10 Helmholtz Equation Equation
This equation governs the motion of time dependent harmonic waves where k is a frequency
parameter. Application includes the propagation of acoustics waves.

 2 u + k 2 u = ( 2 + k 2 )u = 0 Eq. 2.15
2.4.11 Summary List of Exact Solution Methods
The solution is obtained from the list provided below. This list by no means exclusive and many more
exists in literature. They are commonly considered Boundary Value Problems (BVP).
1. Method of Characteristics
2. Shock Capturing Methods
3. Similarity Solutions
4. SCM (Split Coefficient Method)
5. Methods for solving Potential Equation

38 Caty Fairclough, “COMSOL Blog”, December 20, 2017.

 52

6. Methods for solving Laplace equation

7. Separation of Variable
8. Complex Variables
9. Superposition of Non-Linear Equation
10. Transformation of Variables
11. Manufacturing Solutions
 53

3 Solving System of Linear Algebraic Equation

In the study of numerical methods, the distinction can be made between a set of methods such as
solving linear systems of equations, solving matrix eigenvalue problems, interpolation,
numerical integration and finding the roots or zeros of equations, which can be somewhat
considered as the building blocks for larger that arise in engineering/applied mathematics/physics.
As an example, the problem of solving ordinary differential equations, optimization and solving
integral equations. But from the point of view of applied mathematics or engineering, perhaps the
most significant problems in numerical methods is the solution of partial differential equations by
Finite Difference Methods, Finite Element Methods or Boundary Element Methods. The study of the
behavior of numerical methods is called numerical analysis. This is a mathematical subject that
considers the modelling of the error in the processing of numerical methods and the subsequent re-
design of methods39.

3.1 Interpolation (Morphing)

Interpolation or morphing is a method of finding new values for any function using the given set of
values. The unknown value at a particular point can be found using many interpolation formula's. If
the new value has to be found from the two given points then the linear interpolation formula is
used whereas if 'n' set of numbers are available the new value is found using the Lagrange's
interpolation formula. The Linear Interpolation Formula is given as,

y − y1 y2 − y1
=
x − x1 x2 − x1
Eq. 3.1
Among non-linear interpolation schemes, Lagrange interpolating polynomial is the prominent
one which can be expressed as the polynomial P(x) of degree ≤ (n - 1) that passes through the n points
(x1, y1 = f(x1), x2,y2 = f(x2), , , , , , , , xn,yn= f(xn)) , and is given by

n n
x − xk
P(x) = ∑ Pj (x) → Pj (x) = yj ∏
xj − xk
j=1 k=1
k≠j
Eq. 3.2

Figure 3.1 Bi-linear (middle) and Bi-Cubic (far right) Interpolation

39 Numerical Methods.com
 54

Cubic splines are other non-linear example such as Non-Rational B-Splines (NURBS) or Bezier
functions. Interpolation schemes are vast and different. For example, Multivariate interpolation is
the interpolation of functions of more than one variable40. Methods include bilinear interpolation and
bi-cubic interpolation in two dimensions, and trilinear interpolation in three dimensions. They can
be applied to gridded or scattered data as depicted in Figure 3.1.
An example of linear morphing can be mathematically expressed as Eq. 3.3 and each step is
characterized by one value of a single parameter (α), called interpolation or morphing factor, as
shown at the bottom of Figure 3.2 where Sˆ2 (Pres. Bush) and Sˆ1 (Pres. Obama).

M(α, t) = α(t)Ŝ1 + [1 − α(t)]Ŝ2

Eq. 3.3
The morphing factor should vary between 0 and 1, such that α = 1 and α = 0 produce source and
target respectively. Convex combinations of more than two objects (images, sounds) are also
possible, as well as using a time varying morphing factor, giving rise to dynamic transformations.

Figure 3.2 Depiction of Image Morphing Using a single Parameter α

3.2 Numerical Integration

Numerical integration is the numerical approximation of the integral of a function. For a function of
one variable, it amounts to finding the area under the graph of the function. That is finding the
integral I where the numerical method for integration generally replace the integral by a weighted
sum of n weights and n function evaluations, so that

b n

I = ∫ f(x)dx ≈ ∑ wi f(xi )
a i=1
Eq. 3.4
For a function of two variables it is equivalent to finding an approximation to the volume under the
surface. Numerical integration is often also referred to as quadrature or sometimes cubature for
functions of two or more variables. Returning to the one variable case, numerical integration involves
finding the approximation to an integral of a function f(x) through its evaluation at a set of discrete
points. There are two distinct approaches to this. Firstly methods like the trapezium rule or
Simpson's rule determine the integral through evaluating f(x) at regularly spaced points. These are
generally referred to as Newton-Cotes formulae. Alternative methods termed Gaussian Quadrature

40 Wikipedia,
 55

methods have arisen that select irregularly-placed evaluation points, chosen to determine the
integral as accurately as possible with a given set of points.
Gaussian Quadrature methods are important as they often lead to very efficient methods. In
numerical integration the efficiency of the method relates to the accuracy obtained with respect to
the number of evaluations of the function f(x). In intensive methods such as the boundary element
method integrations may need to be performed millions of times so the efficiency of the methods
needs to be considered sometimes. In general, care must be taken to match the numerical integration
method to the expected nature of the function f(x). Typically, it may be known that f(x) is regular. On
the other hand f(x) may be singular or oscillatory and will then need special treatment. Often a special
method called a product integration method can be developed for the integration of functions of the
form f(x) = w(x)g(x) where w(x) is a pre-set function and the function and g(x) is known to be a
relatively nice function.

3.3 Algebraic Eigenvalues & Eigenvectors Problems

Before we attend the next section about the solution of model PDE’s, we better familiar ourselves
with the concept of Eigenvalues and Eigenvectors because they play an essential rule in solving the
PDE’s. In linear algebra, an eigenvector or characteristic vector of a linear transformation is a non-
zero vector that does not change its direction when that linear transformation is applied to it. In other
words, if v is a vector that is not the zero vector, then it is an eigenvector of a linear transformation T
if T(v) is a scalar multiple of v. This condition can be written as the equation:

T( v ) = λv Eq. 3.5

Where λ is a scalar known as the eigenvalue or characteristic value associated with the eigenvector
v. If the linear transformation [T] is expressed as a square matrix [A] then the equation can be
expressed as the matrix multiplication:

Ax = λx Eq. 3.6

Where x is a column vector. There is a correspondence between n by n square matrices and linear
transformations from an n-dimensional vector space to itself. For this reason, it is equivalent to define
eigenvalues and eigenvectors using either the language of matrices or the language of linear
transformations. Geometrically, an eigenvector corresponding to a real, nonzero eigenvalue points
in a direction that is stretched by the transformation and the eigenvalue is the factor by which it is
stretched. If the eigenvalue is negative, the direction is reversed. The hyperbolic equations have a
both positive distingue eigenvectors, parabolic has one, while the elliptic has complex one.

a11 − λ a12 .... a1n   x1   0 

 a .... a 2n  x   0 
 21 a 22 − λ  2 =   or ( A − λI) x = 0
 .... .... .... ....  ....  ....
     
 a n1 a n2 .... a nn − λ  x n   0 
Eq. 3.7
Where I is the identity matrix. As shown in Cramer's rule, a linear system of equations has nontrivial
solutions if the determinant vanishes, so the solutions of Eq. 3.7 are given by
 56

( A − λI) = 0 or (1 -  )(2 -  ) ...... (n -  ) = 0 Eq. 3.8

Each λi may be real but in general is a complex number. The numbers λ1, λ2…. λn, which may not all
have distinct values, are roots of the polynomial and are the eigenvalues of A. This equation is known
as the characteristic equation of A, and the left-hand side is known as the characteristic polynomial.
3.3.1 Eigenvalues Calculation
The eigenvalues of a matrix [A] can be determined by finding the roots of the characteristic
polynomial. Explicit algebraic formulas for the roots of a polynomial exist only if the degree n is 4 or
less. According to the Abel–Ruffini theorem there is no general, explicit and exact algebraic formula
for the roots of a polynomial with degree 5 or more. It turns out that any polynomial with degree n is
the characteristic polynomial of some companion matrix of order n. Therefore, for matrices of order
5 or more, the eigenvalues and eigenvectors cannot be obtained by an explicit algebraic formula, and
must therefore be computed by approximate numerical methods. In theory, the coefficients of the
characteristic polynomial can be computed exactly, since they are sums of products of matrix
elements; and there are algorithms that can find all the roots of a polynomial of arbitrary degree to
any required accuracy. However, this approach is not viable in practice because the coefficients
would be contaminated by unavoidable round-off errors, and the roots of a polynomial can be an
extremely sensitive function of the coefficients41.
3.3.1.1 Eigenvector Calculation
The eigenvalues must be determined first. Once these are known, the corresponding eigenvectors
can be calculated directly from the linear system. It should be noted that if vector v is an eigenvector.

(𝐀 − λ𝐈)𝐯⃗ = 0
Eq. 3.9
3.3.1.2 Right and Left Eigenvectors
Given an eigenvalue λ, The eigenvector r that satisfies [A]r = λr is sometimes called a (right)
eigenvector for the matrix [A] corresponding to the eigenvalue λ. If λ1, λ2, ..., λr are the eigenvalues and
r1, r2, ..., rr are the corresponding right eigenvectors, then is easy to see that the set of right
eigenvectors form a basis of a vector space. If this vector space is of dimension n, then we can
construct an n × n matrix [R] whose columns are the components of the right eigenvectors, which has
the property that [A][R] = [R][Λ] where [Λ] is the diagonal matrix whose diagonal elements are the
eigenvalues as shown in Eq. 3.10. By appropriate numbering of the eigenvalues and eigenvectors,

1 0 0 .... 0 
0  0 .... 0 
 2

[ Λ ] =  0 0 3 .... 0 
 
.... .... .... .... .... 
 0 0 0 .... n 
Eq. 3.10
it is possible to arrange the columns of the matrix [R] so that λ1 ≥ λ2 ≥ … ≥ λn. In the same spirite, If
this vector space is of dimension n, then we can construct an n × n matrix [L] whose rows are the
components of the left eigenvectors, which has the property that [L][A] = [Λ][L]. This is easily done
if we define [L ] = [R]−1 and define the components of the left eigenvectors to be the elements of the

41 From Wikipedia.
 57

respective rows of [L]. Beginning with [A][R] = [R][Λ] and multiplying both sides on the left by [R]−1,
we obtain [R]−1[A][R] = [Λ] and multiplying on the right by [R]−1, we have [R]−1[A] = [Λ][R]−1 which
implies that any row of [R]−1 satisfies the properties of a left eigenvector42 or [L][A][R]=[Λ].
3.3.1.3 Diagonalization of a Matrix
Given an n by n matrix [A], we say that [A] is diagonalizable if there is a matrix [X] so that [X]−1[A][X]
= [Λ]. It is clear from the above discussions that if all the eigenvalues are real and district, then we
can use them as matrix of right eigenvectors [R] instead of [X].
3.3.1.4 Case Study 1 - Eigenvalues and Eigenvectors of 3D Euler Equation
The Eigen system/eigenvalues and eigenvectors of the Euler equations of inviscid flow form the
basis of total variation diminishing (TVD) algorithms in computational fluid dynamics (CFD)43.
Whether the conservation equations are solved in a finite difference or finite volume format, the
matrices of right and left eigenvectors that can be found in the literature are generally decomposed
along the directions of a global (x, y, z) or local (ξ, η, ζ) coordinate system. Such matrix decomposition,
however, is not necessary. The eigenvalues and eigenvector matrix of 3D inviscid flow can be
expressed along any given direction, e.g. through a unit vector (nx, ny, nz) normal to a surface. The
resulting expression is relatively simple and allows for more efficient code implementation in finite
volume solvers.
3.3.1.5 Governing Equations
The 3D unsteady Euler equations of inviscid flow, a system of integral conservation equations for
mass, momentum, and energy, can be written in vector notation as the sum of a volume and surface
integral,
 ρ   ρv n 
 ρu   ρuv + pn 
    n x

 QdV + CS FdA = 0

t CV
where Q =  ρv 
 
, F = ρvv n + pn y 

 
 ρw  ρwv n + pn z 
ρe 0   ρh 0 v n 

v n = v.n̂ = un x + vn y + wn z , n 2x + n 2y + n 2z = 1

e0 = e + ek , h 0 = h + ek , ek =
2
(
1 2
u + v2 + w 2 )
Eq. 3.11
with ek being the kinetic energy per unit mass. Static energy, enthalpy, and pressure can all be
expressed in terms of the local speed of sound a , a function of temperature, and the ratio of specific
heats γ,

a2 a2 ρa 2 cp
e= , h= , p= , a 2 = γRT , λ = Eq. 3.12
γ(γ − 1) (γ − 1) γ cv
3.3.1.6 Transformation Matrix
The first step in determining the Eigen system of the above conservation equations is to derive the
corresponding Jacobian or transformation matrix, which can be found by taking partial derivatives

42 Kenneth I. Joy, “Eigenvalues and Eigenvectors”, Visualization and Graphics Research Group Department of
Computer Science University of California, Davis.
43 Axel Rohde, “Eigenvalues and Eigenvectors Of The Euler Equations In General Geometries”, AIAA 2001-2609.
 58

of the flux

 0 nx ny nz 0 
 (γ − 1)e n − uv v n (γ − 2)un x un y − (γ − 1)vn x un z − (γ − 1)wn x (γ − 1)nx 
F 
k x n

A= =  (γ − 1)ek n y − vv n vn x (γ − 1)un y v n − (γ − 2)vn y vn z − (γ − 1)wn y (γ − 1)n y 

Q  
(γ − 1)ek n z − wv n wn x (γ − 1)un z wv y − (γ − 1)vn y v n − (γ − 2)wn z (γ − 1)nz 
[(γ − 1)ek − h 0 ]v n h 0 n x (γ − 1)uvn h 0 n y − (γ − 1)vn y h 0 n z − (γ − 1)wn n γ v n 

Eq. 3.13
We can now rewrite the Euler equations in the format of a general wave equation,


 Q dV + CS F(Q) dA = 0
t CV
where F(Q) = [ A] Q Eq. 3.14

The transformation matrix [A] can be interpreted as a wave speed with local and directional
dependence for a nonlinear multi-dimensional wave. The multidimensional character is really
twofold: (1) we are working in a 3D flow field, where waves can travel in any direction; (2) there are
different types of waves, all traveling at their own characteristic speeds, which are determined by the
eigenvalues of the matrix [A]. The eigenvalues of the transformation matrix [A] are the roots λ of the
characteristic equation,
det (𝐀 − λ𝐈) = 0
Eq. 3.15
where [I] is the identity matrix. It turns out that three eigenvalues are distinct and two are repeated

λ = {Vn , Vn , Vn , Vn + a, Vn − a} , a is speed of sound

Eq. 3.16
where the left and right eigenvector of A is provided in44. Choosing the first set of eigenvectors, both
left and right, the matrix of right eigenvectors for 2D flow is obtained after eliminating the fourth row
and fifth column from the general result, whereas the left eigenvector matrix is found by deleting the
fifth row and fourth column from its original 5x5 matrix. It is interesting to note that after applying
the above 2D definitions, the singularities in the last row of the new 4x4 left eigenvector matrix
disappear. It was demonstrated earlier that two of the five right eigenvectors form a 2-dimensional
subspace, within the general 5-dimensional space spanned by all right eigenvectors, and that every
member of this subspace is itself an eigenvector. This phenomenon was attributed to the fact that
their corresponding eigenvalues are repeated, which creates a “symmetry” within the eigenvector
space. Although it may seem difficult to visualize any symmetry within a 5-dimensional vector space,
part of this symmetry reveals itself when we geometrically interpret the 2-dimensional subspace as
a plane. The Euler equations contain three types, or families of waves, one for every distinct
eigenvalue. Each family of waves carries a different signal. The waves traveling at the speed of the
flow (vn) are called entropy waves, their signal being entropy, whereas waves traveling at the speed
of sound relative to the flow (vn+_ a) are called acoustic waves. Unfortunately, the signal carried by
acoustic waves is not quantifiable in simple thermodynamic terms, but let us just say that they carry
acoustic information45.

44 Axel Rohde, “Eigenvalues and Eigenvectors Of The Euler Equations In General Geometries”, AIAA 2001-2609.
45 See previous.
 59

3.3.2 Case Study 2 – Non-Reflecting Boundary Condition in 2D

When calculating a numerical solution to an unsteady, hyperbolic, partial differential equation on an
infinite domain, it is normal to perform the calculation on a truncated finite domain. This raises the
problem of choosing appropriate boundary conditions for this far-field boundary. Ideally, these
should prevent any non-physical reflections of the outgoing waves, and should be straightforward to
implement numerically. Also, they must produce a well-posed analytic problem since this is a basic
requirement for the corresponding numerical approximation to be consistent and stable. Here the
aim is at turbomachinery flows. In some ways, these flows are more complex than the flow past
isolated airfoils. Wherever the far-field boundary for isolated airfoils can be many chords away from
airfoils, with turbomachinery the far-field can be typically less than one chord away from the blade.
Consequently, whereas for isolated airfoils the steady-state far-field can be modeled as a vortex
correction to the free stream flow, in turbomachinery the far-field contains a significant component
of several different spatial wavenumbers. This is particularly true for flows which are supersonic in
the flow direction but subsonic in axial direction, in which case shocks propagate indefinitely and can
be reflected by improper boundary conditions. Thus, one of the two aims here to correct formulation
of steady-state non-reflecting boundary conditions which will not produce artificial reflections of
steady waves such as shocks46. The other objective is the formulation of accurate non-reflecting
boundary conditions for unsteady waves. Here again isolated airfoils generally have few issues. The
reason is that the primary concern for isolated airfoils is unsteady flow caused by either airfoils
motion (Airfoil flutter) or a fluid dynamics instability (transonic buzz or install). In either case the
unsteadiness originates in the vicinity of the airfoil and radiates outward. Typically the grids on
which such calculations are tend to perfumed becoming progressively coarser as the waves moves
outwards toward the far field boundary, until a radius is reached at which the wavelength of the
unsteady wave is of the order of a few mesh cells. At this point the numerical viscosity will dissipate
the wave and so the unsteadiness will not reach the far-field boundary and accurate non-reflecting
boundary conditions are unnecessary. One of the main concerns in turbomachinery is the
unsteadiness caused by incoming shock waves and wakes from upstream blade rows. The need to
retain an accurate representation of these incoming waves’ represents the use of coarser grids in the
far-field, an instead one must concentrates on accurate non-reflecting boundary conditions.
Unsteady flows can be split into two classes, nonlinear and linear, depending on the amplitude of the
unsteadiness. If the amplitude is sufficiently small that the disturbance everywhere can be
considered to be linear perturbations to a steady flow, then by the principal of superposition the
solution can be decomposed into a sum of modes with different temporal frequencies and different
inter-blade phase angles. Each of these modes can be analyzed separately and so the problem is
reduced to finding the complex amplitude of the harmonic disturbance. This can be achieved be
either a direct method or a pseudo-time marching method, in either case, there is a need for accurate
boundary condition, and it is found that because there is only a single frequency it is possible to
construct the ext. non-reflecting boundary conditions. In nonlinear unsteady flow there are regions
where the amplitude of the unsteadiness is great enough for a second order effects to become
extremely important. This produces a coupling between the different frequencies, and so they cannot
be separated. In the far-field however it is again assumed that the unsteadiness amplitude are small
so that linear theory can be applied. It is no longer possible to construct exact non-reflecting
boundary conditions which can be implemented numerically, but approximate boundary conditions
can be derived instead.
3.3.2.1 General Fourier Analysis and Eigenvectors
Consider the following general unsteady, two dimensional, hyperbolic PDE,

46 M. Giles, “Non-Reflecting Boundary Conditions for the Euler Equations”, CFDL-TR-88-1, Feb. 1988.
 60

U U U
+A +B =0 Eq. 3.17
t x y

Where U is an n-component vector, A and B are constant N x N matrices. Fourier analysis considers
wave like solutions of the form

U(x, y, t) = u ei(kx + ly−ωt) Eq. 3.18

Substituting this into the PDE, results in

(−ωI + kA + lB) u = 0 or det(−ωI + kA + lB) = 0 Eq. 3.19

This is called the dispersion relation and is a polynomial of degree N in each ω, k, and l. The right
eigenvectors requires that

(−ωI + kA + lB)u R = u L (−ωI + kA + lB) = 0

Eq. 3.20
Where uR and uL are right and left eigenvectors of (kA+lB) with eigenvalues ω. By pre-multiplying
Eq. 3.20 by A-1 the left eigenvectors we obtain:

u L (−ωA −1 + kI + lA −1B) = v L A −1 (−ωI + kA + lB) = 0 Eq. 3.21

From above relation it is obvious that uL is equivalent to vLA-1 and uR is right eigenvector of –k also.
The difference between two sets of left eigenvectors lies in their orthogonality relation with the right
eigenvectors. Since uL is the left eigenvector of (kA+lB), it is orthogonal to the all of the right
eigenvectors of the same matrix except for the ones with the same eigenvalue ω. The key point here
is that the orthogonality for the same k and l but different ω. Thus if ωn and ωm are two different roots
of the dispersion equation for the same values of k and l, then

v L (ω n , k, l) u R (ω m , k, l) = 0 Eq. 3.22

Normally in discussing wave motion one is concerned with propagation on an infinite domain, and
so usually one considers a group of waves with the same k and l and different ω. In that case uR and
uL would be relevant right and left eigenvectors. In analyzing boundary conditions however, a general
solution U at the boundary x=0 can be decomposed into sum of Fourier modes with different values
of ω and l. Each of these modes is then a collection of waves with the same value of ω and l and
different
v L (ω, kn , l ) u R (ω, km , l ) = 0 Eq. 3.23

3.4 Solving Systems of Linear Algebraic Equations

The system is often written in the form [A] x = b where [A] is an M x N matrix, x is an N-vector and b
is M-vectors.
 61

3.4.1 Direct Methods

The full description of these methods are provided in details in47. Here, a brief overview are given for
familiarizing the user and as depicted in Error! Reference source not found..
3.4.1.1 Cramer’s Rule
This is one of the most elementary methods used. Unfortunately the algorithm is immensely time
consuming. The number of operations being approximately proportional to (N+1)! Where there are
N unknowns. Therefore, Cramer’s rule should be avoided.
3.4.1.2 Gaussian Elimination
This is very useful and efficient tool for solving systems of algebraic equations, particularly for special
cases of tridiagonal systems. However, the method is not as fast as some others. Approximately, N3
multiplication are required in solving N equations. Also, round-off errors which can be accumulate
through the many algebraic operation when N is large. Rearranging the equations to extend possible
in order to put the coefficients of the largest on the main diagonal (pivoting) will tend to improve
accuracy. The main objective is to transform the system to upper triangle array by eliminating some
of unknowns.
3.4.1.3 Thomas Algorithm
Referring the tridiagonal matrix of coefficients, the system is put into an upper triangle form. Some
flexibility exists in the way in which boundary conditions are handled.
3.4.1.4 Advanced-Direct Methods
Direct methods for systems of algebraic equations which are faster than Gaussian elimination exists.
Unfortunately, none of these methods are general. That is, they are applicable only special class of
equations and associated boundary conditions. Many of these Methods are “field size” limits.
Examples of these are “Error Vector Propagation (EVP), and “Odd-Even Reduction” methods. Clearly,
the fast direct methods should be considered for the problem where overriding consideration is
expected to be CPU time. Again these algorithms are complex and limited.
3.4.2 Iterative Methods
This class of methods is referred as “Relaxation Methods”. It may be further broken into point or
Explicit iterative, and block or Implicit
iterative methods 48. The implementation of
an elliptic equation like Laplace Equation can
be best suited for this purpose. While
iterative methods can sometimes be viewed
as integration in time, the fact that we are
only interested in the “steady-state” solution
allows us to take “short cuts” to get there as
fast as possible as shown in
Figure 3.3. We will cover two types of
iterative methods. Stationary methods are
older, simpler to understand and implement,
but usually not as effective. Nonstationary
methods are a relatively recent
development; their analysis is usually harder
to understand, but they can be highly Figure 3.3 Iteration Methods Path
effective. The nonstationary methods we

47 Anderson, Dale A; Tannehill, John C; Plecher Richard H; 1984:”Computational Fluid Mechanics and Heat
Transfer”, Hemisphere Publishing Corporation.
48 See previous.
 62

present are based on the idea of sequences of orthogonal vectors. (An exception is the Chebyshev
iteration method, which is based on orthogonal polynomials). The rate at which an iterative method
converges depends greatly on the spectrum of the coefficient matrix. Hence, iterative methods
usually involve a second matrix that transforms the coefficient matrix into one with a more favorable
spectrum. The transformation matrix is called a preconditioner. A good preconditioner improves the
convergence of the iterative method, sufficiently to overcome the extra cost of constructing and
applying the preconditioner. Indeed, without a preconditioner the iterative method may even fail to
converge49. Below are short descriptions of each of the methods to be discussed, along with brief
notes on the classification of the methods in terms of the class of matrices for which they are most
appropriate. In later sections of this chapter more detailed descriptions of these methods are given.
3.4.2.1 Stationary Iterative Methods
Iterative methods that can be expressed in the simple form

x k = [𝐁]x k−1 + 𝐜
Eq. 3.24
where neither [B] nor c depend upon the iteration count k are called stationary iterative methods.
(see Eq. 3.24).
3.4.2.1.1 Jacobi Method
The Jacobi method is a method of solving a matrix equation on a matrix that has no zeros along its
main diagonal [Bronshtein & Semendyayev]50. Each diagonal element is solved for, and an
approximate value plugged in. The process is then iterated until it converges. This algorithm is a
stripped-down version of the Jacobi transformation method of matrix diagonalization. The Jacobi
method is easily derived by examining each of the n equations in the linear system of equations Ax =
b in isolation. If, in the i-th equation

n
bi −  a ijx kj −1
a x
j=1
ij j = bi  x ik =
ji

a ii Eq. 3.25

Solve for the value of xi while assuming the other entries of x remain fixed.
3.4.2.1.2 Gauss-Seidel Iteration
Among the many iteration schemes available, Gauss-Seidel in one of the most efficient and useful
point-iterative procedure for large system of equations. The method is extremely simple but
converges under certain conditions related to “diagonal dominance” of the coefficients matrix.
Fortunately, the differencing of many steady-state conservation statements provide this diagonal
dominance. The method uses explicit use of the sparseness of matrix coefficients. A sufficient
condition for convergence of GS procedure is

n
a ii   a ij Eq. 3.26
j=1
j i

49Numerical-Methods.com
50 Bronshtein, I.
N. and Semendyayev, K. A.,”Handbook of Mathematics “, 3rd ed. New York: Springer-Verlag, p. 892,
1997.
 63

The simplicity of the procedure will be demonstrated below,

• Make initial guess of all unknowns (except one).
• Solve each equation for the unknowns whose coefficients are largest in magnitude using
guessed value initially, and the most recent it computed value thereafter.
• Repeat iteratively until changes in unknown becomes small (x = xcalculated + error).
Finally, for a general system of equations, the multiplications per iteration could be as great as N2 but
could be much less if matrix is spares.
3.4.2.1.3 Successive Over-Relaxation (SOR)
Successive Over-Relaxation (SOR) is a technique which can be used to accelerate any iterative
procedure, chief among them Gauss Seidel method. As we apply Gauss-Seidel iteration to a system
of equations, we expect to make several recalculations or iterations before convergence to the
acceptable level. Why not go head and make a correction before the next iteration, hopefully,
accelerating the convergence. According to the form:

n +1*
(
u i, j = u i,n j + ω u i,n +j 1 − u i,n j
* *
)
n = iteration level , u i,n +j 1 = the most recent value , u i,n j = value from previous iteration
*
Eq. 3.27

u i,n +j 1 = newly adjusted and 1 ω  2

*

Where ω is the relaxation parameter and Iterative Method Number of iterations

when 1 < ω < 2, over-relaxation is being
Jacobi 1989
employed. In some problems under-
Gauss-Seidel 986
relaxation 0 < ω < 1 is appears to be
SOR 91
most appropriate when the convergence
is taking oscillatory pattern tends to Table 3.1 Comparison of Iterative Schemes
over-shoot the final solution. As a
comparison on convergence for the 2D
Laplace equation with Dirichlet boundary conditions on, been made by [Grétar Tryggvason-2013]
and the results are for average absolute error (0.001) is presented in
Table 3.1.
3.4.2.1.4 SOR by Line
Although this procedure is workable with almost any
iterative schemes, it make the most sense to work with
Gauss-Seidel method with SOR. We can chose either
rows or columns for grouping with equal ease. In the
SOR by line, one iterative cycle is completed when the
tridiagonal inversion has been applied to all the rows.
The process then repeated until convergence is
achieved. The improved convergence rate observed for
block-iterative methods compared with point-iterative
methods might be due to greater influence by boundary
values. Where ω is the relaxation parameter and when Figure 3.4 SOR by line
1 < ω < 2, over-relaxation is being employed. In some
problems under-relaxation 0 < ω < 1 is appears to be most appropriate when the convergence is
taking oscillatory pattern tens to over-shoot the final solution. The final results, shown in Figure 3.4,
will be
 64

u i,n +j 1 = (1 − ω)u i,n j +

ω
2(1+ β )
2
 (
u in++1,1 j + u in−+1,1 j + β 2 u i,n j+1 + u i,n +j−11 ) Eq. 3.28

Where n denotes the iteration level, i, j, are represents rows and columns. Thomas Algorithm can be
used to sweep by rows and then advances to the next row. β = Δx/Δy is the grid aspect ratio and ω ≤
(1+ β) in order to ensures the diagonal dominance.
3.4.2.1.5 Block-Iterative Methods
The Gauss-Seidel iteration method with SOR stands as the best all-around method for finite-
difference solution of elliptic equation. The number of iterations can be usually reduced even further
by use of block-iterative concepts, but number of algebraic operations required per iterative cycle
generally increase. In block iterative group, subgroups of the unknown are singled out and their value
modified simultaneously by obtaining a solution to the simultaneous algebraic equations be
eliminating method. Thus, the block iterative have an implicit nature and sometimes called implicit-
iterative method. The simplest block procedure is SOR by line.
3.4.2.1.6 Strongly–Implicit Methods
Another type of block-iterative procedure has been gaining favor as an efficient method for solving
the system of equations, arising from elliptic PDE. To illustrate,

[ A]u = C Eq. 3.29

Where [A] is the relatively sparse matrix of unknown coefficients, u is the column vector of unknown,
and C is a column vectors on known quantities. The objective is to replace the sparse matrix [A] by a
modified matrix [A+P] which can be decomposed into an upper and lower triangle matrices [U] and
[L]. An iterative procedure is defined by

A + P u n +1 = C + P u n or LU u n +1 = C + P u n

Define : V n +1 = U  u n +1 Eq. 3.30
Step1 : L V n +1 = C + P  u n Step2 : U  u n +1 = V n +1
 65

Which is repeated iteratively. Step 1

consists simply of a forward substation.
This is followed by the backward
substitution of Step 2.
3.4.2.1.7 ADI Methods
The SOR by lines proceeds by taking all
the lines in the same direction in a
repetitive manner. The convergence
rate often improved by following the
sequence by rows by a second sequence
in the column. Thus the complete
iteration cycle would consists of a sweep
over all rows followed sweeps over the
columns. The procedure best illustrated
in Figure 3.5. In numerical analysis, the
Alternating Direction Implicit (ADI) Figure 3.5 ADI Stencil for Implicit Method
method is a finite difference method for
solving parabolic, hyperbolic and elliptic partial differential equations. It is most notably used to
solve the problem of heat conduction or solving the diffusion equation in two or more dimensions. It
is an example of an operator splitting method. This method results in a very complicated set of
equations in multiple dimensions, which are costly to solve. The advantage of the ADI method is that
the equations that have to be solved in each step have a simpler structure and can be solved efficiently
with the tridiagonal matrix algorithm. Alternatively, we could include the over-relaxation as part of
row and column sweeps in two steps as:

Step1 : u i,n +j 1/2 = (1 − ω)u i,n j +

ω
2(1+ β )
2
 (
u in++1,1/2j + u in−+1,1/2j + β 2 u i,n j+1 + u i,n +j−1/2
1 )
Eq. 3.31
Step2 : u i,n +j 1 = (1 − ω)u i,n +j 1/2 +
ω
2(1+ β )
2
 ( )
u in++1,1/2j + u in−+1,1 j + β 2 u i,n +j+11 + u i,n +j−11

To preserve the diagonal dominance in Thomas Algorithm we require that ω ≤ (1+β2) in the sweep
by row and ω ≤ (1+β2)/β2 in the sweep by column.
3.4.2.2 Non-Stationary Iterative Methods
Nonstationary methods differ from stationary methods in that the computations involve information
that changes at each iteration. Typically, constants are computed by taking inner products of
residuals or other vectors arising from the iterative method51.
3.4.2.2.1 Conjugate Gradient (CG)
The conjugate gradient method derives its name from the fact that it generates a sequence of
conjugate (or orthogonal) vectors. These vectors are the residuals of the iterates. They are also the
gradients of a quadratic functional, the minimization of which is equivalent to solving the linear
system. CG is an extremely effective method when the coefficient matrix is symmetric positive
definite, since storage for only a limited number of vectors is required.
3.4.2.2.2 Minimum Residual (MINRES) and Symmetric LQ (SYMMLQ)

51 Numerical-Methods.com.
 66

These methods are computational alternatives for CG for coefficient matrices that are symmetric but
possibly indefinite. SYMMLQ will generate the same solution iterates as CG if the coefficient matrix
is symmetric positive definite.
3.4.2.2.3 Generalized Minimal Residual (GMRES)
The Generalized Minimal Residual method computes a sequence of orthogonal vectors (like
MINRES), and combines these through a least-squares solve and update. However, unlike MINRES
(and CG) it requires storing the whole sequence, so that a large amount of storage is needed. For this
reason, restarted versions of this method are used. In restarted versions, computation and storage
costs are limited by specifying a fixed number of vectors to be generated. This method is useful for
general non-symmetric matrices.

3.4.2.2.4 Bi-Conjugate Gradient (BiCG)

The Bi-conjugate Gradient method generates two CG-like sequences of vectors, one based on a system
with the original coefficient matrix A, and one on AT . Instead of orthogonalizing each sequence, they
are made mutually orthogonal, or “bi-orthogonal”. This method, like CG, uses limited storage. It is
useful when the matrix is non-symmetric and nonsingular; however, convergence may be irregular,
and there is a possibility that the method will break down. BiCG requires a multiplication with the
coefficient matrix and with its transpose at each iteration.
3.4.2.2.5 Quasi-Minimal Residual (QMR)
The Quasi-Minimal Residual method applies a least-squares solve and update to the BiCG residuals,
thereby smoothing out the irregular convergence behavior of BiCG. Also, QMR largely avoids the
breakdown that can occur in BiCG. On the other hand, it does not effect a true minimization of either
the error or the residual, and while it converges smoothly, it does not essentially improve on the
BiCG.
3.4.2.2.6 Conjugate Gradient Squared (CGS)
The Conjugate Gradient Squared method is a variant of BiCG that applies the updating operations for
the A-sequence and the AT -sequences both to the same vectors. Ideally, this would double the
convergence rate, but in practice convergence may be much more irregular than for BiCG. A practical
advantage is that the method does not need the multiplications with the transpose of the coefficient
matrix.
3.4.2.2.7 Chebyshev Iteration
The Chebyshev Iteration recursively determines polynomials with coefficients chosen to minimize
the norm of the residual in a min-max sense. The coefficient matrix must be positive definite and
knowledge of the extremal eigenvalues is required. This method has the advantage of requiring no
inner products52.
3.4.3 Classical (Stationary) Iterative Schemes vs. Krylov Subspace Methods
The iterative schemes discussed so far are belong to Classic or Stationary Iterative methods which
solve a linear system with an operator approximating the original one or relaxation. They based on a
measurement of the error in the result (the residual), form a "correction equation" for which this
process is repeated. While these methods are simple to derive, implement, and analyze, convergence
is only guaranteed for a limited class of matrices. Oppositely, Krylov Subspace iteration methods
which work by forming a basis of the sequence of successive matrix powers times the initial residual
(the Krylov sequence). The approximations to the solution are then formed by minimizing the
residual over the subspace formed. It is an iterative method for the numerical solution of a non-
symmetric system of linear equations, which are trademarks of CFD. The prototypical method in this
class is the Conjugate Gradient method (CG). Other include the Generalized Minimal Residual

52 Numerical-Methods.com.
 67

method (GMRES), the bi-conjugate gradient method (BiCG), and stabilized CG method (CGSTAB).
Conjugate Gradient methods currently are losing favor to more general Krylov subspace methods
based largely on the GMRES algorithm. As with conjugate gradient, this method is based on the
construction of a set of basis vectors, and formally will converge to the exact solution. Rapid
convergence in the initial iterations requires preconditioning of the matrix in both approaches.
GMRES type algorithms have the advantage that residuals decrease monotonically, and that the
algorithms are generally more robust. They have the disadvantage that they must store an additional
basis vector in the Krylov subspace for each iteration. The partial solution to this problem has been
to restart the solution algorithm after some number of iterations. Providing a recommendation for a
"best" solution algorithm is not currently possible. In fact variability of algorithm performance with
machine architecture and problem type suggests that a "best" algorithm exists only in an average
sense53.
Many methods work only for symmetric systems. While the pressure equation is usually symmetric,
implicit methods usually do not lead to symmetric matrices. Usually the system is preconditioned to
make it better behaved54. Generally iterative methods generate a sequence of approximations that
are used to construct a new approximation. Ideally, we only need to keep a few approximations, but
one of the more popular technique, GMRES, requires all the previous iterates. This leads to the
restarted GMRES. At the present time there does not seem to be a “best” Krylov method. In addition
to the relatively simple early methods like the conjugate gradient method, GMRES is fairly popular
(particularly the restarted version) and BiCGSTAB has been used by a number of people. A large
number of pre-written software packages for the solution of elliptic equations are also available on-
line. It is worth mentioning that for set non-linear equations, it is best to use classical Newton’s
method which of course is heavily dependent on initial guess. For further discussion, a variation on
these method is detailed in [ J. M. McDonough]55.

3.5 Converting Conservative to Primitive Variables for 2D Euler Equations

We begin with the Euler equations in the conservative form and following closely the work by
[Nishikawa]56, we have

ρ  ρu   ρv 
 ρu  ρu 2 + p  
U F G  , G =  ρuv 
+ + =0 where U =   , F = 
t x y  ρv   ρuv  ρv 2 + p
      Eq. 3.32
ρE   ρuH   ρvH 
1 p 1 2  p 1 2 2
E= + (u + v 2 ) , H = + (u + v )
 -1  2  -1  2

The vector of primitive variables is given by

53 Class Notes, “Iterative Solution of Linear Equations”.

54 Grétar Tryggvason, “Numerical Methods for Elliptic Equations-III”, lecture slides, spring 2013.
55 J. M. McDonough,” Lectures on Computational Numerical Analysis Of Partial Differential Equations“.
Departments of Mechanical Engineering and Mathematics, University of Kentucky, 1985, 2002, 2008 ©.
56 Hiroaki Nishikawa. “Forms of the Euler Equations”, January 2000.
 68

1 0 0 0 
ρ  u
u   ρ 0 0 
W=  , U = T −1 W , T −1 = v 0 ρ 0  Eq. 3.33
v  q2
  1 
p  ρu ρv 
 2 γ − 1

Now, we can transform the conservation form into the primitive variable form by multiplying:

∂𝐔 ∂𝐅 ∂𝐆 ∂𝐔 ∂𝐅 ∂𝐔 ∂𝐆 ∂𝐔
𝐓 + 𝐓( + ) = 0 , 𝐓 + 𝐓( + )=0
∂t ∂x ∂y ∂t ⏟ ∂x ∂𝐔
∂𝐔 ⏟ ∂y
𝐀 𝐁
∂𝐖 ∂𝐖 ∂𝐖
or 𝐓𝐀𝐓 −𝟏 (
+⏟ 𝐓𝐁𝐓 −𝟏 (
)+⏟ )
∂t 𝐀𝐖
∂x 𝐁𝐖
∂y
u ρ 0 0 v 0 0 p
1 0 v 0 0
0 u 0 1
where 𝐀W = ρ , 𝐁W = 0 0 v
0 0 u 0 ρ
[0 γp 0 v] [0 0 γp v]
Eq. 3.34
3.5.1 Symmetrizing Matrix Variables
This will have a major impact in preconditioning of governing equations in order to expedite the
convergence rate, as we will see later. Again, sympathizing the work by [Nishikawa]57, the vector of
the symmetrizing variables is

∂p 1
0 0 0
∂u 𝜌𝑎
∂𝐔c = ∂u , ∂𝐔c = 𝐓c ∂𝐖 , 0 𝐓c =
1 0 0
∂v 0 0 1 0
∂s [−𝑎2 0 0 1]
∂𝐔c ∂𝐔c ∂𝐔c
or 𝐓𝐀𝐓 −𝟏 (
+⏟ )+⏟𝐓𝐁𝐓 −𝟏 ( )=0
∂t 𝐀𝐜
∂x 𝐁𝐜
∂y
u a 0 0 v 0 a 0
a u 0 0 0 v 0 0
where 𝐀c = [ ] , 𝐁c = [ ]
0 0 u 0 a 0 v 0
0 0 0 u 0 0 0 v
Eq. 3.35
Similar procedures same can be applied to anther symmetrizing variables as dUm={dp/ρa, dq, qdθ,
ds}T .

57 See previous.
 69

3.6 Programming Languages & Paradigms for CFD

As mentioned above, computer programming is one of the essentials of CFD learning. The user do
not need to be a programming specialist, only to be familiar with different programming paradigms.
Therefore, a brief discussion of different programs used in CFD and their comparison is provided
below. Let first start with difference between Object-Oriented (OO) and Procedural programming.
3.6.1 Difference Between Object Oriented (OO) vs. Procedural Programming
3.6.1.1 Object Oriented Programming (OOP)
Object-oriented Programming is a programming language that uses classes and objects to create
models based on the real world environment. It is a programming languages are organized around
'objects' rather than 'actions' and 'data' rather than 'logic'. An Object-oriented Programming
application may use a collection of objects which will pass messages when called upon to request a
specific service or information. Objects are able to pass, receive messages or process information in
the form of data. One reason to use Object-oriented Programming is because it makes it easy to
maintain and modify existing code as new objects are created inheriting characteristics from existing
ones. This cuts down the development time considerably and makes adjusting the program much
simpler.
3.6.1.2 Procedural
Procedural Programming which at times has been referred to as inline programming takes a more
top down approach to programming. Object-oriented Programming uses classes and objects,
Procedural Programming takes on applications by solving problems from the top of the code down
to the bottom. This happens when a program starts with a
problem and then breaks that problem down into smaller OOP Procedural
sub-problems or sub-procedures. These sub-procedures are method procedure
continually broken down in the process called functional
object record
decomposition until the sub-procedure is simple enough to
be solved. The issue that is obvious in Procedural class module
Programming is that if an edit is needed to the program, the message procedure call
developer must edit every line of code that corresponds to
the original change in the code. An example would be if at Table 3.2 Differences between OOP
the beginning of a program a variable was set to equal the vs Procedural
value of 1. If other sub-procedures of the program rely on
that variable equaling 1 to function properly they will also need to be edited. As more and more
changes may be needed to the code, it becomes increasingly difficult to locate and edit all related
elements in the program. The similarities and dissimilarities, as detailed in Wikipedia, are depicted
in
Table 3.2.
3.6.2 Pillars of Object Oriented Programming (OOP)
Due to simplicity of maintenance and reusability, as well as straightforward design, as relates to CFD,
we describe the 4 pillars of Object Oriented Programming (OOP) as envisioned by [Shri Ram Patel],
which outlined below.
3.6.2.1 Abstraction
Abstraction is a process of exposing essential feature of an entity while hiding other irrelevant detail.
abstraction reduces code complexity and at the same time it makes your aesthetically pleasant.
3.6.2.2 Encapsulation
We have to take in consideration that Encapsulation is somehow related to Data Hiding.
Encapsulation is when you hide your modules internal data and all other implementation
details/mechanism from other modules. it is also a way of restricting access to certain properties or
 70

component. Remember, Encapsulation is not data hiding, but Encapsulation leads to data hiding.
3.6.2.3 Inheritance
The ability of creating a new class from an existing class. Like there word Inheritance literally means
it is a practice of passing on property, titles, debts, rights and obligations upon the death of an
individual. in OOP this is somehow true(Except the death of an individual), where The base class(the
existing class sometimes called as the Parent class) has properties and methods that will be inherited
by the sub class (sometimes called a subtype or child class) and it can have additional properties or
methods. Inheritance is also a way to use code of an existing objects.
3.6.2.4 Polymorphism
Just like in biology, Polymorphism refers to the ability to take into different forms or stages. A
subclass can define its own unique behavior and still share the same functionalities or behavior of its
parent/base class. Yes, you got it right, subclass can have their own behavior and share some
behavior from its parent class BUT!! not vice versa. A parent class cannot have the behavior
of its subclass.
3.6.3 Main Language Styles
The following are widely considered the main programming paradigms, as seen when measuring
programming language popularity. There is some overlap between paradigms, inevitably, but the
main features or identifiable differences are summarized in this table58:
➢ Procedural or Structured Languages
• Based on Individual statements, specifies the steps a program must take to reach a
desired state.
• FORTRAN, ALGOL60, ALGOL68, Cobol, Pascal, C, Ada
➢ Functional Languages
• When you tell the computer to do something it does it. Treats programs as evaluating
mathematical functions and avoids state and mutable data.
• LISP, Scheme, CLOS, ML, Haskel
➢ Logic Languages
• Inference engine that drives things. Defines program logic, but not detailed control
flow.
• Prolog, GHC
➢ Object-Oriented Languages
• Bring together data and operations. Organizes programs as objects: data structures
consisting of data fields and methods together with their interactions.
• Smalltalk, C/C++, Eiffel, Sather, Python, Ada95, Java, OCAML
Note that none of the main programming paradigms have a precise, globally unanimous definition,
nor official international standard. Nor is there any agreement on which paradigm constitutes the
best method to developing software. The subroutines that implement OOP methods may be
ultimately coded in an imperative, functional, or procedural style that may, or may not, directly alter
state on behalf of the invoking program
3.6.4 Comparison Between MATLAB, C/C++, Python and FORTRAN
On the subject of comparing programming languages such as FORTRAN, C and C++, John C. Chien,
among others, writes the following distinction:

58 From Wikipedia, the free encyclopedia.

 71

3.6.4.1 FORTRAN
In Fortran, you deal with main program, subroutines, data and variables separately. The main
program calls the subroutines which then operate on the data and variables, to print, to compute etc...
For example, to print a number in Fortran, you first define the number, and then write the number
to the screen. In Fortran, data and related functions are defined separately. For example, to print a
number in Fortran, you first define the number, and then write the number to the screen.
3.6.4.2 C/C++
In C++, you deal with main program, classes and objects and functions separately. The class includes
its data and related functions as members. By changing the data, you can define different objects from
the class. In this area, it is like a parametric representation. Member functions are just functions
which operate on the data members. In C++, you declare a class called, say "class print number", then
inside the class you put in the number as the data member, and the function which print the number
as the member function, say "prt()". Then, in the main function (program) you first create an object
from the class, this is a copy of class with number and function in it. You exercise the member function
of the object to print the number. If you had defined the object as "print number myprint", then
"myprint.prt" will print the number for you. The difference is "packaging", in Fortran, there are parts,
in C++ the parts are packaged into many different classes. In this way you can protect your data
similar to your bank account, or money in your pocket instead of money on the desk top with your
name in front of it. With this concept, you can also derive other classes with additional functions in
it, something like custom made PC with add-on in it.
As you see, Fortran and C++ are different things, especially related to the use of classes and objects.
So, if you treat data and the related functions as separate things, then Fortran and C are the way to
go. On the other hand, if you group the data and the related function into classes and use it as object,
then C++ has some nice features for you. Unfortunately, the transformation from Fortran and C to
C++ is not simple because it requires the change in concept. By the way, JAVA is also object-oriented
,without the complexity of the pointer. If your main interest is the solver and number crunching,
then Fortran is the one designed for that purpose, on the other hand, C++ will improve your
vision of the current picture of computer programming technology, therefore, it is essential to
understand its basis concepts.
Similar point of view is in complementing above is expressed by [Taked] of Department of
Aeronautics and Astronautics University of Southampton England. Fortran 77 is about the fastest
high-level language for doing numerical computation. Good C code can go about as fast, but it is easier
to write fast F77 than fast C. The reason is that F77 compilers can do a better job optimizing the code.
With C, C++ and FORTRAN 90 the existence of pointers can be detrimental for performance. The
compilers cannot optimize memory accesses if the data is spread all over the place in memory.
However, pointers are incredibly useful for creating useful data structures, such as a linked list for
an unstructured grid. Bear in mind that the performance could be abysmal if implemented naively
though. FORTRAN 90 has two very useful features. The first is dynamic memory allocation. The
second is modules. These make program manageability much easier. Therefore, I think that you can
do well by using a mixture of Fortran 90 and 77. F77 for the number crunching numerical kernel
routines, and F90 for job setup and program management. Similarly, C/C++ could be used together
with F77 numerical routines. This seems to give a good balance between performance and flexible
code structure.
3.6.4.3 Python
According to [Michael Prinkey] via CFD online, Python is a wonderful language and NumPy and SciPy
give you a wonderful workshop to quickly prototype. If you are investigating high-level algorithms,
start here. You won't get production speed code from it, but you will quickly glean a thorough
understanding of what algorithms work the best. There is also a undemanding paper published by [
 72

Kardontchik, J. E., 2014]59, which describes step by step programming with Python for CFD functions.
Other sources for Python programming using CFD is also widely available online.
3.6.4.4 MATLAB
MATLAB (matrix laboratory) is a multi-paradigm numerical computing environment and fourth-
generation programming language. A proprietary programming language developed by
MathWorks®, MATLAB allows matrix manipulations, plotting of functions and data, implementation
of algorithms, creation of user interfaces, and interfacing with programs written in other languages,
including C, C++, C#, Java, Fortran and Python. Although MATLAB is intended primarily for numerical
computing, an optional toolbox uses the MuPAD symbolic engine, allowing access to symbolic
computing abilities. An additional package, Simulink, adds graphical multi-domain simulation and
model-based design for dynamic and embedded systems60. In MATLAB, basic data type is matrix.
This would reduce your efforts a lot while writing a CFD code. On the contrary, a program written in
C or Python would give you a complete control over robustness, speed, accuracy and hence efficiency.
Now, MATLAB will solve a Matrix operation for you. C will allow you to write a program that solves
a Matrix operation. Somewhere on the middle ground, C++ is an object-oriented version of C. If your
aim is to want to use CFD as your research tool or to use it in research and Development
environments, it would be rather suggest you to go ahead with MATLAB. Elsewhere, (i.e., working in
industry using commercial codes), bite the bullet and to use C/C++/Python etc. Other consideration
is economics and time. How much money and time do you have? This is more of a commercial
question. If you have money and no time to write your own matrix manipulation code, buy MATLAB
and use it! Of course, you may want to explore Octave, a rebellious brother of MATLAB.
3.6.4.5 Graphics and GUI programming in MATLAB
MATLAB supports developing applications with graphical user interface (GUI) features. MATLAB
includes GUIDE (GUI development environment) for graphically designing GUIs61. It also has tightly
integrated graph-plotting features. For example, the function plot can be used to produce a graph
from two vectors x and y. In addition, MATLAB program can produce 3D graphics using the functions
surf, plot3 or mesh is illustrated in Figure 3.6. Further information regarding the MATLAB could be

[X,Y] = meshgrid(-10:0.25:10,-
10:0.25:10);
f = sinc(sqrt((X/pi).^2+(Y/pi).^2));
mesh(X,Y,f);
axis([-10 10 -10 10 -0.3 1])
xlabel('{\bfx}')
ylabel('{\bfy}')
zlabel('{\bfsinc} ({\bfR})')
hidden off

Figure 3.6 3D Representation of a Sine Function using MATLAB

59 Jaime E. Kardontchik, “Introduction to CFD with Python”, April 16, 2014.

60 From Wikipedia, the free encyclopedia.
61 "MATLAB GUI", MathWorks. 30 April 2011. Retrieved 14 August 2013.
 73

obtained from excellent resource such as 62-63 among many others.

62 David Houcque, “Introduction To Matlab For Engineering Students”, Northwestern University, August 2005.
63 Introduction to MATLAB.
 74

4 CFD Basics
With the advent of Computational Fluid Dynamics (CFD) in recent decades, it has become an
important tool in dealing with complex issues on different disciplines and engineering applications.
It’s relatively low cost, in relation to experimental data, and the limited availability of theoretical
(analytical) data, made it an integral part of design and analysis loops. More importantly, it is
CFD
relatively easy to adapt to new challenges and problems. Although in the beginning it was only used
by limited number of researches and academics in aeronautical and astronautics, today it is an
integral tool for wide variety of applications. But it is still considered somewhat and art and science,
and in some circles, it is recommended
strongly to be used in conjunction with
experimental data, if available.
Moreover, it is also recommended to be
validated with a simplified and sub-set
of theoretical results, whenever
available. These colorations, known as CFD
Verification & Validation are among an
intensive on-going research in CFD field.
Although there are numerous
definitions on CFD, all excellent in their CFD
own rights, the simplest one could be
stated as CFD is the transformation of
the governing equations from
continuum domain into a discrete one.
The PDE’s are approximated by
difference methods (FDE), resulting in
system of algebraic equations to be
solved numerically. The nature of these Figure 4.1 Contributions from other disciplines to CFD
algebraic equations depends on the
character of the problem posed by the original PDE. The equilibrium (BVP), mathematically elliptic,
usually results in a system to be solved simultaneously in conjunction with satisfying the boundary
conditions. While the marching problems (parabolic or hyperbolic) are usually solved one at a time,
conforming to initial boundary conditions. There are three distinctive disciplines contributing to CFD
analysis;
1. The fluid dynamics which describes the physics of flow behavior as described in in any Fluid
Dynamic text.
2. The mathematical side or the Numerical Analysis which transforms the fundamental
governing equation from continuum into a discrete domain.
3. The computer science (CS) which does that transformation using high-level programming
languages.
The question now arises that do we need expertise of three people from these disciplines in order to
carry out a CFD analysis? The answer is more likely that a person that has some proficiency on each
subsets would be sufficient. These inter-relations are depicted in Figure 4.1. Emphasis should be
made here regarding the Verification & Validation of CFD analysis with respect to analytical or
experimental data. In regard to items (1) and (2), we discussed them briefly before. We will concern
our self with number (3), but restrict ourselves with general information since the aim is not to pay
particular homepage to specific languages, but outline the general information available for CFD and
 75

its particulars. The amount of information

and discussion and literature available on
each topic is enormous. Figure 4.2 shows CFD
the three approaches to fluid dynamic
problems and their strong link as these
approaches do not work in isolation. All
three approaches complement each other
and could be used in coordination. While
the experimentalist might prefer of-
course the experimental approach, or the Analytical Experimental
academics the analytical one, the more
realist and deadline conscious analyst Figure 4.2 Linkage between CFD, Experimental &
would probably prefer the CFD in Analytical approach
conjunction with other two.

4.1 Marching vs. Equilibrium Problems

Marching or propagation problems are transient or transient-like where the solution of PDE is
required on an open domain, subject to a set of Initial or Boundary conditions. Mathematically, these
are parabolic or hyperbolic, as discussed previously where the flow of information is upstream.
Typical examples include unsteady-inviscid flow, steady-state inviscid supersonic, transient heat
conduction, as well as boundary layer flows. Alternatively, the equilibrium problem is where the
solution is subjected to BC’s enclosing the domain, therefore, they are classified as Boundary Value
Problems (BVP). Mathematically, they are elliptic in nature as the flow of information is scattered
through the domain and propagates in all direction. Examples of equilibrium problems are steady-
state temperature distribution, incompressible inviscid flows, and equilibrium stress distribution in
solids.

4.2 Non-Linear CFD Equations

The solution methods could be depending on flow characteristics such steady/un-steady,
compressible/incompressible, viscous/inviscid as well as parameters such as Mach number which
govern flow speed for subsonic/transonic/supersonic regions. Each dependency, or their
combination, could alter the mathematical nature of the equations, therefore, influencing the
selection of appropriate method of solution. Moreover, the 2D & 3D versions of same equations could
under similar conditions, exhibit different characteristics. For example, while 2D boundary layer
equations are parabolic in nature and could be solved by marching methods, the 3D are hyperbolic
and require different technique. The time dependency could also play dominant role. Prime example
would be un-steady Euler equation which of course requires time integration as it is hyperbolic for
all flow regions, and could benefit from a time marching procedure. On the other hand, the steady-
state version, although simpler, has elliptic behavior for subsonic regions, parabolic for sonic and
hyperbolic for supersonic flows. Therefore, requiring different strategy on each occasion as
previously noted64.

64Diskin, Boris; Thomas, James: “Comparison of Node-Centered and Cell-Centered Unstructured Finite-Volume
Discretizations: Inviscid Fluxes,” AIAA Journal 2010.
 76

4.3 Coupled vs Segregated Algorithms

The segregated and coupled approaches differ in the way that the continuity, momentum, and (where
appropriate) energy and species equations are solved, separated. The solver solves these equations
sequentially (i.e., segregated from one another – see Figure 4.3), while the coupled solver solves
them simultaneously (i.e., coupled together – see Figure 4.4). Both formulations solve the equations
for additional scalars (e.g.,
turbulence or radiation
quantities) sequentially. The
segregated solver traditionally
has been used for
incompressible and mildly
compressible flows. The coupled
approach, on the other hand,
was originally designed for high-
speed compressible flows. Both
approaches are now applicable
to a broad range of flows (from
incompressible to highly
compressible), but the origins of
the coupled formulation may
give it a performance advantage
over the segregated solver for
high-speed compressible
flows.65 Coupled flow needs
more resources like memory
and computational time as it
solves coupled equations. But
that also means, it's more stable
in cases with high density
fluctuations like supersonic flow Figure 4.3 Segregated Solution
with shocks etc. It's not the best
choice for a standard case as it might be unstable unless you reduce the courant number a lot.
Segregated flow is a good choice for most cases since it runs fast, but can have problems with
supersonic flows. Mathematically, the main difference is that the fully coupled solver operates on the
full Jacobian matrix as one entity. The segregated solver splits the Jacobian matrix into smaller sub-
problems, usually by degree of freedom type. Different solution strategies can then be used for each
sub-problem. The optimal choice between segregated and fully coupled is problem specific. The fully
coupled solver generally requires less iterations but takes up more memory and solution time per
iteration.66

4.4 Explicit vs. Implicit Methods

The solution methods could be characterized into two main category of Explicit and Implicit. While
the Explicit schemes have the advantage of relative ease in implementation, as they don’t require any
matrix inversions, but are bounded by step size and stability considerations. On the other hand, the
implicit schemes are more prone to be stable, but computationally intensive as they mostly require
linearization of systems of equations and matrix solvers. The linearization of the equations are mostly
achieved by Taylor series expansion of non-linear coefficient matrixes. The final resultant expression

65 Solver Panel, Fluent® Inc. 2003.

66 N. Elabbasi, Certified Consultant, Veryst Engineering, COMSOL Discussion Forum.
 77

could be devised as multiple simpler steps to solve using concepts such Approximate Factorization
(AF). The matrix solvers are also could be grouped as Iterative solvers (Guass-Seidel, SOR, ADI, block
iterative) or direct solvers using schemes such as Guassian Elimination or Thomas Algorithm. Multi-
Grid methods such those implicating V or W cycles for interpolating and extrapolating the results
back and forth between different mesh densities. The validity and feasibility of each method depends
to the application. In
Computational Fluid Dynamics
(CFD), the governing equations
are nonlinear, and the number
of unknown variables is
typically very large. Under these
conditions implicitly formulated
equations are almost always
solved using iterative
techniques. Iterations are used
to advance a solution through a
sequence of steps from a
starting state to a final,
converged state. This is true
whether the solution sought is
either one step in a transient
problem or a final steady-state
result. In either case, the
iteration steps resemble a time-
like process. Of course, the
iteration steps usually do not
correspond to a realistic time-
Figure 4.4 Coupled Solutions
dependent behavior. In fact, it is
this aspect of an implicit method
that makes it attractive for steady-state computations, because the number of iterations required for
a solution is often much smaller than the number of time steps needed for an accurate transient that
asymptotically approaches steady conditions. On the other hand, it is also this “distorted transient”
feature that leads to the question, “What are the consequences of using an implicit versus an explicit
solution method for a time-dependent problem?” The answer to this question has two parts. The first
part has to do with numerical stability, and the second part with numerical accuracy.

4.5 Model Finite Difference Schemes

Now we examine briefly the various Finite-Difference Schemes applied to various model equations
such as Wave equations, Heat equation, Laplace equation, and Burgers equation as their exact
solution are available and obtained before. These equations can be used to model the behavior of
more complicated partial difference equations. Reader should refer to67 for further details. Here, we
only present the results in Table 4.2 and Table 4.1 as they unique to two different Implicit and
Explicit methods. Each of model exhibits certain distinctive features that characteristic of the class
of methods. Detail description on each method is available in68. Based on the information presented
here, it is clear that many techniques can be used to solve the same problem. The difference is the

67 D. Anderson, J., Tannehill, R., Pletcher, ”Computational Fluid Mechanics and Heat Transfer”, ISBN 0-89116-
471-5 – 1984.
68 Anderson, Dale A; Tannehill, John C; Plecher Richard H; 1984:”Computational Fluid Mechanics and Heat

Transfer”, Hemisphere Publishing Corporation.

 78

quality of solution produced and its applicability to problem in hand. The selection can be aided by
experience gained in programming the various methods to solve the equations.

Table 4.2 Explicit Discretization Methods of Model Equations

Methods Accuracy Stability Test Case Viscosity Usage
ADE 1st order Yes 2D Heat None Not Recommended
Brailovskaya 1st order Conditional 1D Burges Viscous Not Recommended
DuFort-Frankel 2nd order Yes 1D Heat None Recommended
Euler 1st order No 1D Wave None Not Recommended
FTCS 1st order No 1D Burges Viscous Not Recommended
Hopscotch 1st order Yes 2D Heat None Not Recommended
Iterative Method Explicit Conditional 2D Laplace None Recommended
Lax 1st order Conditional 1D Wave None Not Recommended
Lax-Wendroff 2nd order Conditional 1D Wave None yes Recommended
Leap Frog 2nd order Conditional 1D Wave None yes Recommended
MacCormack 2nd order Conditional 1D Wave None yes Recommended
MacCormack 2nd order yes 1D Burgers Viscous yes Recommended
Richardson’s 2nd order no 1D Heat None Not Recommended
Rusanov 4th order Conditional 1D Burges Viscous Recommended
Simple Explicit 1st order Conditional 1D Heat None Not Recommended
Time -Split Mac 2nd order Conditional 2D Burgers Viscous Recommended
Cormack
Upstream 1st order Conditional 1D Wave None Not Recommended
Upwind 2nd order Conditional 1D Wave None yes
Warming-Kutler- 3rd order conditional 1D Wave None yes
Lomax
Table 4.1 Implicit Discretization Methods of Model Equations
Methods Accuracy Stability Test Case Viscosity Usage
ADI 2nd order Yes 2D Heat None yes
Allen-Cheng 1st order Conditional 1D Burgers Viscous no
Beam-Warming 2nd order Conditional 1D Burgers None yes
Briley-MacDonald 1st order Yes 1D Burgers Viscous yes
Crank-Nicolson 2nd order Yes 1D Heat None yes
Euler 1st order Yes 1D Wave None no
Simple Implicit 1st order Yes 1D Heat None no
Splitting/Fractional-Step 1st order Conditional 2D Heat None no
Trapezoidal 2nd order Yes 1D Wave None yes
Upwind – (Beam Warming) 2nd order Conditional 1D Wave None yes

4.6 General Discretization Scheme (Beam-Warming)

Among prominent and widely used integration techniques, but not limited, are Crank-Nicolson,
Beam-Warming and Mac-Cormack’s predictor-corrector schemes69. Most could be used cross
governing equations and techniques. Other notables are Euler and Trapezoidal implicit method. In
fact, most of legacy methods are interpedently related. For example, the Implicit Beam-Warming

69Anderson, Dale A; Tannehill, John C; Plecher Richard H; 1984:”Computational Fluid Mechanics and Heat
Transfer”, Hemisphere Publishing Corporation.
 79

spatial differencing for vector E is

θ1 + Δx  i Δx  i θ  1  
ΔiE = (Δ E) + (E ) + 2 Δ i −1E + O  θ1 − − θ 2 (x) 2 + (x) 3 
1 + θ 2 x 1 + θ 2 x 1 + θ2  2   Eq. 4.1
where Δ i E = Ei +1 − Ei (6.1)

This general differencing formula, with appropriate choices of parameters θ1 and θ2 reproduces
many of the standard differencing schemes as perceived in Table 4.3. As evident, many different
methods could be
used to solve the Θ1 Θ2 Scheme Order of accuracy
same problem. The 0 0 Euler (explicit) (Δx)2
difference in the 0 -1/2 Leap Frog (explicit) (Δx)3
quality of the 1/2 0 Trapezoidal (implicit) (Δx)3
solutions is 1 0 Euler (implicit) (Δx)2
frequently small and 1 1/2 Three-point backward (implicit) (Δx)3
the selection of the
optimal technique Table 4.3 Choice of Parameters
becomes difficult.
However, the selection process could be aided by experience gained to solve model equations.

4.7 Linearization of Equations

For any implicit method, the equation of fluid flow are non-linear in unknown due to appearance of
quantities at i+1 level in coefficients. Prime example is the term u (∂u/∂x) which is from convective
term in NS equation. A simple forward differencing, has been used to approximated by 1D as70

u u −u
u  u i +1 i +1 i Eq. 4.2
x Δx
Which of course non-linear.
4.7.1 Frozen (Lagging) Coefficient Method
In this method simply ignore the non-linear coefficient and evaluate it at known station as

u u −u
u  u i i +1 i Eq. 4.3
x Δx

The procedure provides a consistent representation (0th order Taylor series expansion) and ensure
that the differencing scheme is formally no better than 1st order accurate in marching coordinates.

4.7.2 Simple Iterative Method

In this method, the lagged coefficient is updated through a simple iteration until a specified
convergence criteria is satisfied71. This is complement to previous method.

u i +1 − u i  ε where   1 Eq. 4.4

70 K. A. Haffmann, S. T. Chiang, “Computational Fluid Dynamics”, 4th edition, Aug. 2000.

71 K. A. Haffmann, S. T. Chiang, “Computational Fluid Dynamics”, 4th edition, Aug. 2000.
 80

4.7.3 Newton Raphson Linearization Method

This is by far the method of choice and used extensively. The Newton’s method also called quasi-
linearization proceeds as follows. Let assume (A) and (B) represent the non-linearity. Define ∆ as
change in variable between two consecutive iterations (n),

 A = A n +1 − Â n and  B = Bn +1 − B̂n
A n +1Bn +1 = (Â n +  A )(B̂n +  B )
Expanding, after some manipulati on, and dropping the 2nd order term( A )( B )
A n +1 Bn +1  Â n Bn +1 + A n +1B̂n − Â n B̂n for n = 0 → A n +1 = Â n
Eq. 4.5
Which is now linear. The carrot denotes an evaluation of variables from previous iteration.
4.7.3.1 Newton Linearization with Coupling
Several investigators have observed that convergence of iteration can be accelerated by solving the
momentum and continuity equations in coupled manner. The v (∂u/∂y) term is linearized by using
Which of course is linear72.

v in +1 = v̂ in +1 + δ v , u in +1 = û in +1 + δ u
after product terms involving δ dropped : Eq. 4.6
n +1 n +1 n +1 n +1
 u   u   û   û 
 v   v̂ n +1   + v n +1   − v̂ n +1  
 y   y   y   y 
4.7.4 Extrapolating the Coefficients
Values of the coefficients can be obtained at n+1 level by extrapolating based on values already
obtained from previous n level. Formally, the truncation error of this procedure can be made as small
as we wish73. For example, we can use

u u u n − u in −1
n n

u n +1
=u +
n
Δx + + ο(x) 2 where = i + ο(x) 2
x i x i
i i
Δx −
Eq. 4.7
u n − u in −1
u n +1
i =u + i n
i Δx + + ο(x) 2
Δx −

A similar procedure can be used for other coefficients needed at n+1 level.
4.7.5 Case Study – Linearization of 1D Unsteady Euler Equation
A widely used method for achieving this linearization was first suggested by Beam and Warming in
1976. For purposes of discussion, let us consider the system of unsteady 1D Euler equation:

∂U ∂F
+ =0
∂t ∂x
Eq. 4.8

72 D. Anderson, J., Tannehill, R., Pletcher, ”Computational Fluid Mechanics and Heat Transfer”, 1984.
73 D. Anderson, J., Tannehill, R., Pletcher, “Computational Fluid Mechanics and Heat Transfer”, 1984.
 81

where F = F(U). Using the Crank-Nicolson differencing scheme, Eq. 4.8 can be written in finite-
difference form as
∆t ∂F n ∂F n+1
Uin+1 = Uin − [( ) + ( ) ]
2 ∂x i ∂x i
Eq. 4.9
(Sometimes the representation of the spatial derivatives as an average between time levels n and n
+ 1, is called the trapezoidal rule). Eq. 4.9, as it stands, is a nonlinear difference equation. However,
the Beam and Warming approach leads to a local linearization as follows. Expand F in a series
expansion around time level n, that is,

∂F n n+1 ∂F n
Fin+1 = Fin + ( ) (Ui − Ui )+. . . . where ( ) = Ani
n
∂U i ∂U i
Eq. 4.10
Substituting and rearranging we get

∆t ∂F n ∂ n n+1
Uin+1 = Uin − [2 ( ) + Ai (Ui − Uin )]
2 ∂x i ∂x
Eq. 4.11
Therefore, we have achieved what we wanted. We have taken a nonlinear difference equation,
namely, Eq. 4.11, and by means of a Taylor series expansion using lagging coefficients have
linearized this equation, obtaining the linear difference equation. This is one way of achieving the
linearization; there are others. However, the purpose of this subsection is to emphasize that implicit
finite-difference solutions of the conservation form of the governing flow equations lead to
nonlinear difference equations which must in some fashion be linearized before a practical
numerical solution can be obtained. It should be noted that a similar idea for linearization was carried
out by [Briley and McDonald]. In contrast to Beam and Warming, who treated the function, [Briley
and McDonald] treated the time derivative; the results are effectively the same.
4.7.6 Multi-Dimensional Problem and Approximate Factorization (AF) Scheme
The difficulty of working with large matrices resulting from straightforward implementation of
implicit schemes to PDEs in higher dimensions has led to the development of the so-called split or
factored schemes. As the name implies, such schemes split a multi-dimensional problem to a series
of one-dimensional ones, which are much easier to solve. Of course, in general, this conversion cannot
be done exactly and some error is incurred. However, as we will show below, the splitting error is of
the same order as the error already incurred in discretizing the problem in space and time. That is,
the splitting approximation does not erode the order of accuracy of the scheme.74 The basic system
uner consideration is of the form (2D) given by

U E F
+ + =0 Eq. 4.12
t x y

Where U is the vector of conservative variables and E and F are vector of function of U. If the
Traezoidal rule ( ϴ1=1/2, Ɵ2=0) is used as the basis integration scheme,

74 Moin, P; “Fundamentals Engineering Numerical Analysis”, Cambridge University Press.

 82

Δt  U   U  
n n +1

U n +1 = U n +   +  
2  t   t  
Δt  E F   E F  
n n +1
n +1
U =U −
n
 +  + +  
2  x y   x y  

Eq. 4.13
This expression provides a 2nd order integration algoritem for the unkown vector, Un+1, at the next
time level. A local Taylors series expasion of direvtives of E and F used to obtain a Linear equation
such that
E
(
E n +1 = E n + A U n +1 − U n ) where A =
U
F
(
F n +1 = F n + B U n +1 − U n ) where B =
U
Eq. 4.14
When the linearization given by Eq. 4.14 is substitued into Eq. 4.13, a linear system for Un+1 results
as
 Δt   
I  +  A n +  Bn U n +1 =
 2  x y 
 
n
Δt  
I  +  A n +  Bn U n − Δt E + F 
 2  x y   x y 
Eq. 4.15
This is a linear system for theunknown Un+1. Direct solution of Eq. 4.14 is usually avioded due to
llarge operation count in treating muti-dimensional system. The path chosen is ususlly to reduce the
multi-dimensional problem into a sequence of one-demensional inversrsions using the method of
Approximate Factorization (AP) in cross flow plane. Eq. 4.15 may be approximatly facotred into


 I  +
Δt 
A n   I  + Δt  Bn  U n +1 =
 2 x  2 y 
n Eq. 4.16

 I  +
Δt 
A n   I  + Δt  Bn  U n − Δt  E + F 
 2 x  2 y   x y 

This equation is much easier and more cost effective to implement than the large system encountered
in the non-factored form. Basically, the multi-dimensional problem is reduced to a series of one-
dimensional problems by ignoring the cross terms while maintaine the formal order of accuracy.
 83

ΔU n = U n +1 − U n
n

 I  +
Δt 
A n   I  + Δt  B n  ΔU n = −Δt  E + F 
 2 x  2 y   x y 
n

A n  ΔU ' = −Δt E + F 

 Δt  Eq. 4.17
 I  +
 2 x   x y 

 I  +
Δt 
B n  ΔU n = ΔU '
 2 y 

The solution of this system is not trival. The x and y sweeps each require the solution of block
tridiagonal system of equations, or better known as TDMA (Thomas Algorithm). Each block is M
x M if there are M elements in the unknown U vector.
4.7.7 Recommendation on Linearization
For many calculations, the linearization introduced by simply lagging the coefficients will cause no
serious deter ration of accuracy. Errors associated with linearization of coefficients are simply
truncation errors which can be controlled by adjustment in marching step size. Many investigators
have been used this procedures satisfactory. For any problem in which this linearization causes
special difficulties, extrapolation or newton’s linearization with coupling is recommended75.

4.8 Convection & Diffusion Terms Discretization

In problems where fluid flow plays a significant role we must account for the effects of convection76.
Diffusion always occurs alongside convection in nature so here we examine methods to predict
combined convection and diffusion. The principal problem in the discretization of the convective
terms is the calculation of the value of transported property Q at control volume faces and its
convective flux across these boundaries. (see Eq. 4.18). We introduced the central differencing
method of obtaining discretized equations for the diffusion and source terms on the right hand side
of equation. It would seem obvious to try out this practice, which worked so well for diffusion
problems. The diffusion process affects the distribution of a transported quantity along its gradients
in all directions, whereas convection spreads influence only in the flow direction. This crucial
difference manifests itself in a stringent upper limit to the grid size (Peclet No.), which is dependent
on the relative strength of convection and diffusion, for stable convection–diffusion calculations with
central differencing.

∂ ∂ ∂ ∂Q
(ρQ) + (ρUj Q) = (ΓQ ) + S⏟Q
⏟
∂t ∂x
⏟j ∂x
⏟j ∂xj
Transient Source
⏟Convection Diffusion
Transport
Eq. 4.18
Naturally, we also present the case for a number of alternative discretization practices for the
convective effects which enable stable computations under less restrictive conditions. For these

See Previous.
75
76H K Versteeg and W Malalasekera, “An Introduction to Computational Fluid Dynamics - The Finite Volume
Method”, Second Edition, © Pearson Education Limited 1995, 2007.
 84

terms, typically, an LUD (Linear Upwind Differencing), QUICK, TVD, NVD, second-order convective
upstream split scheme (CUSP), or other upwind differencing schemes can be used. Upwind
schemes are developed which take the direction of the flow in consideration77. Nowadays,
upwind schemes are the major spatial discretization technique of main research and commercial
codes78-79.
4.8.1 Upwind Differencing For Convection
It uses the propagation of information with the theory of characteristics in constructing the
information traveling in opposite directions in a separate and stable manner. Second order central
schemes require scalar artificial dissipation to damp oscillations generated near the high gradient

Figure 4.5 Schematic physical representation of a propagating wave in the positive x direction
accompanied by the flow field at a given instant of time and numerical representation with and without
an artificial dissipation term in the vicinity of the discontinuous wave interface

77 Korhan Coskun, “Three-Dimensional Laminar Compressible Navier Stokes Solver For Internal Rocket Flow
Applications”, A Thesis Submitted To The Graduate School Of Natural And Applied Sciences of The Middle East
Technical University, 2007.
78 Frink, N. T., "Upwind Scheme for Solving the Euler Equations on Unstructured Grids", AIAA Journal Vol. 30, No.

1, pp. 70-77, January, 1992.

79 Dadone, A., and Grossman, B., "Characteristic-Based, Rotated Upwind Scheme for the Euler Equations", AIAA

Journal Vol. 30, No. 9, pp.2219-2226, September, 1992.

 85

regions. As an 1D wave equation example80, consider the simple illustration of the inviscid Euler
equation in the non-conservative form:

∂u ∂u
+c =0
∂t ∂x
Eq. 4.19
where c is the advective velocity describing the propagation of a wave in the direction of the x axis.
There is a discontinuity in the velocity u across the wave, as described in Figure 4.5. Assuming that
c is positive, properties at grid point I should depend on the upstream flow-field properties at grid
point i -1. Grid point i +1, on the other hand, should not physically influence the point at i; hence the
choice of numerical scheme must reflect the flow physics. If the gradient ∂u/∂x is approximated
using central differencing, such as the traditional approach described above for MacCormack (1969),
the velocity u profile results in an oscillatory behavior near the discontinuous wave front. In some
circumstances, the numerical procedure can lead to an unstable and chaotic solution.
The common remedy, as mentioned, is to introduce an artificial dissipation term. Despite the
numerical result exhibiting a monotone variation (no oscillations), the diffusive property remains an
undesirable element, as illustrated for the numerical representation in Figure 4.5. Higher order
upwind methods require limiter functions for second order accuracy in space. Characteristic theory
is easy to understand in one-dimensional flows, but for 2D and 3D flow problems, the flow direction
is not clearly identified. However, choosing the upwind direction being normal to the face of the
computational cell across which the fluxes are computed is a commonly used way for 2D and 3D
flows. Upwind schemes may be divided into two categories as Flux Vector Splitting schemes and
Flux Difference Splitting (Godunov) schemes.
4.8.1.1 Flux Vector Splitting Schemes
Upwind discretization is obtained by splitting the flux vector into two parts based on information
coming from upwind and downwind of the cell face in flux vector splitting algorithms81. In other
words, the flux terms are split according to the sign of associated with propagation speeds. The
main drawback of flux vector splitting methods is evident in the vicinity of sonic conditions since
the splitting of flux is done with respect to the sign of the Mach number or the velocity vector.
Consider the Euler equations for unsteady, one-dimensional flow written as,

∂U ∂F
+ =0
∂t ∂x
Eq. 4.20
As described by Eq. 4.20, A is the Jacobian of F; A = ∂F/∂U. For an inviscid flow, the flux vector F
can be expressed directly in terms of its Jacobian as

F = AU
Eq. 4.21
Let us define two matrices [λ+] and [λ-] made up of the positive and negative eigenvalues of A,
respectively. For example, if we have a subsonic flow, then we have λ1 = u and λ2 = u + c, both positive
values, and λ3 = u - c, a negative value. Therefore, in this case, by definition

80 JiyuanTu, Guan-HengYeoh, ChaoqunLiu, “Chapter 9 - Some Advanced Topics in CFD- Computational Fluid
Dynamics (3rd Edition) A Practical Approach”, ScienceDirect, 2018.
81 Steger, J. L., and Warming, R. F., "Flux Vector Splitting of the Inviscid Gas Dynamics Equations with Applications

to Finite Difference Methods", Journal of Computational Physics, Vol. 40, pp. 263-293, 1981.
 86

u 0 0 0 0 0
[λ+ ] = [0 −
u + c 0] , [λ ] = [0 0 0 ] , c = speed of sound
0 0 0 0 0 u−c
Eq. 4.22
With this, we can split the flux vector F into two parts, F+ and F-:

+ −
∂U ∂F + ∂F −
F=F +F and + + =0
∂t ∂x ∂x
Eq. 4.23
where F+ and F- are defined by

F + = A+ U → A+ = Tλ+ T −1
F − = A− U → A− = Tλ− T −1
Eq. 4.24
Where T is vector of eigenvectors of eigenvalues λ. Eq. 4.23 is an example of flux-vector splitting
where F+ corresponds to a flux in the positive x direction, with information being propagated from
left to right by the positive eigenvalues λ1 = u and λ2 = u + c. Hence, when ∂F+/∂x is replaced by a
difference expression, a backward (rearward) difference should be used since F+ is associated only
with information coming from upstream of grid point (i, j). Similarly, F - corresponds to a flux in the
negative x direction, with information being propagated from right to left by the negative eigenvalue
λ3 = u - c. Hence, when ∂F/∂x is replaced by a difference expression, a forward difference should be
used since F - is associated only with information coming from downstream of grid point (i, j). This is
why the flux-vector-splitting scheme described by Eq. 4.24 is a type of upwind scheme; flux-vector
splitting is a numerical algorithm which attempts to account for the physically proper transfer of
information throughout the flow. There are various improvisations on flux-vector splitting in the
modem CFD literature. One such example is Van Leer's flux splitting which imposes certain
conditions on F+ and F- to improve the performance of the numerical scheme for local Mach numbers
near 1.
4.8.1.2 Flux Difference Splitting (Godunov) Schemes
In the flux difference splitting schemes, local Riemann problem is solved on each face of cells.
The flow variables are taken as constant over the left and right states of the cell face. Using left
and right states of the face, local Riemann problem is solved to achieve convective fluxes at the
face. In the original Godunov scheme82, the local Riemann problem is solved exactly. Since this
approach is computationally expensive, some approximate Riemann solvers have been built by
[Roe]83, [Osher and Solomon]84, [Toro]85.
4.8.2 Diffusion Term Discretization
The finite-volume discretization which uses meshes made of arbitrary polyhedral control volumes,
Figure 4.6, enabled one to discard with the complicated curvilinear coordinates. There, the diffusive
flux of the variable Φ through an internal cell face f is approximated as

82 Godunov, S. K., "A Difference Method for the Numerical Computation of Discontinuous Solutions of
Hydrodynamic Equations", Math Sbornik, Vol. 47, pp. 271-306, 1959.
83 Roe, P. L., "Discrete Models for the Numerical Analysis of Time Dependent Multidimensional Gas Dynamics",

Journal of Computational Physics, Vol. 63, pp. 458-476, 1986.

84 Osher, S., and Solomon, F., "Upwind Schemes for Hyperbolic Systems of Conservation Laws", Mathematics of

Computation, Vol. 38, No. 158, pp. 339-377, 1982.

85 Toro, E. F., Riemann Solvers and Numerical Methods for Fluid Dynamics, 2nd Edition, Springer, 1999.
 87

Df = ∫ Γϕ gradϕ. ds ≈ Γϕ (gradϕ)∗f . sf
Sf
ϕn − ϕp (grad ϕ)f . 𝐝𝐟 𝐝𝐟
(gradϕ)∗f = (grade ϕ)f + [ − ]
|𝐝𝐟 | |𝐝𝐟 | |𝐝𝐟 |
Eq. 4.25
The second term on the right
hand side (term in brackets)
represents the difference
between the central difference
approximation of the derivative
in the direction of vector df and
the corresponding value
obtained by interpolating cell-
center gradients. This
correction term detects and
smoothed out any unphysical
oscillations that might occur in
the iteration process86. As the
results one gets: Figure 4.6 An Arbitrary Polyhedral Control Volume

𝐝𝐟 . 𝐝𝐬 𝐝𝐟 . 𝐝𝐬
Df ≈ Γϕ (ϕN − ϕp ) + Γϕ [ (gradϕ)f . sf − (gradϕ)f . 𝐝𝐟 ]
𝐝𝐟 . 𝐝𝐟 𝐝𝐟 . 𝐝𝐟
Eq. 4.26
The first part of this term is treated implicitly, and the rest explicitly. Similarly, [Jasak]87 in his PhD
thesis and in less details in the subsequent publication gives the geometric interpretation to Eq. 4.25,
and Eq. 4.26 naming them the minimum correction, orthogonal correction, and over-relaxed
approach, respectively (Figure 4.7). He approximates the diffusive flux as

(ϕN − ϕp )
Df ≈ Γϕ (gradϕ)∗f . sf = Γϕ [ |∆𝐟| + (gradϕ)f . 𝐤 𝐟 ]
|𝐝𝐟 |
Eq. 4.27
where vectors Δf and kf satisfy the following condition

𝐤 𝐟 = 𝐬𝐟 − ∆𝐟
Eq. 4.28
[Jasak] compared the performance of these three approaches and found out that the over-relaxed
approach is more robust and more efficient than the other two approaches, especially in case of
highly non-orthogonal meshes. This is expected because, unlike the other approaches which are
somewhat arbitrary, the over-relaxed approach comes as a result of a direct discretization of the
transport equations.

86 I. Demirdˇzi´c, “On the Discretization of Diffusion Term in the Finite-Volume Continuum Mechanics”, Numerical
Heat Transfer Fundamentals · July 2015.
87 H. Jasak, Error Analysis and Estimation for the Finite Volume Method with Applications for Fluid Flow, PhD

Thesis, University of London, 1996.

 88

𝐝𝐟 . 𝐬𝐟
Minimum Correction ∶ ∆𝐟 = .𝐝
𝐝𝐟 . 𝐝𝐟 𝐟
|𝐬𝐟 |
Orthgonal Correction ∶ ∆𝐟 = .𝐝
|𝐝𝐟 | 𝐟
𝐬𝐟 . 𝐬𝐟
Over − Relaxed Correction ∶ ∆𝐟 = .𝐝
𝐝𝐟 . 𝐬𝐟 𝐟
Eq. 4.29

Figure 4.7 Geometric Interpretation of the Diffusion Term Approximation –[H. Jasak]

4.9 Pressure Based (Incompressible) Schemes

Finding a suitable solution for incompressible NS equations are the central issue in the subject of
CFD, dating back to at least the 60's, when the MAC scheme88 and the Projection method89 were
invented. MAC scheme and Staggered Grid enjoyed immediate success and similar ideas were also
applied to electromagnetics. On the other hand, it wasn't until the late 80's and early 90's, did we
begin to understand the mysteries surrounding the projection method and realize its potential. By
now projection method is by far the most popular method in this field. The main difficulty in
numerically solving incompressible NS equations is the lack of proper evolutionary equation for
pressure. After all for incompressible flows, pressure does not carry its usual thermodynamic
meaning. It is there mainly as a Lagrange multiplier for the constraint of incompressibility. This
translates to a lack of proper boundary condition, when equations for pressure are derived. The

88 F.H. Harlow and J.E. Welch, Numerical calculation of time-dependent viscous incompressible flow of fluid
with free surface, Phys. Fluids, 1965,
89 A. J. Chorin, “On the convergence of discrete approximations of the Navier-Stokes equations”, Math. Comp. 1969.
 89

prime Finite Volume solution methods, are pressure or density based methods90. The pressure
based schemes are mainly developed for incompressible, low Reynolds number application
when pressure value is guessed to begin and updated using Poisson’s equation after solving the
momentum equations. The main advantage is the decoupling of the momentum and energy
equation when the flow variables could be determined in segregated fashion. Among notable
methods are SIMPLE (Semi-Implicit Methods for Pressure-Linked Equations) and PISO. The density
based methods are more rigorous due to
fact that governing equations are solved
in coupled environment where pressure
is obtained through equation of state. In
both cases, additional scalar equations
are solved in a segregated style. Most
commercial CFD vendors try to provide
both methods91.
With segregated methods an equation for
a certain variable is solved for all cells,
and then the equation for the next
variable is solved for all cells, etc. With
coupled methods, for a given cell
equations for all variables are solved, and
that process is then repeated for all
cells92. The segregated solution method
is the default method in most commercial
finite volume codes. It is best suited for
incompressible flows or compressible
flows at low Mach number.
Compressible flows at high Mach
number, especially when they involve
shock waves, are best solved with the Figure 4.8 Accessible Pressure Solvers in Fluent
coupled solver. Here we are focused here
in incompressible (M < 0.3). Equation
system to be solved is the continuity equation together with the momentum equations, while the
continuity equation is considered as a side condition, demanding a divergence-free flow field.
Pressure does not appear in the continuity equation of continuity and equations of momentum and
continuity need to be coupled (see Figure 4.8).
4.9.1 Methods Based on the Vorticity Equation
By taking the curl of viscous incompressible flow, the pressure term vanishes. As for drawbacks it is
limited to 2D applications, but revised 3D approaches are available.
4.9.2 Methods Based on Artificial (Pseudo) Compressibility
The pseudo-compressibility method is a preconditioning technique of overcoming the numerical
issues related to the uncoupling of pressure and velocity field in the incompressible equations

90 Weinan E, “Numerical Methods for Viscous Incompressible Flows: Some Recent Advances”, Department of
Mathematics and Program in Applied and Computational Mathematics, Princeton University, Princeton, NJ.
91 Introduction to ANSYS Fluent, 2010.
92 Georgia Tech Computational Fluid Dynamics, “Solution Methods for Navier Stokes Equations”, Spring 2007.
 90

[Chorin]93-94. The governing equations are made artificially hyperbolic by adding a density time-
derivative to the continuity equation95. Based on physical analogies, the time derivative of density is
then linked to the pressure by introducing an artificial compressibility factor. A number of possible
approaches are listed below:
∇. uε + εpε = 0
∇. uε + εpεt = 0
∇. uε − ε∆pε = 0
∇. uε − ε∆pεt = 0
Eq. 4.30
The first three versions are well known while the last one was recently introduced in 96. We will refer
them as a class of pseudo-compressibility methods for the unsteady incompressible Navier-Stokes
equations. The aim of this paper is to present some recent analyses on this class of pseudo-
compressibility methods as the perturbation parameter ε → 0. An important aspect of this work is
the error analysis of time discretization schemes (they can be employed, in principle, with any
consistent space discretization) associated with the pseudo-compressibility methods97. As an
example investigated by [Nguyen & Park]98, the topic that is still receiving increasing attention,
particularly for applications in many hydraulic and hydrodynamic problems.
4.9.3 Methods Based on Pressure Iterations or Pressure Correction
These are refer to number of methods as:
1. Projection Methods [Chorin & Temam, 1968]
The projection method is an effective means of numerically solving time-dependent
incompressible fluid-flow problems. It was originally introduced by Alexandre Chorin
in 1967 as an efficient means of solving the incompressible Navier-Stokes equations.
The key advantage of the projection method is that the computations of the velocity and
the pressure fields are decoupled99.
• Fractional Step Method [Kim & Moin, 1975]
The method is based on a fractional step, or time-splitting, scheme in
conjunction with the approximate factorization technique.
2. Marker and Cell Method – MAC [Harlow & Welch, 1965]
A technique is described for the numerical investigation of the time‐dependent flow of
an incompressible fluid, with limited capability. The primary dependent variables are
the pressure and the velocity components. Also used is a set of marker particles which
move with the fluid. These called the Marker And Cell (MAC) method.

93 Alexandre Joel Chorin, “Numerical Solution of the Navier-Stokes Equations”, AEC Computing and Applied
Mathematics Center, Courant Institute of Mathematical Sciences.
94 Alexandre Joel Chorin, “The Numerical Solution of the Navier-Stokes Equations for an Incompressible Fluid”,

supported by the U. S. Atomic Energy Commission, Contract No. AT(30-1)-1480.

95 Michele Ferlauto, “A Pseudo-Compressibility Method For Solving Inverse Problems Based On The 3D

Incompressible Euler Equations”, Taylor & Francis in Inverse Problems in Science and Engineering ,2014.
96 “A new pseudo-compressibility method for the Navier-Stokes equations”, SIAM J. Math. Anal, 1994.
97 Jie Shen, “Pseudo-Compressibility Methods for the Unsteady Incompressible Navier-Stokes Equations”,

Department of Mathematics, Penn State University, USA.

98 Van-Tu Nguyen and Warn-Gyu Park, “A free surface flow solver for complex three-dimensional water impact

problems based on the VOF method”, International Journal For Numerical Methods In Fluids Int. J. Numerical
Meth. Fluids -2015.
99 Wikipedia.
 91

3. SIMPLE, SIMPLER, SIMPLEST, PISO, [Patankar and Spalding, 1972, 1981]. See below.
4. Fast Fluid Dynamic [Stam, 1999]. See below.
While 1 & 2 are scarcely used in CFD, the third item (3), known as pressure corrections methods,
are used extensively. The basic idea is to make use of a Poisson equation for the pressure; as pressure
appears only in the momentum equations in form of a partial derivative of first order. This can be
achieved by computing the derivative ∂P/∂xi such that the result yields a divergence free flow field.
An adequate iterative method is formulated that iterates the pressure until the conservation of mass
and momentum is obtained. These methods are also known as projection methods. This is celebrated
Poisson Equation. The RHS of the equation is a function of the partial derivatives of the velocity
components and is obtained from the convective part of the momentum equations. All other terms
i.e. the time derivative and the local change of the molecular transport (i.e. the diffusive type terms)
are removed because of the continuity side condition100. The basic procedure can be visualized as
Computation of a velocity field from the solution of the momentum equations with an initial
(guessed) or no pressure field.
1. Computation of the pressure from the Poisson equation with the previously computed
velocity field.
2. Correction of the velocity field with the “new” pressure.

2
∂ui ∂uj ∂(ui uj ) 1 ∂p μ ∂uj
=0 , + =− + ( )
∂xi ∂t ∂xi ρ ∂xj ρ ∂xi
2
∂ ∂uj ∂(ui uj ) ∂ 1 ∂p μ ∂uj
Take Div. of Momentum → [ + ]= [− + ( ) ]
∂xj ∂t ∂xi ∂xj ρ ∂xj ρ ∂xi
∂ ∂uj ∂ ∂(ui uj ) 1 ∂2 p μ ∂ ∂uj ∂uj
( )+ [ ]= − + [ ][ ]
∂t ⏟∂xj ∂xj ∂xi ρ ∂xj 2 ρ ∂xi ∂xi ⏟ ∂xj
0 0
2
∂ p ∂ ∂(ui uj )
2
= − ρ [ ]
∂x
⏟ j ⏟∂xj ∂xi
∇2 p f(ui )
Eq. 4.31
Additional detailed information regarding this can be obtained from original classic paper by
[Patankar & Spalding]101, as well as excellent discussion regarding pressure-velocity coupling
(SIMPLE, SIMPLER, SIMPLEC, PISO) algorithms by [Versteeg and Malalasekera]102. There is also an
investigation by [Rhie, & Chow]103 to demonstrate the pressure scheme of the 2D incompressible,
steady N-S equations for flow past an airfoil. This method is applied to the turbulent flows over
airfoils with and without trailing edge separation, with the aid of k-ε modeling of the turbulent flow.

100 Georgia Tech Computational Fluid Dynamics, “Solution Methods for Navier Stokes Equations”, Spring
2007.
101 S. V. Patankar and D. B. Spalding,”Calculation Procedure For Heat, Mass And Momentum Transfer In Three-

Dimensional Parabolic Flows”, International Journal Of Hear Mass Transfer, Vol. Is, Pp. 1787-1806, 1972.
102 H K Versteeg and W Malalasekera, “An Introduction To Computational Fluid Dynamics - The Finite Volume

Method”, Second Edition, Pearson Education Limited 1995, 2007

103 C. M. Rhie, W. L. Chow, “Numerical Study of the Turbulent Flow Past An Airfoil With Trailing Edge Separation”,

AIAA Journal, Vol. 21, No. 11, November 1983.

 92

Instead of the staggered grid, an ordinary grid system is employed and a specific scheme is developed
to suppress the pressure oscillations.
4.9.3.1 PISO Algorithm
PISO algorithm (Pressure-Implicit with Splitting of Operators) was proposed by [Issa] in 1986
without iterations and with large time steps and a lesser computing effort104. It is an extension of
the SIMPLE algorithm. PISO is a pressure-velocity calculation procedure for the Navier-Stokes
equations developed originally for non-
iterative computation of unsteady
compressible flow, but it has been adapted
successfully to steady-state problems. PISO
involves one predictor step and two corrector
steps and is designed to satisfy mass
conservation using predictor-corrector steps.
The PISO algorithm with neighbor correction
is highly recommended for all transient flow
calculations, especially when you want to use
a large time step. (For problems that use the
LES turbulence model, which usually requires
small time steps, using PISO may result in
increased computational expense, so SIMPLE
or SIMPLEC should be considered instead).
PISO can maintain a stable calculation with a
larger time step and an under-relaxation
factor of 1.0 for both momentum and
pressure. For steady-state problems, PISO
with neighbor correction does not provide
any noticeable advantage over SIMPLE or
Figure 4.9 PISO algorithm flow chart (Courtesy of
SIMPLEC with optimal under-relaxation
Giannopapa, and G. Papadakis )
factors.
PISO with skewness correction is
recommended for both steady-state and transient calculations on meshes with a high degree of
distortion. When you use PISO neighbor correction, under-relaxation factors of 1.0 or near 1.0 are
recommended for all equations. If you use just the PISO skewness correction for highly-distorted
meshes (without neighbor correction), set the under relaxation factors for momentum and pressure
so that they sum to 1 (e.g., 0.3 for pressure and 0.7 for momentum)105. A PISO algorithm flow chart
can be appreciated in Figure 4.9. The method is not restricted to using the PISO algorithm, e.g. the
Simple algorithm can be used as well106.
4.9.3.2 Case Study - Second Order Non-Iterative PISO-Algorithm vs Iterative PISO
A new variant of the non-iterative PISO-algorithm for the solution of implicitly discretized fluid flow
equations is proposed and investigated by [Tukovic et al.]107. The governing equations are
discretized in space using a second-order accurate finite volume method. A second-order accurate
three-level implicit temporal discretization scheme is used to discretize the momentum equation,

104 Wikipedia.
105 Fluent User Guide.
106 C. G. Giannopapa, and G. Papadakis, “Indicative Results and Progress on the Development of The Unified Single

Solution Method for Fluid-Structure Interaction Problems”, DOI: 10.1115/PVP2007-26420, 2007.

107 Željko Tukovi ´c a , Milovan Peri´c , Hrvoje Jasak, “Consistent second-order time-accurate non-iterative PISO-

algorithm”, Computers and Fluids, 2018.

 93

where the non-linear convection term is linearized using a second- order explicit approximation of
mass flux at the new time level, based on the results from previous two time steps. In order to ensure
temporal consistency on collocated meshes, a consistent Rhie-Chow interpolation for the
computation of mass fluxes is used, which is here extended to moving meshes. The proposed non-
iterative PISO-algorithm is tested on three fixed and moving mesh test cases, demonstrating its
second order accuracy in time.
4.9.3.2.1 Introduction
The non-iterative nature of the algorithm implies that momentum equation is discretized and solved
only once in the predictor step at the beginning of time step calculation. Predictor step is followed by
two or more corrector steps, where the conservation of mass and momentum is enforced by the
solution of pressure equation and explicit momentum correction. In the original PISO-algorithm,
temporal discretization is per- formed using first-order accurate (two-level) Euler implicit scheme
and the overall temporal accuracy is first order. It is emphasized in that the application of second-
order accurate three-level implicit temporal discretization scheme does not yield second-order
temporal accuracy, since in the linearized convection term in the momentum equation mass flux from
the previous time step is used. The second-order accuracy in time can be achieved by using outer
iteration loop enclosing the non-iterative PISO-algorithm. The mass flux in the linearized convection
terms stems then from the previous iteration (in the first iteration, it stems from the previous time
step). Such an iterative PISO-algorithm does not have substantial advantages when compared to
other iterative algorithms. [Park]108 proposed a non-iterative PISO-algorithm of second-order
temporal accuracy, where the non-linear convection term in the momentum equation is advanced in
time using an explicit temporal discretization scheme of higher order. He proposed using the second-
order Adams–Bashforth or the third-order Runge–Kutta scheme. The disadvantage of computing
convection flux fully explicitly is the limit on time-step size due to stability constraints of the explicit
scheme. In this study we propose an alternative approach to achieve second-order temporal accuracy
of the non-iterative PISO-algorithm. Only the explicit part of the linearized convection flux is
approximated at the new time level using a second-order extrapolation from two previous time steps.
This is equivalent to using the second-order Adams–Bashforth scheme to advance the mass flux in
time within the predictor stage of the PISO-algorithm. Owing to the fact that remaining part of the
convection flux is treated implicitly, the stability limit is increased.
4.9.3.2.2 Preliminary Results
The proposed second-order non-iterative PISO-algorithm (here labelled as ePISO) is validated on
flow around fixed and oscillating cylinder in a channel. The same test case are also computed using
the second-order iterative version of the PISO-algorithm (here labelled iPISO) and the non-iterative
version equivalent to ePISO, except that
explicit terms are computed using values
from the previous time step (here labelled
nPISO). Figure 4.10 displays inlet part of
the spatial domain, which is discretized
with 7126 quadrilateral finite volumes.
The Reynolds number based on the
average inlet velocity and cylinder
diameter is Re = 100. At the channel inlet, Figure 4.10 Flow Around Circular Cylinder in a
parabolic velocity profile is specified with Channel
the average velocity u = 1m/s. Calculation is first performed using the iterative PISO-algorithm and
the time step size Δt = 0 . 02s (corresponding to the maximal Courant number around 1.2) until a

108 Park TS . “Effects of time-integration method in a

large-eddy simulation using the PISO algorithm: Part I – Flow
field”. Numerical Heat Transfer Part A 1999;50:229–45 .
 94

periodic response of the lift and drag force is obtained (t = 50s). Afterwards, calculation is continued
from the time instance t = 50 s using the three PISO-algorithms named in the previous test case: iPISO,
nPISO and ePISO. In such a way all algorithms start with the same initial state; the time step is also
kept the same.
Figure 4.11 shows lift force coefficient as a function of time, calculated using different variants of
the PISO-algorithm. The solution obtained using iterative PISO-algorithm (iPISO) can be considered
as the most accurate for a given time step size. One can notice significant departure of the solution
obtained using nPISO, while the present non-iterative PISO-algorithm (ePISO) gives solution with a
negligible difference compared to the solution obtained using the iterative PISO-algorithm. It should
be also noted that numerical results obtained in this study (frequency and amplitude of lift
coefficient) agree well.

Figure 4.11 Lift Force as a Function of Time, using Different PISO-Algorithms for the Flow Around
Fixed Cylinder in a Channel (Courtesy of Tukovic et al.)

4.9.4 Fast Fluid Dynamic (FFD)

As the naming specifies, Fast fluid Dynamic is a fast (50 times by some estimate) method to solve the
incompressible NS equation. It was purposed originally by (Stam)109 for computer graphics and
animation. In recent years it has been also used for indoor airflow simulation. The model would not
be accurate enough for most engineering applications. Indeed, it suffers from too much “numerical
dissipation”, i.e., the flow tends to dampen too rapidly as compared to actual experiments. In a
computer graphical application, on the other hand, this is not so bad, especially in an interactive
system where the flow is “kept alive” by an animator applying external forces. It applies a time-
splitting technique to separate the Navier-Stokes equations into several simple equations. Then those
spitted equations are solved one by one. Thus, FFD can solve the Navier-Stoke equations without a

Add Force Advect Diffuse Project

Figure 4.12 Solution Method for FFD

109 Jos Stam, “Stable Fluids”.

 95

trial correction iterations used in CFD. This is the reason why FFD is much faster than CFD. However,
current FFD model applied many low-order schemes, so its accuracy is poorer than CFD. It splits the
momentum equation over time in 4 steps, (see Figure 4.12).
Because of a linear interpolation used in a semi-Lagrangian approach for advection equation, the FFD
has a significant numerical diffusion. To reduce the numerical diffusion in the FFD model,
improvements have been purposed in two fronts. One is applying a high-order interpolation scheme
in the semi-Lagrangian solver110. The other improvement was to solve the advection equation using
the convectional method used in CFD, such as the Lax-Wendroff scheme and the QUICK. The first
approach is the use of Fast Fluid Dynamics (FFD) that is an intermediate model between the nodal
and CFD models. The FFD, developed for computer flow visualization, can efficiently solve the
incompressible Navier-Stokes equations (top), energy equation (middle) and species transport
equations (bottom):

u u 1 P  2u 1 
= -u − + ν 2 + SF 
t x ρ x x ρ 
T T  2T 
    2
= -u + α 2 + ST  = -u +k 2 +S+G
t x x  t x x
c c  c
2

= -u + kc + S 
t x x 2
c


Eq. 4.32
where S is the source term and G is the pressure term. The FFD method applies a time-splitting
method (Ferziger and Peric 2002) to solve the governing equations (Eq. 4.32). The purpose of the
splitting method is to divide a complex problem (equation) into several simple ones [Ferziger and
Peric 2002; John 1982; Levi and Peyroutet 2001] since solving these simple equations is
mathematically easy and numerically fast111. Then solutions of these simple equations can be
integrated into an approximated solution for the complex equation. The splitted equations in the FFD
are as follows:

∂φ ∂φ ∂2 φ
= −𝐮 +k 2 +S+G
∂t ∂x ∂x
φ(1) − φ(n) φ(2) − φ(1) ∂2 φ(2)
= ⏟ S , =k (2)
,
∆t Source
∆t ⏟ ∂x
Diffusion
φ(3) − φ(2) ∂2 φ(2) φ(n+1) − φ(3)
=𝐮 and Finally = ⏟
G
∆t ⏟ ∂x (2) ∆t Projection (Preesure)
Advection
Eq. 4.33
where superscripts (1), (2), and (3) represent temporary variables. The FFD computes sequentially
the above four equations. First source is added through equation. Then the FFD calculates diffusion
equation by using a first order implicit scheme. After that, advection equation is solved with a semi-

110 Wangda Zuo, Qingyan Chen, “Improvements On The Fast Fluid Dynamic Model For Indoor Airflow
Simulation”, Fourth National Conference of IBPSA-USA, New York City, New York, August 11 – 13, 2010.
111 Wangda Zuo, Qingyan Chen, “Simulations of Air Distributions In Buildings By FFD On GPU”, HVAC&R Research.
 96

Lagrangian solver. For the momentum equation, the FFD solves pressure equation together with
continuity equation by using a pressure-correction projection method [Chorin ]112. It is worth to
notice that there is an extra projection step before the advection step in the implemented FFD code,
which is to provide a divergence-free velocity field for the semi-Lagrangian solver in the advection
equation.

4.10 Density Based (Compressible) Schemes

The system of governing equations for a single-component fluid, written to describe the mean how
properties113, is cast in integral Cartesian form for an arbitrary control volume V with differential
surface area dA as follow


t V
WdV +  [F − G ].dA =  HdV W, F and G are defined as :
V

ρ  ρv i  0 
ρu     
  ρv i u + p î  τ xi 
    Eq. 4.34
W = ρv  , F = ρv i v + pĵ  , G = τ yi  i, j = 1,3
ρw     
  ρv i w + p k̂   τ zi 
ρE  ρv i E + pv  τ ijv j + q 
 

and the vector H contains source terms such as body forces and energy sources. Here ρ, v, E, and p
are the density, velocity, total energy per unit mass, and pressure of the uid, respectively. τ is the
viscous stress tensor, and q is the heat flux. Total energy E is related to the total enthalpy H by

2
V
E = H − P/ρ and H=h+ Eq. 4.35
2

The implicit-time stepping method (also known as dual-time formulation) is available in the density-
based explicit and implicit formulation. When performing unsteady simulations with implicit-time
stepping (dual-time stepping), we use a low Mach number time-derivative unsteady preconditioner
to provide accurate results both for pure convective processes (e.g., simulating unsteady turbulence)
and for acoustic processes (e.g., simulating wave propagation). Here we introduce a preconditioned
pseudo-time-derivative term into Eq. 4.34 as

112 Chorin, A.J. (1967) "A numerical method for solving incompressible viscous flow problems," J. of Comp. Physics.
113 “Fluent Guide, Using the Solver”, 2006.
 97

∂ ∂W ∂
∫ WdV + ∫ Q dV + ∮[F − G].dA = ∫ H dV
∂t V ∂Q ∂τ V
⏟ V
Γ
ρP 0 0 0 ρT
ρP u ρ 0 0 ρT u
∂W
where Q = [p,u,T]T , =Γ= ρP v 0 ρ 0 ρT v
∂Q
ρP w 0 0 ρ ρT w
[ρP H − δ ρu ρv ρw ρT H + ρCP ]
∂ρ ∂ρ
and ρP = | , ρT = | with δ = 1 ideal gas = 0 incompressible
∂P T ∂T P
Eq. 4.36
Where t denotes physical-time and τ is a pseudo-time used in the time-marching procedure. Note
that as τ⇾1, the second term on the left side of Eq. 4.36 vanishes and Eq. 4.34 is recovered. The
time-dependent term in Eq. 4.36 is discretized in an implicit fashion by means of either a first or
second-order accurate, backward difference in time. Here, we used a three point (2nd order)
Backward Differencing scheme of [ Jameson]114. The dual-time formulation is written in semi-
discrete form as follows:

Γ ε0 ∂W 1 1
[ + ] ΔQk+1 + ∮[F − G] .dA = H − (ε0 W k − ε1 W n + ε2 W n−1 )
Δτ Δt ∂Q V Δt
Eq. 4.37
Where (ε0 = ε1 = 1/2; ε2 = 0) gives first-order time accuracy, and (ε0 = 3/2; ε1 = 2; ε2 =1/2) gives
second-order. The k is the inner iteration counter and n represents any given physical-time level.
The pseudo-time-derivative is driven to zero at each physical time level by a series of inner iterations
using either the implicit or explicit time-marching algorithm. Many details of density schemes will
be discussed later.
4.10.1 Case Study 1 - Density Based Solver for Turbulent Compressible Flows
The development of coupled implicit density based solver for compressible flows are considered by
JF¨urst]115. The main flow field variables are updated using lower-upper symmetric Gauss-Seidel
method (LU-SGS) whereas the turbulence model variables are updated using implicit Euler method.
The equations can be expressed in the integral form suitable for finite volume approximation as
follows (assuming fixed control volume Ω)

d

dt Ω
WdΩ +  (F c − F v ) dS = 0
Ω
Eq. 4.38

where W = [ρ, ρU, ρE] is the vector of conservative variables and Fc and Fv represent inviscid and
viscous fluxes. In the case of turbulent flow the above mentioned system of equations is coupled to

114 Jameson, A., "Time Dependent Calculations Using Multigrid, with Applications to Unsteady Flows Past Airfoils
and Wings," AIAA Paper 91-1596, June 1991.
115 Jiˇr´ı F¨ursta, “On the implicit density based OpenFOAM solver for turbulent compressible flows”, EPJ Web of

Conferences 143, 02027- (2017).

 98

an additional turbulence model (e.g. a two-equation SST model116) which provides components of
the Reynolds stress tensor and the turbulent heat flux. For detail description special and temporal
discretization, please consult the [F¨urst]117.
4.10.1.1 Result for Subsonic and Transonic Flow over a Bump
The two-dimensional subsonic flow over a 10% circular bump has been chosen as a first benchmark.
The inviscid flow passes through a channel of height H = 1m and length L = 3 m. The total pressure at
the inlet was pT,in = 100 kPa, the total temperature TT,in =293.15K and the inlet flow direction was
parallel to x - axis. The average outlet pressure corresponds to isentropic Mach number M = 0.1. The
solution was obtained using a structured mesh with 150×50 quadrilateral cells. The second
benchmark is a transonic flow over the bump using the same geometry as in the previous case. The
same boundary conditions were used at the inlet and the average outlet pressure was set to a value
corresponding to isentropic M = 0.675. In this case the flow accelerates over the bump and reaches

Subsonic Transonic

Figure 4.13 Mach number distribution for flow over a Bump

supersonic speed. Then it passes through a shock wave and continues towards the outlet. One can
see that all methods including the transonic variant of SIMPLE solver give similar results.
Nevertheless the resolution of the shock wave is much better with the AUSM or HLLC based LUSGS
scheme than with the SIMPLE solver which spreads the shock over 4 cell due to higher amount of
numerical viscosity. (see Figure 4.13).

4.10.2 Case Study 2 - Transonic Flow Through a Turbine Cascade

The next benchmark is two-dimensional compressible turbulent flow through test turbine cascade
SE-1050. The geometry and boundary conditions correspond to the test from database. The inlet total
pressure and total temperature were ptot,in = 100 kPa and Ttot,in = 293.15K and the inlet angle was αin
= 19.3◦. The average outlet pressure corresponds to isentropic outlet Mach number Mout = 1.2, the
Reynolds number related to outlet flow conditions and blade pitch was Re = 1.2 ×106 and the inlet
turbulence intensity was Tuin = 2%. The hybrid mesh was composed of structured quadrilateral layer
around the blade and unstructured triangular part in the rest of the domain. The mesh contained
approximately 66×103 cells and the near-wall refinement corresponded to y+ ≈ 0.2. Figure 4.15
(left) shows contours of the Mach number in two periods of the mesh obtained using LU-SGS scheme
with HLLC flux. One can see here both branches of the shock emanating from the trailing edge as well
as the interaction of the inner branch of the shock wave with the blade. One can see that both
numerical methods shows very good agreement experimental data.

116 MENTER, F. R. “Two-equation eddy-viscosity turbulence models for engineering applications”. AIAA Journal
32.8 (1994), pp. 1598–1605.
117 same as 109.
 99

Figure 4.15 2D Mach Number Distribution in the Test Turbine Cascade (left) - 3D Pressure Distribution
at the Blades (right)

The next case is the 3D flow through a test turbine cascade with prismatic blades. The geometry of
blades correspond to the previous case and the blade has finite span. The flow regime corresponds
to outlet isentropic Mach number Mout = 1.2 and Reynolds number related to chord length Re =
1.34×106. Inlet conditions include a simple model of boundary layer corresponding to conditions in
the experiment. The numerical solution was obtained on an unstructured mesh with 3.3×106 cells
with near wall refinement to y+≈ 0.2. The Figure 4.15 (right) shows the distribution of the pressure
at the blades and the side wall and the distribution of entropy in the test plane. The regions with
higher level of entropy show re-
distribution of the energy losses in the
vicinity of side walls.
4.10.3 Case Study 3 - Numerical
Study of Compressible Lid
Driven Cavity Flow
Lid-driven cavity flow of Newtonian
fluids is one of the most well-known
problems in CFD literature due to its
peculiar challenges in the form of
singularities in spite of its simple
geometry. In addition, the availability
of both analytical solutions and
experimental results for the lid-driven
cavity flow field has enabled Figure 4.14 Streamlines of the Base Flow in the Symmetry
researchers to test and improve their Plane z = 0.5 and Re = 1000
computational methods through this
 100

benchmark geometry. Recently, [Hussain]118 is investigated a two-dimensional (2D), mathematical

model is adopted to investigate the development of circulation patterns for compressible, laminar,
and shear driven flow inside a rectangular cavity. And the results of a commercial code (Fluent) is
compared with in-house code, as well as the vorticity-stream function formulation of the 2D
incompressible N-S equations [Ghia et al.]119, using a pressure based coupled algorithm. The
simulations are carried out for the unsteady, lid driven cavity flow problem with moving boundary
(bottom) for different Reynolds number (Re = 100, 400, 1000), Mach numbers (M = 0.5), bottom
velocities and high initial pressure and temperature. Similar studies has been conducted by
[Hosseinzadeh, et al.]120. A snap shot of the results provided in Figure 4.14 and Figure 4.16.
Another study done by [Giannetti]121 present the results of a linear stability analysis applied to an
incompressible flow in a 3D lid-driven cavity. The principal goal is the evaluation of the critical
value of the Reynolds number and the main characteristics of the flow instability.

Figure 4.16 Distribution of Velocities (u, v) along Centerline Horizontal Distance for Re = 100

4.10.4 Case Study 4 - 3D Structured/Unstructured Hybrid Navier-Stokes Method for Turbine Blade
Rows
A 3D viscous structured/unstructured hybrid scheme has been developed for numerical computation
of high Reynolds number turbomachinery flows, see [Tsung et al.]122. The procedure allows an
efficient structured solver to be employed in the densely clustered, high aspect-ratio grid around the
viscous regions near solid surfaces, while employing an unstructured solver elsewhere in the flow
domain to add flexibility in mesh generation. Test results for an inviscid flow over an external

118 Amer Hussain, “A Numerical Study of Compressible Lid Driven Cavity Flow with a Moving Boundary”, Thesis
Submitted to the Graduate Faculty of the University of New Orleans in partial fulfillment of the requirements
for the degree of Master of Science in Engineering Mechanical, 2016.
119 Ghia, U., Ghia, K.N., and Shin, C.T., 1982, “High-Re Solutions for Incompressible Flow Using the Navier-Stokes

Equations and a Multigrid Method,” Journal of Computational Physics, 48, pp. 387-411.
120 S. Hosseinzadeh, R. Ostadhossein, H.R. Mirshahvalad and J. Seraj, “Using Simpler Algorithm for Cavity Flow

Problem”, Mechatronics and Applications: An International Journal (MECHATROJ), Vol. 1, No.1, January 2017.
121 Flavio Giannetti, Paolo Luchini, Luca Marino, “Linear stability analysis of three-dimensional lid-driven cavity

flow”, Department of Mechanical Engineering, University of Salerno, Italy.

122 F.-L. Tsung, J. Loellbach, O. Kwon, and C. Hah, “A Three-Dimensional Structured/Unstructured Hybrid Navier-

Stokes Method for Turbine Blade Rows”, Prepared for the 30th Joint Propulsion Conference and Exhibit
cosponsored by AIAA, ASME, SAE, and ASEE Indianapolis, Indiana, June 27-29, 1994.
 101

transonic wing and a Navier-Stokes flow for an internal annular cascade are presented.
4.10.4.1 Introduction
In modern turbomachinery designs, the rotor and stator blade rows often possess extreme turning
angles such that the flow direction deviates greatly from the axial direction. These geometries
frequently cause difficulties in generating structured meshes123. In order to impose point-to-point
periodicity on periodic boundaries in the domain, the computational grids are typically forced into
highly skewed shapes that decrease the accuracy of the solvers. One remedy for reducing the grid
skewness is to forego the point to point correspondence on the periodic boundaries and to
interpolate solution values where necessary. The disadvantage of the non-periodic grid approach is
that the local density of grid lines on either side of a periodic boundary may be vastly different.
Hence, flow features resolved on one side of the domain boundary may be diffused or lost when
interpolated to a coarser grid on the other side. Another solution to grid skewness is to use
unstructured meshes.
However, the cost and technology of present day unstructured solvers still leave much room for
improvement. The computation of both inviscid and viscous flows on unstructured triangular meshes
in two dimensions, and tetrahedral meshes in three dimensions, has matured significantly in recent
years124-125. The ability of unstructured solvers to handle complex geometries has proven to be
valuable for many computations. However, the geometric flexibility of unstructured grid solvers is
also the source of their disadvantages when compared to structured solvers. Due to the naturally-
ordered data connectivity of structured grids, structured solvers require much less memory
(especially for implicit schemes), implicit structured solvers are straight forward to code, and
turbulence modeling is easier to implement.
The shortcomings of both structured and unstructured methodologies are continuously being
addressed and improved. In the meantime, however, it is possible to take advantage of the best
properties of both methodologies. In the present paper, a solution procedure which couples an
efficient structured solver with an unstructured solver is presented. The hybrid procedure takes
advantage of the computational efficiency of structured codes and, at the same time, is able to benefit
from the geometric flexibility of unstructured solvers. With the hybrid approach, densely packed
structured grids can be placed in the highly viscous regions near solid surfaces, and unstructured
grids can be used away from solid surfaces 126. This approach can avoid the severe grid skewness
commonly experienced by fully structured grids around turbine blades with high turning angles. The
procedure has various other applications as well, such as providing a means of connecting multi-body
geometries for Chimera-type grids to insure node-to-node correspondence127-128. The strategy of
coupling structured and unstructured methods has been implemented for two dimensional
turbomachinery computations by many researchers in the past129.

123 Aronone, A., Liou, M-S., and Povinelli, A., "Transonic Cascade Flow Calculations Using Non-Periodic C-type
Grids," CFD Symposium on Aero-propulsion, NASA Conference Publication 3078, 1990.
124 Barth, T.J., "A 3-D Upwind Euler Solver for Unstructured Meshes," AIAA Paper 91-1548, June 1991.
125 Batina, J.T., "A Fast Implicit Upwind Solution Algorithm for Three-Dimensional Unstructured Dynamic Meshes,"

AIAA Paper 92-0447, Jan. 1992.

126 Weatherill, N.P., "Mixed Structured-Unstructured Meshes for Aerodynamic Flow Simulation," Aeronautical

Journal, pp. 111-123, Apr. 1990.

127 Nakahashi, K. and Obayashi, S., "FDM-FEM Zonal Approach for Viscous Flow Computation Over Multiple-

Bodies," AIAA Paper 87-0604, Jan. 1987.

128 Kao, K.H. and Liou, M.S., "Direct Replacement of Arbitrary Grid-Overlapping by Non-Structured Grid," NASA

TM-106601, May 1994.

129 Nakahashi, K., Nozaki, O., Kikuchi, K-, and Tamura, A., "Navier-Stokes Computations of Two- and Three-

Dimensional Cascade Flow fields," Journal of Propulsion and Power, Vol. 5, No. 3, pp.320-326, 1989.
 102

The present work extends this strategy to three-dimensional turbomachinery flows. The hybrid
solution technique is first tested for an external fixed-wing transonic flow case and the result is
compared with a structured solution for validation. The code is then applied to an annular cascade
and the result is compared with experimental data.
4.10.4.2 Governing Equation Unstructured Solver
The time dependent, Reynolds averaged, compressible Navier-Stokes equations, which express the
conservation of mass, momentum, and energy, are solved. The turbulence viscosity is calculated
using the high Reynolds number turbulence model of [Launder and Spalding]130. The equations of
motion, written in an integral form for a bounded domain Ω with a boundary ∂Ω, are

∂
̂ ds = ∭ 𝐒 (𝐐)dV
̂ ds − ∬ 𝐆(𝐐). 𝐧
∭ 𝐐dV + ∬ 𝐅(𝐐). 𝐧
∂t
Ω ∂Ω ∂Ω Ω
Eq. 4.39
where Q is the unknown vector containing the conserved properties

Q = [ρ, ρu, ρv, ρw, e0 , ρk, ρε ] T

Eq. 4.40
In the above equation, ρ is the fluid density. u, v, and w are the Cartesian velocity components in x, y,
and z directions, respectively. eo is the total energy per unit volume. The turbulent kinetic energy and
turbulent kinetic energy dissipation rate are represented by k and ε. The vector F is the convective
flux term, G is the viscous term, and S is the source term containing the production and destruction
of turbulent kinetic energy. A complete description of the governing equations is presented in
[Reference]131 and will not be detailed here for brevity.
4.10.4.3 Inviscid Flux Spatial Discretization
The inviscid flux across each cell face x is computed using Roe's flux-difference splitting formula132
Here, QL and QR are the state variables to the left and right of the interface K. The matrix A is computed
by evaluating

1 ∂𝐅
̅ |(𝐐R − 𝐐L )}k
𝐅𝐤 = {𝐅(𝐐L ) − 𝐅(𝐐R ) − |𝐀 , ̅=
𝐀
2 ∂𝐐
Eq. 4.41
with Roe-averaged quantities so that

̅ [𝐐R − 𝐐L ]
𝐅(𝐐R ) − 𝐅(𝐐L ) = 𝐀
Eq. 4.42
is satisfied. For a first-order scheme, the primitive variables at each cell face are set equal to the cell-
centered averages on either side of the face. For a higher-order scheme, estimation of the state at
each face is achieved by interpolating the solution at each time step with a Taylor series expansion
in the neighborhood of each cell center. The cell-averaged solution gradients required at the cell
center for the expansions are computed using Gauss' theorem by evaluating the surface integral for

130 Launder, B.E. and Spalding, D.B., 'The Numerical Computation of Turbulent Flows," Computer Methods in
Applied Mechanics and Engineering, Vol. 3, 1974.
131 Kwon, O.J. and Hah, C., "Solution of the 3-D Navier-Stokes Equations with a Two-Equation Turbulence Model

on Unstructured Meshes Applied to Turbomachinery," AIAA Paper 94-1833, June 1994.

132 Roe, P.L., "Characteristic-Based Schemes for the Euler Equations," Annual Review of Fluid Mechanics, Vol. 18,

pp. 337-365, 1986.

 103

the closed surface of the tetrahedra. This process can be simplified using geometric invariant features
of tetrahedra. The resulting second-order formula for the flow state at each cell face can be written
as
1 1
𝐪𝐟𝟏,𝟐,𝟑 = qi + [ (q n1 + qn2 + qn3 ) − qn4 ]
4 3
Eq. 4.43
where the subscripts nl , n2 , n3 denote the nodes comprising face f 1,2,3 of cell i, and n4 corresponds
to the opposite node. The expansion also requires the nodal value of the solution, which can be
computed from the surrounding cell center data using a second order accurate pseudo-Laplacian
averaging procedure as suggested by [Homes and Connell]133. The three dimensional extension by
[Frink]134 is adopted in the present calculations. The convective terms of the turbulence equations
are calculated using a first-order accurate scheme in the present paper to reduce the computational
cost and to ensure the numerical stability of the time integration135.
4.10.4.4 Viscous Flux Spatial Discretization
The evaluation of the viscous term G requires first derivatives of the velocity, temperature, and k-E
values at the cell faces. They are achieved by first evaluating the gradient of each required flow
quantity at the cell center from the known primitive variables at each time step. The gradient of the
desired quantity is obtained by applying the gradient theorem,

1
∇𝐐n = ̂dS
∮ 𝐐𝐧
VΩ
∂Ω
Eq. 4.44
where Ω represents the volume of the domain over which the theorem is applied. The scalar quantity
¢ can be the three components of velocity, the temperature, or turbulence quantities. In the present
calculations, the integral domain is defined as the individual tetrahedral cell, and the surrounding
surface area as2 consists of the four triangular surfaces covering the cell. This formulation is
consistent with the numerical procedure of evaluating the convective fluxes of the present cell-
centered scheme. Once the gradients of the desired quantities are known at the cell center, nodal
values are calculated using the Pseudo-Laplacian averaging mentioned earlier for the convective
terms. The flux through each of the triangular faces in Eq. 4.44 is obtained by averaging the three
nodal values for the triangle. Once the gradients of the primitive variables are obtained, the shear
stresses and can be calculated, from which G is evaluated at the cell center. The nodal values of these
quantities are calculated once again by applying the Pseudo-Laplacian averaging of the surrounding
cell center values. The surface flux of these quantities in Eq. 4.39 is obtained by taking the average
of the three nodal values for each triangular face of each cell.
4.10.4.5 Time Integration
The solution vector is integrated in time using the implicit Euler method, which is first-order accurate
in time. The nonlinear inviscid flux vectors are linearized at every time level about their values at the
previous time level using Taylor expansions, e.g.

133 Holmes, D.G. and Connell, S.D., "Solution of the 2D Navier-Stokes Equations on Unstructured Adaptive Grids,"
AIAA Paper 89-1932, June 1989.
134 Frink, N.T., "Recent Progress Toward a Three-Dimensional Unstructured Navier-Stokes Flow Solver," AIAA

Paper 94-0061, Jan. 1994.

135 Mavriplis, D.J., "Multigrid Solution of Compressible Turbulent Flow on Unstructured Meshes Using a Two-

Equation Model," AIAA Paper 91-0237, Jan. 1991.

 104

n+1 n
∂𝐄 n
𝐄 =𝐄 + ∆𝐐n+1 + O(∆𝛕)2 = 𝐄 n + 𝐀𝑛 ∆𝐐n+1
∂𝐐
Eq. 4.45
The viscous terms are evaluated explicitly. Explicit treatment of the viscous terms still permits the
use of large time steps since the Reynolds numbers of interest here are fairly large. To further reduce
computational time and memory, the radial/spanwise flux derivatives are treated explicitly at the
old time level, but the new values are incorporated as soon as they become available. This explicit
treatment of the spanwise flux terms enables the scheme to solve the three-dimensional equations
by solving one spanwise station implicitly at a time. To eliminate any dependency the solution may
have on the spanwise marching direction, the solver reverses the marching direction with every
spanwise sweep. The resulting left-hand-side of the matrix equation is approximately factored into
alternating directions

( 𝐈 + ∆τδξ 𝐀n ) ( 𝐈 + ∆τδζ 𝐂 n )∆𝐐n+1 =

−∆τ [(𝐄 n − 𝐄vn,n+1 )ξ + (𝐅 − 𝐅vn,n+1 )η + (𝐆n − 𝐆vn,n+1 )ζ ]
Eq. 4.46
The implicit operating matrices are then diagonalized using a similarity transformation [27].
4.10.4.6 Spatial Discretization
Standard second-order central differencing is used for the spatial derivatives. Spectral-radius scaled
fourth/second-difference artificial dissipation136 is added for stability and to eliminate oscillations
near shocks. The viscous terms are evaluated using half-point central differencing so that the
computational stencil for the stress terms uses only three nodes in each of the three directions.
4.10.4.7 Hybrid Coupling
Since the structured code has multi-block capability, the coupling procedure treats the unstructured
portion of the hybrid meshes as a separate block. The blocks are solved independently with respect
to each other. Explicit boundary conditions are updated at the end of each time iteration. The two
codes are loosely coupled, essentially running independently of each other. The only interaction
between the structured solver and the unstructured solver is through boundary conditions. The
advantage of this formulation is that almost any two given codes can be coupled together. In this
paper, a central-differencing, finite-difference procedure and an upwind cell-centered finite volume
procedure are coupled. The structured solver is formulated in an inertial frame of reference whereas
the unstructured solver is in a blade-fixed coordinate system. For non-moving-boundary cases, such
as pressure-driven turbine stator flows, there is no difference between the two formulations.
However, for moving-boundary cases, care must be taken when transferring data between the two
solvers depending on how boundary conditions are prescribed. For example, consider a wing
traveling at a constant velocity.
Since the structured solver assumes the wing is moving and the unstructured solver assumes the
domain is fixed in space, the total energy must be converted between the two frames of reference. At
the end of each iteration, mass, momentum, and energy values are communicated between the
structured and unstructured solvers. Since the unstructured solver is a cell-centered finite-volume
scheme and the structured solver is a vertex-based finite-difference scheme, the unstructured and
the structured grids are overlap by one layer at the boundary in order to minimize interpolation and
extrapolation errors at the boundary. By overlapping the grids instead of abutting them against each
other, the boundary nodes for the structured solver are interpolated from the interior cell center

Jameson, A., Schmidt, W., and Turkel, E., "Numerical Solution of the Euler Equations by Finite-Volume
136

Methods Using Runge-Kutta Time-Stepping Schemes," ALAA Paper 81-1259, June 1981.
 105

values of the unstructured solver

Similarly, the cell boundary face
values of the unstructured solver
are interpolated from interior
structured nodes. An illustration
of the overlapped grid in 2D is
shown in Figure 4.17. Once the
grid construction is completed, a
table is generated to index
unstructured nodes to their
corresponding structured nodes,
and vise-versa. The index is
similar to the table generated
between structured block
boundaries for multi-block
calculations.
Figure 4.17 Structured-Unstructured Mesh Overlapping
4.10.4.8 Results and by One Layer – Courtesy of [Tsung et al.]
Discussion
Validation for both the
structured and the unstructured solvers have been documented previously and will not be presented
here. To test the structured-unstructured hybrid analysis, calculations for a transonic inviscid flow
over a fixed wing and a viscous flow for a turbine stator blade are presented. For all calculations in
this paper, a structured grid is first generated for the entire geometry. The unstructured grids are
generated by subdividing a portion of the structured cells each into six unstructured tetrahedra.
4.10.4.8.1 Transonic Wing
The first calculation selected for the hybrid scheme is a transonic wing at a freestream Mach number
of 0.8 and at 1.25 0 angle-of-attack. For this test case, flow periodicity is imposed on the two end wall
boundaries to simulate a two-dimensional flow. A purely structured solution for this case is first
obtained for comparison. An O-H grid topology is used; that is, an O-grid is generated at each
spanwise station. For the structured calculation, each spanwise station consists of 121 (streamwise)
x 41 (normal) points. Six spanwise stations are required due to the memory allocation algorithms for
the multi-block structured solver. The number of spanwise stations used does not alter the results
since the code simulates a two-dimensional problem.
For the structured-unstructured coupled procedure, the grid is divided into three zones. The inner
zone, which wraps around the solid surface (121x12), is calculated using the structured solver. The
middle zone (121x6), which wraps around the inner structured zone, is subdivided into tetrahedra
for the unstructured solver. The outer zone, which extends from the middle zone to the free stream,
is once again calculated using the structured solver. The purpose of dividing the calculation into
three zones is to simulate a Chimera type of grid where a simple background grid is generated for the
entire domain and a structured grid is used around a solid object. Instead of interpolating between
the two grids, an unstructured solver can be placed to connect the two set of grids and obtain node-
to-node correspondence which remove the needs for interpolation. For a finite volume approach, this
also insures flow conservation. This approach is also recommended for multi-body calculations. To
examine the coupling between the structured/unstructured solver across the three-zones, pressure
contour lines are shown in Figure 4.18 for the first-order and the second-order coupling, along with
the single block structured solver for comparison. For the first order coupling, a slight discontinuity
exists across the zonal boundary. However, the discontinuity is small and barely perceptible on a
global scale. For the second-order coupling, the contours show the flow is smooth with little
indication of the presence of the zonal boundaries. The mass flows in and out of the sandwiched
 106

unstructured zone are within 0.03% of each other, indicating that good flow conservation is achieved.
The unstructured shock is somewhat more compact then the structured solver. One reason for this
is that every structured cell becomes six tetrahedra so that in the streamwise direction, the
unstructured cells are twice as dense as the structured cells. The present hybrid procedure converges
considerably slower than the single structured grid due to the explicit formulation of the
incorporated unstructured solver, which requires a low CFL number.

Figure 4.18 Pressure contour lines across the structured-unstructured zonal interface – Courtesy of
[Tsung et al.]

4.10.4.8.2 3D Viscous Flow Through Turbine Blades-Hybrid Procedure

The second test case selected is a 3D annular cascade [29]. The geometry consists of an annular ring
of 36 turbine stator vanes. The blades are 38.10 mm in span, untwisted, and of constant profile with
an axial chord of 38.23 mm. The stator has a tip diameter of 508 mm and a hub-to-tip radius ratio of
0.85. The inlet flow angle is parallel to the axis of the cascade. The Reynolds number based on the
inlet total quantities and axial chord length is 898,650. The inlet total pressure and total temperature
at one axial-chord length upstream of the blade leading edge are known from experiment. The exit
hub static pressure to inlet total pressure ratio is 0.65 at 2.6 axial-chord lengths downstream of the
blade trailing edge.
For the hybrid calculation, the computational domain extends one axial-chord length upstream of the
blade leading edge and 2.6 axial-chord lengths downstream of the blade trailing edge. The exit hub
static pressure to inlet total pressure ratio is known from experiment. The grid has 105 (inlet to exit)
x29 (blade to blade) x15 (hub to tip) nodes and 23,000 cells are subdivided into unstructured cells.
The structured grid wraps around the solid surfaces of the hub and the turbine blade while the rest
of the domain is filled with unstructured cells. For the present grid, only 15 hub to shroud stations
are used and the grid density near the trailing edge is very coarse (only 3 points defining the rounded
trailing edge). While such a coarse grid is not acceptable for accurate prediction of viscous effects, it
is sufficient for monitoring the communication between the structured and unstructured zones and
to predict the overall flow field. A no-slip boundary condition is applied at the hub and on the blade
surfaces, and an inviscid slip condition is applied on the shroud. In the coupled calculation, the
 107

unstructured solver is run without turbulence

modeling because the effects of turbulence
are much reduced outside the boundary layer.
The pressure contour lines across the
structured and unstructured zones are
plotted in Figure 4.19.

4.11 Some Observations on Usage of

Pressure Based vs. Density Based
Schemes
Several researchers commented on CFD blogs
as where to use the Pressure or Density based
solver, and it was mentioned here as
confidence builder measure. Some
interesting ones are:
➢ Strictly speaking, the pressure-based
approach was developed for low-
speed incompressible flows, while the
density-based approach was mainly
used for high-speed compressible
flows. However, recently both Figure 4.19 Pressure Contour Across the
methods have been extended and Structured-Unstructured Interface – Courtesy of
reformulated to solve and operate for [Tsung et al.]
a wide range of flow conditions
beyond their traditional or original intent (see Figure 4.20). In both methods the velocity
field is obtained from the momentum equations. In the density-based approach, the
continuity equation is used to obtain the density field while the pressure field is determined
from the equation of state. On the other hand, in the pressure-based approach, the pressure
field is extracted by solving a pressure or pressure correction equation which is obtained by
manipulating continuity and momentum equations137.

Pressure or Density
Based (Hybrid)

Figure 4.20 Scope of Incompressible Vs. Compressible Flow Solvers

137 Johnson Emmanuel , CFD on line.

 108

➢ The pressure-based solver traditionally has been used for incompressible and mildly
compressible flows. The density-based approach, on the other hand, was originally designed
for high-speed compressible flows. Both approaches are now applicable to a broad range of
flows (from incompressible to highly compressible), but the origins of the density-based
formulation may give it an accuracy (i.e. shock resolution) advantage over the pressure-based
solver for high-speed compressible flows.
➢ The reason many people extend the simple algorithm for compressible flows is for using a
single unified solver which is valid over the entire flow regime. Many algorithms which are
developed for compressible flows are extended to incompressible flows through
preconditioning for the same reason. Also in certain flow scenarios, there may be wide
variations of the Mach number in the flow region and hence an unified solver is a better way
of handling it. [Harish, CFD online].
➢ According to [Mateu], the main difference is the approach for density variation.
the SIMPLE type algorithm seek for pressure correction with some assumption or
simplification. In other words, we first have an approximate velocity field, for this we
wish to correct it such that continuity satisfies. The velocity correction are expressed in
terms of pressure correction and thus we get pressure correction equation, that is solved
to get new pressure field. When the flow is compressible, the SIMPLE type method goes
little further and relate density change to pressure correction. Since flux is related to
density change, this could also be included into pressure correction equation. This way
you get pressure correction and you update velocity and density. Density based solvers
work on flux balances directly I suppose.

4.12 Residual Implication

In short, the non-linear NS equations are linearized and the accuracy of the linearized equations to
the non-linear ones is the residual where the linear solver solves the linear equations. The
discretization of the spatial terms in the governing equations results in a large coupled set of ordinary
differential equations of the form

d(VM W )
+ R(W) = 0 Eq. 4.47
dt
Where V represents the local control volume, M the mass matrix, and R(W) the spatially discretized
terms. For a vertex-based scheme, the mass matrix relates the average value of a control volume to
the values at the vertices of the mesh. Both V and M are constants for static meshes, and can therefore
be taken outside of the time derivative, and in absence of M it becomes

d𝐖
V + 𝐑(𝐖) = 0
dt
Eq. 4.48
For steady-state cases, these equations must be integrated in time towards t → ∞, where the time
derivatives become vanishingly small. Since time accuracy is not a concern in such cases, and each
equation may be advanced with the maximum permissible time step, as determined from local
stability considerations. There are two general methods of solving, as Explicit or Implicit scheme,
defined following.
 109

4.12.1 Explicit Schemes

A simple time integration scheme is obtained by replacing the time derivative by a simple forward
difference and evaluating the residual at the current time level:

W n +1 − W n
V + R(W n ) = 0 Eq. 4.49
Δt
This corresponds to a single-stage explicit scheme, since updates are obtained from one evaluation
of currently available quantities. Like Runge-Kutta schemes for ordinary differential equations,
multistage time-stepping schemes (which are required with higher-order discretization for stability
reasons) involve the combination of updates obtained at multiple explicitly evaluated intermediate
states138. The stage coefficients of these schemes can be optimized either to provide large time steps
at the expense of time accuracy or to enhance high-frequency damping properties of the scheme,
which is required in the context of a multigrid algorithm [Jameson et al]139 and [van Leer et
al]140.These schemes are extremely simple to implement and require little additional computer
storage. However, stability considerations restrict the maximum permissible time step to values
proportional to the local cell size. Thus, finer meshes lead to smaller time steps, which in turn lead to
larger solution times. For very fine meshes, the convergence of explicit schemes becomes
unacceptably slow, and more sophisticated solution strategies must be adopted.
4.12.2 Implicit Schemes
An implicit scheme is obtained by evaluating the spatial residual terms at the new time level n+1.
Since these quantities are not known explicitly, a linearization must be performed about the current
time level:
 V R 
 +  ΔW = −R( W )
n
Eq. 4.50
 Δt W 
Where ∂R/∂w represents the Jacobian and constitutes a large sparse matrix141. At each time step, the
above linear system must be solved for the corrections ∆w = wn+ 1 - wn from which the flow variables
can be updated. Implicit schemes may be classified by the degree to which the true Jacobian matrix
is approximated. If an exact linearization is employed, the method is unconditionally stable and
reduces to Newton’s method for ∆t → ∞. Although the quadratic convergence properties of Newton’s
method generally produce solutions in a very small number of time steps (often of the order of 10)
[Venkatakrishnan & Barth]142, [Barth & Linton]143, [Nielsen]144, each time step requires the inversion
of a large sparse matrix, which becomes prohibitively expensive in terms of storage and CPU-time for
fine meshes. A common simplification is to replace the exact linearization with one based on a first-

138 D. J. Mavriplis, “Unstructured Grid Techniques”, Annu. Rev. Fluid. Mech. 1997. 29:473–514.
139 Jameson A, Schmidt W, Turkel E.,” Numerical solution of the Euler equations by finite volume methods using
Runge-Kutta time stepping schemes”, AIAA Pap. 81-125- 1981.
140 van Leer B, Tai CH, Powell KG.,” Design of optimally-smoothing multi-stage schemes for the Euler equations”,

AIAA Pap. 89-1933,1989.

141 See 148.
142 Venkatkrishnan V, Barth TJ., “Application of direct solvers to unstructured meshes for the Euler and Navier-

Stokes equations using upwind schemes”, AIAA Pap. 89-0364, 1989.

143 Barth TJ, Linton SW.,”An unstructured mesh Newton solver for compressible fluid flow and its parallel

implementation”, AIAAPap. 95-022, 1995.

144 Nielsen EJ, Anderson WK, Walters RW, Keyes DE.,”Application of Newton-Krylov methodology to a three-

dimensional unstructured Euler code”, Proc. AIAA CFD Conf., 12th, San Diego. AIAA Pap. 95-1733-CP- 1995.
 110

order discretization [Mavriplis and Anderson]145. While this considerably reduces the storage
requirements of the linear system and improves its condition number, it precludes attaining
quadratic convergence rates, owing to the use of an inexact linearization. This in turn favors the use
of iterative methods to solve the linear system, which can be used to converge the system only
partially, because exact inversion of the Jacobian is no longer beneficial, owing to the mismatch
between Jacobian and residual. However, rigorous criteria for determining the optimal level of
convergence of the linear system at each time step have yet to be determined. Simple iterative
schemes such as Jacobi and Gauss-Seidel schemes have been utilized successfully in the context of
implicit schemes with first-order linearization. However, since these are still local techniques, their
convergence degrades with grid size. More sophisticated techniques such as preconditioned GMRES
methods provide enhanced convergence rates, particularly when applied with powerful
preconditioners such as Incomplete Lower Upper (ILU) factorization methods, which provide more
global information for the solution of the linear system.
Although 1st order linearization methods require substantially less memory overheads than exact
linearization methods, their memory requirements are still nontrivial. For example, on a 3D
tetrahedral grid, the first-order Jacobian of a vertex-based scheme requires on the order of 350
storage locations per vertex, over three times the number required for the implementation of an
explicit scheme. An ILU preconditioning strategy can require an equivalent additional amount of
memory. Since GMRES methods only require the evaluation of matrix vector products of the form
(∂R/∂w).∆w, it is possible to forgo the storage of the Jacobian, and evaluate the Jacobian vector
product directly by finite-difference techniques:

R R (W + ε ΔW ) − R ( W )
 ΔW = Eq. 4.51
W ε
Where ε represents a small
parameter. This requires multiple
residual evaluations (one for each
matrix vector product) and
represents a classic tradeoff
between storage and computation,
particularly for expensive
nonlinear residual constructions.
However, it also permits the use of
the more accurate second order
linearization in the implicit scheme,
by using the same second-order
residual in the finite difference
evaluation as in the right-hand side
of equation, as is done in the so-
called Newton-Krylov methods.
4.12.2.1 Convergence Rate and
Storage Requirement
Figure 4.21 compares the Figure 4.21 Comparing the Convergence Rates of the ILU-
convergence rates of a Newton- Preconditioned Newton-Krylov Method
Krylov method that employs ILU

145Anderson WK, Rausch R, Bonhaus D.,”Implicit Multigrid Algorithms for Incompressible Turbulent Flows On
Unstructured Grids”, Proc. AIAA CFD Conf., 12th, San Diego. AIAA Pap. 95-1740-CP, 1995.
 111

preconditioning with convergence of a Gauss-Seidel iterative method applied to a first-order

linearization, as well as with the convergence obtained by a multigrid scheme using the same Gauss-
Seidel solver as a multilevel smoother (Solid line: Gauss Siedel; Dotted line: ILU Preconditioned
Newton-Krylov; Dashed line: Multigrid method -using Gauss Siedel as smoother). The depicted
case involves the solution of two-dimensional subsonic inviscid flow over a NACA 0012 airfoil on a
mesh of 8578 points, using a vertex-based second-order upwind scheme146.
The quadratic convergence property of the Newton-Krylov approach is evident, which achieves a
very rapid asymptotic convergence rate in terms of number of iterations. When compared in terms
of overall CPU-time, the Newton-Krylov scheme, although still superior to the Gauss-Seidel implicit
method, is overtaken by the multigrid scheme, due to the expense of each Newton-Krylov iteration.
Additional improvements to Newton-Krylov methods can be achieved through continuation
techniques, which provide better initial guesses (for example by coarse to fine mesh sequencing), in
order to reduce the initial phase of poor convergence of these schemes. However, one of the
drawbacks of these methods is that they still require powerful preconditioners in order to be
effective, and the best known preconditioners are generally matrix based (such as ILU), which entail
storage requirements similar to those of the first order linearization. A compromise, with obvious
limitations, is to use a weaker but low-storage preconditioner, such as block-diagonal with a Newton-
Krylov Method147. For upwind discretization based on reconstruction techniques, higher-order
Jacobian vector products can be formed by multiplying the first-order Jacobian with a higher-order
reconstructed vector. Although the method requires the storage of the first-order Jacobian, it is exact
and, unlike finite-difference techniques, can also be applied directly to the solution of the adjoin
equations, which is required in particular formulations of the design optimization problem.
Additional approximations to the exact Jacobian can be employed to further reduce the memory
requirements of implicit schemes. For example, one may choose to group sub-regions of the mesh
together and only retain the terms of the Jacobian that pertain to coupling within each region148.
These regions may be determined by a variety of methods and may exhibit varying degrees of
overlap. These approximations may be applied to the Jacobian itself but are most often applied to the
matrix-based preconditioners operating on the true Jacobian. Alternatively, the regions may be
reduced to lines through the mesh joining neighboring vertices, as in the method of line lets, which
attempts to approximate the Alternate Direction Implicit (ADI) scheme of structured grids149. On
sequential machines, the regions or lines can be processed individually, thus reducing the maximum
memory requirements to that of the largest region or line (although on parallel machines no savings
are realized if each region is assigned to an individual processor). The numerical coupling between
regions or lines is lost in these approaches, a loss that in turn degrades overall convergence, as the
number of regions increases. However, this degradation can be minimized by judicious choices of the
regions based on the character of the problem, a technique that has not yet been fully exploited. If
these regions are reduced to individual grid points (i.e. all implicit coupling between grid points is
discarded), only the diagonal block matrices of the original Jacobian are retained, and the point-
implicit method is obtained 150-151. These terms represent the coupling between the various fluid
dynamic equations at a grid point. While point-implicit methods can offer increases in efficiency over
regular explicit schemes, they are also plagued by slow convergence for fine grids. Point-implicit

146 Nielsen EJ, Anderson WK, Walters RW, Keyes DE. ,”Application of Newton-Krylov methodology to a three-
dimensional unstructured Euler code”, Proc. AIAA CFD Conf., 12th, San Diego. AIAA Pap. 95-1733-CP, 1995.
147 Shakib F, Hughes TJR, Johan Z., “A multi element group preconditioned GMRES algorithm for non-symmetric

problems arising in finite element analysis”, Computer. Methods Appl. Mech. Eng. 87:415–56, 1989.
148 Chan TF, Mathew TP.,” Domain decomposition algorithms”, Acta Numer. pp. 61–143.
149 Hassan O, Morgan K, Peraire J., “An implicit finite element method for high speed flows”, AIAA Pap. 90-0402.
150 Hassan O, Morgan K, Peraire J., “An implicit finite element method for high speed flows”, AIAA Pap. 90-0402.
151 Thareja RR, Stewart JR, Hassan O, Morgan K, Peraire J., “A point implicit unstructured grid solver for the Euler

and Navier-Stokes equations.”, AIAA Pap. 88-0036, 1988.

 112

methods are sometimes viewed as preconditioning techniques for explicit schemes, where the point-
implicit matrix is the preconditioner.
4.12.2.2 Reflections on the Evolution of Implicit Navier-Stokes Algorithms152
The motivation for implicit viscous flow solvers has been the need to solve complex flows requiring
highly nonuniform grids and with multiple time and length scales. Such problems can present severe
algorithmic challenges when resolving disparate local length scales introduced by geometry, very
thin shear layers, and other localized flow structures. In addition, the differing time scales of
convection, diffusion, sound propagation, and chemical reaction can result in equation stiffness, a
term used for ordinary differential equations whose system matrices have a wide range of
eigenvalues. In both instances, the enhanced stability properties available from implicit schemes can
help by allowing larger time steps, within an objective of either time accuracy or convergence to a
steady solution. We will briefly comment here on the development of our own noniterative time-
linearized, block-coupled, Alternating-Direction Implicit (ADI) scheme reported during 1973-
1977, and then give some personal reflections on the subsequent evolution of these ideas and some
of the progress achieved through related ideas and innovations introduced by others.
4.12.2.2.1 The Noniterative Time-Linearized Block-Coupled ADI Scheme
With much appreciated support from Dr. Morton Cooper at the Office of Naval Research, we began
work that led to our 1973 implicit algorithm for the3D compressible Navier-Stokes equations in
primitive variables. This algorithm combined
➢ a noniterative implicit time or local linearization,
➢ a coupled block-tridiagonal adaptation of the 1963 Douglas-Gunn formalism for generating
n-dimensional ADI schemes from any linear scalar implicit time-marching scheme, and
4.12.2.3 Use of Implicitly Coupled, Characteristic-Compatible Boundary Conditions
Once we recognized the utility of the Douglas-Gunn formalism to generate a 3D block-tridiagonal ADI
scheme for coupled sets of linear parabolic/hyperbolic equations, we needed to develop a suitable
linearization for the Navier-Stokes equations and stable boundary conditions. We did this by
experimenting with different techniques using the trivial one-dimensional test problem of inviscid
uniform flow at constant Mach number. This test problem was ideal because
• the exact solutions to the differential and difference equations are known and identical,
• solutions with central differences are independent of artificial dissipation, and
• it is parametric in Mach number.
We became convinced that the noniterative time linearization was a key ingredient when it provided
stability for large Courant numbers over a full range of Mach numbers without added dissipation,
provided that implicit characteristic-compatible inflow/outflow boundary conditions were used. The
time linearization fully coupled the equations, thus obviating ad hoc decoupling techniques, and it
was also unambiguous in guiding the implicit treatment of nonlinear boundary conditions, for
example the subsonic “wind tunnel” conditions of total pressure and temperature at inflow and static
pressure at outflow. A somewhat related two-dimensional scheme for magnetohydrodynamics was
developed independently in 1973 by [I. Lindemuth and J. Killeen].
They applied the [Peaceman-Rachford] ADI scheme without linearization, and then linearized
progressively about the two successive solutions obtained for each of the half steps. During the
period 1973-1980, our group applied the time-linearized ADI scheme to numerous three-
dimensional steady subsonic flows in straight and curved ducts and pipes, multiphase multispecies

152W. R. Briley and H. McDonald, “Reflections on the Evolution of Implicit Navier-Stokes Algorithms”, UTC-CECS-
SimCenter-2008-04, September 2008.
 113

turbulent combusting flows, horseshoe vortex flows, and discrete-hole cooling jets. It was also used
in 1975 as a spatial marching algorithm for three-dimensional supersonic flow.
It was significant that the noniterative linearization has second-order time accuracy and does not
reduce the order of first or second-order time differencing. Thus Newton iteration to solve a
nonlinear implicit equation was unnecessary in principle because halving the time step could reduce
both linearization and time differencing errors at the same cost as a single Newton iteration.
Furthermore, both of these errors vanish in steady solutions, and the use of a pseudo time iteration
avoided the need for an accurate initial guess for Newton iteration. In fact, the time step could control
the linearization error during iteration toward a steady solution.
The cost of a single time step using the noniterative implicit scheme was only about twice that of
explicit methods, and it generally had much faster convergence to steady solutions. This efficiency
was needed for implicit solution at a time when computers were inadequate for many practical CFD
problems, especially in three dimensions. However, many subsequent improvements in implicit
algorithms and solution methodology would be needed to achieve modern high-fidelity simulations,
as increasingly powerful computers enabled them. In our view, the primary long-term contribution
of our Navier-Stokes algorithm was to introduce the noniterative time or local linearization to
systems of nonlinear partial differential equations and boundary conditions, and to recognize the
value of the scalar Douglas Gunn formalization to generate implicit coupled ADI schemes for the time
linearized equations. This stimulated considerable interest and further research in implicit
algorithms for the Navier-Stokes equations.
4.12.2.3.1 Early Work at NASA Ames Research Center
Our 1974 seminar at NASA Ames helped to generate some of this interest, and many important
contributions were made there. In 1976, Beam and Warming developed an implicit approximate
factorization (AF) algorithm for the Euler equations using conservative equations that admit
discontinuous solutions. They gave a very concise and elegant derivation of their algorithm by
expressing the time linearization in terms of flux Jacobian matrices for conserved variables Q, and
then using approximate factorization in terms of the implicit solution variable Qn+1 to generate the
one-dimensional block-tridiagonal systems. Their 1978 Navier-Stokes algorithm was similar but
introduced the “delta-form” factorization in terms of Δ Q = (Qn+1 - Qn). Although written in a different
form, the approximate factorization in terms of Δ Q and the Douglas-Gunn procedure for generating
ADI methods give the same result for Q n+1; however, approximate factorization is much more direct
and quickly became the standard derivation. [Steger and Kutler] also gave an early adaptation of this
time-linearized AF scheme for incompressible flows in 1977, using [Chorin’s] artificial
compressibility formulation.
[Warming and Beam] did extensive work on stability of implicit schemes, and one important
development was their local stability analysis indicating that the Douglas-Gunn ADI scheme for the
first-order wave equation is unconditionally unstable in three dimensions. In the context of implicit
algorithms, the von Neumann stability analysis assumes linear equations with constant coefficients,
a uniform grid, and periodic boundary conditions. We later performed a matrix stability analysis for
this inviscid equation showing that when the periodic conditions are replaced by characteristic-
based inflow/outflow conditions, the algorithm is conditionally stable for inviscid Courant number
less than about 1.2. This may explain why many researchers including ourselves were able to obtain
numerous steady solutions in three dimensions. Although it is unlikely that any algorithm would be
unconditionally stable for complex nonlinear systems, the lack of unconditional stability for this
model problem properly motivated the subsequent development of much improved two-factor AF
schemes for three dimensions, most notably the lower-upper (LU) factorization.
4.12.2.3.2 Characteristic-Based Upwind Schemes
A major algorithmic advance came in the early 1980s when [Steger and Warming, B. van Leer, A.
Harten, P. L. Roe], and others developed high resolution characteristic-based (upwind) numerical
 114

fluxes. These flux formulas are used with conservative finite-volume or integral formulations that are
capable of preserving discontinuities, with consequent higher resolution properties than finite-
difference schemes for flows with discontinuities and/or thin viscous layers. Upwind flux
formulations are also widely used for artificial compressibility formulations. While perhaps
originally motivated by their discontinuity-preserving properties, their local upwind properties later
were a key to developing more effective two-factor three-dimensional implicit methods.
4.12.2.3.3 Reducing Factorization Error in Time-Linearized Schemes
The efficiency of the noniterative implicit algorithms was procured at expense of both time-
linearization and factorization errors. Both of these errors are absent in converged steady solutions,
which was their primary use at the time, especially in three dimensions. However, it became evident
that stability and convergence were degraded by factorization error, especially by the three-factor
stability limitation and by increased stiffness at very low Mach numbers. Two additional algorithmic
advances were significant in reducing these factorization errors.
[Steger and Warming’s 1981] noniterative implicit lower-upper factorization (LU/AF) algorithm
based on flux-vector splitting was a very important development. The well-posed and stable steps of
this and other LU/AF schemes are enabled by characteristic-based numerical flux formulas and a
suitable technique for obtaining exact or approximate implicit flux Jacobian matrices. They provide
a stable two-factor framework for three dimensions and were eventually a key ingredient in
developing effective implicit schemes for unstructured grids. Riemann-based numerical fluxes such
as [Roe’s 1981] flux-difference scheme are now widely used in conjunction with LU/AF schemes.
Another means for reducing factorization error is the use of low Mach number preconditioning
techniques, and in 1983 we first developed a constant global preconditioning matrix that reduced
factorization error at low Mach number and greatly improved convergence to steady solutions. This
basic idea has evolved into other well-known preconditioning techniques that alter system
eigenvalues locally, as introduced by [E. Turkel (1984)] and [D. Choi and C. L. Merkle (1985)].
4.12.2.3.4 Iterative Time-Linearized Schemes for Structured and Unstructured Grids
More progress came with the use of iteration at each time step to completely eliminate the
factorization error. [S. R. Chakravarthy (1984) and R. W. MacCormack (1985)] began using iterative
relaxation methods that eliminated factorization errors in the time-linearized scheme. Iterative
schemes later became the basis for successful implicit algorithms that are applicable to unstructured
grids. Although line-oriented factorizations are not applicable for unstructured grids, the time
linearized equations can be solved by point or line iterative methods and by LU relaxation or
factorization. Iterative time-linearized implicit schemes for unstructured meshes were introduced in
the early 1990s by [ J. T. Batina] using point-implicit, two-sweep Gauss-Seidel, and two-sweep point-
Jacobi relaxation, by [V. Venkatakrishnan and D. J. Mavriplis] using preconditioned generalized
minimal residual (GMRES) iteration, and by [W. K. Anderson] using multicolor vectorizable Gauss-
Seidel relaxation.
4.12.2.3.5 Iterative Newton-Linearized Unsteady Schemes
Another significant advance came when [Chakravarthy and M. M. Rai (1986)] introduced an
(approximate) Newton linearization with iterative relaxation that, upon convergence, gives a
solution of the nonlinear unsteady discrete approximation. This introduction of Newton-iterative
schemes became feasible with improvements in computer processing speed and memory, which
allowed both larger problem sizes and algorithm improvements. Although the Newton iterations
themselves were solved approximately using relaxation or factorization techniques, the converged
Newton iteration eliminates both time-linearization and factorization errors at each time step.
Newton-linearized iterative schemes were also developed for the incompressible artificial
compressibility equations by [D. Pan and Chakravarthy (1989), and S. E. Rogers and D. Kwak (1990)].
The Newton linearized methods are distinct from time-linearized methods, which approach Newton
 115

linearization of the steady equations as the time step becomes infinite.

In 1991, we began our current long-term collaboration with Dave Whitfield and his group, then at
the National Science Foundation Engineering Research Center at Mississippi State University.
[Whitfield and L. K. Taylor] had developed an iterative Newton-linearized unsteady upwind artificial-
compressibility scheme that incorporated symmetric Gauss-Seidel (LU/SGS) relaxation to solve the
Newton linearization. This method also used numerical differentiation to compute accurate Roe-flux
Jacobian matrices instead of using approximate Jacobians. Their implementation exploited
CPU/memory trade-offs enabled by large memory resources that had become available, at that time
Cray SSD.
By organizing the computation into sequential blocks and storing the residuals and large numerical
Jacobian matrices, the linear LU/SGS iteration sweeps became extremely inexpensive, and the
factorization error could be eliminated at very little cost. They subsequently modified this algorithm
to perform multiple Newton iteration cycles, using LU/SGS sub-iteration to solve each linear Newton
iterate. Another significant advance came in 1998, when [J.C. Newman III, Anderson and Whitfield]
began using complex variable numerical differentiation to compute highly accurate flux Jacobians
and sensitivity derivatives.
There is some algorithmic unity in the fact that symmetric Gauss-Seidel relaxation and two factor AF
are equivalent linear iteration schemes, although they are generally written in different forms. The
Newton-linearized schemes work well for steady solutions and can be used without the added cost
of Newton iterations, but they are especially beneficial for time-accurate simulations, since they
permit fully implicit nonlinear time-integration schemes. Accordingly, the Newton linearization of
the unsteady equations is commonly used instead of the noniterative time linearization.
4.12.2.3.6 Scalable Parallel Algorithms for Structured and Unstructured Grids
In the early 1990s, we began exploring methods for transitioning implicit algorithms to parallel
computers. In 1995, [R. Pankajakshan and Briley] modified Taylor and Whitfield’s iterative Newton-
linearized Navier-Stokes algorithm to enable parallel solution with minimal effect on its algorithmic
performance. The parallel algorithm used a modified block-Jacobi symmetric Gauss-Seidel (BJ-SGS)
sub-iteration scheme with coarse-grained domain decomposition for mapping to distributed-
memory processors. In his 1997 dissertation, [Pankajakshan] developed a parallel iteration
hierarchy for time-accurate solutions that encompassed multigrid, Newton and BJ-SGS iterations, and
evaluated both algorithmic and software performance. A semi-empirical parallel performance model
for his MPICH software implementation indicated constrained-memory scalability to at least 400
processors using contemporaneous distributed-memory computers such as the IBM-SP2 and Cray-
T3E. In 1997, C. Sheng and Whitfield began working with [Kyle Anderson’s] implicit unstructured
time-linearized flow solver and developed a multi-block implementation to reduce memory
requirements on single-processor machines. Building on this work, [D. G. Hyams’s 2000] dissertation
developed a scalable parallel implicit viscous flow solver for highly nonuniform multi-element
unstructured grids. His work addressed issues of parallel iteration hierarchy for the Newton-
linearized method and of the treatment of connectivity and coupling of the sub-domain interfaces.
Finally, after our current group was formed in 2002, [K. Sreenivas, Hyams, D. S. Nichols III,
Pankajakshan, and Taylor] developed a new code (now called Tenasi) based on this same algorithm
but with many refinements, extensions and new capabilities for complex computational engineering
analysis and design applications. Implicit methods that have evolved from our early work continue
to be applied to a wide range of problems. Many of these methods appear sufficiently attractive that
a significant number of researchers are exploring algorithmic development to improve or extend
these methods to an even broader range of practical problems. Our own work in recent years has
focused on various application problems and related algorithm developments. These interests have
included multiphase and chemically reacting flows, the shallow-water and Maxwell’s equations,
 116

preconditioning, treatment of source terms and discontinuities other than shocks, and complex
variable differentiation.

4.13 Listing to CFD Numerical Methods

This section covering the outline of CFD and brought to you by CFD-Wiki and mentioned here as a
recap. The user should refer to this page for further reference. Be aware that this is lengthy and
general list and not all the items are relevant to the problem in hand. Top level is displayed as Red,
next Green, black, Brown, and so on.
• Basic Aspects of Governing Equations as related to CFD
• Classification of Governing Equations (GE)
• Integral form of GE
• Differential form of GE
• Generic Scalar Transport Equation
• Mathematical behavior of Partial Differential Equations
• Classification of physical behavior
• Equilibrium Problems
• Marching Problems
• Mathematical Classification of flows
• Hyperbolic Flows
• Parabolic Flows
• Elliptic Flows
• Mixed Flow
• Stability Analysis
• Fourier Stability Analysis
• von Neuman Stability Analysis
• Courant–Friedrichs–Lewy (CFL) condition
• Properties of Numerical Solution Methods
• Consistency
• Stability
• Conservation
• Boundedness
• Reliability
• Accuracy
• Solution of Euler Equation
• Method of Characteristics
• Shock Capturing Techniques
• Explicit Mac Cormack Method
• Flux Splitting (Stegar-Warming)
• Total Variation Diminishing (TVD) scheme
• Artificial Dissipation
• The SCM
• Beam & Warming Approximate Factorization scheme
• Solution of Poisson's Equation
• Type-Dependent Differencing
• Solution of Navier-Stokes Equations
• Explicit schemes
• Mac Cormack Predictor-Corrector scheme
 117

• Beam & Warming (AF)

• Vorticity-Stream Function
• Velocity-Pressure Coupling
• Rhie-Chow Interpolation
• Fully Coupled Methods - FC
• Density Methods
• De-Coupled Methods - DC
• Pressure Methods
• SIMPLE
• SIMPLEC - SIMPLE Consistent
• SIMPLEM – SIMPLE Modified
• SIMPLEX
• SIMPLEST - SIMPLE Shortened
• SIMPLER - SIMPLE Revised
• PISO - Pressure Implicit with Split Operator
• PRIME - Pressure Implicit Momentum Explicit
• MSIMPLEC, MPISO , SIMPLESSEC , SIMPLESSE
• CPC - Compressible Pressure Correction algorithm
• MCBA - Mass Conservation Based Algorithms
• GCBA - Geometric Conservation Based Algorithms
• IPSA - Inter-Phase Slip Algorithm
• CPI - Consistent Physical Interpolation
• MWIM - Momentum Weighted Interpolation Method
• PWIM - Pressure Weighted Interpolation Method
• ACM - Artificial Compressibility Method
• CIP - Cubic Interpolating Polynomial Method
• Discrete Operator Splitting
• Solution of Boltzmann Equation CFD Related algorithms
• Wall Distance Calculations
• Moore's k-tree algorithm
• Transport equation based wall distance calculation
• Geometrical Calculations
• Area calculations
• Volume calculations
• Calculation on non-orthogonal curvilinear structured grids, finite-volume method
• General Discretization
• Domain (Field) Discretization
• Grid or Mesh Generation
• Structured
• Algebraic
• PDE
• Adaptive Meshes
• Unstructured
• Advancing Front
• Delaney Triangulation
• Octree Decomposition
• Unstructured Hexahedral Meshes
 118

• Hybrid Meshes
• Overset Meshes
• Cartesian Meshes
• Polyhedral Meshes
• Adaptive Meshes
• Governing Equations Discretization
• Finite Difference
• Finite Element
• Spectral Volume
• Discontinuous Galerkin
• Finite Volume
• Generic scalar transport equation
• Source term linearization
• Gradient computation
• Special Discretization
• Discretization of the diffusion term
• Discretization of the convection term
• Discretization of the transient term
• Time Discretization
• Euler Method
• Crank-Nicolson Method
• Predictor-Corrector Methods
• Runge Kutta Methods
• Adams Bashforth Method
• Adams Moulton Method
• Implicit second order Method

• Linear Systems of Equations

• Direct Methods
• Gaussian Elimination
• LU Decomposition
• Tri-Diagonal Matrix Algorithm - (TDMA -Thomas Algorithm)
• Iterative Methods
• Gauss-Seidel Method
• Jacobi Method
• Successive 0ver-Relaxation Method - SOR
• Stone's Method
• Alternating Direction Implicit (ADI) Method
• Conjugate Gradient Methods
• Conjugate Gradient Method of Golub and van Loan
• Bi-Conjugate Gradient Method
• Bi-Conjugate Gradient Stabilized Method
• Matrix Factorization and Preconditioning
• Incomplete LU Factorization - ILU
• Incomplete Cholesky Factorization
• Multigrid Methods
• Geometric Multigrid – Full Approximation Storage (FAS)
• Algebraic Multigrid - AMG
 119

• Efficiency and Stability

• Limiters
• Flux Limiters
• Gradient Limiters

4.14 What Constitute Solver in CFD

According to [Aditya Kashi], the “Solver” in CFD, can mean several things depending on the context
which been used. It can refer to the spatial discretization scheme, like “finite volume solver” or
“spectral element solver”. In less than often, it can refer to the algorithm used for time stepping (also
called “integrator”), such as “explicit Runge-Kutta” etc. But it usually refers to the algorithm used
for solving the system of linear equations that needs to be solved (PDE solver) such as “sparse LU
solver” or “GMRES solver”, etc.
4.14.1 Linear vs. Non-linear CFD Solver153
From time to time, question arises regarding which solver is best suited for my CFD application?
Without going into too much math, it was generally believed that there are two broad classes of PDE
solving you'll see in CFD. One is the "Euler time marching methods", sometimes called "pseudo-
transient continuation" in the math world, which directly deals with the nonlinearity. The other class
does a linearization of some kind, with an "outer" iterative method to handle the nonlinearity.
Discretizing nonlinear PDEs always gives you a system of nonlinear algebraic equations. A
straightforward example is linearizing these algebraic equations, then using Newton's Method. In
this case newton's method requires a series of linear solves. Euler (or any explicit pseudo time
marching process) works for nonlinear systems, but any nonlinear solver can also inherently solve
linear problems as well. Gauss-Seidel/Jacobi/SOR methods are very similar. A Jacobi method with
a weight set to your "time step" is exactly equivalent to forward Euler. Pseudo time marching
processes are used all over the place in CFD because they can be an extremely robust way to tackle
the nonlinearity by forcing the solution to maintain something physical -- newton's method doesn't
care that pressure or density shouldn't be negative, but a time marching process forces the solution
iterations to be on a "physically realizable manifold" the whole time.
There's multiple levels of "iterative methods" which also probably clouds the description, given a
linear system Ax = b, you can use a "direct solver" which is tantamount to finding the inverse of A,
e.g. gaussian elimination or LU decomposition. You can also use an "iterative solver" such as SOR or
Krylov subspace methods (e.g. GMRES) where you iteratively improve an approximate solution to
Ax = b until it meets some accuracy requirement. You always use some kind of nonlinear method for
nonlinear PDEs whether you know it or not. It’s also interesting to note that in the limit of infinitely
large time steps, implicit Euler time-marching actually becomes Newton’s method.

4.15 Major Integration Schemes for CFD

There are several distinct methods of numerical integration as:
➢ Lattice Boltzmann Methods (LBM)
➢ Finite Volume Method (FV)
➢ Vortex Methods and other Particle
➢ Finite Difference Method (FD)
Methods
➢ Finite Element Method (FE)
➢ Boundary Element Method (BEM)
➢ Weighted Residual (WRM)
➢ Polynomial Fitting
➢ Spectral Methods
➢ Integral Method
• Fast Fourier Transform Methods
(FFT)

153 Reddit Blog, “confusion regarding numerical methods for Linear vs. nonlinear PDE”, 2018.
Among those, the Finite Difference (FD), Finite Element (FE) and Finite Volume (FV) methods are
the most prominent ones with FD being the older one. FD solves the partial differential equations
(PDEs) in the strong (or differential) form, and FE and FV solve PDEs in a weak (or integral) form
[YAO et al.]154. Each of the abovementioned methods possesses its own distinct advantages and
disadvantages. Although the FV was for a long time the most popular CFD method, and that might
still be the case now, the FE, with sophisticated stabilization techniques, space–time methods, and
NURBS technology, is now in a very advanced stage.

154Jianyao Yao, G. R. Liu, Dong Qian, Chung-Lung Chen and George X. Xu. “A Moving-Mesh Gradient Smoothing
Method For Compressible CFD Problems”, Mathematical Models and Methods In Applied Sciences Vol. 23, No. 2
(2013) 273–305.
 121

5 Finite Differencing (FD) & Spatial Discretization

5.1 Finite Difference Method (FD)
The Finite Difference can be used merely as a mathematical tool to develop algebraic approximation
to PDE’s with relative ease. It proceed in rather formal mechanical way, using Taylor’s series
expansion of operating PDE with physical reasoning aside. The equation for u x and uxx evaluated at
point i, can be approximated by

∂u ui+1 − ui ui − ui−1 ui+1 − ui−1

| ≈ ≈ ≈
∂x i ⏟ ∆x ⏟ ∆x ⏟ 2∆x
Forward Backward Cetral
2
∂ u ui+1 − 2ui + ui−1
≈
∂x 2 (∆x)2
Eq. 5.1
Where the ux is approximated using a forward, backward, and central differencing. Δx is small but
finite. Using the same reasoning we obtained an expression for second derivatives. For an excellent
reference in FD, user should refer to155, and will be revisited later.
5.1.1 Some Issues Related Finite Differencing
The idea of finite differencing steams from the definition of derivatives as

u(x, y) u(x + Δx, y) − u(x, y)

= lim Eq. 5.2
x Δx → 0 Δx
5.1.1.1 Truncation Error
Assuming u to be continuous, this will be a reasonable approximation to the derivative. With the aid
of Taylor’s series expansion for expansion of u(x+Δx) could be expressed as

u  2 u (x) 2
u(x + Δx, y) = u(x, y) + Δx + 2 + ...... Eq. 5.3
x x 2!

Rearranging this and dividing by Δx,

x - values
∂u u(x + ∆x, y) − u(x, y)
=
Truncatuin Error →

2.6
∂x ∆x
∂2 u ∆x
− … 1.2
⏟2 2
∂x 0.5
Truncation Error= ο(∆x) 0.3
Eq. 5.4 Order of Accuracy →

Figure 5.1 Correlation between

truncation error and order of
accuracy

D. Anderson, J., Tannehill, R., Pletcher,”Computational Fluid Mechanics and Heat Transfer”, ISBN 0-89116-
155

471-5 – 1984.
 122

Where the truncation error is the difference between PDE and FDE representation and characterized
by using the order notation. It is inversely related to the order of accuracy for the equations as
depicted in Figure 5.1 and would be an extremely important criteria in accuracy of discretized
equation as it is directly related to the stability and accuracy consideration. Therefore, for Eq. 5.4, it
could be said that the forward finite difference representation here is 1st order accurate in space (Δx).
For most practical applications, 2nd order accuracy would be sufficient as by increasing the order of
accuracy, the CPU cost would also increase. Free temperature profile of a jet flow would be an
excellent example in that case (see Figure 5.2). In order to be acceptable, the difference
representation needs to meet the conditions of consistency and stability, to be discussed next.

First Order Accurate Second Oreder Accurate

Third Order Accutare Fifth Order Accurate

Figure 5.2 Free Jet Flow profile for different order of accuracy

5.1.1.2 Consistency
Consistency deals with extend on which the finite difference equations approximate the PDEs. It is
directly linked to truncation error by means of showing that in the limit, as mesh been refined
infinitely, the difference between the PDE and FDE vanishes, i.e. lim mesh → 0 (PDE-FDE) = 0 as lim
mesh → 0 (TE) = 0. Therefore, for consistency to be satisfied, on the limit, the truncation error should
vanish. An important question concerning computational solutions is what guarantee can be given
that the computational solution will be close to the exact solution of the partial differential
equation(s) and under what circumstances the computational solution will coincide with the exact
solution. The second part of this question can be answered (superficially) by requiring that the
approximate (computational) solution should converge to the exact solution as the grid spacings At,
Ax shrink to zero. However, convergence is very difficult to establish directly so that an indirect
route, as indicated in Figure 5.3, is usually followed. The indirect route requires that the system of
algebraic equations formed by the discretization process should be consistent with the governing
partial differential equation(s). Consistency implies that the discretization process can be reversed,
through a Taylor series expansion, to recover the governing equation(s). In addition, the algorithm
 123

used to solve the algebraic equations to give the approximate solution, T, must be stable. Then the
pseudo-equation156:
Consistancy + Stability = Convergence
Eq. 5.5
is invoked to imply convergence. The conditions under which Eq. 5.5 can be made precise are given
by the Lax equivalence theorem. It is very difficult to obtain theoretical guidance for the behavior of
the solution on a grid of finite size. Most of the useful theoretical results are strictly only applicable
in the limit that the grid size shrinks to zero. However the connections that are established between
convergence, consistency and stability are also qualitatively useful in assessing computational
solutions on a finite grid.

Figure 5.3 Conceptual Relationship Between Consistency, Stability and Convergence – Courtesy of
Fletcher

5.1.1.3 Stability and Effect of CFL Number

The stability concept is a bit more involved than consistency and in a restrict sense only applicable
to marching problems. A stable numerical scheme is the one that errors do not propagate through
the system. It devises a relation and imposes a restriction between mesh and time step sizes, defined
as CFL number. There are two prominent procedures, namely, the Von Neumann and Fourier for
stability analysis. To better understand, a simple 1D stability criteria could be devised where CFL =
α Δt/Δx with α being a positive constant, as being presented in Figure 5.4. On the right (b), with
reduced Δx, then CFL < 1, point Q is within computational domain of influence (shaded), and could
be differentiated using the neighbor cells, therefore, is stable. But on the left (a), where CFL < 1, Q
lies outside of computational domain of influence, therefore, unstable. But on the left (a), where CFL
≥ 1, Q lies outside of computational domain of influence, therefore, unstable. It is obvious that the size
of mesh, Δx, is crucial in stability here and by reducing it (finer mesh), the stability constrain is
satisfied. Therefore, it is safe to assume that the numerical stability requirement for many explicit
numerical methods for solving hyperbolic PDE’s, is CFL condition which is

156C. A. J. Fletcher, “Computational Techniques for Fluid Dynamics 1 - Fundamental and General Techniques”, 2nd
Ed., Springer.
 124

Δt
CFL = a 1 Eq. 5.6
Δx

Reducing the Δt, if insisting in keeping the same Δx, would achieve that. Consequently, it is needing
finer step sizes on explicit methodologies for stable conditions. It is known that the best results for
hyperbolic systems using the most common explicit methods are obtained with CFL near unity (see
Figure 5.4).

Figure 5.4 1D Stability Analysis Based on CFL Number; (a) CFL < 1 Unstable; (b) CFL ≥ 1 Stable

[Schneider, et al.]157 added some remarks about the CFL condition for explicit time discretization
methods of Adams–Bashforth and Runge–Kutta type. It was shown that for convection-dominated
problems stability conditions of the type Δt ≤ a Δx α are found for high order space discretization’s,
where the exponent α depends on the order of the time scheme. For example, for second order
Adams–Bashforth and Runge–Kutta schemes we find α = 4/3. A more general 3D CFL number using
the empirical formula [Tannehill et al., 1975] defined as:

-1
α(Δt)CFL  u v w 1 1 1 
Δt  wher e (t)CFL   + + +a + + 
1 + 2 / Re   Δx Δy Δz (x )2 (y )2 (z )2 
ρ u Δx ρ v Δy ρ w Δz
with Re Δ = Min (ReΔx , Re Δy , Re Δz ) Re Δx = Re Δy = Re Δz =
μ μ μ
and a = (p /  ) 0.5 , safty factor α  0.9
Eq. 5.7
5.1.1.4 Convergence
A solution of the algebraic equations (Figure 5.3) which approximate a given partial differential
equation is said to be convergent if the approximate solution approaches the exact solution of the
partial differential equation for each value of the independent variable as the grid spacing tends to
zero. Thus we require

157 Kai Schneider, Dmitry Kolomenskiy, and Erwan Deriaz, “Is the CFL Condition Sufficient? Some Remarks”.
 125

̅(xj , t n )
Tjn → T as ∆x , ∆t → 0
Eq. 5.8
The difference between the exact solution of the partial differential equation and the exact solution
of the system of algebraic equations is called the solution error, denoted by enj; that is

̅(xj , t n ) − Tjn
enj = T
Eq. 5.9
The exact solution of the system of algebraic equations is the approximate solution of the governing
partial differential equation. The exact solution of the system of algebraic equations is obtained hen
no numerical errors of any sort, such as those due to round-off, are introduced during the
computation. The magnitude of the error, enj, at the ( j , n)-th node typically depends on the size of
the grid spacings, Dx and Dt, and on the values of the higher-order derivatives at that node, omitted
from the finite difference approximations to the derivatives in the given differential equation. Proof
that a solution to the system of algebraic equations converges to the solution of the partial differential
equation is generally very difficult, even for the simplest cases.
5.1.1.4.1 Numerical Convergence
For the equations that govern fluid How, convergence is usually impossible to demonstrate
theoretically. However, for problems that possess an exact solution, like the diffusion equation, it is
possible to obtain numerical solutions on a successively refined grid and compute a solution error.
Convergence implies that the solution error should reduce to zero as the grid spacing is shrunk
to zero. The solutions have been obtained on successively refined spatial grids, Ax = 0.2, 0.1, 0.05 and
0.025. The corresponding rms errors are shown in Table 5.1 for s = 0.50 and 0.30. It is clear that
the rms error reduces like Δx2 approximately. Based on these results it would be a reasonable

Table 5.1 Solution Error (rms) Reduction with Grid Refinement – Courtesy of Fletcher

inference that refining the grid would produce a further reduction in the rms error and, in the limit
of Δx (for fixed s) going to zero, the solution of the algebraic equations would converge to the exact
solution. The establishment of numerical convergence is rather an expensive process since usually
very fine grids are necessary. As s is kept constant in the above example the timestep is being reduced
by a factor of four for each halving of Dx. In Table 5.1 the solution error is computed at t = 5000 s.
This implies the finest grid solution at s = 0.30 requires 266 time steps before the solution error is
computed. [Fletcher]158.

5.2 Spatial Discretization Schemes

It is worth noting that concept of discretization commonly refers to Domain Discretization (Grid
Generation), Spatial discretization (Governing Equation), and Temporal discretization (Transient
case). Here we are concerned with Spatial Discretization of G.E. From Wikipedia, the free

158C. A. J. Fletcher, “Computational Techniques for Fluid Dynamics 1 - Fundamental and General Techniques”, 2nd
Ed., Springer.
 126

encyclopedia, in mathematics, discretization concerns the process of transferring continuous

functions, models, and equations into discrete counterparts. This process is usually carried out as a
first step toward making them suitable for numerical evaluation and implementation on digital
computers. Among most frequently used discretization schemes are upwind differencing (UD),
Central Differencing (CD), Total Variation Diminishing (TVD), Monotone Upstream-Centered
Schemes (MUSCL), and Quadratic Upwind Interpolation for convection Kinetics (QUICK). Each has its
own characteristics and feasibility, to be discussed below. In addition, order of accuracy
consideration for a jet flow temperature profile has been shown below for 1st – 5th order (see Figure
5.2). Dramatic difference in the character of the solution between the first and second order or
higher is present, but higher order doesn’t mean it’s always more accurate (more CPU time & round
off errors).
5.2.1 1st order Upwind (UD)
Perhaps the most widely used and simplest scheme as shown in Figure 5.5-(a). It is conservative,
bounded and very stable but dissipative and creates false diffusion. Flux limiters are recommended
to encounter that. It is 1st order accurate and ideal for flow exhibiting a primary direction. Usually
recommended for hyperbolic flow regions159.
5.2.2 Central Differencing (CD)
This is a 2nd order accurate scheme as shown in Figure 5.5-(b). It is bounded at lower Peclet (Pe)
number, but not bounded or conservative at higher Pe number. Not recommended for Uni-
directional flows due to its symmetry character. Although it is not dissipative, it is dispersive and
usually recommended for elliptic flow regions. In order to overcome the issues with un-roundness
and wiggles in solution, leading to stability issues, a deferred linear interpolation approach for CD
is proposed a
φf = ⏟
φFUD + ⏟
φFCD − φFUD
implicit explicit
Eq. 5.10
As indicated, the upwind part is treated implicitly while the difference between the central-difference
and upwind values is treated explicitly. Provided that the numerical solution converges, this
approach leads to pure second-order accuracy. On some commercial suites such as Fluent©, the usage
of CD differencing is restricted to pressure-based solvers.
5.2.3 2nd order Upwind Differencing (UD)
This 2nd order accurate scheme as shown in Figure 5.5-(c) is as its first-order cousin, depends on
the flow direction but not bounded. To a lesser extent, it is still dissipative. Flux limiters are advised
on its implementation160.
5.2.4 Power Law Scheme
This is based on the analytical solution of the one-dimensional convection-diffusion equation. The
face value is determined from an exponential profile through the cell values. The exponential profile
is approximated by the following (see Figure 5.5-(d)).

(1 − 0.1 Pe)5
e = P − ( E −  P ) (5.13)
Pe
Eq. 5.11

159 Georgia Tech Computational Fluid Dynamics Graduate Course; spring 2007.
160 Georgia Tech Computational Fluid Dynamics Graduate Course; spring 2007.
 127

Where Pe is the Peclet number and for Pe > 10 the diffusion is ignored and 1st order upwind is used.

(a) 1st Order UD (b) 2nd Order CD (c) 2nd Order UD

(d) Power Law (e) Higher Order Upwind

Figure 5.5 Different Spatial Scheme

5.2.5 Quadratic Upwind Interpolation for Convection Kinetics (QUICK)

This is a 3nd order accurate scheme, independent to flow direction and not bounded. It is based on a
weighted average of second-order-upwind and central interpolations of the variables. It uses
consistent quadratic profile, therefore, conservative. It has a build in transported and is conditionally
bounded.
5.2.6 Monotone Upstream-Centered Schemes for Conservation Laws (MUSCL)
This is a highly accurate scheme for
solutions that exhibit shocks and
discontinuities. It is a combination of
1st and 2nd order accurate. Further, the
2nd order spatial accuracy for the
inviscid fluxes is achieved in by using
MUSCL extrapolation to reconstruct
an approximate value of the primitive
variables, each side of each cell face.
The MUSCL scheme uses a 13 point Invicsid Stencil (13 cells) Viscous Stencil (25 cells)
stencil per cell in three dimensions, as
shown in Figure 5.6 fluxes is Figure 5.6 Viscous & Inviscid Stencil for a 2nd order
achieved in by using MUSCL accurate MUSCL
 128

extrapolation to reconstruct an approximate value of the primitive variables, each side of each cell
face. The MUSCL scheme uses a 13 point stencil per cell in three dimensions, as shown in Figure 5.6.
The viscous flux is calculated using a Green's Theorem approach to calculate the derivatives at cell
faces and central differencing is used to calculate the scalar values. The viscous fluxes require an
additional twelve points to be added to the inviscid stencil for a total of 25 cells in the stencil per cell.
5.2.7 Total Variation Diminishing (TVD)
This is a 2nd order upwind differencing and defined as a system which does not increase the total
variation of the solution,

TV(u n +1 )  TV(u n ) where TV(u) =  u j−1 − u j (5.14)

j
Eq. 5.12
5.2.7.1 Total Variation Diminishing Cubic Interpolated Propagation (TVDCIP)
For special case of fire science, proposed the new scheme, TVDCIP (Total Variation Diminishing Cubic
Interpolated Propagation), that consists of the spatial interpolation and the mathematical algorithm
for switching scheme with high order accuracy. TVDCIP method is applied to solve the nonlinear
hyperbolic partial differential equations numerically such as Navier-Stokes type equations.161 It is
basically used for convection terms. Other special cases including viscoelastic flows, the polymeric
stress derivative term in the momentum we adopt a Fourth-Order Accurate Compact scheme (COM4)
which has no dissipative error.162
5.2.8 Monotone Advection and Recons traction Scheme (MARS)
MARS is a multidimensional second-order accurate differencing scheme that operates in two
separate steps:
5.2.8.1 Reconstruction
A set of monotone gradients are computed using a multidimensional Total Variation Diminishing
(TVD) scheme. The cell flow properties and the gradients completely define a second-order accurate
spatial discretization.
5.2.8.2 Advection
The reconstructed cell-face flow properties are used to compute the face fluxes for all advected
properties using a monotone and bounded advection scheme. This incorporates a variable
compression level which controls the amount of second-order up winding of the scheme without
affecting the order of accuracy of the spatial discretization. MARS does not rely on any problem
dependent parameters to work properly and it can automatically handle all flow problems and mesh
types supported. However, the user can control the ability of the advection scheme to accurately
capture sharp discontinuities in the flow by setting the scheme’s compression level to a value
between 0 and 1. Low values for this parameter result in a computationally efficient scheme at the
expense of sharpness of resolution. High values improve the resolution but result in an increased
number of iterations when steady flows are computed. The default value for this parameter is 0.5
which is a compromise between accuracy and convergence rate. Of all schemes available, MARS
possesses the least sensitivity of solution accuracy to the mesh structure and skewness.
5.2.9 Blended Differencing
In this method a higher-order, non-monotone scheme like CD or LUD is blended with the lower-order
UD scheme to suppress dispersion. However, the blending factor γ is now user-specified, giving

161 T. Yamanashi, H. Uchida, and M. Morita,” Study on the Numerical Accuracy for the CFD”, Tokyo University of
Science, 1-3 Kagurazaka, Shinjuku-ku, Tokyo, 162-8601, Japan.
162 S.K. Lele, “Compact finite difference schemes with spectral-like resolution”, J. Computational Phys. 103 (1992).
 129

 jBD = γ jCD/LUD + (1 − γ) jUD where 0  γ 1 (5.15)

Eq. 5.13
Evidently, this practice will sometimes involve some trial-and-error adjustments for optimal results.
5.2.10 Higher Order Upwind Schemes
Even though the fluxes were computed exactly, the solution was very inaccurate due to numerical
diffusion. Since the advection is exact, the problem must lie with the averaging of the solution and
the assumption of a constant state in each cell. To generate more accurate schemes, we need to
reconstruct a more accurate representation of the function in each cell. A fairly general family of
higher order methods can be constructed by weighing the slopes in each cell; in different ways as
evaluation of the function at the cell boundary (see Figure 5.5-(e)). Different values of k give
different schemes as depicted in Eq. 5.14:

1− k
u Lj+1/2 = u j + (u j − u j−1 ) + 1 + k (u j+1 − u j )
4 4
1− k
u Rj+1/2 = u j+1 − (u j+1 − u j ) − 1 + k (u j+2 − u j+1 )
4 4
1
k = -1 → Second order k = → Third order
3
k = 0 → Fromm's scheme k = 1 → Centeral
1
k= → QUICK
2
Eq. 5.14
To prevent oscillations near shocks when using high order schemes, we have already talked about
artificial viscosity where we attempt to smooth out oscillations around the shocks. The more modern
approach is to prevent the appearance of oscillations by either:
• Limit the slopes when the variables are extrapolated to the cell boundaries.
• Limit the fluxes near shocks to prevent under or over shoot.
• Reduce the cell Reynolds number (Pe < small ⇾ smaller grid size).
To prevent oscillations, apply Limiters to reduce the slopes where they will cause oscillations163.
5.2.10.1 Case Study - Other High Resolution Differencing Scheme For Arbitrarily Unstructured
Meshes

Citation : Jasak, H., Weller, H. and Gosman, A. (1999), High resolution NVD differencing scheme for
arbitrarily unstructured meshes. Int. J. Numer. Meth. Fluids, 31: 431-449.
https://doi.org/10.1002/(SICI)1097-0363(19990930)31:2<431::AID-FLD884>3.0.CO;2-T
Discretization of the convective part of fluid transport equations has proven to be one of the most
troublesome parts of the numerical fluid mechanics164. The objective is to devise a practice that will
produce a bounded, accurate and convergent solution165. Different procedures for assembling higher-
order bounded differencing schemes have been suggested, the most popular being the total

163 Grétar Tryggvason, “Computational Fluid Dynamics Lecture Series”, spring 2013.
164 H. Jasak, H .G. Weller and A. D. Gosman, “High Resolution NVD Differencing Scheme For Arbitrarily
Unstructured Meshes”, Int. J. Numerical Meth. Fluids 31: 431–449 (1999).
165 See Previous.
 130

variation diminishing (TVD)166-167 and the more recent group of normalized variable diagram
(NVD) approaches. All of these schemes introduce some procedure in which the discretization
practice for the convection term is adjusted locally, based on the currently available solution. In both
families, the bounding procedure uses an indicator function that follows the local shape of the
solution and is based on 1D analysis. Often the analysis refers to cells that are more than once
removed from the face under consideration, usually in the ‘upwind’ direction.
One of the requirements of the differencing scheme presented here is its applicability on
unstructured meshes. The convenience of mesh generation for such meshes makes them attractive
for complicated geometries in industrial CFD applications. Arbitrarily unstructured meshing168-169
further simplifies mesh generation in complicated geometries, but requires a more general face-
based discretization procedure. Only in hexahedral unstructured meshes is it possible to ‘recognize’
the distinct directions needed to implement the TVD and NVD-type schemes in 2D and 3D. In

(a) Unstructured (b) Arbitrarily

quadrilateral mesh unstructured mesh

Figure 5.7 Deciding the Upwind Direction in Arbitrary Unstructured Mesh

arbitrarily unstructured meshes, the concept of the ‘far upwind neighbor’ for a face becomes quite
complicated. It is not clear how to determine the far neighbor since the mesh does not have any clear
directionality (see Figure 5.7 (a)-(b)). For example, any of the points U1, U2 or even U3 in Figure
5.7 (b) can be considered to be an appropriate choice of the far upwind node for the face f. The
implementation of TVD- and NVD-type schemes in such circumstances is no longer a straightforward
exercise. In fact, it has also become obvious that the directionality of the hexahedral mesh does not
necessarily offer the appropriate ‘upwind’ information.
On arbitrarily unstructured meshes, the problem is twofold: it is necessary to store the additional ‘far
upwind’ addressing information (which is simply a programming issue) and, more importantly,
determine which of the ‘upwind’ cells represents the right choice, which is neither simple nor unique.
A more general definition of the unboundedness criterion that avoids the use of the ‘far upwind value’

166 S.Osher and S.R. Chakravarthy, ‘High resolution schemes and the entropy condition’, SIAM J. Numer. Anal., 21,
955–984 (1984).
167 B. Van Leer, ‘Towards the ultimate conservative differencing scheme. II Monotonicity and conservation

combined in a second-order scheme’, J. Comp. Phys., 14, 361–370 (1974).

168 A.D. Gosman, ‘Developments in industrial computational fluid dynamics’, Trans. I. Chem. E., (1998).
169 H. Jasak, ‘Error analysis and estimation in the finite volume method with applications to fluid flows’, Ph.D.

Thesis, Imperial College, University of London, 1996.

 131

is needed for both the TVD and NVD family of differencing schemes. See the [ JASAK et al.]170.
5.2.11 Other High-Order Differencing Schemes : ENO and WENO Schemes
In numerical analysis of differential equations, ENO (Essentially Non-Oscillatory) and WENO
(Weighted Essentially Non-Oscillatory) methods are classes of high-resolution schemes. They been
used in the numerical solution of hyperbolic partial differential equations with aim of capturing
discontinuities within. The first WENO scheme is developed by [Liu, Chan and Osher] in 1994. In
1996, [Guang-Sh and Chi-Wang Shu] developed new WENO scheme which is called WENO-
JS. Nowadays, there are many WENO methods. Reconstruction procedure of the ENO schemes is
based on the approximation of mean values of the primitive variables for each cell in the mesh by
polynomials of one order less than spatial order Of accuracy expected. For the construction of
polynomials of η-th order, one must use N(η) = (η+1)(η+2)/2 cells. For a 2D triangular grid, the p(x
, y) polynomials can, then, calculated as the following formulation:

p(x, y) = ∑ rβ1β2 (x − xc )β1 (y − yc )β2

|β|≤n
Eq. 5.15
here lβl = β1 + β2 with βi ∈ {1,2,3, , ,}, xc and yc are Cartesian coordinates of triangle and r β1 β2 are
unknown coefficients that must be determined171.
At the heart of the WENO schemes is actually an approximation procedure, not directly related to
PDEs, hence the WENO procedure can also be used in many non-PDE applications, including
computer vision and image processing172. As mentioned before, at the heart of the WENO schemes is
actually an approximation procedure, not directly related to PDEs. We use the following simple
problem of interpolation to describe this approximation procedure. Assume that we are given a
uniform mesh ⋯< x1 < x2 < x3 <⋯ and the point values of a function ui= u(xi) . We would like to find
an approximation of the function u(x) at a point other than the nodes xi , for example at the half
nodes x i+1/2 . This can be handled by the traditional approach of interpolation. The numerical
interpolation (approximation) can be defined as173:

(r−1)/2
2k
∂2k u
ûi+1/2 = ui+1/2 + ∑ a2k ∆x 2k
| + 𝒪(∆x r+1 )
∂x i+1/2
k=1
Eq. 5.16
where the coefficients a2k can be obtained by Taylor expansion. For example, we could find a unique
polynomial of degree at most two, denoted by p1(x) , which interpolates the function u(x) at the mesh
points in the stencil S1= {xi−2, xi−1 ,xi} . We could then use u(1)i+1/2 ≡ p1(xi+1/2) as an approximation to
the value u(xi+1/2) , which is given explicitly as
3 5 15
ui+1/2 = ui−2 − ui−1 + ui
8 4 8
Eq. 5.17

170 H. Jasak, H .G. Weller and A. D. Gosman, “High Resolution NVD Differencing Scheme For Arbitrarily
Unstructured Meshes”, Int. J. Numerical Meth. Fluids 31: 431–449 (1999).
171 W. R. Wolf, and J. L. F. Azevedo,, “Implementation of ENO and WENO Schemes For Finite Volume

Unstructured Grid Solutions of Compressible Aerodynamic Flows”, Sept. 2018.

172 Chi-Wang Shu, “WENO methods”, Scholarpedia.
173 Chuangchao Ye, Pengjunyi Zhang, Rui Yan, Dejun Sun, Zhenhua Wan, “Accelerating CFD simulation with

high order finite difference method on curvilinear coordinates for modern GPU clusters”, [physics.comp-ph] 14
Jun 2020.
 132

and is third order accurate. If we use the large stencil S = {xi−2, xi−1, xi, xi+1, xi+2} , which is the union of
all three third order stencils S1 , S2 and S3 , then we would be able to obtain an interpolation
polynomial p(x) of degree at most four, satisfying p(xj)=uj for j = i−2 , i−1 , i , i+1, i+2 , and giving an
approximation ui+1/2≡p(xi+1/2) , or explicitly as

3 5 45 15 5
ui+1/2 = ui−1 − ui−1 + ui + ui+1 − u
128 32 64 32 128 i+2
Eq. 5.18
which is fifth order accurate. Some high-order, compact, and conservative spatial discretization
strategy is proposed and implemented for a finite-difference solver, [M. Allahyari et al./JAFM, Vol. 14,
No. 2, pp. 345-359, 2021].
5.2.11.1 Case Study 1 - Adaptive Central-Upwind Weighted Compact Non-linear Scheme with
Increasing Order of Accuracy

Authors : Kamyar Mansour, Kaveh Fardipour

Title : Adaptive Central-upwind Weighted Compact Non-linear Scheme with Increasing Order of
Accuracy
Appeared in : Department of Aerospace Engineering, Amirkabir University of Technology, Tehran, Iran
5.2.11.1.1 Abstract
In this work, effect of using an adaptive central-upwind (ACU) interpolation on weighted compact
non-linear scheme (WCNS) is investigated. Based on the smoothness of solution, this type of
interpolation adapts between central and upwind stencils by a weighting relation and combination
of smoothness indicators of the optimal high-order stencil and its sub-stencils.
5.2.11.1.2 Introduction
Over past three decades there were many efforts for development of high-order numerical methods
that simultaneously have the capability to capture flow discontinuity and resolve small-scale features
of flow. Weighted Essentially Non-oscillatory (WENO) [1] scheme and Weighted Compact
Nonlinear Scheme (WCNS) [2] scheme are two families of such numerical methods.
The WENO scheme is based on Essentially Non-oscillatory (ENO) scheme [3], but instead of using
only one of sub-stencils, it uses a weighted combination of all sub-stencils. This scheme was
developed in finite volume framework by [Liu et al. [4]. [Jiang and Shu [1] extended the WENO
scheme to finite difference framework and proposed a new formulation for nonlinear weights to
increase order of accuracy and later [Balsara and Shu [5] and [Gerolymos et al. [6] studied the high
order behavior of the WENO scheme.
Despite having high order of accuracy and good shock capturing capabilities, the WENO scheme also
has some shortcomings. One of the problems with the original WENO scheme of Jiang and Shu [1] is
loss of accuracy near critical points. Analysis of Henrick et al. [7] showed this loss of accuracy is
because of nonlinear weights and they purposed a mapping method for computation of nonlinear
weights to prevent loss of accuracy. [Borges et al. [8] also purposed a new method for computation
of nonlinear weights of fifth order WENO to avoid loss of accuracy and later expanded it for higher
order of accuracy [9], their method has lower computational cost in comparison to [Henrick et al. [7]
mapping method.
There are several ways to reduce numerical dissipation of the WENO scheme. One them is hybrid
methods which only use the WENO scheme in vicinity of discontinuities and use another scheme with
lower or no numerical dissipation in smooth regions [10]-[12]. Another way is to optimize
dissipation and dispersion error [13]-[15], this usually achieved by finding optimal coefficients or
linear weights by minimizing integral error following optimizing procedures of Tam and Webb [16]
and Zhuang and Chen [17]. A more recent way for reduction of numerical dissipation of the WENO
 133

scheme is adaptive central-upwind WENO (ACU-WENO) scheme [18]-[20]. Based on smoothness of

solution, this new family of WENO scheme can adapt between central and upwind stencils and
achieves higher order of accuracy, numerical resolution and lower dissipation by using a central
stencil in smooth regions of solution.
The WCNS was originally developed by Deng and Zhang [21] and later extended to higher order of
accuracy by [Nonomura et al. [22] and [Zhang et al. [23]. This scheme is a combination of compact
scheme [24] and WENO interpolation [25]. This scheme includes a node-to-midpoint weighted
interpolation and a midpoint-to-node differencing and has three advantages over finite difference
WENO scheme [26]:
1- Slightly higher numerical resolution;
2- Compatibility with different flux treatments;
3- Better performance on general curvilinear grids [27].
[Nonomura and Fujii [28] studied effects of different types of midpoint-to-node differencing methods
on WCNS and they showed it does not significantly change numerical resolution and shock capturing
capabilities of WCNS. [Nonomura and Fujii [26] proposed a new formulation for midpoint-to-node
differencing, which significantly increases robustness of WNCS. Recently [Sumi and Kurotaki [29]
used a sixth order adaptive central-upwind interpolation with robust formulation of a tridiagonal
midpoint-to-node differencing to improve numerical resolution and robustness of original WCNS
[21]. Some studies [22][30]-[32] showed increasing the order of accuracy of numerical method, will
increase computational efficiency. Therefore in this paper we intend to study ACU-WCNS with order
of accuracy higher than sixth.
Note : For construction of numerical schemes and examples, please consult the [Mansour &
Fardipour ]174.
5.2.11.1.3 References
[1] Jiang, G. S. and Shu, C. W. (1996) Efficient Implementation of Weighted ENO Schemes, Journal of
Computational Physics 126(1), 202-228.
[2] Deng, X. and Zhang, H. (2000) Developing High-Order Weighted Compact Nonlinear Schemes,
Journal of Computational Physics 165(1), 22-44.
[3] Harten, A., Engquist, B., Osher, S. and Chakravarthy, S. R. (1987) Uniformly high order accurate
essentially non-oscillatory schemes, III, Journal of Computational Physics 71(2), 231-303.
[4] Liu, X. D., Osher, S. and Chan, T. (1994) Weighted essentially non-oscillatory schemes, Journal of
Computational Physics 115(1), 200-212.
[5] Balsara, D. S. and Shu, C. W. (2000) Monotonicity Preserving Weighted Essentially Non-oscillatory
Schemes with Increasingly High Order of Accuracy, Journal of Computational Physics 160(2).
[6] Gerolymos, G. A., Senechal, D. and Vallet, I. (2009) Very-high-order WENO schemes, Journal of
Computational Physics 228(23), 8481–8524.
[7] Henrick, A. K., Aslam, T. D. and Powers, J. M. (2005) Mapped weighted essentially non-oscillatory
schemes-Achieving optimal order near critical points, Journal of Computational Physics 207(2).
[8] Borges, R., Carmona, M., Costa, B. and Don, W. S. (2008) An improved weighted essentially non-
oscillatory scheme for hyperbolic conservation laws, Journal of Computational Physics 227(6).
[9] Castro, M., Costa, B. and Don, W. S. (2011) High order weighted essentially non-oscillatory WENO-
Z schemes for hyperbolic conservation laws, Journal of Computational Physics 230(5), 1766-1792.
[10] Pirozzoli, S. (2002) Conservative hybrid compact-WENO schemes for shock-turbulence
interaction, Journal of Computational Physics 178(1), 81-117.
[11] Ren, X. Y., Liu, M, and Zhang, H. (2003) A characteristic-wise hybrid compact-WENO scheme for

174Kamyar Mansour¹, Kaveh Fardipour, “Adaptive Central-upwind Weighted Compact Non-linear Scheme with
Increasing Order of Accuracy”, Department of Aerospace Engineering, Amirkabir University of Technology,
Tehran, Iran.
 134

solving hyperbolic conservation laws, Journal of Computational Physics 192(2), 365-386.

[12] Kim, D. and Kwon, J. H. (2005) A high-order accurate hybrid scheme using a central flux scheme
and a WENO scheme for compressible flow field analysis, Journal of Computational Physics 210(2).
[13] Weirs, V. G. and Candler, G. V. (1997) Optimization of Weighted ENO Schemes for DNS of
Compressible Turbulence, AIAA Paper 97–1940.
[14] Wang, Z. J. and Chen, R. F. (2001) Optimized weighted essentially non-oscillatory schemes for
linear waves with discontinuity, Journal of Computational Physics 174(1), 381-404.
[15] Martín, M.P., Taylor, E.M., Wu, M. and Weirs, V.G. (2006) A bandwidth-optimized WENO scheme
for the effective direct numerical simulation of compressible turbulence, Journal of Computational
Physics, 220(1), 270-289.
[16] Tam, C. K. and Webb, J. C. (1993) Dispersion-relation-preserving finite difference schemes for
computational acoustics, Journal of computational physics, 107(2), 262-281.
[17] Zhuang, M. and Chen, R. F. (1998) Optimized upwind dispersion-relation-preserving finite
difference scheme for computational aeroacoustics. AIAA journal, 36(11), 2146-2148.
[18] Yamaleev, N. K. and Carpenter, M. H. (2009) A systematic methodology for constructing high-
order energy stable WENO schemes. Journal of Computational Physics, 228(11), 4248-4272.
[19] Hu, X. Y., Wang, Q. and Adams, N. A. (2010) An adaptive central-upwind weighted essentially
non-oscillatory scheme, Journal of Computational Physics 229(23), 8952-8965.
[20] Hu, X. Y. and Adams, N. A. (2011) Scale separation for implicit large eddy simulation. Journal of
Computational Physics, 230(19), 7240-7249.
[21] Deng, X. and Zhang, H. (2000) Developing high-order weighted compact nonlinear schemes.
Journal of Computational Physics, 165(1), 22-44.
[22] Nonomura, T., Iizuka, N. and Fujii, K. (2007) Increasing order of accuracy of weighted compact
nonlinear scheme, AIAA paper 2007-893.
[23] Zhang, S., Jiang, S. and Shu, C. W. (2008) Development of nonlinear weighted compact schemes
with increasingly higher order accuracy Journal of Computational Physics, 227(15), 7294-7321.
[24] Lele, S.K. (1992) Compact finite difference schemes with spectral-like resolution, Journal of
computational physics, 103(1), 16-42.
[25]Shu, C.W. (2009) High order weighted essentially nonoscillatory schemes for convection
dominated problems, SIAM review, 51(1), 82-126.
[26] Nonomura, T. and Fujii, K. (2013) Robust explicit formulation of weighted compact nonlinear
scheme, Computer & Fluids 85, 8-18.
[27] Nonomura, T., Iizuka, N. and Fujii, K. (2010) Freestream and vortex preservation properties of
high-order WENO and WCNS on curvilinear grids, Computers & Fluids, 39(2), 197-214.
[28] Nonomura, T. and Fujii, K. (2009) Effects of difference scheme type in high-order weighted
compact nonlinear schemes, Journal of Computational Physics, 228(10), 3533-3539.
[29] Sumi, T. and Kurotaki, T. (2015) A new central compact finite difference formula for improving
robustness in weighted compact nonlinear schemes, Computers & Fluids, 123, 162-182.
[30] Shi, J., Zhang, Y.T. and Shu, C.W. (2003) Resolution of high order WENO schemes for complicated
flow structures, Journal of Computational Physics, 186(2), 690-696.
[31] Zhang, Y.T., Shi, J., Shu, C.W. and Zhou, Y. (2003) Numerical viscosity and resolution of high-order
weighted essentially non-oscillatory schemes for compressible flows with high Reynolds numbers,
Physical Review E, 68(4), 046709.
[32] Berland, J., Bogey, C. and Bailly, C. (2008) A study of differentiation errors in large-eddy
simulations based on the EDQNM theory, Journal of Computational Physics, 227(18), 8314-8340.
[33] Shu, C.W. and Osher, S. (1988) Efficient implementation of essentially non-oscillatory shock-
capturing schemes. Journal of Computational Physics, 77(2), 439-471.
[34] Sod, G.A. (1978) A survey of several finite difference methods for systems of nonlinear
hyperbolic conservation laws, Journal of computational physics, 27(1), 1-31.
[35] Lax, P.D. (1954) Weak solutions of nonlinear hyperbolic equations and their numerical
 135

computation, Communications on pure and applied mathematics, 7(1), 159-193.

[36] Einfeldt, B., Munz, C.D., Roe, P.L. and Sjögreen, B. (1991) On Godunov-type methods near low
densities, Journal of computational physics, 92(2), 273-295.
[37] Toro, E. F. (2009) Riemann solvers and numerical methods for fluid dynamics: a practical
introduction, 3rd edn, Springer, Berlin, Germany.
[38] Shu, C.W. and Osher, S. (1989) Efficient implementation of essentially non-oscillatory shock-
capturing schemes, II, Journal of Computational Physics, 83(1), 32-78.
Uniform Flow Preserving Property of High Order Upwind Finite Difference Schemes on Generalized
Coordinate System
5.2.11.2 Case Study 2 – Uniform Flow Preserving Property of High Order Upwind Finite
Difference Schemes on Generalized Coordinate System

Authors : Taku Nonomura , Nobuyuki Iizuka, and Kozo Fujii

Title : Uniform Flow Preserving Property of High Order Upwind Finite Difference Schemes on
Generalized Coordinate System
Appeared in : Springer Link, Computational Fluid Dynamics 2006 pp 131-136.
Source : https://link.springer.com/chapter/10.1007/978-3-540-92779-2_18
Citation : Nonomura, T., Iizuka, N., & Fujii, K. (2009). Uniform Flow Preserving Property of High Order
Upwind Finite Difference Schemes on Generalized Coordinate System.
5.2.11.2.1 Introduction
Recently, interests of many scientists and engineers are shifting from Reynolds- averaged Navier-
Stokes simulations to direct numerical simulations, large eddy simulations and aero-acoustic
simulations. A major problem for these simulations is that they are computationally very expensive
because they require very high spatial resolution to guarantee accuracy of their results.
An approach to solve this problem is to use high-order schemes with less grid points. According to
[Balsara], resolution of eighth-order scheme is four times as high as second-order schemes in each
dimension while it costs only three times as much as second-order one. Thus, for three dimensional
turbulence problem, computational costs can be saved by factor of 44/3(=83) with eighth order
scheme. Moreover, according to [Shu Shu], when high- order scheme is implemented in multi-
dimensional problem, finite difference schemes, that can be constructed by dimension by dimension
procedure, costs ten times as little as finite element methods or finite volume methods which need
multi-dimensional reconstruction.
Therefore various high order finite difference schemes were recently pro- posed. These schemes
include compact scheme proposed by [Lele Le] and adapted to generalized coordinate by [Gaitonde
and Visba], weighted essential no oscillatory (WENO) scheme proposed by [Jiang et al JS96], and
weighted compact non-linear schemes (WCNS) proposed by [Deng and Zhang]. WENO or WCNS
scheme has been developed to solve flow-fields including discontinuity such as shock wave without
numerical oscillation.
When finite difference schemes are adapted to generalized coordinate form, capability of preserving
uniform flow is important because noise from unpreserved uniform flow hides very small oscillation,
such as turbulent flow structure or aero-acoustic wave. So far, for two-dimensional problem,
preservation of uniform flow has been achieved by using same stencils for both flow computation
and evaluation of metrics for transformation of coordinate. For three-dimensional problems1
preservation of uniform flow has been achieved by evaluating metrics of transformation of
coordinate based on finite volume method concepts. However1 present approach based on finite
volume method concepts do not suit for high order schemes.
[Gaitonde and Visba] proposed to use same stencil for both flow and metrics computations by
adopting the idea of conservative form of metric terms proposed by Thomas et al to preserve uniform
flow for compact scheme on three-dimensional generalized coordinate. On the other hands, most of
 136

researches of WENO and WCNS schemes almost have been demonstrated on only Cartesian
coordinate system. Therefore uniform preserving properties of these schemes on generalized
coordinate have not been studied.
Our interest of research is evaluation of noise from rocket engine plume1 whose main noise source
is turbulent flow including shock waves1 around complicated bodies. On computing such flow-fields
high order non-linear schemes such as WENO or WCNS on the generalized coordinate system are
necessary to save the computational costs. In this study uniform preserving and vortex preserving
properties of WENO and WCNS are investigated.
5.2.11.2.2 WENO and WCNS
WENO schemes use rth-order numerical flux fj+1/2 defined implicitly as following expression.

∂𝐟 1
( ) = (𝐟̃ − 𝐟̃j−1/2 ) + 𝛰(∆x r )
∂x j ∆x j+1/2
Eq. 5.19
It should be noted that this numerical flux is different from the physical flux f . The numerical flux is
constructed by weighted function of fluxes of computational nodes. Moreover numerical flux is
divided into f and . Weighted upwind-stencil is used for each divided flux. Thus in this Formulation
it is difficult to use same stencil for both flow computation and metrics evaluation. Detailed
procedure is explained in the reference of [Deng et al]. In this study, an explicit version of WCNS is
investigated because it is more efficient than the compact version]. This WCNS scheme uses rth-
order physical flux defined explicitly by the following expression

𝐟̂j+1/2 = 𝐟(𝐮(𝐱 j+1/2 )) + 𝛰(∆x r−1 )

Eq. 5.20
This physical flux can be constructed by interpolation of flow variables such as MUSCL procedure
instead of finite difference scheme [DZ00]. However as is physical flux simple difference scheme
achieves only second order as follows.

∂𝐟 1
( ) = (𝐟̂j+k+1/2 − 𝐟̂j−k−1/2 ) + 𝛰(∆x 2 )
∂x j ∆x
Eq. 5.21
Where the above expression is same as second order central difference on twice finer computational
grid. Thus to achieve higher order accuracy sth- order staggered finite difference schemes written as
following expression is necessary.

∂𝐟 1
( ) = ∑ bk (𝐟̂j+k+1/2 − 𝐟̂j−k−1/2 ) + Ο(∆x r ) + Ο(∆x s )
∂x j ∆x
Eq. 5.22
Since this high-order staggered finite difference scheme is linear scheme it seems that idea of
[Gaitonde and Visba] can be used to preserve uniform flow. Actual metric treatment to preserve
uniform flow is explained as follows. In WCNS cell-interface metric and high-order staggered finite
difference schemes are used. This cell-interface metric is high-order-interpolated from cell-node
metric. If conservative form of metric terms is computed by same cell-interface interpolation and
same high-order staggered finite difference schemes1 uniform flow seems to be preserved.
5.2.11.2.3 Metric Computation for Each Case
Uniform flow preserving properties and vortex preserving properties of four schemes are computed.
 137

WENO-G is a WENO scheme computed in generalized coordinate system using conservative form of
metric evaluated by compact schemes because it is difficult to use the same stencil both the flow
computation and the metrics evaluation. WENO-C is computed in Cartesian coordinate system as the
following expression with evaluating all terms numerically.
ξx ηx ζx Eξ Fξ Gξ
∂Q
+ [ξy ηy ζy ] [Eη Fη Gη ] = 0
∂t
ξz ηz ζz Eζ Fζ Gζ
Eq. 5.23
Computation using the non-conservative form written as above should preserve uniform flow.
However its computational cost is three times as expensive as WENO-G. WCNS-G1 is computed in
generalized coordinate system using the conservative form of metric evaluated by the same
interpolation and the same finite difference schemes. The metric evaluation should preserve the
uniform flow. WCNS-G2 is computed in generalized coordinate system using conservative form of
metric evaluated by compact schemes. The scheme will not preserve the uniform flow because the
stencil of the metric evaluation is different from that of the flow-computation.

Table 5.2 Results of uniform flow preserving test. is number of the grid points)

5.2.11.2.4 Uniform Flow and Vortex Preservation Tests

Uniform flow and vortex preservation
test has been carried out for each
scheme. The computational grid used
for uniform flow preservation test is
the same wavy grid as the one used in
the research of [Gaitonde and Visbal]
Figure 5.8 (a). Mach number of the
uniform flow in -direction u is 0.5. Thus
y-direction velocity v and z-direction
velocity w are expected to keep
machine-zero. TVD-Runge-Kutta
scheme is used for time integration.
Flow-fields after 100 steps are
examined. Results are shown in Table
5.2. The present result demonstrated
that WENO-C and WCNS-G1 pre- serve Figure 5.8 Results of vortex preserving test on the wavy
uniform flow while WENO-G and WCNS- grid
 138

G2 cannot. Then vortex preserving test is carried out for the above four schemes and MUSCL schemes
for reference. The computational grid used here is also almost same as the one used in the research
of Gaitonde and Visbal GV99] Figure 5.8 (b).

(a) unform flow preserving (b) vortex preserving

Figure 5.9 Testing Grid

An isentropic vortex is given where exact solution should keep its vortex strength. TVD-Runge-Kutta
scheme is used for time integration. Flow-fields after 1000 steps are examined. Pressure distribution
of the results are shown in Figure 5.9. Vortex center computed by WENO-G is moved and the
corresponding suction peak is much weakened. Vortex center of WENO-C keeps its position but its
suction peak is weakened as MUSCL On the other hand WCNS-G1 and WCNS-G2 keep vortex position
and their suction peak.
5.2.11.2.5 Conclusions
Uniform flow and vortex preserving properties on wavy grids of four high- order schemes have been
investigated while idea of [Gaitonde and Visba] is applied to WCNS scheme. The present results
show that WCNS computed in generalized coordinate system using conservative form of metric
evaluated by the same interpolation and the same finite difference schemes as flow computation has
very good properties for both uniform flow and vortex flow preservation.
5.2.11.2.6 References
Balsara, D. and Shu, C.W.: Journal of Computational Physics, 160, pp.405-452 2000)
Shu, C.W.: International Journal of Computational Fluid Dynamics 2003)
Lele, S. K.: , Journal of Computational Physics, 103, pp.16-44 1992)
Gaitonde, D. V. and Visble, M. R.: AIAA Paper 99-0557 1999)
Jiang, G.-S. and Shu, C.-W.: Journal of Computational Physics, 126, pp. 200-212 1996)
Deng, X. G. and Zhang, H.: , Journal of Computational Physics, 165, pp. 22-44 2000)
Deng, X. G. and Mao, M.: , Computational Fluid Dynamics Journal, 13, pp. 173-180 2004)
Deng, X. G., Liu, X. and Zhang, H.: , AIAA paper, 2005-5246, 2005)
Shu C.-W. and Osher, S.: , Journal of Computational Physics, 77, pp. 439-471 1988)
Shu C.-W. and Osher, S.: , Journal of Computational Physics, 83, pp.32- 78 1989)
 139

5.2.11.3 Case Study 3 - Increasing Order of Accuracy of Weighted Compact Non-Linear Scheme

Authors : Taku Nonomura, Nobuyuki Iizuka, and Kozo Fujii

Appeared in : 45th AIAA Aerospace Sciences Meeting and Exhibit 8 - 11 January 2007, Reno, Nevada
Source : AIAA 2007-893
Citation : Taku Nonomura, Nobuyuki Iizuka and Kozo Fujii. "Increasing Order of Accuracy of Weighted
Compact Non-Linear Scheme," AIAA 2007-893. 45th AIAA Aerospace Sciences Meeting and
Exhibit. January 2007.
The coefficients of higher order weighted compact nonlinear scheme (WCNS) and the resolutions of
the family of higher order WCNS are investigated. The coefficients of seventh and ninth order WCNS
are calculated by using MATHEMATICA. Seventh and ninth WCNS can resolve the discontinuity
without numerical oscillations as well as fifth order WCNS. Seventh and Ninth order WCNS have the
higher resolution on the one-dimensional shock-entropy interaction problem with few grid points
than fifth order WCNS. In addition, the resolutions of the explicit and tridiagonal and penta-diagonal
compact “cell-center to cell-node” difference schemes are investigated. The resolutions of these
schemes are almost same. Thus the explicit scheme which is cheapest one is good for WCNS. In
addition, seventh WCNS can solve the two-dimensional problem with the improvement of resolution,
while ninth order cannot solve it.
5.2.11.3.1 Order of WCNS
The order of WCNS is determined by the smaller order of the odd order accurate interpolation and
the even order accurate differentiation. Thus the combination of the interpolation and difference
schemes realize fourth ~ ninth order WCNS using coefficients shown in this paper. Following
sections, the order of cell-center to cell-node difference s is set 2r, where 2r-1 is the order of the
interpolation scheme. Thus WCNS scheme becomes (2r -1)-th order in this study.
5.2.11.3.2 2D Numerical Tests
In this section, two-dimensional numerical tests of higher order WCNS are carried out to verify the
robustness and the resolution of the new WCNS. In addition the effect of difference of flux splitting is
verified. For the test problem, [Woodward and [Colella’s] Mach 3 forwarding step problem is
computed.
In this test problem, only explicit “cell-center to cell-node” difference schemes are tested because the
resolution of explicit scheme is as high as that of compact scheme in previous section. WCNS5E,
WCNS7E and WCNS9E are computed. WCNS5E and WCNS7E can solve the problem without
numerical oscillation, while WCNS9E cannot due to numerical oscillation near boundary. Thus, in this
section, only WCNS5E and WCNS7E are computed. In addition, [SHUS] flux splitting and van Leer’s
FVS are used. [SHUS] is one of AUSM type splitting which uses the momentum flux of Roe’s FDS.
In this test case, 80 x 240 grid points are used and flow-field of time T=4.0 are compared. The results
of test problem are shown in Figure 5.10 where it shows density distribution of WCNS5E and
WCNS7E with SHUS and FVS.
Furthermore, Figure 5.10 show WCNS5E and WCNS7E can capture shock and resolve flow-field well
without the critical numerical oscillation. Comparing with WCNS5E and WCNS7E, WCNS7E have
better resolution at the contact line after triple points than WCNS5E. In addition, compared with
WCNS7E with SHUS and FVS, WCNS7E with [SHUS] have better resolution at the contact line after
triple points than WCNS7E with FVS while WCNS7E have much more small wiggles which is
numerical oscillations than WCNS7E with FVS. As a result, WCNS7E have better resolution than
WCNS5E and flux-splitting have great effect on the resolution and small wiggles even in seventh
order.
5.2.11.3.3 References
 140

[1] Woodward, P. and Colella, P.,

1984, “The numerical simulation of
two dimensional fluid numerical
computation,” Journal of
Computational Physics. 54, pp.115-
173
[2] Shima E. and Jounouchi T.,
1997, “Role of CFD in Aeronautical
Engineering (No14) –AUSM type
Upwind Schemes,” Proceedings of
the 14th NAL Symposium on
Aircraft Computational
Aerodynamics, pp. 7-12.

Figure 5.10 Density of Mach 3 forwarding step problem with

higher order WCNS; (a)WCNS5E with SHUS, (b)WCNS5E
with SHUS, (c) WCNS7E with SHUS, (d)WCNS7E with FVS
 141

5.2.11.4 Case Study 4 - A Novel Improvement of Compact Nonlinear Scheme For Simulating
Compressible Flows
Authors : Huaibao Zhang, Fan Zhang, Guangxue Wang, and Chunguang Xu
Title : A Novel Improvement 0f Compact Nonlinear Scheme For Simulating Compressible Flows
Appeared in : 11th International Symposium on Turbulence and Shear Flow Phenomena (TSFP11)
Southampton, UK, July 30 to August 2, 2019
Source : https://www.researchgate.net/publication/330934828
Weighted compact nonlinear schemes (WCNSs) were developed to improve the performance of
compact high order nonlinear schemes (CNSs) by utilizing the weighting technique originally
designed for WENO schemes, and non-oscillatory shock-capturing computation and high resolution
in smooth flow field are both achieved. Extensive efforts have been given focusing on improving the
performance of WCNSs ever since then. In this work, the ENO like stencil selection procedure of TENO
schemes is introduced for high-order nonlinear interpolations of midpoint variables, targeting
compact nonlinear schemes which fully abandon the oscillatory stencils crossing discontinuities and
eliminate numerical oscillations. The stencil selection procedure also directly applies smooth stencils
with their optimal weights, ensuring that the optimal numerical accuracy is fully recovered in smooth
flow field.
5.2.11.4.1 Introduction
While second-order accurate numerical methods along with RANS simulations are frequently
implemented by commercial codes, and currently dominate most industry related applications, high-
order CFD schemes are still expected candidates when much of the attention is given on problems
containing both discontinuities and complex flow structures, such as shock-boundary layer
interaction, Rayleigh-Taylor instability, and particularly the numerical simulation of compressible
turbulence flows. However, high-resolution simulations of compressible flows containing
discontinuities are still challenging even for current state-of-the-art numerical methods. Therefore,
the development of advanced high-order CFD schemes, targeting non-oscillatory computation for
shock-capturing and high-order accuracy in smooth flow field, is still an active topic with much to be
done.
Compact finite difference schemes have displayed spectral-like resolution [Lele (1992)], which are
therefore highly favored in the simulation of flow problems involving multi-scales phenomena.
Weighted Compact Nonlinear Schemes (WCNSs) [Deng & Zhang (2000)] are a family of high
resolution nonlinear shock-capturing schemes developed based on the key concept of nonlinear
weighting technique and cell-centered compact schemes. Past research has been performed on
WCNSs, notably by [Nonomura et al. (2010)]; [Nonomura & Fujii (2013)]; [Wong & Lele (2017)],
demonstrating that WCNSs have several advantages over the standard finite-difference Weighted
Essentially Non oscillatory (WENO) schemes [Liu et al. (1994)]:
(1) the resolution is slightly higher;
(2) the choice of flux schemes is more flexible, including Roe scheme [Roe (1981)], [van Leer scheme
van Leer (1982)], and AUSM scheme [Liou (1993)];
(3) WCNS performs well on freestream and vortex preservation properties on wavy grids.
The classical WCNS procedure consists of three steps [Deng & Zhang (2000)]:
(1) the node-to-midpoint weighted nonlinear interpolation of flow variables,
(2) the evaluation of fluxes at midpoints, and
(3) midpoint-to-node central flux differencing.
The flux differencing in the third step can be performed by using compact schemes or explicit
schemes. Despite that a compact scheme is used by the classical WCNS, later work of [Deng et al.
 142

(2005)] suggested that for a fourth or fifth-order WCNS, the weighted nonlinear interpolation in step
(1) dominates the resolution property, and explicit central differencing scheme is recommended due
to its simplicity of implementation and superior computation efficiency. Further work of [Nonomura
& Fujii (2009)] demonstrated that the type of flux differencing does not significantly change the
resolution, even for higher-order WCNSs. The classical WCNS uses the strategy by [Jiang & Shu
(1996)], and is therefore referred to as WCNS-JS.
Recently, a family of high-order targeted ENO schemes has been proposed by [Fu et al. (2016)]. One
of the essential feature of the TENO scheme is the use of ENO type stencil selection procedure. Instead
of merely focusing on developing improved nonlinear weights, the stencil selection technique is
incorporated in WCNS, and the so-called fifth-order TCNS is developed in this work.
Note : For a discussion about the numerical methods, please consult the [Zhang et al.]175.
5.2.11.4.2 Test Case : Rayleigh-Taylor Instability
Rayleigh-Taylor instability problem is used to examine the performance of the presented method.
Two sets of grids are used with the resolutions of 128×512 and 256×1024, respectively. The initial
conditions are given by

(2,0, −0.025acos(8πx), 1 + 2y) x ∈ [0.0, 0.25] , y ∈ [0.0,0.5]

(ρ, u, v, p) = {
(1,0, −0.025acos(8πx), 1 + 3/2) x ∈ [0.0, 0.25] , y ∈ [0.5,1.0]
Eq. 5.24
where a is the speed of sound, given by a =√γ p/ρ and a different γ = 5/3 is used for this specific case.
Reflecting boundary conditions are imposed at the left and right side of the domain, and constant
boundary conditions are given for the top and the bottom sides, in details

(1,0,0, 2.5) , y = 1 , ∀t, x

(ρ, u, v, p) = {
(2,0,0, 1.0) , y = 0 , ∀t, x
Eq. 5.25
Two source terms ρ, and ρv are added to the right hand side of the third and the fourth equation,
respectively. Density profiles at t = 1.95 are shown in Figure 5.11. It can be found that the
presented scheme captures much more abundant wave structures compared with WCNS-JS.
Moreover, TCNS on a coarse grid achieves similar or even better result compared with WCNS-JS
on a fine grid.
5.2.11.4.3 Conclusions
A novel compact nonlinear scheme, which applies the ENO-like stencil-selection procedure, is
introduced in this article. This method named as TCNS aims at achieving the optimal linear
interpolation in the node-to-midpoint interpolation step of the compact nonlinear scheme. The ADR
analysis shows that TCNS recovers the underlying linear scheme up to high wave numbers, even
using a relative large cut-off threshold. It is also demonstrated by using the linear advection case that
TCNS is capable of fully recovering the underlying optimal linear scheme in a smooth flow field ,by
directly applying the optimal linear weights in the node to midpoint interpolation procedure.
Moreover, significant improvements are obtained by TCNS in numerical tests such as the shock-
density wave interaction problem and the Rayleigh-Taylor instability case, which are characterized
by broadband fluctuations and rich small scales, respectively, see also a short 2D video of related

175 Huaibao Zhang, Fan Zhang, Guangxue Wang, and Chunguang Xu, “A Novel Improvement 0f Compact Nonlinear
Scheme For Simulating Compressible Flows”, 11th International Symposium on Turbulence and Shear Flow
Phenomena (TSFP11) Southampton, UK, 2019
 143

mesh movements, courtesy of

[Obeysekara]176 of Imperial College of
London. These significant improvements of
TCNS are mostly attributed to the use of the
ENO-like stencil-selection procedure, which
yields considerably low dissipation and
dispersion errors compared against WCNS-
JS.
5.2.11.4.4 References
Borges, Rafael, Carmona, Monique, Costa,
Bruno & Don, Wai Sun 2008 An improved
weighted essentially non oscillatory scheme
for hyperbolic conservation laws. Journal of
Computational Physics 227 (6), 3191–3211.
Deng, X., Liu, X., Mao, M. & Zhang, H. 2005
Investigation on weighted compact fifth-
order nonlinear scheme and applications to
complex flow. In 17th AIAA Computational
Fluid Dynamics Conference. Toronto, Ontario
Canada: AIAA Paper 2005-5246.
Deng, Xiaogang & Zhang, Hanxin 2000
Developing high order weighted compact
nonlinear schemes. Journal of Computational
Physics 165 (1), 22 – 44.
Fu, Lin, Hu, Xiangyu Y. & Adams, Nikolaus A. Figure 5.11 Rayleigh-Taylor instability problem: 30
2016 A family of high-order targeted ENO density contour lines ranging from 0.9 to 2.2.
schemes for compressible-fluid simulations.
Journal of Computational Physics 305, 333–
359.
Jiang, Guang-Shan & Shu, Chi-Wang 1996 Efficient implementation of weighted ENO schemes. Journal
of Computational Physics 126 (1), 202–228.
van Leer, Bram 1982 Flux-vector splitting for the Euler equations. In Eighth International Conference
on Numerical Methods in Fluid Dynamics (ed. E. Krause), Lecture Notes in Physics, vol. 170, pp. 507–
512. Aachen, Germany: Springer Berlin Heidelberg.
Lele, Sanjiva K. 1992 Compact finite difference schemes with spectral-like resolution. Journal of
Computational Physics 103 (1), 16 – 42.
Liou, M. S. 1993 On a new class of flux splitting. In Thirteenth International Conference on Numerical
Methods in Fluid Dynamics (ed. M. Napolitano & F. Sabetta), Lecture Notes in Physics, vol. 414, pp. 115–
119. Berlin, Heidelberg: Springer Berlin Heidelberg.
Liu, Xu Dong, Osher, Stanley & Chan, Tony 1994Weighted essentially non-oscillatory schemes. Journal
of Computational Physics 115 (1), 200–212.
Mao, Meiliang, Yan, Zhenguo, Liu, Huayong, Zhu, Huajun & Deng, Xiaogang 2015 Study of quasi-linear
spectral analysis method of high-order weighted nonlinear schemes. Acta Aerodynamica Sinica 33
(1), 1–9.
Nonomura, Taku & Fujii, Kozo 2009 Effects of difference scheme type in high-order weighted
compact nonlinear schemes. Journal of Computational Physics 228 (10), 3533 – 3539.

176https://www.linkedin.com/posts/kalyana-obeysekara-a33a5719_computationalfluiddynamics-physics-
deeplearning-activity-6694608748352282624-riWd
 144

Nonomura, Taku & Fujii, Kozo 2013 Robust explicit formulation of weighted compact nonlinear
scheme. Computers & Fluids 85, 8 – 18, international Workshop on Future of CFD and Aerospace
Sciences.
Nonomura, Taku, Iizuka, Nobuyuki & Fujii, Kozo 2010 Freestream and vortex preservation
properties of high order weno and wcns on curvilinear grids. Computers & Fluids 39 (2), 197 – 214.
Pirozzoli, Sergio 2006 On the spectral properties of shock capturing schemes. Journal of
Computational Physics 219 (2), 489 – 497.
Roe, Philip L 1981 Approximate Riemann solvers, parameter vectors, and difference schemes. Journal
of Computational Physics 43 (2), 357–372.
Shu, Chi-Wang & Osher, Stanley 1989 Efficient implementation of essentially non-oscillatory shock-
capturing schemes. J. Com. Phys. 83 (1), 32–78.
Wong, Man Long & Lele, Sanjiva K. 2017 High-order localized dissipation weighted compact
nonlinear scheme for shock- and interface-capturing in compressible flows. Journal of Computational
Physics 339, 179 – 209.
Yamaleev, Nail K. & Carpenter, Mark H. 2009 A systematic methodology for constructing high-order
energy stable WENO schemes. Journal of Computational Physics 228 (11), 4248 – 4272.
 145

5.3 Implementation of Boundary Conditions

We have one more item of business before finishing this section on finite difference quotients. We
pose the following question: What happens at a boundary? What type of differencing is possible when
we have only one direction to go, namely, the
direction away from the boundary? For example,
consider Figure 5.12, which illustrates a portion of a
boundary to a flow field, with the y axis perpendicular
to the boundary. Let grid point 1 be on the boundary,
with points 2 and 3 a distance Δy and 2Δy above the
boundary, respectively. We wish to construct a finite-
difference approximation for ∂u/∂y at the boundary.
It is easy to construct a forward difference as

∂u u2 − u1
( ) = + Ο(∆y)
∂y 1 ∆y
Eq. 5.26
which is of first-order accuracy. However, how do we
obtain a result which is of second-order accuracy? The
Figure 5.12 Grid Points at a Boundary
central difference fails us because it requires another
point beneath the boundary, such as illustrated as
point 2’ in Figure 5.12. Point 2’ is outside the domain of computation, and we generally have no
information about u at this point. In the early days of CFD, many solutions attempted to sidestep this
problem by assuming that u2’ = u2 . This is called the reflection boundary condition. In most cases it
does not make physical sense and is just as inaccurate, if not more so, than the forward difference
given by Eq. 5.26. So we ask the question again, how do we find a second-order-accurate finite
difference at the boundary? The answer is straightforward, as we will describe here. Moreover, we
will seize this occasion to illustrate an alternative approach to the construction of finite-difference
quotients-alternative to the Taylor's series analyses presented earlier. We will use a polynomial
approach, as follows. Assume at the boundary shown in Figure 5.12 that u can be expressed by the
polynomial
u = a + by + cy 2
Eq. 5.27
After applying the b,c, and rearranging, (see [Anderson]177), we obtain:

∂u −3u1 + 4u2 − u3
( ) =
∂y 1 2∆y
Eq. 5.28
Applied successively to the grid points in Figure 5.12. Eq. 5.27 yields at grid point 1 where y = 0.
resulting in Eq. 5.28. It is a one-sided finite-difference expression for the derivative at the boundary
called one-sided because it uses information only on one side of the grid point at the boundary,
namely, information only above grid point 1 in Figure 5.12. Also, it was derived using a polynomial
expression, namely, Eq. 5.27, rather than a Taylor series representation. This illustrates an
alternative approach to the formulation of finite-difference quotients; indeed, all our previous results
as summarized could have been obtained using this polynomial approach. It remains to show the

177John D. Anderson, Jr., “Computational Fluid Dynamics-The Basics With Applications”, Department of
Aerospace Engineering University of Maryland, McGraw-Hill Series in Aeronautical and Aerospace Engineering,
1995.
 146

order of accuracy of Eq. 5.28. Here, we have to appeal to a Taylor series again. Consider a Taylor
series expansion about the point 1.

∂u ∂2 u y 2 ∂3 u y 3
u(y) = u1 + ( ) y + ( 2 ) + ( 3) +⋯
∂y 1 ∂y 1 2! ∂y 1 3!
Eq. 5.29
Compare Eq. 5.29 and Eq. 5.28. Our assumed polynomial expression in Eq. 5.28 is the same as
using the first three terms in the Taylor series. Hence, Eq. 5.28 is of O(Δy)3. Thus, we can write from
Eq. 5.28 as
𝜕𝑢 −3𝑢1 + 4𝑢2 − 𝑢3
( ) = + Ο(∆𝑦)2
𝜕𝑦 1 2∆𝑦
Eq. 5.30
This is our desired second-order-accurate difference quotient at the boundary. The Eq. 5.30 is
called one-sided differences, because they express a derivative at a point in terms of dependent
variables on only one side of that point. Moreover, these equations are general; i.e., they are not in
any way limited to application just at a boundary; they can be applied at internal grid points as well.
It just so happens that we have taken advantage of our discussion of Finite difference quotients at a
boundary to derive such one-sided differences. Of course, as we have seen here, one-sided differences
are essentially mandatory for a representation of a derivative at a boundary, but such one-sided
differences simply offer another option when applied internally within the domain of the overall
calculations. Furthermore, Eq. 5.30 displays a one-sided finite difference of second-order accuracy;
many other one-sided difference formulas for a derivative at a point can be derived with higher
orders of accuracy using additional grid points to one side of that point. In some CFD applications, it
is not unusual to see four- and five-point one-sided differences applied at a boundary. This is
especially true for viscous flow calculations. In such calculations, the shear stress and heat transfer
at the wall, due to a flow over that wall, are of particular importance.
 147

6 Finite Volume Method (FVM)

6.1 Discretization of Control Volume
Generally, there are three types of discretization for control volume schemes. The two most popular
approaches are the so-called cell-centered finite volume (CCFV) method and the vertex-centered
finite volume (VCFV) method, are described below. The CCFV approach defines the solution at the
center of the mesh elements (i.e. cells) such that their values represent cell averages of the unknown
quantities. Figure 6.1 represents a CCFV approach where numbers 0-12 and letters A-L denote grid
nodes and primal cell centers,
respectively. The control volume for a
node centered discretization (VCFV)
around the grid node 0 is shaded. The
control volume for a cell-centered
discretization around the cell center A
is hashed178.
Several techniques to accurately
compute the gradient of the solution at
the element faces, based on node
averaging or least squares, have been
proposed and compared. In all cases, a
reconstruction of the gradient of the
solution is required to guarantee
second-order convergence of the
solution error. This is crucial to
guarantee a first-order convergence of
the solution gradient, which is Figure 6.1 Control-Volume Partitioning For Finite-Volume
required to accurately compute Discretization
engineering quantities of interest (e.g.
lift and drag). The accuracy of the
reconstruction is heavily dependent on the quality of the mesh and some approaches fail to provide
a second-order scheme on highly stretched and deformed grids [Sevilla, R]179.
The VCFV strategy defines the solution at the mesh nodes. A control volume is constructed around
each node by using the centroid of the mesh elements and mid-edge points (and face centroids in
three dimensions). The control volumes form a non-overlapping set of subdomains that cover the
whole domain and form the so-called dual mesh. The resulting approximation is locally piecewise
constant on each dual element where the values of the unknowns represent control volume averages.
Similarly to CCFV scheme, the VCFV method requires the reconstruction of the gradient of the
solution at each dual face.
In parallel to the development of FV schemes, a great effort was dedicated during the 1970s to the
application of finite element methods to CFD problems. The difficulties encountered due to the
convection dominated nature of many fluid flow problems prompted the development of the so-
called stabilized finite element techniques20 and discontinuous Galerkin methods. More recently, a
great effort has been dedicated to reinterpret finite volume schemes within a continuous and
discontinuous finite element framework. [Morton and Sonar] motivate their exposition of finite

178 Boris Diskin, James L. Thomas, Eric J. Nielsen, Hiroaki Nishikawa, Jeffery A. White, “Comparison of node-
centered and cell-centered unstructured finite-volume discretization: viscous fluxes”,
https://ntrs.nasa.gov/search.jsp?R=20110002899 2020-05-16T14:49:28+00:00Z
179 Ruben Sevilla, Matteo Giacomini, and Antonio Huerta, “A face-centered finite volume method for second-order

elliptic problems”, International Journal for Numerical Methods in Engineering · August 2018.
 148

volume schemes as Petrov-Galerkin finite element methods flowing to the edibility the latter
approaches show in terms of approximation using unstructured meshes and the solid theoretical
framework developed for their analysis.
Here, an alternative to the discussed finite volume strategies is proposed by defining the unknowns
over the faces of the mesh. As for CCFV and VCFV, the resulting face-centered finite volume (FCFV)
method may be interpreted as a lowest-order finite element method. More precisely, FCFV is derived
from the recently proposed hybrid sable discontinuous Galerkin (HDG) method by Cockburn and co-
workers by imposing a constant degree of approximation. As such, the method requires the solution
of a global system of equations equal to the total number of element faces. The solution and its
gradient in each element are then recovered by solving a set of independent element-by element
problems.
The FCFV method provides first-order accuracy on both the solution and its gradient without the
need to perform a reconstruction of the gradients to accurately compute the fluxes at the element or
control volume boundary. Therefore, its accuracy is not compromised in the presence of highly
stretched or distorted elements. In addition, due to the definition of the unknowns on the element
faces, the global system of equations that must be solved, provides a less degree of coupling of the
information when compared to other finite volume schemes. The application to scalar and vector
second-order elliptic problems is considered, namely the Poisson and the Stokes problems
respectively. [Sevilla, R]. For fundamentals of the second-order FCFV method, please consult the
work by [Giacomini & Sevilla]180.

6.2 Cell Centered Finite Volume (CCFV) Discretization

The Finite (Control) Volume Method (FVM), examines not only the governing PDE but also the
physical laws or the conservation statement which the PDE represents. The governing PDE could be
transformed through divergence form (conservative) by employing the Gauss Divergence Theorem.
This enables the preservation of conservation quantity within each individual control volume. By
adding these individual control volumes, the overall conservation of the domain would be preserved.
It is difficult to appreciate the subtle differences which may occur in different representation using
FD or FV. In many cases, especially for simple, linear equations, the resulting discretization is the
same. The differences are more pronounced when irregular meshes are used or in some boundary
conditions. In general, the CV method is preferred over FD, due to the fact that they conserve the
physical quantities.
∂
∭ 𝐐 dV + ∭ 𝐅dV = 0
∂t V V
Eq. 6.1
where Q is the vector of conserved variables, F is the vector of fluxes (see Euler equations or Navier
Stokes equations), V is the volume of the control volume element, and is A the surface area of the
control volume element. The time-dependent term and the body force term are assumed constant
over the volume of the cell. Using the divergence theorem and integrating over the control surface,
we obtain semi-discrete form as

∂𝐐
V + ∑(𝐅dA)NS = 0
∂t
NS
Eq. 6.2
Where NS represents the number of surfaces surrounding the CV or cells. In general, the FVM

180Giacomini, M. and Sevilla, R.,“A second-order face-centered finite volume method on general meshes with
automatic mesh adaptation”, arXiv:2005.01663v1 [math.NA] 4 May 2020.
 149

involves the following steps181:

1. Decomposition of the problem domain into control volumes.
2. Formulation of integral balance equations for each control volume.
3. Approximation of integrals by numerical integration.
4. Approximation of function values and derivatives by interpolation with nodal values.
5. Assembling and solution of discrete algebraic system.
Under certain conditions FD and FV schemes yields the same results. An example would be a 2D
steady state conduction with Dirichlet boundaries.

6.3 Finite Volume Method using a 1D Example

The key to the method is that the integral form of the conservation law

∂
∫ U d Ω + ∫ ∇ . 𝐅 dΩ = 0
∂t Ω Ω
Eq. 6.3
can be rewritten, using the Gauss Divergence Theorem.

∂
∫ U dΩ + ∮ 𝐅. 𝐧 dΓ = 0 where 𝐅𝐤∗ is numerical flux
∂t Ω ⏟Ω
∑faces 𝐅𝐤∗ .𝐧𝐤
Eq. 6.4
For unsteady 1D problem it becomes

XR
∂ ∂
∫ U d Ω + ∫ 𝐅x dx = ∫ U d Ω + (𝐅R - FL ) = 0
∂t Ω XL ∂t Ω
Eq. 6.5
In other words, the rate of change of mass in the control volume is equal to the net mass flux through
its boundary. For simplicity, a forward Euler discretization of the time derivative will be considered,
leading to

Δt
Uin+1 − Uin = ∑ 𝐅k∗ . 𝐧k where Vi is control volume of cell
Vi
faces
Eq. 6.6
Steady state computations provide a special case. Usually the above is used to iterate to the steady
state. There are many convergence acceleration techniques182. For 3D using

181 Computational Engineering, Introduction to Numerical Methods, http://www.springer.com/978-3-540-

30685-6.
182 Matthew Hubbard, “Finite Volume Schemes: A Tutorial”, University of Leeds, UK.
 150

Δt ∗ ∗ Δt
Uin+1 − Uin = − (𝐅i+1 − 𝐅i−1 )− (𝐆∗j+1 − 𝐆∗j−1 )
∆x 2
,j,k
2
,j,k ∆𝑦 i, 2 ,k i,
2
,k
Δt
− (𝐇 ∗ k+1 − 𝐇 ∗ k-1 )
∆z i,j, 2 i,j,
2
Eq. 6.7
Dimensional splitting is often used here to improve speed and stability, though accuracy may
diminish (Strang, 1968). Clearly conservation is satisfied due to

∑ ( ∑ 𝐅k∗ . 𝐧k ) = ∑ 𝐅k∗ . 𝐧k
Volumes faces i boundary
Eq. 6.8
so the net flux equal to the contribution from the boundary.

6.4 Unsteady 2D Poisson Equation

The unsteady form of 2D Poisson equation can be written as

∂ϕ ∂2 ϕ ∂2 ϕ
− ( 2 + 2) = S
∂t ∂x ∂y
∂ϕ̄i 1 1
= ∑ ∇ϕ .n̂ds + ∬ SdA = −R(ϕ̄)
∂t Ai Vi CVi
Boundary
Eq. 6.9
This unsteady equation can be cast in a finite-volume formulation as above. The flux integral,
representing the spatial discretization and the source-term control-volume average, forms the
residual of the scheme.

6.5 General Steady 2D Transport Equation in Finite Volume

We will outline in detail the individual steps following closely the development in [Schafer]183,
(except the solution of algebraic systems). We will do this by example for the general stationary
transport, namely,
∂ ∂φ
(ρvi φ − Γ ) = f
∂xi ∂xi
Eq. 6.10
for some problem domain Ω. The starting point for a finite-volume discretization is a decomposition
of the problem domain Ω into a finite number of subdomains Vi (i = 1, . . . , N), called control volumes
(CVs), and related nodes where the unknown variables are to be computed. The union of all CVs
should cover the whole problem domain. In general, the CVs also may overlap, but since this results
in unnecessary complications we consider here the non-overlapping case only. Since finally each CV
gives one equation for computing the nodal values, their final number (i.e., after the incorporation of
boundary conditions) should be equal to the number of CVs. Usually, the CVs and the nodes are
defined on the basis of a numerical grid, which, for instance, is generated. In order to keep the usual

183Schafer, M., “Computational Engineering - Introduction to Numerical Methods”, 2006, X, 321 p. 204 ill ..
Softcover ISBN: 978-3-540-30685-6.
 151

terminology of the FVM, we always talk of volumes (and their surfaces), although strictly speaking
this is only correct for the three-dimensional case. For one-dimensional problems the CVs are
subintervals of the problem interval and the nodes can be the midpoints or the edges of the
subintervals (see Figure 6.2). For three-dimensional problems on the basis of hexahedral or
tetrahedral grids similar techniques as in the two-dimensional case can be applied. By integration of
(Eq. 6.10) over an arbitrary control volume V and application of the Gauss integral theorem, one
obtains:
𝜕𝜑
∫ (ρv𝑖 φ − Γ ) 𝑛 dS = ∫ f dV
𝑆 𝜕𝑥𝑖 𝑖 𝑉
Eq. 6.11
where S is the surface of the CV and ni are the
components of the unit normal vector to the
surface. The integral balance equation (Eq.
6.10) constitutes the starting point for the
further discretization of the considered problem
with an FVM. As an example we consider
quadrilateral CVs with a cell-oriented
arrangement of nodes (a generalization to
arbitrary polygons poses no principal
difficulties). For a general quadrilateral CV we
use the notations of the distinguished points
(midpoint, midpoints of faces, and edge points)
and the unit normal vectors according to the so-
called compass notation as indicated in Figure
6.3 (a). The midpoints of the directly
neighboring CVs we denote – again in compass Figure 6.2 Definition of CVs and Nodes for
Triangular Grids with Donald Polygons
notation – with capital letters S, SE, etc. (see
Figure 6.3 (b)).

(a) Quadrilateral control volume (b) Notations for neighboring

with notations control volumes
Figure 6.3 Finite Volume in a 2D Hexahedra Cell – Courtesy of [Schafer]

The surface integral in (Eq. 6.11) can be split into the sum of the four surface integrals over the cell
faces Sc (c = e ,w, n, s) of the CV, such that the balance equation (Eq. 6.11) can be written equivalently
 152

in the form

∂φ
∑ ∫ ( ρv
⏟ iφ − Γ ) nci dSc = ∫ f dV
Sc ⏟ ∂xi V
c convective
diffusivetive
Eq. 6.12
The expression (Eq. 6.12) represents a balance equation for the convective FCC and diffusive fluxes
and FDC through the CV faces, respectively. For a complete description of the analysis, reader should
check out the [Schafer]184.
6.5.1 Approximation of Surface and Volume Integrals
We start with the approximation of the surface integrals in (Eq. 6.12), which for a cell-centered
variable arrangement suitably is carried out in two steps:
(1) Approximation of the surface integrals (fluxes) by values on the CV faces.
(2) Approximation of the variable values at the CV faces by node values.
The integral can be approximated in different ways by involving more or less values of the integrand
at the CV face. The simplest possibility is an approximation by just using the midpoint of the face:

∫ wi nei dSe ≈ g e δSe

Sc
Eq. 6.13
where we denote with ge = weinei the normal component of w at the location e. With this, one obtains
an approximation of 2nd order (with respect to the face length δSe) for the surface integral, which
can be checked by means of a Taylor
series expansion. The integration Name Formula Order
formula (Eq. 6.13) corresponds to the MidPoint Rule δSege 2
midpoint rule known from numerical Trapezoid Rule δSe(gne+gse)/2 2
integration. Other common integration Simpson Rule δSe(gne+4ge+gse)/6 4
formulas, that can be employed for
such approximations are, for instance, Table 6.1 Approximations for Surface Integrals Over the
the trapezoidal rule and the Simpson Face Se - Courtesy of [Schafer]
rule. The corresponding formulas are
summarized in Table 6.1 with their respective orders (with respect to δSe). For instance, by applying
the midpoint rule for the approximation of the convective and diffusive fluxes through the CV faces
in (Eq. 6.12), we obtain the approximations:

∂φ
FCC ≈ ⏟
ρvi nci g e δSe φc and FCD ≈ −Γnci δSc ( )
∂xi c
ṁ c
Eq. 6.14
where, for simplicity, we have assumed that vi, ρ, and Г are constant across the CV. m˙c denotes the
mass flux through the face Sc. Inserting the definition of the normal vector, we obtain, for instance,
for the convective flux through the face Se, the approximation

184Schafer, M., “Computational Engineering - Introduction to Numerical Methods”, 2006, X, 321 p. 204 ill ..
Softcover ISBN: 978-3-540-30685-6.
 153

FCC ≈ ṁe φe = ρ[v1 (yne − yse ) − v2 (xne − xse )]

Eq. 6.15
Before we turn to the further discretization of the fluxes, we first deal with the approximation of the
volume integral in (Eq. 6.12), which normally also is carried out by means of numerical integration.
The assumption that the value fP of f in the CV center represents an average value over the CV leads
to the two-dimensional midpoint rule:

1
∫ fdv ≈ fp δv = fp |(xse − xnw )(yne − ysw ) − (xne − xsw )(yse − ynw )|
⏟
2
v δv
Eq. 6.16
An overview of the most common two-dimensional integration formulas for Cartesian CVs with the
corresponding error order (with respect to δV). It should be noted that the formulas for the two
dimensional numerical integration can be used to approximate the surface integrals occurring in
three dimensional applications. For three-dimensional volume integrals analogous integration
formulas as for the two-dimensional case are available. In summary, by applying the midpoint rule
(to which we will restrict ourselves) we now have the following approximation for the balance
equation (Eq. 6.12):

∂φ
∑ ṁc φc − ∑ Γnci δSc ( )
∂xi c
⏟c ⏟c
conv. fluxes diff. fluxes
= f⏟
p δv
source
Eq. 6.17
In the next step it is necessary to approximate
the function values and derivatives of φ at the
CV faces occurring in the convective and
diffusive flux expressions, respectively, by
variable values in the nodes (here the CV
centers). In order to clearly outline the essential
principles, we will first explain the
corresponding approaches for a two-
dimensional Cartesian CV as indicated in Figure Figure 6.4 Cartesian Control Volume with
Notations in Formulas Analogous to Finite-
6.4. In this case the unit normal vectors nc along
Difference Methods – Courtesy of [Schafer]
the CV faces.
6.5.2 Discretization of Convective Fluxes
For the further approximation of the convective fluxes FCC , it is necessary to approximate φc by
variable values in the CV centers. In general, this involves using neighboring nodal values φE, φP, . . .
of φc. The methods most frequently employed in practice for the approximation will be explained in
the following, where we can restrict ourselves to one-dimensional considerations for the face Se,
since the other faces and the second (or third) spatial dimension can be treated in a fully analogous
way. Traditionally, the corresponding approximations are called differencing techniques. Strictly
speaking, these are interpolation techniques.
6.5.3 Central Differences
For the central differencing scheme (CDS) φe is approximated by linear interpolation with the values
 154

in the neighboring nodes P and E (see Figure 6.5):

𝑥𝑒 − 𝑥𝑝
φ𝑒 ≈ γ𝑒 φ𝐸 + (1 − γ𝑒 )φ𝑝 where γ𝑒 =
𝑥𝐸 − 𝑥𝑝
Eq. 6.18
The approximation (Eq. 6.18) has, for an
equidistant grid as well as for a no equidistant
grid, an interpolation error of 2nd order. This
can be seen from a Taylor series expansion of
φ around the point xP. By involving additional
grid points, central differencing schemes of
higher order can be defined. (see
[Schafer]185).
6.5.4 Upwind Techniques
The simplest upwind method results if φ is
approximated by a step function. Here, φe is
determined depending on the direction of the
mass flux as follows (see Figure 6.6):
Figure 6.5 Approximation of φe with CDS Method
φe = φp if ṁe > 0 – Courtesy of [Schafer]

φe = φE if ṁ e < 0
Eq. 6.19
This method is called upwind
differencing scheme (UDS). Using a
Taylor series expansion of φ around the
point xP, evaluated at the point xe, has an
interpolation error of 1st order. The
leading error term in the resulting
approximation of the convective flux FC
e becomes

∂φ
m
⏟ ̇ e (x e − x p ) ( )
∂x p
Γnum
Eq. 6.20
The error caused by this is called
artificial or numerical diffusion, since
the error term can be interpreted as a
diffusive flux. The coefficient Гnum is a Figure 6.6 Mass flux Dependent Approximation of φe
measure for the amount of the numerical with UDS Method
diffusion. If the transport direction is
nearly perpendicular to the CV face, the
approximation of the convective fluxes resulting with the UDS method is comparably good (the
derivative (∂φ/∂x)P is then small). Otherwise the approximation can be quite inaccurate and for large
mass fluxes (i.e., large velocities) it can then be necessary to employ very fine grids (i.e., xe − xP very

185Schafer, M., “Computational Engineering - Introduction to Numerical Methods”, 2006, X, 321 p. 204 ill ..
Softcover ISBN: 978-3-540-30685-6.
 155

small) for the computation in order to achieve a solution with an adequate accuracy. The
disadvantage of the relatively poor accuracy is confronted by the advantage that the UDS method
leads to an unconditionally bounded solution algorithm.
An upwind approximation frequently employed in practice is the quadratic upwind interpolation,
which in the literature is known as the QUICK method (Quadratic Upwind Interpolation for
Convective Kinematics). Here, a quadratic polynomial is fitted through the two neighboring points
P and E, and a third point, which is located upstream (W or EE depending on the flow direction).
Evaluating this polynomial at point e one obtains the approximation (see also Figure 6.7):

Figure 6.7 Mass flux dependent approximation of φe with QUICK method – Courtesy of [Schafer]

For an equidistant grid one has:

φe = a1 φE − a2 φw + (1 − a1 + a2 )φp if ṁe > 0

where
φe = b1 φp − b2 φEE + (1 − b1 + b2 )φE if ṁ e < 0
(2 − γw )γ2e (1 − γe )(1 − γw )2
a1 = , a2 =
1 + γe − γw 1 + γe − γw
2
(1 + γw )(1 − γe ) γ2ee γe
b1 = , b2 =
1 + γee − γe 1 + γee − γe
Eq. 6.21
In this case the QUICK method possesses an interpolation error of 3rd order. However, if it is used
together with numerical integration of only 2nd order the overall flux approximation also is only of
2nd order, but it is somewhat more accurate than with the CDS method.
6.5.5 Flux-Blending Technique
The principal idea of flux-blending, which goes back to [Khosla und Rubin] (1974), is to mix different
approximations for the convective flux. In this way one attempts to combine the advantages of an
accurate approximation of a higher order scheme with the better robustness and boundedness
properties of a lower order scheme (mostly the UDS method). To explain the method we again
consider exemplarily the face Se of a CV. The corresponding approximations for φe in the convective
flux FCe for the two methods to be combined are denoted by φMLe and φMHe , where ML and MH are
 156

the lower and higher order methods, respectively. The approximation for the combined method
reads:
φe ≈ βφMH ML
e + (1 − β)φe
Eq. 6.22
From Eq. 6.22, for β = 0 and β = 1 the methods ML and MH, respectively, result. However, it is
possible to choose for β any other value between 0 and 1, allowing to control the portions of the
corresponding methods according to the needs of the underlying problem. However, due to the loss
in accuracy, values β < 1 should be selected only if with β = 1 on the given grid no “reasonable”
solution can be obtained and a finer grid is not possible due to limitations in memory or computing
time. Also, if β = 1 (i.e., the higher order method) is employed, it can be beneficial to use the splitting
according to Eq. 6.22 in order to treat the term bφ,e β “explicitly” in combination with an iterative
solver. This means that this term is computed with (known) values of φ from the preceding iteration
and added to the source term. This may lead to a more stable iterative solution procedure, since this
(probably critical) term then makes no contribution to the system matrix, which becomes more
diagonally dominant. It should be pointed out that this modification has no influence on the
converged solution, which is identical to that obtained with the higher order method MH alone.
6.5.6 Discretization of Diffusive Fluxes
For the approximation of
diffusive fluxes it is necessary to
approximate the values of the
normal derivative of φ at the CV
faces by nodal values in the CV
centers. For the east face Se of the
CV, which we will again consider
exemplarily, one has to
approximate (in the Cartesian
case) the derivative (∂φ/∂x)e. For
this, difference formulas as they
are common in the framework of
the finite-difference method can
be used. The simplest
approximation one obtains when
using a central differencing
formula
∂φ φE − φp
( ) ≈ Figure 6.8 Central Differencing Formula for Approximation of 1st
∂x e xE − xp Derivative at CV face – Courtesy of [Schafer]
Eq. 6.23
which is equivalent to the assumption that φ is a linear function between the points xP and xE (see
Figure 6.8). For the discussion of the error of this approximation, we consider the difference of the
Taylor series expansion around xe at the locations xP and xE:

2
∂φ φE − φp (xe − xp ) − (xE − xe )2 ∂2 φ
( ) = + ( 2)
∂x e xE − xp 2(xE − xp ) ∂x e
3
(xe − xp ) − (xE − xe )3 ∂3 φ
− ( 3 ) + TH
6(xE − xp ) ∂x e
 157

Eq. 6.24
One can observe that for an equidistant grid an error of 2nd order results, since in this case the
coefficient in front of the second derivative is zero. In the case of non-equidistant grids, one obtains
by a simple algebraic rearrangement that this leading error term is proportional to the grid spacing
and the expansion rate ξe of neighboring grid spacings:

(1 − 𝜉𝑒 )(𝑥𝑒 − 𝑥𝑝 ) ∂2 φ xE − xe
( 2 ) with ξ𝑒 =
2 ∂x e xe − xp
Eq. 6.25
This means that the portion of the 1st order error term gets larger the more the expansion rate
deviates from 1. This aspect should be taken into account in the grid generation such that neighboring
CVs do not differ that much in the corresponding dimensions. formula for approximation of 1st
derivative at CV face One obtains a 4th order approximation of the derivative at the CV face for an
equidistant grid by
∂φ 1
( ) ≈ (φw − 27φp + 27φE − φEE )
∂x e 24∆x
Eq. 6.26
which, for instance, can be used together with the Simpson rule to obtain an overall approximation
for the diffusive flux of 4th order. Although principally there are also other possibilities for
approximating the derivatives (e.g., forward or backward differencing formulas), in practice almost
only central differencing formulas are employed, which possess the best accuracy for a given number
of grid points involved in the discretization.
Problems with boundedness, as for the
convective fluxes, do not exist. Thus, there is no
reason to use less accurate approximations. For
CVs located at the boundary of the problem
domain, it might be necessary to employ
forward or backward differencing formulas
because there are no grid points beyond the
boundary.
6.5.7 Non-Cartesian Grids
The previous considerations with respect to the
discretization of the convective and diffusive
fluxes were confined to the case of Cartesian Figure 6.9 Central difference approximation
grids. In this section we will discuss necessary of convective fluxes for non-Cartesian control
modifications for general (quadrilateral) CVs. volumes – Courtesy of [Schafer]
For the convective fluxes, simple
generalizations of the schemes introduced (e.g., UDS, CDS, QUICK, . . . ) can be employed for the
approximation of φc. For instance, a corresponding CDS approximation for φe reads:

|xe̅ − xp | |xe̅ − xp |
φ𝑒 = φ𝐸 + φ𝑝
|xE − xp | |xE − xp |
Eq. 6.27
where xē is the intersection of the connecting line of the points P and E with the (probably extended)
CV face Se (see Figure 6.9). For the convective flux through Se this results in the following
approximation:
 158

ṁ𝑒
F𝑒𝐶 ≈ (|xe̅ − xp |φ𝐸 + |xE − xe̅ |φ𝑝 )
|xE − xp |
When the grid at the corresponding face has a “kink”, an additional error results because the points
x˜e and xe do not coincide (see Figure 6.9). This
aspect should be taken into account for the grid
generation. Let us turn to the approximation of the
diffusive fluxes, for which farther reaching
distinctions to the artesian case arise as for the
convective fluxes. Here, for the required
approximation of the normal derivative of φ in the
center of the CV face there are a variety of different
possibilities, depending on the directions in which
the derivative is approximated, the locations where
the appearing derivatives are evaluated, and the
node values which are used for the interpolation.
As an example we will give here one variant and
consider only the CV face Se. Since along the normal Figure 6.10 Approximation of diffusive fluxes
for non-Cartesian control volumes – Courtesy of
direction in general there are no nodal points, the
[Schafer]
normal derivative has to be expressed by
derivatives along other suitable directions. For this
we use here the coordinates ξ and ῆ defined according to Figure 6.10. The direction ˜ξ is determined
by the connecting line between points P and E, and the direction ˜η is determined by the direction of
the CV face. Note that ˜ξ and ˜η, because of a distortion of the grid, can deviate from the directions ξ
und η, which are defined by the connecting lines of P with the CV face centers e and n. The larger
these deviations are, the larger the discretization error becomes. This is another aspect that has to
be taken into account when generating the grid.
A coordinate transformation (x, y) → (˜ξ, ˜η) results for the normal derivative in the following
representation in (see [Schafer]186), which results for the Jacobi determinant in the approximation
where ψ denotes the angle between the direction ˜ξ and ne (see Figure 6.10). ψ is a measure for the
deviation of the grid from orthogonality (ψ = 0 for an orthogonal grid). The derivatives with respect
to ˜ξ and ˜η can be approximated in the usual way with a finite-difference formula. For example, the
use of a central difference of 2nd order gives:

∂φ φE − φp ∂φ φne − φse
≈ and ≈
∂ξ̅ |xE − xp | ∂η̅ δSe
Eq. 6.28
Inserting the approximations, Eq. 6.28, and using the component representation of the unit normal
vector ne we finally obtain the following approximation for the diffusive flux through the CV face Se:

FeD ≈ De (φE − φp ) + Ne (φne − φse ) where

2 2
Γ [(yne − yse ) + (xne − xse ) ]
De =
(xne − xse )(yE − yp ) − (xE − xp )(yne − yse )

186Schafer, M., “Computational Engineering - Introduction to Numerical Methods”, 2006, X, 321 p. 204 ill ..
Softcover ISBN: 978-3-540-30685-6.
 159

Γ [(𝑦𝑛𝑒 − 𝑦𝑠𝑒 )(𝑦𝐸 − 𝑦𝑝 ) + (𝑥𝑛𝑒 − 𝑥𝑠𝑒 )(𝑥𝐸 − 𝑥𝑝 )]

Ne =
(𝑥𝐸 − 𝑥𝑝 )(𝑦𝑛𝑒 − 𝑦𝑠𝑒 ) − (𝑥𝑛𝑒 − 𝑥𝑠𝑒 )(𝑦𝐸 − 𝑦𝑝 )
Eq. 6.29
The coefficient Ne represents the portion that arise due to
the non-orthogonality of the grid. If the grid is orthogonal,
ne and xE − xP have the same direction such that Ne = 0. The
coefficient Ne (and the corresponding values for the other
CV faces) should be kept as small as possible. The values
for φne and φse in Eq. 6.29 can be approximated, for
instance, by linear interpolation of four neighboring nodal
values:

γp φp + γE φE + γN φN + γNE φNE
φne = Figure 6.11 Interpolation of values in
γp + γE + γN + γNE CV edges for discretization of diffusive
Eq. 6.30 fluxes for non-Cartesian CV – Courtesy of
with suitable interpolation factors γP, γE, γN, and γNE (see [Schafer]
Figure 6.11).
6.5.8 Discrete Transport Equation
Let us now return to our example of the general two-dimensional transport equation Eq. 6.12 and
apply the approximation techniques introduced in the preceding sections to it. We employ
exemplarily the midpoint rule for the integral approximations, the UDS method for the convective
flux, and the CDS method for the diffusive flux. Additionally, we assume that we have velocity
components v1, v2 > 0 and that the grid is a Cartesian one. With these assumptions one obtains the
following approximation of the balance equation Eq. 6.12:

ap φp = aE φE + aw φw + aN φN + as φs + bp
Eq, 6.31
with the coefficients

Γ
aE =
(xE − xp )(xe − xw )
𝜌𝑣1 Γ
a𝑤 = +
𝑥𝑒 − 𝑥𝑤 (𝑥𝑝 − 𝑥𝑤 )(𝑦𝑒 − 𝑦𝑤 )
Γ
aN =
(yN − yp )(yn − ys )
ρv2 Γ
as = +
yN − ys (yp − ys )(yn − ys )
ρv1 Γ(xE − xw )
ap = + +
xe − xw (xp − xw )(xE − xp )(xe − xw )
ρv2 Γ(yN − ys )
+
yn − ys (yp − ys )(yN − yp )(yn − ys )
 160

bp = fp
Eq. 6.32
6.5.9 Treatment of Boundary Conditions
In this particular case (Eq. 6.32) coincides with a discretization that would result from a
corresponding finite-difference method (for general grids this
normally is not the case). It can be seen that – independent from
the grid employed – one has for the coefficients in (Eq. 6.32) the
relation aP = aE + aW + aN + aS . This is characteristic for finite-
volume discretization and expresses the conservatively of the
method. We will return to this important property. Eq. 6.32 is
valid in this form for all CVs, which are not located at the boundary
of the problem domain. For boundary CVs the approximation it
includes nodal values outside the problem domain, such that they
require a special treatment depending on the given type of
boundary condition.
We consider the three boundary condition types that most
frequently occur for the considered type of problems (see Chap.
2): a prescribed variable value, a prescribed flux, and a symmetry Figure 6.12 Cartesian
boundary. For an explanation of the implementation of such boundary CV at west boundary
conditions into a finite-volume method, we consider as an with notations
example a Cartesian CV at the west boundary (see Figure 6.12)
for the transport equation (Eq. 6.12). Correspondingly modified approaches for the non-Cartesian
case or for other types of equations can be formulated analogously. Let us start with the case of a
prescribed boundary value φw = φ0. For the convective flux at the boundary one has the
approximation:
C
Fw ≈ ṁ𝑤 φ𝑤 = ṁ𝑤 φ0
Eq. 6.33
With this the approximation of FCw is known (the mass flux m˙w at the boundary is also known) and
can simply be introduced in the balance equation (Eq. 6.17). This results in an additional
contribution to the source term bP. The diffusive flux through the boundary is determined with the
same approach as in the interior of the domain. Analogously, the derivative at the boundary can be
approximated as follows:
∂φ φp − φw φp − φ0
( ) ≈ =
∂x w xp − xw xp − xw
Eq. 6.34
This corresponds to a forward difference formula of 1st order. Of course, it is also possible to apply
more elaborate formulas of higher order. However, since the distance between the boundary point
w and the point P is smaller than the distance between two inner points (half as much for an
equidistant grid, see Figure 6.12), a lower order approximation at the boundary usually does not
influence the overall accuracy that much. In summary, one has for the considered boundary CV a
relation of the form (Eq. 6.32) with the modified coefficients:

aw = 0
ρv1 Γ(xE − xw )
ap = + +
xe − xw (xp − xw )(xE − xp )(xe − xw )
 161

ρv2 Γ(yN − ys )
+
yn − ys (yp − ys )(yN − yp )(yn − ys )
𝜌𝑣1 Γ
b𝑝 = f𝑝 + ( + ) φ0
𝑥𝑒 − 𝑥𝑤 (𝑥𝑝 − 𝑥𝑤 )(𝑦𝑒 − 𝑦𝑤 )
Eq. 6.35
All other coefficients are computed as for a CV in the interior of the problem domain. Let us now
consider the case where the flux Fw = F0 is prescribed at the west boundary. The flux through the CV
face is obtained by dividing F0 through the length of the face xe−xw. The resulting value is introduced
in (Eq. 6.17) as total flux and the modified coefficients for the boundary CV become:

aw = 0
ρv1 Γ
ap = + +
xe − xw (xp − xw )(xE − xp )(xe − xw )
𝜌𝑣2 𝛤(𝑦𝑁 − 𝑦𝑠 )
+
𝑦𝑛 − 𝑦𝑠 (𝑦𝑝 − 𝑦𝑠 )(𝑦𝑁 − 𝑦𝑝 )(𝑦𝑛 − 𝑦𝑠 )
F0
bp = fp +
xe − xw
Eq. 6.36
All other coefficients remain unchanged. Sometimes it is possible to exploit symmetries of a problem
in order to downsize the problem domain to save computing time or get a higher accuracy (with a
finer grid) with the same computational effort. In such cases one has to consider symmetry planes or
symmetry lines at the corresponding problem boundary. In this case one has the boundary condition:

∂φ
n =0
∂xi i
Eq. 6.37
From this condition it follows that the diffusive flux through the symmetry boundary is zero. Since
also the normal component of the velocity vector has to be zero at a symmetry boundary (i.e., vini =
0), the mass flux and, therefore, the convective flux through the boundary is zero. Thus, in the balance
equation (Eq. 6.17) the total flux through the corresponding CV face can be set to zero. For the
boundary CV in Figure 6.12 this results in the following modified coefficients:

aw = 0
ρv1 Γ
ap = + +
xe − xw (xE − xp )(xe − xw )
ρv2 Γ(yN − ys )
+
yn − ys (yp − ys )(yN − yp )(yn − ys )

If required, the (unknown) variable value at the boundary can be determined by a finite-difference
approximation of the boundary condition (Eq. 6.37). In the considered case, for instance, with a
forward difference formula one simply obtains φw = φP. As with all other discretization techniques,
the algebraic system of equations resulting from a finite-volume discretization has a unique solution
 162

only if the boundary conditions at all boundaries of the problem domain are taken into account (e.g.,
as outlined above). Otherwise there would be more unknowns than equations.

6.5.10 Case Study - Numerical Analysis of Turbulent Flow Around 2D Bodies187

Citation : Tarafder, M. & Mursaline, M.(2019). Numerical Analysis of Turbulent Flow Around Two-
Dimensional Bodies Using Non-Orthogonal Body-Fitted Mesh. International Journal of Applied
Mechanics and Engineering,24(2) 387-410. https://doi.org/10.2478/ijame-2019-0024
This paper deals with the numerical simulation of a turbulent flow around two-dimensional bodies
by the finite volume method with non-orthogonal body-fitted grid. The governing equations are
expressed in Cartesian velocity components and solution is carried out using the SIMPLE algorithm
for collocated arrangement of scalar and vector variables. Turbulence is modeled by the k-
turbulence model and wall functions are used to bridge the solution variables at the near wall cells
and the corresponding quantities on the wall. A simplified pressure correction equation is derived
and proper under-relaxation factors are used so that computational cost is reduced without
adversely affecting the convergence rate. The numerical procedure is validated by comparing the
computed pressure distribution on the surface of NACA 0012 and NACA 4412 hydrofoils for different
angles of attack with experimental data. The grid dependency of the solution is studied by varying
the number of cells of the C-type structured mesh. The computed lift coefficients of NACA 4412
hydrofoil at different angles of attack are also compared with experimental results to further
substantiate the validity of the proposed methodology.
6.5.10.1 Introduction and Literature Survey
Turbulence being so ubiquitous in nature has its prominent influence in almost all practical flows,
thus making its computation so important for applied mathematicians and engineers. However, the
fluid mechanics phenomena in the turbulent flow regime are interesting and at the same time
considerably more complicated. In addition to the inherent complicacies of the system of partial
differential equations governing fluid flow, turbulence and geometric complexities of the domain give
rise to further challenges. While turbulence may be dealt with using a suitable turbulence model,
domain complexity may be circumvented using body fitted coordinates.
In the past, various methods were employed to overcome these difficulties associated with numerical
computation of turbulent flow in complex domains. Rhie188 used finite volume method for the
solution of two-dimensional incompressible, steady turbulent flows over airfoils using k -ε
turbulence model and wall functions. Instead of staggered grids, the body fitted grid utilized a
collocated arrangement of variables where the false pressure field was avoided by special
momentum interpolation. Peric189 developed a finite volume method for viscous flow in complex
geometries, discretizing the governing transport equations in terms of Cartesian vector and tensor
components and arbitrary non-orthogonal coordinates.
[Demirdzic et al.]190 provided a complete exposition of a finite volume approach to the calculation of
turbulent flows. [Karki and Patankar ]191 presented a general calculation procedure for computing

187 Md. Shahjada Tarafder and M. Al Mursaline, “Numerical Analysis Of Turbulent Flow Around Two Dimensional
Bodies Using Non-Orthogonal Body-Fitted Mesh”, Int. J. of Applied Mechanics and Engineering, 2019.
188 Rhie C.M. (1981): A Numerical Study of the Flow Past an Isolated Airfoil with Separation. PhD Thesis, Dept.

of Mechanical and Industrial Engineering, University of Illinois at Urbana-Champaign.

189 Peric M. (1985): A Finite Volume Method for the Prediction of Three-Dimensional Fluid Flow in Complex Ducts.

Ph.D. Thesis, University of London.

190 Demirdzic I., Gosman A.D., Issa R.I. and Peric M. (1987): A calculation procedure for turbulent flow in complex

geometries. Computers and Fluids, vol.15, pp.251-273.

191 Karki K.C and Patanker S.V. (1988): Calculation procedure for viscous incompressible flows in complex

geometries. Numerical Heat Transfer, vol.14, pp.295-307.

 163

fluid flow and related phenomenon in arbitrary-shaped domains with physical covariant velocity
components selected as the dependent variables in momentum equations and the coupling between
the continuity and momentum equations ensured using the SIMPLE algorithm.
[Majumdar]192 reported that solutions of steady-state problems from [Rhie and Chow] momentum
interpolation are dependent on the under-relaxation factor. [Choi]193 reported that the solution using
the original [Rhie and Chow] scheme is time step size dependent. He proposed a modified [Rhie and
Chow] scheme for an unsteady problem which is quite similar to the scheme for a steady problem
used by [Majumdar].
Moreover, [Masuko and Ogiwara]194 carried out numerical simulation of viscous flow around ships
having practical hull forms. The governing equations were discretized by finite difference
approximation and solved with SIMPLE algorithm adopting the k-ε turbulence model and standard
wall functions. [Yu et al.]195 discussed different momentum interpolation practices for collocated grid
systems. [Mulvany et al.]196 carried out an assessment of two-equation turbulence models for high
Reynolds number hydrofoil flows using the finite volume method and SIMPLE solution technique.
[Kuzmin and Mierka]197 presented a detailed numerical study of the k-ε turbulence model using
algebraic flux correction to enforce the positivity constraint. Emphasis was laid on a new
implementation of wall functions whereby the boundary conditions for κ and ε were prescribed in a
weak sense. [Demirdzic]198 discussed the discretization of diffusion term in finite volume continuum
mechanics. [Martınez et al.]199 proposed a possible correction for under-relaxation factor
dependency in the Original Momentum Interpolation Method (OMIM).
The aim of the present paper is to simulate the turbulent flow around two-dimensional bodies by the
finite volume method with non-orthogonal body fitted grid. The k-ε turbulence model and wall
functions are used to bridge the solution variables at the near wall cells and the corresponding
quantities on the wall. The solution is carried out using the SIMPLE algorithm with a simplified
pressure correction equation for collocated arrangement for scalar and vector variables. Despite
using a simplified pressure correction equation to reduce computational cost, and facilitate the
solution of linear systems, a satisfactory convergence rate is achieved by proper choice of under-
relaxation factors. The pressure distribution and lift forces on the surface of the hydrofoils far away
from the free surface are compared with the experiment and the agreement is found to be quite
satisfactory which reflects the accuracy of the present numerical methodology.
6.5.10.2 Governing Equations
In the Cartesian co-ordinate system, the steady two-dimensional turbulent flow around a hydrofoil

192 Majumdar S. (1988): Role of under-relaxation in momentum interpolation for calculation of flow with Non-
staggered grids. Numerical Heat Transfer, Part B, vol.13, pp.125-132.
193 Choi S.K. (1999): Note on the use of momentum interpolation method for unsteady flows. Numerical. Heat

Transfer, Part A, vol.36, pp.545-550.

194 Masuko A. and Ogiwara S. (1990): Numerical simulation of viscous flow around practical Hull form. Fifth

International Conference on Numerical Ship Hydrodynamics, pp.211-224.

195 Yu B., Tao W. and Wei J. (2002): Discussion on momentum interpolation method for collocated grids of

incompressible flow. Numerical Heat Transfer, Part B, vol.42, pp.141-166.

196 Mulvany N., Tu J.Y., Chen L. and Anderson B. (2004): Assessment of two-equation turbulence modeling for high

Reynolds number hydrofoil flow. International Journal of Numerical Methods in Fluids, vol.45, pp.275-299.
197 Kuzmin D. and Mierka O. (2006): On the implementation of the k- turbulence model in incompressible flow

solvers based on a finite element discretization. International Conference on Boundary and Interior Layers, Bail
2006, Germany, pp.1-8.
198 Demirdzic I. (2015): On the discretization of diffusion term in finite-volume continuum mechanics. Numerical

Heat Transfer, Part B, vol.68, pp.1-10.

199 Martınez J., Piscagliaa F., Montorfanoa A., Onoratia A. and Aithalb S.M. (2017): Influence of momentum

interpolation methods on the accuracy and convergence of pressure-velocity coupling algorithms in Open FOAM.
Journal of Computational and Applied Mathematics, vol.309, pp.654-673.
 164

for an incompressible fluid is governed by the following time-averaged equations

∂u ∂v
+ =0
∂x ∂y
∂(ρuu) ∂(ρuv) ∂ ∂u ∂ ∂u ∂p
+ = [(μ + μT ) ]+ [(μ + μT ) ]−
∂x ∂y ∂x ∂x ∂y ∂y ∂x
∂(ρuv) ∂(ρvv) ∂ ∂v ∂ ∂v ∂p
+ = [(μ + μT ) ]+ [(μ + μT ) ]−
∂x ∂y ∂x ∂x ∂y ∂y ∂y
Eq. 6.38
where u and v are the mean velocity components in the x and y directions, respectively, ρ is the fluid
density, P is the mean pressure and μ is the laminar viscosity. μT is the turbulent viscosity and is given
by
κ2
μT = Cμ ρ
ε
Eq. 6.39
where k is the turbulent kinetic energy, ε is the dissipation rate of k and Cμ is a constant. In the k -ε
turbulence model, k and ε are governed by the following transport equations in the Cartesian
coordinate system

∂(ρuκ) ∂(ρvκ) ∂ 𝜇 𝑇 ∂𝜅 ∂ 𝜇 𝑇 ∂𝜅
+ = [(μ + ) ]+ [(μ + ) ] + G − ρε
∂x ∂y ∂x 𝜎𝜅 ∂x ∂y 𝜎𝜅 ∂y
∂(ρuε) ∂(ρvε)
+
∂x ∂y
∂ 𝜇 𝑇 ∂𝜀 ∂ 𝜇 𝑇 ∂𝜀 𝜀
= [(μ + ) ] + [(μ + ) ] + (C𝜀1 G − C𝜀2 ρε)
∂x 𝜎𝜀 ∂x ∂y 𝜎𝜀 ∂y 𝜅
Eq. 6.40
where G is the production of k and is given by

∂u 2 ∂v 2 ∂u ∂v 2
G = 2μT [( ) + ( ) ] +μT ( + )
∂x ∂y ∂x ∂y
Eq. 6.41
The standard values of the constants are as follows

C𝜇 = 0.09 , σ𝜅 = 1 , σ𝜀 = 1.3 , C 𝜀1 = 1.44 , C 𝜀2 = 1.92

Eq. 6.42
Eq. 6.38-Eq. 6.41 may be represented in the following generic form

∂(ρuφ) ∂(ρvφ) ∂ ∂φ ∂ ∂φ
+ = [Γ ]+ [Γ ] + R 𝜑 (x, y)
∂x ∂y ∂x ∂x ∂y ∂y
Eq. 6.43
where u and v are the mean velocity components, φ is any generic dependent variable u, v,k ,ε Γ is
an effective diffusion coefficient and Rφ is the source term. Note that for continuity equation φ = 1, Γ
= 0 , Rφ = 0 , for u-momentum equation φ = u , Γ = μ+μT , Rφ = -∂p/∂x and so on. Considering the body
 165

fitted co-ordinate system, ξ = ξ (x, y) , η= η ( x, y) as shown in Figure 6.13 (a and b). Eq. 6.43 can
be transformed into the following form

1 ∂(ρUφ) 1 ∂(ρVφ)
+
J ∂ξ J ∂η
1 ∂ Γ ∂φ ∂φ 1 ∂ Γ ∂φ ∂φ
= [( α − β )] + [( γ − β )] + S φ (ξ, η)
J ∂ξ J ∂ξ ∂η J ∂η J ∂η ∂ξ
∂y ∂x ∂y ∂x
where U = u − v , V=v −u
∂η ∂η ∂ξ ∂ξ
Eq. 6.44
are the contra variant velocity components.

∂x 2 ∂y 2 𝜕𝑥 𝜕𝑥 𝜕𝑦 𝜕𝑦 𝜕𝑥 2 𝜕𝑦 2
α=( ) +( ) , β= + , γ=( ) +( )
∂η ∂y 𝜕𝜉 𝜕𝜂 𝜕𝜉 𝜕𝜂 𝜕𝜉 𝜕𝜉
Eq. 6.45
Sφ(ξ ,η) is the source term in ξ,η coordinates, J is the Jacobian of transformation and is given by

𝜕𝑥 𝜕𝑦 𝜕𝑥 𝜕𝑦
J= +
𝜕𝜉 𝜕𝜂 𝜕𝜂 𝜕𝜉
Eq. 6.46

Figure 6.13 Coordinate System and C-type Control Volume

6.5.10.3 Boundary Conditions

Neglecting the effect of free surface, the boundary conditions for a flow field around a hydrofoil fixed
in a stream of uniform velocity u0 can be written as:

(a) Inflow boundary: The components of the flow variables are provided as

u = u0 v=0 κ = κ 0 ε = ε0
Eq. 6.47
(b) Outflow boundary: The outlet boundary is located far from the region of interest and the Reynolds
 166

number is high, the gradient in the flow direction is taken to be zero. Thus

∂u ∂v ∂κ ∂ϵ
= = = =0
∂n ∂n ∂n ∂n
Eq. 6.48
where ∂/∂n is the derivative parallel to the streamlines;
(c) Solid boundary: The no slip boundary condition is applied on the surface of the hydrofoil.

u=0 , v=0 , κ=0, ε=0

Eq. 6.49
(d) Far-field boundary: The flow variables on the symmetry plane are prescribed as

us = up v=0 κs = κp εs = εp
Eq. 6.50
Moreover, as the standard k-ε turbulence model cannot be applied in the transition layer and also the
viscous sublayer around the hydrofoil, the wall functions are to be adopted.
6.5.10.4 Finite Volume Discretization of Governing Equation
The discretization is performed following a finite control volume approach in which the
computational domain is divided into a number of contiguous quadrilateral cells. A collocated grid
arrangement is used in which all the variables are stored at the geometric center of the cell (see
Figure 6.14).

Figure 6.14 Collocated Grid Arrangement

The locations of the various dependent variables and the associated cells for this grid configuration.
Eq. 6.44 is integrated over the volume of each cell in the computational domain as

∂(ρUφ) ∂(ρVφ)
∫[ + ] dv =
∂ξ ∂η
CV
 167

∂ Γ ∂φ ∂φ ∂ Γ ∂φ ∂φ
∫ [( α − β )] dv + ∫ [( γ − β )] dv + ∫ JS φ (ξ, η) dv
∂ξ J ∂ξ ∂η ∂η J ∂η ∂ξ
CV CV CV
Eq. 6.51
Applying Gauss’s divergence theorem to convert volume integrals to surface integrals, Eq. 6.51
after little rearrangement, may be written as

(ρUφ∆η) 𝑒𝑤 +(ρVφ∆ξ) 𝑛𝑠 =
e n e
Γ ∂φ Γ ∂φ Γ ∂φ
( α ∆η ) + ( γ ∆ξ ) + {− ( β ∆η )
J ∂ξ w
J ∂η s
J ∂η w
n
Γ ∂φ
−( β ∆ξ ) + J(S c + Sp φp )∆ξ∆η}
J ∂ξ s
Eq. 6.52
The cross derivative terms have been added to the source term which in turn has been linearized as
suggested by [Patanker]200. Using the notation of Figure 6.14, the following approximations may be
made for the derivatives at face e:

∂x xE − xp ∂x (xn − xs )e xne − xse

( ) ≈ , ( ) ≈ =
∂ξ e ξE − ξp ∂η e (ηn − ηs )e ηne − ηse
∂y yE − yp ∂y (yn − ys )e yne − yse
( ) ≈ , ( ) ≈ =
∂ξ e ξE − ξp ∂η e (ηn − ηs )e ηne − ηse
Eq. 6.53
Analogous expressions may be derived for other faces. Eq. 6.67 can be written as

[ Fe φe − Fw φw + Fn φn − Fs φs ] =
[ De (φE − φ𝑝 ) − Dw (φp − φ𝑊 ) + Dn (φN − φ𝑝 ) − Ds (φp − φ𝑠 ) ] +
[ Ne (φn − φ𝑠 )𝑒 − Nw (φn − φ𝑠 )𝑤 + Nn (φe − φ𝑤 )𝑛 − Ns (φe − φ𝑤 )𝑠 ] +
[ (S𝑐 − S𝑝 φp )δV ]
Eq. 6.54
Central differencing is used to discretize the diffusion terms and suitable interpolation for the
convective terms is required to express cell face values in terms of nodal values. This is achieved from
[Demirdzic and Peric]201 by blending second-order central (CDS) differencing and first-order
unconditionally stable upwind differencing scheme (UDS) in a deferred correction manner

[ φUDS
e ]imlicit + λ [ φCDS
e − φUDS
e ]explicit
Eq. 6.55
where λ is the blending factor having value between 0 to 1. The explicit part in Eq. 6.55 is obtained
from previous iteration and added to the source term, like the cross derivative terms. Using the above
scheme for convective terms and after little manipulation, Eq. 6.54 can be written in the following
algebraic form

200 Patanker S.V. (1980): Numerical Heat Transfer and Fluid Flow. New York: McGraw-Hill.
201 Demirdzic I. and Peric M. (1990): Finite volume method for prediction of fluid flow in arbitrary shaped domains

with moving boundary. International Journal of Numerical Methods in Fluids, vol.10, pp.771-790.
 168

a𝑝 φ𝑝 = a𝑊 φ𝑊 + a𝐸 φ𝐸 + a𝑁 φ𝑁 + a𝑆 φ𝑆 + S ′
Eq. 6.56
Introducing an under-relaxation factor to slow down changes of the dependent variable in
consecutive iterations, Eq. 6.56 becomes

𝑎𝑝
( ) φ𝑝 = a𝑊 φ𝑊 + a𝐸 φ𝐸 + a𝑁 φ𝑁 + a𝑆 φ𝑆 + S 𝜑
𝛼𝜑
Eq. 6.57
For further information, please consult the development in [Shahjada Tarafder and Al Mursaline ]202.
6.5.10.5 The Wall Functions
To close the statement of the problem, we still need to prescribe the tangential stress as well as the
boundary conditions for k and ε on the solid boundary. Note that the equations of the k - ε model are
invalid in the vicinity of the wall where the Reynolds number is rather low and viscous effects are
dominant. In the case of laminar flow, the no-slip wall boundary condition is directly applied at the
wall. For turbulent flows however, to avoid using finer grids near walls, where steep cross flow
gradients exist, ‘wall functions’ are used. These wall functions are given as follows

1/4 1/2 1/4 1/2

ρvt Cμ κp ρCμ κp np
τw = where n+
p =
1 μ
ln(En+
p)
κ
Eq. 6.58
vt is the wall parallel velocity and nP is the distance of the first computational node normal to the wall.

1/4 1/2 3/4 3/2

Cμ κ p Cμ κp np
G𝑝 = τ𝑤 εp =
κn𝑝 κn𝑝
Eq. 6.59
6.5.10.6 SIMPLE Algorithm
To obtain velocity and pressure fields and values of turbulent quantities, an iterative solution
procedure akin to the SIMPLE method (Patankar and Spalding]203) is used. In the present work the
scalar and vector variables are stored in a collocated manner.
6.5.10.7 Results and Discussion for NACA 0012 Hydrofoil
The flow past an NACA 0012 hydrofoil with 2.5 m chord length (C) is simulated at Rec = 2. 8 x106 with
grid sizes of 50 x 14 , 88 x 20, 176 x 40 and 0, 6, 10 degrees of incidence. A typical grid arrangement
for 176 x40 cells is shown in Fig.3. In the present simulation the inlet boundary at the front of the foil
along with the upper and bottom boundaries are located 4 chord lengths away from the leading edge.
The outlet boundary at the rear side is 8 chord lengths away from the leading edge. The under-
relaxation factors used are αu = αv = αk = αε = 0.7 , αp = 0.1 and the value of blending factor is λ = 0.5 .
A comparison of obtained pressure coefficients with its experimental results for (6 degree) and a grid
size of 176 x 40 is shown in Figure 6.15 (Top). From these figures it is evident that the computed

202 Md. Shahjada Tarafder and M. Al Mursaline, “Numerical Analysis Of Turbulent Flow Around Two Dimensional
Bodies Using Non-Orthogonal Body-Fitted Mesh”, Int. J. of Applied Mechanics and Engineering, 2019.
203 Patanker S.V. and Spalding D.B. (1972): A calculation procedure for heat, mass and momentum transfer in

three dimensional parabolic flows. International Journal of Heat and Mass Transfer, vol.15, pp.1787–1806.
 169

results agree very well with the experiment. In fact, at 0 and 6-degree incidence the result agrees

Figure 6.15 Cp Contours for NACA 0012 Hydrofoil at 6 Degree Incidence

 170

better than that obtained by [Rhie]204 near the leading edge. However, discrepancy can be noted at
the leading and trailing edges which may be attributed partly to inadequate turbulence modeling.
Moreover, the mesh may not be sufficiently fine to capture the very strong gradients existing in those
regions. In addition to the pressure coefficient curves, the pressure contours and streamlines at 6
degrees incidence are shown in Figure 6.15 (Bottom) for a grid size of 176 x 40 at Rec = 2 8 x106 .
The C type structured mesh with three different sizes was issued to investigate the influence on
surface pressure coefficients. At 10 degree angle of attack, pressure coefficients are computed at each
of the three grids. It is evident that refinement of the grid gives increased numerical accuracy of
pressure coefficients in the present case. Significant improvement is found on the suction side of the
hydrofoil. However, the effect on the pressure side (where the discrepancy is small) is less
considerable. For complete analysis, please see [Shahjada Tarafder and Al Mursaline ]205.
6.5.10.8 Conclusions
This paper presents the numerical computation of turbulent flow past hydrofoils using the finite
volume method with a k-ε turbulence model. The following conclusions can be drawn from the
present numerical study:
➢ The surface pressure distributions on a number of hydrofoils such as NACA0012206 and
NACA4412 located far from the free surface are computed and then compared with the
theoretical as well as experimental data. The agreement with established data is found to be
excellent except very close to the leading edge.
➢ Grid refinement is carried out systematically keeping a proper distance between the first
computational node and the wall. Significant improvement in results has been obtained with
a refined grid particularly near the leading edge.
➢ Despite simplifying the pressure correction equation by neglecting cross-diffusion
contributions,
➢ proper under-relaxation factors may be used so that computational cost is reduced without
adversely affecting the convergence rate.
➢ Lift coefficients are computed which exhibit an excellent agreement with experimental data
at lower angles of attack. However, due to poor performance of the k - ε turbulence model, at
higher angles of attack when flow separation takes place, the agreement is found to be less
satisfactory.

6.6 Some Abnormalities Associated with Control Volume (CV)

For fluid flow, there are special considerations. As seen earlier in this section, there are 5 equations
for 5 unknowns (u, v, w, p, T). However, there are two problems with these equations which are
specific to computational fluid dynamics (CFD). First, the governing equations are not only coupled,
but they have non-linear terms, namely the advection or inertia terms. The handling of these terms
has been an ongoing research project for at least the last 40 years. If these terms are not modeled
accurately enough, they will introduce an error known as “Numerical Diffusion”.
For fluid flow, there are special considerations. As seen earlier in this section, there are 5 equations
for 5 unknowns (u, v, w, p, T). However, there are two problems with these equations which are
specific to computational fluid dynamics (CFD). First, the governing equations are not only coupled,

204 Rhie C.M. (1981): A Numerical Study of the Flow Past an Isolated Airfoil with Separation. PhD Thesis, Dept. of
Mechanical and Industrial Engineering, University of Illinois at Urbana-Champaign
205 Md. Shahjada Tarafder and M. Al Mursaline, “Numerical Analysis Of Turbulent Flow Around Two Dimensional

Bodies Using Non-Orthogonal Body-Fitted Mesh”, Int. J. of Applied Mechanics and Engineering, 2019.
206 Gregory N. and O'Reilly C.L. (1970): Low Speed Aerodynamic Characteristics of NACA 0012 Airfoil Section,

Including the Effects of Upper Surface Roughness Simulation Hoarfrost. National Physical Laboratory,
Teddington, England, Aero Report No.1308.
 171

but they have non-linear terms, namely the advection or inertia terms. The handling of these terms
has been an ongoing research project for at least the last 40 years. If these terms are not modeled
accurately enough, they will introduce an error known as “Numerical Diffusion”. As its name
indicates, the errors can completely swamp any physical diffusion and misrepresent the physics of
the real world problem. If you model
the advection terms with the usual
methods of obtaining high accuracy
(central differences, standard
Galerkin schemes), you introduce Artificial
numerical dispersion errors where Viscosity
the numerical solution oscillates
around the true solution. These
dispersion errors can quite easily
Dissaption
lead to divergent solutions, especially &
in turbulent flows. Most commercial Dispersion
finite volume and finite element
methods have discretized these Numerical
terms in some special way which is a Diffusion
compromise of accuracy and
stability. Finite volume methods use
techniques like skew up winding and
QUICK schemes. Successful finite
element methods use some sort of Figure 6.16 Potent Numerical Errors in CFD
streamline upwind element. (Yes,
there are finite element CFD methods
available which do not use this method, but they are not generally applicable). (See Figure 6.16).
The second major difficulty with the governing partial differential equations is that no explicit
equation for pressure exists for incompressible flows. For example, if we use the Navier-Stokes or
momentum equations to solve for the velocities, we have only the continuity equation to solve for
pressure. However, pressure does not appear in the continuity equation. This problem has been side-
stepped by manipulating a combination of these equations. The most predominant method
(commercially, that is) for solving this dilemma of the missing pressure equation was developed for
finite volume methods and is known as SIMPLE or some variant of it. This method is well-explained
in the book by [Patankar]207. Almost all of the commercial finite volume CFD codes use this method
as well as two most popular finite element CFD codes.
6.6.1 Numerical Diffusion Effects
Artificial viscosity tends to reduce all gradients in the solution whether induced physically or

Figure 6.17 Effects a) exact b) 1ST order (Dissipation) c) 2nd order (Dispersion)

207 Suhas V. Patankar, “Numerical Heat Transfer”, Hemisphere Publishing, 1980, ISBN 0-89116-522-3.
 172

numerically. This effect, which is the direct result of

even derivative terms in the truncation error, is
also called Dissipation (upwind differencing).
Another quasi-physical effect of numerical
schemes is called Dispersion which is the direct
result of odd derivative terms in truncation error
(central differencing). The combined effects are
called Diffusion. Figure 6.17 shows these effects.
To show how the impact of the artificial viscosity,
consider a supersonic flow over backward facing
step as demonstrated in Figure 6.18. Now let us
try to find the numerical solution for this problem
by progressively increasing the magnitude of the
artificial viscosity (See Figure 6.18 (a)-(d)). We
observe that adding the artificial viscosity has
thickened the shock for sure. The smoothing (Less
wiggles) of the shock can observed for the
increasing artificial viscosity. See the Fig (d) for the
smooth shock(without wiggles). There are
essentially two main classes of convective flux
approximation in widespread use, namely:
1. ‘Low-order’ schemes, which
characteristically generate discretized
equation forms that are easy to solve,
produce solutions which obey the expected Figure 6.18 Progression of Shock Wave
physical bounds, but sometimes give rise to on Supersonic Flow over a Backward Step
smearing of gradients. The latter effect has
come to be known as ‘Numerical Diffusion- Dissipation effects. This is a form of truncation
error that diminishes as the grid is refined, but at an increased cost of calculation.
2. ‘High-order’ schemes, which better preserve steep gradients, but may result in equations that
are more difficult to solve (and, in extreme cases, may provoke numerical instabilities) and/or
have solutions exhibiting non-physical spatial oscillations (‘wiggles-dispersion effects’). These
oscillations may in some cases, lead to spurious values, e.g. negative species mass fraction or
turbulence kinetic energy. This phenomenon is often termed Numerical Dispersion. It too can
be diminished by grid refinement or by using monotone schemes (e.g. a blending methodology)

The effect of retaining order of terms in TE displaying either the Dissipative or Dispersive effect is
can be illustrated in Figure 6.19.
6.6.2 Balancing the Diffusion vs. Convection
using Peclet Number (Pe) •Lowest-Order term in TE
This is an important property in comparing the Diffusion contain even (Dissiaptive)
CD and UD differencing schemes or other •Lowest -Order term in TE
diffusion related issues. As shown in contains odd (Disspersive)
Figure 6.20 for higher speed flows (large Pe),
the CD differencing becomes oscillatory and Figure 6.19 Effect of retaining TE terms
unstable and not suitable, while UD reveals
more stable and recommended. On the other
hand, for low speed flows (smaller Pe), such as subsonic flows and denser mesh, CD seems
comparably more stable.
 173

advection transport rate LU

PeL = = Re.Pr =
diffusive transport rate α
k
where L = characteritic length , U = velocity and α =
ρC p
Eq. 6.60

Figure 6.20 Correlation Between CD and UD Schemes for Different (Pe) Number

6.6.2.1 Case Study – 1D Convection / Diffusion Problem

To illustrate this more explicitly, solving the 1D convection-diffusion equation on a slab of length L
for variable ϕ subjected to following BC’s

Pe << 1
Diffusion dominated

Convection dominated

Pe = 0
Φ0 <ϕ<ϕL

Pe >> 1

0<X<L

Figure 6.21 Ratio of Convection Vs Diffusion - Effect of (Pe) number

 174

    
(ρ ) =  Γ  subject to BC  x =0 = 0 and  x =L = L Eq. 6.61
x x  x 

Where ρφ and Γ are independent of x. Differentiating results the analytical exponential solution
known as power law (see
Figure 6.21):

x
(Pe )
 (x) −  (0) e L − 1 ρuL
= (Pe) where Pe = Eq. 6.62
 (L) −  (0) e − 1 Γ

Following observations could be made from this plot:

• The ratio of the strengths of convection and diffusion may be measured using the Peclet
number, Pe.
• For high absolute values of Pe number (i.e., convection dominated flow), the value of the
dependent variable at the interface (i.e. x = L/2) can be seen to be nearly equal to the value
at the upwind boundary which is the assumption made by the upwind scheme. Such
assumption could not be hold for small Pe values (diffusion dominated; see
• Figure 6.21).
• When Pe=0, (no flow, or pure diffusion), ϕ could be interpolated using a simple linear average
between the values at x = 0 and x = L.
• The upwind scheme always calculates diffusion assuming a linear relationship between the
dependent variable and distance and therefore over estimates diffusion at large absolute
values of Pe208.
Another point of view of above on a 1D
spatial mesh, for general contours of
variable ϕ at point p, for Pe = 0 i.e., pure
diffusion, the contours are concentric
circles for constant values of ϕ because
diffusion tends to spread out ϕ evenly in all
directions. As Pe increases, the curve
becomes more elliptic and the value of ϕ at
E node is more influenced by the upstream
node. At Pe = infinity i.e., pure convection, Figure 6.22 Spatial effects of (Pe) number on 1D flow
the contours are completely outstretched
and value of ϕ at E is affected only by upstream conditions (see Figure 6.22).

6.7 Some Attributes to Spatial Discretization

6.7.1 Cell-Centered (CC) vs. Node-Centered (NC)
The cell-centered (CC) referred to a discretization that uses primal grid cells as control volumes, as
opposed to node centered (NC) that uses median-dual cells as control volumes. Illustrates the control
volume partitioning of a finite volume discretization. Performance of flow solvers can also vary with
each algorithm and mesh data structure used. While the CC based solvers integrate the fluxes on the
faces of individual cells, NC based solvers integrate the fluxes over cells connected to individual cells.
Though the number of faces is the same in both approaches, the actual finite volume used in CC based

208 “Discretization of Convection –Diffusion type equation”; 10th-Indo–German Winter Academy 2011.
 175

is smaller hence better spatial resolution. On the other hand, the NC based approach works with less
memory, especially for tetrahedral cells when usually the numbers of cells are five-six times of
nodes209. In cell-centered scheme, the control volumes are identical with the grid cells and the flow
variables are associated with the centers of the grid cells as illustrated in Figure 6.23. When we
evaluate the discretized flow equations, we have to supply the convective and the viscous fluxes at
the midpoints of the faces of the control volume. However, in cell-vertex scheme, the flow variables
are associated with the grid nodes (vertices) rather than the cell centroids. Median-dual control
volumes are formed by connecting the centroids, face- and edge-midpoints of all cells sharing the

Figure 6.23 Cell Centered and Node Centered

particular node. The definition of a median-dual control volume results in a polyhedral hull around
each grid node.
6.7.2 Staggered vs. Collocated Grids
This has nothing to do with grid generation techniques but rather spatial discretization. A staggered
grid is a setting for the spatial discretization, in which the variables are not defined at the same
position as in case of the collocated grid. On a
staggered grid the scalar variables (pressure,
density, total enthalpy etc.) are stored in the cell
centers of the control volumes, whereas the
velocity or momentum variables are located at
the cell faces. This is different from a collocated
grid arrangement, where all variables are stored
in the same positions. A staggered storage is
mainly used on structured grids for
compressible or incompressible flow
simulations. Using a staggered grid is a simple
way to avoid odd-even decoupling between the
pressure and velocity. Odd-even decoupling is a
discretization error that can occur on collocated
grids and which leads to checkerboard patterns
in the solutions. The disadvantage of using
staggered grids is that different variable are
stored at different places and this makes it more Figure 6.24 Control Volume for spatial
difficult to handle different control volumes for discretization of staggered grid vs collocated

209Diskin, Boris; Thomas, James: “Comparison of Node-Centered and Cell-Centered Unstructured Finite-Volume
Discretization: Inviscid Fluxes,” AIAA Journal 2010.
 176

different variables and to keep track of the metrics210. Modern codes instead use a collocated storage.
Figure 6.24 shows a regular staggered grid for the discretization of 2D incompressible Navier-Stokes
equations. When solving the incompressible Navier-Stokes in the primitive variables (u, v, and p)
formulation, we need to discretize the pressure and the velocity components. On a staggered grid
these primitive variables are all defined at different location, with pressure p at the cell center and
the velocity components at the center of the cell faces211.
6.7.3 Desired Properties in Discretization Schemes
In general, any discretization scheme should have the following properties:
• Conservativeness - global conservation of properties must be insured, i.e., flux consistency
in control volume (Flux IN = Flux OUT).
• Boundedness - values predicted by scheme should be within realistic bounds. For linear
problems (Heat conduction), those would be the boundary values. For non-linear problems,
the values within the domain could be outside of boundary values.
• Transportiveness - As diffusion works in all direction, but convection only in flow direction,
the numerical scheme should recognize the flow direction since it effects the balance of
convection vs. diffusion (Peclet Number).
The manner in which the convective and diffusive fluxes are expressed in terms of nodal φ values is
one of the key factors determining accuracy and stability, for both steady-state and transient
calculations. At the high Reynolds numbers often encountered in practice, the choice of convective
flux approximation is particularly important.

6.8 Wall Distance Estimation

The non-dimensional y+ quantity, (based on local cell fluid velocity from the wall to the first node),
has a different meanings in terms of Boundary Layer theory and CFD. In Boundary Layer, y+ is simply
a local thickness Reynolds number. In CFD, y+ is a non-dimensional distance from the wall to the first
grid point. Typical range of y+ for a flow over a smooth flat plate with no adverse pressure gradient
is from wall out to about 150. For finer mesh lower values of y+ should be used. The mechanism for
estimating y+ could be followed by using the local Rex and an imperial skin friction correlation as
follow:
ρU  L BL
Cf = 2log10 (Re x ) − 0.65
− 2.3
Re x =
μ
Eq. 6.63
1 τw y +μ
τ w = Cf − ρU 2 2 u = y=
2 ρ ρu 

Now that we know how to obtain non-dimensional y+ (mostly imperially), you might asked why?
Estimating y+ is so important because placing the first layer of mesh had tremendous impact in CFD
and Near Wall analogy. It dictates how many cell needed to resolve flow accurately. As you approach
the wall you will notice that the velocity decreases non-linearly up to a point where the fluid will have
zero velocity at the wall. This is what is termed the "no slip" wall condition in CFD. If we plot a
typical velocity profile in the near-wall region, we can see that we have a large change in velocity in
the wall normal direction and it is important to our CFD simulation that we capture this gradient
correctly. To do this, we need to use inflation layer meshing to accurately capture the boundary layer
region for any wall-bounded turbulent flows. The image below plots the non-dimensional velocity
versus the non-dimensional wall normal distance, with each line from top to bottom demonstrating

210 CFD Online.

211 Nikhil Kumar, PhD candidate - Scientific computing.
 177

the difference between a favorable pressure gradient (APG) through to adverse pressure gradient
(FPG) with flow separation. (See Figure 6.26).
It is clear that the flow behavior in the near
wall region is fairly complex and needs to be
captured appropriately to have any
confidence in our CFD results, especially if we
intend to report key engineering data such as
separation points or pressure drops.
Providing a suitable inflation mesh for the
geometry is strongly tied to the choice of the
turbulence model, and the flow field we
are interested in capturing. We can elect to
resolve the complete profile of the boundary
layer of alternatively we can make use of
empirical wall functions to reduce the cell
count. To use a wall function approach for a
particular turbulence model with confidence,
we need to ensure that our y+ values are
within a certain range. Figure 6.25 Figure 6.26 Near Wall Velocity Profile
exemplifies the relationship between y+ and
first layer of boundary layer. It is considered good practice to include between 10 and 15 layers
situated within the boundary layer of your flow to accurately resolve and predict any separation or
reattachment points212.
6.8.1 Case Study - Influence of Hybrid
Grid and Turbulence Model
To provide guidelines for CFD
simulations for tandem cycling
aerodynamics, we refer to the parametric
study by [Mannion et al.]213. Specific
attention to near-wall grid resolution
and turbulence model choice are
provided. Furthermore, the forces
experienced by pilot and stoker are Figure 6.25 Wall Function Relation in Boundary Layer
investigated separately to provide an
understanding of the drag interaction between the pilot and stoker athletes.
6.8.1.1 Gridding
Two different grid topologies were devised; (i) a tetrahedral-only grid and (ii) a combined prismatic-
tetrahedral grid, with prismatic cells in the boundary layers and tetrahedral cells beyond (hybrid). A
grid-sensitivity study, comprising of a coarse, medium and fine tetrahedral-only grid, was conducted,
with the surface grid systematically refined with each face size halved for the progressive grids. The
boundary layer resolution depended on the tetrahedral cell size at the surfaces of interest for each
grid as no prism layers were used. The grid sizes were 11.1, 24.6 and 64.9 M cells, respectively. The
medium grid was selected as a reference grid for surface face sizing’s for subsequent grids, providing
a compromise between accuracy and computational expense. Cell face sizes varied depending on the

212LEAP CFD Team, Leap Australia CFD Team Blog, 2014.

213 P. Mannion, Y. Toparlar, B. Blocken, M. Hajdukiewicz, T. Andrianne, and E. Clifford, “Improving CFD
prediction of drag on Paralympic tandem athletes: influence of grid resolution and turbulence model”, Sports
Engineering, October 2017.
 178

location of the surface on the athlete or bicycle geometry. The dimensions are normalised by the
diameter of the athlete’s head (0.2 m). A new grid was created, denoted as grid 1, which stemmed
from the medium grid in the grid independence study. Figure 6.27 illustrate segments of the surface
grid, and also the volume grid in a vertical centre-plane. The total number of cells in grid 1 was 20.2
M. For the second grid used in this study, denoted as grid 2, settings for grid 1 were implemented as
the background grid with the addition of prism cells to all wall surfaces in the boundary layers. Note
that y* ≈ 1 and < 5 is required to resolve the thin viscous sublayers to reproduce boundary layer flow
and potential separation. 20 prism cells with a growth ratio of 1.2 were used. The total number of
cells in grid 2 was 33.3 M. Further discussion are provided in 214.

Figure 6.27 Grid 1 - Tetrahedral Cell Growth(Left) vs, Grid 2 – Hybrid Prism Layer Growth (Right)

6.8.1.2 Boundary Conditions

A uniform velocity of 15 m/s with 0.2% turbulence intensity and a hydraulic diameter of 1 m was
applied as a velocity inlet condition. Air with a density of 1.225 kg/m3 and a viscosity of
1.789E−5 kg/m . s was specified as the fluid. Zero static gauge pressure was applied to the outlet
boundary. A symmetry condition was applied for the lateral boundaries, the top boundary, and also
for the ground boundary to represent a free-slip wall. A no-slip wall with zero roughness was applied
for the tandem bicycle surfaces and for the athlete surfaces.
6.8.1.3 Governing Equations and Solver Settings
The simulations were performed using ANSYS Fluent 16®. The RANS equations were solved in
coordination with the shear stress transport (SST) k-ω turbulence model. The Least Squares Cell
Based method was used to compute gradients. The Coupled algorithm was used for pressure–velocity
coupling(SIMPLE). Second-order pressure interpolation was used, along with second-order
discretisation schemes for all equations. Due to the inherent unsteady nature of tandem cycling
aerodynamics, the pseudo-transient solver
within Fluent was used. Averaging was
required for the resulting forces from the
pseudo-transient simulations where steady-
state convergence was unachievable. A study
was conducted to determine a suitable
pseudo-transient time-step, with values
decreasing by one order of magnitude from
0.1 to 1E−05 s. Drag values were averaged over
4500 iterations after an oscillatory phase was
reached. A negligible difference was found
varying time-step size, with a final size of Figure 6.28 Drag Force (N) on the Pilot and Stoker
0.01 s used to allow for sufficient oscillations using Different Grids

214 See Previous.

 179

to occur over 2000 iterations for averaging purposes. All simulations reported were averaged over
2000 iterations after results reached a statistically steady state.
6.8.1.4 Drag findings
The drag coefficient is described by:

FD
CD =
1
ρAV 2
2
Eq. 6.64
where FD is the drag force (N), ρ the density (kg/m3), A the frontal area (m2) and V the velocity (m/s).
The total drag force of bicycle and two riders for grid 1 and grid 2 were 39.6 and 43.2 N, respectively.
It was expected that the stoker would experience a lower drag than the pilot due to drafting.
However, Figure 6.28 shows that grid 1 yielded an opposite drag distribution: 35.9% of total drag
for the pilot and 46.1% for the stoker. Grid 2 however, yielded 52.5% of total drag for the pilot and
26.9% for the stoker. Figure 6.29 shows the differences in surface pressure coefficient for grids 1
and 2. It is clear that the grid resolution in the boundary layer and its impact on the flow separation
locations and resulting wake flow played a critical role in the drag differences between both grids.
predictions. Note that the athletes experienced a larger viscous drag for grid 1 due to the flow staying
attached for longer to the surfaces of the athletes.

Figure 6.29 CP for Different Grids

6.8.1.5 Impact of Turbulence Model

Eight turbulence models were applied to grid1: the T-SST k-ω, the k-kl-ω, the intermittency SST k-ω,
the SST k-ω, the standard k-ε, the realizable k-ε, the renormalization-group (RNG) k-ε, and the 1-
equation Spalart–Allmaras turbulence model. The 1-equation Wolfshtein model was used for low
Reynolds number modelling with the k-ε models. Second-order discretisation schemes were used for
the convective and viscous terms of all equations. The results are summarised in Figure 6.30215.
The T-SST model provided drag evictions for the pilot and stoker that deviated by − 1.0% and
− 14.4%, respectively, from the wind-tunnel results. The T-SST model also failed to predict the
correct direction of the lateral force on the pilot. The intermittency SST k-ω model provided
comparable results to the T-SST model, with drag deviations of 2.4% and − 13.5% for the pilot and
stoker respectively. The k-kl-ω model under-predicted drag for the pilot by − 14.1%, and over-

215 Drag/Lift and Lateral Forces (N) acting on the pilot and the stoker as obtained by
various turbulence models.
Systematic and random errors within a 95% confidence interval are represented by error bars for the wind-
tunnel data.
 180

predicted drag on the stoker by 4.6%. However, it predicted all lateral and lift forces to within the
error region of the force transducers for both the pilot and stoker. The SST k-ω models predicted the
drag of the pilot and stoker to within − 3.7% and − 13.9% of the wind-tunnel values respectively, and
the Spalart–Allmaras model predicted the drag of the pilot and stoker to within − 4.0% and − 15.9%
respectively. The k-ε models all under-predicted pilot drag forces beyond 30%, with the realizable
and standard k-ε models predicting a larger drag force on the stoker than on the pilot.

Figure 6.30 Drag/ Lift and Lateral Forces (N) Acting on the Pilot and the Stoker

6.8.1.6 Concluding Remarks

The aerodynamics of a tandem para-cycling road race setup was simulated using CFD, and was found
to have a full-scale CD of 0.787. A grid with coarse near-wall boundary layer resolution was found to
yield counter-intuitive drag distributions for individual athletes, with the stoker experiencing a
higher drag than the pilot. Wind-tunnel experiments proved the counter-intuitive drag distributions
to be incorrect, with the stoker experiencing 39% less drag than the pilot. In addition to this grid
dependency of achieving an average y* value close to 1, the CFD simulations were also shown to have
a dependency on turbulence models, with the SST k-ω turbulence model providing the most accurate
drag predictions: 4.0% and 4.2% for the pilot and stoker respectively when compared against wind-
tunnel validation data, and modelling the wind-tunnel geometry within the CFD fluid domain. The
realizable k-ε and standard k-ε models predicted the stoker to experience a larger drag than the pilot,
despite the grid meeting the requirement of an average y* value less than 1, and using low Reynolds
number modelling opposed to wall functions. The RNG k-ε model under-predicted the drag on the
pilot beyond 20%. It is recommended that a fine grid with an average y* value of 1 or less be used in
combination with the SST k-ω turbulence model for future tandem cycling aerodynamics research.

6.9 Gradient Smoothing Method (GSM)

Recently, a new CFD solver based on Gradient Smoothing Method (GSM) has been developed by [Liu
and Xu]216 by combining the major features of the standard FVM and some mesh free techniques [YAO

216 G. R. Liu and George

X. Xu,”A gradient smoothing method (GSM) for fluid dynamics problems”, Int. J. Numerical
Meth. Fluids 2008; 58:1101–1133.
 181

et al.]217. It has been found that the GSM is effective for various types of fluid dynamics problems. In
GSM, all the unknowns are stored at nodes (as in mesh free methods and node-centered FVM) and
their derivatives at various locations of interest are approximated using gradient smoothing
operations. Different kinds of compact smoothing domains can be constructed from the primitive (or
background) mesh of the problem domain for calculating the derivatives at different locations of a
cell, such as at nodes, at midpoints of edges, or at the centroid of the cell. Different kinds smoothing
functions can be chosen for approximating the derivatives.
Compared with FVM, the GSM is more like a strong form method since it operates directly on the
governing equations of fluid dynamics as FDM. However, because of the use of gradient smoothing
operation, which approximates the derivatives at a point of interest using an integral form, the GSM
has a weak formulation flavor, and is conservative as the FVM. For this reason, GSM is known as a
weak form-like method. Intensive numerical tests have demonstrated that the GSM often
outperforms the FVM in terms of stability against the grid irregularity, accuracy and efficiency.
Although the GSM has been successfully applied for solving the compressible and incompressible
fluids for both inviscid and viscous cases on fixed meshes, the GSM has not yet been used for solving
fluid problems with moving boundaries or moving meshes, which is quite common in engineering,
and extremely important for fluid structure interaction simulations.
6.9.1 Gradient Smooth Operation
The gradients of a field variable U at a point of interest at xi in domain Ώi can be approximated in the
form of :

U i  U (x i )   U(x) wˆ (x − x i )dx
i

 1/V x  Ωi
U i   U(x)ŵ(x - x ) n ds -  U(x)ŵ(x - x ) dV
 i
i
i
i and ŵ =  i
0 x  Ωi
Eq. 6.65
where ∇ is a gradient operator and ŵ is a smoothing function that can be chosen properly based on
the requirement on the accuracy of the approximation. ∂Ώi represents the external boundary of the
GSD, and n denotes the unit normal vector on ∂Ώi. Using Eq. 6.65, the right-hand side second term
vanishes and Eq. 6.65 then becomes:

1 1
∇Ui ≈ ⃗ ds
∮ U𝐧 , ∇. (∇Ui ) ≈ ⃗ . ∇Uds
∮𝐧
Vi Vi
∂Ωi ∂Ωi
Eq. 6.66
The above mentioned equation gives an approximation of gradients at a point using a local smoothing
domain. Such a gradient smoothing technique is often adopted in many mesh free methods, for
example, in the widely used Smoothed Particle Method (SPM), to stabilize the nodal integrations
and ensure linear conformability. It has also been used in the development of the smoothed FEM. For
spatial and Temporal discretization, readers are encourage to consult the work by [YAO et al.]218 and

217 Jianyao Yao, G. R. Liu, Dong Qian, Chung-Lung Chen and George X. Xu. “A Moving-Mesh Gradient Smoothing
Method For Compressible CFD Problems”, Mathematical Models and Methods In Applied Sciences Vol. 23, No. 2
(2013) 273–305.
218 Jianyao Yao, G. R. Liu, Dong Qian, Chung-Lung Chen and George X. Xu. “A Moving-Mesh Gradient Smoothing

Method For Compressible CFD Problems”, Mathematical Models and Methods In Applied Sciences Vol. 23, No. 2
(2013) 273–305.
 182

[Liu and Xu]219.

6.9.2 Case Study 1 - Implicit Gradient Method for Cell-Centered Finite-Volume Solver on
Unstructured Meshes
Citation : Hiroaki Nishikawa. "An Implicit Gradient Method for Cell-Centered Finite-Volume Solver on
Unstructured Grids," AIAA 2019-1155. AIAA Scitech 2019 Forum. January 2019
An Investigation of an implicit gradient method for a second-order cell-centered finite-volume
method on unstructured grids is done by [Nishikawa]220. In the implicit gradient method, solution
gradients are obtained by solving a global system of linear equations, but they can be computed
iteratively along with an implicit finite-volume solver iteration for the Euler or Navier-Stokes
equations. The cost of the implicit gradient computation is thereby made comparable to or even
cheaper than that of explicit gradient methods such as a least-squares method. Furthermore, a known
stability issue with a face-neighbor gradient stencil is effectively circumvented by expanding the
gradient stencil to the entire domain. We discuss the relative performance of the implicit gradient
method and a least-squares method for various inviscid and viscous flow computations on
unstructured grids.
6.9.2.1 Background and Introduction
Convergence and accuracy of implicit finite-volume solvers widely used in practical computational
fluid dynamics (CFD) codes are greatly affected by gradient computation methods, especially on
unstructured grids. Despite great progress made over the last decades, represented by an extensive
study on least-squares gradients in221, current state-of-the-art unstructured-grid finite-volume
solvers still encounter robustness issues when dealing with highly-distorted grids as typical in
practical applications involving complex geometries. In practical solvers, inconsistent gradient
methods such as the Green-Gauss method are locally employed for robustness222-223-224, or simply
gradients are ignored in problematic regions (e.g., in a cell having a large face angle). Also, stable
computations often require different types of gradient methods for different terms, e.g., unweighted
least-squares (LSQ) gradients for the inviscid terms, and weighted LSQ gradients for the viscous
terms and the source terms. Another known problem is that finite-volume schemes with LSQ
gradients computed over face neighbors are unstable for triangular and tetrahedral grids.
Development of gradient methods for increasing robustness and efficiency without degrading
accuracy, therefore, remains an important subject of research. Recently, in a series of papers, a new
approach called the variational reconstruction (VR) method was proposed. In this method, a globally
coupled system of linear equations for the gradients is derived by minimizing the solution jumps at
faces. The system is iteratively solved along with a finite-volume flow solver. If only one iteration is
performed for the gradients per solver iteration, then the cost per iteration is almost the same as the
explicit methods such as the GG and LSQ methods, or more efficient than LSQ methods with extended

219 G. R. Liu and George X. Xu,”A gradient smoothing method (GSM) for fluid dynamics problems”, Int. J. Numerical
Meth. Fluids 2008; 58:1101–1133.
220 Hiroaki Nishikawa, “An Implicit Gradient Method for Cell-Centered Finite-Volume Solver on Unstructured

Grids”, AIAA Sci tech 2019 Forum.

221 Diskin, B. and Thomas, J. L., “Accuracy of Gradient Reconstruction on Grids with High Aspect Ratio," NIA Report

No. 2008-12 , 2008.

222 Mavriplis, D. J., Revisiting the Least-Squares Procedure for Gradient Reconstruction on Unstructured Meshes,"

Proc. of 16th AIAA Computational Fluid Dynamics Conference, AIAA Paper 2003-3986, Orlando, Florida, 2003.
223 Shima, E., Kitamura, K., and Haga, T., Green-Gauss/Weighted-Least-Squares Hybrid Gradient Reconstruction

for Arbitrary Polyhedral Unstructured Grids," AIAA J., Vol. 51, No. 11, 2013, pp. 2740{2747.
224 Moukalled, F., Mangani, L., and Darwish, M., The _nite volume method in computational fluid dynamics: An

intro-duction with OpenFOAM and matlab, Fluid Mechanics and Its Applications, Volume 13, Springer
International Publishing, 2015.
 183

stencils beyond face neighbors. The stability issue mentioned earlier is effectively circumvented by
the gradient stencil spanning the entire grid. A further advantage lies in the straightforward
extension to high-order through minimizing the jumps of high-order polynomials. Inspired by the VR
method, we developed a similar implicit gradient methodology called the Implicit Green-Gauss
(IGG) gradient method225, which was derived from a second-order finite-volume discretization of a
hyperbolic diffusion model. In226, the IGG gradient method was demonstrated, for model advection-
diffusion problems, to produce accurate gradients on highly distorted grids including highly-curved
thin grids, and also allow implicit finite-volume solvers to converge for a discontinuous solution. In
this paper, we investigate the IGG gradient method for inviscid and viscous flow problems.
6.9.2.2 Finite-Volume Discretization
Consider the compressible Navier-Stokes (NS) equations in 2D with a cell-centered finite-volume
discretization is employed for the discretization, where the residual is defined as an approximation
to the NS system integrated over a computational cell j (see Figure 6.31) with the midpoint rule:

𝐑 j = ∑ 𝛟jk 𝐀jk
k∈{kj}
Eq. 6.67
Where kj is a set of neighbors of the cell j, Ajk is the length of the face across j and k, and φjk is a
numerical flux. In this work, the [Roe] and [HLL] fluxes are used for the inviscid terms, and the alpha-
damping flux for the viscous terms. These numerical fluxes are functions of the primitive-variable
gradients, ∇Wj and ∇Wk, and the primitive variables W = (ρ, u, v, p) linearly extrapolated at the face
midpoint from the two adjacent cells j and k:

𝐖L = 𝐖i + 𝚽j ∇W. ∆𝐱 jm ,
𝐖R = 𝐖k + 𝚽𝐤 ∇W. ∆𝐱 km
Eq. 6.68
where Δxjm = xm - xj , Δxkm = xm - xk,
xm is the face-midpoint position, xj
and xk are the centroid coordinates
of the cells j and k, respectively, and
ϕj and ϕk are the Venkat limiter
functions computed based on the
enforcement at nodes. The limiter
function is defined by the
minimum, as typically done, of
those computed for all primitive
variables. The cell gradients, ∇Wj
and ∇Wk, need to be computed
from the numerical solutions
stored at cells. These gradients are
typically computed by a LSQ Figure 6.31 Stencil for Cell-Centered Finite-Volume
method or the Green-Gauss Discretization – (Courtesy of Nishikawa)
method. Here, the IGG gradient
method is employed and its performance is investigated and compared with a LSQ gradient method.

225 Nishikawa, H., From Hyperbolic Diffusion Scheme to Gradient Method: Implicit Green-Gauss Gradients for
Unstructured Grids," J. Comput. Phys., Vol. 372, 2018, pp. 126-160.
226 See previous.
 184

For the viscous flux, these gradients will be used to evaluate the consistent term of the alpha-damping
viscous flux: for example, the x-velocity gradient is given by

1 α
∇u|face = [∇uj + ∇uk ] + (uR − uL )𝐧
̂jk 𝐞jk = 𝐱 k − 𝐱 j
2 |𝐞jk . ̂𝐧jk |
Eq. 6.69
where the first term is the consistent term (i.e., consistently approximating the gradient), the second
term is the damping term (i.e., responsible for damping high-frequency errors) with the damping
coefficient α given an optimal value: α = 4/3, and njk = (nx , ny) is the unit vector normal to the face
pointing from j to k. Further information can be obtained from [Nishikawa]227.
6.9.3 Case Study 2 - Transonic Flow Over a Joukowsky Airfoil at M = 0.8
The next inviscid test case is a transonic flow over a Joukowsky airfoil of a unit chord at M = 0.8 at
the angle of attack 1.25 degrees. In all calculations, we set CFL = 10. The IGG method was applied
with αg = 4.0 to smooth the gradients across shocks and with and without the Venkat limiter.
Otherwise, the same parameter setting was used as in the previous case. The limiter was applied with
the parameter K = 5.0 for the LSQ gradients and K = 10.0 (less limiting) for the IGG gradients.

Figure 6.32 Inviscid Flow over a Joukowsky Airfoil at M1 = 0.85 using Unstructured Mesh - (Courtesy of
Nishikawa)

227Hiroaki Nishikawa, “An Implicit Gradient Method for Cell-Centered Finite-Volume Solver on Unstructured
Grids”, AIAA Sci tech 2019 Forum.
 185

Convergence histories are shown in [Nishikawa]228 where the (Implicit Defect-Correction - least-
squares (IDC-LSQ) solver stalls while the IDC-IGG solver converges with and without the limiter
(denoted by IGG-Limited and IGG, respectively). Note that the IDC-LSQ solver converges on a
structured quadrilateral/triangular grid without a limiter; the divergence without a limiter is
therefore considered due to the irregularity. The limited version of the IDC-IGG solver is slightly
slower in CPU time apparently due to the limiter calculation. Solutions are compared in Figure 6.32.
As expected, the IGG solution without a limiter has over and under shoots around the shock on the
upper surface (see Figure 6.32(e)); and they are greatly reduced by the limiter. Contours of the
limiter function are plotted in Figure 6.32(g)-(i). It is detected that the limiter acts much less on
the IGG gradients than on the LSQ gradients, thus preserving accuracy better with the IGG gradients.
6.9.4 Concluding Remarks
The implicit Green-Gauss gradient method was tested and compared with the least-squares gradient
method for a cell-centered finite-volume method on unstructured grids for inviscid and viscous
problems. The gradients are computed by iteratively solving a global system of linear equations with
the Gauss-Seidel relaxation scheme. An implicit defect-correction solver was tested, where the Gauss-
Seidel relaxation is performed once for the gradient in each nonlinear iteration. Numerical results
have confirmed that the per iteration cost of the implicit Green-Gauss gradient method is comparable
to the LSQ gradient computation. The implicit solver has been demonstrated to converge to a
specified tolerance for all problems (subsonic, transonic, and hypersonic inviscid flows, and low- and
high-Reynolds-number viscous flows) even with shock waves and limiters whereas it stalled with the
least-squares gradients; and it converged faster in CPU time than the solver with the LSQ gradients
in all cases except the at plate case. The solution, however, was found to be more accurate with the
implicit Green-Gauss gradient method.
Despite the demonstrated robustness, the solver needs further improvements. The need for the
relaxation factor adjustment in the IGG relaxation seems to imply that the IDC-IGG solver can become
unstable if the gradients gain accuracy too fast. Apparently, the implicit Euler/NS solver and the
implicit Green-Gauss gradient relaxation are not always compatible with each other. A fully-coupled
solver is one possible strategy to address the issue. There can be, at least, two variants. One is
Newton's method applied to the entire set of residual equations, where the Jacobian is compact and
exact but larger (i.e., 20 x 20 blocks in 3D). Another would be a Jacobian-Free Newton-Krylov (JFNK)
method with the IDC-IGG solver used as a preconditioner. These solvers will be particularly useful
for allowing parameters to be defined locally to adapt the gradients to local ow features while keeping
different values in the Jacobian/preconditioner for robust convergence. A preliminary study
indicates that the Newton solver is much more robust than the IDC-IGG and JFNK solvers, but
requires a very efficient linear solver (e.g., multigrid) to minimize the overall computing time; it
seems suggest that the hyperbolic Navier-Stokes formulation is better suited since the linear
relaxation converges rapidly by the reduced condition number (due to the elimination of second
derivatives).

228Hiroaki Nishikawa, “An Implicit Gradient Method for Cell-Centered Finite-Volume Solver on Unstructured
Grids”, AIAA Scitech 2019 Forum.
186
 187

7 Boundary Types
7.1 Introduction
According to Wikipedia, in the field of differential equations, a boundary value problem (BVP) is a
differential equation together with a set of additional constraints, called the boundary conditions. A
solution to a boundary value problem is a solution to the differential equation which also satisfies the
boundary conditions. To be useful in applications, a boundary value problem should be well posed.
This means that given the input to the problem there exists a unique solution, which depends
continuously on the input. The specification of proper initial conditions (IC) and boundary
conditions (BC) for a PDE is essential, and they are:

➢ If too many IC/BC are specified, there will be no solution.

➢ If too few IC/BC are specified, the solution will not be unique.
➢ If the number of IC/BC is right, but they are specified at the wrong place or time, the solution
will be unique, but it will not depend smoothly on the IC/BC.
➢ This means that small errors in the IC/BC will produce huge errors in the solution.

In any of these cases we have an ill-posed problem229. As an example, [Huang et al.]230 studied the
well possess of inflow/outflow boundary conditions on a subsonic flow reigns for a flat plate. They
demonstrate that some conventional ways of posing subsonic inflow/outflow boundary conditions
are ill-posed.

7.2 Naming Convention for Different Types of Boundaries

Boundary conditions and their correct implementation are among the most critical aspects of a
correct CFD simulation231. Mathematically, there are four types of Dirichlet, Von Neumann, Mixed,
Robin, Cauchy, and Periodic.
7.2.1 Dirichlet Boundary Condition
Direct specification of the variable value at the boundary. E.g.
setting the distribution of a racer ϕi at a west boundary to
zero: ϕw = 0.
7.2.2 Von Neumann Boundary Condition
Specification of the (normal) gradient of the variable at the
boundary. E.g., setting a zero gradient ∂ϕ i /∂n=0 at a
symmetry boundary.
7.2.3 Mixed or Combination of Dirichlet and von Neumann
Boundary Condition
Direct specification of the variable value as well as its
gradient. It is required to satisfy different boundary
conditions on disjoint parts of the boundary of the domain Figure 7.1 Mixed Boundary
where the condition is stated. (see Figure 7.1). Conditions

229 Macintosh HD, Documents, AOSC614-DOCS:PPTClasses, ch3.1: PDEs Well Posed IC&BC. doc Created on
September 26, 2007.
230 Arthur C. Huang, Steven R. Allmaras, Marshall C. Galbraith and David L. Darmofal, “Well-Posed Subsonic

Inflow-Outflow Boundary Conditions for the Navier-Stokes Equations”, 2018 AIAA Aerospace Sciences Meeting.
231 Bakker André, Applied Computational Fluid Dynamics; Solution Methods; 2002.
 188

7.2.4 Robin Boundary Condition

It is similar to Mixed conditions except that a specification is a linear combination of the values of a
function and the values of its derivative on the boundary of the domain232. Robin boundary
conditions are a weighted combination of Dirichlet boundary conditions and Neumann boundary
conditions. This contrasts to mixed boundary conditions, which are boundary conditions of
different types specified on different subsets of the boundary. Robin boundary conditions are also
called impedance boundary conditions, from their application in electromagnetic problems,
or convective boundary conditions, from their application in heat transfer problems. If Ω is the
domain on which the given equation is to be solved and ∂Ω denotes its boundary, the Robin boundary
condition is:

∂u
au + b =g on ∂Ω
∂n
Eq. 7.1
for some non-zero constants a and b and a given function g defined on ∂Ω. Here, u is the unknown
solution defined on Ω and ∂u/∂n denotes the normal derivative at the boundary. More
generally, a and b are allowed to be (given) functions, rather than constants233.
7.2.5 Cauchy Boundary Condition
In mathematics, a Cauchy boundary conditions augments an ordinary differential equation or a
partial differential equation with conditions that the solution must satisfy on the boundary; ideally
so to ensure that a unique solution exists. A Cauchy boundary condition specifies both the function
value and normal derivative on the boundary of the domain. This corresponds to imposing both a
Dirichlet and a Neumann boundary conditions. It is named after the prolific 19th-century French
mathematical analyst Augustin Louis Cauchy234.
7.2.6 Periodic (Cyclic Symmetry) Boundary Condition
Two opposite boundaries are connected and their values are set equal when the physical flow
problem can be considered to be periodic in space. They could be either physical or non-physical in
nature. Among non-physical conditions, inflow, outflow, symmetry plane, pressure and for physical
the wall (fixed, moving, impermeable, adiabatic, etc.). Some vendors choose their boundary to be
reflected by above description, (OpenFOAM®); and some (i.e., CD-Adapco® and Fluent®) to use
their own particular naming, depending to application in hand.
7.2.7 Generic Boundary Conditions
The most widely used generic B.C’s are:

• Walls (fixed, moving, impermeable, adiabatic etc.)

• Symmetry planes
• Inflow
• Outflow
• Free surface
• Pressure
• Scalars (Temperature, Heat flux)
• Velocity
• Internal

232 Gustafson, K., (1998). Domain Decomposition, Operator Trigonometry, Robin Condition, Contemporary
Mathematics, 218. 432–437.
233 Wikipedia,
234 From Wikipedia, the free encyclopedia.
 189

• Pole
• Periodic
• Porous media
• Free-Stream
• Non-Reflecting
• Turbulence-Intensity
• Immersed
• Free Surface

Among others and excellent descriptive available through literature for each.

7.3 Wall Boundary Conditions

All practically relevant flows situations are wall-bounded and near walls the exchange of
mass, momentum and scalar quantities is largest. At a solid wall Stokes flow theory is valid
i.e. the fluid adheres to the wall and moves with the wall velocity. Different treatment for the
different variables in the Navier-Stokes equations is required.
7.3.1 Velocity Field
The fluid velocity components equal the velocity of the wall. The normal and tangential velocity
components at an impermeable, non-moving wall are:

v t = v wall = 0 ; v n = 0 Eq. 7.2

Mass fluxes are zero and hence convective fluxes are zero.

Cwall = 𝑚̇ 𝜑 = 0
Eq. 7.3
Diffusive fluxes are non-zero and result in wall-shear stresses.

→
D wall =  τ n ds Eq. 7.4
ij

7.3.2 Pressure
The specification of wall boundary conditions for the pressure depends on the flow situation. In a
parabolic or convection dominated flow a von Neumann boundary condition is used at the wall:

P
=0 Eq. 7.5
n wall

In a flow with complex curvilinear boundaries, at moving walls, or in flows with considerably large
external forces there may exist large pressure gradients towards the walls. The most common
treatment of such boundaries is a linear extrapolation form the inner flow region. If the exact value
of the pressure at the boundaries is not of interest no boundary conditions are needed when a
staggered grid is used. When a pressure correction method is used, wall boundary conditions are also
needed for pressure correction variable p’. Conservation of mass is only ensured when p’=0 at the
walls. For the purpose of stability this is usually accomplished by a zero gradient condition. The
 190

boundary conditions for the pressure and for the velocity components are valid for both laminar and
turbulent flows. In the case of a turbulent flow near wall gradients are significantly larger and a very
high resolution is required particularly for high Reynolds number flows. Therefore, wall functions
were invented that bridge the near wall flow with adequate (mostly empirical) relationships.
7.3.3 Scalars/Temperature
Direct specification of the scalar/temperature at the wall boundary (Dirichlet Boundary condition)

T(x, t) = Twall Eq. 7.6

Specification of a scalar/temperature gradient i.e. specification of a scalar/temperature flux (von

Neumann Boundary condition):

 T( x , t )
q wall ( x , t ) = − λ Eq. 7.7
n wall

7.3.3.1 Common Inputs for Wall Boundary Condition

• Thermal boundary conditions (for heat transfer calculations).
• Wall motion conditions (for moving or rotating walls).
• Shear conditions (for slip walls, optional).
• Wall roughness (for turbulent flows, optional).
• Species boundary conditions (for species calculations).
• Chemical reaction boundary conditions (for surface reactions).
• Radiation boundary conditions.
• Discrete phase boundary conditions (for discrete phase calculations).
• Wall adhesion contact angle (for VOF calculations, optional).

7.4 Symmetry Planes

Used at the centerline (y= 0) of a 2-D axisymmetric
grid. Can also be used where multiple grid lines
meet at a point in a 3D O type grid. They used in CFD
simulations to reduce the numerical effort (see Figure
7.2). Must be used carefully and only when both
geometry and flow are symmetrical. Unsteady flows
around symmetrical obstacles are always asymmetric:
e.g. flow around a square obstacle. Steady flows in
symmetrical diffusers or channel expansions can be
asymmetric and symmetry conditions should only be
used when an asymmetric flow can be excluded a Figure 7.2 Symmetry Plane to Model
one Quarter of a 3D Duct
priori. At a symmetry boundary the following
conditions apply:

➢ The boundary normal component of the velocity disappears and the flux through the
boundary is zero:

⃗Vn = 0 , Csym = ṁ φ = 0
Eq. 7.8
 191

➢ Scalars have all zero gradients. Consequently the diffusive fluxes of the scalars are also
zero:
φ
= 0 ; D sym = 0 Eq. 7.9
n
➢ The boundary normal gradient of tangential velocity components is also zero. As a result,
the shear stresses disappear

7.5 Inflow Boundaries

An inflow boundary is an artificial boundary that is used in CFD simulations because the
computational domain must be finite. Proper use of inflow boundary conditions can reduce the
numerical effort and need to be selected carefully so that the flow physics is not altered. At the inflow
usually variables are specified directly i.e. Dirichlet condition. The convective fluxes can be computed
and are added to source term. Diffusive fluxes are computed and added to the central coefficient AP.
Common inflow boundaries are: Pressure inlet, Velocity inlet, Mass flow inlet, among others.
7.5.1 Velocity Inlet
This types of boundary conditions are used to define the velocity and scalar properties of the flow at
inlet boundaries. The contribution inputs usually includes:

• Velocity magnitude and direction or velocity components

• Rotating (Swirl) velocity (for 2D axisymmetric problems with swirl)
• Temperature (for energy calculations)
• Turbulence parameters (for turbulent calculations)
• Radiation parameters
• Chemical species mass fractions (for species calculations)
• Mixture fraction and variance (for non-premixed or partially premixed combustion
calculations)
• Discrete phase boundary conditions (for discrete phase calculations)
• Multiphase boundary conditions (for general multiphase calculations)
7.5.2 Pressure Inlet
These boundary conditions are used to define the total pressure and other scalar quantities at flow
inlets. Required inputs are:

• Total (stagnation) Pressure

• Total (stagnation) Temperature
• Flow direction
• Static pressure
• Turbulence parameters (for turbulent calculations)
• Radiation parameters
• Chemical species mass fractions (for species calculations)
• Mixture fraction and variance (for non-premixed or partially premixed combustion
calculations)
7.5.3 Mass Flow Inlet
These boundary conditions are used in compressible flows to prescribe a mass flow rate at an inlet.
It is not necessary to use mass flow inlets in incompressible flows because when density is constant,
velocity inlet boundary conditions will fix the mass flow. Some of the common inputs are:
 192

• Mass flow rate, mass flux, or (primarily for the mixing plane model) mass flux with average
mass flux
• Total (stagnation) temperature
• Static pressure
• Flow direction
• Turbulence parameters (for turbulent calculations)
• Radiation parameters
• Chemical species mass fractions (for species calculations)
• Mixture fraction and variance (for non-premixed or partially premixed combustion
calculations)
• Discrete phase boundary conditions (for discrete phase calculations)
• Open channel flow parameters (for open channel flow calculations using the VOF multiphase
model)
7.5.4 Inlet Vent
boundary conditions are used to model an inlet vent with a specified loss coefficient, flow direction,
and ambient (inlet) total pressure and temperature.

7.6 Outflow Boundaries

An outflow boundary is also an artificial boundary that is used in CFD simulations because the
computational domain must be finite. The location of the outflow boundary must be sufficiently
downstream of the region of interest. At the outlet boundary recirculation zones may not be present
and streamlines should be parallel. The mathematical formulation of the boundary condition may not
influence the flow in the inner part of the domain. Zero gradient conditions are most widely used for
all variables. The outlet boundary is usually used to check global mass conservation during an
iterative process. Commonly used outflow boundaries include: Pressure outlet, Pressure far-field,
Outlet vent, and Exhaust fan.
7.6.1 Pressure Outlet
These boundary conditions are used to define the static pressure at flow outlets (and also other scalar
variables, in case of back flow). The use of a pressure outlet boundary condition instead of an out
flow condition often results in a better rate of convergence when back flow occurs during iteration.
The contributions inputs requires are:
• Static pressure
• Backflow conditions
• Total (stagnation) Temperature (for energy calculations)
• Backflow direction specification method
• Turbulence parameters (for turbulent calculations)
• Chemical species mass fractions (for species calculations)
• Mixture fraction and variance (for non-premixed or partially premixed combustion
calculations)
• Multiphase boundary conditions (for general multiphase calculations)
• Radiation parameters
• Discrete phase boundary conditions (for discrete phase calculations)
• Open channel flow parameters (for open channel ow calculations using the VOF
• multiphase model)
• Non-reflecting boundary (for compressible density-based solver)
• Target mass flow rate (not available for multiphase flows)
 193

7.6.2 Pressure Far-Field

boundary conditions are used to model a free-stream compressible flow at in unity, with free-stream
Mach number and static conditions specified. This boundary type is available only for compressible
flows. Inputs are:
• Static pressure.
• Mach number.
• Temperature.
• Flow direction.
• Turbulence parameters (for turbulent calculations).
• Radiation parameters.
• Chemical species mass fractions (for species calculations).
• Discrete phase boundary conditions (for discrete phase calculations).
7.6.3 Outflow
Boundary conditions are used to model flow exits where the details of the flow velocity and pressure
are not known prior to solution of the flow problem. They are appropriate where the exit flow is close
to a fully developed condition, as the outflow boundary condition assumes a zero normal gradient
for all flow variables except pressure. They are not appropriate for compressible flow calculations.
7.6.4 Outlet Vent
boundary conditions are used to model an outlet vent with a specified loss coefficient and ambient
(discharge) static pressure and temperature. The inputs are:
• Static pressure
• Backflow conditions
• Total (stagnation) temperature (for energy calculations)
• Turbulence parameters (for turbulent calculations)
• Chemical species mass fractions (for species calculations)
• Mixture fraction and variance (for non-premixed or partially premixed combustion
calculations)
• Multiphase boundary conditions (for general multiphase calculations)
• Radiation parameters
• Discrete phase boundary conditions (for discrete phase calculations)
• Loss coefficient
• Open channel flow parameters (for open channel flow calculations using the VOF
multiphase model)
7.6.5 Exhaust Fan
Boundary conditions are used to model an external exhaust fan with a specified pressure jump and
ambient (discharge) static pressure.

• Static pressure
• Backflow conditions
• Total (stagnation) temperature (for energy calculations)
• Turbulence parameters (for turbulent calculations)
• Chemical species mass fractions (for species calculations)
• Mixture fraction and variance (for non-premixed or partially premixed combustion
calculations)
• Multiphase boundary conditions (for general multiphase calculations)
• User-defined scalar boundary conditions (for user-defined scalar calculations)
 194

• Radiation parameters
• Discrete phase boundary conditions (for discrete phase calculations)
• Pressure jump
• Open channel flow parameters (for open channel ow calculations using the VOF
multiphase model).

7.7 Free Surface Boundaries

7.7.1 Velocity Field and Pressure
Free surface boundaries can be rather complex and the location of the free surface is usually not
known a-priori. E.g. the swash of a fluid in a tank, the pouring of liquid into a glass. Only at the initial
time the position of the free surface is known and in the following an additional transport equation
to determine the location of the free surface is needed. Two boundary conditions apply at the free
surface boundary:

• Kinematic boundary condition - Fluid cannot flow through the boundary. i.e. the normal
component is equal to the surface velocity.
• Dynamic boundary condition - All forces that are acting on the free surface have to be in
equilibrium. These include shear stresses from the fluid below the surface and possibly
from a second fluid on the other side fluid and surface tension235.

In many CFD applications the free surface is treated as a flat plane where the symmetry condition is
applied.
7.7.2 Scalars/Temperature
Treated in an analogue manner as the wall boundary condition. Direct specification i.e. Dirichlet
boundary conditions or von Neumann boundary conditions or a combination of both.

7.8 Pole (Axis) Boundaries

Used at the centerline (y = 0) of a 2-
D axisymmetric grid (see Figure
7.3). It can also be used where
multiple grid lines meet at a point in
a 3-D O-type grid. No other inputs
are required.

7.9 Periodic Flow Figure 7.3 Pole (Axis) Boundary

Boundaries
Periodicity simply corresponds to matching conditions on the two boundaries. The velocity field is
periodic BUT the pressure field is not. The pressure gradient drives the flow and is periodic. A
pressure JUMP condition on the boundary must be specified236. Used when physical geometry of
interest and expected flow pattern and the thermal solution are of a periodically repeating nature
(see Error! Reference source not found.).

235 Georgia Tech Computational Fluid Dynamics Graduate Course; spring 2007.
236 Solution methods for the Incompressible Navier-Stokes Equations.
 195

Figure 7.4 Periodic Boundary

7.9.1 Case Study – Boundary Condition on A 2-D HPT Blade Subject to Strong Pressure Gradient
and Curvature Effects
Here, we present an study done by [Zhao an Sandbergd]237 as it investigate the bypass transition in
an HPT working at conditions representative of a modern aircraft engine (Sandberg & Michelassi

Figure 7.5 Schematic for HPT case setup. The computational grid is showing every fifteenth
lines in x and y directions.

Yaomin Zhao, Richard D. Sandberg, “Bypass transition in boundary layers subject to strong pressure gradient
237

and curvature effects”, Journal of Fluid Mechanics · April 2020.

 196

2019). By performing highly

resolved simulations with up to
8.6 × 108 grid points on a O-H
type multiblock grid, the flow
physics can be investigated in
detail. In particular, the effects of
the (Free Stream Turbulence) FST
with strong intensities and large
integral length scales. A linear
cascade setup is used to replicate
the experiments performed at the
Von-Karman Institute (VKI). In
the present simulations, only one
passage is calculated, and the
computational domain is
enclosed by the red lines on the
axial and pitch wise (x − y) plane Figure 7.6 Overview of the Flow Field
cut (see Figure 7.5). Contours of
the mean velocity magnitude U, along with the isolines of the mean pressure p, are shown in Figure
7.6. Through the HPT vane, the flow accelerates significantly due to the pressure ratio between the
inlet and the outlet. From the blade leading edge, it first passes through a region of high-curvature
and strong pressure gradient marked by the arrow M1. Then the flow acceleration continues until
the minimum of p, marked by the arrow M2, which is followed by the APG region on the blade suction
side. Moreover, a pressure dipole marked by M3 is caused by the vortex shedding at the blade trailing
edge.

7.10 Non-Reflecting Boundary Conditions (NRBCs)

Many problems in computational fluid dynamics occur within a limited portion of a very large or
infinite domain. Difficulties immediately arise when one attempts to define the boundary condition
for such a system. Such boundary conditions are necessary for the problem to be well-posed, but the
physical system under consideration has no boundary to model. One needs to define an artificial
boundary whose behavior models the open edge of the physical system. Such a boundary definition
is often called a non-reflecting boundary condition (NRBC), as its primary function is to permit wave
phenomena to pass through the open boundary without reflection. The standard pressure boundary
condition, imposed on the boundaries of artificially truncated domain, results in the reflection of the
outgoing waves. As a consequence, the interior domain will contain spurious wave reflections. Many
applications require precise control of the wave reflections from the domain boundaries to obtain
accurate flow solutions. Non-reflecting boundary conditions provide a special treatment to the
domain boundaries to control these spurious wave reflections. The method is based on the Fourier
transformation of solution variables at the non-reflecting boundary238. The solution is rearranged as
a sum of terms corresponding to different frequencies, and their contributions are calculated
independently. While the method was originally designed for axial turbomachinery, it has been
extended for use with radial turbomachinery. In many applications of CFD such as Turbomachinery
because of close approximately of blades and the physical conditions, it is warranted to use NRBC’s.
Another prime candidate is Computational Aero-Acoustics (CAA) which is concerns with propagation
of traveling sound waves. In other word, by restricting our area of interest, we effectively create a
boundary where none exists physically, dividing our computational domain from the rest of the

M. Giles, “Non-Reflecting Boundary Conditions for the Euler Equations.”, Technical Report TR 88-1-1988,
238

Computational Fluid Dynamics Laboratory, Massachusetts Institute of Technology, Cambridge, MA.

 197

physical domain. The challenge we must overcome, then, is defining this boundary in such a way that
it behaves computationally as if there were no physical boundary239.
7.10.1 Case Study 1 - Turbomachinery Application of 2D Subsonic Cascade
The first test case is an axial turbine blade where both the in- and outflow are subsonic and the NRBC
will be compared to the Riemann boundary conditions. In the short flow-field simulations the in- and
outflow boundaries are positioned at 0.4 times the chord from the airfoil. For the long flow-field
simulation this distance becomes 1.5 times the chord. Figure 7.7 shows contour plot of the pressure
of the flow. The field of interest is the flow-field close to the boundary240. To give a detailed look at
that part of the flow, the pressure contours are put in close proximity. Unfortunately this means the
flow-field at the suction side becomes less clear. The subsonic flow means that any reflections diffuse
fairly quickly. Therefore there are almost no observable differences when the long flow-field is
considered. For the short flow-field the reflections become more apparent when Riemann boundary
conditions are used. At the outflow the pressure contours are clearly deflected away from the
boundary and never cross it. At the inflow the opposite happens and the pressure contours are bend
towards the boundary. This behavior is not observed when looking at the NRBC. Clearly these
boundary conditions are successful in removing the reflections from the flow. One can have a closer
look at the boundary itself to further clarify this comparison. The pressure at the outflow boundary
presented in Figure 7.7, where we notice that the NRBC do a better job of simulating the pressure
at the outflow, although it should be noted that on the absolute scale, all the differences are very
small.

Figure 7.7 Pressure contours plot for 2nd order spatial discretization scheme

239 John R. Dea, “High-Order Non-Reflecting Boundary Conditions for the Linearized Euler Equations”, Monterey,
California, 2008.
240 F. De Raedt, “Non-Reflecting Boundary Conditions for non-ideal compressible fluid flows”, Master of Science at

the Delft University of Technology, defended publicly on December 2015.

 198

7.10.2 Case Study 2 - CAA Application of Airfoil Turbulence Interaction Noise Simulation
The instantaneous contours of the non-dimensional pressure that is radiated from the airfoil due to
the turbulence interaction mechanism (see Figure 7.8). In each case, the entire simulated domain
is shown. It is qualitatively displays that the acoustic pressure waves do not appear to be acted by
the edges of the domain, and are not acted by the changes in domain size between the two
simulations. An exception to this is at the domain edge directly downstream of the airfoil. In this
region, unphysical pressure disturbances can be seen that correspond to the vortical turbulence
encountering the NRBC’s region. However, because these pressure disturbances appear inside the
zonal NRBC region, they are contained and do not radiate back into the domain241.

Domain X Domain X/2

Figure 7.8 Aero-Acoustics Application for NRBC’

7.11 Turbulence Intensity Boundaries

When turbulent flow enters domain at inlet, outlet, or at a far-field boundary, boundary values are
required for242:
➢ Turbulent kinetic energy k.
➢ Turbulence dissipation rate ε.
Four methods available for specifying turbulence parameters:
➢ Set k and ε explicitly.
➢ Set turbulence intensity and turbulence length scale.
➢ Set turbulence intensity and turbulent viscosity ratio.
➢ Set turbulence intensity and hydraulic diameter.
7.11.1 Turbulence Intensity
The turbulence intensity I defined as:
√2/3k
I=
u
Eq. 7.10
Here k is the turbulent kinetic energy and u is the local velocity magnitude. Intensity and length scale
depend on conditions upstream:
➢ Exhaust of a turbine. (Intensity = 20%. Length scale=1-10 % of blade span).

241 James Gill, Ryu Fattah, and Xin Zhangz, “Evaluation and Development of Non-Reactive Boundary Conditions
for Aeroacoustics Simulations”, University of Southampton, Hampshire, SO16 7QF, UK.
242 Bakker, Andre, ”Applied Computational Fluid Dynamics; Lecture 6 - Boundary Conditions”, 2002.
 199

➢ Downstream of perforated plate or screen (intensity=10%. Length scale = screen/hole size).

➢ Fully-developed flow in a duct or pipe (intensity= 5%. Length scale = hydraulic diameter).

7.12 Free Surface Boundary

Free surfaces occur at the interface between two fluids. Such interfaces require two boundary
conditions to be applied243:
• A kinematic condition which relates the motion of the free interface to the fluid velocities at
the free surface and
• A dynamic condition which is
concerned with the force balance
at the free surface.
7.12.1 The Kinematic Boundary
Condition
The position of a free surface can always
be given in implicit form as F(xj , t) = 0.
For instance, in Figure 7.9 the height of
the free surface above the x-axis is
specified as y = h(x, t) and an appropriate
function F(x, y, t) would be given by F(x,
y, t) = h(x, t) − y. Fluid particles on the
free surface always remain part of the
free surface, therefore we must have: Figure 7.9 Sketch Exemplifying the conditions at a Free
Surface Formed by the Interface Between Two Fluids
DF ∂F ∂F
= + uk
Dt ∂t ∂xk
Eq. 7.11
This is the kinematic boundary condition. For surfaces whose position is described in the form z =
h(x, y, t), the kinematic boundary condition becomes

∂h ∂h ∂h
w= +u +v
∂t ∂x ∂y
Eq. 7.12
where u, v, w, are the velocities in the x, y, z directions, respectively. For steady problems, we have
DF/Dt = 0 and the kinematic boundary condition can be written as uini = 0 or symbolically u · n = 0,
where n is the outer unit normal on the free surface. This condition implies that there is no flow
through the free surface (but there can be a flow tangential to it!).
7.12.2 The Dynamic Boundary Condition
The dynamic boundary condition requires the stress to be continuous across the free surface which
separates the two fluids (e.g., air and water). The traction exerted by fluid (1) onto fluid (2) is equal
and opposite to the traction exerted by fluid (2) on fluid (1). Therefore we must have t(1) = −t(2).
Since n(1) = −n(2) (see Figure 7.9) we obtain the dynamic boundary condition as

τ1ij nj = τ2ij nj
Eq. 7.13

243 “Viscous Fluid Flow: Boundary and initial conditions”, Lecture Series, Manchester, UK.
 200

where we can use either n(1) or n(2) as the unit normal. On curved surfaces, surface tension can
create a pressure jump across the free surface. The surface tension induced pressure jump is given
by
1 1
∆p = σκ , κ= +
R1 R 2
Eq. 7.14
In this expression σ is the surface tension of the fluid and κ is equal to twice the mean curvature of
the free surface, where, R1 and R2 are the principal radii of curvature of the surface (for instance, κ
= 2/a for a spherical drop of radius a and κ = 1/a for a circular jet of radius a). Surface tension acts
like a tensioned membrane at the free surface and tries to minimize the surface area. Hence the
pressure inside a spherical drop (or inside a circular liquid jet) tends to be higher than the pressure
in the surrounding medium. If surface tension is important, the dynamic boundary condition has to
be modified to
τ1ij n𝑗1 + σκn1𝑖 = τ2ij n𝑗1
Eq. 7.15
where κ > 0 if the centers of curvature lie inside fluid (1).

7.13 Other Boundary Conditions

Other boundary conditions can occur in special applications. For instance, the presence of an elastic
boundary leads to fluid-structure interaction problems in which the fluid velocity has to be equal to
the velocity of the elastic wall, while the elastic wall deforms in response to the traction that the fluid
exerts on it. At porous walls, the no-penetration condition no longer holds: the volume flux into the
wall is often proportional to the pressure gradient at the porous surface. Non-uniformly distributed
surfactants (substances which reduce the surface tension) can induce tangential stresses at free
surfaces, etc.

7.14 Further Remarks

For an incompressible fluid, the boundary conditions need to fulfill the overall consistency condition

∮ ui ni dS = 0
∂V
Eq. 7.16
where ∂V is the surface of the spatially fixed volume in which the equations are solved. If there are
no free surfaces (and associated dynamic boundary conditions), the pressure is only defined up to an
arbitrary constant as only the pressure gradient (but not the pressure itself) appears in the Navier-
Stokes equations. For initial value problems, the initial velocity field (at t = 0) already has to fulfill the
incompressibility constraint. These remarks are particularly important for the numerical solution of
the Navier-Stokes equations244.

244 “Viscous Fluid Flow: Boundary and initial conditions”, Lecture Series, Manchester, UK.
201
 202

8 Temporal Discretization and Time Marching

To account for transient effects, the governing equations must be discretized in time245. As it turns
out, the temporal discretization is slightly easier to deal with than that for the spatial effects. Since
the governing equation is hyperbolic/parabolic in time, the solution at time t depends upon its
history and not on its future. Temporal discretization is considered as one sided, while spatial is
two sided. Considering heat conduction where the temperature change in one point influences all
spatial neighbors. Whereas time is always one way and as an example, temperature change at any
point is only affected by current and past temperatures. Transient effects are usually dealt with by
using a time stepping procedure, with an initial condition provided. The time dimension is divided
into a set of discrete time steps, each of size ∆t. The solution algorithm therefore marches forward
in time, computing a solution at each time step. The spatial discretization for the time-dependent
equations is identical to the steady-state case. Temporal discretization involves the integration of
every term in the differential equations over the time step ∆t. The integration of transient effects
takes several forms, each yielding a different accuracy. For simplicity, we express the time dependent
transport of a scalar ϕ

∂(ρϕ)
= L(ϕ)
∂t
Eq. 8.1
Where the function L is an operator that incorporates all of the non-transient terms, namely:
diffusion, convection, and source terms. Integrating the above equation over a control volume yields

𝛛(𝛒𝛟)
∫ dV = ∫𝐋(𝛟)dV = 𝐋(𝛟)
𝐕 𝛛𝐭 𝐕

Eq. 8.2
After the spatial discretization has been performed using the techniques described in the other
sections, we obtain where L denotes the spatial discretization operator (the discretized diffusion,
convection, and source terms), and V denotes the volume. At this point, we make an important
observation. The temporal discretization of the transient term (LHS) need not be the same as that
of the discretized diffusion, convection and source terms (RHS). Each term can be treated
differently yield different accuracies. We are now ready to perform the transient discretization. In
the framework of the finite volume method, there are various methods that can be used to perform
this task, the most popular of which are the Euler Implicit, Crank-Nicolson, Fully Implicit schemes,
Runge Kutta Method and the 2nd order Backward Difference Formula (BDF2). The temporal
discretization of the diffusion, convection, and source terms will be presented first, followed by the
methods used for the transient term. We assume the values at a given control volume are known at
an initial time t and we are interested in obtaining the values at time t+∆t. This method states that
the time integral of a given variable is equal to a weighted average between existing and future values.
For a given control volume ϕ, the RHS of the general discretized equation L(ϕ) contains terms
involving values at ϕ and its neighboring control volumes. The temporal discretization is carried out
through an integration over time of the RHS where each unknown is involved in the process. An
integration over time of the RHS where each unknown is involved in the process.

245 CFD on Line discussion

 203

𝐭+Δt
∫ φ dt ≈ [f φ𝐭+Δt + (𝟏 − 𝐟) φ𝐭 ] Δt
𝐭
Eq. 8.3
Where f is a weighing factor between 0 and 1. The following holds:
• f = 0 results in the fully explicit scheme
• f = 1 results in the fully implicit scheme
• f = 0.5 results in the Crank-Nicolson scheme
The product of most frequently schemes are described below. Other notable schemes are time spilt
McCormack method, Lax-Wendorff, and ADI, as well as many more as described in various text
books.

8.1 Transient 1D Diffusion with Source Term

Retaining the time derivative in the conservation equation we get,

d  dU  U
k  + Qv = Eq. 8.4
dx  dx  t

The right hand side includes effects of the source term and the diffusion term.

t + Δt
U U
 (f + (1 − f) t ) Δt
d  dU 
k  + Qv = or =  (x, t) = t + Δt
Eq. 8.5
dx  dx  t t t

where the right hand side includes effects of the source term and the diffusion term. The transient
term on the left can expressed simple explicit method as

U U
1 t

=  dtdx  (Ut1 - U t 0 ) x Eq. 8.6

t t0
t

Or, one of following methods.

8.1.1 Euler Implicit
The 1st order accurate Euler Implicit is probably the most used techniques in transient simulation
because of its simplicity. It is
 n +1 −  n
= L( n +1 ) Eq. 8.7
Δt

Where L denote the spatial discretization. This is unconditionally stable and results in a tridiagonal
system of algebraic equations to be solved by Thomas algorithm.
8.1.2 Crank-Nicolson
This is also an implicit scheme and like previous method it is unconditionally stable, but 2nd order
 n +1 −  n
accurate in time. The equations can be written as = L( n ,  n +1 ) Eq. 8.8
Δt
 204

Figure 8.1 Numerical Stability of 1D Heat Equation using Explicit & Implicit Algorithms (Wikipedia)
 205

Where again L is an operator of spatial discretization and the resulting equations yields a tridiagonal
system of linear algebraic equations.
8.1.3 Simple Explicit
The following explicit scheme which is 1st order accurate in time is presented as

 n +1 −  n
= L( n ) Eq. 8.9
Δt

Although the explicit schemes are not recommended because of the obvious reasons, nevertheless, it
is mentioned as completeness.
Case Study – Numerical Solution of 1D Heat Model Equation
Generally, the explicit schemes tend to become more unstable than the implicit ones. For example,
consider a 1D Heat equation as

1
ut = αuxx where α = , B. C. ∶ u(0, t) = u(1, t) = 0
π2
Eq. 8.10
with aid of Separation of Variables we get the exact solution as

1 −t
u(x, t) = e sin(πx)
π2
Eq. 8.11
Using Eq. 8.10 and Eq. 8.11, together with Table 4.2 and Table 4.1, we get the result that using
explicit schemes would be not be stable, while implicit schemes (i.e., Euler implicit or Crank
Nicholson) yield to numerically stable solution, as demonstrated Figure 8.1.
8.1.4 Runga-Kutta Method
In numerical analysis, the Runge–Kutta methods are a family of implicit and explicit iterative
methods, which includes the well-known routine called the Euler Methods (yn+1 = yn+ hf(xn,yn)), used
in temporal discretization for the approximate solutions of ordinary differential equations246. These
methods were developed around 1900 by the German mathematicians [C. Runge] and [M. W. Kutta].
8.1.4.1 Explicit Treatment
The family of explicit Runge–Kutta methods is a generalization of the method mentioned above. It is
given by:

k1 = f(tn , y n ),

k 2 = f(tn + c 2 h, y n + h(a 21k1 )),
s

y n +1 = y n + h  bi k i k 3 = f(tn + c3h, y n + h(a 31k1 + a 32 k 2 ))
i =1 . . . .

k s = f(tn + cs h, y n + h(a s1k1 + a s2 k 2 + ....... + a s,s −1k s −1 ))
Eq. 8.12

246 “Runge–Kutta methods”, From Wikipedia, the free encyclopedia.

 206

To specify a particular method, one needs to provide the integer s (the number of stages), and the
coefficients aij (for 1 ≤ j < i ≤ s), bi (for i = 1, 2,..., s) and ci (for i = 2, 3,.., s). The matrix [aij] is called the
Runge–Kutta matrix, while the bi and ci are known as the weights and the nodes defined by

i −1

a
j=1
ij = ci for i = 2,......, s Eq. 8.13

By far the most often used is the classical 4th order Runge-Kutta formula RK4, which has a certain
sleekness of organization about it247. Therefore, fourth-order Runge-Kutta method requires four
evaluations of the right hand side per each step size h. (see Eq. 8.14).

k 1 = h f (x n , y n ) 
 
k = h f  x + h , y + k 1  


2

n
2
n
2 

y n +1 = y n + h f(x n , y n )  
k = h f  x + h , y + k 2 
 3 
n
2
n
2 
 
k 4 = h f (x n + h, y n + k 3 ) 
 

y n +1 = y n +
k1 k 2 k 3 k 4
6
+
3
+
3
+
6
+  h5 ( )
Eq. 8.14
8.1.4.2 Implicit Runge-Kutta Methods
All Runge-Kutta methods mentioned up to now are explicit methods. Explicit Runge–Kutta methods
are generally unsuitable for the solution of stiff equations because their region of absolute stability
is small; in particular, it is bounded. This issue is especially important in the solution of partial
differential equations. The instability of explicit Runge–Kutta methods motivates the development of
implicit methods. An implicit Runge–Kutta method has the form

s  s 
y n +1 = y n +  b i k i where k i = f  t n + ci h, y n + h  a ijk i  Eq. 8.15
i =1  j=1 
The difference with an explicit method is that in an explicit method, the sum over j only goes up to i-
1. The coefficient aij of an explicit method is lower triangular. In an implicit method, the sum over j
goes up to s and the coefficient matrix is not triangular. The consequence of this difference is that at
every step, a system of algebraic equations has to be solved. This increases the computational cost
considerably. If a method with s stages is used to solve a differential equation with m components,
then the system of algebraic equations has ms components. This can be contrasted with implicit
linear multistep methods (the other big family of methods for ODEs): an implicit s-step linear
multistep method needs to solve a system of algebraic equations with only m components, so the size
of the system does not increase as the number of steps increase.

247“Numerical Recipes In C: The Art of Scientific Computing”, (ISBN 0-521-43108-5), Copyright (C) 1988-1992
by Cambridge University Press.
 207

8.1.4.3 Case Study - Time Integration Schemes for the Unsteady NS Equations
The focuses here is on the efficiency of higher-order Runge-Kutta schemes in comparison with the
popular Backward Differencing Formulations (BDF). For this comparison an unsteady two-
dimensional laminar, using a thin-layer N-S equation (TLNS3D) code is chosen with a test case of flow
around a circular cylinder at Re=1200. It is concluded that for realistic error tolerances, 4th and 5th
order Runge Kutta schemes are the most efficient. Therefore, for reasons of robustness and computer
storage, the 4th order Runge-Kutta method is recommended. The efficiency of the 4th order Runge-
Kutta scheme exceeds that of 2nd order Backward Difference Formula (BDF2) by a factor of 2.5 at
engineering error tolerance levels (10-1 - 10-2). Efficiency gains are more dramatic at smaller
tolerances. For a constructive discussion, readers should follow the paper by [Bijl, et al.]248.
8.1.5 Piecewise Analytical Method
The new method of solving DE is called piecewise analytic method (PAM). The PAM is based on
dividing the solution interval into subintervals. PAM gives an approximate analytic solution which is
very accurate and can be applied to each subinterval successively [Abassy]249 . The piecewise analytic
method can be made to be of any order of accuracy. PAM gives very interesting results especially
when the solution contains singular points.

8.2 Transient Simulation

Alternatively, the transient (un-steady) scheme could be visualized as series of quasi-Steady-State
solutions to be advanced temporally until pre-defined run time reached (dual time stepping), depicts
such solution propagation though time steps250. Unlike steady-state where the solution is advanced
until residuals fall below pre-defined tolerance level, in transient analysis, there is really no pre-set
convergence criteria defined expect the duration time. The transient residuals although should fall
but could exhibit certain small oscillations, characteristic of unsteady flow. As long as that oscillation
are small enough to limit the temporal approximation errors. In fact the solution sequence which
outlined in Figure 8.2-(left) could be translated to Figure 8.2-(right) which displays residual plots
for a transient flows, .”,

Figure 8.2 Transient Solution Sequences

248 Hester Bijl, Mark H. Carpenter, Veer N. Vatsa, “Time Integration Schemes for the Unsteady Navier-Stokes
Equations”, AIAA-2001.
249 Tamer A. Abassy, “Piecewise Analytic Method”, Int. Journal of Applied Mathematical Research, 2012.
250 Bakker André, Applied Computational Fluid Dynamics; Solution Methods; 2002.
 208

8.2.1 Time Step Size

The choice of time step is of course most important and should be elected carefully. It is restricted by
CFL number on explicit methods. For fully implicit methods, although in principal it could be any
value, in practice other considerations impose limit as previously indicated. This limit, produces in
terms of Δt < h/v where h is the size of smallest cell, and v is typical characteristic flow velocity.
Usually there are options of static (constant) step size, or more robust dynamic evaluation as dictated
internally by the solver.
8.2.2 Steady-State Vs. Transient Run
"How do I know in advance whether to perform a steady-state or an unsteady CFD simulation?" The
simple answer is, "you don't know", so it is imperative to provide some help on when to use unsteady
(also known as transient or time-dependent) simulations251. When you run a steady-state simulation
it uses an iterative scheme to progress to convergence. If you see persistent oscillations in the
residuals plot (solver diagnostics) or oscillations in a key monitor, such as a drag force monitor, with
increasing iterations then that's a good indicator the flow may be unsteady (transient) and the
simulation needs to be run as an unsteady. When you are unsure as to whether your simulation is
unsteady or steady-state it is always worth running a steady-state simulation first because it typically
takes an order of magnitude less CPU time to complete. If the steady-state simulation is sufficient
then you'll save a lot of time over running an unsteady simulation. As an example let us examine the
residual and force monitors for an unsteady simulation of vortex shedding behind a cylinder (see
Figure 8.3 (a)) and compare them with those generated by the same simulation running,
incorrectly, in steady-state mode. Simply put, the rule is that the flow is transient unless proven
otherwise, as many flows in nature are transient.

(a) Pressure
Contours

(b) Steady-State
Force Monitor for
a flow pass a Circle

(c) Steady-State
Residual Monitor
for a flow pass a
Circle

Figure 8.3 Vortex shedding behind a cylinder run wrongly as a Steady State

251 “Steady-State or Unsteady CFD Simulation?” Symscape, Computational Fluid Dynamics Software for All.
 209

8.2.2.1 Case Study 1 - Vortex Shedding Behind a Cylinder - Ran as Steady State (Wrong!)
Remember that vortex shedding behind a cylinder is unsteady as described in the previous section,
so it is interesting to see what happens when you try to run the same simulation in steady-state mode.
Clearly after only 20 updates (200 iterations) there is something wrong, as expected, indicated by
the widely oscillating lift force. For a steady-state simulation you'd expect some oscillations in the lift
force values when you first start the simulation and then see them damp out to a fixed value with
increasing iterations. Something is also wrong, as expected, with the residuals. For a steady-state
simulation you should see all residuals reduce relatively smoothly toward or below 1e -3 with
increasing iterations (see Figure 8.3: (b)-(c)).
8.2.2.2 Case study 2 - Double Sided Membrane - Ran as Steady State (Right!)
For a steady-state simulation the residuals and force profile should be something similar to those in
the tutorial "Flow Over a Double-Sided Membrane". Note the converged lift and drag force values
with increasing iterations252. The relatively smooth reduction below 1e-3 in all the residuals with
increasing iterations. In conclusion, if you run a steady-state simulation for an unsteady flow then
you will get poor results, but as I just pointed out there is usually a good indication from the steady
state results that an unsteady simulation was necessary. However, if you find that that your
simulation has a steady-state then you can avoid the unnecessary expense of performing an unsteady
simulation (see Figure 8.4).

(a) Double Sided

Membrance

(b) Steady-State
Force Monitor

(c) Steady-State
Residual Monitor

Figure 8.4 Steady-State Force and Monitoring for a flow over a double sided membrane case

8.2.2.3 Case Study 3 – Evolution of the Wake of 3 Inline Square Prisms

Citation : Zheng, Qinmin and Alam, Md. Mahbub. “Evolution of the wake of three inline square prisms”,
Phys. Rev. Fluids, 4, 10, 2019, American Physical Society, DOI : 10.1103/PhysRevFluids.4.104701,
https://link.aps.org/doi/10.1103/PhysRevFluids.4.104701

252 “Steady-State or Unsteady CFD Simulation?” Symscape, CFD Software for All.
 210

Expanding on that, a study of the flow around three tandem square prisms, conducted by [Zheng &
Mahbub Alam]253, may provide us a better understanding of complicated flow physics related to
multiple closely spaced structures. In this paper, a numerical investigation on the flow around three
tandem square prisms at a Reynolds number Re =(U∞W/ν) = 150 is conducted for L/W = 1.2–10.0,
where U∞ is the freestream velocity, W is the prism width, L is the prism center-to-center spacing,
and ν is the kinematic viscosity of the fluid. Extensive analyses are done of flow structures, Strouhal
numbers St, fluid forces, and vorticity, velocity, and pressure fields. Four distinct flow regimes and
their ranges are identified, viz., single bluff-body flow
➢ Regime I : 1.2 ≤ L/W < 3.0), alternating reattachment flow
➢ Regime II: 3.0 < L/W < 4.3), synchronous co-shedding flow
➢ Regime III: 4.3 < L/W < 7.3) and asynchronous Co-shedding flow
• Regime III A: 4.3 < L/W < 5.1) and dual St flow
• Regime III B: 5.1 < L/W < 7.3).
➢ Regime IV: 7.3 < L/W ≤10.0).
The dependence on L/W of fluctuating and time-mean fluid forces and St of the three prisms in each
regime is studied in detail and connected to the flow structures. A secondary vortex street following
the primary vortex street is observed for Regimes III B and IV. The detailed physics of the evolution
of the primary vortex street to the secondary is imparted. The inherent frequency associated with
the secondary vortex street is smaller than that with the primary. The evolution process of the
primary vortex street to the secondary leads to a tertiary frequency. Dynamic mode decomposition
analysis is proposed for the first time as a useful and quantitative tool to identify and quantify the
secondary vortex street and its onset position. (See Figure 8.5 and [Zheng & Mahbub Alam]254.

Regime I Regime II

Regime III B Regime IV

Figure 8.5 Contours of instantaneous vorticity field showing vortex shedding from the prisms

8.2.3 Concept of Dual Time Stepping

When solving an unsteady fluid flow, inadvertently, two time steps come into the picture. One a real
physical time integration (outer loop), the other is pseudo timing mainly for iterative schemes on
spatial coordinates (inner looping). The idea is to use an implicit scheme with a large stability region
and to solver the implicit equations at each time step, while by inner iterations using an accelerated

253 Qinmin Zheng and Md. Mahbub Alam, “Evolution of the wake of three inline square prisms”, Physical Review
Fluids 4, 104701 (2019).
254 Qinmin Zheng and Md. Mahbub Alam, “Evolution of the wake of three inline square prisms”, Physical Review

Fluids 4, 104701 (2019).

 211

evolution scheme in artificial time or pseudo-time [Jameson]255 as :

d d𝐖
∫ 𝐖dV + ∮ 𝐧𝐢 . 𝐟i . dS = 0 or V + 𝐑(𝐖)
⏟ =0
dt ⏟dt
Cell Cell Boundary 𝐏seudo Time
Real Time
Eq. 8.16
Where w now denotes the average value of the state in the cell. V is the cell volume, or in the two
dimensional case, the cell area. R(w) is the residual resulting from the space discretization.
Applications to external aerodynamic typically use grids in which the cell area or volume varies by
many orders of magnitude between the body and the far field, and this is a principal reason for using
implicit schemes for time accurate simulations.
8.2.3.1 Backward Differencing Formula (BDF)
Introducing superscripts n to denote the time level, the second order Backward Difference Formula
(BDF2) for time integration is

V 3 n+1 1
( 𝐖 − 2𝐖 n + 𝐖 n−1 ) + 𝐑(𝐖 n+1 ) = 0
dt ⏟2 2
𝐑∗ (𝐖)
Eq. 8.17
In the dual time stepping scheme this equation is solved by marching the equation:

d𝐖
+ 𝐑∗ (𝐖) = 0
dτ
Eq. 8.18
In solving Eq. 8.18 one is free to use every available acceleration technique for fast steady state
solutions without regard for time accuracy. It was shown that Multigrid techniques can be very
effective for this purpose. The dual time stepping approach has been quite widely adopted,
particularly in conjunction with the (BDF2) scheme. Recently there has been considerable interest in
whether Implicit Runge-Kutta schemes can achieve better accuracy for a given computational cost
than the Backwards Difference Formulas. Most of the studies to date have focused on Diagonal
Implicit Runge-Kutta (DIRK) schemes, sometimes called semi-implicit schemes, in which the stages
may be solved successively.

8.3 Essence of Local Preconditioning in CFD

It is essentially two types of preconditioning:
1. PDE level and its discretization ,
2. Linear system level (or purely mathematics).
Here we are mainly concern with type (1) which is a time derivative preconditioning to modify the
way the solution evolves in pseudo-time towards convergence (or at each time step, for unsteady).
Traditional time-marching algorithms that use the physical time derivatives in their formulation have
been very successful in the computation of transonic and supersonic flows as argued by [Colin,
Deniau and Boussuge]256. However, a major shortcoming in most compressible flow solvers is
their inability to solve efficiently problems of mixed flow involving very low Mach numbers in

Antony Jameson, “Time Integration Methods In Computational Aerodynamics”, 2003.

255

Yann Colin, Hugues Deniau and Jean-Fran¸cois Boussuge,”A Robust Low Speed Preconditioning Formulation:
256

Application to Air Intake Flows”, CERFACS, Toulouse, France.

 212

the flow. At low Mach numbers, most of these codes encounter degraded convergence speeds due to
the wide disparity that exists between the particle and acoustic wave speeds. Local preconditioning
techniques have been first introduced to enable the simulation of an incompressible flow thanks to a
compressible flow solver. They alter the time derivatives of the compressible Euler and Navier-
Stokes (NS) equations to control the propagation velocities of the various modes. The goal of
preconditioning methods is to reduce the disparity between the particle and acoustic wave speeds
so that good convergence properties may be obtained at all speeds. The idea is thus to modify the
time-marching behavior of the equations without altering the steady state solutions.
The use of preconditioning does not only reduce the stiffness of the system of equations; it also
improves accuracy at low speed. The modification of the fluxes required to take into account the new
characteristics of the preconditioned system results in a well-conditioned dissipation formulation
and ensures reliable accuracy. So far, preconditioning techniques have proven to be a powerful tool.
Unfortunately, the gains in convergence and accuracy are accompanied by reduced robustness, which
constitutes the major shortcoming of these methods for their use in an industrial context. The loss of
robustness can be explained for two different flow phenomena. First, for Euler computations, the
loss of robustness is due to stagnation point regions, which result in high local pressure disturbances.
Then, for turbulent computations, the presence of low Reynolds number regions may also lead to
instabilities. Indeed, the preconditioning techniques have been developed to eliminate analytic
stiffness arising from the propagative disparities, inherent to the hyperbolic system of Euler
equations. However, for viscosity dominated flows, the NS system becomes parabolic and the
boundary layer region is dominated by diffusion processes257. [Birken et al.]258 studied the multigrid
methods for finite volume discretization of the Reynolds Averaged Navier-Stokes (RANS) for both
steady and unsteady flows, using a preconditioned smoothers.

8.3.1 Pseudo (Artificial) Compressibility

The objective here is to propose a formulation of the preconditioning parameter to carry out robust
computations for both internal and external flows. Since we are only interested in steady solutions,
we will modify the time. Following the reasoning given by [Turkel]259, the simplest such modification
is the pseud-compressibility approach which adds a pressure time derivative to the continuity
equation. Then all the equations can be marched in time until a steady state is reached. Since there is
no decay mechanism except for boundaries we can accelerate to a steady state only by increasing the
allowable time step. By normalizing the fastest speed it is shown that we accelerate the convergence
when all the speeds are close together in absolute value. Conversely, the worst convergence occurs
when the speeds of the system differ by orders of magnitude. It is also shown in that in order to have
a well-posed problem that is compatible with the steady state, especially in terms of boundary
conditions, it is desirable to have a symmetric hyperbolic system. For a symmetric hyperbolic system,
when the preconditioning matrix is positive definite we are guaranteed that we have not changed the
appropriate number of boundary conditions and that we have not introduced any nonphysical time
reversals. The advantage of this method is its low computational cost and the reasonable
convergence speed. The efficiency of the method relies on the approximation of the inviscid and
viscous Jacobians by their spectral radii to eliminate large block matrix inversions. In this context,
the Block-Jacobi preconditioner may not be useful since its simplification within this framework
leads to classical local time stepping procedure.

257 See Previous.

258 Philipp Birken, Jonathan Bull, Antony Jameson, “Preconditioned smoothers for the full approximation scheme
for the RANS equations”, October 16, 2017.
259 Eli Turkel, “Preconditioned Methods For Solving The Incompressible and Low Speed Compressible Equations”,

Journal of Computational Physics, 1987.

 213

8.3.2 Case Study 1 – Preconditioned Methods for Solving 2D Incompressible Flow

Here, we consider the incompressible inviscid equations. Extensions to the viscous equations will
be considered later260. We will only consider time-independent solutions and follow closely the
development in261. Nevertheless, since we shall discuss time-marching algorithms we begin with the
time-dependent incompressible inviscid equations,

1 αu
pt + u x + v y = 0 , p + ut + uux + vu𝑦 + p𝑥 = 0
β2 β2 t
αv
p + ut + uvx + vv𝑦 + p𝑦 = 0
β2 t
Eq. 8.19
Here, α and β are functions to be determined. When, α = 0, we recover the standard pseudo-
compressibility method and we need only determine β. To form a conservation system we multiply
the first equation by u and also v and then add to the second and third equations, respectively. We
first rewrite Eq. 8.19 in vector form as

1
0 0
β2
αu p 0 1 0 p 0 0 1 p
1 0 (u) + [1 u 0] (u) + [0 v 0] ( u ) = 0
β2 ⏟ ⏟0 0 u ⏟
αv
v v ⏟0 0 v ⏟ v
0 1 𝐖t 𝐀𝟎 𝐖𝐱 𝐁𝟎 𝐖𝐲
[ β2
⏟ ]
𝐄 −1
Eq. 8.20
p 0 𝛽2 0 p 0 0 𝛽2 p
(u) + [1 (1 − α)u 0] (u) + [0 v −αu ] (u) = 0
⏟v ⏟0 −αv u ⏟ v ⏟1 0 (1 − α) ⏟
v
𝐖t 𝐀 𝐖𝐱 𝐖𝐲
𝐁
Eq. 8.21
In order to consider the wave speeds of Eq. 8.21, we Fourier transform the system [Turkel]262. We
note that the optimal β for α ≥ 1 is gotten by choosing an equality in Eq. 8.22 rather than an
inequality. Hence, if one wishes the system to be both close to optimal and symmetrize, should
choose β2 slightly larger than u2 + v2. Furthermore, for α ≤ 1, Eq. 8.22 implies automatically. For α
≤ 0 , Eq. 8.22 is always satisfied for all β.

β2 > α(u2 + v 2 )
Eq. 8.22
We can time integrate using an implicit scheme on Eq. 8.19 such that:

260 See Previous.

261 See Previous.
262 Eli Turkel, “Preconditioned Methods for Solving the Incompressible and Low Speed Compressible Equations”,

Journal of Computational Physics -1987.

 214

∂ n ∂
[𝐄 −1 + ∆t ( 𝐀0 + 𝐁0n )] ∆𝐖 = −∆t(𝐟xn + 𝐠 ny ) , ∆𝐖 = 𝐖 n=1 − 𝐖 n
∂x ∂y
𝐟 = [u , u2 + p , uv]T , 𝐠 = [u , uv , v 2 + p]T
∂ ∂
𝐄 −1 [𝐈 + ∆t 𝐄 𝐧 ( 𝐀n0 + 𝐁0n )] ∆𝐖 = −∆t(𝐟xn + 𝐠 ny )
∂x ∂y
Eq. 8.23
We now apply an approximate factorization and ignore errors in the conservation form of the left-
hand side to get

∂ ∂
𝐄−1 [𝐈 + ∆t ( 𝐀)] [𝐈 + ∆t ( 𝐁)] ∆𝐖 = −∆t(𝐟xn + 𝐠 ny )
∂x ∂y
Eq. 8.24
Since the matrices [A] = [E][A0] and [B] = [E][B0] are well conditioned, there is no way that the
splitting error can slow down the convergence compared with the standard AD1 splitting. Further
details and formulation for compressible flow are obtainable from [Turkel]263.

∂𝐖 ∂(𝐟 − 𝐟𝐯 ) ∂(𝐠 − 𝐠 v ) ∂(𝐡 − 𝐡v )

+ + + =0
∂t ∂x ∂y ∂z
ρ ρu 0
ρu ρu + p
2
τxx
𝐖 = ρv , 𝐟= ρuv , 𝐟v = τxy
ρw ρuw + p τxz
[ Et ] [(Et + p)u] [uτxx + vτxy + wτxz − qx ]
ρv 0
ρuv τxy
𝐠 = ρv + p
2
, 𝐠v = τyy
ρvw τyz
[(Et + p)v] [uτxy + vτyy + wτyz − qy ]
ρw 0
ρuw τxz
𝐡= ρvw , 𝐡v = τyz
ρw + p
2 τzz
[(Et + p)w] [uτxz + vτyz + wτzz − qz ]
Eq. 8.25
Additional information can also be obtained from studies by [Turkel and Vatsa]264 and [Turkel &
Vatsa]265

263 Eli Turkel, “Preconditioned Methods for Solving the Incompressible and Low Speed Compressible Equations”,
Journal of Computational Physics -1987.
264 Eli Turkel and Veer N. Vatsa, “Choice of Variables and Preconditioning for Time Dependent problems”, NASA

Langley Research Center, Hampton, VA, AIAA Paper.

265 Eli Turkel and Veer N. Vatsa, “Local Preconditioners For Steady And Unsteady Flow Applications”,

Mathematical Modelling and Numerical Analysis.

 215

8.3.3 Case Study 2 – Preconditioned Methods for 3D Unsteady Low Speed Compressible N-S
The governing equations are the unsteady compressible Navier-Stokes equations which describe the
conservation of mass, momentum and energy of the flow field. Following closely the development in
[Colin et al.]266, the conservative form can be expressed in 3D Cartesian coordinates (x, y, z) as Error! R
eference source not found..
8.3.3.1 Low Speed Preconditioning Formulation
Preconditioning techniques involve the alteration of the time-derivatives used in time-marching CFD
methods with the primary objective of enhancing their convergence. The alteration of the
propagation velocities is done by multiplying the time-derivative by a preconditioning matrix ГW as

∂𝐖 ∂𝐖 ∂𝐖 ∂𝐖
𝚪𝐖 + 𝐀𝐖 + 𝐁𝐖 + 𝐂𝐖 = 𝐑𝐖
𝐕 (𝐖)
∂t ∂x ∂y ∂z
∂𝐟 ∂𝐠 ∂𝐡
where 𝐀𝐖 = , 𝐁𝐖 = , 𝐁𝐖 =
∂𝐖 ∂𝐖 ∂𝐖
Eq. 8.26
are the inviscid Jacobians expressed in conservative variables and RWv (W) denotes the differential
operator for the viscous term. An issue related to the study or comparison of preconditioning
matrices relies on the choice of the system variables. Merkle, Turkel or Van Leer thus make different
variable choices. Besides the choice of the conservative variables set W is not well suited for
preconditioning analysis due to the complex formulation. The analysis is generally simplified by
considering the Euler conservative variables and its differential form of the entropy variables as:

T
T
dp
e 2
𝐖 = [ρ, ρu, ρv, ρw, E] , d𝐖 = [ , du, dv, dw, dp − c dρ]
ρc
Eq. 8.27
where c is speed on sound and superscript e assigns entropy267. In this formulation, the system of
preconditioned Navier-Stokes equations becomes:

∂𝐐 ∂𝐐 ∂𝐐 ∂𝐐
+ 𝚪𝐐−𝟏 [𝐀𝐐 + 𝐁𝐐 + 𝐂𝐐 ] = 𝚪𝐐−𝟏 𝐑 𝐕 (𝐐)
∂t ∂x ∂y ∂z
ε
ε 0 0 0 −δ
ρc
where 𝚪𝐐−𝟏 = 0 1 0 0 0
0 0 1 0 0
0 0 0 1 0
[0 0 0 0 1 ]
Eq. 8.28
where AQ, BQ, and CQ are symmetric matrixes. The Г-1Q is the preconditioning matrix where a
generic Weiss-Smith/Choi-Merkle preconditioner (WSCM) was implemented. In this formulation, ԑ
~ M2 is the preconditioning parameter, to be defined, and δ is a free parameter varying from 0 to 1.

266 Yann Colin, Hugues Deniau and Jean-Fran¸cois Boussuge,”A Robust Low Speed Preconditioning Formulation:
Application to Air Intake Flows”, CERFACS, Toulouse, France.
267 Kaveh Hosseini and Juan J. Alonso, “Practical Implementation and Improvement of Preconditioning Methods

for Explicit Multistage Flow Solvers”, AIAA 2004{0763.

 216

For δ = 0, the preconditioner is the Weiss-Smith preconditioner, whereas for δ = 1 it reduces to the
Choi-Merkle one. At this point, we recall that there are two other types of preconditioners which try
to modify the inviscid flux Jacobian eigenvalues. The first one is the preconditioner by Turkel which
introduces a new parameter to improve the condition number to unity at the low Mach number limit.
The second one is the Van Leer-Lee-Roe preconditioner (VLR) which is designed to optimize wave
speeds across all Mach number ranges. Concerning convergence issues, the natural candidate turns
to be the VLR preconditioner which yields the most optimal condition number obtainable. The WSCM
does not achieve the optimal conditioning at low Mach numbers and possesses a condition number
of 2.61 as the Mach number tends to zero.
8.3.3.2 Numerical Integration
The numerical code solves the compressible Navier-Stokes equations on structured meshes using a
finite volume method where the preconditioning is used as:

W
V t dV + Γ W S Μ.ndS = 0 where M = (f - f v , g - g v , h - h v )
-1
Eq. 8.29

and n is the outward unit normal. These equations are discretized in space using a cell centered finite
volume method. The discretized equations reduce to the following system applied to one cell:

W
V + Γ -1
WR = 0 Eq. 8.30
t
where R is the residual of the discretized convective and viscous fluxes. On structured meshes, a large
number of schemes have been developed in order to get an approximation of the convective fluxes.
Most of them may be written using a central discretization in conjunction with an artificial dissipation
model. Therefore, the inviscid part of the numerical flux Fi+1/2 at the cell face i + 1/2 is written as:

Fi +1/2 = Mi +1/2.si+1/2 - Γ W,i +1/2di+1/2 Eq. 8.31

where S = (SX, SY, SZ) is the corresponding surface vector and d is the dissipative flux. The use of
preconditioning does not only reduce the stiffness of the system of equations, it also improves
accuracy at low speeds. It turns out that the artificial dissipation fluxes become extremely large for
very low velocities and are responsible for the loss of accuracy of the original convective schemes.
The modification of the fluxes required to take into account the new characteristics of the
preconditioned system results in a well-conditioned dissipation formulation and ensures reliable
accuracy. The artificial dissipation flux d of [Jameson et al.] consists of a blend of second- and fourth-
order differences. Other relative details can be found in [Colin, et al. ]268.
8.3.4 Case Study 3 - Numerical Analysis of a Dual-Time Pseudo-Compressibility Method

Citation : Van-Tu Nguyen, Duc-Thanh Vu, Warn-Gyu Park, Young-Rae Jung, Numerical analysis of water
impact forces using a dual-time pseudo-compressibility method and volume-of-fluid interface tracking
algorithm, Computers & Fluids, Volume 103, 2014, Pages 18-33, ISSN 0045-7930,
https://doi.org/10.1016/j.compfluid.2014.07.007.

Yann Colin, Hugues Deniau and Jean-Francois Boussuge, “A Robust Low Speed Preconditioning Formulation:
268

Application to Air Intake Flows”, CERFACS, Toulouse, France.

 217

An implicit algorithm based on a dual-time pseudo-compressibility method is developed to compute

water impact forces on bodies, by [Nguyen et al.]269 and [Nguyen & Park]270. Flow fields of
incompressible viscous fluids are solved using unsteady Reynolds-averaged N–S equations. Pseudo-
time derivatives are introduced into the equations to improve computational efficiency. A second-
order volume-of-fluid interface tracking algorithm is developed in a generalized curvilinear
coordinate system to track the interface between the two phases in the computational domain. A grid
refinement study of the dam-break flow is performed as a validation, and the obtained solutions
agreed well with the experimental data and with the results of other numerical simulations.
Numerical analysis of water impact forces on a hemisphere, two cones, and a wedge through free
falling in one degree of freedom is then performed. Free surface deformation, pressure coefficients,
impact velocities, and vertical accelerations during impact are compared with available experimental
data and theoretical results. Good agreement with these results is obtained.
8.3.4.1 Introduction and Background
The computational fluid dynamic simulation of free surface flow for water impact problems is a vast
topic that is still receiving increasing attention, particularly for applications in many hydraulic and
hydrodynamic problems. The important examples for hydraulic problems in civil engineering are the
potential risks of failures of levees, dams, reservoirs, and flood management. The main computational
challenges when solving such highly nonlinear, complex problems, include the determination of the
dynamic pressure loads by water impact and the very violent motions of the interface, including
turbulence, discontinuities, splashing, wave breaking, and mixing or entrapment of one fluid within
another. The challenges have motivated researchers to develop novel, efficient, and accurate
numerical models for high resolution, 2D, and 3D flow simulations over the years. A class of particle
methods adopts a Lagrangian framework for the description of the fluid, such as the smoothed
particle hydrodynamics (SPH), the meshless finite element method, or the particle finite element
method. These methods have been developed by many authors for the simulation of an extensive
class of problems involving complex interaction between fluids and solids. Among the Lagrangian
approaches, the earliest method is the SPH technique, originally proposed by pioneers [Gingold and
Monaghan]271 and [Lucy]272.
The technique subsequently received widespread interest and success within the scientific
community, especially for its applications in both hydraulic and hydrodynamic problems because of
its simplicity and high computational efficiency273. However, this method faces some drawbacks, such
as a lack of consistency, and a tensile instability in distortion of the material domain that can lead to
a low accuracy and to a prominent problem in proving the convergence. Based on the best features
of the particle method, together with the finite element method, both the meshless finite element and
the particle finite element methods, in which meshless and mesh-based shape functions are
respectively used, were later introduced. The methods are easy to use and introduce the boundary
condition and have proven their capability for simulating a broad class of problems. Moreover, the
tensile instability in the distortion of the material domain still occurs and leads to a poor accuracy in
simulations using the methods.

269 Van-Tu Nguyen, Duc-Thanh Vu, Warn-Gyu Park, Young-Rae Jung, “Numerical analysis of water impact forces
using a dual-time pseudo-compressibility method and volume-of-fluid interface tracking algorithm”, Computers
& Fluids 103 (2014) 18–33.
270 Van-Tu Nguyen and Warn-Gyu Park, “A free surface flow solver for complex three-dimensional water impact

problems based on the VOF method”, Int. J. Numerical Meth. Fluids (2015).
271 Gingold RA, Monaghan JJ. “Smoothed particle hydrodynamics: theory and application to non-spherical stars”.

Monthly Notices of the Royal Astronomical Society 1977; 181:375–389.

272 Lucy LB. “A numerical approach to the testing of the fission hypothesis”. Astronomical Journal 1977.
273 Prakash M, Rothauge K, Cleary PW. “Modelling the impact of dam failure scenarios on flood inundation using

SPH”. Applied Mathematical Modelling 2014; 38:5515–5534.

 218

For the use of Eulerian approaches, the Level-Set Method (LSM) and the Volume-Of-Fluid (VOF)
method are the most popular and are frequently used by researchers for the prediction of the
problems. The LSM method was originally proposed by [Osher and Sethian]274. The interface is
represented as a smooth function of the distance from the interface, and highly deformed interfaces
can be treated. Because of the simplicity and ability to capture the interface, the method has been
developed and applied to a wide range of problems, such as the Rayleigh Taylor instability, vortex
motion, bubbles and drops, free surface motion, and the interaction between free surface flow and
rigid bodies. However, the drawback with mass conservation of the method is an unphysical
loss/gain of the fluid in simulations. As an alternative to the LSM in capturing the interface, the VOF
technique is a good tool for numerical simulations of free surface flows275. As a result of the high
accuracy of mass conservation, many different novel, accurate, VOF interface tracking algorithms
have been developed and have been used extensively to compute bubbles, droplets, and the Rayleigh
Taylor instability, free surface flows, water impact problems, and multiphase flows of incompressible
fluids. An overview of the methods available can be found in our previously published work276.
The aim here is to develop a robust numerical method that allows the predictive 3D simulation of
free surface flows and dynamic pressure loads by the water impact on complex domains. A dual-time
pseudo-compressibility (NS) model, and a VOF interface tracking algorithm that was implemented in
a generalized curvilinear coordinate system. The NS equations are expressed in the vector form
system, which is discretized on a general curvilinear grid using a class of lower-upper symmetric
Gauss–Seidel method. The Monotonic Upstream-Centered Scheme for Conservation Laws (MUSCL)-
Roe procedure is employed in the discretization of the convective terms to achieve a high-order
approximation. To evaluate the accuracy of the VOF method on curvilinear grids, the multiple
numerical benchmark tests, including 2D and 3D deformation cases, are first performed. The
numerical procedure is then validated by computing water entries of the free-falling hemisphere and
cone, the dam-break impact flows, and the impact of the wave on a container and on a tall structure
where dam failure is assumed to occur further to the collapse of a water column and a subsequent
impact with the obstacles. The obtained results are compared with the experimental data as well as
with the results of other numerical simulations in the literature.
8.3.4.2 Governing Equations
The dual-time, pseudo-compressibility, homogeneous flow model is developed on the basis of the
unsteady incompressible Reynolds-averaged N–S equations. All dependent variables are non-
dimensionalized using the free-stream conditions, such as free-stream velocity u1, free-stream
density q1, freestream viscosity μ1, and characteristic length of the body D. The governing equations
can be expressed in generalized curvilinear coordinates as follows277-278:

̂ ∂Q
∂Q ̂ ∂Êj ∂Êjv
Γe + + + = Ŝ
∂t ∂τ ∂ξj ∂ξj
Eq. 8.32

274 Osher S, Sethian JA. “Fronts propagating with curvature-dependent speed: algorithms based on Hamilton–
Jacobi formulations”, Journal of Computational Physics 1988; 79:12–49.
275 C. W. Hirt and B. D. Nichols, “Volume of Fluid (VOF) Method for the Dynamics of Free Boundaries”, Journal of

Computational Physics 39, 201-225 (1981).

276 Nguyen V-T, Vu D-T, Park W-G, Jung Y-R. “Numerical analysis of water impact forces using a dual-time pseudo-

compressibility method and volume-of-fluid interface tracking algorithm”. Computers and Fluids 2014.
277 Nguyen V-T, Vu D-T, Park W-G. “3D numerical simulation of water entry of free falling objects using Navier–

Stokes computations and moving chimera grid scheme”. 10th Asian CFD Conference, Korea. 2014.
278 De Jouëtte C, Laget O, Le Gouez JM, Viviand H. “A dual time stepping method for fluid–structure interaction

problems”, Computers and Fluids 2002; 31:509–537.

 219

where the primitive solution variable, convective, viscous flux, and source vectors are written as
follows:

̂ = Q = 1 [ρ̅, ̅u,
Q ρ ρ̅w]T
ρ ̅v,
J J
1
̂j = [ρm Uj , ρm uUj + ξj,x p , ρm vUj + ξj,y p , ξj,x p , ρm wUj + ξj,z p]T
E
J
1
̂jv = [0 , ξj,k τx,k , ξj,k τy,k , ξj,k τz,k ]T
E
J
1 T
Ŝ = [0 , ρ𝑚 g 𝑥 , ρ𝑚 g 𝑦 , ρ𝑚 g 𝑧 ]
J
Eq. 8.33
and t is the physical time, s is the pseudo-time, and ĝx, ĝy and ĝz are the accelerations due to gravity
and hydrodynamic forces in the x, y and z directions, respectively. The accelerations can be
determined using the dynamic equation of the body. The pseudo-time derivatives involve a pseudo-
density ρ∼; and the pressure field is calculated on the basis of an additional pseudo-state equation:

ρ̃
p = ρm U02 ln ( ) + p∞
ρ∞
Eq. 8.34
where the parameters are set in accordance with U0 = U∞ or U0 = √u2 + v2 + w2, in which u, v, and w
are the local values of the respective velocities obtained at a previous iteration step in pseudo-time.
The mixture density and mixture viscosity are defined as follows:

ρm = α1 ρl + (1 − α1 )ρg , μm = α1 μl + (1 − α1 )μg
Eq. 8.35
where α is the volume fraction, and the subscripts l and g denote liquid and gas phases, respectively.
The contravariant velocities are given as follows:

Uj = ξj,t + ξj,x u + ξj,y v + ξj,z w

Eq. 8.36
And the matrix Гe is given as
0 0 0 0
ρm −u 1 0 0
𝚪e = [ ]
ρ̃ −v 0 1 0
−w 0 0 1
Eq. 8.37
8.3.4.3 Time Integration
The governing equations expressed by Eq. 8.32 are discretized on a general, structured grid, using
the lower-upper symmetric Gauss–Seidel method. An upwind difference scheme, based on the
characteristic information of the governing equations, is used to compute the convective flux
derivatives. For this purpose, the flux Jacobian matrix is split into two sub-vectors associated with
the nonnegative and non-positive eigenvalues. The convective flux in Eq. 8.32 can be linearized as
 220

follows279:

̂ ∂𝐐
∂𝐐 ̂ ∂𝐐 ̂
Γe + + 𝐀𝒋 = 𝐒̅
∂t ∂τ ∂ξ𝑗
−Uj ξj,x ξj,y ξj,z
∂𝐄̂𝐣 ρ −2uUj + U02 ξj,x uξj,x + Uj uξj,y uξj,z
̂𝐣 =
𝐀 =
∂Q̂ ρ̃ −2vUj + U02 ξj,y vξj,x vξj,y + Uj vξj,z
2
[ −2vw + U0 ξj,z wξj,x wξj,y wξj,z + Uj ]
and flow velocity = U0 = √u2 + v 2 + w 2
Eq. 8.38
The physical time derivative is approximated by a second-order backward difference, and the
implicit Euler’s finite difference formula is used for the pseudo-time derivative. The governing
equation, expressed by Eq. 8.32, can be written in a difference form as follows:

∂Âj 𝚪𝐞 ̂ n+1,k − 4𝐐
3𝐐 ̂n + 𝐐 ̂ n−1
[𝐈 + ∆τ ( + 1.5 − 𝐒)] ∆𝐐 ̂ n+1,k
− ∆τ𝚪e
∂ξj ∆t 2∆t
∂𝐄̂j − ∂𝐄̂j
v
= ∆τ ( − 𝐒̂) where ∆𝐐 ̂ n+1,k = 𝐐̂ n+1,k+1 − 𝐐̂ n+1,k
∂ξj
Eq. 8.39
Here, the superscript n represents the physical time index, and the superscript k represents the
pseudo-time index. The pseudo-time step is determined in accordance with the local pseudo-time
step, which is defined by the largest system eigenvalue. At each physical time step, a pseudo-time
iterative procedure is applied, such that ΔQn+1,k /Δτ →0 at convergence.
8.3.4.4 Six-Degree-of-Freedom Rigid Body Motion
A 6 DOF model is integrated into the numerical solver for considering the arbitrary motions of rigid
bodies. The flowchart of the strong combination of the NS/VOF flow solver and 6DOF model is shown
in Figure 8.6. Hydro-forces and moments are computed after each pseudo-time step (inner loop) of
the NS/VOF solver, and then, the 6DOF motions of a rigid body are solved. Next, the Chimera grid
system is redefined (moved) based on the motions of the rigid body. At this step, the constraints
based on the difference between two sequential iterations of calculated loads are checked. In addition
to the computational Chimera grid, the fulfillment of such conditions allows for the continuation of
the procedure and the movement of the body. This inner loop provides a strongly coupled solution
in the entire computational domain between the rigid body motions and flow field. The 6DOF rigid-
body motion equations are required to describe the position and orientation of a rigid body during
simulations. Among these, three DOFs provide the location of a point fixed on the body. The other
three DOFs provide the orientation of the body in a fixed reference frame. The position of the body

279 NguyenV-T, Vu D-T, Park W-G, Jung Y-R. “Numerical analysis of water impact forces using a dual-time pseudo
compressibility method and volume-of-fluid interface tracking algorithm”. Computers and Fluids 2014.
 221

in a reference frame can be obtained by

integrating the velocity in that
frame280,281.
8.3.4.5 3D Flow Evolution of a Dam
Break
3D simulations of interfacial structures
under wave-front propagation in a dam-
break problem are accessible by
[Nguyen & Park]282. The corresponding
experiments were conducted by
[Fraccarollo and Toro]283. In the
experiment, still water with a total
depth of H = 0.6 m was initially placed in
a plane-bottom tank with a width of 2 m
and a length of 1 m. The tank had a 0.4 m
wide gate and was connected to a
downstream area with dimensions of 2
m × 2 m, which had open boundaries on
three sides. Figure 8.7 shows an image
of the experimental setup and the snap
shot of the 3D dam-break flow
simulation that was performed using a
grid comprising 1.5 M points. The water Figure 8.6 Flowchart of the Strong coupling of NS/VOF
front flows through the gate and moves Flow Solver and 6DOF Model (Courtesy of Nguyen & Park)

Figure 8.7 Snapshots of 3D dam-break flow: (a) experimental image and (b) simulation result at 0.5 s
(Courtesy of Nguyen & Park)

280 P.M. Carrica, R.V. Wilson, R.W. Noack, F. Stern, Ship motions using single-phase level set with dynamic overset
grids, Computational Fluids 36 (2007)1415–1433.
281 N.C. Prewitt, D.M. Belk, W. Shyy, Parallel computing of overset grids for aerodynamic problems with moving

objects, Prog. Aerospace Sci. 2000.

282 Van-Tu Nguyen, and Warn-Gyu Park, “Enhancement of Navier–Stokes solver based on an improved volume-

of-fluid method for complex interfacial-flow simulations”, Applied Ocean Research 72 (2018) 92–109.
283 L. Fraccarollo, E.F. Toro, Experimental and numerical assessment of the shallow water model for two-

dimensional dam-break type problems, J. Hydraulic Res. 1995.

 222

downstream to half the length of the rectangular tank in a relatively short time of t = 0.5 s.
As the water flows past the gate and moves down-stream, the water height in the reservoir reduces,
which can be clearly observed from the change in the color of the interface with respect to time. The

Figure 8.8 Image sequences of the 3D evolution of a dam break in the case of closed downstream walls,
where the color of the free surface denotes the water level (red, z = 0.6 m; green, z = 0) (Courtesy of
Nguyen & Park)
 223

free surface becomes smoother after the wave front reaches the bounds. The water level and
velocities of the flow at various positions inside and outside the tank were measured in the
experiment. A comparisons between the predicted and experimental data of the water levels as a
function of time is been made (see [Nguyen & Park]284). The obtained numerical results are similar to
the experimental measurements in terms of the oscillation frequency and amplitude of the velocity
signals. The two aforementioned quantitative comparisons show that the proposed method is
capable of accurately simulating complex interface structure flows. A simulation of the same problem
is conducted to observe other scenarios of dam-break failure. However, in this simulation, closed
boundaries are applied to the downstream area.
Figure 8.7 shows the snapshots of the dam-break problem during the simulation at time set = 0.25,
0.5, 1.0, 1.5, 2.0, 3.0, 4.0, and 12.15 s. At t = 0.25 s and 0.5 s, longitudinal crests appear, and the
structure of the interface formed is similar to that in the case of a downstream area with open
boundaries. The interfacial structure remains similar until the wave front reaches the bounds. The
water progressively propagates and hits the lateral and downstream walls, thus resulting in complex
waves reflecting from the walls. The back waves return and interact with the primary waves, forming
a complex interfacial structure. Finally, the water returns to an approximately hydrostatic
distribution, as shown at time t = 12.15 s.
8.3.4.6 Free-Surface Wave Around Surface-Piercing NACA Foils
To further validate the numerical method for unsteady free-surface wave flows, the experiments
conducted by [Metcalf and Stern]285 for a surface-piercing NACA 0024 foil are used for comparison
with the numerical results. The hydrofoil has a chord length of 1.2 m and a beam length of 0.29 m.
Figure 8.9 shows the visual comparisons of the wave profiles around the surface-piercing NACA
0024 foil for Fr = 0.37. A quantitative wave pro-file is calculated close to the wall. data for Fr = 0.37
and 0.55. It can be observed that the numerical simulations predict the location of the toe (x < 0.4 in
the case of Fr = 0.37 and x < 0.5 in the case of Fr = 0.55)accurately, but fail to predict the subsequent
regions with sharp rise (0.4 < x < 0.6 in the case of Fr = 0.37 and x > 0.5 in the case of Fr = 0.55). The

Figure 8.9 Visual Comparison of Simulated and Experimental Free-Surface Wave Profiles
(Courtesy of Nguyen & Park)

284 Van-Tu Nguyen, and Warn-Gyu Park, “Enhancement of Navier–Stokes solver based on an improved volume-
of-fluid method for complex interfacial-flow simulations”, Applied Ocean Research 72 (2018) 92–109.
285 B. Metcalf, J. Longo, S. Ghosh, F. Stern, Unsteady free-surface wave-induced boundary-layer separation for a

surface-piercing NACA 0024 foil: towing tank experiments, J. Fluids Structure. 22 (2006) 77–98.
 224

present results
and the solutions
of [Kandasamy et
al.]286 agree with
the data. Studies
on grid
refinement and
numerical
convergence were
performed using
this large-scale
problem. The
wave profiles
around the
surface-piercing Figure 8.10 Wave profile Around a Surface-Piercing NACA 0024 Hydrofoil at Fr
NACA 0024 foil = 0.37obtained from Grid Refinement
for Fr = 0.37 were
simulated using grids with five different grid sizes of 180,000 (grid I), 400,000(grid II), 710,000 (grid
III), 1,050,000 (grid IV), and 1,820,000 (grid V)points. The wave profiles for three different grid sizes
do not agree with the experimental data. The discrepancies between the predicted results for
medium and fine grids are small, and the present results agree well with the experimental data as
shown in Figure 8.10.
8.3.4.7 Computational Model and on a NACA 66 Hydrofoil
The proposed model is used to simulate a wave breaking around a surface-piercing NACA 66
hydrofoil moving at a speed of 5.33 m/s. The chord and span of the hydro-foil are 0.15 m and 0.45
m, respectively. The hydrofoil is submerged in water at a depth of 0.15 m, and 0.30 m is out of the
water. Figure 8.11 shows the profile of the wave breaking due to the hydrofoil and corresponding
computational grid. The water starts splashing up from the original water surface at the leading part
and extends down-stream, forming a pair of thin liquid sheets. The sheets stretch and subsequently

(a) Grid (b) Waves

Figure 8.11 Grid and Wave Breaking by a NACA 66 Hydrofoil (Courtesy of Nguyen & Park)

286 M. Kandasamy, T. Xing, F. Stern, Unsteady free surface wave-induced separation: vortical structures and
instabilities, J. Fluid Struct. 25 (2009)343–363.
 225

pinch-off from the liquid sheet with a long wake. The water is separated on two sides of the hydrofoil.
In addition to the two liquid sheets in the wake region, a third liquid sheet, which is at a level lower
than the water level, is formed in the middle region between the two liquid sheets because of gravity.
At the end of the wake, the liquid sheets fall and tend to return to the initial waterline.

8.3.4.8 Water Entry of Free-Falling Rigid Objects

8.3.4.8.1 Hemisphere
Before applying the model for the simulation of a free-falling sphere and considering the complicated
behavior of a cavity, a model of the 6DOF motion coupled with the numerical solver was used to
simulate a transient free-falling hemisphere entering the free surface of water. This was done to
check and validate the computational model using a simpler problem (presented in287). The
experimental results of [De Backer et al.]288 were compared with simulation results. The diameter
and mass of the hemisphere were 0.3 m and 11.5 kg, respectively. It was freely released into the
water, and it reached a velocity of 4.0 m/s immediately prior to touching the free surface. Figure
8.12 shows the temporal pressure-distribution coefficients defined as Cp = p/(0.5ρU2entry). The case
is simulated using three different grid sizes, and the obtained results are compared with the
experimental data, the numerical simulation results of [Wang and Soares]289, and the asymptotic
results obtained on assuming a constant water impact velocity as in. The numerical results of [Wang

Figure 8.12 Impact Pressure Distributions over the Surface of a Water Entry Hemisphere over time
(Courtesy of V.T. Nguyen, W.-G. Park)

287 V.-T. Nguyen, W.-G. Park, A free surface flow solver for complex three-dimensional water impact problems
based on the VOF method, Int. J.Numer. Methods Fluids 82 (2015) 3–34.
288 G. De Backer, M. Vantorre, C. Beels, J. De Pré, S. Victor, J. De Rouck, et al., Experimental investigation of water

impact on axisymmetric bodies, Appl. Ocean Res. 31 (2009) 143–156.

289 S. Wang, C. Guedes Soares, Numerical study on the water impact of 3D bodies by an explicit finite element

method, Ocean Eng. 78 (2014) 73–88.

 226

and Soares]290 contain numerous spurious oscillations and the asymptotic solutions exhibit
remarkable overestimation, while the present results agree well with the experimental data. It was
also shown the experimental and theoretical values of the penetration depth of the free-falling
hemisphere (see [Nguyen & Park]291). The comparisons show that the numerical model is successful
in determining the impact loads due to the interaction of the free surface and structure and the
trajectory of the water-entry body.
8.3.4.8.2 Sphere
Next, the hydrodynamics of a steel sphere freely falling into water are simulated. [Arist off et al.]292
studied this problem experimentally. The diameter of the sphere is 0.0254 m, and the ratio of the
densities of the sphere and water is 7.86. The sphere enters the water at a velocity of 2.17m/s. For
computational convenience, Chimera grids comprising a body-fitted sphere grid with a size of 130 ×
30 × 50 and a curvilinear background grid with a size of 400×110×40 are used, as shown in Figure
8.13. Simulations are performed for the falling of the sphere at three different entry velocities.
depths increase linearly with entry velocity. The comparison of the penetration depths of the sphere
at various entry velocities is shown in Figure 8.14. The curves in the figure show that when entry
velocity is higher, the sphere penetrates the water faster and cavity pinch-off occurs at a higher depth.
For Further and complete discussion, please consult the [Nguyen & Park]293.

Figure 8.13 Chimera Grids for Water Entry of a Figure 8.14 Center-depth variation for a free-
Sphere: (a) Curvilinear Background Grid (b) Body- falling sphere with time at various entry
Fitted Grid (Courtesy of Nguyen & Park) velocities (Courtesy of Nguyen & Park)

290 S. Wang, C. Guedes Soares, Numerical study on the water impact of 3D bodies by an explicit finite element
method, Ocean Eng. 78 (2014) 73–88.
291 Van-Tu Nguyen, and Warn-Gyu Park, “Enhancement of Navier–Stokes solver based on an improved volume-

of-fluid method for complex interfacial-flow simulations”, Applied Ocean Research 72 (2018) 92–109.
292 J.M. Aristoff, T.T. Truscott, A.H. Techet, J.W.M. Bush, The water entry of decelerating spheres, Phys. Fluids 2

(2010) 032102.
293 Van-Tu Nguyen, and Warn-Gyu Park, “Enhancement of Navier–Stokes solver based on an improved volume-

of-fluid method for complex interfacial-flow simulations”, Applied Ocean Research 72 (2018) 92–109.
227
 228

9 Panel Methods
9.1 Preliminaries
The panel method is an analysis method that can be used to arrive at an approximate solution for the
forces acting on an object in a flow294. The method is based on inviscid flow analysis, so it is limited
to the resultant pressure forces over the surface. The panel method is basically a numerical
approximation that relies on using discrete elements on the surface of an object and then prescribing
a flow element (such as a vortex or doublet or source or sink) on each element that will satisfy certain
boundary conditions (like no flow crosses the surface of the object). The interaction of the elements
are accounted for and must also satisfy the condition that far from the object the flow should be equal
to the free stream velocity approaching the object. There are a number of books and papers written
that describe the method in very general terms and even the inclusion of viscous forces to some
degree. But here we are just introducing the method to get a feel for its usefulness in external flows,
so we will use a simply geometry with a simply distribution of flow elements. More complicated
models exist but they all
are based on the
simplified form
presented here.
We will assume that we
have potential flow such
that the governing
equation for the flow
field is the Laplace of the
velocity potential, ∇2ϕ =
0. The boundary
condition at an
impermeable surface,
where the velocity
normal to the surface is
zero, is ∇ϕ⋅n = 0. Also,
we can put our frame of
reference on the object Figure 9.1 Physical domain for Laplace's equation
so fluid flow approaches
the object. Keep in mind that since it is inviscid there may be a nonzero velocity component tangent
to the surface. Also, the goal is to determine the velocity on the surface, and once this is found the
Bernoulli equation can be used to find the local pressure distribution. The pressure can then be
integrated over the surface to find the force by the fluid flow. Without deriving this it can be shown
that the following defines the velocity potential at any point P in the flow field (using Green’s
Identify):
1 ∇ϕ 1
ϕ(P) = ∫( − ϕ∇ ) 𝐧 ∙ dS
4π r r
Eq. 9.1
where the integral is over the surface area of the flow (assuming 2D flow), S, and

r = √(x − ξ)2 + (y − η)2

294Intermediate Fluid Mechanics by James Liburdy is licensed under a Creative Commons Attribution-Non-
Commercial-ShareAlike 4.0 International License.
 229

Eq. 9.2
This equation indicates that to solve for the velocity potential we must evaluate the integral on the
flow boundaries (both the solid surface and infinitely far away). Please also refer to [Tannehill, Dale
A. Anderson, Richard H. Pletcher, Computational Fluid Mechanics and Heat Transfer, 2nd edition, page
431].

9.2 Gridding (Paneling)

The general approach is to select a “grid”
which is a series of “panels” that form the
surface. Here we take the panels as
straight flat surfaces arranged over the
real surface. In the limit of very small
panels the constructed surface will
simulate the actual surface. On each panel
we place a distribution of flow elements
(like sources, sinks, vortices, etc.) that
when combined together will result in a
flow field that will satisfy the surface
boundary condition. There are lots of ways
to identify which elements to use and how
they may be distributed on the panels.
Here we will use vortex elements, with one Figure 9.2 Illustration of a panel geometry; any three
placed on each panel. The net flow is the dimensional shape can be constructed; shown here is a
result of superposition of the flow set up surface of what could be a three dimensional object such
by each vortex on each panel element. So as an entire airplane
at each point in the field we add together
the flow caused by all of the panel elements using the superposition rule. The panels that are far away
from a given point will have less and less influence on the flow because the strength of the flow caused
by a flow element decreases with distance from the element origin. For instance for a “source” the
velocity decreases with increasing radial position because the flow is spreading out away from the
source. But the influence never really goes to zero. Figure 9.2 illustration of a panel geometry; any
three dimensional shape can be constructed; shown here is a surface of what could be a three
dimensional object such as an entire airplane.
In placing a series of panels over the surface we first need to specify the size of each panel. We place
a vortex of some strength somewhere on the panel (whose location is shown below) and we must
identify points on the surface where we want to make sure that the velocity is zero across the surface.
To be clear, individual points on each panel are used to evaluate the element (vortex) flow field –it
needs its own origin, or coordinate system, to write an equation for the flow generated by this
element. We also only pick a point on the panel to check to make sure that the net sum of
contributions from all elements results in zero flow across the surface. The fact that we only satisfy
the condition at one point on each panel will be satisfactory if the panels are made to be reasonably
small. These points are called “collocation points” on each panel. In the end each panel will have
coordinates that define its location on the surface, coordinates for the element location on the surface
and coordinates for the collocation point.

9.3 Method of Solving the Potential Equation

The full potential equation could be either conservative or non-conservative form. Here we deal with
non-conservative form which obtained using the continuity equation and some manipulation via
momentum equation. The 3D velocity potential equation in non-conservative for
 230

  
u= , v= , w=
x y z
  x2    y2   2  2 x y 2  2 y z
Continuity : 1 - 2   xx + 1 − 2  yy + 1 − z2  zz −  xy − x2 z  xz − 2  yz = 0
 a  2
 a     a  a a a
u 2 + v 2 + w2 dp
Momentum & Energy : + = constant →
2 d
  x2 +  y2 +  z2  ρ   x +  y +  z 
2 2 2

dp = −ρd   =− 2d
 2  a  2 
  
γ −1 2
whe re a = a 0 − V and V = u 2 + v 2 + w 2 , a 0 = speed of sound
2
9.3
Note that for an incompressible flow, a → ∞, the velocity potential reduces to linear Laplace’s
equation and could be solve with relative ease. It represents a combination of continuity, momentum
and energy equations and could be solved for velocities (i.e., Mach number) and then temperature,
pressure and density could be obtained using the isentropic relations, previously defined for local
values. It should be noted that the total quantities are known and obtained from free-stream
conditions. For a 2D flow, after substituting for u, v, we obtain:

 u2   v2  2uv
1 - 2   xx + 1 − 2   yy − − 2  xy = 0 Eq. 9.4
 a   a  a

After the landmark paper by295 , it points to type dependent differencing. The idea is:

u+v u+v
− 1  0 → hyperbolic , − 1  0 → elliptic Eq. 9.5
a a
If the flow is subsonic, elliptic equation prevails and central differencing are used for the derivatives.
On the other hand, if the flow is supersonic, the equation is hyperbolic and upwind differencing could
be used.
9.3.1 Conservative Forms
The conservation form potential equation for 2D may be written:

295 Murman, E, M, and Cole, J, D, “Calculation of Plane Steady Transonic Flows”, AIAA Journal, Volume 9, pp. 114-

121, 1971.
 231

 ( u)  ( v)  
+ +0 wh ere u = ,v =
x y x y
1/γ −1
 γ −1 2 2 
from Energy equation ρ = 1 − M  (u + v 2 − 1
 2 
 A1 = ξ 2x + ξ 2y
 U  V 
 ρ  +  ρ  = 0 where U = A1 ξ + A 2 η , V = A 2 ξ + A 3 η  A 2 = ξ x ηx + ξ y ηy
 J ξ  J η  A = η2 + η2
 3 x y

 ( , )
J= = ξ x η y − ξ y η x where ξ x = Jyη , ξ y = −Jx η , η x = −Jyξ , η y = Jx ξ
 (x, y)
1/(γ −1)
 γ −1 2 
ρ = 1 − M  (U x + V y − 1)
 2 
Eq. 9.6
Where U and V are contravariant velocity components.

9.4 Method for Transonic Flow

The full potential equation is useful for describing transonic flows when Shock wave’s strength is
small. We use the scheme of Retarded density as an example. A second-order finite differencing
approximation can be written for a 2D conservative potential equation in conservative form as:

 ρU   ρV 
ξ   + η   =0
 J  ,j
i +1  J  i, j+1
2 2

1
U i +1/2, j = (A1 ) i +1/2, j ( i +1, j −  i, j ) +(A 2 ) i +1/2, j ( i +1, j+1 −  i +1, j−1 +  i, j+1 −  i, j−1 )
4
1
V i, j+1/2 = (A3 ) i, j+1/2 ( i, j+1 −  i, j ) + (A 2 ) i, j+1/2 ( i +1, j+1 −  i −1, j+1 +  i +1, j −  i −1, j )
4
 U   ρV 
 ξ ~ ρi    + η   =0
  J  i +1/2, j   J  i, j+1
2

where retarded density ρ = (1 − ν ~ )ρ +ν ρ

i i + k, j i +1/2, j i + k, j i + 2k −1/2, j

0 U i +1/2, j  0   1 
k= ν = max 0, C1 1 − 
M 2 
with and
1 U i +1/2, j  0  
Eq. 9.7
Where the constant C1 is unity for small regions of supersonic flow but must be increased in regions
where shocks strength is appreciable. This information may be used throughout the flow to solve the
full potential equation at points which are either elliptic or hyperbolic. The solution procedure used
to solve the set of resulting difference equations may take many forms. We choose the Approximate
Factorization (AF) scheme. The full potential equation may be written as

N(ϕn+1 − ϕn ) + ωL(ϕn ) = 0
Eq. 9.8
 232

Where ω is the relaxation parameter, L(ϕn) represents the residual of resulting equation, and N is
operator which determines the iteration method consist of N = N1N2. The N1 and N2 must be selected
so their product approximates L. The procedure uses simple matrix operation and the overall scheme
is stable.
9.4.1 Case Study - Panel Methods for Solving the Laplace’s Equation
Potential flow plays an important historical role in the theory of flight since 50s and before the CFD
applications become fashionable. Numerical models based on this approach are known as panel
methods in the aerodynamics community. Panel methods are numerical models based on simplifying
assumptions about the physics and properties of the
flow of air over an aircraft296. It is well known fact that
the Boing® used panel techniques for design of original
747s where they are still in service. The viscosity of air
in the flow field is neglected, and the net effect of
viscosity on a wing is summarized by requiring that the
flow leaves the sharp trailing edge of the wing smoothly.
The compressibility of air is neglected, and the curl of
the velocity field is assumed to be zero (no vorticity in
the flow field). Under these assumptions, the vector
velocity describing the flow field can be represented as
the gradient of a scalar velocity potential, to solve the
problem of potential flow over a solid object. Laplace’s
equation must be solved subject to the boundary
condition that there be no flow across the surface of the
object. This is usually referred to as the tangent-flow
boundary condition. Additionally, the flow far from the
object is required to be uniform. The results of solving
Laplace’s equation subject to tangent-flow boundary Figure 9.3 Cp Contours for a Rotor using
Panel Method
conditions provide an approximation of cruise
conditions for an airplane. The basic solution procedure
for panel methods consists of discretizing the surface of the object with flat panels and selecting
singularities to be distributed over the panels in a specified manner, but with unknown singularity-
strength parameters. Since each singularity is a solution to Laplace’s equation, a linear combination
of the singular solutions is also a solution. The tangent-flow boundary condition is required to be
satisfied at a discrete number of points called collocation points. This process leads to a system of
linear algebraic equations to be solved for the unknown singularity-strength parameters. Details of
the procedure vary depending on the singularities used and other details of problem formulation, but
the end result is always a system of linear algebraic equations to be solved for the unknown
singularity-strength parameters. Panel methods are applicable to 2 and 3D flows (see Figure 9.3)
and has one distinctive feature. That is a solution for body pressure distribution can be obtained
without solving the entire domain. For flow over a 2D, the flat panels become straight lines, but can
be thought of as infinitely long rectangular panels in the three-dimensional interpretation.

9.5 Vortex Lattice Method (VLM)

The Vortex Lattice Method, (VLM), is a numerical method used in computational fluid dynamics,
mainly in the early stages of aircraft design and in aerodynamic education at university level297. The
VLM models the lifting surfaces, such as a wing, of an aircraft as an infinitely thin sheet of discrete

296 R. L. Fearn, “Airfoil Aerodynamics Using Panel Methods”, The Mathematica Journal, 2008.
297 From Wikipedia.
 233

vortices to compute lift and induced

drag. (see Figure 9.4). The influence
of the thickness, viscosity is
neglected. The vortex lattice method
is built on the theory of ideal flow,
also known as Potential flow. Ideal
flow is a simplification of the real flow
experienced in nature, however for
many engineering applications this
simplified representation has all of
the properties that are important
from the engineering point of view. Figure 9.4 Simulation of an Airplane Based in the VLM
This method neglects all viscous
effects. Turbulence, dissipation and boundary layers are not resolved at all. However, lift induced
drag can be assessed and, taking special care, some stall phenomena can be modelled. By the above
assumptions the flow field is conservative vector field, which means that there exists a perturbation
velocity potential φ such that the total velocity vector V is given by

V = V∞ + ∇ϕ
Eq. 9.9
and that φ satisfies Laplace's equation. Laplace’s equation is a second order linear equation, and
being so it is subject to the principle of superposition. Which means that if φ1 and φ2 are two solutions
of the linear differential equation, then the linear combination c 1 φ1 + c2 φ2 is also a solution for any
values of the constants c1 and. As [Anderson] put it "A complicated flow pattern for an irrotational,
incompressible flow can be synthesized by adding together a number of elementary flows, which
are also irrotational and incompressible.”. Such elementary flows are the point source or sink, the
doublet and the vortex line, each being a solution of Laplace’s equation. These may be superposed in
many ways to create the formation of line sources, vortex sheets and so on. In the Vortex Lattice
method, each such elementary flow is the velocity field of a horseshoe vortex with some strength Γ298.
9.5.1 Case Study - Transonic and Viscous Potential Flow Method Applied to Flexible Wing
Transport Aircraft
Authors : Daniel Chaparro1, Gustavo E. C. Fujiwara2, Eric Ting3 and Nhan Nguyen3
Affiliations : 1MORI Associates, Moffett Field, CA 94035, 2University of Washington, Seattle, WA 98195,
3NASA Ames Research Center, Moffett Field, CA 94035

Title of Article : Transonic and Viscous Potential Flow Method Applied to Flexible Wing Transport
Aircraft
Source : AIAA 2017-4221, AIAA Aviation Forum, 5-9 June 2017, Denver, Colorado, 35th AIAA Applied
Aerodynamics Conference, DOI: 10.2514/6.2017-4221
Adaption : None except minor modification for formatting
9.5.1.1 Abstract
The need to rapidly scan large design spaces during conceptual aerodynamic design calls for
computationally efficient tools such as the vortex lattice method (VLM). Although some VLM tools,
such as Vorview/Vorlax have been extended to model fully supersonic flow, VLM solutions are
typically limited to inviscid, subcritical flow regimes. Many transport aircraft operate in the transonic
speed regime, which limits the applicability of the VLM for such applications. This paper presents a
novel approach to augment potential flow methods with a correction method for transonic and

298 From Wikipedia.

 234

viscous aerodynamic effects present for many transport aircraft. The method leverages a directly
coupled transonic small disturbance (TSD) and integral boundary layer (IBL) framework to
capture transonic and viscous flow physics. The viscous and transonic flow corrections are
integrated into a static aeroelastic modeling framework and applied to the NASA General Transport
Model (GTM) equipped with a novel control surface known as the Variable Camber Continuous
Trailing Edge Flap (VCCTEF).
In the same manner, the aeroelastic framework is also adapted to couple MSES, a fully-simultaneous
Euler/IBL solver, with the VLM in the loop. Aeroelastic solutions and computational wall clock time
from the TSD/IBL and MSES transonic viscous flow potential flow models are compared with higher
fidelity 3D aerodynamic solvers. The wall clock time of the transonic viscous flow potential flow
model is two orders of magnitude less than the three-dimensional Euler solver Cart3D. It is found
that that GTM lift slope calculated by the transonic and viscous potential flow method agrees with
Reynolds Averaged Navier-Stokes (RANS) to within 2.0% and the cruise lift over drag ratio agrees to
within 3.6%. A large design space of candidate VCCTEF configurations is evaluated using the TSD/IBL
and MSES transonic and viscous corrections and the best performing VCCTEF candidates are
compared. The VCCTEF candidate with the largest drag reduction is predicted to reduce cruise drag
by 7.04% and 5.71% by the TSD/IBL and MSES models, respectively. Furthermore, both models agree
on the best performing VCCTEF configuration.

Figure 9.5 VCCTEF Configuration on a Generic Transport Model (GTM)

9.5.1.2 Introduction
The commercial transport industry is trending towards incorporating composite materials and other
lightweight materials with the goal of achieving more energy-efficient aircraft. Weight reduction
leads to lower lift requirements and subsequent reductions in drag and thrust requirements.
Ultimately, lower drag and thrust translate into higher efficiency and lower fuel consumption. The
additional structural flexibility of these lightweight materials can, however, lead to aeroelastic
 235

interactions that can degrade aerodynamic performance at off-design conditions. Recovering off-
design performance through the use of active wing shaping control can maximize the benefits of
employing lightweight/highly flexible aerospace materials in wing structures. In 2010, a conceptual
study titled “Elastically Shaped Future Air Vehicle Concept”[1] was conducted by NASA to investigate
multiple active wing shaping control concepts for flexible wings. The study proposed the Variable
Camber Continuous Trailing Edge Flap (VCCTEF) control effector as a means to elastically control the
wing washout twist and wing bending deflection to change the local angle of attack to reduce drag
[1–4]. Under the Fixed Wing project Active Aeroelastic Shape Control (AASC) element, NASA and
Boeing conducted a
joint study to
investigate the
application and
potential of the
VCCTEF [5-6]. As
shown in Figure 9.5,
the VCCTEF is
composed of multiple
spanwise flap Figure 9.6 VCCTEF Section with 3 Camber Segments (Green) Compared to a
sections connected by Traditional Flap System (Blue)
flexible elastomer
material. Unlike traditional flap systems, elastomer material is incorporated as a means to prevent
vorticity generation between flap sections that occurs if a geometric discontinuity exists. The VCCTEF
configuration for the General Transport Model (GTM) is comprised of three chordwise segments per
spanwise flap, as shown in Figure 9.6 where each segment can finely tune the camber of the wing at
any point within the flight envelope.
The aircraft used in this study is based on the NASA Generic Transport Model (GTM), which is a
research model that originated from NASA Langley Research Center. The model is selected because
subsonic wind tunnel data is available for validation of computational models. The GTM is a notional
single-aisle, mid-
size, 200-passenger
aircraft.1 The
design lift
coefficient is ¯CL =
0.51 at the design
mid-cruise
condition of M =
0.797 at 36000 ft,
however, the
stiffness properties
of the GTM wing are
scaled by half to
represent current
trends towards
lightweight and Figure 9.7 Illustration of the GTM Aircraft Equipped with the VCCTEF
flexible wing
designs. Therefore, the GTM with the flexible wing is modeled as having a cruise ¯CL = 0.497 to
account for the weight benefit of a lighter, but more flexible wing. The GTM equipped with the
VCCTEF is shown in Figure 9.7.
Previous studies of the GTM equipped with the VCCTEF have examined the drag reduction benefits of
the VCCTEF by estimating skin friction drag and predicting induced drag through the use of the
 236

VLM.1, 2 Since VLM does not account for shock effects, wave drag was not optimized. Since the GTM
cruise speed is Mach = 0:797, it is necessary to account for wave drag and transonic effects on lift and
induced drag. In a follow up study by ([Lebofsky et al,)[7] an Euler solver was leveraged to create
two-dimensional (2D) lift and drag databases from which to sample and augment the VLM spanwise
lift and drag distributions. The study also refined the estimate for skin friction drag by using a flat
plate analytical approximation. The framework was efficient and accounted for transonic effects on
lift and drag. A limitation of the approach is that it relied on having a database of high-fidelity airfoil
data. For example, if one wished to optimize the layout of the VCCTEF, it would be necessary to create
an exhaustive database of all the configurations to be evaluated a priori. Creating the database can
be time consuming and may yield erroneous predictions for large wing deformations. The approach
also did not capture viscous boundary layer effects on lift and drag. In 2016, a single-pass framework
that coupled 2D Transonic Small Disturbance (TSD) and Integral Boundary Layer (IBL) solutions
in the loop with the VLM was developed by Chaparro et al8 to capture transonic and viscous effects
on lift and drag.
This paper expands on the viscous and transonic corrections presented in [2016 Van Dam et al]
presented a methodology to couple 2D and 3D aero models for high lift predictions [9,10]. The
algorithm is adapted to couple the VLM with TSD/IBL by altering the incidence angle across the wing
in the VLM model to capture viscous and transonic effects. The iterative re-twisting scheme is
incorporated into an aeroelastic framework that is optimized to efficiently couple the Finite Element
Analysis (FEA), VLM and TSD/IBL models. The framework is also adapted to use MSES, a fully
simultaneous Euler/IBL solver, instead of TSD/IBL, and the results are compared to the TSD/IBL
model and other high fidelity 3D aerodynamic solvers.
9.5.1.3 Aero-Structural Modeling
9.5.1.3.1 Vortex Lattice Model
Vorlax is an aircraft aerodynamic modeling tool
based on the VLM. The VLM is an extension of
Prandtl’s lifting line theory that is applicable to a
broader range of lifting surfaces including swept
and low aspect ratio wings. However, limitations
associated with the VLM in general apply to
Vorlax aerodynamic analysis. For example, Vorlax
is limited to subcritical or fully supersonic
potential flow, consequently, the drag prediction
from Vorlax accounts only for induced drag. Wave
drag and viscous drag must be modeled by other
methods. Vorview is the pre/post processor for
the Vorlax code that is used to generate the lattice
Figure 9.8 GTM Aircraft Wing-Body Model in
model and analyze results [11]. Once the VLM Vorview
model is generated by Vorview, it can be more
computationally efficient to deform the VLM
panels in Vorlax without using Vorview if the airfoil shape can be assumed to not deform. The Vorlax
model of the GTM has been validated by wind tunnel data [12]. The Vorlax GTM wing-body
configuration is composed of 118 spanwise panels with 12 chordwise segments each and is shown
Figure 9.8.
9.5.1.3.2 Transonic Small Disturbance Model / Integral Boundary Layer Model
TSFOIL is a TSD solver for flow past lifting airfoils [13]. TSFOIL is chosen for its rapid solve time, ease
of use, and its open-source architecture. The method has been shown to compare well with other
Euler solvers for moderate angles of attack. More detail about the TSD method as it pertains to the
present framework as well as 2D validation cases are available in previous work [8]. A mesh
 237

sensitivity study is conducted

to ensure that lift and drag do
not change if the mesh is
further refined, and the results
are shown in Figure 9.9. Lift
is invariant across the three
analyzed meshes. Some small
differences are seen at high lift,
but this study focuses on cruise
conditions. As expected, drag
is more sensitive to the mesh
than lift. Coarsening the
baseline mesh by half does
change the drag polar, but
refining it to the maximum
allowable mesh size in TSFOIL
does not appreciably alter the
drag prediction.
The IBL model is presented in
detail in prior work [14]. The
IBL model employs a direct
sequential coupling of the TSD
and viscous boundary layer
equations with special
treatment near the trailing
edge to prevent the boundary
layer solution from becoming
singular, known as Goldstein’s
singularity [15]. A Goldstein
singularity results from strong
interactions between the
boundary layer and the
inviscid flow near the trailing
edge separation point where
strong adverse pressure
gradients are present. For a
given Cp distribution across the
airfoil surface, the model
marches from the leading edge
stagnation point to the trailing
edge calculating the boundary
layer displacement thickness Figure 9.9 TSFOIL Mesh Sensitivity Study; Lift (a) and Drag (b)
at each node. In the laminar
region, the model uses the compressible formulation of Thwaites’ method [16] to march the solution
towards the laminar to turbulent transition point. Michel’s method [17] is used to locate the
transition point. After transition, the compressible formulation of Head’s method [18] is used to
march to the trailing edge. The displacement thickness over the aft 5% of the chord is linearly
extrapolated to avoid singular solutions near the trailing edge. Flow separation is determined when
the Head shape factor exceeds 2.2. The IBL solution is not valid once the flow is separated; therefore,
when flow separation is predicted, the displacement thickness is linearly extrapolated beginning 5%
 238

upstream of the separation point. Further detail of the IBL model can be found in Ref. [14]. The IBL
model has been shown to agree well with XFOIL19 for subsonic flow and with RANS for transonic
flow [14]. The TSD/ IBL loop typically converges on lift within four to five iterations for the GTM at
cruise conditions.
9.5.1.3.3 Euler-IBL Model (MSES)
MSES is a widely accepted airfoil analysis code developed by Dr. Mark Drela [20] that is capable of
capturing transonic and boundary layer physics. MSES is a fully simultaneous solver, meaning that it
solves the Euler and boundary layer equations together using the global Newton method. Details
regarding the implementation and theory behind MSES can found in Ref. [20]. MSES is integrated in
the loop to handle arbitrary
VCCTEF configurations
without a need for look-up
tables. MSES is also
integrated into the
aeroelastic framework by
automating the mesh
coarsening/refinement steps
listed in Table 9.1 in order
to minimize the number of
airfoils that fail to converge.
The mesh refinement
sequence is tuned for Table 9.1 Automated Mesh Refinement Sequence for MSES
robustness such that the
number of GTM airfoils that
do not converge at cruise conditions is minimized; however, it may be worthwhile to further tune the
sequence for other aircraft and/or operating conditions. Care is also taken to initialize MSES from a
previous solution whenever possible to reduce computation time. A mesh sensitivity study is
conducted and the results are shown in Figure 9.10. Lift is essentially insensitive to the three
meshes. While drag changes between the coarse and baseline mesh; the change between the baseline
and refined meshes is negligible. As a result, MSES is allowed to break out of the automatic mesh
refinement sequence as soon as it converges, but in order to ensure consistent drag predictions, MSES

Figure 9.10 MSES Mesh Sensitivity Study ; Lift (a) and Drag (b)
 239

is always re-run with the baseline mesh when the outer transonic and viscous correction loop
converges.
9.5.1.3.4 Static Aeroelastic Modeling Framework
A static aeroelastic modeling framework is used to couple the aerodynamics models to a finite
element model to appropriately model the wing aero-structural interaction at a given operating
condition. The structural model is a 3D beam finite element model developed by (Nguyen, Ting et
al.) [21,22]. Following prior work [7], the stiffness properties of the GTM wing are scaled by half to
represent current trends towards lightweight and flexible wing designs. The user can opt to specify
the aircraft angle of attack and solve for the aerodynamic loads, known as angle of attack mode.
Alternatively, the user can specify the aircraft lift coefficient and solve for the angle of attack and
aerodynamic loads, known as fixed lift coefficient mode. The framework is depicted in both fixed
angle of attack and fixed lift coefficient mode in Figure 9.11. The geometry deformer deflects the
VCCTEF and generates the geometry input files for the VLM and TSD/IBL models. Vorview/Vorlax is
used to generate the vortex lattice model of the undeformed wing prior to applying the transonic and
viscous corrections and coupling with the FEA model. In fixed angle of attack mode, the VLM accounts
for the structural deformation by translating the wing VLM panels along the three cartesian
directions and by twisting the wing panels about the pitch axis. The VLM model is executed and the
aerodynamic loads corresponding to the deformed wing are then input into the FEA model. The two
solvers iterate until the wing tip twist converges. Similarly, in the fixed lift coefficient mode, the
aeroelastic lift matching loop iterates between Vorlax and FEA until the tip twist converges while
simultaneously converging to the specified aircraft lift coefficient by solving for the angle of attack
using to the aircraft lift curve slope.

Figure 9.11 Static Aeroelastic Modeling Framework for Fixed Angle of Attack (α) and Fixed Lift
Coefficient (b)

The transonic and viscous correction loop couples TSD/IBL with the VLM and FEA models. The
methodology for the transonic and viscous corrections to VLM is described in detail in Section II.E.
It is worthwhile to note that the TSD/IBL correction is the most computationally expensive step in
 240

the framework because it is applied on each spanwise section, but placing the TSD/IBL correction
outside of the aeroelastic loop minimizes the amount of TSD function calls and thus significantly
reduces overall computation time. In order to move the TSD/IBL correction outside of the aeroelastic
loop, it is assumed that the airfoils only twist about the pitch axis. However, twist about the yaw and
roll axis have a relatively minor impact on the aerodynamic loads.
9.5.1.3.5 Transonic and Viscous Corrections for the VLM
The transonic and viscous corrections leverage strip theory to couple the VLM with higher fidelity 2D
solutions. In this study, a total of 33 streamwise sections along the span of the wing, as shown in
Figure 9.12, are analyzed using either TSD/IBL or MSES to augment the VLM to capture transonic
and viscous effects.

Figure 9.12 Streamwise Slices of the GTM wing for 2D Aerodynamic Analysis

The flow chart shown in Figure 9.13 highlights the major building blocks of the transonic and
viscous correction method. Let y be the spanwise coordinate along the wing. The transonic and
viscous corrected aerodynamic loads are calculated using the following process:
1 The virtual twist g, and elastic twist Фy, for each section along the span are initialized to zero.
2 The incidence angle for each VLM panel is set to equal the sum of the geometric incidence
angle αinc, the elastic twist Фy, and the virtual twist angle:

̃inc (y) = αinc (y) + ϕy (y) + γ(y)

α
Eq. 9.10
Positive incidence angle is nose up.
3 The VLM and FEA models iterate at the specified input flight condition and until Фy converges.
The aeroelastic loop yields the wing deformation and the aircraft aerodynamic load
coefficients including spanwise distributions of lift coefficient cl(y), induced drag coefficient
cdi(y), and quarter-chord moment coefficient cmqc (y).
4 The 3D/2D transformation calculates the effective angle of attack:

cl vlm
α2D (y) = α0 (y) + − γ(y)
clα
Eq. 9.11
for each section where a0 is the zero lift angle of attack, and clvlm (y) is the section lift
 241

Figure 9.13 Transonic and Viscous Correction Method Flow Chart

coefficient at y from the VLM model. cla is the 2D lift slope with the Prandtl-Glauert
compressibility correction such that

2π
cla =
√1 − MΛ2 c/2
Eq. 9.12
where MΛc/2 is the sweep-corrected Mach number with respect to mid-chord.
5 If the TSD/IBL model is selected, steps 5(a) and 5(b) are repeated until the lift coefficient
converges for each spanwise section before proceeding to step 7.
(a) The TSD model is executed for each section at the effective 2D angle of attack. Surface
pressure and velocity are calculated for the upper and lower airfoil surfaces.
(b) The local pressure and velocity distributions are used by the IBL model to calculate
the boundary layer displacement thickness. The displacement thickness is then added
to the airfoil coordinates, effectively thickening the airfoil.
6 If the MSES model is selected, MSES is executed for each section at the effective 2D angle of
attack.
7 The virtual twist angle of each section is updated as
 242

cl2Di (y) − clvlmi (y)

γi+1 (y) = γi (y) +
cla
Eq. 9.13
where cl2D is the airfoil lift coefficient computed by either TSD/IBL or MSES.
8 Steps 2 through 7 are repeated until the VLM sectional lift and the airfoil lift from either
TSD/IBL or MSES converges for all sections.
9 Viscous and wave drag from the 2D solutions are integrated over the span to calculate the
total viscous and wave drag of the aircraft whereas the lift, pitching moment, and induced
drag are calculated by the re-twisted VLM model.

Table 9.2 Comparison of 3D Inviscid Lift Curve and Drag Polar Metrics of Interest for the Rigid GTM

9.5.1.4 Results
9.5.1.4.1 Rigid Clean Wing GTM Results
1. Comparison of Inviscid Aerodynamic Solutions
The GTM is first modeled without the VCCTEF deflected, aeroelastic deformation, and viscous
correction. Five inviscid solutions are compared in Figure 9.14. The standalone VLM model is
considered the lowest order model out of the five because it does not capture transonic effects or
lifting surface thickness effects. The other solutions are VLM coupled with TSD, VLM coupled with
MSES and two solutions from a 3D Euler solver, Cart3D [27]. The two Cart3D solutions correspond
to a coarse mesh for exploratory studies that could be run on a local work station and a finer mesh
for higher fidelity drag calculations that must be run on a high performance cluster [28].
Unfortunately only one data point for the fine-mesh Cart3D model is available. Table 9.2 compares
the 3D aircraft CLa , CL0 and the lift over drag ratio (L/D) at cruise. The VLM+TSD and VLM+MSES
models under-predict CLa by 8.7%, and 7.4%, respectively, when compared to Cart3D (coarse). CL0 is
under-predicted by 8.2% and 9.6% by the VLM+TSD and VLM+MSES models respectively. The L/D
ratio at cruise predicted by the VLM+TSD and VLM+MSES models agree with the fine-mesh Cart3D
model within 0.28% and 5.5% respectively.

Figure 10. Lift Curve (Left) and Drag Polar (Right) for the Rigid GTM Using Inviscid Aero Solvers
9.5.1.4.2 Comparison of Viscous Aerodynamic Solutions
The GTM without the VCCTEF deflected and no aeroelastic deformation is modeled with viscous
effects. Three viscous solutions
are compared in Fig.11 and
Table 9.3: VLM+TSD/IBL,
VLM+MSES, and RANS
(LAVA)[28,29]. LAVA is a grid-
flexible computational fluid
dynamic solver with specialized
modules for launch vehicle
Table 9.3 Comparison of 3D Viscous Lift Curve and Drag Polar
applications developed at NASA
Metrics of Interest for the Rigid GTM
AmesFigure
Research Center.
9.14 Lift Both and Drag Polar (Right) for the Rigid GTM Using Inviscid Aero Solvers
Curve (Left)
 243

transonic and viscous potential flow models compare favorably with the RANS model. Although a
slight offset is seen between the RANS and the transonic and viscous flow VLM lift curves, the lift
slopes of the VLM+TSD/IBL and VLM+MSES models agree with RANS to within 2.0% and 0.3%
respectively as shown in Table 9.3. The cruise L/D predicted by the VLM+TSD/IBL and VLM+MSES
models agree with RANS to within 3.6% and 2.1% respectively.

Figure 9.15 Lift Curve (Left) and Drag Polar (Right) for the Rigid GTM Using Viscous Aero Solvers

9.5.1.4.3 Aeroelastic Clean Wing GTM Results

The full aeroelastic framework described in Section 7 is exercised to model the flexible GTM with the
transonic and viscous corrections. The lift curves and drag polar for the flexible GTM calculated using
the TSD/IBL and MSES corrections are compared in Figure 9.16. The cruise condition predicted
with RANS by (Denison et al)[28] is also included for comparison. The offset in lift between RANS
and the transonic and viscous potential flow models seen in the rigid analysis is also present in the
aeroelastic results. However, the cruise L/D predicted by the TSD/IBL and MSES corrections agree
with RANS to within 2.4% and 0.1%, respectively. The aeroelastic framework is able to capture how
the wing deformation produces a more triangular lift distribution as shown in Figure 9.17. The

Figure 9.16 Lift Curve (a) and Drag Polar (b) for the Aeroelastic GTM with Transonic and Viscous Flow
VLM Models
 244

pressure coefficient contour plots in Figure 9.18 and the lift distributions from the two transonic
and viscous flow VLM models agree well with each other, which suggests that the TSD/IBL model can
capture viscous and transonic physics with similar fidelity as MSES.

Figure 9.17 Aeroelastic Effect on Lift Distribution using TSD/IBL (a) and MSES (b) at C L = 0.497

9.5.1.4.4 Rigid GTM Wing with VCCTEF Deployed

The transonic and viscous corrections are leveraged to rapidly evaluate a large number of VCCTEF
candidate configurations for the rigid wing GTM model at the baseline cruise ¯CL = 0.51. The analysis
is conducted with the TSD/IBL correction and then repeated with MSES to compare the drag
prediction capability of the two models. For this study, the 16 VCCTEF flaps are composed of three
chordwise segments that are constrained such that

δ𝑖
δ𝑘𝑖 = k
3
Eq. 9.14
where δi is the total flap deflection for ith flap and
δki is the deflection for the kth chordwise
segment of the ith flap. The total VCCTEF
deflections for the spanwise segments are
parameterized according to the first four
Chebyshev polynomials:

1
2ηi
δi = [c1 c2 c3 c4 ]
4η2i − 1
[8η3i − 4ηi ]
Figure 9.18 Contour Plots of Pressure
i−1 Coefficient Obtained with the Transonic and
where ηi = −1 + Viscous Potential Flow Models for the Flexible
nf − 1 GTM at
Eq. 9.15 CL = 0.497
 245

where nf = 16. A total of 85 VCCTEF candidates are selected by finding combinations of the
polynomial parameters, cj, subject to the following constraints

|δi − δi+1 | ≤ 2° , 1° ≤ δ𝑖 ≤ 6° , − 10 ≤ c𝑗 ≤ 10
Eq. 9.16
The drag reduction achieved by the
VCCTEF, ΔCD, is defined as

∆CD = CDclean − CDvcctef

Eq. 9.17
Of the 85 VCCTEF candidates
evaluated, the average drag
reduction predicted by the TSD/IBL
model is 2.3 drag counts larger than
the MSES model. The flap deflections
for the top five performing VCCTEF
configurations according to both
models are plotted in Fig. 15 of
[Chaparro et al. “Transonic and
Viscous Potential Flow Method
Applied to Flexible Wing Transport
Aircraft”, 2017]. Note that the Table 9.4 Drag Reduction (counts) for the Top Five GTM
results from the TSD/IBL and MSES VCCTEF Configurations
models are very similar. The two
models agree on the first, third and fourth best configurations, and their fifth best configurations have
the same shape but are offset by 0.75 degrees. The drag reduction predicted by the TSD/IBL and MSES
models is tabulated for the top five performing VCCTEF configurations in Table 9.4. The total drag

Figure 9.19 Lift Distribution with the Best Performing VCCTEF Candidate Calculated with TSD/IBL (a)
and MSES (b)
 246

reduction estimated for the best configuration is 7.04% and 5.71% according to the TSD/IBL and
MSES models, respectively.

Figure 9.20 Change in Cp due to the Best Performing VCCTEF Candidate Calculated with the TSD/IBL
(a) and MSES (b) Models

Figure 9.19 shows how the lift distribution of the rigid GTM changes with the best performing
VCCTEF candidate. Note that lift distribution over the inboard half of the wing does not move towards
the elliptical lift distribution to minimize induced drag. By coupling the transonic and viscous
corrections to the VLM, the tradeoff between wave and induced drag becomes apparent. In this study,
the best performing VCCTEF candidate is reducing the total aircraft drag by accepting an induced
drag penalty in order to reduce wave drag, which is highest near the root where the airfoils are
thickest. Figure 9.20 compares the pressure coefficient contours between the clean wing and the
wing with the best performing VCCTCF candidate. The pressure contours demonstrate that deploying
the VCCTCF reduces wave drag by delaying the shock onset.
9.5.1.4.5 Wall Clock Time Comparison
The motivation for developing the approach is to develop a VLM-based method that captures
transonic and viscous physics while maintaining relatively short wall clock times. Figure 18 shows a
wall clock comparison between six aerodynamics models calculating a single fixed lift aeroelastic
solution: VLM, VLM+TSD, VLM+TSD/IBL, VLM+MSES(Euler), VLM+MSES(Viscous), and Cart3D. Note
that the calculations are conducted on a Macbook Pro with 16GB of memory and a 2.5 GHz Intel i7
processor, which has 4 cores/8 threads. Unfortunately, Vorlax does not have the ability to run on
multiple cores, but the TSD, MSES and Cart3D models are executed on all 4 cores. The VLM+TSD/IBL
run time is longer than the VLM+TSD time because the number of TSFOIL executions increases when
the IBL code is coupled. Nonetheless, the VLM+TSD/IBL is approximately three times faster than
VLM+MSES and two orders of magnitude faster than Cart3D. The shorter wall clock time is especially
important during conceptual design when there is a need to explore large design spaces with
reasonable accuracy. The computation time for the TSD/IBL and MSES models can be further reduced
by dynamically building a database of 2D solutions and interpolating lift as function angle of attack
and VCCTEF deflection. While it is not practical to run aeroelastic RANS cases for this study on a
 247

Macbook Pro, it is
worth noting that a
single rigid-wing LAVA
(RANS) case ran on
320 processors on the
NASA Pleiades
supercomputer
converges in about
four hours, which
highlights the
practicality of the
proposed transonic
and viscous potential
flow method for
flexible wing aircraft.
9.5.1.5 Conclusion
A novel approach to Figure 9.21 Wall Clock Comparison Between Multiple Aerodynamic Codes
for a Single Fixed Lift Aeroelastic Solution
efficiently augment
potential flow
methods with corrections for transonic and viscous effects using TSD coupled with IBL is developed
and applied to the GTM equipped with a novel control surface known as the Variable Camber
Continuous Trailing Edge Flap (VCCTEF). The transonic and viscous potential flow model is shown to
be in close agreement with the LAVA RANS solver; the lift curve slope agrees to within 2.0% and the
cruise L/D agrees to within 3.6%. The framework is also adapted to couple MSES, a fully
simultaneous Euler- integral boundary layer solver, with vortex lattice in the loop and the results are
compared to the TSD/IBL model. A large design space of VCCTEF candidates is evaluated using the
TSD/IBL and MSES augmentations for the VLM. The study demonstrated the VCCTEF’s ability to
reduce the overall drag of the rigid-wing GTM by 7.04% and 5.71% according to the TSD/IBL and
MSES VLM models, respectively. The study highlights the VCCTEF’s ability to tailor the camber of the
wing to reduce wave drag, which cannot be modeled with the VLM alone. The study also shows that
the TSD/IBL and MSES corrections agree on the best VCCTEF candidates, which suggests that the
TSD/IBL method can capture transonic and viscous flow physics with similar fidelity as the widely
accepted MSES code. Both transonic and viscous flow VLM models are orders of magnitude faster
than 3D Euler and RANS solvers. The MSES +VLM wall clock time is 3.1 times larger than the TSD/IBL
VLM model. The fast wall clock time and close agreement with RANS makes the VLM+TSD/IBL model
ideal for rapid design optimization studies during early design, which would otherwise be
impractical with RANS or Euler solvers especially when accounting for aeroelastic deformation of the
aircraft structures.

Acknowledgments
The authors would like to thank Marie Denison at Science and Technology Corporation in the NASA
Computational Aero-science Branch for providing the LAVA and Cart3D data used to compare with the
VLM modeling results. The authors would also like to thank the Advanced Air Transport Technology
(AATT) Project under the Advanced Air Vehicles Program (AAVP) of NASA Aeronautics Research Mission
Directorate (ARMD) for funding support of this work. The authors also would like to acknowledge the
work done by Boeing Research & Technology (BR&T) and the University of Washington for their
collaboration with NASA on this work under tasks NNL11AD25T and NNL12AD09T entitled
“Development of Variable Camber Continuous Trailing Edge Flap System” and “Development of Variable
Camber Continuous Trailing Edge Flap System for B757 Configured with a More Flexible Wing”.
 248

9.5.1.6 References
[1] Nguyen, N., “Elastically Shaped Future Air Vehicle Concept,” NASA Innovation Fund Award 2010
Report, http://ntrs.nasa.gov/archive/nasa/casi.ntrs.nasa.gov/20110023698_2011024909.pdf,
Submitted to NASA Innovative Partnerships Program Office, October 8, 2010.
[2] Nguyen, N., and Urnes, J., “Aeroelastic Modeling of Elastically Shaped Aircraft Concept via Wing
Shaping Control for Drag Reduction,” AIAA Atmospheric Flight Mechanics Conference, AIAA-2012-
4642, August 2012.
[3] Nguyen, N., Trinh, K., Reynolds, K., Kless, J., Aftosmis, M., Urnes, J., and Ippolito C., “Elastically
Shaped Wing Optimization and Aircraft Concept for Improved Cruise Efficiency,” 51st AIAA
Aerospace Sciences Meeting including the New Horizons Forum and Aerospace Exposition, AIAA-
2013-0141, January 2013.
[4] Ippolito, C., Nguyen, N., Totah, J., Trinh, K., Ting, E., “Initial Assessment of a Variable-Camber
Continuous Trailing-Edge Flap System for Drag-Reduction of Non-Flexible Aircraft in Steady-State
Cruise Condition,” AIAA Infotech@Aerospace (I@A) Conference, AIAA-2013-5143, August 2013.
[5] Boeing Report No. 2010X0015, "Development of Variable Camber Continuous Trailing Edge Flap
System,” October 4, 2012.
[6] Urnes, Sr., J., Nguyen, N., Ippolito, C., Totah, J., Trinh, K., Ting, E., “A Mission-Adaptive Variable
Camber Flap Control System to Optimize High Lift and Cruise Lift-to-Drag Ratios of Future N+3
Transport Aircraft,” 51st AIAA Aerospace Sciences Meeting including the New Horizons Forum and
Aerospace Exposition, AIAA-2013-0214, January 2013.
[7] Lebofsky, S., Ting, E., Nguyen, N., “Multidisciplinary Drag Optimization of Reduced Stiffness
Flexible Wing Aircraft With Variable Camber Continuous Trailing Edge Flap,” 56th
AIAA/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference, AIAA-2015-1408,
January 2015.
[8] Chaparro, D., Fujiwara, G.E.C., Ting, E., and Nguyen, N., "Aerodynamic Modeling of Transonic
Aircraft Using Vortex Lattice Coupled with Transonic Small Disturbance for Conceptual Design", AIAA
Aviation 2016 Conference, Washington, D.C., AIAA paper 2016-3418, June 2016.
[9] Van Dam, C. P., Vander Kam, J.C., Paris, J.K., “Design-Oriented High-Lift Methodology for general
aviation and civil transport aircraft,” Journal of Aircraft, Vol. 38, No. 6, Nov.-Dec. 2001.
[10] Fujiwara, G.E.C., Nguyen, N., “Aero structural Design Optimization of a Subsonic Wing with
Continuous Morphing Trailing Edge”, 35th AIAA Applied Aerodynamics Conference, Denver, CO, June
5-9, 2017 (submitted for publication), 2017.
[11] Miranda, L.R., Elliot, R.D., and Baker, W.M., “A Generalized Vortex Lattice Method for Subsonic
and Supersonic Flow Applications,” NASA CR-2865, 1977.
[12] Nguyen, N., Nelson, A., Pulliam, T., “Damage Adaptive Control System Research Report,” NASA
Ames Research Center Internal Report, April, 2006.
[13] Stahara, S. S., “Operational Manual for Two-Dimensional Transonic Code TSFOIL,” NASA
Contractor Report 3064, December, 1978.
[14] Fujiwara, G. E. C., Chaparro, D., and Nguyen, N., "An Integral Boundary Layer Direct Method
Applied to 2D Transonic Small-Disturbance Equations," AIAA Aviation 2016 Conference, Washington,
D.C., AIAA paper 2016-3568, June 2016.
[15] Goldstein, Sidney. "On laminar boundary-layer flow near a position of separation." The Quarterly
Journal of Mechanics and Applied Mathematics 1.1, 1948, pp. 43-69.
[16] Thwaites, B., “Approximate Calculation of the Laminar Boundary Layer,” Aero. Quart., Vol. 1, pp.
245-280, 1949.
[17] Michel, R., “Etude de la Transition sur les Profiles d’Aile; Etablissement d’un Critere de
Determination de Point de Transition et Calcul de la Trainee de Profile Incompressible,” Tech. rep.,
ONERA, 1951. Report 1/1578A.
[18] Head, M.R., “Entrainment in the Turbulent Boundary Layer,” Aeronautical Research Council
Reports and Memoranda Report 3152, September 1958.
 249

[19] Drela M., “Newton solution of coupled viscous/inviscid multielement airfoil flows,” AIAA Fluid
Dynamics, Plasma Dynamics and Lasers Conference, AIAA-1990-1470, June 1990.
[20] Drela, M. and Giles, M. B., “Viscous-Inviscid Analysis of Transonic and Low Reynolds Number
Airfoils,” AIAA Journal, Vol. 25, No. 10, Oct. 1987.
[21] Nguyen, N., Ting, E., Nguyen, D., Dao, T., and Trinh, K., “Coupled Vortex-Lattice Flight Dynamic
Model with Aeroelastic Finite-Element Model of Flexible Wing Transport Aircraft with Variable
Camber Continuous Trailing Edge Flap for Drag Reduction,” AIAA Atmospheric Flight Mechanics,
AIAA- 2013-4746, Boston, MA, August 2013.
[22] Ting, E., Nguyen, N., and Trinh, K., “Static Aeroelastic and Longitudinal Trim Model of Flexible
Wing Aircraft Using Finite-Element Vortex-Lattice Coupled Solution,” 55th
AIAA/ASME/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials Conference, AIAA-
2014-0837, National Harbor, MD, January 2014.
[23] Lebofsky, S., Ting, E., Nguyen, N., “Aeroelastic Modeling and Drag Optimization of Aircraft Wing
with Variable Camber Continuous Trailing Edge Flap,” 32nd AIAA Applied Aerodynamics Conference,
AIAA-2014-2443, June 2014.
[24] Ting, E., Nguyen, N., Lebofksy, S., “Static Aeroelastic Modeling of a Sub-Scale Wind Tunnel Model
with Novel Flap Concept,” 56th AIAA/ASCE/AHS/ASC Structures, Structural Dynamics, and Materials
Conference, AIAA-2015-1407, January 2015.
[25] Fujiwara, G. E. C. and Nguyen, N., "Adaptive Aeroelastic Wing Shape Optimization for High-Lift
Configurations," AIAA Aviation 2015 Conference, Dallas, TX, AIAA paper 2015-2420 , 2015.
[26] Ting, E., Reynolds, K.W., Nguyen, N. and Totah, J., "Aerodynamic Analysis of the Truss-Braced
Wing Aircraft Using Vortex-Lattice Superposition Approach", AIAA Aviation 2014 Conference, AIAA
paper 2014-2597, 2014.
[27] Aftosmis, M.J., Berger, M.J., and Melton, J.E., “Robust and Efficient Cartesian Mesh Generation for
Component- Based Geometry,” AIAA Journal, Vol. 36, No. 6, 1998, pp. 953-960.
[28] Denison, M., Housman, J.A., Ting, E., Nguyen, N., “Comparison of Viscous and Inviscid Loads in a
Static Aeroelastic Model of the Variable Camber Continuous Trailing Edge Flap Concept in the
Transonic Regime”, AIAA Aviation 2016 Conference, AIAA paper 2016-3571, 2016.
[29] Kiris, C., Barad, M., Housman, J., Sozer, E., Brehm, C., and Moni-Yeta, S., “The LAVA computational
fluid dynamics solver,” 52nd Aerospace Sciences Meeting, AIAA paper 2014-0070, 2014.

9.6 VLM vs DLM

The DLM can be regarded as an extension of the Vortex Lattice Method (VLM), which is used for
steady load calculation. The subsonic Doublet Lattice Method (DLM) has been the industry standard
method for the calculation of unsteady air loads for the past three decades. A constant-doublet panel
has exactly the same velocity field as a vortex-ring panel, so there is really no fundamental difference
between the two methods. There are some differences in implementation details, however. Most
documented VL methods place a horseshoe vortex at the ¼ chord location of each surface panel, so
that each panel has its own trailing vortices which extend into the wake. No special treatment of the
wake is then necessary. Putting the vortex at the ¼ chord location allows accurate results with very
few chord-wise panels. Just one or two chord-wise panels are often sufficient. Most DL methods put
a constant-doublet distribution on each surface AND wake panel. This is equivalent to putting a ring
vortex around the perimeter of each panel rather than the 1/4-chord location. The requires more
chord-wise panels to get the center of lift position roughly correct. On the other hand, DL is more
convenient for unsteady flows, where the wake doublet strengths are not constant and so the
horseshoe vortex representation is not very natural. For further consideration, please consult [L. H.
van Zyl]299.

299 L. H. van Zyl, ”Robustness of the subsonic doublet lattice method”, The Aeronautical Journal, 2003.
 250

9.7 Vortex Panel Method (VPM) vs. DLM

The Vortex Panel Method (VPM) is a very simple and computationally inexpensive method to solve
the incompressible/inviscid flow past airfoils. It was investigated by [Sulton et al.]300 and have shown
that it is capable of solving the flow past a thin airfoils for subsonic and laminar flows. For transonic
and turbulent flows, as the angle of attack increased, the method lacks the precision, specially near
leading and trailing edges. As a comparison, vortex lattice methods are:
➢ Similarities
• singularities are placed on a surface
• the non-penetration condition is satisfied at a number of control points
• a system of linear algebraic equations is solved to determine singularity strengths
➢ Differences
• Oriented toward lifting effects, and classical formulations ignore thickness
• Boundary conditions (BCs) are applied on a mean surface, not the actual surface (not
an exact solution of Laplace’s equation over a body, but embodies some additional
approximations, i.e., together with the first item, we find ΔCp, not Cp upper and Cp lower)
• Singularities are not distributed over the entire surface
• Oriented toward combinations of thin lifting surfaces (recall Panel methods had no
limitations on thickness)301.

300 Khalid M. Sultan, Anas M. Elshabli, Mohammed A. Kashbour, Seraj A. Ben. Ateiga, “Performance assessment
of the vortex panel method”, Libyan Journal of Science & Technology, (2019).
301 Aerodynamics of 3D Lifting Surfaces through Vortex Lattice Methods, 1998.
 251

10 Solution Methods for Inviscid (Euler) Equations

10.1 Background
The interest in Euler equations arises from the fact that in many primary design the information
about the pressure alone is needed. In boundary layer where the skin friction and heat transfer is
required, the outer edge condition using the Euler. The Euler equation is also of interest because of
interest in major flow internal discontinuities such as shock wave or contact surfaces. Solutions
relating to Rankine-Hugonist
Equations are embedded in Euler Subsonic Sonic Supersonic
Flow
equation. The Euler equations govern M<1 M=1 M>1
the motion of an Inviscid, Non-heat- Steady Elliptic Parabolic Hyperbolic
Conducting flow have different
Unsteady Hyperbolic Hyperbolic Hyperbolic
character in different regions. If the flow
is time-dependent, the flow regimes is
Table 10.1 Classification of the Euler Equation on
hyperbolic for all the Mach numbers and Different Regimes
solution can be obtained using marching
procedures. The situation is very different when a steady flow is assumed. In this case, Euler
equations are elliptic when the flow is subsonic, and hyperbolic when the flow is supersonic. For
transonic flows, has required research and development for many years.
Table 10.1 shows the different flow regimes and corresponding mathematical character of the
equations.

10.2 Method of Characteristics

For inviscid flows, method of Characteristics is the oldest and most nearly exact method which still
used to solve parabolic PDE’s. Other notable methods are Shock-Capturing which could be used to
capture large pressure gradients within a supersonic flow using a predictor-corrector scheme such
as Mac Cormack’s. The Euler equations for steady-state, isentropic, irrotational can be combined into
a single equation as potential equation. This could be treated using a type-dependent differentiation
where for supersonic regions where the flow is hyperbolic an upwind differential recommended. For
subsonic region, where the flow is elliptic, a central differencing recommended. Evidently, the
solution methods are dependent on several parameters confined to problem at hand. The description
for some legacy discretization methods and their characteristic behavior are available in literature.
Closed form solutions of non-linear hyperbolic partial differential equation do not exists for general
cases. In order to obtain the solution to such an equations we are required to resort to numerical
methods. The method of characteristics is the oldest and most nearly exact method in use to solve
hyperbolic PDEs. Even though this technique is been replaced by newer finite difference method. A
background in characteristic theory and its application is essential. The method of characteristics is
a technique which utilizes the known physical behavior of the solution in each point in the flow.
10.2.1 Linear Systems
Consider Steady Supersonic of Inviscid, Non-heat conducting of small perturbation for 2D perfect
gas302 as depicted in Figure 10.1.

D. Anderson, J., Tannehill, R., Pletcher,”Computational Fluid Mechanics and Heat Transfer”, ISBN 0-89116-
302

471-5 – 1984.
 252

 
(1 − M  ) xx +  yy = 0 denoting (1 − M  ) = β 2 and u = ,v=
x y
u v v u w w
β2 − =0 , − =0 writ e in vector form + [ A] =0
x y x y x y
 1
u   0 − 2
where w =   and [ A] = β
v  
− 1 0 
Eq. 10.1
The eigenvalues of this system are the eigenvalues of [A]. These are obtained by extracting the roots
of characteristics equation of [A] as

1
−λ −
[ A] − λ[I ] = 0 or β2 = 0 ,
−1 −λ Eq.
1 1 1
λ2 − = 0 , λ1 = , λ2 = −
β 2
β β
10.2

This is pair of roots from the differential equation

of characteristics. Next we determine the
compatibility equation. These equations are
obtained by pre-multiplying the system of
equations by left eigenvectors of [A]. This Figure 10.1 Characteristics of Linear Equation
effectively provides a method for writing the
equations along the characteristics. Let L1
represents the left eigenvectors of [A]
corresponding to λ1 and L2 represents the left eigenvectors corresponding to λ2. Drive the
eigenvectors of [A]:

T
[Li ] [𝐀 − λi 𝐈] = 0
1 1
− − 2
l β β −β β
L1 = [ 1 ] → ⏟
[l1 l2 ] =0 L1 = [ ], L2 = [ ]
l2 1 1 1
LT −1 −
[
⏟ β]
𝐀
Eq. 10.3
The compatibility equations along λ1 is obtained from
 253

L  w    w 
T
T
i
x + [A]w y = 0 or Li x + λi w y = 0
 1 
u x + β u y 
compatabilty along λ1 is obtained [-β 1]  =0 Eq. 10.4
vx + 1 vy 
 β 

(βu − v ) + 1  (βu − v ) = 0 in similar manar  (βu + v ) − 1  (βu + v )
x β y x β y

It is expressed the fact that quantity (βu-v) is constant along λ1, and (βu+v) is constant along λ2. The
quantities are called Riemann Invariants. Since these two quantities are constant and opposite pair
of characteristics, it is easy to determine u and v at a point. If at a point we know (βu-v) and (βu+v),
we can immediately compute both u and v.
10.2.2 Non-Linear Systems
The development presented so far is for a system linear equations for simplicity. In more complex
nonlinear settings, the results are not as easily obtained. In the general case, the characteristics slopes
are not constant and vary with fluid properties303 shown in Figure 10.2. For a general nonlinear
problem, the characteristics equation must be integrated numerically to obtain a complete flow field
solutions. Consider a 2D supersonics flow of a perfect gas over a flat surface. The Euler equation
governing this inviscid flow as a matrix form

w w
+ [ A] =0
x y
 v 
 uv − a2 − 0 
u  
p

v u − a2
( )
2
1  0 
v 2
u − a2
where w =  
0
and [A] = 2  u ρu 
p u − a2  
  − ρva ρua 2
2
uv 0 
e   − ρv

ρu
u
v
u
(
v 2
u −a 2 

)
Eq. 10.5
The eigenvalues of [A] determine the characteristics direction and are304

v v uv + a u 2 + v 2 − a 2 uv − a u 2 + v 2 − a 2
λ1 = , λ 2 = , λ 3 = , λ4 = Eq. 10.6
u u u2 − a2 u2 − a2
The matrix of left eigenvectors associated with these values of λ may be written as

See previous.
303

D. Anderson, J., Tannehill, R., Pletcher,”Computational Fluid Mechanics and Heat Transfer”, ISBN 0-89116-
304

471-5 – 1984.
 254

 ρu ρv 
 0 1
a2 a2
 ρu ρv 1 0
−1
 
[ T] =  − 1
+
u 1 1
0 Eq.
 u 2 + v2 − a 2 v u 2 + v2 − a 2 ρva 
 1 u 1 1 
 −v 0
 u + v − a
2 2 2 v u 2 + v2 − a 2 ρva 
10.7
We obtain the compatibility relations by pre-multiplying the original system by [T]-1. These relations
along the wave fronts are given by:

du dv β dp dy
−v +u + = 0 along = λ3
ds 3 ds 3 ρ ds 3 dx
Eq. 10.8
du dv β dp dy
v −u + = 0 along = λ4
ds 4 ds 4 ρ ds 4 dx

These are an ordinary differential equations which holds along the characteristic with slope λ3, λ4,
while arc length along this characteristics is denoted by s3, s4. In contrast to linear example, the
analytical solution for characteristics is not known for the general nonlinear problem. It is clear that
we must numerically integrate to
determine the shape of the
characteristics in step by step manner.
Consider the characteristic defined by
λ3. Stating at an initial data surface, the
expression can be integrated to obtain
the coordinates of next point at the
curve. At the same time, the
differentials equation defining the
other wave front characteristics can be
integrated. For a simple first-order
integration this provide us with two
equations for wave front
characteristics. From this expressions,
we determine the coordinate of their
intersection, point A. Once the point A
is known, the compatibility relations,
are integrated along the characteristics
to this point. This provide a system of Figure 10.2 Characteristics of Nonlinear solution point
equations at point A. This is a first- (exaggerated)
order estimate of the both the location
of point A and the associated flow variables. In the next step, the new intersection point B can be
calculated which now includes the nonlinear nature of the characteristic curve. In a similar manner,
the dependent variables at point B are computed. Since the problem is nonlinear, the final
intersection point B does not necessary appear at the same value of x for all solution points.
Consequently, the solution is usually interpolated onto an x-constant surface before the next
 255

integration step. This requires additional logic and added considerably to the difficulty in turning an
accurate solution305

10.3 Shock Capturing Method

10.3.1 Explicit McCormack Method
The shock capturing schemes are most used techniques for computing inviscid flows with shocks. In
this approach, the Euler’s equations are cast in conservative form for and shocks wave predicted as
part of solution. The shock waves predicted by these methods are smeared over several grid intervals
bur the simplicity of the approach may outweigh the slight compromised in the results. First we
consider 2D steady supersonic flow. We assume the x-axis forms the body surface and marching
direction. The equations can be written as

 ρu   ρv 
E F
+ = 0 where E = p + ρu 2  and F =  ρuv  Eq. 10.9
x y
 ρuv  p + ρv 2 

This is a hyperbolic PDE and explicit Mac Cormack scheme would be a good choice as

̅̅̅̅̅̅ ∆x n 1 ̅̅̅̅̅̅ ∆x ̅̅̅̅̅̅ ̅̅̅̅̅̅

𝐄in+1 = 𝐄in − (𝐅i+1 − 𝐅in ) → 𝐄in+1 = [𝐄in + 𝐄in+1 − (𝐅in+1 − 𝐅i−1
n+1
)]
⏟ ∆y 2
⏟ ∆y
predictor corrector
Eq. 10.10
At the end of predictor/corrector step, E must be decoded to obtain the primitive variables. This way,
the flux vector can be evaluated for the next integration step. The y-components of velocity v is
immediately known

2
E E γ E2  γ E2  γ −1
v= 3 , ρ= 1 , u=    − (2H − v 2 ) , p = E 2 − ρu 2
E1 u γ + 1 E1  γ + 1 E1  γ + 1
Eq. 10.11

Having completed this process, F can be

recalculate and the next step can be
implemented. Readers also should
consult the work by [Mahdi and Al-
Kwarizmi]306 for shock capturing on a
2D compressible unsteady Euler
equation. Using a Mac Cormack's time
marching method that an explicit finite-
difference technique. The test case Figure 10.3 Supersonic Flow Over Circular Arc with Inlet
chosen is that of a transonic and M=1.4 - Courtesy of [Mahdi and Al-Kwarizmi]

305 D. Anderson, J., Tannehill, R., Pletcher,”Computational Fluid Mechanics and Heat Transfer”, ISBN 0-89116-
471-5 – 1984.
306 Ahmed Fouad Mahdi, Al-Kwarizmi, “Shock Wave Capturing Numerically in Two-Dimensional Supersonic Wind

Tunnel for Different Configurations”, Eng. & Tech. Journal, Vol. 30, No.3 , 2012.
 256

supersonic flow through a channel with a circular arc bump on the lower wall, half wedge and
extended compression corner.
10.3.2 1D Upwind Flux-Splitting Scheme (Steger-Warming)
This is usually used as a Shock Capturing Method and belongs to a class of solutions as AUSM
(Advection Upstream Splitting Method). It is developed as a numerical inviscid flux function for
solving a general system of conservation equations. It is based on the upwind concept and was
motivated to provide an alternative approach to other upwind methods, such as the Godunov
method, flux difference splitting methods by [Roe]307, [Solomon and Osher], flux vector splitting
methods by [Van Leer]308, and [Steger and Warming. The AUSM first recognizes that the inviscid flux
consist of two physically distinct parts, i.e., convective and pressure fluxes. The former is associated
with the flow (advection) speed, while the latter with the acoustic speed; or respectively classified as
the linear and nonlinear fields. Currently, the convective and pressure fluxes are formulated using
the eigenvalues of the flux Jacobian matrices. To illustrates the flux splitting concepts, consider
the one-dimensional system of hyperbolic PDE,

U E  U  E+  E−
+ = 0 define E = E + + E − so + + =0 Eq. 10.12
t x t x x

Where the plus indicates a backward differencing, a minus forward differencing is required. The split
flux can be used either for explicit or implicit algorithms. For example, a second-order upwind,
predictor/corrector scheme (Beam & Warming, 1975) used as

Predicter U nj +1 = U nj −
Δt
Δx
(
 E +j + Δ E −j )
Correcter U nj +1
1 n
= U j + U nj +1 −
2
Δt
Δx
( )
 2 E +j n +  E nj +1 +
Δx
(
Δt 2 − n
)

Δ E j − ΔE −j n +1 

Eq. 10.13
An implicit algorithm using the trapezoidal rule is derived


I +
Δt
2x
([A j ]+ + Δ[A j ]− )ΔU nj = −
 Δt
Δx
E+ + ΔE−  
 
Eq. 10.14
This algorithm is 1st order accurate in space, 2nd order accurate in time. The left hand side can be AF
and ignoring the 2nd order cross terms


 I +
Δt


[A j ]+   I +
Δt


Δ[A j ]− ΔU nj = −
Δt
Δx
E + + ΔE −  
 2 x  2 x 

 I +
Δt
2x

[A j ]+ ΔUj = −
Δt
Δx
E + + ΔE − 
 
 Δt 
 I + Δ[A j ]− ΔU nj = ΔUj
 2x 
Eq. 10.15

307 Roe, Flux Differencing Scheme Details.

308 Bram van Leer, “Flux-Vector Splitting for the 1990s”, The University of Michigan, N91-21073.
 257

In this equations, each one dimensional sweep required the solution of two block bi-diagonal (here
1) system. Usually the advantages of AP is more pronounced in multi-dimension problem. The use of
split flux techniques for shock capturing applications produces somewhat better results than
standard central difference schemes.
10.3.3 Total Variation Diminishing (TVD) as other Upwind Schemes
In numerical methods, Total Variation Diminishing (TVD) is a property of certain discretization
schemes used to solve hyperbolic partial differential equations. To capture the variation fine grids
(∆x = very small) are needed and the computation becomes heavy and therefore un-economic. The
use of coarse grids with central difference scheme, upwind scheme, hybrid difference scheme, and
power law scheme gives false shock predictions. TVD scheme enables sharper shock predictions on
coarse grids saving computation time and as the scheme preserves monotonicity there are no
spurious oscillations in the solution. In systems described by partial differential equations, such as
the following 1D hyperbolic advection equation,

U U E
+ [ A] where [ A] =
t x U
Eq. 10.16
U
TV = 
x
dx =  U j+1 − U j subject to TV U ( n +1
)  TV (U )
n

Conventional shock capturing schemes for the solution of nonlinear hyperbolic conservation laws
are linear and L2-norm stable when considered in the constant coefficient case309. There are three
major difficulties in using such schemes to compute discontinuous solutions of a nonlinear system,
such as the compressible Euler equations:
1. Schemes that are second (or higher) order accurate may produce oscillations wherever the
solution is not smooth.
2. Nonlinear instabilities may develop in spite of the &-stability in the constant coefficient case.
3. The scheme may select a nonphysical solution.
It is well known that monotone conservative difference schemes always converge and that their limit
is the physical weak solution satisfying an entropy inequality. Thus monotone schemes are
guaranteed not to have difficulties (2) and (3). However, monotone schemes are only first-order
accurate. Consequently, they produce rather crude approximations whenever the solution varies
strongly in space or time. When using a second- (or higher) order accurate scheme, some of these
difficulties can be overcome by adding a hefty amount of numerical dissipation to the scheme.
Unfortunately, this process brings about an irretrievable loss of information that exhibits itself in
degraded accuracy and smeared discontinuities. Thus, a typical complaint about conventional
schemes which are developed under the guidelines of linear theory is that they are not robust and/or
not accurate enough. The class of TVD schemes contains monotone schemes, but is significantly
larger as it includes second-order accurate schemes.
10.3.3.1 The Upwind Connection to Explicit Artificial Dissipation
If the viscosity isn’t large enough, velocity oscillations about the correct mean velocity are observed
to develop behind a shock. These oscillations can be interpreted as a macroscopic version of heat
energy, i.e., fluctuating kinetic energy in place of fluctuating molecular energy310. These upwind
schemes all claim (with good justification) to be physically consistent since they follow in some sense

309 H. C. Yee, NASA, “Implicit Total Variation Diminishing (TVD) schemes for steady-state calculations”, Journal of
Computational Physics, March 1985.
310 Flow Science Blog, “What are Artificial and Numerical Viscosities?”
 258

the characteristics of
the flow. They in
general can be shown
to produce sharp
oscillation free shocks
without added
artificial dissipation.
Figure 10.4 shows
the coefficient of
pressure for a shock
without artificial
dissipation (left);
with artificial Figure 10.4 Coefficient of Pressure for a Shock
dissipation (right).
Also these schemes have an inherent amount of internal dissipation, due to the one sided differences,
which cannot be modified or decreased. It may be advantageous to have the flexibility of a simple
central difference scheme with a controllable amount of artificial dissipation. It can be shown that
the upwind schemes have an equivalence to central difference schemes with added dissipation. The
central schemes are much simpler and more flexible and are therefore desirable if the dissipation can
be added in an analogous fashion to the upwind schemes311. This is often called the Implicit Artificial
Dissipation as opposed to Explicit one which purposely added to a difference equation and, been
discussed previously. In general, the flux of any scheme can be written as

⏟f + f − fFTCS
⏟ FTCS = Foward Time & Central Space
FTCS flux Flux due to Artificial Viscosity
Eq. 10.17
10.3.4 3D Unsteady Euler Equation Solutions Using Flux Vector Splitting
To compute the flow of a store released from an aircraft, it is desirable to solve the unsteady Euler
equations on a grid that moves with the store along its trajectory312. The objective here is to solve the
3D unsteady Euler equations on a time-dependent grid. The computations presented here are for
bodies whose motion is prescribed. As the solution advances in time, body motion could be
determined from the Euler equations by using force and moment coefficients obtained from the Euler
solution in the dynamic equations of motion for the body to determine the trajectory of the body. The
objective of this report is to present and verify dynamic-grid Euler equations computations.
Following the development in313, the strong conservation law of the Euler equations in curvilinear
coordinates can be written as:

311 T. H. Pulliam, “Solution Methods in Computational Fluid Dynamics”, NASA Ames Research Center, USA.
312 David, L., Whitfield, “Three-Dimensional Unsteady Euler Equation Solutions Using Flux Vector Splitting”,
NASA-CR-173254 19840008789.
313 See Previous.
 259

Q F G H
+ + + =0 (7.18)
τ ξ η ζ
ρ  ρU   ρV   ρW 
 ρu   ρuU + ξ p   ρuV + η p   ρuW + ζ p 
   x   x   x 
Q = J  ρv  , F = J  ρvU + ξ y p  , G = J  ρvV + η y p  , H = J  ρvW + ζ y p 
       
ρw   ρwU + ξ z p   ρwV + ηz p   ρwW + ζ z p 
 ρe   U(e + p) − ξ t p V(e + p) − η t p  W(e + p) − ζ t p
 ( , ,  )
and ξ = ξ (x, y, z) , η = η (x, y, z) , ζ (x, y, z) , t =  and J =
 ( x, y , z )
Matrices of transformation are :
ξ x = J (yη z ζ − y ζ z η ) , ξ y = J (x η z ζ − x ζ z η ) , ξ z = J (x η y ζ − x ζ y η )
η x = J (yξ z ζ − y ζ z ξ ) , η y = J (x ξ z ζ − x ζ z ξ ) , ηz = J (x ξ y ζ − x ζ x ξ )
 x = J (yξ z − y z ξ ) ,  y = J (x ξ z − x z ξ ) ,  z = J (x ξ y − x y ξ )
with contvariant velocity components
U = ξ x u + ξ y v + ξ z w , V = ηx u + η y v + ηz w , W = ζ x u + ζ y v + ζ z w
Eq. 10.18
The 3D unsteady Euler equations,
Eq. 10.18, are a hyperbolic system of five equations and hence have five characteristic velocities in
each of the three spatial directions. These characteristic velocities are determined from the
quasilinear form of
Eq. 10.18. A finite volume discretization of
Eq. 10.18 balances the increase of the conserved quantity in a computational cell, or volume, with
the flux of the quantity through the surface of the cell. Assuming the dependent variables are
constant in the interior of cell i, j, k, and that the flux vectors F, G, and H are constant over the constant
ξ , constant η and constant surfaces of the ζ , respectively, an explicit discretization yields:

n+1 n n n
⏟ i,j,k − Q i,j,k ) ∆ξ∆η∆ζ = − (Fi+1,j,k − Fi−1.j.k ) ∆η∆ζ
(Q
2 2
∆Q
∆τ

− (Gn 1 − Gn 1 ) ∆ξ∆ζ − (H n 1 − Hn 1 ) ∆ξ∆η

i,j+ ,k i,j− ,k i,j,k+ i,j,k−
2 2 2 2
Eq. 10.19
and time step Δτ is given in314, or using the classical Runge-Kutta scheme. The central difference
operator is been used in Eq. 10.19 indicates the flux vectors are evaluated at cell faces in this finite
volume formulation. The numerical scheme used is a finite volume version of the second-order
upwind scheme of [Beam & Warming]315, while the present scheme is an extension of that used by

314 David, L., Whitfield, “Three-Dimensional Unsteady Euler Equation Solutions Using Flux Vector Splitting”,
NASA-CR-173254 -19840008789.
315 Warming, R. F. , and Beam, K. M., "Upwind Second·-Order Difference Schemes .md Applications in Aerodynamic

Flows," AIAA ,Journal, Vol. 14. No.9, September 1976, pp. 1241-1249.
 260

Deese316.
10.3.4.1 Flux Splitting
Hyperbolic partial differential equations, such as the Euler equations, are characterized by the
existence of a limited domain of dependence. The solution at a point does not depend on every
other point in the field; this means that information travels only in certain characteristic directions.
Numerical schemes intended to solve hyperbolic equations are usually enhanced by insuring that
the numerical method propagates information in the direction specified by the partial differential
equation. This can be done by using an upwind method, or one in which the difference operator is
taken in the direction from which the information should come. Stability properties are often
improved by unwinding, and it is usually unnecessary to add smoothing terms or artificial viscosity
to an upwind method. The 3D Euler equations,
Eq. 10.18, are a hyperbolic system of five equations and hence have five characteristic velocities in
each of the three spatial directions. These characteristic velocities are determined from the
quasilinear form of
Eq. 10.18. Details of splitting and test cases can be obtained from317. It is suffice to show that five
eigenvalues are:

λ1k = λ2k = λ2k = ⏟

ku + kv + kz , λ4k = θk + c|∇k| , λ5k = θk − c|∇k|
θk
1/2
where k = {ξ, η, ζ} and |∇k| = (k 2x + k 2y + k 2z )
Eq. 10.20
And c is speed of sound. As evident, two of the five right eigenvectors form a 2D subspace, within the
general 5-dimensional space spanned by all right eigenvectors, and that every member of this
subspace is itself an eigenvector. This phenomenon was attributed to the fact that their
corresponding eigenvalues are repeated, which creates a “symmetry” within the eigenvector space.
Although it may seem difficult to visualize any symmetry within a 5-dimensional vector space, part
of this symmetry reveals itself when we geometrically interpret the 2-dimensional subspace as a
plane. The Euler equations contain three types, or families of waves, one for every distinct eigenvalue.
Each family of waves carries a different signal. The waves traveling at the speed of the flow (λk1,2,3 )
are called entropy waves, their signal being entropy, whereas waves traveling at the speed of sound
relative to the flow ( λk4,5 ) are called acoustic waves. Unfortunately, the signal carried by acoustic
waves is not quantifiable in simple thermodynamic terms, but let us just say that they carry acoustic
information318.

10.4 The Riemann Problem

The Riemann problem consists of a conservation law together with piecewise constant data having a
single discontinuity (i.e., shocks)[Kong]319. Here we will discuss the problem for a linear system, and
then discuss how the Riemann problem for the Euler equations, addressing specific problems that
will be focused on during comparisons of schemes. The initial state of the system is defined as

316 Deese, J. E, "NumericalExperi1D:ents with the Split-Flux··Vector Form of the Euler Equations," AIAA Paper No.
83-0122, January 1983.
317 David, L., Whitfield, “Three-Dimensional Unsteady Euler Equation Solutions Using Flux Vector Splitting”,

NASA-CR-173254 -19840008789.
318 Axel Rohde, ”Eigenvalues And Eigenvectors Of The Euler Equations In General Geometries”, AIAA 2001-2609.
319 Charlotte Kong, “Comparison of Approximate Riemann Solvers”, A dissertation submitted in partial fulfilment

of the requirement for the degree of Master of Science in Mathematical and Numerical Modelling of the
Atmosphere and Oceans, 2011.
 261

uL for x ≤ 0
u(x, t = 0) = {
uR for x ≥ 0
Eq. 10.21
To put into perspective, the initial state is constant for all negative x, and constant for all positive x,
but differs between left and right. In the one-dimensional case we can consider this problem as a gas
with one temperature and density located to the left of a removable wall and another gas with
another temperature and density to the right of the wall. At time t = 0 the wall is instantly removed
and the results are observed. In numerical analysis Riemann problems appear in a natural way in
finite volume methods for the solution of conservation law equations due to the discreteness of the
grid. For this we use approximate Riemann solvers, since iterative schemes are too costly some
assumptions must be made, which will be discussed further in the following section.
10.4.1 Roe Approximate Riemann Solver
The Roe solver, devised by [Roe]320, is an approximate Riemann solver based around the Godunov
scheme and works by looking for an estimate for the inter cell numerical flux or Godunov flux at the
interface between two computational cells. It essentially determine the approximate solution by
solving a constant coefficient linear system instead of the original nonlinear system. We study
the break-up of a single discontinuity. Those cases where F(u) is linear are well-known and
essentially trivial. Those cases where u is scalar and F is non-linear can be surprisingly complicated,
but have been thoroughly investigated. If u is a vector and F is non-linear, then the problem involves
non-linear algebraic equations together with, usually, logical conditions which express the fact that
a given member of the wave system may be present either as a shockwave or as an expansion fan. In
general, the most efficient way to solve these equations will depend on the system of conservation
laws from which they derive; ingenuity is required to exploit special features of each individual
system. The usual way of incorporating the Riemann problem into the numerical solution is to take
(un, un+i), for each i in turn, as pairs of states defining a sequence of Riemann problems, which are
then thought of as providing information about the solution within each interval (i , i+1). Various
individual methods are then distinguished by the way in which this information is put to use. Here,
we consider approximate solutions which are exact solutions to an approximate problem:

u t + Au x = 0  ut + Aˆ u = 0 , A = F
u
x

whe re A ˆ = 1 ( A + A ) = A 1 (u + u )
L R L R
2 2
1 ~ ~p ~ p  ~p
f Roe = F(u L ) + F(u R ) -  λ p α R  where α ~p = L (u R - u L ) Eq. 10.22
2 p 
~ p ~ p ~p ˆ
and λ , R , L being the eigenvalue s and the right and left eigenvectors of A
and p = number of equations

For further information, reader should consult the [Roe]321.

320 P. L. Roe, “Approximate Riemann Solvers, Parameter Vectors, and Difference Schemes”, Journal of
Computational Physics 43, 357-372 (1981).
321 P. L. Roe, “Approximate Riemann Solvers, Parameter Vectors, and Difference Schemes”, Journal of

Computational Physics 43, 357-372 (1981).

 262

10.4.2 Case Study – Further Discussion Regarding the Shock Capturing Techniques to Evaluating the
Discourteous Solution
Author : Joseph J. S. Shang
Affiliation : Wright State University, 3640 Colonel Glenn Highway, Dayton, OH 45435-0001, USA
Title of Paper : Landmarks and new frontiers of computational fluid dynamics
Original Appearance : Advances in Aerodynamics (Springer Nature)
Source : https://doi.org/10.1186/s42774-019-0003-x
A major pacing item for CFD adopting shock capturing techniques is evaluating discontinuous
solution generated by shock waves and slip streams from shock interceptions. A breakthrough by
Godunov demonstrates a multi-dimensional flow field that contains shock waves and contact
surfaces can still be analyzed [16]. The discontinuities of the hyperbolic differential systems are
treated as a piecewise continuous data distribution within a control volume and to be solved across
the singular point as the Riemann problem.
The underpinning principle is the monotonicity preserving property of the hyperbolic difference
equation; namely, temporal increment/decrement of dependent variable is monotonic. Based on this
property, Harten originates the total variation diminishing (TVD) scheme and spans off a huge
amount of research on TVD schemes and a variety of flux limiters for analyzing piecewise
discontinuous solutions for CFD [23].
From the physics viewpoint, the treatment of shock jump by flux splitting can be easily understood
through the concept of zone of dependence for supersonic flows. By solving a set of Riemann
problems over the entire computational domain according to their distinctive characteristics, this
approach actually honors the physics of domain of influence. The directional wave propagation is
constructed according to the phase velocity from the permissible database. In an outstanding work
by Steger and Warming, it has shown a systematic relationship of the real eigenvalue and eigenvector
for the split flux formulation. They also demonstrate the Euler equations, together with the equation
of state for gas, possessing the homogeneous function of degree one property [24].
The solving procedure for the split equation is by applying one-side differencing approximation to
achieve the approximate Riemann problem. The basic issue is that the split inviscid flux components
are not differentiable at singular sonic points. This behavior is also the peculiarity of the approximate
governing equations. The continuous viscous terms are solving simultaneously by spatially central
scheme. An incisive summary for using the approximated Riemann approximations can be found
from the work of Roe [25].
In spite of the rational treatments of discontinuous numerical solution, the undesirable artifact of
numerical oscillation or the Gibbs phenomenon is always presented around a singular point. A series
of excellent algorithms for maintaining computational stability and yielding sharp definition of a
piecewise continuous numerical solution are the ENO and WENO (weighted ENO) schemes [26, 27].
The fifth-order WENO scheme is supported by an overall stencil of five points; the smoothness of a
solution is measured by the sum of normalized squares of the scaled L2 norm for derivatives from the
lower-order polynomials. The conditioned information is incorporated into the weights definition to
improve the convergence at the critical points. It is revealed that the enhancement of the fifth-order
scheme is derived from a large weight assigned to the discontinuous stencils, but not from their
superior order of convergence at critical points.
From the lessons learned, the desired feature of a numerical scheme may be better selected from the
optimization in the Fourier space rather than by focusing on the lowest possible truncation error.
Along this line of reasoning, compact-difference scheme becomes a viable method to achieve high
resolution. The basic algorithm is an implicit procedure for evaluating derivatives, which has a small
stencil dimension and yet can maintain a lower level of dispersive and dissipative errors than the
conventional numerical schemes. The basic formulation of compact-difference approximation is
 263

derived from the Hermite’s generalization of the Taylor series [28]. The compact differencing
formulations for evaluating the first-order and second-order derivative have been given by Lele [29].
The formulas are three-point spatially central scheme and require attention on boundaries. Since
the scheme is inapplicable on the immediately adjacent grid point next to a boundary, a transitional
operator between the boundary and the interior domain is required. The transitional boundary
scheme is not only required to transmit data from the boundary to the interior domain but also must
preserve the stability and accuracy for the global solution.
Although the high-resolution scheme is stable in the classic sense but the transition operator is one
of the sources that contributes to spurious high-frequency oscillations known as time-delay
instability. The time-instability is incurred by positive real eigenvalue components which dominate
the numerical result.
A very effective remedy to control the
time-delay instability has been
demonstrated by using a low-pass filter
[30]. The spectral function is a
symmetric numerical filter that contains
no imaginary part and has the low-pass
amplitude response. In other words, the
low-pass filter modifies only the
amplitude but not the phase relation
among all Fourier components.
In Figure 10.5, the accuracy and
effectiveness of the 4th and 6th order
compact-difference scheme
with/without including some fine tuning
is depicting to reduce significantly
dispersive and isotropic errors in
comparison with conventional Figure 10.5 Comparison of dispersion error of compact-
numerical methods such as the 2nd difference scheme with others methods
order MUSCL and the simple 4th order
explicit schemes. The detailed
comparison on dispersive error is display in the wave numbers range of π/4 < w < 2π.
The superior behavior of compact-difference schemes are noted, but the fine tuning may be
counterproductive [31]. The reduction for isotropic error for multi-dimensional computation by
compact-difference schemes is displaying in the wave numbers range from π/8 to 3π/4 in four
quadrants.
In most CD simulations, a high resolution solution is frequently needed in the high gradient regions
like shock jump, media interface, and flame front. A local grid refinement approach appears to be
very attractive; especially the grid refinement is independent from the global mesh system. The
spectral-like polynomial grid refinement method introduced by Korpriva that meets the
requirement, and the high resolution result is derived from the Gauss quadrature [32]. By this
approach, there is no need to reconstruct the overall grid system, but by just increasing the degrees
of polynomials within the refined grid block to capture the fine-structure features. The local grid
refinement numerical procedure is equally applicable to the temporal advancement of a time-
dependent problem. In some cases, it may even be possible to examine a problem that has
discontinuity between the integral intervals by relegating the singularity to the weighting function.
A unique behavior of the recursive formula for derivative computation is that the result depends on
all discretized points or the roots of the polynomial within the refined grid block. In fact, all high-
resolution schemes striving to achieve a spectral-like accuracy are by employing all discretized
points in an array to mimic the Gauss-quadrature formulation.
 264

The spectral-like polynomial refinement for the entire computational domain is by dividing it into
grid-refining blocks. The local high resolution solution is generated within each grid-refining block
by the unequal-spacing roots from any of the classic orthogonal polynomials (Legendre, Laguerre,
Chebyshev, Hermit, Gegenbauer, Jacobi, Meixner-Poluckzek). For discontinuity capturing, an artificial
dissipative term may not be necessary within the sub grid domain for suppressing numerical
oscillations, but just by reducing the grid-refining space dimension or by increasing the order of the
polynomial. The local grid refinement approach has also been extended for solving the conservation
laws on unstructured grids [33].
10.4.3 References
[16] Godunov SK (1959) Finite-difference method for numerical computational of discontinuous
solution of the equations of fluid dynamics. Mat Sb 47:271–306
[18] Peaceman DW, Rachford HH (1955) The numerical solution of parabolic and elliptic differential
equations. J Soc Ind Appl Mat 3:28–41
[19] Richtmyer RD, Morton KW (1967) Differential methods for initial-value problem, 2nd ed. Inter
science Publishers, Wiley, New York
[20] Shang JS (2009) Computational fluid dynamics application to aerospace science. Aeronaut J
113(1148)
[21] Brandt A (1973) Multi-level adaptive technique (MALT) for fast numerical solution to boundary
value problem, lecture notes in physics, vol 18. Springer-Verlag, Berlin, pp 82–89
[22] Delaunay, B., Sur la Sphere Vide, Bull. Acad. Science, USSR, VII, Class. Sci. Mat. Nat. 1934.
[23] Harten A (1983) High-resolution schemes for hyperbolic conservation Laws. J Comp. Phys
49:375–385
[24] Steger JL, Warming RF (1981) Flux vector splitting of the inviscid Gas dynamics equations with
application to finite difference methods. J Comp. Phys 40:263–293
[25] Roe PL (1981) Approximate Riemann solvers, parameter vectors and difference schemes. J
Comp. Phys 43:357–372
[26] Balsara D, Shu CW (2000) Monotonicity preserving weighted essentially non-oscillatory
schemes with increasingly high order of accuracy. J Comp Physics 160:405–452
[27] Shu CW, Osher S (1989) Implementation of essentially non oscillatory shock capture scheme II.
J Comp physics 83:32–78
[28] Collatz L (1966) The numerical treatment of differential equations. Springer-Verlag, New York.
[29] Lele SK (1992) Finite difference schemes with spectral-like resolution. J Comp Phys. 103:16–14
[30] Gaitonde D, Shang JS (1997) Optimized compact-difference-based finite-volume schemes for
linear wave phenomena. J Comp Physics 138:617–643
[31] Gaitonde DV, Shang JS, Young JL (1999) Practical aspects of higher-order numerical schemes for
wave propagation phenomena. Int J Num. Methods Eng. 45:1849–1869
[32] Korpriva D (1994) Multidomain spectral solution of compressible viscous flows. J Comp Physics
115:184–199
[33] Korpriva D (1996) A conservative staggered-grid Chebyshev multi-domain method for
compressible flows, II. A semi structure method. J Comp Physics 129:475–488
 265

10.5 Euler Equation With 2-Species322

To model two-fluid flows, the seven-equation model proposed by [Baer and Nunziato [25] is the most
complete model. In Baer and Nunziato's model, conservation equations of the mass, momentum and
energy are solved for each species and an additional transport equation is solved to describe the
topology of the fluid-fluid interface. However, their model is computationally very expensive and
arguably retains redundant information.
The simplest family of models to describe two-fluid flows is the four-equation model which consists
of equations of mass, momentum and energy for the mixture of fluids as a whole and one transport
equation. In order to suppress pressure oscillations across material interfaces, different quantities
were proposed for the transport equation in nonconservative advection form. Abgrall [26] and Shyue
[27] respectively suggested 1/(Υ - 1) or Y to be solved in the transport equation for pressure
equilibrium across material interfaces, where is the ratio of specific heats of the mixture and Y is the
mass fraction of one of the species. However, Abgrall's model has a technical problem that interfaces
cannot be described if both fluids have the same value of Υ and Shyue's model has a strong
assumption that molecular masses of the two fluids are the same. Worse still, neither of the models
conserves mass of each species discretely. Another family of reduced models that are able to conserve
the mass of each species and maintain pressure equilibrium at interfaces is the five-equation model.
The five-equation model proposed by [Allaire et al. [24] for two immiscible, inviscid, and non-
conducting fluids in the following form is used in present work:

∂Z1 ρ1 ∂
+ (Z ρ u ) = 0
∂t ∂xj 1 1 j
∂Z2 ρ2 ∂
+ (Z ρ u ) = 0
∂t ∂xj 2 2 j
∂ρui ∂
+ (ρui uj + pδij ) = 0
∂t ∂xj
∂E ∂
+ [u (E + p)] = 0
∂t ∂xj i
∂Z1 ∂Z1
+ uj =0
∂t ∂xj
Eq. 10.23
where ρ1 and ρ2 are the densities of fluids 1 and 2 respectively. ρ, ui, p, and E are the density, velocity
vector, pressure, and total energy per unit volume of the mixture respectively. Z1 is the volume
fraction of fluid 1. The volume fractions of the two fluids Z1 and Z2 are related by:

322Man Long Wong, Sanjiva K. Lele ,” High-Order Localized Dissipation Weighted Compact Nonlinear Scheme for
Shock- and Interface-Capturing in Compressible Flows”, Department of Aeronautics and Astronautics, Stanford
University, Stanford, USA, 2019.
 266

Z2 = 1 − Z1
Eq. 10.24
The ideal equation of state is used to close the system. By using the isobaric assumption, we are able
to derive an explicit mixture rule for the ratio of specific heats of the mixture:
1 Z1 Z2
= −
Υ − 1 Υ1 − 1 Υ2 − 1
Eq. 10.25
where Υ1 and Υ2 are the ratios of specific heats of fluids 1 and 2 respectively. In the absence of surface
tension, the isobaric assumption is consistent with pressure equilibrium across material interfaces.
The transport equation of volume fraction is solved in advection form. Following the approach
proposed by [ Johnsen et al. [21] and extended by [Coralic et al. [23], the following mathematically
equivalent form of the advection equation is used for the adaptation of a HLLC-type Riemann solver
to compute fluxes at midpoints between cell nodes:

∂Z1 ∂(Z1 uj ) ∂uj

+ = Z1
∂t ∂xj ∂xj
Eq. 10.26
10.5.1 Numerical Methods
In vector notation, Eq. 10.23, can be written as:

∂𝐐 ∂𝐅(𝐐) ∂𝐆(𝐐) ∂𝐇(𝐐)

+ + + = 𝐒(𝐐)
∂t ∂x ∂y ∂z
Eq. 10.27
Where Q, F, G, H, and S are the vectors of conservative variables, fluxes in the x, y, and z directions,
and sources, respectively. For two-species flow with five-equation model,

Z1 ρ1 u1 Z1 ρ1 v1 Z1 ρ1 w1 0
ρ Z2 ρ2 u1 Z2 ρ2 v2 Z2 ρ2 w2 0
ρu 2
ρu + p ρuv ρuw 0
𝐐 = ρv 𝐅 = ρv 𝐆 = ρv 2 + p 𝐇= ρwv 𝐒= 0
ρw ρw ρvw ρw 2 + p 0
[E] u(E + p) v(E + p) w(E + p) 0
[ Z1 u ] [ Z1 v ] [ Z1 w ] [Z1 ∇. u]
Eq. 10.28
u, v, and w are the components of velocity u in the x, y, and z directions respectively. For details on
solution scheme, please refer to [Wonga &Lele ]323.
10.5.2 References
[21] E. Johnsen, T. Colonius, Implementation of WENO schemes in compressible multicomponent
flow problems, Journal of Computational Physics 219 (2) (2006).

323Man Long Wong, Sanjiva K. Lele ,” High-Order Localized Dissipation Weighted Compact Nonlinear Scheme for
Shock- and Interface-Capturing in Compressible Flows”, Department of Aeronautics and Astronautics, Stanford
University, Stanford, USA, 2019.
 267

[22] T. Nonomura, S. Morizawa, H. Terashima, S. Obayashi, K. Fujii, Numerical (error) issues on

compressible multicomponent flows using a high-order differencing scheme: Weighted compact
nonlinear scheme, Journal of Computational Physics 231 (8) (2012).
[23] V. Coralic, T. Colonius, Finite-volume WENO scheme for viscous compressible multicomponent
flows, Journal of Computational Physics 274 (2014).
[24] G. Allaire, S. Clerc, S. Kokh, A five-equation model for the simulation of interfaces between
compressible fluids, Journal of Computational Physics 181 (2) (2002).
[25] M. Baer, J. Nunziato, A two-phase mixture theory for the deflagration-to-detonation transition
(DDT) in reactive granular materials, International Journal of Multiphase Flow 12 (6) (1986).
[26] R. Abgrall, How to prevent pressure oscillations in multicomponent ow calculations: a quasi
conservative approach, Journal of Computational Physics 125 (1) (1996).
[27] K.-M. Shyue, An effcient shock-capturing algorithm for compressible multicomponent problems,
Journal of Computational Physics 142 (1) (1998).
[28] T. Nonomura, K. Fujii, Robust explicit formulation of weighted compact nonlinear scheme,
Computers & Fluids 85 (2013).
[29] E. M. Taylor, M. Wu, M. P. Martin, Optimization of nonlinear error for weighted essentially non
oscillatory methods in direct numerical simulations of compressible turbulence, Journal of
Computational Physics 223 (1) (2007).
 268

11 Solution Methods for Viscous (Navier-Stokes) Equations

The Navier-Stokes equation are very difficult to solve in their complete form. In general, a very large
amount of computer resources is needed to obtain a solution. This is particularly true for the
compressible N-S equations which are a mixed set of elliptic-parabolic equations, while the unsteady
incompressible N-S equations are a mixed set of hyperbolic-parabolic equations. As a consequence,
different numerical techniques must be used to solve them. The time dependent solution is normally
used when a steady-state flow is computed. That is, the unsteady N-S solutions are integrated in time
until a steady-state solution is achieved. Thus, for a three-dimensional flow field a four-dimensional
(3 space +1 time) problem must be solved. Besides being very CPU intensive, it needs a very large
amount of storage. That is why whenever possible, the complete compressible N-S equation should
be avoided. The compressible N-S and energy equations in Cartesian coordinates without body
force or external heat addition, in vector notation can be written as:
U E F G
+ + + =0 where U, E, F, and G are given by
t x y z
ρ  ρu 
 ρu   ρu + p − τ xx
2 
   
U =  ρv  , E= ρuv − τ xy 
   
ρw   ρuw + p − τ xz 
 E t  (E t + p)u − uτ xx − vτ xy − wτ xz + q x 
 
 ρv   ρw 
 ρuv − τ xy   ρuw − τ xz 
   
F= ρv 2 + p − τ yy  , G= ρvw − τ yz 
   
 ρvw − τ yz   ρw + p − τ zz
2

(E t + p)v − uτ xy − vτ yy − wτ yz + q y  (E t + p)w − uτ xz − vτ yz − wτ zz + q z 
   
where
2  u v w  2  v u w  2  w u v 
τ xx = μ 2 − −  , τ yy = μ 2 − −  , τ zz = μ 2 − − 
3  x y z  3  y x z  3  z x y 
 u v   w u   v w 
τ xy = τ yx = μ +  , τ xz = τ zx = μ +  , τ yz = τ zy = μ + 
 y x   x z   z y 
T T T  u 2 + v2 + w 2 
q x = −k , q y = −k , q z = −k and E t = ρ e + 
x y z  2 

Eq. 11.1
By setting density to constant, we obtain the incompressible N-S equation for Newtonian flow as:
 269

.V = 0
u gx
V  
+ (V. )V + p = ν 2 V + g where V = v and g = g y
t w g
  z
Eq. 11.2

11.1 General Transformation

These equations can be expressed in terms of a generalized non-orthogonal curvilinear coordinated
system

ξ = ξ (x, y, z) , η = η (x, y, z) , ζ (x, y, z) , t = t

x x x
 ( , ,  ) 1 1
J= = = = 1/ y y y
 ( x, y, z ) J −1  (x, y, z)
z z z
 ( , ,  )
where matrices are :
ξ x = J (y η z ζ − y ζ z η ) , ξ y = J (x η z ζ − x ζ z η ) , ξ z = J (x η y ζ − x ζ y η )
η x = J (yξ z ζ − y ζ z ξ ) , η y = J (x ξ z ζ − x ζ z ξ ) , η z = J (x ξ y ζ − x ζ x ξ )
 x = J (yξ z − y z ξ ) ,  y = J (x ξ z − x z ξ ) ,  z = J (x ξ y − x y ξ )
Eq. 11.3
The generalized transformation to the compressible N-S equations (
Eq. 11.1) written in vector form as324

𝐔𝐭 + ξx 𝐄𝛏 + ηx 𝐄𝛈 + ζx 𝐄𝛇 + ξy 𝐅𝛏 + ηy 𝐅𝛈 + ζy 𝐅𝛇 + ξz 𝐆𝛏 + ηz 𝐆𝛈 + ζz 𝐆𝛇 = 𝟎

𝐔 𝐄ξx + 𝐅ξy + 𝐆ξz 𝐄ηx + 𝐅ηy + 𝐆ηz 𝐄ζx + 𝐅ζy + 𝐆ζz

( ) +( )+ ( )+( )=0
⏟𝐉 ⏟ 𝐉 ⏟ 𝐉 ⏟ 𝐉
U1 t E1 F1 G1
∂𝐔1 ∂𝐄1 ∂𝐅1 ∂𝐆1
+ + + =0
∂t ∂ξ ∂η ∂ζ
Eq. 11.4
By no means is this form of representation is conclusive. Many texts and researchers are attain their
own representation as they see to be relevant. But the concepts should be the same. This is the strong
conversation form of governing equation which is best suited for finite differencing schemes. It
should be noted that the vectors E1, F1and G1 contain partial derivatives of viscous and heat transfer
terms which also has to be transformed. Alternatively, the vectors E, F, and G can be split into an

324Anderson, Dale A; Tannehill, John C; Plecher Richard H; 1984:”Computational Fluid Mechanics and Heat
Transfer”, Hemisphere Publishing Corporation.
 270

inviscid (i) and a viscous parts (v). Reason for doing this becomes evident later, and also this make it
more modular and easy to handle.

U  (Ei - E v )  (Fi - Fv )  (G i - G v )
+ + + =0
t x y z
  U   1
 +  
 ξ x (Ei − E v ) + ξ y (Fi − Fv ) + ξ z (G i − G v ) 
t  J  ξ  J



+
 1
 
 η x (Ei − E v ) + η y (Fi − Fv ) + η z (G i − G v ) 
η  J



+
 1
  
 ζ x (Ei − E v ) + ζ y (Fi − Fv ) + ζ z (G i − G v )  = 0
ζ  J 
ρ   ρu   0 
 ρu   ρu 2 + p   τ xx 
     
U =  ρv  , E i =  ρuv  , E v =  τ xy 
     
ρw   ρuw + p   τ xz 
 E t  (E t + p)u  uτ xx + vτ xy + wτ xz − q x 
 
 ρv   0 
 ρuv    xy 
   
Fi =  ρv 2 + p  , Fv =   yy 
   
 ρvw    yz 
(E t + p)v u xy + v yy + w yz − q y 
 
 ρw   0 
 ρuw    xz 
   
G i =  ρvw  , Gv =   yz  Eq. 11.5
   
 ρw + p 
2
  zz 
(E t + p)w  u xz + v yz + w zz − q z 
 

11.2 Solution Methods for Compressible (Density Based) N-S Equation

The unsteady compressible N-S equations are a mixed set of hyperbolic-parabolic equations, while
the unsteady incompressible is a mixed set of elliptic-parabolic. As a consequence, different
numerical techniques should be used to solve N-S equations in compressible vs incompressible flow
regimes. We start with compressible methods first and divide the categories to Explicit and Implicit
schemes.
11.2.1 Explicit Scheme (Mac Cormack)
The Mac Cormack Explicit, the predictor/corrector which gained a lots of popularity among
research cycles in 70s and early 80s. For an excellent discussion on these and many more, reader is
 271

advised review325. The compressible N-S equations results:

∂U ∂E ∂F ∂G
+ + + =0
∂t ∂x ∂y ∂z
𝑛+1 𝑛 Δt 𝑛 𝑛 Δt 𝑛 𝑛 Δt 𝑛 𝑛
Predictor: Ui,j,k = 𝑈i,j,k − (𝐸𝑖+1,j,k − 𝐸i,j,k ) − (𝐹i,j+1,k − 𝐹i,j,k ) − (𝐺i,j,k+1 − 𝐺i,j,k )
Δx Δy Δz
n+1
Corrector: Ui,j,k
1 n n+1 Δt n+1 n+1 Δt n+1 n+1
= [Ui,j,k + Ui,j,k − (Ei,j,k − Ei−1,j,k ) − (Fi,j,k − Fi,j−1,k )
2 Δx Δy
Δt n+1 n+1
− (Gi,j,k − Gi,j,k−1 )]
Δz
Eq. 11.6
Where x= i Δx, y= j Δy and z = k Δz. This is 2nd order accurate in both space and time. The choice of
Δt for stability consideration is obtained from326 as:

-1
α(Δt)CFL u v w 1 1 1 
Δt  wher e (t)CFL   + + +a + + 
1 + 2 / Re   Δx Δy Δz (x ) (y ) (z )2 
2 2

ρ u Δx ρ v Δy ρ w Δz
with Re Δ = Min (ReΔx , Re Δy , Re Δz ) Re Δx = Re Δy = Re Δz =
μ μ μ
and a = (p /  ) 0.5 , safty factor α  0.9
Eq. 11.7
The explicit scheme is 2nd order accurate in both space and time. In the present form, forward
differencing are used for all special derivatives in the predictor step while backward differences are
used in the corrector step. The forward and backward differencing can be alternated between
predictor and corrector steps as well as between the three spatial derivatives in sequential fashion.
This eliminates any bias due to the on sided differencing. Moreover, the derivatives appearing in
viscous terms of E, F and G must be differenced correctly in order to maintain 2nd order accuracy. The
x derivative terms appearing in E are differenced in the opposite direction to that for dE/dx while the
y-derivative and z are approximated with central differencing. Likewise, the y derivative terms
appearing in F and z derivative terms appearing in G are differenced in opposite direction to that
used for dF/dy and dG/dz respectively. Cross-derivative terms in F and G are approximated with
central differencing. After each predictor or corrector step, the primitive variables (ρ, u, v, w, e, p, T)
can be found be decoding the U vector. McCormack modified the original method by splitting the
original McCormack scheme into a sequence of one-dimensional operations327. Thus, it become
possible to advance the solution in each direction (Δtx, Δty, and Δtz) with large differences in mesh
spacing (Δx, Δy, Δz). The explicit Mac Cormack algorithm is a suitable method for solving both steady
and unsteady flows at moderate to low Reynolds numbers. However, it is not satisfactory method for
solving high Reynolds number flows where the viscous regions become very thin. For these flows,

325 Anderson, Dale A; Tannehill, John C; Plecher Richard H; 1984:”Computational Fluid Mechanics and Heat
Transfer”, Hemisphere Publishing Corporation.
326 Tannehill, J, C., Holst, T, L, and Rakich, J, V, “Numerical Computation of Two-Dimensional Viscous Blunt Body

Flows and Impinging Shock”, AIAA Paper 75-154, Pasadena, Ca, USA, 1975.
327 MacCormack, R, W, “Numerica solution of the Interaction of a shock wave with Laminar boundary layer”,

Proceding of 2nd International Conference in Numerical Methods in Fluid Dynamics, pp. 151-163.
 272

the mesh must be highly refined to resolve the viscous regions. This leads to small time steps and
longer computer time.
11.2.2 Case Study – Dual Block Applied to Navier-Stokes Equations in 2D
A 2D Navier-Stokes equation using explicit Mac Cormack method on multi-block structured mesh
are investigated by [Almeida et al.]328, for steady state and unsteady state compressible fluid flows.
The multi-block technique and generalized coordinate system are used to develop a numerical solver
which can be applied for a large range of compressible flow problems on complex geometries without
modifying the governing equations and numerical method. Besides that the numerical method is
based on a finite difference approach and the generalized coordinates introduced allow the
application of the boundary conditions easily. The subsonic flow over a backward facing step and
supersonic flow over a curved ramp are presented, and the results are compared with the
experimental and numerical data.
11.2.2.1 Governing Equations
The two-dimensional compressible Navier-Stokes equations in generalized coordinates system (x ,h)
without body forces, mass diffusion, finite-rate chemical reactions, or external heat addition can be
written in non-dimensional conservative law form as [Anderson et al. ]329.

̂ ∂(𝐄̂𝐢 − 𝐄̂𝐯 ) ∂(𝐅̂𝐢 − 𝐅̂𝐯 )

∂𝐐
+ + =0
∂t ∂ξ ∂η
Eq. 11.8
where Q is the state vector of conservative variables defined as below and Êi and ˆFi are the inviscid
flux vectors,

ρ ρU ρV
ρU ρuU + ξx P ρuV + ηx P
̂=[ ] ,
𝐐 𝐄̂𝒊 = 𝐉 −1 ρvU + ξ P , 𝐄̂𝒊 = 𝐉 −1 [ρvV + η P]
ρV y y
Et [ (Et + p)U] (Et + p)V
Eq. 11.9
The Ev and Fv are the viscous flux vectors in the x and h directions, which are given below

0 0
−1 −1
Re (ξx τxx + ξy τxy ) Re (ηx τxx + ηy τxy )
𝐄̂𝐯 = 𝐉 −1
Re−1 (ξx τxy + ξy τyy ) , 𝐄̂𝐢 = 𝐉 −1
Re−1 (ηx τxy + ηy τyy )
−1 −1
[ Re (ξx βx + ξy βy ) ] [ Re (ηx βx + ηy βy ) ]
Eq. 11.10
where J is the Jacobian of the transformation, r is the density, u and v are the velocity components in
the x and h coordinate directions, U and V are the contravariant velocities, Et is total energy per unit
of volume, ξx, ξy, ηx and ηy are the metrics of transformation, βx = τxxu + τxyv, and βy = τxyu + τyyv, p is
the static pressure and t describes the stress components for viscous flow. The Navier-Stokes

328 Jeferson Osmar de Almeida, Diomar Cesar Lobão, Cleyton Senior Stampa, Gustavo Benitez Alvarez, “Multi-
block Technique Applied to Navier-Stokes Equations in Two Dimensions”, Original Article, DOI: 10.5433/1679-
0375.2018v39n2p115.
329 Anderson, D.; Tannehill, J.; Pletcher, R. Computational fluid mechanics and heat transfer, Hemisphere

Publishing Corporation - McGraw Hill, 1984.

 273

equations are based on the universal law of conservation of mass, conservation of momentum and
conservation of energy. However, to complete this system of equations is necessary to add an
equation of state that can be written as:

1
p = (γ − 1) [E𝑡 − ρ (u2 + v 2 )]
2
Eq. 11.11
in which the fluid is considered a perfect gas. In the above equation, γ denotes the ratio of the specific
heats. The dimensionless of the variables is performed to eliminate the scale problems. Please see
[Anderson et al. ]330.
11.2.2.2 Numerical Method
The Mac Cormack method331 is used for solving the governing equations as given by Eq. 11.8. The
numerical method is an explicit predictor-corrector scheme based on a finite difference formulation
that has been used for compressible flow, and it has second-order of accuracy in both space and time.
When the method is applied to the two-dimensional compressible Navier-Stokes equations given by
Eq. 11.8, it can be written as

Δt n Δt n
Predictor: ̂ n+1
Q i,j
̂ ni,j −
=𝐐 (𝐄̂i+1,j − 𝐄̂i,jn ) − (𝐅̂ − 𝐅̂i,jn )
Δξ Δη i,j+1
1 n ̅̅̅̅̅̅ Δt n+1̅̅̅̅̅̅ ̅̅̅̅̅̅ Δt n+1 ̅̅̅̅̅̅ ̅̅̅̅̅̅
Corrector: ̂ n+1
Q i,j = [̂ ̂ n+1
𝐐i,j + 𝐐 i,j − (𝐄̂i,j − 𝐄̂i-1,j
n+1
)− (𝐅̂i,j − 𝐅̂i,j-1
n+1
)]
2 Δξ Δη
Eq. 11.12
where Ê = Êi -Êv and ˆF = ˆFi -ˆFv. The explicit Mac Cormack method requires the predictor be calculated
first to n = 0 (initial conditions). As stated by [Roe]332 any flow which goes through Mach 1.0, the
classical numerical methods (Finite Difference/Finite Volume) present physical discontinuities, so
needs special numerical treatment. In the present work the flow is much greater than Mach 1.0, well
developed supersonic flow. In this case the Mac Cormack method does not require such special
treatment. Note that Mac Cormack method does not carry any special treatment for the convective
terms (inviscid) or any other terms in the formulation, it is a very naive method. Here, the boundary
conditions are defined as follows: inflow, outflow and lower and upper solid wall. The inflow
condition, all variables are prescribed as Dirichlet boundary conditions in supersonic regime. For
outflow condition, with subsonic flow, the value of the static pressure p is prescribed as Dirichlet
boundary conditions and all other variables are extrapolated, but if the flow is supersonic, all other
variables are extrapolated. The solid wall condition are represented by non-slip boundary condition
u = 0 for viscous flow and free-slip boundary condition U. n = 0 for inviscid flow, where in the last
case the velocity components are calculated through of the contravariant velocity components.
However, in both cases all other variables are extrapolated.
11.2.2.3 Block Interface
A multi-block structured mesh with two blocks is used to clarify the block interface boundary
condition treatment. The first block, which is in red, is where the fluid enters. The second block, which
is in blue, is where the fluid from block1 becomes its inlet. The mesh of each block has 60 x 50 nodes.

330 Anderson, D.; Tannehill, J.; Pletcher, R. Computational fluid mechanics and heat transfer, Hemisphere
Publishing Corporation - McGraw Hill, 1984.
331 Mac Cormack, R. The effect of viscosity in hypervelocity impact cratering. AIAA., 1969. p. 69–354, 1969.
332 ROE, P. Approximate Riemann solvers, parameter vectors, and difference schemes. Journal of Computational

Physics, 1981. v. 43, p. 357–372, 1981.

 274

For the treatment of the boundary between blocks1 and 2, it is observed that they are physically
coincident. For the exchange of information between the blocks, the multi-block structured mesh is
coincident at the boundary. At the boundary between the blocks, the boundary of block1 is
considered as an outflow boundary condition, since the fluid leaves this block. In block2, the
boundary is the inlet of the fluid exiting block1, being considered an inlet boundary condition. In
order to avoid a numerical discontinuity of the solution at the boundary between the blocks, after
each iteration a numerical boundary condition is applied at this border, where the values of the
variables are determined through the variables neighboring of such boundary. For this numerical
condition, please consult the work by [Almeida et al.]333. With the input and initial conditions and
the boundary conditions mentioned above, it is possible to obtain satisfactory numerical results by
implementing the Mac Cormack original method and the use of multi-block structured mesh in the
cases discussed as follow.
11.2.2.4 Stability Consideration
The monitoring of the stability of the numerical scheme is performed by observing the residue at all
internal points of the discretized domain, according to the equation below:

CFL γP 2
(Δt) =
2v
where CFx = |ξx u + ξy v | + √ √ξx + ξ2y
+ CFx + CFy ρ
Re
γP 2
and CFy = |ηx u + ηy v | + √ √ηx + η2y
ρ
Eq. 11.13
Among the conserved variables of ˆQ, the one used in the above equation for determining the residue
is the density ρ. To keep the numerical scheme stable, the value of the time integration step can be
defined by the following empirical formula in generalized coordinates [Anderson et al. ]334, [Mac
Cormack]335, where CFL is the Courant-Friedrichs-Lewy number. For the explicit Mac Cormack
scheme the CFL must be less than or equal to 0.5. The values of CFx and CFy for the above equation
can be determined through Eq. 11.13.
11.2.2.4.1 Test 1 - Flow Over a Backward-Facing Step
The first test is the well-known subsonic flow over a backward facing step that is often used as
benchmark problem in computational fluid dynamics. The main feature of this flow is that it has a
simple geometry that generates an interesting and complex flow field, such as flow separation,
reattachment zone and recirculation bubbles on the upper and lower wall of the channel. These
characteristics are dependent on the Reynolds number and the geometrical parameters336; [Biswas

333 Jeferson Osmar de Almeida, Diomar Cesar Lobão , Cleyton Senior Stampa, Gustavo Benitez Alvarez, “Multi-
block Technique Applied to Navier-Stokes Equations in Two Dimensions”, Original Article, DOI: 10.5433/1679-
0375.2018v39n2p115.
334 Anderson, D.; Tannehill, J.; Pletcher, R. Computational fluid mechanics and heat transfer, Hemisphere

Publishing Corporation - McGraw Hill, 1984.

335 Mac Cormack, R. The effect of viscosity in hypervelocity impact cratering. AIAA., 1969. p. 69–354, 1969.
336 Armaly, B.; Durst, F.; Pereira, J.; Schonung, B. Experimental and theoretical investigation of backward facing

step flow. J. Fluid Mech, 1983. v. 127, p. 473–496, 1983.

 275

et al.]337, [Breuer and Durst]338, [Saleel et al.]339. For the numerical simulations, the following initial
conditions are assumed: density r0 = 1.21kg/m3, pressure p0 = 1.01x105N/m2, coefficient of dynamic
viscosity μ = 1.81x105 kg/(m.s), and ratio of the specific heats γ = 1.4. The block1 (upstream block )
has a mesh size of 25x25 and the block2 (downstream block ) has size of 160 x 50. The Figure 11.1
(a-b) shows the velocity contours of the steady state flow field for two different Reynolds number
(Re = 50 and 100) for expansion ratio H/h=1.9423 (⋍2). In all cases shown in figures, a vortex is
found in the concave corner behind the step and the maximum velocity is located on the upstream
side of the channel, as found in the experiments performed by, [Biswas et al.]340, [Saleel et al.341]. It
can also be seen that the size of the
recirculation region increases with
Experimental xr Numerical xr
increasing Reynolds number. The
Table 11.1 describes the variation of
Re REF-1 Authors REF-2 REF-3
reattachment length (xr) for two
different Reynolds numbers (Re = 50 50 1.70 1.61 1.55 1.55
and 100), which clearly shows that the 100 3.06 2.82 2.80 2.81
reattachment length increase with
increasing Reynolds number, as found Table 11.1 Comparation of the Results For Flow Over a
in the references. In the second column Backward Facing Step
are presented experimental values

(a) Re = 50

(b) Re = 100

Figure 11.1 Velocity Contours for Flow Over a Backward Facing Step

337 Biswas, G.; Breuer, M.; Durst, F. Backward-facing step flows for various expansion ratios at low and moderate
Reynolds numbers. Journal of Fluids Engineering, 2004. v. 126, p. 362–374, 2004. DOI: 10.1115/1.1760532.
338 Baird, S.; Mcguirk, J. J. Multi-block parallel simulation of fluid flow in a fuel cell. In: High-Performance

Computing and Networking. HPCN-Europe 1999. Springer Lecture Notes in Computer Science, 1999. v. 1583.
339 Saleel, C.; Shaija, S.; Jayaraj, S. On simulation of backward facing step flow using immersed boundary method.

American Journal of Fluid Dynamics, 2013.

340 See 283.
341 Saleel, C.; Shaija, S.; Jayaraj, S. On simulation of backward facing step flow using immersed boundary method.

American Journal of Fluid Dynamics, 2013.

 276

obtained in [Armaly et al.]342, and denoted by REF-1. The third column presents the results obtained
in this work, and denoted by Authors. The fourth and fifth columns correspond to values obtained by
numerical simulations in [Biswas et al.]343 and [Saleel et al.]344, and denoted by REF-2 and REF-3
respectively. The results obtained by the present work are quite compatible with results found in the
references. As can be clearly seen in Figure 11.1 (a-b) and Table 11.1, the results are in agreement
with the numerical data especially with respect to the reattachment length.
11.2.2.4.2 Test 2 - Flow Over a Curved Ramp
The second test case is a transient supersonic flow over a curved ramp at Mach number 1.5 that can
be used as benchmark problem for the supersonic Euler equations. This flow is characterized by the
formation of a detached bow shock in front of the curved ramp and of an expansion wave. Moreover,
other feature is the wave shock reflection from the solid wall [Lobao]345-346, [Allen]347. For the
numerical simulation, the initial conditions are as follows: density r0 = 1.21kg/m3, pressure p0 = 1.01
x 105N/m2, and ratio of the specific heats γ = 1.4. The mesh for the multi-block solutions of this
problem. The mesh sizes in the first and second blocks are 110 x 80 and 60 x 80, respectively. The
pressure and Mach number contour over a curved ramp are shown in Figure 11.2 (a-b)
respectively, for Mach number 1.5. The detached bow shock in front of the curved ramp can be seen
in the figures as well as the shock reflection from the solid wall. The Figure 11.2 (b) shows the
expansion wave, where is possible to observe the subsonic, transonic and supersonic zones before of
the curved ramp. As can be observed in Figure 11.2 (a-b) there is no spurious oscillation at the
interface boundary between the two blocks even in presence of a very strong shock wave structure.

(a) Pressure contour (b) Mach number contour

Figure 11.2 Flow Field Over a Curved Ramp

342 Armaly, B.; Durst, F.; Pereira, J.; Schonung, B. Experimental and theoretical investigation of backward facing
step flow. J. Fluid Mech, 1983. v. 127, p. 473–496, 1983.
343 Biswas, G.; Breuer, M.; Durst, F. Backward-facing step flows for various expansion ratios at low and moderate

Reynolds numbers. Journal of Fluids Engineering, 2004. v. 126, p. 362–374, 2004

344 Saleel, C.; Shaija, S.; Jayaraj, S. On simulation of backward facing step flow using immersed boundary method.

American Journal of Fluid Dynamics, 2013.

345 Lobao, D. High Resolution Schemes Applied to the Euler Equations. (PhD thesis), University of Bristol, 1992.
346 Lobao, D. Numerical simulations of Navier-stokes for transient flows in 2d. In: Mechanical Computational.

Buenos Aires, Argentina: [s.n.], 2010.

347 Allen, C. An Efficient Euler Solver for Predominantly Supersonic Flows with Embedded Subsonic Pockets. (PhD

thesis, University of Bristol, 1992.

 277

The numerical interface boundary condition is well suited for this simulation which demonstrates its
applicability. The present numerical results are compared with the numerical data presented in
[Lobao]348.
11.2.2.5 Conclusions
In this work, the discretization of the Navier-Stokes and Euler equations on multi-block structured
mesh is realized using explicit Mac Cormack method. The numerical results for steady state and
unsteady state compressible fluid flow in two-dimensional geometries are presented for two
different test cases in subsonic and supersonic regimes, the flow over a backward facing step and
flow over a curved ramp, respectively, which indicated good agreement with the references data.
Therefore, it is shown the capability of the methodology to obtain numerical results for these types
of subsonic and supersonic flows in two-dimensional complex geometries.
11.2.3 Other Explicit Schemes
In addition to Mac Cormack scheme, other explicit methods can be used to solve the Compressible N-
S equation, including:
• Hopscotch Method
• Leapfrog/DuFort-Frankel Method
• Brailovskaya Method
• Allen-Cheng Method
• Lax-Wendroff Method

These as discussed earlier might be used for Heat equation and viscous Burger’s equation, but
difficult to apply to more complicated equation like compressible N-S. For one thing, all the above,
except the Lax-Wendroff, are 1st order accurate in time. So they cannot be used for accurately
compute the time evolution of a flow field. In addition, all have stability restriction which limits the
maximum time steps.
11.2.4 Implicit Schemes
11.2.4.1 Beam-Warming
In the Beam-Warming scheme, the solution is marched in time using:

θ1Δt  n Δt  n θ  1  
ΔnU = (Δ U) + (U ) + 2 Δ n −1U +   θ1 − − θ 2 (t)2 + (t)3 
1 + θ 2 t 1 + θ 2 t 1 + θ2  2  
Δ n U = U n +1 − U n
Eq. 11.14
This is a general difference formula (revisited) with appropriate choice of parameters ϴ1 and ϴ2,
represents many of the standard difference scheme. For compressible N-S equations, either the Euler
implicit scheme (ϴ1 =1, ϴ2 =0), which is first order accurate in time, or a three point backward implicit
scheme (ϴ1=1, ϴ2=1/2), which is 2nd order accurate in time, is normally used. The difference formula
so called delta form is linearized using truncated Taylor series expansion. For example

348 Lobao, D. High Resolution Schemes Applied to the Euler Equations. (PhD thesis)”, University of Bristol, 1992.
 278

 E 
n
n +1 n +1
E = E +
n
 (U − U ) + Ο[( t) ]
n 2

 U 
Δ n E = [ A]Δ n U + Ο[(Δt)2 ] wher e E = Ei - E v
Similary Eq. 11.15
Δ F = [B]Δ U + Ο[(Δt) ]
n n 2

Δ n G = [C]Δ n U + Ο[(Δt)2 ]

Where [A], [B], and [C] are the Jacobian matrix which their definition can be obtained from different
sources, such as [Chung]349. The details for a 2D compressible N-S equation is provided in350.
11.2.4.2 Mac Cormack
Mac Cormack (1981) also developed an implicit algorithm analog to his explicit method. This new
method consists of two stages. In the first stage, uses the original Mac Cormack scheme while the
second stage employs an implicit algorithm which eliminates any stability restriction. The resulting
matrix equations are either lower or upper block equations with details in351. The implicit Mac
Cormack method is defined by:

 λΔt  n +1 λΔt n +1
1 + Δu i = (u i ) explicit + Δu i +1
n
Predictor:
 Δx  Δx
Δu in +1 = u in +1 − u in ( )
(u in ) explicit = u in +1 explicit
− u in
 λΔt  n +1 n +1 λΔt n +1
Corrector: 1 + Δu i = (u i ) explicit + Δu i −1
 Δx  Δx
1
(
u in +1 = u in + u in +1 + Δu in +1
2
) ( ) 1
 ( )
u in +1 explicit = u in + u in +1
2 explicit
(
+ Δu in +1 )
explicit

 2μ Δx  
λ   c + −  , 0.0 c,μ  0
 Δx Δt  
Eq. 11.16
This method is unconditionally stable and 2nd order accurate in both space and time, provided that
μΔt/(Δx)2 is bounded as Δt and Δx approach zero. For detail explanation of method and accuracy, as
well as a 2nd order N-S example, reader could refer to352.

349 T. J. Chung, “Computational Fluid Dynamics”, University of Alabama in Huntsville, Cambridge University
Press 2002.
350 Anderson, Dale A; Tannehill, John C; Plecher Richard H; 1984:”Computational Fluid Mechanics and Heat

Transfer”, Hemisphere Publishing Corporation.

351 Anderson, Dale A; Tannehill, John C; Plecher Richard H; 1984:”Computational Fluid Mechanics and Heat

Transfer”, Hemisphere Publishing Corporation.

352 Anderson, Dale A; Tannehill, John C; Plecher Richard H; 1984:”Computational Fluid Mechanics and Heat

Transfer”, Hemisphere Publishing Corporation.

 279

11.3 Solution Methods for Incompressible N-S Equations (Pressure Based)

11.3.1 Explicit Schemes
The incompressible N-S equations are tougher to solve due to their elliptic nature as wells as
decoupling of pressure from continuity.
11.3.1.1 Vorticity-Stream Function Approach (2D)
Taking curl of incompressible momentum equation above, the vorticity potential equation is derived
as
∂v ∂u ∂ψ ∂ψ Dω
ω= − , u= , v=− , ⏟ + ν∇2 ω
= ω∇V
∂x ∂y ∂y ∂x Dt 0
Eq. 11.17
Where the scalars such as pressure vanish. While vorticity is not a primary variable in flow analysis,
it is still extremely instructive to examine the impact of vorticity vector on N-S equation. The first
term on the right arises from the convective derivative and is called vortex stretching term. The
second term is obviously a viscous-diffusion term. The vorticity–stream function approach is one of
the most popular methods for solving 2D incompressible N-S equations. This parabolic PDE is called
Vorticity Transport Equation. An additional equation involving the new dependent variables ω and
ψ can be obtained

Step 1
Set initial ω
and ψ at t=0
Step 2
Step 6 Solve
Solve
for Pressure - if
Vorticity
solution not
Transport
converged
Eqation for ω
return to step 2
at t+Δt
Vorticity-
Stream
Function
Approach
Step 3
Step 5
Iterate for
detemine
new ψ
value of ω on
(Poisson
boundries
Equation)

Step 4
u=ψx v=ψy

Figure 11.3 Vorticity-Stream Function Approach For 2D Incompressible N-S Equations

 280

 2ψ  2ψ
+ = −ω Eq. 11.18
x 2 y 2
This an elliptic equation (Poisson equation) which can be solved readily. As a result of change
variables, we have been able to separate the mixed elliptic-parabolic 2D incompressible N-S equation
into one parabolic equation and one elliptic one. In order to determine the pressure, we have
differentiate the x-momentum equation with respect to y (d/dy), and y-momentum with x (d/dx) and
adding to obtain:
 u v u v 
 2 p = 2ρ −  Eq. 11.19
 x y y x 

Which is a Poisson equations for pressure. These equations are normally solved using time marching
procedure. (see Figure 11.3).
11.3.1.2 Primitive-Variable Approach
The vorticity-stream function approach for solving the incompressible N-S equations losses it
attractiveness when applied to 3D flow because a single scalar stream function does not exist in that
case. One of the early techniques proposed for solving incompressible N-S equation in primitive
variables form is the artificial compressibility method of 353. In this method, the continuity equation
is modified to include an artificial compressibility term which vanishes when a steady-state solution
is reached. With the additional of this term, to continuity equation, the resulting N-S equations are
mixed set of hyperbolic-parabolic equations which can be solved with standard time dependent
approach.
ρ u v w
+ + + =0 Eq. 11.20
t  x y z
Where ρ’ is an artificial density and t’ is a fictitious time which is analogous to real time in a
compressible flow. The artificial density is related to the pressure by an artificial equation of state
p=ρ’/β where β is an artificial compressibility factor to be determine later. In general, an implicit
finite-differencing scheme is recommended over an explicit one. The artificial compressibility
method is one of technique for solving the incompressible N-S equations. By far the most common
primitive-variable approach is using a Poisson equation for pressure in place of the continuity
equation, known as one of the Projection Method. The other one is Fractional Step method. The
algorithms generally defined as354:
1. First the system is progressed in time to a mid-time-step position, solving the above transport
equations for mass and momentum using a suitable advection method. This is denoted the
predictor step.
2. At this point an initial projection may be implemented such that the mid-time-step velocity
field is enforced as divergence free.
3. The corrector part of the algorithm is then progressed. These use the time-centered estimates
of the velocity, density, etc. to form final time-step state.
4. A final projection is then applied to enforce the divergence restraint on the velocity field. The
system has now been fully updated to the new time.

353 Chorin. A, J, “A Numerical Method for Solving Incompressible Viscous Flow Problem”, Journal of Computational

Physics, Volume 2, pp. 12-26, 1967.

354 From Wikipedia, the free encyclopedia.
 281

11.3.2 Implicit Schemes

11.3.2.1 SIMPLE (Semi-Implicit Method for Pressure Linked Equation) for 2D N-S Equation
Since the SIMPLE procedure was described in detail previously, here an outline of it would be
sufficient. It is suffice to say that it can also be applied to the incompressible N-S equation. This
procedure based on a cyclic series of guess-and-correct operations to solve the governing equations.
The velocity is first calculated from the momentum equation using a guess pressure. The pressure
and velocities are then corrected so as to satisfy continuity. This process is continued until the
solution converges. In this procedure, the actual pressure p, u, v is written as

p = p 0 + p
u = u 0 + u Eq. 11.21
v = v 0 + v

Where p0, u0, v0 are the estimated values of velocity and p’, u’ and v’ are the velocity corrections. The
pressure corrections are related to the velocity corrections by approximate form of momentum
equations as
u  p Δt p
ρ =− → u = −
t x ρ x
Eq. 11.22
v p Δt p
ρ =− → v = −
t y ρ y

Where Δt’ is a fictitious time increment. After combining and substituting in continuity equation

∂u ∂v ∂u0 ∂v0 ∆t ′ ∂2 p′ ∂2 p′
( + )−( + )+ ( + 2) = 0
⏟∂x ∂y ∂x ∂y ρ ∂x 2 ∂y
0
∂u0 ∂v0 ∆t ′ 2 ′
( + ) = ( )∇ p
∂x ∂y ρ
Eq. 11.23
Where the Poisson equation can be solved for the pressure correction. The procedure can be
described in fig with steps starting with step 1 until it converges. The SIMPLE procedure has been
used successfully to solve a number of incompressible flow problems. However, for certain cases, it
is found that the convergence rate is not satisfactory. This id due to the pressure correction equation
tends to overestimate the value of p’ even though the corresponding velocities correction are
reasonable. That is why an under relaxation constant ωp is been used for pressure correction as
indicated below:
p = p 0 + ωp p Eq. 11.24

Figure 11.4 shows the necessary steps in SIMPLE procedure. The SIMPLE procedure has been
revised to improve the rate of convergence. The new procedure is called SIMPLER (SIMPLR Revised).
In SIMPLER the velocity corrections are computed in the same manner as the SIMPLE, but a complete
Poisson equation for the pressure is used. Also, the velocity field is guessed initially instead of
pressure.
 282

Step 1
Guess p0

Step 5
Replace Step 2
p0,u0,v0, Solve for
with p , u, u0,v0
v

Step 4 Step 3
Correct Solve for
the pressure
pressure correction
p=p0+p' p'

Figure 11.4 SIMPLE Procedure as applied to 2D Equation

11.3.3 Projection Methods

Other incompressible methods such as Projection Methods of [Chorin, 1967] as well as Fractional
Step , MAC and Fast Fluid Dynamics (FFD) as detailed before.

11.4 Approximation (Thin-Layer) of N-S Equation

For Thin-Layer approximation to the N-S equations, the viscous terms containing derivatives in the
direction parallel to the body surface are neglected in unsteady N-S equations. But all other terms in
momentum equation are retained. Although the thin-layer N-S equations are considerably less
complicated the complete N-S, a substantial amount of computer effort is still required to solve these
equations. The thin-layer N-S equations are a mixed set of hyperbolic-parabolic PDE in time. As a
consequence, the time dependent approach can be applied in the same manner of full N-S. Thus, the
methods of solving complete N-S equations previously discussed can be applied here also. (See Eq.
11.25).
U E F G
+ + + =0
t x y z
ρ  ρu   ρv   ρw 
 ρu   ρu 2 + p   ρuv − τ xy   ρuw 
       
U =  ρv  E=  ρuv  F=  ρv 2
+ p − τ yy
 G=  ρvw 
       
ρw   ρuw   ρvw − τ yz   ρw 2
+ p 
 E t  (E t + p)u  (E t + p)v − uτ xy − vτ yy − wτ yz + q y  (E t + p)w 
 
 283

Eq. 11.25

11.5 Parabolized N-S Equation

The parabolized N-S (PNS) equations Have recently gained popularity because the can predict
complex 3D, steady, supersonic, viscous flow field in an efficient manner. This efficiency achieved
because the equations can be solved using a space marching finite-difference technique as oppose
the time marching technique which is normally employed for the full N-S equations. The PNS
equations are for an entire supersonic flow field similar somewhat to solving either the inviscid
portions using an Euler equations or the viscous portion using the Boundary Layer equations. The
PNS are mixt set of hyperbolic-parabolic equations provided that certain conditions provided. These
condition include the requirement that the inviscid outer region of the flow field be supersonic and
stream wise velocity component be everywhere positive355.
An additional constraint is caused by the presence of the stream wise pressure gradient in the stream
wise momentum equation. If this term included everywhere in the flow field, then upstream influence
can occur in the subsonic portion of the boundary layer and a space marching method of the solution
is not well-poised. This leads to exponentially growing solutions which are often called departure
solutions. Several techniques are available to circumvent this difficulty. One of the novel technique
for handling the stream wise pressure gradient term was proposed by Vigneron.356. In this approach,
a fraction of pressure gradient term, ω (dp/dx) in the stream wise momentum equation is retained
in the subsonic viscous region and the remainder (1 - ω) (dp/dx) is either omitted or is evaluated
explicitly using a back-ward differencing for the “sub-layer approximation” techniques. The
parameter ω can be obtained by analyzing the 2D PNS equation, where ω was inserted for pressure
gradient (ω p). By ignoring the viscous term and obtaining the eigenvalues of the remaining Euler
equation, an expression can be obtained for ω as

γM2x σλM 2x
ω = Eq. 11.26
1 + (γ − 1)M2x 1 + (γ − 1)M2x

Where σ is safety factor. The PNS equations in generalized coordinates can be obtained by simply
dropping the unsteady and viscous terms containing partial derivatives with respect to stream wise
direction ξ. The resulting equations become:

E3 F3 G 3

+

+

=0 where E3 =
1
J

ξ x Ei + ξ y Fi + ξ zG i 
1

F3 = η x (Ei − Ev ) + η y (Fi − Fv ) + ηz (G i − Gv )
J

1

G 3 = ζ x (Ei − Ev ) + ζ y (Fi − Fv ) + ζ z (G i − Gv )
J

Eq. 11.27
And the prime indicates that the terms containing viscous partial derivatives w.r.t ξ have been
ignored. The stress and heat flux terms are reduced to

355 Anderson, Dale A; Tannehill, John C; Plecher Richard H; 1984,”Computational Fluid Mechanics and Heat
Transfer”, Hemisphere Publishing Corporation.
356 Vigneron, Y, C, Rakich, J, V, and Tannehill, J, C, “Calculation of Supersonic Viscous Flow over Delta Wings with

Sharp Subsoinc Leading Edge”, AIAA paper 78-1137, Seattle, Washington, 1978.
 284

 xx =
2
3
 
 2( x u +  x u ) − ( y v +  y v ) − ( z w +  z w )
2

 yy =  2( y v +  y v ) − ( x u +  y u ) − ( z w +  z w )
3

2

 zz =  2( z w +  z w ) − ( x u +  y u ) − ( y v +  y v )
3


 xy =  yx =   y u +  y u +  x v +  y v 
 xz =  zx =   u
z  +  z u +  x w +  y w 


 yz =  zy =   v
z  +  z v +  y w +  y w 


qx = -k(  xT +  xT ) , qy = -k(  yT +  yT ) , qz = -k(  zT +  zT )
Eq. 11.28
Bear in mind that for many applications, the thin–layer approximation can also be applied to the PNS
equations.
11.5.1 3D Compressible Implicit Numerical Solution of PNS Equation
As previously indicated, the PNS equations are mix set of hyperbolic-parabolic equations in the
stream wise direction provided following has been met:

• Inviscid flow is supersonic.

• Stream wise velocity component is everywhere greater than zero.
• Stream wise pressure gradient term in stream wise momentum equation is either omitted or
the departure behavior is suppressed.

Until recently, the PNS equations have been solved using iterative, implicit finite-differencing
schemes. [Vigneron] were the first to employ a more efficient non-iterative, implicit approximation
factorization (AF) scheme to solve the PNS equations. Let us apply 3D compressible PNS equations
written in Cartesian coordinates (x is stream wise direction) as

 u   0 
 u 2 + p  (1 −  ) p 
E p F G    
+ + + =0 where E =  uv  p= 0 
x x y z    
  uw   0 
( Et + p )u   0 
E = Ei + p , F = Fi − Fv , G = G i − G v
Eq. 11.29
Where we incorporate the pressure gradient term in stream wise direction separately. The solution
is marched in x-direction using the Beam and Warming marching technique as
 285

θ1 + Δx  i Δx  i θ  1  
ΔiE = (Δ E) + (E ) + 2 Δ i −1E + O  θ1 − − θ 2 (x) 2 + (x) 3 
1 + θ 2 x 1 + θ 2 x 1 + θ2  2  
where Δ i E = Ei +1 − Ei , x = ix
substituting governing equations in BW marching formula:
θ1 + Δx   (i F)  (i G)  Δx   (F i )  (G i )  θ
ΔE=
i
 +  +  +  + 2 Δ i −1E
1 + θ 2  y z  1 + θ 2  y z  1 + θ 2
where Δ i E = Δ i E + Δ ip , Δ i F = Δ i Fi + Δ i Fv , Δ iG = Δ iG i + Δ iG v
Eq. 11.30
Where the inviscid delta terms ΔiE, ΔiF and ΔiG are linearized using the Taylor series expansion. In
order to linearize the inviscid delta terms, we make use of the fact that E’, Fi, and Gi are function of
the vector:
 U3 
 U 2 U3 
 ρ   U1   
 U1 
 ρu   U 
   2  U 32 ~ 
+U   U 22 + U 32 + U 24 
For example, Fi =  U1
~
U =  ρv  =  U 3  
wh ere U = (γ − 1) U 5 − 
       2U 
ρw   U 4   U3U 4 
1

 E t   U 5   U1 
 ~ U 
 
 U5 + U 3 
 U1 
 E  i
i

(E) i +1 = (E) i +    U + (x)

2

 U 
Δ i E = [Q]i Δ i U
 F 
i

(Fi ) i +1 = (Fi ) i +  i  i U + (x) 2 or Δ i Fi = [R ]i Δ i U
 U  Δ iG = [S]i Δ i U
 i

 G i  i
i
i +1
(G i ) = (G i ) + 
i
  U + (x)
2

 U 
Eq. 11.31
Where [Q], [R], and [S] are the Jacobian of matrices dE’/dU, dFi/dU, and dGi/dU which are given
in357. The viscous delta terms can be linearized using a method suggested by Steger358. In order to
apply this linearization method, the coefficients of viscosity (μ) and thermal conductivity (k) are
assumed to be locally independent of U, and the cross derivative viscous terms are neglected. As the
result of these assumptions Fv and Gv have the general form of

357 Anderson, Dale A; Tannehill, John C; Plecher Richard H; 1984,”Computational Fluid Mechanics and Heat
Transfer”, Hemisphere Publishing Corporation.
358 Steger, J, L, “Implicit finite difference Simulation of the flow about Arbitrary Geometries with application to

Aitfoils”, AIAA paper 77-665, Albuquerque, NM, 1077.

 286

   5  β k  i 
fk = αk (β k ) , f i +1 = f i + α ik     Δ L  + Ο(Δx)
2
 U
y y  L =1  U L  
 i 
 5  β k  i 
  Δ U L  + Ο(Δx) 2
i +1
gk = αk (β k ) , g = g + α ki

z z  L =1  U L  Eq. 11.32

α ik is independent of U and β k is a function of U. We can write :
Δ i Fv = [V ]i Δ i U , Δ i G v = [ W ]i Δ i U
Where the [V] and [W] are the Jacobian matrices for viscous terms. We now ready to put everything
in our marching order equation (BW) and the results

 
i
  R V
  S w

i

 E  θ Δx 
  Fi Fv   G i G v 
  i
  + 1  y  U − U  + z  U − U  Δ U =
   1 + θ 2
U        
 Q    A   B  
Δx  Fi G i  θ2 i −1
−  y + z  + 1 + θ Δ E − Δ p
i

1+ θ2   2
i −1 Eq. 11.33
 p =  p + (x)
i 2

 θ Δx  [ A] [B]   i
i

or [Q]i + 1  +  Δ U = R.H.S.
 1 + θ 2  y y  

Using Appriximat e Factoriziation :
 i θ1Δx  [ A] i   i θ Δx  [B] i  i
[Q] +

  
1 + θ 2  z  
i -1 
  (Q )  [Q] + 1   Δ U = R.H.S.
1 + θ 2  y  
 

The partial derivative d/dy and d/dz are approximated by 2nd order accurate central differencing.
The algorithm is implemented in the following manner:

 θ1Δx  [ A]    θ1Δx  [B]  

 
i i
   
 (Q )
-1
[Q ]i
+  
i
 [Q] +i
  Δ i U = R.H.S.

 1 + θ 2  z    1 + θ 2  y 
 

 θ1Δx  [ A]  
i
 i
[Q] +   Δ U 1 = R.H.S.
i
Step 1 :

 1 + θ 2  z   
Step 2 : Δ U 2 = [Q] Δ U1
i i i Eq. 11.34

 θ1Δx  [B]  
i

Step 3 : [Q] +
i
  Δ i U = Δ i U 2
 1 + θ 2  y  
 
i +1
Step 4 : U =U =ΔU i i

In steps 1 and 3, the system of equations can be solved using a block tridiagonal solve. As the result
of manipulation in step 2, the inverse matrix does not have to be determined in the solution process.
 287

Finally, in step 4, the vector of unknown at station i+1 (Ui+1) is determined and the primitive variable
can be obtained in the following manner

n +1 i +1 n +1 U i2+1 n +1 U i3+1 n +1 U i4+1

ρ =U
1 , u = i +1 , v = i +1 , w = i +1 ,
U1 U1 U1
Eq. 11.35
n +1 U i5+1 (u i +1 ) 2 + (vi +1 ) 2 + (w i +1 ) 2
e = i +1 −
U1 2

11.5.2 Case Study - New-Generation Hyperbolic Navier-Stokes Schemes: Speed-Up and Accurate
Viscous/Heat Fluxes 359
We introduce a 1st order hyperbolic system model for viscous flows, and propose a unique way of
computing steady viscous flows: integrate the hyperbolic Navier-Stokes system in time toward the
steady state. We construct an upwind finite-volume scheme for the hyperbolic Navier-Stokes system
and demonstrate remarkable advantages of the resulting Navier-Stokes code: O(1/h) speed-up over
traditional Navier-Stokes codes, where h is the mesh spacing, and the capability of simultaneously
computing the viscous stresses and the heat fluxes to the same order of accuracy as that of the main
variables on irregular grids. The paper concludes with discussions on the future developments and
the potential impact on the future algorithm development for computational fluid dynamics.
Although the 2D development is in the emphasis here, for 3D, readers are encourage to consult [Li et
al.]360.
11.5.2.1 Introduction
We extend the 1st order hyperbolic system method developed for model equations in a series of
papers361-362 to the Navier-Stokes equations. We thereby propose a non-traditional way of computing
viscous flows: integrate an equivalent first-order hyperbolic system in time toward the steady state.
The first-order system is deliberately designed such that it reduces to the original Navier-Stokes
equations in the steady state; its viscous part is a hyperbolic system on its own. Navier-Stokes codes
arising from the proposed method will be radically different from those currently used.
11.5.2.2 Simplified Discretization
Robust and accurate viscous discretization is made simple because the viscous term is hyperbolic
just like the inviscid term. Methods developed for the inviscid term such as upwind fluxes and
limiters, are directly applied to the viscous term. The first-order hyperbolic system method has a
great potential for overcoming many difficulties associated with the viscous discretization
encountered particularly in unstructured grids and high-order methods (see 363-364 and references

359 Hiroaki Nishikawa, “New-Generation Hyperbolic Navier-Stokes Schemes: O(1/h) Speed-Up and Accurate
Viscous/Heat Fluxes”, 20th AIAA, June 2011, Hawaii.
360 Lingquan Li, Hong Luo, Jialin Lou and Hiroaki Nishikawa, “High-Order Hyperbolic Navier-Stokes

Reconstructed Discontinuous Galerkin Method for Unsteady Flows”, AIAA Aviation Forum, TX, 2019.
361 Nishikawa, H., A First-Order System Approach for Diffusion Equation. I: Second-Order Residual-Distribution

Schemes," Journal of Computational Physics, Vol. 227, 2007, pp. 315-352.

362 Nishikawa, H., A First-Order System Approach for Diffusion Equation. II: Unification of Advection and

Diffusion," Journal of Computational Physics, Vol. 229, 2010, pp. 3989-4016.

363 Nishikawa, H., Beyond Interface Gradient: A General Principle for Constructing Diffusion Schemes," 40th AIAA

Fluid Dynamics Conference and Exhibit, AIAA Paper 2010-5093, Chicago, 2010.
364 Nishikawa, H., Robust and Accurate Viscous Discretization via Upwind Schemes, I: Basic Principle," Computers

and Fluids, 2011, in press.

 288

therein). Note that the proposed method is different from the mixed finite-element method365 and
other first-order system methods 366-367 in that our system is hyperbolic in time while their systems
have no such characterization. It is the hyperbolicity that brings a drastic change in
the viscous discretization.
11.5.2.3 Speed-up
Because the system is hyperbolic, the explicit time step is determined by the CFL condition, leading
to an O(h) time step where h is the mesh spacing. It implies O(1/h) speed-up over traditional schemes
with an O(h2) time step. For implicit time-stepping schemes, the advantage comes in the condition
number of the linearized system to be inverted at every time step: O(1/h) versus O(1/h2). It brings
O(1/h) speed-up for iterative methods for solving the linearized system. In either case, the
acceleration factor grows for larger-scale problems. In 3D simulations, it will be O(100) for 1M grid
points, and O(1000) for 1 Billion grid points. Such an orders-of-magnitude improvement in the
algorithm will not only allow us to overcome the current hardware limit, but also bring a
continuously-growing advantage for larger-scale problems along with the increase in computing
power.
11.5.2.4 Accurate Viscous/Heat Fluxes
The first-order hyperbolic system method generates a new class of Navier-Stokes codes that are
capable of computing the viscous stresses and the heat fluxes simultaneously with the main ow
variables to the same order of accuracy on irregular grids. The use of irregular grids is common in
practical applications due to complex geometries; the irregularity is hard to avoid particularly once
we start adapting grids for viscous flows.
On such grids, the physical quantities sought by the viscous simulation, such as the viscous stresses
and the heat fluxes, can be obtained only with a lower order of accuracy, typically one order less than
that of the main flow variables. They also often exhibit erroneous behavior. For example, current
state-of-the-art Navier-Stokes codes are known to produce erratic viscous stress and heating
distributions368-369. The proposed method has a great potential for overcoming these difficulties.
Also, it can be extended to produce accurate vorticity as well as the viscous stresses and the heat
fluxes. The core of the proposed method lies in the construction of a hyperbolic model for viscous
flows. Therefore, it is fully compatible with virtually any numerical method (e.g., finite-
volume/element, residual-distribution, and other modern higher-order methods such as
discontinuous Galerkin and spectral-volume/difference methods).
Many advantages such as those described above are intrinsic properties of the proposed method.
Here, we present an example of such a hyperbolic model for the compressible Navier-Stokes
equations, and show how it can be discretized by the finite-volume method. Presented here
preliminary results obtained by the resulting Navier-Stokes code, and demonstrate the O(1/h) speed-
up and accurate viscous/heat fluxes on irregular grids.
11.5.2.5 First-Order 2D Hyperbolic Navier-Stokes System
Consider the 2D compressible Navier-Stokes equations:

365 Girault, V. and Raviart, P. A., An analysis of a mixed finite element method for the Navier-Stokes equations,"
Numerische Mathematik, Vol. 33, 1979, pp. 235-271.
366 Heys, J., Lee, E., Manteuffel, T., and McCormick, S., An alternative least-squares formulation of the Navier-

Stokes equations with improved mass conservation," Journal of Computational Physics, Vol. 226, 2008.
367 Jiang, B.-N., The Least-Squares Finite Element Method, Springer, 1998.
368 Gnoffo, P. A., Multi-Dimensional, Inviscid Flux Reconstruction for Simulation of Hypersonic Heating on

Tetrahedral Grids," AIAA Paper 2009-599, 2009.

369 Kitamura, K., Nakamura, Y., and Shima, E., An Evaluation of Euler Fluxes II: Hypersonic Surface Heating

Computation," 38th AIAA Fluid Dynamics Conference and Exhibit, AIAA Paper 2009-3648, Seattle, 2008.
 289

∂ρ ∂(ρu) ∂(ρv)
+ + =0
∂t ∂x ∂y
∂(ρu) ∂(ρu2 + p − τ𝑥𝑥 ) ∂(ρuv − τ𝑦𝑥 )
+ + =0
∂t ∂x ∂y
∂(ρv) ∂(ρuv − τxy ) ∂(ρv 2 + p − τyy )
+ + =0
∂t ∂x ∂y
∂(ρE) ∂(ρuH − τxx u − τxy v + qx ) ∂(ρuH − τyx u − τyy v + qy )
+ + =0
∂t ∂x ∂y
Eq. 11.36
E is the specific total energy, and H = E +p/ρ is the specific total enthalpy. Also, τxx, τxy, and τyy are the
viscous stresses, and qx and qy are the heat fluxes defined as

2 ∂u ∂v ∂u ∂v 2 ∂u ∂v
τxx = μ (2 − ) , τ𝑥𝑦 = τyx = μ ( − ) , τyx = μ (2 − )
3 ∂x ∂y ∂y ∂x 3 ∂x ∂y
μ ∂T μ ∂T
qx = ( ) , qy = ( )
Pr(γ − 1) ∂x Pr(γ − 1) ∂y
Eq. 11.37
We construct a first-order Navier-Stokes system by replacing Eq. 11.37 by the following evolution
equations for the viscous stresses and the heat fluxes:

𝜕𝜏𝑥𝑥 μ𝑣 ∂u 1 ∂v 𝜏𝑥𝑥
= ( − − )
𝜕𝑡 T𝑣 ∂x 2 ∂y 𝜇𝑣
𝜕𝜏𝑥𝑦 μ𝑣 3 ∂u 3 ∂v 𝜏𝑥𝑦
= ( − − )
𝜕𝑡 T𝑣 4 ∂y 4 ∂x 𝜇𝑣
𝜕𝜏𝑦𝑦 μ𝑣 ∂v 1 ∂u 𝜏𝑦𝑦
= ( − − )
𝜕𝑡 T𝑣 ∂y 2 ∂x 𝜇𝑣
∂qx μv 1 ∂T qx
= (− − )
∂t Tv γ(γ − 1) ∂x μh
∂qy μv 1 ∂T qy
= (− − )
∂t Tv γ(γ − 1) ∂y μh
Eq. 11.38
Where μv and μh are the scaled viscosities,

4 γμ ρL2 ρL2
μv = μ , μ = , Tv = , Tℎ =
3 Pr μv μh
Eq. 11.39
Tv and Th are relaxation times associated with the viscous stress and the heat flux, and L is a length
scale of O(1) defined as suggested in370. Note that the first-order system has been deliberately

370 Nishikawa, H., A First-Order System Approach for Diffusion Equation. II: Unification of Advection and
Diffusion," Journal of Computational Physics, Vol. 229, 2010, pp. 3989-4016.
 290

constructed such that it reduces to the original Navier-Stokes system in the steady state for arbitrary
Tv and Th. Traditionally, the Navier-Stokes system is discretized in two steps: the inviscid term is
discretized by a method suitable for hyperbolic systems (e.g., upwind differencing) followed by the
discretization of the viscous term by a method suitable for parabolic equations (e.g., central
differencing). We challenge the tradition by proposing the following first-order system model. As in
one dimension, we cast the first-order system in the form of a preconditioned conservative system:

∂𝐔 ∂𝐅 ∂𝐆
𝐏 −1 + + =𝐒
∂t ∂x ∂y
Eq. 11.40
Where
ρu
ρu2 + p − τxx
ρ
ρu ρv 2 − τxy
ρv ρuH − τxx u − τxy v + qx
ρE −u
−3
𝐔 = τxx , 𝐅= v ,
τxy 4
u
τyy
qx 2
a2
[ qy ]
γ(γ − 1)
[ 0 ]
ρv
ρuv − τxy
ρv 2 + p − τyy
ρuH − τxy u − τyy v + qy
−v
𝐆= −3
u
4
−v
0
a2
[ γ(γ − 1) ]
 291

0 1 0 0 0 0 0 0 0 0
0 0 1 0 0 0 0 0 0 0
0 0 0 1 0 0 0 0 0 0
0 0 0 0 1 0 0 0 0 0
𝐏 = τxx /μv , 𝐒 = 0 0 0 0 Tv /μv 0 0 0 0
τxy /μv 0 0 0 0 0 Tv /μv 0 0 0
τyy /μv 0 0 0 0 0 0 Tv /μv 0 0
qx /μh 0 0 0 0 0 0 0 Th /μh 0
[ qy /μh ] [0 0 0 0 0 0 0 0 Th /μh ]
Eq. 11.41
The wave structure of the system can be analyzed by the Jacobian matrix projected along an arbitrary
vector, (nx; ny):
∂𝐅 ∂𝐆
𝐏𝐀𝐧 = 𝐏 ( nx + n )
∂𝐔 ∂𝐔 y
Eq. 11.42
Again, we split the Jacobian into the inviscid and viscous parts:

𝐏𝐀𝐧 = 𝐏𝐀𝐢𝐧 + 𝐏𝐀𝐯𝐧

Eq. 11.43
As Ain and Avn are the projected Jacobians derived from the inviscid and viscous fluxes, respectively.
The Eigen-structure of the inviscid Jacobian is well known: the eigenvalues are

λ1i = u𝑛 − a , λi2 = u𝑛 , λi3 = u𝑛 , λi4 = u𝑛 − a , λi5,6,7,8 = 0

Eq. 11.44
where un = unx + vny. The corresponding linearly independent right-eigenvectors are denoted by Ri,k
, k = 1, 2, 3, , , , 9. The first four eigenvectors represent the inviscid acoustic, shear, and entropy waves,
constituting the inviscid subspaces. The other eigenvectors, Ri k, k = 5, 6, 7, 8, 9, are the vectors having
1 in the k-th component and 0 elsewhere. The viscous Jacobian has the following eigenvalues,

λ1v = −anv , λv2 = anv , λv3 = −amv , λv4 = amv , λv5 = −ah ,
λv6 = ah , λv7,8,9 = 0
Eq. 11.45
Where

νv 3νv νh
anv = √ , amv = √ , λv4 , ah = √
Tv 4Tv Th
Eq. 11.46
The speed, anv, is associated with the normal viscous stress; it is called the normal viscous wave. On
the other hand, amv is associated with the shear viscous stress; it is called the shear viscous wave. As
in one dimension, ah is the speed of the heating wave. The corresponding linearly independent right-
eigenvectors can be found and denoted by Rik, k = 1, 2, 3, , , , , 9. Thus, the viscous part is a hyperbolic
system by itself, describing the isotropic normal/shear viscous and heating waves. In both the
inviscid and viscous parts, the corresponding left-eigenvectors denoted by Li k and Lvk, k = 1, 2, , , , 9
 292

can be found. Then, the full Jacobian can be expressed as a sum of 10 (not necessarily orthogonal)
subspaces:
4 6

𝐏𝐀 = ∑ λik ∏ik + ∑ λvk ∏vk

k=1 k=1
Eq. 11.47
where Πik and Πvk are the pure inviscid and viscous projection matrices,

∏ik = 𝐑ik 𝐋ik , ∏vk = 𝐑vk 𝐋vk

Eq. 11.48
As in one dimension, the subspaces are orthogonal only within the inviscid and viscous parts, not
orthogonal across them.
11.5.2.6 Discretization
Given a computational grid composed of quadrilaterals and/or triangles, we store the solution data
at nodes, and discretize the hyperbolic Navier-Stokes system by the edge-based finite-volume
method:
j
d𝐔𝐣 1 1
𝐏j−1 = − ∑ 𝚽jk 𝐀jk + ∫ 𝐒dv
dt Vj Vj
k=1 Ωj
Eq. 11.49
where Uj is the solution vector at the node j,
Vj is the median dual volume, Kj is a set of
neighbors of the node j, Φjk is a numerical
flux along the directed area vector (see
Figure 11.5),

𝐧𝐣𝐤 = 𝐧𝐋𝐣𝐤 + 𝐧𝐑𝐣𝐤

Eq. 11.50
defined at the midpoint of the edge, and Ajk
is the magnitude of the directed area vector
(i.e., Ajk =∣ njk∣). For second-order accuracy,
we compute the solution gradients by the
unweighted least-squares method at nodes,
and reconstruct a linear solution within
each dual control volume. On the boundary,
a suitable boundary flux is applied with the Figure 11.5 Dual control volume for node-centered
linearity-preserving quadrature formulas finite-volume schemes with unit normal associated
(see Appendices of Ref.371 for a with an edge {j , k}.
comprehensive list of linearity-preserving
quadrature formulas in both two and three dimensions). We define the numerical flux as an upwind
flux of the form:

371Nishikawa, H., Beyond Interface Gradient: A General Principle for Constructing Diffusion Schemes," 40th AIAA
Fluid Dynamics Conference and Exhibit, AIAA Paper 2010-5093, Chicago, 2010.
 293

1 1
𝚽jk = [𝐇jk (𝐔R ) + 𝐇jk (𝐔L )] − 𝐏 −1 |𝐏𝐀𝐧 | Δ𝐔
2 2
Eq. 11.51
where ΔU = UR - UL, and UR and UL are the extrapolated solution vectors at the midpoint of the edge,
{j , k}, Hjk is the physical flux projected along the directed area vector,

𝒏𝑗𝑘
𝐇jk = [𝐅, 𝐆]. 𝐧
̂𝑗,𝑘 , ̂𝒋𝒌 =
𝐧
|𝒏𝑗𝑘 |
Eq. 11.52
and An = ∂Hjk/∂U. The dissipation matrix is again computed by the simple approximation:

4 6

𝐏𝐀𝒏 ≈ ∑|λik |∏ik + ∑|λvk | ∏vk

k=1 k=1
Eq. 11.53
The interface quantities are computed similarly as in one dimension; and thus the resulting
numerical flux will reduce to the Roe flux in the inviscid limit. Also, we simplified the viscous part of
the dissipation term and implemented it as a single vector. Note that the resulting numerical flux is
upwind for all Reynolds numbers. The Number of Nodes Iteration semi-discrete equation is then
integrated in time by the forward-Euler time-stepping scheme. The global time-step Δt, is defined as
the minimum of the local time-step, Δtj , restricted by the local CFL condition:

2Vj
∆t j = CFL , CFL ≤ 1.0
∑k∈{Kj } 𝐀jk (|un | + a + ah )j
Eq. 11.54
Again, the time step is O(h), not
O(h2); it is an intrinsic property of
numerical schemes solving the
hyperbolic Navier-Stokes system.
11.5.2.7 Results
Preliminary results are available
for a viscous shock-structure
problem whose exact solution can
be obtained by numerically
solving a pair of ordinary
differential equations for the
velocity and the temperature372. Figure 11.6 Irregular Triangular Grid for the Viscous Shock-
The program used to generate the Structure
exact solution in this study can be
downloaded at http://www.cfdbooks.com/cfdcodes.html. In all computations, we take M1 = 3.5, Pr
= 3/4, γ = 1.4, Re∞ = 25, and T∞ = 400 [k]. We consider the 1D viscous shock-structure problem in a
2D rectangular domain. Results are available for irregular triangular grids generated from 21x5,
41x9, 61x13, 81x17, 101x21 regular grids by random perturbation, random diagonal splitting, and
stretching. In each grid, the nodes are clustered over the viscous shock as shown in Figure 11.6.

372 Xu, K., A

Gas-Kinetic BGK Scheme for the Navier-Stokes Equations and Its Connection with Artificial Dissipation
and Godunov Method," Journal of Computational Physics, Vol. 171, 2001, pp. 289-335.
 294

Again, the exact solution is used as the initial solution. Also, a similar internal pressure condition and
boundary conditions are applied as in one dimension. Time integration is performed with CFL= 0.99
until the divided residual is reduced by six orders of magnitude in the L1 norm. The hyperbolic
Navier-Stokes scheme is compared with a traditional Navier-Stokes scheme based on the Roe flux for
the inviscid term and the face-tangent average-least-squares scheme for the viscous term.
Figure 11.7 shows the convergence results for the hyperbolic Navier-Stokes scheme and the
traditional Navier-Stokes scheme. As clearly seen in Figure 11.7 (a), the traditional scheme takes
orders of magnitude larger number of iterations to reach the steady state than the hyperbolic scheme.
As predicted, and confirmed by Figure 11.7 (b), it increases linearly with the total number of nodes,
denoted by N, for the traditional scheme while it is proportional to the square root of the number of
nodes for the hyperbolic Navier-Stokes scheme. Note again that the speed-up factor grows with the
number of nodes. The hyperbolic scheme gets faster and faster than the traditional scheme for larger-
scale problems.

Figure 11.7 Comparison of Convergence for the 2D Problem. Stars: the Traditional Scheme. Circles:
the Hyperbolic Scheme

Error convergence results are given in Figure 11.8. For the traditional scheme, the viscous stress
and the heat flux were computed once at the end of the iterations by the unweighted least-squares
reconstruction. The traditional scheme has some irregularity in the error convergence. It is
considered as a result of the mesh irregularity; we observed a uniform second order convergence for
different sets of grids. Figure 11.8 (a-b) shows the error convergence in the viscous stresses and
the heat fluxes (τyy is equivalent to τxx, and thus the result is not shown). As expected, the hyperbolic
scheme achieved second-order accuracy in these quantities. The reconstructed values for the
traditional scheme are only 1st order accurate. This difference in the order of error convergence
implies orders of magnitude smaller errors in further grid refinement. We emphasize that this
 295

superior accuracy in the viscous stresses and the heat fluxes comes with the O(1/h) faster iterative
convergence. Even if second-order accuracy is achieved by high-order or implicit reconstruction for
the traditional scheme, it remains extremely slower than the hyperbolic scheme.

Figure 11.8 Error Convergence Results for the Viscous Stresses and the Heat Fluxes in the 2D problem

11.5.2.8 Concluding Remarks and Future Developments

We have introduced a first-order hyperbolic Navier-Stokes system to compute the steady state
solution of the compressible Navier-Stokes equations. A finite-volume scheme has been constructed
for the hyperbolic system based on an upwind flux. Preliminary results demonstrated the advantages
of the resulting Navier-Stokes code over traditional Navier-Stokes codes: the O(1/h) speed-up by the
explicit time-stepping scheme and the second-order accuracy of the viscous stresses and the heat
fluxes on fully irregular unstructured grids.
For a simple and systematic discretization, we have introduced a preconditioned conservative form
of the hyperbolic Navier-Stokes system. This particular approach greatly simplifies the analysis of
the wave structure of the system by avoiding complications arising from the nonlinearity. It also
paves the way for further improvements by setting a stage for the local preconditioning technique to
ensure the accuracy in the low-Mach number limit as well as to accelerate the convergence to the
 296

steady state373. If the full Eigen-structure were available for the hyperbolic system in the steady form,
the optimal local preconditioning matrix could be constructed for the whole system in one and two
dimensions374. This preconditioned-form approach is a general approach applicable to other
nonlinear equations. It makes the proposed method available to a wide range of practical
applications. Also, we have introduced an approximate evaluation of the upwind dissipation matrix:
the full absolute Jacobian is approximated by a sum of the inviscid and viscous absolute Jacobians.
This approximate approach simplifies the construction of the interface flux for the hyperbolic Navier-
Stokes system. Although the approach has been found successful for the problem considered here, it
may be more sensible to weight them by the Reynolds number as suggested in Ref. 375. Yet, we may
simply view it as a sum of the Roe inviscid flux and the upwind viscous flux, and explore other choices
for each ux. For example, we can employ a more robust inviscid flux such as the Rotated-RHLL flux376
for applications involving strong shocks.
For flows with shock waves, a mechanism to prevent oscillations such as a limiter needs to be
incorporated, perhaps, not only for the main flow variables but also for the extra variables. It can be
considered as introducing a means to control the nonlinear stability such as the positivity of the
viscous discretization. Positivity is considered as a very important property for the viscous
discretization, but it remains difficult to devise a positive viscous discretization on general
unstructured grids. In the first-order hyperbolic system method, if we have a positive scheme for the
inviscid term, we almost immediately have a positive scheme for the viscous term.
The development of shock-capturing hyperbolic Naiver-Stokes schemes is underway. In practical
applications, the computational grid is typically highly-stretched for resolving viscous layers. For
such grids, it would be desired to have implicit time-stepping schemes available. In this case, the
residual Jacobian, which needs to be inverted at every time step, has an O(h) times smaller condition
number, and thus we expect O(1/h) faster iterative convergence. The development of implicit time-
stepping schemes is a very important subject of future work. Eventually, we will extend the method
to time-accurate simulations by employing the dual-time stepping technique where the Navier-
Stokes code arising from the proposed method can be used in the inner iteration.
A rapid convergence is expected over each physical time step. It should be noted that the first-order
hyperbolic system for the Navier-Stokes equations is by no means unique. There are many other
possible choices. For example, an alternative system can be constructed such that it leads to not only
accurate viscous stresses and heat fluxes but also accurate vorticity on unstructured grids. The
development of such alternative systems is currently underway.
The first-order hyperbolic system method is a general method applicable to various partial
differential equations involving second and higher order derivatives. As shown, O(1/h) times faster
iterative methods can be constructed for the Laplace/Poisson type equations with accurate solution
gradients simultaneously computed. Applications of the resulting fast elliptic solvers include the
elliptic grid generation/adaptation, the Poisson equation for the pressure in the incompressible
Navier-Stokes equations, etc. In Ref.377, the method was extended to the advection-diffusion equation,
and in particular a uniform iterative convergence over a wide range of Reynolds numbers was
demonstrated for boundary-layer type problems on stretched grids. The resulting hyperbolic

373 Nishikawa, H., Roe, P., Suzuki, Y., and van Leer, B., A General Theory of Local Preconditioning and Its
Application to the 2D Ideal MHD Equations," 16th AIAA Computational Fluid Dynamics Conference, AIAA Paper
2003-3704, Orlando, 2003.
374 See Previous.
375 Nishikawa, H., A First-Order System Approach for Diffusion Equation. II: Unification of Advection and

Diffusion," Journal of Computational Physics, Vol. 229, 2010, pp. 3989-4016.

376 Nishikawa, H. and Kitamura, K., Very Simple, Carbuncle-Free, Boundary-Layer Resolving, Rotated-Hybrid

Riemann Solvers," Journal of Computational Physics, Vol. 227, 2007, pp. 2560-2581.
377 Nishikawa, H., A First-Order System Approach for Diffusion Equation. II: Unification of Advection and

Diffusion," Journal of Computational Physics, Vol. 229, 2010, pp. 3989-4016.

 297

advection-diffusion scheme can be readily applied, for example, to scalar turbulence model
equations.
Application to higher-order derivative terms is also possible: construct an extended 1st order
hyperbolic system and solve it by an upwind scheme. That is, partial differential equations of
arbitrary order can be discretized by methods for hyperbolic systems, and it leads to orders of
magnitude more efficient and accurate solvers. Finally, the method has been extended to nonlinear
systems in this paper. Opportunities for practical applications are now wide open: incompressible
Navier-Stokes equations, thermal flow and heat transfer phenomena in nuclear applications,
resistive magneto-hydrodynamic simulations, ground-water simulations in hydro-geology, chemical
diffusion, and so on. The impact of the proposed method on the future CFD development is expected
to grow larger as the grid size gets larger and the geometry gets more complex.

11.6 Multigrid Methods

Multigrid methods (MGM) are effective techniques that can deliver fast convergence rates with
minimal memory overheads. The basic idea of a multigrid method is to accelerate the solution of a
set of fine grid equations by time stepping on a sequence of fine and coarse grids using a simple
explicit scheme. Multigrid methods capitalize on the effectiveness of explicit methods in reducing
high frequency error modes. A high-frequency error can be thought of as a solution component that
oscillates between neighboring grid points. While explicit methods are effective at eliminating such
errors, they are relatively ineffective at reducing low-frequency or global error components. This is
to be expected, because explicit schemes rely exclusively on local information supplied through the
residual stencil.
The multigrid strategy consists in explicitly time stepping the fine grid equations until the high-
frequency errors have been conquered. The solution is then transferred to a coarser grid, where the
remaining low-frequency errors now manifest themselves as high-frequency errors and can be
effectively damped by the same explicit method. The process is repeated recursively through a
complete sequence of grids, where each level is responsible for a particular bandwidth of errors.
When the coarsest grid of the sequence is reached, the corrections are interpolated back to each
successively finer grid. This entire multigrid cycle is then repeated until overall convergence is
achieved. For structured grids, simply get rid of every other line of grid and make sure that grid is
sufficient for coursing. But for unstructured meshes, the issue is more complicated. Given a fine
unstructured mesh, there is no simple procedure for constructing uniformly coarser levels, as in the
structured-grid case. One of the drawbacks of this approach is that it still requires the user to
generate multiple grids for a single solution, which is at best tedious. With this in mind, several efforts
have been pursued to automate the process. But first we consider the typical (i.e. V &W) cycle which
is the essence of the multigrid process.
11.6.1 Multigrid Cycle & Definitions
Usually iterative methods would reduce the high-frequency (i.e., oscillatory) errors but fails to reduce
the low-frequency (i.e., smooth) errors thus will results poor rate of convergence378. High-frequency
and low-frequency are the names of the errors that are referring to the error in the coarser and finer
mesh. In the case of the iterative methods, the solution in the finer grid looks smoother but if we
transfer the solution to the coarser grid it will be oscillating. To solve this problem we need to do
sampling of the solution from finer grid to coarser grid where a few iterations (iterative methods for
example, Gauss-Seidel) need to be done to reduce the high-frequency error.
All multigrid methods require a definition of a succession of coarse grids, based on the original "fine"
grid. The process of defining the coarser grids involves what is called Agglomeration, i.e. the
combination of several nodes or control volumes or coefficients from the original grid.

378 Ezhilmathi, “Magic behind the Most of the CFD solvers for HPC “, Scientific Computing blog, 2013.
 298

Transferring the solution from finer grid to coarser grid is called Restriction. After the few steps
of the restriction are done, the error of high-frequency will be reduced. Later this solution is
transferred to finer grid for further calculation and by then the low-frequency error is also
minimized. Transferring the solution from coarser grid to finer grid is called the Prolongation
or Interpolation, (see Figure 11.9).
On the finer grid it is suggested to do a few more iterations to keep the high-frequency error still
small. So the solution that is in the finer grid will be having less high-frequency and less low-
frequency error. These cycles have to be continued until the solution meets the desired convergence
criteria. Prolongation, restriction and less iteration in the finer grid gives a faster convergence rate
and quicker solution than the normal stationary iterative methods379. An investigation into a Multi-
gridding algorithm by [Gargoloff]380 shows that the maximum value of the residuals for a typical
multigrid solver were reduced much faster than the maximum value of the residuals for the one-level
grid solver. The 2D case was run for a NACA 0012 airfoil. Further discussion regarding smoother
terminology and their usage can be obtained from [Birken et al.]381.

Figure 11.9 Multi-Gridding Cycle

11.6.2 Classification382
Multigrid methods (MGM) are classified into two branches
➢ Geometric multigrid - FAS
➢ Algebraic multigrid - AMG
In the Geometric Multigrid (GMG), agglomeration of the nodes (cells, elements, or control volumes)
takes place on the geometric level, and a set of new data structures representing the coarse grids
need to be constructed for each level. This method is usually difficult to deal with when using the
finite volume method since the FVM is based on a cell centered discretization which makes it difficult
to define irregularly shaped control volumes. On the other hand, the Algebraic Multigrid (AMG)
performs the agglomeration using the coefficient matrix only (hence the name algebraic). In the AMG
method, specially chosen coefficients of the matrix are combined together to form a new coefficient
matrix representing the coarser grid and no new discretization is required on the coarse grids.
Agglomeration techniques could vary from algorithm to algorithm. It is recommended that (AMG) is
best suited for unstructured mesh, while (GMG) is appropriate for structure mesh. Agglomeration

379 Same as above.

380 A Dissertation by Joaquin Ivan Gargoloff, “A Numerical Method For Fully Nonlinear Aero-elastic Analysis”,
Submitted to the Office of Graduate Studies of Texas A&M University in partial fulfillment of the requirements
for the degree of DOCTOR OF PHILOSOPHY, 2007.
381 Philipp Birken, J onathan Bull, and Antony Jameson, “A note on terminology in multigrid methods”, Proc. Appl.

Math. Mech. 16, 721 – 722 (2016).

382 CFD online.com
 299

techniques could vary from algorithm to algorithm. For that, see different sources such as [Ghia ]383.
11.6.3 Unstructured Mesh Cycling
Methods that remove selected
mesh points, using graph-based
algorithms, and re-triangulate the
remaining subset of vertices have
been demonstrated as effective
tools for generating coarse-level
Methods that remove selected
mesh points, using graph-based
algorithms, and re-triangulate the
remaining subset of vertices have
)been demonstrated as effective
tools for generating coarse-level
multigrid meshes. However, for
very complex geometries, the
generation of coarse meshes that
conform to the original geometry
either manually or automatically,
becomes an unpleasant task. A
usual and tedious approach to
unstructured multigrid methods is
to generate a sequence of
completely independent coarse
and fine meshes and use linear
interpolation to transfer
variables back and forth between
the various meshes of the
sequence, within a multigrid cycle
384-385. The meshes may be
generated using any feasible grid
generation technique and will
generally be non-nested, and may
even not contain any common
points.
The only requirement is that they Figure 11.10 Sequence of gridding in unstructured multigrid
scheme
conform to the same domain
boundaries. This technique is more flexible than the nested subdivision approach, since the fine and
coarse meshes are not constrained, and may be optimized independently for accuracy and speed of
convergence respectively. Furthermore, this approach can be applied to a problem with a pre-
specified fine mesh. The inter grid transfer operators can be determined in a preprocessing operation
but require smart search techniques to determine the patterns between coarse and fine grid
elements, which in principle may overlap randomly. Once these have been determined and stored,

383 U. Ghia, K. N. Ghia, and C. T. Shin, “High-Re Solutions for Incompressible Flow Using the Navier-Stokes
Equations and a Multigrid Method, Journal of Computational Physics 48, 387-411 (1982).
384 Mavriplis DJ., “Three-dimensional multigrid for the Euler equations”, AIAA , 1992.
385 Leclercq MP., “Resolution des equations d’Euler par des methods multigrilles conditions aux limites en regime

hypersonique”, PhD thesis. Dep. Appl. Math, Univ. de Saint- Etienne.

 300

grid transfers are simply implemented as a weighted gather and scatter of data between coarse and
fine grid arrays.
Application of this technique to a three-dimensional inviscid flow is depicted in Figure 11.10 where
two grids from the sequence of grids employed in the overset mesh multigrid scheme for the
computation of transonic flow over an aircraft configuration. Yet another approach is provided by
agglomeration (collection) multigrid methods. Based on the control-volume formulation, these
methods form coarser-level meshes by blending or agglomerating fine-level control volumes
together with their neighbors386. The resulting coarse mesh contains a smaller number of larger and
more complex control volumes. The precise manner in which control volumes are agglomerated can
be controlled through a graph-based algorithm similar to the vertex removal procedure described
above. The flow solver must be modified to run on arbitrarily shaped control volumes on the coarse
levels. For inviscid-flow control-volume formulations, this presents little difficulty, since the
equations are generally discretized as fluxes over individual control-volume faces, which can be used
to build contour integrals about arbitrarily shaped control volumes. The observed convergence rate
of the agglomeration multigrid method is almost identical to that obtained with the overset-mesh
method.
11.6.4 Viscous Flow Consideration
For viscous flows, the discretization of diffusion terms on arbitrarily shaped control volumes is no
longer straightforward. An algebraic interpretation of agglomeration multigrid provides a
mechanism for dealing with equations sets that contain diffusion terms. Borrowing from the
algebraic multigrid literature [Ruge & St¨uben 1987], a coarse-grid operator may be constructed by
projecting the fine-grid operator onto the space spanned by the coarse-grid basis functions. In
general, the multigrid convergence rates achieved for viscous flow cases are substantially slower
than those achieved for inviscid cases. The slower convergence in the viscous flow cases is principally
due to the anisotropic stretching of the mesh in the boundary-layer and wake regions. Unstructured
multigrid methods offer the possibility of designing schemes that are insensitive to anisotropic
effects.
The graph-based agglomeration or vertex coarsening algorithms described above can be modified
to merge only those neighboring control volumes or delete only those neighboring vertices that are
the most strongly coupled to the current point. By coarsening or agglomerating only in the direction
of strongest coupling, as determined by the magnitude of the coefficients in the fine-grid discrete
equations, rather than eliminating all possible neighbors, coarse levels that are optimal for the
problem at hand can be generated. An alternate strategy for improving the convergence of multigrid
methods is to employ a stronger smoother as the base grid solver. Any of the implicit methods
described above may be substituted for an explicit method. Of course, most. There is an obvious
duality between the construction of sub regions for implicit methods and coarsening strategies for
multigrid methods that should be exploited in the future to better couple locally implicit strategies
with multigrid methods.
The above discussion on multigrid methods centers on the Full Approximation Storage (FAS)
technique, where multigrid is applied directly to the nonlinear form of the governing equations.
Multigrid may also be employed as a solver for the linear system arising at each time step of an
implicit scheme, or even as a preconditioner for an iterative solution strategy. While these strategies
may yield desirable robustness characteristics, they are plagued by the high memory requirements
of the implicit linearization, which are avoided in the FAS approach.

Lallemand M, Steve H, Dervieux A., “Unstructured multi gridding by volume agglomeration: current status”,
386

Computational. Fluids 21(3):397–433.

 301

11.6.5 Case Study - Development and Application of Agglomerated Multigrid Methods for Complex
Geometries
Authors : Hiroaki Nishikawa, Boris Diskin and James L. Thomas
Appearance : 40th Fluid Dynamics Conference and Exhibit, 28 June - 1 July 2010, Chicago, Illinois
Source : AIAA 2010-4731. http://arc.aiaa.org | DOI: 10.2514/6.2010-4731
[Nishikawa et al.]387 reports on progress been made on the development of agglomerated multigrid
techniques for fully un-structured grids in 3D. Building upon two previous studies, focused on
efficiently solving a model diffusion equation. It demonstrates a robust fully-coarsened agglomerated
multigrid technique for 3D complex geometries, incorporating the following key developments:
consistent and stable coarse-grid discretization, a hierarchical agglomeration scheme, and line-
agglomeration/relaxation using prismatic-cell discretization’s in the highly-stretched grid regions. A
significant speed-up in computer time over state-of-art single-grid computations is demonstrated for
a model diffusion problem, the Euler equations, and the Reynolds-averaged Navier-Stokes equations
for 3D realistic complex geometries.
11.6.5.1 Introduction
Multigrid techniques [1] are used to accelerate convergence of current Reynolds-Averaged Navier-
Stokes (RANS) solvers for both steady and unsteady ow solutions, particularly for structured-grid
applications. [Mavriplis et al. ][2, 3, 4, 5] pioneered agglomerated multigrid methods for large-scale
unstructured-grid applications. During the present development, a serious convergence degradation
in some of the state-of-the-art multi-grid algorithms was observed on highly-refined grids. To
investigate and overcome the difficulty, we critically studied agglomerated multigrid techniques [6,
7] for two- and three-dimensional isotropic and highly-stretched grids and developed quantitative
analysis methods and computational techniques to achieve grid-independent convergence for a
model equation representing laminar diffusion in the incompressible limit. It was found in Ref. [6]
that it is essential for grid-independent convergence to use consistent coarse-grid discretization. In
the later Ref. [7], it was found that the use of prismatic cells and line-agglomeration/relaxation is
essential for grid-independent convergence on fully-coarsened highly-stretched grids.
Here, we extend and demonstrate these techniques for inviscid and viscous flows over complex
geometries. The paper is organized as follows. Finite-volume discretization employed for target grids
are described. Details of the hierarchical agglomeration scheme are described. Elements of the
multigrid algorithm are then described, including discretization on coarse grids. Multigrid results for
complex geometries are shown for a model diffusion equation, the Euler equations, and the RANS
equations. The final section contains conclusions and recommendations for future work.

387Hiroaki Nishikawa, Boris Diskin and James L. Thomas, “Development and Application of Agglomerated
Multigrid Methods for Complex Geometries”, 40th Fluid Dynamics Conference and Exhibit 28 June - 1 July 2010,
Chicago, Illinois, AIAA 2010-4731.
 302

11.6.5.2 Discretization
The discretization method is a finite-volume discretization (FVD) centered at nodes. It is based on
the integral form of governing equations of interest:

̂) dΓ = ∬ s dΩ
∮(𝐅. 𝐧
Γ Ω
Eq. 11.55
where F is a flux tensor, s is a source term, Ω is a control
volume with boundary Г, and n^ is the outward unit
normal vector. For the model diffusion (Laplace)
equation, the boundary conditions are taken as
Dirichlet, i.e., specified from a known exact solution
over the computational boundary. Tests are performed
for a constant manufactured solution, U(x; y; z) = 10.0,
Figure 11.11 Illustration of a node-
with a randomly perturbed initial solution. For inviscid
centered median-dual control volume
flow problems, the governing equations are the Euler (shaded). Dual faces connect edge
equations. Boundary conditions are a slip-wall midpoints with primal cell centroids.
condition and inflow/outflow conditions on open Numbers 0-4 denote grid nodes
boundaries. For viscous flow problems, the governing
equations are the RANS equations with the Spalart-
Allmaras one-equation model [8]. Boundary conditions are non-slip condition on walls and
inflow/outflow conditions on open boundaries. The source term, s, is zero except for the turbulence-
model equation (see Ref. [8]).
The general FVD approach requires partitioning the domain into a set of non-overlapping control
volumes and numerically implementing Eq. 11.55 over each control volume. Node-centered
schemes define solution values at the mesh nodes. In 3D, the primal cells are tetrahedra, prisms,
hexahedra, or pyramids. The median-dual partition [9, 10] used to generate control volumes is
illustrated in Figure 11.11 for 2D. These non-overlapping control volumes cover the entire
computational domain and compose a mesh that is dual to the primal mesh.
The main target discretization of interest for the model diffusion equation and the viscous terms of
the RANS equations is obtained by the Green-Gauss scheme [11, 12], which is a widely-used viscous
discretization for node-centered schemes and is equivalent to a Galerkin finite-element discretization
for tetrahedral grids. For mixed-element cells, edge-based contributions are used to increase the h-
ellipticity of the operator [11, 12].
The inviscid terms are discretized by a standard edge-based method with unweighted least-squares
gradient reconstruction and Roe's approximate Riemann solver [13]. Limiters are not used for the
problems considered in this paper. The convection terms of the turbulence equation are discretized
with first-order accuracy.
11.6.5.3 Agglomeration Scheme
As described in the previous papers [6,7], the grids are agglomerated within a topology-preserving
framework, in which hierarchies are assigned based on connections to the computational
boundaries. Corners are identified as grid points with three or more boundary-condition-type
closures (or three or more boundary slope discontinuities). Ridges are identified as grid points with
two boundary-condition-type closures (or two boundary slope discontinuities). Valleys are
identified as grid points with a single boundary-condition-type closure. Interiors are identified as
grid points without any boundary condition. The agglomerations proceed hierarchically from seeds
within the topologies | first corners, then ridges, then valleys, and finally interiors. Rules are enforced
 303

to maintain the boundary condition types of the finer grid within the agglomerated grid. Candidate
volumes to be agglomerated are vetted against the hierarchy of the currently agglomerated volumes.
In this work, we use the rules summarized in Table 11.2.
In order to enable a valid non-degenerate stencil for linear prolongation and least-squares gradients
near boundaries [7], the rules reflect less agglomerations near boundaries than in the interior.
Corners are never agglomerated, ridges are agglomerated only with ridges, and valleys are
agglomerated only with valleys. A typical boundary agglomeration generated by the above rules is
shown in Figure 11.13.
The conditional entries denote that further inspection of the connectivity of the topology must be
considered before agglomeration is allowed. For example, a ridge can be agglomerated into an
existing ridge agglomeration if the two boundary conditions associated with each ridge are the same.
For valleys or interiors, all available neighbors are collected and then agglomerated one by one in the
order of larger number of edge-connections to a current agglomeration until the maximum threshold
of agglomerated nodes (4 for valleys; 8 for interiors) is reached. The prolongation operator P1 is
modified to prolong only from hierarchies equal or above the hierarchy of the prolonged point.
Hierarchies on each agglomerated grid are inherited from the finer grid.
For the results reported in this paper, we employ agglomeration scheme II described in previous
papers [6 , 7]. It has been modified to deal with viscous meshes using implicit-line agglomeration. It
performs the agglomeration in the following sequence:

1. Agglomerate viscous boundaries (bottom of implicit lines).

2. Agglomerate prismatic layers through the implicit lines (implicit-line agglomeration).
3. Agglomerate the rest of the boundaries.
4. Agglomerate the interior.

Table 11.2 Admissible Agglomerations

 304

The second step is a line-agglomeration step where volumes are agglomerated along implicit lines
starting from the volume directly above the boundary volume. Specifically, we first agglomerate
volumes corresponding to the second and third entries in the implicit-line lists associated with each
of the fine-grid volumes contained in a
boundary agglomerate. The line
agglomeration continues to the end of the
shortest line among the lines associated
with the boundary agglomerate. This line-
agglomeration process preserves the
boundary agglomerates. Figure 11.12
illustrates typical implicit line-
agglomeration near a curved solid body.
The implicit line-agglomeration preserves
the line structure of the fine grid on coarse
grids, so that line-relaxations can be
performed on all grids to address the grid
anisotropy. If no implicit lines are defined,
typical for inviscid grids, the first two steps
are skipped.
In each boundary agglomeration (steps 1
and 3), agglomeration begins with corners, Figure 11.13 Trailing-edge area of a 3D wing
agglomerated
Figure 11.12 byTypical
the hierarchical scheme. Primal grid
implicit line-agglomeration
creates a front list defined by collecting is
volumes adjacent to the agglomerated showing a curved solid body surface grid
shown by thin lines; agglomerated is shown
on the left andby
a
thick lines.
symmetry plane on the right. The projection of the line-
corners, and proceeds to agglomerate
agglomerations can be seen on the symmetry plane.
volumes in the list (while updating the list
 305

Figure 11.14 Grids and convergence of the model diffusion equation for the F6 wing-body
combination

as agglomeration proceeds) in the order of ridges and valleys. During the process, a volume is
 306

selected from among those in the same hierarchy that has the least number of non-agglomerated

Figure 11.15 Grids and Convergence of the Model Diffusion Equation for the DPW-W2 case
 307

neighbors, thereby reducing the occurrences of agglomerations with small numbers of volumes. A
heap data-structure is utilized to efficiently select such a volume. The agglomeration continues until

Figure 11.16 Grids and Convergence for the wing-ap inviscid case.
 308

the front list becomes empty. Finally, for both valleys and interiors, agglomerations containing only
a few volumes (typically one) are combined with other agglomerations. Figure 11.14 and Figure
11.15 show primal grids and agglomerations for the F6 wing-body combination and the DPW-W2
[14] grids. These grids are viscous grids; the primal grid has prismatic viscous layers around the body
and the wing. Coarsening ratios are indicated by r k (k = 1 ; 2 ; 3 ; 4) in the parenthesis. Line
agglomeration was applied in these regions. Figure 11.16 show primal grids and agglomerations
for a wing-body combination, a wing-ap combination, and a 3D wing with a blunt trailing edge, all
are pure-tetrahedral inviscid grids.
11.6.5.4 Single-Grid Iterations
Single-grid iteration scheme is based on the implicit formulation:

Ω ∂𝐑̂∗
( + ̂ (𝐔)
) δ𝐔 = −𝐑
∆τ ∂𝐔
Eq. 11.56
where ^R (U) is the target residual computed for the current solution U, Δτ is a pseudo-time step,
∂Ř*/∂U is an exact/approximate Jacobian, and δU is the change to be applied to the solution U. An
approximate solution to Eq. 11.56 is computed by a certain number of iterations on the linear
system (linear-sweeps). Update of U completes one nonlinear iteration. The RANS equations are
iterated in a loosely-coupled formulation, updating the turbulence variables after the mean-ow
variables at each nonlinear iteration. The left-hand-side operator of Eq. 11.56
includes an exact linearization of the viscous (diffusion) terms and a linearization of the inviscid
terms involving first-order contributions only. Thus, the iterations represent a variant of defect
correction.
Typically in single-grid FUN3D RANS applications, the first-order Jacobian corresponds to the
linearization of Van Leer's flux-vector splitting. For inviscid cases, we consider using the linearization
of Roe's approximate Riemann solver. Jacobians are updated after each iteration. The linear sweeps
performed before each nonlinear update include νp sweeps of the point multi-color Gauss-Seidel
relaxation performed through the entire domain followed by νl line-implicit sweeps in stretched
regions. The line-implicit sweeps are applied only when solving the model diffusion or the RANS
equations. In a line-implicit sweep, unknowns associated with each line are swept simultaneously by
inverting a block tridiagonal matrix [7]. For RANS simulations, νp = νl = 15 for the mean-flow
equations and νp = νl = 10 for the turbulence equation. For the model diffusion equation, only one
linear sweep is performed per nonlinear iteration, i.e., νp = νl = 1, and the exact Jacobian computed
only once at the beginning of the entire calculation. In spite of linearity of the model diffusion
equation, computations of Ř(U) in Error! Reference source not found. do not employ the exact J
acobian, thus, providing a better similarity to nonlinear computations.
11.6.5.5 Multigrid
Elements of the multigrid algorithm are presented in this section. In this study, we do not explore
various algorithmic options, relying on the methods that proved effective from the previous studies.
11.6.5.5.1 Multigrid V-Cycle
The multigrid method is based on the full-approximation scheme (FAS) [1, 15] where a coarse-grid
problem is solved/relaxed for the solution approximation. A correction, computed as the difference
between the restricted fine-grid solution and the coarse-grid solution, is prolonged to the finer grid
to update the fine-grid solution. The two-grid FAS is applied recursively through increasingly coarser
grids to define a V-cycle. A V-cycle, denoted as V (ν1 ; ν2), uses ν1 relaxations performed at each grid
before proceeding to the coarser grid and ν2 relaxations after coarse-grid correction. On the coarsest
 309

grid, relaxations are performed to bring two orders of magnitude residual reduction or until the
maximum number of relaxations, 10, is reached.
11.6.5.5.2 Inter-Grid Operators
The control volumes of each agglomerated grid are found by summing control volumes of a finer grid.
An operator that performs the summation is given by a conservative agglomeration operator, R0,
which acts on _ne-grid control volumes and maps them onto the corresponding coarse-grid control-
volumes. Any agglomerated grid can be defined, therefore, in terms of R0 as

Ωc = R 0 Ωf
Eq. 11.57
where superscripts c and f denote entities on coarser and finer grids, respectively. On the
agglomerated grids, the control volumes become geometrically more complex than their primal
counterparts and the details of the control-volume boundaries are not retained. The directed area of
a coarse-grid face separating two agglomerated control volumes, if required, is found by lumping the
directed areas of the corresponding finer-grid faces and is assigned to the virtual edge connecting the
centers of the agglomerated control volumes. Residuals on the fine grid, ^Rf , corresponding to the
integral Eq. 11.55 are restricted to the coarse grid by the conservative agglomeration operator, R0,
as
̂ ̂f
𝐑c = R 0 𝐑
Eq. 11.58
where ^Rc denotes the fine-grid residual restricted to the coarse grid. The fine-grid solution
approximation, Uf , is restricted as
𝐑 0 (𝐔f Ωf )
𝐔0c =
Ωc
Eq. 11.59
where Uc 0 denotes the fine-grid solution approximation restricted to the coarse grid. The restricted
approximation is then used to define the forcing term to the coarse-grid problem as well as to
compute the correction, (δU)c:
(δ𝐔)c = 𝐔 c − 𝐔0c
Eq. 11.60
where Uc is an updated coarse-grid solution obtained directly from the coarse-grid problem. The
correction to the finer grid is prolonged typically through the prolongation operator, P1, that is exact
for linear functions, as
(δ𝐔)f = P1 (δ𝐔)c
Eq. 11.61
The operator P1 is constructed locally using linear interpolation from a tetrahedra defined on the
coarse grid. The geometrical shape is anchored at the coarser-grid location of the agglomerate that
contains the given finer control volume. Other nearby points are found by the adjacency graph. An
enclosing simplex is sought that avoids prolongation with non-convex weights and, in situations
where multiple geometrical shapes are found, the first one encountered is used. Where no enclosing
simplex is found, the simplex with minimal non-convex weights is used.
11.6.5.5.3 Coarse-Grid Discretization
For inviscid coarse-grid discretization, a first-order edge-based scheme is employed. For the model
equation and the viscous term in the RANS equations, two classes of coarse-grid discretization were
previously studied [6, 7]: the Average-Least-Squares (Avg-LSQ) and the edge-terms-only (ETO)
schemes. The consistent Avg-LSQ schemes are constructed in two steps: first, LSQ gradients are
computed at the control volumes; then, the average of the control-volume LSQ gradients is used to
 310

approximate a gradient at the face, which is augmented with the edge-based directional contribution
to determine the gradient used in the ux. There are two variants Inviscid Viscous (Diffusion) Primal
grid Second-order edge-based reconstruction Green-Gauss Coarse grids First-order edge-based
reconstruction Face-Tangent Avg-LSQ.
Coarse grids Exact or Approximate (edge-terms only) of the Avg-LSQ scheme. One uses the average-
least-squares gradients in the direction normal to the edge (edge-normal gradient construction). The
other uses the average-least-squares gradients along the face (face-tangent gradient construction).
The ETO discretization are obtained from the Avg-LSQ schemes by taking the limit of zero Avg-LSQ
gradients. The ETO schemes are often cited as a thin-layer discretization in the literature [2, 3, 4];
they are positive schemes but are not consistent (i.e., the discrete solutions do not converge to the
exact continuous solution with consistent grid refinement) unless the grid is orthogonal [13, 16]. As
shown in the previous papers [6, 7], ETO schemes lead to deterioration of the multigrid convergence
for refined grids, and therefore are not considered in this paper. For practical applications, the face-
tangent Avg-LSQ scheme was found to be more robust than the edge-normal Avg-LSQ scheme. It
provides superior diagonal dominance in the resulting discretization [6, 7]. In this study, therefore,
we employ the face-tangent Avg-LSQ scheme as a coarse-grid discretization for the model equation
and the viscous term.
For excessively-skewed faces (over 90◦ angle between the outward face normal and the
corresponding outward edge vector), which can arise on agglomerated grids, the gradient is
computed by the Avg-LSQ scheme and edge contributions are ignored. The Galerkin coarse-grid
operator [1], which was considered in a previous study, is not considered here since the method was
found to be grid-dependent and slowed down the multigrid convergence for refined grids [6]. For
inviscid discretization, we employ a first-order edge-based discretization on coarse grids. Table 11.3
shows a summary of discretization used.

Table 11.3 Summary of discretization used to define the residual, ^R

11.6.5.5.4 Relaxations
Relaxation scheme is similar to the single-grid iteration described in 11.6.5.4 with the following
important differences. On coarse grids, the Avg-LSQ scheme used for viscous terms has a larger
stencil than the Green-Gauss scheme implemented on the target grid and its exact linearization has
not been used; instead relaxation of the Avg-LSQ scheme relies on an approximate linearization,
which consists of edge terms only. For inviscid cases, the first-order Jacobian is constructed based on
Roe's approximate Riemann solver, and thus it is exact on coarse grids where the first-order scheme
is used for the residual. For RANS cases, the first-order Jacobian is constructed based on Van Leer's
flux-vector splitting, but the inviscid part of the residual is computed by Roe's approximate Riemann
solver. Therefore, the Jacobian is approximate on both the primal and coarse grids. Table 11.5
summarizes the Jacobians used for inviscid and viscous (diffusion) terms on the primal and coarse
grids. In multigrid nonlinear applications, Jacobians are evaluated at the beginning of a cycle and
frozen during the cycle. For inviscid and RANS flow simulations, significantly fewer linear sweeps are
used in a multigrid relaxation than in a single-grid iteration: νp = νl = 5 for both the mean ow and
turbulence relaxations. For the model diffusion equation, still only one sweep is performed per
relaxation.
 311

11.6.5.5.5 Cost of Multigrid V-Cycle

All of the computations in the paper use FAS multigrid. For the linear model diffusion equation, the
computer time would be reduced if the corresponding correction scheme (CS) cycle is used. To
estimate relative cost of multigrid cycles in comparison with single-grid iterations, the cost of
nonlinear residual valuations, relaxation updates, and Jacobian evaluations needs to be taken into
account. Suppose that a nonlinear relaxation and a Jacobian evaluation cost σ and J times a nonlinear
residual evaluation, respectively. Then, the cost of a single-grid iteration relative to the cost of a
nonlinear residual evaluation is given by

W SG = σSG + J
Eq. 11.62
where the superscript SG denotes single-grid iterations. On the other hand, a multigrid cycle involves
ν1 + ν2 nonlinear relaxations, a nonlinear residual evaluation before restriction, and a Jacobian
evaluation per cycle per grid. A residual evaluation on coarse grids is also required to form the FAS
forcing term. The cost of a multigrid cycle, MG, relative to the cost of a fine-grid nonlinear residual
evaluation is given by

W MG = [C(ν1 + ν2 )σMG + J + 1] + C − 1
Eq. 11.63
where C is a coarse-grid factor,

1 1 1
C= + + +⋯
r1 r1 r2 r1 r2 r3
Eq. 11.64
Here, rk is the agglomeration ratio of the k-th agglomerated grid. The relative cost, WMGSG , of a V -cycle
is therefore given by

MG
W MG
WSG = SG
W
Eq. 11.65
Table 11.4 shows
values of WMGSG , σ and J
for each equation set
within the single-grid Table 11.4 Summary of costs and typical numbers of linear-sweeps. The
iteration and the multigrid cycle is a 5-level V (2; 1) with a typical coarsening ratio 8. The
multigrid method. The numbers in parenthesis denote the number of point and line sweeps,
values for σ and J are respectively, and the second set for RANS denotes the number of point and
line sweeps of the turbulent equation.
based on measured
computer times
associated with residual evaluation, Jacobian evaluation, and linear-sweeps on the primal grid for

Table 11.5 Summary of Jacobians, ∂^R*/∂U

 312

particular configurations. The corresponding values on a per node basis vary from the tabulated
values on the coarser grids and across configurations. Thus, Eq. 11.65 serves as a reasonable
approximation to the expected code performance. Note that the Jacobian computation has been
ignored for the model diffusion equation. This is because the Jacobian is constant for the linear
problem and therefore it is computed only once and never updated. Observe also that σ is much
smaller in the multigrid cycle than in the single-grid iteration for the nonlinear cases. This saving
comes from much fewer linear-sweeps in the multigrid method. We experimentally found that the
multigrid convergence did not depend heavily on the number of linear-sweeps. Increasing them
further does not reduce the number of cycles for convergence, but it merely increases the CPU time.
The numbers of linear-sweeps shown for the single-grid method are typical numbers considered
sufficient for robust computations with a reasonably large CFL number. The relative cost (Eq. 11.65)
computed based on the measured σ and J are shown in the third column of the table. Considering a
5-level V (2 ; 1) cycle with a coarsening ration of 8, the relative cost is found to be 4 : 5 for the diffusion
equation, 1: 8 for the inviscid equation, and 1 : 5 for the RANS equations.
11.6.5.6 Results for Complex Geometries
All calculations presented in this paper were performed with a single processor. Parallelization of
the multigrid algorithm is currently underway. The multigrid cycle is a 5-level V (2 ; 1) for all cases.
11.6.5.6.1 Model Diffusion Equation
The multigrid method was applied on grids generated for two practical geometries: the F6 wing-body
and the DPW-W2 wing-alone cases [14]. Both grids are tetrahedral, but prisms are used in a highly-
stretched viscous layer near the solid boundary. Pyramidal cells are also present around the
transitional region. The multigrid V (2; 1) cycle is applied and compared with single-grid iterations.

Table 11.6 Cost of V-cycle relative to a single-grid iteration and speed-up factor. The expected speed-
up factors have been computed with the actual coarsening ratio

The CFL number is set to infinity. For the F6 wing-body grid (1,121,301 nodes), the grids and
convergence results are shown in Figure 11.14. The speed-up factor is 63 in CPU time. A similar
result was obtained for the DPW-W2 grid (1,893,661 nodes) as shown in Figure 11.15. The speed-
up factor is nearly 22 in this case. The cost of one V -cycle computed according to Eq. 11.65 with
actual coarsening ratios is shown for each case in the fourth column of Error! Reference source not f
ound.. It shows that one V-cycle costs nearly 4 single-grid iterations. The fifth column is an expected
speed-up factor based on the number of single-grid iterations (NSG), the number of multigrid cycles
(NMG), and the factor WMGSG
MG
N𝑆𝐺 ⁄NMG WSG
Eq. 11.66
 313

The last column is the actual speed-up factor computed as a ratio of the total single-grid CPU time to
the total multigrid CPU time. A fairly good agreement can be observed between the expected and
actual speed-up factors.

Table 11.7 Summary of grid sizes and parameters for the inviscid cases

11.6.5.6.2 Inviscid Flows

The multigrid method was applied to three inviscid cases: low-speed subsonic flows over a wing-
body configuration, a wing-ap configuration, and a NACA15 wing with a blunt trailing edge. Table
11.7 shows a summary of grid sizes and parameters. Nramp denotes the number of first
iterations/cycles over which the CFL number is ramped from 10 to 200 for single-grid/multigrid
calculations. The multigrid cycle is V (2 ; 1) for these cases.
The convergence results for the wing-ap configuration is given in Figure 11.16 (f). It shows that
the multigrid converges (to machine zero) nearly 2 times faster in CPU time than the single-grid

Figure 11.17 Residual versus CPU time for the F6 wing-body case (RANS)
 314

iterations. For the NACA15 wing case, the solution does not fully converge in either single-grid or
multigrid computations apparently due to an unsteady behavior near the blunt trailing edge.
In all three cases, the ratio of the number of multigrid cycles to the number of single-grid iterations
is about twice the speed-up factor in terms of the CPU time. It implies that the cost of one multigrid
V (2 ; 1) cycle is close to the cost of two single-grid iterations. These results are in good agreement
with the estimates of the cost of one V-cycle computed according to Eq. 11.65 and shown in the
fourth column of Error! Reference source not found.. The estimated cost of one V -cycle is 1:8 of t
he single-grid iteration cost for all inviscid cases. The estimated speed-up shown in the fifth column
agrees well with the actual speed-up shown in the last column.
11.6.5.6.3 Turbulent Flows (RANS)
We applied the multigrid algorithm to a RANS simulation on the F6 wing-body grid shown in Figure
11.14. The inflow Mach number is 0:3, the angle of attack is 1 degree, and the Reynolds number is 2
: 5 M. For this case, a prolongation operator that is exact for a constant function is used. The P1
prolongation operator encountered a difficulty on a boundary for this particular configuration, and
it is currently under investigation. The CFL number is not ramped in this case, but set to 200 for the
mean-flow equations and 30 for the turbulence equation. Convergence results are shown in Figure
11.17. As can be seen, the multigrid achieved four orders of reduction in the residual 5 times faster
in CPU time than the single-grid iteration. For this case, neither the multigrid nor single-grid method
fully converges seemingly due to a separation near the wing-body junction. Four orders of magnitude
reduction is just about how far a single-grid is run in practice for this particular configuration. The
comparison of the number of cycles with the number of single-grid iterations in the figure implies
that the CPU time for a multigrid V (2 ; 1) cycle is less than the CPU time for two single-grid iterations.
As shown in Error! Reference source not found., one multigrid V-cycle actually costs 1 : 6 single-g
rid iterations, indicating a good agreement between the expected and actual speed-up factors.
11.6.5.7 Concluding Remarks
An agglomerated multigrid algorithm has been applied to inviscid and viscous flows over complex
geometries. A robust fully-coarsened hierarchical agglomeration scheme was described for highly-
stretched viscous grids, incorporating consistent viscous discretization on coarse grids. Results for
practical simulations show that impressive speed-ups can be achieved for realistic flows over
complex geometries. Parallelization of the developed multigrid algorithm is currently underway to
expand the applicability of the developed technique to larger-scale computations and to demonstrate
grid-independent convergence of the developed multigrid algorithm.
11.6.5.8 References
[1] Trottenberg, U., Oosterlee, C. W., and Schuller, A., Multigrid, Academic Press, 2001.
[2] Mavriplis, D. J., Multigrid Techniques for Unstructured Meshes," VKI Lecture Series VKI-LS 1995-
02, Von Karman Institute for Fluid Dynamics, Rhode-Saint-Genese, Belgium, 1995.
[3] Mavriplis, D. J., Unstructured Grid Techniques," Annual Review of Fluid Mechanics, Vol. 29, 1997,
pp. 473-514.
[4] Mavriplis, D. J. and Pirzadeh, S., Large-Scale Parallel Unstructured Mesh Computations for 3D High-
Lift Analysis," Journal of Aircraft, Vol. 36, No. 6, 1999, pp. 987{998.
[5] Mavriplis, D. J., An Assessment of Linear versus Non-Linear Multigrid Methods for Unstructured
Mesh Solvers," Journal of Computational Physics, Vol. 275, 2002, pp. 302-325.
[6] Nishikawa, H., Diskin, B., and Thomas, J. L., Critical Study of Agglomerated Multigrid Methods for
Diffusion," AIAA Journal, Vol. 48, No. 4, 2010, pp. 839-847.
[7] Thomas, J. L., Diskin, B., and Nishikawa, H., A Critical Study of Agglomerated Multigrid Methods for
Diffusion on Highly-Stretched Grids," Computers and Fluids, 2010, to appear.
[8] Spalart, P. R. and Allmaras, S. R., A One-Equation Turbulence Model for Aerodynamic Flows," AIAA
paper 92-0439, 1992.
 315

[9] Barth, T. J., Numerical Aspects of Computing Viscous High Reynolds Number Flows on Unstructured
Meshes," AIAA Paper 91-0721, 1991.
[10] Haselbacher, A. C., A Grid-Transparent Numerical Method for Compressible Viscous Flow on Mixed
Unstructured Meshes, Ph.D. thesis, Loughborough Universit, 1999.
[11] Anderson, W. K. and Bonhaus, D. L., An implicit upwind algorithm for computing turbulent flows
on unstructured grids," Computers and Fluids, Vol. 23, 1994, pp. 1-21.
[12] Diskin, B., Thomas, J. L., Nielsen, E. J., Nishikawa, H., and White, J. A., Comparison of Node-Centered
and Cell-Centered Unstructured Finite-Volume Discretization. Part I: Viscous Fluxes," 47th AIAA
Aerospace Sciences Meeting, AIAA Paper 2009-597, January 2009.
[13] Diskin, B. and Thomas, J. L., Accuracy Analysis for Mixed-Element Finite-Volume Discretization
Schemes," NIA Report No. 2007-08 , 2007.
[14] “The DPW-II Workshop for the geometry," http://aaac.larc.nasa.gov/tsab/cfdlarc/aiaa-
dpw/Workshop2/workshop2.
[15] Briggs, W. L., Henson, V. E., and McCormick, S. F., A Multigrid Tutorial, SIAM, 2nd ed., 2000.
[16] Thomas, J. L., Diskin, B., and Rumsey, C. L., Towards Verification of Unstructured Grid Methods,"
AIAA Journal, Vol. 46, No. 12, December 2008, pp. 3070-3079.
 316

12 Finite Element Method (FE)

The finite element method (FEM) is used in structural analysis of solids, but is also applicable to
fluids. However, the FEM formulation requires special care to ensure a conservative solution. The
FEM formulation has been adapted for use with fluid dynamics governing equations. Although FEM
must be carefully formulated to be conservative, it is much more stable than the finite volume
approach. However, FEM can require more memory and has slower solution times than the FVM. In
this method, a weighted residual equation is formed:

R i = ∭ Wi QdV ε
Ri
Eq. 12.1
Where Ri is the equation residual at an element vertex i, Q is the conservation equation expressed on
an element basis, Wi is the weight factor, and Vε is the volume of the element388. This is very general
statements and to illustrate we resort to simple examples to follow. But first some terminology and
definition of them.
The most powerful method in finite elements is the Galerkin method. Here, the solution is first
expanded in a set of basic functions and the residual is made orthogonal to a set of test functions.
When the basis and the test functions are the same, the method is termed a Galerkin method;
otherwise it is called a Petrov-Galerkin method. The standard Galerkin method leads to a centered
scheme and is unconditionally unstable for hyperbolic problems when combined with forward Euler
discretization in time. Therefore, artificial viscosity has to be added in some form to stabilize the
procedure. In contrast with the finite volume methods, finite element practitioners have always
preferred cell vertex schemes since the global function is usually expressed as a summation of trial
functions multiplied by the values at the vertices. These trial functions are typically assumed to be 1
at a vertex and zero at all other vertices.

12.1 Continuous & Discontinuous Galerkin Finite Element Methods

In mathematics, in the area of numerical analysis, Galerkin Methods are a class of approaches for
converting a continuous operator problem (such as a differential equation) to a discrete problem. In
principle, it is the equivalent of applying the method of variation of parameters to a function space,
by converting the equation to a weak formulation. Typically one then applies some constraints on
the function space to characterize the space with a finite set of basic functions. Galerkin's method
provides powerful numerical solution to d]ifferential equations and modal analysis. The approach is
usually credited to [Boris Galerkin] but the method was discovered by [Ritz], to whom Galerkin refers.
The basic idea of the Galerkin method is to multiply the system of partial differential equations
by arbitrary test functions. In Discontinuous Galerkin Methods we perform a partial integration
over each element resulting in a weak formulation. When we discretize the weak formulation the
discontinuous Galerkin methods specifically uses a numerical flux at the boundaries of the element
that concerns the two neighboring elements. The discontinuity may occur at the element boundaries.
The advantage of this method that we can refine hp locally, which means that we may refine the
element size h of one element or that we may adjust the order of our polynomial expansion p in an
element.
We can consider a piecewise linear discretization of a function u(x) in one dimension, see Figure
12.1-(top), in which the nodal values of uh(x) determine uh everywhere else. Hence, this finite
element discretization is based on function values at the nodes and is continuous across elements. In
contrast, in a discontinuous Galerkin finite element method a piecewise linear discretization is

388 Wikipedia, the free encyclopedia.

 317

found on each element and the

limit or trace values
approaching the nodes from the
element left or right of a node
are not assumed to be
continuous, see Figure 12.1-
(bottom). Hence, this
discretization is continuous in
each element but discontinuous
across elements. While more
degrees of freedom are required
in the discontinuous
discretization, it offers generally
more flexibility and more
accuracy.

12.2 Steps in the Finite Figure 12.1 Continuous vs. Discontinuous Galerkin Finite
Element Element Method
Discretization
For the both continuous and discontinuous Galerkin Finite Element, discretization usually contain
the following steps389:
1. Derive weak formulation: Each equation is multiplied by its own arbitrary test function,
integrated over the domain of validity entirely or as a sum of integrals over all elements, and
integrated by parts to obtain the weak formulation.
2. Form discretized weak formulation/algebraic system: The variables are expanded in the
domain or in each element in a series in terms of a finite number of basis functions. Each
basis function has compact support over neighboring elements (for continuous finite
elements) or within each element (for discontinuous finite elements). This expansion is then
substituted into the weak formulation, and a test function is chosen alternately to coincide
with a basis function, to obtain the discretized weak formulation. The resulting system is a
linear or nonlinear algebraic system.
3. Evaluate of integrals in local coordinate system: A local or reference coordinate system is
used to evaluate the integrals. In the continuous finite element discretization global matrices
and vectors are assembled in the assembly routine.
4. Solve algebraic system: The resulting algebraic system is solved (iteratively) using forward
time stepping methods or linear algebra routines.
Steps 1 and 2 are often combined in discontinuous Galerkin methods. If in step 3, we have not
discretized time, we choose a time discretization of the ordinary differential equations resulting after
the spatial finite element discretization. If in step 4 the algebraic system is nonlinear, we choose an
iterative solution method which essentially will solve a linear system at each iteration step. Compact
support means that the test functions are only nonzero locally over one or a few neighboring
elements. In the continuous finite element method (we consider), the basic functions are zero at the
edge of their domain of influence, while in the discontinuous case the basis function is (generally)
nonzero within an element including the element boundary and zero elsewhere. The careful
definition of function spaces for the test and basis functions is common practice in finite element

Onno Bokhove and Jaap J.W. van der Vegt, “Introduction to (dis)continuous Galerkin finite element methods”,
389

Department of Applied Mathematics, University of Twente.

 318

methods. This is often perceived to be complicated, but we will see that all the function spaces make
sense, also intuitively.

12.3 Concept of Shape (Basis) Function

The shape function is the function which interpolates the solution between the discrete values
obtained at the mesh nodes. Therefore, appropriate functions have to be used and, as already
mentioned, low order polynomials are typically chosen as shape functions (see Figure 12.2). For
three-dimensional finite element simulations it is convenient to discretize the simulation domain
using tetrahedrons, as depicted. Thus, linear shape functions must be defined for each tetrahedron
of the mesh, in order to apply the Galerkin method. For one dimensional shape function, If a
polynomial is selected as basis function, then

1 , i = j
N i (x) = a1 + a 2 x + a 3 x 2 + . . . . . . . .
N i (x j ) = δ ij = 
0 , i  j
1 - Linear Case : For two nodes of x1 and x 2 we have N i (x) = a i1 + a i2 x
2 - Quadratic Case : x 1 , x 2 , and x 3 N i (x) = a i1 + a i2 x + a i3x 2
3 - Cubic Case : x 1 , x 2 , x 3 and x 4 N i (x) = a i1 + a i2 x + a i3x 2 + ai 4 x 3
Eq. 12.2

Linear Quadratic

Cubic Lagrangian

Figure 12.2 Illustration of Different Shape, Basis, and Blending Functions

12.3.1 Lagrangian Elements

To avoid solving this so complex system of equations, the well-known properties of the Lagrangian
polynomials can be used where it compasses all these shape functions. This equation is easier to
 319

implement, as can be checked using this Matlab code. The graphic representation of each case show
case in Figure 12.2390.

12.4 Simple 1D Discretization Example using Shape (Basis) Function

This definition is applicable to every bit as of FV method. The description of the laws of physics for
space and time-dependent problems are usually expressed in terms of partial differential equations
(PDEs). The Finite Element Method (FEM) is used to compute such approximations391. Take, for
example, a function u that may be the dependent variable in a PDE (i.e., temperature, electric
potential, pressure, etc.) The function u can be approximated by a function uh using linear
combinations of basis (shape) functions according to the following expressions:

u  uh wher e u h =  u i ψi Eq. 12.3

i

Here, ψi denotes the basis functions and ui denotes the coefficients of the functions that approximate
u with uh. The principle for a 1D problem. u could, for instance, represent the temperature along the
length (x) of a rod that is non-uniformly heated. Here, the linear basis functions have a value of 1 at
their respective nodes and 0 at other nodes. In this case, there are seven elements along the portion
of the x-axis, where the function u is defined (i.e., the length of rod). One of the benefits of using the
finite element method is that it offers great freedom in the selection of discretization, both in the
elements that may be used to discretize space and the basis functions. Depending on the problem at
hand, other functions may be chosen instead of linear functions.

12.5 Mathematics Behind Finite Element Method

To enlighten ourselves without details, we resort to the following statements. According to [Jeff H
Peterson]392, the mathematical basis for the FEM, lies in the math of Hilbert spaces. A Hilbert space
is a way of treating a function like a vector so that we can do some vector math tricks with it.
Remember that a vector is a series of values multiplied by a set of orthogonal basis vectors (like the
unit vectors that define the x, y, and z directions... think i, j, k unit vectors). We can use a similar
technique to define a function. First, we pick a set of basis functions instead of vectors (the functions
need to be orthogonal to each other Orthogonal functions but that's getting too much into the math)
and then we can define the original function as a sum of coefficients multiplied by those basis
functions like this:

u = αk ψk where ψ is the basis function Eq. 12.4
k =1

The next step is to convert our differential equation into something called the "weak formulation".
This is basically multiplying by a test function and then integrating over the space. Without going into
the specifics, this is how we convert our differential equation into a form we can use our vector math
on. This form is an "inner product" (the equivalent of a dot product except we're dealing with
functions now, see Inner product space) of our solution and the test function. This inner product is
an integral and we can use Integration by parts to convert it to more manageable form (that also
includes some nice ways to impose boundary conditions). After this, we recognize that we're actually

390 CFD online, 2009.

391 Detailed Explanation of the Finite Element Method (FEM), COMSOL Inc.
392 Jeff H Peterson, “What exactly is Finite element analysis – Quora”, PhD in mathematical modeling of crystal

growth systems.
 320

dealing with a finite-dimensional space. Thus the function vectors aren't infinite and we're not really
summing to infinity. This is what is meant by discretization in the FEM. Remember our basis
functions we picked earlier? We use a finite number of these to represent the solution of our
differential equation like this:
N
u =  α k ψ k Eq. 12.5
k =1

The only difference between this and what's above is that now our sum is finite. The next trick is to
let our test function be a basis function. We also make sure to choose basis functions such that they
don't overlap. This ensures that they are "orthogonal" like we wanted earlier and gives us a really
easy way to approximate our solution over the domain of interest. These basis functions are often
polynomials (especially quadratic polynomials). It almost seems like we've made the problem harder
by adding all this abstraction and mathematics but at the end of all of this, what have we really done
here? We have converted the problem into its matrix form and we can now solve it using matrix
algebra. If the problem was linear to begin with, we're simply solving

𝐀𝐱 = 𝐛
Eq. 12.6
which, is what most numerical methods classes teach you how to solve. Of course even if the problem
wasn't linear, we can linearize it with something like Newton's Method and still solve it. For a simple
problem such as Poisson's equation, the matrix A is very easy to compute and is often called the
"Stiffness Matrix" in homage to the FEM's beginnings in elasticity problems. This matrix is formed
by the inner product of the basic functions with themselves (very sparse and diagonal dominant if
you constructed your problem rationally) multiplied by whatever constant is in your original
equation. The solution vector then is a list of the coefficients being multiplied by the basic functions,
and in order to plot your solution and get real values out, you multiply this by your series of basis
functions. This provides a function (albeit a long, complicated one) that approximates the solution to
your problem.

12.6 Summary of Math Procedure for Finite Element Analysis

So in short we are performing these steps:
➢ Convert our problem into a vector problem using the properties of Hilbert spaces.
➢ Create the weak formulation so that now we can pose our problem as an inner product.
➢ Discretize our domain and choose basis functions that don't overlap elements.
➢ Convert our inner product into a set of linear equations.
➢ Solve the linear set of equations and get your solution.

12.7 Example with Heat Source

Say there is a solid with time-varying temperature but negligible variations in space. In this case, the
equation for conservation of internal (thermal) energy may result in an equation for the change of
temperature, with a very small change in time, due to a heat source

dT
c p = g(T, t) Eq. 12.7
dt
Here, ρ denotes the density and cp denotes the heat capacity. Temperature, T, is the dependent
 321

variable and time, t, is the independent variable. The function may describe a heat source that varies
with temperature and time. The ODE in Eq. 12.7 states that if there is a change in temperature in
time, then this has to be balanced (or caused) by the heat source. Oftentimes, there are variations in
time and space. The temperature in the solid at the positions closer to a heat source may, for instance,
be slightly higher than elsewhere. Such variations further give rise to a heat flux between the different
parts within the solid. In such cases, the conservation of energy can result in a heat transfer equation
that expresses the changes in both time and spatial variables (x), such as:

T
c p + .q = g(T, t, x) Eq. 12.8
t

As before, T is the dependent variable, while x (x = (x, y, z)) and t are the independent variables. The
heat flux vector in the solid is denoted by q = (qx, qy, qz) while the divergence of q describes the change
in heat flux along the spatial coordinates referred to as Fourier’s law. Here, the derivatives are
expressed in terms of t, x, y, and z. When a differential equation is expressed in terms of the
derivatives of more than one independent variable, it is referred to as a partial differential equation
(PDE), since each derivative may represent a change in one direction out of several possible
directions. Further note that the derivatives in ODEs are expressed using d, while partial derivatives
are expressed using the more curly ∂. Rather than solving PDEs analytically, an alternative option is
to search for approximate numerical solutions to solve the numerical model equations. The finite
element method is exactly this type of method; a numerical method for the solution of PDEs. Similar
to the thermal energy conservation referenced above, it is possible to derive the equations for the
conservation of momentum and mass that form the basis for fluid dynamics. Further, the equations
for electromagnetic fields and fluxes can be derived for space and time-dependent problems, forming
systems of PDEs. Continuing this discussion, let's see how the so-called weak formulation can be
derived from the PDEs.
12.7.1 Steady State Heat Sink; Weak Formulation; Basis Functions & Test Functions
Assume that the temperature distribution in a heat sink is being studied, given by Eq. 12.8, but now
at steady state, meaning that the time derivative of the temperature field is zero. The domain
equation for the model domain, Ω, is the following:

.(-kT) = g (T, x) in Ω
with B.C. T = T0 on Ω1
Eq. 12.9
(−kT).n = h (T − Tamb ) on Ω 2
(−kT).n = 0 on Ω 3

further, assume that the temperature along a boundary (∂Ω1) is known, in addition to the expression
for the heat flux normal to some other boundaries (∂Ω2). On the remaining boundaries, the heat flux
is zero in the outward direction (∂Ω3). The boundary conditions at these boundaries then becomes
where h denotes the heat transfer coefficient and Tamb denotes the ambient temperature. The
outward unit normal vector to the boundary surface is denoted by n. The situation is best describe in
Figure 12.3. The next step is to multiply both sides of Eq. 12.8 by a test function φ and integrate
over the domain Ω:
 322

 .(-kT)  dV =  g(T, x) dV

Ω Ω
in Ω
Eq. 12.10

Figure 12.3 Mathematical Model of a Steady State Heat Sink.

12.7.2 Test Function

The test function φ and the solution T are assumed to belong to Hilbert spaces. A Hilbert space is an
infinite dimensional function space with functions of specific properties. It can be viewed as a
collection of functions with certain nice properties, such that these functions can be conveniently
manipulated in the same way as ordinary vectors in a vector space. For example, you can form linear
combinations of functions in this collection (the functions have a well-defined length referred to as
norm) and you can measure the angle between the functions, just like Euclidean vectors. Indeed, after
applying the finite element method on these functions, they are simply converted to ordinary vectors.
The finite element method is a systematic way to convert the functions in an infinite dimensional
function space to first functions in a finite dimensional function space and then finally ordinary
vectors (in a vector space) that are tractable with numerical methods.
12.7.3 Weak Formulation
The weak formulation is obtained by requiring Eq. 12.9 to hold for all test functions in the test
function space instead of Eq. (5.8) for all points in Ω. A problem formulation based on Eq. (5.8) is
thus sometimes referred to as the pointwise formulation. In the so-called Galerkin Method, it is
assumed that the solution T belongs to the same Hilbert space as the test functions. This is usually
written as φ ϵ H and T ϵ H, where H denotes the Hilbert space. Using Green’s first identity (essentially
integration by parts), the following equation can be derived from (5.9):
 323

 kT. dV −  (−kT) . n dS =  g (T, x)  dV

Ω 2Ω Ω
Eq. 12.11
The weak formulation, or Variational formulation, of Eq. 12.9 is obtained by requiring this
equality to hold for all test functions in the Hilbert space. It is called “weak” because it relaxes the
requirement Eq. 12.9, where all the terms of the PDE must be well defined in all points. The relations
in Eq. 12.11 and Eq. 12.10 instead only require equality in an integral sense. For example, a
discontinuity of a first derivative for the solution is perfectly allowed by the weak formulation since
it does not hinder integration. It does, however, introduce a distribution for the second derivative
that is not a function in the ordinary sense. As such, the requirement (Eq. 12.9) does not make sense
at the point of the discontinuity. A distribution can sometimes be integrated, making Eq. 12.10 well
defined. It is possible to show that the weak formulation, together with boundary conditions (5.8), is
directly related to the solution from the pointwise formulation. And, for cases where the solution is
differentiable enough (i.e., when second derivatives are well defined), these solutions are the same.
The formulations are equivalent, since deriving (5.10) from (5.8) relies on Green’s first identity,
which only holds if T has continuous second derivatives. This is the first step in the finite element
formulation. With the weak formulation, it is possible to discretize the mathematical model equations
to obtain the numerical model equations. The Galerkin method one of the many possible finite
element method formulations can be used for discretization. First, the discretization implies looking
for an approximate solution to Eq. 12.11 in a finite-dimensional subspace to the Hilbert space H so
that T ≈ Th. This implies that the approximate solution is expressed as a linear combination of a set of
basis functions ψi that belong to the subspace:

Th (x) =  Ti ψi (x)
i
Eq. 12.12
The discretized version of Eq. 12.12 for every test function ψj therefore becomes:

 T  kψ .ψ dV +   (−kT ψ ).nψ dS =  g T ψ

i
i i j
i
i i j
i
i i ψ jdV
Ω 2Ω Ω
Eq. 12.13
The unknowns here are the coefficients Ti in the approximation of the function T(x). Eq. 12.13 then
forms a system of equations of the same dimension as the finite-dimensional function space. If n
number of test functions ψj are used so that j goes from 1 to n, a system of n number of equations is
obtained according to Eq. 12.13. From Eq. 12.12, there are also an unknown coefficients (Ti).
Once the system is discretized and the boundary conditions are imposed, a system of equations is
obtained according to the following expression:

ATh = b Eq. 12.14

where T is the vector of unknowns, T h = {T1, .., Ti, …, Tn}, and A is an NxN matrix containing the
coefficients of Ti in each equation j within its components Aji. The right-hand side is a vector of the
dimension 1-n. A is the system matrix, often referred to as the stiffness matrix, harkening back to the
finite element method’s first application as well as its use in structural mechanics. If the source
function is nonlinear with respect to temperature or if the heat transfer coefficient depends on
temperature, then the equation system is also nonlinear and the vector b becomes a nonlinear
function of the unknown coefficients Ti. One of the benefits of the finite element method is its ability
to select test and basis functions. It is possible to select test and basis functions that are supported
 324

over a very small geometrical region. This implies that the integrals in Eq. 12.14 are zero
everywhere, except in very limited regions where the functions ψj and ψi overlap, as all of the above
integrals include products of the functions or gradients of the functions i and j. The support of the
test and basis functions is difficult to depict in 3D, but the 2D analogy can be visualized.

12.8 Effect of Basis Function

Assume that there is a 2D geometrical domain and that linear functions of x and y are selected, each
with a value of 1 at a point i, but zero at other points k. The next step is to discretize the 2D domain
using triangles and depict how two basis functions (test or shape functions) could appear for two
neighboring nodes i and j in a triangular mesh. Two neighboring basis functions share two triangular
elements. As such, there is some overlap between the two basis functions, as shown in Figure 12.4
(A - B). Further, note that if i = j, then there is a complete overlap between the functions. These
contributions from the coefficients for the unknown vector T that correspond to the diagonal
components of the system matrix Ajj. If the two basis functions are now a little further apart where
these functions do not share elements but they have one element vertex in common. As the Figure
12.4 (A) indicates, they do not overlap. When the basis functions overlap, the integrals in Eq. (5.12)
have a nonzero value and the contributions to the system matrix are nonzero. When there is no
overlap, the integrals are zero and the contribution to the system matrix is therefore zero as well.
This means that each equation in the system of equations for (5.12) for the nodes 1 to n only gets a
few non-zero terms from neighboring nodes that share the same element. The system matrix A in Eq.
(5.13) becomes sparse, with nonzero terms only for the matrix components that correspond to
overlapping Tij. The solution of the system of algebraic equations gives an approximation of the
solution to the PDE. The denser the mesh, the closer the approximate solution gets to the actual
solution393.

(B) Not overlapping

(A) Overlapping

Figure 12.4 Basis functions sharing two triangular elements in 2D

12.9 Different Elements

As mentioned above, the Galerkin Method utilizes the same set of functions for the basic functions
and the test functions. Yet, even for this method, there are many ways (infinitely many, in theory) of
defining the basis functions (i.e., the elements in a Galerkin finite element formulation). Let's review
some of the most common elements. The linear basis functions, as defined in a triangular mesh that
forms triangular linear elements, are depicted in Figure 12.5. The basic functions are expressed as
functions of the positions of the nodes (x and y in 2D and x, y, and z in 3D). In 2D, rectangular elements
are often applied to structural mechanics analyses. They can also be used for boundary layer meshing

393 Detailed Explanation of the Finite Element Method (FEM), COMSOL Inc.
 325

in CFD and heat transfer modeling.

Their 3D analogy is known as the
hexahedral elements, and they are
commonly applied to structural
mechanics and boundary layer
meshing as well. In the transition
from hexahedral boundary layer
elements to tetrahedral elements,
pyramidal elements are usually
placed on top of the boundary layer
elements. The corresponding second-
order elements (quadratic elements)
are shown in the figure below. Here,
the edges and surfaces facing a
domain boundary are frequently
curved, while the edges and surfaces Figure 12.5 Finite element approximation of the
facing the internal portion of the temperature field in the heat sink
domain are lines or flat surfaces. Note,
however, that there is also the option to define all edges and surfaces as curved. Lagrangian and
serendipity most common element types in 2D and 3D. It is difficult to depict the basis of the
quadratic basis functions in 3D inside the elements above, but color fields can be used to plot function
values on the element surfaces. When discussing FEM, an important element to consider is the error
estimate. This is because when an estimated error tolerance is reached, convergence occurs. Note
that the discussion here is more general in nature rather than specific to FEM.

12.10 Error Estimation

The finite element method gives an approximate solution to the mathematical model equations. The
difference between the solution to the numerical equations and the exact solution to the
mathematical model equations is the error: e = u - uh}. In many cases, the error can be estimated
before the numerical equations are solved (i.e., an a priori error estimate). A priori estimates are
often used solely to predict the convergence order of the applied finite element method. For instance,
if the problem is well posed and the numerical method converges, the norm of the error decreases
with the typical element size h according to O(hα), where α denotes the order of convergence. This
simply indicates how fast the norm of the error is expected to decrease as the mesh is made denser.
An a priori estimate can only be found for simple problems. Furthermore, the estimates frequently
contain different unknown constants, making quantitative predictions impossible. An a posteriori
estimate uses the approximate solution, in combination with other approximations to related
problems, in order to estimate the norm of the error.

12.11 Additional Finite Element Formulations

In the examples above, we have formulated the discretization of the model equations using the same
set of functions for the basis and test functions. One finite element formulation where the test
functions are different from the basis functions is called a Petrov-Galerkin Method. This method is
common, for example, in the solution of convection-diffusion problems to implement stabilization
only to the streamline direction. It is also referred to as the streamline upwind/Petrov-Galerkin
(SUPG) method. In the solution of coupled systems of equations, different basis functions may be
used for different dependent variables. A typical example is the solution of the Navier-Stokes
equations, where the pressure is often more smooth and easy to approximate than the velocity.
Methods where the basis (and test) functions for different dependent variables in a coupled system
belong to different function spaces are called mixed finite element methods.
 326

12.12 Case Study - A High Order Discontinuous Galerkin – Fourier Incompressible

3D Navier–Stokes Solver with Rotating Sliding Meshes394

Citation : Esteban Ferrer, Richard H.J. Willden, A high order Discontinuous Galerkin – Fourier
incompressible 3D Navier–Stokes solver with rotating sliding meshes, Journal of Computational Physics,
Volume 231, Issue 21, 2012, Pages 7037-7056, ISSN 0021-9991,
https://doi.org/10.1016/j.jcp.2012.04.039.
We present the development of a sliding mesh capability for an unsteady high order (order P ≥3) h/p
Discontinuous Galerkin solver for the 3D incompressible Navier–Stokes equations. A high order
sliding mesh method is developed and implemented for flow simulation with relative rotational
motion of an inner mesh with respect to an outer static mesh, through the use of curved boundary
elements and mixed triangular–quadrilateral meshes. A second order stiffly stable method is used to
discretize in time the Arbitrary Lagrangian–Eulerian form of the incompressible Navier–Stokes
equations. Spatial discretization is provided by the Symmetric Interior Penalty Galerkin formulation
with modal basis functions in the x–y plane, allowing hanging nodes and sliding meshes without the
requirement to use mortar type techniques. Spatial discretization in the z-direction is provided by a
purely spectral method that uses Fourier series and allows computation of spanwise periodic 3D
flows. The developed solver is shown to provide high order solutions, second order in time
convergence rates and spectral convergence when solving the incompressible Navier–Stokes
equations on meshes where fixed and rotating elements coexist.
In addition, an exact implementation of the no-slip boundary condition is included for curved edges;
circular arcs and NACA 4-digit airfoils, where analytic expressions for the geometry are used to
compute the required metrics. The solver capabilities are tested for a number of two dimensional
problems governed by the incompressible Navier–Stokes equations on static and rotating meshes:
the Taylor vortex problem, a static and rotating symmetric NACA0015 airfoil and flows through three
bladed cross-flow turbines. In addition, three dimensional flow solutions are demonstrated for a
three bladed cross-flow turbine and a circular cylinder shadowed by a pitching NACA0012 airfoil.
12.12.1 Introduction
Problems where the forces on rotating/oscillating geometries are to be predicted are common in
engineering and fluid–structure interaction problems. Examples are flows around isolated rotating
bodies and airfoils, turbomachinery applications, insect flight aerodynamics, unmanned air vehicles
and more recently flows around renewable energy devices; wind and tidal turbines. These flows are
characterized by long wake structures, vortex shedding and stalled flows associated with flow
unsteadiness. To accurately simulate wakes and vortex structures and their evolution, high order
(typically P≥3) numerical methods (i.e. h/p conformal spectral and h/p non-conformal
Discontinuous Galerkin methods) are preferred since dissipation and dispersion errors are
minimum395 when compared to low order (typically 63) methods. Furthermore, for smooth
problems, the exponential decay of the error with polynomial enrichment (p-refinement) in high
order methods as opposed to the fixed decay rate characteristic of low order methods (i.e. h-
refinement only) renders high order methods particularly attractive to obtain accurate solutions for
flows where viscosity limits discontinuities (i.e. elliptic type equations) as in the incompressible
Navier–Stokes (NS) equations.
For conformal discretization (i.e. classic low order finite elements or h/p spectral type methods), if

394 Esteban Ferrer, Richard H.J. Willden, “A high order Discontinuous Galerkin – Fourier incompressible 3D
Navier–Stokes solver with rotating sliding meshes”, Department of Engineering Science, U. of Oxford, UK.
395 G. Karniadakis, S. Sherwin, Spectral h/p Element Methods for Computational Fluid Dynamics, Oxford Science

Publications, 2005.
 327

geometric discontinuities arise (e.g. hanging nodes), a geometric incompatibility396 is created since
C0 continuity cannot be ensured across elements. Similarly, when neighboring elements have
different numbers of degrees of freedom (e.g. mixed triangular–quadrilateral mesh), a functional
incompatibility is created. Over the past few decades methods to overcome these difficulties have
been developed in the framework of conformal discretization, examples are iterative patching,
constrained approximation or mortar patching (see 419 for a review). A natural alternative to
overcome these incompatibilities is to relax the continuity condition across elements, which leads to
the Discontinuous Galerkin (DG) finite element approach.
The DG finite element method can be seen as an extension of h/p spectral methods where the C0
continuity requirement across element boundaries is relaxed or also as a high order finite volume
method with compact stencil. Contrary to conformal finite elements or h/p spectral methods, DG
methods are locally conservative by construction. As in spectral methods, high-order polynomials
can be used within each element allowing exponential convergence, whilst discontinuities in the
solution are permitted at element interfaces.
It has been argued that DG methods are prohibitively expensive, when compared to conformal spatial
discretization (e.g. h/p spectral methods), since additional degrees of freedom (DOF) arise from the
discontinuities between elements. However, the relative number of the additional boundary degrees
of freedom to internal degrees of freedom for each element, decreases rapidly for high polynomials.
To exemplify this, let us consider a two dimensional problem and a mesh with Nel triangular elements
with polynomial spaces of order k. On the one hand, if a h/p spectral discretization is considered, the
number of global degrees of freedom required is DOF spectral ≈ (Nel /2) k2 (see397). On the other
hand, a DG discretization leads to a global number of DOF: DOFDG ≈ (Nel2/2) (k+1)(k+2). The last
expression shows that for high polynomial orders k, the number of degrees of freedom for the DG
discretization tends to DOFDG ≈ (Nel2/2)/k2. This simple analysis shows that for high polynomial
orders DOFDG ≈ DOFspectral and the additional cost of DG methods is not overwhelming.
DG formulations to solve the incompressible NS equations have seen increased popularity over
recent years as evidenced by the number of publications on the topic. Previously, the authors
presented the development of a DG code that uses the Symmetric Interior Penalty Galerkin (SIPG)
formulation for solving the unsteady 2D NS equations using straight sided triangular elements.
Simulation results from the solver were shown to be in good agreement with experimental results
and results from a h/p conformal spectral code. The present implementation is an extension of the
previous work to enable solutions on hybrid meshes (i.e. mixed triangular and quadrilateral
elements) with rotationally sliding meshes and with curved boundary conditions. In addition, Fourier
series applied orthogonally to the 2D DG plane are used to enable 3D flow solutions for problems
with spanwise geometric homogeneity. These developments have been implemented and tested and
results are reported herein.
The sliding mesh technique allows for mesh motion where an inner mesh zone region rotates with
respect to an outer static mesh. This relative motion creates hanging nodes at the interface between
static and rotating elements and boundaries (i.e. walls) curved edged elements are essential. To
clarify these concepts before continuing, we depict in Figure 12.6 an example of a 2D mesh, where
the static and rotating subdomains, the external curved boundary for a symmetric NACA airfoil and
the curved sliding mesh interface with the associated hanging nodes have been highlighted.
The present work shows that in the DG context, the geometric incompatibility arising from the
hanging nodes due to mesh rotation does not cause loss of exponential convergence properties.
Further, we show that the functional incompatibility originating from the use of triangular and
quadrilateral element types (these elements have different numbers of local degrees of freedom) in

396See Previous.
397P.E. Vos, S. Sherwin, R. Kirby, From h to p efficiently: implementing finite and spectral h/p element methods
to achieve optimal performance for low and high-order discretization, J. of Computational Physics (2010).
 328

combination with orthogonal modal basis functions, does not present a problem. Finally, the sliding
mesh implementation shows high accuracy when solving the incompressible NS equations in two and
three dimensions.
To account for the relative mesh movement of the inner mesh with respect to the static outer mesh,
it is advantageous to write the equations for fluid motion in their Arbitrary Lagrangian–Eulerian
(ALE) form398. The ALE description was first introduced for finite difference methods and
unstructured meshes for fluid simulation399 and subsequently extended to finite elements. The ALE
approach is generally used for dynamically deforming mesh elements (i.e. arbitrary node movement)
and has been widely explored in combination with DG methods to solve hyperbolic type equations
including the compressible NS equations. As for the incompressible ALE form of the NS equations

Figure 12.6 Mixed Triangular–Quadrilateral Mesh for a Symmetric Airfoil with Curved Boundaries and
a Circular Sliding Mesh. Static and Rotating Subdomains are Distinguished

with arbitrary mesh movement, work was limited for a long time to conformal h/p spectral
discretization400. However, very recent work combines this technique with a DG approach.
Deforming element techniques (e.g. ALE for deforming elements) require either generally small body
motions or remeshing for large motions that would have otherwise lead to unacceptably element
distortions. To avoid these limitations, an appealing approach is provided by combining the ALE
approach with the sliding mesh technique. This method is particularly suitable to problems where
the mesh movement is known a priori; e.g. rigid body rotation without mesh deformation. We chose
to follow this approach, the ALE formulation with sliding meshes for non-deforming elements, and
summarize some of its advantages below:
• Inertially fixed and rotating objects can be present in the same simulation.

398 J. Donea, A. Huerta, Finite Element Methods for Flow Problems, John Wiley & Sons Ltd., 2005.
399 C. Hirt, A. Amsden, J. Cook, An arbitrary Lagrangian–Eulerian computing method for all flow speeds, Journal
of Computational Physics 14 (3) (1974).
400 A. Beskok, T. Warburton, An unstructured hp finite-element scheme for fluid flow and heat transfer in moving

domains, Journal of Computational Physics 174 (2) (2001) 492–509.

 329

• No remeshing is required (with its associated computational cost) as mesh elements do not
distort, enabling unlimited rotation as opposed to non-sliding or deforming ALE methods in
which large distortions and subsequent remeshing need to be handled.
• Subsequently, no projection of the solution into a new mesh is required to advance the
solution in time, which is generally a non-conservative process.
• In our approach, no interpolation through the sliding interface is required (as in low order
methods) which would introduce large numerical errors, destroying the high order
properties of the method.
The ALE approach with sliding mesh interfaces, has been previously studied in the context of h/p
conformal spectral methods that require mortar techniques for subdomain linking. Recently, a sliding
mesh capability has been described for the incompressible NS equations401, where the authors used
is geometric analysis and a low order conformal discretization (using NURBS) coupled with a special
treatment for the sliding curved interface exploiting ideas from DG methods. Purely DG discretization
in conjunction with the sliding mesh approach, have been used for the solution of electromagnetic
problems (i.e. Maxwell equations). It was shown innumerous references that to solve the Maxwell
eigensystem with sliding mesh interfaces, mortar techniques are beneficial.
To the authors’ knowledge, the work presented herein details the first high order (order P≥3) DG
solver with sliding meshes for the solution of the incompressible NS equations. We present a novel
approach where non-conformal DG is used in all elements and the curved sliding mesh interface is
approximated through an analytical mapping for the description of the circular interface. We show
that our approach and implementation does not require of mortar type techniques to accurately solve
the incompressible NS equations. In addition, to account for curved external boundary surfaces (i.e.
walls) representing NACA 4-digit airfoils, we introduce an analytical mapping as is required to obtain
smooth solutions for high order techniques. An efficient three dimensional extension of the 2D DG
solver is included, which enables 3D flow solutions on rotating geometries that present a geometrical
homogeneity in the spanwise (Fourier) direction.
12.12.2 Methodology
We are interested in solving the incompressible NS equations in primitive variable formulation in a
unique domain composed of non-overlapping static and moving subdomains:

Ω(t) = Ωsta ∪ Ωrot (t)

Eq. 12.15
Where Ωstat ∩ Ωrot(t) = Γsta-rot and Γsta-rot defines the circular sliding mesh interface. The non-
dimensional incompressible NS equations in ALE form402 can be written (in the absence of body
forces) as:

∂𝐮 1
= ((𝐮 − 𝐰). ∇)𝐮 = −∇p + ∇2 𝐮 ; ∇. 𝐮 = 0 in Ω(t)
∂t Re
Eq. 12.16
where u = (u, v)T and w are non-dimensionalised using the characteristic free stream flow velocity
magnitude U and represent the flow and mesh velocities, respectively. Re is the Reynolds number
(i.e. Re = UL/ν where L is a characteristic length scale and ν is the kinematic viscosity), t represents
the dimensionless time and p is the non-dimensional pressure (normalized using upstream dynamic

401 J. Cottrell, T. Hughes, Y. Bazilevs, Isogeometric Analysis: Toward Integration of CAD and FEA, John Wiley &
Sons, Ltd, 2009.
402 J. Donea, A. Huerta, Finite Element Methods for Flow Problems, John Wiley & Sons Ltd., 2005.
 330

pressure q = ρU2, where ρ is the fluid density). We distinguish between the static and rotating
subdomain meshes (see Figure 12.6) by setting:

0 ∈ Ωsta
𝐰={
̅ × 𝐱 ∈ Ωrot (t)
𝛚
Eq. 12.17
where el are the mesh elements in Ω(t), ϖ = (0, 0, Lω/U) is the non-dimensional mesh rotational
velocity and ω represents the dimensional mesh rotational speed. To solve the described system, we
select to discretize temporally using finite differences (here a dual splitting method), and a high order
DG-Fourier method for spatial discretization.
12.12.2.1 Arbitrary Lagrangian–Eulerian Temporal Discretization
We perform the temporal discretization using a second order splitting scheme developed in403 and
studied previously by the authors in conjunction with a DG method for the incompressible NS
equations in Eulerian form. The ALE version of this algorithm has been previously used in the context
of h/p spectral discretization. Within this scheme the nonlinear terms are treated explicitly whilst
viscous and pressure terms are handled implicitly. The resulting temporally discretized momentum
equation is given by:

γ0 𝐮𝐧+𝟏 − α0 𝐮𝐧 − α1 𝐮𝐧−𝟏 1 2 n+1

= −∇pn+1 + ∇ 𝐮 −
∆𝐭 Re
β0 [((𝐮n − 𝐰 n ). ∇)𝐮n ] − β1 [((𝐮n−1 − 𝐰 n−1 ). ∇)𝐮n−1 ]
Eq. 12.18
Using second order Adams–Bashforth values for the coefficients, we set: γ0 = 3/2, α0 = 2, α1 = -1/2,
β0 = 2 and β1 = -1. This second order method is not self-starting and we select: γ0 = α0 = β0 = 1 and α1
= β1 = 0 which reduces to the first order explicit forward Euler method for the first time step. For
detailed information such as rigid mesh motion, special discretization, mapping, users encourage to
access with [Ferrer and Willden]404.
12.12.2.2 3D Fourier Extension
The selected tensorial representation of the mappings from physical to computational space in
combination with tensorial orthogonal basis functions allow for 3D extensions of the method without
further modifications. In this section, we describe a 3D extension where the 2D Discontinuous
Galerkin capabilities presented in the previous sections are extended to account for three
dimensional flow features using Fourier series. The selected Fourier extension enables the use of the
Fast Fourier Transform (FFT) algorithm and provides an efficient 3D solver for problems involving
spanwise geometric homogeneity. The use of Fourier series, to efficiently extend two dimensional
codes and account for three dimensional flows, has been previously reported for h/p conformal
spectral methods. To the authors’ knowledge, this type of extension has not been used together with
DG discretization. However, this technique leads to a series of 2D DG linear solves that are coupled
through the non-linear terms (see details below), in a similar manner to conformal implementations.
This feature renders these Fourier extensions similar. Let us introduce the three dimensional DG-
Fourier velocities and pressure as:

403 J. Hesthaven, T. Warburton, Nodal Discontinuous Galerkin Methods: Algorithms, Analysis, and Applications,
Springer, 2008.
404 Esteban Ferrer, Richard H.J. Willden, “A high order Discontinuous Galerkin – Fourier incompressible 3D

Navier–Stokes solver with rotating sliding meshes”, Department of Engineering Science, U. of Oxford, UK.
 331

M M
−1 −1
2 2

𝐮(x, y, z) = ∑ 𝐮(x, y)eiβmz , p(x, y, z) = ∑ 𝐮(x, y)eiβmz

M M
m=− m=−
2 2
Eq. 12.19
i=√-1, b = 2p/Lz represents the wavenumber in the z-direction and Lz defines the length of the domain
in the periodic z-direction. In addition, M defines the number of planes in physical space (and the
number of modes in the Fourier space) and m denotes the m-th mode of the Fourier transformation.
Let us note that um(x,y) and pm(x,y) have been previously defined (see [Ferrer and Willden]405) and
represent the x–y plane DG solution for the m-th Fourier mode. Applying the Fourier transform to
the momentum and continuity equations in ALE form Eq. 12.16 and using its linearity property, one
obtains:

M
−1
2
∂𝐮m 1 2
∑ { ̂m pm − ∇
+∇ ̂m 𝐮m + FFT[(𝐮 − 𝐰). ∇ 𝐮]} eiβmz = 0
∂t Re
M
m=−
2
M
−1
2
̂m 𝐮m }eiβmz = 0
∑ {∇ in Ω(t)
M
m=−
2
Eq. 12.20
where FFTm{_} represent the m-th mode of the Fourier transform of the argument and the Fourier
operators ⟑ are defined as:

T
∂ ∂
̂m = (
∇ , , iβm) = (∇2D , iβm)T
∂x ∂y
∂2 ∂2
̂
∇𝑚 = ( 2 + 2 − β2 m2 ) = (∇22𝐷 − β2 m2 )
2
∂x ∂y
Eq. 12.21
where ⊽ 2D and ⊽22D are the two dimensional gradient (nabla) and Laplacian operators, respectively.
The system described by Eq. 12.20 is equivalent to solving an independent equation for each mode
m:

∂𝐮m 1
{ ̂m pm −
+∇ ̂2m 𝐮m + FFT[(𝐮 − 𝐰). ∇ 𝐮]} = 0 ; ∇
∇ ̂m 𝐮m = 0 in Ω(t)
∂t Re

The coupling between modes is ensured by the non-linear terms which are computed in physical
space. It should be noted that this Fourier extension only requires real to complex transforms and
hence only M/2 + 1 modes need to be computed at each time step.

Esteban Ferrer, Richard H.J. Willden, “A high order Discontinuous Galerkin – Fourier incompressible 3D
405

Navier–Stokes solver with rotating sliding meshes”, Department of Engineering Science, U. of Oxford, UK.
 332

12.12.2.3 Rotating Cases

In this section we consider cases where the NACA 4-digit boundary is used conjointly with the sliding
mesh technique (mesh depicted in Figure 12.6 and solution snapshots in Figure 12.7). A
NACA0015 airfoil under impulsive start conditions is rotated around its quarter chord point for a
range of geometric angles of attack (i.e. the angle between the free stream velocity and the airfoil
chord line) AOA = 0–35 degrees and four rotational speeds Lx/U = 0.005, 0.01, 0.05 and 0.1, where L

Figure 12.7 Snapshots of the Rotating NACA0015 for Polynomial Order k = 5 and Rotational Speed
Lx/U = 0.05 for (a) AOA = 17.2 deg. and (b) AOA = 28.6 deg

= c and c represents the airfoil chord. The Reynolds number is fixed to Re = 100. We use a triangular–
quadrilateral mesh constituted by 1462 elements and fix the polynomial order to k = 5. We show in
Figure 12.7 snapshots of the unsteady simulation for Lx/U = 0.05 at two geometric AOAs (AOA =
17.2 and 28.6 deg.). We observe that the contours are smooth across the interfaces between
triangular and quadrilateral elements and also across the sliding mesh interface. In addition, the time
averaged forces for the static cases for AOA = 0, 10, 20 and 30 degrees have been included.
Comparison between static cases and dynamic simulations at low rotational speed show very good
agreement demonstrating the low error level introduced by the rotating mesh interface. We also note
that dynamic effects cause an increase in the vertical and horizontal components of the forces with
increasing rotational speed.
12.12.3 Case Study - Three Bladed Cross-Flow Turbine Case Study
This section presents the solver’s ability to compute flows through cross-flow turbines (Darrieus
type) in two dimensions. This type of turbine exhibits azimuthal changes in blade aerodynamics,
leading to complex flow-phenomena such as stalled flows, vortex shedding and blade-vortex
interaction. We consider two configurations: an unconstrained three bladed turbine and a three
bladed turbine placed in an elliptically shaped duct (constrained or ducted case). In addition to the
classic unconstrained configuration, the ducted case has been selected to demonstrate the ability of
the sliding mesh technique to simulate the relative rotational motion of the inner mesh (rotating with
the turbine blades) with respect to the outer static mesh (stationary with respect to the duct walls).
The three bladed turbine configuration consists of NACA0015 airfoils located at a radial distance
from the center of rotation of two chords R = 2c. The relative angle between blades is fixed at 120_ as
shown in Figure 12.9. The rotational speed is set to Lx/U = 0.5, where U is the free stream velocity
 333

magnitude and L = c. The Reynolds number based on the free stream velocity and the blade chord is
Re = 50. We note that at this low Reynolds number, three dimensionalities may be disregarded,
enabling purely 2D flow simulations.
The polynomial order is fixed to k = 3, the time step Dt to 0.0005 and we use two triangular–
quadrilateral meshes with curved NACA boundaries constituted of: 3320 elements for the

Figure 12.9 Solution Snapshots of a 3 Bladed Cross-Flow Turbine for Polynomial Order k = 3 for (a)
Unconstrained and (b) Ducted Cases.

unconstrained three bladed turbine and 3614 for the ducted three bladed turbine. We depict in
Figure 12.9 (a) and (b), snapshots of the velocity magnitude for the unconstrained and ducted
devices respectively. It can be seen that the DG solver is able to resolve the complex flow physics
arising from the rotational motion, loading and unloading of the blades, and the interaction with the
boundary layers of the inertially fixed duct walls.

Figure 12.8 Solution Snapshots of Flow Through a 3D Cross-Flow Turbine: (a) 11 pressure contours
[_1.6:0.8] and (b) iso-surfaces of z-velocity w = ±1 x10-6
 334

12.12.4 3D Flows
In this last section, we provide two examples of three dimensional flow simulations using the Fourier
extension of the DG solver with rotating sliding meshes. Firstly, the three bladed cross-flow turbine
geometry described is simulated with a rotational speed of Lx/U = 0.1 at a Reynolds number (based
on the freestream velocity and blade chord) of Re = 200 (i.e. large enough to trigger three
dimensionalities in the flow). The spatial discretization consists of polynomials of order k = 4 in the
x–y DG plane and 16 Fourier planes in the z-direction to discretize a spanwise length of Lz/c = 2,
where c is the blade chord. Figure 12.8 (a) shows pressure contours and Figure 12.8 (b) depicts
iso-surfaces of z-velocity showing the three dimensional character of the solution.
Secondly, we present a problem where a circular cylinder is shadowed by a pitching NACA0012
airfoil. This test case is presented to illustrate the possibility of locating static and rotating 3D
geometries together in a single mesh domain. In addition, all of the features presented in this paper
are included in a single simulation: high order sliding meshes and external curved boundary
condition for NACA airfoils and circular cylinders. The NACA airfoil is pitching with a rotational speed
of Lx/U = 0.5 and the Reynolds number based on the airfoil chord (which is equal to the cylinder
diameter) is Re = 200. As in the previous 3D case, the spatial discretization consists of polynomials
of order k = 4 in the x–y DG plane and 16 Fourier planes in the z-direction. The spanwise length is, in
this case, set to Lz/c = 2.5. Figure 12.10 shows the pressure contours and iso-surfaces of z-velocity.
It can be seen that the sliding mesh technique does not damage the numerical solution and captures
the pressure interactions between static and rotating geometries. In addition, Figure 12.10 (b)
shows that three dimensionalities are present in the wake of both geometries.

Figure 12.10 Solution Snapshots of a 3D Circular Cylinder Shadowed by a Rotating NACA Blade (a)
Pressure Contours (b) Iso-surfaces of z velocity w = ±1.5 x 10-6

12.12.5 Conclusions
This work details the formulation, implementation and verification of a novel approach to solve the
three dimensional incompressible NS equations using a high order DG-Fourier formulation with
rotating sliding meshes. The solver allows accurate solutions with curved elements that conform to
either arced surfaces (internal or external) and the profile of NACA 4-digit airfoils. Hybrid
unstructured meshes (i.e. with triangular and quadrilateral elements) can be handled without
damaging the solver’s high order numerical properties. To account for relative motion, the NS
equations are written in Arbitrary Lagrangian–Eulerian form and the sliding interface is given special
treatment that maintains the temporal convergence rates and spectral properties of the code when
rotational motion is applied to a region of the mesh. In addition, it has been shown that the high order
 335

properties are maintained when using body fitted NACA 4-digit boundaries. We have demonstrated
that geometric and functional incompatibilities do not represent a problem in the DG context and
that sliding mesh interfaces can be handled naturally without the necessity of mortar type
techniques. The solver accuracy has been verified with three test cases for two dimensional flows:
the Taylor vortex problem, a static and rotating NACA0015 airfoil and a three bladed cross flow
turbine under unconstrained and ducted flow conditions. These test cases have enabled verification
of the code as well as demonstrating its capabilities to simulate flow problems in which accuracy is
mandatory.
Three dimensional flow features are captured through a Fourier extension that allows high order 3D
solutions for rotating geometries that present a spanwise geometric homogeneity. 3D flow
simulation examples are provided for a three bladed cross-flow turbine and a pitching airfoil
shadowing a circular cylinder. The presented two and three dimensional simulations have
demonstrated the solver’s capability to compute complex flow physics for problems where static and
rotating geometries coexist in a single domain. We conclude that the novel approach to account for
sliding mesh interfaces and curved external boundaries provides a promising framework for
studying fluid–structure interaction problems involving rotation. For further evidence, please see the
[Ferrer and Willden]406.

12.13 Conceptual Differences Between 4 Most used Prediction Methods

The partial differential equations shown earlier in this section must be discretized or transformed
into a set of algebraic equations which can be solved digitally. There are numerous methods available
to do this discretization. The three most popular, based on the number of commercial (CFD) codes
are:
• Finite Difference Method (FDM)
• Finite Volume Method (FVM)
• Finite Element Method (FEM)
• Discontinuous Galerkin Methods (DS-FEM)
Rendering to [TAZ Engineering]407, finite element and finite volume methods (FEM and FVM) are
both based on dividing the flow domain into small cells, or volumes. These may possibly take any
shape (triangles, quadrilaterals, etc. in 2D; tetrahedra, prisms, cubes, polyhedral etc. in 3D). They are
thus more suitable for application considering complex flow geometries. Although may be applied to
numerically solve both solid mechanics and CFD problems, traditionally FEM found its way mostly to
solid mechanics but nonetheless may be applied (sometimes with a very sound basis as in the case
of the Galerkin procedure applied to the N–S Equations) to fluid problems, FVM tends to be more
popular in CFD and generally used in most noticeable major commercial packages. Finite difference
schemes can generally be applied to regular-shaped domains using body-fitted grids renders
problems of which large grid distortions cannot be avoided, as problematic to apply hence not easily
applied for very complex flow geometry shapes encountered in most industrial engineering oriented
fluid problems. Despite not being generally used in industrial codes, finite difference schemes are
useful for introducing the ideas of accuracy, truncation error, stability and boundlessness in a well-
defined and fairly transparent way and in the following paragraphs shall do just that. (see Figure
12.11).

406 Esteban Ferrer ⇑, Richard H.J. Willden, “A high order Discontinuous Galerkin – Fourier incompressible 3D
Navier–Stokes solver with rotating sliding meshes”, Department of Engineering Science, U. of Oxford, UK.
407 Tomer’s (Tom) Blog, ”Numerical Schemes in CFD: Up winding and The Cell-Reynolds Problem”, TAZ-

Engineering: Thermal Management, Computational Fluid Dynamics (CFD) and optimization to the level of art.
 336

Figure 12.11 Conceptual Differences Between Three Most used Prediction Methods

12.13.1 Finite Differencing Method (FDM)

In the Finite Difference Method (FD), the partial derivatives are replaced with a series expansion
representation, usually a Taylor series. The series is truncated usually after 1 or 2 terms. The more
terms you include, the more accurate the solution. However, more terms in the expansion causes the
complexity and number of discrete points or nodes of the solution to increase dramatically. Applying
this method to a regularly shaped geometry is straightforward. However, for irregularly shaped
geometries, the equations must be transformed before the Taylor series can be applied. This
transformation introduces all sorts of problems in terms of additional cross-coupling of equations,
mesh generation and general convergence408.
12.13.2 Finite Volume Method (FVM)
In the Finite Volume Method (FV), the governing equations are integrated over a volume or cell
assuming a piece-wise linear variation of the dependent variables (u, v, w, p, T). Again the piece-wise
linear variation determines both the accuracy and the complexity. Using these integrations, you
essentially balance fluxes across the boundaries of the individual volumes. The flux is calculated at
the mid-point between the discrete nodes in the domain. Hence, you must calculate a flux between
all neighboring nodes in the domain. In a topologically regular mesh (same number of divisions in
any one direction), this flux calculation is quite straightforward. In an irregular mesh (as in an
automatically generated tetrahedral mesh), this calculation will lead to an excruciating amount of
fluxes and a major bookkeeping effort to make sure all the fluxes have been calculated properly.
12.13.3 Finite Element Method (FEM)
In the Finite Element Method (FE), Galerkin's method of weighted residuals is generally used. In
this method, the governing partial differential equations are integrated over an element or volume
after having been multiplied by a weight function. The dependent variables are represented on the
element by a shape function, which is the same form as the weight function. The shape function may
take any of several forms. The main advantage as well as the main disadvantage of finite elements is

408 “Finite Element vs Finite Volume”, Autodesk Knowledge Network.

 337

that it is a mathematical approach that is difficult to put any physical significance on the terms in the
algebraic equations. In the finite volume method, you are always dealing with fluxes, not so with finite
elements. However, the application of finite elements on any geometric shape is the same. Also, the
boundary conditions which must be added after the fact for finite volume methods are an integral
part of the discretized equations. While it is true that early (FE) CFD methods struggled with
modeling high speed flows as many of the successfully applications demonstrated (FV) techniques to
the (FE) discretization method has produced a highly robust means of predicting not only high speed
turbulent flows, but compressible flows, as well. All of this has been accomplished with the strictest
application of Galerkin's method of weighted residuals. For simple geometries, you can show that
all 3 of these methods produce the exact same solution matrix or digital representation. All 3 methods
can produce similar sets of discretized equations for the governing equations of fluid flow and heat
transfer.
12.13.4 Discontinuous Galerkin Methods (DG)409
The discontinuous Galerkin (DG) methods are locally conservative, stable, and high-order accurate
methods which can easily handle complex geometries, irregular meshes with hanging nodes, and
approximations that have polynomials of different degrees in different elements. These properties,
which render them ideal to be used with hp-adaptive strategies, not only have brought these methods
into the main stream of computational fluid dynamics, for example, in gas dynamics, compressible
and incompressible flows, turbomachinery, magneto-hydrodynamics, granular flows, semiconductor
device simulation, visco-plastic crack growth and chemical transport, viscoelasticity and transport of
contaminant in porous media, but have also prompted their application to a wide variety of problems
for which they were not originally intended like, for example, Hamilton-Jacobi equations, second-
order elliptic problems, elasticity, and Korteweg-de Vries equations.
An introduction to DG methods can be found in the short monograph410. A history of their
development and the state of the art up to 1999 can be found in. Finally, a fairly complete and updated
review is given in. This paper is a short essay on DG methods which differs from the above mentioned
references in that it is intended for a wider audience and focuses on the exposition of the ideas behind
the devising of these methods as well as in the mechanisms that allow them to perform so well in
such a variety of problems. Table 12.1 compares and summarizes observation on 4 commercially
available predication methods as documented by [Hesthaven]411 of Brown University. Accordingly,
the scheme combines the high-order flexible element of FEM, with local statement of FVM.

Scheme Complex High-Order Explicit Semi- Conversation Elliptic

Geometries Accuracy Discrete Form Laws Problem
FDM No Yes Yes Yes Yes
FVM Yes No Yes Yes Yes
FEM Yes yes No Yes Yes
DG-FEM Yes Yes Yes Yes Yes

Table 12.1 Summarizing the Discontinuous Galerkin Finite Element Method

409 Bernardo Cockburn. Discontinuous Galerkin methods. Journal of Applied Mathematics and Mechanics /
Zeitschrift für Angewandte Mathematik und Mechanik, Wiley-VCH Verlag, 2003, 83 (11), pp.731-754.
10.1002/zamm.200310088. hal-01352444.
410 B. Cockburn, Discontinuous Galerkin methods for convection-dominated problems, High-Order Methods for

Computational Physics, edited by T. Barth and H. Deconink, Lecture Notes in Computational Science and
Engineering, Vol. 9 (Springer Verlag, Berlin, 1999), pp. 69–224.
411 Jan S Hesthaven, “Discontinuous Galerkin methods Lecture 1”, RMMC 2008.
 338

12.13.5 Comparing FV and FE for Spatial Convergence of 3D Benchmark Cases

Recently, [Park & Anderson]412 investigated the FV and FE schemes, both implemented within the
same flow solver to evaluated for several test cases. Different grid densities and types were used. The
cases include subsonic flow over a hemisphere cylinder, subsonic flow over a swept bump
configuration, and supersonic flow in a
square duct. The FV and FE schemes are
both used to obtain solutions for the
first two cases, whereas only the FV
scheme is used for the supersonic duct.
Altogether, the FE scheme produces
results on biased tetrahedral mesh
topology that are closer to the reference
solution than the FV scheme. But the
reader is advised not to take these
results as universal rule. The simulation
used FUN3D, a NASA research code, but
similar results is obtained by other
NASA linked codes such as CFD3D and
USM3D.
To solve nonlinear flow equations,
FUN3D uses a Hierarchical Adaptive
Nonlinear Iterative Method (HANIM).
The main objective in using (HANIM) is
the preconditioner alone (PA) baseline
solver technology to advance nonlinear
solution in pseudo-time where (HANIM) Figure 12.12 Propagation Mean Flow Residuals on
preconditioner based on defect- Different Mesh Density (Courtesy of Pandya, et al.)
correction scheme and point-implicit
Gauss-Seidel solution method413. It effectively results in speed up iterative convergence on finer grids
as shown in Figure 12.12. On the innermost level HANIM uses a preconditioner based on a defect-
correction method and iterates on a simplified first-order Jacobian with a pseudo-time term. One
preconditioner iteration involves a point- or line-implicit multi-color pass through the domain. The
number of preconditioner iterations may vary for different nonlinear iterations. A Generalized
Conjugate Residual (GCR) method uses the preconditioner solutions to converge linear residuals and
to compute solution correction [Diskin, et al.]414. Details regarding this, as well as other benchmark
cases can be found in415.
12.13.5.1 Case Study - Subsonic Flow Over 3D Swept Bump Configuration
The test case is a swept 3D bum where the finite-element and finite-volume solutions are obtained
on pure tetrahedral and pure hexahedral meshes ranging in size from just over 18 K nodes, up to
over 966 K nodes. Comparisons are made with simulations that have been computed on hexahedral
meshes with almost 59 M nodes. The 3D bump depicted in shows the surface mesh for representative

412 Michael A. Park_ and W. Kyle Anderson, “Spatial Convergence of Three Dimensional Turbulent Flows”, DOI:
10.2514/6.2016-0859.
413 Mohagna J. Pandya, Boris Diskin, James L. Thomas, and Neal T. Frink, “Assessment of Preconditioner for a

USM3D Hierarchical Adaptive Nonlinear Iteration Method (HANIM) (Invited)”, AIAA Science and Technology
Forum and Exposition (SciTech 2016), San Diego, California, January 4-8 2016.
414 Boris Diskin, James L. Thomas, Mohagna J. Pandya, Christopher L. Rumsey, “Reference Solutions for

Benchmark Turbulent Flows in Three Dimensions”, NASA Langley Research Center, Hampton, VA.
415 See previous.
 339

tetrahedral and hexahedral meshes, respectively. The flow conditions correspond to a freestream
Mach number of 0.2 and a Reynolds number of 3 M based on unit grid length. In the results that
follow, comparisons are again made with the results described in [Diskin et al.]416 for drag
coefficients, pressure distributions, and profiles of u-velocity and the turbulence working variable. A
summary of the mesh sizes used for this study is given in Table 12.2. Note that for the finest mesh,

Table 12.2 Tetrahedral Grids for Finite-Volume and Finite-Element Solutions

finite-volume results on hexahedra from [Diskin et al.]417 are used as reference solutions. Because
the tetrahedral were not used beyond mesh sizes of 966 K nodes, the number of tetrahedral for the
finest two meshes is designated with “NA." Pressure distributions, velocity profiles, and profiles of
the turbulence working variable computed using the finite-element and finite-volume results on
tetrahedral meshes have been obtained on grid levels 3-5, and are compared to corresponding results
on hexahedral meshes of the same density.
12.13.5.2 Results and Discussion
The tetrahedral meshes are generated directly from the hexahedral meshes by splitting each
hexahedra into six tetrahedral with a consistent stencil418. During this process, all the surface
triangles have been generated by splitting the surface quadrilaterals in identical directions, thereby
introducing a strong bias in the mesh that apparently has a stronger effect on the finite-volume
solutions than on the finite-element solutions. Meshes with more random dissection of the hexahedra
or meshes typically produced by unstructured grid generation, may mitigate the strong bias of these
tetrahedral grids. Surface pressure distributions along the y = 0 symmetry plane are shown in Figure
12.13 (a-b) which displays the Cp for finite-element and finite-volume discretization on tetrahedral
meshes ranging in size from 18 K nodes to 966 K nodes. The finite-volume reference solution, which
is computed on the mesh with almost 59 M nodes.

416 Diskin, B., Thomas, J. L., Pandya, M. J., and Rumsey, C. L., “Reference Solutions for Benchmark Turbulent Flows
in Three Dimensions," AIAA SciTech Forum and Exposition (SciTech 2016), AIAA.
417 See Previous.
418 Dompierre, J., Labb_e, P., Vallet, M.-G., and Camarero, R., “How to Subdivide Pyramids, Prisms, and Hexahedra

into Tetrahedra," 8th International Meshing Roundtable, 1999.

 340

(a) - Finite Element

(b) - Finite Volume

Figure 12.13 Convergence of Cp along symmetry plane (y = 0) using Tetrahedral cells

 341

13 Spectral and Other High Degree Methods

In the field of CFD, low-order methods are generally robust and reliable; as a result, they are
routinely employed in practical calculations. For the same computing cost, high-order methods can
provide considerably more accurate solutions, but they are more complicated and less robust419.
Spectral method, among the most widely used, are powerful methods for the solution of partial
differential equations. Unlike finite difference methods, spectral methods are global methods, where
the computation at any given point depends not only on information at neighboring points, but on
information from the entire domain. The numerical solution is expressed as a finite expansion of
some set of basis functions. When the PDE is written in terms of the coefficients of this expansion,
the method is known as a Galerkin Spectral Method. Spectral methods converged exponentially,
which makes them more accurate than local methods. Global methods are preferable to local
methods when the solution varies considerably in time or space, when very high spatial resolution is
required, and also when long time integration is needed420. In the past several years, the activity on
both theory and application of spectral methods have been concentrated on collocation spectral
methods. One of the reasons is that collocation methods deals with nonlinear terms more easily than
Galerkin Methods. The nonlinear terms are treated on a FD manner, via grid points values
multiplication. The underlying idea of a collocation spectral method is to approximate the
unknown solution in the entire computational domain by an interpolating high order
polynomial at the collocation points. The spatial derivatives of the solution are approximated by
the derivatives of the polynomial and the time derivative, if it exists, is solved through classical finite
differences schemes. Periodic and non-periodic problems are respectively treated with
trigonometric and algebraic polynomials. Some of the methods commonly used in the literature are
the Fourier collocation methods for periodic domains and the Jacobi polynomials for non-periodic
domains, with the Chebyshev and Legendre polynomials as special cases421.
Spectral methods and finite element methods are closely related and built on the same ideas; the
main difference between them is that spectral methods use basis functions that are nonzero over the
whole domain, while finite element methods use basis functions that are nonzero only on small
subdomains. In other words, spectral methods take on a global approach while finite element
methods use a local approach. Partially for this reason, spectral methods have excellent error
properties, with the so-called "exponential convergence" being the fastest possible, when the
solution is smooth. However, there are no known three-dimensional single domain spectral shock
capturing results (shock waves are not smooth). In the finite element community, a method where
the degree of the elements is very high or increases as the grid parameter h decreases to zero is
sometimes called a spectral element method. Spectral methods are computationally less expensive
than finite element methods, but become less accurate for problems with complex geometries and
discontinuous coefficients422.

13.1 Spectral Difference Method (SD)

The spectral difference (SD) method combines the prominent features of structured and
unstructured grid methods to achieve high computational efficiency and geometric flexibility. It
utilizes the concept of discontinuous and high-order local representations to achieve conservation
and high accuracy. Universal reconstructions are obtained by distributing unknown and flux points
in a geometrically similar manner for all unstructured cells. In this thesis, we consider first order (p0),

419 H.T. Huynh a,⇑, Z.J. Wangb, P.E. Vincent, “High-ordermethods for computational fluid dynamics: A brief review
of compact differential formulations on unstructured grids”, Computers & Fluids · June 2013.
420 Sigal Gottlieb and David Gottlieb (2009), Scholarpedia.
421 Bruno Costa, “Spectral Methods for Partial Differential Equations”, A Mathematical Journal Vol. 6, No 4,2004.
422 From Wikipedia.
 342

second order (p1) and third order (p2) schemes. The unknowns are updated using the differential
form of the conservation law equations by approximating the flux derivatives at these unknown
points. In order to obtain the flux derivatives, we use a polynomial reconstruction of the fluxes from
their values at available flux points to the unknown points. As a result, the method is defined as a
difference method. The SD method is similar to the multi-domain spectral method developed by
[Kopriva] and can be viewed as an extension of the multi-domain spectral method to a simplex
unstructured grid423. If the unsteady 2D Euler equations are considered, Q is a vector of conservative
variables, the viscous fluxes are zero vectors and F and G are the inviscid fluxes,

ρ  ρu   ρv 
ρu   ρu 2 + p   ρuv 
Q F G    
+ + =0 wher e Q = , F= , G= 2 
t x y ρv   ρuv   ρv + p 
     
E u(E + p)   v(E + p) 
Eq. 13.1
The current cell residual term R(Q) can be evaluated once the neighboring three cells are known. We
can denote the unknown and flux points for cell i as rj,i , and rk,i, respectively. The solutions of Q at
flux points can be conveniently constructed using a Lagrange-type polynomial basis function ,L j,i(r)
as
Np
Q(rk, j ) =  L j,i (rk,i )Q j,i Eq. 13.2
j=1

where Np=(n+1)(n+2)/2 is the number of unknown points required to support a degree p polynomial
construction as already illustrated in Figure 13.1. As a result, Q(rk) is continuous inside individual
cell element, while across the element interfaces, it is discontinuous and the inviscid fluxes F(rk) and
G(rk), are not uniquely determined.

Figure 13.1 Placement of Unknown ( Red Dot) and Flux (Blue ) Points for a Triangular Element; From
left to right - Case (a): First Order; Case (b): Second Order; Case (c): Third Order.

423Ravishekar Kannan, “High Order Spectral Volume and Spectral Difference Methods on Unstructured Grids”, A
Dissertation submitted to the graduate faculty in partial fulfillment of the requirements for the degree of Doctor
of Philosophy, Iowa State University, Ames, Iowa 2008.
 343

13.2 Spectral Volume Method (SV)

Spectral volume is a numerical method for conservation laws. It has all the desired properties of a
higher order scheme. The spectral volume (SV) method can be interpreted as a derivative of the finite
volume method where each cell (called spectral volume) is sub divided into a structured set of
smaller sub-cells (called control volume). This scheme exhibits many of the desirable properties of
the Discontinuous Galerkin (DG) method. It achieves high-order accuracy on unstructured grids and
has a compact stencil (as each SV only communicates only with its nearest neighbors), thus making
it suitable for parallel applications. As the DOFs are the control volume averages, the limiting
phenomenon can be applied at a control volume level. Thus the SV scheme can attain a higher
resolution for discontinuities than the DG method. In this thesis, we consider 2nd , 3rd and 4th order
SV schemes. Here, we following the development by [Kannan] closely.
13.2.1 Basic Formulation
Eq. 13.1 is considered in domain Ω with the appropriate initial and boundary conditions. Domain Ω
is discretized into I non-overlapping triangular (2D) cells. In the SV method, the simplex grid cells are
called SVs, denoted Si, which are further partitioned into CVs, denoted Cij, which depend on the
degree of the polynomial reconstruction. Figure 13.2 shows linear, quadratic and cubic partitions
in 2D.

Figure 13.2 Partitions of a Triangular SV . Linear, Quadratic and Cubic Reconstructions Publicized in
Case (a), Case(b) and Case(c) respectively

We need N unknown control volume solution averages (or DOFs). Np is calculated using the below
formula Np = (m+1)(m+2)/2 where m is the degrees of the polynomial, constructed using the CV
solution averages. The CV averaged conserved variable for Cij is defined as

1
Q (rkj ) =
Vij  QdV
Cij
, j = 1, , , N p , i = 1, , I Eq. 13.3

Where Vi,j is the volume of Cij. Given the CV averaged conserved variables, a m-th degree polynomial
can be constructed such that it is (m+1)- th order approximation to Q. In other words, we can write
the polynomial as
Np
pi (x, y) =  L j (x, y)Qi, j Eq. 13.4
j=1

where the shape functions Lj(x,y) satisfy

 344

1
Vij L
Cij
n (x, y)dV = δ jn Eq. 13.5

The G.E. (Eq. 13.1) is integrated over the Cij. This results in the below equation

K
∂𝐐 1
+ ∑ ∫ (𝐇⃗⃗ . 𝐧
⃗ ) dA = 0
∂t Vij Ar r=1
Eq. 13.6
where H = (F - Fv, G - Gv) , Ar represents the rth face of Cij, n is the outward unit normal vector of Ar
and K is the number of faces in Cij. The surface integration on each face is done using a (m+1)th order
accurate Gauss quadrature formula. The fluxes are discontinuous across the SV interfaces.

13.3 Fast Fourier Transform Methods (FFT)

The most accurate method for calculating spatial derivatives is to use Fast Fourier Transforms.
Instead of being (Δx)2, (Δx)6 or (Δx)n , the spectral, semi-spectral or Galerkin Methods converge to
the exact solution faster than any algebraic value of n < ∞. As Δx decreases, the rate of convergence
is exponential. That does not mean however, that for some finite value of Δx that the spectral method
will give a perfect solution424. The French mathematician [Fourier] discovered that periodic
waveforms can be modeled as the sum of sine and cosine waves called a Fourier series. The
mathematical expression is given in Eq. 13.1 for a generic signal with a known range. Application of
Eq. 13.7 for a square wave with increasing values of “n” yields to Figure 13.13, where it is apparent
that as n -> ∞ the square wave is more accurately reproduced.

1 
  nx   nx 
f(x) = a 0 +  a n cos   + b n sin  
2 n =1   L   L 
Eq. 13.7
The Fourier transform is used to represent a periodic function by a discrete sum of complex
exponentials. Instead of Eq. 13.7 above, we could equally write the summation in complex form:

 inπn L inπn
1
f(x) =  e L
, cn =  f(x) e L dx Eq. 13.8
− 2L −L

If we define a new variable k = 2πn⁄α we can define the Fourier Integral (after some math):


f(k) =  f(x) eikx dx (5.7)
−
Eq. 13.9
If we want a result for f(k), which is called the Fourier Transform f(x), we apply Eq. 13.9 in the limit
as approaches infinity with the Fourier series coefficients cn = ckα/2π. The Fourier transform is
closely related to a Fourier series and is used to represent a general function by a continuous
superposition (or integral) of complex exponentials. The Fourier transform decomposes an arbitrary

424 Computational Fluid Dynamics lecture series in Spectral Methods.

 345

waveform into its sine components, thus revealing its frequency content that might else be difficult
to detect. A Discrete-Time Fourier transform (DTFT) is a form of the Fourier transform that is
applicable to uniformly sampled continuous functions. The Fast Fourier Transform (FFT) takes
advantage of powers of two and is a numerically efficient implementation of computing DTFTs.
Simply stated, the FFT is an efficient numerical algorithm that allows for the transformation of a time
dependent signal into the frequency domain (or vice-versa)425.
13.3.1 Case Study - Flow and Fast Fourier Transform (FFT) Analyses for Tip Clearance Effect in an
Operating Kaplan Turbine426

Citation : Kim H-H, Rakibuzzaman M, Kim K, Suh S-H. Flow and Fast Fourier Transform Analyses for
Tip Clearance Effect in an Operating Kaplan Turbine. Energies. 2019; 12(2):264.
https://doi.org/10.3390/en12020264
The Kaplan turbine is an axial propeller-type turbine that can simultaneously control guide vanes
and runner blades, thus allowing its application in a wide range of operations (see Figure 13.3).
Here, turbine tip clearance plays a crucial role in turbine design and operation as high tip clearance
flow can lead to a change in the flow pattern,
resulting in a loss of efficiency and finally the
breakdown of hydro turbines. This research
investigates tip clearance flow characteristics and
undertakes a transient fast Fourier transform
(FFT) analysis of a Kaplan turbine. In this study,
the computational fluid dynamics method was
used to investigate the Kaplan turbine
performance with tip clearance gaps at different
operating conditions. Numerical performance was
verified with experimental results. In particular, a
parametric study was carried out including the
different geometrical parameters such as tip
clearance between stationary and rotating
chambers. In addition, an FFT analysis was Figure 13.3 Runner Vane of the Kaplan
performed by monitoring dynamic pressure Turbine
fluctuation on the rotor. Here, increases in tip
clearance were shown to occur with decreases in efficiency owing to unsteady flow. With this study’s
focus on analyzing the flow of the tip clearance and its effect on turbine performance as well as
hydraulic efficiency, it aims to improve the understanding on the flow field in a Kaplan turbine.
13.3.1.1 Background and Literature Survey
Kaplan hydro turbines are widely used in hydropower plants because of their high efficiency over a
wide range of operating conditions. A propeller-type turbine is suitable when the load on the turbine
remains constant. For a Kaplan turbine’s design and operation, tip clearance, which is formed by the
rotating runner blades and the stationary runner chamber427-428 is essential. Because of the clearance

425 A Guide to Random Data Analysis for Computational Fluid Dynamics, Published by Computer Aided
Engineering Associates Inc.
426 Hyoung-Ho Kim , Md Rakibuzzaman, Kyungwuk Kim and Sang-Ho Suh, “Flow and Fast Fourier Transform

Analyses for Tip Clearance E_ect in an Operating Kaplan Turbine”, Energies 2019.
427 Fox, R.W.; Mcdonald, A.T.; Pritchard, P.J.; Leylegain, J.C. Fluid Mechanics, 8th ed.; John Wiley and Sons, Inc.

Hoboken, NJ, USA, 2012

428 Wu, H.; Feng, J.J.;Wu, G.K.; Luo, X.Q. Numerical investigation of hub clearance flow in a Kaplan turbine. IOP

Conf. Ser. Earth Environ. Sci. 2012.

 346

gap between the blade tips and turbine casing of axial turbomachinery rotor blades, tip leakage flows
are expected. In both compressors and turbines, the tip region flow tends to include a pressure-
driven, oblique leakage flow from the pressure side to the suction side of the blade, and the roll-up of
a tip vortex in the corner bounded by the casing and the blade on the suction side. Leakage flows then
cause efficiency loss because of the increase in tip clearance in Kaplan turbines429.
An unsteady state analysis of hydro-turbines can be useful in predicting and analyzing the instability
caused by the unsteady flow field and in developing mitigating techniques to minimize the effects of
these phenomena430. [Wang et al.]431 investigated the characteristic frequencies in the unsteady
hydraulic behavior of a hydraulic turbine experimentally wherein their results showed that the
pressure fluctuation in a draft tube is stronger than that in the upstream flow passage. A more recent
study by [Su et al.]432 investigated the chaotic dynamic characteristics of pressure fluctuation signals
in hydro-turbines, and their results revealed that the main energy pressure fluctuations in a draft
tube are located at low-frequency regions. Also recently, [Glowacz]433 investigated the fault
diagnosis of a single-phase induction motor based on acoustic signals and then proposed a signal

Figure 13.4 3D Geometry of the prototype Kaplan Turbine

429 Roussopoulos, K.; Monkewitz, P.A. Measurements of Tip Vortex Characteristics and the Effect of an Anti-
Cavitation Lip on a Model Kaplan Turbine Blade. Flow Turbulence. Combust. 2000, 64, 119–144.
430 Anup, K.C.; Thapa, B.; Lee, Y.-H. Transient numerical analysis of rotor–stator interaction in a Francis turbine.

Renew. Energy 2014, 65, 227–235

431 Wang, F.J.; Li, X.Q.; MA, J.M.; Yang, M. Experimental investigation of characteristic frequency in unsteady

hydraulic behavior of a large hydraulic turbine. J. Hydrodyn. 2009.

432 Su, W.T.; Li, X.B.; Lan, C.F.; Ah, S.; Wang, J.S.; Li, F.C. Chaotic dynamic characteristics of pressure fluctuation

signals in hydro-turbine. J. Mech. Sci. Technol. 2016, 30, 5009–5017.

433 Glowacz, A. Fault diagnosis of single-phase induction motor based on acoustic signals. Mech. Syst. Signal

Process. 2018, 117, 65–80.

 347

processing method for early fault diagnosis of electrical and mechanical faults of rotating machines.
At present, the conventional methods for monitoring the stability of the fluid machinery include
pressure fluctuation, acoustic, output, and other inspections. In the running monitoring, the pressure
fluctuation is the major focus to identify the operating conditions. To extract the characteristics of
fluctuated signals, mathematical tools like the fast Fourier transformation (FFT) analysis is often
utilized, wherein the signals’ features in the time domain, frequency domain, or amplitude can be
obtained. Such methods are
significantly helpful for the
working stability and state
inspection of hydro-turbines.
Therefore, the unsteady
pressure fluctuations
characteristics in the tip
leakage flow play an
important role in load
instabilities.
With this, this study focuses
on the investigation of the tip
clearance flow of a Kaplan
turbine through numerical
simulations. Furthermore,
unsteady three-dimensional
turbulent flow throughout
the full domain of a Kaplan
turbine was investigated
through simulations, and the
pressure pulsation in the Figure 13.5 The Block Diagram
runner and generator (hub)
was predicted and analyzed using the FFT analysis. Geometrical Model and Meshing The 3D geometry
of the horizontal prototype Kaplan turbine was selected to analyze the flow characteristics of the
hydrological aberration as shown in Figure 13.4. The runner shape was scanned using a noncontact
portable 3D scanner from the operation site. During scanning, the 3D scanner generated the 3D
geometrical shape of the runner blade through the laser detection technique connected to the
computer; however, the 3D runner blade shape generated was not smoothened out. Therefore, the
gaps were filled, and the rough geometry was modified using CAD software ANSYS ICEM to get the
original 3D runner
shape. The block
diagram of the research
study is represented in
Figure 13.5.
The 3D geometry of the
casing, generator, guide
vane, runner hub, and
draft tube domains was
designed by ICEM-CFX
from the 2D drawing
information provided by
Chungju 2 Hydropower
Plant, K-Water, Korea. In Figure 13.6 Tip Cearance Gap of the Kaplan Turbine
this study, two model
 348

cases (cases 1 and 2) were invested. The tip clearance gap of case 1 was 1.75 mm and 6 mm (only
two blades) for case 2. Figure 13.6 shows the tip gap between the runner and runner chamber
(adopted from K-Water). In this figure, A, B, and C represent the gap height between the runner and
the runner housing. The tip gap of A was 6.05
mm, 4.80 mm for B, and 3.35 mm for C. The Description Dimension
runner tip clearance was considered the Runner outlet diameter 1648.25 mm
uniform gap, because it is difficult to make the Head 9.2 m
real shape as well as good grid of the runner. Flow Rate 75.3 m3/s
Table 13.1 shows the specifications of the Max. Power 6000 KW
model turbine. Rotational Speed 171.4 rpm
The model was meshed by ANSYS ICEM-CFX Runner Blade 4
based on a finite volume methods FVM) . 434 Guide Vane 16
Because of the flexibility of the complex
design of the hydraulic turbine when solving Table 13.1 Main Design Parameters of the
complex geometries, the unstructured prism prototype Kaplan Turbine
tetrahedron grid system was employed to
make the grid. The total meshing grids of case 1 were 3,167,233 nodes and 16,506,970 elements, and
4,937,129 nodes and 27,291,793 elements for case 2. To precisely simulate the flow in a whole
turbine channel, further grid refinement around the blades’ edges is required. In the blade end
surfaces, the volumes’ sizes were controlled as shown in [Hyoung-Ho Kim]435. Because of the complex
prototype geometry, the grid becomes large, which is needed for a comparatively fine grid, as
numerical simulations lead to a considerable amount of computational data. To reduce the influence
of grid number on the computational results, a grid independence test is important to check the
convergence of the simulation. The grid independence test was performed at the rated operating
condition (GV 67° and RV 23°), and the results found that the efficiency deviation was less than 1%
as shown in Figure 13.7. The mesh independence test was carried out based on the most accepted
grid convergence index (GCI) method
[30–32]. With this, the approximate and
extrapolated relative error can be
expressed as:

εnew − εold
εa = × 100%
εnew
Eq. 13.10
The grid convergence index (GCI) can be
written as:

1.25 × εa
GCI =
r2 − 1
Eq. 13.11
where εa is the relative error, and r is the Figure 13.7 Mesh Independence Test of the Kaplan
mesh ratio. The finite volume approaches Turbine
of the cells near the wall boundary are
irregular, thus possibly requiring a particular treatment. Prisms can first create a layer of regular

434 Ansys Inc. ANSYS-CFX, “Introduction, CFX Reference Guide, CFX Tutorials, CFX-Pre User’s Guide, CFX-Solver
Manager User’s Guide, Theory Guide”, Release 16.00; Ansys Inc.: Cannonsburg, PA, USA, 2016.
435 Hyoung-Ho Kim , Md Rakibuzzaman, Kyungwuk Kim and Sang-Ho Suh, “Flow and Fast Fourier Transform

Analyses for Tip Clearance E_ect in an Operating Kaplan Turbine”, Energies 2019.
 349

prisms near the wall and then

mesh the remaining volume with
tetrahedrons [33,34]. This grid
approach enhances the near
walls and gives better solutions
and convergence of
computational methods436. The
quality of the model turbine
grids is represented in Table
13.2. The estimated numerical
Table 13.2 Mesh Quality of the Kaplan Turbine
uncertainties in the hydraulic
turbine are shown in437. It shows,
the 3,167,233 grid density showed lower uncertainties. Also, the total meshed element and nodes
were different for each guide vane and runner vane opening angle as represented in [Hyoung-Ho
Kim]438 for cases 1 and 2, respectively.
13.3.1.2 Governing Equations
Numerical analysis of the fluid flow was based on the continuity and momentum equations439, which
are expressed as Eq. 13.12 as follows:

∂ui ∂ui ∂ui ∂p ∂ ∂ui

=0 , ρ( + uj )=− + (μ ̅̅̅̅̅
− ρu ′ u′ )
i j
∂xi ∂t ∂xj ∂xi ∂xj ∂xj
Eq. 13.12
where ρ and μ are density and dynamic viscosity respectively, p is the pressure scalar, ρu’iu’j is the
apparent turbulent stress tensor. For numerical simulation, the turbine domain was considered with
its steady-state, incompressible flow. The runner domain was rotating on the z-axis at a given
rotating speed, and the casing and draft tube were on a stationary domain. The turbine domain is
shown in Figure 13.8. The analysis boundaries were imposed on the inlet and the outlet as static
pressure of 90,221 Pa and 0 Pa on each domain model. All boundary walls were assumed to be
smooth with a nonslip condition. A frozen rotor was applied to couple the rotation and stationary
domain for steady analysis. The rotational speed was 171.4 rpm. Menter’s shear stress transport
(SST) turbulence model was used to solve the turbulence behavior of the flowing fluid440. The residual
value was of 1 x 10-5 was controlled by convection criteria. The unsteady simulation was carried out
to investigate the dynamic behavior of the model Kaplan turbine. The time step of the 3_ rotation of
the runner blade was selected. In general, 4–5 cycles would be enough to get the stable unsteady flow,
but because of the limitation of computer resources, two rotation cycles were made for case 1, and
five cycles for case 2. Therefore, the time step was 7.29 x 10-3 s. The total time for case 1 was 0.70 s,
and 1.75 s for case 2. The transient stator–rotor was accounted to couple the rotation and stationary
interface for the unsteady analysis. Furthermore, the FFT analysis of the Kaplan turbine investigated
the stability of the operating conditions.

436 Friziger, J.H.; Peric, M. Computational Methods for Fluid Dynamics, 3rd ed., Springer: New York, NY, USA, 2002.
437 See Previuos.
438 Hyoung-Ho Kim , Md Rakibuzzaman, Kyungwuk Kim and Sang-Ho Suh, “Flow and Fast Fourier Transform
Analyses for Tip Clearance E_ect in an Operating Kaplan Turbine”, Energies 2019.
439 Drtina, P.; Sallaberger, M. Hydraulic turbines—Basic principles and state-of-the art computational fluid

dynamics applications. Proc. Inst. Mech. Eng. Part C. 1999.

440 Wilcox, D.C. Turbulence Modeling for CFD, 1st ed., DCW Industries, Inc.: La Cañada Flintridge, CA, USA, 1994.
 350

Figure 13.8 Turbine Domain for Computational Analysis

13.3.1.3 Validation of Numerical Results

To confirm the numerical results, simulation results were validated with experimental data. Figure
13.9 show the comparison of the experiment and the numerical values at different operating
conditions for the tip clearance interval of 1.75 mm. It is seen from that figure that both results had
almost similar trends. Therefore, it was revealed that the results of the computer simulations agree
with experimental results.

Figure 13.9 Comparison Between Computed and Experimental Results as a Function of Runner Vane
Opening Angle

13.3.1.3.1 Performance Characteristics

Computer simulations were conducted at different flow rates by changing guide vane and runner
vane opening angles of the Kaplan turbine for cases 1 and 2. The guide vane opening angles varied
 351

from 23.5 to 72.0 deg, and the runner vane opening angles varied from -4 to 25 deg for cases 1 and 2.
The simulations were conducted with output ranging from 1000 to 6000 kW. For details regarding
the performance issues, please refer to [Hyoung-Ho Kim]441.

Figure 13.10 Tip Clearance (a) Velocity and (b) Pressure Profiles Between the Runner Blade and the
Shroud (case 2)

13.3.1.3.2 Effect of Tip Clearance

The analysis of the internal
flow characteristics was
performed at the rated
condition 67 deg and RV 23
deg), which is mostly
operated during power
generation. Tip clearance
flow creates a vortex that
separates from the runner
blade suction side and gives
rise to a local pressure
reduction near the tip on the
suction side of the runner
blade and at the center of the
tip vortex. In this region, if
the static pressure falls
below the vapor pressure of
the liquid water flowing
through the turbine, then
cavitation will occur. Figure
13.10 shows the tip
clearance velocity and Figure 13.11 Blade-to-Blade Velocity Contours

Hyoung-Ho Kim , Md Rakibuzzaman, Kyungwuk Kim and Sang-Ho Suh, “Flow and Fast Fourier Transform
441

Analyses for Tip Clearance E_ect in an Operating Kaplan Turbine”, Energies 2019.
 352

pressure profiles distribution in a plane view. It is observed that the tip clearance vortex impacts on
the shroud of the runner, where pressure suddenly drops into the center of the runner; here, pressure
is gradually increased, and velocity is velocity is decreased near the runner blade trailing edge. In this
case, a possible cavitation scenario could be formed inside the tip vortex.
Figure 13.11 shows the blade-to-blade velocity contour for the models for cases 1 and 2 at rated
conditions; here, for case 1, the flow velocities inside the turbine are well predicted. No recirculation
zone was found inside the runner and shroud tip gap. However, for case 2, the flow instability
occurred near the tip of the runner, velocity is increased at the center of the runner suction side, and
the flow is highly unsteady. Moreover, the tip clearance gap increases the turbulent kinetic energy
and reduces the output of the turbine. Figure 13.12 shows the turbulent kinetic energy in the
vicinity of the draft tube of cases 1 and 2 at a rated condition. It was observed that for case 1, the
turbulent kinetic energy at the exit of the runner outlet was symmetrical in the up and down
directions, and for case 2, it was asymmetric, which may be because of the instability of the flow near
the uneven tip gap of the runner.

Figure 13.12 Distribution of the Turbulence Kinetic Energy at Draft Tube

13.3.1.4 Conclusions
This study was based on the steady and unsteady flow analyses with varying tip gaps between the
runner and the discharge ring of the Kaplan turbine through a computerized flow analysis. For
numerical analysis, three-dimensional modeling of the existing Kaplan turbine was performed from
the drawing and scanning shape data. The performance analysis of the existing turbine was
accomplished by changing the runner vane opening from -4 to 25 deg and the guide vane opening
from 23.5 to 72.0 deg. The computed results were compared with the experimental data provided
by the manufacturer to verify the validity of the simulation. The performance of the standard tip gap
(1.75 mm) and that of an abnormal tip gap (6 mm) of the Kaplan turbines were also compared. It was
confirmed that the output of the turbine decreased when the interval was generated. In the
comparison of the results, in both cases, the flow rate of case 2 was 1.46% less than of the
conventional turbine, and the output difference was a maximum of 81 kW. In the case of efficiency,
the difference between both cases was only 4.59% within the rated range. The internal flow field and
the turbulent kinetic energy distribution was also observed and found that unsteady flow occurred
as the gap became higher than usual, and pressure fluctuation and period were confirmed at the
peripheral part of the vane. Through the FFT analysis, it was confirmed that the vibration caused by
the flow instability of the runner vane was transmitted to the generator. To reduce the dynamic
vibration effect of the turbine, the tip gap of the machine should be minimized. Therefore, repairing
 353

the runner and discharge ring gap within the normal range should be considered because this flow
may cause severe vibrations to the runner. The study notes that the simulation results did not
consider the bearing and mechanical loss of the turbine that occurred in the machine. The thrust
bearing, shaft seals, and guide bearings into the model could be a great interest for further research
combined with the FFT analysis.

13.4 Finite Differences vs. Spectral Methods

Finite difference methods are obtained by approximating a function u(x) by a local polynomial
interpolant, and the derivatives of u(x) are then approximated by differentiating this local
polynomial. Here local refers to the use of nearby grid points to approximate the function or its
derivative at a given point. For slowly varying functions, the use of local polynomial interpolants
based on a small number of interpolating grid points is very reasonable. Indeed, it seems to make
little sense to include function values far away from the point of interest in approximating the
derivative. However, using low-degree local polynomials to approximate solutions containing very
significant spatial or temporal variation requires a very fine grid in order to accurately resolve the
function. Clearly, the use of fine grids requires significant computational resources in simulations of
interest to science and engineering. In the face of such limitations we seek alternative schemes that
will allow coarser grids, and therefore fewer computational resources. Spectral methods are such
methods; they use all available function values to construct the necessary approximations.

Figure 13.13 Partial Sums of the Fourier Series of a Square Wave

13.5 Weighted Residual Method (WRM)

Weighted residual methods (WRM) assume that a solution can be approximated analytically or
piecewise analytically. In general a solution to a PDE can be expressed as a superposition of a base
set of functions
N
T(x, t) =  a j (t) j (x) Eq. 13.13
j=1
 354

where the coefficients aj are determined by a chosen method. The method attempts to minimize the
error, for instance, finite differences try to minimize the error specifically at the chosen grid points.
WRM’s represent a particular group of methods where an integral error is minimized in a certain
way and thereby defining the specific method. Depending on the maximization WRM generate
• Finite Volume Method,
• Finite Element Methods,
• Finite Difference Methods.
• Spectral Methods,
13.5.1 General Formulation
The starting point for WRM’s is an expansion in a set of base or trial functions. Often these are
analytical in which case the numerical solution will be analytical

N
T(x, y, z, t) = T0 (x, y, z, t) +  a j (t) j (x, y, z) Eq. 13.14
j=1

with the trial functions φj(x; y; z). T0(x; y; z; t) is chosen to satisfy initial or boundary conditions and
the coefficients aj(t) have to be determined. possible trial functions are φj(x) = xj-1 or φj(x) = sin(jπx).
The expansion is chosen to satisfy a differential equation L(͠T) = 0 (where T is the exact solution)

T 2T
L(T ) = −α 2 = 0 Eq. 13.15
t x
However, the numerical solution is an approximate solution, i.e., T≠ ̅T such that the operator L
applied to T produces a residual L (T) = R . The goal of WRM’s is to choose the coefficients aj such that
the residual R becomes small (in fact 0) over a chosen domain. In integral form this can be achieved
with the condition

 W m (x, y, z) R dx dy dz = 0 Eq. 13.16

where Wm is a set of weight functions (m = 1 , , , , M) which are used to evaluate above equation. The
exact solution always satisfies (6.2) if the weight functions are analytic. This is in particular true also
for any given subdomain of the domain for which a solution is sought. There are four main categories
of weight or test functions which are applied in WRM’s, namely,
• Subdomain Method
• Collocation Method
• Least Squares Method
• Galerkin Method
For excellent discussion regarding these method and general information, readers should consult442.

13.6 Boundary Element Method (BEM)

The boundary element method (BEM), also known as panel methods, is a numerical computational
method of solving linear partial differential equations which have been formulated as integral
equations (i.e. in boundary integral form). It can be applied in many areas of engineering and

442 Weighted Residual Methods, Chapter 6.

 355

science including fluid mechanics, acoustics, electromagnetics, and fracture mechanics. The integral
equation may be regarded as an exact solution of the governing partial differential equation. The
boundary element method attempts to use the given boundary conditions to fit boundary values into
the integral equation, rather than values throughout the space defined by a partial differential
equation. Once this is done, in the post-processing stage, the integral equation can then be used again
to calculate numerically the solution directly at any desired point in the interior of the solution
domain. BEM is applicable to problems for which Green's functions can be calculated. These usually
involve fields in linear homogeneous media. This places considerable restrictions on the range and
generality of problems to which boundary elements can usefully be applied. Nonlinearities can be
included in the formulation, although they will generally introduce volume integrals which then
require the volume to be discretized before solution can be attempted, removing one of the most
often cited advantages of BEM. The benefits of this approach is that it takes a 2D or 3D PDE and
reduces it to an algorithm that involves 1D or 2D computations only, respectively. In 2008,
[Baltazar]443 investigated a low-order potential-based panel method for the calculation of the
incompressible steady for his thesis. The potential flow around wings and marine propellers is
presented, with an on the modelling of the potential flow wings and marine propellers using a
boundary element method. In another attempt, same author presents the mathematical formulation
and the numerical implementation of a Boundary Element Method for the calculation of the steady
incompressible potential flow around delta wings with leading-edge vortex sheet separation, where
a partial wake relaxation model with given wake geometry is used [Baltazar]444.
13.6.1 Comparison to Other Methods
The boundary element method is often more efficient than other methods, including finite elements,
in terms of computational resources for problems where there is a small surface/volume ratio.
Conceptually, it works by constructing a mesh over the modelled surface. However, for many
problems boundary element methods are significantly less efficient than volume-discretization
methods (finite element method, finite difference method, finite volume method). A good example of
application of the boundary element method is efficient calculation of natural frequencies of liquid
sloshing in tanks. Boundary element formulations typically give rise to fully populated matrices. This
means that the storage requirements and computational time will tend to grow according to the
square of the problem size. By contrast, finite element matrices are typically banded (elements are
only locally connected) and the storage requirements for the system matrices typically grow quite
linearly with the problem size.

13.7 Dispersion-Relation-Preserving (DRP) Algorithms for 3D Problems

The Dispersion-Relation-Preserving (DRP) algorithms is to minimize the numerical dispersion
error in large-scale (3D) Finite-Difference Time-Domain (FDTD), as explored by [Wang and
Teixeira]445. The dispersion error is first expanded in spherical harmonics in terms of the
propagation angle and the leading order terms of the series are made equal to zero. Frequency-
dependent FDTD coefficients are then obtained and subsequently expanded in a polynomial (Taylor)
series in the frequency variable. An inverse Fourier transforation is used to allow for the

443 J. Baltazar, “On the Modelling of the Potential Flow About Wings and Marine Propellers Using a Boundary
Element Method”, University of Lisbon, 2008.
444 J. Baltazar, “Leading-Edge Vortex Flow Modelling Around Delta Wings Using a Boundary Element Method”,

September 2009.
445 S. Wang and F. L Teixeira, “Dispersion-Relation-Preserving FDTD Algorithms for Large-Scale Three-

Dimensional Problems”, IEEE Transactions on Antennas and Propagation, Vol. 51, No. 8, August 2003.
 356

incorporation of the new

coefficients into the FDTD updates.
Readers should also consult the
work by [Sriramkrishnan]446, where
an incompressbile NS on channel
flow been investigated for Re = 800
& 1200. In that particular article,
author was interested in transition
to turbulence. (See Figure 13.14).
Other prominent references in this
topic would be [Wang et al.]447.
13.7.1 Effects of AeroAcoustics
In contrast with (CFD), which has a
history of fast development,
computational aeroacoustics
(CAA) has only recently emerged as
a separate area of study448. CAA
focuses on the accurate prediction of
aerodynamically generated sound
as well as its propagation. Both
aspects of the problem, sound
generation and propagation, are Figure 13.14 Typical Cross Sectional View of a Square
Channel Showing Wall Bisectors and Corner Bisectors
enormously demanding in terms of
(Courtesy of Sriramkrishnan)
time-domain computation due to
the large number of grid points
normally required. For aeroacoustic simulation to become more practical, CAA schemes must thus
be higher order accurate and numerically optimized in order to reduce the required number of grid
points per wavelength while still ensuring tolerable levels of numerically induced dissipation and
dispersion. Numerical dissipation and numerical dispersion are the two primary sources of error
associated with computational schemes. Because of these errors, classical CFD schemes have been
found to be unsatisfactory for the study of wave propagation over long distances and large time
intervals. Many CFD schemes, such as the MacCormack scheme, upwind schemes, and essentially
nonoscillatory (ENO) schemes have been extended to high order using more stencil points for
application to acoustic problems. Recent reviews of computational aeroacoustics by [Tam]449and
[Wells and Renaut]450 have discussed various numerical schemes currently popular in CAA. In many
computational aeroacoustics applications, the DRP schemes have often been favored for their
accuracy and efficiency. DRP schemes, however, are implemented only on uniform Cartesian grids.
Practical problems in aeroacoustics, however, are seldom confined to uniform Cartesian grids with
the associated computing grids often being nonuniform Cartesian or curvilinear. Grid-optimized,

446 Sriramkrishnan M, “DNS of Three-Dimensional Incompressible Channel Flow”, A Report Submitted in Partial
Fulfillment of the Requirements for the Degree of Master of Technology, Department of Aerospace Engineering
Indian Institute of Technology, Kanpur May, 2014.
447 J. L. Wang, Q. B. Huang, Z. X. Liu, and K. Li., “Explicit High Accuracy Maximum Resolution Dispersion Relation

Preserving Schemes for Computational Aeroacoustics”, Hindawi Publishing Corporation,

448 Cheolung Cheong and Soogab Lee, “Grid-Optimized Dispersion-Relation-Preserving Schemes on General

Geometries for Computational Aeroacoustics”, Journal of Computational Physics 174, 248–276 (2001).
449 C. K. W. Tam, “Computational aeroacoustics: Issues and methods”, AIAA J. 33, 1788 (1995).
450 V. L. Wells and R. A. Renaut, “Computing aerodynamically generated noise”, Annual. Rev. Fluid Mechanic,

1997.
 357

dispersion-relation-preserving (GODRP) finite difference schemes are proposed, based on

optimization that gives finite difference equations locally the same dispersion relation as the original
partial differential equations on the grid points in the nonuniform Cartesian or curvilinear mesh451.

13.8 Lattice Boltzmann Method (LBM)

Lattice Boltzmann Methods (LBM) (or Thermal Lattice Boltzmann Methods (TLBM)) is a CFD
methods for fluid simulation. Instead of solving the Navier–Stokes equations, the discrete Boltzmann
equation is solved to simulate the flow of a Newtonian fluid with collision models such as
Bhatnagar–Gross–Krook (BGK). By simulating streaming and collision processes across a limited
number of particles, the intrinsic particle interactions evince a microcosm of viscous flow behavior
applicable across the greater mass452. It is a modern approach in Computational Fluid Dynamics and
often used to solve the incompressible, time-dependent Navier-Stokes equations numerically. Its
strength lies however in the ability to easily represent complex physical phenomena, ranging from
multiphase flows to chemical interactions between the fluid and the surroundings. The method finds
its origin in a molecular description of a fluid and can directly incorporate physical terms stemming
from a knowledge of the interaction between molecules. For this reason, it is an invaluable tool in
fundamental research, as it keeps the cycle between the elaboration of a theory and the formulation
of a corresponding numerical model short. At the same time, it has proven to be an efficient and
convenient alternative to traditional solvers for a large variety of industrial problems453. In LBM, the
fluid is replaced by fractious particles. These particles stream along given directions (lattice links)
and collide at the lattice sites. The LBM can be considered as an explicit method. The collision and
streaming processes are local. Hence, it can be programmed naturally for parallel processing
machines. Another beauty of the LBM is handling complex phenomena such as moving boundaries
(multiphase, solidification, and melting problems), naturally, without a need for face tracing method
as it is in the traditional CFD.

13.8.1 Approaches
There are two main approaches in simulating the transport equations (heat, mass, and momentum),
continuum and discrete454. In continuum approach, ordinary or partial differential equations can be
achieved by applying conservation of energy, mass, and momentum for an infinitesimal control
volume. Since it is difficult to solve the governing differential equations for many reasons
(nonlinearity, complex boundary conditions, complex geometry, etc.), therefore finite difference,
finite volume, finite element, etc., schemes are used to convert the differential equations with a given
boundary and initial conditions into a system of algebraic equations. The algebraic equations can be
solved iteratively until convergence is insured. Let us discuss the procedure in more detail, first the
governing equations are identified (mainly partial differential equation). The next step is to discretize
the domain into volume, girds, or elements depending on the method of solution. We can look at this
step as each volume or node or element contains a collection of particles (huge number, order of
1016). The scale is macroscopic. The velocity, pressure, temperature of all those particles
represented by a nodal value, or averaged over a finite volume, or simply assumed linearly or bi-
linearly varied from one node to another. The properties such as viscosity, thermal conductivity, heat
capacity, etc. are in general known parameters (input parameters, except for inverse problems). For
inverse problems, one or more thermos-physical properties may be unknown.

451 Cheolung Cheong and Soogab Lee, “Grid-Optimized Dispersion-Relation-Preserving Schemes on General
Geometries for Computational Aeroacoustics”, Journal of Computational Physics 174, 248–276 (2001).
452 CFD online
453 NIST is an agency of the U.S. Commerce Department, “Lattice Boltzmann Methods”, 2002.
454 A. A. Mohamad, “Lattice Boltzmann Method”, Fundamentals and Engineering Applications with Computer

Codes, ISBN 978-0-85729-454-8.

 358

On the other extreme, the medium can be considered made of small particles (atom, molecule) and
these particles collide with each other. This scale is microscale. Hence, we need to identify the inter-
particle (inter-molecular) forces and solve ordinary differential equation of Newton’s second law
(momentum conservation). At each time step, we need to identify location and velocity of each
particle, i.e., trajectory of the particles. At this level, there is no definition of temperature, pressure,
and thermo-physical properties, such as viscosity, thermal conductivity, heat capacity, etc. For
instance, temperature and pressure are related to the kinetic energy of the particles (mass and
velocity) and frequency of particles bombardment on the boundaries, respectively. This method is
called Molecular Dynamics (MD) simulations.
13.8.2 Dilute Gas Regimes
The rarefied gas dynamics can be best classified by Knudsen Number (Kn) where it is a measure of
collisions in a gas flow, and equals to mean free path of a molecule λ divided by L the flow length
scale. For rarefied gas, the Monte Carlo simulation has been used where it described as: Direct
simulation Monte Carlo (DSMC) method is the Monte Carlo method for simulation of dilute gas
flows on molecular level, i.e. on the level of individual molecules. To date DSMC is the basic
numerical method in the kinetic theory of gases and rarefied gas dynamics. It is basically a
generic numerical method for a variety of mathematical problems based on computer
generation of random numbers. Applications of DSMC simulations would be:
➢ Satellites and space crafts on LEO and in deep space
➢ Re-entry vehicles in upper atmosphere
➢ Nozzles and jets in space environment
➢ Dynamics of upper planetary atmospheres
➢ Fast, non-equilibrium gas flows (laser ablation, evaporation, deposition)
➢ Flows on microscale, microfluidics
to name a few455. The flow regimes of diluted gas, with the aid of Knudsen number, is demonstrated
in Figure 13.15.

Free molecular Transitional Flow Continuum Flow

(collisionless) Flow • Non-equilibrium • Local equilibrium
• Non-equilibrium flows • Kn << 1 (Kn < 0.01)
flows • Kn ~ 1
• Kn >> 1 (Kn > 10)

Figure 13.15 Flow Regimes for Diluted Gas

455 G.A. Bird, “Molecular gas dynamics and the direct simulation of gas flows”. Clarendon Press, 1994
 359

13.8.3 Middle of the Scale

As usual, in the process of , the book keeping plays a major role. We need to identify location (x, y, z)
and velocity (cx, cy, and cz are velocity components in x; y and z direction, respectively) of each
particle. Also, the simulation time step should be less than the particles collision time, which is in the
order of (10-12 s). Hence, it is impossible to solve large size problems (order of cm) by MD method.
At this scale, there is no definition of viscosity, thermal conductivity, temperature, pressure, and
other phenomenological properties. Statistical mechanics need to be used as a translator between
the molecular world and the macroscopic world. The question is, is the velocity and location of each
particle important for us? For instance, in this room there are billions of molecules traveling at high
speed order of 400 m/s; like rockets, hitting us. But, we do not feel them, because their mass
(momentum) is so small. The resultant effect of such a ‘‘chaotic’’ motion is almost nil, where the air
in the room is almost stagnant (i.e., velocity in the room is almost zero). Hence, the behavior of the
individual particles is not an important issue on the macroscopic scale, the important thing is the
resultant effects. What about a middle man, sitting at the middle of both mentioned techniques? the
lattice Boltzmann method (LBM). The main idea of Boltzmann is to bridge the gap between micro-
scale and macro-scale by not considering each particle behavior alone but behavior of a
collection of particles as a unit, Figure 13.16. The property of the collection of particles is
represented by a distribution function. The keyword is the distribution function. The distribution
function acts as a representative for collection of particles. This scale is called meso-scale. The
mentioned methods are illustrated in Figure 13.16. LBM enjoys advantages of both macroscopic
and microscopic approaches, with manageable computer resources. LBM has many advantages. It is
easy to apply for complex domains, easy to treat multi-phase and multi-component flows without a
need to trace the interfaces between different phases. Furthermore, it can be naturally adapted to
parallel processes computing. Moreover, there is no need to solve Laplace equation at each time step
to satisfy continuity equation of incompressible, unsteady flows, as it is in solving Navier–Stokes (NS)
equation. However, it needs more computer memory compared with NS solver, which is not a big
constraint. Also, it can handle a problem in micro and macro scales with reliable accuracy.

Molecular Dynamics Lattice Boltzmann Continuum

• Microscopic scale Method • Macroscopic Scale
• Hamilton s Equation • Mesoscopic scale • Finite Volume, Finite
• Boltzman Equation Element, etc
• Navier-Stokes Equations

Figure 13.16 Simulations Spectrum of Rarefied Gas

13.8.4 Maxwell Distribution Function

In 1859, Maxwell (1831–1879) recognized that dealing with a huge number of molecules is difficult
to formulate, even though the governing equation (Newton’s second law) is known. As mentioned
before, tracing the trajectory of each molecule is out of hand for a macroscopic system. Then, the idea
of averaging came into picture. The idea of Maxwell is that the knowledge of velocity and position of
each molecule at every instant of time is not important. The distribution function is the important
 360

parameter to characterize the effect of the molecules; what percentage of the molecules in a certain
location of a container have velocities within a certain range, at a given instant of time. The molecules
of a gas have a wide range of velocities colliding with each other’s, the fast molecules transfer
momentum to the slow molecule. The result of the collision is that the momentum is conserved. For
a gas in thermal equilibrium, the distribution function is not a function of time, where the gas is
distributed uniformly in the container; the only unknown is the velocity distribution function. For a
gas of N particles, the number of particles having velocities in the x-direction between cx and cx + dcx
is Nf (cx)dcx . The function f(dcx) is the fraction of the particles having velocities in the interval cx and
cx dcx; in the x-direction. Similarly, for other directions, the probability distribution function can be
defined as before. Then, the probability for the velocity to lie down between cx and cxdcx; cy and cy dcy;
and cz and cz dcz will be N f(cx) f(cy) f(cz) dcx dcy dcz: It is important to mention that if the above
equation is integrated (summed) over all possible values of the velocities, yields the total number of
particles to be N,

 f(c ) f(c ) f(c ) dc dc dc

x y z x y z =1
Eq. 13.17
Since any direction can be x, or y or z, the distribution function should not depend on the direction,
but only on the speed of the particles. Therefore,

f(cx ) f(cy ) f(cz ) =  (c2x + c 2y + c 2z )

Eq. 13.18
where φ is another unknown function, that need to be determined. The value of distribution function
should be positive (between zero and unity). Hence, in Eq. 13.18, velocity is squared to avoid
negative magnitude. The possible function that has property of Eq. 13.18 is logarithmic or
exponential function. After some manipulation and math, as well as help from kinetic theory, the
final form of distribution is

3  mc2 
−  
 m  2 
f(c) = 4π   ce
2  2kT 
Eq. 13.19
 2ππk 
Note that this function increases parabolically from zero for low speeds, reaches a maximum value
and then decreases exponentially. As the temperature increases, the position of the maximum shifts
to the right. The total area under the curve is always one, by definition. This equation called Maxwell
or Maxwell–Boltzmann Distribution function. For detail information, see 21.
13.8.5 Boltzmann Transport Equation
A statistical description of a system can be explained by distribution function f (r, c, t) ; where f (c, r,
t) ; c is the number of molecules at time t positioned between r and dr which have velocities between
c and cdc; as mentioned before. An external force F acting on a gas molecule of unit mass will change
the velocity of the molecule from c to c +F dt and its position from r to r +c dt . (see Figure 13.17).
The number of molecules, f (r, c, t) before applying the external force is equal to the number of
molecules after the disturbance, f (r + c dt , c + F dt , t + dt), if no collisions take place between the
molecules. Hence,

f(r + cdt , c + Fdt , t + dt) dr dc − f(r,c, t) dr dc = 0 Eq. 13.20

 361

However, if collisions take place between the molecules there will be a net difference between the
numbers of molecules in the interval drdc: The rate of change between final and initial status of the
distribution function is called collision operator, Ώ. Hence, the equation for evolution of the number
of the molecules can be written as,

f(r + cdt , c + Fdt , t + dt) dr dc − f(r,c, t) dr dc = Ω(f) dr dc dt Eq. 13.21

Dividing the above equation by dt

dr dc and as the limit dt → 0; yields

df (r, c, t)
= Ω(f)
dt
Eq. 13.22
The above equation states that the
total rate of change of the
distribution function is equal to
the rate of the collision. The Ώ is a
function of f and need to be
determined to solve the
Boltzmann equation. For system
without an external force, the Figure 13.17 Position and velocity vector for a particle after and
Boltzmann equation can be before applying a force, F
written as,
f (r, c, t)
+ c.f = Ω(f) wh ere c and f are vectors
t
Eq. 13.23
Eq. 13.23 is an advection equation with a source term (Ώ), or advection with a reaction term, which
can be solved exactly along the characteristic lines that is tangent to the vector c, if Ώ is explicitly
known. The problem is that Ώ is a function of f and Eq. 13.23 is an integral-differential equation,
which is difficult to solve. Therefore, there are several approximations available for Ώ. The relation
between the above equation and macroscopic quantities such as fluid density, q; fluid velocity vector
u, and internal energy e, is as follows

ρ(r, t) =  m f(r, c, t) dc
ρ(r, t) u(r, t) =  m c f(r, c, t) dc
1 3
ρ(r, t) e(r, t) =
2  m u a2 f(r, c, t) dc and e =
2m
k BT
Eq. 13.24
where m is the molecular mass and ua the particle velocity relative to the fluid velocity, the peculiar
velocity, ua = c – u. The conservation of mass, momentum, and energy are shown Eq. 13.24
respectively.
13.8.6 The BGKW Approximation
It is difficult to solve Boltzmann equation because the collision term is very complicated. The outcome
of two body collisions is not likely to influence significantly, the values of many measured quantities
 362

(Cercignani, 1990). Hence, it is possible to approximate the collision operator with simple operator
without introducing significant error to the outcome of the solution. Bhatnagar, Gross and Krook
(BGK) in 1954 introduced a simplified model for collision operator. At the same time Welander
(1954), independently, introduced similar operator. The collision operator is replaced as,

1 eq
Ω(f) = ω(f eq - f) = (f - f)
τ
Eq. 13.25
The coefficient ω is called the collision frequency and τ is called relaxation factor (ω =1/τ). The local
equilibrium distribution function is denoted by feq , which is Maxwell–Boltzmann distribution
function. After introducing BGKW approximation, the Boltzmann equation (Eq. 13.25), without
external forces can be approximated as,

f 1
+ c.f = (f eq - f)
t τ
f i (r + ci Δt , t + Δt) = f i (r, t) +
τ

Δt eq
f i (r, t) − f i (r, t) 
Eq. 13.26
The above equation is the working horse of the Lattice Boltzmann Method and replaces Navier–
Stokes equation in CFD simulations. It is possible to derive Navier–Stokes equation from Boltzmann
equation. The local equilibrium distribution function with a relaxation time determine the type of
problem needed to be solved. The beauty of this equation lies in its simplicity and can be applied for
many physics by simply specifying a different equilibrium distribution function and source term
(external force). Adding a source term (force term) to the above equation is straightforward.
However, there are a few concerns, which will be discussed in the following chapters. Also, the details
of implementing the above equation for different problems, such as momentum, heat and mass
diffusion, advection–diffusion without and with external forces. It is possible to use finite difference
or finite volume to solve partial differential. Some authors used this approach to solve fluid dynamic
problems on non-uniform grids. The main emphasis is to solve Eq. 13.26 in two steps, collision and
streaming.
13.8.7 Lattices & DnQm Classification
Lattice Boltzmann models can be operated on a number of different lattices, both cubic and
triangular, and with or without rest particles in the discrete distribution function456.A popular way
of classifying the different methods by lattice is the DnQm scheme. Here "Dn" stands for "n
dimensions", while "Qm" stands for "m speeds". For example, D3Q15 is a 3Dimensional Lattice
Boltzmann model on a cubic grid, with rest particles present. Each node has a crystal shape and can
deliver particles to 15 nodes: each of the 6 neighboring nodes that share a surface, the 8 neighboring
nodes sharing a corner, and itself (The D3Q15 model does not contain particles moving to the 12
neighboring nodes that share an edge; adding those would create a "D3Q27" model). Real quantities
as space and time need to be converted to lattice units prior to simulation. Non-dimensional
quantities, like the Reynolds number, remain the same. (see Figure 13.18).

456 From Wikipedia, the free encyclopedia.

 363

Figure 13.18 Real Molecules versus LB particles

13.8.8 Lattice Arrangements

In general, two models can be used for lattice arrangements, called D1Q3Q and D1Q5Q, as shown in
Figure 13.19. D1Q3 is the most popular one. The black nodes are the central node, while the gray
nodes are neighboring nodes. The factitious particles stream from the central node to neighboring
nodes through linkages with a specified speed, called lattice speed.
13.8.8.1 1D Lattice Boltzmann Method (D1O2)
The kinetic equation for the distribution function (temperature distribution, species distributions,
etc.), fk(x,t) can be written as:

f k (r, t) f (r, t) x
+ ck . k =Ω k ck = Eq. 13.27
t x t
where k = 1 , 2 (for one dimensional problem, D1Q2). The left-hand side terms represent the
streaming process, where the distribution function streams (advects) along the lattice link. The right-
hand term, X k , represents the rate of change of distribution function, f k , in the collision process. BGK
approximation for the collision operator can be approximated by

Ωk =
1
τ

(f k (x, t) - f keq (x, t)  Eq. 13.28

The term s represents a relaxation time toward the equilibrium distribution (f eq k ), which is related
to the diffusion coefficient on the macroscopic scale. The above equation is the working horse for the
diffusion problem in one dimensional space, which can be reformulated as,

f k (x + Δx, t + Δt) = f k (x, t)[1 - ω] + ω f keq (x, t)

Eq. 13.29
where ω=Δt/τ is the relaxation time. Error! Reference source not found. represents a number of e
quations for different k values (k ¼ 1 and 2), in each direction. The schematic diagram for three nodes
 364

with necessary linkages, central and neighboring nodes, where c1 = c, c2= -c. The dependent variable
can be related to the distribution function fi, as,

2 2
 ( x, t ) =  f k (x, t) , f eq
k = w k (x, t) and w k =1 Eq. 13.30
k =1 k =1

Figure 13.19 Lattice Arrangements for Velocity Vectors for Typical 1D, 2D and 3D Discretization

13.8.8.2 2D Lattice Boltzmann Method (D2Q9)

The process can be modelled using the Boltzmann transport equation, which is

f ( x, t)
+ u.f = Ω Eq. 13.31
t
where f(x,t) is the particle distribution function, u is the particle velocity, and is the collision
operator457. The LBM simplifies Boltzmann's original idea of gas dynamics by reducing the number
of particles and confining them to the nodes of a lattice. For a two dimensional model, a particle is
restricted to stream in a possible of 9 directions, including the one staying at rest. These velocities
are referred to as the microscopic velocities and denoted by ei, where i = 0, , , , 8. This model is
commonly known as the D2Q9 model as it is two dimensional and involves 9 velocity vectors. For
each particle on the lattice, we associate a discrete probability distribution function fi(x, ei, t) or
simply fi(x, t), i = 0 , , , 8, which describes the probability of streaming in one particular direction. The

457 Yuanxun Bill Bao & Justin Meskas,” Lattice Boltzmann Method for Fluid Simulations”, April 14, 2011.
 365

macroscopic fluid density and velocity can be defined as a summation of microscopic particle
distribution function,

8 8
1
 (x, t) =  f i (x, t) , u(x, t) =  f (x, t) Eq. 13.32
i =0  i =0
i

The key steps in LBM are the streaming and collision processes which are given by

Equilibrium
[fi (x, t) − fi (x, t)]
fi (x + ∆x, t + ∆t) − fi (x, t) = −
⏟ ⏟ τ
Steaming
Collision
Eq. 13.33
In the actual implementation of the model, streaming and collision are computed separately, and
special attention is given to these when dealing with boundary lattice nodes. In the collision term of
(Eq. 13.33), feq i (x, t) is the equilibrium distribution, and τ is considered as the relaxation time
towards local equilibrium. For simulating single phase owes, it success to use Bhatnagar-Gross-
Krook (BGK) collision, whose equilibrium distribution feqi is defined by

 e i .u 9 (e i .u) 2 3 u.u 
f (x, t) = w i ρ + ρs i (u(x, t))
i
eq
, s(u) = w i 3 + − 
 c 2 c2 2 c2 
 4/9 i=0
 Δx
w i =  1/9 i = 1,2,3,4 , c= is the lattice speed
1/36 i = 5,6,7,8 Δt

Eq. 13.34
The fluid kinematic viscosity in the D2Q9 model is relate d to the relaxation time by

2 - Streaming 3 - Compute
1 - Initialize ρ, step: move fi → macroscopic ρ
u, fi and feq fi* in the and u from f*i
direction of ei using Eq. (4.29)

5 - Collision step:
calculate the updated
6 - Repeat step 2 4 - Compute feq i
distribution function fi
to 5 using Eq. (4.20)
= f*i - 1/τ (f*i-feqi)
using Eq. (4.20)

Figure 13.20 Schematics of Solving 2D Lattice Boltzmann Model

 366

2 - 1 (x) 2
 = Eq. 13.35
6 t
The algorithm can be summarized in
Figure 13.20. Notice that numerical issues can arise as τ → 1/2. During the streaming and
collision step, the boundary nodes require some special treatments on the distribution functions
in order to satisfy the imposed macroscopic boundary condition.
13.8.9 Case Study 1 – Solving 2D Conduction with Heat Flux Boundary Using The Lattice
Boltzmann Method (LBM)458

Authors : Raoudha CHAABANE, Faouzi ASKRI, Sassi Ben NASRALLA

Appeared in : International Renewable Energy Congress, November 5-7, 2009- Sousse Tunisia
The lattice Boltzmann method (LBM) has been developed over the last decade as an alternative
promising tool for fluid flows. It has been widely used in many kinds of complex flows such as
turbulent flow, solar collectors, multiphase flow and micro flow. This article deals with the
implementation of the lattice Boltzmann method (LBM) for the solution of conduction problems with
heat flux and temperature boundary conditions. Problems in 2D rectangular geometries have been
considered. In the 2D geometry, the south and the north boundary are subjected to constant heat flux
condition. The remaining boundaries are at prescribed temperatures. The energy equation is solved
using the LBM. The results of the LBM have been found to compare very well with those available in
the literature.
13.8.9.1 Formulation and Kinetic Equation
We have considered transient heat conduction heat in a 2D rectangular geometry. Thermo-physical
properties of the medium are assumed constant. The system is initially at prelature T E . For time t >
0, the south and the north boundaries are subjected to heat fluxes qT ,S and qT ,N , respectively. The east
and the west boundaries are kept at temperatures TE and TW , respectively. For the problem under
consideration, and in the absence of convection and radiation, the energy equation is given by

∂T
= α∇2 T + Q
∂t
Eq. 13.36
Where α is the thermal diffusivity. The starting point of the LBM is the kinetic equation satisfies a
discretized evolution equation of the form 459

∂fi (r, t) 1
+ e̅i . ∇fi (r, t) = [fi (r, t) − fi0 (r, t)] , i = 1,2,3, , , , , b
∂t τ
Eq. 13.37
fi is the particle distribution function denoting the number of particles at the lattice node ȓ at time t
moving in direction i with velocity êi along the lattice link Δȓ = êi Δt connecting the nearest neighbors.
b is the number of directions in a lattice through which the information propagates. The basis of the

458 Raoudha Chaabane, Faouzi Askri, Sassi Ben Nasrallah, “Application of the lattice Boltzmann method for
solving conduction problems with heat flux boundary condition”, International Renewable Energy Congress
November 5-7, 2009 - Sousse Tunisia.
459 S. Succi, “The Lattice Boltzmann Method for Fluid Dynamics and Beyond”, Oxford University Press, New York,

2001.
 367

discrete velocity model is a finite set of virtual velocities êi or equivalently, of virtual fluxes of the
considered scalar field T(ȓ, t) which given by
𝑏

T(r, t) = ∑ 𝑓𝑖 (𝑟, 𝑡)
𝑖=0
Eq. 13.38
The observed flux is expressed by

∑ 𝑓𝑖 (𝑟, 𝑡) e⃗𝑖
𝑖=0
Eq. 13.39
The well-known D2Q9 lattice model (Figure
13.21) will be considered here. In that model,
the set of êi is such that they connect the point,
on which the lattice stencil is centered, to its
nearest neighbors on a spatial grid with
uniform spacing in both coordinate directions.
Any LBM advances the probability densities fi (ȓ Figure 13.21 Schematic Diagram of the D2Q9
, t) in time and thereby computes the evolution Lattice
of the considered scalar. In the absence of
external sources or fluxes for the scalar, the corresponding discrete evolution equation can be
written in the following general form (Eq. 13.37). It is a single-relaxation-time model with relaxation
constant that can be related, via Chapmann–Enskog analysis, to the diffusivity of the medium f0i is
the equilibrium distribution function. The relaxation time can be related with the thermal diffusivity,
the lattice velocity C and the time step 460 by the following relation

3α ∆t
τ= +
|c|2 2
Eq. 13.40
For the D2Q9 model in particular, the 9 velocities êi and their corresponding weights wi are the
following

i−1 i−1
e⃗0 = (0,0) , e⃗i = [cos ( π) , (sin π)] . c i = 1,2,3,4
2 2
2i − 1 2i − 1
e⃗i = √2 [cos ( π) , (sin π)] . c i = 5,6,7,8
2 2
4 1 1
w0 = wi = i = 1,2,3,4 wi = i = 5,6,7,8
9 9 36
Eq. 13.41
It is to be noted that in the above equations, C = Δx/Δt = Δy/Δt and the weights satisfy the relation
∑wi = 0. After discretization, and considering heat generation, equation can be written as

460 D.A. Wolf-Gladrow,“Lattice Gas Cellular Automata and Lattice Boltzmann Models : An introduction”, Springer
Verlag, Berlin-Heidelberg, 2000.
 368

∆t
f (r, e𝑖 ∆t, t + ∆t) = fi (r, t) = [f (r, t) − fi0 (r, t)] + w𝑖 ∆tQ∗
τ i
Eq. 13.42
where Q* is the non-dimensional heat generation and wi is the weight in corresponding direction. To
process equation, an equilibrium distribution function is required. For heat conduction problems,
this is given by
fi0 (r, t) = wi T(r, t)
Eq. 13.43
13.8.9.2 Results and Discussion
We consider transient heat conduction problems in 2D Cartesian geometry with the following
conditions:
13.8.9.2.1 Case Study 1 - Four Boundaries are at Known Temperatures
The initial and the boundary conditions for cases (1) are the following

T(x, y, 0) = Tref , T(x, 0, t) = 0.25Tref

T(x, Y, T) = T(0, y, t) = T(X. y, t) = Tref
Eq. 13.44

Figure 13.22 Centerline(x/X=0.5) Temperature Evolution for Different Instants (Case 1)

Courtesy of [Chaabane et al.]
 369

Steady state conditions were assumed to have been achieved when the temperature difference
between two consecutive time levels at each lattice center did not exceed10-6. Non dimensional time
was defined as ξ = αt/L2 where L is the characteristic length. Δξ was taken as 10-4. To check the
accuracy of the present LBM algorithm, the same problem was solved using the finite volume method
and the results given by the two algorithms are compared with those available in the literature. In
Figure 13.22, the non-dimensional centerline (x/X = 0.5) temperature has been compared at
different instants ξ for the case (1).
13.8.9.2.2 Case Study 2 - Effects of Heat Generation and the Four Boundaries are at Specified
Temperatures

Citation : Subhash C. Mishra, Bittagopal Mondal, Tanuj Kush, B. Siva Rama Krishna, Solving transient
heat conduction problems on uniform and non-uniform lattices using the lattice Boltzmann method,
International Communications in Heat and Mass Transfer, Volume 36, Issue 4, 2009, Pages 322-328,
ISSN 0735-1933, https://doi.org/10.1016/j.icheatmasstransfer.2009.01.001.
In Figure 13.23, the effects of volumetric heat generation are shown. The non-dimensional
volumetric heat generation is taken as unity. Effect of heat generation is very less in the beginning
compared to steady state because it takes some time to influence the temperature profile. For the 2D
geometry, the number of iterations for a 50 x 50 grid is (3719) compared to that cited at the literature
(3257)461.

Figure 13.23 Comparison of Centerline (x/X=0.5) Temperature in the Presence/Absence of Heat

Generation (Case 2) - Courtesy of [Chaabane et al.]

461S.C. Mishra, M. Bittagopal, K. Tanuj, B.S.R. Krishna, “Solving transient heat conduction problems on uniform
and non-uniform lattices using the Lattice Boltzmann Method”, International Communications in Heat and Mass
Transfer, 36, pp.322-328, 2009.
 370

13.8.9.2.3 Case Study 3 - The Bottom and Top Boundaries are at Prescribed Fluxes and Remaining
Two Boundaries at Known Temperatures
It is seen from the Figure 13.24 that the
steady state results match exactly which
each other.

T(x, y, 0) = T0
q(x, 0, t) = qs
q(x, Y, t) = qn
T(x, y, T) = T(X, y, t) = T0
Eq. 13.45
13.8.9.3 Conclusions
The LBM is used to solve transient heat
conduction problems in two
dimensional geometries with uniform
lattices having constant temperature
and/or flux boundary conditions. Effect
of heat generation is also studied. The Figure 13.24 Centerline(x/X=0.5) Temperature
same problems are solved using the Evolution for Different Instants (Case3) - Courtesy of
finite volume method. The results given [Chaabane et al.]
by the two numerical approaches are
compared with those available in the
literature and good agreement is
obtained. On the other hand, the effect of
lattice size is highlighted via the number
of iterations and the CPU time. The
considered 2D geometry is a simple one,
to allow simple validation. Advection and
radiation are omitted. Thus, it remains to
demonstrate the viability of the LBM as
heat diffusion-advection solver. For
further details, please consult the work by
[Chaabane et al.]462 and [Hamila et al.]463.
13.8.9.3.1 Case Study 4 - Lid-Driven
Cavity Flow
In this simulation, performed by [Bao &
Meskas]464, we have a 2D fluid flow that is
driven by a lid at the top which moves at a
speed of ũ(x ,1, t) = Vd in the right Figure 13.25 Color plot of the norm of the velocity: Re =
direction. The other three walls have 1000, ν = 1/18, τ = 2/3, 256 x 256 lattice

462 Raoudha Chaabane, Faouzi Askri, Sassi Ben Nasrallah, “Application of the lattice Boltzmann method for
solving conduction problems with heat flux boundary condition”, International Renewable Energy Congress
November 5-7, 2009 - Sousse Tunisia.
463 Rihab Hamila, Raoudha Chaabane, Faouzi Askri, Abdelmajid Jemni, Sassi B. Nasrallah, “Lattice Boltzmann

method for heat transfer problems with variable thermal conductivity”, International Journal of Heat And
Technology, June 2017.
464 Yuanxun Bill Bao & Justin Meskas, “Lattice Boltzmann Method for Fluid Simulations”, April 14, 2011.
 371

bounce-back BCs for the distribution function and no-slip BCs for the velocity, u = 0 and v = 0. The
initial conditions state that the velocity field is zero everywhere and the initial distribution function
is set by the weights, fi = wi. This results in an initial condition that ρ = 1 from. The only exception is
the velocity of the fluid on the top is set to be Vd. The two top corner lattice points are singular points
and are considered as part of the moving lid. Simulations were done with a 256 x 256 lattice, ν = 1/18,
τ = 2/3 and with Re = 400 and 1000. Figure 13.25 shows a color visualization of the simulation for
Re = 1000.

13.9 Immersed Boundary Method (IBM)

The immersed boundaries (IB) method allows
one to greatly simplify the grid generation and
even to automate it completely. The governing
equations are solved directly on a grid in their Fluid Cells
simplest form by means of very efficient
numerical schemes. The grid generator
detects the cell faces that are cut by the body
surface and divides the cells into three types:
solid and fluid cells, whose centers lie within Interface Cells
the body and within the fluid, respectively;
and fluid/solid interface cells, which have at
least one of their neighbors inside the
body/fluid. Then, the centers of the fluid and
Solid Cells
solid-interface cells are projected onto the
body surface along its normal direction, so as
to obtain fluid-cells projection points and
solid-cell projection points, (see Figure
Figure 13.26 Immersed Boundaries
13.26).
13.9.1 Case Study - Immersed Boundary
Methods (IBM)
Author : Henry Bandringa,
Acknowledgments : Master Thesis in Applied Mathematics, Institute of Mathematics and Computing
Science, University of Groningen, The Netherlands, 2010.
Usually, a body conforming grid is being used for computing the flow around an arbitrary body. This
approach requires coordinate transformations and/or complex grid generation. Especially for
moving bodies, in every time step a new mesh has to be generated, which requires a lot of computing
time. In contrast, an immersed boundary method (IBM) uses a non-body conforming Cartesian grid.
In the immersed boundary method the fluid is represented on an Eulerian coordinate and the
structure is represented on a Lagrangian coordinate. In general a body does not align with the grid,
so some of the computational cells will be cut. The cut cells can be treated in several ways. These
treatments can be globally categorized into a continuous forcing approach, a discrete forcing
approach and cut-cell methods. Many different immersed boundary methods have been developed
over the years and one of my tasks was to recommend a few of these methods, which could be
implemented in the CFD code.
13.9.1.1 Introduction
An (Cartesian grid-based) immersed boundary method is a methodology for dealing with a body
which does not necessarily have to fit conform a Cartesian grid, see for example Figure 13.27. If
that's the case, the solid boundary will cut through this grid. Because the grid does not conform to
the solid boundary, imposing the boundary conditions will require modifying the governing
 372

equations in the vicinity of the

boundary. What these
modifications exactly are, will be
treated in the subsequent
sections. The objective of this
thesis is to find an immersed
boundary method that can
simulate unsteady viscous
turbulent flows with high
Reynolds numbers and also
works in three dimensions. And
if possible, this method should
also be able to simulate moving
boundaries. This method should
be implemented on a Cartesian Figure 13.27 Immersed Boundary illustration; Eulerian Mesh (¯x)
grid with a staggered and Lagrangian Mesh ( ¯ xk) [53]
positioning of the discrete
variables.
13.9.1.2 General Considerations
One of the advantages of using an immersed boundary method is that grid generation is much easier,
because a body does not necessarily have to fit conform a Cartesian grid. Another benefit is that grid
complexity and quality are not significantly affected by the complexity of the geometry when carrying
out a simulation on a non-boundary conforming Cartesian grid. Also, an immersed boundary method
can handle moving boundaries, due to the stationary non-deforming Cartesian grid. As a result of the
above remarks, an immersed boundary method uses less memory and CPU compared to the usual
method, a body fitted grid and the thereby belonging transformations. In comparison with structured
curvilinear body-fitted grids, Cartesian grids reduce the per-grid-point operation count due to
absence of additional terms associated with grid transformations.
When comparing to unstructured curvilinear body-fitted grids, Cartesian grid-based immersed
boundary (IB) methods are agreeable to powerful line-iterative techniques and geometric multigrid
methods, leading to a lower per-grid-point operation count. Also multi-phase and multi-material
problems, where the interface is between different materials, can be regarded as immersed boundary
problems. A disadvantage is that imposing of the boundary conditions is not straightforward
compared to the traditional methods. Also, the ramifications of the boundary treatment on accuracy
and conservation properties of numerical schemes are not trivial. Another drawback is the following.
Alignment between grid lines and body surface in boundary conforming grids allows for better
control of the grid resolution in the vicinity of the body and this has implications for the increase of
grid size with increasing Reynolds numbers. However, a substantial fraction of grid points can be
inside the solid body, i.e. where the fluid flow equations need not to be solved.
13.9.2 Categories of IB Methods
The way the boundary conditions are imposed on the immersed boundary determines for the most
part how an IB algorithm will look like. It is also what distinguishes one IB method from another. The
simulation of a viscous incompressible flow past a body is described by the Navier-Stokes equations
(the governing equations)

∂𝐮
ρ( + 𝐮. ∇𝐮) + ∇p − μ∆𝐮 = 0
∂t
 373

∇. 𝐮 = 0 in ΩΓ and u = uΓ on Γb
Eq. 13.46
where u(x, t) is the fluid velocity and p(x, t) is the pressure.
The coefficients ρ and μ are the constant fluid density and
viscosity, respectively. The solid body occupies the domain
b, with boundary denoted by Гb, and the surrounding fluid IBM District
domain denoted by f . In an IB method, Eq. 13.46 will be
discretized on a non-boundary conforming Cartesian grid
and the boundary condition will be imposed indirectly
through modifications of Eq. 13.46. In general, the Forcing Cut-Cell
modification takes the form of a forcing function in the Function Approach
governing equations that reproduces the effect of the
boundary. There is also another approach available: the so-
called cut-cell approach. The first category will be
explained first in this section. The second category will be Continuous
shortly described further on in this section. Introducing a
forcing function, leads to a division of IB methods into two
groups [42], namely continuous forcing and discrete (or Figure 13.28 Hierarchy of IB
direct) forcing. See Figure 13.28. Sometimes these Methods
approaches are called diffuse respectively sharp interface
methods. In the first approach, the forcing function,
denoted here by f, is included into the momentum equation leading to the equations

∂𝐮
ρ( + 𝐮. ∇𝐮) + ∇p − μ∆𝐮 = 𝐟
∂t
Eq. 13.47
which are then applied to the entire domain (Ωf + Ωb). Many methods have been developed for
choosing the external body force f. Some of these methods will be described further on in this report.
After choosing an appropriate forcing function, the equations are subsequently discretized on a
Cartesian grid, and the equations are solved in the entire domain. In the discrete forcing approach,
the governing equation Eq. 13.47 are discretized on a Cartesian grid neglecting the immersed
boundary, resulting in a set of discretized equations. After that, the discretization in the cells near the
IB is adjusted to account for their presence, i.e. the grid points in the vicinity of the immersed
boundary will be computed using a interpolation scheme.
In the cut-cell approach the boundary conditions at the immersed boundary are not imposed by a
forcing function. Instead, this approach requires truncating the Cartesian cells at the immersed
boundary to create new cells which conform to the shape of the surface. This reshaping may result in
very small cells, which has a negative impact on the numerical stability. One remedy for this problem
is using a cell-merging strategy.
13.9.3 Continuous Forcing Approach With Boundary Imposition
Several methods make use of a continuous forcing approach. Examples are methods for elastic
boundaries, methods for rigid boundaries (i.e. boundaries which are fixed) and the distributed
Lagrange multiplier method, see for instance [24] for more methods belonging to this category. These
three methods will be discussed in more detail.
13.9.3.1 Elastic Boundaries
[Peskin] introduced in 1972 the concept of immersed boundary methods [48, 49], where he used this
method to compute ow patterns around heart valves. Peskins method is a mixed Euler-Lagrangian
finite-difference method for computing the flow interaction with a flexible immersed boundary. In
 374

this method the fluid flow is governed by the incompressible Navier-Stokes equations and these are
solved on a stationary Cartesian grid. The IB is represented by a set of massless elastic fibers and the
location of these fibers is tracked in a Lagrangian fashion by a collection of massless points that move
with the local fluid velocity
∂𝐗
(s, t) = 𝐮(𝐗(s, t))
∂t
Eq. 13.48
Here, the boundary configuration is described by the curve X(s, t), where s is a parameter chosen in
such a way that a given value of s represents a given physical point of the boundary for all times t.
Figure 13.27 shows such a configuration.
[Peskin] defines the force density, f (x, t), by a δ-function layer that represents the force applied by
the immersed boundary to the fluid. The problem of this definition is that the location of the fibers
does not generally coincide with the nodal points of the Cartesian grid. Therefore, the forcing is
distributed over a band of cells around each Lagrangian point (see Figure 13.29(a)) and this
distributed force will be used in the momentum equations of the surrounding nodes. By replacing
the sharp δ-function with a smooth distribution function, denoted by d, this new forcing function will
be more suitable for use on a discrete mesh. Due to the fibers, the forcing at any grid point x is then
given by

𝐟(𝐱, t) = ∫ 𝐅(s, t)δ(𝐱 − 𝐗(s, t))ds

Γb
Eq. 13.49
The boundary force, F(s,t), on the particular segment at time t is determined by the boundary
configuration at time t
𝐅(s, t) = 𝐒(𝐗(, t), t)
Eq. 13.50
where the S satisfies a generalized Hooke's law if the boundary is elastic. There are more approaches

Figure 13.29 Effect of Explicitly Adding a Forcing Function, [42].

 375

for the distribution function developed over the years and some of them are shown in Figure 13.29
(b).
13.9.3.2 Rigid Boundaries
The first approach for rigid boundaries is called virtual boundary method, used by Goldstein et al.
[21]. The main idea of the virtual boundary method is to treat the body surface as a virtually existent
boundary embedded in the fluid. This boundary applies force on the fluid so that the fluid will be at
rest on the surface (no-slip condition). Let's denote the boundary Гb by { Xe(s) : 0 ≤ s ≤ Lb }. The force
F(s, t) on the boundary is determined by the requirement that the fluid velocity u(x, t) should satisfy
the no-slip condition on the boundary

∂𝐮
ρ( + 𝐮. ∇𝐮) + ∇p − μ∆𝐮 = 𝐟
∂t
𝐟(𝐱, t) = ∫ 𝐅(s, t)δ(𝐱 − 𝐗 𝒆 (s))ds
Γb
∇. 𝐮 = 0 in ΩΓ and u = uΓ on Γb and
0 = 𝐮(𝐗 𝒆 (s), t) ≡ ∫ 𝐮(x, t)δ(𝐱 − 𝐗 𝒆 (s))dx
Ω
Eq. 13.51

Figure 13.30 (a) Schematic showing a generic body past which flow is to be simulated. The body
occupies the volume Ωb with boundary Гb. The body has a characteristic length scale L, and a boundary
layer of thickness δ develops over the body. (b) Schematic of body immersed in a Cartesian grid on which
the governing equations are discretized.

Since the body force is not known a priori, it must be calculated in some feedback way in which the
velocity on the boundary is used to determine the desired force. In the virtual boundary formulation,
the force is expressed as
𝑡
𝐅(s, t) = α ∫ 𝐮(s, τ)dτ + β𝐮(s, t)
0
Eq. 13.52
 376

where u is the fluid velocity at these surface points. The particular form given in Eq. 13.52, can be
seen as a PI controller, where P stands for the proportional part and I for the integrating part. When
there is also a differentiating part included, the formula can be described as a PID controller. This
construction of Eq. 13.52 seems reasonable given that this formula is a feedback mechanism. When
α and β are chosen negative and large enough in magnitude, then u will stay close to its prescribed
value. To avoid interpolating the velocity field from grid points to the boundary points, Goldstein et
al. let the boundary points coincide with grid points.
However, in order to generate a smooth surface rather than a step-like surface, the boundary force is
multiplied by a narrow Gaussian distribution so that the nearby grid points can receive a part of the
force influences. Although this local smoothing will blur the location of the surface within one grid
cell, the method can produce promising results if sufficient spatial resolution is used [34]. A
disadvantage of feedback forcing is that this not only may induce spurious oscillations but also
restricts the computational time step associated with numerical stability. Especially for highly
unsteady flows, stability problems arise due to considerable stiffness. Saiki et al. [50] extended this
feedback forcing approach, such that the spurious oscillations caused by the applied feedback forcing
term at the boundary are eliminated. They modified Eq. 13.52 into the area-weighted average
function
𝑡
𝐅(𝐱 𝑠 , τ) = α ∫ [𝐮(𝐱 𝑠 , τ) − 𝐯(𝐱 𝑠 , τ)]dτ + β[𝐮(𝐱 𝑠 , t) − 𝐯(𝐱 𝒔 , t)]
0
Eq. 13.53
where the velocity of the body itself is controlled by specifying v at the boundary points. By
employing this function a better interpolation of the fluid velocity at the boundary points is
developed. They used fourth-order central difference approximations. The use of finite-differences
avoids the appearance of spurious flow oscillations at the boundary.
Also an appropriate distribution of nodal boundary forces at these grid points has been made using
this formula. If the body moves, i.e. v ≠ 0, then the position of the boundary points at each time step
is computed by integration of v = dxs/dt . They showed that the feedback-force IB method is capable
of handling the solid boundary problems including also moving boundaries. Details pf other
approaches such as a special version of the virtual boundary method used by [Lai et al. [34], a
penalty method which has been developed by [Khadra et al. [30], is the immersed interface method
(IIM), [Lee et al. [36], and solving linear system of [Le et al. [35], can be obtained from the
[Bandringa]465.
13.9.3.3 General Considerations
An advantage of the continuous forcing approach is that the above described methods are
independent of the underlying spatial discretization in contrast to methods that are based on a
discrete forcing approach. Therefore, this approach can be implemented into an existing Navier-
Stokes solver with relative ease. A disadvantage of these methods is that the smoothing of the
forcing function inherently leads to an inability to provide a sharp representation of the
immersed boundary and therefore these methods are not useful for high Reynolds number flows.
Another drawback of the continuous approach is that they all require the solution of the governing
equations inside immersed body. With increasing Reynolds numbers the proportion of grid points
inside the IB also increases.
13.9.4 Direct Forcing Approach (Discrete Approach)
The discrete approach is better suited for higher Reynolds numbers, due to imposing the velocity
boundary conditions at the immersed boundary, without introducing or computing any forcing

465Henry Bandringa, “Immersed boundary methods”, Master Thesis in Applied Mathematics, Institute of
Mathematics and Computing Science, University of Groningen, The Netherlands, 2010.
 377

term. The methods that will be discussed here in detail are the direct-forcing method and extensions
of it, like the ghost-cell method, and the hybrid Cartesian/immersed boundary method. The
governing equations are most of the time discretized as follows. A second-order Adams-Bashforth
scheme is employed for the convective terms, while the diffusion terms are discretized using an
implicit Crank-Nicolson scheme. This eliminates the viscous stability constraint, which can be quite
severe in simulation of viscous flows [69].
13.9.4.1 General Idea
The incompressible flows governed by the non-dimensional Navier-Stokes equations, including the
body force term are given by

∂𝐮 1
+ 𝐮. ∇𝐮 + ∇p − ∆𝐮 = 𝐟
∂t Re
∇. 𝐮 = 0 in ΩΓ and u = uΓ on Γb
Eq. 13.54
The forcing term f functions as a velocity corrector for the grid points inside the IB. This term is
prescribed at each time step to establish the desired boundary moving velocity Vib. For a time-
stepping scheme, this force can be expressed as

un+1
i − uni
= RHSi + f i
∆t
Eq. 13.55
where the upper index of u indicates the time and the lower index the space. If fi must yield

Vin+1 − uni
un+1
i = Vin+1 → fi = − RHSi
∆t
Eq. 13.56
Thus the body force is defined like

1 1
fi = {𝐮. ∇𝐮 + ∇p − ∆𝐮 + (𝐕ib − 𝐮n ) near Γb
Re ∆t
0 elsewhere
Eq. 13.57
This approach only holds when the immersed boundary coincide with the grid. In general this is not
the case and the question arises: how to prescribe the boundary condition? This algorithm will be
explained using the idea of [Balaras et al. [1] below
1. First, compute u* in the discretized Navier-Stokes equations and omitting the forcing term fn+1.
The resulting u* will not satisfy the boundary conditions on the immersed boundary.
2. Then, compute fn+1 from Eq. 13.54. The value of the velocity Vib on the forcing points is
computed using an interpolation procedure. These forcing points can be placed outside and
inside the body, which is used in a ghost cell method, (Following).
3. Compute u* from the discretized Navier-Stokes equations with the forcing term. The resulting
velocity will satisfy the desired boundary conditions on the immersed boundary.
4. Compute the pressure using the Poisson equation.
5. Update the velocity and pressure.
6. Go to step 1.
13.9.4.2 Direct Forcing
 378

The (spectral) method of Mohd-Yusof [43]

uses a forcing term, which is determined by
the difference between the interpolated
velocities in the boundary points and the
desired (physical) boundary velocities. The
forcing term, generated in this manner, thus
directly compensates the errors between the
calculated velocities and the desired velocity
profile on the body surface. If the boundary is
stationary, a tangentially opposite direction
flow field to the external layer flow field is
specified in the internal layer inside the
immersed boundary surface. Where the
internal/external layer is defined in a layer of
grid points immediately inside/outside the Figure 13.31 Schematic Interpolation of Mohd-Yusof
immersed boundary. The force is thus ethod, [70].
determined by pairing the velocity at the
internal point to the velocity at the external point with a weighted linear interpolation, to enforce the
desired tangential velocity on the boundary, i.e. the method mirrors the velocity field across the
immersed boundary. An example of such a pair is illustrated in Figure 13.31 between Point 1 and
Point 2 [70].
Methods based on mirroring satisfy the velocity boundary condition with the accuracy of the
interpolation method. [Kim et al. [32] reported that in practice, the accuracy may be lower, because
of incompatibility with the continuity equation. Internal treatment is required for this method to
alleviate the problem of spurious oscillations near the boundary. Due to the reversed velocity field,
problems with the mass conservation arise in the boundary cell [40]. Methods based on mirroring
satisfy the velocity boundary condition with the accuracy of the interpolation method. Kim et al. [32]
reported that in practice, the accuracy may be lower, because of incompatibility with the continuity
equation. Internal treatment is required for this method to alleviate the problem of spurious
oscillations near the boundary. Due to the reversed velocity field, problems with the mass
conservation arise in the boundary cell [40].
13.9.4.3 Improvements
(Fadlun et al.)[11] further implemented the discrete-time forcing approach, as suggested by [Mohd-
Yusof [43] to a three-dimensional finite-difference method on a standard marker-and cell (MAC)
staggered grid and showed that the approach was more efficient than feedback forcing. There is a
difference between these approaches. In [Faldun et al.] the velocity at the first grid point external to
the body (external forcing) is obtained by linearly interpolating the velocity at the second grid point
(which is obtained by directly solving the Navier-Stokes equations) and the velocity at the body
surface, which conceptually corresponds to applying the momentum forcing inside the flow field. On
the other hand, momentum forcing is applied only on the body surface or inside the body in [Mohd-
Yusof's method [43] (internal forcing).
The interpolation direction (i.e., the direction towards the second grid point) taken by [Fadlun et al.]
is either the streamwise (x) or the transverse direction (y), but the choice of interpolation direction
is arbitrary, which can generate problems in complex configurations [1]. This approach thus does not
rely on the mirrored velocity field in the solid region, used by [Mohd-Yusof]. Therefore, the accuracy
of the interpolation remains the same. The linear interpolation formula used here is equivalent to an
implicit forcing formulation. Note, this method is sometimes called the standard reconstruction
method, i.e. a combination of linear interpolation with the standard mass conservation [27], see
Figure 13.35 (a).
 379

Since the velocity boundary condition is enforced with implicit forcing, there is no severe limit on the
time step. Another advantage is that the velocity components from the regions across the immersed
boundary are decoupled [27]. [Fadlun et al.] did not include a special constraint on the mass
conservation. So the original method for enforcing mass conservation results in coupling between
the solutions across the immersed boundary via discretized operators, which disobeys the pressure
decoupling constraint. Therefore, this method fails to predict the flow fields correctly. Another
drawback is that the velocity boundary condition is exactly satisfied in the momentum solution step,
but a finite error is introduced during the projection to the divergence-free velocity field. This is
because the velocity equal to the intermediate velocity is not enforced at the immersed boundary
[29].
This method works well for bodies that are aligned with the grid lines. For geometrically complex
immersed boundaries, however, the choice of the reconstruction direction may not be unique,
because often more than one grid line passing through a near-boundary node may intersect the
boundary [7].

(a) Fadlun et al.

(b) Balaras

Figure 13.32 Treatment of the Interface Cells; Forcing is Applied on the Filled Circles. [1].

[Balaras [1] proposed a better reconstruction scheme, based on the method of [Mohd-Yusof] and
[Fadlun et al.], which performs the reconstruction along the well-defined line normal to the body. By
taking a provisional explicit step (advancing both viscous and diffusive terms with an Adams-Bash
forth scheme for example) an inversion of a large sparse matrix can be avoided. The algorithm
eliminates the ambiguities associated with interpolation along grid lines, like in [Fadlun et al. [11],
see Figure 13.32.
However, this method is restricted to flows with immersed boundaries that are aligned with one
coordinate direction, for example two-dimensional or axisymmetric shapes [19]. Based on the ideas
of [Balaras, Gilmanov et al. [19] developed a new reconstruction scheme, which is applicable to
arbitrarily complex, three-dimensional immersed boundaries. The proposed methodology maintains
a sharp fluid/body interface by discretizing the body surface using an unstructured, triangular mesh.
The solution in the vicinity of boundary nodes is reconstructed via linear interpolation along the local
normal to the body. Unfortunately, this method is only applicable to stationary bodies of simple
(convex) shape [18]. An improved version of [Mohd-Yusof']s method has been made by [Zhang et al.
[70]. They improved the method of [Mohd-Yosuf] by implementing a bilinear
interpolation/extrapolation function to interpolate the direct force, which ensures more accurate
boundary forcing expressions.
 380

In Figure 13.33, a sketch of the interpolation/extrapolation is shown between immersed boundary

points and the grid points near the immersed boundary. This function is the same as the function
introduced by Saiki et al. [50]. For the method of
[Mohd-Yusof], only the tangential component of
the interpolated velocity was used for specifying
the force, while the method of [Zhang et al].
included both the tangential and the normal
velocity components. The advantages are that it
is easy for coding and the number of boundary
points can be increased independently of the
computational grid. Therefore, the accuracy in
the vicinity of the boundary can be enhanced
without using higher-order schemes, or
arranging complicated grids near the boundary.
A drawback is that the proposed [Saiki et al.]
interpolation/extrapolation method needs user-
Figure 13.33 Schematic interpolation of Zhang
defined parameters as input. [Choi et al. [4] made method, which is the same as Saiki et al. [50],
a finite-volume approach, which is based on [70].
[Fadlun et al]. They developed a more general
immersed boundary method that is valid at all Reynolds numbers and is suitable for implementation
on arbitrary grid topologies. Choi et al. introduced the concept of tangency correction by
decomposing the velocity into tangential and normal component along the outward normal direction
to the immersed surface. The tangential velocity component is expressed as a power-law function of
the wall normal distance.
However, the choice for using a power-law function seems somewhat arbitrary. The well-known
pressure Poisson equation approach is applied by [Sheu et al. [51], to eliminate the pressure gradient
terms from the momentum equations by performing a curl operator to the momentum equations.
They propose a quadratic interpolation scheme. First step in their method is to calculate the
intermediate velocity, then correct this intermediate velocity using the proposed quadratic
interpolation method. After that, they impose the forcing term F in the solid-fluid cells and finally the
real velocity will be computed.
13.9.4.4 Fulfilling Conservation Laws
The problem of most of the treated methods, is that they do not fulfil (some of) the conservation laws,
such as conservation of the wall condition. The treated immersed boundary methods violate the wall
condition in the discrete equation system during time advancement. This problem arises from the
inconsistency of the pressure with the velocity interpolated to represent the solid wall. Therefore,
[Ikeno et al. [25] developed a scheme to maintain the consistency between pressure and velocity for
the immersed boundary method to achieve precisely the desired wall condition, like the no-slip wall
condition or a non-zero velocity at the wall. They accomplished this by deriving a modified pressure
equation based on the interpolated pressure gradient. [Ikeno et al.] uses a two-point forcing strategy,
i.e. in the fluid as well as in the body region forcing is applied. Forcing in the body is employed,
because extrapolating the velocity at the grid points nearest to the wall increases the accuracy of the
velocity gradient. They demonstrated theoretically the conservation of the wall condition, mass,
momentum and energy.
Due to satisfaction of all the conservation laws, this method works fine for high Reynolds numbers
even in combination with turbulent flows. [Ikeno et al.] showed that, unlike their method, [Fadlun et
al. [11], [Ye et al. [69], [Tseng et al. [58] and [Kim et al. [32] do not satisfy the conservation properties
of the wall condition.
 381

Domenichini [8] found out that direct forcing schemes, treated so far, are not able to satisfy the
impenetrability condition on the fixed and moving wall. This fact appears to be strictly related to the
use of fractional step methods. He suggests that improvements can be obtained with the iterative
solution of the irrotational part of the flow, when spectral methods are used and local modification
of the discrete differential operators are difficult to be implemented.
13.9.4.5 Conservation of Mass
In some approaches, like [Fadlun et al., mass conservation at the immersed boundary is satisfied by
the velocity fields both in the fluid and solid regions, Figure 13.35 (a). The pressure is coupled by
this construction. In this case the non-physical velocity field in the solid becomes important because
it afects the pressure and velocity distribution through the velocity divergence across the immersed
boundary. This issue can become more serious in the reconstruction methods (e.g. [11, 18, 19]), since
treatment of the velocities at the first grid points into the solid region is undefined [27].

Figure 13.35 Different schemes for defining control volumes for mass conservation near the
immersed boundary. [27]

In general, as the interpolation scheme is defined without regard to the continuity equation, the
velocities obtained from the interpolation scheme will not satisfy conservation of mass for the cell.
This will cause the magnitude of the pressure in these mass conservation cells to slowly increase
without bound, in other words a pressure build-up. However, as this pressure does not appear in any
discrete momentum equation (as a result of the two point pressure gradient stencil and the selection
criteria for immersed boundary points), the solution will
not diverge. This pressure is therefore, completely
decoupled from all other discrete variables [44].
To alleviate the problem of incorrect pressure
distribution along the immersed boundary, Li et al. [37]
suggested to impose a zero normal gradient condition of
pressure. This approach can be interpreted as an
alternative way to enforce continuity since the Poisson
equation is derived from the continuity equation.
In Figure 13.34 node A has two fluid neighbor nodes B
and D. The zero normal pressure gradient condition on
point P can be obtained by setting pA = pP1 . Thus mass
conservation at point P will be satisfied since there will Figure 13.34 Implementation of the
be no flow across the boundary. zero gradient pressure condition on a
Muldoon et al. [44] among others, observed the non- collocated grid [37].
physical and unbounded behavior of the pressure at
 382

certain locations near the immersed boundary. They developed two new methods. The first is called
PVR (pseudo velocity reduced) for non-moving immersed boundaries. For moving boundaries, the
method is called CV (constrained velocity). In both methods, no distinction is made between the
inside or outside of an immersed boundary. [Muldoon et al.] managed to maintain mass conservation
in their methods, but due to using a pseudo-velocity the solution will not satisfy the prescribed
velocity at the immersed boundary, but satisfy the pseudo-velocity.
The objective of Kang et al. [27] is to assess the accuracy and efficiency of the immersed boundary
method to correctly predict the wall-pressure fluctuations in turbulent flows. This will be achieved
by introducing additional constraints. Firstly, a compatibility for the interpolated velocity boundary
condition related to mass conservation and secondly the formal decoupling of the pressure on this
surfaces. Their starting point is the method of [Fadlun et al].
This approach is referred to as the reconstruction of the interpolation method. The immersed
boundary-approximated domain method (IB-ADM) was developed, to satisfy the pressure
decoupling constraint. This decoupling process allows discontinuous solutions across the interface
and is similar to the jump condition used in the immersed interface method [36], and the ghost fluid
method [12]. A schematic sketch of the IB-ADM is shown in Figure 13.35 (c). In the IB-ADM, the
velocity equal to the intermediate velocity is enforced at the approximated boundary Гa instead of
ГIB. It is very important to satisfy the pressure decoupling constraint, because this assumptions leads
to
∂(pk − pk−1 )
=0 at Γ𝑎
∂n
Eq. 13.58
i.e. (strict) mass conservation and no accumulation of the pressure error. When Гa is very close to ГIB,
the original condition is recovered. The linear interpolation is included with the effect of the pressure
gradient term (revised linear interpolation method of [28]), resulting in a slightly reduced velocity
error. Kang et al. showed that by satisfying the pressure decoupling constraint, the IB-ADM is
successfully in handling very thin solid objects. For a complete discussion, please consult the
[Bandringa]466.
13.9.5 Cut-Cell Finite-Volume Approach
The basic formulation of the cut-cell method will first be described. Then some improvements for
this approach will be treated. And finally, the problem of using a staggered grid instead of a collocated
grid for the cut-cell method will be discussed.
13.9.5.1 Basic Formulation
[Ye et al. 69] proposed a different approach for simulating convection-dominated flows on a
collocated (non-staggered) grid called a cut-cell method (in the past also named Cartesian grid
method), which does not use the concept of momentum forcing. They used a central difference
interpolation scheme near the immersed boundary that gives second-order spatial accuracy. In this
method cells in the Cartesian grid that are cut by the IB are identified, and the intersection of the
boundary with the sides of these cut-cells is determined. Next, cells cut by the IB, whose cell center
lies in the fluid, are reshaped by discarding the portion of these cells that lies in the solid. Pieces of
cut-cells whose centers lie in the solid are usually absorbed by neighboring cells to prevent stability
problems. This results in the formation of control volumes, which are trapezoidal in shape, as shown
in Figure 13.36.
Details of this reshaping procedure can be found in [Uday kumar et al. [61, 64]. The approach
proposed by Ye et al. is to express a given ow variable Φ in terms of a two-dimensional polynomial

466Henry Bandringa, “Immersed boundary methods”, Master Thesis in Applied Mathematics, Institute of
Mathematics and Computing Science, University of Groningen, The Netherlands, 2010.
 383

interpolating function in an appropriate region and evaluate the fluxes f based on this interpolating
function. For instance, in order to approximate the flux on the southwest face, fsw, Φ (in the shaded
trapezoidal region shown in Figure 13.37 (b)) is expressed in terms of a function that is linear in x
and quadratic in y
ϕ = c1 xy 2 + c2 y 2 + c3 xy + c4 y + c5 x + c6
Eq. 13.59
where c1 to c6 are six unknown
coefficients. Eq. 13.59
represents the most compact
function that allows at least a
second-order accurate
evaluation of Φ at the sw
location.
The presence of immersed
boundaries alters the
conditioning of the linear
operators and this can slow
down the iterative solution of
the these equations.
Therefore a preconditioned
Conjugate Gradient method is
used for accelerating the
convergence. Since the inside
of the immersed boundary is
also gridded, this method also
has the capability to solve a
different set of equations
inside the immersed
boundary. For instance,
equations of heat conduction
could be solved inside the
body. While Ye et al. show that
their interpolation scheme is Figure 13.36 Schematic of computational domain with immersed
itself second-order accurate, boundaries. [69].
in their test for the accuracy of
the overall scheme they use simulations of Wannier flow in which there is no contribution from the
advection (convection) terms [33].
A disadvantage of cell-merging is that it generally entails a considerable increase in complexity as
fluxes between diagonally adjacent cells must also be calculated and the computational molecule for
merged boundary cells becomes different to that used for the standard cells. There are also significant
problems associated with the formulation of a systematic merging algorithm in three dimensions
[33], because of complex polyhedral cells.
The discretization fulfils the conservation laws due to the finite-volume approach. [Mittal et al. [41]
stated that successful implementation of the cut-cell method (with cell merging) to three-dimension
geometries has not yet been accomplished, due to the above described difficulties. Kirkpatrick et al.
[33] noted that the matrix condition number increases significantly when the size of the reshaped
control volume is very small.
[Uday kumar et al. [63] extended the method of Ye et al. in order to allow for motion of the immersed
boundary. The moving boundary is represented as a sharp interface using an Eulerian-Lagrangian
 384

Figure 13.37 Schematic of interpolation for cell face values and derivative at boundary cell; Points
in the six-point stencil for fsw [69].

approach and the interface tracking procedure from [Uday kumaret al. [64]. The curved immersed
boundary is represented using marker particles which are connected by piecewise quadratic curves
parametrized with respect to the arclength. They use an implicit treatment, i.e. the boundary and
flow are advanced in time simultaneously in a fully coupled manner instead of a sequentially manner
which is the case for an explicit treatment. The primary advantage of this approach instead of the
explicit treatment, is that it removes any stability constraints associated with the boundary motion.
A multigrid method is used for accelerating the convergence of the pressure Poisson equation. The
finite-volume discretization in a given cell can be written in the general form

∑ a k ϕk = b
k=1
Eq. 13.60
where ak(for k = 1, , , ,M) denote the coefficients accompanying the nodal values Φk within a stencil
consisting of M neighbors and b is the source term that contains the explicit terms as well as the
terms involving boundary conditions. For cells away from the interface, M = 5 corresponds with the
central-difference spatial discretization. [Uday kumar et al. [62] demonstrated the versatility of the
numerical method [63] by applying it to some challenging physical problems in solidification and
fluid-structure interactions.
13.9.5.2 Improvements
[Kirkpatrick et al. [33] developed a method for representing curved boundaries for the solution of
the Navier-Stokes equations on a non-uniform, staggered, three-dimensional Cartesian grid. Solid
boundaries are defined as quadratic surfaces in a Cartesian coordinate system. Kirkpatrick et al.
circumvent the problems related to cell-merging by introducing a novel cell-linking method. This
method overcomes problems associated with the creation of small cells while avoiding the
complexities involved with other merging approaches. Rather than merging two cells to form a single
 385

cell, the two cells are linked as a

"master/slave" pair in which the two
nodes are coincident while each cell
remains a distinct entity. A master
cell and slave cell is shown in Figure
13.38467. As a result, the fluxes, wall
shear stress, volumetric and surface
information are calculated in exactly
the same way for the master and
slave cells as they are for the
standard boundary cells. There are
some problems when using a
staggered grid in combination with a
cut-cell approach, as pointed out by
[Kirkpatrick et al.]. Using cell-linking,
is one of the solutions for some of
these problems.
However, the application of Figure 13.38 A master and slave cell are shown for the u
Kirkpatrick et al. method to moving component of velocity.
objects is not so easy because of the
presence of extremely small-size
cells [56]. Also, the way the problem of linking a master and a slave velocity node has been handled
is not elegant. The slave node is placed a small distance from the master node. Then the diffusion flux
between the two nodes automatically becomes extremely high and forces the two velocities to take
the same value. This situation is not ideal for the case of high Reynolds number flows.
[Dr”oge et al. [9] introduced a different kind of cut-cell discretization. The cut-cells are not merged
or linked as in the previous cases, there are just left alone. Their cut-cell discretization preserves the
symmetry of convection and diffusion. Convection is discretized by a skew-symmetric operator and
diffusion is approximated by a symmetric, positive finite coefficient matrix. This formulation for
discretization conserves the kinetic energy, if the dissipation is turned off, and is stable on any grid.
This scheme is second-order accurate, when the dissipation is turned off. The cut-cell approach of
Chung [5] simulates two-dimensional unsteady viscous incompressible with (moving) rigid bodies
of arbitrary shape on a collocated Cartesian grid.
The pressure-free projection method is used as the incompressible flow solver. This solves the
coupling problem between the divergence-free constraint and the pressure gradient in the
momentum equation. In the work of [Ye et al. [69], the value of variables or fluxes at the center of the
merged cell face or the merged cutting surface of the merged cell is directly interpolated from the
neighboring positive centers or centers of the merged boundary segment. However, in Chung's
method, the merged cell face is regarded as a combination of a full cell face and a cut-cell face.
Likewise, the merged cutting surface is a combination of two boundary segments. This approach has
particular benefit in three-dimensional problems, due to easier and more systematic making of both
the location of the cut face center and the construction of the interpolation stencil. Further, the
merging of boundary segments in Ye et al. will reduce the shape resolution of the immersed
boundary, more than the method of Chung. See for example Figure 13.39 for such a situation. The
observed order of accuracy in the spatial discretization is super-linear, though not quadratic.

467A master and slave cell are shown for the u component of velocity. The slave cell velocity has only one
pressure node associated with it. It is moved to the same position as the master cell node [33].
 386

Figure 13.39 Difference between Chung and Ye [5].

13.9.5.3 LS-STAG
[Cheny et al. [3] developed a new immersed boundary method, which is based on the MAC method
for staggered Cartesian grids and where the irregular boundary is sharply represented by its level-
set function (signed distance function), hence it is called LS-STAG method. The flow variables are
computed in the cut-cells and not interpolated like in e.g. [11]. Furthermore, the fluxes in Cartesian
and cut-cells are discretized in a consistent and unified fashion, so ad hoc treatments for the cut-cells
are no longer needed. The LS-STAG is based on the symmetry preserving finite-volume method by
Verstappen et al. [66], which has the ability to preserve on non-uniform staggered Cartesian grids
the conservation of total mass, momentum and kinetic energy of the original MAC method [22].
The discretization in the LS-STAG method preserves the five-point Cartesian structure of the stencil,
resulting in a highly computationally efficient method. The LS-STAG method can also handle moving
immersed boundaries, but the first attempt of [Cheny et al.] is a simple procedure, which does not
guarantee the discrete conservation of momentum or kinetic energy. Therefore, the developers will
in the future improve their procedure by using in the freshly cleared cells (fluid cells which were
inside the solid at the previous time step) a fully implicit semi-Lagrangian time-stepping. By
performing an accuracy test, they discovered that their method is super-linear accurate in space.
[Cheny et al.] stated that by using a level-set function, the immersed boundary can be sharply
represented. However, [Berthelsen et al. [2] are of the opinion that the level-set formulation fails to
describe sharp corners and infinitely thin plates.
13.9.5.4 General Considerations
An advantage of a cut-cell method is that is possible to accurately impose the boundary conditions
on the body. Another benefit is that the cut-cell method is based on finite volume, so strict
conservation of mass and momentum is guaranteed even in the vicinity of the immersed boundary.
Furthermore, the application of adaptively refined grids does not complicate the implementation of
this technique. One of the drawbacks of this approach is the following. Implementing the boundary
conditions in irregular cells requires a large number of "special treatments", which could result in
complex coding logistics. When using cut-cells, these cell should not become too small. Otherwise this
could not only lead to stability problems, but also lead to slow convergence of the Poisson solver.
 387

Finally, an iterative solution procedure may be required due to the irregular stencil near the
immersed boundary [4].
13.9.6 Case Study - Flow Past a Sphere468
[Yun et al. (2003)] conducted large-eddy simulations (LESs) of flow past a sphere at the Reynolds
numbers of 3700 and 104, based on the freestream velocity and sphere diameter (d). The IB method
was implemented in a cylindrical coordinate system, and momentum forcing and mass sources/sinks
were introduced inside the IB to satisfy the no-slip condition on the sphere surface and continuity
for the cell containing the IB, respectively (Kim et al. 2001). A hybrid spatial discretization scheme
was used wherein a third-order compact upwind scheme was employed before separation to avoid
dispersion errors and the second-order central difference scheme was applied to the wake region
together with a dynamic sub grid scale model. The computational domain was −15 ≤ x/d ≤ 15, 0 ≤
r/d ≤ 15 and 0 ≤ θ < 2π, where x , r , and θ are the streamwise, radial, and azimuthal directions,
respectively. The numbers of grid points used were 577(x)×141(r )×40 (θ).
Figure 13.40 shows
computed vortical
structures visualized
using particle tracers
for Re = 104, together
with an experimental
flow visualization at
Re = 1.5×104 (Werl´e
1980). Vortex rings
are observed forming
immediately behind
the sphere and the
computed wake
structure for Re = Figure 13.40 Flow structures in the near wake behind a sphere: (a) Re = 104
104 (Figure 13.40 (IB simulation); (b) Re = 1.5 × 104 (experiment).
a) is very similar to
that observed in the experiment (Figure 13.40 b). Simulation results are summarized in Error!
Reference source not found. where the computed time-averaged drag coefficient (¯Cd ), base
pressure coefficient (¯Cpb ), and Strouhal number (St) are presented together with previous
experimental and numerical data. In general, the results from the IB simulations are in good
agreement with these other studies, thereby further validating the fidelity of the IB simulations.
13.9.7 References
[1] E. Balaras. Modeling complex boundaries using an external force field on fixed cartesian grids in
large eddy simulations. Computers & Fluids, 33(3):375{404, 2004. doi:10.1016/S0045-
7930(03)00058-6.
[2] P.A. Berthelsen and O.M. Faltinsen. A local directional ghost cell approach for incompressible
viscous flow problems with irregular boundaries. Journal of Computational Physics,
227(9):4354{4397, 2008. doi:10.1016/j.jcp.2007.12.022.
[3] Y. Cheny and O. Botella. The LS-STAG method: A new immersed boundary/level-set method for the
computation of incompressible viscous flows in complex moving geometries with good conservation
properties. Journal of Computational Physics, 229(4):1043{1076, 2010.
[4] J.I. Choi, R.C. Oberoi, J.R. Edwards, and J.A. Rosati. An immersed boundary method for complex
incompressible flows. Journal of Computational Physics, 224(2):757{784, 2007.

468 Rajat Mittal, Gianluca Iaccarino, “Immersed Boundary Methods”, Ann. Rev. Fluid Mech. 2005. 37:239–61
 388

[5] M.H. Chung. Cartesian cut cell approach for simulating incompressible ows with rigid bodies of
arbitrary shape. Computers & Fluids, 35(6):607{623, 2006. doi:10.1016/j.compfluid.2005.04.005.
[6] A. Cristallo and R. Verzicco. Combined Immersed Boundary/Large-Eddy-Simulations of
Incompressible Three Dimensional Complex Flows. Flow Turbulence Combust, 77(1-4):3{26, 2006.
[7] J. Deng, X.M. Shao, and A.L. Ren. A new modification of the immersed-boundary method for
simulating flows with complex moving boundaries. Int. J. Num. Meth. Fluids, 52(11):1195{1213, 2006.
[8] F. Domenichini. On the consistency of the direct forcing method in the fractional step solution of the
Navier-Stokes equations. Journal of Computational Physics, 227(12):6372{6384, 2008.
[9] M. Dr•oge and R. Verstappen. A new symmetry-preserving Cartesian-grid method for computing
flow arbitrarily shaped objects. Int. J. Num. Meth. Fluids, 47(8-9):979{985, 2005. doi:10.1002/ fld.924.
[10] R. Eymard, T. Gallouet, and R. Herbin, Editors: P.G. Ciarlet and J.L. Lions. The finite volume
method, Handbook of Numerical Analysis, volume VII. 2000.
[11] E.A. Fadlun, R. Verzicco, P. Orlandi, and J. Mohd-Yusof. Combined Immersed-Boundary Finite-
Difference Methods for Three-Dimensional Complex Flow Simulations. Journal of Computational
Physics, 161(1):35{60, 2000. doi:10.1006/jcph.2000.6484.
[12] R.P. Fedkiw, T. Aslam, B. Merriman, and S. Osher. A Non-oscillatory Eulerian Approach to
Interfaces in Multi material Flows (the Ghost Fluid Method). Journal of Computational Physics,
152(2):457{492, 1999. doi:10.1006/jcph.1999.6236.
[13] F.N. Felten and T.S. Lund. Kinetic energy conservation issues associated with the collocated mesh
scheme for incompressible flow. Journal of Computational Physics, 215(2):465{484, 2006.
[14] R. Franke. Scattered Data Interpolation: Tests of Some Methods. Math. Comp., 1982.
[15] T. Gao, Y.H. Tseng, and X.Y Lu. An improved hybrid Cartesian/immersed boundary method for
fluid-solid flows. Int. J. Num. Meth. Fluids, 55(12):1189{1211, 2007. doi:10.1002/fld.1522.
[16] R. Ghias, R. Mittal, and H. Dong. A sharp interface immersed boundary method for compressible
viscous flows. Journal of Computational Physics, 225(1):528{553, 2007. doi:10.1016/j.jcp.2006.
[17] F. Gibou, R.P. Fedkiw, L.T. Cheng, and M. Kang. A Second-Order-Accurate Symmetric Discretization
of the Poisson Equation on Irregular Domains. Journal of Computational Physics, 176(1), 2002.
[18] A. Gilmanov and F. Sotiropoulos. A hybrid Cartesian/immersed boundary method for simulating
flows with 3d, geometrically complex, moving bodies. Journal of Computational Physics,
207(2):457{492, 2005. doi:10.1016/j.jcp.2005.01.020.
[19] A. Gilmanov, F. Sotiropoulos, and E. Balaras. A general reconstruction algorithm for simulating
flows with complex 3d immersed boundaries on Cartesian grids. Journal of Computational Physics,
191(2):660{669, 2003. doi:10.1016/S0021-9991(03)00321-8.
[20] R. Glowinski, T.W. Pan, and J. Priaux. Distributed Lagrange multiplier methods for incompressible
viscous flow around moving rigid bodies. Com. Methods Appl. Mech. Eng, 151(1):181-194, 1998.
[21] D. Goldstein, R. Handler, and L. Sirovich. Modeling a No-Slip Flow Boundary with an External
Force Field. Journal of Computational Physics, 105(2):354{366, 1993. doi:10.1006/jcph.1993.1081.
[22] F.H. Harlow and J.E. Welch. Numerical Calculation of Time-Dependent Viscous Incompressible
Flow of Fluid with Free Surfaces. Phys. Fluid, 8(12):2182{2189, 1965. doi:10.1063/1.1761178.
[23] W.X Huang and H.J. Sung. Improvement of mass source/sink for an immersed boundary method.
Int. J. Num. Meth. Fluids, 53(11):1659{1671, 2007. doi:10.1002/fld.1367.
[24] G. Iaccarino and R. Verzicco. Immersed boundary technique for turbulent flow simulations. Appl.
Mech. Rev., 56(3):331{347, 2003. doi:10.1115/1.1563627.
[25] T. Ikeno and T. Kajishima. Finite-difference immersed boundary method consistent with wall
conditions for incompressible turbulent flow simulations. Journal of Computational Physics,
226(2):1485{1508, 2007. doi:10.1016/j.jcp.2007.05.028.
[26] G. Kalitzin and G. Iaccarino. Toward immersed boundary simulation of high Reynolds number
flows. Center for Turbulence Research Annual Research Briefs, pages 369{378, 2003.
[27] S. Kang, G. Iaccarino, F. Ham, and P. Moin. Prediction of wall-pressure fluctuation in turbulent
flows with an immersed boundary method. Journal of Computational Physics, 2009.
 389

[28] S. Kang, G. Iaccarino, and P. Moin. Accurate and efficient immersed-boundary interpolations for
viscous flows. Center for Turbulence Research Annual Research Briefs, pages 31{43, 2004.
[29] S. Kang, G. Iaccarino, and P. Moin. Accurate Immersed-Boundary Reconstructions for Viscous Flow
Simulations. AIAA Journal, 47(7):1750{1760, 2009. doi:10.2514/1.42187.
[30] K. Khadra, P. Angot, S. Parneix, and J.P. Caltagirone. Fictitious domain approach for numerical
modelling of Navier-Stokes equations. Int. J. Num. Meth. Fluids, 34(8):651{684, 2000.
[31] D. Kim and H. Choi. Immersed boundary method for ow around an arbitrarily moving body. Journal
of Computational Physics, 212(2):662{680, 2006. doi:10.1016/j.jcp.2005.07.010.
[32] J. Kim, D. Kim, and H. Choi. An Immersed-Boundary Finite-Volume Method for Simulations of Flow
in Complex Geometries. Journal of Computational Physics, 171(1):132{150, 2001.
[33] M.P. Kirkpatrick, S.W. Arm_eld, and J.H. Kent. A representation of curved boundaries for the
solution of the Navier-Stokes equations on a staggered three-dimensional Cartesian grid. Journal of
Computational Physics, 184(1):1{36, 2003. doi:10.1016/S0021-9991(02)00013-X.
[34] MC. Lai and C.S. Peskin. An Immersed Boundary Method with Formal Second-Order Accuracy and
Reduced Numerical Viscosity. Journal of Computational Physics, 160(2):705{719, 2000.
[35] D.V. Le, B.C. Khoo, and K.M. Lim. An implicit-forcing immersed boundary method for simulating
viscous flows in irregular domains. Comp. Methods Appl. Mech. Eng. 197(25-28):2119{2130, 2008.
[36] L. Lee and R. Leveque. An immersed interface method for incompressible Navier-Stokes equations.
SIAM J. Sci. Com., 25(3):832{856, 2003. doi:10.1137/S1064827502414060.
[37] C.W. Li and L.L. Wang. An immersed boundary finite difference method for LES of ow around bluff
shapes. Int. J. Num. Meth. Fluids, 46(1):85{107, 2004. doi:10.1002/fld.749.
[38] C.C. Liao, Y.W. Chang, C.A. Lin, and J.M. McDonough. Simulating flows with moving rigid boundary
using immersed-boundary method. Computers & Fluids, 39(1):152{167, 2010.
[39] S. Majumdar, G. Iaccarino, and P. Durbin. Rans solvers with adaptive structured boundary
nonconforming grids. Center for Turbulence Research Annual Research Briefs, pages 353{366, 2001.
[40] A. Mark and B.G.M van Wachem. Derivation and validation of a novel implicit second-order
accurate immersed boundary method. Journal of Computational Physics, 227(13):6660-6680, 2008.
[41] R. Mittal, H. Dong, M. Bozkurttas, F.M. Najjar, A. Vargas, and A. von Loebbecke. A versatile sharp
interface immersed boundary method for incompressible flows with complex boundaries. Journal of
Computational Physics, 227(10):4825{4852, 2008. doi:10.1016/j.jcp.2008.01.028.
[42] R. Mittal and G. Iaccarino. Immersed boundary methods. Ann. Rev. Fluid Mech., 37:239{261, 2005.
[43] J. Mohd-Yusof. Combined immersed-boundary/B-spline methods for simulations of flow in complex
geometries. Center for Turbulence Research Annual Research Briefs, 161(1):317{327, 1997.
[44] F. Muldoon and S. Acharya. A divergence-free interpolation scheme for the immersed boundary
method. Int. J. Num. Meth. Fluids, 56(10):1845-1884, 2008. doi:10.1002/fld.1565.
[45] P. de Palma, M.D. de Tullio, G. Pascazio, and M. Napolitano. An immersed-boundary method for
compressible viscous flows. Computers & Fluids, 35(7):693{702, 2006.
[46] D. Pan and T.T. Shen. Computation of incompressible flows with immersed bodies by a simple ghost
cell method. Int. J. Num. Meth. Fluids, 60(12):1378-1401, 2008. doi:10.1002/fld.1942.
[47] N. Peller, A.L. Duc, F. Tremblay, and M. Manhart. High-order stable interpolations for immersed
boundary methods. Int. J. Num. Meth. Fluids, 52(11):1175{1193, 2006. doi:10.1002/fld.1227.
[48] C.S. Peskin. Flow patterns around heart valves: A numerical method. Journal of Computational
Physics, 10(2):252-271, 1972. doi:10.1016/0021-9991(72)90065-4.
[49] C.S. Peskin. The fluid dynamics of heart valves: experimental, theoretical, and computation
methods. Ann. Rev. Fluid Mech., 14:235{259, 1982. doi:10.1146/annurev.fl.14.010182.001315.
[50] E.M. Saiki and S. Biringen. Numerical Simulation of a Cylinder in Uniform Flow: Application of a
Virtual Boundary Method. Journal of Computational Physics, 123(2):450{465, 1996.
[51] T.W.H. Sheu, H.F. Ting, and R.K. Lin. An immersed boundary method for the incompressible Navier-
Stokes equations in complex geometry. Int. J. Num. Meth. Fluids, 56(7):877-898, 2008.
 390

[52] A.F. Shinn, M.A. Goodwin, and S.P. Vanka. Immersed boundary computations of shear- and
buoyancy driven flows in complex enclosures. International Journal of Heat and Mass Transfer, 52(17-
18):4082-4089, 2009. doi:10.1016/j.ijheatmasstransfer.2009.03.044.
[53] A.L.F.L.E. Silva, A. Silveira-Neto, and J.J.R. Damasceno. Numerical simulation of two-dimensional
flows over a circular cylinder using the immersed boundary method. Journal of Computational Physics,
189(2):351{370, 2003. doi:10.1016/S0021-9991(03)00214-6.
[54] S.W. Su, M.C. Lai, and C.A. Lin. An immersed boundary technique for simulating complex flows with
rigid boundary. Computers & Fluids, 36(2):313{324, 2007. doi:10.1016/j.compfluid.2005.09.004.
[55] K. Taira and T. Colonius. The immersed boundary method: A projection approach. Journal of
Computational Physics, 225(2):2118{2137, 2007. doi:10.1016/j.jcp.2007.03.005.
[56] K. Takizawa, K. Yabe, Y. Tsugawa, T.E. Tezduyar, and H. Mizoe. Computation of free-surface flows
and fluid-object interactions with the CIP method based on adaptive meshless soroban grids. Com. Mech,
40(1):167{183, 2007. doi:10.1007/s00466-006-0093-2.
[57] F. Tessicini, G. Iaccarino, M. Fatica, M. Wang, and R. Verzicco. Wall modeling for large-eddy
simulation using an immersed boundary method. Center for Turbulence Research Annual Research
Briefs, pages 181-187, 2002.
[58] Y.H. Tseng and J.H. Ferziger. A ghost-cell immersed boundary method for flow in complex
geometry. Journal of Computational Physics, 192(2):593-623, 2003. doi:10.1016/j.jcp.2003.07.024.
[59] M.D. de Tullio, P. de Palma, G. Iaccarino, G. Pascazio, and M. Napolitano. An immersed boundary
method for compressible flows using local grid refinement. Journal of Computational Physics,
225(2):2098-2117, 2007. doi:10.1016/j.jcp.2007.03.008.
[60] M. Tyagi and S. Acharya. Large eddy simulation of turbulent flows in complex and moving rigid
geometries using the immersed boundary method. Int. J. Num. Meth. Fluids, 48(7):691{772, 2005.
[61] H.S. Udaykumar, H.C. Kan, W. Shyy, and R. Tran-Son-Tay. Multiphase dynamics in arbitrary
geometries on fixed Cartesian grids. Journal of Computational Physics, 137(2):366{405, 1997.
[62] H.S. Udaykumar, R. Mittal, and P. Rampunggoon. Interface tracking finite volume method for
complex solid-fluid interactions on fixed meshes. Com. Num. Meth. Eng. 18(2):89{97, 2002.
[63] H.S. Udaykumar, R. Mittal, P. Rampunggoon, and A. Khanna. A Sharp Interface Cartesian Grid
Method for Simulating Flows with Complex Moving Boundaries. Journal of Computational Physics,
174(1):345-380, 2001. doi:10.1006/jcph.2001.6916.
[64] H.S. Udaykumar, R. Mittal, and W. Shyy. Computation of solid-liquid phase fronts in the sharp
interface limit on fixed grids. Journal of Computational Physics, 153(2):535{574, 1999.
[65] A.E.P. Veldman and A. Velick_a. Stromingsleer. Rijksuniversiteit Groningen, Groningen, 2010.
[66] R.W.C.P. Verstappen and A.E.P. Veldman. Symmetry-preserving discretization of turbulent flow.
Journal of Computational Physics, 187(1):343-368, 2003. doi:10.1016/S0021-9991(03)00126-8.
[67] R. Verzicco, G. Iaccarino, M. Fatica, and P. Orlandi. Flow in an impeller stirred tank using an
immersed boundary method. Center for Turbulence Research Annual Research Briefs,2000.
[68] G. Yang, D.M. Causon, and D.M. Ingram. Calculation of compressible flows about complex moving
geometries using a three-dimensional Cartesian cut cell method. Int. J. Num. Meth. Fluids,
33(8):1121{1151, 2000. doi:10.1002/1097-0363(20000830)33:8<1121::AID-FLD45>3.0.CO;2-H.
[69] T. Ye, R. Mittal, H.S. Udaykumar, and W. Shyy. An Accurate Cartesian Grid Method for Viscous
Incompressible Flows with Complex Immersed Boundaries. Journal of Computational Physics,
156(2):209{240, 1999. doi:10.1006/jcph.1999.6356.
[70] N. Zhang and Z.C. Zheng. An improved direct-forcing immersed-boundary method for finite
difference applications. Journal of Computational Physics, 221(1):250-268, 2007.
391
 392

14 Components of CFD Simulation

Having positioned CFD, and its importance, in the global technological world of virtual prototyping,
we should now look at the main components of a CFD system469. We wish to answer the following
question: What are the steps you have to define in order to develop, or to apply, a CFD simulation?
We make no difference at this stage between these two options, as it is similarly essential for the User
of a CFD code to understand clearly the different options available and to be able to exercise a critical
judgment on all the steps involved. At this juncture, it is imperative to define difference between
Modeling and Simulation, since both terms used interchangely in CFD.

Figure 14.1 A typical CFD Simulation

469 Introduction: An Initial Guide to CFD and to this Volume; page 1, 2007.
 393

14.1 Simulation vs. Modeling

A model is a physical, mathematical, or otherwise logical representation of a system, or process. The
Simulation is a method of implementing a model over time470. A typical CFD component of simulation
is illustrated in Figure 14.1, and steps are defined below.
14.1.1 Part I
It selects the mathematical model, defining the level of the approximation to reality that will be
simulated. This could be include what kind of equation would be sufficient for simulation, Euler, N-S,
Heat Transfer, etc. For most commercially available CFD codes this step is automatically defined
within the code and the no need for end-users intervention. Then the discretization of G.E. which
includes examples of CD, UD, QUICK, MARS etc.
14.1.2 Part II
It covers the discretization phase, which has three main components, namely Spatial Discretization,
Temporal Discretization, and Field Discretization as defined by the grid generation which discussed
in details and quite tedious. Spatial and Temporal discretization of the governing equation in space
and time respectively.
14.1.3 Part III
The numerical scheme must be analyzed and its properties of stability and accuracy have to be
established. The immediate following step II is to decide upon the type of the solver needed (density
based, pressure based, steady, unsteady, etc.) to address the specific problem. Later, it is required to
select appropriate physical model (turbulence, multiphase, species transport, combustion or
aerodynamic) based on the problem defined. The selection of solver and physical model should be
followed by defining the material properties, operating conditions and boundary conditions. One
should also establish convergence monitors in order to avoid extra computing time. Once the solver
and physics are set up, the flow field should be initialized. Numerical stability and consistency criteria
would be stablished for different flow solver (Implicit vs Explicit).
14.1.4 Part IV
The solution of the numerical scheme has to be obtained, by selecting the most appropriate time
integration methods, as well as the subsequent resolution method of the algebraic systems, including
convergence acceleration techniques. Monitoring solutions as well as revising the solution control
during the run also important.
14.1.5 Part V
Graphic post-processing of the numerical data to understand and interpret the physical properties of
the obtained simulation results. This is made possible by the existence of powerful visualization
software. Whenever possible, it must verify the result. Repeat the process to obtain the mesh and
other flow sensitivities. Alternatively, another way is to be envisioned as following (see Figure
14.2). They are simulation process which can be further streamlined. The solver and solver setting,
as well as, pre-processing, including, mesh generation and CAD date (solid molding), and post
processing are represented. It is well known in CFD circuits that about %70 of engineer times is
spend in pre-processing (CAD data and Mesh Generation). So that it is a big field by itself and requires
keen specialties.

470 Grand Challenges on the Theory of Modeling and Simulation

 394

Material
Physical Properties
Models: Boundary
Turbulence Conditions
Combustion Solver
Radiation Settings
Mutiphase
Moving Mesh
Transport Equations:
Mass - Momentum
Energy
Equation of state
Supporting physical Solid
Post models Modling
Processing

Material Properties
Boundary Conditions Mesh Generation
Initila Conditions

Figure 14.2 Streamlined CFD Simulation Process

14.2 Simulation Components as Envisioned by NASA

Finally, NPARC (alliance for CFD verification and validation) of NASA describes the CFD simulation
process as below. Bear in mind that this is a detailed and lengthy endower and not all the components
are needed all the time, but the user should be confident with the concepts at least.

1. Formulate the Flow Problem

2. Model the Geometry and Flow Domain
3. Establish the Boundary and Initial Conditions
4. Generate the Grid
5. Establish the Simulation Strategy
6. Establish the Input Parameters and Files
7. Perform the Simulation
8. Monitor the Simulation for Completion
9. Post-process the Simulation to get the Results
10. Make Comparisons of the Results
11. Repeat the Process to Examine Sensitivities

Further details include:

14.2.1 Formulate the Flow Problem
The first step of the analysis process is to formulate the flow problem by seeking answers to the
following questions:
➢ What is the objective of the analysis?
➢ What is the easiest way to obtain those objective?
 395

➢ What geometry should be included?

➢ What are the freestream and/or operating conditions?
➢ What dimensionality of the spatial model is required? (1D, quasi-1D, 2D, axisymmetric, 3D)
what should the flow domain look like? What temporal modeling is appropriate? (steady or
unsteady)
➢ What is the nature of the viscous flow? (inviscid, laminar, turbulent)
➢ How should the gas be modeled?
14.2.2 Model the Geometry and Flow Domain
The body about which flow is to be analyzed requires modeling. This generally involves modeling the
geometry with a CAD software package. Approximations of the geometry and simplifications may be
required to allow an analysis with reasonable effort. Concurrently, decisions are made as to the
extent of the finite flow domain in which the flow is to be simulated. Portions of the boundary of the
flow domain coincide with the surfaces of the body geometry. Other surfaces are free boundaries
over which flow enters or leaves. The geometry and flow domain are modeled in such a manner as to
provide input for the grid generation. Thus, the modeling often takes into account the structure and
topology of the grid generation.
14.2.3 Establish the Boundary and Initial Conditions
Since a finite flow domain is specified, physical conditions are required on the boundaries of the flow
domain. The simulation generally starts from an initial solution and uses an iterative method to reach
a final flow field solution.
14.2.4 Generate the Grid
The flow domain is discretized into a grid. The grid generation involves defining the structure and
topology and then generating a grid on that topology. Currently all cases involve multi-block,
structured grids; however, the grid blocks may be abutting, contiguous, non-contiguous, and
overlapping. The grid should exhibit some minimal grid quality as defined by measures of
orthogonality (especially at the boundaries), relative grid spacing (15% to 20% stretching is
considered a maximum value), grid skewness, etc... Further the maximum spacing’s should be
consistent with the desired resolution of important features. The resolution of boundary layers
requires the grid to be clustered in the direction normal to the surface with the spacing of the first
grid point off the wall to be well within the laminar sublayer of the boundary layer. For turbulent
flows, the first point off the wall should exhibit a y+ value of less than or equal to 1.0.
14.2.5 Establish the Simulation Strategy
The strategy for performing the simulation involves determining such things as the use of space-
marching or time-marching, the choice of turbulence or chemistry model, and the choice of
algorithms.
14.2.6 Establish the Input Parameters and Files
A CFD codes generally requires that an input data file be created listing the values of the input
parameters consisted with the desired strategy. Further, a grid file containing the grid and boundary
condition information is generally required. The files for the grid and initial flow solution need to be
generated.
14.2.7 Perform the Simulation
The simulation is performed with various possible with options for interactive or batch processing
and distributed processing.
 396

14.2.8 Monitor the Simulation for Completion

As the simulation proceeds, the solution is monitored to determine if a "converged" solution has been
obtained, which iterative convergence is. Further discussion can be found on the page entitled
Examining Iterative Convergence.
14.2.9 Post-Process the Simulation to get the Results
Post-Processing involves extracting the desired flow properties (thrust, lift, drag, etc.) from the
computed flow field.
14.2.10 Make Comparisons of the Results
The computed flow properties are then compared to results from analytic, computational, or
experimental studies to establish the validity of the computed results.
14.2.11 Repeat the Process to Examine Sensitivities
The sensitivity of the computed results should be examined to understand the possible differences
in the accuracy of results and / or performance of the computation with respect to such things as:
• dimensionality
• flow conditions
• initial conditions
• marching strategy
• algorithms
• grid topology and density
• turbulence model
• chemistry model
• flux model
• artificial viscosity
• boundary conditions
Further information can be found on the pages entitled Verification & Validation Assessment.
14.2.12 Document
Documenting the findings of an analysis involves describing each of these steps in the process. Again,
the user normally does not have the luxury of time to perform all of these, but it is mentioned here
for completeness.

14.3 Comparison of Some Commercially Available CFD Software

There are multitude of commercially available CFD tools and in-house software designed for special
needs. On a commercially available front, there are about 30 packages and three of them, general
purpose products, covering about %50 percent of market. They are ANASYS© CFX and FLEUNT, and
CD© Star-CD, and its sister product Star-CCM+ with comparison of these 3 codes been made on Table
14.1. An OpenFOAM is another arrival which is open source and free to use. The OpenFOAM project
emerged from the Field Operations And Manipulations (FOAM) concept at Imperial College London
in the late-1980s. It has one disadvantage: namely lack of GUI interface but on the plus side it has
open-source routine which are easily to manipulate. Each has its own strength and shortcoming and
the user is advised to choose according to specific needs. But in general, they are yields to the same
results for general application. A good source of what is available and their associated problems is
the CFD-online site. Beside these, there are number of research oriented codes which attribute to
specific problems and needs.
 397

One of latest study in inter code comparison is study to be done by [Bayraktar et al.]471. There,
underlying benchmark problem which have been defined in 1996, is a steady simulation of flow
around a cylinder at Re = 20. The benchmark problems are studied with three CFD software packages,
OpenFOAM, CFX and in-house code FeatFlow which employ different numerical approaches to the
discretization of the incompressible Navier-Stokes equations. It sets for finite volume method
(OpenFOAM), element based finite volume method (CFX), and finite element method (FeatFlow)
respectively. As for result, the CFD software package with high order finite element approximation
has been found to be computationally more efficient and accurate than the ones adopting low order
space discretization methods472.

Supported Pressure Density Based Multi- System Discretization

Grid Type Based grid Solver
Fluent Structured SIMPLE Coupled Algebraic Gauss- UD
Unstructured SIMPLEC Implicit Geometric Seidel TVD
Hybrid PISO Preconditioned QUICK
Star- Structured SIMPLE Conjugate UD
CD Unstructured SIMPLEC ----------- ----------- Gradient TVD
Star- Hybrid PISO OUICK
CCM+ CD
CFX Structured SIMPLE Algebraic UD
Unstructured ----------- Coupled ILU TVD
Hybrid QUICK
CD

Table 14.1 An Overview of some commercially available CFD software

14.3.1 AcuSolve© (www.altair.com)

It is a leading general-purpose Computational Fluid Dynamics (CFD) solver that is capable of solving
the most demanding industrial and scientific applications. AcuSolve’s robust and scalable solver
technology empowers users by providing unparalleled accuracy on fully unstructured meshes.
Applications ranging from steady RANS simulations to complex, transient, multi-physics simulations
are handled with ease and accuracy.
14.3.2 OpenFOAM© (www.openfoam.com)
OpenFOAM is a free, open source CFD software package produced by a commercial company,
OpenCFD Ltd. It has a large user base across most areas of engineering and science, from both
commercial and academic organizations. OpenFOAM has an extensive range of features to solve
anything from complex fluid flows involving chemical reactions, turbulence and heat transfer, to solid
dynamics and electromagnetic.
14.3.3 OpenFlower (sourceforge.net/projects/openflower)
OpenFlower is a free and open source CFD code (for Linux and Windows) mainly intended to solve
the turbulent incompressible Navier-Stokes equations with a LES approach. It can deal with arbitrary
complex 3D geometries with its finite volume approach.

471 E. Bayraktar_, O. Mierka and S. Turek, “Benchmark Computations of 3D Laminar Flow Around a Cylinder
with CFX, OpenFOAM and FeatFlow”, Institute of Applied Mathematics (LS III), TU Dortmund Vogelpothsweg
87, D-44227, Dortmund, Germany.
472 See Pervious.
 398

14.3.4 FLASH (flash.uchicago.edu)

A modular, parallel adaptive-mesh code initially designed for thermonuclear runaway problems but
now capable of a wide variety of astrophysical problems. Includes modules for MHD, nuclear burning,
radioactive cooling, self-gravity, particle dynamics, and cosmological expansion. I can give you the
most help with this code since I am one of the principal authors. FLASH is described in Fryxell et al.
2000, ApJS, 131, 273; extensive current documentation is available at their web site.
14.3.5 GADGET© (www.mpa-garching.mpg.de/~volker/gadget)
A parallel N-body tree plus smoothed particle hydrodynamics (SPH) code. Gadget can be used to
address a wide array of astrophysical interesting problems, ranging from colliding and merging
galaxies, to the formation of large-scale structure in the Universe. With the inclusion of additional
physical processes such as radioactive cooling and heating, Gadget can also be used to study the
dynamics of the gaseous intergalactic medium or to address star formation and its regulation by
feedback processes. Gadget is described in Springel 2005, MNRAS, 364, 1105 and in documentation
available at its web site.
14.3.6 HYDRA© (hydra.mcmaster.ca/hydra)
Adaptive particle-particle-particle-mesh (AP3M) plus smoothed-particle hydrodynamics (SPH) code.
Designed for (but not restricted to) cosmological simulation (includes a cosmological initial
conditions generator).. Current documentation is available with the Hydra source code.
14.3.7 ZEUS-MP (lca.ucsd.edu/portal/software)
Parallel, non-adaptive hydro- and magneto hydrodynamics code with self-gravity and radiation.
Zeus-MP is extensively discussed in astro-ph/0005109 and references therein.
14.3.8 ANSYS CFX© (www.ansys.com/Products/Simulation+Technology/Fluid+Dynamics)
ANSYS CFX computational fluid dynamics (CFD) software delivers best-in-class technology no matter
what level of CFD is required. Using ANSYS Design Modeler, ANSYS CFX works with your CAD
package. The ANSYS Workbench platform is used through all stages of a complete CFD simulation
providing consistency, persistence of information and a convenient work flow. ANSYS CFX guides the
user through the physics setup process. Post-processing provides graphics and hard numbers as well
as animations that can be used by themselves or in presentations. ANSYS CFX also includes a freely-
distributable 3D viewer for sharing 3D designs with colleagues. In addition, the ANSYS CFX suite
contains industry-leading CFD solver capabilities using a unique multi-grid linear solver, essentially
linear parallel CPU scalability and access to a broad set of models that provide accurate answers to
deliver real world results. With ANSYS CFX it is simple to add more functionality while continuing to
seamlessly support and leverage the investments in existing CAE assets. ANSYS CFX delivers the
ability to apply the most powerful and precise CFD technology to virtually every fluid engineering
problem. Explore in depth the technology and advantages of ANSYS CFX using the chart below or
menu on the left.
14.3.9 ANSYS ICEM CFD© (www.ansys.com/Products/Other+Products/ANSYS+ICEM+CFD)
From CAD to mesh generation for analysis, ANSYS ICEM CFD provides sophisticated geometry
acquisition, mesh generation, mesh editing, a wide variety of solver outputs and post-processing.
ANSYS ICEM CFD is the only Universal pre-processor for analysis including FEA, CFD and other CAE
applications such as particle transport and computational electro-magnetic. Used for engineering
applications such as computational fluid dynamics and structural analysis, ANSYS ICEM CFD’s mesh
generation tools offer the capability to parametrically create grids from geometry in multi-block
structured, unstructured hexahedral, tetrahedral, hybrid grids consisting of hexahedral, tetrahedral,
pyramidal and prismatic cells; and Cartesian grid formats combined with boundary conditions.
ANSYS ICEM CFD is the Swiss army knife of meshing, our product vision is to provide a variety of
flexible tools that can generate “any” type of mesh on “any” type of geometry for “any” solver or “any”
 399

application.
14.3.10 FLUENT© (www.fluent.com)
Replaced by Ansys Workbench. The broad physical modeling capabilities of FLUENT have been
applied to industrial applications ranging from air flow over an aircraft wing to combustion in a
furnace, from bubble columns to glass production, from blood flow to semiconductor manufacturing,
from clean room design to wastewater treatment plants. The ability of the software to model in-
cylinder engines, aero acoustics, turbo machinery, and multiphase systems has served to broaden its
reach.
14.3.11 COMSOL Multi-physics© (www.comsol.com/products/multiphysics)
COMSOL Multi-physics is a modeling package for the simulation of any physical process you can
describe with partial differential equations (PDEs). It features state-of-the-art solvers that address
complex problems quickly and accurately, while its intuitive structure is designed to provide ease of
use and flexibility. You can easily model most phenomena through predefined modeling templates.
Modifying these to specific applications is possible through equation-based modeling capabilities. To
deal with the increasing demand for realistic representations of the world around us, you can easily
model systems of coupled physics phenomena. COMSOL Multi-physics provides a friendly, fast and
versatile environment for multi-physics modeling. Fast results and unprecedented flexibility make
COMSOL Multi-physics the ideal modeling and simulation software for research, product
development, and education.
14.3.12 CFDRC© (www.cfdrc.com)
CFDRC offers unique capabilities for Multiphysics, Multiscale, and Coupled Simulations of fluid,
thermal, chemical, biological, electrical, and mechanical phenomena for real-world applications.
CFDRC’s technologies, products, and services enable better understanding of complex problems, and
lead to better decisions resulting in better concepts, designs, products and systems.
14.3.13 STAR-CD/STAR-CCM© (www.cd-adapco.com)
The STAR-CD solver provides one of the most effective numerical methodologies available in an
industrial CFD code with the high level of accuracy needed for complex unstructured meshes. This is
delivered with the speed, efficiency and robustness demanded by engineering design and
development cycles. STAR-CD uses state-of-the-art, proprietary numerical schemes to achieve the
highest levels of accuracy in both steady and transient simulations, making this solver one of the least
sensitive to mesh type and quality, including distorted tetrahedral meshes. Remarkably, this has been
achieved without sacrificing efficiency or robustness. So, whatever the choice of mesh or engineering
application, the STAR solver will provide the best solution in the shortest time. A particular feature
of STAR-CD is its fast CPU performance for transient flows. As the first to introduce moving mesh into
a CFD code, we have always been technology leaders in this area. The meshes can not only move and
deform, but they can also slide along non-matching interfaces; furthermore, selected cells or cell
regions can be deleted or added, detached and again attached to the core model.
14.3.14 FLOW3D© (www.flow3d.com)
FLOW-3D is a powerful modeling tool that gives engineers valuable insight into many physical flow
processes. With special capabilities for accurately predicting free surface flows, FLOW-3D is the ideal
software to use in your design phase as well as in improving production processes. FLOW-3D is an
all-inclusive package. No special additional modules for meshing or post-processing are needed. An
integrated graphical user interface ties everything together, from problem setup to post-processing.
14.3.15 TACOMA© (www.ge.com)
Has been under development at General Electric Global Research for16 years and it been used for
turbomachinery application. It comprises almost 500,000 lines of code, from twenty or more authors.
The equations solved are the steady, Reynolds-averaged Navier-Stokes equations (RANS), or the
 400

unsteady, Reynolds-averaged Navier-Stokes equations (URANS). TACOMA is a density-based, multi

block, time-marching code using the well-established Jameson, Schmidt, Turkel (JST) algorithm.
Available turbulence models include several flavors of Wilcox’s (k, ω) model and Menter’s SST model.
A two-equation transition model due to Menter is available.
14.3.16 CONVERGE™ ( www.convergecfd.com)
This is a general purpose 3D CFD Code with focus on engine simulation. It claims to be is very efficient
and accurate, where complex geometries and/or moving geometries and chemistry are to be
modeled. Although it is a general purpose CFD tool, but is primarily developed as Internal
Combustion Engine simulations (ICE). There are many and flexible ways to adjust the simulation
grid resolution in space and in time, including using a stationary and orthogonal mesh in the interior
of your domain, whose grid density automatically varies to resolve gradients using Adaptive Mesh
Refinement (AMR). CONVERGE is loaded with the physical models needed to accurately simulate
HCCI, Diesel and spark ignited engines. Included are advanced spray and combustion models,
including the SAGE detailed chemistry solver. Of course, CONVERGE runs well in parallel as well.

14.4 NASA Developed CFD Software

14.4.1 FUN3D
FUN3D is a finite-volume, node-centered, unstructured-grid RANS solver, which is widely used for
high-fidelity analysis (see Figure 14.3). FUN3D solves the governing flow equations on mixed-
element grids; the elements are tetrahedral, pyramids, prisms, and hexahedra. At median-dual
control-volume faces, the inviscid fluxes
are computed using an approximate
Riemann solver. For second-order
accuracy, face values are obtained by a
MUSCL scheme, with unweighted least-
squares gradients computed at the nodes.
The viscous fluxes use full approximation
of viscous stresses. For tetrahedral
meshes, the viscous fluxes are discretized
using the Green-Gauss (cell-based)
gradients. For non-tetrahedral meshes,
the edge-based gradients are combined
with Green-Gauss gradients. The diffusion
term in the turbulence model is handled in
the same fashion as the mean flow viscous
terms. FUN3D uses the SA-neg variant of
the SA turbulence model that admits Figure 14.3 Projectile Simulation (BMI Corporation)
negative values for the Spalart turbulence
variable. This variant was designed for improved numerical behavior. The SA-neg model is identical
to the original SA model for positive values of the Spalart turbulence variable.
To solve nonlinear flow equations, FUN3D uses a Hierarchical Nonlinear Iterative Method (HANIM).
On the innermost level HANIM uses a preconditioner based on a defect-correction method and
iterates on a simplified first-order Jacobian with a pseudo-time term. One preconditioner iteration
involves a point- or line-implicit multi-color pass through the domain. The number of preconditioner
iterations may vary for different nonlinear iterations. A Generalized Conjugate Residual (GCR)
method uses the preconditioner solutions to converge linear residuals and to compute solution
correction. A nonlinear controller assesses the correction computed by the linear solver. The
controller is responsible for the CFL adaptation strategy and for deciding when to update the
Jacobian. As a result of this assessment, the suggested correction can be applied fully, partially, or
 401

completely discarded; the current Jacobian may be updated or reused in the next iteration; and the
current CFL number may increase, decrease, or stay the same. The nonlinear iterations can be tightly
or loosely coupled, i.e., operate on the mean flow and turbulence equations collectively or separately.
Initially, the CFL number is ramped over a prescribed number of iterations, but then it automatically
changes within prescribed bounds.
14.4.2 USM3D
USM3D is an unstructured finite-volume cell-centered RANS solver that has been widely used
within NASA, other U.S. government agencies, and industry as a workhorse for aerodynamic analysis
of complex configurations. A fully-implicit formulation is implemented implying that the auxiliary
solution variables at the grid nodes and boundary faces as well as the cell gradients are computed
solely from the current solution variables defined at the cell centers. Solution values at the nodes are
averaged from solutions at surrounding cells using a pseudo-Laplacian method, which is equivalent
to a least-squares minimization procedure with a linear fit. The second-order spatial discretization
of inviscid fluxes is accomplished by reconstructing solutions at the cell faces; the reconstruction is
based on solution gradients computed within cells. The reconstruction scheme corresponds to the
MUSCL scheme with k = 0. For this study, the cell gradients are evaluated with the Green-Gauss
integration using solution values at the nodes.
Inviscid fluxes are computed at each cell face using the upwind Roe’s FDS scheme. Face gradients
required for evaluation of viscous fluxes are computed from the Mitchells stencil. The convective
term of the SA turbulence model equation is approximated with the first-order accuracy. The velocity
gradients contributing to the source term of the SA model are computed at cell centers using the face
area average of face gradients. In the presence of grid lines, accuracy of the velocity gradients is
improved by augmentation with a directional gradient based on the line mapping. A HANIM solver
described in473 is used for nonlinear iterations. The USM3D preconditioner uses a defect correction
scheme. The approximate Jacobian for the mean flow equations is formed using the linearization of
the first-order FDS inviscid fluxes and a thin-layer approximation for the viscous fluxes. The
approximate Jacobian for a turbulence-model equation includes the contributions from the
advection, diffusion, and source terms. The advection term is linearized with a first-order
approximation. A thin-layer approximation is used for the diffusion term. The entire contribution
from the linearized source term is added to the diagonal.
The preconditioner equations are solved with point- or line-implicit Gauss-Seidel (G-S) iterations.
The mean flow and turbulence model preconditioner G-S iterations are loosely coupled. Residual
reduction targets are used for an earlier termination of G-S iterations to improve the runtime
efficiency. HANIM uses CFL adaptation as a comprehensive tool to improve robustness and efficiency
of nonlinear iterations. It provides two additional hierarchies over the USM3D preconditioner solver.
The hierarchies are an enhanced linear solver for the exact linearization of RANS equations and a
nonlinear control of the solution update. The linear solver uses GCR-based matrix-free methods,
couples residuals of the mean-flow and turbulence model equations, and is expected to meet certain
residual reduction targets. The nonlinear solution update strategy automatically checks solution
reliability (positive pressure and density) at cell centers and adapts the under-relaxation parameter
and pseudo-time step.
14.4.3 CFL3D
CFL3D is a structured-grid multi-block cell-centered finite-volume code widely applied for
analysis of complex flows. It has been used in many recent workshops involving complex turbulent
flows. It uses second-order, upwind-biased spatial differencing scheme (a MUSCL scheme
corresponding to k = 1/3 that allows a third-order accuracy in one dimension for the convective and

473Pandya, M. J., Diskin, B., ames L. Thomas, and Frink, N. T., “Assessment of Preconditioner for a USM3D
Hierarchical Adaptive Nonlinear Method (HANIM),” AIAA 2016-????, 2016.
 402

pressure terms, and second-order differencing for the viscous terms; it is globally second-order
accurate. Roe’s FDS method is used to obtain inviscid fluxes at the cell faces. The option to model the
full Navier-Stokes mean-flow equations is exercised for all cases. CFL3D uses the SA-neg scheme to
model eddy viscosity. In distinction from the other two codes that use a first-order approximation
for the convection term in the SA model, CFL3D uses a second-order approximation. The turbulence-
model diffusion term uses the thin-layer approximation. The iteration scheme is loosely coupled, i.e.,
first, the mean-flow equations are advanced with the eddy-viscosity fixed, then the turbulence-model
equation is advanced with the mean-flow solution fixed. CFL3D employs local time-step scaling, grid
sequencing, and multigrid to accelerate convergence to steady state.
14.4.4 Case Study - Grid Convergence for 3D Benchmark Turbulent Flows
Grid convergence studies are performed by [Diskin et al.]474 to establish reference solutions for
benchmark three dimensional turbulent flows in support of the ongoing turbulence model
verification and validation effort NASA. The benchmark cases are a subsonic flow around a
hemisphere cylinder and a transonic flow around the ONERA M6 wing with a sharp trailing edge. The
study applies widely-used CFD codes developed and supported at the NASA Langley Research Center:
FUN3D, USM3D, and CFL3D. Reference steady-state solutions are computed for the RANS in
conjunction with the Spalart-Allmaras turbulence model on families of consistently-refined grids
composed of different types of cells. Coarse-to-fine and code-to-code solution variation is described
in detail. For further details, readers should consult the work by 475.
14.4.4.1 Subsonic Flow Around a Hemisphere Cylinder
Five grid families are generated for this study. Unstructured grids of families 1 to 4 have triangular
faces on the hemisphere surface and no polar singularity. Each family has four levels of nested grids;
L1 is the finest grid level, L2 is the second finest grid level, etc. Statistics of grids from families 2 (Tet),
4 (prism/hex), and 5 (structure) are shown in Table 14.2. These grids have a polar singularity at
the axis attached to the apex of the hemisphere, i.e., along this polar axis, hexahedral cells degenerate
into prismatic cells. Unstructured grids corresponding to the same level have the same distribution

Table 14.2 Statistics of four finest grids for hemisphere cylinder grid families.

of grid nodes. In comparison with unstructured grids at the same level, family 5 (structure) grids
have the same number of surface elements on the hemisphere surface, the same distribution of nodes
on the cylinder surface, and more nodes on the hemisphere surface. All the L1 grids have the near-
surface normal spacing approximately corresponding to y+ = 0.5. FUN3D solutions are computed on
grids of families 2 (Tet) and 4 (prism/hex); SFE solutions are computed on family 2 (Tet) grids,
USM3D solutions are computed on grids of families 2 (tet) and 5 (structure), and CFL3D solutions are
computed on family 5 (structure) grids.
14.4.4.2 Geometry, Flow Parameters, and Boundary Conditions

474 B. Diskin, W. K. Andersony, M. J. Pandyaz, C. L. Rumseyx, J. L. Thomas , Yi Liuk, and H. Nishikawa, “Grid
Convergence for Three Dimensional Benchmark Turbulent Flows”, AIAA Aerospace Sciences Meeting, 2018.
475 See Previous.
 403

The geometry is taken from the experimental study reported by [Tsieh]476. In the experiment, the
radius of the hemisphere was 0.5 in., the body length was 10 in., and the unit Reynolds number per
foot was 4.2x106. Thus, in the computational domain with the unit length taken as 1 in., the
hemisphere radius is 0.5, the cylinder length is 9.5, and the Reynolds number is Re = 3.5 x 105 per
unit length. The reference solutions have been computed at the following flow conditions: the
reference Mach number Mref = 0.6, angles of attack of 0, 5, 10, 15, and 19 degrees, and the reference
temperature Tref = 540 degree R. Here, presents only solutions corresponding to the 19 degrees angle
of attack. The origin of the coordinate system is located at the apex of the hemisphere. The positive x
direction is the stream wise direction collinear with the axis of the hemisphere and cylinder. Figure
14.4 sketches the layouts of boundary conditions and shows the global view of a computational grid
with half-plane symmetry. The downstream computational boundary is located at the back of the
cylinder, x = 10. The outflow conditions specified at the downstream boundary are constant pressure
conditions corresponding to P = Pref = 1. The far field boundary is a hemisphere with radius of 100
units centered at x = 10; y = 0; z = 0.

Figure 14.4 Global view of hemisphere cylinder geometry and boundary conditions

14.4.4.3 Results for Hemisphere Cylinder

For solution visualization, Figure 14.5 presents the FUN3D solution computed on the prism/hex L1
grid. The pressure contours and streamlines are shown in two planes corresponding to y = 0, and x =
6.0. The pressure is non-dimensional by ρref a2ref , where ρref and aref are the dimensional freestream
density and speed of sound, respectively. In the symmetry plane corresponding to y = 0, the cross-
stream separation is characterized by downward ow velocity. The separation occurs behind the
hemisphere-cylinder junction and continues for the entire cylinder length. A minimum pressure is

Tsieh, T., “An Investigation of Separated Flow About a Hemisphere Cylinder at 0 to 19 Degrees Incidence in the
476

Mach Number Range from 0.6 to 1.5," AEDC-TR-76-112, 1976.

 404

observed on the leeside, upstream of the hemisphere-cylinder junction. A large primal vortex and a
smaller secondary vortex are shown in the cross flow planes corresponding to x = 6.0. The separation
locations of these primal and secondary vortices are similar to those documented in the experiment.
An off-body vortex is seen in the shear layer of the primal vortex, outboard of the secondary one.

Figure 14.5 Global View of Hemisphere Cylinder Pressure Contours using L1 grid at surfaces y = 0
(left) and x = 6 (right)

14.4.4.4 Forces and Pitching Moment

Grid convergence plots of the lift, total drag (including pressure and viscous components), and
pitching moment coefficients and the maximum eddy viscosity are shown in Figure 14.6. The value
of the characteristic grid spacing, h, is computed as h = N -1/3, where N is the number of degrees of
freedom (cells for USM3D and CFL3D, nodes for FUNFV477 and SFE). The aerodynamic coefficients
computed with different codes on different grid families are generally converging to the same limit
with grid refinement. Convergence of the maximum eddy viscosity is less clear, mainly because of the
disagreement between limit projections from FUNFV (prism/hex) solutions and other solutions,
even though the SFE and FUNFV (prism/hex) solutions agree well on the finest L1 grids. Overall code-
to-code aerodynamic coefficient variation from the L4 grids to the L1 grids is up to 20%. In this
estimate and in the rest of the paper, relative variation is computed with respect to the middle of the
variation range. Extrapolation to the infinite-grid limit is problematic because no reliable order of
convergence can be established. No solution appears to converge uniformly in all quantities. Three
solutions, USM3D (structure), FUNFV(prism/hex) and SFE, converge monotonically. Considering lift,
USM3D (structure) solutions show less than first-order convergence, i.e., the lift approaches the limit
from above with a concave shape. The FUNFV (prism/hex) and SFE lift curves approach the limit
from above with convex shapes, indicating a convergence order that is higher than first order. The
FUNFV (tet) lift appears to converge with first order on the three finer grids. The USM3D (tet) lift
converges from above and changes the curve shape from concave to convex.
Considering pressure drag convergence, the FUNFV (prism/hex) and SFE convergence curves
approach the limit with convex shapes from above, but intersect. Lacking an exact solution, we use a
quantitative characterization of observed solution variation to evaluate accuracy. Variation of the

477Two different discretization available in FUN3D are employed: the baseline finite-volume discretization
(FUNFV) and a recently implemented stabilized finite-element discretization (SFE) based on a Streamlined
Upwind Petrov-Galerkin formulation.
 405

aerodynamic coefficients computed on the L1 grids is described in Table 14.3. The largest relative
difference among all solutions is observed for the pitching moment and does not exceed 4.4%.
Accuracy of aerodynamic coefficients improves proportionally to degrees of freedom used in CFD
computations. This property is the foundation of all grid refinement studies. It also justifies the
expectation of accuracy benefits from tetrahedral-grid cell-centered formulations that provide more

Figure 14.6 Grid Convergence of Aerodynamic Forces for Hemisphere Cylinder

degrees of freedom on grids of the same level. The USM3D solutions use about six times more degrees
of freedom on grids of family 2 (tet) than other solutions on grids of the same level. Because grid
convergence shown in Table 14.3 is not regular, quantitative assessments of accuracy
improvements due to additional degrees of freedom are difficult and imprecise. Qualitatively, the
aerodynamic coefficients computed by USM3D(tet) on the L2 grid are within the variation range of
the L1 solutions. Looking at the grid convergence on the three finer grids, the maximum and
minimum values of integrated aerodynamic quantities have been generally exhibited by the CFL3D
solutions and the FUNFV (tet) solutions. (The only exception is that, for the viscous drag coefficient,
the minimum is exhibited by the SFE solutions). Relative variation among the core-group L1 solutions
is also shown in Table 14.3.
 406

The deviations of the CFL3D

solutions from the core-
group solutions may be
attributed to the thin-layer
approximation. The
abnormalities in the FUNFV
(tet) solutions observed on
the current grids are harder
to explain. In the limit of grid
refinement, all FUNFV and
SFE solutions are expected to
converge to the same Table 14.3 Hemisphere Cylinder: Variation of Aerodynamic
“infinite-grid" solution. On Coefficients on L1 Grids
the current grids,
nonphysical oscillatory solution modes resembling checker-board instabilities were observed in the
FUNFV (tet) solutions with the default MUSCL scheme coefficient, κ = 0.0. The FUNFV solutions
computed on grids of other families are smooth. Note that the default value of the MUSCL scheme
coefficient on non-tetrahedral grids is κ = 0.5. In this study, an increased coefficient of κ = 0.75 is used
for FUNFV (tet) solutions. Solutions with κ = 0.75 do not exhibit nonphysical oscillations, but appear
to be somewhat less accurate. Although not shown, FUNFV solutions with κ = 0.0 were computed on
tetrahedral grids by using the approximate mapping discretization method for inviscid fluxes478-479.
Approximate-mapping solutions do not exhibit nonphysical oscillations and provide aerodynamic
coefficients well within the core-group variation range.
14.4.4.5 Fine Grid Surface Pressure, Skin Friction, and Off-Body Variation
In this section, surface pressure and skin friction are shown for four sets of solutions: USM3D (tet),
FUNFV(prism/hex), SFE, and CFL3D solutions. Hereafter, only these four hemisphere-cylinder
solutions of the available set are shown mainly for conciseness and presentation clarity. First, global
views of solution variation on the L1 grids are shown. Figure 14.7 displays the surface pressure and
the x-component of skin friction at the symmetry plane corresponding to y = 0. The pressure
maximum indicating the leading edge stagnation is located near x = 0.03 on the windward side of the
hemisphere. The pressure minima are observed on the leeside near x = 0.3 and on the windward side
near x = 0.45. A zone of low pressure is also observed on the leeside at x > 4. Near the outflow
boundary, the leeside pressure increases and the windward pressure decreases, creating a small
negative-lift zone. The L1-grid surface pressure distributions computed with different codes are
almost indistinguishable. The fine-grid code-to-code differences in the surface pressure at the local
extrema located on the hemisphere are within 0.5%. Although comparisons with experimental data
are not the focus of this paper, the computed surface pressure agrees qualitatively with the
experimental measurements. Figure 14.7-(b) shows the x-component of the skin friction vector.
Note that the y and z components of the skin friction are zero at the symmetry plane. The skin-friction
profiles computed with different codes are similar in most places. The largest discrepancy of about
15% is observed on the leeside, near x = 6.5. Other places of noticeable but more local discrepancies
are downstream from the hemisphere-cylinder junction at x = 0.5 and near the outflow are non-
dimensionalized by aref . The solutions are plotted along a vertical line attached to the upper surface
of the cylinder at x = 5 and y = 0.21. The view is chosen to show solution variation across the core of
the primal crossflow vortex located in this area. All off-body profiles are over plotted in the global

478 Diskin, B. and Thomas, J. L., “Comparison of Node-Centered and Cell-Centered Unstructured Finite-Volume
Discretizations: Inviscid Fluxes," AIAA J., Vol. 49, No. 4, 2011, pp. 836-854.
479 Diskin, B. and Thomas, J. L., “Erratum: Comparison of Node-Centered and Cell-Centered Unstructured Finite-

Volume Discretizations: Inviscid Fluxes," AIAA J., Vol. 51, No. 1, 2013, pp. 277.
 407

view.

Figure 14.7 Global View of Surface Pressure and Skin Friction at symmetry plane (y = 0) for
Hemisphere Cylinder

14.4.4.6 Effect of Grid Refinement on Surface Pressure and Skin Friction

Grid refinement and zooming have been applied to study solution variation near nontrivial flow
features and near regions of the largest solution differences observed in global views. Although not
shown, global views of the surface pressure profiles in the three planes (y = 0, z = 0, and x = 5)
computed on four grids in corresponding families are hardly distinguishable. The results in [Diskin
et al.]480 illustrates local grid convergence of the leeside surface pressure near the location of the
minimum pressure. Even in the zoomed view, only the coarsest L4 grid solutions are clearly
distinguishable. All codes appear to approach the same limit. Variation between L1 solutions is less
than 0.07%. The minimum pressure coefficients computed with USM3D, FUNFV, and CFL3D decrease
monotonically with grid refinement and show similar coarse-to-fine grid variation of about 7%. The
SFE minimum pressure coefficients show remarkably small coarse-to-fine grid variation of less than
0.5%, but converge non-monotonically with grid refinement. For complete discussion, please consult
the work by [Diskin et al.]481.
14.4.4.7 Transonic Flow Around an M6 Wing
The ONERA M6 experiment has been widely used for validation of CFD solvers482. A relatively simple,
well-documented geometry and a rich experimental database for a large variety of flow conditions
provide a unique combination for practical and inexpensive benchmark studies. Reference solutions
for transonic flows around the M6 wing are presented in this section. Authors believe that the
solutions computed on grids with more than 360 million degrees of freedom represent the largest
M6 computations conducted to date. The grid generation, coarsening, partition, and multigrid

480 B. Diskin, W. K. Andersony, M. J. Pandyaz, C. L. Rumseyx, J. L. Thomas , Yi Liuk, and H. Nishikawa, “Grid
Convergence for Three Dimensional Benchmark Turbulent Flows”, AIAA Aerospace Sciences Meeting, 2018.
481 See Previous.
482 Schmitt, V. and Charpin, F., “Pressure Distribution on the ONERA-M6-Wing at Transonic Mach Numbers," In

Experimental Data Base for Computer Program Assessment. Report of the Fluid Dynamics Panel Working
Group 04, AGARD AR 138, 1979.
 408

capabilities for the M6 model are described in an accompanying paper483. The M6 wing geometry
used in this study has been slightly redefined for numerical analysis of turbulence model simulations.
(See Figure 14.8).
14.4.4.8 Geometry, Flow Parameters and Boundary Conditions
Recently, a group at ONERA has considered the M6 model and its past experiments in greater
detail484-485. As part of this effort, the group has created a new CAD geometry for the wing. In this
geometry, the trailing edge of the
wing has been made sharp for the
purpose of this particular CFD
exercise. The reference solutions
for the OM6 wing are computed at a
freestream Mach number 0.84,
Reynolds number 14.6 x 106 based
on the unit root chord, and the angle
of attack of 3.06 degrees. The far-
field boundary in the shape of a
hemisphere is located at 100 unit
chords. The symmetry condition is
assigned at the plane containing the
root airfoil. Note that the
experiment used a splitter plate
near the wing root, which is not
modelled by CFD codes. This
discrepancy is believed to be the
cause of disagreement between CFD
solutions and experiment
measurements at inboard sections. Figure 14.8 M6 wing: pressure contours computed by USM3D
on family 4 prism/hex L1 grid
14.4.4.9 Grids for M6 Wing
The M6 grids used in this study are topologically equivalent to the full-geometry (y = 0 symmetry
plane) hemisphere-cylinder grids described before. The cylinder surface is mapped on the wing
surface with the specified wing section, and the hemisphere surface is mapped on the rounded wing
tip. Five nested grid families have been generated for the M6 geometry by using input profiles

Table 14.4 Statistics of Grids for OM6 Wing Grid Families

483 Nishikawa, H. and Diskin, B., “Customized Grid Generation and Processing for Benchmark Three-Dimensional
Flows," SciTech-2018, Kissimmee, FL, Jan., 2018, To be published as AIAA Paper.
484 Gleize, V., Dumont, A., Mayeur, J., and Destarac, D., “RANS simulations on TMR test cases and M6 wing with the

ONERA elsA flow solver (Invited)," AIAA Paper 2015-1745, 2015.

485 Mayeur, J., Dumont, A., Gleize, V., and Destarac, D., “RANS simulations on TMR 3D test cases with the ONERA

elsAfl ow solver," AIAA Paper 2016-1357, 2016.

 409

available at the TMR website. Statistics of L4 - L1 grids from families 1 (prism), 2 (tet), 4 (prism/hex)
and 5 (structure) are shown in Table 14.4. The far field boundary grids are not shown because they
look similar to the full-geometry extension of those for the hemisphere-cylinder configuration. The
surface grids have a moderate stretching toward the leading and trailing edges resulting in a
relatively coarse grid spacing in the mid-chord region. All the L1 grids have the first node off the
surface located at an average of approximately y+ = 0.5.
14.4.4.10 Results for M6 Wing
Figure 14.8 presents the contours of the surface pressure computed by USM3D on the prism/hex
L1 grid of family 4. The pressure is non-dimensionalized by ρref a2ref. A lambda shock is clearly visible
on the surface with the shock intersection located at about 80% of the wingspan. Grid convergence
of aerodynamic coefficients is described next. USM3D solutions have been computed on grids of
families 2 (tet) and 4 (prism/hex); FUNFV solutions have been computed on grids of families 1
(prism) and 4 (prism/hex);
and CFL3D solutions have
been computed on
structured grids of family 5
(structure).
All computations have been
conducted with no flux
limiters. Figure 14.9 (a)-
(b) show convergence of the
lift, total drag. No solution
converges monotonically for
all plotted quantities; thus,
no order property can be
deduced from the observed
convergence. Nevertheless,
all solutions approach the
same aerodynamic
coefficient values in the limit
of grid refinement. The
slopes of pitching moment
convergence curves shown in
Figure 14.10 are highly
irregular for solutions on grid
families 4 (prism/hex) and 5
(structure). For example, the
pitching moment coefficient
computed from the family 4
USM3D (prism/hex)
solutions decreases initially
with grid refinement from L4
grid to L3 grid, increases on
L2 grid, and decreases again
on L1 grid. Lift and pitching
moment convergence
observed for FUNFV (prism)
and USM3D (tet) solutions is
more regular. The differences Figure 14.9 M6 Grid Convergence of Aerodynamic Forces CL, CD
 410

among lift and pitching-moment

coefficients computed by all codes
on all grids do not exceed 6%. Drag
coefficients appear to be
converging with more regular
slopes, but do not provide
convergence patterns suitable for
the infinite-grid extrapolation. The
total and pressure drag coefficients
computed from CFL3D and FUNFV
solutions change the direction of
convergence on the L1 grids. The
viscous drag coefficient computed
from the USM3D (tet) solution
changes the direction of
convergence on the L1 grid. Only
USM3D (prism/hex) solutions
Figure 14.10 M6 Grid Convergence of Pitching Moment
converge monotonically for the
three drag coefficients. Relative
variation of drag coefficients computed on different grids is more significant than variation of the lift
and pitching moment coefficients; pressure and viscous drag coarse-to-fine variation is
approximately 30% and 16%, respectively. To establish solution accuracy, Table 14.5 shows code-
to-code variation of the forces, pitching moment, and maximum eddy viscosity on the L1 grid. Among
all integral aerodynamic coefficients, the maximum relative difference of 0.94% is observed for the
pressure drag. Maximum eddy viscosity variation exceeds 10%, indicating considerably higher
uncertainty than in integrated quantities.

Table 14.5 Variations of Aerodynamic Coefficients

For quantities that converge regularly in grid refinement, e.g., lift (Figure 14.9-(a)) and pitching
moment (Figure 14.10), the USM3D (tet) solutions appear to provide significant accuracy benefits
on same-level grids. Variation of surface pressure coefficients computed on the L1 grids at the
measurement sections used in the 9 of 35 experiment is shown in [Diskin et al.]486. Only three
computations, FUNFV (prism/hex), USM3D (prism/hex), and CFL3D (structure), are used in this

486B. Diskin, W. K. Andersony, M. J. Pandyaz, C. L. Rumseyx, J. L. Thomas , Yi Liuk, and H. Nishikawa, “Grid
Convergence for Three Dimensional Benchmark Turbulent Flows”, AIAA Aerospace Sciences Meeting, 2018.
 411

section for conciseness. The three codes

extract surface pressure at the same
span wise locations specified. The
wingspan is taken as b =
1.47601797621980 and the relative
axial position was computed as x/c = (x
- xmin)/(xmax - xmin). In the global view,
the L1 pressure profiles from different
codes are in close agreement. Small
oscillations in FUNFV solutions are
observed near the shocks at sections 1-
4. All solutions place shocks at the same
locations and identify the same
pressure minima on the lower and
upper wing surfaces. The pressure
profiles at leading and trailing edges
are indistinguishable. As compared to
other studies, an improved agreement
with the experiment is observed at
section 4. This improvement is
observed to be due to the increased grid
resolution provided by the L1 grids.
Figure 14.11 show a global view of
leeside pressure grid refinement at
sections 1 (η = 0.2). The pressure plots
show significant variation with grid
refinement. The mid chord grid spacing
on L4 and L3 grids is too coarse to
represent details of the pressure
profiles; the corresponding coarse-grid
solutions miss most of the shock
structure and are significantly different
from the solutions obtained on the fine
grids. All solutions computed on L2 and
L1 grids represent the shock details and
agree to each other remarkably well.
The grid convergence patterns of
USM3D and CFL3D solutions are quite
similar, as expected, because both
codes use cell-centered formulations.
FUNFV uses a node-centered
formulation and exhibits a different
convergence pattern.
All codes identify the minimum of
pressure at the same location, x/c ≈ Figure 14.11 M6 section (η=x/c=0.2) Global View of
0.39. The code-to-code discrepancy in leeside Pressure Grid Refinement
the minimum-pressure value is about
0.09%. Minimum pressure computed from all solutions converges monotonically with grid
refinement and demonstrates at least a second-order convergence rate. Most of L4 and L3 solutions
completely miss the double-shock structure in this region. Only the FUNFV L3 solution indicates a
 412

presence of a shock structure; USM3D and CFL3D L3 solutions miss it. However, all the L1 solutions
predict a double-shock structure in this region and agree well with each other. All the L1 solutions
predict a pressure plateau between two shocks at 0.3 < x/c < 0.35. The normalized x-direction grid
spacing at this location is Δx/c ≈ 0.02, providing just four grid nodes across the plateau. In spite of
the minimal grid resolution, the maximum code-to-code difference between pressure values on this
plateau is less than 6%.
14.4.4.11 Concluding Remarks
Detailed grid-convergence studies for two benchmark (3D) flows have been conducted to establish
reference solutions for Reynolds-averaged Navier-Stokes (RANS) equations using a “negative"
variant of the Spalart-Allmaras turbulence model [Diskin et al.]487. The benchmark flows are a
subsonic flow around a hemisphere cylinder and a transonic flow around the ONERA M6 wing (M6)
with a sharp trailing edge. The reference solutions have been computed with three widely used CFD
codes developed at NASA Langley, FUN3D, USM3D, and CFL3D. The codes use different discretization
approaches and iterative solution methods. Two different unstructured-grid second-order node-
centered discretization available in FUN3D are used for the hemisphere-cylinder computations: the
FUNFV discretization uses a standard finite-volume scheme and the SFE discretization uses a
recently added stabilized finite-element formulation. SFE is not used for M6 computations. USM3D
uses an unstructured-grid second-order cell-centered finite-volume formulation. CFL3D uses a
second-order cell-centered structured-grid formulation. Five families of consistently-refined nested
grids of different topology have been generated for the studies, including both structured grids and
unstructured grids with various types of elements. The finest family grids provide from 60 M to over
400 M degrees of freedom. To eliminate iterative errors, all solutions on all grids have converged to
near machine-zero residual levels. Although turbulence model validation is not the focus of the paper,
the reference solutions have been compared with available experimental data. The main thrust of the
paper is assessing variation between CFD solutions computed with different codes on different
families of consistently-redefined grids. The assessment includes two characteristics:
(1) relative coarse-to-fine variation of solutions computed on grids of different families and
(2) relative code-to-code variation of solutions computed on the finest family grids.
All codes show close agreement in predicting aerodynamic coefficients for the separated flow around
the hemisphere-cylinder configuration. The code-to-code discrepancy among all aerodynamic
coefficients computed on the _nest family grids is less than 4.5% and variation among a core group
of four solutions is less than 0.75%. The coefficients appear to converge to the same limit with grid
refinement, but no convergence order can be discerned for the observed convergence. There is more
uncertainty about the grid convergence limit of the maximum eddy viscosity. The surface pressure
and skin friction in different _ne-grid solutions over plot in most global views. A local disagreement
among the codes of about 15% is observed in the vicinity of the leeside just past the middle section
of the cylinder. Various off-body solution components probed outside of this region also over plot.
Local solution characteristics, such as surface pressure minima and the circumferential angle of
vortex separation and reattachment locations, also converge to the same limit with grid refinement.
The reference solutions compare well with available experimental data.
The reference solutions for a transonic flow around M6 have been computed using the three
formulations, FUNFV, USM3D, and CFL3D. The aerodynamic coefficients computed by different codes
on the finest grids of different families agree well; the maximum difference among all coefficients
does not exceed 0.73%. The difference in maximum eddy viscosity is 10.3%, which is much larger
than the corresponding difference in the aerodynamic coefficients. The surface pressure computed
with the three codes have been compared at seven OM6 wing sections. The pressure profiles

487B. Diskin, W. K. Andersony, M. J. Pandyaz, C. L. Rumseyx, J. L. Thomas , Yi Liuk, and H. Nishikawa, “Grid
Convergence for Three Dimensional Benchmark Turbulent Flows”, AIAA Aerospace Sciences Meeting, 2018.
 413

computed on the finest grids over plot in the global views. Away from shocks, all the pressure profiles
computed on the two finest grids are close to each other, within a 1-2% range. Increased grid
resolution allows for an improved resolution of the lambda-shock feature that was a challenge in past
M6 computations. As compared to previous studies available in the literature, the present solutions
on the finest grids provide an improved agreement with the experiment. Further details is available
in [Diskin et al.]488.

14.5 Symbolic Math Packages

14.5.1 Maxima (maxima.sourceforge.net)
Open-source symbolic math package based on the classic Macsyma computer algebra system.
14.5.2 Mathematica (www.wolfram.com)
Commercial symbolic math package, available on department machines (try man Mathematica). Also
available inexpensively through a UIUC site license.
14.5.3 Maple (www.maplesoft.com)
Another commercial symbolic math package.
14.5.4 MatLab (www.mathworks.com)
Yet another famous commercial symbolic math package, a bit more geared toward engineering and
numerical work than Mathematica or Maple.
14.5.5 MATHGL (http://mathgl.sourceforge.net/doc_en/index.html)
MathGL is a huge open-source plotting library for scientific graphics. It can be used for many
programming languages, in particular for C++. Mainly we will use the Figure library (implemented
by J. Gacon, student of CSE@ETH)

488B. Diskin, W. K. Andersony, M. J. Pandyaz, C. L. Rumseyx, J. L. Thomas , Yi Liuk, and H. Nishikawa, “Grid
Convergence for Three Dimensional Benchmark Turbulent Flows”, AIAA Aerospace Sciences Meeting, 2018.
 414

15 Best Guidelines for Optimal CFD Simulation

CFD does not provide an exact solution to all problems, but is in many cases a reliable tool that can
provide useful results when an experienced user employs it489. Still is some circles, considered an art
that science. An inexperienced user, on the other hand, may obtain very nice graphs that are very far
from being a prediction of the stated problem. Some of the problems arise from the many default
settings in commercial CFD codes, since the user may obtain results without knowing what the code
is doing by accepting settings that are not appropriate for the specific problem. The user must make
an active decision regarding each setting because many problems can arise from a user failing to
understand what the proper settings should be. CFD simulation contains both errors and
uncertainties. An error is defined as a recognizable deficiency that is not due to a lack of
knowledge, whereas an uncertainty is a potential deficiency that is due to a lack of knowledge.
Poor simulation results may arise for many different reasons; the user may make mistakes in
formulating the problem to solve or in formulating the geometry and meshing. Numerical errors may
occur due to lack of convergence and a poor choice of discretization methods. The models for
turbulence, reactions and multiphase flow are not exact descriptions of the real world, and poor
results may be introduced by not selecting the best model for the specific case. In the end, the
robustness and reliability of the simulations should be analyzed by sensitivity analysis, verification,
validation and calibration.

Figure 15.1 Component of Uncertainty within a Simulation

Figure 15.1 shows components of uncertainty within a simulation where Up is all the other flow
parameters which contribute to the uncertainty, i.e., boundary conditions, initial conditions, spatial
and temporal discretization, choice of turbulence, auxiliary equations, etc. In general, it is good
practice to study the obtained simulation results critically. CFD is not (yet) at the stage at which it
can be treated as a black box. Simulation results should be verified with experimental findings, fluid-
mechanics theory and, sometimes, instinct. It may also help to understand the limitations of the CFD
model if a simulation is performed for a case regarding which the results are known, prior to
exploring new ground with the model. Some common errors and recommendations for best practice
are given following.

489 Computational Fluid Dynamics for Engineers, Cambridge University Press.

 415

15.1 Element of Uncertainty

15.1.1 Geometry & Grid Generation
➢ Make sure that the CAD geometry is complete for the flow simulation. The CAD drawing
should be kept as simple as possible. All details smaller than the computational cells can in
most cases be removed, but small details on the surface can sometimes be important for the
flow, e.g. a welded joint may induce flow separation.
➢ Symmetric boundary conditions restrict the solution to a symmetric solution, and no
transport is allowed across the symmetry plane, e.g. bubble columns and fluidized beds are
poorly described by 2D simulations.
➢ When the inflow and outflow are not known exactly, they should be put far from the region
of interest.
➢ With constant-pressure outflow, specify the direction of the outflow (e.g. normal to the plane)
to minimize the pressure difference across the surface.
➢ Use pressure outflow for multiple outflow boundaries.
➢ Avoid having non-orthogonal cells close to the boundaries. The angle between the grid lines
and the boundary should be close to 90◦. Use body-fitted grids when the grid cannot be
aligned with the surface.
➢ Avoid the use of highly skewed cells. The angles should be kept between 40◦ and 140◦. The
maximum skewness should be < 0.95 and the average below 0.33. The aspect ratio should be
below 5, but may be up to 10 inside the boundary layer.
➢ The squish index describes how the cell faces are oriented, where 0 is a perfect cell and the
upper limit is 1. The recommendation is that the maximum squish index should be below
0.99.
➢ Perform a grid check to avoid problems due to incorrect connectivity. This can often be done
in the mesh-generator software.

15.1.2 Convergence
➢ Use different convergence criteria for different variables.
➢ Monitor integral quantities of solution-sensitive variables.
➢ Make global balances for mass, momentum and energy.
➢ Monitor the solution at specific important points.
➢ Test for steady state by switching to a transient solver.
➢ Plot the residual to evaluate whether the solution is poor in some regions of the
computational domain.
➢ Use more robust numerical schemes, e.g. initially employ first-order upwind, changing to a
higher order for the final iterations.
➢ Reduce the under-relaxation or CFL number initially.
➢ Examine the local residual. The convergence problem might be localized to one small region.
Use grid adaptation to refine or coarsen the mesh in areas where it is needed.
➢ Using a fine grid throughout may diminish the convergence rate.
➢ Solve steady-state problems transiently.
➢ Try different initial guesses, e.g. obtain an initial guess from a short transient simulation or
obtain an initial guess from a steady-state simulation.
➢ Solve for only a few variables at a time, e.g. solve for the flow field first and keep the velocities
constant whilst solving concentration and chemical reactions. Finally solve for all variables.
➢ Use the coupled solver for high-speed compressible flows, highly coupled flows with strong
body forces or flows being solved on very fine meshes. Keep in mind that the coupled solver
requires 1.5–2 times more memory than needed by a segregated solver.
 416

15.1.3 Numerical
➢ The exact inlet conditions for the turbulence properties are usually not known exactly. The
value of k/ε times the average velocity gives an order-of-magnitude estimation of how far
into the system the settings at the inlet will survive.
➢ Exact inlet conditions for LES are not possible. Modern CFD programs can generate simple
turbulent eddies at the inlet, but the inlet should be far from the area of interest in order to
allow a proper statistical distribution of eddies to develop. Periodic boundary conditions
should be used when possible, but may introduce unphysical correlations. As always, it is
important to verify the obtained solution critically.
➢ For RANS models, the lower level of y+ should typically be between 20 and 30 at the walls.
Some commercial CFD programs can be made to accept lower y+ levels by adjusting the
appropriate model. (Check the manual for actual values.) The upper limit of y+ is usually in
the range 80–100.
➢ LES models require additional treatment at no-slip walls.
➢ For low-Re turbulence models the first grid should be at y+< 4, preferably at y+ ≈ 1 with 5 –
10 mesh points below y+=20. Standard wall functions are not recommended for flow with a
negative pressure gradient, e.g. with flow separation at the wall.
15.1.4 Reaction
➢ Check the Dahmk¨ohler number. Slow chemical reactions with Da ≪ 1 are straightforward.
Very fast, mixing-controlled, isothermal chemical reactions with Da ≫ 1 can be modeled
rather accurately. Most other conditions with Da ≈ 1 and reactions that lead to heat formation
and changes in density will give uncertain results.
➢ Very low residuals as a convergence criterion for concentration are often required, and
monitoring integral quantities for mass balances or steady local concentrations usually
provides more reliable indicators than low residuals.
15.1.5 Multiphase Flows
Generally speaking, multiphase flows are more challenging than single-phase flows, and errors in
multiphase flow simulations are typically larger than those in single-phase simulations and
propagating faster. These errors may have a number of different origins. They are:
15.1.5.1 Not Knowing the Most Important Physical Mechanisms
Because of the wide variety of multiphase flow types, there is not one ‘generic’ model for multiphase
flows. Before attempting to model a multiphase flow system, it is very important to understand the
physical mechanisms occurring in the flow system. This includes understanding the most important
forces and mechanisms in the flow and the properties of the fluid(s) and/or solids, as well as a good
estimate of the length scales and timescales of the physical processes. Only with this knowledge can
appropriate models be selected and their shortcomings in a simulation estimated.
15.1.5.2 Closure Models
Most errors in multiphase flow simulations arise from shortcomings of the closure models employed.
Most closure models are empirically determined, which makes them applicable, strictly speaking,
only under conditions similar to those for the data they are built from. Analytical closure models are
developed for ‘ideal’ conditions, which are hardly ever met in reality. Using closure models for
conditions or regimes different from those to which they are applicable may be asking for trouble.
15.1.5.3 Time-Scale and Length-Scale Separation
In the derivation of turbulence models, a separation between the timescales and length scales of the
‘large’ eddies carrying the energy and the scale(s) at which energy is dissipated is assumed. Similar
assumptions are required in order to derive governing and closure models for multiphase flows. For
 417

instance, many closure models require the particles to be smaller than the large-scale flow structures
in a flow.
15.1.5.4 Choice of Model and Governing Equations
It is important to understand the implications and assumptions of each multiphase model, and under
what conditions it may be employed. For instance, in the derivation of the Euler–Euler or two-fluid
model, the pressure gradient over large interfaces, which is important in separated-flow situations,
is neglected. This means that such a model will not capture the dynamics of free surfaces very
accurately.
15.1.5.5 Numerical Errors
Many numerical errors potentially arising in single-phase calculations may also arise in multiphase
flow calculations. Therefore, it is important to check the best-practice guidelines for single-phase
flow computations. However, there may be additional problems. Many of the ideas employed in
solving multiphase flows arise from single-phase flows, leading to slow convergence, or, worse, an
erroneous result.
15.1.5.6 Avoiding Risks in Multiphase Flows
To minimize the potential problems occurring when performing multiphase flow simulations, the
following checklist may be employed,

I. If possible, start the simulation with a single-phase flow situation resembling the system. This
simulation can be optimized in terms of grid size, time step, etc. by using the best-practice
guidelines for single-phase flow.
II. Determine the regime of the multiphase flow in terms of dimensionless parameters (Re, We,
St,) . . . This enables the choice of suitable closure models and may give insight into the
expected flow situation.
III. Make an estimate of the forces acting on bubbles, particles or droplets and under which
conditions these forces will occur.
IV. Make a suitable selection of the turbulence model and decide which terms (and coupling to
the dispersed phase) are important.
V. If possible, start with a geometry, flow properties and dispersed-phase properties similar to
those of a system of which you know the behavior or for which experimental data are
available. This creates confidence in the models employed.
VI. If there is a large size distribution of the dispersed phase, a multi-fluid approach might be
required for the dispersed phase. This allows the use of a range of size classes, which can be
monitored separately. Size distributions can have a big effect on the flow.
VII. First-order-accurate models, such as the VOF model and the surface-tension models, require
a very fine mesh – in these cases a relatively mesh-independent solution is very important.
VIII. Make sure that iterations are well converged. Many popular commercial CFD solvers will start
with a new time step when a specified maximum number of iterations is reached, regardless
of convergence criteria. This may be detrimental in terms of the quality of results obtained.

15.2 Verification, Validation, Calibration and Certification

Verification is a procedure to ensure that the program solves the equations correctly. Validation is
done to test how accurately the model represents reality, and calibration is often used to adjust the
simulation to known experimental data in order to study parameter sensitivity (‘what if’) in the
design process. More detailed discussion of V&V, Uncertainty, Errors and Sensitivity analysis would
present later on. The follow-on work to determine the degree to which results from a CFD code are
an accurate representation of the real world from the perspective of the intended uses of that code
 418

has been referred to as “calibration”, “certification”, or just further validation. [Melnik, et. al,]490
describe a process for “certification” which is indicative of our current practices. The reality is that
these processes for determining usability of a CFD process for an intended purpose are “…ongoing
activities that do not have a clearly defined technical completion point”. Rather than referring to
these processes as “calibration” or “certification” we will just continue to call them “validation” and
validation for an intended purpose 491.
15.2.1 Validation for an Intended Purpose.
A CFD code by itself can never be validated for an intended use because there are just too many
variables. These variables include those user accessible “knobs” within the solver that control the
discretization, dissipation, convergence strategy, turbulence model, etc. and the BIG external variable
called the grid. As a minimum it is the process of how a particular CFD solver is run along with its
related grid generation that must be validated. And as previously stated, using wind tunnel data for
CFD validation is best done on a trend and increment basis which means we will be running CFD to
compare those trends and increments. The question is not can CFD give me a great answer for one or
two test cases, but can the CFD “processes” give me good answers for a range of cases when run by a
competent engineer? This is what validation for an intended purpose is all about 492.

15.3 Simulation for Everyone

Putting simulation into the hands of the non-expert is a subject that elicits much discussion within
CFD committees493. A crucial element is the relationship between the simulation expert and the non-
expert, where the responsibility of the expert ends and the ability for the non-expert to be able to use
simulation safely starts. This requires simulation experts to design smart simulation applications and
is somewhat analogous to the traditional handbooks where experts would develop solutions in a
parametric form and a working engineer didn’t have to know how those solutions were developed
but would use a formula and perhaps some graphs in conjunction with the formula to come to some
kind of a prediction. For this approach to work with simulation, it is extremely important for the
simulation application to have incorporated into it a solution validation & verification procedure,
since inexperienced and non-expert users are less likely to be able to judge accuracy.
The development of the most effective smart applications requires both external software and
internal company design processes. In this case, the experience and knowledge of the in-house
application experts can be embedded into the interface for the simulation tool. Whilst an expert in
simulation is required to develop these context-specific tools, most small/medium enterprises often
do not employ dedicated simulation engineers so need to rely on external consultants. The most
difficult part of developing an application for non-expert users is getting the technical requirements
from the organization defined in such a way that they can be distilled into an application that can
then be released to the general population. It requires a process that has a clear statement of what
senior engineering staff expect these applications to do and how do they define success and
acceptance for them. The benefits are huge once the tool is created as it can then be inserted into the
workflow, saving engineering time and making solutions operator-independent but it requires
acceptance into the culture of the organization.
According to Jean-François Imbert , “A smart application can’t be commercial software only. It is
a tool which combines commercial software and embedded knowledge-based processes and

490 Melnik, R.E., Siclari, M.J., Barber, T., and Verhof, V., “A Process for Industry Certification of Physical Simulation
Codes,” AIAA-1994-2235, June 1994.
491 Edward N. Tinoco, “CFD Uncertainty and Validation for Commercial Aircraft Applications”, Boeing

Commercial Airplanes, Seattle, USA

492 Tinoco, Edward N. and Bussoletti, John E., “Minimizing CFD Uncertainty for Commercial Aircraft Applications

(Invited),” AIAA-2003-407, January 2003.

493 This article, written by Althea de Souza , first appeared in NAFEMS Benchmark Magazine, April 2017 edition.
 419

rules in order to secure the product simulation and design.” This is easier said than to be done.
From previous experience, all the commercial vendors, has their own deadline and customers. And
their cooperation regarding this, would be hard to come by. Rendering to Oxford Dictionaries,
democratization is the action of making something available to everyone. In the context of
engineering simulations, it is making simulation tools available to everyone. It could be argued (and
has been) that the tools have been available to everyone for some time, including free and open
source codes, tools that can be run on laptop computers with relatively modest specifications and
tools with easy to use interfaces and GUIs. The limiting factor though is the ability to be able to use
those tools effectively and responsibly. The complexities of the simulation are generally hidden from
the user (no need to see the mesh or select the turbulence model) and the terminology used is that
of the application engineers, rather than simulation engineers.
There are now a number of organizations offering CAE app-building facilities, whether they are
mainstream code vendors allowing users to create apps from their more general purpose code or
independent organizations. There has been a growing call for software vendors to provide users with
tools for solution verification. These tools can be incorporated into the approved workflow for use
by non-experts. There is still a problem with validation; it is directly tied to the user. A strong
argument is that this is as it should be, since validation entirely depends on how the tool is used and
has little to do with the mathematical solution procedure. This is particularly important with general
purpose codes that are used for a wide range of applications. Can vendors put anything into the tools
to guide the users? Asking for a demonstration of simulation validation from non-expert users will
always prove challenging and so tools designed for this class of user should be built very carefully
around a proven simulation approach and tightly restricted to prevent use outside the scope of
applicability.
Distinguishing between the development of design rules and the application of design rules can help
to differentiate when an expert or non-expert user can undertake an activity. Developing design rules
requires validation and expert involvement, whereas applying design rules can be done by non-
experts using smart tools. Smart applications should have been verified and have embedded
modelling rules so users are not expected to carry out verification. Validation of both the software
and the tightly constrained modelling approach is a pre-requisite for smart tools and should be
conducted before they are released to users494.

15.4 CFD on a Deadline

Have you ever feared a deadline because your model took longer to solve than expected? What is an
Optimal Workflow for Complex Modeling Projects on a Deadline? Unfortunately not an easy answer
available. But there are common causes to avert them. There are few suggestion, but not too many.
A full 3D model might take anywhere from minutes to several hours to run. Thus, every mistake in a
complex model takes longer to find out about. Moreover, you can’t be sure what causes the issue.
[Ruud Börger & Andrew Griesmer] of COMSOL Multiphysics® exploring this question and came up
with some suggestion. The first option is to bite the bullet and try to debug or optimize the 3D model.
This is of course more reliable and error proof approach, but harder to implement. The other, if your
model geometry allows that, is to create a 2D or asymmetric model of the original one, and try to use
it for debugging purpose.
15.4.1 Optimizing your Modeling Workflow
There are many situations where a 3D model with complicated physics is required. A common
approach is to take all of the expected “ingredients”, put them together, and reach your goal. But
what if something unexpected happens? What if the model doesn’t converge? Any of the following
could be the problematic:

494 David Quinn, “The Big Issues in Engineering Simulation- Democratization”, May 25, 2017.
 420

➢ For multi-physics cases, any boundary condition that you used in the first, second, third, etc.
on physics interface
➢ A coupling between any of the physics
➢ Material behavior
➢ The mesh requirements for any of these factors
➢ An unexpected or unknown phenomenon

This solution is more geared toward multi-physics models. A full 3D model might take anywhere from
minutes to several hours to run. Thus, every mistake in a complex model takes longer to find out
about. Moreover, you can’t be sure what causes the issue. There are remedies pipelined and following
suggestion to implement:
1. Using achieved cases with similar phenomena (i.e., educating by example).
2. Start your own model in 2D or 2D axisymmetric, if applicable.
3. Test the major physical principles of your model in a single-physics environment.
4. After each successful step, add complexity, such as multi-physics effects, nonlinear materials,
etc.

Not to be outdone, STAR-CCM+® like many other commercial vendors, provides a simple way for 2D
simulation to test out designs and boundary conditions. In STAR-CCM+ there are two new features
that will put that problem to rest. STAR-CCM+ is a 3D based tool so most of the controls are based on
surface and volumetric selections thus we needed a way to bridge the gap between a traditional 2D
meshes and the inputs for STAR-CCM+. To do this all parts that are to be run in 2D need to have a
planar surface on the Z=0 plane, from there it could be a revolver, extrude or frankly any shape.
“Badge for 2D”; is the feature flags the surfaces that lie on the Z = 0 plane and marks the perimeters
as 2D boundaries, thus preventing unneeded part surfaces from being sent to regions. There are
other issues to be considered when using 2D instead of 3D models. Beside geometry, the first thing
to recall is turbulence. Mixing and more generally diffusion phenomena are never purely 2D. If you
are in turbulent conditions and you are using RANS models, 3D diffusion might be accounted for by
the transport equation of your turbulence quantities.

15.5 Myths of Computational Fluid Dynamics

It is an interesting topics which haunts the CFD users, or potential CFD users, to the date. Although
the original posting is with regards to commercial CFD and its market, nevertheless, the benefits
outweigh the commercial usage. Therefore, we included here to clear some of common
misperceptions in regard to CFD495.
15.5.1.1 CFD is Difficult and Take Too Long to be Used in the Design Process
This myth has a historical basis. Like FEA codes in the distant past, CFD codes of the 1980s and 1990s
were difficult to use496. Fit-for-purpose meshing, choice of solution numeric, turbulence modeling,
achieving and judging solution convergence, assuring result fidelity, and correct result interpretation
were all once expert-only activities. Today, the skills a mechanical designer needs to operate the CFD
software are simply knowledge of the CAD system and the physics pertaining to the product, both of
which the majority of design engineers already possess. The greatest time sink for CFD has always
been the meshing process, with a considerable amount of manual intervention needed to achieve
acceptable mesh quality by eliminating gaps and overlaps, reducing skewness, aspect ratio, warpage,
and controlling the volume of individual cells (cell size ratio to neighbor cells, smallest cell size, and

495 Ivo Weinhold & John Parry,” The 10 Myths Of Computational Fluid Dynamics”, Mentor Graphics White Paper,
Mechanical Analysis Division, 2014.
496 “The Five Myths of Computational Fluid Dynamics”, NAFEMS BenchMARK magazine, April 2008, pp. 28-29.
 421

mesh distribution). As design inherently involves changing geometry, this semi-manual process had
to be repeated for each design iteration. All of these steps can now be fully automated using native
3D CAD data directly for fluid flow simulations without the need for translations or copies. New parts
and features resulting from design changes can be meshed in a matter of minutes, dramatically
reducing the time required for analysis.
15.5.1.2 Accuracy has to be sacrificed to use CFD during the Design Process
Traditional CFD has taken the approach of attempting to resolve everything on the volume mesh.
This was not always the case. Years ago, limited computing power meant that wall functions were
the only way to represent the effect of the boundary layer between the solid surface and the bulk
flow. These wall functions were originally applied only in the near-wall cell. Using mesh to resolve
the boundary layer was prohibitively expensive, and the result quality achievable on any tractable
mesh for all but the simplest of situations was far inferior to what could be achieved using empirically
based wall functions, which could be further refined to account for surface roughness497.
Small discrepancies between physical reality and simulation result for a mesh cell were transmitted
to the neighbor cells by convection and diffusion as a natural part of the solution. Compounded across
the mesh, this compromised overall solution accuracy. Solution convergence was also much harder
to achieve. The effect of the variation in mesh quality was such that the physical effects of minor
geometric variations were masked. This issue has plagued traditional body-fitted CFD ever since. The
industry has, of course, risen to the challenge in various ways: Increases in computing performance
have allowed ever finer meshes to be used. More mesh allows the variation in size and shape from
one mesh cell to the next to be reduced, improving orthogonality. Considerable effort has continued
to be expended on the development of automated mesh generators, allowing mixtures of hexahedra,
tetrahedral, and prisms to be used. More recently, polyhedral meshes, which have better
orthogonality properties for any arbitrary flow direction4, have been constructed, e.g., by
aggregating tetrahedral. Finally, more sophisticated numerical schemes that improve the estimation
of cell face fluxes and pressure coupling have been devised.
The new generation of CFD software, however, comes with key technologies that allow high-quality
results to be obtained fast. In a matter of a few minutes, it is possible to construct an octree Cartesian-
based mesh that automatically refines around solid–solid and solid–fluid interfaces, sidestepping the
need to separately generate surface and volume meshes. Designers can control mesh density with a
single slider, maximizing ease-of-use. Being Cartesian, the mesh has the highest possible numerical
mesh quality, because the cells in Cartesian and Cartesian-based meshes are perfectly
orthogonal498,499. Hence, it is not possible to improve the mesh quality through manual intervention.
This unparalleled ease-of-use actually leads to a significant improvement of result accuracy, because
the fidelity of the CFD simulation no longer has to be limited by project time constraints. In an
industrial setting, engineers who are using traditional CFD do not have time to refine y+ values across
the whole model even when the geometry is relatively simple. The more complex the geometry, the
greater the benefit obtained from using octree Cartesian meshing. Octree Cartesian meshes are highly
suited to solution-adaptive mesh refinement, where the mesh self-refines as the solution progresses
to ensure that gradients are adequately captured, for example, in shock capture.
15.5.1.3 Experts are Needed to Get Accurate CFD Simulation Results
Building in ease-of-use lowers the level of simulation-specific expertise needed, thereby allowing
engineers to focus on the engineering problem. The numerical and physical modeling expertise of the
software developers is accessed by the design engineer via the automation provided within the tool.

497 M. R. Malin and J. D. Parry, “Turbulent Heat and Momentum Transfer in Rough Tubes”, Journal of
computational Fluid Dynamics and its Applications, Vol. 1, No. 1, January 1988, pp. 59-80.
498 John Parry and David Tatchell “Flomerics’ EFD Meshing Technology: A White Paper”.
499 “Advanced Immersed Boundary Cartesian Meshing Technology in FloEFD” MGC 02-11 TECH9690-W
 422

Inputs are limited to what needs to be known about the problem being described: boundary
conditions, materials, etc. applied directly to the native 3D geometry within the CAD system. Thus,
non-simulation experts can get reliable, repeatable, high-quality simulation results. We are in no way
ignoring the value of analysis experts. Such people are the very reason that CFD has grown to become
what it is today, pioneering new applications that have driven the development of new physical and
numerical models, etc. In late design, when the geometry is stable, traditional CFD can be applied,
leveraging the availability of analysis-ready CAD which is refined using a CAD-embedded CFD
solution, to provide further design verification if desired. Alternatively, analysis experts can
themselves use CAD-embedded CFD during design. CAD-embedded CFD and traditional CFD can, and
arguably should, complement one another. CAD-embedded CFD also integrates well with other CAE
tools used in product design, allowing temperature data to be exported as a thermal load for use in
Simulation and NASTRAN-based FEA solvers, among others, helping to accelerate other aspects of
the design.