Elements of Numerical Methods for Compressible Flows
The purpose of this book is to present the basic elements of numerical methods
for compressible flows. It is suitable for an advanced undergraduate or graduate
course, and for specialists working in high-speed flows. The book focuses on the
unsteady one-dimensional Euler equations, which form the basis for development
of numerical algorithms in compressible fluid mechanics. The book is restricted
to the basic concepts of finite volume methods, and is intended to provide the
foundation for further study and application by the reader. The text is supplemented
by end-of-chapter exercises.
Doyle D. Knight is Professor of Aerospace and Mechanical Engineering in the
Department of Mechanical and Aerospace Engineering at Rutgers – The State
University of New Jersey. He received his PhD in Aeronautics at the California
Institute of Technology in 1974. After two years service in the United States Air
Force as an Aeronautical Engineer and a one-year postdoctoral fellowship at the
California Institute of Technology, he joined the faculty of the Department of
Mechanical and Aerospace Engineering at Rutgers in 1977. He is a member of
the American Institute of Aeronautics and Astronautics, the American Society
of Mechanical Engineers, and the European Mechanics Society.

Cambridge Aerospace Series
Editors
Wei Shyy
and
Michael J. Rycroft
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.

J. M. Rolfe and K. J. Staples (eds.): Flight Simulation
P. Berlin: The Geostationary Applications Satellite
M. J. T. Smith: Aircraft Noise
N. X. Vinh: Flight Mechanics of High-Performance Aircraft
W. A. Mair and D. L. Birdsall: Aircraft Performance
M. J. Abzug and E. E. Larrabee: Airplane Stability and Control
M. J. Sidi: Spacecraft Dynamics and Control
J. D. Anderson: A History of Aerodynamics
A. M. Cruise, J. A. Bowles, C. V. Goodall, and T. J. Patrick: Principles
of Space Instrument Design
G. A. Khoury and J. D. Gillett (eds.): Airship Technology
J. Fielding: Introduction to Aircraft Design
J. G. Leishman: Principles of Helicopter Aerodynamics, second edition
J. Katz and A. Plotkin: Low Speed Aerodynamics, second edition
M. J. Abzug and E. E. Larrabee: Airplane Stability and Control: A History
of the Technologies that Made Aviation Possible, second edition
D. H. Hodges and G. A. Pierce: Introduction to Structural Dynamics
and Aeroelasticity
W. Fehse: Automated Rendezvous and Docking of Spacecraft
R. D. Flack: Fundamentals of Jet Propulsion with Applications
E. A. Baskharone: Principles of Turbomachinery in Air-Breathing Engines
Doyle D. Knight: Numerical Methods for Compressible Flows

Elements of Numerical Methods for Compressible Flows DOYLE D. KNIGHT Rutgers – The State University of New Jersey .

no reproduction of any part may take place without the written permission of Cambridge University Press.CAMBRIDGE UNIVERSITY PRESS Cambridge. New York. Melbourne. Cape Town.org Information on this title: www. . ISBN-13 978-0-521-55474-9 hardback ISBN-10 0-521-55474-8 hardback Cambridge University Press has no responsibility for the persistence or accuracy of URLs for external or third-party Internet Web sites referred to in this publication and does not guarantee that any content on such Web sites is. Madrid. accurate or appropriate.cambridge. Subject to statutory exception and to the provisions of relevant collective licensing agreements. First published 2006 Printed in the United States of America A catalog record for this publication is available from the British Library.org/9780521554749 C Cambridge University Press 2006  This publication is in copyright. New York. or will remain. Singapore. São Paulo Cambridge University Press 40 West 20th Street.cambridge. NY 10011-4211 USA www.

and Caitlin .To Kelly. Courtney.

.

9 General Riemann Problem 2.1 Case 1.5 Expansion Fan 2.2 Differential Forms of One-Dimensional Euler Equations 2.2. Shock and Expansion: p1 < p∗ and p4 > p∗ 2.2.Contents List of Illustrations List of Tables Preface page xiii xvii xix 1 Governing Equations 1.4 Vector Notation 1.9.6 Speed of Sound 1.2 Case 2.3 Characteristic Form 2.4 Method of Characteristics 2.1 Conservative Form 2.3 Case 3.1 Introduction 2. Two Shock Waves: p1 < p∗ and p4 < p∗ 2.7 Alternate Forms 1 1 2 3 3 3 4 5 2 One-Dimensional Euler Equations 2.3 Discontinuous Waves 2.7 Shock Formation 2.9.3 Convective Derivative 1.5 Entropy 1.2 Conservation Laws 1. Expansion and Shock: p1 > p∗ and p4 < p∗ ix 9 9 9 10 11 14 17 20 22 25 26 30 32 34 35 36 .6 Domains of Dependence and Influence 2.2.8 Shock Formation from Sinusoidal Disturbance 2.2 Nonconservative Form 2.1 Introduction 1.9.

4.4 Case 4: ΔQi+ 12 ≤ 0.4 Shock Waves and Weak Solutions 3.1 A Simple Flux Quadrature 3. and Stability 3.6 Results 4.5 A Class of Discrete Approximations 3.5.x Contents 2.8.5 Summary 4.1 Introduction 3.8.4.3 No New Extrema 4.8.9 Stability 3.4.11 Conclusion 45 45 45 46 47 48 49 51 53 53 56 59 60 61 61 66 68 69 4 Reconstruction 4.2 The Problem 3.10 Convergence 3.8 Flux Quadrature and Stability 3.1 Introduction 4.2.2 Another Simple Flux Quadrature 3.2.1 Case 1: ΔQi+ 12 ≥ 0.1 A Simple Flux Quadrature 3.1 Algorithm 5.3 Case 3: ΔQi+ 1 ≤ 0.1 Introduction 5.2 Case 2: ΔQi+ 12 ≥ 0.2 Another Simple Flux Quadrature 3.4.6 Accuracy 3.5. Convergence.9. Two Expansions: p1 > p∗ and p4 > p∗ 2.5 Essentially Non-Oscillatory Methods 4. Consistency.4 Case 4. ΔQi− 21 ≤ 0 4.8.4. ΔQi− 21 ≥ 0 4.3 Discretization 3.10 Riemann Shock Tube 37 39 3 Accuracy.9.2 Stability 105 105 105 106 107 .2 Reconstruction Using the Primitive Function 4. ΔQi− 21 ≥ 0 4.4 Modified Upwind Scheme for Conservation Laws 4.4.4 Four Issues 3.2 Godunov’s Method 5. ΔQi− 1 ≤ 0 2 2 4.7 Consistency 3.2 Results 73 73 75 79 81 84 87 89 89 90 92 92 97 99 5 Godunov Methods 5.1 Determination of the Value of a 4.3 Numerical Domain of Dependence 3.9.

3 Van Leer’s Method 6.3.2.3.2 Stability 6.1 Introduction 7.2.2 Accuracy.2 Total Variation 8.2 Explicit Methods 7.1 Algorithm 6.3 Accuracy.3 Accuracy.1 Introduction 8.2 Stability 5.3.1 Algorithm 5.4.3 Accuracy.1 Beam-Warming 7.2 Stability 5.2 Stability 6. Consistency.4 Stability of Selected Methods 7.3.1 Algorithm 8.4. Consistency.1 Introduction 6.3 Accuracy. and Convergence 185 185 187 191 191 209 211 212 213 216 5.3.3 Accuracy.4 Entropy Fix Osher’s Method 5.2. Consistency.1 Algorithm 6.4 MUSCL-Hancock Method 8. and Convergence Roe’s Method 5.2.2.3. Consistency.1 Runge-Kutta 7.3. Consistency.3 Flux Corrected Transport 8.4.4.4.3 5.3 Total Variation 8.3.2 Beam-Warming 170 170 171 171 172 173 175 175 180 8 TVD Methods 8.3.2 Steger and Warming’s Method 6.1 Algorithm 8.4.3 Implicit Methods 7.3.1 Runge-Kutta 7.2 Accuracy. and Convergence 6.1 Algorithm 5. Consistency.4. and Convergence 108 109 110 117 120 122 127 127 139 141 6 Flux Vector Splitting Methods 6.4 .3.Contents xi 5. and Convergence 8. and Convergence 5. and Convergence 147 147 148 148 153 155 157 157 163 164 7 Temporal Quadrature 7. Consistency.

3 Notes Bibliography Index Total Variation 216 224 240 244 .xii Contents 8.4.

16 Pressure for Riemann Shock Tube for p4 /p1 = 2 and T4 /T1 = 1 2. 2.9 Formation of shock wave 2. The figure is drawn for a > u > 0. 2.6 Example of instability associated with a shock wave 3. Vi .4 The three characteristics intersecting xi+ 12 are shown with positive slopes 3.7 The three characteristics emanating from point p.18 Velocity for Riemann Shock Tube for p4 /p1 = 2 and T4 /T1 = 1 2.List of Illustrations 1.5 Left-moving expansion wave 2.27) 3.14 General Riemann Problem: Case 3 (expansion and shock) 2.55) 3.2 2.1 2.4 Right-moving expansion wave 2. Vi+1 3. 2.7 Example of instability with Euler explicit 3.15 General Riemann Problem: Case 4 (two expansions) 2.3 Control volume Curve x(t) Control volume affixed to wavefront The three characteristics intersecting at point p.5 Computed and exact solutions for the flux quadrature (3. p) for γ = 1. The figure is drawn for a > u > 0.8 Convergence study xiii 2 15 18 21 23 25 25 26 27 31 32 33 34 36 37 38 39 40 40 41 46 47 57 58 59 60 66 69 .3 Computed and exact solutions for the flux quadrature (3.12 General Riemann Problem: Case 1 (two shock waves) 2.19 Entropy for Riemann Shock Tube for p4 /p1 = 2 and T4 /T1 = 1 3.1 Grid of control volumes Vi−1 .1 2.11 Riemann function f (p∗ .13 General Riemann Problem: Case 2 (shock and expansion) 2. The figure is drawn for a > u > 0.17 Temperature for Riemann Shock Tube for p4 /p1 = 2 and T4 /T1 = 1 2.4 2.2 Domain of dependence for Qi+ 12 3.10 Initial condition for the General Riemann Problem 2.8 Isocontours of density 2.6 The three characteristics intersecting at point p.

22 4.13 5.2 5.3 4.and second-order reconstructions Initial condition for General Riemann Problem Roe solution to the General Riemann Problem Roe solution to the General Riemann Problem Convergence for Roe’s method using a first-order reconstructions Computed and exact solutions using a first-order reconstruction Computed and exact solutions using first.4 4.26 5.6 4.16 4.5 5.5 4.11 Cells used to define for κ = +1 4.3 5.1 4.xiv List of Illustrations 4.6 5.24 4.14 ΔQ i+ 2 i− 2  4.12 5.4 5. ΔQi− 12 ≤ 0 ΔQi+ 12 ≤ 0.23 4.12 Cells used to define 4.1 5.13 ΔQi+ 12 ≥ 0.10 5.15 Qli+ 1 2 Qri− 1 2 Qli+ 1 2 Qri− 1 2  i+ 1 for Case 1 ΔQ 2 ΔQi+ 12 ≥ 0.17 4. ΔQi− 12 ≥ 0  1 for Case 1  1 and ΔQ 4.8 5. ΔQi− 12 ≤ 0 ΔQi+ 12 ≤ 0.11 5.7 5.25 4.9 4.2 4.and second-order reconstructions Convergence for Roe’s Method using a second-order reconstruction Waves for Riemann Shock Tube Pressure for Riemann Shock Tube for p4 /p1 = 2 and T4 /T1 = 1 Temperature for Riemann Shock Tube for p4 /p1 = 2 and T4 /T1 = 1 86 87 87 88 90 90 92 93 93 94 98 100 100 106 107 109 109 110 111 114 115 121 121 122 123 124 125 125 .9 5.7 Exact function Q(x) = sin x and cell averaged Qi Reconstruction to interior faces of cell i Cells used to define Qi (x) Dependence of Fi− 12 and Fi+ 12 Reconstruction of sine using ten cells Reconstruction using ten cells Cells used to define Qli+ 1 for κ = 13 74 75 78 78 79 80 82 4.15 ΔQi− 1 for Case 1 83 83 86 4.19 4. ΔQi− 12 ≥ 0 Reconstruction of sine using ten cells and κ = 13 with a limiter Reconstruction using ten cells and κ = 13 with a limiter Reconstruction using ten cells and κ = 1 Discontinuous function with ten cells Choices for cells to define Qi (x) Reconstruction of sine using ten cells Reconstruction using ten cells A typical component of Qi at time tn Possible flow structure for Godunov’s Second Method Convergence for Godunov’s Second Method using first-order reconstruction Computed and exact solutions using first-order reconstruction Computed and exact solutions using first.8 Cells used to define Qri− 1 for κ = 82 2 2 1 3 for κ = −1 for κ = −1 83 83 for κ = +1 4.20 4.21 4.10 Cells used to define Cells used to define 2 4.18 4.14 5.

and no limiter 186 Velocity at κao t = 14 for Godunov with a third-order reconstruction and min-mod limiter 186 Computed solution using Van Leer’s Method with a first-order reconstruction 187 Total Variation of ρ.List of Illustrations 5. ρu.4 6.17 5.14 6.and second-order reconstructions 143 Convergence for Osher’s Method using a second-order reconstruction 144 Case 1 150 Case 2 151 Case 3 152 Case 4 152 Convergence for Steger-Warming Method using a first-order reconstruction 156 Computed and exact solutions using a first-order reconstruction 157 Convergence for Steger-Warming Method using a second-order reconstruction 158 Computed and exact solutions using first.12 6.15 7.9 6.20 5.16 5.1 8.and second-order reconstructions 166 Convergence for Van Leer’s Method using a second-order reconstruction 167 Lima¸con of Pascal for a = 0.10 6.6 8.10 8. and ρe 189 189 Velocity at κao t = 14 TV(ρu) for different flux algorithms 190 TV(ρu) for Godunov with a third-order reconstruction with and without reconstruction limiter 190 Class 1 194 Class 2 197 Class 3 198 Class 4 199 Class 5 200 Class 6 201 Class 7 202 .2 8.13 6.8 8.3 6.2 6.1 8.7 6.and second-order reconstructions 159 M + and M − 160 2 + 2 − 162 (γM + 1) and (γM + 1) 163 M [(γ − 1)−1 + 12 M 2 ]+ and M [(γ − 1)−1 + 12 M 2 ]− Convergence for Van Leer’s Method using a first-order reconstruction 165 Computed and exact solutions using a first-order reconstruction 166 Computed and exact solutions using first.5 8. a third-order reconstruction.11 6.13 8.7 8.4 8.3 8.1 6.11 8.22 6.5 6.8 179 Velocity at κao t = 14 using Godunov’s Method.18 5.8 6.6 6.21 5.12 8.9 8.19 5.14 xv Velocity for Riemann Shock Tube for p4 /p1 = 2 and T4 /T1 = 1 126 126 Entropy for Riemann Shock Tube for p4 /p1 = 2 and T4 /T1 = 1 General Riemann Problem: Case 4 (two expansions) 130 Convergence for Osher’s Method using a first-order reconstruction 142 Computed and exact solutions using a first-order reconstruction 142 Computed and exact solutions using first.

ρu.29 8.37 203 203 204 205 206 207 207 208 209 210 211 211 212 213 213 214 214 215 217 217 218 218 219 .18 8.35 8.17 8.32 8.19 8.30 8. and ρe (κao Δt = 0.25 8.31 8.36 8.0125 and 200 cells) 8.025 and 100 cells) Total Variation of ρ.025 and 100 cells) Velocity at κao t = 14 for MUSCL-Hancock (κao Δt = 0.16 8.025 and 100 cells) Velocity at κao t = 14 for MUSCL-Hancock (κao Δt = 0.33 8.025) TV(ρu) for FCT with different flux algorithms (second-order RungeKutta with Δt = 0.025) Computed and exact solutions (κao Δt = 0.34 8.15 8.00125) Velocity at κao t = 14 for FCT with Godunov’s Method and third-order reconstruction (first-order Euler with Δt = 0.28 8.025) TV(ρu) for FCT with different flux algorithms (first-order Euler with Δt = 0.025) Velocity at κao t = 14 for FCT with Godunov’s Method and third-order reconstruction (second-order Runge-Kutta with Δt = 0.24 8.20 8.00125) Convergence for MUSCL-Hancock Computed and exact solutions (κao Δt = 0.00125) TV(ρu) for FCT with different flux algorithms (first-order Euler with κao Δt = 0.23 8.21 8.22 8.025) Velocity at κao t = 14 for FCT with Godunov’s Method and third-order reconstruction (second-order Runge-Kutta with Δt = 0.26 8.27 Class 8 Class 9 Class 10 Class 11 Class 12 Class 13 Class 14 Class 15 Class 16 Convergence for Flux Corrected Transport Computed and exact solutions (κao Δt = 0.xvi List of Illustrations 8.

1 7.2 8.4 5.3 5.1 4.List of Tables 2.1 4.2 4.3 4.1 7.1 5.2 5.4 6.1 One-Dimensional Rankine-Hugoniot Conditions Reconstruction of Q for Computing Fi± 12 Reconstruction Derivative of Reconstructed Function in Cell i Algorithm for Determining Value of a Requirements for Integral Curves Evaluation of Integrals Inadmissible Cases Flux Formulas for Osher’s Method Four Cases for Eigenvalues Implicit Temporal Integration Methods Cases Classes for Flux Corrected Transport xvii 19 75 83 98 99 133 138 138 138 150 173 178 193 .

.

Several noteworthy texts on numerical methods for compressible flows are cited herein. The focus is on the unsteady one-dimensional Euler equations which form the basis for numerical algorithms in compressible fluid mechanics. and even in this regards is not intended to be exhaustive in its treatment. NJ xix . New Brunswick.Preface The purpose of this book is to present the basic elements of numerical methods for compressible flows. The book is restricted to the basic concepts of finite volume methods. I would like to express my appreciation to Florence Padgett and Peter Gordon (Cambridge University Press) and Robert Stengel (Princeton University) for their patience. Any omissions or errors are mine alone.

.

3) where p is the static pressure.1) where T is the static temperature. the fluid is assumed to be thermally perfect. The static enthalpy h is defined as p (1. are constant. p = ρRT (1. 1 . Thus. A. i. First.† The following assumptions are made. the fluid is assumed to be calorically perfect. in Schreier (1982) and White (1974) . with the statements of the four basic physical laws governing such motions.2) h = ei + ρ Second.1 Introduction This chapter presents the governing equations of one-dimensional unsteady flow of a compressible fluid without derivation. either directly or indirectly. for example. respectively..4) † The full equations of three-dimensional compressible viscous flow are presented. and R is the species gas constant.e. Shapiro (1953) 1. the specific heats cv and cp at constant volume and pressure. the internal energy per unit mass ei is ei = cv T (1. R = cp − cv (1. ρ is the density.1 Governing Equations All analyses concerning the motion of compressible fluids must necessarily begin.

. . momentum. and energy are d dt d dt   ρ dV + V  ρudV + V  ρundA = 0 (1.. and the fluid is assumed to be homogeneous (i....... .... uniform molecular composition). ..9) (1.. u .. .. .....1... inviscid unsteady flow in a tube of constant crosssectional area A. ..12) † On the left face n = −1.. ...................... ....6) A    ρu2 + p ndA = 0 (1. ...1. and on the right face n = +1. Control volume The integral conservation equations for mass...2 Governing Equations It follows that h = cp T (1. .11) ∂ρe ∂ (ρe + p) u + =0 ∂t ∂x (1.. ... ... .......5) Third...... .... . 1.. ........ the fluid is assumed to be inviscid. We define a control volume V as shown in Fig..... .. 1. ..10) ∂ρu ∂ρu2 ∂p + =− ∂t ∂x ∂x (1... 1. n .. .. ... ..2 Conservation Laws Consider one-dimensional. x A V Fig. ..... ..... Fourth...... radiation effects and chemical reactions are omitted...... ...e.. . ... .8) dt V A where n is the unit vector in the outward direction† and the total energy e is e = ei + 12 u2 The differential forms of the conservation laws are ∂ρ ∂ρu + =0 ∂t ∂x (1... .. n ..... ..7) A  d ρedV + (ρe + p) undA = 0 (1..

3 Convective Derivative The convective derivative of a function f is defined as Df ∂f ∂f = +u (1.19) 1.13) Dt ∂t ∂x and represents the rate of change of the variable f with respect to time while following a fluid particle.10) to (1. ∂ρ ∂ρu + =0 ∂t ∂x to obtain  (1.14)  ∂p ∂u ∂u ρ = − +u ∂t ∂x ∂x   ∂pu ∂e ∂e = − +u ρ ∂t ∂x ∂x (1.4 Vector Notation The differential form of the conservation laws (1.5 Entropy The change in entropy per unit mass s is T ds = dei − p dρ ρ2 (1.12) may be written in a compact vector notation as ∂Q ∂F + =0 ∂t ∂x where Q= (1.1.16) 1.17) ⎧ ⎫ ⎪ ⎨ ρ ⎪ ⎬ (1.18) ρu ⎪ ⎩ ρe ⎪ ⎭ ⎧ ⎪ ⎨ ⎫ ⎪ ρu ⎬ ρu2 + p F= ⎪ ⎩ ρeu + pu ⎪ ⎭ (1.15) (1. The differential form of the conservation laws for momentum and energy can be rewritten in terms of the convective derivative using the conservation of mass.3 Convective Derivative 3 1.20) .

1953) may be expressed as δQ T dS ≥ (1.e. For an ideal gas.23) Therefore.28) The Mach number is the ratio of the flow velocity to the speed of sound: M= |u| a (1. this implies that dS ≥ 0 for an adiabatic process. 1.6 Speed of Sound The speed of sound is the velocity of propagation of an infinitesimal disturbance in a quiescent fluid and is defined by   ∂p  ∂ρ s a= (1. and T is the static temperature.24) (1. for isentropic flow between two states.22) s − s1 (1.21) Using (1.3)..27) where the partial derivative is taken at constant entropy s. δQ is the heat added to the system.25) The Second Law of Thermodynamics (Shapiro. this equation may be integrated to obtain s − s1 = cv ln T ρ − R ln T1 ρ1 (1.26) where S is the entropy of a system (i. an identifiable mass of fluid). p2 p1 p2 p1   ρ2 γ = ρ1  γ/(γ−1) T2 = T1 (1. two alternate forms may be obtained: T p − R ln T1 p1 p ρ = cv ln − cp ln p1 ρ1 s − s1 = cp ln (1.4 Governing Equations For a thermally perfect gas.29) . a=  γRT (1. In particular.

16).10). the energy and momentum equations may be utilized to obtain an equation for the internal energy ei .15) is the expression on the left side: ∂ρu2 ∂ρu + ∂t ∂x = = = ∂u ∂ρ ∂u ∂ρu +u + ρu +u ∂t ∂t ∂x  ∂u   ∂ρ ∂x∂ρu  ∂u +u + ρ +u ∂t ∂x ∂t ∂x   ∂u ∂u +u ρ ∂t ∂x ρ using (1.31) ∂t ∂x ∂t Alternately.11) and (1.33) It is noted that equations (1.32) ∂t ∂x ∂x Also.15) using (1.33) can be rewritten in terms of the convective derivative using (1.7 Alternate Forms The total enthalpy H is H =e+ p ρ (1.20). A similar derivation applies to (1. .35)  =0 (1.34) (1.11) and (1.30) The conservation of energy may be expressed as an equation for the total enthalpy: ∂ρHu ∂p ∂ρH + = (1.31) to (1. ∂u ∂ρei ∂ρei u + = −p (1. the energy may be rewritten in terms of the entropy. Using (1.36) Exercises 1.7 Alternate Forms 5 1.1 Derive the alternate form of the momentum equation (1. ∂ρs ∂ρsu + =0 ∂t ∂x (1.10). solution The difference between (1.10):  ∂H ∂H +u ρ ∂t ∂x  ∂ei ∂ei +u ρ ∂t ∂x  ρ  = ∂p ∂t  ∂s ∂s +u ∂t ∂x = −p ∂u ∂x (1.1.

Derive a conservation equation for the static enthalpy h.3 Governing Equations Derive the enthalpy equation (1.2 1.12) to yield ∂ρH ∂ρHu ∂p + = ∂t ∂x ∂t using (1.33). Derive the internal energy equation (1. this may also be written as ρ Dh Dt = Dp Dt † The equation is also known as the mechanical energy equation. 2 Subtract from ∂p ∂ (−pu) + u ∂x ∂x ∂p ∂u ∂p −u −p +u ∂x ∂x ∂x ∂u −p ∂x = = = Multiply the mass equation (1.30).15) by u: ρu  ∂u ∂t   ∂ ∂ 1 2 ρ u +u ∂t 2 ∂x +u 1 2 ∂u ∂x u2  =  = ∂p ∂x ∂p −u ∂x −u which represents an equation for the kinetic energy† per unit mass the energy equation (1.4 1.32).6 1. Multiply the mass equation by (see above) to yield ∂ ∂t 1 2  ρu2 + ∂ ∂x 1 2 1 2 u 2 and add to the kinetic energy equation  ρu2 u = −u Subtract from the previous equation to yield ∂ρh ∂ρhu + ∂t ∂x = ∂p ∂p +u ∂t ∂x Using the mass equation.5 −p = ∂u ∂x Derive the entropy equation (1.31).10) by ei and add to the above to yield ∂ρei u ∂ρei + ∂t ∂x 1. ∂p ∂x . solution Multiply the momentum equation (1.16) using (1.9) to yield  ρ ∂ei ∂ei +u ∂t ∂x  1 2 u . solution Add ∂p/∂t to both sides of the energy equation (1.

7 7 In the presence of a body force per unit mass f . cv Dp cp Dρ Ds = − Dt p Dt ρ Dt Thus.  po To  (γ − 1) 2 γ/(γ−1) = p 1+ M 2   (γ − 1) 2 = T 1+ M 2 .23). the total temperature To at a point is defined as the static temperature achieved by bringing a fluid particle at that point to rest adiabatically. homogeneous flow. 1 Dp 1 Ds γ Dρ = + p Dt cv Dt ρ Dt The term on the left is the normalized rate of change of the static pressure following a fluid particle. The second term on the right is the rate of change of the fluid particle density (divided by ρ/γ).8 The total pressure po at a point is defined as the static pressure achieved by bringing a fluid particle at that point to rest isentropically. the momentum and energy equations become ∂p ∂ρu ∂ρu2 + = − + ρf ∂t ∂x ∂x ∂ρe ∂ (ρe + p) u + = ρf u ∂t ∂x The mass equation is unchanged. 1. at locations where the derivatives of the flow variables are not defined).Exercises 1. solution From the entropy equation (1.6 1. Thus.e. For an inviscid. Show that the total enthalpy equation is ∂ρHu ∂p ∂ρH + = + ρf u ∂t ∂x ∂t Derive the following equation: 1 Ds γ Dρ 1 Dp = + p Dt cv Dt ρ Dt and provide a physical interpretation of each of the terms.. Therefore. Similarly.36). s − s1 = cv ln p ρ − cp ln p1 ρ1 Differentiating. an increase in entropy of the fluid particle acts to increase its static pressure. Thus. The first term on the right is the rate of change of entropy (divided by cv ) following a fluid particle. except when the fluid particle crosses a discontinuity (i. the rate of change of entropy following a fluid particle is zero from (1. an increase in fluid particle density acts to increase the static pressure.

solution The left-hand side of the equation is the time rate-of-change of the kinetic energy per unit mass following a fluid particle. the entropy definition (1. the kinetic energy decreases because the pressure gradient decelerates the flow. Provide a physical interpretation of each of the terms.10 Derive an equation for the convective derivative of the Gibbs free energy g = h − Ts . The right-hand side is the work done on the unit mass by the pressure gradient.e. u∂p/∂x > 0. which implies either a) u > 0 and ∂p/∂x > 0 or b) u < 0 and ∂p/∂x < 0).8 Governing Equations Thus. The mechanical energy equation (see Problem 1... u∂p/∂x < 0). the kinetic energy increases because the pressure gradient accelerates the flow.9 where ρo = po /RTo . If the pressure decreases in the direction of the flow (i.3) is u ∂p D  1 2 u =− Dt 2 ρ ∂x Provide a physical explanation. If the pressure increases in the direction of the flow (i. 1.e.25) may be written s − s1 = cp ln To po − R ln To1 po1 Show that ∂p ρ Dpo Ds = − ρTo ρo Dt ∂t Dt 1.

the reader may consult. Friedrichs (1948) 2. but also in many other cases of practical interest. Courant and K. It is therefore essential to understand the development and implementation of these algorithms in their original onedimensional context. The presentation herein is necessarily brief. An understanding of these properties is essential to the development of numerical algorithms. of which three are particularly useful in the development 9 . 2. not only in the flow of compressible fluids.2 One-Dimensional Euler Equations Nature confronts the observer with a wealth of nonlinear wave phenomena. This chapter describes the principal mathematical properties of the onedimensional Euler equations.1 Introduction The remainder of this book focuses on numerical algorithms for the unsteady Euler equations in one dimension. Courant and Friedrichs (1948) and Landau and Lifshitz (1958). Although the practical applications of the one-dimensional Euler equations are certainly limited per se. R.2 Differential Forms of One-Dimensional Euler Equations The one-dimensional Euler equations can be expressed in a variety of differential forms. virtually all numerical algorithms for inviscid compressible flow in two and three dimensions owe their origin to techniques developed in the context of the one-dimensional Euler equations. For further details. for example. O.

2.3) Q2 Q22 Q2 + (γ −1) Q3 − 12 2 Q1 Q1 ⎪ ⎪ ⎪ ⎪ Q Q Q ⎪ 2 3 2 ⎪ + (γ −1) Q3 − ⎪ ⎪ ⎪ Q1 ⎩ Q1 . t) is ⎧ ⎫ ⎪ ⎨ F1 ⎪ ⎬ ⎧ ⎪ ⎨ ⎫ ⎪ ρu ⎬ F2 = ρuu + p F= ⎪ ⎪ ⎪ ⎪ ⎩ F3 ⎭ ⎩ ρeu + pu ⎭ The static pressure p is obtained from p = (γ −1) ρe − 12 ρu2 (2. flow solutions may exhibit discontinuities that require separate treatment.10 One-Dimensional Euler Equations of numerical algorithms. The flux vector F(x. t) is the vector of dependent variables. However.4) The flux vector F is a function of Q: F= ⎧ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎨ (2. as will be discussed later in Section 2.1 Conservative Form The one-dimensional Euler equations in conservative differential form are ∂Q ∂F + =0 ∂t ∂x (2. and e is the total energy per unit mass.3. These forms are applicable where the flow variables are continuously differentiable.2) ρu ⎪ ⎩ ρe ⎪ ⎭ where ρ is the density.1) where Q(x.2. Q= ⎧ ⎫ ⎪ ⎨ Q1 ⎪ ⎬ Q 2 ⎪ ⎩ Q ⎪ ⎭ 3 ⎧ ⎫ ⎪ ⎨ ρ ⎪ ⎬ = (2. u is the velocity component in the x-direction.

2 1 Q2 2Q 1 ⎫ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎬ .

 L ∂Q ∂F dx = 0 + ∂t ∂x 0 .1) over this region. momentum. ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎭ (2.5) The term conservative form arises from the observation that in a finite domain the mass. Integrating (2. and energy are strictly conserved in this formulation. Consider a region 0 ≤ x ≤ L.

shock waves and contact surfaces) must be accurately simulated.† to an arbitrary precision depending on the fineness of the spatial and temporal meshes. t) − F(L.2 Differential Forms of One-Dimensional Euler Equations 11 Thus ∂ ∂t L 0 Qdx = F(0. showed that numerical algorithms that are both conservative and convergent can approximate the exact solution.6) where A is the Jacobian matrix defined by A= ∂F ∂Q (2.e.2.1). of course.2 Nonconservative Form Equation (2.3.7) and given by A= ⎧ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎨ 0 (γ −3) 2  Q2 Q1 2 ⎪   ⎪ ⎪ Q2 3 Q2 Q3 ⎪ ⎪ ⎪ + (γ −1) −γ ⎪ ⎩ Q21 Q1 1 (3−γ) (γ −1) ⎫ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎬ Q2 γ Q1 ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎭ 0 Q2 Q1  Q2 Q3 3 γ − 2 (γ −1) Q1 Q1 2 (2. in their study of conservation laws..1) can also be written as ∂Q ∂Q +A =0 ∂t ∂x (2. Lax and Wendroff (1960). This is the correct statement of the physics. at all points x in a finite domain except for a finite number of discontinuities where the flow variables satisfy the jump conditions discussed in Section 2.g. including discontinuities.2. 2. . Many numerical methods are patterned after the conservative form (2.1) almost everywhere.8) † The exact solution satisfies (2. The principal motivation is the requirement that flows with discontinuities (e.. t) which indicates that the net increase in Q per unit time within the region is due to the net flux into the region. i.

H = cp T + 12 u2 the Jacobian can be written in terms of u and H. using the definition of the total enthalpy per unit mass H. A= ⎧ ⎪ ⎪ ⎪ ⎪ ⎨ 0 (γ −3) 2 u 2 ⎪ ⎪ ⎪ ⎪ ⎩ −γeu + (γ −1)u3 1 0 (3−γ)u (γ −1) γe − 32 (γ −1)u2 γu ⎫ ⎪ ⎪ ⎪ ⎪ ⎬ (2.12) Equation (2.6) is hyperbolic since all of the eigenvalues of A are real (Garabedian. It is straightforward to show that its eigenvalues are λ1 = u λ2 = u + a λ3 = u − a (2.12 One-Dimensional Euler Equations Using the definition of Q.11) There are a number of important properties of the Jacobian matrix A. 1 r2 = u+a ⎪ ⎩ H + ua ⎫ ⎪ ⎬ ⎪ ⎭ ⎧ ⎪ ⎨ .9) ⎪ ⎪ ⎪ ⎪ ⎭ Alternately.10) Also. The corresponding right eigenvectors are ⎧ ⎪ ⎨ 1 r1 = u ⎪ ⎩ 1 u2 2 ⎫ ⎪ ⎬ ⎪ ⎭ ⎧ ⎪ ⎨ . 1 r3 = u−a ⎪ ⎩ H − ua ⎫ ⎪ ⎬ ⎪ ⎭ (2.13) . A= ⎧ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎨ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎩ 0 (γ −3) 2 u 2 (γ −1) 3 u − Hu 2 ⎫ ⎪ ⎪ ⎪ ⎪ ⎪ ⎬ (γ −1) ⎪ 1 0 (3−γ)u H − (γ −1)u2 γu ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎭ (2. 1964). the Jacobian may be written in terms of u and e. A= ⎧ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎨ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎩ 0 (γ −3) 2 u 2 ua2 (γ −2) 3 − + u (γ −1) 2 ⎫ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ (γ −1) ⎪ ⎬ 1 0 (3−γ)u 2 a (3−2γ) 2 + u (γ −1) 2 γu ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎭ (2. using e= a2 + 1 u2 γ(γ − 1) 2 the Jacobian may be written in terms of u and a.

⎧ ⎪ ⎨ λ1 Λ= 0 ⎪ ⎩ 0 ⎫ 0 0 ⎪ ⎬ λ2 0 ⎪ 0 λ3 ⎭ (2. This is true since A has three linearly independent eigenvectors (Franklin. ⎧ ⎪ ⎨ 1 1 1 T = u u+a u−a ⎪ ⎩ 1 u2 H + ua H − ua 2 ⎫ ⎪ ⎬ (2. 1968). − − .16) The matrix‡ T is the concatenation of the right eigenvectors of A. ‡ The matrix T is not to be confused with the static temperature.15) where Λ is a diagonal matrix of eigenvalues of A. Denote ⎧ ⎫ ⎪ ⎨ rk1 ⎪ ⎬ rk = for k = 1.17) ⎪ ⎭ The inverse T −1 is the concatenation of the left eigenvectors of A. 3 (2.2.14) The Jacobian matrix A can be expressed as† A = T ΛT −1 (2. (γ − 1) 2 . 2 2 4 a 2a 2 a 2a 2a2 (γ − 1) u2 (γ − 1) u u 1 (γ − 1) + . .2 Differential Forms of One-Dimensional Euler Equations and the corresponding left eigenvectors are  l1 =  l2 =  l3 = (γ − 1) u2 u (γ − 1) 1− .19) rk2 ⎪ ⎩ r ⎪ ⎭ k3 † Equation (2. T −1 ⎧ (γ − 1) u2 ⎪ ⎪ ⎪ 1 − ⎪ ⎪ 2 a2 ⎪ ⎪ ⎪ ⎪ ⎨ (γ − 1) u2 1u − = 2 ⎪ 4 a 2a ⎪ ⎪ ⎪ ⎪ (γ − 1) u2 ⎪ 1u ⎪ ⎪ ⎪ + ⎩ 2 4 a 2a u (γ − 1) 2 a (γ − 1) u − + 2 a2 (γ − 1) u − − 2 a2 1 2a 1 2a (γ − 1) − a2 (γ − 1) 2a2 (γ − 1) 2a2 ⎫ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎬ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎭ (2.18) Two relationships between the left and right eigenvectors can be derived. − 2 a2 a a2 13  u 1 (γ − 1) (γ − 1) u2 (γ − 1) u − .15) indicates that A is similar to the diagonal matrix Λ. − + . 2. 2 2 4 a 2a 2 a 2a 2a2   (2.

22) The second relationship between the left and right eigenvectors can be found from the identity T T −1 = I. Consider the scalar equation ∂f ∂f +c =0 (2.23) k=1 It can also be shown by direct substitution (Exercise 2.5) that F = AQ (2. t) is a scalar function and c is a positive constant.20) Then it may be directly verified† 3  rik ljk = δij (2. 2. 2. the identity T T −1 = I yields 3  rki lkj = δij (2.25) is f (x.3 Characteristic Form The one-dimensional Euler equations may be written in a form that exhibits their wavelike character.21) k=1 where δij is the Kronecker delta defined by  δij = 1 if i = j 0 otherwise (2. 0) = g(x).2.24) This is known as Euler’s Identity or the homogeneity property and is the basis of some of the Flux Vector Splitting methods that will be discussed in Chapter 6. t) = g(x − ct) (2. the solution to (2.25) ∂t ∂x where f (x. 3 (2. 1966).21) follows directly from the fact that the eigenvalues of A are distinct (Isaacson and Keller.14 One-Dimensional Euler Equations and lk =  lk1 lk2 lk3  for k = 1. Since T is the concatenation of the eigenvectors rk (taken as column vectors) and T −1 is the concatenation of the eigenvectors lk (taken as row vectors). .26) † The relationship (2. For an unbounded domain −∞ < x < ∞ and initial condition f (x. It is first useful to describe wavelike behavior.

. df =0 dt on dx =c dt (2.. . .. ... .. t) along a curve x(t) as shown in Fig... .31) where the energy equation has been replaced by the equation for entropy ..1..2 Differential Forms of One-Dimensional Euler Equations 15 This represents a wave traveling to the right with velocity c without a change of form.27)....... 2... . t) = + dt dt ∂x dt ∂t (2..... . ....... . .1.. ......25) can also be discerned by considering the behavior of f (x.. x(t) . .. . 2.. The one-dimensional Euler equations can be recast in a form similar to (2. . . .. we then have df d ∂f dx(t) ∂f = f (x. . .29) ∂p ∂x (2... .. The equations are ∂ρ ∂ρu + ∂t ∂x ∂ρu ∂ρu2 + ∂t ∂x ∂s ∂s +u ∂t ∂x = 0 = − = 0 (2...... . . .... .. .. ..26).... ... .. The curves x − ct = constant are known as the characteristics of (2.. The wavelike character of (2. Curve x(t) Denoting the derivative of f with respect to t along the curve x(t) by df /dt. ..30) (2. . .. . x Fig..28) which indicates that the value of f is unchanged on the curve dx/dt = c or x − ct = constant as seen in (2.25)..... .27) where ∂f /∂x denotes the partial derivative of f with respect to x holding t constant. and likewise ∂f /∂t indicates the partial derivative of f with respect to t holding x constant. . .... .. t ...... .25) and (2..27). ..... ... ... .. Comparing (2. .. .2.....

 p s − s1 = cv log p1 Then ∂s ∂s cv +u = ∂t ∂x p    ρ − cp log ρ1  cp ∂p ∂p − +u ∂t ∂x ρ and thus ∂p γp ∂p +u − ∂t ∂x ρ   ∂ρ ∂ρ +u ∂t ∂x  ∂ρ ∂ρ +u ∂t ∂x   =0 From the conservation of mass.32) dt dt which implies that the entropy is conserved following a fluid particle. the conservation of momentum is rewritten as ∂u ∂u ∂p ρ + ρu =− ∂t ∂x ∂x and multiplying this equation by a and adding (2. and thus ρ and a are functions of p alone. ∂u ∂p ∂p +u + a2 ρ =0 (2. We seek a transformation of the remaining equations to a form similar to (2.36) assuming inviscid. from (2.33) ∂t ∂x ∂x Using the conservation of mass. It is evident that the entropy equation is already in the form of (2.34) Assuming that the entropy at some initial time is uniform. t) (2.28) and can be written as ds dx(t) = 0 on = u(x(t). Hence.32) the entropy remains uniform. (2.27). ∂ρ ∂ρ ∂u +u = −ρ ∂t ∂x ∂x and since a2 = γp/ρ. From the definition of entropy (1.23). nonheat conducting flow.  ∂u ∂p ∂u ∂p + (u + a) + ρa + (u + a) ∂t ∂x ∂t ∂x  or 1 dp du + =0 ρa dt dt dx =u+a dt on (2.34) may be rewritten as d dt   dp +u =0 ρa .16 One-Dimensional Euler Equations (1.33).

where δV has width dx and height dy.e. d dt   2 a + u = 0 on (γ −1) Similarly.2. The conservation of mass (1. the solutions for discontinuous waves (Section 2. it is possible to derive d dt   2 a − u = 0 on (γ −1) dx =u+a dt dx =u−a dt The unsteady one-dimensional Euler equations may therefore be rewritten as   dx d 2 = 0 on a+u =u+a dt (γ −1) dt   dx d 2 a−u =u−a = 0 on dt (γ −1) dt ds dx = 0 on =u (2. i..3). However. . Relative to an inertial reference frame. respectively. and dx/dt = u are the characteristics. The control volume is affixed to the wavefront and is therefore moving at velocity uw relative to the reference frame.6) applied to δV yields† ∂ ρ¯ dx dy + ρ2 (u2 −uw ) dy − ρ1 (u1 −uw ) dy = 0 ∂t † Strictly. Consider a wavefront with velocity uw (t). The curves dx/dt = u + a. Equations (2.e.35) imply that the Riemann invariants are convected without change along their respective characteristics..2. As shown in Fig. Note that. Equations (2.3 Discontinuous Waves The one-dimensional unsteady Euler equations admit waves wherein one or more flow variables are discontinuous across the wavefront. u and a are functions of x and t.3 Discontinuous Waves 17 Since p/p1 = (ρ/ρ1 )γ and a/a1 = (p/p1 )(γ−1)/2γ . to the lowest order. the fluid has velocities u1 and u2 to the left and right of the wavefront. the Euler equations admit wavelike solutions. i. 2. a small control volume δV spans the wavefront. dx/dt = u − a. 2.35) dt dt These are known as Riemann invariants.35) permit multiple-valued solutions that are unphysical and hence must be supplemented with additional constraints. in general.

ρ1 (u1 −uw ) − ρ2 (u2 −uw ) = 0 or [ρ (u−uw )]w = 0 (2..36) to (2..... dx ...38) The conservation of energy likewise yields [ρe (u−uw ) + pu]w = 0 (2... u2 ...... .2.. Dividing by dy and taking the limit dx → 0... ... . They admit two different types of wavelike solutions...39) are the one-dimensional Rankine-Hugoniot conditions and are summarized in Table 2....18 One-Dimensional Euler Equations . . u1 dy uw ..39) Equations (2....37) where we define where f1 and f2 represent the value of the function f on either side of the wave surface.. . Fig.36) [f ]w = f1 − f2 (2.. . .. 2..40) † A third type of discontinuous solution (vortex sheet) is possible in two and three dimensions... Control volume affixed to wavefront where ρ¯ = 12 (ρ1 + ρ2 ).....7) yields ∂ρu dx dy + (ρ2 u2 (u2 −uw ) + p2 ) dy − (ρ1 u1 (u1 −uw ) + p1 ) dy = 0 ∂t and hence taking dx → 0 yields [ρu (u−uw ) + p]w = 0 (2......1. Note that f1 is not necessarily equal to f2 .† The first is a contact surface defined by u1 = u2 = uw (contact surface) (2. ....... The conservation of x-momentum (1.

From (2. where the parameter may be chosen to be the shock pressure ratio† σl = p2 /p1 that cannot be less than one. it is possible that ρ1 = ρ2 and ρ1 e1 = ρ2 e2 .. It can be shown that this implies σl ≥ 1 (Liepmann and Roshko 1957).39) admit a one-parameter family of solutions † The subscript l denotes a wave moving to the left. Therefore. ‡ The Second Law of Thermodynamics requires the entropy change to be nonnegative as the flow crosses the discontinuity.41) There are two cases.39) is also satisfied.3 Discontinuous Waves 19 This solution satisfies (2. Equation (2.36).44) (γ − 1) (γ + 1) + σl 2γ 2γ 1/2 (2. Note that there is no condition imposed on [ρ]w or [ρe]w .42) (2.36) to (2. One-Dimensional Rankine-Hugoniot Conditions Equation Condition Mass Momentum Energy [ρ (u−uw )]w = 0 [ρu (u−uw ) + p]w = 0 [ρe (u−uw ) + pu]w = 0 The second is a shock wave for which u1 = u2 = uw (shock wave) (2. Table 2.36) to (2. . p1 = p2 . the wave speed is negative relative to a frame of reference traveling at u1 ).44) can be rewritten as u2 = u1 − a1 (σl − 1)  (γ+1) γ σ + (γ−1) 2γ l (2.38).e.46) 2γ In the second case. u1−uw < 0. which corresponds to a wave moving to the left (i.‡ The result is1 ρ2 ρ1 T2 T1 = = u2 = (γ − 1) + (γ + 1)σl (γ + 1) + (γ − 1)σl ρ1 σl ρ2    ρ1 ρ2 u1 + uw −1 ρ2 ρ1  uw = u1 − a1 (2. with the latter implying that T1 = T2 . Equations (2.45) Note that Equation (2. Equations (2.1. which corresponds to a wave moving to the right. In the first case. corresponding to the sign of u1 − uw .2. u1 −uw > 0.39) admit a one-parameter family of solutions for given u1 and a1 .43) (2.

4 Method of Characteristics The characteristic form (2. we consider an unbounded domain −∞ < x < ∞.53) (2.47) (2.49) 1/2 (2.54) For simplicity. Initial continuous profiles of u(x. where the parameter may be chosen to be the shock pressure ratio σr = p1 /p2 . which cannot be less than one: ρ1 ρ2 T1 T2 = = u1 = (γ − 1) + (γ + 1)σr (γ + 1) + (γ − 1)σr ρ2 σr ρ1    ρ2 ρ1 u2 + uw −1 ρ1 ρ2  uw = u2 + a2 (γ − 1) (γ + 1) + σr 2γ 2γ (2. The initial entropy is uniform and denoted by so .51) 2.20 One-Dimensional Euler Equations for given u2 and a2 .52) (2. t) are specified as u(x. t) and a(x. 0) = ao (x) Consider a point† p at an infinitesimal time dt > 0 as shown in Fig.35) of the one-dimensional Euler equations offers a direct method for solution. .3. o) = uo (x) a(x.49) can be rewritten as u1 = u2 + (σr − 1) a2  γ (γ+1) σ + (γ−1) r 2γ 2γ (2. The equations are   2 d a+u = 0 dt (γ −1)   d 2 = 0 a−u dt (γ −1) ds = 0 dt dx =u+a dt dx on =u−a dt dx on =u dt on (2. 2.50) Note that Equation (2.48) (2. There are three characteristics that emanate from t = 0 and intersect p: dx dt = u 1 + a1 † The point p is not to be confused with the static pressure.

... .. ... ... . .. . .. .. . . . ..52) and (2... .. .. . ... . .. ... .. .. p dt .. .. ... . ... . ...... (2...... . .59) . ... . . . .4 Method of Characteristics dx dt dx dt 21 = u 2 − a2 = u3 where the slope of each characteristic is evaluated at its origin at t = 0. . ... . . sp = s o (2. .. . .. .. . .. .... . ..... . . . . . .. .. . . ..... ...53). The three characteristics intersecting at point p.. . .. . .. ...... . .. Therefore. .. .....3. . . 2 x Fig. . . . .. . . .... . . .. ... . .. .. .. . 2... .. .. . 0 3 1 . .. .. 2 ap + up = (γ − 1) 2 ap − up = (γ − 1) 2 a1 + u1 (γ − 1) 2 a2 − u2 (γ − 1) which may be solved to yield (γ − 1) (u1 − u2 ) 4 ap = 1 2 up = 1 (a1 − a2 ) + (γ − 1) (a1 + a2 ) + 1 2 (u1 + u2 ) (2...... . . .. ....54). .... .. The figure is drawn for a > u > 0.55) (2.. . .. . . . .. ... From (2. ..2.... . .  pp ρp where the subscript r  ap 2γ/(γ−1) = pr ar  2/(γ−1) ap = ρr ar indicates a reference value.. .. ... .... t . ...57) and thus the flow remains isentropic.58) (2.. ... .... .56) From (2.. .....

pressures and temperatures.55) to (2.59) cannot be used solely in all cases because they do not guarantee that there will be only three characteristics intersecting point p. a modification is required which is the introduction of a discontinuous wave (i. 2 2 aa + ua = ab + ub (γ − 1) (γ − 1) (2.e. Figure 2. i.53).62) Since points c and d are in the uniform region 1. shock) when necessary. This process could presumably then be repeated and thus the solution could be obtained in principle.5 Expansion Fan The Method of Characteristics can be used to determine the flow solution within a simple wave joining two regions of uniform entropy at different velocities. Since this is physically nonsensible. Indeed.. 2. However. In fact.e. This issue will be discussed in further detail in Section 2. Consider points a and b on a positive characteristic dx/dt = u + a located within the expansion fan. i. and likewise a negative characteristic from point d will intersect point a..4 illustrates the first type of simple wave bounded by the positive characteristics dx/dt = u1 + a1 and dx/dt = u2 + a2 bordering two regions of uniform flow denoted by subscripts 1 and 2.22 One-Dimensional Euler Equations It would appear that equations (2.e.60) From (2. these equations could be employed to determine the solution on a discrete set of points p. The region between these two positive characteristics is denoted by an expansion fan. multiplevalued) solutions are mathematically possible in certain regions depending on the initial conditions and in the absence of any additional constraints. nonunique (i. 2 aa − ua = (γ − 1) 2 ab − ub = (γ − 1) 2 ad − ud (γ − 1) 2 ac − uc (γ − 1) (2.63) ..52). they do not guarantee the uniqueness of the solution. Equations (2. this is not the case.61) (2. 2 2 aa − ua = ab − ub (γ − 1) (γ − 1) (2.59) provide an algorithm for the complete solution to the flowfield at time dt.e.. From (2. A negative characteristic dx/dt = u − a drawn from a point c will intersect point b.55) to (2.7.

the equation can be integrated to obtain   (γ − 1) (γ + 1) ua − u 2 + a2 t x= 2 2 and thus  u = a =  2 x (γ − 1) + u2 − a2 (γ + 1) t 2   (γ − 1) x 2 − u2 + a2 (γ + 1) t (γ + 1) (2.. ...... 2 2 aa − ua = a2 − u2 (γ − 1) (γ − 1) and thus (γ − 1) (ua − u2 ) + a2 2 Therefore.. . . . ... .. ... the characteristic inside the expansion fan is aa = dx (γ − 1) (γ + 1) = ua − u2 + a2 dt 2 2 Since ua is constant on the characteristic..... ...... . ........... ... . ..... .......... ... ... . .... ...... .5 Expansion Fan 23 t. .. ..... . ....61).. ............. . . .. . 2 .... .. ..... from (2............. ................ . ....... . . . ... ....... .. ... .. .............. ua = u b aa = ab and thus every characteristic dx/dt = u + a within the expansion fan is a straight line..... .. 1 ............. . ..60) and (2.......... ... ... ..... .. . ............ ........ . 2. ............ Right-moving expansion wave From (2. . ... .. .... .. .... .. . ... .. . . ........ .... .... .... ......... . . ............... ....65) ....... . . ... . ... ..... .......... ... ............................. ... ..... ........ .. ..... . . .. .... ....... . . .. dx/dt = u + a1 . 1 .. . .... ....... ...... .... . .... . . ... .. . ...... . .......... .....4... ........ .. .... . .. ....... ... .. ...... .... ....... ........... ............. . .. .. ........ .. dx/dt = u2 + a2 u a b r a r r c r d u a x Fig..................2... .. .. . .... ............... ....... ...... . 2 ..... .. .. . .. .63).............. .................. 1 ....... ................. . ........ .. ... ........ ............ .64) (2..... .............. ......... . .. .... .... .... ... ........ .... ....... . ..... ........ ............. . ....... Now.......

by extending point a into region 1 and using Equation (2.66). although Fig. The remaining flow variables can be obtained from the isentropic relations p p1 T T1 ρ ρ1   a 2γ/(γ−1) a1  2 a = a1  2/(γ−1) a = a1 = (2. An analogous conclusion holds for Fig.5 illustrates the second type of simple wave bounded by the negative characteristics dx/dt = u1 − a1 and dx/dt = u2 − a2 bordering two regions of uniform flow denoted by subscripts 1 and 2.24 One-Dimensional Euler Equations which represents the solution for u and a within the expansion fan.68) Figure 2. In addition. this becomes 2a2 u1 = u2 + (γ − 1)  p1 p2 (γ−1)/2γ  −1 (2.66). the flow conditions in regions 1 and 2 are related by u 1 = u2 + 2 (a1 − a2 ) (γ − 1) (2.70) The remaining flow variables can be obtained from the isentropic relations in Equations (2. the flow conditions in regions 1 and 2 are related by 2 u 2 = u1 + (2.4 shows that both characteristics dx/dt = u1 + a1 and dx/dt = u2 + a2 move to the right. this becomes 2a1 u2 = u1 − (γ − 1)  p2 p1 (γ−1)/2γ  −1 (2.66) In addition.5. this need not be the case.69) (2. 2. Equation (2. . An analysis similar to the first type yields the following solution within the expansion fan:  u = a =  x (γ − 1) 2 + u 1 + a1 (γ + 1) t 2   x 2 (γ − 1) − + u1 + a1 (γ + 1) t (γ + 1) (2. 2.71) (a1 − a2 ) (γ − 1) Using the isentropic relations in Equation (2.66).67) Using the isentropic relations.72) It should be noted that.61).

.... ...... .e.. .6 Domains of Dependence and Influence 25 t... ... . .... .. .. ... ... . .. . . .. . ...... ...... .. ...6.... .... ..... . . . .. ..... ..2.. . .... ........ ... .. ...... ..... .... ... ..... x2 ....... ... .. . . ... . .. ....... ... ............ ... ........ ... . . . x3 . .. . . x2 ..... ....... . . . . . Similarly. ....... .. .. .. ..... u a b r u1 a1 ra c r d r . ....6. . .. ....... .. . . which are the intersection of the three characteristics emanating from a single point p at t = 0 ... .. .. ... . .. ... dx/dt = u2 − a2 dx/dt = u1 − a1 . ... . ....... ..... ... . .... x3 ) This region is known as the domain of dependence of the solution at point p..... . .. .. ... .... ..... .... .... .. ............ ........ ....... . ............ .. .......... ...... . . .. ............ ... ... . ... .. ....... ....... . t . . .. .. ..... ... . ... 2. .... .. ....... .. .... . . ....5....... . ... . ..... ...... . .... i... . . .... . 2.. x2 . . x3 ) ≤ x ≤ max(x1 .. ..... it is possible to define the points x1 .. ... . .. .. . .. ..... . . .. x Fig... ..... ... ... ... .. ... . ... ...... ...... .... . ... . ...... ..... ..... . 1 3 2 domain of dependence .... . .. ... .. . ....... .... .... ... 2. . .. 2 .. . ... .. . .... .... Left-moving expansion wave 2... .. . .. . . ... ... . . .. .... ... .. . ..... ...... ... . ... . .... .. ..55) to (2. .... ...... .... . .. ...... . . . ..... ... .. .... . .. ... dt 0 p ... ...... ..... . The figure is drawn for a > u > 0... The three characteristics intersecting at point p... . ....... . . ....... ... .58) that the flow at p depends only on a limited range of data at t = 0 in x.. .... . .... ... .. x Fig. . ...... ..... . ... . ... . .... .. .. . . . . . . ... .. .......... ... ..... .. .............. 2 . . ...... . .. .. .. .. .. ... ..... ..... . . .... . .. . .. .. .. .... ..... . as illustrated in Fig... .. . .6 Domains of Dependence and Influence It is evident from (2. ... ... ... ..... . ... ..... .... . the region min(x1 .... ... . .... . . . ... ..... ... .. .. . .... .. .. . .. .

. .......... A change of flow conditions at p at t = 0 will only influence the flow within this region at t = dt.... . .. ...... ........... .. x3 ) at t = dt defines the domain of influence of point p.. .....74) .... Specifically. ... ... . inviscid flow with no shocks..... t .... . ....... The figure is drawn for a > u > 0. . . ... .... ..... ....... .... . i. .. .. ... the formal application of the Method of Characteristics (Section 2...... The region min(x1 ........ ...7 Shock Formation Depending on the initial conditions... ...7. ... .4) can yield nonunique solutions. p x Fig. x2 ...... .. .. .26 One-Dimensional Euler Equations with t = dt... ... . and therefore one or more of the assumptions employed in deriving the characteristic equations (2. The entropy s is assumed uniform at t = 0. dt 2 3 1 ... ..... . ..... .7.e. shock wave) appears. .... This is physically impossible. .. ......35) must be violated..... . .. .... ....... ... ........... . The conditions for the formation of a shock wave and its position in space and time can be determined... the assumption that the flowfield is continuous and differentiable is violated at discrete locations where a discontinuous solution of the Euler equations (i.... ............... ... .. Consider an adiabatic.. . 2........ . .. 2.. .. .... ...... ... ....... . .. .... ..... x2 . . . We follow the derivation of Landau and Lifshitz (1959)..... The Euler equations are ∂ρ ∂ρu + ∂t ∂x ∂ρu ∂ρu2 + ∂t ∂x = 0 = − (2... more than three characteristics intersect at some point p at some instant in time...... . . . ..... ... .. ........... .. . ... ............ ... ... .... domain of influence 0 . . ........ .. .e. .73) ∂p ∂x (2.. x3 ) ≤ x ≤ max(x1 .. ... . . as illustrated in Fig. 2.. The three characteristics emanating from point p...

. .. respectively..7 Shock Formation ∂s ∂s +u ∂t ∂x 27 = 0 (2... . . . .. ∇ρ .. ....... .. ... .... . . ... ... ...77) ... ...8. .... . where ex and et are unit vectors in the x and t directions. .... ..75) From the entropy equation.... .8).. . .... . The vector δxex + δtet is constructed to be parallel to the isocontour.. .. ... 2........ . ... ....... By definition. Therefore.. ......76) t .. .... ... .... Isocontours of density Consider the isocontours of ρ in the x − t plane (Fig...... ρ = constant δt δx 0 .... s remains constant for t > 0 until the possible formation of any shock waves.. ....2. . .... . the gradient ∇ρ = ∂ρ ∂ρ ex + et ∂x ∂t is orthogonal to the isocontour...... .. . x Fig. .. .. . . The conservation of mass becomes ∂ρ dρu ∂ρ + =0 ∂t dρ ∂x and therefore  ∂ρ ∂t ∂ρ ∂x −1 =− dρu du (2.. . the inner product of these vectors is zero:   ∂ρ ∂ρ ex + et · (δxex + δtet ) = 0 ∂x ∂t from which ∂ρ ∂t  ∂ρ ∂x −1  =− ∂x  δx =− δt ∂t ρ (2... ...... We seek finite amplitude waves wherein the velocity u is a function of density ρ.... 2... .. .

 dρu ∂x  = ∂t ρ dρ (2.  ∂ρ/∂x = du dρ −1 ∂u ∂x and the momentum equation can be rewritten as ∂u ∂t  ∂u ∂x −1  a2 =− u+ ρ  du dρ −1  (2. Since u is assumed to be a function of ρ.28 One-Dimensional Euler Equations From (2. ∂u ∂u 1 ∂p +u + =0 ∂t ∂x ρ ∂x Now   ∂p  ∂ρ ∂p  ∂s ∂p = +  ∂x ∂ρ s ∂x ∂s ρ ∂x The second term is zero.77). and using (1.82) Since the flow is isentropic.78) and (2. where a is the speed of sound.76) and (2. using the conservation of mass. using (1.77).24).79).78) From the conservation of momentum.  ∂u ∂x −1  ∂x  =− ∂t u  (2.28).27) the first term is a2 ∂ρ/∂x. dρu a2 =u+ dρ ρ  du dρ −1 and thus du a =± dρ ρ Therefore u=± (2.79) By an argument similar to that leading to (2. and (1. ∂u ∂t and since u is a function of ρ.81) a dρ ρ (2. 2 u=± (γ − 1) da .80)  ∂x  ∂x  =  ∂t u ∂t ρ From (2.25). (1.

85). it is evident that the characteristics are straight lines whose slopes depend on the value of u at the x intercept at t = 0.86) (γ + 1) ∂x  = ±ao + u  ∂t u 2 (2. .86).87) and using (2. Equations (2. x = f (u).87).88) where f (u) is the constant of integration. Consider a right traveling wave. From (2.85) where ρo and po are the density and pressure at the location where u = 0.83) and (2.84). and (2. For each characteristic.2. Thus.83). From (2.80).88) yield two solutions corresponding to waves traveling to the left (−) and right (+).79).84) 2γ/(γ−1) (2. and (2. Therefore  ∂x  =u±a ∂t u    (γ + 1) x = ±ao + u t + f (u) 2 (2.83) and therefore  ρ = ρo 1 ±  p = po 1 ± (γ − 1) u 2 ao (γ − 1) u 2 ao 2/(γ−1) (2. each point on the initial wave travels to the right with a velocity that † This can always be assured by means of a Galilean transformation. From (2. the slope of a characteristic is  ∂x  (γ + 1) = ao + u  ∂t u 2 (2.† Then u=± 2 (a − ao ) (γ − 1) where ao is the speed of sound at the location where u = 0.89) Thus. (2. the value of u is a constant.81).7 Shock Formation 29 We assume that there is some point within the gas at t = 0 where the velocity is zero. The formation of a shock wave depends on the initial conditions. Note that at t = 0. (2. a = ao ± (γ − 1) u 2 (2. (2.

2.8 Shock Formation from Sinusoidal Disturbance The following specific example will be employed in later chapters for evaluating various numerical algorithms. Since this is physically unfeasible. This is shown† schematically in Fig. sin −1  u ao  (2.92) with κ = 2π/λ.1.91) holds and the time ts .25). 2.88). ts = − 2 df (γ + 1) du and d2 f =0 du2 (2.88).30 One-Dimensional Euler Equations is a monotonically increasing function of u. The initial condition for the speed of sound a corresponding to a wave traveling to the right is (γ − 1) u (2. if u decreases with increasing x over a portion of the initial wave. a shock wave must occur at ts which satisfies the shock conditions in Table 2.93) a = ao + 2 The initial conditions for the remaining flow variables are obtained from (1. Now −1 f (u) = κ † Only one period is shown for clarity. At this point. At the instant ts of the formation of the shock wave. Thus. where λ is the wavelength and is a dimensionless coefficient. there will occur a crossing of the characteristics at some time ts . The location of the shock formation is defined by (2.90) From (2. 0) = ao sin κx (2. The initial condition for u is chosen to be u(x.24) and (1.  ∂x  = 0 and ∂u t  ∂ 2 x   =0 ∂u2 t (2. implying a multiple-valued solution for t > ts . It is evident therefore that the formation of a shock(s) requires one or more inflection points us in the initial velocity distribution where d2 f /du2 = 0 and a negative slope df /du at u = us .9 for an initial periodic profile for u.91) which defines the value of u at which (2. the slope ∂u/∂x becomes infinite.94) .

. .......... . ......... . .......... . ........ .... (b) At time of shock formation (t = ts ) u . ......... ........... ....... . ....... ......... .9......... ........... . .. ....... ... .. ... .. .............. . ...... ... . . .... . ...... . . . ... ........ .. . ........... ............... d2 f = 0 for du2 u=0 (2. . ... .... ... . ... .. .. ............. ......... Formation of shock wave Taking into account the sign... . ... .. ... .... .. ... . .. ... ....... .... .. .... .... . .. . . ⎧  −1/2 ⎪ ⎨ −κ−1 ( ao )2 − u2 df =  −1/2 du ⎪ ⎩ κ−1 ( ao )2 − u2 for π 2 for − π 2 ≤ κx ≤ ≤ κx ≤ 3π 2 π 2 (2.... ... ... ........ ... .. ..... . .. . .. ..... . . x . ............. . ........... ... ...... ... . . . ....... . . .. . ...... .. .... ...... .. ........ .. ... ... .. ....... . .. .. .. ... .. ....2... ... ... . ....8 Shock Formation from Sinusoidal Disturbance 31 u ... .. ... . . . .....95)..... .....96) .... .... . .. .. ... . ... .. x (a) Initial condition (t = 0) u ....... ... .... .... ...... .. ... . .... ..... ............ . .. .. ...... ........ .. . .. .. .......... ... . ..... ... . .. . .... ... . 2.. ...... . . (c) After shock formation (t > ts ) Fig. ..95) From (2. .... ...... ... . .. .. ... . .... ... x . ... ......... . ............. ... ...... . .... ... .......... .. ...

comprised of left. . .. a contact surface separates the fluid that was initially on either side of the interface at t = 0. static pressure p1 ..... . ‡ A fifth solution.. 2.... ... . . Since the initial condition is periodic. .. . a countably infinite number of shocks initially appear at ts at locations ao ts + nλ for n = 0. .and right-moving expansion waves and two contact surfaces with a vacuum in between (Gottlieb and Groth..15. .. x Fig. the shock forms a distance ao ts to the right of the point where u = 0 in the initial condition. Initial condition for the General Riemann Problem The specific solution is determined by the contact surface pressure p∗ ..12 to 2.. the velocity u1 . The solution of the General Riemann Problem can be constructed from the individual solutions for a propagating normal shock (Section 2. At time t = 0...91). . 1988) is physically implausible. u1 p1 T1 . and static temperature T4 are likewise specified..9 General Riemann Problem The General Riemann Problem. namely.5). static pressure p4 . .. and static temperature T1 are given... . In each case.. which are illustrated in Figs.. ... 2) one shock wave and one expansion wave. . 1) two shock waves. the velocity u4 . 3) one expansion wave and one shock wave. ... . . Depending on the left and right states.† illustrated in Fig.. two different states of the flow exist and are separated by a contact surface at x = 0..97) From (2.. ts = 2 (γ + 1) κao (2...88).10. and 4) two expansion waves. .. .. The solution may be easily extended to allow for different γ on either side of the contact surface. † We assume identical values of the ratio of specific heats γ. there are four‡ possible solutions for t > 0.. 2.. is a useful paradigm for the development and testing of numerical algorithms for the one-dimensional Euler equations. For x < 0. 2.32 One-Dimensional Euler Equations and from (2. . For x > 0. . ..10. diaphragm u4 p4 T4 . 2. ±1. .3) and an expansion wave (Section 2.

2.9 General Riemann Problem

33

which is defined by the transcendental equation§
a1 f (p∗, p1 ) + a4 f (p∗ , p4 ) − u1 + u4 = 0
where

f (p , p) =



1
γ


2
(γ−1)


γ+1 p∗
γ−1 −1/2
+
2γ p
2γ

(γ−1)/2γ
p

p∗
p − 

1

−1

p

(2.98)

for p∗ ≥ p

(2.99)

for p∗ ≤ p


and a = γRT is the speed of sound. The behavior of f (p∗ , p) for γ = 1.4
is shown in Fig. 2.11.
f (p∗, p)
3

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

2
1
0
−1
−2
−3
−4
−5

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

1

2

3

4

p∗ /p

5

Fig. 2.11. Riemann function f (p∗ , p) for γ = 1.4

The four solutions are defined below and shown in Figs. 2.12 to 2.15. For
each solution, there are four regions, each of which is separated by a simple
wave (e.g., shock wave, contact surface, or expansion wave). Although the
simple waves are shown in Figs. 2.12 to 2.15 with a specific direction (e.g.,
the left shock in Fig. 2.12 moves to the left), the waves may travel in either
direction†.
§ The solution can be obtained using Newton’s method. The initial guess may be taken to be
(Gottlieb and Groth, 1988)
p∗ =

(ρ1 a1 p4 + ρ4 a4 p1 + ρ1 ρ4 a1 a4 (u1 − u4 ))
(ρ1 a1 + ρ4 a4 )

Note that if p∗ < p1 and p∗ < p4 (Case 4), the equation may be solved directly:
p∗ =  

(a1 + a4 ) +

(γ − 1)
(u1 − u4 )
2

−(γ−1)/2γ

a1 p 1

† Of course, the simple waves cannot cross each other.

−(γ−1)/2γ

+ a4 p4 

−1 2γ/(γ−1)

34

One-Dimensional Euler Equations

2.9.1 Case 1. Two Shock Waves: p1 < p∗ and p4 < p∗
The velocities of the left and right shock waves (csl ) and (csr ) and contact
surface (cc ) are2 

csl

= u1 − a1 

csr

= u4 + a4

cc = u1 −

a1
γ

(γ + 1)

(γ + 1)
2γ    

p∗
−1 +1
p1

(2.100)

p∗
−1 +1
p4

(2.101)  

p∗
−1
p1

(γ + 1) p∗ (γ − 1)
+
2γ p1
2γ 

−1/2

(2.102)

The flow variables in Region 2 (csl < x/t < cc ) are
u2 = cc
p2 = p

(2.103)

∗ 

ρ2 = ρ1 (γ − 1) + (γ + 1) 

p∗

p1

(γ + 1) + (γ − 1) 

p∗ −1

p1

(2.104)
(2.105)

The static temperature T2 can be found from the ideal gas equation. The
flow variables in Region 3 (cc < x/t < csr ) are
u3 = cc
p3 = p

(2.106)

∗ 

ρ3 = ρ4 (γ − 1) + (γ + 1) 

p∗

p4

(γ + 1) + (γ − 1) 

p∗ −1

p4

t
.....
.. ...

shock
..

contact
surface

shock

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

2

1

3

4

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

x

Fig. 2.12. General Riemann Problem: Case 1 (two shock waves)

(2.107)
(2.108)

2.9 General Riemann Problem

35

2.9.2 Case 2. Shock and Expansion: p1 < p∗ and p4 > p∗
The velocities of the left shock (cs ), the left and right boundaries of the right
expansion fan (cl and cr ), and the contact surface (cc ) are3 

cs = u1 − a1 

cl = u4 + a4
cr = u4 + a4
cc = u1 −

a1
γ

(γ + 1)

(γ + 1)
(γ − 1) 

p∗
−1
p1   

p∗
−1 +1
p1

p∗
p4  

(γ−1)/2γ

(2.109)

2

γ−1 

(γ + 1) p∗ (γ − 1)
+
2γ p1

(2.110) 
−1/2

(2.111)
(2.112)

The flow variables in Region 2 are
u2 = cc
p2 = p

(2.113)

∗ 

ρ2 = ρ1 (γ − 1) + (γ + 1) 

p∗

p1

(γ + 1) + (γ − 1) 

p∗ −1

p1

(2.114)
(2.115)

and the flow variables in Region 3 are
u3 = cc
p3 = p

(2.116)

∗ 

ρ3 = ρ 4

(2.117)

∗ 1/γ

p
p4

(2.118)

Within the expansion fan cl ≤ x/t ≤ cr , 

u =

x (γ − 1)
2
+
u 4 − a4
γ+1 t
2 

p = p4

(γ − 1) 1
(γ + 1) a4 

ρ = ρ4

p
p4 

1/γ   

x
2
− u4 +
t
(γ + 1)

(2.119) 
2γ/(γ−1)

(2.120)
(2.121)

36

One-Dimensional Euler Equations
t
contact
surface
.

shock
..

.
.....
... ..

expansion fan

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

2

3

1

4

x

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

Fig. 2.13. General Riemann Problem: Case 2 (shock and expansion)

2.9.3 Case 3. Expansion and Shock: p1 > p∗ and p4 < p∗
The velocities of the right shock (cs ), the left and right boundaries of the
left expansion fan (cl and cr ), and the contact surface (cc ) are4 

cs = u4 + a4
cl = u1 − a1
cr = u1 + a1
cc = u4 +

a4
γ 

(γ + 1)
2γ  

p∗
−1 +1
p4

2
(γ + 1)

γ − 1 (γ − 1) 

p∗
−1
p4  

∗ (γ−1)/2γ

p
p1

(2.122)
(2.123) 

(γ + 1) p∗ (γ − 1)
+
2γ p4

(2.124) 
−1/2

(2.125)

The flow variables in Region 2 are
u2 = cc
p2 = p

(2.126)

∗ 

ρ2 = ρ 1

(2.127)

∗ 1/γ

p
p1

(2.128)

and the flow variables in Region 3 are
u3 = cc
p3 = p

(2.129)

∗ 

ρ3 = ρ4 (γ − 1) + (γ + 1) 

p∗

p4

(γ + 1) + (γ − 1)

Within the expansion fan cl ≤ x/t ≤ cr , 

u =

x (γ − 1)
2
+
u 1 + a1
γ+1 t
2 

p∗ −1

p4

(2.130)
(2.131) 

(2.132)



p = p1

2.9 General Riemann Problem  

x
(γ − 1) 1
u1 −
(γ + 1) a1
t 

ρ = ρ1

p
p1

+ 

1/γ

37

2
(γ + 1) 

2γ/(γ−1)

(2.133)
(2.134)

t
.
......
.. ..

expansion fan

contact
surface

shock

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

2

3

1

4

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

x

Fig. 2.14. General Riemann Problem: Case 3 (expansion and shock)

2.9.4 Case 4. Two Expansions: p1 > p∗ and p4 > p∗
The velocities of the left and right boundaries of the left expansion fan (cl1
and cr1 ), the right expansion fan (cl2 and cr2 ), and the contact surface (cc )
are5
cl1

= u1 − a1

cr1

= u1 + a1  

cl2

= u4 + a4

cr2

= u4 + a4

cc = u1 +

2
(γ + 1)

γ − 1 (γ − 1)
(γ + 1)
(γ − 1)  

p∗
p4   

∗ (γ−1)/2γ

p
p1 

(γ−1)/2γ

(2.136)
2

γ−1 

p∗ (γ−1)/2γ

2a1
1−
(γ − 1)
p1

(2.135)  

(2.137)
(2.138)
(2.139)

The flow variables in Region 2 are
u2 = cc
p2 = p

(2.140)

ρ2 = ρ 1  

p∗ 1/γ

p1

(2.141)
(2.142)

.. .. .............. ............. .. ..... .. ..... 1 contact surface expansion fan no..... . .... .....................  u = x (γ − 1) 2 + u 1 + a1 γ+1 t 2  p = p1  x (γ − 1) 1 u1 − (γ + 1) a1 t  ρ = ρ1 p p1  +  (2.... 2 1 3 4 ... .. .. ............ .......... . ..... ..... ....... .. .... .. ....... ... ..... ......... .. . . ... .. . ............ ...... .. ........ . . ... ........ .. ......... .... ..... . ..... ............. ...... .... expansion fan no.......... ...... .... ......... ..... . . ... . ....... .... ... ........ ... .. ... .. .... .... . .......... ..150) 1 γ (2.. .... . .......... .. . . ............. 2 .. ... ....... . ............ .. .. .151) t ... ....149)  x 2 − u4 + t (γ + 1) 2γ/(γ−1) (2....... . . ... ......... ..... . ... . .... . .... .143)  (2. ..... . . .. .....146) 2 (γ + 1) 2γ/(γ−1) (2.................  u = x (γ − 1) 2 + u 4 − a4 γ+1 t 2  p = p4 (γ − 1) 1 (γ + 1) a4  ρ = ρ4 p p4   (2... . . .......... .. .. ..... x Fig............ . .... . .144) ∗ 1/γ p p4 (2.... ............. . ......... ......... ...........38 One-Dimensional Euler Equations and the flow variables in Region 3 are u3 = cc p 3 = p∗ ρ3 = ρ 4 (2... ........ .. ............. . 2.. .. .. ............. ... .. . . ....148) and within the right expansion fan cl2 ≤ x/t ≤ cr2 .. .... ... ... .. ....... .. General Riemann Problem: Case 4 (two expansions) .. ........... .. .... . . . ... ......15. ... ... ........ .147) 1 γ (2.... ..... .. ....... . . .... ....... ...... ... ............... ... .. ..... ..145) Within the left expansion fan cl1 ≤ x/t ≤ cr1 . ... ..... .. . .. . . . .

....... ......... ....... contact surface ..... Equations (2.. .............. ..... without loss of generality..................... 2.....99) can then be rewritten as (Liepmann and Roshko.............. ...... ...... ........... ......... ..................... .......... shock 0 −2 −1 0 1 2 .......... We may assume. . that p4 > p1 and hence the configuration corresponds to Case 2....... ...... .......................... ... .................... 2 1 expansion fan .................. .....................121)........................ ....... p/p1 .........112) as cs = − a1 γ  p∗ −1 p1  (γ + 1) p∗ (γ − 1) + 2γ p1 2γ −1/2 The flow variables in Regions 2 and 3 and the expansion fan are given by Equations (2..............109) as  cs = −a1 (γ + 1) 2γ   p∗ −1 +1 p1 The velocity of the left and right boundaries of the expansion fan are given by (2. . 1957) p4 p∗ = p1 p1  (γ − 1)(a1 /a4 )(p∗ /p1 − 1) 1− √  2γ 2γ + (γ + 1)(p∗ /p1 − 1) −2γ/(γ−1) from which the shock pressure ratio p2 /p1 = p∗ /p1 can be found by iteration. ........................... ....... ..................111) as  cl = a4 (γ + 1) (γ − 1)  p∗ p4 (γ−1)/2γ 2 − (γ − 1)  cr = a4 where p∗ /p4 = (p∗ /p1 )/(p4 /p1 )............10 Riemann Shock Tube The Riemann Shock Tube Problem is a special case of the General Riemann Problem wherein the initial velocities u1 and u4 are zero..............10 Riemann Shock Tube 39 2.... ....... ........ .....110) and (2. .... x/a1 t Fig...................... . .... .....113) to (2..98) and (2................ . The velocity of the contact surface is obtained from (2........ ....................2........ Pressure for Riemann Shock Tube for p4 /p1 = 2 and T4 /T1 = 1 . ...... The shock speed is obtained from (2.. .......................... .16...

.......... ........ ... . ..... ... . ..... ....19 for the initial conditions p4 /p1 = 2 and T4 /T1 = 1 with γ = 1..... −0.... ..... ......... ..... Velocity for Riemann Shock Tube for p4 /p1 = 2 and T4 /T1 = 1 The static pressure and temperature....... ... ..2 ....... ..... ... .. ...... At the contact surface.... . . ........ ..... .. ....... ............................. . 2. ............. .... ...... ............. and entropy are shown in Figs... . .........3 Fig........ .. ............ . ..... 2.... ...... ... ...... ................ velocity....... ........ ... ..................... . ......702978 and cr /a1 = 1.. ..... .......... . ........ ........ . .... ...... † This is an example of a conical flow wherein there is no physical length scale........ ............0 .. ..9 .......1 0....... ...17.............. 2...... The pressure... Within the expansion fan........................ ... .................. ........ .... the pressure and velocity are continuous while the temperature (and hence the density) and entropy change discontinuously............ .... . .. .. ..... ............ ....... ........18...... ....... . The expansion fan moves to the right with left and right velocities cl /a1 = 0......... The abscissa is x/t normalized by a1 . . ........... ....... .. ...... ....... ... ...... .....0 0....... . .. ... . ..... .............. ... shock 1............................ ... .... temperature..† A shock moves to the left at velocity cs /a1 = −1..... contact surface expansion fan 0...... . velocity........ .... ..........................16 to 2... ....1 1.. ... .... Temperature for Riemann Shock Tube for p4 /p1 = 2 and T4 /T1 = 1 u/a1 ............... ... ................................................ and entropy change discontinuously across the shock...... 0.................. .. ...... ... ......... .... ........ ............ .. ........... ..... ...... ................ .... .... ............. ............. ............ .......................... .. ......4.......................1 −0...... ....... ......... .... . ... whereby the solution at any time t can be obtained.2 −1 0 1 2 ........ .... .. .......40 One-Dimensional Euler Equations T /T1 1... ..0... x/a1 t shock contact surface expansion fan −0.. .159479......... ............... all flow variables change continuously......8 −2 −1 0 1 2 .. ...... ........ ............... .. x/a1 t Fig.

. .. .. expansion fan contact surface −0.... .... . Entropy for Riemann Shock Tube for p4 /p1 = 2 and T4 /T1 = 1 Exercises 2.......................... . .... ......... ....... ...... ....3 2 = Q2 Q1 Q2 Q1 3 2 Derive the eigenvalues of A. ..... . ........... ................ −0...... ........ 2. ............ x/a1 t 0...... .... ........................... .. ........12) are distinct (Franklin..8).. .......... ..Exercises 41 (s−s1 )/cv . ..... ........................ ..... ..... 1968)....... . solution The Jacobian is obtained by differentiation of the flux vector F = (F1 ..3 Fig............... ............. ................... F2 ...... F3 )T using (2........ Let A be an n × n matrix........1 −0......... ............ ....19.... .. . .................................... .... ....... . .... .. solution A formal procedure for determining the eigenvectors can be defined using the property that the eigenvalues (2.... . .. shock . .... .. ......... ......... .... . . ........ ...... .... .... ................... ............... ..... Let .... . Denoting the element in the ith row and j th column of A by Aij .......2 ......0 .. −1 0.....5)............... .. we obtain A11 = A12 = A13 = A21 = A22 = A23 = A31 = ∂F1 ∂Q1 ∂F1 ∂Q2 ∂F1 ∂Q3 ∂F2 ∂Q1 ∂F2 ∂Q2 ∂F2 ∂Q2 =0 =1 =0 (γ − 3) Q2 2 Q1 Q2 = (3 − γ) Q1 = (γ −1) ∂F3 Q2 Q3 = −γ + (γ −1) ∂Q1 Q21 ∂F3 Q3 3 =γ − (γ −1) ∂Q2 Q1 2 ∂F3 Q2 = =γ ∂Q3 Q1 A32 = A33 2. ..... ... ......... ............... ... ......... . .. . .....1 Derive the Jacobian matrix (2..... .. Derive the right eigenvectors of A...1 0 1 2 ......... ...........2 2....

solution Denoting F = AQ and using (2. Then (E2.2) 1 u2 . 1968). solution From the definition of entropy. 12 u2 )T . . . The remaining right eigenvectors are obtained in a similar manner. . in agreement with (2.4 2. 2 The solution is e2 = u and e3 = Thus. F1 = Q2 F2 = (γ −1)Q3 − F3 = γQ2 Q3 − Q1 (γ −3) Q22 2 Q1 (γ −1) Q32 2 Q21 which is equal or equivalent to (2.5 Derive the left eigenvectors of A Prove Euler’s Identity (2.5). show that s2 − s1 > 0 implies σl > 1. . . . At least one of the Aˆq is nonsingular (Franklin. q−1. . An eigenvector e = (e1 . s2 − s1 = cv log p 2 p1 − cp log ρ 2 ρ1 Denote σ = p2 /p1 . en )T can be found by setting the q th element of e equal to 1 and solving the nonsingular system of equations n  (Aij − λδij ) ej = −Aiq for i = 1. .15) is the concatenation of the right eigenvectors of A. s2 − s1 > 0 implies that the argument of log must be greater than one.1) yields (A22 − λ) e2 + A23 e3 A32 e2 + (A33 − λ) e3 = = −A21 −A31 (E2.42). .8). and the matrix T −1 is the concatenation of the left eigenvectors of A? Hint: Are the right and left eigenvectors unique? For the case u1−uw > 0 in Section 2. . 2. . From (2. we may take the 1/γ root of the argument and therefore  (γ + 1) + (γ − 1)σ  (γ − 1) + (γ + 1)σ σ1/γ > 1 . for q = 1. . .24). q+1. n. .6 2. . (γ − 1) + (γ + 1)σ ρ2 = ρ1 (γ + 1) + (γ − 1)σ Thus s2 − s1 = cv log  (γ + 1) + (γ − 1)σ (γ − 1) + (γ + 1)σ γ  σ Therefore.42 One-Dimensional Euler Equations Aˆq be the (n−1) × (n−1) matrix formed by crossing out row q and column q of A. u. . the eigenvector corresponding to λ1 = u is r1 = (1.7 Is it true in general that the matrix T in a general similarity form (2.1) j=1 j=q Consider the eigenvalue λ = u.13). n (E2.3. Let q = 1. Since σ > 0 and γ > 1. 2.

Thus. u2 = u1 + 2 a1 (γ − 1) and hence the left expansion fan is the region u1 − a1 ≤ x 2 ≤ u1 + a1 t (γ − 1) . and therefore a2 = 0. For ν > 1. d2 f /dν 2 > 0. 2) −a < u < 0. Thus. since this would require a change in sign of d2 f /dν 2 . Then the above condition becomes f (ν) > 1 where f (ν) = ν (γ+1) − (γ + 1) γ (γ + 1) ν + ν (γ − 1) (γ − 1) Therefore df dν d2 f dν 2 d3 f dν 3  γ 1 ν (γ−1) + (γ − 1) (γ − 1) = (γ + 1) ν γ − = γ(γ + 1)ν (γ−2) (ν − 1) = γ(γ + 1)ν (γ−3) [(γ − 1)ν − (γ − 2)]  Thus. f < 1 for ν < 1. Thus. 2.6 and 2. a Taylor’s series expansion for f (ν) about ν = 1 yields f (ν) = 1 + γ(γ + 1) (ν − 1)3 + O((ν − 1)4 ) 6 The Taylor series indicates that there exists a sufficiently small neighborhood of ν = 1. f > 1 requires ν > 1 and hence s2 − s1 > 0 requires σl > 1.8 2. By a similar argument. where f > 1 if ν > 1 and f < 1 if ν < 1.7 for the other three possible cases: 1) u > a. from above. and 3) u < −a. 2. solution The configuration is illustrated in the figure below. By assumption.9 Redraw Figs. Determine the solution to the General Riemann Problem corresponding to a vacuum and two expansion fans. there is a vacuum adjacent to the right boundary of the left expansion fan. The solution in the left expansion fan is  x (γ − 1) + u1 + a1 t 2  u = 2 (γ + 1) a = (γ − 1) x 2 a1 u1 − + (γ + 1) t (γ + 1) p p1 = a a1   2γ/(γ−1) which is valid in the region u1 − a1 ≤ x/t ≤ u2 − a2 .Exercises 43 Let ν = σ1/γ . it is not possible for f to become less than one for ν > 1.

........ . .. ... 2......... ..... ....... . ...... . . . . .......... . . 2 .. ......... 3 .. .. .. .... .. .. . .. ...... ...... . . ... . ..... . .... ... ..... .. ..... . Thus. 1 contact surface ... .... . . ...... ..... ..... ..... ... . ...... u4 − u1 > 2 (a1 + a4 ) (γ − 1) t expansion fan no.. .. ....... .... ........... ... ...... ... ...... ..... ..44 One-Dimensional Euler Equations Similarly. . ... .. ........ . ... ...... .. .. ....... .... ...... .. 2 .... .. . ......... .... By assumption.. The solution exists provided that the contact surfaces are separated.. .. ... .. 2 u3 = u4 − a4 (γ − 1) and hence the left expansion fan is the region u4 − 2 x a4 ≤ ≤ u4 + a4 (γ − 1) t The contact surfaces coincide with the inner boundaries of the expansion fans.... ........... .. ........ .............10 .... ........ ........... .. .......... there is a vacuum adjacent to the right boundary of the left expansion fan.. 1 4 ............. .. ...... ..... . .......... .......... . . from above... .......... ...... ........ . ......... contact surface expansion fan no. . .. .. . ........... .. . ....... ............... . ...... ... ..... .. . . . .. . ... ..... ....... ... . . . ...... .. ...... .. .................. . . ......... .. . ............ .... and therefore a3 = 0..... ......... .. ... . ..... ..... .. vacuum ....... .. .. ..... . ... . . ..... x Solve the General Riemann Problem using the linearized Euler equations ∂Q ∂Q + A¯ =0 ∂t ∂x where ⎧ ⎪ ⎨ ⎫ ⎪ 0 1 0 ⎬ A¯ = (γ −1)¯ u2 /2 (3−γ)¯ u (γ −1) ⎪ ⎪ ⎩ −H ¯ −(γ −1)¯ ¯u u2 γu ¯ ⎭ ¯ +(γ −1)¯ u3 /2 H where u ¯ = ¯ = H u1 + u4 2 H1 + H4 2 .... ........... ...... .. . .. . ... ... ... ... ............. ............ . .. . .... .... ... .. ................. ... .......... ... ... ........ ..... ... .... .... .... ... . .... ... the solution in the right expansion fan is u = 2 (γ + 1) a = (γ − 1) (γ + 1) p p4 = a a4   (γ − 1) x + u4 − a4 t 2   2 x − u4 + a4 t (γ + 1) 2γ/(γ−1) which is valid in the region u3 + a3 ≤ x/t ≤ u4 + a4 .. ... ....

Q= ⎧ ⎫ ⎪ ⎨ Q1 ⎪ ⎬ Q 2 ⎪ ⎩ Q ⎪ ⎭ 3 45 ⎧ ⎫ ⎪ ⎨ ρ ⎪ ⎬ = ρu ⎪ ⎩ ρe ⎪ ⎭ (3. Practical discretization methods are presented in subsequent chapters. and stability in the framework of the one-dimensional unsteady Euler equations. consistency.1) where Q is the vector of dependent variables. Convergence. and Stability It is perfectly obvious that a revolution of some sort became due in the field of numerical methods with the advent of modern computing machines.2 The Problem The fundamental problem is to define an accurate algorithm for integrating the one-dimensional Euler equations in control volume form: d dt  V Q dxdy +  ∂V Fdy = 0 (3. convergence. A simple discretization is derived for the purposes of illustration.1 Introduction The chapter presents the concepts of accuracy. Robert Richtmeyer (1967) 3. Consistency.2) . 3.3 Accuracy.

3) 3. .. which is some function f of a set of the volume-averaged variables Qi in the neighborhood of xi+ 12 . 3. . are denoted by xi+ 12 for i = 0. i = 1.3 Discretization The conservation equations (3. . located midway between the adjacent centroids. where tn+1 = tn + Δtn (3. Vi . Convergence. .7) . M . Qi+ 1 = f (Qi−m .1) are applied to a discrete set of control volumes Vi . .46 Accuracy. .5) where Vi = ΔxΔy and Δy is the (constant) height of each cell. . . In the discretization. The spatial (flux) quadrature involves faces i + 12 and i − 12 . . Vi+1 Assume a discretization of time into discrete levels tn . Qi−m+1 . Qi+n ) 2 (3..1. The cell faces. . . The flux vector F depends on Q. denote the volume-averaged vector of dependent variables by Qi (t) = 1 Vi  Q dxdy Vi (3. Consistency.. . x xi+ 12 Fig.4) For volume i.. i−1 i i+1 xi− 12 . . . M as illustrated in Fig. Qi+n−1 . Grid of control volumes Vi−1 ..6) 2 where Ai+ 12 = Δy is the surface area of the face at xi+ 12 . . Qi . the dependence on Q can be replaced by an assumed dependence on Qi+ 12 .. and Stability and F is the vector of fluxes ⎧ ⎫ ⎪ ⎨ F1 ⎪ ⎬ ⎧ ⎪ ⎨ ⎫ ⎪ ρu ⎬ = F= F2 ρuu + p ⎪ ⎪ ⎪ ⎪ ⎩ F3 ⎭ ⎩ ρeu + pu ⎭ (3. . namely. 3. .. n = 1. .. ... .1.. Assume a uniform discretization of the x-axis into M cells of length Δx with centroids xi . Denote Fi+ 12 = 1 Ai+ 12  xi+ 1 Fdy (3.. The solution requires specification of a set of control volumes and algorithms for the temporal and spatial quadratures.

10) The problem therefore reduces to defining the temporal and spatial quadrature algorithms for evaluating the second term on the right side of (3.. 3....... Holt 1984)...... .. .. M at time tn .. ... dQi + dt Fi+ 12 − Fi− 12 Δx =0 (3... The third .. Toro.. . ... .... .10)..... .. Then the Euler equations (3..... ... . i+n Fig. ........ ..... 1967.. This approach is also known as the Method of Lines (Hirsch.. . . . Hoffman. ... Morton and Mayers. .. .... i+1 i i−1 xi− 12 xi+ 12 . ...... the solution at time tn+1 can be obtained by integration: Qn+1 = Qni − i 1 Δx  tn+1 tn Fi+ 12 − Fi− 12 dt (3.. . .. ..8) Since Vi is assumed to be independent of time. i−m ......10). . . .. . . Domain of dependence for Qi+ 12 as illustrated in Fig.. Given the solution Qi for i = 1. ..4 Four Issues In developing algorithms for the solution of (3. .. . . ... . ... .... which is the fidelity of the numerical simulation in representing the actual solution of the governing partial differential equations. 1997). 1989. 1988. . Fletcher.. . ..1) are transformed into a system of ordinary differential equations... which is the quantitative measure of the agreement between the numerical simulation and the exact solution on a given grid. ...... . . 3. ..9) This is the semi-discrete method by which the Euler partial differential equations (2. ............ .. 1988.. . ..4 Four Issues 47 Qi+ 12 ..... The first is accuracy. ..3.2.2. there are four issues which must be addressed (Richtmeyer and Morton... 3. Anderson et al.. . ........ . . 1994. .. .1) become dQi Vi + Fi+ 12 Δy − Fi− 12 Δy = 0 dt (3. 1984... The second is consistency. ..

. we consider a two-level . stability. we determine the necessary conditions for the consistency of this class of discrete approximations. which is the property that the numerical solution converges to a solution when the spatial and temporal grid spacings are reduced to arbitrarily small values. i.10) for the purposes of examining accuracy. we consider the accuracy of this specific class of discrete approximations. Additionally.6). consistency. which represents the absence of temporally unbounded oscillations of a nonphysical nature in the simulation. we present a class of discrete approximations to (3. and Stability is stability. Consistency.9) viewed as an ordinary differential equation. We show that the spatial flux quadrature algorithm must properly represent the physics of the flow to prevent the occurrence of unphysical oscillations that grow unbounded in time.48 Accuracy. we consider the issue of convergence. Second. and convergence. Convergence. according to (3. Finally. specific problems associated with the formation of shock waves are identified. Fi+ 12 = 1 Δy  Fdy xi+ 1 2 and thus Fi+ 1 = F (Qi+ 1 ) = F(Qi+ 1 ) 2 2 2 The class is limited in scope for the purposes of discussing the issues of accuracy. 3. and convergence. stability. First. consistency. Third.5 A Class of Discrete Approximations We consider a class of discrete approximations to (3. and extensions are presented in later chapters.6) on the stability of (3.10): = Qni − Qn+1 i 1 Δx  tn+1 tn Fi+ 12 − Fi− 12 dt where. Fourth. These four issues will be made more precise in subsequent sections. without discretization of the time derivative. First. we examine the effect of the spatial flux quadrature algorithm (3. We assess the numerical stability of the class of discrete approximations introduced earlier.e. The approximation proceeds in two steps. The fourth is convergence.

6 Accuracy 49 approximation to the time integral.  .3.

Qi . . . we replace Qi+ 12 in (3. The discrete approximation to (3. . Second.10) is therefore = Qni − Qn+1 i −  Δt  . . . . . tn+1 tn Fi+ 12 dt ⇒ θF (Qn+1 ) + (1 − θ)F (Qni+ 1 ) Δt i+ 1 2 2 (3. Qi+n−1 .12) where Qi+ 1 is defined according to (3. This is the generalized trapezoidal method.11) by Qi+ 12 ⇒ Qi+ 12 (3. Qi−m+1 .11) where θ is a constant and ⇒ indicates that the integral is replaced by the discrete expression on the right. .7) as 2 Qi+ 12 = f (Qi−m .8). Qi+n ) and the specific form of the function f is yet to be determined (see Section 3.

n+1 θ F (Qn+1 1 ) − F (Qi− 1 ) i+ 2 2 Δx .

6 Accuracy In this section. .. . We  i exact anticipate† that E can be expanded in a Taylor series. . Δt) = aΔx + bΔt + cΔx2 + dΔxΔt + eΔt2 +   O(Δxi Δt3−i ) i=0.7. E(Δx.3 + .15) where the coefficients a.10) by (3.. e are functions of x and t. we determine the accuracy of the specific class of discrete approximations (3...10) and E depends on Δx and Δt. Δx) (3.13):   Qn+1  i     discrete = Qn+1  i exact + E(Δt. (3. We define the error E in approximating (3.14) where Qn+1 is defined by (3.13) 3.. ‡ More precisely.  Δt  (1 − θ) F (Qni+ 1 ) − F (Qni− 1 ) 2 2 Δx (3. The notation for the remainder O(Δxi Δt3−i ) indicates that the term is proportional‡ to † See Section 3.13). .. f (x) = O(g(x)) as x → xo if lim x→xo f (x) =α g(x) .

Consistency. and Stability ΔxiΔt3−i . Convergence. We can assume that . The leading terms of E are O(Δx) and O(Δt) since E must vanish as Δx and Δt approach zero for any reasonable discretization.13) by an expression involving Q. The first step is to replace Q in (3.50 Accuracy.

7).16) xi+ 1 2 where R1 and R2 depend on the function f in (3. Using a Taylor expansion† F (Qi+ 12 ) =   F (Q(x) + ΔxR1 (x) + Δx2 R2 (x) + O(Δx3 )) = F (Q(x)) + ∂Q xi+ 1 2 ∂F . Qi+ 21 = Q(x) + ΔxR1 (x) + Δx2 R2 (x) + O(Δx3 ) (3.

13). The second step is to expand the flux difference in (3.13): F (Qni+ 1 ) − F (Qni− 1 ) = F (Qni+ 1 ) − F (Qni− 1 ) 2 2 2 ∂ + Δx2 ∂x  2  ∂F R1 + O(Δx3 ) ∂Q (3.17) into the discrete approximation (3. ΔxR1 (x) + Δx2 R2 (x) + O(Δx3 ) 2 +O(Δx ) where the right side is evaluated at xi+ 12 . which yields Qn+1 = Qni − i − −  Δt  .17) The third step is to substitute (3.

n+1 θ F (Qn+1 ) − F (Q ) 1 1 i+ 2 i− 2 Δx .

 Δt  (1 − θ) F (Qni+ 1 ) − F (Qni− 1 ) 2 2 Δx    n+1  n  Δt ∂ ∂ ∂F ∂F Δx2 θ R1 + (1 − θ) R1 Δx ∂x ∂Q ∂x ∂Q + O(Δx2 Δt) (3. It may be shown (Mathews and Fink.11).18) The fourth step is to incorporate the error in the time discretization (3. 1998) that .

θF (Qn+1 ) + (1 − θ)F (Qni+ 1 ) Δt = i+ 1 2 2 where α is nonzero and finite. An additional notation is f (x) = o(g(x)) as x → xo if lim x→xo † Note that R1 and R2 are vectors and ∂F/∂Q is a matrix. f (x) =0 g(x) .

the algorithm is temporally second-order accurate.† If R1 = 0.7 Consistency The discrete approximation (3. and . θ = 0 (explicit Euler). Substituting into (3.19) where T1 is expressed in terms of F and θ. 3. Δt) are a = 0 b = 0 c = 0  ∂ d = − θ ∂x  e = −(θ − 12 ) ∂F R1 ∂Q n+1 ∂ + (1 − θ) ∂x ∂T1 ∂x  ∂F R1 ∂Q n  (3.7 Consistency  51 tn+1 tn Fi+ 12 dt + (θ − 12 )Δt2 T1 (xi+ 12 ) + O(Δt3 ) (3.20) The discrete system (3. 1 2 (Crank-Nicholson or trapezium).10) if it yields the integral equation (3. it is temporally first-order accurate.13) is therefore equivalent to Qn+1 = Qni − i  1 Δx  tn+1 tn Fi+ 12 − Fi− 12 dt +E(Δx. and if R1 = 0.15) in the error E(Δx.13) is consistent with the exact equation (3.22) The error is described in terms of its temporal and spatial components.21)  Qn+1 |exact i where the coefficients (3. then the algorithm is spatially second-order accurate.10) in the limit of Δt → 0 and † Three typical values are θ = θ = 1 (implicit Euler).18).3. and if θ = 12 . then it is spatially first-order accurate. Qn+1 i = Qni 1 Δx −  tn+1 F (Qi+ 12 ) − F (Qi− 12 ) dt tn ∂T1 + O(Δt2 Δx) + O(Δt3 ) ∂x   n+1  n  ∂ ∂F ∂ ∂F R1 + (1 − θ) R1 − ΔxΔt θ ∂x ∂Q ∂x ∂Q − (θ − 12 )Δt2 + O(Δx2 Δt) (3. Δt)  (3. If θ = 12 .

Δt)..9) as Fi+ 12 − Fi− 12 dQi + dt Δx =0 which implies E =0 Δt→0 Δt Δx fixed (3...52 Accuracy.. This is determined by the nature of the error E(Δx. and from (3.1).. Then   Qn+1  i discrete − Qni |exact Δt 1 + ΔtΔx Fi+ 12 − Fi− 12 dt that the discrete expression (3.23) as Δt → 0 and Δx → 0. Consistency.1) in the limit of Δx → 0 and Δt → 0.25) and so forth.10) provided a = 0 b = 0 O(Δx Δt i 3−i   ) c = 0 i=0.3 is   O(Δxi Δt3−i ) i=0. The additional requirement is consistency with the differential equation (2.. we have E(Δx.. from (3.14). and Stability the differential equation (2. Δx) where. this must approach (3...23) Taking the limit Δt → 0.. From (3. the condition is lim Δt→0 Δx→0 E =0 Δt (3. Convergence.3 + .15).   Qn+1  i   discrete = Qn+1  i exact + E(Δt.24) lim Thus...13) is consistent with the integral equation (3. the discrete approximation (3.13)  tn+1 Fi+ 12 − Fi− 12 dt = tn E Δt (3.1 (3. Δt) = aΔx + bΔt + cΔx2 + dΔxΔt + eΔt2 +   O(Δxi Δt3−i ) i=0. Taking the limit of (3.10) we have Qn+1 i = Qni 1 − Δx  tn+1 tn are exact and Assume the values for Qn+1 i n+1 is used to determine Qi |discrete .26) ...

We seek to determine the stability of (3.. Similarly.1 A Simple Flux Quadrature We begin with a simple and straightforward approach to illustrate the concept.8. although necessary.7). Thus. Qi+ 21 is the simple spatial average. In the following subsections. 3.27) using the analysis developed by von Neumann (1950)2 .25).7) which is used to define Fi+ 12 using (2. For purposes of simplicity.10). These methods are chosen for their obvious simplicity‡ to illustrate a fundamental requirement. namely.5).e. namely. We make two assumptions. ‡ Albeit they would not likely be employed in any practical computation. To this end. we show by example that. an approximate value for Q is obtained at xi+ 12 from (3. We also show that satisfying this requirement. Fi+ 12 = F(Qi+ 12 ) Qi+ 1 = 2 1 2 (Qi + Qi+1 ) (3. we present two different flux quadrature methods.. Certain discretizations will yield exponentially growing solutions that are physically implausible. 3.† Such discretications are denoted unstable.27) The first assumption states that the flux quadrature is based on an interpolated value Qi+ 12 .8 Flux Quadrature and Stability 53 which is already satisfied by (3. Chandrasekhara. the discrete approximation (3.8. . irrespective of the temporal quadrature method. discretizations that avoid such exponentially growing solutions are denoted stable.e.9) has a significant impact on the accuracy of the solution. for example.4). 1981).3. i. i. the spatial flux quadrature must respect the physical domain of dependence.13) is consistent1 with (3. is not sufficient to guarantee an acceptable solution everywhere (Section 3.8 Flux Quadrature and Stability The choice of spatial discretization for (3. the failure to meet this requirement leads to exponentially growing perturbations and an unsatisfactory solution. we assume that the † Here we distinguish purely numerical instabilities from physically unstable solutions of the linearized equations of motion as those that arise in hydrodynamic stability theory (see. The second assumption defines the interpolation (3.

Fi+ 12 = F (Qi ) + 12 Ai (Qi+1 − Qi ) (3. The quantity O((ΔQ)2 ) represents the terms that are quadratic in the difference ΔQ = Qi+ 1 − Qi . Consider Q(x.27). t) is Q(x. 2 Fi+ 12 ≈ F (Qi ) + Ai Qi+ 12 − Qi (3.54 Accuracy. The term ∂F/∂Q|i is denoted Ai and is A(Qi ).34) to linear terms.31)) and on t. Consistency. t) = Qi (t) (3. (3.31) ˆ k (t) are complex vectors whose subscript k indiThe Fourier coefficients Q ˆ k (t) indicates cates an ordering with respect to the summation index l. Using (3.33)  ∂F  2 = F (Qi+ 12 ) = F (Qi ) + Q 1 − Qi + O (ΔQ) i+ 2 ∂Q i (3. Thus. Q1 = Q M (3.32) dQi + =0 dt Δx is expanded in a Taylor series about Qi .35) where the ≈ sign will hereafter be replaced by an equals sign. and Stability flow is periodic in x over a length L = (M −1)Δx.e.36) . Given the values of Qi at some ˆ k at tn are obtained from time tn . Q dependence on l (through (3.28) and assume M is odd with M = 2N +1. the Fourier coefficients Q i=2N  ˆ k (tn ) = 1 Qi (tn )e−ιkxi Q 2N i=1 Consider the discrete Euler equations Fi+ 12 − Fi− 12 The flux vector Fi+ 12 Fi+ 12 (3.29) Then the Fourier series for Q(x.. i. t) = where ι = √ l=N  ˆ k (t)eιkx Q (3. t) to be a continuous vector function that interpolates Qi : Q(xi . Convergence.34) where we have substituted F for F using (3.27). truncating (3.30) l=−N +1 −1 and the wavenumber k depends on l according to k= 2πl L (3.

3.8 Flux Quadrature and Stability 55 and similarly Fi− 12 = F (Qi ) − 12 Ai (Qi − Qi−1 ) (3. l=N .30). Substituting for Qi from the (interpolating) Fourier series (3.37) Fi+ 12 − Fi− 12 = 12 Ai (Qi+1 − Qi−1 ) (3.39) dt 2Δx Assume A is slowly varying in x and hence can be treated as a constant to a first approximation.33) becomes dQi (Qi+1 − Qi−1 ) + Ai =0 (3.38) Thus and (3.

A = T ΛT −1 (3. . and therefore ˆk dQ ˆ k = 0 for l = −N +1.47) gives .41) can be solved exactly.45) Multiplying (3.41) + GQ dt or equivalently ˆk dQ ˆ k = 0 for k = + GQ dt π (−N +1) π . . ˆk  1 dQ ιkxi ˆ k eιk(xi +Δx) − eιk(xi −Δx) e A Q + dt 2Δx l=−N +1 l=−N +1 l=N  (3.42) where G = αA (3. Using (2.40) The coefficients of the individual Fourier terms eιkxi must vanish.46) ˜k dQ ˜k = 0 + αΛQ dt (3. .44) α= Δx Equation (3. Δx Δx N (3. .43) implying that each element of the matrix A is multiplied by the complex coefficient α defined as ι sin kΔx (3. . . N (3.15). . .41) by T −1 and defining ˜ k = T −1 Q ˆk Q (3.

N .51) Writing where ωmr and ωmi are the real and imaginary parts of ωm . The initial condition at κao t = 0 and the solution at κao t = 1. Periodic boundary conditions are imposed on the left and right boundaries.e. and Stability Defining ˜k = Q ⎧ ⎫ ˜ ⎬ ⎪ ⎨ Qk1 ⎪ ˜ Q k2 ⎪ ⎩ ˜ ⎪ Qk ⎭ (3.49) for m = 1.1. The profile at κao t = 1.48) 3 gives ˜ k (t) = Q ˜ k (0)e−ωm t Q m m (3. .52) Thus. the solution does not grow exponentially in time) is ωmr ≥ 0 (3.92) with = 0.8. .25 shows close agreement with the exact solution.2 Another Simple Flux Quadrature We further illustrate the concept of spatial stability by considering a different formula for the flux quadrature Fi+ 12 = F(Qi+ 12 ) † However.50) ωm = ωmr + ι ωmi (3. (3.12).25 are shown.8. gives ˜ k (t) = Q ˜ k (0)e−ωmr t e−ιωmi t Q m m (3.54) This is shown in Fig. The initial condition is defined by (2.56 Accuracy.† 3. Convergence. Consistency. the flux quadrature formula (3. 2.27) is stable. the agreement degenerates when the shock wave forms (Section 3. and ωm = αλm (3. where ωm are complex constants. .8.44) and (2. 3. 3 and l = −N+1. .3 for the problem described in Section 2.53) From (3. the condition for stability (i. we have ω mr = 0 sin kΔx λm ωmi = Δx Therefore. ..4). and the domain 0 < κx < 2π is discretized into 100 uniform cells where κ = 2πL−1 .

.. .. ..... ....... ........ . ... .... ... .. ... .... ... . .... . ..... ... ......... . .. ... ... ... ....10 κx −0.60) (3...3..... . . . ..... κa t = 0 0... . . Proceeding as before.. .. 3. ....... .. . ... .. .... ... .. .. . .. .. . . . . ... .. ... .. ... ..... .... .. ... .15 ......... .................. . .... ...... .... .. .... ...... ....... . ... ... ... .. .. .. .. . . . ..... .. ........ . .. . . . .... ..... . . . .... .... κa t = 1. ....57) G = αA (3.. ....41) by ˆk dQ ˆ k = 0 for l = −N +1. ... . .. 3..27) Qi+ 21 = Qi (3. ..10 0. ....... .... ......... .. .... .....56) dt Δx and Equation (3....... . . .... . .39) is replaced by (Qi − Qi−1 ) dQi + Ai =0 (3... ... . .. .3. . .. ........ . .. . . . ... .. .. .... ... . .. ... ωmr ≥ 0 requires ωmr = λ1 ≥ 0 (3. .. .58) where and sin kΔx (1 − cos kΔx) +ι Δx Δx Following the same analysis used previously.. . . .... . . .... .. . .. ... ... .. α= (1 − cos kΔx) λm Δx sin kΔx ω mi = λm Δx for m = 1... ........... . . . . ... .....59) (3... .... . ..... ..... .. Equation (3. . . .. . .. . . .. .. .... . ... ...... .. .. ....... ... ... .... .. ................. .. . . .. .00 ◦◦ ◦◦ ◦◦ ◦◦ ◦ ◦ ◦◦ ◦◦ ◦◦ ◦◦ ◦ ◦ ◦◦ ◦◦ ◦◦ ◦◦ ◦◦ ◦◦ ◦◦ −0.... .. . ...... . ... . . .. . 2. .. ... ....... . .. .. Computed and exact solutions for the flux quadrature (3. ...... . .... ..... .61) .. . ..05 ◦◦◦ ◦◦ ◦ ◦ ◦ ◦ ◦ ◦◦ ◦◦ ◦◦◦ ◦ ◦◦ ◦◦ ◦◦◦ ◦◦◦ ◦◦◦ ◦◦◦◦◦ ◦◦◦◦◦◦◦◦◦◦◦◦ ◦◦◦◦◦ ◦◦◦◦◦◦ ◦◦◦◦◦ −0..... ..... . ...55) This corresponds to using the Qi of the cell to the left side of the cell face at xi+ 21 to determine Fi+ 12 . Since 1 − cos kΔx ≥ 0 for all k.... . .8 Flux Quadrature and Stability 57 u/ao .. .... ..... .... ... ....... . . .. . ... .. ...... ...... . ... ....05 ◦◦ ◦◦ ◦◦ ◦◦ ◦◦ ◦◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ 1 ◦◦◦ 2 3◦◦ 4 ◦◦◦ 5 6 ◦ ◦ 0.. ....... .. .... . ..... ◦ computed ◦◦◦◦◦ ◦◦◦◦ ◦◦◦◦◦◦ ◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦ ◦◦◦◦◦◦◦ ◦◦◦ ◦◦◦ exact ◦◦◦ ◦ ◦ ◦ ◦◦ ◦◦◦ ◦◦ ◦◦ ◦ ◦ ◦ ◦◦ ◦◦ 0. . . ... .... ... . .. . . .. ....... . .. ... ...25 o o .. ........ ....... .. N + GQ dt (3... .....15 Fig..... . . ....

..... .. ... ... . . .. ...55) is stable only for fluid traveling to the right at or above the speed of sound. . ...... . . ... ..... . x Fig. ..35) have positive slope as illustrated in Fig.62) u≥a (3. ... .. . .63) which implies Thus... . The three characteristics intersecting xi+ 12 are shown with positive slopes An example of the formation of the instability is shown in Fig.. Periodic boundary conditions are imposed on the left and right boundaries...... This is consistent with flow physics. . . . . ...... .55) is of limited usefulness.. . ..... . . the method in (3..... 3. ...4.. . . . .. ... . ........ .. 3......25 wherein the computed velocity displays a pointwise fluctuation. The initial condition is defined by (2. . .. . .. .. The formula (3. ....... .. t .. . .. .. . ... . Convergence. . .. . ... . . .92) with = 0. . ... .... If it were to be employed. . 3. .... . ..63) is not met............ . ... . ..... .. . .. Consistency.. The instability is expected as the condition (3.. .. . . .. The initial condition at κao t = 0 and the solution at κao t = 1.55) for the flux Fi+ 12 uses information only from the cell to the left of the face at i + 12 ...5 for the problem described in Section 2... . ... .. ... . .... . .1. ..58 Accuracy. the flow at xi+ 12 only depends on conditions at x < xi+ 12 . ... .. .. . .... ... ... ..... . ... . . . ... .. .. ... . . .... i. and Stability λ2 ≥ 0 λ3 ≥ 0 (3. ... .. . ..... .. .. . . . ... . 1 3 2 xi+ 12 ..4. ...... . .. . . .. .. it would be necessary to test for satisfaction of this condition at every timestep and every face and employ some other method if this condition were not met.. and the domain 0 < κx < 2π is discretized into 100 uniform cells. The algorithm (3......63). .. . . dt 0 p . ..... however.. . .. .. .. .. .... . ... . . . ... ... . . due to the condition (3.... . ....e.. A numerical instability is apparent at κao t = 1...8.. . . ... This is physically sensible only if all three characteristics defined by (2. ... .. ... .25 are shown.. .. . .

. ................. † In a heuristic. . .... .55) 3. .. ..........5. . . .. ... ..... . ..... ... . ... .. . ... ...... .. . ........... .............. ....... . . ... ... .....27) incorporates more information than necessary under certain circumstances. Computed and exact solutions for the flux quadrature (3.. ... and admitedly nonrigorous.. 3............. . ... .......... . . ..... ... . .... ............... .. ... ........ ............ ... ....... ........ .. .......... This is a principle reason for the development of flux algorithms that are strongly motivated by flow physics.. . .. ... ............ . ..... .. .... ... ....... . . . ...... . algorithm (3. ...35).. .. .. .......... .. . ... ......... . .. .... . ... .. . ..... .. . ...... ... ... .. . ... ..... .. .. .... . . .. In other words..... ... ........ .. ... .. ...... ...00 ◦◦ ◦◦ ◦ ◦◦ ◦◦ ◦◦ ◦◦ ◦◦ ◦◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦◦ ◦ ◦ ◦◦ ◦◦ ◦◦◦◦◦◦◦ ◦ −0..... . .... ..... ... .... .... .. ...... .............. ...... .. . In summary. .. ...... ........... . .... .... ... .... ....10 ◦ ◦ ◦◦ ◦ ◦◦ ◦ ◦ ◦◦ −0. .. .. .... ..... . .. .... . . . . ... . ..... . .. . ... . .. . .. ... .. .... .... .. .. . ....... .... Regardless of the magnitude of u relative to a and the sign of u.... .... .. ....... .... . .... . .. .. ....... . ....... ...27) does not exhibit a constraint similar to (3. .. . . ... .. .25 o o .... . ...... . ... ... . .. ... .... . ..... . ..... .... .......... . The explanation is again based on flow physics....... .. . . ........ . ....... ..... However. ....... .......... . ........... .. .. . ......3.... .... . . ....... ... ........ . .. . .... . ... ... . .... manner.. . .. .... .. ........ . .. . . .... ..05 ◦ ◦◦◦◦ ◦ ◦◦ ◦ ◦ ◦ ◦ ◦◦◦ ◦ ◦◦◦ ◦ ◦◦◦ ◦◦◦ ◦ ◦◦◦◦◦ ◦◦◦◦◦◦◦◦ ◦ ◦◦◦◦◦◦ −0...... . .. ... .. . ...... . ... ..... .. .. . .. ......... .. ..... . ....... .... ... ...8.... ... ...................... ....63).... ...................... . .. .. . .... . ......... ..... .. . ..... ....... . ........ . . Indeed. . .... .. we have observed† that the flow physics must be fully respected by the spatial flux quadrature..... ......... .. .. .. ...... ..... .. ... . ..... . .... . ... ..... .... . . an excess of information regarding the flowfield is not destabilizing...... ..... . . .... .. . ... ... ....... .. .....8 Flux Quadrature and Stability 59 u/ao .. ............ . ◦ computed ◦◦◦◦ ◦ ◦◦◦◦◦◦ ◦◦◦◦◦◦◦◦ ◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦ ◦◦◦ ◦◦◦ ◦◦◦ exact ◦ ◦ ◦ ◦◦◦ ◦◦◦ ◦◦ ◦◦◦ ◦ ◦ ◦◦ ◦◦ 0.. whereas an absence of relevant flow information is destabilizing. .... . κa t = 0 κa t = 1... ... .. ..... . . . . ... .. ... .... .... .. .10 κx Fig.... . ...05 ◦◦ ◦◦ ◦◦ ◦◦ ◦◦◦ ◦◦ ◦ ◦ ◦ ◦ ◦◦ ◦◦ ◦◦◦ 2 1 3 4 ◦◦◦ 5 6 ◦ ◦ ◦ ◦ ◦ ◦ ◦ 0.. ... ... ... ... . .. .. the algorithm (3..... .. .... 0.. ... . ... . .... .... . ..... .. ...3 Numerical Domain of Dependence It is useful to consider why the algorithm (3... ... .... ...... . . .15 0..... .27) incorporates flow information corresponding to all three flow characteristics in (2...... . This requirement is independent of whatever algorithm is chosen for the temporal integration.. ....... ... .... ..... .. .... .....15 ..... .. .. . . We shall discuss some of these algorithms in Chapters 5 and 6. ... ...... . . the numerical domain of dependence must include the physical domain of dependence......... .............. . ....

....... .. ... . .....53).. . ..... . .. . .... ... ... that the simple flux algorithm Fi+ 12 = F(Qi+ 12 ) Qi+ 21 = 1 2 (Qi + Qi+1 ) is not adequate in general.... .. . 3... . . . .. .. . ..... .. ... .10 −0.. .. .. ... .. ... . . . .. ....05 ◦◦ ◦◦◦◦◦◦ ◦◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦◦ 1 2◦◦◦◦◦◦ 3◦◦ 4 5 6 ◦◦ ◦ ◦ 0.. despite satisfying the stability condition (3. .... ....e... . .. .. .. .. ........ . . ...... ..... . . .. . 3. ... As noted in Section 2... .. ... ... and Stability 3... .......... . . . .. . ......... ....... . ..15 . . ... . . .. ...... . . . . ..... .. .. . A solution to this problem is presented in Chapter 8... . ..... .. . ... ... . 0.....8..10 0. .. . . ... . . .. .... .... ..6... .....33 ◦◦ ◦◦ .. The computed profile displays severe oscillations immediately downstream of the shock wave (i... . . ..... ...... ... . ... ..4 Shock Waves and Weak Solutions It must be noted. . ...05 ◦ ◦ ◦ ◦◦ ◦◦ ◦◦◦ ◦◦◦◦ ◦◦◦◦◦ ◦◦◦◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦◦◦◦◦◦ ◦◦◦◦◦◦ ◦◦◦◦ ◦◦◦◦◦◦ ◦◦ ◦◦◦◦ ◦◦◦◦ −0. . ..... . . ...... .. .. ... .... . . .... .. ........... ..... .. .. ... ...... ...... . . the initial sinusoidal disturbance forms a train of shock waves in a finite time. Consistency. .. . .... . . ...... .. Convergence. ........ . ... .. . .. ...... .... . ..... ..... ....... .... ..... .. . . ........ for example. .. . . . .. .. .. ... . ... ...... ...... .6 shows the computed and exact profiles for the velocity for the same conditions as Fig. . . .... ... The appearance of the oscillations is analogous to the Gibbs phenomenon in the Fourier series representation of a discontinuous function (see. .... .. . . ....00 ◦◦ ◦◦ ◦◦ κx ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦◦ ◦◦ ◦◦◦◦ ◦◦ ◦◦ ◦◦◦◦ ◦ ◦ ◦ −0.. ... ... . . however. .......... . .... .8........... . . ... ... .. These oscillations are numerical in origin and represent a fundamental shortcoming of the flux algorithm...... .... . ..... ............. . ... .. . .... Example of instability associated with a shock wave ..... . as viewed by an observer traveling with the shock wave).... ............ . . ... .. . . .. .. . .97)... . ..... .... . ...... ... .. ......... ........ .... .. .... ..... .. .... . . .......3 at the time of initial formation of the shock wave in (2.. ..... ..... . . .. ......... .. ... . .... ... ◦◦ ◦ ◦◦◦◦◦◦◦◦◦◦◦◦◦ ◦◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ computed ◦◦◦◦ ◦◦◦ ◦ ◦ ◦ exact ◦ ◦ ◦◦◦ ◦ ◦◦ ◦◦ ◦ ◦◦◦ ◦◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦◦ ◦◦ ◦ ◦◦ 0.. . ..... ........ = 0 κao t = 8. . . . . ........... .. .... .......... ..... ... .......... . . .. . . ... .. .. Figure 3. .. ........ ........ .. .. . . .. .. . .... ...... namely..... . ... . ...........15 Fig.. ...... .. .. Greenberg..... 1998). ... .. ... .. ..... ........ .. ... .... .. . ... ... .. ... ... .. .... . .. . .60 Accuracy.. κao t..... ....... ..... .. . .... .. .... .. ...... .. . . ... ....... . .. . . ..... u/ao . .... . .... .. ..... . ........ . .. . ...... .... .. the assumption that Q is a continuous function in x.. ....

3.9 Stability

61

3.9 Stability
The selection of a physically consistent spatial flux quadrature is not sufficient to ensure the stability of the discrete approximation (3.13):
= Qni −
Qn+1
i
− 

Δt 

n+1
θ F (Qn+1
)

F
(Q
)
1
1
i+ 2
i− 2
Δx


Δt 
(1 − θ) F (Qni+ 1 ) − F (Qni− 1 )
2
2
Δx

as we shall now illustrate.

3.9.1 A Simple Flux Quadrature
We assume the simple flux quadrature (3.27),
Qi+ 1 =
2

1
2

(Qi+1 + Qi )

This algorithm was shown to be stable for the semi-discrete stability analysis
(Section 3.8.1).
The stability analysis of the discrete approximation is similar to the stability analysis for the semi-discrete approximation presented in Section 3.8.
The flow is assumed periodic in x,
Q1 = QM
where M is odd and M = 2N + 1. Defining Q(x, tn ) to be a semi-discrete
vector function that interpolates Qni , the Fourier series for Q(x, tn ) is
Q(x, tn ) =

l=N 

ˆ nk eιkx
Q

l=−N +1


where ι = −1 and the wavenumber k depends on l according to (3.31).
ˆ n are complex vectors whose subscript k indicates
The Fourier coefficients Q
k
an ordering with respect to the summation index l through (3.31) and the
superscript n indicates a discrete time level tn . Given the values of Qi at
ˆ n at tn are obtained from
some time tn , the Fourier coefficients Q
k
i=2N 

ˆn = 1
Qi (tn )e−ιkxi
Q
k
2N i=1

The flux vector Fi+ 12 is expanded in a Taylor series about Qi ,

Fi+ 12 = F (Qi+ 12 ) = F (Qi ) + Ai Qi+ 21 − Qi + O (ΔQ)2

62

Accuracy, Consistency, Convergence, and Stability

where A = ∂F/∂Q is assumed constant. Thus, (3.13) is approximated by
Qn+1
= Qni −
i

Δt

n+1
A θ Qi+ 1 − Qn+1
1
i− 2
2
Δx

Δt

A (1 − θ) Qni+ 1 − Qni− 1 2 2 Δx which yields = Qni − Qn+1 i − Δt .

.46). Substitute for Qi from the (interpolating) Fourier series and note that the coefficients of the individual Fourier terms eιkxi must vanish: ˆ n+1 = [I − Δt(1 − θ)αA] Q ˆn [I + ΔtθαA] Q k k (3.64) can be rewritten as ˜ nk ˜ n+1 = [T − Δt(1 − θ) α T Λ] Q [T + Δt θ α T Λ] Q k Multiplying by T −1 .44). ˜ n+1 = [I − Δt(1 − θ) α Λ] Q ˜n [I + Δt θ α Λ] Q k k and therefore ˜ n+1 = GQ ˜ nk Q k (3. A = T ΛT −1 where Λ is the diagonal matrix of eigenvalues λi defined in (2. From (2.15). Using (3. n+1 A θ Qi+1 − Qn+1 i−1 2Δx    Δt A (1 − θ) Qni+1 − Qni−1 2Δx which is a discrete analog of (3. ˆk ˜ k = T −1 Q Q Then (3.65) G = [I + Δt θ α Λ]−1 [I − Δt(1 − θ) α Λ] (3.64) where I is the identity matrix† and α= ι sin kΔx Δx which is the same as (3.39).12).66) where † The identity matrix has 1’s on the diagonal and 0’s elsewhere.

G Gn = GG   (3. .3.72) i=1 and C is a constant independent of k. .9 Stability 63 It is straightforward to show that ⎧ ⎪ ⎨ γ1 G= ⎫ 0 0 ⎪ ⎬ γ2 0 ⎪ 0 γ3 ⎭ 0 ⎪ ⎩ 0 (3.68) ˜ n+1 is therefore The solution for Q k ˜ 1k ˜ n+1 = Gn Q Q k (3. The norm ||G|| of a matrix G is defined as4 |Gx| k.x=0 |x| ||G|| = max (3.x=0 |x| max (3.73) where the maximum is taken over all possible k and all nonzero complex vectors x. Q k sensible instability) for any arbitrary initial condition.70) n times ˜ n+1 must remain bounded (in the absence of a physically Intuitively.74) and thus (3. 3 (3. Thus ˜ 1 | for all k ˜ 1 | ≤ C|Q |Gn Q k k or |Gn x| ≤C k.71) where |x| indicates the norm3 of the vector x: |x| = i=n  !1/2 |xi | 2 (3.67) where γi are the eigenvalues of G given by γi = [1 − Δt(1−θ) α λi ] [1 + Δt θ α λi ]−1 for i = 1. 2.75) .73) is equivalent to ||Gn || ≤ C (3.69) ˜ 1 is the initial condition and where Q k . ˜ n+1 | ≤ C|Q ˜ 1 | for all k |Q k k (3.

76) is somewhat cumbersome to analyze due to the presence of the power n.78) i where γi are the eigenvalues of G. ||Gn || ≤ C ⎧ ⎪ ⎨ 0 < Δt < τ ⎫ ⎪ ⎬ ⎪ ⎩ all k ⎪ ⎭ 0 < nΔt < T for (3. and Stability This is the condition for stability (Richtmyer and Morton.80) becomes |γi | ≤ 1 + O(Δt) for ⎧ ⎪ ⎨ 0 < Δt < τ ⎫ ⎪ ⎬ ⎪ ⎩ all k ⎪ ⎭ 0 < nΔt < T (3. Without loss of generality† we may take C2 ≥ 1. . equivalently. Convergence. 1 Δt R ≤ C2n = C2T Now Δt C2T = exp( Δt T log C2 ) = 1 + O(Δt) Δt Thus C2T is bounded by 1 + C3 Δt for 0 < Δt < τ and hence (3. Additionally. We therefore seek a simpler form. 1967). where C2 is a constant independent of k. a necessary condition for stability is Rn ≤ C2 for (3.79) ⎧ ⎪ ⎨ 0 < Δt < τ ⎫ ⎪ ⎬ ⎪ ⎩ all k ⎪ ⎭ 0 < nΔt < T (3.76) for some positive τ and T .80) for some positive τ .77) where R(G) is the spectral radius of G defined by R(G) = max |γi | (3. for any matrix G.80) would still hold. Consistency. It is straightforward to show that. we could redefine C2 to be equal to some value greater than 1 and the necessary condition (3.81) † If in fact C  < 1. Equation (3. it can be shown (Exercise 3. which may be more precisely stated as n G is uniformly bounded or. Then. ||G|| ≥ R(G) (3.9) that Rn ≤ ||Gn || ≤ ||G||n Then.64 Accuracy.

82) and the von Neumann condition is then clearly sufficient for stability. for example.67) that G is a normal matrix.84) Then (3.94).8 with = 0. This is the von Neumann condition for stability. It can be shown (Franklin. Anderson et al. † See. where ΔtCF L is defined in (3.83) becomes (β Δt λi )2 (1−2θ) ≤ 0 (3..8) Rn = ||Gn || = ||G||n (3. 3. and a time step Δt = ΔtCF L .85) for all i. 1967). It is evident from (3. the term O(Δt) is omitted.86) Thus. for the flux quadrature Qi+ 1 = 2 1 2 (Qi + Qi+1 ) Euler explicit (θ = 0) is unconditionally unstable†. the domain 0 < κx < 2π discretized into 100 uniform cells. The condition for stability is θ≥ 1 2 (3. ‡ A normal matrix is an n × n matrix with n orthogonal eigenvectors. i.83) i assuming there are no physically sensible instabilities. The instability for Euler explicit is shown in Fig.7 for the problem described in Section 2. GG∗ = G∗ G. .3. It is therefore evident that stability for the semi-discrete approximation is not a sufficient condition for stability of the discrete approximation. (1984). where G∗ is the complex conjugate transpose of G.e. The term O(Δt) permits exponential growth where physically sensible (Richtmyer and Morton. Define α = ιβ where β = sin kΔx Δx (3. then (3.79) becomes (Exercise 3.01.9 Stability 65 for all eigenvalues γi of G. 1968) that a normal matrix commutes with its adjoint. If G is a normal matrix‡. otherwise. and therefore the necessary and sufficient condition for stability is    1 − α Δt (1−θ)λi   ≤1  1 + α Δt θλ  (3. while Euler implicit (θ = 1) and Crank-Nicholson (θ = 12 ) are unconditionally stable.

. . .. . .... .... .......... ........ . . ......... .. . . ..... .. . . .. .... . ..90).. ..... ..... Convergence..... ... . . . .. .. ..87) ˆn ˆ n+1 = [I − Δt(1 − θ)αA] Q [I + ΔtθαA] Q k k (3. ..010 ◦ ◦◦◦◦◦ ◦◦◦◦◦◦ ◦◦ ◦ ◦◦◦◦◦ ◦◦ ◦ ◦◦◦◦◦◦◦ exact ◦ ◦ ◦◦ ◦◦ ◦ ◦◦◦ ◦ ◦ ◦ ◦ ◦ 0. ... . . ... . . .. ... .. . ..... .... The result is  Δt λi Δx  "  # Δt λi (1−2θ) − 1 ≤ 0 Δx (3. .... .. .88) α = a + ιb (3........ . .. . . . . . ... . Qi+ 21 = Qi This algorithm was shown to be stable for the semi-discrete analysis (Section 3... ... . ... . . ..89) By a similar analysis. ... .. . ..... ......... .. . .......... ...... .... .... . ...... ......... ... ...90) The necessary and sufficient condition for stability is again (3..005 ◦ ◦ ◦◦◦ ◦ ◦◦◦◦ ◦ ◦ ◦ ◦◦◦◦ ◦ ◦ ◦ ◦◦ ◦ ◦◦◦◦◦◦◦ ◦ ◦◦◦◦◦◦◦◦◦◦ ◦ κa t = 5...... .....2) provided λi ≥ 0 for i = 1............ . . . .. . ... ...... . .. .. .. ..... ... ........... ... . ....... .. ..... . ... ... .... 0... ....9. .. ... .... . ... 3...... . . ....... .... ..... .. ... .. ........... ............ .......... ......... . .... .. . . ... . .. ...... .... .. . .. ........... ... .. .... ......... .....005 ◦ ◦◦◦ ◦ ◦ ◦◦◦ ◦◦ ◦◦ ◦◦ ◦ ◦ ◦ ◦◦ ◦ ◦◦ ◦◦ 1 2 4 5 ◦ 3◦◦◦ ◦ 6 ◦◦ 0.... Consistency.... ......... .... .. κa t = 0 ◦ ◦ ◦◦ ◦◦ ◦ ◦ ◦◦ ◦ ◦ ◦ ◦◦ ◦◦ ◦◦◦◦◦ ◦ ◦ computed 0.. κx Fig.. .015 . .... . ....... ....... ... 2... .... .. .. ...... . ........ ....... . . .. ........ ... .. . ....... ....... . . ............ . . ........ ... . . . .. .. .000 ◦◦ ◦ ◦ ◦ ◦◦ ◦◦ ◦◦ ◦◦◦◦ ◦◦◦ ◦◦◦◦ ◦ ◦◦◦ ◦◦ ◦◦◦ ◦◦ ◦ −0......... . ... .010 ◦ ◦◦ ◦ ◦◦◦◦◦◦◦◦◦◦ ◦ ◦◦ ◦ ◦ ◦◦ ◦ ◦ ◦ −0.... ... . . ..... .. . . .. . ..... .... ... . ... ..... . . ... .... .... ...... .............. . .. ....... .. ...2 Another Simple Flux Quadrature We consider the simple flux quadrature (3.. .. . ......... and Stability u/ao . ........... .. . ......... ..... ... .... .... ... ......... ... . . ....... ... .. .. . . . .......89) and (3. .... ...... .... .. ... .. . .... . 3 (3.... .... .. . .. . . .... . . . . ... . . . .... .. . . .... ...... ....... ... ... . ........... . ......... Example of instability with Euler explicit 3.. . .. ...... .. ... .. . ... ........ .015 o .. . . ..... .. ... ... . ...... .. ... ... ..... ..... .. .... .. . .. ... .. . ... . . ..... . ........ . ...... . . .... .. . ....... .. ..75 −0.....7.... . ... ..... .......... .. .. .. ... ......91) ......... . . . ... . ...... ....... .. . ... .. ... ............... .... . ......... ....... .. .... ..... .... .. .. .. .... . ... ......... ...83) where α is defined by (3............. ... ...... where where a = b = (1−cos kΔx) Δx sin kΔx Δx (3.55)...... ........... ..... .66 Accuracy......... ... ...... ... .. . .. o .... .. . . . ... .......... .... .... . .. .......8. .

is identical to (3. too.97) Note that the first restriction in (3. For Euler implicit (θ = 1).93) is identical to the condition obtained for the semi-discrete stability analysis in (3.95) i More generally. 2.63). 2. 3. i The second restriction in (3.99) For λi ≥ 0.93) where ΔtCF L = min Δx λi (3. the stability condition is  Δt ≥ max − i Δx λi  (3. the algorithm is conditionally stable with the restriction λi ≥ 0 (3.98) This.92) Δt ≤ ΔtCF L (3.94) ΔtCF L = min Δx |λi | (3.9 Stability 67 for i = 1.3.96) and the second restriction can be rewritten as C≤1 (3. The Courant number C is defined as Δt ΔtCF L C= (3. then the algorithm is conditionally stable with a minimum allowable timestep.93) is the Courant-Friedrichs-Lewy (or CFL) condition. Thus. If one or more λi is negative. For Crank Nicholson (θ = 12 ). For Euler explicit (θ = 0). the algorithm is conditionally stable with the restrictions Δx λi ≥ 0 and Δt ≤ λi for i = 1. u ≥ a (3. .63). the algorithm is unconditionally stable. 3.

e i −Qi ||κao Δt = κao Δt 10−3 2 n ||Qn.ρe i=1 1/2 (3. and Stability 3. For a spatially and temporally second-order accurate algorithm.e = Qdxdy (3. Convergence. and ρe as defined in (3.1. Δx ) The results for the simple flux quadrature (3.8 for the problem described in Section 2.100) as  ||Qn.8 is n ||Qn.e i − Qni || $ = i=M  1  −2 n 2 Q∞ (Qn.e i −Qi ||κao Δt=10−3 . It is evident that the computed solution converges quadratically† to the exact solution for sufficiently small Δt.85) to (2.e i − Qi ) M ρ. The exact solution Qni is obtained from (2. and the domain is 0 < κx < 2π.9.68 Accuracy. † The quadratic line in Fig. as it is usually difficult to demonstrate the convergence of the numerical solution except for simple problems † We use the double vertical brace notation || · || to denote the norm of the composite of the solutions for ρ.10 Convergence In the context of a finite volume algorithm. which is prior to the shock formation.102) where the first implies summation over the three conservative variables and Q∞ = (ρo . Δx2 ) Qn+1 i i and therefore n 2 2 ||Qn.92) with = 0.e i − Qi || = O(Δt .e i n lim ||Qn. The Equivalence Theorem of Lax and Richtmyer (1956) provides an important insight to the issue of convergence. The initial condition is defined by (2. ρu.ρu.27) are presented in Fig.100) Δt→0 Δx→0 is the volume average of the exact solution over at a fixed value t where Qn. a numerical solution Qni conif† verges to the exact solution Qn.e i cell i:  1 Qn. The norm is evaluated at κao t = 7.88). 3.e i − Qi || = 0 (3. ρo a2o )T . This is not be confused with the matrix norm introduced in Section 3. ρo ao . Consistency. and Δt ∝ Δx. 3. = Qn+1 + O(Δt2 .101) i Vi Vi We employ the root-mean-square for the norm in (3.102).8.

........ . . . . ... ....11 Conclusion In this chapter.. ........ .. .... ....... .. ... ......3............ lim x→0 sin x cos x = lim =1 x x→0 1 . . .... .. . .... ...... .. . ..... .... .... ...... ... .. ..... ... ...... .. .. .. −2 .. ... the theorem states that the stability of the finite difference approximation is a necessary and sufficient condition for convergence.. and c) x log x = o(1) as x → 0 solution a) By L’Hospital’s rule.. and unstable methods.... . Under these conditions. ◦ ..... ◦ .. .. ....... .. 3....11 Conclusion 69 n log10 ||Qn.. ...... and stability... ..... and a significant number of other algorithms have been developed.. ..........13) to the Euler equations for the purpose of illustrating the concepts of accuracy...... b) cos x = O(x − π/2) as x → π/2.... In other words. .. The theorem applies to linear systems (with properly posed initial conditions) discretized by a consistent finite difference approximation........... ... ...... consistency... .....e i −Qi || 0 −1 −3 −4 .. ◦ . . ........ ....... convergence. . ..... .8.......... ..... .. ... . . ... ◦ ◦ ◦ computed quadratic ◦ ◦ ◦ −5 −3 −2 −1 0 log10 κao Δt Fig. ..... Convergence study whose exact solutions are known.................... . 3....... . ......... .... ... ..... .. ......... .. This particular discretization incorporates both unconditionally and conditionally stable methods... .. .... we have introduced a family of discrete approximations (3... .. ..1 Show that a) sin x = O(x) as x → 0.. .. .. ....... . . . .... . ........ . .. .. ... . It is by no means a full catalog of discretization schemes..5 Exercises 3... ...... consistency and stability imply convergence. .......... ..........

momentum. 3.34). x2 + Δx2 . x log x = o(1) as x → 0. etc. and energy. lim x→π/2 cos x = (x−π/2) lim − x→π/2 sin x = −1 1 Thus. and H3 as the Hessian matrices for the conservation of mass. solution The Taylor series expansion of a function f (x1 + Δx1 . H2 . Consistency.4 = ⎧ ⎪ ⎨ = = ⎪ ⎩ ⎧ ⎪ ⎪ ⎨ ⎪ ⎪ ⎩ 0 0 0 0 0 0 0 0 0  Q2 2 (γ −3) Q Q2 −(γ −3) Q23 1 1 2 (γ −3) Q Q2 0 −(γ −3) Q1 0 0 0 1 1 Q3 2 Q3 2γ QQ − 3(γ −1) Q24 3 1 Q2 3 −γ Q + 3(γ −1) Q23 Q2 1 ⎫ ⎪ ⎬ 0 −γ Q22 Q1 1 1 ⎪ ⎭ Q2 3 −γ Q + 3(γ −1) Q23 Q2 2 −γ Q Q2 2 −3(γ −1) Q Q2 γ Q1 1 γ Q1 1 1 0 1 ⎫ ⎪ ⎪ ⎬ ⎪ ⎪ ⎭ In deriving (3.3 Repeat the derivation in Section 3. Then. x2 . Convergence. x3 + Δx3 ) = f (x1 . Discuss the physical significance.  H1 H2 H3 3.8. it was stated that the coefficients of the individual Fourier terms eιkxi must vanish. x3 ) +  1 2 where H is the Hessian matrix  H= fx1 x1 fx2 x1 fx3 x1 fx1  fx2 Δx1 fx1 x2 fx2 x2 fx3 x2 fx3 Δx2 fx1 x3 fx2 x3 fx3 x3   Δx3 Δx1 Δx2 Δx3   +  Δx1 Δx2 Δx3 H   and fx1 = ∂f /∂x1 . x2 + Δx2 . cos x = O(x−π/2) as x → π/2. x3 ) can be expressed as f (x1 + Δx1 . lim x log x = lim x→0 x→0 x−1 log x = lim − −2 = lim −x = 0 −1 x x→0 x x→0 Thus. c) By L’Hospital’s rule. x3 + Δx3 ) about (x1 . Denote H1 . Find an explicit expression for the O(ΔQ)2 terms in (3. and Stability Thus.2 3. .64). Prove this statement.70 Accuracy. b) By L’Hospital’s rule.2 with Qi+ 12 = Qi+1 and show the method is stable only if u ≤ −a. sin x = O(x) as x → 0.41) and (3. x2 .

solution Equations (3. 1966). then 1 1 ≤ ||(1 − G)−1 || ≤ 1 + ||G|| 1 − ||G|| solution From the identity I = (1 − G) (1 − G)−1 we have ||I|| = || (1 − G) (1 − G)−1 || Now || (1 − G) (1 − G)−1 || ≤ || (1 − G) || || (1 − G)−1 || and || (1 − G) || ≤ (1 + ||G||) Thus. 1 ≤ (1 + ||G||) || (1 − G)−1 || and therefore 1 ≤ || (1 − G)−1 || 1 + ||G|| From (I − G) (I − G)−1 = I we obtain (I − G)−1 = I + G (I − G)−1 and thus || (I − G)−1 || ≤ 1 + ||G|| || (I − G)−1 || Since ||G|| ≤ 1.5 71 Prove that (3. 1 1 − ||G|| . 3.Exercises 3.62) implies (3. || (I − G)−1 || ≤ (Isaacson and Keller. where R(G) is the spectral radius of G.7 Show that ||G|| ≥ R(G).62) are u u+a ≥ ≥ 0 0 u−a ≥ 0 which can be written u u ≥ ≥ 0 −a u ≥ a The third condition implies the first and second and therefore is the requirement.6 3. since ||I|| = 1.63). Prove that if ||G|| < 1.

ein )T of G with eigenvalues λi . Then ||G2 || = = ≥ max |GGx| x = 0 |x| max |GGx| |Gx| x = 0 |Gx| |x| max |GGx| |Gx| x∈V |Gx| |x| where the inequality arises since the set of eigenvectors V is a subset of all nonzero vectors x. . and Stability Show that if G is a normal matrix. Consistency.72 3. . . Rn ≤ ||Gn || ≤ ||G||n .83).9 Accuracy. where R is the spectral radius of G. Show that. . then Rn = ||Gn || = ||G||n . The second inequality follows from the matrix norm property ||AB|| ≤ ||A|| ||B||. . in general.6. Convergence. where R is the spectral radius of G. 3.86) from (3. solution The proof for n = 1 is Exercise 3.10 Derive the stability condition (3. ei2 . Let n = 2 and let V denote the set of right eigenvectors ei = (ei1 .8 3. Furthermore. max |GGx| |Gx| x∈V |Gx| |x|    = max Gλi ei  λi ei  i |λi ei | |ei | = max i |λi |   Gei  = |λi | |ei | max 2 i |λi | max 2 i |λi | = R2 =   |λi | ei  |ei | The proof for the general case follows by induction. where A and B are matrices (see the Endnotes).

we are usually not in the pitiful state of having to interpolate on a table of its values! William Press et al.9) is  dQi + dt Fi+ 12 − Fi− 12  Δx =0 (4. Fi± 1 = Fi± 1 2 2 73 .1 Introduction The semi-discrete form of the Euler equations (3. almost completely useless in day-today work: if we know enough about our function to apply a theorem of any power. (1986) 4. For a one-dimensional flow. These theorems are.1) where Qi (t) is the cell averaged vector of dependent variables. alas.4 Reconstruction There is an extensive mathematical literature devoted to theorems about what sort of functions can be well approximated by which interpolating functions. Qi (t) = 1 Vi  Vi Fi± 12 is the spatial flux quadrature Fi± 12 = 1 Ai± 12 Q dxdy  Fdy xi± 1 2 and Ai± 12 = Δy is the surface area of the face at xi± 12 .

........ .. however. .. .......... ......... ..... ...... in turn...... ....... ....... ......... ... .5 0. .... ....... . .... ... ......... ... ....... ... ..... . ............. . .. The algorithm for the fluxes Fi± 12 must take this discontinuous behavior into consideration. the periodic function Q(x) = sin x for 0 ≤ x ≤ 2π..... .... .. ....... . .... .. .. that a discontinuity may exist at each cell interface... Methods for computing Fi+ 12 are described in Chapters 5 and 6. Three methods are presented in the following sections. .... ...... ... 1....................... 1 2 3 4 5 6 ....... .... This method leads to excessive numerical diffusion. ..... ......... .... .......... . the approximation improves as the size of the cell is reduced........... . 4......... ......... ...... ...... .. ...... the fluxes Fi± 12 . ... .. Within a given cell...... .... . . ...... which....... ...... . ... ..... . for example. . .... .. t).. . which is first-order accurate......... The discretization of the domain (Section 3.. however..e... . . . .. .. ..... .. a local approximate reconstruction Qi (x) of the exact function Q(x) can be formed to compute the fluxes...... .......0 cell average exact solution . . The simplest reconstruction is Qi (x) = Qi . . .. ..... . ... . Q and Qi .... . ........ .....0 −0. .. ... .... .... . .. Of course. ... .1).... and is generally not acceptable. ......... ......... . .. .. The exact function Q(x) and cell averaged values Qi are shown in Fig....... ....... ........... Exact function Q(x) = sin x and cell averaged Qi The time evolution of Qi requires.. .. ... .1... ...... ..... ........... .. ..... .74 Reconstruction where F(Q) is given by (2.. Qi (x) is a function defined in cell i that is in general not a constant. ....5)... the cell averaged value is only an approximation of the exact function Q. .... Assume ten cells are employed.... ... Qi (x) is constructed to provide a closer approximation to Q(x) than is afforded by Qi ...3) and introduction of the volume averaged vector Qi (t) result in a loss of information regarding Q(x.. .. . . .......... ... Within each 2 cell i. . .. x Fig. ... i... .. . ...† Note.... ........ .. .. .... . .... . ...... . ....... 4..0 0. † Note that Qi is the cell averaged value in cell i and is therefore a constant.. . ... .. We therefore seek reconstruction methods of higher order accuracy.5 −1. ..... .. ...... must be computed using the Qi in the vicinity of xi± 1 . .. . Qi (xi+ 12 ) = Qi+1 (xi+ 12 )...... . from (4.. .... Consider............. .. .. ... .... .... .... .... ..... .1...... .. ...

Qi . . .......... pressure... .. and density. § It is also possible to interpolate a different set of variables. Note that I(x) is defined for x within the ‡ Earlier... 4. and Qi+1 to reconstruct the solution in cell i... . . 4. .. Van Leer (1979) developed a second-order accurate method using piecewise linear reconstruction with Godunov’s method.. .2) 2 where Δx is assumed constant.1. For 2 cell i... A similar reconstruction is performed for all cells.4...1. .. x xi+ 32 Fig. the polynomial is employed to estimate the value of Q denoted by Qli+ 1 (Fig... .2 Reconstruction Using the Primitive Function Colella and Woodward (1984) and Harten et al. .. i xi− 12 xi− 32 i+1 xi+ 12 .... We define a primitive function I(x) according to  I(x) = x Q dx xi− 3 for xi− 32 ≤ x ≤ xi+ 32 (4.2). .. In each cell i.. while on the left 2 face at xi+ 12 .... e. . On the right face of cell i at xi− 12 . the flux Fi+ 12 is then determined using Qli+ 1 and Qri+ 1 ... a polynomial is constructed to approximate Q. (1987a) developed methods for higher order reconstruction‡.... . the polynomial is employed to estimate the value§ of Q denoted by Qri− 1 . ..... .2. velocity... This is summarized 2 2 in Table 4. . .. right face at xi− 12 i−1 left face at xi+ 12 . Reconstruction of Q for Computing Fi± 12 Quantity Use reconstruction in cell Qli− 1 2 Qri− 1 2 Qli+ 1 2 Qri+ 1 i−1 i i i+1 2 We present a particular form of a reconstruction method that employs the cell averaged values Qi−1 ... .. and similarly 2 2 the flux Fi− 12 is then determined using Qli− 1 and Qri− 1 . .2 Reconstruction Using the Primitive Function 75 4....g. .. Reconstruction to interior faces of cell i Table 4..

xi− 12 . xi− 12 ] xi+ 12 − xi− 32 I[xi− 1 . xi+ 1 . xi− 12 . xi+ 3 ] − I[xi− 3 .8) . xi− 1 ] 2 2 a2 = I[xi− 32 . and xi+ 32 . xi− 12 .7)  (x−xi− 32 )(x−xi− 12 )(x−xi+ 12 )(x−xi+ 32 ) d4 I  4! where x ˆ depends on x and xi− 32 ≤ x ˆ ≤ xi+ 32 . xi+ 12 . xi+ 12 . xi+ 1 ] 2 2 2 xi+ 23 − xi− 32 2 2 2 (4. i. xi+ 12 ] − I[xi− 32 . xi− 1 . xi+ 12 ] = I[xi− 32 .76 Reconstruction limits of the three cells i−1. Then I(xi− 32 ) = 0 I(xi− 12 ) = Δx Qi−1 I(xi+ 12 ) = Δx (Qi−1 + Qi ) I(xi+ 32 ) = Δx (Qi−1 + Qi + Qi+1 ) (4.  dx4 x=ˆx (4.5) where I[xi− 32 ] = I(xi− 32 ) I[xi− 32 . xi− 12 .4) The coefficients are defined by a0 = I[xi− 32 ] a1 = I[xi− 3 . 1998) P (x) = a0 + a1 (x − xi− 32 ) + a2 (x − xi− 32 )(x − xi− 12 ) + a3 (x − xi− 32 )(x − xi− 21 )(x − xi+ 12 ) (4.3) There is a unique third-order polynomial P (x) that interpolates I(x) at the four points xi− 32 . and i+1. The polynomial may be obtained using Newton’s formula (Mathews and Fink. xi− 12 . xi+ 12 ] a3 = I[xi− 32 . xi+ 32 ] (4. xi− 12 ] = I[xi− 32 . xi+ 12 . xi+ 32 ] = I[xi− 12 ] − I[xi− 32 ] xi− 12 − xi− 32 I[xi− 12 .6) Now I(x) = P (x) + E(x) for xi− 32 ≤ x ≤ xi+ 32 where E(x) is the error defined by E(x) = (4.

2 Reconstruction Using the Primitive Function 77 The reconstruction function† for Q in cell i is denoted as Q(x) and is defined by Qi (x) = dP dx Thus for xi− 12 ≤ x ≤ xi+ 12  (4.10) Defining ΔQi+ 12 = Qi+1 − Qi (4. xi+ 12 ] = (2Δx)−1 (Qi − Qi−1 ) I[xi− 12 .3). I[xi− 32 .14) † Recall that Qi is the cell averaged value in cell i and is therefore a constant.5). and Qi (x) is a function defined in cell i that is in general not a constant. xi+ 32 ] = Qi+1 I[xi− 32 . and (4. xi− 12 ] = Qi−1 I[xi− 12 . xi+ 12 . . xi+ 32 ] = (2Δx)−1 (Qi+1 − Qi ) (4. xi− 12 . (4.9)  Qi (x) = a1 + a2 (x−xi− 12 ) + (x−xi− 32 ) +  a3 (x−xi− 1 )(x−xi+ 1 ) + (x−xi− 3 )(x−xi+ 1 )+ 2 2 2 2  (x−xi− 32 )(x−xi− 12 ) Using (4.11) the coefficients are a1 = Qi−1 a2 = (2Δx)−1 ΔQi− 21 a3 =  6(Δx)2 −1  ΔQi+ 12 − ΔQi− 21  (4.6).4. xi+ 12 ] = Qi I[xi+ 12 .13) and then Qi (x) = Qi − + (ΔQi+ 21 − ΔQi− 12 ) (ΔQi+ 21 24 − ΔQi− 21 ) 2(Δx)2 ξ2 + (ΔQi+ 21 + ΔQi− 12 ) 2Δx ξ (4.12) Define ξ = x − xi (4.

i−1 xi− 32 i i+1 xi− 12 xi+ 12 . 4.15) Therefore... 4..... From (4.3..4. cells used to define Qi (x) for cell i ...4... Therefore. .. cells used to define Fi− 12 .. . Qi and Qi+1 .....2). and similarly for cell i+1.....78 Reconstruction The domain of dependence of Qi (x) for cell i is shown in Fig. By replacing i by i−1... (4. ..... dE dI − dx dx dE = Q(x) − dx Qi (x) = From (4. i+2 xi+ 32 xi+ 52 Fig.. x .. and (4.... cells used to define Fi+ 12 i−1 xi− 32 i xi− 12 i+1 xi+ 12 .. 4. Dependence of Fi− 12 and Fi+ 12 . ..... the computation of the flux contributions to cell i involve cells i−2 to i+2 as shown in Fig... 4...14) can be applied to define the reconstruction within cell i. Qi (x) is a third-order accurate reconstruction of Q within cell i. .8). . .. x xi+ 32 Fig.3... it is evident that dE/dx = O(Δx3 )... . Thus..7).. .. Qli+ 1 = Qi Qri+ 1 2 = Qi+1 − 13 ΔQi+ 21 − 16 ΔQi+ 32 Qli− 1 = Qi−1 + 13 ΔQi− 21 + 16 ΔQi− 32 Qri− 1 = Qi 2 2 2 + 13 ΔQi+ 21 + 16 ΔQi− 12 − 13 ΔQi− 21 − 16 ΔQi+ 12 (4.9)... Cells used to define Qi (x) From (4.14) it is evident that Qi (x) depends on Qi−1 . the reconstruction within cell i−1 is obtained. i−2 xi− 52 ... . Equation (4....

...... for cell i..... 4...17) Q and Qi (x) ..... . ....... . ... .4. . ... .... . ....14)..... 4.... .18) Ten cells are employed. .. .. .. ........... . ........ The function is reconstructed within . ....... . ............ .... . . .. . ... ....... . .     1  3 2 4 5  6  .. .. .... ... .... ........... .... .. .3 No New Extrema 79 An example of the reconstruction of a scalar function is presented in Fig...... . . Ten cells are employed. . . .... ..... The function is reconstructed within each cell i according to (4. .. .. ..... .... ...... . with the symbols  and  indicating the values of Qri− 1 2 and Qli+ 1 ....0  −0... . ..........5. . x    Fig.......6... .. ..5. .. .. . ....... 1.. By denoting δQi+ 1 = Qli+ 1 − Qri+ 1 2 2 we have δQi+ 12 =  1 6 (4.0   reconstructed solution exact solution .... ... ... ...... .. . . . .. 4... It is evident that the reconstructed func2 tion Qi (x) is a close approximation to the exact function Q(x) within each cell. .... ... The exact function is  Q(x) = −1 for x ≤ π +1 for π < x (4.... .......... ....3 No New Extrema A problem arises with the reconstruction (4.11). .. ........... ...... .. respectively..... .. .. ....14) when the function Q exhibits a discontinuity as illustrated in Fig....... . .. . ...... ..... ... ....0 0... ... . with the discontinuity coinciding with the boundary separating the fifth and sixth cells... Reconstruction of sine using ten cells 4....16) 2 ΔQi+ 23 − 2ΔQi+ 12 + ΔQi− 21  and using (4.. ........5 −1. ..... There is a small jump in Qi (x) evident at the face of each cell. ... ..... . ... ..... ... δQi+ 12 = 1 6 (Qi+2 − 3Qi+1 + 3Qi − Qi−1 ) (4.. .... ....... ..... ..... The exact function is Q(x) = sin x and it is assumed periodic...... ... .. .....5 0.

.....0  reconstructed exact 0.... When cell i is immediately to the left of the discontinuity.. ........... .... with the symbols  and  indicating the values of Qri− 1 and Qli+ 1 ........ . . δQ5 12 = 1 3 (QL − QR ) noting that Qi = −1 for i = 1 to 5 and Qi = +1 for i = 6 to 10... .....5 ....0  −1.... ... . ... ...... ..6 can be identified using Fig..... Using (4.. 1... ... the reconstruction algorithm (4.. ........ The reconstruction has introduced fictitious local extrema at x4 12 and x6 12 .... 4. ...17).. Q and Qi (x) ............. ...... . .... ............. ........ 1  2   3  4 5 6 ....... .. ......5 0. −1........... It is clearly insufficient to simply require a given formal order ... the reconstruction yields a variation of Qi (x) with x in the fifth and sixth cells while the exact function Q is a constant.................. . .. ... .... Similarly.. ... x   Fig...........0 −0.. .. .................. . .. .....80 Reconstruction each cell i according to (4........ . . 4. A correct reconstruction would only employ data from the right......... ... ... ......... 2 2 It is evident that the reconstruction (4.............. ... .....14)............... Finally....... the reconstruction algorithm (4..... ....... respectively......... ......... Reconstruction using ten cells The criteria for developing reconstruction algorithms must therefore be expanded........... The source of the problem in Fig.14) is a poor approximation of the exact function Q within the fifth and sixth cells... . .. .. ....... .. for cell i............. ............ ... .... 4. .......... ...14) employs data from both the left and the right of the discontinuity...14) employs data from both the left and the right of the discontinuity......... ................ ........ A correct reconstruction would only employ data from the left.3.. when cell i is immediately to the right of the discontiuity..5       ...... The jump δQ5 12 at the discontinuity (x = π) is incorrect........5   1..6....... ... ..

15) require modification in the presence of discontinuities in Q.15) can be expressed in a general form for cell i as Qli+ 1 2 = Qi + 1 4  (1 − κ)ΔQi− 12 + (1 + κ)ΔQi+ 12  † The constants a and b should be as small as possible. The constants a and b define the neighborhood of cells surrounding (and including) cell i that are utilized in the determination of the local extrema (local maximum and minimum). This approach assumes that the exact function Q possesses continuous derivatives up to some order depending on the order of reconstruction. unphysical oscillations) in the reconstruction. a shock or contact surface in gas dynamics). the exact function Q is not known. and therefore detection of the discontinuity is considerably more difficult.. Thus.4 Modified Upwind Scheme for Conservation Laws 81 of accuracy of reconstruction. . . . In principle. We may restrict our attention to the face values since they determine the flux at the cell face given the flux quadrature algorithm.22 later. . Qi . . Qi+b . Qi−a+1 . Qli+ 1 = Qri− 1 = Qi correspond to a first-order reconstruction that is excessively 2 2 diffusive.g. however. .† The modification to a reconstruction algorithm to incorporate this principle is denoted a limiter. 4. the relevant range is a ≥ 1 and b ≥ 1. An counterexample is presented in Fig. A straightforward approach is to avoid the formation of new local extrema relative to the cell averaged solutions in some neighborhood of cell i. This approach is denoted No New Extrema and may be summarized as follows: No New Extrema (NNE): The reconstructed left and right states Qli+ 1 and Qri− 1 shall not introduce any new extrema 2 2 relative to Qi−a . 4. that the NNE principle does not guarantee that a left or right state is correctly reconstructed at a discontinuity in Q. in solving the Euler equations. . Qi+b−1 .4 Modified Upwind Scheme for Conservation Laws The reconstruction function (4.. The minimum values a = b = 0.4. An example is the Modified Upwind Scheme for Conservation Laws (MUSCL) described in the following section.. However. Equations (4.e. This assumption is not valid when Q manifests a discontinuity (e. since the objective is to avoid the formation of new local extrema (i. where a ≥ 0 and b ≥ 0.14) and face values (4. the reconstruction algorithm could be revised to incorporate an appropriate directional bias in the vicinity of a discontinuity.. It should be emphasized.

. ... i+1 xi+ 12 xi+ 32 Fig. ..... .. and Qi+1 .. . Qi . . ... .. 2 i−1 xi− 32 i xi− 12 x 2 i+1 xi+ 12 ... . as shown in Fig. 4..... and similarly.. 2 which are to the left of xi+ 12 ... ........e.... .. . as shown in Fig.. and thus Qri− 1 (note right face) employs two cells on 2 the right of xi− 12 and one cell to the left.. . . . . Equation (4.. ... .... .. which 2 are centered on xi+ 12 .. .... The dependence of Qli+ 1 and Qri− 1 on the adjacent cells is 2 2 shown in Figs..19) where κ = 13 . ......9. 4. Similarly.82 Qri− 1 2 = Qi − 1 4 Reconstruction  (1 − κ)ΔQi+ 12 + (1 + κ)ΔQi− 12  (4. 4..... .... ..19) is “upwind-biased” since Qli+ 1 2 depends on Qi−1 ... the dependence of Qli+ 1 and Qri− 1 on 2 2 the adjacent cells is the same as shown in Figs. ..11......... ...... ..19) is a second-order upwind-biased reconstruction if κ = 0 (Exercise 4. .....7 and 4.2.. .. . Qri− 1 depends 2 on Qi−1 .. x xi+ 32 Fig.......10.7 and 4. Equation (4.... cells used to define Qli+ 1 ....1).. It is a “centered” reconstruction..7. 4... i+ 12 . i−1 xi− 32 xi− 12 Ql .. 1 3 cells used to define Qri− 1 . i... In this case.. as shown in Fig..... .. . 2 ... . .... .8...3). . and similarly Qri− 1 depends 2 only on Qi−1 and Qi .... . . . .. ............ i... i .. .. Qri− 1 2 depends only on Qi and Qi+1 . .... .... .. .8...... .... ... Qli+ 1 depends only on Qi and Qi−1 ...... . . . . . and thus Qli+ 1 (note left face) employs two 2 cells on the left of xi+ 12 and one cell to the right... . ......e........ . 4.... ... ... . . .... .... . .... .12....... ..... .....4).. Equation (4. which are to the right of xi− 12 ..19) is a second-order reconstruction if κ = −1 (Exercise 4. .. 4.. Equation (4.. ..... as illustrated in Fig. Cells used to define Qli+ 1 for κ = 2 Qri− 1..19) is also a second-order reconstruction if κ = +1 (Exercise 4...... ........... ... ... and Qi+1 ..... .. Qli+ 1 depends only on Qi and Qi+1 . Cells used to define Qri− 1 for κ = 2 1 3 There are three additional interesting values for κ that are summarized in Table 4.. 4...8. which are centered on xi− 12 . ........ .. 4.. Qi . ...... The above formulas can be applied to each cell and thereby define the left and right face values for every face. It is “upwind”....

..... 2 i−1 xi− 32 cells used to define Qli+ 1 . 2 ........ ... ..9.. . .. . ..... ..... . . ... i+1 xi− 12 xi+ 12 x xi+ 32 Fig... .. x xi+ 32 Fig.... .......... . ....... xi+ 32 Fig....... ... .......... .... . ... ...... ...... ... . ..... 4. . Cells used to define Qri− 1 for κ = −1 2 Qli+ 1 .. x xi+ 32 xi+ 12 Fig. .... . Cells used to define Qli+ 1 for κ = +1 2 cells used to define Qri− 1 . 4. .... i 2 ..... ... ....... . .... ..... ...2. . ........ .... 2 i−1 xi− 32 2 i+1 i xi− 12 xi+ 12 . . .... ... .. . . .. ........... ..... . . .. Cells used to define Qli+ 1 for κ = −1 2 Qri− 1 cells used to define Qri− 1 .. . . .. Cells used to define Qri− 1 for κ = +1 2 x .11.. .. ...... . . . ... . .. ........ . ... .4 Modified Upwind Scheme for Conservation Laws 83 Table 4..12.. ........ . ...... .. .... 2 . .. ........ .. i−1 xi− 32 i xi− 12 2 i+1 .... 4....... ... ... . . .. ... . . . . . .. . ..... .... . ....10..4.. Reconstruction κ Order Definition nd 1 2 3rd 2nd 2nd 1 3 0 −1 Centered Upwind-biased Upwind-biased Upwind cells used to define Qli+ 1 Qli+ 1 ......... ... .. .. . ... i−1 xi− 32 i xi− 12 Qri− 1 2 i+1 xi+ 12 .... ... .......... ........... . . .. .... ... .. 4........ . . . .. ... .... . ...... .... ... .. .

23) ≥ min(Qi−1 .21) Qli+ 1 ≥ min(Qi−1 .25) 1 = ΔQ 1 .1977b. provided ΔQi− 1 ≤ b ΔQi+ 1 .e. Qi . (1986) modified (4. Assume that ΔQ i± 2 i± 2 becomes ΔQi− 12 ≤ b ΔQi+ 12 where b is a positive constant defined for κ < 1 by‡ b= 3−κ 1−κ (4. Qi+1 ) (4.20).1 Case 1: ΔQi+ 21 ≥ 0. Qi . a new minimum or maximum) relative to Qi−1 .4.2.1979) developed a combined reconstruction-evolution method entitled Monotone Upwind Schemes for Scalar Conservation Laws denoted by the same acronym (MUSCL). 4. When Thus. Qi+1 ) (4.24) 2 2 This approach is denoted Modified Upwind Scheme for Conservation Laws (MUSCL).13. we can assign ΔQ i± 2 i± 2 2 2 ΔQi− 21 > b ΔQi+ 21 . Qi .21) is satisfied automatically. Van Leer (1977a. This implies four constraints: Qli+ 1 ≤ max(Qi−1 . we write ΔQi− 21 = b ΔQi+ 12 + δ † Earlier. ΔQi− 21 ≥ 0 The relative values of Qi−1 . then (4..84 Reconstruction 1: Anderson et al. . The first constraint (4.† There are four cases to consider for the values of κ in Table 4. Qi .22) Qri− 1 2 Qri− 1 2 ≤ max(Qi−1 . Qi .19) by replacing ΔQi± 12 with ΔQ i± Qli+ 1 = Qi + 1 4 Qri− 1 2 = Qi − 1 4 2  1 + (1 + κ)ΔQ 1 (1 − κ)ΔQ i− i+ 2  (1 − κ)ΔQ i+ 12 2 + (1 + κ)ΔQ i− 12   2 (4. Qi and 2 2 Qi+1 . Hirsch (1988) uses the term Monotone Upstream-Centred Schemes for Conservation Laws.20) 1 = ΔQ 1 except when this would introduce new extrema in where ΔQ i± 2 i± 2 Qli+ 1 or Qri− 1 (i. 4. Then the above condition using (4. ‡ If κ = 1. Qi+1 ) (4.21) yields Qi + 1 4   1 + (1 + κ)ΔQ 1 ≤ Qi+1 (1 − κ)ΔQ i− i+ 2 2 1 = ΔQ 1 . and Qi+1 are illustrated in Fig. Qi+1 ) (4.

2 The third constraint (4.27) .21) is satisfied provided Qli+ 1 = Qi+1 + 2  ΔQi− 12 b ΔQi+ 12 1 ΔQ i− = 1 ΔQ i+ = ΔQi+ 21 2 2 if ΔQi− 21 ≤ b ΔQi+ 12 if ΔQi− 21 > b ΔQi+ 12 (4. the first constraint (4.22) yields Qi + 1 4   1 + (1 + κ)ΔQ 1 ≥ Qi−1 (1 − κ)ΔQ i− i+ 2 2 1 = ΔQ 1 . this condition is satisfied. this constraint becomes. Substituting (1 − κ) δ 4 and δ must therefore be zero. for κ < 1. Therefore.24) yields Qi − 1 4   1 + (1 + κ)ΔQ 1 ≥ Qi−1 (1 − κ)ΔQ i+ i− 2 2 1 = ΔQ 1 . we obtain 1 ΔQ i− = ΔQi− 12 ΔQ = 2  i+ 12 ΔQi+ 21 b ΔQi− 21 1 = ΔQ 1 .4 Modified Upwind Scheme for Conservation Laws 85 where δ ≥ 0.26) 1 = ΔQ 1 .23) yields Qi − 1 4   1 + (1 + κ)ΔQ 1 ≤ Qi+1 (1 − κ)ΔQ i+ i− 2 2 1 = ΔQ 1 . If κ = 1. then ΔQ i± 2 i± 2 The second constraint (4. this constraint becomes Assuming ΔQ i± 2 i± 2 (5 − κ)ΔQi− 12 + (1 + κ)ΔQi+ 12 ≥ 0 Since ΔQi± 1 ≥ 0. If κ = 1. Assuming ΔQ i± 2 i± 2 ΔQi+ 1 ≤ b ΔQi− 1 2 2 By a similar argument as employed for the first constraint. this constraint becomes Assuming ΔQ i± 2 i± 2 (5 − κ)ΔQi+ 12 + (1 + κ)ΔQi− 12 ≥ 0 which is satisfied. The fourth constraint (4. then ΔQ i± 2 i± 2 if ΔQi+ 12 ≤ b ΔQi− 21 if ΔQi+ 12 > b ΔQi− 21 (4.4.

16 for and ΔQ i± 2 i± 2 κ < 1. . . . 4.. .. . .... ..14 to 4.. ......... .. . . ... .. ..... i− 12 ...... x xi+ 12 xi− 12 Fig. ... ...... i−1 ..... ... .. .. . ΔQi+ 12 . .. .. ..  ΔQ i− 21 = 1 ΔQ i+ = 2  ΔQi− 12 b ΔQi+ 12 if ΔQi− 21 ≤ b ΔQi+ 12 if ΔQi− 21 > b ΔQi+ 12 ΔQi+ 12 b ΔQi− 12 if ΔQi+ 21 ≤ b ΔQi− 12 if ΔQi+ 21 > b ΔQi− 12 (4.. .. .. ....... 4.... .... . .. . ....... ... for κ < 1. . .. . .. ....... .......... .. ...... .... ..... .... This is illustrated in Figs. ...... ...26) and (4. ... ΔQi− 12 = b−1 ΔQi+ 12 . ... .... . . .. .... .... . .. .... ........14.. .... . .. . ΔQ ΔQ ΔQ ΔQ = ΔQ = bΔQ = ΔQ = ΔQ ΔQ ΔQ ΔQi− 12 = bΔQi+ 12 = bΔQi+ 12 = ΔQi+ 12 ΔQi− 12 i+ 1 and ΔQ i− 1 for Case 1 Fig........ ..... i i+1 .... ... .....28) 1 = ΔQ 1 for κ = 1. .... ΔQi− 12 ≥ 0 Combining (4.. .. i− 12 .. ......... .. ...... .. ..... .. ......... . ..86 Reconstruction Q .... .... .. . .... i− 12 ... ... . i− 12 i− 12 ....... ........ i− 12 i+ 12 . . . ..... ..13.. ..... . . . ....... . . ... ΔQi+ 12 ≥ 0. .. ..... ... . .... ..... .. . ........ .. ......... ....... . .... .....27) yields.. .... .. .. .. ... .. .. 4. . .. .. ... . i+ 12 i+ 12 .. . .. . ... i+ 12 .... ..... . .... .. . ...... .... ΔQ 2 2 . .. ... . ..... . . ..

. For κ = −1.... . .... .... ..... ...... which is shown for ΔQi+ 12 > −ΔQi− 21 .. ...... . ..... ......... ............ . ........ .... ....... .. ...... . .... ... . ... ..... ... .... . ......... ............... . .... .. . ...... ......... ... ..... . . .......... ....... ......... .... ....... ... . ... ........ ....... . .22) and (4. ... ..... ..... ....... . ........ . ....... ... .... . . ..... .. ... . .... ...... ... . . . i+ 2 .... .... ... .. .. .... .. ............ ..... .... ... .. . .. ... ..... ... 4. ..... ... . . .. . ΔQi− 12 i− 1 for Case 1 Fig.... .... ... ...... ... ...... . . .. . .............. . .. . .... ............... ........ ... ............ .... ... .. .. . ... .. . .. ......4. . ........ .. . ........ ......... ... . . .. .. ... .... ΔQ ...... .... .. .... ..... ..... . ............. .... .. .... .... . . .. .... . .2 Case 2: ΔQi+ 21 ≥ 0.. ... ..... .. ... ... ... ....... ... . ..... .. . . .. .. .. . .. ... ...... ... 4. ....... ...... ..... ....... .... .... . ........ ............. ... ............... .. . . ..... .... ......... ....... i+ 12 ... . . ..2.. ... ....... ..... ....... . . ....... ... ... ΔQi− 12 i+ 1 for Case 1 Fig. .. ... ........ . . .. . . . . ..... ... ..... .. ...... .... .. .. . . ... . ... .. . .... . ....... ............ . .. ..... ... .... .... ... ...... . .. .. ... .. .. .. ... . . . ........ . ........ ... . .. .. ..... . .20) is Qli+ 1 2 1 = Qi + 12 ΔQ i− Qri+ 1 2 1 2 ΔQi+ 12 2 = Qi − 1 = ΔQ 1 ..... . . ... . ..... .... . .. ..... . . . ... ... . ... . . ...17. ... .. . ...... .. .... ... . ... ... ....... .............. ......4. .... .. ... ...... ..... ..... .. . ......... .. .. ... ... ..... ........... ......... . .. .......... ...... ΔQi− 21 ≤ 0 The relative values are illustrated in Fig... .... ... . ....... ....... . . .. . ... . . . .. ...... ..... . .. ... .. ... .... .... . ........ .. .. . . ............... . ... ...... . .. .. .... ...... ... ..... .... .. .. ... . . . ........ . . ...... ........ ...... .. .15.... ..... ..... ......... .. . ..... .. .... . . . ... ...... ........ ... .... .. ... ... . ΔQ 2 4... ....... .. ... ..... . ...... .......... .... .... ....... .. . ...... ........ .... ...... .... .... ..... .. ........ . ............ . .... ... . . . ...... ..... .. .. .. . .. . .. ... .. .. . ........ ......... .. .............. .... . ... .. ...... We consider the four values of κ in Table 4...... .. ........ .. . ........ ........ . ... . .. . ... .. ... . ... ............. . . ... .. .... . . ΔQ ....... ..... ...... .......... ... . ...... ..... ........................ . . . ....24) imply Assume that ΔQ i± 2 i± 2 ΔQi− 12 ΔQi+ 1 2 ≥ 0 ≤ 0 . ..... . ... ... ........ . ...... ... ... ....... ..... ..... . .... . . ... ΔQ 2 1 ΔQ i+ 2 .. ..... .. .. . . . ..... ........... ... . ........... ......... .. . .... . ......... ..... .......... ... ... ...... .... ....... ..... . ........ ....4 Modified Upwind Scheme for Conservation Laws 87 i− 1 ΔQ 2 . . .. ... ....... .......... .. .... .. . . ...... . .... .. .... .................. . ... .... .. . 1 .... . ........ .. .... ........... ............. ... ......... .... . . ........ . ... .... . ..... .. Equations (4. .. . ..... .... ... . .. .. . .... .. ... .... . .... ............. .. .. ..... ........ .... .... .. .. .... ..... . ...... ... ......... .. . ....... .... .. ...... .. . .... .. ....... .... . . .16......... ... . ......... ....... . . .. .. ..... ... . .. .. ........... .. . .... ... . . .. . ... .. .. ..... . Equation (4.. . . .... .... . ............ ... . .. .. ... . 4. ..

. ... . ...... ΔQi− 12 ≤ 0 which is opposite of the conditions for Case 2. Thus.. Equation (4.. Equation (4.17.. ..22) implies Assume that ΔQ i± 2 i± 2 ΔQi+ 21 + ΔQi− 12 ≥ 0 while Equation (4. i−1 . . .. . Thus. ..20) is Qli+ 1 = Qi + 1 6 Qri+ 1 2 = Qi − 1 6 2   1 + 2ΔQ 1 ΔQ i− i+ 2 ΔQ i+ 12 2 + 2ΔQ   i− 12 1 = ΔQ 1 except where a limiter must be employed.... . xi+ 12 xi− 12 Fig... i i+1 . .. Equation (4. . ... 4....... .. . we take 1 ΔQ i− = 0 1 ΔQ i+ = 0 2 2 For κ = 13 ... .. . . .. ... ..20) is Qli+ 1 = Qi + 1 4 Qri+ 1 2 = Qi − 1 4 2  1 1 + ΔQ ΔQ i+ i− 2  ΔQ i+ 12 2 + ΔQ   i− 12 1 = ΔQ 1 . .. ... . . ΔQi+ 12 ≥ 0.... . . .. . . ... . we take 1 ΔQ i− = 0 1 ΔQ i+ = 0 2 2 For κ = 0. . ... Assume that ΔQ i± 2 i± 2 .. . .. ...88 Reconstruction Q . . ... . . . .. . x . .24) implies ΔQi+ 21 + ΔQi− 12 ≤ 0 which are in general contradictory..

4.18.20) is Qli+ 1 2 1 = Qi + 12 ΔQ i+ Qri+ 1 2 1 2 ΔQi− 12 2 = Qi − 1 = ΔQ 1 . The analysis is similar to Case 1. 4. For κ = −1 and 0. ΔQi− 21 ≥ 0 The relative values are illustrated in Fig.24).21) to (4.29) 1 = ΔQ 1 for κ = 1. and ΔQ i± 2 i± 2 4. Equation (4.4 Case 4: ΔQi+ 21 ≤ 0.4 Modified Upwind Scheme for Conservation Laws 89 The following definitions satisfy Equations (4.24):  1 ΔQ i− = 2  ΔQ = i+ 12 ΔQi− 12 −2 ΔQi+ 12 if ΔQi− 21 ≥ −2 ΔQi+ 12 if ΔQi− 21 < −2 ΔQi+ 12 ΔQi+ 12 −2 ΔQi− 12 if ΔQi+ 21 ≤ −2 ΔQi− 12 if ΔQi+ 21 > −2 ΔQi− 12 For κ = 1.19.21) to (4. The analysis is similar to Case 2.3 Case 3: ΔQi+ 21 ≤ 0.4.4. ΔQi− 21 ≤ 0 The relative values are illustrated in Fig. 4.4. the result is 1 ΔQ i− = 0 1 ΔQ i+ = 0 2 2 . Then Assume that ΔQ i± 2 i± 2 Qli+ 1 2 = Qi + 12 ΔQi+ 21 Qri+ 1 = Qi − 12 ΔQi− 21 2 which satisfy the conditions in Equations (4. The result for κ < 1 is  1 ΔQ i− 2 ΔQ i+ 21 =  = ΔQi− 12 b ΔQi+ 12 if ΔQi− 21 ≥ b ΔQi+ 12 if ΔQi− 21 < b ΔQi+ 12 ΔQi+ 12 b ΔQi− 12 if ΔQi+ 21 ≥ b ΔQi− 12 if ΔQi+ 21 < b ΔQi− 12 (4.

.. . . .. .. ... .. .18. ...... . .... .. .. .5 Summary The MUSCL scheme is Qli+ 1 2 = Qi + 1 4  1 + (1 + κ)ΔQ 1 (1 − κ)ΔQ i− i+ 2 2  .. .19.... ΔQi− 12 ≤ 0 For κ = 13 . i i+1 . i−1 ... . .. .. ..90 Reconstruction Q .. .. .. . .. i−1 .... ... . . . ... 4... . .... . .. ... . ΔQi+ 12 ≤ 0.. . . .. .. .. . .. i+1 .. xi− 12 i ...... .. .... 4. the result is 1 ΔQ i− 2 ΔQ i+ 12  =  = ΔQi− 12 −2 ΔQi+ 12 if ΔQi− 21 ≤ −2 ΔQi+ 12 if ΔQi− 21 > −2 ΔQi+ 12 ΔQi+ 12 −2 ΔQi− 12 if ΔQi+ 21 ≥ −2 ΔQi− 12 if ΔQi+ 21 < −2 ΔQi− 12 For κ = 1.. . .. . .... the result is 1 ΔQ i+ 2 = ΔQi+ 12 1 ΔQ i− 2 = ΔQi− 12 Q ... .... . ....... .. . . x xi+ 12 xi− 12 Fig.. . . . ..... ... . .. ... . ... ΔQi+ 12 ≤ 0. ... ... .. . . . .. . . x xi+ 12 Fig.4. . .. .. .. .... ... ΔQi− 12 ≥ 0 4. . . ... .. .. .. ... . . .. .

7). ΔQ For κ = if ΔQi− 21 ≤ b ΔQi+ 12 if ΔQi− 21 > b ΔQi+ 12 with ΔQi+ 21 ≥ 0 and ΔQi− 21 ≤ 0.  ΔQ i− 12 = 1 ΔQ i+ = 2  ΔQi− 12 −2 ΔQi+ 12 if ΔQi− 21 ≤ −2 ΔQi+ 12 if ΔQi− 21 > −2 ΔQi+ 12 ΔQi+ 12 −2 ΔQi− 12 if ΔQi+ 21 ≥ −2 ΔQi− 12 if ΔQi+ 21 < −2 ΔQi− 12 For κ = 0 and −1 (Exercise 4. ΔQi+ 21 ) . 1 ΔQ i− 2 = ΔQi− 12 1 ΔQ i+ 2 = ΔQi+ 12 (4.  For κ = ΔQi− 12 b ΔQi+ 12 ΔQi− 12 b ΔQi+ 12 if ΔQi− 21 ≥ b ΔQi+ 12 if ΔQi− 21 < b ΔQi+ 12 ΔQi+ 12 b ΔQi− 12 if ΔQi+ 21 ≥ b ΔQi− 12 if ΔQi+ 21 < b ΔQi− 12 with ΔQi+ 21 ≤ 0 and ΔQi− 21 ≥ 0.30) For κ = 1 3 with ΔQi+ 21 ≥ 0 and ΔQi− 21 ≥ 0. bΔQi+ 21 ) 1 ΔQ i+ 2 = minmod (bΔQi− 12 .4.4 Modified Upwind Scheme for Conservation Laws   1 + (1 + κ)ΔQ 1 Qr 1 = Qi − 1 (1 − κ)ΔQ i− 2 i+ 2 4 91 i− 2 For κ = 1.  1 ΔQ i− = 2  1 ΔQ i+ = 2 For κ = 1 3 1 ΔQ i− = 2  ΔQ = i+ 12 1 3 if ΔQi+ 21 ≤ b ΔQi− 12 if ΔQi+ 21 > b ΔQi− 12 ΔQi− 12 −2 ΔQi+ 12 if ΔQi− 21 ≥ −2 ΔQi+ 12 if ΔQi− 21 < −2 ΔQi+ 12 ΔQi+ 12 −2 ΔQi− 12 if ΔQi+ 21 ≤ −2 ΔQi− 12 if ΔQi+ 21 > −2 ΔQi− 12  i− 21 = 1 ΔQ i+ =  2 1 3 ΔQi+ 12 b ΔQi− 12 with ΔQi+ 21 ≤ 0 and ΔQi− 21 ≤ 0. 1 ΔQ i− 2 = minmod (ΔQi− 12 .

...... .. .. ...... ...92 Reconstruction where ⎧ ⎪ ⎨ x if |x| ≤ |y| where x and y have the same sign minmod (x... 4.0 .. ... .14). The limiter yields an accurate reconstruction of the left and right states within each cell including the discontinuity at x = π. ....... . and 1. y) = y if |x| > |y| where x and y have the same sign where x and y have opposite signs ⎪ ⎩ 0 4. 4. . ... ... .. ...22. 1 2  reconstructed solution exact solution .5 Essentially Non-Oscillatory Methods In Section 4.....18) with ten cells is shown in Fig.. 5 . . However. . ... . . ..... . . −1. 4........ ... as shown in Fig. the left and right states at the discontinuity are incorrectly reconstructed for κ = 1.. x .. 4.... ..........5 . ... 0... 4 ...0 . uses the cell average values Qi−1 .. 0. ...... ... and thus the limiter has no effect in this case. 3 .. 1..... .. ............... Qi .. . ... ...... given by Equation (4.. 0.. .20) with κ = 13 . 6..  .... 1 3 with a limiter The effect of the limiter on the reconstruction for the discontinuity (4.20. .. ..6 Results The effect of the limiter on the reconstruction for the sin wave with ten cells is shown in Fig... Fig..0  −0. ... ... ... The reconstructed values at the left and right sides of each face are identical to Fig. the reconstruction polynomial Qi (x) within cell i.. Q . .. ... ..... 4. .. .. ... 4. Reconstruction of sine using ten cells and κ = .. . .... .5 ... Similar results1 are obtained for κ = −1.20 using (4.. ......21 using (4......... .. .20) with κ = 13 .5......4... Identical results are obtained with κ = −1 and 0.2... and Qi+1 as illustrated ..

. .. . ....5 0.. ...... .. x  −1............ Reconstruction using ten cells and κ = 1 in Fig...0 0............. ... .......... .......... .5   reconstructed exact       . −1.... . ... .....0 −0.. ... . .... ...0   2   3  4 5 6 ........... ..... .......... . Reconstruction using ten cells and κ = 1 3 with a limiter Q . ...... ... ... 4...... ............. .. .......5 0.. .. 4.... ....0  1   3 2   4 5 6 .. .... ..4... .............5 Essentially Non-Oscillatory Methods 93 Q ... .5 1... 1 −1....... .. . ...... ................ for discontinuous functions. ................. .... ..... ...... .. ..5 1........ ..... . the use of a fixed symmetric stencil of cells in the vicinity of the discontinuity yields unphysical . . However.0 −0..5 Fig........5 Fig....5    reconstructed exact     . . ........ x  −1... ...... 1........ ... ... . .. .............. 4.... . .. 4.0 0. .5). ......g. .................... . .21.. ........ ..... . ... .... Fig.22.. .. ............ ....... . .3. .... ...... .... .. ............. ... For smooth functions...... ......... .. . . ..... .. . this leads to an accurate reconstruction within the cell i (e.. . This stencil of cells is symmetric about cell i and fixed........... . ... 1. .. ..... .. ... ..... ... ....

. . . x 5 −1.g..... .. .......... .. . .. . .. .... ...... ....... ....31) in regions where Q(x) is smooth.. . ... . The MUSCL scheme described in the previous section eliminates the unphysical extrema as indicated in Figs 4. .. . . . ... . .0 .6.. Fig.. ... .5 1. .... ...5 −1. Similarly. ... . .... .. .94 Reconstruction extrema (e. Discontinuous function with ten cells Harten et al..... .... .. . .. .. . . ... ... ... . .. Q .. . .23... .. ..... .... . First..... .. 1 1 2 .... .. .. ... .. .. . ... . . .. ... 9 5 10 6 ..... . . . . . . .. ........ 4. .... The explanation is seen in Fig. 4.. ..... . . .. .... . .. .. Reconstruction of the Q in cell number 5 should utilize only information from cell number 5 and cells to its left. ..  Qi = xi+ 1 2 xi− 1 Qi (x) dx (4.. ... .... .. . . . ..33) . .21 and 4... .... . . .5 Fig. .. ... .5 0.. (1987b). . ... . ... ........ . . .. . .0 −0. . .. Second. ... ...... . . (1987a) developed a reconstruction method that permits an asymmetric stencil of cells. . Qi (x) satisfies the Essentially Non-Oscillatory property T V (Qi (x)) ≤ T V (Q) + O(Δx3 ) (4. 1. .... ... .. reconstruction of Q in cell number 6 should utilize only information from cell number 6 and cells to its right. 3 2 3 4 7 8 4 . . . ... . Use of a symmetric stencil for reconstruction of Q within cells number 5 and 6 leads to unphysical extrema as indicated in Fig.. Q(x) = Q(x) + O(Δx3 ) (4. We develop the method for the case of a secondorder accurate reconstruction of Qli+ 1 and Qri− 1 in cell i. . .22 while retaining a symmetric stencil.... . Qi (x) reconstructs the cell average exactly.... .... ... . .. . .32) 2 Third. 4. ..6).. . .. . .. The reconstruction polynomial Qi (x) is required to satisfy three conditions. ... .. .. . .. .0 exact 0. .. 4.... . .. The extension 2 2 to higher order is described in Harten et al.. ... ..23. 6 .... ...

xi−a+ 12 ] xi−a+ 32 − xi−a− 12 . 1988)  xi+ 1 2 T V (Q) = xi− 1 2    dQ     dx  dx (4. Then. xi−a+ 12 ] a2 = I[xi−a− 1 .5 Essentially Non-Oscillatory Methods 95 where T V is the Total Variation of a function defined as (Hirsch.37) where I[xi−a− 12 ] = I(xi−a− 12 ) I[xi−a− 12 . xi−a+ 3 ] 2 2 2 a3 = I[xi−a− 12 . and i−a+2.34) and constitutes a measure of the oscillatory behavior of Qi (x) within cell i (see also Section 8.35) 2 Note that I(x) is defined for x within the limits of the three cells i−a. xi−a+ 32 ] − I[xi−a− 12 . xi−a+ 12 . i−a+1. xi−a+ 32 . xi−a+ 12 .36) There is a unique third-order polynomial P (x) that interpolates I(x) at the four points xi−a− 12 . xi−a+ 32 .2).4. Following Section 4. xi−a+ 12 . 1998) P (x) = a0 + a1 (x − xi−a− 12 ) + a2 (x − xi−a− 12 )(x − xi−a+ 12 ) + a3 (x − xi−a− 12 )(x − xi−a+ 12 )(x − xi−a+ 32 ) The coefficients are defined by a0 = I[xi−a− 12 ] a1 = I[xi−a− 12 . and xi−a+ 52 . similar to Equation (4. xi−a+ 32 ] = I[xi−a+ 12 ] − I[xi−a− 12 ] xi−a+ 12 − xi−a− 12 I[xi−a+ 12 . xi−a+ 52 ] (4.2) corresponds to a = 1 in Equation (4. I(xi−a− 12 ) = 0 I(xi−a+ 12 ) = Δx Qi−a I(xi−a+ 32 ) = Δx (Qi−a + Qi−a+1 ) I(xi−a+ 52 ) = Δx (Qi−a + Qi−a+1 + Qi−a+2 ) (4. and Equation (4.3).35). The polynomial may be obtained using Newton’s formula (Mathews and Fink. we define the primitive function  I(x) = x Q dx xi−a− 1 for xi−a− 12 ≤ x ≤ xi−a+ 52 (4. xi−a+ 1 .2. xi−a+ 12 ] = I[xi−a− 12 .

38) Now I(x) = P (x) + E(x) for xi−a− 12 ≤ x ≤ xi−a+ 52 where E(x) is the error defined by E(x) = (4.39)  (x−xi−a− 1 )(x−xi−a+ 1 )(x−xi−a+ 3 )(x−xi−a+ 5 ) d4 I  2 2 2 2 4!  dx4 x=ˆx (4. xi−a+ 32 . and (4. xi−a+ 12 . the coefficients are a1 = Qi−a a2 = (2Δx)−1 ΔQi−a+ 12 a3 =  6(Δx)2 −1  ΔQi−a+ 32 − ΔQi−a+ 12  (4. xi−a+ 52 ] = Qi−a+2 I[xi−a− 12 . xi−a+ 32 . (4. xi−a+ 32 ] = (2Δx)−1 (Qi−a+1 − Qi−a ) I[xi−a+ 1 .40) ˆ ≤ xi−a+ 52 .42) Using (4. where x ˆ depends on x and xi−a− 12 ≤ x The reconstruction function for Q in cell i is denoted as Qi (x) and is defined by dP (4. xi−a+ 12 . xi−a+ 5 ] = (2Δx)−1 (Qi−a+2 − Qi−a+1 ) (4. xi−a+ 12 . xi−a+ 32 ] (4.37).36).38). xi−a+ 52 ] =  xi−a+ 52 − xi−a− 12 −1 I[xi−a+ 12 . I[xi−a− 12 . xi−a+ 3 . xi−a+ 12 ] = Qi−a I[xi−a+ 12 .11).43) 2 2 2 Using (4. xi−a+ 52 ]−  I[xi−a− 12 . xi−a+ 32 ] = Qi−a+1 I[xi−a+ 32 .96 Reconstruction  I[xi−a− 12 .41) Qi (x) = for xi− 12 ≤ x ≤ xi+ 12 dx Thus   Qi (x) = a1 + a2 (x−xi−a+ 12 ) + (x−xi−a− 12 ) +  a3 (x−xi−a+ 12 )(x+xi−a+ 32 ) + (x−xi−a− 12 )(x−xi−a+ 32 )+  (x−xi−a− 12 )(x−xi−a+ 12 ) (4.44) .

From (4.40).39). it is evident that dE/dx = O(Δx3 ).45) 2(Δx)2 where ξ = x − xi .24. ⎧ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎨ dQi = ⎪ dx ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎩ ΔQi+ 1 2 Δx ΔQi+ 1 2 Δx ΔQi− 1 + 2 + 2 + Δx ΔQi− 1 Δx + (ΔQi+ 3 −ΔQi+ 1 ) 2 Δx2 2  x − xi+ 21 (ΔQi+ 1 −ΔQi− 1 )  Δx2 (ΔQi+ 1 −ΔQi− 1 )  2 2 Δx2 2 2 (ΔQi− 1 −ΔQi− 3 ) 2 Δx2 2 x − xi+ 12 x − xi− 12  x − xi− 12     for a = 0 for a = 1 for a = 1 for a = 2 (4. Therefore.46) . Qi (x) = Qi − (ΔQi−a+ 32 − ΔQi−a+ 12 ) 24    1  ΔQi−a+ 12 + a − 12 ΔQi−a+ 32 − ΔQi−a+ 12 ξ + Δx (ΔQi−a+ 32 − ΔQi−a+ 21 ) 2 + ξ (4.40).35). 4. Qi (x) is a third-order accurate reconstruction of Q.13).5.33) since Qi (x) =  xi+ 1 2 xi− 1 2    dQi     dx  dx  xi+ 1 2 = xi− 1  ≤ 2 xi+ 1 2 xi− 1  ≤ xi− 1 2 xi+ 1 2 xi− 1 2    dQ d2 E   −   dx  dx dx2     x 1  dQ  i+ 2   dx +  dx     d2 E     2  dx  dx  2    dQ  3    dx  dx + O(Δx ) 4.4.45). The value of a is determined by effectively minimizing T V (Qi (x)). It is also straightforward to verify (4. (4. From (4. dE dI − dx dx dE = Q(x) − dx From (4.1 Determination of the Value of a The three possible choices for the domain of dependence of Qi (x) for cell i are shown in Fig. and (4.5 Essentially Non-Oscillatory Methods 97 Using2 (4.

.. .24. i−2 xi− 52 i−1 xi− 32 i xi− 12 i+1 xi+ 12 i+2 xi+ 32 xi+ 52 cells used to define Qi (x) for cell i for a = 0 ... Choices for cells to define Qi (x) Note that the two expressions for the case a = 1 are equivalent..........3.. . . x . ...... x ....... Table 4....... i−2 xi− 52 i−1 xi− 32 i xi− 12 i+1 xi+ 12 i+2 xi+ 32 . . 4.47) .... ... ...  T V (Qi (x)) = xi+ 1 2 xi− 1 2    dQi     dx  dx β k − ΔQi+ 12 − ΔQi− 12 − ΔQi− 12 − ΔQi− 32 1 1 0 0 (4... Derivative of Reconstructed Function in Cell i a α 0 1 1 2 ΔQi+ 12 ΔQi+ 12 ΔQi− 12 ΔQi− 12 ΔQi+ 32 ΔQi+ 12 ΔQi+ 12 ΔQi− 12 Therefore. ..... and k are defined in Table 4.3.... ... .. We rewrite the above as  dQi α β  x − x = + 1 i+k− 2 dx Δx Δx2 where α............ x xi+ 52 cells used to define Qi (x) for cell i for a = 1 ... β. ... .98 Reconstruction cells used to define Qi (x) for cell i for a = 2 .... i−2 xi− 52 i−1 xi− 32 i xi− 12 i+1 xi+ 12 i+2 xi+ 32 xi+ 52 Fig..

4. 4.48) Following Harten et al.4. This method is denoted Weighted Essentially NonOscillatory reconstruction. The correct values are reconstructed at all faces.18) with ten cells is shown in Fig. The reconstruction of the discontinuity (4.5. Exercises 4.26. (1987a). Instead of selecting the value of a that yields the reconstruction with the minimum oscillation. 4.1 Prove (4. The algorithm is indicated in Table 4. solution Equation (4.19) becomes Qli+ 1 = Qi + 12 ΔQi− 1 Qri− 1 = Qi − 12 ΔQi+ 1 2 2 2 2 Consider the primitive function  x Q dx I(x) = x i− 3 2 for xi− 3 ≤ x ≤ xi+ 1 2 2 .2 Results The reconstruction for the sine wave with ten cells is shown in Fig.25. The reconstructed values at the left and right sides of each face are similar to Fig. we select a to minimize α and β in turn. (1994).4. 4. Table 4. Algorithm for Determining Value of a 1st criterion |ΔQi+ 12 | ≤ |ΔQi− 12 | |ΔQi+ 12 | > |ΔQi− 12 | 2nd criterion |ΔQi+ 32 |ΔQi+ 32 |ΔQi+ 12 |ΔQi+ 12 − ΔQi+ 12 | ≤ |ΔQi+ 12 − ΔQi+ 12 | > |ΔQi+ 12 − ΔQi− 12 | ≤ |ΔQi− 12 − ΔQi− 12 | > |ΔQi− 12 − ΔQi− 12 | − ΔQi− 12 | − ΔQi− 32 | − ΔQi− 32 | k a 1 1 0 0 0 1 1 2 An alternate approach for Essentially Non-Oscillatory reconstruction was developed by Liu et al.19) is second-order accurate if κ = −1.≤  xi+ 1 2 xi− 1 2 = |α| + 1 2 Exercises  x |α| dx + Δx 99 i+ 1 2 xi− 1 2  |β|   x − x 1   dx i+k− 2 Δx 2 |β| (4.5. all possible reconstructions for a given order accuracy are generated and a weighted sum is used as the actual reconstruction.

............ ... . .0  reconstructed solution exact solution .. 0 1 2 3 4 5 6 7 8 9 .. 4..5 0.. . ...... 5 1 1 0 1 1 1 1 0 1 1 ...5 cell a −1.. .... 0. . ... . ... ...... x . .. and xi+ 1 ... .....25. The polynomial may be obtained using Newton’s formula 2 2 2 ... 3 ..... .. .......... .. .. . .. 2 −0.. .. .....0 0..... ..... ....... . . ....... .. . 6...... .. ....... . . ... Fig..... ... ... ..... . ..... ....... .. ...... . .. . 4 . Then I(xi− 3 ) = 0 I(xi− 1 ) = Δx Qi−1 I(xi+ 1 ) = Δx (Qi−1 + Qi ) 2 2 2 There is a unique second-order polynomial that interpolates I(x) at the three points xi− 3 .. .............. ....... .. .........0  . 1......... 1 .. .. .. .. .... .............. ......... −1.......... .. ...... ..... ...... ..... ... . .... . x  −1. ....... . .... .. . .......0 −0... .. .. .. . ......... Reconstruction using ten cells where Δx is constant....   .. ...... . 0...5    reconstructed exact      .. .. . ..2)......... 4..5 1.. . ... ...... xi− 1 . Note that the range of the function is less than (4.. ... . . ..... .......5 Fig. ..26.... Reconstruction of sine using ten cells Q ....0 ...... ........... ..........5 ........ ........ ... .. .. ..... .. ..100 Reconstruction Q ... ........0  1  2   3  4 5 6 .. 1....

xi+ 1 ] are defined by (4.14) yields 1 Qi = Vi  Vi Qi (x) dx . 2 Qri− 3 = Qi−1 − 12 ΔQi− 1 2 2 is a second-order accurate reconstruction of Q to the right face of xi− 3 using cells i − 1 2 and i.2 Verify that the reconstruction Qi (x) in (4. xi− 1 ]. xi− 1 . evaluating Q(x) at xi−3 . I[xi− 3 . and I[xi− 3 . Similarly. 4. 1998) P (x) = a0 + a1 (x − xi− 3 ) + a2 (x − xi− 3 )(x − xi− 1 ) 2 2 2 The coefficients are defined by a0 = I[xi− 3 ] a1 = I[xi− 3 . Thus. Qri− 1 = Qi − 12 ΔQi+ 1 2 2 is a second-order accurate reconstruction of Q to the right face of xi− 1 using cells i and 2 i + 1. xi− 1 . 2 2 2 2 2 2 a0 = a1 = Qi−1 a2 = (2Δx)−1 ΔQi− 1 0 2 Then I(x) = P (x) + E(x) for xi− 3 ≤ x ≤ xi+ 1 2 2 where E(x) is the error defined by E(x) =  (x−xi− 3 )(x−xi− 1 )(x−xi+ 1 ) d3 I  2 2 2  3! dx3  x=ˆ x where x ˆ depends on x and xi− 3 ≤ x ˆ ≤ xi+ 1 . xi+ 1 ] 2 2 2 2 2 2 where I[xi− 3 ]. Thus.6).Exercises 101 (Mathews and Fink. Letting i → i + 1. xi− 1 ] a2 = I[xi− 3 . The reconstruction function for Q in cell 2 2 i is denoted as Qi (x) and is defined by Qi (x) = dP dx xi− 1 ≤ x ≤ xi+ 1 for 2 2 and thus Qi (x) = Q(x) + O(Δx2 ) Now Qi (x) = Qi + ΔQi− 1 2 Δx ξ where ξ = x − xi . Qli+ 1 = Qi + 12 ΔQi− 1 2 2 is a second-order accurate reconstruction of Q to the left face of xi+ 1 using cells i − 1 2 and i.

4 4. solution Equation (4. Qri− 1 = Qi − 12 ΔQi− 1 2 2 4.2. Qli+ 1 = Qi + 12 ΔQi+ 1 2 2 Using the primitive function again.19) becomes Qli+ 1 = Qi + 12 ΔQi+ 1 Qri− 1 = Qi − 12 ΔQi− 1 2 2 2 2 Using the same primitive function as in Exercise 4.19) is second-order accurate if κ = +1.3 Reconstruction Prove (4. xi+ 1 ] 2 2 2 Now I(x) = P (x) + E(x) where E(x) is defined by E(x) = for xi− 1 ≤ x ≤ xi+ 1 2 2  (x − xi− 1 )(x − xi+ 1 ) d2 I  2 2  2! dx2  x=ˆ x 2 .19) is second-order accurate if κ = 0. We define the primitive function  x Q dx I(x) = xi− 1 ≤ x ≤ xi+ 1 for 2 x− 1 2 2 Then I(xi− 1 ) = 0 I(xi+ 1 ) = Δx Qi 2 2 The first-order polynomial P (x) that interpolates I(x) at the two points xi− 1 and xi 1 2 is  P (x) = a0 + a1 x − xi− 1  2 where a0 = I[xi− 1 ] a1 = I[xi− 1 .1 and letting i → i + 1.102 4. Prove the following reconstruction is first-order accurate: Qli+ 1 = Qi Qri− 1 = Qi 2 2 solution The solution is analogous to the derivation in Section 4.5 Prove (4.

2 2 Q(x) = dP dx a1 = Qi = and Q(x) dE dI − dx dx Q(x) + O(Δx) = = 4.14) assuming an arbitrary mesh spacing Δxi . and for 2 2 ΔQi+ 1 ≤ 0 and ΔQi− 1 ≥ 0.6 4. respectively. bΔQi+ 1 ) = 2 2 if ΔQi− 1 ≤ bΔQi+ 1 2 2 if ΔQi− 1 > bΔQi+ 1 ΔQi− 1 2 bΔQi+ 1 2 2 2 which agrees with the result for Case 1.4 for κ = −1 and 0 is equivalent to 1 ΔQ i− 2 = minmod (ΔQi− 21 . bΔQi+ 12 ) 1 ΔQ i+ 2 = minmod (bΔQi− 21 . A similar analysis holds for ΔQ 4. for ΔQi+ 1 ≤ 0 2 and ΔQi− 1 ≤ 0. Thus.7 Derive Equation (4. bΔQi+ 1 ) = 0 2 2 which agrees with the results for Cases 2 and 4. y) = y if |x| > |y| where x and y have the same sign where x and y have opposite signs ⎪ ⎩ 0 solution Consider the expression  i− 1 = minmod (ΔQi− 1 . bΔQi+ 1 ) ΔQ 2 2 2 For ΔQi+ 1 ≥ 0 and ΔQi− 1 ≥ 0. which agrees with the result for Case 3. 2 2 minmod (ΔQi− 1 .29). bΔQi+ 1 ) = 2 2 ΔQi− 1 2 bΔQi+ 1 2 if ΔQi− 1 ≥ bΔQi+ 1 2 2 if ΔQi− 1 < bΔQi+ 1 2 2  i+ 1 . .8 2 Develop the third-order reconstruction analogous to Equation (4. 2  minmod (ΔQi− 1 . Finally. Prove that the limiter in Section 4. 2 2  minmod (ΔQi− 1 . For ΔQi+ 1 ≥ 0 and ΔQi− 1 ≤ 0.Exercises 103 where xi− 1 ≤ x ˆ ≤ xi+ 1 . ΔQi+ 12 ) where ⎧ ⎪ ⎨ x if |x| ≤ |y| where x and y have the same sign minmod (x.

where k is the iteration number beginning with k = 1.46) are equivalent.10 Prove that the algorithm in Table 4. (c) Select the next iterate for the index of the leftmost cell according to ˆik+1 (i) = ⎧ ˆ ⎪ ⎪ ⎨ ik (i) − 1 if |I[xˆik (i)− 3 . . . xˆik (i)− 1 . 1997 ) for determining the leftmost cell of the three cells used to reconstruct Qi (x) in cell i: (a) Denote the index of the leftmost cell used in the reconstruction of cell i as ˆik (i).4 for determining the value of a is equivalent to the following iterative scheme (Harten et al. 4.9 Reconstruction Show that the two expressions for a = 1 in Equation (4. solution The first expression for a = 1 is ΔQi+ 1 (ΔQi+ 1 − ΔQi− 1 ) dQi 2 2 2 = + dx Δx Δx2  x − xi+ 1  2 Using x − xi+ 1 2 = x − xi− 1 − (xi+ 1 − xi− 1 ) = x − xi− 1 − Δx 2 2 2 (E4. 2 2 |I[xˆik (i)− 1 . .. . . . 1987a. . xˆik (i)+ 1 +k ]| 2 otherwise 2 2 .104 4. xˆik (i)+ 1 . . 2. xˆik (i)− 1 +k ]| < 2 ⎪ ⎪ ⎩ ˆi (i) k for k = 1. (b) Since the reconstruction must include cell i.1) 2 the second expression for a = 1 is obtained. see also Chu. set ˆi1 (i) = i.

and 4. for example. “Godunov’s Scheme” would never have been created. 5. For discontinuous functions.2 Godunov’s Method Godunov (1959) introduced the concept of utilizing the solution of the local Riemann problem at each cell face as the basis for determining the flux Fi± 12 105 . the difference between the left and right states can be of the same order as the function value itself as shown in Figs. Van Leer.25. the discontinuity may be slight as illustrated in Figs.5. In this chapter. The class of flux algorithms based on exact or approximate solutions of the Riemann problem (or some generalization thereof) are known as Godunov. 4. 1997).21 and 4. For smooth functions.5 Godunov Methods If I would have read Lax’s paper [Lax 1959] a year earlier. This discontinuity in the reconstructed functions is analogous to the Riemann problem described in Section 2. Riemann or Flux Difference Splitting Methods (Hirsch. The original concept was developed by Godunov (1959). Sergei Godunov (1999) 5. 1988. we present three different methods. in Toro (1997). 4.20.1 Introduction The reconstruction methods described in Chapter 4 inherently give rise to a discontinuity in the reconstructed functions at the cell faces.9 and therefore suggests the development of a flux algorithm based on the solution of the Riemann problem. 4. Additional algorithms are described.26.

. ... . . . ....2). .. .... . .9) at xi+ 12 (Fig.. . . i+2 ...... . .10): = Qni − Qn+1 i 1 Δx  tn+1  tn  Fi+ 12 − Fi− 12 dt 5. .. . . . . . ..... . . i xi− 12 . . . ...... . . .† The solution to the general Riemann problem at xi+ 12 for t > tn depends only on (x−xi+ 12 )/(t−tn ) and not on x or t separately. ... ... . . ... ... .. The left and right states Qli+ 1 2 and Qri+ 1 at xi+ 12 may be taken to be 2 Qli+ 1 = Qi Qri+ 1 2 = Qi+1 2 (5. . . .... M at time tn . ..... . . .... .. ... . .1) which corresponds to a first-order accurate reconstruction of Q to the cell face. . 5. ... . . . a higher order accurate reconstruction can be performed using the methods of Chapter 4. x xi+ 32 Fig.. . .... A typical component of Qi is illustrated in Fig... .... ........ . . .. . . . . . . In general. ... .......1 Algorithm Consider the solution Qi for i = 1.1.. .. . . ..... .. ... .. .. ...... . . Qi−1 i−2 ..... .... . ... there is a discontinuity in Qi at each cell face... . .. . . Q .. ... . . .106 Godunov Methods in the integral form of the Euler equations (3. 5. Consequently.. .... ... .. . ... .. . . . ... . . .... ... . A typical component of Qi at time tn The flux Fi+ 12 is determined from the solution of the local general Riemann problem (Section 2. .. .. . . . Qi+2 Qi+1 Qi−2 .. Qi i−1 xi− 32 .. . . . ... .1. . .. . ... . . .. i+1 xi+ 12 . . . . ... ..... . .. .. . 5.. .... .. .. .2. † Alternately. .. .. ...... . .

. ........2 Stability Intuitively. ... . .. This implies Δt = ΔtCF L ... . .. ... ........... where ΔtCF L = min i Δx cmax (5. ...... ...... . .. . ..2 Godunov’s Method 107 the solution for Q at xi+ 12 is independent of time... .... . ......... . .......... ... .... contact surface contact surface shock t i−1 i i+1 t .... ...... ... . . ... ... . . . . . . ..... .. ....... ...........2) 2 where QR i+ 1 is the solution of the general Riemann problem at xi+ 12 using 2 the reconstructed values Qli+ 1 and Qri+ 1 . .4) where cmax is the maximum absolute wave speed of the waves entering cell i and mini indicates the minimum value over all cells.. ... . .. ..5) .. .. the time step Δt must be chosen to insure that the rightmost wave emanating from the general Riemann problem at the left face does not intersect the right face. .. ... . . . ..... . ..... ... ... .. ..2. ... . .. ... .. . .... ...... .. . .. ..... .. . . . ..... ... .... x ... .. .. .. .....2... ... ...... ....... .... and therefore  tn+1 tn Fi+ 12 dt = Fi+ 12 Δt = F(QR i+ 1 )Δt (5. .. n+1 ........... . ........ . . . ... .... . . .. .... .. Possible flow structure for Godunov’s Second Method 5. ... ......... ... ... . . .. and vice versa. .. ...... . . ...... ..........1 expansion fan .. . ... . . .. ... . .......... ... . ... . .. n shock .3) Qn+1 i i+ 12 ) − F(Qi− 12 ) Δx This is Godunov’s Second Method. .... .5.. . . .. . ... . .. . . .... . . ... ... ... .... . ..... .... . . . . . . . .. ....... . . . Fig.. .. ...  Δt  R = Qni − F(QR (5... .. .. . ...... ......... . .. .. ..... . ...... ........ ..... .. ........ ............ .. . .. .... . . Δt .. ... ....... .. .... . .... .... . ... ....... ... ........ . . . . .... ... .. .... ... . 2 2 Thus. ... . .4) is  ΔtCF L Δx ≈ min min i j |λj |  (5..... . ...... . ....... . ..... .. Δx expansion fan ..... ...... . . A similar result holds for Fi− 12 .......... ...... ....... . ..... . ... .... .. ..... . . . .. .. .. . . . . . The linear approximation to (5.. .. . . .... . ......... ... . ..... . .. ..... .. 5........ . ... . .

1).1. Consistency.1).3. The initial condition is defined by (2.6) where the Courant number C is less than unity. 5.108 Godunov Methods where λj . 2. Δt = CΔtCF L (5.20) with κ = 0 and the minmod limiter. 3 are given by (2. The exact solution Qn.8. The norm.18.7) where the Courant number C = 0. The linear convergence is a direct consequence of the linear reconstruction (5.12) and the minimum is taken over all cells. The speed of the disturbance is accurately predicted.88).46 and ΔtCF L is calculated according to  ΔtCF L Δx = min min i j |λj |  (5.1) is employed.92) with  = 0.e i −Qi ||κao Δt =  κao Δt 5 × 10−4  n ||Qn. is obtained from (2. 5. 5. 5. for κao t = 7 and C = 0. which is prior to the shock formation.3 Accuracy. and consequently a more conservative estimate is employed.g. and therefore Δt and Δx are decreased by the same ratio. and Convergence We now consider the problem described in Section 2. j = 1.. The error in amplitude is substantially reduced by using a secondorder reconstruction as indicated in Fig. † The linear line in Fig. The time step Δt is determined by Δt = CΔtCF L (5.102). The time step Δt is held in fixed ratio to the grid spacing Δx. The computed result (using 100 cells) and exact solution are displayed in Fig.2.e i −Qi ||κao Δt=5×10−4 .8) using the initial condition. where the results are shown using (4. and the domain is 0 < κx < 2π. The first-order reconstruction (5.5. Experience indicates that this criterion may be an overestimate depending on the particular choice of parameters for the algorithm (e.e i The convergence is displayed in Fig.4. however. The significant error in amplitude is attributable to the dissipative nature of the first-order reconstruction (5. defined by (3.85) to (2. The solution converges linearly† to the exact solution for sufficiently small Δt. together with results for the first-order reconstruction. 5.3 is n ||Qn. reconstruction method) and flow. is evaluated at κao t = 7.

. .. . .. .. .. ... ......3 Roe’s Method Roe (1981. ............. ....... .. .. .... ....... . ..... .10 κx −0.. .. ..... . .... .......... ... ..........15 ..... ...... . ............. . . ....... . ...1986) developed an algorithm based on an exact solution to an approximation of the generalized Riemann problem. .. ........... ... . .. . ... . .....3 Roe’s Method 109 n log10 ||Qn... 5. .. .. .... . ............ . .... ... . .. .... .. ...e i −Qi || 0 −1 −2 .. . .. .. .. .... . ◦ ◦ ◦ ◦ ◦ computed linear ◦ ◦ ◦ −3 −4 ◦ −3 −2 −1 0 log10 κao Δt Fig.... ...... Unlike Godunov’s Sec- ..... . ... ...... .15 Fig.. ... . . ............. 5. .05 ◦ ◦◦◦ ◦ ◦◦◦◦ ◦ ◦ ◦4 ◦ 2 3 5 6 ◦◦◦1 ◦ ◦ ◦ ◦ 0. ... ....... ... ...... ... .... . ....3...... . .. . . . . . ...... . ... . . .. .... .4.. ... ..... .... .. . ... . . κa t = 7 o ..... . ..... .. . ... .. . . .. ... . .. .. ..... ... . ... .. .. ..... . . .. .... ...... .... . .. .. .. .......... . ..... ... ...... .. .. .. . .... Computed and exact solutions using first-order reconstruction 5. . . .... ◦ computed ◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦ ◦ exact ◦ ◦ ◦ ◦◦ ◦◦ ◦◦ ◦◦◦◦ ◦ ◦ ◦ 0.... . . ..00 ◦ ◦ ◦ ◦ ◦ ◦ ◦◦ ◦ ◦◦◦◦ ◦◦ ◦◦ −0. ... .05 ◦◦◦◦ ◦ ◦ ◦◦ ◦ ◦ ◦◦◦ ◦◦◦◦ ◦◦◦◦◦◦◦◦◦◦◦◦◦◦ −0... .. . . .. .5.. .... . ..... . .. ..... ................ . ......... ...... . . . .. . . .. .. .. .. ........ .. ...... .. ... .. .... . .. ...... .. ... ..10 0..... . .... ... ... .... . .... . . .......... ..... ....... . . . ..... . .. . ◦ . ...... . . .. ... .......... .... .. ..... .............. ............ ...... ..... .. ... .. . .. . .. Convergence for Godunov’s Second Method using first-order reconstruction u/ao 0.. .. . ..... . .. ... .. . .... .. ... .

... .. . ..... .. Computed and exact solutions using first... . . ..10) and H = e + p/ρ is the total enthalpy. ... .. . . .. .. ..... .10 0. .. . ... 5.10 −0.. . ....11) ˜ l .. .....11) ⎧ ⎪ ⎨ (5.... ... . Roe’s scheme does not require an iterative procedure to find the flux‡..... ...... . ... Qr ) depends on the left and right states Ql and Qr of the where A(Q general Riemann problem (Fig.. .. ‡ Recall that the solution of the General Riemann Problem requires an iteration to find the contact pressure except in the case of expansion-expansion...... . .... ..... . . ....... . .05 •◦•◦•◦•◦ • ◦ ◦ ◦ • ◦ • ◦◦• •◦•◦•◦ 2 3 5 6 ◦•◦•◦1•◦ ◦4 • ◦ • ◦ κx • ◦ 0..... ◦ computed 1st order •••••••••• ••◦•◦◦◦◦◦◦◦◦◦ ••••••• • • • computed 2nd order • ◦ ◦ •◦ ◦◦◦ • •◦•◦•◦•◦◦◦◦ ◦ • • ◦ exact ◦ ◦ • ◦ •◦ ◦◦ • 0... . . .... . ............... . . ...... . .. .. . .. . .. ... ....... .. .. .. . . . ...... .15 Fig... ... Qr ) ∂Q = 0 + A(Q ∂t ∂x (5.05 ◦•◦•◦•◦••• ◦ ◦◦ ◦ ◦ • ◦◦◦•• ◦◦◦ ◦◦◦◦◦◦◦◦◦◦◦•◦•◦•◦•••• • • • • •••••••••••••••• −0.... . ... . 5..6) and is assumed constant...... .. . .... . . .15 ........110 Godunov Methods u/ao . .... .6) in nonconservative differential form. o .. .. . . . ... ... .... .... . .. . ....... . ...... .... ....... .. . . .. ..... .... . ... κa t = 7 0...... ...5.. .. ∂Q ∂Q +A =0 ∂t ∂x where A is the Jacobian (2..00 ◦ •◦• ◦ ◦ • ◦ ◦ • ◦ • ◦ • ◦ ◦ • •◦ ◦••◦•◦• ◦◦◦• ◦◦ −0..... Roe sought a solution of the general Riemann problem using an approximate form of the Euler equations ∂Q ˜ l ... ..... ...... . ... ... .... ........ .. .. . ... ... ..... .... .. .. ... . .... .. .. .. . .. .... ..and second-order reconstructions ond Method. .9) ⎫ 0 1 0 ⎪ ⎬ 2 (γ −3) u /2 A(Q) = (3−γ) u γ −1 ⎪ ⎪ ⎩ −Hu + (γ −1)u3 /2 H −(γ −1)u2 γu ⎭ (5. ... . ... . ...... .... . . . ...... ......... 5... . ..3....1 Algorithm Consider the Euler equations (2. . . ....

.. respectively. i. 5. . . Property (iii) requires satisfaction of the exact jump conditions across a shock wave. (iv) The eigenvectors of A(Q ˜ have Property (i) implies that the individual components of the vector AQ the same units as the corresponding components of the vector F. the arbitrary piecewise constant functions f and g that have left and right states indicated by fl . ˜ several identities are needed. 111 Qr . .e. hyperbolic character of (5. Qr ) → A(Q) as Ql → Qr → Q. Initial condition for General Riemann Problem The matrix A˜ is an approximation to the matrix exact A and satisfies the following four properties (Roe... ˜ l .1): Δ(f + g) = Δf + Δg Δ(f g) = f¯Δg + Δf g¯ Δ(1/f ) = −Δf /fˆ2 (5.13) The matrix A˜ is found in the following manner..6... A(Q ˜ l .. gl and fr . Qr ) is diagonalizable with real. A(Q distinct eigenvalues and linearly independent eigenvectors... .9) for smooth Q... Property (iv) is consistent with the assumed ˜ l . (ii) A(Q ˜ l . Consider In order to determine the matrix A. .12) where Δf = fl − fr and similarly for Δg. . The following identities can be proven (Exercise 5. Qr ) are linearly independent... 1981) (i) A˜ provides a linear mapping from the vector space of Q to the vector space of F.. .11) is a close approximation to the solution of (5.. (iii) For any Ql and Qr . Qr ) × (Ql − Qr ) ≡ Fl − Fr . and f¯ ≡ 12 (fl + fr )  fl fr fˆ ≡ (5. x Fig. Property (ii) implies that the solution to (5.3 Roe’s Method Ql .. .11). .5. A parameterization vector . gr .

Therefore ˜ = AΔQ ˜ CΔν = ABΔν ˜ Δν = C −1 ABΔν ΔF (5.16) ⎪ ⎭ Since Q and F are quadratic in the elements of ν.4) ⎧ ⎪ ⎨ 0 1 0 (3−γ)˜ u (γ −1) A˜ = (γ −3)˜ u2 /2 ⎪ ⎩ −H ˜u ˜ −(γ −1)˜ ˜ +(γ −1)˜ u3 /2 H u2 γu ˜ ⎫ ⎪ ⎬ ⎪ ⎭ (5. It is straightforward to show (Exercise 5. and so on (Exercise 5.18) = CΔν where ΔQ = Ql − Qr . it is possible to find matrices B and C such that ΔQ = BΔν (5.21) . the determination of the matrix A˜ reduces to the problem of finding B and C.20) ⎫ ⎪ ⎬ ⎪ ⎭ (5.17) ΔF (5.112 Godunov Methods ν = (ν1 . Consequently.3) ⎧ ⎪ ν1 ⎨ 2¯ ⎫ 0 0 ⎪ ⎬ B = ν¯2 ν¯1 0 ⎪ ⎪ ⎩ ν¯ /γ (γ −1)¯ ν2 /γ ν¯1 /γ ⎭ 3 ⎧ ⎪ ⎨ ν¯2 ν¯1 0 C = ν2 /γ (γ −1)¯ ν1 /γ (γ −1)¯ ν3 /γ (γ +1)¯ ⎪ ⎩ ν¯2 0 ν¯3 and thus (Exercise 5.14) ⎪ ⎭ ⎧ ⎪ ⎨ ν12 Q= ν1 ν2 ⎪ ⎩ ν ν /γ + (γ −1)ν 2 /2γ 1 3 2 ⎧ ⎪ ⎨ ⎫ ⎪ ⎬ (5. ν3 )T is introduced as ⎧ ⎪ ⎨ √ ρ √ ρu ν= ⎪ ⎩ √ρ H and thus Q and F are ⎫ ⎪ ⎬ (5.19) and thus A˜ = CB −1 .2). ν2 .15) ⎪ ⎭ ν1 ν2 F= (γ −1)ν1 ν3 /γ + (γ +1) ν22 /2γ ⎪ ⎩ ν 2 ν3 ⎫ ⎪ ⎬ (5.

⎧ ⎪ ⎨ 1 e˜1 = u ˜ ⎪ ⎩ 1u 2 2˜ ˜2 = u λ ˜+a ˜. By comparison with (5. Qr ) is the Roe matrix.26) ˜ l .23) ⎫ ⎪ ⎬ ⎪ ⎭ (5. respectively. e˜2 = ⎪ ⎪ ⎭ ⎩ H ˜ +u ˜a ˜ ˜3 = u λ ˜−a ˜ ⎫ ⎪ ⎬ ⎧ ⎪ ⎨ 1 u ˜−a ˜ . Property Property (ii). Qr ) = S˜Λ ˜ S˜−1 A(Q (5. enthalpy. thus satisfying Property (iv). Qr ) is treated as a constant. Finally.24) where a ˜ is the speed of sound based on the Roe-averaged total enthalpy and velocity and is given by  a ˜= 2 ˜ − 1u (γ −1)(H 2˜ ) (5.5). A˜ satisfies ˜ → H as Ql → Qr → Q.3 Roe’s Method where 113 √ √ ρl ul + ρr ur ν¯2 = √ √ ν¯1 ρl + ρr √ √ ρl Hl + ρr Hr ν¯3 = √ √ ν¯1 ρl + ρr u ˜ ≡ ˜ ≡ H (5. The matrix A(Q Clearly.22) ˜ are the Roe-averaged velocity and Roe-averaged total The quantities u ˜ and H ˜ l . A˜ satisfies Property (i).25) It may be directly verified that the eigenvectors are linearly independent. Qr ) ∂Q = 0 + A(Q ∂t ∂x (5. where S˜ is matrix of right eigenvectors of A(Q ⎧ ⎪ ⎨ 1 1 1 S˜ = u ˜ u ˜+a ˜ u ˜−a ˜ ⎪ ⎩ 1u 2 H ˜ ˜ +u ˜a ˜ H −u ˜a ˜ 2˜ ⎫ ⎪ ⎬ ⎪ ⎭ (5. e˜3 = ⎪ ⎪ ⎭ ⎩ H ˜ −u ˜a ˜ (5.5. ˜1 = u λ ˜. the eigenvalues λ eigenvectors e˜i of A˜ may be found directly (Exercise 5.28) . The Roe matrix is now sought. Qr ). since u ˜ → u and H ˜ i and the right (iii) is satisfied by construction. where A(Q may be diagonalized as ˜ l . ⎫ ⎪ ⎬ ⎧ ⎪ ⎨ 1 u ˜+a ˜ .10).27) ˜ l . The exact solution of the approximate General Riemann Problem ∂Q ˜ l .

. The solutions for R1 ..... C2 . . ..32) The solution of (5... .... t t ..... ˜ dx/dt=λ 1 . ... ... .7 assuming 0 < u ˜<a ˜. and C3 are the characteristic curves of (5...30).... . ... Qr ) is treated as a constant............30) is R1 = constant R2 = constant R3 = constant dx ˜ = λ1 = u ˜ dt dx ˜ = λ2 = u on curve C2 defined by ˜+a ˜ dt dx ˜ on curve C2 defined by ˜−a ˜ = λ3 = u dt on curve C1 defined by C1 . The solution is illustrated in Fig. .. 5.. . . l . R2 =R2 R2 =R2 . ... ... . . ... .. ... ....... 5.... 5.....26) by S˜−1 to obtain ∂R ˜ ∂R +Λ =0 (5... .31) 2 ⎪ ⎩ R ⎪ ⎭ 3 0 ⎫ 0 ⎪ ⎬ 0 ⎪ ˜3 ⎭ λ (5. . ....... . ... . .. . Solution for R3 Fig....... .. ˜ .. . ....29) ˜ Since A(Ql . ... and R3 are shown together in Fig. . . . l l . .. .. it is possible to multiply (5...... ..8.. x ... . R3 =R3r l . .... R2 .. x Solution for R1 t ... .. ... . x Solution for R2 . .30) ∂t ∂x where2 R ≡ S˜−1 Q = ˜ is and Λ ˜≡ Λ ⎧ ˜ ⎪ ⎨ λ1 ⎪ ⎩ ⎧ ⎫ ⎪ ⎨ R1 ⎪ ⎬ 0 ˜2 λ 0 0 R (5...... ... . .. . R1 =R1r R1 =R1 .... . ˜ dx/dt=λ 2 .. . .114 Godunov Methods and S˜−1 is ⎧ ⎪ ⎨ ⎫ 1 − (γ −1)˜ u2 /2˜ a2 (γ −1)˜ u/˜ a2 −(γ −1)/˜ a2 ⎪ ⎬ −1 2 2 2 S˜ = (γ −1)˜ u /4˜ a −u ˜/2˜ a −(γ −1)˜ u/2˜ a + 1/2˜ a (γ −1)/2˜ a2 ⎪ ⎪ ⎩ (γ −1)˜ u2 /4˜ a2 + u ˜/2˜ a −(γ −1)˜ u/2˜ a2 − 1/2˜ a (γ −1)/2˜ a2 ⎭ (5. .. . ... .7.. Roe solution to the General Riemann Problem . R3 =R3 ..... dx/dt=λ 3 ..

....... ....... .. ....... ...5..... . R ...... ... ..... .e. .... ... ..... 1l 2l 3l 1r 2r 3r . .. .31).............. ....... . .. . ..... .. .. R ...... . .. .. . .. ..... .. ... ..35) It is straightforward to show that ˜ (Rl + Rr ) + 1 S| ˜˜ Fi+ 12 = 12 S˜Λ 2 Λ|(Rl − Rr ) where ˜ = |Λ| Using (5. ........ .... R ) x Fig..... ..... ................. ........... .. .3 Roe’s Method 115 t........33) The individual components of R at xi+ 12 (i. R . . . ... ....... ......38) . .. . ..... 5..... . ..... . 1r 2l 3r. .. .... ..... . ...... ... ....... . .......36) ⎫ ⎪ ⎬ (5. . .... .. R ) (R . ........... . .. ... .. . . .. .... ... ... . .. ...... ... ....... ... . Roe solution to the General Riemann Problem Consider now the semi-discrete form (3.... ... ... ... ...... .. .1) is  ˜ = S˜Λ ˜ S˜−1 Fi+ 12 = Fi+ 12 = AQ  ˜ SR  i+ 21 ˜ i+ 1 = S˜ΛR 2 (5...3 where sign(λ˜l ) = ⎧ ˜ ⎪ ⎨ +1 if λk > 0 ⎪ ⎩ ˜k < 0 −1 if λ ˜k = 0 0 if λ (5... (R ... . . . ...8) are given by 1 1 ˜ Rk |n+1 (5... . ............ . x = 0 in Fig.. . . . . . ..... . .. .34) i+1/2 = 2 (Rkl + Rkr ) + 2 sign(λk ) (Rkl − Rkr ) for k = 1.. ..... ..... R ) (R . . .... 5. . . .. . . . ˜3 dx/dt = λ ˜ dx/dt = λ ˜2 dx/dt = λ 1 .. . ... ..... Fi+ 12 = 1 2  ⎧ ˜ ⎪ ⎨ |λ1 | ⎪ ⎩ 0 0 0 0 ˜ |λ2 | 0 ˜3 | 0 |λ (5...... .. .9) of the Euler equations  Fi+ 12 − Fi− 12 dQi + dt  =0 Δx The flux Fi+ 12 given by (4. .... R ) (R . R ... ....2.37) ⎪ ⎭  ˜ Λ| ˜ S˜−1 (Ql − Qr ) ˜ S˜−1 (Ql + Qr ) + S| S˜Λ (5....8......... . ..... . .. ... . 1l 2l 3r ... . ............

A more convenient form is ⎡ Fi+ 12 = 1 2 ⎣Fl + Fr + + j=3  ⎤ ˜ j |˜ αj |λ ej ⎦ (5. The flux Fi+ 12 in (5.40) is taken to be an approximate expression for the flux in the exact Riemann Problem.40) where Fl = F (Qli+ 1 ) and Fr = F (Qri+ 1 ). This is Roe’s method for the flux 2 2 F .41) j=1 where  α1 =  α2    (γ −1) Δρe a ˜2     (γ −1) u 1 ˜2 (γ −1) u ˜ u ˜ − = − Δρ + Δρu + 4 a ˜2 2˜ a 2˜ a 2 a ˜2  α3  ˜2 (γ −1) u u ˜ 1− Δρ + (γ −1) 2 Δρu − 2 2 a ˜ a ˜  (γ −1) Δρe 2˜ a2     ˜2 (γ −1) u ˜ u ˜ (γ −1) u 1 + = + Δρ − Δρu + 4 a ˜2 2˜ a 2˜ a 2 a ˜2   (γ −1) Δρe 2˜ a2 (5. and so on. then Fi+ 12 = 1 2   ˜ Λ| ˜ S˜−1 (Ql − Qr ) Fl + Fr + S| (5. strength of the kth wave that has speed λ .39) Approximating the left and right states by the reconstructed values Ql = Qli+ 1 2 Qr = Qri+ 1 2 the flux becomes Fi+ 1 = 2 1 2   ˜ Λ| ˜ S˜−1 (Ql 1 − Qr 1 ) Fl + Fr + S| i+ i+ 2 2 (5. It uses an exact solution to the approximate Riemann Problem (5. The coefficients αk may be interpreted as the ˜k.116 Godunov Methods ˜ S˜−1 Ql = AQ ˜ l is the flux based on the flow variables on the left Since S˜Λ side and similarly for the right side.42) and Δρ = ρl − ρr .26).

Qi . Qri− 1 ) 2 2 2 2 dt where  Ri = − Fi+ 12 − Fi− 12  (5. Qi ). Qi . . X3 . X3 = Qli− 1 . . yielding Ri (Qli+ 1 . 2 X2 = Qri+ 1 . . and A˜ and |A| We rewrite (3. Qri− 1 ) = Ri (Qi . . Qi . 4. Qri+ 1 ) 2 We linearize Fi+ 12 as Fi+ 12 = 1 2  2  ˜ l 1 + Qr 1 ) + |A|(Q ˜ l 1 − Qr 1 ) A(Q i+ i+ i+ i+ 2 2 2 (5. X4 ) = (Qi . are evaluated at (X1 . Qi )+ 2 2 2 2 ∂Ri r ∂Ri l ∂Ri r ∂Ri l (Qi+ 1 − Qi ) + (Qi+ 1 − Qi ) + (Qi− 1 − Qi ) + (Q 1 − Qi )+ 2 2 2 ∂X1 ∂X2 ∂X3 ∂X4 i− 2 O(ΔQ2i ) where ∂Ri /∂Xj . Qi .† The Euler equations (3. Qli− 1 .9) as dQi = Ri (Qli+ 1 . X4 ) We expand Ri in a Taylor series about Qi . Fi+ 12 = F (Qli+ 1 . 2 2 X4 = Qri− 1 2 Thus.5.43) 2 where ˜ = S| ˜ Λ| ˜ S˜−1 |A| ˜ are treated as constants.2 Stability We consider the semi-discrete stability of Roe’s Method. we define X1 = Qli+ 1 . X3 . Qli− 1 .3 Roe’s Method 117 5. X2 .44) Δx For clarity. Ri = Ri (X1 .9) are  Fi+ 12 − Fi− 12 The flux Fi+ 12  dQi + =0 dt Δx depends on the left and right states. Qri+ 1 .3. X2 . j = 1. † A discrete stability analysis of Roe’s Method is presented in Chapter 7. . Qri+ 1 .

45) Moreover.48) Thus. Λ− is a diagonal matrix whose elements min(λm .46) where A+ = A− = 1 2 (A 1 2 (A + |A|) = T Λ+ T −1 − |A|) = T Λ− T −1 (5. the linearized semi-discrete form of the Euler equations is  dQi 1  − A Qi+1 + (A+ − A− )Qi − A+ Qi−1 =− dt Δx (5.118 Godunov Methods From (5. 0)} Λ− = diag {min(λm . ∂Fi+ 12 = ∂X1 ∂Fi+ 12 ∂X3 1 2 = ∂X3  ˜ . Λ+ is a diagonal matrix whose elements max(λm . we assume that the flow is periodic in x over a length L = (M −1)Δx.   ˜ . 0) are thus nonpositive. ∂Fi− 12  ∂X4 1 2 = 0. Thus. A˜ + |A| ∂Fi+ 12 = 0. A˜ = A + O(ΔQ) ˜ = |A| + O(ΔQ) |A| and thus. = 1 2  ˜ A˜ − |A| =0 ∂X2   ˜ A˜ − |A| To proceed further. We select the simple first-order reconstruction Qli+ 1 = Qi Qri+ 1 2 = Qi+1 2 (5. Similarly. Q1 = QM . to the lowest order. 0) are thus nonnegative. A˜ + |A| ∂Fi+ 12 ∂X2 ∂Fi− 12 ∂X1 ∂Fi− 12 ∂X4 = 1 2  ∂Fi− 12 = 0.49) For purposes of simplicity. Ri = −  1  − A Qi+1 + (A+ − A− )Qi − A+ Qi−1 Δx (5.47) where Λ+ = diag {max(λm .40). the spatial reconstruction must be specified. 0)} (5.

53) . time tn .48)..32). Q(xi . t) = Qi (t) Then the Fourier series (3.5.50)  1  ιkΔx − e A + A+ − A− − e−ιkΔx A+ Δx From (5. t) = l=N  ˆ k (t)eιkx Q l=−N +1 √ where ι = −1 and the wavenumber k depends on l according to (3. 0) (5. i. 0) λ− m = min(λm .52) it is evident that the eigenvalues of G are λ Gm = −  1  ιkΔx − − −ιkΔx + e λm + λ+ − λ − e λ m m m Δx Defining ˆk ˜ k = T −1 Q Q (5.30) for Q(x. G=− G = T DT −1 where D is the diagonal matrix D=−  1  ιkΔx − e Λ + Λ+ − Λ− − e−ιkΔx Λ+ Δx (5. Q dependence on l (through (3.e. Given the values of Qi at some ˆ k at tn are obtained from (3. t) is Q(x.31)) and on t. Consider Q(x. t) to be a continuous vector function that interpolates Qi . the Fourier coefficients Q Substituting into (5.31): k= 2πl L ˆ k (t) are complex vectors whose subscript k indiThe Fourier coefficients Q ˆ k (t) indicates cates an ordering with respect to the summation index l.3 Roe’s Method 119 and assume M is odd with M = 2N +1.49) yields ˆk dQ ˆk = GQ dt where the amplification matrix G is (5.47) and (5.51) Defining λ+ m = max(λm .

120

Godunov Methods

Equation (5.50) becomes
˜k
dQ
˜k
= DQ
dt
Defining
˜k =
Q



˜ ⎬

⎨ Qk1 ⎪

˜
Q

k2

⎩ ˜ ⎪
Qk ⎭
3

the solution is
˜k = Q
˜ k (0)eλGm t
Q
m
m

for

m = 1, 2, 3

The condition for stability is therefore
Real(λGm ) ≤ 0

(5.54)

i.e., all of the eigenvalues λGm , m = 1, 2, 3 of the amplification matrix G lie
in the left half or imaginary axis of the complex plane. Now
Real(λGm ) = − 

1  +
λm − λ−
m (1 − cos kΔx)
!
"
!
"
Δx 
≥0

≥0

and hence the stability condition (5.54) is satisfied.

5.3.3 Accuracy, Consistency, and Convergence
We now consider the problem described in Section 2.8. The initial condition
is defined by (2.92) with  = 0.1 and the domain is 0 < κx < 2π. The
first-order reconstruction (5.1) is employed. The norm, defined by (3.102),
is evaluated at κao t = 7, which is prior to the shock formation. The semidiscrete form (3.9) of the Euler equations is employed and the time integration is performed using a second-order Runge-Kutta method (Chapter
7).
The convergence is displayed in Fig. 5.9. The solution converges linearly†
to the exact solution for sufficiently small Δt. The linear convergence is a
direct consequence of the linear reconstruction (5.1). The computed result
(using 100 cells) and exact solution are displayed‡ in Fig. 5.10 for κao t = 7
† The line in Fig. 5.9 is
n
||Qn,e
i −Qi ||κao Δt = 

κao Δt
5 × 10−4 

n
||Qn,e
i −Qi ||κao Δt=5×10−4

‡ More precisely, the timestep κao Δt = 0.025, which corresponds to C = 0.45 at t = 0.

5.3 Roe’s Method

121

and C = 0.45. The significant error in amplitude is attributable to the
dissipative nature of the first-order reconstruction (5.1). The speed of the
disturbance is accurately predicted, however.
n
log10 ||Qn,e
i −Qi ||

0

−1

.
.....

..
.....

.
.....

.
.....

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

−2

.
.....

..
....

..
....

.
.....

computed
linear


−3
−4

−3

−2

−1

0
log10 κao Δt

Fig. 5.9. Convergence for Roe’s method using a first-order reconstructions

u/ao
0.15

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

κa t = 7

o
.
...
...
...
...
.
.
.
....
.........
..... ...
................ ................
.
.
.
.
.
.
.
.
....
.......
...
......
...
......
...
......
...
......
.
.
.
.
.
...
....
.
.
.
.
.
...
....
.
.
.
.
.
..
....
.
...
.
.
.
...
.
...
.
.
.
.
....
...
.
.
.
.
...
....
.
.
.
.
...
....
.
.
.
.
...
....
.
....................
.
.
.
..
....
.
...
.
.
.
...
.
...
.
.
.
...
...
.
.
.
.
.
..
.....
...
......
.
...
.....
...
......
...
......
......
.
...
.
.
.
.
...
......
.....
...
......
...
.......
.....
........
.
.......
.
.
.
.
.
.
.
.
......................

◦ computed





exact









◦◦◦
◦◦
◦◦
◦◦◦◦◦



0.05
◦◦
◦◦◦◦
◦◦
◦◦◦◦



◦◦4


1
2
3
5
6

◦◦

0.00
◦◦




◦◦
◦◦
◦◦◦◦
◦◦
◦◦◦
−0.05
◦◦◦◦
◦◦




◦◦◦
◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦
−0.10
0.10

κx

−0.15
Fig. 5.10. Computed and exact solutions using a first-order reconstruction

122

Godunov Methods

The error in amplitude is significantly reduced by using a second-order reconstruction as indicated in Fig. 5.11, where results are shown using (4.20)
with κ = 0 and no limiter together with results for the first-order reconstruction. The convergence for the second-order reconstruction (using second order accurate integration in time) is shown in Fig. 5.12 and displays
quadratic convergence.†
u/ao
0.15

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

κa t = 7

o
.
...
...
...
...
.
.
.
.....
........
...................
.......
.............
.
.
.
.
....
.
.
.
...
.....
...
.......
......
...
......
.
.
...
.
.
.
....
.
...
.
.
.
.
...
.....
.
.
.
.
...
...
.
.
.
.
.
...
....
.
.
.
.
...
...
.
.
.
.
.
...
...
.
.
.
.
...
.
....
.
.
...
.
.
....
.
.
...
.
.
....
....................
.
...
.
.
.
....
...
.
.
.
.
...
....
.
.
.
.
...
....
.
.
.
.
..
....
...
.....
.
...
.....
...
......
...
......
...
......
.
.
.
.
.
...
......
...
......
...
......
.....
.......
.......
.......
.............................

◦ computed 1st order
•••••
•••◦•◦•◦•◦•◦ •••••



• computed 2nd order


• ◦◦◦ ◦◦◦◦◦ ••
•◦•◦•◦•◦◦◦◦◦
◦◦


exact



◦◦ •
•◦◦
0.05
◦◦•
•◦•◦•◦•◦




•◦
◦◦
•◦•◦•◦
2
3
5
6
•◦•◦1•◦

•◦4




κx

0.00
◦◦
◦•◦














◦◦
◦••◦•◦•
◦◦•◦
• ◦◦◦
−0.05
◦•◦•◦•◦••






• ◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦•◦•◦•◦•◦•••
•••
••
•••••••••••••••
−0.10
0.10

−0.15
Fig. 5.11. Computed and exact solutions using first- and second-order reconstructions

5.3.4 Entropy Fix
It is interesting to compare the exact solution of the Roe equations (5.11)
∂Q
˜ l , Qr ) ∂Q = 0
+ A(Q
∂t
∂x
with the exact solution of the Euler equations (2.1)
∂Q ∂F
+
=0
∂t
∂x
for the Riemann Shock Tube (Section 2.10). The solution to the Roe equations exhibits three waves corresponding to the eigenvalues (5.23). For the
† The line in Fig. 5.12 is
n,e

||Qi

−Qn
i ||κao Δt = 

κao Δt
5 × 10−4 

2

n,e

||Qi

−Qn
i ||κao Δt=5×10−4

5.3 Roe’s Method

123

n
log10 ||Qn,e
i −Qi ||

0
−1

..
...

.
...

.
....

..
...

..
...

.

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

−2

..
...

−3

computed
quadratic

−4

−5

−6
−4

−3

−2

−1

0
log10 κao Δt

Fig. 5.12. Convergence for Roe’s Method using a second-order reconstruction

Riemann Shock Tube,
u
˜ = 0
a
˜ = 

(γ −1)cp 

ρ 1 T1 +  


√ 1/2
ρ4 T4 / ( ρ1 + ρ4 )

˜ . The Roe eigenvalues are therefore
and hence a
˜2 = (γ −1)H
˜1 =
0
λ
˜2 =
a
˜
λ
˜ 3 = −˜
a
λ
The three waves divide the x−t domain into four regions (Fig. 5.13).
The matrix S˜ is


⎨ 1

1
1
˜
S=
0
a
˜
−˜
a

⎩ 0 a
˜2 /(γ −1) a
˜2 /(γ −1)





124

Godunov Methods
t.......

dx/dt = 0

dx/dt = −˜
a

dx/dt = a
˜

.....
.....
.....
.....
.....
.....
.....
.....
.....
.
.
.
.
.....
.....
.....
.....
.....
.....
1l
2l
3r
1r
2l
3r ............
.....
..
.
.
.....
.
.
.....
....
.....
.....
.....
.....
.....
.....
.....
.....
.
.
.
.....
.
....
.....
.....
.....
.....
.....
.....
.....
.....
.....
.
.
.
.
.....
.....
.....
1l
2l
3l
1r
2r
.....
.....
.....
.....
.....
.....
....
.
.
.....
.
.
...
.....
.
.
.
.
.....
..
.....
.....
.....
....
..... .........
..... .....
....

(R , R , R )

(R , R , R )

(R , R , R )

(R , R , R3r )
.........

x

Fig. 5.13. Waves for Riemann Shock Tube

and the vector R is




ρ − (γ −1)ρe/˜
a2

−1
˜
ρu/2˜
a + (γ −1)ρe/2˜
a2
R=S Q=



−ρu/2˜
a + (γ −1)ρe/2˜
a2 ⎭

˜ is
and thus the solution Q = SR
Q=

Q=




⎨ ρ1 ⎪

0


⎩ ρe ⎪

1

for x < −˜
at




a2 ⎪
⎨ ρ1 + (γ −1)(ρe4 − ρe1 )/2˜


Q=

(γ −1)(ρe1 − ρe4 )/2˜
a
1
(ρe
+
ρe
)
1
4
2


for −˜
at < x < 0




a2 ⎪
⎨ ρ4 − (γ −1)(ρe4 − ρe1 )/2˜


(γ −1)(ρe1 − ρe4 )/2˜
a
1
(ρe
+
ρe
)
1
4
2
Q=


for 0 < x < a
˜t




⎨ ρ4 ⎪

0


⎩ ρe ⎪

4

for x > a
˜t

The static temperature and pressure, velocity, and entropy are shown in
Figs. 5.14 to 5.17 for the initial conditions p4 /p1 = 2 and T4 /T1 = 1 with
γ = 1.4. The abscissa is x/t, normalized by a1 , whereby the solution at

............ .................. ... The Roe equations yield cs /a1 = −˜ center wave is a contact surface with exact velocity cc /a1 = −0... . 5.........2 .. .. ..... .. ................................ .............. ...14... .......... x/a1 t Fig... Pressure for Riemann Shock Tube for p4 /p1 = 2 and T4 /T1 = 1 T /T1 1.......... .. .. .......... ..... 2 x/a1 t Fig..0 0........... . ... ..... .............. ... ................... Overall............. .......... ...... ....................... ...... ..... . p/p1 .... .... ........ ....... ................. .... .. Detailed discussions of the concept of the entropy condition are presented in Hirsch (1988) and Laney (1998).. ..... .......... ............ ...... . .... ................. . ...... .......... .... .............................. ............ 0....... ......... The cs /a1 = −1. The right wave is an expansion whose left and right boundaries are defined by the wave speeds cl /a1 = 0.....9 ... .. .. ... ................... ..............702978 and cr /a1 = 1.......... contact surface ..... .. . ........... ... A finite-volume algorithm solution of this example problem based on Roe’s method would not necessarily yield an expansion shock due to the effects of spatial and temporal truncation errors....... ...... ............................................ ..... ............. ................ ........... The Roe equations yield cc /a1 = 0... ...............5. temperature......... . ..... .................. ....... ..................................8 −2 contact surface Roe Exact −1 expansion fan 0 1 2 ................ ... ..... .... .......1 1....... ................. ... ....... ...... ..... . ...... ....... ... ........ ... ............... 2 1 expansion fan ...... ................... ... .............................................. . . ..... ............... the solution of the Roe equations for the pressure.... ... ...... ....... and entropy are an approximation of the exact solution of the Euler equations............ ............ 5..... ......... ....... Roe Exact shock 0 −2 −1 0 1 ..... .... The leftmost wave is a shock with exact speed a/a1 = −1.............. ...... ........................ .. ..... ...... ................ .. ..3 Roe’s Method 125 any time t can be obtained..... .................... ......... ..... The Roe equations yield a discontinuous expansion† (an expansion shock) with wave speed cs /a1 = 1........ ...... .............. ..... .... ........ . ......... ...159479..... ..... .......... .. ......................... ..... ..... .... ......... .... ..... ................... . .... Temperature for Riemann Shock Tube for p4 /p1 = 2 and T4 /T1 = 1 † This example is intended to illustrate the phenomenon of an expansion shock using Roe’s method.15........ .........247519............................... ...... velocity.......... .... ............... ....... .. .. ....................... .............. ...... ......... . shock 1.................... .. ..... ..... ..... .... ..................................... .............

.... . . where Roe eigenvalues |λ # ˆi | = |λ ˜ 2 /4εˆ ˜ i | < 2εˆ λ a + εˆ a for |λ a i ˜ i | ≥ 2εˆ ˜i| for |λ a |λ (5.. .. ...... .. ............ ......................... . ..... 2 ... −0... ...................... ...... . .. ....... . ....... . ..... ..... ........ ......... ..... ....... ... . .......... .. . 1 ...... .......... ......... ... ..... ... ...... . .............. ..... .........55) where a ˆ is a suitable velocity scale (e....... .. ........... ....................3 Fig........ . ...... 0 1 2 ... .............................1 −0... ................. .... . . .... Therefore.......... |λ −1 ˜ ˜ ˜ det(S|Λ|S ) > 0 always.. .. .. . .. The ˜ i | are replaced by the approximate eigenvalues |λ ˆ i |.. ....... ....... .... .. .......... ........... .. −0... ........... ....... .. . ..... ............... ............... ............... .......... .... ...................... ...................... ... ...... ... ..... .... ........... ..... ............... ........... . ......................1 when a ˆ=a ˜).. .......... ..... ......... ........ . ......... ........... .........1 0 0.... ..... ............ −1 0............ .. .... ... ε = 0................ ........... .. ... ......... ............... .... .............. . ............ ..2 −0....... .. 2.................... ..... . ..1 shock x/a1 t Roe Exact contact surface −0... ..... . .............. .. .. ...................... ..... . .................... 5.... .. ..... .... ... .. .. ..... .. .. ... ................. ......... . . ..g... .... .... .... ... . ...... .... .. ..... . .. x/a1 t Roe Exact ..... .. .......g..........0 .... .. ... ...... ........ . ................. .......... . . .............. ..... ..0 ................. .... ........ ......... ........ ........ .... ..3 ........... ..... . .... ....... ..... ....... . .. ...17.... .. ........ Entropy for Riemann Shock Tube for p4 /p1 = 2 and T4 /T1 = 1 Harten (1983) proposed the following remedy to the expansion shock. .. ................ .......... ........ Velocity for Riemann Shock Tube for p4 /p1 = 2 and T4 /T1 = 1 (s−s1 )/cv ... ......... ........... ........................ shock expansion fan contact surface Fig. ........ ................ .......... ... ....... . .......... .. . .......... ................. ........... .......126 Godunov Methods u/a1 ... ....... .................... ............ ........... expansion fan −0................ .. ....... 0..1 −1 0... . ....... ........ ... ... ........ ...16.... . ..... . ...... ......... ......2 .... . ... ... . .... a ˆ =a ˜) and ε is a small positive ˆ i | > 0 for i = 1... ....... ................. ............ ...... ...... 3 and number (e..... ....... 5.. ........

58) The diagonal matrix Λ is split according to Λ = Λ+ + Λ− (5. 5.5. 0) λ− k for k = 1. The Jacobian matrix A defined by A= ∂F ∂Q can be expressed as A = T ΛT −1 where ⎧ ⎪ ⎨ λ1 Λ= 0 ⎪ ⎩ 0 ⎫ 0 0 ⎪ ⎬ λ2 0 ⎪ 0 λ3 ⎭ (5. 2. 1983).57) Therefore. . Osher and Solomon. 0) k = min(λk .4.1 Algorithm Consider the Euler equations (3. 3 (5. Osher and Chakravarthy. − λk = λ+ k + λk for k = 1. The individual eigenvalues are split according to λ+ = max(λk .9) in semi-discrete form:  dQi + dt Fi+ 12 − Fi− 12 Δx  =0 where the flux Fi+ 12 = Fi+ 12 since the problem is one-dimensional. We choose to include it in this chapter on Godunov Methods since it is based on an exact solution of one of the five cases of the General Riemann Problem. 3 (5. 1982.16).56) according to (2.4 Osher’s Method Osher developed an approximate Riemann solver† based on the concept of a series of path integrations in the space of solutions of the Euler equations (Engquist and Osher. 1980.15) and (2. 2.65).59) † It could be also be argued that Osher’s Method is a Flux Vector Split Method (Chapter 6) on the basis of (5.4 Osher’s Method 127 5.

13). The flux is likewise split as F = F+ + F− (5.128 Godunov Methods where Λ+ = and Λ− = ⎧ + ⎪ ⎨ λ1 ⎫ 0 0 ⎪ ⎬ + 0 λ2 ⎪ ⎭ 0 λ+ 3 0 ⎪ ⎩ 0 ⎧ − ⎪ ⎨ λ1 0 0 − 0 λ2 0 λ− 3 0 ⎪ ⎩ 0 As an example. λk are eigen− values of A− .61) ⎫ 0 0 ⎪ ⎬ + 0 u+a 0 Λ = ⎪ ⎪ ⎩ 0 0 0 ⎭ and (5.63) The Jacobian matrix is therefore split according to A = A+ + A− (5.62) ⎫ 0 ⎪ ⎬ Λ− = 0 0 0 ⎪ ⎩ 0 0 u−a ⎪ ⎭ (5. Then ⎧ ⎪ ⎨ u ⎫ ⎪ ⎬ ⎪ ⎭ (5. and the right eigenvector corresponding to both λ+ k and λk is rk given by (2.67) and thus + − Fi+ 12 = Fi+ 1 + Fi+ 1 2 2 (5.60) ⎧ ⎪ ⎨ 0 0 (5.65) where ∂F + ∂Q ∂F − ∂Q = A+ (5.64) where A+ = T Λ+ T −1 A− = T Λ− T −1 − + It is straightforward3 to show that λ+ k are eigenvalues of A .66) = A− (5. assume 0 < u < a.68) .

5. from (5. Fi+ 12 = F + (Ql ) + F − (Qr ) − = F (Q ) + F (Q ) + + l   l ! F (Ql ) Qr Ql ! F − (Qr ) − = F (Q ) + F (Q ) + +  l l "  Qr Ql A− dQ " A− dQ .4 Osher’s Method 129 + The term Fi+ 1 represents the contribution to the flux associated with the 2 waves that move from left to right across the interface xi+ 21 since A+ = T Λ+T −1 and Λ+ is the diagonal matrix of eigenvalues that are positive + at xi+ 12 .69) 2 The essence of Osher’s method is to determine an algorithm for Fi+ 12 that is equivalent to (5.  Qr Ql  + − A dQ = Qr Ql  Qr Ql ∂F + dQ = F + (Qr ) − F + (Ql ) ∂Q ∂F − dQ = F − (Qr ) − F − (Ql ) ∂Q and therefore F (Q ) = F (Q ) − + + l  r F − (Qr ) = F − (Ql ) +  Qr A+ dQ Ql Qr A− dQ Ql Thus. it is reasonable to use Qli+ 1 to determine Fi+ 1 2 2 − and Qri+ 1 to determine Fi+ 1 according to 2 2 2 Fi+ 12 = F + (Qli+ 1 ) + F − (Qri+ 1 ) 2 (5. Consider the integral  Qr Ql A+dQ where we omit the subscript i + 12 on Ql and Qr in the remainder of this section for simplicity. Thus. the term Fi+ 1 represents the contribution to the flux 2 associated with the waves that move from right to left across the interface xi+ 21 since A− = T Λ− T −1 and Λ− is the diagonal matrix of eigenvalues that + are negative at xi+ 1 . Similarly. From (5.66).67).69) but avoids entirely the need to actually compute either F + or F − .  Qr Ql A dQ = Similarly.

... ............. ... ... . ..... . . .. .. .. . .......... . ............. .. ........ A similar expression for Fi+ 12 can be obtained in terms of F(Qr ) and the integral of A+ .. ...... . ...... ...... .. .. .................... . ... ........ ..... . .............. ....9. 5....... ..... .... . .. .... . ....... . .... .. .....18.... . ....... ....... ... ...... ............. . ..... ...18.. ... ...130 Godunov Methods Thus.. ....... ... ....... ... t . .. .... . ... ......... .. . ..4) ⎡ cc 2a1 ⎣ = u1 + 1− (γ − 1) u2 = cc p2 = p ∗ ρ2 = ρ1 u3 = cc & p∗ p1 '1 γ & p∗ p1 ' (γ−1) 2γ (5.. . .... ... ...7) ⎡ p∗ = ⎤ (γ−1) ⎢ a1 + a4 + 2 (u1 − u4 ) ⎥ ⎣ ⎦ (γ−1) (γ−1) − − a1 p1 2γ + a4 p4 2γ 2γ γ−1 The solutions in Regions 2 and 3 are (Section 2...... ................. ..... Expression (5... .. .... .. Fi+ 12 = F(Q ) +  l Qr Ql A− dQ (5........... ............ . ..... ...70) which defines Fi+ 12 without specific mention of either F + or F − . .. . 1 contact surface expansion fan no. ...... .... . .. ...... ....... The wave structure is shown in Fig.. ....... .... ............... . .. ... .. 2 . .. 5.... .... . ........ .... ........ .......... ................... .. . .. .. .. . ............ .... . . ... .. ........... ........ . ..70) is evaluated using the exact solution to the general Riemann problem for two expansions (Section 2... ... . .... . expansion fan no.... ..... ...... .... ...... .. This expression is then assumed to be approximately valid for all cases........ The contact surface pressure p∗ is (Exercise 5..4)..71) ⎤ ⎦ .... The leftmost state (denoted by Q1 ) corresponds to Ql and the rightmost state (denoted by Q4 ) corresponds to Qr . . . .... . ... ............ . . ..... ... . . .. ... ..... ..... .. ...... ...... . .. ... ... . .... .. ... .9....... ... . .... ......... . ................... ....... ... General Riemann Problem: Case 4 (two expansions) A closed-form expression for the entire solution can be obtained.... . ... ........ .... 2 3 1 4 .. .. x Fig.. . .... .. . ........ .. ..

149) to (2. For each integral Ij we consider an integral curve in the space of solutions Q defined by dQ = σj (Q) dζ for 0 ≤ ζ ≤ ζ1j on Ij where σj depends on Q in some manner yet to be specified. We evaluate the integral in (5.5. . and so on.13) of A: σ j = rk where rk is one of the following: ⎧ ⎪ ⎨ ⎫ ⎪ 1 ⎬ r1 = .135) to (2. The boundaries of the expansion fans are given by (2. respectively. u ⎪ ⎩ 1 u2 ⎪ ⎭ 2 ⎧ ⎪ ⎨ ⎫ ⎪ 1 ⎬ r2 = .151).70) by decomposing it into three separate components:  Qr Ql A− dQ =   2 1 A− dQ + ! " I1   3 2 A− dQ + ! I2 "   4 3 A− dQ ! I3 (5.138). u+a ⎪ ⎩ H + ua ⎪ ⎭ ⎧ ⎪ ⎨ ⎫ ⎪ 1 ⎬ r3 = u−a ⎪ ⎩ H − ua ⎪ ⎭ and the value of k depends on the integral Ij in some manner yet to be † Note that the actual integral is in the three-dimensional space of solutions Q. 2. 3 A particularly useful choice for σj is one of the right eigenvectors (2.148) and by (2. Then Ij =  j  = j  = j j+1 A− dQ j+1 A− dQ dζ dζ j+1 A− σj dζ for j = 1.72) " ( where I1 = 12 A− dQ indicates the integral† from Region 1 to Region 2.146) to (2.4 Osher’s Method p3 = p ∗ & ρ3 = ρ4 p∗ p4 131 '1 γ The solutions within the left and right expansion fans are given by (2. not in the x−t plane.

.73) = rk for 0 ≤ ζ ≤ ζ1j dζ Let s denote the location in the space of solutions Q where λk changes sign if such a point exists.74) Assume that λk < 0 for Ij1 and λk ≥ 0 for Ij2.  Ij =  j  = j s A− dQ + ! Ij1 " s  j+1 A− dQ s λ− k rk dζ + ! Ij2  " j+1 s λ− k rk dζ (5. Divide the integral Ij into two segments. 2.. Hence λ− j2 k Ij =  j  = j s λ− k rk dζ s A− dQ  s& = = = = = ' ∂F − A+ dQ j ∂Q  s  s ∂F dQ − A+ dQ ∂Q j j  s  s ∂F dQ − λ+ k rk dζ j ∂Q j  s ∂F − dQ since λ+ k = 0 when λk = 0 j ∂Q Fs − Fj where Fs = F(Qs ) and Fj = F(Qj ) for j = 1.132 Godunov Methods − specified. This is denoted the sonic point. 3 j+1 λ− k rk dζ j Assume furthermore that for each integral Ij there is a particular eigenvector − rk for which λ− k is monotonic. . i. Since λ− k is an eigenvalue of A with a right eigenvector rk . A− rk = λ− k rk and for  Ij = k = 1. λk changes sign at most one time on the integral curve dQ (5.e. . Note that k = j in general. 4. . Then λ− k = 0 for Ij1 and = 0 for I . If λk ≥ 0 for Ij1 and .

Thus. The density decreases monotonically from Region 1 to Region 2 since they are connected by an expansion. Table 5. we may choose ζ = ρ − ρ1 with no loss of generality. then Ij = Fj+1 − Fj In summary. Equation (5. It remains to demonstrate the two properties shown in Table 5.73) it is evident that ζ has the units of density.5. then a similar analysis yields Ij = Fj+1 − Fs If λk ≥ 0 for the entire path in Ij . From (5. Thus.1. Fi+ 12 in (5. then Ij = 0 Similarly.1 for each integral curve Ij and determine expressions for Q at the sonic points.73) then becomes ⎛ ⎞ ⎜ ⎟ dρ ⎜ ⎟ ⎜ dρ ⎟ ⎜ dρu ⎟ ⎟ dQ ⎜ ⎟ =⎜ ⎜ ⎟ dρ dρ ⎜ ⎟ ⎜ ⎟ ⎜ dρe ⎟ ⎝ ⎠ dρ . Requirements for Integral Curves Property Requirement for Ij 1 There exists a particular eigenvector rk such that the integral curve dQ/dζ = rk connects the solution points Qj and Qj+1 The eigenvalue λk is monotonic on this integral curve 2 We proceed by construction for each integral beginning with I1 . ⎧ Fj+1 − Fj ⎪ ⎪ ⎪ ⎨ F −F s j Ij = ⎪ F − Fs j+1 ⎪ ⎪ ⎩ 0 if if if if λk λk λk λk <0 <0 ≥0 ≥0 for for for for Ij Ij1 and λk ≥ 0 for Ij2 Ij1 and λk < 0 for Ij2 Ij (5.69) is simply the sum of evaluations of F at these specific points. if λk < 0 for the entire path in Ij .4 Osher’s Method 133 λk < 0 for Ij2 .75) This form permits the integrals Ij to be expressed in terms of the flux F evaluated at specific points in the space of solutions Q. and all flow variables are then functions of the single variable ρ in the expansion.

the second term is dρu dρ and the third term is dρe dρ du dρ 2 da = u− ρ (γ −1) dρ = u−a = u+ρ & 1 d ρa2 + 12 ρu2 dρ γ(γ −1) a2 + 1 u2 − ua = (γ −1) 2 = H − ua ' = where H is the total enthalpy. 2 da du =− dρ (γ −1) dρ Moreover.76) u+ a = u1 + a1 (γ −1) (γ −1) in the left expansion (Exercise 5. Thus. ⎛ ⎞ 1 dQ ⎜ ⎟ = ⎝ u − a ⎠ = r3 dρ H − ua and Property 1 is satisfied. Next. the speed of sound in the expansion is / a= and therefore γp = ρ 0 γp1 (γ−1)/2 ρ ργ1 (γ −1) a da = dρ 2 ρ Hence. To determine the second element.148). we note that 2 2 (5. u−a= x t for cl1 ≤ x ≤ cr1 t (5.77) . Thus. we note that λ3 = u − a Using (2.8).134 Godunov Methods The first element in dQ/dρ is 1.146) to (2.

80) since ρ3 e3 − ρ2 e2 = = p3 p2 + 12 ρ3 u23 − − 1 ρ2 u22 (γ −1) (γ −1) 2 2 1 2 ρ3 u3 − 12 ρ2 u22 (5. then the flow conditions at this sonic point are (Exercise 5.78) Since a1s > 0.2.79) The four possible cases for I1 are determined using (5. we consider the possibility of λ3 = 0 on the integral curve. the left sonic point exists only if u1 > − 2 a1 (γ −1) (5. To evaluate (5.75) and are presented in Table 5. For I2 we choose ζ = ρ − ρ2 and ⎧ ⎪ ⎨ 1 σ2 = r 1 = Then  ρ3 ρ2  r1 dρ = r1 u ⎪ ⎩ u2 ⎪ ⎭ ρ3 ρ2 ⎫ ⎪ ⎬ dρ since u2 = u3 = (ρ3 − ρ2 )r1 ⎛ ⎞ ρ3 − ρ2 ⎜ ⎟ = ⎝ ρ3 u3 − ρ2 u2 ⎠ 1 ρ u2 − 1 ρ u2 2 3 3 2 2 2 = Q3 − Q2 (5.9) u1s a1s ρ1s p1s (γ −1) 2 u1 + a1 (γ +1) (γ +1) = u1s = & ' a1s 2/(γ−1) = ρ1 a1 & ' a1s 2γ/(γ−1) = p1 a1 (5. If this occurs.75).4 Osher’s Method 135 Hence.81) . λ3 is monotonic on the integral curve dQ = r3 dρ and Property 2 is satisfied.5.

we note that 2 2 a = u4 − a4 (5. Thus. Property 1 is proven.2. the curve is denoted linearly degenerate. For I3 we choose ζ = ρ − ρ3 and consider ⎛ ⎞ dρ ⎜ ⎟ ⎜ dρ ⎟ ⎟ ⎜ ⎜ dρu ⎟ ⎟ ⎜ dQ ⎜ = ⎜ dρ ⎟ ⎟ dρ ⎟ ⎜ ⎟ ⎜ ⎜ dρe ⎟ ⎠ ⎝ dρ The first element in dQ/dρ is 1. the second term is dρu dρ du dρ da 2 ρ = u+ (γ −1) dρ = u+a = u+ρ and the third term is dρe dρ = d dρ & 1 ρa2 + 12 u2 γ(γ −1) ' † If the integral curve vector σj is constant.136 Godunov Methods since p3 = p2 .82) u− (γ −1) (γ −1) in the right expansion (Exercise 5. du 2 da = dρ (γ −1) dρ The speed of sound in the right expansion is / a= γp = ρ 0 γp4 (γ−1)/2 ρ ργ4 and therefore (γ −1) a da = dρ 2 ρ Hence. . The two possible cases for I2 are presented in Table 5. To determine the second element. Thus.8). Since λ1 = u is constant in Regions 2 and 3. λ1 is monotonic† and hence Property 2 is proven.

2.2. u+a= x t for cl2 ≤ x ≤ cr2 t Hence. we note that λ2 = u + a Using (2. then the flow conditions at this point are a4s (γ −1) 2 u4 − a4 (γ +1) (γ +1) = −u4s ρ4s = ρ4 u4s p4s = & ' a4s 2/(γ−1) a4 & ' a4s 2γ/(γ−1) = p4 a4 (5. .85) The four possible cases for I3 are determined using (5.4. there are 32 different possible cases.83) and Property 1 is satisfied.3.149) to (2. Thus.84) Since a4s > 0.75) and are presented in Table 5. If this occurs. four combinations of conditions for Regions 2 and 3 are inadmissible as indicated in Table 5. Next. The admissible cases yield the flux formulas for Fi+ 12 shown in Table 5.5. A sonic point exists if λ2 = 0 on the integral curve. However.151). λ2 is monotonic on the integral curve dQ = r2 dρ and Property 2 is satisfied. ⎛ ⎞ 1 dQ ⎜ ⎟ = ⎝ u + a ⎠ = r2 dρ H + ua (5. the right sonic point exists only if u4 < 2 a4 (γ −1) (5. From Table 5.4 Osher’s Method 137 a2 + 1 u2 + ua (γ −1) 2 = H + ua = where H is the total enthalpy.

however. this approach is inaccurate or fails when |u1 −u4 | 0 (Toro. Inadmissible Cases Case u2 − a2 u2 − a2 u2 − a2 u2 − a2 ≥0 ≥0 ≥0 ≤0 cc cc cc cc Reason <0 <0 ≥0 ≥0 u3 + a 3 u3 + a 3 u3 + a 3 u3 + a 3 ≥0 ≤0 ≤0 ≤0 implies implies implies implies a2 a2 a3 a3 <0 <0 <0 <0 Table 5. 1997). Flux Formulas for Osher’s Method u1 − a1 ≥ 0 u4 + a4 ≥ 0 u1 − a1 ≥ 0 u4 + a4 ≤ 0 u1 − a1 ≤ 0 u4 + a4 ≥ 0 u1 − a 1 ≤ 0 u4 + a 4 ≤ 0 F1 F1 + F4 − F4s F1s F1s + F4 − F4s 0 ≤ cc ≤ a2 F1 + F2 − F1s F1 + F2 − F1s +F4 − F4s F2 F2 + F4 − F4s −a3 ≤ cc ≤ 0 F1 − F1s + F3 F1 − F1s + F3 +F4 − F4s F3 F3 + F4 − F4s cc ≤ −a3 F1 − F1s + F4s F1 − F1s + F4 F4s F4 a2 ≤ cc .2. u2 − a2 u2 − a2 u2 − a2 u2 − a2 ≥0 ≤0 ≤0 ≥0 ≥ 0. ≥ 0. Evaluation of Integrals Integral ( j+1 − A dQ j Case I1 u1 − a1 u1 − a1 u1 − a1 u1 − a1 cc < 0 cc ≥ 0 u3 + a3 u3 + a3 u3 + a3 u3 + a3 I2 I3 ≥ 0. ≤ 0. Table 5. ≤ 0.138 Godunov Methods Table 5. An alternate set of integration paths (the original ordering or O-ordering) was originally proposed by Osher.4. 1997).3. ≥ 0. ≤ 0. ≤ 0. u4 + a4 u4 + a4 u4 + a4 u4 + a4 ≥0 ≥0 ≤0 ≤0 0 F2 − F1 F2 − F1s F 1s − F 1 F3 − F2 0 0 F 4s − F 3 F4 − F4s F4 − F3 These integration paths employed for Ij are denoted the physical ordering or P-ordering (Toro.

The Euler equations (3.87) where A = A(Qi ) is treated as a constant. For purposes of simplicity. 1 dQi = A (Qi−1 − Qi ) dt Δx (5. Then Fi+ 1 2 = F(Qli+ 1 ) Fi− 12 = F(Qli− 1 ) 2 2 We further assume a simple first-order reconstruction: Qli+ 1 = Qi Qli− 1 = Qi−1 2 2 Expanding in a Taylor series about Qi and neglecting higher order terms. Qri+ 1 .4: u1 − a1 ≥ 0 u4 + a4 ≥ 0 ≤ cc a2 ⎫ ⎪ ⎬ ⎪ ⎭ (5.86) and assume that this condition holds at the cell interfaces i ± 12 .4. we assume that the flow is periodic in x over a length L = (M − 1)Δx. Qli− 1 . where M = 2N + 1 and N is an integer.5.9) are  Fi+ 12 − Fi− 12 dQi + dt Δx  =0 The flux Fi+ 12 depends on the left and right states: Fi+ 12 = F (Qli+ 1 .9) as dQi = Ri (Qli+ 1 .4 Osher’s Method 139 5. Qri− 1 ) 2 2 2 2 dt where  Fi+ 12 − Fi− 12 Ri = −  Δx We consider a specific case of Osher’s Method from Table 5. Consider . Qri+ 1 ) 2 2 We rewrite (3.2 Stability We now consider the semi-discrete stability of Osher’s Method.

t) in a Fourier series.87) yields ˆk dQ ˆk = GQ dt where the amplification matrix G is  1  −ιkΔx e −1 A Δx G= Since G is a constant multiple of A. l=N  Q(x. t) = ˆ k (t)eιkx Q l=−N +1 √ where ι = −1 and the wavenumber k depends on l according to k= 2πl L Substituting this into (5. t) to be a continuous vector function that interpolates Qi and expand Q(x. G = T DT −1 where D=  1  −ιkΔx e −1 Λ Δx It is evident that the eigenvalues of G are λGm =  1  −ιkΔx e − 1 λm Δx where the λm are given in (2. Defining ˜ k = T −1 Q ˆk Q we have ˜k dQ ˜k = DQ dt Defining ˜k = Q ⎧ ⎫ ˜ ⎬ ⎪ ⎨ Qk1 ⎪ ⎪ ⎩ Q˜k2 ⎪ Q˜k ⎭ 3 the solution is then ˜k = Q ˜ k (0)eλGm Q m m .12).140 Godunov Methods Q(x.

1). which is prior to the shock formation.20 for κao t = 7 and C = 0.§ † The line in Fig. is evaluated at κao t = 7.e ||Qi −Qn i ||κao Δt=5×10−4 .e i −Qi ||κao Δt =  κao Δt 5 × 10−4 2 n. The linear convergence is a direct consequence of the linear reconstruction (5. § The line in Fig.21. The error in amplitude is significantly reduced by using a second-order reconstruction as indicated in Fig. The convergence is displayed in Fig. The solution converges linearly† to the exact solution for sufficiently small Δt.4 5. which corresponds to C = 0. 2. the algorithm is stable. The convergence for the second-order reconstruction (using second-order accurate integration in time) is shown in Fig. 5. 5. defined by (3.3 Accuracy. The initial condition is defined by (2. Consistency. The computed result (using 100 cells) and exact solution are displayed‡ in Fig.45. where results are shown using (4.92) with  = 0.4 Osher’s Method 141 The condition for stability is therefore Real (λGm ) ≤ 0 for m = 1.e i −Qi ||κao Δt=5×10−4 ‡ More precisely.4. The speed of the disturbance is accurately predicted.102).025. the timestep κao Δt = 0.1). Therefore.1 and the domain is 0 < κx < 2π.86).20) with κ = 0 and no limiter together with results for the first-order reconstruction.22 is n ||Qn. and Convergence We now consider the problem described in Section 2.5.22 and displays quadratic convergence.e i −Qi ||κao Δt =  κao Δt 5 × 10−4  n ||Qn.45 at t = 0. 5. however. The first-order reconstruction (5. 5. The semidiscrete form (3. The norm. The significant error in amplitude is attributable to the dissipative nature of the first-order reconstruction (5.19 is n ||Qn. 5. λm ≥ 0.8.1) is employed.9) of the Euler equations is employed and the time integration is performed using a second-order Runge-Kutta method (Chapter 7). 5.19. 3 Now Real (λGm ) = 1 (cos kΔx − 1) λm ! " Δx  ≤0 From (5.

.... .. . ....05 ◦◦ ◦◦◦◦ ◦◦ ◦◦◦◦ ◦ ◦ ◦ ◦◦4 ◦1◦ ◦ 2 3 5 6 ◦ ◦ ◦◦ ◦◦ 0... ...... .. .. ....... .. .. . .... . .... . Computed and exact solutions using a first-order reconstruction Exercises 5... ◦ ◦ ◦ ◦ computed linear ◦ ◦ ◦ −3 −4 −3 −2 −1 0 log10 κao Δt Fig.. . . ... . . .. .. . . .. ... .. ........ ............ .... ... ... .... .. ............. ... . . . .... .... .... . . . .... .. . ... . . ... .. .. . ........ . ........ 5.... ......... .. ... . .. ......... ....... ... . .. . .. . . .. .. ..... ..00 ◦ ◦ ◦ ◦ ◦ ◦ ◦◦ ◦◦◦◦ ◦◦ ◦◦ ◦◦◦◦ −0.......... ... . .. .... . .... .........1 Prove the identities (5...... ..... . ... . solution a) Δ(f + g) = Δf + Δg . ...... .... ... . . . . ◦ ◦ −2 . .... .15 ..... .. .... .. ....... ..... ...... ... .. ... ....... . . .. ... . . ... .... .. .. ..... ... ........ 0..... . . .. ..... ... . . .12)..10 0....... . ... ◦ ... ... ....... . ...... ...15 Fig......... κa t = 7 o .... ... .. ......... ... . .. . ..05 ◦ ◦◦ ◦ ◦ ◦◦ ◦◦◦ ◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦ −0. .... . .10 κx −0. ... . .. .... ........ ...... .... ...... . . . . ..... . . .... . .... ◦ computed ◦ ◦ ◦ ◦ ◦ exact ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦◦◦ ◦◦ ◦◦ ◦◦◦◦◦ ◦ ◦ ◦ ◦ 0.. .... . ....... . ........ 5. .. .... ..20.... .. .. .. .. . ......... . . .. .. ... ........... .. .... . . .. ... .. Convergence for Osher’s Method using a first-order reconstruction u/ao ....... ..... .. . ....... ....... ... ........ .........142 Godunov Methods n log10 ||Qn.. ... ..... ........ .... . ....... . ....e i −Qi || 0 −1 ...... ... ..... .19.. .. .. ... .. . ....... . . ...... ........ . ... . .. .... ........ .... . ...

..... .....05 ◦•◦•◦•◦•• ◦ ◦ • ◦ • ◦ • ◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦•◦•◦•◦•◦••• ••• •• ••••••••••••••• −0. ...... .... . . . .. . .. ... .... ... ..... ... . ..... . . . ....00 ◦◦ ◦•◦• • ◦ ◦ • ◦ • ◦ • ◦ • ◦ • ◦◦ ◦••◦•◦• ◦◦•◦ • ◦◦◦ −0. ... .. ...... .and second-order reconstructions The proof is straightforward. ... . . o .......Exercises 143 u/ao . ... . . Derive the intermediate matrices B and C used in the derivation of Roe’s method.2 Consider the quadratic function f (ν1 .. . ... .. .. . and d3 depend on the coefficients a.. .. ... . .. .... .. ... ν2 .. . .. . ... .. . ..... . .. .......... . ... ... d2 ... ... . .. . Computed and exact solutions using first. ..... .. ....3 where d1 ........... .. ◦ computed 1st order ••••••••• ••◦•◦•◦•◦•◦◦◦◦◦ •••• • • • • computed 2nd order • ◦ ◦◦◦◦ • •◦•◦•◦•◦◦◦◦◦ ◦◦ • • exact ◦ • ◦ ◦◦ • •◦•◦◦ 0... . ..... ..... . . .... . .... ... .. .. ... . .... .. . .... b) Δ(f g) = f¯Δg + Δf g¯ f¯Δg + Δf g¯ = = 1 (f + 2 l 1 (fl gl 2 = 1 2 = Δ(f g) fr )(gl − gr ) + (fl − fr ) 12 (gl + gr ) − fl gr + fr gl − fr gr + fl gl + fl gr − fr gl − fr gr ) (2fl gl − 2fr gr ) c) Δ(1/f ) = −Δf /fˆ2 Δ(1/f ) = fl−1 − fr−1 = (fr − fl )/fl fr −Δf /fˆ2 = 5.. ...... ci and ν¯1 . . ... .10 0.. ..05 ◦◦• •◦•◦•◦•◦ • ◦ • ◦ ◦◦ •◦•◦•◦ 2 3 5 6 ◦•◦•◦1•◦ •◦4 • ◦ • ◦ κx • ◦ 0. .... .. . ..... ....... ... ..21.. ... . ....... . ...... .... .. . .. ...15 . .. κa t = 7 0...... . . .. ......... ...... . ..... .... ....... . . . ...... . .... ... ν3 ) = a + b1 ν1 + b2 ν2 + b3 ν3 + c1 ν12 + c2 ν1 ν2 + c3 ν1 ν3 + c4 ν22 + c5 ν2 ν3 + c6 ν32 Prove that Δf = d1 Δν1 + d2 Δν2 + d3 Δν3 5........ .... .15 Fig. .... . . . .... ν¯2 and ν¯3 .... .. . .... .. ... bi .. . . ............. ... ..10 −0. ... . 5... .. .....

.. .... ....... .. .144 Godunov Methods n log10 ||Qn..12)... Derive the eigenvalues (5..... . . . . .. ... ...... ..24) of the Roe ma˜ trix A...... # ΔQ = 2¯ ν1 Δν1 ν¯2 Δν1 + ν¯1 Δν2 ν¯3 Δν1 /γ + ν¯1 Δν3 /γ + (γ −1)¯ ν2 Δν2 /γ and therefore # B= Also... . ..... .... .. ... .. ..... ◦ ◦ ◦ ◦ ◦ −3 computed quadratic ◦ −4 ◦ ◦ −5 ◦ −6 −4 −3 −2 −1 0 log10 κao Δt Fig.... .. .. .. . ............... . ... . . .22.. .... .... .. ◦ ........... ...... .. ....20)... . . ...... .............. .... ...... .... . ...... . . ... . . . ... .......... .....5 2¯ ν1 ν¯2 ν¯3 /γ 1 ν¯2 (γ −1)¯ ν3 /γ 0 ν¯1 (γ +1)¯ ν2 /γ ν¯3 0 (γ −1)¯ ν1 /γ ν¯2 1 1 Derive the Roe matrix A˜ using A˜ = CB −1 and (5.... .. Convergence for Osher’s Method using a second-order reconstruction solution From (5... .....4 5.. .. ..... . ..... . . ... ...... ... .. .... . 5... .. . # ΔF = 0 ν¯1 (γ −1)¯ ν2 /γ 0 0 ν¯1 /γ 1 ν¯2 Δν1 + ν¯1 Δν2 (γ −1)¯ ν3 Δν1 /γ + (γ −1)¯ ν1 Δν3 /γ + (γ +1)¯ ν2 Δν2 /γ ν¯3 Δν2 + ν¯2 Δν3 and therefore # C= 5.... ... .. ...e i −Qi || 0 −1 ..... . . .. . . . −2 . .... .. . .. ..23) and eigenvectors (5. ...17) using (5.... ..

2 (γ −1)   a1 p∗ p1 (γ−1)/2γ  + a4 p∗ p4 (γ−1)/2γ  = u1 − u4 + 2 (a1 + a4 ) (γ −1) . The remaining eigenvectors are obtained using a similar approach.21). ˜ = H a ˜2 + 12 u ˜2 (γ −1) which yields ˜ 3 + 3˜ ˜ 2 + (˜ ˜+u −λ uλ a2 − 3˜ u2 )λ ˜3 − u ˜a ˜2 = 0 which directly factors into ˜ −u ˜ − (˜ ˜ − (˜ (λ ˜)(λ u+a ˜))(λ u−a ˜)) = 0 ˜1 = u Denote the eigenvector e˜1 corresponding to the eigenvalue λ ˜ as # e˜1 = υ1 υ2 υ3 1 Then ˜e1 = λ1 e˜1 A˜ Thus. solution From (2.6 Show that the equivalent Rankine-Hugoniot conditions (Section 2.3) for the Roe equations (5. Using (5. the above equations yield υ2 = u ˜ and υ3 = 12 u ˜2 .25).98) and (2. υ2 (γ −3) 2 u ˜ υ1 + (3−γ)˜ uυ2 + (γ −1)υ3 2 (γ −1) 3 ˜u ˜ − (γ −1)˜ u ˜ )υ1 + (H (−H ˜+ u)υ2 + γ u ˜υ3 2 = u ˜υ1 = u ˜υ2 = u ˜υ3 We may arbitrarily choose υ1 = 1.7 Derive the expression for the contact surface pressure (5.Exercises 145 solution By definition.99). ˜ λ 2   ˜ (3−γ)˜ u−λ 2 +   ˜ − (γ −1) H ˜ − (γ −1)˜ γu ˜−λ u2  (γ −3) 3  3 u ˜2 ˜3 ˜ − (γ −1) (γ −1) u ˜u (γ u ˜ − λ) −H ˜ 2 2 =0 From (5. since p∗ < p1 and p∗ < p4 for Case 4. 5.71) for the General Riemann Problem Case 4 (expansion-expansion).25). ˜ =0 det(A˜ − λI) Using (5.11) are ˜ AΔQ − uw ΔQ = 0 5.

10 Derive the admissible cases in Table 5.8 2γ − + a4 p4 − u4 ) (γ−1) 2γ 2γ ⎤ γ−1 ⎦ For the General Riemann Problem Case 4 (expansion-expansion).84). Since the Riemann invariant (5.78). show that in the left expansion u+ 2 2 a = u1 + a1 γ −1 γ −1 and in the right expansion u− 5. solution At the left sonic point. . 2 2 u1 + a1 = u1s + a1 γ −1 γ −1 s These equations may be solved to obtain u1s = (γ −1) 2 u1 + a1 (γ +1) (γ +1) Since the flow is isentropic within the left expansion.82) yields 2 2 u4 − a4 = u4s − a4s γ −1 γ −1 and the right sonic point λ2s = u2s + a2s = 0 yields (5.4.76) applies in the left expansion.146 Godunov Methods Solving for p∗ yields ⎡ p∗ = (γ−1) (u1 2 4+ ⎣ a1 + a(γ−1) − a1 p 1 5.  ρ1s = ρ1 p1s = p1 a1s a1 2/(γ−1)  a 2γ/(γ−1) 1s a1 A similar analysis applied to the right expansion using the Riemann invariant (5.9 2 2 a = u4 − a4 γ −1 γ −1 Show that the left sonic point is defined by (5. λ3s = u1s −a1s = 0. 5.

Laney (1998) 6. Culbert B. Thus it is reasonable to use Qli+ 1 to evaluate F + . The basic concept of flux vector splitting is to decompose the flux vector F into two parts.1 Introduction Consider the Euler equations (3. F = F+ + F− where ∂F + ∂Q ∂F − ∂Q has nonnegative eigenvalues has nonpositive eigenvalues The term F + represents the contribution to the flux associated with waves that move from left to right across the cell interface at i + 12 since the eigenvalues of its Jacobian ∂F + /∂Q are positive (or zero). Similarly. the term F − represents the contri2 bution to the flux associated with waves that move from right to left across 147 . aside from Riemann solvers. as far as characteristic modeling goes.9) in semi-discrete form:  Fi+ 12 − Fi− 12  dQi + =0 dt Δx where the flux Fi+ 12 = Fi+ 12 since the problem is one-dimensional. the only currently available alternative is flux vector splitting.6 Flux Vector Splitting Methods In fact.

1) (γ − 1) − a2 (γ − 1) 2a2 (γ − 1) 2a2 ⎫ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎬ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎭ (6. For additional algorithms.1 Algorithm From (2.148 Flux Vector Splitting Methods the cell interface since the eigenvalues of its Jacobian ∂F − /∂Q are negative (or zero). see Laney (1998) and Toro (1997). Thus it is reasonable to use Qri+ 1 to evaluate F − .2.15). Numerous al2 gorithms for F + and F − have been developed. see also Sanders and Prendergast (1974). We present two methods in this chapter. A = T ΛT −1 where ⎧ ⎪ ⎨ λ1 Λ= 0 ⎪ ⎩ 0 ⎫ 0 0 ⎪ ⎬ λ2 0 ⎪ 0 λ3 ⎭ with λ1 = u λ2 = u + a λ3 = u − a and ⎧ ⎪ ⎨ ⎫ ⎪ 1 1 1 ⎬ T = u u+a u−a ⎪ 1 2 ⎪ ⎩ H + ua H − ua ⎭ 2u T −1 ⎧ (γ − 1) u2 ⎪ ⎪ ⎪ 1 − ⎪ ⎪ 2 a2 ⎪ ⎪ ⎪ ⎪ ⎨ (γ − 1) u2 1u − = 2 ⎪ 4 a 2a ⎪ ⎪ ⎪ ⎪ 2 ⎪ 1u (γ − 1) u ⎪ ⎪ ⎪ + ⎩ 2 4 a 2a u (γ − 1) 2 a (γ − 1) u − + 2 a2 (γ − 1) u − − 2 a2 1 2a 1 2a (6.2 Steger and Warming’s Method Steger and Warming (1981) developed a flux vector split algorithm based on Euler’s identity F = AQ. 6.2) . 6.

− λ i = λ+ i + λi (6.6) Thus Λ = Λ + + Λ− where Λ+ = ⎧ + ⎪ ⎨ λ1 0 0 0 λ+ 2 0 λ+ 3 0 ⎪ ⎩ 0 ⎫ ⎪ ⎬ ⎪ ⎭ and Λ− = ⎧ − ⎪ ⎨ λ1 0 ⎪ ⎩ 0 0 0 0 λ− 2 0 λ− 3 ⎫ ⎪ ⎬ (6. F = AQ = T ΛT −1 Q T −1 Q + .7) ⎪ ⎭ Using (2.2 Steger and Warming’s Method 149 Steger and Warming assumed a wave speed splitting.3) with = λ+ i λ− i = This is equivalent to λ+ i = λ− i = 1 2 1 2 (λi + |λi |) (λi − |λi |) (6.15) and (2.4) λi if λi ≥ 0 0 if λi < 0 (6.24).5) 0 if λi ≥ 0 λi if λi < 0 (6.6.

T Λ−  T −1 Q = .

In the context of Steger and Warming’s method. For Case 1. we define the Mach number M = u/a and interpret a negative Mach number to imply u < 0.2). .8) F− ⎡ ⎤ ⎡ ⎤ 1 1 1 (γ −1) ± ⎢ ρ ±⎢ ρ ±⎢ ⎥ ⎥ ⎥ = ρλ1 ⎣ u ⎦ + λ2 ⎣ u + a ⎦ + λ3 ⎣ u − a ⎦ (6. The fluxes F + and F − can be determined from (6.1 (Exercise 6. all eigenvalues are negative.9) as described below.1) ⎡ F± ⎤ (6.9) γ 2γ 2γ 1 2 H + ua H − ua 2u There are four possible cases for the eigenvalues λ± i depending on the Mach number† M as indicated in Table 6.T Λ+  F+ The terms F ± can be written as (Exercise 6. and the wave speed splitting † The Mach number is strictly a nonnegative quantity defined by M = |u|/a.

.......... All waves intersecting the cell face at xi+ 1 originate from the right as illustrated in Fig.... .. . . .. .. Case 1 For Case 2.. . .... 3. which is drawn for M = − 12 .. ........ . . ..... ..... Four Cases for Eigenvalues Case 1 2 3 4 Range M −1 < M 0<M M ≤ −1 ≤ 0 ≤ 1 ≥ 1 λ+ 1 λ+ 2 λ+ 3 λ− 1 λ− 2 λ− 3 0 0 λ1 λ1 0 λ2 λ2 λ2 0 0 0 λ3 λ1 λ1 0 0 λ2 0 0 0 λ3 λ3 λ3 0 − yields λ+ i = 0 and λi = λi < 0 for i = 1..... . . ......... ............. λ2 = λ2 .... 2 .... λ λ λ xi+ 12 . ... . 3 .. .. .. λ2 = 0. two eigenvalues are negative and one eigenvalue is positive.. ......... ...................... . ........... The wave corresponding to λ2 intersects the cell face at xi+ 12 from the left. 3 λ λ λ xi .. ......... ..... ........ .. .... ... . the flux is evaluated using the flow variables reconstructed to the right face and left face according to ... . 2 . ...10) t.......150 Flux Vector Splitting Methods Table 6... ... 6.. .. .. . . ... ... ....... ... .. .. ...... ... λ1 = λ1 ........ 1 ... .. λ3 = 0.. and λ− 3 = λ3 ..... ...... + + − − The wave speed splitting yields λ+ 1 = 0... .. ........... which 2 is drawn for u = − 32 a.... ...... ... x xi+1 Fig..1........... .. . .... . ... .. ...... .. 2. Thus........ .. the flux is evaluated using the flow variables reconstructed to the right face: ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ 1 1 1 (γ −1) ρr ρr ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ρr λ1r ⎣ ur ⎦ + λ2r ⎣ ur + ar ⎦ + λ3r ⎣ ur − ar ⎦ F= γ 2γ 2γ 1 2 Hr + ur ar Hr − ur ar 2 ur (6........... ... . ... .... .... ..2...... 6......... . ............ .1...... .... ...1... . while the waves corresponding to λ1 and λ3 intersect the cell face at xi+ 21 from the right as illustrated in Fig..... . ..... . Thus. 1 . ... 6.. ...... ...

. . . .. .. . 3 3 . .. . Thus.... ...... .. 6.... ..... 1 2 ... .3........ . 2 .. ............ . .... . . .. 3. ....... . The wave speed splitting yields − λ+ i = λi and λi = 0 for i = 1. ..... . . .. . . . . ... 2. λ3 = 0.. . .. ........... .... 6.... . . λ2 = 0. .. ... λ1 = 0.. . .. The waves corresponding to λ1 and λ2 intersect the cell face at xi+ 12 from the left.. .... .. .... the flux is evaluated using the flow variables reconstructed to the right face and left face according to ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ 1 1 1 (γ −1) ρl ρr ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ F= ρl λ1l ⎣ ul ⎦ + λ2l ⎣ ul + al ⎦ + λ3r ⎣ ur − ar ⎦ γ 2γ 2γ 1 2 Hl + u l al Hr − ur ar 2 ul (6....12) For Case 4.. ... . ..... . . ...2.2 Steger and Warming’s Method ⎡ ⎤ ⎡ 151 ⎤ ⎡ ⎤ 1 1 1 (γ −1) ρl ρr ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ F= ρr λ1r ⎣ ur ⎦ + λ2l ⎣ ul + al ⎦ + λ3r ⎣ ur − ar ⎦ γ 2γ 2γ 1 2 Hl + ul al Hr − ur ar 2 ur (6...... .. . . .. .... . . .... . . .... ... ... ... .. .6.. . ............ . . ...... . ...... . .. .. . . . . . . Case 2 For Case 3.... .. which is drawn for 2 M = 12 ... . . .. .... .. two eigenvalues are positive and one eigenvalue is negative... .... .. ..... which is drawn for M = 32 . .. .. .. .. .. .. .. . . . .. . . . all eigenvalues are positive. ... ....... ..... 6. + + − − The wave speed splitting yields λ+ 1 = λ1 ...... ...11) t. ... . 1 .......... .... .. . Thus... .. while the wave corresponding to λ3 intersects the cell face at xi+ 1 from the right as illustrated in Fig.... .... ... .. ...... .... the flux is evaluated using the flow variables reconstructed to the left face according to . λ2 = λ2 . .. ...... ...... ...... λ λ λ λ λ λ xi xi+ 12 x xi+1 Fig. ..... .......... .... ..... ......... ... − and λ3 = λ3 . All waves intersecting the cell face at xi+ 21 originate from the left as illustrated in Fig.. . . ....... ...........3.........

... . ..... . ...... . . ....3...... ......14) M= al + ar ± The wave speeds λ± i have a discontinuity in slope at λi = 0... . ...... ..... .. 2 2 ..... the value of M is the average Mach number at the interface.... .... ... .... .. .. ... . . ....... ........ . ... .. ....... ............. ........... .. .. ... ... ... .. . ....... . ...... .... ....... . .. ..... . .. which can ... . which may be defined as ul + ur (6. . . ..... . .. .. .. .. . . ...... .. ............... . . 3 ..... . ....... ....13) t... 6. .. .. ........ .. .... .. . . .. .. .. . .. 6.. . . .. .. ... .. ... ....... . .. ... . . ...... . ..... ... . . .. . . ... . . .. .. .. . . . . .... ................ . .. . .. ......... ........ . ....... Case 4 In practice. . ... .. .. .. 1 . λ λ λ λ λ xi xi+ 12 λ x xi+1 Fig.... . ..... ..... .. . .. . . ...... 3 1 .... . ..... . . . ..152 Flux Vector Splitting Methods t........ . . .. . ......4.... ..... .... . 3 . ... ..... .... . .. .. . ......... .. . . 2 .................. ... ..... ..... .. . ......... .. . . . .... . ... . . 1 ........ . ...... ..... . .. .. .. . ...... .. . . ............... .. ... ... ..... .... .... .. ... ... .... .. ...... ... 2 .... ...... ... . ............. ........... 3 .... ... . ......... .... ... ..... ......... λ λ λ λ λ xi λ xi+ 12 x xi+1 Fig... Case 3 ⎡ ⎤ ⎡ ⎤ ⎡ ⎤ 1 1 1 (γ −1) ρl ρl ⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ρl λ1l ⎣ ul ⎦ + λ2l ⎣ ul + al ⎦ + λ3l ⎣ ul − al ⎦ F= γ 2γ 2γ 1 2 Hl + ul al Hl − ul al 2 ul (6....... .. ..... ... .. . .. ....... ......... . ... . ..... . . .. .. 1 .. ... . ..... . ...... . .

Qli− 1 .9) are  Fi+ 12 − Fi− 12 dQi + dt Δx  =0 We rewrite them as dQi = Ri (Qli+ 1 . a ˆ = a) and ε is a small number (e. Fi+ 12 + − l r = Fi+ 1 (Qi+ 1 ) + Fi+ 1 (Qi+ 1 ) 2 2 2 = F + (Qi ) + F − (Qi+1 ) = F + (Qi ) + F − (Qi ) + 2 ∂F − (Qi+1 − Qi ) + O(ΔQ2 ) ∂Q and Fi− 1 2 + − l r = Fi− 1 (Qi− 1 ) + Fi− 1 (Qi− 1 ) 2 + 2 − 2 2 = F (Qi−1 ) + F (Qi ) ∂F + = F + (Qi ) + (Qi−1 − Qi ) + F − (Qi ) + O(ΔQ2 ) ∂Q ..2 Stability We consider the semi-discrete stability of Steger and Warming’s Method.2. λ± i is replaced by (Laney. Qri+ 1 . 6.. Therefore.g.6.15) where a ˆ is a suitable velocity (e. The Euler equations (3.g. ε = 0.2 Steger and Warming’s Method 153 lead to spurious numerical behavior at these locations. Qri− 1 ) 2 2 2 2 dt where  Fi+ 12 − Fi− 12 Ri = −  Δx We assume a simple first-order reconstruction: Qli+ 1 = Qi Qri+ 1 2 = Qi+1 2 Thus. 1998) λ± i =  λi ± 1 2   λ2i + ε2 a ˆ2 (6.1).

.30) for Q(x.31): k= 2πl L ˆ k (t) are complex vectors whose subscript k indiThe Fourier coefficients Q ˆ k (t) indicates cates an ordering with respect to the summation index l. the Fourier coefficients Q Substituting the Fourier series into (6. Q 1 = QM and we assume M is odd with M = 2N + 1.31)) and on t. time tn . i.32). Defining ∂F + ∂Q = T Λ+ T −1 ∂F − = ∂Q = T Λ− T −1 A+ = A− we then have  dQi 1  − A Qi+1 + (A+ − A− )Qi − A+ Qi−1 =− dt Δx (6. Consider Q(x. t) to be a continuous vector function that interpolates Qi . Given the values of Qi at some ˆ k at tn are obtained from (3.16) For simplicity.154 Flux Vector Splitting Methods where we evaluate ∂F ± /∂Q at i and neglect its variation in x.e. t) = ˆ k (t)eιkx Q l=−N +1 √ where ι = −1 and the wavenumber k depends on l according to (3. Q(xi . t) is l=N  Q(x. Q dependence on l (through (3. t) = Qi (t) Then the Fourier series (3.17) .16) yields ˆk dQ ˆk = GQ dt where the amplification matrix G is G=−  1  ιkΔx − e A + A+ − A− − e−ιkΔx A+ Δx (6. we assume that the flow is periodic in x over a length L = (M −1)Δx.

19) Defining ˜ k = T −1 Q ˆk Q gives ˜k dQ ˜k = DQ dt Defining ˜k = Q ⎧ ⎫ ˜ ⎬ ⎪ ⎨ Qk1 ⎪ ˜ Q k2 ⎪ ⎩ ˜ ⎪ Qk ⎭ 3 then the solution is ˜ k (0)eλGm t ˜ km = Q Q m for m = 1.20)  1  + λm − λ− m (1 − cos kΔx)   Δx .6. 2.18) (6. 3 The condition for stability is therefore Real(λGm ) ≤ 0 Now Real(λGm ) = − (6.2 Steger and Warming’s Method 155 Now G = T DT −1 where D is the diagonal matrix  1  ιkΔx − e Λ + Λ+ − Λ− − e−ιkΔx Λ+ Δx The eigenvalues of G are D=− λ Gm = −  1  ιkΔx − − −ιkΔx + e λm + λ+ λm m − λm − e Δx (6.

 .

1 and the domain is 0 < κx < 2π. defined by (3.20) is satisfied. ≥0 ≥0 and hence the stability condition (6. The first-order reconstruction (5. Consistency.2. The semi-discrete . and Convergence We consider the problem described in Section 2.102).8.1) is employed. is evaluated at κao t = 7.92) with = 0. which is prior to the shock formation.3 Accuracy. The norm. 6. The initial condition is defined by (2.

.. ..... .e i −Qi ||κao Δt=5×10−4 . . The convergence is displayed in Fig. . ..... . ... .... 6.......156 Flux Vector Splitting Methods form (3... . . The speed of the disturbance is accurately predicted. . ..... 6.. −1 ◦ −2 ◦ ◦ ◦ ....... The error in amplitude is attributable to the dissipative nature of the first-order reconstruction (5. .... . ... The computed result (using 100 cells) and exact solution are displayed in Fig.... The linear convergence is a direct consequence of the linear reconstruction (5... . .1).. .. ... ... ...... ........ ... .... .. ... 6... .... n log10 ||Qn. .5 is n ||Qn.1).... ... .. . The convergence for the second-order reconstruction (using † The line in Fig......... ..... however.. ........ ◦ ......e i −Qi ||κao Δt =  κao Δt 5 × 10−4  n ||Qn. .. ...20) with κ = 0 and no limiter together with results for the first-order reconstruction... . . . . . .... ..45...... ◦ ◦ ◦ computed linear ◦ ◦ −3 −4 −3 −2 −1 0 log10 κao Δt Fig................ .... ...... . ...5. where results are shown using (4. ..... The solution converges linearly† to the exact solution for sufficiently small Δt................. ......... .. .. . .........9) of the Euler equations is employed and the time integration is performed using a second-order Runge-Kutta method (Chapter 7). 6.. . .... . ...... . . . ........ .. Convergence for Steger-Warming Method using a first-order reconstruction The error in amplitude is significantly reduced by using a second-order reconstruction as indicated in Fig.. .. ...... .... ...... . .... . ..8.. . . ... . ... . . .... .. 6.... . ..... . ..5. .. .. ...... .....e i −Qi || 0 ....6 for κao t = 7 and C = 0..

....3....... .. .. ........... .. . . ..1 Algorithm The flux vector F can be written (Exercise 6.. .. ....... ..... . . ... .4) F= ⎧ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎨ ρaM  ρa2  γM 2 + 1 γ ⎪ ! ⎪ ⎪ 1 ⎪ 3 2 1 ⎪ ⎪ + ρa M M ⎪ ⎩ (γ −1) 2 ⎫ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎬ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎪ ⎭ † The line in Fig.. ............ ...... . . .... . . . ... ..21) . . . ... . .. ....... .e i −Qi ||κao Δt =  κao Δt 5 × 10−4 2 n ||Qn.. . ... 6.. ..00 ◦ ◦ ◦ ◦ ◦ ◦◦ ◦ ◦◦◦◦ ◦◦ ◦◦◦◦ ◦◦ −0. ... ... ....† 6.... ....... .. .. .15 Fig. .... o ... . .... .. . . . .. . .10 κx −0..... ..... . . ... ... . .. . . .. . ... . Computed and exact solutions using a first-order reconstruction second-order accurate integration in time) is shown in Fig.... .. .. ...... ...... .. .. .... ...e i −Qi ||κao Δt=5×10−4 (6. . . .. .. ◦ computed exact ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦◦◦ ◦◦◦◦ ◦ ◦ ◦ ◦ ◦ ◦◦ 0. ..6.. ... ........... ..... ... . .. ..... .. . . . .15 .. . .... . ... . ... 6.7 is n ||Qn.... .. .... . ....3 Van Leer’s Method Van Leer (1982) developed a flux vector split method based on the Mach number...3 Van Leer’s Method 157 u/ao . . . . . ..... . ..... ........... .. ..... . . .. 6.. ......05 ◦◦◦◦ ◦ ◦◦◦ ◦ ◦ ◦ ◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦ 0..... ....05 ◦◦◦◦ ◦◦ ◦◦◦◦ ◦ ◦◦ ◦ ◦ ◦ ◦ ◦ ◦◦4 2 3 5 6 ◦◦1 ◦ ◦ ◦ ◦◦ 0... .. .. κa t = 7 0........10 −0.. ........ ... .... . . ....... . .. ...... ........ . . .. .... .. 6... .. .. .. .. .... ...6. ..... .7 and displays quadratic convergence.. ..

. with ρ and a evaluated using Ql or Qr as appropriate.22) ρu = ρl al M + + ρr ar M − (6. . . . ...158 Flux Vector Splitting Methods n log10 ||Qn.. .. ........ .. ..... . ..... . ..... . ....... .. .. . .. ....... .. . For the mass flux.. .... ......... .. respectively. The term involving the Mach number is split into two parts. . .24) ... . . the term involving the Mach number is simply M and is split according to M = M+ + M− (6...... .......... . .. ◦ . . ... . ... ... Van Leer proposed M + ⎧ ⎪ ⎨ 0 = f1+ ⎪ ⎩ M for M ≤ −1 for − 1 ≤ M ≤ 1 for M ≥ 1 (6... . Convergence for Steger-Warming Method using a second-order reconstruction Each expression in (6....... . . ... . .. .... .. . −2 ........ ...... .. ....21) involves three quantities. ... ... . .......... . . . . .... .... ... the speed of sound a. ...... ... the density ρ. . ....... . ....... namely.... ... .. 6.. .. . and the Mach number M .......... . . ◦ ◦ ◦ ◦ −3 ◦ computed quadratic ◦ −4 −5 ◦ ◦ ◦ −6 −4 −3 −2 −1 0 log10 κao Δt Fig. ..e i −Qi || 0 −1 .. .. .... ...23) The mass flux is taken to be where the subscripts l and r imply that the quantities are evaluated using Ql and Qr . ..... ... ...... . ...... . ... ..... .....7..............

...... .. . .. . ..... implying that all waves are moving to the right......... ... .... . .05 ◦•◦•◦•◦•• ◦ ◦ ◦◦◦◦ ◦ • ◦ • ◦◦◦◦◦◦◦◦◦◦◦•◦•◦•••• •• •• ••••••••••••••••• −0. ... .. .... ... . . o . ..... ..8. ..... .. ........... .. .. . ..00 • ◦ • ◦ ◦ • ◦ • ◦ • ◦ ◦ • • ◦ • ◦◦ ◦••◦•◦• ◦ ◦◦•◦•◦ • ◦◦◦ −0....10 0.. ... ... which may be defined as ul + ur M= (6....... ... .... . .... .. ..28) . Similarly. . .. ..... ... .. ... .. . ........ . ... . . . . ......... .. . .... . ..... .25) where M is the average Mach number at the interface. . . ..... . ... ... Ql is employed to compute ρ and a. ... .... .... ......... ..... Hence..... . ..... ... ... κa t = 7 0..... ..... ... .. .. . They are chosen to satisfy (6. .. . . implying that all waves are moving to the left.. . .... .27) For M < −1 the eigenvalues are negative.. ... ..... .3 Van Leer’s Method 159 u/ao ... ... .... ..... . ... ... ◦ computed 1st order ••••••••••• •••••• • • • • • computed 2nd order • • •◦•◦•◦•◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦ • • • exact • ◦ ◦ •◦ ◦◦ • 0...... .. .and second-order reconstructions and M − ⎧ ⎪ ⎨ M = f1− ⎪ ⎩ 0 for M ≤ −1 for − 1 ≤ M ≤ 1 for M ≥ 1 (6.. .. ..... .........22) and to provide the continuity of M ± and its first derivative with respect to M at M = ±1...... ..26) al + ar This yields ⎧ ⎪ ⎨ ρ r ar M ρu = ρl al f1+ ⎪ ⎩ ρaM l l + ρr ar f1− for M ≤ −1 for − 1 ≤ M ≤ 1 for M ≥ 1 (6. . Computed and exact solutions using first... . ..... ....15 ..5) that f1+ = f1− = 1 4 − 14 (M + 1)2 (M − 1)2 (6.. ..... . .. .... .. .. . . It can be shown (Exercise 6.... It remains to determine the functions f1+ and f1− .. .. 6.. .. .. . It is therefore reasonable to use Qr to compute ρ and a..... . .. ... .05 ◦◦ • •◦•◦•◦•◦◦ • ◦ • ◦ • ◦◦ •◦•◦•◦◦ ◦•◦4 2 3 5 6 ◦•◦•◦1•◦ • ◦ • ◦ κx • ◦ ◦◦ 0.10 −0.. . ... ..6.... ..... . ... . ... . . .. ..... . .. .. .. for M > 1 the eigenvalues are positive.... . . .. ... . . .15 Fig..... ..

... ..... ...... . .5 2.... .. ...... . ..5 M− −1. ..... ............. . ....... 6.......0 M −0.. .. ..... ...... . ...... .. .. ......... ..... ......... .....5 1................ .... ..... ..... ............. .. .... ........... ...... .. . ............... .. .............. ...... .... ... ..... ... ...5 −2........ .. ......... .0 0... ... . .. . . .. .......... ... ..32) ... .. ....... ... .5 1........ ..... ... .... .... ............ ..... . ....0 M+ 0.. ... The mass flux is obtained from (6.. the term involving the Mach number is (γM 2 +1) and is split according to  + (γM 2 + 1) = γM 2 + 1 The momentum flux is taken to be  + ρu2 + p = γ −1 ρl a2l γM 2 + 1  − + γM 2 + 1  (6.. ... .... .5 −1....0 −0......... . .... .9.. ... . .. .. ..... .....9.. .. .. 6.. ..... .....0 1...... ... . ...... .... 1.... .. . .... .............. .. ..23). .. .. .... .. .. . .0 ................ ....0 −1... .......... ...... . ... .......... .. . . ....... M + and M − For the momentum flux... .........5 −2.... . .. .. ......... ..0 0.... .. .. .. .. . ........... ... .. .. ...0 Fig... .160 Flux Vector Splitting Methods The complete expressions for M ± are therefore M + ⎧ ⎪ ⎨ 0 = and M − 1 4 ⎧ ⎪ ⎨ M = (M + 1) ⎪ ⎩ M − 14 ⎪ ⎩ 0 2 (M − 1) for M ≤ −1 for − 1 ≤ M ≤ 1 for M ≥ 1 (6.0 −1.. ... ......... ... .... .. . ... ... ............. ... .. 2...... . . ...... ............ ..... ..........5 0....29) for M ≤ −1 for − 1 ≤ M ≤ 1 for M ≥ 1 2 (6. ... . ... .. .... .... ....... .... .. .. . ... . . ....31) − + γ −1 ρr a2r γM 2 + 1 (6.. .... ..................... ..... .. .. .... ..30) The functions are shown in Fig..... ... ...

31) and to provide the continuity of (γM 2 +1) and its first derivative with respect to M at M = ±1. the term involving the Mach number is M [(γ − 1)−1 + 12 M 2 ] and is split according to M [(γ − 1)−1 + 12 M 2 ] = M [(γ − 1)−1 + 12 M 2 ]+ + M [(γ − 1)−1 + 12 M 2 ]− (6. The momentum flux is obtained from (6.32).6.37) ⎧ 2 ⎪ ⎨ γM + 1 for M ≤ −1 − 14 (M − 1)2 [(γ − 1)M − 2] for − 1 ≤ M ≤ 1 ⎪ ⎩ 0 for M ≥ 1 (6.39) The energy flux is taken to be (ρe + p)u = ρl a3l M [(γ − 1)−1 + 12 M 2 ]+ + ρr a3r M [(γ − 1)−1 + 12 M 2 ]− (6.33) ⎧ 2 ⎪ ⎨ γM + 1 for M ≤ 1 f− for − 1 ≤ M ≤ 1 for M ≥ 1 2 ⎪ ⎩ 0 (6. It can be shown (Exercise 6. 6.35) It remains to determine the functions f2+ and f2− . For the energy flux. They are chosen to satisfy (6.36) The complete expressions for (γM 2 + 1)± are therefore (γM 2 + 1)+ = and (γM 2 + 1)− = ⎧ ⎪ ⎨ 0 ⎪ ⎩ 1 4 (M γM 2 for M ≤ −1 + 1)2 [(γ − 1)M + 2] for − 1 ≤ M ≤ 1 +1 for M ≥ 1 (6.6) that f2+ = f2− = 1 4 (M − 14 (M + 1)2 [(γ − 1)M + 2] − 1)2 [(γ − 1)M − 2] (6.40) .34) ⎧ −1 2 2 ⎪ ⎨ γ ρr ar (γM + 1) for M ≤ −1 γ −1 ρr a2r f2− + γ −1 ρl a2l f2+ for − 1 ≤ M ≤ 1 ⎪ −1 ⎩ γ ρl a2l (γM 2 + 1) for M ≥ 1 (6.38) The functions are shown in Fig.3 Van Leer’s Method Van Leer proposed ⎧ ⎪ ⎨ 0 for M ≤ 1 (γM + 1) = for − 1 ≤ M ≤ 1 ⎪ ⎩ γM 2 + 1 for M ≥ 1 2 + f2+ and (γM 2 + 1)− = This yields ρu2 + p = 161 (6.10.

.. ................... .. ....... . ....... ...... . ...... ... ...... ..... ...0 1. It can be shown (Exercise 6.. ....... . .... ... .. . .. .5 −1... ......... ... ... ...5 2. . ....... .... .... .... .......... ........ . .. 6 5 4 3 (γM + 1) (γM + 1)+ 2 1 −2.... .. ........ .. ........ . .... .... .. . .. .... .... ......... ................. ......... 2 − ...... ..5 0 0............ .... . .. ..... They are chosen to satisfy (6................ ... ...... .42) for M ≤ −1 for − 1 ≤ M ≤ 1 for M ≥ 1 (6........ .... ......0 −0. .......... .......... ...... .. . ....39) and to provide the continuity of M [(γ −1)−1 + 12 M 2 ] and its derivative with respect to M at M = ±1..... . .. ... ........ . ... .. ..........41) ⎧ 1 −1 2 ⎪ ⎨ M [(γ − 1) + 2 M ] for M ≤ −1 f− 3 ⎪ ⎩ 0 ⎧ 1 3 −1 2 ⎪ ⎨ ρr ar M [(γ − 1) + 2 M ] (ρe + p)u = ⎪ ⎩ ρl a3l f3+ + ρr a3r f3− ρl a3l M [(γ − 1)−1 + 12 M 2 ] for − 1 ≤ M ≤ 1 for M ≥ 1 (6...0 0.......10......... . ... .. .... ............... ..... .. ............ ..........162 Flux Vector Splitting Methods .. .. .... ....... ...... .. . . .... ........ . . .... ................. .. .. ..... ....7) that f3+ = f3− = 2 −1 −1 2 1 8 (γ + 1) (γ − 1) (M + 1) [(γ − 1)M + 2] − 18 (γ + 1)−1 (γ − 1)−1 (M − 1)2 [(γ − 1)M − 2]2 The complete expressions for M [(γ − 1)−1 + 12 M 2 ]± are therefore M [(γ − 1)−1 + 12 M 2 ]+ = (6...... ..... .. .... 7 . .... .... ..44) ..... ...... ................ .... ...... 2 . ..... ...... ........ .... . ... .. ... ..... .... ..........43) It remains to determine the functions f3+ and f3− .... ..0 M Fig.. .. ... ... ......... .... ... . ....... .. .. ..0 −1.. ... ...... ...... ... . (γM 2 + 1)+ and (γM 2 + 1)− Van Leer proposed M [(γ − 1)−1 + 12 M 2 ]+ = and M [(γ − 1)−1 + 12 M 2 ]− = This yields ⎧ ⎪ ⎨ 0 for M ≤ −1 for − 1 ≤ M ≤ 1 f3+ ⎪ ⎩ M [(γ − 1)−1 + 1 M 2 ] for M ≥ 1 2 (6. .... .... ......... .... . .5 1............... .... ... 6.......... ... . ... ... .... ..

. .. ... .... 2 − ...... .. . .......... ...... . .... . .....5 M [(γ − 1)−1 + M ] 0 0............ . ..... . .......... .0 1. . ................... . . The energy flux is obtained from (6......0 −1.. ..... ............. .............. . . .... . .... .0 −2 0.. .47) ˆ − T −1 = TΛ (6.. ................. ...45) ⎪ ⎩ for M ≥ 1 M [(γ − 1)−1 + 12 M 2 ] 1 8 (γ and M [(γ − 1)−1 + 1 M 2 ]− = 2 ⎧ −1 + 1 M 2 ] ⎪ M [(γ − 1) ⎨ 2 ⎪ ⎩ − 18 (γ + 1)−1 (γ − 1)−1 (M − 1)2 [(γ − 1)M − 2] 0 2 for M ≤ −1 for − 1 ≤ M ≤ 1 (6..... ..... ..... ...3. ....... ..... ..0 −0.....0 + 12 M 2 ]+ M −4 −6 −8 −10 Fig.... .. .. .. .... ............. 10 . ........... . .... .... .. ....... .2 Stability The stability analysis for Van Leer’s method is identical to that in Section 6....... ........... .40)...... .......... .. ... .. ..... ..... ... 2 . ....... M [(γ − 1)−1 + 21 M 2 ]+ and M [(γ − 1)−1 + 12 M 2 ]− 6....... . 8 6 4 M [(γ − 1) 2 −2.. ... . ....... .. .. .. .. .. ........... ............. . .......... ...... ............ .. ............... . . −1 .. 6. ... .. .2.5 2..... . .. ...... .. .. .... .. ....... ...............5 −1.... ... ... .............. ..11.... . ............. .... ...... ................. .5 1.2 provided that the fluxes F + and F − satisfy ∂F + ∂Q ∂F − ∂Q ˆ + T −1 = TΛ (6.. ..... ...... ........ . ........ ........ ... . ............ .46) for M ≥ 1 The functions are shown in Fig.. ... ... .. ..... . ..... . ........ . .... ......48) .. ........... . .... . .............. . ....... .......6.. .. . ..... ........ .... . .......... . .3 Van Leer’s Method ⎧ ⎪ ⎨ 0 163 for M ≤ −1 + 1)−1 (γ − 1)−1 (M + 1)2 [(γ − 1)M + 2]2 for − 1 ≤ M ≤ 1 (6.11.... ............ ..... 6.. 1 ..

Since M ≥ 1. and Convergence We consider the problem described in Section 2. The semi-discrete form (3.102). 6.3 Accuracy.e i −Qi ||κao Δt =  κao Δt 5 × 10−4  n. the eigenvalues were determined by Van Leer (1982). The norm. Thus. F + = F and the eigenvalues of ∂F + /∂Q are given by (2. is evaluated at κao t = 7. and hence det(∂F + /∂Q) = 0 for −1 < M < 1. For −1 < M < 1.1 and the domain is 0 < κx < 2π. F + = 0 and there is no contribution to the stability analysis. 6.92) with = 0.50) We now consider (6.49) (6. and ˆ of ∂F + /∂Q and λ λ i i ˆ + are nonnegative λ i ˆ − are nonpositive λ i (6. The solution converges linearly† † The line in Fig.1) is employed.164 Flux Vector Splitting Methods ˆ + and Λ ˆ − are diagonal matrices whose elements are the eigenvalues where Λ + ˆ − of ∂F − /∂Q. Consistency. The first-order reconstruction in (5. all eigenvalues are nonnegative. The convergence is displayed in Fig.8. λ + eigenvalue of det(∂F /∂Q) for −1 < M < 1.12 is n ||Qn. 6. There are three possible cases: For M ≥ 1.3. Since the energy flux may be expressed as ρa3 f3+ = + 2 1 3 (f2 ) ρa 2(γ 2 −1) f1+ for − 1 < M < 1 the third row of ∂F + /∂Q is a linear combination of the first and second ˆ 1 = 0 is an rows. defined by (3. The remaining two eigenvalues are solutions of the quadratic equation (Van Leer. 1982) # "   ˆ 2 − 3 a(M + 1) 1 − (γ −1)(M −1) γ(M −1)2 + 2γ(M −1) − 2(γ +3) λ ˆ λ 2 12γ(γ +1) " # (M −1) + 14 a2 (M + 1)3 1 − [4γ(γ −1)(M −1) + (γ +1)(3−γ)] = 0 8γ(γ +1) (6.e ||Qi −Qn i ||κao Δt=5×10−4 .9) of the Euler equations is employed and the time integration is performed using a second-order Runge-Kutta method (Chapter 7). The initial condition is defined by (2. For M ≤ 1.51) Van Leer concluded that both roots of this equation are positive provided 1 < γ < 3. which is prior to the shock formation.12).50).12. respectively.49). A similar analysis proves (6.

..... . ... .e i −Qi ||κao Δt =  κao Δt 5 × 10−4 2 n.. . .. .... 6. . .12. however..15 and displays quadratic convergence.. . ... ........ .... .. ... .......45.. ..1). .... .... ... ..6.... Convergence for Van Leer’s Method using a first-order reconstruction The error in amplitude is significantly reduced by using a second-order reconstruction as indicated in Fig... .....e i −Qi || 0 .. .. . 6. ◦ ◦ ◦ −2 .... .. . .. . ... The computed result (using 100 cells) and exact solution are displayed in Fig... where the results are shown using (4... .... .. . ..... 6........ .... ..... ◦ ... . The speed of the disturbance is accurately predicted... . ... ... .. .........1)... .... .. The error in amplitude is attributable to the dissipative nature of the first-order reconstruction in (5.. ... .... ....... . ....14. . ......... ........ . ... . 6.. The convergence for the second-order reconstruction (using second-order accurate integration in time) is shown in Fig..... .. .3 Van Leer’s Method 165 to the exact solution for sufficiently small Δt.....† † The line in Fig.. .... −1 ..20) with κ = 0 and no limiter together with results for the first-order reconstruction. ..... . ◦ ◦ computed linear ◦ ◦ ◦ −3 ◦ −4 −3 −2 −1 0 log10 κao Δt Fig.. .......e ||Qi −Qn i ||κao Δt=5×10−4 .. .. . ..... .......... .... .... . .. .... . 6. ... The linear convergence is a direct consequence of the linear reconstruction in (5.. ...... ... . .. .. .. . ................ .. . ... n log10 ||Qn.....13 for κao t = 7 and C = 0...15 is n ||Qn. ....... . ... ... . . . ......

. . ..... ... ...... .. .05 ◦ ◦◦◦◦ ◦ ◦◦◦◦ ◦ ◦ ◦4 ◦1◦◦ ◦ 2 3 5 6 ◦ ◦ ◦◦◦ 0. ..... . .......... . ... .. .... ....... . .. .... . ... .... .......10 κx −0.. .. ... .. . ... ...... .. . ...... .... . ... .. . . . ... .. .... solution From (6.... ..... .. .. . ....... . .... ..... .. κa t = 7 0.. ... .. .. . ......15 ... o .. .... .... . ...... ... . .8).10 0. . ..... ....... ... ..... . .. . .. .. .. . ... . . . .. ..... ... .......... . ........ . ...... .... . ... . . . .. .. ..... . . . ........ ... . .. .. ..... ........ . .00 ◦ ◦ ◦ ◦ ◦ ◦◦◦ ◦ ◦◦◦ ◦◦ ◦ −0... ... ... ....... . . ........ .00 • ◦ • ◦ • ◦ • ◦ ◦ • ◦ • ◦ • • ◦ ◦ ◦••◦•◦• •◦•◦ ◦ −0.166 Flux Vector Splitting Methods u/ao ..05 ◦◦◦◦ ◦ ◦ ◦ ◦ ◦◦ ◦◦◦ ◦◦◦◦ ◦◦◦ ◦◦◦◦◦◦◦◦◦◦◦◦◦ −0.... . . ......... . .. ....... .. ..... ....... . Computed and exact solutions using a first-order reconstruction u/ao .. .....14.. .. .... . ..... .....10 0.. ...... ...... . .. ..05 ◦ •◦•◦•◦• ◦ • • ◦ ◦ • ◦ • •◦ ◦•4 •◦•◦•◦ 2 3 5 6 •◦•◦•◦1 ◦ • ◦ ◦ • κx ◦ 0..... .. ..... .... ... ... ...... . .... . ........ . .... ........ ....9).. ... .. . .. .... .. .. . ◦ computed 1st order ••••• •◦•◦•◦•◦•◦•◦•◦◦◦◦◦◦•◦•◦•◦•• • • ◦ ◦ • • ◦ • computed 2nd order • ◦ ◦◦• •◦•◦•◦ • ◦ • ◦ • ◦ ◦• exact ◦•◦•◦•◦ 0.. .... .. .. ..... . κa t = 7 0..... ◦ computed ◦◦◦◦◦◦◦◦◦◦◦◦◦ exact ◦◦◦◦◦◦ ◦ ◦ ◦ ◦ ◦◦ ◦◦ ◦ ◦ ◦ 0. .. . . . ...... .10 −0.... .... .. ...... ... . ... ... ... . .. F ± = T Λ± T −1 Q ... . .. .. ... . ... ... ... . .. .. .. ..... ....... ..... ... . ... . .. ..... ......and second-order reconstructions Exercises 6. .... . .. ...... .. . .... . .. ... .. . . 6.. . ........ ...05 ◦•◦•◦•◦ •◦ • ◦ • ◦ • ◦ ◦• •◦◦◦◦ ◦•◦•◦•◦• ••••◦◦◦◦◦◦◦◦•◦•◦•◦•◦•◦•◦•◦•◦• ••••••••• −0... . . .. ... .. .. ..... ..15 Fig. ... . ....... ..... . . .. .... . ........ .1 Derive the expression (6... ... . ... .... ....15 Fig. . .... ..... .. ..... ........... .15 ..... . . .. .. .. .. . .. Computed and exact solutions using first. . .... .. .. . .... .. o .... . ... . . . ... . . ...13.. ....... .... .. . 6. ..... . ..... .... .....

. .. . . . ....... .............. Convergence for Van Leer’s Method using a second-order reconstruction Using (6. ... (6.. ...... . .. . . . . . ...... 6.. . . Show that (6.12). . ... ... . . ...... .15) provides a continuous slope at λ± i = 0.. . .. . .. . .. ◦ ◦ ◦ ◦ −3 ◦ computed quadratic ◦ −4 −5 −6 −4 ◦ ◦ ◦ −3 −2 −1 0 log10 κao Δt Fig. . dλ+ i dλi dλ− i dλi = d dλi 1 2 (λi + |λi |) = = d dλi 1 2 (λi − |λi |) = % % 1 0 λi > 0 λi < 0 0 1 λi > 0 λi < 0 From (6... . ...3 Derive Table 6. .. ... ... .... .1 using (2. .. .e i −Qi || 0 −1 ..... .. −2 ..... ...... ......9) is obtained... . . . .. .... ... . .... ..Exercises 167 n log10 ||Qn. . ............. .... ... ....... . .. ........ ....... .. the expression for F ± in (6..... . .4). ... .. 6....... .. . ..1)..2).... . ... ◦ . . . ... . . ... solution From (6.. . . . .... ... and (6...15). . T −1 Q= (γ −1)ρ/γ ρ/2γ ρ/2γ $ Using (6. ..... .5). ± Determine the discontinuity in the slope of λ± i at λi = 0...15. . ... ...2 6. .6).. . .. ... ..... dλ± i dλi & = 1 2 ( 1± ' λi λ2i + ε2 a ˆ2 . .... .. ........

45) and for M [(γ − 1)−1 + 12 M 2 ]− in (6. the expressions for M + in (6. dM + dM 6.30) are continuous at M = ±1.7 = 1 2 (M + 1) = 0 at M = −1 = 1 at M = 1 and thus dM + /dM is continuous at M = ±1.39) is obtained by substitution.168 Flux Vector Splitting Methods and thus ) ) ) ) dλ± i dλi 6. are continuous. By inspection. The derivatives of M [(γ − 1)−1 + 12 M 2 ]± can be shown to be continuous at M = ±1 by straightforward differentiation.22) is satisfied.4 6. and have continuous first derivatives with respect to M at M = ±1. 1998): + p =p − p =p ⎧ ⎪ 0 ⎪ ⎨  ⎪ ⎪ ⎩ 1 M +1 2 ⎧ ⎪ 1 ⎪ ⎨  − ⎪ ⎪ ⎩ 0 2 M −1 2 M ≤1 (2 − M ) −1 < M < 1 M ≥1 2 M ≤1 (2 + M ) −1 < M < 1 M ≥1 . By inspection.44) for M [(γ − 1)−1 + 12 M 2 ]± satisfy (6. and have continuous first derivatives with respect to M at M = ±1. 6. f1+ + f1− = = = 1 4 1 4 (M + 1)2 −  1 4 (M − 1)2  M 2 + 2M + 1 − M 2 + 2M − 1 M and therefore (6. and have continuous first derivatives with respect to M at M = ±1.6 6. From (6. the identity (6.8 Prove that the momentum flux splitting in (6. Show that the expressions (6. A similar results holds for M − . the expressions for M [(γ − 1)−1 + 12 M 2 ]+ in (6. Show that the expressions (6. are continuous.5 = λi →0 1 2 Derive the expression (6.29) and for M − in (6.22).44).21). are continuous.28) for M ± satisfy (6. solution From (6.36) for (γM 2 + 1)± satisfy (6.32) implies a pressure splitting (Laney.28).29).46) are continuous at M = ±1. solution From (6.39). Show that the expressions (6.31).

Exercises 6. solution By definition. Find this expression in terms of a.  ∂  ρaM + = a ∂ρ  M + 1 2 2 +ρ  M + 1 2 2 ∂a + ρa ∂ρ  M + 1  ∂M 2 ∂ρ Now ρe = 1 (ρu)2 ρa2 + γ(γ −1) ρ and thus differentiating with respect to ρ gives  ∂a a  = γ(γ −1)M 2 − 1 ∂ρ 2ρ Now ρa2 M 2 = (ρu)2 ρ and thus differentiating with respect to ρ gives M2 ∂M ∂ρa2 + 2ρa2 M =− ∂ρ ∂ρ  Inverting the definition of ρe gives ρa2 = γ(γ −1) ρe − (ρu)2 ρ ρu ρ 2 ! and thus ∂ρa2 = γ(γ −1)u2 ∂ρ Thus. for −1 < M < 1.  ∂M M  =− γ(γ −1)M 2 + 1 ∂ρ 2ρ Substituting into the initial equation gives    a 2 ∂  ρaM + = − M − 1 γ(γ −1)M 2 + 1 ∂ρ 8 6.  ∂ ∂  ρaM + = ρa ∂ρ ∂ρ  M +1 2 2 ! where ∂/∂ρ implies holding ρu and ρe constant. for example.51) are positive for Show that the eigenvalues λ 1 < γ < 3. an expression for ∂(ρaM + )/∂ρ. .10 ˆ + obtained from (6.9 169 Derivation of the eigenvalues of F + for Van Leer’s method requires. M and γ. Thus.

1) Temporal quadrature algorithms can be categorized as explicit or implicit.7 Temporal Quadrature A great many if not all the problems in mathematics may be so formulated that they consist in finding from given data the values of certain unknown quantities subject to certain conditions.1 Introduction We consider algorithms for the temporal quadrature of the semi-discrete Euler equations  Fi+ 12 − Fi− 12 dQi + dt Δx  =0 (7. 170 . For explicit methods. Implicit methods typically are not subject to the Courant-Friedrichs-Lewy condition. explicit methods are restricted by the Courant-Friedrichs-Lewy condition introduced in Section 3. However. This is the simplest approach. † Parallelization is the use of multiple central processing units (cpus) to solve a single problem. albeit with greater effort. A history of parallel computers (through the late 1980s) is presented in Hockney and Jessope (1988). Explicit methods are easily parallelized.† For implicit methods. the values Qi at each timestep (or subiterate within the timestep) may be evaluated independently. Carl Runge (1912) 7.9.2. all values of Qi at each timestep must be solved simultaneously. These methods can also be extended to parallel computation. Recent results in the development of parallel algorithms for computational fluid dynamics are presented in Matsuno (2003).

In general. Probably the most commonly used techniques are the two. 1986.7.2 Explicit Methods 171 We rewrite the semi-discrete Euler equations as dQi = Ri dt with (7. Qri+ 1 ) 2 2 where at i + Qli+ 1 and Qri+ 1 are the left and right 2 2 1 . the residual Ri depends on 2 2 states reconstructed to the face  Ri = Ri Qli+ 1 . 1986). Qri+ 1 . Thus. Jameson et al. The first step computes an intermediate value † Also known as the midpoint method. and Q2i . Qri− 1 2 2 2  2 and hence the semi-discrete form is   dQi = Ri Qli+ 1 . Fi+ 1 = F (Qli+ 1 . Qli− 1 .1 Runge-Kutta The two-stage Runge-Kutta method† is Q0i = Qni Δt 0 R 2 i = Q0i + ΔtRi1 Q1i = Q0i + Q2i Qn+1 = Q2i i (7. Press et al.2)  Fi+ 12 − Fi− 12 Ri = −  (7.2. Additional methods include Bulirsch-Stoer (1970). . Qli− 1 . 1971). 1971.4) 7.3) Δx where Ri is denoted the residual.. 1971.and four-stage Runge-Kutta methods (Abramowitz and Stegun. (1981) and Adams-Bashforth-Moulton (Press et al..2 Explicit Methods There are numerous methods for explicit temporal integration. Gear. 7. Qri+ 1 .5) where we introduce the temporary vectors Q0i . Gear. Q1i . The temporary vector Q0i is identified as Qni . Qri− 1 2 2 2 2 dt (7.

.1). Hence. i. and Q4i ) plus separate storage for the flux quadrature Ri at each step and for the sum of the fluxes in (7. The four-stage Runge-Kutta method is Q0i = Qni Δt 0 R 2 i Δt 1 R Q0i + 2 i Q0i + ΔtRi2 * Δt % 0 Q0i + Ri + 2Ri1 + 2Ri2 + Ri3 6 4 Qi (7. the error in Qn+1 i (Exercise 7. .e. Two time levels of computer storage are required (i.. 7.172 Temporal Quadrature Q1i where Ri0 implies evaluation using Q0i .7) Q2i = (7.6) Q1i = Q0i + (7.e. Q3i .2).8) Q3i = Q4i = = Qn+1 i (7. the minimum storage requirement is approximately 32 M .11) is O(Δt5 ) The algorithm is temporally fourth-order. Two time levels of computer storage are required (i.9) (7. i.e. The algorithm is temporally second-order. Hence. Q0i and an intermediate level that stores Q1i and is overwritten by Q2i ) plus separate storage for Ri . Q2i .3 Implicit Methods There are several methods for implicit temporal integration. The commonly used methods are Beam and Warming (1976) and Briley and McDonald (1973). Q0i .e. the minimum storage requirement is approximately 4 · 3 · M . The vector Q2i is identified final value of Qn+1 i n+1 as Qi .. The second step computes the with Ri evaluated using Q1i .10). the error in Qn+1 is O(Δt3 ) i (Exercise 7.10) (7. and an intermediate level that successively stores Q1i ..

14) n+1 n+1 ) Fi+ 1 − Fi− 1 dQi ))n+1 2 2 = − dt ) Δx  (7.13) Three common temporal integration methods can be defined based on the values of α and β listed in Table 7.1. Implicit Temporal Integration Methods Case Method 1 2 3 Trapezoidal Euler implicit Three point backward α β E 0 0 1 2 O(Δt)3 O(Δt)2 O(Δt)3 1 2 1 1 Into (7. we may expand Qi in time as follows:   α Δt Qni − Qin−1 + (1 + α) (1 + α) + ) Qni +  E = β−α− 1 2  dQi ))n +β dt ) & ) (. We denote 1 is a function of Qi+ 1 and Qi+ 1 evaluated at t 2 2 2 )n+1 ) = lQn+1 Qli+ 1 ) i+ 1 2 (7.1 Beam-Warming Following Beam and Warming (1976).7.3) by comparison with the Taylor series expansion for Qi that the error in this expression is Qn+1 i O(Δt)2 + O(Δt)3 (7.16) 2 n+1 and similarly for Qri+ 1 evaluated at tn+1 .3 Implicit Methods 173 7.3. ) dQi ))n+1 dQi ))n = − dt ) dt ) (7.12) where α and β are constants.17) ) and ∂F ))n Aˆri+ 1 = 2 ∂Qr )i+ 1 2 (7. Table 7. We therefore expand Fi+ 1 as n+1 Fi+ 1 2 ) ∂F ))n n = Fi+ + 1 2 ∂Ql ) Define 2 i+ 12  l Qn+1 i+ 1 2  ∂F ))n ) − l Qni+ 1 + 2 ∂Qr ) i+ 12 ) ∂F ))n Aˆli+ 1 = 2 ∂Ql )i+ 1 2 2  r  r n 2 Qn+1 i+ 1 − Qi+ 1 +O(Δt) 2 2 (7. It may be directly verified (Exercise 7.18) .15) n+1 l r n+1 The flux Fi+ .12) we substitute  n − Fn ) Fi+ 1 dQi ))n i− 12 2 = − ) dt Δx   (7.1.

15) becomes  ) dQi ))n+1 dt ) = − − n − Fn Fi+ 1 i− 1 2  + O(Δt)2 2 Δx 1  ˆl l n+1 l n  ˆl l n+1 l n  A 1 Qi+ 1 − Qi+ 1 − Ai− 1 Qi− 1 − Qi− 1 2 2 2 2 2 Δx . Thus. (7.174 Temporal Quadrature where it is understood that Aˆli± 1 and Aˆri± 1 are evaluated† using the data 2 2 at tn .

i+ 2   terms involving left face at i ± 1 2 1  ˆr r n+1 r n  ˆr r n+1 r n  A 1 Qi+ 1 − Qi+ 1 − Ai− 1 Qi− 1 − Qi− 1 − 2 2 2 2 2 Δx .

i+ 2   terms involving right face at i ± 1 2 (7.24) where the first term arises from (7.20) For any linear reconstruction.22) comprises a blocktridiagonal system of equations for integrating the Euler equations.23) The error in this expression compared to a Taylor series expansion is  E = O((Δx)2 Δt) + β − α − 1 2  O(Δt)2 + O(Δt)3 (7.16).12) becomes   −ν Aˆli− 1 δQni−1 + I + ν Aˆli+ 1 − Aˆri− 1 2 2  2 δQni + ν Aˆri+ 1 δQni+1 =  α Δt δQn−1 − (1 + α) i (1 + α) n Fi+ 1 2 2 n − Fi− 1  2 Δx (7. .21).22) where ν= Δt β (1 + α) Δx (7. † Not to be confused with (4.21) 2 Thus. Equation (7. (7.19) We now introduce the “delta” notation† δQni = Qn+1 − Qni i (7. Aˆli+ 1 2  l  n+1 n Qi+ 1 − l Qi+ 1 − Aˆli− 1 2 2  l n+1 2 2  n Qi− 1 − l Qi− 1 = 2 Aˆli+ 1 δQni − Aˆli− 1 δQni−1 + O((Δx)2 Δt) 2 (7.r are the Jacobians of the flux vector function F and thus depend on the specific † Note that A i± 1 2 choice of the flux algorithm. ˆl.

1 Runge-Kutta We consider the application of the second-order Runge-Kutta algorithm with Roe’s method for the spatial fluxes Q0i = Qni Δt 0 R 2 i = Q0i + ΔtRi1 Q1i = Q0i + Q2i Qn+1 = Q2i i with the simple first-order reconstruction Qli+ 1 = Qi Qri+ 1 = Qi+1 2 2 We consider a von Neumann stability analysis.18).46) as Ri = −  1  − A Qi+1 + (A+ − A− )Qi − A+ Qi−1 Δx where A+ = A− = 1 2 (A 1 2 (A + |A|) = T Λ+ T −1 − |A|) = T Λ− T −1 (7. where A is the Jacobian (2.4 Stability of Selected Methods 175 The Jacobians Aˆl.r i± 12 depend on the specific choice for the flux algorithm.26) .8) 2 2 2 2 ˜ is treated as a constant in (7.4. 7.25) where Ali+ 1 = A(Qli+ 1 ) and Ari+ 1 = A(Qri+ 1 ). here we present the fully discrete stability analyses of selected explicit and implicit temporal integration methods. and the Roe matrix |A| 7.3). The residual Ri is linearized (5.4 Stability of Selected Methods In contrast to the semi-discrete stability analyses presented in Chapters 5 and 6. Aˆli+ 1 2 Aˆri+ 1 2 = 1 2 = 1 2  ˜ i+ 1 Ali+ 1 + |A| 2 2  Ari+ 1 2 ˜ i+ 1 − |A| 2   (7.7. For Roe’s Method (Section 5.

30) is not the same as α in (7. A+ |A| = |A|A+ .30) The solution Qi is expanded as a Fourier series as in Section 5. α2 + + + − − − + − + I − α(A − A ) + (A A + A A − 4A A ) Qni 2   + −α2 (A+ − A− )A+ + αA+ Qni−1 + α2 + + n A A Qi−2 2 (7.2. Q1i = Qni − Qn+1 = Qni − i  Δt  − 1 A Qi+1 + (A+ − A− )Q1i − A+ Q1i−1 Δx (7.12). and A− |A| = |A|A− based on the structure of these matrices.27) (7. 0)} where λm for m = 1.28) Substituting (7.176 Temporal Quadrature and Λ+ = diag {max(λm . 0)} Λ− = diag {min(λm . 3 are the eigenvalues (2.12).29) where‡ α= Δt Δx (7. Substituting into the first Runge-Kutta step.28) yields† = Qn+1 i   α2 − − n A A Qi+2 + α2 (A+ − A− )A− − αA− Qni+1 2+ . substituting into the second Runge-Kutta step. ‡ Note that α in (7. 2.  Δt  − n A Qi+1 + (A+ − A− )Qni − A+ Qni−1 2Δx Similarly. α2 + + + − − − + − + I − α(A − A ) + (A A + A A − 4A A ) 2  +e−ιkΔx −α2 (A+ − A− )A+ + αA+  † Note that A+ A− = A− A+ .3. The Fourier components satisfy ˆ n+1 = GQ ˆn Q k k where the amplification matrix G is G =   α2 ι2kΔx − − e A A + eιkΔx α2 (A+ − A− )A− − αA− 2+ . .27) into (7.

α2 + + + − − − + − (λ λ + λm λm − 4λm λm ) + I − α(λm − λm ) + 2 m m  − + + +e−ιkΔx −α2 (λ+ m − λm )λm + αλm  α2 + + λ λ (7.33) it is evident that λ Gm =   α2 ι2kΔx − − − − − λm λm + eιkΔx α2 (λ+ e m − λm )λm − αλm 2+ . 0) (7.32) Defining λ+ m = max(λm .35) .34) 2 m m This can be compared with the result for the semi-discrete stability analysis in (5.31) G = T DT −1 where D is the diagonal matrix D =   α2 ι2kΔx − − Λ Λ + eιkΔx α2 (Λ+ − Λ− )Λ− − αΛ− e 2+ . Writing b = αλ− = λGm = ωrm + ιωιm (7.7. α2 + + + − − − + − (Λ Λ + Λ Λ − 4Λ Λ ) + I − α(Λ − Λ ) + 2  +e−ιkΔx −α2 (Λ+ − Λ− )Λ+ + αΛ+ +e−ι2kΔx  α2 + + Λ Λ 2 (7.4 Stability of Selected Methods α2 + + A A 2 It is evident that G can be written in the form +e−ι2kΔx 177 (7.53):  1  ιkΔx − − −ιkΔx + λ Gm = − e λm + λ+ − λ − e λ m m m Δx Define Δtλ+ a = αλ+ = Δx +e−ι2kΔx Δtλ− Δx where the subscript m is omitted but implied. 0) λ− m = min(λm .

178

Temporal Quadrature

we have
(a2 + b2 )
cos 2θ + (a − b)[1 − (a − b)] cos θ
2
+1 − (a − b) + 12 (a − b)2 − ab

ωrm =

(b2 − a2 )
sin 2θ − (a + b)[1 − (a − b)] sin θ
2

ωιm =
where

θ = kΔx
There are three possible cases, which are shown in Table 7.2.
Table 7.2. Cases
Case
λm > 0
λm < 0
λm = 0

a

b

>0
0
0

0
<0
0

For λm > 0,
a2
a2
cos 2θ + a(1 − a) cos θ + 1 − a +
2
2
2
a
= − sin 2θ − a(1 − a) sin θ
2

ωrm =
ω ιm

(7.36)

For λm < 0,
ωrm

=

ω ιm

=

b2
b2
cos 2θ − b(1 + b) cos θ + 1 + b +
2
2
2
b
sin 2θ − b(1 + b) sin θ
2

(7.37)

For λm = 0,
ωrm = 1
ω ιm = 0

(7.38)

Letting θ = −θ˜ and b = −˜b, it is evident that the second case is equivalent
to the first.
Stability (3.81) requires
|λGm | ≤ 1

(7.39)

Clearly, the case λm = 0 satisfies this condition. Thus, it is necessary to

7.4 Stability of Selected Methods

179

determine the conditions for which λm > 0 satisfy this condition. Equations (7.36) represent a parametric curve in the complex λG plane with the
parameter θ. This curve can be rewritten as

ωr − (1 − a)
a

!2

ωι
+
a

!2

$2

− [ωr − (1 − a)]

ωr − (1 − a)
a

(1 − a)

2

!2

ωι
+
a

=
!2 $

(7.40)

This is the Lima¸con of Pascal. It is shown in Fig. 7.1 for a = 0.8.
ωι
.
.......
........

1.0

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

0.5

−1.0

−0.5

0.0
0.0

0.5

1.0

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

ωr

−0.5

−1.0

Fig. 7.1. Lima¸con of Pascal for a = 0.8

The condition for stability in (7.39) yields
a≤1

(7.41)

and thus
Δt|λ±
m|
≤1
Δx
+

where λ±
m indicates consideration of both λm and λm . This may be rewritten
as

Δt ≤ ΔtCF L

180

Temporal Quadrature

where
ΔtCF L = min
i

Δx
|λ±
m|

or
C≤1
where C is the Courant number:
C=

Δt
ΔtCF L

7.4.2 Beam-Warming
We consider the application of the Beam-Warming algorithm with Roe’s
method for the spatial fluxes. We assume α = 0 for simplicity (i.e., trape˜ as conzoidal rule or Euler implicit time integration). We treat A and |A|
˜ may be expanded as
stants. The Roe matrix |A|
˜ = |A| + O(ΔQi )
|A|

(7.42)

Thus,
Ali+ 1

=

Ari+ 1

=

2
2

1
2
1
2

(A + |A|) = T Λ+ T
(A − |A|) = T Λ− T

where
Λ+ = diag {max(λm , 0)}
Λ− = diag {min(λm , 0)}
We denote
A+ = T Λ+ T
A− = T Λ− T
Beam-Warming’s algorithm (7.22) becomes   

−νA+ δQni−1 + I + ν A+ − A− δQni + νA− δQni+1 =   

Δt  − n
A Qi+1 + A+ − A− Qni − A+ Qni−1

Δx
The solution Qi is expanded in a Fourier series as in Section 5.3.2:
Q(x, t) =

l=N 

l=−N +1

ˆ k (t)eιkx
Q

(7.43)

7.4 Stability of Selected Methods

181

ˆn = Q
ˆ k (tn ). Substitution into (7.43) yields
where k = 2π/l. We denote Q
k
ˆ n+1 = Gr Q
ˆn
Gl Q
k
k

(7.44)

where     

− 

Gl = −νe−ιkΔxA+ + I + ν A+ − A−
Gr =

μe−ιkΔx A+ + I − μ A+ − A

+ νe+ιkΔx A−
− μe+ιkΔx A−

(7.45)

and
ν = β

Δt
Δx

Δt
Δx
The matrices Dl and Dr can be factored as
μ = (1 − β)

(7.46)

Gl = T Dl T −1
Gr = T Dr T −1

(7.47)

where     

− 

Dl = −νe−ιkΔxΛ+ + I + ν Λ+ − Λ−
Dr =

μe−ιkΔx Λ+ + I − μ Λ+ − Λ

+ νe+ιkΔx Λ−
− μe+ιkΔx Λ−

(7.48)

The matrices Dl and Dr are diagonal and may be written as
%

*

-

.

Dl = diag λDlm
Dr = diag λDrm

(7.49)

where      

λDlm

+

= −νe−ιkΔx λ+
m + 1 + ν λ m − λm

λ D rm

=

+

μe−ιkΔx λ+
m + 1 − μ λ m − λm

+ νe+ιkΔx λ−
m
− μe+ιkΔx λ−
m

(7.50)

Equation (7.44) becomes
˜ n+1 = Dr Q
˜n
Dl Q
k
k

(7.51)

˜ nk = T −1 Q
ˆ nk
Q

(7.52)

where

Defining
˜k =
Q



˜ ⎬

⎨ Qk1 ⎪

˜
Q

k2

⎩ ˜ ⎪
Qk ⎭
3

(7.53)

182

Temporal Quadrature

the solution is then
˜n
˜ n+1 = λDrm Q
Q
km
λDlm km

(7.54)

The necessary condition for stability is therefore
|λDrm | ≤ |λDlm | for

m = 1, 2, 3

(7.55)

Now        


+

= 1 + ν λ+
m − λm (1 − cos θ) + ιν λm + λm sin θ

λDlm


+

= 1 − μ λ+
m − λm (1 − cos θ) − ιμ λm + λm sin θ (7.56)

λDrm

where θ = kΔx. The stability condition (7.55) then becomes 


λ+
m − λm 

2 

(1 − cos θ)2 

Δt
Δx 

2

(2β − 1) 


+ λ+
m − λm (1 − cos θ) 2 


+ λ+
m + λm 

2 

sin2 θ

Δt
Δx 

2

Δt
Δx

(2β − 1) ≥ 0

(7.57)


Now λ+
m −λm ≥ 0 and therefore the second term is nonnegative for all θ (i.e.,
for all kΔx). The first and third terms are nonnegative for β ≥ 12 . Thus,
the necessary condition for stability is satisfied without any constraint on
Δt.

Exercises
7.1

Prove that the two-stage Runge-Kutta method (7.5) is second-order
accurate.
solution
From (7.5),
Q2i = Q0i + ΔtR1i
where R1i , defined in (7.3), is a function of the Q1j in a neighborhood about i based on
the reconstruction method. For example, if the reconstruction (4.15) is used, then R1i
depends on Q0i−2 , Q0i−1 , Q0i , Q0i+1 , and Q0i+2 . We write
R1i = Ri (Q1j )
and thus
R1i = Ri (Q0j +
Expanding in a Taylor series,
R1i = Ri (Q0j ) + 

j

∂Ri
∂Qj

Δt 0
Rj )
2

Δt 0
Rj
2

+ O(Δt)2

21) is O((Δx)2 Δt). specifically.3 Prove that the four-stage Runge-Kutta method (7. Substituting into (7. The error for Cases 2 and 3 may be found in a similar manner.19) is O(Δt)2 .19) is O(Δt)2 .15).4 7. = Qn Qn+1 i + i 7. the error in expansion for Fi+ 1 − Fi− 1 is O(Δx(Δt) ).13) for the truncation error of the expansion (7. Prove that the error in (7.2 7.11) is fourth-order accurate Derive the expression (7. solution Consider Case 1 in Table 7. Thus the error E = O(Δt)3 . Prove that the error in (7. The expansion (7. the error in the expansion for F n+1 is O(Δt)2 . . solution From (7.12) is Δt 2 Qn+1 = Qn i + i Expanding )  ) ) dQi )n dQi )n+1 ) + ) dt dt )n )  )n dQi )n+1 dQi )n d2 Qi )) d3 Qi )) (Δt)2 = + Δt + + O(Δt)3 ) ) 2 ) dt dt dt dt3 ) 2 Substituting into (7.6 2 2 Prove that the error in (7. i+ 1 n+1 Fi+ 1 2 = n Fi+ 1 2 + ˆl 1 A i+ 2  l Qn+1 i+ 1 2 − l Qn i+ 1 2  2 ˆr 1 +A i+ 2  r Qn+1 − r Qn i+ 1 i+ 1 2 2  + O(Δt)2 n+1 n+1 2 Thus. Q2i = Q0i + ΔtRi (Q0j ) + 183  j Now  d2 Qi dRi = = dt2 dt j ∂Ri ∂Qj (Δt)2 0 Rj 2 ∂Ri dQj = ∂Qj dt  j + O(Δt)3 ∂Ri Rj ∂Qj and using dQi = Ri dt then Q2i = Q0i + d2 Qi (Δt)2 dQi Δt + + O(Δt)3 dt dt2 2 Thus. 7.1.17) is O(Δt)2 .12). the error in (7.Exercises Thus. 7.17).12).5 )n ) )n d3 Qi )) (Δt)3 dQi )n d2 Qi )) (Δt)2 + + O(Δt)4 Δt + ) 2 ) dt dt 2 dt3 ) 4 This expression agrees with the Taylor series expansion up to and including O(Δt)2 . the two-stage Runge-Kutta method is second-order accurate.

Prove that the error in the expansion (7. (E7.28). where the Roe matrix S| ∂F = ∂Ql 1 2 ∂F = ∂Qr 1 2 Similarly.7 Temporal Quadrature Derive the expression (7. 7. Thus.1) . ˜ = |A| + O(ΔQi ) |A| 7. solution The expansion is ˜ = |A| + O(ΔQi ) |A| Now ˜ = S| ˜ Λ| ˜ S˜−1 |A| We rewrite the above equation as   ˜ = T + (S˜ − T ) |A|  ˜ − |Λ|) |Λ| + (|Λ| T −1 + (S˜−1 − T −1 )  Thus.42) is O(ΔQi ). 2 solution The flux expression for Roe’s Method in (5.17) and (5. Thus.39) is Fi+ 1 = 2 1 2  2  ˜ Λ| ˜ S˜−1 (Ql − Qr ) Fl + Fr + S| Thus. we have S˜ − T = O(ΔQi ) and similarly for the other differences.10 Derive the stability result for Beam-Warming with α = 12 .184 7.9     ˜ Al + |A| ˜ Ar − |A| Prove that the Lima¸con of Pascal lies within or on the unit circle if (7. ∂F = ∂Ql 1 2  ∂Fl ˜ Λ| ˜ S˜−1 + S| ∂Ql  ˜ Λ| ˜ S˜−1 is treated as constant. ˜ |A| = T |Λ|T −1 ˜ − |Λ|)T −1 + (S˜ − T )|Λ|T −1 +T |Λ|(S˜−1 − T −1 ) + T (|Λ| + terms that are quadratic in the differences Now.25) for the matrices Aˆli+ 1 and Aˆri+ 1 .41) is satisfied. from (2.8 7.

shock waves or contact surfaces). Therefore.8 TVD Methods The theoretical basis of TVD methods is sound for scalar one-dimensional problems only. Figure 8. We consider the problem of Section 2. 8. 8. on a more or less empirical basis. 4. In practical.333. It seems likely that the oscillations in the vicinity of the shock in Fig.1 for a domain 0 < κx < 2π for which a shock wave forms at κao ts = 8. we consider the same algorithm but with the minmod limiter in reconstruction.1 displays the velocity profile at κao t = 14.1 to prevent oscillations in the reconstructed variables (see Fig.6). Eleuterio F.2 displays the velocity profile at 185 . since no limiter was employed in Fig. These oscillations are unphysical (and therefore undesirable) and are attributable to the numerical algorithm. Oscillations have formed in the vicinity of the shock.e.8 with  = 0.1 Introduction It is instructive to examine whether a proper reconstruction method (Chapter 4) with a Godunov (Chapter 5) or flux vector split algorithm (Chapter 6) is capable of computing oscillation-free solutions of the Euler equations in the presence of discontinuities (i.19) with no limiter. We first consider the Godunov algorithm with a third-order (κ = 13 ) reconstruction (4. Toro (1997) 8.. Figure 8.1 are attributable to the reconstruction algorithm. multi-dimensional problems. the accumulated experience of numerous applications has demonstrated that the one-dimensional scalar theory serves well as a guideline for extending the ideas. non-linear.

◦ computed exact 0..... . ◦ ..... ...◦. ◦ .. . ...... and no limiter κao t = 14.. −0. .... .... . .... .. .. .◦ . .... .. ...... .... .◦ . ...... κao t = 14 .◦ . ... ..2.. ....... ◦ . ....... .. ........ ◦ ... .1. . .. κx . ....... . .◦ ...◦.. . .. ◦ .....◦ . ......05 ◦◦◦◦◦◦ ◦ ◦ ◦ ◦ ◦ 1 ◦◦◦◦◦◦◦◦2 3 4 6 ◦ 5 0.. . However.15 Fig. .. ..05 ◦◦◦◦◦◦ ◦◦◦◦◦◦◦ ◦ ◦◦◦◦◦◦◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦◦◦◦ −0...... ..... . .. .. u/ao .. ..◦ ..... ...... ...◦ ....... . ..... . ...10 ◦◦ 0. the elimination of unphysical oscillations in the reconstruction through limiters would be expected to reduce the magnitude of the unphysical oscillations in the computed solution of the Euler equations..... ◦◦◦◦◦◦ ◦◦◦◦◦◦◦◦ ◦◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦◦◦◦◦◦◦ 0.◦ .. .. .◦ ... .. ... ... ... ◦ ..186 TVD Methods u/ao . . . .. . .. . .. .... .. a third-order reconstruction. . .. . ...◦ ...... . ..◦ ... .... .◦.... 0. . Velocity at κao t = 14 using Godunov’s Method.... ..05 ◦.. ...10 κao t = 14 ............. ◦ .◦ ... .. ...◦ .. ◦ .... ....... . .... . . ..◦.. . .. ◦ . .◦ ...15 Fig. .◦ ...... .... ... .15 ... .... the elimination of all unphysical oscillations in the computed solution of the Euler equations requires modification of the fluxes.◦ ....05 . Velocity at κao t = 14 for Godunov with a third-order reconstruction and min-mod limiter It is therefore evident that the elimination of unphysical oscillations in the reconstruction does not guarantee the elimination of unphysical oscillations in the computed solution of the Euler equations. ... ... ◦ computed exact ◦ ◦ . .. ◦ ................ . ...... . . . . . .. . ...◦ . ..◦ ..... . . .......◦ ◦◦ . . . ... ..... . ..... .... ... .. ........ .◦ ........ but on closer inspection it is evident that the oscillations have not been entirely eliminated. ....◦◦2 3 4 5 6 .... .... .◦ ... . .◦ .. ...◦ ... ... . .... .◦ .◦ . .. ......... ... ... .. ......◦ .... ........ .◦ . .. ..◦ . ..◦ ... ....◦ .. .. .. ◦ . .. ......◦ .....◦. ..◦ .. ......... . .... ... .......... . ...◦ ... ....... The magnitudes of the oscillations have been reduced significantly.. . . . . . ..00 ◦◦◦◦◦ ◦ ◦ ◦ ◦ ◦ ◦◦ ◦◦◦◦◦◦ −0.... .... .. ... 1 ..◦ ...◦ . ... ..◦ .◦. .... ..◦ ............ ... ◦ .. . . since the evolution of the solution Qi (t) is directly a consequence of the integral of the fluxes across the control volume ... 0.. ◦ . ◦ .. ..... . . . . ..... ........ .. ..◦ ....◦.... ◦ . . ........ ....◦ . −0. . .. ....◦ −0...◦ . . ... ..... .. .. . ... .◦ ◦◦◦◦ ◦...... ..... . ... ◦ . . ...00 .. ...◦ .. ..... ◦ . ...◦ 0..◦ ... ....◦ . .. .. . ... ............... 8. ........◦.. ..◦ . 8. .10 .... . ..... .... .. .◦ . Nevertheless.... .. . . . . ... ..◦ .... ◦ ..◦ .◦ . ... . ... . .. ◦ ..◦. ... .... . ... . . κx −0..◦........◦ . . .... ...◦ ....15 .◦ . ....◦ ... .....◦ ...◦ .10 0....

... Consider the computed solution in Fig. For the periodic smooth profile shown in Fig.05 ◦◦◦◦◦ ◦ ◦ ◦ ◦ ◦ ◦4 ◦◦◦◦ ◦ 1 2 3 5 6 ... 8....3) where umaxi and umini are the local maxima and minima of u.10 −0...2 Total Variation Harten (1983.10 0. .. It is straightforward to show (Exercise 8... .. ..05 ◦◦◦◦◦ ◦ ◦◦ ◦ ◦ ◦ ◦ ◦◦◦◦◦◦ ◦◦◦◦◦◦◦◦◦◦ ◦◦ −0.. κao t = 7 ◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦ ◦◦◦◦◦◦◦ ◦ ◦ ◦◦ ◦ ◦ ◦ ◦ 0. Thus... 8. .. ◦ ◦ ◦ ◦ ◦ . ..... the appearance of oscillations in the solution increases the number of terms in the above expression and typically increases the value of T V .00 ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦◦ ◦◦◦◦ −0...1) for a periodic function u that T V (u) = 2  (umaxi − umini ) (8.....15 Fig. T V (u) = 2(max ui − min ui ) (8..2 Total Variation according to (3. ..8. .3.9):  dQi + dt Fi+ 12 − Fi− 12 Δx 187  =0 8.1) i The Total Variation of other flow variables can be similarly defined. Computed solution using Van Leer’s Method with a first-order reconstruction ...... .1984) introduced the concept of Total Variation (TV) of a function as both a means of proving the convergence of a discrete system of equations to their exact solution and as a technique for quantifying the extent of oscillations in the discrete solution...15 . 8.. The Total Variation of the velocity ui is defined as T V (u) =  |ui − ui−1 | (8..3.. ...... 0... κx ◦ 0.2) The appearance of oscillations in a solution typically increases the TV.3..... . u/ao .

where B depends only on T V (uo ) and δt satisfies a Courant-Friedrichs-Lewy condition if applicable.6) A related concept is Total Variation Bounded (TVB). † The supremum (denoted sup) of a set is the least upper bound of the set (Ahlfors. however.4) provided that T V (u) is finite. which is attributable to the use of first-order reconstruction. † A more correct nomenclature would be Total Variation Non Increasing (TVNI). and ρe normalized by ρo . . For a smooth function.5.8.1.1). The timestep is defined by (5. ρo ao . and others introduced algorithms for a hyperbolic system of equations that have a nonincreasing value of TV in time. These methods are denoted Total Variation Diminishing (TVD).5) is employed. Boris and Book (1976). Figure 8. the algorithm is TVD.46. implying excessive numerical dissipation. this becomes  T V (u) = ∞ −∞    ∂u     ∂x  dx (8. the Total Variation is significantly below the exact value. 1953). First. and the second-order Runge-Kutta algorithm (7. we consider the first-order reconstruction (5. respectively.4 displays the Total Variation of ρ.8) with C = 0. respectively. A numerical algorithm is TVB if T V (un ) ≤ B for 0≤t≤τ (8. The initial condition is defined by (2. Thus. Periodic boundary conditions are imposed on the left and right boundaries. Considerable diffusion and decay of the profile in the vicinity of the shock is evident. We examine the Total Variation of the solution of the problem described in Section 2. Harten (1984).5) provided ∂u/∂x approaches zero sufficiently fast as |x| approaches infinity. The Total Variation is seen to decrease for all three variables. The velocity profile at t = 14 is shown in Fig.188 TVD Methods The TV concept can be extended to a general function u(x) according to† T V (u) = lim sup δ→0 1 δ  ∞ −∞ |u(x + δ) − u(x)| dx (8. ρu. where ρo and ao are the undisturbed density and speed of sound.7) for all n such that nδt ≤ τ . The flux algorithm is Roe’s method.† A numerical algorithm for a scalar function u is TVD if T V (un+1 ) ≤ T V (un ) (8.92) with  = 0. and ρo a2o . and the domain 0 < κx < 2π is discretized into 100 uniform cells. 8.

...................... ... Total Variation of ρ....s ..... shown in Fig..... ........ ......... . ............... ....... . . 8...... ... .......15 Fig.... ................... ....... . .. ◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦ ◦◦◦◦◦◦◦◦ ◦ ◦◦ ◦ ◦ ◦ ◦ ◦ 1 ◦◦◦◦◦◦◦◦◦2◦ 3 4 ◦◦◦ 5 6 ◦ ◦ ◦◦ 0................ 1............................. ................... . .................. . ρu....... ... ......... .. is essentially constant up to the time of shock formation but then increases in contrast to the exact solution. .. ........ ........... ◦ computed exact 0........................... ................. Velocity at κao t = 14 The Total Variation of ρu is shown in Fig. . .. ............................. ......... . ..... . .75 0... ................ . ........ ....... .. .. 8... ................................. . . .......... TV(ρ) computed ◦ TV(ρu) computed ◦ TV(ρu) exact TV(ρ) exact 1..... ...............................................05 −0... . ...... .... ... ........ The resulting velocity profile at κao t = 14...................... .......... .................. ... t.. . 0.... ......... ...... ........ .. .... 8......2 Total Variation 189 TV ................... shown in .... ......... ... ... .... .... ..... ....8..... .. ............ .............00 ◦ ◦ .....6 for five different flux algorithms using first-order reconstruction.... 0 2 4 6 8 10 12 .... Notwithstanding the quantitative differences.......... ........ ..... ............. .............. ............50 ... ....4... ....................................................................... ..... . .......................10 κx −0.............. .. ..........7......... ............. .........15 .......................00 ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦◦ ◦◦◦ ◦◦◦◦◦◦◦ ◦◦◦◦ −0............... .................25 0............... The flux algorithm is Godunov’s method (i.............. ............ .... ..... .............. ......... ... . .......................... ......................... ........... ... ....................... ............................. ................... ..... .............. 0. Second..........5............... ................ .25 TV(ρe) computed TV(ρe) exact 1........................ .. ...................................... . ........ ........ ................ ............. ......... ..... ..... ....... .... .50 0... ................................ ...... ............ .....10 κao t = 14 ..05 ◦◦◦◦ ◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦ 0................. ............... .. the exact solution of the General Riemann Problem is employed for the flux............ ... ........ ..... ... ................. .. . .................... .............. we consider the third-order upwind-biased reconstruction (4. ................ . it is evident that all methods are TVD and underpredict the exact solution.... ....... .. . .......... ...... .... ...... ........... ........................... .....19) with κ = 13 but no reconstruction limiter..........................e... ............................ 14 κao t Fig........................................ ........00 ............) The behavior of TV(ρu).. .... ........... ..... .... and ρe u/ao ........ ...... .............. ....... .. . ... .. ................................... ............ ..... ................. 8....... ...... .

............... 14 κao t Fig......... ... ............................... .. 8............1 ◦ Exact 0............................ ..... 0.................2 No limiter Min-mod limiter Exact 0................................2 Osher Steger-Warming  Van Leer 0............. ............. ...5 0...........s  ◦ 0...................................2 displays only a slight oscillation in the vicinity of the shock...7.. 8................... .............1........... ..................................................................................................................... Toro (1997) classifies TVD methods into two categories.......................... ◦ Godunov ..... .......... ...... ............. The velocity profile in Fig...................... ................................... .......................3 ◦  ◦ 0........................... which accounts for the increase in TV(ρu)..◦.....................0 0 2 4 6 8 10 12 .................. 0.................. which did not employ a limiter................ Third.......... .. ......... .............................................................. The flux algorithm is Godunov’s Method........... .................... ............................. ............7................................. ...................................4 .....3 ........................ ............... ...... ............................................................ ◦ .................... shown in Fig.........5 ........... ............................................................................ The TV(ρu)...................................................... ..190 TVD Methods TV(ρu) 0................ ....................... .............. TV(ρu) for different flux algorithms TV(ρu) 0...... we consider the third-order upwind-biased reconstruction with the min-mod limiter........... ....................................... .............. 8........  Roe 0..................................................... .............................. ..................6.................................... .................................. 8.............................. ..................................................... ........... ◦............... t................... t.. ............................................................4 ......... ... ............................................... ............................................. .. ............ ................. displays oscillations on both sides of the shock.......0 0 2 4 6 8 10 12 ............... ................................................... .......................... 14 κao t Fig..................................... 8............ TV(ρu) for Godunov with a third-order reconstruction with and without reconstruction limiter Fig.... The first cate- ............ is significantly better than the previous solution............ ...... We conclude that the reconstruction limiter has virtually eliminated the oscillations but that the method is not precisely TVD........ ................1 0.... ...... ................................................................................................ ........... .................... .....s ................................................................. ........... ..

is employed. low order‡) flux solution† Q l algorithm Fi+ 1: 2   ˆ i = Qn − Δt F l 1 − F l 1 Q i i− 2 Δx i+ 2 (8. c In the second step.8) in time and denote the solution at time tn by Qni . 1979).e. which is given by 2 c Fi+ 1 2 a = αi+ 12 Fi+ 1 2 † Referred to as the transported and diffused solution (Zalesak. ‡ Hence the superscript l. we present one example of each of these two approaches. The algorithm is composed of two steps. 8.1 Algorithm Consider the semi-discrete form (3. This approach imposes a limit on the reconstruction to achieve TVD behavior.. we utilize first-order Euler integration of (8. In the first step. 2 ˆ i does not exceed the range of the initial data.8) For simplicity. and Zalesak (1979). an intermediate ˆ i is generated using (typically) a first-order (i.3 Flux Corrected Transport 191 gory is flux limiters. In the following sections. 8.and low-order expressions for the flux.8. This approach is based on a dynamic combination of high.9) l The flux algorithm Fi± 1 is chosen to satisfy the maximum principle. The algorithm is further described in Book et al. i.. The second category is slope limiters. (1975).3. (8.10) . a corrected flux Fi+ 1 .3 Flux Corrected Transport Boris and Book (1973) introduced a flux limiter method that combines a low-order and high-order flux algorithm to achieve a TVD scheme.9) of the Euler equations  dQi + dt Fi+ 12 − Fi− 12  =0 Δx (8.e. This interthe solution Q ˆ i exhibits significant diffusion that leads to an excessive mediate solution Q broadening of discontinuities. Additional examples are presented in Toro (1997). Boris and Book (1976).

ΔQi+1 . Ki+ 1 2 2  (8. These cases may be divided ΔQi−1. F a 1 .14). ΔQi . ΔQ ˆ i+1 . and F a 1 . . . Si+ 1 ΔQi 2 2  (8.. 2 There are 64 (i.13) 2 where     a ˇ ˇ Ki+ 12 = min Fi+ 1  .11) 2 h where Fi+ 1 is a higher order flux (i. ΔQ in Table 8. 26 ) possible cases to evaluate for determining the effect .17) is Flux Corrected Transport (FCT). First.9) to (8. From (8.e.12) ˆ i+2 ).14) 2 with ˆi = Q ˆi − Q ˆ i−1 ΔQ (8. Si+ 1 ΔQi+2 .16)   a Si+ 12 = sign Fi+ 1 (8. generated using a higher order recon2 struction) and αi+ 12 is a coefficient (denoted the hybridization parameter) determined by the requirement that the updated solution given by   ˆi − Δt F c 1 − F c 1 Qn+1 = Q i i− 2 Δx i+ 2 (8. it is evident that Qn+1 depends on the six quantities i ˆ i−1 . . Q does not introduce any new extrema relative to the set (Q The anti-diffusive flux is formed using the intermediate solution. Boris and Book specify the following expression for the corrected flux:  c Fi+ 1 = Si+ 1 max 0. 2 a Fi+ 1. ΔQ ˆ i+1 as indicated ˆ into 16 classes based on the signs of ΔQi−1 .192 TVD Methods a 1 where Fi+ 1 is an anti-diffusive flux given by 2 a h l Fi+ 1 = Fi+ 1 − Fi+ 1 2 2 (8.7) that    c a = sign Fi+ sign Fi+ 1 1 2 2  if c Fi+ 1 = 0 2 (8. ..12) to (8. ΔQ i− 2 i− 2 ˆ i . ΔQ a Fi− 1. ΔQ ˆ i. it may be shown (Exercise 8. ΔQ ˆ i+2 .17) 2 The algorithm in (8. corresponding to the two possible values of the signs of of FCT on Qn+1 i ˆ ˆ ˆ ˆ i+2 .e.18) . We establish three important properties of FCT. ΔQ ˆ i+1 .15) and ˇi = ΔQ and Δx ˆ Δ Qi Δt (8.1. ˆ i−2 .

has the same sign as) the gradient of Q ˆ i+2 on either side of the interface at ˆ i and ΔQ the one-sided differences ΔQ i+ 12 . Fourth. Third. the corrected flux Fi+ 1 exists if and only if the anti-diffusive flux Fi+ 1 2 2 ˆ as defined by is aligned with (i. it may be shown (Exercise 8.19) c a Thus. Second.8. the corrected flux Fi+ 1 (if it is nonzero) has the same character as 2 a the anti-diffusive flux Fi+ 1 . it may be shown (Exercise 8. and thus the corrected flux is set to zero. Assume This may be proven as follows.. If ΔQ is not possible (in the context of FCT) to define a direction of the gradient ˆ at i+ 1 .e. in a more of Q 2 general sense.3 Flux Corrected Transport 193 Table 8. it may be shown that c Fi+ 1 ≥ 0 ˆ i+2 ≥ 0 and ΔQ ˆi ≥ 0 if ΔQ c Fi+ 1 ≤ 0 ˆ i+2 ≤ 0 and ΔQ ˆi ≤ 0 if ΔQ c Fi+ 1 = 0 ˆ i+2 ) = −sign(ΔQ ˆi) if sign(ΔQ 2 2 2 (8.21) ˆ i+2 ≥ 0 and ΔQ ˆ i ≥ 0.9) that c Fi+ if 1 = 0 2 ˆ i+2 ) = −sign(ΔQ ˆ i) sign(ΔQ (8. Classes for Flux Corrected Transport Class ˆ i−1 ΔQ ˆi ΔQ ˆ i+1 ΔQ ˆ i+2 ΔQ 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 + + + + − + + − + − − + − − − − + + + − + + − + − + − − + − − − + + − + + − + + − − + − − + − − + − + + + − − − + + + − − − + − c Thus. Consider ΔQ .8) that 2 c Fi+ 1 = 0 2      ˆ i = sign ΔQ ˆ i+2 = sign F a 1 sign ΔQ i+ iff 2  (8. then it This is a corollary to (8.1.20) ˆ i and ΔQ ˆ i+2 have opposite signs.19).

8.194 TVD Methods c Si+ 21 = 1. For reference. ˆ i+2 and ΔQ ΔQ n+1 c does From (8. Assume Si− 1 = −1. Fi± 1 = 0 for Classes 6. 9. . .e. Ki+ 12 − Si− 21 max 0. Q ˆ i+2 ). . . Qi+2 ) represents the minimum† set from the intermediate . . Q to the set (Q We now examine the remaining classes to ascertain whether new exˆ i−2 . .12) and (8. . we combine (8. . Therefore.. new maximum or minimum) relative ˆ i+2 ) for these classes. The remaining three possible cases for the signs of 2 ˆ i may be considered and the result proven. .13). and 11. Then 2 2 c Ki+ 12 ≤ 0 and Fi+ 1 = 0. . Ki− 12 ] Δx .20). Qi 2 not introduce any new extrema (i. . . ˆ i−2 . Note trema can be introduced at i relative to the set (Q ˆ ˆ that (Qi−2 . solution used to obtain Qn+1 i ˆi − =Q Qn+1 i Δt [Si+ 12 max 0. Then Ki+ 12 ≥ 0 and Fi+ 1 ≥ 0.

 .

.. i−2 i−1 i i+1 i+2 Fig..... .. ... .. .. . .. . ..... ........ . ... . ... . . . ........ .....1. .. ..... .. ... . . ....... ... .. . 8.. ....... ..  Fc (8. the Fi± 1 2 ˆ in the set (Q ˆ i−2 ... ........ Q ˆ i+2 ) is Q ˆ i+2 .. 8. a larger set may be employed. . .... ....... the intermediate solution Q shown in Fig...... c Fi− 1 ≥ 0 2 ... . .......3. Class 1 We first consider the possible creation of a new maximum at i.... ... .. . c Fi+ 1 ≥ 0 2 i+ 12 i− 12 .. 2 .. For Class 1.1 Class 1 ˆ i is monotonically increasing as For Class 1...8. ...... . There are four possible cases depending on the sign of a . .. From a † Depending on the method of reconstruction used to obtain Fi± 1 ...22) Fc i+ 1 2 i− 1 2 8. . . ....... We consider the possible generation of new extrema.. ..... is Q ˆ Q .. . and the smallest value largest value of Q ˆ i−2 ..... . .8........ . . ....

8. 195  ˆ i+2 = −ΔQ ˆ i+1 + Si− 1 max 0.3 Flux Corrected Transport (8.22). Qn+1 −Q i 2 .

Δ1  Δt K 1 Δx i− 2  ˆ i+2 + Si+ 1 max 0. − ΔQ 2 .

If sign(Fi+ 1   2 ˆ i+2 + min Δt F a 1 . If sign(Fi− 1 i− 12 < 0 from (8. Thus. .  Qn+1 i ˆ i−2 = ΔQ ˆ i−1 + Si− 1 max 0. the term Δ2 is negative.) ≤ 0 (8. Δt Ki− 1 −Q 2 2 Δx . then K cases. term Δ1 is nonpositive. . ΔQ ˆ i+1 . b. then which is negative (Fig. 8. ΔQ ˆ i+2 . thereby proving that no new maximum is created. ˆ i−2 . no new maximum ˆ i+2 ). . .   Δt K 1 Δx i+ 2  (8.. . b. .23)  Δ2 We now show that the terms Δ1 and Δ2 are each nonpositive.14) and therefore 2 ˆ i+1 Δ1 = −ΔQ a which is negative (Fig. then K For Δ2 . .22). ΔQ ˆ i−1 Δ1 = −ΔQ Δx i− 2  which is nonpositive since −a + min(a. If sign(Fi− 1 ) = +1. ΔQ ˆi Δ2 = − Δ Q Δx i+ 2  which is negative.24) for any scalar quantities a.8). there are two possible a ) = −1. then 2  ˆ i+1 + min Δt F a 1 . For Δ1 . Thus. . 8.8). From (8. there are two possible cases. Q is created at i relative to the set (Q We next consider the possible creation of a new minimum at i. a ) = −1. If sign(Fi+ 1 i+ 12 < 0 2 from (8. . . Thus.14) and therefore ˆ i+2 Δ2 = −ΔQ a ) = +1.

 Δ1    ˆ i − Si+ 1 max 0. Δt Ki+ 1 + ΔQ 2 2 Δx .

Δ2  (8.25)  .

8. Q minimum is created at i relative to the set (Q 8.9. For Δ1 .3. the largest value of Q ˆ i+2 ). then which is positive (Fig. ΔQ ˆ i−1 Δ 1 = ΔQ Δx i− 2  which is positive.19) and thus Qn+1 = Q i i− 2 a is created. Δ1 is positive. Since sign(ΔQ i− 2 ˆ in the set (Q ˆ i−2 .  Qn+1 i ˆ i+1 = −ΔQ ˆ i+1 + Si− 1 max 0. thereby proving that no new minimum is created. Thus. then K For Δ2 . ˆ i−2 . then K cases. and hence no new extremum F c 1 = 0 from (8. we consider the possible creation of a new 2 maximum. . no new ˆ i+2 ). then 2  ˆ i−1 + min Δt F a 1 . from (8.24). 2 2 ˆ i . ΔQ ˆi F 1 . and For Class 2. . Since sign(ΔQi+2 ) = ˆ i ). −sign(ΔQ i+ 2 ˆ i + Δt F c 1 Qn+1 =Q i Δx i− 2 ˆ i+1 ) = sign(ΔQ ˆ i−1 ) = +1. there are two possible cases.1. . If sign(Fi− 1 ) = +1. . Δ2 is positive.22).21). . 8. If sign(Fi+ 1  ˆ i − min Δ 2 = ΔQ 2 Δt a ˆ i+2 . ΔQ Δx i+ 2  which is positive from (8. Q the smallest value is min(Q a a There are two possible cases depending on sign(Fi− 1 ).8). Thus. If sign(Fi− 1 i− 12 < 0 and therefore 2 ˆ i−1 Δ 1 = ΔQ a which is positive (Fig. If sign(Fi+ 1 i+ 12 < 0 2 and ˆi Δ2 = ΔQ a ) = +1.196 TVD Methods We now show that the terms Δ1 and Δ2 are each nonnegative.8). . thus F c 1 = 0 from (8.21). ˆ i−2 . Q ˆ i+2 ) is Q ˆ i+1 . therefore F c 1 ≥ 0 from (8. a ) = −1. the intermediate solution Q ˆ ˆ ˆ and decreasing from Qi+1 to Qi+2 as shown in Fig. . Therefore. Δt Ki− 1 −Q 2 2 Δx  . Thus. Now.2 Class 2 ˆ i is monotonically increasing to Q ˆ i+1 For Class 2. If sign(Fi− 1 ) = +1. ΔQ ˆ i+1 . 8. there are two possible a ) = −1. If sign(Fi− 1 ) = −1. .

.... . . ........ ˆ i−2 . . . .. .... .... c Fi− 1 2 .. .... . .. . i− 12 .... .. Q ˆ i+2 ). .. Δt Ki− 1 −Q 2 2 .. .... ΔQ ˆ i−1 F 1 .. ˆ in the set (Q ˆ i−2 .. .. . ≥0 .. ........... ..... . ... ...... Q +1.9... .. .. ... ... . ....3 Flux Corrected Transport 197 ˆ Q ....24).......... i−2 i−1 i i+1 i+2 Fig. ... ...... . . . .... ..... . ... . Class 2 and thus  ˆ i+1 = −ΔQ ˆ i+1 + min −Q Qn+1 i Δt a ˆ i+1 . . ........8.... . . The minimum value of Q Now. ...... . .. ..... ... ..  Qn+1 i ˆ i−2 = ΔQ ˆ i + ΔQ ˆ i−1 + Si− 1 max 0... c Fi+ 1 2 =0 i+ 12 . . Q to the set (Q a We consider the possible creation of a new minimum when sign(Fi− 1) = 2 ˆ i+2 ) is min(Q ˆ i−2 .. . .. no new maximum is created relative ˆ i+2 ).. .. 8. ΔQ Δx i− 2  which is nonpositive from (8. Thus..

 Δx Δ1 Now Δ1 > 0 and .

ˆ i−2 . and increasing to Qi+2 as shown in Fig. ΔQ ˆ i+1 . 8. .1. 8.10.   Δ2  ˆ i−1 + min Δt F a 1 . For Class 3.3 Class 3 ˆ i is monotonically increasing to Q ˆi. ΔQ ˆ i−1 Δ 2 = ΔQ Δx i− 2  and thus Δ2 > 0. Now.3. Q ˆ i+2 ) ≥ Qn+1 − Q ˆ i−2 − min(Q Qn+1 i i and thus no new minimum is created. the intermediate solution Q ˆ ˆ ˆ decreasing from Qi to Qi+1 .

Since sign(ΔQ i− 2 ˆ i − Δt F c 1 =Q Δx i+ 2 Qn+1 i ˆ i+2 ) = sign(ΔQ ˆ i ) = +1. ... ΔQ ˆi −Q Δx i+ 2 .. ... If sign(Fi+ 1) = 2 2 +1.... . . ... Now. ...19) and no new extremum is created. . .. . . thus F c 1 = 0. .... the Since sign(ΔQ i+ 2 ˆ i+2 ) is max(Q ˆi.. . ....... . 1 2 ... . ˆ i−2 . . .. Class 3 a ).. Thus.... .. ..... .... . Q largest value of Q ˆ i+1 ).. .. .. .. .  Qn+1 i ˆ i = − min Δt F a 1 .. ...... Q ˆ i+2 ). .. .. ..10. ...... .. ... i−1 i−2 i i+1 i+2 Fig.. For Class 3...... i+ 1 2 c Fi+ 1 ≥ 0 2 ......... ... ... ........ ΔQ ˆ i+2 ........ If sign(F a ) = −1. .. 8.. .... Q smallest value is min(Q ˆ Q . we examine the possible creation of a new extremum. There are two possible cases depending on sign(Fi+ 1 i+ 12 2 c a then Fi+ 1 = 0 from (8. . thus F c 1 ≥ 0.. i− .... .. . .. ...... .. ....198 TVD Methods ˆ i+1 ) = −sign(ΔQ ˆ i−1 )... ... .... . c Fi− 1 = 0 2 ....... .. and the ˆ in the set (Q ˆ i−2 . . .

ΔQ ˆ i+2 . no new maximum is formed. ΔQ ˆi −Q Qn+1 i Δx i+ 2 .   Δ1 where Δ1 > 0. Thus. Also. ˆi.  ˆ i−1 = ΔQ ˆ i − min Δt F a 1 . Q ˆ i+2 ) ≤ Qn+1 − Q ˆi < 0 − max(Q Qn+1 i i Thus.

. Q ˆ i+1 ) = Qn+1 − Q ˆ i−1 + Q ˆ i−1 − min(Q ˆ i−2 . Now. no new minimum is formed.24). ˆ i−2 .   Δ1 and from (8. Q ˆ i+1 ) Qn+1 − min(Q i i Thus. Δ1 ≥ 0.

≥0  .

≥0  .

. ΔQ Δx i− 2 .. the intermediate solution Q ˆ i . . .. If sign(Fi− 1 ) = −1.. .. c Fi− 1 ≥ 0 2 .. Thus. .. .. ˆ i−1 to Q decreasing from Q c ˆ ˆ i+1 ) = ˆ Since sign(ΔQi+2 ) = −sign(ΔQi ). .. ...8.. .. For Class 4...... ....3. .... . ..  ˆ i+1 = −ΔQ ˆ i+1 + min −Q Qn+1 i Δt a ˆ i+1 ... Now...... 8.. .... ... . ΔQ ˆ i−1 F 1 ...... 8.. . Class 4 a a There are two cases depending on sign(Fi− 1 ).. ... Q ˆ i+2 ). and increasing to Q ˆ i+2 as shown in Fig.1..4 Class 4 ˆ i is monotonically increasing to Q ˆ i−1 . .. . .... . i−1 i−2 i i+1 i+2 Fig. Q ˆ i+2 ) is max(Q ˆ i−1 . . . ... For Class 4... i− ... ...... ... .. thus F c 1 ≥ 0.. .. the largest value of Q ˆ ˆ and the smallest value is min(Qi−2 . Since sign(ΔQ 2 ˆ i−1 ) = +1.. ... ...... . sign(ΔQ i− 2 ˆ i + Δt F c 1 Qn+1 =Q i Δx i− 2 ˆ in the set (Q ˆ i−2 ....... thus Fi+ 1 = 0.11.. . .. .. . .. ... .. we examine 2 2 the possible creation of a new extremum. .3 Flux Corrected Transport 199 8....... .. c Fi+ 1 = 0 2 i+ 1 2 1 2 .. If sign(Fi− 1 ) = +1... .. ˆ Q .. . .. ....11......... ...... then 2 2 c a Fi− 1 = 0 and no new extremum is created. ...... . Qi )..... ..

Now.24). ˆ i−1 . Δ1 ≤ 0. Q ˆ i+2 ) = Qn+1 − Q ˆ i+1 + Q ˆ i+1 − max(Q ˆ i−1 . Q ˆ i+2 ) − max(Q Qn+1 i i .   Δ1 where from (8.

≤0  .

≤0 and thus no new maximum is created. Furthermore. ΔQ ˆ i+1 . ΔQ ˆ i−1 −Q Δx i− 2 .  Qn+1 i ˆ i = min Δt F a 1 .

Δ2    .

Now.200 TVD Methods where Δ2 ≥ 0. Q ˆ i ) = Qn+1 − Q ˆi + Q ˆ i − min(Q ˆ i−2 . ˆ i−2 . Q ˆi) − min(Q Qn+1 i i .

≥0  .

. .. . Thus....5 Class 5 ˆ i is monotonically decreasing to For Class 5... .. i−2 i−1 i i+1 i+2 Fig....... ≥0 1 2 ... . . . ................ Qi+2 ) is max(Q ˆ i−1 . ........ . i+ 12 .... .. .. .... 2 2 c a then Fi+ 1 = 0 and no new extremum is created.. −sign(ΔQ i− 2 ˆ i − Δt F c 1 Qn+1 =Q i Δx i+ 2 ˆ i+2 ) = sign(ΔQ ˆ i ) = +1.12. . 8.... c . smallest value is Q ˆ Q . . ... . F i+ 1 2 ..... we 2 2 examine the possibility of a new extremum.. .. .. .. ..... . . the intermediate solution Q ˆ ˆ i+1 ) = ˆ Qi−1 and increasing to Qi+2 as shown in Fig. and the ˆ ˆ largest value of Q in the set (Qi−2 . thus F 1 = 0..  Qn+1 i ˆ i = − min Δt F a 1 ... If sign(Fi+ 1 ) = +1..12... If sign(Fi+ 1 ) = −1. .. . . 8...... Class 5 a a There are two possible cases depending on sign(Fi+ 1 ).. . 8... .. Now... ....... ... ΔQ ˆi −Q Δx i+ 2 . . thus F c 1 ≥ 0... . . i− . the Since sign(ΔQ i+ 2 ˆ ˆ i−2 .. . ... . For Class 5. ..... ...... ...3..... ....... .... c Fi− 1 = 0 2 . ... Since sign(ΔQ c ˆ i−1 ). Q ˆ i+2 ).  ≥0 and thus no new minimum is formed. ... . . . ...1.... ΔQ ˆ i+2 .

Δ1   .

ˆ i−2 . Now. Q ˆ i+2 ) = Qn+1 − Q ˆi + Q ˆ i − max(Q ˆ i−2 .3 Flux Corrected Transport 201 where Δ1 ≥ 0. Q ˆ i+2 ) − max(Q Qn+1 i i .8.

 ≤0 .

ΔQ Δx i+ 2 . Furthermore. Thus no new maximum is formed. ΔQ ˆi F 1 .  ˆ i−1 = ΔQ ˆ i − min Qn+1 −Q i Δt a ˆ i+2 .

.. i+ 1 2 1 2 . . ....... thus Fi− 1 ≥ 0. . .3..... .. .. the intermediate solution Q c ˆ ˆ ˆ Since sign(ΔQi+1 ) = sign(ΔQi−1 ) = +1........ . . thus 2 ˆ i .13.. .14. . . 8. .. . Ki+ 12 − Si− 21 max 0. .. For Class 7. . sign(ΔQ i+ 2 From (8....... Ki− 12 Δx . ..... ... . Since sign(ΔQ ˆ i+1 ) = ˆ i and decreasing to Q Q c ˆ ˆ ˆ −sign(ΔQi−1 ). ... Thus. .... 8.. i−2 i−1 i i+1 i+2 Fig. F c 1 = 0... .13. .. . ..... . thus F c 1 ≤ 0. ...... thus Fi− 1 = 0.6 Class 6 ˆ i is monotonically increasing to For Class 6...3.. ...... Thus no new minimum is formed. .22)... . Δ2 ≥ 0. c Fi+ 1 = 0 2 .. ..1. c Fi− 1 = 0 2 .. ...... . Since sign(ΔQi+2 ) = 2 ˆ i ) = −1.. ˆi − =Q Qn+1 i  Δt  Si+ 21 max 0....... 8... . ......  ≤0   Δ2 where from (8. .... i− ... Qn+1 = Q i i+ 2 ˆ Q ..........7 Class 7 ˆ i is oscillatory as shown in Fig. ..... . Since sign(ΔQi+2 ) = −sign(ΔQi ). Class 6 8.24)... 8...... the intermediate solution Q ˆ i+2 as shown in Fig.. .. and no new maximum or minimum in formed. . ... .. ..1..

... . i+ 1 2 1 2 .. . .... Q which is a new maximum relative to the set (Q mum cannot be formed... c Fi− 1 2 ≥0 ... ....... Assume ˆ i = −1... ˆ i−1 = 5... .. ... . . ΔQ ˆ i+2 = −2... . .. ΔQ ˆ i+1 = 3..... . ..... ...... .. ...... . F 1  > 1 2 Δx Δx i− 2 Then ˆi + 4 =Q Qn+1 1 ˆ i+2 ).. We show by example that a new maximum can be formed..... .. i− ... i−2 i−1 i i+1 i+2 Fig.. .e..... . ..... . . . A new miniˆ i−2 ..... c Fi+ 1 ≤ 0 2 ... . 8.... ... . however... .... ... ........ .... Si− 21 = +1 and Si+ 12 = −1. . since Qn+1 i   . .. ...... ..14.. .. .... Class 7 a Assume that the anti-diffusive fluxes Fi± 1 are aligned with the local gra2 dients. . . .. . .. ..... . ... i.... ΔQ ΔQ Δt a Δt  a  Fi− 1 > 3..202 TVD Methods ˆ Q ... .... .... . .... . .... .. ..

  ˆ i = −Si+ 1 max 0. Δt Ki+ 1 + Si− 1 max 0. Δt Ki− 1 −Q 2 2 2 2 Δx Δx Δ1 .

the intermediate solution Q ˆ i+1 .8 Class 8 ˆ i is monotonically decreasing to For Class 8. Similarly. increasing to Q Q c ˆ ˆ i+2 ) = ˆ sign(ΔQi+1 ) = −sign(ΔQi−1 ).3. and decreasing to Q ˆ i+2 as shown in Fig. 8. thus F c 1 = 0.1.15. since sign(ΔQ i− 2 ˆ i and no new maximum or ˆ i ). Since ˆ i−1 . Qn+1 = Q −sign(ΔQ i+ 2 minimum is formed. 8. Thus. thus F 1 = 0. .   Δ2 and both Δ1 and Δ2 are nonnegative.

. i−1 i−2 i i+1 i+2 Fig. . .. .. ...... ... . . ... .. .. 8. .. ............ i− .... c Fi+ 1 = 0 2 .... 8. decreasing to Q Q ˆ i+1 ) = This is equivalent to the reflection of Class 8 about i.... ˆ i−1 ).. ˆ Q . .... .. . ..... thus F c 1 = 0.. .1. . . Since sign(ΔQ ˆ i+2 ) = −sign(ΔQ ˆ i ). ..8.9 Class 9 ˆ i is monotonically increasing to For Class 9.. ........... . ..15. . .3...... .. ...... . .. .... ... and increasing to Q ˆ i+2 as shown in Fig.... .. ... the intermediate solution Q ˆ i+1 .. . ... .. 8..... ..... . .. 1 2 .... Class 9 i+2 .. . .. .. ... c Fi− 1 = 0 2 .. i−2 i−1 i i+1 Fig.. ...... .... .......... .. since sign(ΔQ −sign(ΔQ i− 2 ˆ i and no new maximum or minimum is thus F c 1 = 0..... . .... Thus.. c Fi− 1 = 0 2 ... ........ . ......... ... .. .......... .... . c Fi+ 1 = 0 i+ 1 2 2 . . . .. ˆ i−1 ........ ...... .. Class 8 8. . .. ... . ..... .. . .. .. ...... ...16..... ... .. . i− . .. .. .16. . ..... .... ....... ..... Qn+1 = Q i+ 2 formed. .... i+ 1 2 1 2 ....3 Flux Corrected Transport 203 ˆ Q . Similarly.... ..

...e. 2 ˆ Q ...... .10 Class 10 ˆ i is oscillatory. .. ... since Qn+1 i   ......... .. ΔQ Δt  a  Δt a F 1 >1 Fi− 1  > 2.... ..... thus Fi+ 1 ≥ 0.. ... however... . ... .. . Si− 21 = −1 and Si+ 12 = +1.... .......... Since sign(ΔQ ˆ i+1 ) = For Class 10... .. . . ΔQ ˆ i = 1.... ...... 2 .... . . ......... Class 10 From (8.. . i−2 i−1 i i+1 i+2 Fig. Assume ˆ i−1 = −2........ .. .. .. thus F c 1 ≤ 0. We show by example that a new minimum can be formed... ..3...1. 2 ..... 1 2 ...17.. ... ...... . . . c Fi− 1 ≤ 0 . i+ 1 2 c Fi+ 1 ≥ 0 ... ..... . the intermediate solution Q ˆ i+2 ) = sign(ΔQ ˆ i) = ˆ i−1 ) = −1. ...... .. Q ˆ i+2 ).... 8....... . . . . Similarly.   ˆ i − Δt Si+ 1 max 0... ..... ΔQ ˆ i+1 = −2.. . . . since sign(ΔQ sign(ΔQ i− 2 c +1... i. i− .. . . A new maxiwhich is a new minimum relative to the set (Q mum cannot be formed. .. Ki+ 1 − Si− 1 max 0. .. ....204 TVD Methods 8. 2 Δx Δx i− 2 Then ˆi − 3 =Q Qn+1 i ˆ i−2 . . Ki− 1 Qn+1 = Q i 2 2 2 2 Δx a Assume that the anti-diffusive fluxes Fi± 1 are aligned with the local gra2 dients..22). . ΔQ ˆ i+2 = 2.. ..

Δt Ki+ 1 + Si− 1 max 0.   ˆ i = −Si+ 1 max 0. . Δt Ki− 1 −Q 2 2 2 2 Δx Δx Δ1 and both Δ1 and Δ2 are nonpositive.

Δ2   .

.. .. . −ΔQ ˆ i+2 ... ..... the intermediate solution Q ˆ ˆ ˆ i−1 )..... Since sign(ΔQi+1 ) = −sign(ΔQ c Fi− 1 = 0. .1. .3. . . −ΔQ ˆi −Q i+ .. 8. since sign(ΔQi+2 ) = −sign(ΔQ 2 ˆ i and no new maximum or minimum is thus F c 1 = 0..... ... .... ˆ Q . ....3..... ... 2 2 c a then Fi+ 1 = 0 and no new extremum is created.....18. Similarly.. i− . . .....19.. the intermediate solution Q ˆ i+2 as shown in Fig.....3 Flux Corrected Transport 205 8... i+ 1 2 .8... . 8. ..12 Class 12 ˆ i−1 and decreasˆ i is increasing to Q For Class 12.... .. . .. Class 11 8. If sign(Fi+ 1 ) = +1.. .11 Class 11 ˆ i is monotonically decreasing to For Class 11..... Thus.. . .... c Fi− 1 = 0 2 .. ... .... c Fi+ 1 = 0 2 1 2 . Now.. Since sign(ΔQ i+ 2 a a There are two possible cases depending on sign(Fi+ 1 ).. .. thus F c 1 ≤ 0. ..1.. i−2 i−1 i i+1 i+2 Fig. we 2 2 examine the possibility of a new extremum.. Since sign(ΔQ ˆ i+1 ) = ˆ i and increasing to Q Q c ˆ ˆ i ). If sign(Fi+ 1 ) = −1. . 2 ˆi − Qn+1 =Q i Δt c F 1 Δx i+ 2 ˆ i+2 ) = sign(ΔQ ˆ i ) = −1... ......... .  Qn+1 i   ˆ i−1 = ΔQ ˆ i + min Δt F a 1  ..18. ˆ −sign(ΔQi−1 )..... .. 8.. .. .. . thus ing to Qi+2 as shown in Fig. ... Qn+1 = Q i+ 2 formed.. ... .. . ... ........ thus Fi− 1 = 0... . . . Thus... . .. .

 2 Δx Δ1 .

Δ2   .

. .... .. .. . . . ... i−1 i−2 i i+1 i+2 Fig... .. . no new maximum is formed.. ...19.......... i+ 1 2 c Fi+ 1 ≤ 0 2 . . ... −ΔQ ˆi =Q Δx i+ 2 ....... .. c Fi− 1 = 0 2 ...... ... ...... . . i− . .. .. .... . ....... −ΔQ ˆ i+2 ... .. .. .. .... .. ... Furthermore.... .... . . ....  Qn+1 i ˆ i + min Δt F a 1 . Class 12 Since Δ2 ≤ −Δ1 ... . ....... .... .... .. .. ... . ... ... ... ...206 TVD Methods ˆ Q ......... ... 1 2 ... ...... 8....

the intermediate solution Q ˆ ˆ i+2 ) = ˆ to Qi .20.   Δ1 Since Δ1 ≥ 0. thus F c 1 ≤ 0. increasing ˆ i is decreasing to Q For Class 13. thus F c 1 = 0. no new minimum is formed. Since sign(ΔQ i− 2 a a There are two possible cases depending on sign(Fi− 1 ). and decreasing to Qi+2 as shown in Fig. −ΔQ ˆ i−1 F 1  . −ΔQ Δx i+ 2 .1.3. If sign(Fi− 1 ) = −1. −sign(ΔQ i+ 2 ˆ i + Δt F c 1 Qn+1 =Q i Δx i− 2 ˆ i+1 ) = sign(ΔQ ˆ i−1 ) = −1. 2 2 c a then Fi− 1 = 0 and no new extremum is created. Since sign(ΔQ ˆ i ). If sign(Fi− 1 ) = +1. we 2 2 examine the possibility of a new extremum. 8.13 Class 13 ˆ i−1 . Thus. Now.  ˆ i = − min Qn+1 −Q i Δt  a  ˆ i+1 . 8.

 Qn+1 i   ˆ i+1 = −ΔQ ˆ i+1 − min Δt F a 1  .   Δ1 Since Δ1 ≥ 0. −ΔQ ˆ i+1 . no new maximum is formed. Furthermore. −ΔQ ˆ i−1 −Q i− 2 .

 Δx Δ1 .

Δ2   .

..... 8........3 Flux Corrected Transport 207 ˆ Q .. .. i−1 i−2 i i+1 i+2 Fig. .... Class 14 a a There are two possible cases depending on sign(Fi+ 1 ). .. Thus..... ..... . . i+ 1 2 . .... .... . ......... ............. ...... .. .. ........ ...... ... ...... no new minimum is formed. .........21.. . ... .. .... . . . .. ...3....... 1 2 .. .... .. 1 2 ..... c Fi− 1 = 0 2 .. 2 2 c a then Fi+ 1 = 0 and no new extremum is created.. . ... . . .... .... 8.. . increasing to For Class 14... If sign(Fi+ 1 ) = −1...... .. . ..20... . .. . . ... .. .. ...8. . ... 2 ..... . .. c Fi+ 1 = 0 2 ... .. . c Fi− 1 ≤ 0 ... ..21....1... .14 Class 14 ˆ i is decreasing to Q ˆ i .. ... . ...... . the intermediate solution Q ˆ i+2 as shown in Fig... ... . ......... i− .. ... 8... ...... 2 ˆi − =Q Qn+1 i Δt c F 1 Δx i+ 2 ˆ i+2 ) = sign(ΔQ ˆ i ) = −1..... .... .... Since sign(ΔQ ˆ i+1 ) = ˆ i+1 ....... i+ 1 2 .... 8... .. ... ... If sign(Fi+ 1 ) = +1. .. ....... . .. .. thus Fi− 1 = 0. .... ......... .. . c Fi+ 1 ≤ 0 2 . i− . . .. Since sign(ΔQ i+ 2 ˆ Q ... . thus F c 1 ≤ 0. and decreasing to Q Q c ˆ −sign(ΔQi−1 ). .. .. .. Class 13 Since Δ2 ≤ Δ1 . i−2 i−1 i i+1 i+2 Fig..... we 2 2 ..... .

 ˆ i = min −Q Qn+1 i Δt  a  ˆ i+2 . −ΔQ ˆi F 1  . Now.208 TVD Methods examine the possibility of a new extremum. −ΔQ Δx i+ 2 .

−ΔQ ˆ i+2 .   Δ1 Since Δ1 ≥ 0. −ΔQ ˆi −Q i+ 2 .  Qn+1 i   ˆ i−1 = ΔQ ˆ i + min Δt F a 1  . Furthermore. no new minimum is formed.

 Δx .

.... . . .. .. Class 15 a a There are two possible cases depending on sign(Fi− 1 ).... . .. .. .22. .. −ΔQ ˆ i−1 −Q i− 2 Δx . .... c Fi− 1 ≤ 0 . ..... .. 2 2 c a then Fi− 1 = 0 and no new extremum is created. . . . If sign(Fi− 1 ) = −1. If sign(Fi− 1 ) = +1. .... ........ .22..... .. 8..1.. .... 8.. 8... Since sign(ΔQi+1 ) = −sign(ΔQ c Fi+ 1 = 0. Δ1   Δ2 Since Δ2 ≤ Δ1 . the intermediate solution Q ˆ ˆ i ). i− .. Since sign(ΔQ i− 2 ˆ Q ........ thus F c 1 ≤ 0. . .  Qn+1 i   ˆ i = − min Δt F a 1  . ... ... ...... . c Fi+ 1 = 0 2 . .. . i−2 i−1 i i+1 i+2 Fig.. .3........ we 2 2 examine the possibility of a new extremum.. ...... .. no new maximum is formed......... .... Now. .... i+ 1 2 1 2 ... ..........15 Class 15 ˆ i is decreasing to Q ˆ i+1 and increasFor Class 15.. ..... ... 2 .. thus ˆ ing to Qi+2 as shown in Fig.... .. ...... .. Thus... 2 ˆ i − Δt F c 1 Qn+1 =Q i Δx i− 2 ˆ i+1 ) = sign(ΔQ ˆ i−1 ) = −1.. −ΔQ ˆ i+1 . ...... . ..

Δ1   .

−ΔQ ˆ i−1 −Q . no new maximum is formed.  Qn+1 i   ˆ i+1 = −ΔQ ˆ i+1 − min Δt F a 1  .8. −ΔQ ˆ i+1 . Furthermore.3 Flux Corrected Transport 209 Since Δ1 ≥ 0.

 Δx i− 2 .

.. no new minimum is formed. . ......... 8........ . . Δ1   Δ2 Since Δ2 ≤ Δ1 ... . and Convergence We consider the problem described in Section 2.... c Fi− 1 ≥ 0 2 . .92) with  = 0..... c Fi+ 1 ≥ 0 2 i+ 12 . . .....1 and the domain is 0 < κx < 2π. .. ..2 Accuracy.... . . ..... is evaluated at κao t = 7.. 8....19) with κ = 13 .. .23. Class 16 8.. . .. ˆ i−2 ... and the high-order flux is computed using a third-order 2 upwind biased reconstruction (4... . ... defined by (3..26) .. .. . .. .. . ... The flux algorithm is Godunov’s Method.. .. . ....23. No new extrema are created relative Fi± 1 2 ˆ i+2 ).. .. ... . . .. i−2 i−1 i i+1 i+2 Fig..102).. ..1. .. This case is similar to Class 1. .. ... The initial condition is defined by (2.... . .....10)... which is prior to the shock formation. 8.. Q to the set (Q ˆ Q ...8.. . . .. . i− 12 .... The proof is left as an exercise (Exercise 8.... ..... Consistency...... ........... The norm.. The lowl l order flux Fi± 1 is computed using a first-order reconstruction (Qi+ 1 = Qi 2 2 and Qri− 1 = Qi )..16 Class 16 ˆ i is monotonically decreasing as For Class 16....3. There are four possible cases depending on the sign of a .. .. The timestep Δt is determined by Δt = CΔtCF L (8.... .... .. .... the intermediate solution Q shown in Fig.. .. ...3.. . .

8... The linear convergence is a consequence of the first-order temporal integration. .. ..24.... .. ........ .. ........ . .. The error in amplitude is associated with the first-order time integration and is significantly reduced by decreasing the timestep (Fig.. 8.. . ..... . The timestep Δt is held in fixed ratio to the grid spacing Δx............ ... ... ... ◦ ...... ... .. .....e i −Qi ||κao Δt=5×10−4 . ..... .........24.. ....... .. .. ..26).. .... ....... .8) using the initial condition. ...210 TVD Methods where the Courant number C = 0.... ..... ......... . . ◦ ◦ computed linear ◦ ◦ ◦ −3 ◦ −4 −3 −2 −1 0 log10 κao Δt Fig... −1 .46 and ΔtCF L is calculated according to (5. .. ... . ........ .. ...... ..25 for κao t = 7 and C = 0.. .. ... ... .. .. .. ..... i The convergence is displayed in Fig...... ... .24 is n ||Qn.......... . .. ....... ...85) to (2. ... ... n log10 ||Qn...... .... .. ....... and therefore Δt and Δx are decreased by the same ratio.. The exact solution Qn.. 8...... ◦ ◦ −2 ◦ .... .. ... .... . ........... .. ........ .. ... .88).. The solution converges linearly† to the exact solution for sufficiently small Δt. 8. . ......e is obtained from (2.. . . ... 8......... The computed result (using 100 cells) and exact solution are displayed in Fig.... .. .e i −Qi || 0 . Convergence for Flux Corrected Transport † The line in Fig... .. ..... ......e i −Qi ||κao Δt =  κao Δt 5 × 10−4  n ||Qn.45.. .... .. ....

.......... . .. .. ... . ........ .. .. . . .. . ...3... .. κa t = 7 o .. .. .... . .... .......10 κx −0. . ...... .... . ........ . .. . .. . . ... ... ..... ....... .....15 Fig. ... ... ........ .. .. ... .10 0...25.. . Computed and exact solutions (κao Δt = 0... ... .. . .... . ... .... . ...... . .... . ... . ... . .. . .....10 κx −0.... ...... .. ........ . .. .. ...... ...... ...... ....... .. .8. .... . ... ...........1 and the domain 0 < κx < 2π is discretized into 100 uniform cells where κ = 2πL−1 .. . . .... ........... ...... .. .... . ... ... .... ..... . ... .. .. 8..... . ..05 ◦◦◦ ◦ ◦ ◦◦◦◦ ◦ ◦ ◦ ◦ ◦1◦◦ ◦ 2 3 5 6 ◦ ◦4 ◦◦◦ 0.. ........ .. .......... ........ .. . . .. .. ...... . . . .. .. ... . ..... .. .3 Total Variation We examine the Total Variation of the solution of the problem described in Section 2... The timestep is defined by (5.. .. ...... . .... ............... . . .... .. .. . . . . The Total Variation of ρu is shown in Fig.05 ◦◦ ◦◦◦ ◦ ◦◦◦ ◦ ◦ 2 3 4 5 6 ◦◦1 ◦ ◦ ◦ 0. .. .... . ... .. . .. . .. . . . .... .. . . . . . .27 for four different high-order flux . 8.. .. . .. . .......... . ......... . . ............26.. . .. .. ...... .... .. ..... . . .. . .. ..00125) 8.. .... . ... . . ......... ... . . ◦ computed ◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦ ◦◦◦◦◦ ◦ exact ◦ ◦ ◦ ◦ ◦ ◦◦◦◦ ◦ ◦ ◦ ◦ 0....... .. ... . ........ . .. ... ◦ computed ◦◦◦ ◦◦◦◦◦◦◦◦ ◦◦◦◦◦◦◦ ◦ exact ◦ ◦ ◦ ◦◦ ◦◦◦ ◦ ◦ ◦ 0........ ..... .... .. . ....... . ........ ... .00 ◦◦◦ ◦ ◦ ◦ ◦◦◦ ◦ ◦◦◦ ◦◦◦ −0..... .. ..15 Fig. .. . .... . κa t = 7 o . . . .. . . .. . ..... .... . .... . .. . .. . . .05 ◦ ◦◦◦◦ ◦ ◦ ◦ ◦ ◦ ◦◦ ◦◦◦◦ ◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦ −0..46. ..... .... .. 8.. ... .8) with C = 0.. ................ . .... .... . .. . .. .... . .. ... ..... .3 Flux Corrected Transport 211 u/ao . .......025) u/ao 0. .. ...... . .... ..... .... ... . . ...... ... Periodic boundary conditions are imposed on the left and right boundaries.. .. .... . . .........15 . . .... 0. .. ...... ..8 with  = 0. . ..... .. .00 ◦ ◦ ◦ ◦ ◦ ◦ ◦◦◦ ◦◦◦ ◦ ◦◦ ◦ −0... . .. . ... . Computed and exact solutions (κao Δt = 0. . ...05 ◦◦◦◦ ◦ ◦ ◦◦ ◦ ◦◦ ◦◦◦ ◦◦◦◦ ◦◦◦◦◦◦◦◦◦◦◦◦◦◦ −0... and a first-order time integration is employed. ... ....... ..15 ..10 0......... . .

.............. Osher..4 MUSCL-Hancock Method The MUSCL-Hancock method (Quirk..................... and Steger-Warming).. The velocity profile is displayed at κao t = 14 in Figs..................... ............................................. ..... ........32................ ... In Fig........................... ....... respectively..... 1985) is a TVD-like algorithm based on the concept of a slope limiter. VanLeer........................... ... ..............3 ........27.... TV(ρu) 0.............s 0 2 Godunov Roe Osher Steger-Warming Exact 4 6 8 10 12 .......................... ........... It is strictly TVD for a linear equation...00125 virtually eliminates the drop in TV as shown in Fig.............. . The decrease in TV is associated with the magnitude of Δt.. A reduction in Δt to Δt = 0........ experience has shown that it achieves TVD or near-TVD performance for the Euler equations (Toro. ....... the TV is shown for a second-order RungeKutta time integration with the same Δt..... 8........025) 8.....27.025 using firstorder Euler and second-order Runge-Kutta integrations...... ............................ The shock wave is diffused over several cells........ .......5 0.............. .. 8.... and the high-order flux is computed using a third-order upwind2 biased reconstruction (4....................... 1994..... ............... 8.......... .... We present here one version of the MUSCL-Hancock method..............................31 for Δt = 0.......30 and 8................... 8..........................................00125 minimizes the diffusion of the shock as shown in Fig.............................. TV(ρu) for FCT with different flux algorithms (first-order Euler with κao Δt = 0................... 8.............. ◦ ◦  ◦  ◦ 0. ......................................... ............... 14 κao t Fig...................... .4 0........ Roe......................... The decrease in TV is virtually the same as in Fig....19) with κ = 13 .... ............. however......................... A reduction in Δt to Δt = 0.............. 1997)....29.......................2 0..........1 0.............................. 8..0 t..............28.. ..................... All flux algorithms are evidently TVD with the incorporation of FCT....................212 TVD Methods algorithms (Godunov......................... The low-order l l flux Fi± 1 is computed using a first-order reconstruction (Qi+ 1 = Qi and 2 2 Qri− 1 = Qi )......................

.. ..1 Algorithm Consider the semi-discrete form (3.............................................................................................................2 0..........9) of the Euler equations  Fi+ 12 − Fi− 12  dQi + =0 (8................3 ◦  ◦ 0.. 0......................................4............... ........ ..................................3 ◦  ◦ 0........ 14 κao t Fig................................................... TV(ρu) for FCT with different flux algorithms (second-order RungeKutta with Δt = 0........28) .....................................................8...................................................1 0.......................4 MUSCL-Hancock Method 213 TV(ρu) 0.......... ........ ......27) dt Δx The MUSCL-Hancock Method proceeds in three steps........... 8.......s .................. .....................................................................0 0 2 Godunov Roe Osher Steger-Warming Exact 4 6 8 10 12 .. t.............................. ............ ◦ .........................5 .............................2 0................... .................................................................. 8....................4 ..................................................s .................................... ..................................4 ...................................... .... 0................................................................................0 0 2 Godunov Roe Osher Steger-Warming Exact 4 6 8 10 12 ........................................... ............ ................... TV(ρu) for FCT with different flux algorithms (first-order Euler with Δt = 0.................. ................................... First............. 14 κao t Fig......................5 ......... .................... ◦  ............................. t.. the flow variables are reconstructed within cell i using a linear expression: Q(x) = Qi + ξ ˆ Δi Δx (8................................................. 0......................................................025) TV(ρu) 0....... 0........................ ............... .....28....... .................29..................................00125) 8......................................................................1 0....◦......

.. .. . . . .... . . .. . . . . .10 0.. .31.. ... . ...... . ..... .. ..... ... ........ ......... . .. ....... ... ... .... . ..... ....... . ....... .. .. ... . .00 ◦ ◦ ◦ ◦ ◦ ◦ ◦◦◦◦ ◦ ◦ ◦ ◦ ◦ ◦◦◦◦ −0. ...15 Fig...... .......... .10 0.. .. .... . ... .. ... . ...... .... . ...... .. ..... κao t = 14 ◦ computed ◦◦◦◦◦◦◦◦◦ exact ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦◦◦◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ 0... ..10 ........ . .. ... . ... .. ... .214 TVD Methods u/ao ... .. . . . .. ..... ... ...... ..... ........ . ......... .. ..... ...... . . ..... .. ...... ..... .... . ...... ... .. . . ..... . .. . .. . . . ... . .. ..... ... ... .. . ....... .05 ◦◦◦◦◦ ◦ ◦◦◦◦◦ ◦ ◦ ◦ ◦ 5 ◦ 1 ◦◦◦◦◦◦◦2 3 4 6 ◦ κx ◦ ◦ 0. . .... .. ...... ...... .. .... .. ... ... .. ...... .. . ... ..... .. .. . . ... ... .. .. .05 ◦ ◦◦◦◦◦ ◦◦◦◦◦ ◦ ◦ ◦ ◦ ◦ 1 ◦◦◦◦◦◦◦2 3 4 6 ◦ 5 ◦ κx ◦ 0.....00 ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦◦◦ ◦ ◦ ◦ ◦ ◦ ◦◦◦◦ ◦ −0. . .. ........ .. . .. . . .. .. .. ..... ..... . .. . .. ... .. .. . .. .. . . . . ... . .. Velocity at κao t = 14 for FCT with Godunov’s Method and third-order reconstruction (second-order Runge-Kutta with Δt = 0.. ... . ... .... ... .. ... ..15 Fig. . ..... ... .29) ˆ i is a limited linear combination of ΔQi− 1 and ΔQi+ 1 according to and Δ 2 2 ˆ i = ς()Δi Δ (8.. . ... 8.. ...... ..... ... .... .. .. Velocity at κao t = 14 for FCT with Godunov’s Method and third-order reconstruction (first-order Euler with Δt = 0..... . . . .. ... .. . .. ...... ....... .. ..... ... .... ...... −0.. 0... ........30) Δi = 12 (1 − κ)ΔQi− 12 + 12 (1 + κ)ΔQi+ 12 (8..... . ... ... .... .. .. ..... ... ... . ...... .. . . .. . .. .. .. . .10 .. ... .. . 8. .... . .. ... .. .. ..025) u/ao .. ..... . ... . ..15 ..05 ◦◦◦◦ ◦ ◦◦ ◦◦ ◦◦◦◦◦◦ ◦◦◦◦◦◦◦◦◦◦◦◦◦ −0. . ....... −0.025) where ξ = x − xi (8... .... .....05 ◦◦◦◦ ◦ ◦◦◦◦◦◦ ◦◦ ◦◦◦◦◦◦◦◦◦◦◦◦◦◦ −0.... .. ....15 ...... .... .. . . . . κao t = 14 ◦ computed ◦◦◦◦◦◦◦ ◦ exact ◦ ◦ ◦ ◦ ◦ ◦ ◦◦ ◦◦◦◦ ◦ ◦ ◦ ◦ ◦ ◦◦ 0.. . .. . ...31) where Δi is . . .... .. .... ..... . ........ .. . ...... ...... .......30..... .. .. ...... . .. .. . .... ... ...... . ... . . . . .. ........ ... ..... ..... .. .. .. ..... ....... ... .. . .... ... .. . .. ...... .. .. . . .... .... ...... . .. ...... ... .. . .. ... .. 0. . . ...

. . . . . .. ... . .. .. . .. . .............. ... .. . ...... . .. .. ....34) where ςˆ = Thus. ... . ... . .. .. ... . .. . . ...... .. . . .. . ....... .. ......... Qli+ 1 2 ˆi = Qi + 12 Δ Qri− 1 ˆi = Qi − 12 Δ 2 (8... . . 2 2 ... .. ... ..... . ..... .. .. ... . ..... the boundary values are updated in time by 12 Δt according to  Δt  Fi+ 12 − Fi− 12 Δx  Δt  − 12 Fi+ 12 − Fi− 12 Δx ˆl 1 Q i+ = Qli+ 1 − ˆr 1 Q i− = Qri− 1 2 2 2 2 1 2 (8. . . ..... ..... ...05 ◦◦◦◦◦ ◦◦◦◦◦ ◦ ◦ ◦ ◦ 1 ◦◦◦◦◦◦◦◦2 3 4 5 6 κx 0.. .. ... ..... ... ... .... .. . . .. .... .... .. ... . .. . ......... ..... κao t = 14 ◦◦◦◦◦◦ ◦ computed ◦◦◦◦◦◦◦ ◦ exact ◦ ◦ ◦ ◦ ◦ ◦◦◦◦ ◦ ◦ ◦ ◦ ◦ 0.. . ..... .. .... .. ..... .00125) and −1 ≤ κ ≤ 1 and ς() is a slope limiter that depends on the ratio† = ΔQi− 12 ΔQi+ 12 (8.. Thus. . . ......36) † The ratio in (8.... .... ... . .. . . ......15 . . ... ..... . ..... ... .05 ◦◦◦◦◦ ◦◦◦◦ ◦ ◦◦◦◦◦◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦◦◦◦◦◦◦◦◦ −0... .. . ..... . . .. ... ...... ................ .. .... . ....32) is evaluated for each of the conservative variables.35) Second. ...... ..10 0. ... Velocity at κao t = 14 for FCT with Godunov’s Method and third-order reconstruction (second-order Runge-Kutta with Δt = 0...32) A slope limiter analogous to the MinMod flux limiter is (Toro... ........ ... 0. −0. ... . ..33) 2 1 − κ + (1 + κ) (8. 8. .......8.10 . . . ... . .... . ... . . ...... . ....... .. . ... .... . ... ......... .. .  = Δρi− 1 /Δρi+ 1 .. .. . ςˆ())  > 1 (8. .. .4 MUSCL-Hancock Method 215 u/ao .. 1997) ⎧ ⎪ ⎨ 0 ς= ≤0  0<≤1 ⎪ ⎩ min(1.. .15 Fig.... .. ....00 ◦◦◦ ◦ ◦ ◦ ◦ ◦ ◦◦◦ ◦ ◦ ◦ ◦ ◦◦◦◦ −0. . .. .. .32.... ... . .. for the conservation of mass... .. . ....

The norm.102).e i −Qi ||κao Δt=5×10−4 .46.1. 8.8) using the initial condition. and the domain 0 < κx < 2π is discretized into 100 uniform cells.2 Accuracy.3 Total Variation We examine the Total Variation of the solution of the problem described in Section 2. 1988).e is obtained from (2.92) with  = 0.34 for κao t = 7 and C = 0. i The convergence is displayed in Fig.46 and ΔtCF L is calculated according to (5.8. re2 2 spectively. 8. The exact solution Qn. the flow variables are updated according to = Qni − Qn+1 i  Δt  ˆ Fi+ 12 − Fˆi− 12 Δx (8. The initial condition is defined by (2. which is prior to the shock formation. 8.37) where Fˆi+ 1 is computed using Godunov’s method based on the flow variables 2 ˆ l 1 and Q ˆ r 1 .e i −Qi ||κao Δt =  κao Δt 5 × 10−4 2 n ||Qn.45.92) with  = 0. The timestep is defined by (5. 8. and ρe are † The line in Fig. Periodic boundary conditions are imposed on the left and right boundaries.85) to (2. The timestep Δt is held in fixed ratio to the grid spacing Δx.88).38) where the Courant number C = 0. The timestep Δt is determined by Δt = CΔtCF L (8. and Convergence We consider the problem described in Section 2.1 and the domain is 0 < κx < 2π. It can be shown that the method Q i+ 2 i+ 2 2 is second-order accurate in time (Hirsch.3) using Qli+ 1 and Qri− 1 . defined by (3. and similarly for Fˆi− 1 .4.216 TVD Methods where Fi+ 12 and Fi− 12 are computed from (2. 8. Consistency. and therefore Δt and Δx are decreased by the same ratio.8. The Total Variation for ρ. is evaluated at κao t = 7.33 is n ||Qn.4. ρu. The solution converges quadratically† to the exact solution for sufficiently small Δt. The initial condition is defined by (2. The computed result (using 100 cells) and exact solution are displayed in Fig. Third.8) with C = 0.33.

...... .. .. ◦ .. . ... ...... . ...10 0.. . ... ....... ... .. ..00 ◦◦◦ ◦ ◦ ◦◦◦ ◦ ◦◦◦ ◦ ◦ −0... ....... .10 κx −0...... . ............. .. .. ..... .. .... .. .. . . ..... .. . ........ .. . . ... .. .. . ... . . . 8............... .. . ......35 compared with the exact solution... .. .......... .. .. . ..... . ........ .. ... .. ...... . ....15 Fig.....05 ◦◦◦◦ ◦ ◦ ◦ ◦ ◦ ◦◦ ◦◦◦◦ ◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦ −0.. .. ... ..... . ... .. ...... . .. ... ... . .. .......... −2 . ..4 MUSCL-Hancock Method 217 n log10 ||Qn.... .. .. ... .... . ..... . .... .. ....36 for κao Δt = .. .... ..... . .... ...... ◦ computed ◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦ ◦◦◦◦ exact ◦ ◦ ◦ ◦ ◦◦ ◦ ◦ ◦ ◦ ◦ ◦ 0. .05 ◦◦◦◦ ◦◦◦ ◦ ◦ ◦ ◦ 2 3 4 5 6 ◦◦1◦ ◦ ◦ ◦ ◦ 0........ . . ..... .. . ...... . .. . . .... . . . ... ........ .. .. .. . ... .. ....... ... .. ... Convergence for MUSCL-Hancock u/ao ... .. . . .. . .34... ................ ...... ... .. . .. . .. .025 and 100 cells) shown in Fig. . 8. The agreement is very good.. . . .. ....... .. . .. ... ....... .. .. ...... ...... . The velocity profile is displayed at κao t = 14 in Fig.... . . . ... κa t = 7 o ....... . . ..... .. .. .......... . ... ...... .. . . . . .. .... ... . ... 8.. . ..... .. .. . ..... .. .. . . . ..... ... ..... ........ ........ . . . . . ..... 0.. ... .... . . .. . ... .... .33. ... ◦ ◦ ◦ ◦ −3 computed quadratic ◦ ◦ −4 ◦ ◦ ◦ −5 ◦ −6 −4 −3 −2 −1 0 log10 κao Δt Fig.15 .. Computed and exact solutions (κao Δt = 0.............. . ... ........ .. .... . .... ... ... .... 8. ...... ... .. .......... . ....... . ... . ... . ......... .. .... ... ..... .. . .8... . ....... ..... ...... . .... . .... . ...... .. . ......e i −Qi || 1 0 −1 .... .

....... ................................ 8. ............................... .................... ....... .. 0... . . ................. ... .. .. ............... and ρe (κao Δt = 0............. . ............ ... .................. . ... ...... ... ......... .........218 TVD Methods TV ..... .........00 .............. ρu........ 0...... .......... .... . .. .... ......... .... ........ ... .025 and 100 cells) ....... ............ ....15 Fig......... ......... ...... ... ... .......... ... .. ................ ........ 0.............025 and 100 cells) 0............................25 TV(ρe) computed TV(ρe) exact 1....37)............ Velocity at κao t = 14 for MUSCL-Hancock (κao Δt = 0. A decrease in the grid spacing and timestep by a factor of two improves the accuracy of the solution in the vicinity of the shock (Fig............ .. ...s ...... ..........025 and 100 cells......... .25 0.. ....... ..... 8. ............... .......... ......... ..... ...... TV(ρ) computed ◦ TV(ρu) computed ◦ TV(ρu) exact TV(ρ) exact 1....................36.......... ......15 ............... ......................... . .. 0 2 4 6 8 10 12 .................... ................................ . .. Total Variation of ρ.. ................. . ... . ................. κao t = 14 ◦ computed ◦◦◦◦◦ ◦◦◦◦◦◦ ◦◦◦ exact ◦ ◦ ◦ ◦ ◦◦◦ ◦◦◦◦◦ ◦ ◦ ◦ 0. ........................ ..05 ◦◦◦◦◦ ◦◦◦◦ ◦◦◦◦◦ ◦ ◦ ◦ ◦ ◦ ◦◦◦◦◦◦◦◦◦◦◦ −0........... ....................... .. ................. ............................05 ◦ ◦ ◦◦◦◦◦ ◦◦◦◦◦ ◦ ◦ ◦ ◦ 1 ◦◦◦◦◦ 2 3 4 5 6 ◦ κx 0........................................... ............ ............. ...... ........... ...... ...... ............... ......... .. .35... ........... .. .... .............. ........ 8... ....................... .. ................ ............ ........ .. .......50 ........................ .................................. . ................ . .. 1............ ...... ............ .... ............................... ....... ........... ... . . ................................ .......... .. .......... ...................... ................. ...... ....................... .......... ......... u/ao .... ........... ......... ............ ........ .............. .... 14 κao t Fig.......... .......50 ◦ .............. .................... ............ ...... ... ............................ ...... ..... . ... ..........75 0.. .. . ..10 . . ......... −0............ ....00 t................................ ... .............. ....... ...... . .... ..... . ..... ... . . ........... .. ... . ...................... ............10 0........ ..... ..... ...00 ◦◦◦◦◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦◦◦◦◦ ◦ −0...... . ............ ........ ....................... .................

.... .... ... ...... .. . .. Consider a single maximum at x1 and minimum at x2 where xo < x1 < x2 < xo + λ.... . ...... ... .. . .......... .... ..0125 and 200 cells) Exercises 8. ......... .. .. ..... .... . . .... ... . ... . . .... .. solution From (8.... .. .. .. .... .. .. ... . . .. .. . ........ . . ........ ..05 ◦◦◦◦ ◦◦◦◦◦◦◦ ◦ ◦◦◦◦◦◦◦◦◦◦◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦◦◦ ◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦ ◦◦◦◦ −0.... . Velocity at κao t = 14 for MUSCL-Hancock (κao Δt = 0.. .... .. . . .. .. .. . . . .. .. . ... .. .. . ... ...00 ◦◦◦◦◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦◦◦◦◦◦◦◦◦ ◦ ◦◦◦◦◦◦◦◦◦◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ −0... .... .. .. A straightforward evaluation of the above integral yields (8... .. . ..... . .05 ◦◦◦◦◦◦◦◦◦◦ ◦◦◦◦◦◦◦◦◦◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ 5 1 ◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦2 3 4 6 0.. .... .. . ... ...10 0... .3)........... . The interval may be subdivided into two intervals in which there is only one local maximum and one local minimum..  TV x1 = xo = = ∂u dx − ∂x  x2 x1 ∂u dx + ∂x  xo +λ x2 ∂u dx ∂x u(x1 ) − u(xo ) − (u(x2 ) − u(x1 )) + u(xo + λ) − u(x2 ) 2 (u(x1 ) − u(x2 )) since u(xo ) = u(xo + λ) from periodicity... . .......... . . .  ∂u = ∂x ≥0 ≤0 ≥0 for xo ≤ x ≤ x1 for x1 ≤ x ≤ x2 for x2 ≤ x ≤ xo + λ Thus.. .. .... . . . ..... . ... ... ... .... Now assume that there are two local maxima (and therefore two local minima) in xo ≤ x ≤ xo + λ......Exercises 219 u/ao .... . .... ... 8... ...... .. ... ....... . .. ..37. . .... Then.... by assumption.  xo +λ T V (u) = xo    ∂u   ∂x  dx where xo is an arbitrary point and λ is the wavelength. .. .. 0. ...... .. .5).. .. ..15 .. .... .. ... κx −0. . .. . ......... . ....1 For a periodic function u.. . ... .15 Fig. The above result can also be shown to be true if x2 < x1 . .. . .......... κao t = 14 ◦◦◦◦◦◦◦◦◦◦◦◦◦◦ ◦ computed ◦◦◦◦◦◦◦◦◦◦◦◦ ◦◦ ◦ exact ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦◦◦◦◦◦◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ ◦ 0. . ... .. . .. . . . . .. ..... .. ..... .. .... . ... .. .. show T V (u) = 2  (umaxi − umini ) where umaxi and umini are the local maxima and minima of u... .10 . .. ...... .. . .... ...... .... . ... ..

Therefore. ui and 2 2 ui+1 .2 8.220 8.10).3 TVD Methods Consider the Riemann Shock Tube Problem (Section 2. 1988) − + and Ci+ Ci+ 1 ≤ 0 1 ≥ 0 2 2 solution Using 1 dui =− dt Δx thus d 1 (ui+1 − ui ) = − dt Δx   C− i+ 1 2 Δui+ 1 + C + 2 i− 1 2  Δui− 1 2 − + − + Ci+ 3 Δui+ 3 + Ci+ 1 Δui+ 1 − Ci+ 1 Δui+ 1 − Ci− 1 Δui− 1 2 2 2 The discrete expression for T V (u) is T V (u) =  2 2 |ui+1 − ui | i and thus  d T V (u) d si+ 1 = (ui+1 − ui ) 2 dt dt i where si+ 1 = sign(ui+1 − ui ). For any function g. Is T V (ρ) constant for all time t > 0 ? Consider the scalar equation ∂u ∂f + =0 ∂t ∂x where f (u) is the flux. Assume that the tube is of finite length with impermeable ends. Show that the necessary conditions for the semi-discrete form to be TVD are (Hirsch. The equation may be rewritten in semidiscrete form as  1 ˆ dui =− fi+ 21 − fˆi− 21 dt Δx where fˆi+ 12 (ui . 2 d dg |g| = sign g dt dt 2 2 2  . we may write − + fˆi+ 21 − fˆi− 21 = Ci+ 1 Δui+ 1 + Ci− 1 Δui− 1 2 2 2 2 where Δui+ 12 = ui+1 − ui − + and Ci+ 1 and Ci− 1 are coefficients that may depend on ui−1 . ui+1 ) is the discrete flux at xi+ 12 and is assumed to depend only on ui and ui+1 .

a sufficient condition for the semi-discrete form to be TVD is either sign φi+ 1 = −si+ 1 2 where  φi+ 1 = si+ 1 2 2 − + Ci+ 1 − Ci+ 1 2 2  2 − + − Ci+ 1 si− 1 + Ci+ 1 si+ 3 2 2 2 2 for all Δui+ 1 or 2 φi+ 1 = 0 2 A total of eight different cases can be considered as indicated in the following table: Case si+ 3 si+ 1 si− 1 1 2 3 4 5 6 7 8 +1 +1 +1 +1 −1 −1 −1 −1 +1 +1 −1 −1 +1 +1 −1 −1 +1 −1 +1 −1 +1 −1 +1 −1 2 For Case 1.Exercises 221 except at g = 0. Consideration of these and the remaining cases (see Exercise 8. Thus. For Case 2. + Ci+ 1 2 i+ 1 2 2 2 2 ≤ 0.4 Evaluate Cases 5 through 8 in Exercise 8. For Case 3. where d|g|/dt is possibly undefined. C − Case 4. C + i+ 1 2 − C− i+ 1 2 ≥ 0. For ≥ 0.3 and show that the necessary conditions are + Ci+ 1 ≥ 0 − Ci+ 1 ≤ 0 2 2 . Then  C− i+ 3 2 i and  C+ i Thus.4) yields the necessary conditions C+ i+ 1 2 − Ci+ 1 ≥ 0 ≤ 0 2 8. d T V (u) 1 = dt Δx Δui+ 3 si+ 1 =  i− 1 2 2 2  2 i Δui+ 1 si− 1 C+ Δui+ 1 si+ 3 i+ 1 2  2 i+ 1 2 i − + Ci+ 1 − Ci+ 1 2  2 2 2 2 2 C− i Δui− 1 si+ 1 = si+ 1  2 2 2 si+ 1 2 2 2 − + − Ci+ 1 si− 1 + Ci+ 1 si+ 3 2 2   2 2 Δui+ 1 2 Therefore. φi+ 1 = 0. d T V (u) 1 = dt Δx   − + Ci+ 1 − Ci+ 1 2 i  − + Δui+ 1 − Ci+ 3 Δui+ 3 + Ci− 1 Δui− 1 2 2 Assume a periodic domain.

Assume ui−1 = ui . if Ki+ 1 > 0. Show that −1 − + Ci+ 1 + Ci+ 1 = [f (ui+1 ) − f (ui )] (ui − ui−1 ) 2 2 solution Consistency of the discrete flux expression fˆi+ 1 (ui .222 8.18)]. Assume the discrete flux is given by a fˆi+ 21 = (ui+1 + ui ) 2 8. then Fi+ 1 = Si+ 1 Ki+ 1 and hence Equation (8. where 2 f (u) is the flux.13). u) = f (u). then Fi+ 1 = 0.7 Is this method TVD? c Prove that the corrected flux Fi+ 1 has the same sign as the anti2 a diffusive flux Fi+ 1 [Equation (8. Then.5 TVD Methods Consider the scalar equation of Exercise 3.8 2 2 c holds. If Ki+ 1 ≤ 0. the flux difference expression fˆi+ 1 − fˆi− 1 = C − 1 Δui+ 1 + C + 1 Δui− 1 i+ i− 2 2 2 2 2 2 yields − fˆi+ 1 − f (ui ) = Ci+ 1 Δui+ 1 2 2 2 Similarly.3. assume ui = ui+1 and obtain + f (ui ) − fˆi− 1 = Ci− 1 Δui− 1 2 2 2 Combining these equations yields − −1 ˆ Ci+ 1 = fi+ 1 − f (ui ) (ui+1 − ui ) 2 + Ci− 1 2 2 = f (ui ) − fˆi− 1 (ui − ui−1 )−1 2 Combining again. where a is a constant. C− i+ 1 2 8. 2 2 c Prove Fi+ 1 = 0 if and only if 2      ˆ i = sign ΔQ ˆ i+2 = sign F a 1 sign ΔQ i+ 2  . 2 solution c From Equation (8.6 + C+ i+ 1 2 = [f (ui+1 ) − f (ui )] (ui − ui−1 )−1 Assume f = au for the scalar equation of Exercise 8. ui+1 ) requires fˆ(u.18) 2 2 8.

. . Therefore. there are four possible a cases depending on the sign of Fi± 1 . Prove that no new extrema are 2 ˆ i+2 ). .14) it is evident that Ki+ 1 < 0 if sign(ΔQ 2 (8.9 223 Prove that ˆ i+2 ) = −sign(ΔQ ˆi) sign(ΔQ c Fi+ if 1 = 0 2 solution ˆ i+2 ) = −sign(ΔQ ˆ i ). .Exercises 8. from From (8. Q created relative to the set (Q .13) c Fi+ 1 = 0 2 8.10 if ˆ i+2 ) = −sign(ΔQ ˆi) sign(ΔQ For Class 16 in Flux Corrected Transport. ˆ i−2 .

the sonic temperature T ∗ and total temperature To are related by T∗ = 2 To (γ + 1) 224 . and thus the total temperature To = H/cp is also conserved..e...3) indicates that the total enthalpy H = h + 12 u2 is conserved across the shock.. respectively....1) where in (8. 1 Normal shock at rest The Rankine-Hugoniot conditions (Table 2. u2 p2 T2 uw = 0 Fig. The component of the velocity normal to the shock is denoted by u..1) and definition of static enthalpy h = cp T have been used. u1 p1 T1 .3) the conservation of mass (8. uw = 0) as shown in Fig. Define the sonic static temperature T ∗ as the static temperature achieved when the flow is compressed or expanded adiabatically to sonic conditions (i.Notes Chapter 2 1 Consider a locally planar shock at rest in the frame of reference (i..3) ρ1 u21 (8.e. Thus. Equation (8. The flow states on the left and right sides are denoted by the subscripts 1 and 2. 1. ..1) yield ρ 1 u1 = ρ 2 u2 + p1 = ρ2 u22 + p2 (8.. at constant total enthalpy).2) h1 + 12 u21 = h2 + 12 u22 (8..

M∗1 M∗2 = 1 (8. Using the definition of a∗ .2). From (8. M∗ u a a a∗ u a ao a a o a∗  T T 1/2 o M To T∗ = = =  = M (γ − 1) 2 M 1+ 2 −1 1/2 (γ + 1) 2 Thus. Since a =  √ γRT = γp/ρ. equivalently.3) may be rewritten as (γ + 1) 2 a21 + 12 u21 = a∗ (γ − 1) 2(γ − 1) and since a∗ is conserved across the shock.4). from (8.Notes to pp. then a21 = a22 = (γ + 1) 2 (γ − 1) 2 a∗ − u1 2 2 (γ + 1) 2 (γ − 1) 2 a∗ − u2 2 2 and therefore a2∗ a2 + u1 = ∗ + u2 u1 u2 which yields the Prandtl relation u1 u2 = a2∗ or.4) where M∗ = u/a∗ . M22 (γ − 1) 2 = 1+ M1 2 γM12 (γ − 1) − 2 −1 (8.5) . 19–19 225 The sonic speed of sound a∗ is defined as a∗ =  γRT∗ and is conserved across the shock. ρ1 u21 + p1 = ρ2 u22 + p2 Divide the left side by ρ1 u1 and the right side by ρ2 u2 (noting ρ1 u1 = ρ2 u2 ). p2 p1 + u1 = + u2 ρ 1 u1 ρ 2 u2 a2 a21 + u1 = 2 + u2 γu1 γu2 Equation (8.

19–19 From (8.3).2). Thus.226 Notes to pp. u1 u2 u21 u1 u2 u21 a2∗ M∗21 (8.6) (γ + 1) 2 (γ − 1) 2 ρ2 = M1 1 + M1 ρ1 2 2 −1 From (8.1).4.5). p2 − p1 = = ρ1 u21 − ρ2 u22 . from (8. ρ2 ρ1 = = = = using (8.

 T2 2γ 2 = 1+ M1 − 1 T1 (γ + 1) From the definition of entropy. s2 − s1 = ln cv  (γ − 1)M12 + 2 (γ + 1)M12  2γ 2 1+ M1 − 1 (γ + 1) 2 + (γ − 1)M12 (γ + 1)M12 γ  The results for the stationary normal shock may be summarized as follows: ρ2 ρ1 u1 u2 p2 p1 = = = T2 T1 = M22 = s2 − s 1 cv (γ +1)M12 (γ −1)M12 + 2 ρ2 ρ1  2γ 2 1+ M1 − 1 (γ + 1)  2γ 2 M1 − 1 (γ + 1) 1+ (γ −1)M12 + 2 2γM12 − (γ −1)  = ln (γ − 1)M12 + 2 (γ + 1)M12  2γ 2 1+ M1 − 1 (γ + 1) 2 + (γ − 1)M12 (γ + 1)M12 γ  (8.1).7) . and thus p2 = 1 + γM12 p1 and thus  ρ1 ρ2 1−   p2 2γ 2 =1+ M1 − 1 p1 (γ + 1) From the ideal gas equation. u2 ρ1 u21 1 − u1 using (8.

From the normal shock relations applied in the frame of reference of the shock wave.. u1 p1 T1 ... uw Fig... 19–19 227 Additionally..Notes to pp. 2 Normal shock moving with velocity uw Case 1: u1 − uw > 0 The fluid moves from left to right across the shock in the frame of reference of the shock.. u2 p2 T2 . where the velocity components are shown in the frame of reference attached to the shock. .8) Now consider a frame of reference in which the shock is moving with velocity uw and the normal components of the velocity are ul and ur on the left and right sides of the shock.. 2γ p2 =1+ p1 (γ + 1) and thus  uw = u1 − a1 u1 − u w a1 ... There are two possible cases as indicated in Fig... respectively. the density and pressure ratio across the shock are related by ρ2 (γ − 1)p1 + (γ + 1)p2 = ρ1 (γ + 1)p1 + (γ − 1)p2 (8. 2.

The density and velocity are related by u1 − uw ρ2 = ρ1 u2 − uw (8.9).10) . (γ − 1) + (γ + 1)σl ρ2 = ρ1 (γ + 1) + (γ − 1)σl From the ideal gas equation. using (8.8). ρ2 u2 = u1 + uw ρ1   ρ2 −1 ρ1 which can be rewritten as u2 = u1 − (σl − 1) a1  γ (γ+1) σl + (γ−1) 2γ 2γ (8.2 −1 γ+1 (σl − 1) + 1 2γ where σl = p2 /p1 is the static pressure ratio across the shock. ρ1 T2 = σl T1 ρ2 and from (8.9) and furthermore. The negative sign is taken for the square root since u1 − uw > 0.

33–34 Case 2: u1 − uw < 0 The fluid moves from right to left across the shock in the frame of reference of the shock. From the normal shock relations. 2γ p1 =1+ p2 (γ + 1) and thus  uw = u2 + a2 uw − u 2 a2 .228 Notes to pp.

The density and velocity are related by ρ1 uw − u2 = ρ2 uw − u1 (8. and f (p∗ . using (8.3. The compatibility conditions are u2 p2 = = u3 p3 Using (2. .108) follow from the results in Section 2. the first condition becomes u1 − a1 f (p∗ . T1 ρ2 = σr T2 ρ1 and from (8.11). p) = 1 γ   p −1 p∗ (γ − 1) (γ + 1) p∗ + 2γ p 2γ −1/2 which correspond to (2.2 −1 γ+1 (σr − 1) + 1 2γ where σr = p1 /p2 is the static pressure ratio across the shock.100) to (2. p1 ) = u4 + a4 f (p∗ . Equations (2.98) and (2.46) and (2. p4 ) where p∗ denotes the static pressure at the contact surface (and hence p2 = p3 = p∗ ).99) for p∗ > p1 and p∗ > p4 .12) 2 The solution consists of four regions divided by the left shock. the contact surface.11) and furthermore. The positive sign is taken for the square root since u1 − uw < 0. u1 = ρ2 u2 + uw ρ1   ρ1 −1 ρ2 which can be rewritten as u1 = u2 + (σr − 1) a2  γ (γ+1) σr + (γ−1) 2γ 2γ (8. (γ − 1) + (γ + 1)σr ρ1 = ρ2 (γ + 1) + (γ − 1)σr From the ideal gas equation.8).51). and the right shock.

The compatibility conditions are u2 p2 = = u3 p3 Using (2.72). Since p∗ < p1 . the expression for f (p∗ . p4 ) where p∗ denotes the static pressure at the contact surface (and hence p2 = p3 = p∗ ). and the right expansion.46) and (2. . 35–37 229 3 The solution consists of four regions divided by the left shock. the contact surface. the contact surface. p1 ) = u4 + a4 f (p∗ . the first condition becomes u1 − a1 f (p∗ . p1 ) = u4 + a4 f (p∗ . p4 ) where p∗ denotes the static pressure at the contact surface (and hence p2 = p3 = p∗ ). p) on the right side of the equation is 1 f (p . and the right shock.134) follow from the results in Section 2.121) follow from the results in Section 2.3. p) = 1 γ   p −1 p∗ (γ − 1) (γ + 1) p∗ + 2γ p 2γ −1/2 Since p∗ < p4 . p) = (γ − 1)  p∗ p  (γ−1)/2γ −1 Equations (2.Notes to pp.122) to (2. the first condition becomes u1 − a1 f (p∗ . The compatibility conditions are u2 p2 = = u3 p3 Using (2. p1 ) = u4 + a4 f (p∗ . the contact surface.3.72). the expression for f (p∗ . p) on the left side of the equation is f (p∗ . p) on the right side of the equation is 2 f (p∗ .68) and (2.68). p) = γ ∗   p −1 p∗ (γ − 1) (γ + 1) p∗ + 2γ p 2γ −1/2 Equations (2. 5 The solution consists of four regions divided by the left expansion.51) and (2. the expression for f (p∗ . p) = (γ − 1)  ∗ p∗ p  (γ−1)/2γ −1 Since p∗ > p4 . the expression for f (p∗ . p4 ) where p∗ denotes the static pressure at the contact surface (and hence p2 = p3 = p∗ ). The compatibility conditions are u2 p2 = = u3 p3 Using (2.109) to (2. Since p∗ > p1 . 4 The solution consists of four regions divided by the left expansion. and the right expansion. p) on the left side of the equation is 2 f (p . the first condition becomes u1 − a1 f (p∗ .

p) = (γ − 1) p∗ p  (γ−1)/2γ −1 Equations (2. Δt) ∂t ∂x where E (Δx.14) Expanding each term in a Taylor series about xi at tn yields n+1 n (Ti+1 − Ti (Tin+1 − Tin−1 ) −κ Δt n−1 − Tin−1 + Ti−1 ) = 2 Δx ∂2T ∂T − κ 2 + E (Δx. 52–52 Since p∗ < p1 and p∗ < p4 . the expression for f (p∗ .135) to (2. Chapter 3 1 Inconsistent discrete approximations are relatively rare in computational science. The classic example is the DuFort-Frankel (1953) algorithm for the heat equation ∂T ∂2T =κ 2 ∂t ∂x (8. the algorithm is n+1 n n−1 − Tin−1 + Ti−1 ) 2 Δx (Ti+1 − Ti (Tin+1 − Tin−1 ) =κ Δt (8.151) follow from the results in Section 2.3.13) Assuming the temperature T is defined at a discrete set of equally spaced points xi . although they are certainly possible to create.230 Notes to pp. p) on the both sides of the equation is  2 f (p∗ . Δt) = κ Δt Δx .

For a grid refinement study employing a succession of smaller spatial grid spacings. Δx1 > Δx2 > . . . . ).13) provided lim Δx→0 Δt→0 Δt =0 Δx This can be achieved. > Δxj > . . this implies that the corresponding timestep must decrease according to Δtj+1 = Δtj  Δxj+1 Δxj 2 for j = 1. . . 2 .15) The discrete approxiation (8.14) is consistent with (8.g. and a corresponding sequence of smaller timesteps Δt1 > Δt2 > . . . for example. > Δtj > . .. Δx1 = Δx. . 2. (e.2 ∂2T + O(Δt)2 + O(Δx2 ) + O ∂t2  Δt4 Δx2  (8. if Δt ∝ κ−1 Δx2 . . . . . Δx2 = 1 Δx.

||A|| = sup x=0 ||Ax||| ||x|| where || · || is a suitable vector norm (see above). .Notes to pp. It was later declassified and published in Crank and Nicholson (1947) and Charney et al. This definition implies (Isaacson and Keller. . 3 The norm of a vector x is a measure of the length of the vector. examples of norms are |x|1 = i=n  |xi | i=1 |x|2 =  i=n   12 2 |xi | i=1 |x|p =  i=n   p1 p |xi | i=1 |x|∞ = max |xi | i where |x|2 is recognized as the Euclidean norm. 4 The norm of a matrix A is a measure of the “stretching” achieved by multiplication of a vector x by the matrix. The norm is denoted |x| and satisfies the following properties (Isaacson and Keller.13). 1988).16) which is a hyperpolic equation with fundamentally different properties than (8. (iii) For any scalar α and any x. 53–63 231 Conversely. 1966): (i) For each vector x. |x + y| ≤ |x| + |y| which is the triangle inequality. (1950) (Hirsch. 2 The Von Neumann Method for stability analysis was developed by John von Neumann at Los Alamos during World War II and was originally classified. The proof of the triangle inequality for these norms is presented in Isaacson and Keller (1966). . Thus. then for Δx → 0 and Δt → 0 the discrete approximation would be consistent with 2 c2 ∂T ∂2T 2∂ T − c + =0 ∂t2 ∂x2 κ ∂t (8. there is a unique real norm |x|. . n. |αx| = |α||x| (iv) For any vectors x and y. 1966) ||Ax|| ||A + B|| ||AB|| ≤ ≤ ≤ ||A|| ||x|| ||A|| + ||B|| ||A|| ||B|| . where c is a constant having dimensions of velocity. For a vector x with elements xi . (ii) The norm is nonnegative for all x and |x| = 0 if and only if x = 0. i = 1. if the timestep was selected according to Δt = c−1 Δx.

.. 3. . .. .... x .... .. 3 Alternating grids for Lax-Wendroff The first step of Lax-Wendroff is equivalent (to second order in Δx) to an explicit Euler integration of (3.. ... .... . Originally developed as a finite-difference algorithm. 1967)..1) on the second set of control volumes Vi for a time interval 1 Δt: 2 n+ 1 2 Qi = Qn i − Δt Fin + 1 − Fin − 1 2 2 2Δx . . ... 1960. Richtmyer and Morton.. . it can be readily interpreted as a finite-volume method using two sets of control volumes.. The volume-averaged vector of dependent variables on the first set of control volumes is defined by  1 Vi Qi (t) = Q dxdy Vi The1 second set of control volumes is a uniform discretization of the x-axis at time tn+ 2 = tn + 12 Δt into M cells of the same length Δx with centroids at xi+ 1 . i −1 tn+1 . . .. 69–69 5 A classic method is the two-step algorithm of Lax-Wendroff (Lax and Wendroff. The first set of control volumes is a uniform discretization of the x-axis at time tn into M cells of length Δx with centroids xi . i i +1 i i−1 .... M as illustrated in Fig. the 2 cells are shifted an amount 12 Δx with respect to the first set. . .... . ... Q dxdy Vi i i−1 tn+ 2  1 Vi Qi (t) = i+1 . .e. .... .. i. . i+1 Fig. i = 1.. . ... . The volume-averaged vector of dependent variables on the first set of control volumes is defined by tn 1 .232 Notes to pp. . .

to O(Δx)2 .1) on the first set of control volumes Vi for a time interval Δt: = Qn Qn+1 i − i Δt n+ 1 n+ 1 Fi+ 12 − Fi− 12 Δx 2 2 . where. Qn i = 1 2 n (Qn i + Qi−1 ) and the flux is evaluated by Fin + 1 = F (Qn i ) 2 The second step of Lax-Wendroff is equivalent to an explicit Euler integration of (3.

.

the algorithm is n+ 1 2 n (Qn i + Qi−1 ) − Qi = 1 2 Qn+1 i = Qn i − Δt n (F (Qn i ) − F (Qi−1 )) 2Δx n+ 1 n+ 1 Δt F (Qi +12 ) − F (Qi 2 ) Δx . 69–69 233 where the flux is evaluated using the results of the first step at tn+ 2 : 1 n+ 1 n+ 1 Fi+ 12 = F (Qi +12 ) 2 Thus.Notes to pp.

Expanding (8.2 and is assumed constant.8) defined in Section 2.2. The stability can be determined using the von Neumann method as described in Section 3.17) in a Taylor series and retaining the lowest order terms.17) (8. (8. n+ 1 2 Qi = 1 2  I− .8. Thus.18) The algorithm is second-order accurate in x and t (for fixed ratio Δt/Δx) and consistent. n+ 1 Δt n n A (Qn Qi 2 = 12 (Qn i + Qi−1 ) − i − Qi−1 ) 2Δx where A is the Jacobian matrix (2.1.

.

Δt Δt A Qn A Qn i + I + i−1 Δx Δx  Expanding (8.18) in a similar manner and substituting yields Qn+1 i = .

.

− Δt Δt A I− A Qn i+1 + I − 2Δx Δx + Δt Δt A I+ A Qn i−1 2Δx Δx Δt Δx .

Thus ˆn ˆ n+1 = GQ Q k k where the amplification matrix is G=I+ Δt Δx .2 A2 Qn i Substituting the Fourier series  l=N Q(x. where ι = √ −1 and k = 2πlL−1 . t) = ˆ k (t)eιkx Q l=−N+1 into the previous equation.

1964) λGi = 1 + α2i (cos kΔx − 1) − ιαi sin kΔx . Since G is a rational function of A.2 (cos kΔx−1) A2 − ι Δt sin kΔxA Δx The von Neumann stability condition is |λGi | ≤ 1 where λGi are the eigenvalues of G. its eigenvalues are (Protter and Morrey.

. ........ . .... 5. .. . .234 Notes to pp... . .. .... .. . . ..... .. . semi-major axis α2 ... ... ... ...... ... . .....3) is = = Real(λGi ) Imag(λGi ) ξ − (1 − α2 ) α2 2  2 + η α ≤1 This represents the interior of an ellipse with center at (ξ. .. . .. .... .. .. .... . and 1 are shown in Figs.... . .... .. . . .. ... ... . . ...... .. .. ................ Fig..... . .... . .. 6 to 8..... .... . .. . .. 0).. ... . . ... . .... .1 are shown in Fig.. . . ... . . ... .. .. ...... .. .. ... 2 . ..... Chapter 4 1 The results for κ = −1............. . .... α q .. ............. ....... .... . . ... 4 Stability diagram for Lax-Wendroff Let ξ η Then (8. .... .. ...... ..... .. ........ ...... . . . . . . .. .. . and semi-minor axis α.. The ellipse lies within the unit circle |λGi | = 1 provided |αi | ≤ 1 which is Δt ≤ min i Δx |λi | The results for the sample problem of Section 3..8.. 92–92 where αi = Δtλi Δx η . .. . . . . ..... ... . . ... ...... ... ... . .. 4.. . . .. . ....... . .. . ..... . . . . ..... η) = (1−α2 . ... . . ... ... . . ... ...... .. . . 0.... . ..... . . . . ...... .. . . .... .... ...... .. ...... . . ....... .. The ellipse is shown in Fig..... . ..... .... ... .. . . . .. . .. . ..... .... ξ 1−α 1 . .. .... . . .. . . ...4. The reconstructed left and .. ... . . .. ...... ..

. .. .15 Fig.. .. .. ... . ....... ... ....05 ◦◦◦ ◦◦◦ ◦ ◦ ◦ ◦ ◦ ◦◦◦ ◦◦ ◦◦◦ ◦◦◦ ◦◦◦◦◦ ◦◦◦◦ ◦◦◦◦ ◦ ◦ ◦ ◦◦◦◦◦◦◦◦◦◦◦ ◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦ −0... ....25 o o . 0..0 ... ...... . .. .. . .... .. ... . .44): Qi = Qi−a + aΔQi−a+ 1 + 2 This follows from 1 Δx  x x a(a − 1) ΔQi−a+ 3 − ΔQi−a+ 1 2 2 2 i+ 1 2 ........ . . .. .. . ..... .. .......... ......... ...... . .. . . ... ....10 −0...................... .. . . . .. . .. .......... .. ..... .. .. . .. . ..... ..05 ◦ ◦ ◦ ◦◦ ◦◦ ◦◦ ◦◦ ◦◦ ◦◦ ◦◦ ◦◦ ◦ ◦ ◦ ◦ 1 2 3 4 5 6 ◦◦ ◦ ◦◦ ◦◦ 0.. .. . . .......... .. ............. .0  .. ..... ....... .... .Notes to pp. . ... . . . ... . . .. . .. ... . .. . ... .. .. .. .... .. .... . .. .. ... ....... ........ x . 96–96 235 u/ao .. ... ... . . ... ...15 ..... ... ..... .  reconstructed solution exact solution ... . .......... −0.. ... ... −1. . . ..... ... . . ..... ..... . ... 5 .. ......... ............. ...... .... . ... ............ ...... ... .. ..... .... ........ .... . . ... .. .10 0... . ..... ..... ..5 ... .. .5 0. . ... .. ..... . . . . .. . . .. .. ........ ..... . . ...... Fig. ..00 ◦ ◦ ◦ ◦◦ ◦ ◦ ◦◦ κx ◦ ◦◦ ◦◦ ◦◦ ◦◦ ◦ ◦ ◦ ◦◦ ◦◦ ◦◦ ◦◦ ◦◦ ◦◦ ◦◦ −0. . . . 1 2  .... 6............ . .. .. ◦ computed ◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦ ◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦◦ ◦◦◦◦ ◦◦◦◦ exact ◦ ◦ ◦ ◦ ◦ ◦◦ ◦ ◦ ◦ ◦◦ ◦◦◦ ◦◦◦◦ ◦◦◦ ◦ 0.... . ..0  .. ............. .... ... .... ...... ....... ...... . ......... ..... ... ..... ....... . 0. . ........... .. . .. This is expected due to the lower order of accuracy.... ...  .. . . . . 1... . .. . . . .... 2 The following identity is used in deriving (4...... . ... .. . ........... . 5 Computed and exact solutions for Lax-Wendroff Q .. .... ..... .... . ..... .... ...... ... . ... ........ . ... . .. .. .... ....... . .. ..... . .... . ..... .... ... . .. ..... ... 3 ...   ...... ..... . 6 Reconstruction of sine using ten cells and κ = −1 with limiter right states are less accurate in comparison with the exact solution than for κ = 13 ... .... ... ...... . ..... ... ... . .... ........... . . 4 .. ...... .... .............  . . ... ... ...  . . .. .. κa t = 0 κa t = 1. . ..... . . .. ......

or 2. which are the only allowable values of a. 1. Q(x) dx = Qi i− 1 2 by construction. It may be verified directly for a = 0. .

. .. .. .. . ...... .... .. 2 −0.... .. .. 1 2 .. ...... Provided that there is no intersection of waves from the left and right face Riemann solutions at tn+1 ...... the solution for Q is therefore known at tn+1 as a combination of two general Riemann solutions and (possibly) an intermediate uniform region.... .. .... −0. 5 .. ... ........ ..   . ... .... 0. ... ... . .... ......... . ..... .......... ..  reconstructed solution exact solution ....  .0  .. .. ....... ... . . . .. .0  ..........5 . 0.. .... Fig.... x .... .... .5 .......... .. . ..... ... x ...0 .... ........ ... ...0 ..... ... .0 . . ..... ..... 8 Reconstruction of sine using ten cells and κ = 1 with limiter Chapter 5 1 Godunov’s First Method is based on the definition = Qn+1 i 1 Δx  Q(tn+1)dx Vi Consider the piecewise constant solution Qi for i = 1.... 1  reconstructed solution exact solution . ... ......5 .. −1.. ........ 3 . ....... 5 . . The left and right n states for Q at xi+ 1 may be taken to be Qn i and Qi+1 . ...  . ...... . ....  0... .. ... .... 6........ . ... −1... ...... ... .. The solution for Qn+1 is i .0 . ... . ..... .... 7 Reconstruction of sine using ten cells and κ = 0 with limiter Q . .5 . ....... . . 3 ..... 4 ..... .... 4 .......  ...... 107–107 Q ..  . . ... .... ............  . . 0. 1.... ... . . 9.. .... .... . ... Fig... ... 1...   ... .. ......  . ... ... .. respectively. . 6... . The general 2 Riemann problem is solved for each face as shown in Fig. M at tn ....... .........236 Notes to pp...  .....

. . . ... . This implies a constraint on the allowable timestep Δt. .. .. .. .. .. ....... .. . ........ .. ...... . ... .. .... . . ... ......... . .... . . . . ........... ... ............ ..... ...... ... . . .. .. ... . . n ... . .. .. ...... and xl and xr are the location of the rightmost wave from 2 QR and leftmost wave from QR at tn+1 . .. ..... .... ..... .... .. . .. . . . ... ...... ... ... .. . . n+1 .. .... . ... .. .... . .... Δx .. .. ..... .... .. ..... Fig...... ... .. .. . . ... ... .... . . . . . . ..Notes to pp. .. . .. ...... . . . ..... ... . .... . . . .... . .... .... ...... ... .. . . . ......... .. ..... ....... . .. ... ... . .. .... . The appropriate i− 1 i+ 1 2 2 modifications when xl < xi− 1 or xr > xi+ 1 are obvious. .... . .... .. .. .. ....... The method implies that 2 2 xl < xr . . . . .... ........ . ............ .. ......... i...... . ..... ... .. contact surface contact surface shock x x expansion fan t i−1 t i . 9 Possible flow structure for Godunov’s First Method 2 The vector R is R1 R2 R3 u − u)2 (γ − 1) (˜ 1 a2 ρ 1− − γa ˜2 2 a ˜2 = (γ − 1) (u − u ˜ )2 1 a2 (u − u ˜) + ρ + 4 a ˜2 2˜ a 2γ a ˜2 = (γ − 1) (u − u ˜ )2 (u − u ˜) 1 a2 + ρ − 4 a ˜2 2˜ a 2γ a ˜2 = and similarly ρ ρu = = ρe = R1 + R2 + R3 u ˜R1 + (˜ u+a ˜)R2 + (˜ u−a ˜)R3 1 2 ˜ +u ˜ −u u ˜ R1 + (H ˜a ˜)R2 + (H ˜a ˜)R3 2 . . . ... ... .... . expansion fan shock .. ... .. . .. ...... .... . .... ..... the wave systems of the two general Riemann problems do not intersect. ...... ..... .............. .. . l ..... . . ...... ...... ... ... x ... ..... . 114–114 therefore Qn+1 i ⎡  xl  xr  1 ⎣ n = QR dx + Q dx + 1 i i− Δx x 2 xl i− 1 2 237 x ⎤ i+ 1 2 R xr Qi+ 1 dx⎦ (8. . respectively......... . . .. . .......19) 2 where QR and QR are the solutions to the general Riemann problems at xi− 1 i− 1 i+ 1 2 2 2 and xi+ 1 ........ . . . . . . ... ... .. .. . ....... .... ......e.. . . . . ........ .... ... . . Δt i+1 ...20) where cmax is the maximum absolute wave speed of the waves entering cell i. ....... .... respectively........ ... ..... . . .... .... ... ... . . . .. . .. ..... ...... ..... ....... which may be estimated by Δt = min i Δx 2 cmax (8. . .. . .. ....... .. .. r .. . .. ..... . .......... . . . . ..... ... .. . . .. ....... ..... ... . . . .. .. ... ... . .. . .. . . . .. . .. . . . .. ... ... ... ... ... ... This method is less appealing than Godunov’s Second Method because the integration in (8. .... . . ... .....19) requires more cputime.

u2 = cc Thus. Also. T Λ+ =  Therefore. a 2 = a1 + (γ −1) (u1 − cc ) 2 Therefore. 4 It is necessary to investigate the condition a2 ≤ cc at the interfaces i ± 12 .19) and lk are the row vectors in (2. From (2. e3 = λ+ 3 r3 λ+ 2 r2 0 0 1 "  λ+ 3 r3  ek λ+ k rk Similarly.238 Notes to pp. For the left expansion. 128–141 3 An eigenvector rk of A is an eigenvector of A+ corresponding to the eigenvalue λ+ k if A+ rk = λ+ k rk Writing A+ T Λ+ T −1 =  = r1  r2   r3 ! l1 l2 l3 Λ+ "   T T −1 where rk are the column vectors in (2.21). u1 + 2 2 a1 = u2 + a2 (γ −1) (γ −1) Furthermore. T Λ+ T −1 rk = = ! " e2 = 0 1 0 λ+ 1 r1 λ+ 2 r2  λ+ 1 r1 ! . T −1 rk = ek where the unit vectors ek are defined as ! e1 = 1 0 0 " .20). the condition a2 ≤ cc becomes u1 + 2 (γ +1) a1 ≤ cc (γ −1) (γ −1) . rk is an eigenvector of A− corresponding to the eigenvalue λ− k.

2 2 a1 = cc + a1 u1 + (γ −1) (γ −1) Noting that p∗ = p2 and p2 = p1 a2 a1  p∗ p1 (γ−1)/2γ . 191–191 239 From the solution in Regions 2 and 3.Notes to pp.

the semi-discrete equation becomes dQi + dt aQi+ 1 − aQi− 1 2 2 Δx . Chapter 8 l 1 The diffusive character of Fi+ 1 can be illustrated for the linear case F = aQ where 2 a > 0 is a constant.2γ/(γ−1) we obtain cc ≥ a2 But since cc = u2 . l Fi+ 1 2 = aQi = aQi+ 1 − a 2 # Δx ∂Q # + O(Δx2 ) # 2 ∂x i+ 12 (8. to first order. u2 − a 2 ≥ 0 This is the same condition assumed for the cells i and i − 1 and therefore may be expected to hold. Using a simple first-order upwind reconstruction.21) Thus.

This equation is analogous to ∂Q ∂aQ ∂2Q + =ν ∂t ∂x ∂x2 (8. The l low order flux Fi+ 1 therefore generates an anti-diffusive contribution to the corrected 2 flux: c Fi+ 1 = αi+ 1 2 2 h l Fi+ 1 − Fi+ 1 2 2 .23) indicating the diffusive effect of a first-order upwind reconstruction for the flux.22) where ν = aΔx/2.  = Δx # # ∂Q # ∂Q # ν −ν # # ∂x i+ 12 ∂x i− 12 −1  (8.

(8.24) .

and Van Leer. 18:248–283. Tannehill. Boris. R. I. J and Nicholson. 22:87–110. (1948) Supersonic Flow and Shock Waves. P. Applied Mathematics Series.-W. Beam.. R. (1984) The Piecewise Parabolic Method (PPM) for Gas-Dynamical Simulations. L. New York. (1976) Flux-Corrected Transport III: Minimal-Error FCT Algorithms. and Woodward. and Pletcher. Technical Report NASA/CR-97-206253 (ICASE Report No. A Fluid Transport Algorithm That Works. SpringerVerlag. Anderson. Fjortoft. Tellus. New York. Journal of Computational Physics. Chandrasekhara. J. B. K. Hemisphere Publishing Corporation. Briley. 11:38– 69. I. J. J. C. 20:397–431.. Chu. J.. H. D. P. and Stegun. In Proceedings of the Fourth International Conference on Numerical Methods in Fluid Dynamics.. (1975) Flux-Corrected Transport II: Generalizations of the Method. DC. Charney. (1973) Solution of the Three-Dimensional Compressible Navier-Stokes Equations by an Implicit Technique. P. Washington. (1984) Computational Fluid Mechanics and Heat Transfer. and McDonald. and Von Neumann. Ahlfors. W. D. D. R. and Book. 54:174–201. K. J. (1971) Handbook of Mathematical Functions. 97-65). (1973) Flux-Corrected Transport. National Bureau of Standards. Anderson. (1950) Numerical Integration of the Barotropic Vorticity Equation. SHASTA. and Friedrichs. Thomas. Journal of Computational Physics. Dover. Springer-Verlag. R. and Warming. (1981) Hydrodynamic and Hydromagnetic Stability. NASA Langley Research Center. AIAA Journal. (1947) A Practical Method for Numerical Evaluation 240 . R. R. J. 24:1453–1460.Bibliography Abramowitz. Crank. New York. Journal of Computational Physics. Colella. M. (1953) Complex Analysis. and Hain. Journal of Computational Physics. McGraw-Hill. New York. D. 2:237–254. New York. Boris. Book. pages 105–110. (1976) An Implicit Finite-Difference Algorithm for Hyperbolic Systems in Conservation Law Form. J. Boris. (1986) Comparison of Finite Volume Flux Vector Splittings for the Euler Equations. volume 55. Journal of Computational Physics. and Book. (1997) Essentially Non-Oscillatory and Weighted Essentially NonOscillatory Schemes for Hyperbolic Conservation Laws. S. Courant.

SIAM Journal of Numerical Analysis. (1968) Matrix Theory. Englewood Cliffs. 49:357–393. (1964) Partial Differential Equations. AIAA Paper No. Proceedings of the Cambridge Philosophical Society. C. 43:50–67. M. H. Palo Alto. (1998) Computational Gas Dynamics. R. New York. Godunov. Schmidt. 7:135–152. C. C. TX. Englewood Cliffs. and Frankel. (1987) Uniformly High Order Accurate Essentially Non-oscillatory Schemes. Engquist. New York. Prentice-Hall. (1988) Assessment of Riemann Solvers for Unsteady One-Dimensional Inviscid Flows of Perfect Gases. Journal of Computational Physics. second edition. Inc.. Inc. 78:437–458. Godunov. New York. and Groth. 36:321–351. Greenberg. A.. S. In Russian. S. SIAM Journal of Numerical Analysis.. P. (1998) Advanced Engineering Mathematics. B. 153:6–25. New York. volume I and II. and Osher. Journal of Computational Physics. Springer-Verlag. E. NJ. C. S. Isaacson. 81-1259. S. Fletcher. John Wiley and Sons. AIAA 14th Fluid and Plasma Dynamics Conference. Journal of Computational Physics. Engquist. K. Cambridge. Mathematics of Computation. Austin. (1988) Numerical Computation of Internal and External Flows. Harten. and Turkel. John Wiley and Sons. 71:231–303. Matematiqeski i Sbornik. S. Prentice-Hall. Jameson. Harten. John Wiley and Sons. NJ. 47:357–393. B. NJ. Journal of Computational Physics. E. Mathematical Tables and Other Aids to Computation. P. Harten. (1981) Numerical Solutions of the Euler Equations by Finite Volume Methods Using Runge-Kutta Time-Stepping Schemes. C. Holt. (1999) Reminiscences About Difference Schemes. C. A. Cambridge University Press. Hoffman. (1959) Fluid Mechanics. Springer-Verlag. E. (1966) Analysis of Numerical Methods. (1980) One Sided Difference Approximations for Nonlinear Conservation Laws. J. Gear. W. 24:279–309. M. and Chakravarthy. 21:1–23. and Keller. III.Bibliography 241 of Solutions of Partial Differential Equations of the Heat Conduction Type. W. Lax. DuFort. Prentice-Hall.. Pergammon Press. (1989) Computational Fluid Dynamics for Engineers. (1983) High Resolution Schemes for Hyperbolic Conservation Laws. Osher. E. (1959) Weak Solutions of Nonlinear Hyperbolic Equations and Their Nu- . Adam Hilger. CA. (1959) A Finite Difference Method for the Computation of Discontinuous Solutions of the Equations of Fluid Dynamics. and Osher. Gottlieb. New York. and Lifshitz. L. (1988) Parallel Computers 2. S. Bristol. Laney. Education Systems. Landau. Franklin. UK. volume I and II. J. (1953) Stability Conditions in the Numerical Treatment of Parabolic Differential Equations. Hockney. New York. (1984) On a Class of High Resolution Total Variation Stable Finite Difference Schemes. A. (1987) Uniformly High-Order Accurate Nonoscillatory Schemes. A. and Jesshope. (1971) Numerical Initial Value Problems in Ordinary Differential Equations. Garabedian. I. A. Harten. Hirsch. Englewood. (1988) Computational Techniques for Fluid Dynamics. (1984) Numerical Methods in Fluid Dynamics.. S.

188:489. P. (1994) An Alternative to Unstructured Grids for Computing Gas Dynamic Flows Around Arbitrarily Complex Two Dimensional Bodies. 1977a. Liu. Upper Saddle River. Steger. Springer-Verlag. Mathematics of Computation. (1912) Graphical Methods. (1964) Modern Mathematical Analysis. K. (1953) The Dynamics and Thermodynamics of Compressible Fluid Flow...-D. Matsuno. (1977a) Towards the Ultimate Conservative Difference Scheme III. (1997) Riemann Solvers and Numerical Methods for Fluid Dynamics.. B. Liepmann. New York. Press. (1957) Elements of Gasdynamics. Runge. and Chan. New York. P. Lax. 9:267–293. (1981) Flux Vector Splitting of the Inviscid Gasdynamic Equations with Applications to Finite-Difference Methods. and Solomon. J. Roe. and Fink. (1986) Characteristic-Based Schemes for the Euler Equations. R. Columbia University Press. J. C. Prentice-Hall. A.242 Bibliography merical Computation. Cambridge. (1956) Survey of the Stability of Linear Finite Difference Equations. and Bulirsch. P. Journal of Computational Physics. K. X. (1994) Weighted Essentially Non-Oscillatory Schemes. F. Communications on Pure and Applied Mathematics. R. NJ. and Wendroff. C. (1986) Numerical Recipes. R. D. S. and Morrey. 50:447–481. Osher. 13:217–237. New York. (1960) Systems of Conservation Laws. 23:125–142. Morton. Communications on Pure and Applied Mathematics. S. Interscience Publishers. T. and Warming. R. Journal of Computational Physics. B. E. (1998) Numerical Methods Using Matlab. and Richtmyer. K. and Morton.. New York. K. Mathews. A. and Prendergast. Japan. 23:263–275. Stoer. Protter. S. Parameter Vectors. 115:200–212. editor. Journal of Computational Physics. K. and Mayers. New York. Schreier. Reading. and Difference Schemes. Shapiro. S. Osher. (1994) Numerical Solution of Partial Differential Equations. Toro. John Wiley and Sons. 40:263–293. (1967) Difference Methods for Initial-Value Problems. SpringerVerlag. John Wiley and Sons. and Roshko. Teukolsky. Van Leer. (1981) Approximate Riemann Solvers. W. Osher. Journal of Computational Physics. S. Addison-Wesley Publishing Co. 43:357–372. Journal of Computational Physics. (1970) Introduction to Numerical Analysis. Quirk. MA. (1982) Upwind Difference Schemes for Hyperbolic Conservation Laws. and Chakravarthy. (1982) Compressible Flow. (1983) Upwind Schemes and Boundary Conditions with Applications to Euler Equations in General Coordinates. 38:339–374. 7:159–193. Communications on Pure and Applied Mathematics. R. P. Annual Review of Fluid Mechanics. J. and Vetterling. H. W. Sanders. second edition. New York.Proceedings of the Parallel CFD 2002 Conference. Flannery. . Cambridge University Press. S. Astrophysical Journal. Upstream-Centered Finite Difference Schemes for Ideal Compressible Flow. M. Lax. B. John Wiley and Sons. Kansai Science City. J. Computers and Fluids. Richtmyer. 18:337–365. New York. (1974) . Cambridge University Press. Roe. (2003) Parallel Computational Fluid Dynamics: New Frontiers and Multidisciplinary Applications . Cambridge.

Berlin. B. Journal of Computational Physics. McGraw-Hill. (1979) Towards the Ultimate Conservative Difference Scheme V. pages 507–512. A Second Order Sequel to Godunov’s Method. 132:1–2. (1982) Flux Vector Splitting for the Euler Equations. Journal of Computational Physics.Eighth International Conference on Numerical Methods in Fluid Dynamics. Van Leer. Van Leer. (1985) On the Relation Between the Upwind-Differencing Schemes of Godunov. F. SIAM Journal on Scientific and Statistical Computing. S. . 5:1–20. 23:276–299. New York. (1977b) Towards the Ultimate Conservative Difference Scheme IV. A New Approach to Numerical Convection. B. 31:335–362. (1974) Viscous Fluid Flow. White. Journal of Computational Physics. In Lecture Notes in Physics . Van Leer. B. Zalesak. 32:101–136. Volume 170. Springer-Verlag.Bibliography 243 Van Leer. B. (1979) Fully Multidimensional Flux-Corrected Transport Algorithms for Fluids. Van Leer. Journal of Computational Physics. B. Enguist-Osher and Roe. (1997) Godunov’s Method for Gas Dynamics: Current Applications and Future Developments.

Index Accuracy 47 Beam and Warming’s Method 173 Stability 180 Consistency 47 Contact surface 18 Control volume 2 Conservation laws Differential form 2 Entropy 5 Integral form 2 Internal energy 5 Convective derivative 3 Convergence 48 Crank-Nicholson 51 Domain of dependence 25 Domain of influence 26 DuFort-Frankel 230 Entropy Definition 3 Equation 4 Equivalence Theorem of Lax and Richtmyer 68 Essentially Non-Oscillatory Methods (ENO) 92 Euler equations Characteristics 17 Conservative form 10 Conservative variables 10 Euler’s Identity 14 Flux vector 10 In terms of Q 10 Homogeneity property 14 Jacobian 11 Diagonal matrix Λ 13 Eigenvalues 12 Eigenvectors (right) 12 Eigenvectors (left) 13 In terms of Q 11 In terms of u and e 12 In terms of u and a 12 In terms of u and H 12 Relationship between left and right eigenvectors 13 Similarity form 13 Matrix T 13 Matrix T −1 13 Nonconservative form 11 Multiple-valued solutions 22 Semi-discrete form 47 Expansion fan 24 Explicit Euler 51 Extrema 80 Flux Corrected Transport 191 Fourier series 54 General Riemann Problem 32 Contact surface pressure 32 f (p∗ . 90 MUSCL-Hancock 212 Newton’s formula 76 No New Extrema 81 Osher’s Method 127 Semi-discrete stability 139 Primitive function 75 Rankine-Hugoniot conditions 18 Reconstruction 74 Riemann invariants 17 Riemann Shock Tube 39 Roe’s Method 109 A˜ 113 Entropy fix 122 Matrix properties 111 Semi-discrete stability 117 244 . p) 33 Expansion and shock 36 Two expansions 37 Two shock waves 34 Shock and expansion 35 Gibbs phenomenon 60 Godunov’s Method 105 Internal energy 1 Lax-Wendroff 232 Limiter 81 Flux limiter 191 Slope limiter 191 Mach number 4 Method of Lines 47 Modified Upwind Scheme for Conservation Laws (MUSCL) 81.

Index Runge-Kutta 171 Stability 175 Shock wave 19 Formation 26 Sinusoidal disturbance 30 Simple wave 22 Specific heat 1 Speed of sound 4 Stability 48 Static pressure 10 Static temperature 1 Steger and Warming’s Method 148 Eigenvalues 149 Semi-discrete stability 153 Wave speed splitting 149 Temporal integration Explicit 171 Implicit 172 Thermally perfect 1 Thermodynamics Second Law 4 Total energy 2 Total enthalpy 5 Total Variation 95. 187 Total Variation Bounded 188 Total Variation Diminishing 188 Trapezium 51 Trapezoidal rule 49 Van Leer’s Method 157 Semi-discrete stability 163 Von Neumann stability analysis 53 245 .

Sign up to vote on this title
UsefulNot useful