This action might not be possible to undo. Are you sure you want to continue?
Harvard Lomax and Thomas H. Pulliam NASA Ames Research Center David W. Zingg University of Toronto Institute for Aerospace Studies August 26, 1999
Contents
1 INTRODUCTION
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.1 Problem Speci cation and Geometry Preparation . . . . . . 1.2.2 Selection of Governing Equations and Boundary Conditions 1.2.3 Selection of Gridding Strategy and Numerical Method . . . 1.2.4 Assessment and Interpretation of Results . . . . . . . . . . . 1.3 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.4 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Conservation Laws . . . . . . . . . . . . 2.2 The NavierStokes and Euler Equations . 2.3 The Linear Convection Equation . . . . 2.3.1 Di erential Form . . . . . . . . . 2.3.2 Solution in Wave Space . . . . . . 2.4 The Di usion Equation . . . . . . . . . . 2.4.1 Di erential Form . . . . . . . . . 2.4.2 Solution in Wave Space . . . . . . 2.5 Linear Hyperbolic Systems . . . . . . . . 2.6 Problems . . . . . . . . . . . . . . . . . . 3.1 Meshes and FiniteDi erence Notation 3.2 Space Derivative Approximations . . . 3.3 FiniteDi erence Operators . . . . . . 3.3.1 Point Di erence Operators . . . 3.3.2 Matrix Di erence Operators . . 3.3.3 Periodic Matrices . . . . . . . . 3.3.4 Circulant Matrices . . . . . . . iii . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1
1 2 2 3 3 4 4 5
2 CONSERVATION LAWS AND THE MODEL EQUATIONS
7 8 12 12 13 14 14 15 16 17
7
3 FINITEDIFFERENCE APPROXIMATIONS
. . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
. . . . . . .
21
21 24 25 25 25 29 30
3.4 Constructing Di erencing Schemes of Any Order . . . . . 3.4.1 Taylor Tables . . . . . . . . . . . . . . . . . . . . 3.4.2 Generalization of Di erence Formulas . . . . . . . 3.4.3 Lagrange and Hermite Interpolation Polynomials 3.4.4 Practical Application of Pade Formulas . . . . . . 3.4.5 Other HigherOrder Schemes . . . . . . . . . . . . 3.5 Fourier Error Analysis . . . . . . . . . . . . . . . . . . . 3.5.1 Application to a Spatial Operator . . . . . . . . . 3.6 Di erence Operators at Boundaries . . . . . . . . . . . . 3.6.1 The Linear Convection Equation . . . . . . . . . 3.6.2 The Di usion Equation . . . . . . . . . . . . . . . 3.7 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
. . . . . . . . . . . .
31 31 34 35 37 38 39 39 43 44 46 47
4 THE SEMIDISCRETE APPROACH
4.1 Reduction of PDE's to ODE's . . . . . . . . . . . . . . . . . . . . . . 4.1.1 The Model ODE's . . . . . . . . . . . . . . . . . . . . . . . . 4.1.2 The Generic Matrix Form . . . . . . . . . . . . . . . . . . . . 4.2 Exact Solutions of Linear ODE's . . . . . . . . . . . . . . . . . . . . 4.2.1 Eigensystems of SemiDiscrete Linear Forms . . . . . . . . . . 4.2.2 Single ODE's of First and SecondOrder . . . . . . . . . . . . 4.2.3 Coupled FirstOrder ODE's . . . . . . . . . . . . . . . . . . . 4.2.4 General Solution of Coupled ODE's with Complete Eigensystems 4.3 Real Space and Eigenspace . . . . . . . . . . . . . . . . . . . . . . . . 4.3.1 De nition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3.2 Eigenvalue Spectrums for Model ODE's . . . . . . . . . . . . . 4.3.3 Eigenvectors of the Model Equations . . . . . . . . . . . . . . 4.3.4 Solutions of the Model ODE's . . . . . . . . . . . . . . . . . . 4.4 The Representative Equation . . . . . . . . . . . . . . . . . . . . . . 4.5 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1 Basic Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 Model Equations in Integral Form . . . . . . . . . . . . . . . . . . . 5.2.1 The Linear Convection Equation . . . . . . . . . . . . . . . 5.2.2 The Di usion Equation . . . . . . . . . . . . . . . . . . . . . 5.3 OneDimensional Examples . . . . . . . . . . . . . . . . . . . . . . 5.3.1 A SecondOrder Approximation to the Convection Equation 5.3.2 A FourthOrder Approximation to the Convection Equation 5.3.3 A SecondOrder Approximation to the Di usion Equation . 5.4 A TwoDimensional Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
51
52 52 53 54 54 55 57 59 61 61 62 63 65 67 68 72 73 73 74 74 75 77 78 80
5 FINITEVOLUME METHODS
71
5.5 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
83 86 87 88 89 90 91 91 92 93 93 95 95 96 97 97 98 99 100 101 102 102 103 105 106 107 110 110 111 112 115 117
6 TIMEMARCHING METHODS FOR ODE'S
6.1 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2 Converting TimeMarching Methods to O E's . . . . . . . . . . . . 6.3 Solution of Linear O E's With Constant Coe cients . . . . . . . . 6.3.1 First and SecondOrder Di erence Equations . . . . . . . . 6.3.2 Special Cases of Coupled FirstOrder Equations . . . . . . . 6.4 Solution of the Representative O E's . . . . . . . . . . . . . . . . 6.4.1 The Operational Form and its Solution . . . . . . . . . . . . 6.4.2 Examples of Solutions to TimeMarching O E's . . . . . . . 6.5 The ; Relation . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.5.1 Establishing the Relation . . . . . . . . . . . . . . . . . . . . 6.5.2 The Principal Root . . . . . . . . . . . . . . . . . . . . . . 6.5.3 Spurious Roots . . . . . . . . . . . . . . . . . . . . . . . . 6.5.4 OneRoot TimeMarching Methods . . . . . . . . . . . . . . 6.6 Accuracy Measures of TimeMarching Methods . . . . . . . . . . . 6.6.1 Local and Global Error Measures . . . . . . . . . . . . . . . 6.6.2 Local Accuracy of the Transient Solution (er j j er! ) . . . 6.6.3 Local Accuracy of the Particular Solution (er ) . . . . . . . 6.6.4 Time Accuracy For Nonlinear Applications . . . . . . . . . . 6.6.5 Global Accuracy . . . . . . . . . . . . . . . . . . . . . . . . 6.7 Linear Multistep Methods . . . . . . . . . . . . . . . . . . . . . . . 6.7.1 The General Formulation . . . . . . . . . . . . . . . . . . . . 6.7.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.7.3 TwoStep Linear Multistep Methods . . . . . . . . . . . . . 6.8 PredictorCorrector Methods . . . . . . . . . . . . . . . . . . . . . . 6.9 RungeKutta Methods . . . . . . . . . . . . . . . . . . . . . . . . . 6.10 Implementation of Implicit Methods . . . . . . . . . . . . . . . . . . 6.10.1 Application to Systems of Equations . . . . . . . . . . . . . 6.10.2 Application to Nonlinear Equations . . . . . . . . . . . . . . 6.10.3 Local Linearization for Scalar Equations . . . . . . . . . . . 6.10.4 Local Linearization for Coupled Sets of Nonlinear Equations 6.11 Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.1 Dependence on the Eigensystem 7.2 Inherent Stability of ODE's . . 7.2.1 The Criterion . . . . . . 7.2.2 Complete Eigensystems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
85
7 STABILITY OF LINEAR SYSTEMS
121
122 123 123 123
. . . . . . . . . . . . . 7.7. . . . . . . . . . .5.3 Relation to Circulant Matrices . . .6. . . . . . .3. . . . . . . . . . 123 124 124 125 125 125 128 128 132 135 135 136 137 141 141 142 143 143 146 8 CHOICE OF TIMEMARCHING METHODS 8. . . . . . . . .8 7. . . . . 7. . . . . . . . . . . . . . . . . . . . . . . . . . 8.1 Sti ness De nition for ODE's . . . . . .7 Problems . . . . . . . . . . . . . . . . . . Numerical Stability Concepts in the Complex Plane . . . . . . .1 Root Traces Relative to the Unit Circle . . . . . . . .2 Driving and Parasitic Eigenvalues . . . . . . . . . . . 8. . . . . . .3 7. . . . . . . . . . . . . . . . . . . . . . . . . 8. . .7. . . . . . . . . . . . . . . 8.2. . . . . . .1 Stability for Large h.7. . . . . . . . . . . . 149 149 149 151 151 152 153 154 154 154 155 158 158 159 160 160 161 . . . . .5 Coping With Sti ness . . . Problems . . . . . . . . .2 Implicit Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . AStable Methods . . . . . .3 Defective Eigensystems . . . . . . . . . . . . . . . . . . .3. . . .3 Defective Eigensystems . . . . . . . 7. . .6. . . 8. . . . . . . . . . . . . .6 7. . . .2 Unconditional Stability. 8. . . . .5. . . . . . . . . .5 7.1 The Basic Procedure .2 Relation of Sti ness to Space Mesh Size . . . Numerical Stability of O E 's . . . . . . . . . . . . . . . . . .1 Explicit Methods . . 7. . . . . . . . . . . .2 Stability for Small t . . . . . . . . . . . . . .5. 8. . . . . . . . . .3 Practical Considerations for Comparing Methods . . . . . . . .2 An Example Involving Di usion . . . . . .2 Some Examples . . . . TimeSpace Stability and Convergence of O E's . 7. . . .2 Complete Eigensystems . . . . . .3 Sti ness Classi cations .4. . . . . . .7. . . . . . . . . . . . . .4. . . . .6 Steady Problems . .1. . . . . . . . . . . . . 8. . . .7 7. . . . . . . . .1 Imposed Constraints . . .4. 8. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1. . . . . . . . .9 7. . . . . . . . . . . . . . . . . 7. . . 7. . . . . . . . . . . . . . . . . . . . . . 7. . . . . . . . . Fourier Stability Analysis .1. . . . . 8. . Consistency .1 The Criterion . . . . . . . . . . . . . . . . .3 An Example Involving Periodic Convection . . .3. . . . . . . . 8. . . . . . . . . .1 Relation to Eigenvalues . . . . . . . . . . . . . 8. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4 7. .4 Comparing the E ciency of Explicit Methods . . . .5. . . . . . . . . . . . . . .6. . . 8. .5. . . . . . . . . . . . . . . . . . . . . . 7. . . . . . .3 A Perspective . 7. . . . . . . 8. . . . .3 Stability Contours in the Complex h Plane. . . . . . . . . . . . . . . . 8. . . Numerical Stability Concepts in the Complex h Plane 7. . . . .
. . . . . . . . . . . . . . . . . . . . . . . . . 9. . . . . . . . . . . . . . . . . . . . . . . . . . . 11.6 Problems . . .2 The GaussSeidel System . . . . . . .3 A TwoGrid Process .2 Properties of the Iterative Method . .3. . . . . . . . . . . . . . .9 RELAXATION METHODS 9. . . . . . . . . . .4. 191 11 NUMERICAL DISSIPATION 203 204 205 207 209 210 212 213 214 12 SPLIT AND FACTORED FORMS 12. . . . . . . . . . . . . . . . . . . . . . . . . . . .1. . . . . . . . . . . . . . . . . . . the Residual. 9. . . . . . . . . . . . . . . . . . . .1 Formulation of the Model Problem . . . . . . . . . . . . . . . 11. . . . . . . . . . .5 Arti cial Dissipation . . . . . . . . . . 9. . .1 Preconditioning the Basic Matrix . . . . . . Upwind Schemes . . . . . . . . . . . . . . . . . . . . . . . . . . .1. . . . . . . . . . . . . . . 9. .3 11. . . . .1 The Concept . . .1 Eigenvector and Eigenvalue Identi cation with Space Frequencies191 10. . . . . . . . . . .2 11. . . . . . . . . .4 OneSided FirstDerivative Space Di erencing The Modi ed Partial Di erential Equation . . . . . . . . . . . . . . . . .3 Factoring Space Matrix Operators in 2{D . . . . . . . . . .3 The Classical Methods . . . . 191 10. . . . . .2. . . . . . . . . . . . .4. . . . . . . . . . . . . .1 The Ordinary Di erential Equation Formulation . . . . . . . . . . . . 163 164 164 166 168 168 168 169 170 170 172 173 174 176 180 182 187 10 MULTIGRID 10. . . . 9. . . .2 The Model Equations . . .1 The PointJacobi System . . . . . . . . . . . . .3 The SOR System . . 11. . . . . . . . . . . . . . . . . . . . . . 9. . . . . . . . . 202 11.2. . . 218 12. . . . . 9. . . . . . . . . 9. . . . and the Error 9. . . . . . . .3 The ODE Approach to Classical Relaxation . . . . . . .1 Motivation . . . . 9. . . . .5 Nonstationary Processes .2 The Converged Solution. .4 Eigensystems of the Classical Methods . . . . . . . . 217 12.1 FluxVector Splitting . . . . . . . . . . . . . . . . .4 Problems . .2 The Basic Process . . . . . . . 11. 9. . . 9. .4. 192 10. . . . . . . . . . . . . . . . . . . .4. . . .1 11. . . . . . . . . . . . . . . . . . . . . . . . . . 9. . . . . . . . . . The LaxWendro Method . . . . . . . .3.1 The Delta Form of an Iterative Scheme . . . . . . . . . 220 217 . . . . . . .4. . . . . . . . . . . . . . . . . . . .2. . . . . . . . . . . 9. . .1. .1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9. . . . .2 ODE Form of the Classical Methods . . . . . . . . . . . . . . . . . . . .2 FluxDi erence Splitting . . . . . . . . . .2 Classical Relaxation . . . . . . . . . .2 Factoring Physical Representations  Time Splitting . 200 10. . 192 10. . . .6 Problems . . . . . . . . . . . . . . .
3. . . .1 The Representative Equation for Circulant Operators . . . Eigensystems of Circulant Matrices . . . . . . . . . . .2 Analysis of a SecondOrder TimeSplit Method . . . . . . . . . .6 12. .2 General Circulant Systems . . . . . . . . . . . . . . . . . . . .2 B. . . . B. De nitions . . . . . . . . . . . . . . . . . .3 Data Base Permutations . .3. . . . .4. . . . . . .2 A. . . . . . . . . . .3 The Factored Delta Form of the Implicit Euler Method . .5 Special Cases Found From Symmetries . . . . .4. . . . . . . A. . . . .3 B. . . . . . . . . . . . . . . .2 The Factored Nondelta Form of the Implicit Euler Method 13. . . . . . . . . . . 13. . . . . .4 Example Analysis of 2D Model Equations . . . . . . . . . . . . . . . . . . . 12. . . . . . . . . . . .3 A.1 B. . . . . . . . . . . . . . . . . . . . . . . 257 . . . . . . . . . Vector and Matrix Norms .4. .4. . . . . .5 B. .6 Special Cases Involving Boundary Conditions . . . . . . . . . . . SecondOrder Factored Implicit Methods . .4 Space Splitting and Factoring . Generalized Eigensystem for Simple Tridiagonals . . . . . . . . .4 Notation . . . .1 Mesh Indexing Convention . . .4 The Factored Delta Form of the Trapezoidal Method . . . . . . . .2 Example Analysis of Circulant Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13. . . 13. . .4 A. . .1 Stability Comparisons of TimeSplit Methods . . . . . . Problems . . . . . . . . . . . .2.3 The Representative Equation for SpaceSplit Operators . . . . . . . 12. . . . . 13. . . . . . . . . . . . . . . . . . . . . . . . . . .4. .5 12. . . . . . . . . . . . . . . . . . The Inverse of a Simple Tridiagonal . . . . . . . . . . . . . . .3. . . . . . . . . . . . . . . . . . . . . . . . . . . . .12. . . . . . . . . . . . . . . .2. Eigensystems . . . 233 233 234 234 237 238 242 242 243 243 244 245 247 249 250 251 251 254 257 258 259 260 260 261 262 263 A USEFUL RELATIONS AND DEFINITIONS FROM LINEAR ALGEBRA 249 B SOME PROPERTIES OF TRIDIAGONAL MATRICES Standard Eigensystem for Simple Tridiagonals . . . B. . . .6 Problems . . . . . . . . . . . .2 Data Bases and Space Vectors . . . . . . . . 13. . . . . . . . . . . . . . . . . 13. . . . .4 12. . . . . . . . Algebra .4. . . . . . . . . 13. 13. . . . . . 13. . . . . . . . . . . . . . . . . . . . . .5 Example Analysis of the 3D Model Equation . . . .7 12. . . . B. . .1 Standard Tridiagonals . . . . . . . . . . . . . . 12. . . . . . . . . . Importance of Factored Forms in 2 and 3 Dimensions The Delta Form . . . . . . . . . . . . . . . . . .3. . . . .1 The Unfactored Implicit Euler Method . . . . . . .1 A. . . B. . . . . . . . . . . . . . . 13. . . . . . . . . . 220 221 221 223 226 226 228 229 13 LINEAR ANALYSIS OF SPLIT AND FACTORED FORMS 13. . . . . .
C THE HOMOGENEOUS PROPERTY OF THE EULER EQUATIONS265 .
can change the form of the basic partial di erential equations themselves. In the eld of aerodynamics. of course. Was there one mathematical structure which could be used to describe the behavior and results of most numerical methods in common use in the eld of uid dynamics? Perhaps the answer is arguable.Chapter 1 INTRODUCTION 1. This. and turbulence. Many of the most important aspects of these relations are nonlinear and. These events are related to the action and interaction of phenomena such as dissipation. the use of numerical methods to solve partial di erential equations introduces an approximation that. The ultimate goal of the eld of computational uid dynamics (CFD) is to understand the physical events that occur in the ow of uids around and within designated objects. motivates the numerical solution of the associated partial di erential equations. all of these phenomena are governed by the compressible NavierStokes equations. convection. Experience has shown that such is not the case. as a consequence. The principal such constraint was the demand for uni cation. in e ect. di usion. which 1 . underlying constraints on the nature of the material began to form. As we shall see in a later chapter. The new equations. but the authors believe the answer is a rmative and present this book as justi cation for that belief. shock waves. At the same time it would seem to invalidate the use of linear algebra for the classi cation of the numerical methods.1 Motivation The material in this book originated from attempts to understand and systemize numerical solution techniques for the partial di erential equations governing the physics of uid ow. slip surfaces. often have no analytic solution. As time went on and these attempts began to crystallize. The mathematical structure is the theory of linear algebra and the attendant eigenanalysis of linear systems. boundary layers.
Mathematically. of course. these errors are often identi ed with a particular physical phenomenon on which they have a strong e ect. and algorithm development in general. if used and interpreted properly. even if the errors are kept small enough that they can be neglected (for engineering purposes). factorization. including the geometry. INTRODUCTION are the ones actually being solved by the numerical process. However. these methods give very useful information. Independent of the speci c application under study. with identifying an error with a physical process. that most numerical methods in practical use for solving the nondissipative Euler equations create a modi ed partial di erential equation that produces some form of dissipation. This means that the errors caused by the numerical approximation result in a modi ed partial di erential equation having additional terms that can be identi ed with the physics of dissipation in the rst case and dispersion in the second. convergence. they can. simulate the physical phenomena listed above in ways that are not exactly the same as an exact solution to the basic partial di erential equation. and the requirements of the simulation. 1. are often referred to as the modi ed partial di erential equations. 1. as long as the error remains in some engineering sense \small". they can lead to serious di culties. if their e ects are not thoroughly understood and controlled. the following sequence of steps generally must be followed in order to obtain a satisfactory solution. physical reality. the theory associated with the numerical analysis of uid mechanics was developed predominantly by scientists deeply interested in the physics of uid ow and. Thus methods are said to have a lot of \arti cial viscosity" or said to be highly dispersive. Since they are not precisely the same as the original equations. and probably will. producing answers that represent little. There is nothing wrong. The geometry may result from .2 Background The eld of computational uid dynamics has a broad range of applicability. However. as a consequence. This motivates studying the concepts of sti ness.2 CHAPTER 1. On the other hand. It is safe to say. if any. Regardless of what the numerical errors are called.2. nor with deliberately directing an error to a speci c physical process. the resulting simulation can still be of little practical use if ine cient or inappropriate algorithms are used.1 Problem Speci cation and Geometry Preparation The rst step involves the speci cation of the problem. these di erences are usually referred to as truncation errors. and consistency. All of these concepts we hope to clarify in this book. This motivates studying the concepts of stability. for example. ow conditions.
Turbulence is an example of a physical process which is rarely simulated in a practical context (at the time of writing) and thus is often modelled. Alternatively. ii) the drag as well as the lift and pitching moment. the turnaround time required. symmetry boundaries. The partial di erential equations resulting from these conservation laws are referred to as the NavierStokes equations. for example. the Reynolds number and Mach number for the ow over an airfoil. Boundary types which may be encountered include solid walls. and the thinlayer NavierStokes equations.2. or mesh.ow equations. . It is generally accepted that the phenomena of importance to the eld of continuum uid dynamics are governed by the conservation of mass. it is always prudent to consider solving simpli ed forms of the NavierStokes equations when the simpli cations retain the physics which are essential to the goals of the simulation. must be selected. However. which is known as a grid. while the NavierStokes equations require the noslip condition. Possible simpli ed governing equations include the potential. Flow conditions might include. the Euler equations require ow tangency to be enforced. in a design context. The success of a simulation depends greatly on the engineering insight involved in selecting the governing equations and physical models based on the problem speci cation.1.3 Selection of Gridding Strategy and Numerical Method Next a numerical method and a strategy for dividing the ow domain into cells. periodic boundaries. one may be interested in i) the lift and pitching moment only. in ow and out ow boundaries. an appropriate set of governing equations and boundary conditions must be selected. and the solution parameters of interest. As an example of solution parameters of interest in computing the ow eld about an airfoil. no geometry need be supplied.2 Selection of Governing Equations and Boundary Conditions 1. or iii) the details of the ow at some speci c location. at a solid wall.2. We concern ourselves here only with numerical methods requiring such a tessellation of the domain. 1. and energy. physical models must be chosen for processes which cannot be simulated within the speci ed constraints. The boundary conditions which must be speci ed depend upon the governing equations. in the interest of e ciency. These may be steady or unsteady and compressible or incompressible. etc. a set of objectives and constraints must be speci ed. The rst two of these requirements are often in con ict and compromise is necessary.2. BACKGROUND 3 measurements of an existing con guration or may be associated with a design study. For example. If necessary. or elements. Once the problem has been speci ed. Instead. The requirements of the simulation include issues such as the level of accuracy needed. momentum. the Euler equations.
when used intelligently. composite. For example. We believe that a thorough understanding of what happens when . or spectral methods. hybrid. Hence their numerical solution can illustrate the properties of a given numerical method when applied to a more complicated system of equations which governs similar physical phenomena. Tannehill. Here again. INTRODUCTION Many di erent gridding strategies exist. the grid can be altered based on the solution in an approach known as solutionadaptive gridding. The choices of a numerical method and a gridding strategy are strongly interdependent. and overlapping grids. and understand most numerical methods used to nd solutions for the complete compressible NavierStokes equations. they have been carefully selected to be representative. and understanding such methods. of di culties and complexities that arise in realistic two. the success of a simulation can depend on appropriate choices for the problem or class of problems of interest. Many of these issues are not addressed in this book. It is critical that the magnitude of both numerical and physicalmodel errors be well understood. and can be aided by sophisticated ow visualization tools and error estimation techniques. unstructured. nitevolume. to develop. These model equations isolate certain aspects of the physics contained in the complete set of equations. Furthermore. niteelement. we present a foundation for developing. This step can require postprocessing of the data. the results of the simulation must be assessed and interpreted. Instead we focus on numerical methods. Although the model equations are extremely simple and easy to solve. The numerical methods generally used in CFD can be classi ed as nitedi erence. which are still evolving.2.and threedimensional uid ow simulations. we can make use of much simpler expressions. 1.3 Overview It should be clear that successful simulation of uid ows can involve a wide range of issues from grid generation to turbulence modelling to the applicability of various simpli ed forms of the NavierStokes equations. Fortunately. analyzing. Rather than presenting the details of the most advanced methods. Some of them are presented in the books by Anderson. with emphasis on nitedi erence and nitevolume methods for the Euler and NavierStokes equations. the use of nitedi erence methods is typically restricted to structured grids. analyze. 1.4 Assessment and Interpretation of Results Finally. the socalled \model" equations.4 CHAPTER 1. including structured. for example calculation of forces and moments. and Pletcher 1] and Hirsch 2].
it should be noted that. Bold type is reserved for real physical vectors. such as velocity. Some of the abbreviations used throughout the text are listed and de ned below. Otherwise. proportional to) . NOTATION 5 numerical approximations are applied to the model equations is a major rst step in making con dent and competent use of numerical approximations to the Euler and NavierStokes equations. however. a scalar quantity in the linear convection and di usion equations. there are many aspects of practical problems which can only be understood in the context of the complete physical systems.e. although we can learn a great deal by studying numerical methods as applied to the model equations and can use that information in the design and application of numerical methods to practical problems.4 Notation The notation is generally explained as it is introduced. 1. usually a vector A term of order (i.4.. kD O( ) ~ bc Partial di erential equation Ordinary di erential equation Ordinary di erence equation Righthand side Particular solution of an ODE or system of ODE's Fixed (timeinvariant) steadystate solution kdimensional space Boundary conditions.S. For example. the variable u can denote a scalar Cartesian velocity component in the Euler and NavierStokes equations.S. The vector symbol ~ is used for the vectors (or column matrices) which contain the values of the dependent variable at the nodes of a grid. and a vector consisting of a collection of scalars in our presentation of hyperbolic systems.1. PDE ODE O E RHS P. the use of a vector consisting of a collection of scalars should be apparent from the context and is not identi ed by any special notation. As a word of caution. S.
6 CHAPTER 1. INTRODUCTION .
2. These equations contain many of the salient mathematical and physical features of the full NavierStokes equations.. and energy. such as the Euler and NavierStokes equations and our model equations. which is natural for nitedi erence schemes. stability. can be written in the following integral form: Z V (t2 ) QdV . mass. The equation is a statement of 7 . and the recurring use of these model equations allows us to develop a consistent framework of analysis for consistency. though.1) In this equation. e. and convergence. in particular. The model equations we study have two properties in common. The equations are then recast into divergence form.g.1 Conservation Laws Conservation laws. The concepts of convection and di usion are prevalent in our development of numerical methods for computational uid dynamics. per unit volume. will be on representative model equations. the integral conservationlaw form. Z V (t1 ) QdV + Z t2 I t1 S (t) n:FdSdt = Z t2 Z t1 V (t) PdV dt (2. and they represent phenomena of importance to the analysis of certain aspects of uid dynamic problems. accuracy. The Euler and NavierStokes equations are brie y discussed in this Chapter. They are linear partial di erential equations (PDE's) with coe cients that are constant in both space and time.Chapter 2 CONSERVATION LAWS AND THE MODEL EQUATIONS We start out by casting our equations in the most general form. momentum. Q is a vector containing the set of variables which are conserved. which is useful in understanding the concepts involved in nitevolume schemes. The main focus. the convection and di usion equations.
is an area A(t) bounded by a closed contour C (t). and z coordinate directions. If all variables are continuous in time. Those methods which make various approximations of the derivatives in Eq. 2. respectively. momentum and energy for a uid. containing the ux of Q per unit area per unit time. On the other hand. a partial derivative form of a conservation law can also be derived.2 The NavierStokes and Euler Equations The NavierStokes equations form a coupled system of nonlinear PDE's describing the conservation of mass.2) dt V (t) S (t) V (t) Those methods which make various numerical approximations of the integrals in Eqs. 2.4) r: i @x @y @z and i j.8 CHAPTER 2.5) @t @x with 2 3 2 3 u 3 2 0 6 6 Q=6 6 6 4 e 7 7 u7 7 7 5 E 6 6 =6 6 6 4 u(e + p) 7 6 7 6 2+p 7. they can be written as @Q + @E = 0 (2. or tensor. or cell. and P is the rate of production of Q per unit volume per unit time. t1 . 2.1 can be rewritten as d Z QdV + I n:FdS = Z PdV (2. 2. The divergence form of Eq. leading to @Q + r:F = P (2. For a Newtonian uid in one dimension.3 and nd a solution for Q on that basis are referred to as nitedi erence methods. CONSERVATION LAWS AND THE MODEL EQUATIONS the conservation of these quantities in a nite region of space with volume V (t) and surface area S (t) over a nite interval of time t2 . then Eq. F is a set of vectors.3) @t where r: is the wellknown divergence operator given. by ! @ +j @ +k @ : (2. Many of the advanced codes written for CFD applications are based on the nitevolume concept. in Cartesian coordinates.2 and nd a solution for Q on that basis are referred to as nitevolume methods.6 u 7 6 7 6 5 4 4 @u 3 @x 4 3 u @u + @x @T @x 7 7 7 7 7 5 (2. the region of space. 2. and k are unit vectors in the x y. In two dimensions.6) .2 is obtained by applying Gauss's theorem to the ux integral.1 and 2. The vector n is a unit vector normal to the surface pointing outward.
In twodimensional Cartesian coordinates. In order . THE NAVIERSTOKES AND EULER EQUATIONS 9 where is the uid density. Thus the conservative form is appropriate for solving ows with features such as shock waves. T is the temperature. they can lead to quite di erent numerical solutions in terms of shock strength and shock speed. and Pletcher 1] and Hirsch 2]. u is the velocity. p is the pressure. Tannehill. Note that the convective uxes lead to rst derivatives in space. This form of the equations is called conservationlaw or conservative form. The ux vectors given above are written in terms of the primitive variables.2. Later on we will make use of the following form of the Euler equations as well: @Q + A @Q + B @Q = 0 (2. v. such as u and p. These equations must be supplemented by relations between and and the uid state as well as an equation of state. u.9) 4 q3 5 4 v 5 4 5 4 5 + 7 q4 e u(e + p) v ( e + p) where u and v are the Cartesian velocity components.8) @t @x @y with 2 3 2 3 2 2 q1 u 3 v 3 6 7 6 7 6 2+ 7 6 7 Q = 6 q2 7 = 6 u 7 E = 6 u uv p 7 F = 6 v2uv p 7 6 7 6 7 6 7 6 (2. Many ows of engineering interest are steady (timeinvariant). The steady solution to the onedimensional NavierStokes equations must satisfy @E = 0 (2.7) @x If we neglect viscosity and heat conduction. and p. we are often interested in the steadystate solution of the NavierStokes equations. or at least may be treated as such. while the viscous and heat conduction terms involve second derivatives. . for example.10) @t @x @y @E @F The matrices A = @Q and B = @Q are known as the ux Jacobians. and is the thermal conductivity. The total energy e includes internal energy per unit volume (where is the internal energy per unit mass) and kinetic energy per unit volume u2=2. the Euler equations are obtained. is the coe cient of viscosity. For such ows. such as the ideal gas law. these can be written as @Q + @E + @F = 0 (2. Nonconservative forms can be obtained by expanding derivatives of products using the product rule or by introducing di erent dependent variables. Details can be found in Anderson.2. with no interest in the transient portion of the solution. Although nonconservative forms of the equations are analytically the same as the above form. e is the total energy per unit volume.
1 q2 7 2 q1 7 7 7 5 3 q3 2 q1 (2. q1 .12) We have assumed that the pressure satis es p = ( . 2 q1 2 q2 q1 . and q4 . as follows: 2 6 6 6 6 6 =6 6 6 6 6 6 4 E1 E2 E3 E4 E 2 3 6 7 6 7 6 7 6 7 6 7 6 7=6 7 6 7 6 7 6 7 6 7 6 5 6 6 4 ( 3 7 7 7 2 2 7 3. . 1)q4 + 2 q1 2 q1 7 7 7 7 q3 q2 7 7 q1 7 7 7 5 3 2q q4 q2 . 1) e . 1 q3 2 . 1)q4 + 3. where is the ratio of speci c heats. q2.1 0 a42 !2 a43 q2 q1 3 7 7 7 7 7 7 7 7 7 7 7 7 7 5 (2.1 q2 + q3 2 2 2 q1 2 q1 q1 q2 (2. q2 q1 q1 3 a41 ! . CONSERVATION LAWS AND THE MODEL EQUATIONS to derive the ux Jacobian matrices. From this it follows that the ux Jacobian of E can be written in terms of the conservative variables as 2 6 6 6 6 6 @Ei = 6 6 6 @qj 6 6 6 6 6 4 0 1 (3 .10 CHAPTER 2. (u2 + v2)=2] from the ideal gas law. q3 . ) q3 q1 q2 q1 0 A= a21 q .13) where a21 = . 3 . 2q . q2 . .1 q3 7 . we must rst write the ux vectors E and F in terms of the conservative variables.11) F 2 6 6 6 6 6 =6 6 6 6 6 6 4 3 2 F1 7 6 7 6 7 6 6 F2 7 6 7 6 7=6 7 6 7 6 F3 7 6 7 6 7 6 5 6 4 F4 q3 q3 q2 q1 ( . ) q2 q1 q q3 1 0 (1 .1 q2 2 3 2 q1 3 7 7 7 7 7 7 7 7 2 7 . cp=cv . + . 2 q4 q 3 q1 2 q3 q1 .
.( . 1)u(u2 + v2) . 1 (3u2 + v2 ) 2 . aspects of convective phenomena associated with coupled systems of equations such as the Euler equations are important in developing numerical methods and boundary conditions. This motivates the choice of our two scalar model equations associated with the physics of convection and di usion. 1) . 3 . The eigenvalues of the ux Jacobian matrices are purely real. Furthermore.16) Derivation of the two forms of B = @F=@Q is similar.14) 3 7 7 7 7 7 7 7 7 7 7 7 5 0 1 0 0 0 a21 (3 . 1 43 q2 a42 = q1 2 q1 ! ! q2 q3 a43 = . This is the de ning feature of hyperbolic systems of PDE's.15) u where a41 = ( . 1)4 ! 11 q2 q1 !3 + q3 q1 2 !2 !2 q2 q1 !3 5.5. THE NAVIERSTOKES AND EULER EQUATIONS a41 = ( 2 . !2 3 5 q4 q1 ! q2 q1 ! q4 . 1 v2 .2. The homogeneous property of the Euler equations is discussed in Appendix C. which are further discussed in Section 2. The NavierStokes equations include both convective and di usive uxes. ue a42 = e . )uv (2.uv a41 a21 = v a42 u a43 (2. )u (1 . . u2 2 2 a43 = (1 . 1) q q 1 1 and in terms of the primitive variables as 2 6 6 6 6 6 A=6 6 6 6 6 6 4 + q3 q1 (2.2. . Thus we also study linear hyperbolic systems of PDE's. )v ( .
as the convection problem. Hence the wave leaves the domain through the out ow boundary without distortion or re ection. This is consistent in terms of the wellposedness of a 1storder PDE. at) (2. (2) In the other type. simply. At any given time. This is referred to as the biconvection problem. what enters on one side of the domain must be the same as that which is leaving on the other.18) The initial waveform propagates unaltered with speed jaj to the right if a is positive and to the left if a is negative. the waveform travels through one boundary and reappears at the other boundary. Now let us consider a situation in which the initial condition is given by u(x 0) = u0(x).12 2. It is the simplest to study and serves to illustrate many of the basic properties of numerical methods applied to problems involving convection. and the domain is in nite.3 The Linear Convection Equation 2.17) @t @x Here u(x t) is a scalar quantity propagating with speed a. we pay a great deal of attention to it in the initial chapters. The manner in which the boundary conditions are speci ed separates the following two phenomena for which this equation is a model: (1) In one type. It is easy to show by substitution that the exact solution to the linear convection equation is then u(x t) = u0(x . Note that the lefthand boundary is the in ow boundary when a is positive. the \out ow" boundary. the ow being simulated is periodic. Hence. corresponding to a wave entering the domain through this \in ow" boundary. without special consideration of boundaries. CONSERVATION LAWS AND THE MODEL EQUATIONS The simplest linear model for convection and wave propagation is the linear convection equation given by the following PDE: @u + a @u = 0 (2. the scalar quantity u is given on one boundary. This type of phenomenon is referred to. It represents most of the \usual" situations encountered in convecting systems.1 Di erential Form CHAPTER 2. With periodic boundary conditions. the process continues forever without any . eventually returning to its initial position. while the righthand boundary is the in ow boundary when a is negative. No boundary condition is speci ed at the opposite side.3. In this case. a real constant which may be positive or negative.
With such an initial condition.23) where the frequency.2. the wavenumber. For M modes. We restrict our attention to initial conditions in the form u(x 0) = f (0)ei x (2. and the phase speed. In order to satisfy the periodic boundary conditions. in a simulation. 2. !. Substituting this solution into the linear convection equation.20 gives the following solution u(x t) = f (0)ei (x. and is the wavenumber.19. we nd that f (t) satis es the following ordinary di erential equation (ODE) df = .25) . 2.2 Solution in Wave Space u(x 0) = M X m=1 fm (0)ei mx (2. It is a measure of the number of wavelengths within the domain. Eq. This means that the phase speed is the same for all wavenumbers.ia f (2.!t) (2.17. THE LINEAR CONVECTION EQUATION 13 change in the shape of the solution. Preserving the shape of the initial condition u0(x) can be a di cult challenge for a numerical method. waves with di erent wavenumbers travel at di erent speeds. The linear relation given by Eq. one obtains 2. We now examine the biconvection problem in more detail. An arbitrary initial waveform can be produced by summing initial conditions of the form of Eq. 2.20) where the time dependence is contained in the complex function f (t). .22) Substituting f (t) into Eq.21) dt which has the solution f (t) = f (0)e.24 is characteristic of wave propagation in a nondispersive medium.ia t (2. 2. a. are related by != a (2.3. must be an integer.24) The relation between the frequency and the wavenumber is known as the dispersion relation.at) = f (0)ei( x. most numerical methods introduce some dispersion that is. As we shall see later.19) where f (0) is a complex constant. the solution can be written as u(x t) = f (t)ei x (2. Let the domain be given by 0 x 2 .3.
Other steadystate solutions are obtained if a source term g(x) is added to Eq. Since the wave equation is linear. 2.4.30) .23.1 Di erential Form Di usive uxes are associated with molecular motion in a continuum uid. In contrast to the linear convection equation.28) @x2 @u = @t " 2 @u @x2 .4 The Di usion Equation 2.29) giving a steady statesolution which satis es @ 2 u . 2. u must vary linearly with x at steady state such that the boundary conditions are satis ed. this parabolic PDE governs the di usion of heat in one dimension. the di usion equation has a nontrivial steadystate solution. g(x) = 0 @x2 (2. the solution is obtained by summing solutions of the form of Eq. as follows: @u = @ 2 u (2. or mixed Dirichlet/Neumann.27) @t @x2 where is a positive real constant. with u representing the temperature. giving u(x t) = M X Dispersion and dissipation resulting from a numerical approximation will cause the shape of the solution to change from that of the original waveform.14 CHAPTER 2. CONSERVATION LAWS AND THE MODEL EQUATIONS where the wavenumbers are often ordered such that 1 2 M . Boundary conditions can be periodic. In the case of Eq.26) 2. 2. Neumann (speci ed @u=@x).at) (2. Dirichlet (speci ed u).27. g(x) # (2.27. m=1 fm (0)ei m(x. the steadystate solution must satisfy @2u = 0 (2. which is one that satis es the governing PDE with the partial derivative in time equal to zero. A simple linear model equation for a di usive process is Therefore. For example.
2 t (2. Let the initial condition be u(x 0) = M X m=1 fm(0) sin m x + h(x) (2.32) @x2 @y2 While Eq. h(x) = ua + (ub .36) Substituting fm(t) into equation 2.37 shows that high wavenumber components (large m) of the solution decay more rapidly than low wavenumber components. Substituting this form into Eq. A solution of the form M X u(x t) = fm(t) sin mx + h(x) (2.4. u( ) = ub. 2. we obtain m=1 u(x t) = M X m=1 m fm (0)e.e.32 is elliptic.4. i. Let the domain be given by 0 x with boundary conditions u(0) = ua. consistent with the physics of di usion. g(x y) = 0 (2.34.34) satis es the initial and boundary conditions. 2.2 Solution in Wave Space We now consider a series solution to Eq. g(x y) (2.27 gives the following ODE for fm : dfm = .35) m m dt and we nd m fm(t) = fm(0)e. 2 f (2. It is clear that the steadystate solution is given by a linear function which satis es the boundary conditions. Eq. The corresponding steady equation is @ 2 u + @ 2 u . Eq. 2.27. 2. 2 t sin m x + h(x) (2.31) @t @x2 @y2 where g(x y) is again a source term. . THE DIFFUSION EQUATION 15 In two dimensions. 2. 2. the di usion equation becomes " # @u = @ 2 u + @ 2 u ..2. ua)x= . and as Laplace's equation for zero g.31 is parabolic.37) The steadystate solution (t ! 1) is simply h(x). The latter is known as the Poisson equation for nonzero g.33) where must be an integer in order to satisfy the boundary conditions.
form a hyperbolic system of partial di erential equations. 2.42) @t + @x With w = X . we obtain =0 (2. and noting that X and X .1.1u See Appendix A for a brief review of some basic relations and de nitions from linear algebra. .39) @t @x @f where f is the ux vector and A = @u is the ux Jacobian matrix.1AX (2. CONSERVATION LAWS AND THE MODEL EQUATIONS The Euler equations.1 are constants. Premultiplying Eq. Many aspects of numerical methods for such systems can be understood by studying a onedimensional constantcoe cient linear system of the form @u + A @u = 0 (2. Such a system is hyperbolic if A is diagonalizable with real eigenvalues.8. Other systems of equations governing convection and wave propagation phenomena.38) @t @x where u = u(x t) is a vector of length m and A is a real m m matrix.41) where is a diagonal matrix containing the eigenvalues of A.5 Linear Hyperbolic Systems CHAPTER 2.1 Thus = X .44) i @t @x 1 @X .43) @t @x When written in this manner.1 . Eq.16 2. and X is the matrix of right eigenvectors. The entries in the ux Jacobian are @f aij = @ui (2.2.1u z } { @ X . this equation can also be written in the form @u + @f = 0 (2. this can be rewritten as @w + @w = 0 (2. are also of hyperbolic type. 2. such as the Maxwell equations describing the propagation of electromagnetic waves. the equations have been decoupled into m scalar equations of the form @wi + @wi = 0 (2.38 by X .40) j The ux Jacobian for the Euler equations is derived in Section 2.1u.1AX X . For conservation laws. postmultiplying A by the product XX .
2. in a supersonic ow. corresponding to the fact that sound waves can travel upstream in a subsonic ow. of course. Therefore. Each characteristic variable satis es the linear convection equation with the speed given by the corresponding eigenvalue of A. The characteristic variables each satisfy the linear convection equation with the wave speed given by the corresponding eigenvalue. where u is the local uid velocity. which corresponds to in ow for these variables. characteristic variables associated with positive eigenvalues can be speci ed at the left boundary. c are associated with sound waves. are q u u + c. While other boundary condition treatments are possible. while u + c and u . where juj > c. The eigenvalues of the ux Jacobian matrix.6. That is. 2.1 7 4 5 0 p u . The onedimensional Euler equations can also be diagonalized. In a subsonic ow. or wave speeds.38 has a solution given by the superposition of waves which can travel in either the positive or negative directions and at varying speeds. we must ensure that our numerical methods are appropriate for wave speeds of arbitrary sign and possibly widely varying magnitudes. Therefore.2. Show that the 1D Euler equations can be written in terms of the primitive variables R = u p]T as follows: @R + M @R = 0 @t @x where 2 3 u 0 M = 6 0 u . where juj < c.6 Problems 1. wave speeds of both positive and negative sign are present. leading to three equations in the form of the linear convection equation. we see that a hyperbolic system in the form of Eq. The signs of the eigenvalues of the matrix A are also important in determining suitable boundary conditions. although they remain nonlinear. Characteristic variables associated with negative eigenvalues can be speci ed at the right boundary. they must be consistent with this approach. While the scalar linear convection equation is clearly an excellent model equation for hyperbolic systems. and c = p= is the local speed of sound. The speed u is associated with convection of the uid. c. all of the wave speeds have the same sign. the boundary conditions can be speci ed accordingly. which is the inow boundary for these variables. Based on the above. PROBLEMS 17 The elements of w are known as characteristic variables. and u .
(Hint: use M = 2 with 1 = 1 and 2 = . in the form @U + A @U = 0 @t @x where U = @u=@x @u=@t]T . (Hint: make use of the matrix S = @Q=@R. 3.) 5. 2. CONSERVATION LAWS AND THE MODEL EQUATIONS Assume an ideal gas. Next consider a solution with boundary conditions ua = ub = 0. in the form of Eq. 4.25.4. Plot the initial condition on the domain 0 x . Write the classical wave equation @ 2 u=@t2 = c2 @ 2 u=@x2 as a rstorder system. The CauchyRiemann equations are formed from the coupling of the steady compressible continuity (conservation of mass) equation @ u+@ v =0 @x @y and the vorticity de nition @v ! = . 2. 9.18 CHAPTER 2. Plot the rst three basis functions used in constructing the exact solution to the di usion equation in Section 2. that is. 2.2. 2. Find the eigenvalues and eigenvectors of A.) Next consider the same initial condition de ned only at x = 2 j=4.2. j = 0 1 2 3.33 with fm(0) = 1 for 1 m 3. p = ( . Show that the two matrices M and A derived in questions 1 and 3. 6. are related by a similarity transform. 8. and initial conditions from Eq. Plot the solution at t = 1 with = 1. nd the necessary values of fm (0). Write the 2D di usion equation.1. Eq. 2. u2=2). respectively. i. write it in the form of Eq..e. 7. Find its eigenvalues and compare with those obtained in question 2.31. Given the initial condition u(x 0) = sin x de ned on 0 x 2 . 1. fm (0) = 0 for m > 3. Find the values of fm(0) required to reproduce the initial condition at these discrete points using M = 4 with m = m . @x + @u = 0 @y . 1)(e . Derive the ux Jacobian matrix A = @E=@Q for the 1D Euler equations resulting from the conservative variable formulation (Eq.5). Find the eigenvalues and eigenvectors of the matrix M derived in question 1. 2.
(c) Determine the conditions (in terms of and u) under which the system is hyperbolic. i.. has real eigenvalues. One approach to solving these equations is to add a timedependent term and nd the steady solution of the following equation: @q + @f + @g = 0 @t @x @y (a) Find the ux Jacobians of f and g with respect to q.v u g = .) . 1 u2 + v2 . we have @f (q) + @g(q) = 0 @x @y where v q= u f = . . 1 2 . (d) Are the above uxes homogeneous? (See Appendix C. (b) Determine the eigenvalues of the ux Jacobians.2.e. PROBLEMS 19 where ! = 0 for irrotational ow. Combining the two PDE's.1 1 Note that the variables have been nondimensionalized.6.u v . For isentropic and homenthalpic ow. the system is closed by the relation = 1 .
20 CHAPTER 2. CONSERVATION LAWS AND THE MODEL EQUATIONS .
¡ ¢ .
¦ H B d 4 Y H h 5 D h H h H H 2 2 5B D T D 5 2 h 2 T Y 7 d H D H F D H 2 n H h 5 D Y R Y DB 7 d H H d R ` n YB F D q 5 2 5B XQW6IkGÆQI6QIÅ8EGVG¢I¬I¢9W6GeIWII6IkGIwGE9W6sI8U EIG¢Å8EGD r 4 H Y 2B 2 i ¥ H d R `B ¤ 2B 2 @ 5 F Y YB H 4 T Y h 2 T H 7B D F D 5 ` 2B by 5 b 2B F Y H 7 D Y Hy 7B Y H F QWIEej pWI8EvXCcIWvXuQVUWeIV9XGW¢ IXx88oXQ96XI9XIut ¶ Ã ¹ » ´ » Ã 9QavÄÂ ± Á ¶ ± À ± ¿ ¹ ½ ¼ ± » ¹ ¶ ¹ ej)6¾hfv6egº¸ · ¶ ´ ² ± ³ ² ewµ± ° ¯ ® c 2 5B D T 4By T y T 4B D 4 T d 8EGVQXEII¢VgXW6¢WI ` 2B Y 5y d 5 H bB D 4B d D Y H d iy R h 2 R ` 2B H D R 5 F DB IX8X988EW6EWGfQ¬EIIpIrIEQ9p8IWEC@ QGfx«VW6Gf¢ªQ6¢UXISQ©¢Iv@ 7 H DY i Y 2 TB Y H Dd T h H 4 T YB R P H 2 T H Y R H YB Y iy T 2 T i 4 T d R 4 4 T ` 2B @ 5yy 5q H F D q 5 F 4 R 7 2B t 2 5 Y T H d YB F D d 5 ¤ Y 2 5B D T R P H h H 7 d 5q Y 2 T d XWx¢U¢6VGI6Q¢IECcXX¢9IW¢I AX§8WVgWwXpG8UeQp8XWVUSQAQ9W8fI¢WGD H F D q 5 Y D 2 HB 4 ¨ H 5 4 Hy TB d T b 5 D 2B h H d 5 Y T YB 2 5B D TB d T b 4B d D H 7 5 H ` H F D yy T 7 d 5qB 2 R YB } pGpVvGo6XQ96XIVWV¢CGoEwQW¢WI¢§E8XWV¢VXWW698Q¢vIW¢XV¢W8IIXhxp3 ¢l¢£xlIu§8EWp8XvuEwCcI¢sxp3V¢l¢£xl#QEX¢Q3VuuXE¢VQWoXB yx x ¢ { H F t ¦ ¥ H d R `B ¤ 2B 2 @ 5 F Y Y T t } yx x ¢ { h Hyy T 4r 5 Y T 2B Yy T b d H D 2 h H 4 T YB R P H F DB @ F Y H 7 2 TB Y H D d T 7 d 5qB 2 R T 5 D h H 7 d 5q Y 2 T d D H 2 T 4 } y 2 Q6¢UXISQGCW6#¢QW¢¡8EpIGAQ9W8fI¢GW9rVgxp3v¢WcxUeXB Y H D T 2B h d 5 5 4 d T H 2ByB b d R 4 y T d H 2 H ` 2B h H Y Y H d H Y H F Y H 7 q 5 Y H Y Y Ty 4 y T d H 2 H ` H DB R P D T F D D 4 T 6GVIXI86¢QIXEXxI6¢G6IQ¢EeQWQIQIQ9vVf6WWVQVGQpQ8uGISCVUW¢q H F D H B Y T F 7 H H 7 H D Y i Y 2 TB Y H D d T T 2B h H D 2 H Y H d YB @ 5y H y TB d H D T 7 H F D q 5 yy pG QEW¢II#6 wQWxe¢XW6GVeXAQWS6W6WIEsEQ¢QGldIWj¢vEi d H D T F 2B h H Y Y R 4 YB h H yyB @ Y 2 5B D T 7B 5 d T H 4 2 H d H gB h r H DB 2 n H Y H F If6GpVUX6WIQXIuhXECI8EGV#XgWpI¢uQpQW6hI3GIUuWQpGD ` 2B iy T d 5q Y HB ` H D T d D d H D T F 2B h H D T H d D H d T Y H bB D T bB d H h H 7B D Y H bB D T bB d H h y TB D T pXxEIp¢u86X86GVWGxCQG¢UXe6GVQWWW¢u68GVVEQIf#EGp6Q¢EGl¢W6Ih¢XGVIWY 2 5 YB d H D T F YB F D 2B YB Y T F 7 H d R Y H bB D T bB d H h H 7B D d 5 Y H bB D T bB d H h y TB D T Y 5 D d H F DB 8XQG¢UEIGXfXWVUI9QpQ68GVVEQIw9XGC8fQ8GVVWQpm¢XGVIWdWf6IGQH h HBy T H 2 T 4 H Y H F t h H D 2 H Y H d YB Y H bB D T bB d H h y TB D d T 5 D Y 2 5B D T 7B 5 d T H 4 2 H d H gB 6XEII¢#sr¢gQIuQWoQW6WIuEwQ¢EWVVEW6Ij¢XGVUWI8XWV#XgWIIVQpQW6hIh r H DB 2 n q 5 D H 4 2 5 4 H F D t d H D T F 4 YB F D 2 1 H d H F h H B Y T F 7 H F 4 T 5 d T H F D YB YB F t h H by 5 GIUm¢p6QI¢QwIGUQ6GpVUfEIG3VWQpQ6X¢UI9Q¢oWpI¢aIGuXCEIu68E¢WY H 2 T 4 F 4B F @ Y 2 5B D T R P H H 4 2 H d H gB h q 5 D H Y T Y H 4 R h 5 d F 4B F @ x } { y { ~ } {yx rVg)EICwp8XWVUSQ9QI6W6hI¢w6wQQpIIrEIC9QUdxxG¢93dzkT 5 D ` 2B h T Hy t D H 2 h H D T 7B 5 d T H d T Y H bB D T bB d H h H 7B D H F t Y 2 5B D T R P H y TB D 2 H d H gB h i d T 2B WwIEI¢QXpQxI#QGldEpgII¢W¢vQ¢EGl¢W6If9XWfIuj6I8EGVUS6VGo6WQsEp¢UEIh r d 5 q 5 7 H D Y i Y T ` 2B 4 R h 5 d i H d H F D t D Y d n Y H bB D T bB d H h y TB D T Y H F D H D T 7B 5 d T 2 T 4 H 2 f8¢pQGfxdepXQpIIo6WQpGhQWIQ¢EGl¢W6ImVGlUW pGuGV#XgWIp¢k¢gwI85 t8Q8GVIWQWEfjQI8EGVIx6QpQW6hIwpXGIW6WfIG¢E8eQIWdI¢9WI¢QI¢GEIwXWWQpGD iy H bB D T 2 d H Dy i Y 2 5B D T R P H H 4 2 H d H gB h ` 2B Dy R Y H d H F D H by 5 Y 2 H F D h 2 T iy Y R 5 H 2 T Dy R 7B Y H Y H F H D T 7B 5 d T 2 T 4 H 2 H 7B D h 2 T H 4 T Y F D 5 5 D D 4 H Y H d F DB @ Y H bB D T bB d H h y TB D d T 2B T D 2 5 GV#XgWIp¢¢gIV#EG I¢C6¢UG¢GQWQ GCvQ¢EWVVEW6Ix¢W¢UwX¢GS¢Q4 Y 2 5B D T R P H y H h 5 7 d R 5 t @ 5 q hB R q i h T H D Y 2 R ` 2B 2 d H b 5 ` Y 2 5B D T R P H H F D F DB @ 2 5 7 7 5 4 2 p8XWVUSQQIx9wI8vusIrXIIpVgGfIIeIEIW68c8aI8XWVUSQ#IGGECA89#86)31 0 £ ' ) $ % ( " #! ' $ & & % © ¨ ¦ §¥ ¤ £ .
D 2 H 2 5 H 2 QoQI8¢T ¢WXI¢IXWXp9Wf6WQpGoQ¢U 7 5 dq DB F YB R ` 2B D YB h 5 D Y H Y H F D 2 H d T F DB @ h H Y 5y 4 2 H iyy T R Y R YB h 2 T D B d 4 Y d H R Y T Y i T @y T YB H 7B D t d H F D H ` 5 D h H Y R 2 H F @ D R WECÅQ8X6IQEE¢UWp X¬I¢9XW6WQIWxg¢E#EGjQ6IG88W¬6WIÆQpCpI t D B d 4 Y R Y T Y T d T H T Y H 4B h 2B H 4 T Y h 2 T H 7B D H F D F D 5 t H 2 5y T h H Y R 2 H F @ D T F D H 4B D 5 QpXQIIWC¢¢Qsp¢uQ6XIpXwQ¢UI¢9XWupGeG¢pVI8E¢6WIQICulUG6XGVfò ä ç ñ õ ø Þà ß î þ ø ßæ ¡p«¬mÝ ä ç ß î þ ø ßæ ¾s ¬vÝ ä ç ñ õ ø Þæ ¡p«mÝ ç ¥ 8¥ pæ ü ¢í ¡£Ý ¤ý ü ¥Icï ü í Ý ä ç ß îû þ ø ù æ ¡xÿsXÆê mÝ ý ü ï Ý ä ç ñ û õ ø ìù æ IgUðëx©9úmÝ Û Y @ 5yy 5q Y T YB ` 2B H h 2B D B d 4 Y d H R Y h 2 T D B d 4 Y R Y d 5q 2 5B D 2 H b 2 5 4 H F QCcXE8u¢CXIEIIEpXQWQIWI¢fpEW6WIIu8¢XGo68o8QwIut Y R 5B b 5 ` 2B 2 T H 7 H F D H ÷ T 7 hy R 5 F 6I8Eo8pXI¢Q#uIW¢¢#XI8pWY DxWo8Qu¢QECp¢WW6QQp96#EG#8#EºXpCe6W8WpI#6IGV#8©QWp¢s8£vì H D 2 5 4 y T 4 5y t ç i d T Y Y H 4 H 2 Y H 7B D H 7 5 Y YB F 4B F @æ Y H Y 5 d R d H F D 5 d 5q h H Y R H d T ö à õ à êà 2 H F i T @ d TyB 7B Y T 2B â h 2 T á d 5q h H Y R H d T ö h 2 T õ d H D T ô D R 5 F ` R 5 d F D Þ î ä ñ D T F Qpª8gv¢XX9XWX#I¢¢QIuW¢jI¢6GVvgp¢I8I8IGóeVIGD HD 5 ò W¢f pwWp8XvXeCcI¢sXdpXQVUWaIGu p¢AEdIEQ¢UwIWuXk W6IC@ ¥ H d R `B ¤ 2B 2 @ 5 F Y Y T t Þ 2B ` 2B 4 T Y H F D Þ î h 2 T ß 2B ` 2B 4 T Y H F D YB ß î H d H F ç pæ ¥ ç ¦ ¥ cEpæ ñ ì ä Þ î ì ä ï ª ða3Þ ß î ê ä í S«uxß ä ä ¡Þ ß hB d ` h H 4 T YB R P H 2 T 2 5 2 H F t ì è YB X86Q¢IWXpx6¢¢Qpujvs¢é¢wEÞ d 5q D T F D h 2 T t ê è YB ß d 5q H h 2B F Y H 7 H F t h H 7 R Y Y T YB d H F D 5 H F D 2 5 H 4 2 H h 2 H H ¢uVUGI¢ëUc¢é¢wXC¢eIpXQ9apu6#IW¢CEf6IGVwpG8 QpQII6s6Ih t h H D 5 2 H h YB Hy TB d T b H 2 5 iy 2 5 2 H F ç Þ à ßæ Ý ä Ý H Y T 4 ã r ¦ H F D iyB d T 7B d d H hB Y 2 5 4 H hQGVIQIXwXI¢XW¢Vwp8I8kQpªuWmåeW¢QÏÔ6eIWdXWVdEWI6IXI8Qv@ Y d H D T F 4 y T d H b H Y D Y d n H F D d 5 ¤ â à á à ß h 2 T t Þ Y Hy TB d T b D 2 H h 2 H H h 2B H F D q 5 Y 2 5B D 4 2 Rq H d W6GpVUh¢WQ¢6WUfIWj8UjpGdIVpjQEI¢XW¢VoQpIQQIpXpG¢vI8EG6pI¢T t Hy 7 T H d 5q t Ý t Y Hy TB d T b D 2 H h 2 H H h H F D ty T d H 2 H ` 2 1 Y 2 5B D T 7B 5 d T H 4 2 H d H gB h H DB 2 ¢Xp#¢p8QQEI¢XW¢VCoQII6s6IIWVGQpQ836I8XWV#XgWIIVuQI6WQsEpGIUn Y H 4 T Y y T 2 5B D T D R 7 5 4 h 2 T y T 4B Y i F Q6Q¢IWhVU8XWVI#¢QI¢¢QXWxojÜ Û ¦ ¥ H d R `B 8puWI¢Xv¤ Ú Ó × Ò Ë É ÙÒ Ø × Í Ê Ê É Ì Ç Ó Ì Í Ì Ñ ÑÒ ÖÕ Ì Ë Ò ÓÒ Ñ Ð Î Í Ì Ë Ê É È u§fj«wÔzmQÔv«SÏmÇ y x 8 .
7 d 5qB 2 R YB ` 2B 4 T Y H F D t ß î d 5 Þ î 2 5 D B d 4 Y R Y 5 2 YB H d H F D 2 H F 8EpIuXIEQ¢UuIGU C8 8fpXQIIWIuEaQIW6Iª ç 9 ¥ ¢%pæ 4D ¢W6H à ï Ý 6 4 ü ï Ý ä ï Ý UÅ85cUªaëî à í ß 6 4 üí ß ä í ß sxr'5xªCx î à ïÞ 6 4 ü ï I3r75c3Þ ä ïÞ a3 î D T F D F 4 R Y h H 2 n H h YB F 4B F @ t î y 5 7 i VUWIQpU6IuEEICIe8 xY H F D YB 2 5B D H 4 H H 2 5 H F t ` 2B 2 T H 7 H YB 4 H d Y DB H 2B 7 d H D H h 5 D h H Y R H d T Y i T @ d H F D pGXk8EGp6Qp8IuuIIEI¢g9WEQ6WIWE pX9WQW6IWk6WI¢wxcQpG S( y 5 7 i Y H F D 2B h HB d d T 4 D 5 2 YB 2 5B D T 7B 5 d T H F D q 5 ç d H h d 5 h 2 Tæ H d R D T 2 H YB 4 H d H F D D R 8swfuIGE6XW¢QC¢IuXd8XWVdEcII¢upG§¢6IW8I¢9WIWVUuWEQ6WIuIWupI ç ¥ 8 pæ 0 ¥21¥ ( à 321 ( 0 t Hy 7 T H d 5q t D T F D F 4 R Y H bB D T bB d H h T 5 D 2 5B D T 7B 5 d T H 4 2 H d H gB h ¢Xp#¢pu8pglUGIWu8GVVWQp9Wd8XWV#XgWIIVw6IQQsEIT D 2 H Y H d H d 5 D h H Y R YB ( y 5 7 i Y H F D D T F D 2 T H 7 H @ YB F D i & H R PB 2 R D 5 2 YB DB ç 2 5B D H 4 H H 2 oQQWpQWGaQWp§Eu)¢s xfvIWmlUGu¢g9vXIWv'VISXIpv¢p§E§Eë8XWpQ6I85 F DB @æ D R i F 5 Y 5yB F H 7 T Y H F D Y @ 5yy 5q Y 2 5B D T 7B 5 d T H 4 2 H d H gB h d 5q 2 5B D T D 5 2 H F WECpI Sp8W8EXIpf9¢GIWCsEX8p8XWVdEcII¢QI6W6hI88XWV¢pIut ß 5 D D 4 H Y H d F DB @ Ý q 5 H bB D T bB d H h D Y d n H F D D 2 H Y H d H d 5 D h H Y R H D 5 2 yyB 9WQW6WGECvV8EWVVEQIUIGfoQW6WI6WGQIs¢phXEC@ ISut Ý Y R F Y H bB D T bB d H h Y Y H d H 5 D h H Y R d H b H 2 H d T Y Hy TB d T b D 2 H h 2 H H h 2 5 Y D B d 4 Y R Y D H D YB F D 2 QQ¢EWVVEW6IuW6WIdGQIQ¢6IaW¢fQEI¢XW¢VoQIpQQI¢GpEWQIIxGCXpG31 ç $ ¥ ¢%pæ Þ Ý " " ä #!Ý H Y R H @ ã q 5 i h R D Y H F D 2B H bB D T bB d H h H 7B D i d T 2B h d 5 H F D d 5 IuuY u¢ppGuIWeX8GVVWQp#EG#¢UEIW¢wpG8U¤ ç ¥ o pæ 4D ¢W6H à ß Ý ä eÝ ¥ à Þ Ý ä eÝ à ß Ý ä eÝ iy T H ` 2 T F 4 d H D 2B H Y R H @ H 7B D d 5 H 4 T Y 2B Y H bB D T bB d H I¢Q8I¢UQWSEaIw9XW8wQ¢UeXCQ¢EWVVEW6Ih y TB D d T d 5q Y R F t Y 2 5B D 2 H b 2 5 4 y T R Y R H F D 5 D ` 2B h d 5 4 4 ¢W¢I8#ISu©QI8EGoQ¢o8QC¢UIIGeWAIEIW¢QQVT QWQIWV#68GVVEQã h H Y Y H d H H d T Y H bB D T bB d H D 2 H 7 H ` 2 T d d T hB d ` H 7B Dr H 4 T goQ9Q¢I¢G¢X89XG6¢U Û pI8Ev¤ ¥ H d R `B © © ¥ Ó × Ò Ë É Ë × u§IawÓ Ì Ç Ó Ì Í Ì Ñ Ñ Ò ÖÕ Ì Ë Ò Ó Ò Ñ Ö Ó É Ú Ì È Ú Ì Ù Ð ¦Ð fvÔmQfÔvvº8§xÎ n1 n+1 n t j2 ∆t j1 j ∆x j+1 j+2 Grid or Node Points x © ¨ .
ç pzæ ¦ ¥ ç zdø ß îæ 4 í Ý 6 4 üí AxÅ5Ý ß î ä í W ß V Ý 7 d 5q H F D 2B h H D 2 H Y H d Dy R Y H d i d T 7 7 R Y H F 8IGeXQGo6WQICEpWQf¢d9IwpGD Y H H Y 2 H Dq 5 H 2 5 h 2 T t 2 5B D T 7B 5 d T H 4 2 H d H gB h h H d H D 2 H 4 D 2B 5 r H H d F D H F D Y T 5 D h H d d Hq H d Y Q6W9QW¢fp8#p¢o8EGV#XgWpI¢fQI6WQsEpeQW6GoQ6oX83QWpGCIG¢aG#6WQ6XB d H D D Ty H F t H bB D T bB d H h D Y d n T 5 D 2 5B D T 7B 5 d T d H h d 5 r h 2 5 4 H Y T YB ç ß QGVpu¢¢EGl¢W6IuUwW#8EGV#XgWIp¢CQI83xI8Q6WwXvs î IÔAxk£5xæ æ rç4 í Ý 6 4 üí Ý u D T F D Y @ 5 F Y ¦ ¦ ¥ P t iy d TyB 7B lUGjCcIWu¢EpSëj8W¢XX9X y í xWw8XWV#XgWIIVQI83fWIX#V6935xæ w u t 5 D 2 5 B D T 7 B 5 d T d H h d 5 r D Y d n T Y B ß î r ç í Ý 6 4 ü í Ý 2 5B Y Y H d H H F D D T F D H H Y H @ 8EWW6WIuIGmVUG6WuvVt ¡ Ep8j8ëvxIG#pGë¢v6IW¢uIGv688jQvcsjfQvcXy ¦ ¥ P 7 5 d ¤ h 5 F D H 7 H F D q 5 d H h d 5 H F D Y H bB ` d H @ 5 D Y H @ 5 H F D 5 D ` 2B 4 T Y hB d ` H F D ` 2B 2B T D 2 5 4 7 d H D d 5 d d H H F t d H h d 5 2B T D d H 4 T 5 D ` 2B 4 T Y hB d pGWIX6¢UrX8dpGIXIE¢o86¾QWw8W6Iu6QI8rE¢fQ6WIX6¢UrX8` H F D 2 5 Y h 2 H H h d 5 d d H H F D D T F D d T Hy 4 5 Yy T YB D1 2 `B Y y T R P H H F D q 5 H hB Y D F `B d H F D 2 pG¬8#pIQQI8W6pGVUW ¢gEQkWE¢X9£Å¢XWC¢Ix6IG¢IXo8XIG¬85 Y d T H T F 4B F @ d 5 d d H q 5 D 2 R 5 7 T 2B T D d H 4 T F DB @ H bB D T bB d H h D Y d n T D 2 H Y H d H d 2 5B D T R P H H F ¢gI¢dXICj8W6¢oI89¢#X¢G6QudWECu8GVVEQIfWUuoQQIQ#8EGVIx6fpGD q 5 H hB Y Dq Hy H F D 2 5 Y 7 d H D H D H d 4 YB h H F D D T F D d T Hy 4 YB DB t d H 2 2 T 7 YB F D 2B h H Y Y H d H 2 H F ¢uIEWC6EapG8C#QGW6W6WEIapGCVUW¢gEQCXCUQQpI¢#EIGeEeQWQI6Iª ç ¦ ¦ ¥ s8pzæ xx8í W ß ¡ V ¦ V ß î Ý ¦ ø í W ß V î ß Ý ¦ ä í W ß ß î V 6 Ý 4 í Ý 6 4 üí ¥x)5WxÝ 2 5B D T R P H ` 2B @ 5yy 5 8XWVUSQwIXCcXE8q H F pGD 8ÆWeEIQApGªI8W¢QWðI¢«9p8¢8GVWQ9ISkIGD X9W6GeIGðI¢Uv 7 d 5q 5 D Dy R Y H d H F D R 5 d ` H d h 2 T ê D R 5 T d 5 D T d H 7 R 2 H F 2B Y 7 d H D H F D h 2 T s î pf¥xÅ75xA8EGV#XgWIp¢w6IQQsEIwQVUWuIW6IXI8QfxXVduXe î ç ß æ r ç 4 í Ý 6 4 ü í Ýæ 2 5B D T 7B 5 d T H 4 2 H d H gB h H 4 T Y H F D d H hB Y 2 5 4 D H ò yy T 7 Y YB ß u y w u t d 5 q 2 5 B D T 7 B 5 d T H y T 2 5 Y T H d T Y B ß î r ç í Ý 6 4 ü í Ý æ 2 5 B Y Y H d H H F D Y R F Y T ` 2 5y Y VI8Ev¢T í xv18w¢XGldEpgII¢vXIVU8WVgWfvE# ¢sQxaq5Wx¢XWQIIWvISut ç ¡ pzæ ¦ ¥ Y Y Y aa`ø í W ß V ç ß î ¡s æ Ý ¦ ø í Wpß ß î V iä í Ý 6 4 üí x)85Ý Ý @ 5 cfò 2B T D 5 5 D ß î i H hB bB h h 2 T í Ý D 4 T d D R X¢Gp89GeSpXxEII¢¢GWpIWY ç h ¥ 8dpæ d T H 2B ¢gpXXy í W í í W b ï ß V s æ 8ì saagø W ß V sæ ø ß ¡s æ xªj5xÝ Y Y Y aa`ø ø Y Y Y V ç ß î ø í Ý ä 4 üí ç ß î ç ß î ¦ ¦ Ý Ý ï Ý ï Û 4 ü í Ý d 5q 2 5B Y 2 T H Y HB d H Y d 5y i T t H F D d H hB Y 2 5 4 t Hy 7 T H d 5 5xu8¢XWp¢UQEWQu8XxgmwpGuQIEWI¢QIVXI9¢pu8U¤ Y Y Y aaSà ¥8f#8fIWxI¢xm¢I8EGVIXI ¢Q4 e à ¦ e ä õ d 5q ý ü í Ý h 2 T í Ý q 5 Y 2 5B D T 2B 7 5 7 5 dq h H 7 d 5q H 2 T 4 H bB D T bB d H h y TB D d T 2 H bB ` T 5 D Y 2 5B D T 7B 5 d T H 4 2 H d H gB h y T 4 5 8WuQ9W8jaVg8GVVEQI¢¢W¢UwQ¢E8CGI¢XGldEpgII¢QpQW6hIo¢Qô í W í í W b ï ß V s pæ 7ì caa`ø W ß V s pzæ ø Xß ¡s pzæ vðCIWxÝ ç8dpmaa`ø c ¥ æ Y Y Y ç ß î õ ø Y Y Y ç ß î õ V ç ß î õ ø í Ý ä ý üí Ý ï ¦ Ý ¦ Ý ï Y H bB ` ß D R 5 T 7 d H D d H D D Ty H F D ` 2B h 2 T ý ü í Ý ä ç ß î Q¢E8up8¢µW6G6GVuIW9IEII¢IpvUsIWxÿs põ ø ß î êæ # xmÝ sð#mAp¢) Sê p¥ pG pQSjTXkQ¢E8A8EGoQ¢S¢Qe8EGVG¢I2 ä ç ß î õ ø ßæ Ý h 2 T ß î ä ß t ¥ 5 D ¦ ¥ Y P 2B 2 H bB ` 2 5B D 2 H b 2 5 4 2 5B D T D 5 H F D ` 2B @ 5yy 5q t 2 H F t h H n Þ F DB @ ç Þ à ßæ Ý d H hB Y 2 5 4 t Hy 7 T H d 5 pGkIXCcXE8uQpuóQp©GECWm#QIEWI86u¢EI#VI#¢ë¤ 8EWI¢IpQEW6WY 2 5B Y 2 T H Y HB d H d 5y i T t Hy 7B Y T q 5 Y 2 T H 7 i h H D T Ry T b H d 5 h H D T d H 2 H ` H 2 T 4 2 5B D T 7B 5 d T H 4 2 H d H gB h ¢XxgvjEI#EWë¢vp¢g9wSaQWVUXV¢m¢ QWVG6IQ8 ¢gu8EGV#XgWpI¢6IQQsEIÆi ² ¶ Ã ¹ » e#6ja´ R ¹ P Ã À C C SQjeIHG ± E ¹ » ´ E ¹ À 9FjwwF± ½ ± Á ´ C wDB @® A Ú Ó × Ò Ë É ÙÒ Ø × Í Ê Ê É Ì Ç Ó Ì Í Ì Ñ ÑÒ ÖÕ Ì Ë Ò ÓÒ Ñ Ð Î Í Ì Ë Ê É È u§fj«wÔzmQÔv«SÏmÇ .
° ¯ § ¢ « ¨ © ¢ « § ¢ ® « ¥ § ¥ ¢ ¬ « ª © ¨ § ¥ ¢ ¦d¦q£¦A¤sF'¦d¥¥dsa1¦£sq7¦d¤£a¡ddvo H F Y H 7 H F D 2B D 2B 5 i d H b H D T $ ¦ ¥ P i 2 H bB ` 2 5B D T 7B 5 d T H 4 2 H d H gB h h H d H D 2 H Q9apGXoX8fQ¢6VqEsëjoQ¢E8e¢XGldEpgII¢aQI6W6hIeQQGo6Q4 H F D HY R h 2 T pGeWI¬IVuÝ 9msÝ ç ¡ mjQp8XWEII¢QÅf¢UII¢sa6EIXX8#EB cfò ä ç æ Ý à ä æ Ý t Y 2 5B DB h 2 5 4 i d T h 2 R 5 D Hy F 4B dB ã H Y 5 7 @ 5 ç ¦ sªø æ r ä ß î F Y H 7 D 2B 5 d R 5 I vhWQ9uoX¢suI¢ë¤ 6 Y ¥ 6 Y 6 ¦ ¦ 6 ¡ ä ê ä eß Y HB d T h 2 R 5 H F 6QEW¢UpI8wpGD ` 2B h Ry 4 H Y D 2B 5 y }x y q 5 d H 7 R 2 H F D 2 T H 7 H @ t F Y H 7 T 2B Y D 2B 5 q 5 d H 7 R 2 H F IEIIEQCGoE8sl3UVf6s pxwIW¢Q#wvëpW6#ECGoX8V6swISaIWD q¢¢gWveQICfVUWwQEG¢ucXQkCcIWuTI¢VGoX8¢UpI8GCWQ97 5 ÷ T H Y H @ 2 H F @ D T F D H 4B D 5 ò @ 5y H 2 @ 5 F Y h 2 T D T Y D 2B 5 i d T h 2 R 5 F DB @ F Y H DSE8p8wpGCQIEWI8h8XWVdEcII¢a9¢GuIWC¢8EGVGS6W6WI6WC8GlGQ8w¢9{ 2B 5 d R 5q H F D d H hB Y 2 5 2 5B D T 7B 5 d T H 7 T Y H F D d 5q 2 5B D T D 2 H Y H d H d d 5 D T d H 5 y x T H bB d H h Y R D Hy @ 5 ò H bB D T bB d H h h 2 5 4 H Y T 5 D 2 5B D T 7B 5 d T H 4 2 H d H gB h D 2B 5 T YB F 4B F ¢EQIuI6Eÿcj¢¢EGl¢W6II8Q6W W#8EGV#XgWpI¢6IQQsEISE8wuXeEIC@ ç $ ¦ ¥ 8qpzæ ç4 í Ý ø í fAxÆvÝ 85Wxæ 6 4 üí Ý ß î ¦ ä íç ( zsÝ ¥ æ 2 5B D Ty H d H F D d H hB Y 2 5 8XWV6WuIGu6IXI8 } r e j  r p y w p t i p r p ng l g r j U¡7mxvu{f o!¡7 o 2 5B D Ty R 7 d 5q h H D T 4B D YB F 5 h8XWVIwW8eQWVgEGfXII¢WY H d 5 7 T Y H dB R P H d 2 5B D T 7 d 5q 2B y T 2 5B DB h h T F 4 R h H 4 d 5q 2 H H d T Y 2 5B DB h 2 5 4 i d T h 2 R 5 W89#QWEISQe8XWVd8pE¢U8EGEpI¢eIxhQQ8p6a¢p8XWEII¢QVUII8 cIF @ 5 d 5 h H 4 2 H d H gB h H d T F Y H 7 H F D 2B Y D 2B 5 d H F D ¢rQ6IQQsEI9WV)W6#9IWAEwWoX8a6IGV5 cIapjXQWaI¢XWQIdQIW¢wQpGE6I2 @ 5 F Y R yy H D Y 2 5B Y Y H d H H Y H F D q 5 d H F DB H t6Q8cQGoE8saIXpII8WI¢#QG9XQ9a#XvoX8W6W6WEII¢Vf8GVVWQpaT d H b H @ 5 Y D 2B 5 ` 2B h 2 R 5 d d R Y q 5 Y 7 d H D 2B F Y H 7 T 2B D 2B 5 H D H d 4 YB h H 2 5 D T H bB D T bB d H h 5 D 2 5B D T 7B 5 d T 2 T H bB ` i H F D H 4 2B Y x } } } } y x y d 5q YB Y T H F D H d T H Y H F W¢XGldEpgII¢d¢¢E896pGf6IXGlÔ¢UhG ¢UlUj8XWVUIWf¢W6Iut ç ¦ ¥ Sgpzæ ç ¥ ¦ ¥ ocpzæ ç z~f¥r75xæ ß îæ ø ç4 í Ý 6 4 üí Ý í W ß ß î ç ß îzæ~øçf4¥íÝ¬øíxÝ 6845üíæ Ý ä V ¦ Ý í W ß V Ý ß î ¦ ä Û Y H bB D T bB d H h h 2 5 4 H Y h 2 T D Y d n H F D d 5q Y 2 5B D T 7B 5 d T H 4 2 H d H gB hr h H d H D 2 H 4 QQ¢EWVVEW6Ip8Q6WI¢fUIG¢I8EGV#XgWIp¢ 6IQQsEI£QQGo6Q4 D 2B 5 r H H d F D H F D H d T Y Ty R 7 d 5q H 4 2 H d H gB hr H DB 2 n q 5 Y Hy 7 T H 2 5 7 7 5 4 D Y 5 7 H F D Y T F d H SE836WIWaIWw¢¢p W86IQQsEI£GIUVf6XI9¢p89#8689apGIVUW6mÜ } r e j  r p y w p t i p r p ng l j ig e h¡7s{zxvusqo2mkshfd ² À Ã » ´ À ± C 9wr ± Á ¶ ± À ± ¿ CejÅ6¹ ½ ¼ ± » ¹ ¶ ¹ hfegº¸ s ® $ Ú Í × Ë É Í Ì Ê fuu× Ì Ç Ó Ì Í Ì Ñ ÑÒ ÖÕ Ì Ë Ò ÓÒ Ñ Ð Ð vÔmQfÔm«xÎ xÎ .
H d T F Y H 7 D 2B 5 r d R 5q T 2 5 Y 2 5B DB h 2 5 W¢W6#uoE8s£fI¢8up8XWEII¢Q4 i d T h 2 R 5 D Hy F 4B dB ã F DB @ H bB D T bB d H h h 2 5 4 H Y h 2 T D Y d n T d 5q Y 2 5B D T 7B 5 d T H 4 2 H d H gB ¢III8u6EIXXGEC¢EGl¢W6I)I86QrI¢wUw8p8XWVdEcII¢#QpQW6hIh ry T d D 2 H 4 D 2B 5 r H H d F D H F D ` 2B D 2 H Y H d H d Y d 5 D T d H 5 B d D T 7 H F t H 7 T Y H F D YB Y H 2By H F D q ¢WGoQ6oX83QWpG IWIEGoQQIQW8WVG6s¢ WWVdIuw¢9¢WIGEQpXXpG¢5 H 2 5 2 t h 5 F D H 7 y T d D 4 H Y T ` 2B D 2 H Y H d H d d 5 D T d H 5 H 4 2 H d H gB h B d D T 7 H F D q 5 H Y T 4 H 7 H d D I¢IUxIG#§¢GW66s#eIEGoQQIQ¢GVWQ8w6IQQsEIWWV#wIWv¢aW¢Qw9QWWxH H F D 2 1 h H nB h 5 7 H B d D T 7 H F D q 5 5 D d pG3QIEIx##GV#uIG¢8W85 ¢GWVsuIW¢gIu¢uVu6IXEwpGVIGGl6EIh 7 5 D D 5 H F D d T H 2 d 5 D T Y H 2By H F D D T F D H D T D 4B i T 7 Y 2 5B DB h 2 5 4 i d T h 2 R 5 t Hy 7 T H d 5 ¤ H 7 T Y H F D iyB d T Y Y H 4 H 2 D 5 2 H d T H 2By 5 D H 2B 7 5 d gdjI¢XGII86VUII8o¢EI9¢pj8UjV#¢WCIG X¢GQQ6I¢Iu¢CIXEuGIEXy ¢Wq h H Y R Y d 5 D T d H 5 D 2B 5 H F D D R t d 5 D T d H 5 H 4 2 H d H gB h D 2B 5 T 2 5 h H Y T YB d 5 D T d H 5 H 4 2 H d H QIW8WVG6s8voX8pGpIxQ8WVG6s8fQI6W6hISE8ae8dQWVUvE8GVWQ86IQQq rqB h B d D T 7 T q 5 H 2By F 4 T d 5 D T d H 5 H 4 2 H d H gB h B d D T 7 T Y H D T d D Y RyyB Hy 7 T H YB F EI WGldv¢aIXE¢½Q¢GVWQ8QI6W6hIWWV#6GVWGpXXE Xp#¢pXIut ¡ ¥ ç ¢ pzæ y A ¼D» e² Ý ¥ ( º² t ø ²Ý ä Y T 9 ¦ ¥ P H DB d @ H d 2 T 4 H ¢ qpëjWEC6W¢Qav@ 6 ¦ ¦ ¦ ç h ¦ ¥ oapzæ ¹ · ·· ¶ !Ý ¹ · ¡ ¡ 6 ¦ ¦ 6 ¦ ³´´ 6 ´µ ß î ¦ ä » h 2 I¢T ç c ¦ ¥ oapzæ ·· ¸¶ !Ý ³´´ ´µ ß î ¦ ä y ¦ t º² à · ·· 4 ¸¶ £Ý Ý ¹ ´µ Ý ´ Ý ³´ ä ² Ý ` 2B 4 R h 5 d D 2 1 7 d 5q y TB 4 H Y i d H b T Y T F B d D T 7 ` 2B Dy R Y H IEQIpGo3W8¢XQ6sQ¢ fVU WWVdIEGIW6Wd H F D D T F D t d H F D d Rq h 2 T t 7 d 5q B d D T 7 r d 5 D 4 H b T 2 pGwVUG§Q6IGfIAp¢§8eGV#3Ô8GQ¢XB QpGwQWppr¢QdvuVUG¢QX6wEwEB 7 H F D Y Y H d H 2 T 4 H @ D T F D d T Hy 4 YB D Ý !ø ç 9 ¦ ¥ 8qpzæ Ý 6 Ý ø ß î r ¢sç ß î r ¢sç ß î r ¢sç ß î r ¢sç Ý Ý 6 ø Ý Ý æ æ ä æ ä ä ä Ý ±£Ý 6 4 ø Ý 4 £Ý 6 ä ä ä ä Ý !µæ ç hÝ ¥ ç hÝ ¥ ç hÝ ¥ 4 hÝ ¥ ç (æ (æ (æ ( æ 7 d 5q H bB D Y H ` ` R Y H d 5 7 H F D 2B Y 2 5B D T R P H H Y H F D ` 2B DB d W8u8GfQ88pWW¢#uIWECI8XWVUSQuQIW IXWEW ç ¦ ¥ ocpzæ ç Ý 3!Æø Ý Ý 6 æ ç ( sÝ ¥ æ ç ( sÝ ¥ æ ç Æø Ý ç Ý ç ÝÆø Ý 6 æ Ý 8£æ 6 4 Ý ¬vÝ Ý ø 4 6 Ý k!æ ç ( sÝ ¥ æ ß î ¦ ß î ¦ ß î ¦ ß î ¦ 4 sÝ ¥ æ ç ( Û Y 2 5B D T R P H d R 5q H F D D T H bB d d 6I8XWVUSQup8IGCVu¢EW¢T Ú Ó × Ò Ë É ÙÒ Ø × Í Ê Ê É Ì Ç Ó Ì Í Ì Ñ ÑÒ ÖÕ Ì Ë Ò ÓÒ Ñ Ð Î Í Ì Ë Ê É È u§fj«wÔzmQÔv«SÏmÇ .
y T 2 5 ` TB h y T d D 2 H z¢U8¢¢IVGGo6Q4 H F D 5 D Y d Hq H d è H 2 5 y T d D 2 H 4 H F D h 2 T d H 7 R 2 2B h h 5 Y i T @y T H d T B d D T 7 h H h 2 T pGdWW66Wc¢é¢ep8v¢WGoQ6 IWI¢Q ISXIfxcEVW¢WWV#eQIIVUdT d¢GoQ9I8¢uIujQ² º c² ë² ¢6WQWppseVgCII¢IIGaI8XVfWo¢fI8Qp8IWEC@ 5q Y D 2 H 7 R ` d T H F t ç à à æ Á Y T h H Y Y H d H H 2 T 4 Y h 2 T H F D ` 2 5y T Y D 2 T D Y 2 5 4 D R 5 F DB B d D T 7 y T 2 5 ` TB hB d D i i DByB B Y Y 5 T YB Y h 2 T q 5 d H 7 R 2 i 2 T F ` R 5 F Dy T B d D T 7 y y GV#¢VU88¢XIXGÅ83XEIXW8f§X§IIVUI¢§Q ISuo¢a8I8pGEVuGV#¢lxc¢x Hy 7B Y T d 5q 2 H bB ` YB 2 5B D T d D Y RyyB H F D h 2 T y T 2 5B D 5 YB D 2 H 7 R ` d T H F D 2 XI9Xw¢Q88E8XWVGWIEXXfIGp¢¢U8EGp8vXS6#p8W¢IG#XB 8¢IW6IC@ q 5 HY R HdH F ç ¥ 8$ pzæ ¹ · ·· ¦ º ·· ·· ¸¶ º Ä º ´´µ ´´ ´´ ³´ ä ç º à à jQÃsÃëÂÛ æ ¤eÁ 2 5B D T D 5 2 H F D H 4 R h 5 d D 2B h 2 T y x { } À T B d D T 7 T F 4 R Y yy T 4 H y T 2 5 ` TB 8EGVG¢IuIG6IIxWGoEp¢w¢99Wx¢e WGldpWE¢gumVU88¢XIh yVGGo6QpG«Ip8W¢ª6W6GpXQW¢ªEICdIp¢UAI8XVeW8pG#¢pQ68Q6E¢¢T T d D 2 H 4 H F D h 2 R 5 d T h H d H D Y Ry 4 H d T F 4B F @ Y h 2 T ` 2 5y T H Y 5 F D d 5q D H 4 H Y 5 d H yy T H d D T F D Y D 2 H 7 Hy H F DB @ B d D T 7 H d T R P Y T YB Y d 5 D T d H 5 H 4 2 H d H gB h B d D T 7 H Y H F D q 5 F 4 T lUGGo6#6XQkWEC WWVd9WVUSWeuXuW¢GVWQ8QI6WQsEpeGV#9WQpG¢r¢ ç ¥ S pzæ ¹ ¦ ¥ ¢r 6 ¦ ¥¢r 6 ¦ ¦ · ·· ¸¶ 6 ¦ ³´´ 6 ´µ ßî ¦ ä ¥ ( i 2 H bB oQ88` YB D F `B d H F D 2 5 2 5B DB h 2 5 4 2 2 T 7 R H ò T d 5q yx {y } ~ } h 2 T Y D 2B 5 Eo8XpG¬8¬8XWEII¢QÆI¢#IQA98el¢98FGoxlW¢lckóp¢9GoX8 d 5B d H D 2B D T H 7 H F 4 Y ` 2B 4 2 H d H gB hry T d D 2 H 4 D 2B 5 r H H d F D T d 5q d 5 D T d H 5 B d D T 7 H F D 2 H F 8XQGoEdV#6IWIX6IQQsEI£¢WGoQ6doX¢s£QIGAd88WVG6s8Æ WWVdIGº6Iut ç ¥ o¥ pzæ Wß V Ý ä ¿¡¾¦ Whß V Ý YB 2 5B DB h 2 5 4 i d T h 2 R 5 H F D H d H F Xe¢XGII86eVUII8uIGuW6IC@ ¥ ç 8 pzæ ç ß ß r6 æ î ¥ 6 W ß V î ¥ Ý Ý ¦ Ý ¦ ä ç ( sÝ ¥ æ D 2B 5 ` 2B @ 5yy 5q H F SE8ÅIXCcXE8kIGD Ûç p p8XW6Qxw6A¢XGII86eVUII8I¢#IQ9C¢8GVWQ85 ¥ 2 5B D 4 H H H æ 2 5B DB h 2 5 4 i d T h 2 R 5 2 2 T 7 R H ò T d 5q d 5 D T d H i ¦ ¥ P 2B H 2By F D d R 5q H F D H 4 Ty H d t Hy 7 T H d H F D d Rq T Y o¬cEfëj`XeIEXðGfI8kIGkQ¢XIQVXI9¢pQIWIÅi ¦ ¦ 6 ¦ ¦ ç ¥ s¦ pzæ ¹ ¦ 6 · ·· ¶ 6 ¦ 6 ´µ ß î ³´´ ¦ ä ¥ ¥à ( ¹ ¦ ¡ ¦ ¦w6 ¡ ¦ · ·· ¶ ¦w6 ¡ ´µ ¦w6 ¡ ³´´ ß î ¦ ä ( 9 Ú Í × Ë É Í Ì Ê fuu× Ì Ç Ó Ì Í Ì Ñ ÑÒ ÖÕ Ì Ë Ò ÓÒ Ñ Ð Ð vÔmQfÔm«xÎ xÎ .
° ¬ « § ª « ÂFÕ¤¥sxd § ® © ª Ö ª ® © ¬ « « § © ¬ « ® © ¢ ¢ « ª § ¬ Ô ¨ ª ª ¬ © ¯ ¬ « Ô § Ñ « ¬ « § ª ¬ Ò Ñ © ÎÐ § ¨ § § Í ¦d¤¥Xv¸£sv1q¦Q¤'qFs¤¤¦ÉS£É£d£ÏFsX£dqÕ¦1kdv¦Ó¦dquk% Ï¥¤p¦« sÌ iDByB T D Y q 5 Y 2 5B D T d H hB Y 2 5 4 2B d H D Ty y Rq H Y R H yyB @ F 4B F @ Y 2 5B D 4 R d D Y 2 5 ¢3EEXI¢Gm¢CI8EGVWQIEWI86eXu6GVX§p6WpsEXCXICuI¢XGIWWI¢Q4 y¢G6IQ8wGEpxVWI pGCGE9W6sQX¢UIp8swIG9IXIpXQpXEQUwpGEGoE8sfI8EW¢Vb T d H 2 H ` H DB R P q 5 H Y R H F D Y DB 7 d H Y HB d T h 2 R 5 H F D ` 2B h Ry 4 2B hy H n H F D 2B Y D 2B 5 Y R 5B d T H F D D T 2 5B D T 7B 5 d T H F D q 5 H d R D T 2 D 4 T H H F D d 5 D T d H 5 H bB D T bB d H h B d D T 7 H F D 2B iqB 4 H pGjVw8XWVdEcII¢CpGU¢uIGVI6¢pupGv8WVG6s¢8GVVEQIa WGlduIWwXwE6QWY 5 D i DByB T H F t Y HB d T h 2 R 5 H F D D T D H 4 H hy H n H F D 2B D 2B 5 i d H b H D T h HBy T 2 H H Y T W#3XXp¢wpuv6QX¢UIp8swIWlp6QX6UIGXCoX¢s68fl6XEII¢QQ¢UF d 5 D T d H 5 D 2B 5 H 7 T Y H F D D T F D d T Hy 4 YB DB h 2 T h H nB 4 H Y iy H R PB 2 R 2 H H H b T F Y 2 5B DB h 2 5 8GVWQ8oE8sw9¢G pGlUGC¢gEQfXfEIVkQUQQWQISXpIeQ6s ¢gUfp8XWEII¢Q4 id T h 2 R 5 H F D t ¥ h 2 T ¢III8apGpph pI¢9 pQSj3xqEsëjDoQ88u8WVG6s8CoE8suIWVUGe8EGD ¥ Y P 2 1 $ ¦ ¥ P i 2 H bB ` d 5 D T d H 5 D 2B 5 H F D 2 T F D 2 5B r T 7 d 5q 2B H d 5 7 i d d T 4 l#W8XW89¢gwh I¢w9 pp6SËo68E¢v8GVWQ8 WWVdfIGVUW6XW¢fò ¥ h 2 T ¥ Y P i 2 H bB ` Y d 5 D T d H 5 B d D T 7 H F D D T F D H 4B D 5 Æ Ê W ß V Ý È ß É î ¥ à Y Y Y aaxà ¡ à ¡ sÝ à ß î ¦ ä ä ä y ¦ t º² ² ² Hd H F W6IC@ Æ ûV¢r 6xà6YaYaxà xà xà ù ¥ Y 6 6 6 Æ V¢r aax8Q8¦ ù û ¥ à Y Y Y à ¦ à ç ¥ oh pzæ y A Çw² Ý sQl² U² eÁ º² t ø ç ¦ à à æ ß î ¦ ä ² Ý ¥ ( h 2 n H @ IUuvpt 8 p¥ h 2 T IV psQSjfE¢¢pXWfo8XapG¢p8XWEII¢Q¢III8IVdI6fwsEWQQIwv£1 ¥ Y P 2B Y T t H hB Y D F `B d H F D 2 5 Y 2 5B DB h 2 5 4 i d T h 2 R 5 2 2 T 7 R H ò H B d 4 Y H d H @ q ç ¥ oc pzæ Æû Ý à Y Y Y à §!Gà ¡ 6aaSà ¡ s!úÝù ß î ¦ ä y A t º² Û Y H hB Y D F `B d h 2 T Dq H 6QIEWCo8EWI¢Xy H F D 2 5 Y 2 5B DB h 2 5 4 i d T h 2 R 5 D Hy F 4B dB ã H F D ` 2B hy 5 F d 5 D 4 H b H F D d 5q Y h 2 T D pG)8wI8EGEpI8QVUII8a6XpXWEeIWIEIX¢I8GQ¢dIGw8wIp¢fY y A ~W6IC@ º² t H d H F ç ¥ 89 pzæ y A ¼w² Ý sgà x8ÔeÁ º² t ø ç ¦ 6 à ¦æ ß î ¦ ä e² Ý ¥ ( i o² Ý ¢¢EGl¢W6II¢QQuIGWVdEcII¢¢Qav@ q 5 H bB D T bB d H h h 2 5 4 H Y H F D H D T 7B 5 d T 2 T 4 H ç ¥ o pzæ ·· ·· Å ¹ Ý ´µ ´ ´ · Ý ´´ Ý ³´´ 4 ·· ¶ £Ý ä ² Ý Y T ` 2B 2 n H ã Y Hy 7 T H Y R 5B b H d d R 5 H By T d H 2 H @ 5 2 2 T 4 H V² Ý IXpU66QXp#¢pp8Xx6ICI86XXVGQpQ8` sp¢Qur Ú Ó × Ò Ë É ÙÒ Ø × Í Ê Ê É Ì Ç Ó Ì Í Ì Ñ ÑÒ ÖÕ Ì Ë Ò ÓÒ Ñ Ð Î Í Ì Ë Ê É È u§fj«wÔzmQÔv«SÏmÇ c .
Ý ¥ ¹ Ý Aº ·· ·· ·· ·· Ý ¸¶ 3 º ç ¥ pzæ ¥ º º 4 xº ´µ ´´ ´´ ´´ Þ¡ ³´ 4 º S 4 ä ç à à Û æ Ü jQ² º s² ² zokÁ i F Y H 7 D 2B 5 r T d 5q 2 H bB ` YB Y HB d D 2 oQ9uoX83s#C8e68E¢uXQXGoQH W8IIS8IWEC@ 7 d 5qB 2 R 2 5 2 F DB d 5 D T d H 5 B d D T 7 y T 2 5 ` TB hB d D 4B h 5B d H y T 4B i D i B d D T 7 y y } T Y T 5 D h H d d Hq H d YB F Y H 8GVWQ8 WWVdV¢U8¢¢IEWWvXIxXQS¢gEIx3¬sGV#h¢QU§¢fGa6WQ6XwWQ97 4B h 5B d H T 2 5 Y 2 5B D T R P H d Ty T 4 Y d 5q Y H 7 H F 4 Y ` 2B 4 2 H d H gB h Y D 2 H Y H d H d D T F D B d D T 7 H F XIxXQw#8I8EGVIx6¢¢QWj8Q9QIuIX6IQQsEIWoQW6WI6WVIG WWVdIut r xs eQ9XpEWQ¢QIXE8uWoX8Co8Xj ä ß î F Y H 7 4B h 5B d H d T H 2By T 2 5 Y D 2B 5 D F `B 6 Y Y aaY ¦ Y c 6 Y 9 6 Y 6 Y $ 6 Y 6 Y ¥ 6 ¦ 6 ¦ ¡ ¡ c 6 9 6 Y Y aaY ä ê ä eß 8EGVQEuIXW¢GjGapXI6QQIvIÚoX¢sCIWlwsIIQH E¢wI8EVW¢GuIXQ IS2 2 5B D T 4 5y ` 2B D d T D Y Y DB ` 2B h H 4 H d D Y R Û D 2B 5 H F D D T Y h 2 DB Y T ` 2 5y Y T Y D d T D Y ` 2B d H 7 R H F D H d H F @ d H D D T 7 iyy T H d 2 Y H 5 h DB t Hy 4 dB 4 T q 5 d H D H 7B d H H F D 2 5 D R 5 hB Ty YB F Y H 7 H F pGW6ICf6GWldEE¢guD ÚWQxIEVXQX6VQW69XQIG8p8E¢XeW6#pGD 2 H F ¥ H d R `B ¤ 2B Y T F Y H 7 d Ty R 4 dB 4 T 2 5 iy H bB D Y H ` ` R Y H d 5 7 d 5 t Y HB d D 2 H h H D T H H d F DB Qpªmp¥ puI8Xm#Ej¢#WQ9¢XIQX6u98 EQ¢EWQ¢8IWW89j8SQ6XWWoQ9QGlgQWWEC@ F Y H 7 d T H 2By T 2 5 h H D 2 H Y H d H d H F DB H 2 T 4 YB F t @ 5y H 2 @ 5 F Y F Y H 7 4B h 5B d H D 2B 5 r D F `B Q9v¢QIXE89QGo6WQIQpGE6¢gvXIumusEQCspWW6#XIxXQvoE8s£o8EQH H F D d H hB Y 2 5 Y H ` 2 T F 4 d 5 D T d H 5 B d D T 7 H F D q pGuQIEWI8eQQ¢I¢Uu8WVG6s8GV#IGv¢5 W¢9IW¢QU ¢wI8XWEIp8Qef¢T 7 d 5q H F D t y y } H d T Y 2 5B DB h 2 5 4 i d r h 2 R 5 H F D q1 h H n H d T Y 2 5B DB h 2 5 4 i d T h 2 R 5 H F D F 4B F @ 2B Y H Y T 4 h H D T d D Y RyyB H b 5 T H F xII8CIGs£jQpuWVI8XWEIp8Qf¢UIp8sIW9EICXQWVgQWVGWIEXEf¢s¢fIut } p tg r j v!¡7 tg Ø eg r p ÙQX!sfd í Ý q 5 D 2 HB 4 ¨ H 5 sxm¢oQX696xQ4 H F D YB D T F D d T Hy 4 DB Y H ÷ T 7 F 4B F @ D 2 H 7 Hy H F Dq n H F D 2B 5 d H T q 5 2 5B DB h h T H F D H D 5 pGwE9ÓlUGw¢gEQ QV¢#ÆEICwo6#6XQÅWIIWrXeQQVr8EGEpI¢eIWdGVfò ç ¦ ¥ sp¥ pzæ à 4 º à à 4 æ Á ä ²Ý ² Ý ç ¡ a¥c¥¡¥à ef5 ( Y XB ¡ ¥ pj7¢CoQE¢VEISQdGV#uIGCISujQ¢GVWQ8 WWVduIWeXCEpWY ¥ P q 5 D 2 Hy T bB R P H B d D T 7 H F D Y R F t d 5 D T d H 5 B d D T 7 H F D 2B DqB F y¢U8¢¢IWwII8W6W8Qw8GlGQ8aoX8dIWAEwIW£¢ê e6I8XWVgxXwI8EW¢V9IGwVT T 2 5 ` TB h T 5 D Y h 2 5 Y H d d 5 4 d 5 D T d H 5 D 2B 5 H F D 2B DqB F Yr i Y 2 5B D T 4 5y Y R 5B d T b H F D D 2 5B D 4 2 Rq H F D q 5 Y H Ry T b H F D 5 D Y D F `B H @ H F D ` 2B bB ` Y D 2 HB 4 ¨ H 5 4 H F D Y D 2 H 7 R ` d T Y DB d 5q Y T 8EGIIIWm¢u6IXV¢pGGuGo8EQvIG pXxE8CWS6X6#6IGuWS6#p8W¢GEu¢¢UF d 5 D T d H 5 B d D T 7 ` 2B h 2 5 Y H d d 5 4 H F t H bB D T bB d H h D Y d n T d 5q d 5 D T d H 5 D 2B 5 H F D H D F `B 8GVWQ8 WWVdwIEII8W6WW¢Qpuv¢¢EWVVEW6IvWIu88WVG6s¢voE8sIWCo8X97 ç ¡ ¥ pzæ ¥ 4 üí Ý4 º ø í Ý ø 4 í Ý4 5Wxfxjxh¥f¡~ø AxÝ ×uhÝ æ í ä íç ( Hy 7 T EI9¢pH d8Uj8EG6pIIW¢6IXV¢uIEW8S8X6I¢#QG#X vSE8QI6WQWfIGvV¢EWVVEW6Ih 5 ¤ 2 5B D 4 2 Rq H F D q 5 Y H Ry T b ` 2B d 5 F `B H 2 q 5 Y 7 d H D 2B ê D 2B 5 H 4 2 H d Hq H d H F D D T H bB D T bB d H H 7 5 Y d 5q 2 H D DB d @ iyy T d H 2 H ` YB d 5 D T d H 5 D 2B 5 98u86GEWC#EE¢G6IQ8CEu8GVWQ8oX8 vQIWQ63se¢XGlQ9XXp¢fQ9GD i 7 H F D 2 H H @D H 2 5B D Ty H d T F YBy T DY H 5 D 2 T D d 5 7B YB DB t Y d 5 D T d H 5 D 2B 5 h 2 T B d D T 7 F D 5 q 5 H Y R Hy T d H hB Y 2 5 4 H ÷ T 7 H @ H 4 2B o¢f8s9X§X§¢Q8GlGQ8§oX8 IVwGV#wW¢sp¢IvEI¢WQIEWI86vV¢#vvQpX h Ú Í × Ë É Í Ì Ê fuu× Ì Ç Ó Ì Í Ì Ñ ÑÒ ÖÕ Ì Ë Ò ÓÒ Ñ Ð Ð vÔmQfÔm«xÎ xÎ .
h 2 T D Y d n T d 5q Y 2 5B D T 7B 5 d T ` 2B 4 2 H d H gB h ry T d D 2 H 4 D 2B 5 r H H d F D h d T h 2 T D Y H F D 2 H F IVuU8p8XWVdEcII¢aIEQI6WQsEp3VGGo6QoX83QWpG#¢UIVfIG#QIª ç ¥ S¢¥ pzæ ¹ · ·· ¦ º º º º ·· ·· ¸¶ ´´µ ´´ ´´ ³´ ä ç º à à 6¥c¥UÂÛ æ Ü xkÁ i 2 H bB oQ88` YB B d D T 7 D 2 Ty R 4 dB 4 y T 2 5 ` TB hB d D T q 5 H Y T 4 y TB 4 H Y H F t DB H b 5 T H 2 5 H F D q 5 D F `B d H F EkGV# o¢XIQX6¢U8¢¢IEWW¢¢g¢6QWeIug8c¢eI¢dIWu¢wo8EWpGD 5 D D 2 H 7 Hy H H 2 5 ç ¥ H d R `B ¤ H H Yæ h H DqB F Y YB B d D T 7 D 2 Ty R 4 dB 4 T q WfoQ9QEQwI8uo¥ pwWI8Ev 6WÅQWEIfEe WGldo¢pQWEQ9¢5 cW¢gfVUWw6XGVfò @ 5 d F 4 T H D T F D H 4B D 5 ç ¥ ¥ o8¥ pzæ ä A æS 4 ¹ 4 ä x±¥ ´´µ 4 ä ³´ · åS±¥ ·· ¸¶ åSÄ¥ 4 ä YB d H h d 5 q 5 B d D T 7 D 2 Ty R 4 dB 4 y T d H 2 H ` D Y 5 7 H F t d H D T E6IW¢v¢dGV#Co¢pQWEQ¢WQI68C¢#uIuj6QGly 7 H F D D R 5 T i T Y 5 D H d 5 7 F 4 R 7 H b T F yyB @ H YB Y iy T 2 T d R 5 2B Hy 5 d y T DB b T i Ty Y H 4B d D T QIW8sVdgWwGu8#dI ¢gUUXECvQEWxEVU¢vI89XCX8U¢xw9gI6QXGV#7 DSVI6WXgo¢GI86W¢aIIVUI8EW¢VIGeI¢X¢wGo6#6XQIW6ICeQ9W8sGV#7 2 Ty R 4 dB D 2 T D Y 2 5 4 H d T Y h 2 T Y R 5B d T b H F D ` 2 5y T Y D 2 H 7 Hy H H F D 2 H F @ h H 7 d 5q t B d D T DSVI6WX6vpGXGV#jXIxXQp¢v6WpIWo¢XQ6sS6Q8c'go¢GI¢Q§¢IWVuGV#7 2 Ty R 4 dB 4 H F D YB B d D T 7 4B h 5B d H T q 5 D H Y R Y y TB 4 H Y T t d H b H @ 5 D 2 T D Y 2 5 4 D 5 2 H d T B d D T 4B h 5B d H T q 5 Yy T 2 5 ` TB h H F D ` 2 5y T Y D 2 H 7 Hy H H F D t Hy 7 T H H F D 2B 2 @ 5 F Y Y T ty T d H 2 H ` 2 XIxXQv¢fE¢U8¢¢IpG#p8X¢uWoQ9QX6 IWh¢EI#VI IWXACcIWV¢WQI68e31 } p tg r j sv£!¡7 F Y H 7 d Ty R 4 dB 4 T 2 5 Y D 2B 5 D F `B W6#¢I6WEQ¢uGoX8Co8Xj j i ã â t rg mÉsáà ß Û ¥ H d R `B p¥ puI8Xm¤ Ú Ó × Ò Ë É ÙÒ Ø × Í Ê Ê É Ì Ç Ó Ì Í Ì Ñ ÑÒ ÖÕ Ì Ë Ò ÓÒ Ñ Ð Î Í Ì Ë Ê É È u§fj«wÔzmQÔv«SÏmÇ 7 6 8 1 ¡ ¥ 5 4 2 3 .
h H D T Ry T b QWVUE¢V6H H 5 D YB H bB D T bB d H h H F D F 4B F @ D T D 2B 5 H F D D R 5 «WeXA8EWVVEQIAIGðXICVoX8AIGp8¢T 6WQWoQQÿ8XW6IpAIWw¢QpX¢Vb h H d H D 2 H 4 2 5B D 4 2 Rq H F D q 5 Y H Ry T H H d F D ` 2B Y R H bB D T bB d H 6WIWIEWI8EWVVEQIh I¢QQru¢Æ8EGV#XgWIp¢T ¢#8XI¢G#8ExcmrWS6W6WI6Wd h 2 5 4 H Y T q 5 2 5B D T 7B 5 d 2 T d 5q Hy T D d 5y i T t T Y D 2 H Y H d H F 4B F @ ¦ ¥ Hy T t d H hB Y 2 5 4 t Hy 7 T H 2 T Y i c ¥ P i 2 H bB ` 2 5B Y 2 T H H F D q 5 H Y XpCpEI¢md6IXI8Q¢Xp#¢pº¢fªpdëjzoº688º8EWI¢IpIW¢IR H bB Y 2 H D H Y H ÷ T 7 F 4 B F @ t H y T D d 5 y i T 8WI6GxQV¢#XICu¸¢Xp¢8Xxgmt 6IGfI8QwGXp¢vEIGfW¢QaW9g#¢f8q T D 4 R d D Y 2 5 4 5 D YB D R 5 YB F D i d d T 4 5 D i T @ h d T @ d 5 D F `B T d D Y h 2 T Hy 7B S¢X¢WGªIVXp#EWY QQpW85 o¢a¢#W8WVG6s¢3oE8sQI6WQsEp3WEII¬pXGIWWI¢Q4 i d H hd i 2 T q 5 Y d 5 D T d H 5r D 2B 5 H 4 2 H d H gB hr H DB 2 n ` 2B D 4 R d D Y 2 5 d 5q Y 2 T H 7 T Y H hB b 5 d D 2B 5 h H n T D R 5 T Y 2 5B D 4 2 Rq q 5 2 5B Y 2 T H Y HB d H Y d 5y i T t H F ¢wI¢Q#6IXxcWI SE8Æ6pp8¢I¢XGIIC¢¬8EWI¢IpQEWQ 8EgmeIut } pã ø  ö r eã ÷  sËgU£õvËö À ± · ¼ vÀ ß sõ ó ¶ G ñ Ã ² ô¾òÓð± R ± ³ Á B ï ¶ ¹ Á ¶ ± À ± ¿¹ ½ ï ¶ ¹ » Á î À » ² ¶ ðgejÅ6¾¡ðgjuejme9Ã í ì® £ Y H by H Y 7 H F D Y H 4B d D T 7 H F D 5 D d 5B d H D 2B yy T YB 2 5B D T 7 d 5q 2B YB F D H 4 2B Y Y d 5 D 4 H b 2 5B DB h 2 5 QQ¢E6W9QIWQQEWWVduIWG8EWQWoXhE¢CXd8XWVd8pEuXIWu6IXf8GQ¢8XWEII¢Q4 i d T h 2 R 5 5 2 H d T H d H F D D T F D H 4B D 5 ò iD d H 5 d y TB 4 H Y YB F D q 5 H ` T D 2 T b h T H ÷ T D H @ 2 ¢III8kIWVWQpG VIGd6XGVf836s¢WIv¢6QW EIGC¢¢¢o¢Vp¢V¢ev)85 d H D T ô Y d 5 D T d H 5 D 2 Ty R 4 dB 4 5 Yy T H d T Y d 5 D T d H 5 4B h 5B d H q 5 Y H Y T 4 y TB 4 H Y H Y H F D t iy d T Hy 6GVaQ8GVWQ8o¢XIQXQeX¢ W¢u8GVWQ8wEIxXW6s¢u6W¢Qv¢XQQWQIWh8W¢QX 6 ¦ ¦ ¦ ¦ ¦ ¦ 6 ¦ ç $ ¥ 83¥ pzæ ß î çsQà 6Sà8¦fæxÜkÁ ¦ ¦ ä ¹ · ·· ¸¶ ¦ 6 ³´´ 6 ´µ ß î ¦ ä Ü ç ¥ æ ( h 2 I¢T ¡ ¦ ¦ ç ¦ à ¦ 6æ Ü sQà ¡ 8w£xkÁ ß î ¦ ä ¹ · ¦w6 ¡ ¦ ¦ ·· ¦ ¸¶ a6 ¦w6 ¡ ´µ ¦w6 ¡ ³´´ ß î ¦ ä Ü ç æ ( 7d 5q H F W¢upGD H ÷ T D i H F D t Y 2 5B DB h 2 5 4 i d T h 2 R 5 4B h 5B d H F DB @ h H Y R H d T t ¥ P H H Y t H bB D T bB d H h h 2 5 4 H V¢a6IW¢QI¢XGII86 f¢UII¢svXIxXQGEC6WIu¢o8¦ poëjXQ8¢¢EGl¢W6I#p8Q6WY ¦ p¥ Í Ì Ö Í fu× ê Ó É Ñ × Ú Ì Ù Ì È Ç ë¬uvvfÚ qÔvwfÔÖ {Ô§Ïmuuóx%xÎ é ÓÒ Ç Ó Ì Í Ì Ñ ÑÒ é Ó Ò Ë Ç è Í Ë Ú Ó × Ç Ð çÐ .
YB Y T 6X¢U QGeo QW8CQEW6WY 7d H D i 7 d H D T 2 5 Y HB d H d 5y i T t q 5 Y 2 5B D T 2B 7 5 4 d T H 2By ` 2B 7 d 5q q 5 i T @ D 2 HB 2 H b 2 5 4 T iy 7B Y YB Hy T D d 5y i T ¢Xxgv¢fI8XWVUEI 86¢QIXEdIE#8¢eguoQEIQ¢o8QEp#EWE XIV8Excmt i í W ß ço8¥ p¥zæ«YaYaY`ø ß V Y î Ý í W ß V Y î ß Ý ¦ í W ß Y çs¦fæoYÉºø ß ç ¦æ Y º V Y î Y sfxkø ¦ Ý ¦ í Wpß ¦ V Y ß î hsÔxkvxº Y ç ¦æ Y º ø í Ý Ý ¦ Y cfxk~ø ç ¦æ Y º ä 4 üí Ý 5º Û5Wx£V9¢XWp¢Uj6XW6Wj8ExcmuIWGII8W6WW¢QEI¢pG X©cWjf¢upG8¢EI9¢pH 4 ü í Ý º q 5 2 5B Y 2 T H Y HB d H Y d 5y i T t H F D 5 D Y h 2 5 Y H d d 5 4 Hy T D H F D 2B @ 5 d D Y Ty H F D t Hy 7 T d8UauceIXIp8sQW86 IGjV6E IWGD WQWwIWVe8XI¢UQXQ8Xxgm IGXB 5 ¤ @ 5 d ` 2B h 2 5 Y H d d 5 4 H F D q 5 Dq Hy H F D 5 7 d H D H F D q 5 2 5B Y 2 T H Y HB d H Y d 5y i T t H F D 2 2 7 Ry 5 4 ` 2B h 2 5 Y H d d 5 4 H F D q 5 5 D H F D D 9IE8QIEII8W6W8QuIWë¢8GpGVT W6GIWVS6X6#6uIGmXfGoQ¢'VXIVD 7 d H D H F D q 5 D 2 HB 4 ¨ H 5 4 H F D YB id D 2 H F 4 T Hy T d 5y i T t H F D R H ÷ T 7 t Y ` 2B h T H F H F D d H h 2 R t H 2 5 D Y 5 7 Dq Hy H F D q 5 D F `B d H F D 5 D Y 2 7 Ry 5 4 H F ¢XxgvuIW9pV¢#oQ8IEI¢QICIGjQpIIS¢I¢j89GXCIWs¢S¢XWCIWGI9IX¢QCIut Y Y Y à ¦ aaxà 8gà ¡ #õ ä í W ý ß V Y î ß ý ý Ý t YB D T F D t ê D 2B 5 H F D D R 5 6XCVUWpUoX8uIGup¢s¢T WQW¢QvV8EWIVUH 7 d H D F 4 T H q 5 2 5B Y 2 T H F D 2B Y d R 4 4 5 D T F D d 5 D 4 Tq 2 5 7 7 5 4 H F D Y d T H T H d H F D 2 7 Ry 5 4 F 4 T H q 5 h T H F H F D D T D T F pGXCIQ6ulUG8W6¢#8#986IG¢gI¢uQIW#IE8Q¢QmVVgIIWuVCVIGD H 4B D 5 2 2 H F t Hy T D H F D 5 D 2B D R H 5 D Y 7 d H D H F D Y HqBy 7B Y F 4B F 6XW¢Iº6Iu«ç VXI¢GpGkGoX#pIsAG9WQWpG QXI9X«EIC@ W6G«¢QWoXB 7d H D F 4 T H 5D 2 h HBy B Dy R 7 2 H H Y T QEXIEGEp Q6sf¢UF h¢WVgXIWXUh8I8IWE¢EI¢GaIW¢6EwIWfV9IX86#XB ß î t H Y T 4 YB F D 2B t F ` R 5 F Dy Tæ Hy T D H F D q 5 Dq Hy H F D D T 2 7 Ry 5 4 T 2 dVgI¢#8XWVUSQfIGX9WQWfIWh¢UE¢vVUW6XGVfe!8EGV#XgWIp¢XIWh¢fWIIGo T H T 2 5B D T R P H H F D 2B Y 7 d H D H F D q 5 yy T D T F D H 4B D 5 ò ù 2 5B D T 7B 5 d T YB F D q 5 H Y R H F D i h H Y R T 4 d 5 d d H y T 4 5y H F D YB D T F @ t iy H 7 T 2 d H @ Y 2 T 2 T 4 Hy T D YB F D q 5 i h R D Y T D T F D Y 2 5B D Y H R QI¢Qj8WQpVgxXuIWXVUCV8E6#VUp6QvI¢9¢QXp¢EIGë¢ppGlUGjI8EGfQISP HpG¢I8wWoQW6WI6WEIuëW¢#¢XGfQISeAGCA8EWQWppV#6W9aXI¢GIG¢5 F D q 5 H 2 5 Y D 2 H Y H d H d YB F t ÷ d T 7 2 5B D Y H R P T F DB @ 2 5B Y Y H d H 2 T H H Y H @ Hy T D H F D q 5 D H F D D i h H nBy 7B Y YB T d H `y T H F D q 5 H 7 5 Y D T F D 5 Y h H D 4 R d D Y 2 5 4 YB Hy T D H F 8WeIG)6UEEI9XWEGI68X¢IWf¢e98#lUGkWº6GIWWI86XeEI¢Iut H bB D T bB d H ¢8GVVWQph h 2 5 4 H Y T 5 D 2 5B D T 7B 5 d T 2 TB ` 2 T d ` T ô D 2B 5 r ¥ h H d H D 2 H 4 d 5q Hy T D d 5y i T t ¦ ¥ Hy T I8Q6WeGr8EGV#XgWpI¢¬¢X8I¢W8¢oX83sºQQWS6Q8EI¢G 8Xxgv¬8pEI¢mt 4 4 Y sfSÉº ç ¦æ Y 4 4 Y sWx ç ¦ræ Y Ý Y sfokº ç ¦æ Y 4 4 Y sfxkº ç ¦æ Y 4 4 Y sx ç ¦ ræ Y ¦ 4 hsfxhº Y ç ¦æ Y º 4 hsGS{ Y ç ¦ ræ Y Ý Y sWx½ ç ¦ræ Y 4 üí Ý Y 5xckº í Ý Y xck 4 í Ý Y ¥Q{ í y ú ¤¦uu t Y î w ú ß í y ü Aut wüu Y î ß í y û ¤Auvt w û u Y î ß í y ú ¤¦uvt w ú u Y î ß í y xuvt w u Y ß î í xÝ ù ä ç4 üí Ýº ø í Ý ø 4 í Ý ×f5Wxjxh¥xSæ ß î ¦ ø í W ß V Ý Ú Ó × Ò Ë É ÙÒ Ø × Í Ê Ê É Ì Ç Ó Ì Í Ì Ñ ÑÒ ÖÕ Ì Ë Ò ÓÒ Ñ Ð Î Í Ì Ë Ê É È u§fj«wÔzmQÔv«SÏmÇ ¥ .
D Y d n T 5 D 2 5B D T 7B 5 d T 2 TB ` 2 T d ` T ô D 2B 5 r UÅ©W«8EGV#XgWIp¢ª¢¢I¢G¢¢oX83l¥ h d T @ ÷ 4 T d 5q Hy T D d 5y i T W¢o¢Ue8EI¢G8Xxgvt ¢8GVVWQph H bB D T bB d H pkEI¢mt ¥ Hy T 4 4 Y sWx7¡ ç ¦ræ Y 4 Y ç xY ræ 4 4Ý Ý Y sGx'¡ ç ¦ ræ Y 4 Y ç xY ræ 4 í y û u w û Au t Y î ß 4 Y cx8¡ ç ¦ ræ Y 4 Y ç xY ræ 4 4 UsS'¡ Y ç ¦ ræ Y 4 44 Uç SY Y ræ ¦ 4 í Ý Y xck 4 í Ý Y 4 ¥xc7¡ ¥xcY í Ý í y xuu t f î w Y ß í y ü u wü Au t Y î ß í y ú u w ú ¤¦u t Y î ß í y xuu t w Y ß î í xÝ í W ß ß î ù ä çí Ý ø 4 í Ý4 f6~hAxf¡ø AxÝ æ í ø V Ý ¦ uXI¢mXCspWuX¢EWVVEW6Ih ¥ Hy T t 2B 2 @ 5 F Y YB H bB D T bB d H D Y d n T ` 2B D 2 H Y H d H d d 5 D T d H 5 ` 2B 4 2 H d H gB h r h d T @ ÷ 4 T D 2B 5 r ¥ T d 5q Hy T D d 5y i T t H F fWUa9IEGoQQIQ8GlGQ8IEQI6W6hI3¢SVUoX¢s£l u8XIV8ExcmIut í W ß î ß V ço¥ pzæ c ¥ ç zd×f5¬xÝ ¥æ ß îæ ä ç4 üí Ý ø í 6 4 í Ý 6 Ý ! ¦ H bB D T bB d H ¢EWVVEW6Ih h 2 5 4 H Y T d 5q d 5 D T d H 5 D 2B 5 ` 2B 4 2 H d H gB h y T d D 2 H 4 D 2B 5 r ¥ d TByB 7 Tq H F D h H bB d H h D Y R Û H b T I86Qw88WVG6s8oX8wIEQI6W6hIh¢WGoQ6oE8s£lj¢XXE#¢IG68WQpuI8gUF H D 2 H D YB Y 2 5 4 7 d H D d 5 d d H H F D H ÷ T 7 5 D F ` R 5 d F D h H hB bB h 2 H H Y T ugoQWEWI86µW6Gu8WQaIG¢Vd#G¢I8WpGQIEI6Q¢IF IWCVUGuGVfò ß î H FD D T F D H D 5 ç 9 ¥ 83¥ pzæ í W ß ¦ V ß Ý î ä í W ß V ß 5 î Ý ¢ £ º ø ¡ ß î ¦ ä xþ ÿ Y h 2 n H 2 5 h 2 T ç i d D H 7 7 i Y i F YB 2 T b yyB @ Y 2 7 Ry 5 IIIup8I¢upG#9xoWEI¢VsEXCI9IX¢Q4 2 H b H yy T Hy 7 T H YB F D d 5qæ Hy T D H F D 2B 2 7 Ry 5 4 F Dq n H F D D T Y d R 4 4 Q¢6uXVEI9¢pXIW9¢XIVIG¬XÆ#pX86«GpIW9lWpQQx5 xðWVg9XIW©31 ÿ þ H Y T 4 YB F D 2 5D d Hq Hd h 2 T ÿ þ H 5 W66WI¢t x½dWD d 5 d d H Y HB d H Y d 5y i T t H F D Y T D Q8W66XQWC8ExcmuIGVEB 7 R Y ` 2B F YB 2 T b 2 5 2 D Y d n H F D H D T 2 `B Y H h H IW9IXpWXp¢VS¢IfWUuIWuWVU8EW6Ia d 5 d d H H F D H D R DB D Y 2 5 4 5 d H 5 D 7 R Y D 5 2 5 h D T F D Y 2 7 Ry 5 4 H F Q8WQuIWGpWEWI86#Q6#W§IWCVI#pCVUGp#IE8QwIut ¹ 6 ¡ ³´µ ·¶ ¡ û ¦ QQà S8¦ ws¥c¥UýS68E¢uXe8EGpE8WwIut 6 à ù 6 ä û º à à ù i 2 H b B ` Y B 2 5 B D R y 5 Y H F ¡ ¦ ¹ º ³´µ ¹ ¦ ¡ w6 ³´µ ä ¦ ¦ ·¶ ·¶ ¦ ¦ ¦ 2 5B D T R P H H F D iq YB D T Y H @ qB 5 d H H d T Y 2 7 Ry 5 8EGVIx6uIGfXWVGuvm#QQ¢CI9IX¢Q4 H H d F D D Y d n H F D q 5 Y 7 R Y H F D D T F D @ 5 F Y iyB Y T H 2 T 4 H 2 5 d H H 5 D Y 7 R Y H Y H F D t 4 h 2 6WIWwfWUIW¢u9IWIWVUWÏcIWEX¢gA¢Q#I#IQQGu#I#QIW¢AI¢T tëpj¢wQE8IQ8WpaIWSëVQW¢I¢fo8EW9GC6Eµ8WQQ6IwvUxIG#wIGm¢5 t T q 5 H 4B 5 F 4 d H 5 d H F D i t H 4 d 5q h 2 T D F `B d 5 D Dq Hy 7 5 dq h H H 4 5 d H @ t h 5 F D H 7 H F D q i 4 T d R 4 4 T q 5 d H h d 5 H F D H B 7B T 7 5 t Y 2 7 Ry 5 4 H Y H F D q 5 F 4 T H q 6¢WIQ6¢u¢6IW¢#IWa6X9Xl#§QI9IE8Q9W6IGv¢VgV5 IWIWw6IXI8 7 R Y H F D d H hB Y 2 5 ¥ 8¥ Í Ì Ö Í fu× ê Ó É Ñ × Ú Ì Ù Ì È Ç ë¬uvvfÚ qÔvwfÔÖ {Ô§Ïmuuóx%xÎ é ÓÒ Ç Ó Ì Í Ì Ñ ÑÒ é Ó Ò Ë Ç è Í Ë Ú Ó × Ç Ð çÐ .
ç pzæ ¥ ý Ýç ßæ ý Ôs8S ä ¾ý ( ä ç ßæ jsmÝ i 2 H bB ` YB y TB 7 5 2 iy 5 2 5B D Ty 5 d H D 2B 2 TB ` 2 T d ` T ô H F o68E¢uX¢98IxE¢s¢XGl8W6GoXV8IVG8¢aIut }ã g e i ÷ã e d i eg j ã e y r p j i ' p j £#¨uõhfæh£7£h{!s¡ÉÂ¡og r p % Ø i  p " i  r "  £sÙ&u¼U$h#Ó! ß õ YB Y iy T 2 QEWxEVU¢T y T 4B d H 7 R 2 2 5 Y D H D i 2 T 7 2B h H Y Y R 4 YB h H d T Y Ty R 7 d 5q H Y H F t Y Ty R 7 d 5q 2 5B D Ty 5 d H D 2 ¢gEWQ9ISº8dGxWÆo¢#«EªQWIQXIW¢9¢XI 8W6Iuÿ6¢p W8Å8EGVX8WQWoXB q 5 @ HB b q 5 D 2B 5 H F D 7 5 dq YB D T F D t i T @ D 2 H d H gB h iy D F `By Y T 2B D 4 H Û R Y H F D F 4 T 5 d ¢§6Xx¢oX8IG¾8WXlUG¢goQQsEIEWS¢XXAÆEdQ¡IpG¬¢oWpI¢T Y R D Hy t YB F D 5 h 5 D d H h d 5 2 1 h H By T d H 2 H ` d H F D d Rq H 2 T 4 DB D T F D D 4 Tq H F D YB t d H b H @ 5 I 6Xj6XIWeIeG 6IW8¬3AQ6XE¢G6IQ86IGfIÆVg VUW 6¢pG XvQ686vcIF tgo¢G89X8 QI¢XWpQ9XIQp8XIfGQpIQWwIVw6GV#8G¢v6Gpp#86wEI8Qm£1 D 2 T D d 5 7B H d 5 Y 2 5B Y 2 H 7B h d H F `B F 5 D h H h 2 H D H h 2 T h H D T 7 5 D R T d H D R 7 5 4 H hy R 5 4 D ¢8GVVWQpo¢8jQI8o¢h¢W8WVG6s¢joX8ë¢GWS6Qj8pQv6VWxW¢o¢US¢f8q H bB D T bB d H h i 2 T d 5q d H h d 5 i 2 T q 5 Y d 5 D T d H 5 D 2B 5 y T d D 2 H 4 d 5 t h H @ H ÷ Y t h d T @ ÷ 4 T t h d T @ d 5 h 2 n 5 D h H Y R H 2 T 4 Y Y H 4 5 d YB F D iy d T Hy d H h d 5 2 H bB ` T q 5 Y 2 5B D T 7B 5 d T H 4 R h 5 d IUGÅQIdÅ¢Q W6QxWIaXIWAW¢QXQQpW8Å68E¢¢uI8EGV#XgWIp¢eQpII 5 D Y Hy T t d 5y i T t q 5 H Y R H F D F ` R 5 d F D h H 2B 7 d H D H h H 5 D Y D 2 HB 4 ¨ H 5 4 H d WaQEI¢mw8Excmu¢dWIpGr8p8WIWrQpX9WQW6I#sGwWS6X6#6d¢T IWdW6IC@ H FD Hd H F ç ¦ ¥ sI pzæ ÿ þ xfä üí xÝ Ü ¾ ø í W ß V Ý Y T Y D 2B 5 ` 2B d 5 F `B H 2 ¦ ø ø q 5 Y 7 d H D 2B D Y T ¢CGoX¢sIEW8S8X6Iuð C¢u9W6GeXVg4 H 2 T 4 ê D 2B 5 hB d ` D T H bB D T bB d H h F D þ H F D 5 D 2 5B D T 7B 5 d T H 4 2 H d H gB h T ty T d H 2 H ` 2 A¢guSE8AEW8ul8EWVVEQIAGIWW8EGV#XgWIp¢d6IQQsEIez¢G6IQ8A31 } ã â r e © p t i p r p ng l § e i eg j  ¥gã  r p i p Ç£Uuvu{fD¨æk7v¦£su8×¤ ç ¡ pzæ ¥ ç fjx¢©h¥x8Å6 ß îæ ä ç í Ý ¥ ø 4 í Ý ß Axæ î 6 í W ß V í Ý ¦ Ý ß oõ Û H bB D T bB d H ¢¢EWVVEW6Ih D Y d n T q 5 2 5B D T 7B 5 d T H 4 2 H d H gB h h d T @ ÷ 4 T d H h d 5r h 2 5 4 H Y T h H bB d H h H b T F H @ Y R F fWU¢)8EGV#XgWIp¢QI6WQsEp«WVo¢U6IW8£I86QA©Q¢EW6I8gUISut ç h ¥ o¥ pzæ í W ß ¥ V Ý î ß ä í W ß ¢ ß V î Ý 4 ¡ 6 6 c 4 ß î ¦ ä xþ ÿ Û 7 d H D d 5 d d H 2 5 B D T 4 2 R d D ` 2 B h T H W6G8W6¢XG lgIIG IEI¢QXy H F D Y H hB b 5 d 2 7 Ry 5 4 F D d R 5q H F D H Y T 4 YB F D 21 û ¥ à 6 à pG#6IXxcWp©9IX¢QÆGp8IW¢g9XpG©3«VGIfS8¦ ù 6 uÃÃà ¸Q88«EIC@ ä û à 4 ù Y H bB ` F 4B F ¹ ·¶ ¡ ¡ ´ ¦ ³µ ¦ ¦ ¦ ¦ ¡ ¹ ¹ ¡ ä ¡ ³µ 4 ´ ·¶ ·¶ ¦ qB 5 d H 5 E9WQ6GD ³´µ 7 R Y Y 2 7 Ry 5 4 H H d F D D Y d n H F D H 7B D YB F IuI#pX86w6WIWCWIupG9XWCEIut ¢¥ Ú Ó × Ò Ë É ÙÒ Ø × Í Ê Ê É Ì Ç Ó Ì Í Ì Ñ ÑÒ ÖÕ Ì Ë Ò ÓÒ Ñ Ð Î Í Ì Ë Ê É È u§fj«wÔzmQÔv«SÏmÇ .
DSE8up8¢u6XW6C8ExcmIEWIQpI¢UufI X¢XpC8«CIVuÆuGoX8 2B 5 D R 5 T Y HB d H d 5y i T t ` 2B Y R h H h 2 T H H D Y R 7 5 Yy T F 4B F @ t ¦ ø ê h 2 T ¦ 6 ê Y D 2B 5 DlCQ8GVVWQpw¢QIIEQIEWE¢fpIhhQGo6WQIQ#oX¢sVu¢EGl¢W6IuIGEEp8¢I2 T Y H bB D T bB d H h H d T h H h Ry 4 2B 5 Yy T D R t h H D 2 H Y H d H d ê D 2B 5 D T H bB D T bB d H h H F D YB iy 2 5 D 5 H d H ¥ Hy T t q 5 5 D H F D D T h H D T d D Y RyyB YB Ty R 7 d 5q Hy 7 T H 2 i ¥ P 5 D Y R 5 ` 5y T 2 Qf7p¥ puXp¢vëV8GIWVwQGlGGfIXEXmXXI W¢vXp#¢p9fmp8 oëQWmI¢88XVU¢T ç $ ¥ 8Â pzæ ÿ þ x`ä üí WxÝ Ü ¾ ø ü í W ß 8 9 ¾ V Ý 7 t B t Y D 2B 5 ` 2B d 5 F `B H 2 D T Y H bB D T bB d H h H h Ry 4 2 ¢H pQGoE8spXW¢sS8EQICV68EWVVEQIIpXQpXB 2 T 4 ¦ ¥ P q 5 2 5B D T By T d H 2 H ` T t ¥ P ¢QI mëjk¢8EGVQXXVGQpQ8eI8 p§ëj ¢6W¢qpIwW88XWV8W6GoXWE9W6f 2 5 h H Y T & Ty R 7 d 5q 2 5B D Ty 5 d H D 2B H DB 7 d H T q 5 H Y R H F D 7 5 dq ` 2B YB d T H bB D T bB d H h D Y d n T d 5q H 7 H F 4 Y ` 2B 4 2 H d H gB h H 4 T Y xy y {y 2 ¢9WIIWp8WpXWEW¢¢EWVVEW6IwUu¢99QIIX6IQQsEI9Q¢IWéÆ¢T d 5q Hy T D d 5y i T t H F D D H 2 d H hB Y 2 5 Y Ty R 7 d 5q 2 5B D Ty 5 d H D 2B 2 TB ` 2 T d ` T ô 7 5 dq Y d 5 D T d ¢wXIV8EgmIGuxIu6IXI86¢IwW8¢XGl8W6GoXV8IVG8¢p8WW¢GVWQH r 5 H 4 2 H d H gB h D 2B 5 xy y } h H D 4 R d D Y 2 5 4 Hy T D d 5y i T t T q 5 Y Hy 7 T H Y R 5B b H d H F 8w6IQQsEIoX¢sé6s©6G6pWGfI86aEI¢u8Excm#¢QXp#¢pI8Ex6WpwIut D H 4 2 5 4 H F D iy 2 5 h H H 2 H @ H d H F D R t Y h 5 F D H gp6QI86uIG#EI8Q6IuuW6ICpIU6IpG697 y¢gEWQ9IS8u6QI6WQWo¢dXeII¢wseVgCX¢X#¢IxE8W6IGj¢e¢XWIQXIwG6EI T 4B d H 7 R 2 2 5 Y H 4 2 H d Hq H d i 2 T 7 2B h 2 R 5q H 2 T 4 Yy TB 7 5 2 iy 5 H Y H F D q 5 2 5B Y Y R 4 YB h H D Hy r 7 5 i H D T D 4B h Y 7 Hy 5 d H F D Y T h H h Ry 4 2B H hy R 5 4 Y H bB D T bB d H h d H h d 5r d H F `B F iy Y R 5B b 864 vVGVG6Xp#6XI¢WIuIW¢QpIX6IXuXI86Q¢EWVVEW6I6IW8£f6I8Xp EI8Eo ç ¥ S8 pzæ ý W ß ç ßæ ý V so Ý ø ý Ýç ßæ ý ëÔsoo ä ç ßæ jsvÝ 2 5B Y Y H d H H F 8EWW6WIwpGD ` 2B 4 R h 5 d t h H Y R H d T H bB D T bB d H h D Y d n Y DB h 2 T 2 5B D 4 2 Rq H F D q 5 Y H Ry T b H D H d 4 YB h F 4B F @ 2 pXQpII6WI9WVd¢EGl¢W6IwfWUuGrIV)8EG6pI9pG¢wQpX¢VG6QXI)EICkXB H Y T 4 H F D d 5q YB 2 5B D T d D Y RyyB d R H 4 T Y 2B Y D 2B 5 2 H bB ` D T 53 } 1yx 1y } xy 2 5B D 4 2 R ¢gfIG8X8EGVWGpXXEvpvVQ¢U#EWSE8dQ¢E8vVu6z4¢2z¢l2xw ¢e8EGIIq H F D q 5 Y H Ry T b H Y R Y Ty R 7 d 5q H DB 7 d H t ç ßæ Ý d 5q y TB 7 5 2 iy 5 T D 4 R d D Y 2 5 4 5 t 2 5B D Ty 5 d H D 2 pGhV6IXV¢CWpj¢IwW8WE9WQ¡sm8S¢98Ix8su6IGp8Qam8EGVX8WQWoXB 2 TB DB 7 d H ` 2B Y R i D R 5 T D F ` R 5 d YB F 4 T 5 d T 2 TB ` 2 T d ` T ô H F D q 5 2 5B D T By T d H 2 H ` ¢WE9WQ{pXWpS8sVo8p8WIE#¢8WIIV#V8IVG8¢IWh¢8EGVQXE¢G6IQ8ªi Y 2 5B D T 7B 5 d QI8EGV#XgWpI¢T 2 TB ` 2 T d ` T ô Y T 5 D h H dd Hq H d H d ¢8p¢G8V§a¢GQWQWW¢T p§ëjå8WAQ¢EW6I9sA¢gwVIGa6#6IW9QpQW6hIh ¥ P 7 5 dq h H bB d H h H 2 T 4 D T F D Y H 7 H F 4 Y H 4 2 H d H gB H DB 2B ¤ h H D 2 H Y H d D Y R Û Y 2 5B D T 7B 5 d T H 4 2 H d H gB h H DB 2 n H F D H bB d H h H d H @ t D 2B 5 H D H d GIXv)QWS6W6WIfIÉI8XWV#XgWIIVQI6WQsEpGIUdIW8WQpQWvvgoE8seW6Q4 r YB h H D TB d 5 d T H F D D T Y H bB D T bB d H h H Y H F D H D T Ry T b H h 2 T t F Y H 7 h H 4 T YB R P H 2 T H Y 5 7 fXpdGlWp8WIp¢IGwVQ¢EWVVEW6IeW6IG#GVUE¢V6ÅIVvhWQ9¬Q6¢UWEISQÅ¢dW89XB tsm¢a8GVVEQIep8Q6W8UwIWw¢¢G j£Q6 IGXaIX¢VwW6QWEI6IG¢o¢T ç ßæ Ý q 5 H bB D T bB d H h h 2 5 4 H Y d 5 D Y d n H F D H ÷ T D H @ q1 D H Y H F D 2B H Ry T b H D H d 4 YB h d H F D 5 i 2 Y H ÷ T D ß 2 H F @ 5d H h 2 T t ý 6¢¢G©¬QICeQQÅp¢¢ëß ä Æ6ICI8 EdëÆ¢g¢oQEQ96xQepG VIGd6XGVfò ß 2 H F @ H 2 5 YB ý Ý F 4 T H q 5 D 2 HB 4 ¨ H 5 4 H F D D T F D H 4B D 5 ç ¥ ¥ oo pzæ ç )'£Qç )U!æ ß 6 4 ßæ ß 6 ä ß Ý ø ç ß 6 4 ßæ ç ß 6 ä ß s)'£Qs)U!æ ßæ ç4 ß 6 ä ß ßæ ç ä ß 6 4 ß çW£)6 Q£)U!æ 4 ß ç ä ß !r6 Q!)7£æ 4 Ý ä Ý ä ç ßæ ß æ ç ß 6 ä ß £¬ø sr6 Qsr7£æ !ªsmÝ ç ß s)6 Qs)U!æ ç ß ßæ ç ß 6 4 ß Y D 2B 5 h H 4 T YB R P Hr 2 5 2 F DB @ ç 2 5B GoX¢sQ6¢UWEISQ£8IGCë8EGD rl8¢WGxu8A8XWV¢sQGoEwEGVWI¢USC8wXI W¢V8IVG8¢uEI#EWwIGD 8WQ¢VD Ty 5 T d D H d 5æ 2 5B D Ty 5 d H D 2B 4B D T d h T R P d 5q Ty R 7 d 5q 2 TB ` 2 T d ` T ô Hy 7B Y H F 7 5dq 2 H ÷ T H 2 T 4 ç ßæ ý H F D q 5 2 5B D 4 R d D Y 2 5 4 H F t ) H H d ` H h q 5 ß 2B Yy TB 7 5 2 iy 5 H d T ç ßæ ý H d H F ¢gs8opGvV8EGIWWI86uIuv)06W86IvVeeXX¢X#8p8w¢us8SW6IC@ $ 3¥ Í Ì Ö Í fu× ê Ó É Ñ × Ú Ì Ù Ì È Ç ë¬uvvfÚ qÔvwfÔÖ {Ô§Ïmuuóx%xÎ é ÓÒ Ç Ó Ì Í Ì Ñ ÑÒ é Ó Ò Ë Ç è Í Ë Ú Ó × Ç Ð çÐ .
ç 9 ¥ 8Â pzæ í W ß ¡ V ¦ V Ý î ß ä xþ ÿ F D B Y H F D Y 2 5B DB h 2 5 4 H Y H F D d H h 2 H û ¦ à ¦ à ¥ 6 à W W#IGuI8EGEpI8Q#W6IGw6III6g8gpfxà ¡ p¥ ù 4 ¦ ¦ ¦ ¦ ¦ 5 D Y 7 R Y 2 7 Ry 5 Gu9IWe9IX¢Q4 " ä û þ à º à à ù 2 5B D Ry 5 Y H F D ` 2 B b T Q¥à s¥c¥U¸«8EGpE8W#IGIXxgUF ¡ ¡ ¡ ¦ ¡ · ¹ ¡ ´µ ´ ¦ ·· w6 ¡ ´´ ¡ ·· ¶ ¡ ³´ · ·· ¦ ¹ ¥ ´µ ´´ ³´´ þ ¹ " ´µ ´´ ä · º ´ ³´ ·· ·· ¶ ·· ¶ ¡ 6 ¥ 6 ¦ ¡ ¦ ¦w6 ¦ ¦w6 ¦ 2 5B D Ty H d H F D iq YB D T Y DY R 7 H @ t i 4 T d R 4 4 T q 5 d H h d 5 H F D H B 7B T 7 5 ¢XGlQuIG#XWVGfI wI¢6¢WIQ6¢V6IW¢wpGf6X9Xl##§t H bB D T bB d H h D Y d n T 5 D 2 5B D T 7B 5 d T 2 TB DB 7 d H D 2B 5 r ¥ y T d D 2 H 4 d 5q Hy T D d 5y i T t ¥ Hy T ¢¢EGl¢W6IWI9W8XWVdEcII¢¢WE9WQpoX83s¢GWS6Qu8EI¢G8Xxgvp¥ puEI¢mt 4 4 4 Y sfSÉº ç ¦æ Y Ý ç ¦æ Y 4 Y sÔxkº Ý Y sWx½ ç ¦ræ Y 4 Y sÔxXþ ç ¦æ Y 4 4 4 Y sfxkº ç ¦æ Y Y sx ç ¦ ræ Y 4 hcfxpþ Y ç ¦æ Y 4 4 4 hsfxhº 4 Y ç ¦æ Y º 4 hsGS{ Y ç ¦ ræ Y þ ¦ ä 4 Y sfxhº ç ¦æ Y 4 ä 4 Y sGS{ ç ¦ ræ Y 4 Y sfSXþ ç ¦æ Y 4 Y sWx ç ¦ræ Y Ý Y sfxpþ ç ¦æ Y " í xuu t y w Y ß î ø í xÝ " ç ¦ ræ 4 Y cGxY í y F u w F ¤Gu t Y î ß Ý Y cxY " ç ¦ ræ 4 í y ü u wü Au t Y î ß " ç ¦ræ 4 Y sWxY í y û u w û ¤Au t Y î ß ß î ¦ ø 4 ç ¦ræ " 4 YhsWxY í y ú u w ú ¤¦u t Y î ß 4 üí Ý Y 5Wxckº í Ý Y xch 4 í Ý Y Axc 4 ü y w u t þ Y ß 5í xuv!q î í !xut î y w u Y ß 4 w Aí y xuu t " î Y ß ù ä ç4 üí Ýº ø í Ý ø 4 í Ý 2jf5zv~¥xSæ 45üWí Wß í ß W þ V ø V Ý Ý 4Aí Wß " V Ý @ 5y H h H 2By D R 5 Y T F 4 T 5 d usEQeQIEXWp8C¢¢oWpI¢T Hy T t d 5y i T t H F D ` 2B Y R h H 2B 7 d H D H h H D Y R 7 F 4B F @ ç T d H `y T H F D iqBy 7B Y 5 D H 2 5 Y EI¢m8XxgvIWApXWp©6IX9W6G6p#I ÆXpCaoGI68XVIWkEXp#EWAGI8¢T 2 H ÷ T D YB D 2B 5 ê H F D 2 5 D 2 HB 4 ¨ H 5 4 H F Dæ Y D 2 HB 4 ¨ H 5 4 H b T F @ 5 2 Y 7 d H D H bB D T bB d H h H F QV¢ EoX89IWÆ89oQEQ96xQIWGoQEQ96xQ8gU§cI#QW8EWVVEQIIut ç ¥ oo pzæ í Ê DEW b ä ¾B ï ß V ï ß sæ 7ì CÈ ç ß î õ ï ¦ ï Ý @ Aä ý ü Ií W ß V Ý Û c ¥ P dpëëj 2B 2 H bB ` 2 5B Y 2 T H Y HB d H Y d 5y i T t H F D q 5 2 5B D T By T d H 2 H ` ` 2B @ 5yy 5q H F D Y H dB R P H d YB F t EkQ88k8XI¢UQXQ8XxgmIG¢¢XGlgXE¢G6IQ¢dIECsEX8 pGQXIS6WXpuaUê Ú Ó × Ò Ë É ÙÒ Ø × Í Ê Ê É Ì Ç Ó Ì Í Ì Ñ ÑÒ ÖÕ Ì Ë Ò ÓÒ Ñ Ð Î Í Ì Ë Ê É È u§fj«wÔzmQÔv«SÏmÇ 8¥ .
° © ¢ ¬ Ò « e ¨ § « e § q i ssd¦qqF$¦sA rh pÐ ¬ ¦ Ð ¤Avg¦sqÕ¦¥UadÕu%q¦ÚvFÕ¤¸q¦£1¤Aqq¦vfq¤ÉddqÕ¦¤S¦£¦¤o¦q¤¦£8d¬ ¦a ¨ § « e Ñ « © Ò ¥ Ò ª ¬ Ò « © ¬ « § ¬ « ª ¯ ¢ ¬ Ò « e ¨ § c ¬ ¬ § ¬ « ª ¢ « § ª ¥ ¨ § © ª © ¨ § b h Y d H by 5 Y h H h 2 T D 2 HB 4 ¨ H q 5 H Y R H F D i h H D 2 H 7 Hy 7B H 2 T 4 i H F D 2 H F @ y T 4B D 4 T d H QQ88WQIp¢UoQX696m¢aIuIGdSe6GoQ9QEI9Xus¢Qe6IW6ICh¢QXG¢Gpw 2 T 4 Y 2 5 B D T 7 B 5 d T `H h T Ü d 5 2 T B D B 7 d H t y T d H 2 H ` 2 1 h H Y R i y H h B @ h 2 T t 2 R d 5 D D 2 H B 4 ¨ ¢QuI8EGV#XgWIp¢w¨¢v8¢XGE9W6f!z¢G6IQ836WIeQIECIVshIW9GoQX696H t H h 5 4 5 D Hy 7B Y YB F 4B F @ t d H by 5 ¢IxQkGXp#EW EÆEICjI688WY ¢I88¢XIXGf¢I¢Q#roÅII8wEB y T 2 5 ` TB hB d D T q 5 Y 2 T H 7 i h 2 R 5q Y ûç ¦ à à ¦æ Á 3sgIW8feýù 4 q 5 2 5B D T Ry T b H H F t y T 2 5 ` TB hB d D T H Y T 4 YB F D 2B t F D hB @ h 2 T yy T 7 Y T ` 2B b T F Y H 4B d D T ¢¬8EGVUE¢V6puº¢I88¢XIXGAW¢Q#EIGÆXuWpXCpp¢UuXVdr)IExcI#6QXGV#7 h H h 2 T q 5 7 d 5q H F D 2B d T H T i H F t Y 2 5B D T 4By T H 4 2 H d H gB h H DB 2 n 2B 2 5 7 7 5 4 i d H b H d QpI¢Uh¢ÏW¢CpGX¢gI¢#Ium6I8EGVQXXpI¢fQI6WQsEpWEIIE8#986#fQ¢C¢T Y d 5 D T d H 5 H Y H F t Y 2 5B D T R P H d T H 2By q 5 D H Y h Hy R 5 4 T q 5 2 5B D Ry 5 Y H F D Y HBy 7B d 5 D T d H 8GlGQ8QIuQp8XWVUSQu¢QIXEu¢Wr6XIp8Qu¢¢XGX8dIWQXEI9Xw8GVWQ85 B d D T 7 H Y d H b 2B 2 i 2 5B D T D H d d H D 2B Hy 7B Y T 2 H bB ` H 2 T 4 yyB D Y DB D R t d 5 D T d H 5 H Y d H b 2 GV#wWW68oX8EGVG6WpWQWoXXp#EWQ¢E8u¢QXEGfCEfpIUQ¢GVWQ8wWQ¢oXB 2 T q 5 2 5B D T Ry T b H H F D ` 2B h 2 T 7 H h YB DB H 4 2B Y t Hy D R Y H d 5 7 YB @ 5 d h 2 5 4 H Y H F D q 5 ` 2B 2 T H ¢m¢d8XWVUXV¢IG pXIIVd6IXEQpXWUVXWpIWW89XÿcWI86QWpG§¢ IEI¢g97 H F t Iu§²W Ý gG¢ WVpQ9WQWoX IWEEIQ IGaW8WIVQp8XWEII¢Q¢III8 IWI¢T i T d d T H D TB h H 7 d H D 2B H F D 2B Dy R Y H d H F D H d 5 D Y h 2 T t Y 2 5B DB h 2 5 4 i d T h 2 R 5 H F D h h tgQI6W6hIC² Ý gWW¢8W668pG¢V¬YxgG¬I9XWd£1 6¢gEQºXp8IWk6IQIS6WY DB H 4 2 H d H gB h t i T d d T d 5 D 4 H b H F D H ÷ T D X Y i T Y iy 7B Y D d T Hy 4 H hy R 5 F Y H 4 2 H R P H YB F D 2B d 5 D 4B h H d H F D q 5 ` 2B 2 T H 7 H F D t 2 5B D T D 2 H 7 Hy 7B y T 4B D 4 T d 5 D D 4 H Y H d F DB EIGX 8W6EIQIeIGC¢IXIVg9IG§8EGVGS6#6XI9Xv¢gEG¢GIG66sQWÅWEª ç ¦ $ ¥ sdpzæ ä ä ²Ý ( ²W Ý ûç ¦ à à ¦æ ù W² Ý ÔcgIG8ÔeýÁf 4 ß y A Çw² Ý cgà ¡ 8w£Á î º² t ø ç ¦ à ¦ 6 æ ¦ H 4 2 H R P H Y d 5 D 4 H d d 5 4rd 5 D 4B h H d 6IQIS6W8W66WW¢Q3Ô8GXI6WI pGeoQW6WI6Wu¢QEIC@ H F D i h H Y Y H d H H d H 2 T 4 F 4B F ç ¡ dpzæ $ ¥ ¢ ß y A Çw² Ý sQà ¡ 8w£eÁ î º² t ø ç ¦ à ¦ 6 æ û ç ¦ à à ¦ æ Áù ä ÔsgpG8f¸Uð² Ý ( 4 ¦ 5 D D 2 Hy T bB R P WuS6X¢VISQH YB YB F D iyy T 4B D T 7 H F D T EfEIGdEXVgXWV#QIWV Q8#EeQQ ¢cIfI¢XGII86¢UpI8fXIEWEkXICXB h H Y 5 7B 2 H H H b T F Y 2 5B DB h 2 5 4 i d T h 2 R 5 D Hy F 4B dB ã F 4B F @ 2 ç h ¥ o3 pzæ y A t º² ø ç ¦ à ¦ 6æ w² Ý cgà ¡ 8w£Á ßî ¦ ä e² Ý scgIG8ÔeÁ (ç ¦ à à ¦æ ¦ YB c ¥ P d 5q 2 5B D T D 5 2 B d D T 7 h H h 2 T X3 pëjp88XWVVI WWVdQpI¢U i Q8WVG6s¢9WWV# §¢5 W8pGdEpG699¢WIGuQWpphEeI6IQI#¢Q WdVgH d 5 D T d H 5 B d D T 7 T q 7 d 5q H F D 2B h 5 F D H 7 H 7 T Y H F D Y Y H d H H @ qB h 2 H F H d 7 5 4 5 D i Y T H DB R P YB 2 5B D T R DB Y H F D t d H b H @ 5 h HBy T H 2 T 4 c ¥ P @ 5 F D Y R Û D Y d n D T D 2 H hB b H D 5 WEISE8EGVUWEIGëQ686vcfhQXEII¢u¢g pIjfcICI'fWUlS6IXxC¢I2 iy T 5 d YB DB d 5 D T d H 5 D 2B 5 T q 5 7 d 5q H F D 21 H h 5 4 d H D R 7 5 4 T 2 7 H F D D 2 H 7 Hy 7B 5 EIVU8WpjXEj¢GVWQ8oX8wh¢Ï8uIG3VI6QGI#¢Qu#EB QIWvoQ9QEI9XwGD d H F D 5 2 T H DB R P h 2 T D H 4 2 5 4 2 TB DB 7 d H H F D ` 2B Y R Y h 5 F D H 7 D 4 R d D Y 2 5 4 5 D ` 2B F D H 2 5 YB D 6IGVI¢aGEpxI¢pQ6I8QeVG#Q{IW#IEWIIxIG#f6pWGfI86dGIEIGuI8CXf£1 }  ã â r e © Up Ø  d § e i e g j  t g ã y y T ã  t g j t  Ç£UuVfD¨ih7vs`QvÚr d ß ß õ T y R 7 d 5 q R} P T Y T 5 D h H d d H q H d 5 Y y T Y B Y B F I 8SouQ¢9GQWQWX¢CEEIut ç c ¥ o3 pzæ ç ß îæ ä ç4 üí Ý ø 4 í Ý 6 zdff5Wx¬¥xfæ ßî ¥ 6 45üí W ß V Ý ø í W ß 4 Aí W ß V ¬ø V Ý Ý Y T h HYY Hd H H 2 T 4 h 5 FD H 7 H F D h 2 ¢k6WQWppus¢QxIW69wpGI¢T 9 3¥ Í Ì Ö Í fu× ê Ó É Ñ × Ú Ì Ù Ì È Ç ë¬uvvfÚ qÔvwfÔÖ {Ô§Ïmuuóx%xÎ é ÓÒ Ç Ó Ì Í Ì Ñ ÑÒ é Ó Ò Ë Ç è Í Ë Ú Ó × Ç Ð çÐ .
YB Y iy T 2 T d 5 d d H d HB d R 5 ¤ ` 2B Y 6XWx¢U¢C8W6uQXI8U9IXIR h H 2B T D 5 H 2 T 4 H 7 H F 4 Y H 4 2 H d H gB hr H DB 2 n T q 5 d 5B b T F H d 5 d d H H F D D R 5 T 2 5B D T 7 d 5q 2 6IX¢Gp8¬¢ge9QpW6IQQsEI£GIUf¢ ¢XxgUQ 8WQepG p¢s¢8EGV#W8XB d6IGfIUj8EGpEW6WQp96G#EXWXVauQpXxcWIvEx¢IW¢Q#sI6IwXXIWEXIªvVXIVD H F D d R ¤ 2 5B D B d 4 Y H h h H DB 7By iy dB Tq T Y H hB b 5 d DB t H d R Y T H 7 y Rq H Y R T YB YB F D HyB F Hy T d¢Xxgvp8WeQpX9WQW6IpWQW8WQepXI¢QXIWvVQI8IG¢9W6GXrQWQIH 5y i T t T 7 5 dq h H 2B 7 d H D H h 7 d H D d 5 d d H ` 2B h T Hy H F D q 5 d H h d 5 H F D q 5 Y 7 d H D 2B h H Y Y H d 2 H Dq 5 YB d 5 D T d H 5 2 T q 5 i 4 T d R 4 4 T H F t Y H 7 H F 4 Y H D T hB h 2 T 4 H F D q 5 i 4 T d R 4 4 T H F D Y Y H Y Y T 5 6G¢Xj¢GVWQ8#V¢¢GI6Q¢fIuQ6#6IWCWVUXpI¢gIWs¢9¢GI6Q¢IWvQW¢aGD Hy T H D Y R 7 H 2 5 2 5B D T 4By T 2 H bB ` T d 5q H 7 H F 4 Y H 4 2 H d H gB hr H DB 2 n T D 4 Hy H Y 5 D d H h d 5 2 XI¢wsfp ap88XWVgEXIIVQ¢E89¢a9QIa6IQQsEI£GIU66X6W9Gf6IW¢31 ² ¹ ² ó ´ g¦¶ ç $ $ ¥ 8dpzæ G À Ã À À À ±¹ À î Ã i9ó6eëe¸ 6 ß ® ß y Wð`² Ý saSpcgà ¡ xcg8w£xkÁ î ¦ ß ît ä ç ¦ 6 à ¦ ¥ 6 à ¦ à ¦ 6æ Ü ¦ ß y Wð×d² Ý swxp¥à ¡ p1S8fxkÁ î ¦ ß ît ä ç ¦ 6 à c à c 6 à ¦æ Ü ¦ ß ² Ý ç $ ¥ SVdpzæ 6 ²Ý Û ç Y H 7 H F 4 Y H Y H F D ` 2B iy T 2B Y H R Y YB y T B 4 2B d H F D q 5 H 2 5 YB YB F D F ` R 5 F Dy T t Y 2 5B DB h 2 5 6#6IWuQIWpXxEIp¢XCQIWE¢UEQIEWIuIWm¢uI¢uXuXpG8I¢IG¢UQp8XWEII¢Q4 id T h 2 R 5 H FD i h HD T Hd ¢III8IWdoeQWVgQ4 QEI8WpIGW8p8XvW6I W¢6¢WIQ6¢§¢6IW¢9¢GpGD 7 Hy 5 d H F D H d 5 2 `B H @ H d H Fæ H d T i 4 T d R 4 4 T q 5 d H h d 5 H 7 T Y H F F DB @ Y H 7 H F 4 Y 2 TB ` 2 T d ` T ô 5 @D t d Ty R 4B D d T 21 Y H 4B h 2B H 4 T Y q 5 h T H d Y d H hB @ T H b T F yyB WECv6#6IWV8IVG8¢w386¢I6XW¢U#3v6QQEIIEQVUWh¢d¢QWIQIECu8gUIEXC@ i H FD D R 6IWvppot dpIVw pQSjËX9Q88IxIG#IG¢9¢GI6Q¢¢QI8#¢WCIG8gUF $ ¥ h 2 T c ¥ Y P 2B 2 H bB ` Y h 5 F D H 7 H F D Y T i 4 T d R 4 4 T q 5 d H h d 5 H 7 T Y H F D H b T D T F D h H bB d H h H 2 T 4 Y 2 5B D T 7B 5 d T 2 TB ` 2 T d ` T ô ` 2B Y R Y H 7 H F 4 Y D 2B 5 H b n D T F D t H Y d R 5 lUG68EQIfsd¢gI8EGV#XgWpI¢#¢8p¢G8V§wIXIuQ9QpWvSE88IVUWp¢WI¢Q4 q 5 t Y R 5B b 5 YB D1 Y D 2 HB h T d ` H H D Y q 5 i DB 2B 4B b H F D 2B h 2 T Y HB d T h 2 R 5 D T Y R 5 H ` T D 2 T b h T H ¢SQp8Xxo8Xv£QWoQXp¢G8dQQWh¢ 3IX6XxpGXdp¢QX¢UIp8sVvI868¢GSV¢¢ 2 T 4 YB F D t 2 5B D T 4By T 2 1 h H D T Ry T b H ` 2B H H d T Y H bB D T bB d H h H F D H d H F @ D 2B 5 H F D 5 D y T 4 5 ¢QCXIWx¢XGlgXEII¢3QWVUE¢V6IX6su¢Q8GVVWQpwpGuWQpCuSE8wpGGhVgxXy yyB D Y YB D T F D 2 5B D T 7 d 5q 2B ` 2B Y R i 4 T d R 4 4 T 2B H Y T H d 4 2B YB F D D H ` i H F D D T F D YB D 2B 5 Hy D R XEGfXVUGe8EGV#W¢pXpXWp¢GI6Q¢X ¢gQIEXpG86IWVUWfEoX¢s EGppWY Hd 5 7 T t d H b H @ 5 W896Q8c 8¢GpQQV©6W¢QW6IXaXQpGC¢WIVgQQ9QIr¢X8I¢W8¢oE8s£l¥ i 4 T d R 4 4 T h H Y T H d 4 2B dB H F D q 5 H Y R T 4 H Y H 7 H F 4 Y 2 TB ` 2 T d ` T ô D 2B 5 r d H b 5 H ` T D 2 T b h T 2 T H b T F i H F D iy d T Hy Y 2 5B D T 7B 5 d T 2 TB DB 7 d H 2 5 h d 5 @ y T 2 n 68cu8¢o¢VpV¢a8gUIGW¢QXvQp8XWVdEcII¢VG#Qf8W8v¢UIi ¦ $ ¥ 5 D c 8dp9G 3 p¥ 6S XwI8XW6WI9IW¢ko¢Gj¢gEGoQpXuX8GVVWQpwfWUu¢aIIXEIW9IWaVIGD Y P 2B Y 2 5B Y Y H d H H F D q 5 i 2 T 5 D y T 4B D 2 H hB YB H bB D T bB d H h D Y d n T q 5 D n H 2By Y H F D D T F H D 5 2 H 2 @ 5 2 ÷ D 5 2 YB D n H 2By Y T 5 D 2B DyB R iDB R 2B D 2 5 4 y T 5y ` h H Y T H d 4 2B H F D i D H Y g W¢IhCspx¢IEuIIEXI GoXvEEIId3EISXWo8Q¢U8E8Q¢gQIEIG#SCWs¢5 YB i 4 T d R 4 4 T 2B 2 5B D 4 R h H d YB F D F 4 R EA6¢WIQ6¢)E)8XW6IpQWaXIWrI 7 cfç z I8wX6¢WIQ6¢C¢aQI8aWEwpI @ 5 ß îæ iy 2 5 YB i 4 T d R 4 4 T q 5 d H h d 5 Y DB D R ß ¢ y ¦ t w² Ý sgà x¢feÁ î ç ¥ $ ¥ o¢dpzæ º² ø ç ¦ 6 à ¦æ û ç ¦ à à ¦ æ Áù ä 3sgIG¢feýÿe² Ý ¥ ( 4 ¦ i o 2 H b B ` Y B D 1 d 5 D T d H 5 H 4 2 H d H g B h B d D T 7 `H h T Ü T q 68E¢vE£mQ8WVG6s8fQI6W6hIGV#¨¢vw¢5 8p8X8XWV#XgWIIVIEXIfpGD 7 d 5q H 2 5 YB 2 5B D T 7B 5 d T H 2By Y H F D T F D h H 2 5B D 2 H 7 H hy R 5 F Y D1 Y 2 5B D T d H 5 y T 2 5 ` TB hB d D iy 2 5 q 5 H Y R Y H ÷ T 7 h 2 T lUG9QI8EGoQ9s#XI8pWv£vQp8XWVG6s8p¢U88VIEWGI8h¢I6¢¢#dIVç hXB ß îæ Y ß ¢ y A t ² Ý sgà x8ÔeÁ î ç dpzæ $ ¥ º² ø ç ¦ 6 à ¦æ ûç ¦ ¦ à ¦æ Ôcgà ¡ g¢feÁ ù å² Ý ¥ ( ¦ ä 4 ¦ Hy 7 T H d 5 ¤ Hy T D d 5y i T Xp#¢pu8U¢Xp¢8Excmt T q 5 Y 2 T H 7 i h H bB d H h iyB Y T H H 5 Yy T 2 T 4 H bB D T bB d H h h 2 5 4 H Y H F D q 5 Y 7 d 5q 2 TB DB 7 d H ¢wI¢g9)oÅ68EQIAXWVgWE¢¬¢Q8GVVEQI¬I¢QQIWu¢w9W8A¢WE9W6f } p p u t r p Ø r w y r p u " xCuXsÂsu¡xg Ú Ó × Ò Ë É ÙÒ Ø × Í Ê Ê É Ì Ç Ó Ì Í Ì Ñ ÑÒ ÖÕ Ì Ë Ò ÓÒ Ñ Ð Î Í Ì Ë Ê É È u§fj«wÔzmQÔv«SÏmÇ % r p u wsuvj w t ß 6õ c ¥ .
H F t H 7 H F 4 Y `H h T Ü d H h d 5 r F D d R 5 q H F D h 2 T H 7 H F 4 Y H 4 2 H d H g B h h H d H D 2 H 4 d H h d 5 r F D d R 5 q h d T IuV#6IW¨¢vvQI83xGp8fIWdI¢#6IW6IQQsEI6WQWoQQQpW8£GfI8d¢Uh r 2 T D Y H F D d 5q Y 2 5B D Ty H d d TyB 7B Y F DB @ ` 2 5y T t ¥ H d R `B x¢fApGe8I8EGVXQ#¢XX9XÿWECÅI8E¢I pkI8Xm¤ Eð6G¢XIE¬%pÿ8EGVUSj 2B h H D D 5y YB c $ ¥ 2 5B D T R P xS ø ß î 6 Ókä ß î ß î 2B UfXY H 4 2B Y t h H D 4 H H H 5 D YB Y T t i 4 T d R 4 4 T d H h d 5r h 2 5 4 H Y 5 D Y H D T 7B 5 d QIEWQW66s9GXC¢ë¢6¢WIQ6¢fQI83p8Q6WGf6GV#XgWpI¢T ç c $ ¥ ovdpzæ ß î ß î 2B ëfEWY ä h vv h jlUGGVfò DT FD HD5 i 2 H bB oQ88` YE6swISQ8gºQUIx#kIWd8GlGQ8kQpQW6hIh 6W6GoQ66IW8£I86QeIG8U¤ B d H 7 R 2 H b T @ h H nB h 5 7 H F D d 5 D T d H 5 H 4 2 H d H gB h Hd H D 2 H 4 d H hd 5r h 2 5 4 HY H F D d 5 2 5B D T 7B 5 d T H F D q 5 i 4 T d R 4 4 T H F D q h8XWVdEcII¢wIWmV6¢WIQ6¢wIG§¢5 HIWVg9X6s px68g¢UW6¢IG§QGldEpgII¢Q IS68gwQUIx#pGXICfGD d R Y T H 7 T YB d H 7 R 2 H b T @ y T R D 4 T H F D Y H D T 7B 5 d T d H 7 R 2 H b T @ h H nB h 5 7 H F D F 4B F @ 5 H6W86IvpGmpxuv8XW6WIv6¢pvIWwEm¢gI¢ViQ6s px68gIWvQICW¢Qsp¢mEB H d ` H h H F D Y R F t 2 5B Y Y H d H D 4 T H H F D 2B Y d T H T t t d H 7 R 2 H b T @ H F D H d H F @ Y d T H T D H Y R T 4 H h H 7 T 2 5 Y YB d H 7 R 2 H b T @ h H nB h 5 7 H F t d H 7 R 2 H b T @ h H nB h 5 7 H F D Y I¢g6saQ9¢UmEjQ ISQ¢guQUIx#vIujQ6s px68gw6UEIx9IW§EB h SW6IC@ HdH F ç 9 $ ¥ 8dpzæ þh ¥v6 ä ä ä ä þ Ôs UfXog6 U8Qª's ëfEWoV9 ë¢Qù ûç ß î 2B Y ß î Y 5 4æ 6 ç ß î 2B Y ø ß î Y 5 4æ ß î ß î þ ß UfXY î 2B ß î ¦ þ sç e ß î þ 6 ¤45Wí ü ß î ¢ e G e þ æ þ ä ¢ ¤4 ¥í e þ 6 4 í Ý 6 4 üí AxÅ5WxÝ t ä íç ( Ý æ d ¦ þ ä í Ý 5 D d 5 D T d H 5 H 4 2 H d H gB h h H d H D 2 H 4 d H h d 5r h 2 5 4 H Y T t Hy 7 T H d 5q t iy T H @ q #xWu¢GVWQ8d6IQQsEI)6W6GoQ6wQpW8£I86QWm¢Xp#¢pw¢§¢EIIVdv£1 þ 6ä ß þ D H ` H @ t ß î ê ä í ß Hd H F 68wI S«uxuW6IC@ ç $ ¥ o¢dpzæ Y XB þ q 5 H b B D T b B d H h D Y d n D 4 T H H F ¢¢EWVVEW6IUC6¢pwIut Y H bB D T bB d H h d H F `B F 5 D Hy T 4By 6Q8GVVWQpuQI8EIGEI¢QXXpI¢T iyy T R P H YB YB Y iy T 2 T H F D F ` R 5 F Dy T t Y 2 5B D T 7B 5 d T H bB D T bB d H h D Y d n 2 5 t H d H F t H D T d D 2 H 4 2 5 EE¢USQXXxE¢UVIW98p8IG¢SQp8XWVdEcII¢f8GVVWQpU#¢p¢QI¢WVGWS6QI¢Q4 yXECwr U¢Q¢EWVVEW6I6GV#XgWpI¢¢GVWQ8 QpQW6hI3GIUQ¢E8vX6@ cIF yB @ H þ q 5 Y H bB D T bB d H h Y H D T 7B 5 d T d 5 D T d H 5 H 4 2 H d H gB h r H DB 2 n 2 H bB ` T yy H @ 5 HpX9¢puG6WQWoXVf86QIWXv£6Q ISQ¢gCIGEd$QICot þ W8uIW9E¢T 2B 7 T H 5 D D Y H d H D 2B q 5 H d 5q H d H F D YB D1 d H 7 R 2 H b T @ H F D YB H d H F @ 7 d 5q H F D 2B H d F 4B F @ t Y D 2 H 2 5 7 5 4 d HB d R 5 ¤ Y DB 5 D 2B h H Y 5 7 5 4 H h H 2 T 4 2 5B D 4 2 Rq 4B h 5B d H i d T d DB d T 2 XpC8QGo6I8#86v6XWp8UGGoX6W898Q6IusVg8EG6pIEIxXW6saVGGIWVi r e j  r p y w ã g j  y U7s{zD7uõ h ¥ e j i eg j  tgã y y ¡ k7v£×T t sÉ6 Ú Ò Ú ê É Ó É Í × Í Í Ì Í Ì Ò Í è × Ñ Ð Ð fxfóCfóÔCmÓ«SxÎ .
{ i s u § { z y s w u Ô « Ñ § e « § ¢ ® ¢ § Ô Ô ¯ § ¬ ¨ § t s ¢ ° « § ¢ y ® ~} rq § Ô 5« § ¢Ñ ¦ § ¢ Ò ¤® d¢ª ¢ Ñ ÉÔ Ô Ô ¯ ¢ © ¤§ q § b ¬ Gi gCExvs¦¤¢qd¬A¤qGÚd©£Adª¤£kÉxd©¤F¦qxfn¦¤¥z £v¦pQ¤vs¢ ¥FS£dqA£'%ªk¤)¦5Éx¤£X¤¬¨ pp A « Ô ¬ ª « © ç h $ ¥ ovdpzæ ø ä ç ß î ¥ 2B ø ß î s ëofEWY U EWY UjX 2B ø ß î 2B Y 4 æ " ß " ën86x4 " æ vä " ù î ç ß î ¥ Y 5 s Upë8Q4 ø ß î U 8Q4 Y 5 ø ß î Y 5 4 ø ¦ h v6 ä íç ( zsÝ 7 æ h 2 I¢T YB d H 7 R 2 H b T @ h H nB h 5 7 ` 2B h 2 5 Y H d d 5 4 H F XfQ ISQ¢cQIEIx#IEII8W6W8QwIut " û í Ý Ôç ¥xÆø Wxæ üí Ý ø ç ¥xÆø í Ý ß Wxæ " f¥¬h5Wx4 " jxGä " ù î üí Ý ø ç4 í Ý ø 4 üí Ýæ ø í Ý ¦ û í Ý 3ç Ax)6 ß æ ç ¥x)6 Wxæ ~fAxÅ5Q¡¸ù î üí Ý ø í Ý üí Ý ø ç4 í Ý 6 4 üí Ýæ 4 ¦ ä íç ( usÝ æ 2 H F D t ¥ ø ê 5 D ¥ 6 ê 7 5 dq ` 2B h 2 H D H Y H 7 H F 4 Y 5 D D Y H d H D 2B d R 5 D 4B d D Y H 6IGI©wW¬½Ï89pXII6GxfQ9QIGC6WQWoXuI8XWW6Wd H @ q vv£1 Q¢GVWQ8 XG#9xuXwzç 7 ©I¢8GVWQ8wEWW69#xEGo¢A¢Eaç W6IC@ d 5 D T d H 5 4B d D H 7 7 i Y T YB í (æ h 2 T d 5 D T d H 5 4B d D H 7 7 i YB D 2 T 2 T YB í (æ H d H F í (æ ø í ( zç 7 ÿzç æ ä í ( Cç æ 7 d 5q H F D 2B 2 H D DB d @ H 2 T 4 Y d 5 D T d H 5 H 4 2 H d H gB hr H DB 2 n ty T d H 2 H ` 2 W¢upGeXQWWWCs¢QuW8WVG6s8u6IQQsEI£GIUpz¢G6IQ¢31 l t H B Y 2 H bB ` T q 5 F Y H 7 T 2B Hy TyB T b j l ë m¡ ¢6X68E¢v¢W6#XfXI¢XX¢VgT ß î H ` 2 T d d H 7 R 2 H b T @ H dB D 2 H H F D d 5q h H by 5 Y H d YB 2 5B D Ry 5 Y H F D q 5 D 2 H 2 5 7 5 4 d H 7 R 2 H b T ¢I¢Gu6s px68gwWEGoQ pGf¢k688W6WE8EGpE8W pGj¢o6I8#86uQ ISQ¢c@ 2 H bB ` T F 4B F @ F DB @ i 4 T d R 4 4 T H F D Y H hB b 5 d d H 7 R 2 H b T @ h H nB h 5 7 H F D d 5q 2 5B Y Y H d 68E¢¬XIC)WECr¢GpQQVeIG 6IXxcWp96swISQ8grQIEIx#dIGw¢¬¢XWQIH Y H 7 H F 4 Y Y R 5B d T b d 5q d H 7 R 2 H b T @ h H nB h 5 6Q9QIuI8X¢V8f6swISQ8geQUI 0 0 0.5 1.5 1 2 3 2 Central nd κ∆ x 1.5 th 4 Central 4 Pade th 2 2.5 3 ¡ .5 1 Û ¥ H d R `B I uWI8Ev¤ Ú Ó × Ò Ë É ÙÒ Ø × Í Ê Ê É Ì Ç Ó Ì Í Ì Ñ ÑÒ ÖÕ Ì Ë Ò ÓÒ Ñ Ð Î Í Ì Ë Ê É È u§fj«wÔzmQÔv«SÏmÇ κ*∆ x 0.5 2.
¨ § c ¬ § Ò ¥ ¬ uddAÕ¦¥£q¦ ° ¬ Ñ ¨ Sds§ ¸q© 642VQÐ c §§ ¯ e ¬ ¨ ¢ u © ¨ª Ò© © Ô¢ § § ¨§ § Í ¦0dd¬d½xsda¦¤A xz Ð ¦S¦ÏU¤£¤¥¤¦« s v " ß î h if ¢ ëfEWY fä Þ " 6 ä 6 ß î 2B Û ç Þæ d 5q h H 2B T D 5 YB ã f¢6IXVp8uE© ` 2B @ 5yy 5q H F D t 7 d H D y TB D T Y H F D 5 D h HBy T H d T Y H 4 2 H d H gB h h H d H D 2 H 4 d H h d 5 r h 2 5 4 H Y q IXCcEX8uIGUW6Gh¢XGVIWwIG9GeQEXIIVw¢6QI6W6hIeQQGo6QQI83p8Q6Wj£1 ¤ ¡ ¢ þ æ ä çÞà ßæ sç ¡ jWvÝ Þ Y T 2 5B D Ry 5 Y D 4 T H H F D Y H bB V8XWpX8¢puIWC688` ¡ ëjëGoE9pXGGEWIpWI¢uvf8 IXx8 ¥ P 5 D 2B ` 2B D R DB D Y R Y h 2 T ç Þæ d 5q ` 2B by 5 6 £if¼ä ç ¡ pzæ ¥ " " ã H F D Y H n YB D T Y ç Þæ H d H F aIWC6UEGVWW6IC@ þçÞæ ä çÞà ßæ sjWvÝ YB d H 7 R 2 H b T @ F DB @ 2 5B D 4 2 Rq 4B 2 5 7 d T F T i h H D TB DB 2 E6swISQ8geGECe8EGIIEI89W¢Uo6GVXGEpXB 2 5B D Ry 5 Y T D T F ¢XGX8VUWD p¥ 8XW6QxÄ¢WX¢QQC GD $6 8eIXIpQGxrXVd8p¬85 2 5B D 4 H 7 5 dq yy T 4 H 5 7 5 dq ` 2B h 2 H D H 2B T 7 5 h T 2 ß Þ ¡ ä ø Ý Ý 7d 5 8q H F D 2B 2 5B D T R P H 2 5B D 4 H b 2 5 4 d T H 2By H F D 2B T ` T H 4 2 5 d H hB Y 2 5 2 5B D T D H d d H D 2B y T 4B Y i F pGE8EGVUS6e8XW6Q¢o8Q¢gpXXaIGXVo¢6I86IXI88EGVG6WpWQWoX§¢gEWxoIdT 2 H bB ` H 2 T 4 Y d 5 d d H H F D t 2 5B D T R P H 2 5B D 4 H b 2 5 4 d T H 2By H F D q 5 D H D 2 5 4 H F D 21 Y 2 5B D T 7 68E¢#r¢Qa8WQ#IG8EGVUS6)8XW668o8QVgIEX9IGV xWS¢Q9IGA3#6I8XWV#XB r 5 d T H 4 2 H d H gB h ` 2B Y Y H Y Y T d 5q y 5 5 D y Rq H Y R T Y H hB b 5 d d H 7 R 2 H b T @ h H nB h 5 7 H F SgWpI¢QpQW6hIeIEWW6W¢88xGvp6WpduQIExsIuQ ISQ¢gQUI9Iut ß îçs U8Q¬ø æ ß î Y 5 4 ß î 2B Y UfEWor¥ ç ä h £ Y H bB ` F 4B F 688EIC@ e þ á6 e þ æ ß î ¥ ä e þh ø h £mv6o¬ø G e þh ¥£ Y H n YB D T Y H 7 H F 4 Y YB F D d 5q d H 7 R 2 H b T @ h H nB h 5 7 H F 6UEGVWa9QIuXIW8f6swISQ8gQUIx#wIut ç4 í Ý 6 4 üí Ý f¥Å5xæ ß î ¥ ä 4 üíç (æ ø íç (æ ø 4 íç ( 5WhÝ ÿjzÝ UÆAÝ æ i 2 H b B ` Y B H 7 H F 4 Y `H h T Ü d H h d 5 r F D d R 5 oQ88CXu9QIa¡VvCQpW8£GfI8q H F t d 5 d d H iy H dB D 2 H ` 2B H D 2 H 2 5 7 5 4 i d T 2B ` T 7B H F D F DB @ t Hy 7 5 4 YB d H 7 R 2 H b T Iu¬68WQÅQWEGoQkIEQ9S6I8#¢QA¢IX8¢#XIWÆWECjsEI#¢QX#Q ISQ¢c@ h H nB h 5 7 H F D t D 2 H 2 5 7 5 4 4B d D H 7 7 i Y T Y H h Ry 4 2B d 5 D T d H 5 H F D 2 H F y T H d iy H d R YB d H 6UEp9IWxQoQI898QXG699faQIIEQIEv¢GVWQ8IWQIª§¢QW9E6WIpvEQ r 7 R 2 H b T @ h H nB h 5 7 H F D t ç h H d H D 2 H 4æ 4B d D H 7 7 i YB D 2 T YB d 5 D T d H 5 H 4 2 H d H gB hr H DB 2 n H F D 2 H F ISQ¢c6UEIx9jIGlë6WQWoQQuEWG#9xEGo¢E8GlGQ8j6IQQsEI£GIUIWw6Iª ¦ I Ú Ò Ú ê É Ó É Í × Í Í Ì Í Ì Ò Í è × Ñ Ð Ð fxfóCfóÔCmÓ«SxÎ .
h H dB R P H d H d T Y d 5 D T d H 5 i d T h 2 R 5 hQWEISQw¢CW¢GVWQ8eVUII8 y TB 4 H Y 5 2 t Y 2 5B DB h 2 5 4 i d T h 2 R 5 4B h 5B d H q 5 H Y T 4 H F D 2 1 2 5B Y R gB h h 2 T 2 5B D 4 H b 2 5 4 d 5 ¢6QWdpUQI8EGII86¢UpI8XIxXQ¢aW¢QaIW38EWIsEII¢e¢XGQ¢S¢QC8q Y 2 5B D T R P H y H h 5 7 d R 5 d 5q h H h H H 2 Y d 5 D T d H 5 i d T h 2 R 5 H F D d H hB Y 2 5 4 H @ t 2 5B D 4 H Y YB F D 2 p8XWVUSQ§QIx9I88eQpQQpW¢GVWQ8f¢UIp8saIGfQIEWI86 vs8EGQWfXIW31 QQEW¢III8dIWal VIGrIVE¢#8IdpG¢u8EW6GoX#IGAXk8XWV#XgWIIVQpQW6hIh Y HB d T h 2 R 5 H F D D T D T F D h 2 T 2B T 7 5 h H F D q 5 d 5B d H D 2B H F D 2B 2 5B D T 7B 5 d T H 4 2 H d H gB H F D F D 5 Y H D T d 5 d 5 4 2B d 5 D T d H 5 H 4 2 H d H gB h B d D T 7 T pGªG¢eQWVG8W¢QIE8GlGQ8kQpQW6hI«WWV#¬t p¥ ÿ¢XGQÿEB QWI6WEIi ¥ 2 5B D 4 H 2 h H Y Y R 4 YB h Y ² ± ¹ À ´ · ¶ î 6weeëÃ
¤ » w´
² À Ã » ´ À ± C 9aj
± Á ¶ ± À ± ¿ CejÅ6¹
½
¢® £
88A8XW66X6WIQXICQpGpXCXIum¢IpGEEI9¢XB ¦ ¦ 2 5B D 4 H 2B h H Y Y R 4 YB h d H F D d Rq YB YB F t H h R DBy 7 T 2 d 5 d d H H F D 2 5B D d 5 4B d D H 7 7 i Y H F D h 2 T h H H Y 2B d 5 d d H H F D Y H 2B 7 d H D H h d 5 D T d H 5 H 4 2 H d H gB ¢WW6CIG 8EGf8sXWW69#xfpGI¢dQQWEj8WQCIG6IX9W6GIj8GlGQ8CQpQW6hIh y TB D T Y H F D q 5 2 5B D d 5 4B d D H 7 7 i YB D 2 T H F D Y R F t 2 5B D Ry 5 Y H F D q 5 H h R DBy 7 T H F D 2B d 5 d d ¢WVUIGVd8EG¢sXG699fXWSVIWISuv8EGpE8WIG¢CIpGEEI9¢IWdXv8WQH 2 T 5 D Y h T Hy D d T i d T 2B ` T 7B H F D HyB F @ t h H H Y 2 5B D T ` T 5 d H F D 2B d 5 d d 2 T 5 D Y h T H ¢)G#p¢gEe¢IÅ¢IX8¢#XIGXEIChQQW©8XWVo¢I8WIIG«Ed8W6H ¢ÅG9I¢QXy d H 7 R 2 H b T @ h H nB h 5 7 H F D q 5 D d T y T H d H F D D T F D H H Y H @ t ¦ ¥ P q 5 7 d 5q H F Q IS68g6UEp9IG§¢Cf¢U¢QWIWuVUWu6Wuv8p pëëj8¢§8IGD 8WU¤ 7 5d y H b Hy d 5 d d H H 7 T Q¢6E¢WW6#VGY H F D H b HB F 4 T 5 D iy H bB D 4 H Y H pG86EI¢kW)E68EW66sQWd ¡ ªIVqeWEISQW9QI¨¢v6IW¢3WI8pGD ¦ h 2 T $ ¦ H d B R P H d H 7 H F 4 Y `H h T Ü d H h d 5 r F D d R 5 q H F h 2 T H 7 H F 4 Y h H d H D 2 H 4 d H h d 5r F D d R 5q D 2B 5 r $ H F t D 2 H 4 d H ¦ IV#6IWk6WQWoQQfQI83xGp8CSE83sÓpugoQ6W6s ¡ VUGWQE¢eQ6s WVUI 2 T F D Y Y Hy q 5 h H H Y H Y T F 2B d 5 d d 2 T H 4 R h 5 d 5 Ed8W6H ¢eQpIIrWD 0 ¡ ôQWEISQe#6IW6IQQsEIª6WQWoQQdQI83p8Q6WY c Y H dB R P H d H 7 H F 4 Y H 4 2 H d H gB h h H d H D 2 H 4 d H h d 5r h 2 5 4 H H F D t Hy 7 T H d 5 ¤ y H b Hy h H nB 4 H Y T @ 5y H Y d 5 d d H H 4 R h 5 d 5 D h H dB R P H pG¢Xp#¢p8UÅ686E©6UE6QWeµcEQW8WQ6IIxWIAGÅQXIS6Wd H FD q 0¡IG¢5 Y 7 d H D 2B h H Y Y H d H H 2 T 4 H 7 H F 4 Y T q 5 i 4 2 HB 4 ¨ H ` 2B by 5 Y H d H F t ß î r 9W6G)E«QW6WIds¬¢Q9QpW¢ÅIQEQ96epXxE8QeIu Uo oÅQ88` i 2 H bB YB H by 5 Y H d YB H b T @ 2 H bB ` T F 4B F @ i E)hQ¢E8QaEe8gQ88¬EICkoç ¢xX¢r#UuUlU¢w6swISIut æ x } } 1 è } x y q 5 d H 7 R 2 H F iy Y 5B b H d h H d H hB Y 2 5 4 Y H 7 H F 4 Y H F D d 5q h H H Y H Y T F y T 4B d H 7 R 2 H F D Y @ 5 F Y $ ¥ H d R `B 8EIR8EWI6WQpXWp8QQ9QIuIGm8QQWC¢UI¢gEWQ9ISuIWCcI%uWI¢Xv¤ 7 d 5q y T 2B `B d W8ë¢UE8X85 Y DB Y Y 5y iy T R D H b H Y 2 H 2 5 7 5 4 d H 7 R 2 H b T @ D 2 H d H gB h i 2 T 7 q 5 ` 2B D YB Y 2 5 4 7 d 5q H b T WEuQH8XAXy¢UWo2Q8aWDoQI898Q 6swISQ8gao6WQsEpeo¢d¢pXGfXI8Q¡86¢c@ T t Dy R H d T uQEIYQrYfÿV8EW6HsXII¢5I#EÏC¢UX¢XW¢IWÆ8p8IG¢u¢XWQWEId6QIIxWWoXB i H bB Y d YB h 2 2 YB ã Ü y T 2B `B d 5 H F D F ` R 5 F Dy T t 2 5B Y d H YB h Y H 4 R h 5 d D 2 2 5B D T 7 5 d T y T B d H 7 2 H F t d 7 R 2 H b T @ H F D q 5 2 5B D 4 2 Rq T Y 8XWVdEBcII¢¢g4EWQ9IRSIWDmQ6HswISQ8gIG¢e8EG6pIuXB h ë6IX8ECcEWexGD H 4 2B iy @ 5y Y 5 5 Y6GV8¢U85I«85EGpE85W¢QXWd6#px2eHpGD¢EI9¢pdEIG98¦ l r h fQIE ¢EE¢gEWQ9IS2 H D T ` T d 2 B D Ry Y y T 4B H 7 R F t Hy 7 T H YB F D d 5q H 4 2B iyy T 4B d H 7 R h D T ` T 5 YB 5B D 4 Rq 4B 2 7 d T 6HGVoVU8WdpX28EG6p2IvXp85#¢UFuTaFXBICVaQ6sIGEa6QWW¢Up¢¢QXW6#pxIut 4 F @ D T h H H Y H F D YB h H H Y H Y T F y T 4B d H 7 R 2 H F ß î U ä h ß î 2B UEWY t Hy 7 T H H b 5 T H F D d 5 ¢EI#VIu8csVapG8U¤
i d H 7 R 2 H b T ofQ ISQ¢c@ h H nB h 5 7 H F D 5 D h H D Ty H d YB F 4B F @ t h H H Y H Y T F ç h H nB h 5 7 d 5æ y T 4B d H 7 R 2 H F D Y 6UEp9IWwG#QGlQvEeXICQ6sW¢IIUQUI98v¢gEW6#ISfIGXB h W6IC@ HdH F ç ¦ ¥ sp pzæ r 5 d T y TB D T Y H F SgWpI¢ë¢XGVUCIGD
h
ä
h
¤ ¡
¢
þ æ ä ç Þ à ß y T 4B d H 7 R £ç ¡ kæ ¢QXW6#px2 Ý
2B T D 5 H @ E¢8upt ¡ sëjëGoXIEGpWEWIII¢pEI¢8XWV#XB ¥ P 5 D 2B ` 2B D R DB D Y R Y h 2 T ç iy 2 5 2 5B D T 7 7 5 dq d 5 d d H H F D ` 2B d H hB Y 2 5 4 H d T H @ H 4 2B Yæ iy D 4 T H ã YB F D ` 2B by 5 ¢Wj8WQIGIEW6IXI8QC¢vuQIEWG¢pvXpGIXx8
Ú Ó × Ò Ë É ÙÒ Ø × Í Ê Ê É Ì Ç Ó Ì Í Ì Ñ ÑÒ ÖÕ Ì Ë Ò ÓÒ Ñ Ð Î Í Ì Ë Ê É È u§fj«wÔzmQÔv«SÏmÇ
¹
ç ¥ S¢ pzæ
¡ ¡
´´µ ³´´
¹
à
··
ß î ¦ w6
ä y A t º²
ä Ý ·· ¶ !6
ç ¥ ¥ o8 pzæ
··
¦
·· ¶ y ¦ t º²
ø ²Ý D»
¦
¡ ¦
¦w6 ¡
´´µ
¦ w6 ¡
³´´
ß î ¦ w6 ä
»
F DB WEC@
ä ²Ý £ (
2B T D 5 H @ Y H 4 2 H d H gB h h H d H D 2 H 4 d H h d 5r h 2 5 4 H Y F DB @ t Hy 7 T H d 5 ¤ i d T h 2 R 5 YB F D d 5 X¢Gp8wCQ6IQQsEIQW6GoQ6f6IW¢3I¢QQGEC¢Xp#¢pu8Uv¢¢III8uXIW8q h H dB R P H d YB D 2 H 7 D T H d D y TB 4 H Y 5 2 2 H F D t 4 í Ý h 2 5 i H h 2 H D H D 5 2 Y H 5 h 2 5B D T 7B 5 d QXIS6WwX o6#WVgG¢6QWIrQIWv¥¬I¢86ÅI6Gxw¢Iw6p8EGV#XgWpI¢T H 4 2 H d H gB h d 5B d H D 2B H F D Y T ` 2 5y Y T D T F D d T Hy 4 YB D1 i d T h 2 R 5 @ 5 q 2B H F D D Y d n d H hB Y 2 5 6IQQsEI§8XQWSEIWv¢I¢X¢VUWv¢gEQmXj£m8VUII8ÆcUpXIGfWU6IXI8 h H nB 4 H Y YB ä Ý h 2 6UE6QWuXC!I¢T ç pzæ ¥
Æ aCà û Å Ý
xS
à
C£Ý Ïe² Ý Ý à 4 ù ä
YB Y 2 @ 5 2 ÷ 2 R q 5 d 5 D 4 H b H F D Y R F t i d T h 2 R 5 EfICcISIpv¢8W6Q¢IGCISu8VUII8 @ 5 q D R 5 H F D YB i d T h 2 R 5 h 2 T Fr D F `B d H F D h 2 T t i d T h 2 R 5 @ 5 q 2B H F D YB i d T h 2 R 5 h 2 T cUWp¢IGEVUII8eI¢I3o8EWIWkp¢8f¢UII¢såcUpXIWX¢UpI8Ap¢UF r6EIGpxW¢¢EGW8X QQW98g IGVUG 9IW¢9v WQq¢¢UpI8 cUp85 Dq Hy H F D Y R F D H bB DB Y 5 YB h H H Y H b T @ H F D D T F D H 7 RYY T H @ Hd H id T h 2 R 5 @ 5 qD R H F D D T h H nB 4 H Y YB 2 5B DB h 2 5 4 5 2 HyB F @ t i d T h 2 R 5 @ 5 q 2B H F D D T 2 H bB ` YB 2 5B DB h 2 5 4 i d pGVAQIEQ6sX8XWEII¢QeIaXXpC8VUII8cUpXpGVA68E¢X8XWEII¢Qf¢T rxII8j6EIXXapVW¢Q6GVIG93j¢x3«cUW83cUpE¢8uEIEW6sQpGE6fdVg4 h 2 R 5 D Hy F 4B dB ã T t H Y T 4 d H D D Ty H F D 21 H iD @ 5 q D R 5r @ 5 q 2B q 5 d 5 4B h 5B d H d H F DB H H 2 T 2 5B D T R P H 2 5B D 4 H b 2 5 4 d T H 2By H F D d 5q Y 2 5B DB h 2 5 4 i d T h 2 R 5 H F D t 8EGVIx68EG668o86VgIEXwpGC8p8XWEII¢QVUII8wpGëp¥ 8EG66eW#IEWQ6C 2 5B D 4 H 5 D ` 2B d d Hq H
i eg j  â ¬ « i eg j t p © i h£7uX&kÚhªke 2susëugö à r  p ig ! p u
Y H 7 H F 4 Y Y R 5B d T b d 5q h H H Y H Y T F y T 4B d H 7 R 6Q9QIuI8EW¢V8QQWwWVUIhVgXQ9Ifò
0 0 0.5 1
¨ s
Û $ ¥ H d R `B x%puI8Ev¤
¥ o
Ú Ì Ò Í É Ö Ó è × § Ë É Ú Í × Ë É Í Ì Ê × Ì Ç Ó Ì Í Ì Ñ Ñ Ò Ö Ð ¥Ð ÔfmÓVefCCCó fvÔ«o¦xÎ
a* _ a
0.2
0.4
0.6
0.8
1.2
1
2 Central
nd
4 Central
th
κ ∆x
4 Pade
1.5
th
2
2.5
3
H dB R P H XIS6Wd
q 5 Y @ 5 d 5 @D D Y Ty H F D d 5 D T d H 5 d 5B d H D 2B h H d H D 2 H 4 d H h d 5r F D d R 5q T q 5 H Y T 4 H F D 2 » VCcWv3vf¢fIG8GVWQ88XQWSEQW6GoQ66IW8£GfI8a¢WVgIWd31
2 5B D T 4 nB h 5 ¢XGlgUI97
¹
ç c ¥ o pzæ
· ·· ··
¡ ¡
¡
´µ ´´ ´´ ³´´
¹
ß
î ¦ a6 ä y ¦ t º² à ··
· ·· ·· ·· ¶
¦
¡
6
´µ ´´
¦ ¡
¦ w6 ¦
´´ ´´
¦ w6 ¡
ä Ý ·· ¶ 6
¦a6 ¡ ¦
³´´
ß î ¦ w6 ä
»
F DB @ d 5 D T d H 5 H 4 2 H d H gB h T Y H 4 R h 5 d YB F GC8GVWQ8uQI6W6hI96QIIxWpEIut ç 9 ¥ 83 pzæ ç4 Å Ý 6 Å Ý )Ùæ ß î ¦ ä
Å sÝ æ ç (
Û h H Y R H 2 T 4 Ty R 7 d 5q h d T @ ÷ 4 T d H h d 5r D Y d n ` 2B @ 5yy 5 hQWpw¢g XI 8¢o¢I6IW¢3fWUaIXCcXE8q HpGjQ85WVG6Hs¢ 8XdQWSEQI83p8Q6WeÆGEª)hQWpp p W8e6IQQsEIÆW¢@ F D t d D T d 5 d 5B H D 2B d H h d 5r h 2 5 4 H Y T F DB h H Y R H D Y R 7 Ty R 7 d 5q H 4 2 H d H gB h h d T r ÷ T 84¢U i2 lT6WIRu¢Ip¢gIG5WxuQXIS6WXICxQ8WVG6s¢a6IQQsEI£QQGo6Q4 ä ê D h H Y H D 5 2 2 T 4 t 4 ü í Ý Y H dB R P H d F 4B F @ t d 5 D T d H 5 H 4 2 H d H gB hr h H d H D 2 H d H h 5r h 5 4 H Y F D Y R t 4 Å Ý D R 5 T 2 5B D T 7 d 5q 2B 5 2 F DB 6IWd8£I286QeHIG ISFuk5ü 8sVr8EGV#W¢pXeIrGEC@ DpI VT ß 5Ý WV#XB H h 5 2 D r HD T 7 r 5d T Y R 7 SgWpI¢wDI Hd6UnE6QHWYuXBk8EGEpI8Q¢UpI8Im8f¢UII¢s sIWp89IWai h H B 4 Y 2 5B DB h 2 5 4 i d T h 2 R 5 5 2 t i d T h 2 R 5 @ 5 qD R 5 H FD D H D T d R 4 4 T d H h d 5 r F D d R 5q iyy T 5y ` YB 2 5B D T 7B 5 d T YB F ¢WVGI6Q¢C6IW¢3WI8X¢U¢X8uE8XWV#XgWIIVfEIut
¹
à
ç ¥ o8 pzæ
ä y ¦ t º²
¹ ´´µ î ¦ ß ¡ ³´´ w6 ¦ ä Ý ·· ·· ä Ý ¸¶ !86
··
¦
¦ a6
c
¡ 6
·· ¸¶
c
¡ ¦
c õ6
¦
´´µ c õ6
f6
³´´
ß î ¦ ¦ w6 ä
»
F DB @ ¥ ¥ P q 5 7 d 5 WEC 8 pj7¢§8q H F D Y T F d 5 D T d H 5 H 4 2 H d H gB h ` 2B Dy R Y H d H F t Hy T D d 5y i T t T ` 2B Y R h H bB d H h iyB Y T H YB F 4B F pG¢U8GlGQ8QI6W6hIuIXWEIQWIuVXIVj8XxgmwaIXId68WQp EEW¢QXdEIC@ ç $ ¥ 83 pzæ ç ß î Å6 ¢Æh£¢¬Ý æ 8 Ý Ý ø 4 Ý ¥ 6 ä 6 ¦ ä 4 sÝ æ ç (
Û ¦ ä ê D T d 5 D Td H 8lC8GVWQ85 d H h d 5 r h dB F D ` 2B @ 5yy 5q H F D H Y R 2 T 4 H @ t Y H 4 2 H d H gB h h H d H D 2 H 4 d H h d 5r F D d R 5q F DB @ t Hy 7 T QI83XIWaIECcXX¢IGWpVguI6Q6IQQsEIQW6GoQ6QpW8£GfI8GCp¢EI9¢pH d8U ¢9QpW¢XW6GoXpGj¢lUGvVUSQ§XXC6¢WIQ6¢vVU8X¢ IGeIV¢9QI8EWQWoXB 5 ¤ H 7 H F 4 Y d 5B d H D 2B H F D q 5 D T F D y T R P H yyB @ i 4 T d R 4 4 T y T 5y ` H F D h 2 T t H 7 H F 4 Y d 5B d H D 2 H F D q 5 D T F D 2 T F D } è } } YB F 4B F @ i 4 T d R 4 4 T q 5 d H h d 5 2 T H b T F 2 T 4 t } { } pG¢9VIG©VUGwlXexlehlX©XIC¬¢GI6Q¢¢QI8«V8gU©¢gdUc ¢oxl5¢WdUd¢CcIS68GlGQ8#¢Iv8¢GpQQV¢6IW¢fWVI8WpI¢pGD ¢ À y } { ¢ T Y T 2 @ 5 2 ÷ t d 5 D T d H 5 2 T F 4 R i 4 T d R 4 4 T q 5 d H h d 5 H D TBd 5 d T H F ` 2B b T F HyB F @ t ¦ 6 ê 5 D iy 2 5 Y h 2 H D H F 4B F @ IExcIXXpC§8ÿë#GEI8aII6GxXIC¦ l)QXIS6WuXu8WVG6s8woQQsEIT ä ê D T h H dB R P H d YB d 5 D T d H 5 D 2 H d H gB h í Ý q 5 H Ry T b T Y H dB R P H d ¦ ä ê D T 2 5B D T 7B 5 d H 4 2 H 2B T 7 5 h H F D H hB Y D R 5 YB F 4B F @ 6IQXVd8pwpGIEWG8uXXpCIt Axm¢IE¢V f6WEISQWuåwVe8EGV#XgWpI¢T H F D 2 H F D t $ ¥ P 2 2 H bB ` d 5 D T d H 5 d 5B d H D 2B d H h d 5 r F D d R 5q H F D H Y R H @ qB t d H b H @ 5 pGªQIWCpV%uëgEB Q¢E8¢GVWQ8d8EW6GoX6IW¢3WI8pGWpAvEQQ¢6vcf Ú Ó × Ò Ë É ÙÒ Ø × Í Ê Ê É Ì Ç Ó Ì Í Ì Ñ ÑÒ ÖÕ Ì Ë Ò ÓÒ Ñ Ð Î Í Ì Ë Ê É È u§fj«wÔzmQÔv«SÏmÇ 8
ç 9 ¥ SVdpzæ
W ß 4 ü 5Å W ß V iä V Ý Ý
YB D T F D t ¦ XVUWp8ðø ä ê D T h H nB 4 H Y YB AV¬6UE6QWwX©ß ÚÝ VUWd#pWWVe8EGII86rf¢UII¢s)p¢dpQfw8U¤ r D T F D H 7 R Y Y T H @ t 2 5B DB h 2 5 4 i d T h 2 R 5 2 2 T 7 R H ò T d 5 ¥ P 2 9 pëëjXB 2 H bB ` d 5 D T d H 5 H 4 2 H d H gB h H F D 5 D ` 2B h T Hy t Y HB d T h 2 R 5 d T H 2 h H dB R P H d H d T Y 2 5B D T 4 nB h 5 7 5 68E¢§¢GVWQ86IQQsEIIGGuIEI¢gEV6QEW¢UpI8m¢QI 6WXpx6WW¢I8XWVgIEIx#fI2 ç ¥ 9 ¥ o¢dpzæ ç4 í Ý ø í fAxÆvÝ 85Wxæ 6 4 üí Ý
ß î ¦
ä í sÝ ¥ æ ç (
d 5 D T d H 5 d 5B d H D 2B h H d H D 2 H 4 d H h d 5r h 2 5 4 H 8GlGQ8¢XW6GoXQQGo6QQI83p8Q6WY HpGGª¢8c¢AQWI6WXpe8XWVUSQk8EG668o86 IG8¢UpI8åcUX IW¢QIEXy F D F DB H b 5 T h H Y Y R 4 YB h 2 5B D T R P H 2 5B D 4 H b 2 5 4 H F D d 5q i d T h 2 R 5 @ 5 q 2B H F D Y T Y H 2B H 7 T Y H F D ` 2 5y T Y h H H 4 5 d 2 5B DB h 2 5 4 i d T h 2 R 5 D Hy F 4B dB ã T q 5 D 2 H 7 D T H d D H F t Y 2 5B DB h 2 5 9¢WuIGI¢X¢I6QQxWp 8XWEIp8Qf¢UIp8sm6EIXXuh¢joQ9GVQWWCpuvQp8XWEII¢Q4 i d T h 2 R 5 2 2 T 7 R H ò h 2 T D Hy F 4B dB ã d H hB Y 2 5 4 D Y R 7 H @ 2 5B D T R P H 2 5B Y R gB h H F D ` 2B by 5 Y 2 ¢III8I¢#IQI¢6XpXWEuQpXWp8QCfI v8XWVUSQ¢XWphIIW IXxE¢We31
i eg j  â ¬ « i eg } â ng l p u h£7uX&h£¡uÏo2Ùugö
¨ o
9 ¥ P 5 D y T 4B D 2 H hB YB F 4B F pëjëGh¢QXWS6IXuEEIC@ ç dpzæ 9 ¥ ä ä ç4 Å Ý 6 4 ü Å Ý ¡)855æ ßî ¦ ä
ç4 Å Ý 6 4 Å Ý 6 Å ÔÅ7)ÙÝ æ
ß î çf4ÅÝÅ6ÙÅ5Ýæ ¦ ßî ¦
Å sÝ æ ç (
Y H bB Q¢E8` H h 5 2 D T h HBy T d 5 D T d H 5 H 4 2 H d H gB h h H d H D 2 H 4 d H h d 5 r h 2 5 4 H Y H F D 5 D 2B YB F D ` 2B D R DB D Y R pIVeQEXIp¢8WVG6s¢QI6WQsEp6WQWoQQ6IW¢3I¢QQIG#GoXEIG#IXWpGGpI ç ¦ 9 ¥ sdpzæ
4 Å Ý 6 Å Å2Ý j5Ý ä 4 ü Å
Û 4 ü Å Ý 5 D 2 5B D T 7B 5 d T d H h d 5 r D Y d n ` 2B @ 5yy 5q H F D Y H bB ` F 4B F 59Wd8EGV#XgWIp¢QI83fWIwIXCcEX8IWC688EIC@
ç ¡ dpzæ 9 ¥ ¡
ä 4 Å Ý ø Å ÆðÝ 85ÿ55Ôç 6 4 ü Å Ý ä 4 ü Å Ý
® ø
4
®
pfæ 6 ¦
2B T D 5 H X¢Gp8w@ d~GEC¢Xp#¢p8U8ºë°X8EGD ä F DB @ t Hy 7 T H d 5 ¤ ¦ 6 YB 2 5B r T 7B 5 d T H F D q 5 d H h d 5 H F D h 2 T 4 ü í Ý ä í Ý ® i h H 2 n H h d 5 D T d H 5 DqB F Y H F D Y ® H d H F l#XgWIIVIGVQpW8pGdIV5Wxðw¯oQpU6I8WVG6s8EIIWEB W6IC@ ç h ¥ o pzæ ¡ ä 4 ü Å 5Ý ç
Ü 4
® 6 ¦ Vpfæ
Û hB d ` h H 4 T YB R P H 2 T 2 5 d H h d 5 i d T d DB d T 5 D T D T hXW¢Q6¢UXISQ¢8CQpW8d¢WGEpW¢9G GVUh d8XQGoEvIWó8WuQWV¢s¢WGxvsGh5§CcEX¢Cp W8pXCcXE8vIu¢XGl8¢GWxH 5B d H D 2B H F D 7 5 dq h H D Ty 5 T d D H H 5 D 4 ü Å Ý Y @ 5yy T Ty R 7 d 5q ` 2B @ 5yy 5q H F t 2 5B D Ty 5 T d D H 4 T Y q 5 Y 7 d H D 2B YB Y H 7 H F 4 Y i d T h 2 R 5 q 5 D 2 H 7 5y H b H h H F D 5 D F 4 T 5 d T d H F D 5 2 6¢UWv¢9W6GXXwQ9QIVUII8v¢wS6#p8X686pIGeGe¢8WIIVa6IGVIi
$ Â
Ú Ì Ò Í É Ö Ó è × § Ë É Ú Í × Ë É Í Ì Ê × Ì Ç Ó Ì Í Ì Ñ Ñ Ò Ö Ð ¥Ð ÔfmÓVefCCCó fvÔ«o¦xÎ
ç c 9 ¥ ovdpzæ
ç zdø ß îæ
Ê¤45üÅ W ß V ß ð î Ý
ø 4 Å Ý 6 Å Ý )Ù8
È ¥
¦
ä 4 ü Å j5Ý Y H bB ` 4 ü Å Ý d 5q ` 2B by 5 Q88v5u8 IXx8
ç 9 9 ¥ 8dpzæ
ç z~5¢©ª8Å8æ ß îæ ø ç4 ü Å Ý ¥ ø Å Ý 6 4 Å Ý
ß î ¦ ä
45üÅ W ß V Ý
Û ¦ 8ªø IxICV6XEII¢8EGV#XgWpI¢aQI6WQsEp3¢o¢If6IW8£I86Q6IXI8 H h 5 2 D T h HBy T 2 5B D T 7B 5 d T H 4 2 H d H gB hr h d T @ ÷ 4 T d H h d 5r h 2 5 4 H Y T d H hB Y 2 5 pgIE8EGVX8sVGGx#6¢UWpGIXI¢dp§jTE8GVWQ8IGeE¢p¢eWXVr¢Qr T H hB 2 5B D Ty 5 T d D H H 4 T Y H F D ` 2B Y R 9 ¥ P 2B d 5 D T d H 5 H F D 2B T D 5 5 Yy T 2 T 4 H H 7 H F 4 Y d 5B d H D 2B H F D 2 T F D d H @ 5y d H h d 5 H 2 5 H 2 T 4 F 4B F ¢9QI8EWQWoXuIG¢UWuQvsEuQI8uI8u¢QEIC@ iy Y R 5B b H d h H B d 4 Y H h Y H 7 H F 4 Y i d T h 2 R 5 y T 4B d H 7 R 2 H F D F DB @ Y D Y T d D 2 5 4 YB F t 2 5B D Ty R EI8XxWIQXW6W6If6#6IW¢UpI8¢gEW6#ISuIWWECCGf¢GWS¢QCXIum8EGVXI 7 rf8uGVWIQ6¢fQI83p8Q6W9EE¢U8E89E¢p¢#Wf¢GQQ6ICXuEIGph8XWEII¢QVUII8 d 5q H D T d R 4 4 T d H h d 5r h 2 5 4 H Y iyy T 5y ` T 2B T D 5 5 D i d T Y Y H 4 H 2 YB YB F D t 2 5B DB h 2 5 4 i d T h 2 R 5 2 2 T 7 R H ò T 5 D 2 5B D T 7B 5 d T y T 4B d H 7 R 2 T q 5 H Y T 4 H F D 21 H D T d R 4 4 T d H h d 5r h 2 5 4 H Y Y IVdI6fGk8XWVdEcII¢v¢QXW6#px¢WVgIW)3¢WVGpQQVQI83xI8Q6WaXB d 5 D T d H 5 YB F D D T F D H 4B D 5 ò ¥ P 2B 2 H bB ` d 5 D T d H 5 H 4 2 H d H gB h H F D Y H 4 R h 5 d F 4B F 8GVWQ8XpGVIGw6XGVfaph pëjTXkQ¢E8¢GVWQ8QI6W6hIIG6QIIxWpAEIC@ ç 9 ¥ o¢dpzæ
45üÅ W ß V Ý
ß¢¥ î
ø ç Å Ý Ý æ 6 4 Å
ß
8¥ î ¦
ä
Å sÝ ¥ æ ç (
Û d 5 D T d H 5 ` 2B @ 5yy 5q H F D 2B T D 5 H @ t º h 2 T t t d 5q ` 2B by 5 Q8WVG6s¢ IXCcEX8IWX¢Gp8wpsI¢ëc!Up8 IXx8 2 5B DB h 2 5 4 i d T h 2 R 5 2 2 T 7 R H ò d 5q Hy T D d 5y i T t ¥ Hy T ¢XGII86eVUII8I¢#IQC8XIV8EgmjI puXIVvt
4 4
Y sfokº ç ¦æ Y
º 4 Y ç ¦ ræ Y 4 hsGS{
í y xuu t w Y ß
î
u
Ý Y sÔxkº ç ¦æ Y 4
ç ¦ræ Y 4 Y sWx
í y ü w ü uu t Y
î ß
Ý Y sWx½ ç ¦ræ Y
4 hsÔxkº Y ç ¦æ Y 4 ç ¦ ræ Y 4 Y sx
¦ í y ú w ú uu t Y
î ß ß î
45Wí y ¥u t ÉaY î ü w Y º ß í Ý Y xck 4 í Ý Y ¥Q{ í y ú w ú ut Y î ß u
í y û w û uu t Y î ß
í xÝ
ù ×ä
45üí W ß V Ý
º
ø ç í Ý ø 4 í Ý ~¥xSæ
ß
î ¦
ø
í W ß V Ý
¥ Hy T t 2B 2 @ 5 F Y Y I aEI¢meXCspWC¢T t Hy T D d 5y i T t T ` 2B Y R h H 2B 7 d H D H h H iyB Y T H 2 T 4 F 4B F @ Y D 2 T D Y 2 5 4 H d T º h 2 T t t H d H F ¢EI¢¢Xxgv9IXIQIE#QGIus#EEW¢QVgXICuGo¢GI¢QuW¢fIVësëW6IC@ ç $ 9 ¥ 8dpzæ ß î
45üÅ W ß V Ý
º
ø ç4 Å Ý ø Å Ý f5~ÿoæ
ß
î ¦
ä
Å hÝ ¥ æ ç (
Û 7 d 5q ` 2B @ 5yy 5q H F D 2B YB F 4B F W8wpXCcXE8IGeXuEEIC@
H h 5 2 D T d 5 D T d H 5 2 T 2 `B Y H h H @ Y R F {pIClC8GVWQ8¢8XQIwCISut
o
Ú Ó × Ò Ë É ÙÒ Ø × Í Ê Ê É Ì Ç Ó Ì Í Ì Ñ ÑÒ ÖÕ Ì Ë Ò ÓÒ Ñ Ð Î Í Ì Ë Ê É È u§fj«wÔzmQÔv«SÏmÇ
7 d H D d 5 d d H ` 2B h T Hy H F D h 2B QW8WQIEI¢gEuIGeIEv¤ ç xgh5WxÝ üí Ý þ ø 4 üí " ø í Ý º ø 4 í Ý vx~¥x~ø ß Axoæ î í Ý ¦ ä íç ( uÝ ¥¥ æ 7 d 5q H F D 2B H bB D T bB d H h h dB F D T 5 D 2 5B D T 7B 5 d T H 4 2 H d H gB h r H DB 2 n T H bB d H ã W8IWeX8GVVWQpXIW #W8XWV#XgWIIVuQI6WQsEp3WEIIu¢EW6$ ¡ ä " F DB GC@ 8EG6ISVgQCfI 2 5B D Y H R P D T H H $ ä ê D T 2 H bB ` YB Ý h 2 2l68E¢uXIVT ¡ VQ88WVkß 5Ý I¢Ý ä ê D T 2 H bB ` H d T r h 2 T FD 5 2 H F W¢sk6IC@ 8EG6ISXk68EQI9QpW IWdIEWISQ9W88W668¢XGII864 2 5B D Y H R P 2B h H bB d H h H 7 H F 4 Y H F D ` 2B Y R i h H 7 d 5q d 5 D 4 H b 2 5B DB h 2 5 r i d T h 2 R 5 H F D h 2 T Y H 4B d D T 7 ¸ H F D D R 5 H DB d @ t F Y H 7 D 2B 5 ç d 5B d H D 2Bæ T ` 2B Y H of¢UII¢sIWI¢QQEWWVdsIWp8GECpQ9oX8G8XQWSE pXWfp¥ 7 d H D d 5 d d H ` 2B h T Hy H F D h 2B QW8WQIEI¢gEuIGeIEv¤ ç4 üí f5Ý " ø í Ý º ø 4 í Ý x~Axæ ß î ¦ ä íç (æ ø 4 íç (æ uhÝ ÿhAhÝ 7 d 5q H F D 2B H bB D T bB d H h D Y d n T 5 D 2 5B D T 7B 5 d T H 4 2 H d H gB hr H DB 2 n T H bB d H ã 8IGeX¢EWVVEW6IU #W8XWV#XgWIIVuQI6WQsEp3WEIIu¢EW6 7 d H D d 5 d d H ` 2B h T Hy H F D h 2B QW8WQIEI¢gEuIGeIEv¤ ç4 üí f5WxÝ " ø í Ý º ø 4 í Ý vxzh¥x~ø ß Axoæ î í Ý ¦ ä íç ( uhÝ æ 7d 5 8q H F D 2B H bB D T bB d H h D Y d n T 5 D 2 5B D T 7B 5 d T H 4 2 H d H gB hr H DB 2 n d H h d 5 r h dB F D T H bB d H ã pGXa8GVVEQIufWUdG¢XGldEpgII¢6IQQsEI£WEIUQI83xWXpGd¢EW68¦ ² R ± ¶ Ã Q·YjÀ µ ´® G 9 ¥ P 5 D y T 4B D 2 H hB YB F 4B F pV%pëjëGh¢QXWS6IXuEEIC@ ç ¡ Õpzæ c ¥ 45üÅ WUß V Ý ß 8¥ î ø ç Å 5Ý 85Ý æ 6 4 Å ß 8¥ î ¦ ä ä ä Ê¤45üÅ W ß V ß ð î Ý ç h 9 ¥ ovdpzæ ÈD 8¥ ß î ø4ÅÝÅ62ÅÝ8ÆøðÅÝ¢Å684Å¢¥ Ý ¦ ß î ¦ ç4 Å Ý ø Å ÆªÝ 6 4 ü Å Ý 85æ Å sÝ ¥ æ ç ( Y H bB Q¢E8` ½IpCVQXEII¢u¢EWVVEW6Ih H h 5 2 D T h HBy T H bB D T bB d H h 2 5 4 H Y T d 5q d 5 D T d H 5 H 4 2 H d H gB h h H d H D 2 H 4 d H h d 5 r h 2 5 4 H Y H F D 5 D 2B YB F D ` 2B D R DB D Y R I86Qºe8¢GVWQ8ÅQI6W6hIå6W6GoQ6A6IW¢3I¢QQ)IWºGoXEIG©IXWpGGpI 9 Â Ú Ù Ì § × Í Ê Ð ±Ð ª³«c²xÎ .
H 7 H F 4 Y F 4 T H d 5q d 5 d d H 2 5B D T 4 2 R d D H F D 2B Y 7 d H V#6IW¢gu¢C¢WW68XWVgIpWGIWeXu9WQWD l d 5q ß î Y b ß q 5 Y D d T id T 2B ` T 7 `IXIVgXev3dIW#¢EW6j l U ¹¡ ¢Æ U#6xÆ î h f¢wGVUf¢UX¢¢dEB 2B h T Hy 5 @D H F D H bB d H ã ß î h 2 T y T H d H F D D 5y Ü Y H 7 H F 4 Y H Y H F D q 5 F 4 T H d 5q d H 7 R 2 H b T @ h H nB h 5 7 H F D h 2B IVu¢geIGw¢Xj¬Q6#6IWW6IGC¢Æ¢Q 8Q ISQ¢c)QIEIx#epG)IEv¤ í Ý h ø 4 í Ý c ¦ 6 í Ý ¦ ¦ ç ß î æ r s¢zIç AxÝ 6 Ax¢~hAx¢a«hxg8Ôæ í ä íç ( uhÝ æ í Ý ø 4 í Ý 6 í Ý ¥æ ä íç ( ç ß sî Iç AxÆh¥¢)hx¢ÏuzÝ æ æ r ß î r ç4 í Ý 6 í Ýæ ä íç ( ¢f¥ÅUxÏuÝ æ Û iy H bB D 4 H Y H d t d H h d 5 r h dB F D h 2 Q8GQWQp6QI83XIWp¢T t r h 2 5 4 H Y t r D Y d n H d T F 4B F @ t Y H 7 H F 4 Y ` 2B 4 2 H d H gB h h H hB Yr H 2 5 ` 2B @ 5yy 5q H F D d H hB Y 2 5 GI86QGUf¢eEICpQ6#6IWaIX6IQQsEI#QIEW£I8uIECcXX¢fIWvQpXWp8f ¡ ¦ H bB D T bB d H h D Y d n T 5 D Y 2 5B D T 7B 5 d T h H d H ¢¢EGl¢W6IfWU 9GvI8XWV#XgWIIV6WQWD r 2 H 4 D 4 T 7 5 4 h 2 T D 4 T 7 5 4 2 5 2 d H hd 5r FD BY d 5q D 2 H 4d H ¦ xQQC¢U98QIVC6¢I#86I8Iu6IW¢3W W¢o6QW6su ¡ ¢UWW6Xu8WQQ6sY 2 T F D Y Y Hy d 5 d d H h H H H Y T F T H b HB F 4 T 5 D D 2 H 7 H dB R P H d ç F D ` 2 Hy H b T @r d H r Y D 2B 5 r hB d ` H F D h 2B ¤ ¢UIr86EI¢ÅWoQ9QXISQ 0æ W¢I6XQ¢gfQ3ÔGoX¢s£X8IW©IEvph l d 5q ß î Y j l U m¡ 8 UQxb ß î ß î h f¢Xjp¢XGfQISeE68WQp8WVG6s8QI83WI¢6¢I#86IGp¢I8dp¥ D 5y Ü 2 5B D Y H R P 2B h H bB d H h d 5 D T d H 5 d H h d 5 r F D d R 5 q D 4 T 7 5 4 H F D h 2 T t ç $ $ ëj8WVG6s¢96IW¢3WI896VU#¢QI8pIWvV8EWVVEQIºp8Q6Wk#¢¢GVWQ85 P æ d 5 D T d H 5 d H h d 5r F D d R 5q D 4 T 7 5 4 2 5 2 H F D t H bB D T bB d H h h 2 5 4 H Y T d 5q d 5 D T d H H 4 2 H d H gB h h H d H D 2 H 4 d H h d 5r h 2 5 4 H Y H F D d 5q d H 7 R 2 H b T @ h H nB h 5 7 H F D h 2B ¤ 2 5B 6IQQsEI¬QQWS6Q 6IW¢3I¢QQdIWw¢6swISQ8gA6UEp9IWIEv8XWD r T 7B 5 d T H bB D T bB d H h h 2 5 4 H Y T d 5 l#XgWIIV8EWVVEQIAI86QWu¢q h Q ISQ¢g6UEIx9IGeIEIU6pwpxWD d H 7 R 2 H b T @ h H nB h 5 7 H F D ` 2B 2 n H h Y R F 6 ä í þ h uç e í þ ¥ æ ( Y H bB ` H bB D T bB d H h h 2 5 4 H Y H F D d 5q d 5 D T d H 5 H 4 2 H d H gB h T q 5 2 5B D T 4By 68E¢8EWVVEQII8Q6WuIW88GlGQ8uQpQW6hI¢d8XWVgEXIIfi Y H bB Q¢E8` ß þ $6 ä þ þ 2 5B D 4 2 Rq H F D 5 D d 5 D T d H 5 H bB D T bB d H hr h 2 5 4 H Y H F D q 5 2 5B D T 4By i 8XW6IpupGG8WVG6s¢w¢EWVVEW6I3xI8Q6WuIWm¢d8XWVgEXIIfCpc ç $ ¥ P æ d 5 D T d H 5 h H d H D 2 H 4 d H h d 5 r F D d R 5 q H F SVdpëëw8GlGQ86W6GoQ6uQI83WI¢upGD ¢WeQIE¢8VUGeWECuVU 7 5 dq h H 2B T D 5 D T F D F DB @ D d T l d 5q ß î Y b ß q 5 Y D d T i d T 2B ` T 7B h 2 T y T H y T Hd H F D Hd T 7 5 ¢gIG¢U98j l U m¡ 8 U£6 î h #¢GVU9¢UE8¢#XeIV¢gd ß î H F D D 5y Ü ¦ 2 5B D Y H R P 2B h H bB d H h d 5 D T d H 5 H F D d 5q d H 7 R 2 H b T @ h H nB h 5 7 H F D h 2B ¤ pGVXj8e8EG6ISdE68WQp8WVG6s8IG8uQ ISQ¢gd6UEp9IW#IEv9 7 d H D d 5 d d H ` 2B h T Hy H F D h 2B QW8WQIEI¢gEuIGeIEv¤ ç4 üí Ý º ø í Ý ø 4 í Ý 5Wxv~¥xSæ ß î ¦ ä 4 üíç (æ þ ø íç (æ ø 4 íç ( j5WzÝ ¥ S~jzÝ ¥ ÿAzÝ ¥ æ " 7 d 5q H F D 2B H bB 8IGEu8EWD r T b B d H h h 2 5 4 H Y T 5 D 2 5 B D T 7 B 5 d T H 4 2 H d H g B h r H D B 2 n ç `H h T Ü d 5 æ D 4 T 7 5 4 T H b B d H ã lVEW6Ip8Q6W9Gd8EGV#XgWpI¢uQpQW6hI3GIUc¨¢mC8a¢U98Qu¢EW6p Ú Ó × Ò Ë É ÙÒ Ø × Í Ê Ê É Ì Ç Ó Ì Í Ì Ñ ÑÒ ÖÕ Ì Ë Ò ÓÒ Ñ Ð Î Í Ì Ë Ê É È u§fj«wÔzmQÔv«SÏmÇ c 3 .
these ODE's would be expressed in the form d~ = F (~ t) u ~u (4. we use this form. x 51 2 2 ( +1) ( 1) ( ) +1 ( ) 2 ( ) 1 ( +1) ( 1) 2 ( ) +1 ( ) ( ) 1 . without approximating the time derivative. = 4 2h x or h i ujn = ujn. Another strategy for constructing a nitedi erence approximation to a PDE is to approximate all the partial derivatives at once. + 2h ujn . which we refer to as the semidiscrete approach. but the rest of this chapter is devoted to a more specialized matrix notation described below. In the following chapters. As an example let us consider the model equation for di usion @u = @ u @t @x Using threepoint centraldi erencing schemes for both the time and space derivatives. 2ujn + ujn (4. In the most general notation. in turn. On occasion. can be used for the time advance of the solution at any given point in the mesh. we proceed with an analysis making considerable use of this concept. ujn. we nd 2 n 3 ujn .Chapter 4 THE SEMIDISCRETE APPROACH One strategy for obtaining nitedi erence approximations to a PDE is to start by di erencing the space derivatives only.1) which. This generally leads to a point di erence operator (see Section 3. Di erencing the space derivatives converts the basic PDE into a set of coupled ODE's. uj .2) .1) dt which includes all manner of nonlinear and timedependent possibilities.3. 2ujn + ujn 5 .
As we shall see later on. In this case. we can approximate the space derivatives with di erence operators and express the resulting ODE's with a matrix formulation. the spatial and temporal discretizations are not separable. THE SEMIDISCRETE APPROACH Clearly Eq. that the spatial and temporal discretizations are separable. For the time being. however.1.52 CHAPTER 4. 4. This is a simple and natural formulation when the ODE's are linear. 2 @ j x ( 1) 2 ( ) +1 ( +1) 2 0 ( +1) + ujn.3 is referred to as the DuFortFrankel method. Note. this method has an intermediate semidiscrete form and can be analyzed by the methods discussed in the next few chapters. . 4. 4. Equation 4. stability.2 is a di erence equation which can be used at the space point j to advance the value of u from the previous time levels n and n . and convergence of Eqs. 4.3) which can be solved for u n and time advanced at the point j .2 is sometimes called Richardson's method of overlapping steps and Eq. In this approach. )=2. Another possibility is to replace the value of ujn in the right hand side of Eq. namely (ujn + ujn. It is a full discretization of the PDE. there are subtle points to be made about using these methods to nd a numerical solution to the di usion equation. This results in the formula ( ) ( +1) ( 1) ujn ( +1) un = ujn. we shall separate the space di erence approximations from the time di erencing. We introduce these methods here only to distinguish between methods in which the temporal and spatial terms are discretized separately and those for which no such separation is possible. and no semidiscrete form exists.2 and 4. There are a number of issues concerning the accuracy. + 2h 4ujn . A n 5 + uj .1 Reduction of PDE's to ODE's 4. 4. Thus.3 which we cannot comment on until we develop a framework for such investigations. we reduce the governing PDE's to ODE's by discretizing the spatial terms and use the welldeveloped theory of ODE solutions to aid us in the development of an analysis of accuracy and stability. In these simple cases.1 The Model ODE's First let us consider the model PDE's for di usion and biconvection described in Chapter 2. 1 to the level n + 1.2 with the time average of u at that point. 2 ( 1) ( ) 1 1 3 (4.
5 are the model ODE's for di usion and biconvection of a scalar in one dimension. ~ (t) u u f dt (4. even the Euler and NavierStokes equations can be expressed in the form of Eq. 4.6. that is.4. It is used for the scalar convection model when the boundary conditions are periodic.1.5) dt 2 x p where the boundary condition vector is absent because the ow is periodic. In general. Although the equations are nonlinear.6) Note that the elements in the matrix A depend upon both the PDE and the type of di erencing scheme chosen for the space terms. .2 The Generic Matrix Form The generic matrix form of a semidiscrete approximation is expressed by the equation d~ = A~ . using the 3point centraldi erencing scheme to represent the second derivative in the scalar PDE governing di usion leads to the following ODE di usion model d~ = u B (1 . In such cases the equations are nonlinear.4) dt x ~ with Dirichlet boundary conditions folded into the (bc) vector. Eqs.1 0 1)~ u u (4. the 3point centraldi erencing approximation produces the ODE model given by d~ = . 4. 4. the linear analysis presented in this book leads to diagnostics that are surprisingly accurate when used to evaluate many aspects of numerical methods as they apply to the Euler and NavierStokes equations.1. The vector ~ (t) is usually determined f by the boundary conditions and possibly source terms. They are linear with coe cient matrices which are independent of x and t.3. 2 Model ODE for Biconvection The term biconvection was introduced in Section 2. REDUCTION OF PDE'S TO ODE'S 53 Model ODE for Di usion For example. the elements of A depend on the solution ~ and u are usually derived by nding the Jacobian of a ux vector.4 and 4. In this case.2 1)~ + (bc) u ~ (4. a B (.
I] = 0 We form the righthand eigenvector matrix of a complete system by lling its columns with the eigenvectors ~ m: x h i X = ~ ~ : : : ~M x x x The inverse is the lefthand eigenvector matrix.2. mI]~ m = 0 x (4. 4. and together they have the property that 1 2 X . depends explicitly on t. 4.2 Exact Solutions of Linear ODE's CHAPTER 4.6 in which the coe cient matrix.7) The eigenvalues are the roots of the equation det A . THE SEMIDISCRETE APPROACH In order to advance Eq.e. if @F=@u = A where A is independent of u). have the property that A~ m = m~ m or x x A . the exact solution of Eq. This will lead us to a representative scalar equation for use in analyzing timemarching methods. if A does not depend explicitly on t.. As we shall soon see.1 are said to be linear if F is linearly dependent on u (i. the general solution to Eq. the system of ODE's must be integrated using a timemarching method. The ODE's represented by Eq. A. which exist under certain conditions. An eigenvector. and its x corresponding eigenvalue. when the ODE's are linear they can be expressed in a matrix notation as Eq. In order to analyze timemarching methods. These ideas are developed in the following sections. 4. AX = 1 (4. If A does depend explicitly on t. f . ~ m . This holds regardless of whether or ~ not the forcing function.1 Eigensystems of SemiDiscrete Linear Forms Complete Systems An M M matrix is represented by a complete eigensystem if it has a complete set of linearly independent eigenvectors (see Appendix A) . is independent of u.6 can be written in terms of the eigenvalues and eigenvectors of A. we will make use of exact solutions of coupled systems of ODE's. 4.1 in time. 4. 4. As we have already pointed out.54 4.6 can be written. m. the general solution cannot be written whereas.8) where is a diagonal matrix whose elements are the eigenvalues of A. .
4.2. EXACT SOLUTIONS OF LINEAR ODE'S
55
Defective Systems
If an M M matrix does not have a complete set of linearly independent eigenvectors, it cannot be transformed to a diagonal matrix of scalars, and it is said to be defective. It can, however, be transformed to a diagonal set of blocks, some of which may be scalars (see Appendix A). In general, there exists some S which transforms any matrix A such that S ; AS = J where 2J 3 6 J 7 6 7 6 7 ... 6 7 J =6 7 6 7 Jm 4 5 ... and 2 3 m 1 6 7 1 ... 6 7 ... m n =6 Jm 6 7 . . . 1 7 ... 4 5 m n The matrix J is said to be in Jordan canonical form, and an eigenvalue with multiplicity n within a Jordan block is said to be a defective eigenvalue. Defective systems play a role in numerical stability analysis.
1 1 2 ( )
4.2.2 Single ODE's of First and SecondOrder
FirstOrder Equations
The simplest nonhomogeneous ODE of interest is given by the single, rstorder equation du = u + ae t (4.9) dt where , a, and are scalars, all of which can be complex numbers. The equation is linear because does not depend on u, and has a general solution because does not depend on t. It has a steadystate solution if the righthand side is independent of t, i.e., if = 0, and is homogeneous if the forcing function is zero, i.e., if a = 0. Although it is quite simple, the numerical analysis of Eq. 4.9 displays many of the fundamental properties and issues involved in the construction and study of most popular timemarching methods. This theme will be developed as we proceed.
56
CHAPTER 4. THE SEMIDISCRETE APPROACH
The exact solution of Eq. 4.9 is, for 6= ,
u(t) = c e
1 1
t+
ae t
;
where c is a constant determined by the initial conditions. In terms of the initial value of u, it can be written
t; t u(t) = u(0)e t + a e ; e
The interesting question can arise: What happens to the solution of Eq. 4.9 when = ? This is easily found by setting = + , solving, and then taking the limit as ! 0. Using this limiting device, we nd that the solution to
du = u + ae t dt
is given by
(4.10)
u(t) = u(0) + at]e t As we shall soon see, this solution is required for the analysis of defective systems.
SecondOrder Equations
The homogeneous form of a secondorder equation is given by
d u + a du + a u = 0 (4.11) dt dt where a and a are complex constants. Now we introduce the di erential operator D such that d D dt
2 2 1 0 1 0
and factor u(t) out of Eq. 4.11, giving
2
(D + a D + a ) u(t) = 0
1 0
The polynomial in D is referred to as a characteristic polynomial and designated P (D). Characteristic polynomials are fundamental to the analysis of both ODE's and O E's, since the roots of these polynomials determine the solutions of the equations. For ODE's, we often label these roots in order of increasing modulus as , , , m ,
1 2
4.2. EXACT SOLUTIONS OF LINEAR ODE'S
1 2
57
, M . They are found by solving the equation P ( ) = 0. In our simple example, there would be two roots, and , determined from P( ) = + a + a = 0 (4.12) and the solution to Eq. 4.11 is given by u(t) = c e 1 t + c e 2 t (4.13) where c and c are constants determined from initial conditions. The proof of this is simple and is found by substituting Eq. 4.13 into Eq. 4.11. One nds the result c e 1t( + a + a ) + c e 2t( + a + a ) which is identically zero for all c , c , and t if and only if the 's satisfy Eq. 4.12.
2 1 0 1 2 1 2 1 2 1 1 1 0 2 2 2 1 2 0 1 2
4.2.3 Coupled FirstOrder ODE's
A Complete System
A set of coupled, rstorder, homogeneous equations is given by u0 = a u + a u u0 = a u + a u (4.14) which can be written " # 0 a a ~ = A~ ~ = u u ]T u u u A = (aij ) = a a Consider the possibility that a solution is represented by u = c x e 1t + c x e 2t u = c x e 1t + c x e 2t (4.15) By substitution, these are indeed solutions to Eq. 4.14 if and only if " #" # " # " #" # " # a a x = x a a x = x (4.16) a a x x a a x x Notice that a higherorder equation can be reduced to a coupled set of rstorder equations by introducing a new set of dependent variables. Thus, by setting u = u0 u =u we nd Eq. 4.11 can be written u0 = ;a u ; a u u0 = u (4.17) which is a subset of Eq. 4.14.
1 2 11 21 1 1 12 22 2 2 1 2 11 21 12 22 1 2 1 1 11 21 2 2 12 22 11 12 11 21 21 22 1 11 21 11 12 22 12 22 21 2 12 22 1 2 1 1 1 1 0 2 2
58
CHAPTER 4. THE SEMIDISCRETE APPROACH
1 2
A Derogatory System
Eq. 4.15 is still a solution to Eq. 4.14 if = = , provided two linearly independent vectors exist to satisfy Eq. 4.16 with A = . In this case 0 0 1 = 0 1 0 and 0 0 0 = 1 0 1
provide such a solution. This is the case where A has a complete set of eigenvectors and is not defective.
A Defective System
If A is defective, then it can be represented by the Jordan canonical form
"
# " #" # u0 = 0 u u0 1 u
1 2 1 2
(4.18)
whose solution is not obvious. However, in this case, one can solve the top equation rst, giving u (t) = u (0)e t . Then, substituting this result into the second equation, one nds du = u + u (0)e t dt which is identical in form to Eq. 4.10 and has the solution
1 1 2 2 1
u (t) = u (0) + u (0)t]e t
2 2 1
From this the reader should be able to verify that
u (t) = a + bt + ct e t
3 2
is a solution to
2 03 2 6 u0 7 = 6 1 4u 5 4 u0 1
1 2 3
32 3 76u 7 54u 5 u
1 2 3
if
a = u (0)
3
b = u (0)
2
1 c = 2 u (0)
1
(4.19)
The general solution to such defective systems is left as an exercise.
4.2. EXACT SOLUTIONS OF LINEAR ODE'S
59
4.2.4 General Solution of Coupled ODE's with Complete Eigensystems
Let us consider a set of coupled, nonhomogeneous, linear, rstorder ODE's with constant coe cients which might have been derived by space di erencing a set of PDE's. Represent them by the equation
d~ = A~ ; ~ (t) u u f (4.20) dt Our assumption is that the M M matrix A has a complete eigensystem and can be transformed by the left and right eigenvector matrices, X ; and X , to a diagonal matrix having diagonal elements which are the eigenvalues of A, see Section 4.2.1. Now let us multiply Eq. 4.20 from the left by X ; and insert the identity combination XX ; = I between A and ~ . There results u ~ X ; du = X ; AX X ; ~ ; X ; ~ (t) u f (4.21) dt Since A is independent of both ~ and t, the elements in X ; and X are also indepenu dent of both ~ and t, and Eq. 4.21 can be modi ed to u
1 1 1 1 1 1 1 1 1
d X ; ~ = X ; ~ ; X ; ~ (t) u f dt u ~ Finally, by introducing the new variables w and ~ such that g
1 1 1
~ w = X; ~ u
1
~ (t) = X ; ~ (t) g f
1
(4.22)
we reduce Eq. 4.20 to a new algebraic form
~ dw = w ; ~ (t) ~ g (4.23) dt It is important at this point to review the results of the previous paragraph. Notice that Eqs. 4.20 and 4.23 are expressing exactly the same equality. The only di erence between them was brought about by algebraic manipulations which regrouped the variables. However, this regrouping is crucial for the solution process because Eqs.
In the following, we exclude defective systems, not because they cannot be analyzed (the example at the conclusion of the previous section proves otherwise), but because they are only of limited interest in the general development of our theory.
1
60
CHAPTER 4. THE SEMIDISCRETE APPROACH
4.23 are no longer coupled. They can be written line by line as a set of independent, single, rstorder equations, thus
w0 = ... 0 wm = ... 0 wM =
1
1
w ; g (t)
1 1
m wm ; gm (t) M wM
; gM (t)
(4.24)
For any given set of gm (t) each of these equations can be solved separately and then recoupled, using the inverse of the relations given in Eqs. 4.22: ~ (t) = X w(t) ~ u =
M X m=1
wm(t)~ m x
(4.25)
where ~ m is the m'th column of X , i.e., the eigenvector corresponding to m . x We next focus on the very important subset of Eq. 4.20 when neither A nor ~ has f any explicit dependence on t. In such a case, the gm in Eqs. 4.23 and 4.24 are also time invariant and the solution to any line in Eq. 4.24 is w (t) = c e m t + 1 g
m m m m
where the cm are constants that depend on the initial conditions. Transforming back to the usystem gives ~ (t) = X w(t) ~ u = = = =
M X m=1 M X m=1 M X m=1 M X m=1
wm(t)~ m x cm e m t ~ m + x
m=1 m M X 1 ~ gm xm
cm e m t ~ m + X x
;1 X ;1~ f
1
cm e m t ~ m + A; ~ x f

Transient
{z
} 
Steadystate
{z
}
(4.26)
We begin by developing the concept of \spaces". of Jordan blocks or. REAL SPACE AND EIGENSPACE 1 61 Note that the steadystate solution is A. For the model problems on which we are focusing most of our attention. There is. to a much greater extent. we reduce the partial di erential equations to a set of ordinary di erential equations represented by the generic form d~ = A~ . An alternative. The second group is referred to classically as the particular solution or the particular integral. another very useful frame. we get di erent perspectives of the same solution.1 and 4. composed.and postmultiplication of A by the appropriate similarity matrices transforms A into a diagonal matrix. form of the solution is ~ (t) = c e 1 t ~ + + cme m t ~ m + + cM e M t ~ M + A. it is said u to be in real space.3. In this way. The most natural reference frame is the physical one. That is.3 Real Space and Eigenspace Following the semidiscrete approach discussed in Section 4.1 De nition . ~ u x x x f (4. we can develop some very important and fundamental concepts that underly the global properties of the numerical solutions to the model problems. however. This permits us to say a great deal about such problems and serves as the basis for this section. but entirely equivalent. f The rst group of terms on the right side of this equation is referred to classically as the complementary solution or the solution of the homogeneous equations. ~ u u f (4.27) 1 1 1 4. We saw in Sections 4. in the 4.3. as might be expected. and this can add signi cantly to our understanding. How these relate to the numerical solutions of more practical problems depends upon the problem and. in the most general case. In particular. respectively.4.1. it is more instructive to refer to these groups as the transient and steadystate solutions. We say If a solution is expressed in terms of ~ . we identify di erent mathematical reference frames (spaces) and view our solutions from within each. the elements of A are independent of both u and t.28) dt The dependent variable ~ represents some physical quantity or quantities which relate u to the problem of interest.2 that pre. on the cleverness of the relator. ~ .2. In our application to uid dynamics.
~ g f 1 1 ~ = Xw ~ u ~ = X~ f g The elements of ~ relate directly to the local physics of the problem. THE SEMIDISCRETE APPROACH simplest nondefective case. Equations for the eigenvalues of the simple . we found that Eq.2 Eigenvalue Spectrums for Model ODE's It is instructive to consider the eigenvalue spectrums of the ODE's formulated by central di erencing the model equations for di usion and biconvection.2 and. X . When the forcing function ~ is independent of t. it is said to be in eigenspace (often referred to as wave space). m 4. We say ~ If a solution is expressed in terms of w. for simplicity. the transient portion of the solution in real space consists of a linear combination of contributions from each eigenvector. 4. of scalars.28 had the alternative form ~ dw = w . the transient portion of the solution in eigenspace provides the weighting of each eigenvector component of the transient solution in real space.1. the u ~ are linear combinations of all of the elements of ~ . the solutions in the two spaces f are represented by ~ (t) = X cm e m t ~ m + X .3. ~ u x f and wm(t) = cm e m t + 1 gm m=1 2 M 1 1 for real space and eigenspace. using only complete systems for our examples. respectively. and individually elements of w u they have no direct local physical interpretation. The relations that transfer from one space to the other are: ~ w = X. ~ ~ g dt which is an uncoupled set of rstorder ODE's that can be solved independently for ~ the dependent variable vector w. These model ODE's are presented in Section 4. ~ u ~ = X. At this point we make two observations: 1. and 2. However.62 CHAPTER 4. Following Section 4.
1. to help visualize the matrix structure.30) = sin m x m=0 1 M . The interpretation of this result plays a very important role later in our stability analysis.1 m = x M 2 2 2 where = .4. for the model biconvection equation .2 + 2 cos Mm+ 1 m = x ! m .ia sin 2m m=0 1 M . m = m. From Section B.4.4. we present results for a simple 4point mesh and then we give the general case.3.5. Recall that xj = j x = j =(M + 1). 4. and x = 2 =M .3. Notice that the di usion eigenvalues are real and negative while those representing periodic convection are all pure imaginary. REAL SPACE AND EIGENSPACE 63 tridiagonals.i m a m m=0 1 M .3 Eigenvectors of the Model Equations The Di usion Model Next we consider the eigenvectors of the two model equations.4 sin m=1 2 M (4. the model ODE's for di usion.32) . These follow as special cases from the results given in Appendix B. so in general the relation ~ = X w can be written as u ~ 1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4 uj = M X m=1 wm sin mxj j=1 2 M (4.29) = 2(M + 1) x and. B (M : a b c) and Bp(M : a b c). The righthand eigenvector matrix X is given by 2 sin (x ) sin (2x ) sin (3x ) sin (4x ) 3 6 sin (x ) sin (2x ) sin (3x ) sin (4x ) 7 6 6 sin (x ) sin (2x ) sin (3x ) sin (4x ) 7 7 4 5 sin (x ) sin (2x ) sin (3x ) sin (4x ) The columns of the matrix are proportional to the eigenvectors. from Section B. First. 4. we nd for the model di usion equation with Dirichlet boundary conditions . Consider Eq. are given in Appendix B.1 (4.1 (4.31) x is the modi ed wavenumber from Section 3.
we nd 1 2 sin (x ) 6 sin (2x ) 6 6 sin (3x ) 4 sin (4x ) 1 1 1 1 sin (x ) sin (2x ) sin (3x ) sin (4x ) 2 2 2 2 sin (x ) sin (2x ) sin (3x ) sin (4x ) 3 3 3 3 sin (x ) 3 sin (2x ) 7 7 5 sin (3x ) 7 sin (4x ) 4 4 4 4 1 The rows of the matrix are proportional to the eigenvectors. 4. For our model ODE. The coe cient matrices for these ODE's are always circulant.33. Similarly.33 represents a sine transform of the function u(x) for an M point sample between the boundaries x = 0 and x = with the condition u(0) = u( ) = 0.5.1 X m=0 (2 m=M ) j =0 1 m =0 1 M .1 M . 4. we can write ~ = X w as u ~ uj = wmeimxj j=0 1 (4.1 x = j 2 =M . or lefthand eigenvector matrix X . 4. 1 The Biconvection Model Next consider the model ODE's for periodic convection. Eq. Eq.1 M . For the model di usion equation: ~ w = X . THE SEMIDISCRETE APPROACH For the inverse. the righthand eigenvectors are given by ~ m = ei j x With xj = j M . In general w = X . 4. ~ ~ u gives wm = M X j =1 uj sin mxj m=1 2 M (4. ~ = Xw ~ u is a sine synthesis from wave space back to real space.64 CHAPTER 4.32 represents the sine synthesis that companions the sine transform given by Eq. ~ is a sine transform from real space to (sine) wave u space. In summary.34) . Eq. .33) In the eld of harmonic analysis.
i = . i = e. REAL SPACE AND EIGENSPACE 2w 6w 6 6w 4 w 3 21 1 1 1 7 1 6 1 e. i = e .i = . 4.4. ~ is a complex Fourier transform from real space to u wave space. 1 4.3.35) m m = . x. u For circulant matrices. i = 1 e 2 4 6 4 4 4 4 8 4 4 6 12 18 12 4 1 65 For a 4point periodic mesh. i = e.1 j j This equation is identical to a discrete Fourier transform of the periodic dependent variable ~ using an M point sample between and including x = 0 and x = 2 . ~ ~ u u =0 In general For any circulant system: ~ w = X . In summary. i = 7 4 61 e 5 4 e e .4 Solutions of the Model ODE's The Di usion Equation For the di usion equation. Eq. uj (t) = where M X m=1 cme m t sin mxj + (A. we nd the following lefthand eigenvector matrix from Appendix B.4: 1 2 3 4 4 4 4 32u 76u 76 76u 54 u 1 2 3 4 3 7 7 = X. u e.imxj wm = M m=0 1 M .36) . f )j 1 j=1 2 M (4. i = e.27 becomes We can now combine the results of the previous sections to write the solutions of our model ODE's. i = 7= 6 .3. it is straightforward to establish the fact that the relation ~ = X w represents the Fourier synthesis of the variable w back to ~ . ~ = Xw ~ u is a complex Fourier synthesis from wave space back to real space.4 sin 2(M + 1) x 2 2 ! (4. ~ 7 u 5 1 X 1 M .
i m a (4. we can write the ODE solution as uj (t) = M X m=1 cme. while high wavenumber modes (if they have signi cant amplitudes) can have large errors. evaluated at the nodes of the grid: uj (t) = M X m=1 cm e. f )j 1 j=1 2 M (4.1 X m=0 cme.40) m = .66 CHAPTER 4. m 2 t sin m xj + (A. we obtain uj (t) = where M .31.37) With the modi ed wavenumber de ned as 2 m x m = x sin 2 and using m = m. We can write this ODE solution as uj (t) = M . The modi ed wavenumber is an approximation to the actual wavenumber. The di erence between the modi ed wavenumber and the actual wavenumber depends on the di erencing scheme and the grid resolution.41) with the modi ed wavenumber de ned in Eq. This di erence causes the various modes (or eigenvector components) to decay at rates which di er from the exact solution. THE SEMIDISCRETE APPROACH (4.1 (4. With conventional di erencing schemes. Eq.1 X m=0 cme m t ei mxj j=0 1 M .i m at ei mxj j=0 1 M .42) .38) This can be compared with the exact solution to the PDE. The Convection Equation For the biconvection equation. low wavenumber modes are accurately represented.37.39) We see that the solutions are identical except for the steady solution and the modi ed wavenumber in the transient term.1 (4. 2. 4. m 2 t sin m xj + h(xj ) j=1 2 M (4.
g (t) (4.3. which are related by algebraic manipulation. Our next objective is to nd a \typical" single ODE to analyze. discussed in Chapter 11. The importance of this concept resides in its message that we can analyze timemarching methods by applying them to a single. solved there in their uncoupled form. since is real. we pointed out that Eqs.5. This leads to the following important concept: The numerical solution to a set of linear ODE's (in which A is not a function of t) is entirely equivalent to the solution obtained if the equations are transformed to eigenspace. evaluated at the nodes of the grid: M. For such a purpose Eq. 2. while the actual phase speed is independent of the wavenumber. Eq. topics which are covered in Chapters 6 and 7.20 and 4. . not particular problems.2. 4. m=0 4.44 is not quite satisfactory. However. and then returned as a coupled set to real space. 4.26. THE REPRESENTATIVE EQUATION 1 67 and compare it to the exact solution of the PDE.4. We found the uncoupled solution to a set of ODE's in Section 4. this leads to an error in the speed with which various m modes are convected. As discussed in Section 3. X uj (t) = fm (0)e. In the case of noncentered di erencing. A typical member of the family is dwm = w .44) m m m dt The goal in our analysis is to study typical behavior of general situations.4 The Representative Equation In Section 4. The form of Eq. This is helpful both in analyzing the accuracy of timemarching methods and in studying their stability. the result is erroneous numerical dispersion.23 express identical results but in terms of di erent groupings of the dependent variables.1 (4. Since the error in the phase speed depends on the wavenumber. uncoupled equation and our conclusions will apply in general.42 shows that the imaginary portion of the modi ed wavenumber produces nonphysical decay or growth in the numerical solution.i mat ei m xj j=0 1 M .43) Once again the di erence appears through the modi ed wavenumber contained in .4. the modi ed wavenumber is complex. 4. The role of m is clear it stands for some representative eigenvalue in the original A matrix.
in principle.g(t) = ak eikt X ak eikt k ik . 4. 6x @t @x 2 2 . This leads to w(t) = ce t + The Representative ODE dw = w + ae t dt (4. we note that. Write the resulting semidiscrete ODE form. The exact solution of the representative ODE is (for 6= ): t (4.5 Problems 1. In such evaluations the parameters and must be allowed to take the worst possible combination of values that might occur in the ODE eigensystem. THE SEMIDISCRETE APPROACH the question is: What should we use for gm(t) when the time dependence cannot be ignored? To answer this question.46) w(t) = ce t + ae . 3.68 CHAPTER 4.45) for which Eq. From this we can extract the k'th term and replace ik with . For example X . 2. write the semidiscrete form obtained with periodic boundary conditions on a 5point grid (M = 5). Consider the application of the operator given in Eq. Using this operator to approximate the spatial derivative in the linear convection equation. Write the semidiscrete form resulting from the application of secondorder centered di erences to the following equation on the domain 0 x 1 with boundary conditions u(0) = 0. Consider the nitedi erence operator derived in question 1 of Chapter 3. 3.52 to the 1D di usion equation with Dirichlet boundary conditions. Find the entries in the boundarycondition vector.44 has the exact solution: k which can be used to evaluate all manner of timemarching methods. u(1) = 1: @u = @ u . 4. one can express any one of the forcing terms gm(t) as a nite Fourier series.
1 .) Calculate the corresponding modi ed wavenumbers for the secondorder centered operator from Eq. The eigenvalues of the above matrix A. For initial conditions u(x 0) = sin(mx) and boundary conditions u(0 t) = u( t) = 0.4. 4.Bp(10 .4. Consider the matrix A = .1 0 1)=(2 x) corresponding to the ODE form of the biconvection equation resulting from the application of secondorder central di erencing on a 10point grid. Calculate and plot the corresponding ODE solutions. Using these. can be found from Appendix B. as well as the matrices X and X . PROBLEMS 69 4. Compare with the exact solution of the PDE. Repeat for the initial condition u(x 0) = sin 2x.37. (Plot the solution at the grid nodes only. . Note that the domain is 0 x 2 and x = 2 =10. Calculate the numerical phase speed from the modi ed wavenumber corresponding to this initial condition and show that it is consistent with the ODE solution.5. The grid nodes are given by xj = j x j = 0 1 : : : 9. compute and plot the ODE solution at t = 2 for the initial condition u(x 0) = sin x. plot the exact solution of the di usion equation with = 1 at t = 1 with m = 1 and m = 3. 5. Consider a grid with 10 interior points spanning the domain 0 x .
70 CHAPTER 4. THE SEMIDISCRETE APPROACH .
5. momentum. Next we will give our model equations in the form of Eq. we will show how similar semidiscrete forms can be derived using nitevolume approximations in space. i. 71 . Second. As a result.2. Finitevolume methods are applied to the integral form of the governing equations. This will be followed by several examples which hopefully make these concepts clear.1. 2. and energy are conserved in a discrete sense. we saw that the application of a nitedi erence approximation to the spatial derivatives in our model PDE's produces a coupled set of ODE's. they can be applied on unstructured meshes consisting of arbitrary polyhedra in three dimensions or arbitrary polygons in two dimensions. nitevolume methods do not require a coordinate transformation in order to be applied on irregular meshes. we saw how to derive nitedi erence approximations to arbitrary derivatives. mass. While this property can usually be obtained using a nitedi erence formulation.1) We will begin by presenting the basic concepts which apply to nitevolume strategies. primarily. 2. In this Chapter. it is obtained naturally from a nitevolume formulation. Consistent with our emphasis on semidiscrete methods. either in the form of Eq. we will study the latter form. they ensure that the discretization is conservative. which is d Z QdV + I n:FdS = Z PdV dt V (t) S (t) V (t) (5.1 or Eq. of two advantages. First.. In Chapter 4.Chapter 5 FINITEVOLUME METHODS In Chapter 3.e. This increased exibility can be used to great advantage in generating grids about arbitrary geometries. Finitevolume methods have become popular in CFD as a result.
A nondissipative scheme analogous to centered di erencing is obtained by taking the average of these two uxes. 5. Another approach known as uxdi erence splitting is described in Chapter 11. which are a function of Q.1 Basic Concepts CHAPTER 5. The basic elements of a nitevolume method are thus the following: 1 Timemarching methods will be discussed in the next chapter. Thus after applying a timemarching method.1 can be written as 1 Z QdV V V (5. we note that the average value of Q in a cell with volume V is Q and Eq. FINITEVOLUME METHODS The basic idea of a nitevolume method is to satisfy the integral form of the conservation law to some degree of approximation for each of many contiguous control volumes which cover the domain of interest. The ux is required at the boundary of the control volume. First. Similarly. which is a closed surface in three dimensions and a closed contour in two dimensions. Q can be represented within the cell by some piecewise approximation which produces the correct value of Q. we will consider only control volumes which do not vary with time. each cell will have a di erent piecewise approximation to Q. Let us consider these approximations in more detail. at the controlvolume boundary. Next a timemarching method1 can be applied to nd the value of Z V QdV (5. Examining Eq. Thus the volume V in Eq. 5. This ux must then be integrated to nd the net ux through the boundary. 5. When these are used to calculate F(Q).1 is that of a control volume whose shape is dependent on the nature of the grid. In order to evaluate the uxes.72 5. we have updated values of the cellaveraged quantities Q. In our examples. This is a form of interpolation often referred to as reconstruction. we see that several approximations must be made.2) at the next time step. As we shall see in our examples.1. they will generally produce di erent approximations to the ux at the boundary between two control volumes. .4) for a control volume which does not vary with time. the ux will be discontinuous. the source term P must be integrated over the control volume. that is.3) I Z d V dt Q + S n:FdS = V PdV (5.
4. Since there is a distinct approximation to Q(x y z) in each control volume. Evaluate F(Q) at the boundary. A rQdA = C nQdl I (5.5) or. 3. the following relation between rQ and Q is sometimes used: Z I rQdV = nQdS (5. MODEL EQUATIONS IN INTEGRAL FORM 73 1.2.2. Using this approximation. .2 Model Equations in Integral Form 5. V S Z where the unit vector n points outward from the surface or contour.4. In order to include di usive uxes.1 The Linear Convection Equation A twodimensional form of the linear convection equation can be written as @u + a cos @u + a sin @u = 0 @t @x @y (5.2.5. These ideas should be clari ed by the examples in the remainder of this chapter. The onedimensional form is recovered with = 0. two distinct values of the ux will generally be obtained at any point on the boundary between two control volumes.6) 5. in two dimensions. nd Q at the controlvolume boundary. The order of accuracy of the method is dependent on each of the approximations.7) This PDE governs a simple plane wave convecting the scalar quantity. construct an approximation to Q(x y z) in each control volume. This issue is discussed in Section 11. Given the value of Q for each control volume. u(x y t) with speed a along a straight line making an angle with respect to the xaxis. 2. Integrate the ux to nd the net ux through the controlvolume boundary using some sort of quadrature. Apply some strategy for resolving the discontinuity in the ux at the controlvolume boundary to produce a single value of F(Q) at any point on the boundary. Advance the solution in time to obtain new values of Q.
16) 5. with 5.2.1=2 = xj .15) ! to be the integral form of the twodimensional di usion equation.2 gives the following integral form (5. The integral form of the twodimensional di usion equation with no source term and unit di usion coe cient is obtained from the general divergence form. Control volume j extends from xj . we nd (5. Eq.13) (5. 2.9) (5. Substituting these expressions into a twodimensional form of Eq. as shown in Fig.2 The Di usion Equation Q = u F = . 5. d Z udA = I n: i @u + j @u ds dt A @x @y C (5. 2.3.10) Since Q is a scalar.ru ! @u + j @u = .3. 2.17) . We consider an equispaced grid with spacing x. x=2 to xj + x=2.1. as in the model equations.3 OneDimensional Examples We restrict our attention to a scalar dependent variable u and a scalar ux f .14) (5. F is simply a vector. Eq. x=2 xj+1=2 = xj + x=2 (5. with Q = u F = iu cos + ju sin P = 0 d Z udA + I n:(iu cos + ju sin )ds = 0 dt A C (5.8) (5. the twodimensional linear convection equation is obtained from the general divergence form. i @x @y P = 0 Using these. We will use the following notation: xj.11) where A is the area of the cell which is bounded by the closed contour C . The nodes of the grid are located at xj = j x as usual.12) (5.74 CHAPTER 5. FINITEVOLUME METHODS For unit speed a.
18) (5. the cellaverage value becomes 1 Z xj+1=2 u(x t)dx u (t) j uj 1=2 = u(xj 1=2) x fj 1=2 = f (uj 1=2) (5. fj. xj .3.1: Control volume in one dimension. Hence the cellaverage value and the value at the center of the cell di er by a term of second order.1=2 = 0 .11. the integral form of the linear convection equation.1=2 or (5. 5.1=2 = x j+1=2 Pdx (5.23) x ddtj + fj+1=2 .21) 4 j 1920 @x j and the integral form becomes xj. fj.20) j j +1=2 dt j .1=2 Now with = x . uj = uj + O( x2) 5.5. ONEDIMENSIONAL EXAMPLES j1/2 ∆ x j+1/2 75 j2 j1 L R j L R j+1 j+2 Figure 5.1 A SecondOrder Approximation to the Convection Equation In one dimension. Eq. With these de nitions. 5. becomes u (5. we can expand u(x) in Eq.19) Zx d ( xu ) + f . x=2 j @x j 2 @x2 j 6 @x3 j ! ! x2 @ 2 u + x4 @ 4 u + O( x6 ) = uj + 24 @x2 (5.3.19 in a Taylor series about xj (with t xed) to get 2 3 ! ! ! 1 Z x=2 4u + @u + 2 @ 2 u + 3 @ 3 u + : : :5 d uj x .22) where uj is the value at the center of the cell.
76 CHAPTER 5. We choose a piecewise constant approximation to u(x) in each cell such that u(x) = uj xj. which is cell j + 1. Substituting Eqs.32) dt 2 x p .1=2 .29 and 5.27) . this point operator produces the following semidiscrete form: d~ = . 1 (uj.1=2 = 2 (fjL.1=2 x xj+1=2 (5.1 + uj ) (5.1 0 1)~ u u (5.1=2 = uj.26) +1 Similarly.1=2. gives fjR =2 = uj+1 (5.30) . giving fjR 1=2 = uj (5. 2 where f^ denotes a numerical ux which is an approximation to the exact ux.23. 1 B (.1.28) We have now accomplished the rst step from the list in Section 5.30 into the integral form. and cell j . The cell to the right of xj +1=2. In this example. uj.24) Evaluating this at j + 1=2 gives fjL+1=2 = f (uL+1=2) = uL+1=2 = uj (5. as shown in Fig.29) +1 2 and 1 f^j.25) j j where the L indicates that this approximation to fj+1=2 is obtained from the approximation to u(x) in the cell to the left of xj+1=2 .1 (5. FINITEVOLUME METHODS with f = u. giving fjL. cell j is the cell to the right of xj. Eq. 1 is the cell to the left of xj. 5. the discontinuity in the ux at the cell boundary is resolved by taking the average of the uxes on either side of the boundary. Thus 1 f^j+1=2 = 2 (fjL+1=2 + fjR =2) = 1 (uj + uj+1) (5.1 we have de ned the uxes at the cell boundaries in terms of the cellaverage data.1=2 + fjR 1=2) = 1 (uj.1) = 0 (5. we obtain u 1 u x ddtj + 2 (uj + uj+1) . 5.1 + uj ) = x ddtj + 1 (uj+1 .31) 2 2 With periodic boundary conditions. 5.
x=2 u( )d = u x . 5.1 with a piecewise quadratic approximation as follows u( ) = a 2 + b + c (5. 5. ONEDIMENSIONAL EXAMPLES 77 This is identical to the expression obtained using secondorder centered di erences. we can conclude that the use of the average of the uxes on either side of the cell boundary. x=2 x=2 u( )d = uj (5. The three parameters a. and c.22ux2+ uj.5.3.1 0 1) is relevant to nitevolume methods as well as nitedi erence methods. b.1) j 6 c = . except it is written in terms of the cell average ~ .x j.3.1 + 26uj .36) .1 (5.3 x=2 1 j .3.29 and 5.30. xj . b. the piecewise quadratic approximation produces (5. rather than the nodal values. u u Hence our analysis and understanding of the eigensystem of the matrix Bp(. and c are chosen to satisfy the following constraints: 1 Z .1 0 1) are pure imaginary.uj. uj+1 24 With these values of a. can lead to a nondissipative nitevolume method.34) 1 These constraints lead to x Z 3 x=2 u( )d = uj+1 j a = uj+1 . uj. Since the eigenvalues of Bp(.2 A FourthOrder Approximation to the Convection Equation Let us replace the piecewise constant approximation in Section 5. as in Eqs.35) the following values at the cell boundaries: uL+1=2 = 1 (2uj+1 + 5uj .1 u b = uj+12 .1 x Z x=2 . ~ .33) where is again equal to x .
1=2)] j j = 1 (.78 CHAPTER 5.uj+2 + 8uj+1 .uj+2 + 5uj+1 + 2uj ) j 6 1 uL.2) 12 (5.1 + uj.39) using the notation de ned in Section 5. we describe two approaches to deriving a nitevolume approximation to the di usion equation. Eq. we again use the average of the uxes on either side of the boundary to obtain f^j+1=2 = 1 f (uL+1=2) + f (uR+1=2)] j 2 j 1 (5.8 0 8 .uj+1 + 7uj + 7uj. With periodic boundary conditions. uj.1)~ u u (5. the following semidiscrete form is obtained: d~ = .38) (5. gives u 1 x ddtj + 12 (.1) j 6 uR+1=2 = 1 (.1=2 = 6 (2uj + 5uj.2) = 0 (5. uj.1=2 = 1 (. 5.2) j (5. Recalling that f = u. The rst approach is simpler to extend to multidimensions.1=2) + f (uR.3 A SecondOrder Approximation to the Di usion Equation .1=2 = 2 f (uL.1) and 1 f^j.37) (5. while the second approach is more suited to extension to higher order accuracy.1 .23. In this section.42) This is a fourthorder approximation to the integral form of the equation. 8uj. FINITEVOLUME METHODS uR. 1 B (1 .40) = 12 (.3.41) Substituting these expressions into the integral form. as can be veri ed using Taylor series expansions (see question 1 at the end of this chapter).1. 5.43) dt 12 x p This is a system of ODE's governing the evolution of the cellaverage data.uj+1 + 5uj + 2uj. uj.uj+2 + 7uj+1 + 7uj .1 .3.
1=2 = . u ) (5. 5.ru = . u ) (5.1) dt x j . becomes with f = . Eq. uj.2 1) are relevant to the study of nitevolume methods as well as nitedi erence methods. we obtain x duj = 1 (u . Hence.2 1)~ + bc u u ~ 2 f^j.2. the integral form of the di usion equation. we use a piecewise quadratic approximation as in Section 5.1 j (5.16.49) (5. Eq. d~ = 1 B (1 .5. 1x (uj . we can write ! @u f^ = .46) @u dx = u(b) . For our second approach. to secondorder.50) j +1 This provides a semidiscrete nitevolume approximation to the di usion equation.6 becomes a We can thus write the following expression for the average value of the gradient of u over the interval xj x xj+1: 1 Z xj+1 @u dx = 1 (u .44) (5. ONEDIMENSIONAL EXAMPLES u x ddtj + fj+1=2 . Also.51) . u(a) @x x xj @x x j+1 j From Eq. From Eq. fj. 5. = .33 we have dt x @u = @u = 2a + b @x @ (5. Substituting these into the integral form.45) In one dimension. 5.3. we know that the value of a continuous function at the center of a given interval is equal to the average value of the function over the interval to secondorder accuracy.47) j +1=2 @x j+1=2 x j+1 j Similarly. and we see that the properties of the matrix B (1 .3. 2u + u ) or.1=2 = 0 Zb 79 (5. with Dirichlet boundary conditions.@u=@x. 5.44. Eq.48) (5. 5. 1 (u .22.
53) Notice that there is no discontinuity in the ux at the cell boundary. The resulting semidiscrete form with periodic boundary conditions is d~ = 1 B (1 .3. 1 (u . u ) x j+1 j (5. this gives f R = f L = . 2u + u ) (5. .54) which is identical to Eq. and ` is the length of the sides of the hexagons. FINITEVOLUME METHODS j +1=2 j +1=2 with a and b given in Eq. 5.@u=@x. 5.56) A 3 The twodimensional form of the conservation law is d Z QdA + I n:Fdl = 0 dt A C (5. 1x (uj .1) . uj. we use a piecewise constant approximation in each control volume and the ux at the control volume boundary is the average of the uxes obtained on either side of the boundary.57) . The control volumes are regular hexagons with area A. as shown in Figure 5. Thus our nal example is a nitevolume approximation to the twodimensional linear convection equation on a grid consisting of regular triangles.1=2 = . The following relations hold between `.1 j j +1 dt x2 p 5. With f = . fjR 1=2 = fjL. As in Section 5.49.1.35.55) which is written entirely in terms of cellaverage data.80 CHAPTER 5.2.52) (5.2 1)~ u u (5. This produces duj = 1 (u . and A. The nodal data are stored at the vertices of the triangles formed by the grid.4 A TwoDimensional Example The above onedimensional examples of nitevolume approximations obscure some of the practical aspects of such methods. is the length of the sides of the triangles. dt x2 j . 1 ` = p 3 p A = 3 2 3 `2 ` = 2 (5.
1.i + 3 j)=2 . Since these sides are all straight. 3 j)=2 pi (i + p j)=2 3 (.5. . as shown in Figure 5. 5. 3 j)=2 p Table 5.i p (. the unit normals can be taken outside the integral and the ux balance is given by 5 d Z Q dA + X n dt A =0 Z Fdl = 0 where indexes a side of the hexagon.2: Triangular grid. i and j are unit normals along x and y. where we have ignored the source term. Outward normals. A list of the normals for the mesh orientation shown is given in Table 5.2.4.1. see Fig.i . Side 0 1 2 3 4 5 Outward Normal n (i .2. respectively. A TWODIMENSIONAL EXAMPLE c b 2 3 81 p d 4 1 l ∆ 0 5 a e f Figure 5. The contour in the line integral is composed of the sides of the hexagon.
the twodimensional linear convection equation. Introducing the cell average.2.1=2 (5.82 CHAPTER 5. the integrated time rate of change of the integral of u over the area of the hexagon is known exactly.62) 2 1 A u dA = Aup (5.61) . 5. 5. Z and the piecewiseconstant approximation u = up over the entire hexagon. Weights of ux integrals. FINITEVOLUME METHODS For Eq.`=2 u ( )d (5. 5. in terms of u and the hexagon area A.1=2 u(z)dz (5. the approximation to the ux integral becomes trivial. Making the change of variable z = =`. if the integrals in the equation are evaluated exactly. Side 0 1 2 3 4 5 n (i cos + j sin ) (cos .2. we have for side n Z Fdl = n (i cos + j sin ) Z `=2 . 3 sin )=2 cos p (cos + p sin )=2 3 (. cos + 3 sin )=2 .60) dt A . cos p (.58) where is a length measured from the middle of a side .59) Then. That is. 3 sin )=2 p Table 5. There .`=2 Z `=2 Z 1=2 are no numerical approximations in Eq.60.11.60. we have " # 5 d Z u dA + X n (i cos + j sin ) ` Z 1=2 u(z)dz = 0 (5. cos . see Eq. Taking the average of the ux on either side of each edge of the hexagon gives for edge 1: Z up + ua Z 1=2 dz = up + ua u(z)dz = 2 . one has the expression u( )d = ` .1=2 =0 The values of n (i cos + j sin ) are given by the expressions in Table 5.
u ) + (cos + p3 sin )(u . 5. 5.5 Problems 1. uf )] = 0 (5. that this is a secondorder approximation to the integral form of the twodimensional linear convection equation. 2. Use the following linear approximation: u( ) = a + b . Find the semidiscrete ODE form governing the cellaverage data resulting from the use of a linear approximation in developing a nitevolume method for the linear convection equation. using Taylor series expansions.2.66) (5. PROBLEMS Similarly. we obtain u(z)dz = up + uf 2 0 p u ` A ddtp + 2 (2 cos )(ua . 5.60.63) (5.68) or dup + 1 (2 cos )(u .67) Z Z u(z)dz = up + ud 2 4 u(z)dz = up + ue 2 5 Z Z Substituting these into Eq. Use Taylor series to verify that Eq. ud) + (cos + 3 sin )(ub . u ) a d b e dt 3 p +(. ue) p +(. cos + 3 sin )(uc .65) (5. we have for the other ve edges: Z 83 u(z)dz = up + ub 2 2 u(z)dz = up + uc 2 3 (5.42 is a fourthorder approximation to Eq.64) (5.5. along with the expressions in Table 5.5. uf )] = 0 (5.69) The reader can verify. 5. cos + 3 sin )(uc .23.
derive a nitevolume approximation to the spatial terms in the twodimensional di usion equation on a square grid.x j.1 and use the average ux at the cell interface.3. FINITEVOLUME METHODS u a = uj+12 . . Repeat question 3 for a grid consisting of equilateral triangles. 3. Using the rst approach given in Section 5.3.84 where b = uj and CHAPTER 5. 4.
the goal is simply to remove the transient portion of the solution as quickly as possible timeaccuracy is not required. which will be discussed in Chapter 9. For a steady ow problem. or directly using Gaussian elimination or some variation thereof. This produces an iterative method in which a coupled system of linear algebraic equations must be solved at each iteration. These can be solved iteratively using relaxation methods. one can consider a timedependent path to the steady state and use a timemarching method to integrate the unsteady form of the equations until the solution is su ciently close to the steady solution. spatial discretization leads to a coupled system of nonlinear algebraic equations in the form ~u F (~ ) = 0 (6. 85 d~ = F (~ t) u ~u dt (6.2) As a result of the nonlinearity of these equations. some sort of iterative method is required to obtain a solution.1) . is thus relevant to both steady and unsteady ow problems. For example. When using a timemarching method to compute steady ows. topics which are covered in the next two chapters.Chapter 6 TIMEMARCHING METHODS FOR ODE'S After discretizing the spatial derivatives in the governing PDE's (such as the NavierStokes equations). one can consider the use of Newton's method. we obtain a coupled system of nonlinear ODE's in the form These can be integrated in time using a timemarching method to obtain a timeaccurate solution to an unsteady ow problem. The subject of the present chapter. which is widely used for nonlinear algebraic equations (See Section 6.10.3. This motivates the study of stability and sti ness.). Alternatively. timemarching methods for ODE's.
but the methods themselves are formed by linear combinations of the dependent variable and its derivative at various time intervals. and.3 gives u0n = Fn = F (un tn) tn = nh Often we need a more sophisticated notation for intermediate time steps involving temporary calculations denoted by u. Our rst task is to nd numerical approximations that can be used to carry out the time integration of Eq. 4. but we postpone such considerations to the next chapter. we introduced the convention that the n subscript. They are both commonly used in the literature on ODE's. In earlier chapters.3) . for the purpose of this chapter. We then face a further task concerning the numerical stability of the resulting methods. we reduce our PDE to a set of coupled ODE's represented in general by Eq. 6.1 Notation Using the semidiscrete approach. 6. Application of a timemarching method to an ODE produces an ordinary di erence equation (O E ).4) un+1 = f 1 u0n+1 0u0n . In Chapter 2.3 to some given accuracy. 6.1u0n. For these we use the notation ~ ~ u0n+ = Fn+ = F (~n+ tn + h) ~ u The choice of u0 or F to express the derivative in a scheme is arbitrary. and h represents the time interval t. Combining this notation with Eq. we developed exact solutions to our model PDE's and ODE's. TIMEMARCHING METHODS FOR ODE'S Application of a spatial discretization to a PDE produces a coupled system of ODE's.86 CHAPTER 6.1 0 un .1 un.1 du = u0 = F (u t) dt (6. using this theory. or the (n) superscript. The methods we study are to be applied to linear or nonlinear ODE's. However. etc. In this chapter we will present some basic theory of linear O E's which closely parallels that for linear ODE's. always points to a discrete time value. we need only consider the scalar case Although we use u to represent the dependent variable. They are represented conceptually by (6. the reader should recall the arguments made in Chapter 4 to justify the study of a scalar ODE. rather than w. where accuracy can be measured either in a local or a global sense.1. we will develop exact solutions for the model O E's arising from the application of timemarching methods to the model ODE's. u.
5) (6.8) dt written here in terms of the dependent variable. The second is implicit and referred to as the implicit (or backward) Euler method. For an implicit method.1. will be presented in Section 11. for systems of ODE's and nonlinear problems.6. and therefore the time advance is simple. which is a fullydiscrete method.2. at given time intervals.2 Converting TimeMarching Methods to O E's Examples of some very common forms of methods used for timemarching general ODE's are: un+1 = un + hu0n un+1 = un + hu0n+1 and (6. by applying a timemarching method. the rst and third of these are examples of explicit methods. An explicit method is one in which the new predicted solution is only a function of known data. u0n. The value of this equation arises from the fact that.6) According to the conditions presented under Eq. 6. We refer to them as the explicit Euler method and the MacCormack predictorcorrector method. that is. CONVERTING TIMEMARCHING METHODS TO O E'S 87 With an appropriate choice of the 0s and 0s.1 for a method using two previous time levels. As we shall see. The methods are said to be explicit if 1 = 0 and implicit otherwise.7) du = u0 = u + ae t (6. The method commonly referred to as . we can analytically convert MacCormack's method. these methods can be constructed to give a local Taylor series accuracy of any order. u0n. u.3 1 Here we give only MacCormack's timemarching method. un and un. the new predicted solution is also a function of the time derivative at the new time level. The material presented in Chapter 4 develops a basis for evaluating such methods by introducing the concept of the representative equation un+1 = un + hu0n ~ 1 ~ ~ un+1 = 2 un + un+1 + hu0n+1] (6.1 respectively. for example. implicit methods require more complicated strategies to solve for un+1 than explicit methods.4. These methods are simple recipes for the time advance of a function in terms of its value and the value of its derivative. u0n+1. 6.
Apply the simple explicit Euler scheme. 6. 6. We next consider examples of this conversion process and then go into the general theory on solving O E's.7. 6.11) .5. the predictorcorrector sequence. As another example. with constant coe cients. 6. (1 + h)un = ahe hn ~ 1 u (6. to Eq. Eq. The second line in Eq. 2 (1 + h)~n+1 + un+1 . 6. and just as powerful as. The latter are subject to a whole body of analysis that is similar in many respects to.12) Now we need to develop techniques for analyzing these di erence equations so that we can compare the merits of the timemarching methods that generated them. expressed in terms of the dependent variable un and the independent variable n. 1 un = 1 ahe h(n+1) 2 2 which is a coupled set of linear O E's with constant coe cients. TIMEMARCHING METHODS FOR ODE'S such a linear ODE into a linear O E .7 is simply the explicit Euler method. (1 + h)un = hae hn (6. we nd un+1 = un + h un+1 + ae h(n+1) or (1 .2. Eq. since the predictor step in Eq.8.9 is a linear O E . Note that the rst line in Eq. un = he h ae hn (6. 6. This is demonstrated by repeating some of the developments in Section 4. but for di erence rather than di erential equations. 6. h)un+1 .11 is obtained by noting that u0n+1 = F (~n+1 tn + h) ~ u = un+1 + ae h(n+1) ~ (6.9) Eq. 6.88 CHAPTER 6.11 is identical to Eq. Eq.8.6.3 Solution of Linear O E's With Constant Coe cients The techniques for solving linear di erence equations with constant coe cients is as well developed as that for ODE's and the theory follows a remarkably parallel path. applying the implicit Euler method. gives un+1 . to Eq. 6. .10) As a nal example. There results un+1 = un + h( un + ae hn) or un+1 .9. 6. 6. the theory of ODE's.
and b are. (b = ).10. we use for O E's the di erence operator E (commonly referred to as the displacement or shift operator) and de ned formally by the relations un+1 = Eun un+k = E k un Further notice that the displacement operator also applies to exponents. 6.13) where . a. one can readily show that the solution of the defective case. un = u0 n + a b b .14) d Instead of the di erential operator D dt used for ODE's. n un = c1 n + bab . and since the equations are linear and have constant coe cients.3. The independent variable is now n rather than t. The exact solution of Eq. un+1 = un + a n is h i un = u0 + an . n.1 First.1 n This can all be easily veri ed by substitution.13 is where c1 is a constant determined by the initial conditions. complex parameters. SOLUTION OF LINEAR O E'S WITH CONSTANT COEFFICIENTS 89 6. is not a function of either n or u. 4. thus b bn = bn+ = E bn .3. n SecondOrder Equations The homogeneous form of a secondorder di erence equation is given by un+2 + a1 un+1 + a0un = 0 (6.and SecondOrder Di erence Equations FirstOrder Equations The simplest nonhomogeneous O E of interest is given by the single rstorder equation un+1 = un + abn (6.6. in general. In terms of the initial value of u it can be written Just as in the development of Eq.
Eq. E ) u2 which must be zero for all u1 and u2. there are just two roots and in terms of them the solution can be written un = c1( 1 )n + c2( 2 )n (6. 2 . we label these roots 1 . In the simple example given above. 6.90 CHAPTER 6. 6.15 is known as the operational form of Eq.15) which must be zero for all un. rstorder. The operational form contains a characteristic polynomial P (E ) which plays the same role for di erence equations that P (D) played for di erential equations that is. linear homogeneous di erence equations have the form u(1n+1) = c11u(1n) + c12 u(2n) u(2n+1) = c21u(1n) + c22 u(2n) ~ n = hu(1n) u(2n)iT u c12 (6. Again we are led to a characteristic polynomial. In the analysis of O E's.14 for all c1 c2 and n should be veri ed by substitution. its roots determine the solution to the O E.2 Special Cases of Coupled FirstOrder Equations A Complete System Coupled. ) (c c12 ) = 0 21 22 .17) which can also be written ~ n+1 = C~ n u u (c11 . E I ]. and refer to them as the roots. . They are found by solving the equation P ( ) = 0.14.16 is a solution to Eq. E I ]~ = 0 un c21 (c22 . 6. this time having the form P (E ) = det C .17 can be written u1 (n) = C .14 can now be reexpressed in an operational notion as (E 2 + a1E + a0)un = 0 (6.3. E ) C = c11 c12 c21 c22 The operational form of Eq. The roles of D and E are the same insofar as once they have been introduced to the basic equations the value of u(t) or un can be factored out. 6. 6.16) where c1 and c2 depend upon the initial conditions. 6. The fact that Eq. The roots are found from P ( ) = det (c11c . etc. TIMEMARCHING METHODS FOR ODE'S where can be any fraction or irrational number. . 6. Thus Eq.
2 un = u0 + u0n ^ 0 2 n (6.4 Solution of the Representative O E's 6. one can show that the solution to A Defective System 2 3 2 un+1 7 6 6 un+1 5 = 4 1 4^ un+1 1 32 3 7 6 un 7 ^ 5 4 un 5 un is un = u0 n h i un = u0 + u0n .4.4.20) (6. h)E .6.18) 6.1 The Operational Form and its Solution Examples of the nonhomogeneous. following the logic of Section 4. 6. linear.11. E . the solution of Eq.9 to 6. For example. 1 u n .19) (6. 6. 1) . ~ are its eigenvectors.1 n ^ ^ " # .2.2 for defective ODE's. 1]un = h E ae hn (6.17 is if x 2 ~ n = X ck ( k )n~ k u x k=1 where ck are constants determined by the initial conditions.2 2 2 E . Using the displacement operator.(1 + h) # " u # = h " 1 # ae hn ~ 1E 1 (1 + h)E E . these equations can be written E . are given by Eqs.1 + u n(n .2. The solution of O E's with defective eigensystems follows closely the logic in Section 4. rstorder ordinary di erence equations. SOLUTION OF THE REPRESENTATIVE O E'S 91 Obviously the k are the eigenvalues of C and. (1 + h)]un = h ae hn (1 .21) " . produced by applying a timemarching method to the representative equation.
P ( ) = 0. ~ the important subset of this solution which occurs when = 0. 1 Q(E ) = hE e h. 4. respectively.21.19 to 6. The terms P (E ) and Q(E ) are polynomials in E referred to as the characteristic polynomial and the particular polynomial. the ratio Q(E )=P (E ) can be found by Kramer's rule.45.20. we derive the solutions of Eqs.4. Eq. The general solution of Eq. 6.22 and 6. h)E .22 all manner of standard timemarching methods having multiple time steps and various types of intermediate predictorcorrector families.92 CHAPTER 6. Notice also.23: un = c1(1 + h)n + ae hn For the implicit Euler method.24) and the solution of its representative O E follows immediately from Eq. Keep in mind that for methods such as in Eq. For the explicit Euler method. Eq.23. 6. When determinants are involved in the construction of P (E ) and Q(E ).23) P (e ) k=1 where k are the K roots of the characteristic polynomial. we have P (E ) = (1 . 6. Eq. 6. We can express in terms of Eq. TIMEMARCHING METHODS FOR ODE'S All three of these equations are subsets of the operational form of the representative O E P (E )un = Q(E ) ae hn (6. 6. 6.1. one for un and un and we are usually only interested in the nal solution un. h Q(E ) = h (6.22 can be expressed as K h X un = ck ( k )n + ae hn Q(e h) (6.19. In such a case K X Q(1) un = ck ( k )n + a P (1) k=1 6.2 Examples of Solutions to TimeMarching O E's As examples of the use of Eqs. as would be the case for Eq.21. 1 . 6. we have P (E ) = E .21 there are multiple (two in this case) solutions.25) .22) which is produced by applying timemarching methods to the representative ODE. representing a time invariant particular solution. 6. h h (6. 6. or a steady state.
1 . 1 2 2 2 h Q(E ) = det . h . 1 .(1 + h) = E E . 1 .1 Establishing the Relation 6. h)e h . The former are the eigenvalues of the A matrix in the ODE's found by space di erencing the original PDE.5 The .27. 1 (1 + h)E E .5. 1 (1 E h)E 1 hE = 1 hE (E + 1 + h) 2 2 + 2 The root is found from 1 P( ) = . The complete solution can therefore be written 1 h 1 2h2 n + ae hn 2 h e + 1 + h un = c1 1 + h + 2 (6. 1 6. and there ~ results # " E .26) e h .5. THE . the roots and the roots. 4.21. 1 2h2 2 he h 1 n un = c1 1 . Remembering that t = nh. and the latter are the roots of the characteristic polynomial in a representative O E found by applying a timemarching method to the representative ODE. Relation We have now been introduced to two basic kinds of roots.6.27) x . one can write n u(t) = c1 e 1 h ~ 1 + x + cm e mh n ~m + x + cM e Mh n ~ M + P:S: (6. 2 2 h2 = 0 which has only one nontrivial root ( = 0 is simply a shift in the reference index). h . h + ae hn (1 . There is a fundamental relation between the two which can be used to identify many of the essential properties of a timemarch method. h . RELATION so 93 In the case of the coupled predictorcorrector equations. one solves for the nal family un (one can also nd a solution for the intermediate family u). First we make use of the semidiscrete approach to nd a system of ODE's and then express its solution in the form of Eq. This relation is rst demonstrated by developing it for the explicit Euler method. Eq. 1 2h2 P (E ) = det . 6.
For one of these we nd m q h+ 1+ m 1 = 1 + mh + 2 = 2 h2 . Suppose. 6. 6.4. one root. we see a correspondence between m and e m h.28. 2 hE . so that for every the must satisfy the relation m .94 CHAPTER 6. 1. which is given by = 1 + h. q 1 + 2 h2 .27 and Eq.30) Now we notice that each produces two roots.8 to Eq.5.29) Applying Eq. So if we use the Euler method for the time advance of the ODE's.3. 1 = 0 2 (6. 1 4 h4 + m 2 h2 m (6.29. . 2 mh m . which is de ned by un+1 = un.31) (6. we have the characteristic polynomial P (E ) = E 2 . we use the leapfrog method for the time advance. 3 The error is O( 2 h2 ). TIMEMARCHING METHODS FOR ODE'S where for the present we are not interested in the form of the particular solution (P:S:). instead of the Euler method. Now the explicit Euler method produces for each root. m 2 Based on Section 4. the solution2 of the resulting O E is un = c1( 1 )n ~ 1 + x + cm ( m )n ~ m + x + cM ( M )n ~ M + P:S: x (6. 6.32) mh 8 m This is an approximation to e m h with an error O( 3h3).1 + 2hu0n (6. will be discussed in Section 6. 6.28) where the cm and the ~ m in the two equations are identical and m = (1 + m h). Since the value of e h can be expressed in terms of the series 1 e h = 1 + h + 2 2h2 + 1 3h3 + 6 1 + n! nhn + the truncated expansion = 1 + h is a reasonable3 approximation for small enough h. The other root. . x Comparing Eq.
6. We refer to the root that has the above property as the principal root. which is the leapfrog method: un+1 = un. 6.34.5. always produces one root for every root that satis es the relation (6.30 that the .2 The Principal Root Based on the above we make the following observation: Application of the same timemarching method to all of the equations in a coupled system linear ODE's in the form of Eq. However. and designate it ( m )1 . Consider a solution obtained at a given time T using a secondorder timemarching method with a time step h. RELATION 95 6. This arises simply because of our notation for the timemarching method in which we have multiplied through by h to get an approximation for the function un+1 rather than the derivative as in Eq.5.5. Since the error per time step is O(h3). The above property can be stated regardless of the details of the timemarching method. Thus the principal root is an approximation to e h up to O hk . this is reduced by a factor of eight (considering the leading term only). One of these we identi ed as the principal root which always . consistent with a secondorder approximation.6. while the secondorder timemarching method which results from this approximation. The following example makes this clear. 4.35) has a leading truncation error O(h3). 6. Now consider the solution obtained using the same method with a time step h=2. relation for the leapfrog method produces two roots for each .33) 1 2h2 + + 1 k hk + O hk+1 =1+ h+ 2 k! where k is the order of the timemarching method. Note that a secondorder approximation to a derivative written in the form ( t u)n = 1 (un+1 .34) 2h has a leading truncation error which is O(h2).3 Spurious Roots We saw from Eq. twice as many time steps are required to reach the time T . un.1 + 2hu0n (6. knowing only that its leading error is O hk+1 . Therefore the error at the end of the simulation is reduced by a factor of four.1) (6. THE .6.
u Presumably (i.96 CHAPTER 6. For example. in itself. Following again the message of Section 4.. if one starts the method properly) the spurious coe cients are all initialized with very small magnitudes. The initial vector ~ 0 u does not provide enough data to initialize all of the coe cients. To express this fact we use the notation = ( h). However. . All spurious roots are designated ( m )k where k = 2 3 . This results because methods which produce spurious roots require data from time level n . 1 or earlier to advance the solution from time level n to n + 1. 6. the leapfrog method requires ~ n.5.4. Such roots originate entirely from the numerical approximation of the timemarching method and have nothing to do with the ODE being solved. They are also called onestep methods. many very accurate methods in practical use for integrating some forms of ODE's have spurious roots. and presumably the magnitudes of the spurious roots themselves are all less than one (see Chapter 7). For example.28 must be modi ed. generation of spurious roots does not. relation produced by a timemarching scheme can result in multiple roots all of which.1 and thus cannot be started using u only ~ n. If a timemarching method produces spurious roots. 6. In general. they play virtually no role in accuracy analysis. it is always some algebraic function of the product h. No matter whether a root is principal or spurious.36) Spurious roots arise if a method uses data from time level n . are spurious. if there is one spurious root to a method. after some nite time the amplitude of the error associated with the spurious roots is even smaller then when it was initialized. Then the presence of spurious roots does not contaminate the answer. the . In fact.e. We refer to them as oneroot methods. Thus while spurious roots must be considered in stability analysis. That is. TIMEMARCHING METHODS FOR ODE'S has the property given in 6. we have un = c11 ( 1 )n ~ 1 + + cm1( m )n ~ m + + cM 1 ( M )n ~ M + P:S: 1x 1x 1x n~ + n~ + +c12 ( 1)2 x1 + cm2 ( m )2 xm + cM 2 ( M )n ~ M 2x n~ + n~ + +c13 ( 1)3 x1 + cm3 ( m )3 xm + cM 3 ( M )n ~ M 3x +etc. 1 or earlier. make a method inferior. 6. all of the coe cients (cm )2 in Eq. except for the principal one.4 OneRoot TimeMarching Methods There are a number of timemarching methods that produce only one root for each root. the solution for the O E in the form shown in Eq. if there are more spurious roots (6.36 must be initialized by some starting procedure.33. It should be mentioned that methods with spurious roots are not self starting. The other is referred to as a spurious root and designated ( m )2 ..
It is usually used as the basis for establishing the order of a method. Its . One is the error made in each time step.1~ u x x x f (6. )u0n + u0n+1 6. the socalled method. It is useful for comparing methods.4. it is of no use in: .6. These are ~ (t) = c1 e 1 h n ~ 1 + + cm e m h n ~ m + + cM e M h n ~ M + A. Three oneroot methods were analyzed in Section 6. is given by the formula h i un+1 = un + h (1 . a Taylor series analysis is a very limited tool for nding the more subtle properties of a numerical timemarching method. Notice that when t and n = 0. It is instructive to compare the exact solution to a set of ODE's (with a complete eigensystem) having timeinvariant forcing terms with the exact solution to the O E's for oneroot methods.37) respectively. They have the signi cant advantage of being selfstarting which carries with it the very useful property that the timestep interval can be changed at will throughout the marching process. see Section 3. respectively. h h . ACCURACY MEASURES OF TIMEMARCHING METHODS 97 1 The method represents the explicit Euler ( = 0).1~ u x x x f ~ n = c1( 1 )n ~ 1 + + cm( m )n ~ m + + cM ( M )n ~ M + A. as we shall see in Chapter 8.6 Accuracy Measures of TimeMarching Methods 6. so that all the constants. The other is the error determined at the end of a given event which has covered a speci c interval of time composed of many time steps. The only error made by introducing the time marching is the error that makes in approximating e h. relation is ) = 1 +1(1 .2. A popular method having this property. This is a global error. It is quite common to judge a timemarching method on the basis of results found from a Taylor table. For example.4. these equations are identical.1 Local and Global Error Measures There are two broad categories of errors that can be used to derive and evaluate timemarching methods.6. vectors. This is a local error such as that found from a Taylor table analysis. However.6. and matrices are identical except the ~ and the terms inside u the parentheses on the right hand sides. the trapezoidal ( = 2 ). and the implicit Euler methods ( = 1).
98 CHAPTER 6. which can be written as h un = c1 ( 1)n + ae hn Q(e h ) P (e ) (6.2 Local Accuracy of the Transient Solution (er Transient error er! ) The particular choice of an error measure. This is similar to the error found from a Taylor table. including only the contribution from the principal root. given by t u(t) = ce t + ae . analyzing the particular solution of predictorcorrector combinations. Therefore.relation we saw that all timemarching methods produce a principal root for every root that exists in a set of linear ODE's. evaluating numerical stability and separating the errors in phase and amplitude. (6. The latter three of these are of concern to us here. Our error measures are based on the di erence between the exact solution to the representative ODE.39) j j 6. a necessary condition for the choice should be that the measure can be used consistently for all methods. 1 The leading error term can be found by expanding in a Taylor series and choosing the rst nonvanishing term.6. TIMEMARCHING METHODS FOR ODE'S nding spurious roots. is to some extent arbitrary. However. a very natural local error measure for the transient solution is the value of the di erence between solutions based on these two roots. either local or global. The order of the method is the last power of h matched exactly. and to study them we make use of the material developed in the previous sections of this chapter. . nding the global error. In the discussion of the .38) and the solution to the representative O E's. We designate this by er and make the following de nition er e h.
1( h).ia . is found using = . ACCURACY MEASURES OF TIMEMARCHING METHODS 99 Amplitude and Phase Error Suppose a eigenvalue is imaginary. while a negative value corresponds to phase lead (the numerical phase speed is too large).1 (6. Then the numerical method must produce a principal root that is complex and expressible in the form (6. !h Cn x 6.40) 1 = r + i i ei!h From this it follows that the local error in amplitude is measured by the deviation of j 1 j from unity.3 Local Accuracy of the Particular Solution (er ) The numerical error in the particular solution is found by comparing the particular solution of the ODE with that for the O E. ) . Cn = ah= x.41) Amplitude and phase errors are important measures of the suitability of timemarching methods for convection and wave propagation phenomena. Introducing the Courant number. that is q era = 1 . tan. For such cases it is more meaningful to express the error in terms of amplitude and phase. Let = i! where ! is a real number representing a frequency.6.a . The approach to error analysis described in Section 3.42) er = er! = 1 + tan ( 1)i =( 1)r )] p where ! = . The principal root.5 can be extended to the combination of a spatial discretization and a timemarching method applied to the linear convection equation. ( 1)2 + ( 1 )2 i r er! !h . as follows . Thus one can obtain values of the principal root over the range 0 x for a given value of the Courant number. j 1 j = 1 .iCn x.1 ( 1 )i=( 1 )r )] and the local error in phase can be de ned as (6. we have h = . We have found these to be given by 1 P:S:(ODE) = ae t ( . A positive value of erp corresponds to phase lag (the numerical phase speed is too small).6.6. Such can indeed be the case when we study the equations governing periodic convection which produces harmonic motion. where is the modi ed wavenumber of the spatial discretization. The above expression for er! can be normalized to give the error in the phase speed.
46) (6. More precisely. An illustration of this is given in the section on RungeKutta methods.6. 6.100 and CHAPTER 6.47) . For a measure of the local error in the particular solution we introduce the de nition er P:S: h P:S:(O E) . this order is quite important in determining the true order of a timemarching method by the process that has been outlined. a timemarching method is said to be of order k if er = c1 ( h)k1+1 er = c2 ( h)k2+1 where k = smallest of(k1 k2) (6. is equal to zero. In order to determine the leading error term. )Q e h .43 can be written in terms of the characteristic and particular polynomials as er = co . timemarching methods are usually applied to nonlinear ODE's. where n ( .P e h o (6. and for this case. 6. 1 (ODE ) ( ) (6. many numerical methods generate an er that is also zero.43) The multiplication by h converts the error from a global measure to a local one. Eq.45) (6. The algebra involved in nding the order of er can be quite tedious. A necessary condition for this to occur is that the local accuracies of both the transient and the particular solutions be of the same order. so that the order of er and er are consistent. If the forcing function is independent of time. However.44) co = h!0 h( . and it is necessary that the advertised order of accuracy be valid for the nonlinear cases as well as for the linear ones.h ) lim P e The value of co is a methoddependent constant that is often equal to one.4 Time Accuracy For Nonlinear Applications In practice. TIMEMARCHING METHODS FOR ODE'S h P:S:(O E) = ae t Q(e h) P (e ) respectively.
Let T be the xed time of the event and h be the chosen step size.6. However. and q " ( 1 )2 + ( 1)2 i r N (6. given by the relation T = Nh Global error in the transient A natural extension of er to cover the error in an entire event is given by Er e T . N tan 1i 1r !# (6. These are given by Era = 1 .1 ( ) =( ) ] = !T . we are more concerned with the global error in amplitude and phase. Suppose we wish to compute some timeaccurate phenomenon over a xed interval of time using a constant time step. the analysis based on a linear nonhomogeneous ODE produces the appropriate conditions for the majority of timemarching methods used in CFD. Then the required number of time steps. ( 1 ( h))N (6.49) ( ) Er! N !h . We refer to such a computation as an \event".50) .5 Global Accuracy In contrast to the local error measures which have just been discussed. to derive all of the necessary conditions for the fourthorder RungeKutta method presented later in this chapter the derivation must be performed for a nonlinear ODE. For example. 6.6. we can also de ne global error measures.1 ( 1) i 1r . ACCURACY MEASURES OF TIMEMARCHING METHODS 101 The reader should be aware that this is not su cient.48) Global error in amplitude and phase If the event is periodic. These are useful when we come to the evaluation of timemarching methods for speci c purposes. This subject is covered in Chapter 8 after our introduction to stability in Chapter 7.6. tan. is N.
and the result is expressed in operational form. we have developed the framework of error analysis for time advance methods and have randomly introduced a few methods without addressing motivational.102 CHAPTER 6.K 1 0 1 X k k E Aun = h@ k=1. and they are said to be K step because K timelevels of data are required to marching the solution one timestep.1. we introduce classes of methods along with their associated error analysis. In the subsequent sections. since most of them have historic origins from a wide variety of disciplines and applications.K 1 k hn k E A( un + ae ) (6. TIMEMARCHING METHODS FOR ODE'S Global error in the particular solution Finally. The Linear Multistep Methods (LMM's) are probably the most natural extension to time marching of the space di erencing schemes introduced in Chapter 3 and can be analyzed for accuracy or designed using the Taylor table approach of Section 3. The methods are said to be linear because the 's and 's are independent of u and n. It can be measured by Er ( .1 P eh 6. the global error in the particular solution follows naturally by comparing the solutions to the ODE and the O E. 6. h.51) where the notation for F is de ned in Section 6.K k Fn+k (6. ) Qeh .51 is applied to the representative equation.1 The General Formulation When applied to the nonlinear ODE all linear multistep methods can be expressed in the general form 1 X k=1. 6.4. We shall not spend much time on development or design of these methods. one nds 0 1 @ X k=1. They are explicit if 1 = 0 and implicit otherwise. When Eq.8.7 Linear Multistep Methods In the previous sections.K k un+k = h du = u0 = F(u t) dt 1 X k=1. Eq. 6. developmental or design issues.7.52) .
4. can be designed using the Taylor table approach of Section 3. The threestep AdamsMoulton method can be written in the following form un+1 = un + h( 1u0n+1 + 0 u0n + .1 k=0 The AdamsBashforth family has the same 's with the additional constraint that 1 = 0.(.1 . un h u0n h2 u00 n h3 u000 n h4 u0000 n . labeled 1. LINEAR MULTISTEP METHODS 103 We recall from Section 6.(. 6.2 u0n. referred to as AdamsBashforth (explicit) and AdamsMoulton (implicit). .2)1 . Two wellknown families of them.1 .(.2 . a method should at least be rstorder accurate.2 1 6 Table 6. 6. 6. these methods are often \normalized" by requiring X k=1 Under this condition co = 1 in Eq.1 u0n.1. 6.53) 1=1 0 = . Taylor table for the AdamsMoulton threestep linear multistep method.1 2 .2 that a timemarching method when applied to the representative equation must provide a root.1 1 .un . The AdamsMoulton family is obtained from Eq.51 with k = .2)2 .1 u0n.6. and it is certainly a necessary condition for the accuracy of any time marching method.1 6 .h . to be of any value in time accuracy.51 if X X X k = 0 and k = (K + k .51 can be multiplied by an arbitrary constant.2 . 12 . 1) k Since both sides of Eq.2)0 . that is ! (1 + h) as h ! 0.1 .2 2 . that approximates e h through the order of the method.2 Examples There are many special explicit and implicit forms of linear multistep methods.54) A Taylor table for Eq.2 .2) (6.(.0 1 1 .5.1 . One can show that these conditions are met by any method represented by Eq.7.44.2)3 .2u0n.7.h 1 u0n+1 .1 1 .2 (6.54 can be generated as 1 1 1 un+1 1 1 2 6 24 .h 0 u0n . k k k k 6. 6. . The condition referred to as consistency simply means that ! 1 as h ! 0. 11 6 . We can also agree that.1 .h .1 + .
2 to solve for the 's.4 With 1 = 0 one obtains 2 32 3 2 3 1 1 1 0 7 617 6 0 .2 = 5=12 This is the thirdorder AdamsBashforth method.1 + 2hu0n+1i =3 2ndorder Backward h h5u0n+1 + 8u0n . 16u0n.56) 1 = 9=24 0 = 19=24 . One can verify that the Adams type schemes given below satisfy Eqs.1i = un + 2 h 0 i AB2 h = un + 12 23un .1 5 = 4 1 5 (6.1 + 5u0n. Explicit Methods un+1 un+1 un+1 un+1 = un + hu0n Euler 0 = un.1i = un + 12 AM3 .1 = .2 .32 1 . resulting in (6. 6.2 giving (6.5.1 7 = 6 1 7 5 4 5 4 0 . TIMEMARCHING METHODS FOR ODE'S This leads to the linear system 2 1 1 1 1 32 3 213 1 6 2 0 .5=24 . u0n. A list of simple methods.16=12 .2 . some of which are very common in CFD applications. u0n.4 7 6 0 7 6 1 7 6 7 6 7 6 3 0 3 12 7 6 76 (6. = un + hu0n+1 Implicit Euler 1 hhu0n + u0n+1i = un + 2 Trapezoidal (AM2) 1 h4un .55 and 6.1 = .104 CHAPTER 6.2 AB3 Implicit Methods un+1 un+1 un+1 un+1 4 Recall from Section 6.57) 4 54 0 3 12 1 .4 .2 that a kthorder timemarching method has a leading truncation error term which is O(hk+1 ).57 up to the order of the method.55) 4 5 4 .1 + 2hhun Leapfrog 1 h 3u0n .4 7 6 . is given below together with identifying names that are sometimes associated with them.58) 0 = 23=12 . un. In the following material AB(n) and AM(n) are used as abbreviations for the (n)th order AdamsBashforth and (n)th order AdamsMoulton methods.2 = 1=24 which produces a method which is fourthorder accurate.
1=3 . the methods are 2ndorder accurate) if 1 '= .1=2 .59.1=4 . ' 1=2 1 3=4 1=3 1=2 5=9 0 0 0 1=3 5=12 1=6 0 1 0 0 0 1=2 0 .1 = 0 in Eq.51.1=2. un. see Eq. LINEAR MULTISTEP METHODS 105 6. +2 The class of all 3rdorder methods is determined by imposing the additional constraint =2 .1 (6. K=2 in Eq.5 6 1 Finally a unique fourthorder method is found by setting = .1=2 .59 is given in Table 6.7.e.2. One can show after a little algebra that both er and er are reduced to 0(h3) (i. 6.59) Clearly the methods are explicit if = 0 and implicit otherwise..7. can be written as h i (1 + )un+1 = (1 + 2 )un . This limits the interest in multistep methods to about two time levels.51). 6. + ')u0n .2.' = . . 6.6.2=9 0 1=2 .1=2 0 0 0 0 .1=2 0 . which corresponds to 0 = 0 in Eq.1] + h u0n+1 + (1 . 6.1=6 Method Euler Implicit Euler Trapezoidal or AM2 2nd Order Backward Adams type Lees Type Two{step trapezoidal A{contractive Leapfrog AB2 Most accurate explicit Third{order implicit AM3 Milne Order 1 1 2 2 2 2 2 2 2 2 3 3 3 4 Table 6. Notice that the Adams methods have = 0. 'u0n. The most general twostep linear multistep method (i. are known as Milne methods. Some linear one..51.e.1=6 . that is at least rstorder accurate. 6.1=6 0 .1=2 .3 TwoStep Linear Multistep Methods High resolution CFD problems usually require very large data sets to store the spatial information from which the time derivative is calculated. =3 = 6 .5=6 . which corresponds to . Methods with = .and twostep methods.1=3 0 1=12 . A list of methods contained in Eq.
6. to the following observations. For this to hold for er .44 that the same conditions also make it O(h3). Their use is motivated by ease of application and increased e ciency. Predictorcorrector methods constructed to timemarch linear or nonlinear ODE's are composed of sequences of linear multistep methods. onecorrector example is given by un+ = un + h 0n ~ hu i un+1 = un + h u0n+ + u0n ~ (6. following the example leading to Eq. but it is not di cult to show using Eq.61 that + =1 =1 2 which provides two equations for three unknowns.60) where the parameters and are arbitrary parameters to be determined.26. TIMEMARCHING METHODS FOR ODE'S There are a wide variety of predictorcorrector schemes created and used for a variety of purposes. therefore. Their use in solving ODE's is relatively easy to illustrate and understand.6. One can analyze this sequence by applying it to the representative equation and using the operational techniques outlined in Section 6. For the method to be secondorder accurate both er and er must be O(h3).106 6. One concludes. There may be many families in the sequence. It is easy to show. each of which is referred to as a family in the solution process. where measures of e ciency are discussed in the next two chapters. and usually the nal family has a higher Taylorseries order of accuracy than the intermediate ones. . that h P (E ) = E E . The situation for er requires some algebra. that the predictorcorrector sequence is a secondorder accurate method for any .8 PredictorCorrector Methods CHAPTER 6.63) 5 Such as alternating direction. and hybrid methods. Their use in solving PDE's can be much more subtle and demands concepts5 which have no counterpart in the analysis of ODE's. fractionalstep. 1 u0n ~ (6.4. 1 . Q(E ) = E h E + + h] 2 h2 i (6. by following the discussion in Section 6. it is obvious from Eq.61) (6. un+ = un + hu0n ~ 1 un+1 = un + 2 h 1 u0n+ + 2 . one is led. ( + ) h . 6. 6.62) Considering only local accuracy. A simple onepredictor.
63 with = 1.6. speci c. 6. RUNGEKUTTA METHODS 107 A classical predictorcorrector sequence is formed by following an AdamsBashforth predictor of any order with an AdamsMoulton corrector having an order one higher.1 ~ h i un+1 = un + 1 h u0n + u0n+1 ~ ~ (6.1 (6. presented earlier in this chapter. we will consider only singlestep.66) (6. (6. u0n.63 with = 1=2 is 1 un+1=2 = un + 2 hu0n ~ un+1 = un + hu0n+1=2 ~ and.9 RungeKutta Methods There is a special subset of predictorcorrector methods. The Gazdag method.1 ~ i hh u un+1 = un + 12 5~0n+1 + 8u0n . The AdamsBashforthMoulton sequence for k = 3 is i 1 h un+1 = un + 2 h 3u0n .65) 2 The Burstein method. 6 Although implicit and multistep RungeKutta methods exist. which we discuss in Chapter 8. . nally.6 that produce just one root for each root such that ( h) corresponds explicit RungeKutta methods here. is 1 hu ~ i un+1 = un + 2 h 3~0n .68) 6. secondorder accurate methods are given below. u0n. referred to as RungeKutta methods. The order of the combination is then equal to the order of the corrector. MacCormack's method. is un+1 = un + hu0n ~ 1 un+1 = 2 un + un+1 + hu0n+1] ~ ~ Note that MacCormack's method can also be written as un+1 = un + hu0n ~ un+1 = un + 1 h u0n + u0n+1] ~ 2 from which it is clear that it is obtained from Eq. 6. u0n. we refer to these as ABM(k) methods. obtained from Eq. If the order of the corrector is (k).67) (6.64) Some simple.9.
. It is usually introduced in the form k1 = hF (un tn) k2 = hF (un + k1 tn + h) k3 = hF (un + 1k1 + 1k2 tn + 1 h) k4 = hF (un + 2k1 + 2k2 + 2 k3 tn + 2h) followed by u(tn + h) .108 CHAPTER 6. 6. u(tn) = 1 k1 + 2k2 + 3k3 + 4k4 (6. the choices for the time samplings.71 is b un+ = un + hu0n b un+ 1 = un + 1 hu0n + 1hu0n+ ~ b un+ 2 = un + 2 hu0n + 2hu0n+ + 2hu0n+ 1 ~ 0 + hu0 + hu0 (6.73) 2 = 2+ 2+ 2 . Thus for a RungeKutta method of order k (up to 4th order).70. we prefer to present it using predictorcorrector notation.71) However. To ensure k'th order accuracy. Thus. but.70) and this is much more di cult.6. a scheme entirely equivalent to 6. and 2 . The most widely publicized RungeKutta process is the one that leads to the fourthorder method. are not arbitrary. as we pointed out in Section 6. First of all.71 and 6. We present it below in some detail. the method must further satisfy the constraint that er = O(hk+1) (6. TIMEMARCHING METHODS FOR ODE'S to the Taylor series expansion of e h out through the order of the method and then truncates. the principal (and only) root is given by (6. 6.4.72) un+1 = un + 1hun 2 bn+ 3 ~n+ 1 + 4 hu0n+ 2 Appearing in Eqs. They must satisfy the relations = 1 = 1+ 1 (6. it is not su cient to guarantee k'th order accuracy for the solution of u0 = F (u t) or for the representative equation.69) = 1 + h + 1 2h2 + + 1 k hk 2 k! It is not particularly di cult to build this property into a method.72 are a total of 13 parameters which are to be determined such that the method is fourthorder according to the requirements in Eqs. 1.69 and 6.
we have to ful ll four more conditions 2 2 = 1=3 (3) 2 2+ 3 1+ 4 2 3+ 3 3+ 4 3 = 1=4 (4) 2 1 2 (6. Several choices for the parameters have been proposed. A more general derivation based on a nonlinear ODE can be found in several books. Using Eq.73 to eliminate the 's we nd from Eq. but the equations follow directly from nding P (E ) and Q(E ) and then satisfying the conditions in Eqs.74) These four relations guarantee that the ve terms in exactly match the rst 5 terms in the expansion of e h .69 and 6.74 and 6. As discussed in Section 6. Since the equations are overdetermined. but the most popular one is due to Runge.69 the four conditions 1+ 2 + 3 1+ = 1 2+ 3+ 4 = 1=2 3 1+ 4 2 ( 2 + 1 2 ) = 1=6 4 = 1=24 4 12 (1) (2) (3) (4) (6.7 There are eight equations in 6. RUNGEKUTTA METHODS 109 The algebra involved in nding algebraic equations for the remaining 10 parameters is not trivial. the resulting method would be thirdorder accurate. It results in the \standard" fourthorder RungeKutta method expressed in predictorcorrector form as 1 b un+1=2 = un + 2 hu0n b un+1=2 = un + 1 hu0n+1=2 ~ 2 un+1 = un + hu0n+1=2 ~ h i b un+1 = un + 1 h u0n + 2 u0n+1=2 + u0n+1=2 + u0n+1 ~ (6.75 are all satis ed. the fourth condition in Eq.70.75 which must be satis ed by the 10 unknowns.9.6. 6.6. 6. 7 The present approach based on a linear inhomogeneous equation provides all of the necessary . Thus if the rst 3 equations in 6. two parameters can be set arbitrarily.75 cannot be derived using the methodology presented here. which is based on a linear nonhomogenous representative ODE.74 and the rst equation in 6.75) 2 1 + 4 ( 2 2 + 2 2 ) = 1=12 (4) 3 1 (4) 3 1 1 + 4 2 ( 2 + 1 2 ) = 1=8 The number in parentheses at the end of each equation indicates the order that is the basis for the equation. 6.76) 6 conditions for RungeKutta methods of up to third order.4. To satisfy the condition that er = O(k5). 6.
Higherorder RungeKutta methods can be developed.75. These are covered in this Section. In any event.66 and 6. and thirdorder methods can be derived from Eqs.10 Implementation of Implicit Methods We have presented a wide variety of timemarching methods and shown how to derive their . Our presentation of the timemarching methods in the context of a linear scalar equation obscures some of the issues involved in implementing an implicit method for systems of equations and nonlinear equations. For example.78) . relations. It is clear that for orders one through four. Following the steps outlined in Section 6.74 and the rst equation in 6. we will see that these methods can have widely di erent properties with respect to stability. a fthorder method requires six evaluations to advance the solution one step. 6. but they require more derivative evaluations than their order. respectively.1 Application to Systems of Equations Consider rst the numerical solution of our representative ODE u0 = u + ae t (6.67 are secondorder RungeKutta methods.77) using the implicit Euler method. as Euler Predictor Euler Corrector Leapfrog Predictor Milne Corrector RK 4 One can easily show that both the Burstein and the MacCormack methods given by Eqs. TIMEMARCHING METHODS FOR ODE'S 9 > > = > > Notice that this represents the simple sequence of conventional linear multistep methods referred to.110 CHAPTER 6. We shall discuss the consequences of this in Chapter 8. h)un+1 . un = he h ae hn (6.2. 6. storage requirements reduce the usefulness of RungeKutta methods of order higher than four for CFD applications. RK methods of order k require k evaluations of the derivative function to advance the solution one time step.10. This leads to various tradeo s which must be considered in selecting a method for a speci c application. we obtained (1 . 6. 6. 6.72 by setting 4 = 0 and satisfying only Eqs. In the next chapter.
6. As an example. Now u u the equivalent to Eq. but rather we solve Eq. ~ n = .81) (6.79) This calculation does not seem particularly onerous in comparison with the application of an explicit method to this ODE..aBp (. hA).81 as a linear system of equations. The primary area of application of implicit methods is in the solution of sti ODE's.10. 6.1 ~ n . and hence its solution is inexpensive. but in multidimensions the bandwidth can be very large. h (un + he h ae hn ) 111 (6. IMPLEMENTATION OF IMPLICIT METHODS Solving for un+1 gives 1 un+1 = 1 . hf (t + h)] u u The inverse is not actually performed. For our onedimensional examples. f (t) u u ~ ~ (I .g. requiring only an additional division.80) ~ where ~ and f are vectors and we still assume that A is not a function of ~ or t.82) ~ ~ n+1 = (I .85) dt 2 . 6. the cost per time step of an implicit method is larger than that of an explicit method.84) un+1 = un + hF (un+1 tn+1) This is a nonlinear di erence equation. the system of equations which must be solved is tridiagonal (e.10.2 Application to Nonlinear Equations Now consider the general nonlinear scalar ODE given by Application of the implicit Euler method gives du = F (u t) dt (6. Now let us apply the implicit Euler method to our generic system of equations given by ~ 0 = A~ . A = . consider the nonlinear ODE du + 1 u2 = 0 (6. In general.78 is (6.1 0 1)=2 x). for biconvection. hA)~ n+1 .6.83) (6.hf (t + h) u u or (6. as we shall see in Chapter 8.
u )2 + @ 2 F (u . u )(t . tn)2 + @t2 n (6. can be used. tn) @t n @t2 n n 2 (6. the expansion of u(t) in terms of the independent variable t is ! ! @u + 1 (t . the \initial guess" for this nonlinear problem can be quite close to the solution. A Taylor series expansion about these reference quantities gives ! ! @F (u . which gives 1 un+1 + h 2 u2 +1 = un (6. t )2 @ 2 u + u(t) = un + (t . u ) + @F (t . Designate the reference value by tn and the corresponding value of the dependent variable by un. and Eq.89) .86) n which requires a nontrivial method to solve for un+1. u ) + @F (t . t ) + 2 @u2 n n n @u@t n !n 1 2 + 2 @ F (t . since the \initial guess" is simply the solution at the previous time step. we expand F (u t) about some reference point in time. In practice. In order to implement the linearization. which implies that a linearization approach may be quite successful. An iterative method. TIMEMARCHING METHODS FOR ODE'S solved using implicit Euler time marching.88) If t is within h of tn .112 CHAPTER 6. 6.10. t ) + O(h2) F (u t) = Fn + @u n n @t n n (6.3 Local Linearization for Scalar Equations General Development Let us start the process of local linearization by considering Eq.83. tn)k and (u . 6. both (t . such as Newton's method (see below). 6. t ) F (u t) = F (un tn) + @u n n @t n n ! ! 1 @ 2 F (u . Such an approach is described in the next Section.87 can be written ! ! @F (u .87) On the other hand. There are several di erent approaches one can take to solving this nonlinear di erence equation. un)k are O(hk ).
6. With this we obtain the locally (in the neighborhood of tn) timelinear representation of Eq.6. one nds un+1 = un + 1 h Fn + @F (un+1 . and the trapezoidal time march.90) Implementation of the Trapezoidal Method As an example of how such an expansion can be used. namely ! ! ! ! du = @F u + F .83.93) 1 . relative to the order of expansion of the function. 6.83. Using Eq. The trapezoidal method is given by 1 un+1 = un + 2 h Fn+1 + Fn] + hO(h2) (6. locallylinear approximation to F (u t) that is valid in the vicinity of the reference station tn and the corresponding un = u(tn).6.92 results in the expression 1 un = hFn + 1 h2 @F (6. In many uid mechanic applications the nonlinear function F is not an explicit function " ! ! # " !# ! . t ) + O(h2) n n dt @u n @u n n @t n (6. Thus. other secondorder implicit timemarching methods that can be used. of course.10.89 to evaluate Fn+1 = F (un+1 tn+1). combine to make a secondorderaccurate numerical integration process. A very useful reordering of the terms in Eq. 6. it represents a secondorderaccurate. The use of local time linearization updated at the end of each time step. 6. 2 h @F @u n 2 @t n which is now in the delta form which will be formally introduced in Section 12. There are.92) Note that the O(h2) term within the brackets (which is due to the local linearization) is multiplied by h and therefore is the same order as the hO(h2) error from the Trapezoidal Method. IMPLEMENTATION OF IMPLICIT METHODS 113 Notice that this is an expansion of the derivative of the function. un) + h @F + O(h2) + Fn 2 @u n @t n 2) +hO(h (6. The important point to be made here is that local linearization updated at each time step has not reduced the order of accuracy of a secondorder timemarching process.91) where we write hO(h2) to emphasize that the method is second order accurate. consider the mechanics of applying the trapezoidal method for the time integration of Eq. @F u + @F (t .
It is the product of h and the RHS of the basic equation evaluated at the previous time step. 6. 6. (Very seldom does one nd B . thus u u u ~ n+1 = ~ n + ~ n u u u The solution for ~ n+1 is generally stored such that it overwrites the value of ~ n u u and the process is repeated.95) if we introduce Eq. Solve for the elements of the matrix multiplying ~ n and store in some approu priate manner making use of sparseness or bandedness of the matrix if possible. Solve for the elements of hFn. and remove the explicit dependence on time.94 is usually implemented as follows: ~ ~ 1. it is generally the spacedi erenced form of the steadystate equation of some uid ow problem.93 simpli es to the secondorder accurate expression " 1 1 . Let this storage area be referred to as B . but for our applications.94) Notice that the RHS is extremely simple.114 CHAPTER 6.83. A numerical timemarching procedure using Eq. TIMEMARCHING METHODS FOR ODE'S of t. Solve the coupled set of linear equations B ~n = R u ~ for ~ n. In such cases the partial derivative of F (u) with respect to t is zero and Eq. 6. Find ~ n+1 by adding ~ n to ~ n.96) . rearrange terms. In this example. the basic equation was the simple scalar equation 6. Implementation of the Implicit Euler Method We have seen that the rstorder implicit Euler method can be written un+1 = un + hFn+1 " 1 . and save ~ n. 3. u 4. u 2.1 in carrying out this step).90 into this method. h @F @u (6. we arrive at the form !# n un = hFn (6. store them in an array say R. 2 h @F @u !# n un = hFn (6.
We shall see later that this method is an excellent choice for steady problems.10.96 obtained by dividing both sides by h and setting 1=h = 0.94 and 6.99) Here f represents a dimensionless stream function. we mean that the error after a given iteration is proportional to the square of the error at the previous iteration. 6. " Fn (6.1 n (6. We choose the FalknerSkan equations from classical boundarylayer theory. Newton's Method Consider the limit h ! 1 of Eq.94. Our task is to apply the implicit trapezoidal method to the equations d3f + f d2f + dt3 dt2 0 !21 @1 . Quadratic convergence is thus a very powerful property.87 and 6. @F @u ! n un = Fn @F @u ! #. let us consider an example involving some simple boundarylayer equations. . the error at a given iteration is some multiple of the error at the previous iteration. 6. i.4 Local Linearization for Coupled Sets of Nonlinear Equations In order to present this concept.10. Use of a nite value of h in Eq. where the error is the di erence between the current solution and the converged solution.96 leads to linear convergence. df A = 0 dt (6.6. 6. By quadratic convergence.98) This is the wellknown Newton method for nding the roots of a nonlinear equation F (u) = 0. 6.96. There results or .e. The fact that it has quadratic convergence is veri ed by a glance at Eqs. 6.88 (remember the dependence on t has been eliminated for this case). given by Eq.. Omission of this factor degrades the method in time accuracy by one order of h. IMPLEMENTATION OF IMPLICIT METHODS 115 We see that the only di erence between the implementation of the trapezoidal method 1 and the implicit Euler method is the factor of 2 in the brackets of the left side of Eqs. The reader should ponder the meaning of letting h ! 1 for the trapezoidal method. and is a scaling factor. 6.97) un+1 = un .
and tensor products for the terms of higher order. 6. It is derived from Eq.116 CHAPTER 6.87. 1 . one has 9 2.102 by the following process d~ = F (~ ) u ~ u dt (6. and de ning F n as F (~ n).8 Let us refer to this matrix as A.104) 8 Recall that we derived the Jacobian matrices for the twodimensional Euler equations in Section 9 The Taylor series expansion of a vector contains a vector for the rst term.103) 2 6 6 6 6 A=6 6 6 6 6 4 ~ u The expansion of F (~ ) about some reference state ~ n can be expressed in a way u similar to the scalar expansion given by eq 6.u1 u3 . 6. The required extension requires the evaluation of a matrix. called the Jacobian matrix.100) (6.2 vector for the second term. @F1 @u1 @F2 @u1 @F3 @u1 @F1 @u2 @F2 @u2 @F3 @u2 @F1 @u3 @F2 @u3 @F3 @u3 3 7 7 7 7 7 7 7 7 7 5 (6.99 to a set of rstorder nonlinear equations by the transformations This gives the coupled set of three nonlinear equations u01 = F1 = . u2 2 0 = F =u u2 2 1 0 = F =u u3 3 2 and these can be represented in vector notation as u2 = df dt u3 = f (6. except that this time we are faced with a nonlinear vector function. Omitting the explicit dependency ~ ~ u on the independent variable t.101) Now we seek to make the same local expansion that derived Eq. a matrix times a . rather than a simple nonlinear scalar function. 6.90.102) A = (aij ) = @Fi =@uj For the general case involving a third order matrix this is (6. TIMEMARCHING METHODS FOR ODE'S u1 = d f dt2 2 First of all we reduce Eq.
Find an expression for the nth term in the Fibonacci series. of course.11. 6. (1 . ~ n + O(h2) u u (6. Using this we can write the local linearization of Eq.u1 3 0 0 7 A=6 1 (6. which is given by 1 1 2 3 5 8 : : : Note that the series can be expressed as the solution to a di erence equation of the form un+1 = un + un. What is u25 ? (Let the rst term given above be u0.105) where t . tn and the argument for O(h2) is the same as in the derivation of Eq.93.6. d~ = A ~ + F .102 as \constant" which is a locallylinear. Returning to our simple boundarylayer example. 6. which is given by Eq.88.11 Problems 1. and the manner in which. Thus for the FalknerSkan equations the trapezoidal method results in 2 3 1 + h (u3)n . u3 2 u2 .1. on the nature of the problem. the terms in the Jacobian matrix are updated as the solution proceeds depends. u the solution will be secondorderaccurate in time. 6.106) 6. A ~ +O(h2) u ~ n n un nu dt  {z } (6. we nd the Jacobian matrix to be 2 .or secondorder timemarching method. 6. Without any iterating within a step advance. h(u2)n h (u1)n 72 ( u1)n 3 2 .107) 4 5 0 1 0 The student should be able to derive results for this example that are equivalent to those given for the scalar case in Eq. could be used to integrate the equations without loss in accuracy with respect to order.2 1 We nd ~ n+1 from ~ n + ~ n.) . Any rst.h (u1)n 74 6 5 4 5 1 0 76 ( u2)n 7 = h 6 2 4 5 ( u) (u2)n h 3n 0 .101. and the solution is now advanced one step. The number of times. secondorderaccurate approximation to a set of coupled nonlinear ordinary di erential equations that is valid for t tn + h. PROBLEMS 117 ~ u ~ F (~ ) = F n + An ~ . (u1u3)n . Reevaluate u u u the elements using ~ n+1 and continue. u2)n 3 2 2 6 2 7 6 . explicit or implicit.
Find the di erence equation which results from applying the Gazdag predictorcorrector method (Eq. (d) Perform a root trace relative to the unit circle for both di usion and convection. The 2ndorder backward method is given by i 1h un+1 = 3 4un . Consider the following timemarching method: un+1=3 = un + hu0n=3 ~ un+1=2 = un + hu0n+1=3 =2 ~ un+1 = un + hu0n+1=2 .relation.1 + 2hu0n+1 (a) Write the O E for the representative equation. (b) Derive the . Solve for the roots and identify them as principal or spurious. The trapezoidal method un+1 = un + 2 h(u0n+1 + u0n) is used to solve the representative ODE.1 + 23h (u0n+1 + u0n + u0n. (c) Find er . 5. (a) What is the resulting O E? (b) What is its exact solution? (c) How does the exact steadystate solution of the O E compare with the exact steadystate solution of the ODE if = 0? 3. 6. (c) Find er and the rst two nonvanishing terms in a Taylor series expansion of the spurious root.1) (a) Write the O E for the representative equation. Find the . Identify the polynomials P (E ) and Q(E ).65) to the representative equation. TIMEMARCHING METHODS FOR ODE'S 1 2. (b) Derive the . Identify the polynomials P (E ) and Q(E ). Consider the timemarching scheme given by un+1 = un. relation. un. 4.118 CHAPTER 6.relation. 6.
plot the error given by v uX u M (uj . repeat problem 9 using 4thorder (noncompact) di erences in space. Using the computer program written for problem 8. Show solutions at t = 1 and t = 10 compared to the exact solution. use a Courant number of unity. Using the computer program written for problem 7. Write a computer program to solve the onedimensional linear convection equation with in owout ow boundary conditions and a = 1 on the domain 0 x 1. Find the global order of accuracy from the plot. Use the explicit Euler. PROBLEMS 119 Find the di erence equation which results from applying this method to the representative equation. ah= x. Find er and er . What order is the homogeneous solution? What order is the particular solution? Find the particular solution if the forcing term is xed. 2ndorder AdamsBashforth (AB2). uexact)2 u j t M j =1 where M is the number of grid nodes and uexact is the exact solution. Use only 4thorder RungeKutta time marching at a Courant number of unity. and 4thorder RungeKutta methods. use a Courant number. compute the solution at t = 1 using 2ndorder centered di erences in space coupled with the 4thorder RungeKutta method for grids of 100. Run until a periodic steady state is reached which is independent of the initial condition and plot your solution . For the explicit Euler and AB2 methods. implicit Euler. Find the . Use 2ndorder centered di erences in space and a grid of 50 points. On a loglog scale. Write a computer program to solve the onedimensional linear convection equation with periodic boundary conditions and a = 1 on the domain 0 x 1. of 0.1 for the other methods. and 400 nodes. 10.11.relation. Let u(0 t) = sin !t with ! = 10 . Plot the solutions obtained at t = 1 compared to the exact solution (which is identical to the initial condition). 8. 200. Repeat problem 7 using 4thorder (noncompact) di erences in space.0:5)= ]2 with = 0:08. Find the solution to the di erence equation. 11. 7.6. including the homogeneous and particular solutions. trapezoidal.0:5 (x. use u(x 0) = e. For the initial condition. 9.
6. i. Use a thirdorder forwardbiased operator at the in ow boundary (as in Eq. era . without actually deriving the methods. 1. j . derive and use a 3rdorder backward operator (using nodes j . that obtained using the spatial discretization alone. and 400 nodes and plot the error vs. 3. 1. the number of grid nodes. for the combination of secondorder centered differences and 1st. j . Using the approach described in Section 6. use a 3rdorder backwardbiased operator (using nodes j . 2. . 2nd. 6.2. Find the global order of accuracy. Explain your results. j .69) together with 4thorder RungeKutta time marching. 2.. Use grids with 100. j . and j +1 see problem 1 in Chapter 3).120 CHAPTER 6. Use 2ndorder centered di erences in space with a 1storder backward di erence at the out ow boundary (as in Eq. 12. 200. Also plot the phase speed error obtained using exact integration in time. 3.67). At the last grid node.e. and j ) and at the second last node.69. 13. as described in problem 9. 3rd. nd the phase speed error. Repeat problem 11 using 4thorder (noncompact) centered di erences. and the amplitude error. erp. TIMEMARCHING METHODS FOR ODE'S compared with the exact solution. and 4thorder RungeKutta timemarching at a Courant number of unity. Note that the required roots for the various RungeKutta methods can be deduced from Eq. 3.
Í @Ì° 3448bDG£P9@h8ImP04@37PwXF44@aV0@PTVFjHY@E@o4@c1PIeFXG4EDC4A TAI 0 f 97 G 9W 9gg7 57 T 9727 A27 T2 F2 0 `YHGPFP9c@d7vGh3Ae0Sº76FjVIeT@xc8bs4S· ¯ sy ef1tI80DGe035VwPT4¹AXYP9ctpq4} 1X° ¯ syi@db4VpËVf1tIR FW T G w T7 0 G A27 ¥ 5 R F 9 R 0b { ¥ T7 w T 5 806GV035VwPT4¹AHh3fe0m80SP743Ahef6580P2u q@xP2q3280VGXGVF5 u txSz7XY39e0S87SE3gV0x3A80z7e0e5mV0@U4@PI8Fz7PÊeF6GV035ewu XT4A F F T G F 0 G F 27 F 0W 2 w G 57 T G G 57 T2 G F 60@Eeg4@wXf8Fwc8bdbXw4b X2448bDG HY41mSE8bheIXT@ee0@m~36TP04Sº7eFS@624432l3gcA 42 w7 T A27 0 G 0 G F F TAI 0 f 92I G7 G 97 GI É 9W 57q G 0A 0 0 0 ( wpA G F YW ¥ RI 5 R 0 G F cP23Fc4jVI85P2v} ¯ lyc@ÀVf1tIy48bP2d32e0eGHGeF5 u tÇzY7 u @w@z7È43ge0VGyV01Ç1~I u cP2Xq1@P2F 0WT 97 2 wT T 5I f5I G ( F 9Iq TAI 0 Æ 44eb6Go~ ® Â ½ÅÄ§ ¬ ¨ ² § ¶ 4@l EÃaÀ¨ q8I}4@37P04eb6Gh~65P04ymXG6FPwXF4jDid48bE@X04h@cDi0 A27 ªÂ Á ¿ ¾ G TA 9 AI 0 f 9p 9g 0 0 G 9g f7 5I¸ 0227 f 5 5 Gb F T F R 7 F 5 R 0 0( 0 G G AI 0 f ( F 57 1Eu63544Sq©VA@7 u 1tIeR1X(@87e5VGHYe9P5S½7x#P2¼ef1tI»aiPweF65432@i48bpeI#j4eb6Ghp4X24dbew@f DhPfeF4P2HY44@ 3A4X2@d7c@~XTxef1tIU6580VGSº7d48b4£j4eb6Gh#3ge0VG¹TD41~1E}1HY39@Xq8FDG3wleg6Te05 0 G 7 ( F 9gg7 YW 0 F GWI F 5 R G 9 0 GAI 0 f 02I 7 5I¸ 0 0 ² § ¶ E§² y¨ tz³E§² ® @· ¯ « µ ´ A2 4@7 « ª ¨ ±®X° ¯ V® § ¬ § y©y¦§ ¦ YW G 5 5 57 T2 G7p 0 0b { TAI 0 #6A8032V03TV043gV0e0@~41XI8FS46¥uV03T4}y64jc8b6Ghf ( F 57 0 GRI 2 G 9gg7 YW 0 G G 5 5 0 G f R G 0 0( 0 4P2cdbew@`f Xf8F£Si1PIeFSz7XwPF44SqrT}@XqeFSd732V03TV043gV0m4eb¤1eIt5h3Ae0S87654321T}vc8bG 32c8bh4@B£db@eIc4@xe06Ge065ewXT4PAh3fV0x48b§uT}BxPwVF@Ex4eb¡1eIt53Ae0S8765432@eT}@I 0 G A27 w7 5gg7 F F T 0 G YW ¢ T7W 0 G f R G 0 0( R T 5 R 0 G G 5 5 0 G 0g 0 0A A27 Gg7b T7 G7 f w T Vf1tIw@Xq8Fd732V03Te0c3ge0a4eb 6At@PI39@6q4io4@h~eT35e0c@E @l@D@3T60PwVF85Sdb4Vp8IG 65t06RV0P9X9F u §ij41PIeF6G3wp60e0 4@s§eb@l@w324c32l3g44X2e0@a6T30PwVF85Sd32P06Fh63Iw 5 0 2 0 T A27 GIW RI G 0A 0 0A F 57 G7 f G w 0 0 G F F u A 2 7 u G 0 A 0 0 0 G F w 2 w T 7 W 5 p 5 F c8bdbPw4b u P2BTgpc@BT}324462t6g4`A Xf8Fi8bHGF u PThe26530416I}PwVF@Ey4vrse1I u PTc8bG X23f4ebq1lP9390 @4w4v0 c4j3eTc3g60413Iu4P2eFVG3TV035e0X2q4S@87VG1lhPf~e0@V03T4}zP21lg F 0 G 2I A GI2 IA GpW G w2 w ( G F A27 G2 5Ig F 57 0b { G FI 3AjcdS87uc8beIo6A806GS8785VGSzY7 u @mPT 1PIeFcPp1VImeTXGySxef14w@eIEX2@q@4vS48bP24ebXGF u 0iC 0 f T 0 G G w G7 T 97 F 2 G 9 T F RI 5I2 0 f T F7 fIA G7 G F f R F F7 fIA F w 7 F 9W T 0W G F T F T T 7 T G 0 G 5 w G 1eIt5srXR P2Sq1c#P2@87VG6530qP2uX04@87VGelp8IoPA@e7mXTn3fe0VGVYVf35e0ue06T48b#`2m3eT@8I6GS8785VGS7 A27 G FIg 0iC RI T G F T FA RI F FW T F f72YA Fp F G 4S8TP21lk3Ajc@ihef35e0hP2g3Ae0VT43pVwPT46280fG@eXTs`YHGPFX94@d7vGB33TXwSEcPA44X2VG6Te035e0P2F RI 5 0 92I2 0 G RI T w 0 G 0 T2 w ( 0W 2I 0 fI 0bg 5 9 57 @T @374P2XF414Q4ebu@Qe0@37§c8by`2¤3AV0354XAeF416I4P26Fty1432h146244x@P743pXw8FvG@Eg 0 G 2I TA 0 0A A27 9 0 F2p 5I2 9g T 0 02 F FW T RI 2 F2 0A 9 0 0( c8b§@u4462t6g4sc@r@87VT351Hq44a14iX04hPfeFd35c8bXG6F4aPT`YXGPFX94@87VGUSQ@PI8FHG4EDC4B@87654321)' % $ # ! " $ & ©§¥ ¤¢ ¡ ¨ ¦ £ .
8YR8&©71¨QQ1W10T'7R8VUTRQ©P'7I'H©GED"¨CBA6§9871#'¨6'54¥3%©"¦210)#('&%©"¦!©§¥¤£ B # X ¨ ¨ ¨ # # S 73 F # #F & @ $ # $ # ¨ $ # ¨ ¨¦ 36T4@P74§¢DiX04@3Iw 02 9g 9g f A27 ý 9g f w 0 G GIW F TAI 0 f Tp 57q RI FW T 0 G 0 F f7 0 G G F 0q2 4Ss iP04h16Iq4eb§8bSt§P2d4jc8b6Ghd41PIVF@SU@©`YHGPFX94@d7vGq4ebhcP2h@cDieIe32X0432@@3Iw GXF cEvP9X9F u 0 u 4Sw@XY691XqeF6G6wtVg3Te0· ¯ 4@§X° ¯ 6y 1tIR 4¡c@¼E@o3T4Xp@S321X(3F0 A2C A27 0 0 5 A27 T¥ 5 ÿ A27 ÿý T7 0 97q 0 V03T4ebh3A80E1P(VF3T4x0 u 3VT3580j@Edbe4@PI6qV04`2@`YXGPFX94@87VGp4eb4X2Vf35e06G4©eIe32P06FhceppV0@7 0 G G72 0A Gg7 w Tp F 5g FW T 0 G 0 F 0A G G w T 5 ¯ 4S©X° ¯ 3yþP2h3T60PwVF85S#4eb@q6T4Pp@S32@P(3F48b36Te0@374P2eFVG3TV035e0X2kSqþ`2 A27 T¥ F G7 f 0 G RI 0 97q 0 0 0 G T w ( G F Y27 f 3PTjvY@EP2@@qXfF Fi7 57 F(7 0 G (2 97 9 G A G 0 97q 0 ý 0 G I G7 F fI 2 0q2 w FA 0g 5I c8bwc1PI@mP0XFueIp432e0r6T4Pp@S32@P(3F¹0448bT u Eh3Ae0SEP2h14¹Aj1XI8F6G6w116ImXw4jXIeF65tr1E¸ ü 3XTjÇ@z7V05 Fi7 9 0 G7 02 0 G (2 97 9 G A G 0 97q 0 ý 0 G I G7 F fIA 2 Tp FA 5I @XqeFS3(chc8bq4@PI@iX0PF8I462806T4Pp@S32@P(3F¹044ebT u 4#3A80EP2h14©1XIeF4tXøcv1E¸ ü û( 9 R 0 G YW 57 f f T 0W G b(pI 0 G2 5Ig F 57 0b 4cP2F u PIX91tI48bp3A60PÊVF@qh4Vpdl8I141c32U@87VG1lhPfae0@ DY4}{ 33TX08FvG35l1eI4u`YXGXFP94@87VGÛSrXTeFXYSE@d4P2F u XIP91tI48bP2r3ThPfeFi@qd8I3Ae0V535t0DRe0~e0@41XI8FS876G3wlg 0g 5g FW TRI T 9727 ( 9 R 0 G F 0 G Y27 f G 5 57 T2 G 0 0 0b { F¸ T7 0 fI 0bg 0 G ( F F G2 w T 9W 5g 9 TYbg9 w 5g R iqe06T4}i1H°44X(h30V0Äc432h1432c4V03T4ebp4P2cP2@87@3Ih3fP0c1eI4B@37PwVFj4@z7Xw8F6GS874@I T f 0 T 0 97q 0 0 0 G 5 R 0g 0 0W 9p w G7 2 0q2 w FA 0g A27 2 T h4epe56G6wtVgdcPp@S321X(3Fd48b1tI6A80DG3wlc iwlXA416IS4b u 1PIeF6G6w113IXw4PIeF65to4@o1PIVF4tpR ºXR4d1tIwT}U694jhd41è1eIt5§441tIq0 u cÍ4h1XI8F6G6wjX2SP04h@cDid1E#63TP0eFVG65t1VI4g FA 5 R 0AI f 5pI f R A2p R 2 0 F 9g f7 0 5I¸ 0g 5 9@d732h@Ec4tp}P53Fc8b80SPfeFVG6TEzVGSz7P0~S vY13Ic8bxP2s@37u0 u 1vY@E1XI8FS87VGae0@7 µ 4@7 ª ¹R G 0 f7A2 R 0 G G7 0 T 9 G7 5 0 G F2 w 572 G T 5 A2 G7 F fIA G 5 0gg7 G 9 0 GRI 0g 5g 0 G 07 @e0SEX21ceIU@z7l4@6580eGº7a48bÈ@}6TP08FvG35l1eIcdc8b@@f 2 w T2 G 9gg7 9 w 5g T w F F 02I 0 0A 7 G T w 5 F 57 0W 2 @3741XI8FS37PwPFc4@r@z7Xw8FDG@874e0@37hdbPwcb u X2@c1q@Xq8FDG3wt0DR4o8Ie01XI39oXYP9VF@87e5XG4VW@plsSz7w 6f80vGjvYU80DGP04h16IvrSE8bÈ354cP2h3fe0ebXGF u cÍj41PIeF6G6wj~30V0j66f80VGvY4321X(3F}@Xq8FDG3wt0DR44@7 T T 9g f w 7 G7 G 0A F 5 7 2 0 T T T T 0 0 0 0A A2 806GX04h13Iheb@lx1zqEeE8ba3T60PweFe5S#8Is621Hq1plrX9u P9F u 1PIeFS87654PAVF41©SP04Sd7VGwPTú3f80vGVYT 9g f w GIW 0 7b G7 G G7 f G 0 F( 0W 2 G 0 T2I 9W T F T 7 F G YA T 7 0 w7 5gg7 2 G 9 T 7 F G7 0 G RI G7 0 27 0 F( mXRe0Sd7vG`Tc@3780VGh3T4dbSeI4c@1PIeFcPp@eI©ÃdbXw4b Serù@V4ebm@xe0SPf8FvG3T@h3T@Xq1eXGF G7 G 0( G27qA7 0AA7 G2 5Ig F 0 G T7b T T T 0 0 0 G RI T 972 ' 2 0 E8b1@87@Sc@o3A44S@d7vG1thPfa4eb~SEmh3fe0VGVYc321X(3Fde03Tc8bÈ@PTeFHY@Eyc ¯ 1XI8F6G6wj X2§6Ae0VT43pVwPT4vXTvPTc}dc4P26F432V035tXø430SEeg©4SehPf8Fec8b#8bSt©1462t6g4@z71PIeFXG4EDC4A F T FA F Fb { ( w 0 FA w7 T A27 0 G 0 G GIW 2I A 0 0A 2 w 2 F2 0 `YHGPFP9c@d7vGpÇDd®6T41eXwPF1leW65t XYX9@º76F3wleg3T 0 41XI8FSE6¥Ä@P78F32V035tXø4È@P78FVGSEx8I41XI8FS37PwPFc4@7 FW T 7 02I 9I 0gYb 9 0 T2 G7p 0 9 G 0 FA9 57g G T2 G 9gg 5pI F w T 9g T T Fp¥ GI2 F 2 G7 T 0 G T w F G F 0 I ÷ G7 f c1hP2yS304X2eFhP04hPfVF©eI80XGce@4uXTs1XI8FS48pXGVFp4ebpe0@37uPTc8bsX26351Dq0 u ~#tHiVF85SueTXGF R@¡6f80VGvY4321X(3Fh48b8I©6A80SP739e0hl VIPT@§32e0fG@Q@z7x· ¯ sy}S`YHGPFP9c@d7vG4}h ª @I 6f80G I T T 0 0 0 G G G 5 0W 97 RI 2 w ¥ RI FW T 0b { R rjvY432@P(3Fw48bo1 XY39V0P5eF624462t6g4eH° ¯ syr@x`YHGPFP9c@d7vGe4}· ¯ c@eH° ¯ 6yr@m8T´ vG@Eg T T 0 0 0 G 2I G 0 TA 0 0A ¥ RI FW T 0b { A27 T¥ RI 57 Tc13I46211h14x4eb@e1XIjzqE6btxc8bh1§XY69e0P5eF323AV0@EpV0@§`YXGPFX94@87VG~@e4@PI8FHG4EDC4e4v p 0 0(I fIb 0 G RI 5 F 7 0W 0 G 2I G 0 T7W 57 FW T RI T2 F2 0A 5p öçïô õô êç ò ç ð ê í çì êç ë êç éç È Ä©3óÀñ #ï hî©#©sèæ ±åpã ä × âÓÒ×Ü× Ô ÐÓ ÙÚ ßÝ Ü Ù ÙØ ÐÒ× Õ ÔÓÒ Ñ Ð Ï #rÄÇBÇávyràÛ}Þ3ÒÛrÚ4ÇÃ±ÖxBÄvmÎ SÌ° .
ý ° ý ° ¨ ´ ´ ý E E E 0W Gb F f 57g Tp 0 0(I fIb 0 l}1X(vG@E416I4321@1cdc8bG R@)ef1tIl@z7Xw4`Yue44syS Sz780vG4P2j@X04S4 i1EBsdPI4x@4eA1@8Ip@PI8FE3¥X014X2eFT I 5 R9 Fg G 7 d ¥ RI A T F 9g f7 0 5I¸ w 9W 27 5I 7 G 2 G7p 0 9( 7 1eIt56T14@Edb©@PI8FE3¥h1XqeFSd762e03TV043gV0i48bB@¼ef@tIe48bh3f80vGVY1Hq8F6G6wt06R4xeI#3Ae0Sº769e05 f R 0(27 w 2 G7p 0 0 G G 5 5 0 G RI 5 R 0 G T T T 0 0A G G 60@Eegc321X(3FQ@s`2@PI8FDG3wV0wE8b©P2xH°4 8IpH°4Ç3ys@#XYX9@º76F3wleg6T§4S Íj4a41XI8F6G6wj w7 T 0 0 27 2 T G7 G F G T¥ 2I 9 0 0 A27 T2 0 X2U8THG4ep6Te0m48bq@I u e74v0 u PTc8bB1EB@30SEeg4621P(6FpP2r41XI8FcXp1eIrP56F48bÛSe0c8pSE48b6G3wlegcP2F F 9 5 0 G 2 5A F G 5I¸ w7 T 0 0 F T2 G 9 T 0 GRI 5 G72 0 G 0 T 0 u 66f80vGjvY432@P(3Fh1Hq8FDG3wt06Rc@±q4wE8bTU@o`YXGPFX94@87VGh48bQ4@d7vGeT654448I654eA1Q`2 T T T 0 0 0 0A 0 7b G7 G RI FW T 0 G A2 0A2p G 0 5I 2 1XIeF35e0XGVF35w `YHGPFP9c@d7vGp32V03544X2oc8bgX2P01VIs4zYº7cqeT@8I6G6w132@P(3Fo48bh6f80vGjvY432@P(3Fe06GP0c16IÃebXGF u FW T G 0b F 0 G F 9 5 I2 9g 5 0q 0 0 0 G T T T 0 0 9g f w T}v1tI~SEebwe0@4u0 u HYP9@4P236GhmXTo1XI8FHG4416Iu1±ql@ec8bVI£ u eI@mHG6Tj4~35c8bXG6F42 5 R G7 G GI2 97 F ¸ 0 f F 2 FA2 w 0 IW7 0 G T 5( F 0IA 0 0 GpW g T7 0A GI2 0IA G7 G 0 G F F T w F G FW TRI F7 fIA 0 G j4j h r@xzY373w4@43T4x§ S48bmcep85BPTXGe0@37XT48bq`2Ä1`YHGPFP9c@d7vGÛ@hP2@14m48bhP2F WRfaPTF l¨ ©dbPwcb u 1tIae0SzÇ@º763legd48b~SEeb8®680VGvYo680SEX2@4`Aj@PI8FD3@@3 P2F ù e ~ ý F 5 R T7w 9 Fw0 T 0 G G7 G T f T T A G7 F fI 2 Gw0q2Iw 2 G 0 T2 w9 w 5g G2 5Ig F 27 F F T T7bg 0 0W 9pI T 2 0q2 F @PI8F8735cPAeFc13IÇ@37Pw8FDG@87cSd7VG@thPf @mPT~XG~@ 7 3A60PÊeFSE4h3fdtoPA41cebm¹G}1PIeF6G6w113IXw4W A27 2 Tp FA GIW ( G 5 5 0AI f 0 G 5 R 0 G T F 2 w F G G7 4Ss1XIeF4tXøch8b@l#4X28F32V03TV043gV0uTr694jhqc8ba1tIs3A4VCPTeFS87dPTQ1XIeF35e0XGVF35dPT4ebwS48bG 60e0p0 u 1H°4Û4X(©4X28FDG3wleg4`2@c@º74sDiX04h13I48bQP2dPTj©vY@EX21@Pfqc8bÄrSufGDRP0qc8bG T F ¸ ( 0 T 02 9g ý 9g f w 0 G F Fi7 57 F(7 F 0 G RI 9 0 eIh11xP0XFpvG4eh3T4Xp@S621P(6F0 #4ebv@}X9@@`YXGXFP94@87VGq62e06544P2h1tI3SEeb3Te0Sd7vGXT4}{ G 5I 2I 9 Tp f 0 97q 0 ý 0 G RI 97 FW T G 0b F 5 R G7 G G T Fb  97 5 z x ® ÿ }X9@1tIR {y}4 ý w ® ¯ HRpHY41 4@7 F 92I A2 HRhX04@87VG©HY8932V03544X2V0@dT}q48baSEebÛ@P0c@c id@tIÇ ¯ ·jÄsy¡1VIt5h6041dSaT u PIX91tIR F 9W T G 0b F 57 0 G G7 G 9g f7 0 5 R ¥ f R w2I G7 9 XGde0@37hdb4epo`21XI8FSef1tIvR4@e785`YXGVF@º7X9XfeF q3A30XÊPF@411@P74wl @z7qHiVF85Swc8b4@7 F T w 7 w T 2 G7 5 T2 G 5 F T 7 YW 972I( FA 0W 2 w G7 f 0 G A2 3324462t6g44X2vXYV9@z7cP2PFBV0@VT18IDG3w1621P(6FÈeTXGc@1P9S1 3f80vGVY4621P(6Fre06GX04@3I@4HiVF85SE¹R G 0A 0 0A F 5 0 9 57 5 0q 0 0 FRI 97 T T 0 0 9g f w 7 T7b G7 f 7 3T}aw4ebÄS @PI8FjPp1VI62P0VF4@87e548b1pHY41~4462t6g4q`YHGPFX94@d7vG~62e06544P2E8b80SmÉ 0 G RI 2 G 9 T G T2 G 0 G 2I 92I TA 0 0A FW T G 0b F G7 G GI T # 0 2 I W 4X2@ 7 TF G@ 2 ® 1Í ¯ PTF § ¬ 32cb u rHRPù9e!¹D© )eEDdùDEWfXThoH° %¯g UlSy}3A c4ªA 4i1eFp lUqq@6f 7 le0 o5 § @3 Dhd74@1s3II25T7vG ¸w 0 F f ùd F ¥ 85G 7 f 641XI8FG T2 7p 0 9 G 0 FA 57 F 5I 5 T 2 w FW T 57A2 T 0 G G T 0 ÌE3¥v@P78F32V035tXø4gVYSEP24VA1p1tIR 3AV04p 1XIeF35e0XGVF35g`YXGPFX94@87VG»eA@44@d7vG48b#e0Sd7vGQ0 u e065m÷ 'QW xh 8 d IÌ° s is ts m k j i rWvu2hrq'ponW§'l s is ts m rGvhr¥pny QAQ¡ Vôyñænwxví 6ïzò9t©sppxvBh#©Çf Aepã u õ ò rqï i ï êçgç ð ê ä × c6y}À3ÒÛrÚvcÇgir~Bmràgc1`±ÖÕ b Ó a Ý ß Ý Ü Ù ÙØ ÐÒ× Ò àÓ ÔÓ Ï Ù Ö .
2 0 FqRI G FIg F G T FA 0 G 5 G2 T2 w 0 f T G7 4 ¯ 1PIeF6G6wjoP2F u P0È@P21l}PT4eb~VT43pVwPT4a§4XI8FSe7m@87VG4@3Ia@eISeI6530Ê eIw1(PT@SVq6580P2qeb3Thv4eb@P2@14PT4ebX248bHGF u 1XI8FcXp1eIvc8b@)ef@4h1eIu8I462tc@Eb G I 97 G F 0 f 0 G T7 F7 fIA F G F 2 G 9 T 0 GRI 5I2 0 f T G T 0gg7 G7 I T G F 02 0A F FW T A27 0iC F F7 fIA w7 0 G ' T2 G7 F 5 Eb u RS}Vf3580xX2o6A4E6Cc}PT`YXGPFX94@87VG4@o3AjcUXTxP2@14v60@EVg`TDhPfeFÀ3c1PIeFSqXfj±iVI4g g7 w T A27 0 0 G GIW RI 0 0 0 G Tp 02 9p T 0 T2 w G G 5g 5gg j@#60@7EVgÃ4@hPfeFG c8b8b@lv@8TDG3wt6ø 48b§XYe9c13I4Sd7XGcePfVF654PAVF413IQ8Ixe0Sº7VF41eIc4@7 XTXG 3TV0@7zÀdbcepT `2 S80c1VI806GV035VwPT4XA6fe0Q48bG u PIX91tIR XYX9eF@e7eT6T303wcoS4gcoE8bp4@PI8FjPp9 F F T w w Gp 5 F F T 0 9 5 02 GI2 IA G7 G T2 G T ¢ R G 0 fg 0 0A 0 G f R T T 5 G ¸ F Tp RI 2 F2 0A eIÞB~@Ii32h41XI931Dq4p4eb¡1eIt5uh3fe0VGhe04epS876580HGPF9 UhP23AV04h32e0fG@h1XI8FHG4E6Cc¤' 0p F2p GI2 57 T2 F 0A F T Gg7 w F GRI ( F2 F 0W 0 G G7 G T T S4P¥4c@c0eSc1PIeFXG42EDC4i`YHGPFP9cW@d7vG436580c@4dbXT48bSw4X244X213(l4ebx3Ae0Sd7vGv0 u ' FW T 0 G 5I G G f T7 T G @`YXGXFP94@87VGhPfFeU1PweF@8IcghVY@S7qeI 6Ae0e565t06RV0}3ThPf8FDG@eI}PT`YXGXFP94Sd7VGÄ@x1XI8FHG4E6Cc~XT4}{ 5 0 0 f T F FW TRI 2 F2 0A Fb 6Tghm4eb£@i21XI8FcXp1eIr32X0eFT4S878504eb@IiXY4@4c32l3g4u`YXGPFX94@87VG4@37PwVF35h4E8b30V0~§ 0 GRI G 9 T G 2 G G 2 92I TA 0 0A FW T9 0 fp2 G7 G T 0 h g T # 4A p W T P2F @ 7 0 7 5 2 ® ¯ ¯ PTF § ¶ 324b u rXRPù9e!¹Dx ÁeIe" Dùq2PTF·Å ¯ @sy6A¥ 44µ2 @teFI i UScq@ 3f@V0 l4² §½ @3 ¹DG ©d7vGw4@1s3II25T7 ¸w 0 F f e À e5G 7 f PTiÍ ¯ 626f80S878Ip1XI4@Eh13Ighu4}{ F G 0 G G G 2 F27g f w 0b Vôy ñ ¿nví 6ïzò4t©cp½s·zìBlç ©«» ºpã ¾ w u õ ò rqï i tq òg ö ¼ ä 362e06Te04e0S@PI8FEPgVFeTPTc@B8T4@@@37EXC4@P(eF£XR}604@d7vG1lXfE@z7Xw8F6GS874d4B1UX089eGHGPF£@BPTF G 5g 57 2 G7 FARI G2pI f7 G2 w F2 T F w2 5Ig F9 w 5g I2 5I 9RI ¯ 1XI8FXGc413Ij36516q0 u m3h3fP0c1eI4h1XI8F6G6w116IPw4a8Ii6A80SP739e0B6T4Pp@S32@P(3F4ebe6AP04c3p3wI 2 FA2 w 0 I ÷ T 9W 5g 2 0q2 FW G G 5 0 97q 0 0 0 G YW Fg w 0l8IÄ4432e0idbPw4b u PTjavY@EP2@@qXf}48bdRe28¹ ¯ @PI8FHG4416Ia`YXGXFP94@87VGUc8bS48be0@mÉ W G TA G F Fi7 57 F(7 F 0 G ù ù 2 FA2 w FW T 0 G G7 G GI G7 F G 0W 9p w F 5 0W w 5 6A80SEX2Vf35e0alPA4@3IXGV01tI6Rl~eT3T60jVI4g 2 G 9 T 0 G 0A Gb F f 5( 7 w T Gpg f w 7 2I 5I 0 5p¸ 0A 0 @PI8FjPp1VIx48bhY±VI85VG6T4i1X(8bG u VI1©Ãdb4Vpi35e0c4h13I©k1B@V01Vf354ebVG4E@±Y373w4xc8bG G IW7 T( 5W 27 0 I 0 ¶ 9 G 2Ig 0 0 G 5 0W9 F fI2 9Ig 0 GRI 5( 0 5 jp1tSU14X2eF4xAc@h¸35@±q}3T@Sd7vG·Sº78F624u0 1lc iv4ebe01tIDRts@P7@4XY1lv4ebÈSx8bG u eI1m1e(Sº79 vY35@qt§zYqfe06548bp604P2VFieTT6P04ebVG1I tszYqh@PIeF6580HGeF35xc8bwc1PIeFSz7XwPF44SU@37Pw8FDG@87cP2F 0 7 0W 7 0 G w T 9 5 0W 7 f 2 w 0 G T2 G 9gg7 9 w 5g 6351Dq0 u ~B3µeI6530¹Êj214sP9SU1tIR h y@dz g £ ±³ ²°4¡¥®m604P2VFu¨Í ¯ 32h3fe0Sd7Ä@ve0c32e048bhP2F 0 I ÷ ´ I 2 9 7 5 g T 7 ± ¯ w T G 0 G G ¨ R I T T 0 G `YHGPFP9c@7dvG1tI}32X03FhcVpTPT1PIeFXG4Ac13IPT4ebiXYP9Sz7PweF6GV013I4}yjz lz U1uz 1z ÛHRycHYP9@37PwF FW T 5 R G w F 2 F 2 w F G 9 5 0b { ¬ ® ´ 5I ¬ ® F ® 9 vS87c@E¥1aHYe9@374P2XF 9 4241lj14b8HGF gu VI1iPA4@I u i4@7 ÛdVY@4P21@Pf}V044@tIU304X2eFT G 5A7p 5I 5 0 A pIW GpI u 5( 9p A2 ý 57 F(7 F 5pg 5 R w «P9S1tIR ªUÿ4 ý w 97 5 z © ® ® 1 ¯ ef1tI4ebi8IG 5R0 G 6AEXCcjhtWrVGcew ¯ 1XI8FGXc413Ir3TV0@z7qbd4epy1tIBGS48bm600e~0 u 1PIeFcPp@eIBPT4ebw4X28FDG3wleg4`2 0 FAI f 0 Tp f 2 FA2 w T w w T5 R 7 G T 2 G 9 T F G ( 0 T ® @d ¯ £4¡ 8¨ `z ° Á `z Á z § ¨ ®V ¢ ® ´ ® ® A£4¦¡ ¥r1z ´ Á 1z ¤ ¨ ®V ¢ ® ® £48¡z ¨ V® ¢® ´ ´ 2 1PIeFGcPp91VITa048bG4A4EC04@IodbXw4b u 1tIR T 2 2 F 5 @Ì° × âÓÒ×Ü× Ô ÐÓ ÙÚ ßÝ Ü Ù ÙØ ÐÒ× Õ ÔÓÒ Ñ Ð Ï #rÄÇBÇávyràÛ}Þ3ÒÛrÚ4ÇÃ±ÖxBÄvmÎ 'QW xh 86 .
FW T9 0 fp2 5 R 2 FA2 w 0 G 0W G 0 G 1`YXGXFP94Sd7VG£@z7XweF35h4}1tIp1XI8FHG4416Idc8bteIo32S@d7UXTF XT48bx4SwÞx Æ ® Â ý Æ S48bv30e0@87@41odbXw4b u 3Al1PI6916qce0@44eb6Ghpdbew@`f Xf}cÍ F G A27 ° G7 G G2 57p( F 0g 0 0A 57 TAI 0 f 57 0 F { z x ® w G7 G ( G2 57p( YW 0 9W 0 F 0 GRI FW T G 0b c{yl ý GSEebcP23F6080@e7@E1x 3A4VbPTXF4@d7vG3T}XT~T}uc8bÈ@q`YHGPFX94@d7vG62e035c4`2j 0 f F2C Vb3Th80HG4Ee7 2I 0 G7 0A w7 T 0 G ( G7 F 5gg7 YW G 0q2 w 57 ¢ 0b { 16T1XqeFSSXqVF354w30@EVgwc8be4X28FSPfjzieI4c@ mT}aq8Io3A80vG35@@3Ide0@mT}Bw4}1° û w7 5gg7 ( 9 R 0 G F 0A 0 f T F 0 T2 w 2 T F db@VI44@e4P2F u PIX91tI48btP9@876G4h1eIX23AV0354XAeF4@3Ia0 u 1XIeFeTc3pewXT4A Tp F 5g 0 G G 0 FA 4@PI6qV0448b»`2¤@±Y7 u 32e065£Hø47 P2F `YHGPFX94@d7vGuSc6g304@3I48b©4X2S4 i0 u 4 4 6G0 Õ FW T RI G w2 w 0 G 0 F f7 I2 Tp Vôñ ¿oá©ylhÔh)Ò ©eÑ6ïzò4t©cpÐeGç zöò Î 'Ípã y ¾ w u í çì êç ógç Ó ê í ë ê q õ ò r qï i çì q é iÏ ä 5 ( G 5 A 7 p ¥ 5 I 5 0 9 0 ( ¬¨ ® ´ 5 I ® 0 0 A 2 7 ebG u eI1aPweFS87c@E}1U@374P2XF6G1d0 u z hz }1Uz 354ebXG6Fo4Sw° ¨ Æ Æ XRed ¯ F syxX2mX04S4 i1tI1° ¨ Æ Æ @tIVT3T4c3A44c1t3080@e7@E1@4e2 c}6f80VGvYv1Hq8F6G6wt06R4304X2eFT ¥ F 9g f7 0 5 R 5 R 02 0A2pIW G2 57p( GI IA T T T 0 0A w ®1 ¯ Ç A27  97 5 4S&X9@1tIR ° © Æ ® U4 Æ ®ÿ XTi ¯ eIq1PIeFSz74XC4 3AV0P546¥e05 F G 2 G w FAI f Fp 0c8bwT}aq1Hq8FDG3wt06Rcd@tIdE8bh30e0h§hjd ¯ sysebXGF u `YXGVF@º7X9XfeFa8THGe0@4Q4@ X°jÄsy G 0 0A 5 R G7 G T 0 ¥ 5 F T F GI2 A27 ¥ 0 F f7i 0 0A5 R 2 T FA 0 G 9 57gT T T 0 G5 R 2 T FA 0b cP2h@cÄ3T}@XqeF6G3wt0DR4Û1tIa1PIVFeT46pewXT4y48blPT69P0P9S87@EÈh3f80vGVYre03Tc8b£1tIa1XIeFeTc3pewXT4B4}{ z32hefVT3Te0VT@q`YXGPFX94@87VG@37PwVF35h4dc8bpP2X01eIh4q±YP74g G 0 T7 FW T9 0 fp2 0 G F 9 5 I2 9 eT@8I6G6w132@P(3Fr4ebS 3f80vGVY4621P(6Fre06GX04@3ImÈ1tI36548bvG4ESP04@87VGHYP9@e785c6p4HY41ÈPTa1PIeFcPp@eIT 5 0q 0 0 0 G T T 0 0 9g f w 75 R 0 5p¸ 9W T 9 G 02 92I F 2 G 9 0 G 0 c8be035cb u P069ewP56FXGc4i48b©1XY69e0PT6F3wV04@eIË@EXg3F4X2eF4ic8bB@#1XI8FS37jXIÊ¸6G6we0V513IVwÉc8bG w F2p 0 G 2I 5g GI 5Ì 97 w 5g 0 G RI 2 G w 9 5 ¶ 0 6T30@P74udbXw4b u 694jha1PIeF6G6w113I¹wDXw4jXIeF65tB4ebÇ@tIÈVT43p6wjBe0Sz7r1XqeFXGVF432V04ebvP2@16T} w 9g F 0AI f 2 0q2 FA 0g 0 G5 R 5 wI T w 0 T T 0 G F7( ' 5 R 2 F2 0A FW T 0 G G T2 w F 5 R FW TRI 2 F2 0A Fb @tIe1PIeFXG4EDC4d`YXGXFP94@87VG4ebebXGF u 32e0VGPTVF416IyPTBTghr1tIa`YHGPFP9c@d7vG£@i@PI8FHG4EDC4BPT4}{ 02 9 9g f S4@P74`gt¢DiX04@3Iw 0 G F w F2p 0 G 0 T F 5I 2I 9 Tp f ®Y27 57 0 G F Tp 5p T A27 97 w 5 c8bP2iP069ewX53FaXGc448be4XAeF4X2a1§@eP0XFwvG4ecSe0@e06548brXRÇ3c1PIVF44egQ4@@EPg6F4P2VF4g eb@t W 3T4Xp@S621P(6F0 È4ebwSvP9@m1`YHGPFX94@d7vG@z7XweF654x1tImzSEebx3T80d7VG»1XI8FHG4416I XT4}{ GI 0 97q 0 0 G RI 97 FW T 9 0 fp2 5 R G7 G G T 2 FA2 w Fb Ç A27  97 5 ° ® ÿ oc@&§X9@1tIR Þx Æ ® }4 Æ ® 1 ¯ XRpHY41 4@ÄXRaX04@d7vGqHYP9@37PwVF35h4}PTjc8b6Ghi4X24dbew@f F 92I A27 F 9W T 9 0 fp2 F AI 0 f ( F 57 DhPfeFxwh3f80vGVYQdb4ep1tIt1XYV9@z7X0c1u ÍcÇsyÛ3eT@eI`5lcP0cPg8FHG4ev@tI#1PI4SE@3I8THG4@7 0 G 7 T T T w T 5 R 5 9 ¥ GI 9 9p f 5 R 2 F27g f w F A2 ¥ RI YA T 7 f R 0w2I G7 9 R ¥ F G T 2 FA2Iw 0 G G ( F 5Iww °·csB}@sj48pvGoè1VIt5q34@uSdT PIX91tIÄ ¯ ¯ sB§P2e6e0Q1XI8FXGc41348b 8I 4P24VA136@7 2 w FW T 0b { T T 0 0 9g f w 7 YW 0 0qI( RI T 7 0 T2I 1XIeF35e0XGVF35a`YXGXFP94Sd7VG4} 3f80vGVY4621P(6F80DGP04h16Ieh3A4V235@Ì@Tt@y6GV0y654PAVF41 ¯ Ì° × c}Ý ©urlB}}¤Gvg3ÒÇrÚvcÇ× dtc×qovâÈáÂ±ÖÕ b Ó ß Ý ÓÎ àÓ Å ÔÓÄ à Ý Î a à Ð Ü Ù ÙØ ÐÒ ÓÎ ÐÑ ÃÓ ÙÒ Ö s is ts m k j i rWvu2hrq'ponW§'l 6 s is ts m rGvhr¥pny QAQ¡ 6 .
®° Ì1H° ¯ Ý Æ Ý Æè Ý ² ¦ÆÆ ²§ ¦Æè ² ¦ÆÆ µ çÆÆ x ¦ÆÆ ²§ ¦ÆçÆÆ ² µ çÆÆ x ¦ÆÆ § ² µ ¦ÆÆ ¨ ¦ÆÆ E§ ¦ÆÆ 0 7b @zqEw0 u 34@PI8Fº739V0d`YXGPFSE3¥4X2P56F48bi4@vii 4432l4m~60e0 T hef14¹2cuHiVF85S4@r18IDG3w@v4P2VF44@7 T2 G 5 97p 0 F 0 G A27 ® ' FA 0gg ' T 5I æ G7 f A27 5 0q ( Tp A2 Ý µ § ² À¨ E§² A2C ¥ G 2 5 5 9g T ( F 9gg cE0 u jzX° ¯ syh8I1PIVFeT46p3wV0vP04hPfVF4P2XYc4' 0g 5g G2 5Ig F 27 F F G 5 9 0227 f F 1`YvG35l1eI4USd7VG@thPf@XT}PT4ebc1HYe9@37P0 36544@}PTc8bG X2peI6530h8Ih16g80VGahPfeFo4@q4P26F@EVgeb6Td4eb~@o1XI8FcXp1eIm6G@c iu4ebheI3T 1V(35@@3I~XGXRF F Ê G I( T 0 G A27 ( w7 T 0 f 0 G T7 2 G 9 T w7 0 0 G G 0 0q2 w F EDCåD@dTPTjc8b6Gh»c1 ¯ 1PIeF6G6wjP2p6Ae0eTc3pewXT4B354ebVG4tpXTBPT4}y ±®e0Sz7}PwXF1leW35l~c8bG ù ùä F AI 0 f ' 2 0 F T FA 0 5 R F Fb { T w 9I 0gYb 0 X246Acjd² Ü ebXGF u eI6530ueIw1q3g80vGXf8Fmc8bq4Se4X23F@EVgpeb6Tm48bB@BHGXfPFm48bhP2z®e0462e0T F 0iC Ê G I( T 0 G 0 G A27 ( w7 T 0 f 0 G T7 F 9 0 G F T 6TP0eF65e0m1PIzYu48bX2 F 1eIV535x46T3044eIcHGBXRÈEDrùDPDcdvPT#48bDGf ½ED"åD@dvPTXGF T 5 9 7 { 0 G 5 0 I2 wpA 5g F F F AI 0 ' ù ùä F 32c8b£ED`ùDPDcd4@}47 Õ @we0462e0a4ebX2 F Pù9e!¹DXTj4eb6Gh@37PweF65ciXR43SEeb~6T80S87VGT 0 G A27 ®i RI T T 0 G f F AI 0 f9 0 fp2 7 F G7 G G dbPwcb u iRÀ8ùdE§v â4¹Û¢1c1eI4eby3Te0XTeF@w`YHGPFP9c@d7vG£@1XI8FHG4E6CcyPT4ebt@}604@37EXC4@P(eF~4}{ F ùdã á à b(p 5 G 57 FW TRI 2 F2 0A F GRI w2 w F2 T 0b 2 w FW T F G5 1PIVF35e0XGeF65q`YHGPFX94@d7vG}PT4ebU1tIR 2 FA2 w 5 02 7 @PI8FHG4416IpvY@87VT3T306w4uPTF ¯ ¯ X2q1XIeF35e0XGVF35c8beeIX2@143Ajc~e0P5eF32v4ebU35@Ì6Ie8Ir354VA1I F 2 w 0 G T F7 fIA 0iC G 0 0 G 0q w G 0 5 X2x`YXGcEcP2eI@mvG4es§²643ge0VGhPf8FÈ@35le4a4ebsceI35608I1}PT@Svq3580X2oVb3Thdc8b}@7 F F2C F G I( Tp f ß T T 0 GRI 0W fp2 0 G Ê G I( 97 G F 0 f 0 G T XYV9@z7X0Ä1`YXGXFP94Sd7VGÄ35@u8fdbPw¥Þ7 Õ 1d7 Õ SeIe3Ae0V535t0DRe0u6280fG@ÈPTÄ¹GÇ1`YHGPFP9c@d7vGE@d1PIeFXG4EDC4A 5 9 FW T 0Y G F i 5I i T7 G 5 RI F FW TRI 2 F2 0 9@z7XweFeTSº7394ebvPTaPTc}hj² §Ü 6Acja1tIeI6530 8I4P2VF4ebQeb6T4eb@ÈcE§² 31eI6G3w@#1PIeFcPp@eIT T w 0 G F Fb { 0iC 7 5 R Ê G T 5 T 0 f 0 G T7 5 0q 2 G 9 3Ac441l u3T6044eI4Äz 31eI6G6w1ÄSº78FHG4P2§3A4c41l§@UXRP0c@d7vGqloeI§PAS87aPTzH° ¯ sy 0A2pIW 7 wpA 5g Ý 5 0q9 F F 0A2pIW Y27 F 9W T 0W G F T F ¥ ® zH° ¯ E§² µ ¨ tz³E§² ´ PTeP74eef@tIUPT4ebÄ@vG@E416I432@1@4wc}yj· ¯ ly@I Vf1tI48bpX2eº74eVf1tIR F 9p 5 R F GRI 57g Tp 0 0(I fIb 0b { ¥ R 5 R 0 G F 9p 5 d®6T4@º7c`gDhPfeFI u @H@0 F 6916qX0`9SI u a8I©B}4ebE@u1XI8FSPfjzieIc4@QhBc1}304c3Ae0B imÉ 02 9 0 G G 0 G 7 G ¢ 0 GRI 2 G7 F 5gg7 5pI wp 5 G 0 Ø µ ¨ i ² Ü 2 G7p 0 0 G YW 0 f 0 G iC A27 w7 T A27 0 G GIW F w7 Fp 0 F G7 1PIeFSE3¥a4ebxv½@V§Vb3Thdc8bxc4@w30SEeg4@vXf8F8b@lP26A30@4egPTc6¥}PTS48bG XAeF15 8bHGF u X2@14}PTc8bm651±qd{yþÚ4@þ½ªv×P24ebXGF u 4P2HYxP2@14d30@4egT ( 7 F7 fIA F G 0 I Û x x z A27 Ù x Ø x z F ( F 9 F7 fIA w7 DhPfeFw80HG4Ea6G4Vp85vG413ImVGVTP5@V0432V0w60@Eeg¹TDhPfeFw4eboP2p`YXGPFX94@87VGe446C4m4DGP09 u mÉ 0 G F2C 7 w T2 w F ¸ T T w7 0 G 0 G F FW T 02 0A Tp I Ê w7 5gg7 97 G F 0 f w7 T A27 0 G 0 G GIW T 4eI6530 db@VI44@XT@SVq35e0P2eb6Td30@4eg 4@hPfeFac8bx8b@l@7 VT3T30eIc#dbew@dXf8Fwc8bX2#3Ae0S87654321ue0@m41XI8FcXp1eIdP04S87P5VF3T4SEebu30e0S87@E@Ä£wd±ù w 5g 57 f 0 G 0 G F G 0 0( 57 T2 G 9 T 9W 0A G7 G G2 57p( TX4}p h g Â Å ¨ a@s3A44c1tdPTdS48b1PIeFcPp1VIy@z7XweF654 p1zqEP9P9F u eb6T30@4egT Fb{ T7 0A2pIW F G7 G 2 G 9 T 9 0 fp2 7 0 7b 0 f w7 jrTh1VI©1I Bx7 1eIt5©680Se73461Ûo680vGjvY VY@E@PI8Fd7VGpSE8bw380@e7@E1md±vXT4}{ ù¹ Ö 0 f T 2 ¢ f R A G 5020( f T T 572 G T 7 G7 G 0 G2 57p( ù Fb × âÓÒ×Ü× Ô ÐÓ ÙÚ ßÝ Ü Ù ÙØ ÐÒ× Õ ÔÓÒ Ñ Ð Ï #rÄÇBÇávyràÛ}Þ3ÒÛrÚ4ÇÃ±ÖxBÄvmÎ IÌ° 7 X ë F &û þ ¤ ù 7 # @ï # # # @ ¨F 0 3 vXY R3 C¨ïRC3C#&ÿGûâR&C#)y·"3Ëù 9çþ'ýÊü°Xû µR3C¨# ïõ R@X C3ï 7 #AX&8&$"¨'B©47 #P ø y Q ø &# øQ & 8B# Y ¨· 87# R3 7 ¨ ( ø C¦ËC#C38B'B£ê#¢ '0'YC#9!¡X B 7X©A7¨¨ ây§'B9C#87AC#vI1îQ©²'Xõ ë §1B"#¦ ©"#¨ ("¦vuÿøRTC¨3©C¦11ï)3uF )#(°§§©9C¦ü#¨#'Dï&$ çþyÊü'ýC3°4þ ú¦þ¥¤ù F¥ýB üAø "#sR÷7 3C¨B ïRC3!AX8&$X "¨'B©4©"3'Bê#¥0"#B 3P R§©C¦4(C#PCï&'B"#E¨ö& $ô qó §C¦T7 'òñ B $1¨(°R"&RP§Rê3u8FF'#©I6Êë63C¨A"&9'7"#'B1î48í6'@#©R©RËC¨H7B"3Yì 3 A$1¨P(9°8ð3ïX& "ìhIR43H7##'B 9)ë¨ °I(1îBï ê3rv"#F0¨ 8X# & C¨©y(PP©C¦H#B@H3BF C3 AHR& h¦3C¨PACï"&6©'@¨ §#¦ ©Ré¤X ì "3©H X3ì B&H B 0 # #7 #ì & @ B # ¨F & X ¦ ¨ B .
2 1 0 FW T 5 R G w T GI2 GpW $ 5 02 F 2 FA2 1`YXGXFP94Sd7VG}@tI~62P03FhcVp~@4~c4%vY@87VT3T306w4}PTx@PI8FHG4416Iw T FA 5 9 0 T 0 G G7 f 9 0 0( 5 R Tpb { 2 w Y27 YW 9W T2p cPp4@e7@e785DG3wleg#4eb}33T60PwVF85S@e7354621q@tIp4}À1XIeF35e0XGVF35gSþ#P0c@d7vG44qXTF AI 0 f 0 G02I 27 G G 5( F G7 f ( F 0 I( RI T FA 59 0 T 0 G 4eb6Ghu4eblS41#@48b3580z7e0@mXTHiVF85Sp4X24e2651±q1aE@~@~4Ppc@87È@87e56G6wtVgu4eb¹R ü 63T30XweFe5Sqa4ebG G7 f 0 RI 0 97q 0 0 0 G 2Igp 9 T A 0 0A A27 G7 f w T G 0 0( G7 G TAI 0 @B3TcPp@S321X(3F~48bh1l4eHY39P0@eIp462t6g4qc@B3T60PweFe5Spdb4ep~808735c321SEebBcj4eb6Ghf 5 R G w T A27 5 02 GIW 57 @tI 62P03FhcVp»4@yVY@e7eT6T303wcþ8b@le0@hP7eF6580XGVF35Qe06T4}¤36T30XweF85qx@P743pVwP56Fw c@7 w 0b{ G7 f G2 9 A2 0 f f T7 0 f f T 0A w F Fb { 97 5I2 F G7 f ( F 0qI( 0 G 0 @XweF85DGhVY@@PwVF85DGhVYe6T44Pp694P2wPT4}o@ef@4uXTQHieFe5S©4P24V235@Ì@p48b624b u T T T 572 G T 5 R 9 G 0 F 57 A27 T¥ F h3f80vGVYoVY@E@PI8Fd7VG@tI@z7Xw8F32cPAe0@xÌH° ¯ 4@1 ¯ 3yQP2pº7VF35e0XGeF65`YHGPFX94@d7vG4}{ ü w FW T 0b 6T}Bd1Xq@eI8I 3Ae0c}4jc8b6Ghf ¢ 0 9 T G Tp TAI 0 RI FW T 9 0 fp2 0 G GpIW7 T2 T w2 w G2 5Ig F @o`YXGXFP94@87VGy@37PwVF35h4hc8bac1l@d41XIeF4Xp39416I@d7vG1lXfI u G u 874h0 u c1PIeFSº769e0he03Tc8bG 5A T2 G 5 0 f 5¸ FW T 5 R G w T A27 5 02 GIW 0 f 0W ¥ T w F G 1eIE~1`YHGPFX94@d7vG1tI62P06Fjep©4@#VY@e7eT6T303wc©8b@l6T16I3wlÌH° ¯ syÛ@V0@z7mPT4eb#`2@`YXGF H@E6¥p@U6T@3I3wl@X° ¯ sypX2i`YXGXF@E3¥cP2x416I3wV04eb1@ef1cXT©x624b u @e0@Vf354ebVG4E¸ 97p 0 27 0 f 0W ° ¥ F 97p 0 F A2 T 0 G 97 5I2 F ¾ 0 5I 0 5p 6Vf14tP9@@a²PDf D#Pv48bUXT}4Ppc@87Ç@e7856G6wtVga4}j T 5I2 97 RI ù e 0 G F T FA 59 0 T 0b { TI T2 G F Se01legc@87e5U8TXGebXGF u 3Te0ceh13I}HGF Gp f f w XS0cSqVf14PTpHieFe5Sw4ebo624b u ef1c2 u8TXGXTxHiVF85Sy@4Xp4@87Ç@e785DG3wlegd4}1° F 97 5I2 F G7 f 0 G 0 5I Ù F F G7 f 7 RI T FA 59 0 T 0b { û I2 2 u 4 9P390 u V0@ihef14Qi eFe5Sqc@XF4@e7U@87e56G6wtVgg62300 u 6Gls1XI8FSP739e0q48bej1tSi8TDG@½7©I u { 57 T 5I2 G7 f A27 FA 5 9 0 T 0W 2 G 5 0 G GpIW7 w R ! À ù "A£` ER½ e ù ù ùd ù À ù hPeÁ@!VÄIe½DEuEÁdsDEËqI3uEÁdi8ù@½ D`ùf D¢EÿdeSE8bm@z7X039vXT¹Gw µ @¼ef14¹2cµ æ ù G7 G 5 w F RI 5I 7 G G 5 F FW T ¥ 0A F2(7 f fp Fi7 f RI 0 97q 0 0 F F 8I3Ae0Sº769e0aPTo`YXGXFP94Sd7VGÇj±X° ¯ sys`2S448pHG41@ncePfjShcPp@S321X(3FeTXGÈXS0È R@IaÁ@!VIV½DE 4eb8IQ3Ae0Sº769e0uXT©`YXGXFP94Sd7VGÄc1 ¯ 4@o ¯ 3y`2oÌX° ¯ syhP2s321Hq1( e ù 0 G G G 5 F FW T A27 T¥ ¥ F 0 F G7 E8bGh8bXGF u ¯ 4@e1 ¯ 6BqP2p621Hq14@PI8FHG4EDC4i`YXGXFP94Sd7VGm48b~80SP739V0u8Ip6A3040 »u mÉ A27 T¥ F 0 F( T2 F2 0A FW T 0 G G 5 G 02 u I FW T5 1`YHGPFX94@d7vG}1tIR 2 F 2 w 7 T7 Tp RI F Fb { RI 5I Y27 G 5 @PI8FHG4A41I6EDùDª¥Roa@s6Ae046280fG@vXTPT4}w µ @¼ef@4`2cæ SueT62e06Te046ge05 çÆÆ µ ¦ÆÆ e0654b u 0 ® 1ÌH° ¯ ° Þx ¦ÆÆ µ ¦ÆÆ HRFo6A0444@t}XTm1eI6G6w1p1XI8FcXp1eIa4ebs£3A44c1t}XT18IDG3w1e87SEl@º7eFXG4X2dc8b304X2j A2pIW F 5 0q 2 G 9 T 0 G 0A2pIW F 5 0q G7A9 F F 0 G w F × c}Ý ©urlB}}¤Gvg3ÒÇrÚvcÇ× dtc×qovâÈáÂ±ÖÕ b Ó ß Ý ÓÎ àÓ Å ÔÓÄ à Ý Î a à Ð Ü Ù ÙØ ÐÒ ÓÎ ÐÑ ÃÓ ÙÒ Ö °Ì° 4 3 BIï)H #Pï R ¨ C¨ $C¨ ¨ F©7 $h ¨6 ¨ "¦ R&©ì 7653 37 ¨ "3AXP98ð(úRPC¦â"8#yIì 0A0ËPRïY "#'B¨ (©"&#F §B¦ AX)#&9CB °¥8$"H©©6©"¦RRPê&B 3 "¦0 â'&$c1C¨R'B7 3Q1AP(¥3lCëRRB7 3 C¨C&¥"P'#$8 17"¨8$ÿ©# QP#F C¦©l1XëPCïP¥¥÷rS98$ ('& &ï # B3 7& X # ) 3 # HB 3 X 7 Xï .
$ï 7 # V 7 ¨ F 7 & 7 ï ¨ F # B 0 o F 7 & l X & ï # j 0 3 # X V # D i B 0 g @# # PRvR31¨C¨PEî©G©"¦%Rê3âPR·n1m"3RP98ðkR%©1ïR&s8F©e(ê3sh8BC#'(y8X1ë#Y RRVB C¨¥& 8$H I#)HC#÷B 8÷# 8C#RX 8&GR3C¨7 ¥¨ 8$(9(¥3$©ì "&RPuPC¦2RRf¨¨ µRRì3 3Cïë 10 R8÷B ed ¨ 'B©"¦)©êï1RX1Xyú(Éø©{(C3P!F)#"8"#©(µ'7'#IËRP68@C#)4CëP87R '¨RC##Y RV8&C¨v¥AX8$ê#&¥¨¥0$©°CHF## A0Q3 ©¥8$C¦2A0ÿ# ©T·3©"#)¨ §P¤¦ 'ï C$F # B 0 ¨ @ 3 ¨ ø # ø 7 3 3 7 @ B ô Y B F F & # ì & ¦ # #R $ w F2p 0 G 0 G SP069ewP56FXG4cdc8b§fD±E4ebi8IG 6T80SP739V0m1VIe565ee0@44u4} Ì4csye60e0 T 1eIV535U±ùc4eb@qcP24eb@4cmXTP939e0XG£36G6we0e5@3Iw G 5 5 0 T7bg 0b { ®° ¥ 5 0 d 0 G RI ( F GI2 Tp G F 5 XT#1XI8FS8762e06Te046ge0uc8b@})f ²Ix4ebS48bxXYc14Sz7hdX039VwP53FmXGc4u4ebpdù cHGS48bG F 2 G G 5 5 0 G RI ù e À 0 G G7 G 92I T2 0 f w F2p 0 G e F G7 GD@½7i4} Xu o4X28F32V03TV043gV0PTaXG§324b u 4X28F6GV04eg6580P2X2y4tp6RV0@z7lwvG4e§h63T1zq37P09 w R 0b{ ¡ ( G 5 5 F F 0 ( 5 G F F9 5 w 0W Tp f 0 0 35@6q4BHGqdbXw4b @P069ewX53FrXG4cm4eb£@~30c R 0 02 F F w F2p 0 GRI w Ât065e032h w 40 G A 2 5 7 0T G 0 A I f 2 0 q F cRf 3pIewX53F0 ~48b474@fp eI@GPT691zq' 87e5 4395 41e6GeFFPI 6w11@3I92 7 Xw4W 0 A 8bÂc 9©4peFP2¨( Â 324e0ý 6T69e044j6gA50 hp601@PIw2 87eFe56G14ebq006wG 1y3IG2 3Xw43TF V0a@4V0z70TW 8b3541X25wG tIF 44@@4h44eb8p@HGG 4a14@8b(02IF p4reb0 u@VYg @EpX2¨7 1ÂF @SPfT7ý( pP2@zYFT P(eF7 @u5 du8bPTFG G { G I e05 f R A 2 7 }350 xv h« b G2 h 7 Ic 0 T4VA@7 u eI4@374394hw1XI8FS4PgeFVTPT448bm4P2eF62e06Te046ge060@87e54eb136Te0@37e035cP2F Â t@B448pHG41@f 5 G TA 0b 0AI f 2 G7 FA 0 G ( G 5 5 w G 0 G T w ýRI 0A F2(7 0 G T ' 0 G7 02 A27 9 5 ý 0AI f 2 Tp FA 0 G 5I¸ A27 ¨ 0 c8bwmoS1Hq8FS6(4sc@Sz7e0dPTF Â 39cjh§@PIeFc£Hø4q4ebd@s@° ¨ 4S#z Â e0654b u 04Pp@S4P2eFVG@87VG 48b8T32V03Te0c3ge05 ü 4eb6Te0@37geb@tg`2 1XI8FDG3w116IPw41k1XIeF4tXø46548bHG3F0 97q ( 5 T 0 G G 5 0 G T w GIW 2 0q2 FW 5I 2 Tp FA 0 ( G 5 ¡ YW G 0 0( F F F GI 0 G RI w G w7 0 0 G I T 5p F 4X28F32V03Te0c3ge05 ¥¢ öQ3A808735c321vXTvHGHRv@eI¹5tµ4ebrSi30@e7856GS4 ii48bT u 4VbiX° ¯ e041X(¸ x f u u f x x w u s d p h f d b#£0%yvtrqigec a2 0q2 Ìb@PI8FDG3w@@3Iw Pw4PIVF35l#4@1PIVF4tXø44X2PF39cjh4@PI8FE3¥uc8b1tIR Â 35e06Gh@87@4u4eb@#1XI8FDG44tph@7 FA 0g A27 2 Tp FA ( 0AI f T2 G7p 0 0 G 5 ý 0 f 57g 0 G RI 2 w2 R 7 T 8T@eI¹5t 4ebU@d43pPIh4eb30S8785 8Is6A3060jVI4yP9X9F u 0 u 1eISz7e0dPTc8b1E@4Sº74u9 ¹gtÉDiX04@3Iw GI 0 G RI T w 9 0 G w G G w 5g 2 T 5 F G 5I¸ 02 9g f 0 G F w F2p 0 G G 0 G 5 9 G GI 0 G 0g 0 Gb F f 0 G T7 0 Fp c8bxP2vX039VwP53FUXG44c8be8Iv@XqeFSº769e0X0PFieI8T@eI¹5t48bDG3wlc iU1X(a0 u V0354b eIU@4XA41( G F 0q2 w 0q 7 T7 0 T Fb { 0 97q 0 0 ý 0 G A27 0 G 0W 2 G 5 7 A2 32X0432@@3IpvY35@w}@6T1Vq65e0UXT4}y36T4PpS@621P(6Fac8bq4Sh48bx62300 u 6Glq1XI8FSP739e0w4EC 2 w Tp w7 5gg7 F F T 0 G RI 0 97q 0 0 0 G 57 ÿ 0 G 0 @37w0 u 3AV04P§db@1eI44S80DGe035VwPT4¹AHh3fV0dc8b¹Ry µ @3T4Xp@S621P(6F48bV0@»448bae0654b u Y § Ø Y `Ü Á è Á E§ Ø ÿ qÿ Ü Á è Á § Ø ´ ¥´ Ü ¨ E§ ² Y èè ÿ ² èè ´ ² ² T7 5g 0 0W T 97 2 @©6Ae0VT3Te0cc iatzY7 u @Sz7w vG@E413Ic3211h14m48bd8Ih1PIeFcPp1VI4ebj 3fe0VGVY4621P(6FUXWe06GX04@3Iw@Eb µ XR~e0@Vf354ebVG4E¸ 57g Tp 0 0(I fIb 0 G G 2 G 9 T 0 G T T 0 0 9g f w 7 T7 F 5I 0 5p ² § ¶ E§² µ ¨ tz³E§² « ´ YW G 5 5 0W 2 w T2 G7p 0 w 0 FA972C 0 G ¢ RI T 7 RI 2 G7 #3Ae062e06Te046ge0lo@3741PIeFSE3¥a60432V035tXø4Û@4E48bE3T}B@~DGe0iSp1PIeFSPfF 5gg7 ( w 0 FA 0 G A2C G 0 G T w7 5gg7 F F T 0 G GI2 5I 0 0b zieIc4@w4P26F432V035tXø4~48bh4Ea8Iq62@@87rS7 u db@eIc4@~80DGe035VwPT4¹AHh3fV0m4ebB@4B16548bDG4» 'Q¥ l¡Vh y ËTRrªSQ£ uªQIG 8F k U k P s R PH @ ¥ çê t D #e9qxEÏ C A ç t B6öé )Ò #yï ö í ç ð zêx~©h)Ò 6ïzò4t©cp½s·zìBlç ©«» ò ôï éçì êí õ ò rqï i tq òg ö ¼ × âÓÒ×Ü× Ô ÐÓ ÙÚ ßÝ Ü Ù ÙØ ÐÒ× Õ ÔÓÒ Ñ Ð Ï #rÄÇBÇávyràÛ}Þ3ÒÛrÚ4ÇÃ±ÖxBÄvmÎ zÍ ±° ä pã .
9 7 w 5 g 0 G G 7 G F f R T T 9 W 5 g G 7 F f I A 2 G 7 F Au G 9 g g @EXg3F4X2eF4x4ebeSEeb· ¯ 4X(¸ @eIt5x30V0 u 0 u h3fX041eIch3Ae0SEP2h14s1PIeFSEPgVFeTXT4©8Ih6AP0PFc4@7 32c» ÛyVY35@6qþW 3A6044eI4xT I #V0@e06548b©30cP2eF~48bDGp@jVI`5SI sPTxXT4}{ 0b ý 0 0 Y wpA 5g G 57 0 G w T AI 0 f GI G 7 F Fb d w u x gk¡i ¡yb%e d s x x c T w F G F FW TR @e0Sz7}PTc8bxP2`YXGXFP94@87VGÄ@I 0(2 5 I2 T7b AI 0 f 0 G A27 F2C 97 5 R w F2p 0 G 0 GpI 9 R w G 0 @4@874}SEj4eb6Gh48bx4Ss Â 80XGcEtX9@1tIX039ewX53F}HG44a4ebv4PAVF8Tc1PTX9@½7v30@e785Btlc8bG 2 0q2 FW 5 R Tp 0b 5 0 A279 G YW A2p R F F RI A27 G7 0 0W 1XI8FDG3w116IPw41tIp3AV04x32c»D8®1eIV535p4StSº7eFe5pc41tIBPTXGp6280fG@pc@p6A80SPfeFVG3Tmtu8IG T7b ýRI 0A F2(7 f 0 G 97 z ý x 0(2 5 0 G5 R 9W T F F T w G7 F fI SEht@r44epXG4@@q48bHYP9@EVpT 9 cGx Â þ2vB14@e7U48bÈ1tIX04@d7vGÄXTHGy3Te0Sz7u6A80S4P2h14A 2 G7 FA 5 R Tp 0b AI 0 f 9p 9g 0 0 G 5 R 9 5 I T 5p F j1XI8FSEXgeFVTPT4a1tIs6Ae04§324»h48bDGd35X04ydXG3FXwPF4j i4eb1tIdeTXG4Vp3TV0uT u 4ebx ¯ e041X(¸ d w u x xz h v u fz x v gy%%9ym}i9qg0 ±Í° Ì° ° 1° z° d ¯ Í T2 G 0 Þ « ý 0 fI 9W7 c1PIeFSº769~É dh1jr1H° ¯ X04@Ä{ eb@4P2X9oÆ G 0 F z ¨ ® Â ý Á ° Â ý $ ® Â Â ý ´ ~ ° « « « vÞ z ¨ $ Â bý $ ´ Â a Â $ ý «° « ´ « ý ´ ¨« Â ý ´ « Â þ© dvÞ ý «° « ¨ « z ý « ý þ© (7 Ê7 @E6A@d z Â ý © Â ý Á Á ¥ ´ ® Â @ Á ° « ÍÆ }' z ¨ ® Â ý W Á Â ´ Á Â $ Â Á ¨ ° ý ´ ' ® ´ ý W « ý ´ ° « « ÍÆ ' z Â ý ´ ´ ´ Á ' ® Â ý Á ´ ° ® Â ´ ý W ´ ~ ° eA5S7 u d@U#cS w7 A2 z ¨ Á $ ® Â ý ~ ° « « 97 F 0g 5 @EXA13I6ÊtS87È{ z ¨ ® Â´ Á  Â « « ´ý z ¨° þ®® ýÂ ´ k~ °° q 65P04yUXG3FXwPF4h`f 9p 9g ° « ý « ¨ « Í U' z oÂ ý W ´ © Â ¨ ý a ´ Á ¥ ® Â ý ´ Á ° « }' z Â ´ ý ´ Á ¥ ® Â ý Á ° « ( Rg 1VIt5c@z70 Õ z ¨ þ Â Â 1k« ° © ý « 35P0cB~HG3FXwPF4j 9p 9gi z ¨ þ© ý «° « ° 341XI8FSe785vG4PpX9P9a48bq3Ae0651±q3IV0@U6TP0eFXGPFX94PWVFeT1lg T2 G T F 0 G YW 0 w 57 F TI G2 5Ig F 0 G RI TI Æ T w 0AI f 0 G F 0gg7b 0IA G7 T7 T7 T 9W 5 @d7vG1thPfq48b}@eVG1o§63Te0Sz7r394q48bsX232l4@Eu3T4iSEb u @X9390 u @u6fP041VI4g 6A80S4P2h14`A1PIeF6G3w@13IrPw4PIVF35lÄ@w6A80SEX2@4v®j1XI8FS4PgeFVTPT4È1 I 1PIVF4£Hø4veId6AP0PFc4@UV0@7 G7 F fI 2 0q2 w FA 0g 5I G7 F fIA 2 G7 FA5 2 Tp FA G 9gg7 5 TAI 0 f 0 G F 0gg7b G7 RI 9 Fg G F I T 5 F 7 0W 0 G G7 G 9g F F cj4eb6Ghi48byHR32l4@EÈP9X9F u SEb u @Sz7Pwcj`YmPT2 u 4ebv@PIzqE3bli48bSE8b 3AX0PF4hPfvXT¹G jdc©4@x46y£1PIeF6G6w113I¹wDXw4jXIeF65t©4S#1XIeF4tXø41tIT394he48b8I6AP0PFc4@7 A27 T¥ 2 0q2 FA 0g A27 2 Tp FA 5 R 0AI f 0 G G 9gg V0@q6Y4ebp324b u 4jc8b6Ghr41PIVF@Sx6A3044eIcU8THG4ep6Te0c8b}80Se785vG4PpX9P9uÍ ¯ 4Su· ¯ 6Te041X(¸ 57 0 G 0 TAI 0 f Tp 57q YW wpA 5g 9 5 0 G G T F A27 5p F 3448bDG@`Y6GX0eF@Sm@tIc1PIeFSº769e0Q ce4ebT u 4ebdH° ¯ X04@ÄU3cj4eb6Ghi4P2cdbew@`f Xf8FG TAI 0 f RI 57q 7 5 R T2 G 5 ý 0 G I T 9W7 { TAI 0 f ( F 57 0 97 w7 YW wpA 5g 57 G7 G 0 f T @E8pDG@©3A6044eI4iV0@aS48buh1eI©ebXGF u 6T30@e785SjVI`5tDG@cDii4ebue0@416I4vDGP09 u mÉ w G GI w7 0 0 G 57g f w Tp I p d w u x w x q0y¡iy~yw Xvs }{©ybgv d p d p xz x w ° Ó à ÐÚ Ã tÓÚÑ â Ý Î Ó Ï Ù ×Ò ÑÓÎ à Ý Î Ü Ù ÙØ ÐÒ× Ú Ð Ù ÔÓ âr à Ö j±Í° vÛBÑ§xuÇB}áÄÒ ràoBe}}À3ÒÇrÚvcÇvDÎBsÃqp±ÖÕ .
ωh Ι(σ) R (σ) oo ωh= 0 ±Í° . λh λ h= οο × âÓÒ×Ü× Ô ÐÓ ÙÚ ßÝ Ü Ù ÙØ ÐÒ× Õ ÔÓÒ Ñ Ð Ï #rÄÇBÇávyràÛ}Þ3ÒÛrÚ4ÇÃ±ÖxBÄvmÎ a) Diffusion Ι(σ) .oo R (σ) λ h= 0 σ=e b) Convection i ωh .2 0q2 FW 0 G 5 R A27 Â 97 5 R w F2p 0 G 0 GpI 9 R GI 5 97 w 5 1XI8FDG3w116IPw4xc8bw@tIh4@ycz X9@i1tIhP039VwP56FeXGc4p4ebq4PAVF8Tc@wXTP9@½7d@eIr@EPg6F4P2VF4g 0 G 0 I ÷ Â RI 0(2 5 7 5 R 9W T T F7 5 A27 w 0 G 0 T F T F7 5 GI c8b635@6q0 u ~Q Sx14@e7i1tIqX04@87VGucP2@3fV04@xX039ewX53Fpc8bq4XAeF4X2u4X2@3fe0u@eI5 Tp 5p T 0 G Ê f R 7 T w Â T7 T w G7 G 57 F(7 F 5pg F GI c1PIVF44eg4ebxeI3560n1VIt5©zY7 u d3TV0@z7V0354X2F @hV0@z7aS48b`2d@VY@4P21@PfV044vXT@eI`5cý 0 G 0 G 0 FA Fp¥ F 2 G7 T 0b { 02I 27 G G 5( 0 f 0 A27 c8bs624b u 62e065£Hø480XGcaXT©1PIeFSE8pHGeFh4}qS41Q@Eeba6580S37e01a6T16I3wlW Æ Æ 4@P069ewX53Fw 0 G 0 9 GI 5 Tp 5p T 0 G G FIg G7 G G7 w T ° ý 0 TA 0 (2 5 0 G G7 c8b3T1zq37P0}@eI4@PIeFc4ega4ebUP21lSEeb}Sd304X2eFzd« © Â 324b u 446210 4@e7a48b}S48bG T u 4ebhe0c1Eq4}xSP04Sd7VGtyX9P9F u jc8b6Ghqc8bsdbXw4b u @tIR Â q1Hq8FS6(4r@z7V0}@h14@e7o7 I T 5p(C 0b { 9W T 0W AI 0 f 0 G F 5 ý 0 G7 02 9 5 RI 0(2 5 XTe035c8bÇ@e0@tI6Re0654}ij ¯ 4X(h30V0ÈSP069ewP56FaXG4chc8b©1Q@E8b6548bSe7£X21P(VF1h48bSdeTVG@87VGT F 0 G 5 0b { F ¸ T w F2p 0 G 2I 27 G 0 G 5 F 5I 0 G G7 5 GI 5 Tp 5p T 0 G 0 0 T ( Rg 9 0 G 0 92p GpW AI 0 f GI G 7 97 F Fb @jVI41PIVF44Vgx48br@h3f4dbVw©1eIt5cSz7P04ebp@XPF44rzj4j4eb6Ghi@eI¹5SI u #QeIXT@PTiXT4}{ d w u x w u ds w p p w x d f x 0y¡i¡tyX£¥£k%y#rgy 0bf Gg7b F F G GpIW cu6580c@4X2}PT4eb}c1l@7 XA@87yXT~V01oD±ùcP2B1eIV5354ebyc1l@d4P24eb@4r±Ye7BXGS@44epXGXF4SX2B1eIV535a4r3T30c4jVI4g F T F 5I Æ d F 5 0 0 G GpIW7 ( F GI2 T T F0A 9g f7 F 5 0 I2 wpA 5 A4eb6Gh1VIt5c@37P0uc8bdbXw4b u X2F @d14@e7mPTwe035c8b~SE8bmT u cebuV041Ea4ebo14@48bHG' I 0 f ( Rg 9 0 G F Â RI 0(2 5 7 F 0 G G7 G I T 5p(C 0 G b(pI 9 351VIe535j148bHGF u PT48bDG4ebiSEebB£@z7h@43T4iXT48bB@@Ìl@hc@k6A80P2@t@7 u ' 0 GpI F AI 0 f 0 G G7 G 2 0 f GI2 0IA F G 0qIW7 GpI G FIg T T Â 1°¡x x z 14@e7x48bhX2qX039ewX53FiHG44x4ebh1hHY39V0PT3F6we04wPTX9@½7eSEebþ3T30c4jVI4©eIXT@7 0(2 5 0 G F w F2p 0 G 2I 5g 9 R G7 G 7 wpA 5g 9 XGacp4£@P0cu9 W@d7vGHY41a@4PT©48bDGi4ebt @lf@åhÁÀ8ùj²BXTQs324b u 63Te0Sz71PIeF6G6w113IXw4W F G W T 92I GI2 F AI 0 f 0 G F ý 0 T w 2 0q2 F 5R @tIÇ6350@6q0 I~ecl® Â ý UÁeI1tImXGr@i4XAeFeTc1XTP9@½7#c@iP069ewX53FHG44i4eb#18TvG@d7vG@eI5 ÷ z © w e 5 R F RI 0 GpI 9 R A27 w F2p 0 G 2I 5 T GI Tp 5p 0 G w R GI2 F GI 5 Tp 5p T 0 G GpW ý RI 0(2 5 7 5 R 9W T F GI c1PIVF44egTh4beÈ36G@½7Q`2pz@cdXTd@jVIwc1PIVF44egh4ebdj4 Â @14S87od1tIiX04@d7vGaPTd@eI5 3c1PIeFSE3¥694}1tI8T@eI¹5tËSm6T30@e7856G@cB1H° ¯ e0c1P(¸ T2 G7p 0 0AI f 5 R GI RI w G w7i û 5p F σ = e λ h .
° Â x z 0W 0(2 5 F 9 7 T ¥ GI ¦ Â 5 1Þx v32300 u 6Gl14S87p6A80XGhPfXFivG4Vp©@42 s E@d1tIR 5 0 0A 9g f7 I2 wpA 5g AI 0 f97 F 0g G 0 G w T G 0 G 0W w 0 F @eIe565U44epXGXF4Se4r3T6044eI4hjc8b6GhE@EXA13I6ÊtS87854eb604P2VFeI u c8bp32600 u DGt~30c32e065£Hø4A 5 ¥7 f 7 F 0 G GpW T w ( 0q2 w FA 0g 0 G 5 R 5 0 T7bg 92I ( wpA 5 1VI¡§PTV0354ebj4}@V0@z74X28FDG3w116IxPw4PIVF35l4ebi1tI1VIe565e0SE4HY41scP23Fc4jVI4g RI FW7g w 0 G T7b F AI 0 f ( Rg 9 0 G 0 9 Tp f ùd @Q`YXGXFP94SE@z74ebe@EiHGj4eb6Gh©1eIt5j@z7X0 c8bqSPPFhvG4@ DPù9e!D CeED8ùDEWfPq¤ Re±ùtfÀý d"Ë"z¤ Â uT¤xdsI%ÁeIzvPù9!¹DPmfzD@½7P2£4@xV0@z#cP28FD3@@3iPwcjXIeF3lqc8bG d ý ù eä e ¤ ef Gw R F A 27 T7w ( Gw0q2Iw FA 50 g 0 A27 ( G7 FA 0 G GIW 5 R w F2p 0 G 0 T F 5I 2I 9 R GI F ¸ 4S4P2eFSEPgVFeTXT448b©8bSt1tIeX039VwP53FaHG444ebe4PAVF4P21Q1ÄP9@½7veT@jVI`5teT¹GicÍ ¯ 4X(QP2F 3A808785vG4PpX9P9HYP9@P78FvG@EUV0@BSEeb41VI@z7V01tIc1@P7441levY35@PTuj4eb6GhjSEPA16I3Êl@87e5U4}{ G T F 9 57g 57 G7 G T2 T 5 5 R 02I 5 9pgIg 0q 7 F AI 0 f97 F 0g G 0b Â X9@1tIP04Sd7VGc4}PT48bDGa4eb@PI8FE3¥694jhf 97 5 R 9W T2p F AI 0 f 0 G 2 G7p 0 0AI 3448bDG}c1PIVF@SU1tIU8T@eI¹5tâ@3T30S87Èyj· ¯ e041X(¸ TAI 0 f Tp 57q 5 R GI RI w 5{ û 5p F λ h=1 Diffusion c) AB2 d w u x z d £ x x ¡ gy%5by9 i§¢#he Í Ó à ÐÚ Ã tÓÚÑ â Ý Î Ó Ï Ù ×Ò ÑÓÎ à Ý Î Ü Ù ÙØ ÐÒ× Ú Ð Ù ÔÓ âr à Ö @±Í° vÛBÑ§xuÇB}áÄÒ ràoBe}}À3ÒÇrÚvcÇvDÎBsÃqp±ÖÕ λ h=2 σ2 σ2 σ1 σ1 σ1 a) Euler Explicit b) Leapfrog σ2 Convection σ2 ωh = 1 σ1 σ1 σ1 .
T FA w2 T f w 0 G 0A2p 9W 5g Tp 3AV0eT46pewXT43T30c@d7vG46pewX53Fa48b35c44¤6fP04@eI441XIeF65e0T 7 GI2 F A27 FW T F F f T7 G @4XTÃ4@s`YXGXFP94Sd7VGcP2PAX9h@§8Ig6Ae0V535t06RV0PTh`YHGPFX94@d7vG4P2@k4P2hPT4}þ@PI8FDG3w@@3Iw 5 F FW T F RI A F Fb { 2 0q2 Pw4PIVF35l4P2Xq@1P23TX044epVGa1Hq8FS87XG8F@44@PIeF65e0P2o3AV046230l eb@lw@±q4~cj4eb6GhewvÞ FA 0g ( F 9Iq F FA T 0 G G27p¥ Tp T F Tp 0W GIW 0 7b TAI 0 f 0 G A27 ' 0 G 2 T 5 F G 5I¸ 2 FA TIggI 0 G F T F f f w 0W 0 7 c8b4@a}4eb1VI@z7V0yPT4eb1E1XI8FDG3we0X54}80HGeF1l414ebiP2rVG4Vp¥¥3Ae0eGXGhh13I3230l~1zqEb XA41I u )fE@åhÀ h@87u48by@1eIV535#@c4eX04@d7vG#3AV0@º7693w46230li1zqE#XA41I u jc8b6Ghf 9p ù 0 f T 0 G RI 5 0 27 GpW 9W T 5 0A 0W 0 7b 9p AI 0 0 G w 0 G 0 T F 9 R A7b GI 5 0 G G2 5Ig F2p 0 G 5 0W 7 f FW T c8bvP069ewX53Fa48bvcPAeFcP2x32X0P9@½7@EU@jVIwc8bÈ¹Rz@87VG1lXf44qXY39@XqeFSº769e0lxzYqh`YXGXFP94Sd7VGcP2F TX48bu3AV04@z7y@eIe56548b@V041E}4ebtSmV0432V0~c8biP2jP9@ef}V0@rSEebr46g80vGUXf8Fl@B35le42 F G YW Tp w 5 0 0 G 5p(C 0 GRI T T 0 G F 97 T 57 G7 G T T 0 GRI 0W fp 3Ae0XGhPfPFhX2©3A@Xq1VI3Te0dl@37vVG6Te06580P2yS @PI8FjPp1VIv32X0eF4@e785a48bÄXR635@6q0 u ~ P9@m1tIR F 9 7 F 0 9 5 0W 2 w G F RI 2 G 9 T G T2 G 0 G F 0 I ÷ Â 97 5 X04@87VG44~PTo4@dP069ewX53F~XG4cw4ebdcPAeFeTc1}XTP9@½7@eISEPg6F4P2VF4u4eb}4jc8b6Gheb@luI ~@tImS48bG 9u W T 2 p F A 2 7 w F 2 p 0 G 0 G p I 9 R G I 5 9 7 w 5 g 0 G T A I 0 f G W 5 R G 7 I T R A27 w T F ¸ 9W 5g 2 0q2 FW 7 G 9gg7 57 0 G 0 TAI 0 4ebyÍ ¯ 4@wS ¯ 31X(} 3fP0c1eI41PIeF6G3w@13IXw4q8I3AP0XF44@we0SDY48b#624b 4jc8b6Ghf Þ A27 ' 0 G YW 0 Fg G 57 T w wp 0 F GII f 7 0W 7 f 27 T 0 G 5I dvö4@}ec8b©3AEHC4`Ywe0S3TV0@z7©db4@cepeTXT~@jhqut zYq@EegdhPf8FmVG1cebT XY691Hq8FSP739e0im35@Ìx@PI8FjPp1VI62P0VF4@87e5eu@Xq1VI3Te0eI 6Ae0P5c6¥e0}XTSE8bx48bDGi@`YXGPFX94@87VGT 0 G 5 7 0qI 2 G 9 T G T2 G 7 0 9 5 G Fp 5 F G7 G AI 0 f 7 RI FW 0 G GI 5 97 w 5g 0 G RI T7W 0 G 2I 5 0b ý RI 0 97q 97 T 5 c8bm3@eISEPg6F4P2VF448bd@xPTVF@E4eb»@v@V01tI6RV0354}{ Â i@6T4Pp@S~P9@efp@tIR ¥¢ ¡ 8IG V01PI69©VY651XTSEebv@eIÈ@EXg3F4X2eF4x30c4jVI4v44eb6Ghq4X24dbew@`f Xf8FSE1PIeF32116IX9' T w 0q F G7 G GI 597 w 5g 7 wpA 5g TAI 0 f ( F 57 0 G 972 G 0q2 w 9 ¼ uz 9 `y9S9V u p z Gg7b F 0 T2 w 57 T w 0 jT6580c@4 P26Ae0654PAVF413IaV0@de0@37dc8bG GI2 F F G F 5 R T2 G7 F w2 5Ig F RI F GI ý 0 G 97 RI w 5 w7 G7 G S4mXTmXT48b#dbXw4b u 1tI}41PIeFSEepXGjm@604@87VG1lXfy@~PT~8T@eI¹54e48bÛX9@B@o6Y@e743p6w@SE8beIT P9@efBXT Â @60PÊeF3ge0VG}hPf8F4eb32cb u `YXGXFP94Sd7VG£@q@PI8FvG3T4~48b~4epVT44d8Ia4P2eFVG6Te035e0P2BPTr¹G 97 T F T T 0 G 0 G 0 FW TRI 2 0p¥ 0 G 0 5pg G ( G F F z x® w F 5 ý 9g f w 9 c{t ý ÚdbPw4b u @tIR Â oDiX04@3I£X9@7 5 R @tId1´»© Æ Â ý Æ j1tSXTBXGhPfPFe`YXGPFX94@87VGuvl4ebS48bG u 4ebpSz7v4vB % x 1tIR GpIW7 F F 9 FW T Þ 0 G G7 I T 2 w 02 ³ Â x z 5 P039VwP56FHG44d48ba4PAVF4P24X2@3fe0BlSz7e0654b u P9@m1tIaP04@87VG4cmXTdvö1PIeF6G6w113IXw4W w F2p 0 G 0 T F T F7 5 Þ T 0 Â 97 5 R 9W T2p F Þ 2 0q2 F 5 R A27b 0 GI 0 G 2 G F 9 F Þ T 0 G 0 G 97 TI 97 ( FI @tI4@46548b@m48bhvyjrd8Ie6A80HGXfPFrPTddv@37e0654b u cvdeI±Y7 u c8b4P9SrVG1hPf@w4X211( 335e0Sz7V01XTpBT1tIx`YXGXFP94Sd7VGBSe14@e7e4ebSEebvc4£@1Hq8FS13(4#c@@37e0mPTF Â Q624b u Â ý G 5( F Þ 5 R FW T RI 0(2 5 0 G G7 G GpW0 G7 02 A27 9 5 ý 0 RI 0(2 5 7 5 R 9W T 57 TAI 0 f GIW G7 I T 5p(C 0b F ¸ F I @14S87§h1tIxX04@d7vGoV0@4jc8b6Ghk8b@lqE8bG u 4Vb6Te0414o4}{ jÍ ¯ 4P(gP2Ã2 u cebT V0@448bDGh87eGcD14c~Ô654eA@`IjebVG41tIþ4Sj416I3we04ebq1tIqeT@eI`5lö48ba@3T60@87È{ 57 TAI 0 f Gp 0(2p Þ 0 5 5p R A27 A2 T 0 G 5 R GI 0 G RI w 5 ý 0 F 5I 0 G G7 G w 9 57 GI 5 Tp 5p T GIW G7 G GI É Â 0 cz ¨ #324b u P21X(eF1a4eb}S 3A80z7PIae0@eT@eI41XIeF44Vg8b@lSEebae0@mB ¨ 624b u Se0@37x4P2eF6G6w113IXw4i4ebm1tI#4SÛ ´ ¤© Â §32cb u P069ewX53FHG44e4ebv3T@±q37P0@jVIv41XIeF4cegp7 T w ( 0q2 FW 0 G 5 R A27 « ý 0 w F2p 0 G 0 9 GI 5 Tp 5p T Se0@374P2eFSEPgVFeTXT4u4eb}1EU1`YHGPFX94@d7vGwc8beTXGhPfPF~@eI~41XIeF4ceg£36Te0@371XI8F6G6w116IPw4o4@7 T w ( G7 FA 0 G 5I¸ FW T 0 G F 9 GI 5 Tp 5p T 7 T w 2 0q2 FW A2 2 G 7 F A 0 G G I Wu F ¸ F I T 5 7 0 b { 0 2 I 9 7 w 5 g 0 G G 2 F A A 1PIeFSEPgVFeTXT448b#eb@tQ`2ujÍ ¯ 4P(ÄP2Q2 u cebie0SqV03T4}eS41Ä@EPg6F4P2VF4i4ebeI1XI8FXGc4@7 X2weT@eIuc1PIVF44egI h1zqEevG4edXGÈD1zc4Ä@X° ¯ P0c@ { Xw443p dPTF 1tISº7h14XY1lg F GI 5 Tp 5p T G 0 7b Tp f F ® ° I2 9W7 FW w 7 5 R 9 F fI2 9I Xw8FvGPTeF6580DG@87@4db8THG}304X2jB3@eIe565e0@Ecg u XI~6044eI4eIh6A41X(eF3TcrS7 u 48bDGSE3A@aB4}{ w 57 w F w F 5 0 T7b 9 wpA 5g G 02 0A T AI 0 f (7 Ê7 0b d w u x £ § 0y¡ibk%btx ² ® ¯ ® © p d w u x u u h ® x h x w u d « d f x © ±y£°~uQqgk¡iyktkqsk#rg¡kh k¬y£0q 0kªes º ¹ » 'k%¸G·ApI ''kleb6µ F kR i µ t ¶R × âÓÒ×Ü× Ô ÐÓ ÙÚ ßÝ Ü Ù ÙØ ÐÒ× Õ ÔÓÒ Ñ Ð Ï #rÄÇBÇávyràÛ}Þ3ÒÛrÚ4ÇÃ±ÖxBÄvmÎ @±Í° .
DsAz16I w 448bDGUc1PIVF@S@tI8T@eI¹5tâ@m6T30@e7ÄyjÍ ¯ V041X(¸ ® G2 TAI 0 f Tp 57q 5 R GI RI w 5 { û 5p F Diffusion λ h=.οο σ3 σ2 σ1 λ h=2.8 λ h=2 σ1 σ1 σ1 e) Gazdag d) Trapezoidal f) RK2 g) RK4 dÍ Ó à Ð Ú Ã t Ó Ú Ñ â Ý Î Ó Ï Ù × Ò Ñ Ó Î à Ý Î Ü ÒÙ Ù Ø ÐÒ × Ú Ð ÎÙ Ô Ó â r à Ö Ö ±° vÛBÑ§xuÇB}áÄÒ ràoBe}}À3ÇrÚvcÇvDBsÃqp±Õ Convection σ2 σ3 σ1 σ1 σ1 ωh= 2 σ1 ω h = 2/3 2 .
R@e@4@87Q3A6080@e7@E1ph1zqE#6Y4eb£cz ©Â 1tIX21P(VF1i4ebvSdVGSd7VG4jc8b6Gh41XIeF4cegX9@7 I 0(2 5 G2 57p( 7 0 7b 0 G ¨ 5 R F 5I 0 G G7 5 T TAI 0 f Tp 5p T 9 TAI 0 f 0 G 5 R w F ù È À É 27 T7 cj4eb6Ghe03Tc8bp1tI©30cP2j@9 EÁdDQËuIuÊ@SgeIG 6Ae0V535t06RV0xPTlj`Y3580VGSº7Q4}{ 5 F 0g G G 9 0b £ À·1tI¤ef6580e4X24@37P0a48bxX2£EDùDª{±ùdµEÿdw30XÊXf4P2hx6Yc8bG ¡ 5 R G ( FA 9 0 G F ù F F F f 0 T7 5 R T F G 5 w7 TI f 0 G 9 0 0( 57 TAI 0 f 0 G 0 Ib2 T27g 0 Sv@½7@VI4P2u80Se743p6w@vG1u48bXYX9@8765432@ee0@448bDGiV03T4ebÇ635@6q0 u c1XIeF4@4c iv3TX0eF65e0T 5 9 7 { 7RI T7W 0 G 2 Ê f R 7 T w 5 T7 ý 9g f w 0 f T 5 R 9W T2 @PIzYeÄ@UXTeF@448bxvr4eI6530)1eIt5q±Y7 u ~43A6030eI4g Â @oxDiX04h13Ivh1VIU1tIvP0c@d7vG44p 0 f 0W G 0 9 F AI 0 f 0 G w F2p 0 G 2I 5 T T 97 AI 0 f 0 F Æ 7 5 h16I3wlx8IxXY69@XPFvXT#jc8b6Ghe4ebtSP039VwP56FXG4ce48b#@v8TvG@d7vGjzY7 u @#j4eb6Ghu4X2P9o1tIR GI 5 Tp 5p T 02I T 9 G7 w F ù È ù È 7 T7 G @jVIc1PIVF44eg~41rvG@z7X0S~604P2jr@9 EÁdDQv£ Áe@w8I3Ae0V535t0DRe0BPT~t`Yy35hef@tIm4}{ 5 F 0g G 0 5 R 0b cz g Â 324b u f@å iEÿd©BP9@½7UeT@eI5 0 ù 9 R GI Tc1PIVF44egBX9@pV0@z7dPT4ebs`2 3448bDGo@E6A@d4@s8bVG@tIceb@Tvh@Emu654eA1¹Ij4Sp48bW p 5p T 97 T w F G TAI 0 f (7 Ê7 A27 5 R T7 T f7A ' 0 5 A2 0 G Y 6AEXCXF46fjDiXT}t`Y6548b@4}Äcz g Â 624b u PùReÊfD%fE·EÁdp}PTP9S½7@eIy41XIeF4cegve0Sz7w 0 9g 0 0 F 0g G 0 GI 0b { 0 ù 9 R GI 5 Tp 5p T 7 T TX48b©`2wjc8b6Ghx1VIt5c@z7X048b#3A4XC4`YvPTl`Ye4vh33Tl`Y I u r@iteIG 3fc8b4EC F G AI 0 f ( Rg 9 0 G YW 0 Fg G F 0g G 02 0g G G RI 0W 0 G A2 0 u cÍ ¯ 4S ¯ 3@P(oP2~448bDG~@eIvX04PgeFXG4uw4eb@~3T60@87e5@jVI`5t 4ebmDG3wleg4X2w0 u ¹R A27 T F ¸ F TAI 0 f GI 5 9 9p f 0 G RI w G GI 0 G 0 T F p d w u x w x x ¼ u p w x z qgk¡iyÇ§0byy£k9V 641PIeFSd7j4@3IcP9@~S4ÈXR43VG1h}1tIeT3TX048bvG1I u V0@ 4@dP0c@d7vG44p T2 G Gpg f w 97 GI2 F TI f 5 R 9 5 57 A27 9W T2 HYP9@37Pw44@eI85vG@d7z7q6AP0X9@z7vV0@Ucj4eb6Ghdb4jr6AXf394b u 351±qXYdPwcPTF ® ÿ Ü @mcPp@S4@P78FXGcP2F 9 Fbg T G w 9 w 57 TAI 0 f wp 0 0 0 I 9 Fp¥ RI 0 97q9 F P9@efoVY651§621m3z° 1d e04@373wl~624P2hhPf~PT6580vG@87XT4e30V0#@37u41lj¢8IÄSE3¥0 97 T 0q 7 0q Æ ® Tp 0W G 0 F f F F T FA T 2 w 02Id G 97p XTF 41XIeF44Vgoc8bv@ 41v@)ÝRÅI 44epXG41Sq 48bmXR3651Dq0 u m» Å cP2eFSz7e0654P2g8bHGF u 35P0X9@efT T p 5 p T 0 G R I 0 2 I´ 0 A F 2 ( 7 f 0 G F 0 I ÷ ( T w F 97 DG1Q6Y4ebÈSP04Sd7VGaXT§48bDG48brXR§4@ÇP9@ef§VY651t 8IQ3A30Vw1tIe0@u8T32X03Fh63Ie03Tc8bG 0( 0 G 9W T F AI 0 f 0 G F A27 97 T 0q 0W G 5 R 57 G w 0 w 0 3Ac1P(VF3T4ÛX9390 u XTmVT3T60jVI4cP28FvG@d7vGa48bÄ¹Rc@ÍclsyX2w° G1tIR ® ÿ Ü 48bE68T@eI41XIeF4cegT 02 0A F w 5g ( 5 T 0 G ¥ F Ç 5 0 G GI 5 Tp 5p 0 G RI G w 0 w 0 G 9 F F 0 G 5 T G TI w 5p w 5g9 0 T 0 G A27( c8by@m8T32X03Fh63Iw48b60PÊXF@º7eFXG4X2¼6f48bmVG@87VG8I#32V014dbv6Te04c3A30eIcÈSº73F6wtVgu4eb4@l4cP2F vVG@87VGÄXR69e0@4V0@ 6Yc8b}SEebG u 4syS30PweF6G@e74P2pV1I u 8T@eI41XIeF4ceg8bHGF u 4jc8b6Ghf 5 T T GI2 57 0 G G7 I2 0 w 5g F 5 GI 5 Tp 5p T TAI 0 Iu b A 2 0 A 2 p T p f 0 2 I 9 5 F G R I w 0 p T 2 w 0 G G 7 9 7 0 G 0 5 I 4gc@d7vGeT35c44hVG4ux41rzXGcep3TV0ePTc8b~@p3T60432c6¥e04@3Ix4ebp80SEXp@S6q8Ie654eA@g`2 ÄÄ d I G 9 7 p ¥ 0 0 2 I T p IF 5 p g T 7 7´¥e8Û@E3ES41}4@Pec4eE ×4@E1 8SE30 A 27 ° IG 97 p¥ @41B@4Pg3FcP2eFcq4ebD£3eT@jVI`5toI u he0@xV0354} cz ¨ @d Á ÌÀ ¨ Û »4Eq0 u 02I 97 w 5g 0 G ´ GI G 57 0b { ® ®° « ® Â A2C z ¨ Ç scP2eF@8I6SlUjd « Ã Á Ã ¨ ® 9Ã Â cz Â v1tIS48b}T u PIP9@tI~HG±H° ¯ sB 1eIE¸ ® Â ( 5 Gw7 ¸ ¨ ý 5 R G7 G 9 R F Í ¥ f 5 cz g Â s624b u @º7h14HY1tuPw8FvGPTVF3580DG@87SEdbv8THG4P2c4epVG3A4VbPTPFc16I3w@l XYX9eF@37QSz7w ý 0 9 F fI2 9Ig w 57 w F ( FYA T YW 0 9g f w7 0W T 0 2 XT4}y Â Ç@6T4Pp@S4X9@qVfpvY351B1tIx32@6qi`YHGPFX94@d7vGeTXG6G3wlegcP2U46GX0E36516q0 u myS1Hq8F6GS8785VGS7 Fb { ýRI 0 97q 97 T 0q 5 R 0 0 FW T F 0 T F Tp 9 0 I ÷ 0 w G XT~XGF u X0@mX21loDY@87c3p3wS©@7 1VIt5qeIlXÁÀö4@7 ¢ À%@~Vf3580 P2o8b@lue0S8746p3w@354VA1I F Fq RI G FIg w 5 w7 2 f R T ¡ A2 ¡ RI T G F GIW G 5 w7 0 5 F G F AI 0 f F G G7 G 2 0 F 0 F( TAI 0 f 0 G ( Tp I T 2 w 02 jeAX548buXTj4eb6GhiPT4ebeSEebyc jg@PI8FDG3wjgX2k621Hq1i448bDGx4eb©4P2VF4 u 4VbgSz7x4v ¯ Á ² S eÂ Á ´ ¯ sd Á E1~« ¨ t±E ¿ T´ ² ® ±ÍH° ¯ ¾ ² ² ´³² DûÌH° ¯ X04@Äw60e0 T 48bDGx6g80vGPT8FHG4eUSz74X2PF9 ® 9W7{ AI 0 f 9p f 5 0 £3g80vG`TSI u 4zHG3FXwPF4jDit@e0S8746p3w@mvG1w48bÄ@oc48pvG@37Pw8FHGeF657 1VIt5mh3f80vGT 3fX041VI4y@o4P2 G 9g 0 G 5 w7 TI f 0 GRI YA T9 w f R T 9W 5g RI A F TX48bESÈc1PIeFS87e5VG4XpP9P9BvG3TlU4ebE@}4vy68T@eIc1PIVF44egsSy3Tlj`YeX2@d7vG3560s@3046280VGXTjDiU4ebuW F GRIT2 G T F 0W 0 GRI 02 GI 5 Tp 5p TRI 0g G F wRI w F 0 0 G Y GpIW7 Gb(p 5W 0W 2 w G7 97 T 5 R 9W 5g FW T Tp T 5I f wp f 7 F 0b j1tSr@41eIc~lh@z7rSEebG Â X9@efy1tI©3fP0c1eI4w`YXGXFP94Sd7VGB4@PIeF65e0}V01db4eBPT~e0654}{ ¼ uz ¥9 `y¨yqx ¡£¢ ½ u p f w d u p u × âÓÒ×Ü× Ô ÐÓ ÙÚ ßÝ Ü Ù ÙØ ÐÒ× Õ ÔÓÒ Ñ Ð Ï #rÄÇBÇávyràÛ}Þ3ÒÛrÚ4ÇÃ±ÖxBÄvmÎ 1±Í° .
36580c@4dbm i4a48bxP23Ae0VT43pVwPT4VT3T4tXøeFVGT Gg7 w G 02 0 G F T FA 02 R@aDG3wV04Vpe4ebq8Ic@z7X0v¹G@`YXGXFP94@87VGÈ@z7XweF654B@c4epVGe4ebm1tIx1XI8FSSHq8F@h1VI¥qec8bG I ¥W T 0 G G TA 9 FW T9 0 fp2 RI YA T 0 G 5 R 2 G7 GI f 5 7 f 0 XTÃ1XI8FS48pXGVFxPT4}{ @60@Eegc321X(3FþP2F 6Y@e743p6w@SE4PAHq44P2P56F48bq@tI»VA@6(e05 @xc14ebXGF u F 2 G7 T Fb w7 T 0 0 w 5 w7 97p F FA F 0 G 5 R 57 Y27 GpI VT3T30eIc @PI8FjPp1VIu4eb 1c1eI4ebd@XqVF4qeI VbPTF u 0 u SE8b~eT1eI6G6w1321X(3F 8bHGF u 3Ae0Sº76FjVIeT@7 Æ Â ý Æ w 5g 2 G 9 T 0 G b(p 5 G 0 5A G G7 G 5 0q 0 0 G w T R@e6T4PpS@1e(@P7sHY39@Xq8Fº739V06f80VGvY4321X(3F4ebhP2VGXTjDie06548bi3TV0@z7V03T4ebm@eb@lg`2 I 0 97q 0 5 9 0 G 5 T T T 0 0 0 G F F 0 0 G T w 0 G RI GIW z32X0eF4@e7856A80@7 u 4p G T2 G G2 2 0 G 0A w F G7 G7g 7 ( T 2 G 9 T G YA T 7 A2C G 92I T 0 c8bq6T44Pp694P2qSEebG 8bSE§4P2VF4p 1XI8FcXp1eI80S87VG`TcSz780vGh»4EseIXY41g360e0#§ ü 63T4c8pXGc1@o6A80Se7@E3gV0T 0A F2(7 f G 57 XY694PAF u 4P2zqEu6T4PpS@621P(6FWh8bHGF u 6f80vGjvYh3AX04416ISP04@87VGQXYe932e06544X2u@E4v ü 0 ( F 7b 0 97q 0 0Ìý T T T 9gp w 9W T G 0b F T7b 02 û 0 624b u T2 FA2 w 0A2p 5 wI FW T9 0 fp2 RI 0 T7 G2 5Ig F TI f 0 G 5 R Y 41PIeFXG4c13Iv65444VT43p6wj`YHGPFX94@d7vGÄSz7PwVF35h4B@DG3wleg@@d7vG1thPfVG1he4ebm@½7 641PIeFS87654PAVF416Iq`YHGPFP9c@d7vG X2P01VI1cP2h~6548b87i6T80cepXGeFVG416Io6A1Hq1eI6Te05 T2 G 0 T2 w FW T F 9 5 5I F f 0 G 5 7 G T2 w 0 9 V0@ Vf35e0 62P0eFc@87e5Q4ebwSvP9@7 dbPw4b u X2F @PI8FE8pHGeF48ba3651Dq0 u 4v6448bDGP0c@d7vG44p 57 T G G T2 G 0 G RI 9 F 2 G7 T 0 G 0 Ib TAI 0 f 9W T2 HYP9@37Pw44@eI85vG@d7z7mc14P2XF4ep¤1eIt5a4PAVFU3VG@4hzYP74@37p`YHGPFP9c@d7vGE@e0XT@eI4h13Id4ebo`2 9 Fbg T G w GpI ( 9 5 f R 0 T ' 57g 7 9g 2 w FW T F f 5g f w 0 G 9 TIg T7 0 5 9 T7 T T 0 G 07 f G @X04PWVFeT1l@i1V(@º7S30PÊVF3g80vG48buS@ eI©ebPTF u 0 u PTc8b30XÊXf4P2hx8I©4@1PIeFcPp@eIp7 F G F F F f G A272 G 9 T e0c4h13IeI 62@@87~46g80vG@65tecc8b1mc432l3g4XYX9@8765432@x1PIeFSd7cc16I48bÈ@~VG@3Iw Gpg f w G 0 G T T RI 0W fp2 0 G 2I TA 0 0A 9 0 0( 2 G Gpg f w 0 GRI T 0 G A27b 0 GI 0 G 2 9 TIg T7 97 T T7 T F G RI T 0 G 07 f G T c8bÄ£4@Ed354eb@qc8b§vp@P0cPWeFVT1lw@BX9@efd@3ge0VGaPTc8bU@h30XÊeFh48b@Sqo8I©360e0q0 u 3ge0VG¹TDhPf8Fec8by@vVT3T4X2P9@efu48b#1a4c32l3g4vcj4eb6Ghv41r@ÈP9Sv1tI@PI8FjPp1VI32P0VF4@e785G 0 G 0 G RI 0 97 T 0 G 2I TA 0 0A TAI 0 f 5pI RI 97 5 R 2 G 9 T G T2 0 G RI w 5 w7 0 G w TA27 G T2 G 0 f T 0 9 5 G ( F T 57 9W7 f 5 c8bB@©DY@8746p3w@c8be30cP2eF£4@d32X0eF4S8785w1VIi1Hq1eI6Te0x8I4X2630V0ee0@0 u HY4@4ep6Te0y¢ A 0 5g f w T 0 F R RI 0 97q 97 T 5 R FW T YA T 2 T 5 0b £43246be04h16IkeIG VYSz7kXYV9P5@½7pXTF Â @x3T4Xp@SmX9@qVfx1tI`YXGXFP94@87VGþj48pvGkeIG @eI@37e04}{ Ô pR Ó t ¶R 6Õqr¬ApI ''kleb6µ 86Ò Ï A B çê t Ñ #u9qxD "Ð 6çöé )Ò #yï zêx~©h)Ò 6ïzò4t©cp½s·zìBlç ©«» t öí ç ð ò ôï éçì êí õ ò rqï i tq òg ö ¼ ä pã 0A F2(7 S448pHG41@f 650 u XI ebXGF u 32X03Fh63Ii3AcPA±qeI4xDY@87c3p3wSP2ae0@37e0654P2yf®X9@qVfqHYP9@4ep4 p 48b4XAeFe5651±qI 9 G w 0 w 7 YW 0 F 5g w 5 w7 F T w F 97 T 97 T 0 G 0 0 XYX9@8765432@41XI8FS37PwPFc4@ÃUeX2ÄeT626fe0X543¥e0`YXGXFP94@87VGd4@Sz32X0@SXq46¥vXYX9@8765432@ÈXT6Tt`YG 9 0 0(T2 G 9gg7 ¸ F G 0 Fp 5 FW T A27 G 97 Fp 0 9 0 0( F 0g I u c8bd@þ6YS8746p3w@u@354VA148bmlV1I u @E@PI8Fd7c4h16Iu@4@@y32@Xq11E¸ G 0 G RI w 5 w7 RI 0 5I 0 G 5 972 G Gpg f w RI G2pI f7 0 F( 7 5I @6Y@e743p6w@¤@eIt5@4cP2@z7legwXY691Hq8FSP739e0635tcepu6Y4eb@oVf35e01eIV535y3TX0eF65e01XI±YÄv4X24@37P0}4ebeP2F w 5 w7 f R( F 0 T 0 G 5 0ø T 0 G G 5 0 T 5 9 7{ ( FA 9 0 G G32X03Fh63I}c8bl@4c8pXGc1@~c8bs@yPTVF@E4eb@j6351Dq0 u ~y Â @41462cP9SqVfy1tIa`YXGPFX94@87VGT w 0 w 0 GRI 0A F2(7 f 0 GRI T7W 0 G 2I 0 I ÷ b(pI 0 97 T 5 R FW ¯ zÍmvÛBèÎ ¤uÈB}ÞmÄ¤rà Be}}Þ3ÒÛrÚ4ÛvDÎroÍkªÌ±ÖÕ ° Ó à ÐÚÑ ý tÓÚÑ â ÝÎ Ó ÏÒ Ù ×Ò ÑÓÎ àÝÎ Ü Ù ÙØ ÐÒ× Ú Ð Ù ÔÓ âr à Ö .
31VIe5351XI8FS3744Vp85G 5 0 2 G w2 VG3TX0P9@7ef4eb SE»j4eb6Ghv@EXA13I6ÊtS87854ebcj4eb6Gh©P04Sd7VG¹T~v6354VA1`I4SaX9@i@ S48bG 9 T 0 G T7b AI 0 f 97 F 0g G 0 G TAI 0 f 9W ' 0 5 A2 97 RI G7 0V1hef53c8bVGctpÇ£4@È dùdù28¹£E@e #Æ Õ P0c@d7vG`T~Q@@a354VA148baS48b§6A1±qeIchlW 5I 0 5 R A27 ù Æ 9W ' 27 RI 0 5I 0 G G7 G 0 5g 0 2 @37w~¹Gr6534VA1`I413I6we0@EebU354@P(4DY@8746p3w@ @~@4}4jc8b6Ghae06T48bÈ@a414E8bv60Pw8FSmÉ 0 5 A2 T 27 G 0b Fb w 5 w7 27 T7b TAI 0 f 0 G RI 02I2 G7 G GI 64jc8b6GhXG3FXwPF4hPfr±®P0c@d7vG`T~ ' P04@87VGhXYX9@E1XI8FHG4416I44h1jrj ¯ P0c@{ TAI 0 f 9g F 9W 9W T 972 FA2 w2p 0 fI 9W7 9 7 F 1Hq8F0 6Gg S878516I`w' u I« @ãw« Id 0 w G2 ã ° ã ã @EXA13I6Êt@3Te78560 Õ 3g80vG`TSI { 1ã¡°d« 1ãw« 1ã¡° 0G Í ° @ã °d ° ÍSã¡ S¡zã¡d«« 11ã ¡zã w°° « @i°Í eA@7tug `Yd}UäE7 A' 2 5 0 G @7 h @ A wT f 5o4S z z 1¡° °ã ° 35@X9 0 4Ey@U3IFp79 PA 3ÊHG3FlPw0 XF@4e79g h{ `f gÄ z z ° â à Þ 354VAv 0 5 AI 0 48bDGoÆ 9W7 { F I T F TAI 0 f 9g F 9W ' RI T j ¯ P04SP2o2 u 4eb}XT}4jc8b6GhXG3FXwPF4hPfX04@87VG`TSmDGe0á' «à ÁÞ xâ ® ± H° ¯ ° « ß â ®d 1©H° ¯ ° Á à ß ® zH° ¯ ° á}Þ HRpXYc14@XRP0c@d7vG`T~al8IU448bDGde06T48b444ac14cIdc 1XI8F6G6wj F 92I A27 F 9W ' 0W G TAI 0 f 0 G TA2C 02I 2 0 X2p32@Xq1TQ#Æ Õ 3g80vG`TSI u @reTSº739m48bd8Ir41PIeF6G4ctpsSz7e0~1Hq8FHGeF1lÛ@rVf35e0P2u`YXGXFP94@87VG¹T~' F 0 F( Æ GRI T w 0 G G T2 w2 R9 5 0 TIgRI T G F FW 5 R G 9g T F R 7 ( F 9gg7 Y Fi7 57 F(7 F G 0 0 G T F G2 w F F f @tIvVG6T80wX04hPfeFHYe9X5@½7x4X2XY44S©6PTj VYSEP21SqXfee0X58F32ec8b4X2@d716IvHGBXR~Pù9e!D Ý A27 02 ý 9g f w 0 G F Fi7 9 5 0 G7 02 0 G T F G2 w FW T RI 2 F 5 0 4SÄS4@P74Ü9 g Â Q iP04h13I4eb©P2PTjBSz7e0@Xq8F3(44ebv4P2Sd713Io`YXGXFP94Sd7VG@1XI13(V0hc8bG HRPù9e!¹D ª PTkjc8b6Gh¼' DPù9e!Dm~lseIgPAS87PTk`YVG65t@eI4hPT4ebgebXGF u j4eb6Ghn' F f F AI 0 f f É 0W G F T F 0g 5g F G AI 0 f 9W T G 0b F 57 G7 @X04@87VGxHY8932V03544X2ae0@~S48bG T}cX9@Ä1tIrX04@87VGXTÈXGEHRÇPù9e!¹D~ ÁeIlf@²dXTu48bDG@37PwVF35h4Ã' 975 R 9W T F F F f e F AI 0 f9 0 fp2 û2 F2 0A 0 G 5 T 9g f7 0 5I¸ TAI 0 f ( F 57 0 GRI 5 G 1XI8FHG4E6Ccdc8bx8bHGF u vG@d7vGw0 u SP04h@cDi}1EB64jc8b6Ghe4X24dbewSq¹fDhPf8FÈ@3TV048pz7t0R `YHGPFP9c@d7vG@8765432@48b4EDC44@VT43pVwPT48I~S87t530462e065t06Re062P0P96930jDi@U@6T1vq35e0}XT4}{ FW T9 0 0( 0 G 02 0A A27 T FA G 0 f R w 5 G w 0 27 T7 0 T Fb @4@P74qh iP04h16I~c8bP2yPTjuVYSEP21SqXf~c8br@BfG6RX048ba8Iy11P0PFrvG4eB3T4Xp@S321X(3Fqýu 02 9g ý 9g f w 0 G F Fi7 57 F(7 F 0 G RI 9 0 G G 5I2I 9 Tp f 0 97q 0 0 9P@lS8TvG@EcSz7e048bE8b626fe0P5c6¥e0Uc8beIe3A804@@XG 3fe0VGVYc321X(3Fe06GX04@3IviebXGF 97RI 57g9 5 0 G G7 G G 0 Fp 5 0 G G G2pI f7 F T T 0 0 9g f w 7 eT6Ge0©6AP04c13I~1tI£c@q ¯ 1XI8F6G6wj©X23A4EDC4v@7 u T}aS6GV0qB@`YHGPFP9c@d7vGa62e06544`2 T 9gp w 5 RA27 2 0 F 02 0A T RI T 7 RI FW T G 0b × âÓÒ×Ü× Ô ÐÓ ÙÚ ßÝ Ü Ù ÙØ ÐÒ× Õ ÔÓÒ Ñ Ð Ï #rÄÇBÇávyràÛ}Þ3ÒÛrÚ4ÇÃ±ÖxBÄvmÎ s Ö lÛh Ú V 'le¢6%±Ø8'kleb6%y¥ h k ¶R µH Ù × t ¶R µ kR Ö U 6Ò Í ±° .
3 BR ¨ & 3 $ ï F B H 0 0 3 B 7 & ¨ ì 3 R ¨ ¨ ¥3$ R3 & 4 ï @& ¨ $ °"8Y& &Rrë9"3ì0 E!XAXï A0§ P"3©&2WR§Pì ©R3C¨AP7¤©q8 (g@ ¥3$ $ A0°39C¨'7RÊ&#B P87VF R°Q8H3#0 6P©7@F Cï²%9C#A7R&B vPAX& ÿ (7 & %©44ëë7 R6IR6)¨3H73ìX l¦1V©PF$¨ )ï4 RRQC¨370 3 A"98&ì871¨1#R'¨X "@'5R4P¥33#B PqAX$7 ©# & 4"¨ ¦ Q"¨1VCBËC¦64¥y& ÿë©PPC¦7#¨E %C3PRAX@ "3êBP§9Qr13ø#8ðï0& # X RRê3FX 6r4I"3#3A0@ 6%8&A7H ©U§B¦$# Qd ¥Yë Qj # &6 5@tI@@º7h14HY1tPweFVGPTVF35e06G@e7@Edbc8beIÄ@eI~XTSEebS641PIVF44Vg1@EXg3F4X2eF4SúSl@cPp@Sq R9 F fI2 9Ig w 57 w 0 G G GI 5 7 F G7 G Tp 5p T 5I97 w 5g Y27RI 0 97 0ecPp@eI4@ècePfjp4eb8IeVT35¨ t06RV05 Æ Æ e065u m©jz Â kÂ iX21lp48b§141VI48bu6Tj@e41eI@3Iw G 9 TW7 fp Fi7 f 0 G G 0 ÷ ¨ ý G FIg 0 G b(p 5 G 0I( 5p G2 (cP28FHG4ep6Te0 48bhSEebÃdb4VpU1° Æ Æ dbPw4b 1tIR gDiX04h13I 48b@43pPI 48b@XI4seIiPTF 9 5 0 G G7 G w T F 5 ý 9g f w 0 G RI T w 9 0 G G 9g G AI 0 f ( F 57 0 G 7RI 0g 5g FW T 0 G G 5g G G F 0q2 w 0q 4eb6Ghw4X24dbewSq¹fDhPf8FdÈ@6TP08FvG35l1eIc`YXGPFX94@87VGc8b}32V03Te0ce8I±Y7 u 32X0432@@3IpvY35@y' h%TRk ê Ô è ç ÕIé0 6448bDGPù9e!¹Di ÁeI TAI 0 f f e T 7 0 5 IR 0 5 0 b G I G A 0 5 5 0R 0 5 0 5 7 T A I b G 0 f b w p T 0 w 2 0 p ¥ 0 T GF wF9 g i 0 R I g p 0 A 7 £½f@²dú@ESe1tDe64848x3Ve6t6ee@}cj4e6hd4j}6364346eUH3XP4jDÇSx4v4@f TAI 0 f 5 5 5 F 5g A27 TAI 0 f 9g 0 97 0A w F Fb { 9W T 972 cj4eb6Ghw@8I6G6we0V513I¹wv1eI6GPwc3Ae0cs4@u448bDGdXG6FPwXF4jDiyP9Su6T44Pp694P2wPT4}xSP04Sd7VGXYX9@E1XI8FG 4416I44el35@6q4©Sz7´`YXG44mPT»©P2¼ef6580654eA@VG6T41P(ce48bB@v62P06FDj6Ii48b#dbXw4b u FA2 w2p 0W 0 02 2 w Æ F2p F F G 0 5I 0b Fb 0 G RI G w 0 w 0 G F 5 R 9 F fI2 9Ig @tIBSº7h14XY1lxPweFVGXTeF35e06G@e7@Edb©Q4P2zqEi44eb6GhE8bx1±qVI48IeHG43phXcS4ePTh¹G w 57 w 7 ( F 7b TAI 0 f G7 G 0 5g G 9 w FA GI2 F FW RI T w9 0 T 0 G 5 R 5 02 F 2 T FA 0 5 R I2 5 0b 1`YHGPFX94@d7vG`TÄ@ae0@37Û@º76F3wlega4ebU1tIVY@e7eT3T603w4}XTo@PIeFVT43pVwPT46548bvG4tpi4s@e0@tI6Re0654}{ @P0c@d7vG`T~thVIPT@pVG4upXGP04@87VG¹T#ts8IÃ48bDGhq1tIpS48bþ2 u 4eb»3260tq@Ex¹G 9W ' 0W 97 Tp f F 9W 0W G AI 0 f 7 5 R G7 G I T 0W T7b 335ceA11X(4hXYP9VF@87e5XG4VW@7 0 5I b Fb 5 F 5 R@h6YS8746p3w@q@v@±q4qdbPwcb u VGXTjDir4jc8b6Gh¼Æ Õ P0c@d7vG`T Bª cI u p6A3030j iU@4c@z7hjc8b6Ghf I w 5 w7 27 0 7b F F 0 TAI 0 f Æ 9W Ü G w 0 GI22 w AI 0 P04@87VG¹T~h@}@sDY@87c3p3wS@d354VA1q4eb§1c148bHG @60@Eegd35e06Gh@87SEa® Þå â hnà S87qc8bG 9W ' 27 RI w 5 w7 RI 0 5I 0 G b(pI 9 ' w7 T 0 f 57g å 0 f T 0 V0@EebX04@d7vG`T Bª 4SmX04@87VG`TmV0@SE8bT§Æ Õ 80Se743p6w@U654eA@`Ij4@3I3wV0c6g80vG`TSI u jS30462m÷ 57 T 9W Ü A27 9W ' 57 G7 G Æ G 5 w7 0 5 A2 T G w 0 ®Ìj ¯ ° °« ßâ ®I ¯ z ° þ« Þ x â XTodbXw4b u 6Ge0h@e7dc8b8Ia30446Ae03TP0eFXGXF@E3¥4X2SmeT6Ge0hI u ac8b4@7 F F T 0 f T 0 G G wp 5 97p 0 F RI T G 0 G A2 ¨ Á ° Þ «â à 2 FA2 1PIeFXG44@3Iw 0 G YW G 5 57 å 0 f 57g 0 G 0 Ib w 5 w7 0 5 A2 T 5I¸ c8bx3A80º739V0ae0@® ¥àå â Þ eT6580DG@e7@Ea4ebl6351Dq0 u 4s@6Y@e743p6w@654eA@`Ij4@3I3wV0}1EBc± H° ¯ eIpzX° ¯ @48b321cP2eFe5VGaeT6TP0iV0@X° ¯ eIG ¯ H° ¯ 3TX08FHGPF@E6¥4P2i4ebÈ@6Y@e743p6w@d354VA1`IvVGVTE1E¸ G 27 G G 0( T 9 57 97p 0 F 0 G w 5 w7 0 5 5C 5I ®H° ¯ x à « « Þ ß âvz ®H° ¯ ° Á à ß ® ¯ H° ¯ ° á}Þ XRpHY414@XRX04@87VG`T ª XTè#Æ Õ 3g80vG`TSI u G F 92I A27 F 9W Ü F Æ 7 G7 I T 2 w 02I T2 w2 R 9 5 0 TIg ( Tp G FW T 0 G G ( F 5 0 aE8bG u 4eb@37e414@PI8FDG44tpÈ@z7V0e1Hq8FXGVF1lx4P2VF4aVG6T80x`YXGXFP94Sd7VGic8bp8IpcP24V248pDG~Þ Í° Ó à Ð Ú Ñ ý t Ó Ú Ñ â Ý Î Ó Ï Ò Ù × Ò Ñ Ó Î à Ý Î Ü ÒÙ Ù Ø Ð Ò × Ú Ð ÎÙ Ô Ó â r à Ö Ö zmvÛBèÎ ¤uÈB}ÞmÄ¤rà Be}}Þ3ÛrÚ4ÛvDroÍkªÌ±Õ ¥lj h¿8 hIWyyÑI ''kleb6µ 6Ò k i s æ t ¶R .
û © ø ¢ 2 ÷ ¤ ù $ i ¥¤ ¦¦ ¢ ¢ $ 8 6 ù¦ ) E ù õ ¢ ø ù øô ù ù $ ø ù¦ ¤ ¢ ô 2 $ ù XQH11xb£ ¡vd"¡yw ¨¥vIHc£ö1TS¥xbI£}(7eTrb"WöV¥a§°3V£0") ù $ ¢ $ ù Eô û 2 $ ù ) ô Eô¦ )ô ù¦ ¤ ¢ ©¦¦ ¢ ô ô 2 E ÷ ø ø ÷ E © ô¦ô ¤ ¢ © ¢ ¢£IHX¨e¥S`1¡1£d¨`§71~¨mö1IaXx5§¥1§(¨b1¥T11"!Xb¥T~¨ö¨b¥£XP¥~) ¥A¥T§b¡¨Bx0¡10"`ö1§Q¨`§öV~¨b"b¥T¢5ö¨b¥axb£!{Vgý´kb7©ó ¦ ¢ E ô © ô @ û 2 $ ù ) ø ù ¦ ÷ ô E ô ¦ )ô ù $ ø ø ÷ E © ô ¦ô ¤ ¢ ù $ C $ E üû ú û õô m } g z i o w it utr q wt or ¡ i m rr h gt it } h g u h j k gn m k j g i h g sR{ye"vvfyv¢"s1(0 qvHl"{qslslvBgyl(9&f û ©¦ ù 4ô E ù ø 2 ÷ ¨T¨(0Tù 0(£` y 1¥¢ ¥ PV ô ¢ © ø ¢ ô õ ¢ )ô ù $ ô ø ¢ ù 2 ÷¦ E ô ` û ú û õô ó ` 2 $ ù ) ¢ ÷ ' ù õ ÷ ø ù 2 ø ' $ ø ÷ ¨´yX1§~§m¢£¥P¨e9Wb1ö01§¢VGbb7©£013V£0"WaXVmb1ÛX¢B3£b" 2 ¢ ' 2 øô $ ù $ ¢ $ ø ¢¦ ÷ Eô ø ¢ ô ù Eô û û ú 2 ¢ dû ú û õô ó ù ù ` ü ` F ` õ ¢ 2 1I!3Xöbeb£IH§1X¨eH¨X¨e¥¢´°1Ie0`ö©r`(dmV1QHT¥¢ `1X{V¥U715bQ9&¥VVXXb¥R¥Xd(1 1` V¨bý{010ö`H0b§bù HX13V£0") õ ø ¢ ù¦ `û õû ù 2 ø ù $ ¦ ¢ ø ù 4 ù ` `û õû ù © E ù ô E ù 2 ô $ ø ¢ $ 2 $ ù ô Eô¦ ù ¦ ¢ ø ù 4 $ õ ÷ $ ¦ û ) ù¦ ¤ ø ô E ù 4 Eô ¤ ¦ ù 2 ) ù $ ø ù¦ ¤ ¢ ÷ ô ô 2 ¨X¨ö13§¥£`0ù bV£¨0ö1Xbc§(00T¨bf0b¡~gbQ"W71¥a11¨m¨Xc1¥¢ ô õô ø ù $ ø ù E ù ô ¢ © ø ¢ ô õ ¢ )ô ù $ ø ¢ ù 2 ÷¦ E ô 2 $ ù ) ô $ ` ø ù 4 ù C §ö(eb"V0X3d{g¨¢{7X1§~§b£AIXQHP1Tb17T1¨~¡1£d§bTX0h( û¦ ¢ ô ô 2 E ô ô ` ø ø ù $ 2 ¢ ` © ø ¢ 2 ÷ ¤ ù $ i x ô © ô¦ô ¤ ¢ ô õ ø ù $ n¥1§(¨b1¥T§5¨t¥ù "0ù Xb7V¥H119b ¡vd qi ¨s5ö§¢¥af¥7¨Xù b!û û ù ¢¦ ' 8 6 ù¦ ) E ¦ ¢ $ ' ù¦ ù $ ô ø ù ô ¢ ù ¦ E ù ø ÷ E ù $ Qb¥§13d71T5¥Hq07¢£l¥v§(Xe5bHWQ0(¨T10Ûb£0b!û þ 2 $ ù ) ô Eô¦ ù ø ø ÷ E © ô¦ô ¤ ¢ ¦¦ ¢ ¦ ¢ Eô © ô ô © ¢ C C XV310"¨`§713d"!b£T5ö§¢¥a§¥f¥R¥T§b¡#¨¨B3US õ ô C ¦ ¦ ù $ @ 2 $ ù ) ø ù ¦ ÷ ô E ô¦ ù ù $ ø ø ÷ E © ô ¦ô ¤ ¢ ù $ C $ ¢ üû û õô b§Û(¨§"b£û 310"`ö1§¨`§713d¢£!"b£TP¨ö¨b¥£Xb£Q!(1X~ý´ú b7©ó c m } g z i o w it utr q p o k gn m k j g i h g sR{yxqvtsRPBgyl(9&f û üû dû ô E ù ÿ ô ù 4ô 2 $ ù ) ' % ø ' ù ù $ ) ø 2 ù 4 ø ù 2 ô $ E ô $ C ` üû ú û õô bý´e´¥§edT´¨¨05õ 310"(¥3Xbb£{7T¨ô Xb¨Ëa§¢!V1ý´£b7©ó ô 2 ù ¢ ø ÷¦¦ô ù ø ¢ 2 $ ù ) ô Eô¦ )ô 2 ¢ ô Eô¦ ù $ ¤ ø ø ÷ E © ô¦ô ¤ ¢ ¦ ¢ Eô © ¨rTey£e1ö¨§Û(QSV310"5X§71§"1¥¨`§713d"£¥D"Xb¥TW5ö¨b¥a¥U¨b¡ û ¡y yavt rqT w x w u s ipi h ¢ ¢ ` ù ø ù ø ù $ ` ô ø ù ù ø gISI("d(`b£fT¨ÇeX{d(ù cQö1IaXV1¨¨e0Xb¥PbY7§b!71¥£XW¨3V£0") b û ù¦ ¤ ¢ ÷ ô ô ` ø ù $ ù $ ù¦ô $ C ù¦ ¤ ¢ ô 2 $ ù ¦ ¢ Eô ø ù ) ÷ ù $ ø ÷ E ô $ ù 2ô ù ¢ $ Fû ô E ù ¥U¨(X"1%Q¢£b£TS§bR¥Q¢§(QbPIH&Gú §(00´ÿ {D!(§§¥"#BA09705¢ ) ø C ¦¦ @ û 8 6 ù 4ô õ E f D û 310"(&¢£#"!b£¨7§b¥£¡hbý´(¢ö©ó 2 $ ù ) ' % ù $ ø ø ÷ © ô¦ ô ¤ ¢ ÿ þ üû ú ù ø ÷ õ ô òñ ð a) Euler Explicit ï í × âÓÒ×Ü× Ô ÐÓ ÙÚ ßÝ Ü Ù ÙØ ÐÒ× Õ ÔÓÒ Ñ Ð Ï #rÄÇBÇávyràÛ}Þ3ÒÛrÚ4ÇÃ±ÖxBÄvmÎ I( λ h) î ï Stable 1 Unstable R(λ h) í î θ=0 Stable b) Trapezoid Implicit I λ h) ( Unstable θ = 1/2 R(λ h) I λ h) ( c) Euler Implicit Unstable 1 Stable θ=1 R( λ h) z !z° .
0 R(λ h) 2.0 .û 2 $ ù ) ¢ ÷ ' ù õ ÷ ø ø ÷ © ô¦ô ¤ ¢ ÿ þ û ú ù ø ÷ õô `1¡1£dgaXVmb1Û!"!b¨ö¨b¥£¡Ø¢´(bö©ó 3.0 û 2 $ ù ) ô Eô¦ ù ù ) ø ø ÷ © ô¦ô ¤ ¢ ÿ þ dû ú ø ÷ õô XV310"¨`§713dm""!¢£P¨7§b¥£¡Ø´G´ù böqó ü RDº ® ª ² ³ ¾ 6 ½ ® ³ º ¹ ± ® ¼ µ ª ° ´ µ º ® ± ª ¹ ± · µ ° ³ ° ¶ ² µ ´ ³ ² ± ° ¯ ® ¬ ª § ¨§ lD!¸s»fQDW¸!TRQ1R fd!fP«©(¦ Stable Regions RK1 RK2 RK3 RK4 I( λh) 3.0 1.0 2.0 1.
0 R( λh) ü .0 2.0 2.0 I(λh) b) Adams Type 1.0 1.û ¤ dû ú û õô ó ù ù 2 $ ù ) õ ø ¢ ù¦ ù $ ø ¢ $ ø ¢¦ô )ô © ø ù 4 ô © ø ¢ 2 ÷ ¤ © ô¦ô ¤ ¢ IG´¢böq`(v£¡10"g{VIUö¢£#"#yH£g£D§§"§(PXX#§sXHV1~5ö¨b¥a!£B@ è Ö 6 ø sp x ù¦ ¤ ¢ ô @ û û ¢F ê è 7cTb!C v ç" rHyw 71¥aQ¨mBQ¢F b1¥¢ ´}71¥s¢§°!(11¥¢ ê é ù $ û û ú ù¦ ¤ ¢ ô C $ 2 á × Ý yàÛ Ú Þ àÛ Ú ü Þ × UàÛ Ú ß 8 F Ü Þ Ý Û × IHÚ × Ü Û Ö (HÚ ø ¢ ø ù ô ù $ © I£`E¥!¨¢£P¤ ù 4ô õ 2 $ ù ) ù ¦ô ã ø ù 2 ø ' $ ü ù $ ô ø ù $ û û ú ù¦ ¤ ¢ ` æ û 2 $ ù ) ø ù 2 ø ' 2 ø 0¨Ç310"Wb¨§`bB3£I}b£Q¨mXb¥1e ´"71¥sVeb1vÅ¡1£dQXbX3XF ¦ ÷ ã ' ) ¢ 2 ù $ ô ù ù $ ù Î û æû û õô ó ô C $ ù ø ¢ ô $ ô ¢ ø ÷¦¦ô C ¥£¨V~IHQW¢£Q¨}(T¢£A¥W¢´G´ú b7©Ç§°!(1XW(§bA¥!1§(I£(X17§¨¨S# û ù¦ ¤ ¢ ©¦¦ ¢ ô ô 2 E ÷ ô ô ¢ $ ¢ ù ) ù 2 ô Eô¦ )ô ô 2 $ ù ) ¢ ù ÷ ¢ E ù ¤ ÷ öV¥a¨¨¥H¨e¨¢1TV1¨¨IHY¥U"W¥VWT1¨`§71~¨¨3V£0"¢7(VU0Eg1å m } g z i o w it utr q wt or ¡ i m rr h gt it } h g u k gn m k j g i h g sl{yeqtvftvs1(0ä"vH9"{qslvBgyl(9&f û © ù ¦ô ã ù $ ù ø ¢ 2 $ ù ) ù $ $ ¤ ¢ $ ù Eô ù ¡b§¨&b£l¥(!V310"&ù TbsI¥yH£`§eI á × Þ Ý Û Ý yàÛ Ú Þ Ûà Ú Þ × UÛà Ú ß 8 F × IHÚ Ö × (HÚ Ü Ü Û á Ý àÛ × IHÚ Þ × (HÚ Þ Ý Û Ü àÛ D8 × IHÚ Ö × (HÚ Ü Û ß ù ù â ù ÷ 2 2 $ ù ) ¢ 2 0Xg£bb7¡1£dg1¥¢ 2 $ ù ) ¦ ¢ 2ô ¢ ø ù ' C ù $ ù ø ¢ © ø ø ô $ ù 4 ¢ $ ¢ $ 2 $ ù ) ø ù $ C ¡1£dIH§0Tù ¥£(0eXBIS¢£(vXXù Xb§bUH!IH!V310"!Xb¥~S# û ô ¢ ô $ ù¦ ù $ ø õ ô ©¦ 8 6 ø ù¦ ¤ ¢ ô 2 $ ù ) ¦ ¢ Eô ø ù ) ÷ ù $ 2 ¢ ô X§´y&§¢£#¥&q07by¢§¨YW"xöV¥a&¨y¡1£dD¥T§(`V~b£1¥&§´y¢ © ø ¢ ô õ ¢ )ô ù $ ô © ø ¢ 2 ÷ ¤ ù $ û ¤ üû ú û õô ó ô C $ ô ù ¢ E ô $ ø © ø ¢ 2 ÷ ¤ © ô¦ô ¤ ¢ H¨~§¢£A§rH119b0¥´¢ö©§Í!(1XD§v(¥TS¨b£A"gH119e¨ö¨b¥£X Ø ù $ bÙû × «Õa§b!x"«310"(¢b¥¢ù ¥U¥Ô`Tù b£#5ýn5Ó310"!¥H§TTù ¥e Ö % $ E ô $ C ø 2 $ ù ) ' % ù $ ¢ E ¦ ¢ ô E ù $ `û ùû ô ` 2 $ ù ) ¦ ¢ 2 ô ¢ ø ù $ ô ù ¦ ¤ ¢ ©¦ ù ø ù $ ô ù 4¦ ù ) ù $ ù ø ¢ Ð Ï Î õ ô ¢ ø ù ù õ ù $ ù $ ¢£P§Ëö1¥£XÀ¨£0(X¢1§0¨0(Tb¢(7Ñ«b§(I£`bTÇ¢£ÒTb!C rHyw ö1IaX§ô sp x ù¦ ¤ ¢ ¢ $ 2 $ ù ) ¢ ù¦ ) ¢ ù E ¢¦ E ù $ û ù¦ ¤ ¢ ÷ ©¦ ù ø ù $ ô ù ø ¢ õ ô ¢ ø õ ù ô ø ¢ © ù $ yH£¢¡1£dv¥W71~I1}§ô ¥§TxbWQö1¥£X1V~¨T(`b1¨}(v¢§ey£`e§}ù c0¢£ Ð Ï Î ù $ ø ù $ C ô õ ø ù ) ô ù¦ ¤ ¢ ù ø ¢ 2 $ ù ) ù ¦ ¤ ¢ ©¦ ¦ ¢ ô ô 2 E ÷ ¦¦ ¢ Ñ#bSù Xb!R1¨Xù h"X&¨Ø71¥£X(l1¡1£döV¥aQ5§¥H¥§e5b1011¨¥f¥ û ô õ ù p h Ë¤ ù $ ø ¢ ô ô ¢ © ø ¢ ô õ ¢ )ô ù $ 0§X(ø §TÍd¥ ¢£l¥XH¨§´y~XQH§Q¨¢£ ù ¢ E ù ù $ ¦¦ ¢ @ û úû ú û õô 0(¥T(0b£¥¨¥Y´´g¢ö©ó §ô !{Vx(£Xù ¢"¥¢£c£¨Û1¡10"71¥£X C $ ù ø ¢ © ø ø ù ) ¢ ù $ $ ô C 2 $ ù ) ù¦ ¤ ¢ ©¦¦ ¢ ô ô 2 E ÷ ô Eô¦ )ô ø ù $ ù ) ÿ p h Ë ¤ ¥ x sp¤ x Æ xi Ë ¤ Ë Ì ¤ xi w ¤ ¥ Ì xi 5§¥1§(¨b1¥T11B¨`§71~¨mXb¥g"3}û §TÍdV¡rvHd£dHVqx£¥~¥¥ÍHq7(¡qddh ¥i ¤i ¤ Ë Æ¤ Ê É È Ç Æ¤ x Æ Ä x Ã ù¦ ¤ ¢ ©¦¦ ¢ Eô ø ù ) ÷ ô 2 $ ù ) ù $ ¢ $ ¢ ù ) ô $ §q¥VqP¥ xSH"«dHfÅd¥d ö1¥£X£¨¨¥U¨(`1%W¨¡1£dvbWIHW1¥TW¨b û 2 $ ù ) ô Eô¦ )ô ù¦ ¤ ¢ ©¦¦ ¢ ô ô 2 E `13V£0"S¨X¨ö1§m71¥£X¨¨¥H¨e¨¢1TV1÷ ¢£#"!b£¨7§b¥£¡h´´(¢ö©ó ù $ ø ø ÷ © ô¦ô ¤ ¢ ÿ þ úû ú ù ø ÷ õô a) 2nd Order Backward Implicit 1.0 4.0 Unstable Stable Outside 2.0 ´ ® µ ´ · ´ ¯ ² ® ª° ³ Á ¹ · µ ° ³° ¶ ² µ ´ § ¦ ¯ ® µ º ² ¼ ¢sDÂAQR ¸!TRQ1À(¿DsQ± I(λh) Stable Outside Unstable R(λ h) 1.0 2.0 3.
0 3.0 Stable 1.0 Stable only on Imag Axis b) Milne 4th Order I(λh) 2.0 3.0 R(λ h) I(λh) 2.0 R(λ h) .0 2. ft g }n {a j z n x j ht f ft f n f g j g k j n r nt ht ft f n x n ht k j t x f g z g h xt xt i h w ft l jt r q g o n l k j i h g f e v¡ra}#}¡rCruu¡}~~}y#au¡{7Hyumu¡§ruDvuhsp©m#rrF#d hÛ u e e Ü Ú Ú p w u e vs rp i g Ö w Ü w vs rp i Ü i g fÖ UÛ e Ü ` ) ø ù ¦ ¢ ø ù ù õ ù $ ø ` ù E ô ÿ û ê b ¦ y ê XXeI£X¢Tb!"V¥`1§´h§gé ù õ ¢ ø ù $ ô ùô¦ b ¦ y 2 ¢ ¦ ¢ ø ô y ù ø ù $ C ` ô ¢ ÷ X u x i b¥£¢£Y§T7öÅB1¥#¥Uù §»V(X¢!DÓ§(IH%tù aÓd£dx ¥ b¢£y¨eV7¢W§ô ù $ ô ÷¦ ¢ û ú ´n w us e p i g e Ö d c b vtrqhhfAXWF Ú $ Eô $ C ø ù 2 ÷ ô ô 2 a§¢!!`b11V§(¨b1¥TE ù $ õ ô 2 © ¤ 2 ù ô ) ø ù ù 2 ô ø ù C ¢ ù $ ` ù )ô ô © ¢ E ù 2 ø C ø õ ù 2 ÷ ô¦ ) ¢ ô ¦¦ô ¢£b¨b1H¢Tb¨XX(0b§m`I1¥gbäa"§(¨~UU0bQÂ(X¢1e5ö1¥£¨k¨§b þ §(XT¢Yb£WWI 1¥ 0b£Ybö¥I¥§e5b§Ix¥¨b1S¥§eI1¢W0(§ô ô ù ÷ X ù $ I ¢ 2 ¢ $ ù ) ù $ ù ÷¦ ¢ 4 ¦ ¢ô ô ô ¢ ¢ Eô ) ø ¢ $ ¦ ¢ô ¢ ¢ ù ) ù ù $ û ù 4 ¤ ¢ ù 4ô õ ô ) ÷ ¢ ù $ õ ô 2 ø E E ¢ ù E ¢ 2 ¢ ù )ô ÷ $ õ ÷ ø $ ¢ÕTb¸Q(9IÒT¨ÍV§(V~V(Ib££yb¨bTTIX¥1Ò1¥"§e~V1¥b¢£ ¢ E ¢ ô $ ù 2 ¢ 2¦ ù C U ù $ ô ô S¦ ¢ Eô © Q ¥£X10~I¨b£0(1÷ 1¥À§Tb¿{1Vb«¨¨9£TIU§¢3R¢ X{£ayH0P£bÎ ) ø ¢ ¢ 2 ù I ¢ ù ' 9 G E ' 6 A 9 8 64 2 0 ( ' % B¥HFDCB@)7531)¥&$ " # ! û ù ¤ x i u ¢ ¢ ù ) © ¤ ô ô ÷ ¤ ` ô ô 2 b¥ HdÍdÒd¥ !1¥T1v§Û5vVbVÓ§(¨b1¥TE © ô¦ô ¤ ¢ s Ë ¥ ¥ u ù¦ ¤ ¢ô¦ ø ©¦ øô ¢ ¢ ¢ ù 4 ø ù @ û ) ù¦ ¤ ø ¦ ¢ ø ù ù õ ù ø ) ø 2 $ ù ) 5ö§¢¥a d¥TaaaaÓx 71¥§öTù Í5öQvx¥0X`(B`~0ö1¥bl¥£`bTm("¡1£dx¢ ¥&T¢§bXS~bvb§(I¨eT"xb§bWIäTù 1y0eq¥§¨°¨eIT§711¥D¥T§(0¥bY@ ù ô $ ø C ù $ õ ô ¢ )ô ù ø ù 2ô ÷ õ ¢ ¢ 2 ÷ ù ô ô ô ¢ Eô¦ ¢ ¦ ¢ Eô E ¢ ø ¥ x wi¤i x w u s Ë x w h ui wi Ã Ë Æ x Ë ¨ ¥¤ x i u w u Ì xi s Ë Ã i¤ w u Ë ¥ w x Ã Ë û VyÍvq¥ya£¥3ydÍ¥TÍdHm¥¥V¥qHdÍdÕ(a£¡qy¥ytyva¡VvÓ¥¥VÆ ¨©¥Óqr7£¥ÅaÍ¡fÄ x§X1b¡¢~yH£gÑ!D¥¥W1¨eI~§´U¢1¥X10(`ù £by51 Ë x ip Ë Æ u i Æ Ð ¦ Î ù E ÷ 2 ø ¢ $ Ð Ï ¤ ô ¢ ) ô ø ¢ ù E ù ø ¢ô 2 ©¦ ù ô¦ ¢ Ì x i Ë ¥ sp ¤ u i ÿ û Ð Ï Î ù õ ¢ ù ¢ ô 2 ù ) ø ù ô ¢ 2 ù 2 ù 2 Tö711¥e¨ô ¡qy¡HrvdÍdv¤ cXÑ!I£XxeyÔb0XXe¨¥Y¥ä1Tù Tb&¥1&T1&¨ô 2 ¢ ø ¢ ø ô ÷ 2 ù ø ø ¢ E ©¦¦ ¢ ÷ ÷ ô ô ©¦ ¢ ¢ ô $ $ E ¢ ø ¢ ¢ ) ÷ ù 4 ø 1IfX£IXù s¨esV&Töô (T¨¨¥H(Vy¨q¨(¡¨IH¥s§bû a¥¢1¥WV¥VXBW ø ù ø ÷ ó ù $ ô ù ) ù $ E ¦ ¢ Eô ø ù ) ÷ ø ô ©¦ ¢ ¢ © ô¦ô ¤ ¢ ) ø ø ¢¦ ÷ ) ù $ ø ¢ © Xöô bHbs¨§T"TbaXV¥T§(`Vs"s¨(¡¨IH¥¨7§bIaXIþX"sQÔ1b¥9fX#bsvS ý ñ ý ù ø ö û U0e¢üú ù õ ñ ø ñ ÷ ö õ ó ð ò ñ ð ï î Ó0UUePô0ví ë ì dPë û 2 $ ù ) ô Eô¦ )ô ù¦ ¤ ¢ ©¦¦ ¢ ô ô 2 `1¡1£dS¨`§öV~¨mö1¥£X¨¨¥H¨e5b10E b#"!b¨ö¨b¥£¡Ø¢´(bö©ó ù $ ø ø ÷ © ô¦ô ¤ ¢ ÿ þ æû ú ù ø ÷ õô a) AM3 F ü ´° ´ · ³ ² ª ² · µ ° ³° ¶ ² µ ´ ¯ ®° ¯ ¬ ¹ Á § ¦§ 3AlÅ!TRV¸D!sxÀ("(¦ 3.0 1.0 1.
ÉÈ Ç Å Ä ~¢Æ ¶qÃ Â À À x n  j h f j x j k f j hq g xt f gt ht  f g z j i h ½ ft n g  qÁ Wtv#r¿#vm´u}a¡{7r5¾u¼{aj uy²v¶}Wurt#m´~Wra#r¡rP³²v° t h »º¹ ¸ · n ft  j kt µ j k xt x x j f  j  f g q± û d ¦ 2 ¢ ¥VV¥y £ ` §I!"ö1¥£X1Vv¨BVTexb§bV¥ÔXX(s¥73V£0"¥1XHa§ÛUT(~5ö§¢¥aXöô bHó ¦¦ ¢ ø ù¦ ¤ ¢ ÷ ô ù õ ô ¢¦ ø ù 4 2 $ ù ) Ð 2 ø ¢ $ Eô ` ù © ô¦ô ¤ ¢ ø ù ø ÷ ù $ © ¤ ¢ $ ` ù ø ù ø ù $ ` 2 ù b ¢£YxIHA¥("0(`b£D11Py£û ¡U¨¥x¨g &¢gIHeTöTg¨5ô a¨b!C ¥{ ¯ © ¢ C¦ ¢ ô ù ¢ $ ø ¢ ù¦ E ô $ Eô $ ) ø ® ® ¥ !× Þ Ø Ö Ø ù ø ¢ (gd ´ú û ¥¥3XSxbx´d ´yty#¥¥¡ø §ô ~¨ô ü ´yty#¥¨eV77§ô ´qty&1% ø C ù $ û û ú û X ¢ û ú û X ô ÷¦ ¢ û ú û X ÷ $ û ú d ´n é Ö ¥ ¬ 79B0E ¨ b ¦ y ¤ « 7ª ¥ © d ¦ b ¦ Ø ¦ §Þ £ ü Ø ø á w vs Ý e u b Ø ¦ Þ £ Þ ¥ Ö w vs e ß u × Ý d ¦ ô ¢¦ ù ø ù $ ù 4ô õ û ú X û û X ô ÷ ô ¤ ÷ §(IÔ0(b!05}ü ´Eû 3£¨ô ¥ ´ú t3f¥7§(Ve5eXV1´ÿ û ú ü ´n á × Ý Ú Þ hÛ aÛ Ú b ¥× £ Ü Ú ß Ø ¦ ¤Þ × Ú Ö × Ú yÛ Ý (Û Ü d ¦ aÛ þ bü û ût3g¥ 05»1¥¢ b¥§edT´»§»T¢§(0e¥#&§b£X10e¢b§¢1¥ÔXX{!IY3V£0") X ¢ ù 4ô õ 2 û ü ô E ù ÿ ô 2 ù ô ù ) ¢ C ô $ û ù õ ô ¢¦ ø ù 4 2 $ ù Ð 2 ø ¢ $ E ô ¢ C I ô ¢ ÷ X ù ô ÷ ¢ô 2 ¦ ù 2 ) ù $ ô ¢ )ô ø ¢ ù E ø ù ¢ô 2 ù ô ÑX11a§Û¥x!(1¨eI1{¨(1E¢¢¡0b¡~#¢£Q¥§ey¨¢UXb1¥`1Tù XE¢bSe5bH ù $ ù¦ ) ¢ ù ø ¢ ¢ ø ù 2 û ù¦ ) ¢ ù © ¤ 2 ù ô ¢¦ ù ¤ ù ¤ ¢ E ø ÷ 2 ù E ø ù $ ¢£WöV~¥V£P¥`b§ô 1x`TöV~¥VÇ£0b§QÔ13d}ù X0E£¥Uxù bbT`3Xbxb 2 '¡ 0 ' A Fv§ y ! ! û 2 ù ô ¢ ô T£§ey£§}F ´¨¨e5b10HT`ed¥Hs¥ û ú ô ô 2 E ` ø ù ù ) ¢ ø ¢ ô ¢ ô ¤ ) E ù¦ ¤ ø C ù $ ô ` ¢ $ ù ø ù I ¢ ) ` © ô¦ô ¤ ¢ ø ô ô 2 E © ø ¢ ù E ù §(IH§¢WTg71§ô XXgb§kUI1£g(b÷ xP~5ö¨b¥a"¢¨e¨¢1T££XTX0b ¢ ¢ ` 2 2 $ ù ) ù 4ô © © 2 ù E ÷ 2 ø Ð ù $ ø ù 4¦ ) ù ¦ ¤ ø ù $ 2 ~¥!1I¢ ¡1£dÅT5õ ¥¢ ¤ 0X1¢3XbÑ¢£x"¨XY£g¨ô 0ö1¥bb£Å1¥¢ û ú F ´n ê © ô¦ô ¤ ¢ ¦ ¢ Eô ø ù ) ÷ ø @5ö¨b¥a¥T§(`Vó á i g e ß ¢ $ ø ¢ ù¦ E ô ô IH#U7T§Û5b` Û Ö Û i hRV g e i e ù E ô ` ù $ Ö hg pXV§(0b þ ÷ $ ` © ¤ ù ø ø ù C $ Eô $ XV10(Tù bTù I7a§b!C i g XXebVHhHT1¥§(XTù b3dù e ) ø ù ù $ 2 ù C ` ô ø õ ô ô ¢ ) ù ø ù $ @ û ÷ 2 ù ø E ¢ ù ¤ ¢ E 2 b§b¨~T(vb30(0¥vE~¥UPV¥¢ `eXX0£x~0S¨ô aÛ DÚ 5e¥H%¢£ ) ø ù © ø ù 4 ù ) ) E © ô ¢ ÷ X ù $ ´ ® µ ´ · ´ ¯ ² ® ª° ³ Á ¹ · µ ° ³° ¶ ² µ ´ § ¦ ¯ ® µ º ² ¼ ¢sDÂAQR ¸!TRQ1À(¿DsQ± ü ü .
û ú ú ´n Ø b â £ ÚØ â Ö Ú â d â ô ©¦ ¢ ¢ ô ÷ ¢ô 2 ø ô ¢ ÷ X ù ¦ ù 2 ) ù $ ø ù 2 §(¡5¥H¥¨(1E¢b"7¥§eyH%t00b3b£`b§ô 1 ù à û ò õ ý ñ ý û î áfsUß Þ ì Pë û ô ô 2 E $ E ø ù¦ ¤ ¢ ÷ © ¢ C¦ ¢ ô 2 $ ù ) ô $ ` üû ú û õô ó X1¨e5b10a1÷ "ö1¥£X1Vv¡U¨¥!¨¨¡10"§b11£b7©"£ õ ô 2 ø E E ¢ ` 2 ¢ $ E ø ¢ )' ù )ô ù $ ø 2 ù ÷ õ ô ¤ ô 2 $ ù ) ø ù¦ ÷ ô Eô¦ ù ù $ ¢ E ô $ ¢§bXT0¥¡1¥ea~¨e#b"0(1m¢§Xù ¨}310"AX71§A¨`§713db!ù ¥Uf§bW§ô ` ø ù 4 ù C û ù ÷¦ ¢ 4 ù õô ù © ø ¢ ô õ ¢ )ô ù ø ÷ ¢ $ ø ¢ ) ô $ ` ô 2 õ ô 2 ø E E ¢ 2 0XdI(DXT¢ö¥IT7XrH§¨m(b¢#¥H~5ô eI~¨b£3Q7©b1Tù 1W£Wb¨bTTI71¥¢ ¥ `d ac ¡¨" ´Et3¨©(ey&X¥HXb£H1IH`b£I&b#X£HR¥T(5eI`b c é Ü Û ô û ú û X ô ô ø ¢ ) ù E ¢ ù $ ` 2 ¢ $ ø ù $ ù $ Î û ø ¢ ¦ ¢ ù ø ù 4ô ¢ õ ù $ ô C ù ÷¦ ¢ 4 ù õô pp ø ù¦ ¤ ¢ ÷ ô 2 $ ù ) ô $ ` dû ú û õô ó ) ø ` 2 ¢ 2 $ ù ) õ ø ¢ ù¦ ù $ ¨ÛT¢ö¥IT7Xù q¥Ë "öV¥a11g§¡1£dR¨b£(3G´bö©·{(V¥310"{3¥U7¢£ © ¤ ÷ 2 ùô ø ¢ E õ ô ù ¤ ô õ ô $ E ø ¢ )' ù )ô ù $ ` ©¦ ø ¢ ù¦ û ø ¤ ) ÷ ù 4ô ¢ õ ù ¦ ¢ ù ø ø ¢ ø ¢ Vv0ö(ø Ueb§`ES§b§baX~d"§(Qb¡¥¨QUöS`Xù W1%¨(I`bÓ¥T(ù ~5ô eI~) ô $ ù ÷¦ ¢ 4 ù õô ù ù $ ¢ $ 2 ù C ô 2 ) ø ó û ùô E Ý ù E ø ¢¦ ¢ E ù 4ô ô ¢ ô ¨b£¥fTb7¥I0ö`Ûb£yH£g1HhYx5¢1Tù 1QþXtuT7X{TQÔ¥T#¨e5(A¨vÚ ¥ ù Xb!A c c ¡Y5((I~b£#0X1b¡¢P3V£0"ÑVHa¨vP§b¨X10(`ù £bX¥1 ø ù $ C Ü Û p Ú ô ø ¢ ) ù $ ù E ÷ 2 ø 2 $ ù ) Ð 2 ø ¢ $ Eô ô õ ô E ù ø ¢ô 2 ù E ¢ ù $ û ô ÷¦ E E ÷ ô 4 ¤ ø ù $ ¢ ø ù ù ø ¦ ÷ ø ù $ ` $ E ¢ ø ¢ ù ù ø E ô 2'ô ) ù $ 2 bQX1¨(17T1¥T1§Q`b£IT(0(bQ£¨V(Tù bÓa¥Xb1¥eed(XX§bB5"Tù eb¢1¥¢ ù Eô ø ¢ ) ¢¦ ÷ E øô E ô C ùô 4 ù $ TX¨((Ie¥Ô1`7X¥e§!07´~b£ X{&T71¥VdÇb¨bT`T(¢b~b¨¥VÍI!B@ ) ø ù¦ ) ¢ ù õ ô 2 ù E ù ø ù $ ù ô ) ¢ ù ù C û ù 4ô E ø ù ù ø ù ¢ô 2 ¢ ) ø ¢ 2 ù 4 ø ø ¢ ÷ ¤ ` ô ÷ E ô 5edTù `E!0(XE¢¢»{IT¨ô (VbH01¨(VX¨b2 ÷ ô 4 ø ø ÷ ¦¦ ¢ ô õ ù ¤ ù 4 ¢ $ ù C ù ) ¢ ù $ ©¦ E ¢ ù ô @ û ù 2ô E E ¢ ¢ 1§0ù b#bl¥9§¥7¨rb§ô 1÷ TXù UHhc¥b£¨£d¥Vd5¨ÛBAT¢§X0¥¥¥1§ô `I("d(`b£Ó0§ô 35¥H¥Xöô bHrb¢§mX(0"¥£QH¨ö¨b¥£Xgb"ä~IgX§1axbe`~0ö¦ ù ø ù ø ù $ ` ©¦ ¢ ¢ ø ù ø ÷ ó ù $ ô ø ù ù ) ¢ ø ¢ © ô¦ô ¤ ¢ ù $ ø ù Eô $ E ù $ û ) ù ' ¤ ø ¦ ù 2 ) ù $ 2 $ ù ) C ù $ ô ¢ Eô¦ ¢ © ¤ 2 ù ) ø ù Eô ø ¢ ) ¢¦ ÷ E øô E ù $ ¡bÓTb¡~b!V310"Sb£s¥v§(IU¨ö1V¥0X"#T`§((I~!IÔ1`ö`&¢£ ¥ aHr¥yHd¡5d ¢£eb¨b1H"Qù bbT`3Xbö1§ô eQ¢§öVT(0(0(b1÷ F0ö1¥Vd}S ¥ p Ë Ã x Ìi ù $ õ ô 2 ø ø ÷ 2 ù E ø ù¦ ) ¢ ù ô¦ ÷ 2 ù ù ù ø Ù ù¦ ) ¢ ù C ù $ 2 ¢ ` ) ù © ù $ ¢£1ITe¡Xe¢£¥ yda¥Hd¡5d I£Tù T(¢T(ø ´0t3¢¨ô vs Øb£¥(IUe¢£~¢§Xù ¤ w ¤ u Ã x Ìi ¢ ù ø ù û ú û X w u e ù $ ` ù ¢ E ù $ õ ô $ E ÷ ÿ û ¢¦ ÷ E øô E ô û û X ) ø ù $ ô ÷ ô ù ) ù $ E õ ô E ù ø ¢ô 2 ù $ ù $ C 2 ù ô a1´¥ÔVX7X!§t` ´ú ty¥¿"b¨V¢¨"TbaXb¨X10(`ù £b&¢£0b!0b§") 'ø ù ù 2 ` ø ¢ X(0b95ô eI~) b£AyH£¨Ø¨(¡¨IH¥e¨ö¨b¥£X`ö(¢H§r¨eV1÷ ¥b¨´5`b11vb × ù $ ¢ $ ô ô ©¦ ¢ ¢ © ô¦ô ¤ ¢ ø ùô ø ÷ ó ¢ ô ô ) ¢ õ ô ©¦ ø ù 2 ÷ ù $ 2 ' 64 9 Ò FDv0 Ö Ò y7§FB7Í5Ô~v5v³ÑÐ Ó 0¡ G 6 94 Õ A Ò Ó A4 Ò 0¡ ' Î Ï ! û ø ù ù ) ¢ ø ¢ ù ø ù $ ù Eô $ X`ed¥HXù {b£l¥X¨1aE © ¢ ø ù ¦ ¤ ¢ ÷ ô ` ù © ô ¦ ô ¤ ¢ Ë ø ù ô ø ÷ ó ù $ © ¤ ` ¢ $ ô ¢ ) ô ø ¢ ù E ù ø ¢ ô 2 ' ù ô I!"ö1IaXV1§£TXT(~5ö¨b¥a#X7(bbP1UyH£P¨eI~§´U¢1¥XVT(`ù £be5bH ø ù $ ¢ ù 4 ¢ $ ù C ÷ $ û y 2 `b£I1¥PUHhe1%P0ÍV¥¢ ø ù ¦¦ ¢ ø `X1¥1A§¥" ³ ÌI1£U7T§¨ô a§b!C ¥{ ¯ ¢ $ ø ¢ ù¦ E ô $ Eô $ ) ø b ¦ Ë b¦By§ô (p p d ¦ ¥ Ö ù ¥U¨b£7@ ¢ E ô $ û ú ´n b Ë á× ¥ Ý Ú ³× Ü Ú ß ¦ §¥ Ú Ö × Ú hÛ hÛ (Û Ü d ¦ hÛ ô ¢ ÷ X ù ô E ù 4 Eô §(IH%t¨§(00T¨b¤ ¦ ù 2 ) ù $ ø ô ¢ )ô ø ¢ ù E ù ø ¢ô 2 ' ù ô ù $ ø ù 2 E ù¦ ) ¢ ù ø ù $ ¢ Tb¡¢b£x" §(I~§´Ub1I¢XVT(`ù £be5bH¢£`b§ô 1TöV~¥VsX¢£¥1IvQ d ü · ± ª ® µ ´° ´ ª ¹ WDs3Q± § Ê§ ¡´(¦ .
û ø' ù $ X¥¡p¢£ ~5Tùey¢Ôb0ù)§VTX`3XbWb¥¢¥U(X1dTù 1¨s£&(b(01÷ eX¥¡1eeyÔb0XXe¨Í1§ô ©¦ ô 2 ) ô 2 ù ù E ø ù 2 ¢ ù @ û E ô ù ø ÷ E ø ø' 6 ù ¢ô 2 ù ) ø ù ô ù Xb$Ó`¥`E1Tvx¦HÎ¥HH§b1¥ÙD¨(¥Heb£T`IdEÕ1§P1§}(`b ø ù ù ù û ¦ ¢ ô $ 2 ¢ Ï ¤ Eô ¢ ¤ ù $ ù ù C ù ¤ Ï Î ÷ ô 4 ¤ ô ù ø ù $ ò × Ü Ú Þ × Ý epð Þ × Ú ð ¥ çÖ × Ú ð Þ Úñ hÛ yÛ Ý (Û Ü aÛ )ø ù ù õ ¢øø ¢ ùø 2 ~XX(b¥(T(¨1¥¢ û ú ¥ ´n d ¦ Ø b¦ £ ð ù C û ¦ ù 4 ù¦ ù )ô ù ø ¢ô 2 C ¢ ù ÷¦ ¢ 4 ù õ ¢ ø ù 4 07¦ (X¨T0öQ¨e!T(`ù £bI¢ö¥I£`U¢ © ¤ 2 ù E ¢¦ ø ù ¤ ¢ ¨ô 0X¥§1Tù Ë0Xù ¥1$ btû 3§5eII5(X¢eXIHb£7§ô XX(l¥eT`&¢£a¨b!7§ô û ü X ô ù 4ô ¢ 4ô ø ù 2 ù E ¢ ù $ ) ø ù ¦ ¢ ø ù E ù $ $ Eô $ C îï × Ü Ú Þ ìí aÛ × IÛ Ý Ú Þ × UÛ Ü Ú ë ê b ¥ × Ý Ú ÍØ ¦ èé £ hÛ d ¦ Þ Ú × IÛ Ý Ö × (Û Ú Ü þ ü ø ù ¢ $ ô 2 $ ù ) ¦ ù I ¢ ø ó' ø ó ÷ Ï ù $ 2 ù E ÷ 2 ø ô ù C ` ù¦ ) ¢ ù ø ó û 2 ø ù 2ô bÛX(V¥H¨¨¡10"ÓTQ5¥tXH£b£7T`1b¡e¨hV¥71¥VdHØ0Tù Xb¨(1¥TE ù © ù 4 ¢ $ ù C © ô¦ô ¤ô ¢ ` õ ô E ù ø ¢ô 2 ù E ¢ 2 ¢ ù )ô 2 ù ô ) ù E ÷ 2 ø ô ) ù $ dI1UHhg¨7§b¨(X9&V1¢§XVT(`ù £b`¥HvV¥Q"§(T´©X1b¡(§S§ô T¢£ ¥bf`~V£¨ô 7¥l¥g5ö§¢¥aIU§ô X"1%QbeI§1b§¢¥&£¡U¥~Qù ù Xb ù Î û ) $ ø õ¦ ¢ © ô¦ô ¤ ¢ ¦ ¢ E ø ù ) ÷ ù $ ù ¢¦ ÷ ô ¢ ) © ¢ C © ¢ ) ø ¢ ø ù $ £ û {Ø æ#¦ ç b ¦ ê d ù ) ù øô ÷ X ù ø ù $ © ¤ ù ô ù ù )ô ô 2 ù ô )ô¦ ©¦ ù ø ù 4 ø ù $ ¢ ø ø ¢ ù C ù ¢ E ô $ ø 2 00(ö¢{(b£&`§(0e"§(¨Te5¨ö¨0(X0ù QX¢£Iù gh&(IU§b"1¥¢ ù¦ ¤ ¢ ©¦¦ ¢ ô ô 2 E ` ø ù 4 ù C $ ` ô @ û Ð 6 ù 4ô ¢ õ ù ¦ ¢ ù ø ø ù¦ ¤ ¢ ô ô ù E 71¥£X¢5§¥1§(¨b1¥TDT`0h(1f0§B`Rv¨eyXb¥T("~71¥a&¨5s`1§ô æ ´§t3 û ú û X ù ¢ ø õ ù ô 2 $ ù ) ¢ ÷ ' ù õ ÷ ø ù 2 ø ' 2 (I£Xe¨£e¡10"QaXVmdb1ÛAXbX¡1 b(VI(XbTI¥¨1Tù s´HT"10b1 ù $ ù ÷ ` ù ø ÷ E ` 2¦ ÷ E b û ¢ ù ) ù $ ù õ ¢ ø ô ¤ô $ $ õ ÷ ù ø ¢ 2 ù ø ø ¢ E ù ø ù C ù )ô ù $ ¢ ù¦ ¤ ¢¦ô ¢ 4 ¢ $ E ¢ ø ¢ ©¦ b¥£(X5b§¢Vù £bVTv¢0öô (T¥1x(`I("§(x¢£I&öV¥Ô¨I(£a¥b1I¢51 ù $ ô ¢¦ ÷ E¦ ¢ E 2 ¢ $ ù $ ` ø ù 4 ù C $ ` © ô¦ô ¤ ¢ ¤ ø ¦¦ ¢ @ û © ô 4 ù¦ ù ) ù E ø ÷ ¢ ¢£"1§(IÔVX§IU1¥H!bVT`0h(1¥¥¨7§b¥1bV§¥~50ö"XÓ¥XXbXù ¤ (1 C ¢ E E ¢ ô $ 2 ¢ ` ô Eô 2 ù ø ø ù $ ¢ ù C ø 2 ø ¢ $ Eô © ¤ 2 ÷ ¢ C 2 $ ù ) ù $ ¥T0¥vW¨b£ 1¥¨ed§b0(bXbIUhW"äX11a§Û«ÅTù 1r¥#ä¡1£db û ¤ dû ú û õô ó ù ù ` ù ¢ 0IG´£b7©X(0T(IUE $ E ÷ ¦¦ ¢ ø ù¦ ¤ ¢ ÷ 2 $ ù ) õ ø ¢ ù¦ ù $ ô ø ' ÷ ô ø ÷ ù $ 2 ¢ ù 4ô ¢ õ ù 2 a1(Ó§¥"ö1IaXV1Û§ô 3V£0"{VIUöb7¨Û¥¡1§(¢bb£1¥5eIXb1¥¢ ¦ ¢ ø ¦¦ ¢ ù ø ¢ ¥Uù §§¥~( ` ' ` !¥¥3XVP¢£~`1§ô ~71¥£X11Ë¨¨¥U¨(X"1%§§b&yH£y0ö¥Td(ø ø' 6 ù $ ù E ù¦ ¤ ¢ ÷ ©¦¦ ¢ Eô ø ù ) ÷ ô ô $ ¢ $ 2 ù¦ ¢ ù 4 ù ø ù ¢ $ ô $ ô ©¦ ¢ ¢ ø ÷ Î `eV¥1¨b£«§r§ô ¡¨¥HIbÕû a~v¨eb£g"À0(1÷ 3V£0"c{3¥U7bÅ¨ÛC $ E ø ¢ ) ù )ô ù $ ø 2 ù 2 $ ù ) õ ø ¢ ù¦ ù $ $ ô û ú æ ´n å ® b Ø ¦ Þ Ú ¥ c c vÛ £ ä Ö Ú ã d ä ã þ Ð Ï Î ) ù © ù $ 3áXÑ¥¿0eX¡¢£¥¢ ô ¢ ø ¢ ) ô 2 ù ù ø ù ¤ ¢ E ô 2 $ ù ) ù ø E ô 2'ô ) ù ¢ û ù¦ ¤ ¢ ÷ ù ¤ 2¦ ÷ ¥§eya¥1W5ô eI~¨¨0(XT(¢VBù IU¨310"!e0ù XX§bB5"T(#AöV¥a11vE~§1SC $ E ¢ ø ¢ ô $ ¢ $ ©¦ ù ¢ô 2 ù ) ) C I C ù C 2 ¢ © ¢ 2 ø ¢ ù¦ E ù ô ÷ X ô ù E E ù $ ` ø ù 4 a¥b1I§¢fI1£Q5T(IÔb0§ô (11(1#he1¥&UH¡£U7TSe¨¢³§uVT`1TSb£0X0ù '#(0!(1¥V¥#~¨7§b¥£X1¨qIU§ô X"1%A¥V0X1¥Te¢£¥"§eÓ1XHa¨vc@ C û C I ¢ C © ô¦ô ¤ ¢ ô ¦ ¢ E ø ù ) ÷ ù E E ù $ ` ù )ô Ð 2 ø ¢ $ Eô û úû ú ô E ù ÿ 2 $ ù ) ø ùô ø ÷ ó ù $ © ¤ © ô¦ô ¤ ¢ ô 2 ù ©¦ ¢ ¢ 2 ´´~¥§edT´¨¨ô 310"X7(bHbP5ö§¢¥a£¨¨T`¡¨¥HI1I¢ b£t3§ô û ü û X 2¡10"Q§¢£0eTù Tù beù QE¡!§bA¥1§(IH%t"b¨eIX(¨B"¡1£d¢T(¥9Xb $ ù ) ô $ 2 ù ø b û ù © ô $ ô ¢ ÷ X ù õ ô ¢ ø õ ù ô ø 2 $ ù ) ¢ 2 ù ø 2 ø ¢ $ Eô û ó ô C ù â ` E ¢ ô ` ù¦ ¤ ¢¦ô ¢ 4 ¢ ù ) ¢ E ù ¤ ø ù ÷ ) E ù ø ¤ ø ¢ ù © © ¢ X1QHa§ÛÓh§ÛdgÓ0Iv§tVé V( 71¥§§IUW"¥U0E`eVb0e("0ù T¥ã ä ´ ® µ ´ · ´ ¯ ² ® ª° ³ Á ¹ · µ ° ³° ¶ ² µ ´ § ¦ ¯ ® µ º ² ¼ ¢sDÂAQR ¸!TRQ1À(¿DsQ± ü .
b ¦ d ¦ þ ù ø ù $ (`b!C ´n Fû ú ¢ ô )ô¦ ù $ ù I ¢ I¨"§öyb£YQPa é Fû ú ´n Ød â ÚØ â Ø þ ÿ£ ¥ Ø b â £ ÚØ â Ö Ú â d â 2 ¢ ô ¢ ù ù 4 ¤ ¢ ù $ V¥c1¨(1¥1V(¥äb 2 ù b û VHy¢é Ì~c ¦ ý b ¦ d û ú û X ô ) ø ù ù $ ù E ¢¦ ù $ ô £5ÛC ¥ ´mt3¸¨~~XX(yb££X¥§1TÛ ü b iii Þ Ú vvvi â 3¦ û Ø ¦ ( paap p d aÛ d Ø ¦ ( ¥ Ú w w w w b â Ø ¦ aÛ ü b û i w Þ Ú vi â y¦ ö¥ Ú w â aÛ aÛ d Ø ¦ Ö Þ Þ b Ø ¦ × Ü Ú Þ ¥ DÚ ¥ × DÚ × Ý Ú aÛ × IÛ Ý (Û Ü aÛ ¥ d Ú ñ ¦ ù E ù ø ¢ô 2 ù E ¢ 2 ¢ ù )ô 2 ù ô ) ù $ ø 2 0X10(`ù £bX¥HXV¥"§ev0¢©b#"71¥¢ p p aap ii Ú vvi â Ø ¦ d aÛ ò Ú i Ú aÛ â çÖ × IÛ Ý × (Û Ü ù 4 ¢ $ ù C ù 4ô ¢ 4ô ø ù 2 ù )ô ù $ ø ó û¦¦ ¢ ) UH&h5eII5(X¢¨eb#Hun¨¥~ © ø ù 4 ù ) E ù ¤ ù ô $ ù ) ù $ ¢ õ ô 4¦ ©¦¦ ¢ ÷ E ¢ ù ø ¢ ù C ô ¢ ÷ X ù ¦ ¢ô ø ù ¢ô 2 ¦ ¢ô ø ¢ ù $ XT"T0E!X¨(0!bDIb§¨¥&¨§IHed¥!(hx¥§eyH%tb¥§eTù XE¢b3¥Ô(X1¢£ E ÷ ø E ø 2 ¢ ùô ø ø ¦ © ¢ ¢ û ú X ô ) ø ù ù $ 2 ¢ ù ù C ` ô $ ù 4 ø d1((X10Tù ¨V¥!T7(Xù #¨3UsWv§ô I ´£û 37¨5~XX(b£v1¥1VmI3T§¢£(Xbxs û Ï ¤ ¦ ¢ ô õô ø ù $ $ ô D¥1§7(b7¨ÛC dÓdd§d¥ s u x ¤ ¥i ¦ ù $ ô ô $ ø 2ô ¢ ù 4 ¢ $ ù C ù E ø ù $ ú x w Byau ¥¥§b#§§¢£"P¨HUHhX§ô bb û`§¥§)V35 e¨e I§9XøX(¨ô¥¢§b¥9¥e¥§eyH§beTE qdÍr{# ¥b§ô 1}¡1£d#71¥£X ¦ ¢ô ©¦ ô ¢¦ ù ô õ ¢ ø õ ¢ â ô ¢ ô ¤ ) ¤i u ip ù ø ¢ õ ÷ 2 $ ù ) ù¦ ¤ ¢ © ¦ ¢ ô E ÷ 2 ÷ ù ¢ ù C ù ô ` ù ø ¤ 2 õ ) ù ¦ ø ù 4 ¤ ¢ ù $ 5¦§¥1§ô(¨b21¥T11¥¢11"4(1$hAXE1¨(y¥b÷ eù !£¡3Û¡£RTXù ¨1÷ Tù Ø(9¥Sb ÷§TÍ¤dWsrq¥yÍtq¥yaÓ¡ §ô ¡1£dbyH£#1IU"¨b p h Ë ¥ pp Ë x wi¤i x w u x 2 $ ù ) ù $ ¢ $ ¢ ù ) ô $ û ·õöê ð êxõb¥}ùb£¢¨ô ð ¥T(øÇ¥¢ø" Pù ya¨b!D¥f¨¥¥¡ôã ((`b é ù ¢ø $ ¦ ¢ ù © ê ø ¢ $ Eô $ C ¦¦ ¢ ø' ù ø ¢ ù ø ù $ ð Þ Ö gBq0E ð b ¦ y b ¦ By Ø §( Ø ð ô ¥ ô Eô ¢ ø 2 ¢ ÷ X ù $ ô ÷¦ ù $ §m§(I£¢¥H%Hb£l¥v¨eV7b ¥ ¥ ¥ Þ é ÖA ð 9b¦ó{0E ð Ø 7 ð ø á w vs Ý e u Þ w vs Wß ð u e Þ ¥ ç7 ð Þ Ö × Ý 7 ð ù E ¢ 2 ¢ ù¦ ô $ û û ú ô E ù ÿ ô 2 ù E ÷ 2 ø £XV!I!1¥Tö!§¢A´ú Ë¥§edT´¨0X1b¡(§ô ô ©¦ ¢ ¢ © ô¦ô ¤ ¢ ø ùô ø ÷ ó ù $ ¢ ù ) © ¤ ô ÷ ô $ © ø ¢ E © ¢ C ù¦ )ô ù $ §(¡5¥H¥¢¨ö¨b¥£XX7(b"b¥1IU"c¨V§bXø U7£cUXT71§(b ú iü · ± ª ® µ ´° ´ ª ¹ WDs3Q± § Ê§ ¡´(¦ .
ï é B Cî é é ¥ èA ¥ Ö 9 c ï B Cî ¥ 5 é ¥ 5 é à Ö HÚ ã ã ¥ é é ¥ èA é Ö Ù7 $ ô £5ÛC Ï Î ù $ ø ù 2ô &b£X¢§(Vû ý 5 ò ø ÷ î ð 6TAç4 2 ì 3Pë d 9 @Þ Ú 7 8Ö Ú û 11ø XsXe¨¢¥çÖ XV310"TQ5¥tXHÏ 2 ÷ ¤ ø ø ù © ø ¢ ø ô ¤ ø ¢ ¢ û 2 $ ù ) ¦ ù I ¢ ø ó' ø ó ÷ ' 2 1I¢ !"1¥§e5b107010e§(VT1¥d¥£b`T¥f¥X~~V´Ø¢´ö1If ø ô ô 2 E © E ù ô E 2 ¢ © E ¢ ø ÷ E E ¢ © ø ¢ ) ) ÷ ÿ þ Fû ú ù¦ ¤ ¢ 1 ) b 0¦ % d ¦ w ! £ # ! ù ø ù ø ù $ ("d(X¢ á w £ i ß ' % (&Ø ¨s¨1 ô ©¦ i ! Ö $"0eI~§´UbVQ # ! ù ¢ )ô ø w ! £ # ! # Ö $! $ ô £5ÛC i ! ` ù ô ©¦¦ ¢ ô ô 2 T0e§(V~¨¨¥H¨e¨¢1 ù¦ ¤ ¢ ÿ ©¦¦ ¢ ô ô 2 E 71¥£¡¨¨¥H¨e¨¢1TV ù ©¦ ) ø ô T(X§ô 1¨"¨b © E ù ø d1T(X§ô 1Hó ù¦ ¤ ¢ ÿ ©¦¦ ¢ ô ô 2 ö1Ia¡s¨¨¥H¨e5b1 Ø w õ ê d ¦ ê d #¦ ¦ ù I ¢ø ó ø ó ÷ T¥££´XH£Ï © ô¦ô ¤ ¢ ÿ ø ¨ö¨b¥£¡vHó ¢ ÷ ' ù õ ÷ ø ù 2ø ' 2 £VmdbVv!`bB31 û ù¦ ¤ ¢ ù $ 71¥abv§ô C $ 2 $ ù ) ¢ ÷ ' ù õ ÷ ø ù 2 ø ' 2 E ù ù $ ô ô 2 E © ô¦ô ¤ ¢ ù $ © ¤ 2 ù ) !(1c¡1£dv£Vmb1ÛXbX¡1T0(xbI§(¨b1¥T¨ö¨b¥£Xxb£0(§ô ¢ $ ¢ ù ø ù 4 ¢ Ù $ Eô $ C ù ù )ô ù $ ô ¢ ø E ¢ 2 ¢ ù yH£¥(`0ù ¥X1÷ ~¨ô a¨b!cTeW"§egb¨£(X1¥Tv£1¥Tö¦ ô $ ` ø ù 4 ù C w ¨b£lTX0h( þ û ù E ô 2 $ ù ) ù $ 2 ¢ ` d ¦ XTe§ô 1T&§¡10"Pb£1¥AXP§ô Ø ù $ ` ¢ E ô ¢ ) ù TbU¥£X10e1¨~T(ø i I1£yU¢ ¢ $ © ¢ C $ E ÷ ô a1(v¨é 1Ié yBA¢´ö1Ifv¨TX¨(~1(¨Í§(IH(¨(bD`TX¨(T(X ý b ¦ 2 ¢ ý d ¦ @ û Fû ú ù¦ ¤ ¢ ô 2 ù ô ø ¢ ) ) ÷ ô ô ¢ ÷ ô ù $ û ù ô ù ¦§¥~¨¢§bX§b¥I"¢00ù ¥ry(x#¡yd(cyÍv¥1 ¡ xHqqqAHdd§dVya7rgyU H¡ ¦ ¢ ) ©¦ õ ô $ ô ¢ 4 ø ù 4 Ãp w ¥ w ¤ ¤ w ¤ ¥ Ä x w i¤ Ë Æ¤ Æ¤i Ä ¤ x ¤ ¥i ¥ x w u sp w i x û ú û X ¢ E ¢ $ E ÷ @ û ô ¢ ÷ X ù ô ÷ ¢ô 2 ¢ ` ô ¢ ÷ X ù ù 4 ¢ ¤yÓe§i ¨¦¥£¡ É ù ¥Uca1(0¨eIH%{¨(1E¢¢Q¥1¨eI1{UC ¢ ¥ ´0t3 w x ¥ © §¢ ¤ ¢ w §ô ¢ $ © ¢ C ¨¡1£dP¢£yP¨Xc¨¨¥Hed¥7hy§(IH%tsb£AU¥£X1¥T1§~Tù ø i I1£yU¢ ô 2 $ ù ) ù $ © ¤ ù 4¦ ©¦¦ ¢ ÷ E ¢ ù C ô ¢ ÷ X ù ù $ ` ¢ E ô ¢ ) $ E ÷ ô a1(7¨ré V¥~é m¦ 5BVDû Íryd0dH¡Æ §ô Gú t3Dû ÍrydX¥Vù §ú ´Eû 3 ý b ¦ 2 ¢ ý d ô ÷ $ u ip w h ù s Fû û X u ip w h Ë Ë ô û ú X ´ ® µ ´ · ´ ¯ ² ® ª° ³ Á ¹ · µ ° ³° ¶ ² µ ´ § ¦ ¯ ® µ º ² ¼ ¢sDÂAQR ¸!TRQ1À(¿DsQ± æ ü .
` ù¦ ¤ ¢ '@ ø hö1¥£XBÍ ` ù¦ ¤ ¢ ' f 7 ` ù¦ ¤ ¢ ' ô @ û 2 $ ù ) ù $ ø ô ¢¦ ø ' 6 ù $ ù 4 ø ù Ï ` 2 ø ø ¥71¥aXgeIö1IaXB¨0¡1£d&b£"¥§eyÔTù yÏ1x¢£&¨ô XÅB1Tù (0E ô ù 2 ¨ÛT1 G¢t3§9äd1¥¡H1F¥!T¢ö¥IIH!x£1VsB&` ¢l¢0eISW¨Û@ dû û X ô c 2 ¢ ` % ` b ù ÷¦ ¢ 4 ¢ $ C ` @ ã ã â ù ' C ¢ ô `¨`§71~¨©s¨`§713dq¨uv¡1£d#¢£§u`bfIHÕR¢§1"s§¢£Qb¨(1310") ô Eô¦ )ô ø ô Eô¦ ù ô @ ` 2 $ ù ) ù $ ô ø ù 2 ø ¢ $ b û ¢¦ ÷ ) ø ô $ õ ô ÷ 2 $ ù õ ô $ E ø ¢ )' ù )ô ¢ ô ¢ ¤ ` ` ù 2 ô ¦ ¢ ø ) ù ù $ © ¤ X ù 2 ô ¦ ¢ô ¢ ù $ õ ô E ¢¦ ù ø © b§¢a~¨e#§QaVyfadbV§I£~0ebYdb1¨¥§eI1Ab#b§`¥Ô10(S b ¥ × × Ü Ú ¦ Ý Ú F Ö × Ü Ú w Þ Ú w Þ Ó× Ý Ú w W V Wü V W V þ ù 4ô ¢ 4 ø ù 2 ø ¢ ø ô ¢ )ô ø ¢ ô ' F 2 ø ù ù E E ¢ ) E ù $ ¦¦ ¢ E ù û Q¨(II¨ô XbXX£W!"¨eI~§´U¢1¥!§9ByTù X(0X!d¥HTbÓ§¥TTÛ´d $ õ ÷ ø $ ¢b X¢RIV310"£VmdbVvb££5Û!ù HDbbXb£ XbX ø ù 2 ø 2 $ ù ) ¢ ÷ ' ù õ ÷ ù $ $ ô C ø ¢ ) û ü $ õ ÷ ø $ ø ù 2 ø ¥S13V£0"PX"VX¥#¥HQQb£"#b£TW¨ö¨b¥£Xb¥¨1v1¥b¨XXedbü 2 $ ù ) $ ø $ ¢ ' ) ¢ 2 ù $ ø ø ÷ E © ô¦ô ¤ ¢ ù $ ¦ 2 ¢ ù ô ) ø ù ù Ï û û ù¦ ¤ ¢ ô 2 $ ù ) ù $ $ Eô $ C ø Qö1¥£X¨¨¡1Ë 0"¢£7a§b!#" Û 3¥¢¥£b11VT¨(¡¨¥1¥P5ö¨b¥a × ù õ ¢ ø ù $ 2 ` ô ©¦ ¢ ¢ © ô¦ô ¤ ¢ ø ù ø ÷ ó õ Xöô bH¨b§ô Uk¦ ç 8 ¨kTXù W1%l5"P¥bb£IY!(1F` Û ¶(`b!C û b ô ` ø ¤ ) ÷ â ó ø ¢ ø ÷ ù $ ¢ C I × ù ø ù $ û bü × Þ Ú ¥ × Ü Ú ØÛ × Ý Û Ú Û Û ¥ × Þ × Ü Ú Û × Ý Û Ú Û ¥ Ú Û Ö × (Û Ú Ü þ ù 4ô 2 $ ù X05õ 310") ¢ ø 2 b ¢ â C I ©¦ ù 2ô C ù $ ` ô ¢ ÷ X ù ô E ù 4 E ø ¢ ù ô¦ ù $ 2 ùô¦ ¢ ù $ b û PXb1Tù Ty&!{V¡¨0b§ÛbU¨eIH%{¨edT0ÓTb§7Db£#QT7ö11Ie0bÕ¢F û 2 $ ù 0¡10") b£#"{¥byUù TÛUv ù $ ø ù 4 ¤ ¢ ù $ ¢ ù E ù¦ E øô E ô ÷ ù $ öTXö`S5b1Bb£ û ¦ ù ) ¢ ù $ I§1"¥£b¥ C ¢ ø 2 2 ¢ ù ¢¦ ' ù¦ ) E ù $ ô ø ù $ ù E ¢ ø ù $ ¦ ¤ ¤ Ub1I¢¥Ô1B9¿d71Tb£x§e¥3XbÓ¥XI£eb£D¥¨AÌHv û ù E ¢¦ ¢ ù¦ ¢ ø ù $ ù ÷¦ ¢ 4 ù ÷¦ ¤ ¢ ù $ ù ÷ ) E 2 ¢ é æû é `T`¥Ô1gXIUö!y~#£¥¡bsI0böI¥(Vö¥1¥beVb0V¥xG´ é X{x¢¥!¨¥IXX(§Ë¨~xPsû £T¢!¡1£d¢X"3¥#¥H ) ø d éû é ¦ ¢ 4 ø ù ô ô 8 ù I ¢ Ö 6 ù $ C 2 $ ù ) $ ø $ ¢ ' ) ¢ 2 ø ù 2 ø ' 2 `b31 bSI¥¡¿¢£I0böI¥IU§ô X"1%~¢£Igö1¥£7~e3b~¥t ù $ ø' ù $ ù ÷¦ ¢ 4 ¦ ¢ E ø ù ) ÷ ù $ ù¦ ¤ ¢ ¢ ù ÷ ) ¢ û ` 2 ÷ ¤ ù $ $ ô C ù E ¦ ÷ ø ø ÷ © ø ` © ô¦ô ¤ ¢ ø 8 2 ÷ 11V9ù Tb7¨Û!T(X§ô 10£¨V(Tù Ûbù 755ö¨b¥a"7!1V1¤ 2 ù E ù ù $ ù ø ¢ ¢ $ C ` 2 $ ù ) ù ù ø $ ù $ ø ô ¢¦ ù ø ' 6 ù $ õ ô T(0Tù 3dbP(yH!01¡1£dv`(b7ù TbW"&V§(IÔ0(!Vb£b¨(SUv E û ô ÷¦ © 2 ¢ ù E ¢ ù ù $ $ ô C ô ÷¦ 2 ù ÷ ) E ù $ ù ø ¢ ) 0¨eV773¥U(X0¥V¢£v£5ÛP§(VöX0eV¢~0b£(QH~¥ û ù ù )ô é `1T(X"§(¢é "é ÕÅ1¥~1T(X"§(¢¥d "¢ü é Õ01T(XP"§( ø Ö 8 2 ¢ ù ù )ô é ø Ö 8 ` ù ù )ô é é d ø é Ö 8 ` ù ù )ô é é ¥¥&" G yA01T(Xg"§(v¥é " ¨ é Y(1qT0(¥T~`(bf§¥`1¡ ø Ö 8 ù ÷ ` ù ¢ E ù ù ø $ ¦¦ ¢ @ û 2 '3£d&¥X(I1¥0XöVel5X¨ö1§0X71§A¨`§713d5bb§ô 1é A~"§( $ ù ) I E ¢ ) ø E ¢ ã 2 ¢ ` ø ù¦ ÷ ô Eô¦ )ô ` ø ù¦ ÷ ô Eô¦ ù ù $ õ ÷ Ö d ù )ô ) ø QI { RP c c Ht Ú §(¨b1¥TR¥§e¨¢§b ¥{&(I£Xe¨£eb¡T~ge5(ø PHv G Ö é ô ô 2 E ¦ ¢ô ô ô ù $ ) ø ù ¢ ø õ ù ô ù 2 E ¢ ù ô b ¤ ` ù )ô ô ù 4 ¢ $ ù ¤ ô ÷¦ Ï Î ù $ ù 2 C ` ô ÷¦ ù ¢ h¨ev§UH0Ev¥§e3öXÅ&b£T1¸(¨eV7(Ia ' © 2 ¢ ù ù $ ô ¢ $ b û ù õ ¢ I E ¢ ¦ ¢ E ø ù ) ÷ ¢ õ ÷ 7 ù ÷¦ ¢ 4 ù õô ù ù $ 2 ô ¢ V¥Teb£¨ÛIHÕQy¥HIU§ô X"1%Wxb§ô 1F¥T¢ö¥IT7XmbP1§qó t ü ´ ® ³ ¶ ¹ ¯ º § D§ ¢#DÀE(¦ .
`¦ ¢ ) ¥~ø1ç×§§þX¨BD¢»T71¢`I1¥b»{#XXE¢b¨b£v!v§QÔ13f ô ô û æ ) ù¦ ¤ ø ø ù C ¢ ù $ ) ø ø ù ¢ô 2 ô $ © $ C ô ¢¦ ` © ô¦ ¤ ¢ ô ) © ¢ 2 ù C ¦¦ ¢ ø ¤ ) ÷ ¢ ø ÷ ¨7§ôbIaXQE§eI£V ¡X¥ø"PTh(§§IXù W1%!¥£¢ 1§´y~b§5IH!1¥ù Tb ) ÷ )ô ¢ ) ù $ ô ¢ $ C ` ù $ T(I#¿ûX0böI¢¥0ùö`'9 ùb£ Ù1H¡10"`ö1§¨`§713dù b£v"Ò¨eI§T(ø 2 ù ¢ ù ÷¦ 4 õô ù $ 2 ` 2 $ ù ) ø ù¦ ÷ ô Eô¦ ù $ ø ô ¢¦ ù ' ù $ 16¢b£cõb§ô (û q ¥~0öô eT¨ùb£ 75SV¥«§(Ia¥V 5ô eI~À0b1¥H«§ô e5(ø b ù ø ù $ ù 4ô õ 2 ¢ ô ¢ ø ¢ ) 2 ù 2 ¢ ¤ × ù ô Û yq ¥ HÚ Ö × (HÚ Û × Ü Û ä ` û ù û ô ` ) ø ø Eä ù 4 ' ô ø ¢ ) ô õ ô $ E ø ¢ ) ù ) ô ø ù ¦ ÷ ô E ô ¦ ù ¨Q¢"#£00Í¡©(ä (I7¨}bä §baX~¨eX71§!¨`§713dq¥ ô ¢ Eô¦ ¢ ù $ ) ø ¦ ÷ ù ø $ Eô $ C ¦ Î ù $ ù ô b û ù ÷¦ ¢ 4 ù õô ù' 6 ù $ 2 û §(IU¨ö1V¥e¢£ ¥{£51(0(Ëa¨b!Åx§&b£(¨(ø AX0bö¥ITöX1&bVH1` ô 2 ù õ ô û ) ø ø E ù 4 ' ô ø ¢ ) ô õ ô E ù ø ¢ô 2 ¦ ¢ô ¢ 2 ø ¢ C I E ¢ ¤ ø ù 2 ø ' ø 5b10EVQQ¢§(p¨"l00B©(eyW¨b¨X10(`ù £b3IÔeyHxQ#¥HsXbXXX£ ) ø ¦ ø $ Eô $ C Ð Ï Î {¨1÷ Tù a¨b!Ó¥ T(X¡X!be5(ø fH1V9¥ö(ÛbIx0e5XXù ) ù © ù $ ù ô b û © ø ¢ 2 ÷ ¤ $ õô ø ù $ ¢ 2 ù ô ) ø ¨Í¨e¨¢1TX111xDH1V9#q0ö¢£Iv¥§e5b10PH1190öVa§(7g¢ ô ô ô 2 E © ø ¢ 2 ÷ ¤ û © ø ¢ 2 ÷ ¤ ù¦ ù $ ¢ ô ô 2 E © ø ¢ 2 ÷ ¤ ù¦ $ Eô øô Ï ©¦ û ¨1VQl¢æ 0ö1¥b¨g¥e0Xù Uf5e¨ô #&¨Û§(00T9U¢§öA¢£X¢§(Vû ) ù¦ ¤ ø ô ¢ 2 ù ù 4 ¢ C ù 4ô ¢ $ ô C ô E ù 4 E ø ¢ ù ô¦ ù $ ø ù 2ô û © ô¦ô ¤ ¢ ø ø ¤ ) ÷ ¢ ø ÷ ) ÷ )ô ¢ ) ù $ 2 ô ó 5ö§¢¥a"Xù W1%I£b»1¨´ybv1§©Øû 5e5(7£¨Ûv¨eI1{ù ù 4ô ô $ ô C ô ¢ ÷ X ô E ù 4 E ø ¢ ù ô¦ ù $ 2 ùô¦ ¢ õ ô $ E ø ¢ ) ù )ô ø ù¦ ÷ ô Eô¦ ù $ ô C 2 ù¦ ÷ E õ ¨e000Ub¨ö&b£x£7T7ö11I~¢§baX"§(`ö1§¨`§713d¨£¨ÛT7110~b¨ô ' E ø ù ¢ô 2 2 ø ¢ C I E ¢ ¤ ø ù 2 ø ' ø © ô¦ô ¤ ¢ ù $ ù ©¦ ¢ ¢ ` ©¦ ¢ ¢ ø ù ø ÷ ó õ ô û 1Tù XE¢by¥HXbXXb#¥¢5ö¨b¥a~¢£xX¡¨¥1¥lT§ô ¡¨¥HI&`öô bHË¢§(&¢æ û 2 $ ù ) ù ù $ ù ø ô ô 2 `1¡1£d(0b£l¥b"P¥§e5b10E © ô¦ô ¤ ¢ ù $ 2 ô 5ö§¢¥avb£yV§©ó h§b¡§ô Xù v(e1¨e¨¢1TyX111Pb V¥xT`1Tù XE¢b¥Ô( ` Eô 2 ø ù ø ¢ ô ô 2 E © ø ¢ 2 ÷ ¤ ù $ 2 ¢ ù E ø ù ¢ô 2 ¦ ¢ô ' ¢ ù $ ø 2 ù ÷ ô õ ô E ù ø ù ¢ô {H¢¢£"Ò0(1§»b¨X10(XE¢¢2 0(X(TX~XbX¡1 ¨h¤ §bb¨eIX(¨ô 2 ùø ù ù E ø ù 2ø ' 2 ô ` Ï ô $ õ ô ¢ ø õ ù ¥ Ö (`b!C ø ù¦ ¤ ¢ ô ÷ ù ¤ 2¦ ÷ C 2 $ ù ) õ ô $ E ø ¢ )' ù )ô ô Eô¦ ù $ Eô $ b "71¥a5b(&E§1SQ1¡10"¢§baXBd"§(5X§713d~a¨bÕQû ù ø ù $ Ë Ø b â Ú Ø â Ö Ú â þ Ï ¤ õ ô C ¦¦ ù $ ø ù 2ô û ÛD¢§Û(§¨"b£X¢§(V´ú d â û ô ©¦ ¢ ¢ ø ùô ø ÷ ó õ ô ÷ ¢ ù ù û ô X¨(¡¨IH¥!X7(b"b¨(1vITE0v0§( ' ¢ ÷ X ù ô ¢ô 2 ¦ ù 2 ) ù $ 2 ùô¦ ¢ 2 $ ù ) $ ø $ ¢ ' ) ¢ 2 ø ù 2 ø ' 2 {H%ty¥§(V÷ £bDTb¡~b££0ö711¥ä¡10"y£"VX¥#~IHQ&XbX3V b ù $ 2 ¢ ù E ù ø ù ¢ô 2 2 ù ø ù ù E ø ù 2 ø ' 2 1I0X10(XE¢¢vT(`eT`AXbX¡1 Ig§(IH§¢WTb"~0eX"§(ö1¥£X V) ô ¢ ô ¤ ) E ù $ ø ù ù )ô ù¦ ¤ ¢ ) ÷ ¥ '5´y~eb1HH95ô eI~£IXù ¥Ô(IHXeb¥Tb7¥I0ö`b~b¨(iû ô ¢ ) ù $ 2 ` ø ¢ ) ø ¢ ø ¦ ¢ô ¢ ù $ ù ÷¦ ¢ 4 ù õô ù ù $ õ ô Ö 8 6 $ ô C 2 ù ô ¢ ¤ ô ` ô ¢ ¦ ¢ ù ø ù 4ô ¢ õ ù ù $ E ù ø ù ô ø ÷ E © ô¦ô ¤ ¢ ù $ ù ø ù $ ¨Û0b§QaV&§qT§´yS¥T(s¨(IX¢Pb£&£00(XXe¨bTc5ö§¢¥ab(`b!C ô ù $ `û ùû ô ` 6 ù 4ô ¢ õ ù ¦ ¢ ù ø ø 8 ù ÷¦ ¢ 4 ù¦ ¤ ¢ ) ÷ )ô ¢ ) ù $ ¢ $ C $ ¨eb1¨QERW¨(IX¢lIU(Û"6 ¥¥b7¥I71¥£X»V§´y~&¢£IH£¸(1X é ÙÖ ç 6 8 Þ 7 ç F 8 6 Þ ¥ Ø 2 $ ù ) $ ø $ ¢ §ô ¡1£d£"VI# ' ) ¢ 2 ø ù 2ø ' 2 ¥HQ!`bB31 b"¨eI§T(ø !V&b£yH£7T¨ô û ©((I£IXù ¥§eIHX ù $ ø ô ¢¦ ù ' 6 ù $ ¢ $ ù 4 ô ø ¢ ) ø ¢ ø ¦ ¢ô ¢ ù¢£l¥!Tb7¥I0ö`b£71HVV75b10EVQxb§ô h`1§(¨bVTXHV1§b¡§ô Xù $ ù ÷¦ ¢ 4 ù õô ù ù $ 2 ` üû ô 2 ù õ û ô ô 2 E © ø ¢ 2 ÷ ¤ Eô 2 ø $ ô C ô ¢ ÷ X ù ô ÷ ¢ô 2 ¦ ù 2 ) ù $ ô ù 4ô ¢ 4ô ø ù 2 ¦ ¢ô ¢ ù $ ô ¢ )ô ø ¨Ûä§(IH%t¨(1E¢¢#Tb¡bä§5eII5(X¢¥Ô(IH7b£Y¨eI~§´UbV¥¢ ù E ø ù ¢ô 2 2 ù ø ù ù E ø ù 2 ø ' 2 `1Tù XE¢b¢0(`eT``b31 ¢£gb§511¥Tb¨aV¥ 0e3ÀW¢£e5(ø ¢ ù $ õ ô ©¦ ¢ © ¤ 2 ù ô ¢ ¤ ) ù © Ï Î ù $ ù ô b û ´ ® µ ´ · ´ ¯ ² ® ª° ³ Á ¹ · µ ° ³° ¶ ² µ ´ § ¦ ¯ ® µ º ² ¼ ¢sDÂAQR ¸!TRQ1À(¿DsQ± é ¥d .
1 Relation to Eigenvalues The introduction of the concept referred to as \sti ness" comes about from the numerical analysis of mathematical models constructed to simulate dynamic phenomena containing widely di erent time scales. and. De nitions given in the literature are not unique. and the decision to use some numerical timemarching or iterative method to solve it. 7. The forcing function may also be time varying in which case it would also have a time scale. since this part of the ODE has no e ect on the numerical stability of 149 . The di erence between the dynamic scales in physical space is represented by the di erence in the magnitude of the eigenvalues in eigenspace.Chapter 8 CHOICE OF TIMEMARCHING METHODS In this chapter we discuss considerations involved in selecting a timemarching method for a speci c application.1. 8. but fortunately we now have the background material to construct a de nition which is entirely su cient for our purposes. Any de nition of sti ness requires a coupled system with at least two eigenvalues.1.1 Sti ness De nition for ODE's 8. We start with the assumption that our CFD problem is modeled with su cient accuracy by a coupled set of ODE's producing an A matrix typi ed by Eq. An important concept underlying much of this discussion is sti ness. which is de ned in the next section. we assume that this scale would be adequately resolved by the chosen timemarching method. Examples are given showing how timemarching methods can be compared in a given context. However. In the following discussion we concentrate on the transient part of the solution.
In x many numerical applications the eigenvectors associated with the small j mj are well resolved and those associated with the large j mj are resolved much less accurately. timemarching method is represented by Eq. the homogeneous part. although these eigenvectors must remain coupled into the system to maintain a high accuracy of the spatial resolution. we exclude the forcing function from further discussion in this section.150 CHAPTER 8. . The situation is represented in the complex h plane in Fig.28. CHOICE OF TIMEMARCHING METHODS Stable Region λh I(λ h) 1 R(λ h) Accurate Region Figure 8. if at all. In this gure the time step has been chosen so that time accuracy is given to the eigenvectors associated with the eigenvalues lying in the small circle and stability without time accuracy is given to those associated with the eigenvalues lying outside of the small circle but still inside the large circle. This is given by Eq. Consider now the form of the exact solution of a system of ODE's with a complete eigensystem. The whole concept of sti ness in CFD arises from the fact that we often do not need the time resolution of eigenvectors associated with the large j m j in the transient solution.27 and its solution using a oneroot. 6. For a given time step. 6. 8.1. the time integration is an approximation in eigenspace that is di erent for every eigenvector ~ m .1: Stable and accurate regions for the explicit Euler method.
and the other.1.1. It is important to notice that these de nitions make no distinction between real.2 Driving and Parasitic Eigenvalues For the above reason it is convenient to subdivide the transient solution into two parts. First we order the eigenvalues by their magnitudes. Rephrased. thus j 1j j 2j j Mj (8. numerical stability requirements are dictated by the parasitic ones. complex. we nd that. 8. although time accuracy requirements are dictated by the driving eigenvalues.3 Sti ness Classi cations In particular we de ne the ratio and form the categories The following de nitions are somewhat useful. but their presence must not contaminate the accuracy of the complete solution). An inherently stable set of ODE's is sti if j pj j Mj Cr = j M j = j pj Mildlysti Cr < 102 Stronglysti 103 < Cr < 105 Extremelysti 106 < Cr < 108 Pathologicallysti 109 < Cr It should be mentioned that the gaps in the sti category de nitions are intentional because the bounds are arbitrary. it states that we can separate our eigenvalue spectrum into two groups one 1 ! p] called the driving eigenvalues (our choice of a timestep and marching method must accurately approximate the time variation of the eigenvectors associated with these). Unfortunately. . and imaginary eigenvalues.8. STIFFNESS DEFINITION FOR ODE'S 151 8. called the parasitic eigenvalues (no time accuracy whatsoever is required for the eigenvectors associated with these.1) Then we write p M Transient = X c e m t ~ + X c e m t ~ (8.1. p+1 ! M ].2) xm xm m m Solution m=1 {z m=p+1  } {z }  Driving Parasitic This concept is crucial to our discussion.
we are interested only in the converged steadystate solution. As a result it is quite common to cluster mesh points in certain regions of space and spread them out otherwise..1. Even for a mesh of this moderate size the problem is already approaching the category of strongly sti . the sti ness is determined by the ratio M = 1 . and are greatly a ected by the resolution of a boundary layer since it is a di usion process. and in a boundary layer. Typical CFD problems without chemistry vary between the mildly and strongly sti categories.2. 2 and the ratio is .152 8. this ratio is about 680.e. A simple calculation shows that 4 2 1 = . CHOICE OF TIMEMARCHING METHODS Many ow elds are characterized by a few regions having high spatial gradients of the dependent variables and other domains having relatively low gradient phenomena. but much less so than for di usiondominated problems. For a mesh size M = 40. 4x2 4 =4 M +1 2 2 M= 1 The most important information found from this example is the fact that the sti ness of the transient solution is directly related to the grid spacing. The simplest example to discuss relates to the model di usion equation. We see that in both cases we are faced with the rather annoying fact that the more we try to increase the resolution of our spatial gradients. One quickly nds that this grid clustering can strongly a ect the eigensystem of the resulting A matrix. Consider the case when all of the eigenvalues are parasitic. In order to demonstrate this. negative numbers that automatically obey the ordering given in Eq.2 Relation of Sti ness to Space Mesh Size CHAPTER 8. 4x2 sin2 2 = . the sti er our equations tend to become. In this case the eigenvalues are all real. Under these conditions. 8. Furthermore. 2 sin 2(M + 1) x M ! . i. For the biconvection model a similar analysis shows that j j=j j 1 M x 1 Here the sti ness parameter is still spacemesh dependent.3. in di usion problems this sti ness is proportional to the reciprocal of the space mesh size squared. let us examine the eigensystems of the model problems given in Section 4. 4x2 x 2=. near an airfoil leading or trailing edge. Examples of where this clustering might occur are at a shock wave. Our brief analysis has been limited to equispaced x .
and its computation usually consumes the major portion of the computer time required to make the simulation. discussed in Section 6. The actual form of the coupled ODE's that are produced by the semidiscrete approach is ~ u At every time step we must evaluate the function F (~ t) at least once. and erp discussed earlier. and the accuracy would be related to how much the energy of certain harmonics would be permitted to distort from a given level.3. Such a computation we refer to as an event. capacity. The appropriate error measures to be used in comparing methods for calculating an event are the global ones.5.6. T . PRACTICAL CONSIDERATIONS FOR COMPARING METHODS 153 problems. This function is usually nonlinear. Nevertheless. For example. d~ = F (~ t) u ~ u dt . There is no unique answer to such a question. but in general the sti ness of CFD problems is proportional to the mesh intervals in the manner shown above where the critical interval is the smallest one in the physical domain. The length of the time interval. era . 8. if certain ground rules are agreed upon. and the accuracy required of the solution are dictated by the physics of the particular problem involved. one can wonder how this information is to be used to pick a \best" choice for a particular problem. relevant conclusions can be reached. Let us now examine some ground rules that might be appropriate. It should then be clear how the analysis can be extended to other cases. among other things. rather than the local ones er . For example. Since there are an endless number of these methods to choose from. Let us consider the problem of measuring the e ciency of a time{marching method for computing.3 Practical Considerations for Comparing Methods We have presented relatively simple and reliable measures of stability and both the local and global accuracy of timemarching methods. and technology in uencing this is undergoing rapid and dramatic changes as this is being written. the time interval would be that required to extract a reliable statistical sample. Era . highly dependent upon the speed. it is. in calculating the amount of turbulence in a homogeneous ow. and architecture of the available computer.8. Er and Er! . an accurate transient solution of a coupled set of ODE's. We refer to a single calculation of the ~ u vector F (~ t) as a function evaluation and denote the total number of such evaluations by Fev . over a xed interval of time.
must have the property: du = . First of all. Three methods are compared  explicit Euler.4 Comparing the E ciency of Explicit Methods As mentioned above.2 An Example Involving Di usion Let the event be the numerical solution of from t = 0 to T = . In some contexts. Implications of computer storage capacity and access time are ignored. and RK4. h. this can be an important consideration.3) Er < 0:005 eT .e. 2. must simulate an entire event which takes a total time T . i. The timemarch method is explicit. To the constraints imposed above. AB2. so that CHAPTER 8. 8.u dt (8.4. We judge the most e cient method as the one that satis es these conditions and has the fewest number of evaluations. the allowable error constraint means that the global error in the amplitude. Eq. see Eq. i. 6. The follow assumptions bound the considerations made in this Section: 1.1 Imposed Constraints T = Nh where N is the total number of time steps.48. and must use a constant time step size.154 8. The calculation is to be timeaccurate.3 is obtained from our representative ODE with = . let us set the additional requirement The error in u at the end of the event.t. CHOICE OF TIMEMARCHING METHODS 8.1. must be < 0:5%. the e ciency of methods can be compared only if one accepts a set of limiting constraints within which the comparisons are carried out. a = 0...4. this makes the exact value of u at the end of the event equal to 0. u(T ) = 0:25. Since the exact solution is u(t) = u(0)e.e. 3. the global error. ln(0:25) with u(0) = 1. 8.25. Fev .
1: Comparison of timemarching methods for a simple dissipation problem. Method N h Fev Er 1 Euler 193 :00718 :99282 193 :001248 worst AB2 16 :0866 :9172 16 :001137 RK4 2 :6931 :5012 8 :001195 best Table 8.4.3 An Example Involving Periodic Convection Let us use as a basis for this example the study of homogeneous turbulence simulated by the numerical solution of the incompressible NavierStokes equations inside a cube with periodic boundary conditions on all sides.4.8.4. 8. and the number of these evaluations must be kept to an absolute minimum because of the magnitude of the problem. In this example we see that. Under these conditions a method is judged as best when it has the highest global accuracy for resolving eigenvectors with imaginary eigenvalues.1. Thus the secondorder AdamsBashforth method is much better than the rstorder Euler method. it follows that 155 1 . The main purpose of this exercise is to show the (usually) great superiority of secondorder over rstorder timemarching methods. ( 1 (ln(:25)=N ))N =:25 < 0:005 where 1 is found from the characteristic polynomials given in Table 7. The above constraint has led to the invention of schemes that omit the function evaluation in the corrector step of a predictorcorrector combination. In this numerical experiment the function evaluations contribute overwhelmingly to the CPU time. ln(0:25)=N . In this case. the method with the highest local accuracy is the most e cient on the basis of the expense in evaluating Fev . in addition to the constraints given in Section 8. leading to the socalled incomplete .1 were computed using a simple iterative procedure. since h = T=N = .1. for a given global accuracy. The results shown in Table 8. On the other hand. and the fourthorder RungeKutta method is the best of all. a complete event must be established in order to obtain meaningful statistical samples which are the essence of the solution. we add the following: ~u The number of evaluations of F (~ t) is xed. COMPARING THE EFFICIENCY OF EXPLICIT METHODS Then.
Basically this is composed of an AB2 predictor and a trapezoidal corrector. and the number of evaluations. of course.4) 1 See Eqs.39. The second and fourth order RK methods are 2. However. so that for these methods h = h1 . the time span required for one application of the method increases. For a 1evaluation method the total number of time steps. that more e cient methods will result from the omission of the second function evaluation. However. etc. the time step must be twice that of a one{evaluation method so h = 2h1.38 and 6. Notice that as k increases. Let h1 be the time interval advanced in one step of a oneevaluation method. notice also that as k increases. In general. given in Section 6.relation for the method is shown as entry 10 in Table 7. An example is the method of Gazdag. one evaluation being used for each step. j 1(ik!h1 )jK=k (8. The Gazdag. In order to discuss our comparisions we introduce the following de nitions: Let a kevaluation method be de ned as one that requires k evaluations of ~u F (~ t) to advance one step using that method's time interval. are the same. k=1 j k = 2 j 2h1 k=4 j 4h1 0 T j j j ( h1 )]8 (2 h1)]4 (4 h1)]2 uN Step sizes and powers of for kevaluation methods used to get to the same value of T if 8 evaluations are allowed. respectively. For a 2evaluation method N = K=2 since two evaluations are used for each step. after K evaluations.1.156 CHAPTER 8. and AB2 schemes are all 1evaluation methods.8. the derivative of the fundamental family is never found so there is only one evaluation required to complete each cycle. CHOICE OF TIMEMARCHING METHODS predictorcorrector methods. For a 4evaluation method the time interval must be h = 4h1 . K . This is the key to the true comparison of timemarch methods for this type of problem. in this case. in order to arrive at the same time T after K evaluations. Let K represent the total number of allowable Fev . the global amplitude and phase error for kevaluation methods applied to systems with pure imaginary roots can be written1 Era = 1 . The . N . 6. However. The presumption is. . h.and 4evaluation methods. leapfrog. the power to which 1 is raised to arrive at the nal destination decreases see the Figure below.
4.2: Comparison of global amplitude and phase errors for four methods.4 and 8. K leapfrog AB2 Gazdag RK4 !h1 = :1 100 1:0 1:003 :995 :999 !h1 = :2 50 1:0 1:022 :962 :979 a. for !h = 0:2 the Gazdag method produces a j 1 j = 0:9992276 whereas for !h = 0:8 (which must be used to keep the number of evaluations the same) the RK4 method produces a j 1 j = 0:998324.3:8 .5. 8. The rst three of these are oneevaluation methods and the last one is a fourevaluation method. The event must proceed to the time t = T = 10. Notice that to nd global error the Gazdag root must be raised to the power of 50 while the RK4 root is raised only to the power of 50/4.1 k " 157 (8. Using this. the results shown in Table 8. COMPARING THE EFFICIENCY OF EXPLICIT METHODS Er! = !T . Gazdag. and RK4. The 1 root for the Gazdag method can be found using a numerical root nding routine to trace the three roots in the plane. by some rather simple calculations2 made using Eqs.2. First of all we see that for a oneevaluation method h1 = T=K .3e. 7.8. We consider two maximum evaluation limits K = 50 and K = 100 and choose from four possible methods. It is not di cult to show that on the basis of local error (made in a single step) the Gazdag method is superior to the RK4 method in both amplitude and phase. We idealize to the case where = i! and set ! equal to one. exact = 1. Table 8. AB2.0. we are making our comparisons on the basis of global error for a xed number of evaluations. see Fig. K leapfrog AB2 Gazdag RK4 !h1 = :1 100 . 2 .2:4 :45 :12 !h1 = :2 50 . Phase error in degrees.9:8 1:5 1:5 b. K tan.:96 . although. However. leapfrog. in terms of phase error (for which it was designed) it is superior to the other two methods shown. Amplitude. and the fact that ! = 1. For example. On the basis of global error the Gazdag method is not superior to RK4 in either amplitude or phase.5) 1 (ik!h1 )]imaginary 1 (ik!h1 )]real # Consider a convectiondominated event for which the function evaluation is very time consuming. we nd.
001 is negligible. The time step limit based on stability. We will also assume that c1 = c2 = 1 and that an amplitude less than 0. The coupled equations in real space are represented by u1(n) = c1 (1 .6. Numerical evaluation is altogether di erent. If some h fall outside the small circle but stay within the stable region. 100h)nx11 + c2(1 . For a speci c example. the time histories of the two solutions would appear as a rapidly decaying function in one case. We rst run 66 time steps with h = 0:001 in order to resolve the 1 term. 100h)nx21 + c2(1 . If h is chosen so that all h in the ODE eigensystem fall inside this circle the integration will be numerically stable. Also shown by the small circle centered at the origin is the region of Taylor series accuracy. CHOICE OF TIMEMARCHING METHODS Using analysis such as this (and also considering the stability boundaries) the RK4 method is recommended as a basic rst choice for any explicit timeaccurate calculation of a convectiondominated problem. 8. We have de ned these h as parasitic eigenvalues. is h = 0:02. 8. With this time step the . This de nes a time step limit based on accuracy considerations of h = 0:001 for 1 and h = 0:1 for 2 .1 Explicit Methods The ability of a numerical method to cope with sti ness can be illustrated quite nicely in the complex h plane. and a relatively slowly decaying function in the other.5 and 7. these h are sti .1. depend upon ( m h)]n and no j m j can exceed one for any m in the coupled system or else the process is numerically unstable. If uncoupled and evaluated in wave space. consider the mildly sti system composed of a coupled twoequation set having the two eigenvalues 1 = .1.100 and 2 = . of course.5 Coping With Sti ness 8. In this case the trace forms a circle of unit radius centered at (. Let us choose the simple explicit Euler method for the time march. Numerical solutions.1 0) as shown in Fig. A good example of the concept is produced by studying the Euler method applied to the representative equation. but stable. 7.6) We will assume that our accuracy requirements are such that su cient accuracy is obtained as long as j hj 0:1. h)nx22 + (PS )2 (8. The transient solution is un = (1 + h)n and the trace of the complex value of h which makes j1 + hj = 1 gives the whole story.158 CHAPTER 8. Stability boundaries for some explicit methods are shown in Figs. h)nx12 + (PS )1 u2(n) = c1 (1 .100t quickly becomes very small and can be neglected in the expressions when time becomes large. Analytical evaluation of the time histories poses no problem since e.5. which is determined from 1.
(1 . It is true that in the nal 69 steps one root is 1 . We choose the trapezoidal method. (It is true that for the same accuracy we could in this case use a larger step size because this is a secondorder method. 0:5h nx + (PS ) 2 1 + 0:5h 22 ! (8. However.2 Implicit Methods 1 .4b. we simply replace the Euler with the trapezoidal one and analyze the result. COPING WITH STIFFNESS 2 159 term is resolved exceedingly well.001).001 and that of the 2 term (i. h)n term to zero (i.9361.8. This is the same step size used in applying the explicit Euler method because here accuracy is the only consideration and a very small step size must be chosen to get the desired resolution. After 70 time steps the 1 term has amplitude less than 0.e. Now with the implicit method we can proceed to calculate the remaining part of the event using our desired step size h = 0:1 without any problem of instability. Since this is also a oneroot method..t to below 0.5. h)n) is 0. 50h nx + c u1(n) = c1 1 + 50h 11 2 ! 1 . the amplitude of the 1 term (i. 100h)n. 50(0:1)]= 1 + 50(0:1)] = 0:666 . Its behavior in the h plane is shown in Fig. with h = 0:02.001 and can be neglected. 50h nx + c u2(n) = c1 1 + 50h 21 2 ! 8. we need to use a step size of about h = 0:001. In both intervals the desired solution is secondorder accurate and well resolved.100t . which in just 10 steps at h = 0:1 ampli es those terms by 109. we would like to change the step size to h = 0:1 and continue. 0:5h nx + (PS ) 1 1 + 0:5h 12 ! 1 . (1 . Now let us reexamine the problem that produced Eq.5. with 69 steps required to reduce the amplitude of the second term to below 0. Thus the total simulation requires 405 time steps. about 339 time steps have to be computed in order to drive e. far outweighing the initial decrease obtained with the smaller time step. However.001. It follows that the nal numerical solution to the ODE is now represented in real space by 1 . To drive the (1 .e. Hence the 1 term can now be considered negligible. this is not possible because of the coupled presence of (1 .001...6 but this time using an unconditionally stable implicit method for the time march. 100h)n) is less than 0. We would then have a well resolved answer to the problem throughout the entire relevant time interval. 8. its in uence in the remaining 70 . and this has no physical meaning whatsoever. its in uence on the coupled solution is negligible at the end of the rst 70 steps. but that is not the point of this exercise). After 66 steps. below 0.e. since (0:666 )n < 1. and.7) In order to resolve the initial transient of the term e. 7. the maximum step size that can be taken in order to maintain stability. In fact.
Our sole goal is to eliminate it as quickly as possible. the advantage begins to tilt towards implicit methods.3 A Perspective 8. as compared to about 139 for the trapezoidal method. which is in the stronglysti category. It has enjoyed remarkable success in many practical problems however.5. These explicit methods can be considered as e ective mildly sti stable methods. In the example. the conditionally stable Euler method required 405 time steps. It is important to retain a proper perspective on a problem represented by the above example. . CHOICE OF TIMEMARCHING METHODS steps is even less.1. 2 = . if there are more spurious roots (8. it should be clear that as the degree of sti ness of the problem increases. its behavior for t > 0:07 is identical to that of a stable spurious root.6 Steady Problems In Chapter 6 we wrote the O E solution in terms of the principal and spurious roots as follows: un = c11 ( 1 )n ~ 1 + + cm1( m )n ~ m + + cM 1 ( M )n ~ M + P:S: 1x 1x 1x n~ + n~ + +c12 ( 1)2 x1 + cm2 ( m )2 xm + cM 2 ( M )n ~ M 2x n~ + n~ + +c13 ( 1)3 x1 + cm3 ( m )3 xm + cM 3 ( M )n ~ M 3x +etc. It is clear that an unconditionally stable method can always be called upon to solve sti problems with a minimum number of time steps. as the reduced number of time steps begins to outweigh the increased cost per time step. However.160 CHAPTER 8. However. Therefore. the Euler method is extremely easy to program and requires very little arithmetic per step. For re ned investigations of such problems an explicit method of second order or higher. is recommended. For preliminary investigations it is often the best method to use for mildlysti di usion dominated problems. we need an introduction to the theory of relaxation before it can be presented.10 000. we have no interest whatsoever in the transient portion of the solution. The total simulation requires 139 time steps. about three times as many. The reader can repeat the above example with 1 = . such as AdamsBashforth or RungeKutta methods. 8. Actually. although this root is one of the principal roots in the system..8) When solving a steady problem. There is yet another technique for coping with certain sti systems in uid dynamic applications. This is known as the multigrid method.
What is the maximum allowable time step if all but the rst eigenvector are considered parasitic? . Eq. If we use the explicit Euler timemarching method what is the maximum allowable time step if all but the rst two eigenvectors are considered parasitic? Assume that su cient accuracy is obtained as long as j hj 0:1. For example. Among implicit methods.2) and periodic convection (Section 8. 6. 8.1. the di erence being that the order of accuracy is irrelevant. Therefore the time step can be chosen to eliminate the transient as quickly as possible with no regard for time accuracy. because of their stability properties.2) and periodic convection (Section 8. Consider the di usion equation (with = 1) discretized using 2ndorder central di erences on a grid with 10 (interior) points. Considering the stability bounds for these methods (see problem 4 in Chapter 7) as well as their memory requirements. the implicit Euler method is the obvious choice for steady problems.3) using the 3rd. 3.3) using 2nd. 2. when using the implicit Euler method with local time linearization. Hence the explicit Euler method is a candidate for steady di usion dominated problems. none of the eigenvalues are required to fall in the accurate region of the timemarching method. a nite time step may be required until the solution is somewhat close to the steady solution. Eq.96.8. Referring to Fig.7 Problems 1.4. Repeat the timemarch comparisons for di usion (Section 8. Find and plot the eigenvalues and the corresponding modi ed wavenumbers. which leads to Newton's method. When we seek only the steady solution.7. all of the eigenvalues can be considered to be parasitic. 6.and 3rdorder RungeKutta methods.4.and 4thorder RungeKutta methods.4.98. PROBLEMS 161 the choice of a timemarching method for a steady problem is similar to that for a sti problem. one would like to take the limit h ! 1.and 4thorder AdamsBashforth methods. compare and contrast them with the 3rd. However.4. and the fourthorder RungeKutta method is a candidate for steady convection dominated problems. 8. Repeat the timemarch comparisons for di usion (Section 8.
CHOICE OF TIMEMARCHING METHODS .162 CHAPTER 8.
1. the ODE given by x b Eq. 9. a time derivative is introduced as follows d~ = A~ . which satis es the following coupled system of nonlinear algebraic equations: ~u F (~ ) = 0 (9. Alternatively.4 has a stable steady solution only if A meets the above condition. Following a timedependent path to steady state is possible only if all of the eigenvalues of the matrix A (or . 9. ~ exists as long as A is nonsingular. and choosing timemarching methods. they can be used to solve for the steady solution of Eq. Although the solution ~ = A.2) In the latter case. analyzing.A) have real parts lying in the left halfplane. 1 163 .Chapter 9 RELAXATION METHODS In the past three chapters. we developed a methodology for designing. the unsteady equations are integrated until the solution converges to a steady solution.3) To solve this system using a timemarching method. These methods can be used to compute the timeaccurate solution to linear and nonlinear systems of ODE's in the general form d~ = F (~ t) u ~u (9. ~ x x b (9.4) dt and the system is integrated in time until the transient has decayed to a su ciently low level. The same approach permits a timemarching method to be used to solve a linear system of algebraic equations in the form A~ = ~ x b (9.1) dt which arise after spatial discretization of a PDE.
C~ b = 0 u f Notice that the solution of Eq. If we designate the conditioning matrix by C . at each time step of an implicit timemarching method or at each iteration of Newton's method. our analysis will focus on their application to large sparse linear systems of equations in the form Ab~ . does not depend at all on the eigensystem of the basic matrix Ab .8) which is identical to the solution of Eq. C~ b = A.7 is (9.7 depends crucially on the eigenvalue and eigenvector structure of the matrix CAb. C~ b = A. ~ b u f f b b f 1 1 1 1 1 (9. ~ b = 0 u f (9. exists. In the simplest possible case the conditioning matrix is a diagonal matrix composed of a constant D(b). Using an iterative method.6) b f 1 9.7) ~ = CAb ]. and.164 CHAPTER 9.5 from the left by some nonsingular matrix. Such systems of equations arise.1 Formulation of the Model Problem 9. we seek to obtain rapidly a solution which is arbitrarily close to the exact solution of Eq. equally important. 9. For example.5. 9. we consider iterative methods which are not time accurate at all. and the use of the subscript b will become clear shortly. there are wellknown techniques for accelerating relaxation schemes if the eigenvalues of CAb are all real and of the same sign. the . Such methods are known as relaxation methods.5. ~ b u (9. In this chapter.1. This preconditioning has the e ect of multiplying Eq. 9. for example. which is given by ~ 1 = A. 9.2. While they are applicable to coupled systems of nonlinear algebraic equations in the form of Eq. In the following we will see that our approach to the iterative solution of Eq.5) where Ab is nonsingular. 9.1 Preconditioning the Basic Matrix It is standard practice in applying relaxation procedures to precondition the basic equation. C . 9. the problem becomes one of solving for ~ in u CAb~ . RELAXATION METHODS The common feature of all timemarching methods is that they are at least rstorder accurate. provided C . To use these schemes.
2 0 6 6 1 0 .2 3 2 u 7 6 . this leads to the approximation 82 > 0 1 >6 > 0 6 1 <6 .1 3 7 7 7 7 7 17 5 1 2 0 6 .1 0 6 6 .1 . The result is A = .1.1 .9) The matrix in Eq. consider a spatial discretization of the linear convection equation using centered di erences with a Dirichlet condition on the left side and no constraint on the right side.1 0 76 176 . We then further condition with multiplication by the negative transpose.1 4 1 0 . Using a rstorder backward di erence on the right side (as in Section 3.1 1 1 76 0 76 76 .1 0 >4 : .6).9 has eigenvalues whose imaginary parts are much larger than their real parts. It can rst be conditioned so that the modulus of each element is 1. This is accomplished using a diagonal preconditioning matrix 2 61 60 6 D = 2 x6 0 6 60 4 0 1 0 0 0 0 0 0 1 0 0 0 0 0 1 0 0 0 0 0 1 2 3 7 7 7 7 7 7 5 (9.9. For example.2 1 1 6 0 0 6 = 6 1 0 . A choice of C which ensures this condition is the negative transpose of Ab. FORMULATION OF THE MODEL PROBLEM 165 conditioning matrix C must be chosen such that this requirement is satis ed.11) 1 .2 .1 1 1 ~= 0 u 2 x >6 x 6 >6 .AT A = 2 1 1 2 0 6 .1 32 0 7 6 .10) which scales each row. 9.0 a 7 6 7~ 6 7u+6 0 7 6 17 6 0 5 4 2 0 39 > 7> 7> 7= 7 7> 7> 5> (9.2 4 1 (9.1 54 3 7 7 7 17 7 17 5 1 0 .1 1 1 6 0 6 6 .1 .
. Relaxation methods are applicable to more general matrices. We do not necessarily recommend the use of .2 1 =6 6 6 1 . and the classical relaxation methods can be applied.13) where A is symmetric negative de nite.1.166 1 1 CHAPTER 9. the conditioned matrix .e. as given in Appendix B. symmetric with negative real eigenvalues).AT as a preconditioner we simply wish to show that a broad b range of matrices can be preconditioned into a form suitable for our analysis. Thus even when the basic matrix Ab has nearly imaginary eigenvalues. 2 We use a symmetric negative de nite matrix to simplify certain aspects of our analysis. ~ = 0 f (9.2 . 2 1 6 . We will consider the preconditioned system of equations having the form A~ .2 The Model Equations Preconditioning processes such as those described in the last section allow us to prepare our algebraic equations in advance so that certain eigenstructures are guaranteed. The classical methods will usually converge if Ab is diagonally dominant.2 0 1 7 6 76 176 76 76 054 1 0 .1 which has all negative real eigenvalues. as de ned in Appendix A. RELAXATION METHODS 1 1 If we de ne a permutation matrix P and carry out the process P T . 1 .2 0 1 76 7 6 1 0 . we will thoroughly investigate some simple equations which model these structures.1 0 1 76 76 0 7 6 0 .1 ) just rearranges the rows and columns of a matrix.12) 9. The symbol for the dependent variable has been changed to as a reminder that the physics being modeled is no longer time 2 A permutation matrix (de ned as a matrix with exactly one 1 in each row and column and has the property that P T = P .2 32 32 0 1 0 0 0 0 0 0 1 0 0 0 0 0 1 0 0 1 0 0 17 07 7 07 7 07 5 0 3 1 .2 1 6 1 6 1 .2 1 7 6 54 0 1 1 . In the remainder of this chapter.AT A ]P (which just reorders the elements of A and doesn't change the eigenvalues) we nd 2 60 60 6 60 6 60 4 1 0 0 0 1 0 0 0 0 1 0 0 1 0 0 0 0 7 6 .AT Ab b is nevertheless symmetric negative de nite (i.2 4 3 7 7 7 7 7 17 5 (9.
~ .13.7.5 and 9.15) @ u = g(x) (9. g(x) @t @x This has the steadystate solution 2 2 2 2 (9. FORMULATION OF THE MODEL PROBLEM 1 167 accurate when we later deal with ODE formulations. ~ u u ~ g (9. (bc) g ~ (9.17) dt x ~ where (bc) contains the boundary conditions and ~ contains the values of the source g term at the grid nodes. ~ = A.14) The above was written to treat the general case. A = CAb and ~ = C~ b f f (9. we obtain B (1 .2 1)~ + (bc) . It is instructive in formulating the concepts to consider the special case given by the di usion equation in one dimension with unit di usion coe cient : @u = @ u . If we consider a Dirichlet boundary condition on the left side and f either a Dirichlet or a Neumann condition on the right side.2 1) x ~ fb = ~ . 9.19) ~ where ~ = x fb. Note that the solution of Eq. In the notation of f Eqs. 9.1.9. then A has the form A=B 1~ 1 b 2 2 2 . we nd d~ = 1 B (1 . is guaranteed to exist because A is nonsingular.16) @x which is the onedimensional form of the Poisson equation. In this case 2 Ab = 1 B (1 .2 1)~ = ~ f (9. Introducing the threepoint central di erencing scheme for the second derivative with Dirichlet boundary conditions.18) Choosing C = x I .
1 The Delta Form of an Iterative Scheme +1 We will consider relaxation methods which can be expressed in the following delta form: h i H ~ n . The iteration count is designated by the subscript n or the superscript (n).2 Classical Relaxation 9. or at least easier to solve than the original system. We attempt to do this in such a way. The matrix H is independent of n for stationary methods and is a function of n for nonstationary ones.20) Note that s = .168 CHAPTER 9. RELAXATION METHODS ~ = . ~ = G~ n . ~ 1 = ~ n .2 s]T s = .21 for ~ n gives ~ n = I + H . A]~ n . The converged solution is designated ~ 1 so that ~ 1 = A. ~ f 1 (9.1 is easily obtained from the matrix resulting from the Neumann boundary condition given in Eq.24 using a diagonal conditioning matrix.2 .2 The Converged Solution. and much of the remaining material is devoted to this case.24) Hence it is clear that H should lead to a system of equations which is easy to solve. ~ e f (9. A (9. ~ n = A~ n . A tremendous amount of insight to the basic features of relaxation is gained by an appropriate study of the onedimensional case.25) 1 1 . that it is directly applicable to two.2. The error at the nth iteration is de ned as ~ n ~ n .22) 9. 3.23) where G I + H. ~ f f (9.and threedimensional problems.2 b .2 or . however. H . 9. ~ f (9.2.21) where H is some nonsingular matrix which depends upon the iterative method. A.1 (9. H . 9. the Residual. and the Error +1 +1 1 1 1 Solving Eq.
In all of the above. 9.29 is satis ed using the new values of j and j . 9. The residual at the nth iteration is de ned as ~ n A~ n . Hence the j th row of Eq. The GaussSeidel method is n = 1h n + n . .9. There results the relation between the error and the residual A~ n . 9. This operator comes about by choosing the value of jn such that together with the old values of j. and its eigenvalues.5.27) Finally.29 is satis ed.30) j j j 2 j. The PointJacobi method is expressed in point operator form for the onedimensional case as n = 1 h n + n . ~ n = 0 e r (9. and j .2.2.28) Consequently. and use the de nition in Eq. G is referred to as the basic iteration matrix.2 1)~ = ~ f (9. 9. f i (9. Nonstationary processes in which H (and possibly C ) is varied at each iteration are discussed in Section 9. This operator is a simple extension of the pointJacobi method which uses the most recent update of j.f i (9. ~ r f (9.22.29) as given in Section 9.31) j j j 2 j. the j th row of Eq.26) Multiply Eq. The method of successive overrelaxation (SOR) ( +1) ( ) 1 ( ) +1 ( +1) 1 +1 ( +1) ( +1) 1 ( ) +1 1 1 +1 . and the old value of j . it is not di cult to show that ~ n = G~ n e e +1 (9.3 The Classical Methods Point Operator Schemes in One Dimension Let us consider three classical relaxation procedures for our model equation B (1 .26.25 by A from the left. we have considered only what are usually referred to as stationary processes in which H is constant throughout the iterations. which we designate as m . determine the convergence rate of a method.1. 9. 9.2. CLASSICAL RELAXATION 169 where ~ 1 was de ned in Eq.
32) j ( +1) 1 ( ) +1 ( +1) ( ) ( ) where ! generally lies between 1 and 2.3 The ODE Approach to Classical Relaxation The particular type of delta form given by Eq. ~ ] fb f (9. 9.34) Then the pointJacobi method is obtained with H = .170 CHAPTER 9.1 The Ordinary Di erential Equation Formulation . about which we have already learned a great deal. C A ~ .13 into its diagonal.21 results from the application of the explicit Euler timemarching method (with h = 1) to the following system of ODE's: ~ H d = A~ . h i n = jn + ! ~j .(L + D). which is also easy to solve. and the portion above the diagonal. ! ) n + ! n . jn (9. ~ = H . 9. U . D.D. RELAXATION METHODS is based on the idea that if the correction produced by the GaussSeidel method tends to move the solution toward ~ 1. The GaussSeidel method is obtained with H = .3. It is usually expressed in two steps as h i ~j = 1 jn + jn . ! f (9. then perhaps it would be better to move further in this direction.35) dt This is equivalent to d~ = H . 9. 9.33) j j 2 j. such that A=L+D+U (9. being lower triangular. which certainly meets the criterion that it is easy to solve. fj 2 .21 leads to an interpretation of relaxation methods in terms of solution techniques for coupled rstorder ODE's. One can easily see that Eq. but it can also be written in the single line n = ! n + (1 . 2 j 2 j ( +1) ( +1) 1 ( ) ( ) +1 The General Form The general form of the classical methods is obtained by splitting the matrix A in Eq.36) b dt 1 1 9. ~ f (9. A~ . L. the portion of the matrix below the diagonal.
6.38) which is the solution of Eq.36. Throughout the remaining discussion we will refer to the independent variable t as \time". they are not changed throughout the iteration process. the preconditioning matrix in 9.36 are independent of t. The eigenvalues are xed by the basic matrix in Eq. that is.3. the \time" step. It is clear that. In a stationary method. an iteration process has been decided upon).7.28) ~ n = c ~ (1 + h)n + x 1 1 1 + cm~ m (1 + mh)n + x {z error + cM ~ M (1 + x n M h) } +~ 1 (9. A is nonsingular). 9.13. A are linearly independent. As we shall see. a constant for the entire iteration. and the secondary conditioning matrix in 9. even though no true time accuracy is involved. Hence the numerical solution after n steps of a stationary relaxation method can be expressed as (see Eq. is now referred to in relaxation analysis as the error. We see that the goal of a relaxation method is to remove the transient solution from the general solution in the most e cient way possible. 9. H . ~ f 1 (9. The generalization of this in our approach is to make h. the only free choice remaining to accelerate the removal of the error terms is the choice of h. For this method m = 1 + m h. A have negative real parts (which implies that H . ~ is also independent u f . These are xed by the initial guess. if all of the eigenvalues of H . Suppose the explicit Euler method is used for the time integration. 9. A depends on neither ~ nor t. H and C in Eq. A has been chosen (that is. given by 1 1 1 1 ~ 1 = A. The eigenvalues are xed for a given h by the choice of timemarching method. the three classical methods have all been conditioned by the choice of H to have an optimum h equal to 1 for a stationary iteration process. the solution can be of t.37)  x {z error where what is referred to in timeaccurate analysis as the transient solution.35. then the system of ODE's has a steadystate solution which is approached as t ! 1. and the eigenvectors of H written as ~ = c e 1 t~ + + cM e M t~ M +~ 1 x} (9. THE ODE APPROACH TO CLASSICAL RELAXATION 1 1 1 171 In the special case where H .9. so that we must devise a way to remove them all from the general solution on an equal basis.39) The initial amplitudes of the eigenvectors are given by the magnitudes of the cm . In general it is assumed that any or all of the eigenvectors could have been given an equally \bad" excitation by the initial guess. 1 . Assuming that H .
1: VALUES OF and ! IN EQ.31 and 9.2 1)~ n . With this choice of notation the three methods presented in Section 9.35. ~ f (9.2. 9.2 1)~ n . RELAXATION METHODS 9. We arrive at H (~ n .2.29 into the ODE form 9.40) dt As a start.43 THAT LEAD TO CLASSICAL RELAXATION METHODS ! 0 1 1 Method Equation 6:2:3 6:2:4 6:2:5 1 PointJacobi 1 h i GaussSeidel 2= 1 + sin M + 1 Optimum SOR .B (1 . 1 +1 TABLE 9.B .30. 9.2 ODE Form of the Classical Methods The three iterative procedures de ned by Eqs. let us use for the numerical integration the explicit Euler method 0 (9. However.3 can be identi ed using the entries in Table 9. this is not in the spirit of our study.3.32 obey no apparent pattern except that they are easy to implement in a computer code since all of the data required to update the value of one point are explicitly available at the time of the update.42) +1 +1 It is clear that the best choice of H from the point of view of matrix algebra is .3 is that all the elements above the diagonal (or below the diagonal if the sweeps are from right to left) are zero. since multiplication by the inverse amounts to solving the problem by a direct method without iteration.41) n = n+h n with a step size. ~ f (9.2 1)~ . ~ f (9.2 1) since then multiplication from the left by . equal to 1. ! 0)(~ n .2 1) gives the correct answer in one step. we are led to 2 B (. (1 .1. If we impose this constraint and further restrict ourselves to banded tridiagonals with a single constant in each band. ~ n) = B (1 . Now let us study these methods as subsets of ODE as formulated in Section 9.43) where and ! are arbitrary. h. Insert the model equation 9. 9. Then ~ H d = B (1 .3. ~ n) = B (1 .1.172 CHAPTER 9. The constraint on H that is in keeping with the formulation of the three methods described in Section 9.
The point operator that results from the use of the explicit Euler scheme is ! ! n = ! n + ! (h . ) n . ~ b = 0 u f (9. the three methods are all contained in the following set of ODE's d~ = B . 9. 1) n + !h n . or error.49) where ~ n is the transient. !h f (9. The basic equation to be solved came from some timeaccurate derivation Ab~ . The e f three classical methods. solution of the set of ODE's ~ H d = A~ . and SOR. and ~ 1 A.48) dt This solution has the analytical form ~ n = ~n + ~ 1 e (9. ~ f (9. j 2 j.47) An iterative scheme is developed to nd the converged.2 1)~ . 2 0) B (1 .4. In this light.9. ~ is the steadystate solution.46) This equation is preconditioned in some manner which has the e ect of multiplication by a conditioning matrix C giving A~ . or steadystate. ~ f (9.45) j j.4 Eigensystems of the Classical Methods The ODE approach to relaxation can be summarized as follows. ~ = 0 f (9. our purpose is to examine the whole procedure as a special subset of the theory of ordinary di erential equations. GaussSeidel. However.1. (. (!h . 1 .44 and Table 9. 9. are identi ed for the onedimensional case by Eq. EIGENSYSTEMS OF THE CLASSICAL METHODS 173 The fact that the values in the tables lead to the methods indicated can be veri ed by simple algebraic manipulation.44) dt ! and appear from it in the special case when the explicit Euler method is used for its numerical integration. 2 2 j 2 j 1 ( +1) ( +1) 1 ( ) 1 ( ) ( ) +1 This represents a generalization of the classical relaxation techniques. PointJacobi.
the ODE matrix H .1 since both d and f are zero.51) for the special case 2 0)~ (9.3. 2 The eigensystem can be determined from Appendix B. Note that for h > 1:0 (depending on M ) there is the possibility of instability. To obtain the optimal scheme we wish to minimize the maximum j m j which occurs when h = 1. The eigenvalues are given by the equation m (9. c = 1.30. The three special cases considered below are obtained with a = 1. Fig. 9.2 1)~ m = x m B (.44. . e = 2=!. It is also instructive to inspect the eigenvectors and eigenvalues in the H .4. the asymptotic convergence rate is determined by the eigenvalue m of G ( I + H . the maximum j mj is obtained with m = 1.53) m = . is shown in Fig.31.1 The PointJacobi System 1 If = 0 and ! = 1 in Eq. 9. i.1.2 and for h > 1.50) In this section. The plot for h = 1. Thus Convergence rate j mjmax m=1 2 1 M (9. A reduces to simply B ( 1 .174 CHAPTER 9.1 2 ). 9. B (1 . This amounts to solving the generalized eigenvalue problem A~ m = m H~ m x x (9.32. For h < 1.e. 9. b = . we take M = 5 for the matrix order. To illustrate the behavior. j j = j M j and the best possible convergence rate is achieved: 1 1 j m jmax = cos M + 1 (9. RELAXATION METHODS 1 Given our assumption that the component of the error associated with each eigenvector is equally likely to be excited. and 9.52) ! xm The generalized eigensystem for simple tridigonals is given in Appendix B. 9. 9.54) .2. the optimum stationary case. A matrix for the three methods. and f = 0. A) having maximum absolute value. d = .1 + cos M +1 m = 1 2 ::: M The . we use the ODE analysis to nd the convergence rates of the three classical methods represented by Eqs.2. 9. This relation can be plotted for any h.relation for the explicit Euler method is m = 1 + m h. the maximum j m j is obtained with m = M . Fig. j m j > 1:0. This special case makes the general result quite clear.
EIGENSYSTEMS OF THE CLASSICAL METHODS 1 175 For M = 40. 2 )h]n~ 1 2 2 p 1 . from Eq.57) 4 = .1 + 1 = .1:866 From Eq.9. Again from Appendix B.1:0 = . 3 )h]n~ x 2 1 x + c 1 . (1 . 3=2 7 6 6 6 7 ~ 6 . 3=2 7 5 4 p 5 4 1=2 . 3=2 1 2 3 4 5 (9. 3=2 3 2 p 3 2 3 1= p =2 7 p2 7 6 .0:134 = . A are given by ~ j = (xj )m = sin j m x j = 1 2 ::: M (9. For M = 5.53 1 2 3 = . 9. the numerical solution written in full is 5 p ~ n . we obtain j mjmax = 0:9971.55) M +1 This is a very \wellbehaved" eigensystem with linearly independent eigenvectors and distinct eigenvalues.56) The corresponding eigenvalues are. the eigenvectors can be written as 2 1=2 3 2 3 2 p3=2 3 p 7 p 7 6 3=2 7 6 1 7 6 3=2 7 6 6 0 7 6 ~ = 6 p 1 7 ~ = 6 p 7 ~ = 6 .1 .1. (1 .1 . 1 = .39.23 times its initial level. The rst 5 eigenvectors are simple sine waves. the eigenvectors of H . ~ 1 = c 1 .1:5 2 p (9.0:5 2 = .4. 3=2 7 4 5 4 5 5 4 p 7 1 1=2 .1 7 6 7 x 6 0 7 x 6 7 x 6 6 7 6 6 3=2 7 6 0 7 7 6 .1 = . 9. Thus after 500 iterations the error content associated with each eigenvector is reduced to no more than 0. 23 = . 3=2 7 ~ = 6 p0 7 x = 6 p 7 6 1 7 x 6 6 7 6 3=2 7 7 6 .1 + 23 = .
1 2 0)B (1 .59) 1 which can be studied using the results in Appendix B.0 m=3 1. AGS .0 2. One can show that the H .0 0. h = 1 M = 5.1 2 0)~ m x (9.2 The GaussSeidel System If and ! are equal to 1 in Eq. (. (1 3 4 + c 1 . A matrix for the GaussSeidel method.0 m=5 CHAPTER 9. + c 1 .2 1)~ m = x m B (.4. 9. the matrix eigensystem evolves from the relation B (1 .2 1) = 1 .1: The relation for PointJacobi.2. (1 + 1 )h]n~ 2 x p + c 1 . RELAXATION METHODS h = 1.44.0 Figure 9.58) 9. is AGS B .176 1.0 m=1 Values are equal σm m=4 m=2 Μ=5 0. (1 + 23 )h]n~ x 4 5 )h]n~ x 3 5 (9.
0 h > 1. h = 0:9 M = 5.0 at h De cr ea 1.0 177 1.0 h < 1.2: The relation for PointJacobi. .3: The relation for PointJacobi. Exceeds 1.0 2.9.072 1.0 ng si ea cr In In cr ea si σm ng h h M = 5 0. h = 1:1 M = 5.4.0 λmh Figure 9.0 λ mh ∼ ∼ 1.0 De cr ea h si ng σm ng si Figure 9.0 1.0 0.0 2. EIGENSYSTEMS OF THE CLASSICAL METHODS Approaches 1.0 h 0.0 0.072 / Ustable h > 1.
.. . . The equation for the nondefective eigenvalues in the ODE matrix is (for odd M ) + m ) m = 1 2 ::: M2 1 (9.3=4 1=2 7 2 6 7 6 0 1=8 . is shown in Fig.4. RELAXATION METHODS 2 . . 9.3=4 1=2 7 4 6 7 (9.23. The eigenvectors are quite unlike the PointJacobi set.1 1=2 3 1 6 0 . Furthermore. They are no longer symmetrical. .relation for h = 1.. . The m with the largest amplitude is obtained with m = 1. 7 . 6 .62) M +1 The . producing waves that are higher in amplitude on one side (the updated side) than they are on the other. .1 + cos ( M +1 and the corresponding eigenvectors are given by j. .3=4 1=2 7 3 6 7 6 7 6 0 1=16 1=8 ..60) 6 0 1=32 1=16 1=8 . If M is odd there are (M + 1)=2 distinct eigenvalues with corresponding linearly independent eigenvectors. . For M = 40.3=4 1=2 7 5 6 7 6 . . . A X =J = 6 7 6 X GS GS 1 6 7 6 6 ~ 77 6 6 1 77 4 4 55 ~ 1 1 2 3 3 3 (9. m + ~ m = cos m x sin j M + 1 m = 1 2 ::: M2 1 (9. the error associated with the \worst" eigenvector is removed in half as many iterations. Hence the convergence rate is 2 1 CHAPTER 9.64) . and there are (M . The Jordan canonical form for M = 5 is j m jmax = cos M + 1 2 (9. 7 .. j m jmax = 0:9942. . 1)=2 defective eigenvalues with corresponding 178 Since this is the square of that obtained for the PointJacobi method. .. the optimum stationary case. 250 iterations are required to reduce the error component of the worst eigenvector by a factor of roughly 0. they do not represent a common family for di erent values of M . .61) m = . . 4 5 0 1=2M M The eigenvector structure of the GaussSeidel ODE matrix is quite interesting.principal vectors.63) 3 2 h~ i 7 6 h~ i 7 6 6 7 2~ 37 .
3=32 1 2 3 4 5 The corresponding eigenvalues are = .0 0.1 (4) Defective. 34 )n~ 1 2 3 3 1 1 2 2 (9. 3=16 7 7 607 6 0 7 6 .0 σm Μ=5 2 defective λ m 0.4: The relation for GaussSeidel.65) 6 7 x 6 7 x 6 7 x 6 7 x 6 7 x 6 6 6 7 6 7 6 7 6 9=16 7 7 6 .0 1. linked to (5) Jordan block The numerical solution written in full is thus ~ n . The eigenvectors and principal vectors are all real.66) .1=4 = .0 2. EIGENSYSTEMS OF THE CLASSICAL METHODS 1.1 7 ~ = 6 4 7 (9. h )n~ 4 x h x + c (1 .4 7 4 5 4 p 5 4 5 4 5 4 5 9=32 0 0 1 . ~ 1 = c (1 . For M = 5 they can be written 2 3 2 p 3 2 3 2 3 2 3 3=2 1=2 7 17 0 7 6 3=4 7 6 p3=4 7 607 6 2 7 6 0 7 6 6 7 6 6 6 0 7 ~ = 6 3=4 7 ~ = 6 p0 7 ~ = 6 0 7 ~ = 6 .0 179 h = 1.0 λmh Figure 9. h = 1:0 M = 5.4.3=4 ) = .9.
The generalization to any M is fairly clear.68) 7 7 5 Eigenvalues of the system are given by m + = .4. h)n. 2x + x x . 2x + x x 4 4 3 4 3 4 4 5 2 3 2 3 4 3 4 4 2 2 3 2 3 4 3 4 2 2 3 2 3 4 3 . 9. 2x 4 .2.2x + x x x 2x2x x 6 x + . + c h n(n . !) + ! p = 0 ! is optimum for the stationary case. If 4(1 . and the following conditions hold: 1. The H . is 1 1 2 x 0 0 6 . 2.1 x 0)B (1 . ASOR B . 2x 0 0 0 x 4 4 3 7 7 7 7 (9. Two eigenvalues are real.2 1).44.2x + x x . . ~ x 3 4 1 5 2 2 3 4 5 + c (1 . One can show that this can be written in the form given below for M = 5. 2x + x x . h)n + c h n (1 . If ! is chosen such that 4(1 .1 x 0)B (1 .2x x . !) + ! pm < 0 zm and m are complex.2 1). If M is odd. h)n~ x 5 1! 1 4 5 (9. RELAXATION METHODS " # n + c h n (1 .1 + !pm 2 zm 2 m = 1 2 :::M 2 2 1 2 (9. h)n. If M is even. h) x 1! 2! + c (1 . 3. A matrix for the SOR method. equal and defective. 2x + x x . one of the remaining eigenvalues is real and the others are complex occurring in conjugate pairs.3 The SOR System 1 If = 1 and 2=! = x in Eq. !) + ! pm] = pm = cos m =(M + 1)] 2 2 2 1 If ! = 1.67) 9. + x .180 CHAPTER 9.69) where zm = 4(1 . the remaining eigenvalues are complex and occur in conjugate pairs.2 + x 1 . h)n. the ODE matrix is B . the system is GaussSeidel. ~ + c (1 . x 0 6 1 6 . 1) (1 . (. 2x x x 6 . (. 2x + x 6 x .
relation reduces to that shown in Fig. j mjmax = 0:8578.6 7 1 7 p p )=2 7 6 1=2 7 6 6 3(1 i 2)=6 7 6 6 6 6 7~ 6 9 7~ 6 7 6 7~ 6 0 7 ~ = 6 1=3 7 x = 6 16 7 x = 6 p 7 x = 6 1=3 7 (9. much faster than both GaussSeidel and PointJacobi.4=3 (9.74) 2 1 2 34 5 1 1 3 4 5 !opt p + i p .2=3 (2) Defectivep linked to = .1 j . 1.70) !opt = 2= 1 + sin M + 1 ~m x m = = m. m = 1 . This illustrates the fact that for optimum stationary SOR all the j m j are identical and equal to !opt . For M = 5 they can be written 2 3 2 3 2 p3(1 3 2 3 1=2 7 . For odd M . 1 (9. In practical applications.71) Using the explicit Euler method to integrate the ODE's. the optimum value for the stationary case. and if h = 1.73) 6 7 0p x 6 7 6 7 6 6 6 1=6 7 6 13 7 6 3(5 i 2)=54 7 7 6 0 7 7 4 5 4 5 4p 5 4 5 p 1=18 6 1=9 3(7 4i 2)=162 The corresponding eigenvalues are = . 9. 2p2i)=9 = .5. The remaining linearly independent eigenvectors are all complex.9. the .23 times its initial value in only 10 iterations. there are two real eigenvectors and one real principal vector.(10 + 2 2i)=9 = . the optimum value of ! may have to be determined by trial and error. Hence the worst error component is reduced to less than 0. and the bene t may not be as great. EIGENSYSTEMS OF THE CLASSICAL METHODS One can easily show that the optimum ! for the stationary case is and for ! = !opt 181 (9. Hence the convergence rate is j m jmax = !opt .72) !opt = 2= 1 + sin M + 1 For M = 40.4.(10 . h + h m .1 m sin 2 where m j M +1 (9. p m= m 2 m 2 1 2 q .
5: The relation for optimum stationary SOR. 2p2i)h=9]n~ x n~ c 1 . however. 2h=3)n~ x p c 1 . In our ODE b f approach this could also be considered and would lead to a study of equations with nonconstant coe cients.0 2. 2h=3)n + c nh(1 .5 Nonstationary Processes In classical terminology a method is said to be nonstationary if the conditioning matrices. 9.0 λmh Figure 9.0 1. 2h=3)n. (10 . This process changes the PointJacobi method to Richardson's method in standard terminology. to study the case of xed H and C but variable step size. ~1 = + + + + c (1 .182 1.39 is 1 . RELAXATION METHODS h = 1.0 2 real defective λ m σm 2 complex λ m 1 real λ m Μ=5 0. but it can greatly a ect the convergence rate. (10 + 2 2i)h=9] x c (1 . 4h=3)n~ x 1 2 1 2 3 2 3 4 5 4 5 1 (9. h.0 0.0 CHAPTER 9.75) 9. H and C . This does not change the steadystate solution A. ]~ x c (1 . ~ b . M = 5. It is much simpler. are varied at each time step. The numerical solution written in full is ~n . For the GaussSeidel and SOR methods it leads to processes that can be superior to the stationary methods. The nonstationary form of Eq. h = 1.
76.76) where the symbol stands for product. NONSTATIONARY PROCESSES N ~ N = c ~ Y (1 + hn ) + x 1 1 183 + cm~ m x M hn ) + ~ 1 N Y n=1 n=1 1 (1 + m hn ) + + cM ~ M x N Y n=1 (1 + (9. We will see that a few well chosen h's can reduce whole clusters of eigenvectors associated with nearby 's in the m spectrum.5. for m = 1 2 M . the error term can theoretically be completely eliminated in M steps by taking hm = . we seek b max a j(Pe)N ( )j = minimum with(Pe)N (0) = 1 (9.1= m . This leads to the concept of selectively annihilating clusters of eigenvectors from the error terms as part of a total iteration process.80) . Now focus attention on the polynomial (Pe)N ( ) = (1 + h )(1 + h ) 1 2 (1 + hN ) (9. Let us consider the very important case when all of the m are real and negative (remember that they arise from a conditioned matrix so this constraint is not unrealistic for quite practical cases).9. Consider one of the error terms taken from M N ~ N ~ N .1= m for m = 1 2 : : : M . 9. Now we pose a less demanding problem. the eigenvalues m are generally unknown and costly to compute. Since hn can now be changed at each step. However. we considered making the error exactly zero by taking advantage of some knowledge about the discrete values of m for a particular case.79) treating it as a continuous function of the independent variable . This is the basis for the multigrid methods discussed in Chapter 10. In the annihilation process mentioned after Eq. Let us choose the hn so that the maximum value of (Pe)N ( ) is as small as possible for all lying between a and b such that b a 0.77) and write it in the form cm~ m Pe( m ) cm~ m x x N Y n=1 (1 + m hn ) (9. Mathematically stated.78) where Pe signi es an \Euler" polynomial. It is therefore unnecessary and impractical to set hm = . ~ 1 = X cm~ m Y (1 + e x m=1 n=1 m hn ) (9.
84) b a h 2 b a 2N n Remember that all are negative real numbers representing the magnitudes of m in an eigenvalue spectrum. b a b This problem has a well known solution due to Markov.82) q q N N 1 TN (y) = 2 y + y .76 is expressed in terms of a set of eigenvectors. cos (2n . b TN a. ampli ed by the coe cients cm Q(1 + m hn).1 using only 3 iterations. TN . 9. + ( . RELAXATION METHODS a. ~ m .1 y 1 and 2 2 (9. It is 2 . ) cos (2n . Eq. a. 1) 6 " #! (9. and it leads to the nonstationary step size choice given by ( " #) 1 = 1 .84. In relaxation terminology this is generally referred to as Richardson's method. ! (Pe)N ( ) = where (9. As an example for the use of Eq. y .184 CHAPTER 9. The error in the relaxation process represented by Eq.84 gives us the best choice of a series of hn that will minimize the amplitude of the error carried in the eigenvectors associated with the eigenvalues between b and a . . b ! . let us consider the following problem: Minimize the maximum error associated with the eigenvalues in the interval . 1) n = 1 2 :::N (9.85) hn = 2= 3 . 9. 1 (9.86) .2 . 9. With each x eigenvector there is a corresponding eigenvalue.83) 2 are the Chebyshev polynomials for jyj > 1.81) TN (y) = cos(N arccos y) are the Chebyshev polynomials along the interval . The three values of h which satisfy this problem are (9. 1 + 1 y .
9. 9. Pt: 0 1 1 N B 1 + hn C Y (Pt )N ( ) = B 2 C (9. This was derived from Eq.9. 2 hn under the same constraints as before.89) would result.92) .90) b max a j(Pt)N ( )j = minimum .87 is given in Fig.6 are h = 4=(6 . Eq. NONSTATIONARY PROCESSES and the amplitude of the eigenvector is reduced to (Pe) ( ) = T (2 + 3)=T (3) 3 3 3 185 (9.87) (9. 8] =2 99 3 3 3 A plot of Eq.41. If instead the implicit trapezoidal rule n+1 = n+ is used.2 . 9. 9. 9. The values of h used in Fig. namely that =1 0 1 M N 1+ h ~ N = X cm~ m Y B 2 n B x @ 1 m n 1 . the nonstationary formula 1 h( 0 + 0 ) n 2 n +1 (9.91) @ 1 A n 1 .37 on the condition that the explicit Euler method.6 and we see that the amplitudes of all the eigenvectors associated with the eigenvalues in the range .1 have been reduced to less than about 1% of their initial values. 3) h = 4=(6 . 9.76. p 0) h = 4=(6 + 3) 1 2 3 p Return now to Eq.5.88) where n p p o T (3) = 3 + 8] + 3 . 2 hn =1 =1 1 mC C + ~1 A m (9. This calls for a study of the rational \trapezoidal" polynomial. with (Pt)N (0) = 1 (9. was used to integrate the basic ODE's.
4 −0.6 −1.6: Richardson's method for 3 steps.6 −0.2 .2 0.01 −0.01 0.005 (P ) (λ) e 3 0 −0.6 −1.1 0 −2 −1.2 −1 λ −0.4 0.8 −1.8 −1.186 CHAPTER 9.8 −0. minimization over .2 −1 λ −0.2 0 0.7 0.4 −1. RELAXATION METHODS 1 0.4 −0.8 −0.5 0.6 (Pe )3 (λ) 0. .4 −1.015 0.02 −2 −1.8 0.015 −0.005 −0.2 0 Figure 9.6 −0.1.02 0.9 0.3 0.
7. x x not just the expressions. The error amplitude is about 1/5 of that found for (Pe) ( ) in the same interval of . 1 1 1 . A )xn + b x (I + A )xn = (I . (You do not have to nd H . PROBLEMS ! 187 The optimum values of h can also be found for this problem. A for the SOR method with A = B (4 : 1 . but we settle here for the approximation suggested by Wachspress n.) Find the numerical values.6. The results for (Pt) ( ) are shown in Fig. For a linear system of the form (A + A )x = b. 2 =. A )~ + b x where is a parameter. = N . Gn)A. 2.2 1) and ! = !opt. Show that this iterative method can be written in the form H (xk . Recall that the eigenvalues of H . a n=1 2 N (9.93) b hn b This process is also applied to problem 9.9. 9. nd the eigenvalues of H . b 1 2 2 +1 1 +1 1 2 Determine the iteration matrix G if = . . Then nd the corresponding j m j values. A. 3.6 Problems 1. f show that 1 ~n = Gn ~ + (I . A satisfy A~ m = mH~ m . f ~ 0 1 1 2 where G = I + H . 9.7 are h = 1 p h = 2 h = 2 ( 1) ( 1) 3 3 1 2 3 9. consider the iterative method (I + A )~ = (I .1=2. xk ) = (A + A )xk .2. Using Appendix B. Given a relaxation method in the form ~ H ~n = A~n .85. The values of h used in Fig.
5 −1 −1.6 −0.5 0.7 0.2 0 2 x 10 −3 1.2 −1 λ −0.4 −1.188 CHAPTER 9.6 −0.8 −0.2 −1 λ −0.6 −1.4 −0.1.4 −1.5 1 0.6 −1.1 0 −2 −1.3 0. minimization over .8 −0.2 0.4 0.8 −1. RELAXATION METHODS 1 0. .8 0.2 .9 0.4 −0.7: Wachspress method for 3 steps.2 0 Figure 9.8 −1.5 (Pt )3 (λ) 0 −0.5 −2 −2 −1.6 (P ) (λ) t 3 0.
5. Plot the solution after the residual is reduced by 2. and 4 orders of magnitude. (b) the GaussSeidel method. Determine the asymptotic convergence rate. the number of iterations. use u(x) = 0.6. Use secondorder centered di erences on a grid with 40 cells (M = 39). (c) the SOR method with the optimum value of !. Solve the following equation on the domain 0 x 1 with boundary conditions u(0) = 0. 6x = 0 @x For the initial condition. PROBLEMS 189 4. Compare with the theoretical asymptotic convergence rate. Iterate to steady state using (a) the pointJacobi method. 3. and (d) the 3step Richardson method derived in Section 9.9. Plot the logarithm of the L norm of the residual vs. 2 2 2 . u(1) = 1: @ u .
RELAXATION METHODS .190 CHAPTER 9.
2 1). There are many variations of the process and many viewpoints of the underlying theory. respectively. The viewpoint presented here is a natural extension of the concepts discussed in Chapter 9. The eigenvalues and eigenvectors are given in Sections 4. Notice that as the magnitudes of the eigenvalues increase. That is.1.m sin(mx) (10.3.3.Chapter 10 MULTIGRID The idea of systematically using sets of coarser grids to accelerate the convergence of iterative schemes that arise from the numerical solution to partial di erential equations was made popular by the work of Brandt.2 and 4. if the eigenvalues are ordered such that j j j j 1 2 2 j Mj (10. 10.3.1) then the corresponding eigenvectors are ordered from low to high space frequencies. ~ (10.1 Eigenvector and Eigenvalue Identi cation with Space Frequencies Consider the eigensystem of the model matrix B (1 .2 1)~ = X 1 D(~ ) X .3) xx~ = x x 191 2 2 2 2 1 . Note that @ sin(mx) = .1 Motivation 10.2) @x and recall that 1 B (1 . This has a rational explanation from the origin of the banded matrix. the spacefrequency (or wavenumber) of the corresponding eigenvectors also increase.
the correlation of large magnitudes of m with high spacefrequencies is to be expected for these particular matrix operators. the complete counterexample of the above association is contained in the eigensystem 1 for B ( 2 1 1 ). a sine synthesis. 2 2 2 2 2 10. .5. One nds that 1 M + 1 . 10.m . of the Richardson method described in Section 9. However. This is to be expected of an iterative method which is time accurate. by design.1. the property can be restored by using h < 1. error components corresponding to high space frequencies will be reduced more quickly than those corresponding to low space frequencies. exactly the opposite 2 behavior. We have seen that X .2 Properties of the Iterative Method The second key motivation for multigrid is the following: Many iterative methods reduce error components corresponding to eigenvalues of large amplitude more e ectively than those corresponding to eigenvalues of small amplitude.m m << M (10. from Appendix B.2. As we saw in Section 9. However.2 The Basic Process First of all we assume that the di erence equations representing the basic partial di erential equations are in a form that can be related to a matrix which has certain . 9. the operation 1 D(~ ) represents the numerical approximation of the multiplication of the x appropriate sine wave by the negative square of its wavenumber.1. Therefore. This is consistent with the physics of di usion as well. and X ~ . this method produces the same value of j j for min and max . The classical pointJacobi method does not share this property. for example. It is also true.2 + 2 cos m . For this matrix one nds.4) m= M +1 x Hence. this correlation is not necessary in general. of the GaussSeidel method and. This is the key concept underlying the multigrid process. ~ represents a sine transform. MULTIGRID where D(~ ) is a diagonal matrix containing the eigenvalues. as shown in Fig. In fact. When an iterative method with this property is applied to a matrix with the above correlation between the modulus of the eigenvalues and the space frequency of the eigenvectors.4.192 1 CHAPTER 10.
the vector ~ b represents the boundary conditions and the forcing function.6 is used.11. m . ~ b] r f (10. we are led to the starting formulation C Ab ~ 1 . In Eq. 3. 9. 10. and initial guess for ~ 1 and proceed through n iterations making use of some iterative process that satis es property 5 above.5. The eigenvectors associated with the eigenvalues having largest magnitudes can be correlated with high frequencies on the di erencing mesh. or it can be \contrived" by further conditioning. ~ b ] = 0 f (10. 5. ~ = 0 e r (10. 4. This form can be arrived at \naturally" by simply replacing the derivatives in the PDE with di erence schemes.27. 2. as in the example given by Eq. as in the examples given by Eq. These conditions are su cient to ensure the validity of the process described next. The iterative procedure used greatly reduces the amplitudes of the eigenvectors 1 associated with eigenvalues in the range between 2 j jmax and j jmax.5) where the matrix formed by the product CAb has the properties given above.8) . The eigenvalues. THE BASIC PROCESS 193 basic properties. The m are fairly evenly distributed between their maximum and minimum values. 9. r where ~ = C Ab~ .2. We start with some any.6) Recall that the ~ used to compute ~ is composed of the exact solution ~ 1 and the r ~ in such a way that error e A~ .10. The basic assumptions required for our description of the multigrid process are: 1. of the matrix are all real and negative. the residual. 3. if f ~ 1 is a vector representing the desired exact solution. but for clarity we suppose that the threestep Richardson method illustrated in Fig. Having preconditioned (if necessary) the basic nite di erencing scheme by a procedure equivalent to the multiplication by a matrix C . The problem is linear. We do not attempt to develop an optimum procedure here. At the end of the three steps we nd ~ .7) where A CAb (10.
e In addition. Next we construct a permutation matrix which separates a vector into two parts.9) Thus our goal now is to solve for ~ .12) 1 The operation PAP .~ e CHAPTER 10. partitions the A matrix to form 1 2 3 A 1 A2 " ~ # " ~ # 6 7 ee = re 4 5 ~ ~o r A 3 A4 e o (10.14) . we can be sure that the high frequency content of ~ has been greatly reduced (about 1% or less of its original value in the initial guess). 10. MULTIGRID (10. and the eigenvalues of the Richardson process in the form: M= M ~ = X cm~ m Y ( m hn)] + X cm~ m Y ( mhn)] e x x 2 3 3 m=1 n=1 m=M=2+1  very low amplitude {z n=1 (10. since a permutation matrix has an inverse which is its transpose.7 for ~ then e ~1 = ~ .194 If one could solve Eq. P ]~ = P ~ e r (10. assumption 4 ensures that the error has been smoothed.10) } Combining our basic assumptions.13) Notice that e r e A ~e + A ~o = ~e 1 2 (10. For a 7point example 2 6 6 6 6 6 6 6 6 6 6 6 4 e e e e e e e 2 4 6 1 3 5 7 3 2 7 6 7 6 7 6 7 6 7 6 7=6 7 6 7 6 7 6 7 6 7 6 5 4 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 76 0 76 76 0 76 76 0 76 76 76 0 76 76 0 76 54 1 32 e e e e e e e 1 2 3 4 5 6 7 3 7 7 7 7 7 7 7 7 7 7 7 5 "~ # ee ~ ~o = P e e (10. 10. we can write PA P .7 from the left by P and. one containing the odd entries. We can write the exact solution for ~ in terms of e e the eigenvectors of A. and the other the even entries of the original vector (or any other appropriate sorting which is consistent with the interpolation approximation to be discussed below).11) Multiply Eq.
in this formulation. e no operations on ~ are required or justi ed. THE BASIC PROCESS 195 It is important to notice that ea and eb represent errors on the boundaries where the error is zero if the boundary conditions are given. It is also important to notice that we are dealing with the relation between ~ and ~ so the original boundary conditions and e r forcing function (which are contained in ~ in the basic formulation) no longer appear f in the problem. It is that there is some connection between ~ e and ~ o brought about by the smoothing property of the Richardson e e relaxation procedure.2.16) 0 0 1 With this approximation Eq. At this point we make our one crucial assumption. 10.15) e 2 1 (e + e ) e b 2 1 2 3 2 4 5 4 6 2 7 6 (10.14 reduces to A ~ e + A A0 ~ e = ~ e e e r 1 2 2 (10.19) or Ac~ e . Since the top half of the frequency spectrum has been removed.10. r If the boundary conditions are Dirichlet.17) (10. notice that. Hence. it is reasonable to suppose that the odd points are the average of the even points. For example 1 (e + e ) e 2 a 1 (e + e ) e 2 1 (e + e ) or ~ o = A0 ~ e e e (10. Finally. no aliasing of these functions can occur in subsequent steps. the averaging of ~ is our only approximation. ea and eb are zero.18) (10. ~ e = 0 e r where Ac = A + A A0 ] 1 2 2 . and one can write for the example case 2 1 0 0 16 1 1 0 A02 = 2 6 0 1 1 6 4 3 7 7 7 5 is an exact expression.
2 .55 for Dirichlet conditions.1 are met. It is easy to show that the high spacefrequencies still correspond to the eigenvalues with high magnitudes.1 { 3 7 (10.23) 4 5 0 0 2 1 2 .15 must be changed. 10.18 gives z } 2 . the eigenvector structure is di erent from that given in Eq.1 = 6 PAP 6 1 6 6 6 1 6 6 4 .2 3 7 7 7 17 2A A 3 7 7 6 1 27 7=4 7 5 7 7 A3 A4 7 7 7 5 (10. When eb = eM .2 6 6 6 6 6 6 . All of them go through zero on the left (Dirichlet) side.2 6 . In the particular case illustrated in Fig.2 4 A1 . However.1. 10.16 changes to 2 1 0 03 7 16 A0 = 2 6 1 1 0 7 60 1 17 (10. 10. For Neumann conditions. all of the properties given in Section 10.20) and Eq. If the original A had been B (7 : 1 .2 .1]T .2 { z } 3 2 1 1 7+6 1 1 5 4 A2 1 1 { 3 7 5 2 z } 2 1 16 1 1 6 4 26 1 1 A 0 1 { Ac 3 2 } z . 10. A in the 1D model equation is B (1 ~ 1) where ~ = . is completely determined by the choice of the permutation matrix and the interpolation approximation. in fact. the example in Eq. B.2 1). 1) xjm = sin j 2M + 1 m=1 2 M (10. If Neumann conditions are on the left. The eigensystem is given by b b Eq.1 1=2 7 5 1=2 .1 1=2 7 6 7 = 4 1=2 .196 CHAPTER 10. MULTIGRID The form of Ac.2 .2 1 1 1 1 1 1 . our 7point example would produce 2 . and.2 ::: . 9.19.1.2 1 1 1 . ea = e . In the present case they are given by " !# (2m .21) 5 If the boundary conditions are mixed DirichletNeumann. the matrix on the coarse mesh. the interpolation formula in Eq. 10.2 .22) and are illustrated in Fig.2 . and all of them re ect on the right (Neumann) side. eb is equal to eM .
Therefore.2.2 4 A1 . At this stage.25) For A = B (M : 1 . The permutation matrix remains the same and both A and A in the partitioned matrix PAP .2 6 .2 1)~ = ~ on the e r 1 B (1 ." We will continue with Dirichlet boundary conditions for the remainder of this Section. Given ~ e e e computed on the coarse grid (possibly using even coarser grids). the relationship between e e In order to examine this relationship.1 7 5 1=2 .1: Eigenvectors for the mixed Dirichlet{Neumann case. (10. THE BASIC PROCESS 197 X Figure 10. and thus ~ . we need to consider the eigensystems of A and Ac: A = X X. we have reduced the problem from B (1 .2 { 2 } 3 z 1 1 7+6 1 1 5 4 A2 1 1 { 3 7 5 2 z } 2 1 61 1 16 4 26 1 1 A 0 2 { Ac 3 2 } z . Ac = Xc cXc. are unchanged (only A is modi ed by putting .15.10.9. 10.2 1) the eigenvalues and eigenvectors are m j=1 2 M (10. we can construct the coarse matrix from 1 1 2 4 } z 2 . which will provide us with the solution ~ 1 using Eq. Recall that our goal is ne mesh to 2 e r to solve for ~ . cos m=1 2 M M +1 M +1 1 1 . we can compute ~ o e using Eq.26) ~ m = sin j m x m = .1 in the lower right element). we must now determine e ~ e and ~ .1 1=2 7 6 7 = 4 1=2 .2 1)~e = ~ e on the next coarser mesh. 10.1=2 { 3 1=2 7 (10.2 1 . In order to complete the process.24) 5 which gives us what we might have \expected.
2 1 . one can easily see that (~ c) coincides with ~ at x x every second point of the latter vector. (~ c) e x c r c (10. 5 0 (10. ( c) approaches 2 . the most di cult error mode to eliminate is that with m = 1. The eigenvalue and eigenvector corresponding to m = 1 for the matrix Ac = 1 B (Mc 1 ..27) M +1 For example. In addition. 6 .31) 2 3 1 607 6 7 = Xc . cos M2+ 1 (~ c) = sin j M2+ 1 j = 1 2 x Mc (10. with M = 51. 1)=2 is the size of Ac. we obtain ( c) = . The exact solution on the coarse grid x x satis es ~ e = A.30) since (~ c) contains the even elements of ~ . ~ = ~ .2 1) are 2 ( c) = . 1 . As M increases. then Mc = (M . that is. i. it contains the even elements of ~ . Then the residual is x e x ~ = A~ = ~ r x x (10.198 CHAPTER 10. MULTIGRID Based on our assumptions.32) 2 3 1=( c) 6 0 7 6 7 = Xc 6 .e.28) For M = 51 (Mc = 25). 7 6 .29) 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 and the residual on the coarse grid is ~ e = (~ c) r x 1 1 1 1 (10. 7 c 6 7 4 . Xc. corresponding to ~ = sin j = . ~ e = Xc . 7 4 5 0 (10.. cos M + 1 x j=1 2 M (10. x Now let us consider the case in which all of the error consists of the eigenvector component ~ ..33) 1 1 1 1 1 1 1 1 1 . = . If we restrict our attention to odd M .0:003649.0:007291 = 1:998 .
which gives Ab = x B (M : 1 .36) 1 B (M : 1 . in each case the appropriate permutation matrix and the interpolation approximation de ne both the down. quite important but they are not discussed here. However. in addition to interpolating ~ e to the ne grid e x e (using Eq. This is equivalent to solving or 1A ~ = ~ e r 2 ce e (10. THE BASIC PROCESS = ( c) 1 1 199 (~ c) x 1 1 1 (~ ) 2 xc 1 (10.10. 10. x C x B (Mc : 1 .38) C= x I 2 (10. obviously. the matrix A = B (M : 1 .2 1) 4 c c 2 2 2 (10. and usually is.2 1) = 1 B (Mc : 1 . 10. The remaining steps required to complete an entire multigrid process are relatively straightforward. The details of nding optimum techniques are. since xc = 2 x.2 1) comes from a discretization of the di usion equation. Thus we see that the process is recursive.39) Applying the discretization on the coarse grid with the same preconditioning matrix as used on the ne grid gives.37.37) c 4 In our case. carried to even coarser grids before returning to the nest level.2. we must multiply the result by 2. but they vary depending on the problem and the user.2 1)~ = ~ ee re (10.and upgoing paths.2 1) and the preconditioning matrix C is simply 2 (10.35) Since our goal is to compute ~ = ~ .2 1) = x B (Mc : 1 . The reduction can be.15). .40) which is precisely the matrix appearing in Eq.34) (10. The problem to be solved on the coarse grid is the same as that solved on the ne grid.
f . 9. Perform n iterations of the selected relaxation method on the ne grid.45) 4 5 0 0 0 0 0 1 0 that is. f ~ (10. MULTIGRID ~ We now describe a twogrid process for the linear problem A ~ = f . 1. A (10. Gn1 ) A.200 10.11. f (10. f = AGn1 ~n + A (I . Solve the problem A ~ = ~ on the coarse grid exactly: e r ~ = A. Next compute the residual based on ~ : ~ ~ ~ ~ = A ~ . Call the result ~ . starting with ~ = ~n. Note that the coarse grid matrix denoted A2 here was denoted Ac in the preceding section." Some form of weighted restriction can also be used.44) In our example in the preceding section. the restriction matrix is 2 3 0 1 0 0 0 0 0 R =6 0 0 0 1 0 0 0 7 (10. Transfer (or restrict) ~ to the coarse grid: r ~ =R ~ r r (10. ~ e r 2 (2) (2) (2) 2 1 (2) 2 (10.41) 1 (1) 1 (1) 1 1 1 where G = I + H.46) 1 2 See problem 1 of Chapter 9..g. which can be easily generalized to a process with an arbitrary number of grids due to the recursive nature of multigrid.21). Eq. the rst three rows of the permutation matrix P in Eq. AGn1 A. . This gives ~ = Gn1 ~n + (I . This type of restriction is known as \simple injection.42) and H is de ned as in Chapter 9 (e.3 A TwoGrid Process CHAPTER 10. f r ~ = AGn1 ~n . Gn1 ) A. Extension to nonlinear problems requires that both the solution and the residual be transferred to the coarse grid in a process known as full approximation storage multigrid.43) 1 1 1 1 1 (1) (1) (1) 1 1 1 1 1 1 2. (1) (2) 2 (1) 1 2 1 3. 10.
f ~ ~ +1 1 2 2 1 2 1 1 1 2 2 1 2 1 1 1 1 (10. If this is the coarsest grid in the sequence. . solve exactly. 4. f + A. I A. In the 1 preceding example (eq. where 2 2 3 2 G = I . R A ]Gn2 2 3 2 3 3 1 3 2 2 2 2 2 3 3 2 3 (10. 10. I ~ e +1 (1) 1 (2) 2 (10. . Extension to four or more grids proceeds in similar fashion. R A]Gn1 A.49) (10. 10. I . with (I . I A. R A]Gn1 ~ n . The basic iteration matrix for a threegrid process is found from Eq. the prolongation matrix is 2 6 6 6 6 6 1 I2 = 6 6 6 6 6 6 4 1=2 1 1=2 0 0 0 0 0 0 1=2 1 1=2 0 0 0 0 0 0 1=2 1 1=2 (10. It is at this stage that the generalization to a multigrid procedure with more than two grids occurs. Combining these steps. A TWOGRID PROCESS 2 2 201 Here A can be formed by applying the discretization on the coarse grid. Otherwise. Transfer (or prolong) the error back to the ne grid and update the solution: ~n = ~ . R A]Gn1 1 2 2 1 2 1 1 1 1 The eigenvalues of this matrix determine the convergence rate of the twogrid process.2 1).50 by replacing A.3.50) Thus the basic iteration matrix is I .47) 3 7 7 7 7 7 7 7 7 7 7 7 5 In our example.40). I and R are the transfer operators between grids 2 and 3.51) In this expression n is the number of relaxation steps on grid 2.15. I A.48) which follows from Eq. apply the twogrid process recursively. one obtains ~n = I . G )A. 10. A = 4 B (Mc : 1 . I A. and A is obtained by discretizing on grid 3.10.
2 . and 4 orders of magnitude.51. 2. Plot the logarithm of the L norm of the residual vs. Repeat problem 4 of Chapter 9 using a fourgrid multigrid method together with (a) the GaussSeidel method. Plot the solution after the residual is reduced by 2. Solve exactly on the coarsest grid. Calculate the theoretical asymptotic convergence rate and compare.5. 10. MULTIGRID 1.4 Problems CHAPTER 10.202 10. (b) the 3step Richardson method derived in Section 9. the number of iterations. 3. Derive Eq. Determine the asymptotic convergence rate.
we have emphasized the secondorder centereddi erence approximations to the spatial derivatives in our model equations. for example. We have seen that a centered approximation to a rst derivative is nondissipative. leading. In this Chapter. we discuss some di erent ways of adding numerical dissipation to the spatial derivatives in the linear convection equation and hyperbolic systems of PDE's. this can be insu cient. especially at high Reynolds numbers. it must be carefully controlled such that the error introduced is not excessive. to the production of shock waves. such as the Euler and NavierStokes equations. In processes governed by nonlinear equations. there can be a continual production of highfrequency components of the solution. we have neglected viscous e ects.. However. the production of high frequencies is eventually limited by viscosity. when we solve the Euler equations numerically. some form of added numerical dissipation is required in the numerical solution of the NavierStokes equations as well.e. due to the limited grid resolution which is practical. Thus the numerical approximation must contain some inherent dissipation to limit the production of highfrequency modes. Therefore. i.Chapter 11 NUMERICAL DISSIPATION Up to this point. Since the addition of numerical dissipation is tantamount to intentionally introducing nonphysical behavior. unless the relevant length scales are resolved. 203 . Although numerical approximations to the NavierStokes equations contain dissipation through the viscous terms. In a real physical problem. the eigenvalues of the associated circulant matrix (with periodic boundary conditions) are pure imaginary.
For periodic boundary conditions the corresponding matrix operator is .uj. The antisymmetric component is the secondorder centered di erence operator.a x m 1 .1 OneSided FirstDerivative Space Di erencing We investigate the properties of onesided spatial di erence operators in the context of the biconvection model equation given by CHAPTER 11.uj.ax . . the solution will either grow or decay with time. Hence the forward di erence operator is inherently unstable while the centered and backward operators are inherently stable. With 6= 0.2) The second form shown divides the operator into an antisymmetric component (. the centered di erence operator ( = 0) produces Re( m ) = 0. our choice of di erencing scheme produces nonphysical behavior. and the backward di erence operator produces Re( m) < 0. 2 1 .a x = 2. Consider the following point operator for the spatial derivative term . uj+1)] 2 x (11. )uj+1] = .a (.ax Bp(.a @u @t @x (11.1 .1 + 2 uj + (1 .uj.uj.1 + 2uj .(1 + )uj. NUMERICAL DISSIPATION @u = .204 11. We proceed next to show why this occurs.1+ uj+1)=2 x and a symmetric component (. uj+1)=2 x.a( x u)j = 2.1 + 2uj . ) The eigenvalues of this matrix are m = . In either case.1) with periodic boundary conditions. cos 2M m + i sin 2M for m = 0 1 : : : M . the roles are reversed. the operator is only rstorder accurate. If a is negative. When Re( m) = 0. A backward di erence operator is given by = 1 and a forward di erence operator is given by = . 1 If a is positive.1. the forward di erence operator ( = .1) produces Re( m ) > 0.1 + uj+1) + (.
11. This means that each term in the expansion given by Eq. 11. 2 ( . 11. a x2 @ 3 u + a x3 @ 4 u + : : : (11. when we use a computer to nd a numerical solution of the problem.5)   amplitude phase . (a + 2 ) The solution is composed of both amplitude and phase terms.3) @t @x 2 @x2 6 @x3 24 @x4 This is the partial di erential equation we are really solving when we apply the approximation given by Eq.({z+ 2 )t}] (11. 11.3 is excited to some degree. We proceed next to investigate the implications of this concept.3 as the modi ed partial di erential equation. 2 ) s = .1.3 is consistent with Eq. @x j 205 First carry out a Taylor series expansion of the terms in Eq. We are lead to the expression 2 ! 3 ! ! ! @ 2 u + x3 @ 3 u . 11. Under these conditions there is a wavelike solution to Eq. Consider the simple linear partial di erential equation @u = . Thus a ) u = e.2 to Eq. x can be small but it is not zero.a @u + a x @ 2 u . 11. 11. 2} ei x.4) @t @x @x2 @x3 @x4 Choose periodic boundary conditions and impose an initial condition u = ei x.1.1 gives @u = . THE MODIFIED PARTIAL DIFFERENTIAL EQUATION 11. 2 . We refer to Eq.11. since the two equations are identical when x ! 0.ia . However. Notice that Eq. Substituting this into Eq. x4 @ 4 u + : : :5 x2 @x2 3 @x3 j 12 @x4 j j We see that the antisymmetric portion of the operator introduces odd derivative terms in the truncation error while the symmetric portion introduces even derivatives.2 The Modi ed Partial Di erential Equation ( xu)j = 2 1 x 42 x @u . 2 ({z.2.2. i 3 + 4 or r = . 11. 11.4 of the form u(x t) = ei xe(r+is)t provided r and s satisfy the condition r + is = .a @u + @ 2 u + @ 3 u + @ 4 u (11.
uj+2) 12 x + (uj.4. Note that the use of onesided di erencing schemes is not the only way to introduce dissipation. that the same is true for any centered di erence approximation of a rst derivative.6) The antisymmetric component of this operator is the fourthorder centered di erence operator. one could choose = 1=2 in Eq.5. 4uj+1 + uj+2)] (11.1 + 3uj + 2uj+1) = 1 (uj.3 we have = . Under the same condition.2 . the choice of a backward di erence scheme ( = 1) produces a modi ed PDE with . 11. 11. 11. Referring to the modi ed PDE. .1) is equivalent to deliberately adding a destabilizing term to the PDE.206 CHAPTER 11. the choice of a forward di erence scheme ( = . Therefore.2 . this operator produces fourthorder accuracy in phase with a thirdorder dissipative term. This we refer to as dispersion. If the wave speed a is positive.1 + 6uj . We have seen that the threepoint centered di erence approximation of the spatial derivative produces a modi ed PDE that has no dissipation (or ampli cation).a x2 =6. Furthermore. 2 > 0 and hence the amplitude of the solution decays.2. 6uj. and the phase depends only on a and . by using the antisymmetry of the matrix di erence operators. Our purpose here is to investigate the properties of onesided spatial di erencing operators relative to centered di erence operators. 8uj. NUMERICAL DISSIPATION It is important to notice that the amplitude of the solution depends only upon and .2 . the numerical phase speed is dependent upon the wavenumber .1 + 8uj+1 . By examining the term governing the phase of the solution in Eq. The symmetric component approximates x3 uxxxx=12. The resulting spatial operator is not onesided but it is dissipative. For example. a thirdorder backwardbiased scheme is given by ( xu)j = 6 1 x (uj. One can easily show. 11. Therefore. the coe cients of the odd derivatives. Biased schemes use more information on one side of the node than the other. the coe cients of the even derivatives in Eq. As a corollary. Any symmetric component in the spatial operator introduces dissipation (or ampli cation). 4uj. Eq. any departure from antisymmetry in the matrix di erence operator must introduce dissipation (or ampli cation) into the modi ed PDE. This is tantamount to deliberately adding dissipation to the PDE. we see that the speed of propagation is a + 2 . Therefore. the phase speed of the numerical solution is less than the actual phase speed.
It is a fullydiscrete nitedi erence scheme. THE LAXWENDROFF METHOD 11.9) uj j j +1 j j .8) @t @x @t2 @x2 Now replace the space derivatives with threepoint centered di erence operators. ah 2 1 4. The quantity j ah j is known as the x Courant (or CFL) number.7) First replace the time derivatives with space derivatives according to the PDE (in this case. 1 For j ah j 1 all of the eigenvalues have modulus less than or equal to unity and hence x the method is stable independent of the sign of a. cos 2 m . 2u(n) + u(n) ) (11. 1 ah (u(n) . u(n) ) + 1 ah (u(n) . the corresponding fullydiscrete matrix operator is 0 2 !3 ! 2 ! 31 1 4 ah + ah 25 1 .11. This explicit method di ers conceptually from the methods considered previously in which spatial di erencing and timemarching are treated separately.3 The LaxWendro Method 207 In order to introduce numerical dissipation using onesided di erencing. Next we consider a method which introduces dissipation independent of the sign of the wave speed. .1 2 x This is the LaxWendro method applied to the linear convection equation. the linear convection equation @u + a @u = 0). For periodic boundary conditions. backward di erencing must be used if the wave speed is positive. giving !2 (n+1) = u(n) . and forward di erencing must be used if the wave speed is negative. ah + ah 2 5A ~ ~ n+1 = Bp @ 2 x u un x x 2 x x The eigenvalues of this matrix are ! ah 2 1 . i ah sin 2 m m =1. It is equal to the ratio of the distance travelled by a wave in one time step to the mesh spacing. There is no intermediate semidiscrete stage.3.a @u (11. x M x M for m = 0 1 : : : M . Consider the following Taylorseries expansion in time: 2 u(x t + h) = u + h @u + 1 h2 @ u + O(h3) @t 2 @t2 (11. Thus @t @x @ 2 u = a2 @ 2 u @u = .1 2 x j+1 j. known as the LaxWendro method.
however. The two methods di er when applied to nonlinear hyperbolic systems. 11. a ( x2 . Recall MacCormack's timemarching method. a2h2 ) @ 3 u = . a h8 x (1 . Hence MacCormack's method has the same dissipative and dispersive properties as the LaxWendro method.10) un+1 = 2 un + un+1 + hu0n+1] If we use rstorder backward di erencing in the rst stage and rstorder forward di erencing in the second stage. a2 h2) @ 4 u + : : : @t @x 6 @x3 8 @x4 This is derived by substituting Taylor series expansions for all terms in Eq. For the linear convection equation. a8h ( x2 . u(jn)1) ~ . 11. u(jn+1))] ~ (11. The two leading error terms appear on the right side of the equation. a2 h ( x2 .208 CHAPTER 11. presented in Chapter 6: un+1 = un + hu0n ~ 1 ~ ~ (11. these should be applied alternately. Recall that the odd derivatives on the right side lead to unwanted dispersion and the even derivatives lead to dissipation (or ampli cation. Therefore. ah (~(jn+1) . x u(jn+1) = 1 u(jn) + u(jn+1) . C 2 ) @ 3 u .11) 2 x u +1 ~ which can be shown to be identical to the LaxWendro method. A closely related method is that of MacCormack. a2h2 ) @ 3 u .6 n @x3 @x3 6 The dissipative term is proportional to 2 4 2 2 4 2 @ . . Cn) @xu 4 @x4 This term has the appropriate sign and hence the scheme is truly dissipative as long as Cn 1. 1 Or viceversa for nonlinear problems.1 a dissipative secondorder method is obtained. depending on the sign). a x2 (1 . NUMERICAL DISSIPATION The nature of the dissipative properties of the LaxWendro scheme can be seen by examining the modi ed partial di erential equation. this approach leads to u(jn+1) = u(jn) .8. a2h2 ) @ u = . which is given by @u + a @u = . the leading error term in the LaxWendro method is dispersive and proportional to a ( x2 . ah (u(jn) .9 and converting the time derivatives to space derivatives using Eq.
This is avoided in the LaxWendro scheme.4.4 Upwind Schemes 209 In Section 11. = 0. by adding a symmetric component to the spatial operator. In this section.11.uj. For more subtle aspects of implementation and application of such techniques to .12) @t @x where we do not make any assumptions as to the sign of a. manner. as a result of their superior exibility. that is.1.e.5. the eigenvalues of the ux Jacobian for the onedimensional Euler equations are u u + a u . = a 0. the wave speeds can be both positive and negative. For example. these are of mixed sign.1 + uj+1) + jaj(. We can rewrite Eq. 11. more generally.12 as @u + (a+ + a. 11. we present the basic ideas of uxvector and uxdi erence splitting. whether it is a forward or a backward di erence) or the sign of the symmetric component depends on the sign of the wave speed. we saw that numerical dissipation can be introduced in the spatial di erence operator using onesided di erence schemes or.13) This approach is extended to systems of equations in Section 11.. From Eq. some form of splitting is required.1 + 2uj . Now for the a+ ( 0) term we can safely backward di erence and for the a. but entirely equivalent. ) @u = 0 a = a 2 jaj @t @x + = a 0 and a. some decomposition or splitting of the uxes into terms which have positive and negative characteristic speeds so that appropriate di erencing schemes can be chosen. UPWIND SCHEMES 11.1x a(. In order to apply onesided di erencing schemes to such systems. This is the basic concept behind upwind methods. When the ow is subsonic. The above approach to obtaining a stable discretization independent of the sign of a can be written in a di erent. if a 0. then a a. However. uj+1)] (11. the direction of the onesided operator (i.1 if a 0. we present two splitting techniques commonly used with upwind methods. Alternatively. With this approach. Consider again the linear convection equation: @u + a @u = 0 (11. we see that a stable discretization is obtained with = 1 if a 0 and with = .uj. In the next two sections.a( xu)j = 2. This is achieved by the following point operator: . then a+ = 0 and If a 0.2. schemes in which the numerical dissipation is introduced in the spatial operator are generally preferred over the LaxWendro approach. a. ( 0) term forward di erence. When a hyperbolic system of equations is being solved. These are by no means unique.
1Xw + @X . we need to apply a backward di erence if the wave speed. and the wi's are the characteristic variables.210 CHAPTER 11. i. In order to apply a onesided (or biased) spatial di erencing scheme. are the eigenvalues of the Jacobian matrix. To accomplish this.1 FluxVector Splitting Recall from Section 2.5 that a linear.X . @w = 0 (11. We can use backward di erencing for the + @w term and forward @x di erencing for the .16) +j j . contains the negative eigenvalues. = .1Xw = 0 @t @x @x (11. let us split the matrix of eigenvalues. i. We can now rewrite the system in terms of characteristic variables as @w + @w = @w + + @w + . the reader is referred to the literature on this subject. NUMERICAL DISSIPATION nonlinear hyperbolic systems such as the Euler equations.j j (11. constantcoe cient. A. and a forward di erence if the wave speed is negative. where += (11. @w term. is positive. Premultiplying by X and inserting the product @x X .15) i @t @x where the wave speeds.14) @t @x @t @x can be decoupled into characteristic equations of the form @wi + @wi = 0 (11.19) . + contains the positive eigenvalues and .1X in the spatial terms gives @Xw + @X +X . . 11.4.17) 2 2 With these de nitions. into two components such that = ++ . hyperbolic system of partial di erential equations given by @u + @f = @u + A @u = 0 (11.18) @t @x @t @x @x The spatial terms have been split into two components according to the sign of the wave speeds.
25) A. (11.. as discussed in Appendix C. f = Au. the Jacobians of the split ux vectors are required. f is also equal to Au as a result of their homogeneous property. = X .23) @t @x @x In the linear case. (11. has all negative eigenvalues. . we obtain A. u @u + @f + + @f . and A. 6= A.21) (11.22) f . one must nd and use the new Jacobians given by A++ = @f @u + (11.1 and recalling that u = Xw. Note that f = f+ + f. UPWIND SCHEMES With the de nitions2 211 A+ = X +X .24) Thus by applying backward di erences to the f + term and forward di erences to the f . has all negative eigenvalues.11.26) For the Euler equations.. term. A++ has eigenvalues which are all positive. = 0 (11. the de nition of the split uxes follows directly from the de nition of the ux. = @f @u . we are in e ect solving the characteristic equations in the desired manner. and A.X . For the Euler equations.4. @f + 6= A+ @f . When an implicit timemarching method is used. = A.20) @u + @A+u + @A. 2 With these de nitions A+ has all positive eigenvalues. @u @u Therefore.1 (11. u = 0 @t @x @x Finally the split ux vectors are de ned as f + = A+u and we can write (11. This approach is known as uxvector splitting. In the nonlinear case.
wR ) ^ (11.19. the uxes must be evaluated at cell boundaries. is known as uxdi erence splitting.1X (wL . 11. gj+1=2.4. We again begin with the diagonalized form of the linear. uR) (11. more suited to nitevolume methods. constantcoe cient.34) .28) In order to obtain a onesided upwind approximation.31) 2 Now. as a ^ function of the states to the left and right of the interface. as in Eq. respectively. hyperbolic system of equations @w + @w = 0 (11.33) 2 where jAj = X j jX . which is nondissipative.1X (wL + wR) + 2 X j jX . The centered approximation to gj+1=2.212 CHAPTER 11.29) i i R if i < 0 where the subscript i indicates individual components of w and g.30) or 1 gj+1=2 = 1 (wL + wR) + 2 j j (wL .1X after and j j to obtain 1 1 X gj+1=2 = 2 X X . is given by 1 gj+1=2 = 2 (g(wL) + g(wR)) ^ (11. wR ) ^ (11. we premultiply by X to return to the original variables and insert the product X . (wi)R] g (11. as in Chapter 5. In a nitevolume method. NUMERICAL DISSIPATION 11.1 (11. wL and wR.2 FluxDi erence Splitting Another approach. we consider the numerical ux at the interface between nodes j and j + 1. we require ( (w ) (^i)j+1=2 = i(wi)L if i > 0 g (11. This is achieved with 1 1 (^i)j+1=2 = 2 i (wi)L + (wi)R ] + 2 j ij (wi)L .27) @t @x The ux vector associated with this form is g = w. Now.32) and thus 1 f^j+1=2 = 1 (fL + fR ) + 2 jAj (uL .
we would like our de nition of f^j+1=2 to satisfy 0 f f^j+1=2 = fL if all i 0s > 0 (11.35) R if all i s < 0 giving pure upwinding.37) For the Euler equations for a perfect gas.38) uj+1=2 = p + p L R p H +p H L L Hj+1=2 = (11.36) 2 if Aj+1=2 satis es fL .35 is obtained by the de nition 1 f^j+1=2 = 2 (fL + fR ) + 1 jAj+1=2j (uL . u = Xw. 3 Note that the ux Jacobian can be written in terms of u and H only see problem 6 at the end .37 is satis ed by the ux Jacobian evaluated at the Roeaverage state given by p LuL + p R uR (11. In order to resolve this.39) p L+p R R R where u and H = (e + p)= are the velocity and the total enthalpy per unit mass. this leads to an upwind operator which is identical to that obtained using uxvector splitting. fR = Aj+1=2 (uL . Eq. In the linear. If the eigenvalues of A are all positive. However. constantcoe cient case.1.11. Thus we can generalize the concept of upwinding to include any scheme in which the symmetric portion of the operator is treated in such a manner as to be truly dissipative.A. 11. In this case. uR) (11. jAj = . consider a situation in which the eigenvalues of A are all of the same sign. and A = X X . 11.3 ( 11. uR) (11. We have also seen that such dissipation can be introduced by adding a symmetric component to an antisymmetric (dissipationfree) operator. there is some ambiguity regarding the de nition of jAj at the cell interface j + 1=2. respectively. ARTIFICIAL DISSIPATION 213 and we have also used the relations f = Xg.5 Arti cial Dissipation We have seen that numerical dissipation can be introduced by using onesided differencing schemes together with some form of ux splitting. of this chapter.5. in the nonlinear case. Hence satisfaction of Eq. jAj = A if they are all negative.
1 (11.1 2 x 2 x a s Applying x = x + x to the spatial derivative in Eq.1 + 6uj .44) x (uj.2 . this approach can be related to the upwind approach. x is stable if i 0 and unstable if i > 0. uj.42) x (Au) = x (Au) + x (jAju) or a s (11. 11.43) x f = x f + x (jAju) where jAj is de ned in Eq. The appropriate implementation is thus a s (11.2 . this often provides su cient damping of high frequency modes without greatly a ecting the low frequency modes.40) ( x u)j = uj+1 .1 + 3uj ) 2 x . but is beyond the scope of this book. A secondorder backward di erence approximation to a 1st derivative is given as a point operator by ( xu)j = 1 (uj. Similarly. 4uj. This is particularly evident from a comparison of Eqs. The second spatial term is known as arti cial dissipation. uj. 11.41) i x = i x + j ij x Extension to a hyperbolic system by applying the above approach to the characteristic variables. 11. It is also sometimes referred to as arti cial di usion or arti cial viscosity. 4uj+1 + uj+2) where is a problemdependent coe cient.214 For example.43.15 is stable if i 0 and a s unstable if i < 0.36 and 11.6 Problems 1. applying x = x . as in the previous two sections. For details of how this can be implemented for nonlinear hyperbolic systems. With an appropriate value of . NUMERICAL DISSIPATION s a ( xu)j = . This symmetric operator approximates x3 uxxxx and thus introduces a thirdorder dissipative term. A more complicated treatment of the numerical dissipation is also required near shock waves and other discontinuities. a s With appropriate choices of x and x. gives a s (11. the reader should consult the literature. s ( xu)j = 11. s It is common to use the following operator for x (11.34. let CHAPTER 11.uj+1 + 2uj . 4uj.
Find the modi ed wavenumber for the operator given in Eq. Plot the real and imaginary parts of x vs.6. Plot the real and imaginary parts of x vs. nd the derivative which is approximated by the corresponding symmetric and skewsymmetric operators and the leading error term for each.39 leads to satisfaction of Eq. Find the matrix jAj. x for 0 x .4. then derive the symmetric and skewsymmetric matrices that have the matrix operator as their sum. Consider the spatial operator obtained by combining secondorder centered differences with the symmetric operator given in Eq. and 1=48. Plot the real and imaginary parts of x vs.2. 11. 5. Using Fourier analysis as in Section 6. Show that the ux Jacobian for the 1D Euler equations can be written in terms of u and H .2. Show that the use of the Roe average state given in Eqs. 6. Using Fourier analysis as in Section 6. Find the modi ed wavenumber for this operator with = 0 1=12 1=24. x for 0 x . 11. nd j j for the combination of this spatial operator with 4thorder RungeKutta time marching at a Courant number of unity and plot vs. x for 0 x . 11.3 to see how to construct the symmetric and skewsymmetric components of a matrix.6. 4.38 and 11.2.1. 3. 11.6. (See Appendix A. .44. Form the plusminus split ux vectors as in Section 11.11.6. Find the modi ed wavenumber for the rstorder backward di erence operator. nd j j for the combination of this spatial operator with 4thorder RungeKutta time marching at a Courant number of unity and plot vs. x for 0 x .6. x for 0 x . Consider the hyperbolic system derived in problem 8 of Chapter 2. x for 0 x . Using Fourier analysis as in Section 6.37. PROBLEMS 215 (a) Express this operator in banded matrix form (for periodic boundary conditions). nd j j for the combination of this spatial operator with 4thorder RungeKutta time marching at a Courant number of unity and plot vs. 2.) (b) Using a Taylor table.
216 CHAPTER 11. NUMERICAL DISSIPATION .
Chapter 12 SPLIT AND FACTORED FORMS In the next two chapters.1 The Concept Factored forms of numerical operators are used extensively in constructing and applying numerical methods to problems in uid mechanics.1) . 12. h. Time marching methods are valid only to some order of accuracy in the step size. we present and analyze split and factored algorithms. and \fractional step". This gives the reader a feel for some of the modi cations which can be made to the basic algorithms in order to obtain e cient solvers for practical multidimensional applications. Factored forms are especially useful for the derivation of practical algorithms that use implicit methods. \time split". When we approach numerical analysis in the light of matrix derivative operators. the concept of factoring is quite simple to present and grasp. Matrices can be split in quite arbitrary ways. ~ u u f dt 217 (12. They are the basis for a wide variety of methods variously known by the labels \hybrid". 3. Now recall the generic ODE's produced by the semidiscrete approach d~ = A~ . and a means for analyzing such modi ed forms. 2. Advancing to the next time level always requires some reference to a previous one. Let us start with the following observations: 1.
For the time march let us choose the simple. from observation 3 (allowing us to drop higher order terms .5) ~ ~ n = I + hAd + hAc]~ n + h(bc) + O(h ) u u +1 2 (12. h A A i~ n . +1 1 2 2 ~ n = I + hA ] I + hA ]~ n . Both of these considerations are investigated later. From observation 1 (arbitrary splitting of A) : (12. 12. ~ u u f dt 1 2 2 and consider the above observations. respectively and their sum forms the A matrix we have considered in the previous sections. h~ + O(h ) u u f 12. rstorder. their numerical stability can be quite di erent. Here we seek only to apply to some simple cases the concept of factoring.2 Factoring Physical Representations  Time Splitting Suppose we have a PDE that represents both the processes of convection and dissipation. h~ + O(h ) u u f +1 2 1 2 Finally. .2) where A = A + A ] but A and A are not unique.3 and 12. Choose again the explicit Euler time march so that d~ = A ~ + A ~ + (bc) u ~ cu du dt (12. However.6) 1 Secondorder timemarching methods are considered later.4 have the same formal order of accuracy and.h A A ~ n): u (12. neither one is to be preferred over the other. h~ + O(h ) u u f +1 1 2 2 1 2 2 1 2 (12.4) Notice that Eqs. and techniques to carry out their numerical evaluation can have arithmetic operation counts that vary by orders of magnitude. Then. explicit Euler method.218 CHAPTER 12. in this sense.3) 2 or its equivalent ~ n = h I + hA ] I + hA ] . SPLIT AND FACTORED FORMS d~ = A + A ]~ . The semidiscrete approach to its solution might be put in the form where Ac and Ad are matrices representing the convection and dissipation terms. from observation 2 (new data ~ n in terms of old ~ n): u u 1 2 1 1 +1 ~ n = I + hA + hA ]~ n .
For example. another way to approximate Eq. hAd]. their selection is arbitrary.7 are often applied in a predictorcorrector sequence. on this basis.6 with the same order of accuracy is given by the expression ~n u +1 Original Unfactored Terms where in this approximation we have used the fact that 1 2 2 = = ~ I .6 have identical orders of accuracy in the time approximation. However. I + hAc]~ n + h(bc) u ~ I + hAd + hAc]~ n + h(bc) +O(h ) u  {z } 1 2 (12.8. In this case one could write 2 = = ~ I + hAd ] I + hAc]~ n + h(bc) u ~ u ~ I + hAd + hAc]~ n + h(bc) + h Ad Ac~ n + (bc) +O(h ) (12. hAd ]. hAd]~ n u +1 +1 ~ = I + hAc]~ n + h(bc) u = un ~ +1 (12. Since numerical sti ness is generally much more severe for the di usion process. where jjAd jj is some norm of Ad ]. Therefore.2.7 and the original unfactored form Eq. 12. as before. = I + hAd + h Ad + if h jjAdjj < 1.7) u {z }  2 2  {z } un ~ ~n u +1 +1 = = ~ I + hAc]~ n + h(bc) u I + hAd]~n u +1 (12. and a rstorder timemarch method is usually unsatisfactory.10) The convection operator is applied explicitly.8) Factoring can also be useful to form split combinations of implicit and explicit techniques. 12. requiring a tridiagonal solver if the di usion term is central di erenced. but the di usion operator is now implicit. stability. In practical applications equations such as 12. 12. We have yet to determine its . the important aspect of stability has yet to be discussed. FACTORING PHYSICAL REPRESENTATIONS  TIME SPLITTING 219 Now consider the factored form ~n u +1 Original Unfactored Terms Higher Order Terms and we see that Eq. 2 We do not suggest that this particular method is suitable for use.12.9) I . This time a predictorcorrector interpretation leads to the sequence un ~ I . this factored form would appear to be superior to that provided by Eq. 12.
9 can be derived for a di erent point of view by writing Eq.11) My k 1 13 23 33 43 12 22 32 42 11 21 31 41 1 (12. ~ I . 12. hAd ]un = I + hAc]un + h(bc) +1 12.3 Factoring Space Matrix Operators in 2{D Factoring is widely used in codes designed for the numerical solution of equations governing unsteady two. the number of (interior) y points is 3. is 4 and My . 3 This could also be called a 5 .10. but in these notes we describe the size of a mesh by the number of interior points.12.and 3D requires some reference to a mesh. Thus u represents the value of u at j = 3 and k = 2. 3 32 @u = @ u + @ u @t @x @y 2 2 (12. Let us study the basic concept of factoring by inspecting its use on the linear 2D scalar PDE that models di usion: We begin by reducing this PDE to a coupled set of ODE's by di erencing the space derivatives and inspecting the resulting matrix operator. 6 point mesh if the boundary points (labeled in the sketch) were included. un = A u + A u + (bc) + O(h ) ~ c n d n h +1 +1 2 We should mention here that Eq.1 Mesh Indexing Convention 2 2 12. A clear description of a matrix nitedi erence operator in 2.12) j Mx Mesh indexing in 2D. The numbers 11 12 43 represent the location in the mesh of the dependent variable bearing that index. the number of (interior) x points. We choose the 3 4 point mesh shown in the Sketch 12. 12. SPLIT AND FACTORED FORMS un . 12.6 in the form Then which is identical to Eq.and threedimensional ows.3. In this example Mx .220 CHAPTER 12.
To be speci c about the structure. and one composed of yvectors with U y . depending on the data{base chosen for the U it multiplies.13) where Pyx = PT = P.3 Data Base Permutations U x = Pxy U y ( ) ( ) The two vectors (arrays) are related by a permutation matrix P such that and U y = PyxU x ( ) 1 ( ) (12. 12. and they are related by the same permutation matrix that relates U x to U y . we consider only two: (1).3. (1) and (2) above are referred to as xvectors and yvectors. 12. we label a data{base composed of xvectors with U x . Notice that the matrices are not the same although they represent the same derivative operation.12. and (2). Examples of the order of indexing for these space vectors are given in Eq.14) Axx y = Pxy Axy y Pyx ( ) + ( ) + (12.2 Data Bases and Space Vectors The dimensioned array in a computer code that allots the storage locations of the dependent variable(s) is referred to as a database. which are notations used in the special case of space vectors.3.16 part a and b.15) Ax+y and U .3. xy xy Now consider the structure of a matrix nitedi erence operator representing 3point centraldi erencing schemes for both space derivatives in two dimensions. u 4 Notice that . 12.11 can be written + 4 If it is important to be speci c about the database structure. respectively. consecutively along columns that are consecutive from j = 1 to Mx. Thus ( ) + ( ) + ( ) ( ) dU = A U + (bc) ~ x y dt + (12. consecutively along rows that are themselves consecutive from k = 1 to My . the usual (but ambiguous) representation is given by Ax y . In this notation the ODE form of Eq. When the matrix is multiplying a space vector U . In particular. FACTORING SPACE MATRIX OPERATORS IN 2{D 221 12. however. There are many ways to lay out a database. The symbol U by itself is not enough to identify the structure of the database and is used only when the structure is immaterial or understood. Their structures are similar. We refer to each row or column group as a space vector (they represent data along lines that are continuous in space) and label their sum with the symbol U . Of these. we use the notation Axx y or Axy y .16 part a and b. ( ) ( ) 12. are subsets of A and ~ . Examples are in Eq. used in the previous sections.
Ax y . 41 42 43 31 32 33 21 22 23 11 12 13 (12. for 3 4 mesh shown in Sketch 12. centraldi erence. for 3 4 mesh shown in Sketch 12.. 12 22 32 42 13 23 33 43 11 21 31 41 . .16) b: Elements in 2dimensional. Ax y . for y ! o... matrix operator. Entries for x ! x. + ( ) 2 6o 6 6 6 6 6 6 6x 6 6 6 6 6 6 (y ) (y ) Ax+y U = 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 4 j x j o j x j o j x j o j j j j o j x j x j o o j x j x j o j x j j x j o j x j x j o o j x j x j o j j j j + j x j o j x j o j x j o ( ) 3 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 x7 7 7 7 7 7 7 o5 .. Entries for x ! x. Data base composed of Mx y{vectors stored in U y . SPLIT AND FACTORED FORMS x j o x j o x x j o x j o j j j j j x j o o j x x j o o j x x j o o j x j j o j o j o j o j x j x x j x j x 3 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 o7 7 7 7 7 7 7 7 7 5 x7 .222 2 6x 6 6 6 6 6 6 6 6 6 6o 6 6 (x) (x) Ax+y U = 6 6 6 6 6 6 6 6 6 6 6 6 6 6 4 CHAPTER 12.12. for y ! o.12. centraldi erence. a:Elements in 2dimensional.. . for both ! . matrix operator. Data base composed of My x{vectors stored in U x . . for both ! .
12.3. FACTORING SPACE MATRIX OPERATORS IN 2{D
223
12.3.4 Space Splitting and Factoring
We are now prepared to discuss splitting in two dimensions. It should be clear that the matrix Axx y can be split into two matrices such that
( ) +
Axx y = Axx + Ayx
( ) + ( ) ( ) ( )
( )
(12.17)
where Axx and Ayx are shown in Eq. 12.22. Similarily
Axy y = Axy + Ayy
( ) + ( )
( )
(12.18)
where the split matrices are shown in Eq. 12.23. The permutation relation also holds for the split matrices so
Ayx = Pxy Ayy Pyx
( ) ( )
and
Axx = Pxy Axy Pyx
( ) ( )
The splittings in Eqs. 12.17 and 12.18 can be combined with factoring in the manner described in Section 12.2. As an example ( rstorder in time), applying the implicit Euler method to Eq. 12.14 gives
h i ~ Unx = Unx + h Axx + Ayx Unx + h(bc)
( ) +1 ( ) ( ) ( ) ( ) +1
or
h i ~ I ; hAxx ; hAyx Unx = Unx + h(bc) + O(h )
( ) ( ) ( ) +1 ( ) 2
(12.19) (12.20)
As in Section 12.2, we retain the same rst order accuracy with the alternative
h ih i ~ I ; hAxx I ; hAyx Unx = Unx + h(bc) + O(h )
( ) ( ) ( ) +1 ( ) 2
Write this in predictorcorrector form and permute the data base of the second row. There results
~ ~ I ; hAxx U x = Unx + h(bc) h i y ~ I ; hAyy Un = U y
( ) ( ) ( ) ( ) ( ) ( ) +1
h
i
(12.21)
224
CHAPTER 12. SPLIT AND FACTORED FORMS
Axx
( )
2 x 6x 6 6 6 6 6 6 6 6 6 6 6 6 (x) U =6 6 6 6 6 6 6 6 6 6 6 6 6 6 4
x x x x x x x x
j j j j j x x j x x x j x x x j x x
j j j j j j j j
Ayx
( )
6 6 6o 6 6 U (x) = 6 6 6 6 6 6 6 6 6 6 6 6 6 6 4
j j j j 2 o j o 6 o j o 6 6 6 o j o 6 6 o j o 6 6 j o o j o o j o o j o j o j o j o j o
The splitting of Axx y .
( ) +
7 7 j o 7 7 j o 7 5 j o 7 j o
7 7 j o 7 7 j o 7 Ux 7 j o 7 7 j o7 7 7 7
j x x j x x x j x x j x x 3 j 7 j 7 7 7 j 7 7 j 7 7
7
3 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 5 x7
Ux
( )
(12.22)
( )
12.3. FACTORING SPACE MATRIX OPERATORS IN 2{D
225
Axy
( )
2 x 6 6 6 6 6 6 6 6x 6 6 6 6 6 6 U (y) = 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 4
j x j x j x j x j x j
j j j
j x j x j x j x j x j x j x j x j j x j x j x j x j x j x j x j x j
Ayy
( )
6 6 6 6 6 6 6 6 6 (y ) U =6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 4
j j j 2 o o j 6o o o j 6 6 o o j 6 6
j x j x j x j x j x j x 3 j j 7 j j 7 7 j j 7 7 j j j
7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 o5
3 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 x7 7 7 7 7 7 7 5
Uy
( )
(12.23)
j o o j j o o o j j o o j j j j j j j
j o o j j o o o j j o o j j j j
( ) +
Uy
( )
j o o j o o j o o
The splitting of Axy y .
226
12.4 SecondOrder Factored Implicit Methods
CHAPTER 12. SPLIT AND FACTORED FORMS
Secondorder accuracy in time can be maintained in a certain factored implicit methods. For example, apply the trapezoidal method to Eq. 12.14 where the derivative operators have been split as in Eq. 12.17 or 12.18. Let the data base be immaterial ~ and the (bc) be time invariant. There results 1 1 ~ I ; 2 hAx ; 1 hAy Un = I + 2 hAx + 1 hAy Un + h(bc) + O(h ) (12.24) 2 2 Factor both sides giving 1 I ; 1 hAx I ; 2 hAy ; 1 h AxAy Un 2 4 1 hA I + 1 hA ; 1 h A A U + h(bc) + O(h ) (12.25) ~ = I+ 2 x 2 y 4 x y n Then notice that the combination 1 h AxAy ](Un ; Un) is proportional to h since 4 the leading term in the expansion of (Un ; Un) is proportional to h. Therefore, we can write 1 ~ I ; 1 hAx I ; 1 hAy Un = I + 2 hAx I + 1 hAy Un + h(bc) + O(h )(12.26) 2 2 2 and both the factored and unfactored form of the trapezoidal method are secondorder accurate in the time march. An alternative form of this kind of factorization is the classical ADI (alternating direction implicit) method usually written 1 1 ~ I ; 2 hAx U = I + 2 hAy Un + 1 hFn 2 1 hA U + 1 hF + O(h ) 1 hA U (12.27) I;2 y n = I+2 x ~ 2 n
+1 3 2 +1 2 3 2 +1 3 +1 +1 3 5 +1 +1 3
For idealized commuting systems the methods given by Eqs. 12.26 and 12.27 di er only in their evaluation of a timedependent forcing term.
12.5 Importance of Factored Forms in 2 and 3 Dimensions
When the timemarch equations are sti and implicit methods are required to permit reasonably large time steps, the use of factored forms becomes a very valuable tool
5 A form of the Douglas or PeacemanRachford methods.
12.5. IMPORTANCE OF FACTORED FORMS IN 2 AND 3 DIMENSIONS 227
for realistic problems. Consider, for example, the problem of computing the time advance in the unfactored form of the trapezoidal method given by Eq. 12.24 1 ~ I ; 2 hAx y Un = I + 1 hAx y Un + h(bc) 2 Forming the right hand side poses no problem, but nding Un requires the solution of a sparse, but very large, set of coupled simultaneous equations having the matrix form shown in Eq. 12.16 part a and b. Furthermore, in real cases involving the Euler or NavierStokes equations, each symbol (o x ) represents a 4 4 block matrix with entries that depend on the pressure, density and velocity eld. Suppose we were to solve the equations directly. The forward sweep of a simple Gaussian elimination lls all of the 4 4 blocks between the main and outermost diagonal (e.g. between and o in Eq. 12.16 part b.). This must be stored in computer memory to be used to nd the nal solution in the backward sweep. If Ne represents the order of the small block matrix (4 in the 2D Euler case), the approximate memory requirement is (Ne My ) (Ne My ) Mx oating point words. Here it is assumed that My < Mx. If My > Mx, My and Mx would be interchanged. A moderate mesh of 60 200 points would require over 11 million words to nd the solution. Actually current computer power is able to cope rather easily with storage requirements of this order of magnitude. With computing speeds of over one giga op, direct solvers may become useful for nding steadystate solutions of practical problems in two dimensions. However, a threedimensional solver would require a memory of approximatly
+ +1 + +1 6 7 8
words and, for well resolved ow elds, this probably exceeds memory availability for some time to come. On the other hand, consider computing a solution using the factored implicit equation 12.25. Again computing the right hand side poses no problem. Accumulate the result of such a computation in the array (RHS ). One can then write the remaining terms in the twostep predictorcorrector form 1 ~ I ; 2 hAxx U x = (RHS ) x ~ I ; 1 hAyy Uny = U y (12.28) 2
( ) ( ) ( ) ( ) ( ) +1 ( )
Ne My Mz Mx
2 2 2
practical size the ll is mostly dense. 7 The lower band is also computed but does not have to be saved unless the solution is to be repeated for another vector. 8 One billion oatingpoint operations per second.
6 For matrices as small as those shown there are many gaps in this \ ll", but for meshes of
each with Mx blocks of order Ne . 12. One way that is especially useful. and matrix multiplications do not commute. 2 hAy Un = h Ax y Un + (bc) + O(h ) (12. 2 hAy Un = I + 2 hAx + 1 hAy Un + h(bc) + O(h ) 2 From both sides subtract 1 I . and ~ let the (bc) be time invariant: 1 1 1 ~ I . 2 hAx .14. we see that this is equivalent to solving My onedimensional ~ problems. 2 hAx . 1 hAy Un 2 leaving the equality unchanged.21 but is secondorder time accurate. 12.228 CHAPTER 12.22. Consider the unfactored form of the trapezoidal method given by Eq. 12.30) I . Then. SPLIT AND FACTORED FORMS 9 which has the same appearance as Eq.6 The Delta Form Clearly many ways can be devised to split the matrices and generate factored forms. We arrive at the expression h 1 ~ i (12. is referred to as the \delta form" and we develop it next. 12. it is guaranteed to converge to the correct steadystate solution of the ODE. 12.23 shows that the nal ~ be permuted to U step consists of solving Mx onedimensional implicit problems each with dimension My . +1 3 +1 + 3 2 + 3 replace the scalar ones. 1 hAy Un = h Ax y Un + (bc) + O(h ) 2 This is the delta form of a factored. if Eq. Now we can factor Eq. using the standard de nition of the di erence operator . 12. The rst step would be solved using My uncoupled block tridiagonal solvers . 12. Un one nds h 1 1 ~ i I .29) Notice that the right side of this equation is the product of h and a term that is identical to the right side of Eq. ( ) ( ) 12. Thus. The temporary solution U x would then y and an inspection of the bottom of Eq. our original ODE.24.29 converges.29 and maintain O(h ) accuracy. 9 A block tridiagonal solver is similar to a scalar solver except that small block matrix operations . Inspecting the top of Eq. Un = Un . 2ndorder. for ensuring a correct steadystate solution in a converged timemarch. 2D equation. 2 hAx I . 2 hAx .
Consider the 1D heat equation: @u = @ u @t @x 2 2 0 x 9 Let u(0 t) = 0 and u(9 t) = 0. 12.12. ii. (a) Space vector de nition i.31) In spite of the similarities in derivation. For example.20 and Eq. 12. PROBLEMS 229 The point at which the factoring is made may not a ect the order of timeaccuracy.7 Problems 1. but it can have a profound e ect on the stability and convergence properties of a method. What is the space vector for the natural ordering (monotonically increasing in index). u ? Only include the interior points. hAx . and if it is immediately factored. 12.20. 12. On the other hand. 12. (1) (2) 12 21 x 2 j . Assume that second order central di erencing is used. hAx] I . the factored form is presented in Eq. 2 . 12. If we reorder the points with the odd points rst and then the even points. i.(P .30 and the O(h) method given by Eq. 2u + u ) xx j The uniform grid has x = 1 and 8 interior points.19. ( u) = 1 (u .19 is h ~ i I . u ? iii. hAy ] Un = h Ax y Un + (bc) + 10 and its factored form becomes h ~ i I . write the space vector.e. the unfactored form of a rstorder method derived from the implicit Euler time march is given by Eq..P ). we will see in the next chapter that the convergence properties of Eq.31 is the appearance of the factor 1 on the left side of the O(h2 ) method. the delta form of the unfactored Eq.1 j j +1 10 Notice that the only di erence between the O(h2 ) method given by Eq. Write down the permutation matrices. 12.31 are vastly di erent.7. hAy ] Un = h Ax y Un + (bc) + (12. 12. so that we can simplify the boundary conditions.
We can put such a partitioning to use. Applying implicit Euler time marching. Comment on the form of the resulting implicit matrix operator.230 CHAPTER 12. write the delta form of the implicit algorithm. (Note f = 0. First de ne extraction operators 21 0 0 0 0 0 0 03 60 1 0 0 0 0 0 07 7 6 60 0 1 0 3 0 0 0 0 7 2I 7 6 7 6 0 7 6 60 0 0 1 0 0 0 07=4 7 Io =60 0 0 0 5 6 0 0 0 07 0 7 6 0 6 7 60 0 0 0 0 0 0 07 7 6 40 0 0 0 0 0 0 05 0 0 0 0 0 0 0 0 20 0 0 0 0 0 0 03 60 0 0 0 0 0 0 07 7 6 60 0 0 0 3 0 0 0 0 7 20 7 6 7 6 0 7 6 60 0 0 0 0 0 0 07=4 7 Ie =60 0 0 0 5 6 1 0 0 07 0 7 6 I 7 6 60 0 0 0 0 1 0 07 7 6 40 0 0 0 0 0 1 05 0 0 0 0 0 0 0 1 (1) (2) 1 2 12 21 ( ) 4 4 4 4 ( ) 4 4 4 4 du = A u + f dt Write down the matrix A . The generic ODE representing the discrete form of the heat equation is (1) 1 (1) v. 1 2 (2) 2 (2) 2 2 . (b) System de nition In problem 1a. we de ned u u A A P . and P which partition the odd points from the even points. SPLIT AND FACTORED FORMS iv. due to the boundary conditions) Next nd the matrix A such that du = A u dt Note that A can be written as 2 3 D UT 7 A =6 4 5 U D De ne D and U .
Beginning with the ODE written in terms of u .12. Comment on their form. Write out the matrices Ao and Ae. Examine the implicit operators for the factored delta form. Comment on the order of the error terms. Write down the delta form of the algorithm and the factored delta form. and Ae operates only on the even terms. ii. de ne a splitting A = Ao + Ae. write them in terms of D and U de ned above. such that Ao operates only on the odd terms. Comment on the solution process this gives us relative to the direct inversion of the original system. You should be able to argue that these are now trangular matrices (a lower and an upper). PROBLEMS (2) ( ) ( ) (2) ( ) ( ) (2) 231 which extract the odd even points from u as follows: u o = I o u and ue =Ieu . i. (2) 2 . iii. Also. Apply implicit Euler time marching to the split ODE.7.
SPLIT AND FACTORED FORMS .232 CHAPTER 12.
233 .7. and we discussed circulant eigensystems1 in Section 4. the results found from this analysis are neither necessary nor su cient to guarantee stability. We have seen that under these conditions a semidiscrete approach can lead to circulant matrix di erence operators. In this and the following section we assume circulant systems and our analysis depends critically on the fact that all circulant matrices commute and have a common set of eigenvectors. the method is probably not suitable for practical use.Chapter 13 LINEAR ANALYSIS OF SPLIT AND FACTORED FORMS In Section 4. and convergence properties of timemarching schemes. if the results indicate that a method has an instability. When these methods are applied to practical problems. fractionalstep. The question is: Can we nd a similar equation that will allow us to evaluate the stability and convergence properties of split and factored schemes? The answer is yes  for certain forms of linear model equations.1 The Representative Equation for Circulant Operators Consider linear PDE's with coe cients that are xed in both space and time and with boundary conditions that are periodic. accuracy. 13. 1 See also the discussion on Fourier stability analysis in Section 7. and (approximately) factored.3. The analysis in this chapter is useful for estimating the stability and steadystate properties of a wide variety of timemarching schemes that are variously referred to as timesplit. However.4 we introduced the concept of the representative equation. hybrid. and used it in Chapter 7 to study the stability.
gm(t) . we can use the arguments made in Section 4.2. ~ (t) u (13. (13. ..4) This is to be contrasted to the developments found later in the analysis of 2D equations.234 CHAPTER 13. 0 wM = ( a + b)M wM . This suggests (see Section 4.3 to uncouple the set and form the M set of independent equations 0 w1 = ( a + b)1 w1 . Since both matrices have the same set of eigenvectors. we arrive at a set of ODE's that can be written d~ = A ~ + A ~ .3) that 13. 0 wm = ( a + b)m wm .2. as a result of space di erencing the PDE. gM (t) (13.. circulant systems is du = + + + ]u + ae t a b c dt where a + b + c + are the sum of the eigenvalues in Aa Ab Ac share the same eigenvector.1) apu b pu f dt where the subscript p denotes a circulant matrix..2) The analytic solution of the m'th line is wm(t) = cm e( a+ b)mt + P:S: Note that each a pairs with one.1 Stability Comparisons of TimeSplit Methods Consider as an example the linear convectiondi usion equation: @u + a @u = @ 2 u @t @x @x2 2 (13. LINEAR ANALYSIS OF SPLIT AND FACTORED FORMS Suppose..2 Example Analysis of Circulant Systems 13. and only one2 . eigenvector. g1 (t) .4: b since they must share a common The representative equation for split.
m = 0 1 M . the characteristic polynomial of this method is P (E ) = (1 .8. respectively. h d )E . Using these values and the representative equation 13. 1 . we can analyze the stability of the two forms of simple timesplitting discussed in Section 12. Eq. 13. EXAMPLE ANALYSIS OF CIRCULANT SYSTEMS 235 If the space di erencing takes the form d~ = . where (see Section 4.1 0 1)~ + u u B (1 . (1 + h c) This leads to the principal root 1 + i ah sin m x = h sin2 m 1+4 2 2 x where we have made use of Eq.10. 2.2. can be represented by the eigenvalues c and d.2. the explicitimplicit Euler method. so that 0 m 2 . 12. In this section we refer to these as 1.3.4.13. The ExplicitImplicit Method 2 (13. a B (.2 1)~ u (13. 4 2 sin2 2 x In these equations m = 2m =M . Eq.2): ( c)m = ia sin m x m (13.4. When applied to Eq.5) dt 2 x p x2 p the convection matrix operator and the di usion matrix operator. Now introduce the dimensionless numbers Cn = ah Courant number x mesh Reynolds number R = a x and we can write for the absolute value of q 2 1 + Cn sin2 m j j= C m 1 + 4 R n sin2 2 0 m 1. the explicitexplicit Euler method.6 to quantify the eigenvalues. 13.7) . 12.6) ( d)m = .
2 C = 2/ R ∆ n 0.7 shows that the critical range of m for any combination of Cn and R occurs when m is near 0 (or 2 ).8 C n C 0. A simple numerical parametric study of Eq.8 n 1. 4 Cn sin2 m 2 j j = 1 + Cn sin 0 m 2 R 2 Again a simple numerical parametric study shows that this has two critical ranges of m . From this we nd that the condition on Cn and R that make j j 1 is 2 h i C 2 1 + Cn sin2 = 1 + 4 R n sin2 2 As ! 0 this gives the stability region 2 Cn < R which is bounded by a hyperbola and shown in Fig. The ExplicitExplicit Method An analysis similar to the one given above shows that this method produces " # q 2 m 1 .236 CHAPTER 13.1: Stability regions for two simple timesplit methods.4 0 0 4 R∆ ExplicitImplicit 8 0 0 4 R∆ ExplicitExplicit 8 _ Figure 13. 13.2 C = R /2 n ∆ C = 2/ R n ∆ 0. 13.1. which yields the same result as in the previous example. LINEAR ANALYSIS OF SPLIT AND FACTORED FORMS 1. 2. and the .4 0. one near 0.
which produces the constraint that 1 Cn < 2 R for R 2 The resulting stability boundary is also shown in Fig. as well as the stability. un. This method applies to a ow in which there is a combination of di usion and periodic convection. 1 h d)E 2 . The convection term is treated explicitly using the secondorder AdamsBashforth method.9) Q(E ) = 1 h(E 2 + E ) 2 13. 13.2 Analysis of a SecondOrder TimeSplit Method . (1 + 2 h c + 1 h d)E + 1 h c 2 2 2 The form of the particular polynomial depends upon whether the forcing function is carried by the AB2 method or by the trapezoidal method. In the former case it is 1 Q(E ) = 2 h(3E . timemarching method on the equation u0 = cu + du + ae t First let us nd expressions for the two polynomials.8) and in the latter (13. The characteristic polynomial follows from the application of the method to the homogeneous equation.1.4 which we split in the fashion of Eq. The di usion term is integrated implicitly using the trapezoidal method. thus 1 1 un+1 = un + 2 h c(3un .13. In order to evaluate the accuracy.2. factored method has a much smaller region of stability when R is small. Our model equation is again the linear convectiondi usion equation 13.5. P (E ) and Q(E ). 13.1) + 2 h d (un+1 + un) This produces 3 P (E ) = (1 . Next let us analyze a more practical method that has been used in serious computational analysis of turbulent ows. as we should have expected.2. we include the forcing function in the representative equation and study the e ect of our hybrid. EXAMPLE ANALYSIS OF CIRCULANT SYSTEMS 237 other near 180o. 1) (13. The totaly explicit.
.9. The results are plotted in Fig. 13.3.6. 1 1 + 3h c + 2h 2 2 1 . LINEAR ANALYSIS OF SPLIT AND FACTORED FORMS Accuracy From the characteristic polynomial we see that there are two {roots and they are given by the equation 3 1 + 2h c + 1h 2 = s d The principal root follows from the plus sign and one can show 1 1 1 = 1 + ( c + d )h + 2 ( c + d )2 h2 + 4 3 + c 2 .7. so er = O(h3) Using P (e h) and Q(e h ) to evaluate er in Section 6.3 The Representative Equation for SpaceSplit Operators Consider the 2D model3 equations @u = @ 2 u + @ 2 u @t @x2 @y2 3 (13. 2 h d (13.11) The extension of the following to 3D is simple and straightforward. 13.10 by a parametric study of cn and R de ned in Eq.238 CHAPTER 13. 3 h3 d d c c 1 1 From this equation it is clear that 6 3 = 6 ( c + d)3 does not match the coe cient of h3 in 1 .10) Stability The stability of the method can be found from Eq. 13.2. 13. 13.1. 1h d 2 d 2 1 . This was carried out in a manner similar to that used to nd the stability boundary of the rstorder explicitimplicit method in Section 13. 12.2. For values of R 2 this secondorder method has a much greater region of stability than the rstorder explicitimplicit method given by Eq. one can show er = O(h3) using either Eq. These results show that.8 or Eq. 2h c 1 . 13. 13.1.10 and shown in Fig. 2 d . for the model equation. the hybrid method retains the secondorder accuracy of its individual components.
23 for the 3 4 mesh shown in Sketch 12.1 b0 b1 ).2: Stability regions for the secondorder timesplit method.12. Now nd the block eigenvector .1 I ~0 I B b b where B is a banded matrix of the form B (b. 12.13) x y dt for the space vector U .12) @t x @x y @y Reduce either of these.8 Cn 0. Let us inspect the structure of these matrices closely to see how we can diagonalize Ax + Ay ] in terms of the individual eigenvalues of the two matrices considered separately.3. First we write these matrices in the form 2 3 2 ~ 3 B b0 I ~1 I b A(xx) = 6 B 7 A(yx) = 6 ~. by means of spatial di erencing approximations.13.2 0.1 I ~0 I ~1 I 7 b b 5 4 5 4b ~. The form of the Ax and Ay matrices for threepoint central di erencing schemes are shown in Eqs.22 and 12.4 Stable 0 4 8 ∆ R _ Figure 13. to the coupled set of ODE's: dU = A + A ]U + (bc) ~ (13. and @u + a @u + a @u = 0 (13. THE REPRESENTATIVE EQUATION FOR SPACESPLIT OPERATORS239 1.
b b b ~ that diagonalize the B blocks.1 4b I I . Let B diag(B ) and ~ ~ ~ Next nd the eigenvectors X ~ diag(X ) and form the second transformation ~ X 2~ 3 2~ 3 1 6 7 7 ~2 ~ ~~ 6 ~ ~ 7 ~=6 X . 13.13. see the bottom of Eq. There results h i Pyx X .1 6 ~.1 A(xx+)y X Pxy X = 2 3 1I + ~ 6 7 6 7 2I + ~ 6 7 (13.1 B 76 X 4 54 54 5 4 .1 I ~0 I b b b b One now permutes the transformed system to the yvector database using the permutation matrix de ned by Eq. LINEAR ANALYSIS OF SPLIT AND FACTORED FORMS matrix that diagonalizes B and use it to diagonalize A(xx).15) 6 7 4 5 3I + ~ 4I + ~ diag(X ) 2 ~ b0 X . That is.1 ~0 ~1).1 32 32 3 2 X B X 6 76 7=6 X .1 A(xx) + A(yx) X Pxy = 3 2 3 2 ~ B 1 I 7 6 7 6 ~ 7 B 6 7+6 2 I 6 7 6 7 6 (13.14 is transparent to the transformation. if we Notice that the matrix Ay 3 7 7 7 5 3 2 ~ 3 ~1 I b b0 I ~1 I b ~0 I ~1 I 7 X = 6 ~.1B X = 6 7 4 5 4 5 ~3 ~ This time.1 X B X where 2 3 6 =6 6 4 3 7 5 1 2 set X 4 (x) is transparent to this transformation.1 I ~0 I ~1 I 7 b b 5 b b 5 4b ~.23.1 I ~0 I ~. by the same argument as before.240 CHAPTER 13. 12.14) ~ 4 5 4 B 7 3 I 5 ~ B 4 I ~ where B is the banded tridiagonal matrix B (~.1 Pyx X . Thus 2 . 12. the rst matrix on the right side of Eq. so the nal result is the complete diagonalization of the matrix Ax+y : h ih ih i ~ ~ X .
but this choice was for convenience only. Often we are interested in nding the value of.3.17) x y dt which has the exact solution a u(t) = ce( x+ y )t . + (13. First reduce the PDE to ODE by some choice4 of space di erencing. THE REPRESENTATIVE EQUATION FOR SPACESPLIT OPERATORS241 It is important to notice that: The diagonal matrix on the right side of Eq. 13.16) ~ where B and B are any two matrices that have linearly independent eigenvectors (this puts some constraints on the choice of di erencing schemes). 4 . To obtain the latter property the structure of ~ the matrices is important. The block matrices B and B can be either circulant or noncirculant in both cases we are led to the nal result: The 2{D representative equation for model linear systems is du = + ]u + ae t x y dt where x and y are any combination of eigenvalues from Ax and Ay .18) x y We have used 3point central di erencing in our example. 13. and the convergence rate to. and where Ax and Ay satisfy the conditions in 13. and its use is not necessary to arrive at Eq.13. Although Ax and Ay do commute.15. Now we are ready to present the representative equation for two dimensional systems.15 contains every possible com~ bination of the individual eigenvalues of B and B .16. does not ensure the property of \all possible combinations". by itself. a and are (possibly complex) constants. In that case we set = 0 and use the simpler form du = + ]u + a (13. This results in a spatially split A matrix formed from the subsets A(xx) = diag(B ) ~ A(yy) = diag(B ) (13. this fact. the steadystate solution of the representative equation.
h .19) un = c 1 . 3. LINEAR ANALYSIS OF SPLIT AND FACTORED FORMS 13. h from which x . and when it is applied to the representative equation. " 1 #n a x+ y 2. Is unconditionally stable. this method: 1.19. Converges very rapidly to the steadystate when h is large. 12. 3. steadystate solution.16. In each case we examine 13. 1 Q(E ) = h giving the solution (13.242 CHAPTER 13.5. The form of the method is given by Eq. Produces the exact (see Eq. h y )un+1 = un + ha P (E ) = (1 . Unfortunately.18) steadystate solution (of the ODE) for any h. The accuracy of the xed. . h y )E .1 The Unfactored Implicit Euler Method Consider rst this unfactored. The convergence rate to reach the steadystate. we nd (1 . use of this method for 2D problems is generally impractical for reasons discussed in Section 12. 13. 2. rstorder scheme which can then be used as a reference case for comparison with the various factored ones. In the following we analyze four di erent methods for nding a xed.4.4 Example Analysis of 2D Model Equations 1. steadystate solution to the 2D representative equation 13. h x . x y Like its counterpart in the 1D case. however. The stability. h .
4. Produces a steady state solution that depends on the choice of h.4. h y )(un+1 . + x y x y This method: 13. un) = h( xun + y un + a) which reduces to (1 . h )(1 . h ) .31 to the 2D representative equation.13. x y We see that this method: 1. Converges rapidly to a steadystate for large h. but the converged solution is completely wrong. h y )un+1 = 1 + h2 x y un + ha and this has the solution " #n 1 + h2 x y a un = c (1 . EXAMPLE ANALYSIS OF 2D MODEL EQUATIONS 243 Now apply the factored Euler method given by Eq. h x)(1 . h y )un+1 = un + ha from which P (E ) = (1 . 3.3 The Factored Delta Form of the Implicit Euler Method . However. h )(1 . One nds (1 .20 to the 2D representative equation. 12. h x)(1 . h x)(1 .20) giving the solution " #n 1 un = c (1 . 1 Q(E ) = h (13.2 The Factored Nondelta Form of the Implicit Euler Method 13. h y )E . There results (1 . h ) . h x)(1 .4. 12. if one tries to take advantage of its rapid convergence rate. The method requires far less storage then the unfactored form. Is unconditionally stable. Next apply Eq. + a h x y x y. it is not very useful since its transient solution is only rstorder accurate and. the converged value is meaningless. 2.
The correct steady solution is obtained. 2 h x 1 . but convergence is not nearly as rapid as that of the unfactored form. since j j ! 1 as h ! 1. 12. as discussed in Section 12. Like the factored nondelta form. Is unconditionally stable.27 should be enough to convince the reader that the 's produced by those methods are identical to the produced by this method. 1 h x 1 .30 to the representative equation and one nds 1 1 . 3. In practical codes.244 CHAPTER 13. the value of h on the left side of the implicit equation is literally switched from h to 1 h. 3. and the factored delta form of the implicit Euler method can be used when a converged steadystate is all that is required. Produces the exact steadystate solution for any choice of h. Produces the exact steady{state solution for any choice of h. LINEAR ANALYSIS OF SPLIT AND FACTORED FORMS 1. 1 h y (un+1 . Finally consider the delta form of a secondorder timeaccurate method. 12. All of these properties are identical to those found for the factored delta form of the implicit Euler method. since j j ! 1 as h ! 1. Converges very slowly to the steady{state solution for large values of h. 1 h y un+1 = 1 + 1 h x 1 + 1 h y un + ha 2 2 2 2 and this has the solution 2 3n 1 + 1h x 1 + 1h y 7 6 a 2 2 7 . Converges very slowly to the steady{state solution for large values of h.4. 2. 2h y 2 This method: 1. un = c6 4 5 1 x+ y 1 .4 The Factored Delta Form of the Trapezoidal Method . 2. this method demands far less storage than the unfactored form.5. it can be used when time accuracy is desired. 2 5 13. 1h x 1 .5 A brief inspection of eqs. un) = h( xun + y un + a) 2 which reduces to 1 .26 and 12. Is unconditionally stable. Since it is second order in time. Apply Eq.
13.11 and 13.5 Example Analysis of the 3D Model Equation 245 The arguments in Section 13. nd the root. +a+ x y 6 z (13.3 generalize to three dimensions and.22. 2 h x 1 . 1 h( x + y + z ) un = h ( x + y + z )un + a] 2 Now factor the left side: 1 1 . First apply the trapezoidal method: un+1 = un + 1 h ( x + y + z )un+1 + ( x + y + z )un + 2a] 2 Rearrange terms: 1 1 1 . under the conditions given in 13.23) Eqs. 8 x y z .16 with an A(zz) included. each with an additional term. 1 h y 1 . and write the solution either in the form 2 3n 1 h( + + ) + 1 h2 ( 1 h3 + + ). 13.5. EXAMPLE ANALYSIS OF THE 3D MODEL EQUATION 13.21) x y z dt Let us analyze a 2ndorder accurate. factored. x y z 61 + 4 x y x z y z 8 x y z7 7 un = c6 2 4 5 1 h( + + ) + 1 h2( 1 h3 1 .12.22) 2 2 This preserves second order accuracy since the error terms 1 h2( 1 h3 x y + x z + y z ) un and 4 8 x y z are both O(h3). 2 x y z 4 x y + x z + y z) . 13. 2 h( x + y + z ) un+1 = 1 + 2 h( x + y + z ) un + ha Put this in delta form: 1 . . 1 h z un = h ( x + y + z )un + a] (13. the model 3D cases6 have the following representative equation (with = 0): du = + + ]u + a (13. One can derive the characteristic polynomial for Eq. delta form using this equation.
Furthermore.22 should not be used for the 3D Euler equations. +i i . ) + ( + ) j (1 . In practical cases. the method is stable for all h. LINEAR ANALYSIS OF SPLIT AND FACTORED FORMS or in the form 2 6 un = c6 4 1 1 1 + 1 h x 1 + 2 h y 1 + 1 h z . Now consider what happens when we apply this method to the biconvection model. From the above analysis one would come to the conclusion that the method represented by Eq.23 that. 13. however.7 With regards to stability. First write the root in Eq.24) It is interesting to notice that a Taylor series expansion of Eq. some form of dissipation is almost always added to methods that are used to solve the Euler equations and our experience to date is that. 13. 2 x 1 . In that case since the absolute value of the product is the product of the absolute values j 2 2 2 = (1 . 13. Remember that in our analysis we must consider every possible combination of these eigenvalues. in the presence of this dissipation. if the method converges.24 results in 1 = 1 + h( x + y + z ) + 2 h2( x + y + z )2 (13.25) h i + 1 h3 3 + (2 y + 2 x) + 2 2 + 3 x y + 2 2 + 3 + 2 x 2 + 2 2 y + 3 + y y y y x x 4 z which veri es the second order accuracy of the factored form. Now we can always nd one combination of the 's for which . and are real numbers that can have any sign. . 2 y 1 . 1h z 2 3n x y z7 7 .12 with periodic boundary conditions. 5 x+ y+ z a (13. This makes the method unconditionally stable for the 3D di usion model when it is centrally di erenced in space. the 3D form of Eq. central di erencing causes all of the 's to be imaginary with spectrums that include both positive and negative values. 4 h3 2 2 1h 1h 1 . it converges to the proper steadystate.i . )2 >1 and the method is unconditionally unstable for the model convection problem. the instability disclosed above is too weak to cause trouble. we already knew this because we chose the delta form. 13. it follows from Eq. 7 However. and are both positive.246 CHAPTER 13.23 in the form 1+ = 1. +i where . if all the 's are real and negative. In this case. )2 + ( . 13. clearly.
(d) Repeat 1c for the explicitimplicit scheme of problem 1b. du = Au + f dt we can split A as follows: A = A1 + A2 + A3 + A4. (c) The scalar representative equation is du = ( + + + )u + a 1 2 3 4 dt For the fully implicit scheme of problem 1a. and accuracy of the converged steadystate solution. Applying implicit Euler time marching gives un+1 . .13. Starting with the generic ODE. leave two explicit: un+1 .6 Problems 247 1.6. PROBLEMS 13. convergence. nd the exact solution to the resulting scalar di erence equation and comment on the stability. un = A u + A u + A u + A u + f 1 n+1 2 n+1 3 n+1 4 n+1 h (a) Write the factored delta form. What is the error term? (b) Instead of making all of the split terms implicit. un = A u + A u + A u + A u + f 1 n+1 2 n 3 n+1 4 n h Write the resulting factored delta form and de ne the error terms.
248 CHAPTER 13. LINEAR ANALYSIS OF SPLIT AND FACTORED FORMS .
A general m m matrix A can be written 2 a11 a12 a1m 3 6a a a 7 A = (aij ) = 6 21 22 . . In the present context a vector is a vertical column or string. Thus 2 v1 3 6 7 ~ = 6 v.. .2 7 v 6 7 4 . Given below is some notation and some of the important relations between matrices and vectors. A. 5 vm and its transpose ~ is the horizontal row vT ~ T = v1 v2 v3 : : : vm] ~ = v1 v2 v3 : : : vm ]T v v 2.Appendix A USEFUL RELATIONS AND DEFINITIONS FROM LINEAR ALGEBRA A basic understanding of the fundamentals of linear algebra is crucial to our development of numerical methods and it is assumed that the reader is at least familar with this subject area.1 Notation 1. 2m 7 7 6 5 4 am1 am2 amm 249 .
250APPENDIX A. An alternative notation for A is h i A = ~1 ~2 : : : ~m a a a and its transpose AT is 2 ~T 3 a 6 ~1 7 6 aT 7 AT = 6 . A is symmetric if AT = A. 4. A is skewsymmetric or antisymmetric if AT = . AH is the conjugate transpose of A. where I is the identity matrix.. USEFUL RELATIONS AND DEFINITIONS FROM LINEAR ALGEBRA 3.1 and has the property that A.2 7 6 7 6 . 10. A is diagonally dominant if aii Pj6=i jaij j i = 1 2 : : : m and aii > Pj6=i jaij j for at least one i. 7 4 T5 ~m a 4. 9. P is a permutation matrix if P ~ is a simple reordering of ~ . 6. A. The trace of a matrix is Pi aii.c . v v 7.1 = I . 3.1 = det1 A] . The inverse of a matrix (if it exists) is written A. A is orthogonal if aij are real and AT A = AAT = I 5. If b A= a d c then det A] = ad . (Hermitian).b a . A is normal if AT A = AAT . 8. 2.1A = AA.2 De nitions 1. 11. A is the complex conjugate of A.A. bc and d A. det A] is the determinant of A.
3 Algebra We consider only square matrices of the same dimension. sA = (saij ) where s is a scalar. If it is asymmetric. Any matrix A can be expressed as the sum of a symmetric and a skewsymmetric matrix. .1 (AT ). 4. I ]~ = 0 x x x and the generalized eigenvalue problem. etc. its eigenvalues are all real. they are all imaginary. Thus: 1 A = 2 A + AT + 1 A . 3. ALGEBRA A.1 = B .A. A + (B + C ) = (C + A) + B .1 = (A. Inverse equalities (if the inverse exists): (A. If a square matrix with real elements is symmetric. including the matrix B . In general AB 6= BA. 5. as A~ = B~ or A .1 A. Transpose equalities: (A + B )T = AT + B T (AT )T = A (AB )T = B T AT 6. A and B are equal if aij = bij for all i j = 1 2 : : : m.1 = A (AB ). 2.1 )T 251 7. 1.1). B ]~ = 0 x x x 2. The eigenvalue problem for a matrix A is de ned as A~ = ~ or A .4 Eigensystems 1.3. AT 2 A.
A set of eigenvectors is said to be linearly independent if a ~ m + b ~ n 6= ~ k x x x m 6= n 6= k for any complex a and b and for all combinations of vectors in the set. and with each distinct eigenvalue there is one associated eigenvector. 8. an m m matrix A has n~ linearly independent eigenvectors with x n~ m and n distinct eigenvalues ( i) with n n~ m. If A has m distinct eigenvalues. X . 2 0 03 1 6 . in which case the possibility exists that it cannot be diagonalized.2 . In general. the matrix is said to be derogatory. but all of the eigenvectors are linearly independent. . .e. If a matrix does not have a complete set of linearly independent eigenvectors. .. . i. 4. USEFUL RELATIONS AND DEFINITIONS FROM LINEAR ALGEBRA 3. . 7 6 7 = 6 0 .1AX = where X is a matrix whose columns are the eigenvectors. and A is said to have a complete eigensystem. it cannot be diagonalized.. If the eigenvalues of a matrix are not distinct. its eigenvectors completely span the space. but it can still be diagonalized. The eigenvectors of such a matrix cannot span the space and the matrix is said to have a defective eigensystem. 0 7 4 5 . Gershgorin's theorem: The eigenvalues of a matrix lie in the complex plane in the union of circles having centers located by the diagonals with radii equal to the sum of the absolute values of the corresponding o diagonal row elements. 9. . If A posseses m linearly independent eigenvectors then A is diagonalizable. the eigenvalues of a matrix may not be distinct. x x 5.. In general. 7 6 . then A is always diagonalizable.252APPENDIX A. and this eigenvector cannot be formed from a linear combination of any of the other eigenvectors. . h i X = ~1 ~2 : : : ~m x x x and is the diagonal matrix 0 0 m If A can be diagonalized. 6.. . 7.
The other r . .. such that where there are k linearly independent eigenvectors and Ji is either a Jordan subblock or i. For each Jordan submatrix with an eigenvalue i of multiplicity r. 0 7 Ji = 6 7 i 6. .4..A.7 J =S 6 . 7 6..7 6 . Some of the Jordan subblocks may have the same eigenvalue. EIGENSYSTEMS 2J 0 03 1 . A repeated root in a Jordan block is referred to as a defective eigenvalue... . 0 7 4 5 0 0 Jk 253 10. A Jordan submatrix has the form 2 0 03 i 1 60 . . . . . 6 . . Use of the transform S is known as putting A into its Jordan Canonical form. 1 7 4 5 0 0 0 i 12.. Note that if P is the permutation matrix 2 3 0 0 1 P = 60 1 07 4 5 1 0 0 P T = P .. there exists one eigenvector. . Defective matrices cannot be diagonalized but they can still be put into a compact form by a similarity transform.1 6 0 17P = 61 07 4 5 4 5 0 0 0 1 14. 7 .1 = P then 2 3 2 3 1 0 0 0 P . . the . . S . . 7 6 7 i 1 6 60 0 7 .1 AS = 6 0 J2 .. 13. The complete set of principal vectors and eigenvectors are all linearly independent. 1 vectors associated with this eigenvalue are referred to as principal vectors. ... For example. 11. .
The spectral radius of a matrix A is symbolized by (A) such that (A) = j m jmax where m are the eigenvalues of the matrix A. 2. having: 9 eigenvalues 3 distinct eigenvalues 3 Jordan blocks 5 linearly independent eigenvectors 3 principal vectors with 1 1 principal vector with 2 A. USEFUL RELATIONS AND DEFINITIONS FROM LINEAR ALGEBRA matrix 22 66 64 6 6 6 6 6 6 6 6 6 6 6 4 1 1 1 3 17 5 1 1 1 1 1 2 1 2 3 3 7 7 7 7 7 7 7 7 7 7 7 7 7 5 is both defective and derogatory.254APPENDIX A.5 Vector and Matrix Norms 1. A pnorm of a matrix A is de ned as Av jjAjjp = max jjjjvjjjjp x6=0 p . A pnorm of the vector ~ is de ned as v 0M 1 X pA1=p jjvjjp = @ jvj j j =1 3.
in fact. VECTOR AND MATRIX NORMS 255 4. in this case it is the L2 norm. 8. jjAjj2 = q PM ja j maximum column sum i=1 ij (A T A ) . All matrix norms must have the properties jjAjj 0 jjAjj = 0 implies A = 0 jjc Ajj = jcj jjAjj jjA + B jj jjAjj + jjB jj jjA B jj jjAjj jjB jj 5. is the lower bound of all the norms of A. Special pnorms are jjAjj1 = maxj=1 M jjAjj1 = maxi=1 2 M PM jaij j maximum row sum j =1 where jjAjjp is referred to as the Lp norm of A. The spectral radius of A.5. 7. Let A and B be square matrices of the same order. When A is normal.A. 6. In general (A) does not satisfy the conditions in 4. (A). so in general (A) is not a true norm. (A) is a true norm.
256APPENDIX A. USEFUL RELATIONS AND DEFINITIONS FROM LINEAR ALGEBRA .
2) and is given by j.4. Many of these can be derived by solving the simple linear di erence equations that arise in deriving recursion relations.Appendix B SOME PROPERTIES OF TRIDIAGONAL MATRICES B. It is useful to list some of their properties.1 ~ m = (xj )m = a 2 sin j m x m=1 2 M (B. Let us consider a simple tridiagonal matrix. we nd (by a recursion exercise) det B (M : a b c)] = 0 if p m b + 2 ac cos M + 1 = 0 m=1 2 M From this it follows at once that the eigenvalues of B (a b c) are p m m=1 2 M (B. see Section 3. and c.1) m = b + 2 ac cos M +1 The righthand eigenvector of B (a b c) that is associated with the eigenvalue m satis es the equation B (a b c)~ m = m~ m x x (B.e.3) c M +1 257 .1 Standard Eigensystem for Simple Tridiagonals In this work tridiagonal banded matrices are prevalent. a tridiagonal with constant scalar elements a.b. If we examine the conditions under which the determinant of this matrix is zero. i..
md)(c . m f ) cos M + 1 = 0 m=1 2 If we de ne m m= M + 1 pm = cos m 1 2 3 1 2 3 (B. d b .1 and c = 1 . ... the elements of which are j. f 7 6 .6) B.8) . c m 2 1 = e. SOME PROPERTIES OF TRIDIAGONAL MATRICES These vectors are the columns of the righthand eigenvector matrix. c m 2 1 sin mj m=1 2 . 7 6 76 7 6 76 7 4 54 5 4 54 5 a b xM d e xM In this case one can show after some algebra that det B (a .i m. j.1 a 2 = ei j.1 a 2 sin jm j=1 2 M X = (xjm) = c (B..2 Generalized Eigensystem for Simple Tridiagonals This system is de ned as follows 2 32 32 x 3 2e f x 3 b c 6a b c 76 x 7 6d e f 76 x 7 6 76 7 6 76 7 6 76 7 6 76 6 76 x 7 = 6 76 x 7 7 a b d e 6 .4) m=1 2 M M +1 Notice that if a = .1 and c = 1. . 2 a 1 ( 1) M M (B. f ] = 0 if q m b .5) c ( 1) The lefthand eigenvector matrix of B (a b c) can be written . m e + 2 (a . c 7 6 .7) M (B. 7 6 .. e c .258 APPENDIX B. . . 2 (B. = 2 X j=1 2 M +1 a M +1 In this case notice that if a = .
3. one can show that 0 0 1 2 3 2 3 1 2 2 2 +1 2 +1 2 2 0 1 b DM = (M + 1) 2 !M b = 4ac 2 . thus D = 1 D = b D = b . f 2 2 2 259 2 q m These relations are useful in studying relaxation methods.3 The Inverse of a Simple Tridiagonal The inverse of B (a b c) can also be written in analytic form. af )pm ] m= e . . B. it is simple to derive the rst few determinants. m d 2 sin j ] m =1 2 M ~m = x m j =1 2 M c. 4ac #M 1 b . Its characteristic polynomial P (E ) is P (E . 2(cd + af )pm + 2pm (ec . THE INVERSE OF A SIMPLE TRIDIAGONAL eb .10) Eq.B. fb)(ea . b . acDM . 2 2 b . Let DM represent the determinant of B (M : a b c) DM det B (M : a b c)] De ning D to be 1.2. In the limiting case when b .11) where we have made use of the initial conditions D = 1 and D = b. (B. 2abc (B.10 is a linear O E the solution of which was discussed in Section 4. 4ac M = DM = p . bd) + (cd . 4ac = 0. 4ac : M =0 1 2 (B. 1 " a . B. ac D = b . bE + ac) and the two roots to P ( ) = 0 result in the solution 8" 9 p p " # < b + b .9) One can also nd the recursion relation DM = bDM . 4fdpm The righthand eigenvectors are #j .
n(.c D c D D D .m =DM 1 Lower triangle: m =n+1 n+2 M 1 n=1 2 M .cD .a)m.a D a D .c D . (.n=DM B.cD D c D D .cD D c D a D D .4 Eigensystems of Circulant Matrices B.aD D D D .c D .m =DM Diagonal: n=m=1 2 M dmm = DM .c)n. = 1 D .aD D D D .mDn.1 dmn = DM .1 = aD The general element dmn is 4 0 2 D4 6 . SOME PROPERTIES OF TRIDIAGONAL MATRICES 2 16 6 4 D4 6 Then for M = 4 B.260 APPENDIX B.cD D c D a D .a3 D1 4 cD .4.12) .4.1 Standard Tridiagonals Consider the circulant (see Section 3.cD . DM .cD cD .aD D D D .aD D 3 2 2 3 1 3 1 2 2 1 4 1 1 3 0 2 1 2 2 2 2 1 2 1 3 1 1 2 2 1 3 1 2 3 4 .aD3 1 6 a2 D 6 2 6 D5 6 .cD 3 7 17 7 2 7 7 5 3 Upper triangle: m=1 2 M .1 1 n=m+1 m+2 M dmn = Dm.4) tridiagonal matrix Bp(M : a b c ) (B.aD D 3 2 2 1 3 2 2 1 2 1 1 2 1 3 1 0 2 1 1 2 1 1 2 3 0 2 3 7 7 7 5 and for M = 5 B .aD D D . DM .a D aD .
= (x0 ) = 1 e. the eigenvalues are not.1 X = (xjm) = eij M m =0 1 M . 1 1 1 M .1 The lefthand eigenvector matrix with elements x0 is 2 j m =0 1 M .1 . c) sin 2 m M M (2 ) m=0 1 2 The righthand eigenvector that satis es Bp(a b c)~ m = m~ m is x x ~ m = (xj )m = ei j m=M x j=0 1 M . B. j =0 .im M X mj j =0 1 M . where X is the conjugate transpose of X .15) 4 b b b b 5 b b b b The eigenvectors are always given by eq. Although the eigenvectors of a circulant matrix are independent of its elements. An example of a dense circulant matrix of order M = 4 is 2 b b b b 3 6 b b b b 7 6 7 6 7 (B.4. B.4.2 General Circulant Systems Notice the remarkable fact that the elements of the eigenvector matrices X and X .12 do not depend on the elements a b c in the matrix.1 (B. and the righthand eigenvector matrix has the form 2 m j =0 1 M .13) B.15 they have the general form M.13 is a special case.14) p where i . i(a .B. are symmetric and that X . For the element indexing shown in eq. all circulant matrices of order M have the same set of linearly independent eigenvectors. B.1.1 (B. EIGENSYSTEMS OF CIRCULANT MATRICES The eigenvalues are m 261 = b + (a + c) cos 2 m .1 M 1 Note that both X and X . for the tridiagonal circulant matrix given by eq. B. = M X . In fact.14. X bj ei jm=M = m 1 0 3 2 1 1 0 3 2 2 1 0 3 3 2 1 0 1 (2 ) of which eq. and further examination shows that the elements in these eigenvectors correspond to the elements in a complex harmonic analysis or complex discrete Fourier series. even if they are completely dense.
. SOME PROPERTIES OF TRIDIAGONAL MATRICES Consider a mesh with an even number of interior points such as that shown in Fig. 7 6 . 7 76 . 7 76 . 7 76 .5 Special Cases Found From Symmetries APPENDIX B.. 7 6 .16 are (2m .. 7 6 . 7 m6 . 7 6 .. 1) m = b + 2a cos 2M + 1 ! m=1 2 M (B.. One can seek from the tridiagonal matrix B (2M : a b a ) the eigenvector subset that has even symmetry when spanning the interval 0 x .262 B. B.1. 7 6 . we seek the set of eigenvectors ~ m for which x 2 b 6a 6 6 6 6 6 6 6 4 a b a a . a a b a b 32x 3 1 7 6 x2 7 76 7 76 . a 7 7 a b a 7 5 a b+a 2 6 6 6 6 6 6 6 6 6 6 4 3 x m~ m (B. 7~ a ~ a)~ m = 7 xm = B (M : a b x 7 .. B.17) . 7 76 . 7 54 5 a 7 6 x2 7 = x 2 3 x1 6x 7 6 27 6 . one can show from previous results that the eigenvalues of eq. 7 6 7 4 x2 5 1 x 1 This leads to the subsystem of order M which has the form b a 7 7 a b a 7 7 . For example. ..16) By folding the known eigenvectors of B (2M : a b a) about the center..
1) x 2M + 1 j =1 2 M Imposing symmetry about the same interval but for a mesh with an odd number of points.1 { Symmetrical folds for special cases m=1 2 M ~ m = sin j (2m . B. 6 7 a a B (M : ~ b a) = 6 . 2 2 .1.6 Special Cases Involving Boundary Conditions We consider two special cases for the matrix operator representing the 3point central di erence approximation for the second derivative @ =@x at all points away from the boundaries. leads to the matrix 2 3 b a 6a b a 7 6 7 6 7 . 1 : a b a) about the center. SPECIAL CASES INVOLVING BOUNDARY CONDITIONS and the corresponding eigenvectors are ~ m = sin j (2m .6.. a 7 6 7 6 7 6 4 5 a b a7 2a b By folding the known eigenvalues of B (2M . see Fig. one can show from previous results that the eigenvalues of eq. combined with special conditions imposed at the boundaries. An evennumbered mesh Line of Symmetry x =0 x= j0 = 1 2 3 4 5 M0 j= 1 2 3 M b.. B.. An odd{numbered mesh Figure B.B.17 are (2m . 1) x 2M ! j=1 2 M B.. 1) m = b + 2a cos 2M and the corresponding eigenvectors are ! 263 Line of Symmetry x =0 x= j0 = 1 2 3 4 5 6 M0 j= 1 2 3 M a.
4 sin ( =2) 2 When the boundary conditions are Dirichlet on both sides.2 1 1 .2 1 1 .2 1 1 .264 APPENDIX B.2 h+ 2 cos Mi+ 1 m = sin j M + 1 (B. 2 6 6 6 6 6 6 4 .2 3 7 7 7 7 7 7 5 m ~m x m = .18) When one boundary condition is Dirichlet and the other is Neumann (and a diagonal preconditioner is applied to scale the last equation).2 1 1 .1 3 7 7 7 7 7 7 5 ~m x m = . 1) = sin j 2M + 1 (B. 1) 2M + 1 (2m .2 1 1 .2 1 1 . SOME PROPERTIES OF TRIDIAGONAL MATRICES Note: In both cases m =1 2 M j =1 2 M .2 + 2 cos( ) = .2 1 1 .19) .2 1 1 .2 + 2 cos (2m . 2 6 6 6 6 6 6 4 .
Consider rst the scalar case.Appendix C THE HOMOGENEOUS PROPERTY OF THE EULER EQUATIONS The Euler equations have a special property that is sometimes useful in constructing numerical methods. F is called homogeneous of degree n. we nd Consider next the theorem as it applies to systems of equations.1) for a xed n. F is said to be homogeneous of degree n and we nd " (C. In order to examine this property. If the vector F (Q) satis es the identity u @F + v @F = nF (u v) @u @v (C.3) @F Q = nF (Q) @q 265 # (C. If F (u v) satis es the identity F ( u v) = nF (u v) (C.4) .2) F ( Q) = nF (Q) for a xed n. Di erentiating both sides with respect to and setting = 1 (since the identity holds for all ). let us rst inspect Euler's theorem on homogeneous functions.
7) (C.12 are homogeneous of degree 1.11 and 2.5) E = An Q + O(h2) F = BnQ + O(h2) (C. 6. THE HOMOGENEOUS PROPERTY OF THE EULER EQUATIONS Now it is easy to show. and their Jacobians.4. see eq. The Euler equations are homogeneous if the equation of state can be written in the form p = f ( ). we notice that the expansion of the ux vector in the vicinity of tn which. by direct use of eq. if F is homogeneous of degree 1. Notice also that. 6. E = En + An(Q . Qn) + O(h2) can be written (C.8) @A Q = 0 @x in spite of the fact that individually @A=@x] and Q are not equal to zero.9) Note that this depends on the form of the equation of state. As a nal remark. according to eq. under this condition. are homogeneous of degree 0 (actually the latter is a direct consequence of the former). 2. A and B . C.1 This being the case.105 can be written in general as.6) since the terms En . C. we notice from the chain rule that for any vectors F and Q @F (Q) = @F @Q = A @Q @x @Q @x @x We notice also that for a homogeneous F of degree 1. where is the internal energy per unit mass.106. .3. the constant term drops out of eq. " # " # (C. F = AQ and @F = A @Q + @A Q @x @x @x Therefore. Qn) + O(h2) F = Fn + Bn(Q . that both E and F in eqs. 1 " # (C. AnQn and Fn . BnQn are identically zero for homogeneous vectors of degree 1.266APPENDIX C.
Anderson. volume 1. 2] C. Hirsch. 1988. and R. New York. J. Tannehill. 267 . Pletcher. John Wiley & Sons. 1984. McGrawHill Book Company.Bibliography 1] D.2. New York. Numerical Computation of Internal and External Flows. Computational Fluid Mechanics and Heat Transfer.
This action might not be possible to undo. Are you sure you want to continue?
We've moved you to where you read on your other device.
Get the full title to continue listening from where you left off, or restart the preview.