Appendix B. The Finite Difference Scheme

Similar documents
Lecture 21: Numerical methods for pricing American type derivatives

NON-CENTRAL 7-POINT FORMULA IN THE METHOD OF LINES FOR PARABOLIC AND BURGERS' EQUATIONS

Additional Codes using Finite Difference Method. 1 HJB Equation for Consumption-Saving Problem Without Uncertainty

Module 3 LOSSY IMAGE COMPRESSION SYSTEMS. Version 2 ECE IIT, Kharagpur

CHAPTER 5 NUMERICAL EVALUATION OF DYNAMIC RESPONSE

Convexity preserving interpolation by splines of arbitrary degree

Numerical Heat and Mass Transfer

Lecture 5.8 Flux Vector Splitting

High resolution entropy stable scheme for shallow water equations

Chapter 5. Solution of System of Linear Equations. Module No. 6. Solution of Inconsistent and Ill Conditioned Systems

Transfer Functions. Convenient representation of a linear, dynamic model. A transfer function (TF) relates one input and one output: ( ) system

Formal solvers of the RT equation

2 Finite difference basics

Lecture 12: Discrete Laplacian

4DVAR, according to the name, is a four-dimensional variational method.

CME 302: NUMERICAL LINEAR ALGEBRA FALL 2005/06 LECTURE 13

Handout # 6 (MEEN 617) Numerical Integration to Find Time Response of SDOF mechanical system. and write EOM (1) as two first-order Eqs.

6.3.4 Modified Euler s method of integration

2.29 Numerical Fluid Mechanics Fall 2011 Lecture 12

NUMERICAL DIFFERENTIATION

A Hybrid Variational Iteration Method for Blasius Equation

Modified Mass Matrices and Positivity Preservation for Hyperbolic and Parabolic PDEs

Errors for Linear Systems

Lectures - Week 4 Matrix norms, Conditioning, Vector Spaces, Linear Independence, Spanning sets and Basis, Null space and Range of a Matrix

Numerical Transient Heat Conduction Experiment

Lab session: numerical simulations of sponateous polarization

Report on Image warping

A new Approach for Solving Linear Ordinary Differential Equations

Parametric fractional imputation for missing data analysis. Jae Kwang Kim Survey Working Group Seminar March 29, 2010

page 2 2 dscretzaton mantans ths stablty under a sutable restrcton on the tme step. SSP tme dscretzaton methods were frst developed by Shu n [20] and

Inductance Calculation for Conductors of Arbitrary Shape

New Method for Solving Poisson Equation. on Irregular Domains

The Finite Element Method: A Short Introduction

Positivity-preserving time discretizations for production-destruction equations. with applications to non-equilibrium flows.

EPR Paradox and the Physical Meaning of an Experiment in Quantum Mechanics. Vesselin C. Noninski

Canonical transformations

MATH 5630: Discrete Time-Space Model Hung Phan, UMass Lowell March 1, 2018

Key Words: Hamiltonian systems, canonical integrators, symplectic integrators, Runge-Kutta-Nyström methods.

Chapter 12. Ordinary Differential Equation Boundary Value (BV) Problems

Application of B-Spline to Numerical Solution of a System of Singularly Perturbed Problems

MMA and GCMMA two methods for nonlinear optimization

Consistency & Convergence

DUE: WEDS FEB 21ST 2018

2.29 Numerical Fluid Mechanics

Numerical Solution of Ordinary Differential Equations

AA214B: NUMERICAL METHODS FOR COMPRESSIBLE FLOWS

E91: Dynamics. E91: Dynamics. Numerical Integration & State Space Representation

NUMERICAL METHODS FOR FIRST ORDER ODEs

CHAPTER 14 GENERAL PERTURBATION THEORY

Salmon: Lectures on partial differential equations. Consider the general linear, second-order PDE in the form. ,x 2

PART 8. Partial Differential Equations PDEs

Modelli Clamfim Equazioni differenziali 22 settembre 2016

Inexact Newton Methods for Inverse Eigenvalue Problems

Module 1 : The equation of continuity. Lecture 1: Equation of Continuity

Lecture 2: Numerical Methods for Differentiations and Integrations

Chapter 4 The Wave Equation

Generalized Linear Methods

ENTROPY CONSERVATIVE SCHEMES AND ADAPTIVE MESH SELECTION FOR HYPERBOLIC CONSERVATION LAWS

Numerical Solution of two dimensional coupled viscous Burgers Equation using the Modified Cubic B-Spline Differential Quadrature Method

EEE 241: Linear Systems

Hidden Markov Models & The Multivariate Gaussian (10/26/04)

Singular Value Decomposition: Theory and Applications

CSci 6974 and ECSE 6966 Math. Tech. for Vision, Graphics and Robotics Lecture 21, April 17, 2006 Estimating A Plane Homography

LOW BIAS INTEGRATED PATH ESTIMATORS. James M. Calvin

Note 10. Modeling and Simulation of Dynamic Systems

Perron Vectors of an Irreducible Nonnegative Interval Matrix

Markov Chain Monte Carlo (MCMC), Gibbs Sampling, Metropolis Algorithms, and Simulated Annealing Bioinformatics Course Supplement

SEMI-LAGRANGIAN SCHEMES FOR LINEAR AND FULLY NON-LINEAR DIFFUSION EQUATIONS

FTCS Solution to the Heat Equation

The equation of motion of a dynamical system is given by a set of differential equations. That is (1)

[7] R.S. Varga, Matrix Iterative Analysis, Prentice-Hall, Englewood Clis, New Jersey, (1962).

Aerodynamics. Finite Wings Lifting line theory Glauert s method

Vector Norms. Chapter 7 Iterative Techniques in Matrix Algebra. Cauchy-Bunyakovsky-Schwarz Inequality for Sums. Distances. Convergence.

Global Optimization of Truss. Structure Design INFORMS J. N. Hooker. Tallys Yunes. Slide 1

Global Sensitivity. Tuesday 20 th February, 2018

A quote of the week (or camel of the week): There is no expedience to which a man will not go to avoid the labor of thinking. Thomas A.

Ballot Paths Avoiding Depth Zero Patterns

Week3, Chapter 4. Position and Displacement. Motion in Two Dimensions. Instantaneous Velocity. Average Velocity

Modelli Clamfim Equazione del Calore Lezione ottobre 2014

Lecture 3: Shannon s Theorem

Hongyi Miao, College of Science, Nanjing Forestry University, Nanjing ,China. (Received 20 June 2013, accepted 11 March 2014) I)ϕ (k)

Appendix for Causal Interaction in Factorial Experiments: Application to Conjoint Analysis

arxiv: v1 [math.ho] 18 May 2008

Problem adapted reduced models based on Reaction-Diffusion Manifolds (REDIMs)

Lecture 10 Support Vector Machines II

ANSWERS. Problem 1. and the moment generating function (mgf) by. defined for any real t. Use this to show that E( U) var( U)

Some modelling aspects for the Matlab implementation of MMA

Difference Equations

ELASTIC WAVE PROPAGATION IN A CONTINUOUS MEDIUM

Digital Signal Processing

6) Derivatives, gradients and Hessian matrices

Chapter Newton s Method

8 Derivation of Network Rate Equations from Single- Cell Conductance Equations

1 GSW Iterative Techniques for y = Ax

Suppose that there s a measured wndow of data fff k () ; :::; ff k g of a sze w, measured dscretely wth varable dscretzaton step. It s convenent to pl

Discretization. Consistency. Exact Solution Convergence x, t --> 0. Figure 5.1: Relation between consistency, stability, and convergence.

Chp 3: Scalar Advection and Linear Hyperbolic Systems. By Prof. Dinshaw S. Balsara

Using T.O.M to Estimate Parameter of distributions that have not Single Exponential Family

Physics 181. Particle Systems

The Feynman path integral

Transcription:

140 APPENDIXES Appendx B. The Fnte Dfference Scheme In ths appendx we present numercal technques whch are used to approxmate solutons of system 3.1 3.3. A comprehensve treatment of theoretcal and mplementaton ssues of dscretzaton methods for advecton-dffusonreacton problems are gven n the monograph by Hundsdorfer and Verwer Hundsdorfer, Verwer 003. Very nterestng applcatons of these results for bomedcal problems are descrbed by Gersch and Chaplan Gersch, Chaplan 006, Gersch and Verwer Gersch, Verwer 00 see also references gven n these publcatons. Here we propose to use a lnearzed mplct backward Euler method for the approxmaton of the dffuson-reacton subproblems and the explct forward Euler method for soluton of the advecton subproblem. We have restrcted to the frst order methods due to ther robust stablty. Note that our man goal s to nvestgate the nfluence of a possble ll-posedness of the PDE system to the asymptotcal behavour of the soluton. Appendx B.1. The method of lnes. Dscretzaton n space We use the method of lnes MOL approach see, Gersch, Chaplan 006; Hundsdorfer, Verwer 003. At the frst step we approxmate the spatal dervatves n the PDE by applyng robust and accurate approxmatons targeted for specal physcal processes descrbed by dfferental equatons. Doman [0, 1] s covered by a dscrete unform grd ω h = {x : x = h, = 0,..., N 1}, x N = 1 wth the grd ponts x. On the semdscrete doman ω h k [0, T ] we defne functons U t = Ux, t, V t = V x, t, = 0,..., N 1, here U, V approxmate solutons ux, t, vx, t on the dscrete grd ω h at tme moment t. We also defne the forward and backward space fnte dfferences wth respect to x: x U = U +1 U h, x U = U U 1. h Usng the fnte volume approach, we approxmate the dffuson and reac-

APPENDIXES 141 ton terms n the PDE system 3.1 by the followng fnte dfference equatons: A DR1 U, = D x x U + γru 1 U, U p A DR V, U, = x x V + γ 1 + βu p V. B.57 The stencl of the dscrete scheme requres to use functons defned outsde of the grd ω h. We apply perodcty boundary condtons 3.3 to defne dscrete functons for < 0 or N: U = U N, U N 1+ = U 1 for > 0. B.58 The advecton term n equaton 3.1 depends on the varable velocty ax, t := χ 1 + αv v x, therefore the maxmum prncple s not vald for the respectve transport equaton. But problem 3.1 stll has non-negatve solutons, and ths property can be preserved on the dscrete level by applyng proper upwndng approxmatons. The dscrete spatal approxmaton of the velocty s computed by a + 1 t = χ 1 + αv + V +1 / xv. In the followng we consder the upwnd-based dscrete fluxes Gersch, Chaplan 006; Hundsdorfer, Verwer 003: [ F T U, a, + 1/ = a + 1 U + ψθ ] U +1 U, a+ 1 0, B.59 F T U, a, + 1/ = a + 1 wth the Koren lmter functon ψθ = max [ U+1 + ψ1/θ +1 ] U U +1, a+ 1 0, mn 1, 1 3 + 1 6 θ, θ. The lmter ψ depends on the smoothness montor functon θ = U U 1 U +1 U. < 0,

14 APPENDIXES For ψ = 0 we get the standard frst-order upwnd flux F T UW U, a, + 1/ = max a + 1, 0 U + mn a + 1, 0 U +1. Let us denote the dscrete advecton operator as A T U, V, = 1 h FT U, a, + 1/ F T U, a, 1/. Then we obtan a nonlnear ODE system for the evaluaton of the approxmate sem-dscrete solutons du = A T U, V, + A DR1 U,, x ω h, B.60 dv = A DRV, U,. Appendx B.. Operator splttng methods In order to develop effcent solvers n tme for the obtaned large ODE systems we apply the splttng technques. They take nto account the dfferent nature of the dscrete operators defnng the advecton A T U, V, and the dffuson-reacton A DR1 U,, A DR V, U, terms. The system resolvng the sem-dscrete advecton process can be solved very effcently by usng explct solvers, whle the dffuson-reacton sem-dscrete system s stff and t requres an mplct treatment. Also we are nterested n preservng at the dscrete level the postvty and/or boundedness of the soluton, f such propertes hold for the dfferental ODE system. Frst we consder the symmetrcal splttng method also known as the Strang splttng Strang 1968. Gven approxmatons U n, V n at tme t n, we

APPENDIXES 143 compute solutons at t n+1 = t n + by the followng scheme: du = A T U, V n,, U t n = U n, t n t t n+ 1 = t n + /, B.61 du du dv = A DRV, U,, = A DR1 U,, U t n = U n+ 1, t n t t n+1, B.6 = A T U, V n,, U t n = U n+1, t n+ 1 t t n+1 V t n = V n, t n t t n+1. B.63 Here we also have spltted the gven ODE system nto two blocks wth respect to U and V functons. Lemma 1. Solutons of the splttng ODE problem B.61 B.63 are nonnegatve f U n 0 and V n 0 for all x ω h. Proof. The proof for the advecton subsystems follows from the constructon of the dscrete fluxes by usng the upwndng technque. The proof for the dffuson-reacton subsystems follows from the lemma n Gersch, Verwer 00 that the soluton of an ntal value problem for systems of ODEs dy = F t, Y t, t 0, Y 0 = Y 0 s non-negatve f and only f for all t and any vector V R m and all 1 m v = 0, v 0 for all = f t, V 0. It s easy to see that for the dffuson-reacton subsystems the dffuson parts of the matrces are dagonally domnant and all off-dagonal entres are nonpostve. For the reacton parts the requred estmates are also trvally satsfed. Lemma. If 0 U n C for all x ω h, then a soluton of the splttng ODE problem B.6 s also bounded U maxc, 1. Proof. Here we use the fact that U = 1 s a stable attractor of the reacton functon. Let U n = max U n. Then t follows from the defnton of A DR1 U, that n the worst case du > 0, f C < 1, du = 0, f C = 1,

144 APPENDIXES and The lemma s proved. du < 0, f C > 1. Appendx B.3. Numercal ntegraton of ODEs There are many numercal ntegraton methods for soluton of non-stff and stff ODEs. For detaled dscussons of these schemes we refer the reader to Gersch, Chaplan 006; Harer, Norset, Wanner 1993; Harer, Wanner 1996; Hundsdorfer, Verwer 003. Let ω be a unform tme grd ω = {t n : t n = n, n = 0,..., M, M = T f }, here s the tme step. For smplcty ths step sze s taken constant. In the followng, we consder numercal approxmatons U n, V n to the exact soluton values ux, t n, vx, t n at the grd ponts x, t n ω h ω. Remark 1. In Baronas, Šmkus 011, the explct forward Euler scheme s used to solve problem 3.1. Snce no detals are gven n Baronas, Šmkus 011 on approxmatons of spatal dervatves, we use dscrete operators ntroduced n prevous sectons and wrte the explct forward Euler scheme as: U n+1 V n+1 U n V n = A T U n, V n, + A DR1 U n,, = A DR V n, U n,. We note that ths scheme can be wrtten as a splttng algorthm: U n+ 1 U n U n+1 U n+ 1 V n+1 V n = A T U n, V n,, = A DR1 U n,, = A DR V n, U n,.

APPENDIXES 145 Thus the explct Euler scheme can be consdered as a specal case of splttng algorthms. Despte easy mplementaton and good parallel propertes of explct algorthms, the man drawback of the explct Euler method s that due to the condtonal stablty we must restrct the ntegraton step to Ch for stff dscrete dffuson-reacton subproblems. The Rosenbrock and mplct Runge-Kutta methods are successfully appled for ntegraton of a stff part of the splttng semdscrete-scheme,.e. dffuson reacton equatons B.6, B.63, see Gersch, Chaplan 006; Gersch, Verwer 00; Hundsdorfer, Verwer 003. Here we propose to use a lnearzed mplct backward Euler method for the approxmaton of the dffuson-reacton subproblems and the explct forward Euler method for soluton of the advecton subproblem. We have restrcted to the frst order methods due to ther robust stablty. Note that our man goal s to nvestgate the nfluence of a possble ll-posedness of the PDE system to the asymptotcal behavour of the soluton. We dscretze the semdscrete problem B.61 B.63 wth the fully dscrete scheme U n+ 1 3 U n 0.5 U n+ 3 U n+ 1 3 U n+1 U n+ 3 0.5 V n+1 V n = A T U n, V n,, B.64 = D x x U n+ 3 + γru n+ 1 3 1 U n+ 3, B.65 = A T U n+ 3, V n,, B.66 = A DR V n+1, U n+1,. B.67 We apply two splttngs of the advecton term, because then we use only half of the splttng step sze for the explct method. Ths doubles the stablty and postvty domans of the explct method see Gersch, Verwer 00. Next we prove that statements of Lemma 1 and hold also for solutons of the fully dscrete fnte dfference scheme B.64 B.67 Lemma 3. For a suffcently small tme step 0 solutons of the fnte dfference scheme B.64 B.67 are non-negatve f U n 0 and V n 0 for all x ω h. Proof. The proof for the advecton problems B.64 and B.66 follows from

146 APPENDIXES the constructon of the dscrete fluxes by usng the upwndng technque and selecton of a suffcently small tme step 0. The proof for the dffuson-reacton problems B.65 and B.67 follows from the maxmum prncple Samarsk 001. For example, consder problem B.65. We assume, that U n+ 3 = mn U n+ 3. 0 <N We wrte the dscrete equaton B.65 for U n+ 3 1 + γru n+ 1 3 n an explct form U n+ 3 = 1 + γr U n+ 1 3 + D h U n+ 3 +1 + U n+ 3 1 U n+ 3 Snce U n+ 1 3 0 and U n+ 3 ±1 U n+ 3 we get that U n+ 3 0. Lemma 4. If 0 U n+ 1 3 C for all x ω h, then a soluton of the fnte dfference scheme B.65 s also bounded Proof. U n+ 3 maxc, 1, x ω h. The proof s based on the maxmum prncple and a specal form of the dscrete reacton term. Frst, we consder the case C 1. Let U n+ 3 = max U n+ 3. Then t follows from B.65 that 1+γrU n+ 1 3 U n+ 3 1+γr U n+ 1 3 = U n+ 3 Next we consder the case C > 1. Then we get that 1 n+ 3 1 + γru 1 + γru n+ 1 3 1. U n+ 3 The lemma s proved. 1 n+ 3 1 + γru 1 + γru n+ 1 3 = 1 + U n+ 1 3 1 1 + γru n+ 1 3 < C.