Applied Math for Engineers

Similar documents
ECE257 Numerical Methods and Scientific Computing. Ordinary Differential Equations

Fourth Order RK-Method

Numerical Methods - Initial Value Problems for ODEs

Solving Ordinary Differential equations

Review Higher Order methods Multistep methods Summary HIGHER ORDER METHODS. P.V. Johnson. School of Mathematics. Semester

Initial value problems for ordinary differential equations

Numerical Methods for Differential Equations

MTH 452/552 Homework 3

COSC 3361 Numerical Analysis I Ordinary Differential Equations (II) - Multistep methods

Numerical solution of ODEs

AN OVERVIEW. Numerical Methods for ODE Initial Value Problems. 1. One-step methods (Taylor series, Runge-Kutta)

9.6 Predictor-Corrector Methods

multistep methods Last modified: November 28, 2017 Recall that we are interested in the numerical solution of the initial value problem (IVP):

Numerical Methods for Differential Equations

Initial-Value Problems for ODEs. Introduction to Linear Multistep Methods

CS 450 Numerical Analysis. Chapter 9: Initial Value Problems for Ordinary Differential Equations

Lecture Notes to Accompany. Scientific Computing An Introductory Survey. by Michael T. Heath. Chapter 9

The family of Runge Kutta methods with two intermediate evaluations is defined by

Ordinary differential equations - Initial value problems

Mathematics for chemical engineers. Numerical solution of ordinary differential equations

NUMERICAL SOLUTION OF ODE IVPs. Overview

Scientific Computing: An Introductory Survey

2 Numerical Methods for Initial Value Problems

Jim Lambers MAT 772 Fall Semester Lecture 21 Notes

Numerical Methods for the Solution of Differential Equations

Ordinary Differential Equations

8.1 Introduction. Consider the initial value problem (IVP):

Astrodynamics (AERO0024)

Mini project ODE, TANA22

5.6 Multistep Methods

Chapter 5 Exercises. (a) Determine the best possible Lipschitz constant for this function over 2 u <. u (t) = log(u(t)), u(0) = 2.

Euler s Method, cont d


Solving Ordinary Differential Equations

Numerical Differential Equations: IVP

You may not use your books, notes; calculators are highly recommended.

Linear Multistep Methods I: Adams and BDF Methods

CHAPTER 5: Linear Multistep Methods

Chapter 8. Numerical Solution of Ordinary Differential Equations. Module No. 1. Runge-Kutta Methods

Ordinary differential equation II

Math 128A Spring 2003 Week 12 Solutions

MB4018 Differential equations

Differential Equations

HIGHER ORDER METHODS. There are two principal means to derive higher order methods. b j f(x n j,y n j )

Integration of Ordinary Differential Equations

Math Numerical Analysis Homework #4 Due End of term. y = 2y t 3y2 t 3, 1 t 2, y(1) = 1. n=(b-a)/h+1; % the number of steps we need to take

Multistep Methods for IVPs. t 0 < t < T

4 Stability analysis of finite-difference methods for ODEs

Part IB Numerical Analysis

Scientific Computing with Case Studies SIAM Press, Lecture Notes for Unit V Solution of

Initial value problems for ordinary differential equations

Solving PDEs with PGI CUDA Fortran Part 4: Initial value problems for ordinary differential equations

Ordinary Differential Equations

1 Error Analysis for Solving IVP

Bindel, Fall 2011 Intro to Scientific Computing (CS 3220) Week 12: Monday, Apr 18. HW 7 is posted, and will be due in class on 4/25.

Math 128A Spring 2003 Week 11 Solutions Burden & Faires 5.6: 1b, 3b, 7, 9, 12 Burden & Faires 5.7: 1b, 3b, 5 Burden & Faires 5.

Module 4: Numerical Methods for ODE. Michael Bader. Winter 2007/2008

Consistency and Convergence

1 Ordinary Differential Equations

PowerPoints organized by Dr. Michael R. Gustafson II, Duke University

MATH 350: Introduction to Computational Mathematics

Chap. 20: Initial-Value Problems

NUMERICAL ANALYSIS 2 - FINAL EXAM Summer Term 2006 Matrikelnummer:

ODE Runge-Kutta methods

Lecture 10 Polynomial interpolation

Chapter 8. Numerical Solution of Ordinary Differential Equations. Module No. 2. Predictor-Corrector Methods

PowerPoints organized by Dr. Michael R. Gustafson II, Duke University

Lecture 4: Numerical solution of ordinary differential equations

Butcher tableau Can summarize an s + 1 stage Runge Kutta method using a triangular grid of coefficients

Explicit One-Step Methods

Multistage Methods I: Runge-Kutta Methods

Ordinary Differential Equations II

Solving scalar IVP s : Runge-Kutta Methods

CS520: numerical ODEs (Ch.2)

Ordinary Differential Equations II

2.29 Numerical Fluid Mechanics Fall 2011 Lecture 20

Lecture 8: Calculus and Differential Equations

Lecture 8: Calculus and Differential Equations

Generalized RK Integrators for Solving Ordinary Differential Equations: A Survey & Comparison Study

Numerical Methods for Differential Equations

Math 411 Preliminaries

Southern Methodist University.

4.4 Computing π, ln 2 and e

Introduction to the Numerical Solution of IVP for ODE

Lecture V: The game-engine loop & Time Integration

KINGS COLLEGE OF ENGINEERING DEPARTMENT OF MATHEMATICS ACADEMIC YEAR / EVEN SEMESTER QUESTION BANK

Lecture IV: Time Discretization

Numerical Methods for Partial Differential Equations

Selected HW Solutions

PowerPoints organized by Dr. Michael R. Gustafson II, Duke University

CHAPTER 10: Numerical Methods for DAEs

Efficient Simulation of Physical System Models Using Inlined Implicit Runge Kutta Algorithms

Math 660 Lecture 4: FDM for evolutionary equations: ODE solvers

EXAMPLE OF ONE-STEP METHOD

Embedded 5(4) Pair Trigonometrically-Fitted Two Derivative Runge- Kutta Method with FSAL Property for Numerical Solution of Oscillatory Problems

On interval predictor-corrector methods

Numerical Methods for Partial Differential Equations

Second Order ODEs. CSCC51H- Numerical Approx, Int and ODEs p.130/177

Solving Delay Differential Equations (DDEs) using Nakashima s 2 Stages 4 th Order Pseudo-Runge-Kutta Method

Lecture Note 3: Interpolation and Polynomial Approximation. Xiaoqun Zhang Shanghai Jiao Tong University

Transcription:

Applied Math for Engineers Ming Zhong Lecture 15 March 28, 2018 Ming Zhong (JHU) AMS Spring 2018 1 / 28

Recap Table of Contents 1 Recap 2 Numerical ODEs: Single Step Methods 3 Multistep Methods 4 Method for Systems and Higher Order ODEs 5 Homework Ming Zhong (JHU) AMS Spring 2018 2 / 28

Recap Summary We have discussed: How to solve non-homogeneous linear First Order ODE system. Method of Undetermined Coefficients. Method of Variation of Parameters, u = Y 1 g. How to solve linear ODEs with power series. y + p(x)y + q(x)y = 0, p, q analytic, then y = m=0 a mx m. y + b(x) x y + c(x) x y = 0, b, c analytic at x = 0, then 2 y = x r m=0 a mx m. Ming Zhong (JHU) AMS Spring 2018 3 / 28

Numerical ODEs: Single Step Methods Table of Contents 1 Recap 2 Numerical ODEs: Single Step Methods 3 Multistep Methods 4 Method for Systems and Higher Order ODEs 5 Homework Ming Zhong (JHU) AMS Spring 2018 4 / 28

Numerical ODEs: Single Step Methods Why Numerical Methods? Not all ODEs can be solved analytically. Consider, x i = N φ( xj x )(xj i x i ), for 1 i N. j=1 Here, φ(r) = 1 if 0 r < 1 2 and φ(r) = 0.1 if 1 2 r < 1 and φ(r) = 0 otherwise. Ming Zhong (JHU) AMS Spring 2018 5 / 28

Numerical ODEs: Single Step Methods Notations Consider the Initial Value Problem (IVP) for (f is nice for the IVP to have unique solution), y = f (x, y), y(x 0 ) = y 0. We will try to obtain a numerical solution for x (a, b) containing x 0 : We consider some equidistant mesh points: x 1 = x 0 + h, x 2 = x 0 + 2h,, x n = x 0 + nh. h is called the step size, it is fixed. It is derived from Taylor Series expansion (step by step): y(x + h) = y(x) + hy (x) + h2 2 y + The Euler method (or Euler-Cauchy method) uses only the first derivative: y(x + h) y(x) + hy (x), y n+1 = y n + hf (x n, y n ). Ming Zhong (JHU) AMS Spring 2018 6 / 28

Numerical ODEs: Single Step Methods Error of the Euler Method Local Truncation Error By Taylor series expansion, y(x n ) y n O(h 2 ), assuming y n 1 = y(x n 1 ), called the Local Truncation Error. The number of steps is proportional to 1 h. The total error (global error) is proportional to h 2 1 h = h. The Euler method is a first-order method. There are roundoff errors in computer arithmetics, might accumulate in y 1, y 2,. Ming Zhong (JHU) AMS Spring 2018 7 / 28

Numerical ODEs: Single Step Methods Varaible Step Size Selection Since y = f, y = f = f x + f y y = f x + f y f. We can use it to esimate y, given a certain Tolerance τ, we want 1 y (ɛ n ) 2τ τ, thus h n y (ɛ n ). h 2 n y must not be zero on the interval. We can take the maximum of y over the whole solution interval. Ming Zhong (JHU) AMS Spring 2018 8 / 28

Numerical ODEs: Single Step Methods Improved Euler Method (Heun s Method) Predictor and Corrector Consider the Taylor series expansion of y, y(x + h) = y(x) + hy + h2 2 y + h3 3! y +. Repalcing y with f and y(x + h) = y(x) + hf + h2 2 f + h3 3! f +. However, f requires partial derivatives. The general strategy is to avoid computing f s directly, which brings us the improved Euler method. First, we preditc yn+1 = y n + hf (x n, y n ); use yn+1 to correct, y n+1 = y n + h 2 (f (x n, y n ) + f (x n+1, y n+1)). Ming Zhong (JHU) AMS Spring 2018 9 / 28

Numerical ODEs: Single Step Methods Local Truncation Error of Heun s Method Theorem The local truncation error of the Heun s method is of order h 3. Can we go beyound second order? The Runge-Kutta Methods (RK4 methods) k 1 = hf (x n, y n ). k 2 = f (x n + h 2, y n + k 1 2 ). k 3 = f (x n + h 2, y n + k 2 2 ). k 4 = f (x n + h, y n + k 3 ). y n+1 = y n + 1 6 (k 1 + 2k 2 + 2k 3 + k 4 ). It is a 4 th order method. Ming Zhong (JHU) AMS Spring 2018 10 / 28

Numerical ODEs: Single Step Methods General Explicit RK Methods Consider a s stage RK method of the following, y n+1 = y n + h s i=1 b ik i, where k 1 = f (x n, y n ) k 2 = f (x n + c 2 h, y n + h(a 21 k 1 )) k 3 = f (x n + c 3 h, y n + h(a 31 k 1 + a 32 k 2 )). k s = f (x n + c s h, y n + h(a s1 k 1 + a s2 k 2 + a s,s 1 k s 1 )) To specify a particular method, one needs to provide the integer s (the number of stages), the coefficients a ij (for 1 j < i s), b i (for i = 1,, s), and c i (for i = 2, s). The matrix A = [a ij ] is called the Runge-Kutta matrix, b i s are known as the weights and the nodes. Ming Zhong (JHU) AMS Spring 2018 11 / 28

Numerical ODEs: Single Step Methods Butcher Tableau 0 c 2 a 21 c 3 a 31 a 32..... c s a s1 a s2 a s,s 1 b 1 b 2 b s 1 b s The Runge-Kutta method is consistent if i 1 j=1 a ij = c i for i = 2,, s. order 1 2 3 4 5 6 7 8 min s 1 2 3 4 6 7 9 11 Ming Zhong (JHU) AMS Spring 2018 12 / 28

Numerical ODEs: Single Step Methods Step Size Control Runge-Kutta-Fehlberg Consider any first order method, and two approximations, y h and y 2h, at the same x n obtained from two different step sizes, h and 2h. Then, y yh e h and y y2h e 2h. Moreover, e 2h 2 1 e h, then, y h y 2h e 2h e h e h. So we can estimate e h by taking the difference of y h and y 2h. And, If y h y 2h τ (a given tolerance), we keep y 2h and move on. Otherwise, we calculate y h 2 and compare. Ming Zhong (JHU) AMS Spring 2018 13 / 28

Numerical ODEs: Single Step Methods The RKF Method However the RKF method can do it better. RKF is an embedded RK method. The Fehlberg s fifth-order method, y n+1 = y n + γ 1 k 1 + + γ 5 k 5 + γ 6 k 6. [γ 1,, γ 6 ] = [ 16 134 The Fehlberg s fourth-order method,, 0, 6656 12825, 28561 56430, 9 50, 2 55 ] yn+1 = y n + γ1 k 1 + γ5 k 5. [γ1,, γ 5 ] = [ 25 1408 216, 0, 2565, 2197 4104, 1 5 ] They use the same k i s. Ming Zhong (JHU) AMS Spring 2018 14 / 28

Numerical ODEs: Single Step Methods RKF Method, Cont. It also comes with an error estimator, These are explicit methods, where the evaluation of f s depend on previous information, (x n, y n ). Some problems are very sensitive to step size (explicitly). We need to consider implicit method for stiff problems, the implicit Euler method (Backward Euler), y n+1 = y n + hf (x n+1, y n+1 ). Each step, we need to solve an equation, y n+1 hf (x n+1, y n+1 ) = y n. Use Newton s method (if it is non-linear). It gives only first order accuracy however allows for bigger step size. Ming Zhong (JHU) AMS Spring 2018 15 / 28

Numerical ODEs: Single Step Methods Example Example Numerically solve the IVP, y = 20y + 20x 2 + 2x, y(0) = 1. The true solution is: y = e 20x + x 2, e 20x decays too fast. Using Backward Euler: y n+1 = y n + hf (x n+1, y n1 ). Solve for y n+1, we obtain y n+1 = yn+h(20(xn+h)2 +2(x n+h)) 1+20h. Solution is stable for Backward Euler with h = 0.2. Solution is stable for Euler with h = 0.05 but unstable for h = 0.1. Solution is stable for RK4 with h = 0.1 but unstable for h = 0.2. Ming Zhong (JHU) AMS Spring 2018 16 / 28

Multistep Methods Table of Contents 1 Recap 2 Numerical ODEs: Single Step Methods 3 Multistep Methods 4 Method for Systems and Higher Order ODEs 5 Homework Ming Zhong (JHU) AMS Spring 2018 17 / 28

Multistep Methods Multistep Methods We have discussed, A family of Single Step Methods: they are all part of the Runge-Kutta family (of different orders). They are self-starting, they only need y n 1 to get to y n. The Adams-Bashforth Methods, Consider the IVP, y = f (x, y) with y(x 0 ) = y 0. By integration, y(x n+1 ) y(x n ) = x n+1 x=x n y (x)(= f (x, y)) dx. We only know f at (x 0, y 0 ), (we do have expression for f, but y is unknown). We approximate f using polynomial at unknown (x, y). So y(x n+1 ) y(x n ) x n+1 x=x n p(x) dx. Ming Zhong (JHU) AMS Spring 2018 18 / 28

Multistep Methods Adams-Bashforth Methods, Cont. Continue on, Suppose we are given (x n 3, y n 3 ), (x n 2, y n 2 ), (x n 1, y n 1 ), and (x n, y n ). Let f n = f (x n, y n ) and f n s = f (x n s, y n s ), for s = 1, 2, 3. Suppose also x n 3, x n 2, x n 1, x n, are equidistant points. We will approximate f at the 4 pairs of points using cubic polynomial, p 3 (x), defined as p 3 (x) = f n + r f n + r(r + 1) 2 f n + 2 r(r + 1)(r + 2) 3 f n. 6 Here, r = x xn h, and is the Newton s backward difference formula, f n = f n f n 1, and k f n = k 1 f n k 1 f n 1. y n+1 = y n + h 24 (55f n 59f n 1 + 37f n 2 9f n 3 ), Adams-Bahsforth method of fourth order. Ming Zhong (JHU) AMS Spring 2018 19 / 28

Multistep Methods Adams-Moulton Formula The Adams-Bashforth method uses all past information, we can also include the information at (x n+1, y n+1 ). p 3 (x) = f n+1 + r f n+1 + r(r+1) 2 2 f n+1 + r(r+1)(r+2) 6 3 f n+1. y n+1 = y n + h 24 (9f n+1 + 19f n 5f n 1 + f n 2 ). It is an implicit method. In order to avoid solving an equation, we can use the predictor-corrector of AB4. Predict: y n+1 = y n + h 24 (55f n 59f n 1 + 37f n 2 9f n 3 ). Correct: y n+1 = y n + h 24 (9f (x n+1, y n+1 ) + 19f n 5f n 1 + f n 2 ). This is called the Adams-Moulton s method of fourth order. How do we get started? We are only given (x 0, y 0 ), but we will need (x 1, y 1 ), (x 2, y 2 ), and (x 3, y 2 ) to start the method. Just use any method to come up with those points. Ming Zhong (JHU) AMS Spring 2018 20 / 28

Method for Systems and Higher Order ODEs Table of Contents 1 Recap 2 Numerical ODEs: Single Step Methods 3 Multistep Methods 4 Method for Systems and Higher Order ODEs 5 Homework Ming Zhong (JHU) AMS Spring 2018 21 / 28

Method for Systems and Higher Order ODEs First Order System of ODEs Let us consider the IVP: y = f ( x, y) with y( x 0 ) = y 0. x is of size n 1 and y is of size m 1. f and y have to have the same size, but not necessarily x and y. f is nice so that the IVP has a unique solution on some open interval containing x 0. We only focus on first order ODEs or system of ODEs. For any higher order, y (m) = f (x, y, y, y,, y (m 1) ). Conversion: y 1 = y, y 2 = y y m = y (m 1). Then, y 1 = y 2,, y m 1 = y m, and y m = f (x, y 1, y 2,, y m 1 ). Ming Zhong (JHU) AMS Spring 2018 22 / 28

Method for Systems and Higher Order ODEs Single Step Methods The Euler method, The RK4 method, y n+1 = y n + h f ( x n, y n ). k1 = h f ( x n, y n ) k2 = h f ( x n + h 2, y n + k 1 2 ) k3 = h f ( x n + h 2, y n + k 2 2 ) k4 = h f ( x n + h, y n + k 3 ) y n+1 = y n + 1 6 ( k 1 + 2 k 2 + 2 k 3 + k 4 ) Ming Zhong (JHU) AMS Spring 2018 23 / 28

Method for Systems and Higher Order ODEs Runge-Kutta-Nystrom Methods The RKN methods, RKN methods are direct extensions of RK methods for second order ODEs: y = f (x, y, y ). k 1 = h 2 f (x n, y n, y n) k 2 = h 2 f (x n + h 2, y n + K, y n + k 1 ), K = h 2 (y n + 1 2 k 1) k 3 = h 2 f (x n + h 2, y n + K, y n + k 2 ) k 4 = h 2 f (x n + h, y n + K, y n + L), L = h(y n + k 3 ) y n+1 = y n + h(y n + 1 3 (k 1 + k 2 + k 3 )) y n+1 = y n + 1 3 (k 1 + 2k 2 + 2k 3 + k 4 ) Ming Zhong (JHU) AMS Spring 2018 24 / 28

Method for Systems and Higher Order ODEs RKN without y If y = f (x, y) not containing y, we have a simplify RNK formulas k 1 = h 2 f (x n, y n ) k 2 = h 2 f (x n + h 2, y n + h 2 (y n + k 1 2 ) = k 3 k 4 = h 2 f (x n + h, y n + h(y n + k 2 ) y n+1 = y n + h(y n + 1 3 (k 1 + 2k 2 )) y n+1 = y n + 1 3 (k 1 + 4k 2 + k 4 ) Backward Euler method for stiff systems, y n+1 = y n + h f ( x n+1, y n+1 ). At each step, we will have to solve a (possible) non-linear system for y n+1. Ming Zhong (JHU) AMS Spring 2018 25 / 28

Homework Table of Contents 1 Recap 2 Numerical ODEs: Single Step Methods 3 Multistep Methods 4 Method for Systems and Higher Order ODEs 5 Homework Ming Zhong (JHU) AMS Spring 2018 26 / 28

Homework Homework 7 Please do the following: Chapter 4 Review Questions: 20, 21, 22, 23. Chapter 5 Review Questions: 12, 14, 16, 18. Ming Zhong (JHU) AMS Spring 2018 27 / 28