Energy-Preserving Runge-Kutta methods

Similar documents
High-order symmetric schemes for the energy conservation of polynomial Hamiltonian problems.

High-order Symmetric Schemes for the Energy Conservation of Polynomial Hamiltonian Problems 1 2

ENERGY PRESERVING METHODS FOR VOLTERRA LATTICE EQUATION

SYMMETRIC PROJECTION METHODS FOR DIFFERENTIAL EQUATIONS ON MANIFOLDS

Multiple invariants conserving Runge-Kutta type methods for Hamiltonian problems

Multistage Methods I: Runge-Kutta Methods

1. Introduction. We consider canonical Hamiltonian systems in the form 0 I y(t 0 ) = y 0 R 2m J = I 0

Runge-Kutta Theory and Constraint Programming Julien Alexandre dit Sandretto Alexandre Chapoutot. Department U2IS ENSTA ParisTech SCAN Uppsala

ENERGY PRESERVATION (DISSIPATION) IN NUMERICAL ODEs and PDEs

Finite Differences for Differential Equations 28 PART II. Finite Difference Methods for Differential Equations

Validated Explicit and Implicit Runge-Kutta Methods

Scientific Computing: An Introductory Survey

Improved Starting Methods for Two-Step Runge Kutta Methods of Stage-Order p 3

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

Symplectic integration with Runge-Kutta methods, AARMS summer school 2015

Efficiency of Runge-Kutta Methods in Solving Simple Harmonic Oscillators

Geometric Numerical Integration

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

Numerical solution of ODEs

Review. Numerical Methods Lecture 22. Prof. Jinbo Bi CSE, UConn

Implicit-explicit exponential integrators

Inverse Eigenvalue Problems: Theory and Applications

CHAPTER 10: Numerical Methods for DAEs

Numerical Methods for ODEs. Lectures for PSU Summer Programs Xiantao Li

MATHEMATICAL METHODS INTERPOLATION

About the numerical conservation of first integral by a class of symmetric methods

Ordinary Differential Equations

Numerical Methods for Differential Equations

Reducing round-off errors in symmetric multistep methods

Backward error analysis

Applied Numerical Analysis

A family of A-stable Runge Kutta collocation methods of higher order for initial-value problems

Two Optimized Runge-Kutta Methods for the Solution of the Schrödinger Equation

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

NUMERICAL ANALYSIS 2 - FINAL EXAM Summer Term 2006 Matrikelnummer:

Application of Geometric Integration to some Mechanical Problems

Starting Methods for Two-Step Runge Kutta Methods of Stage-Order 3 and Order 6

The collocation method for ODEs: an introduction

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

8.3 Numerical Quadrature, Continued

ENERGY PRESERVING INTEGRATION OF BI-HAMILTONIAN PARTIAL DIFFERENTIAL EQUATIONS

Copyright is owned by the Author of the thesis. Permission is given for a copy to be downloaded by an individual for the purpose of research and

Semi-implicit Krylov Deferred Correction Methods for Ordinary Differential Equations

PARTIAL DIFFERENTIAL EQUATIONS

Discontinuous Collocation Methods for DAEs in Mechanics

Applied Math for Engineers

Numerical Methods for Engineers. and Scientists. Applications using MATLAB. An Introduction with. Vish- Subramaniam. Third Edition. Amos Gilat.

Initial value problems for ordinary differential equations

Symplectic Runge-Kutta methods satisfying effective order conditions

Module 6: Implicit Runge-Kutta Methods Lecture 17: Derivation of Implicit Runge-Kutta Methods(Contd.) The Lecture Contains:

Solving Ordinary Differential equations

A SIXTH ORDER AVERAGED VECTOR FIELD METHOD *

Four Point Gauss Quadrature Runge Kuta Method Of Order 8 For Ordinary Differential Equations

Chapter 11 ORDINARY DIFFERENTIAL EQUATIONS

Parallel Methods for ODEs

SOLVING ODE s NUMERICALLY WHILE PRESERVING ALL FIRST INTEGRALS

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

EXTENSIONS OF THE HHT-α METHOD TO DIFFERENTIAL-ALGEBRAIC EQUATIONS IN MECHANICS

Runge-Kutta and Collocation Methods Florian Landis

Fourth Order RK-Method

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

AM205: Assignment 3 (due 5 PM, October 20)

SANGRADO PAGINA 17CMX24CM. PhD Thesis. Splitting methods for autonomous and non-autonomous perturbed equations LOMO A AJUSTAR (AHORA 4CM)

Quadratic SDIRK pair for treating chemical reaction problems.

Exam in TMA4215 December 7th 2012

(f(x) P 3 (x)) dx. (a) The Lagrange formula for the error is given by

Solving Orthogonal Matrix Differential Systems in Mathematica

Integration. Topic: Trapezoidal Rule. Major: General Engineering. Author: Autar Kaw, Charlie Barker.

Unique expansions with digits in ternary alphabets

MTH 452/552 Homework 3

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

Arbitrary order trigonometric Fourier collocation methods for multi-frequency oscillatory systems

Numerical Methods for Differential Equations

EAD 115. Numerical Solution of Engineering and Scientific Problems. David M. Rocke Department of Applied Science

Numerical Methods for Engineers

On the construction of discrete gradients

Sixth-Order and Fourth-Order Hybrid Boundary Value Methods for Systems of Boundary Value Problems

Partitioned Runge-Kutta Methods for Semi-explicit Differential-Algebraic Systems of Index 2

Advanced methods for ODEs and DAEs

Suboptimal feedback control of PDEs by solving Hamilton-Jacobi Bellman equations on sparse grids

Energy-preserving variant of collocation methods 1

A SYMBOLIC-NUMERIC APPROACH TO THE SOLUTION OF THE BUTCHER EQUATIONS

Ordinary differential equations - Initial value problems

Numerical Mathematics

An Overly Simplified and Brief Review of Differential Equation Solution Methods. 1. Some Common Exact Solution Methods for Differential Equations

Fourth-order symplectic exponentially-fitted modified Runge-Kutta methods of the Gauss type: a review

ENO and WENO schemes. Further topics and time Integration

Consistency and Convergence

Efficient implementation of Radau collocation methods

Variable Step Runge-Kutta-Nyström Methods for the Numerical Solution of Reversible Systems

Modeling & Simulation 2018 Lecture 12. Simulations

Review for Exam 2 Ben Wang and Mark Styczynski


CHAPTER 5: Linear Multistep Methods

Accuracy Enhancement Using Spectral Postprocessing for Differential Equations and Integral Equations

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

Numerical Methods - Initial Value Problems for ODEs

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

Do not turn over until you are told to do so by the Invigilator.

The Milne error estimator for stiff problems

Transcription:

Energy-Preserving Runge-Kutta methods Fasma Diele, Brigida Pace Istituto per le Applicazioni del Calcolo M. Picone, CNR, Via Amendola 122, 70126 Bari, Italy f.diele@ba.iac.cnr.it b.pace@ba.iac.cnr.it SDS2010, Monopoli, June 8 11, 2010 Fasma Diele, Brigida Pace (IAC-CNR) Energy-Preserving RK methods 11/06/2010 1 / 28

OUTLINE Definition of the problem and motivations s-stage trapezoidal methods Conservation of Hamiltonian function of polynomial type High order conservative methods Energy-Preserving Near-DIRK methods Numerical tests Conclusions Fasma Diele, Brigida Pace (IAC-CNR) Energy-Preserving RK methods 11/06/2010 2 / 28

DEFINITION OF THE PROBLEM We consider the numerical integration of Hamiltonian systems ( ) 0 I ẏ = J H(y), J =, I 0 where the Hamiltonian function H(y) is a polynomial. AIM: Define RK methods that exactly conserve the above Hamiltonian function: H(y n+1 ) = H(y n ) for all n and h > 0 Fasma Diele, Brigida Pace (IAC-CNR) Energy-Preserving RK methods 11/06/2010 3 / 28

MOTIVATIONS (1/2) Many interesting Hamiltonian systems arising from different fields of study are defined by polynomial Hamiltonian functions. Furthermore any nonlinear Hamiltonian function may be in general well approximated by a polynomial. It is well known that symplectic RK methods only conserve quadratic Hamiltonian functions: H(y) = 1 2 y T Cy but, in general, fail to yield conservation for higher degree. So do symmetric methods. Fasma Diele, Brigida Pace (IAC-CNR) Energy-Preserving RK methods 11/06/2010 4 / 28

MOTIVATIONS (2/2) A few years ago, the unfruitful attempts to devise energy-preserving RK methods culminated in the general feeling that they could not exist even in the simpler case of polynomial Hamiltonians (A. Iserles and A. Zanna, Preserving algebraic invariants with Runge-Kutta methods, J. Comput. Appl. Math., 125, 2000, 69 81). Fasma Diele, Brigida Pace (IAC-CNR) Energy-Preserving RK methods 11/06/2010 5 / 28

EXAMPLES Energy function evaluated over the numerical solution obtained by solving the quartic pendulum equation H(p, q) = 1 2 p2 + 1 2 q2 1 24 q4. by the Lobatto IIIA method of order four (left picture) and Gauss method of order six (right picture). Fasma Diele, Brigida Pace (IAC-CNR) Energy-Preserving RK methods 11/06/2010 6 / 28

APPROACH (1/2) Classically, a numerical method in a given class (RK, multistep methods, etc.) is defined by selecting the coefficients in such a way that the order is maximized and a given number of extra conditions are satisfied. Depending on the kind of continuous problem that is to be solved, such extra conditions confer the method specific features, for example: a good conditioning, good linear stability properties, more efficiency in solving the nonlinear systems involved, etc. Fasma Diele, Brigida Pace (IAC-CNR) Energy-Preserving RK methods 11/06/2010 7 / 28

APPROACH (2/2) Accordingly, we wonder whether some parameters defining the method may be selected not to improve the order but to guarantee the conservation of the Hamiltonian function for the discrete solution y n. More specifically, we refer to the class of RK methods. QUESTION: Can we modify a given Runge Kutta formula by adding a number of internal stages in such a way that the resulting formula applied to our problem be conservative? Fasma Diele, Brigida Pace (IAC-CNR) Energy-Preserving RK methods 11/06/2010 8 / 28

SKETCH OF THE IDEA: quadratic Hamiltonian (1/2) We consider one of the simplest RK methods: the trapezoidal method: y n+1 y n = 1 2 J ( H(y n) + H(y n+1 )) Multiplication of both sides by ( H(y n ) + H(y n+1 )) T yields ( H(y n ) + H(y n+1 )) T (y n+1 y n ) = 0 For quadratic Hamiltonian function this is equivalent to the conservation law: H(y n+1 ) = H(y n ), for all times t n Fasma Diele, Brigida Pace (IAC-CNR) Energy-Preserving RK methods 11/06/2010 9 / 28

SKETCH OF THE IDEA: quadratic Hamiltonian (2/2) In fact we consider the segment σ n joining y n to y n+1 σ n (c) = (1 c)y n + cy n+1, with c [0, 1]. and the line integral H(y n+1 ) H(y n ) = H(y)dy = y n y n+1 1 0 σ n(c) T H(σ n (c)) dc 1 =(y n+1 y n ) T H(σ n (c)) dc 0 = 1 2 (y n+1 y n ) T ( H(y n ) + H(y n+1 )) =0 H(y) being linear. Fasma Diele, Brigida Pace (IAC-CNR) Energy-Preserving RK methods 11/06/2010 10 / 28

s-stage TRAPEZOIDAL METHODS The s-stage trapezoidal method is defined as y n+1 = y n + h s b i f (K i ). i=1 where K 0 = y n, K s = y n+1 and, for i = 2,..., s 1, K i are internal stages that are located over the segment σ n (c) = (1 c)y n + cy n+1. In the simplest case they are equally distributed: K i = σ n (c i ), i = 1,..., s, with c i = (i 1)h/(s 1). Fasma Diele, Brigida Pace (IAC-CNR) Energy-Preserving RK methods 11/06/2010 11 / 28

SOME EXAMPLES When s = 2 we obtain the trapezoidal method; for s = 3 and s = 5 we obtain respectively the methods: y n+1 = y n + h ( 6 f (y n ) + 4f ( y n + y n+1 ) + f (y n+1 ) 2 ) and y n+1 = y n + h 90 ( 7f (y n ) + 32f ( 3y n + y n+1 4 +32f ( y n + 3y n+1 ) + 7f (y n+1 ) 4 ) + 12f ( y n + y n+1 ) ) 2 When applied to y = f (t), these become the Newton-Cotes quadrature formulae of order 4 and 6 respectively. On the other hand, when applied to general ODE problems their order reduces to two. Fasma Diele, Brigida Pace (IAC-CNR) Energy-Preserving RK methods 11/06/2010 12 / 28

BUTCHER ARRAY FORMULATION Written as a Runge-Kutta method, an s-stage trapezoidal method is defined by the following Butcher array: 0 0 0...... 0 c 1 c 1 b 1 c 1 b 2...... c 1 b s c 2. c 2 b 1. c 2 b 2....... c 2 b s. c s 1 c s 1 b 1 c s 1 b 2...... c s 1 b s c s b 1 b 2...... b s b 1 b 2...... b s It is easily seen that each method under consideration is symmetric. Fasma Diele, Brigida Pace (IAC-CNR) Energy-Preserving RK methods 11/06/2010 13 / 28

CONSERVATION OF ENERGY (1/2) As done for the trapezoidal method, starting from y n+1 = y n + hj s b i H(K i ), i=1 we obtain ( s b i H(K i ) T ) (y n+1 y n ) = 0. i=1 that will represent our conservation law. In fact assume that H(y) is a polynomial... Fasma Diele, Brigida Pace (IAC-CNR) Energy-Preserving RK methods 11/06/2010 14 / 28

CONSERVATION OF ENERGY (2/2) 1 H(y n+1 ) H(y n ) =(y n+1 y n ) T H(σ n (c)) dc Suppose the number of the internal stages K i be enough to guarantee that the underlying quadrature formula is exact when applied to the polynomial H(σ n (c)). Then ) H(y n+1 ) H(y n ) =(y n+1 y n ) T ( s i=1 =0 0 b i H(K i ) Fasma Diele, Brigida Pace (IAC-CNR) Energy-Preserving RK methods 11/06/2010 15 / 28

NUMERICAL TEST: Polynomial pendulum (1/2) We consider the pendulum equation defined by H(p, q) = 1/2p 2 + 1 cos q and retain a finite number of terms in the Taylor expansion of the cosine. In particular we consider: - H(p, q) = 1 2 p2 + 1 2 q2 1 24 q4, (quartic pendulum oscillator); - H(p, q) = 1 2 p2 + 1 2 q2 1 24 q4 + 1 720 q6, (pendulum oscillator of degree six); - H(p, q) = 1 2 p2 + 1 2 q2 1 24 q4 + 1 720 q6 + 1 40320 q8, (pendulum oscillator of degree eight). Fasma Diele, Brigida Pace (IAC-CNR) Energy-Preserving RK methods 11/06/2010 16 / 28

NUMERICAL TEST: Polynomial pendulum (2/2) 3-stage (first row), 5-stage (second row) and 7-stage (third row) trapezoidal methods applied to the three pendulum problems. Fasma Diele, Brigida Pace (IAC-CNR) Energy-Preserving RK methods 11/06/2010 17 / 28

HIGH ORDER CONSERVATIVE METHODS (1/2) We introduce some example of methods of order four and six that exactly preserve the Hamiltonian function of separable Hamiltonian system, in the case where such function is a polynomial of degree at most three. These methods are symmetric and are defined as where... y n+1 = y n + h s b i f (K i ). i=1 Fasma Diele, Brigida Pace (IAC-CNR) Energy-Preserving RK methods 11/06/2010 18 / 28

HIGH ORDER CONSERVATIVE METHODS (2/2) K 0 = y n, K s = y n+1 and, for i = 2,..., s 1, K i are internal stages that are located over quadratic curve of the phase space: σ n (c) = y n +2(K 2 y n )c +2(y n+1 2K 2 +y n )(c 1 2 )c, to get methods of order four, cubic curve of the phase space: σ n (c) = y n + 3(K 2 y n )c + 9 2 (K 3 2K 2 + y n )(c 1 3 )c + 9 2 (y n+1 3K 3 + 3K 2 y n )(c 1 3 )(c 2 3 )c, to get methods of order six. Fasma Diele, Brigida Pace (IAC-CNR) Energy-Preserving RK methods 11/06/2010 19 / 28

NUMERICAL TEST Methods of order four (first row) and of order six (second row), applied to cubic Hamiltonian (first column) and to quartic Hamiltonian (second column). Stepsize h = 1; number of points n = 1000. Fasma Diele, Brigida Pace (IAC-CNR) Energy-Preserving RK methods 11/06/2010 20 / 28

NOTES (1/2) The class of s-stage Trapezoidal Methods represents the first instance (and the first existence proof) of RK methods capable of providing a precise conservation of the energy, in the case where the Hamiltonian function is of polynomial type. See for example: F. Iavernaro, B. Pace. s-stage Trapezoidal Methods for the Conservation of Hamiltonian Functions of Polynomial Type. AIP Conf. Proc. 936, pp. 603-606, 2007. F. Iavernaro, B. Pace. Conservative Block-Boundary Value Methods for the Solution of Polynomial Hamiltonian Systems. AIP Conf. Proc. 1048, pp. 888-891, 2008. B. Pace. Symmetric schemes for the solution of Hamiltonian problems, PhD thesis, University of Bari (Italy), 2008. F. Iavernaro, D. Trigiante. High-order symmetric schemes for the energy conservation of polynomial Hamiltonian problems. JNAIAM, vol. 4, n. 1 2, pp. 87 101, 2009. Fasma Diele, Brigida Pace (IAC-CNR) Energy-Preserving RK methods 11/06/2010 21 / 28

NOTES (2/2) An alternative approach based on B-series and average vector fields methods has been recently provided in E. Celledoni, Robert I. McLachlan, D.I. McLaren, B. Owren, G. Reinout, W. Quispel and W.M. Wright. Energy-preserving Runge-Kutta Methods, ESAIM: M2AN 43 (2009) 645-649. A modification of collocation methods extending the averaged vector field method of high order has been developed by E. Hairer in Energy-preserving variant of collocation methods, to appear in JNAIAM, 2010. These methods are connected to the methods studied by Iavernaro and Trigiante (JNAIAM 2009) when particular quadrature formulas are chosen. Fasma Diele, Brigida Pace (IAC-CNR) Energy-Preserving RK methods 11/06/2010 22 / 28

ENERGY-PRESERVING NEAR-DIRK METHODS: first results No explicit RK method can be energy-preserving for polynomial Hamiltonians of degree higher than two and all the energy-preserving RK methods provided in literature are fully implicit. QUESTION: Can we define energy-preserving near-diagonally implicit (i.e. with sparse Butcher array) RK symmetric methods that exactly conserve polynomial Hamiltonian function? Fasma Diele, Brigida Pace (IAC-CNR) Energy-Preserving RK methods 11/06/2010 23 / 28

METHOD OF ORDER 2 0 0 0 0 0 1/3 1/12 1/4 0 0 2/3 1/8 3/8 1/8 3/24 1 1/8 3/8 3/8 1/8 1/8 3/8 3/8 1/8 Near-DIRK second order method applied to the cubic Hamiltonian (first row) and to quartic Hamiltonian (second row). Stepsize h = 1; number of points n = 1000. Fasma Diele, Brigida Pace (IAC-CNR) Energy-Preserving RK methods 11/06/2010 24 / 28

METHOD OF ORDER 2 0 0 0 0 0 0 1/4 1/24 1/6 1/24 0 0 1/2 1/12 1/3 1/12 0 0 3/4 1/12 1/3 1/8 1/6 1/24 1 1/12 1/3 1/6 1/3 1/12 1/12 1/3 1/6 1/3 1/12 Near-DIRK second order method applied to the cubic Hamiltonian (first row), to quartic Hamiltonian (second row) and to Hamiltonian of degree five (third row). Stepsize h = 1; number of points n = 1000. Fasma Diele, Brigida Pace (IAC-CNR) Energy-Preserving RK methods 11/06/2010 25 / 28

SKETCH OF THE IDEA We consider γ n joining y n to y n+1 γ n (c) = σn(c) 1 = (1 2c)y n + 2cy n+ 1, with c [0, 1 2 2 ] σ 2 n(c) = (1 2c)y n+ 1 2 and the line integral H(y n+1 ) H(y n ) = H(y)dy = = 1/2 0 =... = 0 y n y n+1 σ 1 n (c) T H(σ 1 n(c)) dc + + 2cy n+1, with c [ 1 2, 1] 1 0 1 1/2 γ n(c) T H(σ n (c)) dc σ 2 n (c) T H(σ 2 n(c)) dc Fasma Diele, Brigida Pace (IAC-CNR) Energy-Preserving RK methods 11/06/2010 26 / 28

METHOD OF ORDER 4 Near-DIRK fourth order methods applied to the cubic Hamiltonian (first column) and to quartic Hamiltonian (second column). Stepsize h = 1; number of points n = 1000. Fasma Diele, Brigida Pace (IAC-CNR) Energy-Preserving RK methods 11/06/2010 27 / 28

CONCLUSIONS The class of s-stage Trapezoidal Methods represents the first instance of RK methods capable of providing a precise conservation of the energy, in the case where the Hamiltonian function is of polynomial type.these are methods of order two that generalize the classical trapezoidal method. After that, the underlying idea, has permitted the construction of energy-preserving RK methods of orders 4 and 6 and has leaded to the discovery of arbitrary high order implicit energy preserving RK methods. The research of energy-preserving near-diagonally implicit RK formulae has being the subject of the current research. Fasma Diele, Brigida Pace (IAC-CNR) Energy-Preserving RK methods 11/06/2010 28 / 28