University of Leeds
Workshop on Geometric Integration, 4th May, 1999


A. Iserles, Cambridge
Email: A.Iserles@damtp.cam.ac.uk

Solving linear differential equations in Lie groups

Abstract

The subject matter of this talk is the numerical solution of differential equations that evolve on Lie groups. Such equations are important in many applications, when it is important to respect invariants and symmetries.

We will demonstrate that the solution can be pushed to the underlying Lie algebra and expanded there in Magnus series. The structure of the series and their construction use graph theory. The discretization of the expansion requires a range of techniques in multivariate integration, combinatorics and the theory of free Lie algebras. The outcome is a technique that allows for a high-precision integration of complicated differential equations at an affordable cost and that has been already applied, e.g. to the determination of Sturm--Liouville spectra and to integration in homogeneous spaces.


C.J. Budd (Bath)
Email: cjb@maths.bath.ac.uk

Incorporating symmetry and the maximum principle into a numerical scheme

Abstract

In the theory of partial differential equations two very important features governing the qualitative behaviour of their solutions are the underlying Lie symmetries and the maximum principle. The first because it allows a reduction of the system to an ordinary differential equation with resulting self-similar solutions. The second because it gives an ordering on the solutions which often impleis that the self-similar solutions are global attractors.

If a numerical method is to accurately reproduce the qualitative behaviour of a partial differential equation it is desirable that it should be both invariant under the action of the symmetry group and also posess a maximum principle. In practice it is hard to achieve both of these two desirable qualities simultaneously. However in certain special case it can be done. We will describe a discretisation of the porous medium equation that not only has symmetry invariance and a maximum principle but also retains certain conservation laws. As a result is has excellent global qualitative features.

This is joint work with Matthew Piggott (Bath)


D.J. Higham (Strathclyde)
Email: d.j.higham@strath.ac.uk

Orthogonal Integration and the Computation of Lyapunov Exponents

Abstract

The problem of computing approximate Lyapunov exponents has been addressed in various application areas, especially in the physics community, but it has received relatively little attention from numerical analysts. Hence the convergence theory is incomplete. In this talk we focus on the discrete QR iteration, which has two main elements:
  1. the solution of an ODE with a built-in orthogonality structure, and
  2. the approximation of an infinite-time integral.
A well-designed algorithm must balance the error contributions from the numerical ODE solver and the quadrature process. To get some insight into how this balancing act may take place, we present a convergence analyse for a simple class of test problems. The analysis raises an intesting subproblem in numerical linear algebra.


S. Reich (Surrey)
Email: S.Reich@mcs.surrey.ac.uk

Multi-Symplectic Integrators: Numerical Methods That Make Waves

Abstract

A wide range of conservative PDEs can be formulated as a multi-symplectic Hamiltonian PDE. Contrary to the "classical" approach to Hamiltonian PDEs as an evolutionary system on an appropriate function space, the multi-symplectic formulation treats both time and space as local quantities leading to local conservation laws for symplecticity, energy, and momentum. In my talk I will present numerical schemes that lead to a discrete conservation of symplecticity and excellent conservation of energy and momentum. This work can be seen as a natural extension of symplectic integration for classical mechanics to Hamiltonian PDEs.


H. Munthe-Kaas (Bergen, Norway)
Email: hans@ii.uib.no or H.Munthe-Kaas@damtp.cam.ac.uk

Canonical integrators and the search for good actions

Abstract

Modern numerical Lie group integrators for differential equations are formulated in an abstract language, which may be interpreted in different concrete settings. We will give an introduction to various views of these methods. Whereas classical integration schemes are always based on stepping forwards by translations on R^n, the new scemes allow a wide range of different actions for advancing the solution. The choice of actions is crucial for the properties of the numerical solution, and the problem of choosing a good action can be compared to the problem of finding a good preconditioner in an iterative solver for a linear system. In both cases, one wants approximations to the original system that are easy to solve and that captures some essential properties of the system we want to integrate. We will through various examples show the importance of finding good actions.


Contact Carsten Knudsen (email: carsten@amsta.leeds.ac.uk ) or Vadim Kuznetsov (email: vadim@amsta.leeds.ac.uk ) for a TEX file of the programme and further details.


<< Move to

Page created by allan@amsta.leeds.ac.uk
Last updated 16th March, 1999.