On the preservation of phase space structure under multisymplectic discretization (Q598165): Difference between revisions

This paper is concerned with the (local and global) preservation properties of multisymplectic schemes when applied to the numerical solution of multisymplectic Hamiltonian partial differential equations (PDEs). In the one dimensional space case a Hamiltonian PDEs is said to be multisymplectic if it can be written in the form (1) \( M z_t + K z_x = \nabla_z S,\) where the state variable \( z = z(t,x) \) is a \(d\)-multidimensional real variable, \(M\) and \(K\) are \(d\)-dimensional real skew-symmetric matrices and \( S=S(z)\) is a smooth function. For this equation we may associate the two forms \( \omega = (1/2) ( dz \wedge M dz )\), \( \kappa = (1/2) ( dz \wedge M dz )\) that satisfy the multisymplectic conservation law \( \partial_t \omega + \partial_x \kappa = 0 \). In this setting multisymplectic discretizations are those which preserve a discrete version of the above conservation law. Thus, application of the symplectic implicit midpoint rule to each independent variable of (1) yields to the multisymplectic centered cell discretization. The authors focus their attention in two PDEs: 1) The one dimensional nonlinear Schrödinger equation: \( i u_t + u_{xx} + 2 | u| ^2 u = 0 \) with periodic boundary conditions that can be written in multisymplectic form with \( z=(p,q,v,w)\), \( u=p+i q\), \( v=p_x\), \( w=q_x\). 2) The sine-Gordon equation: \( u_{tt} - u_{xx} + \sin u = 0 \) also with periodic boundary conditions that is written as a fourth-dimensional state multisymplectic equation by using a Lagrangian formulation with a constraint. The authors present the results of several numerical experiments with the two problems by using the above mentioned centered cell discretization and also a multisymplectic spectral method. From a detailed study of the numerical results they conclude that these geometric integrators preserve the structure of the flow better than other standard integrators with a reasonable computational cost.