nlchains: A fast and accurate time integration of 1-D nonlinear chains on GPUs

Abstract

We present nlchains, a software for simulating ensembles of one-dimensional Hamiltonian systems with nearest neighbor interactions. The implemented models are the α-β Fermi–Pasta–Ulam–Tsingou model, the discrete nonlinear Klein–Gordon model with equal or site-specific masses, the Toda lattice and the discrete nonlinear Schrödinger equation. The integration algorithm in all cases is a symplectic sixth order integrator, hence very accurate and suited for long time simulations. The implementation is focused on performance, and the software runs on graphical processing unit hardware (CUDA). We show some illustrative simulations, we estimate the runtime performance and the effective scaling of the cumulative error during integration. Finally, we give some basic pointers to extend the software to specific needs. Keywords: FPU, Nonlinear chain, GPU, CUDA