Convergence of AMR and SPH simulations - I. Hydrodynamical resolution and convergence tests

TL;DR

This study compares hydrodynamical simulations using AMR finite volume and SPH methods, demonstrating their convergence at high resolution and identifying specific conditions where each method performs well or faces challenges.

Contribution

It provides a detailed comparison of AMR and SPH hydrodynamical methods, highlighting convergence behaviors, strengths, and limitations in various shock and instability tests.

Findings

Methods converge at high resolution in most cases.

Both methods handle adiabatic and well-resolved cooling shocks effectively.

SPH simulations are significantly slower than finite volume simulations at comparable resolutions.

Abstract

We compare the results for a set of hydrodynamical tests performed with the AMR finite volume code, MG and the SPH code, SEREN. The test suite includes shock tube tests, with and without cooling, the non-linear thin-shell instability and the Kelvin-Helmholtz instability. The main conclusions are : (i) the two methods converge in the limit of high resolution and accuracy in most cases. All tests show good agreement when numerical effects (e.g. discontinuities in SPH) are properly treated. (ii) Both methods can capture adiabatic shocks and well-resolved cooling shocks perfectly well with standard prescriptions. However, they both have problems when dealing with under-resolved cooling shocks, or strictly isothermal shocks, at high Mach numbers. The finite volume code only works well at 1st order and even then requires some additional artificial viscosity. SPH requires either a larger value of the artificial viscosity parameter, alpha_AV, or a modified form of the standard artificial viscosity term using the harmonic mean of the density, rather than the arithmetic mean. (iii) Some SPH simulations require larger kernels to increase neighbour number and reduce particle noise in order to achieve agreement with finite volume simulations. However, this is partly due to the need to reduce noise that can corrupt the growth of small-scale perturbations. In contrast, instabilities seeded from large-scale perturbations do not require more neighbours and hence work well with standard SPH formulations and converge with the finite volume simulations. (iv) For purely hydrodynamical problems, SPH simulations take an order of magnitude longer to run than finite volume simulations when running at equivalent resolutions, i.e. when they both resolve the underlying physics to the same degree. This requires about 2-3 times as many particles as the number of cells.