Accuracy and Efficiency of Raytracing Photoionisation Algorithms

TL;DR

This paper compares three photoionisation algorithms in hydrodynamical simulations, highlighting a second-order explicit method as the most efficient and scalable, with implications for improving existing simulation codes.

Contribution

It introduces a second-order explicit algorithm for photoionisation that outperforms existing methods in efficiency and scalability, and provides guidelines for optimal timestepping criteria.

Findings

Second-order explicit algorithm is most efficient.

Implicit methods have larger errors with multi-frequency radiation.

Parallel scaling of the second-order explicit method is significantly better.

Abstract

Three non-equilibrium photoionisation algorithms for hydrodynamical grid-based simulation codes are compared in terms of accuracy, timestepping criteria, and parallel scaling. Explicit methods with first-order time accuracy for photon conservation must use very restrictive timestep criteria to accurately track R-type ionisation fronts. A second-order accurate algorithm is described which, although it requires more work per step, allows much longer timesteps and is consequently more efficient. Implicit methods allow ionisation fronts to cross many grid cells per timestep while maintaining photon conservation accuracy. It is shown, however, that errors are much larger for multi-frequency radiation than for monochromatic radiation with the implicit algorithm used here, and large errors accrue when an ionisation front crosses many optical depths in a single step. The accuracy and convergence rates of the different algorithms are tested with a large number of timestepping criteria to identify the best criterion for each algorithm. With these criteria selected, the second-order explicit algorithm is the most efficient of the three, and its parallel scaling is significantly better than that of the implicit algorithm. The upgrade from first- to second-order accuracy in explicit algorithms could be made very simply to fixed-grid and adaptive mesh-refinement codes which currently use a first-order method.