Smoothed Particle Hydrodynamics in pkdgrav3 for Shock Physics Simulations. I. Hydrodynamics

TL;DR

pkdgrav3 is a high-performance, scalable, parallel tree-SPH code optimized for large-scale hydrodynamic simulations with self-gravity on CPU/GPU architectures, validated through standard tests and used in planetary impact modeling.

Contribution

The paper introduces pkdgrav3, a novel, efficient, and scalable parallel tree-SPH code that combines advanced algorithms for gravity and hydrodynamics on heterogeneous systems.

Findings

pkdgrav3 scales efficiently to thousands of CPU cores and GPUs.

The code demonstrates excellent accuracy in standard hydrodynamic tests.

It has been successfully applied to planetary impact simulations.

Abstract

We present pkdgrav3, a high-performance, fully parallel tree-SPH code designed for large-scale hydrodynamic simulations including self-gravity. Building upon the long development history of pkdgrav, the code combines an efficient hierarchical tree algorithm for gravity and neighbor finding with a modern implementation of Smoothed Particle Hydrodynamics (SPH) optimized for massively parallel hybrid CPU/GPU architectures. Its hybrid shared/distributed memory model, combined with an asynchronous communication scheme, allows pkdgrav3 to scale efficiently to thousands of CPU cores and GPUs. We validate the numerical accuracy of pkdgrav3 using a suite of standard tests, demonstrating excellent agreement with analytic or reference solutions. The code was already used in several peer-reviewed publications to model planetary-scale impacts, where SPH's Lagrangian nature allows accurate tracking of material origin and thermodynamic evolution. These examples highlight pkdgrav3's robustness and efficiency in simulating highly dynamical, self-gravitating systems. pkdgrav3 thus provides a powerful, flexible, and scalable platform for astrophysical and planetary applications, capable of exploiting the full potential of modern heterogeneous high-performance computing systems.