FAST: A Fully Asynchronous Split Time-Integrator for Self-Gravitating Fluid

We describe a new algorithm for the integration of self-gravitating fluid systems using SPH method. We split the Hamiltonian of a self-gravitating fluid system to the gravitational potential and others (kinetic and internal energies) and use different time-steps for their integrations. The time integration is done in the way similar to that used in the mixed variable or multiple stepsize symplectic schemes. We performed three test calculations. One was the spherical collapse and the other was an explosion. We also performed a realistic test, in which the initial model was taken from a simulation of merging galaxies. In all test calculations, we found that the number of time-steps for gravitational interaction were reduced by nearly an order of magnitude when we adopted our integration method. In the case of the realistic test, in which the dark matter potential dominates the total system, the total calculation time was significantly reduced. Simulation results were almost the same with those of simulations with the ordinary individual time-step method. Our new method achieves good performance without sacrificing the accuracy of the time integration.

💡 Research Summary

The paper introduces FAST (Fully Asynchronous Split Time‑Integrator), a novel integration scheme designed to accelerate Smoothed Particle Hydrodynamics (SPH) simulations of self‑gravitating fluids. Traditional individual‑time‑step methods assign each particle its own hydrodynamic timestep but still compute gravitational forces for all particles at every sub‑step, making gravity the dominant cost. FAST tackles this by explicitly splitting the Hamiltonian of the system into two parts: (i) kinetic energy plus internal (pressure) energy, and (ii) the gravitational potential energy. The first part, which governs rapid hydrodynamic changes, is integrated with small, particle‑specific timesteps (Δt_hydro). The second part, which typically evolves more slowly, is integrated with a much larger, global timestep (Δt_grav).

The algorithm follows a mixed‑variable symplectic approach reminiscent of multiple‑time‑scale integrators. At each global time t, the smallest Δt_hydro among all particles is identified; the corresponding particle is advanced in position, velocity, and internal energy by that amount. When t reaches a multiple of Δt_grav, the gravitational forces for the entire system are recomputed and applied. This “fully asynchronous” scheme means that gravitational force evaluations occur only once per Δt_grav, reducing the number of expensive gravity calculations by roughly an order of magnitude. Because each sub‑step uses a second‑order symplectic operator, long‑term energy conservation remains robust, and the authors report only negligible drift in test problems.

Three benchmark tests validate the method. The first is a spherical collapse of a uniform cloud, a classic test of self‑gravity and pressure support. The second is an explosion problem where a central pressure pulse drives a rapid expansion, stressing the coupling between fast hydrodynamics and slower gravity. The third is a realistic galaxy‑merger simulation using initial conditions taken from previous work, in which a massive dark‑matter halo dominates the potential. In all cases FAST reproduces the same density, velocity, and structural evolution as a conventional individual‑time‑step code. Crucially, the number of gravity updates drops by a factor of 9–11, and in the galaxy‑merger run the total wall‑clock time is reduced by about 40 %.

The authors discuss the regimes where FAST is most advantageous: systems where the gravitational field varies slowly compared to hydrodynamic timescales, such as cosmological structure formation or dark‑matter‑dominated halos. They also note potential pitfalls: during violent events (e.g., close encounters or shock fronts) the global gravity timestep may need to be reduced or adapted dynamically to avoid force lag. Possible extensions include adaptive Δt_grav control, higher‑order symplectic operators, and integration with GPU‑accelerated gravity solvers.

In summary, FAST demonstrates that a careful Hamiltonian split combined with fully asynchronous, symplectic integration can dramatically cut the computational cost of self‑gravitating SPH simulations without sacrificing accuracy. This approach opens the door to more efficient large‑scale astrophysical simulations, offering a practical pathway to higher resolution and longer physical timescales in studies of galaxy formation, star cluster dynamics, and cosmological fluid evolution.