Aster brings flexible tools to simulations of gravitational systems

Two ways to calculate gravity

In a preprint, Federico Benevelli introduces Aster, an open-source C++ library for gravitational N-body simulations. Here, “N-body” means a collection of objects whose motions depend on their mutual gravitational attraction. The paper describes the software and presents numerical benchmarks, rather than reporting a new astronomical observation.

Aster offers two main ways to calculate gravitational forces. Direct summation evaluates each body's interaction with every other body. This avoids the grouping approximation used by the alternative solver, but the workload grows roughly as the square of the number of objects. Doubling the population therefore means about four times as many pairwise calculations.

The second approach uses the Barnes–Hut tree algorithm. Nearby particles are organized into a hierarchy, allowing sufficiently distant groups to be represented by their combined mass and center of mass. This reduces the computational burden, with an expected scaling of approximately N log N, at the cost of approximate forces.

An opening parameter controls when a group can be treated as one object. More detailed inspection of the tree can improve force accuracy, but requires more calculations. Choosing that parameter is therefore part of choosing the simulation's balance between speed and precision.

Organizing particles for parallel work

A central feature of Aster's tree implementation is Morton encoding. Each particle receives an identifier derived from its spatial coordinates. Sorting these identifiers places spatially related particles together and helps construct the tree through parallel operations rather than repeated recursive traversal.

Benevelli attributes performance benefits to this organization, including improved memory locality: related data are kept closer together, reducing inefficient memory access. Aster also uses processor vector instructions, parallel execution through TBB and GPU computing through OpenCL.

The library includes supporting tools for collecting simulation data, displaying systems in a real-time three-dimensional renderer and setting up large particle populations. These features make it a collection of numerical and analysis tools rather than a single gravity solver.

Moving forward through time

Calculating a force is only part of a simulation. An integrator uses those forces to advance positions and velocities through successive time steps.

Aster includes symplectic Euler, leapfrog, SABA methods and a Wisdom–Holman planetary integrator. Symplectic methods are designed to preserve an underlying mathematical structure of Hamiltonian dynamics, helping control long-term numerical behavior. They do not make a calculation error-free: the time step, force approximation and suitability of the method still matter.

Wisdom–Holman integration is tailored to systems dominated by a central mass. It separates the main orbital motion around that object from interactions among the smaller bodies. Benevelli's tests illustrate why matching the integrator to the physical system can matter as much as selecting a nominally higher-order method.

What the benchmarks show

For speed comparisons, Benevelli simulated a self-gravitating disk around a one-solar-mass central object, varying the particle count from 1,000 to 100,000. Aster, Rebound and Gadget4 were tested with tree-based force calculations and comparable second-order integration schemes.

The author reports that Aster's approach becomes particularly advantageous around 60,000 particles, while Gadget4 has similar execution times across much of the tested range. These results were obtained using an Intel i5-8250U processor and an NVIDIA GeForce MX130 GPU. They describe those configurations and workloads, not a universal ranking of the packages.

Other demonstrations include an eight-planet-like system evolved for 800 simulated years and tests of increasingly elongated binary orbits. The binary tests favor Wisdom–Holman integration over the tested kinetic–potential splitting methods at high eccentricity, when close approaches produce rapidly changing accelerations.

Aster also implements post-Newtonian corrections through 2.5PN order, extending its force calculations beyond Newtonian gravity. These are approximations, not a full solution of general relativity. Users can supply additional force laws, although the paper warns that arbitrary forces may undermine the properties expected from symplectic integration.

What remains uncertain

The preprint contains inconsistencies that limit precise interpretation. Its energy-error comparison gives different particle counts in the main text and figure caption. The binary-test discussion and caption also give different eccentricity thresholds for Wisdom–Holman's performance.

More broadly, bounded energy error is a useful diagnostic, but it does not by itself establish that every calculated trajectory is accurate. The reported tests support Aster as a flexible numerical toolkit for the examined problems. They do not establish equal accuracy or superior speed across all astrophysical applications.