Software prefetch optimization for event-driven molecular dynamics of hard particles

Event-driven molecular dynamics (EDMD) is the method of choice for simulating hard-particle systems, underpinning studies of phase behavior, nucleation, and glassy dynamics, as well as of non-equilibrium phenomena ranging from granular to active matter. Decades of algorithmic refinement—single-event scheduling, hybrid event calendars, and optimized neighbor searches—have reduced the computational cost per event to the point where, as we show, the dominant bottleneck for large systems is no longer arithmetic but memory-access latency: the irregular, data-dependent access pattern of event processing defeats hardware prefetching once the particle data outgrow the processor cache. We eliminate this bottleneck by exploiting a structural feature that is unique to event-driven simulation: upcoming events are encoded in the event calendar before they are processed. A multi-stage software-prefetch pipeline, staged across consecutive events so as to resolve the pointer-chasing dependency from particle pairs to their neighbor lists, overlaps DRAM transfers with useful computation. The optimization leaves the algorithm, the event ordering, and the trajectories bit-for-bit unchanged, yet accelerates simulations of hard disks and hard spheres by 15–25% for [Formula: see text] across mono- and polydisperse systems in two and three dimensions, outperforming existing EDMD codes. Our results demonstrate that hardware-aware event scheduling opens a new axis of optimization for event-driven algorithms in general.

Authors

Publication Details

Journal
International Journal of Modern Physics C
Published
2026-11-13
DOI
https://doi.org/10.1142/s0129183127501889
Primary Topic
Parallel Computing and Optimization Techniques
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Software prefetch optimization for event-driven molecular dynamics of hard particles

Masaharu Isobe, Nobuaki Murase, Masahiro Miwa
International Journal of Modern Physics C
Parallel Computing and Optimization Techniques
article

Software prefetch optimization for event-driven molecular dynamics of hard particles

Masaharu Isobe, Nobuaki Murase, Masahiro Miwa
article en

Abstract

Event-driven molecular dynamics (EDMD) is the method of choice for simulating hard-particle systems, underpinning studies of phase behavior, nucleation, and glassy dynamics, as well as of non-equilibrium phenomena ranging from granular to active matter. Decades of algorithmic refinement—single-event scheduling, hybrid event calendars, and optimized neighbor searches—have reduced the computational cost per event to the point where, as we show, the dominant bottleneck for large systems is no longer arithmetic but memory-access latency: the irregular, data-dependent access pattern of event processing defeats hardware prefetching once the particle data outgrow the processor cache. We eliminate this bottleneck by exploiting a structural feature that is unique to event-driven simulation: upcoming events are encoded in the event calendar before they are processed. A multi-stage software-prefetch pipeline, staged across consecutive events so as to resolve the pointer-chasing dependency from particle pairs to their neighbor lists, overlaps DRAM transfers with useful computation. The optimization leaves the algorithm, the event ordering, and the trajectories bit-for-bit unchanged, yet accelerates simulations of hard disks and hard spheres by 15–25% for [Formula: see text] across mono- and polydisperse systems in two and three dimensions, outperforming existing EDMD codes. Our results demonstrate that hardware-aware event scheduling opens a new axis of optimization for event-driven algorithms in general.

International Journal of Modern Physics C
Parallel Computing and Optimization Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon