arXiv preprint ·

On the statistical reliability of chaotic simulations at reduced precision

Ross Ah-Weng and Peter V. Coveney

Published
Source
arXiv preprint, posted before peer review
Original
Read it at arXiv
Topics
Computational PhysicsDynamical SystemsNumerical AnalysisChaotic Dynamics

Abstract

Reduced-precision arithmetic is increasingly used to accelerate scientific simulation, while AI workloads are driving its growing prominence in hardware design. Yet small arithmetic errors do not necessarily imply accurate statistics. Finite representation limits spatial resolution, while rounding can merge trajectories that should separate in chaotic systems, progressively concentrating probability on a shrinking collection of computer states. This redistribution can distort statistical averages well before periodicity becomes apparent; furthermore, disjoint computational basins can induce an initialisation bias, even when the intended equilibrium statistics are independent of the initial ensemble. We use functional graphs and their transfer operators to connect these pathologies to bounds on statistical bias, the duration of accurate averaging, and the effect of reducing precision. Exhaustive numerical experiments on four chaotic maps, with 1-24 significand bits and the 24-bit endpoint matching IEEE FP32, show how initially accurate ensembles drift towards incorrect statistics, with lower precision often bringing this deterioration forward. Stochastic rounding substantially extends some useful averaging windows but leaves others severely biased. Longer computed trajectories can therefore worsen inferred statistics, and a finite average can outperform the "infinite-time" computer average by orders of magnitude. This exposes a risk of statistically biased inference at reduced precision and provides a further motivation for collecting statistics from ensembles of shorter trajectories.

The authors’ abstract, as arXiv lists it. arXiv’s listings, abstracts included, are free for anyone to reuse.

Read it on ReadSearch

Each opens after you create an account, and brings you back to this paper.

Already a reader? Sign in

All new papers in Computational Physics

Get the few papers worth your time in Computational Physics every morning.

ReadSearch reads the new papers in your field each day and brings you the few worth reading, each with its key finding in a sentence.

Papers belong to their authors and publishers; every page here links to the original. All research topics