Prob-GParareal: A Probabilistic Numerical Parallel-in-Time Solver for Differential Equations

A research team has introduced Prob-GParareal, a probabilistic parallel-in-time solver for differential equations that quantifies numerical uncertainty while remaining compatible with existing classical solvers, according to a paper posted on arXiv [1]. The method, detailed in a preprint submitted on 4 Sep 2025 and revised on 14 Jun 2026, extends the GParareal algorithm by using Gaussian processes to model the Parareal correction function [1][2]. This approach allows the propagation of numerical uncertainty across time and produces probabilistic forecasts of a system's evolution [2]. The authors state that Prob-GParareal accommodates probabilistic initial conditions and maintains compatibility with classical numerical solvers, which ensures straightforward integration into existing Parareal frameworks [1][2]. The paper includes a theoretical analysis of the computational complexity and derives error bounds for the new algorithm [2]. The authors then demonstrate its accuracy and robustness on five benchmark ordinary differential equation systems, including chaotic, stiff, and bifurcation problems [1][2]. A variant called Prob-nnGParareal, which replaces the Gaussian processes with nearest-neighbors Gaussian processes, is also introduced to illustrate increased performance on an additional partial differential equation example [2]. The work appears on arXiv, an open-access repository of electronic preprints that is not peer-reviewed but is widely used in mathematics, physics, and computer science [6]. As of November 2024, the repository receives about 24,000 submissions per month [6]. The paper is listed under the Statistics and Computation category and is accessible through the arXiv abstract page, which features experimental community tools under the arXivLabs framework [1][4]. arXivLabs, launched in 2020, provides a space for third-party collaborators to develop features such as bibliographic explorers and code finders that add value for readers and authors [4][5]. The submission history shows the initial version was 979 KB and the revised second version is 411 KB [1]. The corresponding author is listed as Guglielmo Gattiglio [1]. The researchers argue that this work bridges a gap in the development of probabilistic counterparts to established parallel-in-time methods [2].