Quantifying Rare Events in Stochastic Reaction-Diffusion Dynamics Using Tensor Networks

TL;DR

This paper introduces a tensor network-based method to efficiently compute macroscopic transition rates in stochastic reaction-diffusion systems, overcoming limitations of traditional sampling techniques and enabling analysis of extremely large state spaces.

Contribution

The authors develop a novel tensor network approach to extract transition rates from the full ensemble evolution of reaction-diffusion systems, bypassing rare event sampling challenges.

Findings

Successfully computed switching rates in a large 1D reaction-diffusion model.

Demonstrated subexponential growth in computational expense with system size.

Showed the method's ability to handle approximately 3×10^{15} microstates.

Abstract

The interplay between stochastic chemical reactions and diffusion can generate rich spatiotemporal patterns. While the timescale for individual reaction or diffusion events may be very fast, the timescales for organization can be much longer. That separation of timescales makes it particularly challenging to anticipate how the rapid microscopic dynamics gives rise to macroscopic rates in the non-equilibrium dynamics of many reacting and diffusing chemical species. Within the regime of stochastic fluctuations, the standard approach is to employ Monte Carlo sampling to simulate realizations of random trajectories. Here, we present an alternative numerically tractable approach to extract macroscopic rates from the full ensemble evolution of many-body reaction diffusion problems. The approach leverages the Doi-Peliti second-quantized representation of reaction-diffusion master equations along with compression and evolution algorithms from tensor networks. By focusing on a Schl\"{o}gl model with one-dimensional diffusion between $L$ otherwise well-mixed sites, we illustrate the potential of the tensor network approach to compute rates from many-body systems, here with approximately $3 \times 10^{15}$ microstates. Specifically, we compute the rate for switching between metastable macrostates, with the expense for computing those rates growing subexponentially in $L$. Because we directly work with ensemble evolutions, we crucially bypass many of the difficulties encountered by rare event sampling techniques$\unicode{x2013}$detailed balance and reaction coordinates are not needed.