no. 16 · Tuesday 17 November 2026
First-order convergence of finite-difference approximations to the minimum action for diffusions with multiplicative noise
Abstract
We prove first-order convergence of finite-difference approximations to the Freidlin–Wentzell minimum action for diffusions with multiplicative noise, which answers the question of the sharp rate left open in [Hong, Jin, Sheng, \emph{IMA J. Numer. Anal.} 45 (2025), 1501–1536]. As shown in that work, a bound on the action controls only the $L^2$ norm of the velocity, which leads to a half-order estimate for the difference between the discrete and continuous actions. To improve it, we use the discrete Euler–Lagrange equation to derive mesh-independent $L^\infty$ bounds on the velocity of piecewise affine interpolants of stationary discrete paths with bounded action. For the case of a unique continuous minimizer, we identify the leading coefficient of the error between the discrete and continuous minimum actions on uniform grids, and a scalar example shows that the first-order error bound is sharp. We also prove strong $H^1(0,T;\mathbb R^d)$ subsequential convergence of interpolated discrete minimizers. Numerical experiments for a two-protein genetic toggle switch confirm first-order convergence of the discrete action for transitions in both directions.
- fields
- Numerical Analysis