Error Bound and Stability Analysis for a Randomized Singly Diagonally Implicit Runge-Kutta Method
arXiv:2607.18928
2026
Dynamics
1 ideas extracted · analyzed Aug 30, 2026
What the math gives to ML
The paper provides a randomized, singly diagonally implicit Runge–Kutta construction whose individual realizations retain algebraic stability while achieving unusually high root-mean-square convergence under low temporal regularity. The transferable asset is the combination of randomized quadrature, shared diagonal implicit solves, and stability for stiff or nonsmooth vector fields. A promising neural-network use is a stochastic SDIRK integrator for neural ODE, state-space, and diffusion probability-flow trajectories, where learned vector fields and score networks are often only piecewise smooth. The first test should compare equal-function-evaluation randomized SDIRK against Heun, midpoint, deterministic SDIRK, and diffusion-specific solvers on stiff and nonsmooth learned vector fields.
Ideas from this paper
Unverified
2026
Replace the explicit Euler, Heun, or fixed-step midpoint update used for a neural ODE or diffusion probability-flow trajectory with a two-stage randomized SDIRK step. Draw one random scalar per time step, use it in both implicit stage equations, and solve each stage with Newton or damped fixed-point iteration. The randomness targets quadrature error caused by nonsmooth score networks, while the singly diagonal structure permits reuse of the same Jacobian preconditioner for both stage solves.
Useful6/10
Difficulty7/10
Novelty6/10