Error bounds improve neural network solutions for heat and phase change problems
Certified neural approximations for parabolic obstacle problems
Machine Learning
Summary
Solving problems where something changes state, like ice melting into water, is tricky because traditional neural networks don’t guarantee how close their answers are to the true solution. The researchers developed a method to measure and certify the error in neural network solutions for these ‘free-boundary’ problems involving heat flow and obstacles. Their approach ensures that the neural network's predictions stay within a known error range, improving trustworthiness in simulations. They tested their method on example problems and showed that the error stays reasonably small and can be computed reliably.
What this means in practice
- •For computational engineers: Certify error bounds for neural network models simulating heat and phase change in complex materials.
- •For mechanical simulation developers: Provide reliable accuracy estimates in software simulating contact and phase change phenomena.
Authors
Davood Damircheli
Abstract
Residual-trained neural surrogates carry no error guarantee, a limitation for free-boundary problems such as phase change and contact. We derive guaranteed error bounds for neural approximations of parabolic obstacle problems in first-order state--flux--multiplier form. The main result bounds the flux error of the network and the energy error of its feasible projection $\max\{v,g\}$ by the constitutive residual, the initial mismatch, the complementarity defect, and the PDE residual weighted by $c_F^2/α$, without the coercivity and norm-equivalence constants of the first-order formulation. It extends the Apushkinskaya--Repin majorant for the heat operator to general coefficients, free multipliers and a flux bound. A penetration-consistency result shows that neural triples with arbitrarily small bound exist. Outward-rounded cellwise Taylor-model enclosures turn the bound into a finite computation valid on any adaptive partition. The flux and gradient errors of the projected output are certified below $0.27$ and $0.33$ of the solution scale for a one-dimensional Stefan problem and, after adaptive refinement, below one half for all three trained networks on a two-dimensional expanding disk. On an expanding ellipse the gradient bound stays above one half; substitution diagnostics trace the floor of the trained networks to the projected state near the free boundary, and the remaining gap to enclosure overhead. Without a closed-form solution, a validated lower bound replaces the solution norm; there the enclosure is nearly exact, but the trained networks have errors above the solution scale and no useful accuracy is certified. The bounds control a state--flux norm, not the multiplier or the free-boundary location.