A Positivity-Preserving Expectation Scheme for Hamilton--Jacobi--Bellman Equations with Oblique Robin Boundary Conditions
arXiv:2608.11936
2026
Architecture
1 ideas extracted · analyzed Sep 1, 2026
What the math gives to ML
The paper constructs a Bellman propagation operator whose coefficients are nonnegative because each update is represented as a conditional expectation over reflected weak-Euler branches. The transferable asset is a monotone, positivity-preserving neural layer: branch probabilities stay fixed and nonnegative, while oblique Robin effects enter through nonnegative attenuation and source factors. This structure can replace unconstrained spatial mixing in PDE surrogates, controlled-dynamics models, or value-function networks. The clearest test is long-horizon rollout stability and monotonicity on a learned reflected diffusion.
Ideas from this paper
Unverified
2026
Replace an unconstrained spatial aggregation in a neural PDE surrogate or controlled-dynamics model with a fixed-branch expectation layer. Each output is a maximum over controls of a nonnegative weighted average of next-state values, with reflected overshoots attenuated by Robin factors. Increasing any input value therefore cannot decrease the output, giving a hard monotonicity and positivity property instead of relying on a penalty.
Useful6/10
Difficulty5/10
Novelty7/10