A Structure-Preserving Method of Fundamental Solutions for the Multi-Phase Mullins-Sekerka Flow
arXiv:2607.12759
2026
Optimization
2 ideas extracted · analyzed Aug 30, 2026
What the math gives to ML
The paper combines a mesh-free harmonic representation with an exact null-space treatment of conserved quantities. The most transferable component is not the method of fundamental solutions itself, but the separation between a soft residual solve and an exactly enforced linear invariant constraint: conservation is maintained by projecting the least-squares solution onto the null space of the discrete constraint matrix. This suggests a neural operator or physics-informed network whose final predicted state is obtained from a constrained least-squares correction, preserving mass, charge, probability, or other batch-level invariants to numerical precision while retaining soft penalties for difficult residuals.
Ideas from this paper
✓✓ Beats tuned baseline
2026
Add an exact linear-constraint projection to the output solve of a neural operator or physics-informed model. The network produces an unconstrained prediction or coefficient vector, while a small constrained least-squares layer removes the component violating known conservation laws and separately penalizes residuals that cannot be enforced exactly.
Useful7/10
Difficulty5/10
Novelty5/10
Unverified
2026
Use a CFL-like step-size controller for neural simulators or neural ODE rollouts, shrinking the integration step when the predicted state changes rapidly and relaxing it when dynamics are smooth. The controller uses the smallest spatial resolution and maximum predicted velocity, rather than a fixed global step chosen for the worst case.
Useful5/10
Difficulty3/10
Novelty4/10