A Randomized PDE Energy driven Iterative Framework for Efficient and Stable PDE Solutions

Researchers introduce RPEF, a training-free PDE solver that replaces finite element matrix assembly and neural network training with energy-driven diffusion iterations plus Gaussian smoothing, enforcing boundary conditions at every step. Tested on 1D Poisson, Heat, and Burgers equations, it achieves stable convergence from arbitrary random initializations. Computational scientists seeking mesh-free, data-free alternatives to FEM or PINNs should evaluate this, though 1D-only validation leaves 3D scalability unproven.