Equivariant Neural Prediction of the Stokes Resistance Tensors for Arbitrary Microparticle Shapes

This paper presents an SO(3)-equivariant neural network for predicting the grand resistance matrix of irregular microparticles in Stokes flow. The 6×6 tensor encodes translational and rotational hydrodynamic response, with its A and C blocks governing settling, diffusion, and orientational transport. Unlike empirical drag correlations that compress the tensor to a single scalar, this approach preserves the full orientation-dependent drag and rotational response. The method targets microfluidics, colloidal science, and particle-based soft matter applications.