Data-driven Whitney forms for structure-preserving control volume analysis

Jonas A. Actor, Xiaozhe Hu, Andy Huang, Scott A. Roberts, Nathaniel Trask

Abstract: Highlights•Partition of unity architecture learns physically-relevant control volumes from data.•Machine-learned Whitney forms preserve underlying topological structure.•Learned models exhibit algebraic convergence comparable to FEM methods.•Learned models reduce degrees of freedom by 40000x for lithium-ion battery problem.