Fast Numerical Coarsening with Local Factorizations

Numerical coarsening methods offer an attractive methodology for fast simulation of objects with high-resolution heterogeneity. However, they rely heavily on preprocessing, and are not suitable when undergo dynamic material or topology updates. We present that largely accelerate the two main processes numerical coarsening, namely training data generation optimization shape functions, as a result we manage to leverage runtime under local To data, propose domain-decomposition solver based substructuring leverages factorizations. computation decoupled smoothness fitting. evaluate quantitatively accuracy performance our proposed methods, show achieve comparable baseline, albeit speed-ups orders magnitude. also demonstrate example simulations