SymX: Energy-based Simulation from Symbolic Expressions
share
Optimization time integrators have proven to be effective at solving complex multi-physics problems, such as deformation of solids with non-linear material models, contact with friction, strain limiting, etc. For challenging problems with high accuracy requirements, Newton-type optimizers are often used. This necessitates first- and second-order derivatives of the global non-linear objective function. Manually differentiating, implementing and optimizing the resulting code is extremely time-consuming, error-prone, and precludes quick changes to the model. We present SymX, a framework based on symbolic expressions that computes the first and second derivatives by symbolic differentiation, generates efficient vectorized source code, compiles it on-the-fly, and performs the global assembly of element contributions in parallel. The user only has to provide the symbolic expression of an energy function for a single element in the discretization and our system will determine the assembled derivatives for the whole model. SymX is designed to be an integral part of a simulation system and can easily be integrated into existing ones. We demonstrate the versatility of our framework in various complex simulations showing different non-linear materials, higher-order finite elements, rigid body systems, adaptive cloth, frictional contact, and coupling multiple interacting physical systems. Moreover, we compare our method with alternative approaches and show that SymX is significantly faster than a current state-or-the-art framework (up to two orders of magnitude for a higher-order FEM simulation).
READ FULL TEXT