MOOSE Newsletter (December 2022)
Support Ended for PETSc 3.5
Full support for PETSc 3.5 has ended, as the few remaining references to 3.5 have been removed from MOOSE. If using PETSc 3.5, please consider updating to the current supported release (3.16.6) or the alternate tested release (3.11.4) using either of the update_and_rebuild_petsc scripts.
MOOSE Improvements
Added new ability to mark a solution as "invalid". This gives developers the ability to say that a solution is "out of bounds" for things like material correlations. A solution is allowed to be invalid _during_ a nonlinear solve - but not once the solve has converged.
MOOSE-Libtorch interface improvements, enabling Deep Reinforcement Learning
The interface between MOOSE and Libtorch has been extended by:
- wrapped objects that can train artificial neural networks - the ability to read neural networks trained in python using a TorchScript format - an implementation of the Proximal Policy Optimization (PPO) algorithm in the Stochastic Tools Module - reporters that can print neural network parameters - controllers which can use neural networks to control simulations