b'Accelerate MultiphysicsAdvanced multiphysics solve techniques using modern Solves Usingmachine learning benefit research in all computational science domains that depend on nonlinear solution strategies.Deep Learning M ultiphysics simulation is essential in computational science. For decades, multiphysics simulation has been extensively investigated because the phenomena it models are encountered in many applications of interest to DOE and other national agencies. However, solution time significantly limits the applicability of multiphysics simulation. In particular, variation in nonlinearity in multiphysics systems is the leading cause of numerical difficulties PROJECT NUMBER:in standard simulation algorithms. 21A1057-027This project explored advanced nonlinear solver techniques, including a nonlinear TOTAL APPROVED AMOUNT:subset accelerator and machine learning techniques. The basic idea of the nonlinear $125,000 over 1 year subset accelerator was to form a nonlinear subproblem using the components with PRINCIPAL INVESTIGATOR:high residual values, which correspond to high nonlinearity. The solution of the Fande Kong nonlinear subset accelerator was employed as the initial guess for each nonlinear iteration of the outer solver. We verified the algorithms effectiveness using a CO-INVESTIGATOR:computational fluid dynamics example, a lid-driven cavity. The new algorithm Alexander Lindsay, INL worked for certain problems with small viscosity while the traditional approach did not. In addition, we initially explored the application of deep learning and machine learning to improve the developed nonlinear solver.Nonlinearity for lid-driven cavity, based on which a nonlinear accelerator was created to accelerate the solver performance.40'