MOOSE Newsletter (September 2025)

Preliminary Kokkos-MOOSE release

A new GPU backend, Kokkos, was added in MOOSE and is available for preliminary testing with NVIDIA GPUs. Kokkos-MOOSE provides programming and user interfaces that are designed to resemble the original MOOSE interfaces to make adaptation easier and reduce porting efforts. Kokkos-MOOSE currently supports the GPU version of Kernels, NodalKernels, BCs, and Materials and will be continuously extended to incorporate more MOOSE capabilities and support more GPU vendors. See this page to get started with Kokkos-MOOSE.

TestHarness --only-tests-that-require option

The capability parameter for test specifications allows for run-time checking of whether a test can or cannot run based on capabilities defined by the application executable. The --only-tests-that-require command line option has been added to the TestHarness, which allows filtering of tests based on whether or not they depend on a capability.

For example, take a situation in which ./run_tests --only-tests-that-require cuda is run along with the following test specification:

[Tests]
  [needs_cuda]
    ...
    capabilities = 'cuda'
  []
  [doesnt_need_cuda]
    ...
    capabilities = 'libtorch'
  []
[]

In this case, the test needs_cuda will be run because it depends on the cuda capability, while the doesnt_need_cuda test will not run because it does not depend on the cuda capability.

Optimizations

Face and neighbor face material properties are now only computed if the boundary conditions and interface kernels active on the relevant boundary have requested material properties, or if any active constraint has requested a material property. The logic for deciding whether to execute on internal sides for DG and HDG kernels was reworked to limit queries in the dedicated warehouses within each subdomain. The expected computational cost reduction for each simulation is highly dependent on the set of active features and the cost repartition. Use profiling to determine the computational cost repartition.

Solid mechanics

The ReactionForceAux was added to leverage the TaggingInterface to output the forces, as applied when computing the residual, at the desired nodes.

Contact

The PenetrationLocator was upgraded to be able to support finding elements that are proximal to but not connected to these nodes. This may occur on non-conforming surface meshes, such as Flex IGA meshes or 3D meshes with hanging nodes from adaptive refinement.

Objects based on the GeometricSearchInterface still default to the old search_method=nearest_node_connected_sides, which only finds paired points on sides connected to the nearest node to a queried point, but with the new search_method=all_proximate_sides option, contact and gap objects can instead use the point locator octree to test all sides touching a point.

Fluid Structure Interaction

The Newmark-Beta time integration scheme can now be used in Arbitrary Lagrangian Eulerian (ALE) simulations. New interface kernels enforce continuity of flux and continuity of the solution via the penalty method across an interface with the solid velocity computed using the Newmark-Beta method.

Phase field

A set of kernels was added to solve the phase field equations coupled with the Navier-Stokes equations for advection. This is notably used for multiphase flow, as it involves the coupling of phase tracking algorithms with the Navier-Stokes equations. The advantage of using the phase field method for this coupling is that it does not require the explicit calculation of the interface.

2025.09.18 Update

• Limited support for untangling of tangled meshes in VariationalMeshSmoother

• Support for smoothing meshes including pyramids with VariationalMeshSmoother

• Added SparseMatrix::read_coreform_hdf5() method for reading CSR sparse matrix data from an HDF5-based format

• Default to syncing C++ streams for more reliable console output in cases like threaded output

• Made libMesh::out and libMesh::err thread-safe even when emitting non-ASCII (unicode, color codes) output

• Made prepare_for_use() optional after mesh stitching. This will enable much faster parallel generation of complex meshes in certain MOOSE workflows.

• Added optional parameter to build_side_list_from_node_list() to allow only some nodesets to be converted rather than all

• Added Parameters::Value::type_info() method

• Updated autotools to their latest versions. This fixes a linker bug that was preventing libMesh compilation with some newer software stacks.

• Use MatGetValue instead of MatGetRow in PetscMatrix::operator(), to improve compatibility and efficiency

• PerfLog performance logging of Netgen tetrahedralization

• Bug fixes for parallel StaticCondensation operation

• Bug fix for certain use cases of PetscVector constructed directly from PETSc Vec

• Bug fix for triangulation of boundaries with quadratic edges

• Bug fix for an adaptive refinement failure case on distributed meshes

• Fixed handling of linearly-dependent data in reduced basis EIM training

• Fixed small PetscViewer memory leaks in print_matlab() methods