MOOSE Newsletter (April 2025)

MFEM backend support

MOOSE now has base support for finite element computations using MFEM. This support was developed by the United Kingdom Atomic Energy Authority (UKAEA), originally as part of their Platypus application, but that support has now been merged upstream into MOOSE. Those wishing to get familiar with use of the MFEM backend can look at test cases as well as the class documentation under the MFEM header on our source index page.

capabilities parameter in tests specs

The MOOSE test harness provides numerous parameters to skip tests depending on capabilities provided by the current executable. We have checks for the petsc_version, required_submodules, the configured ad_size etc. Those checks all require custom python functions that parse configuration files and probe the source tree.

The new capabilities parameter combines all the test harness parameters that probe built-in functionality of the executable, and provides a flexible boolean expression parser to construct abitrary conditions that test execution depends on. Capabilities now relies on an application internal registry that can be queried with the --show-capabilities option. For example

min_ad_size = 150
method = '!dbg'

is now

capabilities = 'ad_size>=150 & method!=dbg'

Application and module developers can register their own capabilities using the addCapability function. Check MooseApp::registerCapabilities() for examples.

Parsed-expression generation of multiple subdomain ids

The ParsedSubdomainMeshGenerator is able to specify the extent of a single subdomain on the mesh using a parsed expression, able to represent the combinatorial logical expression defining its extent. With the ParsedSubdomainIDsGenerator, we can now evaluate a parsed expression on each element and use the integer result of this evaluation to assign a new subdomain id to the element. This added flexibility facilitates distinguishing multiple elements of a lattice a posteriori, after creating the lattice.

Material property outputs system rework

The material property outputs system was reworked to perform consistently using both specific Outputs objects and keywords such as all or none specified in the outputs parameter of Materials objects. Several issues were fixed at that time.

Neumann boundary conditions for linear finite volume discretization

Neumann boundary conditions, used to specify a diffusive heat flux for example, have been added to the linear finite volume discretization. This is part of an effort to support conjugate heat transfer simulations in the Navier Stokes module.

Navier Stokes

Users may now specify separator sidesets on variables. These separator sidesets are boundaries that prevent flow, and decouple equations between the two sides of the boundary. This is helpful for coarse mesh multi-dimensional thermal hydraulic simulations.

The DittusBoelterFunctorMaterial was added to compute the Dittus-Boelter correlation for the wall heat transfer coefficient.

Porous flow

Spatially varying initial properties were added to the Kozeny Carman materials. This will serve to model ground fractures with spatially varying apertures.

Subchannel

The examples in the subchannel module have been reorganized into three categories:

• Verification: Inputs compared between subchannel simulations and either analytical solutions or other subchannel codes,

• Validation: Inputs of simulations compared to experimental data, and

• Examples: Inputs showcasing useful techniques in subchannel, unrelated to verification and validation (V&V) efforts. effort.

Reactor

The HexagonalGridPositions was extended to allow the nesting of another positions object within the hexagonal grid. This allows users to obtain all pin positions in a hexagonal core with multiple hexagonal pin lattices within a single object.

2025.04.02 Update

• Set nullspace for both J and P operators because PETSc may look at either when setting up linear solves

• No longer need to manually set hypre memory location for PETSc 3.23. This was previously required for MOOSE stack builds with GPU enabled because PETSc's hook into hypre assumed that any configuration enabling GPU support should always run with data structures on device

• Just use DIFFERENT_NONZERO_PATTERN in MatCopy in PetscMatrix::operator=. This is much more robust than using SAME_NONZERO_PATTERN

• Assert verify calls where appropriate

• Change API name SparseMatrix::reset_memory to SparseMatrix::restore_original_nonzero_pattern. The new API name better reflects both the focus of the API and also that this API may not perform any memory allocation/deallocation as long as previous fills of the matrix have matched the original preallocation

• Better error message from System::read_serialized_vectors() when encountering a difference between the number of vectors we expect to read vs. the number of vectors declared in the XDR header

• Lock at PetscMatrix level when doing add/set. This will allow user code to remove their mutex locks around matrix assembly and also saves users who did not previously have proper mutex locking in threaded regions of their code

• Add mutex to fparser setEpsilon to prevent thread races

• Add QBase::get_base_order() API allowing comparison of multiple quadrature rules independent of p-refinement level

• Update TIMPI submodule

  • Use MPI-4 APIs for MPI_Count, i.e. 64-bit, size support, when available

  • Fix for uninitialized warning with gcc 13.3

• Remove phony make code which could trigger multiple netgen builds

• Get metaphysicl config include

• Allow instantiation of DenseVector::print with dual numbers

• CheckpointIO can now handle some of the Mesh caches if needed. MOOSE often skips partitioning when doing checkpoint restart, so now at the libMesh level we can do recalculate_n_partitions() and update_post_partitioning() when that is the case.

• Add begin/end members to DenseVector allowing for range based looping

• Avoid creating temporary array for hashword2() overload taking 64-bit inputs

2025.04.17 Update

• Autoconf-submodule (ACSM) update

  • When recent clang versions are detected, the compiler flag -fno-assume-unique-vtables is used. This enables dynamic_cast to work reliably on macOS 15 (Sequoia) environments, which fixes a number of libMesh and MOOSE failures there.

  • More thorough detection for code coverage support when –enable-coverage builds may require it from multiple languages' compilers.

• TIMPI submodule update

  • Fixes and tests for timpi_version.h utilities

  • Unit test fixes for C++14 compilers

  • timpi_pure attribute for pure functions; on C++17 and newer compilers this prevents ignoring the function return values

  • Marked verify() and semiverify() as pure. This prevents the common usage error of not asserting the result when intended.

  • Removed default Attributes values. Code which attempts parallel reductions on user-defined types without first specifying their attributes now fails to compile, instead of compiling but not working correctly.

• Added support for creating and computing on arbitrary polygons in 2D

• Better tolerance computations when stitching meshes along a boundary with discrete NodeElem elements

• Fix for reading of Gmsh files with physicals of different types but the same id numbers

• Migrated more print methods of templated classes into headers, to enable user instantiations with unanticipated types

• Added "-V" option to the meshtool utility app to apply the VariationalSmoother to the input mesh

• Moved demangle() to libmesh_common.h. Any failures in cast_ptr or cast_ref in dbg or devel modes are now reported using demangled typenames.

• Fix for return values of FE::n_quadrature_points(): after using a user-specified list of quadrature points rather than a quadrature-rule-generated list, the length of the specified list is now correctly returned. This fixes undefined behavior in JumpErrorEstimator subclasses with the integrate_slits option enabled, which fixes a valgrind-detected error and a potential test failure in libMesh unit tests.

• Fix for undefined behavior in SimplexRefiner. This similarly fixes a libMesh unit test.

• Fix for a Netgen configuration error: configuring the same Netgen-enabled libMesh build directory for a second time no longer leaves Netgen in an "unbuilt, but marked as built" state.

• Refactoring for simplicity and forward compatibility in Gaussian quadrature code

• More assertions

2025.04.22 Update

• Add option to variational smoother to preserve subdomain boundaries, and add test for new option.

• Drop LOG_SCOPE calls from DofMap::dof_indices()

• Update citations

• Update traffic

• Fixes for VTK 9.4

  • Add check for failures to find VTK macros

  • Detect new VTK macros from new VTK header file

2025.04.02 Update

We updated PETSc to version 3.23. For a summary of changes please see the PETSc release page.

2025.04.22 Update to the moose-dev Package

The moose-dev conda package was updated in order to address MOOSE build issues encountered after the recent MacOS update to Sequoia 15.4. This required a number of dependency package updates and changes. Context for this issue can be found in idaholab/moose#30227, and descriptions of the compiler and dependency package updates can be found below.