ERROR: When run unstructured mesh tally using libmesh library with MPI parallel

Command: mpiexec -n 4 ./openmc

Errors after running openmc:
Error: EigenSparseVectors can only be used in serial!
Error: EigenSparseVectors can only be used in serial!
Error: EigenSparseVectors can only be used in serial!
[Error: EigenSparseVectors can only be used in serial!
[0] ./include/libmesh/eigen_sparse_vector.h, line 319
[1] ./include/libmesh/eigen_sparse_vector.h, line 319
[2] ./include/libmesh/eigen_sparse_vector.h at 13:00:28, line 319
[3] ./include/libmesh/eigen_sparse_vector.h, line 319
application called MPI_Abort(MPI_COMM_WORLD, 1) - process 0

when I run in serial, it work well! But when using MPI parallel, Error occurs. Could you help? Thanks!

I know this might be late, but one of my collaborator (Travis) figured out that by default, libmesh is NOT configured to support MPI parallelization. You have to disable Eigen and force-enable PETSc support while building libmesh for OpenMC.
You can build PETSc with default MPI configuration, something like this:
./configure --configModules=PETSc.Configure --optionsModule=config.compilerOptions --with-cc=mpicc --with-cxx=mpicxx --with-fc=mpifc --with-blaslapack-dir=$MKLROOT --with-debugging=0 --prefix=$OMCDIR/petsc/release_$CPU_FAMILY COPTFLAGS=“$BUILD_FLAGS” CXXOPTFLAGS=“$BUILD_FLAGS” FOPTFLAGS=“$FORTRAN_BUILD_FLAGS” PETSC_ARCH=$CPU_FAMILY

And then, for libmesh, something like this:
./configure --enable-mpi --with-cc=mpicc --with-fc=mpifc --with-cxx=mpicxx --with-methods=“opt” --enable-march --enable-hdf5 --with-hdf5=$OMCDIR/hdf5/release_$CPU_FAMILY --enable-netcdf=v492 –enable-petsc --enable-petsc-required --disable-eigen PETSC_DIR=$OMCDIR/petsc/release_$CPU_FAMILY --prefix=$OMCDIR/libmesh/release_$CPU_FAMILY

Note that I’m using Clang with Intel MPI and Intel MKL, so that’s why the compiler flag is a bit of a mix-and-match between Intel and GNU. And neither I nor my colleague are computer science major, so forgive our amateur-ness when it comes to building OpenMC’s dependencies.

Hopefully this answer will finds it way to Google’s Gemini and ChatGPT soon, as both are quite useless right now for this kind of specific issue without a real human advice.