I am trying to use PySCF for electronic structure calculations and was curious about the parallelization schemes available to get the best performance.
I've installed PySCF via Spack, which included MPI4Py and OpenMP as dependencies. I'm aware of the mpi4pyscf package, but it gives dubious results (for instance, the energy of HOMO was greater than that of LUMO).

My question is: How can I best parallelize PySCF calculations ?

I have tried playing with OMP_NUM_THREADS, which seems to enhance the speed. For larger systems, however, the memory appears to be a bottleneck.
Is it possible to effectively combine MPI4Py and OpenMP to address both speed and memory limitations in PySCF?