CPU backend not found when trying to run solve_sci

[nonhermitian]

Following the examples the SBD available backends is listed as

SBD Device Information
============================================================
✓ GPU Available: CUDA
  GPU Count: 1
✓ CPU Available: Always
============================================================

but when running solve_sci I get:

RuntimeError: Backend 'cpu' not available. Available: []

indicating that I do not have any available backends

[nonhermitian]

I see that I must build the libs myself even when doing pip install. Note that on Ubuntu I had to change the mpicc config checks to

compile_flags = subprocess.check_output(['mpicc', '-compile-info'],
                                                universal_newlines=True).strip().split()
link_flags = subprocess.check_output(['mpicc', '-link-info'],
                                             universal_newlines=True).strip().split()

for it to work

[hfwen0502]

@nonhermitian thanks for the feedback. yes, the setup would only work for OpenMPI. Since you are using MPICH, that's why it breaks. The new release will provide a more robust solution to support different MPI implementation. In the meantime, you can either create a conda env, forcing to use OpenMPI.

conda create -n sbd python=3.12 pybind11 numpy setuptools wheel pyscf pip mpi4py openmpi openblas -c conda-forge
conda activate sbd
pip install sbd-eigensolver qiskit-addon-sqd

Or you can set MPI_HOME environment variable that points to your MPI install before you run pip install sbd-eigensolver. It will automatically pick up $MPI_HOME/bin/mpicc.

FYI @garrison