Incomplete support for Gaussian basis set format

[susilehtola]

The following input

molecule {
0 1
Li
Li 1 1.63
}

basis {
spherical
****
Li 1
S 1 1.0
 20.0 1.0
S 1 1.0
 2.0 1.0
****
Li 2
S 1 1.0
 10.0 1.0
S 1 1.0
 1.0 1.0
****
}

energy('scf')

should give me the first basis set on atom 1 and the second basis set on atom 2. What I get instead is

$ psi4 

QcdbException BasisSetNotFound: BasisSet::construct: Unable to find a basis set for atom 1 for key BASIS among:
  Shell Entries: ['LI']
  Basis Sets: [('anonymous8452ca2a', 'anonymous8452ca2a', None)]
  File Path: /tmp/crc, /tmp/crc, /home/work/psi4/install/share/psi4/basis
  Input Blocks: anonymous8452ca2a



Traceback (most recent call last):
  File "/home/work/psi4/install/bin/psi4", line 271, in <module>
    exec(content)
  File "<string>", line 48, in <module>
  File "/home/work/psi4/install/lib/psi4/driver/driver.py", line 493, in energy
    wfn = procedures['energy'][lowername](lowername, molecule=molecule, **kwargs)
  File "/home/work/psi4/install/lib/psi4/driver/procrouting/proc.py", line 2006, in run_scf
    scf_wfn = scf_helper(name, post_scf=False, **kwargs)
  File "/home/work/psi4/install/lib/psi4/driver/procrouting/proc.py", line 1266, in scf_helper
    base_wfn = core.Wavefunction.build(scf_molecule, core.get_global_option('BASIS'))
  File "/home/work/psi4/install/lib/psi4/driver/p4util/python_helpers.py", line 115, in pybuild_wavefunction
    basis = core.BasisSet.build(mol, "ORBITAL", basis)
  File "/home/work/psi4/install/lib/psi4/driver/p4util/python_helpers.py", line 77, in pybuild_basis
    return_atomlist=return_atomlist)
  File "/home/work/psi4/install/lib/psi4/driver/qcdb/libmintsbasisset.py", line 683, in pyconstruct
    verbose=verbose)
  File "/home/work/psi4/install/lib/psi4/driver/qcdb/libmintsbasisset.py", line 902, in construct
    (at + 1, key, text2))

BasisSetNotFound: BasisSet::construct: Unable to find a basis set for atom 1 for key BASIS among:
  Shell Entries: ['LI']
  Basis Sets: [('anonymous8452ca2a', 'anonymous8452ca2a', None)]
  File Path: /tmp/crc, /tmp/crc, /home/work/psi4/install/share/psi4/basis
  Input Blocks: anonymous8452ca2a

[loriab]

Yes, that's expected. While we want to support Gaussian basis format for all the basis set specification, the connection of that basis to the molecule is a separate part, IMO. Integer labeling of atoms is problematic, especially for SAPT and n-body molecules. So all basis sets for Psi use the Fe 0 scheme (that is, generic per element), then you can assign different basis sets to different atoms per element or per indiv atom via label through assign statements in a basis [name] {...} block, as in the mints2 test case.

[susilehtola]

I feel like this should be addressed by checking on line 185 of libmintsbasissetparser.py that the input atom number is always zero.

[susilehtola]

I don't really understand what the Python code is doing

[loriab]

I agree, it'd be nice to catch those and warn ppl off from that feature of GBS format. In the existing logic https://github.com/psi4/psi4/blob/master/psi4/driver/qcdb/libmintsbasissetparser.py#L183-L184 is already catching only <element symbol(s)> 0 through the zero near the end of https://github.com/psi4/psi4/blob/master/psi4/driver/qcdb/libmintsbasissetparser.py#L115.

This would be a general (to all files read), not specific, catch, but if you change that 0 in the regex to \d+ then add the below after https://github.com/psi4/psi4/blob/master/psi4/driver/qcdb/libmintsbasissetparser.py#L184, does that catch and provide better guidance?

gbsint = int(atom_array.match(line).group(2))
if gbsint != 0:
    raise ValidationError("""Gaussian94BasisSetParser::parse: Unsupported basis to atom by number assignment. Use "assign" statements as in https://github.com/psi4/psi4/blob/master/tests/mints2/input.dat : line %d: %s""" % (lineno, line))