Slow SCF convergence?

It looks like SCF convergence is slower in the current version of PySCF than in previous releases.

For instance, the example calculation on water

from pyscf import gto, scf

mol = gto.Mole()
mol.verbose = 4
mol.atom = '''
O 0 0      0
H 0 -2.757 2.587
H 0  2.757 2.587'''
mol.basis = 'ccpvdz'
mol.symmetry = 1
mol.build()

for guess in ['1e', 'atom', 'minao', 'huckel']:
    mf = scf.RHF(mol)
    mf.init_guess = guess
    mf.kernel()

takes

• 29 iterations to converge from the core guess
• 27 iterations to converge from the atom guess
• 17 iterations to converge from the minao guess
• 26 iterations to converge from the huckel guess

Water in cc-pVDZ should converge in 6-8 iterations. What’s happened?!

[susilehtola]

The same geometry is used in many other examples, as well.

dft/02-gks.py gto/10-atom_info.py gto/11-basis_info.py gto/12-serialization.py scf/01-h2o.py scf/02-ghf.py scf/02-rohf_uhf.py scf/03-level_shift.py

I’m not sure I’d call it H2O when the OH distance is 3.78 Å(!)

[susilehtola]

The odd thing is that the calculations in #1000 don’t work anymore, even though they should work…