Spin densities from adaptive sampling SCF calculations


I ran adaptive sampling SCF calculations on an open-shell (doublet) system. I can see the partial atomic Mulliken charges. But somehow the spin densities in the next column always remain at 0.0000. This is the case regardless of the # of determinants chosen for the calculation and even when I did AS-CI instead. Could you comment on this? Thanks.

Could you post your input file and the fragment of output you are referring to?


1 2 #charge & multip
Cu -0.497099330 -0.002947368 0.000000000


jobtype sp
exchange b3lyp
basis def2-TZVP
CAS_M_S 1 #This is the number of unpaired e-s
CAS_N_ELEC 17 # I am playing around with this
CAS_N_ORB 44 # I am playing around with this
ASCI_NDETS 50000 # I am playing around with this
ASCI_USE_NAT_ORBS TRUE #The DL paper suggests that this is useful
THRESH 5 #ok for now
max_scf_cycles 400
SCF_CONVERGENCE 5 #ok for now
mem_total 80000
mem_static 40000
symmetry false
sym_ignore true
scf_algorithm diis_gdm


Analysis of ASCI wave function for root 0
Distribution of determinants by excitation level
level Ndets weight cum. weight
0 1 9.09871e-01 9.09871e-01 #looks reasonable based on older casscf calcns with molcas
1 205 1.84499e-03 9.11716e-01
2 12705 8.46220e-02 9.96338e-01
3 7527 5.59334e-04 9.96898e-01
4 29562 3.10239e-03 1.00000e+00

                Orbital Energies (a.u.)   #actually occupation #s

Alpha MOs
– Occupied –
2.0000 2.0000 2.0000 2.0000 2.0000 2.0000 2.0000 2.0000
2.0000 2.0000 1.9829 1.9828 1.9826 1.9826 1.9825 1.9803
1.9687 1.9655 0.9938

      Ground-State Mulliken Net Atomic Charges

 Atom                 Charge (a.u.)    Spin (a.u.)

  1 Cu                    0.903850       0.000000
  2 others                -0.300978       0.000000
  3 others
  4 others

Sum of atomic charges = 1.000000
Sum of spin charges = 0.000000

Thanks, the issue is reproducible with Q-Chem 5.3.2. A quick inspection of the ASCI code revealed that it does not save alpha and beta ASCI one-particle density matrices to the file as expected by the analysis routines (charge analyses as well as multipole moments). Therefore, any properties reported there do not correspond to the ASCI wavefunction and probably are not meaningful at all. Sorry for the misleading output, I am reporting this issue to the developers so that it can be rectified in the future.

Excellent. Thank you.

It doesn’t appear to be an issue for just ASCI. It seems to be an issue for conventional CASSCF calculations as well.

Correct, both ASCI and CASSCF calculation types are executed by the same component, and both are affected by the same issue.