Molcas Forum

vincenzo

Member

Registered: 2018-09-26

Posts: 2

rassi calculation

Running a Rassi calculation for computing oscillator strengths between a RASSCF ground state and a number of RASSCF excited states I get the following error message

 RASSI CANNOT CONTINUE. THE PROBLEM AT HAND
  IS PROBABLY NOT SOLUBLE. THE TWO ORBITAL
  SPACES ARE TOO DISSIMILAR.
  LU PARTITIONING IS TROUBLESOME. DIAGONAL
  ELEMENT NR.                    2  IS TOO SMALL:
  IN MATRIX CXA IT IS   1.0000000000000000
  IN MATRIX CYB IT IS   4.1641815018101051E-017
  EVEN AFTER OPTIMAL PIVOT-TRANSFORMATION.

In reality the orbitals of the ground state and the orbitals of the excited states are pretty similar, but in a symmetry there is a switch between two occupied orbitals that are almost orthogonal. In the overlap matrix there are then diagonal terms extremely small and off diagonal terms close to 1.

Any hint on how to solve this problem? Is it possible to reorder the orbitals before they are printed out and saved on the binary file at the end of a RASSCF calculation?

Thanks

Vincenzo

Ignacio

Administrator

From: Uppsala

Registered: 2015-11-03

Posts: 1,085

Re: rassi calculation

Do these orbitals belong to the same space (inactive, RAS1, RAS2, RAS3)? If so, I'd say the "optimal pivot-transformation" should be finding the correct order, and if it doesn't it's either some bug or that the problem refers to another orbital. If they don't, you could try to find a common active space by changing the number of states in CIRoot.

vincenzo

Member

Registered: 2018-09-26

Posts: 2

Re: rassi calculation

Thanks Ignacio for the prompt reply.
No, the two orbitals belong to the same symmetry but to two different spaces (inactive and RAS respectively), that is why, I suppose, the pivot-transformation doesn't work. This is a core excitation calculation where, the core orbital must be active (single occupancy) while the other orbitals  of the same symmetry, are kept inactive. In order to have the core orbital after the inactive ones,  I perform an orbital reordering and the core orbital in the RAS space is then assigned to a supersymmetry to avoid variational collapse. May be there is a different way of doing it, but I don't see, right away, how to avoid the variational collapse (i.e. the "transfer" of the hole from the core orbital to a valence orbital) while keeping the two orbitals of the same symmetry in the same space.
In absence of an explicit answer to my second question, I must suppose there is no way of reordering the two orbitals from the excited state calculation before running RASSI, correct?

Thanks

Vincenzo

Ignacio

Administrator

From: Uppsala

Registered: 2015-11-03

Posts: 1,085

Re: rassi calculation

No, just reordering doesn't help, because you'd need to change the CI coefficients accordingly, and that cannot be done when you exchange orbitals from different spaces. It is precisely for this kind of calculation that the HEXS keyword was introduced in RASSCF, I believe, you could give it a try.