Molcas Forum

lijingbai2009

Member

Registered: 2019-08-28

Posts: 29

doubly excited configurations have lower CASSCF energy

Hi,

Recently I have a problem with the CASSCF state order. Here is the output for my CAS(4,4) 4-state-averaged calculations:

      energy=    -307.495233
      conf/sym  1111     Coeff  Weight
             1  2200  -0.96155 0.92458
             6  2020   0.09257 0.00857
             8  2002   0.10853 0.01178
            10  udud   0.19134 0.03661
            12  0220   0.09813 0.00963
            14  0202   0.08450 0.00714

      printout of CI-coefficients larger than  0.05 for root  2
      energy=    -307.184797
      conf/sym  1111     Coeff  Weight
             6  2020   0.47201 0.22280
             8  2002  -0.38417 0.14758
            12  0220  -0.48786 0.23801
            14  0202   0.37983 0.14427
            15  uudd  -0.49717 0.24718

      printout of CI-coefficients larger than  0.05 for root  3
      energy=    -307.156678
      conf/sym  1111     Coeff  Weight
             2  2ud0  -0.90064 0.81116
             5  u20d  -0.40997 0.16808
            16  u02d   0.09902 0.00980
            19  0ud2   0.10469 0.01096

      printout of CI-coefficients larger than  0.05 for root  4
      energy=    -307.144562
      conf/sym  1111     Coeff  Weight
             3  2u0d  -0.11545 0.01333
             4  u2d0  -0.97572 0.95204
            17  u0d2  -0.12933 0.01673
            18  0u2d  -0.13383 0.01791

You can see the root 2 is contributed by four doubly excited configurations and the energies are:

::    RASSCF root number  1 Total energy:   -307.49523335
::    RASSCF root number  2 Total energy:   -307.18479709
::    RASSCF root number  3 Total energy:   -307.15667820
::    RASSCF root number  4 Total energy:   -307.14456225

However, the CAPST2 energy said:

::    CASPT2 Root  1     Total energy:   -308.51973882
::    CASPT2 Root  2     Total energy:   -308.20166946
::    CASPT2 Root  3     Total energy:   -308.25958877
::    CASPT2 Root  4     Total energy:   -308.26642209

The root 4 is the actual second state.

I understand the CASSCF suffers the state order problem because of missing dynamic correlation. Do these results mean I have this situation? I checked the active orbitals two π and two π* and they look good. I run some tests with CAS(4,3) or RAS(1,1;2:2;1:1) but they didn't help very much. I am just wondering has anyone encounter this before and if there is a way to fix it at the CASSCF level.

Thanks!

Ignacio

Administrator

From: Uppsala

Registered: 2015-11-03

Posts: 1,085

Re: doubly excited configurations have lower CASSCF energy

This is completely normal and happens all the time. The order of states is likely to change between CASSCF and CASPT2.

lijingbai2009

Member

Registered: 2019-08-28

Posts: 29

Re: doubly excited configurations have lower CASSCF energy

Hi Ignacio,

Thanks for your answer. If CASSCF likely has "wrong" order of states, does it mean I have to use CASPT2 for MEP or surface hopping so I can get the right order of states in the beginning? According to my CASPT2 results, the root 4 and root 1 should be S0 and S1 and there will be one intersection between them. But in CASSCF, the root 4 will have intersections to the root 3 and 2 before the root 1. Do I understand correctly? Thanks!

Ignacio

Administrator

From: Uppsala

Registered: 2015-11-03

Posts: 1,085

Re: doubly excited configurations have lower CASSCF energy

It depends on the specific situation, but most likely you'd have to at least take it into account. If it runs out that roots 2 and 3 have no sizable interactions (if there are no hops to these states), you may get away with ignoring them and doing everything at CASSCF level. CASPT2 is not perfect either, so whatever your results, they are dependent on the method you use.

lijingbai2009

Member

Registered: 2019-08-28

Posts: 29

Re: doubly excited configurations have lower CASSCF energy

I want to add an update for my CASSCF state order issue. After checking the configuration, I think CASSCF underestimated the energy of the doubly excited root 2 because of the absence of other lower excited-states. The CASPT2 captured the influence of those missing states, thus corrected the order of states in my 4-state-averaged wavefunction. Then I decided to remove the highest π* from active space to reduce the effect of the doubly excited root 2 as you can see the overall weight of 2002 and 0202 is ~25%. With the new CAS(4,3) active space I can easily average all 6 states without missing important configurations and I finally have the CASSCF state order the same as the CASPT2 order. I hope this trick also helps others.

Besides, the reason I want to modify the active space is I have done TDDFT and EOM-CCSD calculations and they all agreed with CASPT2 results but not CASSCF, which makes me feel really unreasonable if I run surface hopping using CASSCF with the wrong order of states. Now it is good to go!

lijingbai2009

Member

Registered: 2019-08-28

Posts: 29

Re: doubly excited configurations have lower CASSCF energy

To close this topic, I wanted to add a follow-up that eventually solved my problem with doubly excited configuration and inconsistent CASSCF state order compared to CASPT2.

The strong contribution from the doubly excited configuration can be easily removed by taking the highest pi*-orbital out of CAS(4,4).

But CAS(4,3) still suffered state order issues at the CASSCF level if not using 6-state-averaging due to the missing of dynamic correlation. By checking TDDFT results, there is notable excitation from a lower sigma-orbital. After including this sigma-orbital to have CAS(6,4), the CASSCF produced the same state order as CASPT2. Since the sigma-orbital, in general, are not "active" as pi-orbital, so I have to at least use 4-states-averaging to stabilize CAS(6,4).

I hope this could be helpful to everyone.

Last edited by lijingbai2009 (2020-02-26 19:59:29)