[Columbus] MCSCF in COLUMBUS

[Hans Lischka]

Dear Spiridoula,

The Columbus mcscf program stops in an unconventional way, but it is 
correct in stopping. In the mcscfin input the invariance properties are 
specified (FCIORB) and they will show that there are no orbital rotation 
parameters to be optimized. So, there is no mcscf procedure, only the 
CAS orbitals will be resolved, e.g as NOs. I wonder what Gaussian and 
GAMESS did in this case. They should stop mcscf after one iteration, 
isn't it? Could you do gradients and nonadiabatic couplings in this case 
with these two programs?
That Columbus is not able to make a calculation without external 
orbitals is a drawback. Thomas showed the way to circumvent the problem 
by doing CISD (based on a closed shell configuration).
Please try this first together with a gradient calculation, that is simpler.
You wrote: "And the CI part it turns out is not invariant to the 
orbitals even in FCI."
It is not clear to me what you did. You could not do it with Columbus, 
could you?
The orbitals always have to be resolved in invariant subspaces to 
guarantee continuity of the wavefunction for geometrical displacements. 
We have two resolutions, NO and Fock operator. At least gradients are 
independent of this choice, we tried that. I am not sure about the 
derivative couplings.
There is only the CI part of the coupling for the exact eigenfunctions. 
I think the CSF part comes only because of the nonvariational character 
of the MOs in a CI calculation. But here we have the exact result, so I 
wonder what happens in this case with the CSF contribution.

Best regards, Hans

On 3/2/2017 6:37 PM, Spiridoula Matsika wrote:
> Dear Hans and All,
>
> Thank you all for your responses. As some of you suggested there are
> other ways to get the FCI energy. I have done that  using ciudg. As
> Scott just pointed out ciudg will also not work without an external
> space but I have constructed different kinds of CI expansions with a
> smaller active space and double excitations to virtuals. Since there
> are only 2 electrons in the system this still is a FCI calculation.
> And in FCI anyway I do the calculation  I get the same energies and
> gradients. BUT what I want to calculate is the derivative coupling
> and the breakdown into the CI and CSF part of the derivative
> coupling. And the CI part it turns out is not invariant to the
> orbitals even in FCI. I was shocked to see that but then I read more
> carefully the 2004 paper where the f was first implemented in
> Columbus (JCP 120, 7322) and it is explicitly stated there that this
> is the case. So a lot of you understand that much better than I do.
>
> I have succeeded in running the CASSCF as FCI with gamess and
> gaussian. So I was wondering why it will not work in Columbus.
>
> The output is below. As Ron guessed, the program runs the first CI
> iteration and then it stops. The energies are correct even at the
> first iteration.
>
> Thanks, Spiridoula
>
>
>
>
>
>
>> On Mar 2, 2017, at 8:21 AM, Hans Lischka
>> <hans.lischka at univie.ac.at> wrote:
>>
>> Dear Spiridoula,
>>
>> In a FCI there are no orbitals to optimize since the energy is
>> invariant under unitary transformations. The final reason of the
>> breaking is the fact that there are no virtual orbitals, as you
>> suspected. Actually, a simple energy calculation would be
>> sufficient. I am not sure from the error message where the
>> calculations really broke. Did it do so at the very beginning or
>> did go beyond the energy calculation? Could yo send us the mcscfls
>> file from the WORK directory?
>>
>> The alternative is (as Felix wrote) to use the ciudg program and
>> perform the FCI calculation with a corersponding reference space
>> and zero excitations into the virtual (non-existent) space.
>>
>> Ron and Thomas, do you have other suggestions?
>>
>> Best regards, Hans
>>
>> On 3/1/2017 8:46 PM, Spiridoula Matsika wrote:
>>> Dear Columbus friends,
>>>
>>> I am trying to run a benchmark calculation. The system is HeH+
>>> and I am setting up an MCSCF to act as FCI. So the active space
>>> contains all the orbitals in the basis set. But the calculation
>>> always fails. I have tried different versions of Columbus,
>>> different basis sets, etc. but nothing seems to work. The error I
>>> get is:
>>>
>>> *** error *** orthot: na,nb=         0         1 bummer
>>> (fatal):orthot: argument error                      0
>>>
>>>
>>> Is there any reason that the MCSCF in COLUMBUS will not work if
>>> there are no virtual orbitals and all orbitals are included in
>>> the CAS?
>>>
>>> Thank you, Spiridoula Matsika
>>>
>>>
>>>
>>> Spiridoula Matsika Professor of Chemistry Department of
>>> Chemistry Temple University 1901 N.13th Street 242 Beury Hall
>>> Philadelphia, PA 19122
>>>
>>> Phone: 215 204 7703 Fax: 215 204 1532 smatsika at temple.edu
>>> <mailto:smatsika at temple.edu>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>> _______________________________________________ Columbus mailing
>>> list Columbus at lists.osc.edu
>>> https://lists.osu.edu/mailman/listinfo/columbus
>>>
>>
>> _______________________________________________ Columbus mailing
>> list Columbus at lists.osc.edu
>> https://lists.osu.edu/mailman/listinfo/columbus
>