Multiconfigurational trial wave functions from GAMESS-US

General discussion of the Cambridge quantum Monte Carlo code CASINO; how to install and setup; how to use it; what it does; applications.
Post Reply
Andrew Powell
Posts: 4
Joined: Wed Jun 26, 2013 5:31 pm

Multiconfigurational trial wave functions from GAMESS-US

Post by Andrew Powell »

I found the directory in CASINO for the GAMESS example, and I followed the author's examples for N2 in which he did a VMC for a HF trial wave function and another VMC for a MCSCF wave function. The VMC energy that I received using the HF wave function was consistent with his, but the VMC for the MCSCF was not.
According to the README file, he got the following for the MCSCF calculations:

GAMESS gives: -19.4742 a.u.
CASINO gives: -19.470(3) a.u.

When I used his files, I got -19.377(2), which is similar to the VMC HF energy of -19.379(2). Since these two energies were similar, I was wondering why there is such a large gap between his VMC energy for MCSCF and mine. I didn't know if this was due to a bug in CASINO, or if something had changed in the newer versions of it. I sent Mr. Badinski an e-mail, but haven't received a reply.
The computing cluster I use has CASINO v2.12.1 (7 March 2013) and the latest version of GAMESS-US.
Thank you for your consideration.

Regards,
Andrew
Mike Towler
Posts: 239
Joined: Thu May 30, 2013 11:03 pm
Location: Florence
Contact:

Re: Multiconfigurational trial wave functions from GAMESS-US

Post by Mike Towler »

Dear Andrew,

I am not immediately sure of the answer to your question. However, I am currently conducting a review into the status of all the Gaussian basis set codes that CASINO supposedly supports (with the help of Mike Deible and Albert Defusco in Pittsburgh) and I will make sure this issue is thoroughly investigated.

Also you should know that Albert is currently implementing native support for CASINO which will be incorporated directly into the latest version of GAMESS. I understand from him that this will be completed soon, in which case we no longer have to reply on the (bless Alex) rather scrappy converter that is currently in the CASINO distribution (though this will probably be improved and retained so that we support older versions of GAMESS)..

Alex is long gone from QMC I'm afraid - we'll have to do this one on our own!

Mike
Andrew Powell
Posts: 4
Joined: Wed Jun 26, 2013 5:31 pm

Re: Multiconfigurational trial wave functions from GAMESS-US

Post by Andrew Powell »

Dear Mike,

Thank you for your prompt response and attention.

Regards,
Andrew
Andrew Powell
Posts: 4
Joined: Wed Jun 26, 2013 5:31 pm

Re: Multiconfigurational trial wave functions from GAMESS-US

Post by Andrew Powell »

As a follow-up question, I was wondering if there is any evidence of an atomic calculation successfully using multiconfigurational wave functions, either with or without Gaussian basis sets. I am curious to see if the latest version of CASINO is reading these particular wave functions correctly, or if the issue is just with the Gaussian basis sets. One of my goals with GAMESS was to create MCSCF wave functions for atoms, and compare my calculated energies with those reported in the 2011 paper by P. Seth, P. Lopez Rios, and R. J. Needs, published in the Journal of Chemical Physics (http://scitation.aip.org/content/aip/jo ... /1.3554625). Thank you for your consideration and patience.

Regards,
Andrew
Mdeible
Posts: 13
Joined: Tue Jun 11, 2013 2:46 pm

Re: Multiconfigurational trial wave functions from GAMESS-US

Post by Mdeible »

Hi Andrew,

I have had this same issue in the past. CASINO seems to give the HF energy for a multideterminant wave function unless there is an MDET block included in the correlation.data file. An example for the format of the MDET block is in the manual. All though the output from CASINO seems to recognize that it is a multideterminant wave function from the gwfn.data file, it does not appear to consider the other determinants unless it also finds them in the correlation.data file. There is no correlation.data file in the GAMESS directory, so using those files alone probably won't reproduce the MCSCF energy.

I think Dr. Towler could verify this, but this has always seemed to be the problem for me in the past.

Mike
Andrew Powell
Posts: 4
Joined: Wed Jun 26, 2013 5:31 pm

Re: Multiconfigurational trial wave functions from GAMESS-US

Post by Andrew Powell »

Hi Mike,

Thanks for your advice about using the MDET block for the N2 example. I did as you suggested and saw a qualitative lowering in energy so that I now get a value similar to the author's. Now, I am testing this with other systems.

Andrew
Mike Towler
Posts: 239
Joined: Thu May 30, 2013 11:03 pm
Location: Florence
Contact:

Re: Multiconfigurational trial wave functions from GAMESS-US

Post by Mike Towler »

Hi guys,

Sorry for the delay in responding. I was producing the new CASINO web site, which took more effort than I thought it would. See http://vallico.net/casinoqmc.

Basically, as you've seen, multideterminant information can appear in both the xwfn.data file and in the correlation.data file.

It's in xwfn.data when the trial wave function is produced by a multideterminant method such as MCSCF (the xwfn.data file represents the wave function exactly as computed by the external code). This defines the reference configuration.

You include the multideteminant information in correlation.data when you wish to modify the reference configuration produced by the external program. Let's say the trial wave function is a single HF determinant. You can then create an excited state, or a multideterminant expansion, using the HF state as a reference (and the coefficients in front of the determinants can be optimized in a variance minimization or whatever). Remember that correlation.data is supposed to contain all optimizable parameters - the stuff in xwfn.data cannot be optimized.

That said, I can confirm that there currently appears to be some problem with this. The multideterminant data in the GAMESS MCSCF example is being read and printed to the output file, but the energy appears to be the same as the HF case - as Andrew says. This is probably a bug, which I will investigate this evening.

I note in passing that (1) the input files in the GAMESS MCSCF_demo example are obsolete, and that (2) Alex has modified the pseudopotential files in such a way that modern CASINO will no longer read them. I have just fixed both these issues in the current beta distribution.

More later.

Cheers,
Mike
Mike Towler
Posts: 239
Joined: Thu May 30, 2013 11:03 pm
Location: Florence
Contact:

Re: Multiconfigurational trial wave functions from GAMESS-US

Post by Mike Towler »

OK - I see the problem - it's a CASINO bug introduced at the same time as 'alternative sampling' a year or so ago. Sorry about that!

The fix isn't immediately trivial, in the sense that I need to think about it, and I have to leave work now. Hopefully I'll get some time to sort it out later tonight.

Cheers,
Mike
Mike Towler
Posts: 239
Joined: Thu May 30, 2013 11:03 pm
Location: Florence
Contact:

Re: Multiconfigurational trial wave functions from GAMESS-US

Post by Mike Towler »

Hi guys,

OK - I've now fixed this problem. The nature of the error meant that - for any multideterminant reference configuration in xwfn.data - only the first deteminant would be used by CASINO, and that's why you were getting the same answer as the VMC-HF run for the single determinant wave function. As I said in the previous post, the cockup was introduced in May 2012 when someone introduced 'alternative sampling' for VMC.

That was a serious error which I'm surprised hasn't been picked up before. Many thanks to Andrew for pointing it out! (and many apologies to anyone who has been inconvenienced by this).

The fix is in the latest current beta, which you can download from the new website here: http://vallico.net/casinoqmc/update-casino/

Cheers,
Mike
Post Reply