[molpro-user] Hard limits on CAS space sizes?

[Dr Seth OLSEN]

Hi Molpro-users,


Thank you, Mariusz, for your reply.  I agree with you regarding the nature of CASSCF calculations.  My intent regarding this calculation was to serve as a guide to the selection of orbitals (and an improved orbital set) for a reduced calculation with a selected (non-CAS) configuration set.  I cannot help but notice that for the molecule I am working with that the most relevant configurations would seem to consist of a subset of those that would be generated from single and double excitations out of a CAS(2,2) -like reference space.  Unfortunately, the molecule is too large to consider numerical differentiation of an MRCI wavefunction built upon CAS(2,2) or I would certainly be taking that route.  

My plan to proceed after this large benchmark was to generate a subset of the CAS space that included all single and double excitations from the 20, 11, and 02 configurations within the HOMO-LUMO subspace. In this way the CAS(2,2) wavefunction would be correlated in a balanced way with the rest of the valence space (and perhaps beyond) without generating heaps of configurations with little weight in the final 2-state averaged wavefunction, or which affect the energy of both states in nearly identical fashion.    Since this would be done via the MULTI utility, I would thereby be able to extract analytic gradients.  

That is, I was hoping to generate a configuration set in multi that looked somewhat like CAS(2,2) times valence MRCI (with all non-active orbitals cored).  

Right now I'm trying to figure out what the best way  would be within molpro to generate this configuration set.  I had thought of doing a CAS(2,2) x MRCI, saving the configurations and then calling this record into a subsequent MULTI calculation (using a SELECT card).  Would this work?

Does anybody on the list have any insights into the best way to go about doing this?

Cheers,

Seth


ccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccms

Dr Seth Olsen, PhD
Postdoctoral Fellow, Computational Systems Biology Group
Centre for Computational Molecular Science
Chemistry Building,
The University of Queensland
Qld 4072, Brisbane, Australia

tel (617) 33653732
fax (617) 33654623
email: s.olsen1 at uq.edu.au
Web: www.ccms.uq.edu.au 

ccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccms




----- Original Message -----
From: Mariusz Sterzel <msterzel at buffalo.edu>
Date: Tuesday, May 24, 2005 11:15 am
Subject: Re: [molpro-user] Hard limits on CAS space sizes?

> Dear Dr. Olsen
> 
> In a past I was working with CASSCF for some time. To my knowledge and
> experience it not possible to run such HUGE calculations even with 
> 30 Gb
> of RAM, actualy RAM is not a problem, but your No. of determinants is.
> Honestly speaking smaler active space is much better then big one. I
> suggest you examine your active space again and choose smaller one. 
> I was
> working with molecules which has 400 basis functions and more and 
> it was
> allways possible to choose active space which does not exceed 14 
> orbitalsand 14 electrons. Try TDDFT first, print occ. No. with 
> better accuracy
> then 10^{-2} and look at orbitals with bigger ones. Or test small 
> activespaces on CASPT2 level and again look al the occupation No.
> 
> I do not know what type is your computer, but as I remember some 
> time ago
> someone posted that for 32 bit ones active space is limited to 16
> orbitals - check mailing list.
> 
> And again CAS is not a black box, it allways requires patience and 
> care.
> 
> Mariusz
> 
> --
> Mariusz Sterzel
> Email: msterzel/^at^\buffalo.edu
> 
> 
> 
> On Tue, 24 May 2005, Dr Seth OLSEN wrote:
> 
> >
> > Hi Molpro-users,
> >
> > I'm doing CASSCF on a molecule with 274 contracted basis 
> functions and 126 electrons.  I would like to perform a CAS(18,18) 
> calculation on this molecule, averaged over the two lowest states.  
> I'm working on an SMP box with 8 processors and 30GB of ram.  So 
> far I have not had much luck with the calculation, which quits very 
> soon after calling MULTI, before the calculation apparently starts 
> in earnest.  Am I running into a hard-coded limit in molpro?  If 
> so, can I recompile and get around the limit?
> >
> >
> > The error I get reads:
> > ERROR EXIT
> > CURRENT STACK:      MAIN
> >
> > Which is not very descriptive.  I appear to have plenty of 
> scratch space, but this would not appear to be the problem anyway 
> b/c I get the same error running direct.  I note that the error is 
> not your usual 'not enough memory error' either.  This makes me 
> think that there may be a deeper issue.  Am I correct in this?  Is 
> this calculation out of reach?
> >
> > Cheers,
> >
> > Seth
> >
> >
> >
> >
> > ccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccms
> >
> > Dr Seth Olsen, PhD
> > Postdoctoral Fellow, Computational Systems Biology Group
> > Centre for Computational Molecular Science
> > Chemistry Building,
> > The University of Queensland
> > Qld 4072, Brisbane, Australia
> >
> > tel (617) 33653732
> > fax (617) 33654623
> > email: s.olsen1 at uq.edu.au
> > Web: www.ccms.uq.edu.au
> >
> > ccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccmsccms
> >
> >
> >
> >
> >
> >
> >
> >
>