[molpro-user] memory problems in UCCSD-F12

[Toru Shiozaki]

Dear Nino,
this is a rather large calculation and unfortunately requires a large amount of memory.
The error message is reasonable as the memory is not allocated at once.
Perhaps this is the second allocation of a 237MW array (300-237=63mw).

I have tested this input in my environment and with 1500mw per process I got the following error message

>  For minimal I/O algorithm in triples, increase memory by 210.05 Mwords to 1710.33 Mwords.
>  For full I/O caching in triples, increase memory by 2621.47 Mwords to 4121.75 Mwords.
>  A further 2.05 Mwords of memory are needed for the triples to run. Increase memory to 1502.33 Mwords.

Certainly with 1505mw, it at least went through the point it failed.

Hope this helps your customer. More might follow from experts of the CC codes.
Best wishes,
Toru

On Apr 12, 2011, at 6:58 PM, Nino Runeberg wrote:

> Dear molpro-users,
> 
> one of our customers is having problems with memory allocation
> in his uccsd(t)-f12 calculation using molpro.2010.1 pl20 built
> with the tcgmsg-mpi/openib version of ga-5-0-2.
> If we allocate 200 MW per process (memory,200,M)
> the job is terminated immediately after entering the
> UCCSD-F12 part with the message:
> 
> CCSD(T)     terms to be evaluated (factor= 1.000)
> Number of core orbitals:          10 (  10 )
> Number of closed-shell orbitals:  38 (  38 )
> Number of active  orbitals:        1 (   1 )
> Number of external orbitals:     395 ( 395 )
> insufficient memory available - require                 237314025  have
>                199629925
> the request was for real words
> GLOBAL ERROR fehler on processor   0
> 
> The error correctly reports the request of 200 MW and suggests that an
> increase to 300 MW should be sufficient. However, with "memory,300,M"
> an otherwise identical job is terminated at the same point with the
> message:
> 
> Number of external orbitals:     395 ( 395 )
> insufficient memory available - require                 237314025  have
>                 62315900
> the request was for real words
> 
> In this job the memory requirement remains identical but, for some reason,
> the requested 300 MW is now seen as only 63 MW. Any ideas?
> 
> I have attached both an input example and the CONFIG for this build.
> Best regards,
>                        -Nino
> 
> ---
> ---------------------------------------------------------------------------------------------
> Nino Runeberg, PhD, Application Scientist, Science Support/Chemistry,CSC
> P.O. Box 405 02101 Espoo, Finland   tel +358 94572733, fax +358 94572302
> mobile +358 50 381 9721, e-mail: runeberg at csc.fi
> CSC - IT Center for Science, www.csc.fi/chem/
> ---------------------------------------------------------------------------------------------
> 
> <CONFIG><test_300.inp>_______________________________________________
> Molpro-user mailing list
> Molpro-user at molpro.net
> http://www.molpro.net/mailman/listinfo/molpro-user