SEARCH
TOOLBOX
LANGUAGES
Forum Menu

Memory Consumption CISD/CCSD

From NWChem

Viewed 91 times, With a total of 3 Posts
Jump to: navigation, search

Just Got Here
Threads 1
Posts 3
Hello everybody.

I have been assigned the task to design a small lab for our students using post-HF methods. The available workstations have only 8 GBytes of memory, so I am first trying to find out which size of computations are possible at all.

I noticed that I can run a CCSD total energy calculation on a methanediol molecule with a cc-pVTZ basis set (165 basis functions) in core (without writing the integrals to disk) using this input:

 title "Methanediol" 
START
scratch_dir /mnt/local/run
permanent_dir /mnt/local/run
memory total 7800 mb noverify
geometry units atomic autosym
O 2.087648 0.000000 0.732536
O -2.087648 0.000000 0.732536
H 1.737757 0.000000 2.483784
H -1.737757 0.000000 2.483784
C 0.000000 0.000000 -1.158285
H 1.618298 0.000000 -2.586301
H -1.618298 0.000000 -2.586301
end
basis
* library cc-pvtz
end
scf
singlet
rhf
end
tce
scf
ccsd
2eorb
2emet 1
end
task tce energy

The calculation uses about 2.5 GByte (RSS) memory according to top.

However, I have not been able to complete a CISD calculation using the cc-pVTZ basis. As far as I understand the documentation the "2eorb" option only works for CC type calculations, in fact using it with CISD leads to an abort once the CI iterations start. Removing "2eorb", i.e.

tce  
scf
cisd
2emet 1
end

also aborts, I assume due to lack of memory. Just to test another size: Using a aug-cc-pvdz basis with 111 basis functions completes without 2eorb and "2emet 1".

With the cc-pvtz basis the only working input so far is "2emet 2" and "2emet 3", e.g.

tce  
scf
cisd
2emet 3
end

which of course leads to some massive disk access, it is scrubbing the disk for about 4 hours. One of the integral files on disk is 12 GByte in size while the other is 2 GByte in size, so using a tmpfs (ramdisk) is no option either.

I have been playing with the "memory" directive but with no luck, there is apparently really not enough available memory.

Still, before giving up on this: can I do something to fit the original cc-pvtz/CISD calculation into memory or to minimize disk access further ? Somehow I would like to pack the 2 GByte temporary file into memory :-) Especially after noticing that CCSD works in-core.


Best Regards

Christof

  • Karol Forum:Admin, Forum:Mod, NWChemDeveloper, bureaucrat, sysop
    Profile
    Send PM
Clicked A Few Times
Threads 0
Posts 8
Hi Christof,
the 2eorb/2emet options have been implemented for the CCSD/EOMCCSD/CCSD(T)/CR-EOMCCSD(T) methods only. Incorporating these changes in the CISD should not cause any problems and will significantly reduce the memory use for the RHF and ROHF cases. What version of nwchem are you using? I will send you the cisd_c1.F, cisd_c2.F, and cisd_e.F files (I will check them in the development tree).
Best regards,
Karol

karol.kowalski@onnl.gov

Just Got Here
Threads 1
Posts 3
Dear Karol,

thank you very much for that unexpected and very generous offer.

This should really save us quite some administrative effort trying to organize reliable overnight compute jobs
in that computer pool and of course it will reduce the students idle time substantially :-) On top, with three
heavy atoms and a reasonable basis set we should be able to do more interesting examples than just the
canonical two-atomic molecules.

Please email the files to c.koehler@bccms.uni-bremen.de (I could not get an email address verified for some reason,
but it should work). I am using 6.1.1 with gfortran/gcc 4.3.3 and no optimized BLAS (might try some later).
I assume I can simply replace the old ones and recompile ?

It would be great if the change would make it into the next release.


Best Regards

Christof

Just Got Here
Threads 1
Posts 3
Dear Karol,

re-reading your reply: it is no problem for me to check them out of the repository either, assuming I have read access
to the branch.


Best Regards

Christof


Forum >> NWChem's corner >> Running NWChem



Who's here now Members 0 Guests 0 Bots/Crawler 1


AWC's: 2.5.10 MediaWiki - Stand Alone Forum Extension
Forum theme style by: AWC