7:51:33 AM PDT - Fri, Aug 12th 2016 |
|
Hi,
I have been using the ORCA software package for performing calculations, but I wanted to try out NWChem for some features unavailable to ORCA . I can optimize a structure with very similar geometry to what I get in ORCA, however the frequency calculations are absurdly slow compared to what I am used to in ORCA. The frequency calculations in ORCA take ~10 hours to calculate and in NWChem it takes ~72 hours. This is comparing analytical frequency calculations in NWChem to numerical frequency calculations in ORCA, which I imagined should be much slower.
The size of my system is 52 atoms with one manganese center. I used the default optimization parameters in NWChem and the ORCA criteria are similar (deltaE is 5e-6 for ORCA and 1e-6 for NWChem). I used the Lebedev302 grid size in ORCA under its default paramters ('Grid4'). Both utilized COSMO for MeCN. I tried NWChem without COSMO and it was still slow.
Is there a way to speed up the frequency calculations? I was running it with 12 processors for both. I have tried increasing the memory available to NWChem, but it doesn't appear to utilize any more than approximately 16gb. I set memory to 4000 mb per process with a total allotment for the job of 48gb. I can allocate more resources if there is a way to reasonably speed up the calculations.
This was with NWChem 6.3 as it was already configured on the cluster I use. If there's any more info you need, I can do my best to provide it.
Thanks
|
|