Hi everyone!
I've been using the CASTEP code bundled within Materials Studio 8.
I usually perform single point calculations, but I tried to run optimization of a X-ray crystal structure.
I find it surprising because it is taking too long! It has already one week running and it's only completed 7 optimization steps!
Is this normal? The DFT module within MS8 is much faster!!!
It is running with 16 cores.
Thanks in advance!
Well, it of course depends upon the number of atoms you can considered,quality of convergence etc. etc. It's pretty normal that it will take several days or weeks for that matter. You can switch to cluster by the way (you will require linux platform for that). Density functional theory unlike classical molecular dynamics simulation process is indeed time consuming and tedious. But you achieve better results for sure. Cheers!
Dear Monisha Rastogi.
Many thanks for your reply.
I'm running computations In a Windows server. What puzzled me is if I choose a GGA functional (PBE or PW91), In a DFT module, optimization runs smoothly and After a cople of hours is finished, using the same input as the one for CASTEP.
If I use again PW91, this time with CASTEP, it has been a week already and only 7 optimization steps have run. I don't expext the plainwave approach would take so long!
Was I naïve expecting shorter run times?
Hi again,
I have no clue about the system(supercell hopefully) that you have considered. Between did you checked the performance(through task manager)? Make sure that it is showing CPU usage/Memory usage up to 90-100%. That way we can be pretty sure that CASTEP is doing its job and you can wait patiently.
Dear Mauricio,
I'm sorry to hear the calculation is taking so long, but I'm afraid I can't really say why without knowing what system you're trying to simulate. How big is your simulation cell? How many electrons are there?
There are some things you mention in your question that I find a bit confusing. When you say the "DFT module", do you mean DMOL? CASTEP is also a DFT program, not a semi-empirical one. The main fundamental difference between the two programs is the basis set -- CASTEP uses a plane-wave basis set, and DMOL uses a localised basis set. Whilst a plane-wave basis set is numerically efficient and has some nice mathematical properties (a single convergence parameter, periodicity etc), it does mean that if you have large vacuum regions in your simulation cell you will need more basis states (plane-waves), which is not true with a localised basis set.
In any case I'm happy to investigate what's happening if you like; just send me your input files and I'll take a look.
Hope that helps,
Phil Hasnip
(CASTEP developer)
Dear Philip James Hasnip.
Thanks a lot for your reply!
Yes, the DFT module I refer to is DMol3.
And you are totally right. What I wanted to say is the use of pseudopotentials made me expect a fast computation... but the details you give now clarifysome differences between localized basis sets vs plane-wave basis sets.
My mistake about the 'semiempirical' label!
CASTEP is indeed a DFT program.
I am sending the mischievous unit cell as an attachment. Many thanks for your help and attention!
With best regards.
Mauricio Maldonado
Dear Mauricio,
Thanks -- would it be possible to send your CASTEP input files instead of the xsd file? Otherwise I might choose a different cut-off energy, k-point sampling etc. which would defeat the whole point of me looking at it. They should be called something like 3f.cell and 3f.param -- be warned that the ".cell" files are sometimes marked as "hidden" in Windows, so you might need to change the Folder options in order to see them.
All the best,
Phil
Dear Philip
Many thanks for your time and your help!
Unfortunately, I could not locate the files at the moment, and I think it will take some time before I have the opportunity to attempt further computations on those systems.
Thanks, and best regards.
Mauricio
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Graphs