Sciencemadness Discussion Board

DFT calculations at home

Ragbips - 3-10-2012 at 15:33

Is there a software package suitable for performing DFT calculations (preferably with B3LYP exchange-correlation) on my standard crappy home PC?
I have never done these calculations before and would like to know if this is possible. And if so, is it all Linux based? I have been studying these calculations for a while now without having the opportunity to bring it into practice so far.

[Edited on 3-10-2012 by Ragbips]

Polverone - 4-10-2012 at 02:18

Yes, GAMESS-US is free and will do that on a standard PC under Windows. Though of course the size of the systems you can treat will be limited by the hardware. You need to click through an agreement and provide an email address, then you will be able to download a version for Windows: http://www.msg.ameslab.gov/gamess/License_Agreement.html

Firefly is a fast GAMESS-like software that is similarly free and runs on Windows but requires registration: http://classic.chem.msu.su/gran/gamess/index.html

I would recommend GAMESS-US over Firefly just because there are better graphical front-ends for preparing molecules and calculations with GAMESS-US. For example Avogadro looks to have pretty mature GAMESS-US support: http://avogadro.openmolecules.net/wiki/Main_Page

If you have a lot of money to spend there are some other nice options for Windows like Hyperchem, Spartan, and CAChe. I used Hyperchem extensively about 7 years ago. It is very easy to set up and execute calculations, but the downside is that it will make guesses and substitutions without warning if you try to run an ill-defined calculation. Programs like GAMESS will expose more of the problems and won't suggest solutions if you try to set up a calculation improperly, which can be frustrating but at least avoids "magical" behavior.

arsphenamine - 4-10-2012 at 11:20

+1
Caveats:
GAMESS-US is best compiled and run under Linux since the parallelized SMP code is handled with minimal user input, although the compilation setup can be mildly finicky when moving between gfortran revisions. Documentation is verbose but old school -- no indexing, minimal tutorials.

I have yet to get the Windows versions of GAMESS to run parallel computations, or any version of Firefly to do so. For that matter, Firefly doesn't run at all under modern 64-bit Linuces.

Visualizing the molecular orbitals is most easily done using Molden and best done with Jmol as it performs well across all three OS platforms. MacMolPlot may be satisfactory on a Mac but breaks file associations under Windows, and is unstable on a modern 64-bit Linux.

For all its alleged capabilities, I cannot recommend Dalton. Unusual build difficulties and spotty parallelization suggest an absence of professional supervision in its production. "Adult" software does not generate a hardware exception and core dump when it can't find a file.

In short, you can't get performance, capabilities, ease of use, and good docs all in one package unless you pay for it. This is why people pay $$$$ for a Gaussian + GaussView suite license -- there's so much less shit to put up with.

Ragbips - 6-10-2012 at 15:40

I have tried them both now, thanks for the info you both!
I've managed to get an old version of Gaussian (03W) and it seems to work better than GAMESS for now. GAMES came with some installation problems on win7 64bit while Gaussian seems to be doing well.
I've tried a first run on cubane but it came up with an Consistency failure in CalDSu.
This is probably because I tried using B3LYP level of theory on 6-31G(d) basis and these may not combine well, maybe due to guessing on the Harris functional or I should use a denser grid and I have learned make sure that there are grid points on ghost atoms .
My knowledge about DFT is really basic but I've been studying the book Density Functional Theory A Practical Introduction by Sholl and Steckel a few years ago, but I've never managed to do any of the problems and examples.

I'll start over again, and with the help of you guys I'll be able to try the problems myself and get much more understanding of DFT in practice, because doing it only in theory by reading the book is undoable.

Chosen - 8-10-2012 at 03:40

A very nice option is also Dalton or NWChem. There is only one issue with them, you have to compile the source code. But I strongly recommend NWChem. It is really powerful software capable of most frequent used techniques, like DFT, HF, post-HF (MP 2,3,4) CISD, CCSD, also with triples, quadruples... Also time dependent techniques like TD-DFT. It could be compiled in serial or parralell mode. Well, I think, that it has many more features than Gaussian and is absolutely FREE. I also used GAMESS, even on win7 64-bit with no problem at all (of course running with administrative privileges).

If your cubane was simply drawn, it is obvious, that it will not converge (SCF will fail to converge) on 6-31G basis sets and especially when using polarizabilities. You sholud to use more sparse grid and small base, like STO-3G or at most 3-21G. After completing of geometry optimization, you can apply denser grid (ie. the default one) and run again on 3-21G and after this step, using the same grid on 6-31G(d). Alternatively, instead of relaxing the grid, you can optimize using some semiempirical method ie. PM6, and then on the 3021G level using the default grid and then on the 6-31G(d).

Sorry for my poor english.

arsphenamine - 8-10-2012 at 18:46

At B3LYP//631+G(d), GAMESS will resolve cubane geometry in 3-30 minutes
(depending on CPU+memory resources) if you relax the nuclear gradient criterion:

$STATPT OPTTOL=0.0005 $END

Preliminary optimization using the PM6 semi-empirical method provided by MOPAC is a good start.

MOPAC 2012 is available for free under a non-commercial academic license.