Converge, dammit!

One of the biggest irritations in time dependent DFT calculations is getting your functionals to converge. And a certain large molecule that I’m trying to work some calculations on is just being a little carbonaceous bitch. I’ve run the calculation three times now and it’s still giving me “SCF failed to converge”.

Ok. One more try. I resubmitted the file (after tinkering with it) a couple of hours ago and the output’s looking a little wobbly, but otherwise ok so far. Fingers crossed. I feel like an engineer sometimes actually — some of these things are a case of finding the right values by direct experimentation, or in simpler terms, screwing around with it until it works.

Anyway, here’s hoping it works this time… Converge! Converge, I say!


Current DIIS subspace size = 15
REM_GRID_RESTART_DIIS = 0

50 -1262.0106527154 9.86E-04 00000 Convergence failure

Q-Chem fatal error occurred in module scfman.C, line 1724:

SCF failed to converge

(Incidentally, this is from the first of the output files… If after several attempts, I was still only down as far as 10-4, I’d have gone and kicked it by now. And it’s not a good idea to kick a supercomputer…)

About Invader Xan

Molecular astrophysicist, usually found writing frenziedly, staring at the sky, or drinking mojitos.
This entry was posted in Imported from Livejournal and tagged , . Bookmark the permalink.

10 Responses to Converge, dammit!

  1. invaderxan says:

    I know the feeling. It’s not helpful. In fact it’s even more than less than unhelpful…

  2. dr_psycho says:

    mmmmm…. fatal error
    That’s a phrase i see rather too frequently

  3. invaderxan says:

    Ohhhh right. Now I’m with you. :)
    Actually, I think most consoles work out significantly more powerful than comparably priced computers. They’re just more fiddly to set up!
    Big Bang Theory… Meh. :P

  4. lactam says:

    Renderfarms are large collections of networked computers that are used to render CGI images. Industrial Light and Magic (George Lucas!) coined the term. They had rooms filled with tower cases of ‘pooters, rendering … uhm… let’s say … Episode 3.
    Buying a bunch of cheap P4’s is a bit cheaper than the XBox 360’s. Perhaps for only a short while. They have also done the same with the Playstation 3. The nature of the Cell processor in the Playstation 3 makes me wonder …
    From what I understand, the Cell processor is … uhm… complicated… it’s got something like four or eight cores (four PPE’s, eight SPE’s) … which makes it … well, from what I’ve seen at the local Best Buy, pretty much cheaper than most 1U or 2U stackable computer units.
    I’ll go back to watching Big Bang Theory now….

  5. invaderxan says:

    Blimey. No word with “CON” in it is safe anymore, is it? :P
    Erm… sorry I think there’s some terminology mismatch here. Renderfarms?
    The computing facilities here are pretty much self contained. I’m not really using the main cluster yet — I may be working with some big molecules, but they don’t really warrant that much computing power. Plus, there’s usually a pretty sizeable waiting list to join… The departments who use the cluster the most (mine included) have their own, smaller clusters. It’s an Opteron-based system unless I’m mistaken…
    On a related note to cheap Dell-based clusters, I’m sure I read somewhere about a group who hooked up a bunch of XBox 360s to use as a cluster…

  6. lactam says:

    I tried adjusting the convergence parameters a little
    This made me giggle a bit. Some words have different connotations…
    Off topic a bit … do you get time on a renderfarm, or do you submit your computational jobs to big iron? Up here in the cold (really bloody cold right now) Canadian arctic, there’s some interuniveristy supercomputing alliances (the SHARK network, etc) on top of locals doing Beowulf clusters from cheap-ass Dell computers (we can get P4’s for ~$150 or less… yowza!), but I’m really curious as to whether you run your calcs on the second cousin of Big Blue. :)
    Cheers.

  7. invaderxan says:

    Hehehe… Thanks for the advice. :)
    I tried adjusting the convergence parameters a little, but I don’t really want to mess with it too much. It becomes kind of a tightrope walk between getting it to work and getting it to work accurately.
    My latest adjustments are to the initial guess (which, so I’ve read, may be a good idea with this kind of issue)…

  8. lactam says:

    Density Functional Theory? If memory serves me, then there should be some sort of parameter that you could set, in order to allow the algorithm to have boundaries …
    … hehe… you gotta have boundaries. I like your “little carbonaceous bitch!” more, though.

  9. invaderxan says:

    LOL! :)

  10. Anonymous says:

    cant help you with your code… but i promise to use my new favorite phrase as often as applicable: “little carbonaceous bitch!”

Comments are closed.