Sciencemadness Discussion Board

Anyone have access to a computer cluster?

smaerd - 2-7-2015 at 19:07

I'll keep it brief.

Does anyone have access to a computer cluster? :) Edit- Or a wicked hot-iron home computer/server with GAMESS on it :P.

I have a few computational experiments I am keen on running but they exceed my hard-drive space and take days on my computer. I couldn't even do the geometry optimization on the one molecule. It ate up my diskspace almost completely.

edit -
I guess in return I'll end up sharing one of my ideas which has some pretty solid experimental evidence behind it. Which ya is pretty meaningless but its a cool project with potential real world applications etc... All it really lacks is a computation FT-IR or H1-NMR. Well ideally I'd have a 2-D experimental NMR (TOECSY, ROESY, NOESY, etc) but I wasn't able to accomplish those in the time-frame that I did my project in due to having a limited instrument. I don't think I'll be publishing the idea even though I could. It might be more badass to make it public or something.

[Edited on 3-7-2015 by smaerd]

careysub - 3-7-2015 at 07:36

I have an 8 core, 4 GHZ, 32 GB RAM data crunching machine running Spark currently. How much storage did you eat up?

Zombie - 3-7-2015 at 07:53

I have a similar set up, and am willing to volunteer time to do what you need. I have a few 1TB hard drives that are empty, and it takes a second to install them.

A cold beer is payment enough. I'll collect that if your ever in carrabelle.

smaerd - 3-7-2015 at 14:37

Doing a geometry optimization I ate up 200+GB which made the calculation impossible. I'm not sure if RAM was an issue it looked like it was okay with 250k words. Keep in mind geometry optimization is just the first thing so I can get a vibrational analysis. I'm new to GAMESS in general, so I may need some input on good basis set's and configurations etc.


Aye Careysub now that's a beastly computer! Zombie I'm afraid the data storage may end up in the TB range. I don't know though, I think there is a way to get GAMESS to predict that but I am new. So not sure.

Right now I'm in contact with a member who has access to a cluster. If they are allowed, and can run the file then it'll be solved. If not I am going to have to take either of you up on the offer. Either way I think I'm going to make the project public.

Just because I'd like people to build off of the idea, and I'd love some face-to-face criticism of my work. There are a few things I was uncertain of.

Also I wanted to thank everyone who PM'd me offering advice and other things. It means a lot. I want to put the project to rest but my mind will always look back and wonder unless I can get a little bit more information. Which otherwise I may never have the opportunity too. Also Dr. Bob helped supply me with some things that made it possible, so it would only be right to post it here.

[Edited on 3-7-2015 by smaerd]

Polverone - 3-7-2015 at 15:47

I am posting here because smaerd said it was ok to discuss in the public thread. He is looking at a cyclodextrin derivative.

The disk space issue is caused by intermediate integrals taking up too much space. You can get around that by running a direct calculation that recomputes integrals as needed instead of storing them, but that will be slower and this is already going to be quite slow even on an 8 core machine.

The desire is to do geometry optimization and then a vibrational analysis, to predict IR spectrum, possibly followed by NMR shielding prediction.

A cyclodextrin is a fairly large molecule by the standards of ab initio quantum chemistry. Predictions made by electronic structure methods are not directly comparable to typical condensed phase conditions at the laboratory bench. Computational chemistry assumes, by default, an isolated molecule in vacuum. You can run calculations with implicit solvation to partially correct the problem, but it's only a partial fix and it increases the computational cost. You also need to account for electron correlation to produce reasonably accurate geometries. That means a rather more expensive calculation than simple Hartree-Fock if you stick with ab initio methods.

There is a recent Nature Protocols paper, which you can find in References, guiding researchers how to produce computational NMR predictions that are actually useful at the bench level:

The steps can be mimicked with free software instead of the expensive commercial products they recommend, but it will be a fair bit of work for a molecule this size. You need to find multiple conformers and run property calculations on each of them.

I have recommended to smaerd that he obtain MOPAC2012, since it is free for academic use:

The semi-empirical geometry optimization and vibrational analysis with PM7 or PM6 will run much faster than ab initio calculations with GAMESS. Those calculations can run with implicit solvation via COSMO if desired. Structures from MOPAC optimization can be used for NMR shielding calculations in GAMESS if enough computing power is available for that kind of job.

paulr1234 - 3-7-2015 at 23:12

Scarily smart people on this forum. I was happy to correctly calculate the density of sulfur in my Chem 101 class last night!

Zombie - 4-7-2015 at 02:12

I understood "sulfur".

The bottom right corner shows what I have. I can set the computations to run thru the graphics card at a much faster rate. (similar to Bit Coin mining).

If you need help... It's here.

Untitled.jpg - 477kB

Northern - 14-10-2015 at 03:54

Get an Amazon web services account and purchase computer time in units as needed. For data storage they can handle literally anything essentially anything whether you want a relation db (*think sql), something like Mongo that uses a NoSQL key-value pair engine, or just flat file storage which is provided by there S3 service.

I swear i dont work for them but by god have i been using the hell out of their free tier since its well free and I have some spare time to play with all of there toys.

smaerd - 16-10-2015 at 03:11

Thanks Northern,

Hmmm. I'll have to see if they are capable of handling what I need. By what you describe it doesn't sound like they would be able too.

Recently I have migrated to a larger university with a few clusters hanging around. I now know someone who can introduce me to someone whose bread and butter is gaussian and other types of calculations. So I'm going to try that out.