Molecular Dynamics Simulation of Graphene on Rescale | Hacker News Reader

Molecular Dynamics Simulation of Graphene on Rescale

blog.rescale.com

21 pointsgpoort7 comments

Cool demonstration of Rescale; I hadn't come across them before. It's an interesting market, and certainly an underserved one (I say this as someone who's idly considered running similar calculations on high-end AWS instances and been impressed with the cost but less-so with the time it'd take to set it all up). I imagine one of the difficult things for them is getting access to a lot of the academic codes in widespread use at the moment, considering a lot are under fickle and/or expensive licenses. I don't see a DFT[1] code on your tools list at the moment; perhaps Quantum Espresso[2] (GPL-licensed) might be a candidate if you wanted to try one.

Regarding the topic of the blog post, it's a frequent source of frustration to me that a lot of these scientific tools are so obtuse to use. I imagine the set of people who could understand the physics behind them and might want to use (or just play with) them (e.g. the author of this blog post) is much larger than the set of people who are actually capable of doing so. A good proportion of my day is spent solving problems that are toolchain related rather than science related and it really is immensely frustrating.

[1] http://en.wikipedia.org/wiki/Density_functional_theory

[2] http://www.quantum-espresso.org

Hey Osmium, this is John from Rescale. Thanks for the code(Quantum Espresso) suggestion. We'll make sure to add it to our software library soon!

I actually use Quantum Espresso rather extensively in my own research. It should work quite well with an interface like this. It has a sub-program pp.x specifically designed to make common plots out of the programs output.

I can tell you that it tends to be a little slow compared to some other codes such as VASP, and it does tend to take up a good deal more disk space. It has a few other quirks but it is a great program. Also its wavefunctions can be exported to XML files.

If you are looking anytime soon for people with computational physics/chemistry specific knowledge I may be looking for work soon.

> Also its wavefunctions can be exported to XML files.

Surely they'd be huge though? In any case, it's always interesting to hear experiences about other codes. I haven't actually used Quantum Espresso myself, but I know it's Free, so had been considering it... How would you quantify "a little slow"?

The xml files can be huge, but wavefunction files in general tend to be. The main point being that it allows for direct access to the files for custom code work.

I can quantify a little slow with a specific research experience. This isn't the case for all uses of Quantum Espresso, etc....

When I run an HSE or PBE0 calculation on 192 atoms it takes just over 2 weeks of computation time using quantum espresso. When my co-worker did it in vasp using the same computing systems it took just a couple of days. I don't really know the reason for this.

From time to time things like this creep up.

One of the quirks you should be aware of an enforce in all runs where this is relevant is you need to set

wf_collect=.true.

if you do not then each subsequent run will require the same number of processors in the same configuration each time the program is run.

http://www.quantum-espresso.org/wp-content/uploads/Doc/INPUT...

When I was working on this stuff, the most popular GPLed code was ABINIT – do people still use that? (My lab was a SIESTA shop, and there were a lot of CASTEP people around, but not many using VASP...)

No problem; I look forward to seeing it. Also, I just read in your marketing material that you support "connecting multiple simulation codes." As I'm sure you're aware, this is actually a very subtle problem and an active research area in terms of e.g. connecting simulations that operate on different length scales, but even on a much more basic level it can be difficult: just translating atomistic data from one file format to another, for example. With that in mind, you might find these projects useful:

http://openbabel.org/ https://wiki.fysik.dtu.dk/ase/

Both I imagine could be quite useful if integrated tightly into Rescale, and I'm sure the community as a whole would be grateful for any updates/refinements made to those codebases to e.g. support more tools.

Reply on news.ycombinator.com