Difficult to add dpd/tstat pairstyle to GPU package?

All,

I am wondering if the reason pair-style dpd/tstat hasn’t been added to the GPU package is because of the difficulty in rewriting it for GPU use, or if it was simply just a lack of demand.

If it’s the latter, is this something that a motivated user could contribute to?

Thank you,

Bryan A. Rolfe
PhD Candidate in Chemical Engineering
Cornell University

I won't say lack of demand since it has come up several times on the mailing list, but we really haven't looked at it yet.

You might want to check out this thread:

http://sourceforge.net/mailarchive/forum.php?thread_name=CAENuAmEE1n_wZZh2VB97opOfuawjMNhwfLkVDXScf8wBSs%2BQXA%40mail.gmail.com&forum_name=lammps-users

Not sure about the plans to eventually release this code. You might want to e-mail and ask. I did send an e-mail in a direct communication that I will be adding support for 'newton on' next week for another model.

Certainly we would welcome any contribution and can help answer questions or get you started.

It might be good to let me know if you decide to work on this. I think that some changes that affect the entire library are going to be necessary at some point and I don't want to waste anyone's effort...

- Mike

Note that for the dpd pair styles there is the additional issue
of how random #s are used to conserve energy, which is
related to the pair styles requiring the newton flag to be on.

Steve

Michael -

Thank you very much for pointing to that useful, and quite recent thread. I’ll look into this a little more and drop the author of that work a line to see if an implementation that includes the newton flags is being actively pursued. Perhaps my systems are large enough that this effect, as Steve pointed out, will not contribute significantly to the outcome of the simulation.

Hopefully I can make a contribution to this package shortly.

  • Bryan