Is Monte-Carlo integrator ready for MPI implementation?

[Peter Yen]

Dear Developers

I am wondering if the newest version of Vampire supports MPI calculation using the Monte Carlo integrator. I tested with an llg-Heun integrator, and it succeeded, but it seems to have an error for the MC integrator in MPI. Any comments will be appreciated.

with kind regards,
Peter

[gabo...@gmail.com]

In the post at [1], it looks like VAMPIRE used to give an error message "Error - Monte Carlo Integrator unavailable for parallel execution" for the Monte Carlo integrator.  In VAMPIRE 6.0, it looks like the error message was removed from sim.cpp under "case 1: // Montecarlo" starting with commit number 5582e49c99998697af956ecbcb1d65c802e796e5 [2].  So "maybe" it will work as long as the cell used is not non-orthogonal [3].

On the other hand, if you are using the constrained Monte Carlo integrator, sim.cpp under "case 3: // Constrained Monte Carlo" [4], it looks like that will still give:

Error - Constrained Monte Carlo Integrator unavailable for parallel execution

Perhaps the developers will have comments, but hopefully that can help in the meantime.

[1] https://groups.google.com/g/vampire-users/c/RAHCvweTmsA/m/DBTVKgpxORwJ

[2] https://github.com/richard-evans/vampire/commit/5582e49c99998697af956ecbcb1d65c802e796e5

[3] https://groups.google.com/g/vampire-users/c/bL_LH6GncMo/m/b5W22mMqBgAJ

[4] https://github.com/richard-evans/vampire/blob/v6.0/src/simulate/sim.cpp

Best Regards,

Gavin

VAMPIRE user

[Peter Yen]

Dear Gavin

Thank you for your useful information. Hope you all the best in your future career! 

with best regards,
Peter

gabo...@gmail.com 在 2022年8月9日 星期二下午1:55:21 [UTC+8] 的信中寫道：

[milton...@gmail.com]

Correct me if I'm wrong but I thought it was impossible to parallelize Monte-Carlo because of how the energy change upon reorientation of one spin wouldn't be well defined if you're attempting to reorient another spin at the same time (at least if they're close enough to interact directly), and that the only way to "do Monte-Carlo in parallel" is to divide a larger simulation into many smaller ones and simply run them in parallel. If there are other ways to parallelize Monte Carlo for simulations in magnetism then I guess they would have to add a separate "parallel Monte Carlo" integrator.