parallel temper move

[directysj]

Hello @mlund

I am testing parallel temper in Faunus.

I have tried to run temper.yml script example.
The command is used as -- $yason.py temper.yml | mpirun --np 3 --stdin all faunus

It stuck after the command and nothing has happened, no errors. Only creating three files with names mpi0.stdout, mpi1.stdout, and mpi2.stdout. My system (Debian 10) is installed openmpi version 3.1.3 and GCC 10.3.0.

I performed $make test after $make install, which fails for temper.yml.

Can I ask your suggestions?

Thanking you in advance!!

[mlund]

The parallel temper example must be run through the notebook as temper.yml is a template used to generate the three input files for each replica. You'll need to install Jupyter Notebook. Could you describe what is failing with make test and make install? I have it passing on my machine.

[mlund]

I just checked the manual where it says "the move is performed exactly once per Monte Carlo cycle". Can you confirm this in your setup? This may be too often for your system and we could consider implementing a way to do this at a user defined interval. Besides that, I think the main advantage of tempering is that rare events are better sampled in each replica than if they were brute force sampled N times longer.

[directysj]

I just checked the manual where it says "the move is performed exactly once per Monte Carlo cycle". Can you confirm this in your setup? This may be too often for your system and we could consider implementing a way to do this at a user defined interval. Besides that, I think the main advantage of tempering is that rare events are better sampled in each replica than if they were brute force sampled N times longer.

Thank you so much for the quick reply. The temper move is performed once a MC cycle. If you can add user defined interval for it then I can test if this make any speedup. I agree with you that replicas sample rare events better.

[mlund]

OK, could you try the new branch in PR #381? See the list of changes of how to use it.

[directysj]

OK, could you try the new branch in PR #381? See the list of changes of how to use it.

Thank you so much for adding nskip to temper move. I will test it and let you know any differences.

[directysj]

OK, could you try the new branch in PR #381? See the list of changes of how to use it.

I have tested for one run and the speedup is 21% corresponding to skip 100--> temper: {format: xyzq, nstep: 100}