lab-cosmo / i-pi-dev_archive

Development version of i-PI
21 stars 12 forks source link

Cannot resume a simulation with LAMMPS #160

Closed duran closed 7 years ago

duran commented 7 years ago

I have been trying i-PI and unlike LAMMPS vanilla ensembles I cannot split the simulation in several sections, from something like this:

fix ipi ...

run 200

to something like this:

fix ipi ...

run 100

run 100

In the second run the simulation halts. I am not sure if this is the expected behaviour, some problem with my software or a bug.

I am working with:

ceriottm commented 7 years ago

Uhm... we never tried this and in principle I have a hard time seeing when this might be useful, but indeed it should work. I suspect that the socket is not properly closed and reopened when a "run" section ends. Perhaps @robeme who has recently cleaned up the LAMMPS interface can see to this.

robeme commented 7 years ago

Could reproduce the bug and found a solution. init() in the fix class assigns in the second run zero to ipisock due to the if/else construct. I introduced another flag which tells lammps if a socket had been already assigned and check this whenever init() is called.

duran commented 7 years ago

Do you have any plans to patch this bug?

I have been following the repo and trying new versions of the software but I cannot see anything done about this. But maybe I am looking in the wrong branch or whatever.

ceriottm commented 7 years ago

It is actually fixed in the latest github versions of LAMMPS - or so should be.

robeme commented 7 years ago

It should, but the LAMMPS developers changed something in the domain decomposition balancing. This may have an effect on the i-Pi implementation. Which LAMMPS Version you are currently running?BestRAm 14.03.2017 2:50 nachm. schrieb Michele Ceriotti notifications@github.com:It is actually fixed in the latest github versions of LAMMPS - or so should be.

—You are receiving this because you modified the open/close state.Reply to this email directly, view it on GitHub, or mute the thread.

duran commented 7 years ago

I thought the problem was within the server. I did not check the commits in the LAMMPS repository. I will pull and compile the last version later this week.

Thanks for everything, Bruno

ceriottm commented 7 years ago

nono the problem was with LAMMPS.

On 14 March 2017 at 10:34, Bruno Durán notifications@github.com wrote:

I thought the problem was within the server. I did not check the commits in the LAMMPS repository. I will pull and compile the last version later this week.

Thanks for everything, Bruno

— You are receiving this because you were assigned. Reply to this email directly, view it on GitHub https://github.com/cosmo-epfl/i-pi-dev/issues/160#issuecomment-286459540, or mute the thread https://github.com/notifications/unsubscribe-auth/ABESZ5G20K7CIBfWT5ipbkov78FjLz2Qks5rlrNwgaJpZM4Li7bH .