i-PI simulations exploding bonds

[Will Borrelli]

Hi all, 

Thanks in advance for any help. I am trying to connect i-PI to our group's fortran-based code which simulates SPC/fw water with periodic boundary conditions along with an optional quantum mechanical solute. At this point, I am just trying to validate that i-PI works with pure SPC/Fw water classically and 32 bead PIMD before moving to adding solutes. 

I believe that I have everything coded up properly, and my simulation runs for 100s of fs up to several ps, and during this time the observables seem reasonable for both classical water (1 bead) and quantum water (32 bead PIMD). The issue is that, seemingly randomly, one of my waters will explode and the simulation becomes unstable. Up until that point nothing seems out of the ordinary. For single atomic configurations, I have validated that the unedited version of our code returns the same forces and energies as when I run it through i-pi. 

I have a suspicion that it could be periodic boundary condition related, although I don't fully understand why. Our code allows free water diffusion but computes all distances with minimum image convention. I have the i-pi socket with pbc=False since it seems to worsen the instabilities (i-pi wraps atoms not water molecules, which upsets the topology in our code). For running PIMD, I lock the bead to each client so I don't think it would be related to bead indexing issues. 

If anyone has a recommendation for things to check to hunt this down I would greatly appreciate it! Attached is my input.xml and initial xyz file.