i-PI + PLUMED umbrella sampling

[Aran Lamaire]

Dear i-PI developers

When trying to perform umbrella sampling using i-PI in combination with PLUMED, I ran into some problems. If I am not mistaken, the virial tensor in i-PI and PLUMED are defined in a different way, so that the PLUMED results passed to i-PI (in engine/forcefields.py) should probably contain -vir instead of vir. Compared to an older i-pi-dev version (March 2018) I still use, the newest i-PI github version also seems to miss the bias potential and virial term in the integration of the equations of motion (e.g. in barostats.py), so that in spite of the communication with PLUMED i-PI still performs a regular PIMD simulation.

Thank you in advance for your help.

Kind regards,

Aran

[Michele Ceriotti]

So, the virial might be a problem - never tried to bias something that couples significantly to the cell, so any debugging in this compartment would be highly appreciated.

On the other hand, general biasing (should) work. Try the Zundel example in examples/plumed. 

Note that the bias now acts through ensemble.

I also agree that somewhere along the path it looks like the bias contribution to barostat.pstep disappeared. 

Needs to be carefully tested and re-introduced.