[Centroid potential of mean force in I-PI]

[Bruno Duran]

Hello everyone,

I am trying to calculate the energy barrier between two sites using the NEB algorithm. This should be the same as the classical method changing the classical potential energy for the centroid potential of mean force (CPMF)[1] generated by the Hamiltonian of the extended space. I am using I-PI to approximate the values of the CPMF in this way:

1. I set the correct atom position and desired temperature to I-PI and use 28/56 beads
2. I add some seed, initial velocities, timestep and choose an nve ensemble (all of this should not be important)
3. I run 1 step, and extract the potential energy and centroid forces of the step 0.

But doing this, the results are always the same as using the interatomic potentials without PIMD techniques. I should see different energies and values that allow quantum effects and lower energy barriers. Am I misunderstanding something of I-PI?

I attach an input example that shows what I am doing.

Thanks,
Bruno

[1]: doi.org/10.1016/0009-2614(96)01109-8

[Michele Ceriotti]

Dear Bruno,

   If I understand correctly what you're doing you've not misunderstood i-PI, but rather how the CPMF works. You have to let the system sample the PI Hamiltonian and only then compute the distribution of centroid (collective) coordinate and build a free-energy profile. What you're doing now is setting all of the beads at a position along the MEP, compute their (identical) energies and average them. 

Another possibility that is closer to NEB in spirit is to do instanton calculations that have been recently implemented in i-PI by Mariana Rossi, Yair Litman and Jeremy Richardson. But first and foremost I'd say you need to read up a bit about rate theories and PIMD.

Best

Michele

[Bruno Duran]

Dear Michele,

Thank you for your fast reply and sorry for my late one.

I am aware that I have to sample the centroid distribution  and that what I was getting were all the beads in the initial positions. I had absurd ideas about how i-PI was supposed to initialized the beads.

In LAMMPS I specify a set beads to immobilize (or just do not integrate)  and thus impose a centroid constrain. But I am not sure that I can do that in I-PI, the only reference to fix some atoms is in the undocumented options <fixatoms> in the examples and I guess that it inmobilize all the beads.

Has anybody calculated the CPMF with I-PI before? Do you know if it is possible?

Thanks,
Bruno

[Michele Ceriotti]

So to be clear, you want to compute the PMF by running several trajectories with fixed centroid position?

If that's all you want to do it's not implemented by very very easy to implement. If you want to set fancy constraints it's another (more complicated) story

Cheers

M