PBMETAD possible issue with namd2 v2.13

[ztimol]

Possible bug when attempting to run PBMETAD in namd2-cuda v2.13 patched with plumed v2.6.

Tried to run the basic PBMETAD distance calculation from the docs and system crashed with atoms moving too fast error on the first time-step. In the colvar file bias reads 'nan'.

This occurs both when trying to run a PBMETAD on 2 colvars and 1 colvar (if you want to do the latter with PBMETAD for some reason).

Basic METAD on 1 and 2 colvars runs without issue on same system. Other plumed functionality seems to run okay so don't think it's a compilation/patching issue. Also running system under NVT.

Haven't found anything directly related to this in the forum but did note this pbmetad issue

Plumed config file, plumed covlar file and namd logs attached. Hills file is empty.

md_sol_plumed.log plumed_metad_dist.txt COLVAR.txt

Not sure if any other info is required but if so let me know.

Thanks a lot

[GiovanniBussi]

Hi,

as far as I know, the NAMD patch distributed with PLUMED does not contain the commands necessary to implement replica exchange with different Hamiltonians, so I would say this is just not supported. I would be surprised if other multi replica algorithms work correctly, because they also seem not supported

Giovanni

[ztimol]

Hi,

Ah of course - I really should have realised that was the issue. Will port over to gromacs to give it a try at some point.

Thanks a lot for the help.

[carlocamilloni]

sorry maybe I missed something, but this looked like a single replica calculation, or was it actually replica-exchange? PBMetaD works in single replica

[GiovanniBussi]

Arg, my fault. I apologize, this is definitely to be checked.

[carlocamilloni]

anyway the issue is already at time zero when there should not be any force applied, can you try to set the hills height to zero?

[maxbonomi]

Hi,

I have to admit PBMETAD might not have been thoroughly tested with standard metadynamics, as it was conceived to be used mostly in combination with well-tempered metadynamics. Could you try you switched on well-tempered metad by adding for example:

BIASFACTOR=8

to the PBMETAD line?

Max

On Jun 8, 2020, at 16:38, Carlo Camilloni notifications@github.com wrote:

anyway the issue is already at time zero when there should not be any force applied, can you try to set the hills height to zero?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or unsubscribe.

[carlocamilloni]

I also thought the same, but I tested locally and it works, still it is worth checking

[ztimol]

> sorry maybe I missed something, but this looked like a single replica calculation, or was it actually replica-exchange? PBMetaD works in single replica

Apologies for the delayed response. I'm new to PLUMED and this form of analysis in general. Just wanted to make sure I didn't miss something on my part.

As you've already picked up it's a standard MD simulation (that is no tcl scripts were used to employ replica exchange).

I did note this Enhanced Conformational Sampling 2015 paper referenced in the docs. However, I assumed (likely incorrectly) that it would be possible to run PBMETAD in a standard MD simulation. Out of curiosity is it even possible (or even sensible) to run a multidimensional collective variable parallel bias metadynamics simulation without replica exchange?

In reality I don't need to employ PBMETAD and was just testing PLUMED functionality. i looked into it as I was actually having issues with both standard and well-tempered 2D metadynamics on ring puckering that I think I now may have managed to work around.

I observed a strange artifact in the molecular structure. I'm just testing to see if it occurs again and that it wasn't an error on my part. If it is a possible issue will post it here in the next few days.

Thanks for the help and I'm happy to assist in any way I can.