Reweighting and GRID in path-CV metadynamics with with adaptive hills

[Yong Wang]

Hi,

I am doing metadynamics using path CVs with adaptive hills (ADAPTIVE=DIFF) from which I hope to alleviate the boundary issue at the high and low path.s regions (with high density and low fluctuation). But I have two technical questions:

1) The performance drops terribly as the simulation goes longer. So I tried to use GRID, but alway got the WARNING no matter how I set the GRID_SPACING:

PLUMED: WARNING: to use a METAD with a GRID and ADAPTIVE you need to set a Grid Spacing larger than half of the Gaussians

Actually I just ignored this warning and performed a simulation with a high GRID_BIN and the preliminary results looks quite ok. So I am wondering if it is safe to ignore this warning. If so I suggest to remove it in the coming version.

 2) My second question: is it possible to do reweighting to get the PMF on other CVs from the metadynamics with adaptive hills? I got the WARNING that 

reweighting has not been proven to work with adaptive Gaussians. 

As far as I understand, the PMF reconstruction from adaptive-hills metad is a summarisation of two components: PMF*(gamma-1)/gamma + -kbt*log(P) where the latter term is used for correction of the occasionally adding very wide hills. I guess it is the latter term might hamper the implementation of reweighting here. Actually, I did a test on a model peptide system, and found the PMF from the summarization of two components is very similar to that obtained from reweighting on the same CV. Does this test indicate the reweighting process also works for adaptive-hills in practice? Or maybe this is just a lucky case where I found the correction term is quite flat, so only contributes very little to the free energy difference (just 1-2 kJ/mol). In contrast to the convergence error, so the correction error might be ignorable. Does this explanation make sense? 

So even that "reweighting has not been proven to work with adaptive Gaussians", it might be doable in practice in the cases that the correction term has minor contribution or the adaptive hills do not go crazily. Does it make sense? 

Any comments and suggestions are appreciated!

Regards,

Yong

==========================

Yong Wang

Postdoc in Lindorff-Larsen Lab
Structural Biology and NMR Laboratory  &  Linderstrøm-Lang Centre for Protein Science

Department of Biology, University of Copenhagen
Ole Maaløes Vej 5, DK-2200 Copenhagen N, Denmark
https://scholar.google.com/citations?user=LblGaZMAAAAJ&hl=en