Discrepancy between FES from sum_hills and reweighting

531 views
Skip to first unread message

Billy Noonan

unread,
Jun 15, 2021, 2:01:38 AM6/15/21
to PLUMED users
Dear Experts,

Using the COLVAR files as output from my PBMETAD simulations, I get the .png file attached ending in *sum_hills_COLVAR.png

The command to generate this was:
plumed sum_hills --histo COLVAR.0,COLVAR1,COLVAR2 --kt 2.5775 --sigma X,X --idw Y,Z
Where the COLVAR files are output from a PB-MetaD simulation

The equivalent output from the re-weighting procedure also attached as a *histo.png file.

Additionally, I get a lot of 'inf' values for the free energy output from re-weighting, which I just set to 2000 kJ/mol and did not include in the colour scheme.

I have three questions:
(1) Why are there 'inf' values? Is it under-sampling?
(2) Why are the FES different?
(3) The PB-MetaD simulations have not converged. Should these two methods be the same upon convergence?

Best Wishes,
Billy

FES_HB-charged_contacts_8_pep3_sum_hill_COLVAR.png
FES_HB-charged_contacts_8_pep3_histo.png
Message has been deleted

Billy Noonan

unread,
Jun 17, 2021, 2:30:28 AM6/17/21
to PLUMED users

Hi All,

Apologies for the double post. I understand everyone is busy and would appreciate clarification when free...

I think in the first reply here, it is said I can use sum_hills on the PBMETAD simulation output:

I have noticed in these two papers, the re-weighting method is used for PB-MetaD and WT-MetaD with adaptive Gaussians:
https://pubs.acs.org/doi/pdf/10.1021/acs.jctc.5b00846

It is stated that sum_hills can be used with the adaptive Gaussian method if the value of sigma is small enough.

But it is immediately recommended in many other posts on this list to do re-weighting instead, in most places.
Is sum_hills reliable for PB-MetaD if the simulation has converged? Why does everyone recommend to do the re-weighting?

Best Wishes,
Billy

Giovanni Bussi

unread,
Jun 17, 2021, 3:11:27 AM6/17/21
to plumed...@googlegroups.com
Hi,

if you disregard the PB part, the answer is that you should do reweighting if you have adaptive Gaussians. The reason is that Gaussians that have different width in different regions of the CV space change the relationship between histogram and bias, so that the bias is not anymore proportional to F(s).

I guess the same holds for PB with adaptive Gaussians.

Giovanni


--
You received this message because you are subscribed to the Google Groups "PLUMED users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to plumed-users...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/plumed-users/f2eefada-0f0d-476f-8b7f-0a6f382d3d73n%40googlegroups.com.

Billy Noonan

unread,
Jun 17, 2021, 3:24:51 AM6/17/21
to PLUMED users
Hi Giovanni,

Thank you very much for the reply, am very grateful for the help.

I understood from the paper here that if σ
G was sufficiently small then the effect on the FES by inversion of the bias across CV space would be minimal. 
https://pubs.acs.org/doi/pdf/10.1021/ct3002464
Thank you for the clarification :-)

Am I able to use both sum_hills of the COLVAR files output and the reweighting method with just PB-MetaD? Are both okay without adaptive Gaussians?

Appreciate the assistance

Best Wishes,
Billy

Massimiliano Bonomi

unread,
Jun 17, 2021, 3:26:46 AM6/17/21
to plumed...@googlegroups.com
Reweighting is certainly ok with PB-MetaD and adaptive Gaussians. Building histograms with sum_hills or other codes without reweighting is not correct!

Max

Sent from my iPhone

On 17 Jun 2021, at 09:24, Billy Noonan <billy.willi...@gmail.com> wrote:

Hi Giovanni,
G was sufficiently small then the effect on the FES by inversion of the bias across CV space would be minimal. 

Billy Noonan

unread,
Jun 17, 2021, 3:31:08 AM6/17/21
to PLUMED users
Thank you very much for the clarification :-)

Billy Noonan

unread,
Jun 17, 2021, 3:39:25 AM6/17/21
to PLUMED users
One last question.

If I get the COLVAR-REWEIGHT file as described here in Exercise 4:
https://www.plumed.org/doc-v2.6/user-doc/html/lugano-3.html

Can I use sum_hills on that, as below:
plumed sum_hills --histo   COLVAR-REWEIGHT --kt 2.5775 --idw X,Y --sigma A,B --outhisto histo_X_Y.dat

Here, the re-weighting has been done, and sum_hills is being used with the weights as the input.

Apologies if these questions are naive, I really do appreciate this.

Best Wishes,
Billy
Reply all
Reply to author
Forward
0 new messages