path collective variable

[qingh...@gmail.com]

Hello Plumed group,

I am using path collective variable to study a loop motion. I used g_morph from Gromacs to generate the 26 intermediates using two crystal structures at open and close states, respectively. The interpolation of conformations is from -0.1 to 1.1. The RMSDs of backbone of the loop between each pair of neighbor conformations is 0.183 angstrom (alignment was one on the backbone atoms of the whole chain) . Based on the equation from the tutorial, I got a lambda = 6867.9 nm-2 (2.3/0.0183/0.0183 nm-2)

Then, I ran metadynamics simulations on the S_path and Z_path, but the simulations crashed after hundreds picoseconds. I got some error like this:

Fatal error:
653 particles communicated to PME node 8 are more than 2/3 times the cut-off out of the domain decomposition cell of their charge group in dimension x.
This usually means that your system is not well equilibrated.

I used 72 cores for the simulation, and 500ps NVT and 500 ps NPT equilibration have already been performed before the metadynamics simulation.
And the same equilibrated system could be used for umbrella sampling successfully. So I don't think that the system was not well equilibrated.

As I ran a HREX simulation scaled the loop some time ago, I used the same plumed file to calculate the S_path and Z_path on the HREX trajectory file.
At some frames, I got "-nan" and "inf" for S_path and Z_path, respectively. Then I tried with a smaller lambda (2986.0 nm-2), then I didn't get any "-nan, inf".
I tried this lambda=2986.0 nm-2 (1.0/0.0183/0.0183 nm-2) as I saw people in this paper used the inverse of the MSD between successive frames as the lambda.
https://www.nature.com/articles/srep08425

I also tried to run the same metadynamics simulation on the S_path and Z_path using GPUs, the simulation also crashed with some error of Segmentation fault.
I also ran SMD simulations driving the S_path, some runs crashed at around 10 ns because of Segmentation fault.


So does it mean that I used a too high lambda in the simulation? Could you please give me some suggestions on my setup? Any tips will be appreciated!


######   plumed.dat  ####
# Unit setting
 UNITS LENGTH=nm TIME=ps ENERGY=kcal/mol
# make chain AB whole
 chainAB: GROUP ATOMS=1-8258
 WHOLEMOLECULES ENTITY0=chainAB

 path: PATHMSD REFERENCE=refs26.pdb LAMBDA=6867.9

 meta: METAD ARG=path.sss,path.zzz ADAPTIVE=DIFF SIGMA=125 HEIGHT=0.5 TEMP=300 BIASFACTOR=16 PACE=125 FILE=HILLS

 PRINT STRIDE=25 ARG=* FILE=colar.dat

#######




All the best,
Qinghua

[Pratyush Tiwary]

Can you send a plot of s and a variables ' trajectory from start to crash ?

Pratyush 

--
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+unsubscribe@googlegroups.com.
To post to this group, send email to plumed...@googlegroups.com.
Visit this group at https://groups.google.com/group/plumed-users.
To view this discussion on the web visit https://groups.google.com/d/msgid/plumed-users/6d408016-9218-4417-9342-2a3669c1e88e%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[qingh...@gmail.com]

Hello Pratyush,

Thanks very much for your reply. The attached files are the S_path and Z_path during the simulations from beginning to crash. Thanks a lot!


All the best,
Qinghua

To unsubscribe from this group and stop receiving emails from it, send an email to plumed-users...@googlegroups.com.

[Pratyush Tiwary]

Hi Qinghua

Firstly, am I reading it right that you are depositing 2-d hills with width 125 units? The whole CV space for s goes from 0 to 9 and z from 0 to 0.1 The gaussian width can not be comparable to that  - unless I am missing something super obvious, the sigma value is a 1000 times or even 10000 times larger than what it should be. 

Secondly, in the paper you mention lambda is described as proportional to, not the same as, the distance between the inverse of the MSD between frames. The path CV formalism is exact in lambda tends to infinite limit - in practice lambda has to be small and finite (think replacing delta function with a finite width gaussian for practical reasons). I would probably try a value around 100-500 nm^2

Thirdly, if both of these work try 1-d metadynamics biasing only s.

Fourthly, if even that crashes, do 1-d metadynamics but with z variable restrained to <0.04 or so roughly.

Hopefully you will have some progress through these.

- Pratyush

--

Pratyush Tiwary

Post-doctoral Scientist, Columbia University

Department of Chemistry

sites.google.com/site/pratyushtiwary/

To unsubscribe from this group and stop receiving emails from it, send an email to plumed-users+unsubscribe@googlegroups.com.

To post to this group, send email to plumed...@googlegroups.com.
Visit this group at https://groups.google.com/group/plumed-users.

To view this discussion on the web visit https://groups.google.com/d/msgid/plumed-users/eaed1d1f-a838-4797-a89d-0acffc6963e8%40googlegroups.com.

[qingh...@gmail.com]

Hello Pratyush,

Thanks very much for your detailed comments and suggestions. As I used adaptive Gaussian potential (ADAPTIVE=DIFF), which is recommended in the tutorial, 125 for PACE is not the Gaussian width, but number of steps. Actually, I also tried with the standard Gaussian potential with PACE=0.15, 0.15 for S_path and Z_path, the simulations also crashed at some points.

Then I will try to follow your suggestions, reducing the lambda, doing 1D metadynamics. I will post the updates here if I have some progress!

Thanks again for your helpful suggestions!


All the best,
Qinghua

Hi Qinghua

[qingh...@gmail.com]

Sorry, the PACE should be SIGMA!

[Pratyush Tiwary]

Hi

You do have SIGMA=125 as well as PACE=125 just to be clear. So I still think you have mega-gaussians :)

Pratyush

--

Pratyush Tiwary

Post-doctoral Scientist, Columbia University

Department of Chemistry

sites.google.com/site/pratyushtiwary/

To unsubscribe from this group and stop receiving emails from it, send an email to plumed-users+unsubscribe@googlegroups.com.

To post to this group, send email to plumed...@googlegroups.com.
Visit this group at https://groups.google.com/group/plumed-users.

To view this discussion on the web visit https://groups.google.com/d/msgid/plumed-users/81385528-55b4-47ec-a282-a7671fbf435a%40googlegroups.com.

[Davide Branduardi]

Hi

from your plots it looks that your lambda is too large: you have a stepwise evolution in s.

Maybe your frames are not equally spaced? If so just get the largest RMSD to tune lambda.

HTH

D

******************************************************
Davide Branduardi
davide.b...@gmail.com
Web:
https://sites.google.com/site/davidebranduardi
******************************************************

To view this discussion on the web visit https://groups.google.com/d/msgid/plumed-users/CAO8zxx43jB1gqhrFtfV2bhOYZ%2BEvVJDo1sattOgp0uTEuJ04Eg%40mail.gmail.com.

[qingh...@gmail.com]

Hello Pratyush,

As I already set ADAPTIVE=DIFF, the SIGMA should number of timesteps, and it is adaptive-gaussians. Otherwise I need two values for SIGMA.

Anyway, I will do more tests as you suggested. Thanks again for your suggestions!


All the best,
Qinghua

[Pratyush Tiwary]

Ah OK sorry I missed that. Hopefully lambda will solve your problem. 

To unsubscribe from this group and stop receiving emails from it, send an email to plumed-users+unsubscribe@googlegroups.com.

To post to this group, send email to plumed...@googlegroups.com.
Visit this group at https://groups.google.com/group/plumed-users.

To view this discussion on the web visit https://groups.google.com/d/msgid/plumed-users/74d526f0-2a11-4d72-a7dd-da5b1ecf2ab4%40googlegroups.com.

[Davide Branduardi]

A nice way to check if your adaptive gaussians are making sense is plotting with gnuplot 

p "file" u 1:2:3:4 w xyer

where 1:2 should be replacedf by the columns for S and Z in the HILLS file

and 3:4 are the diagonal elements of the width. It is not meant to be realistic (you loose the shape component) but if you do see very tiny bars it means that probably your hills are very small and, as a consequence, your force can be very high and this may cause instabilities in the integration of the eq. of motion.

I feel that's a nice and simple test to do when using adaptive gaussians.

Cheers

Davide

******************************************************
Davide Branduardi
davide.b...@gmail.com
Web:
https://sites.google.com/site/davidebranduardi
******************************************************

To unsubscribe from this group and stop receiving emails from it, send an email to plumed-users+unsubscribe@googlegroups.com.

To post to this group, send email to plumed...@googlegroups.com.
Visit this group at https://groups.google.com/group/plumed-users.

To view this discussion on the web visit https://groups.google.com/d/msgid/plumed-users/74d526f0-2a11-4d72-a7dd-da5b1ecf2ab4%40googlegroups.com.

[qingh...@gmail.com]

Hello Davide,

Thanks very much for your reply! The frames were generated by g_morph of gromacs, and I have check it with vmd, they are equally spaced.
Yeah, maybe my lambda is to big, as suggested by Pratyush. I will give a try on it.


All the best,
Qinghua

To view this discussion on the web visit https://groups.google.com/d/msgid/plumed-users/81385528-55b4-47ec-a282-a7671fbf435a%40googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

--
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 post to this group, send email to plumed...@googlegroups.com.
Visit this group at https://groups.google.com/group/plumed-users.

[qingh...@gmail.com]

Hello Davide,

Thanks for this suggestion, too. I will check the it then!


All the best,
Qinghua

[qingh...@gmail.com]

No problem! Thanks for the discussion!  :-)

[qingh...@gmail.com]

Hello Davide,

I attached the plots for the 200 ps simulations of the hills.

I guess you are right, I have very small bars, which might make the simulations crash. Then I will do more tests on it. Thanks again for your suggestions.


All the best,
Qinghua

On Wednesday, May 10, 2017 at 2:06:36 PM UTC+2, Davide Branduardi wrote:

[qingh...@gmail.com]

Hello again,

Pratyush and Davide suggested that a lambda of 6867.9 nm-2 might be too big in my case, but the lambda is much bigger in the tutorial, 15100 nm-2.
Then how should I think of this difference, because the system of the tutorial is very small? Could someone give a comments on this point? Thanks a lot!


All the best,
Qinghua

[Davide Branduardi]

One of the issues is that you're exploring conformations very far from the path (0.1nm^2 ~3 Ang rmsd!) and the lambda calculation reported in the tutorial assumes you stay close to the input path. When you are far the distance between images is sort of distorted and this contributes to what you see.

Cheers

Davide

******************************************************
Davide Branduardi
davide.b...@gmail.com
Web:
https://sites.google.com/site/davidebranduardi
******************************************************

--
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+unsubscribe@googlegroups.com.

To post to this group, send email to plumed...@googlegroups.com.
Visit this group at https://groups.google.com/group/plumed-users.

To view this discussion on the web visit https://groups.google.com/d/msgid/plumed-users/1ef4fd8a-fcc5-4f56-a74e-02298e0b93bb%40googlegroups.com.

[qingh...@gmail.com]

Hello Davide,

Thanks a lot for your explanation! I understand it now! :-)


All the best,
Qinghua

To unsubscribe from this group and stop receiving emails from it, send an email to plumed-users...@googlegroups.com.

To post to this group, send email to plumed...@googlegroups.com.
Visit this group at https://groups.google.com/group/plumed-users.