Re: timestep querry

From: Abhishek TYAGI (atyagiaa_at_connect.ust.hk)
Date: Thu Apr 02 2015 - 23:07:39 CDT

​Dear Brian,


I understand the point you raised, could you please help me out, how should I check the reliablity of the system and observe these changes. which equation you are talking about.


regards


Abhi

________________________________
From: Brian Radak <brian.radak.accts_at_gmail.com>
Sent: Friday, April 3, 2015 3:17 AM
To: namd-l; Abhishek TYAGI
Subject: Re: namd-l: timestep querry

I think it's important to identify specifically what you mean by "the desired result." I would argue that what one should mean is that you get samples from the desired Boltzmann distribution with the correct weight/frequency. Increasing the time step can violate this criteria. However, it is usually assumed that if energy is well conserved during an NVE simulation, then temperature dependent distributions will be properly sampled. I don't know if that is an exact proof, but it is probably generally accepted.

To be sure, changing the time step does change your simulation results; I believe this is a provable feature of Verlet-type integrators. Whether or not the difference is detectable within statistical error is perhaps a separate question.

Regards,
Brian

On Wed, Apr 1, 2015 at 11:15 AM, Abhishek TYAGI <atyagiaa_at_connect.ust.hk<mailto:atyagiaa_at_connect.ust.hk>> wrote:

​Dear Aron,


Thank you overmuch for the detailed description and clarifying reply.

The system I am using is initially equilibrated in NPT and than I had performed NVT ensemble. whole simulation performed at tilmestep 1.


I tried running the simulation at the same tilmestep 1 for three times, and did not observe any specific change in the MD.


Further, after 50s, I changed to 1.5 just to have a check on the simulation differences. After almost 20 ns, their is change in the structural orientation. This is run for longer duration to check the changes. The RMSD increased of the ligand.


The smaller time step provide the similar structural orientation for about 100 ns, however, change in timestep form 1 to 1.5, helps in getting desired results.


I am agree to all the good points suggested by you, I will try to repeat the simulations for the changed time step 1.5 for many times., but I just want to understand that, if somehow I get the desired result in performing simulation at tilmestep initially for 50ns, further 1.5 ns for next 50s, this will be correct approach or wrong?


Thanks again for the reply


Abhi

________________________________
From: Aron Broom <broomsday_at_gmail.com<mailto:broomsday_at_gmail.com>>
Sent: Wednesday, April 1, 2015 11:51 PM
To: namd-l_at_ks.uiuc.edu<mailto:namd-l_at_ks.uiuc.edu>; Abhishek TYAGI
Subject: Re: namd-l: timestep querry

You can change the timestep in the middle of the simulation, there is nothing theoretically wrong with this. This is no different than simply having as a starting structure whatever you came up with from the first simulation. You do, of course, need to be careful in your analysis if any of the properties you are analyzing have time dependent units and you try to analyze the whole merged simulation at once.

If you observed a difference, are you sure it is actually because of the change in timestep, or is it just something that would have happened if you had run the simulation longer regardless? Particularly, if you think the ligand is moving to a lower energy position, you would expect that to happen in any case.

If there is actually a discrepancy between two different timesteps (this is very hard to pin down, because of the stochastic nature of simulations), then the only sensible conclusion would be that the smaller timestep is the more reliable. But really, you'd have to do a number of runs with these different timesteps, but both having the same total simulation time, in order to be at all convinced that there was a real difference.

I have seen several cases of increasing the timestep after initial rough equilibration. The logic being that early on you may be populating some high energy states as you get rid of odd structural configurations from whatever structure you started with, and these are likely to cause instabilities in the simulation at higher timesteps. Once things have settled, you can go higher and try to get more information with less computational time. The same idea applies to the use of pressure vs. constant volume. Adding a barostat will slow down the simulation, so in some cases it can be effective to use the barostat for some initial time until the volume of the cell equilibrates (from whatever initial deformities/voids were caused by the solvation algorithm), and then move to constant volume.

In the end, I've found there are a lot of "best practices" for MD that you can glean from reading the methods of articles or looking over tutorials written by experienced researchers, but there seem to be few hard rules. Probably the best test in most cases is to run the simulation multiple times, and if the results are all similar and lead to the same conclusion, that is a good sign.

~Aron

On Wed, Apr 1, 2015 at 11:28 AM, Abhishek TYAGI <atyagiaa_at_connect.ust.hk<mailto:atyagiaa_at_connect.ust.hk>> wrote:
Dear Experts,


For a trial I performed a simulation, initially the time step 1 assigned in configuration file, with rigidbond all, further, after 50 ns I changed tilmestep to 1.5, some questions related to this:


1. The change in the tilmestep leads to change in structural orientation, understood as integration time step is bound by the fastest bond vibration

2. Is this changing in time step is correct or wrong in the middle of simulations, I changed to check the system behaviour, as in curiosity, i observed that the receptor-ligand interaction increased.

3. If change in time step provides relevant results, is it possible to use these changes in result output for further analysis.


Therefore, my observation for this work is this, initially at time step 1 less receptor-ligand interaction, but after increasing time step to 1.5 lead to change interaction orientations. This approach is correct or wrong, if wrong, how to make it correct? Can we increase time step after some nano seconds?


I tried to search the mailing list, but not able to find something relevant



--
Aron Broom M.Sc
PhD Student
Department of Chemistry
University of Waterloo



--
Brian Radak
Postdoctoral Scholar
University of Chicago
Department of Biochemistry & Molecular Biology
Gordon Center for Integrative Science, W323A
929 E. 57th St.
Chicago, IL 60637-1454
Tel: 773/834-2812
e-mail: radak_at_uchicago.edu<mailto:radak_at_uchicago.edu>

This archive was generated by hypermail 2.1.6 : Tue Dec 27 2016 - 23:21:02 CST