R: RE: R: RE: R: Re: R: RE: R: RE: R: RE: R:[gmx-users] Tabulated potential - Problem

Fri Oct 9 13:37:53 CEST 2009

OK, I'll answer once more.

You are apparently convinced you are doing the right thing.

But if you have a problem and want someone to help you,

you should report all warnings you get so users do not waste their time

(and your time) helping you. Mentioning this warning right away

would have saved you and us a lot of work.

I had a look at your table and I would guess you force is simply

a factor 10 too large (you might have scaled from Angstroms to nm

or something like that and forgot to scale the force).

The code only determines an approximation of the derivative to

be able to give the warning it gave you.

Berk

Date: Fri, 9 Oct 2009 13:17:45 +0200
From: albitauro at virgilio.it
To: gmx-users at gromacs.org
Subject: R: RE: R: RE: R: Re: R: RE: R: RE: R: RE: R:[gmx-users] Tabulated	potential -	Problem

I asked your help since my tabulated potential reports in the third column the derivatives (minus the derivatives...to be precise) which are analitically exact with respect to the potential reported in the second column. So, in this case, these are the exact forces. If the numerical derivatives calculated by gromacs are so different now it is not a problem of my  input and no information are given to solve this problem...hence my request...
Furthermore, If derivatives are calculated by the code please explain me why I have to report them in the table? and where the code uses one or the others?
Probably, I am stupid or ignorant but since it's the first time I use a tabulated potential I could not find any precise answers to this questions.

I am not blind and I can read both my input/output files and also the manual...so I was aware of the WARNING. 
This does not change the fact that using the analytical derivatives (in numerical form) I have errors and for me this is nonsense.
obviously there should be an answer to this question but where is it? 
I don't think that my request was so stupid as you declare. If there are so many details which could affect the calculations please report and justify them explicitly in the documentation....
anyway, thank you for your kind help.
I won't ask anything else about this problem and I will use my time lo learn reading.

AM

----Messaggio originale----

Da: gmx3 at hotmail.com

Data: 9-ott-2009 9.58 AM

A: "Discussion list for GROMACS users"<gmx-users at gromacs.org>

Ogg: RE: R: RE: R: Re: R: RE: R: RE: R: RE: R:[gmx-users] Tabulated	potential -	Problem

Hi,

I ran your system and mdrun prints a VERY clear warning to stderr and to md.log:
WARNING: For the 999 non-zero entries for table 0 in table_b0.xvg the
forces deviate on average 164% from minus the numerical derivative of
the potential

I understand that Gromacs can be somewhat complicated for new users
and that things might go wrong when you try them the first time.
But you sent a large number of mails about your issues and several people have put time
into helping you out.
The least you could and should do is check your the output of your simulations.
This would have also cost you less time than actually sending questions to the mailing list.

I am fed up with spending time on helping a person find out what seemingly unexplainable crashes are,
when the output clearly tells you what the problem is.

I am very sorry to say so, but I will not help you any more on this topic.
Anyhow I think the problem should be solved now.

Berk

Date: Thu, 8 Oct 2009 15:58:41 +0200
From: albitauro at virgilio.it
To: gmx-users at gromacs.org
Subject: R: RE: R: Re: R: RE: R: RE: R: RE: R:[gmx-users] Tabulated	potential -	Problem

Hi Berk,
the potential I used comes indeed from Boltzmann inversion but I fitted it with a polynomial function of suitable degree which gave an almost perfect agreement on the range chosen (outside this range it reproduces the correct repulsive branches of the potential).  I generated the tabulated potential on the basis of this polynomial fitting and also the derivatives in the third coulmn of the table were calculated on the basis of the analytical derivative of the polynomial. Therefore, all the numerical noise has been removed. Maybe, are there numerical problems of any kind if the potential has a very large repulsive value on the repulsive branches? I any  case in my initial configuration all the bonds share value near the minimum and so they are not exploring these highly repulsive regions.
Thank you

AM

----Messaggio originale----

Da: gmx3 at hotmail.com

Data: 8-ott-2009 9.12 AM

A: "Discussion list for GROMACS users"<gmx-users at gromacs.org>

Ogg: RE: R: Re: R: RE: R: RE: R: RE: R:[gmx-users] Tabulated potential -	Problem

Hi,

The required spacing is basically only related with the second derivative of the potential
and the accuracy you want. In "standard" cases, your spacing of 0.001 nm should be enough.
But if your potential comes from some numerical inversion procedure, your potential could
be very noisy, which could result in nearly random forces, which could cause the system to crash.
Also the maximum time step you can take will depend on the second derivative of the potential.

Berk

Date: Thu, 8 Oct 2009 11:01:39 +0200
From: albitauro at virgilio.it
To: gmx-users at gromacs.org
Subject: R: Re: R: RE: R: RE: R: RE: R:[gmx-users] Tabulated potential -	Problem

Hi,

I used gmxcheck and gxmdump to compare the results of two simulations on an identical system but using two different tabulated potential, one harmonic (the simulation ended correctly) and one not. There are no differences in terms of non bonded interactions, other energy terms and the parameters of the simulation.
Therefore, it is still something related to the tabulated potential. Is there a minimum or a maximum number of (uniformly spaced) points suggested (or required) for a tabulated potential? In this case I used 1001 points to sample an interval from 0 to 1.0 nm. 
Many thanks,

AM

----Messaggio originale----
Da: Mark.Abraham at anu.edu.au
Data: 5-ott-2009 11.19 PM
A: "Discussion list for GROMACS users"<gmx-users at gromacs.org>
Ogg: Re: R: RE: R: RE: R: RE: R:[gmx-users] Tabulated potential - Problem

albitauro at virgilio.it wrote:
> Hi,
> I came back to my original stretching potential in tabulated form and I 
> have still problems...
> when running a geometry optimization the results are like that:
> 
> **************************************************+
> ^MStep 635, Epot=2.259501e+02, Fnorm=4.086e+00, Fmax=1.886e+01 (atom 34)
> ^MStep 636, Epot=2.259499e+02, Fnorm=3.847e+00, Fmax=1.992e+01 (atom 33)
> ^MStep 637, Epot=2.259495e+02, Fnorm=1.002e+01, Fmax=5.408e+01 (atom 33)
> ^MStep 638, Epot=2.259449e+02, Fnorm=1.026e+01, Fmax=4.098e+01 (atom 34)
> ^MStep 639, Epot=2.259393e+02, Fnorm=2.193e+01, Fmax=1.159e+02 (atom 35)
> ^MStep 640, Epot=2.259364e+02, Fnorm=1.469e+01, Fmax=7.064e+01 (atom 34)
> ^MStep 641, Epot=2.259335e+02, Fnorm=1.281e+01, Fmax=7.154e+01 (atom 34)
> ^MStep 642, Epot=2.259315e+02, Fnorm=5.803e+00, Fmax=2.861e+01 (atom 35)
> ^MStep 643, Epot=2.259314e+02, Fnorm=6.332e+00, Fmax=3.116e+01 (atom 34)
> ^MStep 644, Epot=2.259310e+02, Fnorm=3.550e+00, Fmax=1.502e+01 (atom 34)
> 
> Stepsize too small, or no change in energy.
> Converged to machine precision,
> but not to the requested precision Fmax < 0.1
> ************************************************************************************
> 
> as if the system is frozen. My system here is an isolated, linear and 
> finite-length chain.
> When running an md simulation I got the error:
> 
> -------------------------------------------------------
> Program mdrun_mpi, VERSION 4.0.5
> Source code file: bondfree.c, line: 1772
> 
> Fatal error:
> A tabulated bond interaction table number 0 is out of the table range: r 
> 1.815411, between table indices 1815 and 1816, table length 1001
> -------------------------------------------------------
> 
> The tabulated potential is uniformly spaced, the bond type is 8 to 
> exclude LJ interaction between bonded atoms, nrexcl=1 to include 1-3 
> interactions (as required by MARTINI force field). 

That seems like it should work.

> The same system does 
> not have any problem when running simulation with an harmonic stretching 
> potential in both numerical and analytical form.
> Does anyone have any suggestion for possible solutions or error in the 
> input?

Running gmxcheck between various .tpr files may be instructive, e.g. one 
file may have many more nonbonded interactions, or such. Otherwise, 
posting your .mdp file and the start of your .top may help us spot a 
problem.

Mark
_______________________________________________
gmx-users mailing list    gmx-users at gromacs.org
http://lists.gromacs.org/mailman/listinfo/gmx-users
Please search the archive at http://www.gromacs.org/search before posting!
Please don't post (un)subscribe requests to the list. Use the 
www interface or send it to gmx-users-request at gromacs.org.
Can't post? Read http://www.gromacs.org/mailing_lists/users.php

Express yourself instantly with MSN Messenger! MSN Messenger

_________________________________________________________________
What can you do with the new Windows Live? Find out
http://www.microsoft.com/windows/windowslive/default.aspx
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20091009/f960c6dd/attachment.html>