[gmx-users] Fwd: system cooling down

[Erik Lindahl]

Begin forwarded message:

> From: "H.J.Risselada" <H.J.Risselada at rug.nl>
> Date: February 1, 2006 3:44:01 PM GMT+01:00
> To: lindahl at sbc.su.se
> Subject: system cooling down
>
>
> Dear mister Lindahl,
>
> If I may, I would like to ask you your opion about the following:
>
> Something that has puzzled me for months is the fact how something  
> (numerical errors ?) can cause a system cooling down instead of  
> heating up.
>
> I implemented stochastic boundary potentials in the gromacs -3.3  
> source code for the simulation of big vesicles.
> In this case it is made possible to simulate an union-shell-like  
> system, having a vacuum cavity in the inside and a vacuum space on  
> the outside.
> Solvent close to the edge of a shell are experiencing a mean field  
> force representing a continious water envirement.
>
> My test system has an inner radius of 10nm and an outer radius of  
> 22 nm (see attachment 1), so the tickness of the spherical shell is  
> 12 nm. A stochastic heat bath is only applied in the region  
> starting 2.5 nm from the boundary. In the reaction zone (radius  
> 12.5 - 19.5 nm) the dynamics is simple newtonian without T-coupling.
>   The stochastical intergrator I use is the so called LI-(Langevin  
> impulse) leapfrog intergrator. To allow a smooth tranistion from SD  
> to MD the friction coeffient is at some point (10.5 - 21.5) lineary  
> decreased till fric->0.
> (see attachment 1).
>
> I remove both angular - linear momentum, I checked and during the  
> run no rotational energy builds up.
>
> Similar methods (but lacking the internal cavity) where implemented  
> in CHARMM and EGO and never used in systems bigger than 3nm radius.  
> Because it's dam'n slow ;)
>
> I can duplicate their systems getting a beautifil uniform radial  
> temperature profile.
> However when the outer radius becomes > 8 nm, the temperature  
> profile becomes non uniform. In the middle of the shell the system  
> cools as a function of time until it remains at some constant  
> value. (See attachment 2)
>
> Cooling down can physically only occur when trans. energy escaped  
> to some other energy form (rotation).
>
> Nurmerical errors, however normaly tend to heat up the system.
>
> However, the problem seems to scale with the neighbour list update  
> frequency. When the update frequency is 1 the problems seems  
> improved........but to slow for normal usage.
>
> So my question is what causes the energy leak, why does it cool  
> down? and how might I possible solve it the "nice way".
>
> I'm already bussy with some cosmetic solutions for this problem ;)
>
> Thank you in advance,
>
> greetings
>
> Jelger
>
> I use a CG-model, lennard jones interaction only,
> Constaints none ;)
> Shifted potentials,
> The time step I use is 0.04 ps.
>
>
>
>
>
>
>
>

-----------------------------------------------------------
Erik Lindahl  <lindahl at sbc.su.se>     Backup address:  
<erik.lindahl at gmail.com>
Assistant Professor, Computational Structural Biology
Stockholm Bioinformatics Center, Stockholm University, SE 106 91  
Stockholm
Phone: +46 8 5537 8564     Fax: +46 8 5537 8214




-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20060201/4a7fc349/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: profile.png
Type: image/png
Size: 7130 bytes
Desc: not available
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20060201/4a7fc349/attachment.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: spher_temp.ps
Type: application/postscript
Size: 14949 bytes
Desc: not available
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20060201/4a7fc349/attachment.ps>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20060201/4a7fc349/attachment-0001.html>