[gmx-developers] box distortion with isotropic pressure coupling

[Berk Hess]

Hi,

With isotropic scaling each box vector is multiplied each step my the 
scaling factor mu.
For a cubic box all diagonal elements are equal and all others zero.
Multiplying identical numbers with mu again gives identical numbers.
For any non-cubic box, rectangular or trilicinic, there will be 
different elements.
With limited precision multiplying these different number by a factor 
like 1.000001
results in different ratios of different numbers before and after 
multiplication.
In the extreme case one box element is much larger than the others and 
mu is so
1+eps with eps so small that it will only increase this one element, 
whereas the
others would not change.

In general the different box elements will diffuse due to different 
rounding on the least
significant bit.
I don't know if in your case you really see a continuous decrease and 
increase
or if this is just diffusion.

Berk.

Chris Neale wrote:
> I took a look at the cvs code and I wonder if the fix is simply the 
> call to preserve_box_shape()
> that has been added to void berendsen_pscale()?
>
> While the method applied by preserve_box_shape() via do_box_rel() 
> appears quite reasonable,
> this still doesn't explain to me why there would be a problem without 
> this fix, and, specifically,
> why x and y would tend to decrease while z tends to increase in the 
> absence of a call to
> preserve_box_shape().
>
> Any comments at all would be appreciated.
> Thanks,
> Chris.
>
> Chris Neale wrote:
>> Dear developers,
>>
>> I have a 'rectangular' system and many rhombic dodecahedron systems 
>> that, when simulated under
>> isotropic pressure coupling, show a variable ratio of box side 
>> lengths. This does not happen for
>> cubic systems.
>>
>> I gather that this is a known bug, and that it has been fixed for 
>> gromacs 4, but I would like to get some details about the exact 
>> nature of this ex-bug in order to determine how valid my simulations 
>> are. I suspect that it's something to do with using box_x*box_y as a 
>> divisor somewhere for all planes in the pressure coupling, as that 
>> would explain why there is no effect for cubic systems.
>>
>> I do not expect that there would be a serious problem for a protein 
>> in a box of water, but I imagine that it is possible that isotropic 
>> studies on detergent self-aggregation and especially semi-isotropic 
>> studies of lipid bilayers could suffer from significant 
>> non-equilibrium forces. This would be especially
>> relevant for lipid bilayers now that it is fixed since some lipid 
>> parameter refinement was conducted while using earlier versions of 
>> gromacs and may now require re-investigation.
>>
>> I posted this on the users list:
>> http://www.gromacs.org/pipermail/gmx-users/2008-July/035264.html
>>
>> When I mentioned this on the list, I was pointed to some
>> bugzilla comments: http://bugzilla.gromacs.org/show_bug.cgi?id=165  
>> that reference revision comments:  
>> http://www.gromacs.org/gromacs/revisions/
>>
>> Berk Hess
>> 13 Feb 2008
>> box distortion with pressure coupling
>>
>> In all Gromacs versions before 4 the box shape would "diffuse" over  
>> time when using (semi)isotropic pressure coupling. This issue (only  
>> noticable for very long simulations) has been fixed for version 4.0.
>>
>>
>> Which I then reposted to the users list:
>> http://www.gromacs.org/pipermail/gmx-users/2008-July/035270.html
>>
>> If anybody knows the exact nature of the bug then I would be grateful 
>> to hear about it
>> and ideally to see some code snippits or a diff.
>>
>> Many Thanks,
>> Chris.
>>
>>
>>
>>
>> _______________________________________________
>> gmx-developers mailing list
>> gmx-developers at gromacs.org
>> http://www.gromacs.org/mailman/listinfo/gmx-developers
>> Please don't post (un)subscribe requests to the list. Use the www 
>> interface or send it to gmx-developers-request at gromacs.org.
>>
>
> _______________________________________________
> gmx-developers mailing list
> gmx-developers at gromacs.org
> http://www.gromacs.org/mailman/listinfo/gmx-developers
> Please don't post (un)subscribe requests to the list. Use the www 
> interface or send it to gmx-developers-request at gromacs.org.
>
> This email was Anti Virus checked by Astaro Security Gateway. 
> http://www.astaro.com
>