[gmx-users] segfault when running the alchemistry.org ethanol solvation free energy tutorial

[Christopher Neale]

Agreed, the following parameters do not segfault in single or double precision:
sc-alpha                 = 0.5
sc-power                 = 1
sc-r-power               = 6
Same goes for http://bugzilla.gromacs.org/issues/1306 

The following parameters give a segfault in single precision but are ok in double precision
sc-alpha                 = 0.001
sc-power                 = 1
sc-r-power               = 48
Same goes for http://bugzilla.gromacs.org/issues/1306 

Sorry for the confusion earlier. I was using a compilation that I thought was double precision 
but it was actually single. Recompiling in double precision gave me the stability outlined above.

Thank you for your assitance Mark and Michael.

Chris.

-- original message --

Michael Shirts mrshirts at gmail.com 
Fri Sep 27 00:41:17 CEST 2013
Previous message: [gmx-users] segfault when running the alchemistry.org ethanol solvation free energy tutorial
Next message: [gmx-users] segfault when running the alchemistry.org ethanol solvation free energy tutorial
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
And with this change, single is running fine as well.

This was a known issue, but was only documented in the expanded.mdp
files, which was an oversight.  After this, I switched so the default
is less likely to cause problems.  Because of some theory improvements
developed in the group in free energy calculation pathways, the
sc-r-power=48 pathway will now be phased out anyway by 5.1.

On Thu, Sep 26, 2013 at 6:37 PM, Michael Shirts <mrshirts at gmail.com> wrote:
> I thought I had just managed to solve the issue :)
>
> If you look at the soft core parameters, there are two types listed --
> one with sc-r-power = 48, and one with sc-r-power = 6.  The sc-r-power
> are more stable with single precision calculations.
>
> I have changed the files on the website to make the single precision
> ones the default.  Expanded.mdp warned about the issues with
> precision, but left the sc-r-power in place; the Ethanol.mdp did not
> warn about the potential issue.  Now both include the more efficient
> path commented out.
>
> NOTE that this means the exact numbers of the tutorials are not quite
> right anymore; the process is unchanged, as is the final answer, but
> the intermediate dG's will be slightly different.
>
> However, I need to redo them anyway to make them easier in the next
> 1-2 weeks, so I will update them then.
>
> If it still fails with double, that's a different issue -- because I'm
> running them fine with double.
>
>