[gmx-developers] Warnings

[Berk Hess]

Roland Schulz wrote:
>
>
> On Sat, Aug 9, 2008 at 5:04 AM, <hessb at mpip-mainz.mpg.de 
> <mailto:hessb at mpip-mainz.mpg.de>> wrote:
>
>
>
>     >
>     > Regarind the thermostat settings I'm not sure whether one should
>     give
>     > warnings, messages on a info level or a link to the thermostat
>     wiki page
>     > to
>     > encourage people to check these parameters.
>     > The thinks which are often unclear I think are:
>     > - tc_grps: I think often system is used but the wiki page recommends
>     > Protein
>     > Non-Protein  (for the Hot-Solute Cold-Solvent problem)
>     > - Berendsen: It is not well known that it is not NVT/NPT
>     > - tau_t is selected not careful
>     >
>     > I'm happy to do it. Just want to make sure that you agree and
>     I'm not
>     > changing the way warnings should work and don't add warnings
>     because of a
>     > misunderstand on my side. If you have others to add I can
>     quickly do that
>     > too.
>     >
>     > Roland
>
>     We should only generate warnings when we are completely sure that
>     certain combinations would not be useful for certain types of systems.
>     For "standard" protein in solvent simulations one could probably
>     add more checks, but one can simulate any type of system under
>     very varying conditions in Gromacs.
>     I have already removed several warnings in the past year,
>     because some combinations were fine certain cases.
>
>
> My understanding was, that a warning should be given, when it is very 
> uncommon choice unlikely to be a good idea. If it is wrong for certain 
> I thought grompp should give an error.
I agree.
But is should be very uncommon for any type of system.
For example, many people are now also using Gromacs for coarse grained 
simulations.
Here one my be using sigma-epsilon units which makes any absolute value 
check impossible.

I think David's idea of providing proper example md files is a better 
way of helping
people avoid stupid mistakes. Now people just copy mdp files from a 
"random" location.
>
>
>
>     Also one can not check parameter values, since these can vary
>     a lot between different systems.
>
>     Note that in the new version you already get a fatal error with
>     a single warning. This forces people to look at the warnings.
>     Also the note count at the end helps people to actually look
>     for the notes.
>
>     We should add a warning when the comm groups together do not cover
>     the whole system. In some cases one might actually want to do this,
>     but these are so exotic that a warning is fine.
>     When does one get infinity? With multiple groups?
>     Or when some atoms do not belong to any group?
>
>
> It is when atoms do not belong to any group. I'm not sure whether 
> there are also other cases.
>
> Roland
I have fixed this problem a few months ago.
Or is there still a problem like that present in the current CVS?

Berk.