[gmx-users] programmes to have in double precision besides mdrun_d

[Berk Hess]

> Date: Wed, 17 Jun 2009 15:17:53 +0200
> From: erikm at xray.bmc.uu.se
> To: gmx-users at gromacs.org
> Subject: Re: [gmx-users] programmes to have in double precision besides mdrun_d
> 
> Jussi Lehtola skrev:
> > On Wed, 2009-06-17 at 13:45 +0200, Berk Hess wrote:
> >   
> >> Hi,
> >>
> >> Nothing needs to be double precision.
> >>
> >> Why do you want mdrun in double precision?
> >> The only common reason for this is normal mode analysis,
> >> in which case you need all the tools involved in double precision.
> >> For normal MD simulation there is nearly never a need for
> >> double precision.
> >>     
> >
> > If you want to run accurate NVE simulations, double precision is
> > important. If you use thermostats, then there is no need for double
> > precision.
> >
> >   
> I can testify to that. Without double precision I've had problems with 
> drifting total energy when doing non-periodic NVE.
> 

Indeed, this is correct.
But I guess this falls into the category nearly never.
On the other hand, there are so many Gromacs users nowadays,
there nearly never is nearly never zero users.

> /Erik
> > Also, I have found that double precision can be nice for energy
> > minimization since it can handle more pathological cases than single
> > precision. Still, usually it doesn't matter much whether the starting
> > point was prepared in single or double precision.
> >   


That's true.
It would be nice if we could make the single precision code
handle such cases better (force capping?).

Berk



_________________________________________________________________
Express yourself instantly with MSN Messenger! Download today it's FREE!
http://messenger.msn.click-url.com/go/onm00200471ave/direct/01/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20090617/1473e7de/attachment.html>