[gmx-users] single X double precision

[David van der Spoel]

On Fri, 2003-03-07 at 08:26, Anton Feenstra wrote:
> Luciane Vieira de Mello wrote:
> > Dear all,
> > 
> > It is maybe a naive question, but since I am reinstalling Gromacs in my
> > machine, it would be nice to better understand the difference in results
> > we get when using a single or double precision. Is double better but
> > slower? I get the impression we hardly use double precision, by the bit "
> > you never know when you might need.." in the sentence:
> 
> Double precision is indeed slower, but it is not better per se. For some
> specific applications the additional precision is needed (e.g. for normal
> modes analysis), but for regular EM and MD, there are other sources of
> error that are much worse than the limit of the single precision.
And also when using shell MD (polarizable) you need DP for optimization.
 
> It is good, also, to remember that usually MD simulations are run at
> temperatures of 300K (or more). That means there is a lot of thermal noise
> in the system already, the added 'noise' from single precision errors is
> totally neglectable in comparison.
More in particular: when using cut-offs rather than PME you don't have
energy conervation anyway. Otherwise things like Temperature/Pressure
coupling influence the system too.



-- 
Groeten, David.
________________________________________________________________________
Dr. David van der Spoel, 	Dept. of Cell & Mol. Biology
Husargatan 3, Box 596,  	75124 Uppsala, Sweden
phone:	46 18 471 4205		fax: 46 18 511 755
spoel at xray.bmc.uu.se	spoel at gromacs.org   http://xray.bmc.uu.se/~spoel
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++