[gmx-developers] Wrong Ouput from g_density - PBC bug?

[Berk Hess]

Hi,

My fix was lacking one change from real to double which caused 
compilation issues.
Unfortunately two colleagues of mine reverted my change in two steps 
without consulting me.
I have pushed a proper fix to gerrit.gromacs.org to be included in the 
4.5.6 release.

Cheers,

Berk

On 05/17/2012 03:20 PM, Juan M. Vanegas wrote:
> Actually, I should have run some more tests before posting here, so it 
> turns out that there is no issue with the PBC part. The problem is 
> actually with the accumulation of the values. Originally the 
> accumulation of the densities was done with type "real", but if you 
> have gromacs compiled in single precision this array is of type float 
> and you get different results depending on the conditions because of 
> the precision. Berk Hess changed the type of the variable that 
> accumulates the densities explicitly to double sometime in March of 2011
>
> http://repo.or.cz/w/gromacs.git/commit/8c3a82a3caec5dcd04fd11c6f75d2dabbbca86fd 
>
>
> but then it got changed again to real (by mistake?) a few weeks later, 
> and in the latest git version the density storage array is still of 
> type real, but needs to be changed to double or you get spurious results.
>
> Juan