[gmx-users] PME problem in single precision

[DV]

Hello,

I am currently doing some test simulations on a charged protein. I have
solvated and neutralized it ok. I use PME.

When I simulate (MD/EM) the "Protein+Water+Counterions" everything is OK
*PROVIDED* I use mdrun_d (double precision). (Please find below part of the
mdp file)

Otherwise the MD/EM explodes:
"	Warning: Only triclinic boxes ...
" 	            Box[    0]={         NaN,          NaN,          NaN}
............
"	Fatal error: ci = -1 should be in 0 .. -1 [FILE nsgrid.c, LINE 210]

The other way is to use conventional cutoff (no PME) and single precision
and everything is *ok* again.

I have checked this on two machines, one with double and single precision
compilations of gromacs, the other with just a single precision version.

I saw similar problems in the list.

Do you have any idea why this could happen?

How do you compare the reaction field method vs PME for handling Coul
interactions?
(just in case life is not easy and all else fails :-) )

Thanks

Diego Vallejo
Instituto de Física de Líquidos y Sistemas Biológicos.
vallejo at iflysib.unlp.edu.ar

-------------------- mdp file ----------------
coulombtype 	 	= PME
rcoulomb 	 	= 0.9
rvdw 	 		= 1.4
fourierspacing 	 	= 0.12
pme_order 	 	= 4
ewald_rtol 	 	= 1e-5
---
Outgoing mail is certified Virus Free.
Checked by AVG anti-virus system (http://www.grisoft.com).
Version: 6.0.476 / Virus Database: 273 - Release Date: 24/04/2003