[gmx-users] problem with exchange probabilities in remd

Mark Abraham Mark.Abraham at anu.edu.au
Sun Feb 24 03:34:51 CET 2008

> Hi all,
> I am performing a remd simulation of a peptide+water system. I have tried different temperature distributions. I am using an NVT ensemble. My temperature coupling constant is 0.1 ps.
> The problem is that upon increasing the temperature difference between the two replicas more than two (290==>294), the exchange probability decreases sharply, app. to 0.03. 

Yep. That's normal for decent-sized explicit-solvent REMD. You can 
choose a lowest temperature spacing that isn't an integer, of course.

> I have changed the exchange frequency in order to get a better exchange probabilities. However, it has not affected the overall exchange probabilities so much. 

The frequency is independent of the probability, at least until you get 
to periods so short that successive attempts are correlated. See Periole 
and Mark, J Chem Phys 126 014903, and for a counter-view, Sindhikara, 
Meng and Roitberg J Chem Phys 128 024103. Both frequency and probability 
affect the exchange acceptance rate, of course.

> Is there anybody who has encountered such a situation?

Yep. Consider seriously how big a temperature range you might need. 
There's discussion out there that the increased-sampling effect of the 
increased temperature range is outweighed by the ability of the replicas 
   to sort themselves as a function of T. See Periole paper above. Thus, 
there's no need to go to 800K, like you see implicit-solvent REMD 
achieve. 373K is enough to start denaturing proteins in cooking, after 
all :-)


More information about the gromacs.org_gmx-users mailing list