[gmx-developers] Free energy calculations in 3.3_rc1

[Berk Hess]

Erik Lindahl wrote:

> Hi David,
>
> Apart from the PME free energy loops we now use lambda instead of  
> lambda^2 by default in the softcore interpolation. According to Berk  
> this results in less noise, and he's usually right :-)
>
> You can control it with the parameter sc_power - use 2 to imitate the  
> behavior of previous gromacs versions.
>
> (I think it should be documented in share/html/online/mdp_opt.html too).
>
Yes it is.

Another point is that in the older CVS code not 1/r but erfc(beta r)/r 
was soft-cored.
This could make a significant difference in dg/dl values, but it will 
not affect the integral
(except for differences within the normal sampling error margins of course).

Berk.