[gmx-users] pull code, lambda, and soft core availability

[chris.neale at utoronto.ca]

Upon further investigation, it appears that a soft core may help in  
this case. This is in spite of the fact that the problem is not a  
singularity, but that the dgdl term (dgdl=0.5*(Fc_B-Fc_A)*(r-r0)^2)  
gives very large values for Fc_A large enough to be significantly  
restraining and Fc_B = zero, where the sampling in the B state is  
massively unlikely to occur in the A state. I had originally thought  
that the lambda parameter would get included in dgdl  
(http://www.gromacs.org/pipermail/gmx-developers/2009-February/003004.html)  
but that is obviously not the case.

I have found an alternative solution, but in case anyone else wants to  
follow this path in the future, it is possible that a lambda squared  
dependence would give improved behaviour since lambda would then be  
included in the dgdl equation. My quite possibly incorrect attempt at  
differentiation yields dgdl=(lambda-0.5)*Fc_A*(r-r0)^2 in this case  
where Fc_B=0.0, but in any event the (1-lambda)^2 term will ensure  
that lambda hangs around in some form, and this should reduce the  
uncertainties that occur near and at the no-restraint state.

Chris.

-- original message --

Thank you Matt, that is indeed the perfect paper. I'll post back here
once I figure out exactly where the (1-lambda)^2 term would fit into
the pull code US equation.

Chris.

This seems to be a good reference for soft-core interactions.

BEUTLER et al. AVOIDING SINGULARITIES AND NUMERICAL INSTABILITIES IN
FREE-ENERGY CALCULATIONS BASED ON MOLECULAR SIMULATIONS. Chem Phys
Lett (1994) vol. 222 pp. 529-539

On Thu, Feb 12, 2009 at 1:33 PM,  <chris.neale at utoronto.ca> wrote:
   Does anybody know of a good paper that
> describes the underlying equations that define soft core?
>
> Thanks,
> Chris.