[gmx-users] Perturbation Thermodynamic Integration

[Hannes Loeffler]

On Tue, 16 May 2017 10:28:08 -0400
Dan Gil <dan.gil9973 at gmail.com> wrote:

> Thank you for the advice on the cut-off schemes and PME methods.
> 
> What is the physical meaning of a non-interacting final state
> > that has different masses from the initial state?  
> 
> 
> These free energy options was just from me trying to figure out why
> mass has any contributions at all. I am going from molecule A to B
> described in the topology. In the actual simulations, I am planning
> on using this: vdw_lambdas              = 0 0.1 0.2 0.3 0.4 0.5 0.6
> 0.7 0.8 0.9 1 coul_lambdas             = 0 0.1 0.2 0.3 0.4 0.5 0.6
> 0.7 0.8 0.9 1 bonded_lambdas           = 0 0.1 0.2 0.3 0.4 0.5 0.6
> 0.7 0.8 0.9 1 restraint_lambdas        = 0 0.1 0.2 0.3 0.4 0.5 0.6
> 0.7 0.8 0.9 1 mass_lambdas             = 0 0.1 0.2 0.3 0.4 0.5 0.6
> 0.7 0.8 0.9 1
> 
> To get the solvation free energy difference between molecule A and B.

If you do this via decoupling ("absolute" transformation) you do that
once for molecule A and once for molecule B.  I don't see why you would
want to change masses in this case.

If you want to do this via a relative transformation you would not use
couple-moltype at all but fill in the B columns in your topology file
for all force field parameters that change.  In this case you _can_
change the masses but you would have to do the same transformation in
vacuum. Then the mass contributions should cancel perfectly (at least in
the limit of infinite sampling, I guess).  But Michael Shirts has
commented a while back that transforming the masses may interact badly
with bond constraints that are applied to alchemically transformed
bonds (we have seen problems with this too). So it is just simplest to
not use mass-lambdas precisely because of the aforementioned argument.


Cheers,
Hannes.