Thermodynamic integration questions

From: pepe barrientos (confitarlaburra_at_gmail.com)
Date: Tue Dec 06 2011 - 10:39:38 CST

Dear NAMD Developers and community:

I am planning on running some solvation and binding free energy
calculations of a lipopolysacharide (lipid A). I ran some preliminary tests
using TI and the default values for the annihilation of lipidA. For the
uncharge process, there are no issues at all, but for the VdW component I
 observed the "so called" end point problem (from L 0.9 to 1.0. dVdW/dL
 i.e. molecule completely annihilated), even though I am using the soft
core VdW potential (VdWshift 5.0). By looking at the MDs, I can clearly see
that a lot of lipids tails (lipid A has around 300 atoms ) are overlapping
with solvent or protein molecules. (I have restrained one atom of lipid A,
so afterwards I can use a correction for introducing such restrain) which
could explain the rising of the dVdW/dl towards the end of the simulation.
I thought that one way of solving this, was just using a bigger value for
the "softness" (VdWshift 10). This was ,in fact, a solution for the
dVdW/dl issue, but now at lower values of Lambda (< 0.5) while I still have
the electrostatics coupled, the simulation becomes unstable due to an
increase of the Coulombic energies between soft and solvent atoms (very
small VdW repulsive forces, so electrostatics goes to -infinity).

Well after this explanation, my questions are three:

Would you advice to just run the electrostatic and VdW decoupling
separately, or would it be more efficient to add more windows as Lambda
approaches to 1?

How exactly are the dVdW/dL and dE/dL calculated in the TI framework? By
looking at the source code (I am not that familiar with C++), it seems that
instead of using the analytical derivatives, it is just calculating the
values numerically (calculating the slope using previous and forward lambda
values) however I am not 100% sure but the line 02571 in (
http://www.ks.uiuc.edu/Research/namd/doxygen/Controller_8C-source.html) is
pointing out in that direction.

Another possible "bug" that I have found is that the parameter "*alchLambda2"
*is required for TI calculations, even though it is not documented in the
manual. Does it have any effect?

Best

Antonio Garate PhD.

This archive was generated by hypermail 2.1.6 : Mon Dec 31 2012 - 23:21:03 CST