Re: Handling of nonbonded lists in PSF - please clarify

From: Jérôme Hénin (
Date: Sun Feb 25 2018 - 15:10:11 CST

Hi Randy,

On 25 February 2018 at 21:42, Randy J. Zauhar <>

> 8 atoms total
> atoms 1 is unchanging
> atoms 2-3 are final
> atoms 4-6 are initial
> atoms 7-8 are ‘unchanging
> Looking at your alchemify code I would expect the first two lines to be
> 6 !NNB
> 2 3 2 3 2 3
> corresponding to 3 copies of the 2 final atom indices, one set of final
> atoms to be assigned to each of the initial atoms.
> According to the alchemify code (and an actual run on my own system) it
> seems the resulting ‘second section’ would look like this :
> 0 0 0 2 4 6 6 6
> Here, the each of the eight atoms is assigned an offset of some sort into
> the exclusion lists.
> The atoms before the first ‘initial’ atom have 0 offset into the list -
> does that mean the list is being ignored, or it’s being used and we start
> at position 0 (i.e. atoms 2 & 3)?

As long as the index is zero, we're not moving in the list, hence not
adding exclusions to the first 3 atoms.

> The first initial atom has offset 2, which means what? Do we skip the
> first set of two atoms in the exclusion list? In any event, I guess this
> implies it cannot see atoms 2 and 3, which makes sense.

Offset moves from 0 to 2, meaning the first two atoms of the first list (2
3) are being excluded from atom 4. Then we move from 2 to 4, meaning the
next two atoms in the list (again 2 3) are excluded from atom 5. Rince and
repeat for atom 6. The the index stops changing and stays at 6, meaning
we're not adding any further exclusions - as you mentioned, this is the end
of the first list anyway.

> The last ‘initial’ atom gets an offset of 6, meaning what? By my
> interpretation above that would be off the end of the list of exclusions.
> And that setting continues for all the remaining atoms.


> The mind reels.

One of the most expressive statements I've read about this file format.

> I will assume there is some logic in NAMD under which all this makes
> sense.

This predates NAMD: it was inherited from CHARMM, or perhaps XPLOR. I'm
sure it made perfect sense back in the 80's.

The reason I am asking - I was not able to get my ‘mixed’ simulation with
> site solvent as ‘final’ atom set and ligand as ‘initial’ atom set to work,
> it blows up with nonbonded interactions clearly not being excluded. I have
> tried using FEP but with fixed lambda, and without FEP by using an
> ‘alchemified’ PSF, and in both cases the same bad result. I am trying to
> troubleshoot.

That sounds messy. Again, in principle the alchemified PSF should not make
a difference - and it did seem not to make one when you tried.

One more thing to consider: at one end point you'll have ghost water
molecules that might diffuse anywhere. When you start coupling them, you
may run into trouble. This might require some specific restraints.


This archive was generated by hypermail 2.1.6 : Mon Dec 31 2018 - 23:20:53 CST