[gmx-users] ull code with defined negative relative displacements

[chris.neale at utoronto.ca]

Thank you very much Berk,

I'm not trying to be argumentative for arguments sake, but I just  
can't see how exactly one would make pull_geometry=direction work in  
this way. Although I think that I have shown the evidence for this in  
my posts, if I am wrong here then clearly I misunderstand something  
that is pretty important and, as such, I am keen to stick with this  
thread.

If you have the time to post the entire pull-code portion of the .mdp  
options for a pull_geometry=direction run to harmonically restrain  
group 1 at a position of 1 nm more negative z in reference to group 0  
then I will do the testing for that setup.

Nevertheless, I realize that you are very busy and may have more  
pressing priorities.

Thank you for all of your time sticking with this thread for so long,
Chris.

Hi,

I would still say pull_geometry=direction would be the easiest  
solution for this case,
but pull_geometry=position will probably do the same thing.

Berk

[Hide Quoted Text]
Date: Sun, 15 Nov 2009 17:11:43 -0500
From: chris.neale at utoronto.ca
To: gmx-users at gromacs.org
Subject: [gmx-users] pull code with defined negative relative displacements

Hi Berk,

I do not mean the box 0, and I am aware about the pull distance
needing to be less than half of the smallest box length. Please allow
me describe a bit more rigorously what I need to avoid. I will use an
example that is totally fictitious, but is designed to emphasize the
bimodal sampling issue that is relevant to my actual study.

1. Imagine that one wants to calculate the PMF for a water molecule
along the normal to an asymmetric bilayer in which one leaflet is POPC
and the other is lipidA.

2. The free energy for that water residing 0.1 nm away from the
bilayer center in the POPC leaflet (call it -0.1 nm from the bilayer
center) needs to be calculated separately from the free energy for
that water residing 0.1 nm away from the bilayer center in the lipidA
leaflet (call it +0.1 nm from the bilayer center).

3. Let the force constant be 500 kJ/mol/nm^2 and let the user apply
pull_geometry=distance and pull_init1=0.1 while starting the replica
at 0.1 nm from the bilayer center in two simulations where one takes
(a) a starting water position that is 0.1 nm toward the POPC
headgroups and the other takes (b) a starting position of the water to
be 0.1 nm toward the lipidA headgroups.

4. I posit that the probability distributions of the water about the
bilayer normal from simulations (a) and (b) in part 3 above will
converge to the same distribution and that this distribution will be
bimodal with one peak at -0.1 nm and one peak at +0.1 nm. This is
because the simulation that started at -0.1 (0.1 nm closer to the POPC
headgroups) will, even under the biasing force, infrequently cross the
bilayer center of mass and then be closer to the lipidA heagroups. At
this point. the direction of the applied force will the change to draw
the water toward the other possible location at +0.1 nm.

5a. For displacements >> 0.1 nm, this will not be a problem.
5b. For larger force constants, this will not be a problem at 0.1 nm,
but will still be a problem for some closer centers of restraint.

6. pull_geometry=distance is entirely incompatible with this approach.

7. pull_geometry=position does appear to work.


Thanks for your patience,
Chris.

-- original message --

Hi,

Sorry, but I have no clue what you mean with sampling >0
and how you would end up with a bimodal distribution.

You don't mean the box 0, do you?
That is irrelevant.
What can be a problem is that you pull distance should not be more
than half the box length.

Berk
Date: Sun, 15 Nov 2009 12:33:30 -0500
From: chris.neale at utoronto.ca
To: gmx-users at gromacs.org
Subject: [gmx-users] pull code with defined negative relative displacements
I only now noticed Justin mail on g_wham.
You can probably also use pull_geometry=distance and pull_init=1,
if you starting structure has group1 close to 1 nm below group 0.
Agreed, although this will not work when the force constant is not
strong enough to inhibit any sampling >0 -- wherein the distribution
   about 0 would become bimodal and this is something that I can not
allow.

As far as I can tell, pull_geometry=position, pull_init1<0, and
pull_vec1=0 0 0 is the solution, as per my previous message.

Thank you,
Chris.