[gmx-users] pull-code

[Thomas Schlesier]

Yes you are right, should be between 0 and >0.

Do you have a window for a distance equal 0?

This window should behave similar to the RDF-analysis. Because there are 
"no directions".
Or to reformulate the problem.
We make an umbrella window for a distance of 1. If particle stays there 
everything is fine. If particle moves to 0, it should be also fine 
(particle sees a force of k*1). If paticle moves to -1, it should see a 
force of k*2, but instead, the distance is 0 -> no force.
If you have the umbrella window centered at 0 this problem vansihes -> 
if particle move it sees always a force.

But one thing gives me headaches. I don't have this problem in my 
pulling simulation, because the distance between my reference and pulled 
group can not become zero:
But concerning the reaction coordinate it will have a similar flaw like 
the RDF i think: It doesn't matter in wish direction the particle moves 
(left or right) due to the distance we would always say it moves along 
the reaction coordinate. In reality it moves sometimes in the negative 
direction of the reaction coordinate, but we always say it's a positve 
distance -> so positive value on the reaction coordinate.
For an isotropic system this would not matter, but for system which we 
have a anisotrop reaction coordinate it should matter.

Greetings
Thomas



Am 17.02.2012 17:07, schrieb gmx-users-request at gromacs.org:
> Hi Thomas Many thanks for the reply again. At larger distances the two
> curves match up quite well. The curve from the reversible work theorem
> is better behaved and smoother but this could be solely due to
> statistics. I am slightly confused about your statement "If the small
> circle moves between 0 and any value <0 everything should be fine." Do
> you not mean 0 and any value >0 ? Cheers Gavin Thomas Schlesier wrote:
>> >  Hi Gavin,
>> >  if i remember correctly it was a system about pulling a ligand from a
>> >  binding pocket?
>> >  To make the system simpler we have a big circle and in the middle a
>> >  small circle. And we assume that the potential minimum for the
>> >  interaction between both circles is when the small cirlce is in the
>> >  middle of the large circle.
>> >  Now we do the Umbrella sampling. For a window which is centered at a
>> >  distance which is sligthly greater then 0, we will get problems.
>> >  Assume small circle is sligthly shifted to the right. And the other
>> >  windows are also in this dircetion. (->  reaction coordinate goes from
>> >  zero to the right dircetion)
>> >  If the small circle moves between 0 and any value<0 everythig should
>> >  be fine. But if the small circle moves to the left, we will also get a
>> >  positive distance. Problem is from the above defined reaction
>> >  coordinate it should be a negative distance. So we are counting the
>> >  positive distances too much.
>> >  To check this, you could use*g_dist*  to calculate the distance for
>> >  both molecules for the problematic windows. Then project the resulting
>> >  vector onto your reaction coordinate. Then you should see the
>> >  crossings between the right and left side.
>> >
>> >  How do the two free energy curves compare for larger distances, where
>> >  you can be sure, that you do not have this 'crossing problem'?
>> >
>> >  Greetings
>> >  Thomas
>> >
>> >
>> >
>> >  ---------------------------------------------------------------------------------
>> >
>> >
>> >
>> >  Hi all
>> >
>> >  I am returning to a query I had a few weeks ago regarding a discrepancy
>> >  between two free energy curves. One calculated using umbrella sampling,
>> >  the other calculated via the reversible work theorem from the RDF. There
>> >  is sufficient sampling of the dynamics in the RDF so this method is
>> >  viable.
>> >  Anyway in the pull-code I use pull_geometry = dist and pull_dim=Y Y Y.
>> >  The free energy curve from the pull-code method does not give me a
>> >  minimum at the zero value of the order parameter whereas the RDF method
>> >  does. Someone said before about double counting of positive distances at
>> >  small values of the order parameter and therefore information is lost at
>> >  very small distances.
>> >
>> >  Is this correct?
>> >  I am slightly concerned that my curves are not giving me the correct
>> >  information involving a very important state in my reaction coordinate.
>> >
>> >  Also when this dist restraint (which cannot be negative) is implemented
>> >  are there issues with the normalisation of the histograms from g_wham?
>> >
>> >  Cheers
>> >
>> >  Gavin