[gmx-users] PMF in vacuum and pull direction

chris.neale at utoronto.ca chris.neale at utoronto.ca
Wed Jun 9 05:16:10 CEST 2010

Dear Eudes:

Thanks for working with me on this one! What you are getting is  
actually what you are asking for in your .mdp file. You restrain the  
Z-projection of the displacement, but do not affect the X and Y  
components of the displacement with pull_dim = N N Y. If you want to  
pull with separation only in z and want the x and y components of the  
displacement to be close to zero, then you want pull_dim = Y Y Y.

Let's be clear: you say "I should use the components distance (N N Y )  
with direction geometry to get a pulling in a direction z! Please  
correct me if it is wrong.."

And you are literally correct that you should set pull_dim = N N Y if  
you only want to pull in z, by which we mean that you only want to  
apply a biasing force based on the z component of the displacement.  
But what it seems that you want to do is to "pull only in z" but also  
to add a constant bias in x and y such that their displacement  
component remains close to zero.

Now I'm not entirely sure what you want to do, but to me the following  
seems like it will do what you want:

; Pull code
pull = umbrella
pull_geometry = direction
pull_dim = Y Y Y
pull_start = yes
pull_ngroups = 1
pull_group0 = BEN1
pull_group1 = BEN2
pull_vec1 = 0 0 1
pull_rate1 = 0.001
pull_k1 = 1600

Note that I only changed your pull_dim.


-- original message --

Hello Chris, thanks for the reply. Below his message commented walkthrough.
In your email you asked me for quantitative results. All I have so far is
a visual inspection of the pulling trajectory. All I can see are
the two molecules moving linearly away from each other in a rectangular box
of dimensions 3x3x6nm. However this separation does not occur on the z
as I wanted, but in a diagonal way across the faces of the box.


> 1. There is no need to use a virtual atom, the pull code will use the
> center of mass. I strongly suggest that you stop using a virtual atom
> and just use the entire benzene as an argument to the pull code group.
> I have had difficulties with slightly more complicated setups of this
> type.

Ok, now I'm using only the center of mass of the molecules, no virtual

> 2. In test2.jpg, the system without pbc shows a flat PMF after the
> cut-off -- exactly what one would expect. The pbc system shows
> continued interaction -- again what I would expect. So there is
> nothing actually all that strange here. One would not expect to see
> such a drastic difference in a high dielectric such as water, but in
> vacuum I suspect that this is expected.

Alright! In fact the well of both PMFs are not so different.

> 3. Please clarify what your cutoff was. I don't see a cut-off listed
> in your .mdp options and leaving this to the default of 1.0 nm is a
> bad idea because it can lead to confusion a times like this. I might
> assume that it was 0.7 nm based on test2.jpg, but then see the point
> #3 below.

Apesar de eu nao ter incluido o valor do cutoff na mensagem anterior
eu usei 1.2 nm
Although I have not included the cutoff value in previous message
I used it as 1.2 nm.

> 4a. I have no idea what -DPOSRES is actually doing for you since I
> can't see your topology.

A ideia de usar POSRES se referia ao emprego dos sitios virtuais.
Assim eu restringia a distancia das moleculas atraves da restricao do
sitios virtuais sobre seus centros.
The idea of using POSRES is referred to the employment of the virtual sites.
So I restricted the distance of molecules by restricting virtual sites
on their centers.

> 4b. Are you sure that "pull_dim = N N Y" is really what you want?
> Sometimes one wants to average over X and Y, but I am not sure that
> you do in this case.

I want to separate the two molecules from one another so that the
displacement is restricted to the z axis. I can be wrong,
but I think to do it in water, using PBC, I should use the
components distance (N N Y ) with direction geometry to get a
pulling in a direction z! Please correct me if it is wrong.

> 4c. What exactly do you believe pull_r0 and pull_r1 are doing for you?

Here they are worthless, because I have not used the option "cylinder".
I should have removed these options from the original message.

> ### More minor notes:
> 5. regarding test1.jpg: a PMF is correct to an additive constant,
> meaning that you can shift two PMFs relative to one another. These 2
> PMFs are therefore less different than they appear in your compaison
> plot, but they do differ in the slope between 1.0 - 2.0 nm. This is
> probably just a convergence issue and you will always need to do tests
> like this.


> 6. regarding histo.png: can you confirm that the few very short
> gaussians are due to less sampling in a few windows? In any event, the
> overlap looks good.

Yes, there were two histograms whose simulations had crashed for some
It was corrected.

More information about the gromacs.org_gmx-users mailing list