[gmx-users] Umbrella - Force constant

[Thomas Schlesier]

It also depends in some cases strongly on the system.
I have a two-state system in which both states are rather narrow (doing 
a normal pulling simulation, the end-to-end-distance seems nearly 
constant). In these two regions one could use small force constants. but 
both state are seperated by a distance of more then 0.5 nm.
So if one would sample the transition state (between both states) 
accurately one would need really high force constants...
Was too lazy to determine the right force constant for every region, so 
used in the end TI (thermodynamic integration), which one could describe 
as umbrella sampling with constraints.


Am 03.07.2012 16:49, schrieb gmx-users-request at gromacs.org:
> On Tue, Jul 3, 2012 at 2:04 PM, Justin A. Lemkul<jalemkul at vt.edu>  wrote:
>> >
>> >
>> >  On 7/3/12 8:41 AM, Steven Neumann wrote:
>>> >>
>>> >>  Dear Gmx Users,
>>> >>
>>> >>  Do you know or can you suggest some results based on the comparison of
>>> >>  the force constant in Umbrell Sampling? Any literature?
>>> >>
>> >
>> >  That would be lovely, but I've never seen such a thing.  One could probably
>> >  write a book with all the test cases that would be required.  My gut tells
>> >  me that you can't generalize too much in terms of pulling simulations - the
>> >  approach depends on what is being pulled (small molecule, peptide, large
>> >  protein), what the medium is (water, membrane, etc), and what the
>> >  interacting partner is (protein surface, ion channel, binding pocket).
>> >
>> >
>>> >>  As far as I understand when you use the same staring coordinates (from
>>> >>  the same pulling simulation) for windows but you just change the force
>>> >>  constant (e.g. from 500 to 2000 kJ/mol nm2) you should increase number
>>> >>  of windows (for f=2000) as smaller force constant will cover wider
>>> >>  neigboruring distances - that makes sense.
>>> >>  I am curious whether the final result will be the same? I guess with
>>> >>  stronger force it will converge faster but more windows are required.
>>> >>  is it the only one difference?
>>> >>
>> >
>> >  Without a systematic comparison, it's hard to say, but in theory if one
>> >  samples sufficiently and has good overlap between neighboring windows, the
>> >  results should converge to the same answer.
>> >
>> >  If someone knows of some applicable literature that has done such
>> >  comparisons, please post a reference.  I'd love to see it.  Most SMD and US
>> >  methodology is written with hand-waving explanations as to what the authors
>> >  did and why it worked, and I have a suspicion that most reviewers don't have
>> >  a better idea so they can't refute such claims.
>> >
>> >  -Justin
> Thanks Justin. The only thing I found:
>
> Beno?t Roux, The calculation of the potential of mean force using
> computer simulations, Computer Physics Communications, Volume 91,
> Issues 1?3, 2 September 1995, Pages 275-282, ISSN 0010-4655,
> 10.1016/0010-4655(95)00053-I.
> (http://www.sciencedirect.com/science/article/pii/001046559500053I):
>
> " Furthermore, the convergence properties of umbrella
> sampling calculations may be exploited more
> effectively using WHAM. Generating short umbrella
> sampling simulations for a large number of narrow
> windows is computationally more advantageous than
> generating longer simulations with a smaller number
> of wider windows (...) If Nw simulations are used to cover the
> whole range L, the force constant K of the umbrella
> sampling potential must be chosen to insure a proper
> overlap between the adjacent windows, i.e., each
> window should cover a range of dL = L/Nw and the
> value of K should be on the order of kBT/dL^2  based
> on the magnitude of the rms fluctuations. It follows
> that the total simulation time Trot needed to generate
> the Nw windows varies as ,~ L2/Nw. Thus, it is more
> advantageous to run short umbrella sampling simulations
> for a large number of narrow windows (the
> simulation time required to prepare and equilibrate
> the windows, which is also important, is ignored in
> this simple analysis)"
>
> If anyone would find something, please post. that is an interesting
> gap in free energy calculations.
>