Re: [AMBER] TI in Amber and the TUTORIAL A9

From: Marek Maly <marek.maly.ujep.cz>
Date: Thu, 08 Dec 2011 22:48:18 +0100

Hello Thomas,
thanks again for your comments !

Meanwhile I read your interesting article which you recommended:

"Nonlinear scaling schemes for Lennard-Jones interactions in free energy
calculations"

discussing/comparing different ways how to prevent from the start/end
singularities
when particles are vdw coupled/decoupled with/from the rest of the system.
Unfortunately
I didn't find there any part directly related to our discussion.
I checked also several other TI articles but never find some real basics
preferentially
illustrated with some very simple concrete example.

As after your last email I had suspicion that my interpretation of TI is
not
fully correct. I create very simple example with just 3 particles where
one is going
to be uncharged during TI run using the simple linear mixing schema
(please see attached TI.ppt).

On the first slide there is explained my original understanding (probably
not the right one
where V is interpreted just as the potential of unique-common pairs). On
the second slide I
tried to described the whole uncharging process considering V as the total
el. potential.

The both interpretations gives the same expression for the total
lambda-dependend el. energy and it's lambda derivative
(so there is really just difference in V interpretation). But after all I
am still not sure if
any of that interpretations are OK as for example the need for two sander
threads remain unclear
to me considering my actual understanding of how this uncharging process
might be realized
in practice.

Please check it and or correct directly in given ppt document or just
comment in email.
I think it is just the only way how to be able to "speak the same
language" ...

Thank you very much in advance !

   Best wishes,

       Marek




Dne Tue, 06 Dec 2011 10:41:53 +0100 <steinbrt.rci.rutgers.edu> napsal/-a:

> Hi,
>
>> I think it is clear that for example "disappearing" of some molecule X
>> from the solvent brings solvent
>> molecules from neighbourhood of X in closer distances as the space which
>> was previously occupied by
>> molecule X is now more and more free and hence "common"-"common"
>> interaction (here solvent-solvent) changes here dramatically
>> I think.
>
> I am not sure what you mean, but aren't you thinking too complicated
> here?
> What you describe certainly occurs in the simulation, but it is not part
> of the derivative dVdl. Your complete potential energy function
> (including
> *all* common/common, common/SC etc. interactions) can be written out as
> V(lambda). It uses linear mixing for anything non-SC and you can
> therefore
> write down the lambda-derivative of the total potential energy. Again,
> all
> this is written out in our papers (hopefully) clearer than I can make it
> here.
>
>> So I see V(L) in two different contexts.
>>
>> 1) As the L-weighted interaction potential between "unique"-"common"
>> atoms
>> which has analytical formula
>> and so also analytical formula for dV/dL.
>>
>> 2) As the total potential energy of the whole system which is composed
>> of
>> "unique"-"common", "common"-"common", but also
>> "unique"-"unique" atom pairs interactions. Unfortunately for example
>> the
>> dependence of "common"-"common" part of the total potential
>> energy on L is nontrivial and might to be pretty hard to express it
>> analytically as well as it's derivation.
>
> I think (1) has no specific meaning other than being part of your total
> potential energy. (2) is what you want and its dependence on lambda is
> indeed not hard to express analytically.
>
>> From your response is it clear to me, that the values of dV/dL(L) are
>> computed just with respect to the analytically
>> well defined part ("unique"-"common" interacations) of the total
>> potential energy in another words derivated V(L) has here meaning 1).
>
> no, that is not the case in the actual code (and should not be)
>
>> I will for sure check the relevant article and the code as you advised
>> but
>> for this
>> moment I still do not see the necessity of two simultaneous sander runs
>> in
>> case on
>> just one way TI processes (i.e. only charging or only uncharging or only
>> vdw coupling or only vdw decoupling).
>
> There is no necessity for two sander processes (even though it's
> convenient) but you need to do two full energy/force calculations, one
> for
> your start state, one for your end state. Since e.g. Ewald electrostatics
> are not easily pairwise decomposable, changing even one partial charge
> means you need to do the full calculation twice. Now if you have to call
> energy() twice anyway, you might as well do it on two processes to
> improve
> calculation speed.
>
>> Here in principle just one PRMTOP should be enough if I am able to
>> provide the rest information
>> (definition of "unique" atoms and type of the TI change so to define V0
>> and V1 state) as a part of *.in file. For this
>> just one mask (like SCMASK) and another parameter which define one of
>> the
>> 4 possible TI changes
>> should be enough. Is still mystery for me what the second sander thread
>> is
>> calculating in such one way cases
>> but maybe after reading that article and checking that *.f file
>> everything
>> will be clear ( hopefully :)) ).
>
> I believe that you could write a TI code that uses only one prmtop and
> some extra info on the transformation, but I think it would not provide
> any big benefit to the user.
>
> Kind Regards,
>
> Dr. Thomas Steinbrecher
> formerly at the
> BioMaps Institute
> Rutgers University
> 610 Taylor Rd.
> Piscataway, NJ 08854
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
> __________ Informace od ESET NOD32 Antivirus, verze databaze 6687
> (20111206) __________
>
> Tuto zpravu proveril ESET NOD32 Antivirus.
>
> http://www.eset.cz
>
>
>


-- 
Tato zpráva byla vytvořena převratným poštovním klientem Opery:  
http://www.opera.com/mail/



_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber



Received on Thu Dec 08 2011 - 14:00:03 PST
Custom Search