ok, my own personal views, which might not be popular...
Lots of people have tried to do FEP or TI between a QM and an MM
ensemble. Almost everyone of them has failed miserably...
There are some very particular problems where it 'might' work, but in
general, it does not. This is simply because for most interesting
problems, the phase space sampled by QM and MM are not really
overlapping, which makes FEP and TI somewhere between wrong and
impossible to converge.
Warshel's ideas are usually done within an EVB framework, where lots of
assumptions are hidden from plain view, including entropic effects,
differential phase space, etc.
You would be hard pressed to find anyone not from Warshel's group, using
those ideas. At their core, they use a second order expansion of the FEP
formulas, which assumes that the fluctuations in the potentials are
correlated are mostly gaussian.
Lee Woodcock (U south florida) has recently proposed some pretty clever
reweighing ideas to solve this problems, but I would consider them very
experimental at this point.
As far as how to do this technically, computing energies mixing QMm and
MM versus lambda is 'easy'. This should make FEP 'easy' to implement,
even trough a script outside sander. Mixing forces, which is what you
would need for TI, is much harder and not obviously cheaper than just
doing the full calculation in pure QM.
The work by Ross Walker group on adaptive boundaries for QM/MM must have
somewhere a calculation of mixed QM and MM methods, including the
forces. Maybe looking at that code (sander I think), one could extend to TI.
Adrian
On 2/22/16 10:54 AM, Gerald Monard wrote:
> Hi,
>
> In section 22.1.1, it is said "The input (mdin) and starting coordinate
> files must be the same for the two groups.", which makes me thought that
> one could not mix MM and QM/MM.
> I have just tested it (pentadiene test), and having different mdin's
> does not seem to be enforced. Thus, one can indeed mix MM and QM/MM
> using TI.
>
> Thanks,
>
> Gerald.
>
> On 02/22/2016 04:30 PM, David A Case wrote:
>> On Mon, Feb 22, 2016, Ulf Ryde wrote:
>>
>>> Is there any possibility to simulate in Amber with a potential that is a
>>> mixture of pure MM and QM/MM (i.e. of the type
>>> E = (1-k) E_MM + k E_QM/MM
>>> where k is an adjustable parameter (like lambda in standard free-energy
>>> perturbations),
>>> so that you can do free-energy perturbations along this parameter, going
>>> from MM to QM/MM.
>>> Such a potential is employed in Warshel's paradynamics approach (J Phys
>>> Chem B 115 (2011) 7950), which is one of the most promising ways to
>>> include high-level QM in free-energy predictions with a MM reference
>>> potential and a minimum of QM calculations.
>>>
>>> I do not find such an approach in the manual, but I suppose it is simple
>>> to implement for somebody acquainted with with the QM/MM code.
>> This is already implemented--at least I think so, although I have never
>> actually done it myself....
>>
>> In sander, thermodynamic integration is implemented by creating two prmtop
>> files, one for lambda=0 (which would be for E_MM) and one for lambda=1 (which
>> would be for E_QM_MM). Then the standard lambda variable is the "k" in the
>> equation above.
>>
>> In practice, the two prmtop files themselves are actually identical, since
>> turning QM/MM on or off is determined by the ifqnt flag in the mdin file.
>> See section 22.1.1 in the Amber 2015 Reference Manual. Your groupfile
>> would look like the one given there, except that prmtop.0 and prmtop.1
>> could be the same file, but the two "mdin" file would be different, with
>> mdin.0 setting ifqnt=0 and mdin.1 setting ifqnt=1. The mdin.1 file would
>> also include the other qm/mm-related options that are relevant when ifqnt
>> is set to 1. Everything else should proceed as described in Section
>> 22.1.4.
>>
>> Maybe someone on the list with more direct experience can chime in here. But
>> the sort of calculation you describe is *supposed* to be straightforward, and
>> was one of the reasons that TI in sander was implemented in the way that it
>> was.
>>
>> ....regards...dave
>>
>>
>> _______________________________________________
>> AMBER mailing list
>> AMBER.ambermd.org
>> http://lists.ambermd.org/mailman/listinfo/amber
>>
--
Dr. Adrian E. Roitberg
Professor.
Department of Chemistry
University of Florida
roitberg.ufl.edu
352-392-6972
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Feb 22 2016 - 08:30:03 PST
