Welcome to the part of QM/MM that resembles black magic !
We have seen things like that
A possible solution, if you REALLY know your reaction and are willing to
intervene, is to put a very soft, flat restraint, in the bonds that are
going bad but you know they should stay put. I presume you understand
what I mean by flat restraint. Just add it to the nmropt data in a way
that does not allow the atoms to become too close or to far away.
Adrian
On 9/2/16 11:46 AM, Fabrício Bracht wrote:
> Hmmm. Glad to hear that. It makes life a bit easier. Now, I posted a
> question yesterday on the list asking about the use of shake on qm regions.
> I understand that it might seem like a very naive question, but it
> something to do with what we are talking now. Would it be wrong to assume
> that the only atoms that need not to be shaken are the ones involved
> directly in the reaction? I know that once you specify the qm atoms and
> state qmshake = 0 you cannot use shake on any of those atoms. The
> noshakemask has to list all the atoms in the qm region, right? But, what
> would happen if you applied restraints to all the atoms that are not
> directly involved in the reaction.
> I ask this, because I have noticed that, sometimes, a hydrogen from a
> carbon that is adjacent to the group participating in the reaction (for
> example, HB2 linked to CB2 next to the reacting carboxylate of a aspartate
> residue) is the one that explodes first.
>
> Fabrício
>
> 2016-09-02 14:37 GMT-03:00 Adrian Roitberg <roitberg.ufl.edu>:
>
>> Hi Fabricio.
>>
>> Indeed, we have found the same. By the way, this is common in any type
>> of QM/MM calculations, not just umbrella. QM is very finicky and can
>> blow up when some coordinates are just a little bit off.
>>
>> However, when we encounter the issues you found, we do not go all out
>> and minimize + change T slowly. Usually, a VERY small number of
>> minimization steps is enough, since the problem is mostly localized in a
>> single bad contact.
>>
>> Adrian
>>
>>
>>
>> On 9/2/16 11:34 AM, Fabrício Bracht wrote:
>>> Indeed it is. But I have been doing some testings, and was about to ask
>>> this in the list myself. When doing reaction paths with umbrella
>> sampling,
>>> I have found that the coordinates with high energy values on the reaction
>>> path have a really (really really) high tendency on exploding. I tried
>>> loads of different starting configurations and found that, in most cases,
>>> adding a previous minimization, with the umbrella restraints on, solved
>>> this problem. I did this in two different ways. The first was adding a
>>> nmropt coupling to slowly increase the temperature, and the second I only
>>> set the temperature to the final value I wanted and started the
>> simulation
>>> anyways. See that in both cases the simulation time for the QM/MM
>> umbrella
>>> window had to be increased to accommodate the thermalization. I didn't
>>> expect that the simulation with no slow thermalization to finish, but,
>> for
>>> my surprise, those were the ones that worked without any errors. By my
>>> calculations, I lost less time doing this minimization procedure and
>>> increasing the simulation time than repeatedly restarting crashed
>>> simulations.
>>> I did this for a reaction I had already calculated the PMF and the
>> results
>>> were very, very similar. The reaction barrier increased by 0.2 kcal/mol
>> and
>>> the deltaPMF(final-initial) increased by 0.2.
>>> I did this by sheer curiosity, as I had already done this simulation and
>> I
>>> knew the system quite well. But I found the results quite interesting.
>>> What do you guys think?
>>>
>>> Fabrício
>>>
>>> 2016-09-02 14:09 GMT-03:00 Adrian Roitberg <roitberg.ufl.edu>:
>>>
>>>> no, there is not, but this is not a small issue !
>>>>
>>>> But, if you eventually want to run QM/MM MD, as Gustavo said, you will
>>>> have to reheat, re-relax the system at the QM/MM level, which is much
>>>> more expensive than the MM one.
>>>>
>>>> adrian
>>>>
>>>>
>>>>
>>>> On 9/2/16 11:03 AM, Fabrício Bracht wrote:
>>>>> Is there a reason, other than loosing previous velocity information,
>> for
>>>>> *not* minimizing the structures?
>>>>>
>>>>>
>>>>> 2016-09-02 13:17 GMT-03:00 Gustavo Seabra <gustavo.seabra.gmail.com>:
>>>>>
>>>>>> Hi Anna,
>>>>>>
>>>>>>> I have already carried out a classical molecular dynamic of my
>> system.
>>>>>> Continuously, I want to perform a QM/MM minimization starting from one
>>>>>> snapshot of the MD. How can I do it?
>>>>>>
>>>>>> In principle, if you have already run a classical MD on the system,
>> all
>>>>>> you have to do is tweak the mdin file with the new keywords for QM/MM
>>>> and
>>>>>> minimization, then using the restart file from the MD simulation as
>>>> input
>>>>>> structure. Very much as if you were just continuing a classical MD
>>>>>> calculation. (Of course, take care of the points raised earlier by
>>>> Adrian
>>>>>> and Andreas.)
>>>>>>
>>>>>> However, if your goal is to run a QM/MM MD, IMHO you should *not*
>>>> minimize
>>>>>> the geometry all over again. Just turn on the QM hamiltonian and run
>>>> for a
>>>>>> while to adjust the system to the new hamiltonian, keeping the same
>>>>>> ensemble variables as before. (You may nee to adjust dt for a smaller
>>>> step
>>>>>> size.)
>>>>>>
>>>>>> —
>>>>>> Gustavo.
>>>>>>
>>>>>>
>>>>>> _______________________________________________
>>>>>> AMBER mailing list
>>>>>> AMBER.ambermd.org
>>>>>> http://lists.ambermd.org/mailman/listinfo/amber
>>>>>>
>>>>> _______________________________________________
>>>>> AMBER mailing list
>>>>> AMBER.ambermd.org
>>>>> http://lists.ambermd.org/mailman/listinfo/amber
>>>> --
>>>> Dr. Adrian E. Roitberg
>>>> University of Florida Research Foundation 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
>>>>
>>> _______________________________________________
>>> AMBER mailing list
>>> AMBER.ambermd.org
>>> http://lists.ambermd.org/mailman/listinfo/amber
>> --
>> Dr. Adrian E. Roitberg
>> University of Florida Research Foundation 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
>>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
--
Dr. Adrian E. Roitberg
University of Florida Research Foundation 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 Fri Sep 02 2016 - 11:00:04 PDT
