[gmx-users] LINCS vs SHAKE
thompsjj at purdue.edu
thompsjj at purdue.edu
Mon Nov 1 05:06:22 CET 2010
I think Mark's comment still stands in this case: the argument for
keeping the bath and the protein at different, nonequilibrating
temperatures is still unsound. It might not explode right away,
because by lowering the timestep and decreasing temperature, you're
affording the thermostat a better chance to try to scale back
velocities. Explode or not, at 500K the solute is still pumping large
amounts of energy into the 300K solvent, which you're asking the
thermostat to absorb. The physics at the solute-solvent boundary are
questionable at best - real-life atoms don't get to have a special
thermostat to rescale velocities when they touch something hot.
~j
Quoting Sai Pooja <saipooja at gmail.com>:
> Thanks JJ and Mark. I shall revise my algorithm/technique. This was just a
> lazy way of implementing a more robust technique and I shall use the more
> rigorous form.
> However, if one wants to use the current technique with a smaller
> temperature range, the dynamics of the system with say Protein at 500K and
> water at 300K would be unphysical or can that be corrected by using a
> smaller timestep and thermostat type/parameters?
>
> Pooja
>
> On Sun, Oct 31, 2010 at 10:15 PM, <thompsjj at purdue.edu> wrote:
>
>> I concur with Mark - you're pumping heat into the solvent which is
>> physically unsound, particularly at the solute-solvent boundary in your
>> situation. This would be analogous to making a block of metal at 900K
>> instantly "appear" in water at 300K: what would happen?
>> Technically the Nose-Hoover thermostat is going to be better than Berendsen
>> at reducing the noncanonical bias in the replica exchange, but of course
>> using a smaller range of temperatures is going to reduce the pressure on the
>> thermostat to keep up with velocity scaling. This is a different problem
>> entirely, though. To be honest, I see the difference in your management of
>> protein and solvent temperatures as the biggest problem here.
>>
>>
>>
>> Quoting Mark Abraham <mark.abraham at anu.edu.au>:
>>
>>
>>>
>>> ----- Original Message -----
>>> From: Sai Pooja <saipooja at gmail.com>
>>> Date: Monday, November 1, 2010 10:24
>>> Subject: Re: [gmx-users] LINCS vs SHAKE
>>> To: Discussion list for GROMACS users <gmx-users at gromacs.org>
>>>
>>> Some information:
>>>> Except for this note, I do not get any other warning when I generate the
>>>> .tpr files. The cut-offs are in line with the forcefield recommendations
>>>> and
>>>> The largest charge group contains 12 atoms.
>>>> Since atoms only see each other when the centers of geometry of the
>>>> charge
>>>> groups they belong to are within the cut-off distance, too large charge
>>>> groups can lead to serious cut-off artifacts.
>>>> For efficiency and accuracy, charge group should consist of a few
>>>> atoms.
>>>> For all-atom force fields use: CH3, CH2, CH, NH2, NH, OH, CO2, CO, etc
>>>>
>>>
>>> Probably not critical, but IIRC the latest GROMACS imposes single-atom
>>> charge groups for CHARMM, in line with CHARMM developer advice.
>>>
>>>
>>>> @JJ
>>>>
>>>> I am currently using Nose-hoover and it does happen close to a
>>>> replica-exchange. Since I am attempting an algorithm which only scales the
>>>> temperature of the protein and not the solvent across replicas,
>>>> your pointer
>>>> on the thermostat seems very pertinent.
>>>>
>>>
>>> Yep. That's a prime source for problems - IIRC post-exchange velocities
>>> are rescaled globally, not group-wise. Even if the latter were implemented,
>>> you'd still have to reconcile the fact that your simulations
>>> constantly have
>>> heat flowing from the solute heat bath to the solvent heat bath via your
>>> system. Is that sound?
>>>
>>> Also, I'm not sure which temperature GROMACS is taking as the replica
>>> temperature, whether that is dependent on the order of the
>>> temperature-coupling groups in the .mdp file, and whether it shouldn't be
>>> regarding your situation as questionable/illegal and complaining at the
>>> grompp stage.
>>>
>>> Do you think a weaker coupling algo like Berendsen might help resolve
>>>> this problem?
>>>>
>>>> I can also try using a smaller temperature range. I am currently using
>>>> 300-900K (with solvent set at 300 for all 6 replicas). Perhaps if I use a
>>>> smaller temperature range like 300-550, this could be prevented?
>>>>
>>>
>>> If that's the cause, then the answer is "probably", but you should address
>>> the underlying problem/question as above.
>>>
>>> Mark
>>>
>>>
>>>> Pooja
>>>>
>>>> On Sun, Oct 31, 2010 at 7:13 PM, <thompsjj at purdue.edu> wrote:
>>>> Have you looked into changing the thermostat yet? Can you determine if
>>>> the malformed residues are a result of a recent replica exchange? This
>>>> continues to sound like an overheat that the thermostat cannot handle. >
>>>>
>>> >
>>> >
>>> >
>>> >
>>> >
>>> >
>>>
>>>> Quoting Mark Abraham <mark.abraham at anu.edu.au>:
>>>>
>>> >
>>> >
>>> >
>>>
>>>> ----- Original Message -----
>>>> From: Sai Pooja <saipooja at gmail.com>
>>>> Date: Monday, November 1, 2010 8:12
>>>> Subject: Re: [gmx-users] LINCS vs SHAKE
>>>> To: Discussion list for GROMACS users <gmx-users at gromacs.org>
>>>>
>>> >
>>> > Yes, I understand that. But then this is a protein in water simulation
>>> with CHarmm forcefield and I cannot find a reason as to why does this
>>> happen(esp after variable number of steps)?
>>> >
>>>
>>>> MD integrates equations of motion. If the starting configuration, or
>>>> integration procedure is not well-formed, then atoms will experience large
>>>> forces and motions, which lead to more, which eventually lead to a smoking
>>>> pile of debris. See
>>>> http://www.gromacs.org/Documentation/Terminology/Blowing_Up for
>>>> discussion and suggestions
>>>>
>>> >
>>>
>>>> Mark
>>>>
>>> >
>>> > I have checked the .pdb files generated for the steps where Lincs
>>> fails and it seems that H atoms for atleast 1 sidechain are in incorrect
>>> positions(usually clashing). There doesn't seem to be a good way of
>>> analyzing this error. If you have any suggestions I would be eager to try
>>> them.
>>> > >
>>> > Pooja
>>> >
>>>
>>>> On Sat, Oct 30, 2010 at 3:40 PM, Justin A. Lemkul <jalemkul at vt.edu>
>>>> wrote:
>>>>
>>> > >
>>>
>>>> >
>>>>
>>> > Sai Pooja wrote:
>>> > >
>>> > For Replica exchange, is there any advantage of in using SHAKE over
>>> LINCS(other than the stepsize)?
>>>
>>>> I am running an REM simulation and the simulation stops after running
>>>> for variable number of steps (100000, 10000000 etc.) because some
>>>> bond moves
>>>> more than 30 degrees and LINCS gives a warning.
>>>>
>>> > >
>>>
>>>> >
>>>>
>>> > It has been said (check the archive) that LINCS is more stable. Just
>>> because you're getting LINCS warnings does not mean the constraint
>>> algorithm
>>> is to blame, it's just the first algorithm that fails when your model
>>> physics implodes.
>>> > >
>>> > -Justin
>>> > >
>>>
>>>> >
>>>>
>>> > Pooja
>>> > >
>>> > --
>>>
>>>> Quaerendo Invenietis-Seek and you shall discover.
>>>>
>>> > >
>>>
>>>> >
>>>>
>>> > --
>>>
>>>> ========================================
>>>>
>>> > >
>>> > Justin A. Lemkul
>>>
>>>> Ph.D. Candidate
>>>> ICTAS Doctoral Scholar
>>>> MILES-IGERT Trainee
>>>> Department of Biochemistry
>>>> Virginia Tech
>>>> Blacksburg, VA
>>>> jalemkul[at]vt.edu | (540) 231-9080
>>>>
>>> > > http://www.bevanlab.biochem.vt.edu/Pages/Personal/justin
>>>
>>>> >
>>>>
>>> > ========================================
>>>
>>>> --
>>>> gmx-users mailing list gmx-users at gromacs.org
>>>>
>>> > > http://lists.gromacs.org/mailman/listinfo/gmx-users
>>> > Please search the archive at
>>> http://www.gromacs.org/Support/Mailing_Lists/Search before posting!
>>>
>>>> Please don't post (un)subscribe requests to the list. Use the www
>>>> interface or send it to gmx-users-request at gromacs.org.
>>>> Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>>>>
>>> > >
>>> > >
>>>
>>>> --
>>>> Quaerendo Invenietis-Seek and you shall discover.
>>>>
>>> > > --
>>> > gmx-users mailing list gmx-users at gromacs.org
>>> > http://lists.gromacs.org/mailman/listinfo/gmx-users
>>>
>>>> Please search the archive at
>>>>
>>> > http://www.gromacs.org/Support/Mailing_Lists/Search before posting!
>>>
>>>> Please don't post (un)subscribe requests to the list. Use the
>>>> www interface or send it to gmx-users-request at gromacs.org.
>>>> Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>>>>
>>> >
>>> >
>>> >
>>> >
>>>
>>>> --
>>>> gmx-users mailing list gmx-users at gromacs.org
>>>>
>>> > http://lists.gromacs.org/mailman/listinfo/gmx-users
>>>
>>>> Please search the archive at
>>>> http://www.gromacs.org/Support/Mailing_Lists/Search before posting!
>>>> Please don't post (un)subscribe requests to the list. Use thewww
>>>> interface or send it to gmx-users-request at gromacs.org. >
>>>> Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>>>>
>>> >
>>>
>>>> >
>>>> --
>>>> Quaerendo Invenietis-Seek and you shall discover.
>>>>
>>> > --
>>>
>>>> gmx-users mailing list gmx-users at gromacs.org
>>>> http://lists.gromacs.org/mailman/listinfo/gmx-users
>>>> Please search the archive at
>>>> http://www.gromacs.org/Support/Mailing_Lists/Search before posting!
>>>> Please don't post (un)subscribe requests to the list. Use the
>>>> www interface or send it to gmx-users-request at gromacs.org.
>>>> Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>>>>
>>>
>>>
>>
>>
>> --
>> gmx-users mailing list gmx-users at gromacs.org
>> http://lists.gromacs.org/mailman/listinfo/gmx-users
>> Please search the archive at
>> http://www.gromacs.org/Support/Mailing_Lists/Search before posting!
>> Please don't post (un)subscribe requests to the list. Use thewww interface
>> or send it to gmx-users-request at gromacs.org.
>> Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>>
>
>
>
> --
> Quaerendo Invenietis-Seek and you shall discover.
>
More information about the gromacs.org_gmx-users
mailing list