Re: CGMD timestep

From: Jerome Henin (jhenin_at_cmm.chem.upenn.edu)
Date: Mon Dec 08 2008 - 16:00:03 CST

Hi everyone,

About that quote from that Gromacs wiki:
"A reduction of the time step would favor the states of the system
with the lower potential energy. This is equivalent to play with the
temperature."

Am I the only one here to find this idea egregious? One thing is to
tolerate a slightly inaccurate integration, while counting on a good
thermostat to get proper canonical sampling. The other thing is to go
into the regime where integration error just f*cks up your sampling,
and start calling the time step a force field parameter...

Any thoughts on this?
Jerome

On Mon, Dec 8, 2008 at 1:11 PM, Axel Kohlmeyer <akohlmey_at_cmm.upenn.edu> wrote:
> On 12/8/08, BIN ZHANG <zhngbn_at_gmail.com> wrote:
>> If I were correct, NAMD only use the MARTINI FF for Coarse Grained
>> lipid and water, and it has its only CG protein model. So for your
>> question, I'm not really sure(that's why I post on the list. :-) )
>
> ..and i keep repeating myself. the force field parameters and
> the MD code are two _different_ entities.
>
> _you_ have to convince yourself that the parameters _you_ use are
> working "good enough" for your purposes. the problem is that we are
> not working with absolute truths here, but approximations. and an
> approximation that may be a good one for one problem, may be a
> not so good one for the next.
>
> the best way to convince yourself, is to make tests with small(er)
> subsystems and find out, if you can reproduce properties that you
> need to represent well. in CG modelling, poeple take much more
> liberties, but that means that there is less transferability. you just
> have to know what you are doing. i am a bit old-fashioned and thus
> personally would not be comfortable using a parameter set where the
> error from using a very large time step is compensated through the
> parametrization. but then again, since CG modeling is approximate,
> it is a legitimate choice that people make, provided they look at
> properties that are not affected by this approximation.
>
>> In their origin RBCG paper, 25fs is used for their protein-lipid
>> system, but 15fs is used in another application paper.
>
> see above for my view on these issues and again. the fastest
> motion in the system _does_ determine how large a time step
> you can choose, but keep in mind that with a langevin thermostat
> you add a friction term the size of which has an impact on the
> speed of particles. so you have to make choices every where and
> you have to justify them in some way.
>
> cheers,
> axel.
>
>> Bin
>>
>>
>>
>> On Dec 8, 2008, at 9:37 AM, Axel Kohlmeyer wrote:
>>
>>> yes.
>>>
>>> but in how far is this affected by the MD code you are using?
>>>
>>> axel.
>>>
>>> On 12/8/08, BIN ZHANG <zhngbn_at_gmail.com> wrote:
>>>> Hi, Axel
>>>>
>>>> Maybe I understood it wrongly, but it says here
>>>>
>>>> "
>>>> In a all-atom FF you assume the free energy surface the FF describes
>>>> is actually exact and the smaller the time step the more accurate you
>>>> are. You then try to increase the time step to be able to sample bit
>>>> more and increase your statistics.
>>>> "
>>>> And that's normally what we think.
>>>> But for the CG FF:
>>>> "
>>>> Given the CG force field (set of parameters) and thus the potential
>>>> energy surface of the system, the way you are going to explore this
>>>> particular surface is actually depending on many parameters and time
>>>> step is part of them. A reduction of the time step would favor the
>>>> states of the system with the lower potential energy. This is
>>>> equivalent to play with the temperature. This is also true for the
>>>> cutoff -do not modify the cutoff- it is part of the force field.
>>>> Increasing the cutoff will not make your run more accurate but at the
>>>> contrary alter the balance of the different terms.
>>>> "
>>>>
>>>> Bin
>>>>
>>>>
>>>>
>>>> On Dec 8, 2008, at 9:19 AM, Axel Kohlmeyer wrote:
>>>>
>>>>> bin,
>>>>>
>>>>> if i'd read the link, what would it help you?
>>>>>
>>>>> figuring out what is a proper time step for a given
>>>>> problem is an _elementary_ skill in doing MD, so you'll
>>>>> have to convince yourself that you make the right choice.
>>>>>
>>>>> cheers,
>>>>> axel.
>>>>>
>>>>>
>>>>> On 12/8/08, BIN ZHANG <zhngbn_at_gmail.com> wrote:
>>>>>> Hi, Axel:
>>>>>>
>>>>>> I appreciate your comment, but I would be more appreciative if you
>>>>>> would read the link.
>>>>>> Thanks.
>>>>>>
>>>>>> Bin
>>>>>>
>>>>>>
>>>>>>
>>>>>> On Dec 8, 2008, at 1:26 AM, Axel Kohlmeyer wrote:
>>>>>>
>>>>>>> On 12/7/08, BIN ZHANG <zhngbn_at_gmail.com> wrote:
>>>>>>>> Dear all:
>>>>>>>>
>>>>>>>> I just find this link abut MARTINI forcefield (
>>>>>>>> http://wiki.gromacs.org/index.php/Coarse_Grained_Force_Field
>>>>>>>> ), which says "The force field has been parametrized to be
>>>>>>>> valid in
>>>>>>>> the range 0.02/0.04 ps."
>>>>>>>> Does this argument also hold for NAMD?
>>>>>>>
>>>>>>> dear bin,
>>>>>>>
>>>>>>> please refer to your favorate text book on MD and
>>>>>>> see that the maximum allowed time step is determined
>>>>>>> by the fastest motions in your system (i.e. mass of particles
>>>>>>> and stiffness of interactions), the magnitude of error you
>>>>>>> are willing to accept and the stability of the time integration
>>>>>>> algorithm employed. except for the last item everything does
>>>>>>> not depend on the MD implementation and since practically
>>>>>>> all widely used classical MD codes employ one over the Verlet
>>>>>>> variants, also the last point is in practice irrelevant.
>>>>>>> with that in mind, you should be able to answer your question
>>>>>>> by yourself.
>>>>>>>
>>>>>>> cheers,
>>>>>>> axel.
>>>>>>>
>>>>>>>> Thanks a lot.
>>>>>>>>
>>>>>>>> Bin
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>> -------------------------------------------------------------
>>>>>>>> The tree of liberty must be refreshed from time to time with the
>>>>>>>> blood
>>>>>>>> of patriots and tyrants.
>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> --
>>>>>>> =
>>>>>>> =
>>>>>>> =
>>>>>>> =
>>>>>>> =
>>>>>>> ==================================================================
>>>>>>> Axel Kohlmeyer akohlmey_at_cmm.chem.upenn.edu http://
>>>>>>> www.cmm.upenn.edu
>>>>>>> Center for Molecular Modeling -- University of Pennsylvania
>>>>>>> Department of Chemistry, 231 S.34th Street, Philadelphia, PA
>>>>>>> 19104-6323
>>>>>>> tel: 1-215-898-1582, fax: 1-215-573-6233, office-tel:
>>>>>>> 1-215-898-5425
>>>>>>> =
>>>>>>> =
>>>>>>> =
>>>>>>> =
>>>>>>> =
>>>>>>> ==================================================================
>>>>>>> If you make something idiot-proof, the universe creates a better
>>>>>>> idiot.
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>> -------------------------------------------------------------
>>>>>> The tree of liberty must be refreshed from time to time with the
>>>>>> blood
>>>>>> of patriots and tyrants.
>>>>>>
>>>>>>
>>>>>
>>>>>
>>>>> --
>>>>> =
>>>>> =
>>>>> =
>>>>> ====================================================================
>>>>> Axel Kohlmeyer akohlmey_at_cmm.chem.upenn.edu http://
>>>>> www.cmm.upenn.edu
>>>>> Center for Molecular Modeling -- University of Pennsylvania
>>>>> Department of Chemistry, 231 S.34th Street, Philadelphia, PA
>>>>> 19104-6323
>>>>> tel: 1-215-898-1582, fax: 1-215-573-6233, office-tel:
>>>>> 1-215-898-5425
>>>>> =
>>>>> =
>>>>> =
>>>>> ====================================================================
>>>>> If you make something idiot-proof, the universe creates a better
>>>>> idiot.
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> -------------------------------------------------------------
>>>> The tree of liberty must be refreshed from time to time with the
>>>> blood
>>>> of patriots and tyrants.
>>>>
>>>>
>>>
>>>
>>> --
>>> =
>>> ======================================================================
>>> Axel Kohlmeyer akohlmey_at_cmm.chem.upenn.edu http://
>>> www.cmm.upenn.edu
>>> Center for Molecular Modeling -- University of Pennsylvania
>>> Department of Chemistry, 231 S.34th Street, Philadelphia, PA
>>> 19104-6323
>>> tel: 1-215-898-1582, fax: 1-215-573-6233, office-tel: 1-215-898-5425
>>> =
>>> ======================================================================
>>> If you make something idiot-proof, the universe creates a better
>>> idiot.
>>
>>
>>
>>
>>
>>
>>
>>
>>
>> -------------------------------------------------------------
>> The tree of liberty must be refreshed from time to time with the blood
>> of patriots and tyrants.
>>
>>
>
>
> --
> =======================================================================
> Axel Kohlmeyer akohlmey_at_cmm.chem.upenn.edu http://www.cmm.upenn.edu
> Center for Molecular Modeling -- University of Pennsylvania
> Department of Chemistry, 231 S.34th Street, Philadelphia, PA 19104-6323
> tel: 1-215-898-1582, fax: 1-215-573-6233, office-tel: 1-215-898-5425
> =======================================================================
> If you make something idiot-proof, the universe creates a better idiot.
>
>

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:50:13 CST