Re: CG equilibration recurring crashes

From: Axel Kohlmeyer (akohlmey_at_gmail.com)
Date: Sun Dec 27 2009 - 12:44:08 CST

please keep namd-l in the cc.

questions asked to the mailing list should be kept on the list.
so that others get to see the resolution.

[...]

>>> bilayer. To a novice for namd it looks like a problem with the force
>>> field.
>>
>> to me this sounds more like you have a very high
>> energy starting configuration.
>
> This is a few hundred steps before the last crash, which involved two
> atoms from subunit 1 moving too fast (crashes before involved a few
> atoms from subunit 3)

so what?

>> multi-component systems are always tricky to equilibrate
>> without an elaborate protocol. as simple minimization and
>> then heating up to production usually is not sufficient.
>>
>> i would suggest to first use position restraints (called
>> constraints in NAMD-speak, IIRC) to confine most of the
>> system and then let only parts equilibrate independently.
>
> I did that extensively by "fixing" first everything except water, then
> freeing the tails only of bilayer, then the whole bilayer molecule,
> then the protein. I also tried to do that by imposing harmonic
> constraints but I was unable to transfer it to CG.

i don't understand what you mean by this.
have you followed this protocol or not?

>> in case of very high potential energy environments, you
>> may need to alternate between minimization and MD a few
>> times. perhaps you equilibrate first the area away from
>> the protein, i.e. close the "gaps" between the periodic
>> images and let the bilayer "heal", and then pay attention
>> to the proximity of the protein. in any case you have to
>> make sure that you dissipate sound waves from the implosions
>> due to the solvation and embedding into the bilayer.
>> using a high friction coefficient in the langevin thermostat
>> certainly helps with that.
>
> While I am considering all suggestions, could you please write
> explicitly how to set "friction coefficient" and an order of
> magnitude?

the keyword is "langevinDamping"
http://www.ks.uiuc.edu/Research/namd/2.7b2/ug/node31.html#7661
since it is hard to tell how far off you are, i cannot make a recommendation.
i would try a value of 1.0 and crank it up if, if needed and reduce it
if the calculation runs without problems.

>> also i would use a reduced time step to improve
>> the stability of the MD integration and avoid
>> extremely close contacts of fast moving atoms.
>
>
> Less than the 20fs I am using? The information (wrong?) I had was not
> to use less than that.

you have to distinguish between equilibration and production. indeed
the martini force field description says that it was parametrized for
that time step (which i personally find very unsettling, as that would
mean that a certain amount of error in the integration of the equations
of motion would have been factored in, which in turn would lead to
non-conservation of energy, which would make it difficult to check
whether the MD runs stable in the first place. but then again, in coarse
graining many "rules" are relaxed. ).

> As the problems are not with atoms at the border of the box, I have
> not increased the "cut".

so what? the boundaries between protein and solvent
or protein and lipid are just as problematic.

>> there is no single automated way to do this right.
>> one has to use common sense and closely observe the
>> system to react to what is going on.
>
> I am prepared to climb. Surely, however, it was far more
> straightforward with all-atoms amber, once the system was stepwise
> minimized I can't remember of any crash.

it really depends on the individual steps. you have posted many
mails about different steps and sometimes with contradicting
detail information, from which it is difficult to reconstruct what exactly
you have done to get where you are now.

if you are new to NAMD and coarse-grain, it might have
been easier to first learn to use one and then the other.

if you prefer using amber, it should be fairly straightforward
to translate the martini parameters into something that
amber can handle. being a coarse grain force field it
has a very reduced set of interaction site definitions
and parameters.

cheers,
    axel.

> thanks a lot
> francesco
>>
>> axel.
>>

-- 
Dr. Axel Kohlmeyer    akohlmey_at_gmail.com
Institute for Computational Molecular Science
College of Science and Technology
Temple University, Philadelphia PA, USA.

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:53:38 CST