AMBER Archive (2009)

Subject: Re: [AMBER] Strange high temperature in softcore TI

From: steinbrt_at_rci.rutgers.edu
Date: Sun May 24 2009 - 23:13:38 CDT

Hi,

well, there is a good reason for removing charges first and then vdw
interactions.

Even scaled at lambda, the interaction energy of two point charges at the
same place is infinite, but their vdw-softcore-repulsion is finite.
Therefore, doing softcore-vdw on charged atoms inevitably leads to vlimit
errors and crashes as solvent atoms crash into the disappearing region. We
are working on a new version of softcore that would allow such one-step
tansformations for amber11, but for now you will have to stick to the
three step-approach.

It is perfectly normal to see high variance for dvdl in vdw
transformations, this doesnt necessarily mean your result is imprecise.
Check the smoothness of your dvdl-curve for a hint on how converged your
simulation is.

Kind Regards,

Thomas

On Sun, May 24, 2009 11:20 pm, Tom Joseph wrote:
> Thanks very much! Reducing the number of softcore atoms by only
> including one base seems to have helped.
>
> Attempting to do the alchemy in one shot (rather than removing
> charges/mutating/adding charges as in your tutorial), even with the
> reduced softcore region, results in vlimit errors and sander crashing.
> I wanted to do this in hopes that the electrostatics would "stabilize"
> the van der Waals interactions, which alone give a relatively high
> RMSD for DV/DL.
>
> I know even 14 atoms is a decently large change, but why should
> leaving the electrostatics on during the mutation cause such problems,
> since they are scaled by lambda?
>
> Thanks,
> --Tom
>
> On Sun, May 24, 2009 at 10:26 PM, <steinbrt_at_rci.rutgers.edu> wrote:
>> Hi,
>>
>> this sounds like atoms in the TI region are crashing into each other,
>> causing the high temperatures and structural distortions. Odd that you
>> dont see that when you run without TI, though...
>>
>> Always a good thing to try is to test your runs with a smaller TI region
>> and make it subsequently bigger until you reach a point were problems
>> occur. Say only make the one base that actualy changes softcore.
>>
>> Kind Regards,
>>
>> Thomas
>>
>> On Sun, May 24, 2009 9:54 pm, Tom Joseph wrote:
>>> I am trying to do a TI calculation using AMBER 10's softcore
>>> potentials feature but am getting odd results, and was hoping someone
>>> might know what has gone wrong here.
>>>
>>> I am mutating a T:A base pair to A:A in a short DNA:RNA hybrid. The
>>> softcore region includes both bases of the base pair in question, 28
>>> atoms total. After 2000 steps of minimization, I attempt to heat the
>>> system from 100 K (assigned velocities) to 300 K using the mdin file
>>> below. The temperature of the softcore region tends to stay > 500 K
>>> and every few thousand steps suddenly jumps up to some insane number
>>> (sometimes > 10000 K) and a bunch of vlimit errors are produced. The
>>> base pair being alchemically mutated flies apart, distorting the rest
>>> of the structure.
>>>
>>> This happens at all clambdas I've tried. When I minimize and heat
>>> either configuration without TI, it remains stable.
>>>
>>> Would anyone be able to give a clue as to what I'm doing wrong?
>>>
>>> Thanks,
>>> --Tom
>>>
>>> Heating sample mdin:
>>>
>>>  &cntrl
>>>     imin=0, ntx=1, ntpr=500, ntwr=500, ntwx=500, ntwe=500,
>>>     nscm=5000,
>>>     ntf=1, ntc=1,
>>>     ntb=1, ntp=0,
>>>     nstlim=50000, t=0.0, dt=0.001,
>>>     cut=9.0,
>>>     tempi=100.0, ntt=3, gamma_ln=3, ig=12345,
>>>     ntr=0, nmropt=1,
>>> icfe=1, ifsc=1,
>>> clambda=0.33787,
>>> crgmask=':7_at_N1,C6,H6,C5,C7,H71,H72,H73,C4,O4,N3,H3,C2,O2',
>>> scmask=':7_at_N1,C6,H6,C5,C7,H71,H72,H73,C4,O4,N3,H3,C2,O2 |
>>> :19_at_N9,C8,H8,N7,C5,C6,N6,H61,H62,N1,C2,H2,N3,C4',
>>>  &end
>>> Heating from 100 to 300 K
>>>  &wt type='TEMP0', istep1=0,    istep2=10000,  value1=100.0,
>>> value2=300.0,
>>>  &end
>>>  &wt type='TEMP0', istep1=10001, istep2=50000, value1=300.0,
>>> value2=300.0,
>>>  &end
>>>  &wt type='END',  &end
>>>
>>> (crgmask and scmask are adjusted appropriately for the other
>>> configuration)
>>>
>>> _______________________________________________
>>> AMBER mailing list
>>> AMBER_at_ambermd.org
>>> http://lists.ambermd.org/mailman/listinfo/amber
>>>
>>
>>
>> Dr. Thomas Steinbrecher
>> BioMaps Institute
>> Rutgers University
>> 610 Taylor Rd.
>> Piscataway, NJ 08854
>>
>> _______________________________________________
>> AMBER mailing list
>> AMBER_at_ambermd.org
>> http://lists.ambermd.org/mailman/listinfo/amber
>>
>
> _______________________________________________
> AMBER mailing list
> AMBER_at_ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>

Dr. Thomas Steinbrecher
BioMaps Institute
Rutgers University
610 Taylor Rd.
Piscataway, NJ 08854

_______________________________________________
AMBER mailing list
AMBER_at_ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber