Re: ABF - restricting reaction coordinate

From: Subramanian Vaitheeswaran (vaithee_at_umd.edu)
Date: Fri Dec 07 2007 - 14:49:55 CST

Jerome,

Do you mean that the "abf restraintList" clashes with the harmonic potentials at the boundaries defined by xiMin and xiMax? - I can get it to run with relatively few crashes when I don't restrict the RC range.

My choices seem to be:
1. Increase langevinDamping (to say 50 or 100) and keep
"forceConst 0.5", leaving the angle and dihedral in "abf restraintList" unchanged. This _might_ decrease the frequency of crashes enough to get good sampling.

2. Decrease forceConst to, say, 0.1. This will at least partially restrict the range of the RC and hopefully will lead to better sampling than having no restrictions on the range at all (i.e. have only a single window spanning the whole range).

3. Set forceConst to 0 and thus sample in only one window for the range xi E [0,180] degrees.

4. Lift one of the restraints in "abf restraintList" as you suggest. I just don't see how I can do this and still measure the same free energy.

I would greatly appreciate any suggestions.

regards,
Vaithee

---- Original message ----
>Date: Fri, 7 Dec 2007 13:47:33 -0500
>From: "Jerome Henin" <jhenin_at_cmm.chem.upenn.edu>
>Subject: Re: namd-l: ABF - restricting reaction coordinate
>To: "Subramanian Vaitheeswaran" <vaithee_at_umd.edu>
>
>Vaithee,
>From my tests, it looks like you are applying too many restraints: the
>distance + dihedral are conflicting with the ABF restraints. If you
>lift just one of them, the others seem to work. Redesigning your set
>of restraints may solve the problem.
>Jerome
>
>On Dec 7, 2007 1:16 PM, Subramanian Vaitheeswaran <vaithee_at_umd.edu> wrote:
>> Jerome,
>>
>> I am attaching the archive with the files needed. The system is a water-filled nanotube with a methane and a toluene molecule. Please let me know if I can tell you anything more about the system.
>>
>> I greatly appreciate your help, many thanks.
>>
>> Vaithee
>>
>>
>> ---- Original message ----
>> >Date: Fri, 7 Dec 2007 13:02:27 -0500
>> >From: "Jerome Henin" <jhenin_at_cmm.chem.upenn.edu>
>> >Subject: Re: namd-l: ABF - restricting reaction coordinate
>> >To: "Subramanian Vaitheeswaran" <vaithee_at_umd.edu>
>> >
>> >Langevin damping depends a lot on the system size. If you are running
>> >tests on a molecule in gas-phase, for example, it is common that
>> >strong coupling is required.
>> >I think it would be good for me to be able to run tests on this system
>> >to locate the problem.
>> >Jerome
>> >
>> >On Dec 7, 2007 12:59 PM, Subramanian Vaitheeswaran <vaithee_at_umd.edu> wrote:
>> >> Jerome,
>> >>
>> >> You are right - with "forceConst 0.5", the simulation becomes unstable and I have to increase langevinDamping to 20 before the simulation runs for at least 100,000 steps (1fs time step). Even so, xi is not entirely restricted to the range [0.5,45.5]. Here is the raw data for xi:
>> >> **************************************************
>> >> TCL: ABF> Xi at timestep 0 : -95.0825868217
>> >> TCL: ABF> Xi at timestep 10000 : -2.00211414457
>> >> TCL: ABF> Xi at timestep 20000 : 4.77949546132
>> >> TCL: ABF> Xi at timestep 30000 : 8.76225778713
>> >> TCL: ABF> Xi at timestep 40000 : 10.4933641552
>> >> TCL: ABF> Xi at timestep 50000 : -0.706744034879
>> >> TCL: ABF> Xi at timestep 60000 : -3.1966774742
>> >> TCL: ABF> Xi at timestep 70000 : 5.55169580864
>> >> TCL: ABF> Xi at timestep 80000 : -1.68767289768
>> >> TCL: ABF> Xi at timestep 90000 : 3.20040285968
>> >> **************************************************
>> >> This is certainly better than no restriction at all - most data points lie in the desired range.
>> >>
>> >> However, I am a little worried by the fact that langevinDamping has to be so high (20). I am not concerned with kinetics here only thermodynamics, but I am concerned that the system will be relaxing too slowly for the free energies to converge. Do you have any suggestions? I can send you the input files - please let me know.
>> >>
>> >> thanks again,
>> >> Vaithee
>> >>
>> >>
>> >> ---- Original message ----
>> >> >Date: Fri, 7 Dec 2007 12:08:34 -0500
>> >> >From: "Jerome Henin" <jhenin_at_cmm.chem.upenn.edu>
>> >> >Subject: Re: namd-l: ABF - restricting reaction coordinate
>> >> >To: "Subramanian Vaitheeswaran" <vaithee_at_umd.edu>
>> >> >
>> >> >Vaithee,
>> >> >You're right about the manual only including coordinates in Angstrom.
>> >> >But it will have to be made more general at some point.
>> >> >
>> >> >Normally, angle values are handled from -180 to 180, and the
>> >> >periodicity should be fully accounted for. You may need a stronger
>> >> >force constant in your system, maybe 0.5 or 1.0. The only limit is
>> >> >that the simulation becomes unstable when the bias is too stiff.
>> >> >
>> >> >If problems persist, you could send me your input files and I would
>> >> >look into this.
>> >> >Jerome
>> >> >
>> >> >On Dec 7, 2007 12:03 PM, Subramanian Vaitheeswaran <vaithee_at_umd.edu> wrote:
>> >> >> Jerome,
>> >> >>
>> >> >> Actually, it is largely my fault. The user guide only includes the distance based reaction coords - the default value for those (at least for distance-com, which is all that I have used) is indeed 10 kcal/mol/Å^2.
>> >> >>
>> >> >> For the dihedral and dihedral-com RCs that I downloaded from the ABF website, the default is 0. I should have realized that the units of forceConst are different for the angle-based RCs, so the default magnitude is likely to be different too. I should have checked . . . It seems to be working now.
>> >> >>
>> >> >> One other question - how are negative values for the angle handled? - does the code just take the absolute value? That would make sense. When I have "xiMin 0.5" and "xiMax 45.5", here is what I get in the log file:
>> >> >> ****************************
>> >> >> TCL: ABF> Xi at timestep 0 : -95.0825868217
>> >> >> TCL: ABF> Xi at timestep 50000 : -9.76342993541
>> >> >> TCL: ABF> Xi at timestep 100000 : -8.5588270341
>> >> >> TCL: ABF> Xi at timestep 150000 : -10.8294823801
>> >> >> TCL: ABF> Xi at timestep 200000 : -5.35192348943
>> >> >> TCL: ABF> Xi at timestep 250000 : -7.61706445236
>> >> >> TCL: ABF> Xi at timestep 300000 : -9.27184421372
>> >> >> ****************************
>> >> >>
>> >> >> regards,
>> >> >> Vaithee
>> >> >>
>> >> >>
>> >> >> ---- Original message ----
>> >> >> >Date: Fri, 7 Dec 2007 11:42:29 -0500
>> >> >> >From: "Jerome Henin" <jhenin_at_cmm.chem.upenn.edu>
>> >> >> >Subject: Re: namd-l: ABF - restricting reaction coordinate
>> >> >> >To: "Subramanian Vaitheeswaran" <vaithee_at_umd.edu>
>> >> >> >
>> >> >> >Dear Vaithee,
>> >> >> >Thanks for notifying me. I will correct this in the user's guide.
>> >> >> >As a rule, it is always useful to check the summary of ABF parameters
>> >> >> >in the output to see what values are actually used by the code.
>> >> >> >
>> >> >> >I hope everything will work for you now,
>> >> >> >Jerome
>> >> >> >
>> >> >> >
>> >> >> >On Dec 7, 2007 11:33 AM, Subramanian Vaitheeswaran <vaithee_at_umd.edu> wrote:
>> >> >> >> Dear Jerome,
>> >> >> >>
>> >> >> >> As I said in my second mail today, the default value of forceConst is 0 not 10 as the user guide states. So without explicitly defining forceConst, the reaction coordinate is not restricted at all, only the collection of data is. This would certainly explain why xi is out of range.
>> >> >> >>
>> >> >> >> I will try with "forceConst 0.1".
>> >> >> >>
>> >> >> >> thanks,
>> >> >> >> Vaithee
>> >> >> >>
>> >> >> >>
>> >> >> >> ---- Original message ----
>> >> >> >> >Date: Fri, 7 Dec 2007 11:01:47 -0500
>> >> >> >> >From: "Jerome Henin" <jhenin_at_cmm.chem.upenn.edu>
>> >> >> >> >Subject: Re: namd-l: ABF - restricting reaction coordinate
>> >> >> >> >To: "Subramanian Vaitheeswaran" <vaithee_at_umd.edu>
>> >> >> >> >Cc: namdlist <namd-l_at_ks.uiuc.edu>
>> >> >> >> >
>> >> >> >> >Dear Vaithee,
>> >> >> >> >
>> >> >> >> >The problem might come from the fact that the dihedral coordinate uses
>> >> >> >> >force constants in kcal/mol/degree^2, which is a rather large unit. In
>> >> >> >> >many cases, a force constant of 0.1 can be sufficient.
>> >> >> >> >
>> >> >> >> >This does not explain why the coordinate gets out of range. Does it
>> >> >> >> >start within the desired range? If not, it would be interesting to see
>> >> >> >> >the initial behavior. Can you send me the full output (maybe
>> >> >> >> >off-list)?
>> >> >> >> >
>> >> >> >> >Best,
>> >> >> >> >Jerome
>> >> >> >> >
>> >> >> >> >
>> >> >> >> >On Dec 7, 2007 9:55 AM, Subramanian Vaitheeswaran <vaithee_at_umd.edu> wrote:
>> >> >> >> >> Dear Chris and Jerome,
>> >> >> >> >>
>> >> >> >> >> Following both your suggestions a couple of days ago, I split the range of my reaction coordinate (a dihedral angle) into four windows. I specify the following in my .conf file:
>> >> >> >> >> ********************************
>> >> >> >> >> abf coordinate dihedral
>> >> >> >> >> abf abf1 {8 6 12 16}
>> >> >> >> >> abf dxi 1.0
>> >> >> >> >> abf xiMin 0.5
>> >> >> >> >> abf xiMax 45.5
>> >> >> >> >> abf writeXiFreq 50000
>> >> >> >> >> ********************************
>> >> >> >> >> So, I expect that xi will be in the range [0.5, 45.5] degrees for *every* step.
>> >> >> >> >>
>> >> >> >> >> But xi is almost never in this range - here is a small fragment grepped from the log file:
>> >> >> >> >> ********************************
>> >> >> >> >> TCL: ABF> Xi at timestep 4300000 : -119.747074603
>> >> >> >> >> TCL: ABF> Xi at timestep 4350000 : -44.6413954893
>> >> >> >> >> TCL: ABF> Xi at timestep 4400000 : -55.1637937915
>> >> >> >> >> TCL: ABF> Xi at timestep 4450000 : -108.865140314
>> >> >> >> >> TCL: ABF> Xi at timestep 4500000 : -155.734544341
>> >> >> >> >> TCL: ABF> Xi at timestep 4550000 : -57.9804339106
>> >> >> >> >> TCL: ABF> Xi at timestep 4600000 : -160.710021078
>> >> >> >> >> ********************************
>> >> >> >> >> Does this imply that the default value for forceConst of 10 kcal/mol/Å^2 is too small for this system? Do you see any problems with increasing it to an arbitrarily large value?
>> >> >> >> >>
>> >> >> >> >> Thanks for all your help,
>> >> >> >> >> Vaithee
>> >> >> >> >>
>> >> >> >> >>
>> >> >> >>
>> >> >>
>> >>
>>

This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:45:40 CST