Re: [AMBER] about skinnb

From: Robert Duke <rduke.email.unc.edu>
Date: Wed, 28 Apr 2010 17:04:36 -0400

What you are doing in the check that determines whether pairlist rebuild is
needed is simple. If the max distance moved by any one atom is 1/2 skin,
then we basically assume that some other atom has moved close to that
distance. At this point, all atom pair interactions are still covered by
the "skin", but we may have an atom move further in the next step (or not).
At any rate, it is safest to just go ahead and rebuild the pairlist at this
point. This undoubtedly results in pairlist rebuilds that are more frequent
than absolutely necessary, but it keeps the check cheap; everything in code
like this is a tradeoff with regards to performance, and we want the best
performance possible, but within the constraints of insuring accuracy.
Regards - Bob Duke
----- Original Message -----
From: "liu junjun" <ljjlp03.gmail.com>
To: "AMBER Mailing List" <amber.ambermd.org>
Sent: Wednesday, April 28, 2010 4:27 PM
Subject: Re: [AMBER] about skinnb


OK, got one more question. Hopefully I am not bothering you..

The nonbond (nb) pairlist is extended to cut+skinnb, and then those nb pairs
that are within the extended skinnb are ignored when calculating
energies/forces. Meanwhile, 1/2 skinnb is compared with maxdis to decide
whether to rebuild the nonbond pairlist or not, where nb pairlist is not
used here and the maxdis is the maximum displacement among all the atoms in
the system. It looks like those nb pairs within the extended skinnb are not
used. I am curious why extending the nb pairlist. Could you please give me a
hint?

Thanks!

Junjun

On Wed, Apr 28, 2010 at 4:02 PM, liu junjun <ljjlp03.gmail.com> wrote:

> I found the code that filter off those within skinnb of nonbond list. Now
> I
> understand. Thank you very much!
>
> Best!
>
> Junjun
>
>
> On Wed, Apr 28, 2010 at 3:16 PM, Robert Duke <rduke.email.unc.edu> wrote:
>
>> No, not correct. Within the energy and force evaluation routines, you
>> LOOK at all entries on the nonbonded pairlist. BUT you only calculate
>> energies, forces, virials (if needed) on atoms that are within cut of
>> each
>> other. A simple test without even looking at the code. Set skinnb to
>> 2.0
>> and run 50 steps. Then set skinnb to 4.0 and run 50 steps. Compare
>> results
>> - should be the same. Try same with actual change of cut by 2.0
>> angstrom -
>> results will be slightly different.
>> Regards - Bob
>> (but if you look at the code, that should convince you too... ;-))
>>
>> ----- Original Message ----- From: "liu junjun" <ljjlp03.gmail.com>
>> To: "AMBER Mailing List" <amber.ambermd.org>
>> Sent: Wednesday, April 28, 2010 3:05 PM
>> Subject: Re: [AMBER] about skinnb
>>
>>
>>
>> Hi Bob,
>>
>> Thank you for your kind explanation. But I still have question. The
>> manual
>> says:
>> 1. nonbonded list is extended to cut + skinnb
>> 2. van der Waals and direct electrostatic interactions are truncated at
>> cut
>>
>> If nonbond list is extended to cut+skinnb, then how could nonbond
>> interactions be truncated at cut? As I understand by reading subroutine
>> "get_nb_energy", the nonbond energies are calculated by reading
>> whaterever
>> in the nonbond list, which means the nonbond interactions are truncated
>> also
>> at cut+skinnb. Is this correct?
>>
>> Thanks!
>>
>> Junjun
>>
>> On Tue, Apr 27, 2010 at 1:32 PM, Robert Duke <rduke.email.unc.edu> wrote:
>>
>> The skinnb parameter is essentially an optimization parameter; by
>> default
>>> both sander and pmemd add a "skin" to the direct space cutoff length,
>>> monitor the maximum atom movement every dynamics step, and only rebuild
>>> the
>>> pairlist when an atom has moved sufficiently far that an atom within the
>>> actual cutoff distance could be missed in the calculations. If you just
>>> take the default value for skinnb, everything will be fine. At least in
>>> the
>>> past (and maybe currently for sander - I am not sure- but never for
>>> pmemd),
>>> it was possible to specify rebuilding of the pairlist every nsnb
>>> (another
>>> parameter) steps. This is not such a good idea, except perhaps for
>>> minimizations, where I believe with sander it is the default (there is
>>> not a
>>> lot of actual atom movement in a minimization). You can get a more
>>> complete
>>> idea by looking in the manual at the descriptions of the namelist
>>> variables
>>> nsnb, nbflag, and skinnb.
>>> Regards - Bob Duke
>>> ----- Original Message ----- From: "liu junjun" <ljjlp03.gmail.com>
>>> To: <amber.ambermd.org>
>>> Sent: Tuesday, April 27, 2010 1:08 PM
>>> Subject: [AMBER] about skinnb
>>>
>>>
>>>
>>> Hi all,
>>>
>>> I am trying to understand the skinnb parameter. The description for this
>>> parameter in the Manual is:
>>> ----------------
>>> SKINNB
>>> Width of the nonbonded "skin".The direct sum nonbonded list is extended
>>> to
>>> cut + skinnb, and the van der Waals and direct electrostatic
>>> interactions
>>> are truncated at cut. Default is 2.0 .Use of this parameter is required
>>> for
>>> energy conservation, and recommended for all PME runs.
>>> ---------------
>>>
>>> My questions are:
>>> 1. in subroutine "pack_nb_list" or "pack_nb_nogrdptrs" of ew_direct.f,
>>> variable cutoffsq = square of (cut + skinnb)
>>> variable r2 = square of distance between any nonbond pair atoms
>>> The nonbond list is built by setting r2 < cutoffsq. I believe this
>>> corresponds to the description that "The direct sum nonbonded list is
>>> extended to cut + skinnb". However, in subroutine "get_nb_energy", I
>>> notice
>>> that vdw and electrostatic (eel) interactions are calculated by reading
>>> whatever in the nonbond list. It looks like the vdw and eel are
>>> truncated
>>> at
>>> (cut+skinnb) instead of cut as stated in the Manual. Am I missing
>>> something?
>>>
>>> 2. Can anybody please give me any references about "Use of this
>>> parameter
>>> is
>>> required for energy conservation"
>>>
>>> By the way, the source code I am reading is from AMBER8
>>>
>>> Thanks!
>>>
>>> Junjun
>>> _______________________________________________
>>> AMBER mailing list
>>> AMBER.ambermd.org
>>> http://lists.ambermd.org/mailman/listinfo/amber
>>>
>>>
>>>
>>> _______________________________________________
>>> AMBER mailing list
>>> AMBER.ambermd.org
>>> http://lists.ambermd.org/mailman/listinfo/amber
>>>
>>> _______________________________________________
>> AMBER mailing list
>> AMBER.ambermd.org
>> http://lists.ambermd.org/mailman/listinfo/amber
>>
>>
>>
>> _______________________________________________
>> AMBER mailing list
>> AMBER.ambermd.org
>> http://lists.ambermd.org/mailman/listinfo/amber
>>
>
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber



_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Apr 28 2010 - 14:30:03 PDT
Custom Search