Re: expected speed up after fixing atoms?

From: Jose Borreguero (borreguero_at_gmail.com)
Date: Wed Sep 24 2014 - 14:26:06 CDT

A million thanks for bringing this to my attention, sir.

On Wed, Sep 24, 2014 at 3:15 PM, JC Gumbart <gumbart_at_ks.uiuc.edu> wrote:

>
> http://www.ks.uiuc.edu/Research/namd/2.10b1/ug/node27.html#SECTION00086300000000000000
>
>
> - *fixedAtomsForces *[image: $ <$] are forces between fixed atoms
> calculated? [image: $ >$]
> *Acceptable Values: *on or off
> *Default Value: *off
> *Description: *Specifies whether or not forces between fixed atoms are
> calculated. This option is required to turn fixed atoms off in the middle
> of a simulation. These forces will affect the pressure calculation, and you
> should leave this option off when using constant pressure if the
> coordinates of the fixed atoms have not been minimized. The use of constant
> pressure with significant numbers of fixed atoms is not recommended.
>
>
> On Sep 24, 2014, at 3:07 PM, Aron Broom <broomsday_at_gmail.com> wrote:
>
> Yes, that's true that there isn't a point in calculating the forces
> between two atoms that are both fixed. It may be that NAMD will bypass
> that, but I think that it may currently not be coded that way. The reason
> being, that in general, a situation where it would actually be beneficial
> would almost never arise in a reasonable simulation.
>
> For instance, say you have 10,000 atoms, and you want to fix 1000 of
> them. The total number of atom-atom interactions (assuming no cutoff) is
> 10,000^2, and the total number you could potentially save if the
> fixed-fixed ones were ignored is 1000^2. In that case, you are only saving
> on 1% of the calculations. So, implementing a check for the atoms being
> fixed, which itself will take some compute time, isn't really worthwhile
> until you have quite a substantial amount of the system fixed, at which
> point, you are quite likely simulating something so unphysical that the
> results may not be meaningful anyway.
>
> On Wed, Sep 24, 2014 at 2:41 PM, Jose Borreguero <borreguero_at_gmail.com>
> wrote:
>
>> Isn't it the case that calculations involving interactions between fixed
>> atoms are bypassed? What's the point of calculating forces between atoms
>> that won't move?
>>
>> On Wed, Sep 24, 2014 at 2:10 PM, Aron Broom <broomsday_at_gmail.com> wrote:
>>
>>> I don't think you'd expect any speed up actually. Fixing the atoms just
>>> means they don't move, but they are still involved in all the pair-wise
>>> force calculations for the system which is the vast bulk of the
>>> computational work.
>>>
>>> On Wed, Sep 24, 2014 at 1:44 PM, Jose Borreguero <borreguero_at_gmail.com>
>>> wrote:
>>>
>>>> Dear NAMD users,
>>>>
>>>> I have a system where I can fix the positions of more than half the
>>>> atoms without affecting much the dynamics I'm interested in. I was
>>>> wondering what is (approximately) the expected speed up in the simulation
>>>> after I do this. Even more important: What will happen if I run a parallel
>>>> job with this system? Will some of the CPU end up doing no calculations if
>>>> the fixed atoms are assigned to those CPU's?
>>>>
>>>> Best regards,
>>>> Jose Borreguero
>>>>
>>>
>>>
>>>
>>> --
>>> Aron Broom M.Sc
>>> PhD Student
>>> Department of Chemistry
>>> University of Waterloo
>>>
>>
>>
>
>
> --
> Aron Broom M.Sc
> PhD Student
> Department of Chemistry
> University of Waterloo
>
>
>

This archive was generated by hypermail 2.1.6 : Thu Dec 31 2015 - 23:21:15 CST