[gmx-developers] Scaling velocities in Velocity Verlet

[David van der Spoel]

On 2011-12-03 21:37, Shirts, Michael (mrs5pt) wrote:
> Hi, all-
>
>> The problem is that the change in Ekin comes one step too late. I tried
>> to deduce the algorithm from the code (4.5 code base) and deduced the
>> following MD algorithm:
>
> It's rather difficult to follow currently because of the need to bookkeep
> both md and md-vv + plus correct pressure control w/constraints. I have high
> hopes for improving this for 5.0.
>
>> v(t) = v(t-dt/2) + a(t) dt/2
>>
>> v'(t) = lambda v(t)<======= my addition
>>
>> v(t+dt/2) = v'(t) + a(t) dt/2
>>
>> r(t+dt) = r(t) + v(t+dt/2) dt
>>
>> However, the Ekin that is printed is the uncorrected one even though I
>> scale both ekind->ekin and ekind->tcstat[x].ekinf .
>>
>> Is the algorithm correct as I write it above? Is the Ekin(t) already
>> stored at the beginning of the time step?
>
> If you add your modification as an option to update_tcouple, and operate on
> ekind directly (not the velocities), in the same way that berendsen_tcouple
> does, then the bookkeeping should take care of the velocity rescaling
> appropriately.
Close, but no cigar.
Here's the output from g_energy:

@ s0 legend "Potential"
@ s1 legend "Kinetic En."
@ s2 legend "Total Energy"
     0.029600  -35405.431557  6460.241804  -28945.189753
     0.029800  -35407.596247  6462.418642  -28945.177606
     0.030000  -35327.293038  6466.347370  -28860.945667
     0.030200  -35330.858001  6387.271571  -28943.586430
     0.030400  -35334.701690  6391.114750  -28943.586940

The jump in Potential is at t = 0.03, in Kinetic it is at t = 0.302. As 
you see the Total energy has a peak at t = 0.03, but it does come back 
almost to the right level at the next step. This all makes me think I 
should be scaling the previous velocities, hence my questions about the 
algorithm.



>
> If this _doesn't_ work, then there's some tweaks that got lost in the
> different versions that are floating around, and I can either put the
> changes into 4.6 alone, or just add them in with the 4.6 merge.
>
> Best,
> ~~~~~~~~~~~~
> Michael Shirts
> Assistant Professor
> Department of Chemical Engineering
> University of Virginia
> michael.shirts at virginia.edu
> (434)-243-1821
>
>
>
>> v(t) = v(t-dt/2) + a(t) dt/2
>>
>> v'(t) = lambda v(t)<======= my addition
>>
>> v(t+dt/2) = v'(t) + a(t) dt/2
>>
>> r(t+dt) = r(t) + v(t+dt/2) dt
>>
>> However, the Ekin that is printed is the uncorrected one even though I
>> scale both ekind->ekin and ekind->tcstat[x].ekinf .
>>
>> Is the algorithm correct as I write it above? Is the Ekin(t) already
>> stored at the beginning of the time step?
>>
>>
>> --
>> David van der Spoel, Ph.D., Professor of Biology
>> Dept. of Cell&  Molec. Biol., Uppsala University.
>> Box 596, 75124 Uppsala, Sweden. Phone: +46184714205.
>> spoel at xray.bmc.uu.se    http://folding.bmc.uu.se
>> --
>> gmx-developers mailing list
>> gmx-developers at gromacs.org
>> http://lists.gromacs.org/mailman/listinfo/gmx-developers
>> Please don't post (un)subscribe requests to the list. Use the
>> www interface or send it to gmx-developers-request at gromacs.org.
>


-- 
David van der Spoel, Ph.D., Professor of Biology
Dept. of Cell & Molec. Biol., Uppsala University.
Box 596, 75124 Uppsala, Sweden. Phone:	+46184714205.
spoel at xray.bmc.uu.se    http://folding.bmc.uu.se