[gmx-users] Energy minimisation goes to several values

[Justin Lemkul]

On 6/25/15 9:45 PM, Kevin C Chan wrote:
> Dear Users,
>
> I am energy minimising a quite large solvated system containing protein and
> lipids (~800,000 atoms). I used to fix components of the system in order to
> speed-up energy minimisation and sometimes it is easier to debug such
> processes. Here is my protocol:
> 1. fix all except water and so to minimise water
> 2. fix water and then minimise all the rest atoms
> 3. fix nothin and then minimise the whole system
>
> While monitoring the energy of the system thought minimisations, it goes
> fine for step 1 and 2 and converged after just few hundred steps. However
> it goes back to several higher values of energy (bouncing between the
> values) and they started to increase very slowly for step 3. This makes no
> sense to me and did anyone have a similar experience?
>
> There are two unusual points:
> 1. The system energy drops suddenly instead of decreased gradually during
> step2 and then stays at a constant value.
> 2. If I use the resulting structure from step3 to proceed a, say, heating
> process, it simply blows up.
>
> To be clear, my system was solvated and auto-ionized using VMD tools and
> some water inside the membrane has been directly deleted. Backbone of the
> protein and phosphorus atoms of the membrane are under a
> position constraint during all the minimisations. I was choosing conjugate
> gradient for minimization.
>

Does a "normal" minimization (just one overall minimization with nothing fixed) 
yield a stable starting point?  Fixing atoms (using freezegrps?) often actually 
impedes minimization.

-Justin

-- 
==================================================

Justin A. Lemkul, Ph.D.
Ruth L. Kirschstein NRSA Postdoctoral Fellow

Department of Pharmaceutical Sciences
School of Pharmacy
Health Sciences Facility II, Room 629
University of Maryland, Baltimore
20 Penn St.
Baltimore, MD 21201

jalemkul at outerbanks.umaryland.edu | (410) 706-7441
http://mackerell.umaryland.edu/~jalemkul

==================================================