AMBER Archive (2005)

Subject: Re: AMBER: Is the MD simulation normal

From: Bill Ross (ross_at_cgl.ucsf.edu)
Date: Sun May 29 2005 - 10:33:13 CDT

> Seeing form the plot of RMSD as a function of time, it take 300 ps or
> so to reach the equibilbration. I think there may exists problems.

By the way, did you fit to the initial structure? This will
remove any translation and rotation effects.

> I will perform two MD as the control.=20
>
> First, I think I can use amber7 to perform One MD that was earlier
> performed ok. The reason that I paln to perform this MD is that, When
> using amber7, I have meet no problem with the MD.

How much experience are you basing this on? If you want to
compare amber7 and amber8, I'd suggest running a group of
trajectories from slightly different initial conditions using
both programs and comparing the ensemble properties in detail.
This is because every trajectory is an individual walk in the
garden of the system's Hamiltonian, and is subject to minor
chaotic influences such as the order of the arithmetic operations
due to the roundoff behavior of floating point.

> Second, about this MD that we disscussed, I will use constraint NPT MD
> for the first 300 ps and with a following 1 ns NPT MD without
> constraints. If the result generated in this MD is similar to the
> original results, it indicates that the disscussed MD results are
> completely right.

Within some confidence level (i.e. nothing is completely certain
unless we can sample for longer than is humanly possible).

You could also generate different trajectories by using random
initial velocities and varying the random number seed and apply
the same equilibration protocol, to see how variable your system is.
However, even if the results are all the same, you still cannot be
sure that it is just due to one of many interesting local minima.

Perhaps others can suggest references on the sampling problem.

Bill
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu