Re: [AMBER] cpptraj rotdif parameters

From: David A Case <>
Date: Thu, 10 Jul 2014 20:17:44 -0400

On Thu, Jul 10, 2014, Alegrete, Matthew wrote:

> I am currently trying to analyze the rotational diffusion in a few
> simulations, but am having trouble understanding some of the parameters
> that go into the cpptraj script for the rotdif action. I have already
> consulted the documentation as well as other emails in the reflector and
> haven't found anything.
> In the example given in the manual, "nvecs" was given equal to the number
> of frames. Is that generally the way this number should be picked? Was
> there any significance in the example of picking "ncorr" to be 90% of the
> frames?

nvecs is the number of directions to sample; it has nothing to do with the
number of frames in the trajectory. A value of 500 should be plenty, but
you can compare results you get with different values here. (The example of
100 directions is probably enough as well, but I usually use more.)

Time correlation functions are only accurate up to small fraction of the
number of frames. In this case, setting ncorr to a large value doesn't really
hurt, since only values between ti and ti are evaluated. So you just waste a
small amount of time if you compute time correlation functions at values large
than tf.

> Are the times "ti" and "tf" arbitrarily picked limits for integrating the
> correlation function, or should they be the times you want to sample the
> trajectory between? Also, should the "dt" match the time step used in the
> simulation, or is that again an arbitrary integration step?

ti and tf are the times used to evalate the correlation function. Ordinarily,
it should be fine to set ti to zero; you can play some with tf. dt needs to
be the time separation of the snapshots you are using.

The example in the manual *is* misleading in terms of the number of frames
used: with only 100 frames at a 0.002ns spacing, one only has 0.2 ns of total
sampling, which is too short for most analyses. More "usual" would be
to have 50 or more ns of sampling, and to set tf to a larger value, say 5 ns
or more. But a lot depends on the size of the system, and its rotational
correlation time. Don't be afraid to experiment.

> I am suspicious that short trajectories may cause problems with
> convergence. Increasing the number of iterations (itmax) seems to help the
> small anisotropic analysis converge, but the amoeba minimizer is where
> convergence keeps failing.

I don't think I've ever seen the amoeba minimizer fail to converge. That may
be a problem arising from insufficient sampling.


AMBER mailing list
Received on Thu Jul 10 2014 - 17:30:02 PDT
Custom Search