From: Goutham (gouthambs_at_gmail.com)
Date: Sun Sep 20 2009 - 15:23:09 CDT

Hey Axel
THanks for your pointers. So I did the following to get working for my case:

1) Used vmd 1.8.7 ..
2) Split my dcd file into components (proteins seperately from the rest). In
the process, I read the dcd in chunks of 5000 frames, extracted the protein
for that chunk, and later on joined the chunks for the protein part using
catdcd.

Now I had a manageable file, and IRspecGUI worked just fine.

Thanks for your pointers.

Best
Goutham

On Sat, Sep 19, 2009 at 8:14 AM, Axel Kohlmeyer <akohlmey_at_gmail.com> wrote:

> On Fri, 2009-09-18 at 20:25 -0700, Goutham wrote:
> >
> > >...and again "it doesn't work" doesn't help at all!!
> >
> > package require signalproc
> > gives an error message:
> > can't find package signalproc
>
> finally!!!
>
> > However :
> > package require specden
> >
> > loads fine without any error.
> >
> > Also the IRspec GUI works fine!
>
> well, that can only mean, that you are
> using an _obsolete_ version of VMD. with
> the current version 1.8.7 of VMD the
> the specden plugin has been moved from
> the ispecgui to the signal proc package
> as part of a large collection of tools to
> work on data sets. there is more to come
> for the next versions of VMD (if i can
> find the time).
>
> > -----
> >
> >
> > doesn't work how? is there an error message? if yes, which
> > one?
> > which version of VMD do you have? which platform?
> >
> >
> >
> > I have tried VMD1.8.4, VMD1.8.6 on linux. The above mentioned specden
> > package loads up on my Mac running VMD1.8.6.
>
> well, there you go. you cannot expect me to read your mind
> and know that you are using a version of VMD that is not
> up-to-date. nevertheless. the "package require specden" only
> loads the spectral density calculator, even if it has become
> part of the signalproc package. package require signalproc is
> provided as a convenience; its only job is to load all integrated
> subpackages. please have a closer look at the docs.
>
>
> > i am confused! how can it make your life easy, if it doesn't
> > work for you???
> >
> >
> >
> > When I tried using the IRspecGUI on the linux machine, it worked fine.
> > I computed spectral density from the veldcd, and the peaks were as I
> > expected. Which is the part that works. The script doesn't work
> > though.
>
> it does work. it _has_ to work (if you use the specden package), because
> all what the GUI does is to compile the command line arguments and
> then call the specden command. the example scripts are simplified
> variants of the GUI plugin.
>
> > However teh actual case for which I want to compute the spectrum, is
> > very big in size (8GB), and my computer freezes when I load the dcd
> > file. I am thinking if I tried a script version, I probably won't have
> > that issue.
>
> this is a reasoning that i cannot follow. if your machine freezes with
> an 8GB file, then you are running out of memory (how much do you have?).
> please note that to call the specden plugin, you have to first compile
> a tcl list of your data. this will have _additional_ memory requirements
> that are more than double that of the 8GB that your dcd file has in
> size, since tcl internally stores all floating point numbers in double
> precision (i.e. at double the storage). together with the overhead for
> linked lists and the auto-converting data type that tcl uses, you may
> need physical memory in excess of 40GB just to be able to run the whole
> data set in one go.
>
> in these cases it is highly advisable to process that data in pieces
> or write a custom analysis tool that avoids the frequent copying. in
> general for processing ultra-large data sets, one always has to consider
> the extra overhead from tcl/python scripting.
>
> cheers,
> axel.
>
> >
> >
> > Thanks
> > Goutham
> >
> >
> >
> >
> >
> >
> >
> >
> >
> >
> > axel.
> >
> >
> > >
> > > Best
> > > Goutham
> > >
> > >
> > >
> > >
> > >
> > > On Fri, Sep 18, 2009 at 1:36 PM, Axel Kohlmeyer
> > <akohlmey_at_gmail.com> wrote:
> > >>
> > >> On Fri, Sep 18, 2009 at 3:53 PM, Goutham
> > <gouthambs_at_gmail.com> wrote:
> > >> > Hey All,
> > >> >
> > >> > Does any of you know of any packages to calculate
> > (velocity)
> > >> > auto-correlation? I saw a "signalproc" package here:
> > >> >
> > >> > http://www.ks.uiuc.edu/Research/vmd/plugins/signalproc/
> > >> >
> > >> > But this somehow doesn't work. Will be nice if any one
> > can give me
> > >> > pointers
> > >> > to it.
> > >>
> > >> hey goutham,
> > >>
> > >> _what_ doesn't work? and _how_ does it not work?
> > >> just saying that something doesn't work is not very
> > helpful.
> > >> more often than not these problems are a cases of PEBCAC.
> > >>
> > >> the specden command in that package computes spectral
> > >> densities directly from convoluted data and can do this
> > with
> > >> any data, including velocities. but you don't have the
> > intermediate
> > >> VACF but assemble the data directly in fourier space. this
> > is
> > >> very adventageous for large data sets and removes the
> > arbitrariness
> > >> in traditional VACF based spectraldensities (and avoids the
> > O(N**2)
> > >> scaling with correlation length to boot). i have a
> > truckload of those
> > >> spectra lying around on my harddrive, and they all look
> > great and
> > >> have worked fine. ;)
> > >>
> > >> other than that, programming a VACF directly is
> > embarrassingly
> > >> simple in VMD, it just is dog slow.
> > >>
> > >> cheers,
> > >> axel.
> > >>
> > >> >
> > >> > Best
> > >> > Goutham
> > >> >
> > >> >
> > >>
> > >>
> > >>
> > >> --
> > >> Dr. Axel Kohlmeyer akohlmey_at_gmail.com
> > >> Institute for Computational Molecular Science
> > >> College of Science and Technology
> > >> Temple University, Philadelphia PA, USA.
> > >
> > >
> >
> >
> >
> >
> > --
> >
> > Dr. Axel Kohlmeyer akohlmey_at_gmail.com
> > Institute for Computational Molecular Science
> > College of Science and Technology
> > Temple University, Philadelphia PA, USA.
> >
> >
> --
> Dr. Axel Kohlmeyer akohlmey_at_gmail.com
> Institute for Computational Molecular Science
> College of Science and Technology
> Temple University, Philadelphia PA, USA.
>
>