[MMTK] Normal Modes
vanitha at cs.wisc.edu
Fri Mar 24 17:03:15 CET 2006
I ran the normal mode calculation on a protein (hetero dimer) with 400
residues. It's past 24 hours and it's still running. I'm working on a
Pentium 4 with Linux. Is this expected?
> On Mar 23, 2006, at 16:36, vanitha at cs.wisc.edu wrote:
>> 1) The version of MMTK I have does not have a method called
>> EnergeticModes(..). How is this different from NormalModes(..)?
> EnergeticModes is in MMTK 2.5. Just use NormalModes with 2.4 (which
> is VibrationalModes in 2.5 terminology).
>> 2) Also, if I'm understanding you correctly here, if one of the
>> has M atoms and the other has N atoms, then I'm going to end up with a
>> 3(N+M) x 1 vector. In what order will these be in? Would there be a
> The order depends on how you construct the system, on your Python
> version, on your MMTK version, and perhaps even more. Better don't
> make any assumption about atom oder. You don't need to.
>> one-to-one correspondence between the gradient of the force field with
>> respect to the co-ordinates of the molecules and the normal mode
> Yes, obviously. Everything done within one universe has that
> universe's atom order.
>> I just want to to make sure that I'm not adding apples & oranges :-)
> You would get an arror message if you tried :-)
>> 3) I just need the first few non-trivial modes to begin with. Do
>> you know
>> if there is a way to compute just these and save MMTK the trouble of
>> having to compute them all?
> Yes, but you will save time and memory only if your system is quite
> big. There are two options:
> 1) Use an iterative eigenvector algorithm to obtain the first few
> modes. Those are the same modes as with a standard calculation. This
> is of interest mainly if you are running into memory limitations.
> Time-wise it rarely gives a substantial advantage.
> 2) Use a subspace (e.g. FourierSubspace). A well-chosen subspace will
> give you a good approximation to the first normal modes, but they
> will not be exactly the same. However, you do gain CPU time and
> memory for large systems.
> My suggestion is to use the general rule for computing: don't
> optimize before you have a performance problem. If standard normal
> modes are fine for your application, stay with them.
> Konrad Hinsen
> Laboratoire Léon Brilloui
More information about the mmtk