[MMTK] Matrices not aligned

Lars Skjærven larsss at gmail.com
Fri Oct 31 18:04:24 UTC 2008

Hi again, and thanks for your quick reply and testing.

I have been testing this myself a couple of times today without
getting much forward. I managed to compile and run MMTK (with numpy)
successfully on another desktop computer (64-bit linux) with the same
versions mentioned above.

For the server where I have the trouble (also a 64-bit linux box) I
went over the compilation steps again to make note of any error
- The python compilation complained one time: "Can't locate Tcl/Tk
libs and/or headers". However, it continued, and finished
- numpy compilation went fine, and I performed the numpy test (with
nose). ended up with "OK (KNOWNFAIL=1, SKIP=1)".
- scientific python: last line after "python setup.py install":
"error: Invalid argument". However, all test finished successfully.
- MMTK: only "basic_tests.py" succeeded. All other test fails with the
fatal error mentioned above.

I have compiled python and mmtk, but with Numeric instead of numpy, on
this machine plenty of times earlier without encountering trouble.

The benefits from using numpy instead of numeric is the increase of
the allowed size of the force constant matrix, right? at the moment,
using my numeric version of MMTK I encounter "segmentation fault" when
calculating all-atom normal modes on a 7500 atom protein (510
residues) using RigidMotionSubspace and SubspaceNormalModes. Smaller
proteins works fine. By using numpy instead, I was hoping I would end
up with something else than seg.fault (i.e. the modes).

Any tips for solving this are highly appreciated.


On Fri, Oct 31, 2008 at 10:21 AM, Konrad Hinsen <hinsen at cnrs-orleans.fr> wrote:
> On 31.10.2008, at 07:59, Konrad Hinsen wrote:
>>> Both Python 2.5 and 2.4 results in the same error. Have you seen this
>>> before? Is there a bug, or did I choose some incompatible versions ?
>> Not that I know, though I didn't install NumPy 1.2.1 yet.
> I just did, using Python 2.5.1 and everything else in the same version as
> you have. On a 64-bit Linux machine, this works fine; all the MMTK test
> cases give the expected results.
> NumPy 1.2.1 will print a "DeprecationWarning" for a couple of calls made by
> MMTK because they still use the old Numeric C-API that will ultimately
> disappear. The warnings don't do any harm, but if they bother you, you
> should be able to switch them off using the warnings module (though I
> haven't tried).
> Konrad.
> --
> ---------------------------------------------------------------------
> Konrad Hinsen
> Centre de Biophysique Moléculaire, CNRS Orléans
> Synchrotron Soleil - Division Expériences
> Saint Aubin - BP 48
> 91192 Gif sur Yvette Cedex, France
> Tel. +33-1 69 35 97 15
> E-Mail: hinsen at cnrs-orleans.fr
> ---------------------------------------------------------------------

More information about the mmtk mailing list