I'm in the process of re-running the full PDL
install process with cygwin/XP to verify operation
for the latest CVS code and to collect cygwin
specific notes for the upcoming bug-fix release.

Starting with a default build from the current
cvs code, all tests pass except for test #20 in
t/matrixops.t which fails with the following
message:

>> Failure in hqr2 function. Do not trust the given eigenvectors and -values
>> eigensum for the 8x8: nan
>> not ok 20
>> # Failed test 20 in t/matrixops.t at line 118
>> #  t/matrixops.t line 118 is: ok($esum == 61.308);

I'm ignoring the problem as I continue my run-through of
the install process but it looks like the root cause of this
problem can not be solved with a simple fix to eigens.

The eigens() routine does not handle complex results
nor the general matrix case where left- and right-
eigenvectors may be needed (returning NaNs is not
sufficient).

I suggest the following triage:

1. Fix CVS eigens() and documentation to handle symmetric
matrix inputs only (as in 2.4.2).
2. Remove broken support for non-symmetric matrix eigen
calculations.
3. Put general matrix eigen- support on the list for the
next major release.

One way forward would be to use the corresponding functions
from the CLAPACK library with possible addition of ATLAS
support to PDL as in the following links:

http://www.netlib.org/lapack/lug/lapack_lug.html
http://math-atlas.sourceforge.net/faq.html

--Chris


_______________________________________________
Perldl mailing list
[email protected]
http://mailman.jach.hawaii.edu/mailman/listinfo/perldl

Reply via email to