[Scipy-tickets] [SciPy] #1503: isolve and arpack test failures on OS X

SciPy Trac scipy-tickets@scipy....
Fri Aug 26 10:15:36 CDT 2011


#1503: isolve and arpack test failures on OS X
---------------------------------+------------------------------------------
 Reporter:  rgommers             |       Owner:  wnbell
     Type:  defect               |      Status:  new   
 Priority:  normal               |   Milestone:  0.10.0
Component:  scipy.sparse.linalg  |     Version:  0.9.0 
 Keywords:                       |  
---------------------------------+------------------------------------------

Comment(by pv):

 The BiCG failure is probably also to be expected --- it should be more
 sensitive than QMR to stagnation problems.

 The attached dump matrix is indeed not invertible by `gmres` with
 `tol=eps`, which is the default for `eigs`. Using such a tolerance in an
 iterative method is iffy; it could be multiplied with a fudge factor, e.g.
 2, as there is anyway a mismatch between the meaning of `tol` in the inner
 GMRES (tolerance on residual, not on accuracy of solution) and ARPACK
 (tolerance on eigenvalues). I guess it would work if `tol =
 np.finfo(...).eps` is changed to `tol = 2*np.finfo(...).eps` on lines 929,
 960 in `scipy/sparse/linalg/eigen/arpack.py`?

-- 
Ticket URL: <http://projects.scipy.org/scipy/ticket/1503#comment:4>
SciPy <http://www.scipy.org>
SciPy is open-source software for mathematics, science, and engineering.


More information about the Scipy-tickets mailing list