Re: bug in expm

[Ross A. Lippert]

The bug in expm results from an error in the dMatrix.solve() function
called by
expm.  When info==-2 after the fortran call the flow of control takes a
path which
assigns nothing to retval. I am not sure what the correct fix is, but
that is the
heart of the problem.  Should I just do "retval = b" at about line 962
of dMatrix.cc?
or is there a more intelligent fix?

There may be a second issue in how the pade approximation handles
nilpotent matrices
but that will come later.


-r