Closed uhoefel closed 2 years ago
That's right, let me submit your fix.
Could you please verify https://github.com/luhenry/netlib/pull/15. If it seems right to you, I'll do a 3.1.0 release with that fix.
Seems correct as far as I can judge. I think the checks some lines above (e.g. 4962-4965, but also in dgeev) probably should be flipped to "V"
as well:
if (lsame("N", jobvl))
requireNonNull(vl);
if (lsame("N", jobvr))
requireNonNull(vr);
-->
if (lsame("V", jobvl))
requireNonNull(vl);
if (lsame("V", jobvr))
requireNonNull(vr);
There seems to be an issue in
AbstractLAPACK.java
indgeev
. My test input works withnetlib.fommil
and looks like this:To my understanding, that should give the following eigenvalues in
wr
:$$\lambda_{1,2} = \frac{\pm\sqrt{6}+4}{4}$$
My suspicion is that lines 338-341
should read
The documentation of LAPACK states for
vl
that (similar forvr
)so I would at least not intuit that its length matters (or I may very well miss something).
PS: Thanks for fixing the split packages!