Ticket #8710 (closed defect: fixed)
eigenmatrix_right returns inconsistent results for eigenvectors
| Reported by: | rbeezer | Owned by: | jason, was |
|---|---|---|---|
| Priority: | major | Milestone: | sage-4.4 |
| Component: | linear algebra | Keywords: | |
| Cc: | jhpalmieri, AlexGhitza | Work issues: | |
| Report Upstream: | N/A | Reviewers: | William Stein, John Palmieri |
| Authors: | Rob Beezer | Merged in: | sage-4.4.alpha1 |
| Dependencies: | Stopgaps: |
Description
Doctests introduced in #4756 return the negative of certain eigenvectors on certain hardware.
See initital discussion at
http://groups.google.com/group/sage-release/browse_thread/thread/9136569bd1c67f6
Attachments
-
trac_8710-eigenvector-doctest.patch
(6.1 KB) -
added by rbeezer 3 years ago.
Change History
comment:2 Changed 3 years ago by rbeezer
- Cc AlexGhitza added
- Status changed from new to needs_review
- Authors set to Rob Beezer
This patch massages the doctests that were causing failures for 4.4.alpha0 on the Skynet machine, sextus. Its not pretty, but I hope the results are now hardware-neutral.
Alex - you reviewed the original ticket (#4756), so maybe this would be an easy review for you?
John - I don't know if it is easy for you to test this on sextus in advance of merging it? Sounds like it will be a while before I have that kind of access.
Rob
comment:3 follow-up: ↓ 4 Changed 3 years ago by was
- John will have to test, since he has the build on sextus.
- The doctests actually look much nicer normalized to have first entry 1!
(I would give this a positive review if this works.)
comment:4 in reply to: ↑ 3 Changed 3 years ago by rbeezer
Replying to was:
- The doctests actually look much nicer normalized to have first entry 1!
The output looks nicer because this matrix is out of my textbook and was *designed* to have nice-looking answers. I don't like the hard-to-decipher code that gets you there, but that's the way it goes. Thanks for having a look and for the advice on getting this squared away.
Rob
comment:5 Changed 3 years ago by jhpalmieri
- Reviewers set to William Stein, John Palmieri
Tests pass on sextus. I'll test it on a few more machines, and if it works, I'll give it a positive review and merge it into 4.4.alpha1.
comment:6 Changed 3 years ago by rbeezer
Hi John,
Did this eventually past muster on skynet, or does it need more testing?
I still haven't done the SciPy tests I'd like to do skynet yet, but hope to get to that soon.
Rob
