Open dlfivefifty opened 9 years ago
Got to use QZ for numerical stability, which should automatically balance for you(?). This requires writing things as a generalized eigenvalue problem. Yet, I do think Vanni, Yuji, or any members of the Chebfun team would recommend QR for everyday use.
The reason I had to write my own colleaguebalance! Was to use LAPacks Hessenberg eigvals routine. (LAPacks standard eigvals does balancing automatic but the faster Hessenberg routine does not.). I'd expect QZ to be even slower than just using the standard eigvals.
Sent from my iPad
On 17 Apr 2015, at 10:50 pm, Alex Townsend notifications@github.com wrote:
Got to use QZ for numerical stability, which should automatically balance for you(?). This requires writing things as a generalized eigenvalue problem. Yet, I do think Vanni, Yuji, or any members of the Chebfun team would recommend QR for everyday use.
— Reply to this email directly or view it on GitHub.
Should probably be 2^k scaling.
Two papers on the subject of balancing (companion only) are: http://www.siam.org/meetings/la03/proceedings/lemonnid.pdf
Another paper related to colleague matrices: http://eprints.ma.man.ac.uk/2182/01/covered/MIMS_ep2014_49.pdf