QR decomp for complex matrix

[cottrell]

I'm used to the functions dgeqrf and dorgqr for real matrices. Recently I've been working with complex matrices, and I see there's a zgeqrf, but no zorgqr to construct the m x n Q matrix from the raw materials provided by zgeqrf. I guess there must be a reason for that -- can you tell me what it is?

(I've written code which does the job on Q for the complex case, but I'm sure it's horribly inefficient compared with the clever stuff that dorgqr does.)

[Julien Langou]

ZUNGQR

[cottrell]

Perfect! I missed that, needless to say, but then most z* functions have the same name as the corresponding d* ones, apart from the first letter.

[cottrell]

Ah, but now I see: ZUNGQR does not have the same signature as DORGQR (data types aside).

However, it would be helpful if the doc for ZGEQRF had a cross reference to ZUNGQR.

[cottrell]

Oof, sorry, too hasty! Actually the signatures of DORGQR and ZUNGQR are the same, modulo the data types for A, TAU and WORK. So (a less pressing question) is there a reason why the names of the functions differ, other than in their first letter as usual?

[Julien Langou]

OR = Orthogonal
UN = unitary

In general we say orthogonal for real matrices, and we say unitary for complex matrices.

( For the record, my preference would be to use DUNxxx and ZUNxxx. )

[cottrell]

Thanks for the explanation, Julien. My only remaining point is that it would be nice to have cross-references from the primary QR functions to those that extract or construct the actual Q matrix, in the online LAPACK documentation.