Page 1 of 1

MRRR for symmetric EIG(A,B)

PostPosted: Wed Oct 05, 2016 6:16 am
by int128
I would like to contribute routines for solving symmetric EIG(A,B) by MRRR.

Now LAPACK has MRRR and D&C routines to solve standard eigenproblem, but only D&C for generalized.
MRRR for generalized eigenproblem would be a natural addition to LAPACK as it computes eigenvectors faster compared to D&C.
Additionally, I saw that such routines might be of interest in some fields [1].

Guided by this idea, I wrote MRRR-functions: DSYGVR, ZHEGVR (attached).
Both are analogous to existing Divide & Conquer routines: DSYGVD, ZHEGVD.

Would it be interesting to include the routines to LAPACK?

Forking, adding new routines, updating CMake files are easy.
But I have little experience with unit tests in LAPACK.

Of course, I have tested the routines in my environment.

[1] Metallic Systems: A Quantum Chemist's Perspective