I have occasionally seen seg fault error from LAPACK routine:DSTEMR. I tried saving the tridiagonal into a file and loaded with matlab and did ``eig''. Nothing was wrong in matlab. So, I assume that this is a bug in Lapack. It actually is confirmed as a bug. see http://www.netlib.org/lapack/Errata/index2.html#_strong_span_class_red_bug_159_span_strong_dstemr_consolidates_bugs_113_and_126.

So, in ratlanNr and cheblanNr, we comment it out and use dsetv. When this bug is fixed, we may change it back.