Retrieve the LAPACK routine ` sgeevx` from netlib
(see section 11 below on how to do this). Write a
modified version, which we will call ` sgesvn`, which does * not*
do pivoting. Run both routines on randomly generated test matrices,
and compare the error bounds they compute. Modify the test matrices
to make the (1,1) entry very small. How do the error bounds compare now?
Note that LAPACK produces other error bounds besides the one described
here.