Convergence errors in ggev and gges with complex double element types

[pablosanjose]

The generalized Schur and Eigenvalue routines gges and ggev sometimes fail to converge with some matrices A, B of double-precision complex element type. I post some example matrices below.

The error is of the type ERROR: LAPACKException(16) in Julia, or the more informative numpy.linalg.LinAlgError: generalized eig algorithm (ggev) did not converge (LAPACK info=16) in Python. That means insufficient convergence of the ggev or gges routines.

Some empirical observations.

• I didn't encounter the problem with element types other than double-precision complex (i.e. the issue seems to be only in zgges and zgeev)
• Sometimes the failing matrices are themselves real, and casting their element type to single or double precision floats doesn't trigger the error.
• Moreover, the error has been seen to sometimes arise only on macos but not on linux or the converse (for some matrices A,B), although for most failing matrices, the error is triggered in both OSs.
• The error was reproduced with an Intel i7, an Intel Xeon and a Ryzen Threadripper (the latter was not done by me).
• The same error was reproduced calling LAPACK from Julia 1.5.3, Julia 1.7, Python 2.7 and Python 3.9. The same type of error was obtained in all of these, but not in all os/language combinations (see below)
• This type of convergence problem does not arise in any of my tests (including the matrices below) using Intel's MKL implementation (2018 version), or Mathematica's (v12.1)
• I believe there is a way to compute the mutual condition number of each pair of matrices, but I haven't gotten round to figuring that out yet.
• [EDIT] I confirmed the issue is triggered also calling the netlib-LAPACK libraries directly (see post below), so the issue doesn't seem to be in the OpenBLAS libraries bundled with Julia/Python, but rather in the reference zgges et al. family.

I enclose three examples of problematic matrices, both for julia and for python. I tested a total of six environments. Machine "MA" is an Intel(R) Core(TM) i7-8559U CPU @ 2.70GHz running macos Big Sur. Machine "MB" is an Intel(R) Xeon(R) CPU X5650 @ 2.67GHz running Debian Buster. I tested three languages: (1) = Python 2.7 and (2) = Python 3.9 and (3) = Julia 1.5. The results were as follows ("works" means that it does not error, didn't check correctness of the result in all cases)

Example 1

• Fails: MA(1), MA(2), MA(3)
• Works: MB(1), MB(2), MB(3)

Example 2

• Fails: MA(2), MA(3), MB(1), MB(2), MB(3)
• Works: MA(1)

Example 3

• Fails: MB(3)
• Works: MA(1), MA(2), MA(3), MB(1), MB(2)

Matrix examples

The code that errors in Julia is

using LinearAlgebra
eigen(A, B)  # or schur(A, B)

The code that errors in Python is

import numpy as np
from scipy.linalg import eig
eig(A, b=B)

Example 1

Python code

A1 = np.array([[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0], [3.7796350217469814, -3.3125635598133054, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.418270043493963, -6.625127119626611, 0.0, 0.0, 0.0, 0.0, 0.0, -1.0], [-3.312563559813306, 3.779635021746982, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.625127119626612, 6.418270043493964, -1.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 3.7796350217469814, 0.0, 0.0, -3.3125635598133054, 0.0, 0.0, 0.0, -1.0, 6.418270043493963, 0.0, 0.0, -6.625127119626611, 0.0, 0.0], [0.0, 0.0, 0.0, 3.779635021746982, -3.312563559813306, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.418270043493964, -6.625127119626612, 0.0, -1.0, 0.0], [0.0, 0.0, 0.0, -3.3125635598133054, 3.7796350217469814, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.625127119626611, 6.418270043493963, -1.0, 0.0, 0.0], [0.0, 0.0, -3.312563559813306, 0.0, 0.0, 3.779635021746982, 0.0, 0.0, 0.0, 0.0, -6.625127119626612, 0.0, -1.0, 6.418270043493964, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 3.7796350217469814, -3.3125635598133054, 0.0, 0.0, 0.0, -1.0, 0.0, 0.0, 6.418270043493963, -6.625127119626611], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -3.312563559813306, 3.779635021746982, -1.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.625127119626612, 6.418270043493964]]) + 0.0j
B1 = np.array([[1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -3.7796350217469814, 3.312563559813306, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 3.3125635598133054, -3.779635021746982, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -3.7796350217469814, 0.0, 0.0, 3.312563559813306, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -3.779635021746982, 3.3125635598133054, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 3.312563559813306, -3.7796350217469814, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 3.3125635598133054, 0.0, 0.0, -3.779635021746982, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -3.7796350217469814, 3.312563559813306], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 3.3125635598133054, -3.779635021746982]]) + 0.0j

Julia code

A1 = ComplexF64[0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0; 3.7796350217469814 -3.3125635598133054 0.0 0.0 0.0 0.0 0.0 0.0 6.418270043493963 -6.625127119626611 0.0 0.0 0.0 0.0 0.0 -1.0; -3.312563559813306 3.779635021746982 0.0 0.0 0.0 0.0 0.0 0.0 -6.625127119626612 6.418270043493964 -1.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 3.7796350217469814 0.0 0.0 -3.3125635598133054 0.0 0.0 0.0 -1.0 6.418270043493963 0.0 0.0 -6.625127119626611 0.0 0.0; 0.0 0.0 0.0 3.779635021746982 -3.312563559813306 0.0 0.0 0.0 0.0 0.0 0.0 6.418270043493964 -6.625127119626612 0.0 -1.0 0.0; 0.0 0.0 0.0 -3.3125635598133054 3.7796350217469814 0.0 0.0 0.0 0.0 0.0 0.0 -6.625127119626611 6.418270043493963 -1.0 0.0 0.0; 0.0 0.0 -3.312563559813306 0.0 0.0 3.779635021746982 0.0 0.0 0.0 0.0 -6.625127119626612 0.0 -1.0 6.418270043493964 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 3.7796350217469814 -3.3125635598133054 0.0 0.0 0.0 -1.0 0.0 0.0 6.418270043493963 -6.625127119626611; 0.0 0.0 0.0 0.0 0.0 0.0 -3.312563559813306 3.779635021746982 -1.0 0.0 0.0 0.0 0.0 0.0 -6.625127119626612 6.418270043493964]
B1 = ComplexF64[1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -3.7796350217469814 3.312563559813306 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 3.3125635598133054 -3.779635021746982 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -3.7796350217469814 0.0 0.0 3.312563559813306 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -3.779635021746982 3.3125635598133054 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 3.312563559813306 -3.7796350217469814 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 3.3125635598133054 0.0 0.0 -3.779635021746982 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -3.7796350217469814 3.312563559813306; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 3.3125635598133054 -3.779635021746982]

Example 2

Python code

A2 = np.array([[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -1.0, 0.0, 0.0, 0.0, 0.0, 0.0, -2.62, 0.0, 0.0, 0.0, 0.0, 0.0, -1.0, -1.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -2.62, 0.0, 0.0, 0.0, 0.0, 0.0, -1.0, -1.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -1.0, 0.0, 0.0, 0.0, 0.0, 0.0, -2.62, 0.0, 0.0, 0.0, 0.0, 0.0, -1.0, -1.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -2.62, 0.0, 0.0, 0.0, 0.0, 0.0, -1.0, -1.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -1.0, 0.0, 0.0, 0.0, 0.0, 0.0, -2.62, 0.0, 0.0, 0.0, 0.0, 0.0, -1.0, -1.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -2.62, -1.0, 0.0, 0.0, 0.0, 0.0, -1.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -1.0, 0.0, 0.0, 0.0, 0.0, -1.0, -2.62, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, -1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -1.0, -1.0, 0.0, 0.0, 0.0, 0.0, 0.0, -2.62, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -1.0, -1.0, 0.0, 0.0, 0.0, 0.0, 0.0, -2.62, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, -1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -1.0, -1.0, 0.0, 0.0, 0.0, 0.0, 0.0, -2.62, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -1.0, -1.0, 0.0, 0.0, 0.0, 0.0, 0.0, -2.62, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, -1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -1.0, -1.0, 0.0, 0.0, 0.0, 0.0, 0.0, -2.62]]) + 0.0j
B2 = np.array([[1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0]]) + 0.0j

Julia code

A2 = ComplexF64[0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0; 0.0 0.0 0.0 0.0 0.0 0.0 -1.0 0.0 0.0 0.0 0.0 0.0 -2.62 0.0 0.0 0.0 0.0 0.0 -1.0 -1.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -2.62 0.0 0.0 0.0 0.0 0.0 -1.0 -1.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -1.0 0.0 0.0 0.0 0.0 0.0 -2.62 0.0 0.0 0.0 0.0 0.0 -1.0 -1.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -2.62 0.0 0.0 0.0 0.0 0.0 -1.0 -1.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -1.0 0.0 0.0 0.0 0.0 0.0 -2.62 0.0 0.0 0.0 0.0 0.0 -1.0 -1.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -2.62 -1.0 0.0 0.0 0.0 0.0 -1.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -1.0 0.0 0.0 0.0 0.0 -1.0 -2.62 0.0 0.0 0.0 0.0 0.0; 0.0 -1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -1.0 -1.0 0.0 0.0 0.0 0.0 0.0 -2.62 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -1.0 -1.0 0.0 0.0 0.0 0.0 0.0 -2.62 0.0 0.0 0.0; 0.0 0.0 0.0 -1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -1.0 -1.0 0.0 0.0 0.0 0.0 0.0 -2.62 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -1.0 -1.0 0.0 0.0 0.0 0.0 0.0 -2.62 0.0; 0.0 0.0 0.0 0.0 0.0 -1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -1.0 -1.0 0.0 0.0 0.0 0.0 0.0 -2.62]
B2 = ComplexF64[1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0]

Example 3

Python code

A3 = np.array([[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0], [ 0.33748484079831426, -0.10323794456968927, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -2.5940303184033713, -0.20647588913937853, 0.0, 0.0, 0.0, 0.0, 0.0, -1.0], [ -0.10323794456968927, 0.3374848407983142, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -0.20647588913937853, -2.5940303184033713, -1.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.33748484079831426, 0.0, 0.0, -0.10323794456968927, 0.0, 0.0, 0.0, -1.0, -2.5940303184033713, 0.0, 0.0, -0.20647588913937853, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.3374848407983142, -0.10323794456968927, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -2.5940303184033713, -0.20647588913937853, 0.0, -1.0, 0.0], [ 0.0, 0.0, 0.0, -0.10323794456968927, 0.33748484079831426, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -0.20647588913937853, -2.5940303184033713, -1.0, 0.0, 0.0], [ 0.0, 0.0, -0.10323794456968927, 0.0, 0.0, 0.3374848407983142, 0.0, 0.0, 0.0, 0.0, -0.20647588913937853, 0.0, -1.0, -2.5940303184033713, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.33748484079831426, -0.10323794456968927, 0.0, 0.0, 0.0, -1.0, 0.0, 0.0, -2.5940303184033713, -0.20647588913937853], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -0.10323794456968927, 0.3374848407983142, -1.0, 0.0, 0.0, 0.0, 0.0, 0.0, -0.20647588913937853, -2.5940303184033713]]) + 0.0j
B3 = np.array([[1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -0.33748484079831426, 0.10323794456968927, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.10323794456968927, -0.3374848407983142, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -0.33748484079831426, 0.0, 0.0, 0.10323794456968927, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -0.3374848407983142, 0.10323794456968927, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.10323794456968927, -0.33748484079831426, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.10323794456968927, 0.0, 0.0, -0.3374848407983142, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -0.33748484079831426, 0.10323794456968927], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.10323794456968927, -0.3374848407983142]]) + 0.0j

Julia code

A3 = ComplexF64[0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0; 0.33748484079831426 -0.10323794456968927 0.0 0.0 0.0 0.0 0.0 0.0 -2.5940303184033713 -0.20647588913937853 0.0 0.0 0.0 0.0 0.0 -1.0; -0.10323794456968927 0.3374848407983142 0.0 0.0 0.0 0.0 0.0 0.0 -0.20647588913937853 -2.5940303184033713 -1.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.33748484079831426 0.0 0.0 -0.10323794456968927 0.0 0.0 0.0 -1.0 -2.5940303184033713 0.0 0.0 -0.20647588913937853 0.0 0.0; 0.0 0.0 0.0 0.3374848407983142 -0.10323794456968927 0.0 0.0 0.0 0.0 0.0 0.0 -2.5940303184033713 -0.20647588913937853 0.0 -1.0 0.0; 0.0 0.0 0.0 -0.10323794456968927 0.33748484079831426 0.0 0.0 0.0 0.0 0.0 0.0 -0.20647588913937853 -2.5940303184033713 -1.0 0.0 0.0; 0.0 0.0 -0.10323794456968927 0.0 0.0 0.3374848407983142 0.0 0.0 0.0 0.0 -0.20647588913937853 0.0 -1.0 -2.5940303184033713 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.33748484079831426 -0.10323794456968927 0.0 0.0 0.0 -1.0 0.0 0.0 -2.5940303184033713 -0.20647588913937853; 0.0 0.0 0.0 0.0 0.0 0.0 -0.10323794456968927 0.3374848407983142 -1.0 0.0 0.0 0.0 0.0 0.0 -0.20647588913937853 -2.5940303184033713]
B3 = ComplexF64[1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -0.33748484079831426 0.10323794456968927 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.10323794456968927 -0.3374848407983142 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -0.33748484079831426 0.0 0.0 0.10323794456968927 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -0.3374848407983142 0.10323794456968927 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.10323794456968927 -0.33748484079831426 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.10323794456968927 0.0 0.0 -0.3374848407983142 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -0.33748484079831426 0.10323794456968927; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.10323794456968927 -0.3374848407983142]

[martin-frbg]

Can we be certain that all these systems use the reference implementation of BLAS&LAPACK, rather than OpenBLAS or e.g. Apple's Accelerate ?

[pablosanjose]

Since the errors were triggered from various languages and machines, I assume so, but my next step is to try to trigger them calling the reference libraries directly.

EDIT: to be precise, Julia uses OpenBLAS, not sure about Python. What I mean is that I suspect the error is not caused by OpenBLAS version, but would like to confirm that suspicion.

[martin-frbg]

I would certainly be pleased if the problem was not related to OpenBLAS, but numpy is often/usually distributed with OpenBLAS under the hood nowadays, and there could be e.g. error accumulation from use of FMA instructions in its optimized BLAS kernels.

[pablosanjose]

Right, I believe I confirmed the issue using the reference LAPACK library. I obtained and compiled it using homebrew on macos (brew install lapack). I then called the zgeev function in the lapack dylib from Julia, passing first the matrices from example 2, which errored, and then with example 3, which worked.

[pablosanjose]

For reference, I used this code to wrap the call to LAPACK in a Julia function myggev!, and then did myggev!('N', 'V', A, B) with the relevant matrices. Code directly adapted from Julia's stdlib.

[thijssteel]

I did some testing with this and got the following results:

• all tests on my personal machine work
• tests on a xeon machine fail when linked with distribution openblas (only example 1)
• tests on that same xeon machine work when either linked with reference LAPACK or OpenBLAS compiled from source.

without a machine that can reproduce it with a version i can compile myself, i can't really debug. @pablosanjose would you mind running some tests for me?

This gist contains a fortran test file and an edited version of zhgeqz that prints out some logs. If you could clone LAPACK, replace zhgeqz with the one in the gist and run the test file i might be able to identify the problem.

P.S. the line length is a bit long, so you'll have to compile the test file with something like this make lib blaslib gfortran -ffree-line-length-none -ggdb3 -fcheck=bounds -o test test.f90 librefblas.a liblapack.a -lblas -llapack

[martin-frbg]

@thijssteel which version is the distribution openblas in your case ? (Though there is a weird issue with DYNAMIC_ARCH builds made on Sandybridge hardware not running correctly on SkylakeX) If this cannot be reproduced with pure Reference-LAPACK/BLAS I guess we should transfer this issue to the OpenBLAS tracker.

[thijssteel]

OpenBLAS installed with apt.

libopenblas-base/bionic,now 0.2.20+ds-4 amd64 [installed,automatic]
libopenblas-dev/bionic,now 0.2.20+ds-4 amd64 [installed]

I think this is just a very very specific issue with symmetry or underflow that disappears with the slightest change in rounding errors so it should still be treated here. Also, pablos was able to reproduce it using reference LAPACK.

[martin-frbg]

Oops. 0.2.20 is about three and a half years old, has LAPACK 3.7.0, has no specific support for AVX512 xeons, has several known problems with thread safety, register reuse in the assembly kernels etc that have since been fixed (or at least replaced with new ones)

[pablosanjose]

@thijssteel thanks for looking into this. I cloned, compiled and run the test. The results are in this gist.

How should we interpret this?

EDIT: This is in my macbook pro under macos, by the way, what I called MA in the OP.

Let me just mention that the matrices I posted are not fine-tuned or rare in any sense. Similar examples arise all the time for me in my application when I sweep parameters of my model, so the issue is a very real problem for me, not academic. It is crucial of course to ascertain whether this is OpenBLAS-specific or not.

[pablosanjose]

I see the gist is cut off. Let's see if I can link it here directly: result.txt.zip

[thijssteel]

I've spotted the problem.

To solve convergence failures (most often due to symmetries), LAPACK usually selects some random shift to permute the spectrum after a few failed iterations. (see line 370 of the gist). Since that shift is almost zero, it will preserve the symmetry. Eventually, by brute force, we get convergence, but it takes too many iterations and we eventually fail to solve the pencil within the iteration limit. A quick fix is to add more exceptional shift strategies (the QR algorithm has multiple ones for example).

I've also confirmed that this also happens when compiling OpenBLAS 0.2.20 from source. A quick fix seems to solve the issue for me. I've updated the gist with the fix and will make a PR when i've done some more testing.

[thijssteel]

thank you for reporting the problem in such detail and running the tests for me

[pablosanjose]

Fantastic, thank you!! Looking forward to the fix

[pablosanjose]

After much struggle, and much help from the Julia gurus, I managed to compile the OpenBLAS library with the Julia patches plus the patch in #477, and voilá, the example matrices 1, 2 and 3 no longer throw errors!! But unfortunately I soon ran into other matrices for which the fix didn't work. I post here another example that fails on my Macbook Pro (machine MA in the OP), even after #477

A4 = ComplexF64[0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0; 1.7391668762048442 -1.309613611600033 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 2.150333752409688 -2.619227223200066 0.0 -1.0 0.0 0.0 0.0 0.0 0.0 0.0; -1.3096136116000332 1.739166876204844 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -2.6192272232000664 2.150333752409688 -1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 1.739166876204844 0.0 0.0 -1.3096136116000332 0.0 0.0 0.0 0.0 0.0 -1.0 2.150333752409688 0.0 0.0 -2.6192272232000664 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 1.739166876204844 0.0 0.0 0.0 0.0 -1.3096136116000332 0.0 -1.0 0.0 0.0 2.150333752409688 0.0 0.0 0.0 0.0 -2.6192272232000664 0.0; 0.0 0.0 0.0 0.0 1.7391668762048442 0.0 0.0 0.0 0.0 -1.309613611600033 0.0 0.0 0.0 0.0 2.150333752409688 -1.0 0.0 0.0 0.0 -2.619227223200066; 0.0 0.0 -1.309613611600033 0.0 0.0 1.7391668762048442 0.0 0.0 0.0 0.0 0.0 0.0 -2.619227223200066 0.0 -1.0 2.150333752409688 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 1.739166876204844 -1.3096136116000332 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 2.150333752409688 -2.6192272232000664 0.0 -1.0; 0.0 0.0 0.0 0.0 0.0 0.0 -1.309613611600033 1.7391668762048442 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -2.619227223200066 2.150333752409688 -1.0 0.0; 0.0 0.0 0.0 -1.309613611600033 0.0 0.0 0.0 0.0 1.7391668762048442 0.0 0.0 0.0 0.0 -2.619227223200066 0.0 0.0 0.0 -1.0 2.150333752409688 0.0; 0.0 0.0 0.0 0.0 -1.3096136116000332 0.0 0.0 0.0 0.0 1.739166876204844 0.0 0.0 0.0 0.0 -2.6192272232000664 0.0 -1.0 0.0 0.0 2.150333752409688]
B4 = ComplexF64[1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -1.7391668762048442 1.3096136116000332 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.309613611600033 -1.739166876204844 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -1.739166876204844 0.0 0.0 1.309613611600033 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -1.739166876204844 0.0 0.0 0.0 0.0 1.309613611600033 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -1.7391668762048442 0.0 0.0 0.0 0.0 1.3096136116000332; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.3096136116000332 0.0 0.0 -1.7391668762048442 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -1.739166876204844 1.309613611600033 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.3096136116000332 -1.7391668762048442 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.3096136116000332 0.0 0.0 0.0 0.0 -1.7391668762048442 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.309613611600033 0.0 0.0 0.0 0.0 -1.739166876204844]

A5 = ComplexF64[0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0; 6.490384615384624 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.009615384615394 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 11.90076923076925 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -12.019230769230788 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -1.0000000000000007; 0.0 6.490384615384624 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.009615384615394 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 11.90076923076925 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -1.0000000000000007 -12.019230769230788 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 6.490384615384624 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.009615384615394 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 11.90076923076925 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -1.0000000000000007 -12.019230769230788 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 6.490384615384624 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.009615384615392 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 11.900769230769246 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -1.0000000000000007 -12.019230769230784 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 6.490384615384624 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.009615384615394 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 11.90076923076925 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -1.0000000000000007 -12.019230769230788 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 6.490384615384624 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.009615384615394 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 11.90076923076925 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -1.0000000000000007 -12.019230769230788 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 6.490384615384624 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.009615384615394 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 11.90076923076925 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -1.0000000000000007 -12.019230769230788 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.490384615384624 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.009615384615392 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 11.900769230769246 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -1.0000000000000007 -12.019230769230784 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.490384615384624 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.009615384615394 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 11.90076923076925 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -1.0000000000000007 -12.019230769230788 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.490384615384624 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.009615384615394 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 11.90076923076925 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -1.0000000000000007 -12.019230769230788; -6.009615384615392 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.490384615384624 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -12.019230769230784 -1.0000000000000007 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 11.900769230769248 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 -6.009615384615392 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.490384615384624 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -12.019230769230784 -1.0000000000000007 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 11.900769230769248 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 -6.009615384615392 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.490384615384624 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -12.019230769230784 -1.0000000000000007 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 11.900769230769248 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 -6.009615384615393 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.490384615384622 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -12.019230769230784 -1.0000000000000007 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 11.900769230769244 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 -6.009615384615392 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.490384615384624 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -12.019230769230784 -1.0000000000000007 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 11.900769230769248 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 -6.009615384615392 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.490384615384624 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -12.019230769230784 -1.0000000000000007 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 11.900769230769248 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 -6.009615384615392 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.490384615384624 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -12.019230769230784 -1.0000000000000007 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 11.900769230769248 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.009615384615393 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.490384615384622 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -12.019230769230784 -1.0000000000000007 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 11.900769230769244 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.009615384615392 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.490384615384624 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -12.019230769230784 -1.0000000000000007 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 11.900769230769248 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.009615384615392 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.490384615384624 -1.0000000000000007 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -12.019230769230784 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 11.900769230769248];
B5 = ComplexF64[1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.490384615384624 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.009615384615392 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.490384615384624 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.009615384615392 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.490384615384624 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.009615384615392 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.490384615384624 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.009615384615393 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.490384615384624 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.009615384615392 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.490384615384624 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.009615384615392 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.490384615384624 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.009615384615392 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.490384615384624 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.009615384615393 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.490384615384624 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.009615384615392 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.490384615384624 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.009615384615392; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.009615384615394 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.490384615384624 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.009615384615394 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.490384615384624 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.009615384615394 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.490384615384624 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.009615384615392 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.490384615384622 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.009615384615394 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.490384615384624 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.009615384615394 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.490384615384624 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.009615384615394 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.490384615384624 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.009615384615392 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.490384615384622 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.009615384615394 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.490384615384624 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 6.009615384615394 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -6.490384615384624];

Same kind of error as before.

EDIT: I included a second example

[pablosanjose]

I've adapted your Fortran test.f90 program to these new examples. I post here the results

result2.txt.zip

test2.f90.zip

EDIT: ah, sorry, that is without #477. Let me see if I can patch it in and run again. EDIT2: I'm not sure I transferred the Julia matrices correctly to Fortran, because I cannot reproduce the problem now with test2.f90. I will report if I make progress.

[martin-frbg]

I get

 RESULTS OF PENCIL            1  :            8
 RESULTS OF PENCIL            2  :           34

with the current develop branch of OpenBLAS that has #477 included

[pablosanjose]

My bad, you're right. I had not compiled the #477 patch when using test2. It still fails in Julia, though, with the (hopefully correctly patched) OpenBLAS. I need to investigate it further, I might have converted the matrices to Fortran wrong. Just a check, @martin-frbg, did you use test2.f90 or did you actually copy the A4,B4, A5, B5 matrices from the Julia code above?

[martin-frbg]

Oh, I just used your test2.f90. Not sure if I can set up Julia here in time.

[thijssteel]

I think it needs to be said: aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaarg

julian shared some history on this problem in the PR. I'll check if any of those solve the problem.

[thijssteel]

side note, we should add these pencils to the unit tests

[pablosanjose]

Ok. Just let's try to get to the bottom of this before merging. I am not 100% sure I compiled OpenBLAS with #477 correctly, so I'm not sure I'm sounding the alarm too soon or not.

To be clear, what I did:

• I took the #477 diff and adapted it into an OpenBLAS patch
• I compiled OpenBLAS 0.3.13 with the Julia patches plus the #477 patch using Julia's BinaryBuilder. That produced a dylib that I can call from Julia
• I confirmed that examples 1, and 2 fail under Julia 1.7 (which comes bundled with OpenBLAS 0.3.10) in my macbook (MA), but that they did not fail calling the zggev in the patched 0.3.13 dylib from Julia
• I ran my package, patched to call the 0.3.13+#477 dylib until it tripped once more with matrices 4 and 5 above.

There are a number of tricky potential pitfalls here. It could be that I did not correctly apply the #477 patch onto 0.3.13 (first time I try something remotely like this, and the BinaryBuilder process is rather opaque to me), and that 0.3.13 (not yet bundled with Jullia) without the #477 patch does not fail on 1, 2 and 3, but it does on 4 and 5 (remote possibility). And that 0.3.13+#477, correctly applied, does not fail at all (need to reproduce using test2.f90, really). It could also be that the failures are system dependent, and that they don't fail in @martin-frbg's machine. Anyway, let's keep digging for a bit until we clarify.

@thijssteel, would it be possible that you produce a test2.f90 yourself from the examples 4 and 5 above, just like you did for 1,2 and 3? I fear I transferred the matrices incorrectly.

[martin-frbg]

477 should apply cleanly to OpenBLAS (in lapack-netlib). And I (think I) have just reproduced the problem with the python form of testcase 4 in what would be your configuration "MB2" (Provided the matrix ordering is the same for numpy and julia, which was my impression from the early testcases. I basically just peppered case 4 with lots of commas and turned all semicolons into pairs of brackets)

[pablosanjose]

Yes, that's exactly what I did for 1, 2 and 3. So you confirmed failure 4 in Linux+Python2+#477, yes? Does 4 not fail under Python3 + #477, or you didn't try? What about example 5?

[martin-frbg]

Linux, Python3, #477 was what I tried. Have not tried example 5 yet as I got bored of typing commas (and would not have an elegant solution for the problem anyway).

[thijssteel]

You might find the following octave function useful for that (just remove one comma at the end)

function [] = printasfortran (A,name)

[m,n] = size(A);

fprintf( '%s = reshape((/ ', name )
for j=1:n
  for i=1:m
    fprintf( "( %e, %e ),", A(i,j), 0.0 )
  endfor
endfor
fprintf( ' /), shape(%s))\n', name )

endfunction

[pablosanjose]

The python version of 4 and 5:

A4 = np.array([[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0], [ 1.7391668762048442, -1.309613611600033, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 2.150333752409688, -2.619227223200066, 0.0, -1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ -1.3096136116000332, 1.739166876204844, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -2.6192272232000664, 2.150333752409688, -1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 1.739166876204844, 0.0, 0.0, -1.3096136116000332, 0.0, 0.0, 0.0, 0.0, 0.0, -1.0, 2.150333752409688, 0.0, 0.0, -2.6192272232000664, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 1.739166876204844, 0.0, 0.0, 0.0, 0.0, -1.3096136116000332, 0.0, -1.0, 0.0, 0.0, 2.150333752409688, 0.0, 0.0, 0.0, 0.0, -2.6192272232000664, 0.0], [ 0.0, 0.0, 0.0, 0.0, 1.7391668762048442, 0.0, 0.0, 0.0, 0.0, -1.309613611600033, 0.0, 0.0, 0.0, 0.0, 2.150333752409688, -1.0, 0.0, 0.0, 0.0, -2.619227223200066], [ 0.0, 0.0, -1.309613611600033, 0.0, 0.0, 1.7391668762048442, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -2.619227223200066, 0.0, -1.0, 2.150333752409688, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.739166876204844, -1.3096136116000332, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 2.150333752409688, -2.6192272232000664, 0.0, -1.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -1.309613611600033, 1.7391668762048442, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -2.619227223200066, 2.150333752409688, -1.0, 0.0], [ 0.0, 0.0, 0.0, -1.309613611600033, 0.0, 0.0, 0.0, 0.0, 1.7391668762048442, 0.0, 0.0, 0.0, 0.0, -2.619227223200066, 0.0, 0.0, 0.0, -1.0, 2.150333752409688, 0.0], [ 0.0, 0.0, 0.0, 0.0, -1.3096136116000332, 0.0, 0.0, 0.0, 0.0, 1.739166876204844, 0.0, 0.0, 0.0, 0.0, -2.6192272232000664, 0.0, -1.0, 0.0, 0.0, 2.150333752409688]]) + 0.0j

B4 = np.array([[1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -1.7391668762048442, 1.3096136116000332, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.309613611600033, -1.739166876204844, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -1.739166876204844, 0.0, 0.0, 1.309613611600033, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -1.739166876204844, 0.0, 0.0, 0.0, 0.0, 1.309613611600033, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -1.7391668762048442, 0.0, 0.0, 0.0, 0.0, 1.3096136116000332], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.3096136116000332, 0.0, 0.0, -1.7391668762048442, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -1.739166876204844, 1.309613611600033, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.3096136116000332, -1.7391668762048442, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.3096136116000332, 0.0, 0.0, 0.0, 0.0, -1.7391668762048442, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.309613611600033, 0.0, 0.0, 0.0, 0.0, -1.739166876204844]]) + 0.0j

A5 = np.array([[0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0], [ 6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.009615384615394, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 11.90076923076925, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -12.019230769230788, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -1.0000000000000007], [ 0.0, 6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.009615384615394, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 11.90076923076925, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -1.0000000000000007, -12.019230769230788, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.009615384615394, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 11.90076923076925, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -1.0000000000000007, -12.019230769230788, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.009615384615392, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 11.900769230769246, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -1.0000000000000007, -12.019230769230784, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.009615384615394, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 11.90076923076925, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -1.0000000000000007, -12.019230769230788, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.009615384615394, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 11.90076923076925, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -1.0000000000000007, -12.019230769230788, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.009615384615394, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 11.90076923076925, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -1.0000000000000007, -12.019230769230788, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.009615384615392, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 11.900769230769246, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -1.0000000000000007, -12.019230769230784, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.009615384615394, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 11.90076923076925, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -1.0000000000000007, -12.019230769230788, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.009615384615394, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 11.90076923076925, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -1.0000000000000007, -12.019230769230788], [ -6.009615384615392, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -12.019230769230784, -1.0000000000000007, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 11.900769230769248, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, -6.009615384615392, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -12.019230769230784, -1.0000000000000007, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 11.900769230769248, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, -6.009615384615392, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -12.019230769230784, -1.0000000000000007, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 11.900769230769248, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, -6.009615384615393, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.490384615384622, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -12.019230769230784, -1.0000000000000007, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 11.900769230769244, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, -6.009615384615392, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -12.019230769230784, -1.0000000000000007, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 11.900769230769248, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, -6.009615384615392, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -12.019230769230784, -1.0000000000000007, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 11.900769230769248, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.009615384615392, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -12.019230769230784, -1.0000000000000007, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 11.900769230769248, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.009615384615393, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.490384615384622, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -12.019230769230784, -1.0000000000000007, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 11.900769230769244, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.009615384615392, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -12.019230769230784, -1.0000000000000007, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 11.900769230769248, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.009615384615392, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.490384615384624, -1.0000000000000007, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -12.019230769230784, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 11.900769230769248]]) + 0.0j

B5 = np.array([[1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.009615384615392, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.009615384615392, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.009615384615392, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.009615384615393, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.009615384615392, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.009615384615392, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.009615384615392, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.009615384615393, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.009615384615392, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.009615384615392], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.009615384615394, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.009615384615394, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.009615384615394, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.009615384615392, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.490384615384622, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.009615384615394, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.490384615384624, 0.0, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.009615384615394, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.490384615384624, 0.0, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.009615384615394, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.490384615384624, 0.0, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.009615384615392, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.490384615384622, 0.0, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.009615384615394, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.490384615384624, 0.0], [ 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 6.009615384615394, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, -6.490384615384624]]) + 0.0j

[langou]

For reference, I had moved some of the history at #102

[langou]

@thijssteel : how does #421 fare with these pencils? Do you know?

[thijssteel]

DLAQZ0 will defer calculation to DHGEQZ for these pencils ( n < 75)

It is to be excpected that is will fare at least slightly better in a way, because AED can solve a lot of these issues. I also know it will at least be better than most other implementations because of some protection for overflow in the double shift calculation.

On the other hand, i know that DLAQR0 does a lot more to solve convergence issues. It varies AED window sizes, has more advanced exceptional shifts, ... if DLAQZ0 has convergence issues i could add these techniques

[pablosanjose]

I used @thijssteel 's octave function to write the matrices correctly. This is the result I get under macos with #477 patch.

I think they show convergence problems with example 4. Am I reading this correctly?

test2.f90.zip result2.txt.zip

[pablosanjose]

@thijssteel : how does #421 fare with these pencils? Do you know?

Fancy. Is #421 related to https://arxiv.org/abs/2007.03576 ?

[thijssteel]

No, to https://arxiv.org/pdf/1902.10954.pdf (but that paper is more about a theoretical extension of QZ, not about that implementation specifically). I'm quite sure mirko's code beats mine.

[pablosanjose]

This is probably obvious to you, but one common thing I see with all these matrices is that they have a lot of singular values that are equal up to machine precision. I guess your comment about symmetries is precisely this.

[pablosanjose]

Since small pencils are more desirable for unit tests I post here a 12x12 example that fails after #477 for me

A6 = ComplexF64[0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.0; 3.507883011020636 4.043337864077769 0.0 0.0 0.0 0.0 7.883337105374606 8.086675728155537 0.0 -1.0 0.0 0.0; 4.043337864077769 3.5078830110206356 0.0 0.0 0.0 0.0 8.086675728155537 7.883337105374604 -1.0 0.0 0.0 0.0; 0.0 0.0 3.5078830110206356 0.0 0.0 4.043337864077769 0.0 -1.0 7.883337105374604 0.0 0.0 8.086675728155537; 0.0 0.0 0.0 3.5078830110206356 4.043337864077769 0.0 -1.0 0.0 0.0 7.883337105374604 8.086675728155537 0.0; 0.0 0.0 0.0 4.043337864077769 3.507883011020636 0.0 0.0 0.0 0.0 8.086675728155537 7.883337105374606 -1.0; 0.0 0.0 4.043337864077769 0.0 0.0 3.507883011020636 0.0 0.0 8.086675728155537 0.0 -1.0 7.883337105374606];
B6 = ComplexF64[1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 1.0 0.0 0.0 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 -3.507883011020636 -4.043337864077769 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 -4.043337864077769 -3.5078830110206356 0.0 0.0 0.0 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -3.5078830110206356 0.0 0.0 -4.043337864077769; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -3.5078830110206356 -4.043337864077769 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -4.043337864077769 -3.507883011020636 0.0; 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 -4.043337864077769 0.0 0.0 -3.507883011020636];

[thijssteel]

This is probably obvious to you, but one common thing I see with all these matrices is that they have a lot of singular values that are equal up to machine precision. I guess your comment about symmetries is precisely this.

Hmmm, the link between singular values and eigenvalues is nontrivial, at least to me.

Symmetry is more about 2 eigenvalues being equally close to the shift. Then they will converge equally fast. The shift doesn't converge to a single eigenvalue and we eventually never get convergence. This is quite common in real matrices because the eigenvalues are symmetric wrt the real axis.

Other common causes are under/overflow in the calculation of the shift column or rare self destroying vigilant deflations.

The problem we're faced with now is that to disturb the symmetry and lack of convergence, we want to just do a somewhat random shift. But just choosing something random doesn't always work, because it needs to be significantly closer to one eigenvalue than the other. So we base it on the lower entries somehow, but these can in rare cases be zero, or be some linear combination of the trace/determinant of the bottom subpencil which again has special properties that can hamper its effect for some pencils...

So yeah, while i would like to act that i'm going to invent some genious strategy with a lot of derivations. I'm just gonna try a lot of random stuff and see what works. Possibly going to dust off my favourite number: 1302. Thx for all the pencils, gives me a lot of tests to work with.

[pablosanjose]

Thanks for the info. If you come up with more potential fixes I'll try them on my code. Although not the most constructive comment, I note that I haven't been able to make Intel's MKL fail a single time (I've tried!), so there must exist some ideal strategy to solve this that they use.

[langou]

Here is a piece of code that was provided to us by MathWorks a while back. We had problem with this code in our test suite so we never integrated it. My understand is that this code made its way in MKL and in MathWorks LAPACK codes. If someone can try to play with this and see if this helps, this would be terrific.

See: http://math.ucdenver.edu/~langou/zhgeqz--mathworks.f And a diff between MathWorks code (on the left) and LAPACK's current (on the right) is at: http://math.ucdenver.edu/~langou/Diff.html

I am trying to confirm with MathWorks whether the code I am providing above is their correct one. I am not 100% sure. But if someone can try the code and see if this helps on these pencils, this would be helpful.

[thijssteel]

Ok, so yeah. How about we forget all the fancy mathematics i was explaining....

Lines 722-732 don't correctly calculate $AB^{-1}$. In more detail, AB12 is never calculated, instead the code uses AD12.

After fixing that (and for good measure, upgrading the shift calculation slightly), the exceptional shift strategy no longer matters for pablo's examples because we get regular convergence within the iteration limit.

I've updated the gist if pablo wants to confirm, but it seems unlikely that machine dependency is still an issue.

[thijssteel]

This is also likely why MKL and mathwork perform so well

[pablosanjose]

I'm happy to report that with the new fix I can no longer find non-convergent pencils with my code. Well done @thijssteel !! Thank you everybody

EDIT: to be clear, I tested this commit from #477. The following one affects only chgeqz.f, but I'm calling the complex double version, so I guess it doesn't affect my tests.

[pablosanjose]

Not sure what you mean. The original issue has been fixed, so no, you don't need to use other libraries for this now. In any case MKL routines can be used in Julia with MKL.jl and MKLSparse.jl.

[langou]

Yes, MKL (for example) is often faster than reference LAPACK, which means that in turn packages relying on MKL (Mathematica, Matlab, julia, etc.) are faster using MKL than reference LAPACK. The purpose of this GIT repository is to maintain reference LAPACK. The oneMKL team is monitoring our work, commits, etc. and adapting our work to MKL. The oneMKL team is also contributing at times to reference LAPACK with PRs and suggestions and bug reports. My understanding is that some adaption of #477 (for example) will make it to oneMKL soon.