Computing eigendecompositions

Python numerical eigendecompositions

In Python, we first need to load the appropriate packages.

import numpy as np # for numerics
from numpy import linalg as la # for eig
from IPython.display import display, Markdown, Latex # prty
np.set_printoptions(precision=3) # for pretty

Consider the same numerical A matrix from the section above. Create it as a numpy.array object.

A = np.array(
  [
    [-3, 5, 9],
    [0, 2, -10],
    [5, 0, -4],
  ]
)

The numpy.linalg module (loaded as la) gives us access to the eig function.

e_vals,e_vecs = la.eig(A)
print(f'e-vals: {e_vals}')
print(f'modal matrix:\n {e_vecs}')

e-vals: [-11.487+0.j      3.243+4.122j   3.243-4.122j]
modal matrix:
 [[-0.769+0.j     0.122-0.537j  0.122+0.537j]
 [ 0.381+0.j     0.767+0.j     0.767-0.j   ]
 [ 0.514+0.j    -0.095-0.316j -0.095+0.316j]]

Note that the eigenvalues are returned as a one-dimensional array, not along the diagonal of a matrix as with Matlab.

print(f"the third eigenvalue is {e_vals[2]:.3e}")

the third eigenvalue is 3.243e+00-4.122e+00j

Python symbolic eigendecompositions

We use the sympy package for symbolics.

import sympy as sp

Declare symbolic variables.

sp.var('a b c')

(a, b, c)

Define a symbolic matrix A.

A = sp.Matrix([
  [a,b],
  [0,c]
])
display(A)

\(\displaystyle \left[\begin{matrix}a & b\\0 & c\end{matrix}\right]\)

The sympy.Matrix class has methods eigenvals and eigenvects. Let’s consider them in turn.

A.eigenvals()

{a: 1, c: 1}

What is returned is a dictionary with our eigenvalues as its keys and the multiplicity (how many) of each eigenvalue as its corresponding value.

The eigenvects method returns even more complexly structured results.

A.eigenvects()

[(a, 1, [Matrix([
   [1],
   [0]])]), (c, 1, [Matrix([
   [-b/(a - c)],
   [         1]])])]

This is a list of tuples with structure as follows.

(<eigenvalue>,<multiplicity>,<eigenvector>)

Each eigenvector is given as a list of symbolic matrices.

Extracting the second eigenvector can be achieved as follows.

A.eigenvects()[1][2][0]

\(\displaystyle \left[\begin{matrix}- \frac{b}{a - c}\\1\end{matrix}\right]\)