import numpy
import matplotlib.pyplot as plt

%matplotlib inline

85. Eigenvalue problem

Matrix transformaion is written as

\[\mathbf{y} = A\mathbf{x}\]

A different vector in the same direction can be written as scalar multiplication:

\[\mathbf{y} = \lambda\mathbf{x}\]

Equating these \(y\)s yields:

\[A\mathbf{x} = \lambda \mathbf{x} \Rightarrow (A - \lambda I) \mathbf{x} = 0\]

\[\det(A - \lambda I) = 0\]

The eigenvalue problem can also be collected with \(\Lambda\) being a diagonal matrix containing all the eigenvalues and \(X\) containing the eigenvectors stacked column-wise. This leads to the eigenvalue decomposition:

\[A X = X \Lambda \Rightarrow A = X \Lambda X^{-1}\]

with

\[\Lambda = diag(\lambda_i)\]

If we try to find a similar decomposition with different constraints, we can write

\[A = U D V^{H}\]

If \(D\) is a diagonal matrix and \(U\) and \(V\) are unitary, this is the singular value decomposition.

In Skogestad

\[A = U \Sigma V^{H}\]

\[\Sigma = diag(\sigma_i)\]

from ipywidgets import interact

def plotvector(x, color='blue'):
    plt.plot([0, x[0,0]], [0, x[1,0]], color=color)

import matplotlib.patches as patches

Let’s investigate the properties of this matrix:

A = numpy.matrix([[4, 3],
                  [2, 1]])

The eigenvectors and eigenvalues can be calculated as follows. We also calculate the output vectors associated with a unit vector input in the eigenvector directions.

A

matrix([[4, 3],
        [2, 1]])

v = numpy.asmatrix(numpy.random.random(2)).T

v = A*v

v = v/numpy.linalg.norm(v)
v

matrix([[0.89474813],
        [0.44657115]])

lambdas, eigvectors = numpy.linalg.eig(A)
ev1 = lambdas[0]*eigvectors[:, 0]
ev2 = lambdas[1]*eigvectors[:, 1]

The singular values determine the main axes of the translation ellipse of the matrix. Note that the numpy.linalg.svd function returns the conjugate transpose of the input direction matrix.

U, S, VH = numpy.linalg.svd(A)
V = VH.H
ellipseangle = numpy.rad2deg(numpy.angle(complex(*U[:, 0])))

def interactive(scale, theta):
    x = numpy.matrix([[numpy.cos(theta)], [numpy.sin(theta)]])
    y = A*x

    plotvector(x)
    plotvector(y, color='red')
    plotvector(ev1, 'green')
    plotvector(ev2, 'green')
    plotvector(V[:, 0], 'magenta')
    plotvector(V[:, 1], 'magenta')
    plt.gca().add_artist(patches.Circle([0, 0], 1,
                                        color='blue',
                                        alpha=0.1))
    plt.gca().add_artist(patches.Ellipse([0, 0], S[0]*2, S[1]*2,
                                         ellipseangle,
                                         color='red',
                                         alpha=0.1))
    plt.axis([-scale, scale, -scale, scale])
    plt.axes().set_aspect('equal')
    plt.show()
interact(interactive, scale=(1., 10), theta=(0., numpy.pi*2))

<function __main__.interactive>