What is eigenvalue decomposition?

Eigenvalue decomposition, also known as eigendecomposition, is a matrix factorization technique that expresses a square matrix as a product of three matrices:

  • A matrix of eigenvectors

  • A diagonal matrix of eigenvalues

  • The inverse of the matrix of eigenvectors

Mathematically, if AA is a matrix, then its eigenvalue decomposition is given by:

where VV, is the matrix of eigenvectors of AA and DD is a diagonal matrix of the eigenvalues of AA.

Example

Suppose we have the following matrix:

Here, the eigenvalues we got are 77 and 5-5. Hence,

Now, for eigenvectors:

For eigenvalue 77: ([1491][1001].(+7)).v1=0(\begin{bmatrix} 1 & 4\\ 9 & 1 \end{bmatrix} - \begin{bmatrix} 1 & 0\\ 0 & 1 \end{bmatrix}.(+7)).v_{1} = 0

([1491][7007]).v1=0(\begin{bmatrix} 1 & 4\\ 9 & 1 \end{bmatrix} - \begin{bmatrix} 7 & 0\\ 0 & 7 \end{bmatrix}).v_{1} = 0

([6496]).([x1y1])=0(\begin{bmatrix} -6 & 4\\ 9 & -6 \end{bmatrix}).(\begin{bmatrix} x_{1}\\ y_{1} \end{bmatrix}) = 0

6x1+4y1=09x16y1=0 -6x_{1}+4y_{1} = 0 \\ 9x_{1}-6y_{1} = 0 \\

This system of linear equation can have infinitely many solutions. One of the solutions is:

x1=2y1=3 x_{1} = 2\\ y_{1} = 3

Hence, v1=[23] v_{1} = \begin{bmatrix} 2\\ 3 \end{bmatrix}

Normalizing Eigenvectors: v1n=[213313] v_{1n} = \begin{bmatrix} \frac{2}{\sqrt{13}}\\ \frac{3}{\sqrt{13}} \end{bmatrix}

For eigenvalue 5-5: ([1491][1001].(5)).v2=0(\begin{bmatrix} 1 & 4\\ 9 & 1 \end{bmatrix} - \begin{bmatrix} 1 & 0\\ 0 & 1 \end{bmatrix}.(-5)).v_{2} = 0

([1491][5005]).v2=0(\begin{bmatrix} 1 & 4\\ 9 & 1 \end{bmatrix} - \begin{bmatrix} -5 & 0\\ 0 & -5 \end{bmatrix}).v_{2} = 0

([6496]).([x2y2])=0(\begin{bmatrix} 6 & 4\\ 9 & 6 \end{bmatrix}).(\begin{bmatrix} x_{2}\\ y_{2} \end{bmatrix}) = 0

6x2+4y2=09x2+6y2=0 6x_{2}+4y_{2} = 0 \\ 9x_{2}+6y_{2} = 0 \\

This system of linear equation can have infinitely many solutions. One of the solutions is:

x2=2y2=3 x_{2} = 2\\ y_{2} = -3

Hence, v2=[23] v_{2} = \begin{bmatrix} 2\\ -3 \end{bmatrix}

Normalizing Eigenvectors: v2n=[213313] v_{2n} = \begin{bmatrix} \frac{2}{\sqrt{13}}\\ \frac{-3}{\sqrt{13}} \end{bmatrix}

Hence,

Calculating the inverse of VV:

Combining our calculations:

Conclusion

Eigenvalue decomposition empowers us to comprehend complex matrices by breaking them into their constituent eigenvectors and eigenvalue. We use the eigenvalue decomposition to factorize a square matrix, but alternative matrix factorization methods, such as singular value decomposition (SVD), are used for non-diagonalizable or non-square matrices.

Free Resources

Copyright ©2024 Educative, Inc. All rights reserved