Singular value decomposition using NumPy

The function svd() from the numpy.linalg module finds the singular values of a matrix and returns them in an ndarray contained in an SVDResult instance.

Linear Algebra Updated 9 Apr 2025 2 min read

Overview

  • The function svd() from the linalg module of NumPy library calculates the singular values of an input array A and returns the results contained in an SVDResult instance.
  • The first ndarray in the SVDResult instance contains the left singular vector.
  • The second ndarray in the SVDResult instance contains the singular values of the given array A.
  • The third ndarray in the SVDResult instance contains the right singular vector.

Singular Value Decomposition

  • Using singular value decomposition the matrix A of size mxn is factorized into
    • An orthogonal matrix U of dimension mxm.
    • A complex conjugate(also known as Hermitian transpose) of orthogonal matrix V of dimension nxn
    • A diagonal matrix Σ of dimension mxn, containing singular values on its diagonal

      as given by

A = UΣV*

  • Unlike eigenvalues which are defined only for square matrices, singular values can be found for both square matrices and rectangular matrices.
  • The singular values are non-negative real numbers.
  • V*- Complex conjugate of V.

Note:

Orthogonal matrix: An orthogonal matrix is a matrix when multiplied by its transpose results in an identity matrix.

Identity matrix: An identity matrix is a matrix whose diagonal elements are with the value one. An identity matrix is a diagonal matrix.

Diagonal matrix: A diagonal matrix is a matrix whose elements are all ones or zeroes on its diagonal.

Example

python
# Example Python program that obtains the singular values of

            # a square matrix

            import numpy

# Create a 2x2 square matrix

            a1 = numpy.ndarray(shape = (2, 2),

                               dtype = numpy.int32)

a1[0][0] = 2

            a1[0][1] = 5

            a1[1][0] = -7

            a1[1][1] = -3

# Obtain the singular values of the matrix

            svdResults = numpy.linalg.svd(a1)

            print(type(svdResults))

            print(svdResults)

print("Left singular vector:")

            print(numpy.linalg.svd(a1)[0])

print("Right singular vector:")

            print(numpy.linalg.svd(a1)[2])

print("Singular values of the matrix:")

            print(numpy.linalg.svd(a1)[1])

Output

Output
<class 'numpy.linalg._linalg.SVDResult'>

            SVDResult(U=array([[-0.52573111,  0.85065081],

                   [ 0.85065081,  0.52573111]]), S=array([8.71337969, 3.32821489]), Vh=array([[-0.80405286, -0.59455781],

                   [-0.59455781,  0.80405286]]))

            Left singular vector:

            [[-0.52573111  0.85065081]

             [ 0.85065081  0.52573111]]

            Right singular vector:

            [[-0.80405286 -0.59455781]

             [-0.59455781  0.80405286]]

            Singular values of the matrix:

            [8.71337969 3.32821489]

Applications of Singular Value Decomposition

  • Image processing
  • Text analysis