Singular Value Decomposition (SVD) Calculator

Welcome to the Singular Value Decomposition (SVD) Calculator, a powerful linear algebra tool that decomposes any matrix into its fundamental components. SVD factorizes a matrix A = UΣVᵀ and provides step-by-step solutions, interactive visualizations, rank analysis, condition number, low-rank approximation quality, and pseudoinverse computation. Whether you are studying linear algebra, working on machine learning, or analyzing data, this calculator provides professional-grade matrix decomposition.

What is Singular Value Decomposition?

Singular Value Decomposition (SVD) is the factorization of any m×n matrix A into three matrices:

Where:

• A is the original m×n matrix
• U is an m×m orthogonal matrix (left singular vectors, eigenvectors of AAᵀ)
• Σ (Sigma) is an m×n diagonal matrix with non-negative singular values σ₁ ≥ σ₂ ≥ ... ≥ 0
• Vᵀ is an n×n orthogonal matrix (right singular vectors, eigenvectors of AᵀA)

Unlike eigenvalue decomposition, SVD always exists for any matrix, including rectangular and singular matrices. This universality makes it one of the most important factorizations in applied mathematics.

How SVD is Computed

1. Form AᵀA: Compute the n×n symmetric matrix AᵀA
2. Find eigenvalues: Solve det(AᵀA − λI) = 0 to get eigenvalues λ₁ ≥ λ₂ ≥ ... ≥ 0
3. Singular values: σᵢ = √λᵢ (square roots of eigenvalues)
4. Right singular vectors (V): Find eigenvectors of AᵀA, orthonormalize to get columns of V
5. Left singular vectors (U): Compute uᵢ = Avᵢ/σᵢ for each nonzero singular value, extend to a full orthonormal basis

Key Properties

Matrix Rank

The rank of matrix A equals the number of nonzero singular values. This is the most numerically stable way to determine rank, far more reliable than row reduction which can be thrown off by floating-point errors.

Condition Number

The condition number measures how sensitive a linear system Ax = b is to perturbations. A large κ indicates an ill-conditioned matrix; κ = 1 is the ideal case (orthogonal matrices).

Matrix Norms via SVD

• Spectral norm (2-norm): \(\|A\|_2 = \sigma_1\) — the largest singular value
• Frobenius norm: \(\|A\|_F = \sqrt{\sigma_1^2 + \sigma_2^2 + \cdots}\)
• Nuclear norm: \(\|A\|_* = \sigma_1 + \sigma_2 + \cdots\) — sum of all singular values

Applications of SVD

Low-Rank Approximation (Eckart–Young Theorem)

The Eckart–Young–Mirsky theorem states that the best rank-k approximation of A (in Frobenius or spectral norm) is obtained by keeping only the k largest singular values:

The approximation error is: \(\|A - A_k\|_F = \sqrt{\sigma_{k+1}^2 + \cdots + \sigma_r^2}\)

SVD vs Eigenvalue Decomposition

Frequently Asked Questions

What is Singular Value Decomposition (SVD)?

Singular Value Decomposition (SVD) is a matrix factorization that decomposes any m×n real matrix A into three matrices: A = UΣVᵀ, where U is an m×m orthogonal matrix of left singular vectors, Σ is an m×n diagonal matrix of singular values, and Vᵀ is an n×n orthogonal matrix of right singular vectors. SVD always exists for any matrix.

What are singular values used for?

Singular values reveal fundamental properties of a matrix: the rank (number of nonzero singular values), the condition number (ratio of largest to smallest), and matrix norms. They are widely used in data compression (keeping only the largest singular values), principal component analysis (PCA), noise reduction, recommender systems, and solving least-squares problems.

What is the difference between SVD and eigenvalue decomposition?

Eigenvalue decomposition only works for square matrices and requires the matrix to be diagonalizable. SVD works for any m×n matrix (including rectangular ones) and always exists. For a symmetric positive semi-definite matrix, SVD and eigendecomposition coincide. SVD uses two different orthogonal bases (U and V), while eigendecomposition uses one.

How does SVD relate to PCA?

PCA (Principal Component Analysis) is directly computed using SVD. When you center the data matrix X and compute its SVD as X = UΣVᵀ, the columns of V are the principal components (directions of maximum variance), the singular values in Σ encode the standard deviations along each component, and UΣ gives the projected data in the new coordinate system. For N centered observations (N > 1), Var(PCᵢ) = σᵢ²/(N−1), SD(PCᵢ) = σᵢ/√(N−1).

What is a low-rank approximation?

A rank-k approximation of matrix A keeps only the k largest singular values and their corresponding vectors: A_k = U_k Σ_k V_k^T. By the Eckart-Young theorem, this is the best rank-k approximation in both the Frobenius and spectral norms. This is the mathematical foundation behind image compression, latent semantic analysis, and dimensionality reduction.

What is the condition number of a matrix?

The condition number κ(A) = σ_max / σ_min is the ratio of the largest to smallest singular value. It measures how sensitive the solution of a linear system Ax = b is to perturbations. A large condition number means the matrix is ill-conditioned and small errors in input can cause large errors in the solution. A condition number of 1 (orthogonal matrix) is ideal.

Additional Resources

• Low-Rank Approximation - Wikipedia
• Moore-Penrose Pseudoinverse - Wikipedia