← New search

Other meanings of Eigenvalue problem

LINEAR ALGEBRA

Eigenvalue problem

An eigenvalue problem asks for the scalars and nonzero vectors that satisfy Av = λv, or, more generally, T v = λv for a linear operator. The scalar λ is an eigenvalue and v is its eigenvector; together they describe directions or modes whose essential form is preserved by the transformation.

Av = λv
Core equation
Matrix or operator form
det(A − λI) = 0
Finite-dimensional test
Characteristic equation
n eigenvalues
Typical count
An n × n matrix over the complex numbers, counted with multiplicity
1

Definition and formulation

The central task is to find nonzero vectors that a linear transformation changes only by a scalar factor. For a matrix A, the equation Av = λv is equivalent to (A − λI)v = 0; a nonzero solution exists precisely when A − λI is singular, giving the characteristic equation det(A − λI) = 0.1

The resulting roots λ are eigenvalues, while the corresponding null spaces are eigenspaces. Over the complex numbers, every n × n matrix has n eigenvalues when algebraic multiplicity is counted, although some may coincide. Real matrices can have nonreal conjugate pairs, and a repeated eigenvalue can have fewer independent eigenvectors than its algebraic multiplicity, producing a defective matrix.2

2

Structure and solution methods

The method of solution depends strongly on the matrix’s structure. Direct characteristic-polynomial calculations are useful for small matrices, but numerical software generally uses orthogonal or unitary transformations, especially the QR algorithm, because they are more stable than explicitly forming polynomial roots. LAPACK supplies standardized routines for ordinary and generalized eigenvalue problems, including symmetric, Hermitian, and nonsymmetric cases.3

Symmetric real matrices and Hermitian complex matrices have real eigenvalues and orthogonal eigenvectors; they can therefore be diagonalized by an orthogonal or unitary matrix. Normal matrices have an analogous unitary diagonalization, whereas defective matrices require a Jordan form in exact theory. The generalized problem Av = λBv is fundamental when B represents a mass, metric, or covariance matrix; if B is nonsingular, it can sometimes be related to an ordinary problem, though numerical algorithms usually avoid explicitly computing B−1.4

3

Applications and numerical practice

Eigenvalues identify characteristic scales, frequencies, growth rates, and stability thresholds across applied mathematics. In differential equations, eigenvalues of a discretized operator describe vibration modes and decay or amplification; in quantum mechanics, they represent possible measurement values of observables; and in dynamical systems, the locations of eigenvalues relative to the unit circle or imaginary axis help determine stability.5

Large problems rarely require every eigenpair. Power iteration is effective for a dominant eigenvalue under suitable separation conditions, while inverse iteration targets an eigenvalue near a chosen shift. Krylov-subspace methods such as Arnoldi and Lanczos are designed for sparse matrices and can compute a small, informative part of a spectrum. In practice, residuals such as ||Av − λv|| provide a direct check on the accuracy of a computed eigenpair, while conditioning determines how sensitive the answer is to perturbations in the data.

4

Lesser-known aspects

Eigenvalue problems extend beyond finite matrices to operators on infinite-dimensional spaces, where the spectrum need not consist only of isolated eigenvalues. Differential operators can have continuous spectrum, and boundary conditions may determine whether the spectrum is discrete, continuous, or mixed. Sturm–Liouville problems are a particularly important class: under suitable conditions their eigenvalues are real, their eigenfunctions are orthogonal under a weighted inner product, and those functions can represent other functions through series expansions.6

Several edge cases matter in interpretation. The zero eigenvalue means that the transformation has a nontrivial null space and is therefore singular. A matrix may have a complete set of eigenvectors without being symmetric, but nonnormal matrices can still exhibit strong transient amplification even when all eigenvalues suggest asymptotic decay. Singular-value problems are related but distinct: singular values measure stretching between possibly different input and output directions, whereas eigenvalues preserve a single direction up to scale.

Glossary

Algebraic multiplicity
The multiplicity of an eigenvalue as a root of the characteristic polynomial.
Eigenspace
The null space of A − λI associated with an eigenvalue λ.
Geometric multiplicity
The dimension of an eigenvalue’s eigenspace.
Generalized eigenvalue problem
A problem of the form Av = λBv, usually with two matrices A and B.
Spectrum
The set of scalars for which an operator fails to have a bounded inverse; in finite dimensions, it is the set of eigenvalues.

Eigenvalues and eigenvectors depend on the underlying field, inner product, domain, and boundary conditions; statements about reality, orthogonality, completeness, and diagonalization therefore require the relevant hypotheses.