Contents

EigenvaluesEdit

Eigenvalues are a foundational concept in linear algebra and the study of linear systems. They are scalars that reveal how a linear transformation acts along particular directions. When a square matrix A represents a transformation, there are nonzero vectors v, called eigenvectors, such that A v = λ v for some scalar λ. That scalar λ is the eigenvalue corresponding to the eigenvector v. In other words, along the line spanned by v, the transformation simply stretches (or shrinks, or rotates in the complex plane) by a factor λ.

The eigenvalue problem is intimately tied to the determinant and the characteristic polynomial. For a real or complex square matrix A, the eigenvalues are the roots of the characteristic equation det(A − λI) = 0, where I is the identity matrix. Even when A is real, eigenvalues can be complex numbers, and these complex eigenvalues come in conjugate pairs if A has real entries. When A is symmetric (that is, A = A^T) or Hermitian in the complex case, all eigenvalues are real, and the eigenvectors corresponding to distinct eigenvalues are orthogonal, giving a powerful spectral decomposition.

The concept is central not only as a theoretical object but as a practical tool. Eigenvalues inform the qualitative behavior of dynamical systems, the stability of equilibria, and the long-term evolution of processes described by linear models. They also underpin methods in data analysis, physics, and computer science, where the spectrum of a matrix encodes essential structure of the system being studied. For example, in statistics, the eigenvalues of a covariance matrix determine the directions of greatest variance in the data through principal components; in computer science, the eigenvalues of the Google matrix relate to ranking channels on the web; and in physics, energy levels in quantum systems arise as eigenvalues of the Hamiltonian operator. See linear algebra and spectral theorem for broader context, and note that many practical discussions hinge on how the spectrum influences stability and performance in applied settings.

Definition and basic concepts

  • An eigenvalue λ is defined by the existence of a nonzero vector v with A v = λ v, where A is a matrix and v is an eigenvector associated with λ.
  • The set of all eigenvalues of A is the spectrum of A, often denoted by σ(A). This spectrum may be real or complex depending on A.
  • The characteristic polynomial p(λ) = det(A − λI) has degree n for an n × n matrix A, and its roots are precisely the eigenvalues of A.
  • Multiplicity concepts: the algebraic multiplicity counts how many times an eigenvalue appears as a root of p(λ), while the geometric multiplicity is the dimension of the eigenspace corresponding to that eigenvalue.
  • Diagonalizability: A is diagonalizable if there exists an invertible matrix V whose columns are eigenvectors of A such that A = V Λ V^−1, where Λ is a diagonal matrix of eigenvalues. If A is real and symmetric, a stronger form holds: A = Q Λ Q^T with an orthogonal Q.
  • Special cases: if A is real and symmetric, all eigenvalues are real; if A is orthogonal (A^T A = I), eigenvalues lie on the unit circle in the complex plane.

Key relationships and concepts to know: - Av = λv expresses the basic eigenvalue equation; see linear transformation for the action of A on vectors. - The determinant and trace give global checks on the spectrum: det(A) equals the product of eigenvalues, and tr(A) equals the sum of eigenvalues. - The Schur decomposition (see Schur decomposition) provides a nearly diagonal form with eigenvalues on the diagonal, even when A is not diagonalizable. - In the real case, complex eigenvalues occur in conjugate pairs, reflecting the underlying algebra of polynomials with real coefficients.

Computation and algorithms

  • Closed-form solutions exist for small matrices by solving det(A − λI) = 0, but the roots can be intractable analytically as n grows.
  • For large matrices, numerical methods are essential. The most common approaches include:
    • Power method Power method: iteratively amplifies the dominant eigenvalue and its eigenvector.
    • Inverse power method (sometimes with shifts): targets eigenvalues near a chosen value.
    • QR algorithm QR algorithm: a staple in numerical linear algebra for computing the full spectrum by iteratively factoring A into Q R and recombining.
    • Jacobi method for symmetric matrices: iteratively zeroes off-diagonal entries to reveal the spectrum.
    • Schur methods and related decompositions (see Schur decomposition): preserve numerical stability while exposing eigenvalues.
  • Numerical concerns:
    • Conditioning and sensitivity: the eigenvalues of A can be highly sensitive to small perturbations in A, especially for non-normal matrices. This is where the study of the pseudospectrum becomes relevant in understanding how the spectrum behaves under perturbations.
    • Stability of algorithms: modern libraries (for example, those implementing LAPACK routines) rely on robust decompositions to provide accurate eigenvalues and eigenvectors.
  • Practical guidance: for many real-world problems, it is often more important to understand dominant modes (largest eigenvalues by magnitude) or a subset of eigenvalues rather than computing the full spectrum. See discussions around condition number and numerical linear algebra for more detail.

Applications and examples: - In dynamical systems, the sign of the real parts of eigenvalues of the system matrix A determines stability of equilibria. - In data analysis, the largest eigenvalues of a covariance matrix indicate principal directions of variance, a core idea in Principal component analysis. - In Markov processes, eigenvalues of the transition matrix reveal long-run behavior; the stationary distribution is related to the eigenvector associated with the eigenvalue 1 (see Markov chain and stationary distribution). - In physics, the energy levels of a quantum system correspond to eigenvalues of the Hamiltonian operator; the associated eigenvectors describe stationary states (see Hamiltonian (quantum mechanics)).

Interpretation and applications (examples and intuition): - In a physical system, a mode with a large eigenvalue represents a direction in which the system responds strongly to perturbations. - In an optimization context, eigenvalues of the Hessian determine local curvature: positive eigenvalues indicate local minima, negative indicate maxima or saddle points. - In graph theory and data networks, spectral properties of adjacency-like matrices reveal community structure and connectivity patterns.

Theoretical context and limitations

  • Spectral theory studies the full set of eigenvalues across different classes of operators, extending beyond finite matrices to infinite-dimensional spaces.
  • The distinction between eigenvalues and singular values is important in applications: singular values (from the singular value decomposition) always exist for any matrix and provide a robust measure of action, particularly for non-square or badly conditioned problems.
  • For non-normal matrices (where A A^T ≠ A^T A), eigenvalues may not fully capture transient dynamics, and pseudospectral analysis can offer a more accurate picture of behavior under perturbations.
  • In many practical settings, it is useful to complement eigenvalue analysis with related tools: singular value decomposition, Jordan canonical form for understanding defective matrices, and Schur decomposition for stable numerical computation.

See also