Eigenvalue

An eigenvalue of a square matrix $ is a scalar $\lambda$ such that there exists a nonzero vector $ — called an eigenvector — satisfying = \lambda v$. Geometrically, eigenvectors are the directions that a linear transformation stretches or compresses without rotation, and eigenvalues quantify the amount of stretching. A negative eigenvalue indicates reversal; a complex eigenvalue indicates rotation within an invariant plane. The set of all eigenvalues of $ is called its spectrum, and it encodes the matrix's deepest structural properties: stability, resonance, growth, and decay.

The eigenvalue concept is not confined to matrices. It generalizes to any linear operator on a vector space, to differential operators (where eigenfunctions solve Sturm-Liouville problems), to graphs (where the spectral properties of adjacency matrices reveal community structure), and to dynamical systems (where eigenvalues of the Jacobian at a fixed point determine local stability). The eigenvalue is the thread that stitches together linear algebra, analysis, topology, and systems theory.

Definition and Geometry

Formally, $\lambda$ is an eigenvalue of $ if and only if the matrix - \lambda I$ is singular — that is, if $\det(A - \lambda I) = 0$. The polynomial (\lambda) = \det(A - \lambda I)$ is the characteristic polynomial of $, and its roots are precisely the eigenvalues. For an \times n$ matrix, the fundamental theorem of algebra guarantees $ eigenvalues counting multiplicities, though they may be complex even when $ is real.

The geometric multiplicity of an eigenvalue is the dimension of its eigenspace — the subspace of vectors that $ scales by exactly $\lambda$. The algebraic multiplicity is the multiplicity of $\lambda$ as a root of the characteristic polynomial. When geometric and algebraic multiplicities coincide for every eigenvalue, $ is diagonalizable: it admits a basis of eigenvectors and can be written = PDP^{-1}$ where $ is diagonal. When they do not coincide, $ is defective and must be analyzed through its Jordan normal form, which captures the missing eigenvectors through generalized eigenvectors.

Eigenvalues in Dynamical Systems

In a linear dynamical system $\dot{x} = Ax$, the eigenvalues of $ determine every qualitatively important feature of the flow. A real eigenvalue $\lambda < 0$ corresponds to exponential contraction along its eigenvector; $\lambda > 0$ to expansion. Complex eigenvalues $\lambda = \alpha \pm i\beta$ produce spiral trajectories, contracting if $\alpha < 0$ and expanding if $\alpha > 0$. The imaginary part $\beta$ sets the rotation frequency.

This makes eigenvalues the natural language of stability theory. A matrix is Hurwitz stable — all its eigenvalues have negative real parts — if and only if the origin of $\dot{x} = Ax$ is asymptotically stable. The spectral radius $\rho(A) = \max_i |\lambda_i|$ governs the asymptotic growth rate of ^k$ and the convergence of iterative methods. In nonlinear systems, the Hartman-Grobman theorem asserts that the behavior near a hyperbolic fixed point is topologically conjugate to the behavior of its linearization, making eigenvalues the local DNA of dynamics.

The eigenvalue framework extends naturally to control theory. The poles of a transfer function are the eigenvalues of the closed-loop system matrix, and pole placement — the deliberate assignment of eigenvalues through state feedback — is the central technique of linear control design. The Ackermann formula provides an explicit construction for single-input systems.

Spectral Theory and Networks

The eigenvalues of a graph's adjacency matrix or Laplacian encode structural information that is invisible to local inspection. The largest eigenvalue of the adjacency matrix bounds the graph's chromatic number and relates to its expansion properties. The second-smallest eigenvalue of the Laplacian — the algebraic connectivity or Fiedler value — measures how well-connected the graph is; it vanishes precisely when the graph is disconnected. This spectral perspective has become essential in network science, clustering algorithms, and the analysis of complex systems from protein interaction networks to social media graphs.

In quantum mechanics, eigenvalues acquire physical reality as measurable quantities. The eigenvalues of the Hamiltonian operator are the allowed energy levels of a quantum system, and the spectral theorem for self-adjoint operators guarantees that these eigenvalues are real — a mathematical fact with profound physical consequences. The discrete spectrum of the hydrogen atom, the band structure of solids, and the energy gaps that enable superconductivity are all eigenvalue phenomena.

Computation and the Limits of Exactness

Computing eigenvalues is one of the most thoroughly studied problems in numerical analysis. The QR algorithm, introduced independently by John Francis and Vera Kublanovskaya in the early 1960s, iteratively reduces a matrix to upper Hessenberg form and then to quasi-triangular Schur form, revealing the eigenvalues on the diagonal. For large sparse matrices, iterative methods such as the Arnoldi and Lanczos algorithms target specific regions of the spectrum without forming the matrix explicitly.

Yet eigenvalue computation is fundamentally ill-conditioned for non-normal matrices. A small perturbation to a non-normal matrix can produce large changes in its eigenvalues, even when the matrix itself is barely changed. This sensitivity explains why pseudospectra — the sets of points that are eigenvalues of matrices within a given perturbation distance — often provide more reliable information about stability and transient growth than the eigenvalues themselves.

The eigenvalue is perhaps the most misunderstood concept in applied mathematics. We teach it as a number that a matrix assigns to a vector, a scalar that survives multiplication unchanged. But this is the pedagogical view, not the systems view. In reality, eigenvalues are not properties of matrices; they are properties of operators, of dynamics, of graphs, of quantum systems. The matrix is merely one instantiation. The eigenvalue lives at the intersection of algebra and geometry, of local and global, of structure and behavior. To ask for the eigenvalues of a system is to ask how it grows, how it oscillates, how it connects, how it dies. The claim that linear algebra is "done" once you can compute eigenvalues is not just wrong; it is backwards. Eigenvalue computation is where linear algebra begins to become systems theory. Any curriculum that ends with diagonalization has mistaken the airport for the destination.