1 Basic concepts
Eigenvalues describe how a linear transformation stretches or compresses vectors in special directions. Rather than changing every vector in an arbitrary way, a matrix may preserve the direction of certain nonzero vectors while altering their length by a fixed scalar factor. That scalar is called an eigenvalue, and the corresponding vector is an eigenvector. The pair gives a compact way to understand the structure of a linear map.
1.1 Definitions
For a square matrix or linear transformation \(A\), a scalar \(\lambda\) is an eigenvalue if there exists a nonzero vector \(v\) such that \(Av=\lambda v\). This relation says that \(v\) is mapped onto the same line through the origin, scaled by \(\lambda\). The vector \(v\) is then called an eigenvector associated with \(\lambda\).
1.1.1 Eigenvectors
An eigenvector is a nonzero vector whose direction is unchanged by the transformation. The vector may be lengthened, shortened, or reversed, depending on the eigenvalue. Zero vectors are excluded because they satisfy the defining equation for every scalar and therefore carry no directional information.
1.1.2 Eigenspaces
The eigenspace for an eigenvalue \(\lambda\) is the set of all vectors satisfying \(Av=\lambda v\), together with the zero vector. It is a subspace of the underlying vector space. Each eigenspace collects all vectors that share the same scaling behavior under the transformation.
1.2 Matrix representation
When a linear transformation is represented by a matrix, eigenvalues depend on the chosen basis only through the matrix form of the same transformation. If the basis changes, the matrix entries change, but the eigenvalues remain the same for similar matrices. This makes eigenvalues intrinsic to the transformation rather than to a particular coordinate system.
1.3 Geometric interpretation
Geometrically, an eigenvector lies along a direction that the transformation does not rotate away from itself. A positive eigenvalue preserves orientation on that line, while a negative eigenvalue reverses it. An eigenvalue with absolute value greater than 1 stretches the vector, and one between 0 and 1 compresses it. An eigenvalue of 0 collapses the vector to the origin.
1.4 Examples
A diagonal matrix has its diagonal entries as eigenvalues, with the standard basis vectors as eigenvectors. A projection matrix may have eigenvalues 1 and 0, corresponding to vectors kept fixed and vectors sent to zero. A rotation in the plane typically has no real eigenvectors unless it is a half-turn or the zero angle case, though it may have complex eigenvalues.
2 Existence and computation
Eigenvalues are found by solving an equation derived from the condition that nonzero eigenvectors must exist. In practice, exact formulas work for small matrices or special structures, while larger systems often require numerical algorithms. The details depend on the matrix type and the desired accuracy.
2.1 Characteristic equation
The characteristic equation is obtained by requiring that the matrix \(A-\lambda I\) have a nontrivial null space. This condition leads to a polynomial equation in \(\lambda\). Its roots are the eigenvalues of the matrix.
2.1.1 Characteristic polynomial
The characteristic polynomial of a matrix \(A\) is \(\det(A-\lambda I)\) or, in some conventions, \(\det(\lambda I-A)\). It is a polynomial whose degree equals the size of the matrix. The coefficients encode information about traces, determinants, and other structural features.
2.1.2 Determinant condition
A nonzero solution to \((A-\lambda I)v=0\) exists exactly when \(A-\lambda I\) is singular. This happens when its determinant is zero. Thus eigenvalues are precisely the scalars \(\lambda\) for which \(\det(A-\lambda I)=0\).
2.2 Finding eigenvalues
For many matrices, eigenvalues are computed by solving the characteristic equation directly. For larger or sparse matrices, one usually relies on iterative numerical procedures. The choice of method depends on whether exact symbolic expressions or approximate values are needed.
2.2.1 Direct methods
Direct methods include factoring the characteristic polynomial, using algebraic identities, or exploiting matrix structure. These approaches are effective for low-dimensional matrices and those with simple patterns. In dimensions three and above, exact formulas may become cumbersome or impossible to express neatly.
2.2.2 Numerical methods
Numerical methods approximate eigenvalues by repeated computation. Common techniques include the power method, QR iteration, and methods for sparse matrices. These algorithms are designed to be stable and efficient for large-scale problems, though they provide approximations rather than symbolic answers.
2.3 Special cases
Certain matrix classes allow eigenvalues to be read off or identified with minimal work. These cases are especially important because they reveal how matrix structure simplifies spectral behavior. They also serve as benchmarks for more general methods.
2.3.1 Diagonal matrices
For a diagonal matrix, the eigenvalues are exactly the diagonal entries. The corresponding eigenvectors are the standard basis vectors, provided the entries are considered individually. This is the simplest possible spectral situation.
2.3.2 Triangular matrices
For upper or lower triangular matrices, the eigenvalues are the diagonal entries as well. This follows because the characteristic polynomial factors into linear terms determined by the diagonal. The off-diagonal entries affect eigenvectors but not the eigenvalues.
2.3.3 Symmetric matrices
Real symmetric matrices have especially well-behaved eigenvalues. They always have real eigenvalues, and eigenvectors associated with distinct eigenvalues are orthogonal. This makes symmetric matrices central in geometry, optimization, and physical modeling.
3 Properties
Eigenvalues satisfy several algebraic and geometric properties that make them useful in both theory and computation. These properties connect spectral data with invariant quantities such as trace and determinant. They also explain why eigenvalues remain stable under changes of basis.
3.1 Algebraic multiplicity
The algebraic multiplicity of an eigenvalue is the number of times it appears as a root of the characteristic polynomial. An eigenvalue may occur once or several times. Repetition affects the structure of the eigenspace and the matrix’s possible forms under similarity.
3.1.1 Root multiplicity
Root multiplicity refers to the count of repeated roots in a polynomial. For eigenvalues, this count is taken from the characteristic polynomial. A repeated root indicates that the same scaling factor governs multiple dimensions of the transformation, though not necessarily in the same geometric manner.
3.2 Geometric multiplicity
The geometric multiplicity of an eigenvalue is the dimension of its eigenspace. It measures how many linearly independent eigenvectors correspond to that eigenvalue. This quantity is always at least 1 for an eigenvalue and never exceeds its algebraic multiplicity.
3.3 Invariance under similarity
Similar matrices represent the same linear transformation in different bases. They have identical eigenvalues, even though their entries may differ. This invariance is one reason eigenvalues are treated as fundamental characteristics of a transformation rather than of a specific matrix arrangement.
3.4 Trace and determinant relations
The trace of a matrix equals the sum of its eigenvalues, counted with algebraic multiplicity. The determinant equals the product of its eigenvalues, also counted with multiplicity. These relations provide quick checks and connect spectral information to basic matrix invariants.
3.5 Real and complex eigenvalues
A matrix with real entries may have real eigenvalues, complex eigenvalues, or both. Complex eigenvalues often appear in conjugate pairs for real matrices. When the matrix is symmetric, however, all eigenvalues are real, which gives a particularly orderly spectral picture.
4 Types of eigenvalues
Eigenvalues are often classified according to how they appear in the characteristic polynomial or how they influence numerical behavior. These categories are useful in theoretical discussions and in applications such as stability and approximation. Some types are defined algebraically, while others are tied to dominance or degeneracy.
4.1 Simple eigenvalues
A simple eigenvalue occurs with algebraic multiplicity one. It corresponds to a single root of the characteristic polynomial. Such eigenvalues usually lead to a one-dimensional eigenspace and tend to be easier to analyze.
4.2 Repeated eigenvalues
A repeated eigenvalue appears more than once as a root of the characteristic polynomial. Repetition does not automatically imply a large eigenspace, because geometric multiplicity may be smaller than algebraic multiplicity. Repeated eigenvalues often signal a more intricate matrix structure.
4.3 Dominant eigenvalues
A dominant eigenvalue is one whose magnitude exceeds that of the others. In iterative methods, it often controls long-term behavior because repeated application of the matrix amplifies its associated direction. Dominant eigenvalues are important in population models, Markov processes, and numerical algorithms.
4.4 Zero eigenvalues
A zero eigenvalue indicates that the matrix is singular and has a nontrivial null space. Vectors in the corresponding eigenspace are sent to zero. Such eigenvalues are closely related to loss of invertibility and dimensional collapse.
5 Related concepts
Eigenvalues are studied together with several closely connected notions that describe how a matrix can be decomposed or simplified. These ideas help translate abstract spectral information into usable forms. They also clarify when a matrix can be treated as essentially diagonal in some basis.
5.1 Eigenvectors
Eigenvectors provide the directional component of spectral data. Each eigenvalue may have one or more eigenvectors, depending on the dimension of its eigenspace. Together, eigenvalues and eigenvectors reveal invariant directions of the transformation.
5.2 Eigenbases
An eigenbasis is a basis made entirely of eigenvectors. When such a basis exists, the matrix becomes diagonal in that basis. Eigenbases are especially valuable because they reduce complicated transformations to simple coordinate-wise scaling.
5.3 Diagonalization
Diagonalization is the process of expressing a matrix as similar to a diagonal matrix. If successful, it makes repeated powers of the matrix and many computations much easier. The diagonal entries of the simplified matrix are the eigenvalues.
5.3.1 Conditions for diagonalizability
A matrix is diagonalizable when it has enough linearly independent eigenvectors to form a basis. Equivalently, the sum of the dimensions of its eigenspaces must equal the size of the matrix. Distinct eigenvalues help, but repeated eigenvalues may still allow diagonalization if enough eigenvectors exist.
5.3.2 Jordan canonical form
The Jordan canonical form is a nearly diagonal representation that applies when full diagonalization is not possible. It groups equal eigenvalues into blocks that encode generalized eigenvectors and nilpotent structure. This form clarifies how repeated eigenvalues can create more complicated behavior.
5.4 Spectral theorem
The spectral theorem states that certain classes of matrices, especially real symmetric or complex Hermitian matrices, can be diagonalized by an orthogonal or unitary change of basis. It guarantees real eigenvalues in these settings and provides orthonormal eigenvectors. The theorem is one of the most powerful results in linear algebra.
6 Applications
Eigenvalues appear in many mathematical models where a system’s long-term or modal behavior matters. They often determine stability, frequency, or principal directions of variation. Because of this, they are central tools across science and engineering.
6.1 Systems of differential equations
In linear differential equation systems, eigenvalues govern the growth or decay of solutions. The sign of the real part of an eigenvalue often determines whether a mode decays, persists, or grows over time. This makes spectral analysis a standard method for solving and classifying such systems.
6.2 Stability of dynamical systems
Eigenvalues are used to assess whether a fixed point is stable, unstable, or neutrally stable. If relevant eigenvalues lie within a stable region, nearby trajectories tend to return or remain close. This approach is widely used in both continuous and discrete dynamical models.
6.3 Vibration and modal analysis
In mechanical systems, eigenvalues correspond to natural frequencies or squared frequencies of vibration, depending on the formulation. The associated eigenvectors describe mode shapes, showing how a structure oscillates. Modal analysis helps engineers predict resonance and separate complex motion into simpler components.
6.4 Quantum mechanics
In quantum mechanics, observable quantities are represented by operators whose eigenvalues correspond to possible measurement outcomes. Eigenvectors represent states with definite values of the observable. This spectral viewpoint is essential for understanding energy levels and state evolution.
6.5 Principal component analysis
Principal component analysis uses eigenvalues of a covariance matrix to measure how much variance is captured by each principal direction. Larger eigenvalues indicate directions that explain more of the data’s spread. This method is common in statistics, data compression, and feature reduction.
7 Generalizations
The concept of eigenvalue extends beyond finite matrices to broader mathematical settings. In these contexts, the same basic idea persists: a transformation acts by scaling certain special vectors or modes. The details, however, become more subtle in infinite-dimensional or nonlinear environments.
7.1 Linear operators on vector spaces
Eigenvalues are defined for linear operators on vector spaces just as they are for matrices. The matrix case is only a coordinate representation of this more general concept. The operator viewpoint is often more natural in abstract algebra and functional analysis.
7.2 Infinite-dimensional settings
In infinite-dimensional spaces, such as spaces of functions, eigenvalues may behave differently from the finite-dimensional case. Some operators have discrete spectra, while others have continuous spectral components. The study of these operators requires additional analytical tools.
7.3 Generalized eigenvalues
Generalized eigenvalues arise in problems of the form \(Av=\lambda Bv\), where two matrices are involved. Such equations appear in constrained systems, vibration models, and differential discretizations. The pair \((A,B)\) is analyzed instead of a single matrix alone.
7.4 Eigenvalues in non-linear contexts
In nonlinear settings, the term eigenvalue may be extended in specialized ways to describe scaling-like behavior in nonlinear operators or equations. These generalizations are not always unique and often depend on the framework being used. They preserve the core idea of identifying special modes associated with scalar factors.