1 Introduction to Eigenvalues and Eigenvectors
1.1 Eigenvalues as scaling factors
In linear algebra, an eigenvalue of a linear transformation measures how much the transformation scales vectors in a particular direction. If a matrix \(A\) acts on a nonzero vector \(v\) and the result is a multiple of the same vector, then that multiple is the eigenvalue. In symbols, \(Av=\lambda v\), where \(\lambda\) is the eigenvalue.
1.2 Eigenvectors and direction preservation
An eigenvector is a nonzero vector whose direction is preserved by the transformation. The transformation may stretch or shrink the vector, but it does not rotate it away from its original line. This “no change in direction” property is the geometric feature that makes eigenvectors useful for understanding a matrix.
1.3 From vectors to subspaces
While individual eigenvectors already provide insight, the collection of all eigenvectors associated with the same eigenvalue is more structured: it forms a linear subspace called an eigenspace. This shift from single directions to entire subspaces supports systematic computation and theory, including decomposition of vector spaces into independent components.
2 Definition of Eigenspace
2.1 Formal eigenspace definition
Let \(A\) be a linear transformation (represented by a matrix) and let \(\lambda\) be an eigenvalue. The eigenspace corresponding to \(\lambda\), denoted \(E_\lambda\), is the set of all vectors \(v\) such that \[ Av=\lambda v. \] Equivalently, \(E_\lambda=\{v \mid (A-\lambda I)v=0\}\).
2.2 Relationship to the equation \((A-\lambda I)v=0\)
Starting from \(Av=\lambda v\), subtract \(\lambda v\) from both sides to obtain \((A-\lambda I)v=0\). Thus, finding an eigenspace reduces to solving a homogeneous linear system. This connection ties eigenspaces directly to standard methods for computing null spaces.
2.3 Eigenspace as a linear subspace
2.3.1 Nontriviality and existence of eigenvalues
A subspace is called an eigenspace for \(\lambda\) precisely when it contains nonzero vectors, because eigenvalues are defined as scalars for which solutions exist beyond the trivial \(v=0\). When \(\lambda\) is an eigenvalue, the eigenspace \(E_\lambda\) is nontrivial; when \(\lambda\) is not an eigenvalue, the only solution is \(v=0\), so the eigenspace is trivial.
3 Computing Eigenspaces
3.1 Step-by-step method using \((A-\lambda I)\)
To compute \(E_\lambda\), one typically:
- Form the matrix \(A-\lambda I\).
- Solve the homogeneous equation \((A-\lambda I)v=0\).
- Express all solutions as a linear combination of a set of basis vectors.
This procedure relies on Gaussian elimination or other linear-system methods.
3.2 Solving the homogeneous system
The unknown vector \(v\) satisfies a set of linear equations derived from \((A-\lambda I)v=0\). The solution space consists of all vectors that satisfy every equation simultaneously. Because the system is homogeneous, it always includes the zero vector; the presence of additional free parameters determines how large the eigenspace is.
3.3 Describing the solution set with a basis
3.3.1 Dimension of the eigenspace (geometric multiplicity)
Once the system is solved, the free variables can be used to construct basis vectors for \(E_\lambda\). The number of basis vectors equals \(\dim(E_\lambda)\), often called the geometric multiplicity of \(\lambda\). This number indicates how many independent eigenvector directions are available for that eigenvalue.
4 Geometric Interpretation
4.1 Invariant directions under linear maps
An eigenspace consists of directions that remain aligned after applying the transformation. Any vector in the eigenspace is mapped to a vector pointing in the same line (though scaled). As a result, the eigenspace is an invariant set of directions for the transformation.
4.2 How scaling differs by eigenvalue
The scalar \(\lambda\) describes how strongly the transformation scales vectors in that eigenspace. Different eigenvalues correspond to different scaling behaviors: vectors in one eigenspace might be stretched while those in another might shrink, or even be negated if \(\lambda\) is negative.
4.3 Eigenspaces for simple vs. repeated eigenvalues
If an eigenvalue is simple (in algebraic terms, meaning it appears once as a root of the characteristic polynomial), its eigenspace can still have dimension greater than one, though typically it does not. For repeated eigenvalues, the eigenspace may enlarge, reflecting more independent invariant directions. The actual size is captured by geometric multiplicity.
5 Algebraic Properties
5.1 Direct relation to kernels
The defining equation \((A-\lambda I)v=0\) shows that the eigenspace is a kernel: \[ E_\lambda=\ker(A-\lambda I). \] This viewpoint clarifies many properties, since kernels are always subspaces and interact predictably with linear operations.
5.2 Sum and intersection of eigenspaces
Eigenspaces can overlap or combine. For different eigenvalues, their intersection is typically minimal; for the same eigenvalue, they coincide by definition. The structure of sums and intersections helps determine how the overall space decomposes into invariant components.
5.3 Eigenspaces for distinct eigenvalues
5.3.1 Independence of eigenvectors from different eigenspaces
If \(u\) lies in \(E_{\lambda}\) and \(w\) lies in \(E_{\mu}\) with \(\lambda\neq\mu\), then \(u\) and \(w\) cannot be scalar multiples of each other unless they are both zero. More broadly, sets of eigenvectors taken from distinct eigenspaces are linearly independent under appropriate hypotheses, supporting diagonalization procedures.
6 Multiplicity and Dimension Concepts
6.1 Algebraic multiplicity vs. geometric multiplicity
Algebraic multiplicity counts how many times an eigenvalue appears as a root of the characteristic polynomial. Geometric multiplicity is \(\dim(E_\lambda)\), the number of independent eigenvector directions. These quantities are related but need not be equal, especially when matrices are not diagonalizable.
6.2 Inequalities and what they imply
A standard relationship is that geometric multiplicity cannot exceed algebraic multiplicity. When the inequality is strict, it indicates that the transformation lacks enough eigenvectors to fully capture the space via diagonalization, often requiring generalized eigenvectors.
6.3 When eigenspaces are large
Large eigenspaces mean many independent invariant directions exist for that eigenvalue. This can strengthen the possibility of decomposing the space into invariant subspaces aligned with eigenstructure and can simplify computations such as spectral expansions.
6.3.1 Special cases: full geometric multiplicity
When geometric multiplicity equals algebraic multiplicity for each eigenvalue, enough eigenvectors exist to form a complete eigenbasis. In such cases, the matrix is diagonalizable, and the eigenspace decomposition becomes particularly clean.
7 Eigenspaces in Diagonalization
7.1 Diagonalizable matrices and eigenspace decomposition
A matrix is diagonalizable if the space can be spanned by eigenvectors. In that situation, each eigenspace contributes vectors to build a basis for the entire space. The diagonal form arises by expressing the transformation relative to a basis made from eigenvectors.
7.2 Building a basis from eigenvectors
To diagonalize \(A\), one collects a basis for each eigenspace \(E_{\lambda}\) and combines them. If the combined set spans the full vector space and is linearly independent, it forms an eigenbasis. The transformation then acts diagonally on coordinates in this basis.
7.3 Criterion using dimensions of eigenspaces
A common criterion uses dimensions: if the sum of the geometric multiplicities of all distinct eigenvalues equals the dimension of the underlying vector space, then the matrix is diagonalizable. This links the feasibility of diagonalization directly to the sizes of eigenspaces.
8 Eigenspaces and Invariant Subspaces
8.1 Invariant subspaces overview
An invariant subspace is a subspace \(W\) such that applying the linear map to any vector in \(W\) produces another vector still in \(W\). Invariance means the transformation respects the subspace’s internal structure.
8.2 Eigenspaces as invariant subspaces
Every eigenspace is invariant. If \(v\in E_\lambda\), then \(Av=\lambda v\), which is again in \(E_\lambda\) because it satisfies the same eigenvalue equation. This property is often used to understand how the transformation decomposes behavior across different directions.
8.3 Refining decomposition using eigenspaces
Eigenspaces provide a finer breakdown than general invariant subspaces. When multiple eigenvalues exist, one can often decompose the space into direct sums of eigenspaces (in the diagonalizable setting), yielding separate “channels” of action corresponding to different scaling factors.
9 Special Matrix Types (Conceptual Uses)
9.1 Symmetric and orthogonally diagonalizable matrices
For real symmetric matrices, eigenvalues are real and eigenvectors for distinct eigenvalues can be chosen orthogonal. Moreover, the transformation can be diagonalized using an orthonormal basis. This produces eigenspaces that fit naturally with geometric intuition about rotations and projections.
9.2 Normal matrices and orthogonal eigenspaces
Normal matrices satisfy \(A^*A=AA^*\), where \(A^*\) is the conjugate transpose. Such matrices admit a unitary diagonalization, and eigenspaces associated with distinct eigenvalues are orthogonal. This orthogonality simplifies spectral reasoning and computations.
9.3 Consequences for eigenspace structure
For these classes of matrices, eigenspace geometry is well-behaved: orthogonality and complete diagonalization are typical. As a result, eigenspaces can often be treated as clean building blocks for decomposing vectors and analyzing the action of the transformation.
10 Applications and Context
10.1 Understanding long-term behavior in linear systems
In discrete-time linear dynamical systems \(x_{k+1}=Ax_k\), eigenvalues strongly influence growth or decay over time. When eigenvectors and eigenspaces are available, one can express initial states in an eigenbasis and track how each component evolves according to its corresponding eigenvalue.
10.2 Model reduction using dominant eigenspaces
Sometimes only the most significant eigenvalues (often those with largest magnitude) materially affect a system’s behavior. Projecting data onto the corresponding eigenspaces yields a reduced representation that captures the dominant modes while discarding smaller contributions.
10.3 Connections to stability via eigenstructure
Stability analysis uses the idea that eigenvalues with magnitude less than one lead to decaying components, while those with magnitude greater than one lead to growth. Eigenstructure provides a structured way to interpret these outcomes, since each eigenspace describes the evolution of a particular invariant set of directions.
11 Common Pitfalls and Tips
11.1 Confusing eigenvectors with arbitrary vectors
Not every vector that is mapped to a multiple of itself is an eigenvector unless the scalar relationship matches a specific eigenvalue and the vector is nonzero. A frequent mistake is to test only one direction informally rather than verifying \(Av=\lambda v\) for the correct \(\lambda\).
11.2 Mistakes in solving \((A-\lambda I)v=0\)
Errors often arise from forming \(A-\lambda I\) incorrectly (such as sign mistakes on diagonal entries) or from algebra issues while solving the homogeneous system. Because eigenspaces depend on the exact null space, small computational slips can change the result.
11.3 Misreading dimension as number of eigenvalues
11.3.1 Forgetting to list a basis for the eigenspace
The dimension of an eigenspace is the number of independent vectors needed to span it, not the number of distinct eigenvalues. Another common issue is identifying the eigenspace dimension without explicitly constructing basis vectors, leaving the eigenspace conceptually incomplete for later tasks like diagonalization or projection.