How To Find Bases For Eigenspaces

5 min read

Finding a basis for an eigenspace is one of the most fundamental tasks when working with linear operators on vector spaces. In this practical guide, we walk through reliable techniques for identifying the independent eigenvectors associated with each eigenvalue, so you can construct a correct basis for every eigenspace and move forward confidently in your studies. Whether you are preparing for an exam, solving real‑world dynamical systems, or simply curious about the elegance of linear algebra, mastering the process of extracting bases from eigenspaces will greatly enhance your understanding and problem‑solving abilities Still holds up..

Introduction

When a square matrix (A) represents a linear transformation, its eigenvalues (\lambda_i) and corresponding eigenvectors form the core ingredients for decomposing the space into invariant subspaces. An eigenspace for (\lambda_i) is defined as the set

[ E_{\lambda_i}= {,v \in \mathbb{R}^n \mid A v = \lambda_i v ,}. ]

Geometrically, each eigenspace is the span of all eigenvectors belonging to the same eigenvalue. To speak about a basis for (E_{\

For each eigenvalue (\lambda_i) the set

[ E_{\lambda_i}= {,v \in \mathbb{R}^n \mid A v = \lambda_i v ,} ]

is precisely the null‑space of the shifted matrix (A-\lambda_i I). Geometrically this translates to “the line, plane, or higher‑dimensional space spanned by all eigenvectors that share the same eigenvalue.” To speak meaningfully about a basis for (E_{\lambda_i}) we therefore focus on extracting a maximal set of linearly independent solutions to the homogeneous equation

[ (A-\lambda_i I),v = 0 . ]


Step‑by‑step recipe

  1. Obtain the spectrum.
    Determine the eigenvalues (\lambda_1,\dots,\lambda_k) by solving the characteristic equation (\det(A-\mu I)=0). Algebraic multiplicities tell us how many times each root appears; geometric multiplicities will decide whether our basis can reach those counts.

  2. Form the shifted matrix.
    For a given (\lambda_i) construct

    [ B_i := A-\lambda_i I_n . ]

  3. Row‑reduce (B_i).
    Perform Gaussian elimination (or compute the rank) to bring (B_i) to its reduced row‑echelon form. Identify the pivot columns and the corresponding free variables.

  4. Parameterise the solution space.
    Let the number of free variables be (r). Assign each free variable a parameter (t_j) ((j=1,\dots ,r)). Write the general solution as

    [ v = \mathbf{v}_p + t_1\mathbf{v}^{(1)} + \cdots + t_r\mathbf{v}^{(r)}, ]

    where (\mathbf{v}_p) is the particular solution associated with the pivot variables, and (\mathbf{v}^{(j)}) are vectors obtained by setting one free parameter to 1 while keeping the others at 0 Took long enough..

  5. Extract the basis vectors.
    Collect the coefficient vectors of each (\mathbf{v}^{(j)}); these form a basis for the null‑space of (B_i) and hence for (E_{\lambda_i}). Because the null‑space is defined over (\mathbb{R}), any real non‑zero combination of them is also admissible Practical, not theoretical..

  6. Verify.
    Substitute each candidate eigenvector back into (Av) to confirm (Av = \lambda_i v). If the field were complex, ensure the resulting vectors belong to the appropriate complex eigenspace; for real problems, conjugate pairs often generate a two‑dimensional real invariant subspace Most people skip this — try not to..


Practical tips

  • Repeated eigenvalues: When an eigenvalue occurs with algebraic multiplicity larger than its geometric multiplicity, the process above still yields a basis for

Repeated eigenvalues.
When an eigenvalue occurs with algebraic multiplicity larger than its geometric multiplicity, the process described above still yields a basis for the eigenspace

[ E_{\lambda}= {,v\in\mathbb{R}^{n}\mid (A-\lambda I)v=0,}, ]

but the dimension of this space (the geometric multiplicity) is strictly smaller than the algebraic multiplicity. Consequently the matrix cannot be diagonalised by a collection of ordinary eigenvectors alone; one must supplement the ordinary eigenvectors with generalised eigenvectors to obtain a complete basis of (\mathbb{R}^{n}).


Generalised eigenvectors and Jordan chains

A vector (v\neq0) is a generalised eigenvector of rank (k) for the eigenvalue (\lambda) if

[ (A-\lambda I)^{k}v=0\qquad\text{but}\qquad (A-\lambda I)^{k-1}v\neq0 . ]

The smallest such (k) is called the index of the chain.
If (v) is a generalised eigenvector of rank (k), then the vectors

[ (A-\lambda I)^{k-1}v,;(A-\lambda I)^{k-2}v,;\dots,;(A-\lambda I)v,;v ]

form a Jordan chain of length (k). The last vector in the chain, ((A-\lambda I)v), is an ordinary eigenvector, while the preceding vectors are “higher‑order’’ eigenvectors that satisfy the homogeneous equations for higher powers of (A-\lambda I).

The set of all Jordan chains for a given eigenvalue (\lambda) furnishes a basis for the generalised eigenspace

[ G_{\lambda}= {,v\in\mathbb{R}^{n}\mid (A-\lambda I)^{m}v=0\ \text{for some }m\le n,}, ]

which always has dimension equal to the algebraic multiplicity of (\lambda). When the matrix is diagonalisable, each Jordan block has size 1 and (G_{\lambda}=E_{\lambda}).


Constructing a Jordan basis (step‑by‑step)

  1. Compute the minimal polynomial.
    The exponent of each factor ((\mu-\lambda_i)) in the minimal polynomial (p(\mu)) tells the size of the largest Jordan block for (\lambda_i).

  2. Determine the sizes of Jordan blocks.
    Use the partition of the algebraic multiplicity that is consistent with the elementary divisors derived from (p(\mu)). For a single eigenvalue (\lambda) this reduces to finding a partition of its algebraic multiplicity (a) whose largest part equals the index of (\lambda) Turns out it matters..

  3. Find the longest chain.
    Solve ((A-\lambda I)^{k}v=0) for the smallest (k) such that the null‑space dimension jumps (i.e., (\dim\ker (A-\lambda I)^{k} > \dim\ker (A-\lambda I)^{k-1})). Any non‑zero vector in the new kernel that is not

Freshly Written

Current Topics

Curated Picks

Cut from the Same Cloth

Thank you for reading about How To Find Bases For Eigenspaces. We hope the information has been useful. Feel free to contact us if you have any questions. See you next time — don't forget to bookmark!
⌂ Back to Home