Can a defective matrix be inverted?

Yes, provided zero is not an eigenvalue. Defectiveness and invertibility are different properties.

Are all repeated-eigenvalue matrices defective?

No. A scalar multiple of the identity has repeated eigenvalues and a complete eigenvector basis.

Are real symmetric matrices defective?

No. They admit an orthonormal eigenvector basis, including when some eigenvalues repeat.

AI Glossary / Mathematics and optimization / Intermediate

Defective matrix

Definition

A defective matrix is a square matrix that lacks enough linearly independent eigenvectors to form a basis. For an n×n complex matrix, this means fewer than n independent eigenvectors exist, so the matrix cannot be diagonalized. The issue concerns its eigenvector structure, not whether its entries are corrupted or its determinant is zero.

What it means

Diagonalization expresses a matrix as A=PDP⁻¹, where D is diagonal and the columns of P form an eigenvector basis. This can simplify repeated matrix operations and the analysis of linear dynamics. A defective matrix prevents this particular simplification because an invertible eigenvector matrix P cannot be assembled.

There may still be infinitely many eigenvectors along one direction. Multiplying a single eigenvector by different nonzero scalars does not create independent directions. Counting distinct written vectors therefore misses the relevant question: how many independent directions do the eigenspaces supply?

How it works

Compare each eigenvalue's algebraic multiplicity, its repetition as a root of the characteristic polynomial, with its geometric multiplicity, the dimension of its eigenspace. Over the complex numbers, a deficiency occurs when an eigenspace has fewer independent directions than the associated algebraic multiplicity.

Generalized eigenvectors extend ordinary eigenvectors into chains and lead to Jordan form, whose nontrivial blocks describe the missing diagonalization. The matrix can still act on every vector, and many useful computations remain possible.

For numerical work, exact defectiveness is delicate: small perturbations can change eigenvalue multiplicities and restore diagonalization while leaving an ill-conditioned eigenvector basis. Algorithms and tolerances must distinguish exact algebraic statements from floating-point evidence. Lack of a real eigenbasis also requires care; a real rotation can have a complete complex eigenbasis without being defective over the complex numbers.

Three practical examples

1. Linear dynamics: understand why a system's solution may include polynomial factors multiplying exponential terms, rather than only independent exponential modes.

2. Numerical diagnostics: investigate an eigenvector decomposition whose basis is singular or nearly singular instead of assuming the displayed eigenvalues are sufficient.

3. Applied mathematics education: distinguish invertibility, repeated eigenvalues, and diagonalization before using spectral formulas in model analysis.

Think of it this way

Think of trying to describe movement using independent directional arrows. An eigenvector basis would provide enough special arrows that the matrix merely scales each one. A defective matrix supplies too few independent arrows of that kind.

The space itself is not missing dimensions. Ordinary coordinate vectors still span it; only the desired eigenvector coordinate system is unavailable. The analogy therefore concerns a special representation, not a broken geometric space.

Worked example

Consider the original example A=[[3,2],[0,3]]. Its characteristic polynomial is (λ−3)², so the only eigenvalue is 3 with algebraic multiplicity 2.

Solving (A−3I)v=0 gives 2v2=0. Hence v2=0 and every eigenvector is a nonzero multiple of (1,0). The eigenspace has dimension 1, so A is defective. Yet det(A)=9, making A invertible.

Write A=3I+N with N=[[0,2],[0,0]]. Since N²=0, the binomial expansion gives A²=9I+6N=[[9,12],[0,9]]. Direct multiplication produces the same matrix, confirming that defectiveness does not prevent ordinary matrix powers.

For a generalized eigenvector, choose w=(0,0.5). Then (A−3I)w=(1,0), the ordinary eigenvector. This creates a chain rather than a second independent ordinary eigenvector.

By contrast, B=3I has the same repeated eigenvalue and determinant, but every nonzero vector is an eigenvector. Two independent coordinate vectors diagonalize B. Thus repeated eigenvalues alone do not prove defectiveness. These are exact hand calculations, not numerical eigensolver output.

Illustrative case study

Fictional case: Aspen Simulation analyses a two-state linear model and expects two independent eigenmodes. Its report prints the eigenvalue 3 twice, but the two returned eigenvector columns are nearly parallel.

The analyst checks the exact model structure and recognizes the defective example above. Instead of inverting an unsuitable eigenvector matrix, the team uses a method appropriate to the required matrix function and verifies the result directly. It also records how small measurement perturbations affect the interpretation. This scenario illustrates a diagnostic response, not a claim that all nearly parallel eigenvectors prove exact defectiveness.

Common misconceptions and mistakes

Defective does not mean singular: the worked matrix is defective and invertible. Conversely, a diagonal matrix containing zero is singular but already diagonalized.

A repeated eigenvalue is necessary for a complex matrix to be defective, but is not sufficient. Its eigenspace dimension matters.

Do not classify a matrix solely from rounded eigenvalues or the number of columns returned by software. An eigensolver can return n columns without those columns forming a well-conditioned, independent basis. Numerical rank and eigenvector conditioning deserve separate attention.

Business use

The term matters when a technical workflow assumes a full set of independent modes. Detecting a violated assumption can prevent unstable reconstructions or misleading interpretations of system behavior.

Ask what calculation the business actually needs: a prediction step, matrix exponential, sensitivity estimate, or explanatory decomposition. A stable alternative decomposition may be more useful than trying to force diagonalization. Preserve tolerances and the distinction between exact model structure and estimated data so subsequent reviewers can understand the basis of the conclusion.

Check your understanding

Why does A=[[3,2],[0,3]] have only one independent eigenvector direction despite two occurrences of eigenvalue 3?

Answer: The eigenvector equation forces the second coordinate to zero. All eigenvectors lie on the same line, giving geometric multiplicity 1 while algebraic multiplicity is 2.

Related terms

Eigenvalue
Eigenvector
Schur decomposition
diagonalisation
algebraic multiplicity
geometric multiplicity
Jordan form
generalised eigenvector
matrix conditioning

Learn next

Compare a diagonal matrix, the worked defective matrix, and a small perturbation that separates its eigenvalues. Solve the eigenspaces by hand before using software. Then study generalized eigenvector chains and why stable numerical methods often favor Schur form over explicit Jordan-form calculations.

Eigenvector

Sources

Primary sources: UT Austin's ALAFF textbook, “Jordan Canonical Form”, https://www.cs.utexas.edu/~flame/laff/alaff/chapter09-jordan-canonical-form.html ; Steven G. Johnson's MIT notes, https://web.mit.edu/18.06/www/Spring09/generalized-eigenvectors.pdf . The primary textbook's definition and multiplicity discussion were read. Worked matrices, arithmetic and fictional scenario are original; no eigensolver execution is claimed.

Previous: Default rules Next: Defective sample