Concepts and theorems · Intermediate

Eigenvectors: directions preserved by a linear map

Use diagonal maps, shear, and rotation to understand eigenpairs, eigenspaces, diagonalization, and their limits.

Ask Logeia about this articleTry the three questionsReview your answers

Updated

On this page
  1. Exclude the zero vector
  2. Trace a diagonal example
  3. An eigenspace contains combinations
  4. Diagonalization needs enough independent vectors
  5. Check the field and numerical residual
  6. Check every coordinate of an eigenpair
  7. Ask Logeia about this articleCheck your understanding

Exclude the zero vector

For a square matrix A, a nonzero vector v satisfying Av=λv is an eigenvector, and λ is its eigenvalue. The transformation keeps the vector on its original line, not necessarily in the same orientation. A negative eigenvalue reverses direction, while a zero eigenvalue maps the vector to zero. The zero vector is excluded because it satisfies the equation for every scalar and identifies no direction. Specify the underlying field as well: allowing complex vectors and eigenvalues changes what exists. A statement about real eigenvectors is not automatically a statement about all complex eigenvectors of the same matrix.

Trace a diagonal example

For A=diag(3,1), the vector (1,0) maps to (3,0), so its eigenvalue is three. The vector (0,1) has eigenvalue one. But (1,1) maps to (3,1), which is not one scalar multiple of the original vector. Being expressible as a sum of eigenvectors does not make a vector an eigenvector. Likewise (2,0) belongs to the same eigenvalue-three direction as (1,0). Normalization chooses a convenient magnitude; it does not create a different eigenvalue. Verify a candidate by multiplying it and checking that the same scalar works for every coordinate.

An eigenspace contains combinations

For a fixed eigenvalue λ, the solutions of (A−λI)v=0 form its eigenspace, including zero so that the set is a linear subspace. Combinations of eigenvectors for the same eigenvalue remain in that eigenspace; a nonzero combination is another eigenvector. Combining vectors from different eigenvalues generally does not preserve this property. For A=2I, every nonzero vector has eigenvalue two. Thus having only one distinct eigenvalue does not mean there is only one independent eigenvector. Distinguish multiplicity as a root of the characteristic polynomial from the dimension of the corresponding space of eigenvectors.

Diagonalization needs enough independent vectors

Diagonalization requires n independent eigenvectors for an n-by-n matrix. Arranging them as columns of an invertible P gives A=PDP inverse. Finding eigenvalues alone does not supply this basis. The shear matrix with rows (1,1) and (0,1) has only the horizontal eigendirection for eigenvalue one, so two independent eigenvectors cannot be chosen. Multiplying a general vector makes the shear visible. Before using diagonalization to simplify powers or dynamics, verify that the needed basis exists. Real symmetric matrices have the stronger guarantee of an orthonormal eigenbasis, but arbitrary matrices do not inherit that theorem.

Check the field and numerical residual

A real matrix rotating the plane by ninety degrees has no nonzero real eigenvector: every direction moves to a perpendicular line. Complex eigenvalues i and minus i are available when the field is extended. In numerical work, inspect the residual Av−λv for computed candidates. A small residual measures closeness to the equation, but eigenvector sensitivity can still matter, especially for clustered eigenvalues and nonsymmetric matrices. Singular values are a different concept: they are nonnegative and apply to rectangular matrices. Do not substitute them for signed or complex eigenvalues merely because both are produced by matrix-decomposition routines.

Check every coordinate of an eigenpair

Multiplying diag(3,1) by (1,2) gives (3,2). The first coordinate alone suggests λ=3, but the second does not satisfy that same multiplier. Matching one coordinate is not enough. Avoid a coordinate-by-coordinate division test that breaks on zero components. Instead evaluate the full residual Av−λv and interpret it using a scale-appropriate tolerance. This verifies the actual equation while keeping the candidate eigenvalue consistent across every coordinate.

Reading and practice are recorded separately in this browser, for each language.

Learn this in order

Prepare for AI systems · STEP 9

Choose a learning path