Eigenvalues and eigenvectors
Most vectors change direction when a matrix acts on them, but special vectors are only stretched. These eigenvectors, and their stretch factors, reveal the essential behaviour of a matrix.
Definition
A non-zero vector \(\mathbf{x}\) is an eigenvector of a square matrix A if \(A\mathbf{x} = \lambda\mathbf{x}\) for some scalar \(\lambda\), its eigenvalue. A sends the eigenvector to a multiple of itself: stretched if \(|\lambda| > 1\), shrunk if \(|\lambda| < 1\), and flipped if \(\lambda\) is negative. The word 'eigen' is German for 'own' or 'characteristic'.
Checking an eigenvector
To test whether \(\mathbf{x}\) is an eigenvector, compute \(A\mathbf{x}\) and see whether it is a multiple of \(\mathbf{x}\). For \(A = \begin{bmatrix} 3 & 1 \\ 0 & 2 \end{bmatrix}\), \(A(1, 0) = (3, 0) = 3(1, 0)\), so (1, 0) is an eigenvector with eigenvalue 3; and \(A(1, -1) = (2, -2) = 2(1, -1)\), giving eigenvalue 2.
Eigenspaces
For an eigenvalue \(\lambda\), the eigenvectors together with zero form the eigenspace, the null space of \(A - \lambda I\). It is a subspace, often a line but possibly larger. Finding eigenvectors for a known eigenvalue means solving the homogeneous system \((A - \lambda I)\mathbf{x} = \mathbf{0}\) by row reduction.
Zero as an eigenvalue
\(\lambda = 0\) is an eigenvalue exactly when \(A\mathbf{x} = \mathbf{0}\) has a non-zero solution, which happens exactly when A is singular. So A is invertible if and only if 0 is not an eigenvalue, completing another line of the invertible matrix theorem. The eigenspace for 0 is the null space.
Triangular matrices
The eigenvalues of a triangular matrix are the entries on its main diagonal, because A - lambda I is triangular and is singular exactly when a diagonal entry becomes zero. For other matrices, eigenvalues must be found from the characteristic equation, the subject of the next lesson.
Geometric examples
A reflection across a line has eigenvalue 1 for vectors on the line and -1 for vectors perpendicular to it. A projection onto a line has eigenvalues 1 and 0. A shear has only eigenvalue 1, with a single direction of eigenvectors. A rotation of the plane by 90 degrees has no real eigenvectors at all, since every non-zero vector changes direction; its eigenvalues are complex.
Why eigenvalues matter
Eigenvalues describe long-term behaviour of repeated processes, vibration frequencies of structures, stability of dynamical systems, principal directions of data (principal component analysis), energy levels in quantum mechanics, and the importance scores of web pages. Whenever a matrix is applied again and again, its eigenvalues decide what happens.
Worked example
Is \(\lambda = 4\) an eigenvalue of \(A = \begin{bmatrix} 2 & 2 \\ 1 & 3 \end{bmatrix}\)? \(A - 4I = \begin{bmatrix} -2 & 2 \\ 1 & -1 \end{bmatrix}\) has dependent rows, so it is singular and 4 is an eigenvalue. Solving \(-2x_1 + 2x_2 = 0\) gives the eigenspace spanned by (1, 1). Check: \(A(1, 1) = (4, 4) = 4(1, 1)\).
Key takeaways
An eigenvector is a non-zero x with Ax = lambda x; lambda is its eigenvalue, the stretch factor. Check by computing Ax. The eigenspace for lambda is Nul(A - lambda I), found by row reduction. Zero is an eigenvalue exactly when A is singular. Triangular matrices have their diagonal entries as eigenvalues. Reflections have eigenvalues 1 and -1, projections 1 and 0, and plane rotations other than 0 or 180 degrees have no real eigenvalues. Eigenvalues govern repeated processes, vibrations, stability, data analysis, and quantum mechanics.
Flashcard
Non-zero x with A x = lambda x
Check your understanding
For A = [[4, 0], [0, -1]], which vector is an eigenvector with eigenvalue -1?
