Linear Algebra › Section E · Eigenvalues and eigenvectors › Topic E.2
The characteristic equation
Testing one guess at a time is no way to find eigenvalues. A single determinant turns the search into solving a polynomial equation, and the same polynomial explains why the trace and determinant are tied to the eigenvalues.
From guessing to solving
The previous topic, "Eigenvectors and eigenvalues", showed how to check whether a given number \(\lambda\) is an eigenvalue: ask whether \(A - \lambda I\) fails to be invertible. That test is useful only when someone hands you a candidate. To find every eigenvalue, combine it with a fact from Section C: a square matrix is not invertible exactly when its determinant is zero. So \(\lambda\) is an eigenvalue of \(A\) if and only if
This is the characteristic equation of \(A\). Treat \(\lambda\) as an unknown, expand the determinant, and solve.
The characteristic polynomial
When \(\lambda\) is left as a symbol, each entry on the diagonal of \(A - \lambda I\) contains \(-\lambda\), and expanding the determinant multiplies at most \(n\) of them together. So for an \(n \times n\) matrix, \(\det(A - \lambda I)\) is a polynomial of degree exactly \(n\) in \(\lambda\), called the characteristic polynomial. Its roots are the eigenvalues. A polynomial of degree \(n\) has at most \(n\) roots, which confirms that an \(n \times n\) matrix has at most \(n\) eigenvalues. Some roots may be complex even when every entry of \(A\) is real: the 90° rotation gives \(\lambda^2 + 1 = 0\). "Complex eigenvalues" deals with those; here we stay with real roots.
The 2 × 2 shortcut
For a 2 × 2 matrix the polynomial has a pattern worth memorising.
The number \(a + d\), the sum of the main diagonal, is the trace of \(A\), written \(\operatorname{tr} A\), and \(ad - bc\) is \(\det A\). So every 2 × 2 characteristic equation reads
Two quick calculations give the whole equation, and the quadratic formula or factoring finishes the job.
Worked example: a 2 × 2 matrix
Find the eigenvalues and eigenvectors of
1. The trace is \(2 + 1 = 3\) and the determinant is \(2 - 12 = -10\). 2. The characteristic equation is \(\lambda^2 - 3\lambda - 10 = 0\), which factors as \((\lambda - 5)(\lambda + 2) = 0\), so \(\lambda = 5\) or \(\lambda = -2\). 3. For \(\lambda = 5\), \(A - 5I\) has rows (-3, 3) and (4, -4), so \(x_1 = x_2\) and an eigenvector is (1, 1). 4. For \(\lambda = -2\), \(A + 2I\) has rows (4, 3) and (4, 3), so \(4x_1 + 3x_2 = 0\) and an eigenvector is (3, -4). 5. Check: \(A(3, -4) = (-6, 8) = -2(3, -4)\).
Multiplicities
A root can repeat. The algebraic multiplicity of an eigenvalue is the number of times it appears as a root of the characteristic polynomial. Its geometric multiplicity is the dimension of its eigenspace, the number of free variables in \(A - \lambda I\). The geometric multiplicity is always at least 1 and never more than the algebraic multiplicity. The two can differ: the matrix with rows (3, 1) and (0, 3) has characteristic polynomial \((\lambda - 3)^2\), so 3 has algebraic multiplicity 2, but \(A - 3I\) has rows (0, 1) and (0, 0), one free variable, so its geometric multiplicity is only 1. The next topic, "Diagonalization", shows why this gap matters.
Worked example: a 3 × 3 matrix
Find the eigenvalues of \(A\) and their multiplicities.
1. Expand \(\det(A - \lambda I)\) along the first row, which has two zeros: \((2 - \lambda)\big[(3 - \lambda)^2 - 1\big]\). 2. Simplify the bracket: \(\lambda^2 - 6\lambda + 8 = (\lambda - 2)(\lambda - 4)\). 3. So the polynomial is \(-(\lambda - 2)^2(\lambda - 4)\): eigenvalue 2 with algebraic multiplicity 2, and eigenvalue 4 with multiplicity 1. 4. Every row of \(A - 2I\) is a multiple of (1, 1, 1), giving two free variables, so 2 has geometric multiplicity 2. 5. An eigenvector for 4 is (0, 1, 1): \(A(0, 1, 1) = (0, 4, 4)\).
Trace and determinant as checks
Counting each eigenvalue as often as its algebraic multiplicity, and allowing complex values, the eigenvalues of any \(n \times n\) matrix satisfy
The product rule is easy to see: the polynomial factors as \((\lambda_1 - \lambda)(\lambda_2 - \lambda) \cdots (\lambda_n - \lambda)\), and putting \(\lambda = 0\) gives \(\det A\) on one side and the product of the eigenvalues on the other. In the 3 × 3 example, \(2 + 2 + 4 = 8\) is the trace and \(2 \cdot 2 \cdot 4 = 16\) is the determinant. Use these two checks every time.
Similar matrices
Matrices \(A\) and \(B\) are similar if \(B = P^{-1}AP\) for some invertible \(P\). Similar matrices have the same characteristic polynomial, because \(B - \lambda I = P^{-1}(A - \lambda I)P\) and determinants multiply, so the factors \(\det P^{-1}\) and \(\det P\) cancel. They therefore share eigenvalues with multiplicities, trace and determinant, though not eigenvectors: if \(A\mathbf{v} = \lambda\mathbf{v}\), then \(P^{-1}\mathbf{v}\) is the eigenvector of \(B\). The converse fails. The matrix with rows (3, 1) and (0, 3) has the same polynomial as \(3I\), yet \(P^{-1}(3I)P = 3I\) for every \(P\), so nothing but \(3I\) is similar to \(3I\). Similarity is also not row equivalence, since row operations change eigenvalues.
How eigenvalues are computed in practice
Hand calculation through the characteristic polynomial works well for 2 × 2 and 3 × 3 matrices. Beyond that it breaks down twice over. The Abel-Ruffini theorem says there is no general formula in radicals for the roots of polynomials of degree 5 or more. Even with numerical root-finders, the roots of a polynomial can shift a long way when its coefficients are rounded slightly, so forming the polynomial first is a poor strategy. Software instead works on the matrix directly, chiefly through the QR algorithm, published independently by John Francis and Vera Kublanovskaya around 1961. "Iterative and numerical eigenvalue methods" explains how such methods work.
Common mistakes to avoid
First, writing \(\det A - \lambda\) instead of \(\det(A - \lambda I)\). Second, getting the sign of the middle term wrong in the 2 × 2 shortcut: it is \(-(\operatorname{tr} A)\lambda\). Third, computing \(ad + bc\) instead of \(ad - bc\). Fourth, row reducing \(A\) before finding the polynomial, which changes the eigenvalues. Fifth, forgetting repeated roots when checking against the trace and determinant: an eigenvalue of algebraic multiplicity 2 is counted twice.
Key takeaways
The eigenvalues of \(A\) are the solutions of the characteristic equation \(\det(A - \lambda I) = 0\), whose left side is a polynomial of degree \(n\). For 2 × 2 matrices it is \(\lambda^2 - (\operatorname{tr} A)\lambda + \det A\). Algebraic multiplicity counts a repeated root; geometric multiplicity is the dimension of the eigenspace and is never larger. With multiplicity, eigenvalues add up to the trace and multiply to the determinant. Similar matrices share characteristic polynomials and eigenvalues, but equal eigenvalues do not make matrices similar.
Flashcards
Flashcard
det(A − λI) = 0
Test your understanding
Question 1 of 5
What is the characteristic polynomial of the matrix with rows (4, 2) and (1, 3)?
Question 2 of 5
What are the eigenvalues of the matrix with rows (1, 2) and (4, 3)?
Question 3 of 5
A 3 × 3 matrix has characteristic polynomial −(λ − 1)²(λ − 6). Which statement is true?
Question 4 of 5
For the matrix with rows (5, 2) and (0, 5), what are the multiplicities of the eigenvalue 5?
Question 5 of 5
B = P⁻¹AP for some invertible matrix P. Which statement must be true?
