Eigenvalues and eigenvectors, explained visually

An eigenvector of a square matrix A is a nonzero vector that A only stretches, shrinks or flips, never turning it off its own line: Av = λv. The number λ is its eigenvalue, the factor it is scaled by. To find them, solve det(A − λI) = 0 for λ, then solve (A − λI)v = 0 for v.

Av=λvAv = \lambda v

A times v equals lambda times v: applying the matrix A to the eigenvector v gives the same result as multiplying v by the number lambda, its eigenvalue.

Updated

A coordinate grid with two dashed lines through the origin, y = x and y = −2x. Blue arrows v₁ = (1, 1) and v₂ = (−1, 2) lie on them, and green arrows Av₁ = 5v₁, reaching (5, 5), and Av₂ = 2v₂, reaching (−2, 4), lie along the same lines. A blue arrow u = (1, 0) on the x-axis maps to an amber arrow Au = (4, 2), turned off the x-axis, with an amber arc marking the turn.
The matrix A=[4123]A = \begin{bmatrix} 4 & 1 \\ 2 & 3 \end{bmatrix} sends v1=(1,1)v_1 = (1, 1) to (5,5)(5, 5) and v2=(−1,2)v_2 = (-1, 2) to (−2,4)(-2, 4). Each stays on its own dashed line, stretched 5 times and 2 times: they are eigenvectors, with eigenvalues 55 and 22. The vector u=(1,0)u = (1, 0) is turned off the xx-axis to (4,2)(4, 2), so it is not an eigenvector.

Concept map

9 ideas, in the order to learn them. Click one and Zidna draws and narrates an animated lesson on it.

Eigenvalues and Eigenvectors

What eigenvalues and eigenvectors are: the directions a matrix leaves on their own line, and the factors it scales them by. It starts from matrices as motions of the plane and the determinant, finds eigenvalues with the characteristic equation and eigenvectors by solving (A − λI)v = 0, then covers repeated and complex eigenvalues, diagonalization and principal component analysis.

1

Matrices as Linear Transformations

A 2×2 matrix as a rule that moves every vector in the plane, keeping grid lines straight, parallel and evenly spaced.Generate lesson
2

Eigenvectors and Their Eigenvalues

An eigenvector is a nonzero vector that a matrix only scales, Av = λv; its eigenvalue λ is the scale factor.Builds on Matrices as Linear TransformationsGenerate lesson
3

The Determinant

The factor by which a matrix scales area; it is 0 exactly when the matrix flattens the plane onto a line or a point.Builds on Matrices as Linear TransformationsGenerate lesson
4

Finding Eigenvectors

Solving (A − λI)v = 0 for a known eigenvalue λ, to get the line of vectors that the matrix A scales by λ.Builds on Eigenvectors and Their EigenvaluesGenerate lesson
5

Principal Component Analysis

A method that finds the directions in which a data set varies most, as the eigenvectors of its covariance matrix.Builds on Eigenvectors and Their EigenvaluesGenerate lesson
6

The Characteristic Equation

The equation det(A − λI) = 0, whose solutions are exactly the eigenvalues of the matrix A.Builds on Eigenvectors and Their Eigenvalues, The DeterminantGenerate lesson
7

Repeated Eigenvalues

An eigenvalue that is a double root of the characteristic equation, giving either one line of eigenvectors or every vector.Builds on The Characteristic Equation, Finding EigenvectorsGenerate lesson
8

Diagonalization

Writing a matrix as A = PDP⁻¹, with its eigenvectors as the columns of P and its eigenvalues on the diagonal of D.Builds on Finding Eigenvectors, The Characteristic EquationGenerate lesson
9

Complex Eigenvalues

Eigenvalues that are not real numbers, which a real 2×2 matrix has when it turns every direction in the plane.Builds on The Characteristic EquationGenerate lesson

Key ideas

  1. 01

    An eigenvector of a square matrix AA is a nonzero vector vv that AA only scales: Av=λvAv = \lambda v. It stays on its own line through the origin, while most vectors are turned off theirs.

  2. 02

    The eigenvalue λ\lambda is the scale factor: above 11 it stretches the eigenvector, between 00 and 11 it shrinks it, a negative λ\lambda flips it, and λ=0\lambda = 0 sends it to the origin.

  3. 03

    Eigenvalues are the solutions of the characteristic equation det⁡(A−λI)=0\det(A - \lambda I) = 0. For a 2×2 matrix it is a quadratic: λ2−(a+d)λ+(ad−bc)=0\lambda^2 - (a + d)\lambda + (ad - bc) = 0.

  4. 04

    For each eigenvalue, the eigenvectors are the nonzero solutions of (A−λI)v=0(A - \lambda I)v = 0. Any nonzero multiple of an eigenvector is an eigenvector for the same eigenvalue.

  5. 05

    The eigenvalues of a matrix add up to its trace and multiply to its determinant. For [4123]\begin{bmatrix} 4 & 1 \\ 2 & 3 \end{bmatrix} they are 55 and 22: 5+2=4+35 + 2 = 4 + 3 and 5×2=4⋅3−1⋅25 \times 2 = 4 \cdot 3 - 1 \cdot 2.

  6. 06

    A real 2×2 matrix has two different real eigenvalues, one repeated eigenvalue, or a complex pair. A 90° rotation has the complex pair ±i\pm i: it turns every direction, so no real vector keeps its line.

What is an eigenvector, in simple terms?

An eigenvector of a matrix is a direction the matrix does not turn. Multiply an eigenvector by the matrix and the result lies on the same line through the origin, only longer, shorter or pointing the other way. The eigenvalue is the number the eigenvector gets multiplied by.

To see it, treat a 2×2 matrix as a motion of the whole plane. Multiplying by AA sends every vector (x,y)(x, y) to a new vector. The origin stays put, and grid lines stay straight, parallel and evenly spaced, but the grid can be stretched, sheared, rotated, flipped, or even flattened onto a line. The first column of AA is where (1,0)(1, 0) lands, and the second column is where (0,1)(0, 1) lands.

A faint square grid with a slanted purple grid drawn over it. A small blue unit square with corners (0, 0), (1, 0), (1, 1) and (0, 1) maps to a purple parallelogram with corners (0, 0), (4, 2), (5, 5) and (1, 3), labelled area 10. A dashed line y = x passes through both (1, 1) and (5, 5).
Under A=[4123]A = \begin{bmatrix} 4 & 1 \\ 2 & 3 \end{bmatrix}, the unit square becomes the parallelogram with corners (4,2)(4, 2) and (1,3)(1, 3), the columns of AA, and its area grows from 11 to 1010, the determinant of AA. The square's diagonal (1,1)(1, 1) lands on (5,5)(5, 5), still on the dashed line y=xy = x.

Most vectors change direction. Under A=[4123]A = \begin{bmatrix} 4 & 1 \\ 2 & 3 \end{bmatrix}, the vector (1,0)(1, 0) lands on (4,2)(4, 2) and (0,1)(0, 1) lands on (1,3)(1, 3), both turned off their lines. But (1,1)(1, 1) lands on (5,5)(5, 5), which is 5(1,1)5(1, 1): the same line, five times as long. So (1,1)(1, 1) is an eigenvector of AA, with eigenvalue 55.

Av=λvAv = \lambda v

A times v equals lambda times v. Applying the matrix A to the eigenvector v gives the same vector as multiplying v by the number lambda.

  • AA must be a square matrix, so that AvAv and vv have the same size and can be compared.
  • vv must be nonzero. The zero vector satisfies A0=λ0A0 = \lambda 0 for every λ\lambda, so it tells you nothing.
  • λ\lambda can be any number: positive, negative, zero, or even complex.
  • Every nonzero multiple of an eigenvector is also an eigenvector, with the same eigenvalue: (2,2)(2, 2) and (−3,−3)(-3, -3) work as well as (1,1)(1, 1).

An eigenvector really stands for a whole line through the origin: the matrix sends every point of that line to a point on the same line. The eigenvalue says what the matrix does along that line: how much it stretches, and whether it flips.

What do eigenvalues mean?

An eigenvalue tells you what a matrix does along one of its eigenvectors: how much it stretches or shrinks that direction, and whether it flips it. Its size is the stretch factor and its sign says which way the eigenvector ends up pointing.

What an eigenvalue λ does to its eigenvector v
EigenvalueWhat happens to vExample
λ>1\lambda > 1Stretched, same directionλ=5\lambda = 5: (1,1)→(5,5)(1, 1) \to (5, 5)
λ=1\lambda = 1Left exactly where it isA mirror line under a reflection
0<λ<10 < \lambda < 1Shrunk, same directionλ=12\lambda = \tfrac{1}{2}: (4,2)→(2,1)(4, 2) \to (2, 1)
λ=0\lambda = 0Sent to the originThe matrix flattens the plane
λ<0\lambda < 0Flipped to point the other way, then scaled by ∣λ∣|\lambda|λ=−1\lambda = -1: (2,−2)→(−2,2)(2, -2) \to (-2, 2)
Not a real numberNo real eigenvector: every direction is turnedA 90° rotation: λ=±i\lambda = \pm i
A grid with a solid mirror line y = x and a dashed line y = −x. A green arrow to (2, 2) on the mirror is labelled Av₁ = v₁. A blue arrow v₂ to (2, −2) and a green arrow Av₂ = −v₂ to (−2, 2) point in opposite directions along y = −x. A blue arrow u to (−1, −3) and an amber arrow Au to (−3, −1) are mirror images, joined by a dashed amber line.
The reflection [0110]\begin{bmatrix} 0 & 1 \\ 1 & 0 \end{bmatrix} across the line y=xy = x. A vector on the mirror, v1=(2,2)v_1 = (2, 2), stays where it is: eigenvalue 11. A vector across it, v2=(2,−2)v_2 = (2, -2), flips to (−2,2)(-2, 2): eigenvalue −1-1. Any other vector, such as u=(−1,−3)u = (-1, -3), lands on a new line, here at (−3,−1)(-3, -1).

Eigenvalues also describe the whole matrix at once. Their product is the determinant, which measures how the matrix scales area, and their sum is the trace, the sum of the diagonal entries. A matrix with eigenvalues 55 and 22 stretches one direction 5 times and another 2 times, so it scales area by 5×2=105 \times 2 = 10.

Under repeated multiplication, eigenvalues decide what survives. Each time AA is applied, an eigenvector is multiplied by its eigenvalue again, so Anv=λnvA^n v = \lambda^n v. Directions with ∣λ∣>1|\lambda| > 1 grow, directions with ∣λ∣<1|\lambda| < 1 shrink toward the origin, and the direction whose eigenvalue is largest in size usually takes over in the long run.

An eigenvalue of 00 means the matrix sends a whole line of vectors to the origin. That happens exactly when the determinant is 00, so a matrix has an inverse exactly when 00 is not one of its eigenvalues.

How do you find the eigenvalues of a 2x2 matrix?

Solve the characteristic equation det⁡(A−λI)=0\det(A - \lambda I) = 0, where II is the identity matrix. For a 2×2 matrix it is a quadratic equation in λ\lambda, and its solutions are the eigenvalues.

Here is why it works. Subtract λv\lambda v from both sides of Av=λvAv = \lambda v and factor out vv, writing λ\lambda as λI\lambda I so it is a matrix: (A−λI)v=0(A - \lambda I)v = 0. For a nonzero vv to solve this, the matrix A−λIA - \lambda I has to send a nonzero vector to the origin, and a matrix does that exactly when its determinant is 00.

det⁡(A−λI)=0\det(A - \lambda I) = 0

The determinant of A minus lambda times the identity matrix equals zero.

For A=[abcd]A = \begin{bmatrix} a & b \\ c & d \end{bmatrix}, subtracting λI\lambda I takes λ\lambda off each diagonal entry, and the determinant expands to a quadratic built from the trace a+da + d and the determinant ad−bcad - bc:

det⁡[a−λbcd−λ]=λ2−(a+d)λ+(ad−bc)=0\det\begin{bmatrix} a - \lambda & b \\ c & d - \lambda \end{bmatrix} = \lambda^2 - (a + d)\lambda + (ad - bc) = 0

The determinant of A minus lambda I is (a − λ)(d − λ) − bc, which expands to lambda squared, minus (a + d) times lambda, plus (ad − bc). Set it equal to zero.

  1. 1Subtract λ\lambda from each entry on the main diagonal of AA. Leave the other entries alone.
  2. 2Take the determinant: (a−λ)(d−λ)−bc(a - \lambda)(d - \lambda) - bc.
  3. 3Expand and set it equal to 00: λ2−(a+d)λ+(ad−bc)=0\lambda^2 - (a + d)\lambda + (ad - bc) = 0.
  4. 4Solve the quadratic by factoring, or with the quadratic formula when it doesn't factor.
  5. 5Check that the two eigenvalues add up to a+da + d and multiply to ad−bcad - bc.

For A=[4123]A = \begin{bmatrix} 4 & 1 \\ 2 & 3 \end{bmatrix}: (4−λ)(3−λ)−2=λ2−7λ+10=(λ−5)(λ−2)(4 - \lambda)(3 - \lambda) - 2 = \lambda^2 - 7\lambda + 10 = (\lambda - 5)(\lambda - 2), so the eigenvalues are 55 and 22. Check: 5+2=7=4+35 + 2 = 7 = 4 + 3, and 5×2=10=4⋅3−1⋅25 \times 2 = 10 = 4 \cdot 3 - 1 \cdot 2.

The quadratic formula gives both eigenvalues of any 2×2 matrix straight from its trace T=a+dT = a + d and determinant D=ad−bcD = ad - bc:

λ=T±T2−4D2\lambda = \frac{T \pm \sqrt{T^2 - 4D}}{2}

Lambda equals T plus or minus the square root of T squared minus 4D, all over 2, where T is the trace a + d and D is the determinant ad − bc.

Only the diagonal changes in A−λIA - \lambda I. The identity matrix II has 11s on its diagonal and 00s everywhere else, so λI\lambda I has λ\lambda on the diagonal and nothing off it.

How do you find eigenvectors from eigenvalues?

For each eigenvalue λ\lambda, solve (A−λI)v=0(A - \lambda I)v = 0. Because det⁡(A−λI)=0\det(A - \lambda I) = 0, the equations are not independent: for a 2×2 matrix, one row of A−λIA - \lambda I is a multiple of the other, so one equation is enough, and every nonzero solution is an eigenvector.

  1. 1Write A−λIA - \lambda I for the eigenvalue λ\lambda.
  2. 2Take a row with a nonzero entry and write it as an equation in xx and yy.
  3. 3Choose a convenient value for one variable and solve for the other. Whole numbers keep it clean.
  4. 4Check that Av=λvAv = \lambda v.
Eigenvectors of the matrix with rows (4, 1) and (2, 3)
Eigenvalue λA − λIRow 1 saysEigenvector vCheck: Av
55[−112−2]\begin{bmatrix} -1 & 1 \\ 2 & -2 \end{bmatrix}−x+y=0-x + y = 0, so y=xy = x(1,1)(1, 1)(5,5)=5(1,1)(5, 5) = 5(1, 1)
22[2121]\begin{bmatrix} 2 & 1 \\ 2 & 1 \end{bmatrix}2x+y=02x + y = 0, so y=−2xy = -2x(−1,2)(-1, 2)(−2,4)=2(−1,2)(-2, 4) = 2(-1, 2)

A shortcut for a 2×2 matrix [abcd]\begin{bmatrix} a & b \\ c & d \end{bmatrix}: if b≠0b \neq 0, then (b,λ−a)(b, \lambda - a) is an eigenvector for the eigenvalue λ\lambda. For [4123]\begin{bmatrix} 4 & 1 \\ 2 & 3 \end{bmatrix}, where a=4a = 4 and b=1b = 1, λ=5\lambda = 5 gives (1,1)(1, 1) and λ=2\lambda = 2 gives (1,−2)(1, -2), the same line as (−1,2)(-1, 2). If b=0b = 0 but c≠0c \neq 0, use (λ−d,c)(\lambda - d, c) instead.

Eigenvectors are not unique. (1,1)(1, 1), (3,3)(3, 3) and (12,12)(\tfrac{1}{\sqrt{2}}, \tfrac{1}{\sqrt{2}}) all lie on the same eigenvector line, and calculators often scale an eigenvector to length 11. Two answers match if one is a nonzero multiple of the other.

Can eigenvalues be negative, zero, repeated or complex?

Yes. Eigenvalues can be negative, zero, repeated or complex. A negative eigenvalue flips its eigenvector, an eigenvalue of 00 sends its eigenvector to the origin, a repeated eigenvalue is a double root of the characteristic equation, and complex eigenvalues come from a characteristic equation with no real roots. For a 2×2 matrix with trace TT and determinant DD, the sign of T2−4DT^2 - 4D says whether the eigenvalues are two different real numbers, one repeated real number, or a complex pair.

The three cases for a 2×2 matrix with trace T and determinant D
T² − 4DEigenvaluesReal eigenvector linesExample
PositiveTwo different real numbersTwo[4123]\begin{bmatrix} 4 & 1 \\ 2 & 3 \end{bmatrix}: 55 and 22
ZeroOne real number, repeatedOne, or every line if A=λIA = \lambda I[21−14]\begin{bmatrix} 2 & 1 \\ -1 & 4 \end{bmatrix}: 33 twice
NegativeA complex pair p±qip \pm qiNone[3−113]\begin{bmatrix} 3 & -1 \\ 1 & 3 \end{bmatrix}: 3±i3 \pm i
Three small graphs side by side. Two eigenvalues: the parabola λ² − 7λ + 10 crossing the λ-axis at 2 and 5. One, repeated: the parabola λ² − 6λ + 9 touching the λ-axis at 3. No real ones: the parabola λ² − 6λ + 10, whose lowest point is above the λ-axis.
The characteristic polynomials of the three example matrices. λ2−7λ+10\lambda^2 - 7\lambda + 10 crosses the λ\lambda-axis at the eigenvalues 22 and 55; λ2−6λ+9\lambda^2 - 6\lambda + 9 only touches it, at the repeated eigenvalue 33; λ2−6λ+10\lambda^2 - 6\lambda + 10 never reaches it, so its eigenvalues 3±i3 \pm i are complex.

Negative. The reflection across the line y=xy = x has eigenvalue −1-1: it flips every vector on the line y=−xy = -x to point the opposite way. A negative eigenvalue still keeps its eigenvector on its own line.

Zero. The matrix [1224]\begin{bmatrix} 1 & 2 \\ 2 & 4 \end{bmatrix} has determinant 00 and eigenvalues 00 and 55. It sends every multiple of (2,−1)(2, -1) to the origin. An eigenvalue can be 00; an eigenvector can never be the zero vector.

Repeated. The shear [1101]\begin{bmatrix} 1 & 1 \\ 0 & 1 \end{bmatrix} slides each horizontal line of the plane sideways by its height. Its characteristic equation is (1−λ)2=0(1 - \lambda)^2 = 0, so 11 is a repeated eigenvalue, yet only the xx-axis stays on its line. The matrix 2I2I also has a repeated eigenvalue, 22, and every nonzero vector is an eigenvector of it.

Complex. A rotation by 90°, [0−110]\begin{bmatrix} 0 & -1 \\ 1 & 0 \end{bmatrix}, turns every vector, so no real vector keeps its line. Its characteristic equation λ2+1=0\lambda^2 + 1 = 0 has the solutions λ=i\lambda = i and λ=−i\lambda = -i. The complex eigenvalues of a real matrix always come in conjugate pairs like this.

What are the eigenvalues of a diagonal, triangular or symmetric matrix?

For a diagonal or triangular matrix, the eigenvalues are the entries on the main diagonal, so there is nothing to solve. A symmetric matrix of real numbers, one equal to its own transpose, always has real eigenvalues, and its eigenvectors for different eigenvalues are perpendicular.

The diagonal shortcut works because A−λIA - \lambda I is still triangular, and the determinant of a triangular matrix is the product of its diagonal entries. For [2503]\begin{bmatrix} 2 & 5 \\ 0 & 3 \end{bmatrix}, the characteristic equation is (2−λ)(3−λ)=0(2 - \lambda)(3 - \lambda) = 0.

Eigenvalue shortcuts
MatrixEigenvaluesEigenvectors
Diagonal: [300−2]\begin{bmatrix} 3 & 0 \\ 0 & -2 \end{bmatrix}The diagonal entries, 33 and −2-2The axes: (1,0)(1, 0) and (0,1)(0, 1)
Triangular: [2503]\begin{bmatrix} 2 & 5 \\ 0 & 3 \end{bmatrix}The diagonal entries, 22 and 33(1,0)(1, 0) and (5,1)(5, 1)
Symmetric: [1221]\begin{bmatrix} 1 & 2 \\ 2 & 1 \end{bmatrix}Always real: 33 and −1-1Perpendicular: (1,1)(1, 1) and (1,−1)(1, -1)
The identity matrix II11, repeatedEvery nonzero vector
The inverse A−1A^{-1}1λ\tfrac{1}{\lambda} for each eigenvalue λ\lambda of AAThe same as AA's
A power AnA^nλn\lambda^n for each eigenvalue λ\lambda of AAEvery eigenvector of AA, and sometimes more
A shift A+kIA + kIλ+k\lambda + k for each eigenvalue λ\lambda of AAThe same as AA's

Perpendicular means a dot product of zero: (1,1)⋅(1,−1)=1−1=0(1, 1) \cdot (1, -1) = 1 - 1 = 0. The dot product page shows why a zero dot product means a right angle. A 2×2 matrix that is not symmetric can't have two perpendicular eigenvector lines: for [4123]\begin{bmatrix} 4 & 1 \\ 2 & 3 \end{bmatrix}, (1,1)⋅(−1,2)=1(1, 1) \cdot (-1, 2) = 1.

The eigenvalues of any square matrix add up to its trace and multiply to its determinant. It is the fastest way to check an answer.

How do you find eigenvalues of a 3x3 matrix?

Use the same characteristic equation, det⁡(A−λI)=0\det(A - \lambda I) = 0. For a 3×3 matrix the determinant is a cubic in λ\lambda, so there are at most three eigenvalues, and a 3×3 matrix of real numbers always has at least one real one. The work is in expanding the determinant and factoring the cubic.

  1. 1Subtract λ\lambda from the three diagonal entries.
  2. 2Expand det⁡(A−λI)\det(A - \lambda I) along the row or column with the most zeros, so there are fewer terms to multiply out.
  3. 3Factor the cubic. Keep a factor like (2−λ)(2 - \lambda) whole when the expansion hands you one; otherwise try whole numbers that divide the constant term.
  4. 4For each eigenvalue that is not repeated, solve (A−λI)v=0(A - \lambda I)v = 0: the three equations reduce to two, which leaves one variable free to choose.
  5. 5Check that the three eigenvalues add up to the trace and multiply to the determinant.

For [200034049]\begin{bmatrix} 2 & 0 & 0 \\ 0 & 3 & 4 \\ 0 & 4 & 9 \end{bmatrix}, expanding along the first row gives (2−λ)(λ2−12λ+11)=(2−λ)(λ−1)(λ−11)(2 - \lambda)(\lambda^2 - 12\lambda + 11) = (2 - \lambda)(\lambda - 1)(\lambda - 11), so the eigenvalues are 11, 22 and 1111. They add up to 1414, the trace 2+3+92 + 3 + 9, and multiply to 2222, the determinant.

What are eigenvalues and eigenvectors used for?

Eigenvalues and eigenvectors are used wherever a matrix is applied again and again, or a system has natural directions: along an eigenvector, a matrix is plain multiplication by a number. Applying AA to an eigenvector nn times multiplies it by λ\lambda nn times.

Anv=λnvA^n v = \lambda^n v

A to the n, times v, equals lambda to the n, times v: applying A to the eigenvector v n times multiplies it by lambda to the n.

For A=[4123]A = \begin{bmatrix} 4 & 1 \\ 2 & 3 \end{bmatrix}, that means A10(1,1)=510(1,1)=(9,765,625, 9,765,625)A^{10}(1, 1) = 5^{10}(1, 1) = (9{,}765{,}625,\ 9{,}765{,}625), with no matrix multiplication at all.

Where eigenvalues and eigenvectors show up
FieldThe matrixWhat its eigenvectors and eigenvalues give
Statistics and machine learning: principal component analysisA data set's covariance matrixThe eigenvectors are the directions in which the data spreads; each eigenvalue is the variance along its direction
Markov chainsA transition matrix, each column holding one state's probabilities of moving to each stateAn eigenvector with eigenvalue 11, scaled so its entries add up to 11, is a steady state: probabilities one more step leaves unchanged
Web searchThe Google matrix of links between web pagesThe PageRank scores are the entries of its eigenvector with eigenvalue 11
Differential equationsThe matrix AA in x′=Axx' = AxSolutions grow or decay like eλte^{\lambda t} along each eigenvector; the system is stable when every eigenvalue has a negative real part
Engineering: vibrationsThe mass and stiffness matrices of a structureThe eigenvalues give the natural frequencies, and the eigenvectors are the shapes of the vibration modes
Quantum mechanicsThe energy operator, or HamiltonianThe allowed energy levels are its eigenvalues

In principal component analysis, variance is the square of the standard deviation, so the eigenvector of the covariance matrix with the largest eigenvalue is the direction in which the data is most spread out. It is called the first principal component.

Worked examples

Eigenvalues and eigenvectors of a 2×2 matrix

Find the eigenvalues and eigenvectors of A=[4123]A = \begin{bmatrix} 4 & 1 \\ 2 & 3 \end{bmatrix}.

  1. 1Subtract λ\lambda on the diagonal: A−λI=[4−λ123−λ]A - \lambda I = \begin{bmatrix} 4 - \lambda & 1 \\ 2 & 3 - \lambda \end{bmatrix}.
  2. 2Take the determinant: (4−λ)(3−λ)−(1)(2)=λ2−7λ+12−2=λ2−7λ+10(4 - \lambda)(3 - \lambda) - (1)(2) = \lambda^2 - 7\lambda + 12 - 2 = \lambda^2 - 7\lambda + 10.
  3. 3Solve λ2−7λ+10=0\lambda^2 - 7\lambda + 10 = 0. It factors as (λ−5)(λ−2)=0(\lambda - 5)(\lambda - 2) = 0, so λ=5\lambda = 5 or λ=2\lambda = 2.
  4. 4For λ=5\lambda = 5: A−5I=[−112−2]A - 5I = \begin{bmatrix} -1 & 1 \\ 2 & -2 \end{bmatrix}. Row 1 says −x+y=0-x + y = 0, so y=xy = x. Take v1=(1,1)v_1 = (1, 1).
  5. 5For λ=2\lambda = 2: A−2I=[2121]A - 2I = \begin{bmatrix} 2 & 1 \\ 2 & 1 \end{bmatrix}. Row 1 says 2x+y=02x + y = 0, so y=−2xy = -2x. Take x=−1x = -1: v2=(−1,2)v_2 = (-1, 2).
  6. 6Check: A(1,1)=(4+1, 2+3)=(5,5)=5(1,1)A(1, 1) = (4 + 1,\ 2 + 3) = (5, 5) = 5(1, 1), and A(−1,2)=(−4+2, −2+6)=(−2,4)=2(−1,2)A(-1, 2) = (-4 + 2,\ -2 + 6) = (-2, 4) = 2(-1, 2).

Answer: Eigenvalue 55 with eigenvector (1,1)(1, 1), and eigenvalue 22 with eigenvector (−1,2)(-1, 2), or any nonzero multiples of them.

A symmetric matrix with a negative eigenvalue

Find the eigenvalues and eigenvectors of B=[1221]B = \begin{bmatrix} 1 & 2 \\ 2 & 1 \end{bmatrix}, and say what BB does along each eigenvector.

  1. 1The trace is T=1+1=2T = 1 + 1 = 2 and the determinant is D=1⋅1−2⋅2=−3D = 1 \cdot 1 - 2 \cdot 2 = -3, so the characteristic equation is λ2−2λ−3=0\lambda^2 - 2\lambda - 3 = 0.
  2. 2Factor: (λ−3)(λ+1)=0(\lambda - 3)(\lambda + 1) = 0, so λ=3\lambda = 3 or λ=−1\lambda = -1.
  3. 3For λ=3\lambda = 3: B−3I=[−222−2]B - 3I = \begin{bmatrix} -2 & 2 \\ 2 & -2 \end{bmatrix} gives −2x+2y=0-2x + 2y = 0, so y=xy = x and v1=(1,1)v_1 = (1, 1).
  4. 4For λ=−1\lambda = -1: B+I=[2222]B + I = \begin{bmatrix} 2 & 2 \\ 2 & 2 \end{bmatrix} gives 2x+2y=02x + 2y = 0, so y=−xy = -x and v2=(1,−1)v_2 = (1, -1).
  5. 5Check: B(1,−1)=(1−2, 2−1)=(−1,1)=−1⋅(1,−1)B(1, -1) = (1 - 2,\ 2 - 1) = (-1, 1) = -1 \cdot (1, -1). And (1,1)⋅(1,−1)=0(1, 1) \cdot (1, -1) = 0: the eigenvectors are perpendicular, as they must be for a symmetric matrix.

Answer: λ=3\lambda = 3 with (1,1)(1, 1): BB stretches that line 3 times. λ=−1\lambda = -1 with (1,−1)(1, -1): BB flips that line end to end without changing lengths.

A repeated eigenvalue with only one eigenvector line

Find the eigenvalues and eigenvectors of C=[21−14]C = \begin{bmatrix} 2 & 1 \\ -1 & 4 \end{bmatrix}.

  1. 1The trace is T=6T = 6 and the determinant is D=2⋅4−1⋅(−1)=9D = 2 \cdot 4 - 1 \cdot (-1) = 9, so the characteristic equation is λ2−6λ+9=0\lambda^2 - 6\lambda + 9 = 0.
  2. 2It is a perfect square, (λ−3)2=0(\lambda - 3)^2 = 0, so the only eigenvalue is λ=3\lambda = 3, repeated.
  3. 3Solve (C−3I)v=0(C - 3I)v = 0 with C−3I=[−11−11]C - 3I = \begin{bmatrix} -1 & 1 \\ -1 & 1 \end{bmatrix}. Both rows say −x+y=0-x + y = 0, so y=xy = x.
  4. 4Every eigenvector is a multiple of (1,1)(1, 1). Check: C(1,1)=(2+1, −1+4)=(3,3)=3(1,1)C(1, 1) = (2 + 1,\ -1 + 4) = (3, 3) = 3(1, 1).
  5. 5Any vector off that line is turned: C(1,0)=(2,−1)C(1, 0) = (2, -1), which is not a multiple of (1,0)(1, 0).

Answer: One eigenvalue, 33, repeated, with a single line of eigenvectors: the nonzero multiples of (1,1)(1, 1). With only one eigenvector direction, CC cannot be diagonalized.

Complex eigenvalues of a 90° rotation

Find the eigenvalues of R=[0−110]R = \begin{bmatrix} 0 & -1 \\ 1 & 0 \end{bmatrix}, which rotates the plane 90° counterclockwise.

A grid with two dashed lines through the origin. A blue arrow u to (2, 1) is turned 90° to an amber arrow Ru at (−1, 2), and a blue arrow w to (−1, −2) is turned 90° to an amber arrow Rw at (2, −1). Each turn is marked with an amber arc labelled 90°.
RR turns u=(2,1)u = (2, 1) to (−1,2)(-1, 2) and w=(−1,−2)w = (-1, -2) to (2,−1)(2, -1). Every vector is turned a quarter turn off its dashed line, so RR has no real eigenvectors.
  1. 1R−λI=[−λ−11−λ]R - \lambda I = \begin{bmatrix} -\lambda & -1 \\ 1 & -\lambda \end{bmatrix}, with determinant (−λ)(−λ)−(−1)(1)=λ2+1(-\lambda)(-\lambda) - (-1)(1) = \lambda^2 + 1.
  2. 2Solve λ2+1=0\lambda^2 + 1 = 0: λ2=−1\lambda^2 = -1, so λ=i\lambda = i or λ=−i\lambda = -i. There are no real solutions.
  3. 3With no real eigenvalue, no real vector stays on its line, which matches the picture: a quarter turn moves every direction.
  4. 4Over the complex numbers there are eigenvectors. For λ=i\lambda = i, row 1 of R−iIR - iI says −ix−y=0-ix - y = 0, so y=−ixy = -ix and v=(1,−i)v = (1, -i). Check: R(1,−i)=(i,1)=i(1,−i)R(1, -i) = (i, 1) = i(1, -i).

Answer: λ=i\lambda = i and λ=−i\lambda = -i, a complex conjugate pair. RR has no real eigenvectors; its eigenvectors, such as (1,−i)(1, -i) for λ=i\lambda = i, have complex entries.

Eigenvalues and eigenvectors of a 3×3 matrix

Find the eigenvalues and eigenvectors of M=[200034049]M = \begin{bmatrix} 2 & 0 & 0 \\ 0 & 3 & 4 \\ 0 & 4 & 9 \end{bmatrix}.

  1. 1Expand det⁡(M−λI)\det(M - \lambda I) along the first row, which has two zeros: (2−λ)[(3−λ)(9−λ)−16](2 - \lambda)\big[(3 - \lambda)(9 - \lambda) - 16\big].
  2. 2Simplify the bracket: (3−λ)(9−λ)−16=λ2−12λ+11=(λ−1)(λ−11)(3 - \lambda)(9 - \lambda) - 16 = \lambda^2 - 12\lambda + 11 = (\lambda - 1)(\lambda - 11).
  3. 3So (2−λ)(λ−1)(λ−11)=0(2 - \lambda)(\lambda - 1)(\lambda - 11) = 0, and the eigenvalues are 11, 22 and 1111. Check: 1+2+11=14=2+3+91 + 2 + 11 = 14 = 2 + 3 + 9, the trace.
  4. 4For λ=2\lambda = 2: M−2IM - 2I has rows (0,0,0)(0, 0, 0), (0,1,4)(0, 1, 4) and (0,4,7)(0, 4, 7). The last two say y+4z=0y + 4z = 0 and 4y+7z=04y + 7z = 0, so y=z=0y = z = 0 and xx is free: (1,0,0)(1, 0, 0).
  5. 5For λ=1\lambda = 1: M−IM - I has rows (1,0,0)(1, 0, 0), (0,2,4)(0, 2, 4) and (0,4,8)(0, 4, 8). So x=0x = 0 and 2y+4z=02y + 4z = 0, giving y=−2zy = -2z: (0,−2,1)(0, -2, 1).
  6. 6For λ=11\lambda = 11: M−11IM - 11I has rows (−9,0,0)(-9, 0, 0), (0,−8,4)(0, -8, 4) and (0,4,−2)(0, 4, -2). So x=0x = 0 and −8y+4z=0-8y + 4z = 0, giving z=2yz = 2y: (0,1,2)(0, 1, 2).

Answer: Eigenvalue 11 with eigenvector (0,−2,1)(0, -2, 1), eigenvalue 22 with (1,0,0)(1, 0, 0), and eigenvalue 1111 with (0,1,2)(0, 1, 2).

Check yourself

  1. 1.Is (2, 2) an eigenvector of the matrix with rows (4, 1) and (2, 3)? If so, with what eigenvalue?Show answer

    Yes, with eigenvalue 5: the matrix sends (2, 2) to (10, 10), which is 5 × (2, 2).

  2. 2.Is (0, 1) an eigenvector of the matrix with rows (4, 1) and (2, 3)?Show answer

    No. It goes to (1, 3), which is not a multiple of (0, 1).

  3. 3.What are the eigenvalues of the matrix with rows (7, 0) and (0, −3)?Show answer

    7 and −3. A diagonal matrix's eigenvalues are its diagonal entries.

  4. 4.A 2×2 matrix has trace 6 and determinant 8. What are its eigenvalues?Show answer

    4 and 2, the roots of λ² − 6λ + 8 = 0.

  5. 5.What is the characteristic equation of the matrix with rows (2, 5) and (1, −2)?Show answer

    λ² − 9 = 0, since the trace is 0 and the determinant is −4 − 5 = −9. The eigenvalues are 3 and −3.

  6. 6.What are the eigenvalues of the matrix with rows (3, 1) and (1, 1)?Show answer

    2 + √2 and 2 − √2. The characteristic equation λ² − 4λ + 2 = 0 doesn't factor, so the quadratic formula gives them.

  7. 7.5 is an eigenvalue of the matrix with rows (1, 2) and (2, 4). Find an eigenvector for it.Show answer

    (1, 2), or any nonzero multiple: the matrix sends (1, 2) to (5, 10) = 5 × (1, 2).

  8. 8.An eigenvector has eigenvalue −2. What does the matrix do to it?Show answer

    It flips the vector to point the opposite way and doubles its length.

  9. 9.If 3 is an eigenvalue of A, what is the matching eigenvalue of A²?Show answer

    9. If Av = 3v, then A²v = A(3v) = 3Av = 9v.

  10. 10.If 0 is an eigenvalue of a matrix, does the matrix have an inverse?Show answer

    No. It sends a nonzero vector to the zero vector, so its determinant is 0.

Common mistakes

  • Subtracting λ\lambda from every entry: writing A−λIA - \lambda I for [4123]\begin{bmatrix} 4 & 1 \\ 2 & 3 \end{bmatrix} as [4−λ1−λ2−λ3−λ]\begin{bmatrix} 4 - \lambda & 1 - \lambda \\ 2 - \lambda & 3 - \lambda \end{bmatrix}.λI\lambda I has λ\lambda only on the diagonal, so subtract it only from the diagonal: [4−λ123−λ]\begin{bmatrix} 4 - \lambda & 1 \\ 2 & 3 - \lambda \end{bmatrix}.
  • Solving (A−λI)v=0(A - \lambda I)v = 0, getting only v=(0,0)v = (0, 0), and giving that as the eigenvector.The zero vector is never an eigenvector. For a true eigenvalue the equations always have nonzero solutions, so if elimination leaves only (0,0)(0, 0), recheck the eigenvalue or the subtraction.
  • Reading the eigenvalues off the diagonal of any matrix: saying [4123]\begin{bmatrix} 4 & 1 \\ 2 & 3 \end{bmatrix} has eigenvalues 44 and 33.That shortcut works only for diagonal and triangular matrices. Here the off-diagonal entries matter, and the eigenvalues are 55 and 22; they do still add up to 4+3=74 + 3 = 7.
  • Row-reducing AA first, then finding the eigenvalues of the simpler matrix.Row operations change the eigenvalues: [4123]\begin{bmatrix} 4 & 1 \\ 2 & 3 \end{bmatrix} row-reduces to II, whose eigenvalues are 11 and 11, not 55 and 22. Row-reduce A−λIA - \lambda I, and only when solving for eigenvectors.
  • Marking an eigenvector wrong because it differs from the answer key, such as (2,2)(2, 2) instead of (1,1)(1, 1), or (1,−2)(1, -2) instead of (−1,2)(-1, 2).Any nonzero multiple of an eigenvector is an eigenvector for the same eigenvalue. Two answers agree when one is a multiple of the other.
  • Slipping a sign when expanding (a−λ)(d−λ)−bc(a - \lambda)(d - \lambda) - bc, such as adding bcbc instead of subtracting it.Match the pattern λ2−(a+d)λ+(ad−bc)=0\lambda^2 - (a + d)\lambda + (ad - bc) = 0, then check that the roots add up to the trace a+da + d and multiply to the determinant ad−bcad - bc.

Questions students ask

Why are they called eigenvalues and eigenvectors?

Eigen is German for own, in the sense of characteristic or proper: an eigenvector is one of a matrix's own directions. David Hilbert used the word this way in 1904, and the name stuck. Eigenvalues are also called characteristic values or characteristic roots, and the equation that finds them is the characteristic equation.

Can an eigenvector be the zero vector?

No. The zero vector satisfies A0 = λ0 for every number λ, so allowing it would make every number an eigenvalue of every matrix. Eigenvectors are nonzero by definition. An eigenvalue can be 0, though: it means the matrix sends some nonzero vector to the zero vector.

Are eigenvectors always perpendicular?

Not in general. They are guaranteed to be perpendicular for a symmetric matrix, where eigenvectors for different eigenvalues always meet at right angles. The matrix with rows (4, 1) and (2, 3) has eigenvectors (1, 1) and (−1, 2), whose dot product is 1, not 0, so they meet at about 71.6°.

Do only square matrices have eigenvalues?

Yes. The equation Av = λv compares Av with v, so the two must be the same size, and that needs A to have as many rows as columns. Rectangular matrices have singular values instead, which describe how much they stretch along their most important directions.

How many eigenvalues does a matrix have?

An n×n matrix has exactly n eigenvalues if you count repeated ones and allow complex numbers, because its characteristic equation is a polynomial of degree n. It has at most n different eigenvalues. So a 2×2 matrix has two and a 3×3 matrix has three, counted this way.

Why is the determinant the product of the eigenvalues?

The characteristic polynomial det(A − λI) factors as (λ₁ − λ)(λ₂ − λ)···(λₙ − λ), one factor for each eigenvalue. Setting λ = 0 gives det A = λ₁λ₂···λₙ. In pictures, a matrix that stretches one direction 5 times and another 2 times scales every area by 10.

What does an eigenvalue of 1 mean?

Vectors along its eigenvector are left exactly where they are: Av = v. A reflection has eigenvalue 1 along its mirror line. The transition matrix of a Markov chain, with each column holding one state's probabilities of moving to each state, always has eigenvalue 1. An eigenvector for it with no negative entries, scaled so its entries add up to 1, is a steady state: probabilities that one more step leaves unchanged.

Can two different matrices have the same eigenvalues?

Yes. The matrix with rows (4, 1) and (2, 3) and the diagonal matrix with rows (5, 0) and (0, 2) both have eigenvalues 5 and 2; they stretch by the same amounts along different directions. Eigenvalues alone don't pin down a matrix: the identity matrix and the shear with rows (1, 1) and (0, 1) both have eigenvalue 1 twice, yet the identity moves nothing.

Sources