1.15 · AHL · HL onlyPaper 1Paper 2Paper 3

Eigenvalues and eigenvectors

In a minute

An of a is a direction the matrix only stretches: Mv=λvM\boldsymbol{v} = \lambda\boldsymbol{v}, where the number λ\lambda is its . When a 2×22 \times 2 matrix has two different real eigenvalues, they turn any power MnM^{n} into a formula in nn. That is how you predict where a population moving between two towns ends up, years ahead.

You'll be able to

  • Test whether a vector is an eigenvector of a matrix, and find its eigenvalue.
  • Find the of a 2×22 \times 2 matrix and solve it for the eigenvalues, by hand or with a GDC.
  • Find an eigenvector for each eigenvalue of a 2×22 \times 2 matrix.
  • Diagonalize a 2×22 \times 2 matrix with two different real eigenvalues as PDP−1PDP^{-1}.
  • Use Mn=PDnP−1M^{n} = PD^{n}P^{-1} to find powers and long-term behaviour, and interpret them in context.
Before you start3 quick questions on what this builds on
Check yourself
Find det⁡(3514)\det\begin{pmatrix}3 & 5\\1 & 4\end{pmatrix}.
Check yourself
Find (2134)(12)\begin{pmatrix}2 & 1\\3 & 4\end{pmatrix}\begin{pmatrix}1\\2\end{pmatrix}.
Check yourself
Solve x2−1.3x+0.3=0x^{2} - 1.3x + 0.3 = 0.

Revise first: 1.8 Solving equations with technology and 1.14 Matrices, determinants and inverses

Eigenvalues and eigenvectors

In 1.14 you multiplied a matrix by a column vector. Usually the result points in a new direction. For a few special vectors it stays on the same line: the matrix only stretches them.

Take A=(2112)A = \begin{pmatrix}2 & 1\\1 & 2\end{pmatrix}. Then A(10)=(21)A\begin{pmatrix}1\\0\end{pmatrix} = \begin{pmatrix}2\\1\end{pmatrix}, which has turned. But A(11)=(33)=3(11)A\begin{pmatrix}1\\1\end{pmatrix} = \begin{pmatrix}3\\3\end{pmatrix} = 3\begin{pmatrix}1\\1\end{pmatrix}: the same direction, three times as long.

x y 1 1 2 2 3 3 y = x Av = (3, 3) v = (1, 1) Aw = (2, 1) w = (1, 0)
For A=(2112)A = \begin{pmatrix}2 & 1\\1 & 2\end{pmatrix}, v\boldsymbol{v} keeps its direction and is stretched by 33. w\boldsymbol{w} is turned.

So (11)\begin{pmatrix}1\\ 1\end{pmatrix} is an eigenvector of AA with eigenvalue 33. Two facts follow from Av=λvA\boldsymbol{v} = \lambda\boldsymbol{v}.

1. Any non-zero multiple of an eigenvector is also an eigenvector, with the same eigenvalue. (22)\begin{pmatrix}2\\ 2\end{pmatrix} and (−5−5)\begin{pmatrix}-5\\ -5\end{pmatrix} work just as well as (11)\begin{pmatrix}1\\ 1\end{pmatrix}.

2. The zero vector never counts, even though A0=λ0A\boldsymbol{0} = \lambda\boldsymbol{0} for every λ\lambda. An eigenvalue, though, can be 00 or negative. A negative eigenvalue flips the vector to point the opposite way.

To test a vector, work out AvA\boldsymbol{v} and see whether it is a multiple of v\boldsymbol{v}.

Worked exampleLet A=(4−211)A = \begin{pmatrix}4 & -2\\1 & 1\end{pmatrix}. Show that v=(21)\boldsymbol{v} = \begin{pmatrix}2\\1\end{pmatrix} is an eigenvector of AA, and find its eigenvalue. Is w=(12)\boldsymbol{w} = \begin{pmatrix}1\\2\end{pmatrix} an eigenvector of AA?

Try it first, then reveal one step at a time.

  1. Multiply: Av=(4(2)−2(1)1(2)+1(1))=(63)A\boldsymbol{v} = \begin{pmatrix}4(2) - 2(1)\\ 1(2) + 1(1)\end{pmatrix} = \begin{pmatrix}6\\3\end{pmatrix}.
Worked exampleEach year 6%6\% of the people living in Northby move to Southam, and 3%3\% of Southam's people move to Northby. No one else arrives or leaves. Let xnx_n and yny_n be the populations, in thousands, after nn years. Then (xn+1yn+1)=M(xnyn)\begin{pmatrix}x_{n+1}\\ y_{n+1}\end{pmatrix} = M\begin{pmatrix}x_n\\ y_n\end{pmatrix}, where M=(0.940.030.060.97)M = \begin{pmatrix}0.94 & 0.03\\0.06 & 0.97\end{pmatrix}. Show that (12)\begin{pmatrix}1\\2\end{pmatrix} is an eigenvector of MM, find its eigenvalue, and interpret it.

Try it first, then reveal one step at a time.

  1. Read the matrix by columns: column 1 says where Northby's people go (94%94\% stay, 6%6\% move). Column 2 does the same for Southam's.
Check yourself
Which vector is an eigenvector of (3102)\begin{pmatrix}3 & 1\\0 & 2\end{pmatrix}?
Check yourself
(32)\begin{pmatrix}3\\2\end{pmatrix} is an eigenvector of (7−322)\begin{pmatrix}7 & -3\\2 & 2\end{pmatrix}. Find its eigenvalue.

The characteristic polynomial

Guessing eigenvectors is slow. Find the eigenvalues first. Write II for the 2×22 \times 2 , so that λv=λIv\lambda\boldsymbol{v} = \lambda I\boldsymbol{v}.

Then Av=λvA\boldsymbol{v} = \lambda\boldsymbol{v} rearranges to (A−λI)v=0(A - \lambda I)\boldsymbol{v} = \boldsymbol{0}. If A−λIA - \lambda I had an inverse, multiplying by it would give v=0\boldsymbol{v} = \boldsymbol{0}. An eigenvector is never 0\boldsymbol{0}, so A−λIA - \lambda I has no inverse: its determinant is 00.

Subtract λ\lambda from each entry on the leading diagonal, then use ad−bcad - bc:

Learn this
det⁡(a−λbcd−λ)=(a−λ)(d−λ)−bc=λ2−(a+d)λ+(ad−bc)\det\begin{pmatrix}a - \lambda & b\\ c & d - \lambda\end{pmatrix} = (a - \lambda)(d - \lambda) - bc = \lambda^{2} - (a + d)\lambda + (ad - bc)

This is not in the formula booklet, but it is quick to rebuild. A 2×22 \times 2 matrix has at most two eigenvalues. You are only expected to find eigenvalues, by hand or with technology, for 2×22 \times 2 matrices.

Worked exampleFind the characteristic polynomial of A=(6−221)A = \begin{pmatrix}6 & -2\\2 & 1\end{pmatrix}. Hence find the eigenvalues of AA.

Try it first, then reveal one step at a time.

  1. Subtract λ\lambda on the leading diagonal: A−λI=(6−λ−221−λ)A - \lambda I = \begin{pmatrix}6 - \lambda & -2\\ 2 & 1 - \lambda\end{pmatrix}.
Worked exampleA biologist splits a bird population into jnj_n juveniles and ana_n adults, in hundreds, after nn years. Each year an adult raises 1.61.6 juveniles on average, 30%30\% of juveniles become adults, and 70%70\% of adults survive. So (jn+1an+1)=(01.60.30.7)(jnan)\begin{pmatrix}j_{n+1}\\ a_{n+1}\end{pmatrix} = \begin{pmatrix}0 & 1.6\\ 0.3 & 0.7\end{pmatrix}\begin{pmatrix}j_n\\ a_n\end{pmatrix}. Use your GDC to find the eigenvalues of the matrix, to 3 s.f. Interpret the positive eigenvalue.

Try it first, then reveal one step at a time.

  1. Write down the equation you solve: det⁡(−λ1.60.30.7−λ)=0\det\begin{pmatrix}-\lambda & 1.6\\ 0.3 & 0.7 - \lambda\end{pmatrix} = 0.
Check yourself
Find the eigenvalues of (7−430)\begin{pmatrix}7 & -4\\3 & 0\end{pmatrix}.
Check yourself
Use your GDC to find the larger eigenvalue of (0.50.80.60.9)\begin{pmatrix}0.5 & 0.8\\0.6 & 0.9\end{pmatrix}. Give your answer to 3 s.f.

Finding the eigenvectors

Once you have an eigenvalue λ\lambda, find an eigenvector by solving (A−λI)v=0(A - \lambda I)\boldsymbol{v} = \boldsymbol{0} with v=(xy)\boldsymbol{v} = \begin{pmatrix}x\\ y\end{pmatrix}.

1. Write out A−λIA - \lambda I with the value of λ\lambda put in.

2. Write the two equations from (A−λI)(xy)=(00)(A - \lambda I)\begin{pmatrix}x\\ y\end{pmatrix} = \begin{pmatrix}0\\ 0\end{pmatrix}. Because det⁡(A−λI)=0\det(A - \lambda I) = 0, they say the same thing: one is a multiple of the other.

3. Use an equation that isn't 0=00 = 0 to write yy in terms of xx, or xx in terms of yy. Give the free one a simple non-zero value, such as x=1x = 1, or whatever makes both entries whole numbers.

4. Check that Av=λvA\boldsymbol{v} = \lambda\boldsymbol{v}.

Worked exampleA=(6−221)A = \begin{pmatrix}6 & -2\\2 & 1\end{pmatrix} has eigenvalues 22 and 55. Find an eigenvector for each eigenvalue.

Try it first, then reveal one step at a time.

  1. λ=2\lambda = 2: A−2I=(4−22−1)A - 2I = \begin{pmatrix}4 & -2\\ 2 & -1\end{pmatrix}, giving 4x−2y=04x - 2y = 0 and 2x−y=02x - y = 0.
Worked exampleA bike-hire firm keeps 250250 bikes at two stations, A and B. Each day 20%20\% of the bikes at A end the day at B, and 30%30\% of those at B end at A. So the numbers at A and B are multiplied by M=(0.80.30.20.7)M = \begin{pmatrix}0.8 & 0.3\\0.2 & 0.7\end{pmatrix} each day. One eigenvalue of MM is 11. Find an eigenvector for it. Hence find how many bikes should be at each station so that the numbers stay the same from day to day.

Try it first, then reveal one step at a time.

  1. M−I=(−0.20.30.2−0.3)M - I = \begin{pmatrix}-0.2 & 0.3\\ 0.2 & -0.3\end{pmatrix}, giving −0.2x+0.3y=0-0.2x + 0.3y = 0. The second equation is the same multiplied by −1-1.

You meet this balanced state again in 4.19, as the steady state of a Markov chain.

Check yourself
(1322)\begin{pmatrix}1 & 3\\2 & 2\end{pmatrix} has eigenvalues 44 and −1-1. Which is an eigenvector for λ=4\lambda = 4?
Check yourself
A gym's members use morning or evening sessions. Each month 15%15\% of morning members switch to evenings and 45%45\% of evening members switch to mornings, so M=(0.850.450.150.55)M = \begin{pmatrix}0.85 & 0.45\\0.15 & 0.55\end{pmatrix}. An eigenvector for the eigenvalue 11 is (k1)\begin{pmatrix}k\\ 1\end{pmatrix}. Find kk.

Diagonalizing a 2 × 2 matrix

A diagonal matrix has 00s everywhere off the leading diagonal. Diagonal matrices are easy to work with: to multiply two of them, or to raise one to a power, you just work with the diagonal entries.

To diagonalize AA, put its eigenvectors side by side as the columns of a matrix PP. Put the matching eigenvalues on the diagonal of DD, in the same order:

Learn this
P=(v1v2),D=(λ100λ2),A=PDP−1P = \begin{pmatrix}\boldsymbol{v}_{1} & \boldsymbol{v}_{2}\end{pmatrix}, \quad D = \begin{pmatrix}\lambda_{1} & 0\\ 0 & \lambda_{2}\end{pmatrix}, \quad A = PDP^{-1}

You only diagonalize 2×22 \times 2 matrices with two different real eigenvalues. PP is not unique: you may multiply either eigenvector by a non-zero number, or swap the columns. If you swap the columns of PP, swap the eigenvalues in DD too.

Worked exampleDiagonalize A=(6−221)A = \begin{pmatrix}6 & -2\\2 & 1\end{pmatrix}, whose eigenvalues and eigenvectors you found above. Check that PDP−1=APDP^{-1} = A.

Try it first, then reveal one step at a time.

  1. Eigenvalue 22 goes with (12)\begin{pmatrix}1\\2\end{pmatrix}, and eigenvalue 55 with (21)\begin{pmatrix}2\\1\end{pmatrix}.
Worked exampleFor the towns model M=(0.940.030.060.97)M = \begin{pmatrix}0.94 & 0.03\\0.06 & 0.97\end{pmatrix}, find both eigenvalues and an eigenvector for each. Hence write M=PDP−1M = PDP^{-1}, and interpret the second eigenvalue.

Try it first, then reveal one step at a time.

  1. a+d=1.91a + d = 1.91 and ad−bc=0.9118−0.0018=0.91ad - bc = 0.9118 - 0.0018 = 0.91, so the characteristic equation is λ2−1.91λ+0.91=0\lambda^{2} - 1.91\lambda + 0.91 = 0.
Check yourself
A=(2130)A = \begin{pmatrix}2 & 1\\3 & 0\end{pmatrix} has eigenvalue 33 with eigenvector (11)\begin{pmatrix}1\\1\end{pmatrix}, and eigenvalue −1-1 with eigenvector (1−3)\begin{pmatrix}1\\-3\end{pmatrix}. You choose P=(11−31)P = \begin{pmatrix}1 & 1\\-3 & 1\end{pmatrix}. Which DD gives A=PDP−1A = PDP^{-1}?
Check yourself
A=(5412)A = \begin{pmatrix}5 & 4\\1 & 2\end{pmatrix} is diagonalized as PDP−1PDP^{-1}, with (41)\begin{pmatrix}4\\1\end{pmatrix} as the first column of PP. Find the top-left entry of DD.

Powers of a matrix

In a model xn+1=Mxn\boldsymbol{x}_{n+1} = M\boldsymbol{x}_n, the state after nn steps is xn=Mnx0\boldsymbol{x}_n = M^{n}\boldsymbol{x}_0. A GDC finds M10M^{10} for you. To see what happens for every nn, or as nn grows, you need a formula for MnM^{n}.

Diagonalizing gives one. In M2=PDP−1PDP−1M^{2} = PDP^{-1}PDP^{-1}, the middle P−1PP^{-1}P is II, so M2=PD2P−1M^{2} = PD^{2}P^{-1}. The same happens for every power:

In the formula booklet
Mn=PDnP−1M^{n} = PD^{n}P^{-1}

This is in the formula booklet, with PP the matrix of eigenvectors and DD the diagonal matrix of eigenvalues. DnD^{n} is easy: raise each diagonal entry to the power nn. Then multiply out the three matrices.

Learn this
Dn=(λ1 n00λ2 n)D^{n} = \begin{pmatrix}\lambda_{1}^{\,n} & 0\\ 0 & \lambda_{2}^{\,n}\end{pmatrix}

Long-term behaviour comes from the eigenvalues. As n→∞n \to \infty, λn→0\lambda^{n} \to 0 if ∣λ∣<1\lvert\lambda\rvert < 1. If λ=1\lambda = 1, then λn=1\lambda^{n} = 1 for every nn. If ∣λ∣>1\lvert\lambda\rvert > 1, λn\lambda^{n} grows in size without limit, changing sign each step when λ<−1\lambda < -1.

Worked exampleFor Northby and Southam, M=PDP−1M = PDP^{-1} with P=(112−1)P = \begin{pmatrix}1 & 1\\2 & -1\end{pmatrix} and D=(1000.91)D = \begin{pmatrix}1 & 0\\0 & 0.91\end{pmatrix}. Find MnM^{n} in terms of nn. The towns start with 5050 thousand and 4040 thousand people. Find the populations after nn years, and describe the long-term behaviour.

Try it first, then reveal one step at a time.

  1. P−1=1−3(−1−1−21)=13(112−1)P^{-1} = \frac{1}{-3}\begin{pmatrix}-1 & -1\\-2 & 1\end{pmatrix} = \frac{1}{3}\begin{pmatrix}1 & 1\\2 & -1\end{pmatrix}.
0 10 20 30 40 0 10 20 30 40 50 60 70 years, n population (thousands) Northby Southam
Northby, 30+20(0.91)n30 + 20(0.91)^{n}, and Southam, 60−20(0.91)n60 - 20(0.91)^{n}, settle at the 1:21 : 2 split.
Worked exampleOn an island there are fnf_n foxes and rnr_n rabbits, in hundreds, after nn years. A model is (fn+1rn+1)=M(fnrn)\begin{pmatrix}f_{n+1}\\ r_{n+1}\end{pmatrix} = M\begin{pmatrix}f_n\\ r_n\end{pmatrix} with M=(0.60.5−0.21.3)M = \begin{pmatrix}0.6 & 0.5\\-0.2 & 1.3\end{pmatrix}. At the start there are 300300 foxes and 600600 rabbits. Use Mn=PDnP−1M^{n} = PD^{n}P^{-1} to find a formula for fnf_n. Find the number of foxes after 1010 years, in hundreds to 3 s.f., and describe the long-term behaviour.

Try it first, then reveal one step at a time.

  1. Read the model: with no rabbits, the foxes fall to 60%60\% each year. With no foxes, the rabbits grow by 30%30\%. The −0.2-0.2 is rabbits eaten by foxes.
Check yourself
D=(0.5001)D = \begin{pmatrix}0.5 & 0\\0 & 1\end{pmatrix}. Find D3D^{3}.
Check yourself
Deer move between two parks, X and Y. Each year the numbers in X and Y are multiplied by M=(0.760.160.240.84)M = \begin{pmatrix}0.76 & 0.16\\0.24 & 0.84\end{pmatrix}. The eigenvalues are 11 and 0.60.6, with eigenvectors (23)\begin{pmatrix}2\\3\end{pmatrix} and (1−1)\begin{pmatrix}1\\-1\end{pmatrix}. There are 400400 deer in total. In the long term, how many deer are in park X?

Common mistakes

  • Putting the eigenvalues in DD in a different order from their eigenvectors in PP. Column kk of PP must be the eigenvector of the kkth diagonal entry of DD.
  • Writing Mn=P−1DnPM^{n} = P^{-1}D^{n}P. With eigenvectors as the columns of PP, it is PDnP−1PD^{n}P^{-1}: PP first, P−1P^{-1} last.
  • Giving the zero vector, or no vector, as an eigenvector. Choose a non-zero solution of (A−λI)v=0(A - \lambda I)\boldsymbol{v} = \boldsymbol{0} and check that Av=λvA\boldsymbol{v} = \lambda\boldsymbol{v}.
  • Rounding an eigenvalue before raising it to a power. Small errors grow when the power is large, so keep the GDC's full value.
  • Stopping at MnM^{n} or the eigenvalues when the question asks about the context. Say what happens to the quantities, with units, and comment on whether the model is realistic.

Exam tips

  • Describe the long-term behaviour: say which terms tend to 00 because ∣λ∣<1\lvert\lambda\rvert < 1, give the limiting values in context with units, and comment on whether the model stays realistic.
  • Find an eigenvector: any non-zero multiple of a correct eigenvector earns the marks, so choose whole numbers. Show the equation you get from (M−λI)v=0(M - \lambda I)\boldsymbol{v} = \boldsymbol{0}.
  • Write down DD: put the eigenvalues in the same order as their eigenvectors in the columns of PP. If PP is given, first read which eigenvector is in each column.
  • Show that v\boldsymbol{v} is an eigenvector: work out MvM\boldsymbol{v} and write it as λv\lambda\boldsymbol{v}. Work forwards from MM and v\boldsymbol{v}; don't start from the answer.

Key takeaways

Learn thisFor a square matrix AA, λ\lambda is an eigenvalue with eigenvector v\boldsymbol{v} when Av=λvA\boldsymbol{v} = \lambda\boldsymbol{v} and v≠0\boldsymbol{v} \neq \boldsymbol{0}. Any non-zero multiple of v\boldsymbol{v} is also an eigenvector for λ\lambda.
Learn thisdet⁡(A−λI)=λ2−(a+d)λ+(ad−bc)for A=(abcd); its roots are the eigenvalues\det(A - \lambda I) = \lambda^{2} - (a + d)\lambda + (ad - bc) \quad \text{for } A = \begin{pmatrix}a & b\\ c & d\end{pmatrix}\text{; its roots are the eigenvalues}
Learn thisTo find an eigenvector for an eigenvalue λ\lambda of a 2×22 \times 2 matrix AA, solve (A−λI)v=0(A - \lambda I)\boldsymbol{v} = \boldsymbol{0}. The two equations are multiples of each other: pick any non-zero solution.
Learn thisA 2×22 \times 2 matrix with two different real eigenvalues can be written A=PDP−1A = PDP^{-1}: eigenvectors as the columns of PP, and the matching eigenvalues, in the same order, on the diagonal of DD.
In bookletM=PDP−1⇒Mn=PDnP−1,Dn=(λ1 n00λ2 n)M = PDP^{-1} \Rightarrow M^{n} = PD^{n}P^{-1}, \quad D^{n} = \begin{pmatrix}\lambda_{1}^{\,n} & 0\\ 0 & \lambda_{2}^{\,n}\end{pmatrix}
Learn thisIn xn=Mnx0\boldsymbol{x}_n = M^{n}\boldsymbol{x}_0 with M=PDP−1M = PDP^{-1}, the part of x0\boldsymbol{x}_0 along each eigenvector is multiplied by its eigenvalue at every step. As n→∞n \to \infty that part dies away if ∣λ∣<1\lvert\lambda\rvert < 1 and stays the same if λ=1\lambda = 1. If it is not zero and ∣λ∣>1\lvert\lambda\rvert > 1, it grows without limit.

Practise this

26 past-paper questions on Eigenvalues and eigenvectors in the AI HL bank, each with a mark scheme and a worked explanation.

Practise Eigenvalues and eigenvectors →