Given a square matrix A, let us pose the following question: Does there exist a nonzero vector v such that the result Av of multiplying v by the matrix A is simply a scalar multiple of v? Thus we ask whether or not there exist a nonzero vector v and a scalar λ
Section 6.3 and subsequent chapters include interesting applications in which this question arises. The following definition provides the terminology that we use in discussing Eq. (1).
The prefix eigen is a German word that (in some contexts) may be translated as proper, so eigenvalues are sometimes called proper values. Eigenvalues and eigenvectors are also called characteristic values and characteristic vectors in some books.
Consider the 2×2
If v=(2,1)=[21]T,
Thus, v=[21]T
so v=[32]T
Eigenvalues and eigenvectors have a simple geometric interpretation. Suppose that λ
If v=0,
Let λ
so u=kv
We now attack the problem of finding the eigenvalues and eigenvectors of a given n×n
that is, when
For a fixed value of λ
of its coefficient matrix is zero. The equation |A−λI|=0
Now let us see just what sort of equation the characteristic equation in (3) is. Note that the matrix
is obtained simply by subtracting λ
an nth-degree polynomial equation in λ
According to the fundamental theorem of algebra, every nth-degree polynomial equation in one variable has n solutions (counting multiple solutions), but some of them can be complex. Hence, we can say that an n×n
Solving the characteristic equation is almost always easier said than done. In the examples that follow and in the problems, we have chosen matrices for which the characteristic polynomial |A−λI|
Find the eigenvalues and associated eigenvectors of the matrix
Here
so the characteristic equation of A is
that is, (λ+2)(λ−3)=0
Case 1: λ1=−2
Each of the two scalar equations here is a multiple of the equation x+y=0
Case 2: λ2=3
Again, we have only a single equation, 2x+7y=0
Finally, note that it is not enough to say simply that the given matrix A has eigenvalues −2
If the elements of the matrix A are all real, then so are the coefficients of its characteristic equation |A−λI|=0.
The characteristic equation of the matrix
is
Hence the matrix A has the complex conjugate eigenvalues λ=±4i
Case 1: λ1=−4i
The first equation here, 4ix+8y=0
Case 2: λ2=+4i
The 2×2
so I has the single eigenvalue λ=1
so every nonzero vector v=[xy]T
The characteristic equation of the matrix
is (2−λ)2=0
Thus x is arbitrary, but y=0, so the eigenvalue λ=2 corresponds (to within a constant multiple) to the single eigenvector v=[10]T.
Examples 2–5 illustrate the four possibilities for a 2×2 matrix A. It can have either
two distinct real eigenvalues, each corresponding to a single eigenvector;
a single real eigenvalue corresponding to a single eigenvector;
a single real eigenvalue corresponding to two linearly independent eigenvectors; or
two complex conjugate eigenvalues corresponding to complex conjugate eigenvectors.
The characteristic equation of a 3×3 matrix is, by Eq. (5), of the form
Now the value of the continuous function −λ3+⋯+c0 on the left-hand side here approaches +∞ as λ→−∞ and −∞ as λ→+∞. Hence every such cubic equation has at least one real solution, so every 3×3 matrix has (in contrast with 2×2 matrices) at least one real eigenvalue. A 3×3 matrix may have one, two, or three distinct eigenvalues, and a single eigenvalue of a 3×3 matrix may correspond to one, two, or three linearly independent eigenvectors. The remaining two examples of this section illustrate some of the possibilities. The next example shows also that, whereas the zero vector 0 cannot be an eigenvector of a matrix, there is nothing to prevent λ=0 from being an eigenvalue.
Find the eigenvalues and associated eigenvectors of the matrix
The matrix A−λI is
Upon expansion of the determinant along its first row, we find that
Hence the characteristic equation |A−λI|=0 yields the three eigenvalues λ1=0, λ2=1, and λ3=3. To find the associated eigenvectors, we must solve the system (A−λI)v=0 separately for each of these three eigenvalues.
Case 1: λ1=0. We write v=[xyz]T and substitute λ=0 in the coefficient matrix in (8) to obtain the system
From the first of the three equations here, 3x=0, we see that x=0. Then each of the remaining two equations is a multiple of the equation 3y+z=0. The choice y=1 yields z=−3. Thus the eigenvector v1=[01−3]T is associated with λ1=0.
Case 2: λ2=1. Substitution of λ=1 in the coefficient matrix in (8) yields the system
for v=[xyz]T. The first equation 2x=0 implies that x=0. Then the third equation is a multiple of the second equation, 5y+2z=0. The choice of y=2 yields z=−5, so the eigenvector v2=[02−5]T is associated with λ2=1.
Case 3: λ3=3. Substitution of λ=3 in the coefficient matrix in (8) yields the system
In this case, the first equation yields no information, but the result of adding 4 times the second equation to the third equation is −3y=0, so y=0. Consequently, the second and third equations are both multiples of the equation 2x−z=0. The choice x=1 yields z=2, so the eigenvector v3=[102]T is associated with λ3=3.
In summary, we have found the eigenvectors v1=[01−3]T, v2=[02−5]T, and v3=[102]T associated with the distinct eigenvalues λ1=0, λ2=1, and λ3=3, respectively.
Substitution of λ=0 in the characteristic equation |A−λI|=0 yields |A|=0. Therefore, λ=0 is an eigenvalue of the matrix A if and only if A is singular: |A|=0.
Let λ be a fixed eigenvalue of the n×n matrix A. Then the set of all eigenvectors associated with A is the set of all nonzero solution vectors of the system
The solution space of this system is called the eigenspace of A associated with the eigenvalue λ. This subspace of Rn consists of all eigenvectors associated with λ together with the zero vector. In Example 6 we found (to within a constant multiple) only a single eigenvector associated with each eigenvalue λ; in this case, the eigenspace of λ is 1-dimensional. In the case of an eigenspace of higher dimension, we generally want to find a basis for the solution space of Eq. (9).
Find bases for the eigenspaces of the matrix
Here we have
We expand along the first row to obtain
Thus, to find the eigenvalues we need to solve the cubic equation
We look (hopefully) for integer solutions. The factor theorem of algebra implies that if the polynomial equation
with integer coefficients and leading coefficient 1 has an integer solution, then that integer is a divisor of the constant c0. In the case of the cubic equation in (11), the possibilities for such a solution are ±1, ±2, ±3, ±4, ±6, and ±12. We substitute these numbers successively in (11) and thereby find that +1 and −1 are not solutions but that λ=+2 is a solution. Hence λ−2 is a factor of the cubic polynomial in (11). Next, the long division
shows that
Thus we see finally that the given matrix A has the repeated eigenvalue λ=2 and the eigenvalue λ=3.
Case 1: λ=2. The system (A−λI)v=0 is
which reduces to the single equation 2x−2y+z=0. This equation obviously has a 2-dimensional solution space. With y=1 and z=0, we get x=1 and, hence, obtain the basis eigenvector v1=[110]T. With y=0 and z=2, we get x=−1 and, hence, the basis eigenvector v2=[−102]T. The 2-dimensional eigenspace of A associated with the repeated eigenvalue λ=2 has basis {v1,v2}.
Case 2: λ=3. The system (A−λI)v=0 is
The last equation here implies that x=y, and then each of the first two equations yields x=y=z. It follows that the eigenspace of A associated with λ=3 is 1-dimensional and has v3=[111]T as a basis eigenvector.
The typical higher-degree polynomial is not so easy to factor as the one in Example 7. Hence a numerical technique such as Newton’s method is often needed to solve the characteristic equation. Moreover, for an n×n matrix with n greater than about 4, the amount of labor required to find the characteristic equation by expanding the determinant |A−λI| is generally prohibitive; because of the presence of the variable λ, row and column elimination methods do not work as they do with numerical determinants. Consequently, specialized techniques, beyond the scope of the present discussion, are often required to find the eigenvalues and eigenvectors of the large matrices that occur in many applications. Problems 40 and 41 at the end of this section outline a numerical technique that sometimes is useful with matrices of moderate size.
In Problems 1 through 26, find the (real) eigenvalues and associated eigenvectors of the given matrix A. Find a basis for each eigenspace of dimension 2 or larger.
[4−211]
[5−63−4]
[8−63−1]
[4−32−1]
[10−96−5]
[6−43−1]
[10−86−4]
[7−612−10]
[8−102−1]
[9−1020]
[19−1021−10]
[13−156−6]
[2002−2−1−263]
[5004−4−2−2126]
[2−202−2−1−223]
[10−1−23−1−660]
[35−2020021]
[100−68212−15−3]
[36−2010001]
[100−47210−15−4]
[4−312−11002]
[5−636−736−62]
[1222022200320004]
[1040014000300003]
[1010011000200002]
[400−3020000−10600−5]
Find the complex conjugate eigenvalues and corresponding eigenvectors of the matrices given in Problems 27 through 32.
A=[01−10]
A=[0−660]
A=[0−3120]
A=[0−12120]
A=[024−60]
A=[0−4360]
Suppose that λ is an eigenvalue of the matrix A with associated eigenvector v and that n is a positive integer. Show that λn is an eigenvalue of An with associated eigen-vector v.
Show that λ is an eigenvalue of the invertible matrix A if and only if λ−1 is an eigenvalue of A−1. Are the associated eigenvectors the same?
Suppose that A is a square matrix. Use the characteristic equation to show that A and AT have the same eigenvalues.
Give an example of a 2×2 matrix A such that A and AT do not have the same eigenvectors.
Show that the eigenvalues of a triangular n×n matrix are its diagonal elements.
Suppose that the characteristic equation |A−λI|=0 is written as a polynomial equation [Eq. (5)]. Show that the constant term is c0=detA. Suggestion: Substitute an appropriate value for λ.
Problems 38 through 42 introduce the trace of a square matrix and explore its connections with the determinant and the characteristic polynomial of the matrix.
If A=[aij] is an n×n matrix, then the trace Tr A of A is defined to be
the sum of the diagonal elements of A. It can be proved that the coefficient of λn−1 in Eq. (5) is cn−1=(−1)n−1(Tr A). Show explicitly that this is true in the case of a 2×2 matrix.
Suppose that the n×n matrix A has n (real) eigenvalues λ1,λ2,…,λn. Assuming the general result stated in Problem 38, prove that
According to the results stated in Problems 37 and 38, the characteristic polynomial
of a 3×3 matrix A is given by
The remaining coefficient c1 can be found by substituting λ=1 and then calculating the two determinants |A| and p(1)=|A−I|. Use this method to find the characteristic equation, eigenvalues, and associated eigenvectors of the matrix
According to the results stated in Problems 37 and 38, the characteristic polynomial
of a 4×4 matrix A is given by
The remaining coefficients c1 and c2 can be found by substituting λ=±1 and calculating the three determinants |A|, p(1)=|A−I|, and p(−1)=|A+I|. Use this method to find the characteristic equation, eigenvalues, and associated eigenvectors of the matrix
Combine ideas from Problems 37–40 to show explicitly that
If A is a 2×2 matrix then det(A−λI)=(−λ)2+Tr (A)(−λ)+det(A).
If A is a 3×3 matrix then det(A−λI)=(−λ)3+Tr (A)(−λ)2+c1(−λ)det(A), where c1 is the sum of the minors of the diagonal elements of A. (Recall from Section 3.6 the distinction between minors and cofactors.)
3.133.134.151