1

1.1

Eigenvalues and Eigenvectors

Characteristic Polynomial and Characteristic Equation

Procedure. How to nd the eigenvalues? A vector x is an e.vector if x is nonzero and satises Ax = x (A I)x = 0 must have nontrivial solutions (A I) is not invertible by the theorem on properties of determinants det(A I) = 0 Solve det(A I) = 0 for to nd eigenvalues.

Denition. P () = det(A I) is called the characteristic polynomial. det(A I) = 0 is called a characteristic equation.

Proposition. A scalar is an e.v. of a n n matrix if satises P () = det(A I) = 0.

Example. Find the e.v. of A =

0 1 . 6 5

Since A I =

0 1 0 6 5 0

1 , 6 5

we have the characteristic equation det(A I) = (5 ) + 6 = ( 2)( 3) = 0. So = 2, = 3 are eigenvalues of A.

Theorem. Let A be a nn matrix. Then A is invertible if and only if: a) = 0 is not an e.v. of A; or b) det A = 0.

Proof. For b) we have discussed the proof on the determinant section. For a): (): Let A be invertible det A = 0 det(A0I) = 0 = 0 is not an e.v. (). Let 0 be not an e.v of A det(A 0I) = 0 det A = 0 A is invertible.

Theorem. The eigenvalues of a triangular matrix are the entries of the main diagonal.

Proof. Recall that a determinant of a triangular matrix is a product of main diagonal elements. Hence, if a11 a12 . . . a1n 0 a ... a 22 2n A= . . . . . . , . . . . . . 0 0 . . . ann then the characteristic equation is a11 a12 ... a1n 0 a22 . . . a2n det(A I) = . . . ... . . . . . . 0 0 . . . ann = (a11 )(a22 ) . . . (ann ) = 0 a11 , a22 , . . . , ann are the eigenvalues of A. 3 2 3 Example. Find the eigenvalues of A = 0 6 10 0 0 2 3 2 3 Solution. det(A I) = det 0 6 10 . 0 0 2 Thus the characteristic equation is (3)(6)(2) = 0 eigenvalues are 3, 6, 2.

Example. Suppose is e.v. of A. Determine an e.v. of A and A3 . What is an e.v. of An ?

2

Solution. Since is e.v. of A nonzero vector x such that Ax = x AAx = Ax = Ax = 2 x. Therefore 2 is an e.v. of A2 . Analogously for A3 . We have Ax = x and A2 x = 2 x AA2 x = A3 x = A2 x = 2 Ax = 3 x. Thus 3 is an e.v. of A3 . In general, n is an e.v. of An .
1.2 Similar Matrices

Denition. A n n matrix B is called similar to matrix A if there exists an invertible matrix P such that B = P 1 AP .

Theorem. If n n-matrices A and B are similar, then they have the same characteristic polynomial and hence the same eigenvalues. Proof. If B = P 1 AP , then B I = P 1 AP P 1 P = P 1 (AP P ) = P 1 (AI)P . Using the multiplicative property of determinant, we have det(B I) = det(P 1 (A I)P ) = det P 1 det(A I) det P = det(A I). Hence, matrices A and B have the same e.v.

Theorem. Hamilton-Caley. (Without proof. Try to prove it as an exercise) If P () = det(A I) = n + cn1 n1 +. . .+c1 +c0 = 0 then P (A) = An +cn1 An1 + . . . + c1 A + c0 I = 0.
1.3 Algebraic and Geometric Multiplicity of Eigenvalues

Denition. The algebraic multiplicity of an eigenvalue is its multiplicity as a root of the characteristic equation (multa ()). 2 5 Example. Find the polynomial of A = 9 1 and nd e.v. with the algebraic multiplicity. 0 3 1 2 0 0 0 0 3 0 5 1

Solution. The characteristic equation is det(A I) 2 0 0 0 5 3 0 0 = det 9 1 3 0 1 2 5 1 = (2 )(3 )(3 )(1 ) = 0 Thus the e.v. are 1 = 2, 2,3 = 3 and 4 = 1. The algebraic multiplicity of = 3 is 2, or multa (3) = 2.

Denition. The eigenspace E consists of the zero vector and all eigenvectors corresponding to an e. v. .

Denition. The geometric multiplicity of an e.v. is the dimension of the corresponding eigenspace E (multg ()). Recall that the dimension of a vector space is equal to the number of linearly independent vectors it contains.

Example. Find e.v. and their algebraic and geometric 0 1 1 multiplicity for A = 1 0 1. 1 1 0 Solution. The characteristic equation is det(A I) 1 1 = det 1 1 = 3 +3+2 = (2)(+1)2 . 1 1 So the e.v. are 1 = 2, 2,3 = 1. Solving the equation (A i I)x = 0 for i = 1, 2, 3 we nd that 1 E=2 = Span{1} 1

E=1

1 0 = Span{ 0 , 1} 1 1

Thus multa (2)=multg (2) = 1 and multa (1)=multg (1) = 2

Example. Find e.v. and their algebraic and geometric 0 0 1 multiplicity for A = 1 0 3. 0 1 3 Solution. The characteristic equation is det(A I) 0 1 = det 1 3 = 3 32 + 3 1 = ( 1)3 . 0 1 3 So the e.v. are 1,2,3 = 1. Solving the equation (A i I)x = 0 for i = 1, 2, 3 we nd that 1 E=1 = Span{2} 1 Thus multa (1) = 3 and multg (1) = 1.

Multiplicity Theorem.

For any eigenvalue i , i = 1, 2, . . . , n of a n n-matrix A holds multg () multa ().

Proof. Let i be an eigenvalue of A. Let Bi = {v1 , . . . vm } be a basis of the corresponding eigenspace Ei where multg (i ) = m. Note that each vj in Bi is an eigenvector of A corresponding to i . Thus Avj = i vj , j = 1, 2, . . . , m

Extend Bi to form a basis B = {v1 , . . . vm , vm+1 , . . . , vn }. Note that B is now a basis in the whole n-dimensional space (Rn or C n ) while Bi is only a basis in the eigenspace corresponding to e.v. i . Note that eigenspace Ei is only subspace of the whole n-dimensional space (Ei Rn or Ei C n ).

Let Q = (v1 |v2 | . . . |vm |vm+1 | . . . |vn ) be a matrix which columns are vectors v1 , . . . vm , vm+1 , . . . , vn of the vector basis B. Since these vectors are linearly independent, the matrix Q is invertible. Notice that vj = Qej where ej = (0, 0, . . . , 0, 1, 0 . . . , 0)T . Such vector ej is called a j-th ort. Now using the denition of e.v.: Q1 Avj = Q1 i vj = i Q1 vj = i ej , j = 1, 2, . . . , m.
j

Thus A = Q1 AQ = Q1 A[v1 , . . . vm , vm+1 , . . . , vn ] = [i e1 |i e2 | . . . , |i em |Q1 Avm+1 | . . . |Q1 Avn ]] = i Im C , 0 D where Im is the m m-identity matrix. The matrix A is similar to the matrix A since A = Q1 AQ.

Hence using the property of determinant for the block diagonal matrixes (see the assignment 2) PA () = PA = det(A In ) = det(( i )Im )det(D Inm ) = ( i )m PD (). Here In and Inm are n n- and (n m) (n m)identity matrixes respectively. Thus a characteristic polynomial PA () has a root of i of at least degree m, where m = multg (i ). multg () multa (). 1 0 0 1 Example. Find e.v. and eigenspace of A = 0 0 0 0 Dene the algebraic and geometric multiplicities of 0 0 0 0 . 1 1 0 1 e.v.

Solution. Since the matrix A is upper-triangular, its the only e.v. is 1,2,3,4 = 1. Thus multa () = 4. To nd the eigenspace and geometric multiplicity we need to solve the equation (A 1I)x = 0 and nd basis

10

for the null space of (A 1I). 0 0 A 1I = 0 0 Hence the solution is x1 , x2 and x3 are arbitrary numbers, x4 = 0. Thus we can choose 3 linearly independent vectors, for example, 1 0 0 0 1 0 { , , } 0 0 1 0 0 0 Therefore, multg () = 3 Notice that the eigenspace E1 is a 3-dimensional hyperplane in R4 . 0 0 0 0 0 0 0 0 0 0 1 0

11

Example. Suggest a 4 4-matrix with e.v. = 1 and multa (1) = 4 and multg (1) = 3. Solution. From the Multiplicity Theorem we have the following options multg () = 1, multg () = 2 and multg () = 4. From the previous example it is easy to see that 1 0 0 0 0 1 0 0 if A = I = 0 0 1 0 0 0 0 1 0 0 then multa (1) = 4 and A 1I = 0n = 0 0 Hence the solution of (A 1I)x = 0 is x1 , x2 x3 and x4 are arbitrary numbers. Thus we can choose 4 linearly independent vectors, for example, 0 0 0 1 0 1 0 0 { , , , } 0 0 1 0 1 0 0 0
12

0 0 0 0

0 0 0 0

0 0 . 0 0

Therefore, multg () = 4. Notice that for this case the eigenspace E1 coincides with R4 . To get multg () = 2, it is possible to think backward and choose such a matrix A such that the null space of (A 1I) has only 2 linearly independent vectors. This would imply that x3 = 0 and x4 = 0 while x1 and x2 are arbitrary. 0 0 0 0 0 0 1 0 For example, A 1I = 0 0 0 1 and 0 0 0 0 0 0 . 1 1 1 0 0 0 1 Notice that A = 0 0 1 0 0 0 trary and , , = 0. 1 0 hence A = 0 0 0 1 0 0 0 1 1 0

0 0 also works for arbi 1

Exercise. Find a 4 4-matrix with e.v. = 1 and multa (1) = 4 and multg (1) = 3.

13

1.4

Trace of a Matrix

Denition. The trace of an nn-matrix A is dened to be Tr(A) = Sp(A) = n aii , i.e., the sum of the diagonal i=1 elements. (Tr is English, Sp is German from Spur.) Properties. Tr(A) = Tr(AT ) Tr(A) = Tr(A) Tr(A + B) = Tr(A) + Tr(B) Tr(AB) = Tr(BA) Proof as an exercise. Theorem. Let A be a n n-matrix and 1 , 2 , . . . , n be its e.v. Then Tr(A) =
n i=1 i

and det(A) =

n i=1 i .

Proof. Let for simplicity assume that A is similar to a diagonal matrix D = diag{1 , 2 , . . . , n }. Hence A = P 1 DP . From the properties of trace tr(A) = tr(P1 DP) = tr(PP1 D) = tr(D) = From the properties of determinants det(A) = det(P1 DP) = det(P)det(D)det(P1 ) = det(D) =
n i=1 i . 14 n i=1 i .

Example. Find eigenvalues of A = calculation.

a a a a

without

Solution. Notice that det(A) = 1 2 = 0 and Tr(A) = 1 + 2 = 2a 1 = 0 and 2 = 2a.

1.5 Diagonalization

Denition. The matrix is diagonal if all its entries are only on the main diagonal. 1 0 Example. I = . . . 0 0 1 . . . 0 ... ... ... ... 0 1 0 0 0 . , D = 0 2 0 . . . 0 0 3 1

If a matrix is diagonal, it is trivial to compute Dk , det D, etc. 2 0 . Compute D2 , D3 , and 0 3

Example 1. Let D = Dk , k N.

15

2 0 0 3 2 Analogously, D = 0 2k k In general, D = 0 Solution. D =

2 0 4 0 = = 0 3 0 9 0 22 0 23 == 3 0 32 0 0 . 3k

22 0 . 0 32 0 . 33

Example 2. Let A = P DP 1 , nd a formula for Ak , k N. Solution. By induction, A2 = (P DP 1 )(P DP 1 ) = (P D2 P 1 ). If it is true for (n1), then An = (P DP 1 )(P D(n1) P 1 ) = P Dn P 1 . Denition. A n n-matrix is said to be diagonalizable if A is similar to a diagonal matrix, i.e. P invertible such that A = P DP 1 , where D is diagonal. Theorem . A n n-matrix is diagonalizable i A has n independent eigenvectors. In fact, A = P DP 1 , with D a diagonal matrix, i the columns of P are n linearly independent eigenvectors of A. In this case, the diagonal entries of D are e.v. of A, corresponding to eigenvectorscolumns of P .

16

Proof. (): Given A = P DP 1 . Notice that if P is a n n-matrix with columns v1 , . . . , vn , and if D is any diagonal matrix with diagonal entries 1 , . . . , n , then AP = A v1 | v2 | . . . | vn = Av1 | Av2 | . . . | Avn , while 1 0 . . . 0 0 ... 0 2 PD = P . . .. . . . . . . . . 0 0 . . . n = 1 v1 | 2 v2 | . . . | n vn . If A = P DP 1 P A = P D Av1 | Av2 | . . . | Avn = 1 v1 | 2 v2 | . . . | n vn Av1 = v1 , av2 = 2 v2 , . . . , Avn = n vn . Since P is invertible, v1 , v2 , . . . , vn are linearly independent non-zero vectors. Hence by denition 1 , 2 , . . . , n are e.v. of A are v1 , v2 , . . . , vn are corresponding eigenvectors of A.
17

(): Given n linearly independent eigenvectors v1 , v2 , . . . , vn , use them to construct P and use n eigenvalues 1 , 2 , . . . , n (not necessarily distinctive) to construct a diag D. Then by denition of e.v. Av1 = 1 v1 , Av2 = 2 v2 , . . . , Avn = n vn . Av1 | Av2 | . . . | Avn = 1 v1 | 2 v2 | . . . | n vn AP = P D. Since P is invertible (all v1 , v2 , . . . , vn are linearly independent) A = P DP 1 .

Denition. Linearly independent vectors v1 , v2 , . . . , vn form an eigenvector basis in Rn . 2 0 0 Example. Diagonalize A = 1 2 1 if possible. 1 0 1 1) Find the eigenvalues of A:

18

 2 0 0 det(A I) = det 1 2 1 1 0 1 = (2 )2 (1 ) = 0. Thus, 1 = 1, 2,3 = 2 are e.v. 2) Find three linearly independent eigenvectors if possible. By solving (A i I)x = 0, i = 1, 2, 3, we get 0 0 1 v1 = 1 , v2 = 1 , v3 = 0 , 1 0 1 corresponding to 1 = 1, 2,3 = 2. Vectors v1 , v2 , v3 are clearly linearly independent for a basis. 3) Construct P from v1 , v2 , v3 : 0 0 1 P = 1 1 0 . 1 0 1
19

4) Construct D from the corresponding eigenvalues 1 , 2 , 3 : 1 0 0 D = 0 2 0 . 0 0 2 5) Check your results by verifying that AP = P D: 2 0 0 0 0 1 0 0 2 AP = 1 2 1 1 1 0 = 1 2 0 , 1 0 1 1 0 1 1 0 2 0 0 1 1 0 0 0 0 2 P D = 1 1 0 0 2 0 = 1 2 0 . 1 0 1 0 0 2 1 0 2 2 4 6 Example. Diagonalize A = 0 2 2 if possible. 0 0 4 Solution. 1) Since A is triangular, it is clear that 1 = 4, 2,3 = 2 are e.v.

20

2) Solve (A i I)x = 0, i = 1, 2, 3, and nd the eigenbasis. 5 Eigenvector for 1 = 4 is v = 1, 1 1 = 2 is v = 0. 0

Eigenvector for 2,3

Thus dimension of eigenspace corresponding to 2,3 = 1 is 1 (multg (2) = 1 and multa (2) = 1) P is singular (it is not enough eigenvectors to form a basis for R3 ) A is not diagonalizable. 5 0 4 Example. Diagonalize A = 0 3 1 if possible. 0 0 2 Solution. Since it is a triangular matrix, the e.v. are 1 = 5, 1 = 5, 3 = 2. For each 1 , 2 , 3 we can nd corresponding eigenvectors

21

 4 1 0 7 v1 = 0 , v2 = 1 , v3 = 1 . 5 0 0 1 Clearly v1 , v2 , v3 are linearly independent form an eigenvector basis in R3 P = v1 | v2 | v3 is invertible A is diagonalizable and D = diag(5, 3, 2).

2 0 0 Why A = 1 2 1 (1 = 1, 2,3 = 2) and 1 0 1 5 0 4 A = 0 3 1 (1 = 5, 2 = 3, 3 = 2) 0 0 2 are diagonalizable? Theorem 1. If v1 , v2 , . . . , vr are eigenvectors corresponding to distinct eigenvalues 1 , 2 , . . . , r for a n n-matrix A, then v1 , v2 , . . . , vn are linearly independent.
22

Proof. From contradiction. Suppose v1 , v2 , . . . , vr are linearly dependent. Let (p 1) be the last index such that vp is a linear combinations of the preceding linearly independent vectors (p n). Then there exist c1 , c2 , . . . , cp1 such that i, i N, ci = 0 and c1 v1 + c2 v2 + . . . + cp1 vp1 = vp . () Multiply both sides of this equation by A c1 Av1 + c2 Av2 + . . . + cp1 Avp1 = Avp . Using the fact that Avi = i vi , we get c1 1 v1 + c2 2 v2 + . . . + cp1 p1 vp1 = p vp .

()

Multiply () by p and subtract result from (): c1 1 v1 c1 p v1 + c2 2 v2 c2 p v2 + . . . + cp1 p1 vp1 cp1 p vp1 = p vp p vp = 0 c1 (1 p )v1 +c2 (2 p )v2 +. . .+cp1 (p1 p )vp1 = 0. Notice that by construction v1 , v2 , . . . , vp1 are linearly independent by denition of linear independence
23

c1 (1 p ) = c2 (2 p ) = . . . = cp1 (p1 p ) = 0. But i, i N : ci = 0 i, i N : i p = 0 contradiction since by statement of Theorem 1 = 2 = . . . = n . Corollary. If a n n-matrix A has n distinct eigenvalues, A is disgonalizable. Proof. If A has distinct eigenvalues 1 , 2 , . . . , n eigenvectors v1 , v2 , . . . , vn are linearly independent and form an eigenvector basis p = v1 | v2 | . . . | vn is invertible and A is diagonalizable. Diagonalization theorem 2. A n n-matrix A is diagonalizable i multg (i ) =multa (i ), i = 1, 2, . . . , n, i.e. eigenvectors of A form a basis in Rn . Proof. (): Let multg (i ) =multa (i ), i = 1, 2, . . . , n each eigenspace of dimension ni has ni linearly independent vectors but

24

ni = multa (i ) and multa (1 ) + multa (2 ) + . . . + multa (n ) = n

n1 + n2 + . . . + nn = n there are n linearly independent eigenvectors which form an eigenvector basis in Rn A is diagonalizable by Theorem (). (): Let A be diagonalizable. Then by the Theorem* n linearly independent eigenvectors v1 , v2 , . . . , vn which form the matrix P such that A = P 1 DP and D is diagonal. Let B be a set of {v1 , v2 , . . . , vn }. If all e.v. 1 , 2 , . . . , n have algebraic multiplicity 1 (multa i = 1, i = 1, 2, . . . , n) then clearly multa i = multg i = 1, i = 1, 2, . . . , n since dimension of any vector space cannot be less than 1 (trivial case). (General case.) Assume for simplicity that only one eigenvalue such that multa = p and all other e.v. k k 1 , 2 , . . . , n1 , i = k have single algebraic multiplicity (multa i = 1).

25

Since clearly eigenspaces do not intersect, i.e. eigenvector vj corresponding to the e.v. j do not belong to the eigenspace Ei corresponding to the e.v. i (j = j ) (verify it at home) the eigenvectors v1 , v2 , . . . , vnp , i = k, corresponding to 1 , 2 , . . . , n1 , i = k, do not belong to E . k vectors which belong to the set B/{v1 , v2 , . . . , vnp }i=k belong to the eigenspace E . Note that these vectors are k linearly independent by statement of the Theorem and dimB/{v1 , v2 , . . . , vnp }i=k = dimE = p = multa . k k 2 0 0 Example. Why is A = 2 6 0 diagonalizable? 3 2 1 Solution. A has 3 disticnt e.v. 1 = 2, 2 = 6, A is diagonalizable. 2 0 0 2 Example. Diagonalize if possible A = 24 12 0 0 3 = 1 0 0 2 0 0 0 . 0 2

Solution. A has two e.v. of multa = 2: 1,2 = 2, 3,4 = 2. Solve (A i E)x = 0 and nd eigenvectors.

26

Basis for 1,2 multg (2) = 2. Basis for 3,4

1 0 1 0 = 2: v1 = , v2 = . Thus 6 3 6 0 0 0 0 0 = 2: v3 = , v4 = . Thus 1 0 0 1

multg (2) = 2. Hence by The Diagonalization Theorem, since multa (2) = multg (2) = 2 and multa (2) = multg (2) = 2, the matrix A is diagonalizable.

27