You are on page 1of 45

# III.

Matrices

Definition §III.1 An m × n matrix is a set of numbers arranged in a rectangular array having m rows and n
columns. It is written

A11
 A21
A=
 ...

Am1

A12
A22
..
.

···
···
..
.

A1n
A12 
.. 
. 

Am2

· · · Amn

There are two important special cases. A 1 × n matrix (that is, a matrix with 1 row) is called a row vector.
An m × 1 matrix (that is, a matrix with 1 column) is called a column vector. Our convention will be that row
indices are always written before column indices. As a memory aid, I think of matrices as being RC (Roman
Catholic or rows before columns).

§III.1.

Matrix Operations

Definitions
1. Equality. For any two matrices A and B
A = B ⇐⇒ (a) A and B have the same number of rows and the same number of columns and
(b) Aij = Bij for all i, j
2. Addition. For any two m × n matrices A and B
(A + B)ij = Aij + Bij for all 1 ≤ i ≤ m, 1 ≤ j ≤ n
That is, the entry in row i, column j of the matrix A + B is defined to be the sum of the corresponding entries
in A and B. Note: The sum A + B is only defined if A and B have the same number of rows and the same
number of columns.
3. Scalar multiplication. For any number α and any m × n matrix A
(αA)ij = αAij for all 1 ≤ i ≤ m, 1 ≤ j ≤ n
For example  

 
 

2 5
2×1+0 2×2+1
0 1
1 2
=
=
+
2
1 7
2×0+1 2×3+1
1 1
0 3 

4. Matrix multiplication. For any m × p matrix A and any p × n matrix B
(AB)ik =

p
X
j=1

Aij Bjk for all 1 ≤ i ≤ m, 1 ≤ k ≤ n
1

Note (a) AB is only defined if the number of columns of A is the same as the number of rows of B. If A is
m × p and B is p × n, then AB is m × n.
(b) (AB)ik is the dot product of the ith row of A (viewed as a row vector) and the k th column of B
(c) Here is a memory aid. If you write the first factor of AB to the left of AB and the second factor above
AB, then each entry (AB)ik of AB is built from the entries Aij j = 1, 2, · · · of A that are directly to
B is 3 × n

A is m × 3

AB is m × n

its left and the entries Bjk j = 1, 2, · · · of B that are directly above it. These entries are multiplied in
pairs, Aij Bjk , j = 1, 2, · · ·, starting as far from AB as possible and then the products are added up to
P
yield (AB)ik = j Aij Bjk .

(d) If A is a square matrix, then AAA · · · A (n factors) makes sense and is denoted A n .

(e) At this stage we have no idea why it is useful to define matrix multiplication in this way. We’ll get
some first hints shortly.

Example III.2 Here is an example of the product of a 2 × 3 matrix with a 3 × 2 matrix, yielding a 2 × 2 matrix 

 
1 3  

 

0 1 2 
0×1+1×2+2×3 0×3+1×2+2×1
8 4

2 2 =
=
3 4 5
3×1+4×2+5×3 3×3+4×2+5×1
26 22
3 1

Example III.3 Here is an example of the product of a 3 × 3 matrix with a 3 × 1 matrix, yielding a 3 × 1 matrix

  
x1 + x 2 + x 3
x1
1 1 1
 1 2 3   x2  =  x1 + 2x2 + 3x3 
2x1 + 3x2 + x3
x3
2 3 1

Hence we may very compactly write the system of equations
x1 + x 2 + x 3 = 4
x1 + 2x2 + 3x3 = 9
2x1 + 3x2 + x3 = 7
that we dealt with in Example II.1, as A~x = ~b where

1 1 1
A = the matrix of coefficients  1 2 3 
2 3 1

x1
~x = the column vector  x2 
x3

2

 
4
~b = the column vector  9 
7

Basic Properties of Matrix Operations.
Using A, B and C to denote matrices and α and β to denote numbers
1. A + B = B + A
2. A + (B + C) = (A + B) + C
3. α(A + B) = αA + αB
4. (α + β)A = αA + βA
5. (αβ)A = α(βA)
6. 1A = A
7. A + 0 = A

where 0 is the matrix all of whose entries are zero

8. A + (−1)A = 0

(The matrix (−1)A is generally denoted −A.)

9. A(B + C) = AB + AC
10. (A + B)C = AC + BC
11. A(BC) = (AB)C
12. α(AB) = (αA)B = A(αB)
These properties are all almost immediate consequences of the definitions. For example to verify property 9, it
suffices to write down the definitions of the two sides of property 9
[A(B + C)]ik =

X

Aij (B + C)jk =

j

X

Aij (Bjk + Cjk )

j

[AB + AC]ik = [AB]ik + [AC]ik =

X

Aij Bjk +

j

X

Aij Cjk

j

Counterintuitive Properties of Matrix Operations that Cause Numerous Errors
13. In general AB 6= BA. Here are three examples. In the first AB is defined and BA is not defined. In
the second AB and BA are both defined, but are not of the same size. In the third AB and BA are
both defined and are of the same size, but are different.
A = [1 2]

B= 

A = [1 2]

B= 


3
4 

0 0
A=
1 0  

3 5
4 6

0 1
B=
0 0 

AB = [ 11 17 ] BA is not defined

AB = [ 11 ]  

0 0
AB =
0 1

3 6
4 8 

1 0
BA =
0 0 

BA =   

A consequence of this “unproperty” is that (A − B)(A + B) need not equal A 2 − B 2 . Multiplying out
(A − B)(A + B) carefully gives AA + AB − BA − BB. The middle two terms need not cancel. For
3

example 

0 0
A=
1 0  

0 −1
A−B =
1 0 

14. In general AB may be 0 even 

1
A=
1 

0 1
B=
0 0  

0 1
A+B =
1 0  


0
0 0

A −B =
0
0 0 

−1
(A − B)(A + B) =
0
2

2   


0 0
0
=
0 0
0 

0
1

if A and B are both nonzero. For example     

1
1
1
0 0
6= 0, B =
6= 0, AB =
= 0,
1
−1 −1
0 0

A consequence of this is that AB = AC does not force B = C even if every entry of A is nonzero. For
example 

1
A=
−2       

0 0
3 4
1 2
−1
and yet AB = AC =
6 C=
=
, B=
0 0
3 4
1 2
2

Example III.4 (The Form of the General Solution of Systems of Linear Equations Revisited) We
have just seen that any system of linear equations can be written
A~x = ~b
where A is the matrix of coefficients, ~x is the column vector of unknowns, ~b is the column vector of right hand
sides and A~x is the matrix product of A and ~x. We also saw in §II.3 that, if the ranks of [A] and [A | ~b] are the
same, the general solution to this system is of the form
~x = ~u + c1~v1 + · · · + cn−ρ~vn−ρ
where n is the number of unknowns (that is, the number of components of ~x, or equivalently, the number of
columns of A), ρ is the rank of A and c1 , · · · , cn−ρ are arbitrary constants. That ~u + c1~v1 + · · · + cn−ρ~vn−ρ
is a solution of A~x = ~b for all values of c1 , · · · , cn−ρ means that

A(~u + c1~v1 + · · · + cn−ρ~vn−ρ ) = ~b
for all values of c1 , · · · , cn−ρ . By properties 9 and 12 of matrix operations, this implies that
A~u + c1 A~v1 + · · · + cn−ρ A~vn−ρ = ~b
But this is true for all c1 , · · · , cn−ρ if and only if
A~u = ~b

(Set c1 = · · · = cn−ρ = 0)

A~v1 = ~0 (Set c1 = 1, c2 = · · · = cn−ρ = 0 and sub in A~u = ~b.)
..
.
A~vn−ρ = ~0 (Set c1 = · · · = cn−ρ−1 = 0, cn−ρ = 1 and sub in A~u = ~b.)
4

e) Find all square roots of A. find a matrix B obeying B 2 = A. First. β and vectors ~x. ~y . That is. such that T (~x) = MT ~x. ~vn−ρ is a solution of A~x = ~0.In other words ~u is a solution of A~x = ~b and each of ~v1 . 2) Compute A2 = AA and A3 = AAA for   0 a b a) A =  0 0 c  0 0 0   1 1 3) Let A = . map. Exercises for §III.1 1) Define A=  1 2 3 1 2 1    −1 2 B =  −3 1  −2 1 C = [ 2 −2 0 ]   2 D =  −11  2 Compute all pairwise products (AA. we shall see that for each linear transformation T (~x) there is a matrix M T such that T (~x) is the matrix product of MT times the column vector ~x.a. · · · . (Part of this problem is to invent a reasonable definition of eAt . Later in this section. A transformation is said to be linear if T (α~x + β~ y ) = αT (~x) + βT (~y) for all numbers α. BA. In other words. AB. b) Find Ak for all positive integers. 5 . AD.5 A transformation (a.) d) Find a square root of A. A3 . a. function) T from IRn to IRm is a rule which assigns to each vector ~x in IRn a vector T (~x) in IRm . 4 when 0 0 A= 0 0  §III.a. 3. it spits out a new vector T (~x). · · ·) that are defined. c) Find eAt .2 1 0 0 0 0 1 0 0  0 0  1 0 Matrices and Linear Transformations Definition III. A4 . T should be thought of as a machine: if you put a vector ~x into ~x T T (~x) the input hopper. AC.k. 4) Compute Ak for k = 2. 0 1 b)   1 0 a A = 0 1 0 0 0 1 a) Find A2 .k.

however. But if ~v 6= ~0 and α + β 6= 1. Many other geometric operations are linear maps. If translation were linear the two expressions tran~v (α~x + β~ y ) = α~x + β~ y + ~v α tran~v (~x) + β tran~v (~y ) = α(~x + ~v ) + β(~y + ~v ) = α~x + β~ y + (α + β)~v would be equal for all α and β. sin φ] is a unit vector that lies on ~x φ projφ (~x) the line. Example III. The vector ˆb = [cos φ. because the two quantities T (2~x) = (2x2 )2 2T (~x) = 2(x2 )2 are not equal whenever x2 6= 0. In equations. As a result. Another example of a map that is not linear is Example III.6 Translation Define tran~v (~x) be the vector gotten by translating the head of the arrow ~x by tran~v (~x) ~v ~x ~v (while leaving the tail of the arrow fixed). the two expressions are not equal. So projφ (~x) must have direction [cos φ. tran~v (~x) = ~x + ~v . we look at a number of examples. Here are some examples.7 Projection. sin φ] (or its negative) and must have length k~xk cos θ = ~x · ˆb. where θ is the angle between ~x and ˆb. The map T ([x1 . both of transformations that are not linear and transformations that are linear. linear maps play a big role in computer graphics. The unique vector with the right direction and length is projφ (~x) = (~x · ˆb) ˆb 6 . x2 ]) = x22 is not linear. Define projφ (~x) to be the projection of the vector ~x on the line in IR2 that passes through the origin at an angle φ from the x–axis. We have just seen an example of a geometric operation that is not linear.

           1 1 + cos 2φ 1 + cos 2φ sin 2φ 1 0 sin 2φ x1 x1 x1 x1 = − reflφ (~x) = 2 − x2 sin 2φ 1 − cos 2φ x2 0 1 x2 sin 2φ 1 − cos 2φ x2 2          1 + cos 2φ sin 2φ 1 0 cos 2φ sin 2φ x1 x1 = − = sin 2φ 1 − cos 2φ 0 1 x2 sin 2φ − cos 2φ x2 Example III. So reflφ (~x) = ~x + 2[projφ (~x) − ~x] = 2projφ (~x) − ~x We may.8 Reflection. To continue an equal distance in the same direction. once again. you have to add a second copy of projφ (~x) − ~x. rotφ (~x) ~x φ θ 7 . to get from ~x to projφ (~x).9 Rotation. you have to add the vector projφ (~x) − ~x to ~x.It is easy to verify that this really is a linear transformation:  projφ (α~x + β~ y ) = (α~x + β~ y ) · ˆb ˆb = α(~x · ˆb) ˆb + β(~y · ˆb) ˆb = α projφ (~x) + β projφ (~y ) Writing both projφ (~x) and ~x as column vectors    x1 cos2 φ + x2 sin φ cos φ cos φ = x1 sin φ cos φ + x2 sin2 φ sin φ      1 1 + cos 2φ x1 sin φ cos φ x1 sin 2φ = x2 sin2 φ x2 sin 2φ 1 − cos 2φ 2 projφ (~x) = (~x · ˆb) ˆb = (x1 cos φ + x2 sin φ) =  cos2 φ sin φ cos φ  Notice that this is the matrix product of a matrix that depends only on φ (not on ~x) times the column vector ~x. In terms of vectors. Define rotφ (~x) to be the result of rotating the vector ~x by an angle φ about the origin. write this as the matrix product of a matrix that depends only on φ (not on ~x) times the column vector ~x. Example III. You can get from ~x to reflφ (~x) by first walking from ~x to reflφ (~x) projφ (~x) ~x ~x projφ (~x) and continuing in the same direction an equal distance on the far side of the line. Define reflφ (~x) to be the reflection of the vector ~x on the line in IR2 that passes through the origin at an angle φ from the x–axis.

which is one. where A~x is the matrix product of the matrix A times the column vector ~x.. because A(α~x + β~ y ) = α(A~x) + β(A~y ) by properties (9) and (12) of matrix operations. First we consider the case that T (~x) is a linear map from IR2 (that is. two component vectors) to IR2 . Note that every m component vector can be written        x1 1 0 0  x2  0 1 0        ~x =   .If we denote by r the length of ~x and by θ the angle between ~x and the x–axis. Define ~ei to be the column vector all of whose entries are zero.  = x1~e1 + x2~e2 + · · · xm~em  xm 0 0 1 8 .....      x1 cos φ − x2 sin φ r(cos θ cos φ − sin θ sin φ) r cos(θ + φ) = = rotφ (~x) = x2 cos φ + x1 sin φ r(sin θ cos φ + cos θ sin φ) r sin(θ + φ)    cos φ − sin φ x1 = sin φ cos φ x2  Note that in each of examples 7.  + · · · + +xm  . Any vector ~x in IR2 can be written  x1 x2 = x1     1 0 + x2 = x1~e1 + x2~e2 0 1 e1 =   1 0  where ~e2 =   0 1 Because T is linear T (~x) = T (x1~e1 + x2~e2 ) = x1 T (~e1 ) + x2 T (~e2 ) Define the numbers a11 . for each linear map T (~x) there is a matrix A such that T (~x) = A~x. a12 . We shall now show that.. Every map of the form T (~x) = A~x is automatically linear.  + x2  . a22 by  a T (~e1 ) = 11 a21   a12 T (~e2 ) = a22  Then T (~x) = x1 T (~e1 ) + x2 T (~e2 ) = x1        a11 a11 x1 + a12 x2 a12 a11 = = + x2 a21 a21 x1 + a22 x2 a22 a21 a12 a22  x1 x2  The same construction works for linear transformations from IRm to IRn .. a21 . except for the ith .  = x1  . conversely. then  r cos θ ~x = r sin θ  To rotate this by φ we need only replace θ by θ + φ. 8 and 9 there was a matrix A such that the map could be written T (~x) = A~x..

Think of this vector as a column vector. More generally. because  x1    x2   A~x = T (~e1 ) T (~e2 ) · · · T (~em )   . 7) The matrix √ 3+2 1 √ 3√− 2 4 2 √ √  √3 − 2 −√2  3√+ 2 − √ 2 2 12 9 .. Then   define the matrix A = T (~e1 ) T (~e2 ) · · · T (~em ) .2 1) Find the matrices which project on the lines a) x = y b) 3x + 4y = 0 2) Find the matrices which reflect in the lines a) x = y b) 3x + 4y = 0 3) Find the matrices which project on the planes a) x = y b) x + 2y + 2z = 0 4) Find the matrices which reflect in the planes a) x = y b) x + 2y + 2z = 0 5) Find the matrices which rotate about the origin in IR2 by a) π/4 b) π/2 c) π 6) Find the matrix which rotates about the z–axis in IR3 by θ. then the map constructed by first applying T and then applying S obeys  S T (~x) = MS T (~x) = MS MT ~x  so that the matrix associated with the composite map S T (~x) is the matrix product MS MT of MS and MT . so that T (~x) = T ~x. It is traditional to use the same symbol to stand for both a linear transformation and its associated matrix. For example. A is the matrix whose ith column is T (~ei )..For each i between 1 and m. if S and T are two linear transformations with associated matrices MS and MT respectively. Exercises for §III. That is. This is the matrix we want. T (~ei ) is an n–component vector. the matrix associated with the linear transformation T (~x) is traditionally denoted T as well.   xm = x1 T (~e1 ) + x2 T (~e2 ) + · · · + xm T (~em ) (by the definition of matrix multiplication) = T (x1~e1 + x2~e2 + · · · + xm~em ) (by the linearity of T ) = T (~x) This formula T (~x) = A~x is the reason we defined matrix multiplication the way we did.

fix any i and k with 1 ≤ i ≤ m. To help keep straight what is known and what isn’t.10 The m × m identity matrix Im (generally the subscript m is dropped from the notation) is the m × m matrix whose (i. At this stage. 1 ≤ k ≤ n. Definition III. it is said to be singular. Furthermore the Iii that appears in that term takes the value one so (IA)ik = Iii Aik = Aik . By the definition Pm of Iij the only nonzero term in the sum (IA)ik = j=1 Iij Ajk is that with j = i. For example. I’ve made all of the knowns lower case and all of the unknowns upper case. What is A−1 ? It’s the thing you multiply A by to get 1. as desired. If A has an inverse.is a rotation in IR3 . §III. To be an inverse. for any m × n matrix A Im A = AIn = A It is easy to check that this is true. Otherwise. The inverse is generally denoted A−1 . Give another geometric interpretation of this matrix.12 Let’s consider the general 2 × 2 matrix   a b A= c d and see if we can find an inverse for it. What axis does it rotate about and what is the angle of rotation? 8) Find the matrix which first reflects about the line in IR2 that makes an angle φ with the x–axis and then reflects about the line that makes an angle θ with the x–axis.11 A matrix B is called an inverse of the matrix A if AB = BA = I. For example  1 0 I2 = 0 1  The reason we call this the identity matrix is that. then A is said to be invertible or nonsingular. Let’s call the inverse   X Z B= Y W The matrix A is to be treated as a given matrix. j) matrix element is 1 if i = j and 0 if i 6= j. B must obey AB =  a c b d  X Y Z W  =  aX + bY cX + dY 10 aZ + bW cZ + dW  =  1 0 0 1  .3 Matrix Inverses Suppose A is a matrix. What is 1? Definition III. Example III. B is unknown.

there must exist some 1 ≤ j ≤ m such that the augmented matrix [A|~e j ] also has rank m. If A is an n × n matrix then the following are equivalent: (i) A is invertible.Proof: We have a guess for (AB)−1 .e. Consequently. if a given square matrix satisfies any one of properties (i)–(v). (ii)⇒(iii): A~x = ~0 has precisely one solution ⇒ rank A = #unknowns = n . But it is usually computationally much more efficient to solve A~x = ~b by Gaussian elimination than it is to find A−1 and then multiply A−1~b. Only square matrices can be invertible. 12 . So condition (ii) is applicable to B. m. In other words. we merely need to check the requirements of the definition (AB)(B −1 A−1 ) = ABB −1 A−1 = AIA−1 = AA−1 = I (B −1 A−1 )(AB) = B −1 A−1 AB = B −1 IB = B −1 B = I 3. Suppose that A is invertible. then statement (ii) is also true”. where ~ej is the j th column of I. must equal the rank. We shall see in the next section. This implies that there is a matrix C obeying BC = I. (iii)⇒(iv): In the next section. that the j th column of B is solves the system of equations A~x = ~ej . the number of unknowns. Then there exists an n × m matrix B such that AB = I. Then A~x = ~b ⇐⇒ ~x = A−1~b. A has a right inverse). 5. the corresponding system of equations must have a unique solution. Proof: Multiplying the first equation by A−1 gives the second and multiplying the second by A gives the first. (ii) ~x = ~0 is the only solution of A~x = ~0 (iii) A has rank n. which implies that (iii) and subsequently (iv) are also applicable to B. (v)⇒(i): Assume that BA = I. WARNING: This property is conceptually important. (iv)⇒(v): Assume AB = I. So (iv) and (v) apply to A. Then B~x = ~0 ⇒ AB~x = ~0 ⇒ ~x = ~0. Then A~x = ~0 ⇒ BA~x = ~0 ⇒ ~x = ~0. it satisfies all of them. (i)⇒(ii): If A is invertible. So (ii) applies to A. But C = (AB)C = A(BC) = A so BA = I. 4.e. Outline of Proof: Let A be an invertible m × n matrix. (iv) There is a matrix B such that AB = I (i. n. we’ll develop a procedure for solving AB = I for B explicitly. A has a left inverse). To check that the guess is correct. Because the identity matrix I has rank m. By property 1 above. where I is the m × m identity matrix. Proof: The notation “ (i)⇒(ii) ” means “ If statement (i) is true. (v) There is a matrix B such that BA = I (i. then multiplying both sides of A~x = ~0 by A−1 gives ~x = A−1~0 = ~0.

We could apply Gauss reduction separately to the n systems. We shall get another test. For example. (iv) and (v) of property 4 are all tests for invertibility.4 Finding Inverses Suppose that we are given an n × n matrix A and we wish to find a matrix B obeying AB = I. n ~ i ’s are all unknown. for each that is. find its inverse. We just form one big augmented matrix [A|~e1 ~e2 · · · ~en ].10.10.3 1) Determine which of the following matrices are invertible and. the requirement AB = I is ~1 B ~2 · · · B ~ n ] = [~e1 ~e2 · · · ~en ] A[B By the definition of matrix multiplication ~1 B ~2 · · · B ~ n ] = [AB ~ 1 AB ~ 2 · · · AB ~ n] A[B ~ 1 . For example. 13 . (iii). the first column of AB is AB      a11 b11 + a12 b21 a11 b12 + a12 b22 b11 b12 a11 a12 = AB = a21 b11 + a22 b21 a21 b12 + a22 b22 b21 b22 a21 a22 is indeed identical to  a ~ AB1 = 11 a21 a12 a22  b11 b21   a b + a12 b21 = 11 11 a21 b11 + a22 b21  ~1 B ~2 · · · B ~ n ] = [AB ~ 1 AB ~ 2 · · · AB ~ n ] and [~e1 ~e2 · · · ~en ] to be equal.” §III. · · · . Exercises for §III.Items (ii). But because the left hand sides of all n systems are the same. we can solve the n systems simultaneously.       2 4 1 3 1 1 c) b) a) 1 2 −2 2 1 −1 2) Either prove or find a counterexample to the statement “If BA = I and AC = I then B = C. once we have generalized the definition of determinant to matrices larger than 3 × 3: a square matrix A is invertible if and only if det A 6= 0. We must solve n different Recall that A and the ~ei ’s are all given matrices and that the B systems of linear equations. so the requirement AB = I may be expressed as ~ i = ~ei AB for i = 1. the first column of That is. all of their columns For the two matrices A[B must agree. Do not used the canned formula derived in Example II. Denote ~ i the (as yet unknown) ith column of B and by ~ei the column vector having all entries zero except for a by B one in the ith row. when n = 2. when n = 2     b11 b12 b11 ~ B= B1 = b21 b22 b21   b12 ~ B2 = b22   1 ~e1 = 0   0 ~e2 = 1 In this notation. Use the method of Example II.

13 Let 2 −3 2  1 −1 1 A= 3 2 2 1 1 −3  5 2  1 1  Form the augmented matrix .Example III.

2 −3 2 5 .

1 .

 1 −1 1 2 .

0 .

 3 2 2 1.

0 .

5(1) 0 2. 1 1 −3 1 0 It is important to always remember that if we were to erase  0 1 0 0 all  0 0 0 0  1 0 0 1 columns to the right of the vertical line except ~ i = ~ei .5 −4 [A|ei ] at the .5(1)  0 0.5 −1 (4) − 0. So any for the ith . we would have precisely the augmented matrix appropriate for the linear system A B row operation applied to the big augmented matrix [A|~e1 ~e2 · · · ~en ] really is a simultaneous application of the same row operation to n different augmented matrices  2 −3 2 (2) − 0.5(1) 0 6.5 0  (3) − 1.

5 .

1 .

−0.5 .

−0.5 .

−6.5 .

−1.5 .

5 same time. −1.5 −0. Row reduce in the usual way.  0 0 0 1 0 0  0 1 0 0 0 1 .

2 −3 2 5 .

1 0 0 .

5 . 1 0  0 0.5 0 −0.

−0.5 .

 (3) − 13(2) 0 0 −1 0 .

5 −13 1 .

(4) − 5(2) 0 0 −4 1 2 −5 0  .

2 −3 2 5 .

1 0 0 .

1 0  0 0.5 .5 0 −0.

−0.5 .

 0 0 −1 0 .

5 −13 1 .

But it is easier to do all four at the same time by applying more row operations chosen to turn the left hand side into the identity matrix. (4) − 4(3) 0 0 0 1 −18 47 −4   0 0  0 1  0 0  0 1 We could at this stage backsolve the four systems individually. .

  2 −3 2 5 .

1 0 0 0 .

5 0 −0. 0 0  0 0.5 .

−0.5 1 .

  0 .

−5 13 −1 0 −(3) 0 0 1 .

0 0 0 1 −18 47 −4 1 .

 2 −3 2 5 .

1 0 0 0 .

2(2) + (4)  0 1 0 0 .

−19 49 −4 1  .

  0 0 1 0 .

−5 13 −1 0 .

0 0 0 1 −18 47 −4 1  0.5[(1) + 3(2) − 2(3) − 5(4)] 1 0  0 0  0 1 0 0 0 0 1 0 .

0.

.

0.

.

0.

.

Then the above row operations converted . for example. the first column to the right of the vertical line. 1  22 −57 5 −1 −19 49 −4 1   −5 13 −1 0 −18 47 −4 1 What conclusion do we come to? Concentrate on. mentally erase the other three columns to the right of the vertical line in all of the above computations. In fact.

  2 −3 2 5 .

1 .

 1 −1 1 2 .

0  .

  3 2 2 1.

0 .

1 1 −3 1 0 into 14 1 0  0 0  0 1 0 0 0 0 1 0 .

0.

.

0.

.

0.

.

1  22 −19   −5 −18 .

Because row operations have no effect on the set of solutions of a linear system we can conclude that any given ~ 1 obeys B 2 −3  1 −1  3 2 1 1  As 2 1 2 −3    5 1 2 ~ 0  B1 =   1 0 1 0 1 0  0 0 0 1 0 0  we now have that if and only if it obeys 0 0 1 0 1 0  0 0  0 1 0 0 0 0 1 0    0 22 0 ~  −19   B1 =   0 −5 1 −18  0 0 ~ ~1  B1 = B 0 1  22 −19  ~1 =  B   −5 −18  ~ 1 . for each that is.5 invertible and. Repeating this argument with the other columns A−1  22 −57 5 −1  −19 49 −4 1  =  −5 13 −1 0 −18 47 −4 1  Exercises for §III. find its inverse. the first column of A−1 is the first column to the right of the bar in the final augmented In other words.   1 −1 1  −1 2 −1  2 −1 1   2 1 4 3 2 5 0 −1 1   a 0 where ab 6= 0 0 b   1 a b 0 1 c 0 0 1 Determinants – Definition The determinant of a 1 × 1 matrix is defined by det [a11 ] = a11 The determinant of a 2 × 2 matrix is defined by   a a12 det 11 = a11 a22 − a12 a21 a21 a22 15 . B matrix.4 1) Determine which of the following matrices are   2 3 −1 b) 2 3  a)  1 −1 −1 4   1 1 1 c)  1 2 3  d) 1 4 9   1 a f) e) 0 1   1 0 a g)  0 1 0  h) 0 0 1 §III.

(in fact n ≥ 2) the determinant of an n × n matrix A.. starting with a +. is defined by det [A] = n X (−1)1+j a1j det M1j j=1 where M1j is the (n − 1) × (n − 1) matrix formed by deleting..For n > 2.       ~aj   ~ai      .   .14         1 2 3 0 2 1 2 1 0 det  1 0 2  = 1 × det − 2 × det + 3 × det 2 1 3 1 3 2 3 2 1 = 1 × (0 − 4) − 2(1 − 6) + 3(2 − 0) = 12 Exercises for §III. .. The term is a sign times the entry times the determinant of the (n − 1) × (n − 1) matrix obtained by deleting the row and column that contains the entry. . 16 . The sign alternates. There is one term in the formula for each entry in the top row. from the original matrix A.  .6 Determinants – Properties Property E If two rows of an n × n matrix are exchanged. the row and column containing a1j . This formula is called “expansion along the top row”.  = − det  . where each ~ak is an n component row vector..     ..  a11 det  a21 a31 a12 a22 a32   a13 a11 a23  = a11 det  a21 a33 a31 a12 a22 a32   a13 a11 a23  − a12 det  a21 a33 a31 = a11 (a22 a33 − a23 a32 ) a12 a22 a32   a13 a11 a23  + a13 det  a21 a33 a31 − a12 (a21 a33 − a23 a31 ) a12 a22 a32  a13 a23  a33 + a13 (a21 a32 − a22 a31 ) Example III. . .   ~ai   ~aj       . the determinant is multiplied by −1.. whose ij entry is denoted a ij .5 1) Evaluate the determinant of each of the following matrices       0 1 0 −3 1 3 c) b) a) 1 0 2 −1 2 4 d)  cos θ sin θ − sin θ cos θ  2) Evaluate the determinant of each of the following matrices by expanding along the top row         1 a b 7 −1 5 2 1 5 2 0 0 d)  0 1 c  c)  3 4 −5  b)  1 0 3  a)  1 −1 5  0 0 1 2 3 0 −1 2 0 2 3 −1 §III..    det  .

 .Property M Multiplying any single row of a matrix by k multiplies the determinant by k as well. ... ..      det  k~ai  = k det  ~ai   ...      ~ai + k~am   ~ai      .  det  . a11  0  det    0  0 ∗ a22 0 0 ··· ··· . D We will not give a complete proof..    ~a1 ~a1  . .   . A. .   .   . so we end up with a factor k n .   . we have to apply Property M once for each row in the matrix. . But we will give an outline that is sufficiently detailed that you should be able to fill in the gaps.  ~an ~an  To multiply every entry in an n × n matrix by k. ∗ ∗ · · · an−1 n−1 ··· 0  ∗ ∗    = a11 a22 · · · ann  ∗  ann Property P det AB = det A det B Outline of Proof of Properties E..   .   .. M.   .  Property D The determinant of any triangular matrix is the product of its diagonal entries. 17 . For example  k det 0      0 1 0 1 0 = k × det = k × k × det = k2 k 0 k 0 1 Property A Adding any multiple of any row to any other row has no effect on the determinant. ..        ~am  ~am   ..  = det  ..   .

1). 2. then exchanging 2 and 3 to yield (3. 1. For example. 3) can be tranformed into (3. First assume that A is invertible. (1. 2. the above definition is det A = A11 A22 − A12 A21 with the first term being the σ = (1. 2). 2). (2. 1. 1. 1. 3). so the sign of σ = (1. 2) is +1. The sign of a permutation. So the sign of σ = (3. 2. 1). For any n × n matrix A X det A = σ∈Pn sgn σ A1σ(1) A2σ(2) · · · Anσ(n) Here Pn is the set of all orderings of (1. n) can be transformed into σ using an even number of exchanges of pairs of numbers. convert it into the identity matrix. which in turn follows fairly easily from our definition. 2. 2) contribution and the second term being the σ = (2. 1). (1. The symbol “P ” stands for “permutation”. On the other hand. (3. 2. It is also possible to prove the sgn σ is well–defined for all permutations σ. For example. 2). (3. It is possible to prove that the number of exchanges of pairs of numbers used to transform (1. σ(2) = 3. So  P3 = (1. 1. and that the new definition obeys the “expansion along the first row” formula. · · · . That was the basis of the algorithm for computing A−1 that we developed in §III. here denoted sgn σ is +1 if (1. if A is a  1 0  4 0 4 × 4 matrix. 2). Of course (1. (3. in order. 1) and then exchanging 1 and 2 to yield (3. 3. 3. 3). that when applied to A. 2) by first exchanging 1 and 2 to yield (2. To verify that the new and old definitions of determinant agree. 1) We use σ(i) to denote the ith entry of the permutation σ. 2. 3. 2) by just exchanging 2 and 3. 1. Outline of Proof of Property P Once again. 1. 3) can be tranformed into (1. 3). Otherwise. all rows of  0 0 0 1 0 0 A 0 1 0 0 0 1 18 . 1) contribution. 3) to (3. which is the mathematical name for a reordering. the sign is −1. n). 3. If A is a 3 × 3 matrix. (1. 2. which is still even. 2. 1.4. 2). (2. 3). which is trivial. 2). 1) and finally exchanging 1 and 2 to yield (3. (2. the above definition is det A = A11 A22 A33 − A11 A23 A32 − A12 A21 A33 + A12 A23 A31 + A13 A21 A32 − A13 A22 A31 with the terms being the contributions. 1). 2) by first exchanging 1 and 3 to yield (3.All of these properties follow fairly easily from another (in fact the standard) definition of determinant. 2. 3. 1. 3. Any row operation can be implemented by multiplication by a matrix. · · · . For example (1. 2) is −1. 2. 2. when σ = (2. Then there is a sequence of row operations. then exchanging 1 and 3 to yield (2. it suffices to prove that they agree for 1 × 1 matrices. 3). from σ = (1. 3) can also be transformed into (3. (3. 3. (2. This used four exchanges. 1. 1. 2. we will not give a complete proof. 3. If A is a 2 × 2 matrix. 2) is always even. 1).

2) A determinant may be expanded along any row.   .  ~  ~   b   b   . A . The next step is to check that det QC = det Q det C for any elementary matrix Q and any square matrix C. P 1) If any two rows of a matrix A are the same.   .   .   . then det A = 0 because     ~a1 ~a1  .   . The matrices that implement row operations are called elementary matrices.with the exception of row (3) coincide with the corresponding row of A. Qh of row operation implementing matrices.. multiplication by elementary matrices. Row (3) of the product is row (3) of A plus 4 times row (1) of A.   .. So we conclude that if A is not invertible it necessarily has determinant zero.D. Then there is a sequence of row operations. · · · . =⇒ det Qh · · · det Q1 det A = 1 with h = 1 and A replaced by the inverse of Q1 . convert it into a matrix A. We have Qh · · · Q1 A = I for some sequence Q1 . we see that every elementary matrix has nonzero determinant. that when applied ˜ whose last row is identically zero. That is. M.  ~  ~   b   b   . observe that if A fails to be invertible. we have ˜˜ · · · Q ˜ 1 A = A˜ Q h =⇒ ˜ ˜ · · · det Q ˜ 1 det A = det A˜ det Q h ˜ has determinant zero.     = − det  . Then Qh · · · Q1 A = I =⇒ det Qh · · · det Q1 det A = 1 Qh · · · Q1 AB = B =⇒ det Qh · · · det Q1 det AB = det B 1 det AB = det B det A =⇒ Now suppose that A is not invertible... Implementing these row operations using to A. the two rows containing ~b have been interchanged. We shall denote them by Qj here. Implications of Properties E. det   .  .. This is straight forward because we already know the effect that any row operation has on a determinant. the same is true for AB (otherwise B(AB) −1 is an inverse for A) and so both det A = 0 and det AB = 0. Applying Qh · · · Q1 A = I Any matrix that has at least one row identically zero. Finally. like A. det [A] = n X (−1)i+j aij det Mij j=1 19 .   .  ~an ~an On the right hand side.

Example III. M and A.14 along its second row. Then.. combined with Gaussian elimination. the resulting matrix V obeys det U = det V if the row operation is (i) → (i) + k(j) for some j 6= i det U = − det V if the row operation is (i) ↔ (j) for some j 6= i det U = 1 k det V if the row operation is (i) → k(i) for some k 6= 0 These properties. Then A is invertible ⇐⇒ det A 6= 0 Proof: Let R be the triangular matrix that results from the application of Gaussian elimination to the matrix A. .. . . M say that if you take a matrix U and apply to it a row operation. we get        3 2 3 1 3 1 2 2  = −1 × det + 0 × det − 2 × det 2 1 3 1 3 2 1 = −1 × (2 − 6) + 0 − 2(2 − 6) = 12 3) We can now use Gaussian elimination to evaluate determinants.. that is trivially computed using Property D. the matrix of Example III.16  1 2 det  1 1 2 4 3 3 1 0 5 3   1 2 3 A 0 = det  0 6 0 9   (1) 1 2 1 2 0 −2 −1  (2) − 2(1) E 0 = − det   (3) − (1) 0 1 4 4 (4) − (1) 0 1 2 7 1 A 0 = − det  0 0  D   1 (1) 2 1 2 (2) 1 4 4  A 0 = − det   0 (3) 0 −2 −1 0 (4) − (2) 0 −2 3 = − 1 × 1 × (−2) × 4 = 8  2 1 2 1 4 4   0 −2 −1 1 2 7 (1) (3) (2) (4)  (1) 2 1 2 (2) 1 4 4   (3) 0 −2 −1 (4) − (3) 0 0 4 4) The determinant provides a test for invertibility: Let A be an n × n matrix. To get the signs (−1)i+j right. E. Example III. det A = (nonzero number) det R 20 . you  + −  + . allow us to relate the determinant of any given matrix to the determinant of a triangular matrix. . Properties A.where Mij is the (n − 1) × (n − 1) matrix formed by deleting from the original matrix A the row and column containing aij .15 If we expand  1 2 det  1 0 3 2 just have to remember the checkerboard  − + ··· + − ···  − + ··· . . . . by properties E.

−1 m Exercises for §III.3) 5) det A−1 = det A det Am = det A are easy consequences of property P.So det A 6= 0 ⇐⇒ det R 6= 0 ⇐⇒ the diagonal entries Rjj of R are all nonzero (Property D) ⇐⇒ rank R = n (See the definition of rank in §II.3) ⇐⇒ A is invertible (Property 4 of §III.6 1) Evaluate the determinant of each of the following matrices by expanding along some row other than the first  2 1 5  1 0 3 −1 2 0  a) b)  7 −1 5  3 4 −5  2 3 0  c)  1 a b 0 1 c 0 0 1  2) Evaluate the determinant of each of the following matrices using row reduction a) c) 1 1  1 1  3  −2  1 4  −1 2 −2 1 1 −1 −3 1  1 −1 4 8   4 −8 1 1  2 0 5 −2   1 1 2 −3 b) d) 0  −1  2 −3  1 1  1 1  3) For which values of the parameter λ does the matrix  − 21 − λ − 23 1 2 − 52 − λ  have an inverse? 4) For which values of the parameter λ does the system of linear equations 2x1 − x2 = λx1 2x1 + 5x2 = λx2 have a solution other than x1 = x2 = 0? 21 1 0 −1 −2 1 1 2 4 3 9 4 16  −2 3 1 2  0 1 −1 0  1 8   27 64 .

Concise Formulae a) If A is a square matrix with det A 6= 0. A−1  = (−1)i+j ij det Mji det A where Mji is the matrix gotten by deleting from A the row and column containing Aji . A square matrix A is invertible if and only if det A 6= 0. But it is usually computationally much more efficient to solve AB = I by Gaussian elimination than it is to apply this formula. this is the expansion along row j ˜ This other matrix is constructed from A by replacing row j for the determinant of another matrix A. for studying the dependence of A−1 on matrix elements of A. of A by row k of A. If k 6= j.3). Equivalently. This is heavily used in the eigenvalue/eigenvector problem in the next chapter.17 Let   1 1 1 A = 1 2 3 2 3 1 22 . WARNING: This formula is useful. Let B be the matrix with Bij = (−1)i+j det Mji We have to show that (AB)kj is det A if j = k and 0 if j 6= k. for example. This was implication number 2 in the last section. Example III. the system of equations A~x = ~0 has a nonzero solution ~x if and only if det A = 0. Row numbers j and k of A˜ are identical so that det A˜ = 0 by implication number 1 of Property E.§III. (see property 4 of inverses given in §III. Proof: The foundation for this formula is the expansion formula det A = X (−1)i+j Aji det Mji i It is true for any j. But (AB)kj = X Aki Bij = X (−1)i+j Aki det Mji i i If k = j this is precisely the formula for det A given above.7 Determinants – Applications Testing for Invertibility One of the main uses for determinants is testing for invertibility.

18 Consider the system of equations of Example II. x1 + x 2 + x 3 = 4 x1 + 2x2 + 3x3 = 9 2x1 + 3x2 + x3 = 7 This system can be written in the form A~x = ~b with   1 1 1 A = 1 2 3 2 3 1   4 ~b =  9  7 23 .Then det M11 det M21 det M31       2 3 1 3 1 2 = det = −7 det M12 = det = −5 det M13 = det = −1 3 1 2 1 2 3       1 1 1 1 1 1 =1 = −1 det M23 = det = −2 det M22 = det = det 2 3 2 1 3 1       1 1 1 1 1 1 = det =1 det M32 = det =2 det M33 = det =1 2 3 1 3 1 2 so that det A = 1 × M11 − 1 × M12 + 1 × M13 = −7 + 5 − 1 = −3 and A−1   −7 2 1 1  5 −1 −2  = −3 −1 −1 1 b) (Cramer’s rule) If A is a square matrix with det A 6= 0. But it is usually computationally much more efficient to solve A~x = ~b by Gaussian elimination than it is to apply this formula. Example III.2. det A × xi = det A × A−1~b)i = det A X ~ A−1 ij bj = X (−1)i+j bj det Mji j j The right hand side is the expansion along column i of the determinant of the matrix constructed from A by replacing column number i with ~b. Proof: Since ~x = A−1~b. the solution to A~x = ~b is xi = det A with ith column replaced by ~b det A  WARNING: This formula is useful. for example. for studying the dependence of A−1~b on matrix elements of A and ~b.

.According to Cramer’s rule  4 1 x1 = det  9 −3 7  1 1 det  1 x2 = −3 2  1 1 x3 = det  1 −3 2    (1) 1 1 4 1 1 −3 1 2 3  = − det  −3 −1 0  (2) − 3(1) = − =1 3 3 3 1 3 2 0 (3) − (1)    4 1 1 4 1 (1) 1 −3 =1 9 3  = − det  0 5 2  (2) − (1) = − 3 3 7 1 0 −1 −1 (3) − 2(1)    1 4 1 1 4 (1) 1 −6 =2 2 9  = − det  0 1 5  (2) − (1) = − 3 3 3 7 0 1 −1 (3) − 2(1) c) The cross product  ˆı ~a × ~b =  a1 b1 Areas and Volumes ˆ a2 b2  kˆ a3  b3 We have aready seen in §I.2 that areas of parallelograms and volumes of parallelopipeds are given by determinants.

 .

~b = . a volume of parallelogram with sides ~a.

.

det 1 b1 .

 .

a1 .

~c = . ~b. volume of parallelopiped with edges ~a.

.

det  b1 .

c1 .

a2 .

.

b2 .

a2 b2 c2 .

a3 .

.

b3 .

.

c3 .

the standard change of variables formula for integrals in more than one dimension involves a determinant. §III. As a result.8 Worked Problems Questions 1) Compute the following matrix products: (a)  1 0 3 5  0 3 1 4  (b) (g) (j)   2 1 0 4  0 1 1 2 x 2  a b −1 4 y 3  −1 0 1 2 1 0    2 1 1 2 0  0 1 0 −1 1 3 0  2 3 (e) [ 3 1 0 ]  3 0  0 5    1 0  1 0 3 5   (d) 2 3 0 1 1 2 5 0     (h)      3 3 5 3 2 2 −1 0  1 0 01 0 0 0 3 2 0 0 2 0 24    1 2 3 0 4 3 (c)  0 0 1   2 1 1  1 0 0 8 0 4   4 (f ) [ 2 3 ] 5 (i)  1 3 5 4  a c b d  .

(c) Why are the last results in parts (a) and (b) the same? 3) Let  1 2 A= 3 −1   2 6 1 B= 1 −1 0  3 2 C = 1 2 4 1   (a) Compute AB. 50% of all two year cars and all three year old cars. (a) Let ~x(n) be the vector which gives the number of cars that are in each age category during year n. the next day. That is. (AB)C. (b) Find all matrices that commute with B. what is its inverse? 7) State whether each of the following statments is true or false. In each case give a brief reason. (b) Can the order of the factors in the product ABC be changed? 4) Let  −1 2 A= 3 −4   1 −1 B= 0 3  (a) Compute A2 + 2AB + B 2 . (a) No square matrix with real entries can obey A2 = −I. Find a 3 × 3 matrix B such that ~x(n + 1) = B~x(n). (b) Compute (A + B)2 . January 1. 2A + 3B and 3(2A + 3B). (b) Compute 6A. it buys enough new cars to replace those sold the previous day. That is. 6) Suppose the matrix B obeys B 7 − 3B + I = 0.2) Let A=  1 2 3 4 5 6  B=  −3 1 2 −3 2 0  (a) Compute 2A. So it is not necessary to bracket ABC. 5) Define A=  1 0 0 −1  B=  0 1 1 0  (a) Find all matrices that commute with A. find all matrices M that obey AM = M A and BM = M B. 8) Suppose that a taxicab company uses the following strategy to maintain a fleet of fixed size: on December 31 of each year it sells 25% of the cars that are one year old (the lemons). find all matrices M that obey AM = M A. (c) The only 2 × 2 matrices that obey A2 = A are A = 0 and A = I. (b) The only 2 × 2 matrix that obeys A2 = 0 is A = 0. (c) Find all matrices that commute with both A and B. 25 . 9B and 6A + 9B. Is B invertible? If so. find all matrices M that obey BM = M B. BC and A(BC) and verify that (AB)C = A(BC). That is. (c) Account for the difference between the answers to parts a and b. 3B.

0. z) = [x + y. ˆ to kˆ and kˆ to −ˆı. z) = −2y + 9z − 12 (c) f (x. Fill in the missing entires. What proportion of the fleet will consist of third year cars in each of the following four years? (c) Find all equilibrium fleet vectors. 15) A solid body is rotating about an axis which passes through the origin and has direction Ω ~ radians per second.  4 3 2 • 4 3 1   5   −2 −2 −1 −1 11 6 4 3  26  • −1 1 0 −3 1 −1   7   −10 5 • • −8 3 • 1  . (e) Are there any fleet vectors which repeat themselves every two years. That is ~x(n + 2) = C~x(n). 3]. 16) Find the matrix of the linear transformation on IR2 which (a) rotates each vector ~x clockwise by 45◦ . −1]. x − y. (c) reflects each vector ~x about the line x = y and then projects the result onto the x–axis. y. What vector does it map [5. Find the axis of rotation. z) = 3x − 2y + 5z (b) f (x. (c) Find the angle of rotation. (b) The linear transformation is a rotation. 3] and [5. (a) Find the matrix of the linear transformation. 4] to [1. y) = [x + 2y. 8] to [3. Find a matrix A such that ~v = A~x. all vectors ~x(n) obeying ~x(n + 1) = ~x(n). x − y. That is. Denote by ~x the coordinates. [3. 2] to [1. y − x. 14) Determine the matrices of the reflections in the following planes in IR 3 (a) x + y + z = 0 (b) 2x − 2y − z = 0 ˆ ~ = Ω1ˆı +Ω2 ˆ+Ω3 k. 7] and [1. y.(b) Suppose the company starts with a fleet consisting of entirely third year cars. z) = [x. 2. z 2 ] 10) Suppose that a linear transformation maps [1. but not every year? 9) Determine whether or not each of the following functions is a linear transformation. 17) The matrix on the left below is the inverse of the matrix on the right below. (d) Show that this linear transformation really is a rotation. (a) f (x. (b) reflects each vector ~x in the x–axis and then rotates the result counterclockwise by 90 ◦ . at some fixed time. y. y) 13) A linear transformation maps ˆı to −ˆ . 1. z) = [x − y − z. y. z) (d) g f (x. y. 1] to [4. 6]? 12) Find the matrices representing the linear transformations (a) f (x. x + y] (b) g(x. y. z) = x 2 + y 2 + z 2 (d) f (x. 14] to? 11) Is it possible for a linear transformation to map [1. z) = [x + y. 2y + 5z − 3x]   (c) f g(x. 0] (e) f (x. 1] (f) f (x. y. y. y. (d) Find a matrix C that moves the fleet ahead two years. of a point The rate of rotation is kΩk fixed to the body and by ~v the velocity vector of the point at that time. −1] to [8.

if possible. the inverses of each of the following matrices. (b) Find the inverse of I − A. (In series ∞ X An n=0 21) Suppose that some square matrix obeys An = 0 for some positive integer. Compute (a) det(4A) (b) det(A 2 ) (c) det(4A2 ) 26) Suppose that the 6 × 6 matrix A obeys A4 = 2A. if it exists. 27) Evaluate 1 a det  2 a a3  a a2 a3 1 27 a2 a3 1 a  a3 1  a a2 .   3 2 p 23) Suppose that det  0 p 1  = 10.) Show that P∞ n=0 (I − A)−1 = An = I + A + A2 + A3 + · · · converges. A0 is defined to be I. for some square matrix A. 25) Suppose that the 3 × 3 matrix A obeys det A = 5. for those values of p. the series notation. Find all possible values of det A. What are the possible values of p? 1 0 2 24) Let 1 0 A= 0 3 3 4 1 ∗  5 0 0 7  ∗ 6  2 8 where the ∗’s denote unknown entries. Prove that its inverse. 20) Suppose that.18) Find. (a)  1 4 2 7    1 2 3 (f )  4 5 6  7 8 9 (b)  2 5 4 8 0 0 (g)  0 3   0 0 2 0 (c) 1 0 0 0  3 4 4 −3   1 2 −1 (d)  2 5 3  1 3 9   0 1  0 0   1 2 3 (e)  2 3 4  3 4 6 19) Let   −1 2 p B =  0 −1 1  2 1 0 (a) For which values of p does B have an inverse? (b) Find B −1 . Find all possible values of det A. (a) Find the inverse of A. is also linear. 22) Suppose that L is a linear transformation from IRn to IRn .

Solutions 1) Compute the following matrix products: (a)  1 0 3 5  0 3 1 4  (b)    1 0  1 0 3 5   (d) 2 3 0 1 1 2 5 0 (g)  2 1 0 4  0 1 1 2 (j)  x 2  a b −1 4 y 3     2 1 1 2 0  0 1 0 −1 1 3 0    2 3 (e) [ 3 1 0 ]  3 0  0 5 −1 0 1 2 1 0  (h)       3 3 5 3 2 2 −1 0  1 0 01 0 0 0 3 0 2 0 2 0    1 2 3 0 4 3 (c)  0 0 1   2 1 1  1 0 0 8 0 4   4 (f ) [ 2 3 ] 5 (i)  1 3 5 4  a c b d  Solution. (a)  1 0 3 5       0 3 1×0+0×1 1×3+0×4 0 3 = = 5 29 3×0+5×1 3×3+5×4 1 4 (b)  (c)       2 1  1×2+2×0+0×3 1×1+2×1+0×0 2 3 1 2 0   = 0 1 = 0×2−1×0+1×3 0×1−1×1+1×0 3 −1 0 −1 1 3 0     1×0+2×2+3×8 1×4+2×1+3×0 1×3+2×1+3×4 0 4 3 1 2 3 0 0 12 1 1 = 0 × 0 + 0 × 2 + 1 × 8 0 × 4 + 0 × 1 + 1 × 0 0 × 3 + 0 × 1 + 1 × 4 1×0+0×2+0×8 1×4+0×1+0×0 1×3+0×1+0×4 8 0 4 1 0 0   28 6 17 =  8 0 4  0 4 3  (d)     1×1+0×0 1×0+0×1 1×3+0×1 1×5+0×2 1 0  2 3 1 0 3 5 = 2×1+3×0 2×0+3×1 2×3+3×1 2×5+ 3×2 0 1 1 2 5×1+0×0 5×0+0×1 5×3+0×1 5×5+0×2 5 0   1 0 3 5 =  2 3 9 16  5 0 15 25  (e)  (f)  2 3 [3 1 0]3 0 = [3 × 2 + 1 × 3 + 0 × 0 3 × 3 + 1 × 0 + 0 × 5] = [9 9] 0 5 [2 3]   4 = [ 2 × 4 + 3 × 5 ] = [ 23 ] 5 28 .

(g)  2 1 0 4  0 1 1 2        −1 0 1 1 4 −1 0 1 7 4 1 = = 2 1 0 4 8 2 1 0 12 8 4 (h)  (i)      3 2    3 2 3 3 5 5 6 10  2 −1 0  41 10 1 0 = 1 0 01 0 = 0 6 0 0 0 3 6 0 2 0 2 0 0 2 0    1 3 5 4  y 3     b a + 3c b + 3d = d 5a + 4c 5b + 4d a c (j) x 2    ax − y a b = 2a − 3 −1 4 bx + 4y 2b + 12  2) Let A=  1 2 3 4 5 6   B= −3 1 2 −3 2 0  (a) Compute 2A. 3B. 2A + 3B and 3(2A + 3B). (c) Why are the last results in parts (a) and (b) the same? Solution. (AB)C. 9B and 6A + 9B.1. 3 5 3(2A + 3B)=3(2A) + 3(3B)=(3 × 2)A + (3 × 3)B = 6A + 9B 3) Let  1 2 A= 3 −1   2 6 1 B= 1 −1 0    3 2 C = 1 2 4 1 (a) Compute AB. we have. So it is not necessary to bracket ABC. (a)          −7 7 12 −9 3 6 2 4 6 −9 3 6 2 4 6 = + 2A + 3B = 3B = −1 16 12 −9 6 0 8 10 12 −9 6 0 8 10 12     −21 21 36 −7 7 12 3(2A + 3B) = 3 = −3 48 36 −1 16 12 2A =  (b)    −27 9 18 6 12 18 9B = 6A = −27 18 0 24 30 36       6 12 18 −27 9 18 −21 21 36 6A + 9B = + = 24 30 36 −27 18 0 −3 48 36  (c) Applying properties 3 and 5 of the “Basic Properties of Matrix Operations” given in §III. BC and A(BC) and verify that (AB)C = A(BC). (b) Can the order of the factors in the product ABC be changed? 29 . (b) Compute 6A.

4) Let A=  −1 2 3 −4  B=  1 −1 0 3  (a) Compute A2 + 2AB + B 2 . (a)  4 4 1 AB = 5 19 3  16 17 BC = 2 0      3 2  4 4 1  20  1 2 = (AB)C = 5 19 3 46 4 1     20 1 2 16 17 A(BC) = = 46 3 −1 2 0  17 51 17 51   Yup. (c) Account for the difference between the answers to parts a and b.Solution. (b) Compute (A + B)2 . (a) A 2 2AB B2 A2 + 2AB + B 2     7 −10 −1 2 −1 2 = = −15 22 3 −4 3 −4        −1 2 1 −1 −1 7 −2 14 =2 =2 = 3 −4 0 3 3 −15 6 −30      1 −1 1 −1 1 −4 = = 0 3 0 3 0 9         6 0 1 −4 −2 14 7 −10 = + + = −9 1 0 9 6 −30 −15 22  (b)      0 1 1 −1 −1 2 = + A+B = 3 −1 0 3 3 −4      0 1 0 1 3 −1 2 (A + B) = = 3 −1 3 −1 −3 4  (c) (A + B)2 = A(A + B) + B(A + B) = A2 + AB + BA + B 2 . They’re the same. 5) Define A=  1 0 0 −1  B= 30  0 1 1 0  . (b) The matrix product BAC is not defined because the matrix product BA is not defined owing to a matrix size mismatch. Solution. so the answer to part a minus the answer to part b ought to be       −1 2 1 −1 1 −1 −1 2 − 3 −4 0 3 0 3 3 −4       3 1 −4 6 −1 7 = − = −6 −3 9 −12 3 −15 (A2 + 2AB + B 2 ) − (A2 + AB + BA + B 2 ) = AB − BA = This is indeed the difference.

For example A=  0 a b 0  =⇒ So if ab = −1 (e. we need b = c = 0 and a = d. So we need M to be of the form numbers a and b. find all matrices M that obey AM = M A. find all matrices M that obey BM = M B. (b) The only 2 × 2 matrix that obeys A2 = 0 is A = 0. a = 1. 31 A2 =  ab 0 0 ab  .g. So   a 0 for some numbers a and d. find all matrices M that obey AM = M A and BM = M B. That is. (c) Find all matrices that commute with both A and B. (a) In order for both AM and M A to be defined. That is. (b) Find all matrices that commute with B. what is its inverse? Solution. Let M= Then  a c b d       1 0 a b a b = 0 −1 c d −c −d      a b 1 0 a −b MA = = c d 0 −1 c −d AM = These are the same if and only if b = −b and c = −c which in turn is the case if and only if b = c = 0. 0 1 6) Suppose the matrix B obeys B 7 − 3B + I = 0. Let M= Then BM = MB =   a c b d     c d b = a b d     b a 0 1 b = d c 1 0 d 0 1 1 0 a c   a c These are the same if and only if c = b and a = d.  a b b a  for some (c) In order to satisfy the conditions of both part a and part b. we need M to be of the form 0 d (b) In order for both BM and M B to be defined. That is. so M must   1 0 be of the form a for some number a. M must be a 2 × 2 matrix. 7) State whether each of the following statments is true or false. (c) The only 2 × 2 matrices that obey A2 = A are A = 0 and A = I. the matrix B is invertible with inverse 3I − B 6 .(a) Find all matrices that commute with A. Solution. Is B invertible? If so. Solution. In each case give a brief reason. (a) This statement is false . A2 = −I. M must be a 2 × 2 matrix. Because B(3I − B 6 ) = 3B − B 7 = I. b = −1). (a) No square matrix with real entries can obey A2 = −I.

So x1 (n + 1) = . it buys enough new cars to replace those sold the previous day. (a) Pretend that it is now December 31 of year n. During year n + 1.25x1 (n) + . So x2 (n + 1) = . these cars are all between one and two years old.0x3 (n) 2 = x2 (n + 1) = .g. January 1. (a) Let ~x(n) be the vector which gives the number of cars that are in each age category during year n. A2 = A. Find a 3 × 3 matrix B such that ~x(n + 1) = B~x(n). During year n + 1.50x2 (n) + 1.25x1 (n) + .0x3 (n) Of the x1 (n) cars that had their first birthday on January 1. b = 0). x(n)2 cars that are two years old and x(n)3 cars that are three years old.g. b = 0). That is. but not every year? Solution.0x3 (n).50x2 (n) + 1. all vectors ~x(n) obeying ~x(n + 1) = ~x(n). (b) Suppose the company starts with a fleet consisting of entirely third year cars.0x3 (n) cars. What proportion of the fleet will consist of third year cars in each of the following four years? (c) Find all equilibrium fleet vectors. 8) Suppose that a taxicab company uses the following strategy to maintain a fleet of fixed size: on December 31 of each year it sells 25% of the cars that are one year old (the lemons). these cars are all between zero and one year old. B~x(n) B~x(n) B~x(n)    1 = x1 (n + 1) = . So.25x1 (n) + .75x1 (n) Similarly. (d) Find a matrix C that moves the fleet ahead two years. January 1 of year n + 1 we buy the same number.50x2 (n) In summary. We have x(n)1 cars that are (almost) one year old. The next day.50x2 (n) + 1. a = 1. .25x1 (n) + . For example A=  0 a b 0  =⇒ A2 = ab 0 0 ab  =⇒ a2 A = 0 0 b2   So if ab = 0 (e. (e) Are there any fleet vectors which repeat themselves every two years. That is ~x(n + 2) = C~x(n). A2 = 0.(b) This statement is false . on December 31 of year n. of replacement cars.75x1 (n) 3 = x3 (n + 1) = .75x1 (n) remain with the company. a = 1. 50% of all two year cars and all three year old cars. . we sell . For example  a 0 A= 0 b  2  So if a2 = a and b2 = b (e. the next day. x3 (n + 1) = .50x2 (n) + 1.50x2 (n) 32 . (c) This statement is false .

25 . 3 · · · 3 (d) We wish to find a matrix C obeying ~x(n + 2) = C~x(n).5 1 16 q 64 q 3       ~x(5) = B~x(4) = . γ has to be a positive multiple of 3.5 1 ~x(2) = B~x(1) =  .25 . 32    1  q .75 .75 0 0 = 21 16 q 64 q 3 6 0 .5 1 4q 16 q 3  3      = 16 q ~x(4) = B~x(3) = . Let γ = 3p. α. (c) In equilibrium (using I to denote the identity matrix) ~x(n + 1) = ~x(n) B~x(n) = I~x(n) (B − I)~x(n) = 0 If we call the components of ~x(n).5 1 α 0 (B − I)~x(n) =  . 3 3 8 .So   .5 0 q q  8 64 .5 0 0      q 0 . β. So    1/4 2/4 1 1/4 2/4 1 C = B 2 =  3/4 0 0   3/4 0 0  = 0 2/4 0 0 2/4 0 33   7 10 4 1  3 6 12  16 6 0 0 .5 1 4q ~x(3) = B~x(2) =  . γ respectively      −. As ~x(n + 1) = B~x(n) and ~x(n + 2) = B~x(n + 1).75 0 0  0 . we have ~x(n + 2) = B~x(n + 1) = B 2 ~x(n).75 0 0   0  =  34 q  0 0 . 2.5 0    7   37  .75 0 0 4q 3 0 . (b) If there are q cars in the fleet.5 −1 γ 0 The last equation forces β = 2γ and the second equation forces α = 43 β = 38 γ.5 0 0 q  8 The proportions are 0.75 0 0   0  =  0  0 q 0 .25 . The general solution is   8 ~x(n) = p  6  . In order to have an integer numbers of cars. p = 0.5 0 does the job. 1. we start with   0 ~x(1) =  0  q so that  1   7  .25 .25 .75 −1 0   β  =  0  0 . 0.5 1 B =  .

y. ~x0 to stand for [x0 . z) = −2y + 9z − 12 (c) f (x. z) = [x + y. 9) Determine whether or not each of the following functions is a linear transformation. We use the notations ~x to stand for [x. β in IR. the two expressions f (α~x + β~x0 ) = −2(αy + βy 0 ) + 9(αz + βz 0 ) − 12 αf (~x) + βf (x~0 ) = α(−2y + 9z − 12) + β(−2y 0 + 9z 0 − 12) have to be equal for all ~x. y. y 0 . These are equal only when α + β = 1. (b) f (~x) = −2y + 9z − 12 is a function from IR3 to IR. y. ~x0 in IR3 and all α. They aren’t. β in IR. So this f is not linear . For it to be linear. 0] (f) f (x. z 0 ] and f (~x) to stand for f (x. p = 0. z) = [x + y. For it to be linear. The general solution is once again   8 ~x(n) = p  6  . The last equation minus twice the second equation forces 20β − 40γ = 0 or β = 2γ just as in part c. because when y = z = x0 = y 0 = z 0 = 0 the first expression reduces to α2 x2 and the second reduces to αx2 . For it to be linear. ~x0 in IR3 and all α. y.(e) If C~x(n) = ~x(n). because when x = y = z = x0 = y 0 = z 0 = 0 the first expression reduces to −12 and the second reduces to −12(α + β). 1] Solution. αz + βz 0 ]. y. z) = 3x − 2y + 5z (b) f (x. z) = [x. x − y. So this f is not linear . just as in part c. z 2 ] (e) f (x. 1. 34 . y. (c) f (~x) = x2 + y 2 + z 2 is a function from IR3 to IR. y. y. 2. z). (a) f (x. ~x0 in IR3 and all α. They aren’t. These are equal only when α2 = α or when x = 0. (a) f (~x) = 3x − 2y + 5z is a function from IR3 to IR. so this f is linear . Observe that α~x + β~x0 = [αx + βx0 . 3 · · · 3 No fleet vector repeats itself every two years but not every year. then      −9 10 4 α 0 1  (C − I)~x(n) = 3 −10 12   β  =  0  16 6 0 −16 γ 0 The last equation forces α = 83 γ. y. the two expressions f (α~x + β~x0 ) = (αx + βx0 )2 + (αy + βy 0 )2 + (αz + βz 0 )2 2 2 2 αf (~x) + βf (x~0 ) = α(x2 + y 2 + z 2 ) + β(x0 + y 0 + z 0 ) have to be equal for all ~x. x − y. z]. the two expressions f (α~x + β~x0 ) = 3(αx + βx0 ) − 2(αy + βy 0 ) + 5(αz + βz 0 ) αf (~x) + βf (x~0 ) = α(3x − 2y + 5z) + β(3x0 − 2y 0 + 5z 0 ) have to be equal for all ~x. They are. β in IR. αy + βy 0 . z) = x 2 + y 2 + z 2 (d) f (x.

1] − 29 [1. z 2 ] is a function from IR3 to IR3 . 0] αf (~x) + βf (x~0 ) = α[x + y. For it to be linear. 1] = [αx + αy + βx0 + βy 0 . 10) Suppose that a linear transformation maps [1. (f) f (~x) = [x. 2  = [2. 7] and [1. 0] + 14[0. 14] = 5[1. They are. So this f is not linear . −1] to [8. 1] + [1. 0] is a function from IR3 to IR3 .(d) f (~x) = [x + y. If z 0 = 0 the third component of the first expression reduces to α2 z 2 while that of the second expression reduces to αz 2 . x0 − y 0 . β in IR. 3] = 4 106 2. ~x0 in IR3 and all α. 7] − 92 [8. z 2 ] + β[x0 . x − y. z 0 ] 2 = [αx + βx0 . y. ~x0 in IR3 and all α. (αx + βx0 ) − (αy + βy 0 ). For it to be linear. 0] + β[x0 + y 0 . the two expressions f (α~x + β~x0 ) = [(αx + βx0 ) + (αy + βy 0 ). rather than row vectors. the two expressions f (α~x + β~x0 ) = [(αx + βx0 ) + (αy + βy 0 ). This matrix must obey  a c b d       4 a+b 1 = = 7 c+d 1  35 a c b d       8 a−b 1 = = 3 c−d −1 . β in IR. so this f is linear . What vector does it map [5. 53] Solution 2. αy + βy 0 . The third components are equal only for α + β = 1. αy + βy 0 . x − y. (e) f (~x) = [x + y. For it to be linear. so this f is not linear . Write the vectors as column. β in IR. 3]. 1] − [1. −1]) = 19 2 [1. −1] is mapped to 19 2 [4. y. the two expressions f (α~x + β~x0 ) = [αx + βx0 . 1] αf (~x) + βf (x~0 ) = α[x + y. ~x0 in IR3 and all α. The vector [5. Let  a c b d  be the matrix of the linear transformation. 1] + β[x0 + y 0 . These agree only for z = 0 or α = 0. (αx + βx0 ) − (αy + βy 0 ). y 0 . 0] = [αx + αy + βx0 + βy 0 . αx − αy + βx0 − βy 0 . (αz + βz 0 )2 ] 2 αf (~x) + βf (x~0 ) = α[x. αz 2 + βz 0 ] have to be equal for all ~x. 1. x − y. 1] is a function from IR3 to IR3 . x0 − y 0 . 0] have to be equal for all ~x. 1] = 25 ([1. αx − αy + βx0 − βy 0 . 14] to? Solution 1. x − y. α + β] have to be equal for all ~x. −1]) + 14 2 ([1. 1] to [4.

8] = 2[1.  a c b d          2 5 6 −2 5 30 − 28 = = = 53 14 5 2 14 25 + 28 11) Is it possible for a linear transformation to map [1. Both systems are easy to solve a+b=4 a−b=8 =⇒ a=6 b = −2 c+d =7 c−d=3 =⇒ c=5 d=2 Now that we know the matrix of the linear transformation. 4] to [1. 2y + 5z − 3x]   (c) f g(x. e = 2. rather than row vectors. 0. 0. x + y] (b) g(x. Writing column. −1] + [1. it is not possible . [5. f = 1. b = −1. 6] = 2[1. So for the map to be linear. it is necessary that [3. 2] to [1. c = −1. So the matrix is  1 −1 −1 −3 2 5  (c) Substituting into       x    x 1 −1 −1 u y = g  y   = −3 2 5 v z z     1 2   u u f =  −1 1  v v 1 1 36 . d = 1. (a)       x + 2y a b   ax + by  y − x  =  c d  x =  cx + dy  y x+y e f ex + f y provided we choose a = 1. So. 2. 2] + [3. So the matrix is   1 2  −1 1  1 1 (b)       x ax + by + cz a b c   x−y−z y = = dx + ey + f z d e f 2y + 5z − 3x z   provided we choose a = 1. we just have to apply it to the specified input vector. 6]? Solution. y) Solution.This consists of a system of two equations in the two unknowns a and b as well as a system of two equations in the two unknowns c and d. 1]. 3] = [3. f = 5. 8] to [3. y) = [x + 2y. which is false. y. b = 2. z) (d) g f (x. d = −3. 1. [3. 1. 12) Find the matrices representing the linear transformations (a) f (x. z) = [x − y − z. c = −1. e = 1. −1]. 2. 2. 4]. 3] and [5. y − x. y.

(d) Show that this linear transformation really is a rotation. rotate by angles that are strictly 37 . Solution. Observe that the bottom and top edges of the book rotate by 45◦ . Rotate the book. ˆ to kˆ and kˆ to −ˆı. we just need to find a nonzero vector ~n obeying T ~n = ~n. (b) The linear transformation is a rotation. Observe that the spine of the book does not move at all. Under a rotation of θ ◦ . about its spine. by 45 ◦ . (c) Find the angle of rotation. Vectors lying on the axis of rotation. do not change when the rotation is executed. vectors perpendicular to the axis of rotation rotate by θ ◦ . about its spine.gives            1 2  x 1 2 x 1 f g  y  =  −1 1  g  y  =  −1 1  −3 1 1 z 1 1 z        x −5 3 9 x −1 −1   y =  −4 3 6   y  2 5 z −2 1 4 z (d) Substituting       u 1 2   x x v=f =  −1 1  y y w 1 1 into     u  u 1 −1 −1 v g  v   = −3 2 5 w w  gives    1 2               x 1 −1 −1  x 1 −1 −1 x 1 0 x = −1 1  = f = g f y −3 2 5 y −3 2 5 y 0 1 y 1 1 13) A linear transformation maps ˆı to −ˆ . (a) The matrix must obey      a b c 1 0  d e f   0  =  −1  g h i 0 0      a b c 0 0 d e f 1 = 0 g h i 0 1      a b c 0 −1 d e f 0 =  0  g h i 1 0 The desired matrix is  0 0 −1 T =  −1 0 0  0 1 0  (b) Perform the following little experiment. (a) Find the matrix of the linear transformation. Take a book. Find the axis of rotation. To find the axis of rotation of T . by 45◦ again.     n1 0 0 −1 n1  −1 0 0   n2  =  n2  n3 n3 0 1 0  =⇒    n1 −n3  −n1  =  n2  n3 n2  =⇒    n1 1  n2  = c  −1  for any c n3 −1  (c) Rotate the book. Vectors that are neither perpendicular to nor parallel to the axis of rotation.

(Repeat the book experiment a few times. The lengths k(a + b)ˆı + aˆ ˆ = This vector is again perpendicular to ˆı − ˆ − k. between ˆı + ˆ and the vector −ˆ  + kˆ that it is mapped to obeys ˆ = kˆı + ˆk k − ˆ + kk ˆ cos θ (ˆı + ˆ) · (−ˆ  + k) =⇒ 2 cos θ = −1 =⇒ θ = 120◦ Remark. The (a + b)2 + a2 + b2 and k − bˆı − (a + b)ˆ  + akk dot product ˆ · [−bˆı − (a + b)ˆ ˆ = −b(a + b) − a(a + b) + ab [(a + b)ˆı + aˆ  + bk]  + ak] = −a2 − b2 − ab ˆ − bˆı − (a + b)ˆ ˆ = − 21 k(a + b)ˆı + aˆ  + bkkk  + akk shows that the angle between the two vectors is 120◦ . That is. ˆ has the same length as (a + b)ˆı + aˆ ˆ ˆı − ˆ− k. The vector (a + b)ˆı + aˆ  + bkˆ gets mapped to      0 0 −1 a+b −b  −1 0 0   a  =  −a − b  0 1 0 b a ˆ by the dot product test. θ.e. that every vector parallel to ˆı −ˆ  − kˆ gets mapped to itself.  + bkk p p p ˆ = b2 + (a + b)2 + a2 both equal 2(a2 + b2 + ab). in part b. because [1. vector perpendicular to ˆı − ˆ− k.  +bkˆ obeying (cˆı +aˆ  +bk)·(ˆ ˆ should get mapped to a vector which is perpendicular to i. If the linear transformation really is a rotation then every ˆ i. every vector of the form (a + b)ˆı + aˆ  + bk.between 0◦ and θ◦ .  + bkˆ and makes an angle 120◦ with respect to (a + b)ˆı + aˆ  + bk. 14) Determine the matrices of the reflections in the following planes in IR 3 (a) x + y + z = 0 (b) 2x − 2y − z = 0 Solution. 1. −1. concentrating on vectors that are almost parallel to the spine and then on vectors that are almost perpendicular to the spine.e. We have already seen. −1]. −1. the linear transformation does not move it at all. ~v . (a) The vector ~n = ˆı + ˆ + kˆ is perpendicular to the given plane. [1. every vector cˆı +aˆ ˆ ı − ˆ− k) ˆ = c−a−b = 0.) So to determine the angle of rotation. we select a vector. The projection of any vector ~x on ~n is   1 ~n · ~x x+y+z   proj~n (~x) = ~ n = 1 k~nk2 3 1 38 . perpendicular to the axis of rotation. The vector ~v = ˆı + ˆ is perpendicular to the axis of rotation. and compute the angle between ~v and T~v. 0] · [1. −1] = 0. to convince yourself that this is true. It gets mapped to      0 0 −1 1 0  −1 0 0   1  =  −1  0 1 0 0 1 The angle.

should be mapped to themselves. ˆ both of which are parallel to the plane. Denote by ~x the coordinates. should be mapped to themselves.      2 2 1 8 4 1 8 1 −4   −2  = −  −2  9 −1 −1 4 −4 7      1 1 1 8 4 1 8 1 −4   1  =  1  9 0 0 4 −4 7      1 8 4 1 1 1 8 1 −4   0  =  0  9 4 −4 7 2 2 ˆ ~ = Ω1ˆı +Ω2 ˆ+Ω3 k. which is perpendicular to the plane. Find a matrix A such that ~v = A~x. the vectors ˆı − ˆ and ˆ − k. 39 .      1 −2 −2 1 1 1 −2 1 −2   1  = −  1  3 −2 −2 1 1 1      1 −2 −2 1 1 1 −2 1 −2   −1  =  −1  3 −2 −2 1 0 0      0 0 1 −2 −2 1 −2 1 −2   1  =  1  3 −1 −1 −2 −2 1 (b) The vector ~n = 2ˆı − 2ˆ  − kˆ is perpendicular to the given plane. ˆ both of which are parallel to the plane. of a point The rate of rotation is kΩk fixed to the body and by ~v the velocity vector of the point at that time. 15) A solid body is rotating about an axis which passes through the origin and has direction Ω ~ radians per second. On the other hand.The reflection is refl(~x)           x 2 x − 2y − 2z 1 −2 −2 x 1 x + y + z 1  2  =  y − 2x − 2z  =  −2 1 −2   y  ~x−2proj~n (~x) =  y − proj~n (~x) 3 3 3 z 2 z − 2x − 2y −2 −2 1 z ~x Check: The vector ~n. On the other hand. should be mapped to −~n. at some fixed time. The projection of any vector ~x on ~n is   2 ~n · ~x 2x − 2y − z  proj~n (~x) = −2  ~n = k~nk2 9 −1 The reflection is           x 2 x + 8y + 4z 1 8 4 x 2x − 2y − z 1 1  −2  =  y + 8x − 4z  =  8 1 −4   y  ~x − 2proj~n (~x) =  y  − 2 9 9 9 z −1 7z + 4x − 4y 4 −4 7 z Check: The vector ~n. should be mapped to −~n. the vectors ˆı + ˆ and ˆı + 2k. which is perpendicular to the plane.

ˆı is mapped to ˆ and ˆ is mapped to ˆı.7 that the velocity vector is  ˆı ~ × ~x =  Ω1 ~v = Ω x ˆ Ω2 y     Ω2 z − Ω 3 y 0 kˆ Ω3  =  −Ω1 z + Ω3 x  =  Ω3 Ω1 y − Ω 2 x −Ω2 z −Ω3 0 Ω1    Ω2 x −Ω1   y  z 0 16) Find the matrix of the linear transformation on IR2 which (a) rotates each vector ~x clockwise by 45◦ . So. (c) The vector ˆı is reflected to ˆ and then projected to ~0 and the vector ˆ is reflected to ˆı and then projected to ˆı. So.  4 3 2 • 4 3 1   5   −2 −2 −1 −1 11 6 4 3   • −1 1 0 −3 1 −1   7   −10 5 • • −8 3 • 1  Solution. 2 The matrix  does the job. Fill in the missing entires.Solution. 17) The matrix on the left below is the inverse of the matrix on the right below. The matrix  0 1 1 0  does the job. We saw in §I. The matrix  0 1 0 0  does the job. (b) reflects each vector ~x in the x–axis and then rotates the result counterclockwise by 90 ◦ . (a) The vector ˆı should be mapped to √1 (ˆ ı − ˆ) 2 and the vector ˆ should be mapped to √1 2  1 1 −1 1 √1 (ˆ ı + ˆ). The matrices must obey    1 b −1 1 0 4 3 2 a 7 −3 1 −1 5 4 3 1  0   =   0 −10 5 c d −2 −2 −1 −1 0 −8 3 e 1 11 6 4 3  40 0 1 0 0 0 0 1 0  0 0  0 1 . Solution. in the end. (c) reflects each vector ~x about the line x = y and then projects the result onto the x–axis. ˆı is mapped to ~0 and ˆ is mapped to ˆı. in the end. (b) The vector ˆı is reflected to ˆı and then rotated to ˆ and the vector ˆ is reflected to −ˆ  and then rotated to ˆı.

1) : =⇒ d=1 (a)(1) = 0 =⇒ a=1 (−2)(b) + (−2)(7) + (−1)(−10) + (−1)(−8) = 0 =⇒ b=2 (2. e = −3 . 4) : (4)(0) + (3)(−1) + (3. (a)  1 4 2 7  (b)   1 2 3 (f )  4 5 6  7 8 9  2 5 4 8 0 0 (g)  0 3   0 0 2 0  (c) 1 0 0 0 3 4 4 −3  1 2 −1 (d)  2 5 3  1 3 9    0 1  0 0  1 2 3 (e)  2 3 4  3 4 6  Solution. if possible. b = 2. just for practice. We have a canned formula for the inverse of 2 × 2 matrices. using (m.In particular. d = 1. c = −2. But I’ll use row reduction. 18) Find. 3) : (−2)(1) + (−2)(1) + (1)(e) = 0 =⇒ 3c + e = −9 (−1)(c) + (−1)(e) = 1 =⇒ c + e = −5 The solution is a = 1. n) to denote the equation that matches the matrix element in row m and column n of the product of the two matrices on the left with the corresponding matrix element of the identity matrix on the right (3. (a)  Check: . 4) : (−2)(0) + (−2)(−1) + (1. the inverses of each of the following matrices. 3) : (5)(1) + (−1)d + (−1)(1) = 0 (2)d + (4)(1) + (3)(c) + (3.

 1 4 .

.

1 0 2 7.

0 1 (1) (2) − 2(1) .

 1 4 .

.

1 0 0 −1 .

−2 1   1 4 2 7  .

 2 5 .

.

1 0 0 −2 .

−2 1  (1) + 4(2) −(2) "    −7 4 1 0 = 2 −1 0 1 # .

1 0 .

.

−7 4 0 1 .

2 −1 (b)  Check: .

 2 5 .

.

1 0 4 8.

0 1 (1) (2) − 2(1)  (1)/2 + 5(2)/4 −(2)/2    1 0 −2 5/4 = 0 1 1 −1/2  2 5 4 8  .

 1 4/3 .

.

1/3 0 0 −25/3 .

−4/3 1 " # .

1 0 .

.

−2 5/4 0 1 .

1 −1/2 (c)  Check: .

 3 4 .

.

1 0 4 −3 .

0 1 (1)/3 (2) − 4(1)/3  3 4 4 −3  (1) + 4(2)/25 −3(2)/25    1 0 3/25 4/25 = 0 1 4/25 −3/25 41 " # .

1 0 .

.

3/25 4/25 0 1 .

4/25 −3/25 .

(d) .

 1 2 −1 .

.

1 0 0  2 5 3 .

0 1 0 .

1 3 9 .

0 0 1 .

  (1) 1 2 −1 .

.

1 0 0 (2) − 2(1)  0 1 5 .

.

−2 1 0  (3) − (1) 0 1 10 .

−1 0 1  .

  (1) + (3)/5 1 2 0 .

.

6/5 −1/5 1/5 (2) − (3)  0 1 0 .

.

−3 2 −1  (3)/5 0 0 1 .

1/5 −1/5 1/5 Check: .

  0 0 (1) 1 2 −1 .

.

1  0 1 5 .

−2 1 0  (2) .

(3) − (2) 0 0 5 .

1 −1 1   .

(1) − 2(2) 1 0 0 .

.

36/5 −21/5 11/5   2 −1  (2)  0 1 0 .

.

−3 (3) 0 0 1 .

1/5 −1/5 1/5      1 2 −1 36/5 −21/5 11/5 1 0 0  2 5 3   −15/5 10/5 −5/5  =  0 1 0  1 3 9 1/5 −1/5 1/5 0 0 1 (e) .

 1 2 3 .

.

1 0 0  2 3 4.

0 1 0 .

3 4 6.

0 0 1  .

  (1) 1 2 3 .

.

1 0 0 (2) − 2(1)  0 −1 −2 .

.

−2 1 0  (3) − 3(1) 0 −2 −3 .

−3 0 1 .

  1 2 0 .

.

−2 6 −3 (1) − 3(3) 3 −2  −(2) − 2(3)  0 1 0 .

.

0 0 0 1 .

1 −2 1 (3) Check:   .

1 (1) − 2(2) 1 0 0 .

.

−2 0   3 −2  (2)  0 1 0 .

.

0 (3) 0 0 1 .

1 −2 1   1 2 3 −2 0 2 3 4 0 3 3 4 6 1 −2 (f) .

 1 2 3 .

.

1 0 0  4 5 6.

0 1 0 .

7 8 9.

0 0 1    1 1 0 0 −2  =  0 1 0  1 0 0 1 .

  (1) 1 2 3 .

.

1 0 0 (2) − 4(1)  0 −3 −6 .

.

−4 1 0  (3) − 7(1) 0 −6 −12 .

−7 0 1  .

  (1) 0 0 1 2 3 .

.

1  0 −1 −2 .

−2 1 0  (2) .

(3) − 2(2) 0 0 1 .

1 −2 1 .

  (1) 1 2 3 .

.

1 0 0  0 −3 −6 .

−4 1 0  (2) .

(3) − 2(2) 0 0 0 .

observe that      1 2 3 1 0  4 5 6   −2  =  0  7 8 9 1 0 No invertible matrix can map a nonzero vector to the zero vector. There is no inverse . As a check. 1 −2 1 The last equation has a zero left hand side and nonzero right hand side and so cannot be satisfied. (g) 0 0  0 3  Check: 0 0 2 0 1 0 0 0 .

0.

.

1.

.

0.

.

0 0 0  0 3  1 0 0 0 0 0 2 0 0 1 0 0 1 0 0 0 0 0 1 0  0 0  0 1  0 0 10  0 1 0 0  (4)/3 1 0 0 (3)/2  0 1 0 (1)  0 0 1 0 0 0 (2) .

0.

.

0.

.

0.

.

1 0 0 1 0   1 0 0 0 1/3 0 1/2 0   0 1 = 0 0 0 0 0 0 0 1 0 0 42  0 0 1/3 0 1/2 0   0 0 0  1 0 0 0 0 1 0  0 0  0 1 .

for those values of p. .19) Let  −1 2 p B =  0 −1 1  2 1 0  (a) For which values of p does B have an inverse? (b) Find B −1 . Solution.

  −1 2 p .

.

1 0 0 (1) −1 2  0 −1 1 .

0 1 0   0 −1 (2) .

2 1 0.

0 0 1 (3) + 2(1) 0 5 .

  (1) −1 2 p .

.

1 0 0  0 −1 (2) 1 .

.

0 1 0  (3) + 5(2) 0 0 2p + 5 .

2 5 1  .

 p .

.

1 0 0 1 .

.

0 1 0  2p .

In this case. 2 0 1 There is an inverse if and only if 2p + 5 6= 0 . we can continue .

 1 −2 −p .

.

−1 −(1)  0 1 −1 .

0 −(2) .

2 1 .

2p+5 (3)/(2p + 5) 0 0  .

(1) + 2(2) 1 0 0 .

.

 (2)  0 1 0 .

.

(3) 0 0 1.

5 2p+5 p 2p+5 −2p 2p+5 5 2p+5 −1 2p+5 2 2p+5 2 2p+5  .

(1) + p(3) 1 −2 0 .

.

 (2) + (3)  0 1 0 .

.

(3) 0 0 1.

n=0 43 n=1 .  0 0  0 −1 1 2p+5 p+2 2p+5 1 2p+5 1 2p+5  P∞ n=0 (I − A)−1 = For the left hand side (I − A) The second sum P ∞ n=0 ∞ X An n=0 n=0 The argument for n=0 n=0 n=0 n=0 An+1 = A + A2 + A3 + · · · = (I − A) P n=0  P  ∞ An = An (I − A) = I  ∞ ∞ ∞ ∞ P P P P An = An − A An = An − An+1 ∞ P so P ∞ P ∞ n=0 p 2p+5 1 2p+5 1 2p+5    An = I + A + A2 + A3 + · · · converges. (In series Solution. the series notation. We have to verify that (I − A) 5p 2p+5 −2p 2p+5 5 2p+5   20) Suppose that. for some square matrix A.) Show that −5 2p+5 2 2p+5 2 2p+5 ∞ P n=0 An n=1  ∞ ∞ P P An = An − An = I n=0  ∞ n A (I − A) is similar. A0 is defined to be I.

Solution. which in turn implies M (α~x + β~x0 ) = α~u + β~u0 = αM (~x) + βM (~x0 ). Let ~x and ~x0 be two arbitrary vectors in IRn and let α and β be two arbitrary numbers. is also linear. (b) From problem 20. All the remaining terms in the series I + A + A2 + · · · vanish because of An = 0.  3 23) Suppose that det  0 1 Solution. Prove that its inverse. if it exists. Thus. We are to assume that there exists a map M such that L(~x) = y~ if and only if M (~y ) = ~x. M is linear . Find all possible values of det A. What are the possible values of p? 0 2 along the last row       3 2 p 3 2 2 p det  0 p 1  = det + 2 det 0 p p 1 1 0 2 = (2 − p2 ) + 2(3p − 0) = −p2 + 6p + 2 For this to be 10 −p2 + 6p + 2 = 10 ⇐⇒ p2 − 6p + 8 = 0 ⇐⇒ p = 2. 44 . (b) Find the inverse of I − A. we would guess (I − A)−1 = I + A + A2 + · · · + An−1 . To verify that the guess is correct we multiply out (I − A)(I + A + A2 + · · · + An−1 ) = (I + A + A2 + · · · + An−1 ) − A(I + A + A2 + · · · + An−1 ) = (I + A + A2 + · · · + An−1 ) − (A + A2 + A3 + · · · + An ) = I − An = I The same argument also gives (I + A + A2 + · · · + An−1 )(I − A) = I 22) Suppose that L is a linear transformation from IRn to IRn . As L is linear L(α~u + β~u0 ) = α~x + β~x0 . 4 24) Let 1 0 A= 0 3  3 4 1 ∗ 5 0 0 7  ∗ 6  2 8 where the ∗’s denote unknown entries. (a) Trick question!! A has no inverse. ~u = M (~x) ⇒ ~x = L(~u) and ~u0 = M (~x0 ) ⇒ ~x0 = L(~u0 ). Solution.21) Suppose that some square matrix obeys An = 0 for some positive integer. By hypothesis. (a) Find the inverse of A. Define ~u = M (~x) and ~u0 = M (~x0 ). Expanding  2 p p 1  = 10. If A had an inverse then multiplying both sides of A n = 0 by (A−1 )n would give A−1 · · · A−1 A · · · A = 0 (with n A−1 ’s and n A’s) and then I = 0.

assuming that A has real matrix entries. 25) Suppose that the 3 × 3 matrix A obeys det A = 5. Find all possible values of det A. So   0 = (det A4 ) − 64 det A = det A (det A)3 − 64 Consequently det A = 0 or (det A)3 = 64 or det A = 0.     4 0 0 4 0 0 det(4A) = det  0 4 0  A = det  0 4 0  det A = 43 × 5 = 320 0 0 4 0 0 4  det(A2 ) = det A det A = 5 × 5 = 25 det(4A2 ) = det(4A) det A = 320 × 5 = 1600 26) Suppose that the 6 × 6 matrix A obeys A4 = 2A.Solution. 1 0 det  0 3    ∗ 1 6 0  = det  2 0 8 0   4 0 6 0 = det  1 0 2  = det  1 ∗ −8 ∗ ∗  3 4 1 ∗ 5 0 0 7  1 0 = −2 det ∗ −8  3 5 4 0 1 0 ∗ −8  ∗ (1) 6 (2)  2 (3) ∗ (4) − 3(1)  0 −2 (1) − 4(2) 0 2  (2) −8 ∗ (3) = 16 for all values of the ∗’s. 4 . 45 . Compute (a) det(4A) (b) det(A 2 ) (c) det(4A2 ) Solution. Solution. third and fourth equalities. 27) Evaluate 1 a det  2 a a3  Solution. A4 = 2A =⇒ det(A4 ) = det(2A) =⇒ (det A)4 = det(2I) det A As in the last question det(2I) = 26 = 64. a a2 a3 1 a2 a3 1 a  a3 1  a 2 a    1 a a 2 a3 1 a a2 a3 (1) 2 3 4 1 0 0 1 − a  (2) − a(1) a a a 0 det  2 = det    a a3 1 a 0 0 1 − a4 a − a5 (3) − a2 (1) 3 2 4 5 2 6 a 1 a a 0 1−a a−a a −a (4) − a3 (1)     0 0 1 − a4 0 1 − a4 4 4 5   = det 0 1−a a−a = (1 − a ) det 1 − a4 a − a5 1 − a 4 a − a 5 a2 − a 6    = (1 − a4 ) − (1 − a4 )(1 − a4 ) = −(1 − a4 )3 = (a4 − 1)3 We expanded along the first row to achieve each of the the second.