Algebra, II, m2p2
Algebra, II, m2p2
Algebra, II, m2p2
Algebra 2
ii
These notes are based on a course of lectures given by Professor Liebeck during Spring Term 2006 at Imperial College London. In general the notes follow the lectures very closely and only few changes were made mostly to make the typesetting easier. These notes have not been checked by Professor Liebeck and should not be regarded as ofcial notes for the course. In particular, all the errors are made by me. However, I dont take any responsibility for their consequences; use at your own risk (and attend lectures, they are fun). Please email me at as1005@ic.ac.uk with any comments or corrections. Anton Stefanek October 2008
CONTENTS
iii
Contents
I Groups
1 Introduction 2 Symmetry groups 2.1 Symmetry groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 More on D 8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1
3 5 5 7
3 Isomorphism 9 3.1 Cyclic groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 4 Even and odd permutations 13 4.1 Alternating groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 5 Direct Products 17
6 Groups of small size 21 6.1 Remarks on larger sizes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 6.2 Quaternian group Q 8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23 7 Homomorphisms, normal subgroups and factor groups 25 7.1 Kernels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 7.2 Normal subgroups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 7.3 Factor groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 8 Symmetry groups in 3 dimensions 9 Counting using groups 35 37
II Linear Algebra
10 Some revision 11 Determinants 11.1 Basic properties . . . . . . . . . . . 11.2 Expansions of determinants . . . . 11.3 Major properties of determinants 11.3.1 Elementary matrices . . . .
39
41 43 44 46 47 48
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
. . . .
12 Matrices and linear transformations 53 12.0.2 Consequences of 12.1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 12.1 Change of basis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 12.2 Determinant of a linear transformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
iv
CONTENTS
13 Characteristic polynomials 59 13.1 Diagonalisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 13.2 Algebraic & geometric multiplicities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61 14 The Cayley-Hamilton theorem 65 14.1 Proof of Cayley-Hamilton . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 15 Inner product Spaces 15.1 Geometry . . . . . . . . . . . . . . . . . . 15.2 Orthogonality . . . . . . . . . . . . . . . 15.3 Gram Schmidt . . . . . . . . . . . . . . . 15.4 Direct Sums . . . . . . . . . . . . . . . . 15.5 Orthogonal matrices . . . . . . . . . . . 15.6 Diagonalisation of Symmetric Matrices 69 70 71 72 72 73 73
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
. . . . . .
Part I
Groups
1. INTRODUCTION
Chapter 1
Introduction
1. INTRODUCTION
2. SYMMETRY GROUPS
Chapter 2
2. SYMMETRY GROUPS
So we have a wast collection of new examples of groups G (). Example. 1. Equilateral triangle (= ) Here G () contains 3 rotations: e = 0 , = 2/3 , 2 = 4/3 , 3 reections: 1 = l 1 , 2 = l 2 , 3 = l 3 . Each of these corresponds to a permutation of the corners 1, 2, 3: e
2
1 2 3
Any isometry in G () permutes the corners. Since all the permutations of the corners are already present, there cant be any more isometries in G (). So the Symmetry group of equilateral triangle is e , , 2 , 1 , 2 , 3 ,
dihedral group
called the dihedral group D 6 . Note. Easy to work out products in D 6 : e.g. 3 2. The square Here G = G () contains 4 rotations: e , , 2 , 3 where = /2 , 4 reections: 1 , 2 , 3 , 4 where i = l i . So |G | 8. We claim that |G | = 8: Any g G permutes the corners 1, 2, 3, 4 (as g preserves distance). So g sends 1 i, 2 j, neighbour of i , 3 opposite of i , 4 opposite of j . So |G | (num. of choices for i ) (for j ) = 4 2 = 8. So |G | = 8. Symmetry group of the square is e , , 2 , 3 , 1 , 2 , 3 , 4 , called the dihedral group D 8 . (4 choices of i ) (2 choices for j ) (1 2 3)(1 2) = (1 3) 2 .
2. SYMMETRY GROUPS
Note. Can work out products using the corresponding permutations of the corners. e
2 3
e, (1 2 3 4), (1 3)(2 4), (1 4 3 2), (1 4)(2 3), (1 3), (1 2)(3 4), (2 4).
1 2 3 4 For example 3 1
(1 4 3 2)(1 4)(2 3) = (1 3) 2 .
Note. Not all permutations of the corners are present in D 8 , e.g. (1 2).
2.2 More on D 8
Dene H to be the cyclic subgroup of D 8 generated by , so H = = e , , 2 , 3 . Write = 1 . The right coset H = , , 2 , 3 is different from H . H H
So the two distinct right cosets of H in D 8 are H and H , and D 8 = H H . Hence H = = So the elements of D 8 are e , , 2 , 3 , , , 2 , 3 . To work out products, use the magic equation (see Sheet 1, Question 2) = 1 . Example. 3. Regular n -gon Let be the regular polygon with n sides. Symmetry group G = G () contains n rotations: e , , 2 , . . . , n 1 where = 2/n , n reections 1 , 2 , . . . , n where i = l i . , , 2 , 3 {1 , 2 , 3 , 4 } .
2. SYMMETRY GROUPS
So |G | 2n . We claim that |G | = 2n . Any g G sends corners to corners, say 1 i, 2 j neighbour of i . (n choices for i) (2 choices for j )
Then g sends n to the other neighbour of i and n 1 to the remaining neighbour of g (n ) and so on. So once i , j are known, there is only one possibility for g . Hence |G | number of choices for i , j = 2n . Therefore |G | = 2n . Symmetry group of regular n -gon is D 2n = e , , 2 , . . . , n , 1 , . . . , n , the dihedral group of size 2n . Note. Again can work in D 2n using permutations 1 4. Benzene molecule C 6 H6 . Symmetry group is D 1 2. 5. Innite strip of Fs ... What is symmetry group G ()? G () contains translation (1,0) : v v + (1, 0). Write = (1,0) . Then G () contains all translations n = (n ,0) . Note G () is innite. We claim that G () = = innite cyclic group. Let g G (). Must show that g = n for some n . Say g sends n | n Z , (1 2 3 n ) (2 n )(3 n 1)
F
1
F F ...
0 1
F at 0 F at n .
Note that n sends F at n to F at 0. So n g sends F at 0 to F at 0. So n g is a symmetry of the F at 0. It is easy to observe that F has only symmetry e . Hence n g
n n
= = =
e n n .
g g
Note. Various other gures have more interesting symmetry groups, e.g. innite strip of Es, square tiling of a plane, octagons and squares tiling of the plane, 3 dimensions platonic solids. . . later.
3. ISOMORPHISM
Chapter 3
Isomorphism
Let G H 1 G: 1 1 1 1 = = C 2 = {1, 1} , S 2 = {e , a } . 1 1 1 H: e a e e a a a e
These are the same, except that the elements have different labels (1 e , 1 a ). Similarly G H 1 G: 1 2 1 1 2 = = 2 1 C 3 = 1, , 2 , a = e , a , a 2 . 2 2 e H: 1 a a2 e e e a2 a a a2 e a2 a2 1 a
e, a, a2.
In these examples, there is a relabelling function : G H such that if g1 g2 then g 1 g 2 h1 h2 . Denition. Let G , H be groups. A function : G H is an isomorphism if (1) is a bijection, (2) (g 1 )(g 2 ) = (g 1 g 2 ) for all g 1 , g 2 G . If there exists an isomorphism : G H , we say G is isomorphic to H and write G = H. Note.
isomorphic G =H isomorphism
h1 , h2 ,
10
3. ISOMORPHISM
1. If G = H then |G | = | H | (as is a bijection). 2. The relation = is an equivalence relation: G = G (identity function), G = G (1 : H G isomorphism), =H H G = H, H =K G = K ( : G K isomorphism). Example. Which pairs of the following groups are isomorphic? G1 G2 G3 = = = C 4 = i = {1, 1, i , i } , symmetry group of a rectangle = e , , 1 , 2 , cyclic subgroup of D 8 = e , , 2 , 3 .
1
i.e. : i n n . To check that is an isomorphism (1) is a bijection, (2) for m , n Z (i m i n ) = = = = So is an isomorphism and G 1 = G3. Note. There exist many bijections G 1 G 3 which are not isomorphisms. 2. G 2 = G 3 or G 2 = G 1 ? Answer: G 2 = G1. By contradiction. Assume there exists an isomorphism : G 1 G 2 . Say (i ) (1) Then (1) = = as g 2 = e for all g G 2 . Similarly (1) = = (1 1) = (1)(1) y 2 = e. (i 2 ) = (i i ) = (i )(i ) x2 = e = = x G2, y G2. (i m +n ) m +n m n (i m )(i n ).
So (1) = (1), a contradiction as is a bijection. In general, to decide whether two groups G , H are isomorphic: If you think G = H , try to dene an isomorphism : G H .
3. ISOMORPHISM
11
If you think G = H , try to use the following proposition. Proposition 3.1. Let G , H be groups. (1) If |G | = | H | then G = H. (2) If G is abelian and H is not abelian, then G = H. (3) If there is an integer k such that G and H have different number of elements of order k , then G = H. Proof. (1) Obvious. (2) We show that if G is abelian and G = H , then H is abelian (this gives (2)). Suppose G is abelian and : G H is an isomorphism. Let h 1 , h 2 H . As is a bijection, there exist g 1 , g 2 G such that h 1 = (g 1 ) and h 2 = (g 2 ). So h2 h1 = = = = (3) Let Gk Hk = = g G | ord(g ) = k , {h H | ord(h ) = k } . (g 2 )(g 1 ) (g 2 g 1 ) (g 1 )(g 2 ) h1 h2 .
We show that G = H implies |G k | = | Hk | for all k (this gives (3)). Suppose G = H and let : G H be an isomorphism. We show that sends G k to Hk : Let g G k , so ord(g ) = k , i.e. g k = eG , g i = eG for 1 i k 1. Now (eG ) = e H , since (eG ) (eG )
1
= = = =
(eG ) eH
(g )(g ) (g )
i times
(g ) .
12
3. ISOMORPHISM
In other words, (g ) has order k , so (g ) Hk . So sends G k to Hk . As is 1-1, this implies |G k | | Hk |. Also 1 : H G is an isomorphism and similarly sends Hk to G k , hence | Hk | |G k |. Therefore |G k | = | Hk |. Example. 1. Let G = S 4 , H = D 8 . Then |G | = 24, | H | = 8, so G = H. 2. Let G = S 3 , H = C 6 . Then G is non-abelian, H is abelian, so G = H. 3. Let G = C 4 , H = symmetry group of the rectangle = e , , 1 , 2 . Then G has 1 element of order 2, H has 3 elements of order 2, so G = H. 4. Question: (R, +) = (R , )? Answer: No, since (R, +) has 0 elements of order 2, (R , ) has 1 element of order 2.
for all r . Then is an isomorphism, so G = (Z, +). Note. This proposition says that if we think of isomorphic groups as being the same, then there is only one cyclic group of each size. We say: up to isomorphism, the only cyclic groups are C n and (Z, +). Example. Cyclic subgroup 3 of (Z, +) is {3n | n Z}, innite, so by the proposition 3 = (Z, +).
13
Chapter 4
for e , (1 2 3), (1 3 2) : +, for (1 2), (1 3), (2 3) : . Generalize this: Denition. For arbitrary n 2, dene =
i<j
xi x j ,
a polynomial in n variables x 1 , . . . , x n . If we let each permutation g S n permute the variables x 1 , . . . , x n just as it permutes 1, . . . , n then g sends to . Denition. For g S n , dene the signature sgn(g ) to be +1 if g () = and 1 if g () = . So g () = sgn(g ). The function sgn : S n {+1, 1} is the signature function on S n . Call g an even permutation if sgn(g ) = 1, and odd permutation if sgn(g ) = 1. Example. In S 3 e , (1 2 3), (1 3 2) are even and (1 2), (1 3), (2 3) are odd. Example. Given (1 2 3 5)(6 7 9)(8 4 10) S 10 , whats its signature? Aim: To answer such questions instantaneously. This is the key: Proposition 4.1. (a) sgn(x y ) = sgn(x ) sgn( y ) for all x , y S n (b) sgn(e ) = 1, sgn(x 1 ) = sgn(x ). (c) If t = (i j ) is a 2-cycle then sgn(t ) = 1. Proof.
signature sgn
14
= =
sgn(x ), sgn( y ).
and hence sgn(x ) = sgn(x 1 ). (c) Let t = (i j ), i < j . We count the number of brackets in that are sent to brackets (x r x s ), r > s . These are (x i x j ), (x i x i +1 ), . . . , (x i x j 1 ), (x i +1 x j ), . . . , (x j 1 x j ). Total number of these is 2( j i 1) + 1, an odd number. Hence t () = and sgn(t ) = 1. To work out sgn(x ), x S n : express x as a product of 2-cycles use proposition 4.1 Proposition 4.2. Let c = (a 1 a 2 . . . a r ), an r -cycle. Then c can be expressed as a product of (r 1) 2-cycles. Proof. Consider the product (a 1 a r )(a 1 a r 1 ) (a 1 a 3 )(a 1 a 2 ). This product sends a 1 a 2 a 3 a r 1 a 1 . Hence the product is equal to c . Corollary 4.3. sgn(r -c ycl e ) = = (1)r 1 1 1 if r is odd if r is even.
= =
4.1(c)
15
Corollary 4.4. Every x S n can be expressed as a product of 2-cycles. Proof. From rst year, we know that x = c1 cm , a product of disjoint cycles c i . Each c i is a product of 2-cycles by 4.2. Hence so is x . Proposition 4.5. Let x = c 1 c m a product of disjoint cycles c 1 , . . . , c m of lengths r 1 , . . . , r m . Then sgn(x ) = (1)r 1 1 (1)r m 1 . Proof. sgn(x )
4.1(a)
= =
4.3
Importance of signature
1. Well use it to dene a new family of groups. 2. Fundamental in the theory of determinants.
An alternating group
=
4.1(a )
An A n (1 2)
= =
An e, {x (1 2) | x A n } .
These cosets are distinct (as (1 2) A n (1 2) but (1 2) A n ), and have equal size (i.e. | A n | = | A n (1 2)|). We show that S n = A n A n (1 2): Let g S n . If g is even, then g A n . If g is odd, then g (1 2) is even (as sgn(g (1 2)) = sgn(g ) sgn(1 2) = 1), so g (1 2) = x A n . Then g = x (1 2) A n (1 2).
1 So | A n | = 1 2 |S n | = 2 n !.
16
Example. 1. A 3 = {e , (1 2 3), (1 3 2)}, size 3 = 1 2 3!. 2. A 4 cycle shape in A 4 no. Total | A 4 | = 12 = 1 2 4!. 3. cycle shape in A 5 no. Total | A 5 | = 60 = 1 2 5!. e yes 1 (2) no (3) yes 20 (4) no (5) yes 24 (2, 2) yes 15 (3, 2) no e yes 1 (2) no (3) yes 8 (4) no (2, 2) yes 3
5. DIRECT PRODUCTS
17
Chapter 5
Direct Products
So far, weve seen the following examples of nite groups: C n , D 2n , S n , A n . Well get many more using the following construction. Recall: if T1 , T2 , . . . , Tn are sets, the Cartesian product T1 T2 Tn is the set consisting of all n-tuples (t 1 , t 2 , . . . , t n ) (t i Ti ).
Now let G 1 , G 2 , . . . , G n be groups. Form the Cartesian product G 1 G 2 G n and dene multiplication on this set by (x 1 , . . . , x n )( y 1 , . . . , y n ) = (x 1 G 1 y 1 , . . . , x n G n y n ) for x i , y i G i . Denition. Call G 1 G n the direct product of the groups G 1 , . . . , G n . Proposition 5.1. Under above dened multiplication, G 1 G n is a group. Proof. Closure True by closure in each G i . Associativity Using associativity in each G i , (x 1 , . . . , x n )( y 1 , . . . , y n ) (z 1 , . . . , z n ) = = = = = Identity is (e 1 , . . . , e n ), where e i is the identity of G i .
1 1 Inverse of (x 1 , . . . , x n ) is (x 1 , . . . , xn ).
direct product
(x 1 y 1 , . . . , x n y n )(z 1 , . . . , z n ) (x 1 y 1 )z 1 , . . . , (x n y n )z n x 1 ( y 1 z 1 ), . . . , x n ( y n z n ) (x 1 , . . . , x n )( y 1 z 1 , . . . , y n z n ) (x 1 , . . . , x n ) ( y 1 , . . . , y n )(z 1 , . . . , z n ) .
Example. 1. Some new groups: C 2 C 2 , C 2 C 2 C 2 , S 4 D 36 , A 5 A 6 S 297 , . . . , Z Q S 13 , . . . . 2. Consider C 2 C 2 . Elements are {(1, 1), (1, 1), (1, 1), (1, 1)}.
e a b ab
e e a b ab e a b ab
a a e ab b
b b ab e a
ab ab b a e
18
5. DIRECT PRODUCTS
G = C 2 C 2 is abelian and x 2 = e for all x G . 3. Similarly C 2 C 2 C 2 has elements (1, 1, 1), size 8, abelian, x 2 = e for all x . Proposition 5.2. (a) Size of G 1 G n is |G 1 ||G 2 | |G n |. (b) If all G i are abelian so is G 1 G n . (c) If x = (x 1 , . . . , x n ) G 1 G n , then order of x is the least common multiple of ord(x 1 ), . . . , ord(x n ), i.e. o (x ) = lcm(ord(x 1 ), . . . , ord(x n )). Proof. (a) Clear. (b) Suppose all G i are abelian. Then (x 1 , . . . , x n )( y 1 , . . . , y n ) = = = (c) Let r i = ord(x i ). Recall that
k =e xi
(x 1 y 1 , . . . , x n y n ) ( y 1 x1 , . . . , y n xn ) ( y 1 , . . . , y n )(x 1 , . . . , x n ).
r i |k .
Let r = lcm(r 1 , . . . , r n ). Then xr = = For 1 s < r , r i |s for some i . So x is = e . So x s = (. . . , x is , . . . ) = (e 1 , . . . , e n ). Hence r = ord(x ). Example. 1. Since cyclic groups C r are abelian, so are all direct products C r1 C r2 C rk . 2. C 4 C 2 and C 2 C 2 C 2 are abelian of size 8. Are they isomorphic? Claim: NO. Proof. Count the number of elements of order 2 : In C 4 C 2 these are (i , 1) except for (1, 1), so there are 3. In C 2 C 2 C 2 , all the elements except e have order 2, so there are 7. So C 4 C 2 = C2 C2 C2. Proposition 5.3. If hcf(m , n ) = 1, then C m C n = C mn .
r r (x 1 , . . . , xn )
(e 1 , . . . , e n ) = e .
5. DIRECT PRODUCTS
19
Proof. Let C m = , C n = . So ord() = m , ord() = n . Consider x = (, ) C m C n . By 5.2(c) ord(x ) = lcm(m , n ) = mn . Hence cyclic subgroup x of C m C n has size mn , so is whole of C m C n . So C m C n = x is cyclic and hence C m C n = C mn by 3.2. Direct products are fundamental to the theory of abelian groups. Theorem 5.4. Every nite abelian group is isomorphic to a direct product of cyclic groups. Proof. Wont give one here. Reference: [Allenby, p. 254]. Example. 1. Abelian groups of size 6: by theorem 5.4, possibilities are C 6 , C 3 C 2 . By 5.3, these are isomorphic, so there is only one abelian group of size 6 (up to isomorphism). 2. By 5.4, the abelian groups of size 8 are: C 8 , C 4 C 2 , C 2 C 2 C 2 . Claim: No two of these are isomorphic. Proof. Group | {x | ord(x ) = 2} | C2 C2 C2 7 C4 C2 3 C8 1
20
5. DIRECT PRODUCTS
21
Chapter 6
Groups of size 1, 2, 3, 5, 7
By 6.1, only such groups are C 1 , C 2 , C 3 , C 5 , C 7 .
Groups of size 4
Proposition 6.3. The only groups of size 4 are C 4 and C 2 C 2 . Proof. Let |G | = 4. By Lagrange, every element of G has order 1, 2 or 4. If there exists x G of order 4, then x is cyclic, so G = C 4 . Now suppose ord(x ) = 2 for all x = e , x G . So x 2 = e for all x G . Let e , x , y be 3 distinct elements of G . Now suppose xy x a contradiction. So suppose xy y = = x, e, = = e, y 1 = y ,
22
e e x y xy e x y xy
x x e xy y
y y xy e x
xy xy y x e
Groups of size 6
We know the following groups of size 6: C6, D6, S3. Recall D 6 is the symmetry group of the equilateral triangle and has elements e , , 2 , , , 2 . The following equations hold 3
2
= = =
e, e 2 .
The whole multiplication table of D 6 can be worked out using these equations. e.g. = 2 = 2 .
Proposition 6.4. Up to isomorphism, the only groups of size 6 are C 6 and D 6 . Proof. Let G be a group with |G | = 6. By Lagrange, every element of G has order 1, 2, 3 or 6. If there exists x G of order 6, then G = x is cyclic and therefore G = C 6 by 3.2. So assume G has no elements of order 6. Then every x G , (x = e ) has order 2 or 3. If all have order 2 then x 2 = e for all x G . So by Q4 Sheet 2, |G | is divisible by 4, a contradiction. We conclude that there exists x G with ord(x ) = 3. Also by 6.2, there is an element y of order 2. Let H = x = e , x , x 2 . Then y H so H y = H and G = H H y = e , x , x 2 , y, x y, x 2 y . What is y x ? Well, yx yx yx yx = = = = e x x2 y y y y x = = = = x 1 e x e a contradiction.
If y x = x y , lets consider the order of x y ( x y )2 = x y x y = xx y y = x2 y 2 = x2. Similarly ( x y )3 = x 3 y 3 = y = e . So x y does not have order 2 or 3, a contradiction. Hence y x = x y . Conclude that y x = x 2 y . At this point G = e , x , x 2 , y, x y, x 2 y , x3 = e, x2 = e, y x = x2 y . As for D 6 , can work out the whole multiplication table for G using these equations. It will be the same as the table for D 6 (with x , y instead of , ). So G = D6. Remark. Note that |S 3 | = 6, and S 3 = D6. ( as y x = x y )
23
Summary
Proposition 6.5. Up to isomorphism, the groups of size 7 are Size 1 2 3 4 5 6 7 Groups C1 C2 C3 C4, C2 C2 C5 C6, D6 C7
Sheet 3 Q2: |Q 8 | = 8. Q 8 is a subgroup of GL (2, C) and is not abelian and Q 8 = D 8 . Call Q 8 the quaternian group. Sheet 3 Q3 The only non-abelian groups of size 8 are D 8 and Q 8 . Yet more info Size 9 10 11 12 13 14 15 16 Groups only abelian (Sh3 Q4) C 10 , D 10 C 11 abelian, D 12 , A 4 + one more C 13 C 14 , D 14 C 15 14 groups
quaternian group
24
25
Chapter 7
(so sends rotations to +1 and reections to 1). Then is a homomorphism: ( r s )( t u ) = = since s t = ( r t s +u ) (1)s +u = ( r s )( r u ) t t s = 0, s = 1.
26
Proposition 7.1. Let : G H be a homomorphism (a) (eG ) = e H (b) (x 1 ) = (x )1 for all x G . (c) ord((x )) divides ord(x ) for all x G . Proof. (a) Note (eG ) = (eG eG ) = (eG )(eG ). Multiply by (eG )1 to get e H = (eG ). (b) By (a), eH = = So (x 1 ) = (x )1 . (c) Let r = ord(x ). Then (x )r = = = Hence ord((x )) divides r .
image Im
(x ) (x ) (x x ) (x r ) = (eG ) = e H .
Denition. Let : G H be homomorphism. The image of is Im = (G ) = (x ) | x G H . Proposition 7.2. If : G H is a homomorphism, then Im is a subgroup of H . Proof. (1) e H Im since e H = (eG ). (2) g , h Im then g = (x ) and h = ( y ) for some x , y G . gh so g h Im . (3) g Im then g = (x ) for some x G . So g 1 = (x )1 = (x 1 ) (7.1(b)) so g 1 Im . Hence Im is a subgroup of H . Example. 1. Is there a homomorphism : S 3 C 3 ? Yes, (x ) = 1 for all x S 3 . For this homomorphism, Im = {1}. = (x )( y ) = (x y ),
27
2. Is there a homomorphism : S 3 C 3 such that Im = C 3 ? Suppose : S 3 C 3 is a homomorphism Consider (1 2). By 7.1(c), (1 2) has order dividing ord(1 2) = 2. As (1 2) C 3 , this implies that (1 2) = 1. Similarly (1 3) = (2 3) = 1. Hence (1 2 3) = = and similarly (1 3 2) = 1. Weve shown that (x ) = 1 for all x S 3 . So there is no surjective homomorphism : S 3 C 3 . ((1 3)(1 2)) (1 3)(1 2) = 1
7.1 Kernels
Denition. Let : G H be a homomorphism. Then kernel of is Ker = x G | (x ) = e H . Example. 1. If : G H is (x ) = e H for all x G then Ker = G . 2. For sgn : S n C 2 , Ker(sgn) = = 3. If : (R, +) (C , ) is for all x R. Then Ker = = 4. Let : D 2n C 2 , Then Ker = . Proposition 7.3. If : G H is a homomorphism, then Ker is a subgroup of G . Proof. (1) eG Ker as (eG ) = e H by 7.1. (2) x , y Ker then (x ) = ( y ) = e H , so (x y ) = (x )( y ) = e H ; i.e. x y Ker . (3) x Ker then (x ) = e H , so (x )1 = (x 1 ) = e H , so x 1 Ker . In fact, Ker is a very special type of subgroup of G known as a normal subgroup. x R | e 2 i x = 1 Z. x S n | sgn(x ) = 1 A n , the alternating group.
kernel Ker
(x ) = e 2i x
( r s ) = (1)s .
28
Denition. Let G be a group, and N G . We say N is a normal subgroup of G if (1) N is a subgroup of G , (2) g 1 N g = N for all g G , where g 1 N g = g 1 ng | n N .
for all g G . Also subgroup G itself is normal, i.e. G G , as g 1G g = G for all g G . Next lemma makes condition (2) a bit easier to check. Lemma 7.4. Let N be a subgroup of G . Then N G g 1 N g N for all g G . Proof. Clear. Suppose g 1 N g N for all g G . Let g G . Then g 1 N g Using g 1 instead, we get (g 1 )1 N g 1 gNg Hence N g 1 N g . Therefore g 1 N g = N . Example. Claim A n S n . Need to show that g 1 A n g A n for all g S n (this will show A n S n by 7.4). For x A n , sgn(g 1 xg )
4.1 1
N.
N N.
= =
4.1
g 1 A n g A n .
29
= =
(1 3)(1 2)(1 3) (2 3) N .
So (1 3)N (1 3) = N and N S 3 . Example. If G is abelian, then all subgroups N of G are normal since for g G , n N , gn n Hence g 1 N g = N . Example. Let D 2n = e , , . . . , n 1 , , , . . . , n 1 . Fix an integer r . Then r D 2 n . Proof sheet 4. (key: magic equation = 1 , . . . , n = n ). Proposition 7.5. If : G H is a homomorphism, then Ker G . Proof. Let K = Ker . By 7.3 K is a subgroup of G . Let g G , x K . Then (g 1 xg ) = = = (g 1 )(x )(g ) (g )1 e H (g ) eH . = = ng g 1 ng .
So g 1 xg Ker = K . This shows g 1 K g K . So K G . Example. 1. We know that sgn : S n C 2 is a homomorphism, with kernel A n . So A n S n by 7.5. 2. Know : D 2n C 2 dened by ( r s ) = (1)s
is a homomorphism with kernel . So D 2n . 3. Heres a different homomorphism : D 8 C 2 where ( r s ) = (1)r . This is a homomorphism, as (( r s )( t u )) = = = The kernel of is Ker = = Hence e , 2 , , 2 D 8 . r s | r even e , 2 , , 2 . ( r t s +u ) (1)r t = (1)r (1)t ( r s )( t u ).
30
into a group in a natural way. Here is a natural denition of multiplication of these cosets: (N x )(N y ) = N (x y ). Does this denition make sense? To make sense, we need: Nx = Nx Ny = Ny Nxy = Nx y ( )
for all x , y , x , y G . This property may or may not hold. Example. G = S 3 , N = (1 2) = {e , (1 2)}. The 3 right cosets of N in G are N = Ne , N (1 2 3), N (1 3 2). Also N N (1 2 3) N (1 3 2) According to ( ), (N (1 2 3)) (N (1 3 2)) = N (1 2 3)(1 2 3) = N (1 3 2). But ( ) also says that (N (2 3)) (N (2 3)) So ( ) makes no sense in this example. ) makes sense? The condition is that N G . Key is to prove the following: = N (2 3)(2 3) = Ne . = = = N (1 2) N (1 2)(1 2 3) = N (2 3) N (1 2)(1 3 2) = N (1 3).
Proposition 7.6. Let N G . Then for x 1 , x 2 , y 1 , y 2 G N x1 = N x2 N y1 = N y2 (Hence denition of multiplication of cosets ( N x1 y 1 = N x2 y 2 . ) makes sense when N G .)
nx 2
N . Similarly
1 y1 y2
N . Observe
1 1 x1 y 1 y 2 x2 N .
N x 1 y 1 = N x 2 y 2 x 1 y 1 (x 2 y 2 )1 N ( )
1 1 1 (x 1 x 2 )( x 2 y 1 y 2 x2 ) N N
Now
1 y1 y2
N and N G , so
1 x2 N x2
= N , so
1 1 x2 y 1 y 2 x2 N .
).
31
So when N G then the denition of multiplication of cosets (N x )(N y ) = N x y for x , y G makes sense. Theorem 7.7. Let N G . Dene G/N to be the set of all right cosets N x (x G ). Dene multiplication on G/N by (N x )(N y ) = N x y. Then G/N is a group under this multiplication. Proof. Closure obvious. Associativity Using associativity in G (N xN y )N z = = = = = Identity is Ne = N , since N xNe NeN x Inverse of N x is N x 1 as N xN x 1 = N xx 1 = Ne , the identity. = = N xe = N x Nex = N x . (N x y )N z N (x y )z N x(y z) (N x )(N y z ) N x (N y N z ).
Denition. The group G/N is called the factor group of G by N . Note. |G/N | = Example. 1. A n S n . Since
|S n | | An | |G | |N | .
S n/A n
( A n (1 2))2
= =
A n (1 2) A n (1 2) A n (1 2)(1 2) = A n .
2. G any group. We know that G G . What is G/G ? Ans: G/G has 1 element, the identity coset G . So G/G = C1. Also e = {e } G . What is G/e ? Coset e g = g , and multiplication e g ( e h ) = e g h . So G/e = G (isomorphism g e g ).
32
3. G = D 12 = e , , . . . , 5 , , , . . . , 5 where 6 = 2 = e , = 1 . (a) Know that D 12 . Factor group D 12/ has 2 elements , so D 12/ = C2. (b) Know also that 2 = e , 2 , 4 D 12 . So D 12/ 2 has 4 elements, so
D 12/ 2 = C4
or C 2 C 2 .
Which? Well, let N = 2 . The 4 elements of D 12/N are N , N , N , N . We work out the order of each of these elements of D 12/N : ( N )2 (N )2 (N )
2
= = = = = =
N N = N 2 N, N N = N 2 N, N ()2 N.
So all non-identity elements of D 12/N have order 2, hence D 12/ = C2 C2. (c) Also 3 = e , 3 D 12 . Factor group D 12/ 3 has 6 elements so is = C 6 or D 6 . Which? Let M = 3 . The 6 elements of D 12/M are M , M , M 2 , M , M , M 2 . Let x = M and y = M . Then x3 y
2
= = = = = =
( M )3 = M M M = M 3 M, (M 02 = M 2 M, M M = M = M 1 = M 1 M x 1 y.
yx
So D 12/M = identity, x , x 2 , y , x y , x 2 y and x 3 = y 2 = identity, y x = x 1 y . So D 12/ 3 = D6. Heres a result tying all these topics together: Theorem 7.8 (First Isomorphism Theorem). Let : G H be a homomorphism. Then
G/Ker = Im .
Proof. Let K = Ker . So G/K is the group consisting of the cosets K x (x G ) with multiplication (K x )(K y ) = K x y . We want to dene a natural function G/K Im . Obvious choice is the function K x (x ) for x G . To show this is a function, need to prove: Claim 1. If K x = K y , then (x ) = ( y ). Proof. Suppose K x = K y . Then x y 1 K (as x K x x = k y for some k K x y 1 = k K ). Hence x y 1 K = Ker , so (x y 1 ) (x )( y
1
= = = =
e e e ( y ).
(x )( y )
(x )
33
By Claim 1, we can dene a function : G/K Im by (K x ) = (x ) for all x G . Claim 2. is an isomorphism. Proof. (1) is surjective: if (x ) Im then (x ) = (K x ). (2) is injective: (K x ) (x ) (x )( y )1 (x y so x y 1 Ker = K and so K x = K y . (3) Finally ((K x )(K y )) = = = = Hence is an isomorphism. Hence G/K = Im . Corollary 7.9. If : G H is a homomorphism, then |G | = | Ker | | Im |. [Group theory version of the rank-nullity theorem] Example. 1. Homomorphism sgn : S n C 2 . By 7.8
S n/Ker(sgn) = Im(sgn), 1
= = = =
(K y ) ( y ) e e,
(K x y ) (x y ) (x )( y ) (K x )(K y ).
so
S n/A n
= C2.
(x ) = e 2i x .
= = = =
34
3. Is there a surjective homomorphism from S 3 onto C 3 ? Shown previously No. A better way: suppose there exist such . Then Im = C 3 , so by 7.8,
S 3/Ker = C3.
So Ker is a normal subgroup of S 3 of size 2. But S 3 has no normal subgroups of size 2 (they are (1 2), (1 3), (2 3)). Given a homomorphism : G H , we know Ker G . Converse question: Given a normal subgroup N G , does there exist a homomorphism with kernel N ? Answer is YES: Proposition 7.10. Let G be a group and N G . Dene H = G/N . Let : G H be dened by (x ) = N x for all x G . Then is a homomorphism and Ker = N . Proof. (x y ) = = = so is a homomorphism. Also x Ker iff (x ) Nx iff x N . Hence Ker = N . Example. From a previous example, we know 2 = e , 2 , 4 D 12 . We showed that D 12/ 2 = C 2 C 2 . So by 7.10, the function (x ) = 2 x (x D 12 ) is a homomorphism D 12 C 2 C 2 which is surjective, with kernel 2 . = = eH N Nxy (N x )(N y ) (x )( y )
Summary
There is a correspondence normal subgroups of G homomorphisms of G . For N there is a homomorphism : G G/N with Ker = N . For a homomorphism , Ker is a normal subgroup of G . Given G , to nd all H such that there exist a surjective homomorphism G H : (1) Find all normal subgroups of G . (2) The possible H are the factor groups G/N for N G . Example. G = S 3 . (1) Normal subgroups of G are e , G , A 3 = (1 2 3) (cyclic subgroups of size 2 (1 2) are not normal). (2) Factor groups:
S 3/e = S3, S 3/S 3
= C1,
S 3/A 3
= C2/
35
Chapter 8
(where i xes corner i ), (3) rotations of order 2 about an axis joining the mid-points of opposite sides 12,34 , 13,24 , 14,23 . So |R | 12. Also |R | 12: can rotate to get any face i on bottom (4 choices). If i is on the bottom, only 3 possible congurations. Hence |R | 4 3 = 12. Hence |R | = 12. Claim 1. R = A4. Proof. Each rotation r R gives a permutation of the corners 1, 2, 3, 4, call it r : e i , 2 i 12,34 13,24 e = identity permutation all 8 3-cycles in S 4 (1 2 3), (1 3 2), . . . (1 2)(3 4) (1 3)(2 4).
Notice that {r | r R } consists of all the 12 even permutations in S 4 , i.e. A 4 . The map r r is an isomorphism R A 4 . So R = A4. Claim 2. The symmetry group G is S 4 .
36
Proof. Obviously G contains a reection with corresponding permutation = (1 2). So G contains R R . So |G | |R | + |R | = 24. On the other hand, each g G gives a unique permutation g S 4 , so |G | |S 4 | = 24. So |G | = 24 and the map g g is an isomorphism G S 4 .
37
Chapter 9
(B B R ) (B RB ) (RB B ).
Divide the set C into subsets called orbits of G : two colourings c , d are in the same orbit if there exists g D 6 sending c to d . The orbits are the sets (1) - (4) above. The number of distinguishable colourings is equal to the number of orbits of G .
orbit
General situation
Suppose we have a set S and a group G consisting of its permutations (e.g. S = C , G = D 6 above). Partition S into orbits of G , by saying that two elements s , t S are in the same orbit iff there exists a g G such that g (s ) = t . How many orbits are there? Lemma 9.1 (Burnsides Counting Lemma). For g G , dene x(g ) = = Then number of orbits of G = number of elements of S xed by g s S | g (s ) = s . 1 x(g ). |G | g G
x
38
Example. (1) C = set of 8 colourings of the equilateral triangle. G = D 6 . Here are the values of x(g ): g x(g ) e 8 2 2 2 4 4 2 4
By 9.1, number of orbits is 1 6 (8 + 2 + 2 + 4 + 4 + 4) = 4. (2) 6 beads coloured R, R, W, W, Y, Y are strung on a necklace. How many distinguishable necklaces are there? Each necklace is a colouring of the hexagon. The necklaces are indistinguishable if there is a rotation or reection sending one to the other. Let D be the set of colourings of a hexagon and G = D 12 . g x(g ) g x(g ) So by 9.1 number of orbits = So the number of distinguishable necklaces is 11. (3) Make a tetrahedral die by putting 1, 2, 3, 4 on the faces. How many distinguishable dice are there? Each die is a colouring (colours 1, 2, 3, 4) of the tetrahedron. Two such colourings are indistinguishable if there exists a rotation of the tetrahedron sending one to the other. Let E be the set of colourings, and G = rotation group of tetrahedron (so |G | = 12, G = A 4 by Chapter 8). Here for g G 24 if g = e , x(g ) = 0 if g = e . So by 9.1, number of orbits is
1 12 (24) = 2. 6 2
e
4 2
0 2 6
2 0
3 6
4 0 4 6
5 0 5 6
3 6
39
Part II
Linear Algebra
41
Chapter 10
Some revision
From M1GLA: Matrices, Linear equations; Row operations; echelon form; Gaussian elimination; Finding inverses; 2 2, 3 3 determinants; Eigenvalues & eigenvectors; Diagonalization. From M1P2: Vector spaces; Subspaces; Spanning sets; Linear independence; Basis, dimension; Rank, col-rank = row-rank; Linear transformations; Kernel, image, rank-nullity theorem; Matrix [T ]B of a linear transformation with respect to a basis B ; Diagonalization, change of basis .
42
11. DETERMINANTS
43
Chapter 11
Determinants
In M1GLA, we dened determinants of 2 2 and 3 3 matrices. Recall the denition of 3 3 determinant: a 11 a 21 a 31 a 12 a 22 a 32 a 13 a 23 a 33 = a 11 a 22 a 23 a 11 a 23 a 32 a 12 a 21 a 33 + a 12 a 23 a 31 + a 13 a 21 a 32 a 13 a 22 a 31 .
This expression has 6 terms. Each term (1) is a product of 3 entries, one from each column, (2) has a sign . Property (1) gives for each term a permutation of {1, 2, 3}, sending i j if a i j is present. Term a 11 a 22 a 33 a 11 a 23 a 32 a 12 a 21 a 33 a 12 a 23 a 31 a 13 a 21 a 32 a 13 a 22 a 31 Notice: the sign is sgn(permutation), all 6 permutations in S 3 are present. So | A| = Heres a general denition: Denition. Let A = (a i j ) be n n . Then the determinant of A is det( A ) = | A | =
S n S 3
Permutation e (2 3) (1 2) (1 2 3) (1 3 2) (1 3)
Sign + + +
determinant
Example. n = 1, A = (a 11 ). As S 1 = {e }, det( A ) = a 11 . a 11 a 12 n = 2, A = , S 2 = {e , (1 2)}. So | A | = a 11 a 22 a 12 a 21 . a 21 a 22 Check above that the new denition agrees with M1GLA. Aim: to prove basic properties of determinants. These are: (1) to see the effects of row operations on the determinant, (2) to prove multiplicative property of the determinant: det( AB ) = det( A ) det(B ).
44
11. DETERMINANTS
a 1,(1) a n ,(n ) .
So any result about determinants concerning rows will have an analogous result concerning columns. Proposition 11.2. Suppose B is obtained from A by swapping two rows (or two columns). Then |B | = | A |. Proof. We prove this for columns (follows for rows using 11.1). Say columns numbered r and s are swapped. Let = (r s ), 2-cycle in S n . Then if B = (b i j ), b i j = a i ,( j ) . So |B | = =
S n S n
As runs through all elements of S n so does . So |B | = | A |. Proposition 11.3. (1) If A has a row (or column) of 0s then | A | = 0. (2) If A has two identical rows (or columns) then | A | = 0. (3) If A is triangular (upper or lower) then | A | = a 11 a 22 a nn . Proof. (1) Each term in | A | has an entry from every row, so is 0.
11. DETERMINANTS
45
(2) If we swap the identical rows, we get A again, so by 11.2 | A | = | A |. Hence | A | = 0. (3) The only nonzero term in | A | is a 11 a 22 a nn .
Example. By (3), | I | = 1. Can now nd the effect of doing row operations on | A |. Theorem 11.4. Suppose B is obtained from A by using an elementary row operation. (1) If two rows are swapped to get B , then |B | = | A |. (2) If a row of A is multiplied by a nonzero scalar k to get B , then | B | = k | A |. (3) If a scalar multiple of one row of A is added to another row to get B , then |B | = | A |. (4) If | A | = 0, then |B | = 0 and if | A | = 0 then |B | = 0. Proof. (1) is 11.2 (2) Every term in | A | has exactly one entry from the row in question, so is multiplied by k . Hence |B | = k | A |. (3) Suppose c row k is added to row j . So a 11 |B | = a j i + c ak 1 . . . . . . a 1n ajn +c a 1n a j n + c a kn
a 11 = aji
. . . . . .
a 11 ak 1 ak 1
. . . . . .
a 1n a kn a kn
| A| + 0
46
11. DETERMINANTS
Example. 1 2 2 1 3 5 3 6 2 3 2 4 2 2 5 2 1 0 0 0 1 0 0 0 3 1 3 3 3 1 0 0 2 1 2 2 2 1 1 0 2 6 1 5 2 6 19 4
by 11.3(3)
by 11.4(3)
= 4.
by 11.3(3)
S n
a 11 | A 11 |.
Then |B | = | A | by 11.2. Terms in |B | with a 12 add to a 12 | A 12 . So terms in | A | with a 12 add to a 12 | A 12 |. Terms with a 13 . Swaps columns 2 and 3 of A , then swaps columns 1 and 2 to get B = a 13 a 23 . . . an3 a 11 a 21 . . . an1 a 12 a 22 . . . an2 .
11. DETERMINANTS
47
Continuing like this, see that | A | = a 11 | A 11 | a 12 | A 12 + which is expansion by the rst row. For expansion by i th row: Do i 1 row swaps in A to get B = ai 1 a 11 a 21 . . . ai n a 1n a 2n .
Theorem 11.6. Let A be n n . The following statements are equivalent. (1) | A | = 0. (2) A is invertible. (3) The system Ax = 0 (x Rn ) has only solution x = 0. (4) A can be reduced to I n by elementary row operations.
(1) (4): Suppose | A | = 0. Reduce A to echelon form A by elementary row operations. Then | A | = 0 by 11.4(4). So A does not have a zero row. Therefore A is upper diagonal and hence can be reduced further to I n by row operations.
(4) (1): Suppose A can be reduced to I n by row operations. We know that | I n | = 1. So | A | = 0 by 11.4(4).
Theorem 11.8. If A , B are n n then det( AB ) = (det A )(det B ). To prove this need
48
11. DETERMINANTS
r = 0,
..
Bi j
2 + 5r 5
Proposition 11.9. Let A be n n . An elementary row operation on A changes it to E A , where E is an elementary matrix. v1 . . Proof. Let A = . . vn (1) Row operation v i r v i sends A to v1 . . . r vi . . . vn = A i (r ) A .
11. DETERMINANTS
49
(2) Row operation v i v j sends A to B i j A . (3) Row operation v i v i + r v j sends A to C i j (r ) A . Proposition 11.10. (1) The determinant of an elementary matrix is nonzero and | A i (r )| = r , |B i j | = 1, |C i j (r )| = 1. (2) The inverse of an elementary matrix is also an elementary matrix: A i (r )1 = A i (r 1 ), B ij1 = B i j , C i j (r )1 = C i j (r ). Proposition 11.11. Let A be n n , and suppose A is invertible. Then A is equal to a product of elementary matrices, i.e. A = E 1 E k where each E i is an elementary matrix. Proof. By 11.6, A can be reduced to I by elementary row operations. By 11.9 rst row operations changes A to E 1 A with E 1 elementary matrix. Second changes E 1 A to E 2 E 1 A , E 2 elementary matrix . . . and so on, until we end up with I . Hence I = E k E k 1 E 1 A ,
1 1 where each E i is elementary. Multiply both sides on left by E 1 Ek E 1 to get 1 k 1 1 E1 Ek = A.
Each E i1 is elementary by 11.10(2). Remark. The proof shows how to nd the elementary matrices E i of which A is the product. Example. A = 1 1 2 0 2 0 Express A as a product of elementary matrices. 1 0 1 0 1 0 2 2 0 2 0 1 .
1 1
So A = = Towards 11.8: Proposition 11.12. If E is an elementary n n matrix, and A is n n , then det(E A ) = (det E )(det A ). v1 . . Proof. Let A = . . vn 1 1 1 1 0 1 0 1
1
1 0 1 0
1 1 1 1
1 0 0 2 .
0 1/2
1 0
50
11. DETERMINANTS
(1) If E = A i (r ), EA= v1 . . . r vi . . . vn .
So |E A | = | A | by 11.4 and so |E A | = |E || A | by 11.10. (3) If E = C i j (r ) then EA= So |E A | = |E || A | by 11.4 and 11.10. Corollary 11.13. If A = E 1 . . . E k , where each E i is elementary, then | A | = |E 1 | | E k | . Proof. | A | = |E 1 E k | = |E 1 ||E 2 E k | . . . = |E 1 ||E 2 | |E k |. by 11.12 v1 . . . vi + r v j . . . vn .
Proof of Theorem 11.8. (1) If | A | = 0 or |B | = 0, then | AB | = 0 by Sheet 6, Q5. (2) Now assume that | A | = 0 and |B | = 0. Then A , B are invertible by 11.6. So by 11.11, A = E1 Ek , where all E i , F i are elementary matrices. By 11.13, | A| |B | Also AB = E 1 E k F 1 F l , so by 11.13 | AB | = = |E 1 | |E n ||F 1 | |F k | | A ||B |. = = |E 1 | |E k |, |F 1 | |F k |. B = F1 Fl
11. DETERMINANTS
51
(2) det(P 1 AP ) = det( A ) for all n n matrices A . Proof. (1) det(P ) det(P 1 ) = det P P 1 = det I = 1 by 11.8. (2) det(P 1 AP ) = det(P 1 ) det A det P = det A by 11.8 and (1).
52
11. DETERMINANTS
53
Chapter 12
a 1n v 1 + + a nn v n .
[T ] B
[T ]B = Example. T : R2 R2 , T If E = {e 1 , e 2 }, e 1 = 1 0 , e2 = 0 1 x1 x2 = then T (e 1 ) T (e 2 ) so
a 11 . . . an1
a 1n . . . . a nn 2 1 1 2 x1 x2
2x 1 x 2 x 1 + 2x 2
= =
2e 1 + e 2 , e 1 + 2e 2 , 2 1 1 2
f1 + 2 f2, f1 + 3 f2, 1 2 1 3
Proposition 12.1. Let S : V V and T : V V be linear transformations and let B be a basis of V . Then [ST ]B = [S ]B [T ]B , where ST is the composition of S and T .
54
[ v ]B
Before proof, recall from M1P2: If B = {v 1 , . . . , v n } is a basis of V , and v V , can write v = r 1 v 1 + r n v n (r i scalars in F = R or C) and dene r1 . n [ v ]B = . . F . rn 4.8 of M1P2 says that: if T : V V is a linear transformation, then [T (v )]B = [T ]B [v ]B . Example. V = polynomials of degree 2, T (p (x )) = p (x ), B = 1, x , x 2 . Then 0 1 0 [T ]B = 0 0 2 . 0 0 0 Let p (x ) = 1 x + x 2 . Then 1 [p (x )]B = 1 . 1 So by ( ), 0 [T (p (x ))]B = 0 0 So T (p (x )) = 1 + 2x . Proof of 12.1. Let S , T : V V and B a basis of V . For v V , [ST (v )]B Also [ST (v )]B = [S (T (v ))] = = Hence [S ]B [T ]B [ v ]B 1 0 0 . . ,..., . for every column vector [v ]B . Applying this for [v ]B = . . 0 . = 0 1 [ST ]B [v ]B [S ]B [T (v )]B [S ]B [T ]B [ v ]B . = [ST ]B [v ]B . 1 0 0 0 1 1 2 1 = 2 . 0 1 0 ( )
Note. Proposition 12.1 actually proves the associativity of matrix multiplication in a natural way. Heres how. Take 3 linear transformations S , T , U : V V . We know that (ST )U = S (T U ) since (ST )U (v ) S (T U )(v ) for all v V . So by 12.1, if B is a basis of V , ([S ]B [T ]B )[U ]B = [S ]B ([T ]B [U ]B ). These matrices are arbitrary n n since given any n n matrix A , the linear transformation T : v Av (Rn Rn ) has [T ]B = A where B is the standard basis. = = S (T (U (v ))), S (T (U (v )))
55
Then ST (p (x )) = S (p (1 x )) = p (1 x ). If B = 1, x , x , 0 1 0 [S ]B = 0 0 2 , 0 0 0 1 1 1 [T ]B = 0 1 2 . 0 0 1 So 0 [S ]B [T ]B = 0 0 1 0 0 2 2 . 0
[T ]B [T ]B = A 2
= Ak .
= q ( A ).
= 2. Dene
0 0 0 0 0 0
1 0 0
2 0 0 2 0 0 0 1 2 0 2 . 0 0
1 0 0
0 2 0
GL (V )
GL (V ) = {all invertible linear transformations V V } . Then GL (V ) is a group under composition, T [T ]B is an isomorphism from GL (V ) to GL (n , F ) (the group of all n n invertible matrices).
GL (n , F )
56
fn Dene
p 1n e 1 + + p nn e n .
p 11 . P = . . p n1
change of basis matrix
p 1n . . . p nn
the change of basis matrix from E to F . Theorem 12.2. (1) P is invertible. (2) If T : V V is a linear transformation [T ]F = P 1 [T ]E P. Proof. (1) Write e1 = . . . = q 11 f 1 + + q n 1 f n ,
en and dene
q 1n f 1 + + q nn f n
q 11 . Q = (q i j ) = . . qn1
q 1n . . . . q nn
Then (by Sheet 7) QP = I , so Q = P 1 and P is invertible. (2) For v V v = r1e1 + + rn en , r1 . so [v ]E = . . and rn v = s1 f 1 + + sn f n , s1 . so [v ]F = . . . Then sn [ v ]E = P [ v ]F . So [v ]F = P 1 [v ]E . ( )
57
So [T ]F [v ]F = P [T ]E P [v ]F for every column vector [v ]F . Hence [T ]F = P 1 [T ]E P. x1 x2 2 1 1 2 [T ]E P So 12.2 says that [T ]F = = = agrees with earlier example 12. P 1 [T ]E P 1 1 1 2 0 1 1 3 2 1 1 2 1 1 0 1 x1 x2 = = 1 0 1 2 0 1 . , 0 1 1 1 0 1
Example. T : R R2 , T
,E= 2 1 1 1
,F =
. Here
58
59
Chapter 13
Characteristic polynomials
Recall from M1P2: let T : V V be a linear transformation. We say v V is an eigenvector of T if (1) v = 0, (2) T (v ) = v where is a scalar. The scalar is an eigenvalue of T . Denition. The characteristic polynomial of T : V V is det( I T ) where I : V V is the identity linear transformation. By the denition of determinant, this polynomial is equal to det( I [T ]B ) for any basis B . Example. V = polynomials of degree 2. T (p (x )) = p (1 x ), B = 1, x , x 2 . The characteristic polynomial of T is 1 1 1 1 1 0 +1 2 = ( 1)2 ( + 1). det I 0 1 2 = det 0 0 1 0 0 1 From M1P2: Proposition 13.1. (1) The eigenvalues of T are the roots of the characteristic polynomial of T . (2) If is an eigenvalue of T , the eigenvectors corresponding to are the nonzero vectors in E = {v V | ( I T )(v ) = 0} = ker( I T ). (3) Matrix [T ]B is a diagonal matrix iff B consists of eigenvectors of T . Note. Note E = ker( I T ) is a subspace of V , called the -eigenspace of T . Example. In previous example, eigenvalues of T are 1, 1. Eigenspace E 1 is ker( I T ). Solve 0 0 0 1 2 0 1 2 0 0 0 . 0
-eigenspace eigenvalue characteristic polynomial eigenvector
60
Eigenspace E 1 . Solve 2 0 0 1 0 0 1 2 2 0 0 . 0
c Solutions are vectors 2c . So E 1 = {c 2c x | c F }. 0 Basis of E 1 is 1, x , x 2 . Basis of E 1 is 1 2x . Putting these together, get basis B = 1, x x 2 , 1 2x of V consisting of eigenvectors of T , and 1 [T ]B = 0 0 0 1 0 0 0 . 1
Proposition 13.2. Let V a nite-dimensional vector space over C. Let T : V V be a linear transformation. Then T has an eigenvalue C. Proof. The characteristic polynomial of T has a root C by the Fundamental theorem of algebra. Example. Proposition 13.2 is not necessarily true for vector spaces over R. For example T : R2 R2 , 1 T (x 1 , x 2 ) = (x 2 , x 1 ). Characteristic polynomial is = 2 + 1, no real roots. 1
13.1 Diagonalisation
Basic question is: How to tell if there exists a basis B such that [T ]B is diagonal? Useful result: Proposition 13.3. Let T : V V be a linear transformation. Suppose v 1 , . . . , v k are eigenvectors of T corresponding to distinct eigenvalues 1 , . . . , k . Then v 1 , . . . , v k are linearly independent. Proof. By induction on k . Let P (k ) be the statement of the proposition. P (1) is true, since v 1 = 0, so v 1 is linearly independent. Assume P (k 1) is true, so v 1 , . . . , v k 1 are linearly independent. We show v 1 , . . . , v k are linearly independent. Suppose r 1 v 1 + + r k v k = 0. Apply T to get 1 r 1 v 1 + + k r k v k = 0 Then (2)-k (1) gives r 1 (1 k )v 1 + + r k 1 (k 1 k )v k 1 = 0. As v 1 , . . . , v k 1 are linearly independent, all coefcients are 0. So r 1 (1 k ) = . . . = r k 1 (k 1 k ) = 0. As the i are distinct, 1 k , . . . , k 1 k = 0. Hence r 1 = . . . = r k 1 = 0. Then (1) gives r k v k = 0, so r k = 0. Hence v 1 , . . . , v k are linearly independent, completing the proof by induction. (2) (1)
61
Corollary 13.4. Let dim V = n and T : V V be a linear transformation. Suppose the characteristic polynomial of T has n distinct roots. Then V has a basis B consisting of eigenvectors of T (i.e [T ]B is diagonal). Proof. Let 1 , . . . , n be the (distinct) roots, so these are the eigenvalues of T . Let v 1 , . . . , v n be corresponding eigenvectors. By 13.3, v 1 , . . . , v n are linearly independent, hence form a basis of V since dim V = n . Example. Let A= 1 0 . . . 0 2 .. . 0 n
|x I A | =
i =1
(x i )
which has roots 1 , . . . , n . Hence by 13.4, A can be diagonalized, i.e. there exists P such that P 1 AP is diagonal. Note. Not necessarily true if diagonal entries are not distinct, e.g. 1 0 1 1 cannot be diagonalized.
where is not a root of q (x ). Call a () the algebraic multiplicity of .The geometric multiplicity of is g () = dim E , where E = ker( I T ), the -eigenspace of T . Similar denition for n n matrices. Example. A = 1 0 1 , 2 a (1) a (2) A= 1 0 1 , 1 a (1) g (1) = = 2, 1. = = 1 = g (1), 1 = g (2).
62
Proof. Let r = g () = dim E and let v 1 , . . . , v r be a basis of E . Extend to a basis of V : B = {v 1 , . . . , v r , w 1 , . . . , w s } . We work out [T ]B : T (v 1 ) = . . . = = . . . = v 1 ,
T (v r ) T (w 1 )
v r , a 11 v 1 + + a r 1 v r + b 11 w 1 + + b s 1 w s ,
T (w s ) So
a 1s v 1 + + a r s v r + b 1s w 1 + + b ss w s .
0 . . . 0 0 . . . . . . 0
0 . . . 0
.. .
0 0 0 . . . . . . 0
[T ]B
a 11 . . . . . . ar 1 b 11 . . . . . . bs1
a 1s . . . . . . ar s b 1s . . . . . . b ss
Clearly the characteristic polynomial of this is p (x ) = det By Sheet 7 Q3, this is p (x ) = det((x ) I r ) det(x I s B ) = (x )r q (x ). Hence the algebraic multiplicity a () r = g (). Criterion for diagonalisation: Theorem 13.6. Let dim V = n , T : V V be a linear transformation and let 1 , . . . , r be the distinct eigenvalues of T , and characteristic polynomial of T be
r
(x ) I r 0
A x Is B
p (x ) =
i =1
(x i )a (i )
(so
r i =1 a (i ) = n ).
= n.
(3) g (i ) = a (i ) for all i . Proof. To prove(1) (2), (3): Suppose (1) holds. Each vector in B is in some E i , so
r
dim E i |B | = n .
i =1
63
By 13.5
r r
dim E i =
i =1 i =1
g (i )
r i =1
a (i ) = n .
Hence r i =1 dim E i = n and g (i ) = a (i ) for all i . Evidently (2) (3), so it is enough to show that (2) (1). Suppose r i =1 dim E i = n . Let B i be a basis of E i and let B = r B , so | B | = n (the following shows that B s are disjoint). We claim B is a basis of V , i i i =1 hence (1) holds: Its enough to show that B is linearly independent (since |B | = n = dim V ). Suppose there is a linear relation v v + + z z = 0.
v B 1 z B r
Write v1 =
v B 1
v v ,
. . . vr =
z B r
z z ,
so v i E i and v 1 + + v r = 0. As 1 , . . . , r are distinct, the set of nonzero v i s is linearly independent by 13.3. Hence v i = 0 for all i . So vi =
v B i
v v = 0.
As B i is linearly independent (basis of E i ) this forces v = 0 for all v B i . This completes the proof that B is linearly independent, hence a basis of V . Using 13.6 we get an algorithm to check if a given n n matrix or linear transformation is diagonalizable: 1. nd the characteristic polynomial, factorise it as (x i )a (i ) . 2. Calculate each g (i ) = dim E . 3. If g (i ) = a (i ) for all i , YES. If g (i ) < a (i ) for some i , NO. 3 1 1 Example. Let A = 7 5 1 . Check that 6 6 2 (1) Characteristic polynomial is (x + 2)2 (x 4). (2) For eigenvalue 4: a (4) = 1, g (4) = 1 (as it is a (4)). For eigenvalue 2: a (2) = 2, g (2) = dim E 2 = 1. So A is not diagonalizable by 13.6.
64
65
Chapter 14
Theorem 14.1 (Cayley-Hamilton Theorem). Let V be nite-dimensional vector space, and T : V V a linear transformation with characteristic polynomial p (x ). Then p (T ) = 0, the zero linear transformation. Proof later. Corollary 14.2. If A is a n n matrix with characteristic polynomial p (x ), then p ( A ) = 0. Proof. Apply 14.1 to the linear transformation T : F n F n (F = R or C) given by T (v ) = Av . Note. 14.2 is NOT obvious as p (x ) = det(x I A ), put x = A , get p ( A ) = det( AI A ) = 0; rubbish. Example. 1. 14.2 is obvious for diagonal matrices A= This is because the i are the roots of p (x ), so p (1 ) p ( A) = 1 .. . n .
.. . p (n ) = 0.
Corollary 14.2 is also quite easy to prove for diagonalisable matrices (Sheet 8 Q2). 2. For 2 2 matrices A = a c b d , the characteristic polynomial is x a c b x d = x 2 (a + d )x + ad bc .
66
Let T : V V be a linear transformation with characteristic polynomial p (x ). Aim: for v V , show that p (T )(v ) = 0. Strategy: Study the subspace vT = = Span(v, T (v ), T 2 (v ), . . . ) Span(T i (v ) | i 0).
T -invariant
Denition. A subspace W of V is T -invariant if T (W ) W , i.e. T (w ) W for all w W . Proposition 14.3. Pick v V and let W = v T = Span(T i (v ) | i 0). Then W is T -invariant. Proof. Let w W , so w Then T (w ) = a 1 T i 1 +1 (v ) + + a r T i r +1 (v ), so T (w ) W . Example. V = polynomials of deg 2, T (p (x )) = p (x + 1). Then xT = = Clearly this is T -invariant. Span(x , T (x ), T 2 (x ), . . .) Span(x , x + 1) = subspace of polynomials of deg 1. = a 1 T i 1 (v ) + + a r T i r (v ).
TW
restriction
for all w W . Then TW is a linear transformation, the restriction of T to W . Proposition 14.4. If W is a T -invariant subspace of V , then the characteristic polynomial of TW divides the characteristic polynomial of T . Proof. Let BW = {w 1 , . . . , w k } be a basis of W and extend it to a basis B = {w 1 , . . . , w k , x 1 , . . . , x l } of V . As W is T -invariant, T (w 1 ) = . . . = a 11 w 1 + + a k 1 w k ,
T (w k ) Then
a 1k w 1 + + a kk w k .
[T W ]B W
a 11 . . . ak 1
a 1k . . . = A a kk
67
and [T ]B The characteristic polynomial of TW is p W (x ) and characteristic polynomial of T is p (x ) = = = So p W (x ) divides p (x ). Example. V = polynomials of deg 2, T (p (x )) = p (x + 1), W = x T = Span (x , x + 1). Take basis BW = {1, x }, B = 1, x , x 2 . Then [T ]B W = [T ]B = 1 0 1 0 0 1 1 1 1 0 , 1 2 . 1 det x Ik A 0 X x Il Y = det(x I k A ) = A 0 X Y .
det(x I k A ) det(x I l Y ) p W (x ) q (x ).
Characteristic polynomial of TW is (x 1)2 , characteristic polynomial of T is (x 1)3 . Proposition 14.5. Let T : V V be a linear transformation. Let v V , v = 0, and W = v T = Span T i (v ) | i 0 . Let k = dim W . Then is a basis of W . Proof. We show that v, T (v ), . . . , T k 1 (v ) is linearly independent, hence a basis of W . Let j be the largest integer such that the set v, T (v ), . . . , T j 1 (v ) is linearly independent. So 1 j k . Aim to show that j = k . Let S = v, T (v ), . . . , T j 1 (v ) and X = Span(S ). Then X W and dim X = j . By the choice of j , the set v, T (v ), . . . , T j 1 (v ), T j (v ) is linearly dependent. This implies that T j (v ) Span(S ) = X . Say T j (v ) So T j +1 (v ) = b 0 T (v ) + b 1 T 2 (v ) + + b j 1 T j (v ) X . Similarly T j +2 (v ) X , T j +3 (v ) X and so on. Hence T i (v ) X = b 0 v + b 1 T (v ) + + b j 1 T j 1 (v ). v, T (v ), T 2 (v ), . . . , T k 1 (v )
68
W = Span(T i (v ) | i 0) X .
As X W this means X = W , so j = dim X = dim W = k . Hence v, T (v ), . . . , T k 1 (v ) is linearly independent. Proposition 14.6. Let T : V V , let v V and W = v T = Span T i (v ) | i 0 , with basis BW = v, T (v ), . . . , T k 1 (v ) as in 14.5. Then (1) there exist scalars a i such that a 0 v + a 1 T (v ) + + a k 1 T k 1 (v ) + T k (v ) = 0, (2) the characteristic polynomial of TW is p W (x ) = x k + a k 1 x k 1 + + a 1 x + a 0 , (3) p W (T )(v ) = 0. Proof. (1) Clear, since T k (v ) W and BW is a basis of W . (2) Clearly [T W ]B W 0 1 0 . . . 0 0 0 1 . . . 0 .. . 0 0 0 . . . 1 a 0 a 1 a 2 . . . a k 1
(for the last column T (T k 1 (v )) = T k (v ) = a 0 v a 1 T (v ) a k 1 T k 1 (v )). By the Sheet 8, the characteristic polynomial of this matrix is p W (x ) = x k + a k 1 x k 1 + + a 0 . (3) Clear from (1) and (2). Completion of the proof of Cayley-Hamilton 14.1. Have T : V V with characteristic polynomial p (x ), let v V , let W = v T with basis v, T (v ), . . . , T k 1 (v ) . Let p W (x ) = x k + a k 1 x k 1 + + a 0 to be the characteristic polynomial of TW . By 14.6(3), p W (T )(v ) = 0. By 14.4, p W (x ) divides p (x ), say p (x ) = q (x )p W (x ), so p (T ) = q (T )p W (T ). Then p (T )(v ) = = = (q (T )p W (T ))(v ) q (T )(p W (T )(v )) q (T )(0) = 0.
Repeat for all v V (and corresponding W s) to get p (T )(v ) = 0 for all v V , i.e. p (T ) = 0.
69
Chapter 15
= x .x =
n 2 1 |x i | , etc.
Generalisation of these:
Denition. Let F = R or C and let V be a vector space over F . Suppose that to each pair x , y V we assign a scalar (x , y ) F . This mapping is an inner product on V if, for all x , y , z V and , F , the following axioms hold: (1) (x + y , z ) = (x , z ) + ( y , z ). (2) ( y , x ) = (x , y ) (complex conjugate) (3) (x , x ) > 0 if x = 0. Call V equipped with an inner product ( , ), an inner product space (real if F = R, complex if F = C). Note. (x , x ) R by (2), hence (3) makes sense. (0, y ) = (0x , y ) = 0(x , y ) = 0 for all y V . If F = R then (2) says ( y , x ) = (x , y ). (x , y + z ) = (x , y ) + (x , z ). Example. 1. V = Rn , (x , y ) = x . y is an inner product: (1), (2) clear, for (3),(x , x ) = x .x = 2. V = Cn , (x , y ) = x . y = x i y i is an inner product: (1) clear; for (2), ( y , x ) = for (3) (x , x ) = x i x i = |x i |2 > 0 if x = 0. 3. Let V be the vector space of all continuous functions f : [a , b ] R. Dene
b 2 xi > 0 if x = 0.
(x , y ) inner product
y i xi =
x i y i = (x , y );
f ( x ) g ( x )d x .
a
(x ) + g (x ))h (x )d x =
b a
f ( x )h ( x ) +
b a g (x )h (x ) = ( f , h ) + (g , h ).
70
(3) ( f , f ) =
b a
f (x )2 d x > 0 if f = 0.
if x = 0.
15.1 Geometry
length x unit vector distance d
Let V be an inner product space with inner product ( , ). Dene the length of x V by x = Say x is a unit if x = 1. Distance d (x , y ) = x y . . Theorem 15.1 (Cauchy-Schwarz inequality). For any x , y V , | ( x , y )| x y . (x , x ).
Proof. Clear if y = 0. So assume that y = 0. Write = (x , y ). For t R consider the vector x t y . Now 0 x t y
2
=
(1)
(x t y , x t y ) (x , x t y ) t ( y , x t y ) (x , x ) t (x , y ) t ( y , x ) t ( y , y ) (x , x ) t t + t 2 ( y , y ) t 2 ||2 y
2 2
= = =
t 2||2 + x
||2 y
Example. 1. Apply the Cauchy-Schwartz inequality to Cn with x , y = x1 , . . . , xn , y 1 , . . . , y n x i y i . For any complex numbers
(x 1 y 1 + + x n y n )2 (|x 1 |2 + + |x n |2 )(| y 1 |2 + + | y n |2 ). 2. Let V be the space of continuous functions [0, 1] R with f , g = Schwartz
1 2 1 1 0
f (x )g (x )d x . By Cauchy-
f ( x ) g ( x )d x
0
f ( x )2 d x
0
g ( x )2 d x .
3. Take x , y = x 1 y 1 x 1 y 2 x 2 y 1 + 3x 2 y 2 R2 . By Cauchy-Schwartz
2 2 2 2 ( x 1 y 1 x 1 y 2 x 2 y 1 + 3 x 2 y 2 )2 ( x 1 2x 1 x 2 + 3x 2 )( y 1 2y1 y2 + 3y2 )
71
15.2 Orthogonality
Denition. Let V be an inner product space. Then the vectors x , y V are orthogonal if x , y = 0. For W V , the orthogonal complement of W is W
Example. For V = R3 with dot product as the inner product, let W = {(1, 1, 1, 1)}. Then W = = x R3 | x (1, 1, 1, 1) = 0 the plane x 1 + x 2 + x 3 = 0.
Proposition 15.2. For W V , W is a subspace of V . Proof. (1) 0 W as (0, v ) = 0 for all v V . (2) For x , y W , (x , w ) = y , w = 0 for all w W . Therefore x + y , w = (x , w ) + y , w = 0; hence x + y W . (3) For x W , F , (x , w ) = (x , w ) = 0; x W . Denition. A set {v 1 , . . . , v n } is orthogonal if v i , v j = 0 for all i = j . It is orthonormal if vi , v j = Example. 1. For V = Rn with dot product, The standard basis e 1 , . . . , e n is an orthonormal basis of V . 2. For V = set of continuous functions, f : [, ] R with inner product f , g = set of orthogonal functions in V is {1, cos x , cos 2x , . . . , sin x , sin 2x , . . . }. Proposition 15.3. Let {u 1 , . . . , u r } be an orthonormal set of vectors. Then (1) {u 1 , . . . , u r } is linearly independent, (2) for v V , the vector w = v (v, u 1 ) u 1 (v, u 2 ) u 2 (v, u r ) u r is orthogonal to every u i . Proof. (1) Let 1 , . . . , r F be such that 1 u 1 + + r u r = 0. Take inner product with u i (1 u 1 + + r u r , u i ) 1 (u 1 , u i ) + + r (u r , u i ) i (u i , u i ) = = = (0, u i ) 0 0
orthogonal orthonormal
0 1
if j = i , if j = i .
f ( x ) g ( x )d x , a
as the set is orthogonal. Hence (repeating for all u i ) all i = 0 and so {u 1 , . . . , u r } is linearly independent. (2) (w , u i ) = = = ( v , u i ) ( v , u i ) ( v , u 1 ) (u 1 , u i ) ( v , u r ) (u r , u i ) ( v , u i ) ( v , u i ) (u i , u i ) 0.
72
Example. V = R3 with dot product, u 1 = 1 (1, 1, 0), u 2 = 1 (1, 1, 1) so that {u 1 , u 2 } is orthonormal. 2 2 Want to nd a vector orthogonal to both u 1 and u 2 . Let v = (1, 0, 0). Then w = = = is orthogonal to both u 1 and u 2 . v (v, u 1 ) u 1 (v, u 2 ) u 2 1 1 u1 (1, 0, 0) u2 3 2 1 1 1 , , 6 6 3
Example. Let V = R2 with x , y = x 1 y 1 x 1 y 2 x 2 y 1 + 3x 2 y 2 . Find an orthonormal basis. Use GramSchmidt to get u1 w2 = = = = u2 Therefore the orthonormal basis is (1, 0), = (1, 0) = (1, 0), (1, 0) (0, 1) ((0, 1), u 1 ) u 1 (0, 1) + u 1 (1, 1), 1 (1, 1). 2 .
1 2(1,1)
Proposition 15.5. Let V be a nite dimensional inner product space and {u 1 , . . . , u r } an orthonormal set of vectors in V . Then there exists an orthonormal basis of V containing u 1 , . . . , u r . Proof. By 15.3, u 1 , . . . , u r are linearly independent. Therefore they can be extended to a basis of V . Applying Gram-Schmidt, an orthonormal basis can be obtained containing u 1 , . . . , u r .
then we write V = U W and say that V is a direct sum of U and W . Example. V = R3 . If U = Span(1, 0, 0) and W = Span {(0, 1, 0), (0, 0, 1)} then V = U W . Note that dim V = dim U + dim W . Proposition 15.6. The following statements are equivalent: (1) V = U W . (2) dim V = dim U + dim W and U W = {0}. (3) If BU is a basis of U and BW a basis of W then B = BU BW is a basis of V .
73
Proposition 15.7. Let V be a nite dimensional inner product space and let W be a subspace of V . Then V = W W and therefore dim W = dim V dim W . Proof. By 15.4, W has an orthonormal basis {w 1 , . . . , w r }. By 15.5, there exists an orthonormal basis of V {w 1 , . . . , w r , v 1 , . . . , v s }. Each v i is orthogonal to w 1 , . . . , w r , so v i W . Let v V . Then v = 1 w 1 + + r w r + 1 v 1 + + s v s , so v = w + x for w W , x W . If v W W , then (v, v ) = 0 and so v = 0. Hence W W = {0} and therefore V = W W .
= ( Ax )T ( Ax ) = x T A T Ax = x T I x = x
.
Hermitean
x1 . x = . . , y = xn Then
T (x ), y = x , T ( y ) for all x , y V
74
for T : V V with T (x ) = Ax for all x V . Let A Rn n be symmetric and T : V V , T (x ) = Ax for all x V . Then T (x ), y = = = = since A is symmetric. Since A is real we have T (x ), y = = x T ( Ay ) x, T (y ) . ) is called a Ax , y ( Ax )T y x T AT y x T Ay
symmetric operator
Theorem 15.10. Let V be an inner product space and T : V V be a linear transformation such that T (x ), y = x , T ( y ) for all x , y V . Then (1) the eigenvalues of T are real, (2) V has an orthonormal basis consisting of eigenvectors of T . Remark. Applying 15.10 to the linear transformation T (x ) = Ax proves 15.9 for x Rn . Proof of 15.10. (1) Let C be an eigenvalue of T and v a corresponding eigenvector; so T (v ) = v . Then (T (v ), v ) = = = By ( ), this is equal to (v, T (v )) = v therefore R.
2
( v , v ) (v, v ) v
2
(2) We prove that V has a basis of eigenvectors of T by induction on dim V = n . If n = 1, result is clear. Now assume result for all inner product spaces of dimension n 1. Let dim V = n > 1. By 13.2, T has an eigenvalue C and by (1), R. Let v 1 be a corresponding eigenvector, so 1 T (v 1 ) = v 1 . Dene unit vector u 1 = v v 1 . Let W = Span(u 1 ). By 15.7, V = W W . We claim that W is T -invariant, i.e. T (w ) W for all w W :Let w W . Then (u 1 , w ) = 0 and so (u 1 , T (w )) = (T (u 1 ), w ) = (u 1 , w ) = (u 1 , w ) = 0 and therefore T (w ) W .
Thus we can dene the restriction TW : W W , TW (w ) = T (w ) for all w W . Space W is an n 1 dimensional inner product space and linear transformation TW satises ( ). By induction, W has an orthonormal basis u 2 , . . . , u n of eigenvectors of TW , hence also of T . These vectors are orthonormal to u 1 . Hence u 1 , . . . , u n is an orthonormal set of vectors. It is linearly independent by 15.3 and hence a basis of V consisting of eigenvectors of T .