MATH 247: Honours Applied Linear Algebra Anna Brandenberger, Daniela Breitman May 11, 2018
This is a transcript of the lectures given by Prof. Axel Hundemer during the winter semester of the 2017-2018 academic year (01-04 2018) for the Honours Applied Linear Algebra class (math 247). Subjects covered are: Matrix algebra, determinants, systems of linear equations; Abstract vector spaces, inner product spaces, Fourier series; Linear transformations and their matrix representations; Eigenvalues and eigenvectors, diagonalizable and defective matrices, positive definite and semidefinite matrices; Quadratic and Hermitian forms, generalized eigenvalue problems, simultaneous reduction of quadratic forms; and Applications.
1 Abstract Vector Spaces
1.1 Group Theory lecture 01/09 Definition 1.1. Group: Let G be a set with a binary operation G × G → G such that: (a, b) 7→ a + b
• associativity a + (b + c) = (a + b) + c, ∀a, b, c ∈ G.
• there exists a neutral element ∃0 ∈ G s.t. a + 0 = 0 + a = a, ∀a ∈ G.
• there exists an (additive) inverse element ∃ − a ∈ G s.t. a + (−a) = (−a) + a = 0, ∀a ∈ G.
Definition 1.2. A group (G, +)1 is called Abelian if commutativity is 1 A group G with operation +. satisfied:
• ∀a, b ∈ G, a + b = b + a.
Examples 1.1. of Abelian groups:
• (R, +), (C, +), (Q, +), (Z, +) standard addition.
• (R\{0}, ·), (C\{0}, ·) standard multiplication. However, N = {1, 2, 3 . . . } with stan- dard addition is not a group because • Rn, Cn with component-wise addition: there is no inverse element in N for any n ∈ N. (x1, x2,..., xn) + (y1, y2,..., yn) = (x1 + y1,..., xn + yn)
• Mat(m × n, R), Mat(m × n, C): m × n matrices with coefficients in R or C with component-wise addition. 2 anna brandenberger, daniela breitman
• the set F(R) of all real-valued functions with domain R, with:
– addition: ( f + g)(x) ≡ f (x) + g(x) – neutral element: zero function 0(x) ≡ 0 – inverse element: (− f )(x) ≡ − f (x), ∀x ∈ R
Example 1.2. of a non-Abelian group:
• GL(n, R), n ≥ 2: "General Linear Group" i.e. the set of all invert- ible n × n matrices with coefficients in R, with matrix multiplica- tion. " 1 0 # . – neutral element: In = .. s.t. In · A = A · In = A, 0 1 ∀A ∈ GL(n, R). −1 −1 −1 – inverse of A is A since A · A = A · A = In.
Theorem 1.1. Cancellation Law: Let (G, +) be a group and a, b, c ∈ G:
(a)a + b = a + c =⇒ b = c
(b)b + a = c + a =⇒ b = c
Proof of (a):
Exercise 1.3. Prove (b). a + b = a + c =⇒ −a + (a + b) = −a + (a + c) existence of an inverse =⇒ (−a + a) + b = (−a + a) + c associative =⇒ 0 + b = 0 + c definition of neutral element =⇒ b = c neutral element
1.2 Vector Spaces lecture 01/11 Definition 1.3. Let V be a set, and let K be either R or C together + : V × V → V vector addition with two operations: , · : K × V → V scalar multiplication such that (V, +) is an abelian group. If the following axioms for scalar multiplication hold ∀k, l ∈ K and u, v ∈ V:
(kl)v = k(lv) associativity ( + ) = + The neutral element is called the zero k l v kv lv distributivity of scalars over vectors vector. k(u + v) = ku + kv distributivity of vectors over scalars 1 · v = v neutral element of scalar multiplication
then (V, +, ·) is called a vector space over K. K stands for Körper ≈ field in German. math 247: honours applied linear algebra 3
Examples 1.4. of vector spaces:
• (Rn, +, ·) where + and · are standard addition of tuples and stan- dard scalar multiplication.
• Mat(n × m, K) with + standard addition of matrices, and · stan- dard multiplication by scalars.2 2 Checking axioms: (Mat(n × m, K), +) is an abelian • Let s be the set of all real sequences, i.e. (a1, a2,... ) : ai ∈ R to- group: obvious. axioms for scalar multiplication also gether with component-wise addition and scalar multiplication. satisfied. The axioms of VS follow immediately from the fact that the opera- =⇒ vector space. tions are component-wise.
• Let c be the set of all convergent real sequences together with component-wise + and ·. The only thing one must check is that c is closed under + and ·3, which is proven in analysis 1. 3 i.e. that the sum of two convergent sequences is convergent and the scalar • Let c0 be the set of all real sequences converging to 0, together multiple of a convergent sequence is convergent. with component-wise + and ·. This set is closed under + and · since the sum of two sequences with limit 0 has limit 0, same for scalar multiplication.
4 4 • Let c00 be the set of all real sequences that are eventually 0 , with i.e. all but finitely many terms of these component-wise + and ·. are different from 0.
• Let I ⊆ R be an interval. Let F(I) be the set of all real-valued functions on the interval I, together with + and · defined by:
( f + g)(x) ≡ f (x) + g(x) (k f )(x) ≡ k f (x)
• The set of all solutions to y00 + y = 0 or, more generally, the set of n n−1 solutions to any linear differential equation any + an−1y + 1 0 ··· + a1y + a0y = 0, ai ∈ R. Just need to verify that the solutions 5 5 00 are closed under + and ·. Let y1, y2 be two solutions of y + y = 00 00 0, i.e. y1 + y1 = 0 and y2 + y2 = 0, and k an arbitrary constant. ome consequences S of the axioms of a vector space: 00 00 00 (y1 + y2) + (y1 + y2) = y1 + y2 + y1 + y2 = (y00 + y ) + (y00 + y ) Theorem 1.2. Let (V, +, ·) be a VS over K. Then 1 1 2 2 = 0 00 00 (a)k · 0 = 0 ∀k ∈ K (ky1 ) + (ky1) = k(y1 + y1) = 0 (b) 0 · v = 0 ∀v ∈ V =⇒ y1 + y2 and ky1 are solutions of y00 + y = 0. (c)k · v = 0 ⇐⇒ k = 0 ∨ v = 0 4 anna brandenberger, daniela breitman
Proof. of (a)
k · 0 = k · (0 + 0) neutral el of + = k · 0 + k · 0 distribution of vectors =⇒ 0 +¨k ·¨0 = ¨k ·¨0 + k · 0 neutral element =⇒ 0 = k · 0 cancellation law
Proof. of (b)
0 · v = (0 + 0) · v neutral element = 0 · v + 0 · v distribution of scalars =⇒ 0 +¨0 ·¨v = ¨0 ·¨v + 0 · V neutral element =⇒ 0 = 0 · v cancellation law
Proof. of (c): k · v = 0. "⇐" follows from (a) and (b). "⇒": if k = 0, we are done. So let k 6= 0: must show that v = 0. 1 1 kv = 0 =⇒ (kv) = · 0 = 0 k k 1 = ( k)v = 1 · v = v associative k =⇒ v = 0
Theorem 1.3. Let (V, +, ·) be a VS over K. Then:
(a)n · v = v + ··· + v (n times), ∀n ∈ N
(b) 0 · v = 0
(c) (−1) · v = −v, ∀v ∈ V
Proof. of (a): by Induction. Base case n = 1: 1 · v = v is an axiom. Inductive step n → n + 1: assume nv = v + ··· + v for some n ∈ N.
(n + 1)v = nv + 1 · v = (v + ··· + v) +v = v + ··· + v | {z } | {z } n times n + 1 times Proof. of (b) has already been done.
Proof. of (c)
v + (−1)v = 1 · v + (−1) · v = (1 + (−1))v associative = 0 · v = 0 prev. proven
=⇒ (−1)v is the additive inverse of v, i.e. (−1)v = −v. math 247: honours applied linear algebra 5
1.3 Subspaces lecture 01/16 Definition 1.4. Let (V, +, ·) be a VS over K and let W ⊆ V. W is called a subspace of V, in symbols W ≤ V, if (W, +, ·) is a VS over K.
Examples: 1. all subspaces of R2 ({0}, all lines Theorem 1.4. Let (V, +, ·) be a VS over K and let W ⊆ V. Then W ≤ V through the origin and R2 ({0}). iff W is closed under + and · and 0 ∈ W. 2. V ≡ Mat(n × n, K) U ≤ V, where U is the set of all Proof. upper triangular n × n matrices with coefficients in K: ("⇒") Since (W, +, ·) is a VS over K, W is closed under + and ·. U ≡ {A = (aij) ∈ Mat(n × n, K) : Furthermore, 0 ∈ W since (W, +) is an (abelian) group, aij = 0, ∀i > j, 1 ≤ i, j ≤ n}
therefore contains a neutral element. 3. sequences c00 ≤ c0 ≤ c ≤ s
("⇐") Must check the axioms: 4. let Pn(K) be the set of all polyno- mials of degree at most n. Let P(K) • addition: associativity and commutativity hold on W because they be the set of all polynomials with standard + and ·. Then: hold on V. P0(K) ≤ P1(K) ≤≤ · · · ≤ Pn(K) ≤ · · · ≤ P(K) • neutral element: 0 ∈ W by condition.
• additive inverse: Let u ∈ W. Since W closed under ·, we have (−1)u = −u ∈ W.
All axioms of addition hold. All axioms of multiplication hold on W since they hold on V. =⇒ W is a VS w.r.t + and ·, i.e. W ≤ V.
Theorem 1.5.
(a) Let U, W ≤ V VS over K. Then U ∩ W ≤ V. n T (b) Let U1,..., Un ≤ V. Then U1 ∩ · · · ∩ Un = Ui ≤ V. i=1 T (c) I an arbitrary index set Ui ≤ V ∀i ∈ I. Then ≤ V. i∈I
Unions are not closed under +! For example, take V = R2 and let U and W Proof. of (a) [(b) and (c) left as exercises] be respectively vectors on the x and y axes. Their union is not closed under +: •0 ∈ U, 0 ∈ W =⇒ 0 ∈ U ∩ W adding vectors from U and W can yield any vector in R2. • Let v1, v2 ∈ U ∩ W, especially v1, v2 ∈ U =⇒ v1 + v2 ∈ U. U ∪ W ≤ V ⇐⇒ U ⊆ W ∨ W ⊆ U Similarly, v1 + v2 ∈ W. So, v1 + v2 ∈ U ∩ W =⇒ U ∩ W is closed under +. v1 ∈ U =⇒ kv1 ∈ U ∀k ∈ K • v1 ∈ W =⇒ kv1 ∈ W ∀k ∈ K =⇒ kv1 ∈ U ∩ W So U ∩ W is closed under ·.
=⇒ U ∩ W ≤ V 6 anna brandenberger, daniela breitman
Definition 1.5. Union. Let V be a VS over K, U, W ≤ V, then U + W ≡ {u + w : u ∈ U, w ∈ W}.
Theorem 1.6. Let V be a VS over K, U, W ≤ V, then
(a)U + W ≤ V
(b)U + W is the smallest subspace of V containing both U ∪ W i.e if V˜ is any subspace of V, then U + W ⊆ V˜
Proof. of a)
•0 ∈ U, 0 ∈ W =⇒ 0 ∈ U ∪ W ∵ 0 + 0 = 0
• Let v1, v2 ∈ U + W, then ∃u1, u2 ∈ U, w1, w2 ∈ W such that (v1 = u1 + w1) + (v2 = u2 + w2) = v1 + v2 = (u1 + u2) ∈ U + (w1 + w2) ∈ W ∈ U + W.
• kv1 = k(u1 + w1) = ku1(∈ U) + kw1(∈ W) ∈ U + W ⇒ U + W ≤ V
Proof. of b) Let V˜ be any subspace of V ⊂ U ∪ W. Let u ∈ U, w ∈ W be arbitrary, then u + w ∈ U + W. u + w ∈ V˜ =⇒ U + W ⊆ V˜ . So U + W is the smallest subspace of V that contains U ∪ W.
In a similar way we define u1 + ··· + un = {u1 + ··· + un : uk ∈ Uk∀1 ≤ k ≤ n}, where u1, ··· , un ≤ V, then u1 + ··· + un ≤ V.
1.4 Span and Linear Independence
Definition 1.6. Linear Combination. Let V be a VS over K, v1,..., vn ∈ V, k1,..., kn ∈ K. Then an expression k1v1 + ··· + knvn is a linear combination of v1,..., vn.
This notion can be extended to infinitely many vectors: k1v1 + k2v2 + 6 Since infinitely many additions are not . . . where all but finitely many of the ki are 0, i.e. their sum is defined. actually finite.6
∈ Definition 1.7. Span. Let V be a VS over K, v1,..., vn V. Then Note that: span{v1,..., vn} = {k1v1 + ··· + knvn : ki ∈ K, 1 ≤ i ≤ n}. span{v1,..., vn}
= span{v1} + ··· + span{vn} For infinitely many vectors v1, v2, · · · ∈ V, span{v1, v2,... } = {k1v1 + = {k1v1 : k1 ∈ K} + ··· + {knv1 : k1 ∈ K} k2v2 + ··· : ki ∈ K : all but finitely many are 0}. = {k1v1 + ··· + knvn : ki ∈ K} ≤ V Note that span{v1,..., vn} is the smallest subspace of V that contains all vectors v1,..., vk. Exercise 1.5. Prove directly that Examples 1.6. span{v1,..., vn} ≤ V math 247: honours applied linear algebra 7
e1 = (1, 0, 0, ··· , 0) = ( ··· ) 2 e2 0, 1, 0, , 0 2 1. V = R , then R = span{e1,..., en}. ··· en = (0, 0, 0, ··· , 0)
2 n 2 n 2. Consider VS Pn(K) = {1, x, x ,..., x }, then span{1, x, x ,..., x } = n , {k01 + x1x + ··· + xnx : ki ∈ x 0 ≤ k ≤ n} = Pn(K).
3. Consider VS s of all sequences with coefficients in R. s = (1, 0, 0, . . . ) 1 s2 = (0, 1, 0, . . . ) Define =⇒ span{s1, s2,... } = c00. Note that {s1, s2,... } does not span s: s3 = (0, 0, 1, . . . ) every LC of s results in a sequences that ··· eventually constantly equals 0.
Definition 1.8. Linear independence and Dependence. Let V be a VS over K, v1,..., vn ∈ V. We say that v1,..., vn are linearly inde- pendent if k1v1 + ··· + knvn = 0 ⇔ k1 = k2 = ··· = kn = 0. We say that v1,..., vn are linearly dependent if they are not linearly independent.
Theorem 1.7. Let V be a VS over K. lecture 01/18 (a) any (finite) subset of V that contains 0 is LD (linearly dependent).
(b) If v1,..., vn are LD, there exists at least one j, 1 ≤ j ≤ n such that vj is a LC of v1,..., vj−1, vj+1,..., vn. In that case, span{v1,..., vn} = span{v1,..., vj−1, vj+1,..., vn} = span{v1,..., vˆj,..., vn}.
Proof. (a) Consider 0, v1,..., vn. Then 1 · 0 + 0 · 1 + ··· + 0 · vn = 0. Non trivial LC of 0, v1,..., vn since the coefficient of the zero vector is non-zero.7 7 Remark: Even {0} is LD since 1 · 0 = 0 which is a non-trivial LC of the zero- (b) Let v1,..., vn be LD. Thus ∃k1,..., kn ∈ K, where not all of them vector resulting in the zero-vector. 0, such that k1v1 + ··· + kjvj + ··· + knvn = 0. Let kj 6= 0.
=⇒ kjvj = −k1v1 − · · · − kj−1vj−1 − kj+1vj+1 − · · · − knvn (1) ˆ k1 kj kn =⇒ vj = − v1 − · · · − vj − · · · − vn (2) kj kj kj
Thus vj is a LC of {v1, vˆj, vn}. Regarding spans: obviously span{v1,..., vˆj,..., vn} ⊆ span{v1,..., vn}. We still need to prove that span{v1,..., vn} ⊆ span{v1,..., vˆj,..., vn}. Let v = a1v1 + ··· + ajvj + ··· + anvn ∈ span{v1,..., vn}. 8 anna brandenberger, daniela breitman
Then by (3):
ˆ k1 kj kn v = a1v1 + ··· + aj(− v1 − · · · − vj − · · · − vn) + ··· + anvn kj kj kj
k1 kj−1 =⇒ v = (a1 − v1)v1 + ··· + (aj−1 − )vj−1 + (aj+1− kj kj−1
kj+1 kn )vj+1 + ··· + (an − )vn kj+1 kj
which is a LC of v1,..., vˆj,... vn.
Theorem 1.8. Let v1,..., vn ∈ V, linearly independent. Let vn+1 ∈/ span{v1,..., vn}. Then v1,..., vn, vn+1 are LI.
Proof. Let k1v1 + ··· + kjvj + ··· + knvn = 0. We need to show that k1 = ··· = kn = kn+1 = 0. Assume that kn+1 6= 0. Then,
kn+1vn+1 = −k1v1 − · · · − knvn
k1 kn =⇒ vn+1 = − v1 − · · · − vn kn+1 kn+1
=⇒ vn+1 ∈ span{v1,..., vn} Remark: Let V be a VS over K, W ⊆ V. Thus the assumption was wrong. =⇒ kn+1 = 0. Assume that V is finite dimensional. =⇒ k = ··· = k = k = v v We currently cannot conclude that 1 n n+1 0 since 1,..., n are LI. W is finite dimensional. This will be =⇒ v1,..., vn, vn+1 are LI. proven later.
1.5 Bases and Dimension
Definition 1.9. Let V be a VS over K. V is called finite dimensional if Caution: it is not enough to construct V has a finite spanning set, i.e. there are finitely many v1,..., vn ∈ V an infinitely spanning set for V to prove s.t. V = span{v1,..., vn}. V is called infinite dimensional if it is not that it is inf. dim! rather, we need to finite dimensional. show that no finite spanning set can possibly exist.
Examples 1.7. of finite and infinite dimensional VSes:
1. Rn is finite dim. n R = span{e1,..., en}
2. c00 is inf. dim.
Proof. Assume that c00 is finite dim; let s1,..., sn ∈ c00 s.t. c00 = span{s1,..., sn}. ≤ ≤ ∃ = ∀ ≥ sj, 1 j n, is in c00. Thus Nj s.t. sjk 0 k Nj. ≡ { } = ( ) Let N max N1,..., Nk and sj sj1 , sj2 ,... . Then: math 247: honours applied linear algebra 9
= ∀ ≤ ≤ ∀ ≥ sjk 0 1 j n, k N. The same holds for every s ∈ span{s1,..., sn} i.e. s = (s1, s2,..., sn−1, sn), then sk = 0 ∀k ≥ N. th Thus e.g. s˜ = (0, 0, . . . , 0, 1, 0, . . . , 0) ∈ c00 (1 at the N position) is not in the span of s1,..., sn.
Theorem 1.9. Steinitz Exchange Lemma.
Let V be a finite dimensional VS over K. Let u1,..., um ∈ V be LI and i.e. we exchanged m of the vectors let v1,..., vn be spanning, i.e. V = span{v1,..., vn}. Then, m ≤ n and v1,..., vn by u1,..., um without chang- there exist finitely many indices k1,..., kn−m s.t. u1,..., um, vk1 ,..., vkn−m ing the span. is spanning.
Proof. This is proven iteratively, in m steps, replacing one of the vj by one of the u-s at each step.
1. step: Since span{v1,..., vn} = V, ∃a1,..., an ∈ K s.t. u1 = a1v1 + ··· + anvn. Note that u1 6= 0 since u1,..., um are LI. Thus at least one of the coefficients a1,..., an 6= 0. By potentially reordering v1,..., vn, we may w.l.o.g. assume that a1 6= 0.
=⇒ a1v1 = +u1 − a2v2 − · · · − anvn 1 a2 an v1 = + u1 − v2 − · · · − vn a1 a1 a1
=⇒ v1 is a LC of u1, v2,..., vn. =⇒ span{v1, v2,..., vn} = V = span{u1, v2,..., vn}.
2. step: Since u1, v2,..., vn is spanning, ∃b1, a2,..., an ∈ K s.t. u2 = b1u1 + a2v2 + ··· + anvn. We want to show that at least one of the aj is non-zero. Assume a2 = ··· = an = 0. Then, u2 = b1u1 =⇒ b1u1 − 1 · u2 = 0 which is a non-trivial combination of u1, u2 resulting in the zero vector. But u1, u2 are LI, thus at least one aj 6= 0. w.l.o.g assume that a2 6= 0.
a2v2 = u2 − b1u1 − a3v3 − · · · − anvn
1 b1 a3 an v2 = u2 − u1 − u3 − · · · − un a2 a2 a2 a2
=⇒ v2 is a LC of u1, u2, v3,..., vn. Thus V = span{u1, v2, v3,..., vn} = span{u1, u2, v3,..., vn}. 10 anna brandenberger, daniela breitman
··· th After the m step, we obtain a spanning set {u1,..., um, vm+1,..., vn} of V. Especially m ≤ n. lecture 01/23
Theoretical Applications of the Steinitz Exchange Lemma
Theorem 1.10. Every subspace of a finite dimensional VS over K is finite dimensional. Examples 1.8. Proof. Let V be a finite dimensional VS, U ≤ V.V is finite dimen- 1. The VSs of all sequences with coeffi- cients in K is infinite dimensional. sional, thus it has a finite spanning set {v1, ··· , vn }. We need to show 2. We already know that c00 is infinite that U has a finite spanning set. We proceed recursively. dimensional c00 ≤ s. If s were finite Assume this is not the case: dimensional then so would be c00 6= {~ } 6= ∈ { } by previous Thm but c00 is infinite Then U 0 . Let u1 0, u1 U. Then u1 is linearly indepen- dimensional =⇒ s is infinite dent but not spanning. Thus ∃u2 ∈ U with u2 ∈/ span{u1}. By the dimensional. previous theorem, {u1, u2} is linearly independent; however, it is not 3. Similarly, c0 is also infinite dimen- sional since c00 ≤ c0. spanning, thus ∃u3 ∈/ span{u1, u2}, etc. After n+1 iterations, we’ve obtained a linearly independent set {u1, ··· , un+1} ⊆ U ⊆ V. But V has a spanning set {u1, ··· , un} containing n elements! =⇒ contradiction to Steiniz! Our as- sumption was wrong. Thus U is finite dimensional.
Definition 1.10. Basis. Let V be a VS over K. A subset B of V is Exercise 1.9. Prove that any superspace of an infinite dimensional VS over K is called a basis of V if: infinite dimensional.
(i) B is linearly independent.
(ii) B is spanning. 8 Showing that B is a basis for V:
Examples 1.10. Proof.B is spanning. It is also LI: Let a , a , ··· , a ∈ K n 0 1 n = R = { ··· } 2 1. V , B e1, , en such that a0 + a1x + a2x + ··· + n an x = 0 ( =⇒ zero polynomial) n 2 n 2. V = C , B = {e1, ··· , en} = 0 + 0 · x + 0 · x + ··· + 0 · x ∀x ∈ K Two polynomials are identically equal 2 n 8 3. V = Pn, Polynomials of degree ≤ n, B = {1, x, x , ··· , x } iff they have the same coefficients. 0 if (k, l) 6= (i, j) 4. V = mat(mxn, K). Let Mij = (mij), where mkl 1 if (k, l) = (i, j)
9 9 then {Mij}, 1 ≤ i ≤ m, and 1 ≤ j ≤ n is a basis for V. See assignment 3.
5. m = n = 2 : {4’identity’ matrices} is a basis for Mat(2 × 2, K). Q: Does every VS have a basis? A: Yes but we’ll prove this only in the Theorem 1.11. Every finite dimensional VS has a basis. finite dimensional case. math 247: honours applied linear algebra 11
Proof. By Casting Out Algorithm (Steiniz proof), let V be finite dimen- sional, by definition of finite dimensional, V has a finite spanning set {u1, ··· , un}. If {u1, ··· , un} is linearly independent, it is a basis and done. If {vi, ··· , vn} is linearly dependent ∃j : 1 ≤ j ≤ n such that vj is a linear combination of v1, ··· , vˆj, ··· , vn. W.l.o.g., assume that j = n. As proven in previous theorem, we then have that span{v1, ··· , vn} = span{v1, ··· , vn−1} =⇒ {v1, ··· , vn−1} spans V. After repeating this arguments finitely many times, we ob- tain v1, ··· , vk, k ≤ n such that {v1, ··· , vk} is spanning and linearly independent and thus a basis. Q: What is a basis for {0}? Another version of the proof: Bottom Up Algorithm: A: B = {}, since span{} = {0}. Let V be finite dimensional VS over K. Say V has a spanning set of n elements. If V = {0}, B = {} and were done. If not, pick any v1 ∈ V, v1 6= 0, then {v1} is linearly independent. It if is also spanning, we’re done. If not, let v2 ∈ V be any vector with v2 ∈/ span{v1}, then by previous theorem, {v1, v2} is linearly independent if {v1, v2} is spanning, we’re done. If not, ∃v3 ∈ V : v3 ∈/ span{v1, v2}, etc... until spanning. So by Steiniz, this algorithm has to terminate after at most
n steps. We obtain {v1, ··· , vk}, k ≤ n, linearly independent and spanning, i.e. a basis.
Example 1.11. Polynomials of degree ≤ 2, coefficients ∈ K. We know that {1, x, x2} is a basis and thus spanning. Finding another basis.
Proof. Let v0 ≡ 2, then span{v0} 6= P2. Let v1 ≡ 3 + 5x, then v1 ∈/ span{v0} =⇒ {v0, v1} is linearly independent, not spanning (no polynomial of degree 2 is in the span). 2 Let v2 ≡ 7 − 5x + 11x , then v2 ∈/ span{v0, v1} =⇒ {v0, v1, v3} is linearly independent. It needs to be spanning as well since otherwise ∃v3 ∈/ span{v0, v1, v2} =⇒ {v0, v1, v2, v3} linearly independent =⇒ contradiction to Steinitz. =⇒ {v0, v1, v2} is linearly independent and spanning, =⇒ {v0, v1, v2} is a basis for P2.
Theorem 1.12. Let V be a finite dimensional VS over K. Let {u1,..., um} and {u1,..., un} be bases for V. Then m = n, i.e. any 2 bases for a finite dimensional vector space contain the same number of elements.
Proof. Especially {u1, ··· , um} is linearly independent and {u1, ··· , un} is spanning. By Steiniz =⇒ m ≤ n. But we also have that {u1, ··· , un} is linearly independent and {u1, ··· , um} is spanning. By Steiniz =⇒ n ≤ m. ∴ n = m. Remark: the notion of dimension is well-defined by previous theorem. 12 anna brandenberger, daniela breitman
Definition 1.11. Dimension. Let V be finite dimensional VS over K. Let {v1, ··· , vn} be any basis for V. Then n is called the dimension of V. lecture 01/25 Examples 1.12. of dimension:
n 1. R as VS over R has dimension n since {e1,..., en} is a basis.
n 2. C as VS over C has dimension n since {e1,..., en} is a basis.
2 n 3. Pn(K) VS over K has dimension n + 1 since {1, x, x ,..., x } is a basis with n + 1 elements.
4. Mat(m × n, K) VS over K has dimension m · n since 1 if (k, l) = (i, j) {Eij = (mkl) where mkl = } 0 if (k, l) 6= (i, j) is a basis with m · n elements.
Some Theorems on Bases and Dimension
Theorem 1.13. Let V be a finite dimensional VS over K, B = {v1,..., vn} a basis for V. Let u ∈ V. Then there exist uniquely determined
k1,..., kn ∈ K s.t. v = k1v1 + ··· + knvn.
Proof.B is spanning, thus v can be written as a LC of v1,..., vn. Let v = k1v1 + ··· + knvn and v = l1v1 + ··· + lnvn.
=⇒ 0 = (k1v1 + ··· + knvn) − (l1v1 + ··· + lnvn)
= (k1v1 − l1v1) + ··· + (knvn − lnvn)
= (k1 − l1)v1 + ··· + (kn − ln)vn
Since v1,..., vn are LI we have that k1 − l1 = 0, . . . , kn − ln = 0. =⇒ k1 = l1,..., kn = ln. =⇒ v can be expressed in exactly one way as a LC of v1,..., vn.
Theorem 1.14. Let V be a finite dimensional VS over K, let U ≤ V. Then dimU ≤ dimV.
Remark: Note that U is finite dim since V is finite dim. Thus, dimU is defined. Proof. Let {u1,..., um} be a basis for U, {v1,..., vn} be a basis for V. Note that {u1,..., um} is LI and {v1,..., vn} is spanning. By Steinitz (Theorem 1.9), m ≤ n =⇒ dimU ≤ dimV.
Theorem 1.15. Let V be a finite dimensional VS over K with dimV = n.
Let B = {v1,..., vn} be LI. Then B is a basis for V. math 247: honours applied linear algebra 13
Proof.B is LI. We need to show it is spanning. Assume it is not. Then ∃ vn+1 ∈ V, vn+1 ∈/ span{v1,..., vn} =⇒ {v1,..., vn, vn+1} is LI. But since dimV = n, V has a spanning set of length n. contra- diction with Steinitz. Thus B is also spanning and ∴ a basis.
Theorem 1.16. Let V be a finite dimensional VS over K with dimV = n.
Let B = {v1,..., vn} be spanning. Then B is a basis for V.
Proof.B is spanning. We need to show that B is LI. Assume that B is LD. Then ∃j : 1 ≤ j ≤ n s.t. vj is a LC of v1, ··· , vˆj, ··· , vn and V = span{v1, ··· , vn} = span{v1, ··· , vˆj, ··· , vn}. i.e. V has a spanning set of n − 1 vectors. But since dimV = n, it Example 1.13. V = P2(K), dimV =3., has a LI set of length n, contradiction with Steinitz. B = {3, 5 − 9x, 7 + 3x + 13x2}. =⇒ B is also LI and thus a basis. B is LI for degree reasons, thus B is also spanning.
Theorem 1.17. Let V be a finite dimensional VS over K, U ≤ V. Let
Bu = {u1,..., um} be a basis for U. Then Bu can be extended to a basis Bv for V, i.e. ∃ um+1,..., un ∈ V s.t. Bv ≡ {u1,..., um, um+1,..., un} is a basis for V.
Proof. This is the same argument we used for proving the existence of a basis via the Bottom Up algorithm: we start with Bu in the initial step.10 10 Exercise: Work out the details!
Theorem 1.18. Let V be a finite dimensional VS over K, U ≤ V,W ≤ V. Then dim(U + W) = dimU + dimW − dim(U ∩ W).
Proof. Let {v1, ··· , vk} be a basis for U ∩ W. U ∩ W ≤ U. ∴ by the previous theorem, we can extend this to a basis {v1, ··· , vk, uk+1, ··· , um} for U. Similarly, we can find wk+1, ··· , wn s.t. {v1, ··· , vk, wk+1, ··· , wn} is a basis for W.
Now consider the set S ≡ {v1, ··· , vk, uk+1, ··· , um, wk+1, ··· , wn}. We will show that S is a basis for U + W.
1. step: S spans U + W, i.e. span{S} = U + W. First show "⊆". Let v ∈ spanS be arbitrary. Then:
v = a1v1 + ··· + akvk ∈ U ∩ W
+ bk+1uk+1 + ··· + bmum ∈ U
+ ck+1wk+1 + ··· + cnwn ∈ W
where (ai, bj, cl) are scalars.
=⇒ v ∈ U + W ∴ span{S} ⊆ U + W Next show "⊇": Let v ∈ U + W be arbitrary.
Then ∃u ∈ U, w ∈ W s.t. v = u + w ∈ span{v1, ··· , vk, wk+1, 14 anna brandenberger, daniela breitman
··· , wn} ∈ span{v1, ··· , vk, uk+1, ··· , un} ∈ spanS. =⇒ U + W ⊆ spanS =⇒ U + W = span{S} i.e. S spans U + W.
2. step: S is LI. Let the as, bs and cs ∈ K be such that:
0 = a1v1 + ··· + akvk
+ bk+1uk+1 + ··· + bmum (*) + ck+1wk+1 + ··· + cnwn
=⇒ a1v1 + ··· + akvk ∈ U ∩ W
+ bk+1uk+1 + ··· + bmum ∈ U
= −ck+1wk+1 − · · · − cnwn ∈ U ∩ W (since =)
0 0 Since {v1,..., vk} is a basis for U ∩ W, ∃a1, ··· , ak ∈ K s.t. 0 0 −ck+1wk+1 − · · · − cnwn = a1v1 + ··· + akvk
=⇒ 0 · v1 + ··· 0 · vk − ck+1wk+1 − · · · − cnwn 0 0 = a1v1 + ··· + akvk + 0 · wk+1 + ··· + 0 · wn
Note that {v1, ··· , vk, wk+1, ··· , wn} is a basis for W and thus especially LI. Thus, the coefficients in the previous equations are uniquely determined. 0 0 =⇒ [a1 = ··· = ak = 0] and ck+1 = ··· = cn = 0. Plugging this into (*):
0 = a1v1 + ··· + akvk + bk+1uk+1 + ··· + bmum
But {v1, ··· , vk, uk+1, ··· , um} is a basis for U. Thus, a1 = ··· = ak = 0 and bk+1 = ··· = bm = 0. This proves that S is LI. Hence, S is a basis for U + W. 11 Proof. Planes through the origin are the 2-dim subspaces of R3. Let U, W be dim(U + W) = |S| (size of S) 2 such planes. Then: = k + (m − k) + (n − k) (vs, us, ws) dim(U + W) = dimU + dimW − dim(U ∩ W) dim(U ∩ W) = dimU + dimW − dim(U + W) = m + n − k = 2 + 2 − dim(U + W) = dimU + dimW − dim(U ∩ W) = 4 − dim(U + W) (U + W ≤ R3) Example 1.14. Any two planes through the origin in R3 have non ≥ 4 − 3 = 1 trivial intersection, i.e. they intersect in at least a line through the =⇒ the intersection of the planes is a origin. 11 subspace of R3 of dim ≥ 1 and is thus at least a line.
1.6 Direct Sums lecture 01/30 Recall: V VS over K, and U ≤ V, W ≤ V, then U + W ≡ {u + w : u ∈ U, w ∈ W} and U + W ≤ V math 247: honours applied linear algebra 15
Definition 1.12. A sum U + W of subspaces U, W of V is called direct - in symbols, U ⊕ V) if ∀v ∈ U + W, the representation v = u + w, with u ∈ U, w ∈ W is uniquely determined. Similarly a sum U1 + ··· + Un of subspaces of V is called direct if ∀v ∈ U1 + ··· + Un, the representation v = u1 + ··· + un, uj ∈ Uj ∀1 ≤ j ≤ n is uniquely defined.
Theorem 1.19. U + W of subspaces of V is direct iff U ∩ W = {0}.
Proof. " ⇒ " Let U + W be direct. Let v ∈ U ∩ W, then: 0 = 0 + 0 = v + −v |{z} |{z} |{z} |{z} |{z} ∈U+W ∈U ∈W ∈U ∈W Since the representation is unique, v = 0 =⇒ U ∩ W = {0}. " ⇐ " Let v ∈ U + W be arbitrary. Let v = u1 + w1 and v = u2 + w2, for u1, u2 ∈ U, w1, w2 ∈ W.
v = u1 + w1
v = u2 + w2
0 = (u1 − u2) + (w1 − w2)
=⇒ u1 − u2 = w2 − w1 | {z } | {z } ∈U ∈W | {z } | {z } ∈U∩W ∈U∩W
=⇒ u1 − u2 = 0 = w2 − w1
=⇒ u1 = u2 ∧ w1 = w2
=⇒ v is uniquely determined =⇒ U + W is direct.
Exercise 1.15. Let v1, ··· , vn ∈ V be LI, then span{v1} + ··· + span{vn} is direct. 16 anna brandenberger, daniela breitman
2 Linear Maps lecture 01/30 2.1 Mappings and Associated Subspaces
Definition 2.1. Let V and W be VS over K. A function L : V → W is called linear if: (Same K for both V and W.)
1. L(v1 + v2) = L(v1) + L(v2) v1, v2 ∈ V L(v1), L(v2) ∈ W 2. L(kv1) = kL(v1)
Theorem 2.1. Let L : V → W be linear, then L(0 ∈ V) = 0 ∈ W.
Proof.
– version 1. L(0) = L(0 + 0) = L(0) + L(0) definition (1.) ¨ ¨ =⇒ 0 +¨L(¨0) = L(0) +¨L(¨0) cancellation =⇒ L(0) = 0
– version 2. L(0) = L(0 · 0) = 0 · L(0) definition (2.) = 0 theorem
Warning 1. In calculus, the so-called Definition 2.2. Subspaces associated with linear maps. Let L : V → "linear" maps f : R → R, x 7→ ax + b are W be linear. Define: only linear if b = 0!
(a) The kernel, ker L is defined as ker L ≡ {v ∈ V : L(v) = 0} ⊆ V.
(b) The image im L is defined as im L ≡ {L(v) : v ∈ V} ⊆ W
Remark 2. Axler calls ker L the nullspace Theorem 2.2. Let L : V → W be linear, then: of L and im L the range of L.
1. kerL ≤ V
2. imL ≤ W
Proof.
1.0 ∈ ker L since L(0) = 0. Let v1, v2 ∈ ker L, then L(v1 + v2) = L(v1) + L(v2) = 0 + 0 = 0 =⇒ v1 + v2 ∈ ker L and L(kv1) = kL(v1) = k · 0 = 0 =⇒ kv1 ∈ ker L ker L ≤ V
2. Since L(0(∈ V)) = 0(∈ W) =⇒ 0(∈ W) ∈ im L. Let w1, w2 ∈ im L. Thus ∃v1, v2 ∈ V with L(v1) = w1 and L(v2) = w2, then L(v1 + v2) = L(v1) + L(v2) = w1 + w2 =⇒ w1 + w2 ∈ im L and L(kv1) = kL(v1) = kw1 =⇒ kw1 ∈ im L =⇒ im L ≤ W. math 247: honours applied linear algebra 17
Examples 2.1.
1. L defined as: L : Kn → km is linear: x 7→ A · x |{z} |{z} m×n n×1 | {z } 12 m×1 A(x + x˜) = Ax + Ax˜ =⇒ L(x + x˜) = L(x) + L(x˜) • ker L = nullspace(A) A(kx) = k(Ax) =⇒ L(kx) = kL(x) • im L = colspace(A) (c.f. assign. 2) =⇒ L is linear.12
2. D defined as: D : P(R → P(R) is linear: 0 p 7→ p (derivative) 13 ( + ) = ( + )0 = 0 + 0 = ( ) + ( ) D n q n q n q D n D q • ker D = P0(R) 0 0 D(kn) = (kn) = kn = kD(n) • im D = P(R) =⇒ D is linear.13 14
3. (a) D defined as: D : Pn(R → Pn−1(R) is linear as shown • ker D = P0(R) 0 p 7→ p • im D = Pn−1(R) above.14 D is surjective.
(b) Same function defined on D : Pn(R) → Pn(R) is not surjec- tive.15 15
• ker L = P0(R) = (R) ∈ R 4. Integration: V P , for a, b • im L = Pn−1(R) Pn(R) (a) L defined as: L : P(R) → R is linear: Z b p 7→ pdx a
Z b Z b Z b L(p + q) = (p + q)dx = pdx + qdx = L(p) + L(q) a a a Z b Z b L(kp) = kpdx = k pdx = kL(p) a a =⇒ L linear. (b) a ∈ R: L : P(R) → P(R) ker L = {0} Z x p 7→ p(t)dt im L = Set of all polynomials with a a root at x = a. L is linear as shown above.
(c) L defined as follows is linear: L : P(R) → Pn+1(R) Z x r 7→ rdt a
Assignment 4: V = s, s = {(x1, x2, x3, ··· ) : xi ∈ R, ∀i} Left-shift: (x1, x2, x3, ··· ) 7→ (x2, x3, x4, ··· ) Right-shift:(x1, x2, x3, ··· ) 7→ (0, x1, x2, ··· ). 18 anna brandenberger, daniela breitman
Definition 2.3. I ⊆ R interval. c0(I)≡ Set of all continuously functions on I. c1(I)≡ Set of all continuously differentiable functions ··· ··· cn(I)≡ Set of all n-times continuously differentiable functions on I. ··· c∞(I) ≡ Set of all functions on I with derivatives of all orders.
Exercise 2.2. c0(I), cn(I), c∞(I) are VS over R. Show c∞(I) ≤ · · · ≤ cn(I) ≤ · · · ≤ c1(I) ≤ c0(I).
− D : cn(I) → cn 1(I) or D : c∞(I) → c∞(I) are all linear. lecture 02/01
Theorem 2.3. Let V, W be VS over K, L : V → W linear. Then L is injec- tive iff ker L = {0}.
Proof. "⇒" Let L be injective16 and let v ∈ ker L. Then: 16 injective = one to one L(v) = 0 = L(0) =⇒ v = 0 =⇒ ker L = {0}. "⇐" Let ker L = {0}. Let v, v˜ ∈ V s.t. L(v) = L(v˜). =⇒ L(v) − L(v˜) = 0 = L(v − v˜) =⇒ v − v˜ ∈ ker L =⇒ v − v˜ = 0 =⇒ v = v˜ =⇒ L is injective.
Theorem 2.4. Let V be a finite dimensional VS over K, W a VS over K
and {v1, ··· , vn} a basis for V. Let w1, ··· , wn be arbitrary vectors in W. Then there exists a uniquely determined linear map L : V → W s.t. L(v1) = w1, ··· , L(vn) = wn.
Proof. Definition of L: Let v ∈ V. ∃k1, ··· , kn ∈ K s.t. v = k1v1 + ··· + knvn. Define L(v) ≡ k1w1 + ··· + knwn. Note that L is a function since k1, ··· , kn are uniquely determined.
Linearity of L: Let v, v˜ ∈ V.
v = k1v1 + ··· + knvn
v˜ = l1v1 + ··· + lnvn
v + v˜ = (k1 + l1)v1 + ··· + (kn + ln)vn
=⇒ L(v + v˜) = (k1 + l1)w1 + ··· + (kn + ln)wn
= (k1w1 + ··· + knwn) + (l1w1 + ··· + lnwn) = L(v) + L(v˜) math 247: honours applied linear algebra 19
and
kv = (kk1)v1 + ··· + (kkn)vn
=⇒ L(kv) = (kk1)w1 + ··· + (kkn)wn
= k[k1w1 + ··· knwn] = kL(v)
=⇒ L is linear.
Uniqueness: Let L˜ : V → W be linear with L˜ (v1) = w1, ··· , L˜ (vn) = wn.
=⇒ L˜ (k1v1) = k1w1, ··· , L˜ (knvn) = knwn
=⇒ L˜ (k1v1 + ··· + knvn)
= L˜ (k1v1) + ··· + L˜ (knvn)
= k1w1 + ··· + knwn
= L(k1w1 + ··· + knvn) =⇒ L˜ (v) = L(v) ∀v ∈ V =⇒ L˜ = L
=⇒ L is uniquely determined.
Theorem 2.5. Dimension Formula. Let V be a finite dimensional VS over K, W VS over K. Let L : V → W be linear. Then dim ker L + dim im L = dimV.
Proof. Let {u1, ··· , uk} be a basis for ker L and let {w1, ··· , wn} be a basis for im L. Since w1, ··· , wn ∈ im L ∃v1, ··· , vn ∈ V with Figure 1: Representation of Thm 2.5
L(v1) = w1, ··· , L(vn) = wn. Note that the vi are not necessarily uniquely determined! Let B ≡ {u1, ··· , uk, v1, ··· , vn}. claim: B is a basis for V.
1.Show B is LI:
Let a1 u1 + ··· + ak uk + b1 v1 + ··· + bn vn = 0 (?). Applying L to both sides:
=⇒ L(a1u1 + ··· akuk + b1v1 + ··· + bnvn) = L(0) = 0
=⇒ a1L(u1) + ··· + ak L(uk) + b1L(u1) + ··· + bn L(vn) = 0 | {z } | {z } | {z } | {z } =0 =0 =w1 =wn
=⇒ b1w1 + ··· + bnwn = 0
since w1, ··· , wn are LI.
It follows that b1 = ··· = bn = 0 in (?): a1u1 + ··· + akuk = 0. Since u1, ··· , uk are LI, =⇒ a1 = ··· = ak = 0, =⇒ B is LI.
2.Show B is spanning: 20 anna brandenberger, daniela breitman
Let v ∈ V be arbitrary. Let w ≡ L(v). Then w ∈ im L. Thus ∃b1, ··· , bn ∈ K s.t. w = b1 w1 + ··· + bn wn. Consider v˜ ≡ b1 v1 + ··· + bn vn. Then:
L(v − v˜) = L(v) − L(v˜) = 0 |{z} |{z} =w =b1w1+···+bnwn=w
=⇒ v − v˜ ∈ ker L =⇒ ∃a1, ··· , aks.t.
v − v˜ = a1u1 + ··· + akuk
=⇒ v = a1u1 + ··· + akuk + b1v1 + ··· + bnvn =⇒ B is spanning
=⇒ B is a basis for V. =⇒ We have: dimV = k + n = dim ker L + dim im L
Examples 2.3.
1. D : Pn(K) → Pn(K) differentiation. ker D consists of all constant polynomials, i.e. ker D = P0(K). dim ker D = 1. im D consists of all polynomials of degree at most n − 1. i.e. im D = Pn−1(K) =⇒ dim im D = n.
? dim Pn(K) = dim ker D + dim im L assignment 2: 7→ m → n + = + L : x Ax, L : K K n 1 1 n X nullspace(A) = ker L colspace(A) = im L 2. Let A ∈ Mat(n × m, R). since dim ker L + dim im L = dim V =⇒ dim nullspace(A) +rank(A) = m Consider Ax = 0. | {z } nullity(A) Learned in previous linear algebra courses that: dim nullspace(A) + dim colspace(A) = m (dim colspace(A) = dim rowspace(A) = rank(A))
a11 ··· a1n 3. tr : Mat(n × n, K) → K "trace": ······ 7→ a11 + ··· ann. an1 ··· ann tr is linear: a11+b11 ··· a1n+b1n tr(A + B) = tr ······ an1+bn1 ··· ann+bnn
= (a11 + b11) + ··· + (ann + bnn)
= (a11 + ··· + ann) + (b11 + ··· + bnn) = trA + trB
ka11 ··· ka1n tr(kA) = tr ······ kan1 ··· kann
= ka11 + ··· + kann = k(a11 + ··· + ann) = k · trA math 247: honours applied linear algebra 21
ker L is the set of all matrices where the sum of all elements along the main diagonal is 0. im L = K, dim im L = 1 dim Mat(n × n, K) = n2 =⇒ dim ker L = n2 − 1.
4. Let A ∈ Mat(n × m, K) with rows = n < columns = n. The lin. hom. system Ax = 0 has non-trivial solutions. Let L : x 7→ Ax : Km → Kn Then Ax = 0 has non-trivial sol. iff L has a non-trivial kernel, i.e. iff dim ker L > 0. By dimensional Formula, dim ker L + dim im L = dimKm
=⇒ dim ker L = dim Km − dim im L = m − dim colspace(A) = m − dim rowspace(A) = m − n > 0 | {z } ≤n =⇒ ker L is non-trivial. =⇒ Ax = 0 has non-trivial solutions. lecture 02/06
2.2 Endomorphisms and Isomorphisms
Definition 2.4. An endomorphism is a linear map L : V → V, V is VS over K.
Figure 2: Injective (non surjective) map. Theorem 2.6. Let V be finite dimensional VS over K, L : V → V be an en- domorphism, then L is bijective iff L is surjective iff L is injective.
Lemma 2.7. Lemma: Let V be a finite dimensional VS over K, and let U ≤ V s.t dim U = dimV, then U = V. Proof. Let {v , ··· , v } be a basis for U =⇒ {v , ··· , v } is LI, 1 n 1 n Figure 3: Surjective (non injective) map. where n = dim U = dim V =⇒ {v1, ··· , vn} is a basis for V, especially, it spans V =⇒ U = V
Proof. of Theorem 2.6. There are 2 non-trivial implications:
1. injective =⇒ surjective: Figure 4: Bijective map. By Dimension formula, dim ker L + dim imL = dim V. L injective =⇒ kerL = {0}. =⇒ dim kerL = 0 =⇒ dim imL = dim V, where imL ≤ V. By Lemma 2.7, im L = V =⇒ L is surjective.
2. Surjective =⇒ injective: Let L be surjective =⇒ im L = V =⇒ dim im L = dim V dim ker L = 0 =⇒ ker L = {0} =⇒ L is injective. 22 anna brandenberger, daniela breitman
Remark 3. Comparable results are wrong in almost any other context, e.g.: f : R → R is injective but not surjective, 7→ x x e Remark 4. The result is also wrong in while f : R → R is surjective but not injective. general for endomorphisms between infinite dimensional VS (See Ass. 5). x 7→ x3 − x
Coordinate Vectors
All bases in this section are considered as ordered. B = (v1, v2, ··· , vn) i.e bases will have a first, second, etc. element. Vectors in Kn will al- ways be considered as column vectors in this section.
Definition 2.5. Let V be finite dimensional VS over K, B = (v1, v2, ··· , vn) is its basis. Let v ∈ V; then ∃ uniquely deter- mined a1, ··· , an ∈ K s.t v = a1v1 + ··· + anvn. The column vector a1 ··· ∈ Kn uniquely represents v. It is called the coordinate vector of an v with respect to B, written as [v]B.
Many textbooks use {v1, ··· , vn} even if they mean ordered bases. Theorem 2.8. Let V be finite dimensional, B an ordered basis for V, then n the coordinate map []B : V → K , v 7→ [v]B is linear and bijective (this is an isomorphism).
Proof. 1.Linearity of []B: Let B = (v1, v2, ··· , vn). 1 0 [v1]B = 1 · v1 + 0 · v2 + ··· + 0 · vn = ··· = e1 0 0 ! 1 [v2]B = 0 = e2 ··· ··· 0 0 0 [vn]B = ··· = en 1 By the Existence Theorem (Theorem 2.4) for linear maps, ∃ a uniquely determined linear map L : V → Kn with: L(v1) = e1, ··· , L(vn) = en, where a1 a2 L(a1v1 + ··· + anvn) = a1e1 + ··· + anen = ··· an =⇒ L = [ ]B =⇒ []B is linear. Exercise: Prove this directly, i.e with- out the use of the Existence Theorem.
2.Bijectivity of []B: 0 Injectivity: [v]B = ··· ⇐⇒ v = 0 · va + ··· + 0 · vn = 0 i.e 0 ker[]B = {0} =⇒ []B is injective. a 1 n Surjectivity: Let ··· ∈ K be arbitrary. Let v ≡ a1v1 + ··· + anvn, an a1 then [v]B = ··· =⇒ []B is surjective. an =⇒ []B is bijective =⇒ isomorphism. math 247: honours applied linear algebra 23
Theorem 2.9. on isomorphisms: Let L : V → W be an isomorphism. Since L is bijective, it has an inverse map L−1 : W → V, then L−1 is linear.
Proof. Let w1, w2 ∈ W, v1, v2 ∈ V with L(v1) = w1 and L(v2) = w2, then
−1 −1 L (w1 + w2) = L (L(v1) + L(v2)) −1 = L (L(v1 + v2)) −1 −1 = v1 + v2 = L (w1) + L (w2) −1 −1 −1 =⇒ L (w1 + w2) = L (w1) + L (w2)
Let k ∈ K, then
−1 −1 L (kw1) = L (kL(v1)) −1 −1 =⇒ L (L(kv1)) = kL (w1)
=⇒ L−1 linear.
Theorem 2.10. Let V be VS over K of dimension n, then V is isomorphic to Kn.
Proof. Let B be ordered basis for V, then [] : V → Kn is an isomor- phism by Theorem 2.8.
2.3 Matrix Representation of Linear Maps Q: Does there exist a n × m ma- trix A with coefficients in K Idea: Let V, W be finite dimensional VS over K, L : V → W, let B be that mimics the action of L, i.e. [ ( )] 0 = ·[ ] ∀ ∈ 0 = = L v B A v B v V ordered basis for V, B ordered basis for W. dim V m, dim W n. |{z}× |{z} n m m×1 For v ∈ V, w ∈ W. [v] ∈ Km, [w] ∈ Kn. Before we construct A in | {z } B B n×1 the general case, let’s revesit the classical case of V = Km, W = Kn, B A: Such a matrix, if it exists, would standard basis on Km, and B0 standard basis on Kn. rightfully be called the matrix represen- tation of L wrt B and B0.
a11 ... a1m ! = . . . A . .. . an1 ... anm
The linear maps L : Km → Kn are of the form L(x) = Ax, where A is an n × m matrix. What can we say about the coefficients of A?
a11 ! = ( ) = . Ae1 L e1 . i.e the first column of A. |{z} a ∈Km n1 i.e. ··· A = (L(e1) | L(e2) | · · · | L(em)) a1m ! Ae = L(e ) = . m A m m . i.e the th column of . |{z} a ∈Km nm lecture 02/08 24 anna brandenberger, daniela breitman
Definition 2.6. Let L : V → W linear, V, W VS over K with dim V = m, dim W = n, B = {v1, ··· , vm} basis for V, {w1, ··· , wn} basis for W. Define [L]B0,B the matrix representation of L with respect to B and B0 by:
m columns z }| { [L]B0,B = [L(v1)]B0 | · · · | [L(vm)]B0 | {z } | {z } n×1 n×1 | {z } n×m matrix
Theorem 2.11. L : V → W linear, dim V = m, dim W = n, B = 0 {v1, ··· , vm} basis for V, B = {w1, ··· , wn} basis for W. Then:
[L(v)]B0 = [L]B0,B [v]B ∀v ∈ V | {z } | {z } |{z} n×1 n×m m×1 | {z } n×1
Proof. Let v ∈ V be arbitrary, a1 ! v = a v + ··· + a v =⇒ [v] = . 1 1 m m B . . Then: am
a1 ! [L] 0 [v] = ([L(v )] 0 | · · · | [L(v )] 0 ) · . B ,B B 1 B m B . am
= a1[L(v1)]B0 + ··· + am[L(vm)]B0
= [a1L(v1)]B0 + ··· + [am L(vm)]B0 n Note that []B0 : W → K and L are linear. = [a1L(v1) + ··· + am L(vm)]B0
= [L(a1v1 + ··· + amvm)]B0 = [L(v)]B0
Theorem 2.12. Let L : V → W,V, W finite dimensional, L linear, B 0 basis for V and B basis for W, A ≡ [L]B0,B. Then ker L is isomorphic to nullspace(A) and im L is isomorphic to colspace(A). The isomorphism is given by:
ker L → nullspace(A)
v 7→ [v]B and im L → colspace(A)
w 7→ [w]B0
Proof. Let v ∈ ker L ⇔ L(v) = 0 ⇔ [L(v)]B0 = 0 = [L]B0,B[v]B ⇔ [v]B ∈ nullspace(A) Let w ∈ im L ⇔ ∃v ∈ V : L(v) = w ⇔ [L(v)]B0 = [w]B0 = [L]B0,B[v]B ⇔ [w]B0 ∈ colspace(A) math 247: honours applied linear algebra 25
Example 2.4. D : P2(R) → P2(R) endomorphism [ ] = 0 = ( 2) ( ) ≡ 0 We can use D B,B to differentiate a B B 1, x, x , D p p polynomial via matrix multiplication: 2 0 [D]B,B = ? [D]B,B = [D]B Let p(x) = 3 − 2x + 5x , p (x) = 3 −2 + 10x, [p]B = −2 5 2 0 1 0 3 −2 D(1) = 0 = 0 · 1 + 0 · x + 0 · x 0 0 2 −2 = 10 0 0 0 5 0 2 0 1 0 D(x) = 1 = 1 · 1 + 0 · x + 0 · x =⇒ [D]B,B = 0 0 2 which is the coordinate vector of the 0 0 0 D(x2) = 2x = 0 · 1 + 2 · x + 0 · x2 polynomial −2 · 1 + 10 · x = −2 + 10x = p0(x)
ker D is the set of all constant polynomials, i.e. P0(R) im D = P1(R) 0 1 0 0 1 0 nullspace[D]B,B: 0 0 2 ∼ 0 0 1 with columns a0, a1, a2 0 0 0 0 0 0 =⇒ a1 = a2 = 0, a0 free parameter. n a0 o n 1 o a0 =⇒ nullspace(A) = 0 : a0 ∈ R = span 0 0 is exactly the coordinate vectors of 0 0 0 constant polynomials.
n 1 0 o n 1 0 o colspace([D]B0,B) = span 0 2 = span 0 1 0 0 0 0 n a0 o = a1 : a0, a1 ∈ R 0 which is precisely the set of all coordinate vectors of linear polynomi- als.
Example 2.5. L : Mat(2 × 2, R) → Mat(2 × 2, R) t A 7→ A − A Note that L is linear since: L is linear. Now let: L(A + B) = (A + B) − (A + B)t = A + B − At − Bt = 0 = { 1 0 0 1 0 0 0 0 } B B 0 0 , 0 0 , 1 0 , 0 1 t t | {z } | {z } | {z } | {z } = (A − A ) + (B − B ) M11 M12 M21 M22 = L(A) + L(B) L(kA) = kA − (kA)t = kA − kAt 1 0 0 0 = k(A − At) = kL(A) L 0 0 = 0 0 = 0 · M11 + 0 · M12 + 0 · M21 + 0 · M22 0 1 0 1 =⇒ L is linear. L 0 0 = −1 0 = 0 · M11 + 1 · M12 + (−1) · M21 + 0 · M22 0 0 = 0 −1 = · + (− ) · + · + · L 1 0 1 0 0 M11 1 M12 1 M21 0 M22 0 0 0 0 L 0 1 = 0 0 = 0 · M11 + 0 · M12 + 0 · M21 + 0 · M22 0 0 0 0 ! 0 1 −1 0 =⇒ [L]B,B = 0 −1 1 0 0 0 0 0
≡ 2 1 ( ) = 2 1 − 2 3 = 0 −2 Let A 3 −1 , L A 3 −1 1 −1 2 0
0 2 [ ( )] = −2 [ ] = 1 L A B 2 , A B 3 0 −1
0 0 0 0 ! 2 0 0 1 −1 0 1 = −2 = [ ( )] 0 −1 1 0 3 2 L A B 0 0 0 0 −1 0 26 anna brandenberger, daniela breitman
ker L: compute nullspace([L]B,B):
0 0 0 0 ! 0 1 −1 0 ! 0 1 −1 0 0 1 −1 0 = 0 −1 1 0 ∼ 0 0 0 0 0 −1 1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
with columns a11, a12, a21, a22. a12 − a21 = 0 ⇔ a21 = a12 ⇔ a11, a12, a22 free parameters.
1 0 0 a1 =⇒ = 0 1 0 = a2 ∈ R nullspace span 0 , 1 , 0 a3 : a11, a12, a22 0 0 1 a4 a11 a12 ∈ R which are the coordinate vectors of the matrices a12 a22 : a11, a12, a22 which is exactly the set of all 2 × 2 symmetric matrices. Check: 0 0 t L(A) = 0 0 = 0 = A − A ⇔ A = At ⇔ A is 2 × 2 symmetric.
0 0 ([ ] ) = { 1 −1 } im L : colspace L B,B span −1 , 1 0 0 0 0 ! = span{ 1 } = { a12 : a ∈ R} −1 −a12 12 0 0 These are the coordinate vectors of the matrices: { 0 a12 : a ∈ R} A = −At −a12 0 12 This is the set of all 2 × 2 skew-symmetric matrices. lecture 02/13 We continue the above example: Let L : Mat(2 × 2), 7→ Mat(2 × 2, R) A 7→ A − At
0 0 0 0 ! 1 0 0 1 0 0 0 0 0 1 −1 0 B = { 0 0 , 0 0 , 1 0 , 0 1 } [L]B,B = 0 −1 1 0 0 0 0 0 0 1 0 0 1 0 0 0 1 B = { 0 0 , 1 0 , 0 1 , −1 0 } | {z } | {z } | {z } | {z } 0 0 0 0 0 0 0 2 7→ 7→ 7→ 7→ 0 1 − 0 −1 = 0 2 0 0 0 0 0 0 −2 0 −1 0 1 0 −2 0 0 0 0 0 = 0 1 0 + 0 0 1 0 0 0 0 0 0 1 0 [L] 0 = B ,B 0 0 0 0 + 0 0 + 0 1 0 0 0 2 0 0 1 2 −1 0
This matrix is diagonal, and simpler than the previous representa- tion (See last example). Shows advantage of choosing different bases.
Compositions of Linear maps
0 00 Let U, V, W be finite dimensional VS over K with bases B, B , B .L1 : UB0 7→ VB0 , L2 : VB0 7→ WB00 , L2 ◦ L1 : UB 7→ WB00 . How are these three linear maps related? math 247: honours applied linear algebra 27
Theorem 2.13. Let U, V, W be finite dimensional VS over K with bases 0 00 B, B , B .L1 : UB 7→ VB0 , L2 : VB0 7→ WB00 . Then L2 ◦ L1 : UB 7→ WB00 . If dim U = m, dim V = k, dim W = n, then:
[L2 ◦ L1]B00,B = [L2]B00,B0 · [L1]B0,B | {z } | {z } | {z } n×m n×k k×m
Proof. Let u ∈ U be arbitrary vector, then
[L2 ◦ L1]B00,B[u]B = [(L2 ◦ L1)(u)]B00
= [L2(L1(u))]B00
= [L2]B00,B0 [L1(u)]B0
= [L2]B00,B0 [L1]B0,B[u]B∀u ∈ U.
=⇒ [L2 ◦ L1]B00,B = [L2]B00,B0 · [L1]B0,B
Matrix Representation of Isomorphisms
Theorem 2.14. V, W, finite dimensional VS over K. L : V 7→ W linear, 0 B, B bases for V, W respectively. L isomorphism ⇐⇒ [L]B0,B invertible.
Remark 5. Consider id: V 7→ V the identity map, dim V = n, B basis for V, then [id]B,B = In (n × n identity matrix). If B = (v1, ··· , vn), then [id]B,B = [id(v1)]B | · · · | [id(vn)]B 1 0 ··· 0 0 1 ··· 0 = [ ] | · · · | [ ] = v1 B vn B . . .. In | {z } . . . 0 =1·v1+0·v2+···+0·vn 0 0 0 1
Proof.
("⇒") Let L : VB 7→ WB0 be an isomorphism. =⇒ L is bijective −1 =⇒ L has an inverse L : WB0 7→ VB which is linear. We have −1 −1 L ◦ L : VB 7→ VB, L ◦ L = idV then:
−1 −1 [L ◦ L]B,B = I = [L ]B,B0 [L]B0,B = I =⇒ [L]B0,B invertible. −1 Alternatively, [L]B0,B · [L ]B,B0 = I =⇒ [L]B0,B invertible.
("⇐") Let [L]B0,B be invertible n × n matrix. We first prove that L is injective. Let v ∈ ker L:
=⇒ L(v) = 0 =⇒ [L(v)]B0 = 0 [L]B0,B [v]B = [L(v)]B0 = 0 | {z } invertible =⇒ [v]B = 0 =⇒ v = 0 =⇒ ker L = {0} =⇒ L is injective. 28 anna brandenberger, daniela breitman
Next, we prove L is surjective.
dim V = dim W = n dim ker L + dim imL = dim V = n =⇒ dim imL = n = dim W =⇒ imL = W =⇒ L surjective.
=⇒ L is bijective =⇒ L is an isomorphism.
2.4 Change of basis
0 Q1: V finite dimensional VS over K, B, B bases for V , let v ∈ V. Given [v]B, how can we compute [v]B0 ? Consider id : VB 7→ VB0 . [id]B0,B[v]B = [id(v)]B0 = [v]B0
Definition 2.7. So given id : VB 7→ VB0 , we have [v]B0 = [id]B0,B[v]B. The matrix [id]B0,B is called the transition matrix or change-of-basis matrix from B to B0. caution: Schaum’s mistakenly calls 0 0 Q2: L : V 7→ V linear, B, B bases for V. Given [L]B,B, how can we this the transition matrix from B to B. compute [L]B0,B0 ? id L id Consider: VB0 −→ VB −→ VB −→ VB0 . Then [id ◦ L ◦ id] = [id]B0,B[L]B,B[id]B,B0 = [L]B0,B0 . Let P ≡ [id]B,B0 .
0 id : VB 7→ VB −1 −1 then [id]B0,B = [id]B,B0 = P −1 =⇒ [L]B0,B0 = P [L]B,BP
where P is the change-of-basis matrix from B0 to B.
Definition 2.8. Two n × n matrices A, B with coefficients ∈ K are sim- ilar if there exists an invertible n × n matrix P with B = P−1 AP, thus, any two matrix representations of the same linear map L : V 7→ V with respect to bases B, B0 are similar. Note that converse also holds.
n Special case: V = K with two bases: standard basis S and arbitrary 2 e.g. (added) in R , B = {v~1, v~2}. n t basis B. If we consider the vectors in K as row vectors, [v]S = v , a ~a = c1v~1 + c2v~2 | | ! c column vector. ~ = v~ t v~ t 1 a 1 2 c2 [v] [v] = [id] [v] | | How can we compute B? Invert matrix: B B,S S | {z } [id] is the transition matrix from S to B. P B,S i.e. [~a]S = P[~a]B t t i.e. P = [id] change-of-basis matrix. If B = (v1, ··· , vn) then [id]S,B = (v1 | · · · | vn). And we have S,B −1 −1 t t [id]B,S = ([id]S,B) =⇒ [v]B = P [v]S , where P = (v1 | · · · | vn). |{z} =vt Examples 2.6. lecture 02/15 math 247: honours applied linear algebra 29
1. V = R2, S standard basis, B = ((1, 1), (3, 2))