Let V be a vector space over F with dim(V) = n. Also, let be an ordered basis of V. Then, in the last section of the previous chapter, it was shown that for each x ∈ V, the coordinate vector [x] is a column vector of size n and has entries from F. So, in some sense, each element of V looks like elements of Fn. In this chapter, we concretize this idea. We also show that matrices give rise to functions between two finite dimensional vector spaces. To do so, we start with the definition of functions over vector spaces that commute with the operations of vector addition and scalar multiplication.
Definition 4.1.1. [Linear Transformation, Linear Operator] Let V and W be vector spaces over F. A function (map) T : V → W is called a linear transformation if for all α ∈ F and u,v ∈ V the function T satisfies
Definition 4.1.2. [Equality of Linear Transformation] Let S,T ∈(V, W). Then, S and T are said to be equal if S(x) = T(x), for all x ∈ V. DRAFT
We now give examples of linear transformations.
Remark 4.1.4. Let A ∈ Mn(ℂ) and define TA : ℂn → ℂn by TA(x) = Ax, for every x ∈ ℂn. Then, verify that TAk(x) = k times(x) = Akx, for any positive integer k.
Also, for any two linear transformations S ∈ (V, W) and T ∈ (W, ℤ), we will interchangeably use T ∘ S and TS, for the corresponding linear transformation in (V, ℤ).
We now prove that any linear transformation sends the zero vector to a zero vector.
Proposition 4.1.5. Let T ∈(V, W). Suppose that 0
Proof. Since 0
V) = 0˙W. _
From now on 0 will be used as the zero vector of the domain and codomain. We now consider a few more examples.
The next result states that a linear transformation is known if we know its image on a basis of the domain space.
DRAFT Lemma 4.1.7. Let V and W be two vector spaces over F and let T ∈ (V, W). Then T is known, if the image of T on basis vectors of V are known. In particular, if V is finite dimensional and = (v1,…,vn) is an ordered basis of V over F then, T(v) = [v].
Proof. Let be a basis of V over F. Then, for each v ∈ V, there exist vectors u1,…,uk in and scalars c1,…,ck ∈ F such that v = ∑ i=1kciui. Thus, by definition T(v) = ∑ i=1kciT(ui). Or equivalently, whenever
| (4.1.1) |
Thus, the image of T on v just depends on where the basis vectors are mapped. In particular, if [v] = then, T(v) = [v]. Hence, the required result follows. _
As another application of Lemma 4.1.7, we have the following result. The proof is left for the reader.
Corollary 4.1.8. Let V and W be vector spaces over F and let T : V → W be a linear transformation. If is a basis of V then, Rng(T) = LS(T(x)|x ∈).
Recall that by Example 4.1.3.6, for each a ∈ Fn, the map T(x) = aT x, for each x ∈ Fn, is a linear transformation. We now show that these are the only ones.
DRAFT Corollary 4.1.9. [Reisz Representation Theorem] Let T ∈ (ℝn, ℝ). Then, there exists a ∈ ℝn such that T(x) = aT x.
Proof. By Lemma 4.1.7, T is known if we know the image of T on {e1,…,en}, the standard basis of ℝn. As T is given, for 1 ≤ i ≤ n, T(ei) = ai, for some ai ∈ ℝ. So, consider the vector a = [a1,…,an]T . Then, for x = [x1,…,xn]T ∈ ℝn, we see that
Before proceeding further, we define two spaces related with a linear transformation.
Definition 4.1.10. [Range and Kernel of a Linear Transformation] Let V and W be vector spaces over F and let T : V → W be a linear transformation. Then,
Example 4.1.11. Determine Rng(T) and Ker(T) of the following linear transformations.
For computing, Rng(T), recall that {ij|1 ≤ i,j ≤ 2} is a basis of M2(ℝ). So, for example,
Furthermore, if V is finite dimensional then DRAFT
Example 4.1.13. In each of the examples given below, state whether a linear transformation exists or not. If yes, give at least one linear transformation. If not, then give the condition due to which a linear transformation doesn’t exist.
To construct one such linear transformation, let {(1,1)T ,u} be a basis of ℝ2 and define T(u) = v = (v1,v2)T , for some v ∈ ℝ2. For example, if u = (1,0)T then
From now on, we will just write
The readers are advised to see Exercise 3.3.13.9 and Theorem 3.5.9 for clarity and similarity with the results in this section. We start with the following result.
Theorem 4.2.1. Let V and W be two vector spaces over F and let T ∈(V, W).
Proof. As S is linearly dependent, there exist k ∈ ℕ and vi ∈ S, for 1 ≤ i ≤ k, such that the system ∑ i=1kxivi = 0, in the variable xi’s, has a non-trivial solution, say xi = ai ∈ F,1 ≤ i ≤ k. Thus, ∑ i=1kaivi = 0. Now, consider the system ∑ i=1kyiT(vi) = 0, in the variable yi’s. Then,
The second part is left as an exercise for the reader. _ DRAFT
Definition 4.2.2. [Rank and Nullity] Let V and W be two vector spaces over F. If T ∈ (V, W) and dim(V) is finite then we define Rank(T) = dim(Rng(T)) and Nullity(T) = dim(Ker(T)).
We now prove the rank-nullity Theorem. The proof of this result is similar to the proof of Theorem 3.5.9. We give it again for the sake of completeness.
Theorem 4.2.3 (Rank-Nullity Theorem). Let V and W be two vector spaces over F. If dim(V) is finite and T ∈(V, W) then,
Proof. By Exercise 4.1.12.2.2a, dim(Ker(T)) ≤ dim(V). Let be a basis of Ker(T). We extend it to form a basis of V. As, T(v) = 0, for all v ∈, using Corollary 4.1.8, we get
Let, if possible, the claim be false. Then, there exists v1,…,vk ∈\ and a = [a1,…,ak]T such that a≠0 and ∑ i=1kaiT(vi) = 0. Thus, we see that DRAFT
As an immediate corollary, we have the following result. The proof is left for the reader.
Corollary 4.2.4. Let V and W be vector spaces over F and let T ∈(V, W). If dim(V) = dim(W) then, the following statements are equivalent.
Proof. The prove of Part 2 is omitted as it directly follows from Part 1 and Theorem 4.2.3.
Part 1: We first prove the second inequality. Suppose v ∈ Ker(T). Then
By Theorem 4.2.3, Nullity(S) ≤ Nullity(ST) is equivalent to Rng(ST) ⊆ Rng(S). And this holds as Rng(T) ⊆ V implies Rng(ST) = S(Rng(T)) ⊆ S(V) = Rng(S).
To prove the first inequality, let {v1,…,vk} be a basis of Ker(T). Then {v1,…,vk}⊆ Ker(ST). So, let us extend it to get a basis {v1,…,vk,u1,…,uℓ} of Ker(ST).
Claim: {T(u1),T(u2),…,T(uℓ)} is a linearly independent subset of Ker(S).
Clearly, {T(u1),…,T(uℓ)}⊆ Ker(S). Now, consider the system c1T(u1) + + cℓT(uℓ) = 0 in the variables c1,…,cℓ. As T ∈(V), we get T = 0. Thus, ∑ i=1ℓciui ∈ Ker(T). Hence, ∑ i=1ℓciui is a unique linear combination of v1,…,vk, a basis of Ker(T). Therefore,
| (4.2.1) |
for some scalars α1,…,αk. But by assumption, {v1,…,vk,u1,…,uℓ} is a basis of Ker(ST) and hence linearly independent. Therefore, the only solution of Equation (4.2.1) is given by ci = 0, for 1 ≤ i ≤ ℓ and αj = 0, for 1 ≤ j ≤ k. Thus, we have proved the claim. Hence, Nullity(S) ≥ ℓ and Nullity(ST) = k + ℓ ≤ Nullity(T) + Nullity(S). _
We start with the following definition.
DRAFT Definition 4.2.7. [Sum and Scalar Multiplication of Linear Transformations] Let V, W be vector spaces over F and let S,T ∈(V, W). Then, we define the point-wise
Theorem 4.2.8. Let V and W be vector spaces over F. Then (V, W) is a vector space over F. Furthermore, if dim V = n and dim W = m, then dim(V, W) = mn.
Proof. It can be easily verified that for S,T ∈(V, W), if we define (S + αT)(v) = S(v) + αT(v) (point-wise addition and scalar multiplication) then (V, W) is indeed a vector space over F. We now prove the other part. So, let us assume that = {v1,…,vn} and = {w1,…,wm} are bases of V and W, respectively. For 1 ≤ i ≤ n,1 ≤ j ≤ m, we define the functions fij on the basis vectors of V by
| (4.2.2) |
Thus, fij ∈(V, W).
Claim: {fij|1 ≤ i ≤ n,1 ≤ j ≤ m} is a basis of (V, W).
So, consider the linear system ∑ i=1n ∑ j=1mcijfij = 0, in the variables cij’s, for 1 ≤ i ≤ n,1 ≤ j ≤ m. Using the point-wise addition and scalar multiplication, we get
Now, let us prove that LS = (V, W). So, let f ∈(V, W). Then, for 1 ≤ s ≤ n, f(vs) ∈ W and hence there exists βst’s such that f(vs) = ∑ t=1mβstwt. So, if v = ∑ s=1nαsvs ∈ V then, using Equation (4.2.2), we get
Before proceeding further, recall the following definition about a function.
Definition 4.2.9. [Inverse of a Function] Let f : S → T be any function.
Remark 4.2.10. Let f : S → T be invertible. Then, it can be easily shown that any right inverse and any left inverse are the same. Thus, the inverse function is unique and is denoted by f-1. It is well known that f is invertible if and only if f is both one-one and onto.
Lemma 4.2.11. Let V and W be vector spaces over F and let T ∈(V, W). If T is one-one and onto then, the map T-1 : W → V is also a linear transformation. The map T-1 is called the inverse linear transform of T and is defined by T-1(w) = v, whenever T(v) = w.
Proof. Part 1: As T is one-one and onto, by Theorem 4.2.3, dim(V) = dim(W). So, by Corollary 4.2.4, for each w ∈ W there exists a unique v ∈ V such that T(v) = w. Thus, one defines T-1(w) = v.
We need to show that T-1(α1w1 + α2w2) = α1T-1(w1) + α2T-1(w2), for all α1,α2 ∈ F and w1,w2 ∈ W. Note that by previous paragraph, there exist unique vectors v1,v2 ∈ V such that T-1(w1) = v1 and T-1(w2) = v2. Or equivalently, T(v1) = w1 and T(v2) = w2. So, T(α1v1 + α2v2) = α1w1 + α2w2, for all α1,α2 ∈ F. Hence, for all α1,α2 ∈ F, we get DRAFT
Definition 4.2.13. [Singular, Non-singular Transformations] Let V and W be vector spaces over F and let T ∈(V, W). Then, T is said to be singular if 0 ⊊ Ker(T). That is, Ker(T) contains a non-zero vector. If Ker(T) = {0} then, T is called non-singular.
Example 4.2.14. Let T ∈(ℝ2, ℝ3) be defined by T = . Then, verify that T is non-singular. Is T invertible? DRAFT
We now prove a result that relates non-singularity with linear independence.
Theorem 4.2.15. Let V and W be vector spaces over F and let T ∈(V, W). Then the following statements are equivalent.
Proof. 1⇒2 Let T be singular. Then, there exists v≠0 such that T(v) = 0 = T(0). This implies that T is not one-one, a contradiction.
2⇒3 Let S ⊆ V be linearly independent. Let if possible T(S) be linearly dependent. Then, there exists v1,…,vk ∈ S and α = (α1,…,αk)T ≠0 such that ∑ i=1kαiT(vi) = 0. Thus, T = 0. But T is nonsingular and hence we get ∑ i=1kαivi = 0 with α≠0, a contradiction to S being a linearly independent set.
3⇒1 Suppose that T is not one-one. Then, there exists x,y ∈ V such that x≠y but T(x) = T(y). Thus, we have obtained S = {x - y}, a linearly independent subset of V with T(S) = {0}, a linearly dependent set. A contradiction to our assumption. Thus, the required result follows. _
Definition 4.2.16. [Isomorphism of Vector Spaces] Let V and W be two vector spaces over F and let T ∈(V, W). Then, T is said to be an isomorphism if T is one-one and onto. The vector spaces V and W are said to be isomorphic, denoted VW, if there is an isomorphism from V to W.
We now give a formal proof of the statement in Remark 3.6.9.
Proof. Let {v1,…,vn} be a basis of V and {e1,…,en}, the standard basis of Fn. Now define T(vi) = ei, for 1 ≤ i ≤ n and T = ∑ i=1nαiei, for α1,…,αn ∈ F. Then, it is easy to observe that T ∈(V, Fn), T is one-one and onto. Hence, T is an isomorphism. _
As a direct application using the countability argument, one obtains the following result
Corollary 4.2.18. The vector space ℝ over ℚ is not finite dimensional. Similarly, the vector space ℂ over ℚ is not finite dimensional.
We now summarize the different definitions related with a linear operator on a finite dimensional vector space. The proof basically uses the rank-nullity theorem and they appear in some form in previous results. Hence, we leave the proof for the reader.
Theorem 4.2.19. Let V be a vector space over F with dim V = n. Then the following statements are equivalent for T ∈(V).
Exercise 4.2.20. Let V and W be two vector spaces over F and let T ∈(V, W). If dim(V) is finite then prove that
In Example 4.1.3.8, we saw that for each A ∈ Mm×n(ℂ) there exists a linear transformation T ∈(ℂn, ℂm) given by T(x) = Ax, for each x ∈ ℂn. In this section, we prove that if V and W are vector spaces over F with dimensions n and m, respectively, then any T ∈(V, W) corresponds to a set of m×n matrices. Before proceeding further, the readers should recall the results on ordered basis (see Section 3.6).
So, let = and = be ordered bases of V and W, respectively. Also, let A = [v1,…,vn] and B = [w1,…,wm] be the basis matrix of and , respectively. Then, using Equation (3.6.1), v = A[v] and w = B[w], for all v ∈ V and w ∈ W. Thus, using the above discussion and Equation (4.1.1), we see that for T ∈(V, W) and v ∈ V,
Therefore, [T(v)] = [v] as a vector in W has a unique expansion in terms of basis elements. Note that the matrix , denoted T[,], is an m×n matrix and is unique with respect to the ordered basis as the i-th column equals [T(vi)], for 1 ≤ i ≤ n. So, we immediately have the following definition and result.
Definition 4.3.1. [Matrix of a Linear Transformation] Let = (v1,…,vn) and = (w1,…,wm) be ordered bases of V and W, respectively. If T ∈(V, W) then the matrix T[,] is called the coordinate matrix of T or the matrix of the linear transformation T with respect to the basis and , respectively. When there is no mention of bases, we take it to be the standard ordered bases and denote the corresponding matrix by [T].
Note that if c is the coordinate vector of an element v ∈ V then, T[,]c is the coordinate vector of T(v). That is, the matrix T[,] takes coordinate vector of the domain points to the coordinate vector of its images.
Theorem 4.3.2. Let = (v1,…,vn) and = (w1,…,wm) be ordered bases of V and W, respectively. If T ∈(V, W) then there exists a matrix S ∈ Mm×n(F) with
DRAFT Remark 4.3.3. Let V and W be vector spaces over F with ordered bases 1 = (v1,…,vn) and 1 = (w1,…,wm), respectively. Also, for α ∈ F with α≠0, let 2 = (αv1,…,αvn) and 1 = (αw1,…,αwm) be another set of ordered bases of V and W, respectively. Then, for any T ∈(V, W)
We now give a few examples to understand the above discussion and Theorem 4.3.2.
| (4.3.1) |
T[,] | = = = and | ||
T[,] | = = = |
as = B-1 and = B-1. Also, verify that T[,] = B-1T[,]B.
Example 4.3.5. [Finding T from T[,]]
Solution: Let B be the basis matrix corresponding to the ordered basis . Then, using Equation (3.6.1) and Theorem 4.3.2, we see that
DRAFT
| (4.3.2) |
Let V be a vector space over F with dim(V) = n and ordered basis . Then any T ∈(V) corresponds to a matrix in Mn(F). What happens if the ordered basis needs to change? We answer this in this subsection.
Theorem 4.4.1 (Composition of Linear Transformations). Let V, W and ℤ be finite dimensional vector spaces over F with ordered bases , and , respectively. Also, let T ∈ (V, W) and S ∈(W, ℤ). Then S ∘ T = ST ∈(V, ℤ) (see Figure 4.2). Then
Proof. Let = (u1,…,un), = (v1,…,vm) and = (w1,…,wp) be the ordered bases of V, W and ℤ, respectively. Then using Theorem 4.3.2, we have
As an immediate corollary of Theorem 4.4.1 we have the following result.
DRAFT Theorem 4.4.2 (Inverse of a Linear Transformation). Let V is a vector space with dim(V) = n. If T ∈(V) is invertible then for any ordered basis and of the domain and co-domain, respectively, one has -1 = T-1[,]. That is, the inverse of the coordinate matrix of T is the coordinate matrix of the inverse linear transform.
Proof. As T is invertible, TT-1 = Id. Thus, Example 4.3.4.8a and Theorem 4.4.1 imply
Exercise 4.4.3. Find the matrix of the linear transformations given below.
Let V be a finite dimensional vector space. Then, the next result answers the question “what happens to the matrix T[,] if the ordered basis changes to ?”
Theorem 4.4.4. Let = (u1,…,un) and = (v1,…,vn) be two ordered bases of V and Id the identity operator. Then, for any linear operator T ∈(V)
| (4.4.1) |
Proof. As Id is an identity operator, T[,] as (Id ∘ T ∘ Id)[,] (see Figure 4.3 for clarity). Thus, using Theorem 4.4.1, we get
Let V be a vector space and let T ∈(V). If dim(V) = n then every ordered basis of V gives an n × n matrix T[,]. So, as we change the ordered basis, the coordinate matrix of T changes. Theorem 4.4.4 tells us that all these matrices are related by an invertible matrix. Thus, we are led to the following definitions.
DRAFT Definition 4.4.5. [Change of Basis Matrix] Let V be a vector space with ordered bases and . If T ∈(V) then, T[,] = Id[,] ⋅ T[,] ⋅ Id[,]. The matrix Id[,] is called the change of basis matrix (also, see Theorem 3.6.7) from to .
Definition 4.4.6. [Similar Matrices] Let X,Y ∈ Mn(ℂ). Then, X and Y are said to be similar if there exists a non-singular matrix P such that P-1XP = Y ⇔ XP = PY .
Example 4.4.7. Let = 1 + x,1 + 2x + x2,2 + x and = 1,1 + x,1 + x + x2 be ordered bases of ℝ[x;2]. Then, verify that Id[,]-1 = Id[,], as
Definition 4.5.1. [linear Functional] Let V be a vector space over F. Then a map T ∈(V, F) is called a linear functional on V.
Definition 4.5.3. [Dual Space] Let V be a vector space over F. Then (V, F) is called the dual space of V and is denoted by V*. The double dual space of V, denoted V**, is the dual space of V*.
We first give an immediate corollary of Theorem 4.2.17.
Corollary 4.5.4. Let V and W be vector spaces over F with dim V = n and dim W = m.
Exercise 4.5.5. Let V be a vector space. Suppose there exists v ∈ V such that f(v) = 0, for all f ∈ V*. Then prove that v = 0.
So, we see that V* can be understood through a basis of V. Thus, one can understand V** again via a basis of V*. But, the question arises “can we understand it directly via the vector space V itself?” We answer this in affirmative by giving a canonical isomorphism from V to V**. To do so, for each v ∈ V, we define a map Lv : V*→ F by Lv(f) = f(v), for each f ∈ V*. Then Lv is a linear functional as
We use the above argument to give the required canonical isomorphism.
Theorem 4.5.6. Let V be a vector space over F. If dim(V) = n then the canonical map T : V → V** defined by T(v) = Lv is an isomorphism.
Proof. Note that for each f ∈ V*,
Thus, T gives an inclusion (one-one) map from V to V**. Further, applying Corollary 4.5.4.2 to V*, gives dim(V**) = dim(V*) = n. Hence, the required result follows. _
We now give a few immediate consequences of Theorem 4.5.6.
Corollary 4.5.7. Let V be a vector space of dimension n with basis = {v1,…,vn}.
Proof. Part 1 is direct as T : V → V** was a canonical inclusion map. For Part 2, we need to show that
Let V be a finite dimensional vector space. Then Corollary 4.5.7 implies that the spaces V and V* are naturally dual to each other.
We are now ready to prove the main result of this subsection. To start with, let V and W be vector spaces over F. Then, for each T ∈(V, W), we want to define a map : W*→ V*. So, if g ∈ W* then, (g) a linear functional from V to F. So, we need to be evaluate (g) at an element of V. Thus, we define (v) = g, for all v ∈ V. Now, we note that ∈(W*, V*), as for every DRAFT g,h ∈ W*,
2. Theorem 4.5.8. Let V and W be two vector spaces over F with ordered bases = (v1,…,vn) and = (w1,…,wm), respectively. Also, let * = (f1,…,fn) and * = (g1,…,gm) be the corresponding ordered bases of the dual spaces V* and W*, respectively. Then,
Proof. Note that we need to compute [*,*] = and prove that it equals the transpose of the matrix T[,]. So, let
| (4.5.1) |
Now, recall that the functionals fi’s and gj’s satisfy (vt) = ∑ k=1nαk = αt, for 1 ≤ t ≤ n and = ejT , a row vector with 1 at the j-th place and 0, elsewhere. So, let B = and evaluate (gj) at vt’s, the elements of .
Remark 4.5.9. The proof of Theorem 4.5.8 also shows the following.