where the third and the fourth equalities are the linearity of the integral, Remark 5.13. Hence T is linear.
To prove that T is continuous, we prove that it is bounded. For a linear operator T from a normed space (X,∥⋅∥X) to a normed space (Y,∥⋅∥Y) this means, as in Definition 13.5 (ii), that there exists C>0 such that ∥Tf∥Y≤C∥f∥X for every f∈X. Here we need C>0 such that ∣Tf∣≤C∥f∥∞ for every f∈C([−1,1]). Let f∈C([−1,1]). Then, since ∣f∣≤∥f∥∞ and ∣g∣=1,
Hence T is bounded, and T is continuous by Theorem 13.8. We now compute ∥T∥L, the norm of T in L(C([−1,1]),R).
Remark 14.2. Let (X,∥⋅∥X) and (Y,∥⋅∥Y) be normed spaces and let T∈L(X,Y) with T=0. By Proposition 13.10, ∥T∥L is the least constant C>0 such that ∥Tx∥Y≤C∥x∥X for every x∈X. In general, to compute ∥T∥L: (i) try to estimate ∥Tx∥Y≤C∥x∥X for every x∈X, keeping C as small as possible; (ii)(a) find x0∈X with x0=0 such that ∥Tx0∥Y=C∥x0∥X, or (b) find (xn)⊂X∖{0} such that ∥Txn∥Y/∥xn∥X→C.
For Example 14.1 we choose (b) of Remark 14.2. For every integer n≥1 let fn∈C([−1,1]) be equal to −1 on [−1,−1/n], linear on [−1/n,1/n] and equal to 1 on [1/n,1], as in Figure 14.1, so that ∥fn∥∞=1.
Figure 14.1. The drawing of Example 14.1: the function fn for the n of the slider, equal to −1 on [−1,−1/n], linear on [−1/n,1/n] and equal to 1 on [1/n,1].
Hence, by (1) and (a) of Remark 14.2, ∥T1∥L(L∞,L∞)=log2. For i=2, that is, X2=C([−1,1]), we take the functions fn of Figure 14.1, with ∥fn∥∞=1. Then
Indeed, ∣x∣/(1+x2)≤1 on (−1,1), so the integrals of xfn/(1+x2) and of xH/(1+x2) differ at most by ∥fn−H∥1, which tends to 0. Hence, by (1) and (b) of Remark 14.2, ∥T2∥L(C,C)=log2.
(3) Let g(x)=x2 for x∈(−1,1). Is g∈Ker(T1)? Is T1 injective? We compute
T1g=∫−111+x2xx2dλ=∫−111+x2x3dλ=0,
because the function x3/(1+x2) is odd. Hence g∈Ker(T1).
Remark 14.5. Let T be a linear operator between normed spaces. Then T is injective if and only if Ker(T)={0}.
By Remark 14.5, since g=0, the operator T1 is not injective.
For the last equality, Figure 14.2 draws the graphs of e−x, which decreases from 1, and of 1−e−x, which increases from 0. Hence, by (1) and (a) of Remark 14.2, ∥T∥L=1−e−1.
Figure 14.2. The drawings of (2) in Example 14.6: on the left the graph of e−x, on the right the graph of 1−e−x, for x∈[0,1].
(3) Is T injective? The idea is to impose f∈Ker(T), that is, Tf=0, and to look for necessary conditions on f. For f∈L∞((0,1)) let
F(x):=∫0xeyf(y)dyfor x∈[0,1],
so that (Tf)(x)=e−xF(x) for x∈(0,1). Let f∈Ker(T). Then e−xF(x)=0 for almost every x∈(0,1).
In fact, e−xF(x)=0 for every x∈(0,1). The function y↦eyf(y) belongs to L∞((0,1))⊂L1((0,1)), so F is continuous by the first fundamental theorem of calculus, Theorem 9.9. Hence Tf is continuous, and it vanishes at every point because f∈Ker(T). Indeed, a continuous function that vanishes almost everywhere on (0,1) vanishes at every point of (0,1), since every nonempty open interval has positive measure.
Since e−x>0, it follows that F(x)=0 for almost every x∈(0,1). Moreover, since y↦eyf(y) belongs to L1((0,1)), the function F belongs to AC([0,1]), and F′(x) exists and equals exf(x) for almost every x∈(0,1). In detail, F vanishes at every point of (0,1), because Tf does, so F′=0 on (0,1). Hence 0=exf(x) for almost every x, where ex>0. So f(x)=0 for almost every x, that is, f=0 in L∞((0,1)). Therefore Ker(T)={0}, and T is injective by Remark 14.5.
(4) Is T surjective? The idea is to look for necessary conditions on Tf as f varies in L∞((0,1)).
(i) Let f∈L∞((0,1)). In (Tf)(x)=e−xF(x) the factor e−x is continuous, and F∈C([0,1]) by the first fundamental theorem of calculus, because eyf(y) belongs to L1((0,1)). Hence Tf∈C([0,1]) for every f∈L∞((0,1)). Therefore R(T)⊂C([0,1])⊊L∞((0,1)), and T is not surjective.
(ii) We can observe that R(T)⊂AC([0,1]). Both factors of (Tf)(x)=e−xF(x) belong to AC([0,1]): the function e−x, and the function F of (3).
Remark 14.7. Let a<b be real numbers, let v1,v2∈AC([a,b]) and let v:=v1v2. Then v∈AC([a,b]).
Proof (complete). Let ε>0. By Definition 9.11, applied to v1 and to v2 with ε/(1+∥v1∥∞+∥v2∥∞) in place of ε, for i=1,2 there exists δi>0 such that, for every finite family {(aj,bj):j=1,…,N} of pairwise disjoint subintervals of [a,b],
In the first inequality, v1(bj)v2(aj) is added and subtracted inside each absolute value. The second uses ∣v1(bj)∣≤∥v1∥∞ and ∣v2(aj)∣≤∥v2∥∞, and the third the choice of δ. The last inequality holds because ∥v1∥∞+∥v2∥∞<1+∥v1∥∞+∥v2∥∞. Hence v∈AC([a,b]).■
By Remark 14.7, applied on [0,1] to the functions e−x and F, the function Tf belongs to AC([0,1]) for every f∈L∞((0,1)). That is, R(T)⊂AC([0,1]).
(2) What is ∥T∥L? Let f0∈L1((0,∞)) be a function with f0=0 that vanishes outside [2,3], as in Figure 14.3.
Figure 14.3. The drawing of (2) in Example 14.8, an instance: a function f0 of L1((0,∞)) with f0=0, equal to 0 outside [2,3], with the parts where it is 0 thick on the axis.
(3) Is T injective? Let f∈Ker(T), that is, Tf=0. Then (2+χ[1,2](x))f(x+1)=0 for almost every x>0, where 2+χ[1,2](x)>0. Hence f(x+1)=0 for almost every x>0. That is, f(y)=0 for almost every y>1. Then we take a function f0∈L1((0,∞)) with f0=0 that vanishes on (1,∞), as in Figure 14.4.
Figure 14.4. The drawing of (3) in Example 14.8, an instance: a function f0 of L1((0,∞)) with f0=0, equal to 0 on (1,∞), where it is thick on the axis.
Then (Tf0)(x)=(2+χ[1,2](x))f0(x+1) for almost every x>0. Notice that f0(x+1)=0 for every x>0. Hence (Tf0)(x)=0 for almost every x, that is, Tf0=0 in L1((0,∞)). So f0∈Ker(T) and f0=0, and T is not injective.
(4) Is T surjective? Let u∈R(T), so that there exists f∈L1((0,∞)) with Tf=u. Then u(x)=(2+χ[1,2](x))f(x+1) for almost every x>0, and, since 2+χ[1,2](x)>0,
f(x+1)f(y)=2+χ[1,2](x)u(x)=2+χ[1,2](y−1)u(y−1)for almost every x>0,for almost every y>1,
where the second line is the first with y=x+1. Then, if u∈L1((0,∞)), consider
f(y)=⎩⎨⎧02+χ[1,2](y−1)u(y−1)for almost every y∈(0,1),for almost every y>1.
By Note 14.10, only the indices k≥1 count, and we check that k/(1+k2)≤1/2 for every k≥1. This holds if and only if 2k≤1+k2, that is, 0≤k2−2k+1=(k−1)2, which is true. Hence