Section 10.4

The Generalized Minkowski Inequality

[5/40%]
Show
LaTeX
Problem 10.4.1

Generalized Minkowski Inequality: Let (Ω1,F1,μ1)\left(\Omega_{1}, \mathcal{F}_{1}, \mu_{1}\right) and (Ω2,F2,μ2)\left(\Omega_{2}, \mathcal{F}_{2}, \mu_{2}\right) denote σ\sigma-finite measure spaces, let 1≤p<+∞1 \leq p < +\infty, and let f:Ω1×Ω2→R‾f: \Omega_{1} \times \Omega_{2} \rightarrow \overline{\mathbb {R}} be F1⊗F2/B∗\mathcal{F}_{1} \otimes \mathcal{F}_{2} / \mathcal{B}^{*}-measurable and such that

∫Ω2[∫Ω1∣f(ω1,ω2)∣p dμ1(ω1)]1/pdμ2(ω2)<+∞. \int _{\Omega _{2}} \left[\int _{\Omega _{1}} \left|f\left(\omega _{1}, \omega _{2}\right)\right|^{p} \, d\mu _{1}\left(\omega _{1}\right)\right]^{1 / p} d\mu _{2}\left(\omega _{2}\right) < +\infty .

Then

[∫Ω1∣∫Ω2f(ω1,ω2) dμ2(ω2)∣pdμ1(ω1)]1/p≤∫Ω2[∫Ω1∣f(ω1,ω2)∣p dμ1(ω1)]1/pdμ2(ω2). \left[\int _{\Omega _{1}} \left|\int _{\Omega _{2}} f\left(\omega _{1}, \omega _{2}\right) \, d\mu _{2}\left(\omega _{2}\right)\right|^{p} d\mu _{1}\left(\omega _{1}\right)\right]^{1 / p} \leq \int _{\Omega _{2}} \left[\int _{\Omega _{1}} \left|f\left(\omega _{1}, \omega _{2}\right)\right|^{p} \, d\mu _{1}\left(\omega _{1}\right)\right]^{1 / p} d\mu _{2}\left(\omega _{2}\right) .

Consider the σ\sigma-finite measure space (Ω,F,μ)\left(\Omega , \mathcal{F}, \mu \right) and consider the finite measure space (Ω2,F2,μ2)\left(\Omega_{2}, \mathcal{F}_{2}, \mu_{2}\right), where Ω2={a,b}\Omega_{2} = \left\{ a, b\right\}, F2=2Ω2\mathcal{F}_{2} = 2^{\Omega_{2}}, and μ2\mu_{2} denotes counting measure. Let f,g∈Lp(Ω,F,μ)f, g \in L^{p}\left(\Omega , \mathcal{F}, \mu \right), where 1≤p<+∞1 \leq p < +\infty. Define h:Ω×Ω2→R‾h: \Omega \times \Omega_{2} \rightarrow \overline{\mathbb {R}} by writing h(ω,a)=f(ω)h\left(\omega , a\right) = f(\omega ) and h(ω,b)=g(ω)h\left(\omega , b\right) = g(\omega ) for each ω∈Ω\omega \in \Omega. Here, the Generalized Minkowski Inequality becomes as follows: Let f,g∈Lp(Ω,F,μ)f, g \in L^{p}\left(\Omega , \mathcal{F}, \mu \right). Then f+g∈Lp(Ω,F,μ)f + g \in L^{p}\left(\Omega , \mathcal{F}, \mu \right) and ∥f+g∥p≤∥f∥p+∥g∥p\left\| f + g\right\|_{p} \leq \left\| f\right\|_{p} + \left\| g\right\|_{p}.

?
Problem 10.4.2

Let 0<q≤p<+∞0 < q \leq p < +\infty. Without worrying about measurability conditions, obtain the inequality

[∫Ω1(∫Ω2∣f(ω1,ω2)∣q dμ2(ω2))p/qdμ1(ω1)]1/p≤[∫Ω2(∫Ω1∣f(ω1,ω2)∣p dμ1(ω1))q/pdμ2(ω2)]1/q. \left[\int _{\Omega _{1}} \left(\int _{\Omega _{2}} \left|f\left(\omega _{1}, \omega _{2}\right)\right|^{q} \, d\mu _{2}\left(\omega _{2}\right)\right)^{p / q} d\mu _{1}\left(\omega _{1}\right)\right]^{1 / p} \leq \left[\int _{\Omega _{2}} \left(\int _{\Omega _{1}} \left|f\left(\omega _{1}, \omega _{2}\right)\right|^{p} \, d\mu _{1}\left(\omega _{1}\right)\right)^{q / p} d\mu _{2}\left(\omega _{2}\right)\right]^{1 / q} .
?
Remark.
?

We have 1≤p/q<+∞1 \leq p / q < +\infty. Play with exponents and use the Generalized Minkowski Inequality.

Problem 10.4.3

Let L\mathcal{L} denote the Lebesgue subsets of (0,+∞)(0, +\infty ), and let λˉ\bar{\lambda } denote Lebesgue measure on the Lebesgue subsets of (0,+∞)(0, +\infty ). Consider the σ\sigma-finite measure space ((0,+∞),L,λˉ)\left((0, +\infty ), \mathcal{L}, \bar{\lambda }\right). Given a function f∈L1((0,+∞),L,λˉ)f \in L^{1}\left((0, +\infty ), \mathcal{L}, \bar{\lambda }\right), associate the function F:(0,+∞)→[0,+∞)F: (0, +\infty ) \rightarrow [0, +\infty ) given by F(x)=1x∫0xf(y) dλˉ(y)F(x) = \frac{1}{x} \int_{0}^{x} f(y) \, d\bar{\lambda }(y).

?
(a)

For 1<p<+∞1 < p < +\infty we have F∈Lp((0,+∞),L,λˉ)F \in L^{p}\left((0, +\infty ), \mathcal{L}, \bar{\lambda }\right) and ∥F∥p≤p′∥f∥p\left\| F\right\|_{p} \leq p^{\prime } \left\| f\right\|_{p}.

(b)

The constant p′=pp−1p^{\prime } = \frac{p}{p - 1} is the best [smallest] possible constant for the inequality in (a).

(c)

We have excluded the p=1p = 1 case so far. Show that if f∈L1((0,+∞),L,λˉ)f \in L^{1}\left((0, +\infty ), \mathcal{L}, \bar{\lambda }\right) is positive, then F∉L1((0,+∞),L,λˉ)F \notin L^{1}\left((0, +\infty ), \mathcal{L}, \bar{\lambda }\right).

Remark.
?

(a) Replace yy by xtxt, and then use the main result of this section.

(b) Consider f(x)=xϵ−1p  1(0,1)(x)f(x) = x^{\epsilon - \frac{1}{p}} \; \mathbb {1}_{(0, 1)}(x) for each x∈(0,+∞)x \in (0, +\infty ).

Problem 10.4.4

This exercise is a lemma for the next exercise. Let (Ω,F,μ)\left(\Omega , \mathcal{F}, \mu \right) denote a measure space with μ(Ω)<+∞\mu (\Omega ) < +\infty. Every F/B∗\mathcal{F} / \mathcal{B}^{*}-measurable function f:Ω→[0,+∞]f: \Omega \rightarrow [0, +\infty ] satisfies

lim⁡p→∞[∫Ωf(ω)p dμ(ω)]1/p=ess⁡sup⁡ω∈Ωf(ω). \lim _{p \rightarrow \infty } \left[\int _{\Omega } f(\omega )^{p} \, d\mu (\omega )\right]^{1 / p} = \operatorname {ess} \sup _{\omega \in \Omega } f(\omega ) .
?
Remark.
?

Suppose ff is not essentially bounded. Then μ({ω∈Ω:f(ω)≥M})>0\mu \left(\left\{ \omega \in \Omega : f(\omega ) \geq M\right\} \right) > 0 has positive μ\mu-measure for each M>0M > 0. Show for each M>0M > 0 and p>1p > 1 that

0<Mμ({ω∈Ω:f(ω)≥M})1/p≤[∫Ω∣f(ω)∣p dμ(ω)]1/p 0 < M \mu \left(\left\{ \omega \in \Omega : f(\omega ) \geq M\right\} \right)^{1 / p} \leq \left[\int _{\Omega } \left|f(\omega )\right|^{p} \, d\mu (\omega )\right]^{1 / p}

by splitting the integral into the integral over the set of ω\omega where f(ω)≥Mf(\omega ) \geq M and the set of ω\omega where f(ω)<Mf(\omega ) < M. Next, lim⁡p→∞μ({ω∈Ω:f(ω)≥M})1/p=1\lim_{p \rightarrow \infty } \mu \left(\left\{ \omega \in \Omega : f(\omega ) \geq M\right\} \right)^{1 / p} = 1 since μ(Ω)<+∞\mu (\Omega ) < +\infty. Therefore, given any large MM, we may choose pp so large that (∫Ω∣f(ω)∣p dμ(ω))1/p\left(\int_{\Omega } \left|f(\omega )\right|^{p} \, d\mu (\omega )\right)^{1 / p} exceeds, say, M/2M / 2. This gives the result when ff is not essentially bounded. If on the other hand ff is essentially bounded, then f∈Lp(Ω,F,μ)f \in L^{p}\left(\Omega , \mathcal{F}, \mu \right) for every p≥1p \geq 1 by the fact that μ(Ω)<+∞\mu (\Omega ) < +\infty. We have shown earlier that ∥f∥p→∥f∥∞\left\| f\right\|_{p} \rightarrow \left\| f\right\|_{\infty } in this case; see Exercise 4 in Section 8.5.

Problem 10.4.5

We proved the Generalized Minkowski Inequality for 1≤p<+∞1 \leq p < +\infty. There is a version for the case p=+∞p = +\infty. Let (Ω1,F1,μ1)\left(\Omega_{1}, \mathcal{F}_{1}, \mu_{1}\right) and (Ω2,F2,μ2)\left(\Omega_{2}, \mathcal{F}_{2}, \mu_{2}\right) denote σ\sigma-finite measure spaces, and let f:Ω1×Ω2→[0,+∞]f: \Omega_{1} \times \Omega_{2} \rightarrow [0, +\infty ] be F1⊗F2/B∗\mathcal{F}_{1} \otimes \mathcal{F}_{2} / \mathcal{B}^{*}-measurable.

?
(a)

The mapping ω2↦ess⁡sup⁡ω1∈Ω1f(ω1,ω2)\omega_{2} \mapsto \operatorname {ess} \sup_{\omega_{1} \in \Omega_{1}} f\left(\omega_{1}, \omega_{2}\right) is F2/B∗\mathcal{F}_{2} / \mathcal{B}^{*}-measurable, hence we may meaningfully discuss ∫Ω2ess⁡sup⁡ω1∈Ω1f(ω1,ω2) dμ2(ω2)\int_{\Omega_{2}} \operatorname {ess} \sup_{\omega_{1} \in \Omega_{1}} f\left(\omega_{1}, \omega_{2}\right) \, d\mu_{2}\left(\omega_{2}\right).

From now on, assume that ∫Ω2ess⁡sup⁡ω1∈Ω1f(ω1,ω2) dμ2(ω2)<+∞\int_{\Omega_{2}} \operatorname {ess} \sup_{\omega_{1} \in \Omega_{1}} f\left(\omega_{1}, \omega_{2}\right) \, d\mu_{2}\left(\omega_{2}\right) < +\infty.

(b)

For μ1\mu_{1}-a.e. ω1∈Ω1\omega_{1} \in \Omega_{1} the mapping ω2↦f(ω1,ω2)\omega_{2} \mapsto f\left(\omega_{1}, \omega_{2}\right) is in L1(Ω2,F2,μ2)L^{1}\left(\Omega_{2}, \mathcal{F}_{2}, \mu_{2}\right). That is, for μ1\mu_{1}-a.e. ω1∈Ω1\omega_{1} \in \Omega_{1} we have ∫Ω2f(ω1,ω2) dμ2(ω2)<+∞\int_{\Omega_{2}} f\left(\omega_{1}, \omega_{2}\right) \, d\mu_{2}\left(\omega_{2}\right) < +\infty.

(c)

The [nonnegative] mapping ω1↦∫Ω2f(ω1,ω2) dμ2(ω2)\omega_{1} \mapsto \int_{\Omega_{2}} f\left(\omega_{1}, \omega_{2}\right) \, d\mu_{2}\left(\omega_{2}\right) defined on Ω1\Omega_{1} is F1/B∗\mathcal{F}_{1} / \mathcal{B}^{*}-measurable, hence we may meaningfully discuss ess⁡sup⁡ω1∈Ω1∫Ω2f(ω1,ω2) dμ2(ω2)\operatorname {ess} \sup_{\omega_{1} \in \Omega_{1}} \int_{\Omega_{2}} f\left(\omega_{1}, \omega_{2}\right) \, d\mu_{2}\left(\omega_{2}\right).

(d)

We have the inequality

ess⁡sup⁡ω1∈Ω1∫Ω2f(ω1,ω2) dμ2(ω2)≤∫Ω2ess⁡sup⁡ω1∈Ω1f(ω1,ω2) dμ2(ω2). \operatorname {ess} \sup _{\omega _{1} \in \Omega _{1}} \int _{\Omega _{2}} f\left(\omega _{1}, \omega _{2}\right) \, d\mu _{2}\left(\omega _{2}\right) \leq \int _{\Omega _{2}} \operatorname {ess} \sup _{\omega _{1} \in \Omega _{1}} f\left(\omega _{1}, \omega _{2}\right) \, d\mu _{2}\left(\omega _{2}\right) .