2

Suppose that $g:[a,b]\to\mathbb R$ is Lebesgue-integrable and $f:\mathbb R\to\mathbb R$ is continuous. Furthermore, $|f(x)|<a+b|x|$ for some constants $a$ and $b$. It follows that $f\circ g$ Lebesgue-integrable.

Conversely, suppose that $f$ is continuous; for all integrable function $g$ on $[c,d]$, $f\circ g$ is integrable on $[c,d]$. We claim that, there exist constants $a$ and $b$, such that, $|f(x)| < a+ b|x|$.

(This is exercise 71 from Royden 4th edition page no. 134.)

How should we prove the claim?

Sasha Queequeg
  • 357
  • 1
  • 10
Sushil
  • 2,881
  • I handle the case of a set with infinite measure (the whole real line) here: http://math.stackexchange.com/questions/913687/composition-of-lp-functions. It should not be too hard to adapt the argument to the current setting. – PhoemueX Nov 24 '15 at 17:33
  • sorry but argument will be adapted if yn belong to [0.1] but this is not compulsory at all. How argument can be adapted. Can you please explain? @PhoemueX – Sushil Nov 24 '15 at 21:59
  • @Sushil: I will elaborate this evening. Right now, I don't have the time :( – PhoemueX Nov 25 '15 at 11:54
  • 2
    @PhoemueX sorry to disturb. but I think you forget to elaborate. I am not able to see how your proof can be modified in given setting – Sushil Nov 30 '15 at 22:17
  • Are you sure this question is from Royden, 4th edition, page 134? Can't seem to find it there. – Olorun Dec 02 '15 at 06:14
  • @Olorun ROyden Real Analysis. Yes it is on page 134 exercise 71 – Sushil Dec 02 '15 at 12:07
  • @Sushil: Yes, I am very sorry. I now have the modified argument (hand)written. I will type and post it today! – PhoemueX Dec 02 '15 at 12:14
  • Thanks @PhoemueX I am actually stuck at this from last few days. – Sushil Dec 02 '15 at 12:28
  • @Sushil: I just posted the proof. Note that instead of an interval, I consider an arbitrary set $\Omega \subset \Bbb {R}^n $. For the case of an interval, the proof simplifies a bit, in particular the construction of the sets $\Omega_n $ is straightforward. – PhoemueX Dec 02 '15 at 13:46

1 Answers1

9

Since the Lebesgue measure $\lambda$ on $\mathbb{R}^{d}$ is atom-less, it has the following property: For every measurable set $A$ and any $0<\alpha<\lambda\left(A\right)$, there is a measurable subset $B\subset A$ with $\lambda\left(B\right)=\alpha$. This can be proven for example by considering the function $$ f:\left[0,\infty\right)\to\left[0,\infty\right),t\mapsto\lambda\left(A\cap\left[-t,t\right]^{d}\right). $$ Note that $f$ is continuous since for $0\leq t\leq s$, we have \begin{eqnarray*} \left|f\left(s\right)-f\left(t\right)\right| & = & f\left(s\right)-f\left(t\right)\\ & = & \lambda\left(\left[A\cap\left[-s,s\right]^{d}\right]\setminus\left[A\cap\left[-t,t\right]^{d}\right]\right)\\ & \leq & \lambda\left(\left[-s,s\right]^{d}\setminus\left[-t,t\right]^{d}\right)\\ & = & \left(2s\right)^{d}-\left(2t\right)^{d}. \end{eqnarray*} Now apply the intermediate value theorem.

Let us set $\alpha:=\lambda\left(\Omega\right)\in\left(0,\infty\right)$. By applying the above property, we see that there are pairwise disjoint(!) subsets $\left(\Omega_{n}\right)_{n\in\mathbb{N}}$ of $\Omega$ satisfying $\lambda\left(\Omega_{n}\right)=\frac{\alpha}{2^{n}}$ for all $n\in\mathbb{N}$.

Now, I assume that $\varphi:\mathbb{R}\to\mathbb{R}$ has the property that the map $$ \Phi_{\varphi}:L^{p}\left(\Omega\right)\to L^{p}\left(\Omega\right),f\mapsto\varphi\circ f $$ is well-defined. To prove your claim, it suffices to show two separate claims, namely:

  1. $\forall\beta>0\exists C_{\beta}>0\forall\left|x\right|\leq\beta:\:\left|\varphi\left(x\right)\right|\leq C_{\beta}$.

  2. $\exists\beta>0\,\exists K>0\,\forall\left|x\right|>\beta:\:\left|\varphi\left(x\right)\right|\leq K\cdot\left|x\right|$.

In fact, if we combine both claims, we get for $\left|x\right|\leq\beta$ that $\left|\varphi\left(x\right)\right|\leq C_{\beta}\leq K\left|x\right|+C_{\beta}$ and for $\left|x\right|>\beta$ likewise $\left|\varphi\left(x\right)\right|\leq K\left|x\right|\leq K\left|x\right|+C_{\beta}$.

Now, assume that the first property is false. This yields some $\beta>0$ such that the following holds: $$ \forall n\in\mathbb{N}\exists x_{n}\in\mathbb{R}\text{ with }\left|x_{n}\right|\leq\beta\text{ and }\left|\varphi\left(x_{n}\right)\right|\geq2^{n/p}. $$ Now, define $f:=\sum_{n=1}^{\infty}x_{n}\cdot1_{\Omega_{n}}$. Since the $\left(\Omega_{n}\right)_{n\in\mathbb{N}}$ are pairwise disjoint(!), we have $$ \left\Vert f\right\Vert _{L^{p}}^{p}=\sum_{n=1}^{\infty}\left|x_{n}\right|^{p}\lambda\left(\Omega_{n}\right)\leq\beta^{p}\alpha\cdot\sum_{n=1}^{\infty}2^{-n}=\beta^{p}\alpha<\infty $$ and hence $f\in L^{p}\left(\Omega\right)$. But we also have (using again the disjointness) $\varphi\circ f=\sum_{n=1}^{\infty}\varphi\left(x_{n}\right)1_{\Omega_{n}}$ and thus $$ \infty>\left\Vert \Phi_{\varphi}\left(f\right)\right\Vert _{L^{p}}^{p}=\left\Vert \varphi\circ f\right\Vert _{L^{p}}^{p}=\sum_{n=1}^{\infty}\left|\varphi\left(x_{n}\right)\right|^{p}\lambda\left(\Omega_{n}\right)\geq\sum_{n=1}^{\infty}2^{n}\frac{\alpha}{2^{n}}=\sum_{n=1}^{\infty}\alpha=\infty, $$ a contradiction.

Hence, we have established the first property.

Now, we will show that the second property also holds, with $\beta=2^{1/p}$. Assume that this is false. Hence, taking $K=n^{2/p}$ for arbitrary $n\in\mathbb{N}$, we get $$ \forall n\in\mathbb{N}\exists x_{n}\in\mathbb{R}\text{ with }\left|x_{n}\right|>2^{1/p}\text{ and }\left|\varphi\left(x_{n}\right)\right|\geq n^{2/p}\left|x_{n}\right|\geq2^{1/p}n^{2/p}. $$ In particular, we get $\left|\varphi\left(x_{n}\right)\right|\xrightarrow[n\to\infty]{}\infty$. In view of property 1 from above, this easily entails $\left|x_{n}\right|\xrightarrow[n\to\infty]{}\infty$, since property 1 means that $\varphi$ is bounded on bounded sets.

Now, define $$ N_{n}:=\left\lceil \log_{2}\left(\left|x_{n}\right|^{p}\cdot n^{2}\right)\right\rceil \xrightarrow[n\to\infty]{}\infty. $$ Since the sequence diverges to $\infty$, we can extract a strictly increasing subsequence $\left(N_{n_{\ell}}\right)_{\ell\in\mathbb{N}}$. Because this sequence is strictly increasing, the sequence $\left(\Omega_{N_{n_{\ell}}}\right)_{\ell\in\mathbb{N}}$ is pairwise disjoint. Now, define $$ f:=\sum_{\ell=1}^{\infty}x_{n_{\ell}}1_{\Omega_{N_{n_{\ell}}}} $$ and note as above \begin{eqnarray*} \left\Vert f\right\Vert _{L^{p}}^{p} & = & \sum_{\ell=1}^{\infty}\left|x_{n_{\ell}}\right|^{p}\lambda\left(\Omega_{N_{n_{\ell}}}\right)\\ & = & \sum_{\ell=1}^{\infty}\left|x_{n_{\ell}}\right|^{p}\cdot\frac{\alpha}{2^{N_{n_{\ell}}}}\\ \left(\text{since }N_{n}\geq\log_{2}\left(\left|x_{n}\right|^{p}\cdot n^{2}\right)\right) & \leq & \sum_{\ell=1}^{\infty}\left|x_{n_{\ell}}\right|^{p}\cdot\frac{\alpha}{2^{\log_{2}\left(\left|x_{n_{\ell}}\right|^{p}\cdot n_{\ell}^{2}\right)}}\\ & = & \alpha\cdot\sum_{\ell=1}^{\infty}\frac{1}{n_{\ell}^{2}}\\ & \leq & \alpha\sum_{n=1}^{\infty}\frac{1}{n^{2}}<\infty. \end{eqnarray*} Now, note that $$ N_{n}\leq\log_{2}\left(\left|x_{n}\right|^{p}\cdot n^{2}\right)+1 $$ and thus $$ 2^{N_{n}}\leq2\cdot\left|x_{n}\right|^{p}\cdot n^{2}. $$ Thus, we get \begin{eqnarray*} \left\Vert \varphi\circ f\right\Vert _{L^{p}}^{p} & = & \sum_{\ell=1}^{\infty}\left[\left|\varphi\left(x_{n_{\ell}}\right)\right|^{p}\cdot\lambda\left(\Omega_{N_{n_{\ell}}}\right)\right]\\ & \geq & \sum_{\ell=1}^{\infty}\left[n_{\ell}^{2/p}\left|x_{n_{\ell}}\right|\right]^{p}\frac{\alpha}{2^{N_{n_{\ell}}}}\\ & \geq & \sum_{\ell=1}^{\infty}\left[n_{\ell}^{2/p}\left|x_{n_{\ell}}\right|\right]^{p}\frac{\alpha}{2\cdot\left|x_{n_{\ell}}\right|^{p}\cdot n_{\ell}^{2}}\\ & = & \frac{\alpha}{2}\cdot\sum_{\ell=1}^{\infty}1=\infty, \end{eqnarray*} in contradiction to $\Phi_{\varphi}\left(f\right)=\varphi\circ f\in L^{p}\left(\Omega\right)$. Thus, we have also established property 2 from above.

As explained above, this completes the proof.

PhoemueX
  • 36,211