The following is from page 31 of Stein and Shakarchi's Real Analysis. My question is about an aspect of the proof of the following theorem.

Theorem 4.1 Suppose $f$ is a non-negative measurable function on $\mathbb R^d$. Then there exists an increasing sequence of non-negative simple functions $\{\varphi_k\}_{k=1}^\infty$ that converges pointwise to $f$, namely, $$ \varphi_k(x) \le \varphi_{k+1}(x)\quad\text{and}\quad\lim_{k\to\infty}\varphi_k(x)=f(x),\ \text{for all $x$.} $$ Proof. We begin first with a truncation. For $k\ge 1$, let $Q_k$ denote the cube centered at the origin and of side length $k$. Then we define $$ F_k(x) = \begin{cases} f(x) & \text{if $x\in Q_k$ and $f(x)\le k$,} \\ k & \text{if $x\in Q_k$ and $f(x)> k$,}\\ 0 & \text{otherwise.} \end{cases} $$ Then $F_k(x)\to f(x)$ as $k$ tends to infinity for all $x$. Now, we partition the range of $F_k$, namely $[0,k]$ as follows. For fixed $k,j\ge 1$, we define $$ E_{\ell,j}=\left\{x\in Q_k:\frac{\ell}{j}<F_k(x)\le\frac{\ell+1}{j}\right\},\quad\text{for}\ 0\le\ell<kj. $$ Then we may form $$ F_{k,j}(x) = \sum_{\ell=0}^{kj-1}\frac{\ell}{j}{\large{\chi_{E_{\ell,j}}}}(x) $$ [where $\large{\chi_{E_{\ell,j}}}$ is the indicator function of $E_{\ell,j}$].

Each $F_{k,j}$ is a simple function that satisfies $0\le F_k(x)-F_{k,j}(x)\le 1/j$ for all $x$. If we now choose $j=k$, and let $\varphi_k = F_{k,k}$, then we see that $0\le F_k(x)-\varphi_k(x)\le 1/k$ for all $x$, $\color{red}{\underline{\color{black}{\text{and $\{\varphi_k\}$ satisfies all the desired properties.}}}}$

I do not see why $\varphi_k(x)\le\varphi_{k+1}(x)$ for all $x$. Can someone explain that?


Unless I have misunderstood something, I suspect there is a slight error in the proof. Consider $f(x)=x$ and consider the approximation of the point $x_0=.6$. Then for $k=2$ we have $x_0\in E_{1,2}$ and so $F_{2,2}(.6) = \frac{1}{2}$. However, for $k=3$ we have $x_0 \in E_{1,3}$ so that $F_{3,3}(.6)=\frac{1}{3}$. If we consider the sequence

$$\varphi_k(x) = F_{2^k,2^k}(x)$$

then I believe we have the desired result. Indeed, in this case we use the construction of $F_{k}$ to deduce that $F_{k-1}\leq F_{k}$ and the fact that the "vertical mesh" is finer for larger $j$ to deduce that $F_{2^k,2^{k-1}} \leq F_{2^k,2^k}.$ With these observations together we conclude

$$F_{2^{k-1},2^{k-1}}\leq F_{2^{k},2^{k-1}}\leq F_{2^k,2^k}.$$

Edit: to prove $F_{2^{k},2^{k-1}}\leq F_{2^k,2^k}$, fix $x$ and observe that if $x\in E_{\ell,2^{k-1}}$ then either $x\in E_{2\ell,2^{k}}$ or $x\in E_{2\ell+1,2^k}.$ We conclude

$$F_{2^{k},2^{k-1}}(x) = \frac{\ell}{2^{k-1}} \leq \min\{\frac{2\ell}{2^{k}},\frac{2\ell+1}{2^{k}}\}\leq F_{2^k,2^k}(x).$$

  • $\begingroup$ I think you're right! Would you mind proving that first inequality $\large F_{2^k,2^{k-1}}\le F_{2^k,2^{k}}$? I feel like it should be true, but I am having a hard time putting it together formally. $\endgroup$ – Alex Ortiz Apr 18 '17 at 1:52
  • $\begingroup$ @AOrtiz I have updated my answer to prove this inequality. I have some pictures in my head about how these inequalities should work, but I am subject to human error; be sure to check me! $\endgroup$ – Matt Apr 18 '17 at 2:09
  • $\begingroup$ Yes! I just checked it. Your strategy dyadically partitions up the range, and the characteristic functions just keep getting larger because with each additional power of $2$, you refine the same subset of the range. Thanks very much. $\endgroup$ – Alex Ortiz Apr 18 '17 at 2:27
  • $\begingroup$ You're welcome! It was a good question... I also learned from Stein and Shakarchi, and I am sure I did not notice this error when learning this result. $\endgroup$ – Matt Apr 18 '17 at 2:29

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.