3

Problem

Let $[a,b]$ be a closed and bounded interval, and let $C[a,b]$ be the vector space of all continuous real-valued functions on $[a,b]$. Define the function $\|\cdot\|_1:C[a,b]\to\mathbb{R}$ by letting $$ \|f\|_1=\int_a^b|f|d\lambda. $$ Prove that if a function $f\in C[a,b]$ satisfies $\|f\|_1=0$, then $f=0$.

(Note that $\lambda$ is the Lebesgue measure.)

My Attempt

Let $f\in C[a,b]$. If $f=0$, then $\|f\|_1 = \int_1^b0d\lambda = 0$. Now suppose that $$\|f\|_1=\int_a^b|f|d\lambda= \sup\left\{\int hd\lambda:h\in\mathscr{S}_+\ \text{and}\ h\leq|f|\chi_{[a,b]}\right\} = 0.$$ Assume to the contrary that $f(x)\neq0$ for some $x\in[a,b]$. Then $|f(x)|>0$. Since $f$ is continuous, $|f|$ must be continuous and there must exists a closed and bounded interval $[c,d]\subseteq[a,b]$ such that $|f(z)|>0$ for all $z\in[c,d]$. Define a function $h$ by letting $h(z)=0$ for all $z\notin[c,d]$ and $h(z)=\min_{z\in[c,d]}|f(z)|$ (note that the minimum exists because $|f|$ is a continuous function on a compact set $[c,d]$). Then $h\in\mathscr{S}_+$ and $h\leq|f|\chi_{[a,b]}$, but $\int hd\lambda>0$, which is a contradiction. Thus, $f=0$.

Some definitions:

Definition$\quad$ Let $(X,\mathscr{A})$ be a measurable space. $\mathscr{S}_+$ is the set of simple nonnegative real-valued $\mathscr{A}$-measurable functions on $X$.

Definition$\quad$ $\chi_A$ is the characteristic function of the set $A$.

My Question

I am not sure if my attempt is correct. Could someone please help me check? Especially, I claimed that "there must exists a closed and bounded interval $[c,d]\subseteq[a,b]$ such that $|f(z)|>0$ for all $z\in[c,d]$" without proof. (In fact, I am not sure how to prove this rigorously, but I seriously believe this is correct.)

Thank you very much!

Beerus
  • 2,939
  • 1
    Your proof is perfect. For the existence of $[c,d]$, simply argue for instance, letting $m:=|f(x)|$: by continuity of $|f|$ at $x$, $|f|\ge m/2$ on some neighborhood of $x$. I won't edit this comment as an answer because I am sure this is a megaduplicate. – Anne Bauval May 08 '24 at 18:10
  • 1
    Does this answer your question? Why must the integral of a nonzero continuous function be nonzero? (I know it does. This is just an automatically generated message, as a consequence of my vote for closing as a duplicate.) – Anne Bauval May 08 '24 at 18:31
  • @AnneBauval The post you suggested is not quite duplicate with mine. First, that post is Riemann integral, but mine is Lebesgue integral. For a bounded real-valued function on a closed and bounded interval, it is only true that Riemann integrability implies Lebesgue integrability, but not the reverse. Second, my post is also about proof writing, of which the center part is about arguing the existence of an interval $[c,d]$ as a subset of $[a,b]$ such that $|f(z)|>0$ for all $z\in[c,d]$. – Beerus May 08 '24 at 19:26
  • The answer to that post solved precisely your question (the existence of such an interval), independently of the context (Riemann/Lebesgue), by the same argument as in my first comment above. + For the sake of future readers, please don't undo again the clarification of title and body of your post. – Anne Bauval May 08 '24 at 21:13
  • You did (undo again the clarification of title and body): what a pity! + insisting on Lebesgue integral (instead of just Riemann) seems quite pedantic when dealing with continuous functions + if you nevertheless do, restore the correct notation $\int_{[a,b]}$, not $\int_a^b$ + it seems contradictory to comment "my post is also about proof writing" and to remove your corresponding tag. – Anne Bauval May 09 '24 at 07:56

2 Answers2

0

Just a hint to get you in the right direction, but you can make the argument a bit more rigorous and fill in the fact about the interval $[c,d]$ by appealing to the $(\epsilon,\delta)$ definition of continuity.

For example, if $f(z) = \alpha > 0$ then since $f$ is a continuous function we can set $\epsilon = \alpha/2 > 0$ and there must be a $\delta > 0$ such that for every $z' \in [z-\delta,z+\delta] \cap [a,b]$ it holds $$|f(z) - f(z')| < \epsilon = \alpha/2$$ In particular this implies for every $z$ in the interval $$f(z') > f(z) - \alpha/2 = \alpha - \alpha/2 = \alpha/2 > 0$$ This gives you both an interval $[z-\delta, z+\delta] \cap [a,b]$ and a strictly positive lower bound $\alpha/2$ of $f$ on that interval.

There are still a few details that I'll leave to you to fill in to handle the absolute value.

0

It seems that we may argue in this way: Let $f\in C[a,b]$ be such that $\int_{a}^{b}|f|d\lambda=0$, then $f=0$ a.e. (For, let $A=\{x\in[a,b]\mid|f(x)|>0\}$, and $A_{n}=\{x\in[a,b]\mid|f(x)|>\frac{1}{n}\}$. Note that $A=\cup_{n}A_{n}.$ If $\lambda(A_{n})>0$ for some $n$, then $\int_{a}^{b}|f|d\lambda\geq\frac{1}{n}\lambda(A_{n})>0$, which is a contradiction. Therefore, $\lambda(A_{n})=0$ for all $n$ and hence $\lambda(A)=0$.)

Let $A=\{x\in[a,b]\mid|f(x)|>0\}$. Since $\lambda(A)=0,$ $A$ does not contain any interior point (with respect to the relative topology of $[a,b]$). In another word, for each $x\in A$, there exists a sequence $(x_{n})\in[a,b]\setminus A$ such that $x_{n}\rightarrow x$. We argue that $A=\emptyset$. Prove by contradiction. Suppose that there exists $x\in A$. Choose a sequence $(x_{n})\in[a,b]\setminus A$ such that $x_{n}\rightarrow x$. Note that $f(x_{n})=0$, so $f(x)=\lim_{n\rightarrow\infty}f(x_{n})=0$, a contradiction.

  • Thanks for your help! I did not down-vote your answer, and I don't know why it is down-voted. But I have a quick question about your answer: You claimed, in the second paragraph, that for each $x\in A$ there exists a sequence ${x_n}\in[a,b]\setminus A$ such that $x_n\to x$. Could you please explain why? – Beerus May 08 '24 at 19:35
  • Firstly, $A$ cannot contains any interior point (relative to the relative topology of $[a,b]$). For, if there exists $x\in A$ and $\delta>0$ such that $(x-\delta,x+\delta)\cap[a,b]\subseteq A$, then obviously $A$ will have positive Lebesgue measure. (Of course, we assume that $a<b$). – Danny Pak-Keung Chan May 08 '24 at 22:19
  • Now, the negation of the above is: $\forall x\in A,,\forall\delta>0$ $(x-\delta,x+\delta)\cap[a,b]\nsubseteq A$. Let $x\in A$. Put $\delta=1$, then there exists $x_{1}\in(x-1,x+1)\cap[a,b]\setminus A$. – Danny Pak-Keung Chan May 08 '24 at 22:19
  • In particular, $x\in A^{c}$. Note that $x\in A$ while $x_{1}\in A^{c}$, so $|x-x_{1}|>0$. Now, suppose that $x_{1},x_{2},\ldots,x_{n}\in A^{c}$ have been chosen such that $0<|x-x_{n}|<|x-x_{n-1}|<\ldots < |x-x_{1}|$ and $|x-x_{k}|<\frac{1}{k}$ for $k=1,\ldots,n$. Put $\delta=\min(\frac{1}{n+1},|x-x_{n}|)>0$, then we obtain $x_{n+1}\in(x-\delta,x+\delta)\cap[a,b]\setminus A$. Clearly $x_{n+1}\in A^{c}$, $|x-x_{n+1}|<\delta$. Again, $x_{n+1}\in A^{c}$ and $x\in A$, so $0<|x-x_{n+1}|$. By recursion, we obtain a sequence $(x_{n})$ in $A^{c}$ which has the desired property. – Danny Pak-Keung Chan May 08 '24 at 22:19
  • If you are satisfied, please give me upvote to compensate the downvote. – Danny Pak-Keung Chan May 08 '24 at 22:22
  • Thank you very much! I have upvoted it. – Beerus May 08 '24 at 23:26