Here is my attempt of proving $cl(C[a,b])=L^p[a,b]$. I would be immensely thankful to anybody reading it, pointing out the mistakes or improving it. Since I just began studying functional analysis, it's important to me that my understanding is right. Thanks!

Let $\Sigma$ be Borel $\sigma$-Algebra of $[a,b]$.
We can assume $cl(span\{ \chi_A|A \in \Sigma\})=L^p$.

Claim 1. $span\{\chi_O|O \subseteq [a,b] \text{ open} \}$ is dense in $span\{ \chi_A|A \in \Sigma\}$.

Proof: Let $x=\sum_{i=1}^n a_i \chi_{A_i}\in span\{ \chi_A|A \in \Sigma\}, \epsilon>0$ be arbitrarily chosen.
Define $y:=\sum^n_{i=1}a_i \chi_{O_i}$ where $O_i$ are open sets such that $$ \lambda(O_i\setminus A_i)\leq \max_i\lambda(O_i\setminus A_i) < \frac{\epsilon}{n\cdot \max_i |a_i|}$$ This construction is possible since Lebesgue measure is outer regular. Indeed \begin{align} \Vert x-y \Vert_p^p&= \Vert \sum_{i=1}^n a_i (\chi_{A_i}- \chi_{O_i}) \Vert_p^p\\ &\leq \max_i |a_i| \sum_{i=1}^n \Vert \chi_{A_i}- \chi_{O_i} \Vert_p^p\\ &\leq n\cdot \max_i |a_i| \cdot \max_j \Vert \chi_{A_j}- \chi_{O_j} \Vert_p^p\\ &\leq n\cdot \max_i |a_i| \cdot \max_j \lambda(O_j\setminus A_j) <\epsilon \end{align} q.e.d.

Claim 2. $span\{\chi_O|O \subseteq [a,b] \text{ open} \}$ is dense in $L^p$.

Proof: By claim 1 and idempotency of closure we have
$$ cl(span\{\chi_O|O \subseteq [a,b] \text{ open} \})=cl(span\{ \chi_A|A \in \Sigma\})=L^p$$ q.e.d.

Claim 3. $span\{\chi_I| I\subset [a,b] \text{ open interval}\}$ is dense in $span\{\chi_O|O \subset [a,b] \text{ open} \}$ and hence in $L^p$.

Proof: Let $x=\sum_{i=1}^m b_i \chi_{O_i}\in span\{\chi_O|O \subset [a,b] \text{ open} \}, \epsilon>0$ arbitrary. Each $O_i$ can be written as the disjoint union of open intervals $I_{ij}$ with $j$ ranging over not necessarily finite index set $J_i$. Now for each $O_i$ we pick enough number $n(i)$ of those intervals such that $$\lambda(O_i \setminus \cup_{j=1}^{n(i)} I_{ij})\leq \max_{i} \lambda(O_i \setminus \cup_{j=1}^{n(i)} I_{ij})\leq \frac{\epsilon}{m\cdot \max_k |b_k|}$$ Define $y:=\sum_{i=1}^m a_i \sum^{n(i)}_{j=1} \chi_{I_{ij}} \in span\{\chi_I| I\subset [a,b] \text{ open interval}\}$, since it's finite linear combination. Furhermore

\begin{align} \Vert x-y \Vert_p^p&=\Vert \sum_{i=1}^m b_i (\chi_{O_i}-\sum^{n(i)}_{j=1} \chi_{I_{ij}}) \Vert_p^p\\ &\leq \max_k |b_k|\cdot m \cdot \max_{i} \lambda(O_i \setminus \cup_{j=1}^{n(i)} I_{ij})\leq \epsilon \end{align} q.e.d.

Claim 4. $C[a,b]$ is dense in $span\{\chi_I| I\subset [a,b] \text{ open interval}\}$ and hence in $L^p$.

Claim 4. $span\{\chi_I| I\subset [a,b] \text{ open interval}\} \subseteq cl(C[a,b])$ and hence dense in $L^p$.

Proof: Let $x=\sum_{i=1}^n c_i \chi_{I_i}$, $\epsilon>0$. We need to show $B_\epsilon(x)\cap C[a,b]$ is not empty.
For it we construct such function $f$ that tightly enough (in sense of $\Vert f-x \Vert_p^p <\epsilon$) and continuously encapsulates graph of $x$. (I am not sure how to explicitly define it.)

  • 1
    $\begingroup$ Claim 4 does not make sense. $C[a,b]$ is not contained in this span. Instead, you can show that the span is contained in the closure of $C[a,b]$. $\endgroup$ – tomasz Sep 3 '18 at 0:15
  • $\begingroup$ @tomasz You are right, I don't know what I was thinking... Thanks for taking a look! $\endgroup$ – user3342072 Sep 3 '18 at 1:02

$C[a,b]$ is not dense in $L^{\infty}[a,b]$. So I am assuming $1 \le p < \infty$. If $C[a,b]$ is not dense in $L^{p}[a,b]$, then there exists $\Phi \in (L^{p})^*$ such that $\Phi=0$ on $L^{p}[a,b]$, but $\Phi \ne 0$. However, $\Phi(f)=\int_{a}^{b}f(t)\phi(t)dt$ for all $f\in L^p$ and some $\phi \in L^{q}=(L^{p})^*$. Because all characteristic functions $\chi_{E}$, where $E$ is a measurable subset of $[a,b]$, are in $L^p$, then $\int_{E}f(t)dt=0$ for all such $E$. From this it follows that $f=0$.


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.