0

I thought I had a proof that utilized the dominated convergence theorem, but I realized that I was mistaken in my use of the theorem, and my proof never really used that $f(x)\rightarrow0$ as $x\rightarrow1$.

So all I've really been able to show is that $nx^nf(x)\rightarrow0$ point wise on $[0,1)$. Beyond that I'm stuck.

One thing I've considered is using the fact that $f\in L^1([0,1])$ means that there is some measurable $A\subseteq[0,1]$ such that $m(A)<\epsilon$ and $f$ is bounded on $[0,1]\sim A$ ($\epsilon$ can be arbitrarily small). This would at least give me uniform convergence on $[0,1]\sim A$ and thus prove the claim for $[0,1]\sim A$. But I'm not totally clear on where I would proceed from there, or if it's even the best approach.

Any advice would be greatly appreciated. Thanks in advance.

Bears
  • 684

2 Answers2

1

Let $\varepsilon > 0$. Because $f(x) \rightarrow 0$ as $x \rightarrow 1$, there exists $0 < \eta < 1$ such that for all $x \in [1-\eta, 1]$, $|f(x)| \leq \varepsilon/2$. So you have $$\left| n \int_0^1 x^n f(x) \mathrm{dx}\right| \leq \left|n \int_0^{1-\eta} x^n f(x) \mathrm{dx}\right| + \left|n \int_{1-\eta}^1 x^n f(x) \mathrm{dx}\right|$$

So $$\left| n \int_0^1 x^n f(x) \mathrm{dx}\right| \leq n (1-\eta)^n \int_0^{1-\eta} |f(x)| \mathrm{dx} + n \int_{1-\eta}^1 x^n \frac{\varepsilon}{2} \mathrm{dx}$$

$$ \leq n (1-\eta)^n || f ||_1 + n \frac{\varepsilon}{2} \int_0^1 x^n \mathrm{dx}$$

$$\leq n (1-\eta)^n || f ||_1 + \frac{\varepsilon}{2} \frac{n}{n+1}$$

The first term $n (1-\eta)^n || f ||_1$ tends to $0$, and the second term tends to $\varepsilon /2$, so the whole term tends to $\varepsilon /2$. You deduce that there exists $N \in \mathbb{N}$ such that for $n \geq N$, $$\left| n \int_0^1 x^n f(x) \mathrm{dx}\right| \leq \varepsilon$$

TheSilverDoe
  • 29,720
  • Okay. This makes a lot of sense.

    So, my first thought when I saw this problem was to use one of the main convergence theorems (mainly, dominated convergence). The proof you outlined just directly shows that the integral goes to zero. Was there something about this problem that made you think a direct approach would be the best way to go?

    – Bears Aug 12 '20 at 18:42
  • Always try direct approaches before thinking about big theorems :) Here, you can see that the $x^n$ will make the expression crash to $0$... except near $1$. But fortunately, you have something that let you control what happens near $1$ ! This sort of considerations can make you think that a $\varepsilon-\eta$ proof will be sufficient. – TheSilverDoe Aug 12 '20 at 18:51
0

Firstly, we show that for each $c\in[0,1)$, $\lim_{n\rightarrow\infty}\int_{0}^{c}nx^{n}f(x)dx=0$. Observe that \begin{eqnarray*} \left|\int_{0}^{c}nx^{n}f(x)dx\right| & \leq & \int_{0}^{c}nc^{n}|f(x)|dx\\ & \leq & nc^{n}\int_{0}^{1}|f(x)|dx\\ & \rightarrow & 0 \end{eqnarray*} as $n\rightarrow\infty.$ Suppose that $\lim_{x\rightarrow1-}f(x)=\xi$. Let $\varepsilon>0$. Choose $c\in[0,1)$ such that $|f(x)-\xi|<\varepsilon$ whenever $x\in[c,1)$. Choose $N_{1}$ such that $\left|\int_{0}^{c}nx^{n}f(x)dx\right|<\varepsilon$ whenever $n\geq N_{1}$. By direct calculation, we have $\int_{c}^{1}nx^{n}\xi dx=\xi\cdot\frac{n}{n+1}(1-c^{n+1})\rightarrow\xi$ as $n\rightarrow\infty$. Choose $N_{2}$ such that $\left|\int_{c}^{1}nx^{n}\xi dx-\xi\right|<\varepsilon$ whenever $n\geq N_{2}$. Note that \begin{eqnarray*} \int_{c}^{1}|nx^{n}f(x)-nx^{n}\xi|dx & \leq & \int_{c}^{1}nx^{n}\varepsilon dx\\ & = & \varepsilon\cdot\frac{n}{n+1}(1-c^{n+1})\\ & < & \varepsilon. \end{eqnarray*} Now, for any $n\geq\max(N_{1},N_{2})$, we have \begin{eqnarray*} \left|\int_{0}^{1}nx^{n}f(x)dx-\xi\right| & \leq & \left|\int_{0}^{c}nx^{n}f(x)dx\right|+\int_{c}^{1}|nx^{n}f(x)-nx^{n}\xi|dx+\left|\int_{c}^{1}nx^{n}\xi dx-\xi\right|\\ & < & 3\varepsilon. \end{eqnarray*} This show that $\lim_{n\rightarrow\infty}\int_{0}^{1}nx^{n}f(x)dx=\xi:=\lim_{x\rightarrow1-}f(x)$.