25

Suppose $f:[0,1]\to \mathbb{R}$ is continuous. Show that 

$$\lim_{n\to\infty} n\int_0^1 f(x)x^n\,dx = f(1).$$

My answer so far: First I want to assume that $f\in C^1$. Then 

$$n\int_0^1f(x)x^n\,dx = \left[\frac{n}{n+1}x^{n+1}f(x)\right]_0^1 - \frac{n}{n+1}\int_0^1 x^{n+1}f'(x)\, dx\\ \frac{n}{n+1}f(1) - \frac{n}{n+1}\int_0^1 x^{n+1}f'(x)\, dx,$$

which goes to $f(1)$ because the last integral goes to zero.

But approximating $f$ by $\phi\in C^1$ won't necessarily work, because $\phi(1)$ may not equal $f(1)$... how can we finish the argument?

Mittens
  • 46,352
user88203
  • 577
  • 1
    I don't think you could argue by proving it in the case when $f$ is continuously differentiable and then somehow using that to prove it when $f$ is just continuous - you'd have to argue via other methods. – Andrew D Jul 29 '13 at 22:06
  • 2
    Hint: $f(1)=(n+1)\int_{0}^{1} f(1)x^n dx$. – Amitesh Datta Jul 29 '13 at 22:09
  • 1
    For a different approach, see the first answer here. – David Mitra Jul 29 '13 at 22:09
  • 1
    You can simplify your part of the argument: Only polynomials, rather than arbitrary $C^1$ functions, need to be considered. This case is trivial, as we only need to argue for monomials $x^k$, and everything is explicit here. The uniform convergence argument as in the answer below then goes unchanged. – Andrés E. Caicedo Jul 29 '13 at 22:25
  • https://math.stackexchange.com/q/128823/321264 – StubbornAtom Sep 06 '20 at 18:26

10 Answers10

22

Here is a more elementary method than you proposed:

First, note that if $f$ is continuous on $[0,1]$, then it is necessarily bounded on $[0,1]$; say $\lvert f(x)\rvert\leq M$ for all $x\in[0,1]$. If we define $\delta_n:=\frac{1}{\sqrt{n}}$, then $$ \left\lvert n\int_0^{1-\delta_n}f(x)x^n\,dx\right\rvert\leq Mn\int_0^{1-\delta_n}x^n\,dx=\frac{n}{n+1}\left(1-\frac{1}{\sqrt{n}}\right)^{n+1}\rightarrow0\text{ as }n\rightarrow\infty. $$ Now, let $\epsilon>0$ be given. Continuity of $f$ at $1$ implies that there exists $\delta>0$ such that $\lvert 1-x\rvert<\delta$ implies $\lvert f(x)-f(1)\rvert<\epsilon$. Choose $N\in\mathbb{N}$ such that $0<\delta_n<\delta$ for all $n\geq N$. Then for $n\geq N$, $$ n\int_{1-\delta_n}^1(f(1)-\epsilon)x^n\,dx\leq n\int_{1-\delta_n}^1 f(x)x^n\,dx\leq n\int_{1-\delta_n}^{1}(f(1)+\epsilon)x^n\,dx. $$ Computing the left integral $$ \frac{n}{n+1}\left(1-\left(1-\frac{1}{\sqrt{n}}\right)^{n+1}\right)\left(f(1)-\epsilon\right)\rightarrow f(1)-\epsilon\text{ as }n\rightarrow\infty; $$ the right integral yields the same, except with $f(1)+\epsilon$. Then $$ f(1)-\epsilon\leq\liminf_{n\rightarrow\infty}\ n\int_0^1f(x)x^n\,dx\leq\limsup_{n\rightarrow\infty}\ n\int_0^1 f(x)x^n\,dx\leq f(1)+\epsilon. $$ But, this holds for any $\epsilon>0$; so, letting $\epsilon\rightarrow0$, we get the desired result.

Nick Peterson
  • 33,058
  • 1
    This is a great answer, Nicholas. Thanks for your contribution! (I would upvote it except that I've exhausted my daily upvote quota (sorry!). I'll return here in a couple of hours and upvote!) – Amitesh Datta Jul 29 '13 at 22:36
17

We can finish the argument as follows. (Note: We'll assume that the limit in question exists for $f$ and establish that it's equal to $f(1)$. Technically, we should prove that this limit exists as Peter Tamaroff notes below (thanks!). A minor modification of the following argument simultaneously establishes the existence of the limit and its value but we'll leave that as an exercise to the reader.) Let $\epsilon>0$. Choose $\phi\in C^1$ such that $\left|f(x)-\phi(x)\right|<\epsilon$ for all $x\in [0,1]$. You've proven that $$\lim_{n\to\infty} n\int_{0}^{1} \phi(x)x^n=\phi(1).$$ Therefore,

$$\begin{align}\left|\lim_{n\to\infty} n\int_{0}^{1} f(x)x^n dx -\lim_{n\to\infty} n\int_{0}^{1} \phi(x)x^n dx\right|&=\left|\lim_{n\to\infty} n\int_{0}^{1} (f(x)-\phi(x))x^n dx\right|\\ &\leq \lim_{n\to\infty} n\int_{0}^{1} \left|(f(x)-\phi(x))x^n\right| dx\\ &< \lim_{n\to\infty} n\int_{0}^{1} \epsilon x^n dx\\ &=\lim_{n\to\infty} \epsilon \frac{n}{n+1}\\ &=\epsilon\end{align}$$

Therefore,

$$\begin{align}\left|\lim_{n\to\infty} n\int_{0}^{1} f(x)x^n dx - f(1)\right|\leq \left|\lim_{n\to\infty} n\int_{0}^{1} f(x)x^n dx - \phi(1)\right| + \left|\phi(1)-f(1)\right|&<\epsilon + \epsilon\\&=2\epsilon\end{align}$$

Since $\epsilon>0$ was arbitrary, we conclude that $$\lim\limits_{n\to\infty} n\int_{0}^{1} f(x)x^n dx=f(1)$$ for all continuous functions $f:[0,1]\to \mathbb{R}$.

Amitesh Datta
  • 21,429
  • 1
    Why'd there exist such $\varphi \in C^1$? – Kunnysan Jul 29 '13 at 22:22
  • 2
    @Kunnysan I'm not sure which tools you'd like to use but this follows, e.g., from the Stone-Weierstrass theorem. – Amitesh Datta Jul 29 '13 at 22:27
  • 1
    Yeah of course. You could have chosen even a polynomial. – Kunnysan Jul 29 '13 at 22:32
  • 1
    I was writing exactly same solution. Discarded it seeing yours, so +1 instead, :) – Kunnysan Jul 29 '13 at 22:36
  • Thanks @Kunnysan and I'm really sorry that you had to discard your answer! (The same thing happens to me sometimes. In a few years when there are more users on this website, it will hardly be possible to read the question, I suspect, before someone posts an answer!) – Amitesh Datta Jul 29 '13 at 22:39
  • Amitesh: I disagree with the fact you keep $\lim\limits_{n\to\infty}$ in each step. This is unnecessary and slightly informal. For example, in your last step, you can just write $$=\epsilon\frac{n}{n+1}<\epsilon$$ since the quotient is $<1$. – Pedro Jul 29 '13 at 22:54
  • Could you answer on the above? You're using $\lim$ when you still haven't shown the limit exists, Amitesh. – Pedro Jul 30 '13 at 00:46
  • Thanks for your comment, @Peter! (I wasn't online when you posted the last two comments.) I did observe when writing my answer that one needs to show that the limit does indeed exist before any computations may be carried out. As you point out, one can show this by simply removing the $\lim_{n\to\infty}$ in each step (and this still establishes the claim as well as the existence of the limit). So, I was sloppy but I was hoping the OP would understand that one does need to prove the limit exists and figure out how to do that. I'll add a note to my answer to mention this. (+1 to your answer.) – Amitesh Datta Jul 30 '13 at 01:24
  • @AmiteshDatta I would simply avoid the use of $\lim$ during estimations. Just estimate and then argue! =) – Pedro Jul 30 '13 at 01:25
  • @Peter I agree with you! – Amitesh Datta Jul 30 '13 at 01:28
  • It's interesting to me that in this solution the integral is an inner product between $f$ and $g_n : = nx^n$, but wee can't use the Cauchy-Schwarz inequality to estimate $$n\int_0^1 (f(x) - \phi(x))x^n dx$$ because $|g_n|^2$ is not bounded in $n$. – user88203 Jul 30 '13 at 14:53
12

First, note that $$\int_0^1 x^n f(x)dx\to 0$$

since $f$ is bounded, so we can prove that $$(n+1)\int_0^1 x^n f(x)dx\to f(1)$$

But note $$\left( {n + 1} \right)\int_0^1 {x^n}f (1)dx = f(1).$$ so it suffices to consider the case $f(1)=0$.

THM Suppose that $f:[0,1]\to \Bbb R$ is continuous and $f(1)=0$. Then $$\mathop {\lim }\limits_{n \to \infty } \left( {n + 1} \right)\int_0^1 f (x){x^n}dx = 0$$

P Let $\epsilon >0$ be given. By continuity, there exists a neighborhood $[1-\delta,1]$ such that $$|f(x)|<\frac\varepsilon2$$ whenever $x\in[1-\delta,1]$. Write $$\left( {n + 1} \right)\left| {\int_0^1 f (x){x^n} dx} \right| \leqslant \left( {n + 1} \right)\left| {\int_0^{1 - \delta } f (x){x^n} dx} \right| + \left( {n + 1} \right)\left| {\int_{1 - \delta }^1 f (x){x^n} dx} \right|$$ so that $$\left( {n + 1} \right)\left| {\int_{1 - \delta }^1 {f\left( x \right){x^n} dx} } \right| \leqslant \left( {n + 1} \right)\frac{\varepsilon }{2}\int_{1 - \delta }^1 {{x^n} dx} \leqslant \left( {n + 1} \right)\frac{\varepsilon }{2}\int_0^1 {{x^n} dx} = \frac{\varepsilon }{2}$$

On the other hand, $|f|$ attains a maximum on $[0,1-\delta]$ and we have $$\left( {n + 1} \right)\left| {\int_0^{1 - \delta } {f\left( x \right){x^n}{\mkern 1mu} dx} } \right| \leqslant \left( {n + 1} \right)\int_0^{1 - \delta } {\left| {f\left( x \right)} \right|{x^n}{\mkern 1mu} dx} \leqslant M\left( {n + 1} \right)\int_0^{1 - \delta } {{x^n}{\mkern 1mu} dx} \leqslant M{\left( {1 - \delta } \right)^{n + 1}}$$

Since $1-\delta <1$, this goes to $0$; so the claim follows. Note we could have also used that $(n+1)x^n$ converges to zero uniformly on $[0,1-\delta]$ for any $0<\delta <1$ $\blacktriangle$

OBS Note how the proof works: $x^n$ crunches everything away from $1$, and continuity of $f$ plus $f(1)=0$ crunches everything near $1$.

Pedro
  • 125,149
  • 19
  • 236
  • 403
  • @PeterTamaroff your proof is similar to Robert's proof here http://math.stackexchange.com/questions/168163/find-functions-family-satisfying-lim-n-to-infty-n-int-01-xn-fx-f1/168170#168170. There are other approaches there you might want to know .. – user 1591719 Jul 31 '13 at 15:06
9

First, note that $$ (n+1)\color{#C00000}{\int_0^ax^n\,\mathrm{d}x}=a^{n+1}\tag{1} $$ and $$ (n+1)\color{#00A000}{\int_0^1x^n\,\mathrm{d}x}=1\tag{2} $$ Pick an $\epsilon>0$. Since $f$ is continuous, there is a $\delta>0$ so that for all $x\in[1-\delta,1]$, we have $|f(x)-f(1)|\le\epsilon$. Since $f$ is continuous on $[0,1]$, there is an $M$ so that $|f(x)|\le M$ for $x\in[0,1]$. Furthermore, there is an $N$ so that for $n\ge N$, we have $2M(1-\delta)^{n+1}\le\epsilon$.

Thus, for $n\ge N$ $$ \begin{align} &\left|f(1)-(n+1)\int_0^1x^nf(x)\,\mathrm{d}x\right|\\ &=(n+1)\left|\int_0^1x^n(f(1)-f(x))\,\mathrm{d}x\right|\\ &=(n+1)\left|\color{#C00000}{\int_0^{1-\delta}x^n(f(1)-f(x))\,\mathrm{d}x} +\color{#00A000}{\int_{1-\delta}^1x^n(f(1)-f(x))\,\mathrm{d}x}\right|\\ &\le\color{#C00000}{2M(1-\delta)^{n+1}}+\color{#00A000}{\epsilon}\\ &\le2\epsilon\tag{3} \end{align} $$ Thus, $$ \lim_{n\to\infty}(n+1)\int_0^1x^nf(x)\,\mathrm{d}x=f(1)\tag{4} $$ Since $\lim\limits_{n\to\infty}\dfrac n{n+1}=1$, we get $$ \lim_{n\to\infty}n\int_0^1x^nf(x)\,\mathrm{d}x=f(1)\tag{5} $$

robjohn
  • 353,833
  • Nice, but it can be written a bit easier: $$\lim\limits_{n\to+\infty}n\int_0^1x^nf(x)dx =\lim\limits_{\delta\to 0^{+}}\lim\limits_{n\to+\infty}n\int_0^{1-\delta}x^nf(x)dx +\lim\limits_{\delta\to 0^{+}}\lim\limits_{n\to+\infty}n\int_{1-\delta}^1x^nf(x)dx$$ For any $\delta\in (0,1] \lim\limits_{n\to+\infty}n\int_0^{1-\delta}x^nf(x)dx=0$ while $$\inf{f(x):x\in[1-\delta,1]}\leqslant\lim\limits_{n\to+\infty}n\int_{1-\delta}^1x^nf(x)dx\leqslant\sup{f(x):x\in[1-\delta,1]}$$ so, when $\delta\to 0^+$ then $$\lim\limits_{\delta\to 0^+}\lim\limits_{n\to+\infty}n\int_{1-\delta}^1x^nf(x)dx=f(1)$$ – Darius Jun 23 '14 at 06:53
9

By changing the variable, let $ x=t^{\frac{1}{n}}$ and we have $$n\int_0^1 x^n f(x)dx=\int_0^1 f\left(t^{\frac{1}{n}}\right)t^{\frac{1}{n}}dt,$$

and by dominated convergence theorem we conclude: $$\lim_n n\int_0^1 x^n f(x)dx=f(1).$$

6

Thought I'd write out the polynomial method @Kunnysan mentioned in the comments.

Consider an arbitrary polynomial $p(x) = a_0 + a_1x + \dots + a_kx^k$. We can calculate \begin{align} \lim_{n \rightarrow \infty} n \int_0^1 p(x)x^n \, dx &= \lim_{n \rightarrow \infty} n \int_0^1 a_0x^n + a_1x^{n+1} + \dots + a_kx^{n+k} \, dx \\ &= \lim_{n \rightarrow \infty} \left( \frac{n}{n+1} a_0 + \frac{n}{n+2} a_1 + \dots + \frac{n}{n+k+1} a_k \right) \\ &= a_0 + a_1 + \dots + a_k \\ &= p(1) \end{align}

By the Weierstrass approximation theorem there exists a sequence of polynomials $\{p_m\}$ such that $p_m(x) \rightarrow f(x)$ uniformly.

We then write \begin{align} \lim_{n \rightarrow \infty} n\int_0^1 f(x)x^n \, dx &= \lim_{n \rightarrow \infty} n \int_0^1 \lim_{m \rightarrow \infty} p_m(x)) x^n \, dx \\ &= \lim_{n\rightarrow n} n \int_0^1 \lim_{m \rightarrow \infty} p_m(x)x^n \, dx \\ &= \lim_{n \rightarrow \infty} \lim_{m \rightarrow \infty} \int_0^1 p_m(x)x^n \, dx \end{align} where the interchanging of the limit and the integral is valid because the sequence $p_m(x)x^n$ converges uniformly to $f(x)x^n$.

Since $p_m(x)$ is a polynomial, we can use our preliminary work to write $$ \lim_{n \rightarrow \infty} \lim_{m \rightarrow \infty} \int_0^1 p_m(x)x^n \, dx = \lim_{n \rightarrow \infty} \lim_{m \rightarrow \infty} \left( \frac{n}{n+1} a_0 + \frac{n}{n+2} a_1 + \dots + \frac{n}{n+k+1} a_k \right)$$ where $a_i$ are the coefficients of the polynomial $p_m(x)$.

Lastly we interchange limits again, first taking the limit as $n \rightarrow \infty$:

$$\lim_{n \rightarrow \infty} \lim_{m \rightarrow \infty} \left( \frac{n}{n+1} a_0 + \frac{n}{n+2} a_1 + \dots + \frac{n}{n+k+1} a_k \right) = \lim_{m \rightarrow \infty} p_m(1) = f(1)$$

5

Since $f$ is continuous, it is bounded on the compact interval $[0,1]$, say $|f(x)|<M$ for all $x\in[0,1]$. Also, for any $\epsilon>0$, we find delta such that $|f(x)-f(1)|<\epsilon$ for all $x>1-\delta$. Then $$\int_0^1 x^nf(x)\,dx = \int_0^{1-\delta} x^n f(x)\,dx+\int_{1-\delta}^1 x^n f(1)\,dx+\int_{1-\delta}^1 x^n (f(x)-f(1))\,dx$$ The first summand can be estimated by $$\left|\int_0^{1-\delta} x^n f(x)\,dx\right|\le \int_0^{1-\delta}\left| x^n f(x)\right|\,dx\le M\int_0^{1-\delta}x^n\,dx=\frac1{n+1} M(1-\delta)^{n+1}.$$ The second is just $$\int_{1-\delta}^1 x^n f(1)\,dx=\frac{f(1)}{n+1}\cdot(1-(1-\delta)^{n+1}).$$ The last can be estimated as $$\left|\int_{1-\delta}^1 x^n (f(x)-f(1))\,dx\right|\le \int_{1-\delta}^1 \left|x^n (f(x)-f(1))\right|\,dx\\\le\epsilon\int_{1-\delta}^1x^n=\frac\epsilon{n+1}\cdot(1-(1-\delta)^{n+1}).$$ As $n\to\infty$, we have $(1-\delta)^{n+1}\to 0$. If you stick these results together, you'll find that $$\lim_{n\to\infty}n\int_0^1x^nf(x)\,dx=f(1).$$

3

Hint: Try $f(x)=x^k$, then a polynomial, and then a general continuous function.

Beni Bogosel
  • 23,891
0

This answer is almost identical to the answer of user @Daniel Watkins. The difference is that it will be shorter by using his proof for functions $C^{1}$ on the interval $[0,1]$. Fixed any continuous function $f:[0,1]\to \mathbb{R}$, by Weierstrass' approximation theorem there exists a sequence of polynomials $p_{k}:[0,1]\to \mathbb{R}$ that converges uniformally to $f:[0,1]\to \mathbb{R}$. In particular $\lim_{k\to\infty}p_{k}(1)=f(1)$. From what has already been worked out for $C^{1}$ functions in this question ( polynomials are $C^{1}$ functions ) we have \begin{align} \lim_{n\to \infty}\left(n\int_0^1f(x)x^n\,dx\right) =& \lim_{n\to \infty}\left(\frac{n}{n+1}f(1) - \frac{n}{n+1}\int_0^1 x^{n+1}f'(x)\, dx\right), \\ =& \lim_{n\to \infty}\left(\frac{n}{n+1}\left(\lim_{k\to \infty}p_{k}(1)\right) - \frac{n}{n+1}\int_0^1 x^{n+1}f'(x)\, dx\right), \\ =& \lim_{n\to \infty}\;\lim_{k\to \infty}\;\frac{n}{n+1}\left(p_{k}(1) - \frac{n}{n+1}\int_0^1 x^{n+1}f'(x)\, dx\right), \\ =& \lim_{k\to \infty}\;\lim_{n\to \infty}\;\frac{n}{n+1}\left(p_{k}(1) - \frac{n}{n+1}\int_0^1 x^{n+1}f'(x)\, dx\right), \\ =& \lim_{k\to \infty}p_{k}(1) \\ =& f(1) \end{align}

Elias Costa
  • 15,282
0

In this answer we apply the useful inequality

$$ \frac{t}{1+t}\leq \ln( 1+t ) \leq t \qquad \mbox{ for all } \qquad t\geq -1 $$

In particular we have for $t=(-1/n)$ and $n>1$ $$ \color{\red}{\frac{-1}{n-1}=}\frac{(-1/n)}{1+(-1/n)}\leq \ln( 1+(-1/n) ) \leq (-1/n) $$ This inequality implies $$ e^{\frac{-1}{n-1}}\leq e^{\ln(1-1/n)}\leq e^{-1/n} $$ and $$ e^{\frac{-\sqrt{n}}{n-1}}\leq e^{\sqrt{n}\cdot \ln(1-1/n)}\leq e^{-\sqrt{n}/n} \color{\red}{=e^{-1/\sqrt{n}}} $$ Now, note that $$ n\int_{0}^{1}f(x)\cdot x^{n} \, \mathrm{d} x = n\int_{0}^{\frac{1}{n}}f(x)\cdot x^{n} \, \mathrm{d} x + n\int_{1/n}^{\left(1-\frac{1}{n}\right)}f(x)\cdot x^{n} \, \mathrm{d} x + n\int_{\left(1-\frac{1}{n}\right)}^{1}f(x)\cdot x^{n} \, \mathrm{d} x $$ There is $M>0$ such that $|f(x)|<M$ for all $x\in [0,1]$ and we have \begin{align} \left| n\int_{1/n}^{\left(1-\frac{1}{n}\right)}f(x)\cdot x^{n} \, \mathrm{d} x \right| \leq & n\int_{1/n}^{\left(1-\frac{1}{n}\right)}|f(x)|\cdot x^{n} \, \mathrm{d} x \\ \leq & n\int_{1/n}^{\left(1-\frac{1}{n}\right)}|f(x)|\cdot x^{\sqrt{n}} \, \mathrm{d} x \\ \leq & n\int_{1/n}^{\left(1-\frac{1}{n}\right)}M \cdot x^{\sqrt{n}} \, \mathrm{d} x \\ = & n\int_{1/n}^{\left(1-\frac{1}{n}\right)}M \cdot e^{\sqrt{n}\cdot ln x} \, \mathrm{d} x \\ \leq & n\int_{1/n}^{\left(1-\frac{1}{n}\right)}M \cdot e^{\sqrt{n}\cdot ln \left(1-\frac{1}{n}\right)} \, \mathrm{d} x \\ \leq & n\int_{1/n}^{\left(1-\frac{1}{n}\right)}M \cdot e^{\sqrt{n}\cdot ln \left(1-\frac{1}{n}\right)} \, \mathrm{d} x \\ \leq & (\sqrt{n})^{2}\int_{1/n}^{\left(1-\frac{1}{n}\right)}M \cdot e^{-1/\sqrt{n}} \, \mathrm{d} x \\ =& (\sqrt{n})^{2}M \cdot e^{-1/\sqrt{n}}\left[\left(1-\frac{1}{n}\right) -\frac{1}{n} \right] \overset{n\to \infty}{\longrightarrow}0 \end{align}

It is easy to verify that $$\lim_{n\to\infty} n\int_{0}^{\frac{1}{n}}f(x)\cdot x^{n} \, \mathrm{d} x=0$$ Finally, by Dominated Convergence Theorem and Lebesgue's Differentiation Theorem $$ \lim_{n\to \infty} \left( n\int_{\left(1-\frac{1}{n}\right)}^{1}f(x)\cdot x^{n} \, \mathrm{d} x \right) = \lim_{n\to \infty} \left( \cfrac{\int_{\left(1-\frac{1}{n}\right)}^{1}f(x)\cdot x^{n} \, \mathrm{d} x}{\frac{1}{n}} \right) = f(1) $$

Elias Costa
  • 15,282