Let $f$ be a continuous function. I wish to show $$\lim_{n \rightarrow \infty} \int_0^1 n x^n f(x) \; dx = f(1)$$ I can try to split up the integral over intervals $[0, 1-\delta]$, $[1-\delta, 1]$. The integral over $[0, 1-\delta]$ vanishes as $n \rightarrow \infty$, but I am having difficulty estimating the integral over $[1-\delta, 1]$. Thanks!


Your idea is great. One can do the following. First, note that for any integrable $f$ $$\int_0^1 t^nf(t)dt\to 0$$ so we may consider $n+1$ instead of $n$ inside the integrand. This helps, because now $f(1)=\displaystyle\int_0^1(n+1)t^nf(1)dt$. Pick $1>\delta>0$, let $g(t)=f(t)-f(1)$ and note $g(t)\to 0$ as $t\to 1$, $g(1)=0$, and write as you say $$\begin{align}\left|\int_0^1 (n+1)t^n(f(t)-f(1))dt\right|&=\left|\int_0^{1-\delta} (n+1)t^n g(t)dt+\int_{1-\delta}^1 (n+1)t^n g(t)dt\right|\\ &\leqslant \left|\int_0^{1-\delta} (n+1)t^n g(t) dt\right|+\left|\int_{1-\delta}^1 (n+1)t^n g(t) dt\right|\end{align} $$

The first integrand goes to zero since $(n+1)t^n (f(t)-f(1))\to 0$ uniformly over that interval for any $\delta >0$. Given $\varepsilon>0$; can you make the second integral $<\varepsilon$ knowing that $f(t)-f(1)\to 0$ as $t\to 1$?


Given $\varepsilon>0$; choose $\delta>0$ so that $|f(t)-f(1)|<\varepsilon$ if $t\in[1-\delta,1]$. Then $$\begin{align}\left|\int_{1-\delta}^1 (n+1)t^n(f(t)-f(1))dt\right|&\leqslant \int_{1-\delta}^1 (n+1)t^n|f(t)-f(1)|dt\\ &\leqslant \varepsilon\int_{1-\delta}^1 (n+1)t^n dt \\ &\leqslant \varepsilon\int_0^1 (n+1)t^n dt =\varepsilon\end{align}$$

Note no $n$ appeared! This is why $n+1$ is much more convenient that $n$, namely, because it normalizes the integral.

  • $\begingroup$ @user8123123 It should read $f(1)$ in the integrand. Fixing. $\endgroup$ – Pedro Tamaroff Feb 16 '14 at 2:56
  • $\begingroup$ Ok. Using your hint, I think I have the idea for the second integral. For $\delta$ sufficiently small, the integral over $[1-\delta, 1]$ is approximately $f(1) \int_{1-\delta}^{1} n x^n = f(1)\frac{n}{n+1} x^{n+1} \Big|^1_{1-\delta} \rightarrow f(1)$ as $n \rightarrow \infty$. $\endgroup$ – user8123123 Feb 16 '14 at 3:06
  • $\begingroup$ @user8123123 You can make that more rigorous. See what I added. $\endgroup$ – Pedro Tamaroff Feb 16 '14 at 3:36

Alternatively, let me give you an idea which works remarkably well for problems like this:

1) Prove the result for $f(x)$ a polynomial (easy)

2) Use the fact that polynomials are dense in $C[0,1]$ to deduce the general result.


Let $r\in (0,1).$

Let $M=\sup_{x\in [0,1]}|f(x)|.$

For $r\in (0,1)$ let $U(r)=\sup_{x\in [r,1]}f(x)$ and $L(r)=\inf_{x\in [r,1]}f(x).$

For all $n\in \Bbb N$ we have $$(I). \quad\frac {-n}{n+1}r^{n+1}M\le \int_0^rnx^nf(x)fdx\le \frac {n}{n+1}r^{n+1}M.$$ $$(II). \quad \frac {n}{n+1}(1-r^{n+1})L(r)\le \int_r^1nx^nf(x)dx\le \frac {n}{n+1}(1-r^{n+1})U(r).$$

For brevity let $B(r,m)= r^{m+1}M+(1-r^{m+1})U(r).$

Then $\lim_{m\to \infty}B(r,m)=U(r).$ So $$\lim_{m\to \infty}\sup_{n\ge m}B(r,m)=U(r).$$

For brevity let $I(n)=\int_0^1nx^nf(x)dx.$

By $(I)$ and $(II) $ we have $$\lim_{m\to \infty}\sup_{n\ge m}\frac {n+1}{n}I(n)\le$$ $$\le \lim_{m\to \infty}\sup_{n\ge m}r^{n+1}M+(1-r^{n+1})U(r)=$$ $$=\lim_{m\to \infty}\sup_{n\ge m}B(r,m)=$$ $$=\lim_{m\to \infty}B(r,m)=U(r).$$

This holds for every $r\in (0,1),$ so $$(III).\quad \lim_{m\to \infty}\sup_{n\ge m}\frac {n}{n+1}I(n)\le \lim_{r'\to 1^-} \sup_{r\in (r',1)}U(r)=f(1).$$

By similar techniques we obtain $$(IV).\quad \lim_{m\to \infty}\inf_{n\ge m}\frac {n}{n+1}I(n)\ge \lim_{r'\to 1^-}\inf_{r\in (r',1)}L(r)= f(1).$$

By $(III)$ and $(IV)$ we have $\lim_{n\to \infty}\frac {n}{n+1}I(n)=f(1).$ So $\lim_{n\to \infty} I(n)=f(1).$


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.