Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

It can be shown that

$$ n \int_0^1 \frac{x^n - (1-x)^n}{2x-1} \mathrm dx = \sum_{k=0}^{n-1} {n-1 \choose k}^{-1}$$

(For instance see my answer here.)

It can also be shown that $$\lim_{n \to \infty} \ \sum_{k=0}^{n-1} {n-1 \choose k}^{-1} = 2$$

(For instance see Qiaochu's answer here.)

Combining those two shows that

$$ \lim_{n \to \infty} \ n \int_0^1 \frac{x^n - (1-x)^n}{2x-1} \mathrm dx = 2$$

Is there a different, (preferably analytic) proof of this fact? Please do feel free to add a proof which is not analytic.

share|cite|improve this question
What do u mean by an analytic proof – anonymous Oct 11 '10 at 18:20
@Chandru: Using theorems in real/complex analysis etc. For instance, without the n factor, the limit is zero, and that follows from the dominated convergence theorem. – Aryabhata Oct 11 '10 at 18:22
OK! Understood – anonymous Oct 11 '10 at 18:25
Why did you change your name? :) – AD. Jun 29 '11 at 20:49
@AD.: I was thinking of changing it and the comments here helped:… :-) – Aryabhata Jun 29 '11 at 20:52
up vote 14 down vote accepted

Let $x=(1+s)/2$, so that the expression becomes $$ \frac{n}{2^{n+1}} \int_{-1}^1 \frac{(1+s)^n-(1-s)^n}{s} ds = \frac{n}{2^{n}} \int_{0}^1 \frac{(1+s)^n-(1-s)^n}{s} ds. $$ (The integrand is an even function.) Fix some $c$ between 0 and 1, say $c=1-\epsilon$. Then the integral from 0 to $c$ will be small in comparison to $2^n$, since the integrand is bounded by a constant times $(1+c)^n$, so as far as the limit is concerned it is enough to look at the integral from $c$ to 1, and in that integral we can neglect the term $(1-s)^n$ since it will also contribute something much smaller than $2^n$. The surviving contribution therefore comes from $$ \int_c^1 \frac{(1+s)^n}{s} ds, $$ which lies between $$ \int_c^1 \frac{(1+s)^n}{1} ds $$ and $$ \int_c^1 \frac{(1+s)^n}{c} ds, $$ that is, $$ \frac{2^{n+1} - (1+c)^{n+1}}{n+1} < \int_c^1 \frac{(1+s)^n}{s} ds < \frac{1}{c} \frac{2^{n+1} - (1+c)^{n+1}}{n+1} .$$ Multiplying by $n/2^n$ and letting $n\to\infty$ shows that the liminf is at least 2 and the limsup is at most $2/c$. But since this holds for any $c$ between 0 and 1, it follows that liminf=limsup=2, hence the limit is 2.

share|cite|improve this answer
Thanks for this! Though, one thing that seems to be missing is the proof that the limit exists, which I suppose won't be too hard to show (it is not exactly sandwich theorem, but perhaps a 2D variant of it). – Aryabhata Oct 12 '10 at 16:08
@Moron: Doesn't the last step show that the liminf is at least 2, and the limsup is at most 2/c, for all c<1? Hence liminf=limsup=2. – Hans Lundmark Oct 12 '10 at 17:13
Yes that proves it :-), but the last paragraph talks about 'the limit' before doing the squeeze between 2 and 2/c as c->1. I suggest you change the wording to talk about liminf and limsup instead of 'the limit', to make that clear. Sorry, i guess it is just a nitpick. – Aryabhata Oct 12 '10 at 17:25
btw, the only reason for the request is that it was not obvious to me (until you pointed it out :-)) and I am guessing spelling it out might help others in the future. – Aryabhata Oct 12 '10 at 17:48
@Moron: Done! Thanks for pointing it out. Being a nitpick is a good thing in mathematical analysis. :) – Hans Lundmark Oct 13 '10 at 6:43

Here's another idea.

Let $P_n(x)$ be the polynomial defined by $$P_n(x) = \frac{x^n – (1-x)^n}{2x-1},$$ and note that $P_n(x)=(1-x)P_{n-1}(x) + x^{n-1} \qquad (1), $ where $P_1(x)=1.$

Define $I_n = \int_0^{1} P_n(x) dx.$ Integrating (1) between 0 and 1 we obtain

$$I_n = \int_0^{1} (1-x)P_{n-1}(x) dx + \frac{1}{n},$$

and using the symmetry of $P_n(x)$ about 1/2 we note that $\int_0^{1} (1-x)P_{n-1}(x) dx = \frac{1}{2}I_{n-1}.$ Hence we obtain the recurrence relation for $n > 1$ $$I_n = \frac{1}{2}I_{n-1} + \frac{1}{n}, \textrm{ where } I_1=1. \qquad (2)$$

From which it follows that if $\lim_{n \rightarrow \infty} n I_n$ exists then it must equal 2.

To show that the limit exists we can solve (2) to obtain

$$I_n = \sum_{k=1}^n \frac {1}{k2^{n-k}},$$

and straightforward algebra shows that $(n-1)I_{n-1} – nI_n > 0$ for sufficiently large $n$ (in fact, $n>6$), and since $nI_n$ is bounded below by 0 the result follows.

Just for completeness:

$$(n-1)I_{n-1} – nI_n = \left\lbrace \sum_{k=1}^{n-2} \frac{k}{2^k(n-k)(n-k-1)} \right\rbrace - \frac{n}{2^{n-1}} \quad \textrm{ for } n>2.$$

Comparing the first term in the summation with the negative term we have

$$ \frac{1}{2(n-1)(n-2)} > \frac{n}{2^{n-1}} \quad \textrm{ for } n \ge 13.$$

Hence $\lbrace nI_n \rbrace$ eventually forms a strictly decreasing sequence.

share|cite|improve this answer
Nice. Thanks... – Aryabhata Oct 13 '10 at 1:28

Here's a sketch of an elementary analytic demonstration.

(1) The integrals all exceed $2/n$ when $x$ exceeds 1: subtract $2/n = \int_0^1{(t^{n-1} + (1-t)^{n-1}) \ dt}$ to obtain a positive integrand (using elementary inequalities).

(2) The integrals are bounded by a sequence decreasing to the desired limit. By looking at the Taylor series of the log of the integrand near 1, for example, and assuming $n \ge 3$, you can easily establish an upper bound of $\exp((n-2)(x-1))$ in the interval $(1/2, 1]$, with a similar bound in the interval $[0, 1/2)$ by symmetry. The estimate (integrating over one of those two intervals) equals $(1 - \exp(2-n)) / (n-2)$ which, when doubled and multiplied by $n$, converges to $2$.

share|cite|improve this answer
Interesting. Is it possible for you to add a little more detail for part 2? It would make it easier for anyone (not just me) who reads your answer. – Aryabhata Oct 11 '10 at 20:26

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.