# Find $\lim\limits_{{n \to \infty}} \frac1{2^n} \sum\limits_{k=1}^n \frac1{\sqrt{k}} \binom nk$

Find $$\lim_{n \to \infty} \frac {1}{2^n} \sum_{k=1}^n \frac{1}{\sqrt{k}} \binom{n}{k}.$$

First time I thought about Stirling's approximation but didn't get anything by applying it. I would also think about a Riemann Sum, but no idea how to rewrite...

The answer is $0$.

Let $a_k = k^{-1/2}$. Notice that $(a_k)$ decreases to $0$. Then for each fixed $m \geq 1$ and for all $n \geq m$,

$$\frac{1}{2^n} \sum_{k=1}^{n} \binom{n}{k} a_k \leq \frac{1}{2^n} \underbrace{\sum_{k=1}^{m} \binom{n}{k} (a_k - a_m)}_{= \mathcal{O}(n^m)} + \frac{1}{2^n} \underbrace{\sum_{k=1}^{n} \binom{n}{k} a_m}_{=(2^n - 1)a_m}.$$

Taking limsup as $n\to\infty$, it follows that

$$\limsup_{n\to\infty} \frac{1}{2^n} \sum_{k=1}^{n} \binom{n}{k} a_k \leq a_m$$

Since $a_m \to 0$ as $m\to\infty$, this proves

$$\lim_{n\to\infty} \frac{1}{2^n} \sum_{k=1}^{n} \binom{n}{k} a_k = 0.$$

Addendum. I just saw that OP is a high-school student. Here is a little tweak of the argument above that does not use any fancy analysis stuffs.

Let $m_n = \lfloor \log n \rfloor$. Then for $n \geq 3$, we always have $1 \leq m_n \leq n$. Then

\begin{align*} 0 \leq \frac{1}{2^n} \sum_{k=1}^{n} \binom{n}{k} \frac{1}{\sqrt{k}} &= \frac{1}{2^n} \sum_{k=1}^{m_n} \binom{n}{k} \frac{1}{\sqrt{k}} + \frac{1}{2^n} \sum_{k=m_n + 1}^{n} \binom{n}{k} \frac{1}{\sqrt{k}} \\ &\leq \frac{1}{2^n} \sum_{k=1}^{m_n} n^k + \frac{1}{2^n} \sum_{k=m_n + 1}^{n} \binom{n}{k} \frac{1}{\sqrt{m_n}} \tag{1} \\ &\leq \frac{n^{1+m_n}}{2^n} + \frac{1}{\sqrt{m_n}}. \tag{2} \end{align*}

For $\text{(1)}$ I utilized the fact that $\binom{n}{k} \leq n^k$ and $\frac{1}{\sqrt{k}}$ is decreasing. For $\text{(2)}$ I utilized the geometric sum formula and the identity $\sum_{k=0}^{n} \binom{n}{k} = 2^n$.

Now taking $n\to\infty$ and applying the squeezing lemma proves the claim.

• Both your arguments are easy to understand. But +1 for putting extra effort to match the level of OP. – Paramanand Singh Apr 23 '17 at 8:30

It is enough to apply Laplace's method.

Through the inverse Laplace transform we have $\frac{1}{\sqrt{k}}=\int_{0}^{+\infty}\frac{e^{-ks}}{\sqrt{\pi s}}\,ds$, hence

$$\sum_{k=1}^{n}\binom{n}{k}\frac{1}{\sqrt{k}} = \int_{0}^{+\infty}\frac{-1+(1+e^{-s})^n}{\sqrt{\pi s}}\,ds =\frac{2}{\sqrt{\pi}}\int_{0}^{+\infty}\left[(1+e^{-s^2})^n-1\right]\,ds$$ where the last integrand function behaves like $(2^n-1) e^{-ns^2/2}$.
It follows that $$\sum_{k=1}^{n}\binom{n}{k}\frac{1}{\sqrt{k}} \approx (2^n-1)\sqrt{\frac{2}{n}}$$ and the wanted limit is simply zero.

Using Abel's summation and the fact that $\sum \limits_{k=1}^{n} \binom{n}{k} = 2^{n}-1$ you can solve the above question.

From the Abel's summation we can say that $\sum \limits_{k=1}^{n} \frac{\binom{n}{k}}{\sqrt{k}} \approx \frac{\sum \limits_{k=1}^{n} \binom{n}{k}}{\sqrt{n}}-\int \limits_{1}^{n} ( \sum \limits_{k=1}^{t} \binom{t}{k} \frac{d}{dx}(\sqrt{t})) dt$

By substituting we arrive at : $\frac{2^{n}-1}{\sqrt{n}}-\int \limits_{1}^{n} ( 2^{t}-1) \frac{d}{dx}(\sqrt{t}) dt$

It easy by comparison tests to bound the integral part by $\frac{2^n}{\sqrt{n}}$.

So its at most between $0$ and $2\frac{2^n}{\sqrt{n}}$ which both approach $0$ when divided by $2^n$ when $n \to \infty$.

• It might be a good answer but... I don't really understand all these things. I'm in high school. – Liviu Apr 22 '17 at 20:54


In line \eqref{1}, I used the Laplace Method to 'arrive' to line \eqref{2}.

• Why are the last two lines equal? – πr8 Apr 23 '17 at 12:24
• @πr8 That's an application of 'Laplace Method'. Thanks for your remark. – Felix Marin Apr 24 '17 at 0:51
• It's not clear to me how that justifies replacing $(1+e^{-t^2})^n-1$ with the other exponential? – πr8 Apr 24 '17 at 0:56
• $\ln\left(\left[1 + \,\mathrm{e}^{-t^{2}}\right]^{n} - 1\right) = \ln\left(2^{n} - 1\right) - {2^{n - 1} \over 2^{n} - 1}\,nt^{2} + \,\mathrm{O}\left(t^{4}\right)$. Please, visit the Laplace Method link. – Felix Marin Apr 24 '17 at 1:04
• Thanks - I'd think of that step as basically just being a Taylor expansion (Laplace's method to me is the step of actually giving the integral asymptotics) but I suppose this is mostly semantics. – πr8 Apr 24 '17 at 9:04

The limit is part of a broad class of limits for which the classical analysis designed Toeplitz theorem. The theorem says that given an array of real numbers, $\lambda_{n,k}: 1\le k\le n,\ n\ge1$, such that: $i) \ \lim_{n\to\infty}\lambda_{n,k}=0, \ \forall k\in \mathbb{N}$, $ii) \lim_{n\to\infty} \sum_{k=1}^n \lambda_{n,k}=1$ and $iii)$ there exists $C>0$ such that for all positive integers $n$, $\sum_{k=1}^n |\lambda_{n,k}|\le C$, then for any convergent sequence $\delta_k$, we have

$$\lim_{n\to\infty} \sum_{k=1}^n \lambda_{n,k} \delta_k=\lim_{n\to\infty}\delta_n .$$

In your problem set $\displaystyle \lambda_{n,k}=\frac{1}{2^n}\binom{n}{k}$ and $\displaystyle \delta_k=\frac{1}{\sqrt{k}}$, and since the conditions are satisfied, we have

$$\displaystyle \lim_{n \to \infty} \frac {1}{2^n} \displaystyle \sum_{k=1}^n \frac{\displaystyle \binom{n}{k}}{\sqrt{k}}=\lim_{n\to\infty} \frac{1}{\sqrt{n}}=0,$$ which ends the little proof.

$$\lim_{n \to \infty} \frac {1}{2^n} \sum_{k=1}^n \frac{{n}\choose{k}}{\sqrt{k}}=0$$

I offer a sketch proof. The basic idea is as follows: for large $n$, the small values of $k$ will make small contributions, because $\frac{{n \choose k}}{2^n}$ will be small there, and the large values of $k$ will make small contributions because $\sqrt{\frac{1}{k}}$ will be small there. Thus, the whole sum should be small too.

• Write $f(x)=x^{-1/2}\mathbb{I}(x\ge 1)$, and let $X$ denote a binomial random variable with parameter $1/2$ and $n$ trials.
• Then, the term we are taking limits of is $\mathbb{E}f(X)$
• The Central Limit Theorem says that, for large $n$, we have that $X\approx N(n/2,n/4)$ as distributions.
• We can thus (handwaving a bit) deduce that $$\frac{{n}\choose{k}}{2^n} \approx\sqrt{\frac{2}{\pi n}}\exp\left(-\frac{(2k-n)^2}{2n}\right)$$
• One could also establish this via Stirling's formula, or likely other methods, if need be.
• Take some $\epsilon\in(0,1/3)$, and split the sum up into $1\le k\le (1-\epsilon)\frac{n}{2}, k>(1-\epsilon)\frac{n}{2}$.
• In the first case, we have $$(2k-n)^2 \geq \epsilon^2n^2 \implies \exp\left(-\frac{(2k-n)^2}{2n}\right) \le \exp\left(-\frac{n\epsilon^2}{2}\right).$$ Thus, approximately, $$\frac {1}{2^n} \sum_{k=1}^{(1-\epsilon)\frac{n}{2}} \frac{{n}\choose{k}}{\sqrt{k}} \le \sum_{k=1}^{(1-\epsilon)\frac{n}{2}}\sqrt{\frac{2}{\pi n k}}\exp\left(-\frac{n\epsilon^2}{2}\right) \le(1-\epsilon)\sqrt{\frac{n}{2\pi}}\exp\left(-\frac{n\epsilon^2}{2}\right)\to 0$$
• For the remainder of the sum, we have $$\frac {1}{2^n} \sum_{k>(1-\epsilon)n/2} \frac{{n}\choose{k}}{\sqrt{k}} \le \sum_{k>(1-\epsilon)n/2} \frac{{n}\choose{k}}{2^n}\sqrt{\frac{2}{{(1-\epsilon)n}}} \le \sum_{k\ge0} \frac{{n}\choose{k}}{2^n}\sqrt{\frac{2}{{(1-\epsilon)n}}}=\sqrt{\frac{2}{{(1-\epsilon)n}}}\to 0.$$

Thus, as desired, the limit is $0$.

• Finally a good correct answer for me to upvote lol – Simply Beautiful Art Apr 22 '17 at 22:56