# Looking for a proof of an interesting identity

Working on a problem I have encountered an interesting identity:

$$\sum_{k=0}^\infty \left(\frac{x}{2}\right)^{n+2k}\binom{n+2k}{k} =\frac{1}{\sqrt{1-x^2}}\left(\frac{1-\sqrt{1-x^2}}{x}\right)^n,$$ where $n$ is a non-negative integer number and $x$ is a real number with absolute value less than 1 (probably a similar expression is valid for arbitrary complex numbers $|z|<1$).

Is there any simple proof of this identity?

• How/where did you meet this identity? Is the LHS just the Taylor series for the RHS? – Daniel Robert-Nicoud Dec 14 '17 at 23:38
• Anything to do with a random walk? Why did you call this interesting, btw? – Mathemagical Dec 14 '17 at 23:44
• Reminiscent of the explicit form for Chebyshev polynomials of the first kind, hence they idea of enforcing the substitution $x=\sin\theta$ and applying the residue theorem looks like a promising one ;) – Jack D'Aurizio Dec 14 '17 at 23:56
• Or you may prove that both sides are the terms of an Appell / Fibonacci sequence of polynomials. – Jack D'Aurizio Dec 14 '17 at 23:57
• @tired This is in regard to the hitting time of a random walk (more particularly, the probability generating function of the hitting time of the walk at the barrier x=1. Please see the equations 15,16, and 17 in the linked document. galton.uchicago.edu/~lalley/Courses/312/RW.pdf The details of the computation are not listed here, but they are shown in greater detail in Michael Steele's book, Stochastic Calculus and Financial applications (pages 8 and 9). – Mathemagical Dec 16 '17 at 15:25

## 3 Answers

Using

$$\binom{n}{k}=\frac{1}{2 \pi i}\oint_C\frac{(1+z)^{n}}{z^{k+1}}dz$$ we get (integration contour is the unit cicrle)

$$2\pi iS_n=\oint dz \sum_{k=0}^{\infty}\frac{(1+z)^{n+2k}x^{n+2k}}{z^{k+1}2^{n+2k}}=\oint dz \frac{(1+z)^n x^n}{z2^n}\sum_{k=0}^{\infty}\frac{(1+z)^{2k}x^{2k}}{2^{2k}z^k}=\\ 4\frac{x^n}{2^n}\oint dz \underbrace{\frac{(1+z)^n}{4z-(1+z)^2x^2}}_{f(z)}$$

for $|x|<1$ only we have just one pole of $f(z)$ inside the unit circle namely $z_0(x)=\frac2{x^2}-\frac{2\sqrt{1-x^2}}{x^2}-1$ , so

$$S_n=4\frac{x^n}{2^n}\text{res}(f(z),z=z_0(x))=4\frac{x^n}{2^n}\left[ \frac{1}{4 \sqrt{1-x^2}}\left(2\frac{1-\sqrt{1-x^2}}{ x^2}\right)^n\right]$$

or

$$S_n=\frac{1}{\sqrt{1-x^2}}\left(\frac{1-\sqrt{1-x^2}}{ x}\right)^n$$

• Thanks Jack :-) – tired Dec 15 '17 at 0:55
• @tired: Thank you very much for the nice proof. It contains however a misprint. $4$ shall be replaced with $4^{1/n}$ in the denominator of the residue value. But probably simpler and more correctly were to write it just as $$\text{res}(f(z),z=z_0)=\frac{1}{4\sqrt{1-x^2}} \left(2\frac{1-\sqrt{1-x^2}}{x^2}\right)^n.$$ – user Dec 15 '17 at 11:56
• @user355705 you are absolutly right – tired Dec 16 '17 at 15:07

Extracting coefficients on the RHS we get the integral (coefficient on $x^{n+2k}$)

$$\frac{1}{2\pi i} \int_{|z|=\epsilon} \frac{1}{z^{n+2k+1}} \frac{1}{\sqrt{1-z^2}} \left(\frac{1-\sqrt{1-z^2}}{z}\right)^n \; dz.$$

Now we put $(1-\sqrt{1-z^2})/z = w$ so that $z = 2w/(1+w^2).$ This has $w = \frac{1}{2} z + \cdots$ so the image in $w$ of the contour in $z$ can be deformed to a small circle enclosing the origin in the $w$-plane. (Moreover we see that the exponentiated term starts at $z^n$ which justifies the corresponding offset in the series.) We get $dz = 2/(1+w^2) - 4w^2/(1+w^2)^2 \; dw = 2(1-w^2)/(1+w^2)^2 \; dw.$ We also have $1-z^2 = 1 - 4w^2/(1+w^2)^2 = (1-w^2)^2/(1+w^2)^2.$ All of this yields

$$\frac{1}{2\pi i} \int_{|w|=\gamma} \frac{(1+w^2)^{n+2k+1}}{2^{n+2k+1} w^{n+2k+1}} \frac{1}{(1-w^2)/(1+w^2)} w^n \frac{2(1-w^2)}{(1+w^2)^2} \; dw \\ = \frac{1}{2^{n+2k}} \frac{1}{2\pi i} \int_{|w|=\gamma} \frac{(1+w^2)^{n+2k}}{w^{2k+1}} \; dw.$$

This evaluates by inspection to

$$\frac{1}{2^{n+2k}} [w^{2k}] (1+w^2)^{n+2k} = \frac{1}{2^{n+2k}} [w^{k}] (1+w)^{n+2k} \\ = \frac{1}{2^{n+2k}} {n+2k\choose k}$$

which is the claim.

I plan to insert this approach into the next (2019) version of my notes.
The even powers of $\arcsin(x)$ all have a nice Maclaurin series (see eqs (19),(20),(21),(22)), and this can be seen as a side effect of the Lagrange–Bürmann inversion theorem. I am going to implement the same trick here, by computing the derivatives at the origin of the RHS (the opposite way has already been investigated by tired).

$$[x^m]\left[\frac{1}{\sqrt{1-x^2}}\left(\frac{1-\sqrt{1-x^2}}{x}\right)^n\right]=\frac{1}{2\pi i}\oint\frac{(1-\sqrt{1-x^2})^n}{x^{n+m+1}\sqrt{1-x^2}}\,dx$$ where the integral is performed along a small circle enclosing the origin. What about enforcing the substitution $x=\sin z$? The sine function is holomorphic and injective in a neighbourhood of the origin, hence the RHS is simply converted into $$\frac{1}{2\pi i}\oint\frac{(1-\cos z)^n}{\left(\sin z\right)^{n+m+1}}\,dz =\frac{1}{2\pi i}\oint\frac{2^n \sin^{2n}\frac{z}{2}}{2^{n+m+1}\sin^{n+m+1}\frac{z}{2}\cos^{n+m+1}\frac{z}{2}}\,dz$$ or $$\frac{1}{2\pi i}\oint\frac{\tan^{n-m-1}z}{2^{m}\cos^{2m}z}\,dz=\frac{1}{2^{m+1}\pi i}\oint u^{n-m-1}(1+u^2)^{m-1}\,du$$ via $z\to 2z$ and $z\to\arctan u$. Now the RHS is trivially given by a binomial coefficient multiplied by $\frac{1}{2^m}$ (provided by $m\geq n$) and the claim is proved.