Looking for a short proof of a harmless looking binomial identity

binomial-coefficientssummation

I managed to prove for this MSE post the rather harmless looking binomial identity for natural $1\leq k\leq n$:
\begin{align*}
\color{blue}{\sum_{j=0}^k\binom{2n}{2j}\binom{n-j}{k-j}=\binom{n+k}{n-k}\frac{4^kn}{n+k}}\tag{1}
\end{align*}

using the coefficient of operator method. Admittedly, there are a lot of intermediate steps used to show the validity of (1).

Question: I'm wondering if there is a more direct, less lengthy derivation than the one I've provided below.

We obtain for $1\leq k\leq n$:
\begin{align*}
\color{blue}{\sum_{j=0}^k}&\color{blue}{\binom{2n}{2j}\binom{n-j}{k-j}}\\
&=\sum_{j=0}^n\binom{2n}{2j}\binom{n-j}{n-k}\tag{2}\\
&=\sum_{j=0}^n\binom{2n}{2j}[z^{n-k}](1+z)^{n-j}\tag{3}\\
&=[z^{n-k}](1+z)^n\sum_{j=0}^n\binom{2n}{2j}\frac{1}{(1+z)^j}\\
&=\frac{1}{2}[z^{n-k}](1+z)^n\left(\left(1+\frac{1}{\sqrt{1+z}}\right)^{2n}+\left(1-\frac{1}{\sqrt{1+z}}\right)^{2n}\right)\\
&=\frac{1}{2}[z^{n-k}]\left(\left(1+\sqrt{1+z}\right)^{2n}+\left(1-\sqrt{1+z}\right)^{2n}\right)\\
&=\frac{1}{2}[z^{n-k}]\left(1+\sqrt{1+z}\right)^{2n}\tag{4}\\
&=\frac{1}{2}[z^{-1}]z^{-n+k-1}\left(1+\sqrt{1+z}\right)^{2n}\tag{5}\\
&=\frac{1}{2}[w^{-1}]\left(w^2-1\right)^{n-k-1}(1+w)^{2n}2w\tag{6}\\
&=[w^{-1}]w(w-1)^{-n+k-1}(w+1)^{n+k-1}\\
&=[u^{-1}](u+1)u^{-n+k-1}(u+2)^{n+k-1}\tag{7}\\
&=\left([u^{n-k}]+[u^{n-k-1}]\right)\sum_{j=0}^{n+k-1}\binom{n+k-1}{j}u^j2^{n+k+1-j}\\
&=\binom{n+k-1}{n-k}2^{2k-1}+\binom{n+k-1}{n-k-1}2^{2k}\tag{8}\\
&=\binom{n+k}{n-k}\frac{2k}{n+k}2^{2k-1}+\binom{n+k}{n-k}\frac{n-k}{n+k}2^{2k}\tag{9}\\
&\,\,\color{blue}{=\binom{n+k}{n-k}\frac{4^kn}{n+k}}
\end{align*}

and the claim follows.

Comment:

  • In (2) we use the binomial identity $\binom{p}{q}=\binom{p}{p-q}$. We also set the upper index to $n$ without changing anything, since we are adding zeros only.

  • In (3) we use the coefficient of operator method.

  • In (4) we skip $\left(1-\sqrt{1+z}\right)^{2n}=cz^{2n}+\cdots$ since it has only powers of $z$ greater than $n$ and does not contribute to $[z^{n-k}]$.

  • In (5) we apply the rule $[z^{p-q}]A(z)=[z^p]z^qA(z)$.

  • In (6) we use the transformation of variable formula $[z^{-1}]f(z)=[w^{-1}]f(g(w))g^\prime(w)$ with $1+z=w^2, \frac{dz}{dw}=2w$.

  • In (7) we use the transformation of variable formula again, with $w-1=u, \frac{dw}{du}=1$.

  • In (8) we select the coefficients accordingly.

  • In (9) we use the binomial identities $\binom{p-1}{q}=\binom{p}{q}\frac{p-q}{p}$ and $\binom{p}{q}=\binom{p-1}{q-1}\frac{p}{q}$.

Best Answer

The result follows from the equality of two different expressions for the Chebyshev polynomials of the first kind. We have $$ \begin{aligned} T_N(x)&=\sum_{j\ge0}\binom{N}{2j}(x^2-1)^j x^{N-2j}\\ &=\frac{1}{2}\sum_{r\ge0}(-1)^r\frac{N}{N-r}\binom{N-r}{r}(2x)^{N-2r}, \end{aligned} $$ where the first equality holds for $N\ge0$ and the second for $N\ge1$. Expanding the binomial factor in the first expression gives $$ \begin{aligned} &\sum_{j\ge0}\binom{N}{2j}\sum_{r=0}^j\binom{j}{r}(-1)^r x^{2j-2r}x^{N-2j}\\ &\quad=\sum_{r\ge0}(-1)^rx^{N-2r}\sum_{j\ge r}\binom{N}{2j}\binom{j}{k}\\ &\quad=\sum_{r\ge0}(-1)^rx^{N-2r}\sum_{j\ge 0}\binom{N}{2r+2j}\binom{r+j}{r}. \end{aligned} $$ Comparing coefficients yields $$ \sum_{j\ge 0}\binom{N}{2r+2j}\binom{r+j}{r}=\frac{1}{2}\frac{N}{N-r}\binom{N-r}{r}2^{N-2r} $$ Setting $N=2n$ and $r=n-k$ gives your identity.

Of course for this to be a proof, we really have to prove that the two expressions for $T_N(x)$ hold. We define $T_N(x)$ by the condition $\cos(N\theta)=T_N(\cos\theta)$. The first expression for $T_N(x)$ follows by taking the real parts of $$ \cos(N\theta)+i\sin(N\theta)=e^{iN\theta}=\sum_{k=0}^N\binom{N}{k}(i\sin(\theta))^k\cos^{N-k}\theta $$ and recognizing that $(i\sin\theta)^{2j}=(\cos^2\theta-1)^j$.

The factor $\frac{N}{N-r}\binom{N-r}{r}$ in the second expression is the number of $r$-matchings on $C_N$, the cycle graph of $N$ vertices. Equivalently, it's the number of ways of placing $r$ non-overlapping dominoes on the edges of an $N$-gon. What does this have to do with expressing $\cos(N\theta)$ as a polynomial in $\cos\theta$? The idea is to add powers of $2\cos\theta=(e^{i\theta}+e^{-i\theta})$ up to the $N^\text{th}$ power, with coefficients chosen so the only the $e^{iN\theta}$ and $e^{-iN\theta}$ terms survive, and then multiply by $\frac{1}{2}$ to get $\cos(N\theta)$. To eliminate the unwanted terms, we use the principle of inclusion-exclusion, as follows. Represent a term in the expansion of $(e^{i\theta}+e^{-i\theta})^N$ by the sequence of signs in the exponent. So the term $e^{i\theta}e^{i\theta}e^{-i\theta}e^{i\theta}$ in the expansion of $(e^{i\theta}+e^{-i\theta})^4$ would be represented by the sign sequence $++-+$. We want to keep the terms $+++\ldots+$ and $---\ldots-$, and discard everything else. Define $S_j$ to the the set of sequences in which a plus at position $j$ is followed by a minus at position $j+1$, where $j$ ranges from $0$ to $N-1$ and $j+1$ is computed $\mod N$ (so that the sequence is considered to be wrapped on a circle). Since the terms $e^{i\theta}$ and $e^{-i\theta}$ at positions $j$ and $j+1$ cancel, the sum of the terms corresponding to sequences in $S_j$ is $(e^{i\theta}+e^{-i\theta})^{N-2}$. So, from $(e^{i\theta}+e^{-i\theta})^N$, we subtract, for each $j$, the quantity $(e^{i\theta}+e^{-i\theta})^{N-2}$. But if a term has a sequence in which $+$ is immediately followed by $-$ at two different positions, say $j$ and $k$, that term will have been subtracted twice, and therefore needs to be added back in. This requires adding $(e^{i\theta}+e^{-i\theta})^{N-4}$ for every such pair $j$, $k$. By the principle of inclusion--exclusion, we continue in this way, alternately adding and subtracting the terms $(e^{i\theta}+e^{-i\theta})^{N-2r}$ corresponding to sequences in $S_{j_1}\cap S_{j_2} \cap S_{j_3}\cap\ldots\cap S_{j_r}$. It only remains to determine how many non-empty intersections there are of $r$ sets. There is only one condition we need to worry about: if $+$ at $j$ is followed by $-$ at $j+1$, then it certainly is not the case that $+$ is followed by $-$ at positions $j+1$ and $j+2$, so any intersection containing $S_j\cap S_{j+1}$ is empty. This is precisely the non-overlapping domino condition, and the second expression for $T_N(x)$ follows.