6
$\begingroup$

Consider the well-known sequence $C_k=\frac1{k+1}\binom{2k}k$ of Catalan numbers. I came across the below identity while working with certain generating functions. I thought it might be of interest to some folks here. Define the finite sum $$f_n:=\sum_{k\geq0}\binom{2n}{n-3k-2}.$$

QUESTION. Is this true? If so, is there a combinatorial proof to it? $$f_{n+2}-5f_{n+1}+4f_n=C_n.$$

$\endgroup$
6
  • 2
    $\begingroup$ Using hypergeometric functions we have $f_n = {\binom{2 n}{n + 2}} {{}_{4}F_{3}\left(\begin{matrix} 1, \frac{2}{3} - \frac{n}{3}, 1 - \frac{n}{3}, \frac{4}{3} - \frac{n}{3} \\ \frac{n}{3} + 1, \frac{n}{3} + \frac{4}{3}, \frac{n}{3} + \frac{5}{3} \end{matrix}\middle| {-1} \right)}$. It seems likely this identity could be proved using a WZ pair, but I don't know how to do that. $\endgroup$ Commented Dec 27, 2023 at 4:22
  • 1
    $\begingroup$ Have you tried throwing Gosper's algorithm/WZ at it? That may not give the combinatorial proof but it certainly should serve to prove the relation. $\endgroup$ Commented Dec 27, 2023 at 4:22
  • 2
    $\begingroup$ There are a bunch of other identities like this: setting $g_{a,b}(n)=\sum_{k\geq0}\binom{2n}{n-ak-b}$ we have: $4g_{1,0}(n)-g_{1,0}(n+1)=C_n,4g_{2,0}(n)-g_{2,0}(n+1)=C_n,-5g_{3,2}(n+1)+4g_{3,2}(n)+g_{3,2}(n+2) = C_n \text{(this is your identity)}, 6g_{4,1}(n+1)-8g_{4,1}(n)-g_{4,1}(n+2) = C_n, 4g_{5,2}(n)-13g_{5,2}(n+1)+7g_{5,2}(n+2)-g_{5,2}(n+3) = C_n$, and if there is an equation for $g_{a, b}$, there is one for $g_{a, a-b}$ with negated coefficients. For other $0\leq k<n\leq5$ there aren't any such equations, at least of small degree. $\endgroup$ Commented Dec 27, 2023 at 4:40
  • $\begingroup$ @CommandMaster :yes, it seems doable but I'm looking for a combinatorial approach. $\endgroup$ Commented Dec 27, 2023 at 4:41
  • 1
    $\begingroup$ For $g_{1, 0}$ (A032443) we have $g_{1, 0}(n) = \frac12 (4^n + \binom{2n}{n})$, so $4g_{1,0}(n) - g_{1,0}(n+1) = 2\binom{2n}{n} - \frac12 \binom{2n+2}{n+1}$. Now $\binom{2n+2}{n+1} = \binom{2n}{n+1} + 2\binom{2n}{n} + \binom{2n}{n-1} = 2(\binom{2n}{n+1} + \binom{2n}{n})$, so $4g_{1,0}(n) - g_{1,0}(n+1) = \binom{2n}{n} - \binom{2n}{n+1} = C_n$. This might be generalizable by looking at the binomial expansion of $(1 + \zeta)^{2n}$, where $\zeta$ is an $a$-th root of unity. $\endgroup$ Commented Dec 27, 2023 at 5:03

2 Answers 2

9
$\begingroup$

We have $$f_{n+1} = \sum_{k\geq 0}\binom{2n+2}{n-3k-1} = \sum_{k\geq 0}\left(\binom{2n}{n-3k-3}+2\binom{2n}{n-3k-2}+\binom{2n}{n-3k-1}\right)$$ $$f_{n+2} = \sum_{k \geq 0} \binom{2n+4}{n-3k} = \sum_{k \geq 0} \left(\binom{2n}{n-3k-4}+4\binom{2n}{n-3k-3}+6\binom{2n}{n-3k-2} + 4\binom{2n}{n-3k-1} + \binom{2n}{n-3k}\right)$$ So $$f_{n+2} + 4f_n - 5f_{n+1} = \sum_{k \geq 0} \left(\binom{2n}{n-3k-4}- \binom{2n}{n-3k-3} - \binom{2n}{n-3k-1} + \binom{2n}{n-3k}\right)$$ and it can be seen that this sum telescopes to $\binom{2n}{n} - \binom{2n}{n-1} = C_n$.

$\endgroup$
1
  • $\begingroup$ This is a nice elementary approach. Thanks. $\endgroup$ Commented Dec 27, 2023 at 21:06
7
$\begingroup$

Here's a sketch of a generating function proof. Recall that \begin{equation*} \sum_{m=0}^\infty \binom{2m+k}{m} x^m =\frac{c(x)^k}{\sqrt{1-4x}}, \end{equation*} where $c(x) = \sum_{n=0}^\infty C_n x^n$. Then \begin{align*} \sum_{n=0}^\infty f_n x^n &= \sum_{n,k\ge0}\binom{2n}{n-3k-2} x^n \\ &=\sum_{m,k\ge0}\binom{2m+6k+4}{m}x^{m+3k+2}\\ &=\sum_{k\ge0}x^{3k+2}\frac{c(x)^{6k+4}}{\sqrt{1-4x}}\\ &=\frac{x^2c(x)^4}{\sqrt{1-4x}(1-x^3 c(x)^6)} \end{align*} Simplifying (with the help of Maple) yields $$\frac{x^2 c(x)}{1-5x+4x^2}$$ from which the recurrence follows.

Update: Here is an explanation and generalization for the mysterious simplification described above. Let $\def\sq{\sqrt{1-4x}}$ \begin{align} p_n(x) &=\frac 1{\sq}\left[\left(\frac{1+\sq}{2}\right)^n\ -\left(\frac{1-\sq}{2}\right)^n\,\right]\tag{1}\\ &= \frac{c(x)^{-n}-(xc(x))^n}{\sq} = c(x)^{-n}\frac{1-x^nc(x)^{2n}}{\sq}.\tag{2} \end{align} By expanding $(1)$ in powers of $\sq$, we see that for $n>0$, $p_n(x)$ is a polynomial in $x$ of degree $\lfloor (n-1)/2\rfloor$. It can be shown that for $n>0$, $$p_n(x) =\sum_{i=0}^{\lfloor(n-1)/2\rfloor}(-1)^i \binom {n-i-1}{i}x^i.$$ We can express $p_n(x)$ for $n>0$ in terms of the Chebyshev polynomial of the second kind $U_n(x)$ by $$ p_n(x) = (-1)^{n-1}x^{(n-1)/2}U_{n-1}\left(-\frac{1}{2\sqrt x}\right). $$ It follows from $(2)$ that $$\frac{1}{p_n(x)} = \sq \frac{c(x)^n}{1-x^nc(x)^{2n}}$$ so \begin{equation} \frac{1}{(1-4x)p_n(x)} = \frac{c(x)^n}{\sq(1-x^nc(x)^{2n})}.\tag{3} \end{equation} Since $p_3(x) = 1-x$ and $(1-4x)p_3(x) = 1-5x+4x^2$, the case $n=3$ of $(3)$ is \begin{equation*} \frac{1}{1-5x+4x^2}=\frac{c(x)^3}{\sq(1-x^3 c(x)^6)}. \end{equation*}

$\endgroup$
4
  • $\begingroup$ This is a nice and classical set of tools. Thanks. $\endgroup$ Commented Dec 27, 2023 at 21:07
  • $\begingroup$ It might be slightly easier, though of course less general, to simplify as follows. Let $C=C(x)=1+xC^2$, $B=B(x)=\dfrac{1}{\sqrt{1-4x}}$, then $B=\dfrac{C}{1-xC^2}$ (straightforward to see bijectively) and $1+xC^2+(xC^2)^2=(1+xC^2)^2-xC^2=(1-x)C^2$, so that $$\frac{x^2BC^4}{1-(xC^2)^3}=x^2B\frac{C}{1-xC^2}\frac{C^3}{1+xC^2+(xC^2)^2}=\frac{x^2B^2C^3}{(1-x)C^2}=\frac{x^2B^2C}{1-x}=\frac{x^2C}{(1-x)(1-4x)}.$$ $\endgroup$ Commented Dec 28, 2023 at 20:58
  • $\begingroup$ It's a nice comment. I wish you write it as an "answer" if you are able to provide the bijective claim you noted. This would help readers instead of your ideas being buried in the comment box. $\endgroup$ Commented Dec 29, 2023 at 19:38
  • $\begingroup$ I'm not sure it's worth a separate answer, since it only expands on a part of Ira's answer. The bijective proof that $B=\dfrac{C}{1-xC^2}=1+2\dfrac{xC^2}{1-xC^2}$ is well-known: $B$ enumerates grand Dyck paths, which can be partitioned into a possibly sequence of nonempty Dyck paths ($xC^2$) above and below the $x$-axis. Such a path is either empty ($1$) or goes "above-below-above-below-..." or "below-above-below-above-..." the $x$-axis. $\endgroup$ Commented Dec 30, 2023 at 5:35

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .