8
$\begingroup$

As part of a proof in finite group theory, I'm looking for a closed form for the expression

$$\sum_{i=j+1}^{n} \binom{\binom{i}{j}}{2}.$$

Any help — especially with reference or proof — would be appreciated. In the group theory context, there is strong evidence that the overall result is correct, but having a closed form here would be the most obvious way to construct a proof.

I don't know if a closed form exists, but computation of special cases suggests the sum is equal to

$$\frac{(n-j)(n-(j-1))\dotsm(n-1)n(n+1)p_j(n)}{(2j+1)!}$$

where $p_j(n)$ is a polynomial with integer coefficients of degree $j-1$. In fact, the leading coefficient of $p_j$ appears to be $\binom{2j-1}{j-1}$. In general $p_j$ is not irreducible.

I really just need to know that factor of $(n+1)$ is present, but to prove that by induction on $n$ it seems I would need a precise formula for $p_j$. Since I can compute $p_j$ for given $j$, I know this formula is correct for $j=1$, $2$, $3$, and $4$.

$\endgroup$
2
  • 2
    $\begingroup$ Why do you think there is a closed form? $\endgroup$
    – Igor Rivin
    Commented Jun 1, 2015 at 20:04
  • $\begingroup$ Why was this question closed as "off-topic"? $\endgroup$ Commented Jun 3, 2015 at 7:21

5 Answers 5

7
$\begingroup$

A closed form is \begin{equation*} \sum_{i=j+1}^{n} \binom{\binom{i}{j}}{2} =\sum_{k=1}^j \frac{1}{2}\binom{j}{k}\binom{j+k}{k}\binom{n+1}{j+k+1}.\tag{1}\label{474985_1} \end{equation*} For fixed $j$ this is indeed a closed form; for example, for $j=3$ it gives \begin{equation*} \sum_{i=4}^n \binom{\binom{i}{3}}{2}=6\binom{n+1}{5}+15\binom{n+1}{6}+10\binom{n+1}{7}. \end{equation*} Also, each term in \eqref{474985_1}, as a polynomial in $n$, is divisible by $(n+1)n(n-1)\dots (n-j)$.

First note that we may start the sum on the left side of \eqref{474985_1} at $i=0$ instead of $i=j+1$. Then by the well-known identity $\sum_{i=0}^n \binom{i}{k} = \binom{n+1}{k+1}$, \eqref{474985_1} will follow from \begin{equation*} \binom{\binom nj}{2} = \sum_{k=1}^j \frac12\binom jk \binom{j+k}{k}\binom{n}{j+k}. \tag{2}\label{474985_2} \end{equation*}

To prove \eqref{474985_2} we have $\binom{j+k}{k}\binom{n}{j+k} = \binom{n}{j}\binom{n-j}{n-j-k}$, so \begin{align*} \sum_{k=1}^j \frac12\binom jk \binom{j+k}{k}\binom{n}{j+k} &=\frac12\binom{n}{j}\sum_{k=1}^j \binom {j}{k}\binom{n-j}{n-j-k} \\ &=\frac12\binom{n}{j}\biggl[\sum_{k=0}^j \binom {j}{k}\binom{n-j}{n-j-k} -1\biggr] \\ &=\frac12\binom{n}{j}\left[\binom{n}{n-j} -1\right]=\binom{\binom nj}{2} \end{align*} by Vandermonde's theorem.

$\endgroup$
4
  • 1
    $\begingroup$ (2) also admits a bijective proof: The left hand side, times $2$, counts ordered pairs of distinct $j$-element subsets of $\{1,\dots,n\}$. The union of an pair of $j$-element subsets is a $j+k$ element subset for some $ k \leq j$, with $k>0$ if the subsets are distinct, and the additional data needed to recover the two $j$-element subsets is the $k$-element complement of one and then, inside the one, the $k$-element complement of the other. $\endgroup$
    – Will Sawin
    Commented Jul 14 at 13:45
  • $\begingroup$ (2) is also equivalent to $A_{t,2,k}=\frac{1}{2} {t \choose k-t}{k \choose k-t}, t<k \le 2t$ which is the number of t-uniform weighted hypergraphs on 2 edges and k vertices.mathoverflow.net/q/468413. Is there a similar formula for ${ {n \choose j} \choose d}$ for $d \ge 3$ ? It will give an explicit formula for $A_{t,d,k}$ hypergraphs on $d$ edges and k vertices. $\endgroup$
    – CHUAKS
    Commented Jul 14 at 17:28
  • $\begingroup$ The numbers $\frac12\binom jk \binom{j+k}{k}$ are sequence A092370 in the OEIS (oeis.org/A092370). That's how I found the formula. $\endgroup$
    – Ira Gessel
    Commented Jul 14 at 18:28
  • $\begingroup$ @IraGessel See my recent answer. $\endgroup$
    – Fred Hucht
    Commented Jul 14 at 22:34
3
$\begingroup$

Equation \eqref{474985_1} from Ira Gessel's nice answer can be summed using Mathematica 14, and the result can be simplified to \begin{align} \sum_{i=j+1}^{n} \binom{\binom{i}{j}}{2} &=\sum_{k=1}^j \frac{1}{2}\binom{j}{k}\binom{j+k}{k}\binom{n+1}{j+k+1}\\ &=\frac 1 2 \binom{n+1}{j+1}\left[\binom{n}{j} -1+\frac{j(n-j)}{j+2} \,{}_3F_2 \left(\begin{matrix} 1-j,1-(n-j),j+2\\2,j+3 \end{matrix};1 \right)\right]. \end{align}

$\endgroup$
1
$\begingroup$

$$\sum_{i=j+1}^n {i \choose j}$$ has a simple closed form.

It appears that $$ \sum_{i=j+1}^n {i \choose j}^2 = \lim_{x \to 1-} \left(\left( j+1 \right) ^{2}{x}^{j+1} {\mbox{$_3$F$_2$}(1,j+2,j+2;\,2,2;\,x)}- {n+1\choose j} ^{2}{x}^{n+1}{\mbox{$_3$F$_2$}(1,2+n,2+n;\,2+n-j,2+n-j;\,x)} \right)$$

$\endgroup$
1
$\begingroup$

It is not hard to show that the factor $n+1$ is present.

Theorem. Let $p(i)$ be a polynomial. Then there exists a polynomial $P(n)$ such that for $i\ge0$ we have $\sum_{i=0}^n p(i)= P(n)$, and $P(n)$ is divisible by $n+1$.

Proof. The existence of the polynomial $P(n)$ is well known. To show that $P(n)$ is divisible by $n+1$ it suffices to show that $P(-1) = 0$. But $$P(n) = -p(-1) +\sum_{i=-1}^n p(i),$$ and this formula is valid for $n\ge-1$. Thus $P(-1) = -p(-1) +p(-1) =0$.

A slightly different point of view makes this result seem more obvious. Let $p$ be a polynomial. Then by finite differences or otherwise there is a unique polynomial $Q$ such that $Q(i+1)-Q(i)=p(i)$ for all $i$ (all we need is $i\ge0$) and $Q(0) = 0$. By induction, we have $\sum_{i=0}^{n-1} p(i) = Q(n)$ for all $n\ge0$, where, as usual, $\sum_{i=0}^{-1}$ is 0 by definition. Since $Q(0) =0$, $Q(n)$ is divisible by $n$, so $\sum_{i=0}^n p(i) = Q(n+1)$ is divisible by $n+1$.

Another (slightly more complicated, but self-contained) way to prove the theorem is to note that it holds for the basis of polynomials $p(i) = \binom{i}{k}$, since $\sum_{i=0}^n \binom{i}{k}=\binom{n+1}{k+1}$, which is divisible by $n+1$.

$\endgroup$
1
  • 1
    $\begingroup$ A more sophisticated argument is to use the fact that if $p(i)$ is a polynomial of degree $d$, then $\sum_{i\geq 0}p(i)x^i=A(x)/(1-x)^{d+1}$, where $A(x)$ is a polynomial of degree at most $d$, and the coefficient of $x^d$ in $A(x)$ is $\pm p(-1)$. Then note that $\sum_{n\geq 0}P(n)x^n=A(x)/(1-x)^{d+2}$. $\endgroup$ Commented Jul 14 at 14:04
0
$\begingroup$

Mathematica says:

$$ \frac{1}{2} \left(\frac{\frac{\Gamma (-j-1) \Gamma (-j) \binom{n+1}{j} \Gamma (-j+n+2)}{\Gamma (-j+n+1)}+\Gamma (-2 j-1)}{\Gamma (-j)^2}-\binom{n+1}{j}^2 \, _3F_2(1,n+2,n+2;-j+n+2,-j+n+2;1)\right) $$ I am not quite sure what to make of the Gammas with negative integer arguments, but there you have it.

$\endgroup$
1
  • 1
    $\begingroup$ (revised comment) Mathematica's Binomial function works for fairly arbitrary inputs, and Mathematica will attempt to simplify symbolic expressions using Binomial in whatever manner has been coded into it as the "natural" thing to assume you meant. In this case, it is assuming by default that $j$ is not a positive integer (since that is the case that's problematic). If you force it to assume this, with say Assuming[Element[j,Integers]&& j>0, rest of code ], then it will not produce this output and will just leave it exactly as calls to Binomial. $\endgroup$ Commented Apr 5, 2017 at 19:17

You must log in to answer this question.

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