4
$\begingroup$

For any given positive integers $m$, $n$ and $q$, such that $m\leq n$ the following sum $$S_p= \sum_{p=0}^m(-1)^{p+q} \binom mp \binom mq \binom np \binom nq\frac{p! q! (m+n-p-q)!}{m! n!}$$ is equal to $1$ if $q=0$ and $0$ otherwise.

The result can be obtained remarking that $S_p$ of the form $$ \sum_{p=0}^m(-1)^p\binom mp P(m-p)\tag{1}$$ where $P$ is a polynomial of degree $m-q$. One then uses the following property of the form (1) for polynomials $P$ of degree at most $m$, that the sum is equal to $m! a_m$, where $a_m$ is the coefficient of $X^m$ in $P(X)$.

  • What is the name of this property (or a reference, I only know it from wikipedia) ?
  • Is there a simpler way to demonstrate the result without using this particular property ?
$\endgroup$
5
$\begingroup$

We need to calculate the sum

$$S = \sum_{p \ge 0}t_p = \sum_{p \ge 0}(-1)^{p+q} \binom mp \binom mq \binom np \binom nq\frac{p! q! (m+n-p-q)!}{m! n!}$$

where $t_p = 0$ for $p > m$ because $\binom{m}{p} = 0$ for $p > m$.

The term ratio $$\frac{t_{p+1}}{t_p} = \frac{(-1)(m-p)(n-p)(p+1)}{(p+1)(p+1)(m+n-p-q)} = \frac{(p-m)(p-n)}{(p+q-m-n)(p+1)}$$

is a rational function of the summation index $p$.

Thus $S$ is a hypergeometric series given by

$$\begin{align}S &= {_2}F_1\left[\begin{matrix} -m & -n\\q-m-n\end{matrix} \text{ ; } 1\right] \cdot t_0\end{align}$$

which fits the Gauss hypergeometric identity to give

$$S = \frac{\Gamma(q)\Gamma(q-m-n)}{\Gamma(q-m)\Gamma(q-n)} \cdot t_0$$

We know that

$\begin{align}t_0 &=(-1)^q \binom{m}{q}\binom{n}{q}\frac{q!(m+n-q)!}{m!n!}\\&=(-1)^q\frac{\Gamma(m+n-q+1)}{\Gamma(q+1)\Gamma(m-q+1)\Gamma(n-q+1)}\end{align}$

after writing the factorials and binomial coefficients in terms of the Gamma function.

Substituting this into $S$ gives

$\begin{align}S &= \color{darkblue}{(-1)^q} \frac{\Gamma(q)}{\color{darkblue}{\Gamma(q+1)}}\frac{\Gamma(q-m-n)\color{darkblue}{\Gamma(m+n-q+1)}}{\color{darkgreen}{(}\Gamma(q-m)\color{darkblue}{\Gamma(m-q+1)}\color{darkgreen}{)}\cdot \color{darkgreen}{(}\Gamma(q-n)\color{darkblue}{\Gamma(n-q+1)}\color{darkgreen}{)}}\end{align}$

Pairs of terms find themselves amenable to the Euler reflection formula, giving an expression in terms of the sine function.

$\begin{align}S &= (-1)^{q} \frac{1}{q}\frac{\frac{\pi}{\sin{\pi(q-m-n)}}}{\frac{\pi}{\sin{\pi(q-m)}}\frac{\pi}{\sin{\pi(q-n)}}} \\\\ &= \frac{(-1)^{q}}{q\pi }\frac{\sin{(q\pi - m\pi)}\sin{(q\pi - n\pi)}}{\sin{(q\pi - (m+n)\pi )}} \end{align}$

The trigonometric identities for symmetry and periodicity reduce the sum to

$\begin{align}S &= \frac{(-1)^{q}}{q\pi}\frac{(-1)^{m}\sin{( q\pi)}(-1)^n\sin{(q \pi)}}{(-1)^{m+n}\sin{(q\pi)}}\\ &= \frac{(-1)^{q}\sin{( q\pi)}}{q\pi} \\ &= \begin{cases} 0 & q \ne 0 \\ 1 & q = 0 \end{cases}\end{align}$

In terms of the Kronecker delta, the sum becomes $$\color{darkred}{S = \delta_q}$$

Note: I referenced chapter 3 - The Hypergeometric Database of the wonderful book A=B for this problem.

$\endgroup$
  • $\begingroup$ Thank you very much for this answer. I definitely learned something reading it. And thank you also for the reference to the book A=B. $\endgroup$ – Tom-Tom Apr 22 '14 at 19:44
2
$\begingroup$

I really do not know how much this could help : using a CAS, what I found is that $$S=\sum_{p=0}^m(-1)^{p+q} \binom mp \binom mq \binom np \binom nq\frac{p! q! (m+n-p-q)!}{m! n!}=\frac{(-1)^q \sin (\pi q)}{\pi q}$$

$\endgroup$
  • $\begingroup$ So the question is how did it get to this expression... $\endgroup$ – Tom-Tom Apr 17 '14 at 10:15
  • $\begingroup$ I really do not know either ! I was just fascinated by the expression; so I tried. Sorry for not helping you more with this problem. The result is quite amazing. Cheers. $\endgroup$ – Claude Leibovici Apr 17 '14 at 10:18
  • $\begingroup$ The result on the polynomial concerning the expression (1) is proved in the following paper: An Algebraic Identity Leading to Wilson's Theorem Sebastián Martín Ruiz The Mathematical Gazette Vol. 80, No. 489 (Nov., 1996) (pp. 579-582). It is absolutely not easy. I doubt there is another demonstration of the result skipping this part. But after your comment, I think there cold be an hypergeometric identity hidden behind it (the way your CAS has followed, probably). $\endgroup$ – Tom-Tom Apr 17 '14 at 11:57

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.