5
$\begingroup$

Suppose $(a_1,\ldots, a_k)$ is an integer partition of $n$, and $(b_1,\ldots,b_k)$ is a rearrangement of the $a$-sequence. Prove the following identity (preferably combinatorially): $$ \sum_{j_2+\cdots+j_k=l,\atop l\geq 0, \, a_t>j_t\geq 0} \quad\frac{(-1)^l l!}{(a_1+l+1)_{l+1}} {a_2\choose j_2}\cdots {a_k \choose j_k}=\sum_{j_2+\cdots+j_k=l,\atop l\geq 0, \, b_t>j_t\geq 0} \quad\frac{(-1)^l l!}{(b_1+l+1)_{l+1}} {b_2\choose j_2}\cdots {b_k \choose j_k}, $$ where $(x)_k$ is the falling factorial.

$\endgroup$

1 Answer 1

4
$\begingroup$

I assume you meant $j_t\leq a_t$ (not $j_t<a_t$).

It's sufficient to prove that for any analytic function $f(x)$, the function $$F(a_1,a_2) := \sum_{l\geq 0} \sum_{j=0}^{a_2} \frac{(-1)^ll!}{(a_1+l+1)_{l+1}} \binom{a_2}{j} [x^{l-j}]\ f(x)$$ is symmetric, i.e. $F(a_1,a_2) = F(a_2,a_1)$.

Using the property of beta function, we have

\begin{split} F(a_1,a_2) &=\sum_{l\geq 0} \frac{(-1)^ll!}{(a_1+l+1)_{l+1}} [x^l]\ (1+x)^{a_2}f(x) \\ &=\sum_{l\geq 0} (-1)^l \int_0^1 t^l (1-t)^{a_1} {\rm d}t\ [x^l]\ (1+x)^{a_2}f(x)\\ &=\int_0^1 (1-t)^{a_1} (1-t)^{a_2} f(-t) {\rm d}t, \end{split} which is clearly symmetric.

$\endgroup$
8
  • $\begingroup$ Thanks Max! Actually, it has to be $j_t < a_t$, otherwise the identity does not hold. This can be checked, for instance, using the integer partition $(2,1)$. $\endgroup$
    – ricky chen
    Apr 4, 2022 at 11:16
  • $\begingroup$ @rickychen: I do not quite follow. When $k=2$, the sum over $j_2=l\leq a_2$ gives $\frac{1}{a_1+a_2+1}$, which is symmetric. $\endgroup$ Apr 4, 2022 at 18:49
  • $\begingroup$ Let me proceed carefully, assuming $j_2 \leq a_2$. For $(2,1)$, the left-hand side gives $$\frac{(-1)^0 0!}{(2+0+1)_1}{1\choose 0}+ \frac{(-1)^1 1!}{(2+1+1)_2}{1\choose 1}$$, while for $(1,2)$, the right-hand side gives $$\frac{(-1)^0 0!}{(1+0+1)_1} {2\choose 0} +\frac{(-1)^1 1!}{(1+1+1)_2}{2\choose 1} +\frac{(-1)^2 2!}{(1+2+1)_3}{2\choose 2}$$. They are not equal! But, if we assume $j_2<a_2$, then they are equal. $\endgroup$
    – ricky chen
    Apr 5, 2022 at 0:04
  • $\begingroup$ please see my detailed computation above, one gives $\frac{1}{3}-\frac{1}{12}$, the other gives $\frac{1}{2}-\frac{1}{6}+\frac{2}{24}$. BTW, why I could not see "@max" that I put at the very beginning. $\endgroup$
    – ricky chen
    Apr 5, 2022 at 0:41
  • $\begingroup$ You have a numerical error in the second term of the second sum. It should be $\frac{1}{2} - \frac{2}{6} + \frac{2}{24} = \frac{1}{4}$, which matches the first sum and the formula $\frac1{a_1+a_2+1}$. $\endgroup$ Apr 5, 2022 at 1:03

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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