Crossposted from


In playing around with some formulas, I have come up with the following conjecture. I have checked it for a lot of cases, and have good reason to believe it to be true. If anyone could help, I would sincerely appreciate it, and I would be happy to include as co-author.

Some notation:

  • The exponent $R(i)$ means the formal raising by $2$s for $i$ terms. For example, $(k-7)^{R(3)} = (k-7)(k-5)(k-3)$.

  • The exponent $D(i)$ means the formal descending by $2$s for $i$ terms. For example, $k^{D(4)} = (k)(k-2)(k-4)(k-6)$.

  • $k^{D(1)}:=k$, $k^{D(0)}:=1$ and $k^{D(-1)}:=\frac{1}{k+2}$.

  • !! denotes the double factorial, for example $9!! =$

  • $B_i$ denotes the ith Bernoulli number, with the convention that $B_1 = 1/2$.

In what follows, $M$ is a positive odd integer bigger than or equal to $3$, $k$ is a formal variable. For a given odd $M\geq 3$, the following is a polynomial in $k$. My conjecture is that it has roots $1,3,5,.., M-2$.


For example, when $M=3$, we get

$$\binom{2}{1}[(k-1)^{R(1)}+1!!]B_{1}k^{D(0)}+\binom{2}{2}[(k-1)^{R(2)}-3!!]B_{0}[k^{D(-1)}] $$

$$=2[k]B_1+[(k-1)(k+1)-3]\frac{B_0}{k+2}, $$ which has a root of $1$.

When $M=5$ we get



Putting $k=1$ gives

$$-3B_2-9B_1+5B_0 = 0,$$

putting $k=3$ gives

$$9B_2-9B_1+3B_0 = 0.$$

As another examples, when $M=9$ we get

$$ \binom{5}{1}[(k-7)+1]B_4[(k)(k-2)(k-4)] + \binom{5}{2}[(k-7)(k-5)-1.3]B_3[k(k-2)] + \binom{5}{3}[(k-7)(k-5)(k-3)+1.3.5]B_2[k] + \binom{5}{4}[(k-7)(k-5)(k-3)(k-1)-]B_1 + \binom{5}{5}[(k-7)(k-5)(k-3)(k-1)(k+1)+]\frac{B_0}{k+2}$$

Putting $k=7$, using that $B_3=0$ and factoring out, we get

$$\binom{5}{1}B_4+\binom{5}{3}B_2-\binom{5}{4}B_1+\binom{5}{5}B_0 = -\frac{5}{30}+\frac{10}{6}-\frac{5}{2}+1 = 0$$

In special cases the conjecture boils down to well known identities involving Bernoulli numbers and binomial coefficients. I would appreciate an elementary proof, but really anything will do. Perhaps someone into combinatorics or generating functions can help? Thanks again.

  • 3
    $\begingroup$ You should definitely put the whole question into the text here, and not make people visit the other site to see it. $\endgroup$ May 4 at 22:11
  • 1
    $\begingroup$ @Fox you can hit 'edit' on the other question and copy and paste the source (then cancel the edit), rather than re-type the thing. $\endgroup$ May 4 at 22:54
  • 1
    $\begingroup$ @David, Sam: thanks! $\endgroup$
    – Fox Mulder
    May 5 at 0:41
  • 1
    $\begingroup$ Strengthening a conjecture usually makes it easier to prove, so it is worth noting that empirically the given polynomial for $M = 2m+1$ appears to be $(m+1) (k - 1)^{D(m)}$ (and, FWIW, this holds for $M=1$ too). $\endgroup$ May 7 at 16:52

UPDATE. My earlier answer was incorrect due to miscalculation. Below I give a proof of the conjecture.

First, we notice that $k^{R(i)} = (k/2+i-1)_i\cdot 2^i = \binom{k/2+i-1}i\cdot i!\cdot 2^i$. Similarly, $k^{D(i)} = (k/2)_i\cdot 2^i = \binom{k/2}i\cdot i!\cdot 2^i$ and $(2i-1)!!=\frac{(2i)!}{i!2^i}=\binom{2i}{i}\cdot \frac{i!}{2^i}$. Then the proposed conjecture follows from the following polynomial (in $x$) identity: $$\sum_{i=0}^m \big(\binom{x+\frac12-m+i}i + (-1)^{i+1}\binom{2i}{i} \frac{1}{4^i}\big) B^+_{m-i} \binom{x+1}{m-i} = (x+1) \binom{x-\frac12}{m-1}$$ by taking $m = \frac{M+1}2$ and $x=\frac{k}2$. Noticing that $\binom{x+\frac12-m+i}i = (-1)^i \binom{m-x-\frac32}i$ and $B^+_{m-i} = (-1)^{m-i} B^-_{m-i}$, we rewrite the above identity as $$(\star)\qquad \sum_{i=0}^m \big(\binom{m-x-\frac32}i - \binom{2i}{i} \frac{1}{4^i}\big) B^-_{m-i} \binom{x+1}{m-i} = (-1)^m (x+1) \binom{x-\frac12}{m-1}.$$

Proof. Since for a fixed $m$ the l.h.s. and r.h.s. of $(\star)$ are polynomials in $x$, it is enough to prove $(\star)$ for $x$ being a nonnegative integer. We notice that $$\binom{m-x-\frac32}i - \binom{2i}{i}\frac{1}{4^i} = [z^i]\ \big( (1+z)^{m-x-\frac32} - (1-z)^{-\frac12}\big)$$ while $$B^-_{m-i} \binom{x+1}{m-i} = [z^{m-i}]\ z^{x+1}{\cal B}_{x+1}(\frac1z),$$ where ${\cal B}_{x+1}(t)$ is the $(x+1)$-st Bernoulli polynomial and $[z^n]$ is the operator extracting the coefficient of $z^n$.

It follows that the l.h.s. of $(\star)$ equals $$[z^m]\ \big( (1+z)^{m-x-\frac32} - (1-z)^{-\frac12}\big) z^{x+1}{\cal B}_{x+1}(\frac1z).$$

Using Lagrange–Bürmann formula, we conclude that $$[z^m]\ (1+z)^{m-x-\frac32} z^{x+1}{\cal B}_{x+1}(\frac1z) = [z^m]\ (1-z)^{-\frac12} z^{x+1} {\cal B}_{x+1}(\frac1z-1).$$ Then, by the properties of Bernoulli polynomials, $${\cal B}_{x+1}(\frac1z-1) = {\cal B}_{x+1}(\frac1z) - (x+1) (\frac1z-1)^x.$$

So, the Bernoulli polynomials in the l.h.s. of $(\star)$ cancel out, and it reduces to $$-(x+1)\cdot [z^m]\ z (1-z)^{x-\frac12} = (-1)^m (x+1) \binom{x-\frac12}{m-1}.$$ QED

  • $\begingroup$ Hi Max, thanks for your answer, maybe I'm doing something incorrect, but it seems like when M=9, k=7 is indeed an integer root, I put the calculation in the body of the question at the end, it was too long for a comment. Thanks again for looking into this, greatly appreciated! $\endgroup$
    – Fox Mulder
    May 7 at 13:58
  • $\begingroup$ @FoxMulder: You seem to be correct, and there was a typo in my code causing miscalculation. I'm taking my answer down for now. $\endgroup$ May 7 at 16:12
  • $\begingroup$ @FoxMulder: I've added a proof of your conjecture. $\endgroup$ May 8 at 16:05
  • $\begingroup$ Thanks Max, greatly appreciated! If it's ok with you, I'll email you to chat a bit about this and the wider problem, thanks again! $\endgroup$
    – Fox Mulder
    May 9 at 1:52
  • $\begingroup$ Sure, you are welcome to email me to maxal@gwu.edu $\endgroup$ May 9 at 2:24

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.