4
$\begingroup$

Let $n$ be a positive integer and $1\leq j\leq n$. Consider the following polynomial:

$$p_{n,j}(x)=\frac{\prod\limits_{i=1}^{n+1}\frac{x^{i}+1}{x+1}}{\prod\limits_{i=1}^{j}\frac{x^{i}+1}{x+1}\prod\limits_{i=1}^{n-j+1}\frac{x^{i}+1}{x+1}}\in\mathbb{F}[x]$$

This polynomial can be computed for each $j,n$ given. What I want to do is to understand in which cases $p(1)=0$ and in which cases $p(1)=1$ when working over a field $\mathbb{F}$ of characteristic $2$. I checked some cases by hand and this is what I got: $$p_{2,1}(1)=p_{2,2}(1)=1$$ $$p_{3,1}(1)=p_{3,2}(1)=p_{3,3}(1)=0$$ $$p_{4,1}(1)=p_{4,4}(1)=1~~p_{4,2}(1)=p_{4,3}(1)=0$$ $$p_{5,2}(1)=p_{5,4}(1)=1~~p_{5,1}(1)=p_{5,3}(1)=p_{5,5}=0$$ $$p_{6,1}(1)=p_{6,2}(1)=p_{6,3}(1)=p_{6,4}(1)=p_{6,5}(1)=p_{6,6}(1)=1$$ So apparently there is no easy pattern. It is also easily expressible as a product of cyclotomic polynomials, but I'm not sure if this is of any help. I would like (if possible) to have a way to know if $p_{n,j}(1)=0$ or $p_{n,j}(1)=1$ just in terms of $n$ and $j$.

If you are courious, these are the coefficients that has appeared in the computation of the homology groups of a certain family of Artin groups. I want to understand the behaviour of these polynomials over a field of characteristic $2$ in $x=1$ because this is the situation where I can assure that the homology groups are infinite dimensional.

Thanks for your help.

$\endgroup$
6
  • $\begingroup$ Isn't $p_{n.j}(0)$ always $1$, because $(x^i+1)/(x+1)$ is $1$ at $0$? $\endgroup$ Jan 18 at 11:55
  • $\begingroup$ @OfirGorodetsky ups, right... it is everything at $x=1$ $\endgroup$
    – Marcos
    Jan 18 at 12:02
  • $\begingroup$ Should $\frac{x^i+1}{x+1}$ be $\frac{x^i-1}{x-1}$ everywhere, to (a) get the cyclotomics mentioned in the title; (b) get polynomials rather than rational polynomials? $\endgroup$ Jan 18 at 12:48
  • $\begingroup$ @PeterTaylor, re, we're in characteristic $2$. $\endgroup$
    – LSpice
    Jan 18 at 14:55
  • $\begingroup$ @LSpice, in my opinion it's cleaner to give the definition without specifying the characteristic of $\mathbb{F}$ and then specialise later. The way the question's phrased seems to me to fit this model, except that $p_{n,j}$ isn't necessarily a polynomial except in characteristic 2. $\endgroup$ Jan 18 at 15:13

2 Answers 2

5
$\begingroup$

Claim: $p_{n,j}(1)$ vanishes modulo $2$ if and only there is a carry when adding $j$ and $n+1-j$ in base-$2$.

Proof: Let us write $x^n+1 \equiv x^n-1= \prod_{d \mid n} \Phi_d(x)$ where $\Phi_d$ is the $d$th cyclotomic polynomial. Then, since there are $\lfloor m/d\rfloor$ multiplies of $d$ in $\{1,2,\ldots,m\}$, $$p_{n,j}(x) = \prod_{d=2}^{n} \Phi_d(x)^{\lfloor \frac{n+1}{d} \rfloor- \lfloor \frac{j}{d}\rfloor - \lfloor \frac{n-j+1}{d}\rfloor }.$$ Note that $\lfloor a+b \rfloor \ge \lfloor a\rfloor + \lfloor b \rfloor$ so your expression is necessarily a polynomial (and not a strictly rational function).

The value of $\Phi_d(x)$ at $x=1$ is well-understood. One can use the property $x^n-1= \prod_{d \mid n} \Phi_d(x)$ to show $\Phi_d(1)$ is $1$ unless $d$ is a prime power $p^k$ in which case $\Phi_d(1)=p$ (this is classical, see here for proofs). This implies $$p_{n,j}(1) \equiv \prod_{p^k \le n} p^{\lfloor \frac{n+1}{p^k} \rfloor- \lfloor \frac{j}{p^k}\rfloor - \lfloor \frac{n-j+1}{p^k}\rfloor }.$$ Since $2$ is the only even prime, $$p_{n,j}(1) \equiv \prod_{2^k \le n} 2^{\lfloor \frac{n+1}{2^k} \rfloor- \lfloor \frac{j}{2^k}\rfloor - \lfloor \frac{n-j+1}{2^k}\rfloor }.$$ So your expression vanishes modulo $2$ if and only if $$ \sum_{2 \le 2^k \le n} \left\lfloor \frac{n+1}{2^k} \right\rfloor- \left\lfloor \frac{j}{2^k}\right\rfloor- \left\lfloor \frac{n-j+1}{2^k}\right\rfloor\ge 1.$$

Let us express this criterion using base-$2$ representation. If $j = \sum_{i \ge 0} a_i 2^i$, $n-j+1 = \sum_{i \ge 0} b_i 2^i$ and $n+1 = \sum_{i \ge 0} c_i 2^i$ then $$ \left\lfloor \frac{n+1}{2^k} \right\rfloor- \left\lfloor \frac{j}{2^k}\right\rfloor- \left\lfloor \frac{n-j+1}{2^k}\right\rfloor =\sum_{i \ge k} (c_i-a_i-b_i) 2^{i-k} $$ so that $$\begin{align} \sum_{2 \le 2^k \le n} \left\lfloor \frac{n+1}{2^k} \right\rfloor- \left\lfloor \frac{j}{2^k}\right\rfloor- \left\lfloor \frac{n-j+1}{2^k}\right\rfloor &= \sum_{i \ge 0} (c_i-a_i-b_i)(2^{i-1}+2^{i-2}+\ldots+1) \\ &= \sum_{i \ge 0} (c_i-a_i-b_i)(2^{i}-1)\\ &= n+1 - (j + (n+1-j)) - \sum_{i \ge 0} (c_i-a_i-b_i)\\ &= \sum_{i \ge 0} (a_i+b_i-c_i). \end{align}$$ Let $s_2(m)$ be the sum of digits of $m$ in base-$2$. We see that $p_{n,j}(1)\equiv 0$ if and only if $s_2(j) + s_2(n+1-j) > s_2(n+1)$.

In general, $s_2(a)+s_2(b)-s_2(a+b)$ is the number of carries when adding $a$ and $b$ in base-$2$ (can be proved by induction, goes back to Kummer).

$\endgroup$
4
$\begingroup$

The question title talks about cyclotomic polynomials, so consider instead $$q_{n,j}(x)=\frac{\prod_{i=1}^{n+1} \Psi_i(x)}{\prod_{i=1}^j \Psi_i(x) \prod_{i=1}^{n-j+1} \Psi_i(x)}$$ where $\Psi_k(x) = \frac{x^k - 1}{x - 1} = 1 + x + \cdots + x^{k-1}$. Then $\Psi_k(1) = k$ and $$q_{n,j}(1) = \frac{(n+1)!}{j!(n+1)!} = \binom{n+1}{j}$$

In characteristic 2, $q_{n,j} = p_{n,j}$ so you can apply Kummer's theorem to find $p_{n,j}(1)$.

$\endgroup$

Your Answer

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

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