14

A friend recently asked me the following question, which I've been finding tough:

Original question: Given $n \ge 5$, consider the polynomial $p_n(x) = x^n + x^4 + x^2 + x + 1$ (where the lack of $x^3$ is intentional). When is it irreducible (over $\mathbf Z$)?

Partial working: I noticed the following:

  1. When $n \equiv 3 \mod 5$, note that the exponents range across all integers (mod $5$), and so $p_n(\zeta) = 0$ for any primitive $5$th root of unity $\zeta$. It follows that the $5$th cyclotomic polynomial (of degree $4$) divides $p_n$, and thus $p_n$ is reducible.

  2. Something similar happens with $n \equiv 4 \mod 6$, since if $\zeta$ is a primitive $6$th root of unity then $x^n + x^4 + x^2 + x + 1 = (1 + \zeta^2 + (\zeta^2)^2) + \zeta(\zeta^{n-1} + 1) = 0$, since $\zeta^{n-1} = \zeta^3 = -1$.

This proves one direction of the following

Conjecture: $p_n$ is reducible iff ($n \equiv 3 \mod 5$ or $n \equiv 4 \mod 6$).

Doing some computations with Wolfram Alpha has verified this for at least $5 \le n \le 100$, but I'm unsure how to make further progress - i.e. to show $p_n$ is irreducible for other $n$. I'm also unsure how this periodicity in $n$ would come about. I hoped I'd find some prime $\ell$ modulo which the conjecture is true and then I'd be done, but this has also been in vain.

Question: are there any ideas on how to (dis?)prove this conjecture? (In particular: is there any good reason why irreducibility of polynomials should be 'periodic' mod $n$ like this?)

Sam Moore
  • 614
  • For, say, $n\equiv 3\bmod 5$ you can show that $x^4+x^3+x^2+x+1$ is an irreducible factor, so $p_n$ is reducible. This shows one direction of the conjecture. – Dietrich Burde Dec 21 '24 at 21:41
  • Yes, indeed I have proven that direction of the conjecture (in the partial working); the other direction is the hard bit! I'll edit to make that clearer – Sam Moore Dec 21 '24 at 21:44
  • I've checked your conjecture up through $n=600$ with sympy, and it remains true. – Carl Schildkraut Dec 22 '24 at 03:51
  • 1
    I've written out a proof. Periodicity ends up coming from when $f_n$ and its reversal have common roots. Where did you find this question? It's quite a good one. – Joshua P. Swanson Dec 22 '24 at 10:18
  • 1
    @JoshuaP.Swanson Just asked - apparently it's a question in the latest edition of Stewart's Galois Theory - which I'm surprised by, since I'm unsure how Galois theoretic ideas would factor into this. – Sam Moore Dec 22 '24 at 13:00
  • 1
    The (seemingly canonical?) argument I worked out is certainly related to Galois theory, though I'd be very impressed to see an argument using the fundamental theorem and the like, or really any "conceptual" proof. I wonder what Stewart had in mind. – Joshua P. Swanson Dec 22 '24 at 13:38

1 Answers1

7

Strategy: See KCd's outline, which applies here mutatis mutandis. The idea for proving irreducibility is as follows. (A) Show that $f(x)$ and its reversal $\tilde{f}(x) = x^n f(1/x)$ have no common roots. (B) Argue directly that $f \tilde{f}$ has no degree $n$ factors besides the obvious ones. (C) Use (B) to argue that a factor of $f$ divides both $f$ and $\tilde{f}$, so from (A) it must be constant.

The modular conditions become apparent in step (A), which also proves the reducible cases.

Step A: Let $f_n(x) = x^n + x^4 + x^2 + x + 1$, so $\tilde{f_n}(x) = 1 + x^{n-4} + x^{n-2} + x^{n-1} + x^n$. We will show \begin{align*} n \equiv_5 3 &\Rightarrow \Phi_5 \mid f_n \qquad\text{(reducible)} \\ n \equiv_6 4 &\Rightarrow \Phi_6 \mid f_n \qquad\text{(reducible)} \\ \gcd(f_n, \tilde{f_n}) &= 1 \text{ otherwise}. \end{align*}

A little algebra gives \begin{equation} f_n(x) = \Phi_5 + x^3 (x^{n-3} - 1) = \Phi_3 \Phi_6 + x (x^{n-1} + 1) \qquad(1) \end{equation} where $\Phi_d$ is a cyclotomic polynomial, so that \begin{equation} \tilde{f_n}(x) = x^{n-4} \Phi_5 - (x^{n-3} - 1) = x^{n-4} \Phi_3 \Phi_6 + (x^{n-1} + 1). \end{equation} Note that $$ \tilde{f_n} - x^{n-4} f_n = -(x^{n-3} - 1)(x^{n-1} + 1), $$ so $$\gcd(f_n, \tilde{f_n}) = \gcd(f_n, (x^{n-3} - 1)(x^{n-1} + 1)). \qquad(2) $$

Irreducible contributions to (2) are either divisors of $x^{n-3} - 1$ or $x^{n-1} + 1$.

  • If $g \mid x^{n-3} - 1$ and $g \mid f_n$, then from the first equality in (1), $g$ must be $\Phi_5$. But $\zeta_5^{n-3} - 1 = 0$ if and only if $n \equiv_5 3$.
  • If $g \mid x^{n-1} + 1$ and $g \mid f_n$, then from the second equality in (1), $g$ must be $\Phi_3$ or $\Phi_6$. Now $\zeta_3^{n-1} + 1$ is never zero, whereas $\zeta_6^{n-1} + 1 = 0$ precisely when $n \equiv_3 1$ and $(n-1)/3$ is odd, which is precisely when $n \equiv_6 4$.

The claim follows.

Step B: Suppose $f_n \tilde{f_n} = k \tilde{k}$ for some $k \in \mathbb{Z}[x]$ of degree $n$. We will show $k = \pm f_n$ or $\pm \tilde{f_n}$.

Let $k(x) = k_0 + k_1 x + \cdots + k_n x^n$. Compare coefficients of $x^n$ in $f_n \tilde{f_n} = k \tilde{k}$ to get $5 = \sum_{i=0}^n k_i^2$. Hence either $k(x)$ is a sum of $5$ monomials with coefficients $\pm 1$, or $k(x)$ is a sum of two monomials with one coefficient $\pm 2$ and the other $\pm 1$. Since $f_n \tilde{f_n}$ is monic with constant coefficient $1$, we find $k_0 k_n = \pm 1$, so indeed $k(x)$ is a sum of five monomials. Replacing $k$ with $-k$ if necessary, we may assume $k_0 = k_n = 1$.

For $5 \leq n \leq 8$, the result may be checked directly, so now take $n \geq 9$. We have \begin{align*} f_n \tilde{f_n} &= x^{2n} + x^{2n-1} + x^{2n-2} + x^{2n-4} + x^{n+4} + \cdots \\ k \tilde{k} &= (x^n + \epsilon_1 x^{n_1} + \epsilon_2 x^{n_2} + \epsilon_3 x^{n_3} + 1) \\ &\ \ \ \cdot (x^n + \epsilon_3 x^{n-n_3} + \epsilon_2 x^{n-n_2} + \epsilon_1 x^{n-n_1} + 1). \end{align*}

Comparing coefficients of $x^{2n-1}$ forces ($n_1 = n-1$ and $\epsilon_1 = 1$) or ($n_3 = 1$ and $\epsilon_3 = 1$). We may replace $k$ with $\tilde{k}$ if necessary and assume $n_3=1$ with $\epsilon_3 = 1$, so $\tilde{k}(x) = x^n + x^{n-1} + \cdots + 1$.

If $n_1 = n-1$, the coefficient of $x^{2n-1}$ would be $2$ or $0$, so $n_1 \leq n-2$. If $n-n_2 < n-2$, then the only possible contribution to $x^{2n-2}$ is from $\epsilon_1 x^{n_1} \cdot x^n$ if $n_1=n-2$ and $\epsilon_1=1$. At this stage, the coefficient of $x^{2n-3}$ is $0 = 1 + \epsilon_2 [n_2=n-3] + \epsilon_2 [n_2=3]$ (using Iverson brackets). Hence $\epsilon_2=-1$ and either $n_2=n-3$ or $n_2=3$. If $n_2=n-3$, the coefficient of $x^{2n-4}$ is $-1 \neq 1$, and if $n_2=3$, the coefficient of $x^{2n-5}$ is $-1 \neq 0$. Thus $n-n_2 \geq n-2$ after all, forcing $n_2 = 2$.

At this stage $\tilde{k} = x^n + x^{n-1} + \epsilon_2 x^{n-2} + \epsilon_1 x^{n-n_1} + 1$ for $n_1 \leq n-2$. The coefficient of $x^{2n-2}$ is $1 = \epsilon_2 + \epsilon_1 [n_1=n-2] + \epsilon_1 [n_1=n-3]$. Hence $\epsilon_2 = 1$ and $n_1 \leq n-4$. If $n_1 = 3$, then the coefficient of $x^{2n-3}$ would be $\epsilon_1 \neq 0$, so $n_1 > 3$. If $n_1 \geq 5$, then $n-n_1 \leq n-5$, and the coefficient of $x^{2n-4}$ would be $0 \neq 1$. Hence $n_1=4$, in which case the coefficient of $x^{2n-4}$ is $\epsilon_1 = 1$. That is, $k = f_n$.

Step C: Suppose that $n \not\equiv_5 3$ or $n \not\equiv_6 4$ and let $f_n = gh$ where $g, h \in \mathbb{Z}$ have non-zero constant term. Clearly $\tilde{f_n} = \tilde{g}\tilde{h}$. Let $k=g\tilde{h}$, which is of degree $n$. Now $k\tilde{k} = g\tilde{h}\tilde{g}h = f_n\tilde{f_n}$, so from (B), $k = \pm f_n, \pm \tilde{f_n}$. If $k = \pm f_n$, then $\tilde{h} \mid f_n$, so $\tilde{h} \mid \gcd(f_n, \tilde{f_n}) = 1$, and $h$ is constant. Similarly if $k = \pm \tilde{f_n}$, then $g$ is constant. Thus $f_n$ is irreducible.

  • Brilliant, thank you - that's an interesting method! Did you come to (1) just by dividing through $\phi_6$ and $\phi_3$ and playing around with that? – Sam Moore Dec 22 '24 at 12:59
  • 1
    For (1), I computed the gcd for a few values of $n$, which gave the $\Phi_5$ factor, then pulled out the most obvious piece that looked like $\Phi_5$ and simplified the leftovers. I knew I was on the right track since you can read off divisibility by $\Phi_5$ immediately. The $\Phi_6$ version was a little weirder--a little playing around made it clear that I needed to pull out $\Phi_3 \Phi_6$ instead of just $\Phi_6$. The lovely form of the difference made it clear I had made the right choices. – Joshua P. Swanson Dec 22 '24 at 13:45
  • Thanks, that makes sense. – Sam Moore Dec 22 '24 at 14:06