34
$\begingroup$

What I mean to ask is this:

given an irreducible cubic polynomial $P(X)\in \mathbb{Z}[X]$ is there always a quadratic $Q(X)\in \mathbb{Z}[X]$ such that $P(Q)$ is reducible (as a polynomial, and then necessarily the product of 2 irreducible cubic polynomials)?

I did quite some testing and always found a $Q$ that does the job. For example:

$P=aX^3+b,\quad Q=-abX^2,\quad P(Q)=-b(a^2bX^3-1)(a^2bX^3+1)$

$P=aX^3-x+1,\quad Q=-aX^2+X,\quad P(Q)=-(a^2X^3-2aX^2+X-1)(a^2X^3-aX^2+1)$

and a particular hard one to find:

$P=2X^3+X^2-X+4,\quad Q=-8X^2+5X+1,\quad P(Q)=(16X^3-18X^2+X+3)(64X^3-48X^2-11x-2)$

Could there be a formula for $Q$ that works for all cases?

It feels to me that this may have a really basic Galois theoretic proof or explanation, but I can't figure it out.

Update. Maybe a general formula for $Q$ is close. For $P=aX^3+cX+d$ taking $Q=-adX^2+cX$ works.

$\endgroup$
7
  • $\begingroup$ Sorry, but why can't it be a product of an irreducible quadratic and an irreducible quartic? $\endgroup$ Jul 15, 2021 at 20:27
  • 3
    $\begingroup$ @mathworker21: Then $P(Q)$ would have a quadratic root $\alpha$ and therefore $P$ would have a quadratic root $Q(\alpha)$, contradicting the irreducibility of $P$ (which means that a root would generate a field of degree $3$ over $\mathbb{Q}$). $\endgroup$ Jul 15, 2021 at 20:33
  • 2
    $\begingroup$ There are a number of interesting results known about "composition factorization" of polynomials. I realize this is not what you're asking, but one could ask "which cubic polynomials $f$ and $g$ have the property that $f\cdot g$ has a composition factorization. In any case, here are two reference that you might find useful: Beardon, A. F. Ng, T. W. On Ritt's factorization of polynomials. J. London Math. Soc. (2) 62 (2000), no. 1, 127–138. [MR1771856] Beardon, A. F. Composition factors of polynomials. Complex Variables Theory Appl. 43 (2001), no. 3-4, 225–239. [MR1820924] $\endgroup$ Jul 15, 2021 at 22:12
  • 2
    $\begingroup$ The case $P=aX^3+cX+d$ can be transformed through variable changes and rescalings, into the case $P=aX^3+bX^2+d$, giving then $Q=-(81a^3d+6ab^3)X^2-3b^2X-b/(3a)$. While I believe that there are always particular solutions over $\mathbb{Z}$, it's possible that general solutions can only be found over $\mathbb{Q}$... $\endgroup$ Jul 15, 2021 at 23:09
  • 3
    $\begingroup$ Over $\mathbb{Q}$, for $P = ax^3+bx^2+cx+d$ you get $Q = \frac{-27da^2 + 9bca - 2b^3}{27a}x^2 + \frac{3ca - b^2}{3a}x - \frac{b}{3a}$ by eliminating the $x^2$ term and applying your formula. $\endgroup$
    – Aurel
    Jul 16, 2021 at 7:52

1 Answer 1

33
$\begingroup$

You should refer to Lemma 10 (page-233) in this paper by Schinzel where he proves that for any polynomial $F(x)$ of degree $d$ we have a polynomial $G(x)$ of degree $d-1$ such that their composition is reducible.

$\endgroup$
7
  • 8
    $\begingroup$ To save others the scrolling: Lemma 10 is on page 233 of the paper. $\endgroup$
    – Alex B.
    Jul 16, 2021 at 12:02
  • 1
    $\begingroup$ @AlexB. Indeed... I wish I saw your comment earlier! $\endgroup$ Jul 16, 2021 at 14:21
  • $\begingroup$ Sorry @Yaakov Baruch I should have added page number. I will add right now. $\endgroup$ Jul 16, 2021 at 14:23
  • $\begingroup$ The integrality of coefficients is achieved there by a suitable choice of some $k$; I'm not sure, but it seems to me that $k$ could be a chosen to be a rational function of the coefficients (maybe a very messy one, using sums of squares of many things to avoid using the max function), in which case the resulting $H$ (our $Q$) would be given by one formula in the coefficients. $\endgroup$ Jul 16, 2021 at 14:40
  • 6
    $\begingroup$ See also my paper with Bober, Fretwell, and Wooley, Theorem 3.2, to see that there are infinitely many such quadratic polynomials $G(x)$. (The method is certainly based on that of Schinzel.) For those with a background in abstract algebra, I believe our proof has more intuition behind it. $\endgroup$ Jul 16, 2021 at 19:15

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.