8
$\begingroup$

Background: A quadratic polynomial splits over a field $k$ iff its discriminant is a square in $k$. Squares in $\Bbb R$ are just the elements $\ge 0$, and it is also quite easy to recognise squares in a $p$-adic field $\Bbb{Q}_p$. (With a little oddity for $p=2$, which nevertheless can be done and has nice special cases.)

In particular it is easy to find, for each prime $p$, an irreducible quadratic $f_p \in \Bbb{Q}[x]$ which factors completely in both $\Bbb{Q}_p$ and $\Bbb R$. Actually, $f_p = x^2-(1+p^n)$ works for $n\ge 2$ for odd $p$, resp. $n \ge 4$ for $p=2$ (where $n=3$ is only excluded because that's the only case where $1+p^n$ is already a square in $\Bbb Q$).

So let's get cubic, let $$f(x) = ax^3+bx^2+cx+d$$ ($a\neq 0$) be irreducible in $\Bbb{Q}[x]$. Over $\Bbb R$, $f$ factors completely iff its discriminant

$$\Delta(f) = 18abcd - 4b^3d + b^2c^2 - 4ac^3 - 27a^2d^2$$

is a square (i.e. $\ge 0$). But this is treacherous. Over a general field $k\supset \Bbb Q$, $\Delta(f)$ being a square is a necessary criterion for $f$ to factor completely, but not sufficient. What makes it work in $\Bbb R$ is that by the Intermediate Value Theorem, $f$ has at least one root $\alpha \in \Bbb R$ to begin with. Then the other roots are expressions of $\sqrt{\Delta(f)}$ and $\alpha$ -- compare Expressing the roots of a cubic as polynomials in one root and Express one root of depressed cubic equation via another and square root of discriminant). In other words, the splitting field of $f$ over $\Bbb Q$ is $\Bbb Q[\alpha, \sqrt{\Delta(f)}]$, so its splitting field over $\Bbb R$ is $\Bbb R[\alpha, \sqrt{\Delta(f)}] = \Bbb R$.

With this argument, a necessary and sufficient criterion for $f$ to factor completely in $k$ is that $\Delta(f)$ is a square in $k$, and $f$ has at least one root in $k$.

But how to check whether $f$ has at least one root in, say, $\Bbb Q_p$? Well, Hensel's Lemma of course. This seems to work somewhat for "big" $p$, but I run into trouble for $p=2,3$.

Question 1: What are nice necessary and sufficient criteria for $f$ to factor completely in $\Bbb{Q}_p[x]$? (With special attention to $p=2,3$, although of course a general approach that does not even have to handle these cases differently would be optimal.)

Question 2: For each prime $p$, is there a cubic $f_p$, irreducible over the rationals, which splits completely both in $\Bbb{Q}_p[x]$ and in $\mathbb{R}[x]$?

Edit: Lubin and Franz Lemmermeyer have given a nice construction for question 2 in the comments, thank you for that. My stance on it would be to set

$$f_p := \begin{cases} x^3-x+pk & \text{for odd }p\\ x^3-x+8k & \text{for }p=2\\ \end{cases}$$ where $k$ is rational with $v_p(k)\ge 0$. Notice that $x^3-x=x(x-1)(x+1)$ has at least one simple root modulo $p$, so Hensel gives us at least one root in $\Bbb Q_p$. Further $$\Delta(f_p) = \begin{cases} 4(1-\frac{3^3k^2}{2^2}\cdot p^2) & \text{for odd }p\\ 4(1-3^3k^2 \cdot 2^4) & \text{for }p=2\\ \end{cases}$$ is a square in $\Bbb Q_p$, so $f_p$ splits completely in $\Bbb{Q}_p$. To make sure it also splits completely in $\Bbb R$, we have to ensure that $\Delta(f_p) \ge 0$, which we do by choosing $k$ with real absolute value $$|k| \le \begin{cases} \frac{2\sqrt3}{9p} & \text{for odd }p\\ \frac{\sqrt 3}{36} & \text{for }p=2.\\ \end{cases}$$ Now the last thing we want to ensure is that $f_p$ is irreducible over $\Bbb Q$. "Generically", it is. Does anybody see a nice way to enforce this? Funnily, my first attempt for $p=2$, namely $k=\frac{1}{21}$ (since $\frac{1}{21}<\frac{\sqrt 3}{36}<\frac{1}{20}$), already gives the $\Bbb Q$-irreducible $$f_2 = x^3-x+\frac{8}{21}$$ even though its discriminant happens to be the rational square $\frac{4}{49}$.

By the way, I have found this site quite useful for playing around with polynomials over $p$-adics: https://math.la.asu.edu/~jj/localfields/

$\endgroup$
  • 4
    $\begingroup$ For #2, it seems to me that if you take $f(x)=(x-a)(x-b)(x-c)$ for distinct integers $a,b,c$, and look at $g=f(x)+\varepsilon$, where $\varepsilon$ is rational and small both in the $p$-adic sense and the real sense, then $g$ should work, if you were not careless in your choice of $\varepsilon$: after all, a “generic” $\Bbb Q$-polynomial will be irreducible. $\endgroup$ – Lubin Jan 24 '18 at 4:39
  • $\begingroup$ I see. Certainly I should have thought of Krasner besides Hensel. $\endgroup$ – Torsten Schoeneberg Jan 24 '18 at 5:26
  • 2
    $\begingroup$ The decomposition law in nonnormal cubic extensions with discriminant $D$ tells you that if $(D/p) = -1$, then $p$ either splits completely or is inert. Thus if $f$ has one root modulo $p$, it must split completely in the $p$-adic completion. For #2, take $f = (x-1)(x-2)(x-3) + p^m$ for a large power of $p$ and show that one of these polynomials is irreducible (I see that's what Lubin said, except that $\epsilon$ does not have to be small in the real sense since cubic polynomials always have a real root). $\endgroup$ – franz lemmermeyer Jan 24 '18 at 13:11
  • 1
    $\begingroup$ To tell the truth, I wasn’t thinking of Krasner either. For $a$, $b$, and $c$ mutually incongruent modulo $p$, Hensel does it for you perfectly well. $\endgroup$ – Lubin Jan 24 '18 at 20:16
  • $\begingroup$ @franzlemmermeyer re second part: If $\epsilon$ is too big in the real sense (concretely, $>\frac{2}{9}\sqrt{3}$ with your values, which $p^m$ certainly is), the polynomial has only one real root ($\Leftrightarrow$ the discriminant becomes negative). It's supposed to factor completely both over $\Bbb R$ and over $\Bbb{Q}_p$. But certainly $\epsilon =(p/q)^m$ for high enough $m$ and $q>p$ with $gcd(q,p)=1$ does the job. $\endgroup$ – Torsten Schoeneberg Jan 25 '18 at 4:11
2
$\begingroup$

$\def\QQ{\mathbb{Q}}$You might know this, but it wasn't explicit: for a polynomial $f$ of degree $d$ over a $p$-adic field, then $\operatorname{Gal}(f) \subseteq A_d$ (the alternating group) if and only if $\Delta(f)$ is a square. In the case $d=3$, then $A_3 = C_3$ (cyclic) and so the only possible groups are $C_3$ (and $f$ is irreducible) or the trivial group (and $f$ splits). The latter cases are therefore distinguished by testing if $f$ has a root in the ground field. Hence your statement that $f$ splits if and only if the discriminant is a square and $f$ has a factor.

Some answers to #1:

Answer 1. The general answer for finding roots of polynomials over $p$-adic fields is "Panayi's root finding algorithm". One place where it is documented is in section 8 of "On the computation of all extensions of a $p$-adic field" by Pauli and Roblot.

Answer 2. Something a little more direct, I suppose, is that you could use the explicit expression for the roots of a cubic polynomial. This will involve testing if various quantities are cubes, so it's quite analogous to the $d=2$ case, and Hensel's lemma will give you a criterion for whether or not a number is a cube.

Answer 3. Recall the discriminant is defined as $$\Delta = \prod_{i \neq j} (\alpha_i - \alpha_j)$$ where $\alpha_i$ are the roots of $f(x)$, and so testing if it has a square root is really testing if $\sqrt{\Delta} = \prod_{i < j} (\alpha_i - \alpha_j) \in \QQ_p$. If we have a square-root, then we know that the Galois group is inside $C_3$. If $\omega\in\QQ_p$ is a primitive cube root of unity (this exists if and only if $3 \mid p-1$), then similarly defining $$\Phi=\prod_{i=1}^3 \alpha_{i \bmod 3}+\omega\alpha_{i+1\bmod3}+\omega^2\alpha_{i+2\bmod3}$$ then you can expand this out and express it in terms of coefficients of $f$ and $\sqrt\Delta$ (just like you can express $\Delta$ in terms of coefficients of $f$). Then Fact: $f$ has a root if and only if $\Phi$ has a cube root. Therefore $f$ splits if and only if $\sqrt{\Delta}$ and $\sqrt[3]{\Phi}$ both exist.

For things like this, the luxury of being able to test for a $n$th root are generally restricted to fields containing primitive $n$th roots of unity. Hence the discriminant is so well-known because every field has square roots of unity!

$\endgroup$
  • $\begingroup$ Thank you for this. Certainly the nicest case is $p \equiv 1$ mod 3 where one can exploit the third roots of unity. Do you happen to have a formula for $\Phi$ in the coefficients? As for Panayi's algorithm, which is new to me, it seems that it is rather employed to classify extensions of $\Bbb Q_p$, whereas I want something to be factored completely in $\Bbb Q_p$ (vaguely said, it lies "under" the $p$-adics, not "over" them). Or were you thinking of using it indirectly, as in "If $f$ did not split over $\Bbb Q_p$, its splitting field would be ..."? $\endgroup$ – Torsten Schoeneberg Jan 25 '18 at 9:04
  • $\begingroup$ Expressing $\Phi$ is an exercise to the reader, I don't know the formula. $\endgroup$ – Doris Jan 25 '18 at 15:46
  • 1
    $\begingroup$ Panayi's algorithm is an algorithm taking $f(x) \in \mathbb Q_p[x]$ and returning the roots of $f(x)$ in $\mathbb Q_p$. So a necessary and sufficient condition of having a root is "Panayi's algorithm says so". As flippant as this sounds, this approach has been used to prove parts of local class field theory explicitly. $\endgroup$ – Doris Jan 25 '18 at 15:48

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.