I have a task where I need to find a polynomial $R \in \mathbb{Q}[X]$ that has roots $\alpha + \beta$ where $\alpha$ are arbitrary roots of $X^5 - 2$ and $\beta$ of $X^4 - 3\enspace.$

The general technique to use is obvious to me, the polynomial is given by $$ R = \prod_{i, j} (x - (\alpha_i + \beta_j)) $$ with the corresponding roots $$ \begin{align*} \alpha_1 &= \sqrt[5]{2}\\ \alpha_2 &= -\sqrt[5]{-2}\\ \alpha_3 &= (-1)^{\frac{2}{5}}\sqrt[5]{2}\\ \alpha_4 &= -(-1)^{\frac{3}{5}}\sqrt[5]{2}\\ \alpha_5 &= (-1)^{\frac{4}{5}}\sqrt[5]{2}\\ \beta_1 &= \sqrt[4]{3}\\ \beta_2 &= -\sqrt[4]{3}\\ \beta_3 &= i\sqrt[4]{3}\\ \beta_4 &= -i\sqrt[4]{3}\enspace. \end{align*} $$ (compare to questions like Find polynomial whose root is sum of roots of other polynomials)

Now I know that one can argue that $R$ is indeed in $\mathbb{Q}[X]$, but we didn't yet step into Galois-theory so I probably can't use that, nor do I know the details of how to prove this.

The problem with manually computing $R$ is that the resulting equation is huge, really huge. After some hours of Mathematica magic I could resolve the above to $$ \begin{align*} S &= X^{20} - 15X^{16} -8X^{15} + 90X^{12} -1560X^{11} + 24X^{10} - 270X^8 - 11160X^7\\ &\qquad- 4080X^6 - 32X^5 + 405X^4 - 7560X^3 + 3960X^2 - 480X - 227\enspace. \end{align*} $$

The problem is that, if I go from this direction, I would need to check all $5 \cdot 4 = 20$ roots. But that is again a huge task.

At this point I'm stuck, both directions aren't eligible for a manual approach. I probably need some kind of fancy argument with which I can follow that $R = S$ or that $S$ has those roots (despite the fact that $S$ fall out of the sky).

Alternatively, arguments that $R$ is in $\mathbb{Q}[X]$ due to construction. But in this case I would need some guidance because, as said, I'm not sufficiently acquainted with Galois-theory.

  • 7
    $\begingroup$ It's quite easy to write down a $20$-by-$20$ matrix whose characteristic polynomial is your polynomial. Do you like computing characteristic polynomials of large matrices? $\endgroup$ – Lord Shark the Unknown Dec 15 '17 at 3:31
  • 4
    $\begingroup$ After some hours of Mathematica magic ... Or in WA resultant[ x^5-2, (z-x)^4 - 3, x ], or minimalpolynomial[ surd(2,5)+surd(3,4), z ]. $\endgroup$ – dxiv Dec 15 '17 at 3:44
  • 3
    $\begingroup$ I thijk that you need to know just about symmetric polynomials (which would be MUCH less than Galois Theory). $\endgroup$ – Wlod AA Dec 15 '17 at 3:55

I'm not sure if this this is the argument you want. To be precise, what I want to explain is that for monic irreducible polynomials $f(x)$ and $g(x)$, and their roots $\alpha_i$, $\beta_j$ are the conjugates of the roots of them, respectively, the polynomial $$ p(x) = \prod_i \prod_j \left( x-\alpha_i -\beta_j \right) $$ is in $\mathbb{Z}[x]$ without referring the Galois theory.

Viewed as a polynomial over $\mathbb{Z}[\alpha_1, \alpha_2, \ldots , \alpha_n]$, the coefficients of $p(x)$ are symmetric polynomials of $\beta_j$s, say $\sigma_k$ be the degree $k$ elementary symmetric polynomial of $\beta_j$s. Therefore coefficients of $p(x)$ are of the form $B \left( \sigma_1 , \sigma_2 , \ldots , \sigma_m , \alpha_1 , \alpha_2 , \ldots , \alpha_n \right) $. Hence $B$ is a polynomial of $\alpha_i$s with integral coefficients. However, $B$ is also symmetric of $\alpha_i$. Therefore, is an integer. This proves that $p(x)$ is in $\mathbb{Z}[x]$.


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.