If I have a polynomial, P, with root $a$ and a polynomial, Q, with root $b$, is there a way to construct polynomial R such that $a+b$ is a root of R?

Here's a concrete example. a = $\sqrt2$. $P(x) = x^2 -2$ and $P(a) = 0$.

b = $\sqrt3$. $Q(x) = x^2 - 3$ and $Q(b) = 0$.

From another method, I know that $x^4 -10x^2 + 1$ has $\sqrt2 + \sqrt3$ as a root.
Is there a way to construct $R(x)$ from $P(x)$ and $Q(x)$? $P$, $Q$, and $R$ are polynomials with integer coefficient.

Thank you.


Gaussian elimination is the key. The ring $$ \mathbb{Q}[x,y]/(P(x),Q(y)) $$ is a vector space over $\mathbb{Q}$ with dimension $\partial P\cdot \partial Q$, and a base given by $x^i y^j$ for $i\in[0,\partial P-1],j\in[0,\partial Q-1]$. By representing $1,(x+y),(x+y)^2,\ldots,(x+y)^{\partial P\cdot\partial Q}$ in such a base we get $\partial P\cdot\partial Q+1$ elements in a vector space with dimension $\partial P\cdot\partial Q$: it follows that $(x+y)^{\partial P\cdot \partial Q}$ can be represented as a linear combination of $1,(x+y),\ldots,(x+y)^{\partial P\cdot \partial Q-1}$ with rational coefficients: that provides a polynomial that vanishes at $a+b$.

In our concrete example, $P(x)=x^2-2$ and $Q(y)=y^2-3$, so from:

$$\begin{pmatrix} 1 \\ x+y \\ (x+y)^2 \\ (x+y)^3 \\ (x+y)^4\end{pmatrix}=\begin{pmatrix} 1 & 0 & 0 & 0\\ 0 & 1 & 1 & 0 \\ 5 & 0 & 0 & 2 \\ 0 & 11 & 9 & 0 \\ 49 & 0 & 0 & 20\end{pmatrix}\begin{pmatrix}1 \\ x \\ y \\ xy\end{pmatrix}$$ we get $(x+y)^4 = 10(x+y)^2 - 1$, and $R(t)=\color{red}{t^4-10t^2+1}$ is a polynomial that vanishes at $\sqrt{2}+\sqrt{3}$. By studying the rank of the previous matrix, we also have that $R(t)$ is the minimal polynomial of $\sqrt{2}+\sqrt{3}$ over $\mathbb{Q}$.


Hint $ $ We do a slightly less trivial case $\,a=\sqrt2,\,b=\sqrt[3]3.\,$ $\Bbb Q(a,b)\cong \Bbb Q\langle1,b,b^2, a,ab,ab^2\rangle$ as a vector space. $\,x\mapsto (a\!+\!b)\,x\,$ is a linear map on this vector space. Compute its matrix, then apply Cayley-Hamilton, to compute a (characteristic) polynomial having $\,a\!+\!b\,$ as root (e.g. by computing a determinant).

Remark $\ $ Such "determinant tricks" generalize to arbitrary rings. More efficiently (and precisely), one can use Grobner bases to do the elimination (e.g. via ideal-contraction).

Mechanically, we can use resultants to eliminate $\,z,y\,$ from $\,x = y+z,\ x^2=2,\ y^3 = 3.$ First eliminating $\,z\,$ we have ${\rm res}(x\!-\!y\!-\!z,z^2\!-\!2,z) = (x\!-\!y)^2\!-\!2,\,$ then eliminating $\,y\,$ we have $\,{\rm res}((x\!-\!y)^2\!-\!2,y^2\!-\!3,y) = x^6\!-\!6x^4\!-\!6x^3\!+\!12x^2\!-\!36x\!+\!1\,$ (e.g. computed via Alpha)

Alternatively we can contract the ideal $\,(x\!-\!y\!-\!z,z^2\!-\!2,y^3\!-\!3)\,$ to $\,\Bbb Q[x],\,$ using Grobner bases.

The same methods generalize to roots of higher degree polynomials.


We have $\alpha=\sqrt{2}+\sqrt{3} \in K=\mathbb Q(\sqrt2, \sqrt3)$.

$\mathcal B = \{1, \sqrt2, \sqrt3, \sqrt6\}$ is a basis for $K$ over $\mathbb Q$.

Write the matrix of $\mathbb Q$-linear map $\mu : x \mapsto \alpha x$ with respect to $\mathcal B$.

Since $\mu^k : x \mapsto \alpha^k x$, if $p$ is a polynomial over $\mathbb Q$, then $p(\mu)x=p(\alpha)x$ for all $x$. Therefore, $\alpha$ is a root of every polynomial $p$ such that $p(\mu)=0$. In particular, $\alpha$ is a root of the characteristic polynomial of $\mu$ because of the Cayley–Hamilton theorem.

The matrix of $\mu$ is $$ \pmatrix{ 0 & 2 & 3 & 0 \\ 1 & 0 & 0 & 3 \\ 1 & 0 & 0 & 2 \\ 0 & 1 & 1 & 0 } $$ and its characteristic polynomial is $x^4-10 x^2+1$. Therefore, $\alpha=\sqrt{2}+\sqrt{3}$ is a root of that polynomial.

(Although not relevant here, this polynomial happens to be irreducible; see Show that $x^4-10x^2+1$ is irreducible over $\mathbb{Q}$.)

  • $\begingroup$ See also math.stackexchange.com/questions/1444989/…. $\endgroup$ – lhf Jun 28 '16 at 1:32
  • $\begingroup$ As @Bill mentioned, you can also use elimination; see WA. $\endgroup$ – lhf Jun 28 '16 at 1:46
  • $\begingroup$ I don't understand how you computed the matrix of $\mu$, could you please elaborate? $\endgroup$ – i9Fn Dec 30 '16 at 12:00
  • $\begingroup$ @i9Fn, as I compute the matrix of any linear transformation: by listing the images of each basis element as columns. In this case, compute $\alpha 1, \alpha\sqrt2, \alpha\sqrt3, \alpha\sqrt6$. $\endgroup$ – lhf Dec 30 '16 at 12:08

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.