5

Let $i$ be a root of the equation $ x ^2 + 1 = 0 $and let $ω$ be a root of the equation $x^2 + x + 1 = 0$. Construct a polynomial $f(x) = \sum_k a_k x^k$ with $a_k \in \mathbb{Z}$ such that $f(i+w)=0$. Source

After some long thought, I got a 'partial' construction for $f$: let $q(x) =x^2 +1$ and $p(x) = x^2 + x+1$, then:

$$ f(x) = q(x- \omega) + p(x- i)$$

However, the above doesn't necessarily satisfy the 'integer' coefficients criteria and may contain an imaginary part, so I modified the construction a bit to fix the second problem and got:

$$ f(x) = q(x- \omega) + p(x-i) + \overline{ p(x-i) + q( x - \omega)}$$

Problem: The above has no guarantee to have integer coefficients and the expression also has terms of $\overline{x}$ i.e: complex conjugate of $x$.

  • I don't want to do it for you, but here's a better approach: let $\alpha=i+w$, then $\alpha-i$ satisfies the quadratic $? x^2+ ? x+ ?$; now rearrange that to get an equation for $i$ and pop that into $X^2=-1$. – ancient mathematician Jul 05 '21 at 10:50
  • Sure, that is a construction but how is that guaranteed to give integer coefficients @ancientmathematician – tryst with freedom Jul 05 '21 at 10:52
  • Do you know Galois theory? – Aryaman Maithani Jul 05 '21 at 10:53
  • Hmm no, but I'd appreciate an answer based on it, I will look at the answer when I study it (eventually^tm) @AryamanMaithani – tryst with freedom Jul 05 '21 at 10:54
  • One constructive criticism of your question, which otherwise I regard as excellently presented: if I am interpreting your question correctly, (which may not be the case), then you are overloading the variable $i$ as both $\sqrt{-1}$ and the index of the coefficients of the polynomial. I suggest using a different variable (e.g. perhaps $k$) for the index of the polynomial coefficients, to help with the question's clarity. – user2661923 Jul 05 '21 at 10:55
  • Noted and fixed! @user2661923 – tryst with freedom Jul 05 '21 at 10:57
  • Great! I have added an answer which uses Galois theory to construct it. Even if you don't understand why the construction works, you still can compute the polynomial. (Although it would be understandably uninteresting if you can't see how you would apply this to other examples.) – Aryaman Maithani Jul 05 '21 at 11:25

3 Answers3

4

Following ancient mathematician's comment: $$ \alpha-i = \omega$$ and we know $\omega$ satisfies $x^2 + x+1=0$, upon substitution:

$$ (\alpha-i)^2 + ( \alpha -i ) +1=0$$

Simplifying:

$$\frac{ \alpha^2 + \alpha}{1+ 2 \alpha}= i$$

Then following the second suggestion from AncientMathematician (perhaps derived from their experience ), we square both sides:

$$ \frac{\alpha^4 + \alpha^2 + 2 \alpha^3}{1 + 4 \alpha^2 + 4 \alpha } =-1$$

Finally:

$$ \alpha^4 + 5 \alpha^2 + 4 \alpha + 1 + 2 \alpha^3= 0$$

If I had made no algebra mistakes, this should be the construction.

  • To be explicit: I sort of don't like this approach because as I see it , it was just luck that we ended up in a polynomial with integer co-efficient – tryst with freedom Jul 05 '21 at 11:02
  • Of course it wasn't luck! It's clear that all the steps are at least rational, and if the final coefficients are not integers just multiply up the the lcm of the denominators. – ancient mathematician Jul 05 '21 at 11:38
  • 1
    But here's a guaranteed way of finding a polynomial satisfied by $\alpha+\beta$ where $\alpha$ is a root of $f$ of degree $m$ and $\beta$ a root of $g$ of degree $n$: just compute (Smith N Form) the characteristic polynomial of $I_n\otimes C(f)+C(g)\otimes I_m$ where $C(f)$ is the companion matrix of $f$ etc. – ancient mathematician Jul 05 '21 at 11:46
  • 1
    Good point in the first comment, I didn't actually think of that. I think it was that I felt the pieces fell into place tooo perfectly haha @ancientmathematician. I don't know much abstract algebra yet, but I'll hopefully return to this comment and Aryaman's comment when I learn – tryst with freedom Jul 05 '21 at 16:54
2

$\newcommand{\Q}{\Bbb Q}$Consider the tower of fields $$\Q \subsetneq \Q(\iota) \subsetneq \Q(\iota, \omega).$$ It is easy to see that each inclusion is strict. Since $\iota$ and $\omega$ both satisfy quadratic equations over $\Q$, it follows that each extension above is of degree $2$. Thus, there are exactly four $\Q$-embeddings $\sigma_1, \ldots, \sigma_4 : \Q(\iota, \omega) \to \Bbb C$. These are given as (taking combinations of):

  1. Send $\iota$ to either $\iota$ or $-\iota$,
  2. Send $\omega$ to either $\omega$ or $\omega^2$.

We can do these independently, thanks to the tower above and the fact that the irreducible polynomial of $\omega$ over $\Q(\iota)$ is the same as that over $\Q$.

Now, we have the element $\alpha = \iota + \omega \in \Q(\iota, \omega)$. It has four Galois conjugates: $\sigma_1(\alpha), \ldots, \sigma_4(\alpha)$. Then, the polynomial $$f(x) = (x - \sigma_1(\alpha)) \cdots (x - \sigma_4(\alpha)) \tag{1}$$ has coefficients in $\Bbb Q$. If need be, you can scale it up to have coefficients in $\Bbb Z$.


So even if you don't understand why the above works, here is how you get the polynomial: Expand out $(1)$. Here are the values that you can use:

  1. $\sigma_1(\alpha) = \iota + \omega$,
  2. $\sigma_2(\alpha) = -\iota + \omega$,
  3. $\sigma_3(\alpha) = \iota + \omega^2$,
  4. $\sigma_4(\alpha) = -\iota + \omega^2$.

Note that $\sigma_1(\alpha) = \overline{\sigma_4(\alpha)}$ and $\sigma_2(\alpha) = \overline{\sigma_3(\alpha)}$, if that helps with calculations.


Side note: the fact that all the $\sigma_i(\alpha)$ turn out to be distinct tells us that we can't do better than a four degree polynomial.

1

Multiplying $\,\alpha=\omega+i\,$ by $\,i\,$ and using that $\,i^2=-1\,$:

$$ \begin{align} \begin{cases} \alpha &= \omega+i \\ i \,\alpha &= i \omega - 1 \end{cases} \end{align} $$

Eliminating $\,i\,$ between the two equations, for example by taking $\,i=\alpha-\omega\,$ from the first equation and substituting into the second one:

$$ \alpha^2 - 2\omega\alpha + \omega^2+1=0 $$

Using that $\,\omega^2=-\omega-1\,$:

$$ \alpha^2 - 2\alpha\omega - \omega=0 $$

Multiplying by $\,\omega\,$ and using that $\,\omega^2=-\omega-1\,$:

$$ \require{cancel} \begin{align} \begin{cases} \alpha^2 - \omega (2\alpha + 1) &= 0 \\ \omega\,\alpha^2 + (\omega+1)(2\alpha+1) &= 0 \end{cases} \;\;\iff\;\; \begin{cases} - \omega\,(2\alpha+1) + \alpha^2 &= 0 \\ \omega\,(\alpha^2+2\alpha+1) + 2\alpha+1&=0 \end{cases} \end{align} $$

Eliminating $\,\omega\,$ between the two equations:

$$ (\alpha^2+2\alpha+1)\alpha^2+(2\alpha+1)^2 = 0 \;\;\;\;\iff\;\;\;\; \alpha^4 + 2 \alpha^3 + 5 \alpha^2 + 4 \alpha + 1 = 0 $$

With $\,P(x)=x^2+1\,$ and $\,Q(x)=x^2+x+1\,$, the above procedure is equivalent to calculating the polynomial resultant of $\,P(x)\,$ and $\,Q(\alpha-x)\,$, as verified in WA. A similar procedure can be used for higher degrees to produce a polynomial having as roots the sums of roots of $\,P\,$ and $\,Q\,$ with coefficients in the same ring as those of $P,Q$, in this case $\mathbb Z\,$.

dxiv
  • 76,497