8

I was doing an exercise in Brzezinski's Galois Theory Through Exercises and needed to factor the polynomial $x^6+5x^2+x+1=x^6+x+1$ in $\mathbb{F}_5[x]$. Is there a quick way to do this?

I can see it has no roots and therefore no linear factors, so it must either factor as the product of a quadratic and a quartic or two cubics, but listing the irreducible quadratics and cubics mod $5$ is laborious. I found all of the irreducible quadratics, which are \begin{multline} x^2\pm 2, \, x^2+x+1, \, x^2+x+2, \, x^2+2x-2, \, x^2+2x-1, \\ x^2-2x+2, \, x^2-2x-2, \, x^2-x+1, x^2-x+2 \end{multline} and checked that none of these are factors, so it must factor as the product of two cubics, which I know I could find either by finding all irreducible cubics or by writing $$ x^6+x+1 = (x^3+ax^2+bx+c)(x^3+dx^2+ex+f) $$ and then equating coefficients.

I used a computer to perform the Berlekamp algorithm which yielded the factorisation $$ x^6+5x^2+x+1 = (x^3-2x^2-1)(x^3+2x^2-x-1) $$ but it would take me a very long time to do this by hand. Can anyone suggest any tricks that could be applied here?

Sammy Black
  • 25,273
A. Goodier
  • 10,964
  • It is not so difficult to compute the irreducible cubic polynomials over $\Bbb F_5$ by hand (because you only need to find a root or see that there is none). Then you can easily multiply those to see if you can get $x^6+5x^2+x+1$. – Dietrich Burde Apr 12 '23 at 16:07
  • @DietrichBurde Thanks for your comment. Aren't there rather a lot of irreducible monic cubics over $\mathbb{F}_5$? If I write the polynomial as $x^3+ax^2+bx+c$, $c\neq0$, there are $5\cdot5\cdot4=100$ possibilities to check. – A. Goodier Apr 12 '23 at 16:16
  • There are $p(p-1)$ monic irreducible cubics over $\Bbb F_p$, see here. So yes, we have $20$ of them for $p=5$, but they are already known. So you don't have to work for it. Finally, why not use a computer? Berlekamp is great! If you are required for an exam to do it by hand, then you will have a very special polynomial, where you don't have to compute too much. – Dietrich Burde Apr 12 '23 at 16:19
  • 5
    this is something really special about your polynomial: its roots satisfy $a^5=f(a):=-1-1/a$ and you can notice that $f(f(f(t)))=t$ and so every root of your polynomial gives (at most) a cubic extension – user8268 Apr 12 '23 at 16:20
  • @DietrichBurde That's useful to know, but the exercise is intended to be done by hand. The book has separate exercises to be done with a computer. – A. Goodier Apr 12 '23 at 16:24
  • 3
    I am not really convinced that applying "tricks" is so desirable. If it is a homework, I wouldn't be concerned to use the known list of cubic polynomials. For certain polynomials, you can apply a "trick", but in general it is not possible. – Dietrich Burde Apr 12 '23 at 18:25

2 Answers2

9

The observation made by user8268 is the key to the following paper and pencil factorization. Not ruling out the existence of other possibilities — that trick simply works like charm.


Let $\alpha$ be a root of $p(x)=x^6+x+1$ in some extension field of $\Bbb{F}_5$. By Galois theory it follows that $$ \alpha^5=\frac{\alpha^6}{\alpha}=-\frac{\alpha+1}{\alpha}=-1-1/\alpha $$ is another root. The projective linear trasformation (aka a Möbius transformation) $$f(t)=-1-1/t$$ is "famous" for having order three. Meaning that its third iterate is the identity: $f(f(f(t)))=t$. It follows that the Frobenius automorphism $\alpha\mapsto \alpha^5$ cycles back to $\alpha$ after three rounds. Therefore the minimum polynomial of $\alpha$ is (a factor of) $$ m(x)=(x-\alpha)(x-f(\alpha))(x-f(f(\alpha)))=(x-\alpha)(x+1+\frac1\alpha)(x+\frac1{1+\alpha}). $$ Here I used the fact that $f(f(t))=-1/(1+t)$.

Expanding $m(x)$ gives $$ m(x)=x^3+Ax^2+Bx-1, $$ where $A=(1+3\alpha-\alpha^3)/(\alpha+\alpha^2)$ and $B=(1-3\alpha^2-\alpha^3)/(\alpha+\alpha^2).$

How does this help the factorization? The key is that, upon inspection, we see that $A-B=3$. This narrows down the family of putative factors to those of the form $$ x^3+Ax^2+(A-3)x-1. $$

All of the above, together with the OPs observation that the polynomial has no roots in $\Bbb{F}_5$, implies that we are looking for a factorization of the form $$ x^6+x+1=(x^3+Ax^2+(A-3)x-1)(x^3+Cx^2+(C-3)x-1)\qquad(*) $$ with $A,C\in\Bbb{F}_5$. Looking at the terms of degree five (or the linear terms) on the R.H.S. of $(*)$ tells us that $A+C=0$, so $C=-A$. Expanding further, and looking at the quadratic terms we get the equation $$ 0=-A+(A-3)(C-3)-C=AC-4(A+C)+9=AC-1=-A^2-1. $$ This has two solutions $A=\pm2$ in $\Bbb{F}_5$. It is straight forward to check that this gives the desired factorization.


The same observation allows us to factor for example $g(x)=x^8+x+1$ in $\Bbb{F}_7[x]$. Testing shows that $g(2)=0=g(4)$. The above process, this time starting from $\alpha^7=f(\alpha)$, leads to a full factorization $$g(x)=(x-2)(x-4)(x^3-3x-1)(x^3-x^2+3x-1).$$ We see the same difference of three in the coefficients of the quadratic and linear terms of the two cubic factors.

See this older question and its answers for an earlier incarnation of this projective linear transformation in this context. There was no need to identify the cubic factors there. I only posted this because I had never noticed that difference of three. Undoubtedly that is a known fact given that this particular Möbius transformation appears regularly.

Jyrki Lahtonen
  • 133,153
  • 1
    Galois theory of rational function fields does imply that the coefficients of $m(x)\in F(\alpha)[x]$, in the case where $\alpha$ is transcendental over a base field $F$, belong to a subfield $K\subset F(\alpha)$ such that $F(\alpha)/K$ is a cyclic Galois extension. By Lüroth we know that $K=F(\beta)$ for some $\beta$ (also transcendental over $F$). Thus it was obvious that there would be relation between the coefficients of $m(x)$. I don't recall ever realizing exactly how simple the relation is :-) – Jyrki Lahtonen Apr 13 '23 at 07:37
  • +1 for a beautiful solution (and also because of being glad that my little observation could be used to actually solve the problem :) – user8268 Apr 13 '23 at 14:07
  • @user8268 Thanks. I feel a bit bad for stealing your well deserved starring role. I wasn't sure whether you want to post something about this, and decided to do it in the end. I have been criticized for posting "answers in the comments". – Jyrki Lahtonen Apr 13 '23 at 14:10
  • no worries in this case; answering in the comments is certainly criminal, but this time I just suggested where one might look – user8268 Apr 13 '23 at 14:30
  • 1
    So two of the irreducible cubics factor $x^5=f(x)$, and the other two $x^5 = f\circ f(x)$ ? I think I understood the solution, and it is lovely! – orangeskid Apr 21 '23 at 03:22
3

My approach is almost certainly not what the author had in mind, but this technique seems quite effective:

If $\alpha$ is a root of a cubic factor in some extension, then by Vieta the constant term of the cubic factor is $- \alpha \cdot \alpha^5 \cdot \alpha^{25} = -\alpha^{31}.$

So compute $x^{31}$ modulo $x^6+x+1$ in $\mathbb{F}_5[x]$: $$\begin{align*} x^6 &\equiv -x - 1\\ x^{30} &\equiv -x^5 -1 \\ x^{31} &\equiv -x^6 - x \\ &\equiv 1, \end{align*}$$

This means $-\alpha^{31} = -1$ for every root $\alpha$ of $x^6+x+1$, so both cubic factors have constant term $-1$. (This computation also rules out the possibility of quadratic or linear factors: all such factors divide $x^{25}-x$, and only $x-1$ divides both $x^{25}-x$ and $x^{31}-1$.)

So each cubic factor looks like $x^3 + xL(x) -1$, where $L(x)$ is a degree $1$ polynomial in $\mathbb{F}_5[x]$.

This means we are looking for linear polynomials $L_1(x)$ and $L_2(x)$ such that $$ \begin{align*} x^6 + x + 1 &= (x^3 + x\,L_1(x) - 1)\cdot (x^3 + x\,L_2(x)-1) \\ &= (x^3-1)^2 + x(x^3-1)\cdot(L_1(x)+L_2(x)) + x^2\cdot L_1(x)L_2(x). \end{align*} $$ It's worth noting that a decomposition of the form $$ (x^3-1)^2 + x(x^3-1)\cdot \text{ linear } + x^2 \cdot \text{ quadratic }$$ is unique, as the linear term can be determined from the $x^5$ and $x^1$ coefficients, and then the quadratic factor can be solved for. Here we have $$x^6 + x + 1 = (x^3-1)^2 + x(x^3-1)\cdot (-1) + x^2\cdot(x^2+2x)$$ so $$ \begin{align*} L_1(x) + L_2(x) &= -1 \\ L_1(x) \cdot L_2(x) &= x^2 + 2x \end{align*} $$ and we can use (for example) the quadratic formula to find $L_1$ and $L_2$: $$ \begin{align*} L_1(x), \; L_2(x) &= \frac{-1 \pm \sqrt{-4x^2-8x+1}}{2} \\ &= \frac{-1\pm\sqrt{x^2+2x+1}}{2} \qquad \text{ (reminder: coefficients are in $\mathbb{F}_5$)} \\ &= -2x, \; 2x-1. \end{align*} $$ So $x^6+x+1 = (x^3-2x^2-1)(x^3+2x^2-x-1)$ in $\mathbb{F}_5[x]$.