# Uniqueness of prime-power fields

I'm still stuck on the proof of the following theorem. I've asked two questions so far to get to where I am even at this point.

Theorem: Let $p$ be a prime and let $n\in\mathbb{Z}^{+}$. If $E$ and $E'$ are fields of order $p^{n}$, then $E\cong E'$.

Proof: Both $E$ And $E'$ have $\mathbb{Z}_{p}$ as prime fields (up to isomorphism). By Corollary 33.6, $E$ is a simple extension of $\mathbb{Z}_{p}$ of degree $n$, so there exists an irreducible polynomial $f(x)$ in $\mathbb{Z}_{p}[x]$ such that $$E\cong \mathbb{Z}_{p}[x] / \langle f(x)\rangle$$

The next line stumps me again. "Since the elements of $E$ and $E'$ are exactly the roots of $x^{p^{n}} - x$, $f(x)$ is a factor of $x^{p^{n}} - x$ in $\mathbb{Z}_{p}[x]$."

The rest of the proof:

"Because $E'$ also consists of zeros of $x^{p^{n}} - x$, we see that $E'$ also contains zeros of the irreducible $f(x)$ in $\mathbb{Z}_{p}[x]$. Thus because $E'$ also contains $p^{n}$ elements, $E'$ is also isomorphic to $\mathbb{Z}_{p}[x] / \langle f(x)\rangle$."

It follows from work already done that elements of $E$ and $E'$ are zeros of $x^{p^{n}} - x$. But I don't follow how we get $f(x)$ as a factor of $x^{p^{n}} - x$ from this. This is actually what prompted my Question about algebraic field extensions as I thought that might be relevant; but it turns out not to be the reason.

-
I would say that "prime-power fields" is not a very good terminology, the term "finite fields" is more appropriate in my opinion (note that a field is finite iff it has number of elements that is a power of a prime) – Belgi Aug 4 '12 at 20:13

Are you missing the following bits?

1. The polynomial $f(x)\in \mathbb{Z}_p[x]$ of degree $n$ has a root in the field $E$. Because the field extension $E/\mathbb{Z}_p$ is Galois, and $f(x)$ is irreducible, all the roots of $f(x)$ are distinct and in $E$. So $$f(x)=(x-a_1)(x-a_2)\cdots (x-a_n)$$ for some elements $a_1,a_2,\ldots,a_n\in E$.
2. The elements of $E$ are exactly the zeros of the polynomial $p(x)=x^{p^n}-x$. In other words $$p(x)=\prod_{a\in E}(x-a).$$

The troubling claim follows from this. The zeros $a_i,i=1,2,\ldots,n,$ are among the zeros of $p(x)$, so $f(x)\mid p(x)$.

In particular, the polynomial $f(x)$ also has $n$ zeros in $E'$, because $p(x)$ has $p^n$ roots there, and the roots of $f(x)$ are among those.

Edit: Proving my first claim. This depends heavily on the properties of the so called Frobenius homomorhpism $F:E\to E, x\mapsto x^p$. This is a homomorphism, because obviously $F(1)=1$ and $F(xy)=(xy)^p=x^py^p=F(x)F(y)$ for all $x,y\in E$. Less obvious is that $F$ respects addition as well, i.e. $$F(x+y)=(x+y)^p=x^p+y^p=F(x)+F(y)$$ for all $x,y\in E$. This follows from the binomial formula together with the observation that the binomial coefficients ${p\choose i}$ are all divisible by $p$, when $1\le i\le p-1$.

From little Fermat it follows that $F(x)=x^p=x$ for all the elements $x$ of the subfield $\mathbb{Z}_p$. We need to also make the observation that $x^p=x$ only when $x\in\mathbb{Z}_p$. This is because the polynomial equation $x^p-x=0$ can have at most $p$ solutions in the field $E$, and we already found $p$ solutions.

So we assume that $f(x)=x^n+f_{n-1}x^{n-1}+f_{n-2}x^{n-2}+\cdots+f_1x+f_0\in \mathbb{Z}_p[x]$ is irreducible, and has a root $a_1$ in $E$ (=the coset of $x$ in $\mathbb{Z}_p[x]/\langle f(x)\rangle$). In other words $a_1\notin\mathbb{Z}_p$ and $$a_1^n+f_{n-1}a_1^{n-1}+\cdots f_1a_1+f_0=0.$$ Let's apply the mapping $F$ to this equation. Remember that $F(f_i)=f_i$ for all $i$. We get $$a_1^{pn}+f_{n-1}a_1^{p(n-1)}+\cdots f_1a_1^p+f_0=0,$$ or, upon inspection, $f(a_1^p)=0$. Because $a_1\notin\mathbb{Z}_p$, $a_1^p\neq a_1$. Therefore we have found another zero $a_2=a_1^p$ of $f(x)$ in $E$.

We can repeat the argument and keep finding roots of $f(x)$: $a_3=a_2^p$, $a_4=a_3^p$ et cetera. Because $f(x)$ can have at most $n$ roots in $E$, this sequence of roots will have to start repeating at some point. Because $F$ is injective (its kernel is trivial), the repetition must start from $a_1$, in other words $a_1=a_1^{p^k}$ for some $k, 2\le k\le n$.

The polynomial $$g(x)=(x-a_1)(x-a_1^p)\cdots (x-a_1^{p^{k-1}})$$ is stable under $F$, so its coefficients are in $\mathbb{Z}_p$. Furthermore, $g(x)\mid f(x)$. But $f(x)$ was irreducible, so we must have $g(x)=f(x)$, and $k=n$. But all the roots of $g(x)$ are distinct and in $E$ by construction. Therefore the same holds for $f(x)=g(x)$.

-
Part $1$ is novel to me. I'm not familiar with what it means to be Galois, but I suspect this may result in circular development since the point of the chapter was to prove (in the previous theorem) the existence of the finite field $GF(p^n)$ (which I believe is called the Galois Field?) for each prime $p$ and $n\geq 1$. Galois Theory is developed in a later section. Is there a more elementary way to verify that the roots of $f(x)$ are all contained within $E$? (It is clear now that this is what the problem reduces to.) – roo Aug 4 '12 at 16:39
@Kyle: Yes, there is another way, but it is essentially a way of doing Galois theory without calling it by that name. It will not be circular, though. Normally the uniqueness of a field of $p^n$ elements is presented as a consequence of the uniqueness of a splitting field of a polynomial ($p(x)$ in this case). If you want me to attempt to explain it, I can give it a try (in an hour or so, I need to do some sport next). – Jyrki Lahtonen Aug 4 '12 at 17:15
Thanks for being so generous with your time. I would appreciate that, and there is indeed no rush so please take your time. Just as a side note, splitting fields are also not defined until later in the book as well. Also, I've been a bit bothered by the fact that I feel like there is an easier argument (which probably doesn't work otherwise I'm sure the author would have used it instead). I will post it as an answer; perhaps you could point out my logic error if there is one? – roo Aug 4 '12 at 17:27
@Kyle: I added some background for item 1. Does this help? – Jyrki Lahtonen Aug 4 '12 at 20:44
It definitely helps, thanks for the time you put into that. I will still have to spend more some time reading this carefully. Thanks again I really appreciate what you have done for me here. – roo Aug 5 '12 at 18:43

If $E$ is a finite field with $q=p^f$ elements, then we know that its multiplicative group is cyclic (of order $q-1$). Thus the elements of $E$ are exactly the roots of $X^q-X$ in a (fixed) algebraic closure of $\Bbb F_p$.

Note that the polynomial $X^q-X$ has distinct roots (also) in characteristic $p$.

-

Note: This answer may very well be incorrect, but I cannot yet see why. It looks messy only because I am insisting on being explicit about identifications as I am still uncomfortable with ignoring them at this point.

Since $E$ and $E'$ have $p^{n}$ elements, they are degree $n$ finite extensions of appropriate subfields which are isomorphic to $\mathbb{Z}_p$ (any other degree would yield a cardinality different from $p^{n}$).

Roughly speaking, they are isomorphic because they are both $n$ dimensional vector spaces over isomorphic scalar fields.

To make this more precise, I will take a typical element of $E$ and and by converting coefficients and basis elements (via given isomorphisms) I will map it to an appropriate element of $E'$.

Proof: $E$ and $E'$ are degree $n$ finite extensions of appropriate subfields which are isomorphic to $\mathbb{Z}_p$. Denote these subfields by $E_p$ and $E_p'$. Then letting $\phi:E_{p}\to \mathbb{Z}_{p}$ and $\phi':E_{p}'\to \mathbb{Z}_{p}$ be the respective isomorphisms.

Then let $B = \{b_1, ..., b_n\}\subset E$ and $B' = \{b_1', ..., b_n'\}\subset E'$ be $n$-element bases for $E$ as a vector space over $E_p$ and $E'$ as a vector space over $E_p'$, respectively.

For each $x\in E$, write $x = \sum_{i=1}^{n}a_{i}b_{i}$ for some choice of $a_{i}\in E_{p}$. Then define the map $f:E\to E'$ by $f(x) = \sum_{i=1}^{n}a_{i}'b_{i}'$, where $a_{i}' = [\phi'^{-1}\circ \phi](a_{i})$.

-
This explains why they are isomorphic as additive groups, but I think you need to be more careful with the choice of basis in order for products to commute with $f$. – Erick Wong Aug 4 '12 at 21:08
Not only being more careful with that: it is not true that two fields over the same field with the same dimension over that subfield are isomorphic as fields, as the simple example $\,\Bbb Q(\sqrt 2)\,\,,\,\Bbb Q(\sqrt 3)\,$ , both over $\,\Bbb Q\,$ , shows. – DonAntonio Aug 5 '12 at 2:26