\documentclass[10pt]{article}
\usepackage{amsfonts}
\usepackage{amsmath}
\usepackage{amssymb}
\newtheorem{define}{Definition}
%\newcommand{\Z}{{\bf Z}}
\usepackage{psfig}
\oddsidemargin=0.15in
\evensidemargin=0.15in
\topmargin=-.5in
\textheight=9in
\textwidth=6.25in
\begin{document}
\input{preamble.tex}
\lecture{6}{September 28, 2005}{Madhu Sudan}{Arnab Bhattacharyya}
In the last lecture, we saw an algorithm to find roots of a polynomial
in a finite field. In particular, we noticed that if a
polynomial $f(x) \in F_q[x]$ has any roots, then
gcd$(f(x),x^q-x)$ is not 1. This is true because $\prod_{\alpha \in
F_q} (x-\alpha) = x^q - x$ and, so, a nontrivial common
divisor between $x^q-x$ and $f(x)$ indicates that $f(x)$ has linear
factors. We can use this gcd test to determine irreducibility of a
cubic polynomial, since a cubic factors iff one of its factors is linear. But, in general, if
$f(x) \in F_q[x]$ is a monic, degree $d$ polynomial, how do
we determine if it is irreducible? We will answer this question in
today's lecture.
\section{Some Properties of Finite Fields}
\subsection{Order of a finite field}
Let $F$ be a finite field. Then, we claim that $|F|=p^t$ for some
prime $p$ and some positive integer $t$. Let us see why this is
true.
First of all, define the characteristic of a ring $R$ to be the
smallest integer $n$ such that for all $\alpha \in R$, $n \cdot \alpha
= 0$ (if such an $n$ exists). For a field $F$, the characteristic is
equivalently\footnote{Let $m$ be the
smallest integer such that $m\cdot 1=0$. Then, since $n\cdot 1 = 0$,
$n\geq m$. Conversely, for any $\alpha \in F$, $m \cdot \alpha = (m\cdot
1)\alpha = 0$ and, so, $m\geq n$. Therefore, $n=m$.} the smallest
integer $n$ such that $n\cdot 1 =0$. Now suppose $n$ is not prime.
Let $n = pq$ with $p\leq q < n$. Then, $p\cdot 1 \neq 0$ and $q \cdot
1 \neq 0$ but $(p\cdot 1)(q \cdot 1) = (pq)\cdot 1 = 0$, contradicting
the fact that $F$ is an integral domain. So, the characteristic of
$F$ is always a prime.
From the above, it easily follows that $\mathbb{Z}_p \subseteq F$; in other
words, $F$ is an extension field of $\mathbb{Z}_p$. We have the following
fact about extensions of finite fields:
\begin{claim} Let $K$ and $L$ be finite fields with $K \subseteq
L$. Then, $L$ can be viewed as a finite-dimensional vector space over
$K$.
\end{claim}
\begin{proof-idea}
Addition in the $K$-vector space is the addition law in $L$ and scalar
multiplication of an element $\alpha$ in $L$ by an element $c$ of $K$
is defined to be the product $c\alpha$ as multiplied in $L$. Since
$L$ is finite, $L$ must be finite dimensional over $K$.
\end{proof-idea}
So, there must exist a finite set of bases elements $b_1,\dots,b_t$ in
$L$ such that $L = \{\sum_i \alpha_i b_i | \alpha_i \in K\}$ and,
furthermore, if there exists $\alpha_1,\dots,\alpha_t$ such that
$\sum_i \alpha_i b_i = 0$, then $\alpha_1 =\cdots = \alpha_t = 0$.
Thus, if $F$ is a $t$-dimensional vector space over $\mathbb{Z}_p$, $|F|=p^t$
because there are $p$ choices for each of the $t$ $\alpha_i$'s. This
is what we wanted to show.
\subsection{Extension fields and subfields}
The following constrains the number of subfields of a field:
\begin{claim}Let $K,F,L$ be fields such that $K\subseteq F \subseteq
L$, then the dimension of $F$ over $K$ must divide the dimension of
$L$ over $K$.
\end{claim}
\begin{proof-idea}
Suppose $\{\alpha_i\}$ is a set of bases elements of $F$ over $K$ and
$\{\beta_j\}$ a set of bases elements of $L$ over $F$. Then,
$\{\alpha_i\beta_j\}$ is a set of bases elements of $L$ over $K$.
\end{proof-idea}
Let $L$ be a $t$-dimensional vector space over $K$. Then, if $|K| =
q$, $|L|=q^t$. The following claims that there is no other subfield
of $L$ isomorphic to $K$.
\begin{claim}
If $K \subseteq L$, then
\begin{itemize}
\item there exists a unique isomorph of $K$ in $L$
\item given $\alpha \in L$, one can tell if $\alpha$ is in $K$ or not
\end{itemize}
\end{claim}
\begin{proof-idea}
Every $\alpha \in K$ satisfies $\alpha^q = \alpha$, and at most $q$
elements in $L$ satisfy this equation. So, there cannot be another
subfield of $L$ of order $q$, and $\alpha \in K$ if and only if
$\alpha^q = \alpha$.
\end{proof-idea}
The next question that we want to address is whether there are
efficient ways to go from the larger field, $L$, to the smaller field,
$K$. In general, we don't know any good way of enumerating all the
elements of $K$ other than by checking each element of $L$ to see if
it is a fixed point of the map $x \rightarrow x^q$. But there are
efficient maps from $L$ to $K$:
\begin{enumerate}
\item \textbf{Trace}: Let $Tr_K:L \rightarrow L$ be the map defined by
\begin{equation*}
Tr_K(x) = x + x^q + x^{q^2} + \cdots + x^{q^{t-1}}
\end{equation*}
\begin{claim}
Then,
\begin{itemize}
\item For all $\alpha \in L$, $Tr_K(\alpha) \in K$
\item For all $a,b \in K$ and $\alpha,\beta \in L$,
$Tr_K(a\alpha+b\beta) = a\cdot Tr_K(\alpha) + b\cdot Tr_K(\beta)$
\end{itemize}
\end{claim}
\begin{proof}
Let us prove the second part first. First of all, note that for any
$x$ and $y$ in $L$ and for any positive integer $i$, $(x+y)^{q^i} =
x^{q^i} + y^{q^i}$; this fact is used very frequently in finite field
calculations. So,
\begin{align*}
Tr_K(a\alpha+b\beta) &= \sum_{i=0}^{t-1}(a\alpha+b\beta)^{q^i} \\
&= \sum_{i=0}^{t-1}a^{q^i}\alpha^{q^i} +
\sum_{i=0}^{t-1}b^{q^i}\beta^{q^i}\\
&= \sum_{i=0}^{t-1}a\alpha^{q^i} +
\sum_{i=0}^{t-1}b\beta^{q^i}\\
&= a\cdot Tr_K(\alpha) + b\cdot Tr_K(\beta)
\end{align*}
To see that $Tr_K(\alpha) \in K$ for all $\alpha \in L$, note that
\begin{align*}
(Tr_K(\alpha))^q &= \alpha^q+\alpha^{q^2}+\cdots+\alpha^{q^t} \\
&= \alpha^q+\alpha^{q^2}+\cdots+\alpha^{q^{t-1}} +
\alpha \\
&= Tr_K(\alpha)
\end{align*}
\end{proof}
\item \textbf{Norm: } Let $N_K:L\rightarrow L$ be the map defined by
\begin{equation*}
N_K(x) = x^{1+q+q^2+\cdots+q^{t-1}}
\end{equation*}
\begin{claim}
Then,
\begin{itemize}
\item For all $x,y\in L$, $N_K(x\cdot y) = N_K(x) \cdot N_K(y)$
\item $N_K(0) = 0$ and $N_K$ maps $L^\ast$ to $K^\ast$
\end{itemize}
\end{claim}
\begin{proof-idea}
Similar to above proof of the properties of trace.
\end{proof-idea}
\end{enumerate}
\section{Representing Extension Fields}
Suppose we have a way to represent elements of field $K$, and we would
to like to represent elements of an extension field $L \supseteq K$.
We will describe three representations, the last of which, using
irreducible polynomials, is the most standard one.
\subsection{(Partial) Representation 1}
We argued above that $L$, a field of order $q^t$, is isomorphic
to a $t$-dimensional (additive) vector space over $K$, a field of order $q$. So,
one way to represent an element $\alpha\in L$ is as a vector $V_\alpha
\in K^t$. This is suitable for addition because $V_{\alpha+\beta} =
V_\alpha + V_\beta$, but it is not suitable for multiplication.
\subsection{Representation 2}
Let us modify the above representation in order to encode the field
multiplication table. For each $\alpha \in L$, consider the linear map
$A_\alpha:K^t\rightarrow K^t$ defined to be $A_\alpha(V_\beta) =
V_{\alpha\beta}$. Also, note that $A_\alpha(V_\beta + V_\gamma) =
A_\alpha V_\beta + A_\alpha V_\gamma$, satisfying the distributive
law.
Each $A_\alpha$ can be encoded
by a $t$-by-$t$ matrix over $K$. Hence, this representation is
suitable for doing addition and multiplication over $L$.
However, it is often the case that we cannot afford the $O(t^2)$
elements required to represent each element in $L$.
\subsection{Representation 3}
We will show that for all positive integers $t$, there exists a polynomial $g(x) \in
K[x]$ of degree $t$ that is monic and irreducible. Furthermore, $L
\cong K[x]/(g(x))$. Thus, an element of $L$ can be represented as a
polynomial in $K[x]$ of degree less than $t$.
\begin{claim} Given a field $K$ of order $q$ and any positive integer
$t$, there exists a field $L\supseteq K$ of order $q^t$.
\end{claim}
\begin{proof-idea}
Consider the polynomial $f(x) = x^{q^t} -x$. Let $F$ be the extension field
of $K$ over which this polynomial splits completely into linear
factors. We can always find $F$ by repeatedly adjoining roots to
$K$. Now, we will show that $L = \{\alpha \in F |
\alpha^{q^t}-\alpha=0\}$ is the desired field of order $q^t$.
First of all, since $f'(x) = -1$ and, so, gcd$(f,f')=1$, $f(x)$ has no
repeated roots. Combining this with the fact that $f(x)$ has at most
$q^t$ roots leads us to $|L| = q^t$.
Next, we need to see that $L$ is a field. This is pretty
straightforward. If $\alpha, \beta \in L$, then clearly $\alpha\beta$
and $\alpha^{-1}$ are in $L$; also, $\alpha+\beta \in L$ because
$(\alpha+\beta)^{q^t} = \alpha^{q^t} + \beta^{q^t}$ and $-\alpha \in
L$ because $(-1)^{q^t} = -1$ for odd $q$ and $-1=1$ for a field of
characteristic 2.
\end{proof-idea}
It is also true that any two fields of the same order are isomorphic,
although we shall not prove this claim here. (Therefore, in many places, we
will say ``a field'' where ``the field'' is also true.)
Next, we define the concept of a minimal polynomial $g_\alpha(x) \in K[x]$ for an
$\alpha \in L$. If $\alpha \in K$, then its minimal polynomial is
$g_\alpha(x)=x-\alpha$. Otherwise, if $\alpha \in L-K$, consider the smallest set
$\{1, \alpha, \alpha^2,\dots,\alpha^d\}$ such that there exists
$c_0,\dots,c_d\in K$ with $\sum_{i=0}^{d} c_i \alpha^i = 0$ and $c_d =
1$. Then
call $g_\alpha(x) = \sum_{i=0}^d c_i x^i \in K[x]$ the minimal
polynomial for $\alpha$. Note that $d \leq t$ because there can be at
most $t$ linearly independent elements in $L$ regarded as a $K$-vector
space.
\begin{claim}
For any $\alpha \in L$, $g_\alpha(x)$ is an irreducible polynomial
over $K$.
\end{claim}
\begin{proof}
Suppose $g_\alpha(x)$ is not irreducible; that is, $g_\alpha(x) =
h_1(x) h_2(x)$ for some $h_1, h_2 \in K[x]$ and $\deg(h_1)\leq
\deg(h_2) < \deg(g_\alpha)$. Then, since
$g_\alpha(\alpha) = 0$ and $K[x]$ is an integral domain, $h_1(\alpha)
= 0$ or $h_2(\alpha) = 0$. But this is a contradiction to the
minimality of $d=\deg(g_\alpha)$.
\end{proof}
Next, we show that there do exist irreducible polynomials of degree
$d$ for all positive integers $d\leq t$.
\begin{claim}
For any field $K$, there exist at least $\frac{q^d-1}{d}-q^{d/2}$
irreducible polynomials in $K[x]$ of degree $d$.
\end{claim}
\begin{proof}
Consider a field $L$ of order $q^d$ as guaranteed by Claim 6. Then,
we can construct minimal polynomials from each nonzero $\alpha \in L$.
Some of these polynomials may be the same, but since a polynomial of
degree $d$ has at most $d$ roots, each polynomial can repeat at most
$d$ times. Hence there are at least $\frac{q^d-1}{d}$ irreducible
polynomials of degree less than or equal to $d$. Next, note that the
degree of each irreducible polynomial must divide $d$. This is so,
because if an irreducible polynomial $g$ has degree $r$, then
$K[x]/(g(x))$ has dimension $r$ over $K$ and hence, by Claim 2, $r$ must divide
$d$, the dimension of $L$ over $K$. Therefore, the next highest
degree of an irreducible polynomial after $d$ is $d/2$ and there are a
total of $q^{d/2}$ polynomials of degree $d/2$. So, the number of
irreducible polynomials of degree strictly $d$ is at least
$\frac{q^d-1}{d}-q^{d/2}$.
\end{proof}
The above claim should cover most combinations of $q$ and $d$; for
those not covered, the requisite irreducibles are listed in some book!
Note that this claim is analogous to the version of the prime-number
theorem which states that the probability that an integer of $n$ or less
bits is prime is approximately $1/n$.
So, we have shown that given an extension field $L$ of dimension $t$
over $K$, it is always possible to find an irreducible polynomial
$g(x) \in K[x]$ of degree $t$. Also, it is clear that $K[x]/(g(x))$
is a field of order $q^t$ since $g(x)$ is irreducible. By the
isomorphism theorem mentioned above, $L \cong K[x]/(g(x))$. Thus, we
can always represent a field element of $L$ as a polynomial in $K[x]$
modulo an irreducible $g(x)$.
\section{Testing Irreducibility}
Given a $g(x) \in K[x]$ that is monic and of degree $d$, how do we
efficiently decide if it is irreducible? We have now acquired the
tools necessary to give the algorithm.
\begin{claim}
If $g(x) \in K[x]$ is irreducible of degree $d$, then $g(x) \mid x^{q^d}
-x$.
\end{claim}
\begin{proof}
Let $L = K[x]/(g(x))$. (Note this is one of those places where
introducing an extension field is useful even if there was no mention
of one in the original problem.) Then, if $\alpha \in L$,
$\alpha^{|L|} = \alpha^{q^d} = \alpha$. Representing $\alpha$ as a
polynomial $p(x)$ in $K[x]/(g(x))$, we have that $p(x)^{q^d} = p(x)$
(mod $g(x))$. Letting $p(x)=x$, we have what we wanted.
\end{proof}
\begin{claim}
If $g(x)$ is irreducible of degree $d$, then for all $d' < d$, $g(x)
\nmid x^{q^{d'}} - x$.
\end{claim}
\begin{proof}
Suppose $g(x) \mid x^{q^{d'}} -x$ for some $d' < d$. Then, consider a splitting field
$L$ of $x^{q^{d'}}-x$; $|L| = q^{d'}$. So, $g$ must have a root
$\alpha$ in $L$. Let $h_\alpha$ be the minimal polynomial in $K[x]$ for
$\alpha \in L$. Note that $\deg(h_\alpha) \leq d' < d = \deg(g)$; so,
$h_\alpha \neq g$. Now, because
$x-\alpha$ divides both $h_\alpha$ and $g$, gcd($h_\alpha, g$) is
nontrivial. Also, the gcd is a polynomial in $K[x]$, a contradiction
of the fact that $g$ and $h_\alpha$ are irreducible polynomials.
\end{proof}
Thus, we get the following algorithm for testing irreducibility:
\begin{itemize}
\item $d' \leftarrow d-1$
\item Decrement $d'$ until $d' = 1$
\begin{itemize}
\item If gcd($g(x),x^{q^{d'}}-x) \neq 1$, output {\tt reducible}
\end{itemize}
\item If $g(x) \mid (x^{q^d} -x)$, output {\tt irreducible}. Else,
output {\tt reducible}.
\end{itemize}
\noindent
If $g(x)$ is irreducible, then by the above two claims, the algorithms
outputs {\tt irreducible}. If $g(x)$ is reducible, say $g(x) =
h_0(x)h_1(x)$ where $h_0$ is irreducible of degree $d_0 < d$. Then,
by Claim 9, $h_0(x)$ divides $x^{q^{d_0}}-x$ and, so,
gcd($g(x),x^{q^{d_0}}-x) \neq 1$ and the algorithm returns {\tt reducible}.
\end{document}