Click on the colored portions to expand or collapse content.
Algebraic Closure
A polynomial $f\in K[X]$ splits if all its roots are in $K$, that is to say of $f$ factorizes in the form $f(X)= c(X-\alpha_1)\cdots(X-\alpha_n)$ where $c,\alpha_i\in K$.The fundamental example of algebraically closed field is $\C$. If $\Omega$ is algebraically closed containing $k$, then the set of elements of $\Omega$ algebraic over $k$ is a field, and this set is also an algebraic closure of $k$.
Example: Let $A=\{\alpha\in\C\text{ such that }\alpha$ algebraic over $\Q\}$, also known as the set of algebraic numbers in $\C$, then $A$ is a subfield of $\C$ and is algebraic closure of $\Q$, $A$ is denoted as $\wbr{\Q}$.
Lemma
Proof Let $f\in\wbr{k}[X]$ be a non constant polynomial. It suffices to show that there is a root in $\wbr{k}.$ The field $L$ generated by the coefficients of $f$ is of finite dimension on $k$, since the coefficients of $f$ are algebraic over $k$. Thus, the $k$ algebra $A=L[X]/\lr{f}$ is of finite dimension over $k$, where the dimension is $\deg f\cdot\dim_kL$ (via tower law). Consider the evaluation map $k[T]\ra A$, the morphism is never injective since $\dim_k(k[T])=\infty$. Thus there is a non zero $g\in k[T]$ which will get mapped to zero, or in other words $f|g$, in particular $g$ is non constant (since $\deg f>0$). But, $g$ has coefficients in $k$ and thus will split in algebraic closure $\wbr{k}$. Thus, it is the same for $f$ which divides $g$.
Steinitz Theorem
Note that the isomorphism whose existence is given by the theorem is far from unique as we will see: we can even prove that an algebraically closed field admits a infinity of automorphisms. We will first prove the existence, then the uniqueness which follows from fundamental theorem of extension of morphisms.
Let us start by building a huge algebra $A$ in which every polynomial has a root. Let $c(P)$ denote the dominant coefficient of every non zero polynomial. Start by considering the algebra of polynomials with large number of indeterminates (indeterminate for each possible root of $P$) $$ A=k[X_{P,i}] \text{ where }1\leq i\leq \deg P$$ Let $\gamma(i,P)$ denote the coefficients of the following polynomial in $X$, $$P(X)-c(P)\prod_{i=1}^{\deg P}(X-X_{P,i}) $$ Let $I$ be the ideal generated by $\gamma(i,P)$ with $P$ non constant. In case $P$ is a non zero constant, we have $\gamma(0,P)=0$.
$I\neq A$, else we would get a contradiction as we now show. Let $I=A$ then, $$\begin{equation}\label{q1}\sum_{j,P}Q_{P,i_j}\gamma(i_j,P)=1\text{ with }Q_{P,i_j}\in A \end{equation}$$ The coefficients $\gamma(0,P)$ of constant polynomial being zero, only contribute in the sum of Polynomials of strictly positive degree. Choose an extension of fields $K/k$ such that the non constant $P$ (that are finite in number, which we will match with $Q$ in $\eqref{q1}$ later) split with roots $(x_{P,i})$ with $1\leq i\leq\deg P$ in the extension $K$. Let $\phi:A\ra K$ be the morphism of $k$ algebras sending the $X_{P,i}$ to $x_{P,i}$ and other indeterminates to zero. Then $\phi$ induces a morphism $A[X]\ra K[X]$ that sends Polynomials $$P(X)-c(P)\prod_{i=1}^{\deg P}(X-X_{P,i}) $$ to $$P(X)-c(P)\prod_{i=1}^{\deg P}(X-x_{P,i}) $$ by construction $$\phi(\gamma(i,P)) =0\text{ for all }i.$$ Plugging this in $\eqref{q1}$ we can deduce that $0=1$ in $K$, leading to a contradiction. Hence, the ideal is always proper.
Let $J$ be a maximal ideal of $A$ containing $I$ (use Zorn's Lemma for construction) and $L$ be the field $A/J$. By construction, all non constant polynomials $P\in k[X]$ split in $L$, the roots being the images of $X_{P,i}$. All these roots are algebraic on $k$. But they generate $L$ as $k$-algebra. So we have $L$ algebraic on $k$.
Extension of Morphisms
Proof Let $E$ be a non empty subset of couple $(L,\sigma)$ where $L$ is a subfield of $K$ containing $k$ and $\sigma$ is an embedding $\sigma:L\ra \Omega$ making $\Omega$ an $L$ algebra. The extension of the embedding defines an order on $E$ that is clearly an inductive set. Let then $(L,\sigma)$ be a maximal element.
We now show $L=K$. Let $x\in K$, as $x$ is algebraic over $k$ it is also on $L$. Let $P(X)=\sum_ia_iX^i$ be the minimal polynomial of $x$ on $L$ so that the evaluation at $x$ is identfied with $L[X]/\lr{P}=L[x]$. Let $y$ be a root of $P^\sigma(X)=\sum\sigma(a_i)X^i$ in $\Omega$. There exists a unique $L$ morphism $L[X]/\lr{P}\ra\Omega$ that sends $X$ to $y$ because the image of $P$ in $\Omega$ is by definition $P^\sigma(y)=0$, where the $k$ morphism $L[x]\ra\Omega$ extends $\sigma$. By, maximality of $L$, we deduce $x\in L$.
Remark: $σ$ allows us to identify $L$ with $σ (L)$. From now on, we will not distinguish between $L$ and $σ (L)$. Note also that if $K$ is a finite extension, then the notion of dimension makes it possible to show the existence of $L$ without invoking Zorn's lemma.
Uniqueness of Algebraic Closure
Proof Consider $K_1$ as algebraic and $K_2$ as algebraically closed, the theorem above assures the existence of an embedding of $K_1$ in $K_2$. With the preceding notations, the choice of an embedding of $K_1$ in $K_2$ permits to see $K_1$ as a subfield of $K_2$. Now consider $(k,K,\Omega)$ and $(K_1,K_2,K_1)$, we deduce the existence of $\tau\in\Hom_{K_1}(K_2,K_1)$, in other words the following diagram is commutes.
Proof If $y\in\Omega$ is a conjugate of $x$, there exists $\sigma\in \Hom_k(k[x],\Omega)$ such that $\sigma(x)=y$ (this follows from the lemma which gives the correspondence $\Root_f\Omega\leftrightarrow \Hom_k(k_f,\Omega) $, in our case $k_f=k[X]/P=k[x]$ where $x$ is root of the polynomial $P$). Now we just need to extend $\sigma$ to all of $K$, which is possible by the Theorem. Inversely, if $\sigma\in\Hom_k(K,\Omega)$, it leaves minimal polynomial of $x$ on $k$ invariant. Thus, it only permutes the roots, which are the conjugates of $x$ by definition.
Let $K$ be a sub field of $\Omega$ algebraically closed. Let $P$ be a polynomial of $k[X]$ not necessarily irreducible. The
The philosophy is to fix an algebraic closure in which we work, this then allows us to talk about the field of roots.
The reference for this page is INTRODUCTION À LA THÉORIE DE GALOIS by Yves Lazlo.