Motivation
Splitting fields and Galois groups are often discussed inside an algebraically closed field. It is not enough to assume silently that all the roots needed at every stage happen to live together. We will construct an algebraic closure of an arbitrary field, prove that it is unique up to an isomorphism fixing the base, and identify the foundational choice principle used by the construction.
Definition 2.2.1
(Algebraic closure)
An extension is an algebraic closure if every element of is algebraic over and is algebraically closed: every nonconstant polynomial over has a root in . A single root for each polynomial suffices, because after dividing out a linear factor one repeats the argument finitely many times. Thus every polynomial splits into linear factors over an algebraically closed field.
We use Zorn's lemma, the polynomial division algorithm, existence of irreducible factors, the fact that a maximal ideal has a field quotient, and transitivity of algebraicity in towers as background. The one-root embedding-extension lemma used later is proved in the companion embeddings unit. Zorn's lemma is a set-theoretic choice principle; this chapter does not derive it from the other field axioms.
Common mistake
(An algebraically closed field containing need not be an algebraic closure of )
The field contains and is algebraically closed, but it is not an algebraic closure of . The set of complex numbers algebraic over is countable, because is countable and each nonzero polynomial has finitely many roots, while is uncountable; hence contains elements transcendental over . Both conditions of the definition are needed. The algebraic numbers in do form an algebraic closure of : a complex root of a polynomial with algebraic coefficients is algebraic over by transitivity of algebraicity.
Adjoining roots of all base polynomials at once
For each monic nonconstant polynomial , introduce a distinct formal variable . Let
be the polynomial ring in this set of variables. Each individual polynomial in uses only finitely many variables. Let be the ideal generated by the elements for all such . In the quotient , the class of is a root of ; the challenge is to prove that the quotient does not collapse to the zero ring.
Theorem 2.2.2
(The simultaneous-root ideal is proper)
The ideal does not contain .
Proof
(Proof)
If , the definition of an ideal generated by a set gives a finite expression
Only finitely many variables appear in this expression. Starting from , successively adjoin a root of each of the finitely many polynomials . At each step choose an irreducible factor over the current field and quotient its polynomial ring by that factor; this produces a field containing a root. The resulting field is a finite algebraic extension of .
Evaluate each distinguished variable at the chosen root. Assign zero to every other variable appearing in the coefficients . Evaluation is a ring homomorphism from the finitely generated polynomial subring containing the displayed expression to the constructed field. The right side evaluates to zero, while the left side evaluates to one, a contradiction. Thus .
The proof uses only finitely many polynomials because any alleged expression for is finite. It does not assume that an infinite family of roots is already present in a common algebraic closure. Roots are assembled only for that finite list, precisely where the contradiction needs them.
Worked example 2.2.3
(A finite part of the construction)
Over , consider only the requirements and . Adjoin first and then a root of in a further extension. The evaluation , makes both generators vanish in . Hence no finite combination of those two generators with polynomial coefficients can equal one. The general proof applies the same idea to whatever finite list appears in a proposed contradiction.
Common mistake
(A quotient by a proper ideal need not be a field)
The fact that is proper only guarantees that is nonzero. It need not itself be a field. The next step enlarges to a maximal ideal; only then does the quotient become a field. The enlargement must remain proper so that the map from stays injective.
A field containing roots of every base polynomial
Proposition 2.2.4
(One simultaneous root extension)
For every field there is an algebraic extension containing a root of every nonconstant polynomial in .
Proof
(Proof)
By Zorn's lemma, the proper ideal lies in a maximal proper ideal of . For completeness, the chain condition is that the union of a chain of proper ideals containing is an ideal and is still proper: if it contained one, some ideal in the chain would already contain one. The quotient is a field.
The composite map is injective. Its kernel is an ideal of the field and cannot contain one, so it is zero. We identify with its image. For each monic nonconstant , the class of in is a root of because . Dividing any nonconstant polynomial by its leading coefficient covers the nonmonic case.
Finally, is algebraic. Every element of is represented by a polynomial involving finitely many variables . Their residue classes are algebraic over , since each is a root of its corresponding nonzero polynomial. A finite tower generated by algebraic elements is finite over , so every polynomial expression in these classes is algebraic over . This proves the claim even though the full polynomial ring has infinitely many variables.
Worked example 2.2.5
(Why finite support matters)
A residue class represented by depends on only two root classes. Both lie in the finite extension , so the expression is algebraic over . There may be infinitely many other variables in the ring, but none is needed to describe this particular element. The argument is elementwise; it does not assert that has finite degree.
Having one root of each polynomial over is not yet enough to make algebraically closed: a polynomial with coefficients in may not have its coefficients in . We therefore repeat the construction over the new field.
Iterating and taking a union
Theorem 2.2.6
(Existence of algebraic closures)
Every field has an algebraic closure.
Proof
(Proof)
Set . Inductively choose an algebraic extension containing a root of every nonconstant polynomial over , and let
The union of a nested sequence of fields is a field. Every element of lies in some and is algebraic over by transitivity of algebraicity through the finite tower .
Let be nonconstant. Its finitely many coefficients all lie in one common , since there are only finitely many of them and the fields are nested. By construction has a root in . Hence is algebraically closed and algebraic over , as required.
Worked example 2.2.7
(The first stage need not be the last)
Take . A field containing roots of every polynomial over contains , so in this special case it contains and is already algebraically closed if it is algebraic over ; this uses the fundamental theorem of algebra as an external fact. The general proof does not rely on that special fact. It iterates because the one-step construction by itself only promises roots for polynomials over the preceding field.
Common mistake
(A union must contain its coefficients together)
The reason a polynomial over appears at some finite stage is that it has finitely many coefficients and the fields form a nested chain. An arbitrary union of unrelated fields need not be a field, and an infinite list of elements of the union need not all lie in one finite stage. The proof uses exactly the finite-coefficient statement.
Extending an embedding through an algebraic extension
The same one-root lemma that compares finite splitting fields extends to arbitrary algebraic extensions when the target is algebraically closed. The finite-tower proof cannot simply be repeated a finite number of times when the extension has infinitely many elements; Zorn's lemma supplies a maximal compatible partial embedding.
Theorem 2.2.8
(Embedding extension into an algebraically closed field)
Let be algebraic and let be an embedding into an algebraically closed field . Then extends to an embedding .
Proof
(Proof)
Consider pairs where is a field and is an embedding extending . Order them by extension of both the field and the map. Every chain has an upper bound: take the union of its fields and the union of its compatible maps. The union is a field, and the map is a well-defined embedding because any two elements and their algebraic relations occur together in one member of the chain.
By Zorn's lemma there is a maximal pair . If , choose . It is algebraic over because it is already algebraic over . Transport its monic minimal polynomial over coefficientwise through . This irreducible polynomial has a root . The one-root lemma extends to an embedding sending to , contradicting maximality. Hence .
Proof X-ray
Where the two hypotheses enter the extension proof
Each hypothesis is used exactly once. Algebraicity of guarantees that an element has a minimal polynomial over at all, because it is algebraic over the smaller field . Algebraic closedness of supplies a root of the transported polynomial, and the one-root lemma converts that root into an extension of .
The key move is the use of Zorn's lemma. An infinite algebraic extension may need infinitely many one-root steps, and no list of them is given in advance. Maximality turns the statement that one more step is always possible into the statement that no step remains. The chain condition holds because checking that a map is a well-defined embedding involves finitely many elements at a time, and these lie in a single member of the chain. The argument proves existence only: the choice of the root is where different extensions arise.
Worked example 2.2.9
(The target condition matters)
The identity embedding of does not extend to an embedding because the target has no root of . It does extend into an algebraic closure of , where a root exists. The theorem gives existence of an extension, not uniqueness: the two roots of give two possible embeddings into a target containing both.
Counterexample mode
A transcendental extension cannot be embedded
Claim: an embedding into an algebraically closed field extends to every extension field of . Take , let be an algebraic closure of with the inclusion, and let be the rational function field.
Suppose extends . The element is algebraic over , so for some nonzero . Since fixes , it sends to . But is a nonzero element of , and a field homomorphism is injective. This contradiction shows that no extension exists. The repaired statement is the theorem as stated: must be algebraic, which is exactly where the proof found a minimal polynomial for each new element.
Uniqueness up to base isomorphism
Theorem 2.2.10
(Uniqueness of algebraic closures)
If and are algebraic closures, there is a field isomorphism fixing .
Proof
(Proof)
Extend the identity on to an embedding using the previous theorem. Its image is algebraically closed: any nonconstant polynomial with coefficients in the image lifts coefficientwise to one over , has a root there, and the image of that root is a root of the original polynomial.
Every element is algebraic over , and hence algebraic over the larger subfield containing . Its minimal polynomial over is irreducible and has a root in that algebraically closed field, so it has degree one. Therefore . The embedding is surjective and hence an isomorphism fixing .
The isomorphism is generally not unique. An algebraic closure of contains both roots of ; the one-root lemma and the extension theorem give automorphisms extending the identity on that send to either root. Uniqueness means that any two choices have an isomorphism over the base, not that there is only one such isomorphism.
Concept lensStructural
Largest algebraic, smallest algebraically closed
An algebraic closure of is the same as an algebraic extension with no proper algebraic extension, and also the same as an algebraically closed field with no proper algebraically closed subfield containing . For the first, an irreducible polynomial over of degree above one would give a proper finite extension. For the second, the elements of algebraic over form an algebraically closed subfield, as for the algebraic numbers in ; conversely the uniqueness argument shows that an algebraically closed with is all of .
Worked example 2.2.11
(An algebraic closure of a finite field)
Inside a fixed algebraic closure of , let be the roots of . The finite-field factorization theorem shows because . Every algebraic element over has some finite degree and lies in the field with elements; choosing makes , so the element belongs to . Hence the algebraic closure is the nested union . The stages are finite fields, but the union is infinite.
Practice and solutions
Quick check
(Why is the ideal generated by all proper, even though there are infinitely many generators?)
Solution · Answer
Any expression claiming is in the ideal is a finite combination, so it uses only finitely many generators . Their roots can be adjoined successively in a finite extension of , and evaluating at those roots (and zero elsewhere) would give , a contradiction.
Quick check
(Why does a maximal ideal containing that ideal give an extension field of ?)
Solution · Answer
The quotient is a field because is maximal. The kernel of is an ideal of the field , so it is zero or all of ; it cannot contain because is proper. Hence the kernel is zero and embeds in .
Quick check
(Why is the one-stage field not obviously algebraically closed?)
Solution · Answer
The construction guarantees roots only for polynomials whose coefficients lie in the original field . A polynomial over may use coefficients introduced in the extension, so a further stage, or another argument, is needed.
Quick check
(Why do all coefficients of one polynomial over the nested union lie in a common finite stage?)
Solution · Answer
There are finitely many coefficients, each belongs to some stage , and because the fields are nested the stage with the largest index among these finitely many contains them all.
Quick check
(In the embedding-extension proof, where is algebraic closedness of the target used?)
Solution · Answer
It guarantees that the transported minimal polynomial of a new element has a root in the target. The one-root lemma then extends the maximal partial embedding to , which gives the contradiction with maximality.
Quick check
(Why is an embedding between two algebraic closures over the same base automatically onto?)
Solution · Answer
The image is algebraically closed. Every element of the target is algebraic over the base and hence over the image, so its minimal polynomial over the image is irreducible with a root in the image; it therefore has degree one, and the element lies in the image.