CONSTRUCTING ALGEBRAIC CLOSURES

KEITH CONRAD

Let K be a field. We want to construct an algebraic closure of K, i.e., an algebraic

extension of K which is algebraically closed. It will be built as the quotient of a polynomial
ring in a very large number of variables.
For each nonconstant monic polynomial f (X) in K[X], let its degree be nf and let
tf,1 , . . . , tf,nf be independent variables. Let A = K[{tf,i }] be the polynomial ring generated
over K by independent variables doubly indexed by every nonconstant monic f ∈ K[X]
and 1 ≤ i ≤ nf . This is a very large polynomial ring containing K.
Let I be the ideal in A generated by the coefficients of all the difference polynomials
nf
Y
f (X) − (X − tf,i ) ∈ A[X]
i=1
as f runs over nonconstant monic polynomials in K[X]. Working modulo I we have f (X) ≡
Q
i (X − tf,i ), so f (X) splits into linear factors in (A/I)[X]. We want to use a maximal
ideal in place of I since working modulo a maximal ideal would give a complete splitting of
every f (X) from K[X] over a field.
Lemma 1. The ideal I is proper: 1 6∈ I.
Proof. This will follow from the existence of a splitting field of any nonconstant polynomial
in K[X].
Suppose 1 ∈ I. Then we can write 1 as a finite sum m
P
j=1 aj cj , where cj ∈ I and aj ∈ A.
Each cj is a coefficient in some difference
nj
Y
(1) fj (X) − (X − tfj ,i ),
i=1
where fj (X) is monic in K[X] and nj = deg(fj ). There is a (finite) field extension L/K
Qnj
in which the finitely many fj ’s all split completely, say fj (X) = i=1 (X − rj,i ) in L[X].
(Some numbers in the list rj,1 , . . . , rj,nj might be repeated.) We can use the roots rj,i
of f1 (X), . . . , fm (X) to construct a ring homomorphism ϕ from A = K[{tf,i }] to L by
substitution: ϕ fixes K, ϕ(tfj ,i ) = rj,i for 1 ≤ i ≤ nj , and ϕ(tf,i ) = 0 if f is not one of the
fj ’s. Extend ϕ to a homomorphism A[X] → L[X] by acting on coefficients. The polynomial
in (1) is mapped by ϕ to
nj
Y
fj (X) − (X − rfj ,i ) = 0 in L[X],
i=1
so every coefficient
Pin (1) is mapped by ϕ to 0 in L. In particular, ϕ(cj ) = 0. Thus ϕ sends
the equation 1 = m j=1 aj cj in A to the equation 1 = 0 in L, and that is a contradiction. 

Since I is a proper ideal, Zorn’s lemma guarantees that I is contained in some maximal
ideal m in A. (I suspect I itself is not a maximal ideal, but I don’t have a proof of that.)
1
2 KEITH CONRAD

The quotient ring A/m = K[{tf,i }]/m is a field and the natural composite homomorphism
K → A → A/m of rings let us view the field A/m as an extension of K (ring homomorphisms
out of fields are always injective).
Theorem 2. The field A/m is an algebraic closure of K.
Qnf
Proof. For any nonconstant monic f (X)Q ∈ K[X] we have f (X) − i=1 (X − tf,i ) ∈ I[X] ⊂
m[X], so in (A/m)[X] we have f (X) = i (X − tf,i ), where tf,i denotes tf,i mod m. Each
tf,i is algebraic over K (being a root of f (X)) and A is generated as a ring over K by the
tf,i ’s, so A/m is generated as a ring over K by the tf,i ’s. Therefore A/m is an algebraic
extension field of K in which every nonconstant monic in K[X] splits completely.
We will now show A/m is algebraically closed, and thus it is an algebraic closure of K.
Set F = A/m. It suffices to show every monic irreducible π(X) in F [X] has a root in F .
We have already seen that any nonconstant monic polynomial in K[X] splits completely in
F [X], so let’s show π(X) is a factor of some monic polynomial in K[X]. There is a root α
of π(X) in some extension of F . Since α is algebraic over F and F is algebraic over K, α is
algebraic over K. That implies some monic f (X) in K[X] has α as a root. The polynomial
π(X) is the minimal polynomial of α in F [X], so π(X) | f (X) in F [X]. Since f (X) splits
completely in F [X], α ∈ F . 
Our construction of an algebraic closure of K is done, but we want to compare it with
another method to put the construction in context. The idea of building an algebraic
closure of K by starting with a large polynomial ring over K whose variables are indexed
by polynomials in K[X] goes back to Emil Artin. He used a large polynomial ring (somewhat
smaller than the ring K[{tf,i }] we started with above) modulo a suitable maximal ideal to
obtain an algebraic extension K1 /K such that every nonconstant polynomial in K[X] has
a root in K1 (not, a priori, that they all split completely in K1 [X]). Then he iterated
this construction with K1 in place
S of K to get a new algebraic extension K2 /K1 , and so
on, and proved that the union n≥1 Kn (or, more rigorously, the direct limit of the Kn ’s)
contains an algebraic closure of K [2, pp. 544-545]. With more work, treating separately
characteristic 0 and characteristic p, it can be shown [3] that Artin’s construction only needs
one step: K1 is an algebraic closure of K (so Kn = K1 for all n, which is not obvious in
Artin’s own proof). In other words, the following is true: if F/K is an algebraic extension
such that every nonconstant polynomial in K[X] has a root in F then every nonconstant
polynomial in F [X] has a root in F , so F is an algebraic closure of K. Theorem 2 and its
proof, due to B. Conrad, modifies Artin’s construction by using a larger polynomial ring
over K in order to adjoin to K in one step a full set of roots – not just one root – of each
nonconstant monic in K[X], rather than one root for each polynomial. This makes it easier
to prove the constructed field is an algebraic closure of K. A similar construction, using a
maximal ideal in a tensor product, is in [1, Prop. 4, p. A V 21].
At the end of the proof of Theorem 2, the polynomial f (X) in K[X] with α as a root
can be taken to be irreducible over K, so we could build an algebraic closure of K by
defining the ideal I using just the monic irreducible f (X) in K[X] rather than all monic
f (X) in K[X]; the proofs of Lemma 1 and Theorem 2 carry over with no essential changes
other than inserting the word “irreducible” in a few places. Finally, if we restrict the f in
the construction of I to run over the monic separable polynomials in K[X], or the monic
separable irreducible polynomials in K[X], then the field A/m turns out to be a separable
closure of K. The proof of Lemma 1 carries over with the fj being separable (or separable
irreducible), and in the proof of Theorem 2 two changes are needed: A/m is a separable
 CONSTRUCTING ALGEBRAIC CLOSURES 3

algebraic extension of K since it would be generated as a ring over K by roots of separable

polynomials in K[X], and we need transitivity of separability instead of algebraicity (if
F/K is separable algebraic then any root of a separable polynomial in F [X] is separable
over K).

References
[1] N. Bourbaki, “Algebra II: Chapters 4-7,” Springer-Verlag, New York, 1990.
[2] D. Dummit, R. Foote, “Abstract Algebra,” 3rd ed., Wiley, New York, 2004.
[3] R. Gilmer, A Note on the Algebraic Closure of a Field, Amer. Mathematical Monthly 75, 1968, 1101-1102.