The construction of a splitting field involves choices: which irreducible factor do we use first, and which root do we adjoin? To compare different constructions, equality is too strong: their elements may live in unrelated ambient fields. The right assertion is that a field isomorphism fixing the original coefficients identifies them. We first prove the one-root step on which this assertion rests.
Transporting a minimal polynomial
Definition 2.1.1
(Transported polynomial)
Let be a field isomorphism. For , write . Applying coefficient by coefficient is an isomorphism , so it preserves degrees, products and irreducibility. The symbol concerns coefficients; it does not assert that is already defined on any root of .
Theorem 2.1.2
(The one-root extension lemma)
Let be algebraic over , with monic minimal polynomial . Let be an isomorphism, and let be a root of in an extension of . Then there is a unique isomorphism extending and sending to .
Proof
(Proof)
Evaluation at has kernel : division by shows that a polynomial vanishing at is divisible by , since a nonzero remainder would have smaller degree than . Thus . Here the quotient is a field because is irreducible, and its image contains both and .
Since is irreducible and monic, it is the minimal polynomial of its root over . Coefficient transport induces an isomorphism
Composing with evaluation at the two roots gives the required isomorphism. Explicitly, it sends to ; the quotient argument proves this is well defined even if an element has several polynomial descriptions. It extends and sends to . Any such homomorphism has the same value on every polynomial in and is therefore unique.
In particular, when and is the identity, the possible images of under an -embedding of into a given extension are exactly the distinct roots of in . Each root gives one embedding, and every embedding sends to a root because it fixes the coefficients. Multiplicity does not count as a new image.
Worked example 2.1.3
(Two images of a square root)
The polynomial is irreducible over , and both roots lie in . The lemma gives exactly two -embeddings , namely
Both maps have image and hence are automorphisms. We did not simply substitute a sign into an arbitrary expression: irreducibility and the quotient construction ensure every relation is respected.
Common mistake
(A root must be present in the target)
For , is irreducible over , but its only real root is . There is just one -embedding . There are three into , whose images are , and , where is a primitive cube root of unity. An embedding into need not be an automorphism of the original subfield of .
A cyclotomic example by translation
Eisenstein's criterion cannot be applied directly to every polynomial, but a change of variable often exposes the pattern. The substitution is a ring automorphism of with inverse , so is irreducible over exactly when is.
Worked example 2.1.4
(The fifth cyclotomic polynomial)
Let , so that . Then . Since , we obtain .
The prime divides , does not divide the leading coefficient , and does not divide the constant term . The polynomial is primitive because it is monic. By Eisenstein's criterion is irreducible over , hence so is . Let be a complex number with and . Then and , so . Being a monic irreducible polynomial vanishing at , is its minimal polynomial, and with basis .
Extending through a finite tower
Theorem 2.1.5
(Extension into an algebraically closed field)
Let be a finite algebraic extension, an isomorphism, and an algebraically closed field containing . Then extends to an embedding .
Proof
(Proof)
Choose a finite -basis of and use its elements as generators: . Write , with . Inductively suppose is an embedding. Regard it as an isomorphism . The minimal polynomial of over is carried to an irreducible polynomial over , and it has a root in because is algebraically closed. The one-root lemma extends to . After finitely many steps the resulting map embeds in and extends .
The theorem guarantees an embedding into the stated large target, not into an arbitrary extension of . For example, cannot embed over into , since has no rational root. The finiteness assumption permits a finite chain of choices here; extending an embedding from an arbitrary algebraic extension requires a further argument.
Uniqueness of splitting fields
Theorem 2.1.6
(Splitting fields are unique up to base isomorphism)
Let be nonconstant, let be an isomorphism, and let and be splitting fields of and over and respectively. There is an isomorphism extending .
Proof
(Proof)
Choose an algebraic closure of ; its existence is proved in the unit on algebraic closures. A splitting field of a finite-degree polynomial is a finite extension, since it is generated by finitely many algebraic roots. By the finite-tower theorem, extends to an embedding . Every root of in maps under to a root of . As splits in , all these images lie in . They generate over , so .
Conversely, factor in , listing roots with multiplicity. Applying gives in . These are all the roots of in , with multiplicities, by uniqueness of factorization. In particular every root in lies in . Since is generated over by those roots, . Thus is onto as well as injective.
Proof X-ray
Two inclusions inside one ambient field
The key move is to place both fields in a single ambient field: is chosen to contain , and the finite-tower theorem supplies . After that, the proof is two inclusions with different ingredients.
The inclusion uses that is generated over by the roots of , and that splits in , so that every root of in already lies in . The reverse inclusion uses the other two halves of the definition: splits in , and transporting that factorization by lists all roots of in by unique factorization in ; then is generated over by those roots. Multiplicities are carried along rather than counted, which is why no separability assumption is needed.
The isomorphism is generally not unique. For over , the identity and the sign-changing automorphism are two valid choices.
Counterexample mode
Containing all the roots is not enough
Claim: any two extensions of in which splits are isomorphic over . Take , , and . Both fields contain the roots . If with rational , squaring gives , so and ; then forces and forces , and neither has a rational solution. Hence while , and an isomorphism fixing would preserve the -dimension. The repair is the second condition in the definition: a splitting field must be generated over the base by the roots. Here is strictly larger than the field generated by the roots.
Counting the isomorphisms by induction
The proof above uses an algebraically closed field containing . An induction on the degree avoids that construction, and it also counts the isomorphisms that extend .
Theorem 2.1.7
(Counting isomorphisms of splitting fields)
In the situation of the uniqueness theorem, the number of isomorphisms extending satisfies . If has no repeated root in , then .
Proof
(Adjoin one root of a nonlinear factor and induct)
We use induction on , for all fields, isomorphisms and polynomials at once. If , then and with all . Applying gives , so by unique factorization in the roots of in are the . As is generated over by these roots, , and itself is the only extension. Thus .
Now let . Since is generated over by the roots of , some root of is not in . Its minimal polynomial over divides and has degree . Then divides , which splits in , so splits in . Let be its distinct roots in , where . For each the one-root lemma gives an isomorphism extending with .
The field is a splitting field of over : splits in , and is generated over by the roots of . Likewise is a splitting field of over . Since , the induction hypothesis says that the number of isomorphisms extending satisfies .
Every isomorphism extending sends to a root of , hence to exactly one ; by the uniqueness part of the one-root lemma it then restricts to on . Conversely every extension of extends . Therefore , which gives .
Suppose finally that has no repeated root in . The polynomial divides and splits in , so it has distinct roots there. We already know that some isomorphism extends . It maps the distinct roots of to distinct roots of , so . The polynomial still has no repeated root when it is regarded over , so the induction hypothesis gives for every . Hence .
Taking , and gives a statement about automorphisms: if is a splitting field of over , then , with equality when has no repeated root in .
Worked example 2.1.8
(Six automorphisms without listing them first)
Let and let be a primitive cube root of unity. The splitting field of over is . Here , and is a nonreal root of while , so . The three roots are distinct. The theorem gives exactly six automorphisms of over . Each is determined by the images of and , and there are at most possible pairs of images, so every pair occurs.
The hypothesis on repeated roots is sufficient but not necessary. The polynomial has repeated roots, but its splitting field still has automorphisms, since it is also the splitting field of . By contrast, for over in the repeated-root test below, the count is while the degree is .
Counting embeddings through a tower
The finite-tower construction gives more than existence. It also explains why an embedding count cannot exceed the extension degree. At each stage, an embedding already chosen on the smaller field has only finitely many possible extensions: the image of the next generator must be a root of its transported minimal polynomial. The number of distinct roots is at most the polynomial's degree. Multiplying these bounds and using the tower law gives a global bound. This argument counts choices that respect the relations already imposed; it does not allow the images of different generators to be chosen independently without checking those relations.
Proposition 2.1.9
(A degree bound for embeddings)
Let be finite and let be a field containing . Then the number of distinct -embeddings is at most . No normality or separability assumption is required for this upper bound.
Proof
(Proof)
Write and put , . Every -embedding of restricts to one of . Fix an embedding of into . An extension to is determined by the image of , which must be a root in of the transported minimal polynomial over the image of . Its degree is , so it has at most that many distinct roots. Induction bounds the total number by the product of the successive degrees. The tower law identifies this product with . Some roots may be absent from the target, or several roots may coincide, so neither inequality in this counting argument need be equality.
Worked example 2.1.10
(Four compatible sign choices)
Let . The square-root calculation in the simple-extension unit shows that , so the tower has successive degrees two and two. Start with either embedding of into . The minimal polynomial of over that field is , and its coefficients are fixed by both embeddings. Both roots and are available in the target. The one-root lemma therefore supplies four extensions, one for each independent sign choice. The degree bound says there can be no fifth embedding. Each image is again , since changing signs does not change the generated subfield, so all four are automorphisms.
Worked example 2.1.11
(A dependent generator does not add choices)
Now present the same quadratic field redundantly as . Once an embedding sends to one of its two conjugates, it must send to the negative of that chosen image. There are only two embeddings, not four. Equivalently, the second adjunction has degree one: already lies in the first field. Multiplying the degree of twice would ignore the relation and give an incorrect extension degree, or an unnecessarily loose upper bound on the embedding count.
Common mistake
(Use the minimal polynomial over the current field)
Let be a primitive cube root of unity and . Over the second generator has minimal polynomial , but over it does not:
and is a root of the quadratic factor, which is irreducible over the real field because both its roots are nonreal. Using at the second step would permit sending to the image already chosen for , which no injective map can do, and would give the loose bound . The correct bound is .
The target field matters even for a separable polynomial. If the target is , an embedding can use only real roots, while also contains nonreal roots. The degree bound remains valid in either target, but need not be attained in both. Separability will later control repeated roots, and normality will control whether conjugates remain in the original extension. These are different questions: the number of available images, the equality of the image with the original field, and uniqueness of the splitting field should not be treated as the same assertion.
Using the comparison without identifying the models
Suppose two authors construct fields and by adjoining roots in different orders. To invoke uniqueness, first identify the same base field and polynomial, then check two separate conditions: the polynomial splits in each field, and each field is generated by all its roots. The theorem produces an isomorphism fixing the base, but it does not say that unrelated elements have literally become equal. A chosen isomorphism is an additional piece of data.
This distinction matters for calculations. If a root in one construction is sent to a particular root in the other, every polynomial relation over the base is carried to the corresponding relation for . A different base-field isomorphism may send to another conjugate and hence give a different comparison. The uniqueness theorem guarantees that a comparison exists; a root choice specifies which comparison is being used.
When using the bound, distinguish an upper estimate from an actual count. To prove an exact count, one needs both an upper bound and enough explicitly constructed embeddings. A degree-four extension can have fewer than four embeddings into a particular target; its degree records vector-space dimension, while the count records which conjugates the target can receive. In the four-sign example both ingredients were present: the root choices constructed four maps, and the degree bound ruled out all others.
A repeated-root test
Let and let . The polynomial is irreducible by Eisenstein at over , followed by Gauss's lemma. In a field containing , the characteristic- identity gives . Thus is a splitting field of degree , but there is only one -embedding of into any extension containing : the minimal polynomial has only one distinct root. Uniqueness of the splitting field does not require separability, and the degree is not in general the number of embeddings.
Into an algebraically closed target, the bound is attained exactly when every step of the tower has a minimal polynomial with distinct roots.
A natural next question
When is the embedding bound attained?
Into an algebraically closed target, the bound is attained exactly when every step of the tower has a minimal polynomial with distinct roots.
Let be algebraically closed. In the tower proof each step then offers exactly as many images as the transported minimal polynomial has distinct roots, and that number does not depend on the earlier choices. The count of -embeddings is the product of these numbers, so it equals exactly when no step has a repeated root. This is the separability condition developed later; the extension by with above has degree and one embedding.
Exercises and solutions
Quick check
(Let be the positive real root. Find the number of -embeddings of into and into .)
Solution · Answer
Eisenstein at proves irreducible, so it is the minimal polynomial of . Its four distinct complex roots are , and exactly the first two are real. The one-root lemma gives four embeddings into and two into .
Exercise 2.1.12
(Let be a splitting field of over , and let be an extension field of . Prove that every -embedding has image .)
CoreProof
Hint(1)
Show that permutes the finitely many roots of in .
Solution(4)
- 1
Roots go to roots
If with , then , because fixes the coefficients of .
- 2
The roots of in already lie in
Write with . If is a root, then , so for some . Hence maps the finite set of distinct roots of in into itself.
Because: splits in .
- 3
An injective self-map of a finite set is onto
Key step
Since is injective, .
- 4
Conclude from generation by roots
∎Because and fixes , we get .
Quick check
(Suppose is algebraic over and is a root of some polynomial with . Must define an -embedding ?)
Solution · Answer
No. Take , , and . An embedding fixing must carry to , which fails since . The correct condition is that be a root of the minimal polynomial of .
Quick check
(Explain why two splitting fields of the same polynomial over inside one algebraic closure are equal, although the isomorphism of the theorem need not be unique.)
Solution · Answer
The polynomial has a fixed set of roots in the common algebraic closure. Each splitting field is precisely the subfield generated over by that set, so the two subfields are equal. Distinct automorphisms of that one subfield can still fix the base and permute the roots, so the isomorphism is not unique.
Quick check
(Let . How many automorphisms does the splitting field of over have? Describe them by the images of and .)
Solution · Answer
The roots are , so the splitting field is : it contains , and and produce every root. Eisenstein at gives , and gives , so . The four roots are distinct, so the counting theorem gives exactly eight automorphisms. An automorphism sends to one of the four roots and to ; there are at most such pairs, so each pair , with occurs.