A field automorphism fixing preserves every polynomial relation with coefficients in . Thus it can send an algebraic element only to another root of its minimal polynomial. The central insight of Galois theory is that, for the right extensions, these symmetries also recover all intermediate fields.
Motivation: from subfields to subgroups
Suppose we are given a finite extension and asked to list every field with . Viewed directly, this is a question about -subspaces of the vector space that happen to be closed under multiplication and inverses. There is no obvious finite procedure for finding them, and it is not even clear in advance that there are only finitely many. Galois theory answers the question by replacing the field-theoretic problem with a group-theoretic one. When the extension has enough symmetry, the intermediate fields are in bijection with the subgroups of a finite group, and listing the subgroups of a finite group is a terminating computation.
The mechanism is a pair of opposite operations. Given an intermediate field , we can ask which symmetries of leave every element of untouched. Given a set of symmetries, we can ask which elements of every one of them leaves untouched. The first operation produces a subgroup from a field, and the second produces a field from a subgroup. Each operation reverses inclusion, because a larger field imposes more conditions on a symmetry and a larger set of symmetries imposes more conditions on an element. The content of the fundamental theorem is that, for finite Galois extensions, these two operations are mutually inverse.
This unit has four aims. First, we make the two operations precise and record the elementary fact that automorphisms permute roots. Second, we prove that splitting fields of separable polynomials are Galois, relate conjugate intermediate fields to conjugate subgroups, and deduce the fundamental theorem from Artin's fixed-field theorem. Third, we compute completely in two families of examples: the abelian extension and its generalization, and the non-abelian splitting field of , whose group is dihedral of order . The splitting field of , with group , is treated in the unit on cubic and quintic splitting fields. Fourth, we examine what goes wrong when the hypotheses are weakened, so that the word "Galois" is seen to carry real content.
Automorphisms and fixed fields
Definition 6.1.1
(Galois group and fixed field)
For an extension , is the group of field automorphisms of fixing pointwise. For a subgroup of this group, its fixed field is . A finite extension is Galois when it is both normal and separable.
Two small verifications make these objects legitimate. The set is closed under composition and inversion, so it is a group under composition of maps. For the fixed field, each is a ring homomorphism, so if and then and , and if then . Hence is a subfield of . Because fixes pointwise, contains , so it is an intermediate field.
Both operations reverse inclusion. If , then any automorphism fixing pointwise also fixes pointwise, so . If , then an element fixed by every member of is fixed by every member of , so . The top field therefore corresponds to the trivial group, and the bottom field should correspond to the whole group.
Automorphisms permute roots
The opening sentence of this unit is a precise statement that we now prove, because every computation below depends on it.
Theorem 6.1.2
(Roots are permuted and generators determine automorphisms)
Let be a field extension, let , and let . If is a root of , then is also a root of . Consequently, if , then is determined by the -tuple . In particular, if is finite, then is finite.
Proof
(Automorphisms fix the coefficients)
Write with . Since fixes each and respects sums and products, . For the second assertion, every element of is a polynomial expression in the with coefficients in , and acts on such an expression by fixing the coefficients and replacing each by . Every element of is a quotient of two such expressions with , and . Hence the tuple of values determines . Finally, if is finite, choose an -basis of ; each is algebraic over , so it is a root of a nonzero polynomial over and has only finitely many possible images in . There are therefore finitely many possible tuples, hence finitely many automorphisms.
This result gives a concrete way to describe a Galois group: list the generators, list the admissible images of each generator (roots of its minimal polynomial lying in ), and then decide which combinations actually extend to automorphisms. The decision step is where the order formula for Galois extensions, used below, becomes valuable: it tells us how many combinations must survive.
Splitting fields and conjugate fields
Two facts connect the definition of a Galois extension with the computations below. The first says that the extensions one actually writes down, splitting fields, are Galois as soon as no irreducible factor has a repeated root. The second says how an automorphism moves an intermediate field and its subgroup together.
Theorem 6.1.3
(Splitting fields of separable polynomials are Galois)
Let be nonconstant, and suppose that each irreducible factor of in has distinct roots in a splitting field. If is a splitting field of over , then is Galois and . In particular, in characteristic zero every splitting field of a nonconstant polynomial is Galois.
Proof
(Count embeddings, then apply Artin's theorem)
Let be the roots of in , so that , put , and let be an algebraic closure of . We show by induction on that there are exactly embeddings fixing . For there is one. Let fix , let be the minimal polynomial of over , and let be the minimal polynomial of over , which up to a constant is an irreducible factor of . Then divides in , and transporting coefficients by shows that divides . Whether has a repeated root is decided by in , by the derivative test, so has distinct roots in as well. Hence , which splits in the algebraically closed field , has distinct roots there. By the one-root extension lemma of the embeddings unit, the extensions of to correspond to these roots, so there are exactly of them. The tower law completes the induction.
Each embedding fixing sends every to a root of , hence into , so . Being injective and -linear on the finite-dimensional space , it maps onto . Conversely every element of is such an embedding. Therefore has exactly elements. Artin's theorem gives , and , so by the tower law. Finally, the fixed-field unit shows that is normal and separable for every finite group of automorphisms of . Hence is Galois. In characteristic zero every irreducible polynomial has distinct roots, again by the derivative test, so the last sentence follows.
Proposition 6.1.4
(Conjugate fields and conjugate subgroups)
Let be finite Galois with group , let be an intermediate field, and let . Then . Moreover, is normal if and only if for every .
Proof
(Transport the fixing condition)
Let and . Then exactly when . So fixes pointwise exactly when , that is, when . This proves the first assertion.
Suppose is normal and let . The minimal polynomial of over has a root in , so it splits in ; since is one of its roots, . Hence , with equality because both have the same dimension over . Conversely, suppose for all , and let be irreducible with a root . Since is normal, splits in . For each root of in , the proposition on automorphisms moving roots within their orbit in the normality unit supplies with , so . Thus splits in , and is normal.
The fundamental theorem
Theorem 6.1.5
(Fundamental theorem of finite Galois theory)
Let be finite Galois, with . The maps and are inverse inclusion-reversing bijections between intermediate fields and subgroups . For corresponding , one has and . Moreover is Galois exactly when is normal in ; then restriction gives .
The theorem rests on two results proved in the fixed-field unit: a finite normal separable extension has exactly its degree many automorphisms, and Artin's fixed-field theorem says that a finite group of automorphisms of a field satisfies . Together with the conjugation proposition above, they yield the correspondence.
Proof
(Derivation from the two stated results)
For an intermediate , the extension remains normal and separable: the minimal polynomial over divides that over , which splits into distinct linear factors in . Thus . Let . Artin's theorem gives , so the tower law forces . Conversely, for , Artin's theorem gives . Since and the latter group has size at most the extension degree, equality holds. Inclusion reverses directly from the definitions, and the degree formulas follow by the tower law. For the final assertion, let correspond to . By the conjugation proposition, corresponds to . Since the correspondence is a bijection, is normal in exactly when for all , which by the same proposition means that is normal; is automatically separable, since is. Suppose this holds. Each maps onto , so restriction defines a homomorphism , , whose kernel consists of the elements fixing pointwise, namely . Its image has elements, while because is Galois. So restriction is onto and induces .
Below we also use Eisenstein's irreducibility criterion and the tower law.
In practice the degree formulas are the working tools. If we can guess a candidate field that is fixed by , and we can check that , then must equal , because and both have the same degree over . This "inclusion plus degree count" argument replaces tedious coordinate calculations, and it is used repeatedly below.
Worked example 6.1.6
(The two fields inside a quadratic extension)
The polynomial is irreducible over , and its splitting field is . The two automorphisms send to and . Its group has only the trivial subgroup and the whole group, so the correspondence gives exactly two intermediate fields: and . This is an explicit field statement, not merely a count of automorphisms.
Worked example 6.1.7
(A prime-degree finite-field extension)
For over , the Frobenius automorphism has order three and generates the Galois group. A group of prime order has no nontrivial proper subgroup, so the correspondence gives no intermediate field strictly between and . Equivalently, the tower law would force any intermediate degree to divide three.
An example
Worked example 6.1.8
(The biquadratic extension)
Take . It is the splitting field of and is separable in characteristic zero. Its degree is by the biquadratic theorem proved below. The order formula then gives four automorphisms, all determined by their sign choices on the two roots. The four sign changes are therefore automorphisms; they form the Klein four group. The three order-two subgroups fix respectively , , and . For instance, simultaneous sign change fixes ; the fixed field has degree two over , and already has degree two, so they coincide. The full group fixes and the trivial subgroup fixes .
The sketch above asserts that the degree is and that the lattice has exactly five members. The next section proves the general biquadratic result; a companion note checks fixed fields directly in coordinates.
A general biquadratic theorem
The example is a special case of a pattern that holds over any field of characteristic different from , provided three numbers fail to be squares. Stating the hypotheses exactly is instructive, since each one is used in the proof.
Theorem 6.1.9
(Biquadratic extensions)
Let be a field of characteristic different from , and let be such that none of , , is a square in . Let be an extension of with and . Then , the extension is Galois with , and the intermediate fields are exactly , , , , and . All of them are Galois over .
Proof
(Two quadratic steps and a degree count)
Since is not a square in , the polynomial has no root in and is therefore irreducible, so with basis . We claim is not a square in . Suppose with . Expanding gives , and independence of over forces . Because in , either or . If , then is a square in , contrary to hypothesis. If , then , so is a square in , again contrary to hypothesis. Hence is irreducible over and .
Next, is the splitting field of , whose roots are and . Since is not a square, , so , and since we have . The same holds for . Thus both quadratic factors are separable; by the theorem on splitting fields of separable polynomials, is Galois. By the order formula, for .
By the first theorem of this unit, each satisfies and with , and is determined by . Because is fixed by every automorphism in , the assignment is an injective group homomorphism . Both groups have order , so it is an isomorphism, and .
The Klein four group has exactly three elements of order and hence three subgroups of order . Let , where negates only ; let where negates only ; and let . The fixed field of contains , and has degree over , as does ; hence . Likewise . The subgroup fixes , because . The element has square , which is not a square in , so and . Again the degree count gives . The subgroups and give and . By the fundamental theorem these five subgroups yield all intermediate fields. Every subgroup of an abelian group is normal, so every intermediate field is Galois over .
Over with and the hypotheses reduce to the irrationality of , , , so the theorem recovers every claim in the earlier worked example.
Proof X-ray
How each hypothesis is consumed
Note how each hypothesis was consumed: " not a square" gave the first degree-two step, " not a square in " together with " not a square" gave the second, and "characteristic not " was used to separate from and to conclude . The identification of the three quadratic subfields then needs no coordinates: each candidate field is fixed by the relevant subgroup and has the right degree over , so the inclusion-plus-degree count makes it the whole fixed field.
A dihedral example: the splitting field of
In the biquadratic case the group is abelian, so every intermediate field is Galois. The splitting field of over is the standard example in which the group is non-abelian, the lattice is large, and normal and non-normal subgroups occur side by side.
Worked example 6.1.10
(The splitting field of over )
Let be the positive real fourth root of . The roots of are , so the splitting field is : it contains , and conversely and produce all four roots. By Eisenstein's criterion at , . Since does not contain the root of , . Hence , and by the theorem on splitting fields has order .
The group. An automorphism sends to one of the four roots of and to , and it is determined by these two images. There are combinations and automorphisms, so every combination occurs. Let and be the automorphisms with
Then , so has order , and , the restriction of complex conjugation, has order . Moreover and , so . For the eight elements act by
Thus is the dihedral group of order ; it permutes the four roots as the symmetries of the square with vertices in the complex plane.
The subgroups. By Lagrange's theorem a proper nontrivial subgroup has order or . The elements of order are and the four elements , since . So there are five subgroups of order . A subgroup of order either contains an element of order , and is then , or consists of and three elements of order . In the second case it contains two distinct elements and , whose product must have order at most ; hence , the product is , and the subgroup is or . Together with and this gives ten subgroups, so has exactly ten intermediate fields.
The fixed fields. In each case we exhibit an element fixed by the subgroup and generating a field of degree ; by the inclusion-plus-degree argument that field is .
- fixes and . The field has degree by the biquadratic theorem with , (none of is a rational square), and .
- fixes , and fixes , since . Both and are roots of the irreducible , so and have degree .
- sends and , so it fixes . Similarly exchanges and , so it fixes . Here , and . The polynomial is irreducible over (shown below), so and have degree .
- fixes , giving of degree .
- fixes , giving .
- fixes , because and ; this gives .
To see that is irreducible over : it has no rational root, since is impossible for real . A factorization into monic rational quadratics has the form , because the coefficient vanishes. Comparing coefficients gives , and . If , then and , which is impossible. If , then , which has no rational solution.
The lattice can be summarized as follows; the fields in each row have the stated degree over .
| Subgroup | Order of | Fixed field | Degree | normal? |
|---|---|---|---|---|
| yes | ||||
| yes | ||||
| , | , | no | ||
| , | , | no | ||
| , , | , , | yes | ||
| yes |
The normality column is computed in the group. The element commutes with and with (because ), so is normal; subgroups of index are always normal. On the other hand gives , so and . Thus and are conjugate, as are and , and none of these four is normal.
The conjugation proposition predicts the corresponding field statements, and they can be checked directly. Since , the field is , which is the fixed field of . Likewise , so . The real field contains the root of but not the root , so it is not normal over , in agreement with the table. The six normal subgroups give the six Galois subextensions , , , , and . For example has order and every element of it squares to the identity, since and ; so it is a Klein four group, in agreement with from the biquadratic theorem.
Hypotheses and counterexamples
Common mistake
(The hypotheses control the dictionary)
The correspondence is not a claim about every finite extension. The nonnormal extension has only the identity automorphism: the other roots of are nonreal. Its automorphism group has order , while its degree is . Nor does every intermediate field of an arbitrary Galois extension have to be normal over the base: normality corresponds precisely to a normal subgroup.
Two further failures are worth recording because they show that normality and separability each play an independent role.
Common mistake
(Normality is not transitive)
The extension is Galois, and is Galois because it has degree in characteristic zero. Yet is not normal. The polynomial is irreducible over by Eisenstein's criterion and has the root in the real field , but its root is not real and so is not in that field. Consequently has order at most , which is smaller than the degree . In the language of the correspondence, a tower of Galois extensions need not be Galois over the base; the intermediate field is normal over , but the top extension is not.
Common mistake
(Normal but not separable)
Let be a prime, let , and let be a root of in an extension. In characteristic we have , so the extension is the splitting field of , hence normal. The polynomial is irreducible over , because is not a -th power in ; the normality unit proves this with Eisenstein's criterion at the prime and Gauss's lemma. Therefore . Yet the only root of is , so by the first theorem the only automorphism of over is the identity. The extension is normal but inseparable, and its Galois group has order , not . The extension is not Galois, so the order formula and correspondence theorem do not apply; here the automorphism group has order .
Counterexample mode
Dropping the condition that the product ab is not a square
Claim: if and are not squares in , then has degree over . If , , and , then is a square and , so has degree rather than , and there is only one quadratic subfield rather than three. Over the larger base with and , the product is a square in , since . Here , so . This has degree over , because : if with , then , and both and are impossible in . The third hypothesis of the theorem, that is not a square in , is exactly what excludes both examples.
The dictionary is thus sensitive to the base field, as it should be, since the group changes when changes.
History of an idea
From roots to their symmetries
Galois studied equations through transformations of their roots and related these transformations to solvability by radicals. His papers on these ideas appeared in print in 1846, after his death. The modern field-theoretic formulation and its proofs were developed further by later mathematicians; the correspondence stated here uses that later language rather than claiming to reproduce his original notation.
Practice and solutions
Quick check
(In the biquadratic example, determine the subgroup fixing .)
Solution · Answer
It consists of the identity and the automorphism fixing while sending to .
Quick check
(Let be finite Galois with group . How many degree-three intermediate fields occur?)
Solution · Answer
A degree-three intermediate field corresponds to a subgroup of index three, hence order two. The three transpositions generate the three distinct order-two subgroups, so there are three such fields. None is Galois over , since these subgroups are not normal in .
Quick check
(Show that , and find the minimal polynomial of over .)
Solution · Answer
Put . Then , so . Also , and therefore . Thus , and then . Hence , which has degree . From we get , so is a root of . This monic polynomial has degree , so it is the minimal polynomial.
Quick check
(Let be finite Galois with cyclic Galois group of order . How many intermediate fields are there, what are their degrees over , and how many are Galois over ?)
Solution · Answer
The subgroups of a cyclic group of order correspond to the divisors , one subgroup for each, so there are six intermediate fields. Since , their degrees over are , each occurring exactly once. In particular there is a unique field of degree , corresponding to the subgroup of order . A cyclic group is abelian, so every subgroup is normal and all six intermediate fields are Galois over .
Quick check
(Determine and say which hypothesis of the biquadratic theorem fails.)
Solution · Answer
Since , we have . This field is the splitting field of and has degree , so its Galois group has order and consists of the identity and the map negating . The hypothesis that fails is that is not a square: here . (The hypothesis that is not a square in holds, as does that for .) Consequently the degree is , not , and only one quadratic subfield occurs.
Quick check
(Let be the splitting field of over . List the intermediate fields that contain .)
Solution · Answer
A field contains exactly when its subgroup is contained in . The subgroups of this group are , , , and the whole group, so the fields are , , , and . As a check, and .
Quick check
(For the same , determine and list the fields between and .)
Solution · Answer
The automorphisms fixing are , so is cyclic of order . Its subgroups are , and , giving the fields , and . Exactly one field lies strictly between and . Since and , this also shows that stays irreducible over .
Quick check
(Let be finite Galois with group , let be an intermediate field and . Show that the number of distinct fields with equals , where . Check the answer for in the example above.)
Solution · Answer
Each is again an intermediate field, so acts on the set of intermediate fields. By the conjugation proposition , and the correspondence is a bijection, so exactly when . Thus the stabilizer of is , and the orbit-stabilizer theorem gives distinct fields . For we have . The elements commute with , while ; so has index , and the two conjugate fields are and .
Quick check
(Show that by determining which automorphisms fix .)
Solution · Answer
Write , a positive real number. An automorphism sends to , which equals only if , that is . An automorphism sends to , and this equals only if . For the left side is ; for it is the real number ; for and it is , whose real part is nonzero. None equals . So only the identity fixes , which means , and the correspondence gives .