The correspondence between subgroups and intermediate fields is useful only if fixed fields have the degrees the correspondence predicts. One way to understand that prediction is to start with a finite group of symmetries and ask how much of the field remains visible to every symmetry at once. We will prove the key degree formula directly. The two inequalities use different ideas: a small linear system produces fixed coefficients, while independence of automorphisms prevents the group from being too large.
Throughout, fields are commutative and automorphisms preserve . If is a finite group of automorphisms of a field , its fixed field is
It is a field: sums, products and inverses of fixed elements are fixed. We write ; no degree or characteristic hypothesis on is assumed at this stage.
Distinct automorphisms are independent
A list of distinct automorphisms may look like a list of functions rather than vectors. Nevertheless, values of the functions lie in , so it makes sense to ask whether an -linear combination of those functions is identically zero.
Theorem 5.1.1
(Independence of distinct field embeddings)
Let and be fields and let be distinct field embeddings. If
with , then every is zero.
Proof
(Proof)
Suppose a nontrivial relation exists, and choose one with the smallest possible number of nonzero coefficients. Delete zero terms and relabel, so are all nonzero. A one-term relation is impossible: evaluation at would give its nonzero coefficient as zero. Hence .
Because and are distinct, choose with . Apply the relation to and subtract times the relation at . For every this gives
The last term vanished, while the first coefficient remains nonzero because is a field. We have constructed a nontrivial relation with fewer than nonzero terms, contradicting minimality. Thus no nontrivial relation exists.
The coefficients are allowed to lie in the target , not just in the field fixed by the maps. Independence is stronger than saying that the maps are pairwise different: arbitrary distinct functions need not be linearly independent. The multiplicative property of each embedding is exactly what makes the subtraction argument work.
Worked example 5.1.2
(Identity and conjugation)
The identity and complex conjugation are distinct embeddings . If for every complex , put to get , and put to get . Since , both coefficients vanish. These two substitutions illustrate the conclusion in a particularly small case; the theorem works for any finite list of distinct embeddings.
The fixed-field degree theorem
Theorem 5.1.3
(Artin's fixed-field theorem)
If is a finite group of automorphisms of a field and , then . In particular the degree is finite, even if no finite-dimensionality was assumed in advance.
Proof
(Proof)
Put . First show that any elements of are linearly dependent over . Consider the homogeneous equations
in the unknowns over . Elementary linear algebra supplies a nonzero solution. Choose one whose support, the set of indices with nonzero coefficient, has smallest cardinality. Select an index in its support and scale the solution so that the coefficient there is .
For any , applying to each equation shows that the tuple is also a solution: as ranges over , the compositions range over . Subtract the original solution. The difference is a solution, is zero at the normalized index, and is zero at every index outside the original support. If it were nonzero, it would have smaller support, contrary to our choice. Therefore for all and every . The lie in , are not all zero, and the equation for the identity automorphism becomes . Hence .
For the reverse inequality, write and choose an -basis . Every element of is an -linear map . Such a map is determined by its values on this basis, so the -vector space has dimension . The first theorem says that the distinct elements of are linearly independent in this space. Thus , proving .
Proof X-ray
Minimal support forces fixed coefficients
The proof of the upper bound is the delicate step. A solution to the equations over need not initially have coefficients in . Minimal support, followed by normalization, is what forces every coefficient to be fixed. The proof does not divide by the group order, so it remains valid when the characteristic divides .
Worked example 5.1.4
(Complex conjugation and the real field)
Let and , where . The fixed elements are exactly the real numbers, so . The theorem gives , agreeing with the basis . Notice that the theorem uses a finite group; the field itself is infinite, which presents no difficulty.
Worked example 5.1.5
(A fixed field of rational functions)
Let , where is an indeterminate, and let send to while fixing . This is an automorphism of order two, so its fixed field has index two in . The field is fixed. Also : if for a rational function , applying would give , impossible in characteristic zero. Since satisfies over , its degree is at most two and is not one. Thus . Both and lie between and , and both have index two in ; the tower law gives . This example shows how inclusion and a degree count identify a fixed field.
Worked example 5.1.6
(Frobenius on four elements)
Let and put . The four elements are . Squaring fixes and swaps with , because and . Thus has order two and its fixed field is . The theorem gives . Here the characteristic divides the group order, confirming that division by is not part of the argument.
Worked example 5.1.7
(A translation group in characteristic three)
Let and let . Since and neither nor fixes , this automorphism generates a group of order three. Put . In characteristic three, , so is contained in the fixed field. The element satisfies over , giving degree at most three. Its minimal polynomial has all three distinct roots : applying and to a relation over the fixed subfield preserves its coefficients. Therefore the degree is at least three and hence exactly three. Both and the fixed field have index three in , so they are equal. This is another example where the characteristic divides the group order; it also makes the difference between an invariant and a complete description of the fixed field visible.
Orbit polynomials and why the extension is Galois
The proof of the degree theorem did not start by assuming that was algebraic. There is another useful way to see algebraicity element by element. Given , its orbit is the finite set of distinct elements
Form a polynomial with one factor for each distinct orbit element:
Each automorphism in permutes the orbit, so it fixes every coefficient of . Consequently , and . The factors are distinct by construction. This does not by itself count a basis for the entire extension: having a small polynomial for each selected element is a statement about individual elements, while a dimension bound must control all their combinations at once. The linear-system proof supplied that global control.
Proposition 5.1.8
(A finite automorphism group gives a Galois extension)
If is a finite group of automorphisms of and , then is finite, normal and separable. Its full -automorphism group is exactly .
Proof
(Proof)
Artin's theorem gives . For any , its minimal polynomial over divides , since division by the minimal polynomial leaves a remainder of smaller degree that vanishes at . The polynomial splits into distinct linear factors in . Hence the minimal polynomial also splits in with distinct roots. This proves that every element is separable over and gives normality by the irreducible-polynomial characterization. The number of -automorphisms is at most , by the embedding-count bound; the subgroup already has exactly that many elements. Thus the full automorphism group is .
For the sign change on , the orbit polynomial of is . For translation by on , it is . In each example the coefficients are invariant, and the orbit already contains every root of the minimal polynomial. In general an orbit may have fewer than elements because a nonidentity automorphism can fix the chosen element; the degree theorem is about the whole extension, not the size of every individual orbit.
Counting automorphisms in a finite Galois extension
The fixed-field theorem begins with a group and identifies the degree over its fixed field. To use it for a given base field, one must still show that the automorphisms fix no additional elements. Normality and separability supply that step.
Proposition 5.1.9
(Automorphism count for finite Galois extensions)
Let be a finite normal separable extension and . Then and .
Proof
(Proof)
An -automorphism is an -embedding into an algebraic closure, and the embedding-count bound gives at most such maps, so is finite. Every element of is fixed by definition. Let , with minimal polynomial over . Its degree is greater than one. Separability makes its roots distinct, and normality puts all those roots in ; choose a root .
The one-root extension lemma gives an -embedding sending to . Extend it through the finite extension into an algebraic closure of . For any , its image under this embedding is a root of its minimal polynomial over , so normality places the image in . Thus the embedding maps into itself. It is injective and -linear on a finite-dimensional -space, hence surjective. It is an automorphism in that moves . Therefore no element outside is fixed by all of , so . Artin's theorem now gives .
Proof X-ray
Two uses of normality and one of separability
Separability is used once: it supplies a root of . Normality is used twice: to place in , and to make the extended embedding send into itself. Finite dimension then turns an injective -linear self-map into an automorphism. The count is not obtained by listing automorphisms. The proof shows only that fixes nothing outside ; Artin's theorem converts into .
The proof uses the finite-tower embedding-extension lemma and existence of an algebraic closure as background. It also uses the normality characterization that every irreducible polynomial over with a root in splits in . The fixed-field degree theorem itself did not assume normality or separability; those hypotheses enter when the base field is specified before the automorphism group is chosen.
Common mistake
(Finite degree alone does not count automorphisms)
The real field has degree three over , but only the identity automorphism over : the other two roots of are nonreal. The extension is separable but not normal. The fixed field of its trivial automorphism group is the whole cubic field, not . There is no contradiction with Artin's theorem, which uses the actual fixed field.
Counterexample mode
Without separability the count fails
Claim: every finite normal extension has automorphisms. Let with an indeterminate, and . Then is a root of , so is normal. An -automorphism sends to a root of this polynomial, hence to ; the automorphism group is trivial. Yet , since would equate degrees congruent to and modulo . The minimal polynomial of is with , and its -coefficient lies in only for . So . Repair: add separability.
Practice and solutions
Quick check
(Let be the trivial automorphism group of a field . What does the fixed-field theorem say?)
Solution · Answer
Every element is fixed, so and .
Quick check
(In , find the fixed field of and check its degree directly.)
Solution · Answer
The fixed field is ; every complex number has a unique expression with , so the degree is two.
Quick check
(Explain where the upper-bound proof uses that is closed under composition.)
Solution · Answer
After applying to the system, its rows are indexed by . Closure and invertibility make a permutation of , so the transformed tuple satisfies the same system.
Quick check
(Why does separability matter in the proof that a finite normal extension has enough automorphisms to move every element outside the base field?)
Solution · Answer
For , its minimal polynomial has degree greater than one. Separability ensures it has another distinct root; normality places that root in . Without a distinct root, the argument cannot choose an image different from .
Quick check
(In the rational-function example, can belong to the fixed field? Why?)
Solution · Answer
No: the automorphism sends to , which differs from in characteristic zero. The fixed field is .
Quick check
(Does the fixed-field theorem require the characteristic not to divide ? Give an example.)
Solution · Answer
No. The Frobenius group on above has order two in characteristic two and fixed field ; the degree is two.