Motivation
The cubic formula constructs roots by repeatedly adjoining square and cube roots. For a general polynomial, such a construction may be hidden: intermediate expressions need not be roots of the original polynomial, and a radical extension need not itself be normal over the starting field. The Galois criterion converts the existence of some radical tower into a condition on the finite group that permutes the roots. Its proof has two directions, and each needs a separate construction.
Throughout, is a field of characteristic zero, all extensions are taken inside a fixed algebraic closure, and is nonconstant. A radical tower is a finite sequence
for positive integers . A polynomial is solvable by radicals over if one such tower contains all its roots. The tower may contain extra elements and need not be normal over . Adjoining a root of unity is a radical step because .
A group is solvable if its derived series , eventually reaches the identity. In this unit all groups are finite. We use the finite Galois correspondence, extension and uniqueness of embeddings, the independence of distinct field embeddings, the tower law and existence of algebraic closures as background. The companion fixed-field unit proves embedding independence and the finite Galois automorphism count; it does not construct algebraic closures. We also use the standard fact that the compositum of two finite Galois extensions is Galois and that restriction maps in the finite Galois correspondence are onto.
The group operations used in a tower
Proposition 7.2.1
(Solvability passes through subgroups, quotients and extensions)
Subgroups and quotients of a solvable group are solvable. If and both and are solvable, then is solvable.
Proof
(Derived series under subgroups, quotients and extensions)
If , induction gives : every commutator formed in is also a commutator in . Thus the derived series of reaches the identity no later than that of .
For a surjective homomorphism , the image of the commutator subgroup is the commutator subgroup of the image, because and every element of has a preimage. Inductively , so a quotient of a solvable group is solvable.
Finally, if , the quotient calculation gives . If , then . Thus is solvable. This proof uses normality to form the quotient and says exactly where each assumption enters.
A finite solvable group also admits a subnormal series whose successive quotients are cyclic of prime order. To see this, take a derived series, whose quotients are finite abelian, and refine each quotient. Every nontrivial finite abelian group contains an element of prime order: choose a nonidentity element of order , choose a prime , and raise it to the power . Its cyclic subgroup is normal because the group is abelian. Induct on the size of the quotient and lift the resulting subgroup chain through the quotient map. The lifted consecutive quotients are cyclic of prime order. Normality is needed only between consecutive members; the intermediate groups need not all be normal in the original group.
Worked example 7.2.2
(A solvable group with nonnormal subgroups)
The symmetric group has the series with quotients of orders three and two. Both quotients are cyclic, so is solvable. A subgroup generated by a transposition is not normal in ; solvability does not require every subgroup to be normal. The splitting field of has Galois group , consistent with its roots being expressible using a cube root and a cube root of unity.
A cyclic prime extension becomes one radical step
The next lemma supplies the construction in the direction from groups to radicals. The base field must already contain the relevant root of unity; we will adjoin those roots before applying it.
Theorem 7.2.3
(Cyclic prime extensions over a field with roots of unity)
Let have characteristic zero and contain a primitive th root of unity , where is prime. If is Galois with cyclic Galois group of order , then for some satisfying .
Proof
(A nonzero resolvent)
Choose a generator of . For define the resolvent
This is not zero for every : otherwise it would be a nontrivial -linear relation among the distinct embeddings , contradicting their independence. Choose with .
Reindexing the sum and using and gives . Hence , and every power of fixes . The fixed field of the full Galois group is , so . Since and , , so . The tower law makes a divisor of ; it is greater than one, hence is . Thus .
Concept lensAlgebraic
The resolvent as an eigenvector of the generator
View as an -dimensional -vector space and as a -linear map with . The map satisfies and is the identity on the -eigenspace of , so projects onto that eigenspace up to the factor . The lemma says that is an eigenvalue of , and any eigenvector is a valid . Since is a -eigenvector, all eigenvalues occur and .
Worked example 7.2.4
(The quadratic case)
Let and let change the sign of . Take , and . The resolvent is . Its square is , and it generates the same quadratic extension. A different choice of , such as , would give zero; the independence argument guarantees that some choice works, not that every choice works.
Common mistake
(The root of unity is a hypothesis)
A degree- extension need not be generated by an element whose th power lies in the base merely because its degree is . The lemma assumes a cyclic Galois extension and a primitive th root of unity in the base. Its proof uses that root both in the coefficients of the resolvent and in the equation .
Counterexample mode
A cyclic cubic over the rationals that is not a pure cubic
Claim: a Galois extension with cyclic group of prime order is always generated by an th root. Over , let , a root of . This cubic has no rational root ( give and ) and discriminant , so is Galois and cyclic of degree three. Its three roots are real, so . If with , normality would put the conjugate , hence the nonreal , in . So no such exists. The repair is the lemma's hypothesis: over , the cyclic cubic is generated by a cube root.
Solvable group implies a radical tower
Proposition 7.2.5
(Solvable Galois group gives radical expressions)
If the splitting field of has solvable Galois group, then is solvable by radicals over .
Proof
(Adjoin roots of unity, then follow a composition series)
Let . Adjoin to a primitive th root of unity for each prime dividing , and call the resulting field . This is a finite radical extension of : each adjoined element satisfies . The compositum is the splitting field of over and is finite Galois there. Restriction gives an injection
because an automorphism fixing both and fixes the field they generate. Its image is a subgroup of the solvable group , so is solvable.
Choose a subnormal series with cyclic prime-order quotients. Galois correspondence reverses it to a tower
Each successive extension is cyclic Galois of prime degree . Every such prime divides and hence , so , and therefore every field in the tower, contains a primitive th root of unity. The cyclic-prime lemma says that each step can be generated by one element whose th power is in the preceding field.
Concatenate the initial roots-of-unity steps with this tower. It is a radical tower over containing , so it contains every root of .
Proof X-ray
Why the roots of unity are adjoined first
The order of operations is the key move. Roots of unity are adjoined before any Galois theory is done, so that the cyclic-prime lemma is available at every later step. Adjoining them changes the group, but only to the subgroup of , and solvability passes to subgroups. A finite solvable group has a subnormal series with cyclic quotients of prime order, and the correspondence reverses that series into a tower of cyclic steps of prime degree. Each prime that occurs divides , hence , which is why it was enough to adjoin th roots of unity only for the primes dividing . The lemma then makes every step a single radical, and the top of the tower is , which contains every root of .
Worked example 7.2.6
(A cyclic cubic after adjoining a cube root of unity)
Suppose contains a primitive cube root of unity and is Galois cyclic of degree three. The lemma gives and . The three conjugates are . This conclusion uses the stated hypotheses; it does not assert that every cubic extension over an arbitrary base is Galois or has this form without first adjoining a root of unity.
Making a radical step normal without losing solvability
The reverse direction is subtler because a given radical tower need not be Galois over . We enlarge it one step at a time, retaining a solvable Galois group. At each stage, adding all conjugates of a radicand repairs normality; adding roots of unity makes the new kernel abelian.
Proposition 7.2.7
(A normal solvable enlargement of one radical step)
Let be a finite Galois extension with solvable Galois group, and let for an integer . There is a finite Galois extension containing whose Galois group over is solvable.
Proof
(Adjoin all conjugate radicals)
If , then and we can take . Assume and put for a primitive th root of unity. The cyclotomic field is a splitting field of over and is Galois in characteristic zero. Its Galois group embeds in the abelian group by the image of . The compositum is Galois, and restriction embeds into , so this kernel is abelian. Its quotient is ; the solvable-extension lemma therefore makes solvable.
Put . If , then and works. Otherwise let be the distinct conjugates of under . Choose one root of for each , choosing for the conjugate equal to , and let . Since , all roots of each lie in .
The product has coefficients fixed by , hence in . The extension is finite Galois, so it is a splitting field of a finite product of separable polynomials over . The field is the splitting field over of that product together with . It is therefore finite Galois over .
Every automorphism of fixing sends each nonzero to and is determined by these images. The resulting map embeds into a product of copies of the abelian group of th roots of unity: the multipliers multiply under composition because the automorphisms fix . Hence this kernel is abelian. Its quotient is the solvable group , so is solvable. By construction contains .
Worked example 7.2.8
(Why conjugate radicands appear)
Start over with the Galois extension and adjoin . The radicand has conjugate . Adjoining only the real root gives the real field , which is not normal over : it omits . Adding and roots over both conjugate radicands gives the splitting field of . This small example shows why a proof cannot simply declare every radical step normal over the original base.
The radical-solvability criterion
Theorem 7.2.9
(Galois criterion for solvability by radicals)
Let be nonconstant over a field of characteristic zero, and let be its splitting field over . Then is solvable by radicals over if and only if is solvable.
Proof
(Two directions through the Galois correspondence)
The direction from a solvable Galois group to a radical tower was proved above. Conversely, suppose a radical tower contains . Begin with , which is finite Galois over itself with trivial, hence solvable, group. Inductively suppose a finite Galois extension with solvable group contains . The next tower generator satisfies . The normal-enlargement proposition gives a finite Galois with solvable group containing , and hence .
At the end . Since is itself finite Galois, restriction maps onto . The latter is a quotient of a solvable group and is therefore solvable. This proves the reverse implication and completes the equivalence.
The criterion applies to the splitting field of the specified polynomial over the specified base. It is not a claim that every degree-five equation is unsolvable: is solvable by adjoining a fifth root of and a fifth root of unity. The companion cubic-and-quintic unit constructs an explicit polynomial with Galois group . Combining that calculation with the additional group-theoretic fact that is not solvable yields a polynomial not solvable by radicals. The nonsolvability of is not proved in this unit.
Practice and solutions
Quick check
(Why does the cyclic-prime lemma require choosing an with nonzero resolvent rather than taking ?)
Solution · Answer
For a nontrivial primitive th root, , so . Independence of the distinct automorphisms guarantees that the resolvent map is not identically zero, and hence some choice of works.
Quick check
(If and , why is fixed by every power of ?)
Solution · Answer
Applying gives . A generator fixing an element makes all its powers fix it.
Quick check
(Why can adjoining a primitive root of unity be counted as a radical step?)
Solution · Answer
A primitive th root satisfies in the previous field, exactly the radical-step condition.
Quick check
(In the normal-enlargement proof, why do all roots of lie in after one root and have been adjoined?)
Solution · Answer
If , its roots are for in characteristic zero. Both factors already lie in .
Quick check
(Why does a quotient of a solvable group remain solvable?)
Solution · Answer
A surjection carries each commutator subgroup onto the commutator subgroup of its image, so it carries the derived series term by term. Once the original series reaches the identity, so does the quotient's series.
Quick check
(Does the theorem say that every quintic is not solvable by radicals?)
Solution · Answer
No. For example, splits after adjoining and a primitive fifth root of unity, both radical steps. The criterion depends on the particular Galois group, not degree alone.