The quadratic formula expresses roots using arithmetic and square roots. A natural question is whether repeated extraction of roots can solve every polynomial. Before discussing the general obstruction, a cubic calculation shows both the power of radicals and the care needed when choosing them.
This unit has four aims. The first is to make the phrase "solvable by radicals" precise and to see why its definition is shaped the way it is. The second is to carry out Cardano's method for the cubic completely, including a proof that one compatible pair of cube roots produces all three roots, and a proof of what the quantity under the square root says about real roots. The third is to explain Cardano's radicals as Lagrange resolvents, linking them to the discriminant and to the symmetries of the roots, and to prove that for an irreducible cubic with three real roots the detour through nonreal numbers cannot be avoided. The fourth is to state, with its hypotheses, the Galois criterion that explains why some equations of degree five cannot be solved by radicals. Wherever a result is used without proof, it is marked as such.
Radical towers and the question being asked
Definition 7.1.1
(Radical extension)
Let have characteristic zero. A radical tower is a finite sequence in which and for some positive integer . A polynomial over is solvable by radicals over if some radical tower contains all its roots. Roots of unity may occur in the tower; the definition does not require every intermediate extension to be Galois.
Each step of a tower adjoins a single element that is an -th root of something already constructed. A formula built from field operations and root extractions, such as the quadratic formula, is exactly an element of such a tower: one reads the formula from the inside outward and adjoins each radical as it appears. The definition therefore captures the informal phrase "the roots can be written using and ". The condition is the only constraint, so the same tower may be traversed with several different choices of , and the definition asks only that some choice exist.
The remark that intermediate extensions need not be Galois is not a technicality, and the first radical step one would write down for the cubic already shows it.
Counterexample mode
A radical step need not be a Galois extension
Claim: every step of a radical tower is a Galois extension. Take and , where is the real cube root, so . The polynomial is irreducible over by Eisenstein's criterion at the prime , and contains only one of its three roots; the other two, and , are nonreal. A Galois extension that contains one root of an irreducible polynomial contains all of them, so is not Galois. The repair is to adjoin roots of unity first. If contains a primitive -th root of unity and , then splits in , so is a splitting field over and, in characteristic zero, the step is Galois. For instance, is such a step. This is why the Galois criterion stated later cannot be proved by simply following an arbitrary tower.
Depressing the cubic
For over a characteristic-zero field, set . Direct expansion removes the quadratic term and gives , where and . Write . Then
It is worth checking the expansion rather than trusting it. Since ,
Adding these and the constant , the terms cancel, the coefficient of is , and the constant term is . This agrees with the stated and . The hypothesis of characteristic zero is used because division by and is required; in characteristic the substitution is unavailable. The substitution is a translation, so it matches roots of the original and depressed cubics one to one and never changes multiplicities.
The second identity above is the expansion , with the terms added. Its purpose is to replace one unknown by two unknowns , which leaves freedom to impose two conditions. We impose the two that make the right side vanish.
Choose and . Thus and are the roots of
The reason is Vieta's relations: the sum of is and their product is . The conditions are therefore equivalent to a quadratic equation, which we can solve with a square root, and the cubic has been reduced to extracting one more cube root.
Definition 7.1.2
(Resolvent quadratic and its discriminant)
For the depressed cubic , the polynomial is its resolvent quadratic, and is the discriminant of that quadratic. In this notation the roots of the resolvent are , with and . As a notational remark, equals times the usual discriminant of the cubic.
Choosing compatible cube roots
These equations give a construction, but independently choosing arbitrary cube roots need not satisfy . If , choose a nonzero cube root of one value and then set ; the equation for ensures that is the other value. If , solve directly. Multiplying and by opposite powers of a primitive cube root of unity gives the other roots, with coincidences possible when the cubic has repeated roots.
Worked example 7.1.3
(A cubic solved with compatible radicals)
For , one has , , and the two values of are , namely and . Choose and as real cube roots. Their product is , so is a root. Indeed . The remaining roots are and , where .
Common mistake
(Independent root choices)
Taking cube roots independently in Cardano's calculation can destroy . For the example above, replacing by while leaving unchanged gives product , so the cancellation used in the derivation fails. Adjust the second cube root by the inverse factor.
The mistake has a concrete algebraic form. Whenever holds, the identity displayed earlier reduces to for . So is a root precisely when or . Compatibility of the pair is therefore the entire content of the condition , and an incompatible pair gives a number that is generally not a root. The exercise on below carries this out for the example above.
All three roots from one compatible pair
The worked example asserts that the other two roots are obtained by twisting and with cube roots of unity. We now prove this in a form that also handles repeated roots.
Theorem 7.1.4
(Cardano's construction produces every root)
Let be a field of characteristic zero and let with . Suppose contains an element with , a nonzero root of , and an element with . Put and . Then and , and for with ,
In particular are the roots of counted with multiplicity.
The hypothesis guarantees that a nonzero root exists: if both roots of the resolvent were zero, then and . The excluded case is , which is trivial.
Proof
(Proof of the construction theorem)
Since is a root of the resolvent and , we get . Then , and is the definition of . Hence .
From we get , and because otherwise . So and . Write , , and compute the elementary symmetric functions.
First, .
Second, for , . Over the pairs the values of are , so the coefficients add to , and the coefficients add to . In each pair is not divisible by , so . Hence the sum over pairs is .
Third, , so .
Therefore .
Proof X-ray
Where each hypothesis enters the coefficient comparison
The proof never solves for a root. It multiplies out a candidate factorization and compares its three coefficients with those of , and each hypothesis is used at exactly one point. The condition makes , so that is defined; this is the only reason is excluded. Characteristic zero gives , hence and , so exponents of may be read modulo . The coefficient vanishes from alone, for any and . Compatibility is used only in the coefficient, where the cross terms add to , and is used only in the constant term, through the factorization . Because the conclusion is an identity of polynomials, multiplicities need no separate argument: a repeated root simply means that two of the coincide. For an incompatible pair with , the same computation gives , which differs from the cubic by exactly the term found earlier.
The theorem has a direct consequence for the definition at the start of the unit.
Theorem 7.1.5
(Every cubic is solvable by radicals)
Every nonconstant polynomial of degree at most three over a field of characteristic zero is solvable by radicals over .
Proof
(Proof of solvability of the cubic)
Degrees one and two are handled by the linear and quadratic formulas, the latter using one square root. For a cubic, divide by the leading coefficient and depress it to with ; this uses only arithmetic in . If the only root is . Otherwise let with ; since this is a radical step. Let with . Replacing by if necessary, . Let with . Then , and by the previous theorem all roots lie in . Translating back by gives the roots of the original cubic in .
Here the tower has three radical steps with exponents , and . Note that the proof does not pass through the Galois criterion; the criterion below is consistent with it, since every subgroup of is trivial, of order , of order , or itself, and each of these has a subnormal chain with abelian quotients.
Real roots and the sign of the discriminant
For a cubic with real coefficients, the formula forces a surprising feature: the radicals may leave the real numbers even when every root is real. The next result says exactly when this happens.
Theorem 7.1.6
(Real roots of the depressed cubic)
Let and .
(a) If , then has exactly one real root and two nonreal complex conjugate roots.
(b) If , then it has three real roots counted with multiplicity, although are nonreal.
(c) If and is the real number with , then .
Proof
(Proof of the real-root classification)
Work in with , so that .
(a) Here is real and nonzero, so are distinct real numbers, and we may choose . Let be the real cube root of and put , which is real, with . Then gives . The root is real, while has nonzero imaginary part and . By the previous theorem these are all the roots.
(b) Here are nonreal conjugates, and , so . Choose with ; then , and since the positive real cube root is unique, . Therefore . Each root is , which is real.
(c) Here , so , that is . Real cube roots are unique, so and . Then .
Part (b) is the situation where the real answer is a sum of two cube roots of nonreal numbers. The section on the casus irreducibilis below proves that, for an irreducible cubic, no choice of real radicals avoids this passage through nonreal numbers.
Two further computations
Worked example 7.1.7
(Nonreal radicals, three real roots)
Consider . Here and , and the resolvent is , because . Then , so by part (b) all three roots are real. The resolvent roots are . Since , take . Then , and . Indeed .
For the others, and , so . Similarly , so . As a check, , the product is , and the sum of pairwise products is . The roots are and .
Worked example 7.1.8
(A cubic that is not depressed)
Solve . With , , we get and . Setting gives , which can be confirmed by expanding . Now , so there is exactly one real root. The resolvent is with roots . Take ; since , we have . The real root is
and the other roots are and , a nonreal conjugate pair. A numerical check: , and .
Lagrange resolvents and the discriminant
Cardano's two cube roots look like auxiliary quantities introduced to make an identity work. In fact they are determined by the roots themselves, and this explains why the method succeeds.
Theorem 7.1.9
(Cardano's radicals are Lagrange resolvents)
In the situation of the construction theorem, with for ,
Moreover the product of root differences satisfies
Proof
(Proof of the resolvent identities)
Using and ,
and similarly .
For , write , and . Now , while and , whose sum is . Hence
and multiplying by gives . Finally , and . Therefore .
An expression of the form with a cube root of unity is called a Lagrange resolvent. The identities show what each step of Cardano's tower computes. Let have characteristic zero, let have distinct roots , and let be the field generated over by the roots and . Then . An automorphism of fixing permutes the roots, and the resolvent formulas give its effect on and :
- if permutes the roots cyclically, , then and similarly ;
- if fixes and exchanges , then and .
So and are fixed by the cyclic permutations and exchanged by the transpositions, which is why they satisfy a quadratic over . The element changes sign under each transposition and is fixed by the cyclic permutations; since and the characteristic is not , the automorphisms of over fixing are exactly those acting by even permutations. By the fundamental theorem of the Galois correspondence unit, is the fixed field of that subgroup, and because is a nonzero element of .
Concept lensStructural
Cardano's tower follows the chain of subgroups
Over the roots are permuted by a subgroup of . Adjoining passes to the fixed field of the even permutations, and adjoining the cube root passes to the field fixed by the identity alone. The radical steps of Cardano's formula therefore follow the chain , with a square root for the quotient of order and a cube root for the quotient of order . The Galois criterion stated below says that this is not a peculiarity of the cubic: radical formulas exist exactly when such a chain with abelian quotients exists.
The casus irreducibilis
Part (b) of the real-root theorem left open whether the nonreal cube roots could be avoided by some cleverer formula. For an irreducible cubic they cannot. To make this precise, call a radical tower real if every is contained in . The proof needs one irreducibility fact about pure equations of prime degree.
Proposition 7.1.10
(Pure equations of prime degree)
Let be a field of characteristic zero, let be a prime and let . If is not an th power in , then is irreducible over .
Proof
(Proof of the prime-degree criterion)
Since , we have . In a splitting field of , the roots are distinct, because the derivative has only the root , which is not a root of . Fix a root . Each root has the form with , since for every root . Suppose in with monic of degree , where . The roots of are of the roots , so is their product, which equals for some with . Thus lies in , and . Since there are integers with , and then
with , because . This contradicts the hypothesis, so no such factorization exists.
The prime hypothesis cannot be dropped. Over the number is not a fourth power, since fourth powers of rationals are nonnegative, yet .
Theorem 7.1.11
(Casus irreducibilis)
Let be a field and let be an irreducible cubic whose three roots are all real. Then no real radical tower over contains a root of .
Proof
(Proof of the casus irreducibilis)
Prime exponents. If a step adjoins with and with prime and , insert the intermediate field : then and , and is real when is. Repeating this, and deleting steps with , which add nothing, we may assume every step of a real radical tower has prime exponent.
The root differences. Let be the roots of ; they are distinct because is irreducible in characteristic zero. Put . It is real and nonzero. After dividing by its leading coefficient and translating, becomes a depressed cubic with ; a translation does not change root differences, and since is reducible. The resolvent theorem, applied in , gives , because reordering the roots changes at most in sign. So is a subfield of with . The cubic has no root in , since a root would give dividing . Hence is irreducible over .
One root gives all roots. Let be any field with , and let be a root of . Write and label so that . The sum equals , where is the coefficient of , so it lies in . Also , so , the sign depending only on how the roots were labelled in . Hence , and is a splitting field of over .
The contradiction. Suppose a real radical tower with prime exponents contains a root of , where and . Then the fields form a real radical tower over with the same generators, and contains a root. Let be the least index such that has a root in ; then . Put , , and . Since has no root in , it is irreducible over , and for a root the tower law shows that divides . In particular . Then is not an th power in : if with , then (else ), with real, so and . By the proposition, is irreducible over , so . Since and is prime, , and have the same degree over , so . By the previous step is a splitting field over , hence normal over by the normality unit. The irreducible polynomial has the root in , so it splits there: its roots , , all lie in . But and , so . This contradiction proves the theorem.
Proof X-ray
Where the three real roots are used
The hypothesis that all three roots are real is used exactly once: it makes real, so that can be adjoined without leaving . For a cubic with one real root, and is purely imaginary; then the real tower of part (a) of the real-root theorem does reach the real root. Irreducibility over is used to keep irreducible over , and it fails for , whose roots and need only a real square root. The other ingredients are the tower law, the prime-degree proposition and the fact that a splitting field is normal. Characteristic zero enters through the existence of three distinct roots and through the proposition.
What the group detects
Definition 7.1.12
(Solvable group)
A finite group is solvable if it has a finite chain of subgroups in which each is normal in and each quotient is abelian.
For example, is such a chain. This definition concerns the existence of a chain; the entire group need not be abelian.
In the example, has index in and is therefore normal, the quotient has order and is abelian, and has order and is cyclic, so its quotient by the identity is abelian. Yet itself is not abelian, since two transpositions need not commute. Solvability is a property of the whole chain and is strictly weaker than commutativity.
Theorem 7.1.13
(Galois criterion for radicals)
Let be a nonconstant polynomial 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.
The theorem is proved in the radical-criterion unit. Its proof requires adjoining suitable roots of unity and relating cyclic extensions to radical adjunctions, as in the unit on cyclic radical extensions; the correspondence theorem alone does not establish it. The criterion explains why a splitting field and its whole automorphism group are relevant, rather than only a formula for one root. In particular, the theorem together with the group-theoretic fact that is not solvable for , proved in the unit on symmetric groups, shows that a polynomial with Galois group in this range is not solvable by radicals. This is a conditional conclusion: not every polynomial of degree at least five has Galois group , and some higher-degree equations, such as , are solvable by radicals.
Neither the Galois criterion nor the nonsolvability of for is used for the cubic: everything about the cubic in this unit is proved directly.
Common mistake
(Degree five means unsolvable)
It is false that every equation of degree at least five fails to be solvable by radicals. The criterion concerns the Galois group of the particular polynomial. The polynomial has degree five and is solvable by radicals over , as the cubic and quintic continuation shows. Some quintics do have Galois group and hence fail to be solvable by radicals. The statement about a universal formula for generic coefficients is proved in the unit on the generic polynomial.
Practice and solutions
Quick check
(Solve by factoring, and compare its roots with the radical construction.)
Solution · Answer
The factorization is . Here , , and the resolvent has twice. The compatible choice gives , while , (still ) gives , and so does , . The repeated root appears twice, as the construction theorem predicts.
Quick check
(Why does a quintic with Galois group fail to be solvable by radicals, while is solvable by radicals over ?)
Solution · Answer
The radical criterion applies in characteristic zero, and is not solvable, so a polynomial with Galois group is not solvable by radicals. For , first adjoin a fifth root of and then a primitive fifth root of unity ; both are radical adjunctions, since and . All roots are for , so they lie in this tower.
Further exercises with complete solutions
Quick check
(Solve over .)
Solution · Answer
Here , , , so and . With the equation is , which can be seen directly from . The roots are and hence for . In the notation of Cardano's construction, , so one may take , .
Quick check
(Solve and identify the repeated root.)
Solution · Answer
Here , , and . The resolvent is , so , , and . The roots are and . Thus the roots are , and confirms it. The repeated root is .
Quick check
(Using the sign of , say how many real roots each of , and has.)
Solution · Answer
For : , so three real roots. For : , so one real root. For : , so one real root. Checks: the function has derivative and so is strictly increasing; and takes the values at , giving three sign changes.
Quick check
(Prove that is not a Galois extension of .)
Solution · Answer
The polynomial is irreducible over by Eisenstein at , so it is the minimal polynomial of . The field lies in , so it cannot contain the nonreal roots and . A Galois extension containing one root of an irreducible polynomial contains all of them, so the extension is not Galois.
Quick check
(In the worked example , show that is not a root.)
Solution · Answer
Put and . Then , so , and the identity gives . Since , this equals . Now , and because is nonreal. Hence and is not a root.
Quick check
(For with roots , , , compute the Lagrange resolvents and , and verify .)
Solution · Answer
With we have and . Then , so the first resolvent is ; similarly and the second is . These are the values , found in the worked example. For the discriminant, , and , so and . On the other hand . The formula also checks, since .
Quick check
(Show that is a root of , and deduce that lies in no real radical tower over . Then express by Cardano's formula.)
Solution · Answer
From with we get for . Putting gives , so . The only possible rational roots are , and the values there are and , so the cubic has no rational root and is irreducible over . Its , so all three roots are real. By the casus irreducibilis no real radical tower over contains , hence none contains either. Cardano's formula still applies: the resolvent is with , so , , and .
Quick check
(Use the proposition on pure equations of prime degree to show that is irreducible over .)
Solution · Answer
By the proposition it suffices to show that is not a cube in . If with , then ; but is irreducible over by Eisenstein at , so , which does not divide . Hence is not a cube in and is irreducible over .
Summary of hypotheses and dependencies
The following points summarize what was proved and what was assumed.
- Depressing a cubic needs invertible, which holds in characteristic zero.
- Cardano's construction needs a compatible pair with ; independent choices fail, as shown by the incompatible pair , in the exercises.
- A nonzero resolvent root exists whenever , and then the three numbers are all the roots with multiplicity; this was proved above.
- For real coefficients, the sign of decides between one real root, three real roots and a repeated real root, and this was proved above.
- Radical towers need not consist of Galois steps, as shows.
- Cardano's cube roots are the Lagrange resolvents and , and the product of root differences satisfies .
- For an irreducible cubic with three real roots over a subfield of , no real radical tower contains a root; this casus irreducibilis was proved above from the tower law, the irreducibility of for prime , and normality of splitting fields.
- The Galois criterion and the nonsolvability of for are stated here and proved in the radical-criterion and symmetric-group units; the cubic results above do not depend on them.