Recovering information without solving for the roots
Suppose a cubic has roots . Its coefficients determine , and , even when no individual root has been calculated. They also determine , because this expression is . Is that an isolated trick, or can every polynomial expression unchanged by reordering the roots be recovered from the coefficients?
The symmetric-polynomial theorem answers this question and gives a procedure for doing the recovery. This supplies a foundational step behind resolvents and equations for expressions in roots. The companion generic-polynomial chapter establishes the symmetric Galois group by counting automorphisms; here we prove the constructive statement about expressions themselves. Polynomial arithmetic over a field is sufficient. The proof works in positive characteristic as well as characteristic zero.
Formal symmetry and elementary coefficients
Fix a field and an integer . Initially are independent indeterminates, not particular algebraic numbers. Put . A permutation of the indices acts on this ring by permuting its variables and extends to its fraction field.
Definition 1.8.1
(Symmetric polynomial and elementary symmetric polynomials)
A polynomial is symmetric if every permutation of the variables leaves it unchanged as a polynomial. For , define
Thus and . Each is symmetric and homogeneous of total degree .
Choosing either or from each factor gives the identity
The alternating signs belong to this product expansion; in characteristic two the minus signs equal plus signs. Substituting actual roots into a polynomial identity is always allowed. No division by root differences is involved.
Formal symmetry is stronger than an accidental equality after substitution. Over , the nonzero polynomial vanishes at every pair in , but it is not symmetric in . Equal values on a finite field do not establish equality of polynomials. We prove the theorem in the polynomial ring before evaluating it.
A leading term tells us what to subtract
Order exponent vectors lexicographically with : compare the exponent of first, then the exponent of if necessary, and so on. The leading monomial is the largest monomial with nonzero coefficient in the polynomial. For example, the leading monomial of is .
Suppose a nonzero symmetric polynomial has leading term
Then . Indeed, if , swapping those two variables produces a monomial with the same nonzero coefficient but a larger lexicographic exponent vector, contradicting maximality. Set . The polynomial
has leading monomial with coefficient one. Subtracting times this product cancels the leading term while preserving symmetry. All exponents are nonnegative because of the inequalities just proved.
Theorem 1.8.2
(The fundamental theorem of symmetric polynomials)
Every symmetric has a unique expression
If is homogeneous of total degree , the expression can be chosen with every monomial of weighted degree .
Proof
(Existence by cancellation; uniqueness by distinct leading monomials)
For existence, repeatedly perform the subtraction above. If the initial total degree is at most , each product subtracted has total degree . Hence no monomial of degree greater than is introduced. At every nonzero step the leading monomial strictly decreases. There are only finitely many monomials of total degree at most , so the process terminates at zero. The recorded subtractions give . If was homogeneous of degree , every remainder and every subtracted product is homogeneous of degree , giving the weighted-degree assertion. The zero polynomial is represented by .
For uniqueness, the leading monomial of has exponent vector
Its consecutive differences recover the . Thus distinct monomials in the give distinct leading monomials after substitution. In a nonzero , choose the term whose substituted leading monomial is largest. No other substituted term can cancel it: each has a strictly smaller leading monomial and all its other monomials are smaller still. Therefore . Two different expressions for would contradict this by subtraction.
Proof X-ray
Why a decreasing leading term really terminates here
The useful finite set is the set of monomials of total degree at most the initial degree bound. Every cancellation stays inside that set. Saying only that a monomial becomes smaller would leave the termination mechanism unexplained. The proof also never divides by a positive integer, so a characteristic dividing creates no exception.
Worked example 1.8.3
(Carrying out the subtraction for three variables)
Consider
Its leading monomial is , with exponent vector , so subtract . Expanding that product gives each of the six terms of once, together with . The remainder is , and therefore . In characteristic three this becomes ; the identity survives, with coefficients interpreted in .
What uniqueness means after evaluating the variables
The uniqueness assertion takes place in the polynomial ring with independent variables. It says that two polynomials in formal symbols cannot become the same polynomial in after substituting the elementary expressions, unless the two original polynomials were equal. This is also a proof that the formal are algebraically independent.
Common mistake
(Evaluated elementary values can satisfy extra relations)
Actual roots need not be independent. If are all the roots of a depressed cubic , listed with multiplicity, then the evaluated is zero. Therefore the two distinct formal polynomials and take the same value at . This does not contradict the theorem: before evaluation, their difference becomes , a nonzero polynomial. Evaluation then sends that polynomial to zero.
There are consequently two different tasks. To obtain a universal identity in roots, work with independent variables and use the theorem; the result can subsequently be evaluated in every extension field. To simplify an expression for the roots of one particular polynomial, one may also use relations satisfied by its actual coefficients. Those additional relations can shorten the answer but remove uniqueness of its displayed form. Keep these tasks separate when checking a proposed identity: numerical agreement at one tuple cannot replace a universal polynomial identity.
Symmetric rational functions need a denominator argument
Common mistake
(Numerator and denominator need not be symmetric)
A rational function can be invariant without its displayed numerator and denominator being symmetric. For instance, . We therefore cannot apply the polynomial theorem separately to the original numerator and denominator.
Theorem 1.8.4
(The symmetric rational functions)
The elements of fixed by all variable permutations form exactly .
Proof
(Replace the denominator by a product of its permutations)
Every rational expression in the is fixed. Conversely, let an invariant rational function be with and . Let be the set of all permutations and form
Each factor is nonzero, and the polynomial ring is a domain, so . Permuting variables only permutes the factors of , making symmetric. The identity permutation contributes the factor , so is a polynomial. Both and are invariant, hence is a symmetric polynomial as well. The polynomial theorem writes and with nonzero denominator. Thus belongs to .
An average over the group would require division by and would fail in some characteristics. The product proof avoids that operation. When particular roots are substituted into the resulting rational expression, its chosen denominator must be nonzero. Polynomial identities have no such restriction; a rational identity does not authorize division by zero at a special tuple.
From coefficients to an equation for new elements
Let a monic cubic over split in an extension field as
where the symbols now denote the evaluated elementary expressions. Roots are listed with multiplicity. We can compute a polynomial for their squares without solving the cubic.
Worked example 1.8.5
(An equation whose roots are the squared roots)
The elementary expressions in are
For the second identity, expand . Its cross terms are . Therefore
Here the product has one factor for each entry in the root list, including repeated entries. For over , we have , giving . Its only possible rational roots are , where its values are . Thus the cubic is irreducible over . For each root of , the square has minimal polynomial and degree three. This last conclusion uses the additional irreducibility check, not just the displayed identity.
Worked example 1.8.6
(Pairwise products give a different cubic)
The three expressions have sum , sum of pairwise products , and product . Hence
For , the equation is . Its values at are , so it is irreducible over and is the minimal polynomial of each pairwise product. These products also equal , since . This gives a second check: substituting into and multiplying by gives the same equation. The substitution is legitimate because the constant term of is nonzero, so no root is zero.
An equation can have unnecessary factors
Counterexample mode
Transforming distinct roots can make them coincide
Over , let , with roots . Their squares are . The squared-root formula gives . This is a valid equation for each square, but the minimal polynomial of is , not . Even a separable original polynomial can produce a transformed polynomial with repeated roots. The transformation has identified two roots.
Common mistake
(A transformed equation need not be minimal)
Symmetry guarantees that the new equation has coefficients in the base field. It does not guarantee that its degree is the degree of any chosen transformed element, that all transformed values are distinct, or that the equation is irreducible. To determine a minimal polynomial, combine the equation with irreducibility, independent coordinates, or a proved conjugate count. This is the same logical distinction used in the minimal-polynomials-from-coordinates companion.
Power sums and Newton's identities
The question above stated Newton's identities without proof. They are polynomial identities with integer coefficients, so they hold over every field; only solving them for requires dividing by .
Theorem 1.8.7
(Newton's identities)
In put for (so ), and for . Then
Proof
(Differentiate the root product)
Write with . Fix and put for , and . Then and for , while because . Multiplying out therefore gives , so .
The product rule gives . Compare the coefficients of for . On the left it is . On the right, summing over gives , whose last term is . Hence
For , sum the equations over : this gives , and the last term is again , so the same displayed relation holds for . Substitute and write ; multiplying by gives the first identity.
For , multiply by to get , and sum over . This gives , which is the second identity after substituting .
The first identity determines from and the earlier exactly when is invertible in , which is the condition in the question above. The second identity needs no division at all: once are known, every later power sum follows from the coefficients.
Worked example 1.8.8
(Power sums of the roots of a cubic)
Let be the roots of , so that , and . The first identity gives , then , then . For the second identity reads , so
As a check, is the sum of the squares of the roots of found earlier, namely . The value also agrees with the practice formula below.
Two families of roots at once
The proof of the fundamental theorem never divides by a coefficient. It subtracts times a product whose leading coefficient is one, and its uniqueness argument compares leading terms with coefficient one. Therefore the theorem, with the same proof, holds with the field replaced by any commutative ring : every symmetric polynomial in is uniquely a polynomial in with coefficients in . This version lets us treat two independent families of variables.
Theorem 1.8.9
(Polynomials symmetric in two families)
Let be unchanged by every permutation of the and by every permutation of the . Then is a polynomial over in the elementary symmetric polynomials of the and those of the .
Proof
(Apply the theorem over a coefficient ring, then use uniqueness)
Put and regard as an element of . It is symmetric in the , so the ring version gives a unique with . Let permute the , and apply it to the coefficients of . Because the do not involve the , we get . Uniqueness gives . Thus every coefficient of is a symmetric polynomial in the , and by the fundamental theorem over it is a polynomial in . Substituting these expressions into gives the required polynomial.
Proposition 1.8.10
(Equations for sums and products of roots)
Let be monic of degrees , and suppose and in for some extension . Then
lie in . Consequently, if and are algebraic over , then so are and .
Proof
(Compute generically, then evaluate)
In form . A permutation of the , or of the , only permutes the factors, so each coefficient of satisfies the hypothesis of the preceding theorem and is a polynomial over in the and . Now apply the ring homomorphism sending and . It sends to the first displayed product, and it sends to times the coefficient of in , which lies in ; similarly for and . Hence the evaluated coefficients lie in . The product case is identical with in place of .
For the consequence, let and be the minimal polynomials of and , and take for a splitting field of over . Then is one of the , so it is a root of the first product, a monic polynomial over ; the same holds for and the second product.
This gives a second proof, by explicit equations, that the algebraic elements form a field, once one also notes that and, for , are roots of the nonzero polynomials and over , where is a nonzero polynomial of degree over vanishing at ; the unit on algebraic towers proves it by degree counting instead. As with the transformed-root equations above, the resulting polynomial need not be irreducible.
Worked example 1.8.11
(An equation for the sum of two square roots)
Take and over . The four sums are . Grouping the two sums with and the two with ,
This is the minimal polynomial of obtained by degree counting in the algebraic-towers unit; here it arises with rational coefficients guaranteed in advance by the proposition.
Practice
Quick check
(Express in the for . Explain the characteristic-two case.)
Solution · Answer
Expanding gives , so the expression is . In characteristic two the cross terms vanish and the answer is . No division by two is being used.
Quick check
(For three variables, express in without dividing by three.)
Solution · Answer
Expansion gives , where . Using yields . In characteristic three it reduces to , still a valid identity.
Quick check
(For two independent variables, express in elementary symmetric functions. State the denominator condition after evaluation.)
Solution · Answer
The expression is . For values the chosen denominator is , so both must be nonzero. Formal nonzeroness of alone does not ensure nonzeroness at every tuple.
Quick check
(In the cancellation algorithm, which product has leading monomial ? Give its total degree.)
Solution · Answer
The exponent differences are , , and . Use . Its total degree is , matching .
Quick check
(Let over . Find the squared-root polynomial and prove that it is the minimal polynomial of the square of any root.)
Solution · Answer
Here , so the polynomial is . A rational root of this monic cubic would be an integer whose cube is four; none exists. It is irreducible, so it is the minimal polynomial of every squared root. If , then , also proving .
Quick check
(For , find the pairwise-product polynomial. Show that its positive real root generates .)
Solution · Answer
The polynomial is again . If the roots of are with and a primitive cube root of unity, their pairwise products are . The positive real one is , and gives the claimed field equality.
Quick check
(Why does the rational-invariant proof still work when divides ? Where would an averaging argument fail?)
Solution · Answer
The product of finitely many nonzero denominator polynomials is nonzero in a domain, regardless of the characteristic. No integer is inverted. Averaging would divide a sum over by , which is zero in the field when the characteristic divides .
Quick check
(Over , apply the squared-root formula to . Check irreducibility and explain why the resulting polynomial equals .)
Solution · Answer
In characteristic two, . The formula gives . It has value one at both elements of , so the cubic is irreducible. Squaring is Frobenius and permutes the roots: the identity holds because the coefficients belong to . Thus the unchanged polynomial has a field-theoretic explanation as well as an algebraic check.
Quick check
(Let over . Use Newton's identities to compute the power sums of its roots, and check the answer with the roots .)
Solution · Answer
Here and . The first identity gives , and . For the second identity reduces to , so and . Directly, , which is when and otherwise; this gives and .
Quick check
(For and , compute the product polynomial of the proposition on sums and products. Is it the minimal polynomial of ?)
Solution · Answer
The four products are twice and twice, so the product polynomial is . It has rational coefficients, as the proposition guarantees, but it is not irreducible. The minimal polynomial of is the factor , which is irreducible because is irrational. The coincidences among the products are the same phenomenon as the coinciding squares in the counterexample above.