Motivation
A factorization over the rational numbers may contain denominators even when the original polynomial has integer coefficients. It is tempting to say that clearing denominators solves the problem, but clearing them can introduce scalar factors that must be accounted for. The content of a polynomial records exactly that scalar information. Gauss's lemma explains why it behaves predictably when coefficients have unique factorization.
This leads beyond integer polynomials. If a domain is a UFD, so is , and repeated application gives unique factorization in any finite number of variables. The argument does not give principal ideals: a polynomial ring in two variables over a field is already a useful counterexample. We will separate the fraction-field construction, primitive factorization, and the final uniqueness argument, so each hypothesis has a visible role.
Throughout, is a commutative domain with identity. The UFD and prime-element facts from the chapter on Euclidean domains, PIDs and UFDs, and polynomial division/unique factorization over a field from the chapter on polynomials and irreducibility, are named prior results. The fraction field is constructed here rather than silently identified with a familiar numerical field.
Definitions
Definition 8.1.1
(The fraction field of a domain)
On pairs with and , define when . The class of is written . Define
The resulting field is , and embeds in it. Domain cancellation is essential to this construction.
Here are the construction checks. Reflexivity and symmetry of the relation are immediate. If and , multiply the first equality by , use the second, and cancel nonzero to obtain ; this proves transitivity. Replacing representatives in either displayed operation gives equal cross-products after expansion, so both operations are well-defined. Their ring laws follow on a common nonzero denominator from the laws in . The additive zero is , the identity is , and is zero exactly when . A nonzero fraction has inverse . Finally, implies , proving the embedding. We henceforth regard as a subring of .
Definition 8.1.2
(Content and primitive polynomials over a UFD)
Assume now that is a UFD. For a nonzero polynomial , its content is a gcd of all its coefficients, defined up to multiplication by a unit. Its existence follows by taking the minimum exponent of each prime in the nonzero coefficients. A polynomial is primitive if its content is a unit, equivalently no prime divides all its coefficients. Write with primitive.
Zero coefficients cause no difficulty: every element divides zero, so the nonzero coefficients determine the minimum exponents. We do not assign a primitive part to the zero polynomial. The units of are exactly the constant units of , since degrees add for nonzero polynomials over a domain. “Primitive” does not mean monic; is primitive over the integers.
Theorems
Theorem 8.1.3
(Gauss's lemma and descent of factorization)
Over a UFD, a product of primitive polynomials is primitive, and is associated to . Every nonzero polynomial in is a nonzero scalar in times a primitive polynomial in .
Two primitive polynomials in differing by a scalar in differ by a unit in . Consequently a primitive polynomial of positive degree is irreducible in exactly when it is irreducible in . If a primitive polynomial divides an element of in , it already divides it in .
Theorem 8.1.4
(Polynomial rings preserve unique factorization)
If is a UFD, then is a UFD. Its irreducibles are the constant irreducibles of and the primitive positive-degree polynomials irreducible over , up to multiplication by units. Every such irreducible is prime in .
In particular and are UFDs for every field and finite . No claim of principal ideals or Euclidean division in these general rings follows.
Proof sketch or proof idea
Reduction modulo a prime controls content
Proof
(Primitive products and the scalar normalization)
Let be primitive. If their product were not primitive, some prime would divide every coefficient of . In the domain , the reduced polynomials are both nonzero, because neither coefficient list is entirely divisible by . Their product cannot be zero in the polynomial ring over that domain. This contradicts the supposed divisibility of all product coefficients. Thus is primitive.
Writing , , the primitive-product result shows that the content of their product is associated to . This is a statement up to a unit, since contents themselves have only that normalization. For , multiply by a common nonzero denominator of its finitely many coefficients, then extract the content of . This gives with and primitive.
Concept lensAlgebraic
Gauss’s lemma as additivity of a prime-by-prime valuation
Fix a prime of . For a nonzero element let be the exponent of in its factorization, and for nonzero put over its nonzero coefficients. The numbers determine the content up to a unit, and Gauss’s lemma is equivalent to for every prime . Writing with not dividing all coefficients of reduces this to the primitive-product step modulo . For , the values recover content six.
The scalar between primitive polynomials must be a unit
Proof
(Unit scalars, irreducibility and divisibility descent)
Suppose with primitive and chosen coprime in the UFD. Then . If a prime divided , it would not divide , so coefficientwise primality would force to divide every coefficient of . This contradicts primitivity. Hence is a unit. Interchanging the roles shows that too is a unit. The scalar is therefore a unit of .
Now suppose a primitive positive-degree factors as over , both factors of positive degree. Normalize , to primitive over . Their product is primitive, and , so the preceding scalar argument makes a unit of . This produces a nontrivial factorization over . Conversely a factorization of a primitive over cannot contain a nonunit constant factor: that would divide every coefficient. Both nonunit factors must have positive degree and remain nonunits over . This proves the irreducibility equivalence with its primitive hypothesis intact.
For divisibility descent, the zero dividend is immediate. Write a nonzero dividend as with primitive. If primitive divides over , it divides there, say . Normalize with primitive. Since is primitive, comparison with forces to be a unit of . Thus in . This stronger conclusion will transfer primality from the field polynomial ring.
Existence and uniqueness over the original coefficients
Proof
(Existence and primality of the listed irreducibles)
For nonzero nonunit , factor its content into irreducibles of . Factor its primitive part in the UFD . Normalize each positive-degree field irreducible to a primitive polynomial . Their product is primitive, so its scalar difference from the primitive part of is a unit of . Every is irreducible over by the descent argument. Constant irreducibles of remain irreducible in because a factorization of a nonzero constant can only involve constants. This proves existence of the listed factorization.
A constant prime is prime in , because its quotient is the domain . A primitive positive-degree irreducible is prime in . If it divides in , field primality makes it divide or over , and divisibility descent puts that quotient back in . Thus every listed irreducible is prime. Given two factorizations, primality matches one irreducible with an associated factor on the other side; cancel and repeat. Domain cancellation and finite factor lists give uniqueness up to units and order. Applying the theorem repeatedly proves the multivariable consequence.
Proof X-ray
Why clearing denominators is only the beginning
Clearing denominators gives integral coefficients, but not a factorization of the original primitive polynomial without an extra scalar. Primitive products stay primitive, and two primitive polynomials can differ only by an original-ring unit. Those two facts control that scalar. Only then can field factorization and field primality be transported back to the coefficient ring.
Eisenstein's criterion over an arbitrary UFD
Theorem 8.1.5
(Eisenstein's criterion over a UFD)
Let be primitive, , and let a prime satisfy , for all , and . Then is irreducible in and .
Proof
(Reduction to a monomial modulo )
Otherwise Gauss descent gives over , both factors of positive degree. Their leading coefficients are not divisible by , because their product is . Reduction modulo therefore preserves both positive degrees and gives a factorization of the monomial . Over a domain the smallest and largest nonzero degrees of a product are the sums of the corresponding degrees of the factors. A monomial product consequently forces each factor to be a monomial. Both reduced factors have positive degree, so both constant coefficients vanish modulo . Their product is then divisible by , a contradiction. This is the same prime-content mechanism in a form useful for polynomial coefficient rings.
Worked examples
Worked example 8.1.6
(Content records the missing scalar factors)
In , has content six and primitive part . Its irreducible factorization is . Both linear factors are primitive and irreducible over , hence over . The alternative rational factorization of the primitive part does not itself have integral factors. Normalizing its factors gives and ; the scalars cancel. A nonmonic primitive factor is entirely compatible with the theorem.
Worked example 8.1.7
(Irreducibility with a polynomial parameter)
Let be any field, and . The coefficient ring is a UFD because it is Euclidean. Regard as a polynomial in over . The prime divides its constant coefficient but its square does not, divides the zero linear coefficient, and does not divide the leading coefficient one. Eisenstein therefore proves irreducibility in and , in every characteristic. It does not make a PID: the ideal is proper by evaluation at , while a common divisor of must be a unit. Were it principal, its generator would be such a unit, contradicting properness.
Worked example 8.1.8
(A reducible reduction gives no converse test)
The primitive polynomial has no rational root: the rational-root criterion permits only , neither of which is a root. A quadratic factorization over a field would have a linear factor, so it is irreducible over and hence over . Yet modulo three it is . A same-degree irreducible reduction can certify irreducibility under the usual primitive/leading-coefficient hypotheses, but a reducible reduction does not certify reducibility over the fraction field. Reduction is a useful sufficient test, not an equivalence.
Counterexample mode
The primitive-product statement needs the coefficient hypothesis
In , put . Each of and has no nonunit common divisor of its coefficients. Indeed, such a divisor would have positive integer norm dividing both four and six, hence norm two unless it were a unit; but has no integer solution. Their product is , whose coefficients all have the nonunit divisor two. Thus the analogous primitive-product claim fails in this domain. The UFD hypothesis supplies prime coefficient tests that this example lacks.
Common mistakes
Common mistake
(Compare irreducibility only for primitive polynomials)
Irreducibility over and over are compared for primitive positive-degree polynomials. The constant two is irreducible over but a unit over , while loses its scalar nonunit obstruction after passing to . Neither violates the theorem.
Common mistake
(A UFD need not be a PID)
A UFD need not be a PID, and polynomial unique factorization does not create a division algorithm with arbitrary leading coefficients. Content is defined only up to a unit unless a normalization has been chosen. Zero and unit polynomials require separate boundary handling; they are not extra irreducible factors.
Summary
The fraction field permits field polynomial factorization. Content and primitive-product control then remove the denominators without losing scalar information. Irreducibility and divisibility descend for primitive polynomials, making every coefficient or polynomial irreducible prime and establishing unique factorization in polynomial rings over UFDs.
Exercises
Quick check
(If is an injective unital homomorphism into a field, prove there is a unique field homomorphism extending it.)
Solution · Answer
The only possible value of is , since a field homomorphism preserves inverses and by injectivity. Equality makes these values equal after multiplying by the nonzero denominators. The fraction operations show this is a unital homomorphism, and a nonzero fraction maps to a nonzero value, so it is injective. This constructs the extension and proves uniqueness. Requiring an embedding matters: a map with a nonzero kernel could send an allowed denominator to zero.
Quick check
(Show directly that divides in . Explain why the primitive hypothesis cannot be deleted from general divisibility descent.)
Solution · Answer
The quotient is , since . For failure without primitivity, the constant polynomial two divides one in but not in . Clearing the denominator of its quotient changes the equality rather than proving the original divisibility.
Quick check
(Prove is a UFD but not a PID. Specify why the ideal is proper and why it cannot be principal.)
Solution · Answer
Apply the polynomial UFD theorem twice to . Evaluation at maps the ideal to zero but maps one to one, so the ideal is proper. Since is prime (the quotient by is the domain ), its only nonunit divisors are associates of . But does not divide , as evaluation at shows. A common divisor of is therefore a unit. A principal generator for their ideal would divide both and be a unit, contradicting properness.
Quick check
(Prove is irreducible over and in . What happens under the specialization ?)
Solution · Answer
In the UFD , use the prime . The polynomial is primitive, its leading coefficient is one, every lower coefficient is divisible by , and its constant coefficient is not divisible by . The proved Eisenstein criterion gives both irreducibility claims. At it becomes , which is reducible. Irreducibility over a rational-function field does not assert preservation at every parameter value.