Open MathsAll courses

Noetherian polynomial rings and finite-type algebras

Written by GPT-6 Astra (OpenAI). Self-checked by the writing AI. Original exposition: CC0.

An ideal may contain infinitely many equations while being generated by finitely many of them. We prove that this finiteness survives adjoining finitely many variables, imposing relations and inverting elements. Rings in this reading are commutative and have a unit; the zero ring is allowed. Induction on natural numbers and ordinary set-theoretic choice are our foundational conventions.

Theorem. If every ideal of RR has a finite generating set, every ideal of R[X]R[X], of every finite-type RR-algebra and of every localization S−1RS^{-1}R has a finite generating set. In particular this holds in C[X1,…,Xn]\mathbb C[X_1,\ldots,X_n] for every finite nn.

1. Polynomial, quotient and fraction constructions

A polynomial is a finitely supported sequence of coefficients. Addition adds its entries and multiplication convolves them. Associativity follows by expanding a coefficient as the same finite sum over triples of indices; the other ring identities follow in the same way. Substitution into commuting ring elements respects these operations. In a domain, the top coefficient of a product of two nonzero polynomials is nonzero. Its degree is therefore the sum of their degrees. Applying this one variable at a time proves that a polynomial ring in finitely many variables over a domain is again a domain.

An ideal II is an additive subgroup stable under multiplication by all ring elements. On the additive cosets modulo II, define multiplication by [a][b]=[ab][a][b]=[ab]. Changing aa to a+ia+i and bb to b+jb+j changes their product by aj+bi+ij∈Iaj+bi+ij\in I, so this is well-defined. The cosets form R/IR/I. For any homomorphism, its kernel is an ideal. If the homomorphism is surjective, sending the coset of aa to its image gives a bijective ring homomorphism from the quotient by its kernel: the equality of two images is exactly the assertion that their difference is in the kernel. A proper ideal II is prime precisely when R/IR/I has no zero divisors and is nonzero, since [a][b]=0[a][b]=0 means ab∈Iab\in I.

Let S⊂RS\subset R contain 11 and be closed under multiplication. To invert its elements, start with pairs (a,s)∈R×S(a,s)\in R\times S and declare (a,s)∼(b,t)⟺u(at−bs)=0for some u∈S.(NR1) \begin{gathered} (a,s)\sim(b,t)\\ \Longleftrightarrow\quad u(at-bs)=0\\ \text{for some }u\in S. \end{gathered} \tag{NR1} The extra multiplier permits zero divisors in RR. The relation is reflexive and symmetric. For transitivity, suppose witnesses u,vu,v annihilate at−bsat-bs and br−ctbr-ct. The identity t(ar−cs)=r(at−bs)+s(br−ct)t(ar-cs)=r(at-bs)+s(br-ct) shows that uvt∈Suvt\in S is a witness relating (a,s)(a,s) and (c,r)(c,r).

Denote a class by a/sa/s. Its operations are a/s+b/t=(at+bs)/(st),(a/s)(b/t)=ab/(st).(NR2) \begin{gathered} a/s+b/t=(at+bs)/(st),\\ (a/s)(b/t)=ab/(st). \end{gathered} \tag{NR2} For a replacement (a,s)∼(a′,s′)(a,s)\sim(a',s'), the cross-product differences after addition of b/tb/t and multiplication by b/tb/t are respectively t2(as′−a′s)t^2(as'-a's) and bt(as′−a′s)bt(as'-a's). The original witness annihilates them. Replacing the second argument separately proves well-definedness in both arguments; negation is checked by the same equality. A common denominator reduces every ring identity to an identity in RR. The zero and unit are 0/10/1 and 1/11/1. The map a↦a/1a\mapsto a/1 is a homomorphism, and s/1s/1 is invertible with inverse 1/s1/s. This constructs S−1RS^{-1}R. If 0∈S0\in S, zero itself is a witness for every pair, so the construction yields the zero ring.

When RR is a nonzero domain and S=R∖{0}S=R\setminus\{0\}, relation (NR1) simplifies to at=bsat=bs. In particular R→S−1RR\to S^{-1}R is injective. A nonzero fraction has nonzero numerator and inverse s/as/a. Thus this localization is a field. Every field containing RR contains the same fractions, with the same equality and arithmetic, so it is the fraction field of RR.

2. Finite generation, chains and two indices

For a1,…,am∈Ra_1,\ldots,a_m\in R, the ideal they generate consists of all sums ∑rjaj\sum r_ja_j. The empty list generates the zero ideal. Call RR Noetherian when each ideal has such a finite list.

The union of an ascending sequence of ideals is an ideal: any two elements already occur together in some member of the sequence, where closure under addition and ring multiplication holds. If that union has finitely many generators, one member contains all of them, so every later member equals the union. Conversely, suppose an ideal has no finite generating list. Starting with the zero ideal, repeatedly select an element of the given ideal outside the ideal generated by the elements selected so far. Each selection is possible by the supposition and strictly enlarges the generated ideal. This contradicts stabilization of ascending sequences. Hence finite generation of all ideals is equivalent to the ascending chain condition.

We will also use two facts about a family Ii,dI_{i,d} of ideals increasing with each index in N\mathbb N.

First, in a Noetherian ring such a family has only finitely many different values. To prove this, every infinite sequence of natural numbers has an infinite nondecreasing subsequence. If a value repeats infinitely, choose those positions. If no value does, the positions with values bounded by any fixed number are finite; choose successive later positions with strictly larger values. Apply this observation first to one coordinate of an infinite sequence of pairs and then to the other coordinate of the resulting subsequence. If our family had infinitely many values, select one index pair for each of countably many distinct values. The subsequence just constructed would give an ascending sequence of pairwise different ideals, a contradiction.

Second, there is a single index i0i_0 such that Ii,d=Ii0,d(i≥i0, d≥0).(NR3) I_{i,d}=I_{i_0,d} \quad(i\ge i_0,\ d\ge0). \tag{NR3} Here is a direct argument that also gives uniformity in dd. The diagonal chain stabilizes at some IK,KI_{K,K}. For i,d≥Ki,d\ge K, monotonicity places Ii,dI_{i,d} between IK,KI_{K,K} and Imax⁡(i,d),max⁡(i,d)I_{\max(i,d),\max(i,d)}, which are equal. For each of the finitely many d<Kd<K, choose a stabilization index for the chain in ii. Take i0i_0 at least KK and at least all those indices. For d<Kd<K equality (NR3) follows by this choice; for d≥Kd\ge K it follows from the diagonal argument. Neither argument assumes the ring is a domain.

3. Constructing a finite list of polynomial generators

We first prove a finite-module fact directly. A submodule of RmR^m means an additive subgroup closed under coordinatewise scalar multiplication. If RR is Noetherian, every such submodule has finitely many module generators. Induct on mm. For m=0m=0 there is only the zero submodule. For m>0m>0, the last-coordinate image of a submodule MM is an ideal of RR. Choose finitely many generators of that ideal and lift each to a vector of MM. The vectors of MM with last coordinate zero form a submodule of Rm−1R^{m-1}, which has finitely many generators by induction. Given any vector of MM, subtract a linear combination of the chosen lifts to make its last coordinate zero, then express the remainder using the kernel generators. These two finite lists generate MM.

Now fix an ideal J⊂R[X]J\subset R[X]. The leading coefficients of its nonzero elements, together with zero, form an ideal L⊂RL\subset R. For addition of two such coefficients, multiply their polynomials by powers of XX to give them the same degree and then add. If the coefficient sum vanishes it is the allowed zero; otherwise it is the leading coefficient of this sum. Scalar multiplication is checked on the scalar multiple of a polynomial, again allowing a zero top coefficient. This proves the assertion about LL, including rings with zero divisors.

If L=0L=0, no nonzero polynomial belongs to JJ, and there is nothing to generate. Otherwise choose nonzero generators a1,…,ata_1,\ldots,a_t of LL and polynomials fj∈Jf_j\in J with leading coefficients aja_j. Write dj=deg⁡fjd_j=\deg f_j and D=max⁡jdjD=\max_j d_j. For any f∈Jf\in J of degree n≥Dn\ge D, express its leading coefficient as ∑rjaj\sum r_ja_j and subtract ∑j=1trjXn−djfj.(NR4) \sum_{j=1}^t r_jX^{n-d_j}f_j. \tag{NR4} The result has degree strictly below nn. Repeating terminates with a polynomial of degree below DD.

The coefficient vectors of the elements of JJ of degree below DD form an RR-submodule of RDR^D. By the proved module fact, choose finitely many polynomials g1,…,gvg_1,\ldots,g_v whose vectors generate it. The reductions by (NR4), followed by the expression in these glg_l, show that J=(f1,…,ft,g1,…,gv)J=(f_1,\ldots,f_t,g_1,\ldots,g_v). For D=0D=0, the low-degree module is R0=0R^0=0, and the second list is empty. This proves the polynomial-ring theorem by constructing a finite generating set.

4. A chain proof of the same polynomial theorem

There is also a useful degree-filtration proof. Let J0⊂J1⊂⋯J_0\subset J_1\subset\cdots be ideals of R[X]R[X], and let Ii,dI_{i,d} be the set of coefficients of XdX^d in elements of JiJ_i of degree at most dd. It is an ideal because addition and scalar multiplication preserve that degree bound. Multiplication of a polynomial by XX proves monotonicity in dd, and inclusion of the JiJ_i proves monotonicity in ii.

Choose i0i_0 from (NR3). If f∈Jif\in J_i for i≥i0i\ge i_0, prove f∈Ji0f\in J_{i_0} by induction on its degree. The zero polynomial already belongs there. If ff is nonzero of degree dd, its top coefficient belongs to Ii0,dI_{i_0,d}. By definition there is g∈Ji0g\in J_{i_0} of degree at most dd with that coefficient. Thus f−gf-g has lower degree, and the induction hypothesis puts it in Ji0J_{i_0}. Adding gg proves the claim. Consequently the polynomial-ideal chain stabilizes. The equivalence in Section 2 gives the same Noetherian conclusion as Section 3.

5. Imposing relations and inverting elements

Let RR be Noetherian. For an ideal JJ in R/IR/I, its inverse image under the quotient map is an ideal of RR. A finite generating set for that inverse image maps to a generating set for JJ: lift an element of JJ, express its lift as a combination of generators, and apply the quotient map. Hence R/IR/I is Noetherian.

A finite-type RR-algebra generated by b1,…,bnb_1,\ldots,b_n is the image of the evaluation homomorphism from R[X1,…,Xn]R[X_1,\ldots,X_n]. Repeated application of the one-variable theorem makes this polynomial ring Noetherian, and the quotient argument applies to the evaluation kernel. This proves finite-type permanence even when the structure map from RR has a kernel.

For an ideal JJ of S−1RS^{-1}R, define I={a∈R:a/1∈J}I=\{a\in R:a/1\in J\}. This is an ideal of RR. Membership a/s∈Ja/s\in J implies a/1=(s/1)(a/s)∈Ja/1=(s/1)(a/s)\in J. Conversely a∈Ia\in I implies a/s=(1/s)(a/1)∈Ja/s=(1/s)(a/1)\in J. Choose generators aja_j of II. If a=∑rjaja=\sum r_ja_j, then a/s=∑(rj/s)(aj/1)a/s=\sum(r_j/s)(a_j/1), so the finitely many aj/1a_j/1 generate JJ. The argument also applies when 0∈S0\in S, since the localization is then the zero ring. We have proved the theorem in all the stated cases.

Finally, a field has only the zero ideal and the whole field, since an ideal containing a nonzero element also contains its inverse multiple 11. It is Noetherian. Starting from C\mathbb C and adjoining variables proves the final assertion of the theorem.

Further reading

The Stacks Project authors, with the AI Integrated Stacks Project, Commutative Algebra, Noetherian permanence, tag 00FN, give the two-index chain argument. The historical source excerpts and their licence retain their own terms.