You are on page 1of 5

RINGS OF INTEGERS, DEDEKIND DOMAINS, AND SCHEME-THEORETIC CURVES

BRIAN OSSERMAN

1. Dedekind Domains We assume throughout that A is a Noetherian domain, with fraction eld K , and that all our rings are Noetherian domains. Any use of p will denote a non-zero prime ideal of A. De nition 1. We say that A is a Dedekind domain if it is normal of dimension 1. Recall that normal means integrally closed, so we could state this as: A is Dedekind if it is integrally closed, and every non-zero prime ideal is maximal. We also recall from commutative algebra: Proposition 1. A local ring A is a discrete valuation ring if and only if it is normal of dimension 1. Proof: See Se], Prop 3. That is to say, a local ring is a discrete valuation ring if and only if it is a Dedekind domain. It turns out that the property of being a Dedekind ring is completely local. That is: Proposition 2. A Noetherian domain A is Dedekind if and only if Ap is Dedekind for all non-zero prime ideals p of A (and, to rule out the case where A is a eld, at least one non-zero prime ideal exists). Proof: Recall that if S is any multiplicative subset of A which contains 1 but not 0, then there is a bijection of primes ideals of S 1 A with prime ideals of A disjoint from S , given in one direction by intersection with A, and in the other by taking the ideal generated in S 1 A (see exercises). Then suppose that A is Dedekind, and take p a non-zero prime ideal of A. Then Ap certainly has dimension 1, as prime ideals of Ap correspond to prime ideals of A contained in p via intersection with A. Further, Ap must be integrally closed: denote by K the eld of fractions of A, and suppose x 2 K is integral over Ap . Now x satis es some monic polynomial xn +an 1 xn 1 +: : : a0 = 0 with each ai 2 Ap , and we can therefore clear denominators to get an equation sxn + a0n 1 xn 1 + : : : a00 = 0 where s 2 A r p, and a0i 2 A. Multiplying through by sn 1 gives that sx is integral over A and hence in A, so x is in Ap as desired. Conversely, if all the Ap are dimension 1, then given any non-zero prime ideal p, Ap has dimension 1, so the only prime ideal contained in p must be 0. An element x 2 K integral over A is integral over every Ap , so must actually be an element of Ap by hypothesis. The following lemma then completes the proof:
1

BRIAN OSSERMAN

non-zero prime ideals of A. Proof: Take x = a=b 2 \p Ap . Let a = fy 2 A : ya 2 bAg. This is certainly an ideal of A, and it cannot be contained in any prime ideal p0 of A, since we can write a=b = c=d for some d 62 p0 , thanks to the hypothesis that x 2 Ap0 , and then d 2 a. Since a is not contained in any prime ideal, it must be all of A, and in particular, 1 a 2 bA, so x = a=b 2 A, as desired. De nition 2. A fractional ideal of A is an A-submodule a of K with the property that there exists a d 2 A (called a denominator for a) such that da 2 K . Proposition 3. The fractional ideals of a Dedekind domain A form a group under multiplication (with A as the identity). Proof: For fractional ideals a; b of A, denote by (a : b) the fractional ideal fx 2 K : xb ag (see exercises). We then claim that if a is an ideal of A, a0 = (A : a) is the inverse of a. Now, a0 a is a fractional ideal of K , and it is contained in A by construction. But we claim that it is not contained in any prime p of A, and must therefore actually be A, as desired. The point is that a0 aAp = aA0 aAp , which we claim is just Ap . The rst equality is left an exercise. p The second is trivial, making use of the fact that as a DVR, Ap is in particular a principal ideal domain, making the calculation of (a : b) extremely explicit. But since a0 aAp = Ap , then certainly a0 a 6 p, so we get the desired result. Corollary 1. A fractional ideal a of A is determined by its local images. Proof: First let a be an ideal of A. The previous proof shows that a fractional ideal a0 is an inverse of a whenever a0 a A, and a0 is an inverse of a everywhere locally. But I claim the second condition implies the rst: take x 2 a; x0 2 a0 . Then by hypothesis, aApa0 Ap = Ap , so in particular, xx0 2 Ap for all p, and by the earlier lemma, xx0 2 A. But now if a1 and a2 are the same everywhere locally, we know that a1 1 = a2 1, so a1 = a2 . It immediately follows that any fractional ideal which is the inverse of an ideal of A is also determined by its local images. Now, let a be a fractional ideal, and choose d so da A. Then da and (d 1 ) are determined by their local images, and so must be a. Lemma 2. If p p1 pk , then p = pi for some i. Proof: A=p is an integral domain, and in it, p1 pk = 0. Now, the product of non-zero ideals can never be 0 in an integral domain, so some pi = 0, and is contained in p. But pi is maximal, so we actually have pi = p, as desired. De nition 3. If a is an ideal of A, the valuation of a at p, p(a), is de ned to be k, where aAp is the kth power of the maximal ideal of Ap . p is clearly a homomorphism from the group of fractional ideals to Z. Lemma 3. Given x 2 A, x is contained in only nitely many prime ideals. Proof: Note that ideals containing x satisfy the descending chain condition, since if A a1 a2 : : : (x), since the fractional ideals form a group we get an inverse chain (preserving strict inclusion) A a1 1 a2 1 : : : (x 1 ). But (x 1 ) = A as A-modules, and is in particular Noetherian, so this chain must stabilize. Now, if p1; p2 ; p3 ; (x), then we get p1 p1 \ p2 p1 \ p2 \ p3 : : : (x), which must stabilize, meaning for some k, and all i > k, pi p1 \ p2 \ : : : pk

Lemma 1. In a domain A, A = \pAp , where the intersection is taken over all

RINGS OF INTEGERS, DEDEKIND DOMAINS, AND SCHEME-THEORETIC CURVES

this isn't relevant to the proof). Corollary 2. For any fractional ideal a of A, p(a) = 0 for all but nitely many p. Proof: First suppose that a is an actual ideal of A. It is clear that p > 0 , p a, and choosing any non-zero element x 2 a, applying the above lemma and using a (x) immediately gives the desired result. But now if a is a fractional ideal, choose d 2 A so that da A, and we nd that both p (d) and p (da) are non-zero at all but nitely many places, and then it follows that p (a) also has nite support. Corollary 3. Every fractional ideal a of A can be written uniquely as pe1 pe . 1 k Proof: Setting ei = p (a), we know by the previous corollary that we only need nitely many pi , and we had shown earlier that a fractional ideal of A is determined by its local images. Corollary 4. a b if and only if bja. Proof: Containment is preserved by localization, so since the multiplicity of each prime factor of a and b is determined by localizing, each prime factor of b has exponent less than or equal to that of a, so bja. The converse is trivial.
k i

p1 p2 pk , which, by the above lemma, implies that pi = pj for some j k (in fact, p1 \ : : : pk = p1 pk , as the pj , being maximal, must be relatively prime, but

2. Rings of Integers Let K be a number eld (i.e., a nite extension of Q , and OK its ring of integers (i.e., the integral closure of Z in K ). We wish to show that OK is a Dedekind domain (and in particular, it is a ring). Lemma 4. Z is a Dedekind domain. Proof: In fact, any principal ideal domain is a Dedekind domain (see exercises). Proposition 4. If K is a nite separable extension of F , and F is the fraction eld of a Dedekind domain A, and B is the integral closure of A in K , then B is nitely generated as an A-module and a Dedekind domain (in particular, B is a ring). Proof: See Se] Props 9 and 10 for all but the fact that B is integrally closed, and A-M] Cor 5.5 for that. Example 1. Z p 3] p p We show that the ring Z 3] is not the ring of integers of Q ( 3) by producing an ideal which cannot be factored into prime ideals, thus showing that it isn't Dedekind (of course, a much simpler method would be to simply produce an element p p 3) p of Q ( p r Z 3] which is integral over Z, such as 1+ 2 3 ). Consider thep p p 3) 2 (2), but neither (1+ ideal (2) in Z p 3]. It is not prime, as (1+ 3)(1 3) nor (1 3) is in the ideal.pOn the other hand, the only prime ideal containing p p it is (2; 1 + 3), and (2; 1 + 3)2 = (4; 2 + 2 3) which is strictly contained in (2).

BRIAN OSSERMAN

3. Scheme-Theoretic Curves It turns out that from a geometric standpoint, Dedekind domains are precisely what one wants to study if one is interested in nonsingular curves. This makes sense intuitively, as being dimension 1 corresponds to the curve having dimension 1, and being integrally closed in the eld of fractions corresponds to being nonsingular (this is much less obvious, and is in fact only true in dimension one; in higher dimensions, a variety can be normal but still singular). This motivates the following de nition: De nition 4. A Noetherian integral separated scheme S is called a (schemetheoretic) curve if it can be covered by a ne schemes Spec Ai, with each Ai a Dedekind domain. Note that since we have shown that being Dedekind is a local property, this de nition actually implies that any a ne open of S must be a Dedekind domain, and it is equivalent to the requirement that every local ring (except at the generic point) should be a discrete valuation ring. We will see that we can use this general de nition to simultaneously make constructions and prove theorems which apply with an arithmetic avor to rings of integers and a geometric avor to algebraic curves over a eld. Geometrically, ideals of an a ne variety correspond to closed subschemes, and unique factorization into prime ideals is saying, in essence, that every closed subscheme has a well-de ned multiplicity in each irreducible component, and moreover that it is determined by these multiplicities. We give two non-examples, corresponding to a ring failing to be Dedekind due to singularities (non-normality) or being of dimension greater than 1: Example 2. The singular plane curve y2 = x3. In k x; y]=(y2 x3 ), we have the ideal (x); the only prime ideal containing it is (x; y) (and this is a strict containment), but (x; y)2 = (x2 ; xy; y2 ) = (x2 ; xy; x3 ), which is strictly contained in (x). Thus, (x) doesn't have a well-behaved multiplicity at (0,0). Example 3. The a ne plane. In k x; y], we have (x2 ; y), which is contained only in (x; y), but strictly contains (x; y)2 = (x2 ; xy; y2). So (x2 ; y) also doesn't have a nicely behaved multiplicity at (0; 0). Se] Serre, Local Fields. A-M] Atiyah and MacDonald, Introduction to Commutative Algebra. 5. Exercises 1) Show that if A is a domain, and S a multiplicative subset containing 1 but not 0, then there is a one to one correspondence between primes of S 1A and primes of A disjoint from S , and that it is given by sending p to pS 1A in one direction, and p0 to p0 \ A in the other. 2) Show that ideal multiplication and the (:) operator both commute with localization. 4. References

RINGS OF INTEGERS, DEDEKIND DOMAINS, AND SCHEME-THEORETIC CURVES

3) Show if A is a Dedekind domain, unique factorization of elements into irreducibles implies that A is a principal ideal domain. 4) Show that any principal ideal domain is a Dedekind domain. 5) Show that if A has fraction eld K , an A-submodule a of K is a fractional ideal if and only if it is nitely generated. Show also that if a and b are fractional ideals, (a : b) is a fractional ideal as well.

You might also like