Commutative Algebra Lecture 4: Supports of Modules • At the end of last lecture, we discussed localization of a ring R with respect to a prime ideal p, and how there is an order-preserving bijection o n o n primes of R contained in p ∼ = primes of Rp sending a prime q ⊆ p to the ideal qRp generated by its image in Rp . (This uses the general notation for multiplying an ideal I by a module M : we denote by IM the submodule of M generated by {im : i ∈ I, m ∈ M }. Then for any R-algebra A, the ideal IA of A is the one generated by the image of I in A.) In particular, Rp has a unique maximal ideal, pRp . A ring with a unique maximal ideal is called a local ring. • Here’s an extended example of a local ring that hopefully explains the terminology “local” (and “localization” more generally). Example 1. Let X be the unit interval [0, 1], although many topological spaces will do, and let R = C(X, R) be the ring of continuous functions X → R. Let x ∈ X be some point. Then m := {f ∈ R : f (x) = 0} is a maximal ideal, because it is the kernel of the surjective ring homomorphism R → R : f 7→ f (x). I claim that the local ring Rm is the ring of germs of continuous functions X → R at x, i.e. equivalence classes of pairs (U, f ) where U is an open neighborhood of x and f : U → R is a continuous function, and where (U, f ) = (V, g) if there is some neighborhood W ⊆ U ∩ V of x on which f |W = g|W . Germs can be added, subtracted, and multiplied, and hence form a ring themselves, the ring of “local” information about continuous functions near x. Every continuous function f : X → R gives a germ (X, f ), and this constitutes a ring homomorphism R → {germs}. Since any f ∈ R \ m is nonzero on an open neighborhood of x, and on that open neighborhood it has a multiplicative inverse, the germ of such an f is a unit and we obtain a map from the localization Rm → {germs}. 1 This map is injective: if f ∈ R and g ∈ R \ m, and the germ of f /g near x is zero, then f vanishes on some neighborhood U of x. Let b be a “bump function”: a continuous function that is zero outside of U but nonzero at x. Then b ∈ R \ m and b · f = 0, so f /g = 0 in the localization Rm . The map is also surjective: given any germ (U, f ), we can use our bump function b to define a new function ( f (y)b(y) if y ∈ U g : y 7→ 0 if y ∈ /U which is continuous on all of X. Then the germ of g/b at x is equal to (U, f ), by restricting both f and g to the neighborhood of x where b 6= 0. So (U, f ) is in the image of Rm . • Given a continuous function f : X → R, we can define its support as the set of points x ∈ X where the germ of f at x is nonzero. This set is always closed, since “having zero germ at x” is an open condition on x: if it’s true at x it’s also true on a neighborhood of x. There is an analogous definition for the support of an element of any module over a ring: Definition 2. Let R be a ring and M an R-module. Let m ∈ M . The support of m, denoted Supp(m), is the set of prime ideals p ⊆ R such that m 67→ 0 in Mp . (For those of you who haven’t already taken some algebraic geometry, the set of prime ideals of a ring R is called its spectrum and denoted Spec(R). This set is given a topology where the closed sets are those of the form V (I) := {primes p ⊆ R that contain I} for some ideal I. We think of elements of R as functions on this topological space; the subset V (I) is the set of points where all the functions in I vanish.) • Today’s main theorems are about these supports of module elements. The first result is that an element is zero if and only if its support is empty, just as a function is identically zero if and only if its support is empty: Lemma 3. Let R be a ring and M an R-module with an element m. Let f1 , . . . , fn be elements of R generating the unit ideal. Then the following are equivalent: 1. m = 0. 2. m 7→ 0 in Mfi for each i ∈ {1, . . . , n}. 3. m 7→ 0 in Mp for each prime ideal p ⊆ R (i.e. Supp(m) = ∅.) 4. m 7→ 0 in Mm for each maximal ideal m ⊆ R. Proof. Each statement cyclically implies the next: 2 1⇒2. If m = 0, then m 7→ 0 in every localization Mf . 2⇒3. Given a prime ideal p, not every fi can be contained in p, so choose fi ∈ / p with m 7→ 0 in Mfi . The key is that Mp ∼ = (Mfi )p , since (Mfi )p ∼ = Rp ⊗R (Rfi ⊗R M ) ∼ = (Rp ⊗R Rf ) ⊗R M i ∼ = (Rp )fi ⊗R M but (Rp )fi = Rp since fi is already a unit in Rp , so (Mfi )p ∼ = Rp ⊗ R M ∼ = Mp . Then since m 7→ 0 in Mfi , it maps to zero in Mp as well. (Or in other words, since fi acts invertibly on Mp , the canonical homomorphism M → Mp factors via the homomorphism M → Mfi sending m to zero, so m 7→ 0 in Mp . Or in still other words, knowing that m/1 = 0 in Mfi , we can find a power fik of fi for which fik m = 0 in M ; then fik ∈ / p since p is prime, so m 7→ 0 in Mp .) 3⇒4. Every maximal ideal is prime, so statement 4 is just a weaker version of statement 3. 4⇒1. For each maximal m, the fact that m 7→ 0 in Mm tells us that there is fm ∈ /m such that fm · m = 0. Then the ideal generated by all the fm is contained P in no maximal ideal, hence is the unit ideal and 1 can be written as 1 = ni=1 ri fmi for some Pn r1 , . . . , rn ∈ R and specific maximal ideals m1 , . . . , mn ⊆ R. Then m = i=1 ri fmi m = 0. • The support of an element is closely related to its annihilator : Definition 4. Let R be a ring, M an R-module, and m ∈ M an element. The annihilator of m in R is the ideal AnnR (m) := {r ∈ R : r · m = 0}. Sometimes we will leave off the R subscript if there can be no confusion. Lemma 5. Let R be a ring and p ⊆ R a prime ideal. Let M be an R-module and m ∈ M . Then p ∈ Supp(m) if and only if p ⊇ Ann(m). Proof. Suppose that m 67→ 0 in Mp , that is, m/1 6= 0/1 in Mp . This is the same as saying that for all f in the multiplicative subset R \ p, we have f · m 6= 0. Or taking the contrapositive, this is the same as saying that for all f ∈ R such that f · m = 0, we must have f ∈ p, i.e. Ann(m) ⊆ p. 3 A geometric consequence of this lemma is that the support of any single element of an R-module is a closed subset of Spec(R), namely V (Ann(m)). Note also that the knowledge that the annihilator of an element m ∈ M is always an ideal, which is easy to verify, can clean up the proof of 4 ⇒ 1 in Lemma 3: if m 7→ 0 in Mm , then Ann(m) 6⊆ m; if this holds for all maximal ideals m, then Ann(m) must be the unit ideal (1) and so 1 · m = 0. • We have the following module-wide version of Lemma 3: Lemma 6. Let R be a ring and M an R-module. Let f1 , . . . , fn ∈ R together the unit ideal. Then the following are equivalent: 1. M = 0. 2. Mfi = 0 for each i ∈ {1, . . . , n}. 3. Mp = 0 for each prime ideal p ⊆ R. 4. Mm = 0 for each maximal ideal m ⊆ R. Proof. The implications 1 ⇒ 2 ⇒ 3 ⇒ 4 run as before. For 4 ⇒ 1, let m ∈ M . Them m 7→ 0 in each localization Mm at a maximal ideal (because each of these localizations is itself zero), so m = 0. Since this holds for each element of M , we have M = 0. • We can also make a module-wide definitions of support and annihilator: Definition 7. Let R be a ring and M an R-module. The support of M , denoted Supp(M ), is the set of prime ideals p ⊆ R such that Mp 6= 0. The annihilator of M is the ideal AnnR (M ) := {r ∈ R : rM = 0}. • For example, M = 0 if and only if Ann(M ) = (1) and, and by Lemma 6, if and only if Supp(M ) = ∅. Is it always the case that Supp(M ) = V (Ann(M )), as in Lemma 5? No: Example 8. Consider the support and annihilator of the Z-module M M= Z/pZ. p∈N prime For which prime ideals p ⊆ Z is Mp nonzero? Localizing at p is the same as tensoring with Zp , which preserves direct sums, so we have M Mp = (Z/pZ)p . p∈N prime If p = (p) for some prime number p, then since everything in Z \ (p) already acts invertibly on Z/pZ, we have (Z/pZ)(p) = Z/pZ 6= 0. So these prime ideals are all in the support of Z. 4 But the prime ideal (0) ⊆ Z is not in the support: Z(0) = Q, and (Z/pZ) ⊗ Q = 0 for all prime numbers p, so M(0) = 0. The annihilator of M , however, is zero: Given any nonzero n ∈ Z, choose a prime number p not dividing n, which thus has n · Z/pZ = Z/pZ 6= 0. So the only element of Z that kills all of M is zero: Ann(M ) = 0. Then in particular: Ann(M ) ⊆ (0) but (0) ∈ / Supp(M ), so Supp(M ) 6= V (Ann(M )). In fact, Supp(M ) is not even a closed subset of Spec(R), since the only closed subsets are of the form V ((0)) = Spec(R) and, for each n 6= 0, the subset V ((n)) = {(p) : p is a prime dividing n}. • We shall see that the equivalence between p ∈ Supp(M ) and p ⊇ Ann(M ) can only fail if M is not finitely generated, like the direct sum above with infinitely many summands. The first step is a pair of dual identities given a list of generators for our module: Lemma 9. Let R be a ring and M an R-module, and suppose {mi : i ∈ I} is a generating set for M . Then T 1. Ann(M ) = i∈I Ann(mi ). S 2. Supp(M ) = i∈I Supp(mi ). Proof. For the first claim, T we have an easy containment Ann(M ) ⊆ Ann(mi ) for each i, so Ann(M ) ⊆ i Ann(mi ). Conversely, if r ∈ Ann(mi ) for all i ∈ I, then r · mi = 0 for all i ∈ I. Given any m ∈ M , we can write m as a finite linear combination of the mi : n X rj · mij , m= j=1 and then r · m = P j rj (r · mij ) = 0. So rM = 0 and r ∈ Ann(M ). S For the second claim, we again have the easy inclusion Supp(M ) ⊇ i Supp(mi ). For the reverse containment, suppose p ∈ / Supp(mi ) for all i ∈ I. Then each mi 7→ 0 in Mp ; since the mi generate M (and hence their images generate Mp as an Rp -module), we must have Mp = 0 and p ∈ / Supp(M ). As an easy corollary, if p ∈ Supp(M ) then p ∈ Supp(m) for some m ∈ M , so p ⊇ Ann(m) ⊇ Ann(M ). So we always have the inclusion Supp(M ) ⊆ V (Ann(M )). • So we find that if {mi : i ∈ I} generates M , then p ∈ Supp(M )Tif and only if p ⊇ Ann(mi ) for some i ∈ I. And p ⊇ Ann(M ) if and only if p ⊇ i∈I Ann(mi ). Are these equivalent? We know they can’t be in general, but if I is finite they are: 5 Lemma 10. Let I and J be ideals of a ring R, and let p ⊆ R be a prime ideal. The following are equivalent: 1. p ⊇ I ∩ J. 2. p ⊇ IJ. 3. p ⊇ I or p ⊇ J. T Thus by induction, a prime ideal p containing a finite intersection ni=1 Ii is equivalent to p containing oneTof the ideals Ii . Note that this is not true for infinite intersections in general: p∈N prime (p) = (0), but (0) contains no prime p. Also, infinite products of ideals don’t make sense. Proof. (1 ⇒ 2): Follows from I ∩ J ⊇ IJ. (2 ⇒ 3): If J 6⊆ p, then choose j ∈ J \ p. Then ij ∈ p for all i ∈ I, so i ∈ p for all i ∈ I and I ⊆ p. (3 ⇒ 1): If p contains I or J, it certainly contains the smaller ideal I ∩ J, which is contained in both I and J. (Another proof of 1 ⇔ 2: since (I ∩ J)2 ⊆ IJ ⊆ I ∩ J, they have the same radical and are contained in the same prime ideals.) • So we obtain the following corollary analogous to Lemma 5: Corollary 11. Let R be a ring and p ⊆ R a prime ideal. Let M be a finitely generated R-module. Then p ∈ Supp(M ) if and only if p ⊇ Ann(M ). The geometric consequence is that if M is a finitely generated A-module, the support of M is a closed subset of Spec(A), namely the closed subset V (Ann(M )). Proof. We said already that if {m1 , . . . , mn } generates M as an A-module, then p ∈ Supp(M ) if and only if pTcontains Ann(mi ) for some i ∈ {1, . . . , n}. This is true if and only if p contains ni=1 Ann(mi ) = Ann(M ). 6