i NOTES ON LOCAL COHOMOLOGY C. W. WILKERSON Lecture One These notes began as an attempt to understand the statement of the Benson-Carlson theorem in group cohomology. The information was mostly gleaned from the book ”Twenty-four hours of local cohomolgy” by Iyengar, Leuschke, Leykin, Miller, Miller, Singh, and Walther. 0.1. Theorem. Let G be a finite group, k a field, and M a k[G] module. Then there is a spectral sequence ∗,∗ Hm (H ∗ (G, M)) =⇒ Σ−d H∗ (G, M) . ∗,∗ Here Hm is the (graded) local cohomology with respect to the homogeneous augmentation ideal and d is the Krull dimension of H ∗ (BG, k). The convention here is that homology modules have negative gradings while cohomology modules have positive gradings. 0.2. Corollary. If H ∗ (G, k) is Cohen-Macaulay, the spectral sequence ∗,∗ above collapses, Hm (H ∗ (G, k)) = Σ−d H∗ (G, k), and H ∗ (G, k) is Gorenstein. These notes will not discuss the construction of the spectral sequence but should allow us to see how Cor. 0.2 follows from the main theorem. In particular, local cohomology and its connections to Cohen-Macaulay and Gorenstein rings will be discussed. Let R be a commutative ring with unit, M a left R-module, and a an ideal of R. In the applications R is often Noetherian. This is also required for many of the constructions. The a-torsion torsion elements Γa(M) = {m ∈ M|an m = 0 for n >> 0} is the set of elements in M annihilated by some power of a. 0.3. Lemma. Γa(−) is a left exact additive functor from R-modules to R-modules. That is 1 2 C. W. WILKERSON a) f : M → N induces Γa(f ) : Γa(M) → Γa(N) . b) If M is isomorphic to A ⊕ B, then Γa(M) is isomorphic to Γa(A) ⊕ Γa(B). c) If 0 → A → B → C → 0 is exact then 0 → Γa(A) → Γa(B) → Γa(C) is exact. This may fail to be exact on the right. For example R = Z, a = (2), and 0 → Z → Z → Z/2Z → 0 . Since Γa(−) is left exact, it has right derived functors {Rj Γa(−)}. Here R0 Γa(M) = Γa(M). These R-modules are denoted as {Haj (M)} and for each short exact sequence as above produce a long exact sequence 0 → Γa(A) = Ha0 (A) → . . . Ha0 (C) → Ha1 (A) → . . . → Ha1 (C) → . . . → Haj (A) → Haj (B) → Haj (C) → . . . More explicitly, take a resolution {I ∗ } of M by injective R-modules, a chain complex {I ∗ = 0 → I 0 → I 1 → I 2 → . . .} with H 0 (I ∗ ) = M and H j (I ∗ ) = 0 for j > 0. Then Haj (M) = H j (Γa(I ∗ )). In the context of sheaves over schemes, let X = Spec(R), Y = V (a) the subscheme associeted to a, and M̃ the sheaf corresponding to the R-module M. Then Γa(M) = ΓY (X, M̃ ) and the local cohomology modules are isomorphic to the sheaf cohomology modules with support in Y , HY∗ (X, M̃ ). Calculating from first principles is only practical for small examples for which the injective resolutions are apparent. Recall that if I is an injective R-module, then it is divisible. That is, if a ∈ I, r ∈ R a non-zero-divisor, there exists b ∈ I with rb = a. Conversely, if R is a PID, then divisible modules are injective. Our first examples treat this case. 0.4. Example. Let R = Z and a = (p), for some prime p. a) Let M = Z/nZ where n is a nonzero integer. Then 0 → Q/Z → Q/Z → 0 is an injecxtive resolution, where the nonzero map is multiplication by n. Then Γp(Q/Z) = Z/p∞ Z and the induced map is NOTES ON LOCAL COHOMOLOGY 3 multiplication by n. Hence Hp0 (Z/dZ) = Z/pe Z and Hpj (Z/dZ) = 0 if j > 0. Here pe is the largest power of p that divides n. b) Let R = Z, p = (p), and M = Z. Then 0 → Q → Q/Z → 0 is an injective resolution, since Q and Q/Z are divisible abelian groups. Γp(Q) = 0 and Γp(Q/Z) = Z/p∞ Z = Z[1/p]/Z. Z/p∞ Z is generated over Z by {1/p, 1/p2, . . . 1/pn , . . .}, or indeed any infinite subsequence of these. Thus Hp0 (Z) = 0 and Hp1 (Z) = Z/p∞ Z. c) Let R = Z, a = (0), and M any abelian group. Since Γ(0) (A) = A j 0 for all A, H(0) (M) = M and H(0) (M) = 0 for j ≥ 1. That is, Γ(0) (−) is the identity functor, hence exact, and therefore has trivial higher derived functors. d) Let R = Z, a = (1), and M any abelian group. Then Γ(1) (A) = 0 for j all A, and H(1) (M) = 0 for j ≥ 0. That is, Γ(1) (−) is the zero functor, hence exact, and therefore has trivial higher derived functors. Note that in example b) the nonzero local cohomology group is not finitely generated. This non-finite generation is common for local cohomology. Note also that Z/p∞ Z is Artinian (descending chains of submodules terminate). This too is typical for local cohomology. The next example of a polynomial algebra on one variable is directly applicable to the main theorem in the case G = Z/2Z and k = F2 . It is also analogous to the calculations above with Z. 0.5. Example. Let R = k[x] where k is a field and m is the maximal ideal generated by x. Then L, the fraction field of R,i k(x) is divisible and hence injective. a) Let M = k = R/m. Then 0 → L/R → L/R → 0 is an injective resolution of M, where the nonzero map is multiplication by x. Then Γm(L/R) = R[1/x]/R, which is generated over R by any infinite subsequence of {1/x, 1/x2 , . . . 1/xn , . . .}. The kernel of multiplication by x is generated by the equivalence class of {1/x} and is isomorphic to 0 k with the trivial action of k[x] on it. Thus Hm ((k) = k. On the other hand multiplication by x is a surjective endomorpism of R[1/x]/R, so 1 Hm (k) = 0. 4 C. W. WILKERSON b) Let M = R. Then 0 → L → L/R → 0 is an injective resolution of 0 R. Γm(L) = 0 and as before Γm(L/R) = R[1/x]/R. Thus Hm (k[x]) = 0 1 and Hm(k[x]) = k[x][1/x]/k[x]. This can be thought of as the sub-k[x]module of k(x)/k[x] generated by the classes {1/x, 1/x2 , . . . , 1/xn , . . .}. 0.6. Example. The constructions involved in defining local cohomology can be done in the context of graded algebras and modules, with homogeneous ideals. So let’s consider the graded polynomial algebra k[x], where the grading of x is 1. Take m = (x) and M = R = k[x]. Now in the graded category the injective hull of k[x] is k[x, x−1 ], so one has the injective resolution 0 → k[x, x−1 ] → k[x, x−1 ]/k[x] → 0. These modules are now Z-graded. Γm(k[x, x−1 ]) = 0 and Γm(k[x, x−1 ]/k[x]) = 0 1 k[x, x−1 ]/k[x]. Hence Hm (k[x]) = 0 and Hm (k[x]) = k[x, x−1 ]/k[x]. It can be visualized as the ideal yk[y] where k[y] is the polynomial algebra on a generator of degree −1. The k[x]-action is xy a = y a−1 if a − 1 > 0 and 0 otherwise. Thus the local cohomology has a bigrading. If bidegree is (i, j) and i 6= 1, the component is 0. The component for (1, −j) is k for j ≥ 1 and 0 otherwise. 0.7. Example. (The example G = Z/2Z, k = F2 ) H ∗ (Z/2Z, F2 ) = F2 [x] where x has degree 1. The previous example calculated the local cohomology as yk[y] where y has degree −1. Thus the Benson-Carlson theorem holds by calculation in this example. NOTES ON LOCAL COHOMOLOGY 5 Lecture Two : Computations with Koszul and Čech complexes If R is a Noetherian ring, then each injective is a direct sum of indecomposable injectives, {E(R/p)}, where p ranges over the prime ideals of R, and E(R/p) is the injective hull. This is a minimal essential injective super-module and is unique up to isomorphism. Thus the injective resolution of an R-module M is formed from copies of these basic modules. 0.8. Theorem. Let R be a Noetherian ring and a an ideal of R. Then for each prime ideal p of R, Haj (E(R/p)) = 0 if j ≥ 1 and Ha0 (E(R/p)) = E(R/p) if a ⊂ p and is zero otherwise. The multiplicities of the minimal injectives in a resoltion of a module M are called the Bass numbers of M and are subject to much research. Fortunately, one does not always need a resolution to calculate the local cohomology. This lecture will discuss some other methods of calculation. The sequence of maps . . . R/an → R/an−1 → . . . R/a2 → R/a gives rise to a direct limit HomR (R/a, M) → HomR (R/a2 , M) . . . HomR (R/an , M) → . . . n and one has Γa(M) = − lim →n HomR (R/a , M) in a functorial manner. Applying this to an injective resolution of M and using the exactness of filtered direct limits yields the following: ∗ n 0.9. Theorem. Ha∗ (M) = − lim → ExtR (R/a , M). n One can regard this as a hint of constructions we’ll see later in which M is replaced by its R/a-approximation, CellR/a(M) which has as its homotopy groups the local cohomology groups {Haj (M)}. In the direct limit above the sequence {an } can be replaced with any cofinal sequence of ideals. For example if n is a nilpotent ideal, in the 6 C. W. WILKERSON limit one gets the zero ideal and hence Γn(−) is the identity functor. Likewise in general, an ideal can be replaced by its radical for the purposes of local cohomology. 0.10. Corollary. If R is Noetherian, any element of Haj (M) is annihilated by some power of a. By the previous theorem, each element in local cohomology is in the image of one from ExtjR (R/an , ), which is annihilated by an . We need methods of calculating local cohomology that are more directly related to the structure of a and R. 0.11. Definition. a) Let x ∈ R. The Koszul complex K ∗ (x, R) is the sequence x 0→R− →R→0 , where the first R is in homological degree −1, the second in degree 0, and the map is multiplication by x. H j (x, M) is H j (K ∗ (x, R) ⊗R M). b) Let x = {x1 , x2 , . . . xr } be a sequence of elements in R. O K ∗ (x, R) = K ∗ (xi , R) R,i is the Koszul complex on x. c) H j (x, M) = H j (K ∗ (x, R) ⊗R M). 0.12. Definition. a) x is a regular element in R if it is not a zerodivisor and xR 6= R. If M is an R-module, x is a M-regular element if xm = 0 implies m = 0 in M and xM 6= M. b) x = {x1 , x2 , . . . xr } is a regular sequence in R if x1 is regular and each xk is regular for R/(x1 , . . . , xk−1 )R. 0.13. Corollary. x is M-regular if and only if H −1 (x, M) = 0 and H 0 (x, M) 6= 0. Note that H 0 (x, M) = M/xM. In general, for x = {x1 , . . . , xd }, H −d (x, M) = (0 :M x), the elements in M annihilated by the x and H 0 (x, M) = M/xM. 0.14. Theorem. Let (R, m) be a local ring, {x1 , . . . xn } ⊂ m, and M a finitely generated R-module. Then H j (x, M) = 0 for j < n if and only if x is a regular M-sequence. NOTES ON LOCAL COHOMOLOGY 7 The second complex needed is a limit of Koszul complexes, in a sense to be explained later. 0.15. Definition. The Čech complex Č ∗ (x, R) is the complex 0 → R → Rx = R[1/x] → 0 where the non-trivial map is localization at the multiplicative set S = {1, x, x2 . . .}. The leftmost R is in homological degree 0 and the next in degree 1. We also have Č ∗ (x, R) ⊗R M = {0 → M → Mx → 0} . N ∗ Similarly, for x = {x1 , x2 , . . .}, define Č ∗ (x, R) = R Č (xi , R) with the appropriate sign convention on the differentials. Notice that we have M → ⊕i Mxi at the 0-stage of Č ∗ (x, R) ⊗R M. 0.16. Lemma. Let M be a R-module and a generated by {x1 , x2 , . . .}. Then H 0 (Č ∗ (x, R) ⊗R M) is naturally isomorphic to Γa(M). This lemma together with showing that the higher Čech cohomology vanishes on injective modules shows that it must coincide with the local cohomology. This requires the Noetherian condition in general. 0.17. Theorem. Let R be a Noetherian ring and a = (x1 , x2 , . . .). Then Haj (M) is natually isomorphic to H j (Č ∗ (x, R) ⊗R M). Finally, we wish to relate the Čech complex to the Koszul complex. First, for a given x ∈ R we have the chain map K ∗ (xt+1 , R) → K ∗ (xt , R) given by 0 −−−→ y xt+1 R −−−→ x y R −−−→ = y 0 y 0 ←−−− R ←−−− R ←−−− x x x = x 0 x xt 0 −−−→ R −−−→ R −−−→ 0 Applying HomR (−, R), we get xt+1 xt 0 ←−−− R ←−−− R ←−−− 0 8 C. W. WILKERSON That is, we get, after observing that x x Rx = R[1/x] = − lim → ... − → . . .}, →{R − ∗ t that Č(x, R) = Σ1 lim −→t K (x , R). This argument generalizes to a sequence x of elements. In general, the KoszulNcomplex has a duality, in that HomR (K ∗ (x, R), M) = Σ+c (K ∗ (x, R) R M), where c is the cardinality of x. Thus local cohomology inherits the properties of Koszul homology for regular sequences. 0.18. Corollary. If a contains a M-regular sequence of length d, then Haj (M) = 0 for 0 ≤ j ≤ d − 1. If a is generated by n elements, then Haj (M) = 0 for j > n. 0.19. Example. Let R = k[x1 , . . . , xd ] be the polynomial algebra on d generators of degree 1 over a field k and m the maximal ideal generated j by the {x′i s}. Let M = R. Then Hm (R) = 0 for j < d , for j > d, and d Hm (R) = (y1 y2 . . . yd )k[y1 , y2, . . . yd ] where each yk has degree −1. The R-action is given by xi y1a1 . . . yiai . . . ydad = y1a1 . . . yiai −1 . . . ydad if ai ≥ 2 and 0 otherwise. NOTES ON LOCAL COHOMOLOGY 9 Lecture Three: Cohen-Macaulay, Gorenstein, and Duality In the hierarchy of commutative local rings, regular local rings occupy the highest rung. These are the local analogue of polynomial algebras in the graded world. One step down are the complete intersections, that is, quotient rings of the form polynomial or regular local mod an ideal generated by a regular sequence. There are two useful lower rungs that are not so well known outside of commutative algebra. We discuss the most general case first. 0.20. Definition. Let (R, m, K) be a local ring. If m contains a regular sequence {x1 , . . . , xd } of length d = dim(R), R is said to be CohenMacaulay. In general, the maximal length of an R-regular or M-regular sequence is the depth of R or M We use the analogous definitions in the graded case. 0.21. Example. For k a field, the polynomial algebra k[x1 , . . . , xd ] is a graded Cohen-Macaulay ring. Any regular local ring is Cohen-Macaulay, as is any complete intersection. 0.22. Theorem. (Noether Normalization) Let R be a connected graded finitely generated commutative algebra over the field k. There exists a sequence of homogeneous elements {x1 , . . . , xd } of R such that k[x1 , . . . , xd ] is a polynomial subalgebra of R and R is a finitely generated module over k[x1 , . . . , xd ]. These can be chosen so that R is a free module over k[x1 , . . . , xd ] if and only if R is a Cohen-Macaulay ring. 0.23. Corollary. Let R∗ be a graded connected algebra over the field k P ∞ and Pt (R∗ ) = i=0 dimk (Ri ) the Poincare-Hilbert series for R∗ . If R∗ is Cohen-Macaulay, then Y Pt (R∗ ) = g(t)/ (1 − tni ) where g(t) is a polynomial with non-negative integer coefficients. 0.24. Example. Let R = k[x, y/(xy). dim(R) = 1, and taking z = x+y, R is freely generated over k[z] by {1, y}. The regular sequence can be taken to be {z}. Pt (R) = (1 + t)/(1 − t). Cohen-Macaulay also has a characterization in terms of local cohomology: 10 C. W. WILKERSON 0.25. Theorem. Let (R, m) be a local ring. R is Cohen-Macaulay if j and only if Hm (R) = 0 for all j 6= dim(R). In particular, this implies that in the Benson-Carlson spectral sed quence, there is only one non-zero vertical line, Hm . For local rings, the rung between Cohen-Macaulay rings and complete intersections is occupied by Gorenstein local rings. Identified first by Bass and Serre, Gorenstein rings have many characterizations. The standard one is somewhat non-obvious. 0.26. Definition. A local ring (R, m, K) is Gorenstein if and only if it has a finite injective resolution over itself. The first case to understand is when R has zero Krull dimension. 0.27. Theorem. Let (R, m, K) be a zero-dimensional local ring. The following conditions are equivalent: a) R is Gorenstein; b) R is injective as an R-module; c) dimK soc R = 1 ; d) ER (R/m) = R; Recall that socR M = HomR (R/m, M) = (0 :M m). For example, if X is a compact closed oriented manifold, its cohomology algebra H ∗ (X, Q) satisfies Poincaré duality. That is, the cohomology is i=n M R= Ri i=0 and the cup product gives a perfect pairing Ri ⊗K Rn−i → Rn for all relevant i. 0.28. Lemma. If R is a graded commutative algebra over the field K, finite dimensional as a K-vector space, and satifies Poincaré duality, then R is a Gorenstein ring. This follows immediately from the theorem above, since the socle is generated by the top class of R. NOTES ON LOCAL COHOMOLOGY 11 We can understand Gorenstein rings inductively, via the next theorem: 0.29. Theorem. Let (R, m, K) be a local ring and {x1 , . . . , xc } a regular sequence. Then R is Gorenstein if and only if R/(x1 , . . . , xc ) is Gorenstein. 0.30. Theorem. Let (R, m, K) be a Noetherian local ring and M a finitely generated R-module. Then a) depthR M = min{i|ExtiR (K, M) 6= 0}. b) injdimR M = sup{i|ExtiR (K, M) 6= 0}. c) depthR M ≤ dim M and injdimR M ≥ dim M. d) if injdimR M is finite, injdimR M = depthR R. Thus by c) and d), if injdimR R is finite, depthR R = dim R, so Gorenstein implies Cohen-Macaulay. 0.31. Example. a) Regular local rings are Gorenstein. b) Complete interection rings are Gorenstein. c) Gorenstein rings are Cohen-Macaulay, but Cohen-Macaulay rings need not be Gorenstein. 0.32. Theorem. If (R, m, K) is Gorenstein, then it is Cohen-Macaulay. j In particular (R, m, K) is Gorenstein if and only if Hm (R) = 0 for d j 6= d = dim(R) and Hm(R) = ER (K). ER (K) can be used to form a type of dual for R-modules: 0.33. Definition. Let (R, m, K) be a local ring and M an R-module. Then M ˇ= HomR (M, ER (K)). Note that M 6= Mˇˇ in general. Rather, it is the m-adic completion of M. For example, with R = Z and m = (p), EZ (Fp ) = Z/p∞ Z, and Zˇˇ= HomZ (Z/p∞ Z, Z/p∞ Z) = Zpˆ, the p-adic integers. 0.34. Theorem. (Local Duality) Let (R, m, K) be a d-dimensional local Gorenstein ring and M a finitely generated R-module. Then i Hm (M) ≈ Extd−i R (M, R)ˇ . 12 C. W. WILKERSON Alternately, i R Hm (M) ≈ T ord−i (ER (K), M) . The second form of the duality follows directly from realizing that the ∗ Čech complex Č(x, R) gives a shifted flat resolution of Hm (R) = ER (K) since R is Gorenstein. Hence tensoring with M and taking cohomology computes a shifted T or module. If X is a non-singular affine variety, it has a cotangent vector bundle T ∗ X and associated line bundle ∧n T ∗ X. The sections of this line bundle are a locally free rank one module over the coordinate ring of the variety. It has remarkable properties and has been generalized to the singular case. 0.35. Definition. Let (R, m) be a Cohen-Macaulay local ring. A canonical module is a finitely generated R-module ω for which a) depth ω = dim(R) b) typeR ω = 1. Alternately, HomR (ω, ω) = R. c) injdimR ω is finite. These require some explanation. The depth of a module M is the length of the longest M-regular sequence. The type of a module M is dimK ExttR (K, M), where K = R/m and t is the depth of M. This is analogous to the rank one projective module in the geometric case. Note that if R is Gorenstein, then R itself is a canonical module. For local rings, the canonical module is essentially unique, in that any two are isomorphic. There is a form of local duality for Cohen-Macaulay rings with canonical modules: Not every Cohen-Macaulay local ring has a canonical module. Those that are the surjective image of a Gorenstein local ring do. In particular, complete local rings and finitely generated graded K-algebras qualify. 0.36. Theorem. Let R, m, K) be a d-dimensional local Cohen-Macaulay ring with canonical module ω and let M be a finitely generated Rmodule. Then for 0 ≤ i ≤ d there are isomorphisms i Hm (M) ≈ Extd−i R (M, ω)ˇ . d In particular, Hm (R) ≈ ωˇ. NOTES ON LOCAL COHOMOLOGY 13 In the graded case, there are some small changes. Recall that if M is graded, M[a] is the module with M[a]i = Ma+i . In the topological world, M[a] = Σ−a M. In this context the dual HomR (M, ER (K)) becomes HomK (M, K). L 0.37. Theorem. Let R = n≥0 Rn be a finitely generated graded algebra over R0 = K, a field.Then the graded injective hull of K is M ER (K) = HomK (R, K) = HomK (Rn , K) . n≥0 ER (K)−i = HomK (R, K)−i = HomK (Ri , K). If R is Gorenstein and j d-dimensional then Hm (R) = 0 for j 6= d and d Hm (R) = HomK (R, K)[a] = Σ−a HomK (R, K) for some integer a. Here the action of R on HomK (R, K) is by r(f )(m) = f (rm). We think of HomK (R, K) as being negatively graded. We have seen examples of this already in the polynomial algebra case. The point is that K → HomK (R, K) is an essenital extension. That is, any non-zero Rsubmodule of HomK (R, K) must intersect K non-trivially. Also, since HomK (R, K) is dual to the free module R, it is injective. 0.38. Example. Let R = K[x1 , . . . , xd ] where the {xi } have degree 1. Then d Hm (R) = HomK (R, K)[d] = Σ−d HomK (R, K) . 0.39. Theorem. Let R be a graded ring as above and R Gorenstein. Then the Poincaré-Hilbert series Pt (R) satifies the functional equation Pt (R) = (−1)d tN P1/t (R) for d = dim(R) and some integer N. Notice that the Corollary to the Benson-Carlson theorem now follows by calculation, since the spectral sequence of the theorem has only one d line, and must therefore collapse. The conclusion is that Hm (R) = ER (k)[d], and hence R is Gorenstein. 14 C. W. WILKERSON Section Four - Residues I’d like to go now a little beyond the Benson-Carlson material to residues. I’ll interpret this widely to mean mappings or pairings from cohomology and homology modules to the ground field. Let’s start with the nicest example. Let R be the polynomial algebra K[z] in the variable z over the field K. Think of R as being a regular local ring, either by localizing so that (x) becomes the only maximal ideal, or by putting the grading on R with degree z = 1 and using only homogeneous ideals and maps. 0.40. Definition. ΩR/K is the module of Kahler differentials of R to R over K. It’s generated over R by the symbols {dr}, subject to the relations d(rr ′) = rdr ′ + r ′ dr for r, r ′ ∈ R and dk = 0 for k ∈ K. 0.41. Proposition. For R = K[z] as above, ΩR/K is the free R-module generated by dz, i.e. R dz. P 1 0.42. Proposition. For R = K[z] and m = (z), Hm (ΩR/K ) = i<0 ai z i dz, ai ∈ K, subject to the relations z(z −1 ) = 0 and z(z −i ) = z −i+1 if i > 1. This is basically the same proof as earlier calculations that showed 1 Hm (R) = R[1/z]/R. One can use the Čech complex. 0.43. Definition. For R = K[z] and m = (z), the residue map 1 res : Hm (ΩR/K ) → K X is res( ai z i dz) = a−1 . i<0 Similar definitions and calculations can be done for the cases of R the formal power series ring in one variable, or, over C, the convergent power series ring. Thus one sees the connection to the residues of complex variable theory. There are also similar calculations for R a regular local ring of dimension d. Here ΩR/K is replaced by its d-fold d exterior power and one has res : Hm (∧d (ΩR/K )) → K. Department of Mathematics, Purdue University Department of Mathematics, Texas A & M University E-mail address: cwilkers@purdue.edu E-mail address: cwilkers@math.tamu.edu