AN INTRODUCTION TO SPECTRAL SEQUENCES STEVEN DALE CUTKOSKY 1. Introduction In these notes we give an introduction to spectral sequences. Our exposition owes much to the treatments in Eisenbud’s book [2] and in Rotman’s book [4]. 2. Complexes and Differential Modules Suppose that R is a commutative ring with identity. A complex of R-modules is a sequence of R-module homomorphisms (1) F : ··· → F d→i+1 F →di F → ··· i+1 i i−1 where d2 = 0, or a a sequence of R-module homomorphisms (2) F(cid:48) : ··· → Fi−1 d→i−1 Fi →di Fi+1 → ··· where d2 = 0. The differential d has degree -1 in (1) and degree +1 in (2). We will often write F∗ (or (F(cid:48))∗) to denote these complexes. Suppose that M is an R-module, and (3) ··· → F → F → F → M → 0 2 1 0 is a free resolution of M. Suppose that N is an R-module. Then Hom (F ,N) → Hom (F ,N) → ··· R 0 R 1 is a complex of degree 1, and ··· → Hom (N,F ) → Hom (N,F ) R 1 R 0 is a complex of degree -1. The homology of F at F in (1) is H (F) = ker d /im d . The homology of F(cid:48) at Fi i i i i+1 in (2) is Hi(F(cid:48)) = ker d /im d . i i−1 The complex (1) (with differential of degree -1) can be converted to a complex of degree +1 by setting F−i = F . For instance, (3) becomes the exact complex i ··· → F−2 → F−1 → F0 → M → 0 of degree 1. A differential module (F,ϕ) is an R-module F with and endomorphism ϕ such that ϕ2 = 0. The homology of F is H(F) = ker ϕ/imϕ. The differential module F is exact of H(F) = 0. If (F,ϕ) and (G,ψ) are differential modules, then a homomorphism of differentiable modules Λ : F → G is an R-module homomorphism which satisfies Λ◦ϕ = ψ◦Λ. Suppose that F is a complex (which we will assume has degree -1, although the con- struction with “upper indices” is valid when the degree is 1). Then F can be viewed as (cid:76) a differential module F, by setting F = F . The map ϕ : F → F is defined by the i∈Z i ∼ (cid:76) differentials d : F → F . We have that H(F) = H (F). i i i−1 i∈Z i 1 If β 0 → F(cid:48) →α F → F(cid:48)(cid:48)→ 0 is a short exact sequence of differential modules, then there is a connecting homorphism δ : H(F(cid:48)(cid:48)) → H(F(cid:48)), making H(F(cid:48)) →α H(F) (4) δ (cid:45) (cid:46) β H(F(cid:48)(cid:48)) an exact triangle; that is ker α = im δ, ker δ = im β and ker β = im α. Th construction of δ is as in the construction of the connecting homomorphism of a short exact sequence of complexes. If (4) is a short exact sequence of complexes, then δ is the direct sum of the connecting homorphisms of the homology modules of the complexes. 3. Spectral Sequences Definition 3.1. A Spectral sequence is a sequence of modules rE for r ≥ 1, each with a differential d : rE → rE satisfying d d = 0, such that r+1E ∼= ker d /im d = H(rE). r r r r r Suppose that rE is a spectral sequence. We will define submodules 0 = 1B ⊂ 2B ⊂ ··· ⊂ rB ⊂ ··· ⊂ rZ ⊂ ··· ⊂ 2Z ⊂ 1Z = 1E suchthatiE = iZ/iB foreachi. Todothis,let1Z = 1E and1B = 0,sothat1E = 1Z/1B. Having defined iB and iZ for i ≤ r, we define r+1Z as the kernel of the composite map rZ → rZ/rB = rE →dr rE = rZ/rB and let the image of this map be r+1B/rB. We have that r+1Z/r+1B = H(rE) = r+1E and rB ⊂ r+1B ⊂ r+1Z ⊂ rZ as required. Having defined all the iZ and iB, we set ∞Z = ∩Z , ∞B = ∪B . r r We define the limit of the spectral sequence to be ∞E = ∞Z/∞B. We say that the spectral sequence collapses at rE if rE = ∞E, or equivalently, the differentials d = 0 for i ≥ r. i 4. Exact Couples Definition 4.1. An exact couple is a triangle α A → A (5) γ (cid:45) (cid:46) β E where A and E are R-modules and α,β,γ are R-module homomorphisms which are exact in the sense that ker α = im γ, ker γ = im β and ker β = im α. Let d : E → E be the composite map d = βγ. Since γβ = 0, we have that d2 = 0, so E is a differential module, with homology H(E) = ker d/im d. 2 Proposition 4.2. Let (5) be an exact couple. Then there is a derived exact couple α(cid:48) αA → αA (6) γ(cid:48) (cid:45) (cid:46) β(cid:48) H(E) where (1) α(cid:48) is α restricted to αA, (2) β(cid:48) is βα−1 : αA → H(E), taking αa to the homology class of βa for a ∈ A, (3) γ(cid:48) is the map induced by γ on ker d (which automatically kills im d). β(cid:48) is well defined in the Proposition 4.2, since ker α = im γ is taken to im d by β. We construct the spectral sequence of the exact couple (5) by iterating (6), so that 1E = E with differential d = d = βγ 1 2E = H(E) with differential d = β(cid:48)γ(cid:48) from the derived couple 2 3E = H(H(E)) from the derived couple of the derived couple . . . Proposition 4.3. With the notation after the definition of a spectral sequence, we have r+1Z = γ−1(im αr) r+1B = β(ker αr). Thus ∞E = ∞Z/∞B = γ−1(∩im αr)/β(∪ker αr). Proof. We prove the proposition by induction on r, the case r = 0 being immediate. Assume that r ≥ 1, and that rZ = γ−1(im αr−1) and rB = β(ker αr−1). We have that r+1Z = {x ∈ rZ | βα1−rγ(x) ∈ rB}. Suppose that x ∈ γ−1(im αr). Then γ(x) = αr(y) for some y ∈ A, so that x ∈ γ−1(im αr−1) = rZ and βα1−rγ(x) = βα(y) = 0. Thus γ−1(im αr) ⊂ r+1Z. Now suppose that x ∈ r+1Z. Then βα1−rγ(x) ∈ rB implies βα1−rγ(x) = β(y) for some y ∈ ker αr−1, so α1−rγ(x)−y ∈ ker β = im α and α1−rγ(x) = y+α(z) for some z ∈ A. Thus γ(x) = αr−1(y)+αr(z) = αr(z) ∈ im αr, so that x ∈ γ−1(im αr). We have established that r+1Z = γ−1(im αr). We have that r+1B = {βα1−rγ(x) | x ∈ γ−1(im αr−1)}. Suppose that x ∈ β(ker αr). Then x = β(y) for some y ∈ ker αr. Thus αr−1(y) = γ(z) for some z ∈ E, so that z ∈ γ−1(im αr−1) = rZ. Now y = α1−rγ(z) implies x = βα1−rγ(z) ∈ r+1B. Thus β(ker αr) ⊂ r+1B. Suppose that z ∈ r+1B. Then z = βα1−rγ(x) for some x ∈ γ−1(im αr−1). There exists y ∈ A such that γ(x) = αr−1(y). We have 0 = αγ(x) = αr(y) implies y ∈ ker αr. We have z = βα1−rγ(x) = β(y) ∈ β(ker αr). Thus r+1B ⊂ β(ker αr). (cid:3) We have that d : rE = rZ/rB → rZ/rB = rE r is defined by d (x+rB) = βα1−rγ(x)+rB r for x ∈ rZ. 3 Suppose that (F,d) is a differential module, and let α : F → F be a monomorphism. Set F = F/αF. The differential d of F induces a differential on F, giving a short exact sequence of differential modules α β 0 → F → F → F → 0. By the construction of (4), we have an induced exact couple α H(F) → H(F) (7) γ (cid:45) (cid:46) β H(F) The boundary map γ is defined by (8) γ(x) = [α−1d(z)] where z is a lift to F of a representative of x in F. β is the map on homology induced by the projection of F onto F. The induced spectral sequence of this exact couple is called the spectral sequence of α on F. We have 1E = H(F) = [{z ∈ F | dz ∈ αF}/αF]/[{dz | z ∈ F}+αF/αF] = {z ∈ F | dz ∈ αF}/αF +dF, and H(F) = {z ∈ F | dz = 0}/dF. α : H(F) → H(F) is defined by α(x+dF) = α(x)+dF for x ∈ {z ∈ F | dz = 0}. β : H(F) → H(F) is defined by β(x+dF) = x+(αF +dF) for x ∈ {z ∈ F | dz = 0}. γ : H(F) → H(F) is defined by γ(x) = α−1dx+dF for x ∈ {z ∈ F | dz ∈ αF}. d = βγ : 1E → 1E 1 is defined by d (x+(αF +dF)) = α−1dx+(αF +dF) 1 for x ∈ {z ∈ F | dz ∈ αF}. By Proposition 4.3, we have r+1Z = γ−1(im αr) (9) = {z+(αF +dF) | z ∈ F and dz ∈ αr+1F} = {z ∈ F | dz ∈ αr+1F}+αF/αF +dF and r+1B = β(ker αr) = {z+(αF +dF) | z ∈ F and αrz ∈ dF} (10) = {z+(αF +dF) | z ∈ α−rdF} = α−rdF +αF/αF +dF. 4 Thus r+1E = r+1Z/r+1B = {z ∈ F | dz ∈ αr+1F}+αF/α−rdF +αF with differential d : r+1E → r+1E defined by r+1 (11) d (x+(α−rdF +αF)) = (α−r−1dx+(α−rdF +αF)) r+1 for x ∈ {z ∈ F | dz ∈ αr+1F}. 5. Filtered Differential Modules A filtered differential module is a differential module (G,d) together with a sequence of submodules Gp for p ∈ Z satisfying G ⊃ ··· ⊃ Gp ⊃ Gp+1 ⊃ ··· that are preserved by d; that is, dGp ⊂ Gp for all p. Generally, we have that Gp = G for p ≤ 0. Suppose that G is a filtered differential module. We will associate a spectral sequence to G. Let F = (cid:76) Gp. The sum of the inclusion maps Gp+1 → Gp defines a map p∈Z α : F → F that is a monomorphism. Its cokernel is (cid:77) coker α = gr G = Gp/Gp+1. p∈Z From the exact sequence of differential modules α β 0 → F → F → gr G → 0 we obtain the exact couple (7), with associated spectral sequence. Thus the spectral sequence of α on F starts with (cid:77) (cid:77) 1E = H(gr(G)) = H(Gp/Gp+1) = 1Ep. p∈Z p∈Z We have that (12) 1Ep = H(Gp/Gp+1) = {z ∈ Gp | dz ∈ Gp+1}/Gp+1+dGp. Further, H(F) ∼= (cid:76) H(Gp) and p∈Z H(Gp) = {z ∈ Gp | dz = 0}/dGp. The maps α,β,γ are graded; we have γ : H(Gp/Gp+1) → H(Gp+1) is defined by γ(x+(Gp+1+dGp)) = dx+dGp for a representative x ∈ {z ∈ Gp | dz ∈ Gp+1}, β : H(Gp) → H(Gp/Gp+1) is defined by β(x+dGp) = x+(Gp+1+dGp) for a representative x ∈ {z ∈ Gp | dz = 0}. α : H(Gp+1) → H(Gp) is defined by α(x+dGp+1) = x+dGp 5 for a representative x ∈ {z ∈ Gp+1 | dz = 0}. Since α−r(Gp) = Gp−r and αr+1(Gp) = Gp+r+1, and by (9) and (10), we have r+1Zp = {z+(Gp+1+dGp) ∈ 1Ep | z ∈ Gp and dz ∈ Gp+r+1} (13) = {z ∈ Gp | dz ∈ Gp+r+1}+Gp+1/Gp+1+dGp} and r+1Bp = {z+(Gp+1+dGp) ∈ 1Ep | z ∈ Gp and z = dy for some y ∈ Gp−r} = (Gp∩dGp−r)+Gp+1/Gp+1+dGp. We thus have that (14) rZp/rBp = {z ∈ Gp | dz ∈ Gp+r}/Gp∩dGp−r+1+Gp+1. We have by (11) that d : rEp → rEp+r is given by r d (z+(Gp∩dGp−r+1+Gp+1)) = dz+(Gp+r ∩dGp+1+dGp+r+1) r for z ∈ Gp such that dz ∈ Gp+r. ThemoduleH(G)isfilteredbythesubmodules(H(G))p = im H(Gp) → H(G). Writing the associated graded module as gr H(G) = (cid:76) (H(G))p/(H(G))p+1, and writing Kp = p∈Z {z ∈ Gp | dz = 0}, we have exact sequences 0 → dG∩Gp/dGp → H(Gp) = Kp/dGp → (H(G))p → 0 so that (HG)p/(HG)p+1 = Kp/(Kp+1+(dG∩Gp)) = Kp+Gp+1/(Gp∩dG)+Gp+1. since Kp∩((Gp∩dG)+Gp+1) = Kp+1+(dG∩Gp). We have ∞Zp/∞Bp = ∩ ({z ∈ Gp | dz ∈ Gp+r}+Gp+1)/∪ ((Gp∩dGp−r)+Gp+1). r r Writing the quotient on the right as Mp/Np, we have Mp ⊃ Kp + Gp+1 and Np ⊂ (Gp∩dG)+Gp+1. Taking the direct sum over all p, we see that gr H(G) is a quotient of a submodule of ∞Z/∞B. Definition 5.1. The spectral sequence of the filtered differential module G converges and for any term rE of the spectral sequence we write rE ⇒ gr H(G) if gr H(G) = ∞Z/∞B; that is, for each p we have ∩ ({z ∈ Gp | dz ∈ Gp+r}+Gp+1) = {z ∈ Gp | dz = 0}+Gp+1, r and ∪ ((Gp∩dGp−r)+Gp+1) = (Gp∩dG)+Gp+1. r The second of these conditions is relatively trivial; it will be satisfied as soon as G = ∪ Gp. p 6 6. Double Complexes Definition 6.1. A double complex C is a commutative diagram . . . . . . ↑ ↑ ··· → Ci,j+1 d→hor Ci+1,j+1 → ··· d ↑ ↑ d vert vert ··· → Ci,j d→hor Ci+1,j → ··· ↑ ↑ . . . . . . (extendinginfinitelyinallfourdimensions)whereeachrowandeachcolumnisanordinary complex. Let C be a double complex. The total complex of C is the complex tot(C) defined by tot(C)k = (cid:76) Ci,j, with differential d : tot(C)k → tot(C)k+1 defined by d(x) = i+j=k (−1)jd (x)+d (x) for x ∈ Ci,j. hor vert LetC beadoublecomplex. Therearetwoimportantfiltrationsontot(C). Thefirstone is { Gp}, where Gp = (cid:76) Ci,j. The second is { Gp}, where Gp = (cid:76) Ci,j. hor hor j≥p vert vert i≥p Gp is a subcomplex of tot(C), with ( Gk)p = (cid:76) Ci,j. Also, Gp is a hor hor i+j=k,j≥p vert subcomplex of tot(C), with ( Gp)k = (cid:76) Ci,j. vert i+j=k,i≥p Let{Gp}beoneofthesetwofiltrationsoftot(C). LetF = (cid:76) Gp, andletα : F → F p∈Z be the map determined by the inclusions of Gp+1 into Gp. This data determines an exact couple (7) α H(F) → H(F) (15) γ (cid:45) (cid:46) β H(F/α(F)) . Since the Gp are subcomplexes of the complex tot(C), (15) determines long exact co- homology sequences (16) ··· → Hi(Gp+1) →α Hi(Gp) →β Hi(Gp/Gp+1) →γ Hi+1(Gp+1) → ··· WecanconsiderH(F)andH(F/α(F))asbigradedmodulesbysettingH(F)p,q = Hp+q(Gp) and H(F/α(F))p,q = Hp+q(Gp/Gp+1). The maps α, β, γ in (15) are then of respective bidegrees (−1,1), (0,0) and (1,0). WethenhavethattherE arebigraded, andd = βα−(r−1)γ isbihomogeneousofdegree r r in the p grading and of degree −(r−1) in the q grading; that is, d is the direct sum of r maps dp,q : rEp,q → rEp+r,q−r+1. r The differential d goes “r steps to the right and r−1 steps down”. r Let rDp−r+1,q+r−1 = αr−1Hp+q(Gp) = {im Hp+q(Gp) → Hp+q(Gp−r+1)}. Taking the successive derived couples of (15), we obtain long exact sequences (17) ··· → rDp−r+2,q+r−2 →α rDp−r+1,q+r−1 βα−→(r−1) rEp,q →γ rDp+1,q → ··· 7 Let us consider the case when we are considering the horizontal filtration, Gp = Gp. hor Then (counting the total degree p+q of an element of Cq,p in the first homology group, and the relative degree q of an element of Cq,p in the second homology group) we have 1Ep,q = Hp+q(Gp/Gp+1) = {ker d : Cq,p → Cq+1,p}/{im d : Cq−1,p → Cq,p} hor hor = Hq (C∗,p) = Hq(C∗,p) d hor is the homology of d . hor From our calculation (12), we have 1Ep = {z ∈ Gp | dz ∈ Gp+1}/Gp+1+dGp →d1 1Ep+1 = {z ∈ Gp+1 | dz ∈ Gp+2}/Gp+2+dGp+1 is defined by d ([x]) = [d(x)] for x ∈ {z ∈ Gp | dz ∈ Gp+1}. Suppose that x ∈ Cq,p, 1 with dx ∈ Gp+1, represents an element of 1Ep,q. Then d (x) = 0, so d ([x]) ∈ 1Ep+1,q is hor 1 represented by the element d (x) ∈ Cq,p+1. We thus have that vert 2Ep,q = {ker d : 1Ep,q → 1Ep+1,q}/{im d : 1Ep−1,q → 1Ep,q} vert vert = Hp Hq (C∗,∗). vert hor From (14), we have the map d : 2Ep → 2Ep+2, 2 {z ∈ Gp | dz ∈ Gp+2}/Gp∩dGp−1+Gp+1} →d2 {z ∈ Gp+2 | dz ∈ Gp+4}/Gp+2∩dGp+1+Gp+3, is defined by d ([x]) = [d(x)] for x ∈ {z ∈ Gp | dz ∈ Gp+2}. Suppose that x ∈ Cq,p 2 represents an element of 2Ep,q. Then there exists z(cid:48) ∈ Gp+1 such that d(x+z(cid:48)) ∈ Gp+2. We can then achieve this condition with a choice of z(cid:48) ∈ Gq−1,p+1. We necessarily then have that d (x) = 0 and (−1)p+1d (z(cid:48)) = −d (x). so d (x) ∈ 1Ep+1,q is represented hor hor vert 2 by the element d (z(cid:48)) ∈ Cq−1,p+2. vert The map d takes the homology class of z to the homology class of d (z(cid:48)). For this 2 vert to be zero means that d (z(cid:48)) = (−1)p+2d (z(cid:48)(cid:48)) for some z(cid:48)(cid:48), and in this case d carries vert hor 3 the homology class of z to d (z(cid:48)(cid:48)). vert Theorem 6.2. Associated with the double complex C are two spectral sequences r E and hor r E, corresponding, respectively, to the horizontal and vertical filtrations of tot(C). The vert 1E terms are bigraded with the components given by 1 Ep,q = Hq(C∗,p), 1 Ep,q = Hq(Cp,∗). hor vert If Ci,j = 0 for all i < 0 or for all j > 0, then the horizontal spectral sequence converges, so that ∞ E = gr H(tot(C)). hor hor Symmetrically, if Ci,j = 0 for all i > 0 or for all j < 0, then the vertical spectral sequence converges. Proof. We have 1 Ep,q = H( Gp/ Gp+1)p,q = Hp+q(gr (C)p) = Hq(C∗,p). hor hor hor tot The case for the vertical spectral sequence is similar. Theproofofconvergenceusesthebigrading. Wewillprovecovergenceforthehorizontal spectral sequence. The proof for the vertical spectral sequence is similar. Let {Gp} be the horizontal filtration of tot(C). We must show that ∩ ({z ∈ Gp | dz ∈ Gp+r}+Gp+1) = {z ∈ Gp | dz = 0}+Gp+1, r and ∪ ((Gp∩dGp−r)+Gp+1) = (Gp∩dG)+Gp+1. r 8 Since tot(C) = ∪ Gp, the second condition is trivially satisfied. The first condition p means that if z ∈ Gp and for each r there is an element y ∈ Gp+1 such that d(z−y ) ≡ r r 0 mod Gp+r, then there is an element y ∈ Gp+1 such that d(z − y) = 0. Breaking z up into a sum of elements of the same total degree, we may assume that z ∈ (Gq)p = (cid:76) Ci,j for some q. We can then assume (by possibly replacing y with its part i+j=q,j≥p r which is homogeneous of total degree q) that (cid:77) d(z−y ) ∈ (Gq+1)p+r = Cij. r i+j=q+1,j≥p+r If Ci,j = 0 for j > 0, then (Gq+1)p+r = 0 for r > −p. If on the other hand, Ci,j = 0 for i < 0, then (Gq+1)p+r = 0 if r > q+1−p. Thus in either case d(z−y ) = 0 for suitable r r, and we may take y = y for this value of r. r (cid:3) 7. Balanced Tor SupposethatM andN areR-modules. WewillshowthatTorR(M,N)canbecomputed i from a free resolution of either M or N. Let P : ··· → P →ϕi P → ··· → P i i−1 0 and Q : ··· → Q →ψi Q → ··· → Q i i−1 0 be free resolutions of M and N respectively. We will show that ∼ ∼ H(P ⊗ N) = H(tot(P ⊗ Q)) = H(M ⊗ Q) R R R as R-modules. Since TorR(M,N) computed from a free resolution of M is the first of i these, and TorR(M,N) computed from a free resolution of N is the last, this will suffice. i Let E be the vertical spectral sequence associated with the third quadrant double vert complex F = P ⊗ Q, which may be written with upper indices, using the convention R that Pi = P and Qj = Q . −i −j . . . . . . ↑ ↑ ··· → Pi⊗Qj+1 ϕ→⊗1 Pi+1⊗Qj+1 → ··· P ⊗Q : 1⊗ψ ↑ ↑ 1⊗ψ ··· → Pi⊗Qj ϕ→⊗1 Pi+1⊗Qj → ··· ↑ ↑ . . . . . . We have 1 E = 1 E−i,−j = H−j(P−i⊗ Q∗) = H (P ⊗ Q ). vert i,j vert R j i R ∗ ··· → P ⊗Q 1⊗→ψj P ⊗Q → ··· → P ⊗Q → P ⊗N → 0 i j i j−1 i 0 i is exact since P is free. Thus H (P ⊗ Q ) = 0 for j > 0. Tensoring the short exact i j i R ∗ sequence ∼ 0 → ϕ (Q ) → Q → H (Q ) = N → 0 1 1 0 0 ∗ 9 with P , we obtain the short exact sequence i ∼ 0 → P ⊗ϕ (Q ) → P ⊗Q → P ⊗H (Q ) = P ⊗N → 0, i 1 1 i 0 i 0 ∗ i from which we compute ∼ H (P ⊗ Q ) = P ⊗ N. 0 i R ∗ i R Thus the only nonzero 1 E terms are those with j = 0. The differential d is induced vert i,j 1 by d = ϕ⊗1. Thus 1 E is the complex P ⊗ N. Thus 1 E is the complex P ⊗ N hor vert R vert R and (cid:26) H (P ⊗ N) for j = 0 2 E = 2 E−i,−j = i R vert i,j vert 0 for j (cid:54)= 0. We have d−i,−j : 2 E−i,−j → 2 E−i+2,−j−1 2 vert vert so d = 0. We further calculate 2 3 E−i,−j = (ker 2 E−i,−j →d2 2 E−i+2,−j−1)/(im 2 E−i−2,−j+1 →d2 2 E−i,−j vert vert vert vert vert so that 3 E−i,−j = 0 if j (cid:54)= 0. Since vert d−i,−j : 3 E−i,−j → 3 E−i+3,−j−2 3 vert vert we have that d = 0. Continuing in this way, we calculate that d = 0 for k ≥ 2. 3 k It follows that the spectral sequence degenerates at 2E; that is ∞ E = 2 E. Since all vert vert nonzero terms have j = 0, (cid:77) ∞ E = ∞E , vert i,j k,0 i+j=k and the filtration of H(tot(P ⊗ Q)) has only one nonzero piece. Thus we get R H(P ⊗ N) = gr (H(tot(P ⊗ Q))) = H(tot(P ⊗ Q)). R vert R R By symmetry (calculating the horizontal spectral sequence of P ⊗ Q) we get R H(tot(P ⊗ Q)) = H(M ⊗ Q). R R 8. Exact sequence of terms of low degree Proposition 8.1. If Ci,j is a third quadrant double complex (Ci,j = 0 if i > 0 or j > 0), then writing E for E, we have vert a. H0(tot(C)) ∼= H0 H0 (C∗,∗). hor vert b. There is a 5-term exact sequence H−2(tot(C)) → H−2H0 (C∗,∗) →d2 H0 H−1 (C∗,∗) → H−1(tot(C)) → H−1H0 (C∗,∗) → 0 hor vert hor vert hor vert Proof. We first prove a. We have an exact sequence (17) 2D0,0 → 2D−1,1 → 2E0,0 → 2D1,0. Since Gp = 0 for p > 0, we compute that 2D0,0 = 2D1,0 = 0. 2D−1,1 = im H0(G0) → H0(G−1) = C0,0/d(C0,−1)+d(C−1,0) = H0(tot(C)). Since 2E0,0 = H0 H0 (C∗,∗), a follows. hor vert We now prove b. From (17), we have exact sequences 2D−3,1 → 2E−2,0 →γ 2D−1,0 → 2D−2,1 → 2E−1,0 → 2D1,0 2D0,−1 → 2D−1,0 β→α−1 2E0,−1 → 2D1,−1 10