p,q p,q Geometria Superiore Reference Cards for k > 3 we can choose Bk+1(E2 ) and Zk+1(E2 ) as the p,q p,q subobjects of Zk(E2 ) containing Bk(E2 ) and identified by Spectral Sequences means of the isomorphism of the previous step k to the sub- c 2000 M. Cailotto, Permissions on last. v0.0 objects im(dp−k,q+k−1) and ker(dp,q) of Ep,q respectively; in Send comments and corrections to [email protected] k k k this way we have Z (Ep,q)B (Ep,q) ' Ep,q . Generalities. k+1 2 k+1 2 k+1
Definitions Let A be an abelian category. Define Af the So we have a long chain of inclusions: quasi-abelian category of filtered objects of A: objects are ob- p,q p,q p,q p,q E = Z2(E ) ⊇ ··· ⊇ Zk(E ) ⊇ Zk+1(E ) ⊇ ··· ⊇ jects A of A with a (decreasing) filtration, i.e. a family of 2 2 2 2 subobjects (F p(A)) of A s.t. F p+1(A) ⊆ F p(A) for every ··· ⊇ B (Ep,q) ⊇ B (Ep,q) ⊇ ··· ⊇ B (Ep,q) = 0 p∈Z k+1 2 k 2 2 2 p ∈ Z; morphisms ϕ : A−→B are morphisms of A respecting the filtrations, i.e. ϕ(F p(A)) ⊆ F p(B). and the definition continues with: p,q p,q p (d) for every p,q ∈ Z two subobjects B∞(E2 ) and Z∞(E2 ) of The filtration (F (A)) of A is: p,q p,q p,q p∈Z E2 such that Z∞(E2 ) ⊇ B∞(E2 ) and for every k T p (i) separated if p∈ F (A) = 0, p,q p,q p,q p,q Z Zk(E2 ) ⊇ Z∞(E2 ) ⊇ B∞(E2 ) ⊇ Bk(E2 ) (ii) coseparated (or exhaustive) if S F p(A) = A, p∈ p,q p,q p,q Z and define: E = Z∞(E ) B∞(E ); p q ∞ 2 2 (iii) discrete if there exists p ∈ Z s.t. F (A) = 0 (so F (A) = 0 n (e) a family (E )n∈ of objects of A ; for every q > p), Z f ∼ p q (f) a family (βp,q) of isomorphisms βp,q : Ep,q→gr (Ep+q). (iv) codiscrete if there exists p ∈ Z s.t. F (A) = A (so F (A) = p,q∈Z ∞ p A for every q 6 p), Remarks on the definition of spectral sequences: (v) finite if it is discrete and codiscrete. (1) the construction of chains of subobjects in (c) can be per- Remark that discrete (resp. codiscrete) filtrations are sepa- formed for every r in the spectral sequence, defining for rated (resp. exhaustive). p,q p,q p,q every k > r the subobjects Bk(Er ) and Zk(Er ) of Er . p p+1 p,q For A ∈ Af : grp(A) := F (A) F (A) ∈ A, and if A admits (2) for a fixed n ∈ Z, the family {E∞ }p+q=n gives a measure L of the skips in the filtration of En: in fact, if the terms arbitrary direct sums, define: gr(A) := grp(A) ∈ A, the p∈Z L p,q∼ n exist, we have E∞ =gr(E ). associated graded object. This gives a functor gr : Af → A. p+q=n p,q n 0 A morphism of two spectral sequences E = (Er ,E ) and E = Let ϕ : A−→B be a morphism in Af ; then: 0p,q 0n p,q n (E r ,E ) is the datum of f = (fr ,f ) where: (1) if grϕ is a monomorphism, and the filtration of A is ex- (a) fp,q : Ep,q −→ E0p,q are morphisms of A, haustive and separated, then ϕ is a monomorphism; r r r (b) compatibility with the differentials: commutation of (2) if grϕ is an epimorphism, and the filtration of B is exhaus- p,q tive and discrete, then ϕ is an epimorphism; p,q fr 0p,q Er −−−−−−−−−−−→ E r (20) if grϕ is an epimorphism, the filtration of B is exhaustive p,q 0p,q dr y y d r and separated, the filtration of A induces a complete topol- p+r,q−r+1 0p+r,q−r+1 Er −−−−−−−→ E r ogy, then ϕ is an epimorphism; p+r,q−r+1 fr (3) if grϕ is an isomorphism, the filtration of A is exhaustive (c) compatibility with the isomorphisms: and separated, and the filtration of B is exhaustive and p,q p,q Z (E ) f¯p,q Z (E0 ) discrete, then ϕ is an isomorphism. r+1 r −−−−−→r r+1 r p,q 0p,q Br+1(Er ) Br+1(E r ) SpS(A): the category of spectral sequences of A. p,q 0p,q αr y y α r p,q n p,q 0p,q A spectral sequence E = (E ) r 2 ,(E ) in A is the E −−−−−−−−−→ E r > n∈Z r+1 p,q r+1 p,q∈Z f data of: r+1 ¯p,q p,q commutes, where the fr define the morphisms in cohomology; (a) a family (Er ) of objects of A defined for r > 2 and p,q ∈ Z, p,q p,q p,q p+r,q−r+1 (d) compatibility with the ∞ level (b) a family of morphisms (dr ) where dr : Er → Er fp,q B (Ep,q) ⊆ B (E0p,q), fp,q Z (Ep,q) ⊆ Z (E0p,q) p,q p−r,q+r−1 2 ∞ 2 ∞ 2 2 ∞ 2 ∞ 2 and dr ◦dr = 0 for all indices; in such a way that we induce morphisms: f¯p,q : Ep,q → E0p,q; p,q p,q ∞ ∞ ∞ so that we can define: Zr+1(Er ) = ker(dr ) and (e) fn : En → E0n morphisms of filtered objects; p,q p−r,q+r−1 Br+1(Er ) = im(dr ) for which p,q p,q p,q (f) finally the compatibility with the limit Br+1(Er ) ⊆ Zr+1(Er ) ⊆ Er and we require: f¯p,q p,q Ep,q −−−−−−−−−−−→∞ E0p,q (c) a family (αr ) of isomorphisms ∞ ∞ p,q 0p,q ∼ β β αp,q : Z (Ep,q)B (Ep,q) −→ Ep,q ; y y r r+1 r r+1 r r+1 p+q 0p+q grp(E ) −−−−−−→ grp(E ) . gr (f p+q ) we can define for every p,q ∈ Z two families of subobjects of p Ep,q, B (Ep,q) and Z (Ep,q), in such a way that Ep,q ' 2 k 2 k 2 k p,q p,q Remarks on the definition of morphisms of spectral sequences: Zk(E2 ) Bk(E2 ) by induction: p,q p,q p,q (1) if a morphisms f of spectral sequences has the property for k = 2 put B2(E ) = 0 and Z2(E ) = E , p,q 2 2 2 that for a fixed r all fr are isomorphisms, then by (c) for p,q for k = 3 already defined, all s > r the fs are also isomorphisms.
1 GeoSupRC - Spectral Sequences c 2000 by MC n−q(n),q(n) n−q(n),q(n) n−q(n),q(n) (2) a morphism of spectral sequences is an isomorphism if and Er for all s r, so that E∞ = Er p,q > only if: the f2 are isomorphisms for all p,q ∈ Z, in the if the spectral sequence is weakly convergent: the limit of the condition (d) equalities hold, and the fn are isomorphisms sequence can be read at level r. for all n ∈ Z. (2) if moreover the filtrations on the limit are exhaustive and sepa- p n Convergence and regularity conditions: rated, then they are finite and F (E ) = 0 for p > n−q(n) and p n n F (E ) = E for p 6 n−q(n); we have also an isomorphism (o) a spectral sequence is weakly convergent if n n−q(n),q(n) E ' Er . B (Ep,q) = S B (Ep,q) Z (Ep,q) = T Z (Ep,q) ∞ 2 k 2 ∞ 2 k 2 0 (0 ) (Deligne) a spectral sequence E is degenerated in Er0 , for r0 k>2 k>2 p,q > 2, if it is biregular and for any r > r0 we have dr = 0 for all p hold, i.e. if the data of (d) are determined by the previous p p+q p,q and q; then Er0 = E∞ and gr E = Er0 . points. A weakly convergent spectral sequence is: (10) in the category of R-modules of finite length with R a ring, (i) regular if P p,q n (o) p+q=n lgEr > lgE ; (1) for any p,q the decreasing sequence Z (Ep,q) becomes k 2 k (i) equality holds for r = r if and only if E degenerates at r ; stationary, 0 0 n n (ii) if E = 0 except for a finite number of n, then for all r > 2 (2) for any n the filtration of E is discrete and exhaustive. P p+q p,q P n n p,q we have p,q(−) lgEr = n(−) lgE . This implies that for a fixed pair (p,q) we have Z∞(E2 ) = Z (Ep,q) for k 0, but depending on p and q. k 2 Homological and Spherical Sp.Seq. (ii) co-regular if: (1) for any p,q the increasing sequence B (Ep,q) becomes Cohomological spectral sequences: E is cohomological if k 2 k p,q E2 = 0 whenever p < 0 or q < 0. Then under the same condi- stationary, p,q p,q n tions Er = 0 for all r, and E∞ = 0. Properties of the coho- (2) for any n the filtration of E is codiscrete mological spectral sequences: p,q This implies that for a fixed pair (p,q) we have B∞(E2 ) = p,q (1) if the spectral sequence is weakly convergent and the filtra- Bk(E2 ) for k 0, but depending on p and q. tions of the limit are exhaustive and separated, then from (iii) biregular if it is regular and co-regular, i.e. if: the isomorphisms βp,q for p < 0 or q < 0 we deduce that p,q p,q nEn if p 0 (1) for any p,q the sequences Z (E ) and B (E ) F p(En) = 6 k 2 k k 2 k 0 if p > n . became stationary, (2) for any n the filtration of En is finite. In particular we have an exact sequence p,q p,q 1,0 1 0,1 This implies that for a fixed pair (p,q) we have E = E for 0−→E∞ −→E −→E∞ −→0 ∞ k k 0, but depending on p and q. coming from 0 → F 1(E1) → E1 → E1/F 1(E1) → 0. Remarks on the regular spectral sequences: (2) In the previous situation we have the edge morphisms: p,q p,q n,0 n n 0,n (1) if E is regular, then E∞ = lim Er where the induc- Er −→E E −→Er −→r0 tive system is given by: given resp. by the compositions Ep,q ' Zp,qB (Ep,q)−→Zp,qB (Ep,q) ' Ep,q . n,0 n,0 n n n n r ∞ r 2 ∞ r+1 2 r+1 Er E∞ ' grn(E ) = F (E ) ,→ E p,q p,q n,0 n,0 The result follows from B∞ = lim Br(E2 ). where the first map comes from E∞ = lim Er ; and −→r0 −→r (2) dually, if E is co-regular, then Ep,q = lim Ep,q where n n 1 n n 0,n 0,n ∞ r E E /F (E ) = gr0(E ) ' E∞ ,→ Er ←−r0 the projective system is given by: 0,n 0,n T 0,n where the last map comes from E∞ = lim Er = Er . p,q p,q p,q p,q p,q p,q ←−r Er ' Zr(E2 ) B∞ −→Zr+1(E2 ) B∞ ' Er+1. r p,q p,q (3) There exists an exact sequence The result follows from Z∞ = lim Zr(E2 ). ←−r0 d0,1 0−→E1,0 −→E1 −→E0,1 −→2 E2,0 −→E2. (3) Let f be a morphism between spectral sequences E and 2 2 2 p,q 0 p,q E ; we have that f2 isomorphisms for all (p,q) implies f Homological spectral sequences: E is homological if E2 = is an isomorphism in the following two cases: 0 whenever p > 0 or q > 0. Then under the same conditions 0 p,q p,q (i) E and E are biregular; Er = 0 for all r, and E∞ = 0. Properties of the cohomologi- (ii) E and E0 are regular and A is an abelian category where cal spectral sequences: the inductive filtered limits exist and are exact. (1) if the spectral sequence is weakly convergent and the fil- Degeneration of spectral sequences: trations of the limit are exhaustive and separated, nEn if p n (0) (original sense: Godement, Grothendieck) a spectral sequence F p(En) = 6 0 if p > 0 . E degenerates if there exists an r > 2 s.t. for all n ∈ Z there n−q,q n−q,q exists q(n) ∈ Z with Er = 0 for all q 6= q(n) (so Es = 0 In particular we have an exact sequence for all s r and q 6= q(n)). Remarks on degenerate spectral > 0−→E0,−1 −→E−1 −→E−1,0 −→0. sequences: ∞ ∞ (1) if we have q(n+1) > q(n)−r +1 for all n (for example: q(n) = k (2) In the previous situation we have the edge morphisms: n−q(n),q(n) 0,−n −n −n −n,0 or q(n) = n+k with constant k) we can easily see that Es = Er −→E and E −→Er .
2 GeoSupRC - Spectral Sequences c 2000 by MC i,0 i i i−n,n (3) There exists an exact sequence Er −→E and E −→Er . d−2,0 −2 −2,0 2 0,−1 −1 −1,0 (4) There exists a long exact sequence E −→E2 −→ E2 −→E −→E2 → 0. i,0 i i−n,n i+1,0 i+1 ···−→Er −→E −→Er −→Er −→E −→··· . Spectral sequences with spherical base: E is said to have p,q spherical bases if there exists r and n > r for which Er = 0 if Spectral sequence of a filtered complex. p 6= 0,n. Properties of spherical base spectral sequences: (1) if the spectral sequence is weakly convergent and the filtra- Let C(A)f be the category of filtered complexes, i.e. complexes tions of the limit are exhaustive and separated, then the C· of objects of A endowed with a (decreasing) filtration F pC· filtrations are finite and compatible with the differentials (dF pCm ⊆ F pCm+1). Us- (Ei if p 0 ing F p(C·) = Cn ∩F p(C·) to define a filtration on every 6 n∈ F p(Ei) = n,i−n Z E∞ if 0 < p 6 n single object, we have an equivalence of categories with C(Af ). 0 if p > n. The filtration is regular (resp. strongly regular) if for every (2) in the previous situation for every i there is the exact se- n there exists an integer M(n) s.t. for p > M(n) we have quence: Hn(F pC) = 0 (resp. F pCn = 0). n,i−n i 0,i A homotopy between morphisms of filtered complexes is of or- 0−→E∞ −→E −→E∞ −→0 der less than or equal to k if it sends F p into F p−k. coming from the sequence 0 → F n(Ei) → Ei → Ei/F n(Ei) → i n i i 1 i i 0,i 0 considering E /F (E ) = E /F (E ) = gr0E ' E∞ . If (1) We have a functor from C(A)f to SpS(A), defined also for the in the definition we put n < r, then the spectral sequence is levels r = 0,1, for which: constant after the r level, and, under the usual hypothesis (r=0) Zp,q = F p(Cp+q), Zp+1,q−1 = F p+1(Cp+q) so that Ep,q = of weak convergence and filtrations of the limit exhaustive 0 −1 0 p p+q p+1 p+q p,q and separated, we can read the exact sequence with ∞ = r. F (C ) F (C ) with the differentials d0 induced by (3) we find also the morphisms those of the complex C by quotient. p,q n,i−n 0,i (r=1) E = H F p(Cp+q)F p+1(Cp+q) = Hp+q F p(C)/F p+1(C) E −→Ei and Ei −→E 1 r r p,q with the differentials d1 obtained by the short exact sequence given resp. by the compositions 0 → F p+1(C)/F p+2(C) → F p(C)/F p+2(C) → F p(C)/F p+1(C) → 0 n,i−n n,i−n i n i i Er E∞ ' grn(E ) = F (E ) ,→ E using the coboundary morphism: n,i−n n,i−n where the first morphism is given by E∞ = lim Er ; δp+q −→r Ep,q = Hp+q gr (C) −→ Hp+q+1 grp+1(C) = Ep+1,q . and 1 p 1 i i 1 i i 0,i 0,i E E /F (E ) = gr0(E ) ' E∞ ,→ Er (2) Construction of the functor: −∞ ∞ 0,i 0,i (a) Define F C = C and F C = 0. Consider the sequence of where the last morphism is given by E∞ = lim Er . ←−r indexes (p+r,p+1,p,p−r +1); from the exact sequence (4) There exists a long exact sequence 0 → F p+1CF p+rC → F pCF p+rC → F pCF p+1C → 0 n,i−n i 0,i n,i+1−n i+1 ···−→Er −→E −→Er −→Er −→E −→··· we define p,q p,q Zp,q(C) = im Hp+q(F pC/F p+rC)−→Hp+q(F pC/F p+1C) In fact for r < n we can see that Er = Er+1 or all p and r q, so that we can suppose r = n; moreover =∼ker Hp+q(F pC/F p+1C)−→Hp+q+1(F p+1C/F p+rC) ; E0,i ' ker(d0,i) ,→ E0,i ∞ n n then from the exact sequence n,i−n+1 E 0 → F pC F p+1C → F p−r+1C F p+1C → F p−r+1C F pC → 0 En,i−n+1 n ' En,i−n+1 . n 0,i ∞ im(dn ) we define p,q p+q−1 p−r+1 p p+q p p+1 Br (C) = im H (F C/F C)−→H (F C/F C) Spectral sequences with spherical fiber: E is said to have p,q =∼ker Hp+q(F pC/F p+1C)−→Hp+q(F p−r+1C/F p+1C) ; spherical fiber if there exists r and n > r −1 for which Er = 0 if q 6= 0,n. Properties of spherical fiber spectral sequences: p,q p,q we find Br (C) ⊆ Zr (C) and we can use (1) if the spectral sequence is weakly convergent and the filtra- p,q p,q p,q tions of the limit are exhaustive and separated, then the Er (C) = Zr (C) Br (C). filtrations are finite and In particular Zp,q(C) = Hp+q(F pC/F p+1C) and Bp,q(C) = 0 (Ei if p i−n 1 1 6 (remark that the general definition does not work for r = 0). F p(Ei) = i−n,n E∞ if i−n < p 6 i 0 if p > i. (d) similarly (r = ∞) we define: p,q p+q p p+q p p+1 (2) in the previous situation for every i there is the exact se- Z∞ (C) = im H (F C)−→H (F C/F C) p+q p p+1 p+q+1 p+1 quence: =∼ker H (F C/F C)−→H (F C) p,q i,0 i i−n,n B (C) = im Hp+q−1(C/F pC)−→Hp+q(F pC/F p+1C) 0−→E∞ −→E −→E∞ −→0 ∞ =∼ker Hp+q(F pC/F p+1C)−→Hp+q(C/F p+1C) coming from the sequence 0 → F i(Ei) → Ei → Ei/F i(Ei) → i i i i i−n,n p,q 0 considering E /F (E ) = gri−nE ' E∞ . and E∞ (C) as the quotient; we have in fact (3) we find also the morphisms 0 = B1 ⊆ ··· ⊆ Bs ⊆ ··· ⊆ B∞ ⊆ Z∞ ⊆ ··· ⊆ Zr ⊆ ··· ⊆ Z1 = E1 .
3 GeoSupRC - Spectral Sequences c 2000 by MC p,q p,q p+r,q−r+1 0 (b) the differentials dr : Er −→Er are defined by the compositions: ↓ 0 0 K(p+1,p+r) δr Zr/Br Zr/Zr+1 −→Br+1/Br ,→ Zr/Br ↓ ↓ ↓ ∼= 0 −→ K(p+r,p+r +1) −→ K(p,p+r +1) −→ K(p,p+r) −→ 0 ↓ ↓ ↓ so that we have kerd = Z /B and imd = B /B . r r+1 r r r+1 r 0 −→ K(p+1,p+r +1) −→ K(p,p+r +1) −→ K(p,p+1) −→ 0 p,q (c) the isomorphisms αr are induced by ↓ ↓ ↓ ∼ K(p+1,p+r) 0 0 H(Er) = Zr+1/Br Br+1/Br =Zr+1/Br+1 = Er+1 ; ↓ n 0 (e) we consider En(C) = H (C) the cohomology of C as a col- lection of objects with the filtrations defined by F pHn(C) = then constructing the commutative diagram n p n im(H (F (C))−→H (C)) with the morphisms induced in co- HnK(p+1,p+r) p 0 homology from the inclusions F (C) ,→ C; . y p,q ∼ p p+q n n+1 (f) the isomorphisms βn are induced by E∞ (C)=gr H (C) . H K(p,p+r) −−→ H K(p+r,p+r +1) % u (x) the key lemma for the points (b), (c), (f) is: given a diagram y y HnK(p,p+r +1) −−→ HnK(p,p+1) −−→ Hn+1K(p+1,p+r +1) with exact row: 0 P u % c & d p,q 0 y and finally applying the square lemma to im(u) = Zr , im(u ) = L −−→ M −−→ N p,q p+r,q−r+1 0 p+r,q−r+1 a b Zr+1 and im() = Br+1 , im( ) = Br . then b induces an isomorphism im(c)/im(a) → im(b). The square In order to evaluate the terms E∞ we start with the exact of the lemma says that if we have a commutative diagram diagram of complexes 0 L 0 0 0 . y a ↓ ↓ c0 0 −→ K(p+1,∞) −→ K(p+1,∞) −→ 0 P −−→ M0 ↓ ↓ ↓ % c b0 y y 0 −→ K(p,∞) −→ K(∞,−∞) −→ K(−∞,p) −→ 0 L −−→ M −−→ N a b ↓ ↓ ↓ with exact row and column, then we have the isomorphisms 0 −→ K(p,p+1) −→ K(−∞,p+1) −→ K(−∞,p) −→ 0 ↓ ↓ ↓ im(c)/im(a)∼im(bc) = im(b0c0)∼im(c0)/im(a0) = = 0 0 0 0 induced respectively from b and b . then construct the commutative diagram p q HnK(p+1,∞) (y) the key for the points (a), (d) is: let K(p,q) = F /F for p 6 q; 0 for (p1,q1) 6 (p2,q2), i.e. p1 6 p2 and q1 6 q2, we have the . y v commutative diagram u0 HnK(p,∞) −−→ HnK(−∞,∞) K(p ,q ) −−→ K(p ,q ) 2 2 1 2 % u & y y y y Hn−1K(−∞,p) −−→ HnK(p,p+1) −−→ HnK(−∞,p+1) K(p2,q1) −−→ K(p1,q1) v and calling the diagonal map u12, we have for (p2,q2) (p3,q3) p,q 6 and finally apply the square lemma to im(u) = Z∞ , im(v) = the compatibility u13 = u12 ◦u23. Moreover, for p1 6 p2 6 p3 Bp,q and im(u0) = F pEn, im(v0) = F p+1En, where n = p+q. we have the exact sequence ∞ (3) Starting with complexes with regular and exhaustive filtrations, 0−−→K(p2,p3)−−→K(p1,p3)−−→K(p1,p2)−−→0 if A admits exact filtered inductive limits, then we obtain reg- and the associated long exact sequence of cohomology. ular spectral sequences. (4) A family (Bp,q) of objects of A is convergent to a family of (z) The chain of inclusions in (d) can be seen using the sequences p,q∈Z n p,q p+q of index (p,p+1,p+r,p+r +1) and (p−r,p−r +1,p,p+1), and objects (D )n∈ , B =⇒ D , if there exists a complex X ∈ Z p,q the commutative diagrams C(A) with regular filtration s.t. E (X) ' Bp,q and Hn(X) ' n 2 ur+1 vr−1 D for every p,q,n ∈ . K(p,p+r +1) −−→ K(p,p+1) K(p,p+1) −−→ K(p−r +1,p+1) Z (5) Given two morphisms of filtered complexes f,g : C → C0 with y y = vr y y ur = a homotopy s of order less than or equal than k, we have for K(p,p+r) −−→ K(p,p+1) , K(p−r,p+1) −−→ K(p−r,p+1) . the morphisms induced on the spectral sequences f,g : E(C) → E0(C0) that fn = gn for any n ∈ and also fp,q = gp,q for every In fact we have Z = im(Hnu ) ⊇ im(Hnu ) = Z and Z r r r r r+1 r+1 r > k. This is the main motivation for the definition of spectral B = ker(Hnv ) ⊆ ker(Hnv ) = B ; moreover considering r r−1 r r+1 sequence with constraint r 2, instead of r ∈ . the index sequence (p−r +1,p,p+1,p+s), the zero composi- > N tion (of coboundary morphisms) ∂ ∂ Sp. sequences of a double complex. Hn−1K(p−r +1,p)−−→r HnK(p,p+1)−−→s Hn+1K(p+1,p+s) Let C2(A) be the category of bicomplexes with objects in A, gives Zs = ker(∂s) ⊇ im(∂r) = Bs for any r,s, in particular for i.e. C(C(A)) or more explicitly: objects are the collections r = s = ∞. (Xn,m) ,(dn,m) ,(dn,m ) n,m∈Z IX n,m∈Z IIX n,m∈Z The basic isomorphism for the definition of dr can be see start- ing with the exact diagram of complexes where Xn,m ∈ obA,
4 GeoSupRC - Spectral Sequences c 2000 by MC dn,m p IX n+1,m n,m and F (X) codiscrete (resp. discrete) if there exists n0 (resp. Xn,m −→ Xn+1,m d ◦d = 0 II n,m IX IX 00 0 00 n,m d , n,m+1 n,m n ) s.t. for any n > n (resp. n 6 n ) we have X = 0; or if Xn,m −→IIX Xn,m+1 d ◦d = 0 00 0 00 0 IIX IIX there exists m (resp. m ) s.t. for any m 6 m (resp. m > m ) we have Xn,m = 0. n+1,m n,m n,m+1 n,m ·· and dIIX ◦dIX = dIX ◦dIIX . Morphisms between X 2 n,m (5) We define two functors C (A)−→SpS(A): the two spectral ·· n,m f n,m and Y are collections X −→ Y for n,m ∈ Z s.t. sequences associated to a bicomplex X are the spectral se- dn,m ◦fn,m = fn+1,m ◦dn,m quences associated to the simple complex using the two filtra- IY IX tions: I E(X) and II E(X) respectively. They are the same n,m n,m n,m+1 n,m dIIY ◦f = f ◦dIIX . limit (the total cohomology) and we can calculate the first terms: Remark that C2(A) has an internal automorphism (exchange I Ep,q(X) I Ep,q+1(X) II Eq,p(X) II Eq,p+1(X) of the indexes). Standard notions on the category of bicom- 0 0 0 0 k k k k plexes: Xp,q −−−−−−−−−→ Xp,q+1 Xp,q −−−−−−−−−→ Xp+1,q I p,q p,q II q,p p,q (1) Cocycles, coboundary and cohomology operators: d0 =dII d0 =dI ·,m ·,m n,· n,· I Ep,q(X) I Ep+1,q(X) II Eq,p(X) II Eq+1,p(X) ZII := ker(dII ) ZI := ker(dI ) 1 1 1 1 ·,m ·,m−1 n,· n−1,· k k k k BII := im(dII ) BI := im(dI ) p,q p+1,q p,q p,q+1 ·,m ·,m ·,m n,· n,· n,· H (X) −−−−−−−→ H (X) H (X) −−−−−−−−→ H (X) H := Z B H := Z B II I p,q p,q II I II q,p p,q I II II II I I I d1 =dI d1 =dII I p,q p q II q,p q p these structures are complexes w.r.t. the pointed index (with E2 (X) = H H (X) E2 (X) = H H (X) . ·,m n,· I II II I differentials resp. d and d ) so that we define also: I II (6) Two homotopic morphisms of double complexes induce the n m n,m n+1,m m n n,m n,m+1 same morphism on the associated spectral sequence. ZI HII := ker(HII → HII ) ZII HI := ker(HI → HI ) n m n−1,m n,m m n n,m−1 n,m BI HII := im(HII → HII ) BII HI := im(HI → HI ) (7) Given a bicomplex X, if one of the following conditions hold: HnHm := ZnHm BnHm Hm Hn := Zm HnBm Hn . I II I II I II II I II I II I (i) there exists n0 and m0 s.t. for n < n0 or m < m0 (or: n > n or m > m ) we have Xn,m = 0; (2) a morphism f of bicomplexes X and Y is homotopic to zero if 0 0 there exist a pair of families of morphisms for n,m ∈ (ii) there exists n0 and n1 (or: m0 and m1) s.t. for n < n0 or Z n,m sn,m : Xn,m → Y n−1,m n > n1 (or: m < m0 or m > m1) we have X = 0; tn,m : Xn,m → Y n,m−1 then the two associated spectral sequences are biregular (in particular, for k > n −n (or: k > m −m ) the sequences are with the commutation properties 1 0 1 0 constant). sn,m+1 ◦dn,m = dn−1,m ◦sn,m IIX IIY (8) Given a bicomplex X in A category with exact filtrant inductive n+1,m n,m n,m−1 n,m t ◦dIX = dIY ◦t limits, then: n,m and reconstructing f: f is (i) if there exists n0 s.t. for n > n0 (or: there exists m0 s.t. n,m I sn+1,m ◦dn,m +dn−1,m ◦sn,m +tn,m+1 ◦dn,m +dn,m−1 ◦tn,m. for m < m0) we have X = 0 then the sequence E(X) IX IY IIX IIY is regular; Obviously s (resp. t) induces homotopies for the morphism (ii) if there exists n s.t. for n < n (or: there exists m s.t. ·,m ·,m 0 0 0 induced by f between the complexes H (X) and H (Y ) n,m II II II for m > m0) we have X = 0 then the sequence E(X) n,· n,· is regular. (resp. HI (X) and HI (Y )) and so f induces the null map n m n m between the cohomologies H H (X) and H H (Y ) (resp. (9) (Weyl’s lemma) Given a bicomplex C·· with non negative de- m n m n I II I II HII HI (X) and HII HI (Y )). Two maps of bicomplexes are grees (Cn,m = 0 if n < 0 or m < 0) with exact rows and columns, homotopic if the difference is homotopic to zero. except for the first, there exists a canonical isomorphism be- (3) Under one of the following hypothesis: tween the cohomology of the first row and that of the first (i) A admits inductive limits; column. (ii) for objects of C2(A) s.t. for any k the cardinality of {(n,m)| n+m = k, Xn,m 6= 0} is finite; Sp. seq. of a functor w.r.t. a finitely filtered object. we have a functor s : C2(A)−→C(A) acting on the objects by s(X)k = LXn,m and dk : s(X)k −→s(X)k+1 Let A be the abelian category of finitely filtered objects of s(X) ff n+m=k A, i.e. objects of A endowed with a finite filtration; morphisms in this category are the morphisms of A (without reference to where the differentials are induced by Xn,m −→s(X)k+1 de- the filtrations). If A has sufficiently many injective objects, fined for n+m = k as ιk+1 ◦dn,m +(−)nιk+1 ◦dn,m us- · n+1,m I n,m+1 II we have a functor of “filtered resolution”: Aff −→ C(I)f with k+1 i,j k+1 the properties that: ing ιi,j : X −→s(X) for i+j = k +1. By definition the (total) cohomology of a bicomplex is the cohomology of the (1) A· is an injective resolution of A, associated simple complex. (2) F i(A·) is an injective resolution of F i(A), (4) We have two different filtrations on s(X) given by (3) F i(A·) = 0 (resp. = A) if F i(A) = 0 (resp. = A), F p s(X)k = LXn,m and F p s(X)k = LXn,m ; I II (4) given F : A → B and two morphisms u·,v· : A·−→B· exten- n+m=k n+m=k · · n>p m>p sions of f then they are homotopic: u ∼ v , p (5) given f : A → B for which f F i(A) ⊆ F i+s(B) for any i ∈ they are exhaustive and separated; we use the symbols FI (X) p p · · · · i · i+s · and FII (X). Moreover: FI (X) is discrete (resp. codiscrete) Z then f : A → B verifies f F (A ) ⊆ F (B ).
5 GeoSupRC - Spectral Sequences c 2000 by MC Given an additive functor F : A−→B between abelian cate- (2) Given a complex X· of F -acyclic objects (i.e. RqF (Xi) = gories, the first with suff. many injectives, and an object A 0 for every q > 0) then RpF (X·) ' HpF (X·) and moreover · p q · p+q · of Aff , we take the filtered injective resolution A and we R F H (X ) =⇒ H F (X ). In fact the first sp.seq. de- define the spectral sequence of F w.r.t. A as the spectral se- generates in I Ep,0(X·) = Hp R0F (X·) = HpF (X·). quence associated to the filtered complex F (A·) with filtration 2 F i F (A·) = F F i(A·). Limit for this sp.seq. is the coho- (3) By the previous two points, we can calculate the derived func- tors of F using F -acyclic resolutions of the objects of A. mology Hp+q F (A·) = Rp+qF (A) with the filtration (4) Considering two functors F : A−→B and G : B−→C and A ∈ i p+q p+q i · p+q · F R F (A) = im H F F (A ) −→H F (A ) A, if A has sufficiently many injectives we can resolve A with = im Rp+qF F i(A)−→Rp+qF (A) I· ∈ C(I), then apply the spectral sequences of G w.r.t. F (I·); = ker Rp+qF (A)−→Rp+qF A/F i(A) . we have: p,q I E (FI·) = Hp RqG(FI·) II Ep,q(FI·) = RpG Hq(FI·) The terms in the first level are given by: 2 2 = Hp RqG◦F (I·) = RpG(RqF (A)) Ep,q = Hp+q grp F (A·) = Rp+qF (grp(A)) . 1 = Rp (RqG◦F )(A) = (RpG◦RqF )(A) Hypercohomology. sequences with the same limit: the hypercohomology of G◦F w.r.t. A, R·G F (I·). (0) Given a complex C· in C(A), a Cartan-Eilenberg resolution of (5) Leray sp.seq.: if F sends injective objects of A into G-acyclic C· is a bicomplex L·· with L·j = 0 for j < 0 with a morphism objects of B and G is left exact, then R·G F (I·) ' Rp(G◦ of complexes C· −→ L·0 s.t. F )(A) and RpG◦RqF =⇒ Rp+q(G◦F ). In fact the first sp.seq. i (i) for any i the sequence 0−→Ci −→ Li· is exact and induce I p,0 · p 0 p degenerates in E2 (FI ) = R R G◦F (A) = R (G◦F )(A) exact sequences by the coboundary, cocycles and cohomol- by the acyclicity assumption. ogy functors; (6) In the general situation of (4) we can write the edge mor- ·j (ii) for any j the simple complex L admits split short exact phisms: sequences Rp R0G◦F (A)−→RpG F (I·)−→R0 (RpG◦F )(A) 0−→Bi(L·j)−→Zi(L·j)−→Hi(L·j)−→0 i ·j i,j i+1 ·j 0−→Z (L )−→L −→B (L )−→0 (RpG◦R0F )(A)−→RpG F (I·)−→(R0G◦RpF )(A) for any i. and by composition (first up and second down) we have a i,j The resolution is injective if every object L is (so that by the canonical morphism: second condition, also the coboundary, cocycles and cohomol- Rp R0G◦F −→R0G◦RpF. ogy objects are). (1) If A has enough injectives then every complexes admits an in- If G is an exact functor, then this is an isomorphism and we jective Cartan Eilenberg resolution and a morphism between write Rp (G◦F ) ' G◦RpF . In fact the exactness of G implies f · the degeneracy of both sp.seq. and the edge morphisms are complexes C· −→ C0 can be lift to a morphism between the isomorphisms. F ·· resolutions L·· −→ L0 , so that we obtain an exact “resolu- 2 (7) If F is an exact functor, without hypothesis on G, then the tion functor” CE : C(A) → C (I). Moreover homotopic mor- II p,0 · p phisms go to homotopic morphism between resolutions. Ob- second sp.seq. degenerates in E2 (FI ) = (R G◦F )(A) ' viously, bounded complexes admit resolutions bounded in the RpG F (I·) and we have Rp (RqG◦F )(A) =⇒ (Rp+qG◦F )(A) same way. so that by the first sp.seq. which is cohomological we have the (2) Let F : A−→B be a covariant functor between abelian cate- exact sequence of the low degree terms: gories; we (partially) define two functors C(A)−→SpS(B) in 0 → R1(R0G◦F ) → R1G◦F → R0(R1G◦F ) → R2(R0G◦F ) → R2G◦F the following way: starting with a complex C· we take an in- jective CE resolution L··, if there exists the simple complex as- but nothing more. sociated to F (L··) ∈ C2(B) we use the two spectral sequences (8) Let X· ∈ C(A) and I·· ∈ C2(A) be a resolution of X· with the associated to the bicomplex: they are the spectral sequences second degree positive, not necessarily CE, with F -acyclic ob- of hypercohomology of F w.r.t. the complex C·. The limit of jects of A for a functor F : A → B. We define a bicomplex · nXp se q = 0 the spectral sequences is the hypercohomology of F w.r.t. C X·· ∈ C2(A) by Xp,q = . The first spectral se- indicated as RnF (C·) = Hn F (L··) and the terms in the first 0 se q 6= 0 level are: quences associated to X·· and I·· give I p,q ·· p q · II p,q ·· p q · p,q q nH0 (Ip,·) ' Xp if q = 0 p,q E2 (FL ) = H R F (C ) E2 (FL ) = R F H (C ) I E (I··) = H (Ip,·) = II ' I E (X··) 1 II 0 if q 6= 0 1 where RqF (C·) = RqF (Ci) . i∈Z wence the canonical X·· → I·· gives isomorphisms on the limit: (3) the first spectral sequence does not require a CE resolution, n · n ·· n ·· n ·· · H (X ) = Htot(X ) ' Htot(I ) = H s(I ) so that X and I p,q ·· q p and it starts with E1 (FL ) = R F (C ). s(I··) are quasi-isomorphic. Moreover RF (X·) ' RF s(I··) ' ·· ·· n · n ·· Standard results: F s(I ) ' s F (I ) and also H RF (X ) ' Htot F (I ) . · (1) Given A ∈ A and an arbitrary resolution C· (i.e. a positive For the first sp.seq. associated to CE F (X ) we have (terms degree complex s.t. 0 → A → C· is exact) then RpF (C·) ' at level one) RqF (Xp) =⇒ Rp+qF (X·), and for the first sp.seq. p p q · p+q ·· I p,q ·· R F (A) and H R F (C ) =⇒ R F (A). In fact the sec- associated to the bicomplex F (I ) we have E1 F (I ) = ond sp.seq. degenerates: II Ep,q(C·) = RpF Hq(C·) = 0 if q p,· q p 2 HII F (I ) ' R F (X ) and so we have a canonical mor- II p,0 · p 0 · p p+q · p+q ·· p+q · q > 0 and E2 (C ) = R F H (C ) ' R F (A). phism at the limit: R F (X ) ' Htot F (I ) ' H RF (X )
6 GeoSupRC - Spectral Sequences c 2000 by MC (we are used RF (X·) in the sense of derived categories, and the In case that M is R-flat we deduce that symbol ' means quasi-isomorphic, i.e. isomorphism in the de- S ∼ S M ⊗R Torq (T,N)=Torp (M ⊗R T,N); rived category). and if N is S-flat we have Tricomplexes and Hypercohomology of bicomplexes. R ∼ R Torp (M,T ⊗S N)=Torp (M,T )⊗S N. Let A be an abelian category and X··· ∈ C3(A) (category of tricomplexes of A). Under the usual hypothesis we have three In particular if we have a morphism R → S, we can choose ways to reduce X··· to a bicomplex, associating the indices T = S and we have the base change spectral sequence i,j,k as i+j = m, k = n or i+k = m, j = n or i = m, j +k = n. S R R Torp (Torq (M,S),N) =⇒ Torp+q(M,N) . Accordingly, we have six sp.seq. converging to the same limit Exchanging the roles of R and S we can start with N ∈ mod-S, (the total cohomology): T ∈ S-mod-R, M ∈ R-mod; from the associativity property for m n ··· HI,II HIII (X ) N ⊗S T ⊗R M we obtain two spectral sequences m n ··· HIII HI,II (X ) TorS(N,TorR(T,M)) Hm Hn (X···) p q · · I,III II m+n ··· R S =⇒ Hp+q(N ⊗R T ⊗S M ) m n ··· =⇒ Htot (X ) . Torp (Torq (N,T ),M) HII HI,III (X ) m n ··· HI HII,III (X ) In case that N is S-flat we deduce that m n ··· HII,III HI (X ) R ∼ R N ⊗S Torq (T,M)=Torp (N ⊗S T,M); Let F : A → B an additive functor between abelian categories, and if M is R-flat we have A having sufficiently many injectives. Then we define the hy- S ∼ S perderived functors of F on a bicomplex X·· ∈ C2(A) as the Torp (N,T ⊗R M)=Torp (N,T )⊗R M. total cohomology of the tricomplex F (I···) where I··· is an in- ·· In particular if we have a morphism R → S, we can choose jective CE-resolution of X . We have the following spectral T = S and we have the base change spectral sequence sequences with limit the hypercohomology of F on X··: TorS(N,TorR(S,M)) =⇒ TorR (N,M) . Hm(RnF (X··)) p q p+q m n ·· R F (Htot(X )) Kunneth-dual spectral sequences. Let L ∈ C(mod-A) and m n ·· R F (HII (X )) m+n ·· M ∈ C(A-mod); then we have two spectral sequences m n ·· =⇒ R F (X ) . H (R F (X )) p q00 ) II I LExt (H 0 (L),H (M)) m n ·· A q H (R F (X )) 0 00 A I II q +q =q =⇒ Extp+q(L,M) m n ·· p q RII F (HI (X )) Htot(ExtA(L,M)) Suppose now one has G : B → C another functor, with B having and in case L is projective or M is injective we have enough G-acyclic objects. We start with a complex X· ∈ C(A), ·· M p q00 p we take an injective CE-resolution I and calculate the hyper- Ext (Hq0 (L),H (M)) =⇒ H (Hom(L⊗A M)) . ·· 2 A tot cohomology of G on the bicomplex F (I ) ∈ C (B). We have q0+q00=q the following sp.seq. For M ∈ R-mod, T ∈ S-mod-R, N ∈ S-mod, from the adjoint m n · R (R G◦F )(X ) property Hom (M,Hom (T,N))∼Hom (T ⊗ M,N) we ob- m n · R S = S R R G(R F (X )) tain two spectral sequences m n · R G(R F (X )) m+n ·· p q =⇒ R G(FI ) . Ext (M,Ext (T,N)) p+q · ∼ Rm(RnG◦F )(X·) R S H (HomR(M·,HomS(T,N )))= p R =⇒ ∼ p+q · Hm(Rn G(FI··)) Ext (Tor (T,M),N) =H (HomS(M· ⊗R T,N )) . I II S q Rm F (Hn(I··)) II I In case that M is R-projective we deduce that Suppose that F sends injective objects of A to G-acyclic objects q ∼ q of B; then we obtain the following situation: HomR(M,ExtS(T,N)=ExtS(T ⊗R M,N); RmG(RnF (X·)) and if N is S-injective we have RmG(RnF (X·)) p ∼ R =⇒ Rm(R0G◦F )(X·) . Ext (M,HomS(T,N)=HomS(Torp (T,M),N) . Rm(G◦F )Hn(X·) R Hm(Rn(G◦F )(X·)) In particular if we have a morphism R → S, we can choose T = S and we have the Hom-base change spectral sequence Extp (TorR(S,M),N) =⇒ Extp+q(M,N) . Algebraic spectral sequences. S q R Exchanging the roles of R and S we can start with N ∈ S-mod, Kunneth spectral sequences. Let L ∈ C(mod-A) and M ∈ T ∈ R-mod-S, M ∈ R-mod; from the usual adjoint property ∼ C(A-mod); then we have two spectral sequences HomS(N,HomR(T,M))=HomR(T ⊗S N,M) we obtain two spec- 0 00 tral sequences LTorA(Hq (L),Hq (M))) p p q q0+q00=q =⇒ TorA (L,M) Ext (N,Ext (T,M)) Hp+q(Hom (N ,Hom (T,M·)))=∼ p+q S R =⇒ S · R p A p S =∼Hp+q(Hom (T ⊗ N ,M·)) . Htot(Torq (L,M)) ExtR(Torq (T,N),M) R S · and in case L or M is flat we have In case that N is S-projective we deduce that M 0 00 p A q q q p Torp (H (L),H (M)) =⇒ Htot(L⊗A M) . ∼ HomS(N,ExtR(T,M)=ExtR(T ⊗S N,M); 0 00 q +q =q and if M is R-projective we have For M ∈ mod-R, T ∈ R-mod-S, N ∈ S-mod, from the associa- p ∼ S Ext (N,HomR(T,M))=HomR(Torp (T,N),M) . tivity property for M ⊗R T ⊗S N we obtain two spectral se- S quences In particular if we have a morphism R → S, we can choose R S T = S and we have the Hom-base change spectral sequence Torp (M,Torq (T,N)) · · S R =⇒ Hp+q(M ⊗R T ⊗S N ) p q p+q Torp (Torq (M,T ),N) ExtS(N,ExtR(S,M)) =⇒ ExtR (N,M) .
7 GeoSupRC - Spectral Sequences c 2000 by MC Duality spectral sequences. In the situation M ∈ R-mod, Spectral sequences in sheaf theory. T ∈ R-mod-S, N ∈ mod-S, from the canonical morphism Cohomology and Hypercohomology of (global sections HomS(T,N)⊗R M −−→HomS(HomR(M,T ),N) of) Sheaves: we obtain canonical morphisms (1) Hn(X,F ) is the n-th cohomology group of the sheaves F : TorR(Hom (T,N),M)−−→Hom (Ext (M,T ),N) . S S R Hn(X,−) = RnΓ(X,−): Ab(X)−→Ab Let be R left noetherian, M of finite type over R; then if N · is injective, the morphisms are iso. In particular we have the (2) the hypercohomology groups of the global sections on K : following spectral sequences Hn(X,K·) = RnΓ(X,−)(K·) TorR (Extq (T,N),M) p+q · ∼ −q S H (HomS(T,N )⊗R M·)= is a complex for which the two sequences p −q =⇒ ∼ p+q · =H (HomS(HomR(M·,T ),N )) . ExtS(ExtR (M,T ),N) I p,q p q · II p,q p q · E2 = H R Γ(X,−)(K ) E2 = R Γ(X,−) H (K ) Other spectral sequences. In the case M ∈ R-mod, T ∈ = Hp Hq(X,K·) = Hp X,Hq(K·) R-mod-S, N ∈ S-mod, we have a canonical morphism I p,q p q p HomR(M,T )⊗S N −−→HomR(M,T ⊗S N) converge (the first starts with: E1 = C (H (X,K ))). and, in case R is left noetherian and M of finite type over R, we can write the spectral sequences Cohomology and Hypercohomology of Coverings: TorS(Extq (M,T ),N) p+q ∼ (0) Given a topological space X, a presheaf F on X and an open p R Htot (HomR(M·,T )⊗S N·)= q+1 =⇒ covering of X U = (Ui)i∈I , or I = (I0,···,Iq) ∈ I we will Extp (M,TorS(T,N)) ∼ p+q R q =Htot HomR(M·,T ⊗S N·) q write U = T U and also ϕ ∈ F (U ). The complex of alter- Spectral sequences of group cohomology. I I` I I `=0 Let G be a group and M ∈ Z[G]-mod (resp. M ∈ mod-Z[G]). nating q-cochains with values in F is given by: The cohomology (resp. homology) groups of G with values in ( ) [G] M is Hn(G,M) = Extn ( ,M) (resp. H (G,M) = TorZ (M, )). Cq(U,F ) = ϕ ∈ QF (U )| ϕ = sgn(σ)ϕ ∀I,∀σ ∈ Σ(I) [G] Z n n Z I I σ(I) Z q+1 n n I∈I We have that H (G,M) = H (Hom [G](B·(Z[G],Z),M)) and Z q q q+1 Hn(G,M) = Hn(M ⊗ [G] B·(Z[G],Z)) where B·(Z[G],Z) is the with differentials d : C (U,F )−→C (U,F ) sending ϕ = Z q q+2 standard resolution of Z as (trivial) Z[G]-module, i.e. Bi(Z[G],Z) = (ϕI )I∈Iq+1 in d (ϕ) defined for J ∈ I by ⊗i+2 Z⊗ [G] Z[G] with projection : B0(Z[G],Z)−→Z. q+1 Z q P ` d (ϕ)J = (−) ϕ |U The cohomology of G with values in M is determined also (J0,···,Jˆ`,···,Jq+1) J as Hn(G,M) = Hn(C·(G,M)) where C·(G,M) is the complex `=0 n n given by C (G,M) = Hom(G ,M) with differentials If I is ordered we can write Cq(U,F ) = QF (U ). (I0,...,Iq ) n I <···