MATH 101B: ALGEBRA II PART A: HOMOLOGICAL ALGEBRA 23

Transcription

1 MATH 101B: ALGEBRA II PART A: HOMOLOGICAL ALGEBRA Homotopy uniqueness of projective resolutions. Here I proved that the projective resolution of any R-module (or any object of an abelian category with enough projectives) is unique up to chain homotopy. I used diagrams and the (equivalent) equation. First I wrote down the understood standard interpretation of a diagram. Lemma 6.7. Given that the solid arrows (in below) form a commuting diagram, there exists a dotted arrow (the arrow labeled f is supposed to be dotted) as indicates making the entire diagram commute. [This is the understood meaning of this kind of diagram.] The dotted arrow is not necessarily uniquely determined (it is labeled and not!) The assumptions are that P is projective and g : A B is onto. C n+1 P f f C n d n+1 d n 0 C n 1 Proof. By definition of kernel, f lifts uniquely to ker d n. But C n+1 maps onto ker d n. So, by definition of P being projective, f lifts to C n+1. Lemma 6.8. With standard wording as above. The additional assumptions are that the right hand column is exact (i.e., im d C n+1 = ker d C n ), P n+1 is projective and the left hand column is a chain complex (i.e. d P n d P n+1 = 0). f n+1 P n+1 C n+1 P n d P n+1 f n C n d C n+1 d P n d C n P n 1 f n 1 C n 1 Proof. The assumptions implies d C n (f n d P n+1) = 0. By the previous lemma this implies that f n d P n+1 lifts to C n+1.

2 24 MATH 101B: ALGEBRA II PART A: HOMOLOGICAL ALGEBRA Theorem 6.9. Suppose P M 0 is a projective chain complex (augmented) over M and C N 0 is a resolution of N (i.e., an exact sequence). Suppose f : M N is any morphism. Then (1) There exists a chain map f : P C over f. I.e., the following diagram commutes: P ɛ M f f C ɛ N (2) f is unique up to chain homotopy. Proof. (1) Since ɛ : C 0 N is an epimorphism and P 0 is projective, the map f ɛ : P 0 N lifts to a map f 0 : P 0 C 0. The rest is by induction using lemma we just proved. (2) To prove the existence of the homotopy, I first restated Lemma 6.7 as an equation. It says that for any homomorphism f : P C n so that d n f = 0, there exists a homomorphism f : P C n+1 so that d n+1 f = f. We want to show that f is unique up to homotopy. So, suppose f, g are two chain maps over f : M N. Then we want to show that there exists a sequence of morphisms h n : P n C n+1 so that d C n+1 h n + h n 1 d P n = g n f n We set h 1 = 0 by definition. So, for n = 0 we get: d 1 h 0 = g 0 f 0 First, h 0 exists because ɛ(g 0 f 0 ) = (f f)ɛ = 0. If h 0,, h n 1 exist satisfying the above equation then in particular we have: (6.1) d n h n 1 + h n 2 d n 1 = g n 1 f n 1 We want to show that h n exists satisfying the equation d C n+1 h n = g n f n h n 1 d P n The right hand side is the f of Lemma 6.7 and the map that we want (h n ) is the f in Lemma 6.7. So, all we need to do is show that d C n f = 0. d n (g n f n h n 1 d n ) = d n g n d n f n d n h n 1 d n = g n 1 d n f n 1 d n d n h n 1 d n Factoring out the d n and using Equation (6.1) we get: = (g n 1 f n 1 d n h n 1 )d n = h n 2 d n 1 d n = 0 This is 0 since d n 1 d n = 0. Thus h n exists and f g.

3 MATH 101B: ALGEBRA II PART A: HOMOLOGICAL ALGEBRA 25 This gives us the statement that we really want: Corollary In any abelian category with enough projectives, any object A has a projective resolution P A. Furthermore, any two projective resolutions of A are homotopy equivalent. Proof. If there are two projective resolutions P, P then the first part of the theorem above tells us that there are chain maps f : P P and g : P P which cover the identity map on A. Since g f and the identity map are both chain maps P P over the identity of A, the second part of the theorem tells us that f g id P and similarly g f id P. So, P P. The dual argument gives us the following. [In general you should state the dual theorem but not prove it.] Theorem In any abelian category with enough injectives, any object B has an injective coresolution. Furthermore, any two injective coresolutions of B are homotopy equivalent. Following this rule, I should also give the statement of the dual of the previous theorem: Theorem Suppose 0 M Q is an injective cochain complex under M and 0 N C is a coresolution of N (i.e., a long exact sequence). Suppose f : N M is any morphism. Then (1) There exists a cochain map f : C Q under f. I.e., the following diagram commutes: C N f Q f M (2) f is unique up to chain homotopy.

4 26 MATH 101B: ALGEBRA II PART A: HOMOLOGICAL ALGEBRA 6.5. Derived functors. Definition Suppose that A, cb are abelian categories where A has enough injectives and F : A B is a left exact (additive) functor. Then the right derived functors R i F are defined as follows. For any object B of A choose an injective coresolution B Q and let R i F (B) be the ith cohomology of the cochain complex F (Q ): 0 F (Q 0 ) F (Q 1 ) F (Q 2 ) In the case F = Hom R (A, ), the right derived functors are the Ext functors: Ext i R(A, B) := R i F (B) = H i (Hom R (A, Q )) Note that the derived functors are only well-defined up to isomorphism. If there is another choice of injective coresolutions Q then Q Q which implies that F (Q ) F (Q ) which implies that H i (F (Q )) = H i (F (Q )) I pointed out later that, for R-modules, there is a canonical minimal injective coresolution for any module. By definition of F (Q ) we take only the injective objects. The term F (B) is deliberately excluded. But F is left exact by assumption. So we have an exact sequence Thus 0 F (B) F (Q 0 ) F (Q 1 ) Theorem The zero-th derived functor R 0 F is canonically isomorphic to F. In particular, Ext 0 R(A, B) = Hom R (A, B) At this point we tried to do an example: Compute Ext 1 Z(Z/3, Z/2). We took an injective coresolution of Z/2 0 Z/2Z Q/2Z j Q/Z 0 We used that fact that any quotient of a divisible group is divisible. We took Hom(Z/3, ) into the injective part: j : Hom(Z/3, Q/2Z) Hom(Z/3, Q/Z) Then I claimed that this map is an isomorphism. Here is a simpleminded proof. A homomorphism Z/3 Q/Z is given by its value on the generator 1 + 3Z of Z/3Z. This must be a coset a/b + Z so that 3a/b Z. In other words b = 3 and a = 0, 1 or 2. Similarly, a homomorphism Z/3 Q/2Z sends the generator of Z/3 to a coset

5 MATH 101B: ALGEBRA II PART A: HOMOLOGICAL ALGEBRA 27 a/b + 2Z so that 3a/b 2Z. So, b = 3 and a = 0, 2 or 4. So, both of there groups have exactly three elements and a simple calculation shows that j is a bijection delta operator. On of the basic properties of the derived functors is that they fit into an a long exact sequence. Theorem Given any short exact sequence 0 A α B β C 0 there is a sequence of homomorphisms δ n : R n F (C) R n+1 F (A) making the following sequence exact: 0 F (A) F (B) F (C) δ 0 R 1 F (A) R 1 F (B) R 1 F (C) δ 1 R 2 F (A) R 2 F (B) R 2 F (C) δ 2 R 3 F (A) Furthermore, δ n is natural in the sense that, given any commuting diagram with exact rows: we get a commuting square: α 0 A B C 0 f 0 A α B β C 0 R n F (C) h g δ n β h R n+1 F (A) f R n F (C δ n ) R n+1 F (A ) I gave the following construction of these δ operators. First I needed the following lemmas, the first being obvious. Lemma If Q is injective then R n F (Q) = 0 for all n 1. Lemma If 0 A Q K 0 is a short exact sequence where Q is injective, then we have an exact sequence and for all n 1. 0 F (A) F (Q) F (K) R 1 F (A) 0 R n F (K) = R n+1 F (A)

6 28 MATH 101B: ALGEBRA II PART A: HOMOLOGICAL ALGEBRA Proof. We can use Q = Q 0 as the beginning of an injective coresolution of A: 0 A j j 0 j 1 j 2 Q 0 Q1 Q2 Q3 Since coker j = im j 0 = ker j1 = K, we can break this up into two exact sequences: 0 A j Q 0 K 0 j 1 j 2 0 K Q 1 Q2 Q3 The second exact sequence shows that the injective coresolution of K is the same as that for A but shifted to the left with the first term deleted. So, R n F (K) = R n+1 F (A) for all n 1. When n = 0 we have, by left exactness of F, the following exact sequence: 0 F (K) F (Q 1 ) (j 1) F (Q2 ) In other words, F (K) = ker(j 1 ). The image of (j 0 ) : F (Q 0 ) F (Q 1 ) lands in F (K) = ker(j 1 ). The cokernel is by definition the first cohomology of the cochain complex F (Q ) which is equal to R 1 F (A). So, we get the exact sequence F (Q 0 ) F (K) R 1 F (A) 0 We already know that the kernel of F (Q 0 ) F (K) is F (A) so this proves the lemma. My construction of the delta operator proceeded as follows. Start with any short exact sequence 0 A B C 0. Then choose an injective coresolution of A: 0 A j j 0 j 1 j 2 Q 0 Q1 Q2 Q3 Let K = ker j 1 = im j 0 = coker j. Since Q 0 is injective, the map A Q 0 extends to B and cokernels map to cokernels giving a commuting diagram: (6.2) α 0 A B C 0 id A 0 A j Q p 0 K 0 The map g : C K induces a map g : R n F (C) R n F (K) and I defined the connecting homomorphism δ n for n 1 to be the composition: δ n : R n F (C) g R n F (K) = R n+1 (A) f β g

7 MATH 101B: ALGEBRA II PART A: HOMOLOGICAL ALGEBRA 29 I showed that this is independent of the choice of g : C K since, for any other choice g, the difference g g lifts to Q 0 since f f : B Q 0 is zero on A and therefore factors through C. So, g g = R n F (g g ) factors through R n F (Q 0 ) = 0 so g = g. To show independence from the choice of Q 0 I said that there was a canonical choice for Q 0 called the injective envelope of A and I promised to write up the proof of that. What about n = 0? In this case, Lemma 6.17 gives us a 4 term exact sequence: 0 F (A) F (Q 0 ) F (K) R 1 F (A) 0 So, we can define δ 0 : F (C) R 1 F (A) to be the composition δ 0 : F (C) g F (K) R 1 F (A) Again, for any other choice g : C K, the difference g g factors through Q 0. This time F (Q 0 ) 0. But that is OK since F (Q 0 ) is in the kernel of the next map F (K) R 1 F (A) by the 4 term exact sequence Proof of Theorem From your homework you might remember that the sequence (6.2) gives a short exact sequence: 0 B (f β) Q 0 C ( p,g) K 0 Since R n F (Q 0 ) = 0 the top row in the following sequence is supposed to be exact: δ n 1 R n 1 F (K) R n F (B) R n F (C) R n F (K) β = = = = R n F (A) α R n F (B) β R n F (C) δn R n+1 F (A) In the top sequence R n F (C) occurs in position 3n 1 and in the bottom sequence it occurs in position 3n. Therefore, exactness of the bottom sequence for all 0 A B C 0 at position k 1 implies the exactness of the top sequence at position k 1 which implies the exactness of the bottom sequence at position k. So, it is exact everywhere, proving theorem. We just need to check that the diagram commutes. (Actually it doesn t. But that s OK.) The middle square obviously commutes. The right hand square commutes by definition of δ n. The left square anticommutes (i.e., going one way is negative the other way). But that is good enough for the argument to work. This will follow from the way that α and δ n 1 are defined. g

8 30 MATH 101B: ALGEBRA II PART A: HOMOLOGICAL ALGEBRA The morphism α : A B induces a cochain map of injective coresolutions: 0 A j Q 0 Q 1 Q 2 α α 0 α 1 α 2 0 B jb Q B 0 Q B 1 Q B 2 The cochain map α induces the cochain map α : F (Q ) F (Q B ). The induced map in cohomology is α : R n F (A) R n F (B) by definition. If the cokernels of j, j B are K, L we get the commuting diagrams 0 A Q 0 K 0 α j jb α0 0 B Q B 0 L 0 p pb α 0 K Q 1 Q 2 α α 1 α 2 0 L Q B 1 Q B 2 Just as we had R n F (K) = R n+1 F (A) we also have R n F (L) = R n+1 F (B) and the above diagrams show that the maps of injective coresolutions induced by α : A B and α : K L are the same but shifted. In other words, α : R n F (A) R n F (B) is the same as the map α : R n 1 F (K) R n 1 F (L). We need one more commuting diagram: (f β) 0 B Q 0 C K 0 = j B ( p,g) (α0, h) p B α 0 B Q B 0 L 0 Here h : C Q B 0 is the morphism needed to make the diagram commute, i.e., the maps α 0 f, j B : B Q B 0 are not equal. But they agree on A. So their difference factors through C. I.e. h : C Q B 0 so that h β = α 0 f j B The coboundary map δ n 1 : R n 1 F (K) R n F (B) is given by the composition δ n 1 : R n 1 F (K) α R n 1 F (L) = R n F (B) By what I said in the last paragraph, this is the same as α : R n F (A) R n F (B) proving the theorem. (The n = 1 case is slightly different.)

9 MATH 101B: ALGEBRA II PART A: HOMOLOGICAL ALGEBRA Left derived functors. There are two cases when we use projective resolutions instead of injective coresolutions: (1) When the functor is left exact but contravariant, e.g., F = Hom R (, B). (2) When the functor is right exact and covariant, e.g., F = R B. In both cases we take a projective resolution P A 0 and define the left derived functors to be L n F (A) = H n (F (P )) in the first case and L n F (A) = H n (F (P )) in the second case. Definition The left derived functors of F (A) = A R B are called L n F (A) = Tor R n (A, B) review of tensor product. Following Lang, I will take tensor products only over commutative rings. The advantage is that A R B will be an R-module. The tensor product is defined by a universal condition. Definition Suppose that A, B are modules over a commutative ring R. Then a map g : A B C from the Cartesian product A B to a third R-module C is called R- bilinear if it is an R-homomorphism in each variable. I.e., for each a A, the mapping b g(a, b) is a homomorphism B C and similarly g(, b) Hom R (A, C) for all a A. A R B is defined to be the R-module which is the target of the universal R-bilinear map f : A B A B When I say that f is universal I mean that for any other R-bilinear map g : A B C there is a unique R-homomorphism h : A B C so that g = h f. The universal property tells us that A R B is unique if it exists. To prove existence we need to construct it. But this easy. You just take A R B to be the free R-module generated by all symbols a b where a A, b B modulo the relations that are required, namely: (1) (ra) b = r(a b) (2) (a + a ) b = a b + a b (3) a rb = r(a b) (4) a (b + b ) = a b + a b I pointed out that the universal property can be expressed as an isomorphism Hom R (A B, C) = BiLin R (A B, C)

10 32 MATH 101B: ALGEBRA II PART A: HOMOLOGICAL ALGEBRA And the definition of R-bilinear can be expressed as the isomorphisms BiLin R (A B, C) = Hom R (A, Hom R (B, C)) = Hom R (B, Hom R (A, C)) So, we conclude that This is a special case of: Hom R (A B, C) = Hom R (A, Hom R (B, C)) Hom R (F (A), C) = Hom R (A, G(C)) with F = B and G = Hom R (B, ). When we have this kind of isomorphism, F is called the left adjoint and G is called the right adjoint and F, G are called adjoint functors. Lemma Any left adjoint functor is right exact. In particular, tensor product is right exact. Also, any right adjoint functor is left exact. Proof. In the first case, suppose that F is a left adjoint functor and (6.3) 0 A α A β A 0 is a short exact sequence. Then for any C, the left exactness of Hom R (, G(C)) gives an exact sequence 0 Hom R (A, G(C)) Hom R (A, G(C)) Hom R (A, G(C)) By adjunction, this is equivalent to an exact sequence 0 Hom R (F (A ), C)) Hom R (F (A ), C)) Hom R (F (A), C)) The exactness of this sequence for all C is equivalent to the exactness of the following sequence by definition of coker F (α): F (A) F (α) F (A ) F (A ) 0 The left exactness of G is analogous. (Also, the proof uses the left exactness of Hom so the second case is dumb.) Take the sequence (6.3) and suppose that it splits. I.e., A = A A and there is a retraction r : A A so that r α = id A. Then, in the exact sequence F (A) F (α) F (A ) F (A ) 0 F (α) is a monomorphism since F (r) F (α) = F (r α) = id F (A). So, we get a short exact sequence which furthermore splits. This proves the following.

11 MATH 101B: ALGEBRA II PART A: HOMOLOGICAL ALGEBRA 33 Lemma If F is any right (or left) exact functor then F (A A ) = F (A) F (A ). In particular, (A A ) R B = (A R B) (A R B) Another important lemma was the following. Lemma R R B is isomorphic to B as R-modules. Proof. I showed that B satisfies the universal property. Let f : R B B be the map f(r, b) = rb. This is R-bilinear when R is commutative. Suppose that g : R B C is another R-bilinear map. Then we can define h : B C by h(b) = g(1, b). This is R-linear since g is R-bilinear. The required diagram commutes since h f(r, b) = h(rb) = g(1, rb) = rg(1, b) = g(r, b) Furthermore, h is unique since it has no choice but to send b to g(1, b). Since B satisfies the universal property, B = R B. Also the proof gives the isomorphism. r b R B corresponds to rb B. There was one other lemma that I didn t prove because it was obvious. Lemma A B = B A computations. With these lemmas, I did some computations. Suppose that R = Z and A = Z/n. Then a projective resolution of A is given by 0 Z n Z Z/n 0 Since this sequence is exact it gives the following right exact sequence for any abelian group B: Z B n Z B Z/nZ B 0 Using the lemma that R R B = B this becomes: So, we conclude that B n B Z/nZ B 0 Z/nZ B = B/nB More generally, if A is any finitely generated abelian group then A = Z r Z/t 1 Z/t 2 Z/t n and, since tensor product distributes over direct sum we get: A Z B = B r B/t 1 B B/t 2 B B/j n B

12 34 MATH 101B: ALGEBRA II PART A: HOMOLOGICAL ALGEBRA The derived functor Tor Z 1 (Z/n, B) is by definition the kernel of the map Z B n Z B Since Z B = B this is just the map B B given by multiplication by n. So, Tor Z 1 (Z/n, B) = {b B nb = 0} It is the subgroup of B consisting of all elements whose order divides n. It is the n-torsion subgroup of B. Maybe that is why it is called Tor extension of scalars. Suppose that R is a subring of S (e.g., Z R). A homomorphism of free R-modules R n f R m is given by a matrix M(f) = (a ij ) as follows. If the basis elements of R n are e j and the basis elements of R m are e i then m f(e j ) = a ij e i for some a ij R. These numbers determine f since, for an arbitrary element x = x j e j R n we have ( ) f(x) = f x j e j = x j a ij e i = a ij x j e i i,j i,j j since R is commutative. (Take free right R-modules when R is not commutative and this will still work.) This can be written in matrix form: x 1 a1j x j x 1 f x 2 = a2j x j = (a ij) x 2 x n anj x j x n When you tensor with S you get R n R S = (R R S) n = S n R n R S = S n i=1 f id S R m R S = S m The claim is that M(f id S ) = M(f). So, f = f id S is obtained from f by extending scalars to S. If you have an integer matrix, you just take the same matrix and consider it as a real matrix.

13 MATH 101B: ALGEBRA II PART A: HOMOLOGICAL ALGEBRA two definitions of Ext. The last thing I did was to prove that the two definitions of Ext n R(A, B) that we now had were equivalent. Theorem If P A is a projective resolution of A and B Q is an injective resolution of B then H n (Hom(P, B)) = H n (Hom(A, Q )) So, either formula gives Ext n R(A, B). Proof. The theorem is true in the case when n = 0 because both sides are are isomorphic to Hom R (A, B). So, suppose n 1. I gave the proof in the case n = 2. I want to construct a homomorphism H n (Hom(A, Q )) H n (Hom(P, B)) So, take an element [f] H n (Hom(P, B)). The notation means f ker((j 2 ) : Hom R (A, Q 2 ) Hom R (A, Q 3 )) [f] = f + im((j 1 ) : Hom R (A, Q 1 ) Hom R (A, Q 2 )) This gives the following diagram (6.4) d 3 d 2 d 1 P 3 P 2 P 1 P 0 A f 3 f 2 f 1 f 0 f 0 B Q 0 Q 1 Q 2 Q 3 j 0 Since j 2 f = 0, f maps to the kernel K of j 2. But P is a projective resolution of A and B Q 0 Q n 1 K is a resolution of K. So, we proved that there is a chain map from P to this resolution of K which is unique up to chain homotopy. This gives the maps f 0, f 1, etc in the diagram. Note that f 2 d 3 = 0 f 3 = 0. So, f 2 ker((d 3 ) : Hom R (P 2, B) Hom R (P 3, B)) But f 2 is only well defined up to homotopy h : P 1 B. So, we could get f 2 = f 2 + h d + d h. But the second term must be zero since it goes through 0 and the first term This means that h d 2 im((d 2 ) : Hom R (P 1, B) Hom R (P 2, B)) d 0 j 1 [f 2 ] = f 2 + im(d 2 ) is a well defined element of H 2 (Hom(P, B)) and we have a homomorphism: Hom R (A, ker j 2 ) H 2 (Hom(P, B)) j 2

14 36 MATH 101B: ALGEBRA II PART A: HOMOLOGICAL ALGEBRA But this homomophism is zero on the image of (j 1 ) : Hom(A, Q 1 ) Hom(A, Q 2 ) because, if f = j 1 g then we can take f 0 = g d 0 and f 1 = 0 = f 2. Therefore, we have a well defined map H 2 (Hom(A, Q )) H 2 (Hom(P, B)) which sends [f] to [f 2 ]. This is enough! The reason is that the diagram (6.4) is symmetrical. We can use the dual argument to define a map H 2 (Hom(P, B)) H 2 (Hom(A, Q )) which will send [f 2 ] back to [f]. So, the two maps are inverse to each other making them isomorphisms. By the way, this gives a symmetrical definition of Ext n, namely it is the group of homotopy classes of chain maps from the chain complex P A 0 to the cochain complex 0 B Q shifted by n. Elements of Ext n R(A, B) are represented by vertical maps as in Equation (6.4).