Author: Eiko

Tags: Derived Category, Homological Algebra, Derived Functors, Semi-Projective Resolutions, dg-category, Ring Theory, Derived Tensor Evaluation, Derived Homomorphism Evaluation, Dualizing Complexes, Koszul Complex

Time: 2024-08-13 12:00:00 - 2024-08-17 12:00:00 (UTC)

This is the notes I took at the seminar Masterclass in Derived Category Methods in Ring Theory, Aarhus University, 13-16 August 2024. Some of the notations have been altered for my own preference.

Category of Chain Complexes

Let R be a ring acting on the left, and is a k-algebra (e.g. k=Z).

Definition 1.

  1. An R-complex is a graded R-module with an endomorphism d of degree 1, such that the dd=0, Mv+1dv+1MvdvMv1dv1

  2. We use the notation M to mean the underlying graded R-module which is also identified with the complex with zero differential.

  3. A homomorphism of R-complexes MN is a graded homomorphism MN, while a chain map is a homomorphism of complexes, we allow chain map with non-zero degree |α|, that satisfy dNα=(1)|α|αdM

  4. A morphism MN is a chain map of degree 0.

Denote C(R) the category of R-complexes and morphisms of complexes, C(RModS)=C(RkSo) the category of RS-bicomplexes and morphisms of bicomplexes.

Given a complex, we look at certain subcomplexes that are actually graded modules

Definition 2. Z(M):Zv(M)=kerdv B(M):Bv(M)=Imdv+1 C(M):Cv(M)=cokerdv+1=Mv/Bv(M) H(M):Hv(M)=Zv(M)/Bv(M) Note that the differentials restricted to Z(M) and B(M) are automatically zero.

Remarks 1. A morphism at complexes MN restricts to Z(M)Z(N) and B(M)B(N), and induces a morphism H(M)H(N).

Definition 3. An R-complex M is called acyclic if H(M)=0.

Definition 4. A chain map MαN is called null-homotopic if there exists a degree |α|+1 chain map h:MN such that α=dNh+(1)|α|hdM.

Definition 5. Two chain maps MαN and NβM are called homotopic if αβ is null-homotopic.

Proposition 1. α,β:MN are homotopic H(α)=H(β).

The total Hom complex

Given R-complexes M and N, the total hom HomR(M,N)=HomR(M,N) i.e. HomR(M,N)v=iZHomR(Mi,Ni+v) with differential given by dHom(M,N)α=dNα(1)|α|αdM

Proposition 2. Let M, N be R-complexes, Z(Hom(M,N))={chain maps MN} B(Hom(M,N))={null-homotopic chain maps MN} Z0(Hom(M,N))=HomC(R)(M,N)

Definition 6. Given two morphisms α:MM,β:NN, there is a functor map HomR(M,N)Hom(α,β)HomR(M,N) mapping θ(1)|α|(|β|+|θ|)βθα. We denote Hom(M,β)=Hom(idM,β) and Hom(α,N)=Hom(α,idN).

Theorem 1. Hom is a functor C(R)op×C(R)C(k). If we have more structures on the complexes, we can have for example C(RQop)×C(RSop)C(QSop)

Definition 7 (Shift). Let M be an R-complex, define the suspension (ΣsM)v=Mvs,(ΣsM)v=(1)svsM.

Proposition 3. Hom(ΣsM,N)=ΣsHom(M,N) Hom(M,ΣsN)=ΣsHom(M,N)

Total Tensor Product

Given Ro-complex M and R-complex N, the total tensor product is (MRN)=MkN where in degree v, (MRN)v=iZMiRNvi with differential given by dMRN(mn)=dM(m)n+(1)|m|mdN(n).

Definition 8. Given R-complex morphism α:MM and β:NN, there is a morphism αβ:MRNMRN,mn(1)|β||m|α(m)β(n).

Theorem 2. The tensor product is a functor R:C(Ro)×C(R)C(k). if we have more structures on the complexes, we can have for example R:C(QModRo)×C(RModSo)C(QModSo).

Its behavior with shifting is

Proposition 4. Σs(MRN)=ΣsMRN=MRΣsN.

Definition 9. Let M be an R-complex, define supM=sup{iZ:HiM0} infM=inf{iZ:HiM0} define ampM=supMinfM

Definition 10 (Hard Truncations). (Mn)v:={Mvvn0v>n(dMn)v={dvMvn0v>n

(Mn)v:={Mvvn0v<n(dMn)v={dvMvn0v<n

Definition 11 (Soft Truncations). (Mn)v:={Mvv>nZv(M)v=n0v<n

(Mn)v:={0v>nCv(M)v=nMvv<n

Special Morphisms and Natural transformations

Mapping Cone

A mapping cone of a morphism α:MN in C(R) is a complex C(α) defined as C(α)=NΣM with differentials given by dC(α)=(dNαv10dv1M).

Theorem 3. The sequence 0NC(α)ΣM0 is split in C(R) iff α is null-homotopic.

This is a useful tool and a key ingredient in the triangulated structures on the homotopy category K(R).

Special Morphisms

Definition 12. A morphism α:MN in C(R) is a quasi-isomorphism if H(α) is an isomorphism.

Theorem 4. α:MN is quasi-isomorphism C(α) is acyclic.

Example 1. If nsupM, then MMn is a quasi-isomorphism, similarly if ninfM, then MnM is a quasi-isomorphism.

Remarks 2.

  1. If H(M)H(N), there need not exist a quasi-isomorphism MN or NM o.o! In fact we can have HomC(R)(M,N)=0=HomC(R)(N,M).

  2. If there exists quasi-isomorphism MN, there need not exist a quasi-isomorphism in the reversed direction.

Homotopy Equivalences

Definition 13.

  1. A morphism α:MNC(R) is called a homotopy equivalence if there exists a morphism in the reversed direction β:NM such that βαidM and αβidN. Any such beta is called a homotopy inverse of α.

  2. MC(R) is called contractible if idM0, or iff 0N is a homotopy equivalence.

Theorem 5. α:MN is a homotopy equivalence C(α) is contractible.

Remarks 3. If α is a homotopy equivalence, then α is also a quasi-isomorphism. If a complex M is contractible, then M is acyclic (exact).

Example 2. A short exact sequence η:0MMM0 of R-modules can be viewed as a complex ηC(R) which is acyclic. In this way, η is contractible iff η is split exact.

Remarks 4. The homotopy category K(R)=C(R)/ is the category whose objects are the complexes and morphisms are homotopy classes of morphisms in C(R). It could also be defined as a localization with respect to homotopy equivalences, C(R)[1].

Theorem 6. For a functor F:C(R)C(S), the following are equivalent

  1. If αβ, then F(α)F(β).

  2. If α is a homotopy equivalence, then F(α) is a homotopy equivalence.

In this case we say F preserves homotopy.

Example 3. MC(R), we have HomR(M,) and MR preserve homotopy.

Special Natural Transformations

There are some standard isomorphisms and evaluation morphisms.

Definition 14.

  1. Commutativity MRNNRopM,mn(1)|m||n|nm

  2. Associativity (MRN)SPMR(NSP),(mn)pm(np)

  3. Swap HomR(M,HomSop(N,X))SMXNHomSop(N,HomR(M,X)) where there is a sign change (1)|m||n|.

  4. Adjunction HomR(XSN,M)HomS(N,HomR(X,M)) where there is a sign change (1)|x||n|.

Evaluation Morphisms

MδXMHomSop(HomR(M,X),X),δXM(m)(ψ)=(1)|m|ψ(m)

Remarks 5. A derived version of bi-duality is important in at least two cases

  1. R=S, X=RDR, dualizing complex, we have Grothendieck duality.

  2. R=S, X=RRR, derived reflexive complexes.

Homothety evaluation morphism

Definition 15. Homothety evaluation morphism RHomSop(X,X)

Tensor Evaluation

Exercise.

Homomorphism Evaluation

Homomorphism evaluation NSHomR(X,M)ηMXNHomR(HomSop(N,X),M) nψ(φ(1)|n|(|ψ|+|φ|)ψ(φ(n)) There are some conditions under which ηMXN is an isomorphism. For example if N is a bounded complex of finitely presented Sop-modules and M is a complex of injective R-modules.

Sample Application of Homomorphism Evaluation

Recall that a projective module is flat.

Theorem 7. If F is a finitely presented flat R-module, then F is projective.

This is exercise 1.4.17.

Proof. Let MαN be a surjection, we need to show that HomR(F,α) is surjective. As Q/Z is faithfully injective (divisible) abelian group, we have HomR(F,α) is surjective HomZ(HomR(F,α),Q/Z) is injective. By hom evaluation and commutativity, the latter is HomZ(α,Q/Z)F) ◻

Homological Algebra

rendering math failed o.o

There is a fact, an R-complex is projective iff it is a contractible complex of projective R-modules.

Definition 16. An R-complex P is semi-projective (dg-projective) if the total Hom functor HomR(P,) preserves surjective quasi-isomorphisms.

We are not insisting on preserving all surjective morphisms, we only need it to send surjective q-isoms to surjective q-isoms.

The two main examples are

Example 4. A contractible complex of projective modules is semi-projective.

Example 5. A bounded below complex of projective R-modules PvPv1Pr0 is semi-projective.

There are many ways to characterize these complexes,

Theorem 8. For an R-complex P, the following are equivalent

  1. P is semi-projective.

  2. Hom(P,) is exact and preserve q-isoms

  3. For any chain map PN and surjective quasi-isomorphism MN, we can lift the map to PM.

    rendering math failed o.o

  4. Any exact sequence of complexes 0MMP0 where M is acyclic, splits.

  5. P is a complex of projective modules and the functor Hom(P,) preserves acyclic complexes.

Proposition 5. The semi-projectivity satisfy two out of three property. If in an exact sequence 0PPP0 of complexes of projective modules, two out of the three complexes are semi-projective, then the third is also semi-projective.

Definition 17. An R-complex L is semi-free if it is graded free on a basis E of homogeneous elements with E=n0En such that E0Z(L),LEnRi=0n1Ei

Example 6. A bounded below complex of free R-modules is semi-free. LvLv1L00 with Ev a basis for Lv.

Here we see E0 are cycles and indeed LEnRi=0n1Ei.

Proposition 6. A semi-free R-complex is semi-projective.

Definition 18. A semi-projective resolution of an R-complex M is a q-isom from semi-projective complex P to M, PM.

Sketch of the Construction

Here is a sketch of the construction, given an R-complex M, then we take Z0 to be a set of homogeneous generators of H(M), take L0 to be a graded free R-module that surjects on a basis E0 that surjects onto Z0, π0:L0Z(M), L0=0.

Do the same to H(kerπ0), adjust ,π accordingly.

Theorem 9. Let M be an R-complex, there exists a semi-projective resolution π:PM, moreover

  1. One can choose π surjective and P with Pv=0 for v<infM.

  2. One can choose P with Pv=0 for v<infM.

  3. If R is left noetherian and H(M) is bounded below and degree-wise finitely generated, then P can be chosen degree-wise finitely generated with Pv=0 for v<infM.

Definition 19. An R-complex I is semi-injective if HomR(,I) is exact and map surjective quasi-isomorphisms to injective quasi-isomorphisms.

Example 7. A contractible complex of injective R-modules is semi-injective.

Example 8. Bounded above complex of injective R-modules is semi-injective.

Proposition 7. If P is semi-projective then for a faithfully injective k-module E, Homk(P,E) is semi-injective.

Definition 20. A semi-injective resolution of an R-complex M is a q-isom from M to a semi-injective complex I, MqI.

Theorem 10. For every R-complex M, there exists a semi-injective resolution MqI. Moreover

  1. One can choose the map MI to be injective and I with Iv=0 for v>supM.

  2. One can choose I with Iv=0 for v>supM.

Definition 21. An R-complex F is semi-flat if RF preserves injective q-isoms.

Remarks 6. This is equivalent to asking that it preserves injective q-isoms and also preserve exactness.

Example 9. A contractible complex of flat R-modules is semi-flat.

Example 10. A bounded below complex of flat R-modules is semi-flat.

There are some similar characterizations.

Proposition 8. For an R-complex F, the following are equivalent

  1. F is semi-flat.

  2. RF is exact and preserves q-isoms.

  3. The character complex Hom(F,E) is semi-injective, where E is a faithfully injective k-module.

  4. F is a complex of exact modules and the functor RF preserves acyclic complexes.

Corollary 1. A semi-projective complex is semi-flat.

Homotopy Categories

Definition 22. The homotopy category K(R) has the same object as the category of complexes C(R), and the morphisms are homotopy classes of morphisms in C(R), i.e. K(R)(M,N)=Z0(Hom(M,N))/B0(Hom(M,N))=H0(Hom(M,N))

There is a canonical quotient functor QR:C(R)K(R) which maps MM and α[α]. The connection

Proposition 9. Let α be a morphism in C(R),

  1. [α] is zero iff α is null-homotopic.

  2. [α] is an isomorphism iff α is a homotopy equivalence.

Proposition 10. A complex MC(R) is zero in K(R) iff M is contractible.

Theorem 11. The homotopy category K(R) with Σ is a triangulated category with the distinguished triangles being those isomorphic to MN(1,0)tCone(α)(0,1ΣM)ΣM

The homotopy category has a universal property whose special case is useful.

Theorem 12. Let F:C(R)C(S) be a functor that preserves homotopy, then there is a unique functor K(F):K(R)K(S) that makes the following diagram commute rendering math failed o.o

There is an important consequence of splitting,

Proposition 11. Let P be semi-projective, β:MP a q-isom. There exists a γ:PM with 1pβγ.

Corollary 2. A q-isom of semi-projective complexes is a homotopy equivalence.

The Derived Category and Derived Functors

The derived category D(R) is the localization of the homotopy category K(R) with respect to the class of quasi-isomorphisms. It is the category whose objects are the complexes and morphisms are homotopy classes of morphisms in C(R).

  1. There exists a universal functor VR:K(R)D(R),αα11 which maps q-isoms to isoms.

  2. The universality of the functor gives, for every functor F:K(R)U that maps q-isoms to isoms, there is a unique functor F that makes the following diagram commute rendering math failed o.o

Moreover,

  1. D(R) and VR are triangulated.

  2. If U and F are triangulated, then so is F.

Construction of Derived Category (Sketch)

The objects of D(R) are ObjD(R)=ObjK(R) HomD(R)(M,N)={Fractions αφ1} where we invert quasi-isomorphisms, the morphisms (right fractions here) look like the diagrams MφUαN one can also use left fractions.

Some creepy notations used in the book:

  1. homotopy equivalence relation in C(R),

  2. quasi-isomorphism relation in C(R),

  3. isomorphism relation in C(R).

  4. homotopy equivalence relation in C(R), isoms in K(R).

Fact, HomD(R)(M,N)=HomK(R)(P,N) where PqM is a semi-projective resolution of M. This tells us that the hom set is always a set.

Composition in D(R)

We define the composition rule as follows HomD(R)(M,N)×HomD(R)(L,M)HomD(R)(L,N) as sending (αφ1,βψ1)(αγ)(ψχ)1

Triangulated Structure on D(R)

Definition 23. The distinguished triangles in D(R) are defined as the set of isomorphism closure in D(R) of VR(distinguished triangles in K(R)).

The semi-projective resolution functor

Definition 24.

  1. Every R-complex M has a semi-projective resolution πM:P(M)qM.

  2. The lifting properties of semi-projective complexes make P() into an endofunctor of the homotopy category K(R), and π:P()idK(R) is a natural transformation.

Derived Functors

Let F:K(R)K(S) be a functor, we can derive a functor by the universal property.

Definition 25. The total left derived functor LF is the unique functor that makes the diagram commute

rendering math failed o.o

Similarly we may define the right derived functor RF as RF(M)=FI(M)

Remarks 7. P:K(R)K(R) maps q-isoms to isoms.

rendering math failed o.o

where if α is a q-isom, this will force P(α) to be a q-isom.

Abstract Definition to Derived Functors

(LF)VR=VSFPλ=VSFπVSF (LF,λ) is object of a category LF={(F,λ):F:D(R)D(S),λ:FVRVSF} where the morphisms are natural transformations.

Theorem 13. (LF,λ) is the terminal object in LF.

Remarks 8.

  1. One can also derive contravariant functors by switching the roles of P and I.

  2. One can also derive functors in multiple variables.

Now we turn to the most important example, tensor and hom.

Derived Tensor and Hom

Recall R is a k-algebra.

Definition 26.

  1. The Hom functor HomR(,):C(R)op×C(R)C(k) preserves homotopy and induces a functor HomR(,):K(R)op×K(R)K(k) We can right derive this functor to get the derived Hom functor RHomR(,):D(R)op×D(R)D(k) To compute its value RHomR(M,N)=HomR(P(M),I(N))HomR(P(M),N)HomR(M,I(N)).

  2. Similarly we can derive the total tensor functor R:K(R)×K(R)K(k) we write its derived version as RL:D(R)×D(R)D(k) MRLN=P(M)RP(N)P(M)RNMRP(N)

Remarks 9. In fact it suffices to let P(M) or P(N) to be a semi-flat resolution.

Definition 27. Let M and N be R-complexes, ExtRi(M,N):=Hi(RHomR(N,M)) note that we used homology notation, in cohomology notation it will be denoted as Hi(RHomR(N,M)) instead. Similarly ToriR(M,N):=Hi(MRLN)

Standard Isomorphisms in D(R)

Recall that we have commutativity, associativity, swap, and adjunction isomorphisms in C(R).

Proposition 12. These natural isomorphisms exists in derived versions in D(R).

Derived Associativity

The ordinary associativity isomorphism is (MRX)SNMR(XSN) is a natural isomorphism of functors, and it lives naturally in K(R), C(Ro)×C(R-So)×C(S)C(k)K(Ro)×K(R-So)×K(S)K(k)

Now we look at rendering math failed o.o

This is a natural isom of functors D(Ro)×D(R-So)×D(S)D(k)

Boundedness and Finiteness Conditions

We want to know where do we have homology in the RHom and L functors. There are general lowerbounds for non-vanishing of Ext and Tor,

  1. For the derived Hom functor, we have supRHomR(M,N)=inf{i:ExtRi(M,N)0}infMsupN.

  2. For the derived tensor product, we have inf(MRLN)=inf{i:Tori(M,N)0}infM+infN.

Remarks 10. Upper bounds on non-vanishing of Ext and Tor are closely related to homological dimensions of complexes, which will be discussed in later sections.

RHomR(,):D(R)op×D(R)D(k) RL:D(Ro)×D(R)D(k)

In some cases one even has f attached to Df to each of the D in the above, which means all the homology module are finitely generated.

Homological Dimensions

Remarks 11. In order to avoid the confusion caused by using sup and inf in homology and cohomology, I will introduce the following notations, the left bound, lbM={supMfor homology indexinfMfor cohomology index the right bound, rbM={infMfor homology indexsupMfor cohomology index Note that their definition varies depending on what index convention you use. And we have the co-left bound, co-right bounds colbM=lbM,corbM=rbM.

Definition 28. M an R-complex, define pdRM=inf{i:There is a semi-projective res PM with Pv=0 for v>i}=infPMsupP=infPMlbP, which lies in Z{±}.

Remarks 12.

  1. M can be acyclic (exact) complex 0M, pdRM=

  2. If M does not have a semi-projective P with Pv=0 for v0, then the set is empty then by definition pdRM=.

  3. This definition is sensitive on shift, i.e. pdRΣiM=pdRM+i pdRMlbM note that Σ is always left shift whatever index convention you use.

  4. If P is a semi-projective complex, then there is a quasi-isom PqM there exists some isom PM in D(R). so pdRM only depends on the isomorphism. Class of M in D(R).

  5. If M is an R-module, viewed as a complex 0M0, then pdRM is the usual projective dimension of a module.

Remarks 13. It may be tempting to define the projective dimension as the projective dimension of the complex MC(R) viewed as an element in the abelian category C(R), since the category C(R) is abelian with enough projective. But R is not a projective object in the abelian category of complexes as it is not acyclic, in fact if we use this as the definition, we have pdRR=. And it should be warned that this definition is not what we are doing here.

Example 11. M be an R-module with projective resolution. P1P0M0 Let xR be a central element, define K(x,M) to be the complex 0MxM0 then pdRK(x,M)=1.

Here we have a complex rendering math failed o.o

  1. x=1, K(x,M)=0MM00D(R)

  2. x=0 K(x,M)=MΣM,pdK(x,M)=pdM+1

  3. R=Z, M=QZ, x=2 K(x,M)Z/2ZD(Z) pdZK(x,M)=1=pdZM

Theorem 14. For MD(R) and nZ, the following are equivalent

  1. pdRMn

  2. infRHomR(M,N)ninfN for all R-complex N

  3. nsupM and ExtRn+1(M,N)=0 for all N

  4. For some (equivalently every) semi-projective resolution PM, the R-module Cn(P)=cokern+1P is projective.

Furthermore, pdRM=sup{infRHomR(M,N):module N}=sup{i:ExtRi(M,N)0, module N}

Remarks 14. M,ND(R), we have an inequality infMsupNsupRHomR(M,N)=inf{i:Exti(M,N)0}infRHomR(M,N)=sup{i:ExtRi(M,N)0}pdRMinfN.

Theorem 15. Let R be left Noetherian and MD(R), pdRM=sup{infRHomR(M,R/I):IR left ideal} Furthermore, if pdRM<, then pdRM=infRHomR(M,R)

Similarly we can define injective dimension idRM=inf{i:There is a semi-injective res MI with Iv=0 for v>i}=infMI(infI)=infMIcorbI

Minimal semi-injective resolution

Theorem 16. For a semi-injective R-complex I, the following are equivalent

  1. Every q-isom II is an isom in C(R).

  2. Every q-isom IM has a left inverse in C(R).

  3. The only acyclic sub-complex AI is A=0.

  4. For each n, Zn(I)=kerdnI is an essential subset of In.

If any of the above holds, then I is a minimal semi-injective R-complex. A minimal semi-injective resolution of M is a q-isom MI with I minimal semi-injective.

If you have a minimal semi-injective resolution, you can read off the injective dimension.

Theorem 17. If MI is a minimal semi-injective resolution, then the injective dimension idRM=infI=corbI

Example 12. For M=ZC(Z), I=0QQ/Z0 I is a minimal semi-injective resolution of M, idZZ=1

Flat Dimension

Definition 29. The flat dimension of an R-complex M is

  1. fdRM is defined using semi-flat resolution of M.

  2. fdRM controls the homological supremum of ()RM.

Theorem 18. MD(R), we always have fdRMpdRM and equality holds if R is left Noetherian and MD(R).

Flat injective duality

MD(R), E faithfully injective k-module, now HomR(M,E)D(Ro).

  1. Injective dimension idRoHomk(M,E)=fdRM.

  2. fdRoHomk(M,E)=idRM provided that R is left Noetherian and MD(R).

Evaluation Morphisms In Derived Category

Recall that there were the biduality, homothety, tensor evaluation, homomorphism evaluation morphisms. The point is that, these natural morphisms exists in derived versions.

Derived Tensor Evaluation

let MD(R), XD(R-So), ND(S) RHomR(M,X)SLNθMXNRHomR(M,XSLN)

Theorem 19. The crazy thing is that, θMXN is usually an isomorphism o.o! For example for if R is left Noetherian and one of the following holds

  1. MDf(R) with pdRM<, ( M is a perfect R-complex ).

  2. MDf(R),XD(R-So),fdSN<.

Derived Homomorphism Evaluation

NSRHomR(X,M)ηMXNRHomR(RHomSo(N,X),M)

Theorem 20. ηMXN is an isom in D(k) if for example S is right Noetherian and one of the following holds

  1. NDf(So),pdSoN< ( N is a perfect So-complex ).

  2. ND(So),XD(R-So),idRM<.

Dualizing Complexes and Iwanaga-Gorenstein Categories

Derived biduality

Here MD(R),XD(R-So) δXM:MRHomR(RHomR(M,X),X)

To simplify things, we make the blanket assumption in this section: R noetherian, and projective as a k-algebra (module).

Theorem 21. The biduality morphism δRM:MRHomRo(RHomR(M,R),R) is an isomorphism in D(R) if MDf(R) and pdRM<.

Definition 30. An R-Ro complex D is dualizing for R if

  1. H(D) is degree-wise finitely generated over R and Ro.

  2. The injective dimensions idRD<, idRoD<. (because the boundedness

    • χ:RRHomR(D,D) isom in D(R-Ro).

    • χ:RRHomRo(D,D) isom in D(R-Ro).

Example 13. If k is a field, then k is dualizing for k. Let R be a finite dimensional k-algebra, D=Homk(R,k) injective R-module, look at HomR(,D)=HomR(,Homk(R,k))Homk(Rk,k)Homk(,k)

Rχ0HomR(D,D)=HomR(Homk(R,k),Homk(R,k))swapHomR(R,Homk(Homk(R,k),k))

rendering math failed o.o

Theorem 22. Let D be a dualizing complex for R. For MDf(R) δDM:MRHomRo(RHomR(M,D),D) is an isomorphism in D(R).

Theorem 23 (Grothendieck Duality). There is an adjoint equivalence of triangulated categories Df(Ro)RHomRo(,D)Df(R)op Df(Ro)RHomR(,D)Df(R)op unit and counit are biduality. It restricts to Df(Ro)Df(R)opDf(Ro)Df(R)op Df(Ro)Df(R)opIf(Ro)Pf(R)op

Definition 31. We introduce the following notations for the categories of complexes P(R)={MD(R):pdRM<} I(R)={MD(R):idRM<} F(R)={MD(R):fdRM<} If(R)=I(R)Df(R)

Definition 32. An R-complex M is derived reflexive if

  1. The complex itself has to be Df(R).

  2. RHomR(M,R)Df(Ro)

  3. Bi-duality δRM:MqRHomR(RHomR(M,R),R).

Denote M(R)={MD(R):M is derived reflexive}.

Theorem 24. There is an adjoint equivalence of triangulated categories, R(Ro)RHomRo(,R)R(R)op R(Ro)RHomR(,R)R(R)op where Pf(Ro) and Pf(R)op sits inside the arrows.

Definition 33. R is Iwanaga-Gorenstein if idRR< and idRoR<, i.e. injective dimension of R is finite as both a left and right module.

Theorem 25. The following are equivalent

  1. R is Iwanaga-Gorenstein.

  2. Every flat R and Ro-module has finite injective dimension.

  3. Every injective R and Ro-module has finite flat dimension.

Theorem 26. The following are equivalent

  1. R is Iwanaga-Gorenstein.

  2. R is a dualizing complex for R.

  3. R has a dualizing complex D with projective dimension finite over both R and Ro, i.e. pdRD< and pdRoD<.

Now let D be a dualizing complex, look at the adjoint functors D(R)DRLD(R),D(R)RHomR(D,)D(R) The unit αN is an isom if θDDN is an isom, and the co-unit βM is an isom if ηMDD is an isom.

We introduce the following notations

  • A^(R)={ND(R):αN is an isom}

  • A(R)={NA^(R):N,DRLND(R)}

  • B^(R)={MD(R):βM is an isom}

  • B(R)={MB^(R):M,RHomR(M,D)D(R)}

note that F(R)A(R), I(R)B(R).

Theorem 27 (Foxby-Sharp Equivalence). There is an adjoint equivalence of triangulated categories rendering math failed o.o

Proposition 13. Let D be a dualizing complex for R, then Af(R)=R(R)

Theorem 28. The following are equivalent

  1. R is Iwanaga-Gorenstein

  2. F(R)=I(R) and F(Ro)=I(Ro)

  3. R has a dualizing complex and R(R)=Df(R) and R(Ro)=Df(Ro)

  4. R has a dualizing complex and A(R)=D(R)

  5. R has a dualizing complex and B(R)=D(R)

Theorem 29. If D is a dualizing complex for R, and M an R-complex of finite flat dimension, then pdRMmax(supM,idRD+sup(DRLM))<

Homological Invariants in Commutative Algebra

Definition 34. For pSpecR, k(p)=Rp/pRp is the residue field of R at p. For MD(R), suppRM={pSpecR:k(p)RLM0D(R)}. cosuppRM={pSpecR:RHomR(k(p),M)0D(R)}.

Remarks 15. For an R-module M one has the classical support SuppRM={pSpecR:mp0} One always have suppRMSuppRM with equality if M is finitely generated.

Example 14.

  • The suppRk(p)={p}=cosuppRk(p).

  • The injective hall E=ER(R/p) of R/p, which is an injective module and R/p is an essential submodule of E, suppRER(R/p)={p}cosuppRER(R/p)={q:qp}

Theorem 30. For MD(R), one has cosuppRM=suppRM=M=0D(R) (i.e. M acyclic, exact).

Support Formula

Theorem 31. For M,ND(R), one has suppR(MRLN)=suppRMsuppRN. i.e. MRLN=0 iff suppRMsuppRN=.

Example 15.

  • R=Z, M=Z/2Z, N=Q. MRLN=Z/2ZZQ=0

  • SuppZ(0)=

  • SuppZ(Z/2Z)SuppZ(Q)={(2)}Spec(Z)={(2)}

Cosupport Formula

Theorem 32. For M,ND(R) one has cosuppRRHomR(M,N)=suppRMcosuppRN. In particular, RHomR(M,N)=0 in D(R) iff suppRMcosuppRN=.

Koszul Complex

xR be an element of a ring, the Koszul complex is K(x)=0RxR0 where since we are using homology indexing, the degrees are 1,0, for cohomology index, it should be 0,1 instead and the definitions related with it will need to be changed. For a sequence of elements x=(x1,,xn), the Koszul complex is K(x)=K(x1)RRK(xn)

Example 16. K(x1,x2) looks like 0R(x2,x1)tR2(x1,x2)R0

Depth and width

Let a=(x1,,xn) be an ideal of R, MD(R), the a-depth of M is defined as a-depthRM=n+colb(K(x)M)=colbHomR(K(x),M)=colbRHomR(R/a,M)=colbRΓaM

One always have an inequality a-depthRMcolbM and equality holds in some cases.

The a-width is defined as a-widthRM=rb(K(x)RM)=n+rbRHomR(K(x),M)=rb(R/aRLM). Similarly we have a-widthRMrbM and equality holds for example if aJ(R) and MDf(R). Thus for a finitely generated R-module M0, the a-width of M is 0.

Remarks 16. Say aJ(R), M0 is a finitely generated R-module, a-depthRM= the maximal length of an M-regular sequence contained in the ideal a, which is finite.

Example 17. Let RZ,M=Q,x0Z. K(x)ZQ=0QxQ0 which =0 in D(R). The (x)-depth of Q is .

Depth and flat dimension

Definition 35. For a local ring (R,m,k) and MD(R), we define depthRM=m-depthRM and width to be widthRM=m-widthRM.

How does depth and width interact with L and RHom?

  1. You can always compute depthRRHomR(M,N)=widthRM+depthRN

  2. depthR(MRLN)depthRM+depthRNdepthR provided that fdRM< and ND(R).

The Auslander-Buchsbaum Formula

Set (R,m,k) be a local ring, MDf(R). If pdRM<, then pdRM=sup(kRLM)=depthRdepthRM.

Proof. Set N=k in the depth-width formula for MRLN. There are versions of the equalities in the Auslander-Buchsbaum formula for complexes over non-local rings. ◻

Theorem 33. R is any commutative noetherian ring, MD(R) is an R-complex. fdRM=sup{sup(k(p)RLM):pSpecR} If fdRM<, then fdRM=sup{depthRpdepthRpMp:pSpec(R)}

Width and Injective Dimension

  1. You can always compute widthRRHomR(M,N)=widthRM+widthRN

  2. widthR(MRLN)depthRM+widthRNdepthR provided that idRM< and ND(R).

Theorem 34 (The Bass Formula). MDf(R), if idRM<, then idRM=infRHomR(k,M)=depthRinfM. This is Chouinand Formula for injective dimension.

Mathis Duality

Let (R,m,k) be local, E=ER(k) the injective hull of the R-module k, then

Theorem 35. Let (R,m,k) be local and complete, R^=R, for example R=k[[x1,,xn]]/I, then there are equivalences of triangulated categories rendering math failed o.o Here Df/art/l(R) is the full subcategory of D(R) whose objects M satisfy that each homology module Hi(M) is finitely generated / artinian / has finite length. Moreover, if D is a normalized dualizing complex for R (i.e. supD=dimR), then Grothendieck duality Df(R)RHomR(,D)Df(R)op restricts to Mathis duality, i.e. RHomR(M,D)RHomR(M,ER(k)) for MDf(R).

The Derived Category of A Commutative Noetherian Ring

Let a be an ideal and one can look at an obvious sequence of quotients R/a3R/a2R/a

Definition 36. The a-torsion functor Γa=colimu1HomR(R/au,):C(R)C(R) it finds all things that gets killed by some powers of a.

Derived a-torsion RΓa:D(R)D(R)

Definition 37. Han(M)=Hn(RΓa(M))

Theorem 36. Let a be an ideal and MDf(R). One can compute RΓa(M) as RΓa(M)=RHomR(RHomR(M,D),RΓa(D)), where D is a dualizing complex for R. Compare biduality MRHomR(RHom(M,D),D) which is always an isomorphism.

In local algebra, this has a particularly nice interpretation.

Definition 38. The Krull dimension of R is the dimension of Spec(R) as a poset.

Definition 39. A dualizing complex D for R is normalized if the supremum of D is equal to the dimension of R, supD=dimR.

Theorem 37 (Local Duality Theorem). If (R,m,k) is local, D is a normalized dualizing complex for R, then for MDf(R), RΓm(M)=HomR(RHomR(M,D),E(k)) Hmn(M)=HomR(ExtRn(M,D),E(k))

D0dimR/p=dimRE(R/p)E(k)0

Proposition 14. aR an ideal and MD(R). a-depthRM=supRΓa(M)=inf{n:Han(M)0}

Proposition 15. MRΓa(M)suppRMV(a)={pSpecR:am}

Theorem 38. They are commutative up to natural isomorphisms, diagrams of equivalences of triangulated categories rendering math failed o.o

rendering math failed o.o

Definition 40. R is Gorenstein if Rp is Iwanaga-Gorenstein i.e. idRpRp< for each p.

Theorem 39. The following are equivalent

  1. R is Iwanaga-Gorenstein,

  2. R is Gorenstein with dimR<,

  3. R has finite Krull-dimension and R(R)=Df(R).

  4. R has a dualizing complex and the Auslander category A(R)=D(R).

  5. R has a dualizing complex and B(R)=D(R).

  6. R has a dualizing complex and A^(R)=D(R).

  7. R has a dualizing complex and B^(R)=D(R).

Theorem 40. A complex MDf(R) is derived reflexive iff MRHomR(RHomR(M,R),R).

Theorem 41. The following are equivalent

  1. R is Gorenstein,

  2. For every acyclic complex P of projective R-modules also Hom(P,L) is acyclic for every projective R-module L.

  3. For every acyclic complex I of injective R-modules, also HomR(E,I), is acyclic for every injective E.

  4. For every acyclic complex F of flat R-modules, also ERF is acyclic for every injective R-module E.

Theorem 42. For a local ring (R,m,k) the following are equivalent

  1. R is regular,

  2. pdRk<,

  3. Pf(R)=Df(R),

Definition 41. R is regular if Rp is regular for every prime p.

Theorem 43. The following are equivalent

  1. R is regular,

  2. Every acyclic complex of projective modules is contractible,

  3. Every acyclic complex of injective modules is contractible,

  4. Every complex of injective R-modules is semi-injective,

  5. Every complex of projective R-modules is semi-projective,

  6. Every complex of finitely generated projective R-modules is semi-projective.