Homological algebra

Homological algebra studies chain complexes, homology, cohomology, and the derived constructions that measure failure of exactness.

The derived category of sheaves is formed by localizing cochain complexes of sheaves at quasi-isomorphisms, equivalently localizing their homotopy category. The bounded, bounded below and bounded above variants are denoted $D^b(X)$, $D^{+}(X)$ and $D^{-}(X)$. Cohomology sheaves are invariant under quasi-isomorphism and therefore define functors on these categories. Constructible bounded complexes form a natural subcategory on which Verdier duality reverses arrows and, under the usual finiteness hypotheses, is an involution.

A ghost is a morphism inducing zero on every cohomology sheaf, but it need not itself be zero. For the equator $j:S^1\hookrightarrow S^2$ and complement $i:U\hookrightarrow S^2$, the connecting morphism $j_{\ast }{\mathbb{Q}}_{S^1}\to i_{!}{\mathbb{Q}}_U[1]$ is a ghost. Its source and target have cohomology sheaves in different degrees, but its induced map $H^1(S^1)\to H_c^2(U)$ is nonzero, proving that the derived morphism is nonzero.

On bounded constructible complexes over locally compact stratified spaces, Verdier duality is $D_{X}E=R\mathcal{H}om(E,{\omega }_X^{\bullet })$, where ${\omega }_X^{\bullet }$ is the dualizing complex. For an oriented real $N$-manifold, $D_X{\mathbb{Q}}_X\cong {\mathbb{Q}}_X[N]$. Proper direct image commutes with this duality, and $i_x^{!}{D}_{X}E\cong D_{\{x\}}{i}_x^{\ast }E$. These properties exchange stalk support conditions with costalk conditions.

A bounded complex is constructible for a given stratification when each of its cohomology sheaves restricts to a finite-rank local system on every stratum. Constructibility concerns cohomology sheaves, not an arbitrary choice of resolution.

The category $D^{+}(X)$ is obtained from bounded below cochain complexes of sheaves by passing to cochain-homotopy classes of maps and then inverting quasi-isomorphisms. It can equally be modeled by injective resolutions. The cohomology-sheaf functors descend to this localization, and an object is zero exactly when all its cohomology sheaves vanish.

An exact, or distinguished, triangle is one isomorphic to $E\stackrel{f}{\to }F\to \mathrm{Cone}(f)\to E[1]$. For a degreewise short exact sequence $0\to E\to F\to G\to 0$, the map $\mathrm{Cone}(f)\to G$ taking $(v,e)$ to the image of $v$ is a quasi-isomorphism: its kernel is the contractible cone of the identity on $E$. Consequently the short exact sequence defines such a triangle.

For an additive right-exact functor on module categories, apply it to a projective resolution and take homology. Lifts of module maps exist by projectivity; any two lifts are chain homotopic, giving independence and functoriality. The zeroth derived functor is naturally $F$, positive degrees vanish on projective modules, and the horseshoe lemma produces a long exact homology sequence from each short exact sequence. A Tor functor is the left derived functor of a tensor product of modules functor.

For a right-exact additive functor out of an abelian category with enough projectives, apply the functor to a projective resolution and take homology objects. The zeroth derived functor is $F$, positive derived functors vanish on projective objects, and the degreewise split horseshoe lemma produces a long exact sequence from every short exact sequence in an abelian category. The target can be any abelian category, including an opposite category of one.

An exact augmented cochain complex with injective module terms. Module categories have enough injectives: coinduction $J={\mathrm{Hom}}_{\mathbb{Z}}(A,\mathbb{Q}/\mathbb{Z})$ is injective by its character adjunction, and the maps $m\mapsto (a\mapsto \varphi (am))$, over all additive characters $\varphi$ of $M$, embed $M$ into a product of copies of $J$. Products of injectives are injective by coordinatewise extension. Repeat on cokernels to obtain a resolution.

For an additive covariant left-exact functor on module categories, choose an injective resolution $M\to I^{\bullet }$ and set $R^{i}F(M)=H^i(F(I^{\bullet }))$. For a contravariant functor, use a projective resolution instead, as in contravariant right derived functor. Resolution comparison and homotopies give independence and functoriality; short exact sequences give long exact derived-functor sequences. The zeroth functor is naturally $F$.

For an additive contravariant left-exact functor, apply it to a projective resolution of $M$. Reversal of the arrows gives a cochain complex. Comparison maps lift using projectivity, and any two lifts are chain homotopic, so its cohomology is independent of the resolution. The functors are contravariant, vanish on projective modules in positive degrees, and send $0\to M^{\prime }\to M\to M^{\prime \prime }\to 0$ to $0\to F(M^{\prime \prime })\to F(M)\to F(M^{\prime })\to R^{1}F(M^{\prime \prime })\to \cdots$. Degreewise split resolutions from the horseshoe lemma yield this sequence.

A succession of bigraded pages with $E_{r+1}=H(E_r,d_r)$. In the cohomological convention, the differential has bidegree $(r,1-r)$. A filtered complex produces such successive approximations to its cohomology.

For a bounded complex $P$ of projective right modules and a left module $M$ of finite projective dimension, a finite projective resolution of $M$ yields a bounded double cochain complex. Its two spectral sequences give the displayed abutment. Over a noncommutative ring, the handedness of the two tensor factors is essential.

If only the columns $p=-1,0$ are nonzero, a cohomological differential of bidegree $(r,1-r)$ has no possible nonzero source-target pair for $r\ge 2$. Projective boundaries in a complex of projectives give this condition because every cohomology module has projective dimension at most one.

The stable page identifies the associated graded module of a filtration of the abutment. For a finite filtration this is an exhaustive, separated finite extension description. It does not itself determine a canonical splitting. Unbounded filtrations require further convergence hypotheses.

A finite filtration of a cochain complex produces a convergent spectral sequence. The induced abutment filtration is the image of $H^n(F^{p}C)\to H^n(C)$. Finiteness in each degree suffices; a bounded range of cochain degrees is not required.

The Tor functor is the homology of a tensor product of modules with a free resolution: ${\mathrm{Tor}}_i^R(M,N)=H_i(F_{\bullet }{\otimes }_{R}N)$ for a free resolution $F_{\bullet }$ of $M$. It measures the failure of tensoring to preserve exact sequences. For a commutative ring, one may resolve either argument; the double complex of the two resolutions proves that both computations agree.

For a non-zero-divisor $a\in A$, resolve $A/(a)$ by $0\to A\stackrel{a}{\to }A\to A/(a)\to 0$. Tensoring with $M$ gives the kernel and cokernel in the displayed formula; all higher Tor functors vanish. Resolving either argument gives the same homology by the two filtrations of the double complex of the two free resolutions. For $M=A/(a)$, both degrees zero and one equal $A/(a)$.

Start a free resolution of $A/I$ with $P_0=A$ and a free module $P_1$ surjecting onto $I$. The next differential presents the relations in $I$. Right exactness of the tensor product of modules identifies $\mathrm{coker}(P_2{\otimes }_{A}M\to P_1{\otimes }_{A}M)$ with $I{\otimes }_{A}M$. Taking degree-one homology then gives the displayed kernel formula. Hence vanishing for every ideal is equivalent to the flatness criterion using ideals.

A cochain complex is a sequence of abelian groups or modules and maps $d^n:C^n\to C^{n+1}$ satisfying $d^{n+1}{d}^n=0$. Its cohomology is $H^n(C)=\ker d^n/\mathrm{im}{d}^{n-1}$.

A cochain map $f:E\to F$ consists of degreewise maps $f^i:E^i\to F^i$ satisfying $d_F^i{f}^i=f^{i+1}{d}_E^i$. This identity sends cocycles to cocycles and coboundaries to coboundaries, hence induces maps on cohomology. For complexes of sheaves the same identity holds as an identity of sheaf morphisms.

The shift $E[1]$ has $E[1{]}^i=E^{i+1}$ and differential $-d_E^{i+1}$. Thus $H^i(E[1])=H^{i+1}(E)$. In general the differential of $E[r]$ is $(-1{)}^r{d}_E$.

The submodule of cocycles in degree $q$. A projective boundary module $B^{q+1}$ makes $0\to Z^q\to P^q\to B^{q+1}\to 0$ split when $P^q$ is projective, so $Z^q$ is then also projective.

An element killed by the next differential of a cochain complex. Its class modulo coboundaries is a cohomology class.

The submodule of coboundaries in degree $q$. The cohomology group is the corresponding cycle module modulo the boundary module.

An element in the image of the preceding differential of a cochain complex. Every coboundary is a cocycle, since consecutive differentials compose to zero.

A cochain complex with a decreasing filtration by cochain subcomplexes. A filtration finite in each degree is bounded in the degreewise sense; uniform bounds $F^{a}C=C$ and $F^{b}C=0$ are sufficient. The associated spectral sequence starts with the graded quotient complexes.

A sequence of submodules preserved by the differential of a cochain complex. Its inclusion is a map of complexes.

A degreewise short exact sequence of cochain complexes induces a long exact sequence from a coefficient sequence in the coefficient case, and a long exact cohomology sequence generally. Its degree-one connecting homomorphism lifts a quotient cocycle, takes its coboundary, and identifies that result in the subcomplex.

A coboundary map is the degree-one differential $d^n:C^n\to C^{n+1}$ of a cochain complex.

The Hom functor sends two $R$-modules to the abelian group of $R$-linear maps between them. It is contravariant in its first argument and covariant in its second.

The Ext functors are the right derived functors of the Hom functor. They can be computed by applying ${\mathrm{Hom}}_R(-,N)$ to a projective resolution of $M$ and taking cohomology.

For a non-zero-divisor $a\in A$, the free resolution $0\to A\stackrel{a}{\to }A\to A/(a)\to 0$ gives degree-zero kernel and degree-one cokernel, and all higher Ext functors vanish. Directly in an injective resolution of the second variable, multiplication by $a$ is surjective in every term: the map $aA\to I$ taking $ax$ to $xm$ extends to $A\to I$. Thus $0\to {\mathrm{Hom}}_A(A/(a),I^{\bullet })\to I^{\bullet }\stackrel{a}{\to }{I}^{\bullet }\to 0$ is a short exact sequence of complexes and gives the same result. The displayed Hom functor is covariant in $M$; changing it to ${\mathrm{Hom}}_A(M,A/(a))$ changes the variable that is resolved.

A free resolution of $\mathbb{Q}$ has generators $a_n=1/n!$ and independent relations $a_n-(n+1)a_{n+1}$. Thus the Ext functor gives the cokernel of the map on integer sequence productsThe constant sequence one is not in its image. A preimage would require $u_1\equiv {\sum }_{j=1}^{N}j!(\mathrm{mod}(N+1)!)$ for every $N$. For $N\ge 3$, the sum is below $2N!$, tends to infinity, and its distance from $(N+1)!$ also tends to infinity. No fixed integer satisfies all these congruences. Hence ${\mathrm{Ext}}_{\mathbb{Z}}^1(\mathbb{Q},\mathbb{Z})\ne 0$.

Applying the contravariant Hom functor ${\mathrm{Hom}}_R(-,W)$ to $0\to I\to X\to C\to 0$ givesThe connecting map sends $I\to W$ to its pushout of a module extension. Its vanishing means that map extends to $X\to W$. For a hereditary ring, ${\mathrm{Ext}}^2(C,W)=0$, making the indicated restriction on first extension groups surjective. The covariant variable has its corresponding long exact sequence.

An extension group is a value of an Ext functor. In degree one it classifies short exact sequences $0\to N\to E\to M\to 0$ up to isomorphisms fixing the endpoints. The zero class is the split extension; its addition is the Baer sum. Higher degrees can be represented by longer exact extensions, or computed from a projective resolution.

To add two extensions of $M$ by $N$, take their direct sum, form a pullback in a category along the diagonal $M\to M\oplus M$, and form a pushout in a category along addition $N\oplus N\to N$. This produces another extension of $M$ by $N$ and defines addition in ${\mathrm{Ext}}_R^1(M,N)$. In a quiver representation, vertexwise splittings express the two extensions by off-diagonal arrow matrices, and the Baer sum adds those matrices modulo the extension complex of quiver representations coboundaries.

Over a Noetherian ring, a finitely generated module $N$ has a free resolution with every term finitely generated. Hom from a finite free module commutes with localization, and exactness of localization commutes with cohomology. Thus the displayed isomorphism holds for every $i\ge 0$ and arbitrary $M$. The finiteness condition is on the first argument; the second need not be finitely generated.

When $X$ acts as zero, resolve the first module with the Koszul resolution on $(p,X)$. Writing $N=\mathbb{Z}/q$, the resulting groups are $\ker (p:N\to N)$ in degree zero, $(N/pN)\oplus \ker p$ in degree one, and $N/pN$ in degree two, with zero higher groups. This works for composite as well as prime moduli.

For an ideal generated by a finite regular sequence in a commutative ring, the self-Ext functor algebra is exterior on the degree-one dual generators. Contraction chain maps on the Koszul resolution represent those generators, square to zero, and anticommute. The exterior square-zero relations also apply in characteristic two.

Partition the isomorphism classes of simple modules into two sets ${\mathcal{C}}_1,{\mathcal{C}}_2$, with ${\mathrm{Ext}}^1(S,T)={\mathrm{Ext}}^1(T,S)=0$ whenever $S,T$ lie in different sets. Every module $M$ of finite composition length then has a unique decomposition $M=U_1\oplus U_2$ into its largest submodules whose Jordan–Hölder factors belong to the respective sets.

For a simple $S\in {\mathcal{C}}_1$ and a finite-length $N$ with factors in ${\mathcal{C}}_2$, induction through the long exact sequence of the Ext functor gives ${\mathrm{Ext}}^1(S,N)=0$. Induct on the length of $M$: take $0\to N\to M\to S\to 0$, split $N=N_1\oplus N_2$, and suppose $S\in {\mathcal{C}}_1$. The quotient extension $0\to N_2\to M/N_1\to S\to 0$ splits. The inverse image of its $S$-summand complements $N_2$ in $M$ and has only factors in ${\mathcal{C}}_1$. Finally, Hom between modules with factors in different sets is zero, since any nonzero image would have a factor in both sets. Projecting any proposed submodule to the opposite summand therefore proves maximality and uniqueness.

A projective resolution of an $R$-module $M$ is an exact sequencein which every $P_i$ is a projective module.

A finitely generated module over a Noetherian ring admits a possibly infinite free resolution whose every term is finitely generated. Surject a finite free module onto it; the kernel is finitely generated because a finite free module is Noetherian. Repeat this construction on the successive kernels. If the original module has finite projective dimension $n$, its $n$th syzygy is projective by dimension shifting of the Ext functor, giving a length-$n$ resolution with finite projective terms.

For a short exact sequence of modules with projective resolutions of its outer terms, there is a projective resolution of the middle term forming a degreewise split short exact sequence of resolutions. Lift the degree-zero projective cover of the quotient into the middle module and add the projective cover of the submodule; this surjects onto the middle module. Its kernel fits into the corresponding short exact sequence of kernels. Repeating this construction proves the claim. Additive contravariant functors preserve the degreewise split exact sequences with their direction reversed.

Set $A=kQ$, $P_i=A{e}_i$, $X_i=e_{i}X$. There is an exact sequencewhere $d(p\otimes x)=p\rho \otimes x-p\otimes f_{\rho }(x)$ in the source and target summands, and the augmentation is $p\otimes x\mapsto px$. Each $P_i$ is a summand of $A$, so the two terms are projective modules. This also works for infinite-dimensional modules.

Put $C^0={⨁}_i\mathrm{Hom}(X_i,Y_i)$ and $C^1={⨁}_{\rho :i\to j}\mathrm{Hom}(X_i,Y_j)$. The differential has components $g_{\rho }{h}_i-h_j{f}_{\rho }$. Its kernel is ${\mathrm{Hom}}_Q(X,Y)$ and its cokernel is ${\mathrm{Ext}}_Q^1(X,Y)$. In a vertexwise splitting of an extension, the off-diagonal arrow blocks change by these coboundaries when the splitting changes.

Restriction at one arrow induces a surjection ${\mathrm{Ext}}_Q^1(X,Y)\twoheadrightarrow {\mathrm{Hom}}_k(\ker f_{\rho },\mathrm{coker}{g}_{\rho })$. Coboundaries vanish on the kernel after quotienting by the image, while every map between these spaces can be lifted and extended. Nonzero kernel and cokernel therefore force a nonsplit extension. For $X=Y$, rigidity forces every arrow to have maximal rank.

A short exact sequence presenting a module as a quotient of a projective module. The kernel need not be projective. Continuing a projective presentation gives a projective resolution.

For a regular sequence $f_1,\dots ,f_r$ in a commutative ring $R$, its Koszul complex is the finite free projective resolution of $R/(f_1,\dots ,f_r)$ with terms ${\bigwedge }^j{R}^r$ and boundary map contracting by $(f_1,\dots ,f_r)$. Exactness follows by induction: append $f_r$ using the mapping cone for multiplication by $f_r$, which is injective on the preceding quotient. For a polynomial ring viewed as a bimodule, use the regular sequence $X_{i,\ell }-X_{i,r}$ in the enveloping algebra.

A free resolution is a projective resolution in which every resolving module is a free module.

Every finitely generated module with a grading over the polynomial ring $S=k[X_1,\dots ,X_n]$ has a finite graded free resolution of length at most $n$. The Koszul resolution of $k$ has length $n$, so ${\mathrm{Tor}}_i^S(M,k)=0$ for $i>n$. In a minimal graded free resolution, this Tor functor is $F_i/(X_1,\dots ,X_n)F_i$; the graded Nakayama lemma forces $F_i=0$ for $i>n$.

Over $S=k[X_1,\dots ,X_n]$ with its standard grading and $\mathfrak{m}=(X_1,\dots ,X_n)$, a graded free resolution is minimal when $d_i(F_i)\subseteq \mathfrak{m}{F}_{i-1}$ for every $i\ge 1$. It is constructed by choosing minimal homogeneous generators for each successive syzygy module, using the graded Nakayama lemma. After tensoring with $k=S/\mathfrak{m}$, all its differentials vanish, so $F_i/\mathfrak{m}{F}_i\cong {\mathrm{Tor}}_i^S(M,k)$; these dimensions and degrees are intrinsic to $M$.

Given a free resolution $\cdots \to F_1\to F_0\to M\to 0$, its first syzygy module is $\ker (F_0\to M)$, the module of relations among the chosen generators of $M$. Higher syzygy modules are successive kernels. They depend on the choice of resolution; a minimal graded free resolution gives canonical ranks and degrees of generators over a standard graded ring structure on a polynomial ring.

The bar resolution of an associative algebra $A$ over a field has degree-$n$ term $A^{\otimes (n+2)}$ as an $A^e$-module, with the outer factors carrying the left and right actions. Its boundary multiplies each consecutive pair with alternating signs, and its augmentation is multiplication $A\otimes A\to A$. It is exact: the $k$-linear contracting homotopy inserts $1$ at the left. Applying ${\mathrm{Hom}}_{A^e}(-,M)$ gives the Hochschild cochain complex of the bimodule $M$.