Let be a ring and let be the category of modules over . (One can take this to mean either left -modules or right -modules.) For a fixed -module , let for in . (Here is the abelian group of -linear maps from to ; this is an -module if is commutative.) This is a left exact functor from to the category of abelian groups, and so it has right derived functors. The Ext groups are the abelian groups defined by
For each integer i, Exti R(A, B) is the cohomology of this complex at position i. It is zero for i negative. For example, Ext0 R(A, B) is the kernel of the map HomR(A, I0) → HomR(A, I1), which is isomorphic to HomR(A, B).
An alternative definition uses the functor G(A)=HomR(A, B), for a fixed R-module B. This is a contravariant functor, which can be viewed as a left exact functor from the opposite category (R-Mod)op to Ab. The Ext groups are defined as the right derived functors RiG:
Then Exti R(A, B) is the cohomology of this complex at position i.
One may wonder why the choice of resolution has been left vague so far. In fact, Cartan and Eilenberg showed that these constructions are independent of the choice of projective or injective resolution, and that both constructions yield the same Ext groups.[2] Moreover, for a fixed ring R, Ext is a functor in each variable (contravariant in A, covariant in B).
For a commutative ring R and R-modules A and B, Exti R(A, B) is an R-module (using that HomR(A, B) is an R-module in this case). For a non-commutative ring R, Exti R(A, B) is only an abelian group, in general. If R is an algebra over a ringS (which means in particular that S is commutative), then Exti R(A, B) is at least an S-module.
Properties of Ext
Here are some of the basic properties and computations of Ext groups.[3]
Ext0 R(A, B) ≅ HomR(A, B) for any R-modules A and B.
Exti R(A, B) = 0 for all i > 0 if the R-module A is projective (for example, free) or if B is injective.
The converses also hold:
If Ext1 R(A, B) = 0 for all B, then A is projective (and hence Exti R(A, B) = 0 for all i > 0).
If Ext1 R(A, B) = 0 for all A, then B is injective (and hence Exti R(A, B) = 0 for all i > 0).
If R is a commutative ring and u in R is not a zero divisor, then
for any R-module B. Here B[u] denotes the u-torsion subgroup of B, {x ∈ B: ux = 0}. Taking R to be the ring of integers, this calculation can be used to compute for any finitely generated abelian groupA.
Generalizing the previous example, one can compute Ext groups when the first module is the quotient of a commutative ring by any regular sequence, using the Koszul complex.[5] For example, if R is the polynomial ringk[x1,...,xn] over a field k, then Ext* R(k,k) is the exterior algebraS over k on n generators in Ext1. Moreover, Ext* S(k,k) is the polynomial ring R; this is an example of Koszul duality.
By the general properties of derived functors, there are two basic exact sequences for Ext.[6] First, a short exact sequence 0 → K → L → M → 0 of R-modules induces a long exact sequence of the form
for any R-module A. Also, a short exact sequence 0 → K → L → M → 0 induces a long exact sequence of the form
for any R-module B.
Ext takes direct sums (possibly infinite) in the first variable and products in the second variable to products.[7] That is:
The Ext groups derive their name from their relation to extensions of modules. Given R-modules A and B, an extension of A by B is a short exact sequence of R-modules
Two extensions
are said to be equivalent (as extensions of A by B) if there is a commutative diagram:
Note that the Five lemma implies that the middle arrow is an isomorphism. An extension of A by B is called split if it is equivalent to the trivial extension
There is a one-to-one correspondence between equivalence classes of extensions of A by B and elements of Ext1 R(A, B).[9] The trivial extension corresponds to the zero element of Ext1 R(A, B).
The Baer sum of extensions
The Baer sum is an explicit description of the abelian group structure on Ext1 R(A, B), viewed as the set of equivalence classes of extensions of A by B.[10] Namely, given two extensions
Up to equivalence of extensions, the Baer sum is commutative and has the trivial extension as identity element. The negative of an extension 0 → B → E → A → 0 is the extension involving the same module E, but with the homomorphism B → E replaced by its negative.
Construction of Ext in abelian categories
Nobuo Yoneda defined the abelian groups Extn C(A, B) for objects A and B in any abelian categoryC; this agrees with the definition in terms of resolutions if C has enough projectives or enough injectives. First, Ext0 C(A,B) = HomC(A, B). Next, Ext1 C(A, B) is the set of equivalence classes of extensions of A by B, forming an abelian group under the Baer sum. Finally, the higher Ext groups Extn C(A, B) are defined as equivalence classes of n-extensions, which are exact sequences
under the equivalence relation generated by the relation that identifies two extensions
if there are maps for all m in {1, 2, ..., n} so that every resulting square commutes
that is, if there is a chain map which is the identity on A and B.
The Baer sum of two n-extensions as above is formed by letting be the pullback of and over A, and be the pushout of and under B.[11] Then the Baer sum of the extensions is
The derived category and the Yoneda product
An important point is that Ext groups in an abelian category C can be viewed as sets of morphisms in a category associated to C, the derived categoryD(C).[12] The objects of the derived category are complexes of objects in C. Specifically, one has
where an object of C is viewed as a complex concentrated in degree zero, and [i] means shifting a complex i steps to the left. From this interpretation, there is a bilinear map, sometimes called the Yoneda product:
which is simply the composition of morphisms in the derived category.
The Yoneda product can also be described in more elementary terms. For i = j = 0, the product is the composition of maps in the category C. In general, the product can be defined by splicing together two Yoneda extensions.
Alternatively, the Yoneda product can be defined in terms of resolutions. (This is close to the definition of the derived category.) For example, let R be a ring, with R-modules A, B, C, and let P, Q, and T be projective resolutions of A, B, C. Then Exti R(A,B) can be identified with the group of chain homotopy classes of chain maps P → Q[i]. The Yoneda product is given by composing chain maps:
By any of these interpretations, the Yoneda product is associative. As a result, is a graded ring, for any R-module A. For example, this gives the ring structure on group cohomology since this can be viewed as . Also by associativity of the Yoneda product: for any R-modules A and B, is a module over .
For a commutative Noetherian local ringR with residue field k, is the universal enveloping algebra of a graded Lie algebra π*(R) over k, known as the homotopy Lie algebra of R. (To be precise, when k has characteristic 2, π*(R) has to be viewed as an "adjusted Lie algebra".[13]) There is a natural homomorphism of graded Lie algebras from the André–Quillen cohomologyD*(k/R,k) to π*(R), which is an isomorphism if k has characteristic zero.[14]