Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

In group cohomology, one defines $H^i(G;A)$ for $G$ a group and $A$ a $G$-module (an abelian group with a $G$-action) as the $i$-th right derived functor of the functor $$(-)^G: G-mod \rightarrow Ab, A \mapsto A^G.$$ Now, I know that one can also define a chain complex $C^*(G,A)=\{C^n(G,A)\}_{n \geq0}$ with differential $$ (df)(g_1,\ldots,g_{n+1}) =g_1 \cdot f(g_2,\ldots,g_{n+1})+ \sum_{i=1}^{n}(-1)^if(g_1,\ldots,g_i g_{i+1},\ldots,g_{n+1})+ (-1)^{n+1}f(g_1,\ldots,g_n),$$ and that $H^i(C^*(G,A)) = H^i(G;A)$. My question is, are there more differentials on $C^*(G,A)$ that give the same cohomology groups ? For example, I've found various sources where one defines derivations (normally, the kernel of the map $d: C^1(G,A) \rightarrow C^2(G,A)$) as functions satisfying the condition $f(g_1 g_2) = g_2\cdot f(g_1)+f(g_2)$, which aren't in the kernel of $d$, since the kernel of $d$ are the functions that satisfy $f(g_1 g_2) = g_1\cdot f(g_2)+f(g_1)$ (this happens, for instance, in "the Arithmetic of Elliptic curves" by Silverman). If there are more, why do they give the same cohomology groups ?

More particular, let PDer be the principal derivations (functions for which there is an $a \in A$ so that $f(g) = g\cdot a - a$, let LDer be the functions which satisfy $f(g_1 g_2) = g_1\cdot f(g_2)+f(g_1)$ and RDer be the functions which satisfy $f(g_1 g_2) = g_2\cdot f(g_1)+f(g_2)$. Is there an isomorphism

$$LDer/PDer \cong RDer/PDer$$ and if so, why ?

As always, help would be appreciated.

share|improve this question
    
Do you mean different differentials on the same sequence of groups, or do you allow chain complexes with different underlying groups? –  Zhen Lin May 14 '12 at 8:43
    
Different differentials, the chain complex stays fixed (so $G$ and $A$ are fixed) –  KevinDL May 14 '12 at 8:45

1 Answer 1

up vote 3 down vote accepted

Strictly speaking, a chain complex is a sequence of groups equipped with a differential operator, so two different differential operators gives rise to two different chain complexes.

One sufficient condition is the existence of a chain homotopy equivalence. Let's write $C^\bullet$ and $C'^\bullet$ for the two complexes. Then, we need to find chain homomorphisms $f^\bullet : C^\bullet \to C'^\bullet$ and $h^\bullet : C'^\bullet \to C^\bullet$ and a sequence of homomorphisms $\alpha^n : C^n \to C^{n-1}$, $\beta^n : C'^n \to C'^{n-1}$ such that \begin{align} d' \circ f & = f \circ d \\ d \circ h & = h \circ d' \\ d \circ \alpha + \alpha \circ d & = h \circ f - \textrm{id} \\ d' \circ \beta + \beta \circ d' & = f \circ h - \textrm{id} \end{align} It is a standard exercise to show that the homomorphisms in cohomology induced by $f^\bullet$ and $h^\bullet$ are isomorphisms.

In your chain complex, there is an obvious symmetry that we can exploit to construct "alternative" differential operators. Let $C^n = C'^n$ be the group of all functions $f : G^n \to A$, and equip $C^\bullet$ with the standard differential. For each natural number $n$, choose a permutation $\sigma_n \in S_n$. We have an obvious automorphism $\sigma_n^* : C'^n \to C'^n$ and we define $d' : C'^n \to C'^{n+1}$ by the formula below: $$d' = \sigma_{n+1}^* \circ d \circ (\sigma_n^*)^{-1}$$ It is easy to check that this makes $C'^\bullet$ into a chain complex, and $\sigma_\bullet^* : C^\bullet \to C'^\bullet$ is a chain isomorphism. Thus $C^\bullet$ and $C'^\bullet$ have the same cohomology.


One thing you might be wondering about is why the standard resolution even gives the right answer in the first place: after all, the right derived functors are supposed to be defined by injective resolutions, and this isn't one. Secretly, we are using the natural isomorphism $$H^n(G; A) \cong \textrm{Ext}^n_{\mathbb{Z} G}(\mathbb{Z}, A)$$ where $\mathbb{Z}$ is considered as a $G$-module with trivial $G$-action, and the fact that $\textrm{Ext}^\bullet$ can be computed by a projective resolution of the first variable. There is a canonical free resolution of $\mathbb{Z}$, called the bar resolution, and it is constructed as follows. Let $P_n$ be the free $\mathbb{Z} G$ module on $G^{n}$-many generators. (I am assuming $G$ is a discrete but not necessarily finite group here.) If we write the generators of $P_n$ as $[ g_1, \ldots, g_n ]$, then we can define a surjective $G$-equivariant homomorphism $p : P_0 \to \mathbb{Z}$ by $$[] \mapsto 1$$ and clearly $\ker p$ consists of those sums $\sum_i a_i g_i []$ such that $\sum_i a_i = 0$. Notice that it is generated as a $\mathbb{Z}$-module by $\{ (g - 1)[] : g \in G \}$. Let $d : P_{n+1} \to P_n$ be the $G$-equivariant homomorphism defined by $$[g_0, \ldots, g_n] \mapsto g_0 [g_1, \ldots, g_n] + \sum_{i=1}^{n-1} (-1)^i [g_0, \ldots, g_{i-1} g_i, \ldots, g_n] + (-1)^n [g_0, \ldots, g_{n-1}]$$ A standard calculation shows that $P_\bullet$ is a chain complex under this differential operator and that it is exact in positive degrees, with $\operatorname{im} (d : P_1 \to P_0) = \ker p$, so it is indeed a free resolution of $\mathbb{Z}$. Now, taking $C^\bullet = \operatorname{Hom}_{\mathbb{Z} G}(P_\bullet, A)$ gives the standard chain complex for calculating the group cohomology of $A$.

But there's another canonical free resolution $Q_\bullet$, called the normalised bar resolution, where $Q_n$ is the free $\mathbb{Z} G$-module on $(G \setminus \{ e \})^n$-many generators, and the differential $d' : P'_{n+1} \to P'_n$ is constructed essentially the same way. Notice that the chain complex $Q_\bullet$ is naturally a quotient of $P_\bullet$, in such a way that $P_n$ splits as $K_n \oplus Q_n$ as $\mathbb{Z} G$-modules, where $K_\bullet$ is the kernel of the projection $P_\bullet \to Q_\bullet$. Thus, we can construct a chain complex $K_\bullet \oplus Q_\bullet$ which is degreewise isomorphic to $P_\bullet$, but with a subtly different differential. This is another way of getting a non-standard differential on the standard chain complex.

share|improve this answer

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.