# Homology of bar complex vs homology of indecomposables

$\require{AMScd}$

Background: This question is about the bar and cobar constructions, and their relationship with the indecomposables of a dg-algebra. A brief summary of the bar and cobar constructions on ncatlab can be found at [1]. I will also mention the more general case of $A_\infty$ algebras; this case is discussed in this survey [2] or in Proute's thesis [3]. I posted a related question a few months ago in the mathoverflow post [4]. There I called the dg-module of indecomposables the linearized chain complex.'' It seems that the former is more standard terminology so I've switched for the time being.

Setup: Let $(A,\partial,\epsilon)$ be an augmented dg-algebra over a field $k$, i.e. a dg-algebra $(A,\partial)$ along with dg-algebra map $\epsilon:A \to k$. Here $k$ is a dga with trivial grading and differential.

The module of indecomposables $(IA,\partial)$ of $(A,\partial,\epsilon)$ is a dg-module whose underlying graded $k$ vector-space is $IA = \ker(\epsilon)/\ker(\epsilon)^2$. The differential $\partial$ on $A$ descends to a differential $\partial$ on $IA$. Furthermore, the correspondence $(A,\partial,\epsilon) \mapsto (IA,\partial)$ is functorial. That is, given a map: $$f:(A,\partial,\delta) \to (B,\partial,\epsilon)$$ of augmented dg-algebras, i.e. a dg-algebra map with $\epsilon \circ f = \delta$, we get a map of dg-modules: $$If:(IA,\partial) \to (IB,\partial)$$

The bar complex $BA$ of $(A,\partial,\epsilon)$ to a dg-coalgebra whose underlying graded vector-space is: $$BA := k \oplus BA^+ \qquad BA^+ := \bigoplus_{i=1}^\infty \ker(\epsilon)[1]^{\otimes i}$$ The differential combines multiplication and differentiation from $(A,\partial)$ (see [1] for an actual definition) and the coaugmentation $\iota:k \to BA$ is just the inclusion. Similarly, the cobar complex $\Omega C$ of an augmented dg-coalgebra $C$ is an augmented dg-algebra admitting an analogous description to that of $BA$, as a certain tensor algebra. Both constructions are functorial, and there are generalizations of to $A_\infty$ algebras and coalgebras (see [2] or [3]).

The functors $B$ and $\Omega$ form an adjoint pair ($B$ is right adjoint to $\Omega$) and we have natural adjoint maps $\Omega B A \to A$ and $C \to B \Omega C$, which are quasi-isomorphisms of augmented dg-algebras and coaugmented dg-coalgebras, respectively. There is a reference for this in the post [5]. In particular, we get a natural map of dg-modules between the indecomposables: $$I\Omega B A \to IA$$ I believe that it is true that $I\Omega C \simeq \text{coker}(\kappa)$ for any coaugmented dg-coalgebra $(C,\partial,\kappa)$ (an actual isomorphism of chain groups, not just a quasi-isomorphism) so that the above map yields a canonical map of dg-modules and a corresponding map on homology: $$BA^+ \to IA \qquad H(BA^+) \to H(IA)$$

Question: Under what circumstances is the natural map $I\Omega B A \to IA$ an isomorphism on homology? When is it surjective on homology? I'm particularly interested in the case where the underlying graded algebra of $(A,\partial,\epsilon)$ is the free graded commutative algebra over some generators. Any partial answers are much appreciated.

Examples: In some cases this map is simple to understand. For instance, when $A$ is an algebra with the trivial differential and grading, then $H(IA) = IA$ and I believe that the map on homology $H(BA^+) \to H(IA)$ is just projection onto $H^1(BA^+) \simeq IA$.

Another context where something similar happens is when $A = \Omega C$ where $(C,\partial,\kappa)$ is a coaugmented dg-coalgebra (or even an $A_\infty$ coalgebra). Then the map: $$\text{coker}(\iota) \simeq BA^+ \to IA \simeq \text{coker}(\kappa)$$ is just the $A_\infty$ quasi-isomorphism on the cokernels of the coaugmentations induced by the quasi-isomorphism of $A_\infty$ coalgebras of $B\Omega C \to C$, itself induced by the adjunction map $\Omega B \Omega C \to \Omega C$. Since $A_\infty$ quasi-isomorphisms descend to isomorphisms on homology, we know in particular $H(BA^+) \simeq H(IA)$ under the map of interest.

The functor of indecomposables is the left adjoint of a Quillen adjunction between dg-algebras and dg-modules. (For a general reference, see Section 12.1.3 of the book Algebraic Operads by Loday and Vallette, though this was certainly known before the book – I just happen to have it on my desk.) As such it preserves quasi-isomorphisms between cofibrant algebras. The bar-cobar construction on $A$ is always cofibrant, so a sufficient condition for your claim is that $A$ should be cofibrant.
Since you are interested in the case of free symmetric algebra, if $A$ is free symmetric on some space of generators $V$ equipped with a filtration $$0 = F_0 V \subset F_1 V \subset \dots \subset V = \bigcup_{i \ge 0} F_i V$$ such that $\partial(F_i V) \subset S(F_{i-1}(V))$ (i.e. the differential of an element in $F_i V$ is a product of elements in lower filtration), then $A$ is cofibrant. In fact you can assume that the filtration is indexed by an ordinal, not just integers, if you need.
The homology of $I \Omega B A$ is the Hochschild homology of $A$ with constant coefficients $HH_*(A;\Bbbk)$. (Again, see the book of Loday and Vallette.) I wouldn't expect your map to be a quasi-isomorphism if $A$ is not cofibrant – it certainly can happen, but it's basically a coincidence. As for surjective on cohomology... I have no idea.
• Dear Najib: unless I am missing something, the indecomposables of $\Omega BA$ are the generators $BA$. It's homology is $\operatorname{Tor}^A(\mathbb k,\mathbb k)$. To get Hochschild cohomology of $A$, you want to take the complex of derivations of $\Omega BA$ plus $\Omega BA$, i.e. the cone of the map $\Omega BA\to\operatorname{Der}(\Omega BA)$ that includes it as inner derivations. Or is $\Omega BA$ being augmented over something other than $\mathbb k$? – Pedro Tamaroff Jul 16 '18 at 14:13
• @Pedro I should have been more precise. I meant the Hochschild homology of $A$ with constant coefficients, i.e. $HH_*(A; \Bbbk)$. – Najib Idrissi Jul 16 '18 at 14:26