bar-cobar construction

References

Summary

Bar: The bar construction takes a weak algebraic structure (like a homotopical structure) and replaces it by a strict (non-homotopy) structure, by first considering all formal compositions, and then quotienting out by certain relations.

It's called the bar construction, because you write a formal composition of algebraic elements using bars: [x1 ∣ x2 ∣ … ∣ xn]. These formal compositions are then interpretted as simplices. This allows you to turn algebraic structure into spaces

Cobar: Breaks things into formal decompositions. This is very useful with quasicategories: there is no single composition of 1-morphisms — given a 2-simplex, it's really a collection of two different paths that are homoptic (as witnessed by the 2-simplex). The cobar construction then breaks this simplex down into a sum of the different constituent 1-simplices. This allows you to turn spaces into coalgebraic structure

Uses differential graded (co)operads to introduce homotopy.

Two steps:

  1. Bar: a projection
  2. Cobar: a free reconstruction

Requirements

You need to be able to dualize

You need an augmented operad

You typically work with linear operads

You typically assume a connected operad

Definitions

Augmentation

A linear functional ε : A → k on an algebra is called an augmentation if ε ∘ η = 1k

The pair (A, ε) is called an augmented algebra.

The augmentation ideal is ker ε, and we have A ≅ k ⊕ ker ε

Coaugmentation

Let C be a coalgebra. A linear map η : C → k is a coaugmentation if ε ∘ η = 1k

The coaugmented coideal C̄ is the coalgebra coker η, and C ≅ k ⊕ coker η

Reduced comultiplication

Let (C, ε) be a coalgebra with coaugmentation η : k → C, where η(1) ≡ 1 ∈ C. The reduced comultiplication induced by η is Δ̄ = Δ − 1 ⊗ id − id ⊗ 1

Conilpotent coalgebra

A coalgebra C is conilpotent if for all x ∈ C, there is some n such that Δn(x) = 0

Cofree conilpotent coalgebra

Let C be a vector space. The cofree conilpotent coaugmented coalgebra Tc(C) has underlying vector space T(V) , the tensor product on V, and coproduct $$ \Delta (v_{1} \ldots v_{n}) = \sum_{i=0}^{n} v_{1} \ldots v_{i} \otimes v_{i+1} \ldots v_{n} $$ coaugmentation η(1) = 1 ∈ V0 ≅ k, and counit ε(x) = 0 for x ≠ 1 and ε(1) = 1.

Notes

Coaugmented: This isn't just a coalgebra, it's coaugmented. So we can take $ \overline{T^{c} (V)} \cong \bigoplus_{n > 0} V^{\otimes n} $.

Note: The order zero and nth order terms will be 1 ⊗ v1…vn and v1…vn ⊗ 1. Removing these creates the reduced coalgebra, which is conilpotent! This is because the reduced comultiplication will always reduce the number of terms in each tensor factor. After enough time, it'll bring one of the tensor factors down to having only 1, which will cause that term to be removed.

Conilpotent: Tc(V) is not the cofree object in coalgebras! It's only the cofree object in conilpotent coalgebras. The actual cofree object has as underlying vector space ∏nV⊗n which allows formal power series in elements of V.

Bar construction

Let A be an augmented dga. The bar construction on A is the coaugmented coalgebra B(A) = Tc(sĀ) where (sĀ)n ≡ Ān − 1 viewed as a dgcoa, with differential d = d1 + d2, where $$ d_{1}(x_{1} \otimes \ldots \otimes x_{n}) = -\sum_{i=1}^{n} (-1)^{|x_{1}| + \ldots + |x_{i-1}|} x_{1} \otimes \ldots \otimes d_{A}(x_{i} )\otimes \ldots \otimes x_{n} $$

$$ d_{2}(x_{1} \otimes \ldots \otimes x_{n}) = \sum_{i=1}^{n-1} (-1)^{|x_{1}| + \ldots + |x_{i}|} x_{1} \otimes \ldots x_{i} x_{i+1} \otimes \ldots \otimes x_{n} $$

Notes

Because we've shifted the degrees, when we write |xi|, we really mean the shifted degree, which is deg (xi) + 1, where deg  means the original degree

Cobar construction

Let C be an coaugmented dgcoa. The cobar construction on C is the augmented algebra Ω(C) = T(s−1C̄) where (s−1C̄)n ≡ C̄n + 1 viewed as a dgcoa, with differential d = d1 + d2, where $$ d_{1}(x_{1} \otimes \ldots \otimes x_{n}) = -\sum_{i=1}^{n} (-1)^{|x_{1}| + \ldots + |x_{i-1}|} x_{1} \otimes \ldots \otimes d_{C}(x_{i} )\otimes \ldots \otimes x_{n} $$

$$ d_{2}(x_{1} \otimes \ldots \otimes x_{n}) =\sum_{i=1}^{n} \sum (-1)^{|x_{1}| + \ldots + |x_{i, (1)}|} x_{1} \otimes \ldots \otimes x_{i, (1)} \otimes x_{i, (2)} \otimes \ldots \otimes x_{n} $$

Notes

An easier way to write this is that d1 = ddC and d2 = dΔC, using the notation that df sends the linear map f through all of the tensor factors, picking up appropriate signs

Maurer-Cartan equation:

Free operad in vector spaces

Let M be an Smodule in vector spaces. The free operad ℱ[M] in vector spaces has its operations

And composition given by formally grafting trees together

Cofree cooperad in vector spaces

Let M be an Smodule in vector spaces. The cofree cooperad ℱc[M] in vector spaces has its operations the same spaces ℱ[M], but with comultiplication given by the following:

If T is a tree with n levels of composition, where at level i you've attached the trees T1i, …, Tkii then

Δ(T) = ∑iT1i ⊗ … ⊗ Tkii

Notes

In order for this to be a finite sum, we need

Augmented DG Operad

An augmented DGO is an operad in chain complexes 𝒪 along with a map of operads ε : 𝒪 → 𝒦 where 𝒦 is the operad with 𝒦[1] = k and 𝒦[n ≠ 1] = 0

Notes

This allows you to disregard the stuff in arity one from the operad. This is useful for the same reason that it is in the bar-cobar construction in dgas — you don't want identity maps to allow your decompositions to be infinite when you pass to the cooperad.

Theorems

Reduced coalgebra

Let C be a coaugmented coalgebra with unit 1. Then if C̄ is the coaugmented coideal, the reduced comultiplication makes C̄ into a coalgebra, called the reduced coalgebra : Δ̄ : C̄ ⊗ C̄ → C̄

Proof

Bar-cobar adjunction

(Co)fibrant resolutions

Let A be an augmented DGA then $$ \Omega B(A) \xrightarrow{\sim} A $$ (ie they're quasi-isomorphic)

Similarly, if C is a conilpotent DGcoA, then $$ C \xrightarrow{\sim} B \Omega (C) $$

Notes

Resolutions: Since we have these quasi-isomorphisms, we get a resolution of any (co)augmented DGcoA. The main issue is that when you perform this resolution, the space you get out the other side is very unwieldy — it's not messy, but it's still hard to deal with because they're so much (co)free structure

Purpose: What's the point of doing a resolution when it just gives you a more complicated object? The idea is that the more complicated object you get is more complicated because you made it free — it no longer has any relations (the data of the relations basically became the data of all of the ways of breaking stuff down or putting it together). So mapping into/out of it becomes easier

Ext/Tor: For an augmented dga A, the homology of the dual of the bar construction is the same thing as the ext! ExtA(k, k) ≅ H•(B(A)*) and TorA(k, k) ≅ H•(B(A))

Koszul duality: In the case of Koszul algebras, the homology of the bar construction (the Tor) becomes concentrated in a single degree, allowing you to just study a sub-coalgebra, called the Koszul dual coalgebra. And in fact, taking the dual of this coalgebra gives you the Koszul dual algebra A!, whose cohomology is then the Ext!

Adam's cobar theorem

Examples