The Bar Construction

Table of Contents

• 1. Motivation: The Resolution Problem
  • 1.1 Canonical Resolutions and Functoriality
  • 1.2 The Key Idea: Resolve Using the Algebra Itself
  • 1.3 A Unified Perspective
• 2. The Algebraic Bar Construction
  • 2.1 Setup: Augmented Algebras
  • 2.2 The One-Sided Bar Complex B(k, A, k)
  • 2.3 The Two-Sided Bar Construction B(M, A, N)
  • 2.4 Acyclicity: The Contracting Homotopy
  • 2.5 Computing Tor via the Bar Resolution
• 3. The Simplicial Perspective
  • 3.1 The Bar Construction as a Simplicial Object
  • 3.2 Face and Degeneracy Maps
  • 3.3 The Normalized Chain Complex and Dold-Kan
• 4. The Topological Bar Construction: Classifying Spaces
  • 4.1 The Nerve of a Group
  • 4.2 The Classifying Space BG
  • 4.3 The Universal Bundle EG via the Two-Sided Bar
  • 4.4 Homotopy Colimits as Bar Constructions
• 5. The Monadic Bar Construction
  • 5.1 Setup and Definition
  • 5.2 The Canonical Simplicial Resolution
  • 5.3 Connection to Beck Monadicity
• 6. The Bar-Cobar Adjunction and Koszul Duality
  • 6.1 The Cobar Construction
  • 6.2 The Adjunction B Dashv Omega
  • 6.3 Koszul Duality
  • 6.4 Adams’ Cobar Construction for Loop Spaces
• 7. Hochschild Homology and the Cyclic Bar Construction
  • 7.1 The Hochschild Complex as a Bar Complex
  • 7.2 The Cyclic Bar Construction
  • 7.3 Cyclic Homology and the Circle Action
• 8. The Categorical and Infinity-Categorical Perspective
  • 8.1 Bar as Colimit: The Coequalizer Interpretation
  • 8.2 The Infinity-Categorical Bar Construction
  • 8.3 Everything Is One Construction
• 9. The Bar Construction in Spectra: B as Categorical Suspension
  • 9.1 Augmented Ring Spectra and the Spectral Bar Construction
  • 9.2 B as Categorical Suspension, Not Spectrum Suspension
  • 9.3 The Free Algebra Case: When BA Simeq Sigma M
  • 9.4 Iterated Bar and En-Algebras
• References

1. Motivation: The Resolution Problem 🔍

1.1 Canonical Resolutions and Functoriality

Let \(A\) be a ring (or algebra over a field \(k\)) and let \(M\), \(N\) be \(A\)-modules. One of the central problems of homological algebra is the computation of the derived functors \(\mathrm{Tor}_*^A(M, N)\) and \(\mathrm{Ext}^*_A(M, N)\). By definition, these require choosing a projective (or free) resolution of one of the modules. For example, to compute \(\mathrm{Tor}_*^A(M,N)\), one finds a free resolution

\[\cdots \to F_2 \to F_1 \to F_0 \to M \to 0,\]

then applies \((-) \otimes_A N\) and takes homology.

The trouble is that free resolutions are not unique. Different choices of resolution give the same answer — but constructing the comparison chain map and verifying independence requires work. For abstract existence theorems this is fine, but for explicit computation and for functoriality in \(A\) it is not.

The question: Is there a canonical, functorial free resolution — one that requires no choices, depends only on \(A\) and \(M\), and varies naturally with maps of rings or modules?

1.2 The Key Idea: Resolve Using the Algebra Itself

The answer is yes, and the key observation is almost embarrassingly simple: use the algebra \(A\) to build its own resolution. Rather than choosing generators for a free module, observe that \(A\) itself is free of rank one over \(A\). We can then build a resolution of \(k\) (the ground field, viewed as an \(A\)-module via the augmentation \(\varepsilon: A \to k\)) using only tensor products of \(A\) with itself:

\[\cdots \to A \otimes_k A \otimes_k A \to A \otimes_k A \to A \to k \to 0.\]

The differential encodes the multiplication of \(A\): it alternately multiplies adjacent tensor factors or removes boundary factors. This is the bar resolution, and it is entirely determined by the algebra structure of \(A\).

1.3 A Unified Perspective

What makes the bar construction remarkable is that this same idea — resolve an object by freely generated objects built from the generating structure — appears identically in at least five different mathematical contexts:

The unifying principle — section 8.3 — is that every bar construction is the geometric realization of a simplicial object arising from a monad.

2. The Algebraic Bar Construction 🔩

2.1 Setup: Augmented Algebras

Definition (Augmented algebra). An augmented \(k\)-algebra is a \(k\)-algebra \(A\) equipped with a \(k\)-algebra homomorphism \(\varepsilon: A \to k\) called the augmentation. The augmentation ideal is \(\bar{A} = \ker(\varepsilon)\), so \(A \cong k \oplus \bar{A}\) as \(k\)-modules.

The prototypical examples are group algebras \(k[G]\) with \(\varepsilon(g) = 1\) for all \(g \in G\), and free algebras \(T(V) = \bigoplus_{n \geq 0} V^{\otimes n}\) with \(\varepsilon\) projecting to \(V^{\otimes 0} = k\).

2.2 The One-Sided Bar Complex B(k, A, k)

Definition (Bar complex). Let \(A\) be an augmented \(k\)-algebra with augmentation ideal \(\bar{A}\). The bar complex \(B(A) = B(k, A, k)\) is the chain complex with:

• Underlying graded module: \(B_n(A) = \bar{A}^{\otimes_k n}\) in degree \(n \geq 0\).
• Notation: An element \(a_1 \otimes \cdots \otimes a_n \in \bar{A}^{\otimes n}\) is written \([a_1 | a_2 | \cdots | a_n]\).
• Differential: \(\partial: B_n(A) \to B_{n-1}(A)\) is given by

\[\partial[a_1 | a_2 | \cdots | a_n] = \sum_{i=1}^{n-1} (-1)^{i-1} [a_1 | \cdots | a_i a_{i+1} | \cdots | a_n].\]

The map \(\partial\) simply multiplies together adjacent pairs, alternating signs. It is degree \(-1\), and the standard telescoping argument gives \(\partial^2 = 0\) (adjacent terms cancel in pairs).

The complex \(B(A)\) is augmented by the map \(\varepsilon: B_0(A) = \bar{A} \hookrightarrow A \xrightarrow{\varepsilon} k\) (here we use the inclusion \(\bar{A} \subset A\), then augment; equivalently one appends \(k\) in degree \(-1\)).

2.3 The Two-Sided Bar Construction B(M, A, N)

The one-sided bar complex computes the homology of \(k\) over \(A\). For computing \(\mathrm{Tor}_*^A(M, N)\) for general modules, we need the two-sided bar construction.

Definition (Two-sided bar construction). Let \(M\) be a right \(A\)-module and \(N\) a left \(A\)-module. The two-sided bar complex \(B(M, A, N)\) is the chain complex with:

• Degree \(n\) piece: \(B_n(M, A, N) = M \otimes_k \bar{A}^{\otimes_k n} \otimes_k N\).
• Notation: An element \(m \otimes [a_1 | \cdots | a_n] \otimes n'\) is written \(m[a_1 | \cdots | a_n]n'\).
• Differential: \(\partial: B_n(M,A,N) \to B_{n-1}(M,A,N)\) is

\[\partial(m[a_1|\cdots|a_n]n') = ma_1[a_2|\cdots|a_n]n' + \sum_{i=1}^{n-1}(-1)^i m[a_1|\cdots|a_i a_{i+1}|\cdots|a_n]n' + (-1)^n m[a_1|\cdots|a_{n-1}]a_n n'.\]

The first term absorbs \(a_1\) into the right \(A\)-action on \(M\); the last term absorbs \(a_n\) into the left \(A\)-action on \(N\); the middle terms multiply adjacent elements of \(\bar{A}\).

Key observation: The complex \(B(M,A,N)\) is not over \(A\); it is a complex of \(k\)-modules, and its homology in degree \(-1\) (after appending \(M \otimes_A N\) as the augmentation target) is precisely \(\mathrm{Tor}_*^A(M,N)\).

2.4 Acyclicity: The Contracting Homotopy

The bar complex \(B(k,A,k)\) is an acyclic complex of \(k\)-modules — its terms \(\bar{A}^{\otimes n}\) are free over \(k\) but not over \(A\). The actual free left \(A\)-module resolution of \(k\) is \(B(A,A,k)\), with terms \(A \otimes_k \bar{A}^{\otimes n}\) (free over \(A\) on basis \(\bar{A}^{\otimes n}\)); one recovers \(B(k,A,k)\) from it by applying \(k \otimes_A -\). The proof of acyclicity is explicit:

Proposition (Acyclicity). The augmented bar complex

\[\cdots \to B_2(k,A,k) \xrightarrow{\partial} B_1(k,A,k) \xrightarrow{\partial} B_0(k,A,k) \xrightarrow{\varepsilon} k \to 0\]

is exact.

Proof. We exhibit a contracting homotopy \(s: B_n \to B_{n+1}\) defined by

\[s[a_1 | a_2 | \cdots | a_n] = [1 | a_1 | a_2 | \cdots | a_n]\]

(prepend the unit \(1 \in A\) to the bar sequence; note \(1 \notin \bar{A}\), so we must work in the unreduced bar complex where the \(a_i\) range over all of \(A\), not just \(\bar{A}\)). A direct computation gives \(\partial s + s \partial = \mathrm{id}\), hence \(H_n(B(A)) = 0\) for \(n > 0\) and \(H_0 \cong k\). \(\square\)

The same argument applies to the two-sided bar complex \(B(M,A,N)\): the contracting homotopy prepends a unit to the bar sequence, and one obtains:

Corollary. \(B_\bullet(M,A,N) \to M \otimes_A N \to 0\) is a resolution of \(M \otimes_A N\) by free \(k\)-modules.

This exercise makes the contracting homotopy explicit by carrying out the identity \(\partial s + s\partial = \mathrm{id}\) on a concrete element, and identifies the two-sided generalization with the extra degeneracy used in the topological setting.

2.5 Computing Tor via the Bar Resolution

Since \(B(A,A,k)\) is a free left \(A\)-module resolution of \(k\), applying \(M \otimes_A -\) gives a complex computing \(\mathrm{Tor}^A_*(M, k)\). The two-sided complex \(B(M,A,N)\) generalizes this: \(B(M,A,A)\) is a free right \(A\)-module resolution of \(M\) (terms \(M \otimes_k \bar{A}^{\otimes n} \otimes_k A\), free over \(A\)), and

\[\mathrm{Tor}_n^A(M, N) \cong H_n(B(M,A,A) \otimes_A N) = H_n(B(M,A,N)).\]

Note also that \(k \otimes_A k \cong k\) (since \(A\) acts on \(k\) via \(\varepsilon\), the tensor relation becomes trivial), so \(B(k,A,k)\) resolves \(k \otimes_A k = k\) — consistent with the general formula for \(M = N = k\). This is the canonical computation of Tor: no choices of resolution were made.

This exercise computes \(\mathrm{Tor}\) via the bar resolution in the simplest nontrivial case and reveals the periodic structure that arises from the exterior algebra.

3. The Simplicial Perspective 🔺

3.1 The Bar Construction as a Simplicial Object

The bar complex \(B(M,A,N)\) is not just a chain complex — it is naturally a simplicial \(k\)-module. Understanding this simplicial structure is key to seeing why the bar construction generalizes so broadly.

Recall that a simplicial object in a category \(\mathcal{C}\) is a functor \(X_\bullet: \Delta^{\mathrm{op}} \to \mathcal{C}\), where \(\Delta\) is the simplex category with objects \([n] = \{0 < 1 < \cdots < n\}\) and morphisms the order-preserving maps.

Definition (Simplicial bar construction). The simplicial \(k\)-module \(B_\bullet(M,A,N)\) has: - Degree \(n\) term: \(B_n(M,A,N) = M \otimes_k A^{\otimes_k n} \otimes_k N\)

(note: here we use all of \(A\), not just \(\bar{A}\) — the augmentation ideal arises from the normalized chain complex).

3.2 Face and Degeneracy Maps

The structure maps are:

Face maps \(d_i: B_n \to B_{n-1}\) for \(0 \leq i \leq n\):

\[d_i(m \otimes a_1 \otimes \cdots \otimes a_n \otimes n') = \begin{cases} ma_1 \otimes a_2 \otimes \cdots \otimes a_n \otimes n' & i = 0 \\ m \otimes a_1 \otimes \cdots \otimes a_i a_{i+1} \otimes \cdots \otimes a_n \otimes n' & 0 < i < n \\ m \otimes a_1 \otimes \cdots \otimes a_{n-1} \otimes a_n n' & i = n \end{cases}\]

Degeneracy maps \(s_j: B_n \to B_{n+1}\) for \(0 \leq j \leq n\):

\[s_j(m \otimes a_1 \otimes \cdots \otimes a_n \otimes n') = m \otimes a_1 \otimes \cdots \otimes a_j \otimes 1_A \otimes a_{j+1} \otimes \cdots \otimes a_n \otimes n'\]

The degeneracy maps insert a unit \(1_A \in A\) at position \(j+1\).

This exercise verifies the simplicial identities directly from the definitions and identifies which algebraic axiom (associativity vs. unitality) each identity uses.

3.3 The Normalized Chain Complex and Dold-Kan

The relationship between the simplicial bar construction and the chain complex bar construction is given by the Dold-Kan correspondence.

Theorem (Dold-Kan, informal). There is an equivalence of categories between simplicial abelian groups and non-negatively graded chain complexes of abelian groups. Under this equivalence, a simplicial module \(X_\bullet\) corresponds to its normalized chain complex \(N(X)_n = X_n / (\text{images of all degeneracy maps})\), with differential \(\sum_{i=0}^{n} (-1)^i d_i\).

For the simplicial bar construction: the normalized chain complex of \(B_\bullet(M,A,N)\) is precisely the two-sided bar complex \(B(M,A,N)\) defined in section 2.3. (The images of the degeneracy maps — which insert units — exactly account for the difference between using \(A\) and using \(\bar{A}\) in each tensor factor.)

4. The Topological Bar Construction: Classifying Spaces 🌐

4.1 The Nerve of a Group

Let \(G\) be a topological group. The same simplicial formulas define a simplicial topological space \(NG\) (the nerve of \(G\), viewed as a one-object groupoid):

• \(NG_n = G^n\) (the \(n\)-fold product)
• Face maps: \(d_0(g_1, \ldots, g_n) = (g_2, \ldots, g_n)\); \(d_i(g_1, \ldots, g_n) = (g_1, \ldots, g_i g_{i+1}, \ldots, g_n)\) for \(0 < i < n\); \(d_n(g_1, \ldots, g_n) = (g_1, \ldots, g_{n-1})\)
• Degeneracy maps: \(s_j(g_1, \ldots, g_n) = (g_1, \ldots, g_j, 1, g_{j+1}, \ldots, g_n)\)

These are exactly the same algebraic formulas as the bar construction with \(A = k[G]\), \(M = N = k\), now interpreted topologically.

4.2 The Classifying Space BG

Definition (Classifying space). The classifying space of \(G\) is

\[BG = |NG| = B(*, G, *)\]

the geometric realization of the nerve \(NG\). Here \(*\) denotes the one-point space (the trivial \(G\)-space).

Geometric realization replaces each \(G^n\) with a copy of the standard topological \(n\)-simplex \(\Delta^n\), glued together along the face and degeneracy maps:

\[|NG| = \left(\bigsqcup_{n \geq 0} G^n \times \Delta^n\right) \bigg/ \sim\]

where \((d_i(g), t) \sim (g, \delta_i(t))\) and \((s_j(g), t) \sim (g, \sigma_j(t))\), with \(\delta_i, \sigma_j\) the standard face and degeneracy maps on simplices.

Key properties: - When \(G\) is a discrete group, \(\pi_1(BG) \cong G\) and \(\pi_k(BG) = 0\) for \(k > 1\) — so \(BG = K(G,1)\) is the Eilenberg-MacLane space. - \(H_*(BG; \mathbb{Z}) \cong H_*(G; \mathbb{Z})\) — the singular homology of \(BG\) recovers group homology. - A principal \(G\)-bundle \(P \to X\) is classified by a map \(X \to BG\); homotopy classes of such maps are in bijection with isomorphism classes of principal \(G\)-bundles.

This exercise makes the bar construction completely concrete for the two most fundamental groups, connecting the simplicial machinery to familiar classifying spaces.

This exercise shows the bar construction is strictly compatible with products, giving a clean proof that \(K(G \times H, 1) \simeq K(G,1) \times K(H,1)\).

This exercise connects the combinatorics of the bar construction to classical topology via the Euler characteristic, and catches a common off-by-one error in cell counting.

4.3 The Universal Bundle EG via the Two-Sided Bar

The universal principal \(G\)-bundle \(EG \to BG\) also arises from the bar construction. Using \(G\) itself as a left \(G\)-space (via left multiplication):

\[EG = B(G, G, *) = |[n \mapsto G^{n+1}]|\]

The face maps for \(EG\) are: \(d_0(g_0, g_1, \ldots, g_n) = (g_0 g_1^{-1} \cdot \text{something})\)… more precisely, \(EG\) is the realization of the simplicial space with \(n\)-simplices \(G^{n+1}\) and face map \(d_i\) multiplying the \(i\)-th and \((i+1)\)-th entries.

The bundle map \(EG \to BG\) is induced by the simplicial map \(G^{n+1} \to G^n\) that forgets the first factor and records the ratios \(g_0^{-1}g_1, g_1^{-1}g_2, \ldots\)

Theorem. \(EG\) is contractible and \(G\) acts freely on \(EG\) with quotient \(EG/G \cong BG\).

Proof sketch. The contracting homotopy \(s([g_0 | g_1 | \cdots | g_n]) = [e | g_0 | \cdots | g_n]\) (prepend the identity, exactly as in the algebraic case) shows \(EG\) is contractible. The free \(G\)-action is by \((g \cdot (g_0, g_1, \ldots, g_n)) = (gg_0, gg_1, \ldots, gg_n)\). \(\square\)

More generally, the two-sided bar construction \(B(Y, G, X)\) for \(G\)-spaces \(Y\) (right) and \(X\) (left) gives the homotopy quotient (Borel construction):

\[B(Y, G, X) \simeq Y \times_G EG \times_G X.\]

This exercise verifies the contractibility of \(EG\) and freeness of the \(G\)-action from the simplicial structure, giving a complete proof that \(EG \to BG\) is a principal \(G\)-bundle.

This exercise deduces \(\Omega BG \simeq G\) from the fiber sequence, providing the key structural result that explains why the bar construction deloops topological groups.

4.4 Homotopy Colimits as Bar Constructions

The bar construction also computes homotopy colimits of diagrams. Let \(\mathcal{D}\) be a small category and \(F: \mathcal{D} \to \mathbf{Top}\) a diagram. The homotopy colimit of \(F\) is

\[\mathrm{hocolim}_{\mathcal{D}}\, F = B(*, \mathcal{D}, F) = \left\lvert [n \mapsto \bigsqcup_{d_0 \to d_1 \to \cdots \to d_n} F(d_0)] \right\rvert\]

where the coproduct runs over all composable chains of \(n\) morphisms in \(\mathcal{D}\).

• Face map \(d_0\) applies \(F\) to the first morphism \(d_0 \to d_1\) (functoriality).
• Face maps \(d_i\) for \(0 < i < n\) compose adjacent morphisms \(d_i \to d_{i+1}\).
• Face map \(d_n\) forgets the last object.
• Degeneracy maps insert identity morphisms.

This perspective (Malkiewich’s notes on homotopy colimits) shows that \(\mathrm{hocolim}\) is a systematic version of the bar construction with \(G\) replaced by the category \(\mathcal{D}\) and a point replaced by the functor \(F\).

This exercise identifies \(B(*,G,X)\) as the Borel construction \(EG \times_G X\), recovering equivariant homotopy theory as a special case of the bar machinery.

5. The Monadic Bar Construction 🏗️

5.1 Setup and Definition

The algebraic and topological bar constructions are both special cases of the monadic bar construction. See the monads note for background on monads.

Definition (Monad). A monad on a category \(\mathcal{C}\) is a triple \((T, \mu, \eta)\) where \(T: \mathcal{C} \to \mathcal{C}\) is a functor, \(\mu: T^2 \Rightarrow T\) is the multiplication (a natural transformation), and \(\eta: \mathrm{Id} \Rightarrow T\) is the unit, satisfying the associativity and unit axioms.

Definition (Monadic bar construction). Given a monad \((T, \mu, \eta)\) on \(\mathcal{C}\) and an object \(X \in \mathcal{C}\), the bar construction \(B_\bullet(T, X)\) is the simplicial object with:

• Degree \(n\) term: \(B_n(T, X) = T^{n+1}X\)

• Face maps: \(d_i = T^i \mu T^{n-i}: T^{n+2}X \to T^{n+1}X\) for \(0 \leq i \leq n\)

  (apply the monad multiplication \(\mu: T^2 \Rightarrow T\) at the \(i\)-th position)

• Degeneracy maps: \(s_j = T^j \eta T^{n-j}: T^{n+1}X \to T^{n+2}X\) for \(0 \leq j \leq n\)

  (apply the monad unit \(\eta: \mathrm{Id} \Rightarrow T\) at the \(j\)-th position)

If \(X\) is a \(T\)-algebra with structure map \(\xi: TX \to X\), then the augmentation \(d_{n+1} = T^n \xi: T^{n+1}X \to T^n X\) extends this to an augmented simplicial object.

5.2 The Canonical Simplicial Resolution

Theorem. For any \(T\)-algebra \((X, \xi)\), the augmented simplicial object

\[B_\bullet(T,X) \xrightarrow{\xi \circ \mu^n} X\]

is a simplicial resolution of \(X\) by free \(T\)-algebras: each \(T^{n+1}X\) is freely generated (as a \(T\)-algebra) by \(T^n X\).

The geometric realization \(|B_\bullet(T,X)|\) is, when it exists (e.g., when \(\mathcal{C}\) is a model category with enough structure), the cofibrant replacement of \(X\) in the model category of \(T\)-algebras.

The contracting homotopy of the algebraic bar complex has a monadic incarnation: the maps \(s_{-1} = \eta_{T^{n+1}X}: T^{n+1}X \to T^{n+2}X\) (apply \(\eta\) on the outside) provide a simplicial homotopy between the identity and the constant map, showing that \(|B_\bullet(T,X)|\) is contractible before applying the algebra structure.

5.3 Connection to Beck Monadicity

The bar construction is intimately connected to Beck’s monadicity theorem (see Descent and Monadicity). Given an adjunction \(F \dashv U: \mathcal{C} \to \mathcal{D}\) with induced monad \(T = UF\) on \(\mathcal{D}\):

• The bar construction \(B_\bullet(T, X)\) for \(X \in \mathcal{D}\) is the canonical resolution of \(X\) by free \(T\)-algebras.
• The coequalizer of \(UFUFX \rightrightarrows UFX\) is \(X\) itself (when \(U\) reflects isomorphisms).
• Beck’s theorem says \(\mathcal{C}\) is equivalent to \(\mathcal{D}^T\) exactly when \(U\) preserves and reflects coequalizers of \(U\)-split pairs — and the bar construction provides the canonical split pair witnessing this.

The bar construction as the canonical splitting: The pair \((UFUF X, UFX)\) with face maps \(UFU(\varepsilon_X): UFUFX \to UFX\) (counit applied inside) and \(\varepsilon_{UFX}: UFUFX \to UFX\) (counit applied outside) is always \(U\)-split by the bar construction. The geometric realization \(|B_\bullet(T,X)|\) gives the free resolution of \(X\) used in descent theory.

6. The Bar-Cobar Adjunction and Koszul Duality 🔄

6.1 The Cobar Construction

Every bar construction has a dual: the cobar construction, which turns coalgebras into algebras.

Definition (Coaugmented coalgebra). A coaugmented dg-coalgebra is a dg-coalgebra \((C, \Delta, \varepsilon)\) with a map \(\eta: k \to C\) (the coaugmentation) such that \(\varepsilon \circ \eta = \mathrm{id}_k\). The coaugmentation coideal is \(\bar{C} = \ker(\varepsilon: C \to k)\).

Definition (Cobar construction). The cobar construction \(\Omega C\) of a coaugmented dg-coalgebra \(C\) is the dg-algebra:

• Underlying graded algebra: \(T(s^{-1}\bar{C})\), the free associative algebra on the desuspension \(s^{-1}\bar{C}\) (shift of grading by \(-1\)).
• Notation: An element \((s^{-1}c_1) \otimes \cdots \otimes (s^{-1}c_n)\) is written \([c_1 | \cdots | c_n]\) (same bar notation, different meaning).
• Differential: The differential on \(\Omega C\) has two parts:
  1. The internal differential of \(C\), applied to each \(c_i\).
  2. The cobar differential \(\delta[c_1|\cdots|c_n] = \sum_i (-1)^{|c_1|+\cdots+|c_{i-1}|} [c_1|\cdots|\Delta(c_i)|\cdots|c_n]\), using the reduced coproduct \(\bar{\Delta}: \bar{C} \to \bar{C} \otimes \bar{C}\).

6.2 The Adjunction B Dashv Omega

There is an adjunction between the category of augmented dg-algebras and the category of coaugmented conilpotent dg-coalgebras:

\[B \dashv \Omega: \mathbf{dgAlg}_{\mathrm{aug}} \rightleftharpoons \mathbf{dgCoalg}_{\mathrm{coaug}}^{\mathrm{conil}}\]

The bar construction \(BA\) of an augmented dg-algebra \(A\) is the dg-coalgebra with: - Underlying coalgebra: \(T^c(s\bar{A})\), the cofree conilpotent coalgebra on the suspension \(s\bar{A}\). - Differential: The bar differential from section 2.2, now incorporating any internal differential on \(A\). - Coproduct: The deconcatenation coproduct \(\Delta[a_1|\cdots|a_n] = \sum_{i=0}^n [a_1|\cdots|a_i] \otimes [a_{i+1}|\cdots|a_n]\).

The adjunction unit \(A \to \Omega BA\) is the bar-cobar resolution of \(A\).

6.3 Koszul Duality

For a quadratic algebra \(A = T(V)/(R)\) where \(R \subset V^{\otimes 2}\), the Koszul dual coalgebra is \(A^¡ = T^c(sV)/(sR)^{\perp}\). The algebra \(A\) is Koszul if the natural map

\[\Omega(A^¡) \xrightarrow{\sim} A\]

is a quasi-isomorphism — i.e., the bar-cobar resolution through the Koszul dual is minimal (the bar-cobar resolution of a general algebra is always a resolution, but typically far from minimal).

Examples of Koszul algebras: - Polynomial algebras \(k[x_1, \ldots, n]\) (Koszul dual: exterior algebra \(\Lambda[x_1^*, \ldots, x_n^*]\)) - Exterior algebras (Koszul dual: polynomial algebra) - The universal enveloping algebra \(U(\mathfrak{g})\) for a Lie algebra \(\mathfrak{g}\) (Koszul dual: the Chevalley-Eilenberg algebra)

6.4 Adams’ Cobar Construction for Loop Spaces

The original motivation for the cobar construction was purely topological. J. Frank Adams (1956) showed:

Theorem (Adams). For a simply-connected topological space \(X\) with basepoint, there is a quasi-isomorphism of dg-algebras

\[\Omega C_*(X, *) \xrightarrow{\sim} C_*(\Omega X)\]

where \(C_*(X, *)\) is the singular chain coalgebra of \(X\) (with coproduct given by the Alexander-Whitney map), and \(\Omega X\) is the based loop space with its Pontryagin product.

This remarkable theorem says: the cobar construction on the chains of a space models the chains on its loop space. The bar-cobar adjunction is the algebraic shadow of the geometric suspension-loop adjunction \(\Sigma \dashv \Omega\).

7. Hochschild Homology and the Cyclic Bar Construction ♻️

7.1 The Hochschild Complex as a Bar Complex

Let \(A\) be a \(k\)-algebra and \(M\) an \(A\)-bimodule (equivalently, a left \(A \otimes_k A^{\mathrm{op}}\)-module). The Hochschild complex \(C_*(A, M)\) is:

\[C_n(A, M) = M \otimes_k A^{\otimes_k n}, \quad n \geq 0\]

with differential \(b: C_n \to C_{n-1}\) given by

\[b(m \otimes a_1 \otimes \cdots \otimes a_n) = ma_1 \otimes a_2 \otimes \cdots \otimes a_n + \sum_{i=1}^{n-1}(-1)^i m \otimes a_1 \otimes \cdots \otimes a_i a_{i+1} \otimes \cdots \otimes a_n + (-1)^n a_n m \otimes a_1 \otimes \cdots \otimes a_{n-1}.\]

Observation: \(C_*(A, M) = B(M, A, k)\) — the two-sided bar complex with \(N = k\) and the bimodule action folding into the two-sided structure. The Hochschild homology is therefore

\[HH_n(A, M) = H_n(C_*(A, M)) = \mathrm{Tor}_n^{A \otimes A^{\mathrm{op}}}(A, M).\]

For \(M = A\) (with the natural bimodule structure), \(HH_*(A, A) = \mathrm{Tor}_*^{A^e}(A, A)\) where \(A^e = A \otimes_k A^{\mathrm{op}}\).

7.2 The Cyclic Bar Construction

Hochschild homology carries an extra structure: a cyclic symmetry. The key observation is that \(C_n(A,A) = A^{\otimes(n+1)}\) admits a cyclic action by the cyclic group \(C_{n+1}\) via

\[t_n(a_0 \otimes a_1 \otimes \cdots \otimes a_n) = (-1)^n a_n \otimes a_0 \otimes a_1 \otimes \cdots \otimes a_{n-1}\]

(cyclic rotation with sign). This is the cyclic operator of Connes.

Definition (Cyclic bar construction). The cyclic bar construction \(B^{\mathrm{cy}}_\bullet(A)\) is the cyclic object with \(B^{\mathrm{cy}}_n(A) = A^{\otimes(n+1)}\), face maps \(d_i\) as in the Hochschild complex, and the cyclic operator \(t_n\) above.

A cyclic object in \(\mathcal{C}\) is a simplicial object with compatible cyclic actions — formally, a functor from Connes’ cyclic category \(\Lambda\) to \(\mathcal{C}\).

7.3 Cyclic Homology and the Circle Action

The geometric realization \(|B^{\mathrm{cy}}_\bullet(A)|\) carries an \(S^1\)-action (since cyclic objects realize to spaces with circle actions), and:

\[HH_*(A) = \pi_*(|B^{\mathrm{cy}}_\bullet(A)|)\]

The circle action on \(|B^{\mathrm{cy}}(A)|\) gives, via the fibration \(ES^1 \to BS^1 = \mathbb{CP}^\infty\), the cyclic homology \(HC_*(A)\) as the \(S^1\)-equivariant version:

\[HC_*(A) = H_*^{S^1}(|B^{\mathrm{cy}}(A)|)\]

and the periodic cyclic homology \(HP_*(A)\) is the corresponding Tate construction.

8. The Categorical and Infinity-Categorical Perspective 🌌

8.1 Bar as Colimit: The Coequalizer Interpretation

The geometric realization of the bar construction is a colimit. The simplest case: for a \(T\)-algebra \(X\), the realization \(|B_\bullet(T,X)|\) is the coequalizer in \(\mathcal{C}\):

\[T^2 X \underset{T\xi}{\overset{\mu_X}{\rightrightarrows}} TX \to X\]

where \(\mu_X: T^2X \to TX\) is the monad multiplication and \(T\xi: T^2X \to TX\) applies \(T\) to the algebra structure \(\xi: TX \to X\). This coequalizer is the Beck coequalizer and is always split (by the \(T\)-algebra axioms), hence absolute.

More precisely, \(|B_\bullet(T,X)|\) is the geometric realization (= colimit over \(\Delta^{\mathrm{op}}\)) of the entire simplicial diagram, not just the coequalizer of the first two face maps. But for good enough monads, these agree.

Proposition. When \(T\) preserves reflexive coequalizers and \(\mathcal{C}\) has such coequalizers, \(|B_\bullet(T,X)| \cong X\) for any \(T\)-algebra \((X, \xi)\).

Proof sketch. The colimit of \(B_\bullet(T,X)\) is the coequalizer of \(T^2X \rightrightarrows TX\), which by the split pair argument is \(X\) itself. \(\square\)

8.2 The Infinity-Categorical Bar Construction

In the \(\infty\)-categorical setting (Lurie’s Higher Algebra), the bar construction works even better because geometric realization (= \(\infty\)-colimit over \(\Delta^{\mathrm{op}}\)) is better behaved than ordinary coequalizers.

Theorem (Barr-Beck-Lurie). Let \(f: \mathcal{C} \to \mathcal{D}\) be a functor of \(\infty\)-categories admitting a left adjoint \(g\). Then \(f\) is monadic if and only if: 1. \(f\) is conservative (reflects equivalences). 2. \(\mathcal{C}\) admits geometric realizations of \(f\)-split simplicial objects, and \(f\) preserves them.

Under monadicity, the canonical comparison \(|B_\bullet(T, X)| \simeq X\) holds for every \(T\)-algebra \(X\).

The \(\infty\)-categorical bar construction also gives the functor

\[B: \mathbf{Alg}_{\mathcal{O}}(\mathcal{C}) \to \mathbf{CoAlg}_{\mathcal{O}^!}(\mathcal{C})\]

from \(\mathcal{O}\)-algebras to \(\mathcal{O}^!\)-coalgebras for an operad \(\mathcal{O}\) and its Koszul dual \(\mathcal{O}^!\) — Koszul duality for \(\infty\)-operads (Lurie HA §4.4).

8.3 Everything Is One Construction

We can now state the unifying principle:

Every bar construction is the geometric realization of a simplicial object arising from a monad, with the face maps given by the monad multiplication and the augmentation given by an algebra structure map.

Surprisingly, the contracting homotopy that proves acyclicity is always the same map: prepend the unit (of the monad, of the algebra, of the group, …). The sign conventions vary but the idea is universal.

9. The Bar Construction in Spectra: B as Categorical Suspension 🌠

9.1 Augmented Ring Spectra and the Spectral Bar Construction

In the ∞-categorical setting, the bar construction generalizes to ring spectra — \(\mathbb{E}_1\)- or \(\mathbb{E}_\infty\)-algebra objects in the stable ∞-category of spectra \(\mathrm{Sp}\). The algebraic field \(k\) is replaced by the sphere spectrum \(\mathbb{S}\) (the unit for the smash product \(\wedge\)), and the tensor product \(\otimes_k\) is replaced by \(\wedge\).

The input is an augmented \(\mathbb{E}_n\)-ring spectrum: an \(\mathbb{E}_n\)-algebra \(A\) with a map of \(\mathbb{E}_n\)-algebras \(\varepsilon: A \to \mathbb{S}\). The augmentation ideal is the fiber:

\[\bar{A} = \mathrm{fib}(\varepsilon: A \to \mathbb{S}), \quad \text{fitting into } \bar{A} \to A \xrightarrow{\varepsilon} \mathbb{S}.\]

Definition (Spectral bar construction). The bar construction of an augmented \(\mathbb{E}_1\)-ring spectrum \(A\) is the relative smash product:

\[BA = \mathbb{S} \wedge_A \mathbb{S}\]

i.e., the geometric realization of the simplicial spectrum \(B_\bullet(\mathbb{S}, A, \mathbb{S})\) with \(B_n = \mathbb{S} \wedge A^{\wedge n} \wedge \mathbb{S}\), face maps using the multiplication of \(A\) and the augmentation maps at the ends, degeneracy maps inserting the unit \(\eta: \mathbb{S} \to A\).

9.2 B as Categorical Suspension, Not Spectrum Suspension

The key clarification: \(BA\) is not \(\Sigma A\) as a spectrum, but rather the suspension of \(A\) in the ∞-category of augmented \(\mathbb{E}_\infty\)-algebras.

The suspension functor in any pointed ∞-category \(\mathcal{C}\) (with zero object \(0\)) is:

\[\Sigma_\mathcal{C}\, X = 0 \sqcup_X 0 \quad (\text{pushout of two maps } X \to 0).\]

So \(BA\) is the suspension of \(A\) — but only in the ∞-category of augmented \(\mathbb{E}_\infty\)-algebras. The forgetful functor \(\mathrm{Alg}^{\mathrm{aug}}_{\mathbb{E}_\infty} \to \mathrm{Sp}\) does not preserve pushouts, so this algebraic suspension does not coincide with the spectrum-level suspension.

9.3 The Free Algebra Case: When BA Simeq Sigma M

The one case where spectral bar and spectrum suspension coincide is for free algebras.

Theorem. Let \(M\) be a spectrum. For the free \(\mathbb{E}_\infty\)-algebra \(\mathrm{Sym}(M) = \bigoplus_{n \geq 0} (M^{\wedge n})_{h\Sigma_n}\) on \(M\):

\[B(\mathrm{Sym}(M)) \simeq \Sigma M \quad \text{(as spectra)}.\]

Why. The bar spectral sequence degenerates: \(\mathrm{Tor}^{\pi_*\mathrm{Sym}(M)}_{p,*}(\pi_*\mathbb{S}, \pi_*\mathbb{S}) = 0\) for \(p \geq 2\), because the Tor groups of a polynomial algebra over itself vanish above the indecomposables (homological degree 1). The first column is \(\bar{A}/\bar{A}^{\otimes 2} \simeq \Sigma M\) and everything else is zero.

This is the spectral version of the classical fact: the indecomposables of \(k[V] = \mathrm{Sym}(V)\) are just \(V\) (all higher-order terms are decomposable). The bar construction strips away everything but the generators.

9.4 Iterated Bar and En-Algebras

The bar construction can be iterated, and each application “uses up” one level of commutativity. For an augmented \(\mathbb{E}_n\)-algebra \(A\) (for \(n \geq 1\)), the \(n\)-fold bar construction is:

\[B^n A = \underbrace{B \circ B \circ \cdots \circ B}_{n}(A)\]

Each \(B\) lowers the \(\mathbb{E}_k\)-level by one: \(B\) sends augmented \(\mathbb{E}_k\)-algebras to augmented \(\mathbb{E}_{k-1}\)-coalgebras. So \(B^n A\) requires \(A\) to be an \(\mathbb{E}_n\)-algebra (at minimum), and produces an \(\mathbb{E}_0\)-coalgebra (a coaugmented cochain complex / cospectra).

The iterated bar construction is the \(\mathbb{E}_n\)-suspension:

\[B^n A = \Sigma_{\mathrm{Alg}^{\mathrm{aug}}_{\mathbb{E}_n}} A.\]

For free algebras, this degenerates cleanly:

\[B^n(\mathrm{Free}_{\mathbb{E}_n}(M)) \simeq \Sigma^n M.\]

The full hierarchy of structures and their delooping machines:

In the \(\mathbb{E}_\infty\) (fully commutative / stable) case, the ∞-fold bar construction of a grouplike \(\mathbb{E}_\infty\)-space \(A\) is its suspension spectrum \(\Sigma^\infty_+ A\) — the stable homotopy type encoding only the additive structure of \(A\).

References