Algebras and Modules over Operads

Table of Contents

• 1. Algebras over an Operad
  • 1.1 The Structure Maps
  • 1.2 Compatibility Axioms
  • 1.3 Algebras as Operad Maps
  • 1.4 Recovering Classical Examples
  • 1.5 The Category of O-Algebras
• 2. Free Algebras
  • 2.1 Construction and Universal Property
  • 2.2 Classical Cases
  • 2.3 The Operad Monad
• 3. Left and Right Modules
  • 3.1 Left Modules
  • 3.2 Right Modules
  • 3.3 Bimodules
  • 3.4 Classical Recovery for Ass
• 4. The Enveloping Algebra
  • 4.1 Modules over an O-Algebra
  • 4.2 Construction of the Enveloping Algebra
  • 4.3 Classical Cases
• 5. Derivations
  • 5.1 Definition and the Leibniz Rule
  • 5.2 The Module of Derivations
  • 5.3 Operad Derivations and Algebra Derivations
• 6. Kahler Differentials
  • 6.1 The Universal Property
  • 6.2 Construction via Generators and Relations
  • 6.3 Recovery for Com
• 7. A-infinity Algebras
  • 7.1 The Stasheff Relations
  • 7.2 Low-Degree Consequences
  • 7.3 Infinity-Morphisms
  • 7.4 Operadic Perspective
• 8. The Probability Operad Revisited
  • 8.1 P-Algebra Structure on the Positive Reals
  • 8.2 Shannon Entropy as a Derivation
  • 8.3 Connection to BFL Entropy
• References

1. Algebras over an Operad

📐 Throughout, \(k\) is a commutative unital ring, \(\mathbf{Mod}_k\) is the category of \(k\)-modules, and \(\mathcal{O}\) is an operad in \((\mathbf{SymSeq}, \circ, \mathbf{1})\) as developed in Definitions and Examples. We recall that \(\mathcal{O}\) is equipped with a composition morphism \(\gamma: \mathcal{O} \circ \mathcal{O} \to \mathcal{O}\) and a unit \(\eta: \mathbf{1} \to \mathcal{O}\), satisfying the monoid axioms in \((\mathbf{SymSeq}, \circ, \mathbf{1})\).

1.1 The Structure Maps

Definition 1.1 (Algebra over an Operad). An algebra over \(\mathcal{O}\), or an \(\mathcal{O}\)-algebra, is a \(k\)-module \(A\) together with a collection of \(k\)-linear structure maps \[\gamma_A(n) \colon \mathcal{O}(n) \otimes_k A^{\otimes n} \longrightarrow A, \qquad n \geq 0,\] satisfying \(S_n\)-equivariance and the two compatibility axioms stated below.

The \(S_n\)-equivariance condition means: for every \(\mu \in \mathcal{O}(n)\), every permutation \(\sigma \in S_n\), and every tuple \((a_1, \ldots, a_n) \in A^n\), \[\gamma_A(n)(\mu \cdot \sigma;\, a_1, \ldots, a_n) = \gamma_A(n)(\mu;\, a_{\sigma(1)}, \ldots, a_{\sigma(n)}).\] Here \(\mu \cdot \sigma\) is the right \(S_n\)-action on \(\mathcal{O}(n)\).

1.2 Compatibility Axioms

Let \(\mu \in \mathcal{O}(n)\) and \(\nu_1 \in \mathcal{O}(k_1), \ldots, \nu_n \in \mathcal{O}(k_n)\), so that \(\gamma(\mu; \nu_1, \ldots, \nu_n) \in \mathcal{O}(k_1 + \cdots + k_n)\) is the composite operation.

Axiom A (Associativity). For all arities and all elements as above: \[\gamma_A\!\left(\gamma(\mu;\nu_1,\ldots,\nu_n);\; a_{1,1},\ldots,a_{n,k_n}\right) = \gamma_A\!\left(\mu;\; \gamma_A(\nu_1; a_{1,1},\ldots,a_{1,k_1}),\ldots,\gamma_A(\nu_n; a_{n,1},\ldots,a_{n,k_n})\right).\]

Axiom B (Unitality). Writing \(\mathrm{id} = \eta(1_k) \in \mathcal{O}(1)\) for the image of the unit map: \[\gamma_A(\mathrm{id};\, a) = a \quad \text{for all } a \in A.\]

These axioms are most cleanly expressed as commutative diagrams. Let \(\overline{\mathcal{O}}(A)\) denote the symmetric sequence \(\{n \mapsto \mathcal{O}(n) \otimes A^{\otimes n}\}\), so that \(\gamma_A\) is collectively a map of the Schur functor evaluation \(\mathcal{O}(A) \to A\).

The associativity axiom is the outer square of:

The unitality axiom is:

Here \(\mathbf{1}(A) = A\) canonically, and \(\sim\) is this canonical isomorphism.

1.3 Algebras as Operad Maps

🔑 There is a fundamental reformulation that makes the structure entirely explicit.

Proposition 1.2. The data of an \(\mathcal{O}\)-algebra structure on \(A\) is equivalent to an operad morphism \[\rho \colon \mathcal{O} \longrightarrow \mathrm{End}_A,\] where \(\mathrm{End}_A\) is the endomorphism operad of \(A\), defined by \(\mathrm{End}_A(n) := \mathrm{Hom}_k(A^{\otimes n}, A)\) with \(S_n\) acting by precomposition with permutations of tensor factors.

Proof sketch. Given structure maps \(\gamma_A(n)\), define \(\rho_n \colon \mathcal{O}(n) \to \mathrm{Hom}_k(A^{\otimes n}, A)\) by \[\rho_n(\mu)(a_1 \otimes \cdots \otimes a_n) := \gamma_A(\mu; a_1, \ldots, a_n).\] Equivariance of \(\gamma_A\) is equivalent to \(S_n\)-equivariance of \(\rho_n\). Axiom A translates to \(\rho\) commuting with the operad compositions \(\gamma\) and \(\gamma_{\mathrm{End}_A}\). Axiom B translates to \(\rho\) sending \(\mathrm{id} \in \mathcal{O}(1)\) to \(\mathrm{id}_A \in \mathrm{End}_A(1)\). The correspondence is bijective. \(\square\)

1.4 Recovering Classical Examples

Ass-algebras. The associative operad has \(\mathrm{Ass}(n) = k[S_n]\) (the free \(k[S_n]\)-module of rank 1). The equivariance condition and composition axioms force \(\gamma_A\) to give exactly one \(n\)-ary operation for each ordering of inputs, subject only to associativity. Concretely: the generator \(\mathrm{id}_{S_n} \in \mathrm{Ass}(n)\) acts as the \(n\)-fold iterated product \(a_1 \cdots a_n\). An \(\mathrm{Ass}\)-algebra is therefore precisely a unital associative algebra.

Com-algebras. Since \(\mathrm{Com}(n) = k\) with trivial \(S_n\)-action, equivariance forces \(\gamma_A(\mu; a_1, \ldots, a_n) = \gamma_A(\mu; a_{\sigma(1)}, \ldots, a_{\sigma(n)})\) for all \(\sigma\). Combined with associativity, this gives exactly a commutative associative algebra.

Lie-algebras. The Lie operad encodes the antisymmetry \([a,b] = -[b,a]\) and Jacobi identity; a \(\mathrm{Lie}\)-algebra is precisely a Lie algebra in the classical sense.

1.5 The Category of O-Algebras

Definition 1.3 (Morphism of \(\mathcal{O}\)-Algebras). A morphism \(f \colon (A, \gamma_A) \to (B, \gamma_B)\) of \(\mathcal{O}\)-algebras is a \(k\)-linear map \(f \colon A \to B\) that is compatible with all structure maps: \[f\!\left(\gamma_A(\mu; a_1, \ldots, a_n)\right) = \gamma_B\!\left(\mu; f(a_1), \ldots, f(a_n)\right)\] for all \(n \geq 0\), \(\mu \in \mathcal{O}(n)\), \(a_i \in A\).

The resulting category \(\mathcal{O}\text{-}\mathbf{Alg}\) has \(\mathcal{O}\)-algebras as objects and such morphisms as morphisms. It is complete and cocomplete (limits computed in \(\mathbf{Mod}_k\), colimits via reflexive coequalizers). When \(\mathcal{O} = \mathrm{Com}\), this recovers the category of commutative \(k\)-algebras; when \(\mathcal{O} = \mathrm{Lie}\), it recovers the category of Lie algebras.

2. Free Algebras

💡 The category \(\mathcal{O}\text{-}\mathbf{Alg}\) has enough structure to guarantee the existence of free objects, constructed explicitly via Schur functors (see Definitions and Examples, Section 1.3).

2.1 Construction and Universal Property

Definition 2.1 (Free \(\mathcal{O}\)-Algebra). Given a \(k\)-module \(V\), the free \(\mathcal{O}\)-algebra on \(V\) is \[\mathcal{O}(V) := \bigoplus_{n \geq 0} \mathcal{O}(n) \otimes_{S_n} V^{\otimes n},\] equipped with the structure maps \[\gamma_{\mathcal{O}(V)}(\mu;\, [\nu_1 \otimes v_1],\, \ldots,\, [\nu_k \otimes v_k]) := [\gamma(\mu;\nu_1,\ldots,\nu_k) \otimes v_1 \otimes \cdots \otimes v_k]\] where \([\cdot]\) denotes the \(S_n\)-coinvariant class and \(v_i \in V^{\otimes k_i}\).

Theorem 2.2 (Adjunction). The assignment \(V \mapsto \mathcal{O}(V)\) extends to a functor \(\mathcal{O}(-): \mathbf{Mod}_k \to \mathcal{O}\text{-}\mathbf{Alg}\) which is left adjoint to the forgetful functor \(U: \mathcal{O}\text{-}\mathbf{Alg} \to \mathbf{Mod}_k\): \[\mathrm{Hom}_{\mathcal{O}\text{-}\mathbf{Alg}}(\mathcal{O}(V), A) \cong \mathrm{Hom}_{\mathbf{Mod}_k}(V, U(A)),\] natural in \(V\) and \(A\).

Proof sketch. The unit of the adjunction is the inclusion \(\iota_V: V \hookrightarrow \mathcal{O}(V)\) into the \(n=1\) summand: \(V = \mathcal{O}(1) \otimes_{S_1} V \subseteq \mathcal{O}(V)\) (since \(\mathcal{O}(1) \ni \mathrm{id}\) acts as identity). Given \(f: V \to U(A)\), the extension \(\tilde{f}: \mathcal{O}(V) \to A\) is defined on generators by \(\tilde{f}[\mu \otimes v_1 \otimes \cdots \otimes v_n] := \gamma_A(\mu; f(v_1), \ldots, f(v_n))\). Equivariance and the axioms for \(\gamma_A\) ensure \(\tilde{f}\) is a well-defined \(\mathcal{O}\)-algebra morphism, and it is the unique such extension. \(\square\)

2.2 Classical Cases

Com. Since \(\mathrm{Com}(n) = k\) with trivial \(S_n\)-action, \(\mathrm{Com}(n) \otimes_{S_n} V^{\otimes n} = k \otimes_{S_n} V^{\otimes n} \cong V^{\otimes n}/S_n = \mathrm{Sym}^n(V)\). Therefore: \[\mathrm{Com}(V) = \bigoplus_{n \geq 0} \mathrm{Sym}^n(V) = \mathrm{Sym}(V),\] the symmetric algebra on \(V\). This recovers the fact that the free commutative algebra is the symmetric algebra.

Ass. Since \(\mathrm{Ass}(n) = k[S_n]\) with the regular right \(S_n\)-action, \(k[S_n] \otimes_{S_n} V^{\otimes n} \cong V^{\otimes n}\). Therefore: \[\mathrm{Ass}(V) = \bigoplus_{n \geq 0} V^{\otimes n} = T(V),\] the tensor algebra on \(V\). This recovers the fact that the free associative algebra is the tensor algebra.

Lie. The Lie operad is more subtle: \(\mathrm{Lie}(n)\) is the multilinear part of the free Lie algebra on \(n\) generators. One has \(\mathrm{Lie}(V) \cong \mathrm{FreeLie}(V)\), the free Lie algebra, embedded in \(T(V)\) via the Poincaré–Birkhoff–Witt basis.

2.3 The Operad Monad

📐 The free-algebra construction assembles into a monad on \(\mathbf{Mod}_k\).

Definition 2.3 (Operad Monad). The monad associated to \(\mathcal{O}\) is the endofunctor \(T_\mathcal{O} := U \circ \mathcal{O}(-): \mathbf{Mod}_k \to \mathbf{Mod}_k\), i.e., \(T_\mathcal{O}(V) = \mathcal{O}(V)\) as a \(k\)-module, equipped with: - Unit \(\eta_V: V \to T_\mathcal{O}(V)\): the inclusion into the \(n=1\) summand. - Multiplication \(\mu_V: T_\mathcal{O}(T_\mathcal{O}(V)) \to T_\mathcal{O}(V)\): the extension of \(\mathrm{id}: T_\mathcal{O}(V) \to T_\mathcal{O}(V)\) via the adjunction applied with \(A = \mathcal{O}(V)\).

Concretely, \(\mu_V\) is given by “substituting trees into trees”: an element of \(\mathcal{O}(\mathcal{O}(V))\) is a tree of operations with leaves in \(\mathcal{O}(V)\), and \(\mu_V\) composes them using the operad structure \(\gamma\).

Proposition 2.4. The category of Eilenberg-Moore algebras for the monad \(T_\mathcal{O}\) is isomorphic to \(\mathcal{O}\text{-}\mathbf{Alg}\).

Proof sketch. An Eilenberg-Moore algebra is a \(k\)-module \(A\) with a map \(h: T_\mathcal{O}(A) \to A\) satisfying \(h \circ \eta_A = \mathrm{id}_A\) and \(h \circ \mu_A = h \circ T_\mathcal{O}(h)\). Setting \(\gamma_A(n) := h|_{\mathcal{O}(n) \otimes_{S_n} A^{\otimes n}}\) recovers Definition 1.1. The axioms match precisely. \(\square\)

3. Left and Right Modules

📐 Having algebras in hand, we next define modules — objects on which an operad acts from one side. The natural setting is again symmetric sequences, with actions defined via the composition product \(\circ\).

3.1 Left Modules

Definition 3.1 (Left \(\mathcal{O}\)-Module). A left \(\mathcal{O}\)-module is a symmetric sequence \(M \in \mathbf{SymSeq}\) together with a morphism of symmetric sequences \[\lambda: \mathcal{O} \circ M \longrightarrow M\] satisfying: 1. Associativity: \(\lambda \circ (\gamma \circ \mathrm{id}_M) = \lambda \circ (\mathrm{id}_\mathcal{O} \circ \lambda)\) as maps \(\mathcal{O} \circ \mathcal{O} \circ M \to M\). 2. Unitality: \(\lambda \circ (\eta \circ \mathrm{id}_M) = \mathrm{id}_M\) as maps \(\mathbf{1} \circ M \cong M \to M\).

In terms of components, recalling that \((\mathcal{O} \circ M)(n) = \bigoplus_{k \geq 0} \mathcal{O}(k) \otimes_{S_k} \bigoplus_{n_1+\cdots+n_k = n} \mathrm{Ind}_{S_{n_1} \times \cdots \times S_{n_k}}^{S_n}(M(n_1) \otimes \cdots \otimes M(n_k))\), the map \(\lambda\) specifies how \(\mathcal{O}\)-operations compose with operations already in \(M\).

The axioms read as the following commutative diagrams:

3.2 Right Modules

Definition 3.2 (Right \(\mathcal{O}\)-Module). A right \(\mathcal{O}\)-module is a symmetric sequence \(M\) together with \[\rho: M \circ \mathcal{O} \longrightarrow M\] satisfying the mirror axioms: 1. Associativity: \(\rho \circ (\rho \circ \mathrm{id}_\mathcal{O}) = \rho \circ (\mathrm{id}_M \circ \gamma)\) as maps \(M \circ \mathcal{O} \circ \mathcal{O} \to M\). 2. Unitality: \(\rho \circ (\mathrm{id}_M \circ \eta) = \mathrm{id}_M\) as maps \(M \circ \mathbf{1} \cong M \to M\).

3.3 Bimodules

Definition 3.3 (\(\mathcal{O}\)-Bimodule). An \(\mathcal{O}\)-bimodule is a symmetric sequence \(M\) equipped with both a left action \(\lambda: \mathcal{O} \circ M \to M\) and a right action \(\rho: M \circ \mathcal{O} \to M\), satisfying the compatibility condition: \[\lambda \circ (\mathrm{id}_\mathcal{O} \circ \rho) = \rho \circ (\lambda \circ \mathrm{id}_\mathcal{O})\] as maps \(\mathcal{O} \circ M \circ \mathcal{O} \to M\).

This is the operadic analogue of the left-right compatibility condition \(r(am) = a(rm)\) for a classical bimodule.

We denote the categories: - \({}_\mathcal{O}\mathbf{Mod}\): left \(\mathcal{O}\)-modules. - \(\mathbf{Mod}_\mathcal{O}\): right \(\mathcal{O}\)-modules. - \({}_\mathcal{O}\mathbf{Mod}_\mathcal{O}\): \(\mathcal{O}\)-bimodules.

3.4 Classical Recovery for Ass

🔑 The cleanest verification of these definitions is recovering classical module theory.

Proposition 3.4. An \(\mathrm{Ass}\)-bimodule in \(\mathbf{SymSeq}\) concentrated in arity 0 is precisely a classical bimodule over an associative algebra.

Proof sketch. A left \(\mathrm{Ass}\)-module structure on a symmetric sequence \(M\) concentrated in arity 0 (i.e., \(M(0) = N\), \(M(k) = 0\) for \(k \geq 1\)) reduces, by unfolding \((\mathrm{Ass} \circ M)(0) = \bigoplus_k \mathrm{Ass}(k) \otimes_{S_k} N^{\otimes 0}\) — wait, one must be careful: \(M(0) \otimes \cdots\) for \(k\) slots is \(M(0)^{\otimes 0} = k\) for \(k=0\). The action on arity \(0\) from \(\mathrm{Ass}(k)\) gives maps \(A^{\otimes k} \otimes N \to N\) (left multiplication), and the right action from \(N \otimes A^{\otimes k} \to N\) (right multiplication). Associativity and unitality of \(\lambda\) and \(\rho\) are equivalent to the bimodule axioms \((a_1 a_2)n = a_1(a_2 n)\) and \(n(a_1 a_2) = (na_1)a_2\) and \(1 \cdot n = n \cdot 1 = n\). \(\square\)

4. The Enveloping Algebra

📐 For a \(k\)-algebra \(A\), a module over \(A\) in the classical sense is a module over the ring \(A\). For an \(\mathcal{O}\)-algebra, the correct analogue of “module” must be sensitive to the operadic structure; the device encoding this is the enveloping algebra.

4.1 Modules over an O-Algebra

Definition 4.1 (Representation / Module). Let \((A, \gamma_A)\) be an \(\mathcal{O}\)-algebra. A representation of \(A\) (or \(A\)-module in the \(\mathcal{O}\)-sense) is a \(k\)-module \(M\) together with structure maps \[\gamma_M(n, i) \colon \mathcal{O}(n) \otimes A^{\otimes (i-1)} \otimes M \otimes A^{\otimes (n-i)} \longrightarrow M, \quad 1 \leq i \leq n,\; n \geq 1,\] satisfying: - \(S_n\)-equivariance (permuting all slots, with \(M\) in position \(i\) permuted accordingly to position \(\sigma(i)\)). - Compatibility with \(\gamma\): the maps \(\gamma_M(n, i)\) express “applying \(\mu \in \mathcal{O}(n)\) to inputs where all but the \(i\)-th are in \(A\) and the \(i\)-th is in \(M\).” - When all inputs are in \(A\), recovery of the algebra action: \(\gamma_M(n, i)|_{M=A} = \gamma_A(n)\).

4.2 Construction of the Enveloping Algebra

Definition 4.2 (Enveloping Algebra). The enveloping algebra of the \(\mathcal{O}\)-algebra \(A\), denoted \(U_\mathcal{O}(A)\), is the associative \(k\)-algebra defined as the quotient: \[U_\mathcal{O}(A) := T\!\left(\bigoplus_{n \geq 1} \mathcal{O}(n) \otimes_{S_{n-1}} A^{\otimes(n-1)}\right) \Big/ \sim\] where \(T(-)\) denotes the tensor algebra, and \(\sim\) is the equivalence relation generated by: 1. \((\mu \cdot \sigma) \otimes a_1 \otimes \cdots \otimes a_{n-1} \sim \mu \otimes a_{\sigma^{-1}(1)} \otimes \cdots \otimes a_{\sigma^{-1}(n-1)}\) for \(\sigma \in S_{n-1}\) (embedded in \(S_n\) as the stabilizer of position \(n\)). 2. \(\gamma(\mu; \nu_1, \ldots, \nu_n) \otimes \cdots \sim \mu \otimes \cdots \otimes \nu_j \otimes \cdots\) (operadic composition relations). 3. \(\mathrm{id} \otimes 1_k \sim 1_{U}\) (unitality).

The universal property of \(U_\mathcal{O}(A)\) is: \[\mathrm{Der}_\mathcal{O}(A, M) \cong \mathrm{Hom}_{U_\mathcal{O}(A)\text{-}\mathbf{Mod}}(\Omega^1_\mathcal{O}(A), M),\] naturally in the \(A\)-module \(M\) (see Section 6 for \(\Omega^1_\mathcal{O}(A)\)). Equivalently, \(U_\mathcal{O}(A)\)-modules are in bijection with representations of \(A\) in the sense of Definition 4.1.

4.3 Classical Cases

For \(\mathrm{Ass}\). The \(n\)-th summand in the generating module is \(\mathrm{Ass}(n) \otimes_{S_{n-1}} A^{\otimes(n-1)} = k[S_n] \otimes_{S_{n-1}} A^{\otimes(n-1)}\). By Mackey’s induction, \(k[S_n] \otimes_{S_{n-1}} A^{\otimes(n-1)} \cong k[S_n/S_{n-1}] \otimes A^{\otimes(n-1)} \cong A^{\otimes(n-1)} \oplus A^{\otimes(n-1)}\) (left and right cosets). The quotient by the operadic relations gives \(A \otimes A^{\mathrm{op}}\). Thus \(U_{\mathrm{Ass}}(A) = A \otimes_k A^{\mathrm{op}}\), the classical enveloping algebra of an associative algebra.

For \(\mathrm{Lie}\). Here \(\mathrm{Lie}(n) \otimes_{S_{n-1}} A^{\otimes(n-1)}\) assembles under the PBW theorem to the universal enveloping algebra of the Lie algebra \(A\). Thus \(U_{\mathrm{Lie}}(\mathfrak{g}) = U(\mathfrak{g})\), the classical universal enveloping algebra of a Lie algebra.

5. Derivations

📐 Derivations are the fundamental infinitesimal symmetries of algebras; for operadic algebras, the correct notion of derivation is dictated by the multi-linear structure.

5.1 Definition and the Leibniz Rule

Definition 5.1 (\(\mathcal{O}\)-Derivation). Let \((A, \gamma_A)\) be an \(\mathcal{O}\)-algebra and \(M\) a representation of \(A\). A derivation from \(A\) to \(M\) is a \(k\)-linear map \(d: A \to M\) satisfying the operadic Leibniz rule: for all \(n \geq 1\), all \(\mu \in \mathcal{O}(n)\), and all \(a_1, \ldots, a_n \in A\), \[d\!\left(\gamma_A(\mu;\, a_1,\ldots,a_n)\right) = \sum_{i=1}^{n} \gamma_M\!\left(\mu;\, a_1,\ldots,a_{i-1},\, da_i,\, a_{i+1},\ldots,a_n\right).\] Here \(\gamma_M(\mu; a_1, \ldots, a_{i-1}, m, a_{i+1}, \ldots, a_n)\) uses the representation maps of Definition 4.1 with \(m = da_i \in M\) in position \(i\).

5.2 The Module of Derivations

Definition 5.2. The set \(\mathrm{Der}_\mathcal{O}(A, M)\) of all \(\mathcal{O}\)-derivations from \(A\) to \(M\) is naturally a \(k\)-module (pointwise addition and scalar multiplication). When \(M = A\) with its own \(A\)-bimodule structure (via the representation maps coming from \(\gamma_A\) itself), we write \(\mathrm{Der}_\mathcal{O}(A) := \mathrm{Der}_\mathcal{O}(A, A)\).

For fixed \(A\), the assignment \(M \mapsto \mathrm{Der}_\mathcal{O}(A, M)\) is a functor \(A\text{-}\mathbf{Mod} \to \mathbf{Mod}_k\). The universal object representing this functor is the module of Kähler differentials \(\Omega^1_\mathcal{O}(A)\) (Section 6).

5.3 Operad Derivations and Algebra Derivations

💡 Bradley’s paper Entropy as an Operad Derivation works with derivations of the operad \(\mathcal{P}\) itself (not of a \(\mathcal{P}\)-algebra). These two notions are related by a natural evaluation map.

Definition 5.3 (Derivation of an Operad). A derivation of the operad \(\mathcal{O}\) with values in an \(\mathcal{O}\)-bimodule \(M\) is a collection of \(k\)-linear maps \(\partial_n: \mathcal{O}(n) \to M(n)\) satisfying the operadic Leibniz rule: \[\partial_{k_1+\cdots+k_n}\!\left(\gamma(\mu;\nu_1,\ldots,\nu_n)\right) = \sum_{i=1}^{n} \gamma_M\!\left(\mu;\nu_1,\ldots,\nu_{i-1},\partial_{k_i}(\nu_i),\nu_{i+1},\ldots,\nu_n\right)\] where \(\gamma_M\) uses the bimodule action.

Proposition 5.4 (Evaluation Map). Let \(\mathcal{O}\) be an operad, \(A\) an \(\mathcal{O}\)-algebra, and \(N\) an \(A\)-module. There is a natural \(k\)-linear map (the evaluation) \[\mathrm{ev}_A: \mathrm{Der}(\mathcal{O}, \mathrm{End}_A) \longrightarrow \mathrm{Der}_\mathcal{O}(A, N)\] defined as follows: given a derivation \(\partial: \mathcal{O} \to \mathrm{End}_A\) of the operad (with values in the endomorphism bimodule), form \(d: A \to N\) by \(d(a) := \sum_n (\partial_n \mathrm{id}_A)(a)\) in degree \(n=1\).

The significance: this map explains how the operadic derivation of the probability operad \(\mathcal{P}\) (Shannon entropy \(H\), shown by Bradley to be a derivation \(\mathcal{P} \to \mathbb{R}\)-valued bimodule) descends to a derivation of \(\mathcal{P}\)-algebras in the classical sense. See Section 8.

6. Kahler Differentials

📐 The module of Kähler differentials \(\Omega^1_\mathcal{O}(A)\) is the universal linearization of \(\mathcal{O}\)-derivations — it converts derivations into module homomorphisms.

6.1 The Universal Property

Definition 6.1 (Kähler Differentials). Let \((A, \gamma_A)\) be an \(\mathcal{O}\)-algebra. The module of Kähler differentials \(\Omega^1_\mathcal{O}(A)\) is a \(U_\mathcal{O}(A)\)-module equipped with a universal derivation \(d: A \to \Omega^1_\mathcal{O}(A)\), characterized by the natural isomorphism \[\mathrm{Der}_\mathcal{O}(A, M) \cong \mathrm{Hom}_{U_\mathcal{O}(A)}(\Omega^1_\mathcal{O}(A), M)\] for every \(A\)-module (i.e., \(U_\mathcal{O}(A)\)-module) \(M\).

In other words, every \(\mathcal{O}\)-derivation \(\partial: A \to M\) factors uniquely as \(A \xrightarrow{d} \Omega^1_\mathcal{O}(A) \xrightarrow{\tilde\partial} M\) where \(\tilde\partial\) is a \(U_\mathcal{O}(A)\)-module map:

6.2 Construction via Generators and Relations

Proposition 6.2 (Explicit Construction). The module \(\Omega^1_\mathcal{O}(A)\) is the \(U_\mathcal{O}(A)\)-module \[\Omega^1_\mathcal{O}(A) = U_\mathcal{O}(A) \otimes_k A \Big/ \langle \text{Leibniz relations} \rangle,\] where the Leibniz relations are: for all \(n\), \(\mu \in \mathcal{O}(n)\), and \(a_1, \ldots, a_n \in A\), \[d\!\left(\gamma_A(\mu; a_1,\ldots,a_n)\right) \sim \sum_{i=1}^n [\text{element of } U_\mathcal{O}(A) \text{ corresponding to } (\mu; a_1,\ldots,\hat{a}_i,\ldots,a_n)] \cdot d(a_i).\] The universal derivation is \(d: A \to \Omega^1_\mathcal{O}(A)\), \(a \mapsto [1 \otimes a]\).

Proof sketch. The universal property follows from the Freyd adjoint functor theorem applied to the forgetful functor from pairs \((M, \partial: A \to M)\) to \(A\)-modules \(M\); the explicit construction is the coend formula realizing this left adjoint. The Leibniz relations ensure \(d\) is an \(\mathcal{O}\)-derivation; universality follows from the freeness of \(U_\mathcal{O}(A) \otimes A\) as a \(U_\mathcal{O}(A)\)-module. \(\square\)

6.3 Recovery for Com

Proposition 6.3. For \(\mathcal{O} = \mathrm{Com}\) and \(A\) a commutative \(k\)-algebra: \[\Omega^1_{\mathrm{Com}}(A) \cong \Omega^1_{A/k},\] the classical module of Kähler differentials — the \(A\)-module generated by symbols \(da\) for \(a \in A\), subject to \(d(ab) = a \cdot db + b \cdot da\) and \(d(\lambda) = 0\) for \(\lambda \in k\).

Proof sketch. Since \(U_{\mathrm{Com}}(A) = A\) (as noted in Section 4.3), the \(U_{\mathrm{Com}}(A)\)-module \(\Omega^1_{\mathrm{Com}}(A)\) is an \(A\)-module. The Leibniz relations for \(\mathrm{Com}\) at \(n=2\) are exactly \(d(ab) = a \cdot db + b \cdot da\) (using the commutativity of \(A\) to identify both \(\gamma_M(2,1)\) and \(\gamma_M(2,2)\)). The generators \(\{da : a \in A\}\) and the relations match the classical definition. \(\square\)

7. A-infinity Algebras

📐 \(A_\infty\)-algebras, introduced by Stasheff in 1963, are algebras that are homotopy associative in a precise, controlled sense — associative up to an infinite tower of coherent homotopies.

7.1 The Stasheff Relations

Definition 7.1 (\(A_\infty\)-Algebra). Let \(k\) be a field of characteristic zero. An \(A_\infty\)-algebra is a \(\mathbb{Z}\)-graded \(k\)-module \(A = \bigoplus_{n \in \mathbb{Z}} A^n\) equipped with graded \(k\)-linear maps \[m_n: A^{\otimes n} \longrightarrow A, \quad n \geq 1,\] of degree \(|m_n| = 2 - n\) (i.e., \(m_n\) maps \(A^{\otimes n}\) to \(A\) with a degree shift of \(2-n\)), satisfying the Stasheff identities: for all \(n \geq 1\), \[\sum_{\substack{r+s+t = n \\ r, t \geq 0,\; s \geq 1}} (-1)^{rs+t}\; m_{r+1+t}\!\left(\mathrm{id}^{\otimes r} \otimes m_s \otimes \mathrm{id}^{\otimes t}\right) = 0.\]

The sign \((-1)^{rs+t}\) arises from the Koszul sign convention: moving \(m_s\) of degree \(2-s\) past \(r\) inputs each of total degree (tracked cumulatively).

7.2 Low-Degree Consequences

Let us unpack the Stasheff identities for small \(n\).

\(n=1\): The only term has \(r=t=0\), \(s=1\): \(m_1 \circ m_1 = 0\). Thus \(m_1\) is a differential on \(A\), making \((A, m_1)\) a chain complex.

\(n=2\): Terms have \((r,s,t) \in \{(1,1,0),(0,1,1),(0,2,0)\}\): \[(-1)^1 m_2(m_1 \otimes \mathrm{id}) + (-1)^0 m_2(\mathrm{id} \otimes m_1) + (-1)^0 m_1 m_2 = 0.\] Rearranging: \(m_1(m_2(a,b)) = m_2(m_1(a),b) + (-1)^{|a|} m_2(a, m_1(b))\). This says \(m_1\) is a derivation with respect to \(m_2\) — the product \(m_2\) is compatible with the differential.

\(n=3\): The \(n=3\) identity says that the associator \(m_2(m_2(a,b),c) - m_2(a,m_2(b,c))\) equals the boundary of \(m_3(a,b,c)\). Thus \(m_2\) is homotopy-associative, with \(m_3\) the explicit homotopy for associativity.

\(n=4\): The \(n=4\) identity says the pentagon: \(m_3\) satisfies its own coherence up to a homotopy given by \(m_4\).

7.3 Infinity-Morphisms

Definition 7.2 (\(\infty\)-Morphism). An \(\infty\)-morphism \(f: A \rightsquigarrow B\) of \(A_\infty\)-algebras is a collection of graded \(k\)-linear maps \[f_n: A^{\otimes n} \longrightarrow B, \quad n \geq 1,\] of degree \(1-n\), satisfying: for all \(n \geq 1\), \[\sum_{\substack{r+s+t=n \\ s \geq 1}} (-1)^{rs+t} f_{r+1+t}\!\left(\mathrm{id}^{\otimes r} \otimes m_s^A \otimes \mathrm{id}^{\otimes t}\right) = \sum_{\substack{k \geq 1 \\ i_1+\cdots+i_k=n}} (-1)^\epsilon\, m_k^B\!\left(f_{i_1} \otimes \cdots \otimes f_{i_k}\right),\] where \((-1)^\epsilon\) is a Koszul sign.

💡 Surprisingly, an \(\infty\)-morphism with \(f_n = 0\) for \(n \geq 2\) is precisely a strict morphism of \(A_\infty\)-algebras. The full notion is substantially more flexible: it allows morphisms that are only homotopy-compatible with the \(A_\infty\)-structures.

Definition 7.3 (\(\infty\)-Quasi-Isomorphism). An \(\infty\)-morphism \(f: A \rightsquigarrow B\) is an \(\infty\)-quasi-isomorphism if \(f_1: (A, m_1^A) \to (B, m_1^B)\) is a quasi-isomorphism of chain complexes.

7.4 Operadic Perspective

🔑 \(A_\infty\)-algebras fit into the operadic framework via the bar-cobar construction and Koszul duality.

The associative operad \(\mathrm{Ass}\) is Koszul, with Koszul dual cooperad \(\mathrm{Ass}^! = \mathrm{Ass}^{\scriptstyle\vee}\) (the linear dual). The cobar construction \(\Omega(\mathrm{Ass}^!)\) produces a dg-operad whose algebras are precisely \(A_\infty\)-algebras. This is made precise in Koszul Duality and the Bar Construction.

Proposition 7.4. An \(A_\infty\)-algebra structure on \(A\) is equivalent to a morphism of dg-operads \[\Omega(\mathrm{Ass}^!) \longrightarrow \mathrm{End}_A.\]

Heuristic sketch. The cobar construction \(\Omega(\mathrm{Ass}^!)\) is freely generated as an operad by elements \(m_n\) in arity \(n\) and degree \(2-n\), subject to the differential \(\partial(m_n) = \sum_{r+s+t=n} \pm m_{r+1+t} \circ_i m_s\) (the Stasheff relations arising from \(\partial^2 = 0\) in the cobar complex). An operad map \(\Omega(\mathrm{Ass}^!) \to \mathrm{End}_A\) is therefore exactly a choice of maps \(\{m_n \in \mathrm{End}_A(n)\}\) satisfying the Stasheff identities.

8. The Probability Operad Revisited

📐 We return to the probability operad \(\mathcal{P}\) introduced in Definitions and Examples (Section 5.7). Recall: \(\mathcal{P}(n)\) is the topological simplex \(\Delta_n = \{(p_1, \ldots, p_{n+1}) : p_i \geq 0, \sum p_i = 1\}\) (in the topological version of Bradley) or \(\mathcal{P}(n) = \{(p_1, \ldots, p_n) : p_i \geq 0, \sum p_i = 1\}\), with operad structure given by operadic composition of probability distributions.

8.1 P-Algebra Structure on the Positive Reals

Definition 8.1 (\(\mathcal{P}\)-Algebra). A \(\mathcal{P}\)-algebra is a set \(X\) (or \(k\)-module) equipped with structure maps \(\gamma_X(n): \mathcal{P}(n) \times X^n \to X\) satisfying the algebra axioms of Definition 1.1.

Example 8.2 (\([0, \infty)\) as a \(\mathcal{P}\)-algebra). The set \([0, \infty)\) (or \([0,1]\), or \(\mathbb{R}_{\geq 0}\)) becomes a \(\mathcal{P}\)-algebra via the weighted mean structure map: \[\gamma(p_1, \ldots, p_n;\, x_1, \ldots, x_n) := \sum_{i=1}^n p_i x_i.\] The algebra axioms reduce to standard properties of convex combinations: - Associativity: \(\sum_i p_i (\sum_j q_{ij} x_{ij}) = \sum_{i,j} p_i q_{ij} x_{ij}\), i.e., iterated weighted averages compose correctly. - Unitality: \(1 \cdot x = x\).

8.2 Shannon Entropy as a Derivation

🔑 The central result, proved by Bradley in Entropy as an Operad Derivation, is:

Theorem 8.3 (Bradley 2021). Shannon entropy \(H: \Delta_n \to \mathbb{R}\), defined by \[H(p_1, \ldots, p_n) = -\sum_{i=1}^n p_i \log p_i,\] is a derivation of the operad \(\mathcal{P}\) with values in the \(\mathcal{P}\)-bimodule \(\mathbb{R}\) (with the bimodule structure \(\gamma_\mathbb{R}(p; x) = \sum_i p_i x_i\)). Moreover, up to a positive constant, \(H\) is the unique such derivation.

The operadic Leibniz rule for \(H\) (Definition 5.3) gives: \[H(p \circ_i q) = H(p) + p_i H(q),\] where \(p \circ_i q\) denotes the operadic composition inserting distribution \(q\) in position \(i\) of distribution \(p\). This is precisely the chain rule for Shannon entropy.

Proof sketch. By the operadic Leibniz rule applied to the composition \(\gamma_\mathcal{P}(p; q^1, \ldots, q^n) = p \circ (q^1, \ldots, q^n)\): \[H(p \circ (q^1,\ldots,q^n)) = H(p) \cdot \gamma_\mathbb{R}(q^1, \ldots, q^n) + \sum_i p_i H(q^i).\] When only \(q^i \neq \mathrm{id}\) (i.e., all other \(q^j\) are pure point distributions), this reduces to \(H(p \circ_i q) = H(p) + p_i H(q)\). The uniqueness follows from Faddeev’s characterization theorem: any continuous function satisfying this chain rule is a constant multiple of \(H\). \(\square\)

8.3 Connection to BFL Entropy

The broader research thread on entropy and operads is developed in Entropy as an Operad Derivation and Categorical Entropy.

The BFL (Baez-Fritz-Leinster) entropy is a categorification of Shannon entropy using a monoidal functor from a category of finite probability spaces to \([0, \infty)\). The operadic derivation perspective illuminates the BFL result: the chain rule \(H(p \circ_i q) = H(p) + p_i H(q)\) is not a property of entropy but its definition — entropy is the unique (up to scaling) derivation of \(\mathcal{P}\).

The evaluation map of Proposition 5.4 connects these: the derivation of the operad \(\mathcal{P}\) (entropy as an operad-level map) descends, via evaluation on the \(\mathcal{P}\)-algebra \(([0,\infty), \gamma_\mathbb{R})\), to a classical derivation of the \(\mathcal{P}\)-algebra structure. This is the bridge between the operadic characterization and the classical information-theoretic one.

Algorithmic Applications

from itertools import product

def free_ass_algebra(r: int, max_degree: int) -> dict[int, list[tuple]]:
    """
    Returns graded components of T(k^r) up to max_degree.
    Each basis element is a tuple of indices in {1, ..., r}.
    """
    components = {}
    for n in range(max_degree + 1):
        if n == 0:
            components[0] = [()]  # unit element (empty word)
        else:
            components[n] = list(product(range(1, r + 1), repeat=n))
    return components

def multiply(u: tuple, v: tuple) -> tuple:
    """Multiplication in T(k^r) is concatenation."""
    return u + v

# Example: r=2, basis of degree 2 is [(1,1),(1,2),(2,1),(2,2)]
alg = free_ass_algebra(2, 3)
print(f"dim T(k^2)_3 = {len(alg[3])}")  # 8 = 2^3

import numpy as np
from numpy.polynomial import polynomial as P

def sym_sym_to_sym(poly_of_polys: list[list[float]]) -> list[float]:
    """
    Monad multiplication mu: Sym(Sym(V)) -> Sym(V).
    poly_of_polys: list of coefficient lists for each y_n.
    Element of Sym(Sym(k)) = k[y_0, y_1, ...] is a polynomial in the y_n's.
    We evaluate at y_n = x^n.

    Input: poly_of_polys[d] = coefficients of the degree-d monomial's
           polynomial-coefficient in x, i.e., the coefficient of y_0^{d_0} * y_1^{d_1}...
    Simplified: represent an element as sum_n c_n * y_n, a linear polynomial.
    Then mu(sum_n c_n * y_n) = sum_n c_n * x^n.
    """
    # For linear elements sum_n c_n * y_n:
    # result polynomial has coefficient c_n for x^n
    coeffs = poly_of_polys  # [c_0, c_1, c_2, ...]
    return coeffs  # direct substitution: y_n -> x^n

def monad_mult_nonlinear(outer_poly_coeffs, inner_polys):
    """
    General case: outer_poly is a polynomial in y-variables,
    inner_polys[n] = coefficients of x^n polynomial replacing y_n.
    Evaluate by substitution.
    """
    result = [0.0]
    for n, coeff in enumerate(outer_poly_coeffs):
        if abs(coeff) > 1e-12:
            x_to_n = [0.0] * n + [1.0]  # x^n
            term = [c * coeff for c in x_to_n]
            result = P.polyadd(result, term)
    return result

from sympy import symbols, Rational, diff, expand
from sympy import Symbol

x, y = symbols('x y')

def differential(f_expr):
    """
    Compute d(f) in Omega^1_{k[x,y]/(xy)/k}.
    Returns (coeff_dx, coeff_dy) representing df = coeff_dx*dx + coeff_dy*dy.
    Apply the relation xy=0 before computing partial derivatives.
    """
    # Partial derivatives
    df_dx = diff(f_expr, x)
    df_dy = diff(f_expr, y)
    return (df_dx, df_dy)

def apply_relation(coeff):
    """Apply xy = 0 in the quotient ring."""
    return expand(coeff).subs(x * y, 0)

def omega_element_add(w1, w2):
    """Add two Kahler differential elements."""
    return (apply_relation(w1[0] + w2[0]),
            apply_relation(w1[1] + w2[1]))

# The relation from d(xy) = 0: y*dx + x*dy = 0
# So in Omega^1, we have the relation (y, x) ~ 0
# This means x*dy = -y*dx, a key relation between the generators

# Example: d(x^2) = 2x*dx, d(y^2) = 2y*dy
print(differential(x**2))  # (2*x, 0) -- this is 2x*dx
print(differential(y**2))  # (0, 2*y) -- this is 2y*dy

def dga_to_ainfty(dim: int, d_matrix, mult_tensor):
    """
    dim: dimension of underlying graded vector space
    d_matrix: (dim x dim) matrix representing differential d
    mult_tensor: (dim x dim x dim) tensor representing multiplication

    Returns list of structure maps m_n for n >= 1.
    m_1 = d, m_2 = mult, m_n = 0 for n >= 3.
    """
    import numpy as np

    def m1(a):
        return d_matrix @ a

    def m2(a, b):
        # m2(a,b)_k = sum_{i,j} mult_tensor[k,i,j] * a[i] * b[j]
        return np.einsum('kij,i,j->k', mult_tensor, a, b)

    def m_n(n):
        if n >= 3:
            return lambda *args: np.zeros(dim)
        elif n == 2:
            return m2
        elif n == 1:
            return m1

    # Verify n=1 Stasheff: m1(m1(a)) = 0 (i.e., d^2 = 0)
    def verify_n1(a):
        return np.allclose(m1(m1(a)), 0)

    # Verify n=2: m1(m2(a,b)) = m2(m1(a),b) + (-1)^|a| m2(a,m1(b))
    # (sign depends on degree; simplified here for degree 0 inputs)
    def verify_n2(a, b):
        lhs = m1(m2(a, b))
        rhs = m2(m1(a), b) + m2(a, m1(b))
        return np.allclose(lhs, rhs)

    # Verify n=3: m3=0, so the identity reduces to:
    # m2(m2(a,b),c) - m2(a,m2(b,c)) = 0 (strict associativity)
    def verify_n3(a, b, c):
        return np.allclose(m2(m2(a,b),c), m2(a,m2(b,c)))

    return {'m1': m1, 'm2': m2, 'verify': (verify_n1, verify_n2, verify_n3)}

import numpy as np

def entropy(p: np.ndarray) -> float:
    """Shannon entropy H(p) = -sum p_i log p_i."""
    p = p[p > 0]  # avoid 0 * log(0) = 0 by convention
    return -np.sum(p * np.log(p))

def operad_compose(p: np.ndarray, q: np.ndarray, i: int) -> np.ndarray:
    """
    Compute p circ_i q: insert distribution q at position i of p.
    Result has len(p) - 1 + len(q) entries.
    """
    before = p[:i]
    p_i = p[i]
    after = p[i+1:]
    inserted = p_i * q
    return np.concatenate([before, inserted, after])

def verify_entropy_derivation(p: np.ndarray,
                               q: np.ndarray,
                               i: int,
                               tol: float = 1e-10) -> bool:
    """
    Verify H(p circ_i q) = H(p) + p[i] * H(q).
    """
    composed = operad_compose(p, q, i)
    lhs = entropy(composed)
    rhs = entropy(p) + p[i] * entropy(q)
    print(f"H(p ∘_i q) = {lhs:.8f}")
    print(f"H(p) + p_i*H(q) = {rhs:.8f}")
    print(f"Difference: {abs(lhs - rhs):.2e}")
    return abs(lhs - rhs) < tol

# Test: p = (0.5, 0.5), q = (0.3, 0.7), i = 0
p = np.array([0.5, 0.5])
q = np.array([0.3, 0.7])
result = verify_entropy_derivation(p, q, 0)
print(f"Leibniz rule holds: {result}")
# Expected: H(0.15, 0.35, 0.5) = H(0.5, 0.5) + 0.5 * H(0.3, 0.7)

References