Monoidal Coalgebra Modality

• Monoidal coalgebra modality is a categorical framework that models the exponential modality in linear logic, combining comonadic, comonoid, and monoidal structures to ensure resource sensitivity.
• It facilitates the lifting of coalgebra modalities through monads and supports differential structure with coderelictions, deriving transformations, and integral transformations.
• Its algebraic formulation bridges linear logic semantics with additive bialgebra modalities, underpinning models such as Lafont categories and differential linear logic.

A monoidal coalgebra modality is the categorical structure that interprets the exponential modality $!$ of multiplicative–exponential linear logic in a symmetric monoidal setting. Concretely, it combines a comonad $(!,\delta,\varepsilon)$ with a natural cocommutative comonoid structure on each $!A$, together with monoidal structure maps $m_{A,B}:!A\otimes !B\to !(A\otimes B)$ and $m_I:I\to !I$ satisfying coherence, monoidality, and coalgebra-compatibility axioms. In the literature it is also called a linear exponential comonad, and in additive settings it is tightly related to bialgebra modalities, coderelictions, deriving transformations, and models of differential linear logic (Lemay, 2018, Blute et al., 2018).

1. Formal definition and core axioms

Let $(\mathcal{C},\otimes,I)$ be a symmetric monoidal category. A linear category modeling $\mathsf{MELL}$ is a symmetric monoidal closed category $(\mathcal{C},\otimes,I)$ with internal hom $[-,-]$ and evaluation $\mathrm{ev}$, equipped with a monoidal coalgebra modality $(!,\delta,\varepsilon)$0 (Lemay, 2018).

The underlying comonad consists of an endofunctor $(!,\delta,\varepsilon)$1 together with natural transformations

$(!,\delta,\varepsilon)$2

satisfying the comonad laws. In the formulation used in the differential-category literature, these include

$(!,\delta,\varepsilon)$3

as recorded in the revisited treatment of differential categories (Blute et al., 2018).

A coalgebra modality equips each $(!,\delta,\varepsilon)$4 with natural maps

$(!,\delta,\varepsilon)$5

such that $(!,\delta,\varepsilon)$6 is a cocommutative comonoid, and $(!,\delta,\varepsilon)$7 is a comonoid morphism. One standard form of this compatibility is

$(!,\delta,\varepsilon)$8

or, equivalently in the monoidal presentations, the corresponding equations written through the lifted monoidal structure (Lemay, 2018, Blute et al., 2018, Cockett et al., 2017).

The monoidal enhancement adds natural transformations

$(!,\delta,\varepsilon)$9

making $!A$0 a symmetric monoidal endofunctor or symmetric monoidal comonad. The basic coherence conditions are associativity, unit, and symmetry: $!A$1

$!A$2

$!A$3

in the notation of the additive and differential accounts (Blute et al., 2018, Lemay, 19 Feb 2025).

A monoidal coalgebra modality is obtained when the comonad structure, the monoidal structure, and the comonoid structure are all compatible. In particular, $!A$4 and $!A$5 are monoidal transformations, equivalently $!A$6 and $!A$7 are comonoid morphisms. Typical identities are

$!A$8

together with the corresponding unit equalities (Lemay, 2018, Blute et al., 2018, Lemay, 19 Feb 2025).

Semantically, $!A$9 encodes dereliction, $m_{A,B}:!A\otimes !B\to !(A\otimes B)$0 encodes promotion or digging, $m_{A,B}:!A\otimes !B\to !(A\otimes B)$1 gives contraction, and $m_{A,B}:!A\otimes !B\to !(A\otimes B)$2 gives weakening. The monoidality of $m_{A,B}:!A\otimes !B\to !(A\otimes B)$3 and $m_{A,B}:!A\otimes !B\to !(A\otimes B)$4 expresses that these resource-sensitive operations are coherent with tensor, so that $m_{A,B}:!A\otimes !B\to !(A\otimes B)$5 behaves as the exponential “of course” of linear logic (Lemay, 2018).

2. Linear logic, Seely structure, and additive bialgebra modalities

Monoidal coalgebra modalities are standard semantics for the exponential fragment of linear logic, and in symmetric monoidal categories with finite products they are closely related to Seely isomorphisms. A coalgebra modality has Seely isomorphisms when

$m_{A,B}:!A\otimes !B\to !(A\otimes B)$6

are isomorphisms, with

$m_{A,B}:!A\otimes !B\to !(A\otimes B)$7

Under finite products, monoidal coalgebra modalities and Seely-style storage structure determine one another; monoidal storage categories are precisely symmetric monoidal categories with finite products and a coalgebra modality having Seely isomorphisms, and these are equivalent to linear categories (Blute et al., 2018).

In additive symmetric monoidal categories, monoidal coalgebra modalities are equivalent to additive bialgebra modalities. A bialgebra modality augments each $m_{A,B}:!A\otimes !B\to !(A\otimes B)$8 with a commutative monoid structure

$m_{A,B}:!A\otimes !B\to !(A\otimes B)$9

such that $m_I:I\to !I$0 is a commutative bialgebra object, together with additive compatibility

$m_I:I\to !I$1

Theorem 3.7 in the revisited differential-category treatment states that, in an additive symmetric monoidal category, additive linear categories are equivalent to additive symmetric monoidal categories with additive bialgebra modality (Blute et al., 2018).

The passage from monoidal coalgebra modality to additive bialgebra modality is explicit. For example,

$m_I:I\to !I$2

$m_I:I\to !I$3

and conversely one reconstructs $m_I:I\to !I$4 and $m_I:I\to !I$5 from $m_I:I\to !I$6, $m_I:I\to !I$7, and the coalgebra data (Blute et al., 2018).

This equivalence is central for differential linear logic. In additive linear categories there is only one notion of differentiation: coderelictions, deriving transformations, and creation operators coincide. The main theorem of “Differential Categories Revisited” states that in additive linear categories, equivalently in settings with monoidal coalgebra modalities, coderelictions and deriving transformations are in bijection and satisfy the monoidal strength law (Blute et al., 2018). A later development sharpens this by showing that even when additive enrichment is not postulated a priori, a monoidal bialgebra modality equipped with a pre-codereliction induces additive enrichment by bialgebra convolution; this yields a characterization of a differential linear category as a symmetric monoidal category with a monoidal bialgebra modality and a codereliction (Lemay, 19 Feb 2025).

3. Lifting to Eilenberg–Moore categories and Hopf-monadic transport

A central structural result is that monoidal coalgebra modalities can be lifted along suitable monads to Eilenberg–Moore categories. The paper “Lifting Coalgebra Modalities and $m_I:I\to !I$8 Model Structure to Eilenberg-Moore Categories” studies categories of algebras of monads as models of $m_I:I\to !I$9 and introduces the relevant compatibility notion: a symmetric monoidal mixed distributive law

$(\mathcal{C},\otimes,I)$0

between a symmetric comonoidal monad $(\mathcal{C},\otimes,I)$1 and a symmetric monoidal comonad $(\mathcal{C},\otimes,I)$2 (Lemay, 2018).

A mixed distributive law must satisfy the Beck-style compatibility with monad multiplication and unit and comonad comultiplication and counit. In the notation of that paper, the defining equations include

$(\mathcal{C},\otimes,I)$3

$(\mathcal{C},\otimes,I)$4

and a corresponding compatibility with $(\mathcal{C},\otimes,I)$5 (Lemay, 2018).

When both sides carry monoidal structure, $(\mathcal{C},\otimes,I)$6 must also respect $(\mathcal{C},\otimes,I)$7 and $(\mathcal{C},\otimes,I)$8: $(\mathcal{C},\otimes,I)$9 together with the unit equation involving $\mathsf{MELL}$0 and $\mathsf{MELL}$1 (Lemay, 2018). Proposition (liftsymmix) establishes that such laws are in bijection with liftings of symmetric monoidal comonads to Eilenberg–Moore categories of symmetric comonoidal monads, and dually (Lemay, 2018).

For a coalgebra modality $\mathsf{MELL}$2, a mixed law is coalgebraic when each component

$\mathsf{MELL}$3

is a comonoid morphism, where

$\mathsf{MELL}$4

A key simplification is Lemma (biglemma): for a monoidal coalgebra modality, any symmetric monoidal mixed distributive law is automatically coalgebraic (Lemay, 2018).

The main theorem then states that if $\mathsf{MELL}$5 is a $\mathsf{MELL}$6 lifting monad—defined as an exponential lifting monad whose underlying symmetric comonoidal monad is a Hopf monad in the sense of Bruguières–Lack–Virelizier—then the Eilenberg–Moore category $\mathsf{MELL}$7 is again a linear category, and the forgetful functor preserves the full linear structure strictly (Lemay, 2018). The proof factors through three ingredients: symmetric comonoidal monads lift the symmetric monoidal structure, Hopf monads lift the closed structure, and symmetric monoidal mixed distributive laws lift the monoidal coalgebra modality.

The lifted exponential is explicit. For a $\mathsf{MELL}$8-algebra $\mathsf{MELL}$9, define

$(\mathcal{C},\otimes,I)$0

Then

$(\mathcal{C},\otimes,I)$1

and the lifted comultiplication, counit, comonoid structure, and monoidal maps are inherited componentwise from $(\mathcal{C},\otimes,I)$2, $(\mathcal{C},\otimes,I)$3, $(\mathcal{C},\otimes,I)$4, $(\mathcal{C},\otimes,I)$5, $(\mathcal{C},\otimes,I)$6, and $(\mathcal{C},\otimes,I)$7 (Lemay, 2018).

This transport theorem generalizes the Blute–Scott program from multiplicative linear logic to $(\mathcal{C},\otimes,I)$8: categories of modules over Hopf monoids can inherit full linear structure, not only multiplicative structure, provided the relevant mixed distributive law over $(\mathcal{C},\otimes,I)$9 is available (Lemay, 2018).

4. Differential, integral, and calculus structures

A monoidal coalgebra modality is not itself a differential modality, but it is the ambient structure in which differential linear logic becomes natural. In an additive symmetric monoidal category with coalgebra modality, a deriving transformation is a natural family

$[-,-]$0

satisfying constant, Leibniz, linear, chain, and interchange rules (Blute et al., 2018, Cockett et al., 2017). For example, the Leibniz and linear rules are written as

$[-,-]$1

$[-,-]$2

in the revisited formulation (Blute et al., 2018).

In the monoidal setting, a codereliction

$[-,-]$3

is equivalent to a deriving transformation. The correspondence is given by

$[-,-]$4

for additive bialgebra modalities, and Theorems 5.8, 5.10, and 5.11 show that in additive linear categories every deriving transformation automatically satisfies the extra $[-,-]$5-rule, so coderelictions and deriving transformations are in bijection (Blute et al., 2018). A more recent refinement proves that coderelictions are unique and that the additive enrichment itself can be recovered from bialgebra convolution: $[-,-]$6 for parallel maps $[-,-]$7 (Lemay, 19 Feb 2025).

The interaction with integration is developed in the theory of integral and calculus categories. An integral transformation is a natural map

$[-,-]$8

satisfying the integral-of-constants rule, a Rota–Baxter rule, and an interchange rule (Cockett et al., 2017). When a differential category also has an integral transformation satisfying the two categorical fundamental theorems of calculus, one obtains a calculus category (Cockett et al., 2017).

The monoidal case is especially rigid. If the coalgebra modality is monoidal, it is natural to require the monoidal coherence

$[-,-]$9

where $\mathrm{ev}$0 is the coderiving transformation (Cockett et al., 2017). Under this coherence, the integral structure is uniquely determined by the differential structure: in a calculus category with a monoidal coalgebra modality, the integral transformation is necessarily

$\mathrm{ev}$1

and any two monoidal integral transformations compatible with the same differential structure coincide (Cockett et al., 2017).

These developments also delimit the concept. “Differential Categories Revisited” emphasizes two separating examples: differential algebras provide a monoidal coalgebra modality but never induce a differential category, while Rota–Baxter algebras provide a differential category whose coalgebra modality is non-monoidal (Blute et al., 2018). A common misconception is therefore that monoidality of $\mathrm{ev}$2 is either sufficient or necessary for differential structure. The cited examples show that neither implication holds in general.

5. Hopf, group, and additive sources of monoidal coalgebra modalities

One major source of lifting monads compatible with a monoidal coalgebra modality comes from group objects in the Eilenberg–Moore category of $\mathrm{ev}$3-coalgebras. If $\mathrm{ev}$4 is a group object in $\mathrm{ev}$5, with multiplication, unit, and inverse maps all coalgebra morphisms, then the monad

$\mathrm{ev}$6

carries a symmetric comonoidal structure induced by the comonoid structure of $\mathrm{ev}$7, and there is a canonical mixed distributive law

$\mathrm{ev}$8

(Lemay, 2018). When the inverse $\mathrm{ev}$9 equips $(!,\delta,\varepsilon)$00 with a Hopf monoid structure, the monad $(!,\delta,\varepsilon)$01 becomes a Hopf monad, so one obtains a $(!,\delta,\varepsilon)$02 lifting monad (Lemay, 2018).

In a linear category this yields a bijective correspondence between group objects in $(!,\delta,\varepsilon)$03 and cocommutative Hopf monoids in $(!,\delta,\varepsilon)$04 equipped with a compatible symmetric monoidal mixed distributive law; the resulting module category

$(!,\delta,\varepsilon)$05

is again linear (Lemay, 2018). In the special case of Lafont categories, where the exponential modality is free, $(!,\delta,\varepsilon)$06 is equivalent to the category of cocommutative comonoids in $(!,\delta,\varepsilon)$07, so cocommutative Hopf monoids directly produce new Lafont categories of modules (Lemay, 2018).

Additive enrichment with negatives supplies another canonical source. If $(!,\delta,\varepsilon)$08 is additive symmetric monoidal with negatives, then each cofree coalgebra $(!,\delta,\varepsilon)$09 carries a canonical commutative/cocommutative bimonoid structure

$(!,\delta,\varepsilon)$10

and a canonical antipode

$(!,\delta,\varepsilon)$11

Proposition (hopfneg) characterizes precisely when $(!,\delta,\varepsilon)$12 is a Hopf monoid: this happens iff $(!,\delta,\varepsilon)$13 admits negatives (Lemay, 2018). Consequently, the monads

$(!,\delta,\varepsilon)$14

are $(!,\delta,\varepsilon)$15 lifting monads and the corresponding module categories are linear (Lemay, 2018).

The Hopf enhancement also appears in the non-additive codereliction-based theory. A monoidal Hopf coalgebra modality is a monoidal bialgebra modality equipped with an antipode

$(!,\delta,\varepsilon)$16

such that each $(!,\delta,\varepsilon)$17 is a Hopf monoid. In that setting, a pre-codereliction induces Abelian-group enrichment, with additive inverse given by the convolution formula

$(!,\delta,\varepsilon)$18

(Lemay, 19 Feb 2025). This suggests that antipodes are the precise categorical mechanism by which additive enrichment upgrades to enrichment over Abelian groups.

6. Variants, applications, and later extensions

The standard theory is symmetric, but the concept extends beyond symmetric monoidal categories. “Linear Exponential Comonads without Symmetry” develops the non-symmetric analogue on a general monoidal category. There one has a monoidal comonad together with comonoid maps

$(!,\delta,\varepsilon)$19

and a derived family

$(!,\delta,\varepsilon)$20

constructed from $(!,\delta,\varepsilon)$21 and required to satisfy axioms ensuring that the category of coalgebras is cartesian (Hasegawa, 2017). In this setting exchange is unavailable at the multiplicative level, but a controlled symmetry persists inside the image of $(!,\delta,\varepsilon)$22. The main characterization states that a monoidal comonad is a linear exponential comonad iff the induced monoidal structure on the category of coalgebras is cartesian (Hasegawa, 2017).

A related non-symmetric application appears in categorical semantics for Lambek calculus with a relevant modality. There the ambient category is monoidal biclosed rather than symmetric monoidal closed, and the modality is used to model controlled contraction and limited permutation. The categorical structure used is a monoidal biclosed category with a coalgebra modality “very similar to the structure of a Differential Category,” together with restricted symmetry only for $(!,\delta,\varepsilon)$23-marked objects (McPheat et al., 2020). The rule interpretations are explicit: $(!,\delta,\varepsilon)$24

$(!,\delta,\varepsilon)$25

and permutation is interpreted via the restricted symmetry $(!,\delta,\varepsilon)$26 (McPheat et al., 2020). In finite-dimensional vector spaces the paper studies three concrete interpretations of the modality, including identity-based and exterior-algebra-based constructions, and uses them for parasitic-gap semantics (McPheat et al., 2020).

A further extension concerns free completion to differential structure. “Free differential modalities” distinguishes monoidal coalgebra modalities from monoidal differential modalities and proves that, in a suitably well-behaved $(!,\delta,\varepsilon)$27-linear symmetric monoidal category with finite biproducts and algebraically-free commutative monoids, every monoidal coalgebra modality can be freely completed to a monoidal differential modality (Garner et al., 20 Aug 2025). The free completion has the form

$(!,\delta,\varepsilon)$28

with

$(!,\delta,\varepsilon)$29

$(!,\delta,\varepsilon)$30

(Garner et al., 20 Aug 2025). In particular, there exists an initial monoidal differential modality, and in $(!,\delta,\varepsilon)$31 this produces a new model of differential linear logic from the initial monoidal coalgebra modality $(!,\delta,\varepsilon)$32 (Garner et al., 20 Aug 2025).

Taken together, these developments place monoidal coalgebra modalities at the center of a broad categorical program. They serve as the semantics of the exponential modality of linear logic; under additivity they coincide with bialgebra modalities; under suitable mixed distributive laws they lift through Eilenberg–Moore categories; with coderelictions they support differential linear logic; with monoidal integral structure they determine calculus structure uniquely; in non-symmetric settings they support relevant and noncommutative logics; and under algebraically-free completion they generate differential modalities functorially (Lemay, 2018, Blute et al., 2018, Cockett et al., 2017, Hasegawa, 2017, Lemay, 19 Feb 2025, Garner et al., 20 Aug 2025).