🔢 mathematics

Monoidal bicategories, differential linear logic, and analytic functors

This paper advances the theory of monoidal bicategories by introducing bicategorical versions of linear exponential comonads and codereliction transformations, thereby extending Joyal's differential calculus of analytic functors from single-variable to multi-variable contexts between presheaf categories.

Original authors: M. Fiore, N. Gambino, M. Hyland

Published 2026-07-17

📖 7 min read🧠 Deep dive

CC BY 4.0

Original authors: M. Fiore, N. Gambino, M. Hyland

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are trying to understand how the universe is built, not just by looking at the bricks, but by looking at how the bricks talk to each other. In the world of mathematics, there is a field called category theory that does exactly this. Instead of studying individual numbers or shapes, it studies "objects" and the "maps" (or arrows) that connect them. Think of it like a massive, abstract subway map where the stations are ideas and the lines are the rules that let you travel between them.

For a long time, mathematicians have been interested in Linear Logic, a special set of rules for reasoning that treats information like a physical resource. In this logic, you can't just copy a piece of data or throw it away freely; you have to use it exactly once. This is like having a single ticket to a concert: you can't photocopy it for your friends, and you can't leave it in your pocket if you want to get in. To make this logic work, mathematicians use a tool called a comonad (a fancy machine that packages data) and a rule called differentiation (finding how things change).

Recently, scientists discovered that this logic of "using things once" is surprisingly similar to calculus. Just as you can take the derivative of a function to see how it changes, you can take the "derivative" of a logical process. This led to Differential Linear Logic, which allows us to do calculus on logic itself. But there was a problem: all these rules were built for flat, one-dimensional maps. The real world of complex systems often feels more like a 3D sculpture than a flat drawing. This paper asks: What happens if we take these rules and lift them up into a higher dimension, where the connections between connections matter?

The Big Idea: From Flat Maps to 3D Sculptures

This paper, written by Marcelo Fiore, Nicola Gambino, and Martin Hyland, is like a construction manual for upgrading the mathematical "operating system" of logic. The authors take the existing, well-understood rules of Linear Logic and Differential Linear Logic and rebuild them to work in a monoidal bicategory.

In plain English, a bicategory is a mathematical structure where you have objects, maps between them, and also maps between the maps (called 2-cells). If a normal category is a flat subway map, a bicategory is a subway map where the trains themselves can change tracks, merge, or split in specific ways while they are moving. The authors are essentially saying, "Let's stop treating our logical rules as rigid, flat lines and start treating them as flexible, 3D structures."

The New Tools: "Pseudo" Machines and "Codereliction"

To build this new 3D world, the authors invent two main tools:

Linear Exponential Pseudocomonads: Think of a standard "comonad" as a machine that takes a piece of data and wraps it in a special box (the "!" box) that says, "You can use this data as many times as you want, but only after you've processed it." In the flat world, this machine is rigid. In this paper's 3D world, the machine becomes a "pseudocomonad." It's like a smart, squishy box that can stretch and twist. It still does the same job (packaging data for logic), but it has extra "wiggle room" to handle the complex 3D connections. The authors prove that even with this wiggle room, the machine still works perfectly for Linear Logic.
Codereliction: This is the star of the show. In the flat world, "codereliction" is a specific rule that lets you take the derivative of a logical process. It's like a magic wand that turns a "non-linear" process (one that uses data freely) into a "linear" one (one that uses data carefully) to see how it changes. The authors introduce a bicategorical codereliction. Imagine you have a complex, twisting knot of logic. The codereliction is a tool that lets you gently pull on one strand to see how the whole knot reacts, even when the knot is moving and changing shape in 3D space.

The Application: Analytic Functors as "Many-Variable" Functions

The most exciting part of the paper is what they do with these new tools. They apply them to analytic functors.

In the past, mathematician André Joyal showed that certain types of functions (called analytic functors) on sets could be treated like polynomials or Taylor series in calculus. You could take their derivative to see how they changed. However, this only worked for functions with one variable (like $f(x)$ ).

The authors use their new 3D logic tools to extend this to many variables. They show that you can take the derivative of a function that depends on a whole category of inputs, not just a single number.

The Analogy: Imagine you have a recipe for a cake. In the old world, you could only ask, "What happens if I add one more egg?" (one variable). In this new 3D world, the authors show you can ask, "What happens if I change the flour, the sugar, the oven temperature, and the mixing speed all at once, and how do those changes interact?"
They prove that the rules of calculus (like the product rule and the chain rule) still hold true in this complex, multi-dimensional setting. They call these new functions "analytic functors between presheaf categories," which is just a fancy way of saying "complex, multi-variable logical recipes."

What They Found and What They Didn't

The paper proves (it's a mathematical proof, not a guess) that:

You can define these "squishy" 3D machines (pseudocomonads) and they behave exactly as you'd hope for logic.
You can define the 3D "magic wand" (codereliction) and it successfully calculates derivatives for these complex logical recipes.
The famous rules of calculus (Leibniz rule, chain rule) work perfectly in this new 3D environment.

The paper does not claim to have solved every problem in mathematics. It explicitly notes that while they have built a robust framework, there are still "coherence issues" (technical details about how the 3D shapes fit together perfectly) that are tricky. They don't claim this is the only way to do it, but they show it is a valid and powerful way. They also don't claim this solves problems in physics or computer science directly yet; they are building the theory that might one day help those fields. They are laying the foundation, not building the skyscraper.

Why Should You Care?

You might wonder, "Who cares about 3D logic?" The authors suggest that this is crucial for theoretical computer science. In computer programming, we often need to model how code rewrites itself or how different parts of a program interact. The "2-cells" (the maps between maps) in this 3D logic are perfect for modeling these rewriting steps.

By showing that we can do calculus on these complex, 3D logical structures, the authors have opened the door to a new kind of "quantitative semantics." This means we might one day be able to mathematically predict how complex software systems will behave, change, or break, using the same tools we use to predict how a ball rolls down a hill. They have taken the abstract idea of "differentiation" and shown it works even when the world is as messy and interconnected as a 3D sculpture.

Technical Summary: Monoidal Bicategories, Differential Linear Logic, and Analytic Functors

Problem and Context
This paper addresses the gap between low-dimensional category theory (specifically the study of monoidal bicategories) and differential linear logic. While the theory of linear logic models (based on symmetric monoidal categories with a linear exponential comonad) and differential categories (which introduce differentiation via codereliction) are well-established in one dimension, their extension to two dimensions remains fragmented. Existing results in the two-dimensional context, particularly regarding the bicategory of profunctors ( $\mathbf{Prof}$ ) and analytic functors, have often been proved on a case-by-case basis. The authors aim to establish a unified, rigorous bicategorical framework that recovers these results as special cases and extends the differential calculus of Joyal's analytic functors from sets to presheaf categories.

Methodology
The authors employ a "categorification" strategy, lifting standard one-dimensional concepts from linear logic and differential categories to the setting of symmetric monoidal bicategories (specifically, symmetric Gray monoids, justified by strictification theorems). The methodology proceeds through the following stages:

Foundational Definitions: The paper defines linear exponential pseudocomonads as the bicategorical counterpart to linear exponential comonads. This involves equipping a pseudocomonad with a symmetric lax monoidal structure and ensuring that the 2-category of its pseudocoalgebras admits a cartesian monoidal structure.
Structural Analysis: The authors investigate the consequences of additional structural assumptions on the ambient bicategory, specifically the existence of finite products and biproducts. They construct Seely equivalences (isomorphisms relating the tensor product of cofree coalgebras to the cofree coalgebra of a product) and define cocontraction and coweakening maps.
Differentiation: A bicategorical codereliction transformation is defined. This is a 2-categorical version of the creation map used in differential categories. The definition is carefully formulated to handle coherence issues by requiring specific 2-cells (derived from adjunctions and structural maps) to be invertible, rather than imposing complex coherence axioms on non-invertible cells.
Application to Profunctors: The theory is applied to the bicategory of profunctors ( $\mathbf{Prof}$ ). The authors demonstrate that the 2-monad for free symmetric strict monoidal categories on $\mathbf{Cat}$ extends to a pseudomonad on $\mathbf{Prof}$ , which, via duality, yields a linear exponential pseudocomonad. They verify that $\mathbf{Prof}$ satisfies the necessary hypotheses (biproducts, adjoints for counits) to admit a codereliction.

Key Contributions

Linear Exponential Pseudocomonads: The paper introduces and studies the notion of a linear exponential pseudocomonad (Definition 4.12). It proves that the 2-category of pseudocoalgebras for such a structure is a cartesian closed bicategory (Theorem 5.7), generalizing the standard result for linear logic models.
Seely Equivalences: The authors construct Seely equivalences as a sylleptic strong monoidal 2-functor (Theorem 5.1), providing a modular approach to handling coherence in the presence of finite products.
Bicategorical Codereliction: A definition of codereliction is provided (Definition 7.5) that effectively manages coherence by leveraging invertible 2-cells arising from adjunctions. This allows for the derivation of standard differentiation rules (constant, product, chain, and monoidal rules) in the bicategorical setting (Proposition 7.6).
Extension of Analytic Functors: The paper establishes that the bicategory of categorical symmetric sequences (the Kleisli bicategory of the pseudocomonad on $\mathbf{Prof}$ ) admits a codereliction (Theorem 8.6). This allows for the definition of a derivative for analytic functors between presheaf categories.

Results

Theoretical Consistency: The authors prove that the relational model of linear logic and its categorification via profunctors fit into a single theoretical framework. The relational model is recovered as a special case where the bicategory is the category of sets and relations.
Differentiation of Analytic Functors: The derivative of an analytic functor $F: \mathbf{Psh}(A) \to \mathbf{Psh}(B)$ is explicitly defined. For a categorical symmetric sequence $F: !A \to B$ , the derivative $dF$ is a profunctor $A \times !A \to B$ given by the formula $dF(b, (a, \alpha)) = F(b, \alpha \sqcup \langle a \rangle)$ .
Calculus Rules: The paper demonstrates that the axioms of codereliction imply counterparts to the fundamental rules of differential calculus (Leibniz rule, chain rule, etc.) for analytic functors between presheaf categories. This generalizes Joyal's single-variable calculus for analytic functors on $\mathbf{Set}$ to a many-variable setting.

Significance and Claims
The paper claims to provide the first example of a "genuinely two-dimensional model of differential linear logic." Its significance lies in:

Unification: It connects low-dimensional category theory with differential linear logic, offering a precise basis for the analogy between the relational model and the profunctor model.
Generalization: It extends the differential calculus of analytic functors from sets to presheaf categories, analogous to how single-variable calculus extends to multivariable calculus.
Coherence Management: The approach to defining codereliction sidesteps difficult coherence questions by utilizing invertible 2-cells derived from adjunctions, offering a potential guide for future formulations of coherence conditions in higher-dimensional differential structures.

The authors note that while their theory uses non-invertible 2-cells crucially and is not immediately subsumed by $(\infty, 1)$ -category theory, it may be possible to develop an $(\infty, 2)$ -categorical counterpart in the future. The work is presented as a foundational step, establishing the necessary definitions and proving their consistency within the context of monoidal bicategories.