Approaching the Continuous from the Discrete: an Infinite Tensor Product Construction

Basic Information

• Paper ID: 2510.14716
• Title: Approaching the Continuous from the Discrete: an Infinite Tensor Product Construction
• Authors: Antonio Lorenzin (a.lorenzin.95@gmail.com), Fabio Zanasi (University College London)
• Classification: math.CT (Category Theory), cs.LO (Logic in Computer Science)
• Publication Date: October 16, 2025 (arXiv preprint)
• Paper Link: https://arxiv.org/abs/2510.14716

Abstract

In recent years, probabilistic computation has increasingly been studied through the perspective of categorical algebra, particularly via string diagram calculus. While categories of discrete and Gaussian probabilistic processes have been extensively studied with various axiomatization results, more expressive continuous probability categories lack understanding due to inherent difficulties in describing infinite behavior through algebraic methods.

This paper establishes a general construction for adjoining infinite tensor products, enabling continuous probability to be studied from discrete settings. The main result applies this construction to FinStoch (the category of finite sets and stochastic matrices), yielding a category of locally constant Markov kernels whose objects are finite sets together with the Cantor space $2^{\mathbb{N}}$. Any probability measure on the real numbers can be reasoned about within this category. Furthermore, the paper demonstrates how to lift axiomatization results through the infinite tensor product construction, obtaining an axiomatization of continuous probability on countable powers of $2=\{0,1\}$.

Research Background and Motivation

Problem Background

The application of categorical methods to probabilistic computation has attracted widespread attention in recent years, with applications ranging from evidential decision theory to random graphs and active inference. These methods highlight the underlying algebraic structure, provide rigorous semantics, enhance formal clarity and compositional methodology, and allow intuitive description via string diagrams.

Core Problem

Although categories of discrete probability (such as FinStoch and BinStoch) and Gaussian probability have complete axiomatizations, string diagram axiomatization of continuous probability remains a fundamental gap. The main challenge lies in how to directly encode infinite behavior within the existing string diagram algebraic framework.

Research Motivation

1. Theoretical requirement: Need for a complete categorical description of continuous probability
2. Methodological innovation: Connecting discrete and continuous probability through infinite tensor products
3. Practical value: Providing algebraic tools for continuous probability reasoning

Core Contributions

1. General construction: Introduces a general construction for adjoining infinite tensor products to any semicartesian category (Theorem 1)
2. Diagrammatic representation: Provides diagrammatic representation for freely generated categories with infinite tensor products using plate notation
3. FinStoch⊗∞ characterization: Characterizes FinStoch⊗∞ using Stone spaces and locally constant Markov kernels (Theorem 2)
4. Axiomatization: Provides an axiomatization of CantorStoch^{lc}, the restriction of FinStoch⊗∞ to powers of 2 and the Cantor space 2^ℕ (Corollary 2)

Methodology Details

Theoretical Foundation: Semicartesian Categories

Definition 1: A semicartesian category is a symmetric monoidal category (C,⊗,I) where the monoidal unit I is a terminal object.

Key examples:

• FinStoch: Objects are finite sets, morphisms are stochastic functions f: X → Y
• BinStoch: Subcategory of FinStoch with objects being finite powers of 2={0,1}
• BorelStoch: Objects are standard Borel spaces, morphisms are Markov kernels

Definition of Infinite Tensor Products

Definition 2: An abstract infinite tensor product is a functor X: P_(J)^{op} → C, mapping finite subsets F to X_F := ⊗_{j∈F} X_j.

Definition 3: A concrete infinite tensor product X = ⊗_{j∈J} X_j is the limit of the corresponding abstract infinite tensor product, and this limit is preserved by -⊗Y.

General Construction C⊗∞

Core idea: Define morphisms via compatible families satisfying:

• Naturality: Commute with projection morphisms
• Coverage condition: Every target finite subset has a corresponding source finite subset
• Hereditarity: If (F,G) is in the family, then all (F',G) with F' ⊇ F are also in the family

Definition 8: A morphism f: X → Y in C⊗∞ is an equivalence class of compatible families, with composition defined pointwise:

(gf)_{F,H} := g_{G,H} f_{F,G}

Plate Notation

To intuitively represent compatible families, plate notation is introduced:

f_{F,G}
─────────  represents morphism f: X → Y
(F,G) ∈ Λ_f
  X    Y

This notation is similar to plate notation in Bayesian networks but specifically designed for representing compatible families.

Main Theorems

Theorem 1: Universal Property

For a semicartesian category C with erasure deletion, there exists a semicartesian category C⊗∞ with infinite tensor products and a strict symmetric monoidal functor C → C⊗∞ such that for any semicartesian category D with infinite tensor products and symmetric monoidal functor φ: C → D, there exists a unique ITP-preserving symmetric monoidal functor φ̃: C⊗∞ → D making the diagram commute.

Theorem 2: Stone Space Characterization

The ITP-preserving symmetric monoidal functor

φ: FinStoch⊗∞ → StoneStoch^{lc}

is fully faithful, with its essential image being infinite tensor products of finite sets in the topological sense.

Locally constant Markov kernels (Definition 11): A morphism f: X → Y between Stone spaces satisfies:

• f(U|-): X → 0,1 is locally constant for all clopen sets U
• f(-|x): Clopen(Y) → 0,1 is a finitely additive probability measure for all x∈X

Experimental Setup and Applications

Markov Chain Case Study

The paper demonstrates the application of plate notation through Markov chains. Time-homogeneous Markov chains can be represented as:

c_n: X → X^n := f ∘ c_{n-1}

In the infinite tensor product setting, one can define infinite Markov chains c: X → X^ℕ and prove their invariance under prepending steps:

c_n = c_{n-1} ∘ f = c ∘ f

Experimental Results

Expressiveness

Key result: CantorStoch^{lc} contains all probability measures on ℝ. This is because ℝ is an infinite tensor product of 2 in BorelStoch, while 2^ℕ is the corresponding object in CantorStoch^{lc}.

Axiomatization Achievement

Corollary 3: CantorStoch^{lc} is isomorphic to Free^∞(Σ,E), where (Σ,E) is the symmetric monoidal theory of CausCirc.

This means this continuous probability category can be completely characterized using finitely many generators and equations.

Related Work

Categorical Probability Theory

• Markov category theory by Fritz et al. 11
• Axiomatization of discrete probability 21
• String diagrams for Gaussian probability 25

Infinite Tensor Products

• Pioneering work by Fritz and Rischel 16
• Applications in categorical probability 13,14

Innovation of This Paper

Compared to existing work, this paper provides for the first time a systematic construction method from discrete to continuous probability with complete axiomatization.

Conclusions and Discussion

Main Conclusions

1. Establishes a general construction for adjoining infinite tensor products
2. Proves that FinStoch⊗∞ can be characterized as locally constant Markov kernels
3. Provides complete axiomatization of continuous probability
4. Demonstrates a systematic approach from discrete to continuous

Limitations

1. Expressiveness constraints: Cannot recover complete BorelStoch because the degrees of freedom of measurable functions cannot be fully captured through finite operations
2. Locally constant constraint: Can only handle locally constant Markov kernels, not all continuous probabilistic processes
3. Countability restriction: Construction only applies to countably infinite tensor products

Future Directions

1. Extension to general Markov kernels: Investigate how to describe general kernels using locally constant kernels
2. Measure decomposition: Study general constructions guaranteeing conditional existence in Markov categories
3. Stone duality: Study probabilistic morphisms between Boolean algebras via Stone duality
4. Other categories: Apply to Gaussian probability, Gaussian mixtures, and other categories

In-Depth Evaluation

Strengths

1. Theoretical innovation: First systematic connection between categorical descriptions of discrete and continuous probability
2. General construction: Provided general construction applicable to all semicartesian categories
3. Practical tools: Plate notation provides intuitive tools for complex infinite tensor product reasoning
4. Deep results: Proves that Cantor space suffices to represent all probability measures on ℝ

Weaknesses

1. Limited applications: Markov chain case study is relatively simple, requiring more complex applications for demonstration
2. Computational complexity: Does not discuss computational complexity in practical settings
3. Implementation details: Lacks concrete algorithm implementations and computational tools

Impact

1. Theoretical contribution: Provides important theoretical tools for categorical probability theory
2. Methodological significance: Opens new methods for approximating continuous from discrete
3. Application prospects: Provides new theoretical foundations for probabilistic programming and machine learning

Applicable Scenarios

1. Theoretical research: Category theory, probability theory, logic research
2. Probabilistic programming: Provides more rigorous semantic foundations
3. Machine learning: Provides algebraic tools for probabilistic models
4. Formal verification: Provides tools for formal analysis of probabilistic systems

References

The paper cites 29 related references, mainly including:

• 11 Fritz, T.: A synthetic approach to Markov kernels (foundational work on Markov categories)
• 16 Fritz, T., Rischel, E.F.: Infinite products and zero-one laws (original work on infinite tensor products)
• 21 Piedeleu, R. et al.: A complete axiomatisation of equivalence for discrete probabilistic programming (axiomatization of discrete probability)
• 25 Stein, D. et al.: Graphical quadratic algebra (string diagrams for Gaussian probability)

This paper makes important contributions to categorical probability theory, providing a systematic algebraic treatment method for continuous probability. While further development is needed in practical applications, its theoretical value and methodological innovation have profound significance.