Quantum Bayesian Networks: Compositionality and Typing via Linear Logic

Rémi Di Guardia, Thomas Ehrhard, Claudia Faggian · 2026

Read Quantum Bayesian Networks: Compositionality and Typing via Linear Logic by Rémi Di Guardia, Thomas Ehrhard, Claudia Faggian (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-26059.txt.

Introduction

Di Guardia, Ehrhard, and Faggian bring compositional principles and a typing discipline to quantum Bayesian networks. These networks are graphical models for causal relations and measurement probabilities across classical and quantum data. Their compositional semantics matches the standard factor-based reading when all causes are classical. In the purely quantum case it reduces to tensor networks. A typed formalism built on linear logic proof-nets then guarantees well-behaved composition of systems, proved sound and complete with respect to quantum Bayesian networks.

Catalog record

CREATORRémi Di Guardia, Thomas Ehrhard, Claudia Faggian
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE86800
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.26059

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Quantum Bayesian Networks: Compositionality and Typing via Linear Logic Rémi Di Guardia, Thomas Ehrhard, Claudia Faggian Abstract Quantum Bayesian networks [ 10 ] provide a mathematical formalism to describe causal relations, to analyse correlations, and to predict the probabilities of measurement outcomes, in systems involving both classical and quantum data. They generalize Pearl’s Bayesian networks [ 18 ] —prominent graphical models for classical probabilistic reasoning and inference. Our paper brings compositional principles and a typing discipline into this setting. A key feature of our compositional semantics is that when all causes are classical, it coincides with the standard factor-based semantics of Bayesian networks, while in the purely quantum case it reduces to tensor networks. We then propose a typed formalism based on linear logic proof-nets, where types ensure well-behaved composition of systems. \Crefname equationEq.Eqs. \Crefname sectionSec.Sections \Crefname figureFig.Figs. \Crefname theoremThm.Thm. \Crefname propositionProp.Prop. \Crefname remarkRem.Rem. \Crefname corollaryCor.Cor. \Crefname definitionDef.Def. \Crefname exampleEx.Ex. \Crefname propertyPropertyProperties Université Paris Cité, CNRS, Inria, IRIF, F-75013, Paris, France and https://remidig.github.io/ [email protected]://orcid.org/0009-0004-8632-108X Université Paris Cité, CNRS, IRIF, F-75013, Paris, [email protected] Université Paris Cité, CNRS, IRIF, F-75013, Paris, [email protected] ccs Mathematics of computing Bayesian networks ccs Theory of computation Quantum computation theory ccs Theory of computation Linear logic keywords Quantum Bayesian Networks, Quantum Causal Models, Bayesian Networks, Proof-Nets, Linear Logic † † runningtitle: Quantum Bayesian Networks: Compositionality and Typing, via Linear Logic † † runningauthor: R. Di Guardia, T. Ehrhard, and C. Faggian † † copyright: Rémi Di Guardia, Thomas Ehrhard, and Claudia Faggian † † relatedversion: Full version of an accepted but not yet published paper. † † funding: by the Plan France 2030 through the PEPR project EPiQ (ANR-22-PETQ0007) and by the European Union through the MSCA SE project QCOMICAL (Grant Agreement ID: 101182520). 1 Introduction Pearl’s Bayesian networks [ 17 , 18 ] provide a framework for reasoning under conditions of uncertainty and partial knowledge, with a wide range of applications from statistics to epidemiology, economics and computer science. Bayesian networks (BNs) have a dual nature , serving both as probabilistic graphical models for classical probabilistic reasoning and inference, and as causal models, precising the connections between observed data and causal relations. When reasoning on quantum systems , the classical framework is not general enough to account for entanglement and the non-local correlations observed in Bell experiments. The development of quantum causal models (see e.g. [ 1 ] and references therein) is an active research area across quantum information and the foundations of quantum theory, advancing along various axes, whose motivations span from foundational questions and non-locality, to enabling device-independent cryptographic protocols, to facilitating data-driven discovery. In this paper, we focus on Quantum Bayesian networks, a direct generalization of Pearl’s networks, introduced in foundational work by Henson, Lal, and Pusey [ 10 ] . They provide a mathematical framework to describe causal relations, to analyse correlations, and to predict the probabilities of measurement outcomes, in systems involving both classical and quantum data. The formalism builds on previous work by Leifer and Spekkens [ 15 ] , where the perspective is that of quantum theory as a theory of inference . Quantum theory is indeed fundamentally probabilistic at its core, as it is concerned with predicting the probabilities of measurement outcomes on a physical system. In this sense, the prediction task can be framed as a problem of probabilistic inference over models involving both classical and quantum data , as \Cref ex:bell illustrates. Probabilistic inference then offers mathematical and logical tools to comprehend, predict, and control quantum phenomena, essential to both the theoretical understanding and to quantum technologies. Example 1 (Alice & Bob: the Bell set-up) . The directed acyclic graph in \Cref fig:Bell describes the well-known set-up for the Bell experiment. Alice and Bob—who stand in widely separated laboratories—are each able to perform two possible measurements on a qubit (for example, measuring it with respect to two different bases). Their colleague Quentin prepares a pair of (possibly entangled) qubits, sending one to Alice and the other to Bob. When Alice receives her qubit q_{1} , she chooses to randomly perform one of the two possible measurements, by flipping a coin X . When Bob receives his qubit q_{2} , he also performs a measurement, by flipping a coin Y . The result of the experiment is \Pr(a,b\mid x,y)=\Pr(a,b,x,y)/\Pr(x,y) (1) i.e. the probability that the (classical) outcomes of Alice and Bob measurements are respectively a and b , given outcomes x for X and y for Y . A B Q X Y q_{2} q_{1} Figure 1: Bell set-up (from [ 10 ] ) Q X A B Y cut cut \invamp cut X^{+} X^{-} Q^{+} Q^{-} q_{1}^{-} q_{2}^{-} Y^{+} Y^{-} A^{+} B^{+} Figure 2: Bell set-up as a quantum proof-net Bayesian Networks. In the framework of BNs, the causal structure is encoded by a directed acyclic graph (DAG), where nodes represent random variables and edges express conditional dependencies. The strength of the dependencies (or the degree of knowledge) is quantified by conditional probability tables. The structure enables a compact representation of large probability distributions, and efficient inference of marginals or conditionals through factorization. The strength of BNs is to provide efficient inference algorithms (both exact and approximate) that can answer queries about the underlying distribution without explicitly constructing it in full. Critically, the DAG includes both observable variables of interest, and hidden (unobserved) ones; we adopt the common convention of denoting classical hidden variables by shaded nodes. In the quantum setting—as in Bell’s theorems—hidden variables play a central role, and they may correspond to quantum systems. Semantics and Inference. The semantics of a Bayesian network is the probability distribution it defines. More accurately, what one seeks is the marginal distribution over variables of interest . Exact inference computes it precisely. This involves two key operations: product ( composing ) + summing-out ( hiding ) irrelevant variables. The formalization and theory of inference rely on a class of functions, known as factors (see \Cref sec:BN), which is an abstraction of conditional probability distributions. Tractability and efficiency rely on their properties, and specifically on two key aspects: 1. the product of factors inherently shares variables , and 2. product and sum distribute under suitable conditions—pushing the sum on small components reduces the size of computations. Quantum Bayesian Networks, issues. Quantum Bayesian networks are still an emerging field, not as developed as their classical counterparts. A crucial missing feature is the ability to compute the semantics of the model (the desired marginal distribution) through intermediate, partial computations, without ever computing the full joint distribution. Put differently, what is lacking is compositionality , which would enable computing a model’s denotation as a function of its subparts, alongside modular reasoning. A closely related question concerns modularity : when can causal descriptions of systems as subparts be used to construct larger models? A well-established tool to ensure modularity are types , that specify precise contracts ( e.g. input/output behaviors) for the components of a system. The goal of this paper is twofold. • To address the lack of compositionality and modularity in the setting of Quantum Bayesian networks by introducing methods and concepts from denotational semantics and proof theory, thereby enabling compositional principles and a typing discipline. • To develop a framework

Read the complete text