The problem of determining whether a probabilistic program terminates almost surely (i.e. with probability one) is undecidable, and actually Π⁰₂-complete. For this reason, a growing literature has explored classes of programs for which this and related problems can be shown (semi-)decidable. In this work we consider the termination problem for the language of Probabilistic Higher-Order Recursion Schemes (PHORS). Using the weighted relational semantics of linear logic, we translate this problem into the computation of suitable generating functions associated with the program interpreted. This way, we establish the decidability of almost sure termination for a class of programs that extends Li et al.’s affine PHORS via a type discipline with bounded exponentials. To achieve this, we show that the generating functions for such programs are always algebraic, that is, solutions of polynomial equations, yielding an effective method to answer the termination problem.

In the last few years there has been a growing interest towards methods for statistical inference and learning based on computational geometry and, notably, tropical geometry, that is, the study of algebraic varieties over the min-plus semiring. At the same time, recent work has demonstrated the possibility of interpreting higher-order probabilistic programming languages in the framework of tropical mathematics, by exploiting algebraic and categorical tools coming from the semantics of linear logic. In this work we combine these two worlds, showing that tools and ideas from tropical geometry can be used to perform statistical inference over higher-order probabilistic programs. Notably, we first show that each such program can be associated with a degree and a n -dimensional polyhedron that encode its most likely runs. Then, we use these tools in order to design an intersection type system that estimates most likely runs in a compositional and efficient way.

In this paper we introduce several quantitative methods for the lambda-calculus based on partial metrics, a well-studied variant of standard metric spaces that have been used to metrize non-Hausdorff topologies, like those arising from Scott domains. First, we study quantitative variants, based on program distances, of sensible equational theories for the λ-calculus, like those arising from Böhm trees and from the contextual preorder. Then, we introduce applicative distances capturing higher-order Scott topologies, including reflexive objects like the D_∞ model. Finally, we provide a quantitative insight on the well-known connection between the Böhm tree of a λ-term and its Taylor expansion, by showing that the latter can be presented as an isometric transformation.

ArKiv Extended Version https://arxiv.org/abs/2311.15003

We study the interpretation of the lambda-calculus in a framework based on tropical mathematics, and we show that it provides a unifying framework for two well-developed quantitative approaches to program semantics: on the one hand program metrics, based on the analysis of program sensitivity via Lipschitz conditions, on the other hand resource analysis, based on linear logic and higher-order program differentiation. To do that, we focus on the semantics arising from the relational model weighted over the tropical semiring, and we discuss its application to the study of "best case" program behavior for languages with probabilistic and non-deterministic effects. Finally, we show that a general foundation for this approach is provided by an abstract correspondence between tropical algebra and Lawvere’s theory of generalized metric spaces.

The overall purpose of the present work is to lay the foundations for a new approach to bridge logic and probabilistic computation. To this aim we introduce extensions of classical and intuitionistic propositional logic with counting quantifiers, that is, quantifiers that measure to which extent a formula is true. The resulting systems, called cCPL and iCPL, respectively, admit a natural semantics, based on the Borel σ-algebra of the Cantor space, together with a sound and complete proof system. Our main results consist in relating cCPL and iCPL with some central concepts in the study of probabilistic computation. On the one hand, the validity of cCPL-formulae in prenex form characterizes the corresponding level of Wagner's hierarchy of counting complexity classes, closely related to probabilistic complexity. On the other hand, proofs in iCPL correspond, in the sense of Curry and Howard, to typing derivations for a randomized extension of the λ-calculus, so that counting quantifiers reveal the probability of termination of the underlying probabilistic programs.

In this paper we are concerned with understanding the nature of program metrics for calculi with higher-order types, seen as natural generalizations of program equivalences. Some of the metrics we are interested in are well-known, such as those based on the interpretation of terms in metric spaces and those obtained by generalizing observational equivalence. We also introduce a new one, called the interactive metric, built by applying the well-known Int-Construction to the category of metric complete partial orders. Our aim is then to understand how these metrics relate to each other, i.e., whether and in which cases one such metric refines another, in analogy with corresponding well-studied problems about program equivalences. The results we obtain are twofold. We first show that the metrics of semantic origin, i.e., the denotational and interactive ones, lie in between the observational and equational metrics and that in some cases, these inclusions are strict. Then, we give a result about the relationship between the denotational and interactive metrics, revealing that the former is less discriminating than the latter. All our results are given for a linear lambda-calculus, and some of them can be generalized to calculi with graded comonads, in the style of Fuzz.

Programming language semantics traditionally deals with qualitative properties of programs, that is, properties that a program may either satisfy or not, like termination or correctness.Moreover, program semantics generally attribute programs a meaning, typically a function of some kind, so as to be able to identify programs which behave in the same way in all contexts (i.e., which have the same meaning).This can be done in many different ways, from observational equivalence -the coarsest adequate congruence -to various forms of formal systems in the style of equational logic, to denotational semantics.Nevertheless, the past ten years have seen the introduction of a series of logical and semantic frameworks which go significantly beyond this picture: on the one hand, frameworks enabling the expression of quantitative properties, i.e., properties that a program may satisfy to a certain extent or up to a certain error (e.g., probabilistic termination or correctness up to some error probability or some approximation error).Moreover, the meaning attributed to programs may allow the latter to be compared in quantitative ways, that is, as behaving in a similar, although not exactly equivalent, way, or to analyze how sensitive programs are to variations in their input.We refer here, for example, to approaches like behavioral and program metrics, differential semantics, automatic differentiation, sensitivity analysis and its application to differential privacy.These frameworks have progressively led to integrate methods coming from probabilistic programming, approximate and incremental computing, as well as machine learning within several standard theoretical approaches to program semantics.This special issue is meant to collect contributions along these lines and comprises the following five papers:• "Up-To Techniques for Behavioural Metrics via Fibrations, " by Bonchi, König, and Petrisan.This deals with the problem of deriving enhancements to the metric analog of the bisimulation proof method in an abstract way and with how categorical fibrations turn out to be a powerful tool for that.• "Bisimulation and Behavioural Equivalences for Continuous-time Markov Processes," by Chen, Clerc, and Panangaden.This contribution gives a unified view of various notions of behavioral equivalence and bisimulation for probabilistic transition systems whose time evolution is continuous rather than discrete.• "Coherent Differentiation," by Ehrhard.This paper introduces a new categorical framework for higher order program differentiation.In contrast to usual approaches based on the differential λ-calculus, this new framework does not require additivity (hence nondeterminism), and is thus compatible with both deterministic and probabilistic computational models.

We introduce an extension of classical propositional logic with counting quantifiers. These forms of quantification make it possible to express that a formula is true in a certain portion of the set of all its interpretations. Beyond providing a sound and complete proof system for this logic, we show that validity problems for counting propositional logic can be used to capture counting complexity classes. More precisely, we show that the complexity of the decision problems for validity of prenex counting formulas perfectly matches the appropriate levels of Wagner's counting hierarchy.

International audience

We show that an intuitionistic version of counting propositional logic corresponds, in the sense of Curry and Howard, to an expressive type system for the probabilistic event λ-calculus, a vehicle calculus in which both call-by-name and call-by-value evaluation of discrete randomized functional programs can be simulated. In this context, proofs (respectively, types) do not guarantee that validity (respectively, termination) holds, but reveal the underlying probability. We finally show how to obtain a system precisely capturing the probabilistic behavior of λ-terms, by endowing the type system with an intersection operator.

An approximate program transformation is a transformation that can change the semantics of a program within a specified empirical error bound. Such transformations have wide applications: they can decrease computation time, power consumption, and memory usage, and can, in some cases, allow implementations of incomputable operations. Correctness proofs of approximate program transformations are by definition quantitative. Unfortunately, unlike with standard program transformations, there is as of yet no modular way to prove correctness of an approximate transformation itself. Error bounds must be proved for each transformed program individually, and must be re-proved each time a program is modified or a different set of approximations are applied. In this paper, we give a semantics that enables quantitative reasoning about a large class of approximate program transformations in a local, composable way. Our semantics is based on a notion of distance between programs that defines what it means for an approximate transformation to be correct up to an error bound. The key insight is that distances between programs cannot in general be formulated in terms of metric spaces and real numbers. Instead, our semantics admits natural notions of distance for each type construct; for example, numbers are used as distances for numerical data, functions are used as distances for functional data, an polymorphic lambda-terms are used as distances for polymorphic data. We then show how our semantics applies to two example approximations: replacing reals with floating-point numbers, and loop perforation.

In this paper we explore a family of type isomorphisms in System F whose validity corresponds, semantically, to some form of the Yoneda isomorphism from category theory. These isomorphisms hold under theories of equivalence stronger than beta-eta-equivalence, like those induced by parametricity and dinaturality. We show that the Yoneda type isomorphisms yield a rewriting over types, that we call Yoneda reduction, which can be used to eliminate quantifiers from a polymorphic type, replacing them with a combination of monomorphic type constructors. We establish some sufficient conditions under which quantifiers can be fully eliminated from a polymorphic type, and we show some application of these conditions to count the inhabitants of a type and to compute program equivalence in some fragments of System F.

Due to the undecidability of most type-related properties of System F like type inhabitation or type checking, restricted polymorphic systems have been widely investigated (the most well-known being ML-polymorphism). In this paper we investigate System Fat, or atomic System F, a very weak predicative fragment of System F whose typable terms coincide with the simply typable ones. We show that the type-checking problem for Fat is decidable and we propose an algorithm which sheds some new light on the source of undecidability in full System F. Moreover, we investigate free theorems and contextual equivalence in this fragment, and we show that the latter, unlike in the simply typed lambda-calculus, is undecidable.

Generalized metrics, arising from Lawvere's view of metric spaces as enriched categories, have been widely applied in denotational semantics as a way to measure to which extent two programs behave in a similar, although non equivalent, way. However, the application of generalized metrics to higher-order languages like the simply typed lambda calculus has so far proved unsatisfactory. In this paper we investigate a new approach to the construction of cartesian closed categories of generalized metric spaces. Our starting point is a quantitative semantics based on a generalization of usual logical relations. Within this setting, we show that several families of generalized metrics provide ways to extend the Euclidean metric to all higher-order types.

We investigate completeness and parametricity for a general class of realizability semantics for System F defined in terms of closure operators over sets of $\lambda$-terms. This class includes most semantics used for normalization theorems, as those arising from Tait's saturated sets and Girard's reducibility candidates. We establish a completeness result for positive types which subsumes those existing in the literature, and we show that closed realizers satisfy parametricity conditions expressed either as invariance with respect to logical relations or as dinaturality. Our results imply that, for positive types, typability, realizability and parametricity are equivalent properties of closed normal $\lambda$-terms.