POPL 2027
Sun 10 - Sat 16 January 2027 Mexico City, Mexico

Finding denotational models of probabilistic languages that support sampling from continuous distributions is difficult: the existence of continuous distributions means that measure-theoretic objects are needed to define the semantics. A challenging task on top of this is logical reasoning with respect to the semantic model: in the case of probabilistic programs, the desired properties to reason about include the distributions, expected value and the probabilistic (in)dependence of outputs.

This paper introduces a monadic semantics for higher-order probabilistic programs with continuous distributions based on Simpson’s theory of probability sheaves and the internal logic of atomic sheaves. We prove that our semantics forms a model of the $\nu$-calculus. Furthermore, using the well-behavedness of the semantic domain and the internal logic of the topos, we derive the first higher-order probabilistic Hoare logic with rigorous treatment of continuous random variables, higher-order functions, and independence.