Backward through Time, Algebraically

A Compositional Theory of Causally Masked Transformers

What types of decision problems can a causally masked, finite-precision transformer solve for inputs of arbitrary length? Existing answers often rely on idealized arithmetic, but under finite precision, rounding and evaluation order can change what information attention retains and therefore what the model can compute. We develop an algebraic formalization that derives expressivity directly from the model's implemented dynamics. Its central object is its memory; the finite internal state computed by attention that summarizes the information from the prefix available to all future queries. Each attention head updates its own state independently within a layer, while layers compose hierarchically, providing a uniform route from model assumptions to expressivity bounds. Applying this method to transformers without positional embeddings, we obtain an expressivity hierarchy governed by the attention type under specific numerical semantics. Width-one sliding-window attention supports bounded-suffix memory, while a modified form of soft attention supports irreversible, checklist-like state, and combining the two mechanisms provides an interplay of both. Ordinary left-to-right floating-point soft attention can realize more expressive memory operations than any of the above. Algebraically, the four cases correspond to definite, R-trivial, locally R-trivial, and aperiodic semigroups. Under an explicit free-wiring assumption, all four bounds are tight.

Backpropagation for Effectful Languages I: Finite Probability and Discrete Output Algebraic Effects

We analyse reverse-mode automatic differentiation (AD) for discrete probabilistic programs. Our construction is formulated in the framework of Combinatory Homomorphic Automatic Differentiation (CHAD), treating AD as a structure-preserving transformation of programs, guided by a denotational semantics. The main case study is the finite atomic distribution monad, whose computations have finite support and differentiable weights. The key point is that differentiating probabilistic programs requires cotangents to flow backwards not only through deterministic computations, but also through the probabilistic structure itself. We define the corresponding reverse-mode code transformation and prove its correctness, for handled real-output programs, by a categorical logical-relations argument. Although the paper focuses on finite discrete probability, the construction gives a reusable pattern for differentiating discrete-output algebraic effects, including finite multiset non-determinism (e.g., from fork-join parallelism), exceptions, and writer-style accumulation (e.g., for in-place accumulation of high-dimensional vectors). More broadly, we view this work as a foundational step towards extending CHAD to richer probabilistic languages and to other algebraic effects with handlers.

SAT Modulo Well-Founded Semantics

It is shown that the choice operator can be materialized by a SAT solver while propagating the consequences of choices through an extension of the alternating fixpoint algorithm for WFS with conflicts that are propagated back to the SAT solver.

The Infinite, in Finite Time

Linear-time temporal properties, such as those described by Linear-time Temporal Logic, are typically modelled as sets of infinite traces. Yet, in a run-time verification context, such as when testing or monitoring a system, only a finite prefix of the system's behaviour can be observed. For some properties, these finite prefixes may be definitive---a yes or no answer can be given without further observation. By enriching the semantics of LTL with these definitive prefixes, we give a proper inductive accounting of the semantics of LTL$_3$, a multi-valued variant of Linear-time Temporal Logic for run-time verification applications. The semantic descriptions of LTL$_3$ in previous work are given only in terms of their relationship to conventional LTL. We show that the semantics of LTL and of LTL$_3$ are isomorphic. In addition, we formalise the formula progression evaluation technique, popularly used in runtime verification contexts, and show its soundness and completeness up to finite traces with respect to our semantics. Then, we turn to linear-time properties more generally: using our theory of definitive prefixes, we re-prove the well-known safety-liveness decomposition theorem, and reconstruct the topology of infinite traces. We define monitorability for properties, providing neat topological characterisations for various monitorability classes, and arrange them into a hierarchy. All of our definitions and proofs are mechanised in Isabelle/HOL.

Axiomatizability of Alexandrov Dynamic Topological Logic

Sequent-style tableaux for intuitionistic propositional logic

Sequent-style tableaux are a refutation calculus in which each node of the refutation tree carries a finite block of formulae and the structural rules are absorbed into the data structure and the closure criterion. In their original, classical form they rest on an involutive De Morgan negation and on closure upon a complementary pair. We show that both may be dispensed with. Replacing unsigned formulae by signed ones, we obtain a block calculus $\mathbf{B}_{ti}$ for intuitionistic propositional logic in which the whole of intuitionism is carried by one rule, the rule decomposing $\mathsf{F}(A \to B)$, which deletes the $\mathsf{F}$-part of the context on passing to the child block. The rules so obtained are, up to the presentation, those of Fitting's signed tableaux; what is new is the block format, in which the structural rules are absorbed rather than admissible, and what follows from it. We identify the semantic reason for this rule and for the one other anomalous one: of the signed compounds of the language, exactly those governed by the implication fail to be locally decomposable, and the two failures are repaired, respectively, by retaining the principal formula and by purging the context. We prove that $\mathbf{B}_{ti}$ is the multiple-succedent sequent calculus $\mathbf{G}_{ti}$ read upside down, that $\mathbf{G}_{ti}$ admits the structural rules, and that $\mathbf{B}_{ti}$ is sound and complete for Kripke semantics, with the finite model property and a block-theoretic proof of the disjunction property.