Journal of Automated Reasoning最新文献

Distributed clause-sharing SAT solvers can solve challenging problems hundreds of times faster than sequential SAT solvers by sharing derived information among multiple sequential solvers. Unlike sequential solvers, however, distributed solvers have not been able to produce proofs of unsatisfiability in a scalable manner, which limits their use in critical applications. In this work, we present a method to produce unsatisfiability proofs for distributed SAT solvers by combining the partial proofs produced by each sequential solver into a single, linear proof. We first describe a simple sequential algorithm and then present a fully distributed algorithm for proof composition, which is substantially more scalable and general than prior works. Our empirical evaluation with over 1500 solver threads shows that our distributed approach allows proof composition and checking within around 3 $\times$ its own (highly competitive) solving time.

We introduce a single-set axiomatisation of cubical (omega )-categories, including connections and inverses. We justify these axioms by establishing a series of equivalences between the category of single-set cubical (omega )-categories, and their variants with connections and inverses, and the corresponding cubical (omega )-categories. We also report on the formalisation of cubical (omega )-categories with the Isabelle/HOL proof assistant, which has been instrumental in developing the single-set axiomatisation.

We address the challenges of scaling verification efforts to match the increasing complexity and size of systems. We propose a research agenda aimed at building a performant proof engine by studying the asymptotic performance of proof engines and redesigning their building blocks. As a case study, we explore equational rewriting and introduce a novel prototype proof engine building block for rewriting in Coq, utilizing proof by reflection for enhanced performance. Our prototype implementation can significantly improve the development of verified compilers, as demonstrated in a case study with the Fiat Cryptography toolchain. The resulting extracted command-line compiler is about 1000(times ) faster while featuring simpler compiler-specific proofs. This work lays some foundation for scaling verification efforts and contributes to the broader goal of developing a proof engine with good asymptotic performance, ultimately aimed at enabling the verification of larger and more complex systems.

Deep neural networks (DNNs) play a crucial role in the field of machine learning, demonstrating state-of-the-art performance across various application domains. However, despite their success, DNN-based models may occasionally exhibit challenges with generalization, i.e., may fail to handle inputs that were not encountered during training. This limitation is a significant challenge when it comes to deploying deep learning for safety-critical tasks, as well as in real-world settings characterized by substantial variability. We introduce a novel approach for harnessing DNN verification technology to identify DNN-driven decision rules that exhibit robust generalization to previously unencountered input domains. Our method assesses generalization within an input domain by measuring the level of agreement between independently trained deep neural networks for inputs in this domain. We also efficiently realize our approach by using off-the-shelf DNN verification engines, and extensively evaluate it on both supervised and unsupervised DNN benchmarks, including a deep reinforcement learning (DRL) system for Internet congestion control—demonstrating the applicability of our approach for real-world settings. Moreover, our research introduces a fresh objective for formal verification, offering the prospect of mitigating the challenges linked to deploying DNN-driven systems in real-world scenarios.

In Quantified Boolean Formulas QBFs, dependency schemes help to detect spurious or superfluous dependencies that are implied by the variable ordering in the quantifier prefix but are not essential for constructing countermodels. This detection can provably shorten refutations in specific proof systems, and is expected to speed up runs of QBF solvers. The proof system (texttt{QCDCL}) recently defined by Beyersdorff and Boehm (LMCS 2023) abstracts the reasoning employed by QBF solvers based on conflict-driven clause-learning (CDCL) techniques. We show how to incorporate the use of dependency schemes into this proof system, either in a preprocessing phase, or in the propagations and clause learning, or both. We then show that when the reflexive resolution path dependency scheme (texttt{D}^{texttt{rrs}}) is used, a mixed picture emerges: the proof systems that add (texttt{D}^{texttt{rrs}}) to (texttt{QCDCL}) in these three ways are not only incomparable with each other, but are also incomparable with the basic (texttt{QCDCL}) proof system that does not use (texttt{D}^{texttt{rrs}}) at all, as well as with several other resolution-based QBF proof systems. A notable fact is that all our separations are achieved through QBFs with bounded quantifier alternation.

We describe the design, implementation and verification of an automated theorem prover for first-order logic with functions. The proof search procedure is based on sequent calculus and we formally verify its soundness and completeness in Isabelle/HOL using an existing abstract framework for coinductive proof trees. Our analytic completeness proof covers both open and closed formulas. Since our deterministic prover considers only the subset of terms relevant to proving a given sequent, we do the same when building a countermodel from a failed proof. Finally, we formally connect our prover with the proof system and semantics of the existing SeCaV system. In particular, the prover can generate human-readable SeCaV proofs which are also machine-verifiable proof certificates. The abstract framework we rely on requires us to fix a stream of proof rules in advance, independently of the formula we are trying to prove. We discuss the efficiency implications of this and the difficulties in mitigating them.

We present a stepwise refinement approach to develop verified parallel algorithms, down to efficient LLVM code. The resulting algorithms’ performance is competitive with their counterparts implemented in C++. Our approach is backwards compatible with the Isabelle Refinement Framework, such that existing sequential formalizations can easily be adapted or re-used. As case study, we verify a parallel quicksort algorithm that is competitive to unverified state-of-the-art algorithms.

Semilinear maps are a generalization of linear maps between vector spaces where we allow the scalar action to be twisted by a ring homomorphism such as complex conjugation. In particular, this generalization unifies the concepts of linear and conjugate-linear maps. We implement this generalization in Lean’s mathlib library, along with a number of important results in functional analysis which previously were impossible to formalize properly. Specifically, we prove the Fréchet–Riesz representation theorem and the spectral theorem for compact self-adjoint operators generically over real and complex Hilbert spaces, additionally developing the Fourier theory needed to state and prove Parseval’s identity. We also show that semilinear maps have applications beyond functional analysis by formalizing the one-dimensional case of a theorem of Dieudonné and Manin that classifies the isocrystals over an algebraically closed field with positive characteristic.

We present a formal framework for process composition based on actions that are specified by their input and output resources. The correctness of these compositions is verified by translating them into deductions in intuitionistic linear logic. As part of the verification we derive simple conditions on the compositions which ensure well-formedness of the corresponding deduction when satisfied. We mechanise the whole framework, including a deep embedding of ILL, in the proof assistant Isabelle/HOL. Beyond the increased confidence in our proofs, this allows us to automatically generate executable code for our verified definitions. We demonstrate our approach by formalising part of the simulation game Factorio and modelling a manufacturing process in it. Our framework guarantees that this model is free of bottlenecks.

Choice logics constitute a family of propositional logics and are used for the representation of preferences, with especially qualitative choice logic (QCL) being an established formalism with numerous applications in artificial intelligence. While computational properties and applications of choice logics have been studied in the literature, only few results are known about the proof-theoretic aspects of their use. We propose a sound and complete sequent calculus for preferred model entailment in QCL, where a formula F is entailed by a QCL-theory T if F is true in all preferred models of T. The calculus is based on labeled sequent and refutation calculi, and can be easily adapted for different purposes. For instance, using the calculus as a cornerstone, calculi for other choice logics such as conjunctive choice logic (CCL) and lexicographic choice logic (LCL) can be obtained in a straightforward way.