Michael W. Whalen

Sam Bayless, Stefano Buliani, Darion Cassel et al.

Large Language Models perform well at natural language interpretation and reasoning, but their inherent stochasticity limits their adoption in regulated industries like finance and healthcare that operate under strict policies. To address this limitation, we present a two-stage neurosymbolic framework that (1) uses LLMs with optional human guidance to formalize natural language policies, allowing fine-grained control of the formalization process, and (2) uses inference-time autoformalization to validate logical correctness of natural language statements against those policies. When correctness is paramount, we perform multiple redundant formalization steps at inference time, cross checking the formalizations for semantic equivalence. Our benchmarks demonstrate that our approach exceeds 99% soundness, indicating a near-zero false positive rate in identifying logical validity. Our approach produces auditable logical artifacts that substantiate the verification outcomes and can be used to improve the original text.

Andreas Katis, Grigory Fedyukovich, Jeffrey Chen et al.

Diversity in the exhibited behavior of a given system is a desirable characteristic in a variety of application contexts. Synthesis of conformant implementations often proceeds by discovering witnessing Skolem functions, which are traditionally deterministic. In this paper, we present a novel Skolem extraction algorithm to enable synthesis of witnesses with random behavior and demonstrate its applicability in the context of reactive systems. The synthesized solutions are guaranteed by design to meet the given specification,while exhibiting a high degree of diversity in their responses to external stimuli. Case studies demonstrate how our proposed frame-work unveils a novel application of synthesis in model-based fuzz testing to generate fuzzers of competitive performance to general-purpose alternatives, as well as the practical utility of synthesized controllers in robot motion planning problems.

Andreas Katis, Grigory Fedyukovich, Huajun Guo et al.

Automated synthesis of reactive systems from specifications has been a topic of research for decades. Recently, a variety of approaches have been proposed to extend synthesis of reactive systems from proposi- tional specifications towards specifications over rich theories. We propose a novel, completely automated approach to program synthesis which reduces the problem to deciding the validity of a set of forall-exists formulas. In spirit of IC3 / PDR, our problem space is recursively refined by blocking out regions of unsafe states, aiming to discover a fixpoint that describes safe reactions. If such a fixpoint is found, we construct a witness that is directly translated into an implementation. We implemented the algorithm on top of the JKind model checker, and exercised it against contracts written using the Lustre specification language. Experimental results show how the new algorithm outperforms JKinds already existing synthesis procedure based on k-induction and addresses soundness issues in the k-inductive approach with respect to unrealizable results.

Andreas Katis, Grigory Fedyukovich, Andrew Gacek et al.

The realizability problem in requirements engineering is to determine the existence of an implementation that meets the given formal requirements. A step forward after realizability is proven, is to construct such an implementation automatically, and thus solve the problem of program synthesis. In this paper, we propose a novel approach to pro- gram synthesis guided by k-inductive proofs of realizability of assume- guarantee contracts constructed from safety properties. The proof of re- alizability is performed over a set of forall-exists formulas, and synthesis is per- formed by extracting Skolem functions witnessing the existential quan- tification. These Skolem functions can then be combined into an imple- mentation. Our approach is implemented in the JSyn tool which con- structs Skolem functions from a contract written in a variant of the Lus- tre programming language and then compiles the Skolem functions into a C language implementation. For a variety of benchmark models that already contained hand-written implementations, we are able to identify the usability and effectiveness of the synthesized counterparts, assuming a component-based verification framework.

John D. Backes, Michael W. Whalen, Andrew Gacek et al.

English language requirements are often used to specify the behavior of complex cyber-physical systems. The process of transforming these requirements to a formal specification language is often challenging, especially if the specification language does not contain constructs analogous to those used in the original requirements. For example, requirements often contain real-time constraints, but many specification languages for model checkers have discrete time semantics. Work in specification patterns helps to bridge these gaps, allowing straightforward expression of common requirements patterns in formal languages. In this work we demonstrate how we support real-time specification patterns in the Assume Guarantee Reasoning Environment (AGREE) using observers. We demonstrate that there are subtle challenges, not mentioned in previous literature, to express real-time patterns accurately using observers. We then demonstrate that these patterns are sufficient to model real-time requirements for a real-world avionics system.

Elaheh Ghassabani, Andrew Gacek, Michael W. Whalen

Symbolic model checkers can construct proofs of properties over very complex models. However, the results reported by the tool when a proof succeeds do not generally provide much insight to the user. It is often useful for users to have traceability information related to the proof: which portions of the model were necessary to construct it. This traceability information can be used to diagnose a variety of modeling problems such as overconstrained axioms and underconstrained properties, and can also be used to measure completeness of a set of requirements over a model. In this paper, we present a new algorithm to efficiently compute the inductive validity core (IVC) within a model necessary for inductive proofs of safety properties for sequential systems. The algorithm is based on the UNSAT core support built into current SMT solvers and a novel encoding of the inductive problem to try to generate a minimal inductive validity core. We prove our algorithm is correct, and describe its implementation in the JKind model checker for Lustre models. We then present an experiment in which we benchmark the algorithm in terms of speed, diversity of produced cores, and minimality, with promising results.

Andreas Katis, Michael W. Whalen, Andrew Gacek

In previous work, we have introduced a contract-based real- izability checking algorithm for assume-guarantee contracts involving infinite theories, such as linear integer/real arith- metic and uninterpreted functions over infinite domains. This algorithm can determine whether or not it is possible to con- struct a realization (i.e. an implementation) of an assume- guarantee contract. The algorithm is similar to k-induction model checking, but involves the use of quantifiers to deter- mine implementability. While our work on realizability is inherently useful for vir- tual integration in determining whether it is possible for sup- pliers to build software that meets a contract, it also provides the foundations to solving the more challenging problem of component synthesis. In this paper, we provide an initial synthesis algorithm for assume-guarantee contracts involv- ing infinite theories. To do so, we take advantage of our realizability checking procedure and a skolemization solver for forall-exists formulas, called AE-VAL. We show that it is possible to immediately adapt our existing algorithm towards syn- thesis by using this solver, using a demonstration example. We then discuss challenges towards creating a more robust synthesis algorithm.

Andrew Gacek, Andreas Katis, Michael W. Whalen et al.

Virtual integration techniques focus on building architectural models of systems that can be analyzed early in the design cycle to try to lower cost, reduce risk, and improve quality of complex embedded systems. Given appropriate architectural descriptions and compositional reasoning rules, these techniques can be used to prove important safety properties about the architecture prior to system construction. Such proofs build from "leaf-level" assume/guarantee component contracts through architectural layers towards top-level safety properties. The proofs are built upon the premise that each leaf-level component contract is realizable; i.e., it is possible to construct a component such that for any input allowed by the contract assumptions, there is some output value that the component can produce that satisfies the contract guarantees. Without engineering support it is all too easy to write leaf-level components that can't be realized. Realizability checking for propositional contracts has been well-studied for many years, both for component synthesis and checking correctness of temporal logic requirements. However, checking realizability for contracts involving infinite theories is still an open problem. In this paper, we describe a new approach for checking realizability of contracts involving theories and demonstrate its usefulness on several examples.

Andreas Katis, Andrew Gacek, Michael W. Whalen

Virtual integration techniques focus on building architectural models of systems that can be analyzed early in the design cycle to try to lower cost, reduce risk, and improve quality of complex embedded systems. Given appropriate architectural descriptions, assume/guarantee contracts, and compositional reasoning rules, these techniques can be used to prove important safety properties about the architecture prior to system construction. For these proofs to be meaningful, each leaf-level component contract must be realizable; i.e., it is possible to construct a component such that for any input allowed by the contract assumptions, there is some output value that the component can produce that satisfies the contract guarantees. We have recently proposed (in [1]) a contract-based realizability checking algorithm for assume/guarantee contracts over infinite theories supported by SMT solvers such as linear integer/real arithmetic and uninterpreted functions. In that work, we used an SMT solver and an algorithm similar to k-induction to establish the realizability of a contract, and justified our approach via a hand proof. Given the central importance of realizability to our virtual integration approach, we wanted additional confidence that our approach was sound. This paper describes a complete formalization of the approach in the Coq proof and specification language. During formalization, we found several small mistakes and missing assumptions in our reasoning. Although these did not compromise the correctness of the algorithm used in the checking tools, they point to the value of machine-checked formalization. In addition, we believe this is the first machine-checked formalization for a realizability algorithm.