Learning to prove safety over parameterised concurrent systems

Regular model checking is a generic technique for verification of infinite-state and/or parametrised systems which uses finite word automata or finite tree automata to finitely represent potentially infinite sets of reachable configurations of the systems being verified. The problems addressed by regular model checking are typically undecidable. In order to facilitate termination in as many cases as possible, acceleration is needed in the incremental computation of the set of reachable configurations in regular model checking. In this work, we describe how various incrementally refinable abstractions on finite (word and tree) automata can be used for this purpose. Moreover, the use of abstraction does not only increase chances of the technique to terminate, but it also significantly reduces the problem of an explosion in the number of states of the automata that are generated by regular model checking. We illustrate the efficiency of abstract regular (tree) model checking in verification of simple systems with various sources of infinity such as unbounded counters, queues, stacks, and parameters. We then show how abstract regular tree model checking can be used for verification of programs manipulating tree-like dynamic data structures. Even more complex data structures can be handled using a suitable tree-like encoding.

Regular model checking is a technique for the verification of infinite-state systems whose configurations can be represented as finite words over a suitable alphabet. The form we are studying applies to systems whose set of initial configurations is regular, and whose transition relation is captured by a length-preserving transducer. To verify safety properties, regular model checking iteratively computes automata recognizing increasingly larger regular sets of reachable configurations, and checks if they contain unsafe configurations. Since this procedure often does not terminate, acceleration, abstraction, and widening techniques have been developed to compute a regular superset of the reachable configurations. In this paper, we develop a complementary procedure. Instead of approaching the set of reachable configurations from below, we start with the set of all configurations and approach it from above. We use that the set of reachable configurations is equal to the intersection of all inductive invariants of the system. Since this intersection is non-regular in general, we introduce b-invariants, defined as those representable by CNF-formulas with at most b clauses. We prove that, for every $b\geq0$, the intersection of all inductive b-invariants is regular, and we construct an automaton recognizing it. We show that whether this automaton accepts some unsafe configuration is in EXPSPACE for every $b\geq0$, and PSPACE-complete for b=1. Finally, we study how large must b be to prove safety properties of a number of benchmarks.

Regular model checking is a form of symbolic model checking technique for systems whose states can be represented as finite words over a finite alphabet, where regular sets are used as symbolic representation. A major problem in symbolic model checking of parameterized and infinite-state systems is that fixpoint computations to generate the set of reachable states or the set of reachable loops do not terminate in general. Therefore, acceleration techniques have been developed, which calculate the effect of arbitrarily long sequences of transitions generated by some action. We present a systematic method for using acceleration in regular model checking, for the case where each transition changes at most one position in the word; this includes many parameterized algorithms and algorithms on data structures. The method extracts a maximal (in a certain sense) set of actions from a transition relation. These actions, and systematically obtained compositions of them, are accelerated to speed up a fixpoint computation. The extraction can be done on any representation of the transition relation, e.g., as a union of actions or as a single monolithic transducer. Using this approach, we are for the first time able to verify completely automatically both safety and absence of starvation properties for a collection of parameterized synchronization protocols from the literature; for some protocols, we obtain significant improvements in verification time. The results show that symbolic state-space exploration, without using abstractions, is a viable alternative for verification of parameterized systems with a linear topology.KeywordsModel CheckTransition RelationReachable StateLiveness PropertySymbolic Model CheckThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

Tree regular model checking: A simulation-based approach

We propose abstract regular model checking as a new generic technique for verification of parametric and infinite-state systems. The technique combines the two approaches of regular model checking and verification by abstraction. We propose a general framework of the method as well as several concrete ways of abstracting automata or transducers, which we use for modelling systems and encoding sets of their configurations as usual in regular model checking. The abstraction is based on collapsing states of automata (or transducers) and its precision is being incrementally adjusted by analysing spurious counterexamples. We illustrate the technique on verification of a wide range of systems including a novel application of automata-based techniques to an example of systems with dynamic linked data structures.

We address the problem of computing simulation relations over tree automata. In particular, we consider downward and upward simulations on tree automata, which are, loosely speaking, analogous to forward and backward relations over word automata. We provide simple and efficient algorithms for computing these relations based on a reduction to the problem of computing simulations on labelled transition systems. Furthermore, we show that downward and upward relations can be combined to get relations compatible with the tree language equivalence, which can subsequently be used for an efficient size reduction of nondeterministic tree automata. This is of a very high interest, for instance, for symbolic verification methods such as regular model checking, which use tree automata to represent infinite sets of reachable configurations. We provide experimental results showing the efficiency of our algorithms on examples of tree automata taken from regular model checking computations.

We consider the framework of regular tree model checking where sets of configurations of a system are represented by regular tree languages and its dynamics is modeled by a term rewriting system (or a regular tree transducer). We focus on the computation of the reachability set R*(L) where R is a regular tree transducer and L is a regular tree language. The construction of this set is not possible in general. Therefore, we present a general acceleration technique, called regular tree widening which allows to speed up the convergence of iterative fixpoint computations in regular tree model checking. This technique can be applied uniformly to various kinds of transformations. We show the application of our framework to different analysis contexts: verification of parameterized tree networks and data-flow analysis of multithreaded programs. Parametrized networks are modeled by relabeling tree transducers, and multithreaded programs are modeled by term rewriting rules encoding transformations on control structures. We prove that our widening technique can emulate many existing algorithms for special classes of transformations and we show that it can deal with transformations beyond the scope of these algorithms.

In this book we focused on established decidability results in parameterized model checking. As this topic is a lively field, we cannot make any claim about completeness of our survey. In particular, we left out the large body of research that includes invariant-based techniques, regular model checking, symbolic methods, techniques for counter automata, and abstraction-based methods. In this chapter we give a brief overview of these methods and give further references to the interested reader. Our classification of the techniques is unavoidably subjective, as except for several cornerstone papers, many techniques adopted multiple different ideas.

We present an efficient algorithm to reduce the size of nondeterministic tree automata, while retaining their language. It is based on new transition pruning techniques, and quotienting of the state space w.r.t. suitable equivalences. It uses criteria based on combinations of downward and upward simulation preorder on trees, and the more general downward and upward language inclusions. Since tree-language inclusion is EXPTIME-complete, we describe methods to compute good approximations in polynomial time. We implemented our algorithm as a module of the well-known libvata tree automata library, and tested its performance on a given collection of tree automata from various applications of libvata in regular model checking and shape analysis, as well as on various classes of randomly generated tree automata. Our algorithm yields substantially smaller and sparser automata than all previously known reduction techniques, and it is still fast enough to handle large instances.

In reinforcement learning, an agent incrementally refines a behavioral policy through a series of episodic interactions with its environment. This process can be characterized as explicit reinforcement learning, as it deals with explicit states and concrete transitions. Building upon the concept of symbolic model checking, we propose a symbolic variant of reinforcement learning, in which sets of states are represented through predicates and transitions are represented by predicate transformers. Drawing inspiration from regular model checking, we choose regular languages over the states as our predicates, and rational transductions as predicate transformations. We refer to this framework as regular reinforcement learning, and study its utility as a symbolic approach to reinforcement learning. Theoretically, we establish results around decidability, approximability, and efficient learnability in the context of regular reinforcement learning. Towards practical applications, we develop a deep regular reinforcement learning algorithm, enabled by the use of graph neural networks. We showcase the applicability and effectiveness of (deep) regular reinforcement learning through empirical evaluation on a diverse set of case studies.

Within the context of the verification of infinite-statesystems, (ω)-Regular model checking is the name ofa family of techniques in which states are represented by words,sets of states by finite automata on these objects, and transitionsby finite automata operating on pairs of state encodings, i.e.finite-state transducers. If the states are encoded by finitewords, then sets of (pairs of) states can be represented byfinite-word automata. This setting can be used to represent variousclasses of infinite-state systems, [17], including parametricsystems, FIFO-queue systems, and systems manipulating integervariables (those defined in Presburger arithmetic). When the statesare encoded by infinite words, sets of (pairs of) states arerepresented by deterministic weak Buchi automata. This settingis mainly used to represent systems involving both integer and realvariables [4,6], such as linear hybrid systems with a constantderivative.

“Regular model checking” is the name of a family of techniques for analyzing infinite-state systems in which states are represented by words or trees, sets of states by finite automata on these objects, and transitions by finite automata operating on pairs of state encodings, i.e. finite-state transducers. In this context, the central problem is then to compute the iterative closure of a finite-state transducer. This paper addresses the use of regular model-checking like techniques for systems whose states are represented by infinite (omega) words. Its main motivation is to show the feasibility and usefulness of this approach through a combination of the necessary theoretical developments, implementation, and experimentation. The iteration technique that is used is adapted from recent work of the authors on the iteration of finite-word transducers. It proceeds by comparing successive elements of a sequence of approximations of the iteration, detecting an “increment” that is added to move from one approximation to the next, and extrapolating the sequence by allowing arbitrary repetitions of this increment. By restricting oneself to weak deterministic Buchi automata, and using a number of implementation optimizations, examples of significant size can be handled. The proposed transducer iteration technique can just as well be exploited to compute the closure of a given set of states by the transducer iteration, which has proven to be a very effective way of using the technique. Examples such as a leaking gas burner in which time is modeled by real variables have been handled completely within the automata-theoretic setting.

Stochastic continuum modeling of random interphases from atomistic simulations. Application to a polymer nanocomposite

Surface platinum hydride structures may exist and play a potentially important role during electrocatalysis and cathodic corrosion of Pt(111). Earlier work on platinum hydrides suggests that Pt may form clusters with multiple equivalents of hydrogen. Here, using thermodynamic methods and density functional theory, we compared several surface hydride structures on Pt(111). The structures contain multiple monolayers of hydrogen in or near the surface Pt layer. The hydrogen in these structures may bind the subsurface or reconstruct the surface both in the set of initial configurations and in the resulting (meta)stable structures. Multilayer stable configurations share one monolayer of subsurface H stacking between the top two Pt layers. The structure containing two monolayers (MLs) of H is formed at -0.29V vs normal hydrogen electrode, is locally stable with respect to configurations with similar H densities, and binds H neutrally. Structures with 3 and 4 ML H form at -0.36 and -0.44V, respectively, which correspond reasonably well to the experimental onset potential of cathodic corrosion on Pt(111). For the 3 ML configuration, the top Pt layer is reconstructed by interstitial H atoms to form a well-ordered structure with Pt atoms surrounded by four, five, or six H atoms in roughly square-planar and octahedral coordination patterns. Our work provides insight into the operando surface state during low-potential reduction reactions on Pt(111) and shows a plausible precursor for cathodic corrosion.

Functional programs are well suited for concurrent execution due to the Church-Rosser property. However, for many applications there is still need to explicitly specify process systems instead of relying on compilers. Coordination languages provide one viable solution to the problem of specifying concurrent process systems, since they do not interfere with the properties of the language used for the algorithmic specification of the processes. The coordination language K2 to be proposed in this paper is based on a variant of colored Petri nets. It primarily defines process systems with deterministic behavior, but also allows for controlled forms of non-determinism. Specifications based on Petri nets offer several advantages. An underlying net calculus facilitates formal analysis and verification of basic safety and liveness properties. Graphical representations clearly expose structural dependencies among system components. Our variant of colored Petri nets provides several forms of net abstractions which facilitate the systematic construction of complex systems from small subsystems by composition. Recursive specifications can be used to adapt the structure of a process system to actual problem sizes or to varying input parameters.KeywordsInput StreamOutput StreamLiveness PropertyRecursive ProcessHierarchical ProcessThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.