Abstract

This thesis is about specifying and verifying communications among autonomous and possibly heterogeneous agents, which are the key principle for constructing effective open multi-agent systems (MASs). Effective systems are those that successfully achieve applicability, feasibility, error-freeness and balance between expressiveness and verification efficiency aspects. Over the last two decades, the MAS community has advocated social commitments, which successfully provide a powerful representation for modeling communications in the figure of business contracts from one agent to another. While modeling communications using commitments provides a fundamental basis for capturing flexible communications and helps address the challenge of ensuring compliance with specifications, the designers and business process modelers of the system as a whole cannot guarantee that an agent complies with its commitments as supposed to or at least not wantonly violate or cancel them. They may still wish to first formulate the notion of commitment-based protocols that regulate communications among agents and then establish formal verification (e.g., model checking) by which compliance verification in those protocols is possible.

In this thesis, we address the aforementioned challenges by firstly developing a new branching-time temporal logic---called ACTL*c---that extends CTL* with modal operators for representing and reasoning about commitments and all associated actions. The proposed semantics for ACL (agent communication language) messages in terms of commitments and their actions is formal, declarative, meaningful, verifiable and semi-computationally grounded. We use ACTL*c to derive a new specification language of commitment-based protocols, which is expressive and suitable for model checking. We introduce a reduction method to formally transform the problem of model checking ACTL*c to the problem of model checking GCTL* so that the use of the CWB-NC model checker is possible. We prove the soundness of our reduction method and implement it on top of CWB-NC. To check the effectiveness of our reduction method, we report the verification results of the NetBill protocol and Contract Net protocol against some properties. In addition to the reduction method, we develop a new symbolic algorithm to perform model checking ACTL*c.

To balance between expressiveness and verification efficiency, we secondly adopt a refined fragment of ACTL*c, called CTLC, an extension of CTL with modalities for commitments and their fulfillment. We extend the formalism of interpreted systems introduced to develop MASs with shared and unshared variables and considered agents' local states in the definition of a full-computationally grounded semantics for ACL messages using commitments. We present reasonable axioms of commitment and fulfillment modalities. In our verification technique, the problem of model checking CTLC is reduced into the problems of model checking ARCTL and GCTL* so that respectively extended NuSMV and CWB-NC (as a benchmark) are usable. We prove the soundness of our reduction methods and then implement them on top of the extended NuSMV and CWB-NC model checkers. To evaluate the effectiveness of our reduction methods, we verified the correctness of two business case studies.

We finally proceed to develop a new symbolic model checking algorithm to directly verify commitments and their fulfillment and commitment-based protocols. We analyze the time complexity of CTLC model checking for explicit models and its space complexity for concurrent programs that provide compact representations. We prove that although CTLC extends CTL, their model checking algorithms still have the same time complexity for explicit models, and the same space complexity for concurrent programs. We fully implement the proposed algorithm on top of MCMAS, a model checker for the verification of MASs, and then check its efficiency and scalability using an industrial case study.