Designing Multi-Agent Systems from Logic Specifications - Distributed Artificial Intelligence, Agent Technology, and Collaborative Applications - page 2

Although CRM is suitable for modeling MAS, it serves as an operational specification language
for MAS and it requires the designers of MAS to have the understanding of chemistry-inspired com-
putational models. As a matter of fact, logic specification of MAS is better suited as a specification
method in the current understanding, because logic specification focuses on behavioral properties of
the systems without concerns with underlying computational model. We propose a method for generat-
ing MAS specifications in γ-Calculus from their logic specifications. We use the “generate-and-test”
method to design the re-write process. Generally speaking, this process generates data in the domain
of logic specification and creates a verification program in γ-Calculus to verify the logic specification
with the generated data.
When applying this method to MASs, there are some problems to solve. The architectural specification
of a MAS is different from that of a normal program, because for a MAS, we need to consider a collec-
tion of aspects, including distribution, security, performance, etc. This will cause a much more complex
synthesizing process. For example, the distribution aspect will cause the communication pattern to be
considered in the synthesizing process. In our approach, the communications are defined by the logic
specifications of the interfaces of the system in terms of either message passing or shared memory. The
practicability of this method is further strengthened by a transformation method we have proposed to
implement CRM specifications on realistic computational models (Lin, 2004; Lin & Yang, 2006).
MASs are considered as complex systems whose design issues are difficult to be handled by logic
systems. By bridging logic specifications and operational specifications of MASs, our study opens a
path to introducing derivative methods in the higher level architectural design of MASs. This work
will help formalize the design processes and promote the current research endeavor to end the state of
MAS design in case-by-case fashion.

Modeling Mul ti-agent syste Ms By che Mical reaction Models

Gamma (Banatre & Le Metayer, 1990 & 1993) is a kernel language in which programs are described
in terms of multiset transformations. In Gamma programming paradigm, programmers can concen-
trate on the logic of problem solving based on an abstract machine and are free from considering any
particular execution environment. It has seeded follow-up elaborations, such as Chemical Abstract
Machine (Cham) (Berry & Boudol, 1992), higher-order Gamma (Le Metayer, 1994), and Structured
Gamma (Fradet & Le Metayer, 1998).
While the original Gamma language is a first-order language, higher order extensions have been
proposed to enhance the expressiveness of the language. These include higher-order Gamma (Le
Metayer, 1994), hmm-calculus (Cohen & Muylaert-Filho, 1996), and others. The recent formalisms,
γ-Calculi, of Gamma languages combine reaction rules and the multisets of data and treat reactions as
first-class citizens (Banatre, Fradet, & Radenac, 2004, 2005a, & 2005b). Among γ-Calculi, γ 0 -Calculus
is a minimal basis for the chemical paradigm; γ c -Calculus extends γ 0 -Calculus by adding a condition
term into γ-abstractions; and γ n -Calculus extends γ 0 -Calculus by allowing abstractions to atomically
capture multiple elements. Finally, γ cn -Calculus combines both γ c -Calculus and γ n -Calculus. For nota-
tional simplicity, we use γ-Calculus to mean γ cn -Calculus from this point on. Also, we assume that the
readers have the basic knowledge about the syntax and semantics of γ-Calculus.
We found that the dynamic nature of distributed agents makes it an ideal object for modeling us-
ing the Gamma languages. An agent shows a combination of a number of characteristics, such as