This dissertation contributes to two major research areas in safety-critical software systems, namely, software analysis, and software implementation. In reference to the software analysis problem, the main contribution of the dissertation is the development of a novel framework, based on Lyapunov invariants and convex optimization, for verification of various safety and performance specifications for software systems. The enabling elements of the framework for software analysis are: (i) dynamical system interpretation and modeling of computer programs, (ii) Lyapunov invariants as behavior certificates for computer programs, and (iii) a computational procedure for finding the Lyapunov invariants. (i) The view in this dissertation is that software defines a rule for iterative modification of the operating memory at discrete instances of time. Hence, it can be modeled as a discrete-time dynamical system with the program variables as the state variables, and the operating memory as the state space. Three specific modeling languages are introduced which can represent a broad range of computer programs of interest to the control community. These are: Mixed Integer-Linear Models, Graph Models, and Linear Models with Conditional Switching. (ii) Inspired by the concept of Lyapunov functions in stability analysis of nonlinear dynamical systems, Lyapunov invariants are introduced and proposed for analysis of behavioral properties, and verification of various safety and performance specifications for computer programs. In the same spirit as standard Lyapunov functions, a Lyapunov invariant is an appropriately defined function of the state which satisfies a difference inequality along the trajectories. It is shown that variations of Lyapunov invariants satisfying certain technical conditions can be formulated for verification of several common specifications.
展开▼
