Logic programming under the stable model semantics is proposed as a non-monotonic language for knowledge representation and reasoning in artificial intelligence. In this paper, we explore and extend the notion of compatibility and the A operator, which were first proposed by Zhang to characterize default theories. First, we present a new characterization of stable models of a logic program and show that an extended notion of compatibility can characterize stable submodels. We further propose the notion of weak auto-compatibility which characterizes the Normal Forward Chaining Construction proposed by Marek, Nerode and Remmel. Previously, this construction was only known to construct the stable models of FC-normal logic programs, which turn out to be a proper subclass of weakly auto-compatible logic programs. We investigate the properties and complexity issues for weakly auto-compatible logic programs and compare them with some subclasses of logic programs.