In this paper, we investigate the model checking (MC) problem for Halpern and Shoham's modal logic of time intervals (HS) and its fragments, where labeling of intervals is defined by regular expressions. The MC problem for HS has recently emerged as a viable alternative to the traditional (point-based) temporal logic MC. Most expressiveness and complexity results have been obtained by imposing suitable restrictions on interval labeling, namely, by either defining it in terms of interval endpoints, or by constraining a proposition letter to hold over an interval if and only if it holds over each component state (homogeneity assumption). In both cases, the expressiveness of HS gets noticeably limited, in particular when fragments of HS are considered. A possible way to increase the expressiveness of interval temporal logic MC was proposed by Lomuscio and Michaliszyn, who suggested to use regular expressions to define interval labeling, i.e., the properties that hold true over intervals/computation stretches, based on their component points/system states. In this paper, we provide a systematic account of decidability and complexity issues for model checking HS and its fragments extended with regular expressions. We first prove that MC for (full) HS extended with regular expressions is decidable by an automaton-theoretic argument. Though the exact complexity of full HS MC remains an open issue, the complexity of all relevant proper fragments of HS is here determined. In particular, we provide an asymptotically optimal bound to the complexity of the two syntactically maximal fragments AA ̅BB ̅E ̅ and AA ̅EB ̅E ̅, by showing that their MC problem is AEXPpol-complete (AEXPpol is the complexity class of problems decided by exponential-time bounded alternating Turing Machines making a polynomially bounded number of alternations). Moreover, we show that a better result holds for AA ̅BB ̅, AA ̅EE ̅ and all their sub-fragments, whose MC problem turns out to be PSPACE-complete.
