Synthesis of recursive state machines from libraries of game modules
Abstract
This thesis is focused on synthesis. In formal veri cation synthesis can be
referred to the controller synthesis and the system synthesis. This work
combines both this area of research.
First we focus on synthesizing modular controllers considering game on
recursive game graph with the requirement that the strategy for the protagonist
must be modular. A recursive game graph is composed of a set
of modules, whose vertices can be standard vertices or can correspond to
invocations of other modules and the standard and the set of vertices is
split into two sets each controlled by one of the players. A strategy is
modular if it is local to a module and is oblivious to previous module invocations,
and thus does not depend on the context of invocation. We study
for the rst time modular strategies with respect to winning conditions that
can be expressed languages of pushdown automata. We show that pushdown
modular games are undecidable in general, and become decidable for
visibly pushdown automata speci cations. We carefully characterize the
computational complexity of the considered decision problem. In particular,
we show that modular games with a universal B uchi or co-B uchi visibly
pushdown winning condition are Exptime-complete, and when the winning
condition is given as a CaRet or Nwtl temporal logic formula the
problem is 2Exptime-complete, and it remains 2Exptime-hard even for
simple fragments of these logics. As a further contribution, we present a
di erent synthesis algorithm that runs faster than known solutions for large
speci cations and many exits.
In the second part of this thesis, we introduce and solve a new componentbased
synthesis problem that subsumes the synthesis from libraries of recursive
components introduced by Lustig and Vardi with the modular synthesis
introduced by Alur et al. for recursive game graphs. We model the components
of our libraries as game modules of a recursive game graph with
unmapped boxes, and consider as correctness speci cation a target set of
vertices. To solve this problem, we give an exponential-time xed-point
algorithm that computes annotations for the vertices of the library components
by exploring them backwards. We show a matching lower-bound via a
direct reduction from linear-space alternating Turing machines, thus proving
Exptime-completeness. We also give a second algorithm that solves
this problem by annotating in a table the result of many local reachability
game queries on each game component. This algorithm is exponential only
in the number of the exits of the game components, and thus shows that
the problem is xed-parameter tractable.
Finally, we study a more general synthesis problem for component-based
pushdown systems, the modular synthesis from a library of components
(Lms). We model each component as a game graph with boxes as placeholders
for calls to components, as in the previous model, but now the
library is equipped also with a box-to-component map that is a partial function
from boxes to components. An instance of a component C is essentially
a copy of C along with a local strategy that resolves the nondeterminism of
pl 0. An RSM S synthesized from a library is a set of instances along with a
total function that maps each box in S to an instance of S and is consistent
with the box-to-component map of the library. We give a solution to the
Lms problem with winning conditions given as internal reachability objectives,
or as external deterministic nite automata (FA) and deterministic
visibly pushdown automata (VPA) (6). We show that the Lms problem is
Exptime-complete for any of the considered speci cations. [edited by Author]