next.png up.png previous
Next: Symbolic Representation via Constrained Up: An Overview of MSR(): Previous: An example: The Ticket

An Assertional Language for $\mbox{MSR($\ {\mathscr C} $)}$

Since we can easily encode the formalism of two counters machines into $\mbox{MSR($\ {\mathscr C} $)}$ , reachability problems are in general undecidable for MSR($ {\mathscr C} $) specifications. Even for fragments of $\mbox{MSR($\ {\mathscr C} $)}$ that are not Turing powerful, the reachability set of an $\mbox{MSR($\ {\mathscr C} $)}$ specification might be infinite and thus it might be extremely difficult to explore it. The source of infiniteness may be either the increasing size (number of elements) of generated configurations, or the unboundedness of the values attached to atomic formulas (e.g. think about a simple counter incremented at every rule application), or both. The only possibility of automatically analyzing the behavior of this infinite-state specification consists of finding adequate finite, symbolic representations of infinite collections of configurations. In order to achieve this goal, in this section we will introduce an assertional language based on the notion of upward-closed sets of configurations. To explain this idea, let us consider again the Example 1. Sup.pngpose we are interested in proving that the specification of the example discussed in the previous sections satisfies the following invariant: all reachable states satisfy the mutual exclusion property only one process per time is in state use. Instead of proving it directly, we can try to disprove it as follows: we first select all possible configurations that violate it, and then show that they are not reachable from the initial state $init$. The set of interesting violations $U$ consists of configurations like $use(v)~\vert~use(w)~\oplus~U'$ where $U'$ is any possible multiset. Then, we note that $U$ can be generated from the (infinite) set of configurations $M=\{
use(v)~\vert~use(w)~v\geq 0,w\geq 0\}$ by taking the upward closure with respect to multiset inclusion: if a configuration contains any ground instance of $M$, then it violates itself the invariant. As discussed in [1], the negation of an invariant property often enjoys the property of being upward closed with respect to an inclusion ordering defined over sets of configurations. As a first approximation of the problem of exploring an infinite state space, we could then start the exploration from the unsafe states, using minimal violations to represent them, and reason backward by applying the predecessor operator $Pre$ (i.e., computing weakest pre-conditions). Let us formally define the notion of upward closed sets.

Let $\langle { {\mathscr P} , {\mathscr C} , {\mathscr I} , {\mathscr R} } \rangle $ be an $\mbox{MSR($\ {\mathscr C} $)}$ specification, and $S$ a set of configurations. The upward closure of $S$, denoted $\mathit {Up}({S})$, is defined as

\begin{displaymath}\mathit {Up}({S})=\{ {\mathcal N} \ \vert\ {\mathcal M} \preccurlyeq {\mathcal N} , {\mathcal M} \in S\}.\end{displaymath}

We say that $S$ is upward closed if $S=\mathit {Up}({S})$. Upward-closed sets of configurations have interesting properties with respect to the predecessor operator $Pre$.
$Up(Pre(S))\subseteq Pre(Up(S))$\ for any set $S$\ of configurations.
In general the reverse implication does not hold.
Consider $p~\longrightarrow~q_1~\vert~q_2$\ and the singleton s...
whereas the multiset $p$\ belongs to $Pre(Up(S))$.
However, the following property holds.
If $S$\ is upward-closed, then $Up(Pre(S))=Pre(Up(S))$.
In other words, the class of upward-closed sets of configurations is closed under the computation of the pre-image. The previous properties do not suffice to finitely represent and manipulate sets upward closed sets of configurations. In fact, we still have to solve the problem of finitely represent what we called the minimal violations. This is not always possible. However, in the example of the single server ticket algorithm they have a regular structure that we can exploit by translating the extensional description

use(v)~\vert~use(w)~v\geq 0,w\geq 0\}

into the following intensional description, called a constrained configuration


where $x,y$ are free variables. In other words, by annotating a non ground configuration (i.e., in which formulas have free variables) with a constraint we can implicitly represent all unsafe configurations we are interested in for our example. They will correspond to the upward closure of the set of ground instances of the constrained configuration above. It is important to note that this techniques is just an heuristics through which we try to embed as many configurations as possible inside a regular structure. In the following section we will formalize these ideas and show how to use them in combination with the backward reachability approach proposed in [1].

next.png up.png previous
Next: Symbolic Representation via Constrained Up: An Overview of MSR(): Previous: An example: The Ticket
Giorgio Delzanno 2003-02-03