LARA

Problem 1

Galois connection is defined by two monotonic functions $\alpha : C \to A$ and $\gamma : A \to C$ between partial orders $\leq$ on $C$ and $\sqsubseteq$ on $A$, such that

\begin{equation*} \alpha(c) \sqsubseteq a\ \iff\ c \leq \gamma(a) \qquad (*) \end{equation*}

for all $c$ and $a$ (intuitively, the condition means that $c$ is approximated by $a$).

Part a) Show that the condition $(*)$ is equivalent to the conjunction of these two conditions:

\begin{eqnarray*} c &\leq& \gamma(\alpha(c)) \\ \alpha(\gamma(a)) &\sqsubseteq& a \end{eqnarray*}

hold for all $c$ and $a$.

Part b) Let $\alpha$ and $\gamma$ satisfy the condition of Galois connection. Show that the following three conditions are equivalent:

Part c) State the condition for $c=\gamma(\alpha(c))$ to hold for all $c$. When $C$ is the set of sets of concrete states and $A$ is a domain of static analysis, is it more reasonable to expect that $c=\gamma(\alpha(c))$ or $\alpha(\gamma(a)) = a$ to be satisfied, and why?

Problem 2

Suppose you are given a set of predicates ${\cap P} = \{P_0,P_1,\ldots,P_n\}$ in a decidable theory of first-order logic (for example, quantifier-free formulas in the combination of uninterpreted function symbols with integer linear arithmetic) where $P_0$ is the predicate 'false'.

Part a) Consider Conjunctions of Predicates as abstract interpretation domain. Give example showing that it need not be the case that

\begin{equation*} a_1 \leq a_2 \leftrightarrow \gamma(a_1) \subseteq \gamma(a_2) \end{equation*}

Part b) Describe how to construct from $A$ a new, smaller, lattice $B$, where the above equivalence holds. Is there an algorithm to compute $B$ and the partial order on $B$ using a decision procedure for the logic of predicates?

Part c) Suppose that, for the same set of predicates, we use lattice $A$ and lattice $B$ to compute the fixpoints $g_A$ and $g_B$ of the function $F^{\#}$ from Abstract Interpretation Recipe. What can you say about

  • the comparison of numbers of iterations needed to compute the fixpoint $g_A$ and $g_B$ (is one always less than the other, can they be equal, does it depend on set ${\cal P}$ of predicates)
  • the precision of computed information, that is, comparison of sets $\gamma(g_A(p))$ and $\gamma(g_B(p))$ for an arbitrary program point $p \in V$.