Preprocessing for Invariant Validation

Hoare's logic and Dijkstra's predicate transformer calculus have proved adequate for reducing the correctness problem for programs to the validity problem for logical formulas. However, the size of the logical formulas to be validated grows faster than the size of the program , and, even in the propositional case, the validation problem is NP-complete and becomes practically intractable for large programs. We introduce a strategy for dealing with this problem. The principle is to write the formulas in the form (h 1 ^ ^ h n)) c, and to use eeciently computable criteria to select a small subset I f1; : : :; ng such that c remains a logical consequence of H I = fh i : i 2 Ig. These criteria are motivated and the eeciency of the method is investigated.