Next: Bibliography
Up: Heuristics for the Definition
Previous: Saturation of a Theory
Contents
Establishing Complexity Bounds
Three aspects are of importance when using Saturate to
derive complexity bounds for entailment.
- We need to use an ordering in which ground terms have at
most finitely many smaller terms. An lpo is, hence, in most cases
not suitable.
- The treatment of transitive relations by chaining has to be turned off
(cf. flag settings) to invoke ordered
resolution also for transitive relations and, in particular, for
the transitivity axiom itself.
- No selection of negative literals is allowed, cf. command sass.
For establishing a polynomial complexity
bound, the (precedence-modified)
subterm ordering sto should be chosen
with an appropriate
ordering declaration.
Finitely saturating a theory under an instance of this class of partial orderings
(actually one saturates with respect to all possible well-founded
completions of
that instance simultaneously)
proves that the entailment problem for the respective theory
is decidable in polynomial time if the theory is Horn, and in CoNP,
if it is not Horn.
We refer to (Basin & Ganzinger, 1995)
for a detailed explanation of the theoretical basis.
The subdirectory
LocalTheories
of saturate/examples shows examples that we have saturated with
respect to the sto, including the theory of term
embedding,
congruence closure,
various fragments of
propositional calculus,
and the largest PTIME fragment of Allen's interval logic: the theory
of partial nonstrict orderings.
A formulation of sequent calculus
has been saturated modulo AC and with respect to a linear interpretation
function as ordering. The details are explained in the respective file.
Term orderings via linear interpretations are available only
in the Saturate version under SICStus 3.0. Such interpretations
may be useful when one
intends to prove exponential upper bounds for the complexity of
entailment problems.
Proving the locality of a theory usually requires clausal rewriting with
an appropriate setting of
sarp.
in addition to the usual simplification and deletion techniques.
In some cases hereditary ordering constraints with sass([2]) will be useful.
Next: Bibliography
Up: Heuristics for the Definition
Previous: Saturation of a Theory
Contents
Harald Ganzinger
2002-12-04
Imprint | Data Protection