Martin Mundhenk:
Hausdorff Reduktionen zu Mengen mit geringem Informationsgehalt
(Hausdorff reductions to sets of low information content)
(PhD Dissertation'93)

Reductions of $\NP$-complete sets to sparse sets play an important role in structural complexity theory. Since the conjecture of Berman and Hartmanis (1977) that an $\NP$-complete set cannot be sparse, much effort has been spent in research in this topic. Mahaney (1982) proved that sparse $\NP$-hard sets under polynomial-time many-one reducibility cannot exist unless $\P=\NP$. This result was followed by many others deriving under the same hypothesis non-existence of sparse $\NP$-hard sets for more general types of reducibilities. This lead to the independent results of Ogiwara and Watanabe (1991) for bounded truth-table reducibility and Arvind et al. (1992) for the case of conjunctive reducibility. This thesis compares structural properties of sets in the Polynomial Hierarchy with sets which reduce to sparse sets via different types of reducibility. At first the structure of classes defined by reducibilities to sparse sets is considered. The Hausdorff reducibility (Wagner 1980) is for the first time considered in this context. Properties of this reducibility are then used to show that the existence of sparse $\NP$-hard sets under the composition of bounded Hausdorff -- which is the same as bounded truth-table in this case -- and conjunctive reducibility implies $\P=\NP$, generalizing the known results in this area. Weakening from bounded to unbounded Hausdorff reducibility a collaps of the Polynomial Hierarchy to $\P^{\NP}$ is derived. It is shown that this collaps also follows from the existence of sparse $\NP$-hard sets under disjunctive reducibility. All the above mentioned reduction classes to sparse sets are also placed in the $\EL^{p,\theta}_3$-level of the Extended Low-Hierarchy. The results can be interpreted intuitively in a way that the information content of sets reducible to sparse sets is very low compared with sets in the Polynomial Hierarchy. Surprisingly, it also turns out that sets of high information density (defined similar to the ``high sets'' of Book and Lutz 1992) being the natural counterpart of sets of low information content have not more use as oracles. It is shown that sets in ESPACE can be reduced (under the above reducibility types) to sets of high information density if and only if they are reducible to sparse sets.

Back to my homepage