Tel-Aviv University - Computer Science Colloquium

Sunday, Nov 27, 2005, 11:15-12:15

Room 309
Schreiber Building

--------------------------------------------------------------------------------

Nir Piterman

Ecole Polytechnique Fédérale de Lausanne

Title:

Revisiting Safra's Determinisation of Automata on Infinite Words


Abstract:

Determinization and complementation are fundamental notions in

computer science.

When considering finite automata on finite words determinization

gives also a solution to complementation.

Given a nondeterministic finite automaton there exists an

exponential construction that gives a deterministic automaton for the

same language.

Dualizing the set of accepting states gives an automaton for

the complement language.

In the theory of automata on infinite words, determinization and

complementation are much more involved.

Safra provides determinization constructions for B\"uchi and

Streett automata that result in deterministic Rabin automata.

For a B\"uchi automaton with $n$ states, Safra constructs a

deterministic Rabin automaton with $n^{O(n)}$ states and $n$ pairs.

For a Streett automaton with $n$ states and $k$ pairs, Safra

constructs a deterministic Rabin automaton with $(nk)^{O(nk)}$ states

and $n(k+1)$ pairs.

 

Here, we reconsider Safra's determinization constructions.

We show how to construct automata with fewer states

and, most importantly, parity acceptance condition.

Specifically, starting from a nondeterministic B\"uchi automaton with

$n$ states our construction yields a deterministic parity automaton

with $n^{2n+2}$ states and index $2n$ (instead of a Rabin automaton

with $\safrastates$ states and $n$ pairs).

Starting from a nondeterministic Streett automaton with $n$ states and

$k$ pairs our construction yields a deterministic parity automaton

with $n^{n(k+2)+2}(k{+}1)^{2n(k+1)}$ states and index $2n(k+1)$ (instead

of a Rabin automaton with $(12)^{n(k+1)}n^{n(k+2)}(k{+}1)^{2n(k+1)}$

states and $n(k+1)$ pairs).

The parity condition is much simpler than the Rabin condition.

In applications such as solving games and emptiness of tree automata

handling the Rabin condition involves an additional multiplier of

$n^2n!$ (or $(n(k+1))^2(n(k+1))!$ in the case of Streett) which is saved

using our construction.