Translation from Classical Two-Way Automata to Pebble Two-Way Automata

We study the relation between the standard two-way automata and more powerful devices, namely, two-way finite automata with an additional “pebble” movable along the input tape. Similarly as in the case of the classical two-way machines, it is not known whether there exists a polynomial trade-off, in the number of states, between the nondeterministic and deterministic pebble two-way automata. However, we show that these two machine models are not independent: if there exists a polynomial trade-off for the classical two-way automata, then there must also exist a polynomial trade-off for the pebble two-way automata. Thus, we have an upward collapse (or a downward separation) from the classical two-way automata to more powerful pebble automata, still staying within the class of regular languages. The same upward collapse holds for complementation of nondeterministic two-way machines. These results are obtained by showing that each pebble machine can be, by using suitable inputs, simulated by a classical two-way automaton with a linear number of states (and vice versa), despite the existing exponential blow-up between the classical and pebble two-way machines.

💡 Research Summary

The paper investigates the relationship between classical two‑way finite automata (TWFA) and a more powerful variant, two‑way finite automata equipped with a single movable “pebble” (Pebble‑TWFA). Both models recognize exactly the regular languages, yet the pebble extension is known to give a substantial state‑complexity advantage: a pebble automaton can simulate certain regular languages with exponentially fewer states than a classical two‑way automaton. What remained open was whether the converse direction—simulating a pebble automaton by a classical two‑way automaton—necessarily incurs an exponential blow‑up, and how this interacts with the long‑standing open problem of a polynomial trade‑off between deterministic and nondeterministic two‑way automata.

The authors resolve this by constructing a systematic input‑encoding technique that allows a classical two‑way automaton to “virtually” keep track of the pebble’s position. Given a pebble automaton M and an input word w, they build a new word w′ that contains |w|+1 copies of w separated by special delimiters, each copy annotated with a marker indicating a possible pebble location. A classical two‑way automaton A can then read w′, determine which copy it is currently scanning, and update its internal state to reflect pebble moves performed by M. Crucially, the number of states required by A is only a linear function of the number of states of M; the construction does not depend on the length of w or on the number of pebble moves. Conversely, any classical two‑way automaton can be viewed as a pebble automaton that simply never moves its pebble, so the simulation in the opposite direction is trivial and also linear in the number of states.

Armed with these bidirectional simulations, the paper establishes an “upward collapse” phenomenon: if there exists a polynomial‑size state trade‑off between deterministic and nondeterministic classical two‑way automata, then exactly the same polynomial bound holds for the corresponding pebble automata. In other words, a polynomial simulation for the weaker model automatically yields a polynomial simulation for the stronger model. The same reasoning applies to complementation: a polynomial‑size complementation construction for nondeterministic classical two‑way automata implies a polynomial‑size complementation for pebble automata.

The significance of these results is twofold. First, they demonstrate that the apparent exponential gap between classical and pebble two‑way automata is an artifact of naïve constructions; with appropriate input encoding, the gap collapses to linear. Second, they link two major open problems—polynomial determinization and polynomial complementation of two‑way automata—to the same questions for pebble automata. Consequently, any breakthrough (positive or negative) for the classical model immediately transfers to the pebble model, narrowing the landscape of possibilities.

The paper concludes by suggesting that the techniques may extend to automata with multiple pebbles or other limited memory extensions, and that exploring such extensions could reveal whether similar upward collapses hold in broader contexts.