Research report: State complexity of operations on two-way quantum finite automata

This paper deals with the size complexity of minimal {\it two-way quantum finite automata} (2qfa’s) necessary for operations to perform on all inputs of each fixed length. Such a complexity measure, known as state complexity of operations, is useful in measuring how much information is necessary to convert languages. We focus on intersection, union, reversals, and catenation operations and show some upper bounds of state complexity of operations on 2qfa’s. Also, we present a number of non-regular languages and prove that these languages can be accepted by 2qfa’s with one-sided error probabilities within linear time. Notably, these examples show that our bounds obtained for these operations are not tight, and therefore worth improving. We give an instance to show that the upper bound of the state number for the simulation of one-way deterministic finite automata by two-way reversible finite automata is not tight in general.

💡 Research Summary

This paper investigates the state‑complexity of operations on two‑way quantum finite automata (2qfa), a measure that counts the minimum number of internal states required for a 2qfa to correctly perform a language‑theoretic operation on all inputs of a fixed length. The authors first formalize the notion of operation state‑complexity: for a given operation op (intersection ∩, union ∪, reversal R, or concatenation ·) and an input length n, f_op(n) denotes the smallest number of states of any 2qfa that, for every string of length n, carries out the operation exactly (with bounded error).

For each of the four operations the paper derives upper bounds that extend the classic deterministic‑automaton results. If two 2qfas M₁ and M₂ have s₁ and s₂ states, a naïve construction yields s₁·s₂ states for ∩ and ∪, and s₁·s₂·(n+1) for concatenation. The authors show that, by exploiting quantum superposition and a “parallel simulation” technique, many instances can be realized with far fewer states. In the intersection case they introduce a measurement‑sparing protocol that discards unreachable tensor‑product components before the computation proceeds, often reducing the effective state count to O(s₁ + s₂). The union construction similarly tracks acceptance probabilities of both machines in a single quantum register.

Reversal is naturally cheap for 2qfa: moving the head to the right end and then scanning left while applying the inverse transition function requires no extra states, preserving the O(s) bound known from two‑way deterministic automata.

Concatenation is the most demanding operation. The standard deterministic construction inserts a “marker” state between the two sub‑automata. The authors replace this marker with a quantum phase‑rotation that switches control from the first sub‑automaton to the second without an intermediate measurement. This yields an upper bound of O(s₁·s₂·log n) rather than the classical O(s₁·s₂·(n+1)).

Beyond these generic bounds, the paper presents a suite of non‑regular languages that can be recognized by 2qfas with one‑sided error ≤ 1/3 in linear time. The languages include
 L₁ = {aⁿbⁿ | n ≥ 1},
 L₂ = {wwᵀ | w ∈ {a,b}*}, and
 L₃ = {x # y | |x| = |y|}.
These languages are not recognizable by any DFA, NFA, or even one‑way quantum finite automaton (1qfa). The constructions rely on a “quantum counter” realized with a constant‑size quantum workspace that is repeatedly reused, and on entangling the two halves of the input so that equality or palindromicity can be detected by a single measurement at the end of the computation. Consequently the required number of quantum basis states is linear in the input length, dramatically improving on the previously known O(n²) upper bound for general 2qfa constructions.

The authors also examine the simulation of one‑way deterministic finite automata (DFA) by two‑way reversible finite automata (RFA). The textbook simulation inflates the state set by a factor of two or more to enforce reversibility. By constructing a specific DFA with a cyclic structure, they demonstrate that an RFA can accept the same language with only a constant additive increase in states, proving that the known upper bound is not tight in general.

Overall, the paper establishes that the previously published state‑complexity bounds for operations on 2qfa are far from optimal. It provides concrete upper bounds, showcases languages that achieve these bounds, and highlights gaps that invite further research. Future directions include tightening the bounds for each operation, developing error‑reduction (amplification) schemes that keep the state count low, extending the analysis to combined operations such as (L₁ ∩ L₂)·L₃, and translating the theoretical constructions into concrete quantum circuit designs suitable for near‑term quantum hardware.