R1-2H: 1-Limited Reversible 2-Head Automata

• 1-limited reversible 2-head finite automata are defined by single-symbol moves with one head at a time, ensuring both forward and backward determinism.
• They classify states by fixed head transitions, which enforces unambiguous predecessor configurations and limits control flexibility.
• The model establishes a strict hierarchy (2CrevLIN ⊊ 2rev1LIN ⊊ 2revLIN ⊊ 2detLIN), highlighting trade-offs between reversibility and expressive power.

A 1-limited reversible 2-head finite automaton is a two-head device on a single read-only input word in which one head starts at the left end and moves left-to-right, the other starts at the right end and moves right-to-left, the global computation is deterministic in both the forward and backward directions, and every step consumes exactly one symbol by exactly one of the heads. In the notation of "On some Classes of Reversible 2-head Automata" (Nagy et al., 21 Jul 2025), this model is denoted R1-2H, and the associated language family is $2\mathrm{rev1LIN}$. The model occupies a strictly intermediate position inside the hierarchy of deterministic, reversible, and complete reversible 2-head automata, while retaining enough power to recognize some characteristic linear languages, including palindromes (Nagy et al., 21 Jul 2025).

1. Formal model and computation semantics

The underlying 2-head model is defined by a 5-tuple

$A=(Q,q_0,V,\delta,F),$

where $Q$ is a finite set of states, $q_0\in Q$ is the initial state, $F\subseteq Q$ is the set of accepting states, $V$ is the input alphabet, and

$$\delta:Q\times (V\cup\{\lambda\})\times (V\cup\{\lambda\})\to 2^Q.$$

The first head starts at the left end of the input and moves only left-to-right; the second head starts at the right end and moves only right-to-left. A configuration is written as $(q,w')$, where $q$ is the current state and $w'$ is the unread substring between the two heads. If the current configuration is $A=(Q,q_0,V,\delta,F),$0, then a simultaneous inward move of both heads is possible exactly when

$A=(Q,q_0,V,\delta,F),$1

If only one head moves, the labels are $A=(Q,q_0,V,\delta,F),$2 or $A=(Q,q_0,V,\delta,F),$3; the paper abbreviates these by $A=(Q,q_0,V,\delta,F),$4 and $A=(Q,q_0,V,\delta,F),$5, respectively. A computation on input $A=(Q,q_0,V,\delta,F),$6 starts in $A=(Q,q_0,V,\delta,F),$7, ends when the unread substring becomes empty, and accepts precisely when

$A=(Q,q_0,V,\delta,F),$8

The deterministic version, D-2H, is characterized by the absence of $A=(Q,q_0,V,\delta,F),$9-moves, pointwise uniqueness of transitions, and separation conditions preventing conflicting uses of the same left or right symbol from a single state. The family of languages accepted by deterministic 2-head automata is denoted $Q$0 (Nagy et al., 21 Jul 2025).

This formulation is distinctive because the automaton does not scan left-to-right in the ordinary one-way sense. Its state evolution is organized around the shrinking middle segment, and the semantics of reversibility are therefore naturally expressed in terms of unique predecessors of configurations with shorter unread substrings.

2. Reversibility and the 1-limited restriction

Backward determinism requires that every reachable configuration have at most one predecessor configuration. Formally, for all $Q$1, $Q$2, and $Q$3, there is at most one longer word $Q$4 of the form $Q$5, $Q$6, or $Q$7, and at most one state $Q$8, such that

$Q$9

A reversible 2-head automaton, R-2H, is exactly a deterministic 2-head automaton that is also backward deterministic. The resulting language family is $q_0\in Q$0 (Nagy et al., 21 Jul 2025).

For deterministic 2-head automata, reversibility admits a structural characterization. If two transitions lead to the same target state,

$q_0\in Q$1

then reversibility holds exactly when $q_0\in Q$2 or $q_0\in Q$3, and simultaneously $q_0\in Q$4 and $q_0\in Q$5. Here $q_0\in Q$6 denotes the length of the concatenation $q_0\in Q$7; thus the last two conditions exclude ambiguous inverse interpretations arising from both sides being $q_0\in Q$8 in the wrong positions (Nagy et al., 21 Jul 2025).

The 1-limited restriction strengthens this model by forbidding any move in which both heads consume symbols simultaneously. Every transition must be of the form

$q_0\in Q$9

equivalently,

$F\subseteq Q$0

A 1-limited reversible 2-head automaton is therefore an R-2H whose steps are exclusively one-symbol left-head moves or one-symbol right-head moves. This restriction is strict: R1-2H is a proper subclass of the general reversible model (Nagy et al., 21 Jul 2025).

A recurrent misconception is that reversibility already forces one-symbol, one-head behavior. The hierarchy disproves this. General reversible 2-head automata may use genuine two-head moves $F\subseteq Q$1, whereas 1-limited reversible automata categorically may not.

3. State classification in 1-limited reversible automata

A central structural result for R1-2H is that states can be classified according to which head is used by all incoming transitions and which head is used by all outgoing transitions. The paper distinguishes the left head by $F\subseteq Q$2, the right head by $F\subseteq Q$3, and the absence of such transitions by $F\subseteq Q$4 (Nagy et al., 21 Jul 2025).

Class	Meaning
$F\subseteq Q$5	entered by left-head moves, exited by left-head moves
$F\subseteq Q$6	entered by left-head moves, exited by right-head moves
$F\subseteq Q$7	entered by left-head moves, no outgoing transitions
$F\subseteq Q$8	entered by right-head moves, exited by left-head moves
$F\subseteq Q$9	entered by right-head moves, exited by right-head moves
$V$0	entered by right-head moves, no outgoing transitions
$V$1	no incoming transitions, exited by left-head moves
$V$2	no incoming transitions, exited by right-head moves

The ninth formal possibility, $V$3, is excluded by the assumption that states are reachable and useful. Moreover, only one of $V$4 or $V$5 can occur in a given automaton, because there is only one initial state without incoming transitions. Consequently, the state set can be partitioned into at most seven disjoint classes.

Theorem 4.1 states that if an R1-2H accepts a nonempty language, then every state belongs to one such class. The initial state may lie in any of

$V$6

an accepting state may lie in any of

$V$7

and any other state must lie in one of the four central classes

$V$8

The proof mechanism is direct. Backward determinism forces all incoming transitions of a reachable state to use the same head; otherwise some reachable configuration would admit two predecessors of different types. Forward determinism forces all outgoing transitions of a state to use the same head; otherwise a configuration could enable both a left-head and a right-head step. The full characterization in Theorem 4.2 adds symbolwise injectivity conditions: for each state and symbol, there is at most one outgoing transition of the permitted head type, and for each target state and symbol, there is at most one predecessor state of the permitted incoming type (Nagy et al., 21 Jul 2025).

These results make R1-2H unusually rigid. The machine cannot use a state as an arbitrary control nexus; the state itself already encodes a fixed entering head and a fixed leaving head.

4. Expressive power and separating languages

The family $V$9 is strictly smaller than $$\delta:Q\times (V\cup\{\lambda\})\times (V\cup\{\lambda\})\to 2^Q.$$0, but it is not confined to regular languages. The paper explicitly identifies palindromes as a characteristic linear language accepted by reversible 2-head automata, and the discussion of the 1-limited variant states that this can be achieved by alternating left and right moves while storing the required information reversibly in the finite control. It also gives a concrete 1-limited reversible automaton for

$$\delta:Q\times (V\cup\{\lambda\})\times (V\cup\{\lambda\})\to 2^Q.$$1

using two final states $$\delta:Q\times (V\cup\{\lambda\})\times (V\cup\{\lambda\})\to 2^Q.$$2 and $$\delta:Q\times (V\cup\{\lambda\})\times (V\cup\{\lambda\})\to 2^Q.$$3 with transitions

$$\delta:Q\times (V\cup\{\lambda\})\times (V\cup\{\lambda\})\to 2^Q.$$4

This automaton is 1-limited and reversible, but not complete (Nagy et al., 21 Jul 2025).

The main negative result for R1-2H is Lemma 4.3: $$\delta:Q\times (V\cup\{\lambda\})\times (V\cup\{\lambda\})\to 2^Q.$$5 is not accepted by any 1-limited reversible 2-head automaton. In contrast, the same language belongs to $$\delta:Q\times (V\cup\{\lambda\})\times (V\cup\{\lambda\})\to 2^Q.$$6, because a general reversible 2-head automaton may use simultaneous two-head moves such as $$\delta:Q\times (V\cup\{\lambda\})\times (V\cup\{\lambda\})\to 2^Q.$$7 to match a $$\delta:Q\times (V\cup\{\lambda\})\times (V\cup\{\lambda\})\to 2^Q.$$8 in the prefix against a $$\delta:Q\times (V\cup\{\lambda\})\times (V\cup\{\lambda\})\to 2^Q.$$9 in the suffix. This establishes the proper inclusion

$(q,w')$0

A second separation concerns reversibility itself. The regular language

$(q,w')$1

belongs to $(q,w')$2 but is not accepted by any reversible 2-head automaton. Therefore

$(q,w')$3

and, by inclusion, $(q,w')$4 also lies outside $(q,w')$5 (Nagy et al., 21 Jul 2025).

Language	Status	Role
palindromes	in reversible 2-head classes	witnesses non-regular power
$(q,w')$6	in $(q,w')$7, not in $(q,w')$8	separates complete from incomplete 1-limited reversibility
$(q,w')$9	in $q$0, not in $q$1	separates general reversible from 1-limited reversible
$q$2	in $q$3, not in $q$4	separates deterministic from reversible

The resulting picture is that 1-limited reversibility is neither synonymous with regularity nor with the full reversible 2-head power. It recognizes some non-regular linear languages, yet it loses languages that become accessible as soon as simultaneous two-head moves are permitted.

5. Complete reversible automata and the proper hierarchy

A complete reversible 2-head finite automaton is one in which every nonempty configuration has exactly one successor step. Formally, for every nonempty word $q$5 and every state $q$6, there is exactly one pair $q$7 such that

$q$8

This family is denoted $q$9 (Nagy et al., 21 Jul 2025).

Completeness imposes a strong restriction absent from general reversibility. Proposition 5.1 states that no complete reversible 2-head finite automaton can have a transition of the form $w'$0 with $w'$1. Hence every complete reversible 2-head automaton is automatically 1-limited. The converse fails. The language $w'$2 above is accepted by a 1-limited reversible automaton but not by any complete reversible 2-head automaton, so

$w'$3

Theorem 5.2 adds a graph-theoretic condition: every complete 1-limited reversible 2-head automaton has a strongly connected underlying directed graph of states. This strong connectivity is an essential ingredient in the proof that $w'$4 cannot be recognized under completeness. The argument exploits the fact that completeness forces transitions for all relevant leading or trailing symbols, while strong connectivity prevents the machine from isolating nonaccepting behavior in an unreachable or dead region of the graph (Nagy et al., 21 Jul 2025).

The hierarchy established in the paper is therefore

$w'$5

The same discussion also notes that complete reversible automata are closed under complement, and that reversal is available for all four language families considered there.

One useful correction to a common simplification follows immediately: “1-limited” and “complete” are not interchangeable. Completeness implies 1-limitedness, but the inclusion is proper.

6. Relations to linear grammars and adjacent reversible or quantum models

The paper compares these automata with left deterministic linear grammars. A left deterministic linear grammar is a linear grammar

$w'$6

whose productions are of the form

$w'$7

with a determinism condition requiring that if

$w'$8

are both present, then $w'$9 and $A=(Q,q_0,V,\delta,F),$00. The corresponding language family is denoted LDLL. Theorem 6.2 states that every left deterministic linear language is accepted by a deterministic 2-head automaton, and that

$A=(Q,q_0,V,\delta,F),$01

No equality or strict inclusion relation between LDLL and $A=(Q,q_0,V,\delta,F),$02 is claimed. The discussion instead indicates that the classes are different in character: palindromes provide evidence that $A=(Q,q_0,V,\delta,F),$03 reaches beyond ordinary left-deterministic behavior, while some LDLL production patterns do not appear to be realizable by reversible 2-head automata (Nagy et al., 21 Jul 2025).

The broader literature cited in the supplied corpus places R1-2H among several neighboring notions of reversibility. In "Multi-head Watson-Crick quantum finite automata" (Ganguly et al., 2020), one-way reversible multi-head finite automata are defined by injectivity of predecessor configurations together with a fixed movement vector for all transitions entering a state; multi-head Watson-Crick quantum finite automata, by contrast, use unitary evolution on double-stranded inputs and can accept languages beyond all one-way deterministic multi-head finite automata, including

$A=(Q,q_0,V,\delta,F),$04

Similarly, "2-tape 1-way Quantum Finite State Automata" (Ganguly et al., 2016) shows that 2-tape 1-way quantum finite state automata can accept all regular languages and can also accept $A=(Q,q_0,V,\delta,F),$05, while every one-way reversible multi-head automaton is embedded only as a special case of the quantum model.

These comparisons must be interpreted carefully. The R1-2H model processes one input from both ends on a single tape and stops when the heads meet or cross, whereas the automata in (Ganguly et al., 2020) and (Ganguly et al., 2016) are one-way, multi-head or multi-tape, and quantum or Watson-Crick in structure. This suggests that the limitations of $A=(Q,q_0,V,\delta,F),$06 are specific to the interaction between two-ended scanning, backward determinism, and the prohibition of simultaneous two-head moves, rather than to reversibility in the abstract.