2NFAs with Common Guess

• The paper introduces 2NFAs with common guess, a model that uses a single global certificate to separate nondeterministic annotation from deterministic verification.
• It leverages a self-verifying canonical annotation scheme, enabling polynomial state complexity for complementation and efficient simulation of 1-limited automata.
• The model establishes substantial lower bounds in the unary setting, demonstrating a clear separation between common-guess mechanisms and standard nondeterministic automata.

Searching arXiv for the cited papers and closely related work on common-guess automata and limited automata. arXiv_search(query="all:\"2NFA with common guess\" OR all:\"common guess\" automata OR all:\"1-limited automata\" unary", max_results=10, sort_by="submittedDate") arXiv_search(query="id:(Guillon et al., 15 Jul 2025) OR id:(Guillon et al., 11 Apr 2025) OR id:(0811.2586)", max_results=10, sort_by="relevance") A 2NFA with common guess is a two-way finite-state device in which nondeterminism is concentrated into a single shared annotation, or “guess,” fixed for the whole input before the read-only verification phase begins. In one formulation, the machine is a 2NFA equipped with a new kind of nondeterminism that initially annotates each input symbol and then performs a read-only computation over the resulting annotated word (Guillon et al., 11 Apr 2025). In another, closely related formulation, it is a restricted 1-limited automaton operating on an input paired with a global annotation track and verifying that annotation incrementally (Guillon et al., 15 Jul 2025). The defining feature is that the guessed object is common across the computation, rather than re-chosen independently at each nondeterministic branch. A broader antecedent appears in the auxiliary-memory framework of automata with read-only guessed memory, where acceptance means that there exists a single memory content on which an otherwise deterministic machine accepts (0811.2586).

1. Definition and semantic core

The common-guess model is introduced in the recent literature as part of the study of 1-limited automata and two-way automata. In the formulation used in "Nondeterminism makes unary 1-limited automata concise" (Guillon et al., 11 Apr 2025), a 2NFA with common guess ($2\mathrm{NFA}+cg$) is a 2NFA that may initially annotate each input symbol and then run a read-only computation on the annotated word. This separates computation into two phases: an annotation phase and a verification phase.

A more concrete encoding is given in "Polynomial Complementation of Nondeterministic 2-Way Finite Automata by 1-Limited Automata" (Guillon et al., 15 Jul 2025). There the input is viewed as a word

$x \in (\Sigma\times\{0,1\})^*,$

with projection maps $\pi_1(x)$ for the original input and $\pi_2(x)$ for the annotation bits. The “common guess” is the entire second track. It is global and shared across the whole input, rather than produced locally at each step.

This semantic structure distinguishes common-guess automata from ordinary nondeterministic automata. In standard nondeterminism, branching is embedded in the transition relation. In the common-guess setting, the nondeterministic resource is a single guessed object that all later verification steps consult. The auxiliary-memory model of Kintali and others makes this distinction explicit: an automaton with auxiliary memory is deterministic relative to a fixed memory content $\mu$, and a word is accepted iff there exists some $\mu$ making the run accepting (0811.2586). This suggests that common-guess automata are best understood as certificate-verifying two-way devices rather than merely as branching automata.

2. Relation to 1-limited automata and deterministic counterparts

The model is studied in direct connection with 1-limited automata (1-LAs), which are presented as an extension of two-way finite automata characterizing regular languages (Guillon et al., 11 Apr 2025). Within that landscape, 2NFAs with common guess are treated as a restricted form of 1-LA (Guillon et al., 15 Jul 2025).

Model	Characterization in the cited literature
$2\mathrm{NFA}+cg$	2NFA with an initial common annotation phase (Guillon et al., 11 Apr 2025)
$2\mathrm{DFA}+cg$	Deterministic verification phase with a nondeterministic annotation phase (Guillon et al., 11 Apr 2025)
1-LA	Extension of 2NFAs; $2\mathrm{NFA}+cg$ appears as a restricted form (Guillon et al., 15 Jul 2025)

The deterministic counterpart is especially notable. The cited work states that two-way deterministic finite automata with common guess ($2\mathrm{DFA}+cg$) still possess a nondeterministic annotation phase and can be considered a restriction of 1-LAs (Guillon et al., 11 Apr 2025). Accordingly, “deterministic” here does not mean fully deterministic computation in the usual DFA sense. The deterministic aspect applies to the read-only verification phase, while the annotation itself remains guessed.

This distinction matters in descriptional complexity. A model may be deterministic during verification while still drawing substantial succinctness from the guessed annotation. A common misconception is therefore to equate $x \in (\Sigma\times\{0,1\})^*,$0 with ordinary 2DFAs; the cited results rule that out directly, because the common-guess phase carries nontrivial computational content (Guillon et al., 11 Apr 2025).

3. Canonical annotations and self-verification

The most explicit operational account of common-guess automata is the self-verifying construction in (Guillon et al., 15 Jul 2025). Let $x \in (\Sigma\times\{0,1\})^*,$1 be an automaton with state set $x \in (\Sigma\times\{0,1\})^*,$2. For a word $x \in (\Sigma\times\{0,1\})^*,$3, the annotation is organized in blocks of length $x \in (\Sigma\times\{0,1\})^*,$4, and a subset $x \in (\Sigma\times\{0,1\})^*,$5 is encoded by a binary word $x \in (\Sigma\times\{0,1\})^*,$6 of length $x \in (\Sigma\times\{0,1\})^*,$7 such that

$x \in (\Sigma\times\{0,1\})^*,$8

The annotation records reachable-state sets for prefixes of the input. If $x \in (\Sigma\times\{0,1\})^*,$9 denotes the set of states reachable after reading $\pi_1(x)$0, then the $\pi_1(x)$1-th annotation block encodes

$\pi_1(x)$2

The canonical annotation $\pi_1(x)$3 is defined recursively by decomposing $\pi_1(x)$4 with $\pi_1(x)$5, and appending either a full reachable-set encoding or trailing zeros, depending on whether the current prefix ends at a block boundary (Guillon et al., 15 Jul 2025).

Verification is performed by finite-state procedures that check local consistency of the global guess. The paper identifies the subprocedures $\pi_1(x)$6, $\pi_1(x)$7, and $\pi_1(x)$8. The central inductive relation used in the update step is

$\pi_1(x)$9

Thus the common guess does not merely assert acceptance or rejection; it encodes the entire evolution of reachable-state sets and is locally checked block by block.

The same construction is described as self-verifying because correctness is tied to the canonical annotation. The acceptance and rejection requirements are stated as

$\pi_2(x)$0

$\pi_2(x)$1

Malformed annotations need not receive meaningful acceptance semantics. The guessed annotation is therefore a certificate whose role is not merely existential but also verifiable in a strong sense.

4. Simulation and complementation results

The common-guess model is central to recent complementation results. The 2025 paper on polynomial complementation proves that every unrestricted 2NFA can be complemented by a 1-LA with only polynomial increase in size, and that the resulting machine is in fact a restricted 1-LA, namely a self-verifying $\pi_2(x)$2 (Guillon et al., 15 Jul 2025). In the more detailed state-complexity statement extracted from the same work, every $\pi_2(x)$3-state 1-LA has an equivalent self-verifying $\pi_2(x)$4 with polynomially many states, specifically $\pi_2(x)$5, using only $\pi_2(x)$6 annotation symbols (Guillon et al., 15 Jul 2025).

A corollary concerns complementation of 1-LAs. If an $\pi_2(x)$7-state 1-LA recognizes $\pi_2(x)$8, then there exists a 1-LA recognizing $\pi_2(x)$9 with a single-exponential number of states in $\mu$0 and with $\mu$1 work symbols (Guillon et al., 15 Jul 2025). The derivation combines a known simulation of an $\mu$2-state 1-LA by a machine with at most

$\mu$3

states and then applies the polynomial common-guess construction, yielding

$\mu$4

The abstract further states that a single exponential is both necessary and sufficient for complementing 1-LAs (Guillon et al., 15 Jul 2025).

These results make the common-guess model structurally important: it provides a compact certificate format for reachable-set evolution, allowing complementation to be reduced to local verification. A plausible implication is that common-guess structure is not merely a technical variant of 1-LAs but a particularly effective normal form for verification-based constructions.

5. Unary lower bounds and descriptional complexity

The common-guess model is also used to prove strong succinctness separations. The 2025 paper "Nondeterminism makes unary 1-limited automata concise" studies the descriptional complexity of several 1-LA variants and establishes exponential lower bounds for simulations of $\mu$5 by deterministic 1-LAs and by ordinary 2NFAs (Guillon et al., 11 Apr 2025). These lower bounds are derived from a doubly exponential lower bound for simulation of $\mu$6 by one-way deterministic finite automata.

The witnesses are unary languages, i.e. languages over a singleton alphabet (Guillon et al., 11 Apr 2025). This is a technically significant point because it shows that the succinctness gain is not an artifact of alphabet richness. Even in the unary setting, common-guess annotation can compress behavior that is expensive to reproduce in more standard models.

The same work states that this closes a question left open by Pighizzini and Prigioniero concerning the existence of a double-exponential gap between 1-LAs and 1DFAs in the unary case (Guillon et al., 11 Apr 2025). It also proves an exponential lower bound for complementing unary $\mu$7, and hence unary 1-LAs. Together with the complementation upper bounds discussed above, these unary results show that common-guess mechanisms are intertwined with both the power and the cost of simulation.

A common misconception is that such lower bounds would disappear on unary inputs because two-way motion and finite control are then less informative. The cited results directly contradict that expectation: the common guess remains descriptionally potent even over a singleton alphabet (Guillon et al., 11 Apr 2025).

6. Broader lineage: shared guesses, auxiliary memory, and restricted certificates

The idea of a shared guessed object predates the recent common-guess automata papers. In "On models of a nondeterministic computation" (0811.2586), nondeterminism is modeled by a deterministic multi-head two-way automaton with read-only access to an auxiliary memory. A memory model is a graph $\mu$8 with labeled outgoing edges, and a memory content is a mapping

$\mu$9

Acceptance is defined existentially over guesses: $\mu$0

This framework emphasizes precisely the feature that later common-guess automata exploit: the computation is deterministic once a single shared guessed structure is fixed. The paper develops several memory models, including one-way, two-way, and $\mu$1-way tapes, and identifies corresponding complexity classes such as

$\mu$2

as well as restricted-guess characterizations

$\mu$3

(0811.2586).

That paper does not use the term “2NFA with common guess” verbatim, but it provides a conceptual superstructure in which common-guess semantics can be situated. The recent limited-automata literature specializes this general idea to annotation-based verification over finite words, where the guessed object is placed directly on the input as a second track and checked by a finite-state device (Guillon et al., 15 Jul 2025). This suggests a useful taxonomy: ordinary nondeterminism branches locally; auxiliary-memory and common-guess models externalize nondeterminism into a shared certificate.

In current automata theory, 2NFAs with common guess therefore occupy a distinctive position. They are simultaneously a restricted form of 1-limited automaton, an annotation-based extension of two-way automata, and an instance of the broader principle that nondeterminism can be represented by a single globally shared guessed object rather than by unconstrained transition branching (Guillon et al., 11 Apr 2025).