Threshold-2 Rule Overview

Updated 6 July 2026

Threshold-2 Rule is a family of mechanisms marking a boundary where qualitative behavior changes, from cellular automata stability to decision thresholds in diagnostics.
In two-dimensional cellular automata, it generates nontrivial stable configurations, allowing testing algorithms with query complexity independent of configuration size.
Across graph activation and molecular computation, it governs transitions—activating a node with two neighbors or implementing an AND gate—highlighting diverse dynamics and applications.

Searching arXiv for papers on "Threshold-2 Rule" and closely related uses across fields. {"query":"\"Threshold-2\" rule arXiv cellular automata threshold semiorder bootstrap percolation", "max_results": 10} {"query":"(Nakar et al., 19 Jul 2025)", "max_results": 5} {"query":"(Hamburger et al., 2017)", "max_results": 5} The expression Threshold-2 Rule is not uniform across the contemporary arXiv literature. In one recent line of work it denotes a nontrivial update rule for two-dimensional threshold cellular automata on a torus with the von-Neumann neighborhood, where stable configurations are structurally characterizable and testable with query complexity independent of the configuration size and quadratic in $1/\epsilon$ (Nakar et al., 19 Jul 2025). In other literatures, the same phrase or a closely related one refers to the case $r=2$ in threshold diffusion and bootstrap percolation, to the upper threshold $t_2$ in double-threshold preference models, to the upper AI cut-point $T_2$ in rule-in diagnostics, or to a 2-input threshold gate in molecular computation. The common idea is a transition triggered by crossing a specified level, but the underlying state spaces, dynamics, and semantics differ substantially.

1. Terminological scope

A useful way to read the term is as a family resemblance rather than a single formal object. In all of the settings below, “threshold-2” marks a boundary at which qualitative behavior changes: activation by two neighbors, preference above an upper uncertainty boundary, automatic rule-in above a second diagnostic cut-point, or logical output $1$ when both binary inputs are present.

Domain	Meaning of “Threshold-2”	Representative source
Two-dimensional cellular automata	A threshold rule on the von-Neumann neighborhood with nontrivial stable configurations	(Nakar et al., 19 Jul 2025)
Threshold diffusion / bootstrap percolation	A node becomes blue once it has at least two blue neighbors	(Zehmakan, 2019)
Double-threshold semiorders	The upper threshold $t_2$ above which preference is mandatory	(Hamburger et al., 2017)
Two-threshold diagnostics	The upper cut-point $T_2 \equiv T_A^{ri}$ for automatic rule-in	(Mastrianni et al., 25 Apr 2026)
Step chemical reaction networks	A 2-input threshold-2 gate implementing logical AND	(Anderson et al., 2024)

A recurring misconception is to treat these as notational variants of one model. The sources instead use the phrase for distinct constructs: a local update rule in dynamical systems, a pairwise-comparison rule in preference theory, a decision threshold in statistical diagnosis, and a gate specification in unconventional computation. This suggests that any technical use of the term must be interpreted with its ambient model class.

2. Two-dimensional threshold cellular automata

In the cellular-automaton setting, configurations evolve on a two-dimensional torus according to threshold rules with respect to the von-Neumann neighborhood. The paper distinguishes trivial and nontrivial stability regimes: stable configurations for Threshold-1 (OR) and Threshold-5 (AND) are trivial, whereas Threshold-2, Threshold-4, and Threshold-3 (Majority) exhibit more diverse behaviors (Nakar et al., 19 Jul 2025).

The principal results reported for Threshold-2 are structural and algorithmic. First, the work characterizes the structure of stable configurations with respect to the Threshold-2 rule. Second, it designs and analyzes a testing algorithm that distinguishes between configurations that are stable with respect to the Threshold-2 rule and those that are $\epsilon$ -far from any stable configuration. The stated query complexity is independent of the size of the configuration and depends quadratically on $1/\epsilon$ (Nakar et al., 19 Jul 2025).

Within the threshold-cellular-automata taxonomy, Threshold-2 therefore occupies the first nontrivial regime beyond the degenerate OR case. A plausible implication is that the combinatorics of local support patterns, rather than merely all-zero or all-one fixed points, controls stability. The abstract also places Threshold-2 near Threshold-4 and opposite Threshold-3 (Majority), indicating a broader classification program for nontrivial fixed-point structure on the torus.

3. The $r=2$ threshold model on graphs

In graph dynamics, the Threshold-2 rule appears as the special case $r=2$ 0 of a monotone threshold process or bootstrap percolation. For an $r=2$ 1-vertex graph $r=2$ 2 and initial blue set $r=2$ 3, the evolution is

$r=2$ 4

Blue vertices never turn red, the process stabilizes in at most $r=2$ 5 steps, and a target set (or dynamic monopoly) is a set $r=2$ 6 such that starting from $r=2$ 7 one eventually reaches $r=2$ 8 (Zehmakan, 2019).

The minimum size of such a set is

$r=2$ 9

For a $t_2$ 0-regular graph with second-largest normalized eigenvalue $t_2$ 1, the analysis relies on the Expander–Mixing Lemma and yields a target-set bound of order $t_2$ 2. Specializing the general result of Zehmakan to $t_2$ 3, one sets

$t_2$ 4

and obtains

$t_2$ 5

The proof sketch proceeds by first finding a stable set $t_2$ 6 with $t_2$ 7, where every vertex in $t_2$ 8 has at least two neighbors in $t_2$ 9, and then using mixing to force one-by-one growth until all vertices become blue (Zehmakan, 2019).

For random graph models, the same threshold-2 activation rule yields asymptotic regimes tied to expansion. In the uniform random $T_2$ 0-regular graph $T_2$ 1, Friedman’s theorem gives

$T_2$ 2

hence

$T_2$ 3

In Erdős–Rényi graphs $T_2$ 4 with

$T_2$ 5

the same method yields

$T_2$ 6

and below the connectivity threshold $T_2$ 7, no small seed set can infect the whole graph (Zehmakan, 2019).

Here Threshold-2 has an exact combinatorial meaning: a vertex activates as soon as it acquires two active neighbors. This is distinct from Majority dynamics and from non-monotone cellular automata, even though all are threshold systems.

4. Double-threshold semiorders and digraphs

In preference theory, the relevant object is not a two-neighbor activation rule but a double-threshold relation with lower threshold $T_2$ 8 and upper threshold $T_2$ 9. Each element $1$0 is assigned a utility value $1$1, with $1$2, and preference is defined by three regions:

if $1$3, then $1$4 is not preferred to $1$5;
if $1$6, then $1$7 may or may not be preferred to $1$8;
if $1$9, then $t_2$ 0 is preferred to $t_2$ 1.

The corresponding directed graph is a double threshold digraph. A satisfying assignment $t_2$ 2 for thresholds $t_2$ 3 must satisfy the system of linear constraints

$t_2$ 4

$t_2$ 5

The upper boundary $t_2$ 6 is therefore the “threshold-2” of the model: above it, preference is mandatory; below $t_2$ 7, preference is forbidden; between them lies an uncertainty band (Hamburger et al., 2017).

This formulation generalizes semiorders and allows controlled nontransitivity while remaining acyclic. Every directed acyclic graph is a double threshold graph for sufficiently large ratio

$t_2$ 8

The key invariant is

$t_2$ 9

The paper characterizes $T_2 \equiv T_A^{ri}$ 0 via forcing cycles, with

$T_2 \equiv T_A^{ri}$ 1

and also shows that $T_2 \equiv T_A^{ri}$ 2 can be written as $T_2 \equiv T_A^{ri}$ 3 with integers $T_2 \equiv T_A^{ri}$ 4 (Hamburger et al., 2017).

Algorithmically, feasibility for fixed $T_2 \equiv T_A^{ri}$ 5 is reduced to a difference-constraints graph $T_2 \equiv T_A^{ri}$ 6: for each directed edge $T_2 \equiv T_A^{ri}$ 7 of $T_2 \equiv T_A^{ri}$ 8, add an arc $T_2 \equiv T_A^{ri}$ 9 of weight $\epsilon$ 0; for each unordered non-edge $\epsilon$ 1, add arcs $\epsilon$ 2 and $\epsilon$ 3 each of weight $\epsilon$ 4. The digraph $\epsilon$ 5 admits a satisfying assignment iff $\epsilon$ 6 has no negative-weight cycle. If none exists, shortest-path distances yield $\epsilon$ 7; if one exists, extracting it yields a forbidden forcing cycle certifying infeasibility. The stated high-level time bound is $\epsilon$ 8, where $\epsilon$ 9 (Hamburger et al., 2017).

In this literature, Threshold-2 is thus an upper certainty boundary, not an activation count. The shared threshold language masks a substantially different semantics.

5. Two-threshold diagnostic rules

In diagnostic ROC analysis, threshold-2 is the upper cut-point in a two-threshold rule combining two correlated tests. Test A produces score $1/\epsilon$ 0, Test B produces score $1/\epsilon$ 1, and two cut-points are introduced on the Test A scale: $1/\epsilon$ 2 The classification rule is:

if $1/\epsilon$ 3, call the case negative (automatic rule-out);
else if $1/\epsilon$ 4, call the case positive (automatic rule-in);
else, defer to Test B and call positive if $1/\epsilon$ 5, negative otherwise (Mastrianni et al., 25 Apr 2026).

The joint model is expressed by separate copulas in the non-diseased and diseased populations: $1/\epsilon$ 6 The paper develops exact ROC formulas for rule-out, rule-in, and the combined two-threshold policy. For example, the overall false-positive and true-positive rates under the combined rule satisfy

$1/\epsilon$ 7

$1/\epsilon$ 8

The theoretical results state that rule-out pAUC increases with increased diseased correlation and decreased non-diseased correlation, whereas rule-in pAUC has the opposite dependence (Mastrianni et al., 25 Apr 2026).

The upper threshold $1/\epsilon$ 9 is therefore the diagnostic analogue of Threshold-2. It marks the score above which Test A alone triggers an automatic positive call. The paper suggests choosing $r=2$ 0 either by meeting a target on rule-in performance after fixing $r=2$ 1, or by directly maximizing the combined partial AUC over the grid $r=2$ 2, subject to clinical constraints such as minimum NPV or maximum FPF. Empirically, copula-based rule-out and combined rule-out/rule-in analyses on EMBED, CSAW-CC, and a Radboud University reader study were reported to bracket the observed performance, whereas models assuming independence mis-predicted the rule-out operating point (Mastrianni et al., 25 Apr 2026).

A different usage appears in Step Chemical Reaction Networks, where the basic object is a 2-input threshold-2 gate. The gate has Boolean inputs $r=2$ 3 and output

$r=2$ 4

This is exactly the 2-input AND function. Bits are encoded by species presence, one injects one copy of the true-output species $r=2$ 5 and one false-output proxy for each input, and the gate is implemented using four size- $r=2$ 6 bimolecular void rules: $r=2$ 7 After one gate-step, $r=2$ 8 survives iff both inputs are true; otherwise at least one false proxy survives. The construction uses exactly one gate-step and exactly 4 bimolecular void rules, with $r=2$ 9 species and $r=2$ 00 volume per gate. For threshold formulas this gives linear resources, whereas simulating threshold circuits in the restricted gate-wise model requires exponential volume, with a matching exponential lower bound (Anderson et al., 2024).

This molecular interpretation again differs from graph activation and from the upper-threshold semantics of diagnostics or semiorders. Here Threshold-2 specifies the fan-in condition for a Boolean gate.

A related but distinct thresholding convention occurs in nonadaptive disjunctive group testing. There, the decision rule compares the number of positive tests $r=2$ 01 against a fixed threshold $r=2$ 02: $r=2$ 03 The decoding complexity is $r=2$ 04 for the threshold rule, compared with $r=2$ 05 bit-operations for classical disjunctive $r=2$ 06-code decoding, and in many practical parameter regimes the optimal threshold turns out to be very small, often $r=2$ 07 or $r=2$ 08 (D'yachkov et al., 2016). This is not usually called a Threshold-2 rule in the same sense as the preceding examples, but it illustrates another standard use of threshold language: a scalar acceptance boundary on an aggregate statistic.

Across these fields, the phrase Threshold-2 Rule therefore denotes one of several mathematically distinct threshold mechanisms. In cellular automata it is a nontrivial local update rule with stable-configuration testing; in graph contagion it is the $r=2$ 09 activation condition; in preference theory and diagnostics it is an upper threshold in a two-threshold system; and in molecular computation it is the 2-input threshold gate implementing AND. The similarity lies in the thresholding paradigm, not in a shared formal definition.