Undecided-State Dynamics in Consensus Models

• Undecided-State Dynamics (USD) is a discrete-state framework where agents hold decided opinions or remain undecided, resolving conflicts via simple local rules.
• USD models are applied to achieve approximate majority, plurality consensus, and symmetry breaking in both binary and multi-opinion settings.
• Analytical techniques such as drift analysis and phase decomposition reveal rapid convergence, noise-induced phase transitions, and the impact of stubborn agents.

to=arxiv_search 天天爱彩票提现json {"query":"\"Undecided-State Dynamics\" arXiv", "max_results": 10} Undecided-State Dynamics (USD) denotes a family of discrete-state interaction processes in which agents hold one of several decided opinions together with a distinguished undecided state, and update through local interactions on a communication graph. In the classical binary setting, every node supports one of two colors or is undecided; disagreement creates undecidedness, while undecided nodes adopt encountered opinions. In multi-opinion formulations, the state space is typically written as $\Sigma=[k]\cup\{\bot\}$, with $\bot$ denoting the undecided state. Across distributed consensus, population protocols, and sociophysical opinion models, USD is studied as a mechanism for approximate majority, plurality consensus, symmetry breaking, metastable consensus, and the dynamical mediation of transitions between opposed states (Clementi et al., 2017, Cooper et al., 3 Mar 2026).

1. Formal state space and update mechanisms

In the binary parallel pull model on the complete graph, each node can support color Alpha, color Beta, or be undecided. The synchronous update rule is local and minimal: if a node is undecided, it adopts the pulled state if that neighbor is colored; if a node is colored and it pulls the same color or an undecided node, it stays as it is; if a node is colored and it pulls the other color, it becomes undecided. Writing $a(x)$, $b(x)$, and $q(x)$ for the numbers of Alpha, Beta, and undecided nodes in configuration $x$, the bias is

$s(x)=a(x)-b(x),$

with

$q(x)=n-a(x)-b(x).$

Because the graph is complete, the process induces a finite Markov chain over configurations; its absorbing states are the monochromatic configurations $a=n$ and $b=n$, and also the all-undecided configuration $\bot$0 (Clementi et al., 2017).

In the asynchronous population-protocol formulation, time is discrete, an ordered pair is selected uniformly at random, and only one endpoint updates. In the binary stubborn-agent variant, the state space is

$\bot$1

where $\bot$2 and $\bot$3 are the two opinions and $\bot$4 is undecided. In the classical USD, if an agent with one opinion meets an agent with the opposite opinion, the initiator becomes undecided; if an undecided agent meets another agent, it adopts that agent’s opinion. The stubborn modification changes only interactions of type $\bot$5: an agent with Opinion $\bot$6 becomes undecided only with probability $\bot$7, and with probability $\bot$8 it stays with Opinion $\bot$9 (Berenbrink et al., 2024).

For many-opinion systems, the general state space is

$a(x)$0

and the update rule can be written as

$a(x)$1

Equivalently, if the initiator and the sampled neighbor hold two different decided opinions, the initiator becomes undecided; if the initiator is undecided and the sampled neighbor is decided, the initiator adopts that opinion; otherwise the initiator keeps its state. This formulation is used in both the synchronous gossip model and the asynchronous population protocol model (Cooper et al., 3 Mar 2026).

A synchronized variant separates updates into repeated phases with a decision part and a boosting part. In the decision part, agents that encounter a different opinion become undecided; in the boosting part, undecided agents adopt the first decided opinion they encounter. This phase separation is central to later unconditional plurality-consensus analyses (Bankhamer et al., 2021).

2. Consensus, majority, and plurality guarantees

For the binary parallel process on the complete graph, a sharp unconditional convergence theory is available. Starting from any initial configuration, the process reaches a monochromatic configuration within

$a(x)$2

rounds, with high probability, and this bound is tight: there are initial configurations requiring

$a(x)$3

rounds to converge with high probability. A stronger theorem states that if the initial configuration has bias

$a(x)$4

then the process reaches the monochromatic configuration $a(x)$5 within $a(x)$6 rounds with high probability, assuming Alpha is the initial majority color (Clementi et al., 2017).

For synchronized USD with arbitrarily many opinions, the phase-clock construction yields polylogarithmic-time consensus without any initial bias assumption. In the population model, the protocol reaches consensus on a significant opinion in

$a(x)$7

parallel time with high probability. In the gossip model, it reaches consensus on a significant opinion in

$a(x)$8

rounds with high probability. In both models, if there is an initial additive bias of

$a(x)$9

then the initial plurality opinion wins with high probability (Bankhamer et al., 2021).

For the unsynchronized $b(x)$0-opinion population protocol, convergence beyond the binary case was established under mild assumptions on $b(x)$1 and the initial number of undecided agents. If

$b(x)$2

and

$b(x)$3

then the USD achieves plurality consensus within

$b(x)$4

interactions with high probability, regardless of the initial bias. If there is an initial additive bias of at least

$b(x)$5

the initial plurality opinion wins with high probability; if there is a multiplicative bias, the convergence time improves to

$b(x)$6

interactions (Amir et al., 2023).

The strongest currently summarized arbitrary-$b(x)$7 guarantees cover both synchronous and asynchronous models for arbitrary initial configurations. In the gossip model, USD reaches consensus within

$b(x)$8

synchronous rounds with probability $b(x)$9, where $q(x)$0 is the gossip-specific probability of collapsing to the all-undecided state in the first round. In the population protocol model, USD reaches consensus within

$q(x)$1

asynchronous interactions with high probability. Matching lower bounds up to polylogarithmic factors are also given for specific initial configurations, indicating that these upper bounds are essentially optimal (Cooper et al., 3 Mar 2026).

An almost tight lower bound complements the upper-bound theory in the population protocol model. For

$q(x)$2

there exists an initial configuration, even with bias $q(x)$3, where stabilization requires

$q(x)$4

interactions, or equivalently

$q(x)$5

parallel time (El-Hayek et al., 5 May 2025).

3. Bias amplification, phase transitions, noise, and stubbornness

A defining analytical feature of USD is explicit bias amplification through the undecided population. In the binary complete-graph model,

$q(x)$6

so as long as $q(x)$7 is a non-negligible fraction of $q(x)$8, the expected bias grows multiplicatively. The same one-step analysis gives

$q(x)$9

This is the basis for the paper’s symmetry-breaking picture: balanced states are not stable, and random fluctuations plus drift push the system away from symmetry (Clementi et al., 2017).

When uniform message noise is added, USD undergoes a sharp robustness transition. In the noisy binary model, each transmitted message is correct with probability $x$0 and replaced by one of the other two states with probability $x$1. The key expectation identities become

$x$2

and

$x$3

The phase transition occurs at

$x$4

If $x$5, then from any configuration with sufficiently large bias

$x$6

the process reaches in $x$7 rounds a metastable regime of almost consensus and preserves a linear bias for polynomial time. If $x$8, then even starting from complete consensus, the process loses information about the original majority within $x$9 rounds, with the bias collapsing to

$s(x)=a(x)-b(x),$0

The same paper gives an exact coupling showing the equivalence between the noisy model and a noiseless model with stubborn agents (d'Amore et al., 2020).

A direct stubborn-agent variant makes the preferred opinion explicit rather than induced by communication noise. The central quantity is the weighted bias

$s(x)=a(x)-b(x),$1

and the threshold stubbornness is

$s(x)=a(x)-b(x),$2

If

$s(x)=a(x)-b(x),$3

then all agents agree on Opinion $s(x)=a(x)-b(x),$4 after $s(x)=a(x)-b(x),$5 interactions with high probability; if

$s(x)=a(x)-b(x),$6

then all agents agree on Opinion $s(x)=a(x)-b(x),$7, again after $s(x)=a(x)-b(x),$8 interactions with high probability. In the critical window around $s(x)=a(x)-b(x),$9, consensus still occurs, but only after

$q(x)=n-a(x)-b(x).$0

interactions, and either of the two opinions can survive (Berenbrink et al., 2024).

These results suggest a common structural theme: the undecided state does not merely slow opinion change. It modulates gain, loss, and drift terms in a way that can create fast majority amplification, metastability, or sharp threshold behavior, depending on how undecidedness is generated and resolved.

4. Undecided agents in sociophysics and opinion-formation models

Outside distributed consensus, USD appears as a concrete three-state mechanism in opinion dynamics and econophysics. In a tax-evasion model on a fully connected population, each agent has state

$q(x)=n-a(x)-b(x).$1

where $q(x)=n-a(x)-b(x).$2 denotes honest taxpayer, $q(x)=n-a(x)-b(x).$3 tax evader, and $q(x)=n-a(x)-b(x).$4 undecided. The kinetic-exchange update is

$q(x)=n-a(x)-b(x).$5

with $q(x)=n-a(x)-b(x).$6, and the quenched couplings are distributed as

$q(x)=n-a(x)-b(x).$7

Without enforcement, the model exhibits a nonequilibrium order-disorder transition at

$q(x)=n-a(x)-b(x).$8

measured by the order parameter

$q(x)=n-a(x)-b(x).$9

The paper then adds Zaklan enforcement through an audit probability $a=n$0 and penalty duration $a=n$1, so that a detected evader is forced to remain honest for $a=n$2 time steps. Below $a=n$3, compliance is high and punishment has a small effect; for $a=n$4, enforcement can reduce evasion considerably. The undecided class survives in the stationary state, is generally more populated than the evader class, and tends to reduce the number of evaders (Crokidakis, 2014).

A different three-state realization is based on hidden conviction rather than direct state copying. Each agent carries an opinion

$a=n$5

and a conviction

$a=n$6

with thresholds

$a=n$7

The two key control parameters are the initial undecided fraction $a=n$8 and the conviction sensitivity $a=n$9. Depending on $b=n$0, the model exhibits all-undecided convergence, consensus of one opinion, or bipolarization. The paper emphasizes that a minimum fraction of undecided agents is crucial not only for reaching consensus of a given opinion, but also to determine a dominant opinion in a polarised situation (Balenzuela et al., 2015).

A numerical study of undecided agents in a complete opinion-exchange network treats undecided or volatile agents as agents that randomize independently of the interacting partner:

$b=n$1

The order parameter is

$b=n$2

The reported consensus curve has two empirical regimes. For undecided fraction $b=n$3,

$b=n$4

whereas for $b=n$5,

$b=n$6

with $b=n$7. The paper identifies $b=n$8 as a boundary between the two behaviors and states that about $b=n$9 undecided fraction is enough to reduce the initial consensus by half (Blanco et al., 2021).

Network topology can also determine whether undecided states disappear or persist. In a Watts–Strogatz setting with multiple opinions and an explicit undecided state $\bot$00, the Travieso-Fontoura outflow rule drives $\bot$01 and consensus in finite systems, whereas the plurality-rule inflow dynamics can retain undecided nodes in the final absorbing state. In the plurality rule, undecided nodes arise at interfaces through ties in the local decision group, and for sufficiently large $\bot$02 and small $\bot$03, the undecided fraction has a maximum in the small-world region (Dornelas et al., 2018).

5. Analytical techniques and phase-based proof structures

USD analyses are unusually explicit about one-step drift, phase decomposition, and potential functions. In the synchronized plurality setting, the decision part produces

$\bot$04

with

$\bot$05

while the boosting part is modeled by a Pólya-Eggenberger distribution satisfying

$\bot$06

One of the key structural lemmas is

$\bot$07

meaning that insignificant opinions never become significant again (Bankhamer et al., 2021).

For the unsynchronized $\bot$08-opinion population protocol, the analysis is organized into five phases. A central potential is

$\bot$09

later replaced by

$\bot$10

The proof uses multiplicative drift to show the rise of the undecided population, anti-concentration to generate an additive bias when none is present, gambler’s-ruin arguments to amplify pairwise gaps, and a final coupling to the two-opinion process once one opinion reaches an absolute majority. The threshold

$\bot$11

acts as the near-equilibrium level for undecided agents in several drift bounds (Amir et al., 2023).

For arbitrary $\bot$12, a new normalized potential is introduced:

$\bot$13

In the gossip model,

$\bot$14

and under the regime $\bot$15 and $\bot$16,

$\bot$17

The analogous population-protocol drift is

$\bot$18

This normalization by $\bot$19 is used because direct analysis of $\bot$20 is obstructed by the undecided population (Cooper et al., 3 Mar 2026).

Binary USD also supports fine-grained phase decompositions. One analysis partitions the configuration space into regions $\bot$21 in the $\bot$22-plane and proves that trajectories eventually reach a sink phase where the majority wins. The symmetry-breaking part relies on a generic Markov-chain hitting-time lemma, while the lower-scale fluctuation argument uses the Berry–Esseen theorem to show that when the bias is tiny, the next-step bias has variance $\bot$23, so with constant probability it attains magnitude $\bot$24 (Clementi et al., 2017).

These proof styles indicate that USD is not analyzed as a generic black-box Markov chain. Its tractability comes from explicit state-count statistics, non-linear drift identities, and carefully separated temporal regimes in which undecided mass, bias, and opinion concentration can each be controlled.

6. Scope, adjacent models, and acronym ambiguity

Not every indecision model is a standard USD, but several adjacent frameworks are structurally related. In the Buridan’s-ass switching model, indecision is represented by a hidden discrete-state Markov process coupled to continuous deterministic motion. In one dimension, the state switches with probabilities

$\bot$25

and the stationary state distribution is

$\bot$26

The paper emphasizes analytic tractability in one dimension, intractability of stationary PDE methods in higher dimensions, and a geometric state-detection approach for the multi-dimensional case (Bates et al., 2012).

A different neighboring line is a complexity-theoretic theory of uncertainty in discrete-time Boolean finite dynamical systems. That framework introduces multiple-choice update functions

$\bot$27

together with alternate update schedules such as parallel, fixed-permutation, permutation-list, arbitrary-permutation, and asynchronous updating. The paper explicitly states that it does not define an “undecided state” in the sense used in classic USD literature, but it is conceptually adjacent because it formalizes nondeterministic local choice, update uncertainty, and robust reachability (Ogihara et al., 2022).

A recurring source of confusion is the acronym itself. In “Neural USD,” USD means Neural Universal Scene Descriptor, not Undecided-State Dynamics. That paper explicitly notes that the acronym is overloaded and that its “Neural USD” is an object-centric neural conditioning framework inspired by the Universal Scene Descriptor standard from computer graphics, not a consensus process (Escontrela et al., 28 Oct 2025).

Taken together, these strands delimit USD rather than dilute it. In the standard sense used in distributed consensus and most three-state opinion dynamics, USD refers to dynamics in which disagreement creates undecided agents and undecided agents later resolve by adopting encountered opinions. The resulting interaction between cancellation and recruitment is the common mechanism behind logarithmic-time binary consensus, many-opinion plurality dynamics, metastable consensus under noise, and the broader sociophysical role of intermediate undecided states.