Positive Queue: Theory & Applications

• Positive Queue is a queueing model with only positive customers, exemplified by the classical M/M/1 queue in its simplest form.
• Researchers use algebraic structures and random walk techniques to generalize the model, derive traffic equations, and establish stability conditions.
• The concept extends to signalized traffic systems and online moderation, illustrating its versatile application in dynamic service and control environments.

Searching arXiv for the primary source and closely related queueing papers on positive and generalized queues. Positive Queue denotes the standard queueing-theoretic setting in which arrivals are exclusively “positive customers,” namely customers that add one job to the buffer. In the formulation developed for zero-automatic queues, the Positive Queue is recovered as the simplest extremal case, namely the classical $M/M/1$ queue, while the broader framework also recovers Gelenbe’s $G$-queue with positive and negative customers and generalizes both through algebraically structured local interactions at the buffer back-end (0707.3449). In a distinct operational usage in signalized traffic networks, a “positive queue” means a link with queue length $q_i(t)>0$, i.e. a backlog of stationary vehicles, and this positivity condition determines whether outflow runs at capacity or becomes demand-limited (Hosseini et al., 2017). In online moderation systems, “Positive Queue” denotes a dedicated discovery and reward layer that surfaces desirable posts and comments for positive reinforcement rather than punishment-centric review (Lambert et al., 22 Sep 2025). Across these usages, the common motif is that queue positivity is not merely descriptive: it governs dynamics, stability, and control.

1. Queueing-theoretic meaning: positive customers and the standard queue

In queueing theory, “positive customers” are arrivals that add one job to the buffer, while “negative customers” are arrivals that instantaneously remove one job if present (0707.3449). A Positive Queue is thus the standard case with only positive customers, exemplified by the $M/M/1$ queue (0707.3449). This terminology becomes especially useful in comparison with Gelenbe’s $G$-queue, which allows both positive and negative customers, and with the zero-automatic queue framework, which embeds both as special cases inside a more general algebraic model of buffering (0707.3449).

Within the zero-automatic perspective, the Positive Queue corresponds to the free monoid on one generator, $X=\{a\}^*$, with $\Sigma=\{a\}$ and $\nu(a)=1$ (0707.3449). In that case there are no cancellations or mergings, $Next(a)=\{a\}$, and the buffer evolves exactly as in the classical single-class queue (0707.3449). The Twisted Traffic Equations reduce to

$\rho(\lambda+\mu)=\rho^2\mu+\lambda,$

whose solution is $G$0, and the stationary distribution is geometric,

$G$1

with Poisson departures of rate $G$2, recovering Burke’s Theorem (0707.3449).

This identification is significant because it places the standard positive-only queue inside a family of models whose local buffer interactions are encoded by a monoid or group law. A plausible implication is that familiar $G$3 phenomena—geometric stationarity, Poisson output, and a simple load parameter—can be viewed as the degenerate limit of richer interacting-buffer systems rather than as isolated properties.

2. Zero-automatic queues and the algebraic generalization of the Positive Queue

Zero-automatic queues are characterized by a special buffering mechanism evolving like a random walk on some infinite group or monoid $G$4 with finite generating set $G$5 (0707.3449). Each arriving customer carries a class $G$6 and interacts locally with the last class $G$7 already present in the buffer according to the algebraic law $G$8 (0707.3449). If $G$9, the pair cancels and both leave; if $q_i(t)>0$0, they merge into a single customer of class $q_i(t)>0$1; otherwise the new customer is appended (0707.3449).

The buffer content is encoded by a word $q_i(t)>0$2 in the regular language $q_i(t)>0$3 of normal forms, and the state space is $q_i(t)>0$4 (0707.3449). The queue is FIFO: front-end departure removes the first symbol $q_i(t)>0$5, while arrivals act at the back-end on $q_i(t)>0$6 (0707.3449). Arrivals form a rate-$q_i(t)>0$7 Poisson process, classes are i.i.d. with law $q_i(t)>0$8 on $q_i(t)>0$9, service completions occur at rate $M/M/1$0, and empty-buffer incorporation is governed by a boundary condition parameterized by $M/M/1$1 (0707.3449).

This construction generalizes the Positive Queue by replacing passive accumulation with structured local interactions. In the Positive Queue case, the algebra is trivial: no cancellation and no merging occur. In more general monoids or groups, queue evolution reflects both conventional service dynamics and algebraic simplification at the rear of the buffer (0707.3449). The framework therefore unifies classical queueing with interacting multi-class buffers in which workload is not simply the number of customers, but the canonical word representing buffer content.

3. Random walk structure, traffic equations, and stability

With the server blocked, the buffering mechanism is exactly the right random walk $M/M/1$2 on the Cayley graph, whose length process has drift $M/M/1$3 given by a Markovian harmonic measure computed from the Traffic Equations (0707.3449). For $M/M/1$4, the Traffic Equations are

$M/M/1$5

for all $M/M/1$6 (0707.3449). For plain triples, these equations have a unique admissible solution $M/M/1$7 (0707.3449).

It is convenient to define

$M/M/1$8

with $M/M/1$9 and $G$0 (0707.3449). Stability is then determined by the competition between arrival-induced drift and service: $G$1

$G$2

$G$3

(0707.3449).

For the Positive Queue, this criterion collapses to the classical condition $G$4, because the free-monoid case yields $G$5 and no cancellation term (0707.3449). For the $G$6-queue realized on the free group with generator $G$7 and inverse $G$8, the drift becomes

$G$9

and stability becomes

$X=\{a\}^*$0

the standard $X=\{a\}^*$1-queue criterion (0707.3449).

This unified drift condition is one of the main conceptual gains of the zero-automatic viewpoint. It replaces separate ad hoc stability calculations by a single random-walk comparison between the average growth induced by arrivals and the drain induced by service.

4. Product form, quasi-reversibility, and extremal cases

For stable zero-automatic queues, the Twisted Traffic Equations admit a unique admissible solution $X=\{a\}^*$2 with $X=\{a\}^*$3 (0707.3449). The equations are

$X=\{a\}^*$4

for $X=\{a\}^*$5 with $X=\{a\}^*$6 and $X=\{a\}^*$7 (0707.3449). The admissible solution satisfies

$X=\{a\}^*$8

If $X=\{a\}^*$9, then for any word $\Sigma=\{a\}$0, the stationary distribution is

$\Sigma=\{a\}$1

a multiplicative product over the letters of the buffer word (0707.3449). This is a direct generalization of the geometric stationary distribution of $\Sigma=\{a\}$2 (0707.3449).

The departure process in stationarity is Poisson with rate $\Sigma=\{a\}$3, and for each time $\Sigma=\{a\}$4, the queue-content $\Sigma=\{a\}$5 is independent of the departure process up to time $\Sigma=\{a\}$6 (0707.3449). This is a Burke-type theorem characterizing quasi-reversibility in the sense of Chao–Miyazawa–Pinedo (0707.3449).

The two simplest and extremal cases are especially informative.

Case	Algebraic model	Main consequence
Positive Queue	Free monoid $\Sigma=\{a\}$7	$\Sigma=\{a\}$8, geometric $\Sigma=\{a\}$9, Poisson output
$\nu(a)=1$0-queue	Free group $\nu(a)=1$1 with $\nu(a)=1$2	Positive and negative customers with cancellation

In the $\nu(a)=1$3-queue case, boundary conditions select which class persists in the buffer. Two product-form variants are

$\nu(a)=1$4

and

$\nu(a)=1$5

yielding geometric queue-length distributions $\nu(a)=1$6 for the surviving class $\nu(a)=1$7 (0707.3449). Other boundary choices may yield “almost product form” stationary distributions, but the Poisson output property needed for product-form networks is tied to the TTE solution (0707.3449).

The distinction between exact product form and “almost product form” is important. It shows that boundary conditions are not merely technical; they determine whether the queue enjoys full quasi-reversibility or only a weaker stationary structure.

5. Computation, assumptions, and scope of the zero-automatic framework

Computation proceeds through algebraic and random-walk methods. One first solves the Traffic Equations to obtain $\nu(a)=1$8 and the drift $\nu(a)=1$9, then solves the Twisted Traffic Equations to obtain the admissible pair $Next(a)=\{a\}$0 (0707.3449). A compact fixed-point approach substitutes

$Next(a)=\{a\}$1

into the TTE and finds $Next(a)=\{a\}$2 with $Next(a)=\{a\}$3, after which $Next(a)=\{a\}$4 follows from

$Next(a)=\{a\}$5

(0707.3449).

The invariant measure of the generator $Next(a)=\{a\}$6 is

$Next(a)=\{a\}$7

and normalization gives $Next(a)=\{a\}$8 (0707.3449). A QBD approximation is also available by aggregating buffer content to $Next(a)=\{a\}$9, producing a block-tridiagonal generator $\rho(\lambda+\mu)=\rho^2\mu+\lambda,$0; in the level-geometric case, matrix-geometric theory recovers $\rho(\lambda+\mu)=\rho^2\mu+\lambda,$1, consistent with the exact product form (0707.3449).

The strongest results hold for plain triples: infinite plain monoids or groups with natural generators. In that setting the random walk is transient, the successor graph is strongly connected, the Traffic Equations have a unique solution $\rho(\lambda+\mu)=\rho^2\mu+\lambda,$2, and the stable-region TTE have a unique admissible solution $\rho(\lambda+\mu)=\rho^2\mu+\lambda,$3 (0707.3449). For more general non-plain 0-automatic pairs, multiple TTE solutions may exist, the successor graph may be disconnected, and the random walk may be non-transient, as in the free group $\rho(\lambda+\mu)=\rho^2\mu+\lambda,$4 with $\rho(\lambda+\mu)=\rho^2\mu+\lambda,$5 (0707.3449).

The paper also emphasizes several scope conditions. The Poisson output theorem counts front-end service completions, not back-end cancellations; the saturation principle does not hold, since the departure rate satisfies

$\rho(\lambda+\mu)=\rho^2\mu+\lambda,$6

and product form requires choosing $\rho(\lambda+\mu)=\rho^2\mu+\lambda,$7 as the TTE solution, whereas other boundary conditions may yield stationary regimes without Poisson output (0707.3449). Extensions to $\rho(\lambda+\mu)=\rho^2\mu+\lambda,$8 are possible, with stability still comparing $\rho(\lambda+\mu)=\rho^2\mu+\lambda,$9 and $G$00 through appropriate ergodic averages (0707.3449).

6. Positive queue as a state condition in signalized traffic networks

In signalized arterial networks, “positive queue” has a different but precise meaning: for link $G$01, $G$02 means there is backlog at the link and vehicles are waiting to be served by the signal (Hosseini et al., 2017). This condition directly determines the outflow rule. If $G$03, then

$G$04

if $G$05, then

$G$06

where $G$07 is the instantaneous total inflow rate reaching link $G$08 at time $G$09 (Hosseini et al., 2017).

In cumulative form, with

$G$10

the queue length is

$G$11

and when $G$12,

$G$13

(Hosseini et al., 2017). Thus queue positivity is the hinge between saturated discharge and demand-limited discharge.

This distinction drives both transient and steady-state analysis. The model is a delay-differential system,

$G$14

with threshold crossings at times when $G$15 reaches zero (Hosseini et al., 2017). To handle zero travel times, the paper formulates an LP that computes the unique outflow vector $G$16 while enforcing capacity constraints, non-negativity of queues, and instantaneous routing couplings (Hosseini et al., 2017). Under piecewise-constant external inflow and capacity functions, with routing matrix $G$17 sub-stochastic and spectral radius less than $G$18, the DDE with LP-selected outflows admits a unique, non-negative solution $G$19 for all $G$20 (Hosseini et al., 2017).

Under fixed-time control, periodic external inflow and periodic capacity profiles satisfying

$G$21

for some $G$22, there exists a unique $G$23-periodic trajectory $G$24 that globally attracts all trajectories (Hosseini et al., 2017). The iterative computation of this periodic orbit is organized around transition points $G$25, the times when the queue switches from zero to positive in steady state (Hosseini et al., 2017). Here, positive queue is operational rather than algebraic: it signals saturation of service at a link and serves as the switching condition for both analysis and numerical computation.

7. Positive Queue as a moderation interface and positive reinforcement mechanism

In online moderation, Positive Queue refers to a Chrome extension that augments Reddit’s moderator interface with a dedicated “positive” discovery and reward layer (Lambert et al., 22 Sep 2025). Instead of directing attention primarily to harmful or reported content, it surfaces desirable posts and comments and streamlines ways to positively reinforce them (Lambert et al., 22 Sep 2025). The motivation is that punishment-centric tools do not teach users what to do instead, expose moderators to harmful content, and do little to proactively create healthy communities, whereas prior research links positive feedback to higher quality and frequency of contributions, reduced norm violations, and sustained motivation (Lambert et al., 22 Sep 2025).

The system targets Reddit’s Unmoderated queue and injects UI components into the existing modqueue workflow (Lambert et al., 22 Sep 2025). It provides color-coded desirability labels, hover panels with modeled desirability scores and histograms, always-visible filters for post-, author-, and comment-section signals, and enhanced sorting by desirability, score, author age or karma, and comment-section aggregates (Lambert et al., 22 Sep 2025). Reward mechanisms include a “Curate” button for compiling “Best of the week” threads, an “Explain” button for posting moderator replies that articulate why an item is desirable, and improved access to flair, highlight, and award actions (Lambert et al., 22 Sep 2025).

Desirable content is labeled using subreddit-specific XGBoost classifiers with $G$26, trained separately for posts and comments using an $G$27 train-test split and features including LIWC2015, VADER sentiment, Flesch readability, ConvoKit politeness, Detoxify toxicity, and SentenceBERT embeddings (Lambert et al., 22 Sep 2025). Items with Reddit scores in quartile $G$28 are labeled desirable, items in $G$29–$G$30 undesirable, and $G$31 is discarded (Lambert et al., 22 Sep 2025). Averaged across subreddits, performance was reported as $G$32 Accuracy and $G$33 for posts, and $G$34 Accuracy and $G$35 for comments (Lambert et al., 22 Sep 2025).

A controlled user study with five active moderators found that curation was universally favored, sorting and filtering were useful for discovering high-quality content efficiently, and reactions to public explanation comments were mixed (Lambert et al., 22 Sep 2025). The study also found applicability across communities of very different sizes and styles, and moderators sometimes repurposed discovery cues for punitive triage, suggesting dual-use potential (Lambert et al., 22 Sep 2025).

This usage departs entirely from queueing theory, yet retains the same structural idea: a queue is made “positive” by orienting attention and action toward desirable states rather than only toward violations or depletion. A plausible implication is that the phrase has become a design pattern extending beyond stochastic service systems into sociotechnical workflows.