Bartlett's theorem

A Poisson-arrival occupancy theorem for independently moving individuals in a system's regions.

Version

v1 · 2026-09-28 · History

Domain-specific #

8119

Domain group

Formal Sciences

Origin domain

Mathematics

Subdomain

Applied Probability → Mathematics

Core Idea

Bartlett's theorem is an occupancy law for Poisson arrivals whose members evolve independently. Fix a region and observation time. For each possible arrival age, multiply the inflow rate by the chance that an entrant remains in that region, then integrate. The resulting number of individuals is Poisson, and counts in disjoint regions are independent. Kelly and Yudovina give the empty-initial-system formulation and proof; it is stronger than calculating only an expected headcount.

The theorem's one-region infinite-server special case can inform service planning. Garrison and Pecina applied a Poisson-arrival, independent-stay model to a hospital family-medicine service and used λE(B) to predict census. Its 7.44 predicted mean was below the observed 7 a.m. 8.28, reminding readers that a model's conditional theorem does not validate its premises in every real system. Finite beds, shared constraints, time-of-day sampling, or correlated arrivals can matter.

How would you explain it like I'm…

The Random Park Visitors

People wander into a big park at random times, and each one walks around on their own, not following anybody. You can't know exactly how many will be at the playground right now, but Bartlett's theorem tells you exactly how likely each number is, using only how often people arrive and how likely each one is to be at the playground. And the number at the playground has nothing to do with the number at the pond.

Random Arrivals, Random Counts

Suppose visitors arrive at random, like raindrops starting to fall, and once inside each one moves around independently. Bartlett's theorem says the number of visitors you'd find in some area at a given time follows the same kind of randomness as the arrivals, a pattern called a Poisson distribution. To get the average, you add up, over all past moments, how often people arrived then times the chance someone who arrived then is in that area now. Counts in separate areas don't affect each other. But the answer is only as good as its assumptions: if visitors come in groups or run out of room, real counts can differ.

Poisson Occupancy Theorem

Bartlett's theorem is about systems where individuals arrive according to a Poisson process, a common model of purely random arrivals, and then each moves or evolves independently of the others. Starting from an empty system, fix a region and a time. The number of individuals in that region is Poisson distributed, and its mean equals the arrival rate multiplied by the probability that an arrival of each age is in that region, added up (integrated) over all possible arrival ages. Counts in regions that don't overlap are independent. This is stronger than just computing an average headcount, because it gives the whole probability distribution. A special case is a queue with unlimited servers, where the number of customers present has mean λE(B), the arrival rate times the average stay; in one hospital study this predicted a mean census of 7.44 against an observed 8.28, a reminder that the premises must fit the real system.

 

Bartlett's theorem is an occupancy law for Poisson arrival streams whose members evolve independently after entering. Starting from an empty system, fix an observation time t and a region A. The number of individuals in A at time t is Poisson distributed with mean ∫ λ(s)·P(an entrant arriving at time s is in A at time t) ds, integrating over arrival times (equivalently, over arrival ages), and the counts in disjoint regions are mutually independent. Kelly and Yudovina give this empty-initial-system formulation and proof, which yields the full distribution and independence structure rather than only an expected headcount. Taking the region to be 'still in service' gives the infinite-server queue: the number present is Poisson with mean λE(B), where B is the service or stay duration. Garrison and Pecina used such a model for a hospital family-medicine service, predicting a mean census of 7.44 against an observed 7 a.m. value of 8.28. That gap illustrates that the theorem is conditional: finite beds, shared constraints, time-of-day sampling, or correlated arrivals violate its premises.

Structural Signature

Sig role-phrases:

• Poisson arrival process — Independent arrivals supply the random inflow at a specified intensity. It is constitutive. Counterfactual: Arbitrary clustered arrivals need not preserve Poisson occupancy.
• Independent individual paths — Each entrant moves or remains without changing another entrant's trajectory. It is constitutive. Counterfactual: Shared capacity constraints can couple paths and break the result.
• Observation region and time — A subset E and time t define the population count being predicted. It is constitutive. Counterfactual: A timeless population label is not this theorem's random variable.
• Occupancy probability kernel — P(u,E) gives an arrival's chance of being in E after elapsed time u. It is central. Counterfactual: Arrival rate alone cannot determine regional occupancy.
• Poisson mean measure — Integrating inflow times occupancy chance produces the parameter of each count. It is constitutive. Counterfactual: The formula gives more than a first moment because the full count law is Poisson.
• Disjoint-region independence — Counts of distinct regions factor under the same assumptions. It is central. Counterfactual: Overlapping regions do not have independent counts in general.

What It Is Not

• Not Bartlett's test. That statistical test concerns variance homogeneity.
• Not every queue. Shared capacity and interactions can violate independence.
• Not just the mean. The conclusion gives a Poisson count distribution and disjoint-region independence.
• Not automatic empirical fit. Real census data can deviate from model predictions.
• Closest near-miss. A busy finite-bed hospital may resemble the infinite-server occupancy model in normal conditions, but bed blocking and correlated lengths of stay would invalidate a literal independent-path inference.

Scope of Application

• Queueing theory. Predict fixed-time counts in independent-customer networks.
• Population processes. Model independent individuals moving among states.
• Healthcare modeling. Approximate census from admissions and lengths of stay.
• Network planning. Evaluate region-level occupancy under Poisson inflow assumptions.

Clarity

If arrivals are Poisson and individuals move independently, the number found in a region at a fixed time is Poisson. Its mean accumulates arrival rate times each entrant's chance of still being there. Disjoint regions have independent counts. A hospital census model is one application, but data may differ when its assumptions or measurement time do not fit.

Manages Complexity

Poisson input and independent marking preserve the distribution under individual motion, not just the mean. This exact structure disappears when entrants interact or share a binding capacity. The theorem's empty-initial-state, time-dependent statement must also be distinguished from stationary M/G/∞ consequences and empirical approximations.

Abstract Reasoning

1. Specify the Poisson arrival rate and initial state.
2. Define individual movement or service-duration law.
3. Choose a region and fixed observation time.
4. Calculate each arrival cohort's retention probability.
5. Integrate intensity times retention to get the Poisson mean.
6. Check disjointness and independence before extending to joint counts or real forecasts.

Knowledge Transfer

The Poisson-marking mechanism travels across road, migration, service and communication systems if arrivals and paths satisfy its assumptions. A correlated arrival stream or finite queue with interactions is not a literal instance merely because a Poisson curve roughly fits observed counts.

Examples

Canonical

Kelly and Yudovina's Theorem 2.17 starts with an empty system, arrivals at Poisson rate ν, and independent movements on a state space. For a region E at time t, it assigns Poisson mean ν∫₀ᵗP(u,E)du; for disjoint E₁ and E₂ it proves the counts independent. This is the theorem's explicit construction, not a claim about a particular hospital or road.

Mapped back: Poisson arrival process → rate-ν source into an empty system; Independent individual paths → separate trajectory per arrival; Observation region and time → chosen E at fixed t; Occupancy probability kernel → P(u,E) for elapsed duration u; Poisson mean measure → ν∫₀ᵗP(u,E)du; Disjoint-region independence → factorization for E₁ and E₂.

Applied / In Practice

Garrison and Pecina analyzed actual family-medicine inpatient arrivals and census. Treating admissions as Poisson and stays as independent service times in an M/G/∞ approximation gives the one-region occupancy mean λE(B): they predicted 7.44 patients versus observed 7 a.m. mean 8.28. They did not establish exact Bartlett-model fit at the predicted mean, and finite hospital capacity remains a stated approximation limit.

Mapped back: Poisson arrival process → patient admissions modeled as Poisson; Independent individual paths → separate lengths of stay under the infinite-server approximation; Observation region and time → hospital inpatient service census; Occupancy probability kernel → chance an admitted patient remains hospitalized; Poisson mean measure → stationary λE(B), predicted 7.44; Disjoint-region independence → not tested in this one-region study.

Structural Tensions

T1 — Tractable Independence versus Capacity Interaction. Independent paths yield an exact Poisson law, while limited beds or shared servers create dependence.

Diagnostic: Is capacity nonbinding over the measured period?

T2 — Model-Based Forecast versus Observed Time-Specific Census. A stationary mean simplifies planning but a 7 a.m. census can systematically differ from a 24-hour average.

Diagnostic: Does the observation time match the model quantity?

T3 — Region Resolution versus Count Independence. Finer disjoint regions yield independent counts under assumptions; overlapping regions share individuals.

Diagnostic: Are the reported subsets disjoint?

Structural–Framed Character

A provisional portable skeleton is independently marking or thinning random arrivals while preserving a tractable count law. Bartlett's theorem here is the quantified Poisson occupancy result under independent arrivals and path/initialization conditions. Poisson Process is an input object, not a theorem parent.

Evaluative weight: Low; validity depends on assumptions, not desirability of the queue or population. Human-practice-bound: Low formally, though modelers select regions, intensities, and observation time. Institutional origin: Queueing theory names and proves the result; a similar empirical fit does not establish it. Vocabulary travels: The result applies across services, roads, and migration if independent-arrival and marking conditions hold. Import versus recognize: A new model is recognized as an instance by verifying those assumptions and the mean integral; calling interacting finite-server counts Poisson by analogy imports the conclusion.

Its character: A conditional mathematical theorem with transferable counting logic and strict stochastic premises.

Structural Core vs. Domain Accent

Skeletal core. Independently marked random arrivals retain a tractable count law under selection. Domain-bound accent. Queueing/population arrivals, trajectories, regions and occupancy probabilities define Bartlett's theorem. Transfer boundary. Unconstrained marking does not cover interacting finite-server systems or unrelated Bartlett-named statistics.

Instantiates / Related Primes

This entry is a kind of Poisson Process.

• Neighbor: Poisson thinning. It underlies the proof but does not by itself name the occupancy-over-time theorem.

• Neighbor: Little's law. It relates average number, arrival rate and mean duration but alone does not establish a Poisson count distribution.

Relationships to Other Abstractions

Current abstraction Bartlett's theorem Domain-specific

Parents (1) — more general patterns this builds on

• Bartlett's theorem is a kind of Poisson Process Prime

  Bartlett's theorem derives an occupancy law directly from Poisson-arrival, independent-evolution assumptions.

Hierarchy paths (5) — routes to 5 parentless roots

• Bartlett's theorem → Poisson Process → Markov Process → Stochastic Process

Show alternative paths (4)

Neighborhood in Abstraction Space

Bartlett's theorem sits in a moderately populated region (49^th percentile for distinctiveness): it has near-neighbors but no dense thicket of look-alikes.

Family — Queueing, Networks & Concurrent Systems (9 abstractions)

Nearest neighbors

• Jackson's Theorem (Queueing Theory) — 0.87
• Lincoln Index — 0.87
• Network mapping — 0.87
• Evacuation Simulation — 0.87
• Reachability analysis — 0.86

Computed from structural-signature embeddings · 2026-10-08

Not to Be Confused With

• Bartlett's variance test. Tell: A separate statistical homogeneity procedure.
• Burke's theorem. Tell: An output-process result under different queue assumptions.
• Little's law. Tell: An average relation without the full count law.
• General M/G/1 queue. Tell: A single finite server induces waiting interactions absent here.

References

• F. P. Kelly and E. Yudovina, Stochastic Networks, Theorem 2.17 and Remarks 2.18–2.19 — initially empty independent-movement theorem, integrated regional mean, and disjoint-region independence proof.
• P. K. Pollett and colleagues, “Ensemble behaviour in population processes with applications” (2007), §3 — stationary open-population region counts attributed to Bartlett.
• G. M. Garrison and J. L. Pecina, “Using the M/G/∞ queueing model to predict inpatient family medicine service census and resident workload” (2016) — actual hospital occupancy model, 7.44 predicted versus 8.28 observed 7 a.m. mean, and finite-capacity limitation.

The hospital is a one-region stationary special case used as an approximation; its data do not verify the full theorem's assumptions or predicted-mean fit.