Poisson Distribution

A discrete random variable X follows a Poisson distribution with parameter λ > 0 (written X ~ Poisson(λ)) if its PMF is P(X = x) = (λ^x e^(−λ)) / (x!) for x ∈ 0, 1, 2, ṡ. Its expected value is E[X] = λ, and its variance is Var(X) = λ (standard deviation σ = √(λ)). This equality of mean and variance is called 'equidispersion'. If empirical data has variance exceeding the mean, it indicates overdispersion (often modeled by Negative Binomial).

In cloud server engineering, the arrival of inbound HTTP requests to an API gateway per second follows a Poisson distribution. Site Reliability Engineers (SREs) at Google and AWS use λ to size auto-scaling compute pools and provision DDoS mitigation buffers.

French mathematician Siméon Denis Poisson introduced this law in his 1837 work Recherches sur la probabilité des jugements ('Research on the Probability of Judgments'), originally applying it to model erroneous jury convictions in criminal court cases.

The Poisson Signature: 'Mean equals Variance equals Lambda'. If you know the average count λ, you instantly know the whole universe of its probabilities.

Start with the Binomial PMF: P(X=x) = (n!) / (x!(n-x)!) p^x (1-p)^(n-x). Substitute p = λ / n. The formula factors into four distinct pieces:

1. (n(n-1)…(n-x+1)) / (n^x) → 1 as n → ∞.
2. (λ^x) / (x!) (remains constant with respect to n).
3. (1 - λ/n)^n → e^(−λ) by the fundamental definition of Euler's constant.
4. (1 - λ/n)^(-x) → (1 - 0)^(-x) = 1.

Multiplying these limits together produces: lim_(n → ∞) P(X=x) = 1 · (λ^x) / (x!) · e^(−λ) · 1 = (λ^x e^(−λ)) / (x!).

In insurance underwriting, catastrophic hurricanes or nuclear plant leaks are modeled via Poisson because millions of homes (large n) have tiny individual risk (small p), yielding a steady annual expected count λ.

Prussian economist Ladislaus Bortkiewicz famously verified Poisson's law of rare events in 1898 in his book Das Gesetz der kleinen Zahlen ('The Law of Small Numbers'), analyzing 20 years of data on Prussian soldiers killed accidentally by horse kicks.

Limit Rule: 'n goes to infinity, p goes to zero, np stays lambda'. The Binomial combinatorial mess evaporates, leaving λ^x e^(−λ) / x!.

The Binomial framework demands two inputs: total trials n and per-trial probability p. In physical reality, how many 'trials' occur when cars pass an intersection, when Geiger counters click near radioactive uranium, or when patients enter an emergency room? There is no natural discrete trial n. Trying to force Binomial requires arbitrary discretization (e.g. seconds vs milliseconds). Poisson frees us from the artificial trial count n by characterizing the process purely by the average rate of arrival λ per unit interval.

In telecommunications network design, call arrivals at a cell tower cannot be represented as trials of individual phone owners; instead, cellular engineers use Poisson point processes with rate λ calls/minute to dimension transmission trunk lines (Erlang capacity models).

Danish engineer A.K. Erlang in 1909 extended Poisson's distribution to telephony, creating modern queueing theory to determine how many telephone switchboard circuits Copenhagen needed to prevent dropped calls.

Model Selector: 'Know n? Use Binomial. Don't know n, only rate? Use Poisson'.

Examine P(X=x) = (λ^x e^(−λ)) / (x!):

• λ^x: Represents the accumulation of x events occurring at average intensity λ. Higher rate λ amplifies this term.
• e^(−λ): The baseline probability of zero events occurring, P(X=0) = e^(−λ). It scales the entire distribution so that total probability sums to 1.
• x! (x factorial): The permutation count of x indistinguishable arrivals, ensuring the PMF decays rapidly for large x.
• Normalization check: Summing over all x gives ∑_(x=0)^∞ (λ^x e^(−λ)) / (x!) = e^(−λ) ∑_(x=0)^∞ (λ^x) / (x!) = e^(−λ) e^λ = 1.

In genomics, Next-Generation Sequencing (NGS) uses this formula to calculate sequencing depth (coverage): λ is average read depth across the genome, and P(X=0) = e^(−λ) calculates the fraction of unsequenced genomic gaps (Lander-Waterman model).

The appearance of e^(−λ) links Poisson probability directly to the Taylor series expansion of e^x discovered by Isaac Newton and formalized by Leonhard Euler.

Formula Mnemonic: 'Lambda to the x, times e to the minus lambda, over x factorial'. The three ingredients: Power, Decay, Factorial.

A counting process N(t), t ≥ 0 is a Poisson process with rate λ if and only if four axioms hold:

1. Independence: Counts in non-overlapping intervals are mutually independent.
2. Stationarity / Constant Rate: The probability of an event in [t, t+h] depends only on the interval length h, with rate λ h + o(h).
3. Orderliness (Single Arrivals): Two or more events cannot occur at the exact same instant: P(N(t+h) - N(t) ≥ 2) = o(h).
4. Initial State: N(0) = 0.

When these axioms hold, the count of events in any interval of length t follows Poisson(λ t).

In hospital emergency ward management, admissions are checked against these axioms. If a multi-casualty highway accident occurs, the 'single arrival' condition is violated (clustering occurs), requiring compound Poisson models rather than standard Poisson.

Russian mathematician Alexander Khinchin in 1956 proved that the sum of many independent, sparse point processes converges to a Poisson process (Khinchin's Theorem).

The ISOC Rules: Independent, Single, Orderly, Constant rate. If events travel in packs (like bus bunching), Poisson fails.

Traditional probability calculates relative frequency: P = k/n. However, consider radioactive atoms decaying, cosmic rays striking a satellite sensor, or potholes forming along a 100-mile highway. What is n? Is it the number of square meters? The number of seconds? Poisson resolves this by eliminating n entirely. Instead of measuring success per attempt, it measures events per unit exposure: λ = intensity × exposure. This shift from 'fraction of trials' to 'density of arrivals' represents the foundation of modern continuous stochastic calculus.

In astronomical data analysis (e.g. James Webb Space Telescope), photon arrivals on charge-coupled device (CCD) pixel sensors are modeled via Poisson shot noise to distinguish dim exoplanet signals from background cosmic noise.

Walter Schottky in 1918 analyzed current fluctuations in vacuum tubes, identifying the fundamental physical phenomenon of 'shot noise' as discrete electrons arriving according to Poisson statistics.

Mental Hook: 'From Fractions to Rates'. When you cannot count the trials (no n), count the clock and use the rate (λ).

The parameter λ is always bound to a specific interval. Let r be the intrinsic event rate per unit time. Over an observation window of length t, the Poisson parameter is λ = r · t. If a customer support desk receives r = 4 inquiries per hour:

• For 1 hour: λ = 4 × 1 = 4. Mean = 4, Variance = 4.
• For 30 minutes (0.5 hour): λ = 4 × 0.5 = 2. Mean = 2, Variance = 2.
• For an 8-hour workday: λ = 4 × 8 = 32. Mean = 32, Variance = 32.

Furthermore, by the additive property of Poisson variables, if X₁ ~ Poisson(λ₁) and X₂ ~ Poisson(λ₂) are independent, their sum X₁ + X₂ ~ Poisson(λ₁ + λ₂).

In emergency 911 dispatch centers, dispatch supervisors dynamically calibrate λ across morning rush hour, afternoon lull, and weekend surges to schedule operator staffing using queueing formulas (M/M/c queues).

The Greek letter λ (lambda) was chosen by 19th-century European statisticians from the Greek word logos (ratio/proportion) to denote constant arrival intensity.

The Window Rule: 'Double the window, double the lambda'. λ = rate × window. Keep units strictly consistent!