Poisson Processes and Counting Models
Poisson process, counting process, independent increments, exponential waiting time, arrivals
1 Role
This is an early live page in the Stochastic Processes module.
Its job is to introduce the cleanest first counting process:
- events arrive randomly over time
- counts accumulate
- increments over disjoint intervals do not interfere with one another
This is where stochastic processes stop looking only like random state evolution and start looking like random event streams.
2 First-Pass Promise
You can read this page on its own inside the full module spine.
If you stop here, you should still understand:
- what a Poisson process is
- what the rate parameter means
- why event counts over intervals are Poisson-distributed
- why exponential waiting times are tied to the same model
3 Why It Matters
Many random systems are driven not only by states but by arrivals:
- customers entering a queue
- packets arriving at a receiver
- failures occurring in a network
- photons or spikes arriving in a sensor stream
- requests appearing in an online system
The Poisson process is the first major model for that kind of randomness.
It is useful because it is:
- simple enough to analyze exactly
- rich enough to capture random event timing
- close to many first approximations used in queueing, sensing, communication, and reliability
Once a reader sees that, random event timing becomes a reusable object rather than a domain-specific trick.
4 Prerequisite Recall
- from Probability, the Poisson distribution models random counts and the exponential distribution models memoryless waiting times
- from Markov and stochastic-process intuition nearby, a process can be understood through its path over time rather than through one isolated random variable
- from signal or sensing language nearby, event streams are often the right model when the data are spikes, arrivals, or detections rather than smooth waveforms
5 Intuition
5.1 A Counting Process Tracks How Many Events Have Happened
Instead of recording a state value directly, a counting process records cumulative arrivals:
\[ N(t) = \text{number of events observed up to time } t. \]
So the path:
- starts at zero
- jumps upward by one at each event
- stays flat between events
This already makes the object feel different from a smooth signal or a finite-state chain.
5.2 The Rate Controls Average Event Density
The parameter \lambda > 0 is the event rate.
At first pass, it means:
- on average, about
\lambda tevents occur over a time window of lengtht
So larger \lambda means denser arrival streams.
5.3 Independent Increments Mean Separate Time Windows Do Not Interfere
If two time intervals do not overlap, then the event counts in those intervals are independent.
This is one of the defining structural simplifications of the Poisson process.
It says:
- what happened in one window does not bias the random event count in a disjoint future window
5.4 Waiting Times Are Exponential
One of the most useful equivalent views is that:
- the waiting time to the first event is exponential
- successive interarrival times are i.i.d. exponential
So the same process can be understood from two angles:
- counts in windows
- waiting times between events
6 Formal Core
Definition 1 (Definition: Counting Process) A counting process (N(t))_{t \ge 0} is a stochastic process that counts how many events have occurred by time t.
At first pass, it should satisfy:
N(0)=0N(t)is integer-valued and nondecreasing- jumps correspond to events
Definition 2 (Definition: Poisson Process) A Poisson process with rate \lambda > 0 is a counting process (N(t))_{t \ge 0} such that:
N(0)=0- it has independent increments
- for any
s < t, the incrementN(t)-N(s)is Poisson with mean\lambda (t-s)
So event counts over a window depend only on the window length, not on its location.
Theorem 1 (Theorem Idea: Poisson Counts And Exponential Waiting Times Are Two Views Of The Same Process) For a Poisson process with rate \lambda, the waiting times between successive events are i.i.d. exponential random variables with rate \lambda.
At first pass, the key point is:
- window counts and interarrival times are two equivalent ways to describe the same random arrival mechanism
Theorem 2 (Theorem Idea: Independent Poisson Streams Add) If two independent Poisson processes have rates \lambda_1 and \lambda_2, then their superposition is again a Poisson process with rate \lambda_1 + \lambda_2.
This is one reason Poisson models appear so naturally in networks and queueing systems.
7 Worked Example
Suppose packets arrive at a network buffer according to a Poisson process with rate
\[ \lambda = 3 \text{ packets per second}. \]
Then over a 2-second interval, the number of arrivals has distribution
\[ N(2) \sim \text{Poisson}(6). \]
So:
- the mean number of arrivals is
6 - the variance is also
6
The probability of exactly k arrivals is
\[ \mathbb{P}(N(2)=k) = e^{-6}\frac{6^k}{k!}. \]
Now look at waiting times.
The time T_1 until the first packet arrives is exponential with rate 3, so
\[ \mathbb{P}(T_1 > t) = e^{-3t}. \]
That same stream is therefore being described in two equivalent ways:
- by event counts in windows
- by random delays between arrivals
This is the core modeling flexibility of the Poisson process.
8 Computation Lens
When you meet an arrival model, ask:
- is the object really a count over time rather than a continuously varying state?
- does a constant-rate approximation make sense, or is the rate clearly changing over time?
- do we care more about counts in windows, waiting times, or long-run throughput?
- are multiple arrival streams being combined or split?
- is the Poisson model a realistic mechanism, or just a first baseline?
Those questions usually reveal whether the right next step is:
- queueing analysis
- hidden-state modeling
- event detection
- or a more structured nonhomogeneous process
9 Application Lens
9.1 Queueing And Operations
Poisson arrivals are the first default model for customers, jobs, or packets entering a system.
9.2 Communication And Networks
Packet traffic, event arrivals, and independent request streams often begin with Poisson-style intuition before more refined traffic models are introduced.
9.3 Sensing And Event Streams
Photon counts, spike trains, and event detections often use counting-process language more naturally than smooth-signal language.
9.4 Markov And Continuous-Time Bridges
Poisson processes are also a first clean step from discrete-time models toward continuous-time stochastic dynamics.
10 Stop Here For First Pass
If you stop here, retain these five ideas:
- a Poisson process models random event counts over time
- the rate
\lambdacontrols average event density - independent increments are the main structural simplification
- counts in windows and exponential waiting times are two views of the same process
- Poisson models are the first shared language for queues, event streams, sensors, and network arrivals
11 Go Deeper
The strongest adjacent live pages right now are:
- Probability
- Martingales and Optional Stopping Intuition
- Markov Chains and Stationary Distributions
- Controlled Markov Models, Policies, and Cost Functionals
- Signals, Channels, and Noisy Measurements
- State Estimation, Smoothing, and Hidden-State Inference
The natural conditional-expectation bridge in this module is:
12 Optional Deeper Reading After First Pass
The strongest current references connected to this page are:
- MIT 18.445 lecture notes page - official MIT route through martingales, Poisson processes, and continuous-time chains. Checked
2026-04-25. - MIT 18.445 lecture 20 - official MIT note focused on the Poisson-process viewpoint itself. Checked
2026-04-25. - MIT 6.262 Discrete Stochastic Processes - official MIT course hub for arrival processes and countable-state stochastic models. Checked
2026-04-25. - Stanford MS&E 221 - official Stanford stochastic-modeling course page where random arrivals and queueing-style phenomena fit naturally. Checked
2026-04-25.
13 Sources and Further Reading
- MIT 18.445 lecture notes page -
First pass- official MIT lecture hub for a broad stochastic-process route including Poisson processes. Checked2026-04-25. - MIT 18.445 lecture 20 -
First pass- official MIT note for the Poisson-process definition and first structural properties. Checked2026-04-25. - MIT 6.262 Discrete Stochastic Processes -
Second pass- official MIT course hub for discrete stochastic-process and arrival-process intuition. Checked2026-04-25. - Stanford MS&E 221 -
Second pass- official Stanford stochastic-modeling anchor for random arrivals, Markov models, and queueing applications. Checked2026-04-25. - Stanford MS&E 221 references -
Bridge outward- useful official Stanford reference page once you want a broader modeling route beyond the first-pass Poisson baseline. Checked2026-04-25.