Toward measure-theoretic probability

A fair coin has no memory, yet its running proportion of heads settles near one half. What exactly does that mean? A statement about the probability of a large error at time $n$ differs from a statement about almost every infinite sequence of tosses. Neither says that heads and tails must balance after a bad run. The passage from elementary probability to measure-theoretic probability begins when these distinctions become part of the question, rather than technical qualifications added after a calculation.

From calculations to measure

If distributions, expectation, and conditioning are new, start with Charles M. Grinstead and J. Laurie Snell's Introduction to Probability, in the freely available CHANCE Project version. Chapters 1–2 introduce discrete and continuous models; Chapters 4 and 6 develop conditioning and expectation. Then compare Chapter 8 on laws of large numbers with Chapter 9 on central limit theorems. Keep the coin experiment beside the theory: estimate a tail probability, simulate several sample paths if useful, and ask which mathematical statement each picture could illustrate. A plot supplies examples, while a proof controls all the relevant outcomes at once.

You can skip much of this opening if you can calculate an expectation from a distribution, use conditional probability without reversing the conditioning event, and derive the variance of a sum of independent variables. Try those tasks before deciding that another introductory book is needed. Calculus supports continuous distributions, and elementary counting supports discrete ones. For the next stage, add rigorous analysis: sequences of functions, countable operations on sets, and Lebesgue integration. Being quick with densities is useful preparation, but probability measures also describe discrete laws and mixtures that have no density with respect to ordinary length.

The destination text is Rick Durrett's Probability: Theory and Examples, fifth-edition author PDF. Chapter 1 develops the measure and integration framework; Chapters 2–3 address laws of large numbers and central limit theorems. Readers who already know measure theory can test themselves on measurable random variables and convergence theorems, then move directly into probabilistic arguments. Readers meeting integration for the first time should work examples slowly and use the opening chapter as a guide to the analysis they need. A short review cannot replace practice applying monotone and dominated convergence with their hypotheses visible.

A limit that cannot pass through expectation

A useful exercise is to define $X_n$ on $[0,1]$ to equal $n$ on $(0,1/n)$ and zero elsewhere, with uniform probability. For each fixed point, $X_n$ eventually vanishes, but its expectation remains one. Explain why moving a limit through an expectation fails here. Then modify the example so that $|X_n|\leq M$ for one constant $M$ independent of $n$, while retaining pointwise convergence to zero, and check what happens to the expectations. This small calculation exposes a recurring issue in probability: the rare outcomes can become larger while their probabilities shrink. It also gives you something concrete to recall when a proof asks for integrability or uniform control.

Conditioning on information

From there, choose the next question. Durrett's Chapter 4 develops conditional expectation and martingales. For an integrable variable, conditioning on a sigma-algebra produces another random variable that reflects the information retained; it need not reduce to dividing two event probabilities. Start with a finite partition and compute the average on each piece before attempting the abstract definition. If finite-state evolution is more appealing, Grinstead and Snell's Chapter 11 offers a Markov-chain route with matrix calculations. These are meaningful alternatives: one emphasizes information and conditional averages, the other transitions between states.

Keep a comparison sheet for almost sure convergence, convergence in probability, convergence in distribution, and convergence in mean. Supply examples and counterexamples as each appears. Before using a limit theorem, name its assumptions and the quantity being normalized. Later branches include Brownian motion, stochastic integration, interacting systems, and links with ergodic theory. Choose one after you can carry a complete argument from a probabilistic model through its measure-theoretic formulation to a justified conclusion.

Suggest an edit · Report a correction

Original introduction: CC BY-SA 4.0.