Working Through Subrahmaniam's Probability Textbook
Most people grab A Primer In Probability Second Edition Kathleen Subrahmaniam because they need a bridge between basic statistics and formal measure-theoretic treatment. It fills a real gap. I ran into this when someone on a forums I participate in asked about conditional expectation without Lebesgue integration, and I pointed them here after working through it myself. The book is structured in two major parts. The first covers classical probability, discrete and continuous random variables, joint distributions, and transformations. The second shifts toward more rigorous treatment including limits of sequences of events, the law of large numbers, and the central limit theorem with proper convergence arguments. Chapter 6 on properties of expectation is where the book separates itself from lighter introductions. It doesn't just state linearity — it derives it and then shows where intuition fails if you skip the justification. I remember hitting a wall around Chapter 4 when dealing with transformation of random variables. The formula for the change of variables technique assumes a monotonic mapping, but the book presents it in a way that doesn't immediately flag edge cases. I spent maybe two hours on a problem involving the absolute value of a normal variable where I kept getting the support wrong because I wasn't accounting for the fold-over effect. The workaround was to draw the mapping on paper first, label every region, and then check whether the Jacobian was one-to-one across the domain. That visual step probably saved me an hour of algebra errors.
How to actually use this book
Don't read it cover to cover straight through. The notation shifts between chapters without a running legend, and the examples assume fluency with calculus that some readers still need to rebuild. I recommend going through chapters 1 through 5 in order, then looping back to chapter 3 after you've done chapter 5, because the marginal distribution problems in chapter 5 rely on convolution techniques from chapter 3 that you'll have forgotten otherwise. The exercises are where most people get stuck. They're not trivial but they're also not brutal. The harder problems at the end of each chapter sometimes require constructing a counterexample rather than computing a value. I found it useful to sit with a single exercise for at least twenty minutes before looking at any solution manual or hint. The book's answer key only gives final results for most odd-numbered problems, so you need to verify your own work independently. One thing the book does well that other texts miss: it treats covariance and correlation not just as formulas but as geometric objects. The angle interpretation of correlation appears early enough that it sticks. I used this perspective when explaining to a colleague why two variables can be uncorrelated but dependent — the covariance matrix being diagonal doesn't imply independence unless you're working with jointly Gaussian vectors. That distinction trips up everyone at some point.
Where it falls apart
The book doesn't cover stochastic processes, martingales, or Markov chains. If you need those, you'll need a supplementary text. The treatment of characteristic functions in the central limit theorem chapter is adequate but terse — roughly eight pages total. If you're preparing for a qualifying exam that emphasizes Fourier methods, you'll want to supplement with a chapter from Feller or Durrett. The second edition added some new material but the typesetting is inconsistent. Some equations in later chapters have misaligned subscripts that made copy-pasting into LaTeX a minor headache. Also, there's no online errata page I could find that's actively maintained, so if you spot an error, you're on your own to verify it against the first edition or your own calculations.
Get the Full Details
Supplementary resources
Pair this with online lecture notes from a graduate-level probability course for the measure-theoretic sections. The transition from Riemann-style reasoning to Lebesgue reasoning is where the book gets thin. MIT OpenCourseWare has materials that align closely with chapters 6 and 7. For computational practice, working through the same problems in R or Python using simulation helps lock in the intuition before you push further into proofs. The book runs around 400 pages in the second edition. Budget roughly three weeks for a careful first pass if you're doing the exercises, or five weeks if you're working through it alongside a course. Rushing it defeats the purpose — the later chapters reward you for having internalized the earlier ones.
