almost everywhere / almost surely

In measure-theoretic probability you constantly meet statements that are true except on a set so tiny it carries no weight at all. Almost everywhere (in general measure theory) and almost surely (in probability) are the precise words for this: a property holds almost everywhere if the set of points where it fails has measure zero, and it holds almost surely if the event where it fails has probability zero. The exceptions are not forbidden — they are simply invisible to the measure.

Formally, a statement holds almost surely if P(the set where it fails) = 0. The crucial subtlety is that probability zero is not the same as impossible. Pick a uniform random number from [0,1]: the probability of getting exactly 0.5 is zero, yet 0.5 is a perfectly possible outcome — every single specific value has probability zero, even though some value certainly occurs. So almost surely means as good as certain for all probabilistic purposes, while quietly allowing a measure-zero set of exceptions to exist. The phrase shows up everywhere: the strong law of large numbers says the running average converges to the mean almost surely; two random variables are equal almost surely if they differ only on a null event.

Why does this matter so much? Because measure theory is built to ignore null sets, almost-everywhere equality is the right notion of sameness throughout the subject. The Radon-Nikodym derivative is unique only almost everywhere; changing a density on a measure-zero set changes nothing observable; integrals and expectations cannot detect differences on null sets. Treating two functions that agree almost everywhere as the same object is not sloppiness — it is exactly the equivalence the theory respects, and forgetting that probability zero allows real exceptions is a classic beginner's error.