Interactive demo · Probability and Statistical Inference · Chapter 5

Modes of convergence and their counterexamples

Five ways for random variables Xn to approach a limit X, and the sequences that tell them apart. The spike Xn = na I(Un < n−b) is zero except on an event of probability n−b, where it equals na. Its tail probability and moments depend only on a and b; whether its paths settle at zero depends on how the Un are coupled. With one uniform U for the whole sequence the spikes stop once n−b < U; with independent Un and b ≤ 1 they never stop. The top panel shows, for 25 sample paths, every time n at which |Xn − X| > ε; the bottom panel the exact probabilities and moments, with Monte Carlo over 2000 paths.

Sequence

Spike height na: a

Spike probability n−b: b

Coupling

Tolerance ε

Horizon N

Try these

Spike: Xn = na I(Un < n−b), limit 0; "one U" uses Un = U for every n. Normal: Xn = Zn/√n ~ N(0, 1/n) (Example 5.3), limit 0; "one U" means one Z for the whole path. Xn = −X with X ~ N(0, 1), limit X; the coupling does not matter. "Later" in the bottom panel means some m with n ≤ m ≤ N; the readout also gives the exact value for all m ≥ N.

In the lecture notes