Convergence Needs a Reference
What does it mean for a random sequence to converge?
How this lesson unfolds.
Begin with Converges—in what sense, and toward what? The sequence then connects Infinitely many shrinking gaps can reach one target and builds toward Convergence in distribution compares probability laws.
Converges—in what sense, and toward what?
A convergence claim needs an object, a reference target and a rule for measuring closeness. Different rules create genuinely different mathematical statements.
- Learning goalDistinguish convergence notions by their reference object and measure of closeness.
- Helpful foundationssequences · random variables · expectation
- Common trapConvergence is not one universal idea; the reference and measure of closeness must be stated.
- Next connectionInfinitely many shrinking gaps can reach one target
Can infinitely many moves still have a finite destination?
Zeno's half-distance process produces partial sums 1, 1.5, 1.75, 1.875 and so on. The number of moves is infinite, but the remaining distance shrinks to zero and the sums converge to 2.
- Learning goalDistinguish convergence notions by their reference object and measure of closeness.
- Helpful foundationssequences · random variables · expectation
- Common trapConvergence is not one universal idea; the reference and measure of closeness must be stated.
- Next connectionOrdinary convergence follows one deterministic sequence
What does ordinary convergence measure?
For one fixed sequence S_n and target 2, ordinary convergence means the numerical error |S_n-2| tends to zero. One path, one destination, one shrinking gap.
- Learning goalDistinguish convergence notions by their reference object and measure of closeness.
- Helpful foundationssequences · random variables · expectation
- Common trapConvergence is not one universal idea; the reference and measure of closeness must be stated.
- Next connectionRandom repetition creates many possible paths
Why does probability need more than one notion of convergence?
Repeating a random experiment produces different observation sequences and therefore different running-average paths. A probabilistic convergence statement must specify how those possible paths are judged.
- Learning goalDistinguish convergence notions by their reference object and measure of closeness.
- Helpful foundationssequences · random variables · expectation
- Common trapConvergence is not one universal idea; the reference and measure of closeness must be stated.
- Next connectionConvergence in probability reads across experiments
When do large errors become rare?
X_n converges in probability to mu when, for every tolerance epsilon greater than zero, P(|X_n-mu|>epsilon) tends to zero. At each fixed n, look across repeated experiments and count the fraction outside the tolerance band.
- Learning goalDistinguish convergence notions by their reference object and measure of closeness.
- Helpful foundationssequences · random variables · expectation
- Common trapConvergence is not one universal idea; the reference and measure of closeness must be stated.
- Next connectionA sample path extends one experiment
Why follow one growing experiment?
A path X_1,X_2,... is built by extending one growing data stream. Each new observation updates the same running estimate, so eventual path behavior can be discussed coherently.
- Learning goalDistinguish convergence notions by their reference object and measure of closeness.
- Helpful foundationssequences · random variables · expectation
- Common trapConvergence is not one universal idea; the reference and measure of closeness must be stated.
- Next connectionRunning averages form a sequence, not a series
Where does a random sample path come from?
Raw observations arrive, a running average is recomputed after each arrival, and those ordered estimates are plotted against n. The estimates form a sequence; they are not terms being added into an infinite series.
- Learning goalDistinguish convergence notions by their reference object and measure of closeness.
- Helpful foundationssequences · random variables · expectation
- Common trapConvergence is not one universal idea; the reference and measure of closeness must be stated.
- Next connectionAlmost-sure convergence follows paths horizontally
When does one path eventually stay close?
Almost-sure convergence asks whether, with probability 1, each path has a last exit from every tolerance band and then remains inside. Probability-zero exceptional paths may still exist.
- Learning goalDistinguish convergence notions by their reference object and measure of closeness.
- Helpful foundationssequences · random variables · expectation
- Common trapConvergence is not one universal idea; the reference and measure of closeness must be stated.
- Next connectionFive convergence notions measure different objects
What exactly is getting closer?
Ordinary convergence measures numerical error; convergence in probability measures the chance of exceeding a tolerance; almost sure convergence measures eventual path behavior; convergence in distribution measures laws; and Lp convergence measures average powered error.
- Learning goalDistinguish convergence notions by their reference object and measure of closeness.
- Helpful foundationssequences · random variables · expectation
- Common trapConvergence is not one universal idea; the reference and measure of closeness must be stated.
- Next connectionThe power p chooses an error lens
What extra information does Lp add?
For the same approximation error |X_n-mu|, raising it to p changes how strongly large errors are emphasized. Taking expectations produces a new deterministic sequence of average powered errors.
- Learning goalDistinguish convergence notions by their reference object and measure of closeness.
- Helpful foundationssequences · random variables · expectation
- Common trapConvergence is not one universal idea; the reference and measure of closeness must be stated.
- Next connectionLp convergence averages powered error
When does average powered error vanish?
X_n converges to X in Lp when E(|X_n-X|^p) tends to zero. p=1 measures average absolute error, p=2 average squared error, and larger p penalizes extreme errors more strongly.
X_n converges to X in Lp when E(|X_n-X|^p) tends to zero.p=1 measures average absolute error, p=2 average squared error, and larger p penalizes extreme errors more strongly.
- Learning goalDistinguish convergence notions by their reference object and measure of closeness.
- Helpful foundationssequences · random variables · expectation
- Common trapConvergence is not one universal idea; the reference and measure of closeness must be stated.
- Next connectionRepeated samples build a distribution
What changes when each estimate averages more observations?
Across repeated samples, one value per trial can be skewed, averages of five are less skewed, and averages of thirty are smoother and more symmetric. The histogram is a distribution of estimators, not one path.
- Learning goalDistinguish convergence notions by their reference object and measure of closeness.
- Helpful foundationssequences · random variables · expectation
- Common trapConvergence is not one universal idea; the reference and measure of closeness must be stated.
- Next connectionConvergence in distribution compares probability laws
Can values differ while their distributional shapes converge?
Convergence in distribution means the probability law of X_n approaches a target law. Individual outcomes need not approach one value; it is the distributional shape or CDF that converges.
- Learning goalDistinguish convergence notions by their reference object and measure of closeness.
- Helpful foundationssequences · random variables · expectation
- Common trapConvergence is not one universal idea; the reference and measure of closeness must be stated.
- Next connectionCompare implications and counterexamples among convergence notions