List of topics covered in Stat 330/600 in spring 2017
-
Why bother with measure theory?
Advantages and disadvantages of countable additivity.
Sigma-fields are forced upon us. Why Lebesgue was a genius.
-
Generating classes for sigma-fields.
The importance of picking the generating class to suit the problem.
-
Measurable functions. The benefits and disadvantages of allowing
a real function to take infinite values.
-
The cone ℳ+ of [0,∞]-valued measurable functions.
Approximation by simple functions.
-
Integrals as "increasing `linear' functionals on ℳ+ with the monotone convergence property".
ℒ1 as a vector space.
-
Image measures. Distributions of random variables, random vectors, and stochastic processes.
-
Dominated Convergence. Differentiation under integral signs.
-
Negligible sets. ℒ1 versus L1.
-
Convexity-based concepts: ℒp and Orlicz spaces; Hölder inequality;
Jensen's inequality; subgaussianity. The clever converse to Borel-Cantelli.
-
The π–λ theorem for sets.
-
Independence of sigma-fields.
-
Strong law of large numbers with fourth moments. (Better versions later.)
-
Independence and product measures. Lambda spaces of functions (important: replaces lambda cones in UGMTP).
The π–λ theorem for functions. Tonelli and Fubini. Convolutions.
-
Quantile transformations.
-
Construction of measures on product spaces via kernels. Bayes. Brief mention of disintegration.
-
Conditional probability distributions.
- Projections in ℒ2. (Hilbert spaces.) Radon-Nikodym.
Kolmogorov conditional expectations.
-
Measure preserving transformations and the Ergodic theorem. Comment on SLLN.
-
Martingale et al. Stopping times. Information available at a stopping time. The Stopping Time Lemma (a.k.a. the magic lemma). Using STL to prove maximal inequalities. Convergence of positive supermartingales via Dubins's inequality. Applications.
-
Convergence in distribution for metric spaces. Bounded Lipschitz functions.
Continuous Mapping Theorem.
-
Central limit theorem. Lindeberg and beyond.
-
Stochastic order symbols.
- This year the material on Fourier transforms (a.k.a. characteristic functions) fell
off the end of the course.