This preview shows pages 1–3. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: Chapter 30 General Theory of Large Deviations A family of random variables follows the large deviations princi ple if the probability of the variables falling into “bad” sets, repre senting large deviations from expectations, declines exponentially in some appropriate limit. Section 30.1 makes this precise, using some associated technical machinery, and explores a few consequences. The central one is Varadhan’s Lemma, for the asymptotic evalua tion of exponential integrals in infinitedimensional spaces. Having found one family of random variables which satisfy the large deviations principle, many other, related families do too. Sec tion 30.2 lays out some ways in which this can happen. As the great forensic statistician C. Chan once remarked, “Improbable events permit themselves the luxury of occurring” (reported in Biggers, 1928). Large deviations theory, as I have said, studies these little luxuries. 30.1 Large Deviation Principles: Main Defini tions and Generalities Some technicalities: Definition 397 (Level Sets) For any realvalued function f : Ξ → R , the level sets are the inverse images of intervals from∞ to c inclusive, i.e., all sets of the form { x ∈ Ξ : f ( x ) ≤ c } . Definition 398 (Lower SemiContinuity) A realvalued function f : Ξ → R is lower semicontinuous if x n → x implies lim inf f ( x n ) ≥ f ( x ) . 206 CHAPTER 30. LARGE DEVIATIONS: BASICS 207 Lemma 399 A function is lower semicontinuous iff either of the following equivalent properties hold. i For all x ∈ Ξ , the infimum of f over increasingly small open balls centered at x approaches f ( x ) : lim δ → inf y : d ( y,x ) <δ f ( y ) = f ( x ) (30.1) ii f has closed level sets. Proof: A characterbuilding exercise in real analysis, left to the reader. Lemma 400 A lower semicontinuous function attains its minimum on every nonempty compact set, i.e., if C is compact and = ∅ , there is an x ∈ C such that f ( x ) = inf y ∈ C f ( y ) . Proof: Another characterbuilding exercise in real analysis. Definition 401 (Logarithmic Equivalence) Two sequences of positive real numbers a n and b n are logarithmically equivalent , a n b n , when lim n →∞ 1 n (log a n log b n ) = 0 (30.2) Similarly, for continuous parameterizations by > , a b when lim → (log a log b ) = 0 (30.3) Lemma 402 (“Fastest rate wins”) For any two sequences of positive num bers, ( a n + b n ) a n ∨ b n . Proof : A characterbuilding exercise in elementary analysis. Definition 403 (Large Deviation Principle) A parameterized family of ran dom variables, X , > , taking values in a metric space Ξ with Borel σfield X , obeys a large deviation principle with rate 1 / , or just obeys an LDP , when, for any set B ∈ X , inf x ∈ int B J ( x ) ≤ lim inf → log P ( X ∈ B ) ≤ lim sup → log P ( X ∈ B ) ≤  inf x ∈ cl B J ( x ) (30.4) for some nonnegative function J : Ξ → [0 , ∞ ] , its raw rate function . If J is lower semicontinuous, it is just a...
View
Full
Document
This note was uploaded on 12/20/2011 for the course STAT 36754 taught by Professor Schalizi during the Spring '06 term at University of Michigan.
 Spring '06
 Schalizi
 Probability

Click to edit the document details