ECE 250: Stochastic Processes: Week #3 Outline: • Random Variables, Random Vector, and Random Processes • Almost Sure Limit of Random Processes • Distribution of Random Variables • Independence, Independent Processes, and Independent Increment Processes How to characterize a Random Variable • Question: How to show that a function X : Ω→ R is a random variable? • There are several ways: a. By definition, showing that for all B ∈ B, it suffices to show that X−1((−∞, a]) ∈ F for all a ∈ R. b. Practical way: Let g : Rn → R be a continuous mapping and let X1, . . . , Xn be r.v’s. Then X = g(X1, X2, . . . , Xn) is a r.v. This allows us to construct new random variables from the old ones: for example if X, Y are random variables, X+Y , X−Y , X × Y , XY , etc. are all random variables. Example 1. Let X, Y be r.v’s. Then E = {ω|X(ω) = Y (ω)} is an Event. Why? – Let Z = X − Y . – By property (c), Z is a random variable. – {0} is a Borel set (as {0} = ∩∞i=1(−1i , 1i )). – Therefore Z−1({0}) = {ω | Z(ω) = X(ω)− Y (ω) = 0} ∈ F and hence, E = Z−1({0}) is an Event! 1 Limits of Stochastic Processes • Motivation: Early stage of an epidemics dynamics: We have an initial infected popu- lation X0 and at each iteration (day) t ≥ 1, its getting multiplied by a positive random variable wt, i.e., Xt+1 = wtXt, where wt is an independently and identically dis- tributed random variables. Then, if E[log(wt)] > 0, we have limt→∞Xt = ∞ almost surely. • We say that b is an upper bound for a sequence {αk}, if αk ≤ b for all k. Smallest such b is called is the supremum of {αk} and is denoted by supk≥1 αk. We always assume that +∞ is an upper bound for a sequence and hence, supremum always exists. • Similarly, we say that b is a lower bound for a sequence {αk}, if αk ≥ b for all k. Largest such b is called is the infimum of {αk} and is denoted by infk≥1 αk. • We define: lim sup k→∞ αk = inf t≥1 sup k≥t αk lim inf k→∞ αk = sup t≥1 inf k≥t αk. • Note that for a sequence of r.v.s {Xk} and for an ω ∈ Ω, {Xi(ω)} is a sequence in R. Then I. X(ω) = supk≥1Xk(ω) is a random variable, II. X(ω) = infk≥1Xk(ω) is a random variable, III. X(ω) := lim supk→∞Xk(ω) is a random variable, IV. X(ω) := lim infk→∞Xk(ω)is a random variable, V. if X(ω) = X(ω) for almost all ω ∈ Ω, X defined by X = limk→∞Xk(ω) is a random variable (HW 3). 2 Distributions of Random Variables • For a r.v. X, we define the distribution function (or cumulative distribution function (CDF)) of X, to be the mapping FX : R→ [0, 1] defined by F (x) = Pr(X−1 ( (−∞, x])). • Properties of Distribution Functions (see HW 3): a. FX is non-decreasing. b. limx→−∞ FX(x) = 0, and limx→∞ FX(x) = 1. c. FX(·) is right-continuous, i.e., for any x ∈ R, limy→x+ FX(y) = FX(x). d. Define FX(x −) := limy↑x FX(y), then FX(x −) = Pr(X < x) = Pr({ω ∈ Ω | X(ω) < x}). e. For any x ∈ R, we have Pr(X = x) = FX(x)− FX(x−). • If we are given a distribution F , the first question one may ask is Does there exist a probability space (Ω,F ,Pr) and a function X : Ω → R such that X has the given distribution F? And the following theorem is the answer to this question. Theorem 1. Suppose that a function F : R → [0, 1] satisfies the above properties (a), (b) and (c), then there exists a probability space (Ω,F ,Pr) and a r.v. X such that F is the distribution function of X. 3 Expected Value of a Random Variable • How to define expected value? • Many of the constructions in probability theory starts from simple functions (r.v.s): we say that a random variable X is a simple r.v. if X = ∑m i=1 αi1Ai for some finite m ≥ 1, where A1, . . . , Am ∈ F and α1, . . . , αm ∈ R. • We define the expected value of a random variable using the following steps: – Expected value of simple r.v.s: For X = ∑m i=1 αi1Ai, we define E[X] := m∑ i=1 αi Pr(Ai). – For a positive random variable X (i.e., X ≥ 0 almost surely), we define: E[X] := sup{E[Y ] | Y ≤ X and Y is a simple function}. – Define positive and negative side of a random variable as: X+ = 1X≥0X and X− = −1X≤0X. Note that they are both non-negative r.v.s. – We say that the expected value of X exists if either E[X+] < ∞ or E[X−] < ∞ and we let it be E[X] := E[X+]− E[X−]. 4 Properties of Expected Value • If X ≥ 0, then E[X] ≥ 0. • monotonicity : if X ≤ Y , then E[X] ≤ E[Y ]. • E[|X|] = 0 if and only if X = 0 almost surely. • Markov Inequality: For a non-negative random variable X, Pr(X ≥ α) ≤ E[X] α for any α > 0. • Jensen’s Inequality: For a convex function Φ : R→ R, Φ(E[X]) ≤ E[Φ(X)]. Since −Φ is a concave function, the reverse inequality holds for concave functions. • Veryn important result: Monotone Convergence Theorem (MCT): Suppose that X1 ≤ X2 ≤ · · · ≤ X = limk→∞Xk. Then, lim k→∞ E[Xk] = E[ lim k→∞ Xk] = E[X]. 5 PDF, PMF, Continuous, and Discrete Random Variables • We say that X is a continuous r.v., if FX(x) is continuous. If further, FX(x) is differen- tiable, we refer to fX(x) := d dxFX(x) as the probability distribution function (pdf). • Very important: For a (continuous) r.v. X with the pdf fX(x), and any (sufficiently nice) g : R→ R E[X] = ∫ ∞ −∞ xfX(x)dx. • More generally (and important result), for any integrable function g, for the random variable Z = g(X), we have E[Z] = E[g(X)] = ∫ ∞ −∞ g(x)fX(x)dx. • We say that a random variable is a discrete random variable if Pr(X ∈ B) = 1 for a (finite or) countable set B = {bk | k ≥ 1}. • We define the probability mass function p : R → [0, 1] of a discrete random variable X to be defined by: pX(x) = { Pr(X = bk) if x = bk for some k ≥ 1 0 otherwise. • Very important: For a discrete r.v. X (see HW3) E[X] = ∞∑ k=1 bkpX(bk). 6 Independent Random Variables and Processes • Motivation: We have an initial infected population X0 and at each iteration (day) t ≥ 1, its getting multiplied by a positive random wt variable, i.e., Xt+1 = wtXt, where wt is an independently and identically distributed random variables. Then, if E[log(wt)] > 0, we have limt→∞Xt =∞ almost surely. • We say X, Y are two random variables, if X−1(B1) and Y −1(B2) are independent for any Borel sets B1, B2 ∈ B, i.e., Pr(X ∈ B and Y ∈ B) = Pr(X ∈ B) Pr(Y ∈ B). • Important Fact (lemma): X, Y are independent if X−1((−∞, α]) and Y −1((−∞, β]) are independent for all α, β ∈ R, i.e., it suffices to hold the above for sets of the form (−∞, α]. In other words, X, Y are independent if and only if Pr(X ≤ α, Y ≤ β) = FX(α)FY (β). • Similarly, we say that X1, . . . , Xn are independent if for any collection of Borel-sets B1, . . . , Bn, the events X −1 1 (B1), . . . , X −1 n (Bn) are independent. • Again it follows from a result1 that X1, . . . , Xn are independent iff for any selection of real numbers α1, . . . , αn: Pr(X1 ≤ α1, X2 ≤ α2, . . . , Xn ≤ αn) = FX1(α1) · · ·FXn(αn). 1If interested, look for Dynkin’s pi-λ Theorem. 7 Independent and Independent Increment Processes • We say that a DT or a CT random process {Xt} is 1. An independent process if any finite collection Xt1, . . . , Xtn are independent for any n ≥ 2. 2. An independent increment process, if for any n ≥ 2, and a1 < b1 ≤ a2 < b2 ≤ . . . ≤ an < bn (in the respective index set), the increments Xb1 −Xa1, Xb2 −Xa2, ..., Xbn −Xan are independent. 8
欢迎咨询51作业君