Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Aalto University Department of Mathematics and Systems Analysis Exercise set 4 K Kytölä & A Karrila MS-E1602 Large Random Systems, 2017/IV Exercise session: Tue 14.3. at 12-14 Solutions due: Mon 20.3. at 10 Exercise 1. Let n ∈ N, and let C ∈ Rn×n be a symmetric1 and positive definite2, matrix and let m ∈ Rn be a vector. Define a function p : Rn → R by 1 1 > −1 p(x) = exp − (x − m) C (x − m) , Z 2 where Z is a constant. R (a) Calculate Rn p(x) dn x, and show that p is a (correctly normalized) probability p density on Rn if Z = (2π)n/2 det(C). Hint: Do first a change of variables (translation) to reduce to the case m = 0. Then do an orthogonal change of variables to a basis in which C is diagonal. (b) Choose Z as in part (a), and suppose that ξ = (ξ1 , ξ2 , . . . , ξn ) is a random 3 vector in Rn , which has probability p : Rn → R as above. i θ·ξdensity P Calculate the , for θ ∈ Rn where θ · ξ = nj=1 θj ξj denotes characteristic function ϕ(θ) = E e the inner product. (c) Let ξ be the random vector as in (b), and let a1 , . . . , an ∈ R. Show that the P linear combination nj=1 aj ξj is a random number with Gaussian distribution. Exercise 2. Suppose that B = (Bt )t≥0 is a standard Brownian motion and define (k) three other stochastic processes W (k) = (Wt )t≥0 , k = 1, 2, 3, by setting (1) = Bs+t − Bs (2) = λ−1/2 Bλt ( tB1/t , when t > 0 = 0 , when t = 0 Wt Wt (3) Wt where s ≥ 0 and λ > 0 are constants. Show that all these three stochastic processes (k) W (k) = (Wt )t≥0 , k = 1, 2, 3, are also standard Brownian motions. Hint: You may use results about existence and uniqueness of a standard Brownian motion. Exercise 3. Suppose that B = (Bt )t≥0 is a standard Brownian motion. Define a stochastic process X = (Xt )t∈[0,1] by setting Xt = Bt − tB1 . (a) Show that X is a Gaussian process. Calculate the mean function and the covariance function of X, i.e., t 7→ E[Xt ] and (s, t) 7→ Cov[Xs , Xt ] = E[Xs Xt ]−E[Xs ] E[Xt ]. (b) Show that X and B1 are independent. Hint: The distribution of X is determined by the finite dimensional marginals P[Xt1 ∈ A1 , . . . , Xtn ∈ An ], for 0 ≤ t1 < · · · < tn ≤ 1 and A1 , . . . , An ⊂ R Borel. Consider also using the Gaussianity of the processes (recall: Gaussians are independent iff they are uncorrelated). 1Symmetric means Cij = Cji for all i, j = 1, . . . , n. Positive definite means that for vector v ∈ Rn \ {0} we have v > Cv > 0. R any non-zero 3This means that P[ξ ∈ E] = n p(x) d x for all Borel sets E ⊂ Rn . E 2 Exercise 4. Suppose that X = (Xt )t∈[0,1] is a continuous and Gaussian process, for which E[Xt ] = 0 for all t ∈ [0, 1] and Cov[Xs , Xt ] = s(1 − t) for all 0 ≤ s ≤ t ≤ 1. Let Y ∼ N (0, 1) be a random variable independent of X. Define a stochastic process W = (Wt )t∈[0,1] by setting Wt = Xt + tY . (a) Show that W has the same finite dimensional distributions as a standard Brownian motion on the time interval [0, 1]. The event {W1 = y} has zero probability, but it is natural to define conditioning on this event by the following limiting procedure. For any ε > 0, the event {|W1 − y| < ε} has positive probability, so we can consider the conditional distribution of the process W given the event {|W1 − y| < ε}, and then take the (weak) limit of this conditional distribution as ε & 0. (b) Show that the conditional distribution of the process W given {W1 = y} is Gaussian, and calculate their mean and covariance functions. Note: Interpret the distribution of the process as the collection of its finite dimensional distributions, and conditioning as defined by the limiting procedure above. Exercise 5. Let B = (Bt )t∈[0,∞) be a standard Brownian motion. For t ≥ e define p λ(t) = t log(log(t)). √ |Bn | = 2, where the lim sup is taken (a) Show that almost surely lim supn→∞ λ(n) along n ∈ N. Hint: Observe that the restriction to Brownian motion to integer times, (Bn )n∈N , is a random walk with Gaussian steps. Recall the law of iterated logarithm proven earlier. (b) Find a function λ̃ : N → [0, ∞) with the following properties: λ̃(n) λ(n) → 0 as n → ∞, and almost surely maxs∈[n,n+1) |Bs − Bn | ≤ λ̃(n) except for finitely many values of n ∈ N. Hint: You may use the fact that P maxs∈[0,1] Bs > r = √2 2π R∞ r 1 2 e− 2 v dv for r ≥ 0. (c) Show that almost surely we have (with lim sup taken along t ∈ [0, ∞)) |Bt | √ lim sup = 2. t→∞ λ(t) (d) Show that almost surely √ |Bt | lim sup q = 2. t→0 t log | log(1/t)| Hint: Use (c) and the last part of Exercise 2 above. Exercise 6. Let S be a finite or countably infinite set. Define % : S N × S N → [0, ∞) by the formula4 X %(ω, ω 0 ) = 2−i for ω, ω 0 ∈ S N , ω = (ωi )i∈N , ω 0 = (ωi0 )i∈N . i∈N ωi 6=ωi0 (a) (b) (c) (d) 4The Show that % is a metric on the set S N . Show that the projections πi : S N → S defined by πi (ω) = ωi , are continuous. Show that (S N , %) is complete. Show that (S N , %) is separable. notation S N means the set of all functions N → S, i.e., of all sequences ω = (ωi )i∈N such that ωi ∈ S ∀i ∈ N.