关键词 > STAT433/833
STAT433/833 Assignment 0
发布时间:2023-09-27
Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit
ASSIGNMENT
STAT433/833 Assignment 0
1 | Warm-Up
Solve the following exercises from [Dur99]:
a) [1.8]
b) [1.10] (use a computer)
c) [1.56]
c) [1.58]
e) [1.65]
f) [1.72]
g) [1.73]
2 | Some properties of periodic Markov chains
Suppose P is the transition matrix of an irreducible Markov chain on a state space X with period d.
a) Show that there is a partition of X into X0 , X1 , ...Xd−1 disjoint and non-empty so that for each j it holds
that x ∈ Xj and px,y > 0 implies y ∈ Xj+1 mod d .
b) Show that for each j it holds that j = [Pd ]Xj ,Xj is the transition matrix for an irreducible Markov chain
on Xj .
c) Let π(ˆ)0 be a stationary distribution of 0 . Show that π(ˆ)j = π(ˆ)0 [Pj ]X0 ,Xj is stationary for
j for each
1 ≤ j ≤ d − 1.
d) Show that if X is finite then each dth root of unity is an eigenvalue of P. (Hint: construct the corresponding (right-)eigenvectors.
e) Show that if X is finite and if Q is an irreducible transition matrix on X with a (possibly complex) eigenvalue ω that satisfies |ω| = 1 and ω 1, then Q must be periodic. (Hint: if it was aperiodic, seek a contradiction).
3 | Random Walk on a Group
A group is a set 武 together with an operation ◦ : 武 × 武 → 武 such that
• for x,y, z ∈ 武 it holds that (x ◦ y) ◦ z = x ◦ (y ◦ z),
• there exists e ∈ 武 such that for all x ∈ 武 it holds that x ◦ e = e ◦ x = x,
• for each x ∈ 武 there exists an x−1 ∈ 武 such that x ◦ x−1 = x−1 ◦ x = e
For a distribution µ on countable group 武 the random walk on 武 with steps according to µ is the Markov chain with transition matrix given by px,y = µy◦x−1 .
a) Show that a random walk on a group is always doubly stochastic.
b) Show that if µ is symmetric (for all x it holds that µx = µx−1) then a random walk on a group 武 with steps according to µ is reversible.
c) Are random walks on groups always reversible? Prove or provide a counterexample.
4 | Random Walk on a Graph
An (undirected) graph is a set 武 , called vertices, together with a set of pairs of vertices 念 ⊆ {{i, j} : i, j ∈ 武 , i j}, called edges.
The degree of a vertex x ∈ 武 is deg(x) = |e ∈ 念 : x ∈ e| . A graph is d-regular if every vertex has degree d.
A random walk on a graph (武 , 念) is the Markov chain P with transition matrix px,y = I [g()(x(})念] .
a) Show that π ∝ (deg(x))x∈武 is stationary for a random walk on (武 , 念).
b) Show that a random walk on a graph satisfies detailed balance.
c) Show that a random walk on a d-regular graph is doubly stochastic.
d) Are random walks on grpahs always doubly stochastic? Prove or provide a counterexample.
5 | Coupling
Let 发 be countable and P be a Markov transition matrix on 发. Suppose that 发 is irreducible. Suppose there is an x⋆ in 发 and an α > 0 so that for all x ∈ 发 it holds that px,x⋆ ≥ α > 0. The goal of this problem is to show that there is a 0 < ρ < 1, a C > 0 and a distribution π such that for all x ∈ 发
Σ 'py(x) − πy ' < Cρt . (1)
y∈发
a) Show that ETx∗ < ∞. Conclude that the process is positive recurrent and has a stationary distribution. b) Construct a Markov transition matrix, , on 发2 such that if ((Xt, Yt))t≥0 is a Markov chain with that transition kernel then
• if X t and Yt are not both at x⋆ then X and Y each transition independently according to P , and
• if X t and Yt are both at x⋆ then they “couple”, meaning that they remain equal ever after, and transition
(with perfect dependence with each other) according to P.
It is sufficient to describe p(ˆ)(x1 ,y1 ) ,(x2 ,y2 ) for all x1 , x2 , y1 , y2 ∈ 发 .
c) Which states are transient for and which are recurrent?
d) Find a ρ ∈ (0, 1) and a C\ > 0 such that has P(x,y)(T(x*,x*) > t) ≤ C\ρt for all x, y ∈ 发 . e) Bound P(x,y)(Xt Yt).
f) Show that for any random elements A and B on a countable set 政 ,
Σ |P(A = a) − P(B = a)| ≤ 2P(A B). (2)
ae政
g) Show that there is a 0 < ρ < 1, a C > 0 and a distribution π such that for all x ∈ 发
Σ 'py(x) − πy ' < Cρt . (3)
ye发
6 | Discrete Phase-Type Distributions
Let 发 = 社 ∪或 with |社 | ∈ N∪{0}. Let P be a transition matrix on 发 so that all states in 社 are transient. Let (Xt)teNn术0贵 be a Markov chain with transition matrix P and initial distribution α . This question concerns properties of the distribution T或 = min {n ≥ 0 : Xt ∈ 或}.
Let Q = P社 ,社 , q = P社 ,或✶或 . Let d = |社 | .
a) Show that q = (I − Q)✶q .
b) Show that for all i ∈ 社 it holds that Σj qj < 1
c) Show that spec(I − Qd) ⊂ {z ∈ ❈ : |z − 1| < 1}. What does this tell us about spec(Q) and Σt≥0 Qt? (Hint: you can use the Gershgorin circle theorem)
d) Express P(T或 = k) in terms of α社 and Q. Your result should not depend on α或 , P社 ,或, or P或 ,或 .
e) Compute and simplify P(T或 ≤ k).
f) Compute and simplify E[T或].
g) Compute and simplify the moment generating function of T或. Express its domain in terms of spec(Q).
Since the distribution of T或 doesnt depend on the 或 part of the chain, only on the 社-part (per part d)), we denote it it as T或 ∼ DPHd(α社 , Q).
h) Show that if Ti ∼ DPHd i (αi, Qi) for i ∈ {1, 2} and T1 ⊥ T2 then T3 = T1 + T2 ∼ DPHd3 (α3 , Q3 ) for some values of (d3 , α3 , Q3 ). (Hint: d3 = d1 + d2 ).
i) Show that if Ti ∼ DPHd i (αi, Qi) for i ∈ {1, 2}, and I ∼ Ber(b) and ⊥ (T1, T2 , I) then T3 = IT1+(1−I)T2 ∼ DPHd3 (α3 , Q3 ) for some values of (d3 , α3 , Q3 ). (Hint: d3 = d1 + d2 ).
Required of graduate students only!!!
Pick one (1) of the following papers to read and then write a short (≈ 1 page) summary of, focusing on how it applies Markov chain methods.
1. L. Page et al. The pagerank citation ranking: Bring order to the web. Tech. rep. Technical report, stanford University, 1998 http://www.eecs.harvard.edu/~michaelm/CS222/pagerank.pdf
2. D. Aldous and P. Diaconis. “Shuffling cards and stopping times” . The American Mathematical Monthly 93.5 (1986), pp. 333–348 https://statweb.stanford.edu/~cgates/PERSI/papers/aldous86.pdf
3. L. R. Rabiner. “A tutorial on hidden markov models and selected applications in speech recognition” . Proceedings of the IEEE 77.2 (1989), pp. 257–286 https://ieeexplore.ieee.org/abstract/document/18626 4. C. E. Shannon. “A mathematical theory of communication”. The Bell system technical journal 27.3 (1948), pp. 379–423 https://people.math.harvard.edu/~ctm/home/text/others/shannon/entropy/entropy.pdf
References
[AD86] D. Aldous and P. Diaconis. “Shuffling cards and stopping times” . The American Mathematical Monthly 5 (1986) (cit. on p. 4).
[Dur99] R. Durrett. Essentials of stochastic processes . 1999. doi: https://doi.org/10.1007/978-3-319- 45614-0 (cit. on p. 1).
[Pag+98] L. Page, S. Brin, R. Motwani, and T. Winograd. The pagerank citation ranking: Bring order to the web. Tech. rep. Technical report, stanford University, 1998 (cit. on p. 4).
[Rab89] L. R. Rabiner. “A tutorial on hidden markov models and selected applications in speech recogni- tion”. Proceedings of the IEEE 2 (1989) (cit. on p. 4).
[Sha48] C. E. Shannon. “A mathematical theory of communication”. The Bell system technical journal 3 (1948) (cit. on p. 4).