Limit Theorem Exam: Divisible Distributions, Markov Chains, and Conditional Probabilities, Exams of Probability and Statistics

The final exam questions for a limit theorems course. The questions cover topics such as infinitely divisible probability distributions, recurrent markov chains, and conditional probability. Students are required to prove theorems, analyze sequences of random variables, and understand the relationship between different subalgebras of a measure space.

Typology: Exams

2012/2013

Uploaded on 02/21/2013

raivath
raivath 🇮🇳

4.9

(9)

18 documents

1 / 2

Toggle sidebar

This page cannot be seen from the preview

Don't miss anything!

bg1
Limit Theorems
Final Exam
Due 10:30 a.m., December 16
1. Prove that an infinitely divisible probability distribution µon Ris
supported on the rational numbers QR(i.e., µ(Q) = 1) if and only if
some translation of µby a rational number has the form eλ(M) for some
λ > 0 and some probability measure Mwhich is supported on Q.
2. Let {Xi}be a sequence of i.i.d. bounded random variables taking values
in the integer grid Z2. Let Sn=Pn
i=1 Xi. Prove that the sequence Snis a
recurrent Markov chain if and only if E[X1] = 0. Is this still true if we
allow the Xito be unbounded?
3. Can you give an example of a sequence of probability measures µnon R
whose characteristic functions φnconverge point-wise (as ntends to
infinity) to the function 1Z, where Zis the set of integers? What if we
replace 1Zwith 1A, where Ais the set of integers whose absolute values are
perfect squares?
4. Let µbe a probability measure on a measure space (Ω,F). Suppose A,
B, and Care σsubalgebras of Fand that Qais a family of measures on
(Ω,F) (indexed by aΩ) that gives a regular conditional probability for µ
given A. That is, for all S F, the maps aQa(S) are Ameasurable;
Qx(A) = 1Afor all A A; and for all F-measurable functions f, it is
µ-almost everywhere the case that Eµ[f|A] = Rf dQa. Suppose further
that the following hold:
(a) Ais µ-independent of B(i.e., for all A A and B B,
µ(AB) = µ(A)µ(B))
(b) Ais µ-independent of C
(c) For µ-almost all a, the algebras Band Care Qa-independent of each
other.
Prove that Ais µ-independent of the σalgebra generated by B C. Is this
still true without requirement (c)?
5. Let be a (infinite dimensional with a countable basis) separable
Hilbert space and let Bbe the smallest field (not necessarily a σ-field)
containing all subsets of of the form {f: (f, g)A}where g and A
is a Borel subset of R. Let Fbe the σ-field generated by B.
1
pf2

Partial preview of the text

Download Limit Theorem Exam: Divisible Distributions, Markov Chains, and Conditional Probabilities and more Exams Probability and Statistics in PDF only on Docsity!

Limit Theorems Final Exam Due 10:30 a.m., December 16

  1. Prove that an infinitely divisible probability distribution μ on R is supported on the rational numbers Q ⊂ R (i.e., μ(Q) = 1) if and only if some translation of μ by a rational number has the form eλ(M ) for some λ > 0 and some probability measure M which is supported on Q.
  2. Let {Xi} be a sequence of i.i.d. bounded random variables taking values in the integer grid Z^2. Let Sn =

∑n i=1 Xi. Prove that the sequence^ Sn^ is a recurrent Markov chain if and only if E[X 1 ] = 0. Is this still true if we allow the Xi to be unbounded?

  1. Can you give an example of a sequence of probability measures μn on R whose characteristic functions φn converge point-wise (as n tends to infinity) to the function 1Z, where Z is the set of integers? What if we replace 1Z with 1A, where A is the set of integers whose absolute values are perfect squares?
  2. Let μ be a probability measure on a measure space (Ω, F). Suppose A, B, and C are σ subalgebras of F and that Qa is a family of measures on (Ω, F) (indexed by a ∈ Ω) that gives a regular conditional probability for μ given A. That is, for all S ∈ F, the maps a → Qa(S) are A measurable; Qx(A) = 1A for all A ∈ A; and for all F-measurable functions f , it is μ-almost everywhere the case that Eμ[f |A] =

f dQa. Suppose further that the following hold:

(a) A is μ-independent of B (i.e., for all A ∈ A and B ∈ B, μ(A ∩ B) = μ(A)μ(B))

(b) A is μ-independent of C

(c) For μ-almost all a, the algebras B and C are Qa-independent of each other.

Prove that A is μ-independent of the σ algebra generated by B ∪ C. Is this still true without requirement (c)?

  1. Let Ω be a (infinite dimensional with a countable basis) separable Hilbert space and let B be the smallest field (not necessarily a σ-field) containing all subsets of Ω of the form {f : (f, g) ∈ A} where g ∈ Ω and A is a Borel subset of R. Let F be the σ-field generated by B.

(a) Is it possible to construct a finitely additive (though not necessarily σ-additive) probability measure on (Ω, B) that is invariant under all Hilbert space automorphisms of Ω (besides the trivial measure which assigns measure 1 to the origin)? [Hint: try a Gaussian measure. Recall: a Hilbert space automorphism T is a linear bijection from the Hilbert space to itself that preserves the inner product, i.e., (f, g) = (T f, T g)).]

(b) Is it possible to construct a σ-additive probability measure on (Ω, F) that is invariant under all Hilbert space automorphisms of Ω (besides the trivial measure which assigns measure 1 to the origin)?