next up previous
Next: Solution to Exercise 3.3.1. Up: No Title Previous: No Title

Solution to Exercise 3.1.1.

We will follow in part the proof of the Lindeberg CLT given in class. We first derive the characteristic function of a Bernoulli r.v. X with success probability tex2html_wrap_inline616 :

    eqnarray147

Now let tex2html_wrap_inline618 , tex2html_wrap_inline620 be independent Bernoulli random variables with success probability tex2html_wrap_inline622 , and put tex2html_wrap_inline624 = tex2html_wrap_inline628 . The characteristic function of tex2html_wrap_inline624 is

   eqnarray159

We wish to show that this converges to the characteristic function of the Poisson distribution with mean tex2html_wrap_inline632 . Let us derive what that characteristic function is. If T is Poisson with mean tex2html_wrap_inline632 , then

eqnarray169

Thus, our goal is to show

  equation185

Since tex2html_wrap_inline638 = tex2html_wrap_inline642 tex2html_wrap_inline644 tex2html_wrap_inline632 and exponentiation is a continuous function, it follows that

  equation192

Thus, it suffices to show

  equation197

Now

   eqnarray204

Now we would like to apply the Lemma from class that was used in the proof of the Lindeberg CLT, namely:

   lemma224

See Lemma 1 in section 27 of Billingsley. Now

  equation233

since it is the characteristic function of a Bernoulli r.v. (characteristic functions are always bounded by 1 in modulus). Similarly,

  equation238

since it is the characteristic function of a Poisson r.v. with mean tex2html_wrap_inline622 . Thus, we obtain from (10) and Lemma 1.1 without further difficulty that

  eqnarray246

Now with

  equation258

we see that the summands in (14) are simply the remainders in the first order Taylor expansion of tex2html_wrap_inline662 at tex2html_wrap_inline664 . It is not necessarily so easy to estimate what this remainder is since tex2html_wrap_inline666 is a general complex number, i.e. has nonzero real and imaginary parts. We know what the remainder is like for Taylor series expansions of a real variable, but not a general complex variable.

However, let us assume for now that the remainder has the same order, that is

  equation267

Thus, there exists an M such that tex2html_wrap_inline670 and a tex2html_wrap_inline672 such that

  equation271

We wish to apply this result to tex2html_wrap_inline674 = tex2html_wrap_inline678 . We are given in condition (ii) that tex2html_wrap_inline680 tex2html_wrap_inline644 0 as tex2html_wrap_inline686 . Since tex2html_wrap_inline688 and tex2html_wrap_inline690 , condition (ii) is the same as simply tex2html_wrap_inline692 tex2html_wrap_inline644 0. Thus,

  equation279

So there exists tex2html_wrap_inline698 such that for all tex2html_wrap_inline700 ,

  equation286

and hence for all tex2html_wrap_inline700 and all j, tex2html_wrap_inline620 ,

  equation290

Using this back in (14) we obtain

     eqnarray299

This proves (8), so to complete the problem we need to verify (16).

Now we turn to proving (16). This is a well known result in complex variables, but one can show it with a real variable argument applied to the real and imaginary parts. Denote a complex number z by z = tex2html_wrap_inline714 where tex2html_wrap_inline716 and tex2html_wrap_inline718 are both real. Now tex2html_wrap_inline662 = tex2html_wrap_inline724 = tex2html_wrap_inline728 = tex2html_wrap_inline732 . Further, we know

    eqnarray318

Since tex2html_wrap_inline734 , of course tex2html_wrap_inline736 can be replaced with tex2html_wrap_inline738 . Thus,

    eqnarray323

Here, equation (29) follows by just multiplying out the terms. Equation (30) follows since tex2html_wrap_inline740 = tex2html_wrap_inline738 as tex2html_wrap_inline746 , and

eqnarray330

Also, in(30) we used that tex2html_wrap_inline748 , so tex2html_wrap_inline750 .


next up previous
Next: Solution to Exercise 3.3.1. Up: No Title Previous: No Title

Dennis Cox
Mon Oct 13 13:28:22 CDT 1997