Scribd
Upload
248 views2 pages

APMA 1690: Homework 4

This document provides instructions for Homework 4 in APMA 1690, which is due on October 10th by 4:45 PM. Students should put their homework in the homework box in the Division of Applied Mathematics building and not email it or give it to the instructor or TAs. Late homework will be penalized according to policies on the course website. Students can discuss homework with others but must complete their own write-ups and code. The homework includes problems on approximating volumes using the central limit theorem, recursively approximating probability density functions, comparing distributions to the normal distribution, and demonstrating properties of distributions with infinite or nearly infinite variances.

Uploaded by

Jesse Cortez
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
248 views2 pages

APMA 1690: Homework 4

This document provides instructions for Homework 4 in APMA 1690, which is due on October 10th by 4:45 PM. Students should put their homework in the homework box in the Division of Applied Mathematics building and not email it or give it to the instructor or TAs. Late homework will be penalized according to policies on the course website. Students can discuss homework with others but must complete their own write-ups and code. The homework includes problems on approximating volumes using the central limit theorem, recursively approximating probability density functions, comparing distributions to the normal distribution, and demonstrating properties of distributions with infinite or nearly infinite variances.

Uploaded by

Jesse Cortez
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

APMA 1690: Homework 4

Homework 4 is due by 4:45 PM on Friday, October 10. Put your homework into the APMA 1690
homework box in the lobby of the Division of Applied Mathematics at 182 George Street. Do not email
or hand the homework to me or to the TAs, and do not put it into any of our mailboxes.
Homework will be collected shortly after 4:45. Concerning late homework, please read my Policies and
Grading statement on the courses Canvas page.
It is fine by me if you discuss the homework with other students, but the solutions and write-ups should be
done entirely on your own. This includes writing your own code.
Explain your reasoning. Label plots. Attach code at the end. And in general, do what you can to make the job
of the graders as easy as possible.
You are encouraged to use the discussion board and office hours. Please post typos on the discussion board,
even if they are obvious to you.
1. In HW 3, Problems 1c & 1d, you had to integrate a function over [0, 1]100 B(r) where B(r) is the ball of
radius r in R100 . If you used uniform samples for your Monte Carlo, then you should have found that this
worked pretty easily for r = 6 and not so easily for r = 4.
(a) Use the central limit theorem to approximate the volume of [0, 1]100 B(r) for r > 0. [Hint: Express
the volume as a probability related to iid U(0, 1).]
(b) If we try to solve HW 3 Problem 1c & 1d by using Monte Carlo samples from U(0, 1)100 , approximately
what fraction of our samples will land in the region of integration for HW 3 Problem 1c (i.e., [0, 1]100
B(r) for r = 6)? And approximately what fraction for HW 3 Problem 1d (i.e., r = 4)? [Use the CLT
from part (a), not Monte Carlo.]
(c) Estimate the fractions (from part (b) above) using at least 105 Monte Carlo samples. Report approximate
95% confidence intervals. How do your Monte Carlo approximations compare to your CLT approximations?
2. Let X1 , . . . , X20 be iid U(0, 1) and for each m = 1, . . . , 20 define
Sm =

m
X

Xk

k=1

and let fm be the pdf of Sm .


(a) Express fm as an integral over (0, 1) involving only fm1 .
R1
(b) If h(s) = 0 g(x, s)dx, then a simple Riemann approximation of h using 100 bins is
100

h(s)

1 X
g(i/100, s)
100
i=1

Use this approximation applied to your formula from part (a) to recursively approximate and then plot
f1 , . . . , f20 at the points 1/100, 2/100, . . . . [Use something like subplot(5,4,k) to put all 20 plots
onto the same page and save paper.]
(c) For each pdf f1 , . . . , f20 compare it to a normal pdf with the same mean and variance. Plot each normal
pdf on the same graph as its respective fm . Comment on the CLT in this case as n increases from 1 to
20.
1

(d) For the cases m = 2 and m = 20, verify the shape of your Riemman approximated fm by generating at
least 105 Monte Carlo samples of Sm (just sum m iid uniforms) and make a histogram of these samples
using at least 75 bins. (In Matlab, hist(data,75) will work. You can see more complicated uses of
hist in the solutions for HW 2.)
3. Pareto is a collection of continuous probability distributions with one pdf for every pair of > 0 and c > 0:
f (x) =

c
1x>c
x+1

When (0, 1], = and 2 is undefined; when (1, 2], =


c
= 1
and 2 =
plots demonstrating:

c2
(1)2 (2)

c
1

and 2 = ; and when > 2,

. Use iid samples from Pareto distributions with c = 1 and > 0 to make

(a) How the LLN behaves when the mean is infinite [you may find loglog plots useful].
(b) How the LLN works, but the CLT fails when the mean is finite, but the variance is infinite.
(c) How everything works, but can work very slowly, when the variance is finite, but nearly infinite (meaning that a tiny change in the shape of the distribution moves you from finite to infinite variance).
You will need to use a different for each of these cases. Explain and interpret your plots.

You might also like