Design and Analysis of Algorithms May 23, 2012

Massachusetts Institute of Technology 6.046J/18.410J

Profs. Dana Moshkovitz and Bruce Tidor Final Exam

Final Exam
• Do not open this exam booklet until you are directed to do so. Read all the instructions first.
• The exam contains 6 problems, each with multiple parts. You have 180 minutes to earn 180
points.
• This exam booklet contains 17 pages, including this one, and two sheets of scratch paper
which can be detached.
′′
• This exam is closed book. You may use 3 double-sided Letter (8 12 × 11′′ ) or A4 crib sheets.
No calculators or programmable devices are permitted. Cell phones must be put away.
• Write your solutions in the space provided. If you run out of space, continue your answer on
the back of the same sheet and make a notation on the front of the sheet.
• Do not waste time deriving facts that we have studied. It is sufficient to cite known results.
• Do not spend too much time on any one problem. Generally, a problem’s point value is an
indication of how many minutes to spend on it.
• Show your work, as partial credit will be given. You will be graded not only on the correct-
ness of your answer, but also on the clarity with which you express it. Please be neat.
• Good luck!

Problem Title Points Parts Grade Initials

1 True or False 40 8

2 Short Answer 20 4

3 Estate Showing 30 3

4 Credit Card Optimization 30 3

5 Well-Separated Clusters 30 3

6 Looking for a Bacterium 30 2

Total 180

Name:
6.046J/18.410J Final Exam Name 2

Problem 1. True/False/Justify [40 points] (8 parts)

Circle T or F for each of the following statements to indicate whether the statement is true or false,
and briefly explain why. Your justification is worth more points than your true-or-false designation.

(a) T F If a problem has an algorithm that is correct for 2/3 fraction of the inputs, then it
also has an algorithm that is correct for 99.9% of the inputs.

(b) T F If a problem has a randomized algorithm that runs in time t and returns the correct
answer with probability at least 2/3, then the problem also has a deterministic
algorithm that runs in time t and always returns the correct answer.
6.046J/18.410J Final Exam Name 3

(c) T F A perfect hash table that is already built requires 2 hash functions, one for the
first level hash, and another for the second level hash.

(d) T F If Φ is a potential function associated with a data structure S, then 2Φ is also a

potential function that can be associated with S. Moreover, the amortized running
time of each operation with respect to 2Φ is at most twice the amortized running
time of the operation with respect to Φ.

(e) T F If we use van Emde Boas (vEB) to sort n elements, it is possible to achieve
O(n lg lg n) running time. Thus, whenever we need to use traditional O(n lg n)
sorting, we can replace it with vEB sort and achieve a better asymptotic running
time (ignore setup time).
6.046J/18.410J Final Exam Name 4

(f) T F The “union-by-rank” and “path-compression” heuristics do not improve the run-
ning time of M AKE -S ET in the union-find data structure.

(g) T F There is an NP-hard problem with a known polynomial time randomized algo-
rithm that returns the correct answer with probability 2/3 on all inputs.

(h) T F Every special case of an NP-hard problem is also NP-hard.

6.046J/18.410J Final Exam Name 5

Problem 2. Short Answer [20 points] (4 parts)

(a) Recall the forest-of-trees solution to the disjoint-set problem. Suppose that the only
heuristic we use is a variant of the union-by-rank heuristic: when merging two roots
u and v, we compare the number of descendants (rather than the rank), and make u a
child of v if and only if u has fewer descendants. Is this asymptotically worse than the
original union-by-rank heuristic? Explain why or why not.

(b) Suppose we apply Huffman coding to an alphabet of size 4, and the resulting tree is
a perfectly balanced binary tree (one root with two children, each of which has two
children of its own). Find the maximum frequency of any letter.
6.046J/18.410J Final Exam Name 6

(c) In lecture, we saw min-radius clustering, in which the goal was to pick a subset of k
points such that each point formed the center of a cluster of radius r. Suppose instead
that the center of the cluster can be a point not in the original set.
Give an example set of points where it is possible to find k clusters of radius r centered
around arbitrary points, but impossible to find k clusters of radius r centered around
points in the set.

(d) Consider the following algorithm, which is intended to compute the shortest distance
among a collection of points in the plane:

1 Sort all points by their y-coordinate.

2 for each point in the sorted list:
3 Compute the distance to the next 7 points in the list.
4 return the smallest distance found.
Give an example where this algorithm will return a distance that is not in fact the
overall shortest distance.
6.046J/18.410J Final Exam Name 7

Problem 3. Estate Showing. [30 points] (3 parts)

Trip Trillionaire is planning to give potential buyers private showings of his estate, which can
be modeled as a weighted, directed graph G containing locations V connected by one-way roads
E. To save time, he decides to do k of these showings at the same time, but because they were
supposed to be private, he doesn’t want any of his clients to see each other as they are being driven
through the estate.
Trip has k grand entrances to his estate, A = {a1 , a2 , . . . , ak } ⊂ V . He wants to take each of
his buyers on a path through G from their starting location ai to some ending location in B =
{b1 , b2 , . . . , bk } ⊂ V , where there are spectacular views of his private mountain range.
Because of your prowess with algorithms, he hires you to help him plan his buyers’ visits. His goal
is to find a path for each buyer i from the entrance they take, ai , to any ending location bj such that
no two paths share any edges, and no two buyers end in the same location bj .

(a) Trip tells you his idea: find all-pairs shortest paths on G, and then try to select k of
those shortest paths ai ❀ bj such that all k paths start and end at different vertices and
no two paths share any edges.
Give a graph where there exists a set of paths satisfying the requirements, but where
Trip’s strategy won’t find it.
6.046J/18.410J Final Exam Name 8

(b) Rather than using shortest paths, you think that perhaps you can formulate this as a
flow problem. Find an algorithm to find k paths ai ❀ bj that start and end at different
vertices and that share no edges, and briefly justify the correctness and running time
of your algorithm.
6.046J/18.410J Final Exam Name 9

(c) Trip, after trying out the paths found by your algorithm, realizes that making sure the k
paths don’t share edges isn’t enough: it’s possible that some paths will share vertices,
and his buyers might run into each other where their paths intersect.

a1 b1 a1 b1

a2 a2
b2 b2

a1 b1 and a2 b2 share a vertex a1 b1 and a2 b2 share neither vertices nor edges

Modify your algorithm to find k paths ai ❀ bj that start and end in different locations,
and that share neither vertices nor edges.
6.046J/18.410J Final Exam Name 10

Problem 4. Credit Card Optimization [30 points] (3 parts)

Zelda Zillionaire is planning on making a sequence of purchases with costs x1 , . . . , xn . Ideally,
she would like to make all of these purchases on one of her many credit cards. Each credit card has
credit limit ℓ. Zelda wants to minimize the number of credit cards that she needs to use to make
these purchases, without exceeding the credit limit on any card. More formally, she wants to know
the smallest k such that there exists π : {1, . . . , n} → {1, . . . , k} assigning each purchase i to a
credit card j, where π must satisfy the following constraint:
X
∀j ∈ {1, . . . , k} : xi ≤ ℓ.
i∈{1,...,n}
s.t. π(i)=j

Zelda is thinking of using the following algorithm to distribute her purchases:

1 Create an empty list of credit cards L.

2 for each purchase xi :
3 for each credit card j ≤ |L|:
4 if L[j] + xi ≤ ℓ:
5 Purchase xi with credit card j.
6 Set the balance for card j to L[j] = L[j] + xi .
7 Skip to the next purchase.
8 if no existing credit card has enough credit left:
9 Purchase xi with a new credit card.
10 Append a new credit card to L.
11 Set the balance of the new credit card to xi .
12 return k = |L|

(a) Give an example where Zelda’s algorithm will not use the optimal number of credit
cards.
(b) Show that Zelda’s algorithm gives a 2-approximation for the number of credit cards.
HINT: Let OP T be the optimal number of credit cards. In order for the algorithm to
add a new credit card j > OP T , it must reach xi such that xi + ′ min L[j ′ ] > ℓ.
j ∈{1,...,OP T }
It will then set L[j] = xi .
6.046J/18.410J Final Exam Name 11

(c) Show that minimizing the number of credit cards used is NP-hard.
6.046J/18.410J Final Exam Name 12

Problem 5. Well-Separated Clusters [30 points] (3 parts)

Let S = {(x1 , y1 ), . . . , (xn , yn )} be a set of n points in the plane. You have been told that there is
some k such that the points can be divided into k clusters with the following properties:

1. Each cluster has radius r.

2. The center of each cluster has distance at least 5r from the center of any other cluster.
3. Each cluster contains at least ǫn elements.

Your goal is to use this information to construct an algorithm for computing the number of clusters.

(a) Give an algorithm that takes as input r and ǫ, and outputs the exact value of k.
6.046J/18.410J Final Exam Name 13

(b) Given a particular cluster Ci , if you sample t points uniformly at random (with re-
placement), give an upper bound in terms of ǫ for the probability that none of these
points are in Ci .

(c) Give a sublinear-time algorithm that takes as input r and ǫ, and outputs a number of
clusters k that is correct with probability at least 2/3.
6.046J/18.410J Final Exam Name 14

Problem 6. Looking for a Bacterium [30 points] (2 parts)

Imagine you want to find a bacterium in a one dimensional region divided into n 1-µm regions.
We want to find in which 1-µm region the bacterium lives. A microscope of resolution r lets you
query regions 1 to r, r + 1 to 2r, etc. and tells you whether the bacteria is inside that region. Each
time we operate a microscope to query one region, we need to pay (n/r) dollars for electricity
(microscopes with more precise resolution take more electricity). We also need to pay an n-dollar
fixed cost for each type of microscope we decide to use (independent of r).

(a) Suppose that you decide to purchase ℓ = lg n microscopes, with resolutions r1 , . . . , rℓ

such that ri = 2i−1 . Give an algorithm for using these microscopes to find the bac-
terium, and analyze the combined cost of the queries and microscopes.
6.046J/18.410J Final Exam Name 15

(b) Give a set of microscopes and an algorithm to find the bacterium with total cost
O(n lg lg n).
SCRATCH PAPER
SCRATCH PAPER
MIT OpenCourseWare
http://ocw.mit.edu

6.046J / 18.410J Design and Analysis of Algorithms

Spring 2012

For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.