COMP 6651: Final Exam

Fall 2020

Submission through Moodle is due by December 11th at 22:00

Instructions:

• Solve the following problems, write up your solutions, digitize them into a single PDF
file, and upload it on Moodle by the deadline. No late submissions will be accepted.
Double-check your submission to make sure that no pages are missing.

• For any algorithms we have done in class, you do not need to provide the pseudocode.
However, if you are modifying the algorithm, you do need to provide the pseudocode.

• Do not spend too much time on any one problem. Read through all of them first and
attack them in the order that allows you to make the most progress.

• Answers must be brief and precise.

• The 20% rule: You will receive 20% of the points for any (sub)problem for which
you write “I do not know how to answer this question.” You will receive 10% if you
leave a question blank. If instead you submit irrelevant or erroneous answers you will
receive 0 points. You may receive partial credit for the work that is clearly “on the
right track.”
1. (Computational) [10 pts]
Consider the following LP in standard form. Rewrite it in slack form and solve it
using Simplex. For each step of Simplex write down the entering variable, the leaving
variable, and the basic feasible solution.

2
Maximize x
3 1
+ x2
Subject to x1 + x2 ≤8
1
x + x2
2 1
≤ 13
2
x1 ≤5
x2 ≤6
x1 , x2 ≥0

2. (Flow Networks) [10 pts]

You are given a list of courses c1 , c2 , . . . , cm and a student’s list of requirements r1 , . . . , rn .
Requirement ri = (ki , Si ) consists of a natural number ki and a subset of courses
Si ⊆ {c1 , . . . , cm }. The requirement states that the student must take exactly ki differ-
ent courses from the set Si (assume that |Si | ≥ ki ). Each course can be used to satisfy
at most one requirement. The student doesn’t even know whether all the requirements
can be satisfied or not. Help the student determine if all requirements can be satisfied,
and if so, help the student design a schedule of which courses to take so that all the
requirements are met.

(a) State this problem as a flow network G = (V, E), c, s, t. State precisely: what V
is, what E is, what capacities are assigned to edges e ∈ E.
(b) How can you use maximum flow value to decide if it is possible to satisfy all
requirements or not? How can you use maximum flow to design an actual schedule
provided that it is possible to satisfy all requirements?

Example: m = 4 and n = 2. Requirement r1 = (2, {c1 , c2 , c3 }) whereas requirement

r2 = (1, {c3 , c4 }). Then it is possible to satisfy all the requirements by taking course c3
to satisfy requirement r2 and taking courses c1 , c2 to satisfy requirement r1 (note that
since c3 is counted towards r2 , it can’t be counted towards r1 , since each course can be
used to satisfy at most one requirement).
End of Example

3. (Complexity Theory) [20 pts]

(a) For each statement below state whether it is TRUE or FALSE (exclusive). For
each answer provide a brief justification or a small counter-example. Guesses
without justification or with incorrect justification receive 0 marks.
• If L ∈ N P then L∗ ∈ N P.
• If L1 ≤p L2 and L2 ≤p L3 then L1 ≤p L3 .
 (b) Let G = (V, E) be a simple undirected graph that represents a friendship network,
i.e., vertices are people and there is an edge between two people if and only if they
are friends (we assume friendship is reciprocal). A subset of people S ⊆ V form
a tightly-knit community, TKC for short, if all people in S are friends with each
other. The TKC problem asks whether all people in a given friendship network G
can be split into k disjoint TKCs (friendships across different TKCs are allowed).
State the TKC problem as a language LT KC . Prove that LT KC is N P-complete.
Recall that COL = {hG, ki : G has a valid coloring with at most k colors} is
N P-complete.

Example: The following graph can be split into 2 TKCs (circled in the picture), but
not into 1 TKC.

f g
b
a c d e

End of Example

4. (Greedy) [20 pts]

In the Interval Set Cover problem you are given a set of intervals I1 , . . . , In on the real
line. Interval Ij = [sj , ej ) is a half-open interval. You are also given that together they
cover the grand interval [s, e) = nj=1 Ij . However, some intervals might be redundant.
S
We wish to find the smallest subset of intervals that cover the same S grand interval
[s, e). In other words, we are looking for S ⊆ {1, . . . , n} such that j∈S Ij = [s, e) and
|S| is as small as possible. Design an optimal and efficient greedy algorithm for this
task.

(a) Explain your algorithm in plain English.

(b) Describe your algorithm in pseudocode.
(c) Formally prove correctness of your algorithm.
(d) State and justify the runtime of your algorithm.

Example: Let I1 = [4, 10), I2 = [1, 6), I3 = [1, 5) and I4 = [7, 9). Then I1 ∪I2 ∪I3 ∪I4 =
[1, 10). Thus, the grand interval is [1, 10). This interval is covered by two input intervals
I1 ∪ I3 = [1, 10), but not by any single input interval. Thus S = {1, 3} is an optimal
solution.
End of Example
5. (Approximation Algorithms) [20 pts]
Let G = (V, E) be a simple undirected graph. A cycle of length 3 in G is called a
triangle. A subset of edges C ⊆ E is an edge-cover of triangles if every triangle in G
contains at least one edge from C. The Edge-Cover of Triangles problem asks to find
an edge-cover of triangles of minimum size, i.e., minimize |C|. Design a simple and
efficient 3-approximation algorithm for this problem. Hint: borrow inspiration from
the 2-approximation algorithm for Vertex Cover. In Vertex Cover you cover edges by
vertices, in this problem you cover triangles by edges.

(a) Describe your algorithm in plain English.

(b) Describe your algorithm in pseudocode.
(c) Prove that your algorithm achieves approximation ratio 3.

Example: In the following graph, C = {{a, b}, {b, c}, {a, c}} (shown in red) is an
edge-cover of all triangles, because each triangle contains at least one edge from C.

d b e

a c

f
End of Example

6. (Dynamic Programming) [20 pts]

You are designing a roller coaster for an amusement park. You have access to an
infinite supply of roller coaster segments of k different types. Each type is described
by two parameters (ei , mi ), where ei ∈ Z≥0 is how much one segment of this type adds
to excitement and mi ∈ Z is how much one segment of this type adds to (or subtracts
from, since we allow negative mi ) motion sickness. Your job is to decide which segments
to place one after another to form a roller coaster consisting of n segments in total. You
wish to maximize total excitement; however, you cannot just place one exciting segment
after another, as this will make many riders motion sick. At the start of the roller
coaster excitement is 0 and motion sickness is 0, and these two parameters accumulate
additively. The goal is to maximize overall excitement while keeping the accumulated
motion sickness between 0 and 100 (inclusive) during the entire ride. Design an efficient
dynamic programming algorithm for this. Hint: consider subproblems indexed by two
parameters i and m, where 0 ≤ i ≤ n and 0 ≤ m ≤ 100.

(a) State this problem formally in the usual format Input: . . ., Output: . . ..
(b) State the semantic array.
(c) State the computational array.
(d) Briefly justify why the above two arrays are equal.
(e) What running time will an algorithm based on the computational array have?
Why?

Example 1: k = 2, n = 10. The first segment type has e1 = 100, m1 = 50; the second
segment type has e2 = 0, m2 = −10.

Figure 1: This figure illustrates Example 1 of Question 6.

We create a roller coaster consisting of the first two segments of type 1. The excitement
after these two segments is 200, but motion sickness is 100. To reduce the motion
sickness, the next 5 segments are of type 2 each. Thus, the excitement after the first 7
segments is 200, but motion sickness is 50 (reduced due to segments of type 2). This
allows us to place another segment of type 1 as the 8th segment, and to complete the
roller coaster, we use segments of type 2 in the last two places. Thus, the overall
excitement is 300 and motion sickness is maintained between 0 and 100 during the
entire ride.
End of Example
Example 2: k = 2, n = 10. The first segment type has e1 = 100, m1 = 11; the second
segment type has e2 = 10, m2 = 1.
Then we create a roller coaster consisting of the first 9 segments of type 1 and the last
10th segment of type 2. Total excitement is 9e1 + e2 = 910. Motion sickness after the
first segment is m1 = 11, after the second segment is 2m1 = 22, and so on. After the
9th segment accumulated motion sickness is 9m1 = 99, so we cannot use type 1 for the
last segment, as it would push motion sickness beyond 100. Since we use segment of
type 2 in the last place, the motion sickness after the last segment will be 100, which
is within the tolerance level.
End of Example