elie*

2012 2 16

[email protected]

1
1.1 True or False?
True or false? Consider an instance of the Stable Matching Problem in which there exits
a man m and a woman w such that m is ranked first on the preference list of w and w is
ranked first on the preference list of m. Then in every stable matching S for this instance,
the pair (m, w) belongs to S.
S(m, w) S.
w m , (m, w ) S, (m , w) S. m, w
w , w, m m , S S
S,(m, w) S.

1.2 True or False?

True or false? In the stable matching S generated by G-S algorithm, each woman is
paired with her worst valid partner.
w1 worst valid partner m1 . G-S w1
m1 w1 m2 w1 m2 m1 m1 w1
wosrt valid patner S w1 m1 m2 w2 S m2
w1 m2 best valid patner. m2 w1 w2 , w1 m2 m1 .
w1 m2

1.3 True or False

Suppose we have two television stations, A and B. There are n prime-time programming
slots, and each station has n TV shows. Each station wants to assign each show to a distinct
slot so as to attract as much market share as possible. Here is the way we determine how
well the two stations perform relative to each other, given their schedules. Each show has a
fixed score. Well assume that no two shows have exactly the same score. A station wins a
given time slot if the show that it schedules for the time slot has a higher score than the show
the other station schedules for that time slot. The goal of each station is to win as many time
slots as possible.
Suppose Station A reveals a schedule S and Station B reveals a schedule T . Well say
that the pair of schedule (S, T ) is stable if neither station can unilaterally change its own
schedule and win more time slots. That is, there is no schedule S such that Station A wins
more slots with the pair (S , T ) than it did with the pair (S, T ); and symmetrically, there is
no schedule T such that Station B wins more slots with the pair (S, T ). Decide whether you

think the following statement is true or false. If it is true, give an algorithm. If it is false,
give a counterexample.
True or false? For every set of TV shows and scores, there is always a stable pair of
schedules.
n = 2 A a1 , a2 , B
b1 , b2 , a1 > b1 > a2 > b2 .
a1 , b1
a2 , b2 B B
b1 a2 b2 a1 B
a1 b2 a2 b1 A

1.4 Stable Matching

PSL is a shipping company that owns n ships and provide service to n ports. Each ship
has a schedule that says, for each day in some m(m > n) day period, whether the ship is at
sea or staying at one port. During the period, each ship visits each port once, and each visit
lasts exactly one day. For safety reasons, PSL has the following strict requirement:
() No two ships can be in the same port on the same day.
The company wants to perform maintenance on all the ships in the following m day
period. In this case, each ship Si will arrive in one scheduled port and simply remains there
for the rest of the period, ignoring the rest of its schedule. Now the question is: Given
the schedule for each ship, find when each ship should stop its travel so that condition ()
continues to hold. Try to give an algorithm to find the solution.

G-S

1.5 Complexity Analysis

Design a data structure to support the following two operations for a set S of integers:
Insert(S, x) inserts x into set S;
Delete-Larger-Half(S) deletes the largest S/2 elements from S.
Explain how to implement this data structure so that any sequence of m operations runs
in O(m) time.
S.
1. Insert

2. Delete-Larger-Half

9.3

S/2

O(n).

1.6 Complexity Analysis

In stock market, HH-index(historically highest) of the current price is k means that current price is the highest price in the previous k days, but not the highest one in the previous
k + 1 days.
For example, the price changes as showed in the following figure.
8+

HH

7
+

6
+

5
+

3
2

+
6

The price in Day 5 is the highest in Days 5, 4, 3, 2, but not highest one in Days 5, 4, 3, 2, 1,
thus HH(5) = 4;
The price in Day 4 is the highest in Days 4, 3, 2, but not highest one in Days 4, 3, 2, 1,
thus HH(4) = 3;
The price in Day 3 is the highest in Day 3, but not highest one in Days 3, 2, thus HH(3) =
1;
The input in this example is 8, 4, 3, 5, 6, 2, the answer is 1, 1, 1, 3, 4, 1.

Given the prices of n days, please give an algorithm of O(n) time complexity to calculate
the HH-index of all days.
( O(n))
while do
Q
S
k1 = |S|; k2 = |Q|;
while S = do
t = s. pop();
HH[t] = k;
while Q = and v[Q. top()] > v[t] do
tq = Q. pop();
HH[ta ] = 1;
end while
HH[t]+ = |Q| + 1;
end while
while Q = do
tq = Q. pop();
HH[tq ] = 1;
end while
end while

2 NP
2.1 NP-completeness
The subgraph-isomorphism problem takes two graphs G1 and G2 and asks whether
G1 is isomorphic to a subgraph of G2 . Show that the subgragh-isomorphism problem is
NP-complete.
Proof. First we prove that the Subgraph Isomorphism problem is in NP. The certificate is
(G1 = (V1 , E1 ), G2 = (V2 , E2 ), V1 V2 ). The verifying algorithm checks if is a oneto-one function, and for all u, v V1 whether (u, v) E1 if and only if ((u), (v)) E2 .
Secondly, we prove that CIQUE P Subgragh Isomorphism. Let (G = (V, E), k) be
an input instance for CLIQUE. Define G1 to be the complete graph on k vertices, and G2
to be the grapgh G. Then (G1 , G2 ) Subgragh Isomorphism if and only if (G, k)
CLIQUE.

2.2 NP-completeness
Given an integer m n matrix A and an integer m-vector b, the 0-1 Integer Programming problem asks whether there is an integer n-vector x with elements in the set {0, 1}
such that Ax b. Prove that 0-1 Integer Programming is NP-complete.
Proof. First we show that 0-1 Integer-Programming is in NP. The certificate is the nvector x with elements in the set 0, 1. Given the certificate, it can be verified in linear time
(hence, polynomial time) that it satisfies Ax b.
3SAT P 0-1 Integer-Programming
Goal:
Next we want to show that 3SAT is polynomially reducible to the 0-1 Integer-Programming
(IP). That is, we want a polynomial time computable function, which given an instance of
3SAT (a set of variables Y and clauses C) outputs an instance of IP (an m-by-n matrix A and
a m-vector b) such that the clauses C are satisfiable if and only if there is an integer n-vector
x with elements in the set {0, 1} which satisfies Ax b.
Transformation:
For each clause C we write out an inequality as follows. If the literal is variable yi , we replace it by the variable xi ; and if the literal is yi , we replace it by the quantity 1 xi . In what
follows, we will call this quantity that replaces the literal a term. Finally, the sum of these
terms is set 1. For example, if the 3SAT instance consists of two clauses y1 y2 y3 and
y1 y3 y4 , then the inequalities are x1 + (1 x2 ) + (1 x3 ) 1 and (1 x1 ) + x3 + x4 1.
One can obtain the matrix A and vector b from these inequalities quite easily. The entire
6

transformation can be done in linear time (hence polynomial time).

Correctness:
If the clauses are satisfiable then consider the satisfying truth assignment. For each variable
yi , if yi is true, then the corresponding variable xi is assigned value 1, and if yi is false, then
the corresponding variable xi is assigned value 0. Consider any clause c C. Clearly, the
satisfying truth assignment makes at least one literal in the clause c true. If this literal is the
variable yi then xi has value 1 and the inequality corresponding to clause c is satisfied; and
if this literal is yi then xi has value 0, and so 1 xi has value 1 and again the inequality
corresponding to clause c is satisfied. Since the argument applies to any clause c, it implies
that all the inequalities are satisfied by the above assignment. Conversely, if all the inequalities can be satisfied then consider the assignment of values to the variables xi that satisfies
the inequalities. For each variable xi , if xi is 1, then the corresponding variable yi is set to
true, and if xi is 0, then the corresponding variable yi is set to false. Consider any inequality
and the corresponding clause c C. Clearly, if the inequality is satisfied then one of the
three terms must be 1 (since all the terms are 0). If the term is of the form xi then
it implies that xi is 1 and the corresponding literal yi is true; and if the term is of the form
1 xi then it implies that xi is 0 and the corresponding literal yi is true. In either case, the
clause c is satisfied. Our argument applies to any arbitrary clause c, hence the given clauses
are satisfiable.

2.3 NP-completeness
The set-partition problem takes as input a set S of numbers. The question is whether
the numbers can be partitioned into two sets A and B and B = S A such that xA x =
xB x. Show that the set-partition problem is NP-complete.
Proof. First we show that Set-partition is in NP. The certificate for the Set-partition
problem consists of the two sublists S1 and S2 . Given the sublists, in polynomial time we
can compute the sums of the elements in these lists and verify that they are equal. Subset
Sum P set-partition
Goal:
Next we want to show that SubsetSum (SS) is polynomially reducible to the set-partition
problem (PART). That is, we want a polynomial time computable function f , which given
an instance of SS (a set of numbers S = {x1 , x2 , . . . , xn } and a target value t) outputs an
instance of PART (a set of numbers S = {x1 , x2 , . . . , xn }) such that S has a subset summing
to t if and only if S can be partitioned into subsets S1 and S2 that sum to the same value.
Transformation: Observe that the Set-partition problem is a special case of the subset sum
problem where we are trying to find a set of numbers S1 that sum to half the total sum of the
7

whole set. Let T be the sum of all the numbers in S.

T =

xi .

i=1

If t = T /2 then the subset sum problem is an instance of the partition problem, and we are
done. If not, then the reduction will create a new number, which if added to any subset that
sums to t, will now cause that set to sum to half the elements of the total. The problem is that
when we add this new element, we change the total as well, so this must be done carefully.
We may assume that t T /2, since otherwise the subset sum problem is equivalent to
searching for a subset of size T t, and then taking the complement. Create a new element
x0 = T 2t, and call partition on this modified set. Let S be S together with this new
element: S = S {x}. Clearly the transformation can be done in polynomial time.
Correctness: To see why this works, observe that the sum of elements in S is T + T 2t =
2(T t). If there is a solution to the subset sum problem, then by adding in the element
x0 we get a collection of elements that sums to t + (T 2t) = T t, but this is one half
of the total 2(T t), and hence is a solution to the Set-partition problem. Conversely,
if there is a solution to this Set-partition problem, then one of the halves of the partition
contains the element x0, and the remaining elements in this half of the partition must sum to
(T t) (T 2t) = t. Thus these elements (without x0 ) form a solution to the subset sum
problem.

2.4 NP-completeness
In the Half-3SAT problem, we are given a 3SAT formula with n variables and m
clauses, where m is even. We wish to determine whether there exists a truth assignment to
the variables of such that exactly half the clauses evaluate to 0 and exactly half the clauses
evaluate to 1. Prove that the Half-3SATproblem is NP-complete.
Proof. In order to prove that the Half-3SAT problem is NP-Complete, we need to prove
two things:
1. Half 3SAT NP,
2. L NP, L P Half-3SAT, or Half-3SAT NP-Hard
In order to prove that Half-3SAT is in NP, we need to show that a solution to a particular
instance of the problem can be verified in polynomial time. Consider the following witness algorithm which takes an instance of Half-3SAT and a certificate as parameters.
Our algorithm takes a boolean formula in conjunctive normal form, and the certificate or
proposed solution is a list of the boolean variables and their proposed truth assignments.
8

The algorithm walks the formula, evaluating each clause with the proposed truth assignments
and returns true if exactly half the clauses are true, and false if not. This is clearly polynomial
time because the boolean formula is only walked once.
In order to prove that Half-3SAT is NP-Hard, we can consider a reduction from a known
NP-Complete problem to Half-3SAT (or prove it directly, which we won
t even consider).
We choose to reduce from 3SAT to Half-3SAT. Because Circuit-SAT P SAT P 3SAT,
if we are able to show 3SAT P Half 3SAT, then this implies that Circuit-SAT P Half
3-SAT. And since L P Circuit-SAT, L NP, then this implies that L P Half 3SAT,
L NP or that Half 3SAT NP-hard.
Consider the following reduction. First, we need to convert an instance of 3SAT to Half
3SAT. Our approach is create a which contains 4 times as many clauses as . Suppose
contains m clauses. When creating , first we take all of the clauses from . Next, we
create m clauses of the form:
(p p q)
Clearly, these clauses are always true, regardless of the individual truth assignments for p
and q. Next, we create 2m clauses of the form:
(p q r)
These clauses are always true or always false. Therefore, we have created a boolean formula
which contains all of the clauses and m clauses which are always true and 2m clauses
which are either all true or all false. Clearly, this conversion takes polynomial time because
we have only added 3 variables, and 3m clauses.
We also need to show that there exists a yes instance of 3SAT if and only if there exists
a yes instance of Half 3SAT.
= that there exists a truth assignment which causes to be true. Then, the m clauses
which correspond to in are true and there are m clauses which are always true.
Thus, simply let p and r be false, and there exists a truth assignment which satisfies
Half 3SAT, where half the clauses are true and half are false.
= that there exists a truth assignment which causes Half 3SAT to be satisfied, or
a truth assignment that causes half the clauses in to be true and half false. But m
clauses in are always true, which means that the 2m clauses cannot be true if Half
3SAT is satisfied (because then 3m clauses would be true which is more than half).
Thus, the 2m clauses must be false, which means that is true, which means that
3SAT is also satisfied.
Thus, we have shown that a yes instance of 3SAT produces a yes instance of Half 3SAT,
and vice versa, which concludes our proof.
9

2.5 NP-completeness
Show that Undirected-Hamilton-Cycle problem is NP-complete.
Undirected-Hamilton-Cycle(decision problem):
Input: an undirected gragh G
Output: 1 if G has a Hamilton cycle which visits each vertex exactly once and also returns
to the starting vertex, 0 for others.
NP NP-Hard Hamilton
HP, Hamilton HC, Hamilton DHC, Hamilton
DHP.
HP P HC. , HP I: G u0 , v0 V (G),
HC f (I) : G = G + v0 u0 . G Hamilton P (u0 , v0 )
G Hamilton
DHC P DHP ,
DHP P HP. G va , vb V (G ), DHP
, G G
vi V (G ), G vi1 , vi2 , vi3 ,
vi1 vi2 , vi2 vi3 . vi vj E(G ), G vi3 vj1 .f (I) G va1 , vb3
G va1 vb3 Hamilton G
va vb Hamilton G va vb Hamilton
G va1
vb3 Hamilton P (va1 , vb3 ) G Hamilton
va1 P (va1 , vb3 ) P (va1 , vb3 ) vi1 vi2 vi3 2 vi ,
G va vb Hamilton
DHC P DHP P HP P HC, DHC NPC
HC NPC 2

2.6 NP-completeness
In the Half-Clique problem, we are given a graph G with n nodes, where n is even.
We wish to determine whether there exists a subgragh of G with n/2 nodes that is a clique.
Show that the Half-Clique problem is NP-complete.

1. (Half-Clique) NP
H G f , H
n/2, v1 , v2 V (H), (f (v1 ), f (v2 )) E(G).

2. CLIQUE P Half-Clique. G1 , k
k |G|/2
G j = 2k |G| G . G
10

 G k
k < |G|/2
G m = |G| 2k G
G . G 2|G| 2k G |G| k
G k + |G| 2k = |G| k

11

3
3.1 Divide and Conquer
You are interested in analyzing some hard-to-obtain data from two separate databases.
Each database contains n numerical values-so there are 2n values total-and you may assume
that no two values are the same. Youd like to determine the median of this set of 2n values,
which we will define here to be the nth smallest value.
However, the only way you can access these values is through queries to the databases.
In a single query, you can specify a value k to one of the two databases, and the chosen
database will return the k th smallest value that it contains. Since queries are expensive, you
would like to compute the median using as few queries as possible.
Give an algorithm that finds the median value using at most O(logn) queries.

k ak , k bk .
f (k) = ak bnk+1
. f (k) k f (1) 0 a1 bn
n bn . f (n) 0, n an . k

f (k) 0, f (k + 1) 0.

a1 a2 . . . ak bnk+1 ,
b1 b2 . . . bnk ak+1 .
ak+1 bnk+1
A = {a1 , a2 , . . . , ak , b1 , b2 , . . . , bnk } |A| = n.
n A max A = max{ak , bnk }.
k, n
FindNthSmallest(a[1..n], b[1..])
i = 1, j = n;
if a[1] b[n] then
return b[n];
end if
if a[n] b[1] then
return a[n];
end if
while j i > 1 do
12

k = j+i
;
2
temp = a[k];
t = temp b[n k + 1];
if t > 0 then
j = k;
else if t < 0 then
i = k;
else
return temp;
end if
end while
return max{a[k], b[n k]};

3.2 Divide and Conquer

Recall the problem of finding the number of inversions. As in the course, we are given
a sequence of n numbers a1 , . . . , an , which we assume are all distinct, and we difine an
inversion to be a pair i < j such that ai > aj .
We motivated the problem of counting inversions as a good measure of how different
two orderings are. However, one might feel that this measure is too sensitive. Lets call a
pair a significant inversion if i < j and ai > 2aj . Given an O(nlogn) algorithm to count the
number of significant inversions between two orderings.
Sort-and-Count (A)
1: Divide A into two sub-sequence L and R;
2: (RCL , L) = Sort-and-Count(L);
3: (RCR , R) = Sort-and-Count(R);
4: (r, A) = Merge-and-Count(L, R);
5: return (RC = RCL + RCR + r, A);

Merge-and-Count (L, R)
1: InverseCount = 0;
2: i = 1; j = 1;
3: for k = l to r do
4:
5:
6:
7:
8:

if L[i] > R[j] then

A[k] = L[j];
j + +;
else
A[k] = R[i];

13

9:
10:
11:
12:
13:
14:
15:
16:
17:
18:
19:
20:
21:

i + +;
end if
end for
i = 1, j = 1;
for k = Ltor do
if L[i] > 3 R[j] then
InverseCount+ = length(L) i + 1;
+ + j;
else
+ + i;
end if
end for
return InverseCount and A;

3.3 Divide and Conquer

Consider an n-mode complete binary tree T , where n = 2d 1 for some d. Each node
v of T is labeled with a real number xv . You may assume that the real numbers labeling the
nodes are all distinct. A node v of T is a local minimum if the label xv is less than the label
xw for all nodes w that are joined to v by an edge.
You are given such a complete binary tree T , but the labeling is only specified in the
following implicit way: for each node v, you can determine the value xv by probing the
node v. Show how to find a local minimum of T using only O(log n) probes to the nodes of
T.

FindLocalMin(T )
1: vl = probe(T.leftchild());
2: vr = probe(T.rightchild());
3: v = probe(T.root())
4: if v < v1 then
5:
6:
7:
8:
9:
10:
11:

if v < vr then
return T.root();
else
FindLocalMin(T.rightchild());
end if
else
if vl < vr then

14

12:
13:
14:
15:
16:

FindLocalMin(T.leftchild());
else
FindLocalMin(T.rightchild());
end if
end if

probe()

3.4 Divide and Conquer

Suppose now that youre given an n n grid graph G. (An n n grid graph is just the
adjacency graph of an n n chessboard. To be completely precise, it is a graph whose node
set is the set of all ordered pairs of natural numbers (i, j), where 1 i n and 1 j n;
the nodes (i, j) and (k, l) are joined by an edge if and only if |i k| + |j l| = 1.)
We use some of the terminology of the previous question. Again, each node v is labeled
by a real number xv ; you may assume that all these labels are distinct. Show how to find a
local minimum of G using only O(n) probes to the nodes of G. (Note that G has n2 nodes.)
i = 1; i = n; j = 1; j = n; i = (1 + n)/2; j = (1 + n)/2.

3.5 Divide and Conquer

Given a convex polygon with n vertices, we can divide it into several separated pieces,
such that every piece is a triangle. When n = 4, there are two different ways to divide the
polygon; When n = 5, there are five different ways.
Give an algorithm that decides how many ways we can divide a convex polygon with n
vertices into triangles.

Catalan

3.6 Divide and Conquer

Suppose you have n (n is even) coins, and you have a device to decide whether two coins
are made of the same material. (You cannot use the device to compare two piles of coins.)
15

Now you want to decide whether there is a set of more than n/2 of these coins that are made
of the same material.
(a) Give an algorithm using O(n log n) time to solve this problem.
(b) Give an algorithm using O(n) time and O(1) space to solve this problem.
sorry! I have an algorithm but fail to give a proof. So Im not sure whether the algorithm
is correct. Under this circumstance, I decide not to give the solution of this problem and am
hoping someone else will give a satisfying answer. Im looking forward for your answer.

16

4
4.1 Dynamic Programming
Professor Stewart is consulting for the president of a corporation that is planning a company party. The company has a hierarchical structure; that is, the supervisor relation forms
a tree rooted at the president. The personnel office has ranked each employee with a conviviality rating, which is a real number. In order to make the party fun for all attendees, the
president does not want both an employee and his or her immediate supervisor to attend.
Professor Stewart is given the tree that describes the structure of the corporation, using
the left-child, right-sibling representation. Each node of the tree holds, in addition to the
pointers, the name of an employee and that employees conviviality ranking. Describe an
algorithm to make up a guest list that maximizes the sum of the conviviality ratings of the
guests. Analyze the running time of your algorithm.

f( p),
f1 (m) = conv(m) +
f0 (m) = 0 +

ppost(m)

max{f1 (p), f0 (p)}

ppost(m)

conv(m) m Conviviality rating, post(m) m

max f0 (m0 ), f1 (m0 ) m0

O(n), f1 , f2 , f1
f0 f1 2n 1,
f0 n 1 n 1
O(n).

4.2 Dynamic Programming

Suppose you have one machine and a set of n jobs a1 , a2 ,...,an to process on that machine.
Each job aj has a processing time tj , a profit pj , and a deadline dj . The machine can process
only one job at a time, and job aj must run uninterruptedly for tj consecutive time units. If
job aj is completed by its deadline dj , you receive a profit pj , but if it is completed after its
deadline, you receive a profit of 0. Give an algorithm to find the schedule that obtains the
maximum amount of profit, assuming that all processing times are integers between 1 and
n. What is the running time of your algorithm?
Solution:
Recursion Relation: We first order the jobs by increasing deadline so a1 . . . , an are or17

dered by deadline. We then recurse on P (i, j), the maximum profit we can make using
jobs 1, 2, . . . , i in time j. The relation is:

0,
if i = 0,

P (i, j) = max(P (i 1, j), P (i 1, j ti ) + pi ), if j di ,

P (i 1, j),
else.
Running Time: Since all processing times are no more than n, we need only compute j
to n2 . Therefore, P is size O(n3 ). Filling in each square of P requires O(1) time for a total
running time of O(n3 ). Note that the O(n log n) sorting cost has been absorbed.

4.3 Dynamic Programming

To assess how well-connected two nodes in a directed graph are, one can not only look
at the length of the shortest path between them, but can also count th number of shortest
paths.
This turns out to be a problem that be solved efficiently, subject to some restrictions on
the edge costs. Suppose we are given a directed graph G = (V, E), with costs on the edges;
the costs may be positive or negative, but every cycle in the graph has strictly positive cost.
We are also given two nodes v, w V . Give an efficient algorithm that computes the
number of shortest v w paths in G.
Solution:
First we apply the Fullman-Fulkson algorithm to get a shortest path P and the length of the
path. We assume that the shortest path of the graph G is k. For every edge e of the path P ,
we get a graph Ge = G e. We apply the Fulkson-Fulkson algorithm on every of these
graphs and get the shortest paths length, denoted as ke . If there is a ke = k, for some e in
P , the shortest path is not uniqe, otherwise, it is uniqe.

4.4 Dynamic Programming

Youre running a computing system capable of processing several terabytes of data per
day. For each of n days, youre presented with xi terabytes on day i. Note that any unprocessed data is abandoned at the end of the day.
However, the computing system can only process a fixed number of terabytes in a given
day. Besides, the amount of data you can process goes down every day since the most recent
reboot of the system. On the first day after a reboot, you can process s1 terabytes, and so
on, up to sn ; we assume s1 > s2 > s3 > > sn > 0. To get the system back to peak
performance, you can choose to reboot it; but on any day you choose to reboot the system,
you cant process any data at all.
18

Now please give an efficient algorithm that takes values for x1 , x2 , . . . , xn and s1 , s2 , ..., sn
and returns the total number of terabytes processed by an optimal solution.

0 n + 1
f [i, j]
i 1 j + 1 i j
f [1, n] f [1, n]

min{s1 , xi },
i = j,

f [i, j] = 0,
j < i,

max
{f [i, k 1] + f [k + 1, j], g[i, j]}, else.
ikj

g[i, j] i 1 i j i
j
O(n3 ).

F (i, j) i
j n 1
F (n, 1);
F (0, j) = 0 j

F (i, j) = max{F (i 1, j + 1) + min{xni+1 , sj }, F (i 1, 1) + 0}

i xni+1 ,
sj .
F (0, j) = 0, j N+ , i F (i, j) j
n i + 1. F (n, 1)
O(1) O(n2 ) i j.
O(n2 ).

4.5 Dynamic Programming

Given a sequence of n real numbers a1 , ..., an , determine a subsequence (not necessarily contiguous) of maximum length in which the values in the subsequence form a strictly
increasing sequence.

ai Ci , k,
Ck = 1 +

max

i{i|0i<k,ai <ak }

19

Ci .

 a1 , a2 , . . . , an

O(n2 ).

4.6 Dynamic Programming

In an army system, to notify everyone of the postponement, the ranking officer first calls
each of his/her direct subordinates, one at a time. As soon as each subordinate gets the
phone call, he/she must notify each of his/her direct subordinates, one at a time. The process
continues this way until everyone has been notified. Note that each person in this process
can only call direct subordinates on the phone.
We can picture this process as being divided into rounds. In one round, each person who
has already learned of the postponement can call one of his/her direct subordinates on the
phone. The number of rounds it takes for everyone to be notified depends on the sequence in
which each person calls their direct subordinate. In the following example, it will take only
two rounds if A starts by calling B, but it will take three rounds if A starts by calling D.

Give an algorithm that determines the minimum number of rounds needed for everyone
to be notified, and outputs a sequence of phone calls that achieves this minimum number of
rounds.
r(a) a
r 0.
x r x
r x r
n1 n2 . . . , nk , r(x) = max1jk {nj + j}.

20

5
5.1 True or False
Decide whether you think the following statement is true or false. If it is true, give a
short explanation. If it is false, give a counterexample.
True or false? Let G be an arbitrary connected, undirected graph with a distinct cost c(e)
on every edge e. Suppose e is the cheapest edge in G; that is, c(e) < c(e) for every edge
e = e. Then there is a minimum spanning tree T of G that contains the edge e.
e T ,
c(T ) e e
A, B T A, B T

T e . A, B
T
c(T ) < c(T ). T
e .

5.2 Greedy Algorithm

Let us say that a graph G = (V, E) is a near-tree if it is connected and has at most n + 8
edges, where n = |V |. Give an algorithm with running time O(n) that takes a near-tree G
with costs on its edges, and returns a minimum spanning tree of G. You may assume that all
the edge costs are distinct.

G
G

8+1
n 1
O(n).

5.3 Greedy Algorithm

Given a list of n natural numbers d1 , d2 ,...,dn , show how to decide in polynomial time
whether there exits an undirected graph G = (V, E) whose node degrees are precisely the
numbers d1 , d2 ,...,dn . G should not contain multiple edges between the same pair of nodes,
or loop edges with both endpoints equal to the same node.

isGraphsDegrees(d[1..n])
n
1: p =
i=1 d[i];
2: if p > n(n 1)/2 or p is odd then
21

3:
4:
5:
6:
7:
8:
9:
10:
11:
12:
13:
14:
15:
16:
17:
18:
19:
20:
21:
22:
23:
24:
25:
26:

return FALSE;
end if
Sort d[1..n] in ascending order;
while n > 1 and d[n] = 0 do
k = n d[n];
if k < 1 then
return TRUE
end if
for i = k to n 1 do
d[i] = d[i 1];
if d[i] < 0 then
return FALSE;
end if
end for
n = n 1;
if k > 2 and d[k] < d[k 1] then
d = merge(d[1..k 1], d[k..n]);
end if
end while
if d[1] = 0 then
return FALSE;
else
return TRUE;
end if

G d1 , d2 , . . . , dn G ,
d1 , d2 , . . . , dk 1, dk+1 1, . . . , dn1 1, k n dn .
) G , vi

d ,
1 i k;
i

di =
d 1, k i n 1.
i

k = n dn . vn , vn vn vk , vn vk+1 , . . . , vn vn1 dn
d(vi ) = di , i = 1, 2, . . . , n.
) G d(vi ) = di , i = 1, 2, . . . , n. d1 < d2 < < dn . vn
vk , vk+1 , . . . , vn1 , {vk , vk+1 , . . . , vn1 }
vn G1 .
A = {v {vk , vk+1 , . . . , vn1 }|vvn }. A v, d(v) 1.
22

 v , vv , vn vv E(G1 ) vn v , v v .
A
G1 .
dG1 (vn ) = 0; |A| = E(G1 )
dG1 (v) = dG (v) 1, v {vk , vk+1 , . . . , vn1 };
dG1 (v) = dG (v), v {v1 , v2 , . . . , vk1 };
G = G1 vn . G

while

while d1 , d2 , . . . , dn , while
d1 , d2 , . . . , dk 1, dk+1 1, . . . , dn1 1. d1 , d2 , . . . , dn1 1
d1 , d2 , . . . , dn

while d[1] = 0

d[1] = 0 G = ({v}, )

1. ni=1 di n 1
2. d1 , d2 , . . . , dn O(n log n)
3. while n 1 k while for k 1
for

ni=1 di 2n(n 1)/2 = O(n2 ). ni=1 di n(n 1)

FALSE k while
n1
n k O( k=1
(n k)) = O(n2 ).
O(n2 ).
for n Kn
n(n 1)/2
(n2 ). (n2 ).

5.4 Greedy Algorithm

Suppose you have n video streams that need to be sent, one after another, over a communication link. Stream i consists of a total of bi bits that need to be sent, at a constant rate,
over a period of ti seconds. You cannot send two streams at the same time. Besides, there
cannot be any delays between the end of one stream and the start of the next. Suppose your
schedule starts at time 0. We assume that all the values bi and ti are positive integers.
Now, the link imposes the following constraint, using a fixed parameter r:
(*) For each natural number t > 0, the total number of bits you send over the time
interval from 0 to t cannot exceed rt.
23

Note that this constraint is only imposed for time intervals that start at 0, not for time
intervals that start at any other value.
Give an algorithm that takes a set of n streams, each specified by its number of bits bi
and its time duration ti , as well as the link parameter r, and determines whether there exists
a valid schedule. The running time of your algorithm should be polynomial in n.

bi /ti
i
b1 /t1 , b2 /t2 , . . . , bn /tn t0 = 0, Ti =
j=0 tj ; f (t) =
t
bi /ti , t [Ti1 , Ti ); F (t) = 0 f ()d. F (t) 0 t
f (t) F (t) t [0, Tn ]
Tn
n
f (t)dt
bi
i=0
F (t) n
t
t= 0
Tn
i=1 ti

ni=0 bi / ni=1 r t [0, Tn ] rt

n
n
b
/

i
i=0
i=1 r, rt

5.5 Greedy Algorithm

The input consists of n skiers with heights p1 , p2 ,...,pn , and n skies with height s1 ,
s2 ,...,sn . The problem is to assign each skier a ski to minimize the AVERAGE DIFFERENCE between the height of a skier and his/her assigned ski. That is, if the skier i is given
the ski ai , then you want to minimize:
ni=1 (|pi sai |)/n
Solution:
The key observation is that there is no advantage to cross matching reversing the height
if order of skiers and skis. That is if s1 < s2 and p1 < p2 there is no reason to match s1
with p2 and s2 with p1 .To show this, we have to look at all the possible relationships of the 4
heights and show that the cost if matching the shorter of the skiers with the shorter of the
skis is always at least as good as the cost of cross matching them. Without loss of generality
we can assume that the height of s1 is the smallest of the 4 heights. With this assumption,
we have 3 cases (i.e.: the 3 possible orderings of s2 , p1 , and p2 given that p1 < p2 ).
s1 < s 2 < p1 < p 2
|p1 s1 | + |p2 s2 | = p1 s1 + p2 s2 = |p1 s2 | + |p2 s1 |.
s1 < p 1 < s 2 < p 2
|p1 s1 | + |p2 s2 | = p1 s1 + p2 s2 < |p1 s2 | + |p2 s1 | = s2 p1 + p2 s1 .
s1 < p 1 < p 2 < s 2
|p1 s1 | + |p2 s2 | = p1 s1 + p2 s2 < |p1 s2 | + |p2 s1 | = s2 p1 + p2 s1 .
24

To show that there is always an optimal solution with no cross matching, let S be an
optimal matching. If S has no cross matches, we are done. Otherwise, consider two skiers
and two skies in a cross match, and reverse their matching so that the shorter of the two skiers
has the shorter of the two skis. Using the cases above one can show that this change cannot
increase the cost of the match, to the revised solution is at least as good as S.
Simply sort both the skiers and the skis by height, and match the i-th skier with the ith pair of skis. This algorithm is optimal because another assignment would have a cross
match, hence could not have a better cost. The running time of the algorithms is just the time
it takes to sort the two list, O(n log n).

5.6 Greedy Algorithm

The input to this problem consists of an ordered list of n words. The length of the ith
word is wi , that is the ith word takes up wi spaces. The goal is to break this ordered list of
words into lines, this is called a layout. Note that you can not reorder the words. The length
of a line is the sum of lengths of the words on that line. The ideal line length is L. No line
may be longer than L, although it may be shorter. The penalty of having a line of length K
is L K. The total penalty is the sum of the line penalties. The problem is to find the layout
that minimizes the total penalty.

(word):

G. G i i
i + 1
i + 1
i + 1 i + 1
i + 2

O(n),n

25

6
6.1 Prime and Dual
Suppose that we are given a linear program L in standard form, and suppose that for both
L and the dual of L, the basic solutions associated with the initial slack forms are feasible.
Show that the optimal objective value of L is 0.

max f = cT x,
s.t.Ax b;
I s
max f = cT x + 0s
( )
(
) x
s.t. A I
= b.
s
s = b, x = 0, f = cT x = 0.

min f = bT y,
s.t.AT y c;
y = 0 f = bT x = 0.
f = f f = 0.

6.2 True or False

Decide whether you think the following statement is true or false. If it is true, give a
short explanation. If it is false, give a counterexample.
True or false? Suppose that if we allow strict inequalities in a linear programming, and
the linear programming has optimal solution, then there still exists a vertex which takes the
optimal value.

x = (x1 , x2 , . . . , xn ) ai1 x1 + ai2 x2 + +

ain xn < bi ai1 x1 + ai2 x2 + +
ain xn < bi . ai1 x1 + ai2 x2 + + ain xn bi , x

x x LP
26


x LP
x x
x

6.3 Linear-inequality feasibility

Given a set of m linear inequalities on n variables x1 , x2 ,...,xn , the linear-inequality
feasibility problem asks if there is a setting of the variables that simultaneously satisfies
each of the inequalities.
a. Show that if we have an algorithm for linear programming, we can use it to solve the
linear-inequality feasibility problem. The number of variables and constraints that you use
in the linear grogramming problem should be polynomial in n and m.
b. Show that if we have an algorithm for the linear-inequality feasibility problem, we can use
it to solve a linear-programming problem. The number of variables and linear inequalities
that you use in the linear-inequality feasibility problem should be polynomial in n and m,
the number of variables and constraints in the linear programming.

(a)

(b) LPDLP

6.4 Linar Programming Modelling

Integer Linear Programming Problem is different from the classic Linear Programming Problem that some extra constraints such as
xi is an integer, for all i = 1, 2, . . . , n
are added.
A railway station has estimated that at least the following number of staff is needed in
each four-hour interval throughout a standard 24-hour period and the salary per hour for
every person during the different period:

27

Time Period

Staff Needed

5:009:00
9:0013:00
13:0017:00
17:0021:00
21:001:00
1:005:00

S1
S2
S3
S4
S5
S6

Time Period

Salary Per Hour For Every Person

0:008:00
8:0016:00
16:0024:00

C1
C2
C3

All staff works in 8-hour-shifts, which means every person will do continuous work for 8
hours. There are six possible shifts that start on the hour in the beginning of each 4-hour
period in the table. Now if the railway station want to minimize the total salary paid in one
day, please formulate this problem as an integer linear programming problem.
5:00 4 5:009:00 1
9:0013:00 2
i yi i xi
8 i xi2 6

yi = yi1 xi2 + xi Si , i = 1, 2, . . . , 6.

0:008:00

C1 , 0:001:00

y5
1:00-5:00
y6
5:00-8:00
y1

w1 = C1 y5 + 4C1 y6 + 3C1 y1 ;
8:0016:00
123

w2 = C2 (y1 + 4y2 + 3y3 );

16:0024:00
345

w3 = C3 (y3 + 4y4 + 3y5 ).

28

 w = w1 + w2 + w3 .

min w = w1 + w2 + w3 ,
yi = yi1 xi2 + xi Si , i = 1, 2, . . . , 6,
w2 = C1 (y5 + 4y6 + 3y1 ),
w2 = C2 (y1 + 4y2 + 3y3 ),
w3 = C3 (y3 + 4y4 + 3y5 ),
xi

yi .i = 1, 2, . . . , 6.

6.5 Linear Programming Modelling

0-1 Integer Programming Problem is different from the classic Linear Programming
Problem that some extra constraints such as
xi {0, 1}, for all i = 1, 2, . . . , n
are added. There is an undirected graph G with n nodes. We use a matrix M to denote
that whether two nodes are connected by an edge. In other words, mij is 1 if i and j are
adjacent, and 0 otherwise. If you want to paint the nodes with some colors, such that any
two adjacent nodes don
t share the same color, and you try to use as less colors as possible,
please formulate this problem as an integer linear programming problem (see the definition
in Problem 4) or a 0-1 linear programming problem.

Ck

1, k ,
Ck =
0, k .
0-1
min

Ck ,

i=1

m (x + xjk ) mij Ck ,

ij ik
s.t. Ck {0, 1},

xi,k {0, 1}, i, k = 1, 2, . . . , n.

6.6 Linear Programming Modelling

In a directed acyclic graph, you can find no path which start from some node and stop
at the same node. In this case, there are no Hamilton Cycle in this graph. In some practical
problems, we will modify the conditions a little so as to find some approximate solutions.
29

For example, the original target can be modified to finding the minimum number of paths so
that youre able to visit all nodes when you go through all these paths. The beginning and
end of the paths are not restricted. In the following instance, the minimum number of paths
is 2, and the paths are colored red.

Try to formulate this modified problem as a linear programming problem or an integer

linear programming problem (see the definition in Problem 4) or a 0-1 linear programming
problem (see the definition in Problem 5).
(The min cost flow
problem) 1

6.7 Linear Programming Modelling

FBI wants to send a very dangerous criminal from one prison to another. Just in case
that the criminal be rescued, FBI decided to send several cars simultaneously to mislead
his partners. The problem appears here: how to design the routes so as to arrange as many
misleading cars as possible. Suppose all relative streets here are oneway streets. In this
case, the problem can be transformed to finding as many paths as possible in a directed
graph, while all these paths cannot share any node or edge except the starting point s and
the finishing point t. In the following example, the maximum number of paths is 3, and the
available paths are colored red.

Try to formulate this problem as a linear programming problem or an integer linear programming problem (see the definition in Problem 4) or a 0-1 linear programming problem
(see the definition in Problem 5).
1

30

 7.1 1

31

7
7.1 Problem Reduction
In the Network Flow Problem, we add a group of constraints: for every node i, there has
a capacity limitation ki , the flow over the ith node cannot be greater than ki . Please reduce
the problem to the traditional Network Flow Problem.
P = (V, E, k, c, s, t). V E k : V
R+ , k(i) = ki , i V ; c : E R+ s V t V

V = {s, t, v1 , v2 , . . . , vm }, V = {s , t , v1 , v2 , . . . , vm
}, V = V V . (v, v )
E , v V . (u, v) E, (u , v) E
c : E R+

c((u, v)), e = (u , v);

c (e) =
k(u),
e = (u, u ).

P = (V , E , c , s, t)
P c v V
k(v). u, v v V , P
P

7.2

Optimal cover problem

If each vertex of a graph G carries a positive weight, then the weight of a set of S of
vertices is defined as the sum of weights of all the vertices in S and the optimal cover problem
requires finding a cover of the smallest weight. Show that this problem can be turned into
the minimum cut problem whenever G is bipartie.
G V1 , V1 G

w(V1 )
V (G) = X Y . X

Y G : V (G ) = V (G) {s, t}, uv E(G),

(u, v) E(G ); (s, v), (u, t), v X, u Y E(G ) G

+, uv E(G),

c(u, v) = w(v), u = s,

w(u), v = t.
G (S, T ) G
u S, v T , (u, v) E(G ). c(S, T ) = +.
32

({s}, V (G ) s)
G A V (G)

S = (X\A) {s} (Y A),

T = (X A) {t} (Y \A).
c(S, T ) = w(A). (S, T ) G
(S , T ) c(S , T ) < c(S, T ),
c(S , T ) =

c(s, v) +

uS

vT

c(s, v) +

vT X

c(u, t)

c(u, t)

uS Y

B = (T X) (S Y )
c(S , T ) = w(B)
B G e = uv E(G), u T X, e
u
/ T X u X. u X
(T X), v
/ S Y , (u, v) E(G ) u S , v T .
c(S , T ) = + c(S , T ) < c(S, T ) v S Y . e E(G). e B
c(S , T ) = w(B) < c(S, T ) = w(A), A G (S, T )
G

7.3 Problem Reduction

Let M be an n n matrix with each entry equal to either 0 or 1. Let mij denote the entry
in row i and column j. A diagonal entry is one of the form mii for some i.
Swapping rows i and j of the matrix M denotes the following action: we swap the values
mik and mjk for k = 1, 2, . . . , n. Swapping two columns is defined analogously.
We say that M is rearrangeable if it is possible to swap some of the pairs of rows and
some of the pairs of columns (in any sequence) so that, after all the swapping, all the diagonal
entries of M are equal to 1.
(a) Give an example of a matrix M that is not rearrangeable, but for which at least one
entry in each row and each column is equal to 1.
(b) Give a polynomial-time algorithm that determines whether a matrix M with 0-1 entries
is rearrangeable.
33


(a) The matrix

0 1 0

0 1 0
1 0 1

is not rerangeable.
(b)
V = {v1 , v2 , . . . , vn , v11 , v22 , . . . , vnn } {s, t}
E0 = ni=1 {(s, vi )},
E1 = ni=1 {(vi , vkk )|mik = 1},
E2 = ni=1 {(vii , t)},
E = E0 E1 E2 ,

1,
e E0 E2 ,
c(e) =
+, e E
1

P = (V, E, c, s, t). P f f (s, V ) = n,

M

7.4 Unique Cut

Let G = (V, E) be a directed graph, with source s V , sink t V , and nonnegative
edge capacities {ce }. Give a polynomial-time algorithm to decide whether G has a unique
minimum s t cut.
Ford-Fulkerson (S, T ). e
E(G) e = (u, v), u S, v T . (u, v) Ge
fGe (s, V ) > fG (s, V ),

7.5 Problem Reduction

A company has a collection of n software applications, {1, 2, . . . , n}, running on its
old system; and the president would like to port some of these to the new system. If he
moves application i to the new system, they expect a net (monetary) benefit of bi 0.
The different software applications interact with one another; if applications i and j have
extensive interaction, then the company will incur an expense if they move one of i or j to
the new system but not both; lets denote this expense by xij 0.
Due to small but fundamental incompatibilities between the two systems, theres no
way to port application 1 to the new system; it will have to remain on the old system. So,
34

this is the question: Which of the remaining applications, if any, should be moved? Give a
polynomial-time algorithm to find a set S {2, 3, . . . , n} for which the sum of the benefits
minus the expense of moving the applications in S to the new system is maximized.
Hoping others to give a solution...

7.6 Maximum Cohesiveness

In sociology, one often studies a graph G in which nodes represent people and edges
represent those who are friends with each other. Lets assume for purposes of this question
that friendship is symmetric, so we can consider an undirected graph.
Now suppose we want to study this graph G, looking for a close-knit group of people.
One way to formalize this notion would be as follows. For a subset S of nodes, let e(S)
denote the number of edges in S, that is, the number of edges that have both ends in S. We
define the cohesiveness of S as e(S)/|S|. A natural thing to search for would be a set S of
people achieving the maximum cohesiveness.
Give a polynomial-time algorithm that takes a rational number and determines whether
there exists a set S with cohesiveness at least .
Solution:
First we give the solution to the problem. Then we give intuition about how one might
discover the network that solves the problem. Given a graph G = (V, E) and a density ,
we wish to discover whether there is a subset W V such that e(W )/|W | , where
e(W ) is the number of edges with both endpoints in W . We create a network flow to solve
the problem as follows. Add a source node s and connect it to each vertex by an edge of
weight |E|, and add a sink node t and connect it to each vertex vi by an edge of weight
|E|di + 2, where di is the degree of vertex vi . Note that all edges have positive weight,
since di |E|.
Find the minimum weight cut in this graph, and let V1 be the set of vertices connected to
s and V2 be the set of vertices connected to t in this cut. Then if V1 = , there is no subset
of V of density greater than . Otherwise, V1 has density greater than .
We prove this as follows. We know that any cut in the graph must have weight greater
than or equal to the cut that separates V1 from V2 . Hence in particular the cut isolating s has
weight greater than the weight of the minimum cut. We let ci,j = 1 if edge (i, j) is in G.

35

Thus
|V ||E|

iV2

|E| +

(|E| dj + 2) +

jV1

|V ||E| + 2|V1 |

jV1

(dj

ci,j

iV2 ,jV1

ci,j )

iV2

0 2|V1 | 2e(V1 )
e(v1 )
.
|V1 |
On the other hand, if V1 = , we know that the inequality above is reversed, so we know
e(V1 )/|V 1| for any set V1 V .
Why should we have expected this network to solve the problem? The symmetries of
the problem imply that all nodes should be connected to the source and to the sink, and
that the only pieces of information that we can use to distinguish the weights of these links
must be the degrees of the nodes. We wish the procedure to isolate vertices that are highly
connected on one side, and ill connected on the other side. Hence by making the links to
the sink depend on the difference between 2 and the degree, we encourage nodes of high
degree to connect to the source, and nodes of low degree to connect to the sink. We add a
constant to the edges to the source and the edges to the sink to ensure all edges have positive
weight. And the connectivity of vertices inside the graph ensures that nodes of moderate
degree connected to high degree nodes also segregate into the part of the partition with their
high degree neighbors.

36

8
8.1 Approximation Algorithm
Consider the following algorithm for (unweighted) Vertex Cover: In each connected
component of the input graph execute a depth first search (DFS). Output the nodes that are
not the leaves of the DFS tree. Show that the output is indeed a vertex cover, and that it
approximates the minimum vertex cover within a factor of 2.
1 G
G G i Ci ,

|Ci |/|Ci | 2, Ci i ni=1 |Ci | 2 ni=1 |Ci |.

| ni=1 Ci |/|C | 2.
G e = (u, v)
E(G), DFS u, v u A v A A
A e A
|A| 2|C |.

634
APPROX-VERTEX-COVER(G)
1: C ;
2: E E(G);
3: while E = do
4:
5:
6:
7:
8:

let (u, v) be an arbitrary edge of E ;

C C {u, v};
remove from E every edge incident on u or v;
end while
return C;

C. |C|/|C | 2. 4 DFS
(u, v) E , A C, |A| |C|. |A|/|C | |C|/|C | 2.

8.2

Apptoximation Algorithm

Given a graph G = V, E with edge costs and set T V of terminal vertices, the Steiner
Tree Problem is to find a minimum cost tree in G containing every vertex in T (vertices in
V T may or may not be used in T ).
(a)Give a 2-approximation algorithm if the edge costs satisfy the triangle inequality.
(b)Give a 2-approximation algorithm for general edge costs (The graph also need not be
37

complete).
Solution: (a) We call the restricted (by the triangle inequality) problem metric Steiner tree
problem
Let R denote the set of required vertices. Clearly, a minimum spanning tree (MST) on R
is a feasible solution for this problem. We will prove that the cost of an MST on R is within
2 OPT.
Consider a Steiner tree of cost OPT. By doubling its edges we obtain an Eulerian graph
connecting all vertices of R and, possibly, some Steiner vertices. Find an Euler tour of this
graph, for example by traversing the vertices in DFS (depth first search) order:

The cost of this Euler tour is 2 OPT. Next obtain a Hmiltom cycle on the vertices of
R by traversing the Euler tour and short-cutting Steiner vertices and previously visited
vertices of R:

38

Because of triangle inequality, the shortcuts do not increase the cost of the tour. If we
delete one edge of this Hamiltonian cycle, we obtain a path that spans R and has cost at most
2 OPT. This path is else a spanning tree on R. Hence, the MST on R has cost at most
2 OPT.
(b)Now Ill prove that there is an approximation factor preserving reduction from the
Steiner tree problem to the metric Steiner tree problem.
We will transform, in polynomial time, an instance I of the Steiner tree problem, consisting of graph G = (V, E), to an instance I of the metric Steiner tree problem as follows.
Let G be the complete undirected graph on vertex set V . Define the cost of edge (u, v) in
G to be the cost of a shortest u v path in G. G is called the metric closure of G. The
partition of V into required and Steiner vertices in I is the same as in I.
For any edge (u, v) E, its cost in G is no more than its cost in G. Therefore, the cost
of an optimal solution in I dose not exceed the cost of an optimal solution in I.
Next, given a Steiner tree T in I , we will show how to obtain, in polynomial time, a
Steiner tree T in I of at most the same cost. The cost of an edge (u, v) in G corresponds
to the cost of a path in G. Replace each edge of T by the corresponding path to obtain a
subgraph of G. Clearly, in this subgraph, all the required vertices are connected. However,
this subgraph may, in general, contain cycles. If so, remove edges to obtain tree T This
completes the approximation factor preserving reduction.

8.3 Approximation Algorithm

Consider the following maximization version of the 3-Dimensional Matching Problem.
Given disjoint sets X, Y , Z, and given a set T X Y Z of ordered triples, a subset
M T is a 3 dimensional matching if each element of X Y Z is contained in at
most one of these triples. The M aximum 3 Dimensional M atching P roblem is to find
a 3-dimensional matching M of maximum size. (You may assume |X| = |Y | = |Z| if you
want.)
Give a polynomial-time algorithm that finds a 3-dimensional matching of size at least 31
times the maximum possible size.
Solution: The trivial algorithm below approximates the optimal solution within 1/3. (For
any triple in the found matching, consider it and all its neighbor triples. At most three of
these triples can be in the optimal matching.)
T = M ; M = ;
while T = do
t = any element in T ;

39

M = M {t};
T = T {tand all its neighbors in T };
end while;

8.4 Approximation Algorithm

Consider an optimization version of the Hitting Set Problem defined as follows. We are
given a set A = {a1 , a2 , ..., an } and a collection B1 , B2 , ..., Bm or subsets of A. Also, each
element ai A has a weight wi 0. The problem is to find a hitting set H A such that
the total weight of the elements in H, that is, ai H wi , is as small as possible. (H is a hitting
set if H Bi is not empty for each i.) Let b = maxi |Bi | denote the maximum size of any of
the sets B1 , B2 , . . . , Bm . Give a polynomial-time approximation algorithm for this problem
that finds a hitting set whose total weight is at most b times the minimum possible.
Solution:
The hitting set problem is equivalent to the set cover problem: An instance of set cover can be
viewed as an arbitrary bipartite graph, with sets represented by vertices on the left, elements
of the universe represented by vertices on the right, and edges representing the inclusion of
elements in sets. The task is then to find a minimum cardinality subset of left-vertices which
covers all of the right-vertices. In the hitting set problem, the objective is to cover the leftvertices using a minimum subset of the right vertices. Converting from one problem to the
other is therefore achieved by interchanging the two sets of vertices.
Since we have given a 2-approximate algorithm to solve the set cover problem (see Algorithms, T.Cormen et al.), we get a 2-approximate algorithm for the hitting set problem.

8.5 Approximation Algorithm

Suppose you are given a set of positive integers A = {a1 , a2 , . . . , an } and a positive
integer B. A subset S A is called feasible if the sum of the numbers in S does not exceed
B:

ai
ai S

You would like to select a feasible subset S of A whose total sum is as large as possible.
Example: If A = {8, 2, 4} and B = 11, then the optimal solution is the subset S = {8, 2}
Give a polynomial-time approximation algorithm for this problem with the following
guarantee: It returns a feasible set S A whose total sum is at least half as large as the
maximum total sum of any feasible set S A. Your algorithm should have a running tim
of most O(n log n).

40

 x A
B x B/2, B/2 x1 , x2 , . . . , xn ,
B/2

8.6 Approximation Algorithm

Bin Packing problem is as follows: Given n items with sizes a1 , . . . , an (0, 1], find
a packing in unit-sized bins that minimizes the number of bins used.
(a) Give a 2-approximation algorithm for this problem.
(b) Prove the following theorem:
For any > 0, there is no approximation algorithm having a guarantee of
the bin packing problem, assuming P = NP.

3
2

for

You can use any conclusion in Assignment 2.

(a)
Bin b H, Bin c. b H
for i=0 to n do
t = H
if ai H Bin b then
ai b ;
c(b) = c(b) ai ;
else
Bin t ai t
c(t) = 1 ai ;
t H
end if
end for
1/2

1/2,

ki=1 k/2. k 2 ni=1 ai 2 ni=1 ai 2B

1/2,
k1
k1
k
i=1 (1 vk ) + vk > (k 1) (k 2)vk > k/2.
i=1 vi + vk >
i=1 ai =
(b) Hoping others to give a solution....
41