Solution 3.

This file contains the exercises, hints, and solutions for Chapter 3 of the
book ”Introduction to the Design and Analysis of Algorithms,” 2nd edition, by
A. Levitin. The problems that might be challenging for at least some students
are marked by ; those that might be difficult for a majority of students are
marked by .

Exercises 3.1
1. a. Give an example of an algorithm that should not be considered an
application of the brute-force approach.

b. Give an example of a problem that cannot be solved by a brute-force
algorithm.
2. a. What is the efficiency of the brute-force algorithm for computing an
as a function of n? As a function of the number of bits in the binary
representation of n?

b. If you are to compute an mod m where a > 1 and n is a large positive
integer, how would you circumvent the problem of a very large magnitude
of an ?
3. For each of the algorithms in Problems 4, 5, and 6 of Exercises 2.3, tell
whether or not the algorithm is based on the brute-force approach.
4. a. Design a brute-force algorithm for computing the value of a polynomial

p(x) = an xn + an−1 xn−1 + ... + a1 x + a0

at a given point x0 and determine its worst-case efficiency class.

b. If the algorithm you designed is in Θ(n2 ), design a linear algorithm
for this problem.

c. Is it possible to design an algorithm with a better than linear efficiency
for this problem?
5. Sort the list E, X, A, M, P, L, E in alphabetical order by selection sort.
6. Is selection sort stable? (The definition of a stable sorting algorithm was
given in Section 1.3.)
7. Is it possible to implement selection sort for linked lists with the same
Θ(n2 ) efficiency as the array version?
8. Sort the list E, X, A, M, P, L, E in alphabetical order by bubble sort.
9. a. Prove that if bubble sort makes no exchanges on its pass through a list,
the list is sorted and the algorithm can be stopped.

1

b. Write a pseudocode of the method that incorporates this improve-
ment.

c. Prove that the worst-case eﬃciency of the improved version is quadratic.
10. Is bubble sort stable?
11. Alternating disks You have a row of 2n disks of two colors, n dark and
n light. They alternate: dark, light, dark, light, and so on. You want to
get all the dark disks to the right-hand end, and all the light disks to the
left-hand end. The only moves you are allowed to make are those which
interchange the positions of two neighboring disks.

Design an algorithm for solving this puzzle and determine the number of
moves it makes. [Gar99], p.75

2

Hints to Exercises 3.1
1. a. Think of algorithms that have impressed you with their efficiency
and/or sophistication. Neither characteristic is indicative of a brute-
force algorithm.

b. Surprisingly, it is not a very easy question to answer. Mathemati-
cal problems (including those you have studied in your secondary school
and college courses) are a good source of such examples.
2. a. The first question was all but answered in the section. Expressing
the answer as a function of the number of bits can be done by using the
formula relating the two metrics.

b. How can we compute (ab) mod m?
3. It helps to have done the exercises in question.
4. a. The most straightforward algorithm, which is based on substituting x0
into the formula, is quadratic.

b. Analyzing what unnecessary computations the quadratic algorithm
does should lead you to a better (linear) algorithm.

c. How many coefficients does a polynomial of degree n have? Can one
compute its value at an arbitrary point without processing all of them?
5. Just trace the algorithm on the input given. (It was done for another
input in the section.)
6. Although the majority of elementary sorting algorithms are stable, do not
rush with your answer. A general remark about stability made in Section
1.3, where the notion of stability is introduced, could be helpful, too.
7. Generally speaking, implementing an algorithm for a linked list poses prob-
lems if the algorithm requires accessing the list’s elements not in a sequen-
tial order.
8. Just trace the algorithm on the input given. (See an example in the
section.)
9. a. A list is sorted if and only if all its adjacent elements are in a correct
order. Why?

b. Add a boolean flag to register the presence or absence of switches.

c. Identify worst-case inputs first.
10. Can bubble sort change the order of two equal elements in its input?

3

11. Thinking about the puzzle as a sorting-like problem may and may not
lead you to the most simple and eﬃcient solution.

4

Solutions to Exercises 3.1
1. a. Euclid’s algorithm and the standard algorithm for finding the binary
representation of an integer are examples from the algorithms previously
mentioned in this book. There are, of course, many more examples in its
other chapters.

b. Solving nonlinear equations or computing definite integrals are ex-
amples of problems that cannot be solved exactly (except for special in-
stances) by any algorithm.

2. a. M (n) = n ≈ 2b where M(n) is the number of multiplications made by
the brute-force algorithm in computing an and b is the number of bits in
the n’s binary representation. Hence, the efficiency is linear as a function
of n and exponential as a function of b.

b. Perform all the multiplications modulo m, i.e.,

ai mod m = (ai−1 mod m · a mod m) mod m for i = 1, ..., n.

n 2
3. Problem 4 (computes 1 i ): yes

Problem 5 (computes the range of an array’s values): yes

Problem 6 (checks whether a matrix is symmetric): yes

4. a. Here is a pseudocode of the most straightforward version:

Algorithm BruteForcePolynomialEvaluation(P [0..n], x)
//The algorithm computes the value of polynomial P at a given point x
//by the “highest-to-lowest term” brute-force algorithm
//Input: Array P [0..n] of the coefficients of a polynomial of degree n,
// stored from the lowest to the highest and a number x
//Output: The value of the polynomial at the point x
p ← 0.0
for i ← n downto 0 do
power ← 1
for j ← 1 to i do
power ← power ∗ x
p ← p + P [i] ∗ power
return p

We will measure the input’s size by the polynomial’s degree n. The ba-
sic operation of this algorithm is a multiplication of two numbers; the
number of multiplications M (n) depends on the polynomial’s degree only.

5

Although it is not difficult to find the total number of multiplications in
this algorithm, we can count just the number of multiplications in the
algorithm’s inner-most loop to find the algorithm’s efficiency class:
n i n
n(n + 1)
M (n) = 1= i == ∈ Θ(n2 ).
i=0 j=1 i=0
2

b. The above algorithm is very inefficient: we recompute powers of x again
and again as if there were no relationship among them. Thus, the obvious
improvement is based on computing consecutive powers more efficiently.
If we proceed from the highest term to the lowest, we could compute
xi−1 by using xi but this would require a division and hence a special
treatment for x = 0. Alternatively, we can move from the lowest term to
the highest and compute xi by using xi−1 . Since the second alternative
uses multiplications instead of divisions and does not require any special
treatment for x = 0, it is both more efficient and cleaner. It leads to the
following algorithm:

Algorithm BetterBruteForcePolynomialEvaluation(P [0..n], x)
//The algorithm computes the value of polynomial P at a given point x
//by the “lowest-to-highest term” algorithm
//Input: Array P [0..n] of the coefficients of a polynomial of degree n,
// from the lowest to the highest, and a number x
//Output: The value of the polynomial at the point x
p ← P [0]; power ← 1
for i ← 1 to n do
power ← power ∗ x
p ← p + P [i] ∗ power
return p

The number of multiplications here is
n
M (n) = 2 = 2n
i=1

(while the number of additions is n), i.e., we have a linear algorithm.

Note: Horner’s Rule discussed in Section 6.5 needs only n multiplications
(and n additions) to solve this problem.

c. No, because any algorithm for evaluating an arbitrary polynomial of
degree n at an arbitrary point x must process all its n + 1 coefficients.
(Note that even when x = 1, p(x) = an + an−1 + ... + a1 + a0 , which needs
at least n additions to be computed correctly for arbitrary an , an−1 , ..., a0 .)

6

5.
E X A M P L E
A X E M P L E
A E X M P L E
A E E M P L X
A E E L P M X
A E E L M P X
A E E L M P X

6. Selection sort is not stable: In the process of exchanging elements that are
not adjacent to each other, the algorithm can reverse an ordering of equal
elements. The list 2 , 2 , 1 is such an example.

7. Yes. Both operations–finding the smallest element and swapping it–can
be done as efficiently with a linked list as with an array.

8. E, X, A, M, P , L, E
? ?
E ↔ X ↔ A M P L E
?
E A X ↔ M P L E
?
E A M X ↔ P L E
?
E A M P X ↔ L E
?
E A M P L X ↔ E
E A M P L E |X
?
E ↔ A M P L E
? ? ?
A E ↔ M ↔ P ↔ L E
?
A E M L P ↔ E
A E M L E |P
? ? ?
A ↔ E ↔ M ↔ L E
?
A E L M ↔ E
A E L E |M
? ? ?
A ↔ E ↔ L ↔ E
A E E |L
? ? ?
A ↔ E ↔ E ↔ L

The algorithm can be stopped here (see the next question).

9. a. Pass i (0 ≤ i ≤ n−2) of bubble sort can be represented by the following
diagram:
?
A0 , ..., Aj ↔ Aj+1 , ..., An−i−1 ≤| An−i ≤ ... ≤ An−1
in their final positions

7

If there are no swaps during this pass, then

A0 ≤ A1 ≤ ... ≤ Aj ≤ Aj+1 ≤ ... ≤ An−i−1 ,

with the larger (more accurately, not smaller) elements in positions n − i
through n − 1 being sorted during the previous iterations.

b. Here is a pseudocode for the improved version of bubble sort:

Algorithm BetterBubbleSort(A[0..n − 1])
//The algorithm sorts array A[0..n − 1] by improved bubble sort
//Input: An array A[0..n − 1] of orderable elements
//Output: Array A[0..n − 1] sorted in ascending order
count ← n − 1 //number of adjacent pairs to be compared
sflag ← true //swap flag
while sflag do
sflag ← false
for j ← 0 to count−1 do
if A[j + 1] < A[j]
swap A[j] and A[j + 1]
sflag ← true
count ← count −1

c. The worst-case inputs will be strictly decreasing arrays. For them, the
improved version will make the same comparisons as the original version,
which was shown in the text to be quadratic.

10. Bubble sort is stable. It follows from the fact that it swaps adjacent
elements only, provided A[j + 1] < A[j].

5. Here is a simple and efficient (in fact, optimal) algorithm for this problem:
Starting with the first and ending with the last light disk, swap it with
each of the i (1 ≤ i ≤ n) dark disks to the left of it. The ith iteration
of the algorithm can be illustrated by the following diagram, in which 1s
and 0s correspond to the dark and light disks, respectively.

00..011..11010..10 ⇒ 00..0011..1110..10
i−1 i−1 i i

n
The total number of swaps made is equal to i=1 i = n(n + 1)/2.

8

Exercises 3.2
1. Find the number of comparisons made by the sentinel version of sequential
search

a. in the worst case.

b. in the average case if the probability of a successful search is p (0 ≤
p ≤ 1).
2. As shown in Section 2.1, the average number of key comparisons made by
sequential search (without a sentinel, under standard assumptions about
its inputs) is given by the formula

p(n + 1)
Cavg (n) = + n(1 − p),
2

where p is the probability of a successful search. Determine, for a fixed n,
the values of p (0 ≤ p ≤ 1) for which this formula yields the largest value
of Cavg (n) and the smallest value of Cavg (n).
3. Gadgets testing A firm wants to determine the highest floor of its n-
story headquarters from which a gadget can fall with no impact on the
gadget’s functionality. The firm has two identical gadgets to experiment
with. Design an algorithm in the best efficiency class you can to solve
this problem.
4. Determine the number of character comparisons made by the brute-force
algorithm in searching for the pattern GANDHI in the text

THERE_IS_MORE_TO_LIFE_THAN_INCREASING_ITS_SPEED
(Assume that the length of the text–it is 47 characters long–is known
before the search starts.)
5. How many comparisons (both successful and unsuccessful) are made by the
brute-force string-matching algorithm in searching for each of the following
patterns in the binary text of 1000 zeros?
a. 00001 b. 10000 c. 01010
6. Give an example of a text of length n and a pattern of length m that
constitutes the worst-case input for the brute-force string-matching al-
gorithm. Exactly how many character comparisons are made for such
input?
7. Write a visualization program for the brute-force string-matching algo-
rithm.

9

8. In solving the string-matching problem, would there be any advantage in
comparing pattern and text characters right-to-left instead of left-to-right?
9. Consider the problem of counting, in a given text, the number of substrings
that start with an A and end with a B. (For example, there are four such
substrings in CABAAXBYA.)

(a) Design a brute-force algorithm for this problem and determine its
efficiency class.
(b) Design a more efficient algorithm for this problem [Gin04].

10. Word Find A popular diversion in the United States, Word Find, asks
the player to find each of a given set of words in a square table filled with
single letters. A word can read horizontally (left or right), vertically (up
or down), or along a 45 degree diagonal (in any of the four directions),
formed by consecutively adjacent cells of the table; it may wrap around
the table’s boundaries but it must read in the same direction with no
zigzagging. The same cell of the table may be used in different words,
but, in a given word, the same cell may be used no more than once. Write
a computer program for solving this puzzle.
11. Battleship game Write a program for playing Battleship (a classic strat-
egy game) on the computer which is based on a version of brute-force
pattern matching. The rules of the game are as follows. There are two
opponents in the game (in this case, a human player and the computer).
The game is played on two identical boards (10-by-10 tables of squares)
on which each opponent places his or her ships, not seen by the opponent.
Each player has five ships, each of which occupies a certain number of
squares on the board: a destroyer (2 squares), a submarine (3 squares),
a cruiser (3 squares), a battleship (4 squares), and an aircraft carrier (5
squares). Each ship is placed either horizontally or vertically, with no two
ships touching each other. The game is played by the opponents taking
turns “shooting” at each other’s ships. A result of every shot is displayed
as either a hit or a miss. In case of a hit, the player gets to go again and
keeps playing until this player misses. The goal is to sink all the oppo-
nent’s ships before the opponent succeeds in doing it first. (To sink a ship,
all squares occupied by the ship must be hit.)

10

1. Modify the analysis of the algorithm’s version in Section 2.1.
2. As a function of p, what kind of function is Cavg ?
3. Solve a simpler problem with a single gadget first. Then design a better
than linear algorithm for the problem with two gadgets.
4. The content of this quote from Mahatma Gandhi is more thought provok-
ing than this drill.
5. For each input, one iteration of the algorithm yields all the information
you need to answer the question.
6. It will suffice to limit your search for an example to binary texts and
patterns.
7. You may use either bit strings or a natural language text for the visual-
ization program. It would be a good idea to implement, as an option, a
search for all occurrences of a given pattern in a given text.
8. The answer, surprisingly, is yes.
9. a. For a given occurrence of A in the text, what are the substrings you
need to count?

b. For a given occurrence of B in the text, what are the substrings you
need to count?
10. Test your program thoroughly. Be especially careful about a possibility
of words read diagonally with wrapping around the table’s border.
11. A (very) brute-force algorithm can simply shoot at adjacent feasible cells
starting at, say, one of the corners of the board. Can you suggest a better
strategy? (You can investigate relative efficiencies of different strategies
by making two programs implementing them play each other.) Is your
strategy better than the one that shoots at randomly generated cells of
the opponent’s board?

11

1. a. Cworst (n) = n + 1.

b. Cavg (n) = (2−p)(n+1) . In the manner almost identical to the analysis
2
in Section 2.1, we obtain
p p p p
Cavg (n) = [1 · + 2 · + ... + i · + ... + n · ] + (n + 1) · (1 − p)
n n n n
p
= [1 + 2 + ... + i + ... + n] + (n + 1)(1 − p)
n
p n(n + 1) (2 − p)(n + 1)
= + (n + 1)(1 − p) = .
n 2 2

2. The expression
p(n+1)
2 + n(1 − p) = p n+1 + n − np = n − p(n −
2
n+1
2 ) =n− n−1
2 p

is a linear function of p. Since the p’s coefficient is negative for n > 1,
the function is strictly decreasing on the interval 0 ≤ p ≤ 1 from n to
(n + 1)/2. Hence p = 0 and p = 1 are its maximum and minimum points,
respectively, on this interval. (Of course, this is the answer we should
expect: The average number of comparisons should be the largest when
the probability of a successful search is 0, and it should be the smallest
when the probability of a successful search is 1.)

√ √
3. Drop the first gadget from floors
√ n , 2 n , and so on until either the
floor i n a drop from which makes the gadget malfunction is reached
or no such floor in this sequence is encountered before the top of the
building is reached. In the former case, the floor to be found is higher
√ √
than (i − 1) √ and lower than√ n . So, drop the second gadget from
n i
floors (i − 1) n + 1, (i − 1) n + 2, and so on until the first floor
a drop from which makes the gadget malfunction is reached. The floor
immediately preceeding that floor is the floor in question. If no drop in the
first-pass sequence resulted in the gadget’s failure, the floor in question is
√
higher than i n , the last tried floor of that sequence. Hence, continue
√ √
the successive examination of floors i n + 1, i n + 2, and so on until
either a failure is registered or the last floor is reached. The number of
√ √
times the two gadgets are dropped doesn’t exceed
√ n + n , which
puts it in O( n).
4. 43 comparisons.
The algorithm will make 47 − 6 + 1 = 42 trials: In the first one, the G
of the pattern will be aligned against the first T of the text; in the last
one, it will be aligned against the last space. On each but one trial, the
algorithm will make one unsuccessful comparison; on one trial–when the
G of the pattern is aligned against the G of the text –it will make two

12

comparisons. Thus, the total number of character comparisons will be
41 · 1 + 1 · 2 = 43.

5. a. For the pattern 00001, the algorithm will make four successful and one
unsuccessful comparison on each of its trials and then shift the pattern
one position to the right:
0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 1
0 0 0 0 1
etc.
0 0 0 0 1
The total number of character comparisons will be C = 5 · 996 = 4980.

b. For the pattern 10000, the algorithm will make one unsuccessful com-
parison on each of its trials and then shift the pattern one position to the
right:
0 0 0 0 0 0 0 0 0 0 0
1 0 0 0 0
1 0 0 0 0
etc.
1 0 0 0 0
The total number of character comparisons will be C = 1 · 996 = 996.

c. For the pattern 01010, the algorithm will make one successful and
one unsuccessful comparison on each of its trials and then shift the pat-
tern one position to the right:
0 0 0 0 0 0 0 0 0 0 0
0 1 0 1 0
0 1 0 1 0
etc.
0 1 0 1 0
The total number of character comparisons will be C = 2 · 996 = 1, 992.

6. The text composed of n zeros and the pattern 0 ... 01 is an example of
m−1
the worst-case input. The algorithm will make m(n − m + 1) character
comparisons on such input.

7. n/a
8. Comparing pairs of the pattern and text characters righ-to-left can allow
farther pattern shifts after a mismatch. This is the main insight the two
string matching algoirthms discussed in Section 7.2 are based on. (As a
speciﬁc example, consider searching for the pattern 11111 in the text of
one thousand zeros.)

13

9. a. Note that the number of desired substrings that starts with an A at a
given position i (0 ≤ i < n − 1) in the text is equal to the number of B’s
to the right of that position. This leads to the follwing simple algorithm:

Initialize the count of the desired substrings to 0. Scan the text left to
right doing the following for every character except the last one: If an A
is encountered, count the number of all the B’s following it and add this
number to the count of desired substrings. After the scan ends, return
the last value of the count.

For the worst case of the text composed of n A’s, the total number of
character comparisons is

n + (n − 1) + ... + 2 = n(n + 1)/2 − 1 ∈ Θ(n2 ).

b. Note that the number of desired substrings that ends with a B at
a given position i (0 < i ≤ n − 1) in the text is equal to the number of A’s
to the left of that position. This leads to the follwing algorithm:

Initialize the count of the desired substrings and the count of A’s en-
countered to 0. Scan the text left to right until the text is exhausted and
do the following. If an A is encountered, increment the A’s count; if a
B is encountered, add the current value of the A’s count to the desired
substring count. After the text is exhausted, return the last value of the
desired substring count.

Since the algorithm makes a single pass through a given text spending
constant time on each of its characters, the algorihtm is linear.
10. n/a
11. n/a

14

Exercises 3.3
1. Can you design a more efficient algorithm than the one based on the brute-
force strategy to solve the closest-pair problem for n points x1 , ..., xn on
the real line?
2. Let x1 < x2 < ... < xn be real numbers representing coordinates of n
villages located along a straight road. A post office needs to be built in
one of these villages.

a. Design an efficient algorithm to find the post-office location minimizing
the average distance between the villages and the post office.

b. Design an efficient algorithm to find the post-office location minimizing
the maximum distance from a village to the post office.
3. a. There are several alternative ways to define a distance between two
points P1 = (x1 , y1 ) and P2 = (x2 , y2 ). In particular, the Manhattan
distance is defined as

dM (P1 , P2 ) = |x1 − x2 | + |y1 − y2 |.

Prove that dM satisfies the following axioms that every distance function
must satisfy:

(i) dM (P1 , P2 ) ≥ 0 for any two points P1 and P2 and dM (P1 , P2 ) = 0
if and only if P1 = P2 ;

(ii) dM (P1 , P2 ) = dM (P2 , P1 );

(iii) dM (P1 , P2 ) ≤ dM (P1 , P3 ) + dM (P3 , P2 ) for any P1 , P2 , and P3 .

b. Sketch all the points in the x, y coordinate plane whose Manhattan
distance to the origin (0,0) is equal to 1. Do the same for the Euclidean
distance.

c. True or false: A solution to the closest-pair problem does not depend
on which of the two metrics–dE (Euclidean) or dM (Manhattan)–is used.
4. Odd pie fight There are n ≥ 3 people positioned in a field (Euclidean
plane) so that each has a unique nearest neighbor. Each person has a
cream pie. At a signal, everybody hurles his or her pie at the nearest
neighbor. Assuming that n is odd and that nobody can miss his or her
target, true or false: There always remains at least one person not hit by
a pie? [Car79].
5. The closest-pair problem can be posed in k-dimensional space in which
the Euclidean distance between two points P = (x1 , ..., xk ) and P =

15

(x1 , ..., xk ) is defined as

k
d(P , P ) = (xs − xs )2 .
s=1

What is the time-efficiency class of the brute-force algorithm for the k-
dimensional closest-pair problem?
6. Find the convex hulls of the following sets and identify their extreme points
(if they have any).
a. a line segment
b. a square
c. the boundary of a square
d. a straight line
7. Design a linear-time algorithm to determine two extreme points of the
convex hull of a given set of n > 1 points in the plane.
8. What modification needs to be made in the brute-force algorithm for the
convex-hull problem to handle more than two points on the same straight
line?
9. Write a program implementing the brute-force algorithm for the convex-
hull problem.
10. Consider the following small instance of the linear programming problem:

maximize 3x + 5y
subject to x+y ≤4
x + 3y ≤ 6
x ≥ 0, y ≥ 0

a. Sketch, in the Cartesian plane, the problem’s feasible region de-
fined as the set of points satisfying all the problem’s constraints.

b. Identify the region’s extreme points.

c. Solve the optimization problem given by using the following theorem:
A linear programming problem with a nonempty bounded feasible region
always has a solution, which can be found at one of the extreme points of
its feasible region.

16

1. Sorting n real numbers can be done in O(n log n) time.
2. a. Solving the problem for n = 2 and n = 3 should lead you to the critical
insight.

b. Where would you put the post office if it would not have to be at
one of the village locations?
3. a. Check requirements (i)—(iii) by using basic properties of absolute values.

b. For the Manhattan distance, the points in question are defined by
equation |x − 0| + |y − 0| = 1. You can start by sketching the points in
the positive quadrant of the coordinate system (i.e., the points for which
x, y ≥ 0) and then sketch the rest by using the symmetries.

c. The assertion is false. You can choose, say, P1 (0, 0), P2 (1, 0) and
find P3 to complete a counterexample.
4. True; prove it by mathematical induction.
5. Your answer should be a function of two parameters: n and k. A special
case of this problem (for k = 2) was solved in the text.
6. Review the examples given in the section.
7. Some of the extreme points of a convex hull are easier to find than others.
8. If there are other points of a given set on the straight line through Pi and
Pj , which of all these points need to be preserved for further processing?
9. Your program should work for any set of n distinct points, including sets
with many colinear points.
10. a. The set of points satisfying inequality ax + by ≤ c is the half-plane of
the points on one side of the straight line ax + by = c, including all the
points on the line itself. Sketch such a half-plane for each of the inequal-
ities and find their intersection.

b. The extreme points are the vertices of the polygon obtained in part a.

c. Compute and compare the values of the objective function at the ex-
treme points.

17

1. Sort the numbers in ascending order, compute the differences between ad-
jacent numbers in the sorted list, and find the smallest such difference. If
sorting is done in O(n log n) time, the running time of the entire algorithm
will be in
O(n log n) + Θ(n) + Θ(n) = O(n log n).

2. a. If we put the post office at location xi , the average distance between it
1 n
and all the points x1 < x2 < ... < xn is given by the formula n j=1 |xj −
xi |. Since the number of points n stays the same, we can ignore the multiple
1 n
n and minimize j=1 |xj − xi |. We’ll have to consider the cases of even
and odd n separately.

Let n be even. Consider first the case of n = 2. The sum |x1 − x| + |x2 − x|
is equal to x2 − x1 , the length of the interval with the endpoints at x1 and x2 ,
for any point x of this interval (including the endpoints), and it is larger than
x2 − x1 for any point x outside of this interval. This implies that for any even
n, the sum
n
|xj −x| = [|x1 −x|+|xn −x|]+[|x2 −x|+|xn−1 −x|]+...+[|xn/2 −x|+|xn/2+1 −x|]
j=1

is minimized when x belongs to each of the intervals [x1 , xn ] ⊃ [x2 , xn−1 ] ⊃ ... ⊃
[xn/2 , xn/2+1 ]. If x must be one of the points given, either xn/2 or xn/2+1 solves
the problem.
Let n > 1 be odd. Then, the sum n |xj − x| is minimized when x =
j=1
x n/2 , the point for which the number of the given points to the left of it is
equal to the number of the given points to the right of it.
Note that the point x n/2 –the n/2 th smallest called the median–solves
the problem for even n’s as well. For a sorted list implemented as an array,
the median can be found in Θ(1) time by simply returning the n/2 th.element
of the array. (Section 5.6 provides a more general discussion of algorithms for
computing the median.)

b. Assuming that the points x1 , x2 , ..., xn are given in increasing order, the
answer is the point xi that is the closest to m = (x1 + xn )/2, the middle point
between x1 and xn . (The middle point would be the obvious solution if the
post-post office didn’t have to be at one of the given locations.) Indeed, if we
put the post office at any location xi to the left of m, the longest distance from
a village to the post office would be xn − xi ; this distance is minimal for the
rightmost among such points. If we put the post office at any location xi to
the right of m, the longest distance from a village to the post office would be
xi − x1 ; this distance is minimal for the leftmost among such points.
Algorithm PostOffice1 (P )
//Input: List P of n (n ≥ 2) points x1 , x2 , ..., xn in increasing order

18

//Output: Point xi that minimizes max |xj − xi | among all x1 , x2 , ..., xn
1≤j≤n
m ← (x1 + xn )/2
i←1
while xi < m do
i←i+1
if xi − x1 < xn − xi−1
return xi
else return xi−1

The time efficiency of this algorithm is O(n).

3. a. For dM (P1 , P2 ) = |x1 − x2 | + |y1 − y2 |, we have the following:

(i) dM (P1 , P2 ) = |x1 − x2 | + |y1 − y2 | ≥ 0 and dM (P1 , P2 ) = 0 if and
only if both x1 = x2 and y1 = y2 , i.e., P1 and P2 coincide.

(ii) dM (P1 , P2 ) = |x1 − x2 | + |y1 − y2 | = |x2 − x1 | + |y2 − y1 |
= dM (P2 , P1 ).

(iii) dM (P1 , P2 ) = |x1 − x2 | + |y1 − y2 |
= |(x1 − x3 ) + (x3 − x2 )| + |(y1 − y3 ) + (y3 − y2 )|
≤ |x1 − x3 | + |x3 − x2 | + |y1 − y3 | + |y3 − y2 | = d(P1 , P3 ) + d(P3 , P2 ).

b. For the Manhattan distance, the points in question are defined by
the equation
|x − 0| + |y − 0| = 1, i.e., |x| + |y| = 1.
The graph of this equation is the boundary of the square with its vertices
at (1, 0), (0, 1), (−1, 0), and (−1, −1).

For the Euclidean distance, the points in question are defined by the equa-
tion
(x − 0)2 + (y − 0)2 = 1, i.e., x2 + y2 = 1.
The graph of this equation is the circumference of radius 1 and the center
at (0, 0).

c. False. Consider points P1 (0, 0), P2 (1, 0), and, say, P3 ( 1 , 3 ). Then
2 4

2 2
1 3
dE (P1 , P2 ) = 1 and dE (P3 , P1 ) = dE (P3 , P2 ) = + < 1.
2 4
Therefore, for the Euclidean distance, the two closest points are either P1
and P3 or P2 and P3 . For the Manhattan distance, we have
1 3 5
dM (P1 , P2 ) = 1 and dM (P3 , P1 ) = dM (P3 , P2 ) = + = > 1.
2 4 4

19

Therefore, for the Manhattan distance, the two closest points are P1 and
P2 .

4. We’ll prove by induction that there will always remain at least one person
not hit by a pie. The basis step is easy: If n = 3, the two persons with
the smallest pairwise distance between them throw at each other, while
the third person throws at one of them (whoever is closer). Therefore,
this third person remains “unharmed”.
For the inductive step, assume that the assertion is true for odd n ≥ 3,
and consider n + 2 persons. Again, the two persons with the smallest
pairwise distance between them (the closest pair) throw at each other.
Consider two possible cases as follows. If the remaining n persons all
throw at one another, at least one of them remains “unharmed” by the
inductive assumption. If at least one of the remaining n persons throws
at one of the closest pair, among the remaining n − 1 persons, at most
n − 1 pies are thrown at one another, and hence at least one person must
remain “unharmed” because there is not enough pies to hit everybody in
that group. This completes the proof.
5. The number of squarings will be
n−1 n k n−1 n n−1
C(n, k) = 1= k=k (n − i)
i=1 j=i+1 s=1 i=1 j=i+1 i=1
k(n − 1)n
= k[(n − 1) + (n − 2) + .... + 1] = ∈ Θ(kn2 ).
2

6. a. The convex hull of a line segment is the line segment itself; its extreme
points are the endpoints of the segment.

b. The convex hull of a square is the square itself; its extreme points
are the four vertices of the square.

c. The convex hull of the boundary of a square is the region comprised of
the points within that boundary and on the boundary itself; its extreme
points are the four vertices of the square.

d. The convex hull of a straight line is the straight line itself. It doesn’t
have any extreme points.

7. Find the point with the smallest x coordinate; if there are several such
points, find the one with the smallest y coordinate among them. Similarly,
find the point with the largest x coordinate; if there are several such points,
find the one with the largest y coordinate among them.

20

8. If there are other points of a given set on the straight line through Pi and
Pj (while all the other points of the set lie on the same side of the line), a
line segment of the convex hull’s boundary will have its end points at the
two farthest set points on the line. All the other points on the line can
be eliminated from further processing.

9. n/a

10. a. Here is a sketch of the feasible region in question:
y

.

x + 3y = 6

( 0, 2 )
.
.( 3, 1 )
( 0, 0 )
. ( 4, 0 )
. . x

x+y=4

.
b. The extreme points are: (0, 0), (4,0), (3,1), and (0,2).

c.
Extreme point Value of 3x + 5y
(0,0) 0
(4,0) 12
(3,1) 14
(0,2) 10

So, the optimal solution is (3, 1), with the maximal value of 3x + 5y equal to 14.
(Note: This instance of the linear programming problem is discussed further in
Section 10.1.)

21

Exercises 3.4
1. a. Assuming that each tour can be generated in constant time, what will
be the efficiency class of the exhaustive-search algorithm outlined in the
text for the traveling salesman problem?

b. If this algorithm is programmed on a computer that makes one billion
additions per second, estimate the maximum number of cities for which
the problem can be solved in (i) one hour;(ii) 24-hours; (iii) one year; (iv)
one century.
2. Outline an exhaustive-search algorithm for the Hamiltonian circuit prob-
lem.
3. Outline an algorithm to determine whether a connected graph represented
by its adjacency matrix has an Eulerian circuit. What is the efficiency
class of your algorithm?
4. Complete the application of exhaustive search to the instance of the as-
signment problem started in the text.
5. Give an example of the assignment problem whose optimal solution does
not include the smallest element of its cost matrix.
6. Consider the partition problem: given n positive integers, partition
them into two disjoint subsets with the same sum of their elements. (Of
course, the problem does not always have a solution.) Design an exhaus-
tive search algorithm for this problem. Try to minimize the number of
subsets the algorithm needs to generate.
7. Consider the clique problem: given a graph G and a positive integer k,
determine whether the graph contains a clique of size k, i.e., a complete
subgraph of k vertices. Design an exhaustive-search algorithm for this
problem.
8. Explain how exhaustive search can be applied to the sorting problem and
determine the efficiency class of such an algorithm.
9. A magic square of order n is an arrangement of the numbers from 1 to
n2 in an n-by-n matrix, with each number occurring exactly once, so that
each row, each column, and each main diagonal has the same sum.

a. Prove that if a magic square of order n exists, the sum in question
must be equal to n(n2 + 1)/2.

b. Design an exhaustive search algorithm for generating all magic squares
of order n.

c. Go to the Internet or your library and find a better algorithm for

22

generating magic squares.

d. Implement the two algorithms–the exhaustive search and the one
you have found–and run an experiment to determine the largest value of
n for which each of the algorithms is able to find a magic square of order
n in less than one minute of your computer’s time.
10. Famous alphametic A puzzle in which the digits in a correct mathemati-
cal expression, such as a sum, are replaced by letters is called cryptarithm;
if, in addition, the puzzle’s words make sense, it is said to be an al-
phametic. The most well-known alphametic was published by the renowned
British puzzlist H. E. Dudeney (1857-1930):

SEND
+MORE
–––––—
MONEY

Two conditions are assumed: First, the correspondence between letters
and digits is one-to-one, that is each letter represents one digit only and
different letters represent different digits. Second, the digit zero does
not appear as the left-most digit in any of the numbers. To solve an
alphametic means to find which digit each letter represents. Note that
a solution’s uniqueness cannot be assumed and has to be verified by the
solver.

a. Write a program for solving cryptarithms by exhaustive search. As-
sume that a given cryptarithm is a sum of two words.

b. Solve Dudeney’e puzzle the way it was expected to be solved when
it was first published in 1924.

23

1. a. Identify the algorithm’s basic operation and count the number of times
it will be executed.

b. For each of the time amounts given, find the largest value of n for
which this limit will not be exceeded.
2. How different is the traveling salesman problem from the problem of find-
ing a Hamiltonian circuit?
3. Your algorithm should check the well-known conditions that are both nec-
essary and sufficient for the existence of a Eulerian circuit in a connected
graph.
4. Generate the remaining 4! − 6 = 18 possible assignments, compute their
costs, and find the one with the smallest cost.
5. Make the size of your counterexample as small as possible.
6. Rephrase the problem so that the sum of elements in one subset, rather
than two, needs to be checked on each try of a possible partition.
7. Follow the definitions of a clique and of an exhaustive search algorithm.
8. Try all possible orderings of the elements given.
9. a. Add all the elements in the magic square in two different ways.

b. What combinatorial objects do you have to generate here?
10. a. For testing, you may use alphametic collections available on the Inter-
net.

b. Given the absence of electronic computers in 1924, you must refrain
here from using the Internet.

24

1. a. Θ(n!)
For each tour (a sequence of n+1 cities), one needs n additions to compute
the tour’s length. Hence, the total number of additions A(n) will be n
times the total number of tours considered, i.e., n∗ 1 (n−1)! = 1 n! ∈ Θ(n!).
2 2

b. (i) nmax = 15; (ii) nmax = 16; (iii) nmax = 18; (iv) nmax = 20.
Given the answer to part a, we have to find the largest value of n such
that
1
n!10−9 ≤ t
2
where t is the time available (in seconds). Thus, for t = 1hr = 3.6∗103 sec,
we get the inequality
n! ≤ 2 ∗ 109 t = 7.2 ∗ 1012 .
The largest value of n for which this inequality holds is 15 (since 15!
≈ 1.3 ∗ 1012 and 16! ≈ 2.1 ∗ 1013 ).

For the other given values of t, the answers can be obtained in the same
manner.

2. The problem of finding a Hamiltonian circuit is very similar to the travel-
ing salesman problem. Generate permutations of n vertices that start and
end with, say, the first vertex, and check whether every pair of successive
vertices in a current permutation are connected by an edge. If it’s the
case, the current permutation represents a Hamiltonian circuit, otherwise,
a next permutation needs to be generated.

3. A connected graph has a Eulerian circuit if and only if all its vertices
have even degrees. An algorithm should check this condition until either
an odd vertex is encountered (then a Eulerian circuit doesn’t exist) or
all the vertices turn out to be even (then a Eulerian circuit must exist).
For a graph (with no loops) represented by its n-by-n adjacency matrix,
the degree of a vertex is the number of ones in the vertex’s row. Thus,
computing its degree will take the Θ(n) time, checking whether it’s even
will take Θ(1) time, and it will be done between 1 and n times. Hence,
the algorithm’s efficiency will be in O(n2 ).

4. The following assignments were generated in the chapter’s text:

  1, 2, 3, 4 cost = 9+4+1+4 = 18
9 2 7 8 1, 2, 4, 3 cost = 9+4+8+9 = 30
 6 4 3 7  1, 3, 2, 4 cost = 9+3+8+4 = 24
C=
 5
 etc.
8 1 8  1, 3, 4, 2 cost = 9+3+8+6 = 26
7 6 9 4 1, 4, 2, 3 cost = 9+7+8+9 = 33
1, 4, 3, 2 cost = 9+7+1+6 = 23

25

The remaining ones are

2, 1, 3, 4 cost = 2+6+1+4 = 13
2, 1, 4, 3 cost = 2+6+8+9 = 25
2, 3, 1, 4 cost = 2+3+5+4 = 14
2, 3, 4, 1 cost = 2+3+8+7 = 20
2, 4, 1, 3 cost = 2+7+5+9 = 23
2, 4, 3, 1 cost = 2+7+1+7 = 17

3, 1, 2, 4 cost = 7+6+8+4 = 25
3, 1, 4, 2 cost = 7+6+8+6 = 27
3, 2, 1, 4 cost = 7+4+5+4 = 20
3, 2, 4, 1 cost = 7+4+8+7 = 26
3, 4, 1, 2 cost = 7+7+5+6 = 25
3, 4, 2, 1 cost = 7+7+8+7 = 29

4, 1, 2, 3 cost = 8+6+8+9 = 31
4, 1, 3, 2 cost = 8+6+1+6 = 21
4, 2, 1, 3 cost = 8+4+5+9 = 26
4, 2, 3, 1 cost = 8+4+1+7 = 20
4, 3, 1, 2 cost = 8+3+5+6 = 22
4, 3, 2, 1 cost = 8+3+8+7 = 26
The optimal solution is: Person 1 to Job 2, Person 2 to Job 1, Person
3 to Job 3, and Person 4 to Job 4, with the total (minimal) cost of the
assignment being 13.

5. Here is a very simple example:

1 2
2 9

6. Start by computing the sum S of the numbers given. If S is odd, stop
because the problem doesn’t have a solution. If S is even, generate the
subsets until either a subset whose elements’ sum is S/2 is encountered or
no more subsets are left. Note that it will suﬃce to generate only subsets
with no more than n/2 elements.

7. Generate a subset of k vertices and check whether every pair of vertices
in the subset is connected by an edge. If it’s true, stop (the subset is a
clique); otherwise, generate the next subset.

8. Generate a permutation of the elements given and check whether they
are ordered as required by comparing values of its consecutive elements.
If they are, stop; otherwise, generate the next permutation. Since the

26

number of permutations of n items is equal to n! and checking a permu-
tation requires up to n − 1 comparisons, the algorithm’s efficiency class is
in O(n!(n − 1)) = O((n + 1)!).

9. a. Let s be the sum of the numbers in each row of an n-by-n magic square.
Let us add all the numbers in rows 1 through n. We will get the following
equality:

n2 (n2 + 1) n(n2 + 1)
sn = 1+2+...+n2 , i.e., sn = , which implies that s = .
2 2
b. Number positions in an n-by-n matrix from 1 through n2 . Generate a
permutation of the numbers 1 through n2 , put them in the corresponding
positions of the matrix, and check the magic-square equality (proved in
part (a)) for every row, every column, and each of the two main diagonals
of the matrix.

c. n/a

d. n/a

10. a. Since the letter-digit correspondence must be one-to-one and there
are only ten distinct decimal digits, the exhaustive search needs to check
P (10, k) = 10!/(10 − k)! possible substitutions, where k is the number
of distinct letters in the input. (The requirement that the first letter of
a word cannot represent 0 can be used to reduce this number further.)
Thus a program should run in a quite reasonable amount of time on to-
day’s computers. Note that rather than checking two cases–with and
without a “1-carry”–for each of the decimal positions, the program can
check just one equality, which stems from the definition of the decimal
number system. For Dudeney’s alphametic, for example, this equality is
1000(S+M) + 100(E+O) + 10(N+R) + (D+E) = 10000M + 1000O +
100N + 10E + Y

b. Here is a “computerless” solution to this classic problem. First, no-
tice that M must be 1. (Since both S and M are not larger than 9, their
sum, even if increased by 1 because of the carry from the hundred column,
must be less than 20.) We will have to rely on some further insigts into
specifics of the problem. The leftmost digits of the addends imply one of
the two possibilities: either S + M = 10 + O (if there was no carry from
the hundred column) or 1 + S + M = 10 + O (if there was such a carry).
First, let us pursue the former of the two possibilities. Since M = 1, S ≤
9 and O ≥ 0, the equation S + 1 = 10 + O has only one solution: S = 9
and O = 0. This leaves us with

27

END
+ORE
–––—
NEY

Since we deal here with the case of no carry from the hundreds and E and
N must be distinct, the only possibility is a carry from the tens: 1 + E = N
and either N + R = 10 + E (if there was no carry from the rightmost column)
or 1 + N + R = 10 + E (if there was such a carry). The ﬁrst combination
leads to a contradiction: Substituting 1 + E for N into N + R = 10 + E, we
obtain R = 9, which is incompatible with the same digit already represented by
S. The second combination of 1 + E = N and 1 + N + R = 10 + E implies,
after substituting the ﬁrst of these equations into the second one, R = 8. Note
that the only remaining digit values still unassigned are 2, 3, 4, 5, 6, and 7.
Finally, for the righmost column, we have the equation D + E = 10 + Y. But
10 + Y ≥ 12, because the smallest unassigned digit value is 2 while D + E ≤ 12
because the two largest unassigned digit values are 6 and 7 and E < N. Thus,
D + E = 10 + Y = 12. Hence Y = 2 and D + E = 12. The only pair of still
unassigned digit values that add up to 12, 5 and 7, must be assigned to E and
D, respectively, since doing this the other way (E = 7, D = 5) would imply N
= E + 1 = 8, which is already represented by R. Thus, we found the following
solution to the puzzle:

9567
+1085
––––—
10652

Is this the only solution? To answer this question, we should pursue the
carry possibility from the hundred column to the thousand column (see above).
Then 1 + S + M = 10 + O or, since M =1, S = 8 + O. But S ≤ 9, while 8 +
O ≥ 10 since O ≥ 2. Hence the last equation has no solutions in our domain.
This proves that the puzzle has no other solutions.

28

Solution 3.

More Related Content

What's hot

Viewers also liked

Similar to Solution 3.

Solution 3.