This action might not be possible to undo. Are you sure you want to continue?
Discrete Mathematics
Miguel A. Lerma
Contents
Introduction 5
Chapter 1. Logic, Proofs 6
1.1. Propositions 6
1.2. Predicates, Quantiﬁers 11
1.3. Proofs 13
Chapter 2. Sets, Functions, Relations 19
2.1. Set Theory 19
2.2. Functions 27
2.3. Relations 32
Chapter 3. Algorithms, Integers 38
3.1. Algorithms 38
3.2. The Euclidean Algorithm 48
3.3. Modular Arithmetic, RSA Algorithm 52
Chapter 4. Induction, Recurences 59
4.1. Sequences and Strings 59
4.2. Mathematical Induction 62
4.3. Recurrence Relations 65
Chapter 5. Counting 69
5.1. Basic Principles 69
5.2. Combinatorics 71
5.3. Generalized Permutations and Combinations 73
5.4. Binomial Coeﬃcients 75
5.5. The Pigeonhole Principle 77
Chapter 6. Probability 78
6.1. Probability 78
Chapter 7. Graph Theory 82
7.1. Graphs 82
7.2. Representations of Graphs 88
7.3. Paths and Circuits 91
3
CONTENTS 4
7.4. Planar Graphs 97
Chapter 8. Trees 100
8.1. Trees 100
8.2. Binary Trees 102
8.3. Decision Trees, Tree Isomorphisms 104
8.4. Tree Transversal 113
8.5. Spanning Trees 116
Chapter 9. Boolean Algebras 122
9.1. Combinatorial Circuits 122
9.2. Boolean Functions, Applications 127
Chapter 10. Automata, Grammars and Languages 133
10.1. Finite State Machines 133
10.2. Languages and Grammars 137
10.3. Language Recognition 144
Appendix A. 150
A.1. Eﬃcient Computation of Powers Modulo m 150
A.2. Machines and Languages 152
Introduction
These notes are intended to be a summary of the main ideas in
course CS 310: Mathematical Foundations of Computer Science. I
may keep working on this document as the course goes on, so these
notes will not be completely ﬁnished until the end of the quarter.
The textbook for this course is Keneth H. Rosen: Discrete Mathe
matics and Its Applications, Fifth Edition, 2003, McGrawHill. With
few exceptions I will follow the notation in the book.
These notes contain some questions and “exercises” intended to
stimulate the reader who wants to play a somehow active role while
studying the subject. They are not homework nor need to be addressed
at all if the reader does not wish to. I will recommend exercises and
give homework assignments separately.
Finally, if you ﬁnd any typos or errors, or you have any suggestions,
please, do not hesitate to let me know.
Miguel A. Lerma
mlerma@math.northwestern.edu
Northwestern University
Spring 2005
http://www.math.northwestern.edu/~mlerma/courses/cs31005s/
5
CHAPTER 1
Logic, Proofs
1.1. Propositions
A proposition is a declarative sentence that is either true or false
(but not both). For instance, the following are propositions: “Paris
is in France” (true), “London is in Denmark” (false), “2 < 4” (true),
“4 = 7 (false)”. However the following are not propositions: “what
is your name?” (this is a question), “do your homework” (this is a
command), “this sentence is false” (neither true nor false), “x is an
even number” (it depends on what x represents), “Socrates” (it is not
even a sentence). The truth or falsehood of a proposition is called its
truth value.
1.1.1. Connectives, Truth Tables. Connectives are used for
making compound propositions. The main ones are the following (p
and q represent given propositions):
Name Represented Meaning
Negation p “not p”
Conjunction p ∧ q “p and q”
Disjunction p ∨ q “p or q (or both)”
Exclusive Or p ⊕q “either p or q, but not both”
Implication p → q “if p then q”
Biconditional p ↔ q “p if and only if q”
The truth value of a compound proposition depends only on the
value of its components. Writing F for “false” and T for “true”, we
can summarize the meaning of the connectives in the following way:
6
1.1. PROPOSITIONS 7
p q p p ∧ q p ∨ q p ⊕q p → q p ↔ q
T T F T T F T T
T F F F T T F F
F T T F T T T F
F F T F F F T T
Note that ∨ represents a nonexclusive or, i.e., p ∨ q is true when
any of p, q is true and also when both are true. On the other hand ⊕
represents an exclusive or, i.e., p ⊕ q is true only when exactly one of
p and q is true.
1.1.2. Tautology, Contradiction, Contingency.
1. A proposition is said to be a tautology if its truth value is T
for any assignment of truth values to its components. Example:
The proposition p ∨ p is a tautology.
2. A proposition is said to be a contradiction if its truth value is F
for any assignment of truth values to its components. Example:
The proposition p ∧ p is a contradiction.
3. A proposition that is neither a tautology nor a contradiction is
called a contingency.
p p p ∨ p p ∧ p
T F T F
T F T F
F T T F
F T T F
6 6
tautology contradiction
1.1.3. Conditional Propositions. A proposition of the form “if
p then q” or “p implies q”, represented “p → q” is called a conditional
proposition. For instance: “if John is from Chicago then John is from
Illinois”. The proposition p is called hypothesis or antecedent, and the
proposition q is the conclusion or consequent.
Note that p → q is true always except when p is true and q is false.
So, the following sentences are true: “if 2 < 4 then Paris is in France”
(true → true), “if London is in Denmark then 2 < 4” (false → true),
1.1. PROPOSITIONS 8
“if 4 = 7 then London is in Denmark” (false → false). However the
following one is false: “if 2 < 4 then London is in Denmark” (true →
false).
In might seem strange that “p → q” is considered true when p is
false, regardless of the truth value of q. This will become clearer when
we study predicates such as “if x is a multiple of 4 then x is a multiple
of 2”. That implication is obviously true, although for the particular
case x = 3 it becomes “if 3 is a multiple of 4 then 3 is a multiple of 2”.
The proposition p ↔ q, read “p if and only if q”, is called bicon
ditional. It is true precisely when p and q have the same truth value,
i.e., they are both true or both false.
1.1.4. Logical Equivalence. Note that the compound proposi
tions p → q and p ∨ q have the same truth values:
p q p p ∨ q p → q
T T F T T
T F F F F
F T T T T
F F T T T
When two compound propositions have the same truth values no
matter what truth value their constituent propositions have, they are
called logically equivalent. For instance p → q and p ∨ q are logically
equivalent, and we write it:
p → q ≡ p ∨ q
Note that that two propositions A and B are logically equivalent
precisely when A ↔ B is a tautology.
Example: De Morgan’s Laws for Logic. The following propositions
are logically equivalent:
(p ∨ q) ≡ p ∧ q
(p ∧ q) ≡ p ∨ q
We can check it by examining their truth tables:
1.1. PROPOSITIONS 9
p q p q p ∨ q (p ∨ q) p ∧ q p ∧ q (p ∧ q) p ∨ q
T T F F T F F T F F
T F F T T F F F T T
F T T F T F F F T T
F F T T F T T F T T
Example: The following propositions are logically equivalent:
p ↔ q ≡ (p → q) ∧ (q → p)
Again, this can be checked with the truth tables:
p q p → q q → p (p → q) ∧ (q → p) p ↔ q
T T T T T T
T F F T F F
F T T F F F
F F T T T T
Exercise: Check the following logical equivalences:
(p → q) ≡ p ∧ q
p → q ≡ q → p
(p ↔ q) ≡ p ⊕q
1.1.5. Converse, Contrapositive. The converse of a conditional
proposition p → q is the proposition q → p. As we have seen, the bi
conditional proposition is equivalent to the conjunction of a conditional
proposition an its converse.
p ↔ q ≡ (p → q) ∧ (q → p)
So, for instance, saying that “John is married if and only if he has a
spouse” is the same as saying “if John is married then he has a spouse”
and “if he has a spouse then he is married”.
Note that the converse is not equivalent to the given conditional
proposition, for instance “if John is from Chicago then John is from
Illinois” is true, but the converse “if John is from Illinois then John is
from Chicago” may be false.
1.1. PROPOSITIONS 10
The contrapositive of a conditional proposition p → q is the propo
sition q → p. They are logically equivalent. For instance the con
trapositive of “if John is from Chicago then John is from Illinois” is “if
John is not from Illinois then John is not from Chicago”.
1.2. PREDICATES, QUANTIFIERS 11
1.2. Predicates, Quantiﬁers
1.2.1. Predicates. A predicate or propositional function is a state
ment containing variables. For instance “x +2 = 7”, “X is American”,
“x < y”, “p is a prime number” are predicates. The truth value of the
predicate depends on the value assigned to its variables. For instance if
we replace x with 1 in the predicate “x+2 = 7” we obtain “1+2 = 7”,
which is false, but if we replace it with 5 we get “5 + 2 = 7”, which
is true. We represent a predicate by a letter followed by the variables
enclosed between parenthesis: P(x), Q(x, y), etc. An example for P(x)
is a value of x for which P(x) is true. A counterexample is a value of
x for which P(x) is false. So, 5 is an example for “x + 2 = 7”, while 1
is a counterexample.
Each variable in a predicate is assumed to belong to a universe (or
domain) of discourse, for instance in the predicate “n is an odd integer”
’n’ represents an integer, so the universe of discourse of n is the set of
all integers. In “X is American” we may assume that X is a human
being, so in this case the universe of discourse is the set of all human
beings.
1
1.2.2. Quantiﬁers. Given a predicate P(x), the statement “for
some x, P(x)” (or “there is some x such that p(x)”), represented
“∃x P(x)”, has a deﬁnite truth value, so it is a proposition in the
usual sense. For instance if P(x) is “x + 2 = 7” with the integers as
universe of discourse, then ∃x P(x) is true, since there is indeed an
integer, namely 5, such that P(5) is a true statement. However, if
Q(x) is “2x = 7” and the universe of discourse is still the integers,
then ∃x Q(x) is false. On the other hand, ∃x Q(x) would be true if we
extend the universe of discourse to the rational numbers. The symbol
∃ is called the existential quantiﬁer.
Analogously, the sentence “for all x, P(x)”—also “for any x, P(x)”,
“for every x, P(x)”, “for each x, P(x)”—, represented “∀x P(x)”, has
a deﬁnite truth value. For instance, if P(x) is “x + 2 = 7” and the
1
Usually all variables occurring in predicates along a reasoning are supposed to
belong to the same universe of discourse, but in some situations (as in the so called
manysorted logics) it is possible to use diﬀerent kinds of variables to represent
diﬀerent types of objects belonging to diﬀerent universes of discourse. For instance
in the predicate “σ is a string of length n” the variable σ represents a string, while
n represents a natural number, so the universe of discourse of σ is the set of all
strings, while the universe of discourse of n is the set of natural numbers.
1.2. PREDICATES, QUANTIFIERS 12
universe of discourse is the integers, then ∀x P(x) is false. However if
Q(x) represents “(x + 1)
2
= x
2
+ 2x + 1” then ∀x Q(x) is true. The
symbol ∀ is called the universal quantiﬁer.
In predicates with more than one variable it is possible to use several
quantiﬁers at the same time, for instance ∀x∀y∃z P(x, y, z), meaning
“for all x and all y there is some z such that P(x, y, z)”.
Note that in general the existential and universal quantiﬁers cannot
be swapped, i.e., in general ∀x∃y P(x, y) means something diﬀerent
from ∃y∀x P(x, y). For instance if x and y represent human beings and
P(x, y) represents “x is a friend of y”, then ∀x∃y P(x, y) means that
everybody is a friend of someone, but ∃y∀x P(x, y) means that there
is someone such that everybody is his or her friend.
A predicate can be partially quantiﬁed, e.g. ∀x∃y P(x, y, z, t). The
variables quantiﬁed (x and y in the example) are called bound variables,
and the rest (z and t in the example) are called free variables. A
partially quantiﬁed predicate is still a predicate, but depending on
fewer variables.
1.2.3. Generalized De Morgan Laws for Logic. If ∃x P(x) is
false then there is no value of x for which P(x) is true, or in other
words, P(x) is always false. Hence
∃x P(x) ≡ ∀x P(x) .
On the other hand, if ∀x P(x) is false then it is not true that for
every x, P(x) holds, hence for some x, P(x) must be false. Thus:
∀x P(x) ≡ ∃x P(x) .
This two rules can be applied in successive steps to ﬁnd the negation
of a more complex quantiﬁed statement, for instance:
∃x∀y p(x, y) ≡ ∀x∀y P(x, y) ≡ ∀x∃y P(x, y) .
Exercise: Write formally the statement “for every real number there
is a greater real number”. Write the negation of that statement.
Answer: The statement is: ∀x ∃y (x < y) (the universe of discourse
is the real numbers). Its negation is: ∃x ∀y (x < y), i.e., ∃x ∀y (x < y).
(Note that among real numbers x < y is equivalent to x ≥ y, but
formally they are diﬀerent predicates.)
1.3. PROOFS 13
1.3. Proofs
1.3.1. Mathematical Systems, Proofs. A Mathematical Sys
tem consists of:
1. Axioms: propositions that are assumed true.
2. Deﬁnitions: used to create new concepts from old ones.
3. Undeﬁned terms: corresponding to the primitive concepts of the
system (for instance in set theory the term “set” is undeﬁned).
A theorem is a proposition that can be proved to be true. An
argument that establishes the truth of a proposition is called a proof.
Example: Prove that if x > 2 and y > 3 then x +y > 5.
Answer: Assuming x > 2 and y > 3 and adding the inequalities
term by term we get: x +y > 2 + 3 = 5.
That is an example of direct proof. In a direct proof we assume the
hypothesis together with axioms and other theorems previously proved
and we derive the conclusion from them.
An indirect proof or proof by contrapositive consists of proving the
contrapositive of the desired implication, i.e., instead of proving p → q
we prove q → p.
Example: Prove that if x +y > 5 then x > 2 or y > 3.
Answer: We must prove that x + y > 5 → (x > 2) ∨ (y > 3). An
indirect proof consists of proving ((x > 2) ∨ (y > 3)) → (x +y > 5).
In fact: ((x > 2) ∨ (y > 3)) is the same as (x ≤ 2)∧(y ≤ 3), so adding
both inequalities we get x +y ≤ 5, which is the same as (x +y > 5).
Proof by Contradiction. In a proof by contradiction or (Reductio ad
Absurdum) we assume the hypotheses and the negation of the conclu
sion, and try to derive a contradiction, i.e., a proposition of the form
r ∧ r.
Example: Prove by contradiction that if x+y > 5 then either x > 2
or y > 3.
Answer: We assume the hypothesis x +y > 5. From here we must
conclude that x > 2 or y > 3. Assume to the contrary that “x > 2 or
y > 3” is false, so x ≤ 2 and y ≤ 3. Adding those inequalities we get
1.3. PROOFS 14
x ≤ 2 + 3 = 5, which contradicts the hypothesis x +y > 5. From here
we conclude that the assumption “x ≤ 2 and y ≤ 3” cannot be right,
so “x > 2 or y > 3” must be true.
Remark: Sometimes it is diﬃcult to distinguish between an indirect
proof and a proof by contradiction. In an indirect proof we prove an
implication of the form p → q by proving the contrapositive q →
p. In an proof by contradiction we prove an statement s (which
may or may not be an implication) by assuming s and deriving a
contradiction. In fact proofs by contradiction are more general than
indirect proofs.
Exercise: Prove by contradiction that
√
2 is not a rational number,
i.e., there are no integers a, b such that
√
2 = a/b.
Answer: Assume that
√
2 is rational, i.e.,
√
2 = a/b, where a and b
are integers and the fraction is written in least terms. Squaring both
sides we have 2 = a
2
/b
2
, hence 2 b
2
= a
2
. Since the left hand side is
even, then a
2
is even, but this implies that a itself is even, so a = 2 a
.
Hence: 2 b
2
= 4 a
2
, and simplifying: b
2
= 2 a
2
. This implies that b
2
is even, so b is even: b = 2b
. Consequently a/b = 2a
/2b
= a
/b
,
contradicting the hypothesis that a/b was in least terms.
1.3.2. Arguments, Rules of Inference. An argument is a se
quence of propositions p
1
, p
2
, . . . , p
n
called hypotheses (or premises)
followed by a proposition q called conclusion. An argument is usually
written:
p
1
p
2
.
.
.
p
n
∴ q
or
p
1
, p
2
, . . . , p
n
/ ∴ q
The argument is called valid if q is true whenever p
1
, p
2
, . . . , p
n
are
true; otherwise it is called invalid.
1.3. PROOFS 15
Rules of inference are certain simple arguments known to be valid
and used to make a proof step by step. For instance the following
argument is called modus ponens or rule of detachment:
p → q
p
∴ q
In order to check whether it is valid we must examine the following
truth table:
p q p → q p q
T T T T T
T F F T F
F T T F T
F F T F F
If we look now at the rows in which both p → q and p are true (just
the ﬁrst row) we see that also q is true, so the argument is valid.
Other rules of inference are the following:
1. Modus Ponens or Rule of Detachment:
p → q
p
∴ q
2. Modus Tollens:
p → q
q
∴ p
3. Addition:
p
∴ p ∨ q
4. Simpliﬁcation:
p ∧ q
∴ p
5. Conjunction:
1.3. PROOFS 16
p
q
∴ p ∧ q
6. Hypothetical Syllogism:
p → q
q → r
∴ p → r
7. Disjunctive Syllogism:
p ∨ q
p
∴ q
8. Resolution:
p ∨ q
p ∨ r
∴ q ∨ r
Arguments are usually written using three columns. Each row con
tains a label, a statement and the reason that justiﬁes the introduction
of that statement in the argument. That justiﬁcation can be one of the
following:
1. The statement is a premise.
2. The statement can be derived from statements occurring earlier
in the argument by using a rule of inference.
Example: Consider the following statements: “I take the bus or
I walk. If I walk I get tired. I do not get tired. Therefore I take the
bus.” We can formalize this by calling B = “I take the bus”, W =
“I walk” and T = “I get tired”. The premises are B ∨W, W → T and
T, and the conclusion is B. The argument can be described in the
following steps:
step statement reason
1) W → T Premise
2) T Premise
3) W 1,2, Modus Tollens
4) B ∨ W Premise
5) ∴ B 4,3, Disjunctive Syllogism
1.3. PROOFS 17
1.3.3. Rules of Inference for Quantiﬁed Statements. We
state the rules for predicates with one variable, but they can be gener
alized to predicates with two or more variables.
1. Universal Instantiation. If ∀x p(x) is true, then p(a) is true for
each speciﬁc element a in the universe of discourse; i.e.:
∀x p(x)
∴ p(a)
For instance, from ∀x (x+1 = 1+x) we can derive 7+1 = 1+7.
2. Existential Instantiation. If ∃x p(x) is true, then p(a) is true for
some speciﬁc element a in the universe of discourse; i.e.:
∃x p(x)
∴ p(a)
The diﬀerence respect to the previous rule is the restriction in
the meaning of a, which now represents some (not any) element
of the universe of discourse. So, for instance, from ∃x (x
2
= 2)
(the universe of discourse is the real numbers) we derive the
existence of some element, which we may represent ±
√
2, such
that (±
√
2)
2
= 2.
3. Universal Generalization. If p(x) is proved to be true for a
generic element in the universe of discourse, then ∀x p(x) is
true; i.e.:
p(x)
∴ ∀x p(x)
By “generic” we mean an element for which we do not make any
assumption other than its belonging to the universe of discourse.
So, for instance, we can prove ∀x [(x + 1)
2
= x
2
+ 2x + 1] (say,
for real numbers) by assuming that x is a generic real number
and using algebra to prove (x + 1)
2
= x
2
+ 2x + 1.
4. Existential Generalization. If p(a) is true for some speciﬁc ele
ment a in the universe of discourse, then ∃x p(x) is true; i.e.:
p(a)
∴ ∃x p(x)
For instance: from 7 + 1 = 8 we can derive ∃x (x + 1 = 8).
Example: Show that a counterexample can be used to disprove a
universal statement, i.e., if a is an element in the universe of discourse,
1.3. PROOFS 18
then from p(a) we can derive ∀x p(x). Answer: The argument is as
follows:
step statement reason
1) p(a) Premise
2) ∃x p(x) Existential Generalization
3) ∀x p(x) Negation of Universal Statement
CHAPTER 2
Sets, Functions, Relations
2.1. Set Theory
2.1.1. Sets. A set is a collection of objects, called elements of the
set. A set can be represented by listing its elements between braces:
A = ¦1, 2, 3, 4, 5¦. The symbol ∈ is used to express that an element is
(or belongs to) a set, for instance 3 ∈ A. Its negation is represented by
∈, e.g. 7 ∈ A. If the set is ﬁnite, its number of elements is represented
[A[, e.g. if A = ¦1, 2, 3, 4, 5¦ then [A[ = 5.
Some important sets are the following:
1. N = ¦0, 1, 2, 3, ¦ = the set of natural numbers.
1
2. Z = ¦ , −3, −2, −1, 0, 1, 2, 3, ¦ = the set of integers.
3. Q = the set of rational numbers.
4. R = the set of real numbers.
5. C = the set of complex numbers.
Is S is one of those sets then we also use the following notations:
2
1. S
+
= set of positive elements in S, for instance
Z
+
= ¦1, 2, 3, ¦ = the set of positive integers.
2. S
−
= set of negative elements in S, for instance
Z
−
= ¦−1, −2, −3, ¦ = the set of negative integers.
3. S
∗
= set of elements in S excluding zero, for instance R
∗
= the
set of non zero real numbers.
Setbuilder notation. An alternative way to deﬁne a set, called set
builder notation, is by stating a property (predicate) P(x) veriﬁed by
exactly its elements, for instance A = ¦x ∈ Z [ 1 ≤ x ≤ 5¦ = “set of
1
Note that N includes zero—for some authors N = ¦1, 2, 3, ¦, without zero.
2
When working with strings we will use a similar notation with a diﬀerent
meaning—be careful not to confuse it.
19
2.1. SET THEORY 20
integers x such that 1 ≤ x ≤ 5”—i.e.: A = ¦1, 2, 3, 4, 5¦. In general:
A = ¦x ∈ U [ p(x)¦, where U is the universe of discourse in which the
predicate P(x) must be interpreted, or A = ¦x [ P(x)¦ if the universe
of discourse for P(x) is implicitly understood. In set theory the term
universal set is often used in place of “universe of discourse” for a given
predicate.
3
Principle of Extension. Two sets are equal if and only if they have
the same elements, i.e.:
A = B ≡ ∀x (x ∈ A ↔ x ∈ B) .
Subset. We say that A is a subset of set B, or A is contained in
B, and we represent it “A ⊆ B”, if all elements of A are in B, e.g., if
A = ¦a, b, c¦ and B = ¦a, b, c, d, e¦ then A ⊆ B.
A is a proper subset of B, represented “A ⊂ B”, if A ⊆ B but
A = B, i.e., there is some element in B which is not in A.
Empty Set. A set with no elements is called empty set (or null set,
or void set), and is represented by ∅ or ¦¦.
Note that nothing prevents a set from possibly being an element of
another set (which is not the same as being a subset!). For instance
if A = ¦1, a, ¦3, t¦, ¦1, 2, 3¦¦ and B = ¦3, t¦, then obviously B is an
element of A, i.e., B ∈ A.
Power Set. The collection of all subsets of a set A is called the
power set of A, and is represented P(A). For instance, if A = ¦1, 2, 3¦,
then
P(A) = ¦∅, ¦1¦, ¦2¦, ¦3¦, ¦1, 2¦, ¦1, 3¦, ¦2, 3¦, A¦ .
Exercise: Prove by induction that if [A[ = n then [P(A)[ = 2
n
.
Multisets. Two ordinary sets are identical if they have the same
elements, so for instance, ¦a, a, b¦ and ¦a, b¦ are the same set because
they have exactly the same elements, namely a and b. However, in
some applications it might be useful to allow repeated elements in a
set. In that case we use multisets, which are mathematical entities
similar to sets, but with possibly repeated elements. So, as multisets,
¦a, a, b¦ and ¦a, b¦ would be considered diﬀerent, since in the ﬁrst one
the element a occurs twice and in the second one it occurs only once.
3
Properly speaking, the universe of discourse of set theory is the collection of
all sets (which is not a set).
2.1. SET THEORY 21
2.1.2. Venn Diagrams. Venn diagrams are graphic representa
tions of sets as enclosed areas in the plane. For instance, in ﬁgure 2.1,
the rectangle represents the universal set (the set of all elements con
sidered in a given problem) and the shaded region represents a set A.
The other ﬁgures represent various set operations.
A
Figure 2.1. Venn Diagram.
B A
Figure 2.2. Intersection A ∩ B.
B A
Figure 2.3. Union A ∪ B.
2.1. SET THEORY 22
A
Figure 2.4. Complement A.
B
A
Figure 2.5. Diﬀerence A −B.
B
A
Figure 2.6. Symmetric Diﬀerence A⊕B.
2.1.3. Set Operations.
1. Intersection: The common elements of two sets:
A ∩ B = ¦x [ (x ∈ A) ∧ (x ∈ B)¦ .
If A ∩ B = ∅, the sets are said to be disjoint.
2. Union: The set of elements that belong to either of two sets:
A ∪ B = ¦x [ (x ∈ A) ∨ (x ∈ B)¦ .
2.1. SET THEORY 23
3. Complement: The set of elements (in the universal set) that do
not belong to a given set:
A = ¦x ∈ U [ x ∈ A¦ .
4. Diﬀerence or Relative Complement: The set of elements that
belong to a set but not to another:
A −B = ¦x [ (x ∈ A) ∧ (x ∈ B)¦ = A ∩ B.
5. Symmetric Diﬀerence: Given two sets, their symmetric diﬀer
ence is the set of elements that belong to either one or the other
set but not both.
A⊕B = ¦x [ (x ∈ A) ⊕(x ∈ B)¦ .
It can be expressed also in the following way:
A⊕B = A ∪ B −A ∩ B = (A −B) ∪ (B −A) .
2.1.4. Counting with Venn Diagrams. A Venn diagram with
n sets intersecting in the most general way divides the plane into 2
n
regions. If we have information about the number of elements of some
portions of the diagram, then we can ﬁnd the number of elements in
each of the regions and use that information for obtaining the number
of elements in other portions of the plane.
Example: Let M, P and C be the sets of students taking Mathe
matics courses, Physics courses and Computer Science courses respec
tively in a university. Assume [M[ = 300, [P[ = 350, [C[ = 450,
[M ∩ P[ = 100, [M ∩ C[ = 150, [P ∩ C[ = 75, [M ∩ P ∩ C[ = 10. How
many students are taking exactly one of those courses? (ﬁg. 2.7)
10
185
235
65
140
C
60
90
M P
Figure 2.7. Counting with Venn diagrams.
We see that [(M∩P)−(M∩P∩C)[ = 100−10 = 90, [(M∩C)−(M∩
P ∩C)[ = 150−10 = 140 and [(P ∩C) −(M∩P ∩C)[ = 75−10 = 65.
2.1. SET THEORY 24
Then the region corresponding to students taking Mathematics courses
only has cardinality 300−(90+10+140) = 60. Analogously we compute
the number of students taking Physics courses only (185) and taking
Computer Science courses only (235). The sum 60 + 185 + 235 = 480
is the number of students taking exactly one of those courses.
2.1.5. Properties of Sets. The set operations verify the follow
ing properties:
1. Associative Laws:
A ∪ (B ∪ C) = (A ∪ B) ∪ C
A ∩ (B ∩ C) = (A ∩ B) ∩ C
2. Commutative Laws:
A ∪ B = B ∪ A
A ∩ B = B ∩ A
3. Distributive Laws:
A ∪ (B ∩ C) = (A ∪ B) ∩ (A ∪ C)
A ∩ (B ∪ C) = (A ∩ B) ∪ (A ∩ C)
4. Identity Laws:
A ∪ ∅ = A
A ∩ U = A
5. Complement Laws:
A ∪ A = U
A ∩ A = ∅
6. Idempotent Laws:
A ∪ A = A
A ∩ A = A
7. Bound Laws:
A ∪ U = U
A ∩ ∅ = ∅
8. Absorption Laws:
A ∪ (A ∩ B) = A
A ∩ (A ∪ B) = A
9. Involution Law:
A = A
2.1. SET THEORY 25
10. 0/1 Laws:
∅ = U
U = ∅
11. DeMorgan’s Laws:
A ∪ B = A ∩ B
A ∩ B = A ∪ B
2.1.6. Generalized Union and Intersection. Given a collec
tion of sets A
1
, A
2
, . . . , A
N
, their union is deﬁned as the set of elements
that belong to at least one of the sets (here n represents an integer in
the range from 1 to N):
N
¸
n=1
A
n
= A
1
∪ A
2
∪ ∪ A
N
= ¦x [ ∃n(x ∈ A
n
)¦ .
Analogously, their intersection is the set of elements that belong to all
the sets simultaneously:
N
¸
n=1
A
n
= A
1
∩ A
2
∩ ∩ A
N
= ¦x [ ∀n(x ∈ A
n
)¦ .
These deﬁnitions can be applied to inﬁnite collections of sets as well.
For instance assume that S
n
= ¦kn [ k = 2, 3, 4, . . . ¦ = set of multiples
of n greater than n. Then
∞
¸
n=2
S
n
= S
2
∪ S
3
∪ S
4
∪ = ¦4, 6, 8, 9, 10, 12, 14, 15, . . . ¦
= set of composite positive integers .
2.1.7. Partitions. A partition of a set X is a collection S of non
overlapping non empty subsets of X whose union is the whole X. For
instance a partition of X = ¦1, 2, 3, 4, 5, 6, 7, 8, 9, 10¦ could be
S = ¦¦1, 2, 4, 8¦, ¦3, 6¦, ¦5, 7, 9, 10¦¦ .
Given a partition S of a set X, every element of X belongs to exactly
one member of S.
Example: The division of the integers Z into even and odd numbers
is a partition: S = ¦E, O¦, where E = ¦2n [ n ∈ Z¦, O = ¦2n +1 [ n ∈
Z¦.
2.1. SET THEORY 26
Example: The divisions of Z in negative integers, positive integers
and zero is a partition: S = ¦Z
+
, Z
−
, ¦0¦¦.
2.1.8. Ordered Pairs, Cartesian Product. An ordinary pair
¦a, b¦ is a set with two elements. In a set the order of the elements is
irrelevant, so ¦a, b¦ = ¦b, a¦. If the order of the elements is relevant,
then we use a diﬀerent object called ordered pair, represented (a, b).
Now (a, b) = (b, a) (unless a = b). In general (a, b) = (a
, b
) iﬀ a = a
and b = b
.
Given two sets A, B, their Cartesian product AB is the set of all
ordered pairs (a, b) such that a ∈ A and b ∈ B:
A B = ¦(a, b) [ (a ∈ A) ∧ (b ∈ B)¦ .
Analogously we can deﬁne triples or 3tuples (a, b, c), 4tuples (a, b, c, d),
. . . , ntuples (a
1
, a
2
, . . . , a
n
), and the corresponding 3fold, 4fold,. . . ,
nfold Cartesian products:
A
1
A
2
A
n
=
¦(a
1
, a
2
, . . . , a
n
) [ (a
1
∈ A
1
) ∧ (a
2
∈ A
2
) ∧ ∧ (a
n
∈ A
n
)¦ .
If all the sets in a Cartesian product are the same, then we can use
an exponent: A
2
= A A, A
3
= A A A, etc. In general:
A
n
= A A
(n times)
A.
An example of Cartesian product is the real plane R
2
, where R is
the set of real numbers (R is sometimes called real line).
2.2. FUNCTIONS 27
2.2. Functions
2.2.1. Correspondences. Suppose that to each element of a set
A we assign some elements of another set B. For instance, A = N,
B = Z, and to each element x ∈ N we assign all elements y ∈ Z such
that y
2
= x (ﬁg. 2.8).
1
8
9
10
0
2
3 4
5
6
−1
1
2
−2
−3
0
3
Z N
7
Figure 2.8. Correspondence x → ±
√
x.
This operation is called a correspondence.
2.2.2. Functions. A function or mapping f from a set A to a set
B, denoted f : A → B, is a correspondence in which to each element
x of A corresponds exactly one element y = f(x) of B (ﬁg. 2.9).
A B
Figure 2.9. Function.
Sometimes we represent the function with a diagram like this:
f : A → B
x → y
or
A
f
→ B
x → y
2.2. FUNCTIONS 28
For instance, the following represents the function from Z to Z
deﬁned by f(x) = 2x + 1:
f : Z →Z
x → 2x + 1
The element y = f(x) is called the image of x, and x is a preimage
of y. For instance, if f(x) = 2x + 1 then f(7) = 2 7 + 1 = 15. The
set A is the domain of f, and B is its codomain. If A
⊆ A, the image
of A
by f is f(A
) = ¦f(x) [ x ∈ A
¦, i.e., the subset of B consisting
of all images of elements of A
. The subset f(A) of B consisting of
all images of elements of A is called the range of f. For instance, the
range of f(x) = 2x + 1 is the set of all integers of the form 2x + 1 for
some integer x, i.e., all odd numbers.
Example: Two useful functions from R to Z are the following:
1. The ﬂoor function:
x = greatest integer less than or equal to x .
For instance: 2 = 2, 2.3 = 2, π = 3, −2.5 = −3.
2. The ceiling function:
x = least integer greater than or equal to x .
For instance: 2 = 2, 2.3 = 3, π = 4, −2.5 = −2.
Example: The modulus operator is the function mod : ZZ
+
→Z
deﬁned:
x mod y = remainder when x is divided by y.
For instance 23 mod 7 = 2 because 23 = 37+2, 59 mod 9 = 5 because
59 = 6 9 + 5, etc.
Graph: The graph of a function f : A → B is the subset of A B
deﬁned by G(f) = ¦(x, f(x)) [ x ∈ A¦ (ﬁg. 2.10).
2.2.3. Types of Functions.
1. OnetoOne or Injective: A function f : A → B is called one
toone or injective if each element of B is the image of at most
one element of A (ﬁg. 2.11):
∀x, x
∈ A, f(x) = f(x
) ⇒ x = x
.
2.2. FUNCTIONS 29
0
1
2
3
4
y
–2 –1 1 2
x
Figure 2.10. Graph of f(x) = x
2
.
For instance, f(x) = 2x from Z to Z is injective.
A B
Figure 2.11. Onetoone function.
2. Onto or Surjective: A function f : A → B is called onto or
surjective if every element of B is the image of some element of
A (ﬁg. 2.12):
∀y ∈ B, ∃x ∈ A such that y = f(x) .
For instance, f(x) = x
2
from R to R
+
∪ ¦0¦ is onto.
A B
Figure 2.12. Onto function.
3. OneToOne Correspondence or Bijective: A function f : A →
B is said to be a onetoone correspondence, or bijective, or a
2.2. FUNCTIONS 30
bijection, if it is onetoone and onto (ﬁg. 2.13). For instance,
f(x) = x + 3 from Z to Z is a bijection.
A B
Figure 2.13. Bijection.
2.2.4. Identity Function. Given a set A, the function 1
A
: A →
A deﬁned by 1
A
(x) = x for every x in A is called the identity function
for A.
2.2.5. Function Composition. Given two functions f : A → B
and g : B → C, the composite function of f and g is the function
g ◦ f : A → C deﬁned by (g ◦ f)(x) = g(f(x)) for every x in A:
A
x
f
g◦f
B
y=f(x)
g
C
z=g(y)=g(f(x))
For instance, if A = B = C = Z, f(x) = x + 1, g(x) = x
2
, then
(g ◦ f)(x) = f(x)
2
= (x +1)
2
. Also (f ◦ g)(x) = g(x) +1 = x
2
+1 (the
composition of functions is not commutative in general).
Some properties of function composition are the following:
1. If f : A → B is a function from A to B, we have that f ◦ 1
A
=
1
B
◦ f = f.
2. Function composition is associative, i.e., given three functions
A
f
→ B
g
→ C
h
→ D,
we have that h ◦ (g ◦ f) = (h ◦ g) ◦ f.
2.2. FUNCTIONS 31
Function iteration. If f : A → A is a function from A to A, then
it makes sense to compose it with itself: f
2
= f ◦ f. For instance, if
f : Z → Z is f(x) = 2x + 1, then f
2
(x) = 2(2x + 1) + 1 = 4x + 3.
Analogously we can deﬁne f
3
= f ◦f ◦f, and so on, f
n
= f ◦
(n times)
. . . ◦f.
2.2.6. Inverse Function. If f : A → B is a bijective function, its
inverse is the function f
−1
: B → A such that f
−1
(y) = x if and only
if f(x) = y.
For instance, if f : Z → Z is deﬁned by f(x) = x + 3, then its
inverse is f
−1
(x) = x −3.
The arrow diagram of f
−1
is the same as the arrow diagram of f
but with all arrows reversed.
A characteristic property of the inverse function is that f
−1
◦f = 1
A
and f ◦ f
−1
= 1
B
.
2.2.7. Operators. A function from AA to A is called a binary
operator on A. For instance the addition of integers is a binary oper
ator + : Z Z → Z. In the usual notation for functions the sum of
two integers x and y would be represented +(x, y). This is called preﬁx
notation. The inﬁx notation consists of writing the symbol of the bi
nary operator between its arguments: x+y (this is the most common).
There is also a postﬁx notation consisting of writing the symbol after
the arguments: x y +.
Another example of binary operator on Z is (x, y) → x y.
A monary or unary operator on A is a function from A to A. For
instance the change of sign x → −x on Z is a unary operator on Z. An
example of unary operator on R
∗
(nonzero real numbers) is x → 1/x.
2.3. RELATIONS 32
2.3. Relations
2.3.1. Relations. Assume that we have a set of men M and a set
of women W, some of whom are married. We want to express which
men in M are married to which women in W. One way to do that is by
listing the set of pairs (m, w) such that m is a man, w is a woman, and
m is married to w. So, the relation “married to” can be represented
by a subset of the Cartesian product M W. In general, a relation R
from a set A to a set B will be understood as a subset of the Cartesian
product A B, i.e., R ⊆ A B. If an element a ∈ A is related to an
element b ∈ B, we often write a Rb instead of (a, b) ∈ R.
The set
¦a ∈ A [ a Rb for some b ∈ B¦
is called the domain of R. The set
¦b ∈ B [ a Rb for some a ∈ A¦
is called the range of R. For instance, in the relation “married to”
above, the domain is the set of married men, and the range is the set
of married women.
If A and B are the same set, then any subset of A A will be a
binary relation in A. For instance, assume A = ¦1, 2, 3, 4¦. Then the
binary relation “less than” in A will be:
<
A
= ¦(x, y) ∈ A A [ x < y¦
= ¦(1, 2), (1, 3), (1, 4), (2, 3), (2, 4), (3, 4)¦ .
Notation: A set A with a binary relation R is sometimes represented
by the pair (A, R). So, for instance, (Z, ≤) means the set of integers
together with the relation of nonstrict inequality.
2.3.2. Representations of Relations.
Arrow diagrams. Venn diagrams and arrows can be used for repre
senting relations between given sets. As an example, ﬁgure 2.14 rep
resents the relation from A = ¦a, b, c, d¦ to B = ¦1, 2, 3, 4¦ given by
R = ¦(a, 1), (b, 1), (c, 2), (c, 3)¦. In the diagram an arrow from x to y
means that x is related to y. This kind of graph is called directed graph
or digraph.
2.3. RELATIONS 33
a
b
c
1
2
3
4
d
A
B
Figure 2.14. Relation.
Another example is given in diagram 2.15, which represents the
divisibility relation on the set ¦1, 2, 3, 4, 5, 6, 7, 8, 9¦.
1
2
3
4
5
6
7
8
9
Figure 2.15. Binary relation of divisibility.
Matrix of a Relation. Another way of representing a relation R
from A to B is with a matrix. Its rows are labeled with the elements
of A, and its columns are labeled with the elements of B. If a ∈ A
and b ∈ B then we write 1 in row a column b if a Rb, otherwise we
write 0. For instance the relation R = ¦(a, 1), (b, 1), (c, 2), (c, 3)¦ from
A = ¦a, b, c, d¦ to B = ¦1, 2, 3, 4¦ has the following matrix:
1 2 3 4
a
b
c
d
¸
¸
¸
1 0 0 0
1 0 0 0
0 1 1 0
0 0 0 0
2.3.3. Inverse Relation. Given a relation R from A to B, the
inverse of R, denoted R
−1
, is the relation from B to A deﬁned as
b R
−1
a ⇔ a Rb .
2.3. RELATIONS 34
For instance, if R is the relation “being a son or daughter of”, then
R
−1
is the relation “being a parent of”.
2.3.4. Composition of Relations. Let A, B and C be three sets.
Given a relation R from A to B and a relation S from B to C, then
the composition S ◦ R of relations R and S is a relation from A to C
deﬁned by:
a (S ◦ R) c ⇔ there exists some b ∈ B such that a Rb and b S c .
For instance, if R is the relation “to be the father of”, and S is the
relation “to be married to”, then S ◦ R is the relation “to be the father
in law of”.
2.3.5. Properties of Binary Relations. A binary relation R on
A is called:
1. Reﬂexive if for all x ∈ A, x Rx. For instance on Z the relation
“equal to” (=) is reﬂexive.
2. Transitive if for all x, y, z ∈ A, x Ry and y Rz implies x Rz.
For instance equality (=) and inequality (<) on Z are transitive
relations.
3. Symmetric if for all x, y ∈ A, x Ry ⇒ y Rx. For instance on Z,
equality (=) is symmetric, but strict inequality (<) is not.
4. Antisymmetric if for all x, y ∈ A, x Ry and y Rx implies x = y.
For instance, nonstrict inequality (≤) on Z is antisymmetric.
2.3.6. Partial Orders. A partial order, or simply, an order on a
set A is a binary relation “” on A with the following properties:
1. Reﬂexive: for all x ∈ A, x x.
2. Antisymmetric: (x y) ∧ (y x) ⇒ x = y.
3. Transitive: (x y) ∧ (y z) ⇒ x z.
Examples:
1. The nonstrict inequality (≤) in Z.
2. Relation of divisibility on Z
+
: a[b ⇔ ∃t, b = at.
2.3. RELATIONS 35
3. Set inclusion (⊆) on P(A) (the collection of subsets of a given
set A).
Exercise: prove that the aforementioned relations are in fact partial
orders. As an example we prove that integer divisibility is a partial
order:
1. Reﬂexive: a = a 1 ⇒ a[a.
2. Antisymmetric: a[b ⇒ b = at for some t and b[a ⇒ a = bt
for
some t
. Hence a = att
, which implies tt
= 1 ⇒ t
= t
−1
. The
only invertible positive integer is 1, so t = t
= 1 ⇒ a = b.
3. Transitive: a[b and b[c implies b = at for some t and c = bt
for
some t
, hence c = att
, i.e., a[c.
Question: is the strict inequality (<) a partial order on Z?
Two elements a, b ∈ A are said to be comparable if either x y
or y x, otherwise they are said to be non comparable. The order
is called total or linear when every pair of elements x, y ∈ A are com
parable. For instance (Z, ≤) is totally ordered, but (Z
+
, [), where “[”
represents integer divisibility, is not. A totally ordered subset of a par
tially ordered set is called a chain; for instance the set ¦1, 2, 4, 8, 16, . . . ¦
is a chain in (Z
+
, [).
2.3.7. Hasse diagrams. A Hasse diagram is a graphical represen
tation of a partially ordered set in which each element is represented
by a dot (node or vertex of the diagram). Its immediate successors are
placed above the node and connected to it by straight line segments. As
an example, ﬁgure 2.16 represents the Hasse diagram for the relation
of divisibility on ¦1, 2, 3, 4, 5, 6, 7, 8, 9¦.
Question: How does the Hasse diagram look for a totally ordered
set?
2.3.8. Equivalence Relations. An equivalence relation on a set
A is a binary relation “∼” on A with the following properties:
1. Reﬂexive: for all x ∈ A, x ∼ x.
2. Symmetric: x ∼ y ⇒ y ∼ x.
3. Transitive: (x ∼ y) ∧ (y ∼ z) ⇒ x ∼ z.
2.3. RELATIONS 36
1
4
8
6
2
7
9
3
5
Figure 2.16. Hasse diagram for divisibility.
For instance, on Z, the equality (=) is an equivalence relation.
Another example, also on Z, is the following: x ≡ y (mod 2) (“x is
congruent to y modulo 2”) iﬀ x−y is even. For instance, 6 ≡ 2 (mod 2)
because 6 −2 = 4 is even, but 7 ≡ 4 (mod 2), because 7 −4 = 3 is not
even. Congruence modulo 2 is in fact an equivalence relation:
1. Reﬂexive: for every integer x, x−x = 0 is indeed even, so x ≡ x
(mod 2).
2. Symmetric: if x ≡ y (mod 2) then x − y = t is even, but
y −x = −t is also even, hence y ≡ x (mod 2).
3. Transitive: assume x ≡ y (mod 2) and y ≡ z (mod 2). Then
x −y = t and y −z = u are even. From here, x −z = (x −y) +
(y −z) = t +u is also even, hence x ≡ z (mod 2).
2.3.9. Equivalence Classes, Quotient Set, Partitions. Given
an equivalence relation ∼ on a set A, and an element x ∈ A, the
set of elements of A related to x are called the equivalence class of
x, represented [x] = ¦y ∈ A [ y ∼ x¦. Element x is said to be a
representative of class
[x]. The collection of equivalence classes, represented A/∼ = ¦[x] [
x ∈ A¦, is called quotient set of A by ∼.
Exercise: Find the equivalence classes on Z with the relation of
congruence modulo 2.
One of the main properties of an equivalence relation on a set A
is that the quotient set, i.e. the collection of equivalence classes, is
a partition of A. Recall that a partition of a set A is a collection of
2.3. RELATIONS 37
nonempty subsets A
1
, A
2
, A
3
, . . . of A which are pairwise disjoint and
whose union equals A:
1. A
i
∩ A
j
= ∅ for i = j,
2.
¸
n
A
n
= A.
Example: in Z with the relation of congruence modulo 2 (call it
“∼
2
”), there are two equivalence classes: the set E of even integers and
the set O of odd integers. The quotient set of Z by the relation “∼
2
”
of congruence modulo 2 is Z/∼
2
= ¦E, O¦. We see that it is in fact a
partition of Z, because E ∩ O = ∅, and Z = E ∪ O.
Exercise: Let m be an integer greater than or equal to 2. On Z
we deﬁne the relation x ≡ y (mod m) ⇔ m[(y − x) (i.e., m divides
exactly y −x). Prove that it is an equivalence relation. What are the
equivalence classes? How many are there?
Exercise: On the Cartesian product Z Z
∗
we deﬁne the relation
(a, b) R(c, d) ⇔ ad = bc. Prove that R is an equivalence relation.
Would it still be an equivalence relation if we extend it to Z Z?
CHAPTER 3
Algorithms, Integers
3.1. Algorithms
Consider the following list of instructions to ﬁnd the maximum of
three numbers a, b, c:
1. Assign variable x the value of a.
2. If b > x then assign x the value of b.
3. If c > x then assign x the value of c.
4. Output the value of x.
After executing those steps the output will be the maximum of a, b, c.
In general an algorithm is a ﬁnite list of instructions with the fol
lowing characteristics:
1. Precision. The steps are precisely stated.
2. Uniqueness. The result of executing each step is uniquely de
termined by the inputs and the result of preceding steps.
3. Finiteness. The algorithm stops after ﬁnitely many instructions
have been executed.
4. Input. The algorithm receives input.
5. Output. The algorithm produces output.
6. Generality. The algorithm applies to a set of inputs.
Basically an algorithm is the idea behind a program. Conversely,
programs are implementations of algorithms.
3.1.1. Pseudocode. Pseudocode is a language similar to a pro
gramming language used to represent algorithms. The main diﬀerence
respect to actual programming languages is that pseudocode is not re
quired to follow strict syntactic rules, since it is intended to be just
read by humans, not actually executed by a machine.
38
3.1. ALGORITHMS 39
Usually pseudocode will look like this:
procedure ProcedureName(Input)
Instructions...
end ProcedureName
For instance the following is an algorithm to ﬁnd the maximum of
three numbers a, b, c:
1: procedure max(a,b,c)
2: x := a
3: if b>x then
4: x := b
5: if c>x then
6: x := c
7: return x
8: end max
Next we show a few common operations in pseudocode.
The following statement means “assign variable x the value of vari
able y:
x := y
The following code executes “action” if condition “p” is true:
if p then
action
The following code executes “action1” if condition “p” is true, oth
erwise it executes “action2”:
if p then
action1
else
action2
The following code executes “action” while condition “p” is true:
1: while p
2: action
The following is the structure of a for loop:
3.1. ALGORITHMS 40
for var := init to limit
action
If an action contains more than one statement then we must enclose
them in a block:
begin
Instruction1
Instruction2
Instruction3
...
end
Comments are enclose between brackets:
¦This is a comment¦
The output of a procedure is returned with a return statement:
return output
Procedures that do not return anything are invoked with a call
statement:
call Procedure(arguments...)
As an example, the following procedure returns the largest number
in a sequence s
1
, s
2
, . . . s
n
represented as an array with n elements:
s[1], s[2],. . . , s[n]:
1: procedure largest element(s,n)
2: largest := s[1]
3: for k := 2 to n
4: if s[k] > largest then
5: largest := s[k]
6: return largest
7: end largest element
3.1.2. Recursiveness.
Recursive Deﬁnitions. A deﬁnition such that the object deﬁned oc
curs in the deﬁnition is called a recursive deﬁnition. For instance,
3.1. ALGORITHMS 41
consider the Fibonacci sequence
0, 1, 1, 2, 3, 5, 8, 13, . . .
It can be deﬁned as a sequence whose two ﬁrst terms are F
0
= 0,
F
1
= 1 and each subsequent term is the sum of the two previous ones:
F
n
= F
n−1
+F
n−2
(for n ≥ 2).
Other examples:
• Factorial:
1. 0! = 1
2. n! = n (n −1)! (n ≥ 1)
• Power:
1. a
0
= 1
2. a
n
= a
n−1
a (n ≥ 1)
In all these examples we have:
1. A Basis, where the function is explicitly evaluated for one or
more values of its argument.
2. A Recursive Step, stating how to compute the function from its
previous values.
Recursive Procedures. A recursive procedure is a procedure that in
vokes itself. Also a set of procedures is called recursive if they invoke
themselves in a circle, e.g., procedure p
1
invokes procedure p
2
, proce
dure p
2
invokes procedure p
3
and procedure p
3
invokes procedure p
1
.
A recursive algorithm is an algorithm that contains recursive proce
dures or recursive sets of procedures. Recursive algorithms have the
advantage that often they are easy to design and are closer to natural
mathematical deﬁnitions.
As an example we show two alternative algorithms for computing
the factorial of a natural number, the ﬁrst one iterative (non recursive),
the second one recursive.
1: procedure factorial iterative(n)
2: fact := 1
3: for k := 2 to n
4: fact := k * fact
5: return fact
6: end factorial iterative
3.1. ALGORITHMS 42
1: procedure factorial recursive(n)
2: if n = 0 then
3: return 1
4: else
5: return n * factorial recursive(n1)
6: end factorial recursive
While the iterative version computes n! = 1 2 . . . n directly, the
recursive version resembles more closely the formula n! = n (n −1)!
A recursive algorithm must contain at least a basic case without
recursive call (the case n = 0 in our example), and any legitimate
input should lead to a ﬁnite sequence of recursive calls ending up at
the basic case. In our example n is a legitimate input if it is a natural
number, i.e., an integer greater than or equal to 0. If n = 0 then
factorial recursive(0) returns 1 immediately without performing
any recursive call. If n > then the execution of
factorial recursive(n)
leads to a recursive call
factorial recursive(n1)
which will perform a recursive call
factorial recursive(n2)
and so on until eventually reaching the basic case
factorial recursive(0)
After reaching the basic case the procedure returns a value to the last
call, which returns a value to the previous call, and so on up to the
ﬁrst invocation of the procedure.
Another example is the following algorithm for computing the nth
element of the Fibonacci sequence:
3.1. ALGORITHMS 43
1: procedure fibonacci(n)
2: if n=0 then
3: return 0
4: if n=1 then
5: return 1
6: return fibonacci(n1) + fibonacci(n2)
7: end fibonacci
In this example we have two basic cases, namely n = 0 and n = 1.
In this particular case the algorithm is ineﬃcient in the sense that
it performs more computations than actually needed. For instance a
call to fibonacci(5) contains two recursive calls, one to fibonacci(4)
and another one to fibonacci(3). Then fibonacci(4) performs a call
to fibonacci(3) and another call to fibonacci(2), so at this point we
see that fibonacci(3) is being called twice, once inside fibonacci(5)
and again in fibonacci(4). Hence sometimes the price to pay for a
simpler algorithmic structure is a loss of eﬃciency.
However careful design may yield eﬃcient recursive algorithms. An
example is merge sort, and algorithm intended to sort a list of ele
ments. First let’s look at a simple non recursive sorting algorithm
called bubble sort. The idea is to go several times through the list
swapping adjacent elements if necessary. It applies to a list of numbers
s
i
, s
i+1
, . . . , s
j
represented as an array s[i], s[i+1],..., s[j]:
1: procedure bubble sort(s,i,j)
2: for p:=1 to ji
3: for q:=i to jp
4: if s[q] > s[q+1] then
5: swap(s[q],s[q+1])
6: end bubble sort
We can see that bubble sort requires n(n−1)/2 comparisons and
possible swapping operations.
On the other hand, the idea of merge sort is to split the list into
two approximately equal parts, sort them separately and then merge
them into a single list:
3.1. ALGORITHMS 44
1: procedure merge sort(s,i,j)
2: if i=j then
3: return
4: m := floor((i+j)/2)
5: call merge sort(s,i,m)
6: call merge sort(s,m+1,j)
7: call merge(s,i,m,j,c)
8: for k:=i to j
9: s[k] := c[k]
10: end merge sort
The procedure merge(s,i,m,j,c) merges the two increasing sequences
s
i
, s
i+1
, . . . , s
m
and s
m+1
, s
m+2
, . . . , s
j
into a single increasing sequence
c
i
, c
i+1
, . . . , c
j
. This algorithm is more eﬃcient than bubble sort be
cause it requires only about nlog
2
n operations (we will make this more
precise soon).
The strategy of dividing a task into several smaller tasks is called
divide and conquer.
3.1.3. Complexity. In general the complexity of an algorithm is
the amount of time and space (memory use) required to execute it.
Here we deal with time complexity only.
Since the actual time required to execute an algorithm depends on
the details of the program implementing the algorithm and the speed
and other characteristics of the machine executing it, it is in general
impossible to make an estimation in actual physical time, however it
is possible to measure the length of the computation in other ways,
say by the number of operations performed. For instance the following
loop performs the statement x := x + 1 exactly n times,
1: for i := 1 to n
2: x := x + 1
The following double loop performs it n
2
times:
1: for i := 1 to n
2: for j := 1 to n
3: x := x + 1
The following one performs it 1 + 2 + 3 + +n = n(n + 1)/2 times:
3.1. ALGORITHMS 45
1: for i := 1 to n
2: for j := 1 to i
3: x := x + 1
Since the time that takes to execute an algorithm usually depends
on the input, its complexity must be expressed as a function of the
input, or more generally as a function of the size of the input. Since
the execution time may be diﬀerent for inputs of the same size, we
deﬁne the following kinds of times:
1. Bestcase time: minimum time needed to execute the algorithm
among all inputs of a given size n.
2. Wostcase time: maximum time needed to execute the algo
rithm among all inputs of a given size n.
3. Averagecase time: average time needed to execute the algo
rithm among all inputs of a given size n.
For instance, assume that we have a list of n objects one of which is
colored red and the others are colored blue, and we want to ﬁnd the one
that is colored red by examining the objects one by one. We measure
time by the number of objects examined. In this problem the minimum
time needed to ﬁnd the red object would be 1 (in the lucky event that
the ﬁrst object examined turned out to be the red one). The maximum
time would be n (if the red object turns out to be the last one). The
average time is the average of all possible times: 1, 2, 3, . . . , n, which is
(1+2+3+ +n)/n = (n+1)/2. So in this example the bestcase time
is 1, the worstcase time is n and the averagecase time is (n + 1)/2.
Often the exact time is too hard to compute or we are interested
just in how it grows compared to the size of the input. For instance
and algorithm that requires exactly 7n
2
+3n+10 steps to be executed
on an input of size n is said to be or order n
2
, represented Θ(n
2
). This
justiﬁes the following notations:
Big Oh Notation. A function f(n) is said to be of order at most
g(n), written f(n) = O(g(n)), if there is a constant C
1
such that
[f(n)[ ≤ C
1
[g(n)[
for all but ﬁnitely many positive integers n.
3.1. ALGORITHMS 46
Omega Notation. A function f(n) is said to be of order at least
g(n), written f(n) = Ω(g(n)), if there is a constant C
2
such that
[f(n)[ ≥ C
2
[g(n)[
for all but ﬁnitely many positive integers n.
Theta Notation. A function f(n) is said to be of order g(n), written
f(n) = Θ(g(n)), if f(n) = O(g(n)) and f(n) = Ω(g(n)).
Remark: All logarithmic functions are of the same order: log
a
n =
Θ(log
b
n) for any a, b > 1, because log
a
n = log
b
n/ log
b
a, so they always
diﬀer in a multiplicative constant. As a consequence, if the execution
time of an algorithm is of order a logarithmic function, we can just say
that its time is “logarithmic”, we do not need to specify the base of
the logarithm.
The following are several common growth functions:
Order Name
Θ(1) Constant
Θ(log log n) Log log
Θ(log n) Logarithmic
Θ(nlog n) n log n
Θ(n) Linear
Θ(n
2
) Quadratic
Θ(n
3
) Cubic
Θ(n
k
) Polynomial
Θ(a
n
) Exponential
Let’s see now how we ﬁnd the complexity of algorithms like bubble sort
and merge sort.
Since bubble sort is just a double loop its complexity is easy to
ﬁnd; the inner loop is executed
(n −1) + (n −2) + + 1 = n(n −1)/2
times, so it requires n(n −1)/2 comparisons and possible swap opera
tions. Hence its execution time is Θ(n
2
).
The estimation of the complexity of merge sort is more involved.
First, the number of operations required by the merge procedure is
Θ(n). Next, if we call T(n) (the order of) the number of operations
3.1. ALGORITHMS 47
required by merge sort working on a list of size n, we see that roughly:
T(n) = 2T(n/2) +n.
Replacing n with n/2 we have T(n/2) = 2T(n/4) +n/2, hence
T(n) = 2T(n/2) +n = 2(2T(n/4) +n/2) +n = 4T(n/4) + 2n.
Repeating k times we get:
T(n) = 2
k
T(n/2
k
) +kn.
So for k = log
2
n we have
T(n) = nT(1) +nlog
2
n = Θ(nlog n) .
3.2. THE EUCLIDEAN ALGORITHM 48
3.2. The Euclidean Algorithm
3.2.1. The Division Algorithm. The following result is known
as The Division Algorithm:
1
If a, b ∈ Z, b > 0, then there exist unique
q, r ∈ Z such that a = qb + r, 0 ≤ r < b. Here q is called quotient of
the integer division of a by b, and r is called remainder.
3.2.2. Divisibility. Given two integers a, b, b = 0, we say that b
divides a, written b[a, if there is some integer q such that a = bq:
b[a ⇔ ∃q, a = bq .
We also say that b divides or is a divisor of a, or that a is a multiple
of b.
3.2.3. Prime Numbers. A prime number is an integer p ≥ 2
whose only positive divisors are 1 and p. Any integer n ≥ 2 that is not
prime is called composite. A nontrivial divisor of n ≥ 2 is a divisor d
of n such that 1 < d < n, so n ≥ 2 is composite iﬀ it has nontrivial
divisors. Warning: 1 is not considered either prime or composite.
Some results about prime numbers:
1. For all n ≥ 2 there is some prime p such that p[n.
2. (Euclid) There are inﬁnitely many prime numbers.
3. If p[ab then p[a or p[b. More generally, if p[a
1
a
2
. . . a
n
then p[a
k
for some k = 1, 2, . . . , n.
3.2.4. The Fundamental Theorem of Arithmetic. Every in
teger n ≥ 2 can be written as a product of primes uniquely, up to the
order of the primes.
It is customary to write the factorization in the following way:
n = p
s
1
1
p
s
2
2
. . . p
s
k
k
,
where all the exponents are positive and the primes are written so that
p
1
< p
2
< < p
k
. For instance:
13104 = 2
4
3
2
7 13 .
1
The result is not really an “algorithm”, it is just a mathematical theorem.
There are, however, algorithms that allow us to compute the quotient and the
remainder in an integer division.
3.2. THE EUCLIDEAN ALGORITHM 49
3.2.5. Greatest Common Divisor. A positive integer d is called
a common divisor of the integers a and b, if d divides a and b. The
greatest possible such d is called the greatest common divisor of a and b,
denoted gcd(a, b). If gcd(a, b) = 1 then a, b are called relatively prime.
Example: The set of positive divisors of 12 and 30 is ¦1, 2, 3, 6¦.
The greatest common divisor of 12 and 30 is gcd(12, 30) = 6.
A few properties of divisors are the following. Let m, n, d be
integers. Then:
1. If d[m and d[n then d[(m+n).
2. If d[m and d[n then d[(m−n).
3. If d[m then d[mn.
Another important result is the following: Given integers a, b, c, the
equation
ax +by = c
has integer solutions if and only if gcd(a, b) divides c. That is an
example of a Diophantine equation. In general a Diophantine equation
is an equation whose solutions must be integers.
Example: We have gcd(12, 30) = 6, and in fact we can write 6 =
1 30−2 12. The solution is not unique, for instance 6 = 3 30−7 12.
3.2.6. Finding the gcd by Prime Factorization. We have that
gcd(a, b) = product of the primes that occur in the prime factorizations
of both a and b, raised to their lowest exponent. For instance 1440 =
2
5
3
2
5, 1512 = 2
3
3
3
7, hence gcd(1440, 1512) = 2
3
3
2
= 72.
Factoring numbers is not always a simple task, so ﬁnding the gcd
by prime factorization might not be a most convenient way to do it,
but there are other ways.
3.2.7. The Euclidean Algorithm. Now we examine an alter
native method to compute the gcd of two given positive integers a, b.
The method provides at the same time a solution to the Diophantine
equation:
ax +by = gcd(a, b) .
It is based on the following fact: given two integers a ≥ 0 and
b > 0, and r = a mod b, then gcd(a, b) = gcd(b, r). Proof: Divide a by
3.2. THE EUCLIDEAN ALGORITHM 50
b obtaining a quotient q and a remainder r, then
a = bq +r , 0 ≤ r < b .
If d is a common divisor of a and b then it must be a divisor of r = a−bq.
Conversely, if d is a common divisor of b and r then it must divide
a = bq + r. So the set of common divisors of a and b and the set of
common divisors of b and r are equal, and the greatest common divisor
will be the same.
The Euclidean algorithm is a follows. First we divide a by b, obtain
ing a quotient q and a remainder r. Then we divide b by r, obtaining
a new quotient q
and a remainder r
. Next we divide r by r
, which
gives a quotient q
and another remainder r
. We continue dividing
each remainder by the next one until obtaining a zero remainder, and
which point we stop. The last nonzero remainder is the gcd.
Example: Assume that we wish to compute gcd(500, 222). Then we
arrange the computations in the following way:
500 = 2 222 + 56 → r = 56
222 = 3 56 + 54 → r
= 54
56 = 1 54 + 2 → r
= 2
54 = 27 2 + 0 → r
= 0
The last nonzero remainder is r
= 2, hence gcd(500, 222) = 2. Fur
thermore, if we want to express 2 as a linear combination of 500 and
222, we can do it by working backward:
2 = 56 −1 54 = 56 −1 (222 −3 56) = 4 56 −1 222
= 4 (500 −2 222) −1 222 = 4 500 −9 222 .
The algorithm to compute the gcd can be written as follows:
1: procedure gcd(a,b)
2: if a<b then ¦make a the largest¦
3: swap(a,b)
4: while b = 0
5: begin
6: r := a mod b
7: a := b
8: b := r
9: end
10: return a
11: end gcd
3.2. THE EUCLIDEAN ALGORITHM 51
The next one is a recursive version of the Euclidean algorithm:
1: procedure gcd recurs(a,b)
2: if b=0 then
3: return a
4: else
5: return gcd recurs(b,a mod b)
6: end gcd recurs
3.3. MODULAR ARITHMETIC, RSA ALGORITHM 52
3.3. Modular Arithmetic, RSA Algorithm
3.3.1. Congruences Modulo m. Given an integer m ≥ 2, we
say that a is congruent to b modulo m, written a ≡ b (mod m), if
m[(a −b). Note that the following conditions are equivalent
1. a ≡ b (mod m).
2. a = b +km for some integer k.
3. a and b have the same remainder when divided by m.
The relation of congruence modulo m is an equivalence relation. It
partitions Z into m equivalence classes of the form
[x] = [x]
m
= ¦x +km [ k ∈ Z¦ .
For instance, for m = 5, each one of the following rows is an equivalence
class:
. . . −10 −5 0 5 10 15 20 . . .
. . . −9 −4 1 6 11 16 21 . . .
. . . −8 −3 2 7 12 17 22 . . .
. . . −7 −2 3 8 13 18 23 . . .
. . . −6 −1 4 9 14 19 24 . . .
Each equivalence class has exactly a representative r such that 0 ≤ r <
m, namely the common remainder of all elements in that class when di
vided by m. Hence an equivalence class may be denoted [r] or x+mZ,
where 0 ≤ r < m. Often we will omit the brackets, so that the equiva
lence class [r] will be represented just r. The set of equivalence classes
(i.e., the quotient set of Z by the relation of congruence modulo m) is
denoted Z
m
= ¦0, 1, 2, . . . , m−1¦. For instance, Z
5
= ¦0, 1, 2, 3, 4¦.
Remark: When writing “r” as a notation for the class of r we may
stress the fact that r represents the class of r rather than the integer r
by including “ (mod p)” at some point. For instance 8 = 3 (mod p).
Note that in “a ≡ b (mod m)”, a and b represent integers, while in
“a = b (mod m)” they represent elements of Z
m
.
Reduction Modulo m: Once a set of representatives has been chosen
for the elements of Z
m
, we will call “r reduced modulo m”, written
“r mod m”, the chosen representative for the class of r. For instance,
if we choose the representatives for the elements of Z
5
in the interval
from 0 to 4 (Z
5
= ¦0, 1, 2, 3, 4¦), then 9 mod 5 = 4. Another possibility
is to choose the representatives in the interval from −2 to 2 (Z
5
=
¦−2, −1, 0, 1, 2¦), so that 9 mod 5 = −1
3.3. MODULAR ARITHMETIC, RSA ALGORITHM 53
In Z
m
it is possible to deﬁne an addition and a multiplication in
the following way:
[x] + [y] = [x +y] ; [x] [y] = [x y] .
As an example, tables 3.3.1 and 3.3.2 show the addition and multi
plication tables for Z
5
and Z
6
respectively.
+ 0 1 2 3 4
0 0 1 2 3 4
1 1 2 3 4 0
2 2 3 4 0 1
3 3 4 0 1 2
4 4 0 1 2 3
0 1 2 3 4
0 0 0 0 0 0
1 0 1 2 3 4
2 0 2 4 1 3
3 0 3 1 4 2
4 0 4 3 2 1
Table 3.3.1. Operational tables for Z
5
+ 0 1 2 3 4 5
0 0 1 2 3 4 5
1 1 2 3 4 5 0
2 2 3 4 5 0 1
3 3 4 5 0 1 2
4 4 5 0 1 2 3
5 5 0 1 2 3 4
0 1 2 3 4 5
0 0 0 0 0 0 0
1 0 1 2 3 4 5
2 0 2 4 0 2 4
3 0 3 0 3 0 3
4 0 4 2 0 4 2
5 0 5 4 3 2 1
Table 3.3.2. Operational tables for Z
6
A diﬀerence between this two tables is that in Z
5
every nonzero
element has a multiplicative inverse, i.e., for every x ∈ Z
5
such that
x = 0 there is an x
−1
such that x x
−1
= x
−1
x = 1; e.g. 2
−1
= 4
(mod 5). However in Z
6
that is not true, some nonzero elements like
2 have no multiplicative inverse. Furthermore the elements without
multiplicative inverse verify that they can be multiply by some other
nonzero element giving a product equal zero, e.g. 2 3 = 0 (mod 6).
These elements are called divisors of zero. Of course with this deﬁnition
zero itself is a divisor of zero. Divisors of zero diﬀerent from zero are
called proper divisors of zero. For instance in Z
6
2 is a proper divisor
of zero. In Z
5
there are no proper divisors of zero.
In general:
1. The elements of Z
m
can be classiﬁed into two classes:
3.3. MODULAR ARITHMETIC, RSA ALGORITHM 54
(a) Units: elements with multiplicative inverse.
(b) Divisors of zero: elements that multiplied by some other
nonzero element give product zero.
2. An element [a] ∈ Z
m
is a unit (has a multiplicative inverse) if
and only if gcd(a, m) = 1.
3. All nonzero elements of Z
m
are units if and only if m is a prime
number.
The set of units in Z
m
is denoted Z
∗
m
. For instance:
Z
∗
2
= ¦1¦
Z
∗
3
= ¦1, 2¦
Z
∗
4
= ¦1, 3¦
Z
∗
5
= ¦1, 2, 3, 4¦
Z
∗
6
= ¦1, 5¦
Z
∗
7
= ¦1, 2, 3, 4, 5, 6¦
Z
∗
8
= ¦1, 3, 5, 7¦
Z
∗
9
= ¦1, 2, 4, 5, 7, 8¦
Given an element [a] in Z
∗
m
, its inverse can be computed by using
the Euclidean algorithm to ﬁnd gcd(a, m), since that algorithm also
provides a solution to the equation ax +my = gcd(a, m) = 1, which is
equivalent to ax ≡ 1 (mod m).
Example: Find the multiplicative inverse of 17 in Z
∗
64
. Answer: We
use the Euclidean algorithm:
64 = 3 17 + 13 → r = 13
17 = 1 13 + 4 → r = 4
13 = 3 4 + 1 → r = 1
4 = 4 1 + 0 → r = 0
Now we compute backward:
1 = 13 −3 4 = 13 −3 (17 −1 13) = 4 13 −3 17
= 4 (64 −3 17) −3 17 = 4 64 −15 17 .
Hence (−15) 17 ≡ 1 (mod 64), but −15 ≡ 49 (mod 64), so the in
verse of 17 in (Z
∗
64
, ) is 49. We will denote this by writing 17
−1
= 49
(mod 64), or 17
−1
mod 64 = 49.
3.3. MODULAR ARITHMETIC, RSA ALGORITHM 55
3.3.2. Euler’s Phi Function. The number of units in Z
m
is equal
to the number of positive integers not greater than and relatively
prime to m, i.e., the number of integers a such that 1 ≤ a ≤ m and
gcd(a, m) = 1. That number is given by the so called Euler’s phi
function:
φ(m) = number of positive integers not greater than m
and relatively prime to m.
For instance, the positive integers not greater than and relatively prime
to 15 are: 1, 2, 4, 7, 8, 11, 13, 14, hence φ(15) = 8.
We have the following results:
1. If p is a prime number and s ≥ 1, then φ(p
s
) = p
s
− p
s−1
=
p
s
(1 −1/p). In particular φ(p) = p −1.
2. If m
1
, m
2
are two relatively prime positive integers, then φ(m
1
m
2
) =
φ(m
1
) φ(m
2
).
1
3. If m = p
s
1
1
p
s
2
2
. . . p
s
k
k
, where the p
k
are prime and the s
k
are
positive, then
φ(m) = m(1 −1/p
1
) (1 −1/p
2
) . . . (1 −1/p
k
) .
For instance
φ(15) = φ(3 5) = φ(3) φ(5) = (3 −1) (5 −1) = 2 4 = 8 .
3.3.3. Euler’s Theorem. If a and m are two relatively prime
positive integers, m ≥ 2, then
a
φ(m)
≡ 1 (mod m) .
The particular case in which m is a prime number p, Euler’s theorem
is called Fermat’s Little Theorem:
a
p−1
≡ 1 (mod p) .
For instance, if a = 2 and p = 7, then we have, in fact, 2
7−1
= 2
6
=
64 = 1 + 9 7 ≡ 1 (mod 7).
A consequence of Euler’s Theorem is the following. If gcd(a, m) = 1
then
x ≡ y (mod φ(m)) ⇒ a
x
≡ a
y
(mod m) .
1
A function f(x) of positive integers such that gcd(a, b) = 1 ⇒ f(ab) =
f(a)f(b) is called multiplicative.
3.3. MODULAR ARITHMETIC, RSA ALGORITHM 56
Consequently, the following function is well deﬁned:
Z
∗
m
Z
φ(m)
→Z
∗
m
([a]
m
, [x]
φ(m)
) → [a
x
]
m
Hence, we can compute powers modulo m in the following way:
a
n
= a
n mod φ(m)
(mod m) ,
if gcd(a, m) = 1. For instance:
3
9734888
mod 100 = 3
9734888 mod φ(100)
mod 100
= 3
9734888 mod 40
mod 100 = 3
8
mod 100 = 6561 mod 100 = 61 .
An even more eﬃcient way to compute powers modulo m is given
in Appendix A, paragraph A.1.
3.3.4. Application to Cryptography: RSA Algorithm. The
RSA algorithm is an encryption scheme designed in 1977 by Ronald
Rivest, Adi Shamir and Leonard Adleman. It allows encrypting a mes
sage with a key (the encryption key) and decrypting it with a diﬀerent
key (the decryption key). The encryption key is public and can be
given to everybody. The decryption key is private and is known only
by the recipient of the encrypted message.
The RSA algorithm is based on the following facts. Given two
prime numbers p and q, and a positive number m relatively prime to p
and q, Euler’s theorem tells us that:
m
φ(pq)
= m
(p−1)(q−1)
= 1 (mod pq) .
Assume now that we have two integers e and d such that e d = 1
(mod φ(pq)). Then we have that
(m
e
)
d
= m
e·d
= m (mod pq) .
So, given m
e
we can recover m modulo pq by raising to the dth power.
The RSA algorithm consists of the following:
1. Generate two large primes p and q. Find their product n = pq.
2. Find two numbers e and d (in the range from 2 to φ(n)) such
that e d = 1 (mod φ(n)). This requires some trial and error.
First e is chosen at random, and the Euclidean algorithm is
used to ﬁnd gcd(e, m), solving at the same time the equation
ex +my = gcd(e, m). If gcd(e, m) = 1 then the value obtained
3.3. MODULAR ARITHMETIC, RSA ALGORITHM 57
for x is d. Otherwise, e is no relatively prime to φ(n) and we
must try a diﬀerent value for e.
3. The public encryption key will be the pair (n, e). The private
decryption key will be the pair (n, d). The encryption key is
given to everybody, while the decryption key is kept secret by
the future recipient of the message.
4. The message to be encrypted is divided into small pieces, and
each piece is encoded numerically as a positive integer m smaller
than n.
5. The number m
e
is reduced modulo n; m
= m
e
mod n.
6. The recipient computes m
= m
d
mod n, with 0 ≤ m
< n.
It remains to prove that m
= m. If m is relatively prime to p and
q, then from Euler’s theorem we get that m
= m (mod n), and since
both are in the range from 0 to n −1 they must be equal. The case in
which p or q divides m is left as an exercise.
3.3.5. The Chinese Remainder Theorem. Let m
1
, m
2
, . . . , m
k
be pairwise relatively prime integers greater than or equal to 2. The
following system of congruences
x ≡ r
1
(mod m
1
)
x ≡ r
2
(mod m
2
)
. . .
x ≡ r
k
(mod m
k
)
has a unique solution modulo M = m
1
m
2
. . . m
k
.
We can ﬁnd a solution to that system in the following way. Let
M
i
= M/m
i
, and s
i
= the inverse of M
i
in Z
m
i
. Then
x = M
1
s
1
r
1
+M
2
s
2
r
2
+ +M
k
s
k
r
k
is a solution to the system.
Example: A group of objects can be arranged in 3 rows leaving 2
left, in 5 rows leaving 4 left, and in 7 rows leaving 6 left. How many
objects are there? Answer: We must solve the following system of
congruences:
x ≡ 2 (mod 3)
x ≡ 4 (mod 5)
x ≡ 6 (mod 7)
3.3. MODULAR ARITHMETIC, RSA ALGORITHM 58
We have: M = 3 5 7 = 105, M
1
= 105/3 = 35 ≡ 2 (mod 3),
M
2
= 105/5 = 21 ≡ 1 (mod 5), M
3
= 105/7 = 15 ≡ 1 (mod 7); s
1
=
“inverse of 2 in Z
3
” = 2, s
2
= “inverse of 1 in Z
5
” = 1, s
3
= “inverse
of 1 in Z
7
” = 1. Hence the solution is
x = 35 2 2 + 21 1 4 + 15 1 6 = 314 ≡ 104 (mod 105) .
Hence, any group of 104 + 105 k objects is a possible solution to the
problem.
CHAPTER 4
Induction, Recurences
4.1. Sequences and Strings
4.1.1. Sequences. A sequence is an (usually inﬁnite) ordered list
of elements. Examples:
1. The sequence of positive integers:
1, 2, 3, 4, . . . , n, . . .
2. The sequence of positive even integers:
2, 4, 6, 8, . . . , 2n, . . .
3. The sequence of powers of 2:
1, 2, 4, 8, 16, . . . , n
2
, . . .
4. The sequence of Fibonacci numbers (each one is the sum of the
two previous ones):
0, 1, 1, 2, 3, 5, 8, 13, . . .
5. The reciprocals of the positive integers:
1,
1
2
,
1
3
,
1
4
, ,
1
n
,
In general the elements of a sequence are represented with an in
dexed letter, say s
1
, s
2
, s
3
, . . . , s
n
, . . . . The sequence itself can be de
ﬁned by giving a rule, e.g.: s
n
= 2n + 1 is the sequence:
3, 5, 7, 9, . . .
Here we are assuming that the ﬁrst element is s
1
, but we can start at
any value of the index that we want, for instance if we declare s
0
to be
the ﬁrst term, the previous sequence would become:
1, 3, 5, 7, 9, . . .
The sequence is symbolically represented ¦s
n
¦ or ¦s
n
¦
∞
n=1
.
59
4.1. SEQUENCES AND STRINGS 60
If s
n
≤ s
n+1
for every n the sequence is called increasing. If s
n
≥
s
n+1
then it is called decreasing. For instance s
n
= 2n+1 is increasing:
3, 5, 7, 9, . . . , while s
n
= 1/n is decreasing: 1,
1
2
,
1
3
,
1
4
, .
If we remove elements from a sequence we obtain a subsequence.
E.g., if we remove all odd numbers from the sequence of positive inte
gers:
1, 2, 3, 4, 5 . . . ,
we get the subsequence consisting of the even positive integers:
2, 4, 6, 8, . . .
4.1.2. Sum (Sigma) and Product Notation. In order to ab
breviate sums and products the following notations are used:
1. Sum (or sigma) notation:
n
¸
i=m
a
i
= a
m
+a
m+1
+a
m+2
+ +a
n
2. Product notation:
n
¸
i=m
a
i
= a
m
a
m+1
a
m+2
a
n
For instance: assume a
n
= 2n + 1, then
6
¸
n=3
a
n
= a
3
+a
4
+a
5
+a
6
= 7 + 9 + 11 + 13 = 40 .
6
¸
n=3
a
n
= a
3
a
4
a
5
a
6
= 7 9 11 13 = 9009 .
4.1.3. Strings. Given a set X, a string over X is a ﬁnite ordered
list of elements of X.
Example: If X is the set X = ¦a, b, c¦, then the following are ex
amples of strings over X: aba, aaaa, bba, etc.
Repetitions can be speciﬁed with a superscripts, for instance: a
2
b
3
ac
2
a
3
=
aabbbaccaaa, (ab)
3
= ababab, etc.
The length of a string is its number of elements, e.g., [abaccbab[ = 8,
[a
2
b
7
a
3
c
6
[ = 18.
4.1. SEQUENCES AND STRINGS 61
The string with no elements is called null string, represented λ. Its
length is, of course, zero: [λ[ = 0.
The set of all strings over X is represented X
∗
. The set of no
null strings over X (i.e., all strings over X except the null string) is
represented X
+
.
Given two strings α and β over X, the string consisting of α followed
by β is called the concatenation of α and β. For instance if α = abac
and β = baaab then αβ = abacbaaab.
4.2. MATHEMATICAL INDUCTION 62
4.2. Mathematical Induction
Many properties of positive integers can be proved by mathematical
induction.
4.2.1. Principle of Mathematical Induction. Let P be a prop
erty of positive integers such that:
1. Basis Step: P(1) is true, and
2. Inductive Step: if P(n) is true, then P(n + 1) is true.
Then P(n) is true for all positive integers.
Remark: The premise P(n) in the inductive step is called Induction
Hypothesis.
The validity of the Principle of Mathematical Induction is obvious.
The basis step states that P(1) is true. Then the inductive step implies
that P(2) is also true. By the inductive step again we see that P(3)
is true, and so on. Consequently the property is true for all positive
integers.
Remark: In the basis step we may replace 1 with some other integer
m. Then the conclusion is that the property is true for every integer n
greater than or equal to m.
Example: Prove that the sum of the n ﬁrst odd positive integers is
n
2
, i.e., 1 + 3 + 5 + + (2n −1) = n
2
.
Answer: Let S(n) = 1 + 3 + 5 + + (2n −1). We want to prove
by induction that for every positive integer n, S(n) = n
2
.
1. Basis Step: If n = 1 we have S(1) = 1 = 1
2
, so the property is
true for 1.
2. Inductive Step: Assume (Induction Hypothesis) that the prop
erty is true for some positive integer n, i.e.: S(n) = n
2
. We must
prove that it is also true for n +1, i.e., S(n +1) = (n +1)
2
. In
fact:
S(n + 1) = 1 + 3 + 5 + + (2n + 1) = S(n) + 2n + 1 .
4.2. MATHEMATICAL INDUCTION 63
But by induction hypothesis, S(n) = n
2
, hence:
S(n + 1) = n
2
+ 2n + 1 = (n + 1)
2
.
This completes the induction, and shows that the property is true for
all positive integers.
Example: Prove that 2n + 1 ≤ 2
n
for n ≥ 3.
Answer: This is an example in which the property is not true for
all positive integers but only for integers greater than or equal to 3.
1. Basis Step: If n = 3 we have 2n + 1 = 2 3 + 1 = 7 and
2
n
= 2
3
= 8, so the property is true in this case.
2. Inductive Step: Assume (Induction Hypothesis) that the prop
erty is true for some positive integer n, i.e.: 2n + 1 ≤ 2
n
. We
must prove that it is also true for n+1, i.e., 2(n+1)+1 ≤ 2
n+1
.
By the induction hypothesis we know that 2n ≤ 2
n
, and we also
have that 3 ≤ 2
n
if n ≥ 3, hence
2(n + 1) + 1 = 2n + 3 ≤ 2
n
+ 2
n
= 2
n+1
.
This completes the induction, and shows that the property is true for
all n ≥ 3.
Exercise: Prove the following identities by induction:
• 1 + 2 + 3 + +n =
n(n + 1)
2
.
• 1
2
+ 2
2
+ 3
2
+ +n
2
=
n(n + 1) (2n + 1)
6
.
• 1
3
+ 2
3
+ 3
3
+ +n
3
= (1 + 2 + 3 + +n)
2
.
4.2.2. Strong Form of Mathematical Induction. Let P be a
property of positive integers such that:
1. Basis Step: P(1) is true, and
2. Inductive Step: if P(k) is true for all 1 ≤ k ≤ n then P(n + 1)
is true.
Then P(n) is true for all positive integers.
4.2. MATHEMATICAL INDUCTION 64
Example: Prove that every integer n ≥ 2 is prime or a product of
primes. Answer:
1. Basis Step: 2 is a prime number, so the property holds for
n = 2.
2. Inductive Step: Assume that if 2 ≤ k ≤ n, then k is a prime
number or a product of primes. Now, either n + 1 is a prime
number or it is not. If it is a prime number then it veriﬁes the
property. If it is not a prime number, then it can be written as
the product of two positive integers, n + 1 = k
1
k
2
, such that
1 < k
1
, k
2
< n + 1. By induction hypothesis each of k
1
and
k
2
must be a prime or a product of primes, hence n + 1 is a
product of primes.
This completes the proof.
4.2.3. The WellOrdering Principle. Every nonempty set of
positive integers has a smallest element.
Example: Prove that
√
2 is irrational (i.e.,
√
2 cannot be written as
a quotient of two positive integers) using the wellordering principle.
Answer: Assume that
√
2 is rational, i.e.,
√
2 = a/b, where a and
b are integers. Note that since
√
2 > 1 then a > b. Now we have
2 = a
2
/b
2
, hence 2 b
2
= a
2
. Since the left hand side is even, then
a
2
is even, but this implies that a itself is even, so a = 2 a
. Hence:
2 b
2
= 4 a
2
, and simplifying: b
2
= 2 a
2
. From here we see that
√
2 =
b/a
. Hence starting with a fractional representation of
√
2 = a/b
we end up with another fractional representation
√
2 = b/a
with a
smaller numerator b < a. Repeating the same argument with the
fraction b/a
we get another fraction with an even smaller numerator,
and so on. So the set of possible numerators of a fraction representing
√
2 cannot have a smallest element, contradicting the wellordering
principle. Consequently, our assumption that
√
2 is rational has to be
false.
4.3. RECURRENCE RELATIONS 65
4.3. Recurrence Relations
Here we look at recursive deﬁnitions under a diﬀerent point of view.
Rather than deﬁnitions they will be considered as equations that we
must solve. The point is that a recursive deﬁnition is actually a def
inition when there is one and only one object satisfying it, i.e., when
the equations involved in that deﬁnition have a unique solution. Also,
the solution to those equations may provide a closedform (explicit)
formula for the object deﬁned.
The recursive step in a recursive deﬁnition is also called a recurrence
relation. We will focus on kthorder linear recurrence relations, which
are of the form
C
0
x
n
+C
1
x
n−1
+C
2
x
n−2
+ +C
k
x
n−k
= b
n
,
where C
0
= 0. If b
n
= 0 the recurrence relation is called homogeneous.
Otherwise it is called nonhomogeneous.
The basis of the recursive deﬁnition is also called initial conditions
of the recurrence. So, for instance, in the recursive deﬁnition of the
Fibonacci sequence, the recurrence is
F
n
= F
n−1
+F
n−2
or
F
n
−F
n−1
−F
n−2
= 0 ,
and the initial conditions are
F
0
= 0, F
1
= 1 .
One way to solve some recurrence relations is by iteration, i.e., by
using the recurrence repeatedly until obtaining a explicit closeform
formula. For instance consider the following recurrence relation:
x
n
= r x
n−1
(n > 0) ; x
0
= A.
By using the recurrence repeatedly we get:
x
n
= r x
n−1
= r
2
x
n−2
= r
3
x
n−3
= = r
n
x
0
= Ar
n
,
hence the solution is x
n
= Ar
n
.
In the following we assume that the coeﬃcients C
0
, C
1
, . . . , C
k
are
constant.
4.3. RECURRENCE RELATIONS 66
4.3.1. First Order Recurrence Relations. The homogeneous
case can be written in the following way:
x
n
= r x
n−1
(n > 0) ; x
0
= A.
Its general solution is
x
n
= Ar
n
,
which is a geometric sequence with ratio r.
The nonhomogeneous case can be written in the following way:
x
n
= r x
n−1
+c
n
(n > 0) ; x
0
= A.
Using the summation notation, its solution can be expressed like this:
x
n
= Ar
n
+
n
¸
k=1
c
k
r
n−k
.
We examine two particular cases. The ﬁrst one is
x
n
= r x
n−1
+c (n > 0); x
0
= A.
where c is a constant. The solution is
x
n
= Ar
n
+c
n
¸
k=1
r
n−k
= Ar
n
+c
r
n
−1
r −1
if r = 1 ,
and
x
n
= A +c n if r = 1 .
Example: Assume that a country with currently 100 million people
has a population growth rate (birth rate minus death rate) of 1% per
year, and it also receives 100 thousand immigrants per year (which
are quickly assimilated and reproduce at the same rate as the native
population). Find its population in 10 years from now. (Assume that
all the immigrants arrive in a single batch at the end of the year.)
Answer: If we call x
n
= population in year n from now, we have:
x
n
= 1.01 x
n−1
+ 100, 000 (n > 0); x
0
= 100, 000, 000 .
This is the equation above with r = 1.01, c = 100, 000 and A =
100, 000, 00, hence:
x
n
= 100, 000, 000 1.01
n
+ 100, 000
1.01
n
−1
1.01 −1
= 100, 000, 000 1.01
n
+ 1000 (1.01
n
−1) .
4.3. RECURRENCE RELATIONS 67
So:
x
10
= 110, 462, 317 .
The second particular case is for r = 1 and c
n
= c + d n, where c
and d are constant (so c
n
is an arithmetic sequence):
x
n
= x
n−1
+c +d n (n > 0) ; x
0
= A.
The solution is now
x
n
= A +
n
¸
k=1
(c +d k) = A +c n +
d n(n + 1)
2
.
4.3.2. Second Order Recurrence Relations. Now we look at
the recurrence relation
C
0
x
n
+C
1
x
n−1
+C
2
x
n−2
= 0 .
First we will look for solutions of the form x
n
= c r
n
. By plugging in
the equation we get:
C
0
c r
n
+C
1
c r
n−1
+C
2
c r
n−2
= 0 ,
hence r must be a solution of the following equation, called the char
acteristic equation of the recurrence:
C
0
r
2
+C
1
r +C
2
= 0 .
Let r
1
, r
2
be the two (in general complex) roots of the above equation.
They are called characteristic roots. We distinguish three cases:
1. Distinct Real Roots. In this case the general solution of the
recurrence relation is
x
n
= c
1
r
n
1
+c
2
r
n
2
,
where c
1
, c
2
are arbitrary constants.
2. Double Real Root. If r
1
= r
2
= r, the general solution of the
recurrence relation is
x
n
= c
1
r
n
+c
2
nr
n
,
where c
1
, c
2
are arbitrary constants.
3. Complex Roots. In this case the solution could be expressed
in the same way as in the case of distinct real roots, but in
4.3. RECURRENCE RELATIONS 68
order to avoid the use of complex numbers we write r
1
= r e
αi
,
r
2
= r e
−αi
, k
1
= c
1
+c
2
, k
2
= (c
1
−c
2
) i, which yields:
1
x
n
= k
1
r
n
cos nα +k
2
r
n
sin nα.
Example: Find a closedform formula for the Fibonacci sequence
deﬁned by:
F
n+1
= F
n
+F
n−1
(n > 0) ; F
0
= 0, F
1
= 1 .
Answer: The recurrence relation can be written
F
n
−F
n−1
−F
n−2
= 0 .
The characteristic equation is
r
2
−r −1 = 0 .
Its roots are:
2
r
1
= φ =
1 +
√
5
2
; r
2
= −φ
−1
=
1 −
√
5
2
.
They are distinct real roots, so the general solution for the recurrence
is:
F
n
= c
1
φ
n
+c
2
(−φ
−1
)
n
.
Using the initial conditions we get the value of the constants:
(n = 0) c
1
+c
2
= 0
(n = 1) c
1
φ +c
2
(−φ
−1
) = 1
⇒
c
1
= 1/
√
5
c
2
= −1/
√
5
Hence:
F
n
=
1
√
5
¸
φ
n
−(−φ)
−n
¸
.
1
Remainder: e
αi
= cos α + i sin α.
2
φ =
1+
√
5
2
is the Golden Ratio.
CHAPTER 5
Counting
5.1. Basic Principles
5.1.1. The Rule of Sum. If a task can be performed in m ways,
while another task can be performed in n ways, and the two tasks
cannot be performed simultaneously, then performing either task can
be accomplished in m+n ways.
Set theoretical version of the rule of sum: If A and B are disjoint
sets (A ∩ B = ∅) then
[A ∪ B[ = [A[ +[B[ .
More generally, if the sets A
1
, A
2
, . . . , A
n
are pairwise disjoint, then:
[A
1
∪ A
2
∪ ∪ A
n
[ = [A
1
[ +[A
2
[ + +[A
n
[ .
For instance, if a class has 30 male students and 25 female students,
then the class has 30 + 25 = 45 students.
5.1.2. The Rule of Product. If a task can be performed in m
ways and another independent task can be performed in n ways, then
the combination of both tasks can be performed in mn ways.
Set theoretical version of the rule of product: Let A B be the
Cartesian product of sets A and B. Then:
[A B[ = [A[ [B[ .
More generally:
[A
1
A
2
A
n
[ = [A
1
[ [A
2
[ [A
n
[ .
For instance, assume that a license plate contains two letters fol
lowed by three digits. How many diﬀerent license plates can be printed?
Answer: each letter can be printed in 26 ways, and each digit can be
printed in 10 ways, so 26 26 10 10 10 = 676000 diﬀerent plates can
be printed.
69
5.1. BASIC PRINCIPLES 70
Exercise: Given a set A with m elements and a set B with n ele
ments, ﬁnd the number of functions from A to B.
5.1.3. The InclusionExclusion Principle. The inclusionexclusion
principle generalizes the rule of sum to nondisjoint sets.
In general, for arbitrary (but ﬁnite) sets A, B:
[A ∪ B[ = [A[ +[B[ −[A ∩ B[ .
Example: Assume that in a university with 1000 students, 200 stu
dents are taking a course in mathematics, 300 are taking a course in
physics, and 50 students are taking both. How many students are
taking at least one of those courses?
Answer: If U = total set of students in the university, M = set
of students taking Mathematics, P = set of students taking Physics,
then:
[M ∪ P[ = [M[ +[P[ −[M ∩ P[ = 300 + 200 −50 = 450
students are taking Mathematics or Physics.
For three sets the following formula applies:
[A ∪ B ∪ C[ =
[A[ +[B[ +[C[ −[A ∩ B[ −[A ∩ C[ −[B ∩ C[ +[A ∩ B ∩ C[ ,
and for an arbitrary union of sets:
[A
1
∪ A
2
∪ ∪ A
n
[ = s
1
−s
2
+s
3
−s
4
+ ±s
n
,
where s
k
= sum of the cardinalities of all possible kfold intersections
of the given sets.
5.2. COMBINATORICS 71
5.2. Combinatorics
5.2.1. Permutations. Assume that we have n objects. Any ar
rangement of any k of these objects in a given order is called a per
mutation of size k. If k = n then we call it just a permutation of the
n objects. For instance, the permutations of the letters a, b, c are the
following: abc, acb, bac, bca, cab, cba. The permutations of size 2 of
the letters a, b, c, d are: ab, ac, ad, ba, bc, bd, ca, cb, cd, da, db, dc.
Note that the order is important. Given two permutations, they
are considered equal if they have the same elements arranged in the
same order.
We ﬁnd the number P(n, k) of permutations of size k of n given
objects in the following way: The ﬁrst object in an arrangement can
be chosen in n ways, the second one in n − 1 ways, the third one in
n −2 ways, and so on, hence:
P(n, k) = n (n −1)
(k factors)
(n −k + 1) =
n!
(n −k)!
,
where n! = 1 2 3
(n factors)
n is called “n factorial ”.
The number P(n, k) of permutations of n objects is
P(n, n) = n! .
By convention 0! = 1.
For instance, there are 3! = 6 permutations of the 3 letters a, b, c.
The number of permutations of size 2 of the 4 letters a, b, c, d is P(4, 2) =
4 3 = 12.
Exercise: Given a set A with m elements and a set B with n ele
ments, ﬁnd the number of onetoone functions from A to B.
5.2.2. Combinations. Assume that we have a set A with n ob
jects. Any subset of A of size r is called a combination of n ele
ments taken r at a time. For instance, the combinations of the letters
a, b, c, d, e taken 3 at a time are: abc, abd, abe, acd, ace, ade, bcd, bce,
bde, cde, where two combinations are considered identical if they have
the same elements regardless of their order.
5.2. COMBINATORICS 72
The number of subsets of size r in a set A with n elements is:
C(n, r) =
n!
r! (n −r)!
.
The symbol
n
r
(read “n choose r”) is often used instead of C(n, r).
One way to derive the formula for C(n, r) is the following. Let A
be a set with n objects. In order to generate all possible permutations
of size r of the elements of A we 1) take all possible subsets of size
r in the set A, and 2) permute the k elements in each subset in all
possible ways. Task 1) can be performed in C(n, r) ways, and task
2) can be performed in P(r, r) ways. By the product rule we have
P(n, r) = C(n, r) P(r, r), hence
C(n, r) =
P(n, r)
P(r, r)
=
n!
r! (n −r)!
.
5.3. GENERALIZED PERMUTATIONS AND COMBINATIONS 73
5.3. Generalized Permutations and Combinations
5.3.1. Permutations with Repeated Elements. Assume that
we have an alphabet with k letters and we want to write all possible
words containing n
1
times the ﬁrst letter of the alphabet, n
2
times the
second letter,. . . , n
k
times the kth letter. How many words can we
write? We call this number P(n; n
1
, n
2
, . . . , n
k
), where n = n
1
+ n
2
+
+n
k
.
Example: With 3 a’s and 2 b’s we can write the following 5letter
words: aaabb, aabab, abaab, baaab, aabba, ababa, baaba, abbaa, babaa,
bbaaa.
We may solve this problem in the following way, as illustrated with
the example above. Let us distinguish the diﬀerent copies of a letter
with subscripts: a
1
a
2
a
3
b
1
b
2
. Next, generate each permutation of this
ﬁve elements by choosing 1) the position of each kind of letter, then 2)
the subscripts to place on the 3 a’s, then 3) these subscripts to place on
the 2 b’s. Task 1) can be performed in P(5; 3, 2) ways, task 2) can be
performed in 3! ways, task 3) can be performed in 2!. By the product
rule we have 5! = P(5; 3, 2) 3! 2!, hence P(5; 3, 2) = 5!/3! 2!.
In general the formula is:
P(n; n
1
, n
2
, . . . , n
k
) =
n!
n
1
! n
2
! . . . n
k
!
.
5.3.2. Combinations with Repetition. Assume that we have a
set A with n elements. Any selection of r objects from A, where each
object can be selected more than once, is called a combination of n
objects taken r at a time with repetition. For instance, the combinations
of the letters a, b, c, d taken 3 at a time with repetition are: aaa, aab,
aac, aad, abb, abc, abd, acc, acd, add, bbb, bbc, bbd, bcc, bcd, bdd, ccc, ccd,
cdd, ddd. Two combinations with repetition are considered identical
if they have the same elements repeated the same number of times,
regardless of their order.
Note that the following are equivalent:
1. The number of combinations of n objects taken r at a time with
repetition.
5.3. GENERALIZED PERMUTATIONS AND COMBINATIONS 74
2. The number of ways r identical objects can be distributed among
n distinct containers.
3. The number of nonnegative integer solutions of the equation:
x
1
+x
2
+ +x
n
= r .
Example: Assume that we have 3 diﬀerent (empty) milk containers
and 7 quarts of milk that we can measure with a one quart measuring
cup. In how many ways can we distribute the milk among the three
containers? We solve the problem in the following way. Let x
1
, x
2
, x
3
be
the quarts of milk to put in containers number 1, 2 and 3 respectively.
The number of possible distributions of milk equals the number of non
negative integer solutions for the equation x
1
+ x
2
+ x
3
= 7. Instead
of using numbers for writing the solutions, we will use strokes, so for
instance we represent the solution x
1
= 2, x
2
= 1, x
3
= 4, or 2 + 1 + 4,
like this: [[ +[ +[[[[. Now, each possible solution is an arrangement of 7
strokes and 2 plus signs, so the number of arrangements is P(9; 7, 2) =
9!/7! 2! =
9
7
.
The general solution is:
P(n +r −1; r, n −1) =
(n +r −1)!
r! (n −1)!
=
n +r −1
r
.
5.4. BINOMIAL COEFFICIENTS 75
5.4. Binomial Coeﬃcients
5.4.1. Binomial Theorem. The following identities can be easily
checked:
(x +y)
0
= 1
(x +y)
1
= x +y
(x +y)
2
= x
2
+ 2 xy +y
2
(x +y)
3
= x
3
+ 3 x
2
y + 3 xy
2
+y
3
They can be generalized by the following formula, called the Binomial
Theorem:
(x +y)
n
=
n
¸
k=0
n
k
x
n−k
y
k
=
n
0
x
n
+
n
1
x
n−1
y +
n
2
x
n−2
y
2
+
+
n
n −1
xy
n−1
+
n
n
y
n
.
We can ﬁnd this formula by writing
(x +y)
n
= (x +y) (x +y)
(n factors)
(x +y) ,
expanding, and grouping terms of the form x
a
y
b
. Since there are n
factors of the form (x + y), we have a + b = n, hence the terms must
be of the form x
n−k
y
k
. The coeﬃcient of x
n−k
y
k
will be equal to the
number of ways in which we can select the y from any k of the factors
(and the x from the remaining n − k factors), which is C(n, k) =
n
k
.
The expression
n
k
is often called binomial coeﬃcient.
Exercise: Prove
n
¸
k=0
n
k
= 2
n
and
n
¸
k=0
(−1)
k
n
k
= 0 .
Hint: Apply the binomial theorem to (1 + 1)
2
and (1 −1)
2
.
5.4.2. Properties of Binomial Coeﬃcients. The binomial co
eﬃcients have the following properties:
1.
n
k
=
n
n −k
5.4. BINOMIAL COEFFICIENTS 76
2.
n + 1
k + 1
=
n
k
+
n
k + 1
The ﬁrst property follows easily from
n
k
=
n!
k!(n −k)!
.
The second property can be proved by choosing a distinguished
element a in a set A of n + 1 elements. The set A has
n+1
k+1
subsets
of size k + 1. Those subsets can be partitioned into two classes: that
of the subsets containing a, and that of the subsets not containing a.
The number of subsets containing a equals the number of subsets of
A −¦a¦ of size k, i.e.,
n
k
. The number of subsets not containing a is
the number of subsets of A − ¦a¦ of size k + 1, i.e.,
n
k+1
. Using the
sum principle we ﬁnd that in fact
n+1
k+1
=
n
k
+
n
k+1
.
5.4.3. Pascal’s Triangle. The properties shown in the previous
section allow us to compute binomial coeﬃcients in a simple way. Look
at the following triangular arrangement of binomial coeﬃcients:
0
0
1
0
1
1
2
0
2
1
2
2
3
0
3
1
3
2
3
3
4
0
4
1
4
2
4
3
4
4
We notice that each binomial coeﬃcient on this arrangement must
be the sum of the two closest binomial coeﬃcients on the line above it.
This together with
n
0
=
n
n
= 1, allows us to compute very quickly
the values of the binomial coeﬃcients on the arrangement:
1
1 1
1 2 1
1 3 3 1
1 4 6 4 1
This arrangement of binomial coeﬃcients is called Pascal’s Trian
gle.
1
1
Although it was already known by the Chinese in the XIV century.
5.5. THE PIGEONHOLE PRINCIPLE 77
5.5. The Pigeonhole Principle
5.5.1. The Pigeonhole Principle. The pigeonhole principle is
used for proving that a certain situation must actually occur. It says
the following: If n pigeonholes are occupied by m pigeons and m > n,
then at least one pigeonhole is occupied by more than one pigeon.
1
Example: In any given set of 13 people at least two of them have
their birthday during the same month.
Example: Let S be a set of eleven 2digit numbers. Prove that
S must have two elements whose digits have the same diﬀerence (for
instance in S = ¦10, 14, 19, 22, 26, 28, 49, 53, 70, 90, 93¦, the digits of
the numbers 28 and 93 have the same diﬀerence: 8 − 2 = 6, 9 − 3 =
6.) Answer: The digits of a twodigit number can have 10 possible
diﬀerences (from 0 to 9). So, in a list of 11 numbers there must be two
with the same diﬀerence.
Example: Assume that we choose three diﬀerent digits from 1 to
9 and write all permutations of those digits. Prove that among the
3digit numbers written that way there are two whose diﬀerence is a
multiple of 500. Answer: There are 9 8 7 = 504 permutations of
three digits. On the other hand if we divide the 504 numbers by 500
we can get only 500 possible remainders, so at least two numbers give
the same remainder, and their diﬀerence must be a multiple of 500.
Exercise: Prove that if we select n + 1 numbers from the set S =
¦1, 2, 3, . . . , 2n¦, among the numbers selected there are two such that
one is a multiple of the other one.
1
The Pigeonhole Principle (Schubfachprinzip) was ﬁrst used by Dirichlet in
Number Theory. The term pigeonhole actually refers to one of those oldfashioned
writing desks with thin vertical wooden partitions in which to ﬁle letters.
CHAPTER 6
Probability
6.1. Probability
6.1.1. Introduction. Assume that we perform an experiment such
as tossing a coin or rolling a die. The set of possible outcomes is called
the sample space of the experiment. An event is a subset of the sample
space. For instance, if we toss a coin three times, the sample space is
S = ¦HHH, HHT, HTH, HTT, THH, THT, TTH, TTT¦ .
The event “at least two heads in a row” would be the subset
E = ¦HHH, HHT, THH¦ .
If all possible outcomes of an experiment have the same likelihood of
occurrence, then the probability of an event A ⊂ S is given by Laplace’s
rule:
P(E) =
[E[
[S[
.
For instance, the probability of getting at least two heads in a row in
the above experiment is 3/8.
6.1.2. Probability Function. In general the likelihood of diﬀer
ent outcomes of an experiment may not be the same. In that case
the probability of each possible outcome x is a function P(x). This
function veriﬁes:
0 ≤ P(x) ≤ 1 for all x ∈ S
and
¸
x∈S
P(x) = 1 .
The probability of an event E ⊆ S will be
P(E) =
¸
x∈E
P(x)
78
6.1. PROBABILITY 79
Example: Assume that a die is loaded so that the probability of
obtaining n point is proportional to n. Find the probability of getting
an odd number when rolling that die.
Answer: First we must ﬁnd the probability function P(n) (n =
1, 2, . . . , 6). We are told that P(n) is proportional to n, hence P(n) =
kn. Since P(S) = 1 we have P(1)+P(2)+ P(6) = 1, i.e., k1+k2+
+k 6 = 21k = 1, so k = 1/21 and P(n) = n/21. Next we want to
ﬁnd the probability of E = ¦2, 4, 6¦, i.e. P(E) = P(2) +P(4) +P(6) =
2
21
+
4
21
+
6
21
=
12
21
.
6.1.3. Properties of probability. Let P be a probability func
tion on a sample space S. Then:
1. For every event E ⊆ S,
0 ≤ P(E) ≤ 1 .
2. P(∅) = 0, P(S) = 1.
3. For every event E ⊆ S, if E = is the complement of E (“not
E”) then
P(E) = 1 −P(E) .
4. If E
1
, E
2
⊆ S are two events, then
P(E
1
∪ E
2
) = P(E
1
) +P(E
2
) −P(E
1
∩ E
2
) .
In particular, if E
1
∩E
2
= ∅ (E
1
and E
2
are mutually exclusive,
i.e., they cannot happen at the same time) then
P(E
1
∪ E
2
) = P(E
1
) +P(E
2
) .
Example: Find the probability of getting a sum diﬀerent from 10 or
12 after rolling two dice. Answer: We can get 10 in 3 diﬀerent ways:
4+6, 5+5, 6+4, so P(10) = 3/36. Similarly we get that P(12) = 1/36.
Since they are mutually exclusive events, the probability of getting 10
or 12 is P(10) +P(12) = 3/36+1/36 = 4/36 = 1/9. So the probability
of not getting 10 or 12 is 1 −1/9 = 8/9.
6.1.4. Conditional Probability. The conditional probability of
an event E given F, represented P(E [ F), is the probability of E
assuming that F has occurred. It is like restricting the sample space
to F. Its value is
P(E [ F) =
P(E ∩ F)
P(F)
.
6.1. PROBABILITY 80
Example: Find the probability of obtaining a sum of 10 after rolling
two fair dice. Find the probability of that event if we know that at least
one of the dice shows 5 points.
Answer: We call E = “obtaining sum 10” and F = “at least one of
the dice shows 5 points”. The number of possible outcomes is 6 6 =
36. The event “obtaining a sum 10” is E = ¦(4, 6), (5, 5), (6, 4)¦, so
[E[ = 3. Hence the probability is P(E) = [E[/[S[ = 3/36 = 1/12.
Now, if we know that at least one of the dice shows 5 points then the
sample space shrinks to
F = ¦(1, 5), (2, 5), (3, 5), (4, 5), (5, 5), (6, 5), (5, 1), (5, 2), (5, 3), (5, 4), (5, 6)¦ ,
so [F[ = 11, and the ways to obtain a sum 10 are E ∩ F = ¦(5, 5)¦,
[E ∩F[ = 1, so the probability is P(E [ F) = P(E ∩F)/P(F) = 1/11.
6.1.5. Independent Events. Two events E and F are said to be
independent if the probability of one of them does not depend on the
other, e.g.:
P(E [ F) = P(E) .
In this circumstances:
P(E ∩ F) = P(E) P(F) .
Note that if E is independent of F then also F is independent of E,
e.g., P(F [ E) = P(F).
Example: Assume that the probability that a shooter hits a target
is p = 0.7, and that hitting the target in diﬀerent shots are independent
events. Find:
1. The probability that the shooter does not hit the target in one
shot.
2. The probability that the shooter does not hit the target three
times in a row.
3. The probability that the shooter hits the target at least once
after shooting three times.
Answer:
1. P(not hitting the target in one shot) = 1 −0.7 = 0.3.
2. P(not hitting the target three times in a row) = 0.3
3
= 0.027.
3. P(hitting the target at least once in three shots) = 1−0.027 =
0.973.
6.1. PROBABILITY 81
6.1.6. Bayes’ Theorem. Suppose that a sample space S is parti
tioned into n classes C
1
, C
2
, . . . , C
n
which are pairwise mutually exclu
sive and whose union ﬁlls the whole sample space. Then for any event
F we have
P(F) =
n
¸
i=1
P(F [ C
i
) P(C
i
)
and
P(C
j
[ F) =
P(F [ C
j
) P(C
j
)
P(F)
.
Example: In a country with 100 million people 100 thousand of
them have disease X. A test designed to detect the disease has a 99%
probability of detecting it when administered to a person who has it,
but it also has a 5% probability of giving a false positive when given to
a person who does not have it. A person is given the test and it comes
out positive. What is the probability that that person has the disease?
Answer: The classes are C
1
= “has the disease” and C
2
= “does
not have the disease”, and the event is F = “the test gives a positive”.
We have: [S[ = 100, 000, 000, [C
1
[ = 100, 000, [C
2
[ = 99, 900, 000,
hence P(C
1
) = [C
1
[/[S[ = 0.001, P(C
2
) = [C
2
[/[S[ = 0.999. Also
P(F [ C
1
) = 0.99, P(F [ C
2
) = 0.05. Hence:
P(F) = P(F [ C
1
) P(C
1
) +P(F [ C
2
) P(C
2
)
= 0.99 0.001 + 0.05 0.999 = 0.05094 ,
and by Bayes’ theorem:
P(C
1
[ F) =
P(F [ C
1
) P(C
1
)
P(F)
=
0.99 0.001
0.05094
= 0.019434628 ≈ 2%.
(So the test is really of little use when given to a random person—
however it might be useful in combination with other tests or other
evidence that the person might have the disease.)
CHAPTER 7
Graph Theory
7.1. Graphs
7.1.1. Graphs. Consider the following examples:
1. A road map, consisting of a number of towns connected with
roads.
2. The representation of a binary relation deﬁned on a given set.
The relation of a given element x to another element y is rep
resented with an arrow connecting x to y.
The former is an example of (undirected) graph. The latter is an
example of a directed graph or digraph.
a
b
c
d
Figure 7.1. Undirected Graph.
In general a graph G consists of two things:
1. The vertex set V , whose elements are called vertices, nodes or
points.
2. The edge set E or set of edges connecting pairs of vertices. If
the edges are directed then they are also called directed edges
or arcs. Each edge e ∈ E is associated with a pair of vertices.
82
7.1. GRAPHS 83
a
b
c
d
Figure 7.2. Directed Graph.
A graph is sometimes represented by the pair (V, E) (we assume V
and E ﬁnite).
If the graph is undirected and there is a unique edge e connecting x
and y we may write e = ¦x, y¦, so E can be regarded as set of unordered
pairs. In this context we may also write e = (x, y), understanding that
here (x, y) is not an ordered pair, but the name of an edge.
If the graph is directed and there is a unique edge e pointing from
x to y, then we may write e = (x, y), so E may be regarded as a set
of ordered pairs. If e = (x, y), the vertex x is called origin, source or
initial point of the edge e, and y is called the terminus, terminating
vertex or terminal point.
b
c
d
a
Figure 7.3. Graph with parallel edges.
Two vertices connected by an edge are called adjacent. They are
also the endpoints of the edge, and the edge is said to be incident to
each of its endpoints. If the graph is directed, an edge pointing from
vertex x to vertex y is said to be incident from x and incident to y. An
edge connecting a vertex to itself is called a loop. Two edges connecting
the same pair of points (and pointing in the same direction if the graph
is directed) are called parallel or multiple.
7.1. GRAPHS 84
A graph with neither loops nor multiple edges is called a simple
graph. If a graph has multiple edges but no loops then it is called a
multigraph. If it has loops (and possible also multiple edges) then it is
called a pseudograph.
The following table summarizes the graph terminology
Table 7.1.1. Graph Terminology
Type Edges Multiple Edges Allowed? Loops Allowed?
Simple graph indirected no no
Multigraph indirected yes no
Pseudograph indirected yes yes
Directed graph directed no yes
Directed multigraph directed yes yes
The degree of a vertex v, represented deg(v), is the number of edges
that contain it (loops are counted twice). A vertex of degree zero (not
connected to any other vertex) is called isolated. A vertex of degree 1
is called pendant.
The Handshaking Theorem. Let G = (V, E) be an undirected graph
with e edges. Then
2e =
¸
v∈V
deg (v) .
(This applies even if multiple edges and loops are present.)
In a graph with directed edges, the indegree of a vertex v, denoted
deg
−
(v), is the number of edges with v as their terminal vertex. The
outdegree of a vertex v, denoted deg
+
(v), is the number of edges with
v as their initial vertex. (Note that a loop at a vertex contributes 1 to
both the indegree and the outdegree of this vertex.)
Number of vertices of odd degree. An undirected graph has an even
number of vertices of odd degree. Proof: Let V
e
and V
o
respectively
the set of vertices of even degree and the set of vertices of odd degree
in an undirected graph G = (V, E). Then
2e =
¸
v∈V
deg (v) =
¸
v∈Ve
deg (v) +
¸
v∈Vo
deg (v) .
Since deg(v) is even for v ∈ V
e
, the ﬁrst sum in the right hand side of
the equality is even. The total sum must be 2e, which is even, so the
second sum must be even too. But its terms are all odd, so there must
be an even number of them.
7.1. GRAPHS 85
Sum of degrees in an directed graph. Let G = (V, E) be a directed
graph. Then
¸
v∈V
deg
−
(v) =
¸
v∈V
deg
+
(v) = [E[ .
A weighted graph is a graph whose edges have been labeled with
numbers. The length of a path in a weighted graph is the sum of the
weights of the edges in the path.
a
b
d
c
6
3
4
6
7
Figure 7.4. Weighted Graph.
7.1.2. Special Graphs. Here we examine a few special graphs.
The ncube: A graph with with 2
n
vertices labeled 0, 1, . . . , 2
n
− 1
so that two of them are connected with an edge if their binary repre
sentation diﬀers in exactly one bit.
000
001
010 011
100
101
110 111
Figure 7.5. 3cube.
Complete Graph: a simple undirected graph G such that every pair
of distinct vertices in G are connected by an edge. The complete graph
of n vertices is represented K
n
(ﬁg. 7.6). A complete directed graph is
a simple directed graph G = (V, E) such that every pair of distinct
vertices in G are connected by exactly one edge—so, for each pair of
distinct vertices, either (x, y) or (y, x) (but not both) is in E.
7.1. GRAPHS 86
a
b
c d
e
Figure 7.6. Complete graph K
5
.
Bipartite Graph: a graph G = (V, E) in which V can be partitioned
into two subsets V
1
and V
2
so that each edge in G connects some vertex
in V
1
to some vertex in V
2
. A bipartite simple graph is called complete
if each vertex in V
1
is connected to each vertex in V
2
. If [V
1
[ = m and
[V
2
[ = n, the corresponding complete bipartite graph is represented
K
m,n
(ﬁg. 7.7).
A graph is bipartite iﬀ its vertices can be colored with two colors
so that every edge connects vertices of diﬀerent color.
Question: Is the ncube bipartite. Hint: color in red all vertices
whose binary representation has an even number of 1’s, color in blue
the ones with an odd number of 1’s.
b
p
q
r
s
a
c
Figure 7.7. Complete bipartite graph K
3,4
.
Regular Graph: a simple graph whose vertices have all the same
degree. For instance, the ncube is regular.
7.1.3. Subgraph. Given a graph G = (V, E), a subgraph G
=
(V
, E
) of G is another graph such that V
⊆ V and E
⊆ E. If
V
= V then G
is called a spanning subgraph of G.
7.1. GRAPHS 87
Given a subset of vertices U ⊆ V , the subgraph of G induced by
U, denoted 'U`, is the graph whose vertex set is U, and its edge set
contains all edges from G connecting vertices in U.
7.2. REPRESENTATIONS OF GRAPHS 88
7.2. Representations of Graphs
7.2.1. Adjacency matrix. The adjacency matrix of a graph is a
matrix with rows and columns labeled by the vertices and such that
its entry in row i, column j, i = j, is the number of edges incident on i
and j. For instance the following is the adjacency matrix of the graph
of ﬁgure 7.8:
1
a b c d
a
b
c
d
¸
¸
¸
0 1 0 1
1 0 2 0
0 2 0 0
1 0 0 1
b
a
d
c
e1
e2
e3
e5
e4
Figure 7.8
One of the uses of the adjacency matrix A of a simple graph G is
to compute the number of paths between two vertices, namely entry
(i, j) of A
n
is the number of paths of length n from i to j.
7.2.2. Incidence matrix. The incidence matrix of a graph G is a
matrix with rows labeled by vertices and columns labeled by edges, so
that entry for row v column e is 1 if e is incident on v, and 0 otherwise.
As an example, the following is the incidence matrix of graph of ﬁgure
7.8:
1
For some authors if i = j then the entry is twice the number of loops incident
on i; so in the example of ﬁgure 7.8 entry (d, d) would be 2 instead of 1.
7.2. REPRESENTATIONS OF GRAPHS 89
e
1
e
2
e
3
e
4
e
5
a
b
c
d
¸
¸
¸
1 0 0 1 0
1 1 1 0 0
0 1 1 0 0
0 0 0 1 1
7.2.3. Graph Isomorphism. Two graphs G
1
= (V
1
, E
1
), G
2
=
(V
2
, E
2
), are called isomorphic if there is a bijection f : V
1
→ V
2
and a
bijection g : E
1
→ E
2
such that an edge e is adjacent to vertices v and
w if and only if g(e) is adjacent to f(v) and f(w) (ﬁg. 7.9).
a1
b1
c1
d1
e1
a2 b2
c2
d2
e2
Figure 7.9. Two isomorphic graphs.
Two graphs are isomorphic if and only if for some ordering of their
vertices their adjacency matrices are equal.
An invariant is a property such that if a graph has it then all graphs
isomorphic to it also have it. Examples of invariants are their number
of vertices, their number of edges, “has a vertex of degree k”, “has a
simple cycle of length l”, etc. It is possible to prove that two graphs are
not isomorphic by showing an invariant property that one has and the
other one does not have. For instance the graphs in ﬁgure 7.10 cannot
be isomorphic because one has a vertex of degree 2 and the other one
doesn’t.
7.2. REPRESENTATIONS OF GRAPHS 90
a
b
c
e
f
3
4
5 1
2
Figure 7.10. Non isomorphic graphs.
7.3. PATHS AND CIRCUITS 91
7.3. Paths and Circuits
7.3.1. Paths. A path from v
0
to v
n
of length n is a sequence of
n+1 vertices (v
k
) and n edges (e
k
) of the formv
0
, e
1
, v
1
, e
2
, v
2
, . . . , e
n
, v
n
,
where each edge e
k
connects v
k−1
with v
k
(and points from v
k−1
to v
k
if the edge is directed). The path may be speciﬁed by giving only the
sequence of edges e
1
, . . . , e
n
. If there are no multiple edges we can
specify the path by giving only the vertices: v
0
, v
1
, . . . , v
n
. The path
is a circuit (or cycle) if it begins and ends at the same vertex, i.e.,
v
0
= v
n
, and has lenght greater than zero. A path or circuit is simple
if it does not contain the same edge twice.
7.3.2. Connected Graphs. A graph Gis called connected if there
is a path between any two distinct vertices of G. Otherwise the graph
is called disconnected. A directed graph is connected if its associated
undirected graph (obtained by ignoring the directions of the edges) is
connected. A directed graph is strongly connected if for every pair of
distict points u, v, there is a path from u to v and there is a path
from v to u. A connected component of G is any connected subgraph
G
= (V
, E
) of G = (V, E) such that there is not edge (in G) from a
vertex in V to a vertex in V −V
. Given a vertex in G, the component
of G containing v is the subgraph G
of G consisting of all edges and
vertices of g contained in some path beginning at v.
7.3.3. The Seven Bridges of K¨ onigsberg. This is a classical
problem that started the discipline today called graph theory.
During the eighteenth century the city of K¨onigsberg (in East Prus
sia)
1
was divided into four sections, including the island of Kneiphop,
by the Pregel river. Seven bridges connected the regions, as shown in
ﬁgure 7.11. It was said that residents spent their Sunday walks trying
to ﬁnd a way to walk about the city so as to cross each bridge exactly
once and then return to the starting point. The ﬁrst person to solve
the problem (in the negative) was the Swiss mathematician Leonhard
Euler in 1736. He represented the sections of the city and the seven
bridges by the graph of ﬁgure 7.12, and proved that it is impossible to
ﬁnd a path in it that transverses every edge of the graph exactly once.
In the next section we study why this is so.
1
The city is currently called Kaliningrad and is part of the Russian republic.
7.3. PATHS AND CIRCUITS 92
Figure 7.11. The Seven Bridges of K¨onigsberg.
Figure 7.12. Graph for the Seven Bridges of K¨onigsberg.
7.3.4. Euler paths and circuits. Let G = (V, E) be a graph
with no isolated vertices. An Euler path in G is a simple path that
transverses every edge of the graph exactly once. Analogously, an
Euler circuit in G is a simple circuit that transverses every edge of the
graph exactly once.
Existence of Euler Paths and Circuits. The graphs that have an
Euler path can be characterized by looking at the degree of their ver
tices. Recall that the degree of a vertex v, represented deg(v), is the
number of edges that contain v (loops are counted twice). An even
vertex is a vertex with even degree; an odd vertex is a vertex with odd
degree. The sum of the degrees of all vertices in a graph equals twice
its number of edges, so it is an even number. As a consequence, the
number of odd vertices in a graph is always even.
Let G be a connected multigraph. Then G contains an Euler circuit
if and only if G its vertices have even degree. Also, G contains an Euler
path from vertex a to vertex b (= a) if and only if a and b have odd
degree, and all its other vertices have even degree.
7.3. PATHS AND CIRCUITS 93
7.3.5. Hamilton Circuits. A Hamilton circuit in a graph G is
a circuit that contains each vertex of G once (except for the starting
and ending vertex, which occurs twice). A Hamilton path in G is a
path (not a circuit) that contains each vertex of G once. Note that by
deleting an edge in a Hamilton circuit we get a Hamilton path, so if a
graph has a Hamilton circuit, then it also has a Hamilton path. The
converse is not true, i.e., a graph may have a Hamilton path but not
a Hamilton circuit. Exercise: Find a graph with a Hamilton path but
no Hamilton circuit.
c
d
e
f
g
h
b
a
i
j
k
l
m
n
o
p
q
r
s t
Figure 7.13. Hamilton’s puzzle.
In general it is not easy to determine if a given graph has a Hamilton
path or circuit, although often it is possible to argue that a graph has no
Hamilton circuit. For instance if G = (V, E) is a bipartite graph with
vertex partition ¦V
1
, V
2
¦ (so that each edge in G connects some vertex
in V
1
to some vertex in V
2
), then G cannot have a Hamilton circuit if
[V
1
[ = [V
2
[, because any path must contain alternatively vertices from
V
1
and V
2
, so any circuit in G must have the same number of vertices
from each of both sets.
Edge removal argument. Another kind of argument consists of re
moving edges trying to make the degree of every vertex equal two. For
instance in the graph of ﬁgure 7.14 we cannot remove any edge because
that would make the degree of b, e or d less than 2, so it is impossi
ble to reduce the degree of a and c. Consequently that graph has no
Hamilton circuit.
Dirac’s Theorem. If G is a simple graph with n vertices with n ≥ 3
such that the degree of every vertex in G is at least n/2, then G has a
Hamilton circuit.
7.3. PATHS AND CIRCUITS 94
b
c
d
a
e
Figure 7.14. Graph without Hamilton circuit.
Ore’s Theorem. If G is a simple graph with n vertices with n ≥ 3
such that deg(u) +deg(v) ≥ n for every pair of nonadjacent vertices u
and v in G, then G has a Hamilton circuit.
The Traveling Salesperson Problem. Given a weighted graph, the
traveling salesperson problem (TSP) consists of ﬁnding a Hamilton cir
cuit of minimum length in this graph. The name comes from a classical
problem in which a salesperson must visit a number of cities and go
back home traveling the minimum distance possible. One way to solve
the problem consists of searching all possible Hamilton circuits and
computing their length, but this is very ineﬃcient. Unfortunately no
eﬃcient algorithm is known for solving this problem (and chances are
that none exists).
Remark: (Digression on P/NP problems.) Given a weighted graph
with n vertices the problem of determining whether it contains a Hamil
ton circuit of length not greater than a given L is known to be NP
complete. This means the following. First it is a decision problem, i.e.,
a problem whose solution is “yes” or “no”. A decision problem is said
to be polynomial, or belong to the class P, if it can be solved with an
algorithm of complexity O(n
k
) for some integer k. It is said to be non
deterministic polynomial, or belong to the class NP, if in all cases when
the answer is “yes” this can be determined with a nondeterministic
algorithm of complexity O(n
k
). A nondeterministic algorithm is an
algorithm that works with an extra hint, for instance in the TSP, if
G has a Hamilton circuit of length not greater than L the hint could
consist of a Hamilton circuit with length not greater than L—so the
task of the algorithm would be just to check that in fact that length
7.3. PATHS AND CIRCUITS 95
is not greater than L.
2
Currently it is not known whether the class
NP is strictly larger than the class P, although it is strongly suspected
that it is. The class NP contains a subclass called NPcomplete con
taining the “hardest” problems of the class, so that their complexity
must be higher than polynomial unless P=NP. The TSP is one of these
problems.
Gray Codes. A Gray code is a sequence s
1
, s
2
, . . . , s
2
n of nbinary
strings verifying the following conditions:
1. Every nbinary string appears somewhere in the sequence.
2. Two consecutive strings s
i
and s
i+1
diﬀer exactly in one bit.
3. s
2
n and s
1
diﬀer in exactly one bit.
For instance: 000, 001, 011, 010, 110, 111, 101, 100,
The problem of ﬁnding a gray code is equivalent to ﬁnding a Hamil
ton circuit in the ncube.
7.3.6. Dijkstra’s ShortestPath Algorithm. This is an algo
rithm to ﬁnd the shortest path from a vertex a to another vertex z in
a connected weighted graph. Edge (i, j) has weight w(i, j) > 0, and
vertex x is labeled L(x) (minimum distance from a if known, otherwise
∞). The output is L(z) = length of a minimum path from a to z.
1: procedure dijkstra(w,a,z,L)
2: L(a) := 0
3: for all vertices x = a
4: L(x) := ∞
5: T := set of all vertices
6: ¦T is the set of all vertices whose shortest¦
7: ¦distance from a has not been found yet¦
8: while z in T
9: begin
10: choose v in T with minimum L(v)
11: T := T  ¦v¦
12: for each x in T adjacent to v
13: L(x) := min¦L(x),L(v)+w(v,x)¦
2
Informally, P problems are “easy to solve”, and NP problems are problems
whose answer is “easy to check”. In a sense the P=NP problem consist of de
termining whether every problem whose solution is easy to check is also easy to
solve.
7.3. PATHS AND CIRCUITS 96
14: end
15: return L(z)
16: end dijkstra
For instance consider the graph in ﬁgure 7.15.
a
b c
z
d
e
f
2
1
3
2
3 1
2
4
4
Figure 7.15. Shortest path from a to z.
The algorithm would label the vertices in the following way in each
iteration (the boxed vertices are the ones removed from T):
iteration a b c d e f z
0 0 ∞ ∞ ∞ ∞ ∞ ∞
1 0 2 ∞ 3 4 ∞ ∞
2 0 2 3 3 4 ∞ ∞
3 0 2 3 3 4 ∞ 6
4 0 2 3 3 4 ∞ 4
5 0 2 3 3 4 6 4
6 0 2 3 3 4 6 4
At this point the algorithm returns the value 4.
Complexity of Dijkstra’s algorithm. For an nvertex, simple, con
nected weighted graph, Dijkstra’s algorithm has a worstcase run time
of Θ(n
2
).
7.4. PLANAR GRAPHS 97
7.4. Planar Graphs
7.4.1. Planar Graphs. A graph G is planar if it can be drawn
in the plane with its edges intersecting at their vertices only. One such
drawing is called an embedding of the graph in the plane.
A particular planar representation of a planar graph is called a map.
A map divides the plane into a number of regions or faces (one of them
inﬁnite).
7.4.2. Graph Homeomorphism. If a graph G has a vertex v of
degree 2 and edges (v, v
1
), (v, v
2
) with v
1
= v
2
, we say that the edges
(v, v
1
) and (v, v
2
) are in series. Deleting such vertex v and replacing
(v, v
1
) and (v, v
2
) with (v
1
, v
2
) is called a series reduction. For instance,
in the third graph of ﬁgure 7.16, the edges (h, b) and (h, d) are in series.
By removing vertex h we get the ﬁrst graph in the left.
The opposite of a series reduction is an elementary subdivision It
consists of replacing an edge (u, v) with two edges (u, w) and (w, v),
where w is a new vertex.
Two graphs are said to be homeomorphic if they are isomorphic or
can be reduced to isomorphic graphs by a sequence of series reductions
(ﬁg. 7.16). Equivalently, two graphs are homeomorphic if they can be
obtained from the same graph by a sequence of elementary subdivisions.
d
e
a
b d
e
f
c
d
e a
b
a
b
h
c c
Figure 7.16. Three homeomorphic graphs.
Note that if a graph G is planar, then all graphs homeomorphic to
G are also planar.
7.4.3. Some Results About Planar Graphs.
7.4. PLANAR GRAPHS 98
1. Euler’s Formula: Let G = (V, E) be a connected planar graph,
and let v = [V [, e = [E[, and r = number of regions in which
some given embedding of G divides the plane. Then:
v −e +r = 2 .
Note that this implies that all plane embeddings of a given
graph deﬁne the same number of regions.
2. Let G = (V, E) be a simple connected planar graph with v
vertices, e ≥ 3 edges and r regions. Then 3r ≤ 2e and e ≤
3v −6.
3. The graph K
5
is nonplanar. Proof: in K
5
we have v = 5 and
e = 10, hence 3v − 6 = 9 < e = 10, which contradicts the
previous result.
4. The graph K
3,3
is nonplanar. Proof: in K
3,3
we have v = 6 and
e = 9. If K
3,3
were planar, from Euler’s formula we would have
f = 5. On the other hand, each region is bounded by at least
four edges, so 4f ≤ 2e, i.e., 20 ≤ 18, which is a contradiction.
5. Kuratowski’s Theorem: A graph is nonplanar if and only if it
contains a subgraph that is homeomorphic to either K
5
or K
3,3
.
7.4.4. Dual Graph of a Map. A map is deﬁned by some planar
graph G = (V, E) embedded in the plane. Assume that the map divides
the plane into a set of regions R = ¦r
1
, r
2
, . . . , r
k
¦. For each region r
i
,
select a point p
i
in the interior of r
i
. The dual graph of that map is the
graph G
d
= (V
d
, E
d
), where V
d
= ¦p
1
, p
2
, . . . , p
k
¦, and for each edge in
E separating the regions r
i
and r
j
, there is an edge in E
d
connecting p
i
and p
j
. Warning: Note that a diﬀerent embedding of the same graph
G may give diﬀerent (and nonisomorphic) dual graphs. Exercise: Find
the duals of the maps shown in ﬁgure 7.9, and prove that they are not
isomorphic.
7.4.5. Graph Coloring. Consider the problem of coloring a map
M in such a way that no adjacent regions (sharing a border) have the
same color. This is equivalent to coloring the vertices of the dual graph
of M in such a way that no adjacent vertices have the same color.
In general, a coloring of a graph is an assignment of a color to
each vertex of the graph. The coloring is called proper if there are
no adjacent vertices with the same color. If a graph can be properly
7.4. PLANAR GRAPHS 99
a
b
d
c
e
Figure 7.17. Dual graph of a map.
colored with n colors we say that it is ncolorable. The minimum
number of colors needed to properly color a given graph G = (V, E) is
called the chromatic number of G, and is represented χ(G). Obviously
χ(G) ≤ [V [.
7.4.6. Some Results About Graph Coloring.
1. χ(K
n
) = n.
2. Let G be a simple graph. The following statements are equiva
lent:
(a) χ(G) = 2.
(b) G is bipartite.
(c) Every circuit in G has even length
3. Five Color Theorem (Kempe, Heawood) (not hard to prove):
Every simple, planar graph is 5colorable.
4. Four Color Theorem (Appel and Haken, 1976), proved with
an intricate computer analysis of conﬁgurations: Every simple,
planar graph is 4colorable.
Exercise: Find a planar graph G such that χ(G) = 4.
CHAPTER 8
Trees
8.1. Trees
8.1.1. Terminology. A tree is a connected undirected graph with
no simple circuits.
A rooted tree is a tree in which a particular vertex is designated as
the root and every edge is directed away from the root.
We draw rooted trees with the root at the top. The arrows indicat
ing the directions of the edges can be omitted.
r
a
b
c
d
e f
g
h
i
j
k
l m n o p
Figure 8.1. A rooted tree.
The level of a vertex v is the length of the simple path from the root
to v. The height of a rooted tree is the maximum level of its vertices.
Let T be a tree with root v
0
. Suppose that x, y and z are vertices
in T and that (v
0
, v
1
, . . . , v
n
) is a simple path in T. Then:
1. v
n−1
is the parent of v
n
.
2. v
0
, v
1
, . . . , v
n−1
are ancestors of v
n
.
3. v
n
is a child of v
n−1
.
4. If x is an ancestor of y, y is a descendant of x.
5. If x and y are children of z, x and y are siblings.
6. If x has no children, it is called a terminal vertex or leaf.
100
8.1. TREES 101
7. If x is not a terminal vertex, it is an internal or branch vertex.
8. The subtree of T rooted at x is the graph (V, E), where V is x
together with its descendants and E = edges of simple paths
from x to some vertex in E.
8.2. BINARY TREES 102
8.2. Binary Trees
8.2.1. Binary Trees. A binary tree is a rooted tree in which each
vertex has at most two children, designated as left child and right child.
If a vertex has one child, that child is designated as either a left child
or a right child, but not both. A full binary tree is a binary tree in
which each vertex has exactly two children or none. The following are
a few results about binary trees:
1. If T is a full binary tree with i internal vertices, then T has i+1
terminal vertices and 2i + 1 total vertices.
2. If a binary tree of height h has t terminal vertices, then t ≤ 2
h
.
More generally we can deﬁne a mary tree as a rooted tree in which
every internal vertex has no more than m children. The tree is called
a full mary tree if every internal vertex has exactly m children. An
ordered rooted tree is a rooted tree where the children of each internal
vertex are ordered. A binary tree is just a particular case of mary
ordered tree (with m = 2).
8.2.2. Binary Search Trees. Assume S is a set in which elements
(which we will call “data”) are ordered; e.g., the elements of S can be
numbers in their natural order, or strings of alphabetic characters in
lexicographic order. A binary search tree associated to S is a binary
tree T in which data from S are associate with the vertices of T so
that, for each vertex v in T, each data item in the left subtree of v is
less than the data item in v, and each data item in the right subtree of
v is greater than the data item in v.
Example: Figure 8.2 contains a binary search tree for the set S =
¦1, 2, 3, 4, 5, 6, 7, 8, 9, 10¦. In order to ﬁnd a element we start at the root
and compare it to the data in the current vertex (initially the root).
If the element is greater we continue through the right child, if it is
smaller we continue through the left child, if it is equal we have found
it. If we reach a terminal vertex without founding the element, then
that element is not present in S.
8.2.3. Making a Binary Search Tree. We can store data in a
binary search tree by randomly choosing data from S and placing it in
the tree in the following way: The ﬁrst data chosen will be the root of
the tree. Then for each subsequent data item, starting at the root we
8.2. BINARY TREES 103
5
2
1
3 7
6
9
10
8 4
Figure 8.2. Binary Search Tree.
compare it to the data in the current vertex v. If the new data item is
greater than the data in the current vertex then we move to the right
child, if it is less we move to the left child. If there is no such child
then we create one and put the new data in it. For instance, the tree in
ﬁgure 8.3 has been made from the following list of words choosing them
in the order they occur: “IN A PLACE OF LA MANCHA WHOSE
NAME I DO NOT WANT TO REMEMBER”.
A
I
DO
WHOSE
PLACE
OF
IN
LA
REMEMBER
TO
WANT
MANCHA
NAME
NOT
Figure 8.3. Another binary Search Tree.
8.3. DECISION TREES, TREE ISOMORPHISMS 104
8.3. Decision Trees, Tree Isomorphisms
8.3.1. Decision Trees. A decision tree is a tree in which each
vertex represents a question and each descending edge from that vertex
represents a possible answer to that question.
Example: The FiveCoins Puzzle. In this puzzle we have ﬁve coins
C
1
, C
2
, C
3
, C
4
, C
5
that are identical in appearance, but one is either
heavier or lighter that the others. The problem is to identify the bad
coin and determine whether it is lighter or heavier using only a pan
balance and comparing the weights of two piles of coins. The problem
can be solved in the following way. First we compare the weights of
C
1
and C
2
. If C
1
is heavier than C
2
then we know that either C
1
is
the bad coin and is heavier, or C
2
is the bad coin and it is lighter.
Then by comparing say C
1
with any of the other coins, say C
5
, we can
determine whether the bad coin is C
1
and is heavier (if C
1
it is heavier
than C
5
) or it is C
2
and is lighter (if C
1
has the same weight as C
5
). If
C
1
is lighter than C
2
we proceed as before with “heavier” and “lighter”
reversed. If C
1
and C
2
have the same weight we can try comparing C
3
and C
4
in a similar manner. If their weights are the same then we know
that the bad coin is C
5
, and we can determine whether it is heavier or
lighter by comparing it to say C
1
. The corresponding decision tree is
the following:
C1:C5
C1:C2
C3:C4 C1:C5
C3:C5
C1:C5
C3:C5
C1,H
C2,L
C4,H C3,L
C2,H C1,L
C5,H C5,L C4,L C3,H
left
right
balanced
balanced
balanced right
left right
right
balanced
right
left
balanced
left
left
balanced
Figure 8.4. Decision tree for the 5 coins puzzle.
In each vertex “C
i
: C
j
” means that we compare coins C
i
and C
j
by
placing C
i
on the left pan and C
j
on the right pan of the balance, and
each edge is labeled depending on what side of the balance is heavier.
The terminal vertices are labeled with the bad coin and whether it is
heavier (H) or lighter (L). The decision tree is optimal in the sense
that in the worst case it uses three weighings, and there is no way to
solve the problem with less than that—with two weighings we can get
8.3. DECISION TREES, TREE ISOMORPHISMS 105
at most nine possible outcomes, which are insuﬃcient to distinguish
among ten combinations of 5 possible bad coins and the bad coin being
heavier or lighter.
8.3.2. Complexity of Sorting. Sorting algorithms work by com
paring elements and rearranging them as needed. For instance we can
sort three elements a
1
, a
2
, a
3
with the decision tree shown in ﬁgure 8.5
a1<a2?
a1<a3?
a2<a3? a1<a3?
a1,a2,a3
a1,a3,a2 a3,a1,a2
a2,a3,a1 a3,a2,a1
a2,a1,a3 a2<a3?
YES
YES
YES NO
NO YES NO YES
NO
NO
.
Figure 8.5. Sorting three elements.
Since there are 3! = 6 possible arrangements of 3 elements, we need
a decision tree with at least 6 possible outcomes or terminal vertices.
Recall that in a binary tree of height h with t terminal vertices the
following inequality holds: t ≤ 2
h
. Hence in our case 6 < 2
h
, which
implies h ≥ 3, so the algorithm represented by the decision tree in
ﬁgure 8.5 is optimal in the sense that it uses the minimum possible
number of comparisons in the worstcase.
More generally in order to sort n elements we need a decision tree
with n! outcomes, so its height h(n) will verify n! ≤ 2
h(n)
. Since
log
2
(n!) = Θ(nlog
2
n),
1
we have h(n) = Ω(nlog
2
n). So the worse
case complexity of a sorting algorithm is Ω(nlog
2
n). Since the merge
sort algorithm uses precisely Θ(nlog
2
n) comparisons, we know that it
is optimal.
8.3.3. Isomorphisms of Trees. Assume that T
1
is a tree with
vertex set V
1
and T
2
is another tree with vertex set V
2
. If they are
rooted trees then we call their roots r
1
and r
2
respectively. We will
study three diﬀerent kinds of treeisomorphisms between T
1
and T
2
.
1
According to Stirling’s formula, n! ≈ n
n
e
−n
√
2πn, so taking logarithms
log
2
n! ≈ nlog
2
n −nlog
2
e +
1
2
log
2
(2πn) = Θ(nlog
2
n).
8.3. DECISION TREES, TREE ISOMORPHISMS 106
1. Usual graphisomorphism between trees: T
1
and T
2
are isomor
phic if there is a bijection f : V
1
→ V
2
that preserves adjacency,
i.e., f(v) is adjacent to f(w) if and only if v is adjacent to w.
2. Rootedtreeisomorphism: T
1
and T
2
are isomorphic if there is
a bijection f : V
1
→ V
2
that preserves adjacency and the root
vertex, i.e.:
(a) f(v) is adjacent to f(w) if and only if v is adjacent to w.
(b) f(r
1
) = r
2
.
3. Binarytreeisomorphism: Two binary trees T
1
and T
2
are iso
morphic if there is a bijection f : V
1
→ V
2
that preserves adja
cency, and the root vertex, and left/right children, i.e.:
(a) f(v) is adjacent to f(w) if and only if v is adjacent to w.
(b) f(r
1
) = r
2
.
(c) f(v) is a left child of f(w) if and only if v is a left child of
w.
(d) f(v) is a right child of f(w) if and only if v is a right child
of w.
Example: Figure 8.6 shows three trees which are graphisomorphic.
On the other hand as rooted trees T
2
and T
3
are isomorphic, but they
are not isomorphic to T
1
because the root of T
1
has degree 3, while the
roots of T
2
and T
3
have degree 2. Finally T
2
and T
3
are not isomorphic
as binary trees because the left child of the root in T
2
is a terminal
vertex while the left child of the root of T
3
has two children.
T1 T2 T3
Figure 8.6. Trees with diﬀerent kinds of isomorphisms.
Exercise: Find all nonisomorphic 3vertex free trees, 3vertex rooted
trees and 3vertex binary trees. Answer: Figure 8.7 shows all 5 non
isomorphic 3vertex binary trees. As rooted trees T
2
–T
5
are isomorphic,
but T
1
is not isomorphic to the others, so there are 2 nonisomorphic
3vertex rooted trees represented for instance by T
1
and T
2
. All of them
8.3. DECISION TREES, TREE ISOMORPHISMS 107
are isomorphic as free trees, so there is only 1 nonisomorphic 3vertex
free tree.
T1 T3 T2 T4 T5
Figure 8.7. Nonisomorphic binary trees.
8.3.4. Huﬀman Codes. Usually characters are represented in a
computer with ﬁx length bit strings. Huﬀman codes provide an alter
native representation with variable length bit strings, so that shorter
strings are used for the most frequently used characters. As an example
assume that we have an alphabet with four symbols: A = ¦a, b, c, d¦.
Two bits are enough for representing them, for instance a = 11, b = 10,
c = 01, d = 00 would be one such representation. With this encoding
ncharacter words will have 2n bits. However assume that they do not
appear with the same frequency, instead some are more frequent that
others, say a appears with a frequency of 50%, b 30%, c 15% and d
5%. Then the following enconding would be more eﬃcient than the ﬁx
length encoding: a = 1, b = 01, c = 001, d = 000. Now in average an
ncharacter word will have 0.5n a’s, 0.3n b’s, 0.15n c’s and 0.05n d’s,
hence its length will be 0.5n1+0.3n2+0.15n3+0.05n3 = 1.7n, which
is shorter than 2n. In general the length per character of a given en
coding with characters a
1
, a
2
, . . . , a
n
whose frequencies are f
1
, f
2
, . . . , f
n
is
1
F
n
¸
k=1
f
k
l(a
k
) ,
where l(a
k
) = length of a
k
and F =
¸
n
k=1
f
k
. The problem now is,
given an alphabet and the frequencies of its characters, ﬁnd an optimal
encoding that provides minimum average length for words.
Fix length and Huﬀman codes can be represented by trees like in
ﬁgure 8.8. The code of each symbol consists of the sequence of labels of
the edges in the path from the root to the leaf with the desired symbol.
8.3. DECISION TREES, TREE ISOMORPHISMS 108
a
b d
0
1 0
1 0
1
a
b
c d
1 0
1 0
1 0
Huffman code Fix length code
c
Figure 8.8. Fix length code and Huﬀman code.
8.3.5. Constructing an Optimal Huﬀman Code. An optimal
Huﬀman code is a Huﬀman code in which the average length of the
symbols is minimum. In general an optimal Huﬀman code can be made
as follows. First we list the frequencies of all the codes and represent
the symbols as vertices (which at the end will be leaves of a tree).
Then we replace the two smallest frequencies f
1
and f
2
with their sum
f
1
+ f
2
, and join the corresponding two symbols to a common vertex
above them by two edges, one labeled 0 and the other one labeled 1.
Than common vertex plays the role of a new symbol with a frequency
equal to f
1
+f
2
. Then we repeat the same operation with the resulting
shorter list of frequencies until the list is reduced to one element and
the graph obtained becomes a tree.
Example: Find the optimal Huﬀman code for the following table of
symbols:
character frequency
a 2
b 3
c 7
d 8
e 12
Answer: : The successive reductions of the list of frequencies are as
follows:
2, 3
....
5
, 7, 8, 12 → 5, 7
....
12
, 8, 12 → 12, 8, 12
Here we have a choice, we can choose to add the ﬁrst 12 and 8, or
8 and the second 12. Let’s choose the former:
8.3. DECISION TREES, TREE ISOMORPHISMS 109
12, 8
....
20
, 12 → 20, 12
. .. .
32
→ 32
The tree obtained is the following:
a b
d
e
1
1
1
1 0
0
0
0
2
3
7
8
12
5
12
20
32
c
Figure 8.9. Optimal Huﬀman code 1.
The resulting code is as follows:
character code
a 1111
b 1110
c 110
d 10
e 0
The other choice yields the following:
12, 8, 12
....
20
→ 20, 12
. .. .
32
→ 32
a
b
c d
e
1 0
1
1
0
0
1
32
12 20
5
7 8
12
2
3
0
Figure 8.10. Optimal Huﬀman code 2.
8.3. DECISION TREES, TREE ISOMORPHISMS 110
character code
a 111
b 110
c 10
d 01
e 00
8.3.6. Game Trees. Trees are used in the analysis of some games.
As an example we study the following game using a tree: Initially
there are two piles with 3 coins and 1 coin respectively. Taking turns
two players remove any number of coins from one of the piles. The
player that removes the last coin loses. The following tree represents
all possible sequences of choices. Each node shows the number of coins
in each pile, and each edge represents a possible “move” (choice) from
one of the players. The ﬁrst player is represented with a box and the
second player is represented with an circle.
1
0
0
1
0
0
0
0
0
0
1
1
0
1
0
1
2
0
0
0
1
0
0
0
0
0
1
0
0
1
0
0
0
0
1
1
2
1
1
0
0
0
0
0
1
0
0
0
0
0
2
0
3
0
3
1
1 1 1
0 0 0
1
1
0
1
0 0
1
1
1
1
0
1
0 0
0 1
0
1
0
0
1
1
Figure 8.11. Tree of a game.
The analysis of the game starts by labeling each terminal vertex
with “1” if it represents a victory for the ﬁrst player and “0” if it
represents a victory for the second player. This numbers represent
the “value” of each position of the game, so that the ﬁrst player is
interested in making it “maximum” and the second player wants to
make it “minimum”. Then we continue labeling the rest of the vertices
in the following way. After all the children of a given vertex have
been labeled, we label the vertex depending on whether it is a “ﬁrst
player” position (box) or a “second player” position (circle). First
player positions are labeled with the maximum value of the labels of
its children, second player positions are labeled with the minimum
8.3. DECISION TREES, TREE ISOMORPHISMS 111
value of the labels of its children. This process is called the minimax
procedure. Every vertex labeled “1” will represent a position in which
the ﬁrst player has advantage and can win if he/she works without
making mistakes; on the other hand, vertices labeled “0” represent
positions for which the second player has advantage. Now the strategy
is for the ﬁrst player to select at each position a children with maximum
value, while the second player will be interested in selecting children
with minimum value. If the starting position has been labeled “1” that
means that the ﬁrst player has a winning strategy, otherwise the second
player has advantage. For instance in the present game the ﬁrst player
has advantage at the initial position, and only one favorable movement
at that point:
3
1
→
0
1
, i.e., he/she must remove all 3 coins from
the ﬁrst pile. If for any reason the ﬁrst player makes a mistake and
removes say one coin from the ﬁrst pile, going to position
2
1
, then the
second player has one favorable move to vertex
0
1
, which is the one
with minimum “value”.
Alphabeta pruning. In some games the game tree is so complicated
that it cannot be fully analyzed, so it is built up to a given depth only.
The vertices reached at that depth are not terminal, but they can
be “evaluated” using heuristic methods (for instance in chess usually
losing a knight is a better choice than losing the queen, so a position
with one queen and no knights will have a higher value than one with
no queen and one knight). Even so the evaluation and labeling of the
vertices can be time consuming, but we can bypass the evaluation of
many vertices using the technique of alphabeta pruning. The idea is
to skip a vertex as soon as it becomes obvious that its value will not
aﬀect the value of its parent. In order to do that with a ﬁrst player
(boxed) vertex v, we assign it an alpha value equal to the maximum
value of its children evaluated so far. Assume that we are evaluating
one of its children w, which will be a second player (circled) position. If
at any point a children of w gets a value less than or equal to the alpha
value of v then it will become obvious that the value of w is going to
be less than the current alpha value of v, so it will not aﬀect the value
of v and we can stop the process of evaluation of w (prone the subtree
at w). That is called an alpha cutoﬀ. Similarly, at a second player
(circled) vertex v, we assign a beta value equal to the minimum value
of its children evaluated so far, and practice a beta cutoﬀ when one of
its grandchildren gets a value greater than or equal to the current beta
value of v, i.e., we prone the subtree at w, where w is the parent of
that grandchildren.
8.3. DECISION TREES, TREE ISOMORPHISMS 112
4
4 3
5 6 3 4 5 1 ?
v
w
(4)
1
7
Figure 8.12. Alpha cutoﬀ.
8.4. TREE TRANSVERSAL 113
8.4. Tree Transversal
8.4.1. Transversal Algorithms. In order to motivate this sub
ject, we introduce the concept of Polish notation. Given a (not nec
essarily commutative) binary operation ◦, it is customary to represent
the result of applying the operation to two elements a, b by placing the
operation symbol in the middle:
a ◦ b .
This is called inﬁx notation. The Polish notation consists of placing
the symbol to the left:
◦ a b .
The reverse Polish notation consists of placing the symbol to the right:
a b ◦ .
The advantage of Polish notation is that it allows us to write ex
pressions without need for parenthesis. For instance, the expression
a∗(b+c) in Polish notation would be ∗ a+b c, while a∗b+c is +∗a b c.
Also, Polish notation is easier to evaluate in a computer.
In order to evaluate an expression in Polish notation, we scan the
expression from right to left, placing the elements in a stack.
1
Each
time we ﬁnd an operator, we replace the two top symbols of the stack
by the result of applying the operator to those elements. For instance,
the expression ∗ +2 3 4 (which in inﬁx notation is “(2 +3) ∗ 4”) would
be evaluated like this:
expression stack
∗ + 2 3 4
∗ + 2 3 4
∗ + 2 3 4
∗ + 2 3 4
∗ 5 4
20
An algebraic expression can be represented by a binary rooted tree
obtained recursively in the following way. The tree for a constant or
variable a has a as its only vertex. If the algebraic expression S is of
1
A stack or lastin ﬁrstout (LIFO) system, is a linear list of elements in which
insertions and deletions take place only at one end, called top of the list. A queue
or ﬁrstin ﬁrstout (FIFO) system, is a linear list of elements in which deletions
take place only at one end, called front of the list, and insertions take place only
at the other end, called rear of the list.
8.4. TREE TRANSVERSAL 114
the form S
L
◦ S
R
, where S
L
and S
R
are subexpressions with trees T
L
and T
R
respectively, and ◦ is an operator, then the tree T for S consists
of ◦ as root, and the subtrees T
L
and T
R
(ﬁg. 8.13).
o
T
L
T
R
Figure 8.13. Tree of S
1
◦ S
2
.
For instance, consider the following algebraic expression:
a +b ∗ c +d ↑ e ∗ (f +h) ,
where + denotes addition, ∗ denotes multiplication and ↑ denotes ex
ponentiation. The binary tree for this expression is given in ﬁgure 8.14.
+
*
+
d e
f h
*
b c
+
a
Figure 8.14. Tree for a +b ∗ c +d ↑ e ∗ (f +h).
Given the binary tree of an algebraic expression, its Polish, reverse
Polish and inﬁx representation are diﬀerent ways of ordering the ver
tices of the tree, namely in preorder, postorder and inorder respectively.
The following are recursive deﬁnitions of several orderings of the
vertices of a rooted tree T = (V, E) with root r. If T has only one
vertex r, then r by itself constitutes the preorder, postorder and inorder
transversal of T. Otherwise, let T
1
, . . . , T
k
the subtrees of T from left
to right (ﬁg. 8.15). Then:
1. Preorder Transversal : Pre(T) = r, Pre(T
1
), . . . , Pre(T
k
).
8.4. TREE TRANSVERSAL 115
r
T1 T2
...
Tk
Figure 8.15. Ordering of trees.
2. Postorder Transversal : Post(T) = Post(T
1
), . . . , Post(T
k
), r.
3. Inorder Transversal. If T is a binary tree with root r, left sub
tree T
L
and right subtree T
R
, then: In(T) = In(T
L
), r, In(T
R
).
8.5. SPANNING TREES 116
8.5. Spanning Trees
8.5.1. Spanning Trees. A tree T is a spanning tree of a graph G
if T is a subgraph of G that contains all the vertices of G. For instance
the graph of ﬁgure 8.16 has a spanning tree represented by the thicker
edges.
b
c
d
e
f
g
h
a
i
Figure 8.16. Spanning tree.
Every connected graph has a spanning tree which can be obtained
by removing edges until the resulting graph becomes acyclic. In prac
tice, however, removing edges is not eﬃcient because ﬁnding cycles is
time consuming.
Next, we give two algorithms to ﬁnd the spanning tree T of a loop
free connected undirected graph G = (V, E). We assume that the
vertices of G are given in a certain order v
1
, v
2
, . . . , v
n
. The resulting
spanning tree will be T = (V
, E
).
8.5.2. BreadthFirst Search Algorithm. The idea is to start
with vertex v
1
as root, add the vertices that are adjacent to v
1
, then the
ones that are adjacent to the latter and have not been visited yet, and
so on. This algorithm uses a queue (initially empty) to store vertices
of the graph. In consists of the following:
1. Add v
1
to T, insert it in the queue and mark it as “visited”.
2. If the queue is empty, then we are done. Otherwise let v be the
vertex in the front of the queue.
3. For each vertex v
of G that has not been visited yet and is
adjacent to v (there might be none) taken in order of increasing
subscripts, add vertex v
and edge (v, v
) to T, insert v
in the
queue and mark it as “visited”.
4. Delete v from the queue.
8.5. SPANNING TREES 117
5. Go to step 2.
A pseudocode version of the algorithm is as follows:
1: procedure bfs(V,E)
2: S := (v1) ¦ordered list of vertices of a fix level¦
3: V’ := ¦v1¦ ¦v1 is the root of the spanning tree¦
4: E’ := ¦¦ ¦no edges in the spanning tree yet¦
5: while true
6: begin
7: for each x in S, in order,
8: for each y in V  V’
9: if (x,y) is an edge then
10: add edge (x,y) to E’ and vertex y to V’
11: if no edges were added then
12: return T
13: S := children of S
14: end
15: end bfs
Figure 8.17 shows the spanning tree obtained using the breadthﬁrst
search algorithm on the graph with its vertices ordered lexicographi
cally: a, b, c, d, e, f, g, h, i.
a
b
c
d
e
f
g
h
i
Figure 8.17. BreadthFirst Search.
8.5.3. DepthFirst Search Algorithm. The idea of this algo
rithm is to make a path as long as possible, and then go back (back
track) to add branches also as long as possible.
This algorithm uses a stack (initially empty) to store vertices of the
graph. In consists of the following:
1. Add v
1
to T, insert it in the stack and mark it as “visited”.
8.5. SPANNING TREES 118
2. If the stack is empty, then we are done. Otherwise let v be the
vertex on the top of the stack.
3. If there is no vertex v
that is adjacent to v and has not been
visited yet, then delete v and go to step 2 (backtrack). Oth
erwise, let v
be the ﬁrst nonvisited vertex that is adjacent to
v.
4. Add vertex v
and edge (v, v
) to T, insert v
in the stack and
mark it as “visited”.
5. Go to step 2.
An alternative recursive deﬁnition is as follows. We deﬁne recur
sively a process P applied to a given vertex v in the following way:
1. Add vertex v to T and mark it as “visited”.
2. If there is no vertex v
that is adjacent to v and has not been
visited yet, then return. Otherwise, let v
be the ﬁrst nonvisited
vertex that is adjacent to v.
3. Add the edge (v, v
) to T.
4. Apply P to v
.
5. Go to step 2 (backtrack).
The DepthFirst Search Algorithm consists of applying the process just
deﬁned to v
1
.
A pseudocode version of the algorithm is as follows:
1: procedure dfs(V,E)
2: V’ := ¦v1¦ ¦v1 is the root of the spanning tree¦
3: E’ := ¦¦ ¦no edges in the spanning tree yet¦
4: w := v1
5: while true
6: begin
7: while there is an edge (w,v) that when added
8: to T does not create a cycle in T
9: begin
10: Choose first v such that (w,v)
11: does not create a cycle in T
12: add (w,v) to E’
13: add v to V’
14: w := v
15: end
16: if w = v1 then
8.5. SPANNING TREES 119
17: return T
18: w := parent of w in T ¦backtrack¦
19: end
20: end
Figure 8.18 shows the spanning tree obtained using the breadthﬁrst
search algorithm on the graph with its vertices ordered lexicographi
cally: a, b, c, d, e, f, g, h, i.
a
b
c
d
e
f
g
h
i
Figure 8.18. DepthFirst Search.
8.5.4. Minimal Spanning Trees. Given a connected weighted
tree G, its minimal spanning tree is a spanning tree of G such that the
sum of the weights of its edges is minimum. For instance for the graph
of ﬁgure 8.19, the spanning tree shown is the one of minimum weight.
a b
c d
e f
4
1
5
2
6
3
2
6
Figure 8.19. Minimum Spanning Tree.
Prim’s Algorithm. An algorithm to ﬁnd a minimal spanning tree is
Prim’s Algorithm. It starts with a single vertex and at each iteration
adds to the current tree a minimum weight edge that does not complete
a cycle.
8.5. SPANNING TREES 120
The following is a pseudocode version of Prim’s algorithm. If (x, y)
is an edge in G = (V, E) then w(x, y) is its weight, otherwise w(x, y) =
∞. The starting vertex is s.
1: procedure prim(V,w,s)
2: V’ := ¦s¦ ¦vertex set starts with s¦
3: E’ = ¦¦ ¦edge set initially empty¦
4: for i := 1 to n1 ¦put n edges in spanning tree¦
5: begin
6: find x in V’ and y in V  V’ with minimum w(x,y)
7: add y to V’
8: add (x,y) to E’
9: end
10: return E’
11: end prim
Prim’s algorithm is an example of a greedy algorithm. A greedy
algorithm is an algorithm that optimized the choice at each iteration
without regard to previous choices (“doing the best locally”). Prim’s
algorithm makes a minimum spanning tree, but in general a greedy
algorithm does not always ﬁnds an optimal solution to a given problem.
For instance in ﬁgure 8.20 a greedy algorithm to ﬁnd the shortest path
from a to z, working by adding the shortest available edge to the most
recently added vertex, would return acz, which is not the shortest path.
a
b
c
z
1
2
4
100
101
Figure 8.20
Kruskal’s Algorithm. Another algorithm to ﬁnd a minimal span
ning tree in a connected weighted tree G = (V, E) is Kruskal’s Algo
rithm. It starts with all n vertices of G and no edges. At each iteration
we add an edge having minimum weight that does not complete a cycle.
We stop after adding n −1 edges.
8.5. SPANNING TREES 121
1: procedure kruskal(E,w,n)
2: V’ := V
3: E’ := ¦¦
4: while E’ < n1
5: begin
6: among all edges not completing a cycle in T
7: choose e of minimum weight and add it to E
8: end
9: T’ = (V’,E’)
10: return T’
11: end kruskal
CHAPTER 9
Boolean Algebras
9.1. Combinatorial Circuits
9.1.1. Introduction. At their lowest level digital computers han
dle only binary signals, represented with the symbols 0 and 1. The
most elementary circuits that combine those signals are called gates.
Figure 9.1 shows three gates: OR, AND and NOT.
x1
x2
x1
x2
x x
x1
x2
x2
OR GATE
NOT GATE
AND GATE
x1 +
Figure 9.1. Gates.
Their outputs can be expressed as a function of their inputs by the
following logic tables:
x
1
x
2
x
1
+x
2
1 1 1
1 0 1
0 1 1
0 0 0
OR GATE
122
9.1. COMBINATORIAL CIRCUITS 123
x
1
x
2
x
1
x
2
1 1 1
1 0 0
0 1 0
0 0 0
AND GATE
x x
1 0
0 1
NOT GATE
These are examples of combinatorial circuits. A combinatorial cir
cuit is a circuit whose output is uniquely deﬁned by its inputs. They
do not have memory, previous inputs do not aﬀect their outputs. Some
combinations of gates can be used to make more complicated combi
natorial circuits. For instance ﬁgure 9.2 is combinatorial circuit with
the logic table shown below, representing the values of the Boolean
expression y = (x
1
+x
2
) x
3
.
x1
x2
y
x3
Figure 9.2. A combinatorial circuit.
x
1
x
2
x
3
y = (x
1
+x
2
) x
3
1 1 1 0
1 1 0 1
1 0 1 0
1 0 0 1
0 1 1 0
0 1 0 1
0 0 1 1
0 0 0 1
However the circuit in ﬁgure 9.3 is not a combinatorial circuit. If
x
1
= 1 and x
2
= 0 then y can be 0 or 1. Assume that at a given time
y = 0. If we input a signal x
2
= 1, the output becomes y = 1, and
9.1. COMBINATORIAL CIRCUITS 124
stays so even after x
2
goes back to its original value 0. That way we
can store a bit. We can “delete” it by switching input x
1
to 0.
x2
x1
y
Figure 9.3. Not a combinatorial circuit.
9.1.2. Properties of Combinatorial Circuits. Here Z
2
= ¦0, 1¦
represents the set of signals handled by combinatorial circuits, and the
operations performed on those signals by AND, OR and NOT gates are
represented by the symbols , + and respectively. Then their prop
erties are the following (a, b, c are elements of Z
2
, i.e., each represents
either 0 or 1):
1. Associative
(a +b) +c = a + (b +c)
(a b) c = a (b c)
2. Commutative
a +b = b +a
a b = b a
3. Distributive
a (b +c) = (a b) + (a c)
a + (b c) = (a +b) (a +c)
4. Identity
a + 0 = a
a 1 = a
5. Complement
a +a = 1
a a = 0
A system satisfying those properties is called a Boolean algebra.
Two Boolean expressions are deﬁned to be equal is they have the
same values for all possible assignments of values to their literals. Ex
ample: x +y = x y, as shown in the following table:
9.1. COMBINATORIAL CIRCUITS 125
x y x +y x y
1 1 0 0
1 0 0 0
0 1 0 0
0 0 1 1
9.1.3. Abstract Boolean Algebras. Here we deal with general
Boolean algebras; combinatorial circuits are an example, but there are
others.
A Boolean algebra B = (S, ∨, ∧, , 0, 1) is a set S containing two
distinguished elements 0 and 1, two binary operators ∨ and ∧ on S,
and a unary operator on S, satisfying the following properties (x, y,
z are elements of S):
1. Associative
(x ∨ y) ∨ z = x ∨ (y ∨ z)
(x ∧ y) ∨ z = x ∧ (y ∧ z)
2. Commutative
x ∨ y = y ∨ x
x ∧ y = y ∧ x
3. Distributive
x ∧ (y ∨ z) = (x ∧ y) ∨ (x ∧ z)
x ∨ (y ∧ z) = (x ∨ y) ∧ (x ∨ z)
4. Identity
x ∨ 0 = x
x ∧ 1 = x
5. Complement
x ∨ x = 1
x ∧ x = 0
Example: (Z
2
, +, , , 0, 1) is a Boolean algebra.
Example: If U is a universal set and P(U)= the power set of S (col
lection of subsets of S) then (P(U), ∪, ∩, , ∅, U). is a Boolean algebra.
9.1. COMBINATORIAL CIRCUITS 126
9.1.4. Other Properties of Boolean Algebras. The properties
mentioned above deﬁne a Boolean algebra, but Boolean algebras also
have other properties:
1. Idempotent
x ∨ x = x
x ∧ x = x
2. Bound
x ∨ 1 = 1
x ∧ 0 = 0
3. Absorption
x ∨ xy = x
x ∧ (x ∨ y) = x
4. Involution
x = x
5. 0 and 1
0 = 1
1 = 0
6. De Morgan’s
x ∨ y = x ∧ y
x ∧ y = x ∨ y
For instance the ﬁrst idempotent law can be proved like this: x =
x ∨ 0 = x ∨ x ∧ x = (x ∨ x) ∧ (x ∨ x) = (x ∨ x) ∧ 1 = x ∨ x.
9.2. BOOLEAN FUNCTIONS, APPLICATIONS 127
9.2. Boolean Functions, Applications
9.2.1. Introduction. A Boolean function is a function from Z
n
2
to Z
2
. For instance, consider the exclusiveor function, deﬁned by the
following table:
x
1
x
2
x
1
⊕x
2
1 1 0
1 0 1
0 1 1
0 0 0
The exclusiveor function can interpreted as a function Z
2
2
→ Z
2
that assigns (1, 1) → 0, (1, 0) → 1, (0, 1) → 1, (0, 0) → 0. It can also
be written as a Boolean expression in the following way:
x
1
⊕x
2
= (x
1
x
2
) + (x
1
x
2
)
Every Boolean function can be written as a Boolean expression as
we are going to see next.
9.2.2. Disjunctive Normal Form. We start with a deﬁnition.
A minterm in the symbols x
1
, x
2
, . . . , x
n
is a Boolean expression of the
form y
1
y
2
y
n
, where each y
i
is either x
i
or x
i
.
Given any Boolean function f : Z
n
2
→ Z
2
that is not identically
zero, it can be represented
f(x
1
, . . . , x
n
) = m
1
+m
2
+ +m
k
,
where m
1
, m
2
, . . . , m
k
are all the minterms m
i
= y
1
y
2
y
n
such that
f(a
1
, a
2
, . . . , a
n
) = 1, where y
j
= x
j
if a
j
= 1 and y
j
= x
j
if a
j
= 0.
That representation is called disjunctive normal form of the Boolean
function f.
Example: We have seen that the exclusiveor can be represented
x
1
⊕ x
2
= (x
1
x
2
) + (x
1
x
2
). This provides a way to implement the
exclusiveor with a combinatorial circuit as shown in ﬁgure 9.4.
9.2.3. Conjunctive Normal Form. A maxterm in the symbols
x
1
, x
2
, . . . , x
n
is a Boolean expression of the form y
1
+ y
2
+ + y
n
,
where each y
i
is either x
i
or x
i
.
9.2. BOOLEAN FUNCTIONS, APPLICATIONS 128
x1
x2
x2 x1
Figure 9.4. ExclusiveOr.
Given any Boolean function f : Z
n
2
→ Z
2
that is not identically
one, it can be represented
f(x
1
, . . . , x
n
) = M
1
M
2
M
k
,
where M
1
, M
2
, . . . , M
k
are all the maxterms M
i
= y
1
+ y
2
+ + y
n
such that f(a
1
, a
2
, . . . , a
n
) = 0, where y
j
= x
j
if a
j
= 0 and y
j
= x
j
if
a
j
= 1. That representation is called conjunctive normal form of the
Boolean function f.
Example: The conjunctive normal form of the exclusiveor is
x
1
⊕x
2
= (x
1
+x
2
) (x
1
+x
2
) .
9.2.4. Functionally Complete Sets of Gates. We have seen
how to design combinatorial circuits using AND, OR and NOT gates.
Here we will see how to do the same with other kinds of gates. In the
following gates will be considered as functions from Z
n
2
into Z
2
intended
to serve as building blocks of arbitrary boolean functions.
A set of gates ¦g
1
, g
2
, . . . , g
k
¦ is said to be functionally complete
if for any integer n and any function f : Z
n
2
→ Z
2
it is possible to
construct a combinatorial circuit that computes f using only the gates
g
1
, g
2
, . . . , g
k
. Example: The result about the existence of a disjunctive
normal form for any Boolean function proves that the set of gates
¦AND, OR, NOT¦ is functionally complete. Next we show other sets
of gates that are also functionally complete.
1. The set of gates ¦AND, NOT¦ is functionally complete. Proof:
Since we already know that ¦AND, OR, NOT¦ is functionally
complete, all we need to do is to show that we can compute
x +y using only AND and NOT gates. In fact:
x +y = x y ,
hence the combinatorial circuit of ﬁgure 9.5 computes x +y.
9.2. BOOLEAN FUNCTIONS, APPLICATIONS 129
x1 x2
x2
x1
+
Figure 9.5. OR with AND and NOT.
2. The set of gates ¦OR, NOT¦ is functionally complete. The
proof is similar:
x y = x +y ,
hence the combinatorial circuit of ﬁgure 9.6 computes x+y.
x1
x1 x2
x2
Figure 9.6. AND with OR and NOT.
3. The gate NAND, denoted ↑ and deﬁned as
x
1
↑ x
2
=
0 if x
1
= 1 and x
2
= 1
1 otherwise,
is functionally complete.
x1
x1 x2
x2
Figure 9.7. NAND gate.
Proof: Note that x ↑ y = x y. Hence x = x x = x ↑ x, so
the NOT gate can be implemented with a NAND gate. Also the
OR gate can be implemented with NAND gates: x+y = x y =
(x ↑ x) ↑ (y ↑ y). Since the set ¦OR, NOT¦ is functionally
complete and each of its elements can be implemented with
NAND gates, the NAND gate is functionally complete.
9.2.5. Minimization of Combinatorial Circuits. Here we ad
dress the problems of ﬁnding a combinatorial circuit that computes a
given Boolean function with the minimum number of gates. The idea
9.2. BOOLEAN FUNCTIONS, APPLICATIONS 130
x y
x
x
x
y
Figure 9.8. NOT and OR functions implemented with
NAND gates.
is to simplify the corresponding Boolean expression by using algebraic
properties such as (E a) + (E a) = E and E + (E a) = E, where
E is any Boolean expression. For simplicity in the following we will
represent a b as ab, so for instance the expressions above will look like
this: Ea +Ea = E and E +Ea = E.
Example: Let F(x, y, z) the Boolean function deﬁned by the follow
ing table:
x y z f(x,y,z)
1 1 1 1
1 1 0 1
1 0 1 0
1 0 0 1
0 1 1 0
0 1 0 0
0 0 1 0
0 0 0 0
Its disjunctive normal form is f(x, y, z) = xyz + xyz + xyz. This
function can be implemented with the combinatorial circuit of ﬁgure
9.9.
9.2. BOOLEAN FUNCTIONS, APPLICATIONS 131
z
y
x
.
f(x,y,z)
Figure 9.9. A circuit that computes f(x, y, z) = xyz +
xyz +xyz.
But we can do better if we simplify the expression in the following
way:
f(x, y, z) =
xy
. .. .
xyz +xyz +xyz
= xy +xyz
= x(y +yz)
= x(y +y)(y +z)
= x(y +z) ,
which corresponds to the circuit of ﬁgure 9.10.
f(x,y,z)
. y
x
z
Figure 9.10. A simpler circuit that computes
f(x, y, z) = xyz +xyz +xyz.
9.2.6. MultiOutput Combinatorial Circuits. Example: Half
Adder. A halfadder is a combinatorial circuit with two inputs x and
y and two outputs s and c, where s represents the sum of x and y and
c is the carry bit. Its table is as follows:
x y s c
1 1 0 1
1 0 1 0
0 1 1 0
0 0 0 0
So the sum is s = x ⊕y (exclusiveor) and the carry bit is c = x y.
Figure 9.11 shows a halfadder circuit.
9.2. BOOLEAN FUNCTIONS, APPLICATIONS 132
y
s
c
x
Figure 9.11. Halfadder circuit.
CHAPTER 10
Automata, Grammars and Languages
10.1. Finite State Machines
10.1.1. FiniteState Machines. Combinatorial circuits have no
memory or internal states, their output depends only on the current
values of their inputs. Finite state machines on the other hand have
internal states, so their output may depend not only on its current
inputs but also on the past history of those inputs.
A ﬁnitestate machine consists of the following:
1. A ﬁnite set of states S.
2. A ﬁnite set of input symbols I.
3. A ﬁnite set of output symbols O.
4. A nextstate or transition function f : S I → S.
5. An output function g : S I → O.
6. An initial state σ ∈ S.
We represent the machine M = (S, I, O, f, g, σ)
Example: We describe a ﬁnite state machine with two input symbols
I = ¦a, b¦ and two output symbols O = ¦0, 1¦ that accepts any string
from I
∗
and outputs as many 1’s as a’s there are at the beginning of the
string, then it outputs only 0’s. The internal states are S = ¦σ
0
, σ
1
¦,
where σ
0
is the initial state—we interpret it as not having seeing any
“b” yet; then the machine will switch to σ
1
as soon as the ﬁrst “b”
arrives. The nextstate and output functions are as follows:
f g
I a b a b
S
σ
0
σ
0
σ
1
1 0
σ
1
σ
1
σ
1
0 0
133
10.1. FINITE STATE MACHINES 134
This ﬁnitestate machine also can be represented with the following
transition diagram:
start
GFED @ABC
σ
0
a/1
b/0
GFED @ABC
σ
1
a/0
b/0
The vertices of the diagram are the states. If in state σ an input
i causes the machine to output o and go to state σ
then we draw an
arrow from σ to σ
labeled i/o or i, o.
Example: The following example is similar to the previous one but
the machine outputs 1 only after a change of input symbol, otherwise
it outputs 0:
start
GFED @ABC
σ
0
a/0
.· 








b/0
B
B
B
B
B
B
B
B
B
GFED @ABC
σ
1
a/0
b/1
GFED @ABC
σ
2
b/0
a/1
Example: A SerialAdder. A serial adder accepts two bits and out
puts its sum. So the input set is I = ¦00, 01, 10, 11¦. The output
set is O = ¦0, 1¦. The set of states is S = ¦NC, C¦, which stands
for “no carry” and “carry” respectively. The transition diagram is the
following:
start
ONML HIJK
NC
00/0
01/1
10/1
11/0
GFED @ABC
C
01/0
10/0
.
11/1
00/1
10.1.2. FiniteState Automata. A ﬁnitestate automaton is sim
ilar to a ﬁnitestate machine but with no output, and with a set of states
called accepting or ﬁnal states. More speciﬁcally, ﬁnitestate automa
ton consists of:
1. A ﬁnite set of states S.
2. A ﬁnite set of input symbols I.
3. A nextstate or transition function f : S I → S.
10.1. FINITE STATE MACHINES 135
4. An initial state σ ∈ S.
5. A subset F ⊆ S of accepting or ﬁnal states.
We represent the automaton A = (S, I, f, σ, F). We say that an au
tomaton accepts or recognizes a given string of input symbols if that
strings takes the automaton from the start state to a ﬁnal state.
Example: The following transition diagrams represent an automa
ton accepting any string of a’s and b’s ending with an a. The ﬁrst
diagram uses the same scheme as with ﬁnitestate machines, with 1
representing “accept” or “recognize”, and “0” representing “not ac
cept”:
start
GFED @ABC
σ
0
b/0
a/1
GFED @ABC
σ
1
b/0
a/1
The second kind of diagram omits the outputs and represents the
accepting states with double circles:
start
GFED @ABC
σ
0
b
a
GFED @ABC ?>=< 89:;
σ
1
b
a
Two ﬁnitestate automata that accept exactly the same set of strings
are said to be equivalent. For instance the following automaton also
accepts precisely strings of a’s abd b’s that end with an a, so it is
equivalent to the automaton shown above:
start
GFED @ABC
σ
0
b
a
GFED @ABC ?>=< 89:;
σ
1
b
a
GFED @ABC ?>=< 89:;
σ
2
b
a
Example: The following automaton accepts strings of a’s and b’s
with exactly an even number of a’s:
10.1. FINITE STATE MACHINES 136
start
GFED @ABC ?>=< 89:;
E
b
a
GFED @ABC
O
a
b
Example: The following automaton accepts strings starting with
one a followed by any number of b’s:
start
GFED @ABC
σ
0
a
b
GFED @ABC ?>=< 89:;
σ
1
a
b
GFED @ABC
σ
2
a
b
Example: The following automaton accepts strings ending with aba:
start
GFED @ABC
σ
0
a
b
GFED @ABC
σ
1
b
a
b
GFED @ABC
σ
2
a
b
.
GFED @ABC ?>=< 89:;
σ
3
b
a
10.2. LANGUAGES AND GRAMMARS 137
10.2. Languages and Grammars
10.2.1. Formal Languages. Consider algebraic expressions writ
ten with the symbols A = ¦x, y, z, +, ∗, (, )¦. The following are some
of them: “x + y ∗ y”, “y + (x ∗ y + y) ∗ x”, “(x + y) ∗ x + z”, etc.
There are however some strings of symbols that are not legitimate al
gebraic expressions, because they have some sort of syntax error, e.g.:
“(x + y”, “z + +y ∗ x”, “x(∗y) + z”, etc. So syntactically correct al
gebraic expressions are a subset of the whole set A
∗
of possible strings
over A.
In general, given a ﬁnite set A (the alphabet), a (formal) language
over A is a subset of A
∗
(set of strings of A).
Although in principle any subset of A
∗
is a formal language, we are
interested only in languages with certain structure. For instance: let
A = ¦a, b¦. The set of strings over A with an even number of a’s is a
language over A.
10.2.2. Grammars. A way to determine the structure of a lan
guage is with a grammar. In order to deﬁne a grammar we need two
kinds of symbols: nonterminal, used to represent given subsets of the
language, and terminal, the ﬁnal symbols that occur in the strings
of the language. For instance in the example about algebraic expres
sions mentioned above, the ﬁnal symbols are the elements of the set
A = ¦x, y, z, +, ∗, (, )¦. The nonterminal symbols can be chosen to
represent a complete algebraic expression (E), or terms (T) consisting
of product of factors (F). Then we can say that an algebraic expression
E consists of a single term
E → T,
or the sum of an algebraic expression and a term
E → E +T.
A term may consists of a factor or a product of a term and a factor
T → F
T → T ∗ F
A factor may consists of an algebraic expression between parenthesis
10.2. LANGUAGES AND GRAMMARS 138
F → (E),
or an isolated terminal symbol
F → x,
F → y,
F → z.
Those expressions are called productions, and tell us how we can
generate syntactically correct algebraic expressions by replacing suc
cessively the symbols on the left by the expressions on the right. For
instance the algebraic expression “‘y +(x∗y +y) ∗x” can be generated
like this:
E ⇒ E+T ⇒ T +T ⇒ F +T ⇒ y +T ⇒ y +T ∗F ⇒ y +F ∗F ⇒
y+(E)∗F ⇒y+(E+T)∗F ⇒y+(T+T)∗F ⇒y+(T∗F+T)∗F ⇒
y +(F ∗ F +T) ∗ F ⇒ y +(x ∗ T +T) ∗ F ⇒ y +(x ∗ F +T) ∗ F ⇒
y + (x ∗ y +T) ∗ F ⇒ y + (x ∗ y +F) ∗ T ⇒ y + (x ∗ y +y) ∗ F ⇒
y + (x ∗ y +y) ∗ x .
In general a phrasestructure grammar (or simply, grammar) G
consists of
1. A ﬁnite set V of symbols called vocabulary or alphabet.
2. A subset T ⊆ V of terminal symbols. The elements of N =
V −T are called nonterminal symbols or nonterminals.
3. A start symbol σ ∈ N.
4. A ﬁnite subset P of (V
∗
−T
∗
)V
∗
called the set of productions.
We write G = (V, T, σ, P).
A production (A, B) ∈ P is written:
A → B.
The right hand side of a production can be any combination of
terminal and nonterminal symbols. The left hand side must contain at
least one nonterminal symbol.
10.2. LANGUAGES AND GRAMMARS 139
If α → β is a production and xαy ∈ V
∗
, we say that xβy is directly
derivable from xαy, and we write
xαy ⇒ xβy .
If we have α
1
⇒ α
2
⇒ ⇒ α
n
(n ≥ 0), we say that α
n
is derivable
from α
1
, and we write α
1
∗
⇒ α
n
(by convention also α
1
∗
⇒ α
1
.)
Given a grammar G, the language L(G) associated to this grammar
is the subset of T
∗
consisting of all strings derivable from σ.
10.2.3. Backus Normal Form. The Backus Normal Form or
BNF is an alternative way to represent productions. The production
S → T is written S ::= T. Productions of the form S ::= T
1
, S ::= T
2
,
. . . , S ::= T
n
, can be combined as
S ::= T
1
[ T
2
[ [ T
n
.
So, for instance, the grammar of algebraic expressions deﬁned above
can be written in BNF as follows:
E ::= T [ E +T
T ::= F [ T ∗ F
F ::= (E) [ x [ y [ z
10.2.4. Combining Grammars. Let G
1
= (V
1
, T
1
, σ
1
, P
1
) and
G
2
= (V
2
, T
2
, σ
2
, P
2
) be two grammars, where N
1
= V
1
−T
1
and N
2
=
V
2
− T
2
are disjoint (rename nonterminal symbols if necessary). Let
L
1
= L(G
1
) and L
2
= L(G
2
) be the languages associated respectively
to G
1
and G
2
. Also assume that σ is a new symbol not in V
1
∪ V
2
.
Then
1. Union Rule: the language union of L
1
and L
1
L
1
∪ L
2
= ¦α [ α ∈ L
1
or α ∈ L
1
¦
starts with the two productions
σ → σ
1
, σ → σ
2
.
2. Product Rule: the language product of L
1
and L
2
L
1
L
2
= ¦αβ [ α ∈ L
1
, β ∈ L
1
¦
10.2. LANGUAGES AND GRAMMARS 140
where αβ = string concatenation of α and β, starts with the
production
σ → σ
1
σ
2
.
3. Closure Rule: the language closure of L
1
L
∗
1
= L
0
1
∪ L
1
1
∪ L
2
1
∪ . . .
were L
0
1
= ¦λ¦ and L
n
1
= ¦α
1
α
2
. . . α
n
[ α
k
∈ L
1
, k = 1, 2, . . . , n¦
(n = 1, 2, . . . ), starts with the two productions
σ → σ
1
σ , σ → λ.
10.2.5. Types of Grammars (Chomsky’s Classiﬁcation). Let
G be a grammar and let λ denote the null string.
0. G is a phrasestructure (or type 0) grammar if every production
is of the form:
α → δ ,
where α ∈ V
∗
−T
∗
, δ ∈ V
∗
.
1. G is a contextsensitive (or type 1) grammar if every production
is of the form:
αAβ → αδβ
(i.e.: we may replace A with δ in the context of α and β), where
α, β ∈ V
∗
, A ∈ N, δ ∈ V
∗
−¦λ¦.
2. G is a contextfree (or type 2) grammar if every production is
of the form:
A → δ ,
where A ∈ N, δ ∈ V
∗
.
3. G is a regular (or type 3) grammar if every production is of the
form:
A → a or A → aB or A → λ,
where A, B ∈ N, a ∈ T.
A language L is contextsensitive (respectively contextfree, regu
lar) if there is a contextsensitive (respectively contextfree, regular)
grammar G such that L = L(G).
The following examples show that these grammars deﬁne diﬀerent
kinds of languages.
10.2. LANGUAGES AND GRAMMARS 141
Example: The following language is type 3 (regular):
L = ¦a
n
b
m
[ n = 1, 2, 3 . . . ; m = 1, 2, 3, . . . ¦ .
A type 3 grammar for that language is the following: T = ¦a, b¦,
N = ¦σ, S¦, with start symbol σ, and productions:
σ → aσ , σ → aS , S → bS , S → b .
Example: The following language is type 2 (contextfree) but not
type 3:
L = ¦a
n
b
n
[ n = 1, 2, 3, . . . ¦ .
A type 2 grammar for that language is the following:
T = ¦a, b¦, N = ¦σ¦, with start symbol σ, and productions
σ → aσb , σ → ab .
Example: The following language is type 1 (contextsensitive) but
not type 2:
L = ¦a
n
b
n
c
n
[ n = 1, 2, 3, . . . ¦ .
A type 1 grammar for that language is the following:
T = ¦a, b, c¦, N = ¦σ, A, C¦, with start symbol σ, and productions
σ → abc , σ → aAbc ,
A → abC , A → aAbC ,
Cb → bC , Cc → cc .
There are also type 0 languages that are not type 1, but they are
harder to describe.
10.2.6. Equivalent Grammars. Two grammars G and G
are
equivalent if L(G) = L(G
).
Example: The grammar of algebraic expressions deﬁned at the be
ginning of the section is equivalent to the following one:
Terminal symbols = ¦x, y, z, +, ∗, (, )¦, nonterminal symbols = ¦E, T, F, L¦,
with start symbol E, and productions
E → T, E → E +T,
T → F, T → T ∗ F
10.2. LANGUAGES AND GRAMMARS 142
F → (E), F → L,
L → x, L → y, L → z.
10.2.7. ContextFree Interactive Lindenmayer Grammar.
A contextfree interactive Lindenmayer grammar is similar to a usual
contextfree grammar with the diﬀerence that it allows productions of
the form A → B where A ∈ N ∪T (in a context free grammar A must
be nonterminal). Its rules for deriving strings also are diﬀerent. In a
contextfree interactive Lindenmayer grammar, to derive string β from
string α, all symbols in α must be replaced simultaneously.
Example: The von Koch Snowﬂake. The von Koch Snowﬂake is a
fractal curve obtained by start with a line segment and then at each
stage transforming all segments of the ﬁgure into a four segment polyg
onal line, as shown below. The von Koch Snowﬂake fractal is the limit
of the sequence of curves deﬁned by that process.
Figure 10.1. Von Koch Snowﬂake, stages 1–3.
Figure 10.2. Von Koch Snowﬂake, stages 4–5
A way to represent an intermediate stage of the making of the
fractal is by representing it as a sequence of movements of three kinds:
’d’= draw a straight line (of a ﬁx length) in the current direction, ’r’=
turn right by 60
◦
, ’l’= turn left by 60
◦
. For instance we start with a
single horizontal line d, which we then transform into the polygonal
dldrrdld, then each segment is transformed into a polygonal according
to the rule d → dldrrdld, so we get
dldrrdldldldrrdldrrdldrrdldldldrrdld
If we represent by D a segment that may no be ﬁnal yet, then the
sequences of commands used to build any intermediate stage of the
curve can be deﬁned with the following grammar:
10.2. LANGUAGES AND GRAMMARS 143
N = ¦D¦, T = ¦d, r, l¦, with start symbol D, and productions:
D → DlDrrDlD, D → d , r → r , l → l .
Example: The Peano curve. The Peano curve is a space ﬁlling curve,
i.e., a function f : [0, 1] → [0, 1]
2
such that the range of f is the whole
square [0, 1]
2
, deﬁned as the limit of the sequence of curves shown in
the ﬁgures below.
Figure 10.3. Peano curve, stages 1–4.
Each element of that sequence of curves can be described as a se
quence of 90
◦
arcs drawn either anticlockwise (’l’) or clockwise (’r’).
The corresponding grammar is as follows:
T = ¦l, r¦, N = ¦C, L, R¦, with and start symbol C, and productions
C → LLLL,
L → RLLLR, R → RLR,
L → l , R → r , l → l , r → r .
10.3. LANGUAGE RECOGNITION 144
10.3. Language Recognition
10.3.1. Regular Languages. Recall that a regular language is
the language associated to a regular grammar, i.e., a grammar G =
(V, T, σ, P) in which every production is of the form:
A → a or A → aB or A → λ,
where A, B ∈ N = V −T, a ∈ T.
Regular languages over an alphabet T have the following properties
(recall that λ = ’empty string’, αβ = ’concatenation of α and β’, α
n
=
’α concatenated with itself n times’):
1. ∅, ¦λ¦, and ¦a¦ are regular languages for all a ∈ T.
2. If L
1
and L
2
are regular languages over T the following lan
guages also are regular:
L
1
∪ L
2
= ¦α [ α ∈ L
1
or α ∈ L
2
¦
L
1
L
2
= ¦αβ [ α ∈ L
1
, β ∈ L
2
¦
L
∗
1
= ¦α
1
. . . α
n
[ α
k
∈ L
1
, n ∈ N¦ ,
T
∗
−L
1
= ¦α ∈ T
∗
[ α / ∈ L
1
¦ ,
L
1
∩ L
2
= ¦α [ α ∈ L
1
and α ∈ L
2
¦ .
We justify the above claims about L
1
∪L
2
, L
1
L
2
and L
∗
1
as follows.
We already know how to combine two grammars (see 10.2.4) L
1
and
L
2
to obtain L
1
∪ L
2
, L
1
L
2
and L
∗
1
, the only problem is that the rules
given in section 10.2.4 do no have the form of a regular grammar, so we
need to modify them slightly (we use the same notation as in section
10.2.4):
1. Union Rule: Instead of adding σ → σ
1
and σ → σ
2
, add all
productions of the form σ → RHS, where RHS is the right
hand side of some production (σ
1
→ RHS) ∈ P
1
or (σ
2
→
RHS) ∈ P
2
.
2. Product Rule: Instead of adding σ → σ
1
σ
2
, use σ
1
as start
symbol and replace each production (A → a) ∈ P
1
with A →
aσ
2
and (A → λ) ∈ P
1
with A → σ
2
.
3. Closure Rule: Instead of adding σ → σ
1
σ and σ → λ, use
σ
1
as start symbol, add σ
1
→ λ, and replace each production
(A → a) ∈ P
1
with A → aσ
1
and (A → λ) ∈ P
1
with A → σ
1
.
10.3. LANGUAGE RECOGNITION 145
10.3.2. Regular Expressions. Regular languages can be charac
terized as languages deﬁned by regular expressions. Given an alphabet
T, a regular expression over T is deﬁned recursively as follows:
1. ∅, λ, and a are regular expressions for all a ∈ T.
2. If R and S are regular expressions over T the following expres
sions are also regular: (R), R +S, R S, R
∗
.
In order to use fewer parentheses we assign those operations the fol
lowing hierarchy (from do ﬁrst to do last): ∗, , +. We may omit the
dot: α β = αβ.
Next we deﬁne recursively the language associated to a given regular
expression:
L(∅) = ∅ ,
L(λ) = ¦λ¦ ,
L(a) = ¦a¦ for each a ∈ T,
L(R +S) = L(R) ∪ L(S) ,
L(R S) = L(R)L(S) (language product),
L(R
∗
) = L(R)
∗
(language closure).
So, for instance, the expression a
∗
bb
∗
represents all strings of the
form a
n
b
m
with n ≥ 0, m > 0, a
∗
(b +c) is the set of strings consisting
of any number of a’s followed by a b or a c, a(a + b)
∗
b is the set of
strings over ¦a, b¦ than start with a and end with b, etc.
Another way of characterizing regular languages is as sets of strings
recognized by ﬁnitestate automata, as we will see next. But ﬁrst we
need a generalization of the concept of ﬁnitestate automaton.
10.3.3. Nondeterministic FiniteState Automata. A nonde
terministic ﬁnitestate automaton is a generalization of a ﬁnitestate
automaton so that at each state there might be several possible choices
for the “next state” instead of just one. Formally a nondeterministic
ﬁnitestate automaton consists of
1. A ﬁnite set of states S.
2. A ﬁnite set of input symbols I.
3. A nextstate or transition function f : S I → P(S).
4. An initial state σ ∈ S.
10.3. LANGUAGE RECOGNITION 146
5. A subset F of S of accepting or ﬁnal states.
We represent the automaton A = (S, I, f, σ, F). We say that a nonde
terministic ﬁnitestate automaton accepts or recognizes a given string
of input symbols if in its transition diagram there is a path from the
starting state to a ﬁnal state with its edges labeled by the symbols of
the given string. A path (which we can express as a sequence of states)
whose edges are labeled with the symbols of a string is said to represent
the given string.
Example: Consider the nondeterministic ﬁnitestate automaton de
ﬁned by the following transition diagram:
start
GFED @ABC
σ
b
a
GFED @ABC
C
b
b
GFED @ABC ?>=< 89:;
F
This automaton recognizes precisely the strings of the form b
n
ab
m
,
n ≥ 0, m > 0. For instance the string bbabb is represented by the path
(σ, σ, σ, C, C, F). Since that path ends in a ﬁnal state, the string is
recognized by the automaton.
Next we will see that there is a precise relation between regular
grammars and nondeterministic ﬁnitestate automata.
Regular grammar associated to a nondeterministic ﬁnitestate au
tomaton. Let A be a nondeterministic ﬁnitestate automaton given as
a transition diagram. Let σ be the initial state. Let T be the set of
inputs symbols, let N be the set of states, and V = N ∪ T. Let P be
the set of productions
S → xS
if there is an edge labeled x from S to S
and
S → λ
if S is a ﬁnal state. Let G be the regular grammar
G = (V, T, σ, P) .
Then the set of strings recognized by A is precisely L(G).
Example: For the nondeterministic automaton deﬁned above the
corresponding grammar will be:
10.3. LANGUAGE RECOGNITION 147
T = ¦a, b¦, N = ¦σ, C, F¦, with productions
σ → bσ , σ → aC , C → bC , C → bF , F → λ.
The string bbabb can be produced like this:
σ ⇒ bσ ⇒ bbσ ⇒ bbaC ⇒ bbabC ⇒ bbabbF ⇒ bbabb .
Nondeterministic ﬁnitestate automaton associated to a given regu
lar grammar. Let G = (V, T, σ, P) be a regular grammar. Let
I = T.
S = N ∪ ¦F¦, where N = V −T, and F / ∈ V .
f(S, x) = ¦S
[ S → xS
∈ P¦ ∪ ¦F [ S → x ∈ P¦.
F = ¦F¦ ∪ ¦S [ S → λ ∈ P¦ .
Then the nondeterministic ﬁnitestate automaton A = (S, I, f, σ, F)
recognizes precisely the strings in L(G).
10.3.4. Relationships Between Regular Languages and Au
tomata. In the previous section we saw that regular languages co
incide with the languages recognized by nondeterministic ﬁnitestate
automata. Here we will see that the term “nondeterministic” can be
dropped, so that regular languages are precisely those recognized by
(deterministic) ﬁnitestate automata. The idea is to show that given
any nondeterministic ﬁnitestate automata it is possible to construct
an equivalent deterministic ﬁnitestate automata recognizing exactly
the same set of strings. The main result is the following:
Let A = (S, I, f, σ, F) be a nondeterministic ﬁnitestate automaton.
Then Ais equivalent to the ﬁnitestate automaton A
= (S
, I
, f
, σ
, F
),
where
1. S
= P(S).
2. I
= I.
3. σ
= ¦σ¦.
4. F
= ¦X ⊆ S [ X ∩ F = ∅¦.
5. f
(X, x) =
¸
S∈X
f(S, x) , f
(∅, x) = ∅ .
10.3. LANGUAGE RECOGNITION 148
Example: Find a (deterministic) ﬁnitestate automaton A
equiva
lent to the following nondeterministic ﬁnitestate automaton A:
start
GFED @ABC
σ
b
a
GFED @ABC
C
b
b
GFED @ABC ?>=< 89:;
F
Answer: The set of input symbols is the same as that of the given
automaton: I
= I = ¦a, b¦. The set of states is the set of subsets of
S = ¦σ, C, F¦, i.e.:
S
= ¦∅, ¦σ¦, ¦C¦, ¦F¦, ¦σ, C¦, ¦σ, F¦, ¦C, F¦, ¦σ, C, F¦¦ .
The starting state is ¦σ¦. The ﬁnal states of A
are the elements of S
containing some ﬁnal state of A:
F
= ¦¦F¦, ¦σ, F¦, ¦C, F¦, ¦σ, C, F¦¦ .
Then for each element X of S
we draw an edge labeled x from X
to
¸
S∈X
f(S, x) (and from ∅ to ∅):
start
GF ED
@A BC ¦σ¦
a
b
GF ED
@A BC ¦C¦
a
b
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
6
GF ED
@A BC
?> =<
89 :; ¦σ, C, F¦
a
b
GF ED
@A BC
?> =<
89 :; ¦σ, F¦
a
x
x
x
x
x
x
x
x
b
GFED
@ABC
∅
a
b
.
GF ED
@A BC ¦σ, C¦
a
I
I
I
I
I
I
I
I
I
b
GF ED
@A BC
?> =<
89 :; ¦F¦
a
b
GF ED
@A BC
?> =<
89 :; ¦C, F¦
a
I
I
I
I
I
I
I
I
I
I
I
b
We notice that some states are unreachable from the starting state.
After removing the unreachable states we get the following simpliﬁed
version of the ﬁnitestate automaton:
start
GF ED
@A BC ¦σ¦
a
b
GF ED
@A BC ¦C¦
a
b
GF ED
@A BC
?> =<
89 :; ¦C, F¦
a
b
GFED
@ABC
∅
a
b
_
10.3. LANGUAGE RECOGNITION 149
So, once proved that every nondeterministic ﬁnitestate automaton
is equivalent to some deterministic ﬁnitestate automaton, we obtain
the main result of this section: A language L is regular if and only
if there exists a ﬁnitestate automaton that recognizes precisely the
strings in L.
APPENDIX A
A.1. Eﬃcient Computation of Powers Modulo m
We illustrate an eﬃcient method of computing powers modulo m
with an example. Assume that we want to compute 3
547
mod 10.
First write 547 in base 2: 1000100011, hence 547 = 2
9
+ 2
5
+ 2 + 1 =
((2
4
+1) 2
4
+1) 2+1, so: 3
547
= ((3
2
4
3)
2
4
3)
2
3. Next we compute the
expression beginning with the inner parenthesis, and reducing modulo
10 at each step: 3
2
= 9 (mod 10), 3
2
2
= 9
2
= 81 = 1 (mod 10),
3
2
3
= 1
2
= 1 (mod 10), 3
2
4
= 1
2
= 1 (mod 10), 3
2
4
3 = 1 3 = 3
(mod 10), etc. At the end we ﬁnd 3
547
= 7 (mod 10).
The algorithm in pseudocode would be like this:
1: procedure pow mod(a,x,m) ¦computes a^x mod m¦
2: p := 1
3: bx := binary array(x) ¦x as a binary array¦
4: t := a mod m
5: for k := 1 to length(bx)
6: begin
7: p := (p * p) mod m
8: if bx[k] = 1 then
¦if kth binary digit of x is 1¦
9: p := (p * t) mod m
10: end
11: return p
12: end pow mod
150
A.1. EFFICIENT COMPUTATION OF POWERS MODULO M 151
The following is a program in C implementing the algorithm:
int pow(int a, int x, int m) {
int p = 1;
int y = (1 << (8 * size of(int)  2));
a %= m;
while (!(y & x)) y >>= 1;
while (y) {
p *= p;
p %= m;
if (x & y) {
p *= a;
p %= m;
}
y >>= 1;
}
return p;
}
The following is an alternative algorithm equivalent to running
through the binary representation of the exponent from right to left
instead of left to right:
1: procedure pow mod(a,x,m) ¦computes a^x mod m¦
2: p := 1
3: t := a mod m
4: while x > 0
5: begin
6: if x is odd then
7: p := (p * t) mod m
8: t := (t * t) mod m
9: x := floor(x/2)
10: end
11: return p
12: end pow mod
A.2. MACHINES AND LANGUAGES 152
A.2. Machines and Languages
A.2.1. Turing Machines. A Turing machine is a theoretical de
vice intended to deﬁne rigorously the concept of algorithm. It consists
of
1. An inﬁnite tape made of a sequence of cells. Each cell may be
empty or may contain a symbol from a given alphabet.
2. A control unit containing a ﬁnite set of instructions.
3. A tape head able to read and write (or delete) symbols from the
tape.
Tape head
Tape
control
unit
Figure A.1. Turing Machine.
Each machine instruction contains the following ﬁve parts:
1. The current machine state.
2. A tape symbol read from the current tape cell.
3. A tape symbol to write into the current tape cell.
4. A direction for the tape head to move: L = ’move one cell to
the left’, R = ’move one cell to the right’, S = ’stay in the
current cell’.
5. The next machine state.
Turing machines are generalizations of ﬁnitestate automata. A
ﬁnitestate automaton is just a Turing machine whose tape head moves
always from left to right and never writes to the tape. The input of
the ﬁnitestate automaton is presented as symbols written in the tape.
In general we make the following assumptions:
1. An input is represented on the tape by placing the letters of
the strings in contiguous tape cells. All other cells contain the
blank symbol, which we may denote λ.
A.2. MACHINES AND LANGUAGES 153
2. The tape is initially positioned at the leftmost cell of the input
string unless speciﬁed otherwise.
3. There is one starting state.
4. There is one halt state, which we denote by “Halt”.
The execution of a Turing machine stops when it enters the Halt state
or when it enters a state for which there is no valid move. The output
of the Turing machine is the contents of the tape when the machine
stops.
We say that an input string is accepted by a Turing machine if
the machine enters the Halt state. Otherwise the string is rejected.
This can happen in two ways: by entering a state other than the Halt
state from which there is no move, or by running forever (for instance
executing an inﬁnite loop).
If a Turing machine has at least two instructions with the same state
and input letter, then the machine is nondeterministic. Otherwise it is
deterministic.
FiniteState Automata. A ﬁnitestate automata can be interpreted
as a Turing machine whose tape head moves only from left to right and
never writes to the tape.
Pushdown Automata. A pushdown automaton is ﬁnitestate au
tomaton with a stack, i.e., a storage structure in which symbols can be
put and extracted from it by two operations: push (place on the top of
the stack) and pop (take from the top of the stack)—consequently the
last symbol put into the stack is the ﬁrst symbol taken out. Addition
ally there is a third operation, nop, that leaves the stack intact. The
next state function takes into account not only the current state and
the symbol read from the input, but also the symbol at the top of the
stack. After reading the next input symbol and the symbol at the top
of the stack, the automaton executes a stack operation and goes to the
next state. Initially there is a single symbol in the stack.
Linearly Bounded Automata. A linearly bounded automaton is a
Turing machine whose tape is limited to the size of its input string
plus two boundary cells that may not be changed.
Computable Functions. Consider a Turing machine T working on
symbols from an alphabet of only one symbol A = ¦[¦ (“stroke”). Let
f : N → N the function deﬁned so that f(n) = m means that if the
A.2. MACHINES AND LANGUAGES 154
initial input of T consists of a string of n + 1 strokes, the output of T
is a string of m+ 1 strokes. We say that f is computed by the Turing
machine T. A computable function is a function computed by some
Turing machine. A computable function f(n) halts for a given value
of its argument n if T with input n + 1 strokes halts. A computable
function f is total if f(n) halts for every n.
An eﬀective enumeration of a set is a listing of its elements by an
algorithm.
A.2.2. Hierarchy of Languages. Here we mention a hierarchy
of languages that includes (and extends) Chomsky’s classiﬁcation, in
increasing order of inclusion.
1. Regular languages. They are recognized by ﬁnitestate automata.
Example: ¦a
m
b
n
[ m, n = 1, 2, 3 . . . ¦.
2. Deterministic contextfree languages, recognized by determinis
tic pushdown automata. Example: ¦a
n
b
n
[ n = 1, 2, 3 . . . ¦.
3. Contextfree languages, recognized by nondeterministic push
down automata. Example: palindromes over ¦a, b¦.
4. Contextsensitive languages, languages without λ recognized by
linearly bounded automata. Example: ¦a
n
b
n
c
n
[ n = 1, 2, 3 . . . ¦
5. Unrestricted or phrasestructure grammars, recognized by Tur
ing machines.
6. Recursively enumerable languages. A language is recursively
enumerable if there is a Turing machine that outputs all the
strings of the language. Example: ¦a
n
[ f
n
(n) halts¦, where
f
0
, f
1
, f
2
, . . . is an eﬀective enumeration of all computable func
tions.
7. Nongramatical languages, languages that are not deﬁnable by
any grammar and cannot be recognized by Turing machines.
Example: ¦a
n
[ f
n
is total¦.
Contents
Introduction Chapter 1. Logic, Proofs 1.1. Propositions 1.2. Predicates, Quantiﬁers 1.3. Proofs Chapter 2. Sets, Functions, Relations 2.1. Set Theory 2.2. Functions 2.3. Relations Chapter 3. Algorithms, Integers 3.1. Algorithms 3.2. The Euclidean Algorithm 3.3. Modular Arithmetic, RSA Algorithm Chapter 4. Induction, Recurences 4.1. Sequences and Strings 4.2. Mathematical Induction 4.3. Recurrence Relations Chapter 5. Counting 5.1. Basic Principles 5.2. Combinatorics 5.3. Generalized Permutations and Combinations 5.4. Binomial Coeﬃcients 5.5. The Pigeonhole Principle Chapter 6. Probability 6.1. Probability Chapter 7. Graph Theory 7.1. Graphs 7.2. Representations of Graphs 7.3. Paths and Circuits
3
5 6 6 11 13 19 19 27 32 38 38 48 52 59 59 62 65 69 69 71 73 75 77 78 78 82 82 88 91
3.CONTENTS 4 7.1.2. Eﬃcient Computation of Powers Modulo m A.2.5.1.1.2. Automata. Grammars and Languages Finite State Machines Languages and Grammars Language Recognition 97 100 100 102 104 113 116 122 122 127 133 133 137 144 150 150 152 Appendix A. Planar Graphs Chapter 8. Tree Isomorphisms 8.3. 10. Boolean Functions. Machines and Languages .4. A.2. Trees 8. Trees 8. Decision Trees. Spanning Trees Chapter 9. Combinatorial Circuits 9. 10. Boolean Algebras 9. Applications Chapter 10. Tree Transversal 8. Binary Trees 8.4.1. 10.
northwestern. Finally. or you have any suggestions. They are not homework nor need to be addressed at all if the reader does not wish to. McGrawHill. With few exceptions I will follow the notation in the book. Lerma mlerma@math. so these notes will not be completely ﬁnished until the end of the quarter. The textbook for this course is Keneth H. do not hesitate to let me know.Introduction These notes are intended to be a summary of the main ideas in course CS 310: Mathematical Foundations of Computer Science.edu Northwestern University Spring 2005 http://www. I may keep working on this document as the course goes on. if you ﬁnd any typos or errors.edu/~mlerma/courses/cs31005s/ 5 . 2003. Fifth Edition. please.northwestern. Rosen: Discrete Mathematics and Its Applications. These notes contain some questions and “exercises” intended to stimulate the reader who wants to play a somehow active role while studying the subject. I will recommend exercises and give homework assignments separately. Miguel A.math.
Connectives are used for making compound propositions. “2 < 4” (true). “4 = 7 (false)”. the following are propositions: “Paris is in France” (true). Proofs 1. However the following are not propositions: “what is your name?” (this is a question). Writing F for “false” and T for “true”. but not both” Implication p→q “if p then q” Biconditional p↔q “p if and only if q” The truth value of a compound proposition depends only on the value of its components.1. 1. “do your homework” (this is a command). Connectives. “x is an even number” (it depends on what x represents). Propositions A proposition is a declarative sentence that is either true or false (but not both).1. “Socrates” (it is not even a sentence). For instance. “London is in Denmark” (false). we can summarize the meaning of the connectives in the following way: 6 .1. “this sentence is false” (neither true nor false). The truth or falsehood of a proposition is called its truth value.CHAPTER 1 Logic. Truth Tables. The main ones are the following (p and q represent given propositions): Name Represented Meaning Negation ¬p “not p” Conjunction p∧q “p and q” Disjunction p∨q “p or q (or both)” Exclusive Or p⊕q “either p or q.
So.1.1. .e. i. A proposition of the form “if p then q” or “p implies q”. 2. A proposition is said to be a contradiction if its truth value is F for any assignment of truth values to its components. PROPOSITIONS 7 p T T F F q ¬p p ∧ q p ∨ q p ⊕ q p → q p ↔ q T F T T F T T F F F T T F F T T F T T T F F T F F F T T Note that ∨ represents a nonexclusive or. 1. “if London is in Denmark then 2 < 4” (false → true).3. A proposition that is neither a tautology nor a contradiction is called a contingency. For instance: “if John is from Chicago then John is from Illinois”. On the other hand ⊕ represents an exclusive or. Contingency. and the proposition q is the conclusion or consequent. Tautology.. Conditional Propositions. the following sentences are true: “if 2 < 4 then Paris is in France” (true → true). 1. A proposition is said to be a tautology if its truth value is T for any assignment of truth values to its components.. p ∨ q is true when any of p. Example: The proposition p ∨ ¬p is a tautology. i. Note that p → q is true always except when p is true and q is false.1. q is true and also when both are true. represented “p → q” is called a conditional proposition.e. p ¬p p ∨ ¬p p ∧ ¬p T F T F T F T F F T T F F T T F 6 6 tautology contradiction 1. 3. p ⊕ q is true only when exactly one of p and q is true. Contradiction. Example: The proposition p ∧ ¬p is a contradiction.1.2. The proposition p is called hypothesis or antecedent.
For instance p → q and ¬p ∨ q are logically equivalent. and we write it: p → q ≡ ¬p ∨ q Note that that two propositions A and B are logically equivalent precisely when A ↔ B is a tautology. i. is called biconditional.. they are both true or both false. PROPOSITIONS 8 “if 4 = 7 then London is in Denmark” (false → false).1. In might seem strange that “p → q” is considered true when p is false. Logical Equivalence. read “p if and only if q”. Example: De Morgan’s Laws for Logic.4. regardless of the truth value of q. they are called logically equivalent.e. although for the particular case x = 3 it becomes “if 3 is a multiple of 4 then 3 is a multiple of 2”. However the following one is false: “if 2 < 4 then London is in Denmark” (true → false). Note that the compound propositions p → q and ¬p ∨ q have the same truth values: p T T F F q ¬p ¬p ∨ q p → q T F T T F F F F T T T T F T T T When two compound propositions have the same truth values no matter what truth value their constituent propositions have. It is true precisely when p and q have the same truth value. 1. The following propositions are logically equivalent: ¬(p ∨ q) ≡ ¬p ∧ ¬q ¬(p ∧ q) ≡ ¬p ∨ ¬q We can check it by examining their truth tables: .1.1. This will become clearer when we study predicates such as “if x is a multiple of 4 then x is a multiple of 2”. That implication is obviously true. The proposition p ↔ q.
As we have seen. Contrapositive. PROPOSITIONS 9 p T T F F q ¬p ¬q p ∨ q ¬(p ∨ q) ¬p ∧ ¬q p ∧ q ¬(p ∧ q) ¬p ∨ ¬q T F F T F F T F F F F T T F F F T T T T F T F F F T T F T T F T T F T T Example: The following propositions are logically equivalent: p ↔ q ≡ (p → q) ∧ (q → p) Again. for instance. Note that the converse is not equivalent to the given conditional proposition. Converse. but the converse “if John is from Illinois then John is from Chicago” may be false. p ↔ q ≡ (p → q) ∧ (q → p) So.1.1. saying that “John is married if and only if he has a spouse” is the same as saying “if John is married then he has a spouse” and “if he has a spouse then he is married”. the biconditional proposition is equivalent to the conjunction of a conditional proposition an its converse. The converse of a conditional proposition p → q is the proposition q → p.1. this can be checked with the truth tables: p T T F F q p → q q → p (p → q) ∧ (q → p) p ↔ q T T T T T F F T F F T T F F F F T T T T Exercise: Check the following logical equivalences: ¬(p → q) ≡ p ∧ ¬q p → q ≡ ¬q → ¬p ¬(p ↔ q) ≡ p ⊕ q 1.5. for instance “if John is from Chicago then John is from Illinois” is true. .
They are logically equivalent.1.1. For instance the contrapositive of “if John is from Chicago then John is from Illinois” is “if John is not from Illinois then John is not from Chicago”. . PROPOSITIONS 10 The contrapositive of a conditional proposition p → q is the proposition ¬q → ¬p.
Analogously. while the universe of discourse of n is the set of natural numbers. ∃x Q(x) would be true if we extend the universe of discourse to the rational numbers. etc. “X is American”. which is true.2.1. Q(x. namely 5. has a deﬁnite truth value. represented “∃x P (x)”. so it is a proposition in the usual sense. For instance if we replace x with 1 in the predicate “x + 2 = 7” we obtain “1 + 2 = 7”. The symbol ∃ is called the existential quantiﬁer. Quantiﬁers 1. so the universe of discourse of σ is the set of all strings. the statement “for some x. Predicates. “x < y”. the sentence “for all x. P (x)”. represented “∀x P (x)”. Each variable in a predicate is assumed to belong to a universe (or domain) of discourse. P (x)”—. P (x)”—also “for any x. if Q(x) is “2x = 7” and the universe of discourse is still the integers. y). However. 1Usually . For instance in the predicate “σ is a string of length n” the variable σ represents a string. “for every x.1. Predicates. An example for P (x) is a value of x for which P (x) is true.2. PREDICATES. A predicate or propositional function is a statement containing variables. such that P (5) is a true statement. for instance in the predicate “n is an odd integer” ’n’ represents an integer.2. A counterexample is a value of x for which P (x) is false. while 1 is a counterexample. QUANTIFIERS 11 1.1 1. while n represents a natural number. since there is indeed an integer. In “X is American” we may assume that X is a human being.2. then ∃x P (x) is true.2. For instance if P (x) is “x + 2 = 7” with the integers as universe of discourse. so the universe of discourse of n is the set of all integers. but in some situations (as in the so called manysorted logics) it is possible to use diﬀerent kinds of variables to represent diﬀerent types of objects belonging to diﬀerent universes of discourse. but if we replace it with 5 we get “5 + 2 = 7”. P (x)”. then ∃x Q(x) is false. which is false. We represent a predicate by a letter followed by the variables enclosed between parenthesis: P (x). has a deﬁnite truth value. “for each x. if P (x) is “x + 2 = 7” and the all variables occurring in predicates along a reasoning are supposed to belong to the same universe of discourse. So. 5 is an example for “x + 2 = 7”. Given a predicate P (x). The truth value of the predicate depends on the value assigned to its variables. “p is a prime number” are predicates. For instance “x + 2 = 7”. P (x)” (or “there is some x such that p(x)”). Quantiﬁers. so in this case the universe of discourse is the set of all human beings. For instance. On the other hand.
A predicate can be partially quantiﬁed. e. y) . P (x) holds. y) means something diﬀerent from ∃y∀x P (x. (Note that among real numbers x < y is equivalent to x ≥ y.. Hence ¬∃x P (x) ≡ ∀x ¬P (x) .2. Write the negation of that statement. Generalized De Morgan Laws for Logic. Its negation is: ∃x ∀y ¬(x < y). y. then ∀x∃y P (x. for instance: ¬∃x∀y p(x. A partially quantiﬁed predicate is still a predicate. y) represents “x is a friend of y”. but ∃y∀x P (x. then ∀x P (x) is false. The variables quantiﬁed (x and y in the example) are called bound variables. ∀x∃y P (x. z). PREDICATES. P (x) is always false. Answer : The statement is: ∀x ∃y (x < y) (the universe of discourse is the real numbers). In predicates with more than one variable it is possible to use several quantiﬁers at the same time. or in other words. 1. z)”. Thus: ¬∀x P (x) ≡ ∃x ¬P (x) . y) ≡ ∀x¬∀y P (x. but depending on fewer variables. y) ≡ ∀x∃y ¬P (x.2. P (x) must be false. for instance ∀x∀y∃z P (x.1.. This two rules can be applied in successive steps to ﬁnd the negation of a more complex quantiﬁed statement. y. For instance if x and y represent human beings and P (x. but formally they are diﬀerent predicates. t). y) means that everybody is a friend of someone.) . y) means that there is someone such that everybody is his or her friend. i. Note that in general the existential and universal quantiﬁers cannot be swapped. if ∀x P (x) is false then it is not true that for every x. However if Q(x) represents “(x + 1)2 = x2 + 2x + 1” then ∀x Q(x) is true. z. ∃x ∀y (x < y). i. y.3.e.g. in general ∀x∃y P (x. y). QUANTIFIERS 12 universe of discourse is the integers.e. If ∃x P (x) is false then there is no value of x for which P (x) is true. meaning “for all x and all y there is some z such that P (x. Exercise: Write formally the statement “for every real number there is a greater real number”. and the rest (z and t in the example) are called free variables. The symbol ∀ is called the universal quantiﬁer. hence for some x. On the other hand.
Assume to the contrary that “x > 2 or y > 3” is false. An argument that establishes the truth of a proposition is called a proof. 3. a proposition of the form r ∧ ¬r. Deﬁnitions: used to create new concepts from old ones. Proof by Contradiction.e.3. Proofs. so x ≤ 2 and y ≤ 3. Answer : Assuming x > 2 and y > 3 and adding the inequalities term by term we get: x + y > 2 + 3 = 5. From here we must conclude that x > 2 or y > 3.1. and try to derive a contradiction. In a proof by contradiction or (Reductio ad Absurdum) we assume the hypotheses and the negation of the conclusion.. A theorem is a proposition that can be proved to be true.e. An indirect proof consists of proving ¬((x > 2) ∨ (y > 3)) → ¬(x + y > 5). Example: Prove that if x > 2 and y > 3 then x + y > 5. PROOFS 13 1. Example: Prove that if x + y > 5 then x > 2 or y > 3. so adding both inequalities we get x + y ≤ 5. A Mathematical System consists of: 1. An indirect proof or proof by contrapositive consists of proving the contrapositive of the desired implication. Example: Prove by contradiction that if x + y > 5 then either x > 2 or y > 3.1. In a direct proof we assume the hypothesis together with axioms and other theorems previously proved and we derive the conclusion from them. Answer : We assume the hypothesis x + y > 5. In fact: ¬((x > 2) ∨ (y > 3)) is the same as (x ≤ 2)∧(y ≤ 3). 2. Adding those inequalities we get . instead of proving p → q we prove ¬q → ¬p. Mathematical Systems. Proofs 1. Undeﬁned terms: corresponding to the primitive concepts of the system (for instance in set theory the term “set” is undeﬁned). Axioms: propositions that are assumed true. That is an example of direct proof. i. which is the same as ¬(x + y > 5).3. i.3.. Answer : We must prove that x + y > 5 → (x > 2) ∨ (y > 3).
otherwise it is called invalid. . Hence: 2 b2 = 4 a 2 . Consequently a/b = 2a /2b = a /b . pn called hypotheses (or premises) followed by a proposition q called conclusion. pn are true. An argument is a sequence of propositions p1 . . Arguments. . then a2 is even.e.3. Squaring both sides we have 2 = a2 /b2 .. Rules of Inference. From here we conclude that the assumption “x ≤ 2 and y ≤ 3” cannot be right. p2 . . p2 . . so “x > 2 or y > 3” must be true. so a = 2 a . . there are no integers a. Since the left hand side is even. . . and simplifying: b2 = 2 a 2 . where a and b are integers and the fraction is written in least terms. . √ √ Answer : Assume that 2 is rational.e. 2 = a/b.. .2. . b such that 2 = a/b. √ i. which contradicts the hypothesis x + y > 5.3. In an indirect proof we prove an implication of the form p → q by proving the contrapositive ¬q → ¬p. . An argument is usually written: p1 p2 . This implies that b2 is even. so b is even: b = 2b . contradicting the hypothesis that a/b was in least terms. but this implies that a itself is even. PROOFS 14 x ≤ 2 + 3 = 5. Remark : Sometimes it is diﬃcult to distinguish between an indirect proof and a proof by contradiction. 1. i. . √ Exercise: Prove by contradiction that 2 is not a rational number. .1. pn ∴ q or p 1 . In fact proofs by contradiction are more general than indirect proofs. hence 2 b2 = a2 . p2 . In an proof by contradiction we prove an statement s (which may or may not be an implication) by assuming ¬s and deriving a contradiction. . p n / ∴ q The argument is called valid if q is true whenever p1 .
3. Other rules of inference are the following: 1. Addition: p ∴ p∨q 4. Simpliﬁcation: p∧q ∴ p 5.1. Modus Tollens: p→q ¬q ∴ ¬p 3. Modus Ponens or Rule of Detachment: p→q p ∴ q 2. PROOFS 15 Rules of inference are certain simple arguments known to be valid and used to make a proof step by step. For instance the following argument is called modus ponens or rule of detachment: p→q p ∴ q In order to check whether it is valid we must examine the following truth table: p T T F F q p→q T T F F T T F T p T T F F q T F T F If we look now at the rows in which both p → q and p are true (just the ﬁrst row) we see that also q is true. Conjunction: . so the argument is valid.
Hypothetical Syllogism: p→q q→r ∴ p→r 7. The statement is a premise. If I walk I get tired. Modus Tollens Premise 4.3. W → T and ¬T . PROOFS 16 p q ∴ p∧q 6.1. Each row contains a label. Disjunctive Syllogism: p∨q ¬p ∴ q 8. W = “I walk” and T = “I get tired”.3. The statement can be derived from statements occurring earlier in the argument by using a rule of inference. Therefore I take the bus. 2. a statement and the reason that justiﬁes the introduction of that statement in the argument.2. I do not get tired. The argument can be described in the following steps: step statement reason 1) 2) 3) 4) 5) W →T ¬T ¬W B∨W ∴B Premise Premise 1. Disjunctive Syllogism . Example: Consider the following statements: “I take the bus or I walk. The premises are B ∨ W . and the conclusion is B. That justiﬁcation can be one of the following: 1. Resolution: p∨q ¬p ∨ r ∴ q∨r Arguments are usually written using three columns.” We can formalize this by calling B = “I take the bus”.
then p(a) is true for each speciﬁc element a in the universe of discourse.: ∃x p(x) ∴ p(a) The diﬀerence respect to the previous rule is the restriction in the meaning of a. such √ that (± 2)2 = 2. i. 1. 4.: p(x) ∴ ∀x p(x) By “generic” we mean an element for which we do not make any assumption other than its belonging to the universe of discourse. PROOFS 17 1. Example: Show that a counterexample can be used to disprove a universal statement.e. from ∀x (x+1 = 1+x) we can derive 7+1 = 1+7. i. If ∃x p(x) is true. 2. . for instance. but they can be generalized to predicates with two or more variables. for instance. Rules of Inference for Quantiﬁed Statements. i. which now represents some (not any) element of the universe of discourse.e. then ∃x p(x) is true.3. 3. Existential Generalization. So. we can prove ∀x [(x + 1)2 = x2 + 2x + 1] (say.: p(a) ∴ ∃x p(x) For instance: from 7 + 1 = 8 we can derive ∃x (x + 1 = 8). i. for real numbers) by assuming that x is a generic real number and using algebra to prove (x + 1)2 = x2 + 2x + 1. If p(x) is proved to be true for a generic element in the universe of discourse. then ∀x p(x) is true.e.3.3. i. Universal Generalization.e.1. We state the rules for predicates with one variable. then p(a) is true for some speciﬁc element a in the universe of discourse. Existential Instantiation.. Universal Instantiation. So.: ∀x p(x) ∴ p(a) For instance. from ∃x (x2 = 2) (the universe of discourse is the real numbers) we derive the √ existence of some element. If p(a) is true for some speciﬁc element a in the universe of discourse. If ∀x p(x) is true. which we may represent ± 2.e. if a is an element in the universe of discourse.
1. PROOFS 18 then from ¬p(a) we can derive ¬∀x p(x). Answer : The argument is as follows: step statement reason 1) ¬p(a) 2) ∃x ¬p(x) 3) ¬∀x p(x) Premise Existential Generalization Negation of Universal Statement .3.
3. 2. −3. C = the set of complex numbers. 3. if A = {1. S ∗ = set of elements in S excluding zero. 4. 1. Q = the set of rational numbers. A set can be represented by listing its elements between braces: A = {1. called setbuilder notation. 2.1 Z = {· · · . 5. called elements of the set. A set is a collection of objects.1. S + = set of positive elements in S. R = the set of real numbers. −2. for instance 3 ∈ A. its number of elements is represented A.g. −1. 2. Its negation is represented by ∈. 2. Functions. for instance Z + = {1. 2. · · · }. 5}. for instance A = {x ∈ Z  1 ≤ x ≤ 5} = “set of that N includes zero—for some authors N = {1. 4. Some important sets are the following: 1. An alternative way to deﬁne a set. N = {0. 3.CHAPTER 2 Sets. e. · · · } = the set of positive integers. The symbol ∈ is used to express that an element is (or belongs to) a set. 2When 1Note 19 . 1. 4. If the set is ﬁnite. Sets. S − = set of negative elements in S. 3.1. 5} then A = 5. 2. for instance R∗ = the set of non zero real numbers. 3. 2.1. −3. 3. Is S is one of those sets then we also use the following notations:2 1. working with strings we will use a similar notation with a diﬀerent meaning—be careful not to confuse it. −2. 0. for instance Z− = {−1. Setbuilder notation. Relations 2. · · · } = the set of negative integers. e. 7 ∈ A. without zero. · · · } = the set of integers. is by stating a property (predicate) P (x) veriﬁed by exactly its elements. · · · } = the set of natural numbers. 3. 2.g. Set Theory 2. 3.
which are mathematical entities similar to sets. c} and B = {a.e. but with possibly repeated elements. In set theory the term universal set is often used in place of “universe of discourse” for a given predicate. {1. d. {1. and is represented by ∅ or {}. SET THEORY 20 integers x such that 1 ≤ x ≤ 5”—i. b} are the same set because they have exactly the same elements. i. e. In general: A = {x ∈ U  p(x)}. e} then A ⊆ B. 3}} and B = {3. a. represented “A ⊂ B”. 3.3 Principle of Extension.: A = {1. So. if A ⊆ B but A = B. 3Properly . since in the ﬁrst one the element a occurs twice and in the second one it occurs only once. 5}. Exercise: Prove by induction that if A = n then P(A) = 2n .: A = B ≡ ∀x (x ∈ A ↔ x ∈ B) .. 2. in some applications it might be useful to allow repeated elements in a set. 2. Empty Set. where U is the universe of discourse in which the predicate P (x) must be interpreted. b} and {a. 4.e. Power Set. as multisets. so for instance. B ∈ A.. 2}. 3}.2. Multisets. {a. and we represent it “A ⊆ B”. the universe of discourse of set theory is the collection of all sets (which is not a set). 2. {a. {2}. For instance if A = {1.. 3}. b. or A is contained in B. then P(A) = {∅. We say that A is a subset of set B. Two ordinary sets are identical if they have the same elements. a. if A = {1. b. there is some element in B which is not in A. Note that nothing prevents a set from possibly being an element of another set (which is not the same as being a subset!). {2. 3}. then obviously B is an element of A. i.g. speaking. or void set). Two sets are equal if and only if they have the same elements. Subset. a. {3}. or A = {x  P (x)} if the universe of discourse for P (x) is implicitly understood. A set with no elements is called empty set (or null set. {1}. {3.e. if A = {a. b} and {a.1. t}. In that case we use multisets. A is a proper subset of B. i. if all elements of A are in B. For instance. namely a and b. The collection of all subsets of a set A is called the power set of A. t}. A} . and is represented P(A). b} would be considered diﬀerent. c.e. However. {1.
A Figure 2.2. Venn Diagrams. A B Figure 2.1. SET THEORY 21 2.1.1. The other ﬁgures represent various set operations. A B Figure 2. . For instance.3. the rectangle represents the universal set (the set of all elements considered in a given problem) and the shaded region represents a set A. Venn Diagram. Intersection A ∩ B.1.2. Union A ∪ B.2. in ﬁgure 2. Venn diagrams are graphic representations of sets as enclosed areas in the plane.
SET THEORY 22 A Figure 2. A B Figure 2.4.5.1. Set Operations. Complement A. Intersection: The common elements of two sets: A ∩ B = {x  (x ∈ A) ∧ (x ∈ B)} . 2. .3. Union: The set of elements that belong to either of two sets: A ∪ B = {x  (x ∈ A) ∨ (x ∈ B)} . Diﬀerence A − B. A B Figure 2. Symmetric Diﬀerence A ⊕ B.6. 2.2. If A ∩ B = ∅.1. 1. the sets are said to be disjoint.
M ∩ P  = 100. then we can ﬁnd the number of elements in each of the regions and use that information for obtaining the number of elements in other portions of the plane. Complement: The set of elements (in the universal set) that do not belong to a given set: A = {x ∈ U  x ∈ A} . C = 450. It can be expressed also in the following way: A ⊕ B = A ∪ B − A ∩ B = (A − B) ∪ (B − A) . If we have information about the number of elements of some portions of the diagram.4. Physics courses and Computer Science courses respectively in a university.1. P and C be the sets of students taking Mathematics courses. Counting with Venn Diagrams. M ∩ C = 150. 5. 2. P ∩ C = 75. How many students are taking exactly one of those courses? (ﬁg. A Venn diagram with n sets intersecting in the most general way divides the plane into 2n regions. .1. Example: Let M . Assume M  = 300. P  = 350. their symmetric diﬀerence is the set of elements that belong to either one or the other set but not both. Counting with Venn diagrams. 4. A ⊕ B = {x  (x ∈ A) ⊕ (x ∈ B)} . 2. SET THEORY 23 3. (M ∩C)−(M ∩ P ∩ C) = 150 − 10 = 140 and (P ∩ C) − (M ∩ P ∩ C) = 75 − 10 = 65. Diﬀerence or Relative Complement: The set of elements that belong to a set but not to another: A − B = {x  (x ∈ A) ∧ (x ∈ B)} = A ∩ B .2. M ∩ P ∩ C = 10. We see that (M ∩P )−(M ∩P ∩C) = 100−10 = 90.7.7) M 60 90 10 140 65 185 P 235 C Figure 2. Symmetric Diﬀerence: Given two sets.
Identity Laws: A∪∅=A A∩U=A 5. Distributive Laws: A ∪ (B ∩ C) = (A ∪ B) ∩ (A ∪ C) A ∩ (B ∪ C) = (A ∩ B) ∪ (A ∩ C) 4. Bound Laws: A∪U=U A∩∅=∅ 8. Complement Laws: A∪A=U A∩A=∅ 6.1.5. The sum 60 + 185 + 235 = 480 is the number of students taking exactly one of those courses. Involution Law: A=A . Properties of Sets. Idempotent Laws: A∪A=A A∩A=A 7. Commutative Laws: A∪B =B∪A A∩B =B∩A 3. The set operations verify the following properties: 1. 2. Associative Laws: A ∪ (B ∪ C) = (A ∪ B) ∪ C A ∩ (B ∩ C) = (A ∩ B) ∩ C 2.2.1. SET THEORY 24 Then the region corresponding to students taking Mathematics courses only has cardinality 300−(90+10+140) = 60. Absorption Laws: A ∪ (A ∩ B) = A A ∩ (A ∪ B) = A 9. Analogously we compute the number of students taking Physics courses only (185) and taking Computer Science courses only (235).
2.1. SET THEORY
25
10. 0/1 Laws: ∅=U U=∅ 11. DeMorgan’s Laws: A∪B =A∩B A∩B =A∪B 2.1.6. Generalized Union and Intersection. Given a collection of sets A1 , A2 , . . . , AN , their union is deﬁned as the set of elements that belong to at least one of the sets (here n represents an integer in the range from 1 to N ):
N
An = A1 ∪ A2 ∪ · · · ∪ AN = {x  ∃n (x ∈ An )} .
n=1
Analogously, their intersection is the set of elements that belong to all the sets simultaneously:
N
An = A1 ∩ A2 ∩ · · · ∩ AN = {x  ∀n (x ∈ An )} .
n=1
These deﬁnitions can be applied to inﬁnite collections of sets as well. For instance assume that Sn = {kn  k = 2, 3, 4, . . . } = set of multiples of n greater than n. Then
∞
Sn = S2 ∪ S3 ∪ S4 ∪ · · · = {4, 6, 8, 9, 10, 12, 14, 15, . . . }
n=2
= set of composite positive integers . 2.1.7. Partitions. A partition of a set X is a collection S of non overlapping non empty subsets of X whose union is the whole X. For instance a partition of X = {1, 2, 3, 4, 5, 6, 7, 8, 9, 10} could be S = {{1, 2, 4, 8}, {3, 6}, {5, 7, 9, 10}} . Given a partition S of a set X, every element of X belongs to exactly one member of S. Example: The division of the integers Z into even and odd numbers is a partition: S = {E, O}, where E = {2n  n ∈ Z}, O = {2n + 1  n ∈ Z}.
2.1. SET THEORY
26
Example: The divisions of Z in negative integers, positive integers and zero is a partition: S = {Z+ , Z − , {0}}. 2.1.8. Ordered Pairs, Cartesian Product. An ordinary pair {a, b} is a set with two elements. In a set the order of the elements is irrelevant, so {a, b} = {b, a}. If the order of the elements is relevant, then we use a diﬀerent object called ordered pair, represented (a, b). Now (a, b) = (b, a) (unless a = b). In general (a, b) = (a , b ) iﬀ a = a and b = b . Given two sets A, B, their Cartesian product A × B is the set of all ordered pairs (a, b) such that a ∈ A and b ∈ B: A × B = {(a, b)  (a ∈ A) ∧ (b ∈ B)} . Analogously we can deﬁne triples or 3tuples (a, b, c), 4tuples (a, b, c, d), . . . , ntuples (a1 , a2 , . . . , an ), and the corresponding 3fold, 4fold,. . . , nfold Cartesian products: A1 × A2 × · · · × An = {(a1 , a2 , . . . , an )  (a1 ∈ A1 ) ∧ (a2 ∈ A2 ) ∧ · · · ∧ (an ∈ An )} . If all the sets in a Cartesian product are the same, then we can use an exponent: A2 = A × A, A3 = A × A × A, etc. In general: An = A × A ×
(n times)
···
× A.
An example of Cartesian product is the real plane R2 , where R is the set of real numbers (R is sometimes called real line).
2.2. FUNCTIONS
27
2.2. Functions 2.2.1. Correspondences. Suppose that to each element of a set A we assign some elements of another set B. For instance, A = N, B = Z, and to each element x ∈ N we assign all elements y ∈ Z such that y 2 = x (ﬁg. 2.8).
0 2 3 6 7 N 1 4 5 9 3 1 2
0 −1 −2 −3 Z
8 10
√ Figure 2.8. Correspondence x → ± x. This operation is called a correspondence.
2.2.2. Functions. A function or mapping f from a set A to a set B, denoted f : A → B, is a correspondence in which to each element x of A corresponds exactly one element y = f (x) of B (ﬁg. 2.9).
A
B
Figure 2.9. Function. Sometimes we represent the function with a diagram like this: f :A→B x→y or A→B x→y
f
2. all odd numbers. Graph: The graph of a function f : A → B is the subset of A × B deﬁned by G(f ) = {(x.11): ∀x. f (x))  x ∈ A} (ﬁg.. If A ⊆ A.e. the subset of B consisting of all images of elements of A . and B is its codomain. π = 3. etc. The ﬂoor function: x = greatest integer less than or equal to x .5 = −3. 2.. FUNCTIONS 28 For instance.3. The ceiling function: x = least integer greater than or equal to x . For instance: 2 = 2.5 = −2.10). Example: Two useful functions from R to Z are the following: 1. Types of Functions. 1.3 = 2. and x is a preimage of y. the image of A by f is f (A ) = {f (x)  x ∈ A }. i. For instance. For instance: 2 = 2. i. the following represents the function from Z to Z deﬁned by f (x) = 2x + 1: f :Z→Z x → 2x + 1 The element y = f (x) is called the image of x. Example: The modulus operator is the function mod : Z × Z+ → Z deﬁned: x mod y = remainder when x is divided by y. 2. x ∈ A. π = 4.3 = 3. The subset f (A) of B consisting of all images of elements of A is called the range of f . For instance 23 mod 7 = 2 because 23 = 3·7+2. f (x) = f (x ) ⇒ x = x . −2. 59 mod 9 = 5 because 59 = 6 · 9 + 5. the range of f (x) = 2x + 1 is the set of all integers of the form 2x + 1 for some integer x.2. 2.2.e.2. 2. OnetoOne or Injective: A function f : A → B is called onetoone or injective if each element of B is the image of at most one element of A (ﬁg. if f (x) = 2x + 1 then f (7) = 2 · 7 + 1 = 15. 2. For instance. −2. The set A is the domain of f . .
2. or a . A B Figure 2. 3. For instance. FUNCTIONS 29 4 3 y 2 1 –2 –1 0 1 x 2 Figure 2. Graph of f (x) = x2 . For instance.2.12): ∀y ∈ B. Onetoone function. A B Figure 2.11. 2. Onto or Surjective: A function f : A → B is called onto or surjective if every element of B is the image of some element of A (ﬁg. Onto function.12.10. OneToOne Correspondence or Bijective: A function f : A → B is said to be a onetoone correspondence. ∃x ∈ A such that y = f (x) . or bijective. f (x) = 2x from Z to Z is injective. f (x) = x2 from R to R+ ∪ {0} is onto.2.
. For instance. if A = B = C = Z. g(x) = x2 . Function Composition. Bijection. i.2. the function 1A : A → A deﬁned by 1A (x) = x for every x in A is called the identity function for A.4.e. the composite function of f and g is the function g ◦ f : A → C deﬁned by (g ◦ f )(x) = g(f (x)) for every x in A: g◦f f A x G G B y=f (x) g G G @ C z=g(y)=g(f (x)) For instance. Also (f ◦ g)(x) = g(x) + 1 = x2 + 1 (the composition of functions is not commutative in general).13. we have that h ◦ (g ◦ f ) = (h ◦ g) ◦ f .2. f (x) = x + 3 from Z to Z is a bijection. if it is onetoone and onto (ﬁg.2. given three functions A → B → C → D. f g h . 2. 2.5. Given a set A. Some properties of function composition are the following: 1. If f : A → B is a function from A to B. FUNCTIONS 30 bijection. Function composition is associative.2. 2. A B Figure 2. then (g ◦ f )(x) = f (x)2 = (x + 1)2 . Identity Function. we have that f ◦ 1A = 1B ◦ f = f .13). f (x) = x + 1. Given two functions f : A → B and g : B → C. 2.
2. y). For instance. The inﬁx notation consists of writing the symbol of the binary operator between its arguments: x + y (this is the most common). For instance. Operators. then it makes sense to compose it with itself: f 2 = f ◦ f . . Inverse Function. then f 2 (x) = 2(2x + 1) + 1 = 4x + 3. FUNCTIONS 31 Function iteration. and so on.6. The arrow diagram of f −1 is the same as the arrow diagram of f but with all arrows reversed. y) → x · y. A characteristic property of the inverse function is that f −1 ◦f = 1A and f ◦ f −1 = 1B . This is called preﬁx notation.2..2. if f : Z → Z is f (x) = 2x + 1.2. 2. if f : Z → Z is deﬁned by f (x) = x + 3. A function from A × A to A is called a binary operator on A. For instance the change of sign x → −x on Z is a unary operator on Z. its inverse is the function f −1 : B → A such that f −1 (y) = x if and only if f (x) = y.times) ◦f . Analogously we can deﬁne f 3 = f ◦f ◦f . There is also a postﬁx notation consisting of writing the symbol after the arguments: x y +. Another example of binary operator on Z is (x. If f : A → B is a bijective function.7. A monary or unary operator on A is a function from A to A. . 2. f n = f ◦ (n . If f : A → A is a function from A to A. then its inverse is f −1 (x) = x − 3. In the usual notation for functions the sum of two integers x and y would be represented +(x. For instance the addition of integers is a binary operator + : Z × Z → Z. An example of unary operator on R∗ (nonzero real numbers) is x → 1/x.
1. In the diagram an arrow from x to y means that x is related to y. 3.14 represents the relation from A = {a. For instance. As an example. d} to B = {1. (c. i. b) ∈ R. Venn diagrams and arrows can be used for representing relations between given sets.3. .e. (Z. and m is married to w. 1). w is a woman. Representations of Relations. we often write a R b instead of (a. 3). Relations 2. This kind of graph is called directed graph or digraph. 2. for instance. (b. Notation: A set A with a binary relation R is sometimes represented by the pair (A. 3)}. (c. (1. 4). 1). the domain is the set of married men. We want to express which men in M are married to which women in W .3. 2)..3.2. In general. 3. 4)} . 2. 2. The set {b ∈ B  a R b for some a ∈ A} is called the range of R. some of whom are married.3. assume A = {1. So. R). If A and B are the same set. Arrow diagrams.2. R ⊆ A × B. 2). (3. RELATIONS 32 2. One way to do that is by listing the set of pairs (m. (2. Then the binary relation “less than” in A will be: <A = {(x. a relation R from a set A to a set B will be understood as a subset of the Cartesian product A × B. and the range is the set of married women. then any subset of A × A will be a binary relation in A. c. 4). the relation “married to” can be represented by a subset of the Cartesian product M × W . in the relation “married to” above. w) such that m is a man. Relations. (1. ≤) means the set of integers together with the relation of nonstrict inequality. For instance. Assume that we have a set of men M and a set of women W . So. y) ∈ A × A  x < y} = {(1. (2. b. 3). If an element a ∈ A is related to an element b ∈ B. ﬁgure 2. 4}. The set {a ∈ A  a R b for some b ∈ B} is called the domain of R. 4} given by R = {(a.
3. 6.2. d} to B = {1. 2. Given a relation R from A to B. 2. and its columns are labeled with the elements of B. 1). Another example is given in diagram 2. 8. which represents the divisibility relation on the set {1. (b. is the relation from B to A deﬁned as b R−1 a ⇔ a R b . Binary relation of divisibility.14. 7. the inverse of R. 3)} from A = {a. Its rows are labeled with the elements of A. (c. 2). If a ∈ A and b ∈ B then we write 1 in row a column b if a R b. 9}. Relation. b. . 1). For instance the relation R = {(a. (c. Inverse Relation. c. 4} has the following matrix: a b c d 1 1 1 0 0 2 3 4 0 0 1 0 0 0 1 0 0 0 0 0 2. 3. 1 2 4 8 6 7 3 9 5 Figure 2. Matrix of a Relation. 3.3.15. denoted R−1 .15. RELATIONS 33 a c 1 b 2 d 3 4 A B Figure 2.3. 4. 5. otherwise we write 0. Another way of representing a relation R from A to B is with a matrix.
Partial Orders.3. 2. y ∈ A. 2. For instance on Z the relation “equal to” (=) is reﬂexive. RELATIONS 34 For instance. Composition of Relations.2. nonstrict inequality (≤) on Z is antisymmetric. Transitive: (x y) ∧ (y z) ⇒ x z. Transitive if for all x. For instance equality (=) and inequality (<) on Z are transitive relations. Reﬂexive: for all x ∈ A. but strict inequality (<) is not.5. then R−1 is the relation “being a parent of”. if R is the relation “to be the father of”.4. 4. B and C be three sets. x R y ⇒ y R x. x R y and y R z implies x R z. 3. z ∈ A.3. y. Antisymmetric if for all x. 2. an order on a set A is a binary relation “ ” on A with the following properties: 1.6. For instance. Reﬂexive if for all x ∈ A. For instance on Z. x x.3. and S is the relation “to be married to”. x R y and y R x implies x = y. 2. 2. 3. . or simply.3. Examples: 1. then the composition S ◦ R of relations R and S is a relation from A to C deﬁned by: a (S ◦ R) c ⇔ there exists some b ∈ B such that a R b and b S c . x R x. A binary relation R on A is called: 1. 2. Properties of Binary Relations. then S ◦ R is the relation “to be the father in law of”. Given a relation R from A to B and a relation S from B to C. Let A. Antisymmetric: (x y) ∧ (y x) ⇒ x = y. A partial order. if R is the relation “being a son or daughter of”. The nonstrict inequality (≤) in Z. b = at. equality (=) is symmetric. y ∈ A. For instance. Symmetric if for all x. Relation of divisibility on Z+ : ab ⇔ ∃t.
Hence a = att . ).3. 2. Its immediate successors are placed above the node and connected to it by straight line segments. Hasse diagrams. 8. hence c = att . otherwise they are said to be non comparable. ﬁgure 2.3. The only invertible positive integer is 1.8. 2. ). 7. Transitive: (x ∼ y) ∧ (y ∼ z) ⇒ x ∼ z. A totally ordered subset of a partially ordered set is called a chain. As an example. Reﬂexive: a = a 1 ⇒ aa. x ∼ x. Reﬂexive: for all x ∈ A. Set inclusion (⊆) on P(A) (the collection of subsets of a given set A). ac. ≤) is totally ordered. 9}. so t = t = 1 ⇒ a = b. Transitive: ab and bc implies b = at for some t and c = bt for some t .16 represents the Hasse diagram for the relation of divisibility on {1. 8.7.e. 5. for instance the set {1. 3. . 2. 3. 6. Equivalence Relations. Antisymmetric: ab ⇒ b = at for some t and ba ⇒ a = bt for some t . y ∈ A are comparable. Question: is the strict inequality (<) a partial order on Z? Two elements a. 2. 2. 16.. . is not. The order is called total or linear when every pair of elements x. An equivalence relation on a set A is a binary relation “∼” on A with the following properties: 1. which implies tt = 1 ⇒ t = t−1 . As an example we prove that integer divisibility is a partial order: 1.2. Exercise: prove that the aforementioned relations are in fact partial orders. . 4. For instance (Z. . where “” represents integer divisibility. RELATIONS 35 3. 3. A Hasse diagram is a graphical representation of a partially ordered set in which each element is represented by a dot (node or vertex of the diagram).3. i. but (Z+ . Question: How does the Hasse diagram look for a totally ordered set? 2. } is a chain in (Z+ . 4. Symmetric: x ∼ y ⇒ y ∼ x. b ∈ A are said to be comparable if either x y or y x.
Reﬂexive: for every integer x. but y − x = −t is also even. so x ≡ x (mod 2). also on Z. hence y ≡ x (mod 2). 2. i. x−x = 0 is indeed even. the equality (=) is an equivalence relation. RELATIONS 36 8 9 6 5 2 3 7 4 1 Figure 2. For instance. is the following: x ≡ y (mod 2) (“x is congruent to y modulo 2”) iﬀ x−y is even. on Z. 6 ≡ 2 (mod 2) because 6 − 2 = 4 is even.3. From here. Recall that a partition of a set A is a collection of . and an element x ∈ A. Quotient Set. 2. Equivalence Classes. the collection of equivalence classes. The collection of equivalence classes. represented A/ ∼ = {[x]  x ∈ A}. Congruence modulo 2 is in fact an equivalence relation: 1. is a partition of A. Hasse diagram for divisibility.2. is called quotient set of A by ∼. because 7 − 4 = 3 is not even.16. 3. x − z = (x − y) + (y − z) = t + u is also even. hence x ≡ z (mod 2). the set of elements of A related to x are called the equivalence class of x. Element x is said to be a representative of class [x]. represented [x] = {y ∈ A  y ∼ x}. One of the main properties of an equivalence relation on a set A is that the quotient set. Given an equivalence relation ∼ on a set A. Another example.3. Exercise: Find the equivalence classes on Z with the relation of congruence modulo 2. Symmetric: if x ≡ y (mod 2) then x − y = t is even. Transitive: assume x ≡ y (mod 2) and y ≡ z (mod 2). Then x − y = t and y − z = u are even.9. but 7 ≡ 4 (mod 2). Partitions.e. For instance.
What are the equivalence classes? How many are there? Exercise: On the Cartesian product Z × Z∗ we deﬁne the relation (a. Prove that R is an equivalence relation. . b) R (c. . Prove that it is an equivalence relation. of A which are pairwise disjoint and whose union equals A: 1. O}. Example: in Z with the relation of congruence modulo 2 (call it “∼2 ”). . Ai ∩ Aj = ∅ for i = j. and Z = E ∪ O. there are two equivalence classes: the set E of even integers and the set O of odd integers. n An = A. RELATIONS 37 nonempty subsets A1 . We see that it is in fact a partition of Z. On Z we deﬁne the relation x ≡ y (mod m) ⇔ m(y − x) (i. Exercise: Let m be an integer greater than or equal to 2. A3 . A2 . m divides exactly y − x)..2. 2. d) ⇔ ad = bc.3. Would it still be an equivalence relation if we extend it to Z × Z? .e. The quotient set of Z by the relation “∼2 ” of congruence modulo 2 is Z/ ∼2 = {E. because E ∩ O = ∅.
c: 1. Finiteness. Conversely. Generality. programs are implementations of algorithms. 3. The main diﬀerence respect to actual programming languages is that pseudocode is not required to follow strict syntactic rules.1. 4. not actually executed by a machine. Output the value of x. Integers 3. Precision. b.1. 5. 6. In general an algorithm is a ﬁnite list of instructions with the following characteristics: 1. c. The algorithm applies to a set of inputs. since it is intended to be just read by humans. Input. Basically an algorithm is the idea behind a program. The algorithm produces output. Uniqueness. 2. The steps are precisely stated. 38 . 3. Pseudocode. Algorithms Consider the following list of instructions to ﬁnd the maximum of three numbers a. 2. If b > x then assign x the value of b. 3. The algorithm stops after ﬁnitely many instructions have been executed.1. Assign variable x the value of a. Output. If c > x then assign x the value of c. 4. b. After executing those steps the output will be the maximum of a. Pseudocode is a language similar to a programming language used to represent algorithms. The algorithm receives input. The result of executing each step is uniquely determined by the inputs and the result of preceding steps.CHAPTER 3 Algorithms.
b.. ALGORITHMS 39 Usually pseudocode will look like this: procedure ProcedureName(Input) Instructions.3. end ProcedureName For instance the following is an algorithm to ﬁnd the maximum of three numbers a. c: 1: procedure max(a.b.c) 2: x := a 3: if b>x then 4: x := b 5: if c>x then 6: x := c 7: return x 8: end max Next we show a few common operations in pseudocode. otherwise it executes “action2”: if p then action1 else action2 The following code executes “action” while condition “p” is true: 1: while p 2: action The following is the structure of a for loop: .1. The following statement means “assign variable x the value of variable y: x := y The following code executes “action” if condition “p” is true: if p then action The following code executes “action1” if condition “p” is true..
s2 .n) 2: largest := s[1] 3: for k := 2 to n 4: if s[k] > largest then 5: largest := s[k] 6: return largest 7: end largest element 3..1. . . s[n]: 1: procedure largest element(s.3. A deﬁnition such that the object deﬁned occurs in the deﬁnition is called a recursive deﬁnition. s[2]. Recursive Deﬁnitions. ALGORITHMS 40 for var := init to limit action If an action contains more than one statement then we must enclose them in a block: begin Instruction1 Instruction2 Instruction3 . . ..1..) As an example. sn represented as an array with n elements: s[1]. the following procedure returns the largest number in a sequence s1 . end Comments are enclose between brackets: {This is a comment} The output of a procedure is returned with a return statement: return output Procedures that do not return anything are invoked with a call statement: call Procedure(arguments... .2. . . For instance. Recursiveness.
g. 1.1.. an = an−1 a (n ≥ 1) (n ≥ 1) In all these examples we have: 1. the second one recursive. Recursive algorithms have the advantage that often they are easy to design and are closer to natural mathematical deﬁnitions. . procedure p1 invokes procedure p2 . A recursive algorithm is an algorithm that contains recursive procedures or recursive sets of procedures. A recursive procedure is a procedure that invokes itself. 13. It can be deﬁned as a sequence whose two ﬁrst terms are F0 = 0. procedure p2 invokes procedure p3 and procedure p3 invokes procedure p1 . 1: procedure factorial iterative(n) 2: fact := 1 3: for k := 2 to n 4: fact := k * fact 5: return fact 6: end factorial iterative . stating how to compute the function from its previous values. Other examples: • Factorial: 1. 2. ALGORITHMS 41 consider the Fibonacci sequence 0. a0 = 1 2. 3. As an example we show two alternative algorithms for computing the factorial of a natural number. . e. F1 = 1 and each subsequent term is the sum of the two previous ones: Fn = Fn−1 + Fn−2 (for n ≥ 2). .3. the ﬁrst one iterative (non recursive). 2. A Basis. 1. Recursive Procedures. A Recursive Step. n! = n · (n − 1)! • Power: 1. where the function is explicitly evaluated for one or more values of its argument. 8. Also a set of procedures is called recursive if they invoke themselves in a circle. 0! = 1 2. 5.
an integer greater than or equal to 0. n directly. If n = 0 then factorial recursive(0) returns 1 immediately without performing any recursive call. .3.. If n > then the execution of factorial recursive(n) leads to a recursive call factorial recursive(n1) which will perform a recursive call factorial recursive(n2) and so on until eventually reaching the basic case factorial recursive(0) After reaching the basic case the procedure returns a value to the last call. Another example is the following algorithm for computing the nth element of the Fibonacci sequence: . ALGORITHMS 42 1: procedure factorial recursive(n) 2: if n = 0 then 3: return 1 4: else 5: return n * factorial recursive(n1) 6: end factorial recursive While the iterative version computes n! = 1 · 2 · . which returns a value to the previous call. i. and any legitimate input should lead to a ﬁnite sequence of recursive calls ending up at the basic case. In our example n is a legitimate input if it is a natural number. .e. and so on up to the ﬁrst invocation of the procedure. the recursive version resembles more closely the formula n! = n · (n − 1)! A recursive algorithm must contain at least a basic case without recursive call (the case n = 0 in our example).1.
It applies to a list of numbers si . First let’s look at a simple non recursive sorting algorithm called bubble sort. namely n = 0 and n = 1. and algorithm intended to sort a list of elements. the idea of merge sort is to split the list into two approximately equal parts.j) 2: for p:=1 to ji 3: for q:=i to jp 4: if s[q] > s[q+1] then 5: swap(s[q]. so at this point we see that fibonacci(3) is being called twice.. An example is merge sort. ALGORITHMS 43 1: procedure fibonacci(n) 2: if n=0 then 3: return 0 4: if n=1 then 5: return 1 6: return fibonacci(n1) + fibonacci(n2) 7: end fibonacci In this example we have two basic cases. On the other hand. In this particular case the algorithm is ineﬃcient in the sense that it performs more computations than actually needed. For instance a call to fibonacci(5) contains two recursive calls. once inside fibonacci(5) and again in fibonacci(4). s[j]: 1: procedure bubble sort(s. sj represented as an array s[i]. one to fibonacci(4) and another one to fibonacci(3).. sort them separately and then merge them into a single list: .i. However careful design may yield eﬃcient recursive algorithms. . The idea is to go several times through the list swapping adjacent elements if necessary.3..1.s[q+1]) 6: end bubble sort We can see that bubble sort requires n(n − 1)/2 comparisons and possible swapping operations. . . Hence sometimes the price to pay for a simpler algorithmic structure is a loss of eﬃciency. . s[i+1].. Then fibonacci(4) performs a call to fibonacci(3) and another call to fibonacci(2). si+1 .
sj into a single increasing sequence ci .c) merges the two increasing sequences si . .3. 1: for i := 1 to n 2: x := x + 1 The following double loop performs it n2 times: 1: for i := 1 to n 2: for j := 1 to n 3: x := x + 1 The following one performs it 1 + 2 + 3 + · · · + n = n(n + 1)/2 times: . it is in general impossible to make an estimation in actual physical time. This algorithm is more eﬃcient than bubble sort because it requires only about n log2 n operations (we will make this more precise soon). Here we deal with time complexity only. The strategy of dividing a task into several smaller tasks is called divide and conquer.i.m) 6: call merge sort(s. . Since the actual time required to execute an algorithm depends on the details of the program implementing the algorithm and the speed and other characteristics of the machine executing it.1.i.j. .1. .m+1. . In general the complexity of an algorithm is the amount of time and space (memory use) required to execute it. sm and sm+1 .j. . . . cj . . . ALGORITHMS 44 1: procedure merge sort(s.3. For instance the following loop performs the statement x := x + 1 exactly n times.j) 2: if i=j then 3: return 4: m := floor((i+j)/2) 5: call merge sort(s.m.i. sm+2 .m. ci+1 . say by the number of operations performed.c) 8: for k:=i to j 9: s[k] := c[k] 10: end merge sort The procedure merge(s. however it is possible to measure the length of the computation in other ways. Complexity. 3.j) 7: call merge(s. . . si+1 .i.
Often the exact time is too hard to compute or we are interested just in how it grows compared to the size of the input. . or more generally as a function of the size of the input. . assume that we have a list of n objects one of which is colored red and the others are colored blue. . The maximum time would be n (if the red object turns out to be the last one). The average time is the average of all possible times: 1. For instance and algorithm that requires exactly 7n2 + 3n + 10 steps to be executed on an input of size n is said to be or order n2 .1. 2. ALGORITHMS 45 1: for i := 1 to n 2: for j := 1 to i 3: x := x + 1 Since the time that takes to execute an algorithm usually depends on the input. represented Θ(n2 ). its complexity must be expressed as a function of the input. Averagecase time: average time needed to execute the algorithm among all inputs of a given size n. Since the execution time may be diﬀerent for inputs of the same size. . the worstcase time is n and the averagecase time is (n + 1)/2. . Bestcase time: minimum time needed to execute the algorithm among all inputs of a given size n. 2. which is (1+2+3+· · ·+n)/n = (n+1)/2. written f (n) = O(g(n)). 3. A function f (n) is said to be of order at most g(n). So in this example the bestcase time is 1. In this problem the minimum time needed to ﬁnd the red object would be 1 (in the lucky event that the ﬁrst object examined turned out to be the red one). and we want to ﬁnd the one that is colored red by examining the objects one by one. This justiﬁes the following notations: Big Oh Notation. For instance.3. if there is a constant C1 such that f (n) ≤ C1 g(n) for all but ﬁnitely many positive integers n. Wostcase time: maximum time needed to execute the algorithm among all inputs of a given size n. We measure time by the number of objects examined. we deﬁne the following kinds of times: 1. n. 3.
written f (n) = Ω(g(n)). A function f (n) is said to be of order g(n). A function f (n) is said to be of order at least g(n).3. First. Next. if we call T (n) (the order of) the number of operations . if f (n) = O(g(n)) and f (n) = Ω(g(n)). the inner loop is executed (n − 1) + (n − 2) + · · · + 1 = n(n − 1)/2 times. Remark : All logarithmic functions are of the same order: loga n = Θ(logb n) for any a. Theta Notation. The estimation of the complexity of merge sort is more involved. we do not need to specify the base of the logarithm. we can just say that its time is “logarithmic”. As a consequence. because loga n = logb n/ logb a. so they always diﬀer in a multiplicative constant. Since bubble sort is just a double loop its complexity is easy to ﬁnd. so it requires n(n − 1)/2 comparisons and possible swap operations. written f (n) = Θ(g(n)). The following are several common growth functions: Order Θ(1) Θ(log log n) Θ(log n) Θ(n log n) Θ(n) Θ(n2 ) Θ(n3 ) Θ(nk ) Θ(an ) Name Constant Log log Logarithmic n log n Linear Quadratic Cubic Polynomial Exponential Let’s see now how we ﬁnd the complexity of algorithms like bubble sort and merge sort. b > 1. ALGORITHMS 46 Omega Notation. the number of operations required by the merge procedure is Θ(n). Hence its execution time is Θ(n2 ).1. if the execution time of an algorithm is of order a logarithmic function. if there is a constant C2 such that f (n) ≥ C2 g(n) for all but ﬁnitely many positive integers n.
3. ALGORITHMS 47 required by merge sort working on a list of size n.1. we see that roughly: T (n) = 2T (n/2) + n . hence T (n) = 2T (n/2) + n = 2(2T (n/4) + n/2) + n = 4T (n/4) + 2n . So for k = log2 n we have T (n) = nT (1) + n log2 n = Θ(n log n) . Replacing n with n/2 we have T (n/2) = 2T (n/4) + n/2. . Repeating k times we get: T (n) = 2k T (n/2k ) + kn .
2. it is just a mathematical theorem. an then pak for some k = 1. written ba. and r is called remainder. . so n ≥ 2 is composite iﬀ it has nontrivial divisors. 3. (Euclid) There are inﬁnitely many prime numbers. . The Fundamental Theorem of Arithmetic. 3. b = 0. p sk . THE EUCLIDEAN ALGORITHM 48 3. .2.2. There are. If pab then pa or pb. . It is customary to write the factorization in the following way: n = p s1 p s2 . More generally. The following result is known as The Division Algorithm:1 If a. b > 0. Some results about prime numbers: 1. result is not really an “algorithm”. Any integer n ≥ 2 that is not prime is called composite.3.4. 3. 3. Every integer n ≥ 2 can be written as a product of primes uniquely.1. For all n ≥ 2 there is some prime p such that pn. Given two integers a.2. . A nontrivial divisor of n ≥ 2 is a divisor d of n such that 1 < d < n. . Warning: 1 is not considered either prime or composite. algorithms that allow us to compute the quotient and the remainder in an integer division. then there exist unique q. The Division Algorithm.2. . if pa1 a2 . A prime number is an integer p ≥ 2 whose only positive divisors are 1 and p. n. or that a is a multiple of b. Prime Numbers. if there is some integer q such that a = bq: ba ⇔ ∃q. Divisibility. 0 ≤ r < b. 1The . a = bq .2.2. however. Here q is called quotient of the integer division of a by b.2. For instance: 13104 = 24 · 32 · 7 · 13 . We also say that b divides or is a divisor of a.3. we say that b divides a. 2. . b. up to the order of the primes. b ∈ Z. The Euclidean Algorithm 3. r ∈ Z such that a = qb + r. 1 2 k where all the exponents are positive and the primes are written so that p1 < p2 < · · · < pk .
b) = gcd(b. The greatest possible such d is called the greatest common divisor of a and b. Finding the gcd by Prime Factorization. Then: 1. hence gcd(1440. raised to their lowest exponent. In general a Diophantine equation is an equation whose solutions must be integers. if d divides a and b. 1512) = 23 · 32 = 72.7. Factoring numbers is not always a simple task. the equation ax + by = c has integer solutions if and only if gcd(a. b) = product of the primes that occur in the prime factorizations of both a and b. b) divides c. and r = a mod b. 3. Another important result is the following: Given integers a. then gcd(a. That is an example of a Diophantine equation. c. The greatest common divisor of 12 and 30 is gcd(12. We have that gcd(a.2. b are called relatively prime. 2.2. b. r). b). The Euclidean Algorithm.5. If dm then dmn. Example: The set of positive divisors of 12 and 30 is {1. 3. Example: We have gcd(12. d be integers. The method provides at the same time a solution to the Diophantine equation: ax + by = gcd(a. 6}. The solution is not unique. b. 2. b) = 1 then a. 30) = 6. If dm and dn then d(m − n).6. Let m. Greatest Common Divisor. for instance 6 = 3 · 30 − 7 · 12. b) . 3. denoted gcd(a. A positive integer d is called a common divisor of the integers a and b. If dm and dn then d(m + n). 3. Now we examine an alternative method to compute the gcd of two given positive integers a. 1512 = 23 · 33 · 7.2. n. If gcd(a. but there are other ways. THE EUCLIDEAN ALGORITHM 49 3.2. and in fact we can write 6 = 1 · 30 − 2 · 12. so ﬁnding the gcd by prime factorization might not be a most convenient way to do it. Proof: Divide a by . 30) = 6.3. A few properties of divisors are the following. It is based on the following fact: given two integers a ≥ 0 and b > 0. For instance 1440 = 25 · 32 · 5.
If d is a common divisor of a and b then it must be a divisor of r = a−bq. Then we arrange the computations in the following way: 500 222 56 54 = = = = 2 · 222 + 56 3 · 56 + 54 1 · 54 + 2 27 · 2 + 0 → → → → r r r r = 56 = 54 =2 =0 The last nonzero remainder is r = 2.2. and the greatest common divisor will be the same.b) 2: if a<b then {make a the largest} 3: swap(a. The algorithm to compute the gcd can be written as follows: 1: procedure gcd(a. Conversely. Then we divide b by r. obtaining a quotient q and a remainder r. Example: Assume that we wish to compute gcd(500. which gives a quotient q and another remainder r . 0 ≤ r < b. First we divide a by b. we can do it by working backward: 2 = 56 − 1 · 54 = 56 − 1 · (222 − 3 · 56) = 4 · 56 − 1 · 222 = 4 · (500 − 2 · 222) − 1 · 222 = 4 · 500 − 9 · 222 . 222). obtaining a new quotient q and a remainder r . Next we divide r by r . then a = bq + r . and which point we stop. Furthermore.b) 4: while b = 0 5: begin 6: r := a mod b 7: a := b 8: b := r 9: end 10: return a 11: end gcd .3. 222) = 2. hence gcd(500. We continue dividing each remainder by the next one until obtaining a zero remainder. THE EUCLIDEAN ALGORITHM 50 b obtaining a quotient q and a remainder r. if d is a common divisor of b and r then it must divide a = bq + r. So the set of common divisors of a and b and the set of common divisors of b and r are equal. The last nonzero remainder is the gcd. The Euclidean algorithm is a follows. if we want to express 2 as a linear combination of 500 and 222.
3. THE EUCLIDEAN ALGORITHM 51 The next one is a recursive version of the Euclidean algorithm: 1: procedure gcd recurs(a.b) 2: if b=0 then 3: return a 4: else 5: return gcd recurs(b.a mod b) 6: end gcd recurs .2.
. . a and b have the same remainder when divided by m. . . 0. −10 −5 0 5 10 15 20 . . . 4}). . 2. −9 −4 1 6 11 16 21 . . For instance 8 = 3 (mod p). each one of the following rows is an equivalence class: . .. . . 2. where 0 ≤ r < m. if we choose the representatives for the elements of Z5 in the interval from 0 to 4 (Z5 = {0. . Hence an equivalence class may be denoted [r] or x + m Z. so that the equivalence class [r] will be represented just r. Another possibility is to choose the representatives in the interval from −2 to 2 (Z5 = {−2. . Each equivalence class has exactly a representative r such that 0 ≤ r < m. 2}). 1. −7 −2 3 8 13 18 23 . 3. RSA Algorithm 3. The set of equivalence classes (i. . we will call “r reduced modulo m”. . for m = 5. . the chosen representative for the class of r. 1. a = b + km for some integer k.e. 3. Modular Arithmetic.3. written “r mod m”. −6 −1 4 9 14 19 24 . . then 9 mod 5 = 4. .3. m − 1}. Given an integer m ≥ 2. . 2. −1. 1. . Often we will omit the brackets. 2. . RSA ALGORITHM 52 3. while in “a = b (mod m)” they represent elements of Zm . 4}. a and b represent integers. .3. so that 9 mod 5 = −1 . Note that the following conditions are equivalent 1. Remark : When writing “r” as a notation for the class of r we may stress the fact that r represents the class of r rather than the integer r by including “ (mod p)” at some point. the quotient set of Z by the relation of congruence modulo m) is denoted Zm = {0. . For instance. namely the common remainder of all elements in that class when divided by m. a ≡ b (mod m). Congruences Modulo m. . For instance. . The relation of congruence modulo m is an equivalence relation. . It partitions Z into m equivalence classes of the form [x] = [x]m = {x + km  k ∈ Z} . Note that in “a ≡ b (mod m)”.1. 3. MODULAR ARITHMETIC. −8 −3 2 7 12 17 22 . For instance. . 1. Reduction Modulo m: Once a set of representatives has been chosen for the elements of Zm . we say that a is congruent to b modulo m. . Z5 = {0. if m(a − b).3. written a ≡ b (mod m).
3.2.3. i.3. e. Of course with this deﬁnition zero itself is a divisor of zero. However in Z6 that is not true.2 show the addition and multiplication tables for Z5 and Z6 respectively.3. Operational tables for Z6 A diﬀerence between this two tables is that in Z5 every nonzero element has a multiplicative inverse. 2−1 = 4 (mod 5). For instance in Z6 2 is a proper divisor of zero.1. 2 · 3 = 0 (mod 6).1 and 3. some nonzero elements like 2 have no multiplicative inverse.g.e. These elements are called divisors of zero. e. Operational tables for Z5 + 0 1 2 3 4 5 0 0 1 2 3 4 5 1 1 2 3 4 5 0 2 2 3 4 5 0 1 3 3 4 5 0 1 2 4 4 5 0 1 2 3 5 5 0 1 2 3 4 · 0 1 2 3 4 5 0 0 0 0 0 0 0 1 0 1 2 3 4 5 2 0 2 4 0 2 4 3 0 3 0 3 0 3 4 0 4 2 0 4 2 5 0 5 4 3 2 1 Table 3. In Z5 there are no proper divisors of zero. In general: 1. for every x ∈ Z5 such that x = 0 there is an x−1 such that x · x−1 = x−1 · x = 1.3. Furthermore the elements without multiplicative inverse verify that they can be multiply by some other nonzero element giving a product equal zero. The elements of Zm can be classiﬁed into two classes: . + 0 1 2 3 4 0 0 1 2 3 4 1 1 2 3 4 0 2 2 3 4 0 1 3 3 4 0 1 2 4 4 0 1 2 3 · 0 1 2 3 4 0 0 0 0 0 0 1 0 1 2 3 4 2 0 2 4 1 3 3 0 3 1 4 2 4 0 4 3 2 1 Table 3. Divisors of zero diﬀerent from zero are called proper divisors of zero.g.. As an example. MODULAR ARITHMETIC.3. [x] · [y] = [x · y] . RSA ALGORITHM 53 In Zm it is possible to deﬁne an addition and a multiplication in the following way: [x] + [y] = [x + y] . tables 3.
2. 3. Example: Find the multiplicative inverse of 17 in Z∗ . (b) Divisors of zero: elements that multiplied by some other nonzero element give product zero. 7} ∗ Z9 = {1. 7. m). ·) is 49. We will denote this by writing 17−1 = 49 64 (mod 64).3. its inverse can be computed by using m the Euclidean algorithm to ﬁnd gcd(a. so the inverse of 17 in (Z∗ . m) = 1. 2. 4. 8} Given an element [a] in Z∗ . For instance: m Z∗ = {1} 2 Z∗ = {1. 5. which is equivalent to ax ≡ 1 (mod m). 3} 4 Z∗ = {1. 3. . but −15 ≡ 49 (mod 64). Hence (−15) · 17 ≡ 1 (mod 64). 2. 4. 3. All nonzero elements of Zm are units if and only if m is a prime number. The set of units in Zm is denoted Z∗ . 2. since that algorithm also provides a solution to the equation ax + my = gcd(a. 6} 7 ∗ Z8 = {1. An element [a] ∈ Zm is a unit (has a multiplicative inverse) if and only if gcd(a. 5. 5.3. Answer : We 64 use the Euclidean algorithm: 64 17 13 4 = = = = 3 · 17 + 13 1 · 13 + 4 3·4+1 4·1+0 → → → → r r r r = 13 =4 =1 =0 Now we compute backward: 1 = 13 − 3 · 4 = 13 − 3 · (17 − 1 · 13) = 4 · 13 − 3 · 17 = 4 · (64 − 3 · 17) − 3 · 17 = 4 · 64 − 15 · 17 . 3. or 17−1 mod 64 = 49. 5} 6 Z∗ = {1. MODULAR ARITHMETIC. 4} 5 Z∗ = {1. m) = 1. RSA ALGORITHM 54 (a) Units: elements with multiplicative inverse. 2} 3 Z∗ = {1.
psk . If gcd(a. . the positive integers not greater than and relatively prime to 15 are: 1. 3. m2 are two relatively prime positive integers. A consequence of Euler’s Theorem is the following. For instance φ(15) = φ(3 · 5) = φ(3) · φ(5) = (3 − 1) · (5 − 1) = 2 · 4 = 8 . then we have. b) = 1 ⇒ f (ab) = f (a)f (b) is called multiplicative. in fact. 4. 2. function f (x) of positive integers such that gcd(a.3. For instance. then φ(m1 m2 ) = φ(m1 ) φ(m2 ).e.1 3. 1A . If m1 . We have the following results: 1. 8.3. MODULAR ARITHMETIC. hence φ(15) = 8. then φ(ps ) = ps − ps−1 = ps (1 − 1/p). 14. 13. m ≥ 2. Euler’s Theorem. i. That number is given by the so called Euler’s phi function: φ(m) = number of positive integers not greater than m and relatively prime to m . RSA ALGORITHM 55 3. 11. the number of integers a such that 1 ≤ a ≤ m and gcd(a. If a and m are two relatively prime positive integers. Euler’s theorem is called Fermat’s Little Theorem: ap−1 ≡ 1 (mod p) . . 2. The particular case in which m is a prime number p. . (1 − 1/pk ) . For instance. . The number of units in Zm is equal to the number of positive integers not greater than and relatively prime to m.3. then φ(m) = m (1 − 1/p1 ) (1 − 1/p2 ) .2. 27−1 = 26 = 64 = 1 + 9 · 7 ≡ 1 (mod 7). if a = 2 and p = 7. Euler’s Phi Function. If p is a prime number and s ≥ 1. m) = 1.3. In particular φ(p) = p − 1.. where the pk are prime and the sk are 1 2 k positive.3. then aφ(m) ≡ 1 (mod m) . 7. If m = ps1 ps2 . m) = 1 then x ≡ y (mod φ(m)) ⇒ ax ≡ ay (mod m) .
If gcd(e. MODULAR ARITHMETIC. m) = 1 then the value obtained .1. Application to Cryptography: RSA Algorithm. 2. Then we have that (me )d = me·d = m e (mod m) . The RSA algorithm is based on the following facts. For instance: 39734888 mod 100 = 39734888 mod φ(100) mod 100 = 39734888 mod 40 mod 100 = 38 mod 100 = 6561 mod 100 = 61 . So. The decryption key is private and is known only by the recipient of the encrypted message. The RSA algorithm is an encryption scheme designed in 1977 by Ronald Rivest.4. given m we can recover m modulo pq by raising to the dth power. Given two prime numbers p and q. It allows encrypting a message with a key (the encryption key) and decrypting it with a diﬀerent key (the decryption key).3. Assume now that we have two integers e and d such that e · d = 1 (mod φ(pq)). solving at the same time the equation ex + my = gcd(e. Find their product n = pq. [x]φ(m) ) → [ax ]m Hence. Adi Shamir and Leonard Adleman. The RSA algorithm consists of the following: 1.3. paragraph A.3. Euler’s theorem tells us that: mφ(pq) = m(p−1)(q−1) = 1 (mod pq) . m). First e is chosen at random. An even more eﬃcient way to compute powers modulo m is given in Appendix A. m) = 1. m). Generate two large primes p and q. RSA ALGORITHM 56 Consequently. 3. the following function is well deﬁned: Z∗ × Zφ(m) → Z∗ m m ([a]m . This requires some trial and error. and a positive number m relatively prime to p and q. The encryption key is public and can be given to everybody. Find two numbers e and d (in the range from 2 to φ(n)) such that e · d = 1 (mod φ(n)). and the Euclidean algorithm is used to ﬁnd gcd(e. (mod pq) . we can compute powers modulo m in the following way: an = an mod φ(m) if gcd(a.
The private decryption key will be the pair (n. It remains to prove that m = m.3. How many objects are there? Answer : We must solve the following system of congruences: x ≡ 2 (mod 3) x ≡ 4 (mod 5) x ≡ 6 (mod 7) . The number me is reduced modulo n. and since both are in the range from 0 to n − 1 they must be equal. . mk be pairwise relatively prime integers greater than or equal to 2. .3. mk . 4. with 0 ≤ m < n. e). x ≡ r (mod m ) k k has a unique solution modulo M = m1 m2 . 5. RSA ALGORITHM 57 for x is d. 3. and si = the inverse of Mi in Zmi . then from Euler’s theorem we get that m = m (mod n). The Chinese Remainder Theorem. The recipient computes m = m d mod n. in 5 rows leaving 4 left. MODULAR ARITHMETIC.. . The case in which p or q divides m is left as an exercise. m = me mod n. 6. The encryption key is given to everybody. The public encryption key will be the pair (n. . Example: A group of objects can be arranged in 3 rows leaving 2 left. 3. . Let Mi = M/mi . If m is relatively prime to p and q. d). The following system of congruences x ≡ r1 (mod m1 ) x ≡ r2 (mod m2 ) . We can ﬁnd a solution to that system in the following way. m2 . Then x = M 1 s1 r1 + M 2 s2 r2 + · · · + M k sk r k is a solution to the system. Otherwise.5. Let m1 . and in 7 rows leaving 6 left.. while the decryption key is kept secret by the future recipient of the message. and each piece is encoded numerically as a positive integer m smaller than n.3. . e is no relatively prime to φ(n) and we must try a diﬀerent value for e. The message to be encrypted is divided into small pieces.
M2 = 105/5 = 21 ≡ 1 (mod 5). . s3 = “inverse of 1 in Z7 ” = 1.3.3. any group of 104 + 105 k objects is a possible solution to the problem. RSA ALGORITHM 58 We have: M = 3 · 5 · 7 = 105. s2 = “inverse of 1 in Z5 ” = 1. Hence the solution is x = 35 · 2 · 2 + 21 · 1 · 4 + 15 · 1 · 6 = 314 ≡ 104 (mod 105) . MODULAR ARITHMETIC. s1 = “inverse of 2 in Z3 ” = 2. M1 = 105/3 = 35 ≡ 2 (mod 3). M3 = 105/7 = 15 ≡ 1 (mod 7). Hence.
. . . 8. . 2. 4. · · · 2 3 4 integers: 1 . . 1. s2 . . 9. The sequence of positive integers: 1. 5. 5.CHAPTER 4 Induction. Examples: 1. 5. . n2 . but we can start at any value of the index that we want. for instance if we declare s0 to be the ﬁrst term. .··· n In general the elements of a sequence are represented with an indexed letter. . The sequence of positive even integers: 2. 2. .1. . The sequence of powers of 2: 1. 7. The sequence is symbolically represented {sn } or {sn }∞ . . . 8. 7. .1. . . . . The sequence itself can be deﬁned by giving a rule. .1. 2. 4. . . 4. . . 1. . . 3.g. . The sequence of Fibonacci numbers (each one is the sum of the two previous ones): 0. . e. 9. the previous sequence would become: 1. . . The reciprocals of the positive 1 1 1 1. 13. n. s3 . 3. 5. n=1 59 . 16. . . . . Recurences 4. . . 2. say s1 . . . sn . . 2n. 3. A sequence is an (usually inﬁnite) ordered list of elements. . 8. 6. Sequences. 4.: sn = 2n + 1 is the sequence: 3. . Here we are assuming that the ﬁrst element is s1 . 3. Sequences and Strings 4. . .
.g. 2. b. n=3 4.. 1 . · · · . n=3 6 an = a3 · a4 · a5 · a6 = 7 · 9 · 11 · 13 = 9009 . (ab)3 = ababab. 4.g. while sn = 1/n is decreasing: 1. 8. .4. etc. 1 . aaaa. 4. bba. . 5. 9. we get the subsequence consisting of the even positive integers: 2. In order to abbreviate sums and products the following notations are used: 1. . a b a c  = 18. .3.2. Product notation: n ai = am · am+1 · am+2 · · · · · an i=m For instance: assume an = 2n + 1. Given a set X.. 5 . 4. If sn ≥ sn+1 then it is called decreasing. then 6 an = a3 + a4 + a5 + a6 = 7 + 9 + 11 + 13 = 40 . Sum (Sigma) and Product Notation. SEQUENCES AND STRINGS 60 If sn ≤ sn+1 for every n the sequence is called increasing. etc. . e. c}. Repetitions can be speciﬁed with a superscripts. The length of a string is its number of elements. . then the following are examples of strings over X: aba. 3. .1.1. Sum (or sigma) notation: n ai = am + am+1 + am+2 + · · · + an i=m 2. 2 3 4 If we remove elements from a sequence we obtain a subsequence. Strings. . 1 . abaccbab = 8. Example: If X is the set X = {a. for instance: a2 b3 ac2 a3 = aabbbaccaaa. if we remove all odd numbers from the sequence of positive integers: 1.1. . 6. E. 2 7 3 6 . a string over X is a ﬁnite ordered list of elements of X. For instance sn = 2n + 1 is increasing: 3. 7.
the string consisting of α followed by β is called the concatenation of α and β. Given two strings α and β over X. zero: λ = 0. The set of all strings over X is represented X ∗ . SEQUENCES AND STRINGS 61 The string with no elements is called null string.. represented λ. The set of no null strings over X (i. Its length is. of course.e. . all strings over X except the null string) is represented X + .1. For instance if α = abac and β = baaab then αβ = abacbaaab.4.
so the property is true for 1. then P (n + 1) is true. Then the conclusion is that the property is true for every integer n greater than or equal to m. 4. Inductive Step: Assume (Induction Hypothesis) that the property is true for some positive integer n. Let P be a property of positive integers such that: 1. and so on. Then the inductive step implies that P (2) is also true. i. Consequently the property is true for all positive integers. In fact: S(n + 1) = 1 + 3 + 5 + · · · + (2n + 1) = S(n) + 2n + 1 . The validity of the Principle of Mathematical Induction is obvious. Principle of Mathematical Induction. Inductive Step: if P (n) is true. Answer : Let S(n) = 1 + 3 + 5 + · · · + (2n − 1). 2.2.: S(n) = n2 .e. Mathematical Induction Many properties of positive integers can be proved by mathematical induction. MATHEMATICAL INDUCTION 62 4. i. i. Remark : The premise P (n) in the inductive step is called Induction Hypothesis.1.e. Remark : In the basis step we may replace 1 with some other integer m. 1.4. 1 + 3 + 5 + · · · + (2n − 1) = n2 .2. The basis step states that P (1) is true.. By the inductive step again we see that P (3) is true. and 2. S(n) = n2 . Basis Step: If n = 1 we have S(1) = 1 = 12 . Basis Step: P (1) is true. S(n + 1) = (n + 1)2 .2. Then P (n) is true for all positive integers. We must prove that it is also true for n + 1.e. We want to prove by induction that for every positive integer n.. Example: Prove that the sum of the n ﬁrst odd positive integers is n2 . .
2. and 2. and shows that the property is true for all positive integers. 6 • 12 + 22 + 32 + · · · + n2 = • 13 + 23 + 33 + · · · + n3 = (1 + 2 + 3 + · · · + n)2 .. i. hence 2(n + 1) + 1 = 2n + 3 ≤ 2n + 2n = 2n+1 . Inductive Step: Assume (Induction Hypothesis) that the property is true for some positive integer n.e. so the property is true in this case. 2. 2(n + 1) + 1 ≤ 2n+1 . Inductive Step: if P (k) is true for all 1 ≤ k ≤ n then P (n + 1) is true. Basis Step: If n = 3 we have 2n + 1 = 2 · 3 + 1 = 7 and 2n = 23 = 8.e.2.: 2n + 1 ≤ 2n . 1. MATHEMATICAL INDUCTION 63 But by induction hypothesis. Strong Form of Mathematical Induction. Example: Prove that 2n + 1 ≤ 2n for n ≥ 3. . and we also have that 3 ≤ 2n if n ≥ 3.2. Then P (n) is true for all positive integers. and shows that the property is true for all n ≥ 3. 2 n (n + 1) (2n + 1) . This completes the induction. This completes the induction. Basis Step: P (1) is true. Answer : This is an example in which the property is not true for all positive integers but only for integers greater than or equal to 3. Exercise: Prove the following identities by induction: • 1 + 2 + 3 + ··· + n = n (n + 1) . Let P be a property of positive integers such that: 1. hence: S(n + 1) = n2 + 2n + 1 = (n + 1)2 . i. S(n) = n2 . By the induction hypothesis we know that 2n ≤ 2n . 4. We must prove that it is also true for n + 1.4.
4. so the property holds for n = 2. k2 < n + 1. either n + 1 is a prime number or it is not.. and so on. Now we have 2 = a2 /b2 . Since the left hand side is even. Consequently. then k is a prime number or a product of primes. √ √ Example: Prove that 2 is irrational (i.e. such that 1 < k1 . hence n + 1 is a product of primes. From here we see √ that 2 = b/a . Inductive Step: Assume that if 2 ≤ k ≤ n. where a and √ b are integers. contradicting the wellordering √ principle. n + 1 = k1 k2 .2. but this implies that a itself is even. If it is not a prime number.2. hence 2 b2 = a2 . and simplifying: b2 = 2 a 2 . MATHEMATICAL INDUCTION 64 Example: Prove that every integer n ≥ 2 is prime or a product of primes.. If it is a prime number then it veriﬁes the property. then it can be written as the product of two positive integers. By induction hypothesis each of k1 and k2 must be a prime or a product of primes. 2 cannot be written as a quotient of two positive integers) using the wellordering principle. Hence starting with a fractional representation of 2 = a/b √ we end up with another fractional representation 2 = b/a with a smaller numerator b < a. . then a2 is even. Repeating the same argument with the fraction b/a we get another fraction with an even smaller numerator. This completes the proof. So the set of possible numerators of a fraction representing √ 2 cannot have a smallest element. Note that since 2 > 1 then a > b. √ √ Answer : Assume that 2 is rational. 2 = a/b. Hence: √ 2 b2 = 4 a 2 . Answer : 1.e. Basis Step: 2 is a prime number. our assumption that 2 is rational has to be false. The WellOrdering Principle. so a = 2 a . 2.3. 4. Every nonempty set of positive integers has a smallest element. i. Now.
In the following we assume that the coeﬃcients C0 . by using the recurrence repeatedly until obtaining a explicit closeform formula. Ck are constant. We will focus on kthorder linear recurrence relations. i. when the equations involved in that deﬁnition have a unique solution.e. For instance consider the following recurrence relation: xn = r xn−1 (n > 0) . Otherwise it is called nonhomogeneous.3. i. in the recursive deﬁnition of the Fibonacci sequence. So.. . F1 = 1 . C1 . One way to solve some recurrence relations is by iteration. The point is that a recursive deﬁnition is actually a definition when there is one and only one object satisfying it.4. hence the solution is xn = A rn . . RECURRENCE RELATIONS 65 4. If bn = 0 the recurrence relation is called homogeneous. for instance..e. Also.3. and the initial conditions are F0 = 0. which are of the form C0 xn + C1 xn−1 + C2 xn−2 + · · · + Ck xn−k = bn . . . where C0 = 0. The recursive step in a recursive deﬁnition is also called a recurrence relation. Rather than deﬁnitions they will be considered as equations that we must solve. The basis of the recursive deﬁnition is also called initial conditions of the recurrence. . x0 = A . By using the recurrence repeatedly we get: xn = r xn−1 = r2 xn−2 = r3 xn−3 = · · · = rn x0 = A rn . Recurrence Relations Here we look at recursive deﬁnitions under a diﬀerent point of view. the recurrence is Fn = Fn−1 + Fn−2 or Fn − Fn−1 − Fn−2 = 0 . the solution to those equations may provide a closedform (explicit) formula for the object deﬁned.
RECURRENCE RELATIONS 66 4.01n + 100. Example: Assume that a country with currently 100 million people has a population growth rate (birth rate minus death rate) of 1% per year. 000 . The ﬁrst one is xn = r xn−1 + c (n > 0). 000 (n > 0). 000. Find its population in 10 years from now. The nonhomogeneous case can be written in the following way: xn = r xn−1 + cn (n > 0) . 000.4. 000 1. and xn = A + c n if r = 1 . and it also receives 100 thousand immigrants per year (which are quickly assimilated and reproduce at the same rate as the native population).01 + 1000 (1. 000. First Order Recurrence Relations. c = 100.01n − 1 xn = 100. x0 = A . x0 = 100. The solution is n x0 = A . we have: xn = 1.) Answer : If we call xn = population in year n from now. which is a geometric sequence with ratio r. 000 and A = 100.01 xn−1 + 100.01 − 1 n = 100. x0 = A . n (n > 0) . where c is a constant. . its solution can be expressed like this: xn = A r + k=1 n ck rn−k . This is the equation above with r = 1. 00.1.01.01n − 1) . The homogeneous case can be written in the following way: xn = r xn−1 Its general solution is xn = A r n . 000 · 1.3. (Assume that all the immigrants arrive in a single batch at the end of the year. rn − 1 r−1 xn = A r + c k=1 n rn−k = A rn + c if r = 1 . Using the summation notation. 000 · 1.3. hence: 1. We examine two particular cases. 000.
By plugging in the equation we get: C0 c rn + C1 c rn−1 + C2 c rn−2 = 0 . Double Real Root.2. The solution is now n x0 = A . RECURRENCE RELATIONS 67 So: x10 = 110.3. but in . 2. 317 . In this case the general solution of the recurrence relation is n n x n = c 1 r 1 + c 2 r2 . Now we look at the recurrence relation C0 xn + C1 xn−1 + C2 xn−2 = 0 . 462. c2 are arbitrary constants. In this case the solution could be expressed in the same way as in the case of distinct real roots. xn = A + k=1 (c + d k) = A + c n + d n (n + 1) . where c1 . r2 be the two (in general complex) roots of the above equation. They are called characteristic roots. Second Order Recurrence Relations. First we will look for solutions of the form xn = c rn . If r1 = r2 = r. called the characteristic equation of the recurrence: C0 r2 + C1 r + C2 = 0 .4. hence r must be a solution of the following equation. where c1 . Let r1 . We distinguish three cases: 1. Complex Roots. Distinct Real Roots.3. c2 are arbitrary constants. 3. The second particular case is for r = 1 and cn = c + d n. 2 4. the general solution of the recurrence relation is xn = c1 r n + c2 n r n . where c and d are constant (so cn is an arithmetic sequence): xn = xn−1 + c + d n (n > 0) .
4. Its roots are:2 1Remainder: 2φ = √ 1+ 5 2 eαi = cos α + i sin α. Using the initial conditions we get the value of the constants: √ (n = 0) c1 + c2 = 0 c1 = 1/√5 ⇒ c2 = −1/ 5 (n = 1) c1 φ + c2 (−φ−1 ) = 1 Hence: 1 Fn = √ 5 φn − (−φ)−n . √ √ 1+ 5 1− 5 −1 r1 = φ = . F0 = 0. is the Golden Ratio. r2 = r e−αi . k1 = c1 + c2 .3. F1 = 1 . so the general solution for the recurrence is: Fn = c1 φn + c2 (−φ−1 )n . The characteristic equation is r2 − r − 1 = 0 . r2 = −φ = . Example: Find a closedform formula for the Fibonacci sequence deﬁned by: Fn+1 = Fn + Fn−1 (n > 0) . k2 = (c1 − c2 ) i. 2 2 They are distinct real roots. which yields:1 xn = k1 rn cos nα + k2 rn sin nα . Answer : The recurrence relation can be written Fn − Fn−1 − Fn−2 = 0 . RECURRENCE RELATIONS 68 order to avoid the use of complex numbers we write r1 = r eαi . .
Set theoretical version of the rule of sum: If A and B are disjoint sets (A ∩ B = ∅) then A ∪ B = A + B . For instance. How many diﬀerent license plates can be printed? Answer : each letter can be printed in 26 ways. . The Rule of Product.1. 69 . The Rule of Sum. then performing either task can be accomplished in m + n ways. and the two tasks cannot be performed simultaneously. so 26 · 26 · 10 · 10 · 10 = 676000 diﬀerent plates can be printed. More generally. Then: A × B = A · B . and each digit can be printed in 10 ways.1.1. 5. . . if a class has 30 male students and 25 female students. If a task can be performed in m ways and another independent task can be performed in n ways. if the sets A1 .CHAPTER 5 Counting 5. assume that a license plate contains two letters followed by three digits.2. An are pairwise disjoint. Basic Principles 5. A2 . If a task can be performed in m ways. Set theoretical version of the rule of product: Let A × B be the Cartesian product of sets A and B. while another task can be performed in n ways. then the class has 30 + 25 = 45 students. . For instance. then the combination of both tasks can be performed in mn ways. then: A1 ∪ A2 ∪ · · · ∪ An  = A1  + A2  + · · · + An  . More generally: A1 × A2 × · · · × An  = A1  · A2  · · · An  .1.
and for an arbitrary union of sets: A1 ∪ A2 ∪ · · · ∪ An  = s1 − s2 + s3 − s4 + · · · ± sn . For three sets the following formula applies: A ∪ B ∪ C = A + B + C − A ∩ B − A ∩ C − B ∩ C + A ∩ B ∩ C . where sk = sum of the cardinalities of all possible kfold intersections of the given sets. and 50 students are taking both. How many students are taking at least one of those courses? Answer : If U = total set of students in the university. M = set of students taking Mathematics.1. 300 are taking a course in physics. Example: Assume that in a university with 1000 students. 200 students are taking a course in mathematics. The InclusionExclusion Principle.3. ﬁnd the number of functions from A to B. P = set of students taking Physics. BASIC PRINCIPLES 70 Exercise: Given a set A with m elements and a set B with n elements. In general.1. The inclusionexclusion principle generalizes the rule of sum to nondisjoint sets. . 5.5. then: M ∪ P  = M  + P  − M ∩ P  = 300 + 200 − 50 = 450 students are taking Mathematics or Physics. B: A ∪ B = A + B − A ∩ B . for arbitrary (but ﬁnite) sets A.
. 5. ca. Permutations. cd. e taken 3 at a time are: abc. b. Any subset of A of size r is called a combination of n elements taken r at a time. Note that the order is important. For instance. where two combinations are considered identical if they have the same elements regardless of their order. ace. By convention 0! = 1. Combinations. For instance. bac. bce. db. there are 3! = 6 permutations of the 3 letters a. b. abd. c. The permutations of size 2 of the letters a. b. ba. Combinatorics 5. b. bcd. Exercise: Given a set A with m elements and a set B with n elements.2. hence: P (n. bc.2. ﬁnd the number of onetoone functions from A to B. acb. bde.1.2. the second one in n − 1 ways. d.2. dc. cba. ad. cde. cb. If k = n then we call it just a permutation of the n objects. Any arrangement of any k of these objects in a given order is called a permutation of size k. Assume that we have a set A with n objects. We ﬁnd the number P (n. bca. cab. they are considered equal if they have the same elements arranged in the same order.5. The number P (n. bd. Assume that we have n objects. k) of permutations of size k of n given objects in the following way: The ﬁrst object in an arrangement can be chosen in n ways. c. k) = n × (n − 1) × where n! = 1 × 2 × 3 × ··· (k factors) ··· × (n − k + 1) = n! . k) of permutations of n objects is P (n. acd. the combinations of the letters a. 2) = 4 × 3 = 12. c. abe. For instance. da. and so on. c are the following: abc. COMBINATORICS 71 5. b. The number of permutations of size 2 of the 4 letters a. ac. d are: ab. the permutations of the letters a. (n − k)! (n factors) × n is called “n factorial ”. n) = n! . the third one in n − 2 ways. Given two permutations. ade. d is P (4.2. c.
In order to generate all possible permutations of size r of the elements of A we 1) take all possible subsets of size r in the set A. P (r. r) is the following. Task 1) can be performed in C(n. r! (n − r)! The symbol n r (read “n choose r”) is often used instead of C(n. r). r) ways. r) × P (r. COMBINATORICS 72 The number of subsets of size r in a set A with n elements is: n! C(n. r). Let A be a set with n objects. r) = = . r) = C(n. r) n! C(n. By the product rule we have P (n.2. r) ways.5. One way to derive the formula for C(n. r) = . and 2) permute the k elements in each subset in all possible ways. and task 2) can be performed in P (r. r) r! (n − r)! . hence P (n.
bbc. bbaaa. Assume that we have a set A with n elements. GENERALIZED PERMUTATIONS AND COMBINATIONS 73 5. We may solve this problem in the following way. . ddd. . is called a combination of n objects taken r at a time with repetition. task 3) can be performed in 2!. bcc.2. the combinations of the letters a. nk times the kth letter. nk ) = n! . Assume that we have an alphabet with k letters and we want to write all possible words containing n1 times the ﬁrst letter of the alphabet. then 3) these subscripts to place on the 2 b’s.3. . where each object can be selected more than once. generate each permutation of this ﬁve elements by choosing 1) the position of each kind of letter. aab. n2 times the second letter. b. where n = n1 + n2 + · · · + nk . n1 . abd. acd. Example: With 3 a’s and 2 b’s we can write the following 5letter words: aaabb. Permutations with Repeated Elements. n2 . c. . add. . Note that the following are equivalent: 1. d taken 3 at a time with repetition are: aaa. task 2) can be performed in 3! ways. nk ! 5. The number of combinations of n objects taken r at a time with repetition.3. Let us distinguish the diﬀerent copies of a letter with subscripts: a1 a2 a3 b1 b2 . n1 . Two combinations with repetition are considered identical if they have the same elements repeated the same number of times. . abaab. Combinations with Repetition. . Generalized Permutations and Combinations 5. 2) × 3! × 2!. abc. abb. Any selection of r objects from A. . nk ). bdd. aad. . regardless of their order. cdd. ccc. 3. n2 . n 1 ! n 2 ! . bbd. Next. . 2) ways. then 2) the subscripts to place on the 3 a’s.1. baaab. . baaba.. 3. Task 1) can be performed in P (5. ababa. bbb. . bcd. babaa. abbaa. as illustrated with the example above. aabba. acc. ccd. . By the product rule we have 5! = P (5. aabab.3. For instance. aac.3. 2) = 5!/3! 2!. 3.5. In general the formula is: P (n. . How many words can we write? We call this number P (n. hence P (5.
x3 be the quarts of milk to put in containers number 1. r. The number of nonnegative integer solutions of the equation: x1 + x2 + · · · + xn = r .5. GENERALIZED PERMUTATIONS AND COMBINATIONS 74 2. Example: Assume that we have 3 diﬀerent (empty) milk containers and 7 quarts of milk that we can measure with a one quart measuring cup. n − 1) = (n + r − 1)! = r! (n − 1)! n+r−1 . Instead of using numbers for writing the solutions. so the number of arrangements is P (9. The number of ways r identical objects can be distributed among n distinct containers. Let x1 . r . 7. 2) = 9!/7! 2! = 9 . x2 = 1. 7 The general solution is: P (n + r − 1. or 2 + 1 + 4. x2 . x3 = 4. each possible solution is an arrangement of 7 strokes and 2 plus signs. like this:  +  + . 3. Now. so for instance we represent the solution x1 = 2. The number of possible distributions of milk equals the number of non negative integer solutions for the equation x1 + x2 + x3 = 7. In how many ways can we distribute the milk among the three containers? We solve the problem in the following way.3. 2 and 3 respectively. we will use strokes.
4. 5. we have a + b = n. Binomial Theorem. expanding.1.5. called the Binomial Theorem: n (x + y) = k=0 n n n−k k x y k = n n n n−1 n n−2 2 x + x y+ x y + ··· 0 1 2 n n n + xy n−1 + y . Since there are n factors of the form (x + y). n k = n n−k . hence the terms must be of the form xn−k y k . The binomial coeﬃcients have the following properties: 1. which is C(n.4. Properties of Binomial Coeﬃcients.2. Binomial Coeﬃcients 5. k Exercise: Prove n n = 2n k k=0 n a b and k=0 (−1)k n k = 0. k) = n .4. k The expression n is often called binomial coeﬃcient. and grouping terms of the form x y .4. The following identities can be easily checked: (x + y)0 = 1 (x + y)1 = x + y (x + y)2 = x2 + 2 xy + y 2 (x + y)3 = x3 + 3 x2 y + 3 xy 2 + y 3 They can be generalized by the following formula. Hint: Apply the binomial theorem to (1 + 1)2 and (1 − 1)2 . The coeﬃcient of xn−k y k will be equal to the number of ways in which we can select the y from any k of the factors (and the x from the remaining n − k factors). BINOMIAL COEFFICIENTS 75 5. n−1 n We can ﬁnd this formula by writing (x + y)n = (x + y) × (x + y) × (n factors) ··· × (x + y) .
. .4. Using the n sum principle we ﬁnd that in fact n+1 = n + k+1 . i. Those subsets can be partitioned into two classes: that of the subsets containing a.. n . The properties shown in the previous section allow us to compute binomial coeﬃcients in a simple way. The set A has n+1 subsets k+1 of size k + 1. Pascal’s Triangle. BINOMIAL COEFFICIENTS 76 2. Look at the following triangular arrangement of binomial coeﬃcients: 1 0 3 1 0 0 2 1 4 2 1 1 3 2 4 0 3 0 2 0 4 1 2 2 4 3 3 3 4 4 We notice that each binomial coeﬃcient on this arrangement must be the sum of the two closest binomial coeﬃcients on the line above it. The number of subsets containing a equals the number of subsets of A − {a} of size k. allows us to compute very quickly 0 n the values of the binomial coeﬃcients on the arrangement: 1 1 1 1 1 4 3 6 2 3 4 1 1 1 1 This arrangement of binomial coeﬃcients is called Pascal’s Triangle. i. The number of subsets not containing a is k n the number of subsets of A − {a} of size k + 1. n+1 k+1 = n n + k k+1 n k = n! .e. 1 1Although it was already known by the Chinese in the XIV century.3.e. k+1 k 5.5. and that of the subsets not containing a. k+1 .4. This together with n = n = 1. k!(n − k)! The ﬁrst property follows easily from The second property can be proved by choosing a distinguished element a in a set A of n + 1 elements.
5.5. THE PIGEONHOLE PRINCIPLE
77
5.5. The Pigeonhole Principle 5.5.1. The Pigeonhole Principle. The pigeonhole principle is used for proving that a certain situation must actually occur. It says the following: If n pigeonholes are occupied by m pigeons and m > n, then at least one pigeonhole is occupied by more than one pigeon.1 Example: In any given set of 13 people at least two of them have their birthday during the same month. Example: Let S be a set of eleven 2digit numbers. Prove that S must have two elements whose digits have the same diﬀerence (for instance in S = {10, 14, 19, 22, 26, 28, 49, 53, 70, 90, 93}, the digits of the numbers 28 and 93 have the same diﬀerence: 8 − 2 = 6, 9 − 3 = 6.) Answer : The digits of a twodigit number can have 10 possible diﬀerences (from 0 to 9). So, in a list of 11 numbers there must be two with the same diﬀerence. Example: Assume that we choose three diﬀerent digits from 1 to 9 and write all permutations of those digits. Prove that among the 3digit numbers written that way there are two whose diﬀerence is a multiple of 500. Answer : There are 9 · 8 · 7 = 504 permutations of three digits. On the other hand if we divide the 504 numbers by 500 we can get only 500 possible remainders, so at least two numbers give the same remainder, and their diﬀerence must be a multiple of 500. Exercise: Prove that if we select n + 1 numbers from the set S = {1, 2, 3, . . . , 2n}, among the numbers selected there are two such that one is a multiple of the other one.
Pigeonhole Principle (Schubfachprinzip) was ﬁrst used by Dirichlet in Number Theory. The term pigeonhole actually refers to one of those oldfashioned writing desks with thin vertical wooden partitions in which to ﬁle letters.
1The
CHAPTER 6
Probability
6.1. Probability 6.1.1. Introduction. Assume that we perform an experiment such as tossing a coin or rolling a die. The set of possible outcomes is called the sample space of the experiment. An event is a subset of the sample space. For instance, if we toss a coin three times, the sample space is S = {HHH, HHT, HT H, HT T, T HH, T HT, T T H, T T T } . The event “at least two heads in a row” would be the subset E = {HHH, HHT, T HH} . If all possible outcomes of an experiment have the same likelihood of occurrence, then the probability of an event A ⊂ S is given by Laplace’s rule: E P (E) = . S For instance, the probability of getting at least two heads in a row in the above experiment is 3/8. 6.1.2. Probability Function. In general the likelihood of diﬀerent outcomes of an experiment may not be the same. In that case the probability of each possible outcome x is a function P (x). This function veriﬁes: 0 ≤ P (x) ≤ 1 for all x ∈ S and P (x) = 1 .
x∈S
The probability of an event E ⊆ S will be P (E) =
x∈E
P (x)
78
6.1. PROBABILITY
79
Example: Assume that a die is loaded so that the probability of obtaining n point is proportional to n. Find the probability of getting an odd number when rolling that die. Answer : First we must ﬁnd the probability function P (n) (n = 1, 2, . . . , 6). We are told that P (n) is proportional to n, hence P (n) = kn. Since P (S) = 1 we have P (1)+P (2)+· · · P (6) = 1, i.e., k ·1+k ·2+ · · · + k · 6 = 21k = 1, so k = 1/21 and P (n) = n/21. Next we want to ﬁnd the probability of E = {2, 4, 6}, i.e. P (E) = P (2) + P (4) + P (6) = 2 4 6 12 + + = . 21 21 21 21 6.1.3. Properties of probability. Let P be a probability function on a sample space S. Then: 1. For every event E ⊆ S, 0 ≤ P (E) ≤ 1 . 2. P (∅) = 0, P (S) = 1. 3. For every event E ⊆ S, if E = is the complement of E (“not E”) then P (E) = 1 − P (E) . 4. If E1 , E2 ⊆ S are two events, then P (E1 ∪ E2 ) = P (E1 ) + P (E2 ) − P (E1 ∩ E2 ) . In particular, if E1 ∩ E2 = ∅ (E1 and E2 are mutually exclusive, i.e., they cannot happen at the same time) then P (E1 ∪ E2 ) = P (E1 ) + P (E2 ) . Example: Find the probability of getting a sum diﬀerent from 10 or 12 after rolling two dice. Answer : We can get 10 in 3 diﬀerent ways: 4+6, 5+5, 6+4, so P (10) = 3/36. Similarly we get that P (12) = 1/36. Since they are mutually exclusive events, the probability of getting 10 or 12 is P (10) + P (12) = 3/36 + 1/36 = 4/36 = 1/9. So the probability of not getting 10 or 12 is 1 − 1/9 = 8/9. 6.1.4. Conditional Probability. The conditional probability of an event E given F , represented P (E  F ), is the probability of E assuming that F has occurred. It is like restricting the sample space to F . Its value is P (E ∩ F ) . P (E  F ) = P (F )
(6. Two events E and F are said to be independent if the probability of one of them does not depend on the other. P (hitting the target at least once in three shots) = 1−0. The probability that the shooter does not hit the target three times in a row. 5). 2. 2. Now. e.33 = 0. 3.027 = 0. The probability that the shooter does not hit the target in one shot. (5. Find: 1. Note that if E is independent of F then also F is independent of E. Find the probability of that event if we know that at least one of the dice shows 5 points.1. (3. (5. 6. 6). . 1). Hence the probability is P (E) = E/S = 3/36 = 1/12. and the ways to obtain a sum 10 are E ∩ F = {(5. E ∩ F  = 1.973. so E = 3. PROBABILITY 80 Example: Find the probability of obtaining a sum of 10 after rolling two fair dice. 5)}. (2. (5. 2). P (F  E) = P (F ). 4)}. and that hitting the target in diﬀerent shots are independent events. The number of possible outcomes is 6 × 6 = 36.g.6. 5).5.. 5). Independent Events. 5). 5). 4). Answer : 1.g. (5. The probability that the shooter hits the target at least once after shooting three times.7. 6)} . Example: Assume that the probability that a shooter hits a target is p = 0. 3). e. In this circumstances: P (E ∩ F ) = P (E) · P (F ) . 3. so F  = 11. (4. Answer : We call E = “obtaining sum 10” and F = “at least one of the dice shows 5 points”. so the probability is P (E  F ) = P (E ∩ F )/P (F ) = 1/11. (5.027. if we know that at least one of the dice shows 5 points then the sample space shrinks to F = {(1.1. The event “obtaining a sum 10” is E = {(4.3. (5. P (not hitting the target in one shot) = 1 − 0. 5).: P (E  F ) = P (E) . (5. P (not hitting the target three times in a row) = 0. (6.7 = 0. 5).
PROBABILITY 81 6. . P (F  C2 ) = 0.05094 . 000. We have: S = 100. 000. Then for any event F we have n P (F ) = i=1 P (F  Ci ) P (Ci ) P (F  Cj ) P (Cj ) . A person is given the test and it comes out positive.001.05 · 0. A test designed to detect the disease has a 99% probability of detecting it when administered to a person who has it.99.6. Suppose that a sample space S is partitioned into n classes C1 .1. . hence P (C1 ) = C1 /S = 0.) .001 = P (F ) 0.1. 900.019434628 · · · ≈ 2% .99 · 0. P (F ) and P (Cj  F ) = Example: In a country with 100 million people 100 thousand of them have disease X.05094 = 0.999. Bayes’ Theorem. 000.999 = 0. P (C2 ) = C2 /S = 0. What is the probability that that person has the disease? Answer : The classes are C1 = “has the disease” and C2 = “does not have the disease”. . and by Bayes’ theorem: P (C1  F ) = P (F  C1 ) · P (C1 ) 0.05. C1  = 100. C2 . but it also has a 5% probability of giving a false positive when given to a person who does not have it. Also P (F  C1 ) = 0. (So the test is really of little use when given to a random person— however it might be useful in combination with other tests or other evidence that the person might have the disease.99 · 0. 000. Cn which are pairwise mutually exclusive and whose union ﬁlls the whole sample space. C2  = 99. and the event is F = “the test gives a positive”. .001 + 0.6. Hence: P (F ) = P (F  C1 ) · P (C1 ) + P (F  C2 ) · P (C2 ) = 0.
In general a graph G consists of two things: 1. whose elements are called vertices. A road map. If the edges are directed then they are also called directed edges or arcs. The representation of a binary relation deﬁned on a given set. Each edge e ∈ E is associated with a pair of vertices. The edge set E or set of edges connecting pairs of vertices.1. nodes or points. The relation of a given element x to another element y is represented with an arrow connecting x to y.1.1. consisting of a number of towns connected with roads. 2. The latter is an example of a directed graph or digraph. 82 . Undirected Graph.CHAPTER 7 Graph Theory 7.1. Graphs. Graphs 7. a b d c Figure 7. 2. Consider the following examples: 1. The vertex set V . The former is an example of (undirected) graph.
y}.7. the vertex x is called origin. an edge pointing from vertex x to vertex y is said to be incident from x and incident to y. Graph with parallel edges.1. but the name of an edge. GRAPHS 83 a b d c Figure 7.2. If the graph is directed and there is a unique edge e pointing from x to y. Two vertices connected by an edge are called adjacent. y) is not an ordered pair. If the graph is directed. then we may write e = (x.3. . understanding that here (x. so E may be regarded as a set of ordered pairs. They are also the endpoints of the edge. An edge connecting a vertex to itself is called a loop. E) (we assume V and E ﬁnite). Two edges connecting the same pair of points (and pointing in the same direction if the graph is directed) are called parallel or multiple. If the graph is undirected and there is a unique edge e connecting x and y we may write e = {x. If e = (x. and the edge is said to be incident to each of its endpoints. source or initial point of the edge e. terminating vertex or terminal point. y). so E can be regarded as set of unordered pairs. Directed Graph. y). A graph is sometimes represented by the pair (V. In this context we may also write e = (x. a b d c Figure 7. and y is called the terminus. y).
is the number of edges with v as their initial vertex. so there must be an even number of them. But its terms are all odd. so the second sum must be even too.) − Number of vertices of odd degree.1.7.1. The outdegree of a vertex v. Graph Terminology Type Simple graph Multigraph Pseudograph Directed graph Directed multigraph Edges indirected indirected indirected directed directed Multiple Edges Allowed? no yes yes no yes Loops Allowed? no no yes yes yes The degree of a vertex v.) In a graph with directed edges. (Note that a loop at a vertex contributes 1 to both the indegree and the outdegree of this vertex. E). Since deg(v) is even for v ∈ Ve . If a graph has multiple edges but no loops then it is called a multigraph. A vertex of degree zero (not connected to any other vertex) is called isolated. denoted deg (v). which is even. . The total sum must be 2e. Proof: Let Ve and Vo respectively the set of vertices of even degree and the set of vertices of odd degree in an undirected graph G = (V. The following table summarizes the graph terminology Table 7. E) be an undirected graph with e edges. the indegree of a vertex v. If it has loops (and possible also multiple edges) then it is called a pseudograph. Let G = (V.1. the ﬁrst sum in the right hand side of the equality is even. Then 2e = deg (v) . A vertex of degree 1 is called pendant. An undirected graph has an even number of vertices of odd degree. is the number of edges that contain it (loops are counted twice). Then 2e = v∈V deg (v) = v∈Ve deg (v) + v∈Vo deg (v) . represented deg(v). is the number of edges with v as their terminal vertex. denoted deg+ (v). The Handshaking Theorem. GRAPHS 84 A graph with neither loops nor multiple edges is called a simple graph. v∈V (This applies even if multiple edges and loops are present.
for each pair of distinct vertices. The ncube: A graph with with 2n vertices labeled 0. GRAPHS 85 Sum of degrees in an directed graph. The complete graph of n vertices is represented Kn (ﬁg. 1. . Complete Graph: a simple undirected graph G such that every pair of distinct vertices in G are connected by an edge. either (x. Here we examine a few special graphs.1. v∈V v∈V A weighted graph is a graph whose edges have been labeled with numbers. x) (but not both) is in E. Special Graphs. 110 111 010 011 100 101 000 001 Figure 7. Weighted Graph. a 7 6 3 b 6 4 d c Figure 7.7. y) or (y. 2n − 1 so that two of them are connected with an edge if their binary representation diﬀers in exactly one bit. . E) such that every pair of distinct vertices in G are connected by exactly one edge—so.4.1. . 7. Let G = (V. . . The length of a path in a weighted graph is the sum of the weights of the edges in the path. 7.5.2.6). A complete directed graph is a simple directed graph G = (V. 3cube. E) be a directed graph. Then deg− (v) = deg+ (v) = E .
7.n (ﬁg.4 . If V1  = m and V2  = n. Question: Is the ncube bipartite. the corresponding complete bipartite graph is represented Km. E) in which V can be partitioned into two subsets V1 and V2 so that each edge in G connects some vertex in V1 to some vertex in V2 . a subgraph G = (V . For instance. GRAPHS a 86 e b d c Figure 7.1.7).6. E). Complete bipartite graph K3.1. A graph is bipartite iﬀ its vertices can be colored with two colors so that every edge connects vertices of diﬀerent color. If V = V then G is called a spanning subgraph of G. A bipartite simple graph is called complete if each vertex in V1 is connected to each vertex in V2 .7. Complete graph K5 . . 7. p a q b r c s Figure 7. color in blue the ones with an odd number of 1’s.7. Regular Graph: a simple graph whose vertices have all the same degree. Given a graph G = (V. Bipartite Graph: a graph G = (V. Hint: color in red all vertices whose binary representation has an even number of 1’s.3. Subgraph. the ncube is regular. E ) of G is another graph such that V ⊆ V and E ⊆ E.
7. the subgraph of G induced by U . GRAPHS 87 Given a subset of vertices U ⊆ V . and its edge set contains all edges from G connecting vertices in U .1. denoted U . . is the graph whose vertex set is U .
so in the example of ﬁgure 7. is the number of edges incident on i and j.7. column j. Representations of Graphs 7. The adjacency matrix of a graph is a matrix with rows and columns labeled by the vertices and such that its entry in row i. namely entry (i.8 One of the uses of the adjacency matrix A of a simple graph G is to compute the number of paths between two vertices.2.8 entry (d. Incidence matrix. For instance the following is the adjacency matrix of the graph of ﬁgure 7. e3 c e5 7.8: some authors if i = j then the entry is twice the number of loops incident on i.1. i = j. d) would be 2 instead of 1.2.2. 1For .2. the following is the incidence matrix of graph of ﬁgure 7. REPRESENTATIONS OF GRAPHS 88 7.8:1 a b c d a 0 1 0 1 b c d 1 0 2 0 0 2 0 0 1 0 0 1 b e2 e1 a e4 d Figure 7. Adjacency matrix. so that entry for row v column e is 1 if e is incident on v. As an example.2. The incidence matrix of a graph G is a matrix with rows labeled by vertices and columns labeled by edges. and 0 otherwise. j) of An is the number of paths of length n from i to j.
7.2. REPRESENTATIONS OF GRAPHS
89
a b c d
e1 e2 e3 e4 e5 1 0 0 1 0 1 1 1 0 0 0 1 1 0 0 0 0 0 1 1
7.2.3. Graph Isomorphism. Two graphs G1 = (V1 , E1 ), G2 = (V2 , E2 ), are called isomorphic if there is a bijection f : V1 → V2 and a bijection g : E1 → E2 such that an edge e is adjacent to vertices v and w if and only if g(e) is adjacent to f (v) and f (w) (ﬁg. 7.9).
a1 b1 a2 b2
d2
e1 d1
c1
e2
c2
Figure 7.9. Two isomorphic graphs. Two graphs are isomorphic if and only if for some ordering of their vertices their adjacency matrices are equal. An invariant is a property such that if a graph has it then all graphs isomorphic to it also have it. Examples of invariants are their number of vertices, their number of edges, “has a vertex of degree k”, “has a simple cycle of length l”, etc. It is possible to prove that two graphs are not isomorphic by showing an invariant property that one has and the other one does not have. For instance the graphs in ﬁgure 7.10 cannot be isomorphic because one has a vertex of degree 2 and the other one doesn’t.
7.2. REPRESENTATIONS OF GRAPHS
c 3
90
b
e 2 4
a
f
1
5
Figure 7.10. Non isomorphic graphs.
7.3. PATHS AND CIRCUITS
91
7.3. Paths and Circuits 7.3.1. Paths. A path from v0 to vn of length n is a sequence of n+1 vertices (vk ) and n edges (ek ) of the form v0 , e1 , v1 , e2 , v2 , . . . , en , vn , where each edge ek connects vk−1 with vk (and points from vk−1 to vk if the edge is directed). The path may be speciﬁed by giving only the sequence of edges e1 , . . . , en . If there are no multiple edges we can specify the path by giving only the vertices: v0 , v1 , . . . , vn . The path is a circuit (or cycle) if it begins and ends at the same vertex, i.e., v0 = vn , and has lenght greater than zero. A path or circuit is simple if it does not contain the same edge twice. 7.3.2. Connected Graphs. A graph G is called connected if there is a path between any two distinct vertices of G. Otherwise the graph is called disconnected. A directed graph is connected if its associated undirected graph (obtained by ignoring the directions of the edges) is connected. A directed graph is strongly connected if for every pair of distict points u, v, there is a path from u to v and there is a path from v to u. A connected component of G is any connected subgraph G = (V , E ) of G = (V, E) such that there is not edge (in G) from a vertex in V to a vertex in V − V . Given a vertex in G, the component of G containing v is the subgraph G of G consisting of all edges and vertices of g contained in some path beginning at v. 7.3.3. The Seven Bridges of K¨nigsberg. This is a classical o problem that started the discipline today called graph theory. During the eighteenth century the city of K¨nigsberg (in East Pruso 1 sia) was divided into four sections, including the island of Kneiphop, by the Pregel river. Seven bridges connected the regions, as shown in ﬁgure 7.11. It was said that residents spent their Sunday walks trying to ﬁnd a way to walk about the city so as to cross each bridge exactly once and then return to the starting point. The ﬁrst person to solve the problem (in the negative) was the Swiss mathematician Leonhard Euler in 1736. He represented the sections of the city and the seven bridges by the graph of ﬁgure 7.12, and proved that it is impossible to ﬁnd a path in it that transverses every edge of the graph exactly once. In the next section we study why this is so.
1The
city is currently called Kaliningrad and is part of the Russian republic.
3. An Euler path in G is a simple path that transverses every edge of the graph exactly once.12. Let G = (V. Recall that the degree of a vertex v. the number of odd vertices in a graph is always even. Existence of Euler Paths and Circuits. Then G contains an Euler circuit if and only if G its vertices have even degree. represented deg(v).4. As a consequence. o Figure 7.7. G contains an Euler path from vertex a to vertex b (= a) if and only if a and b have odd degree. an odd vertex is a vertex with odd degree. The sum of the degrees of all vertices in a graph equals twice its number of edges. E) be a graph with no isolated vertices. an Euler circuit in G is a simple circuit that transverses every edge of the graph exactly once. Graph for the Seven Bridges of K¨nigsberg. The graphs that have an Euler path can be characterized by looking at the degree of their vertices. An even vertex is a vertex with even degree. so it is an even number. Euler paths and circuits. Also. Let G be a connected multigraph. Analogously. PATHS AND CIRCUITS 92 Figure 7. o 7.11. . The Seven Bridges of K¨nigsberg. is the number of edges that contain v (loops are counted twice). and all its other vertices have even degree.3.
13.3. although often it is possible to argue that a graph has no Hamilton circuit.7.14 we cannot remove any edge because that would make the degree of b. so any circuit in G must have the same number of vertices from each of both sets. PATHS AND CIRCUITS 93 7. Edge removal argument. For instance if G = (V. Another kind of argument consists of removing edges trying to make the degree of every vertex equal two. e or d less than 2. Note that by deleting an edge in a Hamilton circuit we get a Hamilton path. so it is impossible to reduce the degree of a and c. The converse is not true. E) is a bipartite graph with vertex partition {V1 . i. . Dirac’s Theorem.e. so if a graph has a Hamilton circuit. A Hamilton path in G is a path (not a circuit) that contains each vertex of G once. Hamilton’s puzzle. In general it is not easy to determine if a given graph has a Hamilton path or circuit. Consequently that graph has no Hamilton circuit. because any path must contain alternatively vertices from V1 and V2 . If G is a simple graph with n vertices with n ≥ 3 such that the degree of every vertex in G is at least n/2. which occurs twice). then G cannot have a Hamilton circuit if V1  = V2 . then it also has a Hamilton path. A Hamilton circuit in a graph G is a circuit that contains each vertex of G once (except for the starting and ending vertex. V2 } (so that each edge in G connects some vertex in V1 to some vertex in V2 ). Hamilton Circuits. Exercise: Find a graph with a Hamilton path but no Hamilton circuit..3. a t m l b k j c o s r n q p i f e g h d Figure 7. a graph may have a Hamilton path but not a Hamilton circuit. For instance in the graph of ﬁgure 7. then G has a Hamilton circuit.5.
or belong to the class NP.. the traveling salesperson problem (TSP) consists of ﬁnding a Hamilton circuit of minimum length in this graph. One way to solve the problem consists of searching all possible Hamilton circuits and computing their length.3.7.) Given a weighted graph with n vertices the problem of determining whether it contains a Hamilton circuit of length not greater than a given L is known to be NPcomplete. Given a weighted graph. if it can be solved with an algorithm of complexity O(nk ) for some integer k. A decision problem is said to be polynomial.e. A nondeterministic algorithm is an algorithm that works with an extra hint. i. It is said to be nondeterministic polynomial. Remark : (Digression on P/NP problems. This means the following. a problem whose solution is “yes” or “no”. Graph without Hamilton circuit. or belong to the class P. First it is a decision problem. if in all cases when the answer is “yes” this can be determined with a nondeterministic algorithm of complexity O(nk ). Unfortunately no eﬃcient algorithm is known for solving this problem (and chances are that none exists). then G has a Hamilton circuit. The name comes from a classical problem in which a salesperson must visit a number of cities and go back home traveling the minimum distance possible. for instance in the TSP. The Traveling Salesperson Problem. if G has a Hamilton circuit of length not greater than L the hint could consist of a Hamilton circuit with length not greater than L—so the task of the algorithm would be just to check that in fact that length . but this is very ineﬃcient.14. Ore’s Theorem. If G is a simple graph with n vertices with n ≥ 3 such that deg(u) + deg(v) ≥ n for every pair of nonadjacent vertices u and v in G. PATHS AND CIRCUITS 94 b a e c d Figure 7.
The problem of ﬁnding a gray code is equivalent to ﬁnding a Hamilton circuit in the ncube. Every nbinary string appears somewhere in the sequence.7. 011. This is an algorithm to ﬁnd the shortest path from a vertex a to another vertex z in a connected weighted graph. so that their complexity must be higher than polynomial unless P=NP. For instance: 000. .L(v)+w(v. The output is L(z) = length of a minimum path from a to z. s2n and s1 diﬀer in exactly one bit.3.{v} 12: for each x in T adjacent to v 13: L(x) := min{L(x).L) 2: L(a) := 0 3: for all vertices x = a 4: L(x) := ∞ 5: T := set of all vertices 6: {T is the set of all vertices whose shortest} 7: {distance from a has not been found yet} 8: while z in T 9: begin 10: choose v in T with minimum L(v) 11: T := T . In a sense the P=NP problem consist of determining whether every problem whose solution is easy to check is also easy to solve. 101. 110.6. Edge (i. 100. The class NP contains a subclass called NPcomplete containing the “hardest” problems of the class. 111.3. . 2Informally. 1: procedure dijkstra(w. Two consecutive strings si and si+1 diﬀer exactly in one bit. . although it is strongly suspected that it is. s2 .2 Currently it is not known whether the class NP is strictly larger than the class P. . and NP problems are problems whose answer is “easy to check”. j) has weight w(i. s2n of nbinary strings verifying the following conditions: 1. 2.x)} P problems are “easy to solve”. .z. and vertex x is labeled L(x) (minimum distance from a if known.a. PATHS AND CIRCUITS 95 is not greater than L. 010. A Gray code is a sequence s1 . j) > 0. The TSP is one of these problems. 7. otherwise ∞). 001. Dijkstra’s ShortestPath Algorithm. Gray Codes. 3.
connected weighted graph.15. b 2 a 4 e 2 f 3 2 d 1 4 1 c 3 z Figure 7. Dijkstra’s algorithm has a worstcase run time of Θ(n2 ).15.3. Complexity of Dijkstra’s algorithm. Shortest path from a to z. For an nvertex. The algorithm would label the vertices in the following way in each iteration (the boxed vertices are the ones removed from T ): iteration 0 1 2 3 4 5 6 a 0 0 0 0 0 0 0 b c d e f z ∞ ∞ ∞ ∞ ∞ ∞ 2 ∞ 3 4 ∞ ∞ 2 3 3 4 ∞ ∞ 2 3 3 4 ∞ 6 2 3 3 4 ∞ 4 2 3 3 4 6 4 2 3 3 4 6 4 At this point the algorithm returns the value 4. PATHS AND CIRCUITS 96 14: end 15: return L(z) 16: end dijkstra For instance consider the graph in ﬁgure 7. simple. .7.
two graphs are homeomorphic if they can be obtained from the same graph by a sequence of elementary subdivisions.16. Planar Graphs. Planar Graphs 7.4. Three homeomorphic graphs. 7. then all graphs homeomorphic to G are also planar.3. v2 ) is called a series reduction. v) with two edges (u. A graph G is planar if it can be drawn in the plane with its edges intersecting at their vertices only. A particular planar representation of a planar graph is called a map. in the third graph of ﬁgure 7. If a graph G has a vertex v of degree 2 and edges (v. For instance. v). v2 ) with v1 = v2 . A map divides the plane into a number of regions or faces (one of them inﬁnite). (v. where w is a new vertex.4.1. v1 ). Deleting such vertex v and replacing (v.7.16. 7. c c f c b d b d b d a e a e a h e Figure 7. One such drawing is called an embedding of the graph in the plane. d) are in series.2. 7. the edges (h. Two graphs are said to be homeomorphic if they are isomorphic or can be reduced to isomorphic graphs by a sequence of series reductions (ﬁg. PLANAR GRAPHS 97 7. Some Results About Planar Graphs.4. Note that if a graph G is planar. By removing vertex h we get the ﬁrst graph in the left. b) and (h. The opposite of a series reduction is an elementary subdivision It consists of replacing an edge (u. Graph Homeomorphism. v1 ) and (v. v2 ) are in series.4. we say that the edges (v.16). v2 ) with (v1 .4. . v1 ) and (v. w) and (w. Equivalently.
The dual graph of that map is the graph Gd = (V d . pk }. Note that this implies that all plane embeddings of a given graph deﬁne the same number of regions. E) be a connected planar graph. i. If a graph can be properly . The coloring is called proper if there are no adjacent vertices with the same color. . there is an edge in E d connecting pi and pj . 7. Exercise: Find the duals of the maps shown in ﬁgure 7. The graph K3. from Euler’s formula we would have f = 5.e. If K3. Dual Graph of a Map. and r = number of regions in which some given embedding of G divides the plane. For each region ri . . e = E. Assume that the map divides the plane into a set of regions R = {r1 . .4. so 4f ≤ 2e. E) be a simple connected planar graph with v vertices.4. hence 3v − 6 = 9 < e = 10. 7. and for each edge in E separating the regions ri and rj .4. . E) embedded in the plane. Kuratowski’s Theorem: A graph is nonplanar if and only if it contains a subgraph that is homeomorphic to either K5 or K3. Proof: in K3. Graph Coloring. a coloring of a graph is an assignment of a color to each vertex of the graph. . each region is bounded by at least four edges. which is a contradiction. Then 3r ≤ 2e and e ≤ 3v − 6. e ≥ 3 edges and r regions. and prove that they are not isomorphic.3 is nonplanar. rk }. where V d = {p1 . A map is deﬁned by some planar graph G = (V. Then: v − e + r = 2. . . E d ). 4. This is equivalent to coloring the vertices of the dual graph of M in such a way that no adjacent vertices have the same color. On the other hand.7.5. 5. The graph K5 is nonplanar. 3.3 we have v = 6 and e = 9. PLANAR GRAPHS 98 1. and let v = V . Euler’s Formula: Let G = (V.3 were planar. 20 ≤ 18. In general. Warning: Note that a diﬀerent embedding of the same graph G may give diﬀerent (and nonisomorphic) dual graphs. Let G = (V. Proof: in K5 we have v = 5 and e = 10..3 .9.4. select a point pi in the interior of ri . r2 . . which contradicts the previous result. 2. p2 . Consider the problem of coloring a map M in such a way that no adjacent regions (sharing a border) have the same color.
7.4.17. Heawood) (not hard to prove): Every simple. proved with an intricate computer analysis of conﬁgurations: Every simple.4. colored with n colors we say that it is ncolorable. E) is called the chromatic number of G. planar graph is 5colorable. 1. Exercise: Find a planar graph G such that χ(G) = 4. 4. and is represented χ(G). PLANAR GRAPHS 99 a e c d b Figure 7. (b) G is bipartite. The following statements are equivalent: (a) χ(G) = 2.7. χ(Kn ) = n. Four Color Theorem (Appel and Haken. 1976). The minimum number of colors needed to properly color a given graph G = (V. (c) Every circuit in G has even length 3. Five Color Theorem (Kempe. Let G be a simple graph. planar graph is 4colorable.6. Some Results About Graph Coloring. Dual graph of a map. . 2. Obviously χ(G) ≤ V .
A rooted tree is a tree in which a particular vertex is designated as the root and every edge is directed away from the root. . . The arrows indicating the directions of the edges can be omitted. . vn−1 is the parent of vn . Let T be a tree with root v0 . y is a descendant of x. 4. The level of a vertex v is the length of the simple path from the root to v. Trees 8. Suppose that x.CHAPTER 8 Trees 8. vn−1 are ancestors of vn . vn is a child of vn−1 .1. r a d b g h m n o p c e f i j k l Figure 8. 5.1.1. . A rooted tree. The height of a rooted tree is the maximum level of its vertices. If x and y are children of z. If x is an ancestor of y. . vn ) is a simple path in T . 6. v1 . it is called a terminal vertex or leaf. If x has no children.1. A tree is a connected undirected graph with no simple circuits. v0 . 2. v1 . Then: 1. . 3. We draw rooted trees with the root at the top. . y and z are vertices in T and that (v0 . . 100 . Terminology. x and y are siblings.
8. . TREES 101 7. The subtree of T rooted at x is the graph (V. E). 8. If x is not a terminal vertex. where V is x together with its descendants and E = edges of simple paths from x to some vertex in E.1. it is an internal or branch vertex.
The tree is called a full mary tree if every internal vertex has exactly m children. each data item in the left subtree of v is less than the data item in v. Example: Figure 8. but not both. if it is equal we have found it.2 contains a binary search tree for the set S = {1. 7. Binary Trees 8. If a vertex has one child. 2.2. 9.3. 8. The following are a few results about binary trees: 1.2. An ordered rooted tree is a rooted tree where the children of each internal vertex are ordered.1. 6. We can store data in a binary search tree by randomly choosing data from S and placing it in the tree in the following way: The ﬁrst data chosen will be the root of the tree. If T is a full binary tree with i internal vertices. 2. Making a Binary Search Tree. A binary tree is a rooted tree in which each vertex has at most two children. starting at the root we . 8. 8. and each data item in the right subtree of v is greater than the data item in v.g. if it is smaller we continue through the left child. then t ≤ 2h . e. If a binary tree of height h has t terminal vertices. A binary tree is just a particular case of mary ordered tree (with m = 2). If the element is greater we continue through the right child. for each vertex v in T . Assume S is a set in which elements (which we will call “data”) are ordered. If we reach a terminal vertex without founding the element. Then for each subsequent data item. A binary search tree associated to S is a binary tree T in which data from S are associate with the vertices of T so that.2. then T has i+1 terminal vertices and 2i + 1 total vertices. 5.2. then that element is not present in S. designated as left child and right child. or strings of alphabetic characters in lexicographic order.2. Binary Search Trees. A full binary tree is a binary tree in which each vertex has exactly two children or none.2. 10}. 3. Binary Trees. More generally we can deﬁne a mary tree as a rooted tree in which every internal vertex has no more than m children. 4. BINARY TREES 102 8.. In order to ﬁnd a element we start at the root and compare it to the data in the current vertex (initially the root).8. the elements of S can be numbers in their natural order. that child is designated as either a left child or a right child.
the tree in ﬁgure 8.2. For instance. BINARY TREES 103 5 9 3 4 6 7 8 10 2 1 Figure 8.8. If the new data item is greater than the data in the current vertex then we move to the right child. Binary Search Tree. If there is no such child then we create one and put the new data in it. compare it to the data in the current vertex v. IN A I DO LA MANCHA NAME OF PLACE WHOSE WANT TO REMEMBER NOT Figure 8.2. if it is less we move to the left child.3.3 has been made from the following list of words choosing them in the order they occur: “IN A PLACE OF LA MANCHA WHOSE NAME I DO NOT WANT TO REMEMBER”. . Another binary Search Tree.
Then by comparing say C1 with any of the other coins. If their weights are the same then we know that the bad coin is C5 .H C3. C2 . Tree Isomorphisms 8.1. If C1 is heavier than C2 then we know that either C1 is the bad coin and is heavier. The corresponding decision tree is the following: C1:C2 left balanced C1:C5 left balanced left C3:C5 C3:C4 C1:C5 right C1. say C5 . The terminal vertices are labeled with the bad coin and whether it is heavier (H) or lighter (L). but one is either heavier or lighter that the others. Decision Trees.L balanced right right balanced C2. C5 that are identical in appearance.H C1.8. First we compare the weights of C1 and C2 .L left left balanced C3. The problem is to identify the bad coin and determine whether it is lighter or heavier using only a pan balance and comparing the weights of two piles of coins. In this puzzle we have ﬁve coins C1 .L C1:C5 C3:C5 right right balanced C5.H C4.3. and there is no way to solve the problem with less than that—with two weighings we can get . DECISION TREES. Decision Trees. The problem can be solved in the following way. The decision tree is optimal in the sense that in the worst case it uses three weighings.H C2.L C5. C3 . In each vertex “Ci : Cj ” means that we compare coins Ci and Cj by placing Ci on the left pan and Cj on the right pan of the balance. Example: The FiveCoins Puzzle. or C2 is the bad coin and it is lighter. and we can determine whether it is heavier or lighter by comparing it to say C1 . TREE ISOMORPHISMS 104 8.L Figure 8.H C4. and each edge is labeled depending on what side of the balance is heavier. If C1 is lighter than C2 we proceed as before with “heavier” and “lighter” reversed. we can determine whether the bad coin is C1 and is heavier (if C1 it is heavier than C5 ) or it is C2 and is lighter (if C1 has the same weight as C5 ).4. A decision tree is a tree in which each vertex represents a question and each descending edge from that vertex represents a possible answer to that question. If C1 and C2 have the same weight we can try comparing C3 and C4 in a similar manner. C4 .3.3. Decision tree for the 5 coins puzzle.
3. TREE ISOMORPHISMS 105 at most nine possible outcomes. a3 with the decision tree shown in ﬁgure 8. More generally in order to sort n elements we need a decision tree with n! outcomes.a3 YES NO a1<a3? NO YES NO a1<a3? NO . Sorting three elements. Assume that T1 is a tree with vertex set V1 and T2 is another tree with vertex set V2 . 2 1According . a2.a2. Sorting algorithms work by comparing elements and rearranging them as needed.1 we have h(n) = Ω(n log2 n).5 is optimal in the sense that it uses the minimum possible number of comparisons in the worstcase.5.a2 Figure 8. So the worse case complexity of a sorting algorithm is Ω(n log2 n). we know that it is optimal. Recall that in a binary tree of height h with t terminal vertices the following inequality holds: t ≤ 2h . 8.a2.8.a1 a1. For instance we can sort three elements a1 .5 a1<a2? YES a2<a3? YES a1.a3. so taking logarithms log2 n! ≈ n log2 n − n log2 e + 1 log2 (2πn) = Θ(n log2 n). We will study three diﬀerent kinds of treeisomorphisms between T1 and T2 . DECISION TREES.a2 a3.a1. Isomorphisms of Trees.a1 a3. Since log2 (n!) = Θ(n log2 n). n! ≈ nn e−n 2πn.2. Since there are 3! = 6 possible arrangements of 3 elements. 8.a3 a2<a3? YES NO a2.3.a1. a2 . which are insuﬃcient to distinguish among ten combinations of 5 possible bad coins and the bad coin being heavier or lighter.3. Since the mergesort algorithm uses precisely Θ(n log2 n) comparisons. Complexity of Sorting. so the algorithm represented by the decision tree in ﬁgure 8. which implies h ≥ 3.3. so its height h(n) will verify n! ≤ 2h(n) . √ to Stirling’s formula. If they are rooted trees then we call their roots r1 and r2 respectively. Hence in our case 6 < 2h . we need a decision tree with at least 6 possible outcomes or terminal vertices.a3.
2. Binarytreeisomorphism: Two binary trees T1 and T2 are isomorphic if there is a bijection f : V1 → V2 that preserves adjacency.: (a) f (v) is adjacent to f (w) if and only if v is adjacent to w. and left/right children. and the root vertex. T1 T2 T3 Figure 8. i.7 shows all 5 nonisomorphic 3vertex binary trees. (b) f (r1 ) = r2 .8. Exercise: Find all nonisomorphic 3vertex free trees. Example: Figure 8. so there are 2 nonisomorphic 3vertex rooted trees represented for instance by T1 and T2 . f (v) is adjacent to f (w) if and only if v is adjacent to w.e. Trees with diﬀerent kinds of isomorphisms. TREE ISOMORPHISMS 106 1. 3. DECISION TREES. while the roots of T2 and T3 have degree 2. Answer : Figure 8.3. but T1 is not isomorphic to the others. 3vertex rooted trees and 3vertex binary trees. Rootedtreeisomorphism: T1 and T2 are isomorphic if there is a bijection f : V1 → V2 that preserves adjacency and the root vertex. i.. i. As rooted trees T2 –T5 are isomorphic. (d) f (v) is a right child of f (w) if and only if v is a right child of w. All of them .: (a) f (v) is adjacent to f (w) if and only if v is adjacent to w.e. but they are not isomorphic to T1 because the root of T1 has degree 3. (b) f (r1 ) = r2 . (c) f (v) is a left child of f (w) if and only if v is a left child of w. On the other hand as rooted trees T2 and T3 are isomorphic.e. Finally T2 and T3 are not isomorphic as binary trees because the left child of the root in T2 is a terminal vertex while the left child of the root of T3 has two children.6 shows three trees which are graphisomorphic. Usual graphisomorphism between trees: T1 and T2 are isomorphic if there is a bijection f : V1 → V2 that preserves adjacency.6.
f2 . c = 01.05n·3 = 1. b = 01. . . DECISION TREES. k=1 given an alphabet and the frequencies of its characters. Then the following enconding would be more eﬃcient than the ﬁx length encoding: a = 1. ﬁnd an optimal encoding that provides minimum average length for words. b = 10. instead some are more frequent that others. Two bits are enough for representing them. Fix length and Huﬀman codes can be represented by trees like in ﬁgure 8. T1 T2 T3 T4 T5 Figure 8. Nonisomorphic binary trees.15n c’s and 0. 0.4. . Now in average an ncharacter word will have 0.15n·3+0. TREE ISOMORPHISMS 107 are isomorphic as free trees. The code of each symbol consists of the sequence of labels of the edges in the path from the root to the leaf with the desired symbol. . d}.05n d’s.7.5n a’s.8. b 30%.7n. so there is only 1 nonisomorphic 3vertex free tree. fn is n 1 fk l(ak ) .5n·1+0. b. . However assume that they do not appear with the same frequency.3n b’s.3n·2+0.3. hence its length will be 0.8. Huﬀman Codes. . The problem now is.3. F k=1 where l(ak ) = length of ak and F = n fk . for instance a = 11. d = 000. . d = 00 would be one such representation. say a appears with a frequency of 50%. Usually characters are represented in a computer with ﬁx length bit strings. 8. c = 001. an whose frequencies are f1 . As an example assume that we have an alphabet with four symbols: A = {a. . Huﬀman codes provide an alternative representation with variable length bit strings. so that shorter strings are used for the most frequently used characters. 0. c 15% and d 5%. . a2 . In general the length per character of a given encoding with characters a1 . c. which is shorter than 2n. With this encoding ncharacter words will have 2n bits.
we can choose to add the ﬁrst 12 and 8. and join the corresponding two symbols to a common vertex above them by two edges.3. 8. 12 → 12. 7.3. 8. Then we replace the two smallest frequencies f1 and f2 with their sum f1 + f2 .8.5. First we list the frequencies of all the codes and represent the symbols as vertices (which at the end will be leaves of a tree). Fix length code and Huﬀman code. An optimal Huﬀman code is a Huﬀman code in which the average length of the symbols is minimum. 8. DECISION TREES. In general an optimal Huﬀman code can be made as follows. one labeled 0 and the other one labeled 1. Then we repeat the same operation with the resulting shorter list of frequencies until the list is reduced to one element and the graph obtained becomes a tree. Example: Find the optimal Huﬀman code for the following table of symbols: character frequency a 2 b 3 c 7 d 8 e 12 Answer : : The successive reductions of the list of frequencies are as follows: 2. TREE ISOMORPHISMS 108 1 0 1 a 0 1 a 0 1 c 0 d 1 b 1 c 0 0 d b Fix length code Huffman code Figure 8. Constructing an Optimal Huﬀman Code. or 8 and the second 12. 8. 7 . 12 → 5. 3 . Than common vertex plays the role of a new symbol with a frequency equal to f1 + f2 . Let’s choose the former: .8. 12 5 12 Here we have a choice.
12 → 32 20 32 The tree obtained is the following: 32 1 20 1 12 5 1 2 a 1 0 3 b 0 c 7 0 d 8 e 0 12 Figure 8.8. Optimal Huﬀman code 2.10. TREE ISOMORPHISMS 109 12. 12 → 20. 12 → 32 20 32 32 1 12 1 5 1 2 a 0 3 b c 0 7 8 d e 1 0 20 0 12 Figure 8. 12 → 20. The resulting code is as follows: character code a 1111 b 1110 c 110 d 10 e 0 The other choice yields the following: 12. Optimal Huﬀman code 1. 8.3. . 8. DECISION TREES.9.
TREE ISOMORPHISMS 110 character code a 111 b 110 c 10 d 01 e 00 8. The player that removes the last coin loses.3. we label the vertex depending on whether it is a “ﬁrst player” position (box) or a “second player” position (circle).6. The following tree represents all possible sequences of choices. Taking turns two players remove any number of coins from one of the piles. Tree of a game. First player positions are labeled with the maximum value of the labels of its children. This numbers represent the “value” of each position of the game. Trees are used in the analysis of some games.11. Each node shows the number of coins in each pile. DECISION TREES. Game Trees.8. After all the children of a given vertex have been labeled. Then we continue labeling the rest of the vertices in the following way. The analysis of the game starts by labeling each terminal vertex with “1” if it represents a victory for the ﬁrst player and “0” if it represents a victory for the second player. As an example we study the following game using a tree: Initially there are two piles with 3 coins and 1 coin respectively.3. second player positions are labeled with the minimum . 3 1 1 2 0 1 1 1 1 1 0 1 0 1 1 0 0 0 0 1 1 0 2 0 1 1 0 0 1 1 1 0 0 1 0 1 1 0 0 1 2 1 0 1 0 1 0 0 0 0 0 1 3 0 0 1 0 0 0 0 0 0 0 1 0 0 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 1 Figure 8. and each edge represents a possible “move” (choice) from one of the players. The ﬁrst player is represented with a box and the second player is represented with an circle. so that the ﬁrst player is interested in making it “maximum” and the second player wants to make it “minimum”.
If the starting position has been labeled “1” that means that the ﬁrst player has a winning strategy. so it will not aﬀect the value of v and we can stop the process of evaluation of w (prone the subtree at w). If at any point a children of w gets a value less than or equal to the alpha value of v then it will become obvious that the value of w is going to be less than the current alpha value of v. we assign a beta value equal to the minimum value of its children evaluated so far. Alphabeta pruning. In order to do that with a ﬁrst player (boxed) vertex v. we prone the subtree at w. This process is called the minimax procedure. For instance in the present game the ﬁrst player has advantage at the initial position. The idea is to skip a vertex as soon as it becomes obvious that its value will not aﬀect the value of its parent.e. Now the strategy is for the ﬁrst player to select at each position a children with maximum value. so a position with one queen and no knights will have a higher value than one with no queen and one knight).. Even so the evaluation and labeling of the vertices can be time consuming. so it is built up to a given depth only. If for any reason the ﬁrst player makes a mistake and removes say one coin from the ﬁrst pile. at a second player (circled) vertex v. but we can bypass the evaluation of many vertices using the technique of alphabeta pruning. which will be a second player (circled) position.8. Assume that we are evaluating one of its children w. i. TREE ISOMORPHISMS 111 value of the labels of its children. otherwise the second player has advantage. we assign it an alpha value equal to the maximum value of its children evaluated so far. then the 1 second player has one favorable move to vertex 0 . Every vertex labeled “1” will represent a position in which the ﬁrst player has advantage and can win if he/she works without making mistakes. Similarly. That is called an alpha cutoﬀ. he/she must remove all 3 coins from 1 1 the ﬁrst pile.. . i. DECISION TREES. and practice a beta cutoﬀ when one of its grandchildren gets a value greater than or equal to the current beta value of v. and only one favorable movement at that point: 3 → 0 . The vertices reached at that depth are not terminal.e.3. while the second player will be interested in selecting children with minimum value. on the other hand. but they can be “evaluated” using heuristic methods (for instance in chess usually losing a knight is a better choice than losing the queen. which is the one 1 with minimum “value”. In some games the game tree is so complicated that it cannot be fully analyzed. going to position 2 . vertices labeled “0” represent positions for which the second player has advantage. where w is the parent of that grandchildren.
8. .12. TREE ISOMORPHISMS (4) v 112 1 3 4 w 4 3 6 7 5 4 5 1 ? Figure 8. DECISION TREES. Alpha cutoﬀ.3.
4.1. it is customary to represent the result of applying the operation to two elements a. The tree for a constant or variable a has a as its only vertex. we scan the expression from right to left. and insertions take place only at the other end. For instance. Polish notation is easier to evaluate in a computer. In order to evaluate an expression in Polish notation. The reverse Polish notation consists of placing the symbol to the right: ab ◦ . If the algebraic expression S is of stack or lastin ﬁrstout (LIFO) system. called front of the list. In order to motivate this subject. placing the elements in a stack. is a linear list of elements in which deletions take place only at one end. the expression a ∗ (b + c) in Polish notation would be ∗ a + b c. Tree Transversal 8.1 Each time we ﬁnd an operator.8. TREE TRANSVERSAL 113 8. A queue or ﬁrstin ﬁrstout (FIFO) system. called top of the list. Given a (not necessarily commutative) binary operation ◦. while a ∗ b + c is + ∗ a b c.4. called rear of the list. Also. For instance. is a linear list of elements in which insertions and deletions take place only at one end. This is called inﬁx notation. The advantage of Polish notation is that it allows us to write expressions without need for parenthesis. the expression ∗ + 2 3 4 (which in inﬁx notation is “(2 + 3) ∗ 4”) would be evaluated like this: expression stack ∗ + 234 ∗ + 23 ∗+2 ∗+ ∗ 4 3 4 2 3 4 5 4 20 An algebraic expression can be represented by a binary rooted tree obtained recursively in the following way.4. we introduce the concept of Polish notation. Transversal Algorithms. b by placing the operation symbol in the middle: a ◦ b. we replace the two top symbols of the stack by the result of applying the operator to those elements. 1A . The Polish notation consists of placing the symbol to the left: ◦ab.
namely in preorder. Pre(Tk ). where SL and SR are subexpressions with trees TL and TR respectively. postorder and inorder transversal of T . . Then: 1.4. . then r by itself constitutes the preorder. o TL TR Figure 8. where + denotes addition. then the tree T for S consists of ◦ as root. The binary tree for this expression is given in ﬁgure 8. Pre(T1 ). postorder and inorder respectively. .14.15). . and the subtrees TL and TR (ﬁg. The following are recursive deﬁnitions of several orderings of the vertices of a rooted tree T = (V. Tree of S1 ◦ S2 . and ◦ is an operator. Preorder Transversal : Pre(T ) = r. TREE TRANSVERSAL 114 the form SL ◦ SR . . If T has only one vertex r. 8. + + a b * c d e f * + h Figure 8. reverse Polish and inﬁx representation are diﬀerent ways of ordering the vertices of the tree. 8. . E) with root r. Tk the subtrees of T from left to right (ﬁg. Tree for a + b ∗ c + d ↑ e ∗ (f + h).8. . For instance. . consider the following algebraic expression: a + b ∗ c + d ↑ e ∗ (f + h) . Given the binary tree of an algebraic expression. .13.13). its Polish. ∗ denotes multiplication and ↑ denotes exponentiation.14. Otherwise. let T1 .
.. Postorder Transversal : Post(T ) = Post(T1 ). left subtree TL and right subtree TR . r. 3. In(TR ).. Inorder Transversal. Post(Tk ). If T is a binary tree with root r. 2. . . Ordering of trees. Tk Figure 8. TREE TRANSVERSAL 115 r T1 T2 . .15.4.8. r. then: In(T ) = In(TL ). .
we give two algorithms to ﬁnd the spanning tree T of a loopfree connected undirected graph G = (V.16 has a spanning tree represented by the thicker edges. 3. . If the queue is empty. insert it in the queue and mark it as “visited”. In practice. The idea is to start with vertex v1 as root. then the ones that are adjacent to the latter and have not been visited yet. . . 4. Delete v from the queue. Next. however. 2. SPANNING TREES 116 8. then we are done. A tree T is a spanning tree of a graph G if T is a subgraph of G that contains all the vertices of G. For each vertex v of G that has not been visited yet and is adjacent to v (there might be none) taken in order of increasing subscripts.1. c b h d g a f e i Figure 8. Add v1 to T .16. We assume that the vertices of G are given in a certain order v1 . This algorithm uses a queue (initially empty) to store vertices of the graph.8. v ) to T .5. Spanning Trees. removing edges is not eﬃcient because ﬁnding cycles is time consuming.5. E ). add vertex v and edge (v. vn . add the vertices that are adjacent to v1 . E). . Spanning tree.2. insert v in the queue and mark it as “visited”. Spanning Trees 8. BreadthFirst Search Algorithm. The resulting spanning tree will be T = (V . 8. Every connected graph has a spanning tree which can be obtained by removing edges until the resulting graph becomes acyclic.5. . and so on. Otherwise let v be the vertex in the front of the queue. v2 . In consists of the following: 1.5. For instance the graph of ﬁgure 8.
b. f. and then go back (backtrack) to add branches also as long as possible.3. DepthFirst Search Algorithm. c b h d g a f i e Figure 8. in order. Add v1 to T . i. BreadthFirst Search.5. . SPANNING TREES 117 5. e.V’ 9: if (x. In consists of the following: 1. A pseudocode version of the algorithm is as follows: 1: procedure bfs(V. d.y) to E’ and vertex y to V’ 11: if no edges were added then 12: return T 13: S := children of S 14: end 15: end bfs Figure 8. h. insert it in the stack and mark it as “visited”.17 shows the spanning tree obtained using the breadthﬁrst search algorithm on the graph with its vertices ordered lexicographically: a.y) is an edge then 10: add edge (x. Go to step 2. c. 8.5.8. g. This algorithm uses a stack (initially empty) to store vertices of the graph.17. 8: for each y in V .E) 2: S := (v1) {ordered list of vertices of a fix level} 3: V’ := {v1} {v1 is the root of the spanning tree} 4: E’ := {} {no edges in the spanning tree yet} 5: while true 6: begin 7: for each x in S. The idea of this algorithm is to make a path as long as possible.
then we are done. Otherwise let v be the vertex on the top of the stack. Add the edge (v. 2.v) to E’ 13: add v to V’ 14: w := v 15: end 16: if w = v1 then . 4. Otherwise. Go to step 2. Apply P to v .v) 11: does not create a cycle in T 12: add (w. The DepthFirst Search Algorithm consists of applying the process just deﬁned to v1 . If there is no vertex v that is adjacent to v and has not been visited yet. An alternative recursive deﬁnition is as follows. Add vertex v and edge (v.v) that when added 8: to T does not create a cycle in T 9: begin 10: Choose first v such that (w. let v be the ﬁrst nonvisited vertex that is adjacent to v. If there is no vertex v that is adjacent to v and has not been visited yet. 4. 5. A pseudocode version of the algorithm is as follows: 1: procedure dfs(V.E) 2: V’ := {v1} {v1 is the root of the spanning tree} 3: E’ := {} {no edges in the spanning tree yet} 4: w := v1 5: while true 6: begin 7: while there is an edge (w. 3. We deﬁne recursively a process P applied to a given vertex v in the following way: 1. If the stack is empty. 3. let v be the ﬁrst nonvisited vertex that is adjacent to v. SPANNING TREES 118 2. Add vertex v to T and mark it as “visited”. then return. v ) to T . v ) to T . insert v in the stack and mark it as “visited”.8. then delete v and go to step 2 (backtrack ). 5.5. Go to step 2 (backtrack ). Otherwise.
the spanning tree shown is the one of minimum weight.19. f. c. a 2 3 c 6 e 2 1 4 b 5 d 6 f Figure 8. 8. . For instance for the graph of ﬁgure 8. its minimal spanning tree is a spanning tree of G such that the sum of the weights of its edges is minimum. c b h d g a f i e Figure 8.19.5. d. i. SPANNING TREES 119 17: return T 18: w := parent of w in T {backtrack} 19: end 20: end Figure 8. An algorithm to ﬁnd a minimal spanning tree is Prim’s Algorithm. h. Minimal Spanning Trees. b.18 shows the spanning tree obtained using the breadthﬁrst search algorithm on the graph with its vertices ordered lexicographically: a. g. e. DepthFirst Search. Minimum Spanning Tree.5. It starts with a single vertex and at each iteration adds to the current tree a minimum weight edge that does not complete a cycle.4. Given a connected weighted tree G. Prim’s Algorithm.8.18.
8.5. SPANNING TREES
120
The following is a pseudocode version of Prim’s algorithm. If (x, y) is an edge in G = (V, E) then w(x, y) is its weight, otherwise w(x, y) = ∞. The starting vertex is s. 1: procedure prim(V,w,s) 2: V’ := {s} {vertex set starts with s} 3: E’ = {} {edge set initially empty} 4: for i := 1 to n1 {put n edges in spanning tree} 5: begin 6: find x in V’ and y in V  V’ with minimum w(x,y) 7: add y to V’ 8: add (x,y) to E’ 9: end 10: return E’ 11: end prim Prim’s algorithm is an example of a greedy algorithm. A greedy algorithm is an algorithm that optimized the choice at each iteration without regard to previous choices (“doing the best locally”). Prim’s algorithm makes a minimum spanning tree, but in general a greedy algorithm does not always ﬁnds an optimal solution to a given problem. For instance in ﬁgure 8.20 a greedy algorithm to ﬁnd the shortest path from a to z, working by adding the shortest available edge to the most recently added vertex, would return acz, which is not the shortest path.
b 2 a 1 100 c 101 z 4
Figure 8.20 Kruskal’s Algorithm. Another algorithm to ﬁnd a minimal spanning tree in a connected weighted tree G = (V, E) is Kruskal’s Algorithm. It starts with all n vertices of G and no edges. At each iteration we add an edge having minimum weight that does not complete a cycle. We stop after adding n − 1 edges.
8.5. SPANNING TREES
121
1: procedure kruskal(E,w,n) 2: V’ := V 3: E’ := {} 4: while E’ < n1 5: begin 6: among all edges not completing a cycle in T 7: choose e of minimum weight and add it to E 8: end 9: T’ = (V’,E’) 10: return T’ 11: end kruskal
CHAPTER 9
Boolean Algebras
9.1. Combinatorial Circuits 9.1.1. Introduction. At their lowest level digital computers handle only binary signals, represented with the symbols 0 and 1. The most elementary circuits that combine those signals are called gates. Figure 9.1 shows three gates: OR, AND and NOT.
OR GATE
x1 x2
x1 + x2
AND GATE
x1 x2
x1
x2
NOT GATE
x
x
Figure 9.1. Gates.
Their outputs can be expressed as a function of their inputs by the following logic tables: x1 x2 x1 + x2 1 1 1 1 0 1 0 1 1 0 0 0 OR GATE
122
COMBINATORIAL CIRCUITS 123 x1 x2 x1 · x2 1 1 1 1 0 0 0 1 0 0 0 0 AND GATE x 1 0 x 0 1 NOT GATE These are examples of combinatorial circuits. x1 x2 x3 y Figure 9. and . previous inputs do not aﬀect their outputs. If we input a signal x2 = 1. Assume that at a given time y = 0. If x1 = 1 and x2 = 0 then y can be 0 or 1. Some combinations of gates can be used to make more complicated combinatorial circuits. A combinatorial circuit is a circuit whose output is uniquely deﬁned by its inputs.3 is not a combinatorial circuit. A combinatorial circuit.2.9. representing the values of the Boolean expression y = (x1 + x2 ) · x3 . the output becomes y = 1. For instance ﬁgure 9.2 is combinatorial circuit with the logic table shown below.1. They do not have memory. x1 x2 x3 y = (x1 + x2 ) · x3 1 1 1 0 1 1 0 1 1 0 1 0 1 0 0 1 0 1 1 0 0 1 0 1 0 0 1 1 0 0 0 1 However the circuit in ﬁgure 9.
Commutative a+b=b+a a·b=b·a a · (b + c) = (a · b) + (a · c) a + (b · c) = (a + b) · (a + c) 4. each represents either 0 or 1): 1. We can “delete” it by switching input x1 to 0.. + and respectively.3. Two Boolean expressions are deﬁned to be equal is they have the same values for all possible assignments of values to their literals. b. x1 y x2 Figure 9. Properties of Combinatorial Circuits. Here Z2 = {0. Example: x + y = x · y. Associative (a + b) + c = a + (b + c) (a · b) · c = a · (b · c) 2.1. and the operations performed on those signals by AND.9. i. Identity a+0=a a·1=a 5. as shown in the following table: 3.e. OR and NOT gates are represented by the symbols ·.1. Not a combinatorial circuit. Distributive . 9. That way we can store a bit. Complement a+a=1 a·a=0 A system satisfying those properties is called a Boolean algebra. 1} represents the set of signals handled by combinatorial circuits. Then their properties are the following (a. COMBINATORIAL CIRCUITS 124 stays so even after x2 goes back to its original value 0.2. c are elements of Z2 .
Abstract Boolean Algebras. . Distributive x ∧ (y ∨ z) = (x ∧ y) ∨ (x ∧ z) x ∨ (y ∧ z) = (x ∨ y) ∧ (x ∨ z) 4. . 0.3. 0. Commutative x∨y =y∨x x∧y =y∧x 3. ∩. COMBINATORIAL CIRCUITS 125 x 1 1 0 0 y x+y x·y 1 0 0 0 0 0 1 0 0 0 1 1 9. two binary operators ∨ and ∧ on S.9. z are elements of S): 1. A Boolean algebra B = (S. ∨. Complement x∨x=1 x∧x=0 Example: (Z2 . Here we deal with general Boolean algebras. combinatorial circuits are an example. . U ). ∧. 1) is a Boolean algebra. ∪. . +. y. Identity x∨0=x x∧1=x 5. Example: If U is a universal set and P(U )= the power set of S (collection of subsets of S) then (P(U ). satisfying the following properties (x. but there are others. Associative (x ∨ y) ∨ z = x ∨ (y ∨ z) (x ∧ y) ∨ z = x ∧ (y ∧ z) 2. ·. is a Boolean algebra. ∅.1. and a unary operator on S.1. 1) is a set S containing two distinguished elements 0 and 1.
Involution x=x 5. Bound x∨1=1 x∧0=0 x ∨ xy = x x ∧ (x ∨ y) = x 4.4. 0 and 1 0=1 1=0 6. The properties mentioned above deﬁne a Boolean algebra. Idempotent x∨x=x x∧x=x 2. COMBINATORIAL CIRCUITS 126 9. 3. Other Properties of Boolean Algebras.1.9. Absorption . De Morgan’s x∨y =x∧y x∧y =x∨y For instance the ﬁrst idempotent law can be proved like this: x = x ∨ 0 = x ∨ x ∧ x = (x ∨ x) ∧ (x ∨ x) = (x ∨ x) ∧ 1 = x ∨ x.1. but Boolean algebras also have other properties: 1.
Disjunctive Normal Form. That representation is called disjunctive normal form of the Boolean function f . an ) = 1.9. Applications 9. We start with a deﬁnition. Given any Boolean function f : Zn → Z2 that is not identically 2 zero. . BOOLEAN FUNCTIONS. . xn ) = m1 + m2 + · · · + mk . . . deﬁned by the following table: x1 x2 x1 ⊕ x2 1 1 0 1 0 1 0 1 1 0 0 0 The exclusiveor function can interpreted as a function Z2 → Z2 2 that assigns (1. .2. A minterm in the symbols x1 . xn is a Boolean expression of the form y1 · y2 · · · · · yn . 9. This provides a way to implement the exclusiveor with a combinatorial circuit as shown in ﬁgure 9.2.2. .4. x2 . Boolean Functions.1. Conjunctive Normal Form. . Example: We have seen that the exclusiveor can be represented x1 ⊕ x2 = (x1 · x2 ) + (x1 · x2 ). 1) → 1.2. 1) → 0. (0. A Boolean function is a function from Zn 2 to Z2 . .3. 9. . A maxterm in the symbols x1 . . xn is a Boolean expression of the form y1 + y2 + · · · + yn . It can also be written as a Boolean expression in the following way: x1 ⊕ x2 = (x1 · x2 ) + (x1 · x2 ) Every Boolean function can be written as a Boolean expression as we are going to see next.2.2. consider the exclusiveor function. 0) → 1. it can be represented f (x1 . Introduction. APPLICATIONS 127 9. . . For instance. (1. where yj = xj if aj = 1 and yj = xj if aj = 0. . . where each yi is either xi or xi . . . x2 . . where m1 . (0. a2 . where each yi is either xi or xi . . . mk are all the minterms mi = y1 ·y2 ·· · ··yn such that f (a1 . . 0) → 0. . m2 .
. . . . . In fact: x+y = x·y. xn ) = M1 · M2 · · · · · Mk .5 computes x + y.4. NOT} is functionally complete. ExclusiveOr. The set of gates {AND. . Example: The result about the existence of a disjunctive normal form for any Boolean function proves that the set of gates {AND. . . . APPLICATIONS x1 x1 x2 x2 128 Figure 9. it can be represented f (x1 .2. OR. BOOLEAN FUNCTIONS. g2 . . where yj = xj if aj = 0 and yj = xj if aj = 1. We have seen how to design combinatorial circuits using AND. NOT} is functionally complete. an ) = 0. . . . .9. Next we show other sets of gates that are also functionally complete. . OR. .4. all we need to do is to show that we can compute x + y using only AND and NOT gates. A set of gates {g1 . OR and NOT gates. M2 .2. Mk are all the maxterms Mi = y1 + y2 + · · · + yn such that f (a1 . Functionally Complete Sets of Gates. g2 . where M1 . Given any Boolean function f : Zn → Z2 that is not identically 2 one. . a2 . 9. . Example: The conjunctive normal form of the exclusiveor is x1 ⊕ x2 = (x1 + x2 ) · (x1 + x2 ) . . gk } is said to be functionally complete if for any integer n and any function f : Zn → Z2 it is possible to 2 construct a combinatorial circuit that computes f using only the gates g1 . . hence the combinatorial circuit of ﬁgure 9. That representation is called conjunctive normal form of the Boolean function f . In the following gates will be considered as functions from Zn into Z2 intended 2 to serve as building blocks of arbitrary boolean functions. Here we will see how to do the same with other kinds of gates. Proof: Since we already know that {AND. NOT} is functionally complete. gk . . 1.
Proof: Note that x ↑ y = x · y.2. 2. 3. Also the OR gate can be implemented with NAND gates: x+y = x · y = (x ↑ x) ↑ (y ↑ y). The gate NAND.9.2. Here we address the problems of ﬁnding a combinatorial circuit that computes a given Boolean function with the minimum number of gates. hence the combinatorial circuit of ﬁgure 9. APPLICATIONS x1 x2 129 x1 + x2 Figure 9. Since the set {OR. 9.5. denoted ↑ and deﬁned as x1 ↑ x2 = 0 if x1 = 1 and x2 = 1 1 otherwise. is functionally complete. x1 x1 x2 Figure 9. AND with OR and NOT. The set of gates {OR. The idea x2 . so the NOT gate can be implemented with a NAND gate.7. Minimization of Combinatorial Circuits.6 computes x + y. x1 x1 x2 x2 Figure 9. NAND gate. the NAND gate is functionally complete.6. Hence x = x · x = x ↑ x.5. NOT} is functionally complete and each of its elements can be implemented with NAND gates. BOOLEAN FUNCTIONS. The proof is similar: x·y = x+y. OR with AND and NOT. NOT} is functionally complete.
where E is any Boolean expression. z) the Boolean function deﬁned by the following table: x 1 1 1 1 0 0 0 0 y 1 1 0 0 1 1 0 0 z f(x. This function can be implemented with the combinatorial circuit of ﬁgure 9.z) 1 1 0 1 1 0 0 1 1 0 0 0 1 0 0 0 Its disjunctive normal form is f (x. y.2. For simplicity in the following we will represent a · b as ab. Example: Let F (x. NOT and OR functions implemented with NAND gates. . BOOLEAN FUNCTIONS.8.y.9. APPLICATIONS x x 130 x x y y Figure 9. is to simplify the corresponding Boolean expression by using algebraic properties such as (E · a) + (E · a) = E and E + (E · a) = E. so for instance the expressions above will look like this: Ea + Ea = E and E + Ea = E. z) = xyz + xyz + xyz.9. y.
2. z) = xyz + xyz + xyz. Figure 9. y. A circuit that computes f (x. z) = xyz + xyz + xyz. z) = xyz + xyz +xyz = xy + xyz = x(y + yz) = x(y + y)(y + z) = x(y + z) . z Figure 9. y.6.2. which corresponds to the circuit of ﬁgure 9.9. BOOLEAN FUNCTIONS.y. MultiOutput Combinatorial Circuits. that computes 9.9. APPLICATIONS x y 131 f(x. Its table is as follows: x 1 1 0 0 y 1 0 1 0 s 0 1 1 0 c 1 0 0 0 So the sum is s = x ⊕ y (exclusiveor) and the carry bit is c = x · y.10. But we can do better if we simplify the expression in the following way: xy f (x.11 shows a halfadder circuit. Example: HalfAdder.z) . where s represents the sum of x and y and c is the carry bit. A halfadder is a combinatorial circuit with two inputs x and y and two outputs s and c. y. x y z f(x.y. A simpler circuit f (x.10. . Figure 9.z) .
11.2.9. . APPLICATIONS 132 x s y c Figure 9. BOOLEAN FUNCTIONS. Halfadder circuit.
The internal states are S = {σ0 . 4. where σ0 is the initial state—we interpret it as not having seeing any “b” yet. Finite State Machines 10.1. The nextstate and output functions are as follows: f S σ0 σ1 I a b g a b σ0 σ1 1 0 σ1 σ1 0 0 133 . A ﬁnite set of output symbols O.1. I. Combinatorial circuits have no memory or internal states. so their output may depend not only on its current inputs but also on the past history of those inputs. An initial state σ ∈ S.1. FiniteState Machines. O. b} and two output symbols O = {0. A ﬁnite set of states S. 6. We represent the machine M = (S. σ) Example: We describe a ﬁnite state machine with two input symbols I = {a. 5. 3.CHAPTER 10 Automata. their output depends only on the current values of their inputs. A ﬁnitestate machine consists of the following: 1. σ1 }. 1} that accepts any string from I∗ and outputs as many 1’s as a’s there are at the beginning of the string. then the machine will switch to σ1 as soon as the ﬁrst “b” arrives. A nextstate or transition function f : S × I → S. Finite state machines on the other hand have internal states. then it outputs only 0’s. Grammars and Languages 10. A ﬁnite set of input symbols I. f. An output function g : S × I → O. 2. g.
2.1. ﬁnitestate automaton consists of: 1. More speciﬁcally. 10. which stands for “no carry” and “carry” respectively. So the input set is I = {00. A ﬁnite set of states S. If in state σ an input i causes the machine to output o and go to state σ then we draw an arrow from σ to σ labeled i/o or i. FINITE STATE MACHINES 134 This ﬁnitestate machine also can be represented with the following transition diagram: a/1 a/0 start Ö @ABC G GFED σ0 b/0 Ö @ABC G GFED σ1 g b/0 The vertices of the diagram are the states. A serial adder accepts two bits and outputs its sum. Example: The following example is similar to the previous one but the machine outputs 1 only after a change of input symbol. o. The transition diagram is the following: 00/0 01/1 01/0 10/0 start 10/1 Ö Ö 11/0 D GFED r HIJK @ABC G ONML l NC C v 00/1 11/1 10. 3. 11}. otherwise it outputs 0: @ABC G GFED σ0 BB BB b/0  a/0  BB  BB  B2  b/1  ~ D GFED k @ABC @ABC Q GFED m σ1 σ2 start a/0 a/1 b/0 Example: A SerialAdder. C}.1. and with a set of states called accepting or ﬁnal states. A ﬁnite set of input symbols I. A ﬁnitestate automaton is similar to a ﬁnitestate machine but with no output. FiniteState Automata. A nextstate or transition function f : S × I → S. .10. 2. The set of states is S = {N C. 01. The output set is O = {0. 1}.
?>=< σ1 b a Ö @ABC 89:.10. For instance the following automaton also accepts precisely strings of a’s abd b’s that end with an a. We say that an automaton accepts or recognizes a given string of input symbols if that strings takes the automaton from the start state to a ﬁnal state. The ﬁrst diagram uses the same scheme as with ﬁnitestate machines. ?>=< G GFED σ2 Example: The following automaton accepts strings of a’s and b’s with exactly an even number of a’s: . I. FINITE STATE MACHINES 135 4. so it is equivalent to the automaton shown above: b a a b start Ö @ABC G GFED l σ0 D GFED @ABC 89:. f. F). An initial state σ ∈ S.1. σ. We represent the automaton A = (S. Example: The following transition diagrams represent an automaton accepting any string of a’s and b’s ending with an a. ?>=< σ1 Two ﬁnitestate automata that accept exactly the same set of strings are said to be equivalent. and “0” representing “not accept”: b/0 a/1 start Ö @ABC G GFED l σ0 a/1 b/0 Ö D GFED @ABC σ1 The second kind of diagram omits the outputs and represents the accepting states with double circles: b a a b start Ö @ABC G GFED l σ0 Ö D GFED @ABC 89:. A subset F ⊆ S of accepting or ﬁnal states. with 1 representing “accept” or “recognize”. 5.
?>=< G GFED k E Ö C GFED @ABC O start @ABC G GFED σ0 a Ö @ABC 89:. ?>=< σ3 . ?>=< G GFED σ1 b Ö @ABC G GFED σ2 g h b Example: The following automaton accepts strings ending with aba: @ABC G GFED σ0 x @ABC G GFED σ1 p b start a b b a @ABC G GFED l σ2 a a b D GFED @ABC 89:.10.1. FINITE STATE MACHINES b b a a 136 start Example: The following automaton accepts strings starting with one a followed by any number of b’s: b a a Ö @ABC 89:.
: “(x + y”. )}. etc. A term may consists of a factor or a product of a term and a factor T →F T →T ∗F A factor may consists of an algebraic expression between parenthesis . “y + (x ∗ y + y) ∗ x”. Formal Languages. So syntactically correct algebraic expressions are a subset of the whole set A∗ of possible strings over A. “(x + y) ∗ x + z”. or terms (T ) consisting of product of factors (F ). In general. Then we can say that an algebraic expression E consists of a single term E → T. The following are some of them: “x + y ∗ y”.10. we are interested only in languages with certain structure. There are however some strings of symbols that are not legitimate algebraic expressions. ∗. used to represent given subsets of the language. the ﬁnal symbols are the elements of the set A = {x.2. A way to determine the structure of a language is with a grammar. The nonterminal symbols can be chosen to represent a complete algebraic expression (E). etc. 10. ∗. the ﬁnal symbols that occur in the strings of the language. “z + +y ∗ x”. a (formal) language over A is a subset of A∗ (set of strings of A). e. +. The set of strings over A with an even number of a’s is a language over A.1. or the sum of an algebraic expression and a term E → E + T. )}. In order to deﬁne a grammar we need two kinds of symbols: nonterminal. because they have some sort of syntax error. (. Although in principle any subset of A∗ is a formal language.g. y. “x(∗y) + z”. LANGUAGES AND GRAMMARS 137 10.2.2. y. Consider algebraic expressions written with the symbols A = {x.2. For instance: let A = {a. (. z. and terminal. given a ﬁnite set A (the alphabet). Languages and Grammars 10. For instance in the example about algebraic expressions mentioned above. z. +.2. Grammars. b}.
Those expressions are called productions. grammar ) G consists of 1. or an isolated terminal symbol F → x. A production (A. LANGUAGES AND GRAMMARS 138 F → (E). A ﬁnite set V of symbols called vocabulary or alphabet. . 3. and tell us how we can generate syntactically correct algebraic expressions by replacing successively the symbols on the left by the expressions on the right. The right hand side of a production can be any combination of terminal and nonterminal symbols.10. A subset T ⊆ V of terminal symbols. In general a phrasestructure grammar (or simply. For instance the algebraic expression “‘y + (x ∗ y + y) ∗ x” can be generated like this: E ⇒ E +T ⇒ T +T ⇒ F +T ⇒ y +T ⇒ y +T ∗F ⇒ y +F ∗F ⇒ y+(E)∗F ⇒ y+(E+T )∗F ⇒ y+(T +T )∗F ⇒ y+(T ∗F +T )∗F ⇒ y + (F ∗ F + T ) ∗ F ⇒ y + (x ∗ T + T ) ∗ F ⇒ y + (x ∗ F + T ) ∗ F ⇒ y + (x ∗ y + T ) ∗ F ⇒ y + (x ∗ y + F ) ∗ T ⇒ y + (x ∗ y + y) ∗ F ⇒ y + (x ∗ y + y) ∗ x . The elements of N = V − T are called nonterminal symbols or nonterminals. P ). F → z. A start symbol σ ∈ N . B) ∈ P is written: A→B. F → y. σ. 2. 4. The left hand side must contain at least one nonterminal symbol.2. A ﬁnite subset P of (V ∗ −T ∗ )×V ∗ called the set of productions. We write G = (V. T.
P1 ) and G2 = (V2 . T1 . S ::= Tn . σ → σ2 . Productions of the form S ::= T1 . where N1 = V1 − T1 and N2 = V2 − T2 are disjoint (rename nonterminal symbols if necessary). LANGUAGES AND GRAMMARS 139 If α → β is a production and xαy ∈ V ∗ . . Product Rule: the language product of L1 and L2 L1 L2 = {αβ  α ∈ L1 . can be combined as S ::= T1  T2  · · ·  Tn .4. we say that xβy is directly derivable from xαy. for instance.2. 2. and we write α1 ⇒ αn (by convention also α1 ⇒ α1 .2. σ2 . The Backus Normal Form or BNF is an alternative way to represent productions. So.10. .) Given a grammar G.2.3. β ∈ L1 } . T2 . 10. Then 1. . . we say that αn is derivable ∗ ∗ from α1 . Backus Normal Form. Also assume that σ is a new symbol not in V1 ∪ V2 . S ::= T2 . σ1 . Combining Grammars. Let G1 = (V1 . The production S → T is written S ::= T . Let L1 = L(G1 ) and L2 = L(G2 ) be the languages associated respectively to G1 and G2 . Union Rule: the language union of L1 and L1 L1 ∪ L2 = {α  α ∈ L1 or α ∈ L1 } starts with the two productions σ → σ1 . and we write xαy ⇒ xβy . the grammar of algebraic expressions deﬁned above can be written in BNF as follows: E ::= T  E + T T ::= F  T ∗ F F ::= (E)  x  y  z 10. If we have α1 ⇒ α2 ⇒ · · · ⇒ αn (n ≥ 0). P2 ) be two grammars. the language L(G) associated to this grammar is the subset of T ∗ consisting of all strings derivable from σ.
n} 1 1 (n = 1. A ∈ N . G is a contextsensitive (or type 1) grammar if every production is of the form: αAβ → αδβ (i.2. . regular ) if there is a contextsensitive (respectively contextfree.: we may replace A with δ in the context of α and β). . . δ ∈ V .5. β ∈ V ∗ . LANGUAGES AND GRAMMARS 140 where αβ = string concatenation of α and β. a ∈ T . where α ∈ V ∗ − T ∗ . G is a contextfree (or type 2) grammar if every production is of the form: A → δ. G is a phrasestructure (or type 0) grammar if every production is of the form: α → δ. Types of Grammars (Chomsky’s Classiﬁcation). 3. starts with the two productions σ → σ1 σ . .10. B ∈ N . αn  αk ∈ L1 . . 10. . A language L is contextsensitive (respectively contextfree. or A → λ . . k = 1. . σ → λ. starts with the production σ → σ1 σ2 . δ ∈ V ∗ . regular) grammar G such that L = L(G).e.2. The following examples show that these grammars deﬁne diﬀerent kinds of languages. . . δ ∈ V ∗ − {λ}. 1. 2. 2. ∗ where A ∈ N . . G is a regular (or type 3) grammar if every production is of the form: A → a or A → aB where A. ). Let G be a grammar and let λ denote the null string. 1 1 1 1 were L0 = {λ} and Ln = {α1 α2 . 2. where α. 3. . Closure Rule: the language closure of L1 L∗ = L 0 ∪ L 1 ∪ L2 ∪ . 0.
A type 1 grammar for that language is the following: T = {a. Cb → bC . +. with start symbol E. . z. } . . . A → aAbC . } .2. Example: The following language is type 2 (contextfree) but not type 3: L = {an bn  n = 1. 3. T. T →T ∗F . . σ → aAbc . (. 2. with start symbol σ. but they are harder to describe. 3. b}. A. N = {σ}. . 2. and productions E → T. y. 2. Cc → cc . S → b. Two grammars G and G are equivalent if L(G) = L(G ). b. A type 3 grammar for that language is the following: T = {a.2. 2. LANGUAGES AND GRAMMARS 141 Example: The following language is type 3 (regular): L = {an bm  n = 1. and productions: σ → aσ . ∗. . N = {σ. Equivalent Grammars. with start symbol σ. b}. and productions σ → aσb . m = 1. } . c}. with start symbol σ. A type 2 grammar for that language is the following: T = {a. nonterminal symbols = {E. S → bS . . E → E + T. σ → aS . Example: The grammar of algebraic expressions deﬁned at the beginning of the section is equivalent to the following one: Terminal symbols = {x. . )}. F. . σ → ab . Example: The following language is type 1 (contextsensitive) but not type 2: L = {an bn cn  n = 1. L}. T → F. S}.10. and productions σ → abc . A → abC . 10. 3.6. There are also type 0 languages that are not type 1. . . N = {σ. C}. 3 . .
stages 4–5 A way to represent an intermediate stage of the making of the fractal is by representing it as a sequence of movements of three kinds: ’d’= draw a straight line (of a ﬁx length) in the current direction. L → y. as shown below. ContextFree Interactive Lindenmayer Grammar.7. so we get dldrrdldldldrrdldrrdldrrdldldldrrdld If we represent by D a segment that may no be ﬁnal yet. Von Koch Snowﬂake.1. A contextfree interactive Lindenmayer grammar is similar to a usual contextfree grammar with the diﬀerence that it allows productions of the form A → B where A ∈ N ∪ T (in a context free grammar A must be nonterminal). L → z. Figure 10. ’r’= turn right by 60◦ . LANGUAGES AND GRAMMARS 142 F → (E).2. Its rules for deriving strings also are diﬀerent. For instance we start with a single horizontal line d. all symbols in α must be replaced simultaneously. The von Koch Snowﬂake fractal is the limit of the sequence of curves deﬁned by that process. then the sequences of commands used to build any intermediate stage of the curve can be deﬁned with the following grammar: . L → x. stages 1–3. Example: The von Koch Snowﬂake. 10. Figure 10. Von Koch Snowﬂake. The von Koch Snowﬂake is a fractal curve obtained by start with a line segment and then at each stage transforming all segments of the ﬁgure into a four segment polygonal line. to derive string β from string α. which we then transform into the polygonal dldrrdld.10.2. F → L. ’l’= turn left by 60◦ . then each segment is transformed into a polygonal according to the rule d → dldrrdld. In a contextfree interactive Lindenmayer grammar.2.
L. 1] → [0. R → RLR . l}. l → l. The corresponding grammar is as follows: T = {l. i. R}. . Example: The Peano curve. 1]2 .10. r → r. l → l. r. L → RLLLR . deﬁned as the limit of the sequence of curves shown in the ﬁgures below.. Figure 10. and productions: D → DlDrrDlD . Peano curve. stages 1–4. The Peano curve is a space ﬁlling curve. L → l. with start symbol D. r → r. and productions C → LLLL .e. LANGUAGES AND GRAMMARS 143 N = {D}. r}. with and start symbol C.3.2. a function f : [0. R → r. D → d. 1]2 such that the range of f is the whole square [0. N = {C. T = {d. Each element of that sequence of curves can be described as a sequence of 90◦ arcs drawn either anticlockwise (’l’) or clockwise (’r’).
β ∈ L2 } L∗ = {α1 . {λ}. and replace each production (A → a) ∈ P1 with A → aσ1 and (A → λ) ∈ P1 with A → σ1 . P ) in which every production is of the form: A → a or A → aB where A. n ∈ N} . a ∈ T . 3. . L1 L2 and L∗ . B ∈ N = V − T .10.. so we need to modify them slightly (we use the same notation as in section 10. add all productions of the form σ → RHS.3. αn  αk ∈ L1 . add σ1 → λ. σ. 2.2. Language Recognition 10. . ∅. If L1 and L2 are regular languages over T the following languages also are regular: L1 ∪ L2 = {α  α ∈ L1 or α ∈ L2 } L1 L2 = {αβ  α ∈ L1 . 2. i. Product Rule: Instead of adding σ → σ1 σ2 .4 do no have the form of a regular grammar. Regular Languages.e. use σ1 as start symbol. . or A → λ . 1 We already know how to combine two grammars (see 10.3. T. L1 L2 and L∗ as follows.3.1. We justify the above claims about L1 ∪ L2 . αn = ’α concatenated with itself n times’): 1. where RHS is the right hand side of some production (σ1 → RHS) ∈ P1 or (σ2 → RHS) ∈ P2 . Closure Rule: Instead of adding σ → σ1 σ and σ → λ. 1 T ∗ − L1 = {α ∈ T ∗  α ∈ L1 } . use σ1 as start symbol and replace each production (A → a) ∈ P1 with A → aσ2 and (A → λ) ∈ P1 with A → σ2 . Union Rule: Instead of adding σ → σ1 and σ → σ2 . the only problem is that the rules 1 given in section 10.4) L1 and L2 to obtain L1 ∪ L2 .2. / L1 ∩ L2 = {α  α ∈ L1 and α ∈ L2 } . Recall that a regular language is the language associated to a regular grammar. Regular languages over an alphabet T have the following properties (recall that λ = ’empty string’.4): 1. αβ = ’concatenation of α and β’. and {a} are regular languages for all a ∈ T . LANGUAGE RECOGNITION 144 10. a grammar G = (V.2.
etc.3. Next we deﬁne recursively the language associated to a given regular expression: L(∅) = ∅ . A ﬁnite set of states S.3. the expression a∗ bb∗ represents all strings of the form an bm with n ≥ 0.3. Another way of characterizing regular languages is as sets of strings recognized by ﬁnitestate automata. In order to use fewer parentheses we assign those operations the following hierarchy (from do ﬁrst to do last): ∗. R · S.3. L(λ) = {λ} . If R and S are regular expressions over T the following expressions are also regular: (R). (language closure). 2. b} than start with a and end with b. But ﬁrst we need a generalization of the concept of ﬁnitestate automaton. Regular Expressions. R∗ . 10. L(R · S) = L(R)L(S) L(R ) = L(R) ∗ ∗ for each a ∈ T .10. ·. Given an alphabet T . A nextstate or transition function f : S × I → P(S). A ﬁnite set of input symbols I. LANGUAGE RECOGNITION 145 10. R + S.2. ∅. 4. and a are regular expressions for all a ∈ T . a(a + b)∗ b is the set of strings over {a. for instance. So. 3. Regular languages can be characterized as languages deﬁned by regular expressions. +. a∗ (b + c) is the set of strings consisting of any number of a’s followed by a b or a c. λ. 2. We may omit the dot: α · β = αβ. a regular expression over T is deﬁned recursively as follows: 1. (language product). An initial state σ ∈ S. L(a) = {a} L(R + S) = L(R) ∪ L(S) . Nondeterministic FiniteState Automata. Formally a nondeterministic ﬁnitestate automaton consists of 1. m > 0. as we will see next. A nondeterministic ﬁnitestate automaton is a generalization of a ﬁnitestate automaton so that at each state there might be several possible choices for the “next state” instead of just one. .
σ. T. Let P be the set of productions S → xS if there is an edge labeled x from S to S and S→λ if S is a ﬁnal state. Let σ be the initial state. LANGUAGE RECOGNITION 146 5. F). C. n ≥ 0. Then the set of strings recognized by A is precisely L(G). Since that path ends in a ﬁnal state. m > 0. Regular grammar associated to a nondeterministic ﬁnitestate automaton. σ. I. C. P ) . σ. Example: For the nondeterministic automaton deﬁned above the corresponding grammar will be: Õ @ABC G GFED σ Ö @ABC G GFED C b @ABC 89:. We say that a nondeterministic ﬁnitestate automaton accepts or recognizes a given string of input symbols if in its transition diagram there is a path from the starting state to a ﬁnal state with its edges labeled by the symbols of the given string. A subset F of S of accepting or ﬁnal states. ?>=< G GFED F . For instance the string bbabb is represented by the path (σ. let N be the set of states. Let A be a nondeterministic ﬁnitestate automaton given as a transition diagram. the string is recognized by the automaton. f. Next we will see that there is a precise relation between regular grammars and nondeterministic ﬁnitestate automata. Let G be the regular grammar G = (V.10. Let T be the set of inputs symbols. We represent the automaton A = (S. Example: Consider the nondeterministic ﬁnitestate automaton deﬁned by the following transition diagram: b b a start This automaton recognizes precisely the strings of the form bn abm .3. F ). and V = N ∪ T . σ. A path (which we can express as a sequence of states) whose edges are labeled with the symbols of a string is said to represent the given string.
F → λ. f . C → bF . LANGUAGE RECOGNITION 147 T = {a. T. C. f (X. F) be a nondeterministic ﬁnitestate automaton. σ. The idea is to show that given any nondeterministic ﬁnitestate automata it is possible to construct an equivalent deterministic ﬁnitestate automata recognizing exactly the same set of strings. S = P(S). x) = {S  S → xS ∈ P } ∪ {F  S → x ∈ P }. Nondeterministic ﬁnitestate automaton associated to a given regular grammar. F ). Let I = T. σ. where 1. Here we will see that the term “nondeterministic” can be dropped. Then A is equivalent to the ﬁnitestate automaton A = (S . In the previous section we saw that regular languages coincide with the languages recognized by nondeterministic ﬁnitestate automata. σ = {σ}. P ) be a regular grammar. F) recognizes precisely the strings in L(G). N = {σ. The string bbabb can be produced like this: σ ⇒ bσ ⇒ bbσ ⇒ bbaC ⇒ bbabC ⇒ bbabbF ⇒ bbabb . C → bC . b}. with productions σ → bσ . x) . Let G = (V. f. 5. Relationships Between Regular Languages and Automata.10. F }. 3. so that regular languages are precisely those recognized by (deterministic) ﬁnitestate automata. I = I. I . x) = ∅ . σ. and F ∈ V . I. S∈X . where N = V − T . S = N ∪ {F }. x) = f (S. 4. σ . Then the nondeterministic ﬁnitestate automaton A = (S. The main result is the following: Let A = (S.3.3. 10. f. f (∅.4. / f (S. σ → aC . 2. F = {X ⊆ S  X ∩ F = ∅}. F = {F } ∪ {S  S → λ ∈ P } . I.
C. {σ. F }. F }. {F }.10. {C. BC 89 66 @A II v a 6 II II 666 b II 6 a b a II 6 Ø ' GF ?> ED =< GF ?> =< {F ED {C. b}. Ø GF ED G@A BC {σ} y Ø a start a ED GGF {C} @A BC b GF ?> ED a =< G@A {C. After removing the unreachable states we get the following simpliﬁed version of the ﬁnitestate automaton: Ø GF ED G@A BC {σ} b a GF ?> =< ED GF ED G {C} o {σ. i. @A C.: S = {∅. {σ. Then for each element X of S we draw an edge labeled x from X f (S.3. F } 89 BC :. F } @A } 89 BC :. LANGUAGE RECOGNITION 148 Example: Find a (deterministic) ﬁnitestate automaton A equivalent to the following nondeterministic ﬁnitestate automaton A: b b a start Answer : The set of input symbols is the same as that of the given automaton: I = I = {a. The ﬁnal states of A are the elements of S containing some ﬁnal state of A: F = {{F }. ?>=< G GFED F start We notice that some states are unreachable from the starting state. The set of states is the set of subsets of S = {σ. BC @A BC dII y 66 I xY 66 IIa a xxx a b b x 66 III x I 66 xx ED GF ?> =< GFEDr b 666b GF C} P ∅ @ABCdI {σ. F }. {σ. {C}. ED {σ. F }} . Ø a b × GFED G@ABC V ∅ e a b . C. @A 89 BC :. C. {σ. F }. The starting state is {σ}. {C. F }} . x) (and from ∅ to ∅): to S∈X b b Õ @ABC G GFED σ Ö @ABC G GFED C b @ABC 89:. BC @A F } :. C}.e. {σ}. F }. F } 89 :. {σ.
once proved that every nondeterministic ﬁnitestate automaton is equivalent to some deterministic ﬁnitestate automaton.10. we obtain the main result of this section: A language L is regular if and only if there exists a ﬁnitestate automaton that recognizes precisely the strings in L. .3. LANGUAGE RECOGNITION 149 So.
so: 3547 = ((32 · 3)2 · 3)2 · 3. hence 547 = 29 + 25 + 2 + 1 = 4 4 ((24 + 1) 24 + 1) 2 + 1. Eﬃcient Computation of Powers Modulo m We illustrate an eﬃcient method of computing powers modulo m with an example. and reducing modulo 2 10 at each step: 32 = 9 (mod 10). First write 547 in base 2: 1000100011.APPENDIX A A.m) {computes a^x mod m} 2: p := 1 3: bx := binary array(x) {x as a binary array} 4: t := a mod m 5: for k := 1 to length(bx) 6: begin 7: p := (p * p) mod m 8: if bx[k] = 1 then {if kth binary digit of x is 1} 9: p := (p * t) mod m 10: end 11: return p 12: end pow mod 150 . At the end we ﬁnd 3547 = 7 (mod 10). Assume that we want to compute 3547 mod 10. 3 4 4 32 = 12 = 1 (mod 10).x. etc. 32 = 12 = 1 (mod 10). 32 · 3 = 1 · 3 = 3 (mod 10). 32 = 92 = 81 = 1 (mod 10). The algorithm in pseudocode would be like this: 1: procedure pow mod(a. Next we compute the expression beginning with the inner parenthesis.1.
A. int x. if (x & y) { p *= a. } return p.x.2)). EFFICIENT COMPUTATION OF POWERS MODULO M 151 The following is a program in C implementing the algorithm: int pow(int a. a %= m.m) {computes a^x mod m} 2: p := 1 3: t := a mod m 4: while x > 0 5: begin 6: if x is odd then 7: p := (p * t) mod m 8: t := (t * t) mod m 9: x := floor(x/2) 10: end 11: return p 12: end pow mod . } y >>= 1. } The following is an alternative algorithm equivalent to running through the binary representation of the exponent from right to left instead of left to right: 1: procedure pow mod(a.1. while (!(y & x)) y >>= 1. int m) { int p = 1. while (y) { p *= p. p %= m. int y = (1 << (8 * size of(int) . p %= m.
1. A tape symbol read from the current tape cell. All other cells contain the blank symbol. It consists of 1. A ﬁnitestate automaton is just a Turing machine whose tape head moves always from left to right and never writes to the tape.2. Machines and Languages A.2. MACHINES AND LANGUAGES 152 A. 4. A direction for the tape head to move: L = ’move one cell to the left’. 2. Turing Machines. A Turing machine is a theoretical device intended to deﬁne rigorously the concept of algorithm.2. Each cell may be empty or may contain a symbol from a given alphabet. 5. A control unit containing a ﬁnite set of instructions. 3. R = ’move one cell to the right’. S = ’stay in the current cell’. An inﬁnite tape made of a sequence of cells. which we may denote λ. . Turing Machine. 3.1. A tape symbol to write into the current tape cell.A.1. Tape Tape head control unit Figure A. 2. The input of the ﬁnitestate automaton is presented as symbols written in the tape. The next machine state. A tape head able to read and write (or delete) symbols from the tape. An input is represented on the tape by placing the letters of the strings in contiguous tape cells. In general we make the following assumptions: 1. Turing machines are generalizations of ﬁnitestate automata. Each machine instruction contains the following ﬁve parts: The current machine state.
Otherwise the string is rejected. A pushdown automaton is ﬁnitestate automaton with a stack. If a Turing machine has at least two instructions with the same state and input letter. Let f : N → N the function deﬁned so that f (n) = m means that if the . FiniteState Automata.e. nop. a storage structure in which symbols can be put and extracted from it by two operations: push (place on the top of the stack) and pop (take from the top of the stack)—consequently the last symbol put into the stack is the ﬁrst symbol taken out. Otherwise it is deterministic. We say that an input string is accepted by a Turing machine if the machine enters the Halt state.A. After reading the next input symbol and the symbol at the top of the stack. A ﬁnitestate automata can be interpreted as a Turing machine whose tape head moves only from left to right and never writes to the tape. Linearly Bounded Automata. There is one halt state. MACHINES AND LANGUAGES 153 2. This can happen in two ways: by entering a state other than the Halt state from which there is no move. The output of the Turing machine is the contents of the tape when the machine stops. The execution of a Turing machine stops when it enters the Halt state or when it enters a state for which there is no valid move. 3. There is one starting state. A linearly bounded automaton is a Turing machine whose tape is limited to the size of its input string plus two boundary cells that may not be changed. then the machine is nondeterministic. Computable Functions. The next state function takes into account not only the current state and the symbol read from the input. The tape is initially positioned at the leftmost cell of the input string unless speciﬁed otherwise. but also the symbol at the top of the stack. the automaton executes a stack operation and goes to the next state. that leaves the stack intact. or by running forever (for instance executing an inﬁnite loop).2. which we denote by “Halt”. Additionally there is a third operation. 4. Consider a Turing machine T working on symbols from an alphabet of only one symbol A = {} (“stroke”).. i. Pushdown Automata. Initially there is a single symbol in the stack.
} 5. A language is recursively enumerable if there is a Turing machine that outputs all the strings of the language. 1. recognized by deterministic pushdown automata.2. Deterministic contextfree languages. where f0 . recognized by nondeterministic pushdown automata. Here we mention a hierarchy of languages that includes (and extends) Chomsky’s classiﬁcation. We say that f is computed by the Turing machine T . 7. languages that are not deﬁnable by any grammar and cannot be recognized by Turing machines. 2. recognized by Turing machines. Example: palindromes over {a. . Contextsensitive languages. . A computable function f (n) halts for a given value of its argument n if T with input n + 1 strokes halts. Unrestricted or phrasestructure grammars. 4. 2. b}. . }.2. An eﬀective enumeration of a set is a listing of its elements by an algorithm. 2. . is an eﬀective enumeration of all computable functions. Example: {an bn  n = 1. . languages without λ recognized by linearly bounded automata. They are recognized by ﬁnitestate automata. A. 3 . A computable function f is total if f (n) halts for every n. Example: {an  fn is total}. f1 . Hierarchy of Languages. the output of T is a string of m + 1 strokes. . A computable function is a function computed by some Turing machine. 2. in increasing order of inclusion. . . . f2 . 6. Example: {an bn cn  n = 1. Regular languages. Example: {an  fn (n) halts}.A. 3 . Nongramatical languages. MACHINES AND LANGUAGES 154 initial input of T consists of a string of n + 1 strokes. n = 1. .2. Recursively enumerable languages. }. 3 . Contextfree languages. 3. Example: {am bn  m.