\documentclass[12pt]{extarticle}

\usepackage[utf8]{inputenc}
\usepackage{geometry}
\geometry{a4paper, total={170mm, 257mm}, left=26mm, top=26mm,}
\usepackage{amsthm}
\usepackage{textcomp}
\usepackage{amssymb}
\usepackage{amsmath}
\usepackage{tipa}
\usepackage{graphicx}
\usepackage{enumerate}

\usepackage{dsfont}
\usepackage{tikz-cd}
\usepackage{tikz}

\usetikzlibrary{shadings}
\usepackage{hyperref}
\usepackage[usestackEOL]{stackengine}[2013-10-15]
\def\x{\hspace{3ex}} %BETWEEN TWO 1-DIGIT NUMBERS
\def\y{\hspace{2.45ex}} %BETWEEN 1 AND 2 DIGIT NUMBERS
\def\z{\hspace{1.9ex}} %BETWEEN TWO 2-DIGIT NUMBERS
\usepackage{pst-fractal}

\usetikzlibrary{decorations}
\pgfdeclaredecoration{quasi-sirpinski}{do}{%
\state{do}[width=\pgfdecoratedinputsegmentlength, next state=do]{%
\pgfpathmoveto{\pgfpointpolar{-60}{\pgfdecoratedinputsegmentlength/2}}%
\pgfpathlineto{\pgfpointorigin}%
\pgfpathlineto{\pgfpoint{\pgfdecoratedinputsegmentlength/2}{0pt}}%
\pgfpathclose%
}
}

\stackMath
\let\iff\Longleftrightarrow
\let\implies\Longrightarrow
\newcommand{\vsp}{\vspace{.3in}}
\newcommand{\cl}{\centerline}
\newcommand{\N}{\mathbb{N}}
\newcommand{\R}{\mathbb{R}}
\newcommand{\Q}{\mathbb{Q}}
\newcommand{\C}{\mathbb{C}}
\newcommand{\Z}{\mathbb{Z}}
\newcommand{\D}{\mathbb{D}}
\newcommand{\h}{\mathbb{H}}
\newcommand{\done}{\hfill $\diamondsuit$}
\graphicspath{{Images/}}
\newtheorem{theorem}{Theorem}[section]
\newtheorem{corollary}{Corollary}[theorem]
\newtheorem{lemma}[theorem]{Lemma}
\theoremstyle{definition}
\newtheorem{definition}[theorem]{Definition}
\newtheorem{example}[theorem]{Example}

\usepackage{soul}
\usepackage{graphicx}
\graphicspath{ {images/} }
\title{
{MATH 330: Lecture Notes}\\
{\large Binghamton University}}
\author{David P Biddle}
\date{5/24/2022}

\tolerance=1
\emergencystretch=\maxdimen
\hyphenpenalty=10000
\hbadness=10000

\begin{document}

\maketitle

\vspace{.3in}

\tikz
\fill (0,0) -- ++(60:3) -- ++(-60:3) -- cycle;

\tikz[decoration=quasi-sirpinski]
\fill decorate { (0,0) -- ++(60:3) -- ++(-60:3) -- cycle };

\tikz[decoration=quasi-sirpinski]
\fill decorate { decorate {
(0,0) -- ++(60:3) -- ++(-60:3) -- cycle } };

\tikz[decoration=quasi-sirpinski]
\fill decorate { decorate { decorate {
(0,0) -- ++(60:3) -- ++(-60:3) -- cycle } } };

\tikz[decoration=quasi-sirpinski]
\fill decorate { decorate { decorate { decorate {
(0,0) -- ++(60:3) -- ++(-60:3) -- cycle } } } };
\setcounter{section}{0}
\section{Sets and Logic}

\vspace{.2in}

A mathematical proof consists of a sequence of mathematical statements threaded together by rules of

inference. We start by defining the simplest kind of mathematical statements, those that belong to
propositional logic. A \emph{statement} is a meaningful declarative sentence that is unamibiguously
either \emph{true} or \emph{false}. For reference a declarative sentence should have a subject, a
verb/predicate, and should end with a period. We can assign to a statement its truth value, which we
will do here by assigning either $0$ or $1$ to the statement, and we write this value by writing
$tr(P)=1$ if the statement $P$ is true and $tr(P)=0$ if the statement $P$ is false. Propositional logic
concerns itself with a way of putting simple statements together to form what we call \
emph{compound} statements, while at the same time providing a clear method to determine the truth
value of these new statements. The way we put simple statements together is through the use of
connectives; we introduce the fundamental logical connectives now.

\vspace{.2in}

Given statements $P$ and $Q$ we form the statement $P\land Q$, which is said as "$P$ and $Q$", and
we define its truth value by saying $tr(P\land Q)=1$ exactly in the case $tr(P)=1=tr(Q)$ and in no other
cases; the connective $\land$ is called logical \emph{conjunction}. Likewise we can form the
statement $P\lor Q$, which is said as "$P$ or $Q$", and we define its truth value by saying $tr(P\lor
Q)=0$ exactly when both $tr(P)=0=tr(Q)$ and in no other cases; the connective $\lor$ is called
logical \emph{disjunction}. We emphasize here that our $\lor$ is inclusive in that $tr(P\lor Q)=1$ if
both $tr(P)=tr(Q)=1$. There is another connective called the "exclusive disjunction" than we will not
be using in these notes.

\vspace{.2in}

For any statement $P$ we can form the statement $\neg P$, which is said as "not $P$", and we define
its truth value by saying $tr(\neg P)=1-tr(P)$; the connective $\neg$ is called logical \emph{negation}.

\vspace{.2in}

For the most part the three connectives $\land, \lor, \neg$ keep reasonably faithful to how most of us
interpret the words "and", "or", and "not" in our regular use of language. If one says "David has a ferret
and $1+1=3$", one is perfectly happy saying this statement is false since $tr(\text{David has a
ferret})=1$ while $tr(1+1=3)=0$, and we're more or less fine with the idea that the truth value of one of
these simple statements does not affect the truth value of the other. David having ferrets is not the same
subject as basic arithmetic unless we do something quite silly such as "David has a ferret Fergus and a
ferret Freddie and one ferret plus one ferret equals three ferrets". We'll refrain from silliness and ferret
arithmetic for the most part in these notes.

\vspace{.2in}

The next connective, implies, can cause a bit of confusion because of the way we tend to use 'implies'
in our vernacular. Given statements $P$ and $Q$ we form the statement $P\implies Q$, which is said
"$P$ implies $Q$", and we define its truth value by saying $tr(P\implies Q)=0$ exactly when $tr(P)=1$
and $tr(Q)=0$ and in no other cases. So $P\implies Q$ is true unless $P$ is true and $Q$ is false. We
note here that if $tr(P)=0$ then $tr(P\implies Q)=1$ which might seem peculiar at first; we often say
that $P\implies Q$ is true vacuously in this case. We can think of the truth value of $P\implies Q$ as
being a promise rather than causality as in the following example: let $P$ be the statement "your kid
gets an A" and let $Q$ be the statement "you buy them a car". Then we interpret $P\implies Q$ as the
promise "If your kid gets an A then you buy them a car" in the four scenarios as follows:

\begin{itemize}
\item If $tr(P)=1$, your kid gets an A, and $tr(Q)=1$, you buy them a car, you've honored the
promise.
\item If $tr(P)=1$, your kid gets an A, and $tr(Q)=0$, you don't but them a car, you broke your
promise.
\item If $tr(P)=0$, your kid doesn't get an A, and $tr(Q)=1$, you buy them a car, you've honored the
promise because you haven't broken it. You only said you'd buy them a car if they did get an A, you
never said anything about what would happen if they didn't get an A. You might have some parenting
issues however.
\item If $tr(P)=0$, your kid doesn't get an A, and $tr(Q)=0$, you don't buy them a car, there's no
broken promise either.
\end{itemize}

\vspace{.1in}

In $P\implies Q$ there may be absolutely no relationship between $P$ and $Q$ just as in $P\land Q$ or
$P\lor Q$. For example if $P$ represents "$1+1=2$ and $Q$ represents "$\pi$ is irrational" then the
truth value of $P\implies Q$ or "If $1+1=2$ then $\pi$ is irrational" still just depends on the truth
values of $P$ and $Q$ and not any perceived relationship between the statements. There are a lot of
different ways we phrase implications in regular language: we might say "$Q$ is necessary for $P$" or
"$Q$ follows from $P$" or "$P$ only if $Q$" for $P\implies Q$.

\vspace{.2in}

Sometimes implications statements are a bit more subtle. For instance "all triangles are equilateral" and
"the square of a real number is non-negative" might not seem like implications but can be rephrased as
such: "if $T$ is a triangle, then $T$ is equilateral" and "if $x$ is a real number, then $x^2\geq 0$",
have the same meaning as the aforementioned statements.

\vspace{.2in}

We add here one more logical connective that is more of a convenience than a necessity. Given
statements $P$ and $Q$ we form the statement $P\iff Q$, which is said as "$P$ if and only if $Q$, and
we define its truth value by saying $tr(P\iff Q)=1$ if $tr(P)=tr(Q)$ and in no other cases; the
connective $\iff$ is called the logical biconditional. Other ways we may use this connective include
'$P$ happens exactly when $Q$ happens' or '$P$ and $Q$ are logically equivalent'. The phrase "if and
only if" is commonly abbreviated to "iff".

\vspace{.2in}

The use of the numbers $0$ or $1$ as truth values allows us to use a bit of algebra in our determination
of truth values. In order for our arithmetic to make sense we define two operations on the symbols $0$
and $1$ which we call addition and multiplication:

\begin{center}
\begin{tabular}{ c| c | c }
+ & 0 & 1 \\
\hline
0 & 0 & 1 \\
\hline
1 & 1 & 0\end{tabular}
\end{center}

\begin{center}
\begin{tabular}{ c| c | c }
$\cdot$ & 0 & 1 \\
\hline
0 & 0 & 0 \\
\hline
1 & 0 & 1\end{tabular}
\end{center}

\vspace{.2in}

Let $tr(P)=a, tr(Q)=b$. Then we have that $tr(\neg P)=1+tr(P)=1+a$ and $tr(P\land Q)=ab$ since if
either $P$ or $Q$ is false then $P\land Q$ is false and $P\land Q$ is true exactly when both $P$ and
$Q$ are true, the product $ab=a\cdot b$ completely captures these with multiplication. Finding $tr(P\
lor Q)$ we see that $a+b+ab$ does the job since if both $P$ and $Q$ are false this is $0$ and if $P$ is
true and $Q$ is false this is $1+0+0=1$ and likewise if $P$ is false and $Q$ is true this is $0+1+0=1$.
Lastly if $P$ is true and $Q$ is true this is $1+1+1\cdot 1=0+1=1$. Note that the polynomials $ab$ and
$a+b+ab$ are symmetric in $a$ and $b$ in the sense that if we switch the roles of $a$ and $b$ we end
up with the same expression. We expect this is not the same for implication. $tr(P\implies Q)=1+a+ab$
which works since if $P$ is false this is $1$
and if $P$ is true and $Q$ is true this is $1+1+1=0+1=1$ and lastly if $P$ is true and $Q$ is false this
is $1+1+0=0+0=0$. Note that if we switch the roles of $a$ and $b$ in the polynomial $1+a+ab$ we do
not get the same polynomial, that order is very important to understanding this connective.

\vspace{.2in}

What about $tr(P\iff Q)$? This should certainly be symmetric and this is similar to $tr(P\land Q)$ in
that $tr(P\iff Q)=a+b+1$; we leave it to the reader to verify this.

\vspace{.2in}

Suppose we are given a compound statement $Q$ that is constructed using various connectives applied
to the simple statements $P_1, P_2,\dots P_n$. We'd like the determine the truth value of $Q$ based on
the truth values of the $P_i$ fairly systematically. One way of doing this that entertains all possible
truth values of $Q$, is by constructing a \emph{truth table}. We give an example first before describing
the process in general.
\vspace{.2in}

Suppose we have the compound statement:

$$
Q:=((P_1\implies P_2)\iff(\neg P_2\implies \neg P_1)).
$$

We build a table to look at all possible combinations of $tr(P_1),tr(P_2)$:

\vspace{.1in}

$$
\begin{array}{|c|c|c|c|c|c|c|}
P_1 & P_2 & \neg P_1 & \neg P_2 & P_1\implies P_2 & \neg P_2 \implies \neg P_1 & Q\\
\hline
1 & 1 & 0 & 0 & 1 & 1 & 1 \\
1 & 0 & 0 & 1 & 0 & 0 & 1 \\
0 & 1 & 1 & 0 & 1 & 1 & 1 \\
0 & 0 & 1 & 1 & 1 & 1 & 1 \\
\end{array}
$$

\vspace{.2in}

We note the first two columns consist of all $4$ truth combinations of $tr(P_1)$ and $tr(P_2)$ set up in
a way that the rows are ordered \emph{lexicographically} with respect to the order $1>0$ based on the
truth values in these two columns: considering just the row entries in the first two columns, we arrange
the rows so that if a row $R$ has truth values $a,b$ from left to right and $R'$ has truth values $a',b'$
we write $R>R'$ (and put $R$ above $R'$) if either

\begin{enumerate}
\item $a>a'$
or
\item $a=a'$ and $b>b'$.
\end{enumerate}

\vspace{.2in}

In general if $Q$ consists of $n$ simple statements $P_1, P_2, \dots , P_n$ we have $2^n$ possible
combinations of truth values $tr(P_i)$, and we construct the first two columns of the truth table so that
the rows are ordered so that if the row $R$ has truth values $a_1,a_2,\dots, a_n$ and row $R'$ has truth
values $a_1', a_2', \dots , a_n'$ we say $R>R'$ (and put $R$ above $R'$ in the table) if either

$$
a_1>a_1'
$$
or
$$
a_1=a_1'\quad \text{and}\quad a_2>a_2'
$$
or
$$
a_1=a_1', a_2=a_2' \quad \text{and}\quad a_3>a_3'
$$
and so on. We then construct the remainder of the columns of the table by analyzing each connective
systematically.

\vspace{.2in}

\begin{example}Suppose our compound statement is $Q:=((P_1\implies P_2)\lor P_3)\land P_2$. Our

truth table will consist of $2^3=8$ rows and because the statement $Q$ is built up from three
statements and three connectives, the table will have $3+3=6$ columns as follows:

\vspace{.1in}

\begin{displaymath}
\begin{array}{|c|c|c|c|c|c|}
P_1 & P_2 & P_3 & P_1\Longrightarrow P_2 & (P_1\Longrightarrow P_2)\lor P_3 & \quad Q\\
\hline
1 & 1 & 1 & 1 & 1 & \quad1 \\
1 & 1 & 0 & 1 & 1 & \quad1 \\
1 & 0 & 1 & 0 & 1 & \quad0 \\
1 & 0 & 0 & 0 & 0 & \quad0 \\
0 & 1 & 1 & 1 & 1 & \quad1 \\
0 & 1 & 0 & 1 & 1 & \quad1 \\
0 & 0 & 1 & 1 & 1 & \quad0 \\
0 & 0 & 0 & 1 & 1 & \quad0 \\
\end{array}
\end{displaymath}
\end{example}

\vspace{.2in}

\begin{definition}A statement without any explicit connectives is called an \emph{atomic statement}.

A \emph{well-formed statement} is a statement $Q$ that is defined recursively as follows: either $Q$
atomic or $Q$ is formed by applications of the connectives $\neg,\land,\lor,\implies$ to previously
established well-formed statements $P_1$ and $P_2$ in the sense that if we know that $P_1$ and
$P_2$ are well-formed we also know that the following are well-formed:

$$
(\neg P_1)\quad (P_1\land P_2)\quad (P_1\lor P_2)\quad (P_1\implies P_2).
$$

\end{definition}

We encourage the use of parentheses when building well-formed statements each time a connective is
introduced but are content with leaving off an outer pair of parenthesis; note the example $Q:=((P_1\
implies P_2)\lor P_3)\land P_2$ which we consider well-formed if $P_1,P_2,P_3$ are atomic since it
can be constructed in stages: first $(P_1\implies P_2)$, second $((P_1\implies P_2)\lor P_3)$, and
lastly $(((P_1\implies P_2)\lor P_3)\land P_2) $.

\vspace{.2in}

We will only be dealing with well-formed statements in these notes. This avoids a range of issues, for
instance grammatically awkward or possibly ambiguous phrases like "$Q:= S\lor T\land U$"; even if
$S,T,U$ are atomic it's unclear what the truth value of $Q$ is given the truth values of $S,T,U$. Both $
(S\lor T)\land U$ and $S\lor(T\land U)$ on the other hand are well-formed. Truth tables can always be
constructed for well-formed statements, and the truth value of a well-formed statements can clearly and
unambiguously be obtained from the truth value of its atomic components. From this point onward we
will use the word 'statement' to mean 'well-formed statement'.

\vspace{.2in}

\begin{definition}\quad Suppose we have two statements $Q_1$ and $Q_2$ with $tr(Q_1)=tr(Q_2)$.
We say that $Q_1$ and $Q_2$ are logically equivalent or propositionally equivalent and write $Q_1\
equiv Q_2$.
\end{definition}

\vspace{.2in}

We note here that unless $Q_1$ and $Q_2$ are literally the same statement it is bad form to write
$Q_1=Q_2$. This is a very common mistake people make when learning propositional logic, we point
it out to give the reader caution with the similarity of the symbols $=$ and $\equiv$.

\vspace{.2in}

The way we determine whether two statements $Q_1, Q_2$ are logically equivalent is by constructing
a truth table which builds up both statements by considering all possible assignments of truth values in
the atomic statements making up each statement $Q_1,Q_2$ and seeing if the columns under these
statements are identical.

\vspace{.2in}

\begin{example}

We establish that $Q_1:=\neg(P_1\land P_2)$ and $Q_2:=(\neg P_1)\lor (\neg P_2)$ are logically
equivalent:

$$
\begin{array}{|c|c|c|c|c|c|c|}
P_1 & P_2 & P_1\land P_2 & Q_1=\neg(P_1\land P_2) & \neg P_1 &\neg P_2 & Q_2=(\neg P_1)\lor
(\neg P_2)\\
\hline
1 & 1 & 1 & 0 & 0 & 0 & 0\\
 1 & 0 & 0 & 1 & 0 & 1 & 1\\
0 & 1 & 0 & 1 & 1 & 0 & 1\\
0&0&0&1&1&1&1
\end{array}
$$
We see that the columns corresponding to $Q_1$ and $Q_2$ are identical and so $Q_1\equiv Q_2$.
We encourage the reader to build truth tables for $\neg(P_1\lor P_2)$ and $(\neg P_1)\land (\neg P_2)$
to show that these statements are also logically equivalent.
\end{example}

\vspace{.2in}

\begin{definition}

A statement that is always true is called a \emph{tautology}. Thus if the column in a truth table for
such a statement consists solely of $1$'s.
\end{definition}

\vspace{.2in}

There are lots of simple tautologies such as $P\implies P$ and $P\lor \neg P$ for example as well as
more complicated tautologies. Our previous example $((P_1\implies P_2)\iff(\neg P_2\implies \neg
P_1)$ is also a tautology. We could also have verified in this example that the statements $P_1\implies
P_2$ and $\neg P_2\implies \neg P_1 $ are logically equivalent.
We can phrase logical equivalence in terms of the connective $\iff$ in the sense that if $Q_1\equiv
Q_2$ then $Q_1\iff Q_2$ is a tautology and vice a versa. For a list of useful and common tautologies
click \href{https://piazza.com/class_profile/get_resource/hp4txyk89kj6kd/hrhphmx5jar6o}{HERE}.

\vspace{.2in}

A proof or more accurately a \emph{deductive} proof of a statement $Q$ is a process that starts with
statements that are assumed or known to be true and concludes with $Q$ through some valid method of
reasoning. The methods of reasoning we will be most concerned with involve 'rules of inference'. A
rule of inference is a rule that allows us to infer a conclusion from a premise in order to create an
argument. Most rules of inference we will use stem from one that is called \emph{modus ponens} in
Latin. Modus ponens is the rule of inference which allows us to take the information $P$ is true and
$P\implies Q$ is true and allows us to infer that $Q$ is true. One may associate to this rule of inference
the tautology

$$
(P\land (P\implies Q))\implies Q,
$$

which we ask the reader to verify with a truth table. Modus ponens is not the same as this tautology; the
tautology is a statement that is always true which is distinct from the step by step process of deduction,
but we will from time to time blur this distinction between tautology . We can use modus ponens to
deduce other rules of inference: suppose we know that $Q$ is false and that $P\implies Q$ is true; then
we can infer that $P$ must be false as well since if $P$ were true then by modus ponens we would
have $Q$ is true contradicting our assumption that $Q$ is false. This rule of inference is called \
emph{modus tollens} and is simply an application of modus ponens. We will see that this is the basis
for 'proof by contrapositive'. Another natural rule of inference we will use frequently in deductive
reasoning/deductive proofs is as follows: if both $(P\implies Q)$ and $(Q\implies R)$ are true then we
can infer that $(P\implies R)$ is true; this is known as hypothetical syllogism and again stems from a
straightforward tautology. We will not focus so heavily on the names of classical rules of inference
much beyond what has already been stated and a few others discussed below, but the interested reader
can click \href{https://en.wikipedia.org/wiki/List_of_rules_of_inference}{HERE} or check out any
text on mathematical logic for more.

\vspace{.2in}

\begin{example}[Direct Proof]
Suppose we want to prove that "if $x$ is a real number and $x>1$ then $x^2>x$". We proceed as
follows: if $x$ is real and $x>1$ then $x$ and $x-1 $ are both positive,
and hence so is $x(x-1)$. But then $x^2-x>0$ or $x^2>x$.
\end{example}

\vspace{.2in}

In addition to modus ponens there are several other rules of inference which we will briefly outline.
Each of these does have a background tautology in tow and so is 'reasonable' in the sense that modus
ponens is.

\vspace{.2in}

\textbf{Inference/Proof by Contrapositive:} \quad In order to show that $P\implies Q$ is true it is

enough to assume that $Q$ is false and then show that $P$ is false. Associated to this rule of inference
is the tautology

$$
(P\implies Q)\iff (\neg Q\implies \neg P).
$$

\vspace{.2in}

\textbf{Inference/Proof by Contradiction:}\quad Assume that $R$ is a false statement. Then in order to

show that $P\implies Q$ then it is enough to assume $P$ is true and $Q$ is false and derive the false
statement $R$; this is based on the tautology $(P\implies Q)\iff((P\land \neg Q)\implies R)$ where $R$
only takes on the truth value $0$.

\vspace{.2in}

\textbf{Disjunctive Conclusion Proof:}\quad In order to show that $A\implies (B\lor C)$ is true it
suffices to either:

\begin{itemize}
 \item Assume $A$ and $\neg B$ are both true and deduce $C$ is true.
\item Assume $A$ and $\neg C$ are both true and deduce $B$ is true.
\end{itemize}

This rule of inference is associated with the tautologies $(A\implies (B\lor C))\iff ((A\land \neg B)\
implies C)$ and $(A\implies (B\lor C))\iff ((A\land \neg C)\implies B)$

\vspace{.2in}

\textbf{Disjunctive Hypothesis Proof:}\quad In order to show that $A\lor B\implies C$ is true it

suffices to show that both $A\implies C$ and $B\implies C$ are true. This rule of inference is
associated with the tautology $(A\lor B\implies C)\iff((A\implies C)\land(B\implies C) $.

\vspace{.2in}

We encourage the reader to verify the above tautologies. There are other rules of inference besides the
ones listed above, we've simply listed the most common rules of inference/proof methods we'll be
using. Others will be discussed as necessary in these notes when appropriate. To construct a
mathematical proof, namely to show that a 'proposition' or 'theorem' is true, one needs to begin with a
collection of assumed statements (the hypotheses) along with other already understood to be true
statements and definitions, and then find a sequence of statements connected by rules of inference
leading to the conclusion statement.

\vspace{.2in}

In practice the above is often accomplished in a few slightly different but related ways. Let's say one is
trying to establish $P\implies Q$. The first way to do this is to lay out everything you can about
statement $P$ along with the statements of any definitions and any relevant previously established
theorems and deduce from this entirety that the statement $Q$ must be true. This is called the \
emph{forward way} of proving $P\implies Q$. In this proof method one begins by looking at $P$ and
then produces a sequence of statements $P_1, P_2,\dots P_n$ along with compatible rules of inference
so that $P\implies P_1$ and $P_1\implies P_2$, and so on until one infers $P_n\implies Q$ and then
connects these via hypothetical syllogism to get $P\implies Q$. Here the progression is to start at $P$
find $P_1$ then $P_2$ etc.... until arriving at $Q$

\vspace{.2in}

It might be more natural to instead of focusing extensively on $P$ to focus on both $P$ and $Q$ and
look for a statement $P'$ so that one can focus on inferring $P\implies P'$ and $P'\implies Q$ (and
hence $P\implies Q$ as above) and possibly breaking these down further into maybe $P\implies P''$
and $P''\implies P'$ and maybe $P'\implies Q'$ and $Q'\implies Q$ etc... The idea here is that one
works a bit forward from $P$ and a bit backwards from $Q$ and continues until a sequence of
statements and appropriate inferences are made. This is called the \emph{forward-backward way} of
proving $P\implies Q$.

\vspace{.2in}
Often times it is useful for us to have something similar to a statement but where one or more objects in
the statement are indeterminate variables instead of fixed objects. A few examples of these more
general types of statements are: "$x>3$", "$s$ is a number that equals $2t+1$". We can't in general
determine if declarative sentences like these are true or false without being able to substitute in specific
values for the variables. Let's let $Q(x):="x>3"$. Then $Q(2)$ is the atomic statement $"2>3"$, and we
can infer that $tr(Q(2))=0$. Here it is quite reasonable to let $x=2$ here but it would be a bit absurd to
let $x=\text{David}$. We want to make sure that there is a collection of allowable substitutions, and
this gives rise to the concept of a predicate.

\vspace{.2in}

\begin{definition}\quad A \emph{predicate} $P=P(x_1, x_2, \dots ,x_n)$ is a sentence that contains a

finite number of variables and becomes a statement when specific values are substituted for the
variables. The \emph{domain} $D_i$ of a predicate variable $x_i$ is the set of all values that may be
substituted in place of the variable. For each $a_1$ in $D_1$, $a_2$ in $D_2$, etc.... the domain we
get a statement which we write $P(a_1,a_2,\dots)$ so that either $tr(P(a_1,a_2,\dots))=0$ or
$tr(P(a_1,a_2,\dots))=1$.
\end{definition}

\vspace{.2in}

One of the most fundamental objects in mathematics is that of a \emph{set}. A set is, very loosely, a
collection of objects called its members or elements. The key relationship among sets is the relationship
'is a member of' or 'is an element of'. We denote most sets using capital letters like $X,Y, A, B,....$ etc
(though this is not necessary) and the relationship 'is an element of' using the symbol $\in$ and write
$a\in X$ as shorthand for "$a$ is an element of $X$". We'll be discussing a fair number of sets in this
course that you are likely to have already seen in various informal capacities such as the integers or real
numbers but with more of an emphasis on rigorous mathematical properties of these sets. In particular
in these notes we refer to the natural numbers as the set $\mathbb{N}=\{0,1,2,3,\dots\}$, the positive
natural numbers $\mathbb{N}^+=\{1,2,3,\dots\}$, and the integers $\mathbb{Z}=\{\dots,-3,-2,-
1,0,1,2,\dots\}$. The set of rational numbers is written $\mathbb{Q}$ and the set of real numbers $\
mathbb{R}$; we will discuss these sets in much more detail in the course of the notes. Given a
predicate statement $P(x)$ and a set $X$ all of whose elements are in the domain of the predicate
variable, we can form a new set $\{a\in X\, |\, P(a)\}$.

\vspace{.2in}

\begin{example} Let $P(x):="x<3"$ and let the domain $D$ of this predicate be any numbers for
which is makes sense to posit whether that number is less than $3$. Let $X=\mathbb{N}=\
{0,1,2,3....\}$ the set of natural numbers. Then $\{a\in\mathbb{N}\,|\, P(a)\}=\{0,1,2\}$ while if we use
a different set say $\mathbb{R}$ we might build a different set using this predicate: $\{a\in\
mathbb{R}\,|\, P(a)\}=(-\infty,3)$.
\end{example}

\vspace{.2in}

Let's focus for the moment on predicates $P(x), Q(x)$ which we assume to have the same domain (so
the range of objects $x$ for these predicates is the same). We can naturally extend our connectives $\
neg, \land,\lor, \implies$ to predicates $P(x),Q(x)$ by simply forming predicates:
$$
\neg(P(x), P(x)\land Q(x), P(x)\lor Q(x), P(x)\implies Q(x).
$$

\vspace{.2in}

Given a predicate $P(x)$ we would like to be able to build certain statements exerting the existence of
an object $a$ in the domain of the predicate so that $P(a)$ is true; likewise we would also find it useful
to establish that a predicate statement is always true over some collection of objects in the domain of
the predicate. Before we do this we introduce the concept of 'subset' and 'equality' of sets.

\vspace{.2in}

\begin{definition} We say that $X\subseteq Y$ if ($a\in X \implies a\in Y$) is true, and we say $X$ is a
\emph{subset} of $Y$. Two sets $X,Y$ are equal, written $X=Y$, exactly when the statements $X\
subseteq Y$ and $Y\subseteq X$ are both true. Because $((P\implies Q)\land (Q\implies P))\iff (P\iff
Q)$ is a tautology we can say $X=Y \iff ((a\in X)\iff (a\in Y)).$ In this very strong way sets are
determined by their elements.
\end{definition}
\vspace{.2in}

\begin{example} Let $X_1=\{a\in\mathbb{N}\, |\, a>5\}$ and $X_2=\{b\in\mathbb{N}\,|\, \exists c\in\

mathbb{N}, b=c+6\} $. We show that $X_1=X_2$. Let $y\in X_1$ so that $y\in\mathbb{N}$ and
$y>5$. Then $y-5>0$ and $y-6\in\mathbb{N}$. Set $c=y-6$. Then $\exists c\in\mathbb{N}$ so that
$c+6=(y-6)+6=y$ and so $y\in X_2$ and hence $X_1\subseteq X_2$. Let $x\in X_2$. Then $ x\in\
mathbb{N}$ and $\exists c\in\mathbb{N}$ so that $x=c+6$. Since $x=c+6\geq 6$ we know that
$x>5$ and so $x\in X_1$ and we have $X_2\subseteq X_1$ and so $X_1=X_2$. Here we have
$X_1=X_2=\{6,7,8,\dots \}.$
\end{example}

\vspace{.2in}

Suppose we have a predicate statement $P(x)$ with domain $U$. If $X\subseteq U$ we define the truth
value of the statement $(\exists x\in X) P(x)$ to be $1$ if $P(a)$ is true for some $a\in X$ and $0$
otherwise. We call $\exists$ the \emph{existential quantifier} and read $(\exists x\in X) P(x)$ as "there
exists an $x$ in $X$ so that $P(x)$". Similarly we define $(\forall x\in X) P(x)$ to have truth value
$1$ if $P(a)$ is true as $a$ ranges over every single member of the set $X$ and $0$ if $P(a)$ is false
for any $a\in X$. We call $\forall$ the \emph{universal quantifier} and read $(\forall x\in X) P(x)$ as
"for all $x$ in $X$, $P(x)$".

\vspace{.2in}

\begin{example} Let $P(x):="x<-3"$ with domain the set of real numbers $\mathbb{R}$. Let $X_1=\
mathbb{N}, X_2=\mathbb{Z}, X_3=\{-\pi\}$. Then the statement $(\exists x\in X_1) P(x)$ is false
since every natural number is non-negative; $(\exists x\in X_2) P(x)$ is true since $-10\in X_2$ with
$P(-10)$ being true; $(\exists x\in X_3) P(x)$ is also true since $-\pi\in X_3$ and $P(-\pi)$
is true. The statement $(\forall x\in X_1) P(x)$ is false since $0\in X_1$ and $P(0)$ is false; the
statement $(\forall x\in X_2) P(x)$ is false since $0\in X_2$ and $P(0)$ is false; the statement $(\forall
x\in X_3) P(x)$ is true however since there is only one element $-\pi$ of $X_3$ and $P(-\pi)$ is true.
\end{example}

\vspace{.2in}

If the domain of the predicate $P(x)$ say $U$ is clear and we have no interest in restricting the
predicate to a subset of the predicate's domain then we will often just write $(\exists x) P(x)$ and $(\
forall x) P(x)$ instead of $(\exists x\in U) P(x)$ and $(\forall x\in U) P(x)$. We can also extend the idea
of existential and universal quantification to include predicates $P(x_1,x_2,\dots, x_n)$ with more than
one variable ($n\geq 2$) each with respective domains $U_i$ as follows. Let $(\exists x_1) P(x_1,x_2,\
dots,x_n)$ simply be the obvious predicate with one less variable (namely variables $x_2,x_3,\dots,
x_n$).

Let $(\forall x_1) P(x_1,x_2,\dots,x_n)$ be the obvious predicate with one less variable. For example if
our predicate is $P(a,b,c):="a^2+b^2=c^2"$ with domain $U=\mathbb{Z}$ then $(\exists a) P(a,b,c)$
is a predicate with two variables $b,c$ and $(\exists a)(\exists b) P(a,b,c)$ is a predicate with one
variable $c$ and $(\exists a)(\exists b)(\exists c) P(a,b,c)$ is a statement, which happens to be true since
$P(3,4,5)$ is true for instance.

\vspace{.2in}

We will often deal with statements that involve a mix of existential and universal quantifiers and we
want to emphasize that in general the order in which the quantifiers are listed is paramount. For
instance if $P(a,b):="a<b"$ with domain $\mathbb{R}$ we can form eight different statements using
two quantifiers:

$$
S_1:=(\forall a)(\forall b) (a<b)
$$
$$
S_2:=(\forall b)(\forall a) (a<b)
$$
$$
S_3:=(\forall a)(\exists b) (a<b)
$$
$$
S_4:=(\forall b)(\exists a) (a<b)
$$
$$
S_5:=(\exists a)(\forall b) (a<b)
$$
$$
S_6:=(\exists b)(\forall a) (a<b)
$$
$$
S_7:=(\exists a)(\exists b) (a<b)
$$
$$
S_8:=(\exists b)(\exists a) (a<b)
$$

\vspace{.2in}

Note that $S_1$ and $S_2$ are saying the same thing; we will abbreviate these equivalent statements
by writing $(\forall a,b) (a<b)$. Similarly $S_7$ and $S_8$ are saying the same thing;we will
abbreviate these equivalent statements by writing $(\exists a,b) (a<b)$. $S_1, S_2$ are false whereas
$S_7,S_8$ are true. The rest are a bit more delicate. $S_3$ is exerting that given any real number $a$
we can find a bigger real number $b$, which is true (we could let $b=a+1$ for instance). If we switch
the order of these two quantifiers we get statement $S_6$ which claims that there is a real number $b$
that is bigger than every other real number $a$, clearly false statement ($b$ is not bigger than $a=b+1$
for instance). $S_4$ exerts that given any real number $b$ we can find smaller real number $a$, a true
statement as we can always set $a=b-1$. $S_5$ states that there is a real number $a$ that is less than
every other real number $b$, a false statement since we could look at $b=a-1$. Order matters most of
the time here unless all the quantifiers are of the exact same type.

\vspace{.2in}

Looking back at the connectives $\land$ and $\lor$ we see that they are somewhat complementary
when we mix them with the negation connective in the sense that we have DeMorgan's tautologies $\
neg ( P\land Q)\iff (\neg P \lor \neg Q)$ and $ \neg ( P\lor Q)\iff (\neg P \land \neg Q)$; negation
'switches' these two connectives. Something similar happens when we look at how the negation
connective works with universal and existential quantifiers. We note that given a predicate $P(x)$ we
have the following logical equivalences:

$$
[\neg(\forall x) P(x)]\equiv (\exists x) (\neg P(x))
$$
and
$$
[\neg(\exists x) P(x)]\equiv (\forall x) (\neg P(x))
$$

\vspace{.2in}

Let $Q(x,y,z):="x+y=z"$ where our domain is the integers $\mathbb{Z}$. Let's examine the statement
$(\forall x)(\forall z)(\exists y) Q(x,y,z)$ which states that given any two integers $x$ and $z$ we can
find a third integer $y$ so that $x+y=z$ (which happens to be true since we can let $y=z-x$ and the
$y$ defined this way is an integer). The negation of this statement, the statement $\neg (\forall x)(\
forall z)(\exists y) Q(x,y,z)$ is logically equivalent to each of the following statements:

$$
(\exists x)\neg(\forall z)(\exists y) Q(x,y,z)
$$
$$
(\exists x)(\exists z)\neg(\exists y) Q(x,y,z)
$$
and
$$
(\exists x)(\exists z)(\forall y) \neg Q(x,y,z)
$$

where the last statement is $(\exists x)(\exists z)(\forall y)(x+y\ne z)$.

\vspace{.4in}

A key idea when working with sets is that we can build new sets from already understood sets by
means of set operations of various types, several of which have a close relationship to the logical
connectives and quantifiers of predicate logic. Three of the most fundamental set operations are that of
union, intersection and set difference.

\vspace{.2in}

\begin{definition}Let $X,Y$ be sets. Define $X\cup Y:=\{a\,|\, a\in X \lor a\in Y\}$ the union of $X$
and $Y$ and $X\cap Y:=\{a\,|\, a\in X\land a\in Y\}$ the intersection of $X$ and $Y$. Note that
because $P\land Q \iff Q\land P$ and $P\lor Q\iff Q\lor P$ are tautologies, $X\cup Y=Y\cup X, X\cap
Y=Y\cap X$. We define $X-Y:=\{a\in X\,|\, a\notin Y\}$ and call this $X$ minus $Y$. The language
'the difference of $X$ and $Y$ tends to cause a bit of ambiguity, it's not clear whether this refers to $X-
Y$ or $Y-X$, and in general these two sets are not the same. $X-Y$ is a subset of $X$ while $Y-X$ is
a subset of $Y$.
\end{definition}

\vspace{.2in}

\begin{theorem}
Let $A,B,C$ be sets. Then $A\cup(B\cap C)=(A\cup B)\cap (A\cup C)$ and $A\cap (B\cup C)=(A\cap
B)\cup (A\cap C)$.
\end{theorem}

\vspace{.2in}

\begin{proof}
We prove the first equality $A\cup(B\cap C)=(A\cup B)\cap (A\cup C)$ and leave the proof of the
second inequality as a highly recommended exercise for the reader. Let $x\in A\cup(B\cap C)$. This
happens iff $(x\in A)\lor (x\in B\cap C)$ by the definition of union; this in turn is equivalent to $(x\in
A)\lor (x\in B\land x\in C)$ by the definition of intersection. Let $P:="x\in A", Q:="x\in B", R:="x\in
C".$ Then our statement is $P\lor (Q\land R)$. This is logically equivalent to $(P\lor Q)\land (P\lor R)$
or $(x\in A\lor x\in B)\land (x\in A\lor x\in C)$, which in turn is equivalent to $x\in (A\cup B)\cap (A\
cup C)$ by the definition of union and intersection. So we have shown that $x\in A\cup(B\cap C) \iff x\
in (A\cup B)\cap (A\cup C)$ and so $A\cup(B\cap C)=(A\cup B)\cap (A\cup C)$.
\end{proof}
\vspace{.2in}

We say a set $X$ is empty (or an empty set) if $\neg(a\in X)$ or $a\not\in X$ is always true. In other
words a set $X$ is empty if $(\forall a)(a\notin X)$ is true. By our remarks on quantifiers $(\forall a)(a\
notin X)\equiv \neg (\exists a)(a\in X)$.

\vspace{.2in}

\begin{theorem}[Empty Set Theorem]If $X$ and $Y$ are both empty sets then $X=Y$.
\end{theorem}

\vspace{.1in}

\begin{proof} Given empty sets $X$ and $Y$ we need to prove the implications $(a\in X)\implies (a\in
Y)$ and $(a\in Y)\implies (a\in X)$. The implication $(a\in X)\implies (a\in Y)$ is equivalent to its
contrapositive $\neg(a\in Y)\implies \neg(a\in X)$, the latter being a true implication. Similarly we see
that $(a\in Y)\implies (a\in X)$ is true. So we have that $X\subseteq Y$ and $Y\subseteq X$ by the
definition of subset and lastly that $X=Y$ by the definition of set equality.\end{proof}

\vspace{.2in}

The above theorem says that we can refer to \emph{the} empty set, and we will write $\emptyset$ or $\
{\}$ for this unambiguous set. We encourage the reader to provide a simple proof that given any set
$X$ we have $\emptyset\subseteq X$. Is there a 'largest' set? That's a murky question that gets into
various logical issues at foundations of mathematics...the short answer is no. The discussion of sets in
these notes avoids a lot of these important questions to favor a somewhat lighter introduction to the
study of sets. For a more in depth look at set theory, in particular various forms of \emph{axiomatic set
theory} we encourage the reader to click
\href{https://www.math.tamu.edu/~boas/courses/220-2003c/nov14.pdf}{HERE}. In a manner quite
similar to how we sometimes treated domains of various predicates, it's often convenient to restrict
ourselves to some 'universe' or 'universal set' if we are only focusing on objects of a particular type. We
often denote such sets by $U$ and make sure this set is very clearly mentioned or defined. This allows
us to establish another useful set operation.

\vspace{.2in}
\begin{definition}
Given a universal set $U$ and $X\subseteq U$ we define the complement of $X$ (with respect to $U$)
to be the set $X^c:=U-X$, that is, the set of all elements in the universal set not in the set $X$. We see
immediately that $U^c=\emptyset, \emptyset^c=U$ and that for any $X\subseteq U$ $(X^c)^c=X$.
\end{definition}

\vspace{.2in}

\begin{lemma}Let $A,B\subseteq U$. Then $A\subseteq B \iff B^c\subseteq A^c$.

\end{lemma}

\vspace{.1in}

\begin{proof} We note that the contrapositive of the implication $x\in A\implies x\in B$ is the
statement $x\notin B\implies x\notin A$, in other words:

$$
(x\in A\implies x\in B)\iff (x\in B^c \implies x\in A^c)
$$

which is exactly the statement of the lemma by definition of subset.\end{proof}

\vspace{.2in}

\begin{theorem}[DeMorgan's Theorem for sets] Let $A,B\subseteq U$. Then $(A\cap B)^c=A^c\cup
B^c$ and $(A\cup B)^c=A^c\cap B^c$.
\end{theorem}

\vspace{.1in}

We give two proofs of these two equalities.

\vspace{.1in}

\begin{proof}[Proof 1] Let $y\in(A\cup B)^c $. This happens iff $\neg(y\in A\cup B)$ by definition of
complement. In turn this happens iff $\neg(y\in A\lor y\in B)$ by definition of union. This is logically
equivalent to the statement $\neg(y\in A)\land \neg(y\in B)$ by DeMorgan's
tautology. This happens iff $y\notin A\land y\notin B$ which in turn happens iff $y\in A^c\cap B^c $
by definition of intersection and complement. Hence $y\in (A\cup B)^c \iff y\in A^c\cap B^c $ and so $
(A\cup B)^c=A^c\cap B^c$.

\vspace{.2in}

Let $y\in(A\cap B)^c $. This happens iff $\neg(y\in A\cap B)$ by definition of complement. In turn this
happens iff $\neg(y\in A\land y\in B)$ by definition of intersection. This is logically equivalent to the
statement $\neg(y\in A)\lor \neg(y\in B)$ by DeMorgan's tautology. This happens iff $y\notin A\lor y\
notin B$ which in turn happens iff $y\in A^c\cup B^c $ by definition of union and complement. Hence
$y\in (A\cap B)^c \iff y\in A^c\cup B^c $ and so $(A\cap B)^c=A^c\cup B^c$.\end{proof}

\vspace{.2in}

\begin{proof}[Proof 2] We see that $A\cap B\subseteq A,B\subseteq A\cup B$ and so by an earlier

lemma, $(A\cup B)^c\subseteq A^c, B^c\subseteq (A\cap B)^c$. From this we get that $(A\cup B)^c\
subseteq A^c\cap B^c$ and $A^c\cup B^c\subseteq (A\cap B)^c$. Replacing $A,B$ with $A^c, B^c$
respectively in the previous two statements we get $(A^c\cup B^c)^c\subseteq A\cap B$ and $A\cup B\
subseteq (A^c\cap B^c)^c$. We apply the lemma again to get that $(A\cap B)^c\subseteq A^c\cup B^c$
and $A^c\cap B^c\subseteq (A\cup B)^c$ and so we get $(A\cap B)^c=A^c\cup B^c$ and $(A\cup
B)^c=A^c\cap B^c$.\end{proof}

\vspace{.2in}

We can generalize the ideas of union and intersection of two sets to an arbitrary collection of sets.
Suppose we had

$$
S=\{\{1,2,3,4\},\{4,6\},\{1,2,4,6,7,8\},\{4,9\}\}.
$$

We could form a new set which we will denote $\cap S$ to be the intersection of all elements of $S$: $\
cap S=\{4\}$. We could also form another new set which we will denote $\cup S$ to the union of the
elements of $S$: $\cup S=\{1,2,3,4,6,7,8,9\}$. We make these definitions precise:

\vspace{.2in}

\begin{definition} Let $S$ be a collection of sets. We define the intersection of $S$, written $\cap S$,
as follows: $x\in\cap S$ if $\forall A\in S$, $x\in A$. We define the union of $S$, written $\cup S$, as
follows: $x\in\cup S$ if $\exists A\in S$, $x\in A$. We encourage the reader to state and prove more
general forms of DeMorgan's Theorems for union and intersection of collections of sets. We can think
of $A\cup B$ as $\cup\{A,B\}$ and $A\cap B$ as $\cap\{A,B\}$.
\end{definition}

\vspace{.2in}

\begin{example} Suppose we have

$$
S=\{\{1,2,3,4,5,6,7\},\{3,5,7,9,11\},\{4,5,6,7,8,9\}\}
$$. Then $\cap S=\{5,7\}$ since the elements $5$ and $7$ are the only elements common to all three
members of $S$. We see that $\cup S=\{1,2,3,4,5,6,7,8,9,11\}$.
\end{example}

\vspace{.2in}

\begin{example}
Let $X=\R$ be the universal set of real numbers, and let $T=\{(-\infty,1],(-1,2],[1,\infty)\}=\
{A_1,A_2,A_2\}$ be a collection of subsets of $X$. Then $\cap T=(-1,1]$ and $\cup T=\R$. If we look
at the collection of complements $T^c=\{A_1^c,A_2^c, A_3^c\}=\{(1,\infty),(-\infty,-1]\cup(2,\infty),
(-\infty, -1)\}$, we have $\cap T^c=\emptyset$ and $\cup T^c=(-\infty, -1]\cup (1,\infty)$. Note that $\
cap T^c=(\cup T)^c$ and $\cup T^c=(\cap T)^c$.
\end{example}

\vspace{.2in}

Another way to build new sets from old ones is to form ordered pairs or more generally for a given
positive integer $n>0$, ordered $n-tuples$.

\vspace{.2in}

\begin{definition} Let $A,B$ be sets. Then the (Cartesian) product of the sets $A\times B:=\{(x,y)\,|\,
x\in A, y\in B\}$ and more generally, if $A_1, A_2,\dots, A_n$ are sets, the $n-$fold (Cartesian)
product $A_1\times A_2\times\dots\times A_n:=\{(x_1,x_2,\dots, x_n)\, |\, x_1\in A_1, x_2\in A_2,\
dots x_n\in A_n\}.$
\end{definition}

\vspace{.2in}

\begin{example}

\end{example}
Let $A=\{x,1\}, B=\{s,\clubsuit\}$. Then $A\times B=\{(x,s),(x,\clubsuit),(1,s),(1,\clubsuit)\}$.
Furthermore:

$$
A\times B\times A=\{(x,s,x),(x,s,1),(x,\clubsuit,x),(x,\clubsuit, 1),
$$
$$
(1,s,x),(1,s,1),(1,\clubsuit, x),(1,\clubsuit, 1)\},
$$
while

$$
(A\times B)\times A=\{((x,s),x),((x,s),1),((x,\clubsuit),x),((x,\clubsuit), 1),
$$
$$
((1,s),x),((1,s),1),((1,\clubsuit), x),((1,\clubsuit), 1)\}.
$$

Since the ordered pair $ ((1,s),x)$, which is an element of $(A\times B)\times A$, is not an element of
$A\times B\times A$, we see that $(A\times B)\times A$ is not the same set as $A\times B\times A$.
We encourage the reader to write out the sets $B\times A$ and $A\times (B\times A)$ and compare
them to $A\times B$ and $(A\times B)\times A$ respectively to see that in general these are all
different sets.

\vspace{.2in}

\begin{example}
If we let $A_i=A$ for every single $1\leq i\leq n$ where $n\in\N^+$ we simple write $A^n$ for the
product $A_1\times\dots \times A_n$. For instance we write $\R^2$ for the set of ordered pairs of real
numbers, $\R^3$ for the set of triples $(x,y,z)$ of real number, etc...
\end{example}

\newpage

\section{Functions}

\vspace{.2in}

There are many different ways one set may be related to another. One of the most fundamental
relationships between two sets is that of a \emph{function}. We are a bit more rigorous with our
definition of function than one might have seen in a calculus course.

\vspace{.2in}

\begin{definition} A function $f: A\rightarrow B$ consists of three pieces of information: a set $A$,
the domain of the function, a set $B$, the codomain of the function, and an assignment of a unique
element of $B$ to each of the elements of the set $A$. We define $(\exists!x\in S) P(x)$ to mean $(\
exists x\in S) P(x)\land ((y\in S \land P(y))\implies x=y)$ where $P$ is a predicate statement. We can
write the definition of a function rule using quantifiers: $(\forall x\in A)\,(\exists! y\in B)$\,($y=f(x)$).
Sometimes we write $x\overset{f}\mapsto y$ and say $f$ maps the element $x\in A$ to the element
$y\in B.$ We often write $dom(f)=A, codom(f)=B.$ We define

$$
Y^X:=\{f:X\rightarrow Y\}
$$
to be the set of all functions with domain $X$ and codomain $Y$.
\end{definition}

\vspace{.2in}

\begin{example} Let $B=\{y\}$ be a set with a single element and suppose that $A$ is any set. There is
exactly one function from $A$ to $B$ since we would necessarily have $f(x)=y$ given $x\in A$.
\end{example}

\vspace{.2in}

\begin{example} Writing "$f(x)=x^2$" isn't enough to specify a function. Depending on the domain
and codomain we get can many different functions. Two functions are equal if they have the same
domain, the same domain, and the exact same rule. So for instance $h\,:\, \mathbb{R}\rightarrow [0,\
infty)$ given by $h(x)=x^2$ is a different function than $g\,:\, \mathbb{R}\rightarrow\mathbb{R}$
given by $g(x)=x^2$.
\end{example}

\vspace{.2in}

Note that given any function $f\,:\, A\rightarrow B$ we can define a subset of $A\times B$ given by $\
{(x,f(x))\,|\, x\in A\}\subseteq A\times B$ which we call the \emph{graph} of $f$, and we write
$graph(f)$ for this subset of $A\times B$. Often times we blur the distinction between a function and
its graph, but in general if we only knew the graph of the function we would not know its codomain.
For instance if we were told that a function had graph $\{(x,x^2)\,|, x\in \mathbb{R}\},$ we'd know the
domain of the function was $\mathbb{R}$ and what the rule defining the function is, but it is not clear
what the codomain of the function is with just the graph. Both functions $h,g$ in the previous example
have this set as their graphs.

\vspace{.2in}

\begin{definition} Given any set $X$ we can define a function that has domain and codomain $X$
called the identity function on $X$ and written $id_X:,X\rightarrow X$ with rule $id_X(a)=a$ for
every $a\in X$. Note that if $X\not=Y$ then $id_X\not= id_Y$ even though they have the 'same' rule.
\end{definition}

\vspace{.2in}

We can think of an infinite sequence $(a_k)_{k=0}^{\infty}$ where each $a_k\in X$ for some set $X$
as simply a function with domain $\mathbb{N}$ and codomain $X$ by letting $a:\mathbb{N}\
rightarrow X$ be given by $a(k)=a_k$ for each $k\in\mathbb{N}$. A finite sequence of elements of
$X$ can be defined to be a function with domain $\{0,1,2\dots, n\}$ and codomain $\mathbb{N}$,
namely $f:\{0,1,2\dots, n\}\rightarrow X$ gives us the sequence $(f(0),f(1),f(2),\dots, f(n))$ of length
$n+1$.

\vspace{.2in}

\begin{definition} A \emph{binary operation} $p$ on a set $X$ is simply a function $p\,:\,X\times X\

rightarrow X$. Likewise an $n$-ary operation $q$ on a set $X$ where $n\in\mathbb{N}^+$ is a
function $q\,:\, X^n\rightarrow X$. We extend our previous definition of product to include the product
of $0$ copies of $X$ and make $X^0:=\{()\}$ the set which contains the empty tuple. Then we can
define a $0$-ary operation $q$ on a set $X$ as a function $q:X^0\rightarrow X$; since the domain of
such an operation consists of a single element, we can identify $q$ with the element $q(())\in X$ and
think of $0$-ary operations as simply elements of $X$.
\end{definition}

\vspace{.2in}

\begin{example} $+\,:\,\mathbb{Z}\times\mathbb{Z}\rightarrow\mathbb{Z}$ given by $+(a,b)=a+b$

is a binary operation; so is subtraction and multiplication. Note that the function $+:\N\times \N\
rightarrow\N$ given by $+(a,b)=a+b$ is a binary operation on $\N$ but that subtraction is not; $1-3=-
2\not\in\N$. We'll address this later in this section when we discuss \emph{partial functions}.
\end{example}

\vspace{.2in}

\begin{example}
There are lots of natural binary operations, but where do higher order operations show up? Let $T:\R^2\
times \R^2\times \R^2\rightarrow \R^2$ be given by

$$
T((x_1,y_1),(x_2,y_2),(x_3,y_3))=T(A,B,C)=A-B+C=(x_1-x_2+x_3,y_1-y_2+y_3).
$$

If $T(A,B,C)=D$ then $A-B+C=D$ or $A-B=D-C$ tells us that the lengths of the vectors $A-B$ and
$C-D$ are the same, and we encourage the reader to draw a picture of the points $A,B,C,D$ to verify
that they form a parallelogram. If $A,B,C$ are not collinear the parallelogram $A,B,C,T(A,B,C)$ has
non-zero area.

\end{example}

We can
 build new functions out of previously defined functions in several ways, the most important being that
of composition of functions.

\vspace{.2in}

\begin{definition} Let $f: A\rightarrow B, g:C\rightarrow D$ where $B\subseteq C$, in other words we
have one function whose codomain is a subset of the domain of another function. We form $g\circ f\, :\,
A\rightarrow D$ by the rule $g\circ f(a)=g(f(a))$ for each $a\in A$. We read this as $f$ followed by
$g$. Since $im(f)\subseteq B$, we extend the above definition of composition to include any functions
$f:A\rightarrow B, g:C\rightarrow D$ where $im(f)\subseteq C$.
\end{definition}
\vspace{.2in}

\begin{example} Let $f\, :\, \mathbb{R}\rightarrow \mathbb{R}$ be given by $f(x)=e^x$ and $g\,:\,
[0,\infty)\rightarrow [0,\infty)$ given by $g(x)=\sqrt{x}$. Then we can do $f$ first followed by $g$
since $im(f)=(0,\infty)\subseteq [0,\infty)=dom(g)$ and get $g\circ f\,:\, \mathbb{R}\rightarrow [0,\
infty)$ by $g\circ f(x)=g(f(x))=g(e^x)=e^\frac{x}{2}.$ Note we can also compose these functions in
the opposite order since $im(g)=[0,\infty)\subseteq \mathbb{R}=dom(f)$ to get $f\circ g\,:\, [0,\infty)\
rightarrow \mathbb{R}$ by $f\circ g(x)=f(g(x))=f(\sqrt{x})=e^{\sqrt{x}}$. Not only are these rules
different, but the domains and codomains of these functions are not the same: $f\circ g\not=g\circ f$ in
general.
\end{example}

\vspace{.2in}

Let $T:\{1,2,3,4\}\rightarrow \{0,1,2,3,4\}$ be given by the arrow diagram:

$$
T\quad\begin{tikzcd}[column sep=1em,row sep=3ex]
1 \arrow[drr] & 2\arrow[d] & 3 \arrow[dll] & 4 \arrow[dl] \\
0 & 1 & 2 & 3 & 4\end{tikzcd}
$$

and let $S:\{0,1,2,3\}\rightarrow \{0,1,2,3\}$ be given by the arrow diagram:

$$
S\quad \begin{tikzcd}[column sep=1em,row sep=3ex]
0 \arrow[dr] & 1\arrow[dr] & 2 \arrow[dr] & 3 \arrow[d] \\
0 & 1 & 2 & 3 \end{tikzcd}
$$

Since $im(T)=\{0,1,2\}\subseteq dom(S)$ we can form $S\circ T$, $T$ followed by $S$, simply by
stacking the arrow diagrams and then following the arrows from top to bottom:

$$
\begin{tikzcd}[column sep=1em,row sep=3ex]
1 \arrow[drr] & 2\arrow[d] & 3 \arrow[dll] & 4 \arrow[dl]\\
0 \arrow[dr] & 1\arrow[dr] & 2 \arrow[dr] & 3 \arrow[d] & 4\\
0 & 1 & 2 & 3\end{tikzcd}
$$

so $S\circ T$ has arrow diagram

$$
S\circ T\quad \begin{tikzcd}[column sep=1em,row sep=4ex]
1 \arrow[drrr]& 2\arrow[dr] & 3\arrow[dl] & 4\arrow[d]\\
0 & 1 & 2 & 3\end{tikzcd}
$$

Note also that $im(S)=\{1,2,3\}\subseteq dom(T)$ and so we can form the composition $T\circ S$, $S$
follows by $T$ by the same process of stacking the arrow diagrams and recording the result. First we
stack:

$$
\begin{tikzcd}[column sep=1em,row sep=3ex]
0 \arrow[dr] & 1 \arrow[dr]& 2\arrow[dr] & 3\arrow[d]\\
0 & 1\arrow[dr] & 2\arrow[dl] & 3\arrow[dlll] & 4\arrow[dll]\\
0 & 1 & 2 & 3 & 4\end{tikzcd}
$$

and then recording the arrows from top to bottom we get

$$
T\circ S\quad \begin{tikzcd}[column sep=1em,row sep=3ex]
0\arrow[drr] & 1\arrow[d] & 2\arrow[dll] & 3\arrow[dlll]\\
0 & 1 & 2 & 3 & 4\end{tikzcd}
$$

\vspace{.2in}

\begin{theorem} Function composition is associative, namely given $f\,:\, A\rightarrow B, g\,:\,C\

rightarrow D, h\,:\, E\rightarrow F$ where $im(f)\subseteq C, im(g)\subseteq E$ then the functions $h\
circ(g\circ f)$ and $(h\circ g)\circ f$ from $A$ to $F$ are the same function.
\end{theorem}

\vspace{.2in}

\begin{proof} The two functions have the same domain and codomain so we check the rule: let $x\in
A$. Then $h\circ (g\circ f)(x)=h(g\circ f(x))=h(g(f(x))=(h\circ g)(f(x))=(h\circ g)\circ f(x)$, same rule.\
end{proof}

\vspace{.2in}

\begin{example} Let $\alpha\,:\, \mathbb{R}\rightarrow \mathbb{R}$ be $\alpha(x)=2x-1$. If we let

$\beta\,:\, \mathbb{R}\rightarrow\mathbb{R}$ be given by $\beta(x)=\frac{x+1}{2}$ then notice $\
beta\circ \alpha(x)=\beta(\alpha(x))=\beta(2x-1)=\frac{(2x-1)+1}{2}=x=id_{\mathbb{R}}(x)$ and $\
alpha\circ \beta (x)=\alpha(\beta(x))=\alpha(\frac{x+1}{2})=2(\frac{x+1}{2})-1=x=id_{\mathbb{R}}
(x)$ and so we see that $\beta\circ\alpha=id_{\mathbb{R}}=\alpha\circ\beta$. We say that $\alpha$ has
a two-sided inverse, a function that when we compose on the left or the right we get an identity
function. Note that this depends a lot on the sets involved and not just the rule. If $\alpha'\,:\,\
mathbb{Z}\rightarrow \mathbb{Z}$ is given by $\alpha'(x)=2x-1$ then just writing down an analog of
$\beta$ as the rule $\frac{x+1}{2}$ would not be well defined. But if we let

\begin{equation*}
\beta(x)=\left\{\begin{array}{cc}
\frac{x+1}{2} & \quad \exists y\in\mathbb{Z},x=2y-1\\
0 & \quad else\end{array}
\right.
\end{equation*}

Then $\beta(\alpha(x))=\beta(2x-1)=\frac{(2x-1)+1}{2}=x=id_{\mathbb{Z}}(x)$. But one readily sees

that if we compose these functions in the other order we do not get an identity function, for instance $\
alpha(\beta(4))=\alpha(0)=-1\not=id_{\mathbb{Z}}(4)$. We say that $\beta$ is a left inverse for $\
alpha$ but not a right inverse. We will make these terms precise in a moment, but first we introduce
some nice properties of functions which we will connect to the ideas of 'inverses'.

\end{example}

\vspace{.2in}

\begin{definition} We say that $f:A\rightarrow B$ is injective or 1-1 (these are synonyms) if given
$x_1\not= x_2$ elements of $A$ it follows that $f(x_1)\not=f(x_2).$ Equivalently if $f(x_1)=f(x_2)$
then $x_1=x_2$ (why are these statements equivalent?).
\end{definition}

\vspace{.2in}

Looking back at $\alpha\,:\,\mathbb{R}\rightarrow\mathbb{R}$ given by $\alpha(x)=2x-1$ we see that

if $\alpha(x_1)=\alpha(x_2)$ then $2x_1-1=2x_2-1$ certainly implies $x_1=x_2$ and so $\alpha$ is 1-
1. Likewise $\alpha'$ above is 1-1 by a similar argument.

\vspace{.2in}

\begin{definition} We say that $f\,:\, A\rightarrow B$ is surjective or onto (these are synonyms) if
$im(f)=B$, in other words $\forall y\in B\,\, \exists x\in A$ so that $f(x)=y$.
\end{definition}

\vspace{.2in}

\begin{theorem} Let $f:A\rightarrow B$ and $g:B\rightarrow C$. If $f$ and $g$ are $1-1$ then so is
$g\circ f$. If $f$ and $g$ are onto then so is $g\circ f$.

\end{theorem}

\begin{proof}
Suppose that $f,g$ are $1-1$ and that we have $g\circ f(x_1)=g\circ f (x_2)$ for some $x_1,x_2\in A$.
Then $g(f(x_1))=g(f(x_2))$ implies that $f(x_1)=f(x_2)$ since $g$ is $1-1$. But then $x_1=x_2$ since
$f$ is $1-1$. So we have shown $g\circ f(x_1)=g\circ f (x_2)\implies x_1=x_2$ and so $g\circ f$ is $1-
1$.

\vspace{.2in}

Now suppose that $f,g$ are both onto and let $z\in C$. Since $g$ is onto there exists $y\in B$ so that
$g(y)=z$. Since $f$ is onto there exists $x\in A$ so that $f(x)=y$. But then $g\circ
f(x)=g(f(x))=g(y)=z$ and so $g\circ f$ is onto.
\end{proof}

\vspace{.2in}

Looking back at $\alpha\,:\,\mathbb{R}\rightarrow\mathbb{R}$ given by $\alpha(x)=2x-1$ we see that

if $y\in\mathbb{R}=codom(\alpha)$ then $\exists x\in\mathbb{R}=dom(\alpha)$ so that $\
alpha(x)=y$, namely $x=\frac{y+1}{2}$ works ($\alpha(\frac{y+1}{2})=y$) and so $\alpha$ is onto.
On the other hand $\alpha'$ is not onto: there is no $x\in\mathbb{Z}$ so that $\alpha'(x)=0$ since this
amounts to the equation $2x-1=0$ having a solution that is an integer.

\vspace{.2in}

We saw that $\alpha$ had a two-sided inverse $\beta$ while $\alpha'$ only had an inverse on one side.
We formalize these concepts and then connect them to the properties of being injective/surjective.

\vspace{.2in}

\begin{definition} Let $f\, :\, A\rightarrow B$. We say that $L\, :\, B\rightarrow A$ is a \emph{left
inverse} of $f$ if $L\circ f=id_A$. We say that $R\, :\, B\rightarrow A$ is a \emph{right inverse} of
$f$ if $f\circ R=id_B$.
\end{definition}

\vspace{.2in}

\begin{theorem} Suppose a function $f:\rightarrow B$ has both a left inverse say $L:B\rightarrow A$
and a right inverse $R: B\rightarrow A$ then $R=L$.
\end{theorem}

\vspace{.2in}

\begin{proof}e have $R=(id_A)\circ R=(L\circ f)\circ R=L\circ (f\circ R)=L\circ id_B=L$ by the
associativity of function composition.\end{proof}

\vspace{.2in}

\begin{theorem}[Left Inverse Theorem] Let $f: A\rightarrow B$ be a function where $A\not=\

emptyset$. Then $f$ is injective iff $f$ has a left inverse.
\end{theorem}
\vspace{.2in}

\begin{proof} Assume $f$ is 1-1 and take $a_0\in A$. Define $g:B\rightarrow A$ by

\begin{equation*}
g(b)=\begin{cases}
a \quad &\text{if} \, b\in im(f),\quad b=f(a) \\
a_0 \quad &\text{if} \, x \notin im(f) \\
\end{cases}
\end{equation*}

Then we first note that $g$ is well-defined since for each $b\in im(f)$ there is a unique $a\in A$ so that
$f(a)=b$ since $f$ is 1-1. One easily checks that $g\circ f=id_A$ and so $g$ is a left inverse.

\vspace{.2in}

Now let's assume $f$ has a left inverse, call it $h: B\rightarrow A$ so that $h\circ f=id_A$ and assume
that $f(x_1)=f(x_2)$ for $x_1,x_2\in A$. Then applying $h$ to the latter equation we get
$x_1=h(f(x_1))=h(f(x_2))=x_2,$ showing that $f$ is injective. \end{proof}

\vspace{.3in}

\begin{example}
Let's look at an example of finding a left inverse. Let $P:\{1,2,3,4\}\rightarrow \{a,b,c,d,e, f\}$ be
given by the arrow diagram

$$
\begin{tikzcd}[column sep=1em,row sep=2ex]
1 \arrow[d] & 2 \arrow[dr] & 3 \arrow[dr] & 4 \arrow[dr] & \, & \, \\
a&b&c&d&e&f
\end{tikzcd}
$$

It's clear that $P$ is $1-1$; no two arrows point at the same target. In building a left inverse we want to
reverse any existing arrows to get:

$$
\begin{tikzcd}[column sep=1em,row sep=2ex]
a \arrow[d] & b & c \arrow[dl] & d \arrow[dl] & e \arrow[dl] & f\\
1 & 2 & 3 & 4 & \,\end{tikzcd}
$$

The latter is not a function, there are no arrows leaving $b$ and $f$. In order for us to have a function
$Q:\{a,b,c,d,e,f\}\rightarrow\{1,2,3,4\}$ we need to simply define $Q(b)$ and $Q(f)$; any choices will
work, for instance one possible $Q$ is

$$
\begin{tikzcd}[column sep=1em,row sep=2ex]
a \arrow[d] & b\arrow[d] & c \arrow[dl] & d \arrow[dl] & e \arrow[dl] & f \arrow[dll]\\
1 & 2 & 3 & 4 & \,\end{tikzcd}
$$

One
easily checks that for all $x\in \{1,2,3,4\}$ we have $Q(P(x))=x$ so that $Q\circ P=id_{\{1,2,3,4\}}$.
On the other hand, if we look at $P(Q(b))=P(2)=c\not=b$ we see that it's false that $P\circ Q=id_{\
{a,b,c,d,e,f\}}$. Note that the way we built $Q$ was to take any existing arrows from $P$ and reverse
them, and then we added additional arrows in any way we please to ensure that $Q$ is a function.
\end{example}

\vspace{.2in}

\begin{theorem}[Right Inverse Theorem] Let $f\, :\, A\rightarrow B$ be a function. Then $f$ is
surjective iff $f$ has a right inverse.
\end{theorem}

\vspace{.2in}

\begin{proof} Assume that $f$ is surjective. We build a right inverse $p$ as follows: for each $b\in B$
choose an element $a\in A$ so that $f(a)=b$ (there might be many but there is at least one since $f$ is
onto). Define $p(b)=a$ and note that for each $b\in B, f\circ p(b)=f(p(b))=f(a)=b=id_B(b)$, showing
that $p$ is a right inverse of $f$.

\vspace{.2in}

Let's assume that $f$ has a right inverse $q$. Let $b\in B$ and look at $a:=q(b)$. We see that
$f(a)=f(q(b))=b$ and so $f$ is onto as desired. \end{proof}

\vspace{.2in}

\begin{example}
Let's look at an example of finding a right inverse. Let $T:\{1,2,3,4,5\}\rightarrow \{a,b,c\}$ be given
by the arrow diagram

$$
\begin{tikzcd}[column sep=1em,row sep=2ex]
1 \arrow[dr] & 2\arrow[d] & 3 \arrow[dr] & 4 \arrow[dl] & 5 \arrow[dl] \\
\, & a & b & c & \,\end{tikzcd}
$$

$T$ is clearly onto. We'll build a right inverse $S:\{a,b,c\}\rightarrow\{1,2,3,4,5\}$ by going through
the set $\{a,b,c\}$ and for each element picking one of the $T$ arrows and reversing it. One such $S$
is given by

$$
\begin{tikzcd}[column sep=1em,row sep=2ex]
\, & a \arrow[d] & b\arrow[dr] & c \arrow[dr] & \,\\
1 & 2 & 3 & 4 & 5\end{tikzcd}
$$
One checks that for each $x\in\{a,b,c\}$ we have $T(S(x))=x$ so that $T\circ S=id_{\{a,b,c\}}$. Since
$S(T(3))=S(c)=5\not=3$ we do not have $S\circ T=id_{\{1,2,3,4,5\}}$ in this case.

\end{example}

In general finding inverses of any type is tricky business. For instance looking at $f:\R\rightarrow \R$
given by $f(x)=x^3+100\sin(x)$, it's not hard to see that since $f$ is continuous and $\lim_{x\to\pm\
infty}f(x)=\pm \infty$, the intermediate value theorem tells us $im(f)=\R$ and $f$ is onto and hence
has a right inverse. Such a right inverse $g:\R\rightarrow\R$ would have to satisfy $f(g(x))=x$, and so
finding an explicit formula for $g$ would involve finding a solution $y=g(x)$ to the equation
$y^3+100\sin(y)=x$ which is a nightmare. But nonetheless we know at least one such solution exists
by the Right Inverse Theorem. $f$ is not $1-1$ as $f(x)=0$ has many solutions as one can verify by
typing 'solve $x^3+100\sin(x)=0$' into www.wolframalpha.com . By the Left Inverse Theorem no such
left inverse of $f$ exists.

\vspace{.2in}

One might have noticed that the left inverse $q$ of the $1-1$ function $p$ is onto, and the right inverse
$S$ of the onto function $T$ is $1-1$. This is always the case and is fairly immediate from the Left and
Right Inverse Theorems which we leave as an exercise to the reader:

\begin{corollary}
Let $f:A\rightarrow B$ have left inverse $g:B\rightarrow A$. Then $g$ is onto. Suppose $h:B\
rightarrow A$ is a right inverse of $f$. Then $h$ is $1-1$
\end{corollary}

\vspace{.2in}

\begin{definition} A function $f:A\rightarrow B$ that is 1-1 and onto is called a bijection or is said to
be a 1-1 correspondence between $A$ and $B$.
\end{definition}

\vspace{.2in}

\begin{example}
Let $X,Y,Z$ be sets. We show that $f:(X\times Y)\times Z\rightarrow X\times(Y\times Z)$ given by
$f((a,b),c)=(a,(b,c))$ is a bijection. Suppose that $t\in X\times (Y\times Z)$. Then $\exists a\in A, b\in
B, c\in C$ so that $t=(a,(b,c))$. But then $f((a,b),c)=t$ shows that $f$ is onto. If $f(t_1)=f(t_2)$ for
$t_1,t_2\in (X\times Y)\times Z$ then $t_i=((a_i,b_i),c_i)$ for $i=1,2$ and the condition
$f(t_1)=f(t_2)$ is simply that $(a_1,(b_1,c_1))=(a_2,(b_2,c_2))$. But then $a_1=a_2 $ and $
(b_1,c_1)=(b_2,c_2)$ which implies $b_1=b_2, c_2=c_2$ and hence $t_1=t_2$. So $f$ is a bijection.
\end{example}

\vspace{.2in}

Based on the above theorems we know immediately that any function that is a bijection has a unique
inverse, and vice a versa, any function that has both a left and right inverse not only has these one sided
inverses equal, but it must be a bijection. If $f: A\rightarrow B$ is a bijection we write $f^{-1}:B\
rightarrow A$ for the inverse of $f$.

\vspace{.2in}

\begin{definition} A bijection $p:X\rightarrow X$ from a set to itself is called a \emph{permutation of

$X$}. We let $\Sigma_X$ be the set of all permutations of $X$. $id_X\in \Sigma_X$ and if $f,g\in \
Sigma_X$ then $f^{-1}\in\Sigma_X$ and $f\circ g\in\Sigma_X$.

\end{definition}

\vspace{.2in}

\begin{example}
Let $X=\{1,2,3\}$. We first figure out the size of the set $\Sigma_X$. For any permutation $p:X\
rightarrow X$ there are $3$ choices for $p(1)$, and since $p$ is $1-1$ this leaves $2$ choices for
$p(2)$, and $1$ choice for $p(3)$. Thus there are a total of $6=3\cdot 2\cdot 1$ permutations of $X$.
It's not hard to see that if $X$ has $n$ elements the set $\Sigma_X$ has size $n!$. We already know
one of these $6$ elements, $id_X$, so let us find the other $6$.

\vspace{.1in}

Let $\sigma: X\rightarrow X$ be the permutation of $X$ given by the arrow diagram:

$$
\sigma\quad \begin{tikzcd}[column sep=1em,row sep=2ex]
1\arrow[dr] & 2\arrow[dr] & 3\arrow[dll]\\
1 & 2 & 3\end{tikzcd}
$$

If we compose $\sigma$ with itself we first get $\sigma\circ\sigma:=\sigma^2$:

$$
\begin{tikzcd}[column sep=1em,row sep=2ex]
1\arrow[dr] & 2\arrow[dr] & 3\arrow[dll]\\
1\arrow[dr] & 2\arrow[dr] & 3\arrow[dll]\\
1 & 2 & 3\end{tikzcd}
$$

and so

$$
\sigma^2\quad \begin{tikzcd}[column sep=1em,row sep=2ex]
1\arrow[drr] & 2\arrow[dl] & 3\arrow[dl]\\
1 & 2 & 3\end{tikzcd}
$$

Note that $\sigma^2=\sigma^{-1}$ by comparing their arrow diagrams. This tells us that $\sigma\circ\
sigma^2:=\sigma^3$ is simply $id_X$. So we have three permutations so far: $id_X,\sigma, \
sigma^2$. Let $\tau$ be the permutation of $X$ given by the arrow diagram:

$$
\tau\quad\begin{tikzcd}[column sep=1em,row sep=2ex]
1\arrow[dr] & 2\arrow[dl] & 3\arrow[d]\\
1 & 2 & 3\end{tikzcd}
$$

We see immediately that $\tau\circ \tau:=\tau^2=id_X$. Let's look at $\sigma\circ \tau$ by stacking and
composing arrow diagrams:

$$
\begin{tikzcd}[column sep=1em,row sep=2ex]
1\arrow[dr] & 2\arrow[dl] & 3\arrow[d]\\
1\arrow[drr] & 2\arrow[dl] & 3\arrow[dl]\\
1 & 2 & 3\end{tikzcd}
$$

gives us

$$
\sigma\circ\tau\quad\begin{tikzcd}[column sep=1em,row sep=2ex]
1\arrow[d] & 2\arrow[dr] & 3\arrow[dl]\\
1 & 2 & 3\end{tikzcd}
$$

Lastly if we compose $\tau$ and $\sigma$ in the other order to get $\tau\circ\sigma$ we leave it to the
reader to see that one gets

$$
\tau\circ\sigma\quad\begin{tikzcd}[column sep=1em,row sep=2ex]
1\arrow[drr] & 2\arrow[d] & 3\arrow[dll]\\
1 & 2 & 3\end{tikzcd}
$$

Note that each of $\tau, \sigma\circ\tau, \tau\circ\sigma$ are distinct from each other and distinct from
$id_X,\sigma, \sigma^2$, and so altogether we have

$$
\Sigma_X=\{id_X, \sigma, \sigma^2,\tau, \tau\circ\sigma, \sigma\circ\tau\}.
$$
\end{example}

\vspace{.2in}

\vspace{.2in}
The set $\Sigma_X$ with the operation of function composition is an example of a \emph{group}; it is
a set with an associative binary operation that has an identity element and an inverse for each element.
We'll return to our discussion of permutations when we discuss Fermat's Little Theorem in the section
on Number Theory.
\vspace{.2in}

We give a few examples of these concepts in the context of linear algebra and recall some basic facts
about linear transformations (feel free to skip this if you have no background in linear algebra of
course). Let $L=L_A\,:\, \mathbb{R}^n\rightarrow \mathbb{R}^m$ be a linear map given by left
multiplication of an $m\times n$ real matrix $A$ on a column vector $v\in\mathbb{R}^n$, namely
$L(v)=Av$. $L$ is onto exactly when $im(L)=\mathbb{R}^m$ or $rank(A)=dim(im(L))=m$. We also
see that $L$ is 1-1 iff it has trivial kernel: suppose $L$ is 1-1, and $v\in ker(L)$. Then $0=L(0)=L(v)$
implies that $v=0$ and if $ker(L)=\{0\}$ and $L(v_1)=L(v_2)$ then $L(v_1-v_2)=0$ and so $v_1-
v_2\in ker(L)$ and so $v_1=v_2$. By the Rank-Nullity Theorem we can then say $L$ is 1-1 iff
$rank(A)=n$. Now we are ready for some nice examples.

\vspace{.2in}

\begin{example} Let $A=\begin{pmatrix}1 & 2 & 0\\

0 & 1 & 1\end{pmatrix}$, which clearly has $rank(A)=2$ and so the associated linear map $L_A\,:\, \
mathbb{R}^3\rightarrow \mathbb{R}^2$ is onto but not 1-1. Thus $L_A$ has a right inverse but no
left inverse. Let's try to find a right inverse, say $L_B\, :\, \mathbb{R}^2\rightarrow \mathbb{R}^3$
given by some $3\times 2$ matrix $B$. If $L_A\circ L_B=L_{I_2}=id_{\mathbb{R}^2}$ we'd have
$v=L_A(L_B(v))=L_A(Bv)=ABv$. We are looking for a matrix $B$ so that

$$
\begin{pmatrix}1 & 2 & 0\\
0 & 1 & 1\end{pmatrix}\begin{pmatrix} a & b \\
c & d\\
e & f\end{pmatrix}=\begin{pmatrix}a+2c & b+2d \\
c+e & d+f \end{pmatrix}=\begin{pmatrix}1 & 0 \\
0 & 1 \end{pmatrix}
$$

and there are lots of possibilities, one being to set $c=e=0,a=1, b=d=0, f=1$ (call this $B$). We can
easily confirm that no left inverse $L_C$ exists, since if there was such a linear map we would have
$L_C\circ L_A(v)=(CA)v$ for each $v\in \mathbb{R}^3$ and $CA=I_3$. But $rank(I_3)=3$ while
$rank(C)\leq 2$ forces $rank(CA)\leq 2$. One sees what happens if we multiply $BA$:

$$
\begin{pmatrix} 1 & 0 \\
0 & 0\\
0 & 1\end{pmatrix}\begin{pmatrix}1 & 2 & 0\\
0 & 1 & 1\end{pmatrix}=\begin{pmatrix}1 & 2 & 0\\
0 & 0 & 0\\
0 & 1 & 1\end{pmatrix}
$$

which is a rank two matrix not even close to the identity matrix.
\end{example}

\vspace{.2in}

\begin{example} Let $M=\begin{pmatrix}1 & -1 \\

0 & 1 \\
1 & 0 \\
1 & 1\end{pmatrix}$ which clearly has $rank(M)=2$
and so the associated linear map $L_M\,:\, \mathbb{R}^2\rightarrow \mathbb{R}^4$ is 1-1 but not
onto. So $L_M$ has a left inverse but no right inverse, let's find try to find a left inverse say $L_N\,:\, \
mathbb{R}^4\rightarrow \mathbb{R}^2$ where $N$ is $2\times 4$. We would need $NM=I_2$ or

$$
\begin{pmatrix}a & b & c & d\\
e & f & g & h\end{pmatrix}\begin{pmatrix}1 & -1 \\
0 & 1 \\
1 & 0 \\
1 & 1\end{pmatrix}=\begin{pmatrix}
a+c+d & -a+b+d\\
c+g+h & -e+f+h\end{pmatrix}=\begin{pmatrix}1 & 0\\
0 & 1\end{pmatrix}
$$

Again there are infinitely many possibilities for $N$, one being $a=1, b=-1,c=d=0, h=1, g=-1, e=f=0$
(call this matrix $N$). Order really does matter here, $MN$ is

$$
\begin{pmatrix}1 & -1 \\
0 & 1 \\
1 & 0 \\
1 & 1\end{pmatrix}\begin{pmatrix}1 & -1 & 0 & 0\\
0 & 0 & -1 & 1\end{pmatrix}=\begin{pmatrix}1 & -1 & 1 & -1\\
0 & 0 & -1 & 1\\
1 & -1 & 0 & 0\\
1 & -1 & -1 & 1\end{pmatrix}
$$

which is a far cry from $I_4$.

\vspace{.2in}
\end{example}

It's fairly easy to find left and right inverses of linear transformations, but we remind the reader that
finding inverses in general can be pretty horrible.

\vspace{.2in}

\begin{example} Let $p\,:\, \mathbb{R}\rightarrow \mathbb{R}$ be given by $p(x)=x^{11}+5x-\

arctan(x)$. We know that $p'(x)=11x^{10}+5-\frac{1}{1+x^2}$ and that $\frac{1}{1+x^2}\leq 1$ so
that $p'(x)\geq 4>0$ and so $p$ is strictly increasing. By a homework exercise this means that $p$ is 1-
1. We know $\lim_{x\rightarrow \pm \infty}p(x)=\pm\infty$ and $p$ is continuous and so by the
Intermediate Value Theorem we see that $p$ is onto. So $p$ is a bijection and so $p^{-1}\,:\, \
mathbb{R}\rightarrow \mathbb{R}$ exists. Try finding a formula for this and see what trouble you run
into.
\end{example}

\vspace{.2in}

If we are working with a function $f\,:\, A\rightarrow B$ we can form a new function $f^*\,:\, A\
rightarrow im(f)$ given by the same rule $f^*(x)=f(x)$ for any $x\in A$. The function $f^*$ here is
always onto by its very construction. We can also take a function and restrict it to a subset of the
domain as follows:

\vspace{.2in}

\begin{definition} Let $f\,:\, A\rightarrow B$ and let $X\subseteq A$. Define $f|_X\,:\, X\rightarrow
B$ to be the function with rule $f|_X(t)=f(t)$ for all $t\in X$. We call $f|_X$ the restriction of $f$ to
the subset $X$.
\end{definition}

\vspace{.2in}

\begin{example} Let $g\,:\, \mathbb{Z}\rightarrow \mathbb{Z}$ be given by $g(n)=n^2$. It's not hard
to see that $g$ is neither 1-1 nor onto. The function $h=g|_{\mathbb{N}}$ is 1-1 since if
$h(n_1)=h(n_2)$ then $n_1^2=n_2^2$ and so $0=n_1^2-n_2^2=(n_1-n_2)(n_1+n_2)$ and so either
$n_1=n_2$ or $n_1=-n_2$. But since $n_1,n_2\in\mathbb{N}$ the only way $n_1=-n_2$ could
happen is if $n_1=0=n_2$. Either way $n_1=n_2$ and $h$ is 1-1.
\end{example}

\vspace{.2in}

\begin{example} In the previous example we saw that $g$ was not 1-1 onto. We can put the elements
$y\in\mathbb{Z}$ the codomain of $g$ into one of three categories. There are elements like $y=4$ that
have more than one element $x$ in the domain so that $g(x)=y$, for example $g(-2)=4=g(2)$. These
are the perfect squares $y\in\{1,4,9,16,\dots\}$. There are elements like $y=0$ that have exactly one
element $x$ in the domain so that $g(x)=y$, in this case only $x=0$ gives $g(x)=0$. And there are
elements $y$ that aren't in the image of $g$ at all, integers like $-5$ or $7$.
\end{example}

\vspace{.2in}

\begin{example}
The function $s:\R\rightarrow [-1,1]$ given by $s(x)=\sin(x)$ is certainly onto but far from being $1-1$
since $s(k\pi)=0$ for any integer $k$. So $s$ does not have an inverse function. However if we let
$r=s|_{[-\frac{\pi}{2},\frac{\pi}{2}]}$ be the function $s$ restricted to the closed interval $[-\frac{\pi}
{2},\frac{\pi}{2}]$, $r$ \emph{is} $1-1$ since $r$ is strictly increasing on this interval. $r([-\frac{\pi}
{2},\frac{\pi}{2}])=[-1,1]$ and so $r:[-\frac{\pi}{2},\frac{\pi}{2}]\rightarrow [-1,1]$ is a bijection
and has an inverse $r^{-1}:[-1,1]\rightarrow [-\frac{\pi}{2},\frac{\pi}{2}]$. We call the function $r^{-
1}$ the 'inverse sine' or 'arcsine' function. We encourage the reader to revisit their calculus textbook for
other examples of inverse trigonometric functions and see that they are constructed analogously.
\end{example}

\vspace{.2in}

We use the previous example to motivate our next definition. The function $s:\R\rightarrow [-1,1]$
satisfies $s(k\pi)=0$ for any $k\in\Z$. When solving $s(x)=0$ for $x$ we don't get a single value but in
fact an entire set of $x's$ that work. Even though the inverse of $s$ doesn't exists since $s$ is not $1-
1$, we'd like to still be able to talk about the set of all $x$ so that $s(x)=0$ and we will denote this by
$s^{-1}(\{0\})$ or more simply $s^{-1}(0)=\{\dots -2\pi, -\pi,0,\pi, 2\pi, \dots\}$. More precisely we
make the following definition:

\vspace{.2in}

\begin{definition} Let $f\,:\, A\rightarrow B$ and let $Y\subseteq B$. We define the preimage of $Y$
via $f$, sometimes called the inverse image, to be the following subset of $A$:

$$
f^{-1}(Y):=\{x\in A\,|\, f(x)\in Y\}\subseteq A.
$$

Note that $x\in f^{-1}(Y)$ simply means $f(x)\in Y$, it does not mean that $f$ is a bijection and $f^{-
1}$ exists. It is a subset of the domain. Two immediate observations are that $f^{-1}(B)=A$ and $f^{-
1}(\emptyset)=\emptyset$.
\end{definition}

\vspace{.2in}

\begin{example} Revisiting the function $g$ from example $2.12$, $g^{-1}(\{4,5,6,7,8,9\})=\{-

3,2,2,3\}$ and $g^{-1}(\{\dots -2,-1\})=\emptyset$
\end{example}

\vspace{.2in}

\begin{lemma} If $f\,:\, A\rightarrow B$ and $Y_1\subseteq Y_2\subseteq B$, then $f^{-1}(Y_1)\

subseteq f^{-1}(Y_2)$.
\end{lemma}

\vspace{.2in}

\begin{proof}\quad Let $x\in f^{-1}(Y_1)$. Then $f(x)\in Y_1$ and so $f(x)\in Y_2$ and hence $x\in
f^{-1}(Y_2)$. \end{proof}

\vspace{.2in}

\begin{theorem} If $f\,:\, A\rightarrow B$ and $Y_1,Y_2\subseteq B$ then

$$
f^{-1}(B_1\cup B_2)=f^{-1}(B_1)\cup f^{-1}(B_2)
$$
$$
f^{-1}(B_1\cap B_2)=f^{-1}(B_1)\cap f^{-1}(B_2)
$$
\end{theorem}

\vspace{.2in}

\begin{proof} By the previous lemma we know that since $B_1\cap B_2\subseteq B_1, B_2\subseteq
B_1\cup B_2$ we have $ f^{-1}(B_1\cap B_2)\subseteq f^{-1}(B_1), f^{-1}(B_2)\subseteq f^{-1}
(B_1\cup B_2)$ and so $f^{-1}(B_1\cap B_2)\subseteq f^{-1}(B_1)\cap f^{-1}(B_2)$ and $f^{-1}
(B_1)\cup f^{-1}(B_2)\subseteq f^{-1}(B_1\cup B_2) $. If $x\in f^{-1}(B_1)\cap f^{-1}(B_2)$ then
$f(x)\in B_1$ and $f(x)\in B_2$ and so $f(x)\in B_1\cap B_2$ and so $x\in f^{-1}(B_1\cap B_2)$
giving us $f^{-1}(B_1)\cap f^{-1}(B_2)=f^{-1}(B_1\cap B_2)$. Similarly if $x\in f^{-1}(B_1\cup
B_2) $ then $f(x)\in B_1\cup B_2$ and so either $f(x)\in B_1$ or $f(x)\in B_2$. By definition of
preimage this implies $x\in f^{-1}(B_1)$ or $x\in f^{-1}(B_2)$ and so $x\in f^{-1}(B_1)\cup f^{-1}
(B_2)$ and we have $f^{-1}(B_1\cup B_2)=f^{-1}(B_1)\cup f^{-1}(B_2)$.\end{proof}

\vspace{.2in}

\begin{definition} Let $h\,:\, A\rightarrow B$ and $X\subseteq A$. We define a subset of $B$ called
the image of $X$ via $h$ as follows:

$$
h(X)=\{y\in B\,|\, \exists x\in X, h(x)=y\}\subseteq A.
$$
\end{definition}

An immediate consequence of the definition is that $h(A)$ is the image or range of the function $h$.

\vspace{.2in}

\begin{example} Let $g: [0,\infty)\rightarrow \mathbb{R}$ be $g(x)=\sqrt{x}$. Then

$$
g([1,4])=[1,2],
$$
$$
g(\{45\})=\{\sqrt{45}\},
$$
$$
g(\emptyset)=\emptyset,
$$
$$
g([0,\infty))=[0,\infty)=im(g).
$$
\end{example}
\vspace{.2in}

\begin{lemma} Let $h\,:\, A\rightarrow B$ and $X_1\subseteq X_2\subseteq A.$ Then $h(X_1)\
subseteq h(X_2)$.
\end{lemma}

\vspace{.2in}

\begin{proof} Let $y\in h(X_1)$. Then $\exists x\in X_1$ so that $h(x)=y$. Since $X_1\subseteq X_2$
we have that $x\in X_2$ and so $y\in h(X_2)$.\end{proof}

\vspace{.2in}

\begin{theorem} Let $h\,:\, A\rightarrow B$ and let $A_1,A_2\subseteq A$. Then we have:
$$
h(A_1\cap A_2)\subseteq h(A_1) \cap h(A_2)
$$
$$
h(A_1 \cup A_2)=h(A_1)\cup h(A_2),
$$
\end{theorem}

\vspace{.2in}

\begin{proof} Let $y\in h(A_1\cap A_2) $. Then $\exists x\in A_1\cap A_2$ so that $h(x)=y$ by
definition of image. Then by definition of intersection we have $x\in A_1\land x\in A_2$ and so $y\in
h(A_1)\land y\in h(A_2)$ by definition of image. This means $y\in h(A_1) \cap h(A_2)$ by definition
of intersection and so $h(A_1\cap A_2)\subseteq h(A_1) \cap h(A_2)$.

\vspace{.1in}

Let $y\in h(A_1\cup A_2) $. Then $\exists x\in A_1\cup A_2$ so that $h(x)=y$ by definition of image.
Then by definition of union we have $x\in A_1\lor x\in A_2$ and so $y\in h(A_1)\lor y\in h(A_2)$ by
definition of image. This means $y\in h(A_1) \cup h(A_2)$ by definition of intersection and so
$h(A_1\cup A_2)\subseteq h(A_1) \cup h(A_2)$. If we take $z\in h(A_1) \cup h(A_2)$ then $z\in
h(A_1)$ or $z\in h(A_2)$. If $z\in h(A_1)$ then there exists a $a\in A_1$ so that $h(a)=z$ by
definition of image. So $a\in A_1\cup A_2$ by definition of union and so $z=h(a)\in h(A_1\cup A_2)$
by definition of image. The case where $z\in h(A_2)$ is completely analogous. So we have $h(A_1) \
cup h(A_2)\subseteq h(A_1\cup A_2)$ and so $h(A_1\cup A_2)=h(A_1) \cup h(A_2)$.\end{proof}

\vspace{.2in}

\begin{example} Let $f\,:\, \{1,2,3,4\}\rightarrow \{a,b,c,d,e\}$ be given by the following diagram:

$$
\begin{tikzcd}[column sep=1em,row sep=2ex]
1 \arrow[d] & 2 \arrow[dr] & 3 \arrow[dr] & 4 \arrow[d] & \, \\
a&b&c&d&e
\end{tikzcd}
$$

Let $A_1=\{1,3\}$ and $A_2=\{1,4\}$. Then $f(A_1)=\{a,d\}, f(A_2)=\{a,d\}$, $f(A_1\cap A_2)=f(\

{1\})=\{a\}$ and $f(A_1\cup A_2)=f(\{1,3,4\})=\{a,d\}$ and we check $f(A_1)\cap f(A_2)=\
{a,d\}=f(A_1)\cup f(A_2)$. So here we see an example where $f(A_1\cap A_2)\subsetneq f(A_1)\cap
f(A_2)$.
\end{example}

\vspace{.2in}

In summary a function $f:A\rightarrow B$ consists of three pieces of information: a set $A$ the
domain, a set $B$ the codomain, and rule that assigns to each element of $A$
a unique element of the set $B$. As mentioned earlier we can specify the rule by listing the ordered
pairs in the graph of the function. There are many useful relationships in mathematics that require a
much looser framework than that of a function and we turn our attention to the following definition:

\vspace{.2in}

\begin{definition} Let $X$ and $Y$ be sets and let $R\subseteq X\times Y$. Then we say that $R$ is a
relation with domain $X$ and codomain $Y$ or say $R$ is a relation from $X$ to $Y$. Technically the
relation is the triple $(R,X,Y)$ but we often just call the relation $R$ when the domain and codomain
are clear or implied. We will often write $aRb$ to mean $(a,b)\in R$. If $X=Y$ then we say that $R$ is
a relation \emph{on $X$}.
\end{definition}

\vspace{.2in}

We always have $\emptyset\subseteq X\times Y$ for any $X$ and $Y$; we call this the \emph{empty
relation} with domain $X$ and codomain $Y$. Note that even though the empty set is unique there are
many empty relations as we vary domain and codomain. In a similar vein given any sets $X$ and $Y$
we can let $R=X\times Y$ and declare every element of $X$ to be related to every element of $Y$.
Given any set $X$ there is a natural relation $=$ with domain $X$ and codomain $X$ given by $(a,b)\
in =$ to mean $a=b$; we call this the \emph{equals relation} on the set $X$.

\vspace{.2in}

Any function $f:X\rightarrow Y$ is a relation from $X$ to $Y$ if we interpret the function rule as the
set of ordered pairs $graph(f)$. A natural generalization of the function relation is that of a partial
function from $X$ to $Y$; we say that $P:X\rightarrow Y$ is a partial function from $X$ to $Y$ if
$P$ is any relation from $X$ to $Y$ (so $P\subseteq X\times Y$) so that $(x,y_1), (x,y_2)\in P \
implies y_1=y_2$. In other words $P$ is a collection of ordered pairs so that no first coordinate is
related to more than one second coordinate. An example of a partial function is $D:\mathbb{R\times
R}\rightarrow \mathbb{R}$ given by $((x,y),z)\in D$ exactly when $z=\frac{x}{y}$ (we can also
write $D(x,y)=z$ here). This isn't a function since elements like $(3,0)$ in the domain don't correspond
to anything in the codomain, but it is a partial function since if $((x,y),z_1), ((x,y),z_2)\in D$ then
$z_1=\frac{x}{y}=z_2$. Another partial function is $S:\mathbb{N\times N}\rightarrow \mathbb{N}$
given by $((a,b),c)\in S$ exactly when $c=a-b$ (we can also write $S(a,b)=a-b$. This isn't a function
since elements like $(2,4)$ in the domain don't correspond to anything in the codomain.
\vspace{.2in}

In these notes we will be focusing on relations $R$ on a set $X$ that satisfy some well-behaved
properties. For a look at relations in general feel free to click
\hyperlink{https://en.wikipedia.org/wiki/Relation_(mathematics)}{HERE}

\newpage

\section{Relations}

Let $R\subseteq X\times X$ be a relation on a set $X$. We write $(a,b)\in R$ or $aRb$ and say "$a$ is
related to $b$ via $R$" where $a,b\in X$.

\vspace{.2in}

\begin{example}
Let $T=\{O,A,B,AB\}$ and let $BT$ be the relation on $T$ given by the arrow diagram:
$$
\begin{tikzcd}[column sep=1em,row sep=4ex]
O\arrow[d]\arrow[dr] \arrow[drr] \arrow[drrr] & A\arrow[d]\arrow [drr] & B\arrow[d] \arrow[dr] & AB
\arrow[d] \\
O & A & B & AB \end{tikzcd}
$$

This is the 'blood type' relation on the set $T$ of blood types, and the arrows reflect the 'giver-receiver'
dynamic. Note that the arrows are absolutely essential, first and foremost because we make a
mathematical distinction between the domain and the codomain, and secondly because giving the
wrong blood type to someone could be extremely dangerous.

\end{example}

\vspace{.2in}

\begin{example}
Let $W=\{1,2,3,4\}$ and let $L$ be the relation on $W$ given by the arrow diagram:

$$
\begin{tikzcd}[column sep=1em,row sep=4ex]
1\arrow[dr] \arrow[drr] \arrow[drrr] & 2\arrow [drr]\arrow[dr] & 3 \arrow[dr] & 4 \\
1 & 2 & 3 & 4 \end{tikzcd}
$$

One can think of the relation $L$ the following way: we have $xLy$ exactly when $x<y$.

\end{example}
\vspace{.2in}

\begin{definition}We say that $R$ is reflexive if $\forall a\in X, (a,a)\in R.$\end{definition}

\vspace{.2in}

The relation $BT$ above is clearly reflexive, since all four pairs $(O,O),(A,A),(B,B),(AB,AB)$ are
part of the relation. The relation $L$ on the other hand is not reflexive since $(1,1)\not\in L $

\vspace{.2in}

\begin{definition}Let $R_1, R_2$ be relations on $X$. We define the relation:

$$
R_2\circ R_1=\{(a,c)\,|\, \exists b\in X, (a,b)\in R_1, (b,c)\in R_2\}
$$
to be the composition of $R_1$ followed by $R_2$, and we also define
$$
R_1^{-1}=\{(a,b)\,|\, (b,a)\in R_1\}
$$
to be the inverse of the relation $R_1$.
\end{definition}

\vspace{.2in}

We compute $BT\circ BT$ by stacking the relation $BT$ on top of itself and reading off the
composition in the form of an arrow diagram:

$$
\begin{tikzcd}[column sep=1em,row sep=4ex]
O\arrow[d]\arrow[dr] \arrow[drr] \arrow[drrr] & A\arrow[d]\arrow [drr] & B\arrow[d] \arrow[dr] & AB
\arrow[d] \\
O\arrow[d]\arrow[dr] \arrow[drr] \arrow[drrr] & A\arrow[d]\arrow [drr] & B\arrow[d] \arrow[dr] & AB
\arrow[d] \\
O & A & B & AB\end{tikzcd}
$$

Evidently we have $BT\circ BT=BT$. We compute $BT^{-1}$ by simply reversing all the arrows:

$$
\begin{tikzcd}[column sep=1em,row sep=4ex]
O\arrow[d]& A\arrow[dl]\arrow [d] & B\arrow[dll] \arrow[d] & AB \arrow[dlll]\arrow[dll]\arrow[dl]\
arrow[d] \\
O & A & B & AB \end{tikzcd}
$$

\vspace{.2in}
 \begin{definition} We say that $R$ is symmetric if $R=R^{-1}$; this is equivalent to saying $\forall
a,b\in X, (a,b)\in R \iff (b,a)\in R$.\end{definition}

\vspace{.2in}

Since $BT^{-1}$ and $BT$ are different, $BT$ is not symmetric.

\vspace{.2in}

\begin{definition} We say that $R$ is antisymmetric if $\forall a,b\in X$, if $(a,b),(b,a)\in R$ then
$a=b$. In other words $R\cap R^{-1}\subseteq\{(a,a)\,|\,a\in X\}.$\end{definition}

\vspace{.2in}

Since $BT\cap BT^{-1}=\{(O,O),(A,A),(B,B),(AB,AB)\}$ we have that the relation $BT$ is

antisymmetric.

\vspace{.2in}

\begin{definition} We say $R$ is transitive if $R\circ R\subseteq R$; this is equivalent to saying $\
forall a,b,c\in X$, if $(a,b),(b,c)\in R$ then $(a,c)\in R$.\end{definition}

\vspace{.2in}

The relation $BT$ is transitive since $BT\circ BT=BT$. Looking at the relation $L$ we compute $L\
circ L$ by stacking $L$ on top of itself and composing arrows:

$$
\begin{tikzcd}[column sep=1em,row sep=4ex]
1\arrow[dr] \arrow[drr] \arrow[drrr] & 2\arrow [drr]\arrow[dr] & 3 \arrow[dr] & 4 \\
1\arrow[dr] \arrow[drr] \arrow[drrr] & 2\arrow [drr]\arrow[dr] & 3 \arrow[dr] & 4 \\
1 & 2 & 3 & 4 \end{tikzcd}
$$

to get

$$
L\circ L \quad \begin{tikzcd}[column sep=1em,row sep=4ex]
1\arrow[drr] \arrow[drrr] & 2\arrow [drr] & 3 & 4 \\
1 & 2 & 3 & 4 \end{tikzcd}
$$

We see that $L\circ L\subseteq L$ and so $L$ is transitive. Note that we lost some arrows in the
composition here, but since we didn't add any new ones this relation is transitive.

\vspace{.2in}

\begin{example}
Let $X=\{2,3,4\dots\}$ and define a relation $Q$ on $x$ by $mQn$ iff $m$ and $n$ have only the
factors $\pm 1$ in common (we say $m,n$ are coprime or relatively prime). The relation $Q$ is not
reflexive since it's false that $2Q2$ for instance. The relation $Q$ is certainly symmetric, but it is not
antisymmetric since $3Q2$ and $2Q3$ yet $3\not=2$. And the relation $Q$ is not transitive since
$2Q3$ and $3Q2$ yet $2Q2$ is false. Let $Q'$ be the relation on the same set $X$ given by $mQ'n$ iff
$m$ and $n$ share a common factor $p>1$. Then $Q'$ is reflexive since for any $a\in X$ we have $a$
shares the factor $a$ with itself, and so $aQ'a$ is always true. For the same reason that $Q$ is
symmetric, $Q'$ is symmetric. $Q'$ is not antisymmetric since $4Q'6$ and $6Q'4$ yet $4\not=6$.
Lastly $Q'$ is not transitive since $4Q'6$ and $6Q'3$ yet it's false that $4Q'3$.
\end{example}

\vspace{.2in}

We say that $R$ is an \emph{equivalence relation on $X$} if it is reflexive, symmetric, and transitive.
We say $R$ is a \emph{partial order on $X$}, and call $X$ a 'poset', if it is reflexive, antisymmetric,
and transitive.

\vspace{.2in}

\begin{example} If $X=\{1,2\}$, then the relation

\vspace{.15in}

$R_1=\{(1,1),(1,2)\}$ is not reflexive ($(2,2)\notin R_1$), it is not symmetric since $R_1^{-1}\

not=R_1$, but is transitive since $R_1\circ R_1=R_1$ and is antisymmetric since $R_1\cap R_1^{-
1}=\{(1,1)\}\subseteq\{(1,1),(2,2)\}.$

\vspace{.1in}

$R_2=\{(1,2),(2,1)\}$ is not reflexive since $(1,1)\notin R_1$, it is symmetric, it is not transitive since
$R_2\circ R_2=\{(1,1),(2,2)\}$, and is not antisymmetric since $R_2\cap R_2^{-1}=R_2\not\subseteq\
{(1,1),(2,2)\}.$
\end{example}

\vspace{.2in}

\begin{example}Let $X=\mathbb{Z}$ and define $a|b$ to mean $\exists j\in\mathbb{Z}$ so that

$b=aj$. We read $a|b$ as $a$ \emph{divides} $b$ or $b$ is a multiple of $a$. Note that $\forall a\in \
Z$, $a|0$ and $\pm 1|a$. This defines a relation $R$ on the integers where we say $(a,b)\in R\iff a|b$.

\vspace{.2in}

This relation is reflexive since $\forall a\in X, a|a$ and so $(a,a)\in R$. It is not symmetric since $(1,2)\
in R$ but $(2,1)\not\in R$. It is not antisymmetric since $1|-1$ and $-1|1$ so $(1,-1),(-1,1)\in R$ but
$1\not= -1$. It is transitive since if $a|b$ and $b|c$, $\exists j,k\in\mathbb{Z}$ so that $b=aj,c=kb$ and
so $c=kb=k(aj)=(kj)a$, and so $c$ is a multiple of $a$ and hence $a|c$.
\end{example}

\vspace{.2in}
\begin{example} Let $X$ be any set and let $P(X)=\{A\,|\, A\subseteq X\}$ be the collection of all
subsets of $X$ (this is called the power set of $X$). Define a relation $S$ on $P(X)$ by $ASB$ iff $A\
subseteq B$. This relation is reflexive since $A\subseteq A$ for any set $A$ implies $ASA$, it is
antisymmetric since if $ASB$ and $BSA$ then $A\subseteq B, B\subseteq A$ and hence $A=B$.
It is clearly transitive: $ASB, BSC$ implies $A\subseteq B, B\subseteq C$ and so $A\subseteq C$ and
so $ASC$. It is symmetric exactly when $X=\emptyset$; if $X\not=\emptyset$ then $\exists a\in X$
and so $\{a\}\in P(X)$; in this case $\emptyset S\{a\}$ is true but $\{a\} S\emptyset$ is false.
\end{example}

\vspace{.2in}

For instance if $X=\{1,2\}, P(X)=\{\emptyset, \{1\},\{2\}, X\}$ and the above relation is $\{(\emptyset,
\emptyset),(\emptyset, \{1\}),(\emptyset, \{2\}),(\emptyset, \{1,2\}),(\{1\}, \{1\}),
$

$(\{1\}, \{1,2\}), (\{2\}, \{2\}),(\{2\}, \{1,2\}),(\{1,2\}, \{1,2\})\}.$

\vspace{.2in}

If we let $A=\{a,b,c\}$ the relation $S$ on $P(A)$ is best seen through the use of a \emph{Hasse
diagram} as follows:

$$
\begin{tikzcd}[row sep=1.5cm, column sep=1.7cm]
& A \\
\{a, b\}\urar & \{a, c\}\uar \arrow[from=dr]& \{b, c\}\ular \\
\{a \}\uar\urar & \{b\}\ular[crossing over] \urar[crossing over] & \{ c\} \uar\\
& \varnothing\ular\uar \urar
\end{tikzcd}
$$

Here we put the elements of the set $P(A)$ as the vertices, draw arrows to indicate inclusion, and
make the assumptions that every vertex is related to itself (it's reflexive). For a proper definition and
more examples of a Hasse diagram check out \href{https://en.wikipedia.org/wiki/Hasse_diagram}
{HERE}

\vspace{.2in}

\begin{example}Let $B=\{f\,|\, f\,:\, \mathbb{N}\rightarrow \{0,1\}\}$, in other words $B$ is the set of
infinite binary sequences. Define a relation $R$ on $B$ as follows:

$$
fRg \iff f^{-1}(1)\subseteq g^{-1}(1).
$$

Elements of $B$ look like $f=010011010101000010010000\dots$. Then we have two sequences

related if they have $1's$ in exactly the same positions. Letting $\overline{0}=000000\dots$ and $\
overline{1}=111111\dots$ we see that $0100\overline{0} R 01101\overline{0}$ and $01101\
overline{0} R 11101\overline{1}$ etc..

\vspace{.2in}

Clearly the relation is reflexive since checking $f^{-1}(1)\subseteq f^{-1}(1)$ amounts to seeing that
any set is a subset of itself, and the relation is also transitive since $ f^{-1}(1)\subseteq g^{-1}(1)$ and
$g^{-1}(1)\subseteq h^{-1}(1)$ imply that $f^{-1}(1)\subseteq h^{-1}(1)$ and so $fRg\land gRh\
implies fRh$. The relation is not symmetric however; if we define $z(n)=0$ for every $n\in\mathbb{N}
$ and $a(n)=1$ for every $n\in\mathbb{N}$ we have $z^{-1}(1)=\emptyset, a^{-1}(1)=\mathbb{N}$
so $zRa$ but it's false that $aRz$. This relation is antisymmetric since if $f^{-1}(1)\subseteq g^{-1}
(1)$ and $g^{-1}(1)\subseteq f^{-1}(1)$ would imply $f^{-1}(1)= g^{-1}(1)$, and since the codomain
consists of only two elements this forces $f^{-1}(0)= g^{-1}(0)$ as well, so $f=g$. So $R$ is a partial
order on $B$.
\end{example}

\vspace{.2in}

Note in the last example that if we changed the codomain to a different set, say $\{0,1,2\}$ or $\
mathbb{R}$ we could still define a relation on these sets in a completely analogous manner and these
relations would be reflexive and transitive, but with a larger codomain we would lose antisymmetry.

\vspace{.2in}

If $f\,:\, X\rightarrow X$ is a function with domain and codomain $X$ then as we pointed out earlier
$R=graph(f)=\{(a,f(a))\,|\, a\in X\}$ is a relation on $X$. Here $R^{-1}$ makes sense regardless of
whether $f$ has a left/right inverse, it just isn't usually the graph of a function. But $R^{-1}=graph(f^{-
1})$ if $f$ is indeed a bijection. If $R$ is symmetric then $f$ \emph{must} be a bijection but the
converse of this statement is in general false: if $f\,:\, \mathbb{R}\rightarrow\mathbb{R}$ is given by
$f(x)=x^3$, $f$ is a bijection and $R=graph(f)=\{(x,x^3)\,|, x\in\mathbb{R}\}$ while $R^{-1}=\
{(x^3,x)\,|\, x\in \mathbb{R}\}$ is different than $R$ since $(2,8)\in R$ but $(2,8)\notin R^{-1}$.

\vspace{.2in}

We recall that we say $\sim$ is an \emph{equivalence relation} on a set $X$ if it is reflexive,

symmetric, and transitive. We often just say $(X,\sim)$ is an equivalence relation. For each $a\in X$
we define $[a]:=\{b\,|\, a\sim b\}$ to be the \emph{equivalence class} of $a$. Note that since $\sim$ is
reflexive we always have that $a\in[a]$ and so equivalence classes are always non-empty.

\vspace{.1in}
\begin{example} Let $X_1=\{a,b,c,d\}$ and let $\sim_1$ be the equivalence relation:

$$
\{(a,a),(b,b),(c,c),(d,d), (a,b), (b,a), (c,d),(d,c)\}.
$$

Then $[a]=[b]=\{a,b\}$ and $[c]=[d]=\{c,d\}$ are the two distinct equivalence classes. Note that the
same equivalence class can often have many names like in this example.
\end{example}

\vspace{.2in}

\begin{example} Let $X_2=\mathbb{R}$ and define $x\sim_2 y$ iff $x-y\in\mathbb{Q}$. One easily
sees this is an equivalence relation. There are infinitely many equivalence classes, a few examples are
$[0]=[\frac{1}{5}]=[-\frac{43}{642}]=[r]$ where $r$ is any rational number (all the rational numbers
are equivalent). Another class is $[\pi]=[\pi-3.14]=[\pi-3.1415926]$ etc...
\end{example}

\vspace{.2in}

One might also see here that every element of $X$ is in one of these equivalence classes, and that these
classes are either equal or disjoint. This is a general fact about equivalence relations:

\vspace{.2in}

\begin{theorem}[Equivalence Class Theorem]Let $(X,\sim)$ be an equivalence relation. Then $\forall

a,b\in X$ either:

\begin{enumerate}
\item $[a]\cap[b]=\emptyset$ or
\item $[a]=[b]$
\end{enumerate}
\end{theorem}

\vspace{.1in}

\begin{proof} Assume $\exists x\in [a]\cap[b]$. Then $a\sim x$ and $b\sim x$ and so by symmetry $x\
sim b$ and hence by transitivity $a\sim b$ and $b\sim a$. If $y\in[a]$ then $a\sim y$ and combining
this with $b\sim a$ by transitivity we get $b\sim y$ or $y\in[b]$ and hence $[a]\subseteq[b]$. Similarly
$[b]\subseteq[a]$ and so $[a]=[b]$.\end{proof}

\vspace{.2in}

We denote the set of equivalence classes of $(X,\sim)$ by $X/\sim$, namely,

$$
X/\sim\,:=\{[a]\,|\, a\in X\}.
$$
There is a natural surjective function $\pi\,:\, X\rightarrow (X/\sim)$ given by $\pi(a)=[a]$. Looking
back at $(X_1, \sim_1)$ we see that $(X_1/\sim_1)=\{[a],[c]\}$ has two elements and $\pi\,:\, X_1\
longrightarrow (X_1/\sim_1)$ is given by $\pi(a)=[a]=\pi(b), \pi(c)=[c]=\pi(d)$. Note that $\pi^{-1}
([a])=\{a,b\}=[a]=[b]$ and $\pi^{-1}([c])=\{c,d\}=[c]=[d]$ just returns the equivalence classes as pre-
images.

\vspace{.2in}

The Equivalence Class Theorem says that equivalence relations give rise to a \textbf{partition}
of the set involved via its equivalence classes. A partition $\mathbf{P}$ of a set $X$ is a subset of the
power set $P(X)$, excluding the empty set, so that $\cup \mathbf{P}=X$ and if $A\cap B\not=\
emptyset$ are elements of $\mathbf{P}$ then $A=B$. Conversely we have the following:

\vspace{.2in}

\begin{theorem}
Let $\mathbf{P}$ be a partition of a set $X$ and define a relation $\sim$ on $X$ by $a\sim b$ iff $\
exists P\in\mathbf{P}$ so that $a,b\in P$. Then $\sim$ is an equivalence relation an $X$ and $X/\
sim=\mathbf{P}$.
\end{theorem}
\vspace{.2in}

\begin{proof} If $a\in X$ then since $\mathbf{P}$ is a partition there is some $P\in\mathbf{P}$ so

that $a\in P$. But then $a\sim a$, and we see that $\sim$ is reflexive. Suppose $a,b\in X$ with $a\sim
b$. This means there exists some $Q\in\mathbf{P}$ so that $a,b\in Q$. But then $b,a\in Q$ and hence
$b\sim a$ as well; $\sim$ is symmetric. If $a,b,c\in X$ with $a\sim b, b\sim c$ then $\exists P_1, P_2\
in\mathbf{P}$ with $a,b\in P_1$ and $b,c\in P_2$. So $b\in P_1\cap P_2$ implies that $P_1=P_2$
since $\mathbf{P}$ is a partition, and so $a,b,c\in P_1=P_2$ implies $a\sim c$. We leave it to the
reader to readily check that $X/\sim$ is exactly $\mathbf{P}$.\end{proof}

\vspace{.2in}

\begin{example} Let $f\,:\, A\rightarrow B$ be any function. We can build another function based on
$f$ that is a bijection in the following way. First define a relation $\sim$ on $A$ by setting $x\sim y$ if
$f(x)=f(y)$; it's immediate that this is an equivalence relation on $A$. Define $\tilde{f}\,:\, (A/\sim)\
rightarrow im(f)$ by $\tilde{f}([x])=f(x)$. We first make sure this is well defined: if $[x]=[y]$ then
$f(x)=f(y)$ and so $\tilde{f}([x])=f(x)=f(y)=\tilde{f}([y])$ and the definition of $\tilde{f}$ does not
depend on the choice of equivalence class representative. We also see that $\tilde{f}$ is a bijection
since it's clearly onto, and if $\tilde{f}([x_1])=\tilde{f}([x_2])$ then $f(x_1)=f(x_2)$ and so $x_1\sim
x_2$ and $[x_1]=[x_2]$ shows that it is 1-1.
\end{example}

\vspace{.2in}

\begin{example} Let $s\,:\, \mathbb{R}\rightarrow \mathbb{R}$ be $s(x)=\sin(x)$, clearly $s$ is

neither $1-1$ nor onto. $\tilde{s}\,:\, (\mathbb{R}/\sim)\rightarrow [-1,1]$ given by $\tilde{s}[x]=\
sin(x)$ is a bijection.
\end{example}
\vspace{.2in}

\begin{example}
Let $S$ be the relation on $X=\Z\times (\Z-\{0\})$ given by
$$
(a,b)S(c,d) \iff ad=bc.
$$

It's clear that given any $(a,b)\in X$ we have $(a,b)S(a,b)$, so $S$ is reflexive. If $(a,b)S(c,d)$ then
certainly $(c,d)S(a,b)$ since the conditions $ad=bc$ and $cb=ad$ are equivalent. So $S$ is symmetric.
If $(a,b)S(c,d)$ and $(c,d)S(x,y)$ then $ad=bc, cy=dx$ and so

$$
ady=(ad)y=(bc)y=b(cy)=bdx,
$$
and since $d\not=0$ we can cancel it and get $ay=bx$. So $(a,b)S(x,y)$ and $S$ is transitive and an
equivalence relation.

\vspace{.2in}

Define $f:X/S\rightarrow \Q$ by $f([(a,b)])=\frac{a}{b}$. We claim that $f$ is a bijection, but first we
need to establish that $f$ makes sense, that it is well-defined. Suppose that $[(a,b)]=[(c,d)]$ so that $
(a,b)S(c,d)$ or $ad=bc$. Then $\frac{a}{b}=\frac{c}{d}$ and so $f([(a,b)])=f([(c,d)])$. In other
words, our definition of $f$ does not depend on the choice of equivalence class representative and
hence is well-defined. Clearly $f$ is onto. If $f([(a_1,b_1)])=f([(a_2,b_2)])$ then $\frac{a_1}{b_1}=\
frac{a_2}{b_2}$ and so $a_1b_2=a_2b_1$. But then $(a_1,b_1)S(a_2,b_2)$ and $
[(a_1,b_1)]=[(a_2,b_2)]$ and $f$ is $1-1$.
\end{example}

\vspace{.2in}

\newpage

\section{Theory
of Numbers}

\vspace{.3in}

We will spend some time discussing/introducing a subject called arithmetic or number theory, which is
the study of the integers $\mathbb{Z}$ and especially the non-negative integers $\mathbb{N}$. This
will require a type of proof technique that involves proving statements of the following form:

$$
\forall n\in\mathbb{N}, P_n\qquad (*)
$$
where $P_n$ is a predicate statement that depends on the natural number $n$. Some examples of such
predicates are $"n<2^n"$ or $1+3+5+\dots+(2n+1)=(n+1)^2$. We introduce a proof technique to
establish statements like $(*)$ above.

\vspace{.3in}

\begin{theorem}[Principle of Mathematical Induction (version 1)] Let $P_n$ be a predicate statement

that depends on $n\in\mathbb{N}$. Then in order to prove $\forall n\in\mathbb{N}\, P_n$ it suffices to
establish:

$$
1.\, P_0
$$
$$
2. \forall k\in\mathbb{N}\, P_k\implies P_{k+1}
$$
\end{theorem}
\vspace{.2in}

We often refer to establishing $P_0$ above as establishing the \emph{base case} and assuming the
hypothesis $P_k$ of the conditional $P_k\implies P_{k+1}$ as the \emph{inductive hypothesis} or \
emph{inductive assumption}.

\vspace{.2in}

\begin{example} Suppose we want to prove $\forall n\in\mathbb{N}\, n<2^n$. We could use the above
principle, which we will often abbreviate 'PMI', as follows:

\vspace{.1in}

Let $I_n$ be the predicate statement $n<2^n$ where $n\in\mathbb{N}$. We proceed to prove $\forall
n\in\mathbb{N}\, I_n$ by the PMI. We check the statement $I_0$ or $0<2^0$ which we know is true.
Then let's assume $I_k$ holds for some $k\in\mathbb{N}$, i.e, $k<2^k$. Adding $1$ to both sides of
this inequality we get $k+1<2^k+1\leq 2^k+2^k=2^{k+1}$ where we are using the fact that $2^k$ is
an increasing sequence with smallest term $1$. So $I_{k+1}$ follows as a consequence of $I_k$ and
we've established $I_k\implies I_{k+1}$ and so $\forall n\in\mathbb{N}\, I_n$ by the PMI.
\end{example}

\vspace{.2in}

\begin{example}
Suppose we want to prove that $5^{n+1}>n^3+2n+1$ for all $n\in\N$. Let $T_n$ be the predicate
statement $5^{n+1}>n^3+2n+1$. We proceed to prove $\forall n\in\N\, T_n$. The base case is the
statement $T_0$ or $5^1>0^3+2(0)+1=1$ which is certainly true. Let's assume that $T_k$ holds for
some $k\in \N$, i.e. that $5^{k+1}>k^3+2k+1$. Then

$$
5^{k+1}=5(5^k)>5(k^3+2k+1)=5k^3+10k+5
$$
and
since $k^3\geq k^2\geq k\geq 0$ when $k\geq 0$ we know that $5k^3+10k+5=k^3+4k^3+10k+5\geq
k^3+4k^2+10k+5>k^3+3k^2+5k+4=(k+1)^3+2(k+1)+1$. Altogether we have

$$
5^{k+1}>(k+1)^3+2(k+1)+1,
$$
which is exactly the statement $T_{k+1}$. Since $I_k\implies I_{k+1}$ we have that $\forall n\in \N \,
T_n$ by the PMI.

\end{example}

\vspace{.2in}

\begin{example}
Suppose we have an $8\times 8$ checkerboard. We remove one square, any square. We would like to
see if we can tile the remaining board $B$ with one square removed by $3$ square $L$ shaped tiles so
that no two tiles overlap, and that every square in $B$ is covered by one of the $L$ tiles. One can
convince oneself that this can always be done by simply forming all possible $B$'s and then brute force
checking each case, but we turn to induction to find a simpler solution. We think of the $8\times 8$
checkerboard as a $2^3\times 2^3$ checkerboard and as the more general question: if we let $B$ be the
remaining board after a single square is removed from a $2^n\times 2^n$ checkerboard, can $B$
always be tiled with $3$ square $L$ shapes?
\vspace{.2in}

In fact it can for every $n\in N^+$, and we prove this by induction on $n\geq 1$. Let $T_n:=$ "any
$2^n\times2^n$ checkerboard with one square removed can be tiled by $3$ square $L$ shapes". The
statement $T_1$ is that every $2^1\times 2^1$ checkerboard with one square removed can be tiled by a
$3$ square $L$ shape; this statement is clearly true as the remaining board would be exactly a $3$
square $L$ tile. Assume that $T_k$ holds for some $k\geq 1$ and let $B$ be any $2^{k+1}\times
2^{k+1}$ board with a single square removed. Divide the board into quarters $Q_1,Q_2,Q_3,Q_4$
each of size $2^k\times 2^k$. We may assume that the missing square is in $Q_1$ or else we can
renumber the quarters. Then $Q_1$ is a $2^k\times 2^k$
board with one square removed, and so by our inductive hypothesis this can be tiled by $L$ shapes.
$Q_2,Q_3,Q_4$ meet at the center of the board $B$ and so we put an $L$ tile so that it meets these
three squares exactly once. This process covers up exactly one square from each of $Q_2,Q_3, Q_4$
and so by our inductive hypothesis we can tile what remains in these squares. Then the tiling of $Q_1$,
along with the tilings of $Q_2, Q_3, Q_4$, as well as the $L$ tile that meets all three of
$Q_2,Q_3,Q_4$ provides a tiling of all of $B$. So $T_{k+1}$ holds, and we are done by the PMI.
\end{example}

\vspace{.2in}

\begin{example}[Backwards Induction]

There is a tendency to do induction somewhat backward when one is first getting used to writing
proofs. Because this mistake is so common let's discuss it. Suppose we are trying to prove that $3|n^3-
n$ for each $n\geq 0$. Here is what 'backward induction' looks like:
\vspace{.2in}

\begin{proof}[Backward Proof] We proceed to prove $3|n^3-n$ by induction on $n\geq 0$, the base
case being $n=0$, and we check that $3$ divides $0^3-0=0$ which it does. Assume that $3|k^3-k$ for
some $k\geq 0$. This means $\exists a\in\Z$ so that $k^3-k=3a$. Then
$$
3|(k+1)^3-(k+1)
$$
and so
$$
3|k^3+3k^2+3k-k
$$
hence
$$
3|3k^2+3k+3a
$$
which implies
$$
3|3(k^2+k+a)
$$
which is clearly true. Since we arrived at a true statement we are done by the PMI
\end{proof}

\vspace{.2in}

The pieces of a proper proof are all there, but the logic doesn't make any sense. One is working
backwards starting with $3|(k+1)^3-(k+1)$ instead of ending up there. To see what a proper proof by
induction looks like we write out the steps that should follow after one states the inductive hypothesis:

\vspace{.2in}
\begin{proof}[Correct Induction] Assume that $3|k^3-k$ for some $k\geq 0$. This means $\exists a\in\
Z$ so that $k^3-k=3a$. Then

$$
(k+1)^3-(k+1)=k^3+3k^2+3k-k=3(k^2+k+a),
$$

and since $k^2+k+a\in\Z$ we know $3|(k+1)^3-(k+1)$. So our $n=k$ case implies the $n=k+1$ case
and we are done by the PMI.

\end{proof}
\vspace{.2in}

Let is examine another example of 'backwards induction' and how to write things properly. Suppose we
want to prove that $\forall n\in\mathbb{N}$ with $n>2$ that $n^3>2n^2+n+1$ using the PMI. Here is
what we want to avoid:

\begin{proof}[Backward Induction] We proceed by induction on $n>2$, the base case being $n=3$ or
$3^3>2(3^2)+3+1$ or $27>22$ which is true. Let's assume that the inequality holds for $n=k>2$,
namely that $K^3>2k^2+k+1$. We want to prove the case $n=k+1$ or

$$
(k+1)^3>2(k+1)^2+2(k+1)+1
$$
or

$$
k^3+3k^2+3k+1>2k^2+4k+2+2k+2+1
$$
or

$$
k^3+3k^2+3k>2k^2+6k+3
$$
.....
\end{proof}

\vspace{.2in}

This gets nowhere really fast. One has to build the case $n=k+1$ up from the case $n=k$. Here is the
right way to do this proof by induction:

\vspace{.2in}

\begin{proof} We proceed by induction on $n>2$, the base case being $n=3$ or $3^3>2(3^2)+3+1$ or
$27>22$ which is true. Let's assume that the inequality holds for $n=k>2$, namely that
$k^3>2k^2+k+1$. We want to prove the case $n=k+1$ holds so we first note that since $k>2$,
$k^2>2k>4$. Now

$$
(k+1)^3=k^3+(3k^2+3k+1)>(2k^2+k+1)+(3k^2+3k+1),
$$
where we use our I.H to make the inequality. The right hand side (RHS) of this is
$$
2k^2+4k+2+3k^2>2k^2+4k+2+6k=
$$
$$
2k^2+5k+2+5k>2k^2+5k+4,
$$
using the inequalities $k^2>2k>4$. Now $2k^2+5k+4=2(k+1)^2+(k+1)+1$ and so altogether we have
established that $(k+1)^3>2(k+1)^2+2(k+1)+1$ which is precisely the statement $n=k+1$. Thus by the
PMI we are done. \end{proof}

\end{example}

\vspace{.2in}
\begin{definition} let $a,b\in\mathbb{Z}$. Then we write $a|b$ and say $a$ \emph{divides} $b$ or
$b$ is a multiple of $a$ if $\exists k\in\mathbb{Z}$ so that $b=ak$. \end{definition}

\vspace{.2in}

\begin{theorem}[Even Lemma]$\forall n\in\mathbb{N}\, 2|n(n+1)$.

\end{theorem}

\vspace{.2in}

\begin{proof} Let $E_n:="2|n(n+1)"$. We proceed to prove $\forall n\in\mathbb{N}\, E_n$ by the

PMI, the base case or $E_0$ being the statement that $2|0(0+1)$ or $2|0$ which we know is true since
every integer divides $0$. Now assume that $E_k$ holds for some $k\in\mathbb{N}$ so that $\exists t\
in\mathbb{Z}$ so that $2t=k(k+1)=k^2+k$. Then $(k+1)((k+1)+1)=k^2+3k+2=(k^2+k)
+(2k+2)=2t+2k+2=2(t+k+1)=2s$ where $s\in\mathbb{Z}$. This means that $E_{k+1}$ follows and so
we are done by the PMI.
\end{proof}

\vspace{.2in}

\begin{example} Let's prove that $\forall n\in\mathbb{N}$\, $6|n^3-n$. Let $D_n:=6|n^3-n"$ and let's
proceed to prove $\forall n\in\mathbb{N}\, D_n$ by the PMI. First we check that our base case $D_0$
holds, which means checking $6|0^3-0$ or $6|0$ which is true since every integer divides $0$. Now
assume $D_k$ holds for some $k\in\mathbb{N}$ so that $\exists c\in\mathbb{Z}$ so that $6c=k^3-k$
or $k^3=6c+k$. Then $(k+1)^3-(k+1)=k^3+3k^2+2k=6c+3k^2+3k6=6c+3k(k+1)$. But by the Even
Lemma we know that $\exists t\in\mathbb{Z}$ so that $k(k+1)=2t$ and so
$6c+3k(k+1)=6c+3(2t)=6(c+t)$ and so $6|(k+1)^3-(k+1)$ since $c+t\in\mathbb{Z}$. So $D_{k+1}$
follows and we are done by the PMI. Note that since $(-n)^3-(-n)=-(n^3-n)$, we have that for any \
emph{integer} $n$, $6|n^3-n$.
\end{example}

\vspace{.2in}

\begin{example}
Let's look at the sequence $5^n+2\cdot 11^n$ when $n=0,1,2,\dots$. The first few terms are $3,27, 267,
2787, 29907$ which when factored are $3,3^3, 3\cdot 89, 3\cdot 929, 3^2\cdot 3323$, and we guess
that each number of the form $5^n+2\cdot 11^n$ is divisible by $3$. Let $G_n$ be the predicate
statement $3|5^n+2\cdot 11^n$; we want to prove that $\forall n\in\N\, G_n$ holds. The base case is
when $n=0$, and we've already checked that $3|3$. So let's assume
that $G_k$ is true for some $k\in\N$, namely that $3|5^k+2\cdot 11^k$. This means that $\exists t\in\
Z$ so that $3t=5^k+2\cdot 11^k$ or $5^k=3t-2\cdot 11^k$. Then

$$
5^{k+1}+2\cdot 11^{k+1}=5(3t-2\cdot 11^k)+2\cdot 11^{k+1}=15t-10\cdot 11^k+2\cdot 11^{k+1}=
$$
$$
15t-11^k(-10+2\cdot 11)=15t-11^k(12)=3(5t-4\cdot 11^k),
$$
\end{example}

and since $5t-4\cdot 11^k\in \Z$ we know that $G_{k+1}$ holds. So $G_k\implies G_{k+1}$ and so
by the PMI we have that $\forall n\in\N\, G_n$ holds.

\vspace{.2in}

\begin{definition}
A directed graph $\Gamma=(V,E)$ consists of a pair of sets, $V$ are called the vertices of $\Gamma$
and $E$ is a collection of sets of the form $e=(v_1,v_2)$ where $v_i\in V$. A \emph{path} in $\
Gamma$ consists of a collection of vertices $v_1,v_2,\dots , v_k$ where for each $1\leq i<k$ $
(v_i,v_{i+1})\in E$. A path is simple if its vertices are distinct. \end{definition}

\vspace{.2in}

\begin{example}
The Peterson graph $P$ pictured below consists of $10$ vertices (the black dots) and $15$ edges:

\vspace{.3in}

\hspace{1.5in}\begin{tikzpicture}
\begin{scope} [vertex style/.style={draw,
circle,
minimum size=2mm,
inner sep=0pt,
outer sep=0pt,
shade}]
\path \foreach \i in {0,...,4}{%
(72*\i:2) coordinate[vertex style] (a\i)
(72*\i:4) coordinate[vertex style] (b\i)}
;
\end{scope}

\begin{scope} [edge style/.style={draw=gray,double=white}]

\foreach \i in {0,...,4}{%
\pgfmathtruncatemacro{\nextb}{mod(\i+1,5)}
\pgfmathtruncatemacro{\nexta}{mod(\i+2,5)}
\draw[edge style] (a\i)--(b\i);
\draw[edge style] (a\i)--(a\nexta);
\draw[edge style] (b\i)--(b\nextb);
}
\end{scope}

\end{tikzpicture}
\end{example}

\vspace{.2in}

Each relation $R$ on a set $X$ determines a directed graph $\Gamma_R$ as follows: the vertices of $\
Gamma_R$ are the set $X$ and the edges of $\Gamma_R$ are just the ordered pairs $(a,b)\in R$. Let
$R^0=\{(a,a)|a\in X\}$ and $R^{n+1}=R^n\circ R$. Then $(a,b)\in R^k$ for some $k\geq 0$ exactly
when there is a path of length $k$ in $\Gamma_R$ from $a$ to $b$.

\vspace{.2in}

\begin{definition}
Let $R$ be a relation on $X$. Let $R^*=R^0\cup R^1\cup R^2\cup \dots$ be the collection of all
ordered pairs $(a,b)$ so that there is a path from $a$ to $b$. We call $R^*$ the
\emph{reflexive/transitive closure} of $R$.
\end{definition}

\vspace{.2in}

\begin{theorem}
For any $R$ we have that $R^*$ is transitive. Moreover if $R\subseteq S$ and $S$ is reflexive and
transitive, we must have $R^*\subseteq S$, i.e. that $R^*$ is the smallest transitive relation containing
$R$.
\end{theorem}
\vspace{.2in}

\begin{proof}
The fact that $R^0\subseteq R^*$ immediately tells us that $R^*$ is reflexive. Let $(a,b),,(b,c)\in
R^*$. Then $(a,b)\in R^k, (b,c)\in R^l$ for some $k,l\geq 0$. So there is a path of length $k$ from $a$
to $b$ and a path of length $l$ from $b$ to $c$, so by traversing one path followed by the next, we
have a path from $a$ to $c$ that is of length $k+l$ and so $(a,c)\in R^{k+1}\subseteq R^*$. So $R^*$
is transitive. Suppose $S$ is reflexive and transitive with $R\subseteq S$. We prove by induction that
$R^n\subseteq S$ for all $n\geq 0$. The base case is $R^0\subseteq S$, which holds since $S$ s
reflexive. If $R^k\subseteq S$ for some $k\geq 0$ we have that $R^{k+1}=R^k\circ R\subseteq S\circ
S\subseteq S$ as well and we've proven our claim by the PMI. But then $R^*=R^0\cup R^1\cup \dots\
subseteq S$ as well.
\end{proof}

\vspace{.2in}

\begin{example} Let $X=\{1,2,3\}$ and let $R$ be given by the following arrow diagram:

$$
R:\quad\begin{tikzcd}[column sep=1em,row sep=4ex]
1\arrow[d]\arrow[dr] & 2\arrow[dr] & 3\arrow[dl]\\
1 & 2 & 3\end{tikzcd}
$$
Then $R^2$ has diagram

$$
R^2:\quad\begin{tikzcd}[column sep=1em,row sep=4ex]
1\arrow[d]\arrow[drr] & 2\arrow[d] & 3\arrow[d]\\
1 & 2 & 3\end{tikzcd}
$$

In particular since $R^2\not\subseteq R$ then $R$ is not transitive. $R^3$ has diagram

$$
R^3:\quad\begin{tikzcd}[column sep=1em,row sep=4ex]
1\arrow[d]\arrow[dr] & 2\arrow[dr] & 3\arrow[dl]\\
1 & 2 & 3\end{tikzcd}
$$

and so $R^3=R$ and hence $R^4=R^2, R^5=R,$ etc... So $R^*=R^0\cup R\cup R^2$ has arrow
diagram

$$
R^*:\quad\begin{tikzcd}[column sep=1em,row sep=4ex]
1\arrow[d]\arrow[dr]\arrow[drr] & 2\arrow[d]\arrow[dr] & 3\arrow[d]\arrow[dl]\\
1 & 2 & 3\end{tikzcd}
$$

We encourage the reader to check that $R^*\circ R^*\subseteq R^*$ and to draw the graph $\
Gamma_R$.

\end{example}

\vspace{.2in}

There are statements like $\forall n\in\mathbb{N}, n\geq 3, n^2<3^n$ where the truth of the predicate
statement $n^2<3^n$ doesn't become consistent until we reach a certain integer, in this case $n=3$, but
where we would like to use the PMI. We can modify the PMI as follows:

\vspace{.2in}

\begin{theorem}[Principle of Mathematical Induction (version 1*)] Let $P_n$ be a predicate statement

that depends on an integer $n$ and let $a\in\mathbb{Z}$ be a fixed integer. Then in order to prove $\
forall n\in\mathbb{Z}, n\geq a, P_n$ it suffices to establish:

$$
1. P_a
$$
$$
2. \forall k\in\mathbb{Z}, k\geq a,\, P_k\implies P_{k+1}
$$
\end{theorem}

\vspace{.2in}

\begin{example}
Let us prove that $\forall n\in \N, n\geq 3$ that $n^2<3^n$. We proceed by induction on $n\geq 3$, the
base case being when $n=3$, and we check that $3^2<3^3$ holds, which it does. Assume that
$k^2<3^k$ for some $k\geq 3$. Since $k\geq 3$ we have $k^2\geq 3k\geq 9$. Then

$$
3^{k+1}=3(3^k)>3k^2=k^2+2k^2\geq k^2+6k>k^2+2k+1=(k+1)^2,
$$

namely that $(k^2<3^k)\implies (3^{k+1}>(k+1)^2)$ when $k\geq 3$. So by the PMI version $1*$ we
have that $\forall n\in \N, n\geq 3$ that $n^2<3^n$.
\end{example}

\vspace{.2in}

\begin{example}

We show that $(A_1\cup A_2\cup \dots \cup A_n)\cap B=(A_1\cap B)\cup (A_2\cap B)\cup \dots \cup
(A_n\cap B)$ for any sets $A_1,A_2, \dots , A_n, B$. We proved this when $n=2$ in the first chapter,
now we wish to extend the result. We proceed to prove $(A_1\cup A_2\cup \dots \cup A_n)\cap
B=(A_1\cap B)\cup (A_2\cap B)\cup \dots \cup (A_n\cap B)$ by induction on $n\geq 2$, the base case
being a theorem we've already proven. Assume that
$(A_1\cup A_2\cup \dots \cup A_k)\cap B=(A_1\cap B)\cup (A_2\cap B)\cup \dots \cup (A_k\cap B)$
for any sets $A_1,A_2,\dots , A_k, B$ where $k\geq 2$. Looking at $(A_1\cup A_2\cup \dots \cup
A_k\cup A_{k+1})\cap B$ if we set $A'=(A_1\cup A_2\cup\dots \cup A_k)$. Then $(A_1\cup A_2\cup
\dots \cup A_k\cup A_{k+1})\cap B=(A'\cup A_{k+1})\cap B=(A'\cap B)\cup (A_{k+1}\cap B)$. By
our inductive hypothesis $A'\cap B=(A_1\cap B)\cup (A_2\cap B)\cup \dots \cup (A_k\cap B)$ and so
altogether we have:
$$
(A_1\cup A_2\cup \dots \cup A_k\cup A_{k+1})\cap B=((A_1\cap B)\cup (A_2\cap B)\cup \dots \cup
(A_k\cap B))\cup (A_{k+1}\cap B)=
$$
$$
(A_1\cap B)\cup (A_2\cap B)\cup \dots \cup (A_k\cap B)\cup (A_{k+1}\cap B),
$$

and so by the PMI version $1^*$ we are done.

\end{example}

\vspace{.2in}

\begin{theorem}[Well Ordering Principle] Let $\emptyset \not=A\subseteq\N$. Then $A$ has a least
element, namely $\exists x\in A$ so that $\forall y\in A$ we have $x\leq y$.
\end{theorem}

\vspace{.2in}

\begin{proof}
We prove the contrapositive, that if $A$ has no least element that $A$ must be empty. Let $P_k$ be the
statement that $k\in \N-A$ where $k\in\N$. We prove that $\forall k\in \N\, P_k$ by induction on $k\
geq 0$. The base case or $P_0$ is the statement that $0\in \N-A$. This is true because otherwise $0$
would be the least element of $A$ and we are assuming $A$ has no least element. Assume that $P_k$
holds for some $k\geq 0$. If any of the numbers $0,1,2,\dots , k\in A$ then one of these would be the
least element of $A$ (why?). If $k+1\in A$ then $k+1$ would be the least element of $A$ which can't
happen. So $k+1\in \N-A$ and so $P_{k+1}$ holds, and hence by the PMI we know that for all $n\in \
N$ we have $n\in \N-A$. But this means that $\N-A=\N$ or that $A=\emptyset$.
\end{proof}

\vspace{.2in}

\begin{definition} Let $n\in\mathbb{N}$ with $n\geq 2$. Define a relation $\equiv$ on $\mathbb{Z}$
by $a\equiv b$ if $n|(a-b)$. When we want to emphasize the role of the specific natural number $n$,
called the modulus, we write $a\equiv b\pmod{n}$. This gives us a family of equivalence relations for
each $n$ (check this!), and we often say $a$ is congruent to $b$ mod (or modulo) $n$ for $a\equiv b\
pmod{n}$ . We denote the set of equivalence classes of this relation by $\mathbb{Z}/n$.
\end{definition}

We can show there are precisely $n$ distinct equivalence classes in this relation as follows, first
recalling a well known fact (which can be proven by induction on $m$):

\vspace{.1in}

\begin{theorem}[Division Algorithm] Let $n\in\mathbb{N}^+$ and $m\in\mathbb{Z}$. Then $\exists

q,r\in\mathbb{Z}$ with $0\leq r<n$ so that $m=qn+r$. Moreover the $q,r$ here are unique given $m$
and $n$.
\end{theorem}

\vspace{.1in}

\begin{proof} Fix $n\in\mathbb{N}^+$. Case 1: $m\in\mathbb{N}$. We proceed by induction on $m\

in\mathbb{N}$. If $m=0$ we set $q=r=0$ to see the base case holds here. Assume that we can write
$k=qn+r$ where $0\leq r< n$. Then $k+1=qn+r+1$ so if $r<n-1$ then we see that $r+1<n$ and
$k+1=qn+(r+1)$ is the desired expression. If $r=n-1$ then $k+1=(q+1)n+0$ is the desired expression.
Either way we get a non-negative remainder less than $n$. Case 2: $m<0$. Then $-m>0$ and we can
apply case 1 to get $-m=qn+r$ where $0\leq r<n$. If $r=0$ then $m=(-q)n+0$ gives us a remainder of
zero. Otherwise $m=-qn-r=-qn-n+n-r=(-q-1)n+(n-r)$ gives us remainder $0<n-r<0$ since $0<r<n$.
\vspace{.2in}

To prove uniqueness suppose we had $m=qn+r=q'n+r'$ where $0\leq r'\leq r<n$. Then $0\leq r'-r<n$
with $r'-r=n(q'-q)$.
Since $n>0$ this forces $q=q'$ and hence $r=r'$ as well. \end{proof}

\vspace{.2in}

Returning to our example we establish some simple lemmas fixing a modulus $n$.

\vspace{.2in}

\begin{lemma} Given $m\in\mathbb{Z}$, we apply the Division Algorithm to $m$ and $n$ to get
$m=qn+r$ with $0\leq r<n$. Then $[m]=[r]$.
\end{lemma}

\vspace{.2in}

\begin{proof} We know $m-r=qn$ so $n|m-r$ and so $m\equiv r\pmod{n}$.\end{proof}

\vspace{.2in}

\begin{lemma} For $0\leq r<r'<n$ we have $[r]\not=[r']$.

\end{lemma}

\vspace{.2in}

\begin{proof} Assume this is false and that $[r]=[r']$ so that $n|(r'-r)$. Then $\exists k\in\
mathbb{N}^+$ so that $r'-r=nk$ and since $k\geq 1$ we know $nk\geq n$ and so $r'-r\geq n$. But
since $r<r'<n$ we know $0<r'-r<n-r\leq n$, a contradiction. So $[r]\not=[r']$.\end{proof}

\vspace{.2in}

\begin{corollary} $\mathbb{Z}/n$, the set of equivalence classes $mod\, n$, is

$$
\{[0],[1],\dots, [n-1]\}
$$
consisting of $n$ distinct equivalence classes.
\end{corollary}

\vspace{.2in}

\textbf{Proof:}\quad This is a consequence of the above two lemmas: every integer $m$
is in the equivalence class determined by the Division Algorithm, and these classes are distinct.\done

\vspace{.2in}

We now define two operations $+,*$ on the $\mathbb{Z}/n$ as follows. Let $a,b\in\mathbb{Z}$ and
set

$$
[a]+[b]:=[a+b]
$$
$$
[a]*[b]:=[ab].
$$

We need to show that these operations are well-defined. Suppose that $[a]=[a'],[b]=[b']. $ Then $a\
equiv a'\pmod{n}, b\equiv b'\pmod{n}$. Then $a-a'=ns, b-b'=nt$ for some $s,t\in\mathbb{Z}$, and we
check: $[a']+[b']=[a'+b']=[(a-ns)+(b-nt)]=[a+b-(s+t)n]=[a+b]=[a]+[b]$ and $[a']*[b']=[a'b']=[(a-ns)(b-
nt)]=[ab-(s+t)n+stn^2]=[ab]=[a]*[b]$, showing that the two operations are well defined up to
equivalence.

\vspace{.2in}

\begin{example} In $\mathbb{Z}/6$ we have

$$
[3]+[4]=[7]=[1]
$$
$$
[3][4]=[12]=[0]
$$
$$
[5][5]=[-1][-1]=[1]
$$ etc... In $\mathbb{Z}/5$ we have $[2]+[3]=[0], [2][3]=[1], [4][4]=[1],[3][4]=[2]$ etc...
\end{example}

\vspace{.2in}

We make a note here that the statements $a\equiv b\pmod{n}$ and $[a]=[b]$ in $\mathbb{Z}/n$ are
logically equivalent and are often used interchangeably. If there are multiple moduli involved and there
is a risk of confusion, we can always use $[a]_n$ to denote the equivalence class of $a$ modulo $n$.

\vspace{.2in}

\begin{definition} Let $n\in\mathbb{N}$ with $n\geq 2$. We say $n$ is prime if $k|n$ with $k\in\
mathbb{N}$ implies $k=1$ or $k=n$. If $n\geq 2 $ is not prime we say it is composite.\
end{definition}

\vspace{.2in}

\begin{theorem}[Fundamental Theorem of Arithmetic] Every natural number greater than or equal to

two can be factored uniquely as a product of primes.
\end{theorem}
\vspace{.2in}
When we write 'factored uniquely' we mean up to rearrangement of the factors: $(2)(3)$ and $(3)(2)$
are to be considered the same factorization of the composite number $6$. We prove part of the
Fundamental Theorem of Arithmetic shortly that any factorization is necessarily unique, and save the
existence of such a factorization for a later section when we introduce a revamped version of the PMI.
First we need some tools.

\vspace{.2in}

\begin{theorem}There are infinitely many prime numbers.

\end{theorem}

\vspace{.1in}

\begin{proof} Suppose there were finitely many primes $p_1,p_2,\dots, p_n$. Let
$M=p_1p_2...p_n+1$ and note that for each $1\leq k\leq n$ we have $p_k|p_1...p_n$. We also see that
$p_k\not|M$ or else $p_k|(M-p_1p_2...p_n)$ or $p_k|1$ which can't happen since $p_k>1$. So $M$ is
a natural number greater than $2$ that has no prime factors, contradicting the Fundamental Theorem of
Arithmetic.\end{proof}

\vspace{.2in}

A useful took for studying properties of prime and composite numbers is the concept of greatest
common divisor. We give a slightly different definition of this from what most people see in middle
school mathematics which is a bit more useful.

\vspace{.2in}

\begin{definition} Let $a,b\in\mathbb{Z}$ with $a,b$ not both zero. Define:

$$
\gcd(a,b)=min\{k\in \mathbb{N}^+\,|\, \exists s,t\in \mathbb{Z}, k=sa+tb\},
$$
and define $\gcd(0,0)=0$. Note that we needed $a,b$ not both zero so that the above set isn't empty.
\end{definition}

\vspace{.2in}

For example if $a\not= 0$ then $\gcd(0,a)=min \{|a|, 2|a|, 3|a|, ...\}=|a|$. If $a=4, b=-6$ we are looking
at the smallest element of the set $\{4s+(-6)t\in\mathbb{N}^+\,|\, s,t\in\mathbb{Z}\}=\{2,4,6,...\}$
which is $2$ since we can write $2=4(2)+(-6)(1)=4(-1)+(-6)(-1)$ etc...Note that there the $s,t$ are not
unique in general as the last example demonstrates. What we do get right away is that the greatest
common divisor of two integers is a linear combination of the integers with integer coefficients. In fact
the definition is that it is the smallest positive linear combination.

\vspace{.2in}

\begin{theorem}[GCD Theorem]For each $a,b\in\mathbb{Z}$ we have:

\begin{enumerate}
 \item $\gcd(a,b)|a$ and $\gcd(a,b)|b$
\item $k\in\mathbb{Z}, k|a, k|b$ implies that $\gcd(a,b)|k$
\end{enumerate}
\end{theorem}

\vspace{.2in}

\begin{proof} If $a,b$ are both zero the theorem certainly holds, so let's assume $a\not=0$. If $b=0$
then $\gcd(a,b)=|a|$ and it's easy to see that the theorem holds in this case (why?). So let's further
assume $b\not=0$. We first examine the case $a,b>0.$ Let $g:=\gcd(a,b)$. Then $\exists s,t\in\
mathbb{Z}$ so that $g=sa+tb$. By the Division Algorithm we get that $a=gq+r$ where $0\leq r< g$. If
$r=0$ then clearly $g|a$. Suppose that $r\not=0$. Then

$$
r=a-gq=a-(sa+tb)q=(1-sq)a+(-tq)b
$$

is a smaller positive linear combination of $a$ and $b$ than $g$ which is not possible. So $r=0$ and
$g|a$. Similarly $g|b$ and we are done with part 1. Let $k\in\mathbb{Z}$ with $k|a$ and $k|b$. Then
$a=kx,b=ky$ for some $x,y\in\mathbb{Z}$. Then $g=sa+tb=skx+tky=k(sx+ty)$ and so $g|k$ and we
are done with part 2. Let $l\in\mathbb{Z}$. We leave it to the reader to supply the argument when one
or both of $a,b$ are negative.\end{proof}

\vspace{.2in}

The two properties in the GCD Theorem are often taken as the definition of greatest common divisor,
but we derive these properties as consequences of our definition.

\vspace{.2in}

\begin{lemma} If $p$ is a prime, $a\in\mathbb{Z}$ then $\gcd(a,p)=1$ or $\gcd(a,p)=p$. $\

gcd(a,p)=1$ iff $p\not| a$.
\end{lemma}

\vspace{.2in}

\begin{proof} $\gcd(a,p)|p$ and so $\gcd(a,p)=1$ or $p$ since $p$ is prime. If $\gcd(a,p)=1$ then by
definition $\exists s,t\in\mathbb{Z}$ so that $sa+tp=1$ and if $p|a$ we'd have $p|(sa+tp)$ or $p|1$
which can't happen. So $p\not|a$. If $\gcd(a,p)\not=1$ then $\gcd(a,p)=p$ and so $p|a$.\end{proof}

\vspace{.2in}

\begin{lemma}[Euclid's Lemma] Let $p$ be a prime number so that $p|ab$ where $a,b\in\mathbb{Z}$.
Then $p|a$ or $p|b$.
\end{lemma}

\vspace{.2in}

\begin{proof} Suppose that $p\not|a$. Then $\gcd(a,p)=1$ and so by definition of greatest common
divisor $\exists s,t\in\mathbb{Z}$ so that $sa+tp=1$. Then $bsa+btp=b$. Now $p|btp$ and $p|bsa$
since $p|ab$ and so $p|(bsa+btp)$ or $p|b$. \end{proof}

\vspace{.2in}

Note that Euclid's lemma fails for composite numbers: we have $6|(3)(4)$ but $6\not|3$ and $6\not|4$
for instance. Note that the proof of Euclid's lemma depended on our funny way of defining greatest
common divisor.

\vspace{.2in}

\begin{proof}[Proof of Uniqueness, FTA] Suppose $n\in\mathbb{N}$ with $n\geq 2$ has two

factorizations $n=p_1^{\alpha_1}p_2^{\alpha_2}\dots p_k^{\alpha_k}=q_1^{\beta_1}q_2^{\beta_2}\
dots q_l^{\beta_l}$ where $p_1\leq p_2\leq \dots\leq p_k$ and $q_1\leq q_2\leq\dots \leq q_l$ are
primes and assume without loss of generality that $k\leq l$. Since $p_1|q_1^{\beta_1}q_2^{\beta_2}\
dots q_l^{\beta_l}$ we must have by Euclid's lemma that $p_1$ must be one of the primes $q_i$ and
so $p_1\geq q_1$. Since $q_1|p_1^{\alpha_1}p_2^{\alpha_2}\dots p_k^{\alpha_k} $ the same
argument gives us that $ q_1\geq p_1$ and so $p_1=q_1$. So we can divide both sides of $p_1^{\
alpha_1}p_2^{\alpha_2}\dots p_k^{\alpha_2}=p_1^{\beta_1}q_2^{\beta_2}\dots q_l^{\beta_l} $ by
$p_1$ to get $p_1^{\alpha_1-1}p_2^{\alpha_2}\dots p_k^{\alpha_2}=p_1^{\beta_1-1}q_2^{\beta_2}\
dots q_l^{\beta_l}$. We keep dividing by $p_1$ until we run out of $p_1's$ on one side. There can't be
any $p_1's$ left over on either side or we would contradict Euclid's Lemma. So $\alpha_1=\beta_1$.
We continue this dividing process with $p_2$ to get that $q_2=p_2$ and $\alpha_2=\beta_2$. We
continue with dividing process with each successive $p_i=q_i$ until we finish with one side being $1$.
Then the other side must be $1$ as well and this forces $k=l$, and these factorizations to be exactly the
same.\end{proof}

\vspace{.2in}

\begin{definition} A class $[a]\in\mathbb{Z}/n$ is said to be a unit if $[a]x=[1]$ has a solution in $\

mathbb{Z}/n$. We also say $a\in\mathbb{Z}$ is a unit modulo $n$ if $ax\equiv 1\pmod{n}$ has a
solution. Note that these are essentially the same concept, one is expressed using the language of
equivalence classes, the other using the equivalence relation itself. We write $(\mathbb{Z}/n)^*$ for
the set of classes of units. Note that $[1]$ and $[-1]=[n-1]$ are always units and $[0]$ is never a unit.\
end{definition}

\vspace{.2in}

\begin{example} In $\mathbb{Z}/6=\{[0],[1],[2],[3],[4],[5]\}$ there are exactly two units, $[1]$ and

$[5]=[-1]$. We can rule out $[2],[3],[4]$, as follows: suppose $[2]x=[1]$ had a solution. Then $[3]
[2]x=[3][1]$ or $[0]x=[3]$. But the left hand side is $[0]$ for any $x$ and this is a different class than
$[3].$
\end{example}

\vspace{.2in}

\begin{example} In $\mathbb{Z}/10=\{[0],[1],[2],\dots,[8],[9]\}$ we find four units,

$(\mathbb{Z}/10)^*=\{[1],[3],[7],[9]\}.$
\end{example}
\vspace{.2in}

\begin{theorem}$a$ is a unit modulo $n$ iff $\gcd(a,n)=1$.

\end{theorem}

\vspace{.2in}

\begin{proof} Suppose that $ax\equiv 1\pmod{n}$ has a solution. Then $\exists

j\in\mathbb{Z}$ so that $ax-1=jn$. But then $ax+(-j)n=1$ is a positive linear combination of $a$ and
$n$ and since it is one it must be the greatest common divisor by definition. Likewise if $\gcd(a,n)=1$
then $\exists s,t\in\mathbb{Z}$ so that $as+nt=1$. But then $as=1-nt\equiv 1\pmod{n}$ shows that
$x=s$ is a solution to $ax\equiv 1\pmod{n}$ and hence $a$ is a unit modulo $n$.\end{proof}

\vspace{.2in}

\begin{theorem} If $a,b$ are both units modulo $n$ then so is $ab$. If $a_1,a_2,\dots, a_k$ are units
modulo $n$ where $k\in\mathbb{N}^+$ then $a_1a_2....a_k$ is a unit modulo $n$.
\end{theorem}

\vspace{.2in}

\begin{proof} We know $ax\equiv 1\pmod{n}$ and $by\equiv 1\pmod{n}$ both have solutions $x,y\
in\mathbb{Z}$. This means that $\exists j,k\in\mathbb{Z}$ so that $ax-1=jn, by-1=kn$ and so $(ab)
(xy)=(ax)(by)=(1+jn)(1+kn)=1+(j+k)n+(jk)n^2=1+n(\text{stuff})$ where $\text{stuff}\in\mathbb{Z}$
and so $(ab)(xy)\equiv 1 \pmod{n}$ and hence $ab$ is a unit. The second claim is accomplished by
induction on $k$ (do this...).
\end{proof}

\vspace{.2in}

\begin{definition} Let $a\in\mathbb{Z}$ and $n\geq 2$. If $a$ is a unit modulo $n$ (or equivalently if
$[a]$ is a unit in $(\mathbb{Z}/n)^*$ and $ax\equiv 1 \pmod{n}$ or $[a]x=[1]$ in $\mathbb{Z}/n$,
we write $a^{-1}\equiv x\pmod{n}$ or $[a]^{-1}=[x]$ and say $x$ ( or $[x]$) is the \emph{inverse}
of $a$ modulo $n$.
\end{definition}

\vspace{.2in}

\begin{example} Let $n=15$. We know that $(\mathbb{Z}/15)^*=$

$\{[1],[2],[4],[7],[8],[11],[13],[14]\}$ by checking greatest common divisors. We check that

$$
[1]^{-1}=[1],[2]^{-1}=[13],[4]^{-1}=[4],[7]^{-1}=[13],
$$
$$
[11]^{-1}=[11],[13]^{-1}=[2], [14]^{-1}=[14].
$$
\end{example}

\vspace{.2in}

\begin{example}
Let $n=22$.

We know that $(\Z/22)^*=\{[1],[3],[5],[7],[9],[13],[15],[17],[19],[21]\}$. We draw an arrow diagram

connecting each unit with its inverse:

$$
\begin{tikzcd}[column sep=1em,row sep=8ex]
1\arrow[d] & 3\arrow[drrrrr] & 5\arrow[drr] & 7 \arrow[drrrrr]& 9\arrow[dll] & 13\arrow[drr] & 15\
arrow[dlllll] & 17\arrow[dll] & 19\arrow[dlllll] & 21\arrow[d]\\
1 & 3 & 5 & 7 & 9 & 13 & 15 & 17 & 19 & 21\end{tikzcd}
$$

We observe some patterns, the first being that there is left/right symmetry in this picture. This comes
from the identity $xy\equiv 1\pmod{22} \iff (-x)(-y)\equiv 1\pmod{22}$, for instance, we see that $5\
cdot 9\equiv 1\pmod{22}$ and hence $(-5)(-9)\equiv 1\pmod{22}$. But $-5\equiv 17\pmod{22}$ and
$-9\equiv 13\pmod{22}$ and so we see that the arrow $5\to 9$ produces the arrow $13\to 19$ when we
reflect the diagram about its center vertical axis. Note that the function that sends a unit to its inverse is
also a permutation of the set of units.

\end{example}

\vspace{.2in}

If $p$ is a prime number then $1,2,3,\dots, p-1$ are all units modulo $p$ in other words:

$$
(\mathbb{Z}/p)^*=\{[1],[2],\dots,[p-1]\}.
$$

\vspace{.2in}

\begin{lemma}[Cancellation mod $n$] Let $n\geq 2$ and $a\in\mathbb{Z}$ with $\gcd(a,n)=1$. Then
$ax\equiv ay\pmod{n}$ implies that $x\equiv y\pmod{n}$. In other words, we can cancel units modulo
$n$.
\end{lemma}

\vspace{.2in}

\begin{proof} Suppose that $ax\equiv ay\pmod{n}$ so that $ax-ay=a(x-y)=jn$ for some $j\in\
mathbb{Z}$. Since $\gcd(a,n)=1$ we know $a\not=0$ and $a(x-y)$ is a product equal to a multiple of
$n$, non of the prime factors of $a$ can divide $n$ and so they must divide $j$, in other words, $a|j$.
But then $\frac{j}{a}\in\mathbb{Z}$ and so $x-y=\frac{j}{a}n$ gives that $x\equiv y\pmod{n}.$
\end{proof}

\vspace{.2in}

\begin{theorem}[Fermat's Little Theorem] Let $p$ be a prime and let $a\in\mathbb{Z}$ be a unit
modulo $p$. Then $a^{p-1}\equiv 1\pmod{p}.$
\end{theorem}

\vspace{.2in}

\begin{proof} Examine the set $S=\{[1],[2],[3],\dots [(p-1)]\}.$ Each element of this set is a unit
modulo $p$ and no two of these units are equal. Examine the set $S'=[a]S:=\{[1a],[2a],[3a],\dots, [(p-
1)a]\}$ where we multiply each of the $p-1$ units in $S$ by $[a]$. This list consists of units as well
and we claim that no two members of this list are equal. If we had $[ka]=[la]$ then by the cancellation
property we would have $[k]=[l]$. So the list $S'$ also consists of $p-1$ distinct units modulo $p$.
Since there are only $p-1$ distinct units up to congruence modulo $p$, these lists contain the same
members, perhaps in a different order. We multiply all the members of $S$ together to get $[1]\cdot[2]\
cdots\cdot[(p-1)]=[(p-1)!]$. We multiply all members of $S'$ together to get $[1a]\cdot[2a]\cdots\
cdot[(p-1)a]=[a]^{p-1}[(p-1)!]$. These products must be equal so we have

$$
[(p-1)!]=[a]^{p-1}[(p-1)!].
$$

But $[(p-1)!]$ is a product of units and hence a unit, and so we can cancel it from both sides of the
above equation to get

$$
[1]=[a]^{p-1}
$$

which can be rephrased as $a^{p-1}\equiv 1\pmod{p}$.\end{proof}

\vspace{.2in}

\begin{corollary}[Fermat's Little Theorem again] If $p$ is a prime and $a\in\mathbb{Z}$ then $a^p\
equiv a\pmod{p}$.
\end{corollary}

\vspace{.2in}

\begin{proof} If $\gcd(a,p)=p$ then $a^p\equiv a\equiv 0\pmod{p}$ while

if $\gcd(a,p)=1$ this follows from the above form of FLT by multiplying both sides by $a$.\end{proof}

\vspace{.2in}

\begin{example} Let's use Fermat's Little Theorem (or FLT) to find the inverse of the unit $13$
modulo $17$. FLT tells us that $13^{16}\equiv 1\pmod{17}$ or $13(13^{15})\equiv 1 \pmod{17}$
and hence $[13]^{-1}=[13]^{15}$. We can simplify this pretty easily: $13\equiv -4\pmod{17}$ and so
$13^2\equiv (-4)^2\equiv 16\equiv -1\pmod{17}$. Then $13^4\equiv (-1)^2\equiv 1\pmod{17}$ and
$13^8\equiv 1^2\equiv 1\pmod{17}.$ Putting this altogether we find that
$13^{15}=(13)^8(13)^4(13)^2(13)\equiv (1)(1)(-1)(-4)\equiv 4\pmod{17}$. So $[13]^{15}=[4]$ and
hence $[13]^{-1}=[4]$. We check that $(13)(4)=52\equiv 1\pmod{17}$.
\end{example}

\vspace{.2in}

What we saw in the last example is that the inverse of $[a]\in (\mathbb{Z}/p)^*$ is $[a]^{p-2}$ based
on Fermat's Little Theorem.

\vspace{.2in}

\begin{example} Let's find the inverse of the unit $23$ modulo $31$. Using the idea mentioned above
we see that $[23]^{29}=[23]^{-1}$. Now $23\equiv -8\pmod{31} $ and so $23^2\equiv (-8)^2\equiv
64\equiv 2\pmod{31}$ and so $23^4\equiv 2^2\equiv 4\pmod{31}$, $23^8\equiv 4^2=16\pmod{31}$,
$23^{16}\equiv 16^2=256\equiv 8\pmod{31}$. Altogether we get that

$$
23^{29}=(23^{16})(23^8)(23^4)(23)\equiv (8)(16)(4)(-8)\equiv -2^{12}\pmod{31}.
$$

Since $2^5=32\equiv 1\pmod{31}$ we have $-2^{12}=-2^2(2^5)^2\equiv (-4)(1)^2\equiv 27\

pmod{31}$ giving us that $[23]^{-1}=[27]$. We check that $(23)(27)\equiv(-8)(-4)\equiv 32\equiv 1\
pmod{31}$.
\end{example}

\vspace{.2in}

This might seem like a cumbersome way of finding inverses of units, and it is restricted to finding
inverses modulo a prime (we will see shortly that Fermat's Little Theorem has a generalization by Euler
to include composite numbers as well; in fact, Euler gave the first proof of Fermat's Little Theorem
when he generalized it). There are other much older ways we can find inverses, one way in particular
begins with the following famous theorem/algorithm:

\vspace{.2in}

\begin{theorem}[Euclidean Algorithm] Let $a,b\in\mathbb{N}^+$. Apply the Division Algorithm

successively to get:

$$
a=bq_1+r_1 \qquad 0\leq r_1<b
$$
$$
b=q_2r_1+r_2 \qquad 0\leq r_2<r_1
$$
$$
r_1=q_3r_2+r_3 \qquad 0\leq r_3<r_2
$$

\centerline{etc..}

$$
r_{k-1}=q_{k+1}r_k+r_{k+1} \qquad 0\leq r_{k+1}<r_k
$$

Eventually there will be some first $r_n=0$. Then $r_{n-1}=\gcd(a,b).$

\end{theorem}

\vspace{.2in}

\begin{proof} The fact that there will be some first $r_n=0$ follows from the fact that the remainders
in each step get strictly smaller while remaining non-negative. So we just need to establish that $r_{n-
1}=\gcd(a,b)$. We see immediately that $r_{n-2}=q_nr_{n-1}$ and so $r_{n-1}|r_{n-2}$ by
definition. In turn $r_{n-3}=q_{n-1}r_{n-2}+r_{n-1}$ and since $r_{n-1}|r_{n-1}$ and $r_{n-1}|
r_{n-2}$ we know that $r_{n-1}|(q_{n-1}r_{n-2}+r_{n-1})$ or $r_{n-1}|r_{n-3}$. We keep going
with this process to get that $r_{n-1}|r_k$ for each $1\leq k\leq n-1$, and so $r_{n-1}|b$ and $r_{n-1}|
a$ as well. So $r_{n-1}$ is a common factor of $a$ and $b$ and hence by the GCD theorem $r_{n-1}|\
gcd(a,b)$. Let $g=\gcd(a,b)$ and note that $g|(a-bq_1)$ or $g|r_1$. Likewise $g|(b-q_2r_1)$ or $g|
r_2$. Repeating this and working backwards we see that $g|r_k$ for any $k$ and so $g|r_{n-1}$. Since
$g|r_{n-1}$ and $r_{n-1}|g$ we have $g=r_{n-1}.$ \end{proof}

\vspace{.2in}

\begin{example} Let's put this to use: let's compute $\gcd(234,42)$, letting $a=234, b=42$ in the EA
we find that $234=(5)(42)+24,42=(1)(24)+18, 24=(1)(18)+6, 18=(3)(6)+0$ and so $\gcd(234,42)=6$
which we know is correct.
\end{example}

\vspace{.2in}

\begin{example} Let's show that $22$ is a unit modulo $47$ by using the Euclidean Algorithm to show
$\gcd(22,47)=1$. $47=(2)22+3, 22=(7)3+1$ and so $\gcd(22,47)=1.$
\end{example}

\vspace{.2in}

Note that the Euclidean Algorithm doesn't require that we know how to factor $a$ and $b$, it just
requires that we can apply the division algorithm repeatedly, a much much simpler process. We can
also use it to find inverses modulo $n$ as the next example shows.

\vspace{.2in}

\begin{example} In the example above where we compute

$\gcd(22,47)=1$ we can solve for the remainders to get that $1=22-7(3), 3=47-2(22)$. Substituting the
second equation into the first we get $1=22-7(47-2(22))=(15)(22)+(-7)(47)$. Note that what we have
just done is to express our greatest common divisor, in this case $1$, as an explicit linear combination
of $22$ and $47$. Then $1\equiv (15)(22)+(-7)(47)\equiv (15)(22)\pmod{47}$ and so $[22]^{-
1}=[15]$ in $\mathbb{Z}/47.$
\end{example}

\vspace{.2in}

\begin{example} We find $[45]^{-1}$ in $\mathbb{Z}/101$. $101=2(45)+11, 45=4(11)+1$ and so we

get $1=45-4(101-2(45))=(9)(45)+(-4)(101)$ giving us $[45]^{-1}=[9]$. Let's
compute $[2]^{-1}$ in $\mathbb{Z}/101$. $101=2(50)+1$ and so $1=(2)(-50)+(1)(101)$ and so
$[2]^{-1}=[-50]=[51]$ in $\mathbb{Z}/101$.\end{example}

\vspace{.2in}

We can use this method of finding inverses to solve linear congruences modulo $n$ where by a linear
congruence we mean a congruence of the form $ax\equiv b\pmod{n}$ where $a,b\in\mathbb{Z}$. We
first make some general observations about such congruences.

\vspace{.2in}

\begin{example} Suppose we are looking to solve $6x\equiv 4\pmod{10}$. This is a fair bit different
than solving the linear equation $6x=4$ in $\mathbb{Z}$, which has no solution, or the linear equation
$6x=4$ in $\mathbb{R}$ which has a unique solution $x=2/3$. We see that $x=4$ and $x=9$ are both
solutions to $6x\equiv 4\pmod{10}$ and moreover, any number congruent to either of these two
solutions modulo $10$ is also a solution.
\end{example}
\vspace{.2in}

\begin{theorem}Let $g=\gcd(a,n)$ where $a\in\mathbb{Z}$ and $n\geq 2$. Then the linear
congruence $ax\equiv b\pmod{n}$ has no solutions if $g\not| b$ and has $g$ solutions up to
equivalence modulo $n$ if $g|b$.\end{theorem}

\vspace{.2in}

\begin{proof} Solving $ax\equiv b\pmod{n}$ is equivalent to finding a solution to $ax-b=ny$ or $ax-

ny=b$ where $x,y\in\mathbb{Z}$. If $g\not|b$ then this is impossible since $g$ divides $a$ and $n$. If
$g|b$ then we solve $ax-ny=b$ using the Euclidean Algorithm to find a solution $x=x_0,y=y_0$. Then
$x=x_t:=x_0+(\frac{n}{g})t, y=y_t:=y_0+(\frac{a}{g})t$ are solutions where $0\leq t<g$ since $ax_t-
ny_t=a(x_0+(\frac{n}{g})t)-n(y_0+(\frac{a}{g})t)=(ax_0-ny_0)+at\frac{n}{g}-nt\frac{a}{g}=b$.
Moreover these $g$ solutions are not equivalent modulo $n$ in the following sense: if $x_k\equiv x_l\
pmod{n}$ where $0\leq k,l<g$ then $n|(x_k-x_l)$. But $|x_k-x_l|=|(x_0+(\frac{n}{g})k)-(x_0+(\
frac{n}{g})l)|=\frac{n}{g}|k-l|$. Since $0\leq k,l<g$ we know $|k-l|<g$ and so $0\leq|x_k-x_l|<\
frac{n}{g}g=n$ and so $x_l=x_k$ since this is the only multiple of $n$ in this range.
\end{proof}

\vspace{.2in}

Looking back at $6x\equiv 4\pmod{10}$ the above theorem tells us that since $\gcd(6,10)=2$ does
indeed divide $4$ there are two solutions to this congruence up to equivalence. Let's use the Euclidean
Algorithm to find one of these, namely a solution to $6x-10y=4$. Well we can divide both sides by $2$
and solve $3x-5y=2$. $5=3(1)+2$ and so $2=3(-1)-5(-1)$ gives us $x_0=-1=y_0$. Then $x_1=-1+\
frac{10}{2}1,y_1=-1+\frac{6}{2}(1)$ or $x_1=4, y_1=2$. The two solutions up to equivalence to
$6x\equiv 4\pmod{10}$ are $x_0=-1\equiv 9\pmod{10}, x_1=4$

\vspace{.2in}

If we start to examine a \emph{system} of linear congruences things get a bit more complicated. For
example:

$$
x\equiv 3\pmod{4}
$$
$$
x\equiv 10\pmod{15}
$$
$$
x\equiv 5\pmod{7}
$$

does have solutions $x\in\mathbb{Z}$ that satisfy all three congruences at the same time, but the
smallest positive solution is $x=355$. Thankfully under the right circumstances these systems can often
be solved very efficiently due to a theorem first used in 3rd-century China and brought to Europe a
thousand years later by Leonardo De Pisa (Fibonacci). The following theorem gives a sufficient
condition for a system of linear congruences to have a solution, and its proof gives an algorithm for
finding such a solution.

\vspace{.2in}

\begin{theorem}[Chinese Remainder Theorem] Let $n_1,n_2,\dots, n_k$ be a collection of pairwise

relatively prime natural numbers and $b_1,b_2,\dots, b_k\in\mathbb{Z}$. Then the system

$$
x\equiv b_1\pmod{n_1}
$$
$$
x\equiv b_2\pmod{n_2}
$$
\centerline{....}

$$
x\equiv b_k\pmod{n_k}
$$

has a solution $x$ and moreover this solution is unique up to congruence modulo $N=n_1n_2..n_k$.
\end{theorem}

\vspace{.2in}
\begin{proof} Let $N=n_1n_2...n_k$ and set $m_i:=N/n_i$ for $i=1,2,\dots,k$. Then $\
gcd(m_i,n_j)=1$ if $i=j$ and $\gcd(m_i,n_j)=m_i$ if $i\not=j$, in other words $m_i$ is a unit modulo
$n_i$. Let $[m_i]^{-1}=[y_i]$ in $\mathbb{Z}/n_i$ for $i=1,2,\dots, k$. Set
$x=b_1y_1m_1+b_2y_2m_2+\dots +b_ky_km_k$. We check that

$$
x=b_1y_1m_1+b_2y_2m_2+\dots +b_ky_km_k\equiv b_i\pmod{n_i}
$$

for each $1\leq i\leq k$ since if $j\not=i$ then $m_j\equiv 0\pmod{n_i}$ and $b_iy_im_i\equiv b_i\
pmod{n_i}$ since $[y_i][m_i]=[1]$ in $\mathbb{Z}/n_i$.

To show uniqueness up to congruence modulo $N$ let $x$ and $y$ both be solutions. For each $1\leq
i\leq k$, $x-y$ is divisible by $n_i$ since $x-y\equiv b_i-b_i\equiv0\pmod{n_i}$ and since the $n_i$
are relatively prime we get that $x-y$ is divisible by $N$.
\end{proof}

\vspace{.2in}

\begin{example} Let's revisit the linear system

$$
x\equiv 3\pmod{4}
$$
$$
x\equiv 10\pmod{15}
$$
$$
x\equiv 5\pmod{7}
$$

where we see the moduli are relatively prime. In this case $n_1=4,n_2=15, n_3=7$ and so $N=420$
and $m_1=105,m_2=28, m_3=60$. First we find $y_1$ by computing $[105]^{-1}=[1]^{-1}=[1]$ in
$\mathbb{Z}/4$ to get $y_1=1$. We find $y_2$ by computing $[28]^{-1}=[13]^{-1}=[7]$ in $\
mathbb{Z}/15$ to get $y_2=7$. Lastly we find $y_3$ by computing $[60]^{-1}=[4]^{-1}=[2]$ in $\
mathbb{Z}/7$ to get $y_3=2$. We assemble our solution
$x=b_1y_1m_1+b_2y_2m_2+b_3y_3m_3=(3)(1)(105)+(10)(7)(28)+(5)(2)(60)=2875\equiv 355\
pmod{420}$.
\end{example}

\vspace{.2in}

We mentioned earlier that Fermat's Little Theorem's first published proof was by Euler who in fact
generalized the theorem to arbitrary natural numbers; in fact the proof we gave for FLT will easily
generalize with a minimum of modifications. We introduce Euler's totient or phi function for this
purpose.

\vspace{.2in}
\begin{definition} Let $n\geq 2$ be a natural number. Define $\phi(n)$ to be the size of the following
set:

$$
\{k\in\mathbb{N}\,|\, 1\leq k<n, \gcd(k,n)=1\},
$$

in other words $\phi(n)$ is the size of the set $(\mathbb{Z}/n)^*$, the set of units modulo $n$. For
completion we set $\phi(1)=1$ thus defining a function $\phi\,:\,\mathbb{N}^+\rightarrow \
mathbb{N}^+$. This is called Euler's totient or phi function. \end{definition}

\vspace{.2in}

\begin{example}Let $p$ be a prime. Then we have already seen that every number in the set $\{1,2,3,\
dots,p-1\}$ is a unit modulo $p$, there are $p-1$ elements of this set, and so $\phi(p)=p-1$. By looking
at the set $\{1,2,\dots, p,p+1,\dots, 2p, \dots p^k-1\}$ where $k\in\mathbb{N}$ we see that the only
nonunits are $p,2p,\dots (p-1)p$ and so $\phi(p^k)=(p^k-1)-(p-1)=p^k-p=p^{k-1}(p-1)$. Furthermore
we state here without proof the following nice property of $\phi$: if $a,b\in\mathbb{N}^+$ satisfy $\
gcd(a,b)=1$ then $\phi(ab)=\phi(a)\phi(b)$ (the proof of this uses the Chinese Remainder Theorem).
We can check by hand simple facts like $\phi(pq)=(p-1)(q-1)$ where $p,q$ are distinct primes.\
end{example}

\vspace{.2in}

\begin{theorem}
Let $a,b\in \N^+$ with $\gcd(a,b)=1$. Then $\phi(ab)=\phi(a)\phi(b)$.
\end{theorem}

\vspace{.2in}

\begin{proof}
Let $S_1=\{n\, |\,1 \leq n<ab\}$ and note that $S_1$ has $\phi(ab)$ elements by definition of $\phi$.
Let $S_2=\{(k,l)\,|\, 1\leq k<a, \gcd(a,k)=1, 1\leq l<b, \gcd(b,l)=1\}$. Since there are $\phi(a)$ possible
choices for $k$ and $\phi(b)$ possible choices for $l$ we see that $S_2$ has $\phi(a)\phi(b)$ elements.
Define $f:S_1\rightarrow S_2$ by $f(n)=(n\mod a,n\mod b)$. We'll show that $f$ is a bijection. First if
$f(n_1)=f(n_2)$ where $1\leq n_1,n_2<ab$ then
$$
n_1\equiv n_2\pmod{a}
$$
$$
n_1\equiv n_2\pmod{b}
$$
and so $a|n_1-n_2$ and $b|n_1-n_2$. But since $\gcd(a,b)=1$ we know $ab|n_1-n_2$ so $n_1\equiv
n_2\pmod{ab}$. But the only way this can happen is if $n_1=n_2$ and so $f$ is $1-1$. To show $f$ is
onto let $(k,l)\in S_2$. We need to find $n$ so that $n\equiv k\pmod{a}, n\equiv l\pmod{b}$. But such
a solution always exists by the Chinese Remainder Theorem. So $f$ is a bijection. We'll see later in
Chapter $7$ that bijections preserve size and so the sizes of $S_1, S_2$ are the same and hence $\
phi(ab)=\phi(a)\phi(b)$.
\end{proof}

\vspace{.2in}

\vspace{.2in}

\begin{theorem}
Let $n=p_1^{a_1}p_2^{a_2}\dots p_k^{a_k}$ be the prime factorization of $n\geq 2$. Then

$$
\phi(n)=n\prod_{i=1}^k\left(1-\frac{1}{p_i}\right)
$$
\end{theorem}

\vspace{.2in}

\begin{proof}
We have $\phi(n)=\phi(p_1^{a_1}p_2^{a_2}\dots p_k^{a_k})=\phi(p_1^{a_1})\phi(p_2^{a_2})\dots \
phi(p_k^{a_k})$. Dividing by $n$ we get

$$
\frac{\phi(n)}{n}=\frac{(p_1-1)(p_2-1)\dots (p_k-1)p_1^{a_1-1}p_2^{a_2-1}\dots p_k^{a_k-1}}
{p_1^{a_1}p_2^{a_2}\dots p_k^{a_k}}=
$$
$$
\left(\frac{p_1-1}{p_1}\right)\left(\frac{p_2-1}{p_2}\right)\dots \left(\frac{p_k-1}{p_k}\right)=\
prod_{i=1}^k\left(1-\frac{1}{p_i}\right).
$$
\end{proof}

\begin{example}
Let $n=9$ so that the units modulo $9$ are, up to equivalence, $(\Z/9)^*=\{1,2,4,5,7,8\}$. Let $a=2$
and look at the function $p_2:(\Z/9)^*\rightarrow (\Z/9)^*$ defined by $p_2(x)=2x\mod{9}$; here
we're being somewhat relaxed with notation, what we really mean is $p_2([x])=[2x]$, but when there's
no confusion we'll associate $x$ with its equivalence class $[x]$. We draw an arrow diagram of this
function:

$$
p_2\quad \begin{tikzcd}[column sep=1em,row sep=2ex]
1\arrow[dr] & 2\arrow[dr] & 4\arrow[drrr] & 5\arrow[dlll] & 7\arrow[dl] & 8\arrow[dl]\\
1 & 2 & 4 & 5 & 7 & 8\end{tikzcd}
$$
\end{example}

We see that this function is a bijection, i.e. a permutation of the set of units. Let $a=5$ and let $p_5:(\
Z/9)^*\rightarrow (\Z/9)^*$ be defined by $p_2(x)=5x\mod{9}$. It has arrow diagram

$$
p_5\quad \begin{tikzcd}[column sep=1em,row sep=2ex]
1\arrow[drrr] & 2\arrow[dl] & 4\arrow[dl] & 5\arrow[dr] & 7\arrow[dr] & 8\arrow[dlll]\\
1 & 2 & 4 & 5 & 7 & 8\end{tikzcd}
$$
which is clearly another permutation of $(\Z/9)^*$. We encourage the reader to draw diagrams for
$p_4,p_7,p_8$ to see that they are indeed permutations.

\vspace{.2in}

\begin{example} Let $n=15$ so that $(\Z/15)^*=\{[1],[2],[4],[7],[8],[11],[13],[14]\}$. Let $a=4$ so

that $\gcd(4,15)=1$ and look at $p_4:(\Z/15)^*\rightarrow
(\Z/15)^*$ defined by $p_4(x)=4x\pmod{15}$. It has arrow diagram

$$
p_4\quad \begin{tikzcd}[column sep=1em,row sep=3ex]
1 \arrow[drr]& 2\arrow[drrr] & 4\arrow[dll] & 7\arrow[drrr] & 8\arrow[dlll] & 11\arrow[drr] & 13\
arrow[dlll] & 14\arrow[dll]\\
1 & 2 & 4 & 7 & 8 & 11 & 13 & 14\end{tikzcd}
$$

which is another permutation of $(\Z/15)^*$. We encourage the reader to draw $p_7, p_8,p_{11}$
etc...

\end{example}

\vspace{.2in}

We'll see in the next theorem that if $\gcd(a,n)=1$, so that $a$ is a unit mod $n$, the function $p_a:(\
Z/n)^*\rightarrow (\Z/n)^*$ defined by $p_a(x)=ax\pmod{n}$ is a permutation and gives us an
extension of Fermat's Little Theorem:

\vspace{.2in}

\begin{theorem}[Euler's Theorem] Let $n\in\mathbb{N}^+$ and $a\in\mathbb{Z}$ satisfying $\

gcd(a,n)=1$ (so that $a$ is a unit modulo $n$). Then

$$
a^{\phi(n)}\equiv 1\pmod{n}
$$
\end{theorem}

\vspace{.2in}

\begin{proof} Examine the set $S=\{u_1,u_2,u_3,\dots u_{\phi(n)}\}$ where each element of this set
is a unit modulo $n$ so that $1\leq u_i<n$ and no two of these units are equal. Examine the set

$S'=aS:=\{au_1,au_2,au_3,\dots, au_{\phi(n)}\}$ where we multiply each of the $\phi(n)$ units in $S$

by $a$. This list consists of units as well and we claim that no two members of this list are equivalent
modulo $n$. If we had $au_k\equiv au_i\pmod{n}$ then by the cancellation property we would have
$u_k\equiv u_l\pmod{n}$. So the list $S'$ also consists of $\phi(n)$ distinct units modulo $n$. Since
there are only $\phi(n)$ distinct units up to congruence modulo $n$, these lists contain the same
members up to equivalence, perhaps in a different order. We multiply all the members of $S$ together
to get $u_1\cdot u_2\cdots\cdot u_{\phi(n)}:=u$. We multiply all members of $S'$ together to get $
(au_1)\cdot (au_2)\cdots\cdot(au_{\phi(n)})=a^{\phi(n)}u$. These products must be equivalent so we
have

$$
u\equiv ua^{\phi(n)}.
$$

But $u$ is a product of units and hence a unit, and so we can cancel it from both sides of the above
equivalence to get

$$
1\equiv a^{\phi(n)}\pmod{n}.
$$
\end{proof}

\vspace{.2in}

\begin{example} Suppose we want to find the last three digits of the number $13457^{89621}$. First
we note that $13457\equiv 457\pmod{1000}$. We also see that $\phi(1000)=400$ so we know that
since $\gcd(457,1000)=1$ that $457^{\phi(1000)}\equiv 1\pmod{1000}$ or $457^{400}\equiv 1\
pmod{1000}$. Then $457^{89621}=(457^{400})^{224}(457^{21})\equiv (1^{224})
(457^{21})=457^{21}\pmod{1000}$. Now $457^2\equiv 849\equiv-151\pmod{1000}, 457^4\equiv(-
151)^2\equiv 801\equiv -199\pmod{1000}, 457^8\equiv (-199)^2\equiv 601\equiv -399\pmod{1000}$
and so $457^{16}\equiv (-399)^2\equiv 201\pmod{1000}$ giving us that:

$$
457^{21}=(457^{16})(457^4)(457^1)\equiv (201)(801)(457)\equiv 457\pmod{1000}.
$$

Altogether we see that $12457^{89621}\equiv 457\pmod{1000}$ and so the last three digits are $457$.
The number $12457^{89621}$ has $367036$ digits.\end{example}

\newpage

\section{Recursion and Strong Induction}

\vspace{.2in}

The proof of the Fundamental Theorem of Arithmetic as mentioned earlier requires a form of induction
a bit more flexible that our Principle of Mathematical Induction (first versions) from earlier. Sometimes
this second version is called 'strong' induction but it is logically equivalent.

\vspace{.2in}

\begin{theorem}[Principle of Mathematical Induction (version 2)] Let $P_n$ be a predicate statement

that depends on $n\in\mathbb{N}$. Then to prove $\forall n\in\mathbb{N}\, P_n$ is suffices to
establish:

$$
1. \quad P_0
$$
$$
2.\quad \forall k\in\mathbb{N} \land (0\leq j\leq k), P_j\implies P_{n+1}
$$
\end{theorem}

\vspace{.2in}

Condition $2$ can also be written as $\forall k\in\mathbb{N}$, $(P_0\land P_1\land\dots \land P_k)\
implies P_{k+1}$. Notice we are allowed to assume not just $P_k$ as inductive hypothesis but all
previous statements making our inductive hypothesis stronger. If we want an even more general version
that allows us to prove $\forall n\in\mathbb{Z}, n\geq a, P_n$ we can do so by first establishing $P_a$
and then showing that for any $k\in\mathbb{Z}$ so that $a\leq k$ we have that $(P_a\land P_{a+1}\
land\dots\land P_k)\implies P_{k+1}$.

\vspace{.2in}

As our first of many examples to come we prove the existence part of the Fundamental Theorem of
Arithmetic at last.

\vspace{.2in}

\begin{theorem}[Proof of Existence, FTA]Let $n$ be a natural number greater than $1$. Then $n$ can
be written as the product of primes.
\end{theorem}

\vspace{.2in}

\begin{proof} Let $P_n$ be the statement that $n$ can be written as a product of primes. Our base case
is $P_2$, and we check that $2$ can be written as a product of primes. Well $2$ \emph{is} prime and
so itself is the product of a single prime. Assume that $P_j$ is true for all $2\leq j\leq k$ where $k$ is
an integer greater than or equal to $2$. We look at $k+1$. Either $k+1$ is prime, in which case it is a
product of primes, or it is of the form $k+1=ab$ where $a,b$ are natural numbers greater than $2$ and
hence less than $k$ (why?). Well $P_a$ tells us that $a$ can be written as a product of primes and
$P_b$ tells us that $b$ can be written as a product of primes, and so $ab=k+1$ is a product of a
product of primes...and so is a product of primes itself. So $P_{k+1}$ holds and we've established the
existence portion of the FTA.\end{proof}
\vspace{.2in}

Why did we need this second form of induction in the above proof, why wouldn't our first version of
the PMI work here? The factors of $n+1$ generally have no relation to the factors of $n$: think of
$n=99$ which has prime factorization $(3^2)(11)$ and $n+1=100$ which has prime factorization
$(2^2)(5^2)$.

\vspace{.2in}

\begin{example}We use this second form of induction to prove the following statement (you should try
a few examples for yourself): $\forall n\in\mathbb{N}^+$, $n$ can be written as a sum of \
emph{distinct} powers of $2$, i.e., $n=2^{p_1}+2^{p_2}+\dots +2^{p_t}$ where $0\leq p_1<p_2<\
dots <p_t$. Here is the proof: we proceed by induction on $n\in\mathbb{N}^+$, the base case being
that $1$ can be written as a sum of distinct powers of $2$; this is easy $1=2^0.$ Assume that for some
$k\in\mathbb{N}^+$ that each $1\leq j\leq k$ can be written as a sum of distinct powers of $2$ and
look at $k+1$. If $k+1$ is even then $1\leq \frac{k+1}{2}<k+1$ is a natural number and so $\
frac{k+1}{2}=2^{t_1}+2^{t_2}+\dots+2^{t_m}$ where $0\leq t_1<t_2<\dots <t_m$. But then
$k+1=2(2^{t_1}+2^{t_2}+\dots+2^{t_m})=2^{t_1+1}+2^{t_2+1}+\dots+2^{t_m+1}$ is a sum of
distinct powers of $2$ since $1\leq t_1+1<t_2+1<\dots <t_m+1$. Otherwise $k+1$ is odd and so $k$ is
even. Since $k<k+1$ we know that $k$ can be written as a distinct sum of powers of $2$:
$k=2^{a_1}+2^{a_2}+\dots +2^{a_l}$ where $0<a_1<a_2<\dots <a_l$ where here we can specify that
$a_1\not=0$ since $k$ is even. But then $k+1=1+2^{a_1}+2^{a_2}+\dots
+2^{a_l}=2^0+2^{a_1}+2^{a_2}+\dots +2^{a_l}$ is a way of writing $k+1$ as a sum of distinct
powers of $2$. In either case we are done and so the statement holds $\forall n\in\mathbb{N}^+$.
\end{example}

\vspace{.2in}

We can also define sequences using a process very much related to the PMI, the concept of a recursive
sequence which we will articulate with a few examples.

\vspace{.2in}

\begin{example} Let $x_0=0$ and let $x_{n+1}=x_n+(2n+1)$ for $n\geq 0$. Let's compute the first
few terms of this sequence: $x_1=x_0+2(0)+1=1, x_2=x_1+2(1)+1=4, x_3=x_2+2(2)+1=9,
x_4=x_2+2(4)+1=16$. Can you guess a formula for $x_n$ that only depends on $n$?
\end{example}

\vspace{.2in}

The sequence of numbers in the last example is an example of a \emph{recursive} sequence: future
terms of the sequence are defined using previously defined terms. Here's another example:

\vspace{.2in}

\textbf{Collatz Conjecture:}\quad Let $x_0=k\in\mathbb{N}^+$. Define

\begin{equation*}
x_{n+1} = \left\{
\begin{array}{ll}
\frac{x_n}{2} & \quad x_n \quad \text{even} \\
3x_n+1 & \quad x_n \quad \text{odd}
\end{array}
\right.
\end{equation*}

\vspace{.1in}

Say we start with $x_0=20$. Then $x_1=10,x_2=5,x_3=16, x_4=8,x_5=4,x_6=2,x_7=1,x_8=4,

x_9=2,\dots$ where we enter a cycle $4\rightarrow 2\rightarrow 1\rightarrow 4$. Let's try starting
value $x_0=56$. Then the sequence $x_0,x_1,x_2,\dots $ is given by $56\rightarrow28\rightarrow 14\
rightarrow7\rightarrow22\rightarrow11\rightarrow34\rightarrow17\rightarrow52\rightarrow26\
rightarrow13\rightarrow40\rightarrow20\rightarrow10\rightarrow5\rightarrow16\rightarrow8\
rightarrow4\rightarrow2\rightarrow1\rightarrow4\rightarrow2\rightarrow1\rightarrow4\dots$ and we
are right back at that same length three cycle. We encourage the reader to choose a few different $x_0$
and to make a conjecture. Currently it is not known whether all starting values $x_0$ lead to this cycle,
this is an open conjecture know as the Collatz Conjecture or the '$3x+1$ problem'.

\vspace{.2in}

There are many common objects in mathematics that are defined recursively or have natural recursive
definitions, we note a few of these now.

\vspace{.2in}

\begin{definition}We define $0!:=1$ and for $k\in\mathbb{N}$ we define $(k+1)!:=(k+1)k!$. This

defines the sequence of \emph{factorials} $1,1,2,6,24,120,\dots$\end{definition}

\vspace{.2in}

\begin{definition} Let $(a_n)_{n=0}^{\infty}$ be an infinite sequence of real numbers, i.e. $\forall i\

in\mathbb{N}, a_i\in\mathbb{R}$. We define $\sum_{k=0}^0a_k:=a_0$ and define $\
sum_{k=0}^{n+1}a_k:=(a_{n+1})+\sum_{k=0}^na_k.$ If our sequence starts at some $t\in\
mathbb{Z}$, namely that our sequence is $a_t,a_{t+1},a_{t+2}, \dots$ we easily define $\
sum_{k=t}^n a_k$ for $n\geq t$ by setting $\sum_{k=t}^ta_k:=a_t$ and $\
sum_{k=t}^{n+1}a_k:=(a_{n+1})+\sum_{k=t}^na_k.$ \end{definition}

\vspace{.2in}

The Division Algorithm is a special case of a more general

theorem involving polynomials. Let

$$
\R[x]=\{a_0+a_1x+a_2x^2+\dots +a_nx^n: a_i\in \R, n\in\N\}
$$
be the set of all polynomials in $x$ with integer coefficients. Given some $p(x)\in \R[x]$ where
$p(x)=a_0+a_1x+a_2x^2+\dots +a_nx^n$ with $a_n\not=0$ we write $deg(p)=n$ and call this the
degree of $p$. We call $a_n$ the \emph{leading coefficient} of $p$ and $a_nx^n$ the \emph{leading
term} of $p$. We leave undefined the degree of the identically zero polynomial $0(x)=0$. We define
addition, subtraction, and multiplication of polynomials as follows:

\vspace{.2in}

\begin{definition}
Let $p(x)=\sum_{i=0}^na_ix^i, q(x)=\sum_{j=0}^mb_jx^j$. Then

$$
(p\pm q)(x)=\sum_{i=0}^{\max(n.m)}(a_i\pm b_i)x^i,
$$

and

$$
(pq)(x)=\sum_{i=0}^{n+m}c_ix^i,
$$

where $c_i=\sum_{j=0}^ia_jb_{i-j}=a_0b_i+a_1b_{i-1}+\dots +a_{i-1}b_1+a_ib_0$.

\end{definition}

\vspace{.2in}

\begin{lemma}
Let $p,q\in\R[x]$ be non-zero. Then $deg(p\pm q)\leq \max\{deg(p),deg(q)\}$ and
$deg(pq)=deg(p)deg(q)$.
\end{lemma}

\vspace{.2in}

We leave the proof of this lemma to the reader. We are now ready to state and prove our extension of
the Division Algorithm.

\vspace{.2in}

\begin{theorem}[Division Algorithm for Polynomials] If $p,d\in \R[x]$ have $deg(p)=n, deg(d)=m$,

$n\geq m$, and $d\not=0$, then $\exists q,r\in \R[x]$ so that

$$
p(x)=d(x)q(x)+r(x)
$$

where $deg(q)=n-m$ and either $deg(r)<m$ or $r=0$. Moreover $q,r$ satisfying these conditions are
unique.

\end{theorem}
\vspace{.2in}

\begin{proof} Fix $m$ and proceed by induction on $n\geq 0$. The base case is when $n=0$ and since
$n\geq m$ we must have $m=0$ as well. If we let $q=d$ and $r=0$ we see that $p=dq+r$ holds.
Assume the theorem is true for all $k\leq n$ some $n\geq 0$ and let $p\in \R[x]$ have $deg(p)=n+1$.
Let $a_{n+1}\not=0$ and $b_m\not=0$ be the leading coefficients of $p$ and $d$ respectively so that
$p(x)=a_0+\dots +a_{n+1}x^{n+1}$ and $d(x)=b_0+\dots +b_mx^m$. Define $g(x)=p(x)-\
frac{a_{n+1}}{b_m}x^{n+1-m}d(x)$. Then

$$
g(x)=(a_0+\dots +a_{n+1}x^{n+1})-\frac{a_{n+1}}{b_m}x^{n+1-m}(b_0+\dots+ b_mx^m)=
$$
$$
a_0+\dots +a_nx^n+a_{n+1}x^{n+1}-\frac{b_0a_{n+1}}{b_m}x^{n+1-m}-\dots -a_{n+1}x^{n+1}
$$

has degree less than $n+1$ since the leading term of $p$ gets cancelled out. Suppose first that $deg(g)\
geq m$. Then we can apply our inductive hypothesis to $g$ and $d$ to get that there are $a,r\in\R[x]$
so that $g(x)=a(x)d(x)+r(x)$ with $deg(a)=deg(g)-m$ and either $deg(r)<m$ or $r=0$. But then $p(x)-
c(x)d(x)=a(x)d(x)+r(x)$ where $c(x)=-\frac{a_{n+1}}{b_m}x^{n+1-m}$. But this means that
$$
p(x)=(a(x)+c(x))d(x)+r(x)
$$
where $deg(a+c)=n+1-m$ and $deg(r)<m$ or $r=0$, and we are done. If we have $deg(g)<m$ we can
just write $p(x)=c(x)d(x)+g(x)$ and note that $deg(c)=n+1-m$ and $deg(g)<m$. This completes our
induction.

\vspace{.2in}

We now prove that $q, r$ satisfying the conditions in the theorem are uniquely determined. Suppose
that $p=qd+r=q'd+r'$ where $r,r'$ have degree less than $d$ or are the zero polynomial. Then $(qd+r)-
(q'd+r')=0$ and so $(q-q')d+(r-r')=0$. Assume that $q-q'\not=0$. Then $deg((q-q')d)\geq deg(d)$ by the
preceding lemma, and either $deg(r-r')<deg(d)$ or $r-r'=0$. But this cannot be the case since the sum
of two polynomials of different degrees cannot be the zero polynomial. So $q-q'=0$ which in turn
implies $q=q'$ and hence $r=r'$.
\end{proof}

\vspace{.2in}

\begin{corollary} Let $p\in \R[x]$ and $a\in\R$. Then there exists $q\in\R[x]$ so that $p(x)=(x-a)q(x)
+p(a)$.

\end{corollary}

\vspace{.2in}

\begin{proof}
Apply the division algorithm to $p$ and $d(x)=x-a$ to get $q,r$ so that $p=qd+r$ and
$deg(r)<deg(d)=1$ or $r=0$. So $r$ is constant and it's easy to see that $r=p(a)$.
\end{proof}

\vspace{.2in}

\begin{example}Let us prove that $\sum_{k=0}^n 2k+1=(n+1)^2$ by using the PMI where $n\in\
mathbb{N}$. We check that $1=2(0)+1=\sum_0^0 2k+1=(0+1)^2=1$ and so the base case holds.
Assume that $\sum_{k=0}^n2k+1=(n+1)^2$ for some $n\in\mathbb{N}$. Then

$$
\sum_{k=0}^{n+1}2k+1=\sum_{k=0}^n2k+1+2(n+1)+1=(n+1)^2+2(n+1)+2=(n+2)^2,
$$

and so we are done by the PMI. Note how natural it was to prove this statement by induction.
\end{example}

\vspace{.2in}

\begin{example}[Sum Induction Proofs]

Proving a formula of the form $\sum_{i=1}^nf(i)=g(n)$ for some functions $f,g$ by induction follows,
for the most part, a template. One first needs to check the base case which involves checking that $\
sum_{i=1}^1f(i)=f(1)$ equals $g(1)$. Once one has assumed that $\sum_{i=1}^nf(i)=g(n)$ holds for
some $n\geq 1$ one looks at $\sum_{i=1}^{n+1}f(i)$.

\begin{enumerate}
\item Step $1$: evaluate $i=n+1$ to get $\sum_{i=1}^{n+1}f(i)=f(n+1)+\sum_{i=1}^nf(i)$.
\item Step $2$: substitute the inductive hypothesis to get $f(n+1)+\sum_{i=1}^nf(i)=f(n+1)+g(n)$.
\item Step $3$: find a way to manipulate $f(n+1)+g(n)$ so that it equals $g(n+1)$.
\end{enumerate}

Hence through steps $1-3$ one gets $\sum_{i=1}^{n+1}f(i)=g(n+1)$ which completes the induction.
\end{example}

\vspace{.2in}

\begin{theorem}[Difference Theorem] Let $a,b\in\R$ and $n\in\N^+$. Then

$$
a^n-b^n=(a-b)\sum_{j=0}^{n-1}a^jb^{n-1-j}.
$$

\end{theorem}

\vspace{.2in}

\begin{proof}

Suppose that $a\not=b$ so that $(a-b)\not=0 $ (if $a-b$ the statement is obvious). We prove the
statement $\frac{a^n-b^n}{a-b}=\sum_{j=0}^{n-1}a^jb^{n-1-j}$ is true for all $n\geq 1$ by induction
on $n\geq 1$. We check the base case when $n=1$ and see $\frac{a^1-b^1}{a-b}=1$ and $\
sum_{j=0}^{0}a^jb^{-1-j}=1$ which are equal. Assume that $\frac{a^n-b^n}{(a-b)}=\sum_{j=0}^{n-
1}a^jb^{n-1-j}$ for some $n\geq 1$. Then

$$
\sum_{j=0}^na^jb^{(n+1)-1-j}=a^n+\sum_{j=0}^{n-1}a^jb^{n-j}=a^n+b\sum_{j=0}^{n-1} a^jb^{n-
1-j}=
$$
$$
a^n+b\frac{a^n-b^n}{a-b}=\frac{a^n(a-b)}{a-b}+\frac{ba^n-b^{n+1}}{a-b}=\frac{a^{n+1}-
b^{n+1}}{a-b}.
$$
and so we are done by the PMI
\end{proof}

\vspace{.2in}

Most students recognize a few summation formulas from when they see Riemann sums in first
semester calculus (though these are often forgotten). Some examples are:

$$
\sum_{k=1}^n 1=n
$$
$$
\sum_{k=1}^n k=\frac{n(n+1)}{2}
$$
$$
\sum_{k=1}^n k^2=\frac{n(n+1)(2n+1)}{6}
$$
$$
\sum_{k=1}^n k^3=\frac{n^2(n+1)^2}{4}=\left(\sum_{k=1}^n k\right)^2
$$

\vspace{.2in}

Each of these can be proven in an manner completely analogous to our proof that $\sum_{k=0}^n
2k+1=(n+1)^2$. But where do you get these formulas in the first place? They can be obtained a few
different ways, the following is due to the Bernoulli brothers, a pair of Swiss mathematicians, about
$300$ years ago. Let $S_p:=\sum_{k=1}^n k^p$, we wish to see where the above formulas for
$S_0,S_1,S_2,S_3$ come from and see a general way to find formulas for $p=4,5,6\dots$ as well. The
formula for $S_0$ is obvious, the formula for $S_1$ is also pretty easy to get:

$$
S_1=1+2+3+\dots +n
S_1=n+(n-1)+\dots +2+1
$$

Adding these together we get $2S_1=n(n+1)$ and so $S_1=\frac{n(n+1)}{2}$. To get $S_2$ we

proceed to look at the following sum, which might appear a bit strange at first: $\
sum_{k=1}^n(k+1)^3-k^3$. On the one hand

$$
\sum_{k=1}^n(k+1)^3-k^3=\sum_{k=1}^n3k^2+3k+1=
$$
$$
3\sum_{k=1}^nk^2+3\sum_{k=1}^nk+\sum_{k=1}^n=3S_2+3S_1+S_0
$$

But we can also see this is a telescoping sum in the sense that $\sum_{k=1}^n(k+1)^3-k^3=(2^3-
1^3)+(3^3-2^3)+\dots +((n+1)^3-n^3)=(n+1)^3-1=n^3+3n^2+3n$. But then

$$
n^3+3n^2+3n=3S_2+3S_1+S_0=3S_2+3\frac{n(n+1)}{2}+n
$$

enables us to solve for $S_2$:

$$
6S_2=2(n^3+3n^2+3n)-n(n+1)-2n=2n^3+5n^2+3n=
$$
$$
n(2n^2+5n+3)=n(n+1)(2n+3)
$$

and our formula for $S_2$ above follows. Now that we have this nice formula for $S_2$ to find a nice
formula for $S_3$ we look at the sum $\sum_{k=1}^n(k+1)^4-k^4$ and evaluate it in the above ways
and solve for $S_3$. Continuing, once the formulas up to $S_p$ are found for $p\in\mathbb{N}^+$ to
find the formula for $S_{p+1}$ one looks at the sum $\sum_{k=1}^n(k+1)^{p+2}-k^{p+2}$ in the
same manner.

\vspace{.2in}

We encourage the reader to try to find some of these, it's great practice getting more comfortable with
sigma notation and sums in general. One obstacle as one proceeds to look at larger and larger powers
$p$ in the above, one that the Bernoulli's were keenly aware of, is that a bit more elbow grease seems
to be needed in expanding $(k+1)^p$ as $p$ gets larger. No one wants to see an expansion of $
(k+1)^{12}$ by repeatedly multiplying by the term $k+1$. Thankfully there is nice streamlined way to
expand such expressions that is also a product, pun intended, of recursive definitions. We first write out
some of these expansions to get some ideas:

$$
(1+k)^0=1
$$
$$
(1+k)^1=1+k
$$
$$
(1+k)^2=1+2k+k^2
$$
$$
(1+k)^3=1+3k+3k^3+k^3
$$
$$
(1+k)^4=1+4k+6k^2+4k^3+k^4
$$
$$
(1+k)^5=1+5k+10k^2+10k^3+5k^4+k^5
$$

This isn't anarchy, there are some strong patterns here where we've carefully arranged the resulting
polynomials from lowest degree to highest degree. One pattern is that the coefficients as we move from
left to right have a palindromic symmetry: they read the same from left to right as they do right to left.

\vspace{.2in}

We can also just look at the coefficients:

\vspace{.2in}

\Longstack[l]{
n=0\\
n=1\\
n=2\\
n=3\\
n=4\\
n=5\\
n=6\qquad\ \\
}
\Longstack{
1\\
1\x 1\\
1\x 2\x 1\\
1\x 3\x 3\x 1\\
1\x 4\x 6\x 4\x 1\\
1\x 5\y 10\z 10\y 5\x 1\\
1\x 6\y 15\z 20\z 15\y 6\x 1\\
\overline{0\x 1\x 2\x 3\x 4\x 5\x 6}
}

\vspace{.2in}

Wihch is known as \textbf{Pascal's Triangle} with the entries called \emph{binomial coefficients}. We
give another definition of these numbers and then show that it coincides.

\vspace{.2in}
\begin{definition}Let $n\in\mathbb{N}$ and let $0\leq k\leq n$. We define $\binom{n}{k}:=\frac{n!}
{(k!(n-k)!}\in\mathbb{Q}$ and read '$n$
choose $k$.\end{definition}

\vspace{.2in}

It's not immediately clear that this rational number is actually a positive integer, we'll prove that shortly.
But a few things
are readily apparent from this definition.

\vspace{.2in}

\begin{lemma} Let $n\in\mathbb{N}$ and let $0\leq k\leq n$. Then

$$
\binom{n}{k}=\binom{n}{n-k}
$$
and
$$
\binom{n}{0}=\binom{n}{n}=1.
$$
\end{lemma}

\vspace{.2in}

\begin{theorem}[Pascal's Identity]Let $n\in\mathbb{N}$ and let $1\leq k\leq n$. Then

$$
\binom{n+1}{k+1}=\binom{n}{k}+\binom{n}{k+1}
$$
\end{theorem}

\vspace{.2in}

\begin{proof} This is just a calculation; $(n+1)!=(n+1)n!=n!(k+n-k+1)=n!k+n!(n-k+1)$. If we divide

through by $k!$ we get

$$
\frac{(n+1)!}{k!}=\frac{n!}{(k-1)!}+\frac{n!(n-k+1)}{k!},
$$

and dividing this through by $(n-k+1)!$ gives us

$$
\frac{(n+1)!}{k!(n+1-k)!}=\frac{n!}{(k-1)!(n-(k-1))!}+\frac{n!}{k!(n-k)!},
$$
or

$$
 \binom{n+1}{k}=\binom{n}{k-1}+\binom{n}{k}.
$$
\end{proof}
\vspace{.2in}

\begin{theorem} Let $n\in\mathbb{N}$ and let $0\leq k\leq n$. Then $\binom{n}{k}\in\mathbb{N}$,
namely $k!(n-k)!|n!$ \end{theorem}

\vspace{.2in}

\textbf{Proof:}\quad We proceed by induction on $n\in\mathbb{N}$ to show that the statement $k!(n-

k)!|n!$ holds where $0\leq k\leq n$. The base case is when $n=k=0$ and this holds by the above
lemma. Now assume that it holds for some fixed but arbitrary $n\in\mathbb{N}$. If $k=0,n+1$ then
clearly $k!(n+1-k)!|(n+1)!$ so assume $1\leq k\leq n$. Then by assumption $\binom{n}{k},\binom{n}
{k-1}\in\mathbb{N}^+$ so $\exists s,t\in\mathbb{Z}$ so that $sk!(n-k)!=n!, t(k-1)!(n-k+1)!=n!$. So

$$
(n+1)!=n!(n-k+1+k)=n!(n-k+1)+n!k=
$$
$$
sk!(n-k)!(n-k+1)+t(k-1)!(n-k+1)!k=
$$
$$
(s+t)[k!(n+1-k)!],
$$

and we are done by the PMI.

\vspace{.2in}

We next show that the binomial coefficients are indeed the coefficients of $(1+k)^n$. For the sake of
simplicity we define $r^0:=1$ for any $r\in\mathbb{R}$.

\vspace{.2in}

\begin{theorem}[Binomial Theorem] Let $n\in\mathbb{N}$ and $x\in\mathbb{R}$. Then

$$
(1+x)^n=\sum_{k=0}^n\binom{n}{k}x^k.
$$
\end{theorem}

\vspace{.2in}

\begin{proof} We proceed by induction on $t\in\mathbb{N}$, the base case of the theorem being that
$1=(1+x)^0=\sum_{k=0}^0\binom{0}{k}=\binom{0}{0}$ which is true. Assume that $(1+x)^t=\
sum_{k=0}^t\binom{t}{k}x^k $ for some $t\in\mathbb{N}$. Then
$$
\sum_{k=0}^{t+1}\binom{t+1}{k}x^k=1+x^{t+1}+\sum_{k=1}^t\binom{t+1}{k}x^k
$$
$$
1+x^{t+1}+\sum_{k=1}^{t}\binom{t}{k}x^k+\sum_{k=1}^{t}\binom{t}{k-1}x^k=
$$
$$
1+\sum_{k=1}^{t}\binom{t}{k}x^k+x^{t+1}+\sum_{k=0}^{t-1}\binom{t}{k}x^{k+1}=
$$
$$
\sum_{k=0}^{t}\binom{t}{k}x^k+\sum_{k=0}^{t}\binom{t}{k}x^{k+1}=\sum_{k=0}^{t}\binom{t}
{k}x^k+x\sum_{k=0}^{t}\binom{t}{k}x^{k}=
$$
$$
(1+x)^t+x(1+x)^t=(1+x)(1+x)^{t+1}=(1+x)^{t+1},
$$

and we are done by the PMI.

\end{proof}

\vspace{.2in}

\begin{corollary} $2^n=\binom{n}{0}+\binom{n}{1}+\dots +\binom{n}{n}$ for any $n\in\

mathbb{N}.$\end{corollary}

\vspace{.2in}

\begin{proof}Let $x=1$ in the Binomial Theorem.\end{proof}

\vspace{.2in}

\begin{corollary} $\binom{n}{0}-\binom{n}{1}+\binom{n}{2}-\dots +(-1)^n\binom{n}{n}$=0 for

any $n\in\mathbb{N}.$\end{corollary}

\vspace{.2in}

\begin{proof} Let $x=-1$ in the Binomial Theorem.\end{proof}

\vspace{.2in}

\begin{corollary}Let $a,b\in\mathbb{R}$ and $n\in\mathbb{N}$. Then

$$
(a+b)^n=\sum_{k=0}^n\binom{n}{k}a^kb^{n-k}.
$$
\end{corollary}

\vspace{.2in}

\begin{proof}If $b=0$ this is clearly true, so assume $b\not=0$ and set $x=\frac{a}{b}$. Then by the
Binomial Theorem we know that
$$
\left(1+\frac{a}{b}\right)^n=\sum_{k=0}^n\binom{n}{k}\left(\frac{a}{b}\right)^k,
$$

and multiplying both sides by $b^n$ we get

$$
b^n\left(1+\frac{a}{b}\right)^n=b^n\sum_{k=0}^n\binom{n}{k}\frac{a^k}{b^k},
$$

or

$$
(b+a)^n=\sum_{k=0}^n\binom{n}{k}a^kb^{n-k}
$$

as desired.\end{proof}

\vspace{.2in}

\begin{theorem}
Let $f,g$ be $n$ times differentiable functions. Then
$$
(fg)^{(n)}=\sum_{k=0}^n\binom{n}{k}f^{(k)}g^{(n-k)}
$$

where $f^{(0)}=f$.
\end{theorem}

\vspace{.2in}

\begin{proof}
This is essentially the same proof as the Binomial Theorem. One proceeds by induction on $n\geq 1$,
the base case being straightforward. If the theorem holds for some $n\geq 1$ and $f,g$ are $n+1$ times
differentiable, then

$$
(fg)^{(n+1)}=[(fg)^{(n)}]'=\left[\sum_{k=0}^n\binom{n}{k}f^{(k)}g^{(n-k)}\right]'=\sum_{k=0}^n\
binom{n}{k}[f^{(k)}g^{(n-k)}]'=
$$
$$
\sum_{k=0}^n\binom{n}{k}f^{(k+1)}g^{(n-k)}+f^{(k)}g^{(n-k+1)}=\sum_{k=0}^n\binom{n}
{k}f^{(k+1)}g^{(n-k)}+\sum_{k=0}^n\binom{n}{k}f^{(k)}g^{(n-k+1)}.
$$

We note that
$$
\sum_{k=0}^n\binom{n}{k}f^{(k+1)}g^{(n-k)}=f^{(n+1)}g+\sum_{k=0}^{n-1}\binom{n}
{k}f^{(k+1)}g^{(n-k)}=f^{(n+1)}g+\sum_{k=1}^{n}\binom{n}{k-1}f^{(k)}g^{(n-k+1)}
$$
and
$$
\sum_{k=0}^n\binom{n}{k}f^{(k)}g^{(n-k+1)}=fg^{(n+1)}+\sum_{k=1}^{n}\binom{n}
{k}f^{(k)}g^{(n-k+1)},
$$

and so altogether we have

$$
(fg)^{(n+1)}=fg^{(n+1)}+f^{(n+1)}g+\sum_{k=1}^{n}\binom{n}{k-1}f^{(k)}g^{(n-k+1)}+\
sum_{k=1}^{n}\binom{n}{k}f^{(k)}g^{(n-k+1)}=
$$
$$
fg^{(n+1)}+f^{(n+1)}g+\sum_{k=1}^n\binom{n+1}{k}f^{(k)}g^{(n-k+1)}
$$

where we combined sums using Pascal's identity. This is in turn equal to

$$
\sum_{k=0}^{n+1}\binom{n+1}{k}f^{(k)}g^{(n-k+1)}.
$$

which establishes the $n+1$ case, and so we are done by the PMI.
\end{proof}

\vspace{.2in}

\begin{example}
If $n=4$ and $f,g$ are $4$ times differentiable, we have

$$
(fg)^{(4)}=fg^{(4)}+4f'g^{(3)}+6f''g''+4f^{(3)}g'+f^{(4)}g.
$$
\end{example}

\vspace{.2in}

The Binomial Theorem can be used to give a different proof of Fermat's Little Theorem and we ask the
reader to supply a proof of the following:

\vspace{.2in}

\begin{theorem} Let $p$ be a prime and $0\leq k\leq p$. Then $\binom{p}{k}\equiv 0\pmod{p}$ iff
$0<k<p$.\end{theorem}

\vspace{.2in}
\begin{corollary}[Fermat's Little Theorem]Let $p$ be a prime and $n\in\mathbb{Z}$. Then $n^p\equiv
n\pmod{p}$.\end{corollary}

\vspace{.2in}

\begin{proof} We first use induction on $n$ to prove that $n^p\equiv n\pmod{p}$ where $n\in\
mathbb{N}$, the base case being clear. Assume that $n^p\equiv n\pmod{p}$ for some $n\in\
mathbb{N}$. Then

$$
(n+1)^p=\sum_{k=0}^p\binom{p}{k}n^k\equiv n^p+1\equiv n+1\pmod{p},
$$

where the first equality is from the binomial theorem, the second equivalence is from our theorem on
binomial coefficients modulo $p$, and the last is our inductive hypothesis. So we are done by the PMI.\
end{proof}

\vspace{.2in}

\begin{example}[Arithmetic-Geometric Mean Inequality] Given two real numbers $a,b$ we can form
their \emph{arithmetic mean} as $\frac{a+b}{2}$. If $a,b\geq0$ we can also form their \
emph{geometric mean} as $\sqrt{ab}$. Expanding the inequality $(\sqrt{a}-\sqrt{b})^2\geq 0$ we get
$a+b-2\sqrt{ab}\geq 0$ or that $\frac{a+b}{2}\geq \sqrt{ab},$ in other words, the geometric mean
never exceeds the arithmetic mean. It's also straightforward to see that these means are equal exactly
when $a=b$. We define the arithmetic mean of $n$ real numbers $a_1, a_2, \dots, a_n$ as $\
frac{a_1+a_2+\dots +a_n}{n}$ and if each $a_i\geq 0$ we define the geometric mean to be $\sqrt[n]
{a_1a_2\dots a_n}$. \end{example}

\vspace{.2in}

\begin{theorem}[AM-GM Inequality] Suppose that $a_1, a_2, \dots, a_n$ are $n$ non-negative real
numbers. Then

$$
\frac{a_1+a_2+\dots +a_n}{n}\geq \sqrt[n]{a_1a_2\dots a_n}.
$$
\end{theorem}

We prove this by means of two lemmas:

\vspace{.2in}

\begin{lemma}Suppose $a,b\in\R$ with $a<1$ and $b>1$. Then $a+b\geq ab+1$.\end{lemma}

\vspace{.2in}

\begin{proof} We multiply the inequality $b>1$ by the positive real number $1-a$ to get $(1-a)b>(1-a)
$ or $b-ab>1-a$ which is exactly what we want.\end{proof}
\vspace{.2in}

\begin{lemma} Suppose $c_1,c_2,\dots , c_n$ are positive real numbers so

that $c_1c_2\dots c_n=1$. Then $c_1+c_2+\dots + c_n\geq n$, and equality holds precisely when
$c_1=c_2=\dots=c_n=1$.\end{lemma}

\vspace{.2in}

\begin{proof} We proceed by strong induction on $n\geq 2$, where $n$ is the number of positive real
numbers in our product. The base case $n=2$ is covered by the first lemma. Suppose that we know the
lemma holds for all natural numbers $k$ up to and including $n$ where $n\geq 2$ let $c_1,c_2,\dots ,
c_n, c_{n+1}$ be $n+1$ numbers so that $c_1c_2\dots c_{n+1}=1$. Unless each $c_i=1$ we must
have at least one $c_j>1$ and at least one $c_i<1$, and we can re-index so that $c_n<1$ and
$c_{n+1}>1$. Then we have $1=c_1c_2\dots c_{n+1}=(c_1c_2\dots c_{n-1})(c_nc_{n+1})$ and we
look at

$$
c_1c_2\dots c_{n-1}+c_nc_{n+1}\geq (n-1)+2=n+1,
$$

where we've applied our inductive hypothesis to $k=2,n-1$. But we know

$$
c_1+c_2+\dots c_n+c_{n+1}=(c_1+c_2+\dots c_{n-1})+(c_n+c_{n+1})\geq
$$
$$
c_1+c_2+\dots c_{n-1}+c_nc_{n+1}+1\geq (n+1)+1=n+2>n+1,
$$

where we used the first lemma applied to $c_n,c_{n+1}$. Thus the second lemma holds by the PMI.\
end{proof}

\vspace{.2in}

We now give a proof of the AM-GM inequality using the lemmas:

\vspace{.2in}

\begin{proof}[Proof of AM-GM inequality]Let $a_1, a_2, \dots , a_n>0$ and set $g=\sqrt[n]{a_1a_2\
dots a_n}$. Define $x_i=\frac{a_i}{g}$. Then $x_1x_2\dots x_n=1$ and so we can apply the second
lemma to get $x_1+x_2+\dots +x_n\geq n$ or

$$
\frac{a_1}{\sqrt[n]{a_1a_2\dots a_n}}+\frac{a_2}{\sqrt[n]{a_1a_2\dots a_n}}+\dots +\frac{a_n}{\
sqrt[n]{a_1a_2\dots a_n}}\geq n,
$$
which immediately gives us the AM-GM inequality.\end{proof}

\vspace{.3in}

We return to number theory with a discussion of the \emph{arithmetic derivative}, which we first
consider as the function $D:\Z\rightarrow \Z$ defined by the properties:

\begin{enumerate}
\item $D(0)=0$
\item $D(p)=1$ for any prime $p$
\item $D(ab)=aD(b)+D(a)b$ for any $a,b\in \N$
\item $D(-a)=-D(a)$ for any $a\in\N$
\end{enumerate}

Since any natural number $n\geq 2$ can be written uniquely as a product of primes we'll see that the
above conditions allow us to compute $D(n)$ fairly easily once we know how to factor $n$. We also
see that $D(1)=D(1\cdots 1)=1D(1)+D(1)1=2D(1)$ and so $D(1)=0$. Property $4$ allows us to
compute the arithmetic derivative of any integer if we
know how to compute the arithmetic derivative of any natural number. To find $D(125)=D(5^3)$ we
compute:

$$
D(25\cdot 5)=25D(5)+D(25)5=25+5D(5\cdot 5)=25+5[5D(5)+D(5)5]
$$

which gives us $D(125)=75$ or $D(5^3)=3\cdot 5^2$, which looks a lot like the standard power rule of
regular derivatives. It will be convenient for us to write $a'$ for $D(a)$ just like we do for regular
derivatives. We now show how to compute $D(a)$ when $a=p_1^{t_1}p_2^{t_2}\dots p_k^{t_k}$.

\vspace{.2in}

\begin{theorem}Let $a\geq 2$ have prime factorization $a=p_1^{t_1}p_2^{t_2}\dots p_k^{t_k}$.

Then

$$
a'=a\sum_{i=1}^k\frac{t_i}{p_i}.
$$
\end{theorem}

\vspace{.2in}

\begin{proof} We prove this by induction on the number $m$ of not necessarily distinct prime factors
of $a$, the base case is when $m=1$ so that $a=p$ for some prime $p$. Then $a'=1=a\sum_{i=1}^1\
frac{1}{p}$ and so the base case holds. Assume we know the formula works for all natural numbers
greater than one with $m$ prime factors and let $a=bp_{m+1}$ have $m+1$ prime factors. So $b$ has
$m$ prime factors, say $b=p_1p_2\dots p_m$, and we see

$$
a'=(bp_{m+1})'=bp_{m+1}'+p_{m+1}b'=b+p_{m+1}b\sum_{i=1}^m\frac{1}{p_i}=
$$
$$
\frac{a}{p_{m+1}}+a\sum_{i=1}^m\frac{1}{p_i}=a\sum_{i=1}^{m+1}\frac{1}{p_i}.
$$

So our induction is complete. It's straightforward now to see that if $a=p_1^{t_1}p_2^{t_2}\dots

p_k^{t_k}$ that $a'=a\sum_{i=1}^k\frac{t_i}{p_i}$.\end{proof}

\vspace{.2in}

\begin{corollary}We have $(a^n)'=na^{n-1}a'$ for any natural numbers $a$ and $n$.\end{corollary}

\vspace{.2in}

We leave the proof of this as an exercise for the reader. We make a note that even though we have an
analogue of the product rule for the arithmetic derivative, it is not linear, in general we do not have $
(a+b)'=a'+b'$. Take for instance $1=2'$; we can write $2=1+1$ and it does not follow that
$1=2'=1'+1'=0+0=0$. We leave it as an exercise to the reader to show that $a'=1$ iff $a$ is a prime
number. Let $c\in \Z$. We ask the interested reader to solve the 'differential equation $a'=c$. When
$c=1$ we must have $a$ being a prime number, what can you say about $a$ if $a'=2$? What if $a'=6$?
It's interesting to see how $a'$ grows as $a$ grows. We have the following bounds:

\vspace{.2in}

\begin{theorem}For every positive integer $a$ we have

$$
a'\leq \frac{a\log_2(a)}{2},
$$

and if $a$ is composite then $ a'\geq 2\sqrt{a}$. Furthermore if $a$ is a product of $k$ factors then $a'\
geq ka^{\frac{k-1}{k}}$.\end{theorem}

\vspace{.2in}

\begin{proof} If $a=p_1^{t_1}p_2^{t_2}\dots p_m^{t_m}$ then

$$
a\geq 2^{t_1}2^{t_2}\dots 2^{t_m}=2^{t_1+t_2+\dots +t_m}
$$

implies that $\log_2(a)\geq t_1+t_2+\dots +t_m$. Since $a'=a\sum_{i=1}^m\frac{t_i}{p_i}$ we have

$$
a'=a\sum_{i=1}^m\frac{t_i}{p_i}\leq \frac{a\sum_{i=1}^mt_i}{2}\leq \frac{a\log_2(a)}{2},
$$

which proves the first part. If $a=a_1a_2\dots a_k$ then we know

$$
a'=a_1'a_2\dots a_k+a_1a_2'\dots a_k+\dots +a_1a_2\dots a_k'=
$$
$$
a_2a_3\dots a_k+a_1a_3\dots a_k +\dots +a_1a_2\dots a_{k-1}=a\left( \frac{1}{a_1}+\frac{1}{a_2}+\
dots +\frac{1}{a_k}\right)\geq
$$
$$
ak\left(\frac{1}{a_1}\cdot \frac{1}{a_2}\cdots\frac{1}{a_k}\right)^{\frac{1}{k}}=aka^{-\frac{1}
{k}}=ka^{\frac{k-1}{k}},
$$

where we used the $AM-GM$ inequality to complete our proof.\end{proof}

\vspace{.2in}

One can read more about the arithmetic derivative and its deep connection to various conjectures in
number theory such as the Twin Prime Conjecture and the Goldbach Conjecture by clicking \
href{https://www.math.arizona.edu/~ura-reports/063/Sandhu.Aliana/Midterm.pdf}{HERE}

\vsp

We can extend our concept of recursion to define sequences where future terms depend on not just the
previous term but multiple previous terms. Define a sequence of natural numbers by $f_0:=0, f_1:=1$
and for $n\geq 0$ define $f_{n+2}=f_{n+1}+f_n$. So $f_2=f_1+f_0=1+0=1, f_3=f_2+f_1=1+1=2$
etc... We list the first few terms of this sequence:

$$
0,1,1,2,3,5,8,13,21,34,55,89,144,233,377,\dots
$$
$$
f_0,f_1,f_2,f_3,f_4,f_5, f_7,f_8,f_9, f_{10},f_{11},f_{12},f_{13}, f_{14}, \dots
$$

\vspace{.2in}

The sequence $(f_k)_{k=0}^{\infty}$ is known as the \emph{Fibonacci sequence} and is an example

of a doubly recursive sequence; each new term of the sequence depends on the previous two, in this
case it is their sum. A nice way to organize and work with a doubly recursive sequence is to use $2\
times 2$ matrices, so we give a brief introduction to them and some basic facts about them.

\vspace{.2in}

\begin{definition} A $2\times 2$ (real) matrix $A$ is an array $A=\begin{pmatrix}a & b\\

c & d\end{pmatrix}$ where $a,b,c,d\in\mathbb{R}$. We define the product of two matrices as follows:

$$
\begin{pmatrix}a & b\\
c & d\end{pmatrix}\begin{pmatrix}a' & b'\\
c' & d'\end{pmatrix}=\begin{pmatrix}aa'+bc' & ab'+bd'\\
ca'+dc' & cb'+dd'\end{pmatrix}
$$

\vspace{.2in}

We define the determinant of a matrix $A=\begin{pmatrix}a & b\\

c & d\end{pmatrix}$, written $det(A)$, as $ad-bc$ and leave it as exercises to the reader that matrix
multiplication is associative and that $det(AB)=det(A)det(AB)$. We denote by $I$ the matrix $\
begin{pmatrix}1 & 0\\
0 & 1\end{pmatrix}$ and note that $IA=A=AI$ for any matrix $A$.
\end{definition}

\vspace{.2in}

\begin{theorem} Let $F=\begin{pmatrix}1 & 1\\

1 & 0\end{pmatrix}=\begin{pmatrix}f_2 & f_1\\
f_1 & f_0\end{pmatrix}$. Then for any $n\in\mathbb{N}^+$ we have $F^n=\begin{pmatrix} f_{n+1}
& f_n\\
f_n & f_{n-1}\end{pmatrix}$.\end{theorem}

\vspace{.2in}

\begin{proof} We prove this by induction on $n\in\mathbb{N}^+$ the base case $n=1$ simply being
the definition of $F$. Assume that $F^k=\begin{pmatrix} f_{k+1} & f_k\\
f_k & f_{k-1}\end{pmatrix}$ for some $k\in\mathbb{N}^+$. Then

$$
F^{k+1}=(F^k)(F)=\begin{pmatrix} f_{k+1} & f_k\\
f_k & f_{k-1}\end{pmatrix}\begin{pmatrix}1 & 1\\
1 & 0\end{pmatrix}=\begin{pmatrix}f_{k+1}+f_k & f_{k+1}\\
f_k+f_{k-1} & f_k\end{pmatrix}
$$

which is equal to $\begin{pmatrix}f_{k+2} & f_{k+1}\\

f_{k+1} & f_k\end{pmatrix}$ and so we are done by the PMI.\end{proof}

\vspace{.2in}

We can find a wealth of interesting and useful identities involving the Fibonacci numbers simply by
exploiting the above theorem. For instance, if $k,n\in\mathbb{N}^+$ then the fact that
$F^{k+n}=F^kF^n$ or

$$
\begin{pmatrix}f_{k+n+1} & f_{k+n}\\
f_{k+n} & f_{k+n-1}\end{pmatrix}=\begin{pmatrix}f_{k+1} & f_{k}\\
f_{k} & f_{k-1}\end{pmatrix}\begin{pmatrix}f_{n+1} & f_{n}\\
f_{n} & f_{n-1}\end{pmatrix}=
$$
$$
\begin{pmatrix}f_{k+1}f_{n+1}+f_nf_k & f_{k+1}f_{n}+f_kf_{n-1}\\
f_{n+1}f_k+f_nf_{k-1} & f_nf_k+f_{n-1}f_{k-1}\end{pmatrix}
$$

so by equating corresponding matrix entries we get the following corollary:

\vspace{.2in}

\begin{corollary}For $k,n\in\mathbb{N}^+$ we have

$$
f_{k+n}=f_{k+1}f_n+f_kf_{n-1}.
$$

In particular if we set $n=k+1$ we get $f_{2k+1}=f_{k+1}^2+f_k^2$.\end{corollary}

\vspace{.2in}

We leave the following as an exercise to the reader (use induction and the above corollary):

\vspace{.2in}

\begin{corollary} $f_k|f_{nk}$ for any $k,n\in\mathbb{N}$.\end{corollary}

\vspace{.2in}

\begin{corollary} $\gcd(f_k,f_m)=f_g$ where $g=\gcd(k,m)$ where $k,m\in\N$. \end{corollary}

The Fibonacci sequence, although recursive, does have a non recursive description. Let $r=\frac{1+\
sqrt{5}}{2}$ be the positive root of $x^2=x+1$ and $s=\frac{1-\sqrt{5}}{2}$ be the negative root. We
know that $r^2=r+1, s^2=s+1$ and hence that $r=1+\frac{1}{r}$ and $s=1+\frac{1}{s}$.

\vspace{.2in}

We check that $det(F)=-1$ and so $det(F^n)=(det(F))^n=(-1)^n$ and this immediately gives us another
well known identity:

\vspace{.2in}

\begin{corollary}[Cassini's Identity] Let $n\in\mathbb{N}^+$. Then $f_{n+1}f_{n-1}-f_n^2=(-1)^n$.\

end{corollary}

\vspace{.2in}
\begin{theorem}[Binet's Formula] For $n\in\mathbb{N}^+$ we have

$$
f_n=\frac{1}{\sqrt{5}}(r^n-s^n).
$$\end{theorem}

\vspace{.2in}

\begin{proof} We prove $\sqrt{5}f_n=r^n-s^n$ by induction on $n\in\mathbb{N}^+$, the base case

being that $\sqrt{5}=\sqrt{5}f_1=r^1-s^1$ which is true. Assume that $\sqrt{5}f_k=r^k-s^k$ holds for
all $1\leq k\leq n$ where $n\in\mathbb{N}^+$. Then

$$
\sqrt{5}f_{n+1}=\sqrt{5}f_n+\sqrt{5}f_{n-1}=(r^n-s^n)+(r^{n-1}-s^{n-1})
$$

where the first equality is from the definition of the Fibonacci sequence and the second is our inductive
hypothesis applied when $k=n,n-1$. This in turn equals

$$
(r^n+r^{n-1})-(s^n+s^{n-1})=r^{n-1}(r+1)-s^{n-1}(s+1)=r^{n-1}r^2-s^{n-1}s^2
$$

or $r^{n+1}-s^{n+1}$ and we are done by the PMI. \end{proof}

\vspace{.2in}

Note that $0<|s|<1$ and so $s^n\rightarrow 0$ as $n\rightarrow\infty$ and so we can approximate

$f_n$ by $\frac{r^n}{\sqrt{5}}=\frac{(1+\sqrt{5})^n}{2^n\sqrt{5}}$ when $n$ is large. The number
$r$ is very famous in its own right and is often denoted by $\phi=\frac{1+\sqrt{5}}{2}$ and is called
the \textbf{Golden ratio}, an important mathematical and physical constant.

\vspace{.2in}

\begin{theorem} For each $n\in\mathbb{N}$ wive $n\geq 3$ we have $\frac{1.5^n}{\sqrt{5}}<f_n<\

frac{2^n}{\sqrt{5}}$.\end{theorem}

\begin{proof} We first prove $(1.5)^n<\sqrt{5}f_n$ by induction on $n\in\mathbb{N}$ where $n\geq

1$ the base being that $(1.5)^3<\sqrt{5}f_3=2\sqrt{5}$ which is true. Assume that $(1.5)^k<\
sqrt{5}f_k$ for all $1\leq k\leq n$. Then

$$
\sqrt{5}f_{n+1}=\sqrt{5}f_n+\sqrt{5}f_{n-1}>\sqrt{5}((1.5)^n+(1.5)^{n-1})=\sqrt{5}(1.5)^{n-1}
(2.5)>
$$
$$
\sqrt{5}(1.5)^{n-1}(1.5)^2=\sqrt{5}(1.5)^{n+1}>(1.5)^{n+1}.
$$
So by the PMI the first inequality holds for all $n\in\mathbb{N}$ with $n\geq 3$.

Now we prove that $\sqrt{5}f_n<2^n$ for $n\geq 3$ by induction on $n$ where the the base case is $\
sqrt{5}f_3<2^3=8$ which is true. Assume that $\sqrt{5}f_k<2^k$ for all $1\leq k\leq n$. Then $\
sqrt{5}f_{n+1}=\sqrt{5}f_n+\sqrt{5}f_{n-1}<2^n+2^{n-1}<2^n+2^n=2^{n+1}$ and so by the PMI
the second inequality holds for all $n\in\mathbb{N}$ with $n\geq 3$.\end{proof}

\vspace{.2in}

Let's
revisit Pascal's triangle and see if we can find the Fibonacci numbers. We arrange the binomial
coefficients $\binom{n}{k}$ so that we start with a vertical column with $\binom{n}{0}$ for $n\in\
mathbb{N}$ and add binomial coefficients with a slightly different spacing from before as follows:

\vspace{.2in}

\begin{tikzpicture}
\foreach \n in {0,...,6} {
\foreach \k in {0,...,\n} {
\node (\n\k) at (\k,-\n) {${\n \choose \k}$};
}
\pgfmathtruncatemacro{\x}{(\n+1)/2}
\pgfmathtruncatemacro{\y}{\n/2}
\draw (\n0.south west) -- (\x\y.north east);
}
\end{tikzpicture}

\vspace{.2in}

By drawing line segments with slope one as indicated and adding up any binomial coefficients that get
touched we get the sequence:

$$
\binom{0}{0}, \binom{1}{0}, \binom{1}{1}+\binom{2}{0},\binom{2}{1}+\binom{3}{0},\binom{2}
{2}+\binom{3}{1}+\binom{4}{0},
$$
$$
\binom{3}{2}+\binom{4}{1}+\binom{5}{0}, \binom{3}{3}+\binom{4}{2}+\binom{5}{1}+\
binom{6}{0}, \dots
$$

or

$$
1, 1, 2,3,5,8, 13 \dots
$$

which is very clearly the Fibonacci sequence. This can be expressed as

$$
\sum_{k=0}^{\lfloor n/2 \rfloor}\binom{n-k}{k}=f_{n+1},
$$

where $ \lfloor n/2 \rfloor$ is the largest integer less than or equal to $n/2.$

\vspace{.2in}

There are even more interesting relationships between the Fibonnaci numbers, the binomial
coefficients, and even the golden ratio in the form of Binet's formula. Set $f_{-1}:=1$ and recall $r,s$
the positive and negative roots of $x^2=x+1$. We recall that $r=1+\frac{1}{r}$, $-\frac{1}{r}=1-r,
s=1+\frac{1}{s}, -\frac{1}{s}=1-s.$

\vspace{.2in}

\begin{theorem} For any $n\in\mathbb{N}$,

$$
\sum_{k=0}^n(-1)^k\binom{n}{k}f_{k-1}=f_{n+1},
$$
\end{theorem}

\vspace{.2in}

\begin{proof} Oddly this doesn't need a proof by induction; the results we're curated so far using
induction will suffice here. We first note that Binet's formula can be extended to $n=-1$ by checking
that $1=f_{-1}=\frac{1}{\sqrt{5}}(\frac{1}{r}-\frac{1}{s})=\frac{\sqrt{5}}{\sqrt{5}}$. Then

$$
\sum_{k=0}^n(-1)^k\binom{n}{k}f_{k-1}=-\frac{1}{\sqrt{5}}\sum_{k=0}^n\binom{n}{k}((-r)^{k-
1}-(-s)^{k-1})=
$$
$$
\frac{r^{-1}}{\sqrt{5}}\sum_{k=0}^n\binom{n}{k}(-r)^k-\frac{s^{-1}}{\sqrt{5}}\sum_{k=0}^n\
binom{n}{k}(-s)^k
$$

By the Binomial Theorem we know that $\sum_{k=0}^n\binom{n}{k}(-r)^k=(1-r)^n$ and $\

sum_{k=0}^n\binom{n}{k}(-s)^k=(1-s)^n$ and so our sum above is just

$$
\frac{r^{-1}}{\sqrt{5}}(1-r)^n-\frac{s^{-1}}{\sqrt{5}}(1-s)^n
$$

where we know $(1-r)^n=(-\frac{1}{r})^n$ and $(1-s)^n=(-\frac{1}{s})^n$ and so the above becomes

$$
\frac{r^{-1}}{\sqrt{5}}\left(-\frac{1}{r}\right)^n-\frac{s^{-1}}{\sqrt{5}}\left(-\frac{1}{s}\right)^n=-\
frac{1}{\sqrt{5}}(-r)^{-n-1}+\frac{1}{\sqrt{5}}(-s)^{-n-1}.
$$

But $(-s)^{-1}=r$ and $(-r)^{-1}=s$ since $rs=-1$ and so this becomes

$$
\frac{1}{\sqrt{5}}(r^{n+1}-s^{n+1})=f_{n+1}
$$

by Binet's formula once again.\end{proof}

\vspace{.2in}

This formula might seem obtuse so let's look at what's going on for $n=7$ for instance. We have $(-
1)^k,\binom{n}{k},$ and $f_{k-1}$ that we are multiplying together and then adding up:

\vspace{.1in}

$$
\begin{array}{c|cccccccc}
(-1)^k: & 1 & -1 & 1 & -1 & 1 & -1 & 1 & -1 \\
\binom{n}{k}: & 1 & 7 & 21 & 35 & 35 & 21 & 7 & 1\\
f_{k-1}: & 1 & 0 & 1 & 1 & 2 & 3 & 5 & 8\\
(-1)^k\binom{n}{k}f_{k-1}: & 1 & 0 & 21 & -35 & 70 & -63 & 35 & -8\end{array}
$$

and then we add up the bottom row to get $21$ which is indeed $f_8$. Patterns beget patterns.

\vspace{.2in}

\begin{example}
Define a sequence $t_n$ recursively by $t_0=0,t_1=1,t_2=1$ and $t_{n+3}=t_{n+2}+t_{n+1}+t_n$
for $n\geq 0$. This is known as the \emph{tribonacci sequence} and its first few terms are

$$
0,1,1,2,4,7,13,24,44, 81, 149, 274, 504, 927\dots
$$

If one looks at the related sequence $q_n:=\frac{t_{n+1}}{t_n}$ for $n\geq 1$ we see that the first few
terms are

$$
q_3=2, q_4=1.75, q_5=1.857.., q_6=1.846.., q_7=1.83\overline{3}, q_8=1.841.., q_9=1.839..
$$

Just as the case with the Fibonacci sequence, this sequence of quotients converges in very much the
same manner. The number $T$ it converges to is approximately

$$
T\sim1.8392867552141611325518525646532866004241787460975...
$$

or

$$
T=\frac{1+\sqrt[3]{19+3\sqrt{33}}+\sqrt[3]{19-3\sqrt{33}}}{3}
$$

\end{example}

$T$ is one of the solutions to the cubic equation $x^3=x^2+x+1$.

\newpage

\section{Rational and Real Numbers}

\vspace{.2in}

There are many similarities between the real numbers $\mathbb{R}$ and its subset of rational numbers
$\mathbb{Q}$, the most important similarity is that they are both \textbf{ordered fields}.

\vspace{.2in}

\begin{definition}A set $\mathbb{F}$ along with binary operations $+$ and $\cdot$ and partial order
$\leq$ is said to be an ordered field if it satisfies the following properties:

\begin{enumerate}
\item $\forall a,b,c\in\mathbb{F}$ we have $(a+b)+c=a+(b+c)$ and $(a\cdot b)\cdot c=a\cdot (b\
cdot c)$
\item $\forall a,b\in\mathbb{F}$ we have $a+b=b+a$ and $a\cdot b=b\cdot a$
\item There exists elements $0\not=1$ of $\mathbb{F}$ so the $\forall a\in\mathbb{F}$ we have
$0+a=a$ and $1\cdot a=a$
\item $\forall a\in\mathbb{F}\, \exists b\in\mathbb{F}$ so that $a+b=0$
\item $\forall a\in\mathbb{F}\, (a\not=0)$ implies $(\exists b\in\mathbb{F})$ so that $ab=1$
\item $\forall a,b,c\in\mathbb{F}$ we have $a\cdot(b+c)=a\cdot b+a\cdot c$
\item $\forall a,b\in\mathbb{F}$ either $a\leq b$ or $b\leq a$
\item $\forall a,b,c\in\mathbb{F}$ if $a\leq b$ then $a+c\leq b+c$
\item $\forall a,b\in\mathbb{F}$ if $0<a$ and $0<b$ then $0<a\cdot b$

\end{enumerate}
\end{definition}

\vspace{.2in}
[We recall here that $+$ and $\cdot$ being binary operations means they are both functions from $\
mathbb{F}\times\mathbb{F}$ to $\mathbb{F}$, in other words, $\forall a,b\in\mathbb{F}$ we have
$a+b,a\cdot b\in\mathbb{F}$; we also recall that $\leq$ being a partial order means it is reflexive,
transitive, and antisymmetric, and $a<b$ means $a\leq b$ and $a\not=b$]

\vspace{.2in}

Properties $1$ through $6$ describe a structure that is a field, the addition of properties $7$ through
$9$ make $\mathbb{F}$ an ordered field. We will write $xy$ for $x\cdot y$. We often write $-a$ for
the element $b$ in property $4$; we often write $a^{-1}$ or $\frac{1}{a}$ for the element $b$ in
property $5$; we leave it to the reader to show that these elements are uniquely defined. We also will
often write $\frac{a}{b}:=ab^{-1}$ when we have a non-zero element $b\in\mathbb{F}$. One
immediate consequence of property $9$ is that $x^2>0$ for any non-zero element $x$ of an ordered
field. We also have $0<1$ or else if $1<0$ we would have $0<-1$ and hence $0<(-1)^2=1$. This rules
out certain fields from being ordered fields, in particular the field of complex numbers $\mathbb{C}$
of numbers of the form $a+bi$ where $i^2=-1$, since if $\leq$ was such an order for $\mathbb{C}$ we
could not have $0<i$ or else $0<i^2=-1$ and so $1<0$; if $i<0$ then $0<-i$ and so $0<(-i)^2=-1$.

\vspace{.2in}

$\mathbb{Q}\subseteq\mathbb{R}$ are both ordered fields with the usual operations of $+$ and $\
cdot$ and the usual left to right order. There are plenty of distinct proper subsets of $\mathbb{R}$ that
are also fields, for instance $\mathbb{Q}[\sqrt[n]{r}]:=\{a+b\sqrt[n]{r}\,|\,a,b\in\mathbb{Q}\}$ where
$r$ is a positive integer. These are distinct from $\mathbb{Q}$ if $n\geq 2$ and $r\not=s^2$ for any
element $s\in\mathbb{N}$. It is a fact not proven here that if $\mathbb{F}\subseteq\mathbb{R}$ is an
ordered field with the same operations and order as those in $\mathbb{R}$ that $\mathbb{Q}\subseteq\
mathbb{F}$, and so in a very strong sense $\mathbb{Q}$ is the smallest ordered field inside of $\
mathbb{R}$. If we let $m,n\in\mathbb{Z}$ with $n\not=0$ we have $\frac{m}{n}\in\mathbb{Q}$ by
definition. Let $g=\gcd(m,n)$, and set $m'=m/g, n'=n/g$ so that $\frac{m}{n}=\frac{m'g}{n'g}=\
frac{m'}{n'}$. We see that $\gcd(m',n')=1$ (why?) and say that the expression $\frac{m'}{n'}$ is in \
emph{lowest terms}.

\vspace{.2in}

\begin{theorem}
Suppose $\gcd(a,b)=1$ and that $\frac{a}{b}=\frac{c}{d}$. Then $a|c$ and $b|d$.
\end{theorem}

\vspace{.2in}

\begin{proof}
If $\frac{a}{b}=\frac{c}{d}$, we have $ad=bc$, and so any prime power $p^t$ that shows up in the
prime factorization of $a$ must show up in the prime factorization of $bc$. The condition $\
gcd(a,b)=1$ tells us that $p^t$ must divide $c$ and so every prime power dividing $a$ divides $c$. So
$a|c$. Then $c=as$ for some $s\in\Z$ and so $ad=b(as)$ or $d=bs$ and so $b|d$ as well.
\end{proof}

\vspace{.2in}
\begin{definition} The set $\mathbb{R}-\mathbb{Q}$ is called the set of irrational numbers.\
end{definition}

\vspace{.2in}

\begin{theorem}\quad $\sqrt{2}\in\mathbb{R}-\mathbb{Q}$; in other words $\sqrt{2}$ is irrational.\

end{theorem}

\vspace{.2in}

\begin{proof}[Proof 1] Assume to the contrary, that $\sqrt{2}\in\mathbb{Q}$. Then we can write $\

sqrt{2}=\frac{m}{n}$ in lowest terms, i.e. $\gcd(m,n)=1$. So $2=\frac{m^2}{n^2}$ and so
$2n^2=m^2$. Since $2|m^2$ we know by Euclid's Lemma that $2|m$ so $\exists k\in\mathbb{Z}$ so
that $m=2k.$ But then $2n^2=(2k)^2$ or $n^2=2k^2$ and similarly we get that $2|n$ contradicting the
assumption $\gcd(m,n)=1$. So $\sqrt{2}$ is irrational.\end{proof}

\vspace{.2in}

\begin{proof}[Proof 2] Assume to the contrary, that $\sqrt{2}\in\Q$. Then we can write $\sqrt{2}=\
frac{a}{b}$ in lowest terms, i.e. $\gcd(a,b)=1$. So $2=\frac{a^2}{b^2}$ and we have $\frac{a}{b}=\
frac{2b}{a}$. We can apply Theorem $6.2$ to get that $b|a$ and so $\sqrt{2}=\frac{a}{b}\in\N$. This
is a contradiction as $1<\sqrt{2}<2$ is clearly not an integer.

\end{proof}

\vspace{.2in}

\begin{corollary}Let $p$ be a prime. Then $\sqrt[n]{p}$ is irrational for any natural number $n\geq
2$.\end{corollary}

\vspace{.2in}

\begin{proof}
This is more or less the same proof that $\sqrt{2}$ is irrational: assume to the contrary and write $\
sqrt[n]{p}=\frac{a}{b}$ in lowest terms, use Euclid's Lemma to get that $p|a,b$. We leave these details
to the reader.\end{proof}

\vspace{.2in}

\begin{theorem}Let $d\in\mathbb{N}$ so that $\not\exists k\in\mathbb{N}$ so that $d=k^2$. Then $\

sqrt{d}$ is irrational.\end{theorem}

\vspace{.2in}

\begin{proof} There exists $n\in\mathbb{N}$ so that $n^2<d<(n+1)^2$ and so $0<\sqrt{d}-n<1$.

Suppose $\sqrt{d}=\frac{a}{b}\in\mathbb{Q}$ where $a,b>0$ and choose $b$ to be the smallest such
denominator so that $\exists a\in\mathbb{N}$ with $\sqrt{d}=\frac{a}{b}$ (hence $b$ is the smallest
natural number so that $ b\sqrt{d}=a\in\mathbb{N}$). Then $(\sqrt{d}-n)(b\sqrt{d})=bd-nb\
sqrt{d}=bd-nba\in\mathbb{N}$. But since $0<\sqrt{d}-n<1$ we have $(\sqrt{d}-n)b<b$ and so $(\
sqrt{d}-n)b$ is a natural number smaller than $b$ so that $(\sqrt{d}-n)b\sqrt{d}$ is a natural number, a
contradiction. So $\sqrt{d}$ is irrational.\end{proof}

\vspace{.2in}

What distinguishes the real numbers from other ordered fields is that the real numbers are \
emph{complete}; we need a few additional definitions to make this concept precise. We write $t\leq A$
where $t\in\mathbb{R}$ and $A\subseteq\mathbb{R}$ to mean $\forall a\in A, l\leq a$. Analogously
we define $B\leq s$ where $s\in\mathbb{R}$ and $B\subseteq\mathbb{R}$ to mean $\forall b\in B,b\
leq s$. We note that given any $r\in\mathbb{R}$ we have both $r\leq \emptyset$ and $\emptyset \leq r$
since these definitions are satisfied vacuously, so we will in general not include the emptyset in
definitions like the following:

\vspace{.2in}

\begin{definition} We say $\emptyset\not=A\subseteq \mathbb{R}$ is bounded below if $\exists l\in\

mathbb{R}$ so that $l\leq A$; likewise we say that $\emptyset\not=B\subseteq\mathbb{R}$ is
bounded above if $\exists u\in\mathbb{R}$ so that $B\leq u$. We say that $\emptyset\not=A\subseteq\
mathbb{R}$ is bounded if it is bounded below and bounded above. We say $u:=\sup(A)$ is the \
emph{least upper bound} or \emph{supremum} for $\emptyset\not=A$ if $A\leq u$ and if $A\leq u'\
leq u$ implies $u=u'$. We say that $l:=\inf(B)$ is the \emph{greatest lower bound} or \
emph{infimimum} for $\emptyset\not=B$ if $l\leq B$ and if $l\leq l'\leq B$ implies $l=l'$. We leave it
as a simple exercise to the reader to check that if a set has a least upper bound or a greatest lower
bound, these are unique and well-defined.\end{definition}

\vspace{.2in}

\begin{example}$-10\leq[-1,7)$ and $[-1,7)<23$ but $\inf([-1,7))=-1$ and $\sup([-1,7))=7$. If $A=\{\

frac{1}{n}\,|\, n\in\mathbb{N}^+\}=\{1,\frac{1}{2},\frac{1}{3},\dots\}$ then $A\leq 10$ but $\
sup(A)=1$. We have $-7\leq A$ of course but $\inf(A)=0$. Note that in these cases $\inf([-1,7))\in[-
1,7)$ while $\sup([-1,7))\notin [-1,7)$ while $\inf(A)\notin A$ but $\sup(A)\in A$.\end{example}

\vspace{.2in}

\textbf{Completeness Property of $\mathbb{R}$:}\quad Any non-empty subset $A$ that is bounded

above has a least upper bound.

\vspace{.2in}

Note that this is not a property that $\mathbb{Q}$ enjoys: if $A=\{r\in\mathbb{Q}\,|\, r<\sqrt{2}\}$,
clearly $A$ is non-empty and $A\leq 2$. But given an upper bound $u\in\mathbb{Q}$ for $A$ we
have that $u'=\frac{2u+2}{u+2}$ is a rational number that so that $u'<u$ since $u'<u$ is equivalent to
saying $2u+2<u^2+2u$ or $A\leq \sqrt{2}<u$, and $A\leq u'$ since $\sqrt{2}<u'$ is equivalent to
$2(u+2)^2<(2u+2)^2$ or $u^2+4u+4<2u^2+4u+2$ or $\sqrt{2}<u$. So there is no smallest rational
upper bound for $A$. There is a smallest real upper bound for $A$ namely $\sqrt{2}$.
\vspace{.2in}

\begin{theorem} Any non-empty subset $B\subseteq\mathbb{R}$ that is bounded below has a greatest
lower bound.\end{theorem}

\vspace{.2in}

\begin{proof}[Proof 1] Let $L=\{l\in\mathbb{R}\,|\, l\leq B\}$. Then $L$ is a non-empty subset of $\

mathbb{R}$ that is bounded above by every element of $B$. So $t=\sup(L)$ exists by the
Completeness Property and we claim that $t=\inf(B)$. Clearly $t\leq B$ and so let $t\leq t'\leq B$.
Then $t'\in L$ and so $t'\leq t$ by the definition of $t=\sup(L)$ and so $t=t'$ and indeed $t=\inf{B}$
exists.\end{proof}

\vspace{.2in}

\begin{proof}[Proof 2] Let $-B=\{-b\,|\, b\in\mathbb{B}\}$. $B$ is bounded below by say $x$ and so
$-B\leq -x$ is bounded above and also non-empty, and so by the Completeness Property $s=\sup(-B)$
exists. We claim that $t:=-s=\inf(B)$. We know $t\leq B$ since this is equivalent to $-B\leq s$ which if
true. Suppose $t\leq t'\leq B$. Then $-B\leq -t'\leq -t=s$ and since $s=\sup(-B)$ we have $-t'=s=-t$ or
$t=t'$ and we are done.\end{proof}

\vspace{.2in}

Another fairly immediate consequence of the Completeness Property is the following:

\vspace{.2in}

\begin{theorem}[Archimedean Property of $\mathbb{R}$] Given any $r\in\mathbb{R}$ there exists

$n\in\mathbb{N}$ so that $r<n$.\end{theorem}

\vspace{.2in}

\begin{proof} Suppose this is false, that $\mathbb{N}<r$ for some $r\in\mathbb{R}$. Then by the
Completeness Property $s=\sup(\mathbb{N})$ exists. But then $s-1$ is not an upper bound for $\
mathbb{N}$ or else this would contradict the definition of $s$, so $\exists n\in\mathbb{N}$ so that $s-
1<n$. But then $s<n+1$ contradicts $s$ being an upper bound for $\mathbb{N}$ since $n+1\in\
mathbb{N}$.\end{proof}

\vspace{.2in}

A useful consequence of the Archimedean Property is the following:

\vspace{.2in}

\begin{corollary} Given and real number $0<\epsilon$, there is an $N\in\mathbb{N}^+$ so that $0<\
frac{1}{N}<\epsilon$.\end{corollary}

\vspace{.2in}
\begin{proof} By the Archimedian property $\exists N\in\mathbb{N}^+$ so that $\frac{1}{\
epsilon}<N$. Then $0<\frac{1}{N}<\epsilon$.\end{proof}

\vspace{.2in}

We use the last result to show that between any two real numbers there is a rational number:

\vspace{.2in}

\begin{theorem} Let $a,b\in\mathbb{R}$ with $a<b$. Then $\exists q\in\mathbb{Q}$ so that

$a<q<r$.\end{theorem}

\vspace{.2in}

\begin{proof} If $a<0<b$ we can set $q=0$. First assume $0\leq a<b$. If $a=0$ we can use the above
corollary to find $N\in\mathbb{N}$ so that $0<\frac{1}{N}<b$, so let's assume $0<a<b$. By the
corollary above $\exists t\in\mathbb{N}$ so that $0<\frac{1}{t}<b-a$. By the Archimedean property
there exists $s\in\mathbb{N}$ so that $at<s$; choose $s$ to be the smallest such natural number with
this property so that $s-1\leq at<s$. Then $\frac{s}{t}-\frac{1}{t}\leq a<\frac{s}{t}$ and so $\frac{s}
{t}\leq a+\frac{1}{t}<a+(b-a)=b$ gives us altogether $a<\frac{s}{t}<b$ as desired. If $a<b\leq 0$ then
we can look at $0\leq -b<-a$ and apply the previous argument.\end{proof}

\vspace{.2in}

This theorem says that the rational numbers are \emph{dense} in the real numbers, no matter how
small the distance between two real numbers is we can always find a rational number hiding in
between. Likewise, between any two real numbers we can find an irrational number as well:

\vspace{.2in}

\begin{theorem}Let $a,b\in\mathbb{R}$ with $a<b$. Then $\exists t\in\mathbb{R}-\mathbb{Q}$ so

that $a<t<b$.\end{theorem}

\vspace{.2in}

\begin{proof} Since the rational numbers are dense we can find a rational number $p$ so that $a<p<b$
and likewise be can find another rational number $q$ so that $p<q<b$. Let $t=\frac{q-p}{\sqrt{2}}
+p$. We claim that $t\notin\mathbb{Q}$; if $t\in\mathbb{Q}$ then so is $t-p=\frac{q-p}{\sqrt{2}}$
and $\sqrt{2}=\frac{t-p}{q-p}$ contradicting the fact that $\sqrt{2}$ is irrational. We know $p<t$ and
since $\sqrt{2}>1$ we know that $\frac{q-p}{\sqrt{2}}<q-p$ so that $t<(q-p)+p=q$.\end{proof}

\vspace{.2in}

A useful tool and concept when discussing properties of the rational vs real numbers is the concept of a
limit of a sequence. A real sequence $(a_k)_{k=0}^{\infty}$ is simply a function $a\,:\,\mathbb{N}\
rightarrow\mathbb{R}$ where we set $a(k)=a_k$. We want to make precise what it means to say $a_n\
rightarrow L$ as $n\rightarrow\infty$. It's very likely you're never seen a precise formulation of this
concept, more likely something vaguely intuitive yet very imprecise. The following definition was
settled on in the early 19th century several hundred year after calculus had been discovered. We give
the definition first when $L=0$, for sequences that converge to $0$, introduce examples and theorems
for these, and then expand our focus to sequences that converse to an arbitrary real number $L$.

\vspace{.2in}

\begin{definition} Let $(a_k)_{k=0}^{\infty}$ be a real sequence.

We say $\lim_{n\rightarrow\infty}a_n=0$ or write $a_n\rightarrow 0$ as $n\rightarrow\infty$ if for all
$\epsilon>0$ there is a natural number $N$ so that

$$
\forall n\geq N \implies |a_n|<\epsilon
$$\end{definition}

\vspace{.2in}

\begin{definition} Let $(a_k)_{k=0}^{\infty}$ be a real sequence. We say $\lim_{n\rightarrow\

infty}a_n=L$ or write $a_n\rightarrow L$ as $n\rightarrow\infty$ if:

$$
(\forall \epsilon>0)(\exists N\in\mathbb{N})( \forall n\geq N)\,, |a_n-L|<\epsilon.
$$
\end{definition}

\vspace{.2in}

These might look a bit intimidating on first glance, especially the second one, so we introduce some
terminology and facts involving absolute value and inequalities so we can get a better understanding
and focus first on sequences that converge to zero. We also note here that our sequences do not need to
start with $n=0$, we can begin indexing our sequences at any integer, we merely do so here for
convenience.

\vspace{.2in}

\begin{lemma} Let $x,y\in\mathbb{R}$. Then

\begin{enumerate}
\item $-|x|\leq x\leq |x|$ and $|xy|=|x||y|$
\item $|x|^2=x^2$
\item $||x|\pm||y||\leq |x\pm y|\leq |x|+|y|$
\end{enumerate}
\end{lemma}

\vspace{.2in}

\begin{proof} Everything but part 3 is an immediate consequence of the definition of absolute value.
We note that $|x+y|^2=(x+y)^2=x^2+y^2+2xy\leq |x|^2+|y|^2+2|x||y|=(|x|+|y|)^2$ and since $|x+y|$ and
$|x|+|y|$ are both
 non-negative taking square roots gives us $|x+y|\leq|x|+|y|$. Replacing $y$ with $-y$ gives us $|x-y|\
leq|x|+|-y|=|x|+|y|$ and so $|x\pm y|\leq |x|+|y|$. Then we know $|x|=|y+(x-y)|\leq|y|+|x-y|$ or $|x|-|y|\
leq|x-y|$ and also $|y|=|x+(y-x)|\leq |x|+|y-x|$ or $|y|-|x|\leq|x-y|$. So $||x|-|y||\leq |x-y|$. Replacing $y$
by $-y$ in this we get $||x|-|y||\leq|x\pm y|$.\end{proof}

\vspace{.2in}

The third part of the lemma is often called the triangle inequality: if one sets $x=a-b$ and $y=b-c$ it
becomes $|a-c|=|(a-b)+(b-c)|\leq|a-b|+|b-c|$ and we can think of $|a-b|$
as the distance from $a$ to $b$: for example if $a=-1$ and $b=8$ then $|a-b|=|-1-8|=|-9|=9$ is certainly
the length of the interval $[-1,8]$ in the real line. This idea of thinking of $|a-b|$ as the distance from
$a$ to $b$ turns out to be pretty useful.

\vspace{.2in}

\begin{definition} Let $r>0$ be any positive real number and $a\in\mathbb{R}$. Define $D_r(a):=\{x\
in\mathbb{R}\,|\, |x-a|<r\}$ to be the $r$-neighborhood of $a$.\end{definition}

\vspace{.2in}

So $D_r(a)$ is the set of all points that are no more than distance $r$ from the point $a$: solving the
inequality $|x-a|<r$ is the same as solving $-r<x-a<r$ or $a-r<x<a+r$ and so is the open interval $(a-r,
a+r)$. So $D_r(a)=(a-r,a+r)$. Note that if $0<r_1<r_2$ we have $D_{r_1}(a)\subsetneq D_{r_2}(a)$
since the interval $(a-r_1,a+r_1)$ is properly contained in $(a-r_2,a+r_2)$. For instance $D_1(0)\
subseteq D_2(0)$ is simply saying $(-1,1)\subseteq (-2,2)$.

\vspace{.2in}

Let us revisit our definition of $a_n\rightarrow 0$, a sequence converging to zero. What we are saying
is that given any radius $\epsilon>0$ there should exist some time, say $N=N(\epsilon)$, so that after
we reach this time (so for all $n>N$) we have that $a_n\in D_{\epsilon}(0)$. It is not good enough for
us to merely arrive in this $\epsilon$-neighborhood of zero, we have to stay within this $\epsilon$-
neighborhood. A sequence might have some wild behaviour before it settles down and approaches zero.

\vspace{.2in}

\begin{example} Let $(a_n)_{n=0}^{\infty}$ be the sequence given by

\begin{equation*}
a_n = \left\{
\begin{array}{ll}
1 & \quad n \quad \text{even} \\
\frac{1}{n} & \quad n \quad \text{odd}
\end{array}
\right.
\end{equation*}
\vspace{.2in}

The first terms of this sequence are $1,1,1,\frac{1}{3},1,\frac{1}{5},1,\frac{1}{7},1,\frac{1}{9},1,$

$\frac{1}{11},1,\frac{1}{13},1,\frac{1}{15},1,\frac{1}{17},1,\frac{1}{19},1,\frac{1}{21},\dots$.
There are terms in this sequence that do get arbitrarily small, so given any $\epsilon$-neighborhood of
zero $D_{\epsilon}(0)$ we can certainly find $n\in\mathbb{N}$ so that $a_{2n+1}=\frac{1}{2n+1}\in
D_{\epsilon}(0)$. But unless $\epsilon\geq 1$ we would immediately have $a_{2n+2}=1\not\in D_{\
epsilon}(0)$, the sequence would bounce back out of the neighborhood. In this case it is true that the
sequence would have infinitely many terms inside of any $D_{\epsilon}(0)$, but for any significantly
small $\epsilon$ there would not be any consistency.
\end{example}

\vspace{.2in}

\begin{example} Let $(x_n)_{n=1}^{\infty}$ be the sequence $x_n=\frac{4}{n}$. Then we do have

$x_n\rightarrow 0$ since given $\epsilon>0$ we can find ${N}\in\mathbb{N}$ so that $N>\frac{4}{\
epsilon}$ by the Archimedean property. So $\frac{4}{N}<\epsilon$ and moreover, given any $n\geq
N$ we have $\frac{4}{n}\leq\frac{4}{N}$ and so $x_n=\frac{4}{n}\in D_{\epsilon}(0)$ for every $n\
geq N$. Here it was not good enough for us to merely find $N$ so that $x_N\in D_{\epsilon}(0)$ we
needed $x_N, x_{N+1},x_{N+2},x_{N+3},x_{N+4},\dots$ are in this neighborhood.\end{example}

\vspace{.2in}

\begin{example} Let $y_n=\frac{1-n}{1+n+n^2}$ where $n=1,2,3\dots$. We show that $y_n\

rightarrow 0.$ Before we give our proof, let us see what we need to accomplish. We need to find $N$
first and foremost so that $y_N\in D_{\epsilon}(0)$ given an $\epsilon>0$, namely we want $\left|\
frac{1-N}{1+N+N^2}\right|<\epsilon$. Now $\left|\frac{1-N}{1+N+N^2}\right|=\left|\frac{N-1}
{1+N+N^2}\right|<\frac{N}{N^2}$ since this last fraction has a bigger numerator and a smaller
denominator. But this is just $\frac{1}{N}$ and know we can choose $N>\frac{1}{\epsilon}$ here. Let
us put these ingredients into proof form: Let $\epsilon>0$. Choose $N\in\mathbb{N}$ so that $N>\
frac{1}{\epsilon}$. Then for all $n\geq N$ we have

$$
\left|\frac{1-n}{1+n+n^2}\right|<\frac{1}{n}\leq\frac{1}{N}<\epsilon
$$

and so $y_n\in D_{\epsilon}(0)$. Our proof is complete. Note that in our planning we focused on $N$
and $\epsilon$ first and brought $n$ later on; the reason we were able to do this is because of that very
last part of the proof, that $n\geq N$ implied $\frac{1}{n}\leq\frac{1}{N}$.
\end{example}

\vspace{.2in}

Let us look at the negation of the statement $a_n\rightarrow 0$:

$$
\neg((\forall \epsilon>0)(\exists N\in\mathbb{N})(\forall n\geq N)( a_n\in D_{\epsilon}(0))
$$
\vspace{.2in}

The above negation is logically equivalent to

$$
(\exists \epsilon>0)(\forall N\in\mathbb{N})(\exists n\geq N)(a_n\not\in D_{\epsilon}(0)).
$$

In other words a sequence \emph{doesn't} converge to zero if we can find a $D_{\epsilon}$ so that no
matter what $N$ is provided we are able to find an $n\geq N$ with $a_n\not\in D_{\epsilon}(0)$.

\vspace{.2in}

\begin{example} Let's show that that $b_n=\frac{n+1}{n+2}$ does not converge to zero (in fact it
converges to $1$): let $\epsilon=\frac{1}{4}$ and $N\in\mathbb{N}$. Then $\frac{N+1}{N+2}=1-\
frac{1}{N+2}\geq 1-\frac{1}{2}=\frac{1}{2}>\frac{1}{4}$ and so given any $n\geq N$ we would
have $b_n\not\in D_{\frac{1}{4}}(0)$ since $|b_n|\geq |b_N|>\frac{1}{4}$. \end{example}

\vspace{.2in}

\begin{lemma} For any sequence $(a_n)_{n=0}^{\infty}$ we have

$$
(a_n\rightarrow 0)\iff (|a_n|\rightarrow 0)\iff (-a_n\rightarrow 0)
$$
\end{lemma}

\vspace{.2in}

\begin{proof} This is a consequence of $|a_n|=||a_n||=|-a_n|$.\end{proof}

\vspace{.2in}

\begin{theorem} If $a_n\rightarrow 0$ then $\exists M>0$ so that $\forall n$ we have $|a_n|<M$, i.e. $
(a_n)$ is \emph{bounded}.\end{theorem}

\vspace{.2in}

\begin{proof} We know that $\exists N\in\mathbb{N}$ so that $\forall n\geq N$ we have $a_n\in
D_1(0)$. The finite set $\{|a_0|,|a_1|,\dots, |a_N|\}$ is bounded by some $B>0$. Let $M=B+1$. Then
for any $n\in\mathbb{N}$ we have $|a_n|<M$.\end{proof}

\vspace{.2in}

\begin{theorem}[Limit Laws] If $(a_n)$ and $(b_n)$ are sequences with $a_n, b_n\rightarrow 0$ and
$c\in\mathbb{R}$ then:
\begin{enumerate}
\item $a_n\pm b_n\rightarrow 0$
\item $a_nb_n\rightarrow 0$
\item $ca_n\rightarrow 0$
\end{enumerate}
\end{theorem}

\vspace{.2in}

\begin{proof} We prove that $a_n+b_n\rightarrow 0$, the case $a_n-b_n\rightarrow 0$ follows

immediately from this. Let $\epsilon>0$. Since $a_n\rightarrow 0$ there is an $N_1\in\mathbb{N}$ so
that $\forall n\geq N$ we have $a_n\in D_{\frac{1}{2}}(0)$. Since $b_n\rightarrow 0$ there is an
$N_2\in\mathbb{N}$ so that $\forall n\geq N$ we have $b_n\in D_{\frac{1}{2}}(0)$. Let
$N=N_1+N_2$. Then for any $n\geq N$ we have

$$
|a_n+b_n|\leq |a_n|+|b_n|<\frac{\epsilon}{2}+\frac{\epsilon}{2}=\epsilon
$$

and so $a_n+b_n\in D_{\epsilon}(0)$.

\vspace{.1in}

To show $a_nb_n\rightarrow 0$ we take $\epsilon>0$ and then use the fact that since $b_n\rightarrow
0$ the sequence is bounded by some $M>0$. Let $\epsilon'=\frac{\epsilon}{M}$. Then there exists
$N\in\mathbb{N}$ so that for all $n\geq N$ we have $a_n\in D_{\epsilon'}(0)$. Then we also have

$$
|a_nb_n|\leq|a_n|b_n|\leq |a_n|M<\epsilon'M=\epsilon
$$

and so $n\geq N$ implies $a_nb_n\in D_{\epsilon}(0)$.

\vspace{.1in}

The third part is left to the reader.

\end{proof}

\vspace{.2in}

We add two more useful tools to our list of theorems about sequences converging to zero, proving the
first one and leaving the second as an exercise for the reader to enjoy:

\vspace{.2in}

\begin{theorem}[Squeeze Theorem] Let $a_n\leq b_n\leq c_n$ be three sequences so that $a_n,c_n\
rightarrow 0$. Then $b_n\rightarrow 0$ as well.\end{theorem}
\vspace{.2in}

\begin{proof} Let $\epsilon>0$. Choose $N\in\mathbb{N}$ so that if $n\geq N$ we have $a_n,c_n\in

D_{\epsilon}(0)$. Then we know that if $b_n\geq 0$ that $|b_n|\leq c_n<\epsilon$ and if $b_n<0$ that
$|b_n|-b_n\leq -a_n=|a_n|<\epsilon$ so either way $b_n\in D_{\epsilon}(0)$ as well.\end{proof}

\vspace{.2in}

\begin{theorem} Suppose $a_n\rightarrow 0$ and $b_n$ is any bounded sequence, namely $\exists
M>0$ so that $\forall n\in\mathbb{N}$ we have $|b_n|<M$. Then $a_nb_n\rightarrow 0$.\
end{theorem}

\vspace{.2in}

\begin{proof} Let $\epsilon>0$. Choose $N\in\N$ so that if $n\geq N$ we have $a_n\in D_{\frac{\
epsilon}{M}}(0)$. Then if $n\geq N$ we also have $|a_nb_n|=|a_n||b_n|<|a_n|M<\frac{\epsilon}
{M}M=\epsilon$. So $a_nb_n\in D_{\epsilon}(0)$

$$
|a_nb_n|=|a_n||b_n|
$$
\end{proof}

We note that the converse to this theorem is false: let $a_n=\frac{1}{n^2}$ and $b_n=n$. Then
$a_nb_n\rightarrow 0$ but $b_n$ is not bounded.

\vspace{.2in}

\begin{example} Let $x\in\mathbb{R}$. We wish to show that if $|x|<1$ then $x^n\rightarrow 0$.
Let us brainstorm how we might prove this. We want at least to find some $N$ so that $|x|^N<\epsilon$
for a given $\epsilon$. Maybe one thinks of taking a logarithm on both sides, but have not proven
anything about the logarithm, so it will be nice to avoid this at least for now. So we get more creative
here: let $r>0$. Then $(1+r)^N=1+Nr+\sum_{k=2}^N\binom{N}{k}r^k$ by the Binomial Theorem,
and so $(1+r)^N<1+Nr$ since the terms in the sum are positive (this is called \emph{Bernouilli's
Inequality}). Let us focus on $0<x<1$. Then $\frac{1}{|x|}=1+\frac{1-|x|}{|x|}$ and so $\left(\frac{1}
{x}\right)^N=\left(1+\frac{1-x}{x}\right)^N>1+N\left(\frac{1-x}{x}\right)>N\left(\frac{1-x}{x}\
right)$ and so $|x|^N=x^N<\frac{x}{N(1-x)}$. It's this
last quantity that we want to have less than a given $\epsilon$ so we are ready for our proof in the case
$0<x<1$: let $\epsilon>0$. Choose $N\in\mathbb{N}$ so that $N>\frac{x}{\epsilon(1-x)}$. Then if
$n\geq N$ we have:

$$
|x^n|=x^n\leq x^N<\frac{x}{N(1-x)}<\epsilon
$$

and so $x^n\in D_{\epsilon}(0)$ as needed.

We leave the proof that this holds for $-1<x\leq 0$ to the reader.\end{example}
\vspace{.2in}

We turn to looking at sequences converging to arbitrary limits. Our definition of $a_n\rightarrow L$ is

that for any $\epsilon>0$ there is an $N=N(\epsilon)\in\mathbb{N}$ so that if $n\geq N$ we have
$a_n\in D_{\epsilon}(L)$, that given any neighborhood of $L$ however small, our sequence after
some point stays in this neighborhood. Another way of saying that $a_n\rightarrow L$ is to say that the
sequence $b_n:=a_n-L\rightarrow 0$ and so we can translate an awful lot of statements about limits
converging to zero to statements about limits converging to arbitrary real numbers and vice a versa.

\vspace{.2in}

\begin{example} We show that $z_n=\frac{n^2-1}{n^2+1}\rightarrow 1$ by first doing some

preliminary work. We need that $\lvert \frac{N^2-1}{N^2+1}-1\rvert$ is smaller than a given $\
epsilon$. $\lvert \frac{N^2-1}{N^2+1}-1\rvert=\lvert \frac{N^2-1}{N^2+1}-\frac{N^2+1}{N^2+1}\
rvert=\frac{2}{N^2+1}<\frac{2}{N}$ since the last fraction has a smaller denominator. This sets up
our proof: let $\epsilon>0$ and choose $N\in\mathbb{N}$ so that $N>\frac{2}{\epsilon}$. Then if $n\
geq N$ we have

$$
\left| \frac{n^2-1}{n^2+1}-1\right|=\frac{2}{n^2+1}<\frac{2}{n}\leq \frac{2}{N}<\epsilon,
$$

and so $\frac{n^2-1}{n^2+1}\in D_{\epsilon}(1)$.\end{example}

\vspace{.2in}

\begin{example} We show that $w_n=\sqrt{n^2+n}-n\rightarrow \frac{1}{2}$ by again first doing

some preliminary work to determine how we should determine $N=N(\epsilon)$. We need that $|\
sqrt{N^2+1}-N-\frac{1}{2}|$ is smaller than a given $\epsilon>0$. Multiplying by $\frac{\
sqrt{N^2+N}+N}{\sqrt{N^2+N}+N}$ we get that $|\sqrt{N^2+1}-N-\frac{1}{2}|=\lvert \frac{2N}{2\
sqrt{N^2+N}+2N}- \frac{\sqrt{N^2+N}+N}{2\sqrt{N^2+N}+2N} \rvert=\frac{\sqrt{N^2+N}-N}{2\
sqrt{N^2+N}+2N}$ which we again multiply by $\frac{\sqrt{N^2+N}+N}{\sqrt{N^2+N}+N}$ to get
$|\sqrt{N^2+1}-N-\frac{1}{2}|=\frac{N}{2(\sqrt{N^2+N}+N)^2}<\frac{N}{2N^2}=\frac{1}{2N}$
since leaving off the term $\sqrt{N^2+N}$ makes the denominator smaller and hence the fraction
bigger. This we can work with and give our proof: let $\epsilon>0$ and choose $N>\frac{1}{2\epsilon}
$. Then for $n\geq N$ we have that

$$
\left|w_n-\frac{1}{2}\right|=\frac{n}{2(\sqrt{n^2+n}+n)^2}<\frac{1}{2n}\leq\frac{1}{2N}<\epsilon,
$$

and so $w_n\in D_{\epsilon}(\frac{1}{2})$ as needed.\end{example}

\vspace{.2in}

How do we know that a given sequence can't converge to two different real numbers? Why couldn't we
also get $w_n\rightarrow 0$ in the preceding example?

\vspace{.2in}
\begin{theorem}[Uniqueness of Limits] Suppose we have $a_n\rightarrow L$ and $a_n\rightarrow L'$.
Then $L=L'$.\end{theorem}

\vspace{.2in}

\begin{proof} Suppose that $L\not=L'$ and set $\epsilon=\frac{|L-L'|}{2}>0$. Then there is an $N\in\
mathbb{N}$ so that for every $n\geq N$ we have both $a_n\in D_{\epsilon}(L), a_n\in D_{\epsilon}
(L')$. But then $|L-L'|=|L-N+a_N-L'|\leq |L-a_N|+|a_N-L'|<\frac{|L-L'|}{2}+\frac{|L-L'|}{2}=|L-L'|$ a
contradiction. So $L=L'$.\end{proof}

\vspace{.2in}

For some additional practice setting up $(\epsilon, N)$ proofs that various sequences converge we
direct the reader to \href{https://math.mit.edu/~apm/ch03.pdf}{HERE}. We encourage the reader
to state and prove analogs of the Squeeze Theorem and the Limit Laws and to show that sequences that
converge are bounded; we add some additional Limit Law theorems here:

\vspace{.2in}

\begin{theorem}[Add'l Limit Laws] Let $a_n\rightarrow L, b_n\rightarrow M$. Then

\begin{enumerate}
\item $a_nb_n\rightarrow LM$.
\item $\frac{a_n}{b_n}\rightarrow \frac{L}{M}$ if $b_n, L\not= 0$.
\end{enumerate}
\end{theorem}

\vspace{.2in}

\begin{proof} For 1, if $L=0$ this follows from a previous theorem so assume $L\not=0$ and let $B$
be a bound for the sequence $a_n$, i.e., $|a_n|\leq B$ for all $n\in\mathbb{N}$. Let $\epsilon>0$.
Then there exists $N_a\in\mathbb{N}$ so that if $n\geq N_a$ we have $a_n\in D_{\frac{\epsilon}{2|
L|}}(M)$. There also exists $N_b\in\mathbb{N}$ so that if $n\geq N_b$ we have $b_n\in D_{\frac{\
epsilon}{2B}}(L)$. Then for $n\geq N:=N_a+N_b$ we have
$$
|a_nb_n-LM|=|a_nb_n-a_nL+a_nL-LM|\leq |a_n||b_n-L|+|L||a_n-M|
$$
$$
\leq B|b_n-L|+|L||a_n-M|<B\left(\frac{\epsilon}{2B}\right)+|L|\left(\frac{\epsilon}{2|L|}\right)=\frac{\
epsilon}{2}+\frac{\epsilon}{2}=\epsilon.
$$

\vspace{.1in}

For 2, we first prove this when $a_n=1$. Since $b_n$ is bounded and does not converge to zero we
know that there exists $0<B$ so that for all $n\in\mathbb{N}$ we have $B<|b_n|$ (why?). Let $\
epsilon>0$ and choose $N\in\mathbb{N}$ so that if $n\geq N$ we have $b_n\in D_{B|M|\epsilon}(M)
$. Then if $n\geq N$ we have
$$
\left|\frac{1}{b_n}-\frac{1}{M}\right|=\frac{|b_n-M|}{|b_n||M|}< \frac{|b_n-M|}{B|M|}< \epsilon.
$$

\vspace{.2in}

We could also have proven this using the Squeeze Theorem noting that

$$
0\leq\left|\frac{1}{b_n}-\frac{1}{M}\right|<\frac{|b_n-M|}{B|M|}
$$

and noting that $|b_n-M|\rightarrow 0$ as $n\rightarrow\infty$.

\vspace{.1in}

For 2 in general combine this with part 1.\end{proof}

\vspace{.2in}

Often the terms of a sequence grow bigger and bigger, so we define exactly what it means to say $a_n\
rightarrow \infty$:

\vspace{.2in}

\begin{definition} We say $a_n\rightarrow\infty$ if for any $M>0$ there exists $N\in\mathbb{N}$ so

that for all $n\geq N$ we have $a_n>M$. Like wise we say $b_n\rightarrow-\infty$ if $-b_n\
rightarrow\infty.$\end{definition}

\vspace{.2in}

\begin{example} Clearly the sequence $a_n=n$ has the property that $a_n\rightarrow\infty$; this is
simply a restatement of the Archimedean Property. A more interesting example is given by $b_n=\
sqrt{n+\sqrt{n+\sqrt{n+\sqrt{n}}}}$; we note that $\sqrt{n+a}>\sqrt{n}$ for any $a>0$ and so $a_n>\
sqrt{n+\sqrt{n+\sqrt{n}}}>\sqrt{n+\sqrt{n}}>\sqrt{n}$ and so given any $M>0$ we choose $N\in\
mathbb{N}$ so that $N>M^2$. Then if $n\geq N$ we know that $a_n>\sqrt{n}\geq\sqrt{N}>M$ as
needed.\end{example}

\vspace{.2in}

\begin{theorem} Suppose $a_n>0$ is a sequence with positive terms. Then $a_n\rightarrow\infty$ iff
$\frac{1}{a_n}-\rightarrow 0$.\end{theorem}

\vspace{.2in}

\begin{proof} Let $\epsilon>0$ and assume $a_n\rightarrow\infty.$ Then we can find $N\in\
mathbb{N}$ so that $n\geq N$ implies $a_n>\frac{1}{\epsilon}$. But then $|\frac{1}{a_n}|=\frac{1}
{a_n}<\epsilon$ when $n\geq N$ and so $\frac{1}{a_n}-\rightarrow 0$. Likewise if $\frac{1}{a_n}\
rightarrow 0$ and $M>0$ we can find $N\in\mathbb{N}$ so that $n\geq N$ implies $\frac{1}{a_n}<\
frac{1}{M}$ or equivalently $a_n>M$ and so $a_n\rightarrow\infty.$\end{proof}

\vspace{.2in}

An important class of sequences are those that contain terms that are either increasing or decreasing;
we call such sequences \emph{monotone} and establish the following very useful theorem:

\vspace{.2in}

\begin{theorem}[Monotone Convergence Theorem] A monotone sequence converges iff it is bounded.\

end{theorem}

\vspace{.2in}

\begin{proof} We give the proof in the case where we have an increasing sequence $x_n$, i.e. a
sequence so that $\forall n$ $x_n\leq x_{n+1}$. Assume that $x_n$ is bounded. Let $X=\{x_n\,|\, n\in\
mathbb{N}\}$ and note that $X$ is bounded by definition and so $s:=\sup(X)$ exists by the
Completeness Property. We show that $x_n\rightarrow s$. Let $\epsilon>0$. We know that $s-\epsilon$
is not an upper bound for $X$ since $s$ is the \emph{least} upper bound. So there is some $x_N$ so
that $s-\epsilon<x_N$. But then for $n\geq N$ we have

$$
|s-x_n|=s-x_n\leq s-x_N<\epsilon,
$$
since $x_n$ is increasing. So $x_n\in D_{\epsilon}(s)$ and we know that $x_n\rightarrow s$
converges. The other direction of the 'iff' is a consequence of any convergent sequence being bounded.\
end{proof}

\vspace{.2in}

\begin{example} Let $x_0=a\in(0,\frac{1}{2})$ and let $x_{n+1}=\frac{x_n}{2}(1-x_n)$. We prove

first that $\forall k\in\mathbb{N}$ that $x_{k+1}\leq x_k$ and that $x_k\geq 0$. We note first that
$x_1=\frac{a}{2}(1-a)$ and $0<a<\frac{1}{2}$ implies $\frac{1}{2}<1-a<1$ and so $\frac{1}{4}<\
frac{1-a}{2}<\frac{1}{2}$ and so $\frac{a}{2}(1-a)=x_1<\frac{a}{2}<x_0$. We note in general that
$x_{k+1}-x_k=\frac{x_k}{2}(1-x_k)-x_k=-\frac{x_k}{2}-\frac{x_k^2}{2}<0$ and so $x_n$ is
decreasing. Since $x_0>0$ if we assume that $0<x_k<\frac{1}{2}$ for some $k\in\mathbb{N}$ we
know that $\frac{1}{2}<1-x_k<1$ and so $\frac{1}{4}<\frac{1-x_k}{2}<\frac{1}{2}$ and so
multiplying through by $x_k$ we get $\frac{x_k}{4}<x_{k+1}<\frac{x_k}{2}$ and so since $0<\
frac{x_k}{4}$ we know $0<x_{k+1}$ and since $\frac{x_k}{2}<\frac{1}{4}$ we know $x_{k+1}<\
frac{1}{4}$ and so altogether $0<x_{k+1}<\frac{1}{4}<\frac{1}{2}$. So we know that $x_n$ is both
decreasing and bounded and hence has a limit $L$ by the Monotone Convergence Theorem, namely
that $x_n\rightarrow L$. Then $x_{n+1}\rightarrow L$ as well and we know that the limit must
satisfy:

$$
L=\frac{L}{2}(1-L) \quad \text{or}\quad L^2-L=0,
$$

and so the possibilities for $L$ are $L=0$ or $L=1$. $L=1$ is nonsense since our sequence is
decreasing from $x_0=a<\frac{1}{2}$ and so $L=0$.\end{example}

\vspace{.2in}

\begin{example} Let $y_0=y_1=0$ and for $n\in\mathbb{N}$ define $y_{n+2}=\frac{1}

{3}y_{n+1}+\frac{1}{6}y_n+1$. We prove by induction that $y_n<2$ for all $n\in\mathbb{N},$ the
base case following from the definition $y_0=y_1=0<2$. Assume that the bound holds for $y_0,y_1,\
dots y_k, y_{k+1}$ where $k\in\mathbb{N}$. Then $y_{k+2}=\frac{1}{3}y_{k+1}+\frac{1}
{6}y_k+1<\frac{2}{3}+\frac{2}{6}+1=2$ as well. Now
we show that $y_n$ is increasing by induction, again the base case following from the definition; we
note that $y_2=1>0=y_1$ as well. Suppose that $y_0\leq y_1\leq y_2 \leq \dots\leq y_{k+1}$ for some
$k\geq 1$. Then

$$
y_{k+2}-y_{k+1}=\frac{1}{3}(y_{k+1}-y_k)+\frac{1}{6}(y_k-y_{k-1})\geq 0,
$$

so that $y_{k+2}\geq y_{k+1}$ showing that $y_n$ is an increasing sequence. It therefore converges
by the Monotone Convergence Theorem, say $y_n\rightarrow L$. then $L=\frac{L}{3}+\frac{L}
{6}+1$ or $L=2$.\end{example}

\vspace{.2in}

\begin{example} Suppose we want to make sense of an expression like

$$
\sqrt{t+\sqrt{t+\sqrt{t+\sqrt{t+\dots}}}}.
$$

where $t\geq 1$.

We could define a sequence recursively by $a_1=\sqrt{t}$ and for $k\geq 0$ set $a_{k+1}=\
sqrt{t+a_k}$. So $a_2=\sqrt{t+\sqrt{t}}, a_3=\sqrt{t+\sqrt{t+\sqrt{t}}}$ etc... How do we determine if
such a sequence has a limit? First we establish that $a_n$ is an increasing sequence where $a_1=\
sqrt{t}\leq \sqrt{t+\sqrt{t}}=a_2$ follows from the fact that $t\geq 1$. If $a_k\leq a_{k+1}$ for some
$k\in\mathbb{N}^+$ then we know that $a_k+\sqrt{t}\leq a_{k+1}+\sqrt{t}$ and hence $a_{k+1}=\
sqrt{a_k+\sqrt{t}}\leq \sqrt{a_{k+1}+\sqrt{t}}=a_{k+2}$ as well. So we know that this sequence
converges exactly when it is bounded...which is all the time. We claim that for all $k\geq 1$ that
$a_k<2t$, the base case being $a_1=\sqrt{t}<2t$ which is certainly true when $t\geq 1$. Suppose that
$a_k<2t$ for some $k\geq 1$. Then $\sqrt{t}+a_k<\sqrt{t}+2t$ and so $a_{k+1}=\sqrt{a_k+\
sqrt{t}}<\sqrt{\sqrt{t}+2t}<\sqrt{4t}=2\sqrt{t}\leq 2t$. So this sequence has a limit $R$ satisfying
$R=\sqrt{t+R}$ or $R^2-R-t=0$ and so by the quadratic formula $R=\frac{1\pm\sqrt{1+4t}}{2}$.
Since every term of this sequence is positive we know that $R=\frac{1+\sqrt{1+4t}}{2}$ is the correct
limit. Note the fun case when $t=1$:

$$
 \sqrt{1+\sqrt{1+\sqrt{1+\sqrt{1+\dots}}}}=\frac{1+\sqrt{5}}{2},
$$
where this last number is the Golden Ratio.\end{example}

\vspace{.2in}

The Monotone Convergence Theorem is used extensively to establish theorems like the comparison
tests for infinite series. We recall the definition of a convergent infinite series:

\vspace{.2in}

\begin{definition} Let $(a_k)_{k=0}^{\infty}$. Define the $n$-th partial sum to be

$s_n:=\sum_{k=0}^na_k=a_0+a_1+\dots+a_n$. If the sequence $s_n\rightarrow s$ then we write $\

sum_{k=0}^{\infty}a_k=s=\lim_{n\rightarrow\infty}s_n$ and say that the \emph{series} $\sum a_k$
converges. If $\lim s_n$ does not exists we say that the series $\sum a_k$ diverges.\end{definition}

\vspace{.2in}

\begin{example} Let $a_k=r^k$ where $r\in\mathbb{R}$. We leave it as a simple proof by induction

for the reader to show that $(1-r)\sum_{k=0}^nr^k=1-r^{n+1}$ for any $n\in\mathbb{N}$. If $|r|<1$
we know that $\sum_{k=0}^nr^k=\frac{1-r^{n+1}}{1-r}\rightarrow \frac{1}{1-r}$ and so $\
sum_{k=0}^{\infty}r^k=\frac{1}{1-r}$. We can easily adapt this to get that $\sum_{k=t}^{\
infty}ar^k=ar^t\sum_{k=t}^{\infty}r^{k-t}=ar^t\sum_{k'=0}^{\infty}r^{k'}=\frac{ar^t}{1-r}$ when
$|r|<1$. We call this last series the geometric series with first term $ ar^t$ and common ratio $r$.\
end{example}

\vspace{.2in}

\begin{theorem}[Comparison Test] Suppose $0\leq a_n\leq b_n$ and that $\sum_{k=0}^{\

infty}b_k=B$ converges. Then $\sum_{k=0}^{\infty}a_k$ also converges and $\sum_{k=0}^{\
infty}a_k\leq B $.\end{theorem}

\vspace{.2in}

\begin{proof} Since $a_n,b_n\geq 0$ we have that the partial sums $s_n, t_n$ respectively are
increasing since:

$$
s_n\leq s_n+a_{n+1}=s_{n+1},\quad t_n\leq t_n+b_{n+1}=t_{n+1},
$$

and one readily sees that $s_n\leq t_n$. We know that $s_n\leq t_n\leq B$ and so we have that $s_n$ is
an increasing sequence that is bounded above by $B$ and so must converge by the Monotone
Convergence Theorem to a limit no bigger than $B$.\end{proof}

\vspace{.2in}
 \begin{theorem} Suppose that $0< a_n,b_n$ and that $\frac{b_n}{a_n}\rightarrow L$ where $L\
not=0$. Then $\sum_{k=0}^{\infty}a_k$ converges iff $\sum_{k=0}^{\infty}b_k$ converges.\
end{theorem}

\vspace{.2in}

\begin{proof} Suppose that $\sum a_n$ converges. Then since $\frac{b_n}{a_n}\rightarrow L$ there
is an $N\in\mathbb{N}$ so that $n\geq N$ implies that $|\frac{b_n}{a_n}-L|<\frac{L}{2}$ or $-\
frac{L}{2}<\frac{b_n}{a_n}-L<\frac{L}{2} $ so that $ 0<b_n<\frac{3L}{2}a_n$. Since $\sum a_n$
converges $\sum \frac{3L}{2}a_n$ converges and by the Comparison Test $\sum b_n$ converges. The
'iff' comes from the fact that $\frac{a_n}{b_n}\rightarrow \frac{1}{L}$ by the Limit Laws.\end{proof}

\vspace{.2in}

\begin{theorem}[Ratio Test] Suppose that $a_1,a_2, a_3, \dots$ is a sequence of positive numbers so
that $a_{n+1}/a_n\rightarrow L$ with $0\leq L<1$. Then $\sum_{n=1}^{\infty}a_n$ converges.\
end{theorem}

\vspace{.2in}

\begin{proof} Let us take $\epsilon>0$ so that $L+\epsilon<1$ (for instance $\epsilon=\frac{1-L}

{10}$ or something similar). Since $a_{n+1}/a_n\rightarrow L$ there exists $N\in\mathbb{N}$ so that
if $n\geq N$ we have $|\frac{a_{n+1}}{a_n}-L|<\epsilon$ and so $\frac{a_{n+1}}{a_n}<\epsilon+L$.
We have $a_n=\frac{a_n}{a_{n-1}}\frac{a_{n-1}}{a_{n-2}}\dots \frac{a_{N+1}}{{a_N}}a_N<(\
epsilon+L)^{n-N}a_N$ for any $n\geq N$ and so $\sum_{n=1}^{\infty}a_n=\sum_{n=1}^{N-1}a_n+\
sum_{n=N}^{\infty}a_n\leq \sum_{n=1}^{N-1}a_n+\sum_{n=N}^{\infty}(\epsilon+L)^{n-N}a_N$
where $ \sum_{n=N}^{\infty}(\epsilon+L)^{n-N}a_N=a_N\sum_{k=0}^{\infty}(\epsilon+L)^k$ is a
convergent geometric series. Therefore by the Comparison Test we have $\sum a_n$ converges.\
end{proof}

\vspace{.2in}

We make a note here that a series $\sum a_n$ where $\sum |a_n|$ converges is said to be \
emph{absolutely convergent}. We get the following simple corollary of the Comparison Test:

\vspace{.2in}

\begin{theorem}If a series is absolutely convergent it is convergent.\end{theorem}

\vspace{.2in}

\begin{proof} Suppose that $\sum_{n=1}^{\infty}|a_n|$ is convergent. Then since $\forall x\in\

mathbb{R}$ we know $0\leq x+|x|\leq 2|x|$ we know that $\sum_{n=1}^{\infty}a_n+|a_n|$ is
convergent by the comparing it to the convergent series $\sum_{n=1}^{\infty}2|a_n|$ via the
Comparison Test. But $\sum_{n=1}^{\infty}a_n=\sum_{n=1}^{\infty}a_n+|a_n|-\sum_{n=1}^{\infty}|
a_n|$ must also converge being the difference of convergent series (we leave it to the reader to show
that the sum and difference of convergent series are convergent).\end{proof}
 \vspace{.2in}

\begin{example} We will show that $\sum_{k=1}^{\infty}\frac{1}{k^2}$ converges. Note that $\

frac{1}{k^2}<\frac{1}{k(k-1)}=\frac{1}{k-1}-\frac{1}{k}$ and so for $n>1$ we have $\
sum_{k=2}^n\frac{1}{k^2}<\sum_{k=2}^n\frac{1}{k-1}-\frac{1}{k}=\left(1-\frac{1}{2} \right)+\
left(\frac{1}{2}-\frac{1}{3} \right)+\left(\frac{1}{3}-\frac{1}{4} \right)\dots+\left(\frac{1}{n-1}-\
frac{1}{n} \right)=1-\frac{1}{n}<1$ and so $\sum_{k=2}^n\frac{1}{k^2}$ is an increasing sequence
that is bounded above and so converges to a limit no bigger than $1$ by the Monotone Convergence
Theorem and so necessarily $\sum_{k=1}^{\infty}\frac{1}{k^2}$ does as well to a limit no bigger
than $2$. The exact value of this limit is $\frac{\pi^2}{6}$.\end{example}

\vspace{.2in}

\begin{example} We show that the series $\sum_{k=1}^{\infty}\frac{1}{k}$ does not converge (this
series is called the \textbf{harmonic series}). We note that $e^x\geq 1+x$ for all $x\in\mathbb{R}$
since $e^x$ is concave up and $1+x$ is its tangent line at $x=0$ (we are assuming quite a bit here).
Then for any $n\in\mathbb{N}^+$ we know that $e^{\frac{1}{n}}\geq 1+\frac{1}{n}=\frac{n+1}{n}
$. Assume that $s_n=\sum_{k=1}^n\frac{1}{k}\rightarrow L$. Then $e^{s_n}=(e^1)(e^{\frac{1}
{2}})\dots (e^{\frac{1}{n}})\geq (\frac{2}{1})(\frac{3}{2})(\frac{4}{3})\dots(\frac{n+1}{n})=n+1$.
Since $s_n\rightarrow L$ and the function $e^x$ is continuous we have that $e^{s_n}\rightarrow e^L$.
But $e^{s_n}$ is unbounded so cannot converge. Here is another proof: look at $s_{2^n}=1+\frac{1}
{2}+\frac{1}{3}+\frac{1}{4}+\frac{1}{5}+\frac{1}{6}+\frac{1}{7}+\frac{1}{8}+\dots +\frac{1}
{2^n}\geq 1+\frac{1}{2}+\left(\frac{1}{4}+\frac{1}{4}\right)+\left(\frac{1}{8}+\frac{1}{8}+\
frac{1}{8}+\frac{1}{8}\right)+\dots +\frac{1}{2^n})=1+\frac{1}{2}+\frac{1}{2}+\frac{1}{2}+\
dots+\frac{1}{2}=1+\frac{n}{2}$. So $s_{2^n}$ grows without bound and hence so does the
sequence $s_n$. \end{example}

\vspace{.2in}

There are many proofs that the harmonic series does not converge, one might think that the first proof
above contains a bit of cheating in the sense that it relied on properties of continuous functions as well
as the fact that $f(x)=e^x$ is indeed continuous, things we've certainly not established in these notes;
we will not discuss continuity of functions rigorously here. But we would like to be rigorous and give a
rigorous definition of the number $e$ and the function $e^x$ and furthermore show that the number
$e$ is irrational. We first note that for $n>3$ we have that $n!>n^2$ (provide a simple proof by
induction). So $\frac{1}{n!}<\frac{1}{n^2}$ in this range and so $\sum_{k=4}^{\infty}\frac{1}{k!}$
converges by the Comparison Test.

\vspace{.2in}

\begin{definition} Define $e:=\sum_{k=0}^{\infty}\frac{1}{k!}$ and for any $x\in\mathbb{R}$

define $e^x:=\sum_{k=0}^{\infty}\frac{x^k}{k!}$. We leave it as an exercise to the reader to show
that $\sum_{k=0}^{\infty}\frac{x^k}{k!} $ converges for any $x\in\mathbb{R}$ by using the Ratio
Test. By earlier comments it's easy to see that $2<e<3$.\end{definition}

\vspace{.2in}
\begin{theorem} $e$ is irrational.\end{theorem}

\vspace{.2in}

\begin{proof}[Proof (due to Fourier)] Suppose to the contrary that $e=\frac{a}{b}$ where $a,b\in\
mathbb{N}^+$. Define $x:=b!\left(e-\sum_{k=0}^b\frac{1}{k!} \right)=b!\left(\frac{a}{b}-\
sum_{k=0}^b\frac{1}{k!} \right)=a(b-1)!-\sum_{k=0}^b\frac{b!}{k!}$
which is an integer since $k!|b!$ for every $0\leq k\leq b$.

\vspace{.2in}

We see that $x=b!\left(\sum_{k=0}^{\infty}\frac{1}{k!} -\sum_{k=0}^b\frac{1}{k!} \right)=\

sum_{k=b+1}^{\infty}\frac{b!}{k!}>0$. In the sum $\sum_{k=b+1}^{\infty}\frac{b!}{k!}$ each term
$\frac{b!}{k!}=\frac{1}{(b+1)(b+2)\dots (b+(k-b)}\leq \frac{1}{(b+1)^{k-b}}$. So $x=\
sum_{k=b+1}^{\infty}\frac{b!}{k!}\leq\sum_{k=b+1}^{\infty}\frac{1}{(b+1)^{k-b}}=\
sum_{j=1}^{\infty}\left(\frac{1}{b+1} \right)^j $ where the latter is a convergent geometric series
with limit $\left(\frac{1}{b+1}\right)\left(\frac{1}{1-\frac{1}{b+1}}\right)=\frac{1}{b}$. But clearly
$b\geq 2 $ since $2<e<3$ is not an integer, and so we have that $0<x\leq \frac{1}{2}$. But $x$ is an
integer, a contradiction.\end{proof}

\vspace{.2in}

In the above we defined the number $e$ somewhat conveniently so that we could give a proof that it is
irrational. There are several other equivalent 'definitions' of the quantity $e^x$ besides $\sum_{n=0}^{\
infty}\frac{x^n}{n!}$ some of which we state here:

\vspace{.2in}

\begin{theorem} The following definitions are equivalent to our definition of $e^x$ as $\

sum_{n=0}^{\infty}\frac{x^n}{n!}$:

\vspace{.1in}

\begin{enumerate}
\item $e^x:=y$ is the unique number $y$ so that $\int_1^y \frac{dt}{t}=x$.
\item $e^x=\lim_{n\rightarrow\infty}(1+\frac{x}{n})^n$.
\item $e^x:=z$ is the unique function $z(x)$ so that $z'(x)=z(x)$ and $z(0)=1$.
\end{enumerate}
\end{theorem}

\vspace{.2in}

The proofs of these equivalences can be found

\href{https://en.wikipedia.org/wiki/E_(mathematical_constant)}{HERE}.

\vspace{.2in}

We finish this section with a brief look at the relationship between real numbers and their decimal
representations.
\vspace{.2in}

\begin{theorem} Given $r\in\mathbb{R}$ $\exists d_0 \in\mathbb{Z}$ and a sequence $

(d_i)_{i=1}^{\infty}$ where $d_i\in\{0,1,2\dots, 9\}$ so that $r=\sum_{i=0}^{\infty}d_i10^{-i}.$\
end{theorem}

\vspace{.2in}

\begin{proof} Choose $d_0\in\mathbb{Z}$ so that $d_0\leq r<d_0+1$; $d_0$ exists by the

Archimedean property. Next choose $d_1\in\{0,1,2\dots,9\}$ so that $\frac{d_1}{10}\leq r-d_0<\
frac{d_1+1}{10}$; $d_1$ exists because $0\leq 10(r-d_0)<10$. Note that $0\leq r-d_0-\frac{d_1}
{10}<\frac{1}{10}$. Choose $d_2\in\{0,1,2\dots ,9\}$ so that $d_0+\frac{d_1}{10}+\frac{d_2}
{10^2}\leq r< d_0+\frac{d_1}{10}+\frac{d_2+1}{10^2}$; $d_2$ exists because $0\leq 10^2(r-d_0-\
frac{d_1}{10})<10$. Note that $0\leq r-\sum_{i=0}^2d_i10^{-i}<\frac{1}{10^2}$. Continue defining
the sequence $d_k$ recursively so that for each $k\in\mathbb{N}^+$ $d_k$ is chosen so that $\
frac{d_k}{10^k}+\sum_{i=0}^{k-1}d_i10^{-i}\leq r<\frac{d_k+1}{10^k}+\sum_{i=0}^{k-
1}d_i10^{-i}$ and hence $ 0\leq r-\sum_{i=0}^kd_i10^{-i}<\frac{1}{10^k}$. This defines a sequence
$d_k$ of the required form and we immediately get that $r=\sum_{i=0}^{\infty}d_i10^{-i}$ by the
Squeeze Theorem.\end{proof}

\vspace{.2in}

Note that decimal representations are far from being unique (for instance $0.9999999\dots$ is equal to
$1$ and so is $1.0000\dots$), something we don't get into detail here (we encourage the reader to check
out \href{https://en.wikipedia.org/wiki/Decimal_representation}{HERE}). Note in the above proof that
there was nothing special about the number $10$, we could have easily adapted the proof and replaced
$10$ by any base $b\in\mathbb{N}$ where $b\geq 2$ and so that the $d_i\in\{0,1,2\dots ,b-1\}$.

\vspace{.2in}

We've seen that the rational numbers don't satisfy the Completeness Property and that there are
infinitely many irrational numbers. In the next section we make sense of what it means to say there are
much more real numbers than there are rational numbers but that there are as many rational numbers as
there are natural numbers.

\newpage

\section{Cardinality of Sets}

\vspace{.2in}

We start by defining what it means to say that a set is finite. Define $[[0]]:=\emptyset$ and for $n\in\
mathbb{N}^+$ define $[[n]]:=\{1,2,3\dots, n\}$.

\vspace{.2in}

\begin{definition} We say that a set $X$ is finite of size $n\in\mathbb{N}$ if there is a bijection from
$[[n]]$ to $X$. In particular we have that $[[n]]$ is finite of size $n$.\end{definition}

\vspace{.2in}

\begin{example} The set $Y=\{b,l,o,n,d,i,e\}$ is a set of size $7$ since the function $f: [[7]]\rightarrow
Y$ given by $f(1)=b,f(2)=l,f(3)=o, f(4)=n,f(5)=d, f(6)=i, f(7)=e$ is clearly a bijection. In a very strong
sense $f$ is simply counting the elements of $Y$. \end{example}

\vspace{.2in}

What might seem obvious but requires a proof is the fact that if a set $X$ is finite of size $n$ and finite
of size $m$ then $m=n$; what is clear is that if there is a bijection $f:[[n]]\rightarrow X$ and a
bijection $g:[[m]]\rightarrow X$ then there is a bijection $g^{-1}\circ f:[[n]]\rightarrow[[m]]$. So we
need to establish the following:

\vspace{.2in}

\begin{theorem} If $m\not=n$ are natural numbers, there is no bijection from $[[m]]$ to $[[n]]$.\
end{theorem}

\vspace{.2in}

\begin{proof} Assume this is false and let $n\in\mathbb{N}$ be the smallest element of the natural
numbers where there is an $m\in\mathbb{N}$ with $m\not=n$ so that there is a bijection $f:[[n]]\
rightarrow[[m]]$. We first note that $m>n$ since if $m<n$ then $m$ would be smaller and satisfy the
above condition. We then note that $n\geq 1$ since otherwise $f$ would be a bijection between a non-
empty set and the empty set. So $1\in[[n]]$ and so $f(1)\in[[m]]$. Define a new function $g:[[n]]-\{1\}\
rightarrow [[m]]-\{f(1)\}$ by $g(t)=f(t)$ for all $t\in [[n]]-\{1\} $. It's easy to see this is also a
bijection. Let $h:[[n-1]]\rightarrow [[n]]-\{1\}$ be given by $h(t)=t+1$; it's easy to check that $h$ is a
bijection (it has inverse $h^{-1}:[[n]]-\{1\}\rightarrow [[n-1]]$ given by $h^{-1}(s)=s-1$). Lastly let
$p:[[m]]-\{f(1)\}\rightarrow [[m-1]]$ be given by $p(t)=t$ if $1\leq t<f(1)$ and $p(t)=t-1$ if $f(1)<t\
leq m$; $p$ is also a bijection. Then set $q=p\circ g\circ h: [[n-1]]\rightarrow [[m-1]]$ being a
composition of bijections is a bijection with $n-1\not= m-1$ contradicting our assumption that $n$ was
the smallest natural number with this property.
\end{proof}
\vspace{.2in}

We ask the reader to indeed carefully verify that the functions $p,g,h$ in the last proof are bijections.

\vspace{.2in}

\begin{theorem} Let $X$ and $Y$ be finite sets of the same size $n\in\mathbb{N}$ and let $f:X\
rightarrow Y$. Then $f$ is $1-1$ iff $f$ is a surjection iff $f$ is a bijection.\end{theorem}

\vspace{.2in}

\begin{proof} We proceed by induction to prove the statement "if a function from a set with $n$
elements to a set with $n$ elements is injective, then it is bijective (where $n\in\mathbb{N}$)," the
base case being clear (check that this is true for the empty function). Assume it is true for some fixed
but arbitrary $n\in\mathbb{N}$ and let $h:A\rightarrow B$ be a $1-1$ function between two sets
$A,B$ of size $n+1$. Let $x\in A$ and consider the function $h':A-\{x\}\rightarrow B-\{h(x)\}$
defined by $h'(a)=h(a)$ for every $a\in A-\{x\}$. Both $A-\{x\}$ and $B-\{h(x)\}$ have size $n$
(check this!) and $h'$ is certainly $1-1$ and so by our inductive hypothesis is a bijection. Let $y\in B$.
If $y=h(x)$ then $y\in im(h)$ otherwise $y\in B-\{h(x)\}$ and since $h'$ is onto there exists $a\in A-\
{x\}$ so that $h'(a)=y$. But then $h(a)=h'(a)=y$ and so $y\in im(h)$. So $h$ is onto and hence a
bijection and we are done by the PMI.

\vspace{.1in}

We now proceed by induction to prove the statement "if a function from a set with $n$ elements to a set
with $n$ elements is surjective then it is bijective (where $n\in\mathbb{N}$)," the base case being
clear (check that this is true for the empty function). Assume it is true for some fixed but arbitrary $n\
in\mathbb{N}$ and let $g:C\rightarrow D$ be an onto function between two sets of size $n+1$. Since
$C$ has size $n+1$ it is not empty and so $\exists c_0\in C$. Let $g':C-\{c_0\}\rightarrow D-\
{g(c_0)\}$ be defined by $g'(c)=g(c)$ for each $c\in C-\{c_0\}$. Then since $C-\{c_0\}$ and $D-\
{g(c_0)\}$ are finite sets of size $n$ (check this!) and certainly $g'$ is still onto we have by our
inductive hypothesis that $g'$ is a bijection. Suppose $g(x_1)=g(x_2)$ for some $x_1,x_2\in C$. Then
if neither $x_1$ nor $x_2$ is $c_0$ we would have $g'(x_1)=g(x_1)=g(x_2)=g'(x_2)$ and so
$x_1=x_2$ since $g'$ is $1-1$. Otherwise one of $x_1, x_2$ is equal to $c_0$, let's say $x_1=c_0$. If
$x_2\not=c_0$ then $g'(x_2)$ exists with $g'(x_2)=g(x_2)=g(x_1)=g(c_0)$ which is nonsense. So
$x_1=x_2$ and $g$ is $1-1$ and hence a bijection and we are done by the PMI.\end{proof}

\vspace{.1in}

Going back to our function $f:X\rightarrow Y$ assume $f$ is $1-1$. Then by the above we get that $f$
is a bijection and hence onto. If $f$ is onto, then by the above we get that $f$ is a bijection and hence
$1-1$.

\vspace{.2in}

\begin{theorem}[Pigeonhole Principle] If $m>n\geq 1$ are natural numbers then there does not exists a
$1-1$ function from $[[m]]$ to $[[n]]$.\end{theorem}

\vspace{.2in}

\begin{proof} Assume this is false and let $f:[[m]]\rightarrow [[n]]$ be $1-1$. We know that the
function $g:[[n]]\rightarrow [[m]]$ given by $g(x)=x$ is $1-1$ and so $g\circ f:[[m]]\rightarrow[[m]]$
is also $1-1$ and hence a bijection by the previous theorem. But $m\notin im (g\circ f)$ since if $\
exists x\in [[m]]$ so that $g(f(x))=m$ we'd have $f(x)=m$ which makes no sense since the codomain
of $f$ doesn't include $m$.\end{proof}

\vspace{.2in}
For some interesting applications of the seemingly obvious Pigeonhole Principle click
\href{https://www.ukessays.com/essays/mathematics/applications-of-the-pigeonhole-principle-
mathematics-essay.php}{HERE}.

\vspace{.2in}
We make an important note, that if $X$ and $Y$ are finite disjoint sets of size $m$ and $n$ then the
size of $X\cup Y$ is $m+n$; if we have bijections $f:[[m]]\rightarrow
X$ and $g:[[n]]\rightarrow Y$ the function $h:X\cup Y\rightarrow X\cup Y$ given by $h(a)=f(a)$ if
$a\in X$ and $h(a)=g(a)$ if $a\in Y$ is clearly a bijection. This simple fact gives rise to some
important counting methods such as the Inclusion-Exclusion Principle:

\vspace{.2in}

\begin{theorem}[Inclusion-Exclusion Principle] Let $X$ have size $n$ and $Y$ have size $m$. If $X\
cap Y$ has size $k$ then $X\cup Y$ has size $n+m-k$. Moreover if we let $|X|$ be the size of a finite
set $X$ then $|X_1\cup X_2\cup \dots \cup X_t|$ equals

$$
\sum_{i=1}^n|X_i|-\sum_{1\leq i\leq j\leq n}|X_i\cap X_j|+\sum_{1\leq i\leq j\leq k}|X_i\cap X_j\cap
X_k|-\dots +(-1)^{n+1}|X_1\cap X_2\cap \dots \cap X_n|
$$
\end{theorem}

\vspace{.2in}

\begin{proof}
First note that $X\cup Y=(X-(X\cap Y))\cup (X\cap Y)\cup (Y-(X\cap Y))$ where the latter three sets
are pairwise disjoint. So by the comment preceding the theorem we know that

$$
|X\cup Y|=|X-(X\cap Y)|+|X\cap Y|+|Y-(X\cap Y)|.
$$

But $|X-(X\cap Y)|=|X|-|X\cap Y|$ (why?) and $|Y-(X\cap Y)|=|Y|-|X\cap Y|$, and so altogether we
have

$$
|X\cup Y|=(n-k)+k+(m-k)=n+m-k.
$$
The general case can be proven a variety of ways, we direct the interested reader to the link \
href{http://storage.vitasmid.com/misc/iep-proof-for-dummies.pdf}{HERE}
\end{proof}

\vspace{.2in}

\begin{definition}Let $X$ and $Y$ be sets. We say $X$ has \emph{cardinality} less than or equal to
$Y$ if there exists a $1-1$
function $f:X\rightarrow Y$ and write $|X|\leq |Y|$. We write $|X|=|Y|$ if there is a bijection from $X$
to $Y$. If there is a $1-1$ function from $X$ to $Y$ but no bijection from $X$ to $Y$ we will write $|
X|<|Y|$.\end{definition}

\vspace{.2in}
An immediate consequence of the left/right inverse theorems is that $f:X\rightarrow Y$ is $1-1$ iff it
has a right inverse $g:Y\rightarrow X$ so that $g\circ f=id_X$; such a $g$ is necessarily onto (we ask
the reader to provide a proof of this), and so we can also write $|X|\leq |Y|$ is there is a surjection
$g:Y\rightarrow X$.

\vspace{.2in}

\begin{example} Let $n\leq m$ be natural numbers. Then $[[n]]\subseteq [[m]]$ and so the function $\
iota :[[n]]\rightarrow[[m]]$ given by $\iota(t)=t$ is $1-1$ and so $|[[n]]|\leq |[[m]]|$. If $X$ is a set of
size $n$ and $Y$ is a set of size $m$ then there are bijections $f:[[n]]\rightarrow X, g:[[m]]\rightarrow
Y$ and so $g\circ\iota\circ f^{-1}: X\rightarrow Y$ is $1-1$ and so $|X|\leq |Y|$. So the above
definition of cardinality agrees with our idea of size with respect to finite sets. \end{example}

\vspace{.2in}

\begin{example} Let $f:\mathbb{Z}\rightarrow\mathbb{N}$ be given by

\begin{equation*}
f(n)=\left\{\begin{array}{cc}
2n & \quad n\geq 0\\
-2n-1 & \quad n<0\end{array}
\right.
\end{equation*}

and $g:\mathbb{N}\rightarrow\mathbb{Z}$ be given by

\begin{equation*}
g(n)=\left\{\begin{array}{cc}
\frac{n}{2} & \quad n\quad even\\
-\frac{n+1}{2} & \quad n \quad odd \end{array}
\right.
\end{equation*}

\vspace{.1in}

We leave it to the reader to verify that $g\circ f=id_{\mathbb{Z}}$ and $f\circ g=id_{\mathbb{N}}$
and so both $f$ and $g$ are bijections. So $|\mathbb{N}|=|\mathbb{Z}|$. Note that $\mathbb{N}\
subsetneq\mathbb{Z}$ and so just because there are elements of $\mathbb{Z}$ that are not in $\
mathbb{N}$ does not mean that the cardinality of $\mathbb{Z}$ is bigger. In many ways sets that have
the same cardinality as $\mathbb{N}$ are the smallest infinite sets.\end{example}

\vspace{.2in}

\begin{definition} If $|X|\leq|\mathbb{N}|$ we say that $X$ is countable; if $|X|=|\mathbb{N}|$ we

say that $X$ is countably infinite.\end{definition}

\vspace{.2in}
\begin{example} Let $r: \mathbb{N}\times\mathbb{N}\rightarrow \mathbb{N}$ be given by

$$
r(m,n)=2^m(2n+1)-1.
$$

We claim that $r$ is a bijection. Let $k\in\mathbb{N}$. If $k=0$ we have $r(0,0)=k$ and if $k=1$
then $r(1,0)=k$. If $k$ is even then $r(0,\frac{k}{2})=k$ and if $k$ is odd then $k+1$ is even and so
there is a highest power of two dividing $k+1$ say $k+1=2^t(2s+1)$. Then $r(t,s)=k$, and so $r$ is
onto. Suppose that $r(m_1,n_1)=r(m_2,n_2)$ or $ 2^{m_1}(2n_1+1)-1=2^{m_2}(2n_2+1)-1$ which
implies $2^{m_1-m_2}(2n_1+1)=(2n_2+1)$ and since the integer on the right hand side is odd we
must have $m_1=m_2$ and so $2n_1+1=2n_2+1$ and hence $n_1=n_2$. So $r$ is $1-1$ and hence a
bijection. So $\mathbb{N}\times\mathbb{N}$ is countable. We leave it to the reader to show that $\
mathbb{N}^+$ and $\mathbb{Z}\times\mathbb{N}^+$ are also countable.\end{example}

\vspace{.2in}

\begin{example}Let $d:\mathbb{Z}\times\mathbb{N}^+\rightarrow\mathbb{Q}$ be given by

$d(a,b)=\frac{a}{b}$. Then clearly $d$ is onto. Since $\mathbb{Z}\times\mathbb{N}^+$ is countably
infinite there is a bijection $p:\mathbb{N}\rightarrow \mathbb{Z}\times\mathbb{N}^+ $ and so $d\
circ p:\mathbb{N}\rightarrow\mathbb{Q}$ is also a surjection. So $|\mathbb{Q}|\leq|\mathbb{N}|$
and $\mathbb{Q}$ is countable as well.\end{example}

\vspace{.2in}

We say that a set is \emph{infinite} if it is not a finite set. We show now that the natural numbers have
the smallest cardinality of any infinite set.

\vspace{.2in}

\begin{theorem} Let $X$ be an infinite set. Then $|\mathbb{N}|\leq|X|.$\end{theorem}

\vspace{.2in}

\begin{proof} Since $X$ is infinite it is non-empty and so contains some $x$. Define the following
elements of $X$ recursively: $x_0=x$ and $x_{k+1}$ is an element of $X-\{x_0,x_1,\dots, x_k\}$,
where we know $x_{k+1}$ exists since otherwise $X=\{x_0,x_1, \dots, x_k\}$ would be finite of size
$k+1$. So this defines a sequence $(x_i)_{i=1}^{\infty}$ of distinct elements of $X$ and a sequence is
exactly a function from $\mathbb{N}\rightarrow X$. The fact that these elements are distinct implies
that this function is $1-1$.\end{proof}

\vspace{.2in}

There is a philosophical point that is getting somewhat sidestepped in the above argument, something
called the Axiom of Choice; this was also sidestepped much earlier in the lecture notes when we proved
that every surjection has a right inverse. But we won't get into this here other than it is well worth a
Wikipedia search. We mention another way we can distinguish finite sets from infinite sets (this
property is called being \emph{Dedekind infinite}).
\vspace{.2in}

\begin{theorem} A set $X$ is infinite implies it has the same cardinality as a proper subset of itself.\
end{theorem}

\vspace{.2in}

\begin{proof} If $X$ is infinite then there is a $1-1$ function $f:\mathbb{N}\rightarrow X$. Define a
function $h:X\rightarrow X-\{f(0)\}$ by:

\begin{equation*}
h(x)=\left\{\begin{array}{cc}
f(n+1) & \quad \exists n\in\mathbb{N}\,, x=f(n)\\
x & \quad x\not\in im(f)\end{array}
\right.
\end{equation*}

We leave the details that this is a bijection as an exercise to the reader, the idea being that $h$ leaves
the elements of the set $X$ that are not in $im(f)$ alone and then just sends $h(f(0))=f(1),
h(f(1))=f(2), \dots$.\end{proof}

\vspace{.2in}

\begin{lemma} If $|X|\leq|Y|$ and $|Y|\leq |Z|$ then $|X|\leq |Z|$.\end{lemma}

\vspace{.2in}

\begin{proof} This follows from the fact that the composition of injective functions is injective.\
end{proof}

\vspace{.2in}

Suppose we know $|X|\leq |Y|$ and $|Y|\leq |X|$. It's not immediately clear at all that one can conclude
that $|X|=|Y|$ since $|X|\leq |Y|$ and $|Y|\leq |X|$ simply means there exists $1-1$ functions $f:X\
rightarrow Y, g:Y\rightarrow X$, these might not be bijections, and it is careful work to see that given
this information that a bijection exists. It turns out that there always does exist a bijection given these
injections, this is a result known as the Cantor-Bernstein Theorem and we will give a proof of this
towards the end of this section. We've seen that an interesting variety of sets are countably infinite: $\
mathbb{N},\mathbb{Z},\mathbb{N}\times\mathbb{N}, \mathbb{Q}$ for instance. But are there
infinite sets that are not countably infinite? Is for instance $\mathbb{R}$ countably infinite? The
answer to that is a very profound no. We call an infinite set that is not countable an \
emph{uncountable} set.

\vspace{.2in}

\begin{lemma} Let $a<b$ and $c<d$ be real numbers. Then $|(a,b)|=|(c,d)|$ and $|(a,b)|=|\mathbb{R}|
$.\end{lemma}

\vspace{.2in}
\begin{proof} $p:(0,1)\rightarrow (c,d)$ by $p(x)=(c-d)x+d$ is clearly a bijection with inverse $p{-1}:
(c,d)\rightarrow (0,1)$ given by $p^{-1}(x)=\frac{x-d}{c-d}$ and so $|(0,1)|=|(c,d)|$. Likewise $|(0,1)|
=|(a,b)|$ and so $|(a,b)|=|(c,d)|$. The function $t:(-\pi/2,\pi/2)\rightarrow \mathbb{R}$ given by $t(x)=\
tan(x)$ is a bijection from $(-\pi/2,\pi/2)$ to $\mathbb{R}$ and so $|(0,1)|=|(-\pi/2,\pi/2)|=|\mathbb{R}|
$.\end{proof}

\vspace{.2in}

\begin{theorem}[Cantor]$\mathbb{R}$ is uncountable.\end{theorem}

\vspace{.2in}

\begin{proof} We show that $(0,1)$ is uncountable and then the theorem follows from the preceding
lemma. Suppose to the contrary that $(0,1)$ was countably infinite, i.e, there exists a bijection $c:\
mathbb{N}^+\rightarrow (0,1)$. We write out this bijection as a list representing each element of
$(0,1)$ as a decimal; we choose representatives that do not have repeating $9's$, so we choose the
representation $0.005000000\dots$ over $0.0049999\dots$ and so on:

$$
c(1)=0.d_{11}d_{12}d_{13}d_{14}d_{15}\dots
$$
$$
c(2)=0.d_{21}d_{22}d_{23}d_{24}d_{25}\dots
$$
$$
c(3)=0.d_{31}d_{32}d_{33}d_{34}d_{35}\dots
$$
$$
c(4)=0.d_{41}d_{42}d_{43}d_{44}d_{45}\dots
$$

etc....where $d_{ij}$ is the $j$-th decimal place of the $i$-th number on the list. Define a sequence $
(a_k)_{k=1}^{\infty}$ by the rule $a_k=9-d_{kk}$ for each $k\geq 1$. Note that we have $a_k\
not=d_{kk}$ otherwise we would have $a_k+d_{kk}=2d_{kk}=9$ which can't happen since $9$ is
odd.

\vspace{.1in}

Let $a=0.a_1a_2a_3a_4\dots$ where the $k$-th decimal place of $a$ is $a_k$. Then $a\notin im(c)$
since otherwise we would have $0.d_{j1}d_{j2}d_{j3}d_{j4}d_{j5}\dots=0.a_1a_2a_3a_4\dots$
for some $j\in\mathbb{N}^+$ which is impossible since $d_{jj}\not=a_j$ by definition of the
sequence $a_k$. So $c$ is not onto giving us a contradiction. So $(0,1)$ is uncountable.\end{proof}

\vspace{.2in}

\begin{theorem} Let $X$ and $Y$ be countably infinite. Then $X\cup Y$ is countably infinite. \
end{theorem}
\vspace{.2in}

\begin{proof} Let $Y'=Y-X$ and note that $X\cup Y= X\cup Y'$ and $X\cap Y'=\emptyset$. We show
that $X\cup Y'$ is countably infinite. If $Y'$ is finite then we leave it as an exercise to show that $X\
cup Y'$ is countably infinite. Otherwise assume that $X$ and $Y'$ are both countably infinite and let
$f:X\rightarrow \mathbb{N}, g:Y'\rightarrow \mathbb{N}$ be bijections. Define $h: X\cup Y'\
rightarrow\mathbb{N}$ as follows:

\begin{equation*}
h(x)=\left\{\begin{array}{cc}
2f(x) & \quad x\in X\\
2g(x)+1 & \quad x\in Y'\end{array}
\right.
\end{equation*}

which is well-defined since $X\cap Y'=\emptyset$. If $h(x_1)=h(x_2)$ either $x_1,x_2\in X$ or

$x_1,x_2\in Y';$ either way we would get $x_1=x_2$ since $f$ and $g$ are $1-1$. If $n\in\mathbb{N}
$ then if $n=2k$ for some $k\in\mathbb{N}$ then since $f$ is onto $\exists x\in X$ so that $f(x)=k$.
But then $h(x)=2f(x)=2k=n$. If $n=2k+1$ for some $k\in\mathbb{N}$ then since $g$ is onto $\exists
y\in Y'$ so that $g(y)=k$. But then $h(y)=2g(y)+1=2k+1=n$. So $h$ is onto and hence a bijection.\
end{proof}

\vspace{.2in}

\begin{theorem} The set of irrational numbers $\mathbb{R}-\mathbb{Q}$ is uncountable.\

end{theorem}

\vspace{.2in}

\begin{proof} Assume to the contrary that $\mathbb{R}-\mathbb{Q}$ is countable. Then $\

mathbb{R}=\mathbb{Q}\cup (\mathbb{R}-\mathbb{Q})$ would be countable being the union of two
countable sets. So $\mathbb{R}-\mathbb{Q}$ is uncountable.\end{proof}

\vspace{.2in}

Given a set $X$ there is always a way to find another set $Y$ so that $|X|<|Y|$. We first look examine
how this is done with a finite set. We recall from earlier in the course the definition of the power set of
a set: given $X$ we define $P(X):=\{Y\,|\, Y\subseteq X\}$.

\vspace{.2in}

\begin{example} Let $X=[[4]]=\{1,2,3,4\}$. Then $P([[4]])$ contains the following elements arranged
by size:

$$
\emptyset\quad \quad \quad \text{size 0}
$$
$$
\{1\},\{2\},\{3\},\{4\}\quad \text{size 1}
$$
$$
\{1,2\},\{1,3\},\{1,4\},\{2,3\},\{2,4\},\{3,4\} \quad \text{size 2}
$$
$$
\{2,3,4\},\{1,3,4\},\{1,2,4\},\{1,2,3\}\quad \text{size 3}
$$
$$
[[4]]=\{1,2,3,4\}\quad\quad \text{size 4}
$$

\vspace{.2in}

Altogether we have $1+4+6+4+1$ subsets for a total of $2^4$ subsets. If we take $[[n]]$ where $n\in\
mathbb{N}$ we see that there are $\binom{n}{k}$ subsets of $[[n]]$ of size $k$ and so in total there
are

$$
\binom{n}{0}+\binom{n}{0}+\dots +\binom{n}{n}=\sum_{k=0}^n\binom{n}{k}=2^n
$$

subsets and so $|P([[n]])|=|[[2^n]]|$. We note here that $n<2^n$ is true for any $n\in\mathbb{N}$. \
end{example}

\vspace{.2in}

We can say a similar thing about every set regardless of whether it is finite or infinite:

\vspace{.2in}

\begin{theorem}[Cantor]Let $X$ be a set. Then $|X|<|P(X)|$.\end{theorem}

\vspace{.2in}

\begin{proof} Clearly $|X|\leq |P(X)|$ since the function $s:X\rightarrow P(X)$ given by $s(a)=\{a\}$
is $1-1$: $s(a_1)=s(a_2)$ means $\{a_1\}=\{a_2\}$ and hence $a_1=a_2$. Assume to the contrary that
there is a bijection $c:X\rightarrow P(X)$. Define

$$
D=\{a\in X\,|\, a\notin c(a)\}.
$$

Then $D\subseteq X$ or $D\in P(X)=im(c)$ and so $\exists a\in X$ so that $c(a)=D$. But then we
have

$$
a\in D=c(a) \iff a\not\in c(a)=D,
$$
which is clearly absurd. So there is no such bijection and $|X|<|P(X)|$.\end{proof}

\vspace{.2in}

If we start with $\mathbb{N}$ and keep forming successive power sets we get:

$$
|\mathbb{N}|<|P(\mathbb{N})|<|P(P(\mathbb{N}))|<|P(P(P(\mathbb{N})))|<\dots
$$

\vspace{.1in}

Since $|\mathbb{N}|<|\mathbb{R}|$ it's natural to ask whether $|P(\mathbb{N})|\leq |\mathbb{R}|$ or

whether $|\mathbb{R}|\leq|P(\mathbb{N})|$. We will see that $|P(\mathbb{N})|=|\mathbb{R}|$ but
we'll need a few tools first.

\vspace{.2in}

\begin{theorem}[Tarski Fixed Point Theorem] If a function $h:P(X)\rightarrow P(X)$ satisfies $A\

subseteq B\implies h(A)\subseteq h(B)$ then $\exists S\subseteq X$ so that $h(S)=S$.\end{theorem}

\vspace{.2in}

\begin{proof} Let $L=\{A\subseteq X\,|\, A\subseteq h(A)\}$ and let $S=\cup L$. Then any $B\in L$ is
a subset of $S$ and so $B\subseteq h(B)\subseteq h(S)$ and we have $S=\cup L\subseteq h(S)$ and
then $S\in L$. $h(S)\subseteq h(h(S))$ implies that $h(S)\in L$ and so $h(S)\subseteq S$ or $h(S)=S$
as desired.\end{proof}

\vspace{.2in}

\begin{theorem}[Cantor-Bernstein Theorem] If $|X|\leq |Y|$ and $|Y|\leq |X|$ then $X|=|Y|$.\

end{theorem}

\vspace{.2in}

\begin{proof} Let $f:X\rightarrow Y, g:Y\rightarrow X$ be $1-1$ and define $h:P(X)\rightarrow P(X)$

by

$$
h(A)=X-g(Y-f(A)).
$$

Suppose $A\subseteq B\subseteq X$. Then $f(A)\subseteq f(B)$ and $Y-f(B)\subseteq Y-f(B)$ and
$g(Y-f(B))\subseteq g(Y-f(A))$ and lastly $X-g(Y-f(A))\subseteq X-g(Y-f(B))$, so $h(A)\subseteq h(B)
$. So by the lemma there is an $S$ so that $S=h(S)=X-g(Y-f(S))$ or $X-S=g(Y-f(S))$. The function
$p:X\rightarrow Y$ given by
\begin{equation*}
p(x)=\left\{\begin{array}{cc}
f(x) & \quad x\in S\\
g^{-1}(x)& \quad x\in X-S\end{array}
\right.
\end{equation*}

is a bijection since $f$ maps $S$ bijectively onto its image $f(S)$ and $g$ maps $Y-f(S)$ bijectively
onto its image $g(Y-f(S))=X-S$ (i.e. $g^{-1}$ maps $X-S$ bijectively onto its image $Y-f(S)$). \
end{proof}

\vspace{.2in}

\begin{theorem} $|\mathbb{R}|=|P(\mathbb{N})|$.\end{theorem}

\vspace{.2in}

\begin{proof} Let $t:P(\mathbb{N})\rightarrow \mathbb{R}$ be given by $t(S)=\sum_{n\in S}10^{-

n}$. Then if $t(S_1)=t(S_2)$ these real numbers agree at every place since they consist of decimals
with $0$'s and $1$'s in exactly the same places and so $S_1=S_2$ and so $t$ is $1-1$ and $|P(\
mathbb{N})|\leq |\mathbb{R}|$.

\vspace{.2in}

Since $\mathbb{Q}$ is countable $\mathbb{Q}=\{q_0,q_1,q_2,q_3,\dots\}$. Define $s: \mathbb{R}\

rightarrow P(\mathbb{N})$ be given by

$$
s(x)=\{n\in\mathbb{N}\,|\, q_n<x\}.
$$

If $s(x_1)=s(x_2)$ then any $q_i<x_1$ satisfies $q_i<x_2$ and vice a versa and so $x_1=x_2$. So $|\
mathbb{R}|\leq|P(\mathbb{N})|$ and so by the Cantor-Bernstein Theorem we have $|\mathbb{R}|=|P(\
mathbb{N})|$.\end{proof}

\vspace{.2in}

Earlier in the course notes we looked at an example of a partial order on the set of infinite binary
sequences $B=\{f\,|\,f:\mathbb{N}\rightarrow \{0,1\}\}.$ We look now to show that this set has the
same cardinality as $\mathbb{R}$.

\vspace{.2in}

\begin{theorem} $|B|=|\mathbb{R}|=|P(\mathbb{N})|$.\end{theorem}

\vspace{.2in}

\begin{proof} For each $A\subseteq N$ define a function $\chi_A:\mathbb{N}\rightarrow \{0,1\}$ as

follows:

\begin{equation*}
\chi_A(n)=\left\{\begin{array}{cc}
1 & \quad n\in A\\
0 & \quad n\notin A\end{array}
\right.
\end{equation*}

Let $\alpha:P(\mathbb{N})\rightarrow B$ be given by $\alpha(A)=

\chi_A$. Define $\beta: B\rightarrow P(\mathbb{N})$ be given by $\beta(f)=\{n\in\
mathbb{N}\,|\,f(n)=1\}$. Then it's simple to verify that $\alpha\circ \beta=id_B$ and $\beta\circ\
alpha=id_{P(\mathbb{N})}$ and so both are bijections.\end{proof}

\vspace{.2in}

There was absolutely nothing special about the natural numbers in this last theorem, we could replace
$\mathbb{N}$ with any set and still have a theorem:

\vspace{.2in}

\begin{corollary} Let $X$ be any set and let $2^X=\{f\,|\, f:X\rightarrow \{0,1\}\}$ be the set of all
functions from $X$ to $\{0,1\}$. Then $|2^X|=|P(X)|$.\end{corollary}

\vspace{.2in}

Note that when $X$ is finite this is already something we've observed via the Binomial Theorem.

\vspace{.2in}

We can think of the rational numbers $\mathbb{Q}$ as the set of all solutions to linear equations with
integer coefficients, that $\mathbb{Q}=\{x\in\mathbb{R}\,|\, \exists a,b\in\mathbb{Z}\, \, (ax+b=0)\}.$
If we look at all real numbers that show up as roots of polynomial equations with integer coefficients
we get what are called the \emph{real algebraic numbers}.

\vspace{.2in}

\begin{definition} Let $\mathbb{K}\subseteq\mathbb{R}$ be the following subset:

$$
\{x\in\mathbb{R}\,|\, \exists a_i\in\mathbb{Z}, a_nx^n+a_{n-1}x^{n-1}+\dots+a_1x+a_0=0\},
$$

and call the set $\mathbb{K}$ the (real) algebraic numbers. The set $\mathbb{R}-\mathbb{K}$ is
called the set of (real) transcendental numbers.\end{definition}

\vspace{.2in}

Clearly we have $\mathbb{N}\subsetneq \mathbb{Z}\subsetneq\mathbb{Q}\subsetneq \mathbb{K}\

subseteq\mathbb{R}$.

\vspace{.2in}

\begin{example} This might seem like a somewhat esoteric definition but $\mathbb{K}$ contains a lot
of numbers we are already very familiar with like $\sqrt{5}$ since it is a solution to $x^2-5=0$, $\
sqrt[11]{32}$ since it is a solution to $x^{11}-32=0$, and $\sqrt{2}+\sqrt{3}$. It might not be clear
that the last quantity is an algebraic number even though we can see that $\sqrt{2}$ and $\sqrt{3}$ are
algebraic, so we find an appropriate polynomial by being pretty algebraic: if $r=\sqrt{2}+\sqrt{3}$
then $r^2=5+2\sqrt{6}$ and so $(r^2-5)^2=24$ or $r^4-10r^2+1=0$ and so one polynomial with
integer coefficients that has $\sqrt{2}+\sqrt{3}$ as a root of $x^4-10x^2+1$. A more complicated
example is that $\cos(\frac{\pi}{7})$ is algebraic; it is a root of $8x^3-4x^2-4x+1$. There are lots of
algebraic numbers that can't be written in terms of radicals, for instance the real root of $x^5-x+1=0$,
although this is extremely non-trivial to show. In fact $\mathbb{K}$ is an ordered field, a fact that we
won't show in these notes but you can read about
\href{https://www.maths.tcd.ie/pub/Maths/Courseware/428/Primes-III.pdf}{HERE}.
\end{example}

\vspace{.2in}

What is not immediately clear is that there are transcendental numbers and it is beyond the scope of
these notes to actual prove that a given number is
transcendental. It has been shown that $\pi$ and $e$ are both transcendental. The first explicit
transcendental numbers were written down and proven to be transcendental numbers by Liouville in
1844; the interested reader can read about them \href{https://en.wikipedia.org/wiki/Liouville_number}
{HERE}. In particular Liouville showed that the number $\sum_{k=0}^{\infty}10^{-k!}$ is
transcendental, an elementary proof of which can be found
\href{http://deanlm.com/transcendental/construction_of_a_transcendental_number.pdf}{HERE}. We
will however show that there exists countably many algebraic numbers and hence there exists
uncountably many transcendental numbers, a result due to Cantor in 1874. We need a few more tools
regarding countable sets in order to do this.

\vspace{.2in}

Let $C=\{A_0,A_1,A_2, A_3, \dots\}$ be a countable (finite or countably infinite) collection of

countable (finite or countably infinite) sets. We recall that $\cup C=\{x\,|\, \exists A_k, x\in A_k\}$ is
the union of all of the sets in $C$. For example if $C=\{\{1,2,3\},\{3,4,5\},\{5,6,7\}\}$ then $\cup C=\
{1,2,3,4,5,6,7\}$.

\vspace{.2in}

\begin{lemma} Given any countable collection of sets

$C=\{A_0,A_1,A_2, A_3, \dots\}$ we can find another countable collection of pairwise disjoint sets
$C'=\{B_0,B_1,B_2, B_3, \dots\}$ so that $\cup C=\cup C'$.\end{lemma}

\vspace{.2in}

\begin{proof} We define the sequence $B_0,B_1, \dots$ recursively as follows:

Set $B_0:=A_0$;
$B_1:=A_1-A_0$;
$B_2:=A_2-(A_0\cup A_1)$
$B_3:=A_3-(A_0\cup A_1\cup A_2) $ and for $n\geq 1$ we set $B_n:=A_n-(A_0\cup A_1\cup \dots \
cup A_{n-1})\subseteq A_n$. Let $C'=\{B_0, B_1, \dots \}.$ Suppose $s<t$ are natural numbers. Then
$x\in B_s$ implies that $x\in A_s$ and so $x\notin B_t=A_t-(A_0\cup A_1\cup\dots A_{t-1})$ since
$A_s$ is in the list $A_0, A_1, \dots A_{t-1}$. So $B_s\cap B_t=\emptyset$. Since $B_k\subseteq
A_k$ we get immediately that $\cup C'\subseteq \cup C$. Let $x\in\cup C$ so that that $\exists n$ so
that $x\in A_n$. Pick $n_0$ to be the smallest natural number so that $x\in A_{n_0}$. Then $x\in
B_{n_0}$ since $x\notin A_k$ for $k<n_0$ and so $x\in \cup C'$ and $\cup C'=\cup C$.\end{proof}

\vspace{.2in}

For example if $C=\{\{1,2,3\},\{3,4,5\},\{5,6,7\}\}$ where $A_0=\{1,2,3\}, $

$A_1=\{3,4,5\}, A_2=\{5,6,7\}$ then $C'=\{\{1,2,3\},\{4,5\},\{6,7\}\}$. We use this to establish the
following generalization that the union of two countable sets is countable:

\vspace{.2in}

\begin{theorem} Suppose $C=\{A_0,A_1,A_2, A_3, \dots\}$ is a countable collection of countable

sets. Then $\cup C$ is countable. \end{theorem}

\vspace{.2in}

\begin{proof} If we apply the construction of the collection $C'$ from $C$ as in the lemma, each
$B_k\subseteq A_k$ where now we are assuming $A_k$ is countable. Hence the $B_k$ are countable
as well (why?). Since $\cup C'=\cup C$ we show that $\cup C'$ is countable. List the prime numbers
by $p_0=2, p_1=3, p_2=5, \dots$. Since each $B_k$ is countable there is by definition a $1-1$
function $f_k:B_k\rightarrow \mathbb{N}$. Define $f:C'\rightarrow \mathbb{N}$ as follows: if $x\in
B_k$ then set $f(x)=p_k^{f_k(x)+1}$. Since the $B_k$ are disjoint an $x\in C'$ can only belong to
exactly one of them so this is a well defined function. To show $f$ is $1-1$ let $x\not=y$ be elements
of $C'$. Then if $x,y\in B_m$ for some $m$ then we have $f_m(x)\not=f_m(y)$ since $f_m$ is $1-1$
and so $f(x)\not=f(y)$; if $x\in B_m$ and $y\in B_n$ with $n\not=m$ then $f(x)$ is a positive power
of $p_m$ while $f(y)$ is a positive power of $p_n$, and these are different by the Fundamental
Theorem of Arithmetic. So $f$ is $1-1$ and $\cup C'=\cup C$ is countable.\end{proof}

\vspace{.2in}

We are now ready to prove that the set of algebraic numbers is countable; it is an immediate
consequence that the set of transcendental numbers is uncountable since $\mathbb{R}$ being
uncountable is not the union of two countable sets. We will assume as true that a real polynomial of
degree $n$ has at most $n$ real roots.

\vspace{.2in}

\begin{theorem} The set $\mathbb{K}$ of algebraic numbers is countable.\end{theorem}

\vspace{.2in}

\begin{proof} For each $k\in\mathbb{N}$ set $P_k:=\{p=a_nx^n+a_{n-1}x^{n-1}+\dots +a_0\,| \, |

a_n|+|a_{n_1}|+\dots +|a_0|+n=k, a_i\in\mathbb{Z}\}$, i.e., the set of all polynomials with integer
coefficients so that the sum of the degree and the absolute value of the coefficients is exactly $k$. Each
$P_k$ is finite (why?) and hence countable. Let $R_k:=\{r\in\mathbb{R}\,|\, p(r)=0, p\in P_k\}$ be the
set of all real roots of polynomials in $P_k$ which since each polynomial in this set has finitely many
roots must also be a finite set and hence countable. Let $C=\{R_0, R_1, R_2\dots\}$ and observe that
$\mathbb{K}=\cup C$ is a countable union of countable sets and so is countable.\end{proof}

\end{document}