Chapter 1 · Sets, Real Numbers and Countability

1.2 The Real Numbers as a Complete Ordered Field

Thirteen axioms describe $\mathbb{R}$ completely: the field rules, the order rules, and one more — every nonempty set that is bounded above has a least upper bound. Everything in analysis grows from that last axiom.

The rational numbers already form an excellent number system: we can add, subtract, multiply and divide, and we can compare any two of them. Yet $\mathbb{Q}$ is full of holes. The rationals $1,\ 1.4,\ 1.41,\ 1.414,\ 1.4142,\dots$ creep up on a number whose square is $2$, and no rational number is there to be reached. The sequence is increasing and bounded, but in $\mathbb{Q}$ it converges to nothing. Calculus cannot live with such holes: the intermediate value theorem, the existence of maxima, the convergence of monotone sequences and of Cauchy sequences, even the existence of $\sqrt2$ — all of them fail in $\mathbb{Q}$.

This section describes $\mathbb{R}$ by the properties that matter: it is an ordered field (like $\mathbb{Q}$) that is also complete — every nonempty set that is bounded above has a least upper bound. From this single extra axiom we derive the Archimedean property, the density of rationals and irrationals, the existence of roots, the nested interval property, and decimal and binary expansions. In CSIR NET these ideas are examined directly (computing a supremum, deciding which ordered field is complete or Archimedean, which sets are dense) and, far more often, indirectly: nearly every proof in Chapters 2–7 starts with "let $u=\sup S$".

Ordered Fields and Absolute Value

📘 Definition 1.8 — Field and ordered field

A field is a set $F$ with two operations $+$ and $\cdot$ and two distinct elements $0,1$ such that for all $x,y,z\in F$:

Addition$x+y=y+x$; $(x+y)+z=x+(y+z)$; $x+0=x$; each $x$ has a negative $-x$ with $x+(-x)=0$.
Multiplication$xy=yx$; $(xy)z=x(yz)$; $1\cdot x=x$; each $x\ne0$ has an inverse $x^{-1}$ with $xx^{-1}=1$.
Distributive law$x(y+z)=xy+xz$.

An ordered field is a field $F$ with a subset $P$ (the positive elements) such that (O1) for every $x\in F$ exactly one of $x\in P$, $x=0$, $-x\in P$ holds (trichotomy); and (O2) if $x,y\in P$ then $x+y\in P$ and $xy\in P$. We write $x\lt y$ (or $y\gt x$) when $y-x\in P$, and $x\le y$ when $x\lt y$ or $x=y$. Examples: $\mathbb{Q}$, $\mathbb{R}$, $\mathbb{Q}(\sqrt2)=\{a+b\sqrt2:a,b\in\mathbb{Q}\}$ with the order inherited from $\mathbb{R}$.

📐 Theorem 1.18 — Rules for inequalities
In every ordered field: (a) $\lt$ is transitive, and for any $x,y$ exactly one of $x\lt y$, $x=y$, $x\gt y$ holds; (b) $x\lt y\Rightarrow x+z\lt y+z$; (c) $x\lt y$ and $z\gt0$ $\Rightarrow xz\lt yz$, while $x\lt y$ and $z\lt0$ $\Rightarrow xz\gt yz$; (d) $x\ne0\Rightarrow x^2\gt0$; in particular $1\gt0$; (e) $x\gt0\Rightarrow x^{-1}\gt0$, and $0\lt x\lt y\Rightarrow0\lt y^{-1}\lt x^{-1}$; (f) $n\cdot1=1+1+\cdots+1\gt0$ for every $n\in\mathbb{N}$, so an ordered field has characteristic $0$ and contains a copy of $\mathbb{Q}$.

Proof. (a) If $y-x\in P$ and $z-y\in P$ then $z-x=(z-y)+(y-x)\in P$ by (O2); trichotomy for $y-x$ is (O1). (b) $(y+z)-(x+z)=y-x$. (c) $yz-xz=(y-x)z$ is a product of two positives; if $z\lt0$, $xz-yz=(y-x)(-z)\in P$. (d) If $x\in P$, $x^2=x\cdot x\in P$; if $-x\in P$, $x^2=(-x)(-x)\in P$. Since $1\ne0$, $1=1^2\gt0$. (e) If $x^{-1}$ were negative, $1=x\cdot x^{-1}=-\left(x\cdot(-x^{-1})\right)$ would be negative. For $0\lt x\lt y$: $x^{-1}-y^{-1}=(y-x)\,x^{-1}y^{-1}\gt0$. (f) $n\cdot1$ is a sum of positive terms, hence positive and nonzero; so $m\cdot1\ne n\cdot1$ for $m\ne n$, and $\tfrac{m}{n}\mapsto(m\cdot1)(n\cdot1)^{-1}$ embeds $\mathbb{Q}$ in $F$ preserving $+$, $\cdot$ and order. ∎

Two quick consequences are popular exam traps. $\mathbb{C}$ cannot be ordered: in any order $i^2=-1$ would be positive by (d), and also $1=1^2\gt0$, so $0=1+(-1)$ would be positive — contradicting trichotomy. No finite field can be ordered: in $\mathbb{Z}_p$ the positive element $p\cdot1$ equals $0$, contradicting (f).

📘 Definition 1.9 — Absolute value and distance
In an ordered field, $|x|=x$ if $x\ge0$ and $|x|=-x$ if $x\lt0$. The distance between $x$ and $y$ is $|x-y|$. Thus $|x-a|\lt\varepsilon$ means $a-\varepsilon\lt x\lt a+\varepsilon$: "$x$ lies in the $\varepsilon$-neighbourhood of $a$".
📐 Theorem 1.19 — Properties of the absolute value
For all $x,y$ and every $a\ge0$: (a) $|x|\ge0$, with $|x|=0\iff x=0$; $|xy|=|x||y|$; $|-x|=|x|$. (b) $|x|\le a\iff-a\le x\le a$. (c) Triangle inequality: $|x+y|\le|x|+|y|$, with equality iff $xy\ge0$. (d) Reverse triangle inequality: $\big||x|-|y|\big|\le|x-y|$. (e) If $|x|\le\varepsilon$ for every $\varepsilon\gt0$, then $x=0$.

Proof. (a), (b) follow by checking the signs case by case. (c) From $-|x|\le x\le|x|$ and $-|y|\le y\le|y|$, adding gives $-(|x|+|y|)\le x+y\le|x|+|y|$, which is $|x+y|\le|x|+|y|$ by (b). If $xy\ge0$, $x$ and $y$ have the same sign and equality is clear. If $xy\lt0$ then $$(x+y)^2=x^2+2xy+y^2\lt x^2+2|x||y|+y^2,$$ which is $(|x|+|y|)^2$; since both $|x+y|$ and $|x|+|y|$ are $\ge0$ this forces $|x+y|\lt|x|+|y|$. (d) $|x|=|(x-y)+y|\le|x-y|+|y|$ gives $|x|-|y|\le|x-y|$; exchanging $x$ and $y$ gives $|y|-|x|\le|x-y|$; combine with (b). (e) If $x\ne0$, take $\varepsilon=|x|/2\gt0$: then $|x|\le|x|/2$, so $|x|\le0$, contradicting (a). ∎

Part (e) is the analyst's way of proving equalities: to show $a=b$, show $|a-b|\le\varepsilon$ for every $\varepsilon\gt0$.

✏️ Example 1 — An absolute-value inequality
Solve $|x-1|+|x+2|\lt5$. Then explain, with the triangle inequality, why $|x-1|+|x+2|\ge3$ for every real $x$, and find where equality holds.
Split at $-2$ and $1$ For $x\ge1$: $(x-1)+(x+2)=2x+1\lt5\iff x\lt2$, giving $[1,2)$. For $-2\le x\lt1$: $(1-x)+(x+2)=3\lt5$ always, giving $[-2,1)$. For $x\lt-2$: $(1-x)-(x+2)=-2x-1\lt5\iff x\gt-3$, giving $(-3,-2)$.
Answer The union is the open interval $(-3,2)$. Check: at $x=2$ and $x=-3$ the sum is exactly $5$.
Lower bound $|x-1|+|x+2|=|1-x|+|x+2|\ge|(1-x)+(x+2)|=3$. Equality in the triangle inequality holds iff $(1-x)(x+2)\ge0$, i.e. $-2\le x\le1$.
Meaning $|x-1|+|x+2|$ is the sum of the distances from $x$ to $1$ and to $-2$: it is smallest ($=3$, the distance between the two points) exactly when $x$ lies between them.

Completeness: Supremum and Infimum

📘 Definition 1.10 — Bounds, supremum and infimum

Let $S$ be a subset of an ordered field $F$.

  • $u\in F$ is an upper bound of $S$ if $s\le u$ for every $s\in S$; $S$ is bounded above if it has one. Lower bounds and "bounded below" are defined symmetrically; bounded means bounded above and below.
  • $u$ is a supremum (least upper bound) of $S$, written $u=\sup S$, if $u$ is an upper bound of $S$ and $u\le v$ for every upper bound $v$ of $S$. Similarly $\ell=\inf S$ (greatest lower bound).
  • If $\sup S\in S$ it is called the maximum $\max S$; if $\inf S\in S$, the minimum $\min S$.
📘 Definition 1.11 — The completeness axiom
An ordered field $F$ is complete (has the least upper bound property) if every nonempty subset of $F$ that is bounded above has a supremum in $F$. The real number system $\mathbb{R}$ is a complete ordered field. Any two complete ordered fields are isomorphic by a unique order-preserving field isomorphism, so the axioms determine $\mathbb{R}$ completely; $\mathbb{R}$ can be constructed from $\mathbb{Q}$ by Dedekind cuts or by Cauchy sequences (§2.1).
📐 Theorem 1.20 — The $\varepsilon$-characterisation of the supremum
Let $S\subseteq\mathbb{R}$ be nonempty. (a) $S$ has at most one supremum. (b) A real number $u$ is $\sup S$ if and only if
  1. $s\le u$ for every $s\in S$, and
  2. for every $\varepsilon\gt0$ there is $s\in S$ with $s\gt u-\varepsilon$.
(c) If $u=\sup S$, there is a sequence $(s_n)$ in $S$ with $u-\tfrac1n\lt s_n\le u$, so $s_n\to u$. The infimum satisfies the mirror statements: $\ell=\inf S$ iff $\ell\le s$ for all $s\in S$ and for each $\varepsilon\gt0$ some $s\in S$ has $s\lt\ell+\varepsilon$.

Proof. (a) If $u$ and $u'$ are both suprema, each is an upper bound and each is $\le$ every upper bound, so $u\le u'$ and $u'\le u$. (b) Suppose $u=\sup S$ and $\varepsilon\gt0$. Then $u-\varepsilon\lt u$, so $u-\varepsilon$ is not an upper bound (it is smaller than the least one): some $s\in S$ has $s\gt u-\varepsilon$. Conversely, assume 1 and 2. Then $u$ is an upper bound. If $v$ were an upper bound with $v\lt u$, put $\varepsilon=u-v\gt0$; by 2 some $s\in S$ has $s\gt u-\varepsilon=v$, contradicting that $v$ is an upper bound. So $u\le v$ for every upper bound $v$. (c) Apply 2 with $\varepsilon=1/n$ for each $n$. ∎

📐 Theorem 1.21 — The greatest lower bound property
Every nonempty subset $S\subseteq\mathbb{R}$ that is bounded below has an infimum, and $\inf S=-\sup(-S)$, where $-S=\{-s:s\in S\}$.

Proof. If $\ell$ is a lower bound of $S$, then $-\ell$ is an upper bound of $-S$, which is nonempty; let $u=\sup(-S)$. For $s\in S$, $-s\le u$, so $s\ge-u$: $-u$ is a lower bound. If $\ell$ is any lower bound, $-\ell$ is an upper bound of $-S$, so $u\le-\ell$, i.e. $\ell\le-u$. Hence $-u$ is the greatest lower bound. ∎

📐 Theorem 1.22 — Algebra of suprema
Let $A,B\subseteq\mathbb{R}$ be nonempty and bounded above, and write $A+B=\{a+b\}$, $cA=\{ca\}$, $AB=\{ab\}$. Then
  1. $\sup(A+B)=\sup A+\sup B$;
  2. $\sup(cA)=c\sup A$ for $c\gt0$, and $\sup(cA)=c\inf A$ for $c\lt0$ (if $A$ is bounded below);
  3. $A\subseteq B\Rightarrow\sup A\le\sup B$ (and $\inf A\ge\inf B$ when the infima exist);
  4. $\sup(A\cup B)=\max(\sup A,\sup B)$;
  5. if $A,B\subseteq[0,\infty)$ then $\sup(AB)=\sup A\cdot\sup B$.

Proof. (1) Let $\alpha=\sup A$, $\beta=\sup B$. Every $a+b\le\alpha+\beta$, so $\alpha+\beta$ is an upper bound of $A+B$. Given $\varepsilon\gt0$, choose $a\in A$ with $a\gt\alpha-\varepsilon/2$ and $b\in B$ with $b\gt\beta-\varepsilon/2$ (Theorem 1.20); then $a+b\gt\alpha+\beta-\varepsilon$. By Theorem 1.20, $\sup(A+B)=\alpha+\beta$. (2) For $c\gt0$, $x\le u\iff cx\le cu$, so the upper bounds of $cA$ are exactly $c$ times those of $A$, and the least ones correspond. For $c\lt0$, multiplication by $c$ reverses the order and turns lower bounds of $A$ into upper bounds of $cA$. (3) Every upper bound of $B$ is an upper bound of $A$. (4) $m=\max(\sup A,\sup B)$ bounds $A\cup B$; and any upper bound of $A\cup B$ bounds both $A$ and $B$, so it is $\ge\sup A$ and $\ge\sup B$. (5) is Example 3(c). ∎

🧭 Problem-solving strategy — computing $\sup S$ and $\inf S$
  1. Explore. Write out the first terms; split by parity or by the sign of $(-1)^n$; note monotone pieces and their limits. Make a guess $u$.
  2. Bound. Prove $s\le u$ for every $s\in S$ (an inequality, often a one-line estimate).
  3. Approach. For a given $\varepsilon\gt0$ produce an explicit $s\in S$ with $s\gt u-\varepsilon$ — usually by the Archimedean property ("choose $n\gt1/\varepsilon$"). If some $s$ equals $u$, step 3 is automatic and $u=\max S$.
  4. Attained? Decide whether $u\in S$. For multi-parameter sets ($\{1/m-1/n\}$) treat each parameter separately: push one to its best value and let the other run off to infinity.
✏️ Example 2 — Two suprema, one attained
Find the supremum and infimum of (a) $S=\left\{\dfrac1n-\dfrac1m: m,n\in\mathbb{N}\right\}$; (b) $T=\left\{1+\dfrac{(-1)^n}{n}: n\in\mathbb{N}\right\}$. Which of them are attained?
(a) Bound $\tfrac1n\le1$ and $\tfrac1m\gt0$, so $\tfrac1n-\tfrac1m\lt1$; symmetrically $\tfrac1n-\tfrac1m\gt-1$. So $-1$ and $1$ are a lower and an upper bound, and neither value is attained (the inequalities are strict).
(a) Approach Given $\varepsilon\gt0$, by the Archimedean property choose $m$ with $\tfrac1m\lt\varepsilon$; with $n=1$, $1-\tfrac1m\gt1-\varepsilon$. So $\sup S=1$ (Theorem 1.20). Exchanging the roles of $m$ and $n$: $\tfrac1n-1\lt-1+\varepsilon$ for $\tfrac1n\lt\varepsilon$, so $\inf S=-1$. Neither is attained.
(b) Explore The terms are $0,\ \tfrac32,\ \tfrac23,\ \tfrac54,\ \tfrac45,\ \tfrac76,\dots$: even $n$ give $1+\tfrac1n\in(1,\tfrac32]$, decreasing; odd $n$ give $1-\tfrac1n\in[0,1)$, increasing.
(b) Answer Every term is $\le\tfrac32$ (even terms are $\le\tfrac32$, odd terms are $\lt1$) and $\tfrac32\in T$ ($n=2$): $\sup T=\max T=\tfrac32$. Every term is $\ge0$ and $0\in T$ ($n=1$): $\inf T=\min T=0$. Note that the "limit" $1$ of the terms is neither the sup nor the inf.
Interactive 1 — The $\varepsilon$-Band Below a Supremum

The dots are the terms $s_n$ of a set $S=\{s_n\}$. The dashed purple line is $u=\sup S$ (or $\inf S$), and the amber band is $(u-\varepsilon,u]$. Theorem 1.20 says: no dot lies above $u$, and for every $\varepsilon\gt0$ some dot lies in the band. Press Play to shrink $\varepsilon$ and watch the first witness $n$ move — or stay put when the supremum is attained.

bound u—
attained?—
ε—
first witness n—
sₙ—
✏️ Example 3 — Prove or disprove
For nonempty bounded sets $A,B\subseteq\mathbb{R}$, prove or disprove: (a) $\sup(AB)=\sup A\cdot\sup B$; (b) if $A\cap B\ne\varnothing$, $\sup(A\cap B)=\min(\sup A,\sup B)$; (c) if $A,B\subseteq[0,\infty)$, $\sup(AB)=\sup A\cdot\sup B$.
(a) False $A=B=\{-2,1\}$: $AB=\{4,-2,1\}$, so $\sup(AB)=4$, while $\sup A\cdot\sup B=1$. Negative numbers with large absolute value multiply to large positive numbers.
(b) False $A=\{1,3\}$, $B=\{1,2\}$: $A\cap B=\{1\}$, $\sup=1$, but $\min(3,2)=2$. Only the inequality $\sup(A\cap B)\le\min(\sup A,\sup B)$ holds (Theorem 1.22(3) twice).
(c) True Let $\alpha=\sup A$, $\beta=\sup B$. For $a\in A$, $b\in B$: $0\le a\le\alpha$, $0\le b\le\beta$ give $ab\le\alpha\beta$. If $\alpha=0$ then $A=\{0\}$ (its elements are $\ge0$ and $\le0$), so $AB=\{0\}$ and both sides are $0$; similarly if $\beta=0$. Otherwise $\alpha,\beta\gt0$; given $0\lt\varepsilon\lt\min(\alpha,\beta)$ pick $a\gt\alpha-\varepsilon$, $b\gt\beta-\varepsilon$ (both then positive). Then $$ab\gt(\alpha-\varepsilon)(\beta-\varepsilon)=\alpha\beta-\varepsilon(\alpha+\beta-\varepsilon)\gt\alpha\beta-\varepsilon(\alpha+\beta).$$ Since $\varepsilon(\alpha+\beta)$ can be made smaller than any $\eta\gt0$, Theorem 1.20 gives $\sup(AB)=\alpha\beta$.
Lesson Sums behave perfectly under $\sup$; products only for non-negative sets; differences become $\sup(A-B)=\sup A-\inf B$.

The Archimedean Property and Density

📐 Theorem 1.23 — The Archimedean property
(a) $\mathbb{N}$ is not bounded above in $\mathbb{R}$. (b) For every $x\gt0$ and every $y\in\mathbb{R}$ there is $n\in\mathbb{N}$ with $nx\gt y$. (c) For every $\varepsilon\gt0$ there is $n\in\mathbb{N}$ with $\tfrac1n\lt\varepsilon$; hence $\inf\{\tfrac1n:n\in\mathbb{N}\}=0$. (d) For every $x\in\mathbb{R}$ there is a unique integer $m=\lfloor x\rfloor$ with $m\le x\lt m+1$.

Proof. (a) If $\mathbb{N}$ were bounded above, completeness would give $u=\sup\mathbb{N}$. Then $u-1$ is not an upper bound, so some $n\in\mathbb{N}$ has $n\gt u-1$; but then $n+1\in\mathbb{N}$ and $n+1\gt u$, a contradiction. (b) By (a) choose $n\gt y/x$. (c) Take $x=\varepsilon$, $y=1$ in (b). Then $0$ is a lower bound of $\{1/n\}$ and no $\varepsilon\gt0$ is one. (d) By (a) there are integers $k\gt-x$ and so $-k\lt x$: the set $Z_x=\{k\in\mathbb{Z}:k\le x\}$ is nonempty and bounded above by $x$. Let $s=\sup Z_x$; some $m\in Z_x$ has $m\gt s-1$, and every integer $k\gt m$ satisfies $k\ge m+1\gt s$, so $k\notin Z_x$. Thus $m=\max Z_x$: $m\le x$ and $m+1\notin Z_x$, i.e. $x\lt m+1$. If $m,m'$ both work then $|m-m'|\lt1$, so $m=m'$. ∎

Notice that the proof of (a) used only completeness. So every complete ordered field is Archimedean — and an ordered field that is not Archimedean cannot be complete (Example 5).

📐 Theorem 1.24 — Existence of $\sqrt2$; $\mathbb{Q}$ is not complete
There is a unique positive real number $x$ with $x^2=2$, and $x$ is irrational. More generally, for every $a\gt0$ and $n\in\mathbb{N}$ there is a unique $y\gt0$ with $y^n=a$. The set $\{t\in\mathbb{Q}:t\ge0,\ t^2\lt2\}$ is nonempty and bounded above but has no supremum in $\mathbb{Q}$; so $\mathbb{Q}$ is an ordered field that is not complete.

Proof. Let $S=\{t\in\mathbb{R}:t\ge0,\ t^2\lt2\}$. It contains $1$, and $2$ is an upper bound (if $t\gt2$ then $t^2\gt4$). Let $x=\sup S$; then $1\le x\le2$. We show that both $x^2\lt2$ and $x^2\gt2$ are impossible.

If $x^2\lt2$: for $0\lt h\le1$, $(x+h)^2=x^2+h(2x+h)\le x^2+5h$ (as $2x+h\le5$). By the Archimedean property choose $n$ with $\tfrac1n\lt\tfrac{2-x^2}{5}$ and put $h=\tfrac1n$. Then $(x+h)^2\lt2$, so $x+h\in S$ — contradicting that $x$ is an upper bound.

If $x^2\gt2$: for $0\lt h\lt x$, $(x-h)^2=x^2-2xh+h^2\gt x^2-2xh\ge x^2-4h$. Choose $n\ge2$ with $\tfrac1n\lt\tfrac{x^2-2}{4}$ and $h=\tfrac1n$ (then $h\le\tfrac12\lt x$). Then $(x-h)^2\gt2$; for every $t\in S$, $t^2\lt2\lt(x-h)^2$ with $t\ge0$ and $x-h\gt0$, so $t\lt x-h$. Thus $x-h$ is a smaller upper bound — a contradiction.

Hence $x^2=2$. If also $y\gt0$, $y^2=2$ and $y\ne x$, say $y\lt x$, then $y^2\lt x^2$ — impossible; so $x$ is unique. If $x=p/q$ in lowest terms, $p^2=2q^2$ makes $p$ even, $p=2k$, so $q^2=2k^2$ makes $q$ even — contradicting lowest terms. For $n$-th roots the same argument works with the binomial estimate $(x\pm h)^n=x^n\pm h\cdot(\text{bounded})$. Finally, the whole argument above can be run inside $\mathbb{Q}$ (the $h=1/n$ are rational): a rational supremum $r$ of $\{t\in\mathbb{Q}:t\ge0,\ t^2\lt2\}$ would satisfy $r^2=2$, which is impossible. ∎

📐 Theorem 1.25 — Density of $\mathbb{Q}$ and of $\mathbb{R}\setminus\mathbb{Q}$
If $a\lt b$ are real numbers, there is a rational $r$ and an irrational $t$ with $a\lt r\lt b$ and $a\lt t\lt b$. Consequently every open interval contains infinitely many rationals and infinitely many irrationals; we say $\mathbb{Q}$ and $\mathbb{R}\setminus\mathbb{Q}$ are dense in $\mathbb{R}$.

Proof. By the Archimedean property choose $n\in\mathbb{N}$ with $\tfrac1n\lt b-a$, i.e. $nb-na\gt1$. Let $m=\lfloor na\rfloor+1$. Then $m-1\le na\lt m$, so $$na\lt m\le na+1\lt nb,$$ and $r=m/n$ satisfies $a\lt r\lt b$. For the irrational: apply this to $\tfrac{a}{\sqrt2}\lt\tfrac{b}{\sqrt2}$ to get a rational $r$ between them; if $r=0$ (possible only when $a\lt0\lt b$) apply it again to $0\lt\tfrac{b}{\sqrt2}$ to get a rational $r\ne0$. Then $t=r\sqrt2$ lies in $(a,b)$, and $t$ is irrational because otherwise $\sqrt2=t/r$ would be rational. Applying the result to $(a,b)$, then to $(a,r)$, $(a,r')$, … gives infinitely many. ∎

✏️ Example 4 — Producing a rational and an irrational
(a) Use the method of Theorem 1.25 to find a rational number strictly between $\sqrt2$ and $\sqrt2+10^{-3}$. (b) Find an irrational number in $(0,10^{-6})$. (c) Show that between any two irrational numbers there is a rational number.
(a) Choose $n$ We need $\tfrac1n\lt b-a=10^{-3}$, so $n=1001$ works.
Choose $m$ $na=1001\sqrt2=1415.627\ldots$, so $m=\lfloor na\rfloor+1=1416$ and $$r=\frac{1416}{1001}=1.414585\ldots$$ Check: $\sqrt2=1.414213\ldots\lt1.414585\ldots\lt1.415213\ldots=\sqrt2+10^{-3}$. ✓
(b) $t=\dfrac{\sqrt2}{2\cdot10^{6}}=7.07\ldots\times10^{-7}$ is irrational (a nonzero rational multiple of $\sqrt2$) and lies in $(0,10^{-6})$.
(c) Theorem 1.25 applies to any real $a\lt b$, rational or not. Density says nothing about the nature of the end points.
✏️ Example 5 — An ordered field that is not Archimedean
Let $\mathbb{R}(t)$ be the field of rational functions $p(t)/q(t)$ with real coefficients ($q\ne0$). Call $p/q$ positive if the leading coefficients of $p$ and $q$ have the same sign. Show that this makes $\mathbb{R}(t)$ an ordered field in which $t\gt n$ for every $n\in\mathbb{N}$. Deduce that $\mathbb{R}(t)$ is neither Archimedean nor complete.
Well defined Write $\operatorname{lc}$ for the leading coefficient. $f=p/q$ is positive iff $\operatorname{lc}(p)\operatorname{lc}(q)\gt0$; replacing $(p,q)$ by $(pr,qr)$ multiplies this by $\operatorname{lc}(r)^2\gt0$, so the definition does not depend on the representation. Intuitively, $f\gt0$ means "$f(t)\gt0$ for all sufficiently large real $t$".
(O1), (O2) For $f\ne0$ exactly one of $f$, $-f$ is positive. Products: $\operatorname{lc}$ is multiplicative, so positive times positive is positive. Sums: write $f=p/q$, $g=r/s$ with $q,s$ having positive leading coefficients (multiply top and bottom by $-1$ if needed); then $p$, $r$ have positive leading coefficients, $f+g=(ps+rq)/(qs)$, and $ps+rq$ has positive leading coefficient (the top-degree terms cannot cancel because both are positive). So $P$ satisfies the axioms.
$t$ is infinite $t-n$ has leading coefficient $1\gt0$, so $t\gt n$ for every $n$: $\mathbb{N}$ is bounded above by $t$, and the Archimedean property fails. Equivalently, $0\lt\tfrac1t\lt\tfrac1n$ for all $n$: $\tfrac1t$ is a positive "infinitesimal".
Not complete $\mathbb{N}$ is nonempty and bounded above in $\mathbb{R}(t)$; if it had a supremum, the proof of Theorem 1.23(a) would give a contradiction. So no supremum exists. (Directly: if $u$ were $\sup\mathbb{N}$, then $u-1$ would also be an upper bound.)

Nested Intervals, Extended Reals and Expansions

📐 Theorem 1.26 — Nested interval property
Let $I_n=[a_n,b_n]$ $(a_n\le b_n)$ be closed bounded intervals with $I_1\supseteq I_2\supseteq I_3\supseteq\cdots$. Then $\bigcap_{n\ge1}I_n=[\alpha,\beta]\ne\varnothing$, where $\alpha=\sup_n a_n$ and $\beta=\inf_n b_n$. If moreover $\inf_n(b_n-a_n)=0$, the intersection consists of exactly one point.

Proof. Nesting means $a_1\le a_2\le\cdots$ and $b_1\ge b_2\ge\cdots$. For any $m,n$ let $k=\max(m,n)$; then $a_m\le a_k\le b_k\le b_n$. So every $b_n$ is an upper bound of $\{a_m\}$, and $\alpha=\sup a_m$ exists with $\alpha\le b_n$ for all $n$; hence $\alpha$ is a lower bound of $\{b_n\}$ and $\alpha\le\beta=\inf b_n$. If $\alpha\le x\le\beta$ then $a_n\le\alpha\le x\le\beta\le b_n$ for every $n$, so $x\in\bigcap I_n$. Conversely, if $x\in\bigcap I_n$ then $a_n\le x\le b_n$ for all $n$, so $\alpha\le x\le\beta$. Finally $0\le\beta-\alpha\le b_n-a_n$ for every $n$; if the infimum of the lengths is $0$ then $\beta-\alpha\le0$ and $\alpha=\beta$. ∎

All three hypotheses are needed: the open intervals $(0,\tfrac1n)$ and the unbounded closed intervals $[n,\infty)$ are nested with empty intersection, and in $\mathbb{Q}$ the nested rational intervals $[1.4,1.5]\supseteq[1.41,1.42]\supseteq[1.414,1.415]\supseteq\cdots$ (decimal truncations of $\sqrt2$) close in on a point that is not there.

📐 Theorem 1.27 — Nested intervals give completeness
Let $F$ be an Archimedean ordered field in which every nested sequence of closed bounded intervals has nonempty intersection. Then $F$ is complete. Thus, for Archimedean ordered fields, the least upper bound property and the nested interval property are equivalent.

Proof (by bisection). Let $S\subseteq F$ be nonempty and bounded above. Pick $s_0\in S$ and an upper bound $b_1$; put $a_1=s_0-1$, which is not an upper bound. Given $[a_n,b_n]$ with $a_n$ not an upper bound and $b_n$ an upper bound, let $c$ be the midpoint; set $[a_{n+1},b_{n+1}]=[a_n,c]$ if $c$ is an upper bound and $[c,b_n]$ otherwise. The intervals are nested, $b_n-a_n=(b_1-a_1)/2^{n-1}$, and since $2^{n-1}\ge n$ the Archimedean property makes these lengths eventually smaller than any $\varepsilon\gt0$. By hypothesis (and the argument of Theorem 1.26) there is exactly one point $x$ in every $[a_n,b_n]$.

$x$ is an upper bound: if some $s\in S$ had $s\gt x$, choose $n$ with $b_n-a_n\lt s-x$; then $b_n\le a_n+(b_n-a_n)\le x+(b_n-a_n)\lt s$, contradicting that $b_n$ is an upper bound. $x$ is the least one: if $v\lt x$ were an upper bound, choose $n$ with $b_n-a_n\lt x-v$; then $a_n\ge b_n-(b_n-a_n)\ge x-(b_n-a_n)\gt v$, so $a_n$ would be an upper bound — but no $a_n$ is. Hence $x=\sup S$. ∎

The Archimedean hypothesis cannot be dropped: there are non-Archimedean ordered fields (built in non-standard analysis) in which every nested sequence of closed bounded intervals has a common point, yet $\mathbb{N}$ is bounded with no supremum.

📐 Theorem 1.28 — $\mathbb{R}$ is uncountable (nested-interval proof)
For every sequence $(x_n)$ of real numbers there is a real number $y$ with $y\ne x_n$ for all $n$. Hence $\mathbb{R}$, and every interval $[a,b]$ with $a\lt b$, is uncountable.

Proof. Let $I_0=[a,b]$. Given a closed interval $I_{n-1}$ of positive length, divide it into three closed thirds. The point $x_n$ lies in at most two of them (two only if it is a common end point), so at least one closed third does not contain $x_n$; call it $I_n$. The intervals $I_n$ are nested, closed and bounded, so by Theorem 1.26 some $y$ lies in all of them. For each $n$, $y\in I_n$ and $x_n\notin I_n$, so $y\ne x_n$. ∎

Unlike the diagonal proof of §1.1, this one uses no expansions at all — only completeness. It also shows that the uncountability of $\mathbb{R}$ is a consequence of the completeness axiom: $\mathbb{Q}$ fails both.

✏️ Example 6 — Nested intervals in action
(a) Find $\bigcap_n I_n$ for $I_n=[-\tfrac1n,\tfrac1n]$, $J_n=(0,\tfrac1n)$, $K_n=[n,\infty)$, $L_n=[1-\tfrac1n,\,2+\tfrac1n]$. (b) Starting from $[1,2]$, perform six bisection steps to trap $\sqrt2$, keeping at each step the half whose end points have squares on opposite sides of $2$. Read off the first binary digits of $\sqrt2$.
(a) $\bigcap I_n=\{0\}$ (lengths $\to0$); $\bigcap J_n=\varnothing$ (a point $x\gt0$ is outside $J_n$ once $\tfrac1n\le x$ — open intervals); $\bigcap K_n=\varnothing$ (unbounded — no point is $\ge n$ for all $n$); $\bigcap L_n=[1,2]$ (lengths do not tend to $0$, so the intersection is a whole interval $[\sup a_n,\inf b_n]$).
(b) The table
$n$midpoint $c$$c^2$$I_n$digit
11.52.25 > 2[1, 1.5]0
21.251.5625 < 2[1.25, 1.5]1
31.3751.890625 < 2[1.375, 1.5]1
41.43752.06640625 > 2[1.375, 1.4375]0
51.406251.9775390625 < 2[1.40625, 1.4375]1
61.4218752.021728515625 > 2[1.40625, 1.421875]0
Digits "Left half" records the binary digit $0$ and "right half" the digit $1$, so $\sqrt2=1.011010\ldots_2$. The intervals have length $2^{-n}$ and, by Theorem 1.26, intersect in exactly one point — the number whose square is $2$ (Theorem 1.24).
Check $1.011010_2=1+\tfrac14+\tfrac18+\tfrac1{32}=1.40625$, the left end of $I_6$, and $1.40625\le\sqrt2=1.41421\ldots\le1.421875$. ✓
Interactive 2 — Nested Intervals Closing In: Binary and Decimal Digits

Each bar is a closed interval $I_k$ of length $b^{-k}$, cut into $b$ equal parts ($b=2$ or $10$). The part containing $x$ (red marker) becomes $I_{k+1}$, magnified onto the next bar, and its index is the next digit of $x$. The intervals are nested, closed, and their lengths shrink to $0$ — so by the nested interval property exactly one point survives in all of them. Watch which numbers produce repeating digits.

k—
expansion so far—
Iₖ—
length—
digits repeat?—
📘 Definition 1.12 — The extended real line

The extended real line is $\overline{\mathbb{R}}=\mathbb{R}\cup\{-\infty,+\infty\}$ with the order $-\infty\lt x\lt+\infty$ for every $x\in\mathbb{R}$. Arithmetic: $x+(\pm\infty)=\pm\infty$ for real $x$; $(+\infty)+(+\infty)=+\infty$; $x\cdot(\pm\infty)=\pm\infty$ for $x\gt0$ and $\mp\infty$ for $x\lt0$; $x/(\pm\infty)=0$. The expressions $\infty-\infty$ and $\pm\infty/\pm\infty$ are undefined; $0\cdot(\pm\infty)$ is undefined in limits, but set equal to $0$ by convention in measure and integration theory (Chapter 4).

We put $\sup S=+\infty$ if $S\subseteq\mathbb{R}$ is not bounded above, $\inf S=-\infty$ if it is not bounded below, and $\sup\varnothing=-\infty$, $\inf\varnothing=+\infty$. With these conventions every subset of $\overline{\mathbb{R}}$ has a supremum and an infimum in $\overline{\mathbb{R}}$. $\overline{\mathbb{R}}$ is not a field.

📐 Theorem 1.29 — Expansions in base $b$
Let $b\ge2$ be an integer and $x\in[0,1)$. Define the digits $d_k=\lfloor b^kx\rfloor-b\lfloor b^{k-1}x\rfloor\in\{0,1,\dots,b-1\}$. Then $$x=\sum_{k=1}^{\infty}\frac{d_k}{b^k}=\sup_n\sum_{k=1}^{n}\frac{d_k}{b^k},$$ and this expansion does not end in an infinite string of digits $b-1$. Conversely every digit sequence $(e_k)$ defines the point $\sum e_kb^{-k}\in[0,1]$. A number in $(0,1)$ has two expansions if and only if it is of the form $m/b^n$ (one expansion terminates, the other ends in $(b-1)$s); otherwise its expansion is unique. Finally, $x$ is rational if and only if its expansion is eventually periodic.

Proof (sketch with all key steps). Put $t_n=\lfloor b^nx\rfloor/b^n$. Since $d_k=b^k(t_k-t_{k-1})$, the sum telescopes: $t_n=\sum_{k\le n}d_kb^{-k}$. By definition of the floor, $t_n\le x\lt t_n+b^{-n}$. Since $b^{-n}\le\tfrac1n$, the Archimedean property gives, for each $\varepsilon\gt0$, an $n$ with $t_n\gt x-\varepsilon$; with $t_n\le x$ this is Theorem 1.20, so $x=\sup t_n$. The digit bound: for $y=b^{k-1}x$, $b\lfloor y\rfloor\le\lfloor by\rfloor\lt b\lfloor y\rfloor+b$. If $d_k=b-1$ for all $k\gt n$, then $x-t_n=\sum_{k\gt n}(b-1)b^{-k}=b^{-n}$, contradicting $x\lt t_n+b^{-n}$. Two different expansions $(e_k)\ne(f_k)$ of the same number first differing at place $n$ with $e_n\lt f_n$ force $e_n+1=f_n$, $e_k=b-1$ and $f_k=0$ for all $k\gt n$ (otherwise the tails cannot make up the difference $b^{-n}$); this is the case $x=m/b^n$. An eventually periodic expansion is a finite sum plus a geometric series, hence rational. Conversely, for $x=p/q$ the digits come from long division, and the remainders take at most $q$ values, so a remainder repeats and from then on the digits repeat. ∎

✏️ Example 7 — Expansions
(a) Find the binary expansion of $\tfrac1{10}$. (b) Write $0.1\overline{36}=0.13636\ldots$ as a fraction. (c) Give both binary expansions of $\tfrac38$. (d) Show that $0.4999\ldots=0.5$ using suprema.
(a) Double and take integer parts: $0.1\to0.2\to0.4\to0.8\to1.6\to1.2\to0.4\to\cdots$ The digits are $0,0,0,1,1,0,0,1,1,\dots$ and the remainder $0.4$ recurs after four steps: $\tfrac1{10}=0.0\overline{0011}_2$. One-tenth has no finite binary expansion — a fact with real consequences for computers (see the Research–Teaching Nexus).
(b) $x=0.1\overline{36}$: $1000x=136.\overline{36}$ and $10x=1.\overline{36}$, so $990x=135$ and $x=\dfrac{135}{990}=\dfrac{3}{22}$. Check: $3\div22=0.136363\ldots$ ✓
(c) $\tfrac38=\tfrac14+\tfrac18=0.011_2$, and also $0.0101111\ldots_2$ since $\sum_{k\ge4}2^{-k}=\tfrac18$. It is of the form $m/2^n$, so it has exactly two expansions.
(d) The partial sums $s_n=0.4\underbrace{9\cdots9}_{n-1}=0.5-10^{-n}$ satisfy $s_n\lt0.5$, and for each $\varepsilon\gt0$ some $s_n\gt0.5-\varepsilon$ (Archimedean). By Theorem 1.20, $\sup s_n=0.5$, and the infinite decimal means this supremum.
⚠️ Counterexamples and pitfalls
  1. $\mathbb{Q}$ is Archimedean but not complete: $\{t\in\mathbb{Q}:t^2\lt2\}$ is bounded above with no rational supremum.
  2. $\mathbb{R}(t)$ is ordered but neither Archimedean nor complete: $t\gt n$ for all $n$, and $\mathbb{N}$ has no supremum.
  3. Some fields cannot be ordered at all: $\mathbb{C}$ ($i^2=-1$ would be positive) and every finite field $\mathbb{Z}_p$ ($p\cdot1=0$).
  4. A supremum need not be a maximum: $\sup(0,1)=1\notin(0,1)$. A maximum exists iff the supremum belongs to the set.
  5. An upper bound is not the supremum: $2$ bounds $(0,1)$ above but is not least; always check the $\varepsilon$-condition.
  6. Strict inequalities do not survive suprema: $s_n\lt1$ for all $n$ gives only $\sup s_n\le1$ (take $s_n=1-\tfrac1n$).
  7. Products and differences: $\sup(AB)\ne\sup A\sup B$ for $A=B=\{-2,1\}$; $\sup(A-B)=\sup A-\inf B$, not $\sup A-\sup B$.
  8. Nested intervals need closed and bounded: $\bigcap(0,\tfrac1n)=\varnothing$ and $\bigcap[n,\infty)=\varnothing$; but nested open intervals may still meet, e.g. $\bigcap(-\tfrac1n,\tfrac1n)=\{0\}$.
  9. Dense is not uncountable, and dense sets need not be "big": $\mathbb{Q}$ and the dyadic rationals are countable and dense; $\{n\sqrt2:n\in\mathbb{Z}\}$ is infinite but not dense; $\{p/q:p,q\in\mathbb{N}\}$ is dense in $(0,\infty)$ but not in $\mathbb{R}$.
  10. Irrational + irrational can be rational: $\sqrt2+(-\sqrt2)=0$, $\sqrt2\cdot\sqrt2=2$; but rational $+$ irrational and (nonzero rational)$\times$irrational are always irrational.
  11. Two expansions: $0.4999\ldots=0.5$ and $0.0111\ldots_2=0.1_2$. Countably many numbers (the $m/b^n$) have two expansions; a "terminating in decimal" number such as $0.1$ may be non-terminating in binary.
  12. $\pm\infty$ are not real numbers: $\infty-\infty$ is undefined; "$\sup S=\infty$" is shorthand for "$S$ is not bounded above".
  13. Equality in the triangle inequality needs $xy\ge0$; $|x-y|\ge|x|-|y|$, not $|x|+|y|$.

Exercises

I
Ordered fields and absolute value.
  • 1 In an ordered field prove: if $0\lt a\lt b$ then $a^2\lt b^2$; and if $a\lt b$ then $a\lt\tfrac{a+b}{2}\lt b$.
  • 2 Prove $\big||x|-|y|\big|\le|x-y|$, and show that $|x+y|=|x|+|y|$ iff $xy\ge0$.
  • 3 Solve $|2x-3|\lt|x+1|$.
  • 4 Solve $|x-1|\,|x+1|\le3$.
  • 5 Prove that $\mathbb{C}$ admits no order making it an ordered field.
  • 6 Prove that the field $\mathbb{Z}_5$ cannot be ordered.
  • 7 Prove: if $x\le y+\varepsilon$ for every $\varepsilon\gt0$, then $x\le y$.
  • 8 Prove Bernoulli's inequality $(1+x)^n\ge1+nx$ for $x\ge-1$ and $n\in\mathbb{N}$, and use it to show that $\inf\{2^{-n}:n\in\mathbb{N}\}=0$.
1 $b^2-a^2=(b-a)(b+a)$, a product of two positives. $\tfrac{a+b}{2}-a=\tfrac{b-a}{2}\gt0$ and $b-\tfrac{a+b}2=\tfrac{b-a}2\gt0$, using $\tfrac12\gt0$ (Theorem 1.18(e), since $2=1+1\gt0$).
2 This is Theorem 1.19(c), (d); see its proof.
3 Both sides are $\ge0$, so the inequality is equivalent to $(2x-3)^2\lt(x+1)^2$, i.e. $3x^2-14x+8\lt0$, i.e. $(3x-2)(x-4)\lt0$: $x\in\left(\tfrac23,4\right)$.
4 $|x-1||x+1|=|x^2-1|\le3\iff-3\le x^2-1\le3\iff x^2\le4$ (the left inequality always holds): $x\in[-2,2]$.
5 In any ordered field nonzero squares are positive. $i^2=-1$ and $1^2=1$ would both be positive, so $0=1+(-1)$ would be positive, contradicting trichotomy.
6 In an ordered field $1\gt0$, so $1+1+1+1+1\gt0$; but in $\mathbb{Z}_5$ this sum is $0$.
7 If $x\gt y$, take $\varepsilon=\tfrac{x-y}{2}\gt0$: then $x\le y+\tfrac{x-y}2$ gives $\tfrac{x-y}{2}\le0$, a contradiction.
8 Induction: $(1+x)^{n+1}\ge(1+nx)(1+x)=1+(n+1)x+nx^2\ge1+(n+1)x$, using $1+x\ge0$. With $x=1$: $2^n\ge1+n\gt n$, so $0\lt2^{-n}\lt\tfrac1n$, and $\inf\tfrac1n=0$ (Theorem 1.23) gives $\inf2^{-n}=0$.
II
Supremum and infimum. In 1–6 find $\sup$ and $\inf$ (in $\mathbb{R}$) and say whether they are attained.
  • 1 $\left\{1-\dfrac{(-1)^n}{n}:n\in\mathbb{N}\right\}$
  • 2 $\left\{\dfrac{n}{2n+1}:n\in\mathbb{N}\right\}$
  • 3 $\{x\in\mathbb{R}:x^2-3x+2\lt0\}$
  • 4 $\left\{\dfrac mn:m,n\in\mathbb{N},\ m\lt n\right\}$
  • 5 $\left\{\dfrac1m+\dfrac{(-1)^n}{n}:m,n\in\mathbb{N}\right\}$
  • 6 $\{x\in\mathbb{Q}:x^3\lt2\}$ — is it bounded below? Does it have a supremum in $\mathbb{Q}$?
  • 7 Prove that $\sup(cA)=c\inf A$ for $c\lt0$ and $A$ nonempty and bounded below.
  • 8 Let $A$ be nonempty and bounded. Prove $\inf A\le\sup A$, with equality iff $A$ is a single point.
  • 9 For bounded $f,g:D\to\mathbb{R}$ prove $\sup_D(f+g)\le\sup_Df+\sup_Dg$, and give an example with strict inequality.
  • 10 Let $u=\sup S$ with $u\notin S$. Prove that $S$ contains a strictly increasing sequence converging to $u$, and hence that $S$ is infinite.
1 Terms $2,\tfrac12,\tfrac43,\tfrac34,\tfrac65,\dots$: odd $n$ give $1+\tfrac1n\le2$, even $n$ give $1-\tfrac1n\ge\tfrac12$. $\sup=\max=2$ ($n=1$), $\inf=\min=\tfrac12$ ($n=2$).
2 $\tfrac{n}{2n+1}=\tfrac12-\tfrac{1}{2(2n+1)}$ increases with $n$. $\inf=\min=\tfrac13$ ($n=1$); $\sup=\tfrac12$, not attained: given $\varepsilon$, $\tfrac1{2(2n+1)}\lt\varepsilon$ for large $n$.
3 $(x-1)(x-2)\lt0\iff1\lt x\lt2$. $\sup=2$, $\inf=1$, neither attained.
4 The set is $\mathbb{Q}\cap(0,1)$. $\sup=1$ (take $\tfrac{n-1}{n}$), $\inf=0$ (take $\tfrac1n$), neither attained.
5 $\tfrac1m\le1$ and $\tfrac{(-1)^n}{n}\le\tfrac12$, with both equalities at $m=1,n=2$: $\sup=\max=\tfrac32$. Also $\tfrac1m\gt0$ and $\tfrac{(-1)^n}n\ge-1$, so every element is $\gt-1$; with $n=1$ and $m$ large, $\tfrac1m-1\lt-1+\varepsilon$: $\inf=-1$, not attained.
6 Not bounded below (it contains every negative rational). Bounded above (by $2$). In $\mathbb{R}$, $\sup=\sqrt[3]2$, irrational; as for $\sqrt2$ (Theorem 1.24), no rational number is its least upper bound in $\mathbb{Q}$: a rational upper bound $r$ has $r^3\gt2$ (equality is impossible), and a slightly smaller rational is still an upper bound.
7 Let $\ell=\inf A$. For $a\in A$, $a\ge\ell$ and $c\lt0$ give $ca\le c\ell$: $c\ell$ bounds $cA$ above. Given $\varepsilon\gt0$, pick $a\lt\ell+\varepsilon/|c|$; then $ca\gt c\ell-\varepsilon$. By Theorem 1.20, $\sup(cA)=c\ell$.
8 For any $a\in A$, $\inf A\le a\le\sup A$. If $\inf A=\sup A=u$, every $a$ satisfies $u\le a\le u$, so $A=\{u\}$; the converse is clear.
9 For $x\in D$, $f(x)+g(x)\le\sup f+\sup g$; take the supremum over $x$. Example: $f(x)=x$, $g(x)=1-x$ on $[0,1]$: $\sup(f+g)=1\lt1+1=2$. The two suprema are attained at different points.
10 Choose $s_1\in S$ with $s_1\gt u-1$. Having $s_n\lt u$ (as $u\notin S$), choose $s_{n+1}\in S$ with $s_{n+1}\gt\max\!\left(s_n,\,u-\tfrac1{n+1}\right)$ — possible because this maximum is $\lt u$. Then $(s_n)$ is strictly increasing with $u-\tfrac1n\lt s_n\lt u$, so $s_n\to u$, and its terms are distinct.
III
Archimedean property, density and roots.
  • 1 Prove from the Archimedean property that $\inf\left\{\tfrac{1}{n}:n\in\mathbb{N}\right\}=0$ and $\inf\left\{\tfrac{n}{n^2+1}:n\in\mathbb{N}\right\}=0$.
  • 2 Use the method of Theorem 1.25 to find a rational number in $(\sqrt3,\sqrt3+0.01)$.
  • 3 Show that $a+\dfrac{b-a}{\sqrt2}$ is an irrational number between the rationals $a\lt b$.
  • 4 Prove that every open interval contains infinitely many rationals.
  • 5 Prove that $\sqrt3$ and $\sqrt2+\sqrt3$ are irrational.
  • 6 Show that $\{t\ge0:t^2\lt3\}$ has supremum $x$ with $x^2=3$ by adapting the proof of Theorem 1.24.
  • 7 Let $G\ne\{0\}$ be an additive subgroup of $\mathbb{R}$ and $\alpha=\inf(G\cap(0,\infty))$. Prove: if $\alpha\gt0$ then $G=\alpha\mathbb{Z}$; if $\alpha=0$ then $G$ is dense. Deduce that $\{m+n\sqrt2:m,n\in\mathbb{Z}\}$ is dense in $\mathbb{R}$.
  • 8 Show that in the ordered field $\mathbb{Q}$ the set $\{t\in\mathbb{Q}:t\gt0,\ t^2\gt2\}$ has no infimum.
1 $0$ is a lower bound; for $\varepsilon\gt0$ choose $n\gt1/\varepsilon$, then $\tfrac1n\lt\varepsilon$. For the second set, $0\lt\tfrac{n}{n^2+1}\lt\tfrac1n$, so the same $n$ works.
2 Need $\tfrac1n\lt0.01$: $n=101$. $101\sqrt3=174.93\ldots$, $m=175$, $r=\tfrac{175}{101}=1.73267\ldots$, and $1.73205\ldots\lt1.73267\ldots\lt1.74205\ldots$ ✓
3 $0\lt\tfrac{1}{\sqrt2}\lt1$, so the number lies strictly between $a$ and $b$. If it were rational, $\sqrt2=\tfrac{b-a}{t-a}$ would be rational.
4 Find $r_1\in(a,b)$, then $r_2\in(a,r_1)$, $r_3\in(a,r_2)$, …: a strictly decreasing sequence of rationals in $(a,b)$.
5 If $\sqrt3=p/q$ in lowest terms, $p^2=3q^2$; $3\mid p^2$ forces $3\mid p$ ($3$ is prime), then $3\mid q$ — contradiction. If $\sqrt2+\sqrt3=r\in\mathbb{Q}$, then $3=(r-\sqrt2)^2=r^2-2r\sqrt2+2$ gives $\sqrt2=\tfrac{r^2-1}{2r}\in\mathbb{Q}$, a contradiction.
6 $x=\sup$ exists ($1\in S$, $2$ is an upper bound) and $1\le x\le2$. If $x^2\lt3$: $(x+h)^2\le x^2+5h\lt3$ for $h=\tfrac1n\lt\tfrac{3-x^2}5$. If $x^2\gt3$: $(x-h)^2\gt x^2-4h\gt3$ for $h=\tfrac1n\lt\tfrac{x^2-3}{4}$, $n\ge2$, making $x-h$ a smaller upper bound. So $x^2=3$.
7 If $\alpha\gt0$ and $\alpha\notin G$, choose $g\in G$ with $\alpha\lt g\lt2\alpha$ and then $h\in G$ with $\alpha\lt h\lt g$ (both by the $\varepsilon$-characterisation of the infimum); then $g-h\in G$ and $0\lt g-h\lt\alpha$ — contradiction. So $\alpha\in G$; for $g\in G$, $g-\lfloor g/\alpha\rfloor\alpha\in G\cap[0,\alpha)$ must be $0$: $G=\alpha\mathbb{Z}$. If $\alpha=0$, given $a\lt b$ pick $g\in G$ with $0\lt g\lt b-a$; some multiple $kg$ lands in $(a,b)$ (take $k=\lfloor a/g\rfloor+1$). For $G=\mathbb{Z}+\sqrt2\mathbb{Z}$: if $G=\alpha\mathbb{Z}$ then $1=k\alpha$ and $\sqrt2=l\alpha$ would make $\sqrt2=l/k$ rational. So $\alpha=0$ and $G$ is dense.
8 Suppose $r\in\mathbb{Q}$ were the infimum. Every element of the set exceeds $1$ (if $0\lt t\le1$ then $t^2\le1$), so $r\ge1$; and $r^2\ne2$ because $r$ is rational. If $r^2\gt2$, then $r$ is in the set and, as in Theorem 1.24, $r-\tfrac1n$ is still in it for large $n$ — so $r$ is not a lower bound. If $r^2\lt2$, then $r+\tfrac1n$ is still a lower bound for large $n$ — so $r$ is not the greatest one.
IV
Nested intervals, extended reals and expansions.
  • 1 Find the intersection of each nested family: (a) $[0,\tfrac1n]$; (b) $(0,\tfrac1n]$; (c) $[-\tfrac1n,1+\tfrac1n]$; (d) $(1-\tfrac1n,1+\tfrac1n)$.
  • 2 Find the binary expansions of $\tfrac58$ and $\tfrac15$ and the decimal expansion of $\tfrac17$.
  • 3 Write $0.2\overline{45}=0.24545\ldots$ as a fraction in lowest terms.
  • 4 Prove that $0.999\ldots=1$.
  • 5 In $\overline{\mathbb{R}}$ find $\sup$ and $\inf$ of $\varnothing$, $\mathbb{N}$ and $\{-n:n\in\mathbb{N}\}$. Which of $\infty+\infty$, $\infty-\infty$, $(-\infty)(-3)$, $5/\infty$ are defined?
  • 6 Let $(x_n)$ be any sequence in $[0,1]$. Explain how to choose nested closed intervals $I_n\subseteq[0,1]$ with $x_n\notin I_n$, and conclude that $[0,1]$ is uncountable.
  • 7 Applied. An engineer locates the root of $x^3+x-1=0$ in $[0,1]$ by bisection. How many bisections guarantee that the root lies in an interval of length less than $10^{-6}$? Give the root to six decimal places.
1 (a) $\{0\}$; (b) $\varnothing$ (not closed: $0$ is excluded and every $x\gt0$ eventually drops out); (c) $[0,1]$; (d) $\{1\}$ — open intervals, yet the intersection is nonempty (the theorem gives a sufficient condition, not a necessary one).
2 $\tfrac58=\tfrac12+\tfrac18=0.101_2$. For $\tfrac15$ doubling gives $0.4,0.8,1.6\to0.6,1.2\to0.2$, digits $0,0,1,1$ and repeat: $\tfrac15=0.\overline{0011}_2$. $\tfrac17=0.\overline{142857}$.
3 $x=0.2\overline{45}$: $1000x-10x=245.\overline{45}-2.\overline{45}=243$, so $x=\tfrac{243}{990}=\tfrac{27}{110}$. Check $27\div110=0.24545\ldots$ ✓
4 The partial sums are $1-10^{-n}$; their supremum is $1$ by the Archimedean property (as in Example 7(d)).
5 $\sup\varnothing=-\infty$, $\inf\varnothing=+\infty$; $\sup\mathbb{N}=+\infty$, $\inf\mathbb{N}=1$; $\sup\{-n\}=-1$, $\inf\{-n\}=-\infty$. $\infty+\infty=\infty$, $(-\infty)(-3)=+\infty$, $5/\infty=0$ are defined; $\infty-\infty$ is not.
6 As in Theorem 1.28: split $I_{n-1}$ into three closed thirds and let $I_n$ be one that misses $x_n$. A point $y\in\bigcap I_n$ (Theorem 1.26) differs from every $x_n$, so no sequence lists all of $[0,1]$.
7 $f(x)=x^3+x-1$ is increasing with $f(0)=-1\lt0\lt1=f(1)$. After $n$ bisections the interval has length $2^{-n}$; $2^{-19}\approx1.9\times10^{-6}$ and $2^{-20}\approx9.5\times10^{-7}$, so $n=20$ bisections. The root is $x\approx0.682328$.
🎯 Exam corner CSIR Part BCSIR Part CGATE MA

E1. CSIR Part B GATE MA Let $S=\left\{\dfrac1m+\dfrac1n:m,n\in\mathbb{N}\right\}$. Then (A) $\sup S=2$ and $\inf S=0$, both attained; (B) $\sup S=2$ is attained and $\inf S=0$ is not attained; (C) $\sup S=2$ and $\inf S=0$, neither attained; (D) $\sup S=1$ and $\inf S=0$.

Idea Push each parameter to its extreme separately.
Work $\tfrac1m+\tfrac1n\le2$ with equality at $m=n=1$: $\max S=2$. Every element is $\gt0$, and $\tfrac2n\lt\varepsilon$ for $n\gt2/\varepsilon$ (take $m=n$): $\inf S=0$, not attained.
Answer (B).

E2. CSIR Part B Let $F$ be any ordered field. Which statement is necessarily true? (A) $F$ is Archimedean; (B) $F$ contains a subfield order-isomorphic to $\mathbb{Q}$; (C) every nonempty subset of $F$ that is bounded above has a supremum in $F$; (D) every positive element of $F$ has a square root in $F$.

Eliminate (A) fails in $\mathbb{R}(t)$ (Example 5); (C) fails in $\mathbb{Q}$; (D) fails in $\mathbb{Q}$ ($2$ has no rational square root).
Why (B) $n\cdot1\gt0$ for all $n\in\mathbb{N}$ (Theorem 1.18(f)), so $F$ has characteristic $0$ and $\tfrac mn\mapsto(m\cdot1)(n\cdot1)^{-1}$ is an order-preserving embedding of $\mathbb{Q}$.
Answer (B).

E3. CSIR Part B Let $a_n\lt b_n$ for all $n$, with $(a_n)$ strictly increasing and $(b_n)$ strictly decreasing, and let $I_n=(a_n,b_n)$ (open intervals). Then $\bigcap_nI_n$ is (A) always empty; (B) always nonempty; (C) nonempty if and only if $b_n-a_n\to0$; (D) always a single point.

Idea Strict monotonicity pushes $\sup a_n$ strictly inside every open interval.
Work As in Theorem 1.26, $\alpha=\sup a_n\le\beta=\inf b_n$. Since $a_n\lt a_{n+1}\le\alpha$ and $\beta\le b_{n+1}\lt b_n$, we get $a_n\lt\alpha\le\beta\lt b_n$, so $\alpha\in I_n$ for all $n$. (C), (D) fail for $I_n=(-1-\tfrac1n,\,1+\tfrac1n)$, whose intersection is $[-1,1]$.
Answer (B).

E4. CSIR Part C Let $A,B$ be nonempty bounded subsets of $\mathbb{R}$. Which of the following are true? (One or more options may be correct.) (A) $\sup(A\cup B)=\max(\sup A,\sup B)$; (B) if $A\cap B\ne\varnothing$ then $\sup(A\cap B)\le\min(\sup A,\sup B)$; (C) $\sup(A-B)=\sup A-\sup B$, where $A-B=\{a-b\}$; (D) $\inf(AB)=\inf A\cdot\inf B$.

(A) True Theorem 1.22(4).
(B) True $A\cap B\subseteq A$ and $\subseteq B$, so its supremum is at most each of $\sup A$, $\sup B$ (Theorem 1.22(3)).
(C) False $\sup(A-B)=\sup A+\sup(-B)=\sup A-\inf B$. Counterexample: $A=\{0\}$, $B=\{0,1\}$: $A-B=\{0,-1\}$ has supremum $0$, but $\sup A-\sup B=-1$.
(D) False $A=B=\{-1,2\}$: $AB=\{1,-2,4\}$, $\inf(AB)=-2$, but $\inf A\cdot\inf B=1$.
Answer (A) and (B).

E5. CSIR Part C Which of the following subsets of $\mathbb{R}$ are dense in $\mathbb{R}$? (A) $\{m/2^n:m\in\mathbb{Z},\ n\in\mathbb{N}\}$; (B) $\{m+n\sqrt2:m,n\in\mathbb{Z}\}$; (C) $\{n\sqrt2:n\in\mathbb{Z}\}$; (D) $\{p/q:p\in\mathbb{N},\ q\text{ prime}\}$.

(A) Dense Given $a\lt b$, choose $n$ with $2^{-n}\lt b-a$ (Exercise I.8); the multiples $m/2^n$ are spaced $2^{-n}$ apart, so one lands in $(a,b)$ (take $m=\lfloor2^na\rfloor+1$).
(B) Dense It is a subgroup of $(\mathbb{R},+)$ that is not cyclic because $\sqrt2$ is irrational; by Exercise III.7 it is dense.
(C) Not dense Its points are spaced $\sqrt2$ apart; $(0,1)$ contains none of them.
(D) Not dense All its elements are positive, so $(-2,-1)$ contains none. (It is dense in $(0,\infty)$: choose a prime $q\gt1/(b-a)$.)
Answer (A) and (B).

E6. CSIR Part C Let $x\in(0,1)$. Which of the following are true? (A) $x$ has at most two decimal expansions. (B) If $x$ has two distinct binary expansions, then $x$ is rational. (C) The set of $x\in(0,1)$ that have two distinct decimal expansions is uncountable. (D) The binary expansion of $x$ terminates if and only if its decimal expansion terminates.

(A) True By Theorem 1.29, two different expansions of the same number must be "$\ldots d000\ldots$" and "$\ldots(d-1)999\ldots$", determined by $x$; there is no room for a third.
(B) True Two binary expansions occur only for $x=m/2^n$, which is rational.
(C) False These are exactly the numbers $m/10^n$ in $(0,1)$ — a countable set.
(D) False A terminating binary $m/2^n=m5^n/10^n$ terminates in decimal, but $0.1$ terminates in decimal and is $0.0\overline{0011}_2$ in binary.
Answer (A) and (B).
🧑‍🏫 For the teacher

Suggested plan — 2 lectures × 60 min (a third for weaker groups).

  1. Ordered fields and completeness. Axioms (Definition 1.8) quickly; Theorem 1.18 with the two traps ($\mathbb{C}$, $\mathbb{Z}_p$); triangle inequality and Example 1. Supremum, the completeness axiom, and Theorem 1.20 proved in full — then project Interactive 1, first with $1-1/n$ (witness moves) and then with $1+(-1)^n/n$ (witness frozen at $n=2$: attained). Example 2 on the board; Example 3(a) as a quick poll.
  2. Consequences of completeness. Archimedean property (stress that the proof uses only completeness), Theorem 1.24 in full (students should be able to reproduce it), density with Example 4, the non-Archimedean field of Example 5 as a five-minute aside. Nested intervals (Theorem 1.26, Example 6) with Interactive 2 in base 2 for $\sqrt2$, then base 2 for $1/10$ to show periodicity. Finish with Theorem 1.28 and E3 or E5 from the Exam corner.

Discussion questions. (1) Is there a smallest positive real number? A smallest positive rational? What about in $\mathbb{R}(t)$? (2) Which results of this section use completeness, which only the Archimedean property, and which hold in every ordered field? (3) "$\sqrt2=1.41421356\ldots$" — is this a definition of $\sqrt2$? What would we need to know first?

Misconceptions to watch for. Confusing $\sup$ with $\max$; choosing the witness before $\varepsilon$ (quantifier order); believing nested open intervals always have empty intersection, or that the nested interval property needs lengths $\to0$; dividing an inequality by a quantity of unknown sign; thinking $0.999\ldots$ is "just less than" $1$; treating $\pm\infty$ as numbers.

Suggested homework. I(3, 4, 5, 8), II(1, 2, 5, 7, 9, 10), III(2, 5, 7), IV(1, 3, 6, 7); Exam corner E1, E4, E6.

🔬 Research–Teaching NexusResearch-tutored · optional, about 40 minutes

Research question. Which axioms of a complete ordered field survive inside a computer's floating-point numbers — and what happens when one of them fails?

Why it matters. Engineering simulations, weather and climate models, and financial software all compute with IEEE-754 floating-point numbers, a finite set pretending to be $\mathbb{R}$. In 1991 a Patriot air-defence battery in Dhahran, Saudi Arabia, failed to intercept an incoming missile partly because its clock counted time in tenths of a second, and $\tfrac1{10}$ has a non-terminating binary expansion (Example 7): the value stored in a 24-bit register was slightly too small, and the error grew with every tick.

What to do.

  1. In Python, compare (0.1+0.2)+0.3 with 0.1+(0.2+0.3). Which field axiom fails?
  2. Find the largest $k$ for which 1 + 2**-k != 1. Interpret $2^{-k}$ as a kind of "smallest $\varepsilon$" — which property of $\mathbb{R}$ (Theorem 1.19(e), Theorem 1.23) has no analogue here?
  3. Truncate the binary expansion of $\tfrac1{10}$ after $23$ binary places. Compute the error per tick exactly, then the accumulated error after $100$ hours of operation ($3.6\times10^6$ ticks).
  4. Explain why "every bounded set has a supremum" holds trivially for floating-point numbers, and why that does not make them a complete ordered field.

Tools: Python (with fractions.Fraction for exact arithmetic) or a spreadsheet.

Hand in: A one-page note listing the axioms you saw fail, your two computed errors, and one sentence on how interval arithmetic (nested intervals!) is used to make computations rigorous.

Going further. Read about Kahan summation, or about verified computing, in which a computer proves a numerical statement by enclosing every quantity in a small closed interval.

Instructor note: $(0.1+0.2)+0.3$ gives $0.6000000000000001$ and $0.1+(0.2+0.3)$ gives $0.6$ (associativity fails); $k=52$. The 23-place truncation of $\tfrac1{10}$ is $838860/2^{23}$, too small by $2^{-23}\cdot0.8=9.54\times10^{-8}$ s per tick, about $0.34$ s after $100$ hours — long enough for a fast missile to travel several hundred metres. Floating-point numbers form a finite totally ordered set, so every subset has a maximum; they are not a field at all.

🧪 Self-check quiz

0/5 correct

Q1. $\sup\left\{1-\tfrac1n:n\in\mathbb{N}\right\}$ is

Q2. Which ordered field is not Archimedean?

Q3. $u=\sup S$ if and only if $u$ is an upper bound of $S$ and for every $\varepsilon\gt0$ there is $s\in S$ with

Q4. Which nested family has empty intersection?

Q5. The binary expansion of $\tfrac1{10}$ is