Mathematics · Book 3 · Bachelor Year 1

University Mathematics — Year 1

University Mathematics — Year 1 · Bachelor Year 1

12Topology of the Real Line

Limits keep referring to the same geometric vocabulary: points “close to” a set, sets “without boundary leaks”, intervals in which sequences cannot escape. This chapter fixes that vocabulary — open and closed sets, interior and closure, density — on the real line, and proves the compactness of segments in its sequential form. The same notions, in normed vector spaces, are second-year material; on R\R they are within reach and immediately useful for Chapter 13.

12.1 Open sets, closed sets

Definition 12.1 (Neighborhood, open set)

A set VRV \subseteq \R is a neighborhood of xRx \in \R when it contains an interval (xr,x+r)\intoo{x - r}{x + r} for some r>0r > 0. A set URU \subseteq \R is open when it is a neighborhood of each of its points:

xU, r>0,(xr,x+r)U.\forall x \in U,\ \exists r > 0, \quad \intoo{x - r}{x + r} \subseteq U .

Example 12.2

Open intervals are open: for x(a,b)x \in \intoo{a}{b}, take r=min(xa,bx)>0r = \min(x - a,\, b - x) > 0. Half-lines (a,+)\intoo{a}{+\infty} are open; R\R and \emptyset are open (the latter vacuously). [0,1]\intcc{0}{1} is not open: no interval around 00 stays inside.

Proposition 12.3 (Stability of open sets)

Any union of open sets is open; a finite intersection of open sets is open. Infinite intersections may fail: n1(1n,1n)={0}\bigcap_{n \geq 1} \intoo{-\frac 1n}{\frac 1n} = \{0\}, not open.

Proof. Union: if xiUix \in \bigcup_i U_i, then xUi0x \in U_{i_0} for some i0i_0, and the interval provided by Ui0U_{i_0} sits inside the union. Finite intersection: if xU1Ukx \in U_1 \cap \dots \cap U_k, take r=min(r1,,rk)>0r = \min(r_1, \dots, r_k) > 0 of the radii provided by each UjU_j. For the counterexample: any interval around 00 contains some 1n\frac{1}{n} (Archimedes), hence leaves the intersection.

Example 12.4 (Certifying openness with explicit radii)

Is U={xR:x2>2}U = \{x \in \R : x^2 > 2\} open? Yes, and the certificate can be written down: U=(,2)(2,+)U = \intoo{-\infty}{-\sqrt2} \cup \intoo{\sqrt2}{+\infty}, a union of two open half-lines, open by Proposition 12.3. Alternatively, argue point by point: for xUx \in U with x>2x > \sqrt 2, take r=x2>0r = x - \sqrt2 > 0: every y(xr,x+r)y \in \intoo{x - r}{x + r} satisfies y>2y > \sqrt 2, hence y2>2y^2 > 2; symmetrically on the left. Both styles matter — the structural one (build from known open sets by unions and finite intersections) scales better, the ε\varepsilon-style one works when no structure is visible; and Chapter 13 will add a third, the most powerful: UU is the preimage of the open (2,+)\intoo{2}{+\infty} under the continuous xx2x \mapsto x^2.

Definition 12.5 (Closed set)

A set FRF \subseteq \R is closed when its complement RF\R \setminus F is open. By de Morgan and Proposition 12.3: any intersection of closed sets is closed, finite unions of closed sets are closed.

Theorem 12.6 (Sequential characterization of closed sets)

FF is closed if and only if: for every sequence (un)(u_n) of points of FF converging to some R\ell \in \R, the limit \ell belongs to FF. (“Closed== “stable under limits”.)

Proof. (\Rightarrow) Let FF be closed, unFu_n \in F, unu_n \to \ell, and suppose F\ell \notin F. The complement is open: some (r,+r)\intoo{\ell - r}{\ell + r} avoids FF. But convergence puts unu_n in that interval for large nn: contradiction with unFu_n \in F.

(\Leftarrow) Suppose FF is not closed: the complement is not open, so some xFx \notin F has no interval (xr,x+r)\intoo{x - r}{x + r} inside the complement; taking r=1n+1r = \frac{1}{n+1}, choose unFu_n \in F with unx<1n+1\abs{u_n - x} < \frac{1}{n+1}. Then unFu_n \in F, unxFu_n \to x \notin F: the sequential property fails.

Example 12.7 (The sequential test, both ways)

Closed: F=Z{n+1n:n2}F = \Z \cup \bigl\{n + \frac1n : n \geq 2\bigr\}. Let ukFu_k \in F with uku_k \to \ell. The window [1,+1]\intcc{\ell - 1}{\ell + 1} contains only finitely many points of FF (finitely many integers, finitely many n+1nn + \frac1n), and beyond some rank all uku_k lie in it: the sequence then takes finitely many values, and, converging, is eventually constant (as in Exercise 12.3): F\ell \in F. Closed — even though FF contains pairs of points at distance 1n\frac1n, arbitrarily close.

Not closed: G={1m+1n:m,nN}G = \bigl\{\frac1m + \frac1n : m, n \in \N^*\bigr\}. The sequence 1n+1nG\frac1n + \frac1n \in G tends to 00, and 0G0 \notin G (a sum of two positive terms): the sequential test fails, GG is not closed. Interestingly, each 1m\frac1m does belong to GG\overline G \cap G: indeed 1m=1m+1+1m(m+1)G\frac1m = \frac{1}{m+1} + \frac{1}{m(m+1)} \in G. The closing insight: to prove closedness, control all convergent sequences at once (usually via a local finiteness or a closed-formula argument); to disprove it, one well-chosen escaping sequence suffices — the asymmetry makes the negative direction the easy one, and the counterexamples of this chapter all have this one-line shape.

Example 12.8

Segments [a,b]\intcc{a}{b}, half-lines [a,+)\intco{a}{+\infty}, finite sets, Z\Z (a convergent sequence of integers is eventually constant) are closed. (0,1]\intoc{0}{1} is neither open (fails at 11) nor closed (1n0\frac 1n \to 0 \notin the set): most sets are neither. R\R and \emptyset are both open and closed — and they are the only such subsets of R\R (Exercise 12.9).

Example 12.9 (An open set assembled from infinitely many pieces)

RZ=nZ(n,n+1)\R \setminus \Z = \bigcup_{n \in \Z} \intoo{n}{n+1}: an infinite union of open intervals, open by Proposition 12.3 — so Z\Z is closed with no sequential argument needed. Note the division of labor in the stability rules: unions of open sets may be arbitrary (each point only needs its own certificate, supplied by the one set containing it), while intersections must stay finite (certificates must be intersected, and infinitely many radii can shrink to nothing). Exercise 12.10 will show this example is the general shape: every open subset of R\R is a countable disjoint union of open intervals.

12.2 Interior, closure, density

Definition 12.10 (Interior, closure, boundary)

Let ARA \subseteq \R.

  • A point xx is interior to AA when AA is a neighborhood of xx; the interior A˚\mathring{A} is the set of interior points.
  • A point xx is adherent to AA when every neighborhood of xx meets AA; the closure A\overline{A} is the set of adherent points.
  • The boundary is A=AA˚\partial A = \overline A \setminus \mathring A.

Then A˚AA\mathring A \subseteq A \subseteq \overline A.

Proposition 12.11 (Main properties)

  1. A˚\mathring A is the largest open set contained in AA; AA is open iff A=A˚A = \mathring A.
  2. A\overline A is the smallest closed set containing AA; AA is closed iff A=AA = \overline A.
  3. (Sequential characterization of adherence) xAx \in \overline A if and only if xx is the limit of a sequence of points of AA.
  4. Complementation exchanges the notions: RA=(RA) ⁣\R \setminus \overline A = \bigl(\R \setminus A\bigr)^{\!\circ}.

Proof. (4) xAx \notin \overline A     \iff some neighborhood of xx avoids AA     \iff some interval around xx lies in RA\R \setminus A     \iff xx is interior to RA\R \setminus A.

(1) A˚\mathring A is open: if xA˚x \in \mathring A, some (xr,x+r)A\intoo{x-r}{x+r} \subseteq A; every point yy of that interval has a smaller interval around it inside it, hence inside AA: the whole interval is in A˚\mathring A. Any open UAU \subseteq A consists of interior points of AA, so UA˚U \subseteq \mathring A: largest. The characterization of openness follows.

(2) In detail. By (4), RA\R \setminus \overline A is the interior of RA\R \setminus A, an open set by (1): so A\overline A is closed, and it contains AA. Minimality: let FAF \supseteq A be closed. Then RF\R \setminus F is open and contained in RA\R \setminus A, so by the maximality in (1),

RF(RA) ⁣=RA,\R \setminus F \subseteq \bigl(\R \setminus A\bigr)^{\!\circ} = \R \setminus \overline A ,

and taking complements again: AF\overline A \subseteq F. Thus A\overline A is the smallest closed superset. Characterization: if A=AA = \overline A then AA is closed (just shown); if AA is closed, it is itself a closed superset of AA, so minimality forces AA\overline A \subseteq A, and equality.

(3) If unAu_n \in A, unxu_n \to x: every neighborhood of xx contains some unAu_n \in A, so xAx \in \overline A. Conversely, if xAx \in \overline A: each interval (x1n+1,x+1n+1)\intoo{x - \frac{1}{n+1}}{x + \frac{1}{n+1}} meets AA at some unu_n, and unxu_n \to x.

Example 12.12

(0,1)=[0,1]\overline{\intoo{0}{1}} = \intcc{0}{1}; [0,1]˚=(0,1)\mathring{\intcc{0}{1}} = \intoo{0}{1}; (0,1)={0,1}\partial\intoo{0}{1} = \{0, 1\}. For A={1n:nN}A = \{\frac 1n : n \in \N^*\}: A=A{0}\overline A = A \cup \{0\}, A˚=\mathring A = \emptyset, A=A{0}\partial A = A \cup \{0\}. For Q\Q: by density (Theorem 10.14) every real is adherent to Q\Q, so Q=R\overline{\Q} = \R while Q˚=\mathring{\Q} = \emptyset (every interval contains irrationals): the boundary of Q\Q is all of R\R.

Example 12.13 (A full anatomy)

Let A=(0,1](Q(2,3)){4}A = \intoc{0}{1} \,\cup\, \bigl(\Q \cap \intoo{2}{3}\bigr) \,\cup\, \{4\}. We compute the three sets of Definition 12.10, piece by piece.

Interior. A point of (0,1)\intoo{0}{1} has a whole interval inside AA: interior. The point 11: every interval around it leaks right of 11, where AA has nothing until 22: not interior. No point of Q(2,3)\Q \cap \intoo{2}{3} is interior (every interval contains irrationals, Theorem 10.14); neither is the isolated 44. So A˚=(0,1)\mathring A = \intoo{0}{1}.

Closure. Limits of points of AA: all of [0,1]\intcc{0}{1} (0=lim1n0 = \lim \frac1n with 1nA\frac 1n \in A); all of [2,3]\intcc{2}{3} (every real there is a limit of rationals of the interval, density again); and 44. Nothing else: a point outside [0,1][2,3]{4}\intcc{0}{1} \cup \intcc{2}{3} \cup \{4\} has positive distance to that closed set. So A=[0,1][2,3]{4}\overline A = \intcc{0}{1} \cup \intcc{2}{3} \cup \{4\}.

Boundary. A=AA˚={0,1}[2,3]{4}\partial A = \overline A \setminus \mathring A = \{0, 1\} \cup \intcc{2}{3} \cup \{4\}.

The closing insight: the three operations act locally — each piece of AA contributes according to its own nature (a solid interval keeps its inside, a dense-but-porous piece turns entirely into boundary, an isolated point is pure boundary), and a two-line drawing of AA predicts every answer before any proof is written.

Remark 12.14 (Common pitfalls in point-set reasoning)

(i) “Not open” does not mean “closed: most sets are neither ((0,1]\intoc{0}{1}), and two sets are both (\emptyset, R\R) — open and closed are not opposites but duals through complementation. (ii) Interior and closure do not commute: for A=QA = \Q,

A˚==while(A) ⁣=R˚=R:\overline{\mathring A} = \overline\emptyset = \emptyset \qquad\text{while}\qquad \bigl(\,\overline A\,\bigr)^{\!\circ} = \mathring \R = \R :

the two iterated operators differ as much as sets can. (iii) Infinite unions of closed sets can fail to be closed: n1[1n,1]=(0,1]\bigcup_{n\geq1} \intcc{\frac1n}{1} = \intoc{0}{1} — the mirror of the intersection counterexample of Proposition 12.3. (iv) Dense does not mean big: Q\Q is dense, countable, with empty interior, and its complement is dense too; density says “arbitrarily close to everything”, not “almost everything” — the weekend problem’s Cantor set (Problem 12.1) makes the opposite point, a topologically small set that is uncountably big.

Example 12.15 (A closure computed exactly)

Let G={1m+1n:m,nN}G = \bigl\{\frac1m + \frac1n : m, n \in \N^*\bigr\} (from Example 12.7). Claim:

G=G{1m:mN}{0}.\overline G = G \,\cup\, \Bigl\{\frac1m : m \in \N^*\Bigr\} \,\cup\, \{0\} .

(\supseteq) 1m=limn(1m+1n)\frac1m = \lim_n \bigl(\frac1m + \frac1n\bigr) and 0=limn2n0 = \lim_n \frac2n: adherent by the sequential characterization. (\subseteq) Let x=limk(1mk+1nk)x = \lim_k \bigl( \frac{1}{m_k} + \frac{1}{n_k}\bigr); order each pair so that mknkm_k \leq n_k. If (mk)(m_k) is unbounded, a subsequence has mkm_k \to \infty, hence nkn_k \to \infty too and x=0x = 0. Otherwise (mk)(m_k) takes finitely many values, one of them, say mm, infinitely often; along that subsequence 1nkx1m\frac{1}{n_k} \to x - \frac1m: if (nk)(n_k) is bounded it takes some value nn infinitely often and x=1m+1nGx = \frac1m + \frac1n \in G; if not, x=1mx = \frac1m. Every case lands in the announced set. The closing insight: computing a closure is a compactness-style case analysis on indices — bounded index means finitely many values (pigeonhole), unbounded index means a limit escapes — and the answer displays the typical two-layer structure of limit points: the set, its first-generation limits, and their limit 00.

Definition 12.16 (Density, topological form)

AA is dense in R\R when A=R\overline A = \R — equivalently, every nonempty open interval meets AA; equivalently (by Proposition 12.11 (3)), every real is a limit of elements of AA. Examples: Q\Q, RQ\R \setminus \Q, the dyadics (Exercise 10.8), dense subgroups (Exercise 10.9).

Example 12.17 (Density is relative)

Dense” as defined here means dense in R\R; a set can instead be dense in a part of the line only. The dyadics of [0,1]\intcc{0}{1}, i.e. D[0,1]D \cap \intcc{0}{1} (Exercise 10.8), meet every open interval included in [0,1]\intcc{0}{1} but of course miss (2,3)\intoo{2}{3} entirely: they are dense in [0,1]\intcc{0}{1}, meaning D[0,1]=[0,1]\overline{D \cap \intcc{0}{1}} = \intcc{0}{1}. The general phrase “AA is dense in BB” abbreviates BAB \subseteq \overline A — always name the ambient set, since the weekend problem’s endpoints are dense in the Cantor set while being nowhere dense in R\R: the same set, two truthful and opposite-sounding descriptions.

Example 12.18 (Handling density)

Three quick moves that recur constantly. Enlarging: if AA is dense and ABA \subseteq B, then BB is dense (every interval already meets AA). Transporting: if AA is dense, so is λA+μ\lambda A + \mu for λ0\lambda \neq 0 — an interval II meets λA+μ\lambda A + \mu iff the interval Iμλ\frac{I - \mu}{\lambda} meets AA; thus the odd multiples of 10910^{-9}, say, are dense. Intersecting fails: two dense sets can miss each other entirely (Q\Q and RQ\R \setminus \Q): density survives unions and affine maps, never intersections.

12.3 Compactness of segments

Theorem 12.19 (Segments are sequentially compact)

Let aba \leq b. Every sequence of points of [a,b]\intcc{a}{b} has a subsequence converging to a point of [a,b]\intcc{a}{b}.

More generally, the subsets of R\R with this property (every sequence has a subsequence converging in the set) are exactly the closed and bounded sets.

Proof. A sequence in [a,b]\intcc{a}{b} is bounded, so Bolzano–Weierstrass (Theorem 11.16) extracts a convergent subsequence; its limit stays in [a,b]\intcc{a}{b} because segments are closed (Theorem 12.6).

General case. (Closed bounded \Rightarrow compact): let FF be closed and bounded, and (un)(u_n) a sequence in FF. Boundedness of FF bounds the sequence, so Bolzano–Weierstrass extracts uφ(n)u_{\varphi(n)} \to \ell; and F\ell \in F because FF is closed and the subsequence is a convergent sequence of points of FF (Theorem 12.6): the two hypotheses are consumed one each, boundedness for existence of the limit, closedness for its membership. (Compact \Rightarrow closed and bounded): if FF is unbounded, pick unFu_n \in F with unn\abs{u_n} \geq n; every subsequence is unbounded, hence divergent (Proposition 11.4): no convergent subsequence at all. If FF is not closed, take unFu_n \in F with unFu_n \to \ell \notin F (Theorem 12.6): every subsequence converges to F\ell \notin F, so no subsequence converges in FF.

Example 12.20 (Nested compact sets)

A first workout for the theorem. Let K0K1K2K_0 \supseteq K_1 \supseteq K_2 \supseteq \dots be nonempty compact (closed bounded) subsets of R\R. Then nKn\bigcap_n K_n \neq \emptyset. Indeed, pick xnKnx_n \in K_n for each nn: the sequence lives in the compact K0K_0, so a subsequence xφ(n)x_{\varphi(n)} converges to some xx (Theorem 12.19). For every fixed mm, the terms xφ(n)x_{\varphi(n)} with φ(n)m\varphi(n) \geq m all lie in the closed set KmK_m, so the limit xx lies in KmK_m (Theorem 12.6); as mm was arbitrary, xnKnx \in \bigcap_n K_n. A useful companion: if an open set UU contains nKn\bigcap_n K_n, then UKnU \supseteq K_n for some nn — apply the same argument to points xnKnUx_n \in K_n \setminus U; the limit xx would lie in KnU\bigcap K_n \subseteq U, yet UU open forces xφ(n)Ux_{\varphi(n)} \in U eventually, a contradiction. Both statements fail without compactness: n(0,1n)=\bigcap_n \intoo{0}{\frac 1n} = \emptyset and n[n,+)=\bigcap_n \intco{n}{+\infty} = \emptyset. The closing insight: compactness converts an infinite chain of non-emptiness assertions into a single limit point — it is the tool that survives passage to the infinite intersection, and the weekend problem (Problem 12.1) will lean on it twice.

The first four stages of the middle-thirds construction: each segment of C_n loses its open middle third, leaving the 2n+1 segments of C_n+1, of total length (2/3)n+1. The Cantor set C = _n C_n — the subject of the weekend problem  — is the nonempty compact residue guaranteed by the nested-compacts argument above: length zero, yet uncountably many points survive.
The first four stages of the middle-thirds construction: each segment of CnC_n loses its open middle third, leaving the 2n+12^{n+1} segments of Cn+1C_{n+1}, of total length (23)n+1(\frac23)^{n+1}. The Cantor set C=nCnC = \bigcap_n C_n — the subject of the weekend problem Problem 12.1 — is the nonempty compact residue guaranteed by the nested-compacts argument above: length zero, yet uncountably many points survive.

Remark 12.21

This is the engine behind the extreme value theorem (Chapter 13) and Heine’s theorem on uniform continuity. The name “compact” will acquire its general (covering) definition in the second year; on R\R, sequential compactness is all we need, and “compact == closed ++ bounded” is the statement to remember.

Example 12.22 (Boundaries under unions)

Always (AB)AB\partial(A \cup B) \subseteq \partial A \cup \partial B: a point of (AB)\partial(A \cup B) has every neighborhood meeting ABA \cup B (hence AA or BB, infinitely often one of them) and meeting the complement of ABA \cup B, which lies in both complements — a short check then places the point in A\partial A or B\partial B. The inclusion can be spectacularly strict: with A=QA = \Q and B=RQB = \R\setminus\Q,

(AB)=R=,AB=RR=R:\partial(A \cup B) = \partial \R = \emptyset , \qquad \partial A \cup \partial B = \R \cup \R = \R :

two ragged sets can glue into a seamless one, their boundaries annihilating each other. The closing insight: interiors and closures behave monotonically under unions and intersections, but boundaries do not — treat \partial as a derived quantity (AA˚\overline A \setminus \mathring A), never as an operator with algebra of its own.

Remark 12.23 (Perspectives inside this volume)

The vocabulary built here is consumed twice more in this book. In Chapter 13, every theorem is a topology statement in disguise: the intermediate value theorem says continuous maps preserve the interval property, the extreme value theorem says they preserve compactness — and the proofs call Theorems 12.6 and 12.19 by name. In Chapter 25, the same definitions are re-read in R2\R^2 with disks in place of intervals: open sets, closures and compactness transfer word for word, and the two-variable extreme value theorem again rides on Bolzano–Weierstrass (extract on each coordinate). The one notion that does not generalize painlessly is the interval itself — in the plane, connectedness replaces convexity, a story that begins with Exercise 12.9’s “only \emptyset and R\R are open and closed”.

12.4 Exercises

Exercise 12.1

For each set, say whether it is open, closed, both, or neither (with justification): (0,1)(2,3)\intoo{0}{1} \cup \intoo{2}{3};   [0,1)\;\intco{0}{1};   {0}[1,2]\;\{0\} \cup \intcc{1}{2};   RZ\;\R \setminus \Z;   Q(0,1)\;\Q \cap \intoo{0}{1}.

Solution

Solution of Exercise 12.1.

(0,1)(2,3)\intoo{0}{1} \cup \intoo{2}{3}: open (union of open sets), not closed (1n0\frac 1n \to 0 outside).

[0,1)\intco{0}{1}: neither. Not open (no interval around 00 inside); not closed (11n11 - \frac1n \to 1 \notin set).

{0}[1,2]\{0\} \cup \intcc{1}{2}: closed (finite union of closed sets), not open (fails at 00).

RZ\R \setminus \Z: open (Z\Z is closed), not closed: the sequence (1n)\bigl(\frac 1n\bigr) lies in it, but its limit 00 belongs to Z\Z, i.e. escapes the set.

Q(0,1)\Q \cap \intoo{0}{1}: neither. Not open: every interval around a rational contains irrationals. Not closed: it contains sequences tending to the irrational 22\frac{\sqrt 2}{2} (density).

Exercise 12.2

Determine A˚\mathring A, A\overline A and A\partial A for: A=(0,1]{2}A = \intoc{0}{1} \cup \{2\};   A=RQ\;A = \R \setminus \Q;   A={(1)nnn+1:nN}\;A = \bigl\{\frac{(-1)^n n}{n+1} : n \in \N\bigr\}.

Solution

Solution of Exercise 12.2.

A=(0,1]{2}A = \intoc{0}{1} \cup \{2\}: A˚=(0,1)\mathring A = \intoo{0}{1}, A=[0,1]{2}\overline A = \intcc{0}{1} \cup \{2\}, A={0,1,2}\partial A = \{0, 1, 2\}.

A=RQA = \R \setminus \Q: A˚=\mathring A = \emptyset (every interval contains rationals), A=R\overline A = \R (density of the irrationals), A=R\partial A = \R.

A={(1)nnn+1}A = \bigl\{\frac{(-1)^n n}{n+1}\bigr\}: the even terms tend to 11, the odd terms to 1-1, neither belongs to AA. A˚=\mathring A = \emptyset (isolated points), A=A{1,1}\overline A = A \cup \{-1, 1\}, A=A\partial A = \overline A.

Exercise 12.3

Prove that a finite set is closed, first via complements, then via the sequential characterization.

Solution

Solution of Exercise 12.3.

Complements. F={a1<a2<<ak}F = \{a_1 < a_2 < \dots < a_k\}: the complement is the union of the open intervals (,a1)\intoo{-\infty}{a_1}, (ai,ai+1)\intoo{a_i}{a_{i+1}}, (ak,+)\intoo{a_k}{+\infty}open by Proposition 12.3.

Sequences. Let unFu_n \in F, unu_n \to \ell. With ε=min{aiaj:ij}/2>0\varepsilon = \min\{\abs{a_i - a_j} : i \neq j\}/2 > 0 (or any ε\varepsilon if FF is a singleton): beyond some rank, all terms are within ε\varepsilon of \ell, hence within 2ε2\varepsilon of each other, which forces them to be one single aia_i from that rank on; then =aiF\ell = a_i \in F.

Exercise 12.4

Prove that for any A,BRA, B \subseteq \R: AB=AB\overline{A \cup B} = \overline A \cup \overline B. Show by example that AB\overline{A \cap B} can differ from AB\overline A \cap \overline B.

Solution

Solution of Exercise 12.4.

\subseteq: AB\overline A \cup \overline B is closed (finite union) and contains ABA \cup B, so it contains the smallest closed superset AB\overline{A \cup B}. \supseteq: AABA \subseteq A \cup B gives AAB\overline A \subseteq \overline{A \cup B} (closure is monotone: adherent points of AA are adherent to the larger set), and likewise for BB.

Counterexample for intersections: A=(0,1)A = \intoo{0}{1}, B=(1,2)B = \intoo{1}{2}: AB==\overline{A \cap B} = \overline\emptyset = \emptyset but AB={1}\overline A \cap \overline B = \{1\}.

Exercise 12.5 ★★

Let unu_n \to \ell in R\R. Prove that the set {un:nN}{}\{u_n : n \in \N\} \cup \{\ell\} is closed (hence compact if we add that it is bounded — which it is).

Solution

Solution of Exercise 12.5.

Use the sequential characterization (Theorem 12.6). Let S={un}{}S = \{u_n\} \cup \{\ell\} and (vk)(v_k) a sequence in SS with vkmv_k \to m; show mSm \in S. Two cases. If some value vSv \in S is taken by (vk)(v_k) infinitely often, then a constant subsequence gives m=vSm = v \in S. Otherwise every value is taken finitely often; in particular, for each nn, the term unu_n appears finitely often, and \ell too. Then for every NN, the indices kk with vk{u0,,uN,}v_k \in \{u_0, \dots, u_N, \ell\} are finitely many: the remaining vkv_k are terms unu_n with n>Nn > N. Given ε>0\varepsilon > 0, choose NN with unε\abs{u_n - \ell} \leq \varepsilon for n>Nn > N: all but finitely many vkv_k satisfy vkε\abs{v_k - \ell} \leq \varepsilon. Hence vkv_k \to \ell, so m=Sm = \ell \in S.

Exercise 12.6 ★★

Let UU be open and AA arbitrary. Prove that U+A={u+a}U + A = \{u + a\} is open. Deduce that the sum of an open set and any set is open, and contrast: exhibit two closed sets whose sum is not closed. (Try Z\Z and 2Z\sqrt 2\,\Z, with Exercise 10.9.)

Solution

Solution of Exercise 12.6.

U+A=aA(U+a)U + A = \bigcup_{a \in A} (U + a), and each translate U+aU + a is open (translating the interval certificates). A union of open sets is open (Proposition 12.3).

Closed sets: Z\Z and 2Z\sqrt 2\,\Z are closed (as αZ\alpha\Z: convergent sequences are eventually constant, cf. Example 12.8). Their sum Z+2Z\Z + \sqrt 2\,\Z is dense in R\R (Exercise 10.9) but is not R\R (it is countable, or simply: 22Z+2Z\frac{\sqrt 2}{2} \notin \Z + \sqrt2\Z, else 2\sqrt 2 would be rational — writing 22=m+n2\frac{\sqrt2}{2} = m + n\sqrt 2 forces (2n1)2=2m(2n - 1)\sqrt 2 = -2m, so 2Q\sqrt 2 \in \Q unless n=12n = \frac12, impossible). A dense proper subset is not closed: its closure is R\R \neq itself.

Exercise 12.7 ★★

A point xAx \in A is isolated in AA when some neighborhood of xx meets AA only at xx. Prove that every point of Z\Z is isolated in Z\Z, that A={1n}A = \{\frac1n\} has all its points isolated yet AA\overline A \neq A, and that a set whose points are all isolated has empty interior.

Solution

Solution of Exercise 12.7.

Z\Z: the neighborhood (n12,n+12)\intoo{n - \frac12}{n + \frac12} of nn meets Z\Z only at nn.

A={1n:nN}A = \{\frac1n : n \in \N^*\}: around 1n\frac 1n, the interval of radius 1n1n+1=1n(n+1)\frac{1}{n} - \frac{1}{n+1} = \frac{1}{n(n+1)} (halved, say) isolates it from its neighbors — all points isolated. Yet 0AA0 \in \overline A \setminus A: isolated points do not prevent adherent outsiders.

If all points of AA are isolated: no point of AA is interior, since an interior point has a whole interval of AA-neighbors around it (an interval is infinite), contradicting isolation. So A˚=\mathring A = \emptyset.

Exercise 12.8 ★★

Prove that the closure of a bounded set is bounded, and that supA=supA\sup \overline A = \sup A for AA nonempty bounded above. Deduce that supAA\sup A \in \overline A: the supremum is always adherent.

Solution

Solution of Exercise 12.8.

If A[M,M]A \subseteq \intcc{-M}{M}, the closed set [M,M]\intcc{-M}{M} contains AA, hence contains A\overline A (smallest closed superset): A\overline A is bounded.

Let s=supAs = \sup A (finite). Since AAA \subseteq \overline A, supAs\sup \overline A \geq s. Conversely, A(,s]\overline A \subseteq \intoc{-\infty}{s}: the half-line is closed and contains AA; so every element of A\overline A is s\leq s, giving supAs\sup\overline A \leq s. Equality.

sAs \in \overline A: by the ε\varepsilon-characterization (Proposition 10.4), every interval (sε,s+ε)\intoo{s - \varepsilon}{s + \varepsilon} contains an element of AA: ss is adherent.

Exercise 12.9 ★★★

Prove that the only subsets of R\R that are both open and closed are \emptyset and R\R. Hint: suppose AA is open, closed, with AA \neq \emptyset and RA\R \setminus A \neq \emptyset; pick aAa \in A, bAb \notin A, say a<ba < b, and consider s=sup(A[a,b])s = \sup\,(A \cap \intcc{a}{b}); decide whether ss can belong to AA or to its complement.

Solution

Solution of Exercise 12.9.

Suppose AA is open and closed, with aAa \in A and bRAb \in \R \setminus A; without loss of generality a<ba < b. The set B=A[a,b]B = A \cap \intcc{a}{b} is nonempty (aa), bounded: let s=supBs = \sup B. By Exercise 12.8, sBA=As \in \overline B \subseteq \overline A = A (AA closed). Note sbs \leq b, and since bAb \notin A: s<bs < b. Now AA is open: some interval (sr,s+r)\intoo{s - r}{s + r} lies in AA, and we may take r<bsr < b - s. Then s+r2s + \frac{r}{2} belongs to A[a,b]=BA \cap \intcc{a}{b} = B and exceeds ss — contradicting s=supBs = \sup B. Hence no such pair (a,b)(a, b) exists: one of AA, RA\R \setminus A is empty.

Exercise 12.10 ★★★

(Structure of open sets) Let URU \subseteq \R be open, nonempty. For xUx \in U, let IxI_x be the union of all open intervals containing xx and contained in UU. Prove that IxI_x is an open interval, that two sets IxI_x, IyI_y are equal or disjoint, and that UU is a union of countably many pairwise disjoint open intervals (pick a rational in each).

Solution

Solution of Exercise 12.10.

IxI_x is a union of open intervals all containing xx: it is open, and it is an interval, being convex — if u<z<vu < z < v with u,vIxu, v \in I_x, then uu and vv lie in open subintervals JuxJ_u \ni x, JvxJ_v \ni x of UU, and JuJvJ_u \cup J_v is an interval (both contain xx) inside UU containing zz; so zIxz \in I_x (Proposition 10.19).

If IxIyI_x \cap I_y \neq \emptyset: IxIyI_x \cup I_y is then an open interval (convex: two overlapping intervals) contained in UU containing xx and yy, so IxIyIxI_x \cup I_y \subseteq I_x and Iy\subseteq I_y by maximality of each: Ix=IyI_x = I_y.

So UU is the disjoint union of the distinct sets IxI_x (each xUx \in U lies in its own IxI_x). Countability: each nonempty open interval II of the family contains a rational qIq_I (Theorem 10.14), and distinct disjoint intervals get distinct rationals: the family injects into Q\Q, which is countable (it is indexed by pairs of integers). Hence at most countably many intervals.

Exercise 12.11 ★★

A point xRx \in \R is an accumulation point of AA when every neighborhood of xx meets A{x}A \setminus \{x\}; their set is the derived set AA'. Prove that A=AA\overline A = A \cup A', and that AA is closed if and only if AAA' \subseteq A. Determine AA' for A={1n:nN}A = \{\frac 1n : n \in \N^*\}, for A=ZA = \Z, and for A=QA = \Q.

Solution

Solution of Exercise 12.11.

A=AA\overline A = A \cup A'. (\supseteq) AAA \subseteq \overline A always; and if xAx \in A', every neighborhood of xx meets A{x}AA \setminus \{x\} \subseteq A, so xx is adherent. (\subseteq) Let xAx \in \overline A. If xAx \in A, done. If xAx \notin A, every neighborhood of xx meets A=A{x}A = A \setminus \{x\}: xAx \in A'.

Consequently AA closed     \iff A=A=AAA = \overline A = A \cup A'     \iff AAA' \subseteq A.

A={1n}A = \{\frac 1n\}: 00 is an accumulation point (1n0\frac 1n \to 0, terms 0\neq 0); each 1n\frac 1n is isolated (Exercise 12.7), so not in AA'; and a point xA{0}x \notin A \cup \{0\} has a whole interval avoiding AA (between the two neighbors of xx in A{0}A \cup \{0\}, or beyond 11). Hence A={0}A' = \{0\}.

Z=\Z' = \emptyset: every integer is isolated, every non-integer has a neighborhood inside RZ\R \setminus \Z.

Q=R\Q' = \R: every interval around any real contains infinitely many rationals (Theorem 10.14), in particular one different from the center.

Exercise 12.12 ★★★

For ARA \subseteq \R nonempty, define dA(x)=inf{xa:aA}d_A(x) = \inf\{\abs{x - a} : a \in A\}. Prove:

  1. dA(x)dA(y)xy\abs{d_A(x) - d_A(y)} \leq \abs{x - y} for all x,yx, y (dAd_A is 11-Lipschitz);
  2. dA(x)=0d_A(x) = 0 if and only if xAx \in \overline A; in particular, if FF is closed and xFx \notin F, then dF(x)>0d_F(x) > 0;
  3. for every ε>0\varepsilon > 0, the set Vε={x:dF(x)<ε}V_\varepsilon = \{x : d_F(x) < \varepsilon\} is open, contains FF, and ε>0Vε=F\bigcap_{\varepsilon > 0} V_\varepsilon = F for FF closed: every closed set is a countable intersection of open sets.
Solution

Solution of Exercise 12.12.

  1. For every aAa \in A: xaxy+ya\abs{x - a} \leq \abs{x - y} + \abs{y - a}, so dA(x)xy+yad_A(x) \leq \abs{x - y} + \abs{y - a}; taking the infimum over aa: dA(x)xy+dA(y)d_A(x) \leq \abs{x - y} + d_A(y). Exchanging xx and yy gives the other inequality: dA(x)dA(y)xy\abs{d_A(x) - d_A(y)} \leq \abs{x - y}.
  2. dA(x)=0d_A(x) = 0     \iff for every ε>0\varepsilon > 0 there is aAa \in A with xa<ε\abs{x - a} < \varepsilon     \iff every interval around xx meets AA     \iff xAx \in \overline A. If FF is closed and xF=Fx \notin F = \overline F, then dF(x)0d_F(x) \neq 0, i.e. dF(x)>0d_F(x) > 0.
  3. If dF(x)<εd_F(x) < \varepsilon, set r=εdF(x)>0r = \varepsilon - d_F(x) > 0: for yx<r\abs{y - x} < r, part (1) gives dF(y)dF(x)+xy<εd_F(y) \leq d_F(x) + \abs{x - y} < \varepsilon: the interval (xr,x+r)\intoo{x - r}{x + r} lies in VεV_\varepsilon, which is therefore open; it contains FF since dF=0d_F = 0 there. Finally xε>0Vεx \in \bigcap_{\varepsilon>0} V_\varepsilon     \iff dF(x)<εd_F(x) < \varepsilon for all ε\varepsilon     \iff dF(x)=0d_F(x) = 0     \iff xF=Fx \in \overline F = F. Since ε>0Vε=n1V1/n\bigcap_{\varepsilon > 0} V_\varepsilon = \bigcap_{n \geq 1} V_{1/n}, every closed set is a countable intersection of open sets.

12.5 Problem: The Cantor set, small and enormous at once

Problem 12.1

Weekend problem — the Cantor middle-thirds set: length zero, uncountable, perfect, and C+C=[0,2]C + C = \intcc{0}{2}

Remove from [0,1]\intcc{0}{1} its open middle third, then the middle third of each remaining segment, and repeat forever: what survives is the Cantor set CC, the fundamental counterexample factory of analysis. This problem constructs it, reads it through the base-33 machinery of Problem 10.1, and establishes its paradoxical portrait: total length zero, yet uncountable; empty interior, yet no isolated point; totally disconnected, yet C+CC + C fills the whole segment [0,2]\intcc{0}{2}. Formally: C0=[0,1]C_0 = \intcc{0}{1}, and Cn+1C_{n+1} is obtained from CnC_n by deleting the open middle third of each segment of CnC_n; finally C=n0CnC = \bigcap_{n \geq 0} C_n. Throughout, a ternary code of x[0,1]x \in \intcc{0}{1} is any digit string (dk)k1(d_k)_{k\geq1} with dk{0,1,2}d_k \in \{0, 1, 2\} whose value supnk=1ndk3k\sup_n \sum_{k=1}^n d_k 3^{-k} equals xx — improper codes (eventually 22) are allowed; by Problem 10.1 (questions 9–11), every x[0,1]x \in \intcc{0}{1} has one or two codes, two exactly when x=m/3N(0,1)x = m/3^N \in \intoo{0}{1}.

Part I — The construction.

  1. Describe C1C_1 and C2C_2 explicitly as unions of segments, and prove by induction: CnC_n is a disjoint union of 2n2^n closed segments, each of length 3n3^{-n}.
  2. Show that CC is closed, bounded — hence compact (Theorem 12.19) — nonempty, and that every endpoint of every segment of every CnC_n belongs to CC.
  3. The total length of CnC_n is (23)n\bigl(\frac23\bigr)^n. Deduce that for every ε>0\varepsilon > 0, the set CC can be covered by finitely many segments of total length ε\leq \varepsilon: the Cantor set has length zero.
  4. Show that an interval contained in CC has length 3n\leq 3^{-n} for every nn, hence is a singleton or empty: C˚=\mathring C = \emptyset. Being closed with empty interior, CC is nowhere dense.

Part II — The ternary code.

  1. Prove the self-similarity recursion

    Cn+1=13Cn(23+13Cn),henceC=13C(23+13C),C_{n+1} = \tfrac13 C_n \,\cup\, \bigl(\tfrac23 + \tfrac13 C_n\bigr), \qquad\text{hence}\qquad C = \tfrac13 C \,\cup\, \bigl(\tfrac23 + \tfrac13 C\bigr),

    the two pieces being disjoint: CC is two copies of itself at scale 13\frac13.

  2. Prove by induction on nn: xCnx \in C_n if and only if xx has a ternary code whose first nn digits lie in {0,2}\{0, 2\}. Deduce, using the fact that xx has at most two codes: xCx \in C if and only if xx has a code with no digit equal to 11 (a 11-free code).
  3. Codes in action: give 11-free codes for 00, 11, 13\frac13, 23\frac23; show 14=(0.02)3\frac14 = (0.\overline{02})_3 and 34=(0.20)3\frac34 = (0.\overline{20})_3, so both belong to CC; and check that 14\frac14 is not an endpoint of any CnC_n (endpoints have the form m/3nm/3^n).
  4. Show that each xCx \in C has exactly one 11-free code (when xx has two codes, prove that exactly one of the pair contains the digit 11). Conclude: the value map is a bijection from {0,2}\{0,2\}-strings onto CC.
  5. (Diagonal) Let kxkk \mapsto x_k be any map NC\N^* \to C. Build a {0,2}\{0, 2\}-string differing at index kk from the code of xkx_k, and conclude that CC is uncountable — while, by contrast, question 3 says it is metrically negligible.

Part III — Topological portrait.

  1. Assemble the record so far: CC is compact, uncountable, of length zero, nowhere dense. Which single containment CCnC \subseteq C_n carries each property?
  2. (CC is perfect) Let xCx \in C with 11-free code (dk)(d_k). Flipping the digit dnd_n (020 \leftrightarrow 2) produces xnCx_n \in C with xnx=23n\abs{x_n - x} = 2 \cdot 3^{-n}. Conclude that CC has no isolated point: every point of CC is a limit of other points of CC.
  3. Show that the endpoints of question 2 form a countable dense subset of CC (truncate the code after nn digits and continue with 00s; countability as in Exercise 12.10). Conclude: the typical point of CC — like 14\frac14 — is not an endpoint: endpoints are a countable skeleton inside an uncountable body.
  4. (Totally disconnected) Let x<yx < y in CC. Choose nn with 3n<yx3^{-n} < y - x and produce a point z(x,y)z \in \intoo{x}{y} with zCz \notin C. Conclude that the only nonempty intervals contained in CC are singletons.

Part IV — Arithmetic of CC.

  1. Show 1C=C1 - C = C (what does x1xx \mapsto 1 - x do to a 11-free code? recall 1=(0.2)31 = (0.\overline{2})_3).
  2. (Addition of codes) Show that if xx, xx' have codes (ak)(a_k), (bk)(b_k), then x+x=limn(tn+tn)x + x' = \lim_n\,(t_n + t'_n) where tn,tnt_n, t'_n are the partial sums. Deduce: every y[0,1]y \in \intcc{0}{1} is the midpoint of two points of CC — given a code (ek)(e_k) of yy, choose digits ak,bk{0,2}a_k, b_k \in \{0, 2\} with ak+bk2=ek\frac{a_k + b_k}{2} = e_k.
  3. Conclude C+C=[0,2]C + C = \intcc{0}{2} and, with question 14, CC=[1,1]C - C = \intcc{-1}{1}. Concrete instance: write 11 as a sum of the two non-endpoints found in question 7.
  4. Reflect: a set of length zero whose difference set fills [1,1]\intcc{-1}{1}. Why is there no contradiction between “CC is metrically negligible” and “C+CC + C has full length”? (One sentence; think about what length controls and what it does not.)

Part V — Members, rational and irrational.

  1. Combine question 8 with the periodicity criterion of Problem 10.1: a point of CC is rational if and only if its 11-free code is eventually periodic. Run the base-33 long division to check 113=(0.002)3C\frac1{13} = (0.\overline{002})_3 \in C.
  2. Produce an explicitly irrational member of CC: the value of the code with dk=2d_k = 2 at the triangular positions k=j(j+1)2k = \frac{j(j+1)}{2} and dk=0d_k = 0 elsewhere. Justify irrationality by the growing-gaps argument of Problem 10.1 (question 20).
  3. (Onto a full segment) Consider hh mapping the point of CC with 11-free code (dk)(d_k) to the value of the binary string (dk2)\bigl(\frac{d_k}2\bigr), i.e. h(x)=supnk=1ndk22kh(x) = \sup_n \sum_{k=1}^n \frac{d_k}{2}\,2^{-k}. Show that hh maps CC onto [0,1]\intcc{0}{1}. So the negligible CC surjects onto a segment of full length — a second proof that CC is uncountable.
  4. (Self-similar length) Suppose some notion of length LL were defined for CC and its shrunken copies, respecting scaling (L(λA)=λL(A)L(\lambda A) = \lambda L(A)), translation invariance, and additivity over the disjoint decomposition of question 5. Show that then L(C)=23L(C)L(C) = \frac23\,L(C), forcing L(C)=0L(C) = 0: self-similarity alone already sentences CC to length zero.

Part VI — A fat cousin, and the moral.

  1. (Fat Cantor set) Repeat the construction, but at stage nn (n=0,1,2,n = 0, 1, 2, \dots) remove from each of the 2n2^n current segments a central open interval of length 4(n+1)4^{-(n+1)} only. Show that the segment lengths lnl_n obey ln+1=ln4(n+1)2l_{n+1} = \frac{l_n - 4^{-(n+1)}}{2}, ln=2n+124n>0l_n = \frac{2^n + 1}{2\cdot 4^n} > 0, that the resulting K=KnK = \bigcap K_n is compact with empty interior, and that the total removed length is n02n4(n+1)=12\sum_{n\geq0} 2^n 4^{-(n+1)} = \frac12. Admitting the (intuitive, Year 3) additivity of length for finite unions of intervals, and using both statements of Example 12.20, show that any finite family of open intervals covering KK has total length 12\geq \frac12: KK is nowhere dense but not negligible. Smallness has several inequivalent meanings.
  2. (Distances) Show that for a nonempty closed FRF \subseteq \R and xRx \in \R, the infimum d(x,F)d(x, F) is attained (minimizing sequence plus Bolzano–Weierstrass). Then compute

    maxy[0,1]d(y,C)=16,\max_{y \in \intcc{0}{1}} d(y, C) = \frac16 ,

    attained exactly at the center y=12y = \frac12 (a point of a gap created at stage nn is within 3n2\frac{3^{-n}}{2} of the gap’s endpoints, which lie in CC).

  3. (Every point a subsequential limit) Using questions 12 and 2, produce a single sequence in CC whose set of subsequential limits is all of CC. (Compare: for a convergent sequence that set is one point — CC realizes the opposite extreme among compacts.)
  4. Synthesis, one sentence each: (i) which theorems of this chapter did the construction actually consume (stability of closed sets, compactness, sequential characterizations)? (ii) list the four paradoxical pairings of the portrait (length zero/uncountable, closed/empty interior, perfect/totally disconnected, negligible/C+CC+C full); (iii) where does CC resurface later (the devil’s staircase built on hh in the theory of continuity, and the measure theory of the Year 3 volume, where CC separates “countable” from “negligible”)?
Solution

Solution of Problem 12.1.

1. C1=[0,13][23,1]C_1 = \intcc{0}{\frac13} \cup \intcc{\frac23}{1} and

C2=[0,19][29,13][23,79][89,1].C_2 = \intcc{0}{\tfrac19} \cup \intcc{\tfrac29}{\tfrac13} \cup \intcc{\tfrac23}{\tfrac79} \cup \intcc{\tfrac89}{1} .

Induction: if CnC_n is a disjoint union of 2n2^n closed segments of length 3n3^{-n}, deleting the open middle third of each leaves two closed segments of length 3n13^{-n-1} per parent: 2n+12^{n+1} segments, pairwise disjoint (children of distinct parents are separated because the parents were; children of one parent are separated by the removed gap).

2. Each CnC_n is a finite union of segments, hence closed; C=CnC = \bigcap C_n is an intersection of closed sets: closed (Definition 12.5); bounded ([0,1]\subseteq \intcc{0}{1}): compact by Theorem 12.19. Nonempty: 00 lies in the leftmost segment of every CnC_n. Let aa be an endpoint of a segment SS of CnC_n. For mnm \leq n, aCnCma \in C_n \subseteq C_m. For the later stages: the middle-third deletion never removes an endpoint, and aa is again an endpoint of one of the two children of SS (the child touching aa); by induction aCma \in C_m for all mnm \geq n: aCa \in C.

3. Total length of CnC_n: 2n3n=(23)n02^n \cdot 3^{-n} = (\frac23)^n \to 0. Given ε>0\varepsilon > 0, choose nn with (23)nε(\frac23)^n \leq \varepsilon: then CCnC \subseteq C_n, a union of finitely many segments of total length ε\leq \varepsilon.

4. Let ICI \subseteq C be an interval with two distinct points. For every nn: ICnI \subseteq C_n, and II, being convex, must lie inside a single segment of CnC_n (meeting two segments would force II to contain a point of the gap between them, which is outside CnC_n). Hence the length of II is 3n\leq 3^{-n} for all nn: contradiction. So the only intervals inside CC are empty or singletons; in particular no (xr,x+r)\intoo{x-r}{x+r} fits inside CC: C˚=\mathring C = \emptyset. As CC is closed, C=C\overline C = C has empty interior: CC is nowhere dense.

5. Write φ0(x)=x3\varphi_0(x) = \frac x3 and φ2(x)=2+x3\varphi_2(x) = \frac{2 + x}{3}, increasing affine bijections of [0,1]\intcc{0}{1} onto [0,13]\intcc{0}{\frac13} and [23,1]\intcc{\frac23}{1}. Claim: Cn+1=φ0(Cn)φ2(Cn)C_{n+1} = \varphi_0(C_n) \cup \varphi_2(C_n). For n=0n = 0 this is question 1. Induction: an increasing affine map sends the middle third of a segment to the middle third of the image segment, so deleting middle thirds commutes with φ0\varphi_0 and φ2\varphi_2; applying the deletion step to Cn+1=φ0(Cn)φ2(Cn)C_{n+1} = \varphi_0(C_n) \cup \varphi_2(C_n) yields Cn+2=φ0(Cn+1)φ2(Cn+1)C_{n+2} = \varphi_0(C_{n+1}) \cup \varphi_2(C_{n+1}). Intersecting over nn: for x13x \leq \frac13, xC    xφ0(Cn)x \in C \iff x \in \varphi_0(C_n) for all nn     3xCn=C\iff 3x \in \bigcap C_n = C; likewise on [23,1]\intcc{\frac23}{1}; and no point of (13,23)\intoo{\frac13}{\frac23} lies in C1C_1. Hence C=φ0(C)φ2(C)C = \varphi_0(C) \cup \varphi_2(C), disjointly.

6. Induction on nn; the case n=0n = 0 says every x[0,1]x \in \intcc{0}{1} has a code, which is Problem 10.1 (question 9 for x<1x < 1; 1=(0.2)31 = (0.\overline 2)_3). Suppose the equivalence at rank nn. If xCn+1x \in C_{n+1}: by question 5, x=φi(z)x = \varphi_i(z) with zCnz \in C_n and i{0,2}i \in \{0, 2\}; if (ek)(e_k) is a code of zz with first nn digits 11-free, then (i,e1,e2,)(i, e_1, e_2, \dots) has partial sums i3+13kmek3kφi(z)=x\frac i3 + \frac13\sum_{k\leq m} e_k 3^{-k} \to \varphi_i(z) = x: a code of xx with first n+1n + 1 digits 11-free. Conversely, if xx has a code (dk)(d_k) with d1,,dn+1{0,2}d_1, \dots, d_{n+1} \in \{0, 2\}: the shifted string (d2,d3,)(d_2, d_3, \dots) has some value z[0,1]z \in \intcc{0}{1}, its first nn digits are 11-free, and the partial-sum computation read backwards gives x=φd1(z)x = \varphi_{d_1}(z); by induction zCnz \in C_n, so xCn+1x \in C_{n+1} by question 5. Finally: a fully 11-free code puts xx in every CnC_n, hence in CC; conversely if xCx \in C, then for every nn one of the at most two codes of xx (Problem 10.1, question 11) has its first nn digits 11-free; one fixed code must work for arbitrarily large nn (pigeonhole between two codes), and a code whose first nn digits are 11-free for arbitrarily large nn is 11-free outright.

7. 0=(0.0)30 = (0.\overline 0)_3, 1=(0.2)31 = (0.\overline 2)_3, 13=(0.02)3\frac13 = (0.0\overline{2})_3 (the improper twin of (0.1)3(0.1)_3), 23=(0.20)3\frac23 = (0.2\overline{0})_3. Geometric sums:

(0.02)3=j129j=2/911/9=14,(0.20)3=j1239j1=2/311/9=34,(0.\overline{02})_3 = \sum_{j\geq1} \frac{2}{9^{\,j}} = \frac{2/9}{1 - 1/9} = \frac14 , \qquad (0.\overline{20})_3 = \sum_{j\geq1} \frac{2}{3\cdot 9^{\,j-1}} = \frac{2/3}{1 - 1/9} = \frac34 ,

both 11-free: 14,34C\frac14, \frac34 \in C. (The infinite sums abbreviate suprema of partial sums, as in Problem 10.1.) Endpoints of segments of CnC_n are of the form m/3nm/3^n (induction: children endpoints are parent endpoints or differ from one by a multiple of 3n13^{-n-1}). If 14=m3n\frac14 = \frac{m}{3^n} then 3n=4m3^n = 4m, and 43n4 \nmid 3^n: impossible. So 14C\frac14 \in C without ever being an endpoint.

8. Suppose xx had two distinct 11-free codes. Having two codes at all means (Problem 10.1, question 11, base 33) that x=m/3N(0,1)x = m/3^N \in \intoo{0}{1} and the two codes are: the terminating one, with last nonzero digit dN{1,2}d_N \in \{1, 2\} followed by 00s, and its twin, with dN1d_N - 1 at position NN followed by 22s. If dN=1d_N = 1 the first contains a 11; if dN=2d_N = 2 the twin carries dN1=1d_N - 1 = 1. Either way at most one of the pair is 11-free: contradiction. So each xCx \in C has exactly one 11-free code (existence by question 6), and distinct {0,2}\{0,2\}-strings have distinct values. Every {0,2}\{0,2\}-string has value in [0,1]\intcc{0}{1} (partial sums 1\leq 1) with all prefixes 11-free, hence value in every CnC_n, i.e. in CC: the value map is a bijection from {0,2}\{0,2\}-strings onto CC.

9. Let (d(k))(d^{(k)}) be the 11-free code of xkx_k and set ek=2dk(k){0,2}e_k = 2 - d^{(k)}_k \in \{0, 2\}: a {0,2}\{0,2\}-string whose value yy lies in CC and has (ek)(e_k) as its unique 11-free code (question 8). For each kk the codes of yy and xkx_k differ at position kk, so yxky \neq x_k: no map NC\N^* \to C is surjective. An uncountable set of length zero: bigness in cardinality, smallness in measure — simultaneously.

10. Compactness: closedness of the infinite intersection plus boundedness (question 2) — the one property not carried by a single containment. Length zero: CCnC \subseteq C_n with total length (23)n(\frac23)^n (question 3). Nowhere density: CCnC \subseteq C_n forces intervals inside CC to have length 3n\leq 3^{-n} (question 4). Uncountability rides on no containment at all: it needs the full intersection structure, encoded in the bijection of question 8.

11. Flip dnd_n to 2dn2 - d_n: the new string is still a {0,2}\{0,2\}-string, so its value xnx_n lies in CC; the partial sums beyond rank nn differ by exactly 23n2\cdot3^{-n}, so xnx=23n\abs{x_n - x} = 2\cdot3^{-n}. Thus xnxx_n \neq x and xnxx_n \to x: every point of CC is a limit of other points of CCCC is perfect, with no isolated point.

12. By the induction of question 5, the segments of CnC_n are exactly the [t,t+3n]\intcc{t}{t + 3^{-n}} where tt runs over the values of length-nn {0,2}\{0,2\}-strings. Given xCx \in C with code (dk)(d_k), the truncation tnt_n (digits d1dnd_1 \dots d_n then 00s) is therefore a left endpoint, and 0xtn3n0 \leq x - t_n \leq 3^{-n}: endpoints are dense in CC. They form a subset of {m/3n:m,n}\{m/3^n : m, n\}, a set indexed by pairs of integers, hence countable (as for Q\Q in Exercise 12.10). Since CC is uncountable (question 9), all but countably many points of CC are not endpoints — 14\frac14 (question 7) is the visible tip of that iceberg.

13. Pick nn with 3n<yx3^{-n} < y - x. Both x,yCnx, y \in C_n, and they cannot lie in the same segment (length 3n<yx3^{-n} < y - x): the removed gap between their segments provides zz with x<z<yx < z < y and zCnCz \notin C_n \supseteq C. Hence any two points of CC are separated by the complement: the only convex subsets of CC are singletons — CC is totally disconnected.

14. If (dk)(d_k) is the 11-free code of xx, the string (2dk)(2 - d_k) is again a {0,2}\{0,2\}-string, with partial sums

k=1n(2dk)3k=(13n)k=1ndk3k1x:\sum_{k=1}^{n} (2 - d_k)3^{-k} = (1 - 3^{-n}) - \sum_{k=1}^n d_k 3^{-k} \longrightarrow 1 - x :

so 1xC1 - x \in C. Thus 1CC1 - C \subseteq C, and applying the map twice gives 1C=C1 - C = C: the Cantor set is symmetric about 12\frac12.

15. The partial sums tnxt_n \to x and tnxt'_n \to x' (increasing sequences converge to their supremum, i.e. the value), so tn+tnx+xt_n + t'_n \to x + x' by Theorem 11.5. Given y[0,1]y \in \intcc{0}{1} with code (ek)(e_k), choose (ak,bk)=(0,0),(0,2),(2,2)(a_k, b_k) = (0,0), (0,2), (2,2) according as ek=0,1,2e_k = 0, 1, 2: then ak+bk=2eka_k + b_k = 2e_k, the strings (ak)(a_k), (bk)(b_k) are {0,2}\{0,2\}-strings with values x,xCx, x' \in C, and

x+x=limn(tn+tn)=limn2k=1nek3k=2y:x + x' = \lim_n\,(t_n + t'_n) = \lim_n 2\sum_{k=1}^n e_k 3^{-k} = 2y :

every y[0,1]y \in \intcc{0}{1} is the midpoint of two points of CC.

16. Question 15 gives [0,2]=2[0,1]C+C\intcc{0}{2} = 2\,\intcc{0}{1} \subseteq C + C, and C+C[0,1]+[0,1]=[0,2]C + C \subseteq \intcc{0}{1} + \intcc{0}{1} = \intcc{0}{2}: equality. Then, using 1C=C1 - C = C:

CC=C+(C1)=(C+C)1=[1,1].C - C = C + (C - 1) = (C + C) - 1 = \intcc{-1}{1} .

Concrete instance: 1=14+341 = \frac14 + \frac34, a sum of two non-endpoint members of CC.

17. Length measures how much of the line the set itself occupies; it says nothing about the set of sums, which is the image of the two-parameter family C×CC \times C under (x,x)x+x(x, x') \mapsto x + x' — the two digit strings are chosen independently, and that freedom is exactly what fills [0,2]\intcc{0}{2}. No theorem bounds the length of a sumset by the lengths of the summands, and CC is the proof that none can.

18. By Problem 10.1 (question 18), xx is rational iff its proper expansion is eventually periodic. The 11-free code of xCx \in C is either that proper expansion or the improper twin of a terminating one; a terminating string and its twin (eventually constant 22s) are both eventually periodic, so periodicity of the 11-free code is equivalent to rationality of xx. Long division of 113\frac1{13} in base 33 (r0=1r_0 = 1): 3=130+33 = 13\cdot0 + 3, 9=130+99 = 13\cdot0 + 9, 27=132+127 = 13\cdot2 + 1, and the remainder returns to 11: digits 002\overline{002}, so 113=(0.002)3\frac1{13} = (0.\overline{002})_3, 11-free and periodic: a rational member of CC. (Check: 2/2711/27=226=113\frac{2/27}{1 - 1/27} = \frac{2}{26} = \frac1{13}.)

19. The string with dk=2d_k = 2 at the triangular positions k=j(j+1)2k = \frac{j(j+1)}{2} and 00 elsewhere is a {0,2}\{0,2\}-string, so its value xx^* belongs to CC (question 8). It has infinitely many 22s with gaps j+1j + 1 \to \infty between consecutive ones, so it is not eventually periodic (a period TT would eventually force 22s at gaps T\leq T: the growing-gaps argument of Problem 10.1, question 20); by question 18, xQx^* \notin \Q. And by question 9 plus the countability of Q\Q, all but countably many members of CC are irrational: xx^* is the norm, not the exception.

20. Let y[0,1]y \in \intcc{0}{1}: it has a binary code (ck)(c_k) with ck{0,1}c_k \in \{0, 1\} (Problem 10.1, question 9, base 22; y=1y = 1 takes the all-11s string). Then (2ck)(2c_k) is a {0,2}\{0,2\}-string, its value xx lies in CC, and h(x)h(x) is the value of (ck)(c_k), namely yy: hh maps CC onto [0,1]\intcc{0}{1}. If CC were the image of a map from N\N^*, composing with hh would list all of [0,1)\intco{0}{1}, contradicting the diagonal theorem of Problem 10.1 (question 22): CC is uncountable, again. A length-zero set surjecting onto a full segment.

21. By question 5, CC is the disjoint union of φ0(C)\varphi_0(C) and φ2(C)\varphi_2(C), each a translate of the scaled copy 13C\frac13 C. Additivity, scaling and translation invariance give

L(C)=L(φ0(C))+L(φ2(C))=13L(C)+13L(C)=23L(C),L(C) = L(\varphi_0(C)) + L(\varphi_2(C)) = \tfrac13 L(C) + \tfrac13 L(C) = \tfrac23\,L(C),

so 13L(C)=0\frac13 L(C) = 0: L(C)=0L(C) = 0. Self-similarity alone sentences CC to length zero — question 3 merely executed the sentence.

22. A segment of length lnl_n loses a central interval of length 4(n+1)4^{-(n+1)}, leaving two segments of length ln+1=ln4(n+1)2l_{n+1} = \frac{l_n - 4^{-(n+1)}}{2}; from l0=1l_0 = 1, induction confirms ln=2n+124nl_n = \frac{2^n + 1}{2\cdot4^n}: indeed 12(2n+124n14n+1)=2(2n+1)124n+1=2n+1+124n+1\frac12\Bigl(\frac{2^n+1}{2\cdot4^n} - \frac{1}{4^{n+1}}\Bigr) = \frac{2(2^n + 1) - 1}{2\cdot4^{n+1}} = \frac{2^{n+1} + 1}{2\cdot4^{n+1}}, and ln>0l_n > 0 always: the construction never starves. K=KnK = \bigcap K_n is closed and bounded, hence compact; an interval inside KK lies in one segment of KnK_n, of length ln0l_n \to 0: empty interior. Removed length: n02n4(n+1)=14n0(12)n=12\sum_{n\geq0} 2^n \cdot 4^{-(n+1)} = \frac14\sum_{n\geq0}\bigl(\frac12\bigr)^n = \frac12, and each KnK_n has total length 2nln=2n+12n+1>122^n l_n = \frac{2^n + 1}{2^{n+1}} > \frac12. Now let finitely many open intervals have union UKU \supseteq K. By the companion statement of Example 12.20, UKnU \supseteq K_n for some nn; admitting additivity of length on finite unions of intervals, the total length of the covering intervals is at least that of KnK_n, which exceeds 12\frac12. So KK is nowhere dense, yet no cheap cover exists: topological smallness (nowhere dense) and metric smallness (length zero) are genuinely different notions, and KK separates them.

23. Attainment: let d=d(x,F)d = d(x, F) and pick akFa_k \in F with xakd+1k\abs{x - a_k} \leq d + \frac1k: the aka_k are bounded, so Bolzano–Weierstrass (Theorem 11.16) extracts aφ(k)aa_{\varphi(k)} \to a, with aFa \in F (FF closed, Theorem 12.6) and xa=limxaφ(k)=d\abs{x - a} = \lim \abs{x - a_{\varphi(k)}} = d. Now the maximum: if yCy \in C, d(y,C)=0d(y, C) = 0; otherwise yy lies in a gap removed at some stage n1n \geq 1, an open interval of length 3n3^{-n} whose two endpoints belong to CC (question 2), so d(y,C)3n216d(y, C) \leq \frac{3^{-n}}{2} \leq \frac16, with equality requiring n=1n = 1 and yy at the center of the gap (13,23)\intoo{\frac13}{\frac23}, i.e. y=12y = \frac12; and indeed d(12,C)=16d\bigl(\frac12, C\bigr) = \frac16 since C(13,23)=C \cap \intoo{\frac13}{\frac23} = \emptyset and 13,23C\frac13, \frac23 \in C. Hence maxy[0,1]d(y,C)=16\max_{y\in\intcc{0}{1}} d(y, C) = \frac16, attained exactly at 12\frac12.

24. The endpoints form a countable dense subset of CC (question 12): list them as a single sequence (ej)j1(e_j)_{j\geq1}, a sequence in CC. Its subsequential limits all lie in CC (CC closed). Conversely, fix xCx \in C: for each nn, the segments of the CmC_m containing xx (mnm \geq n) have their endpoints within 3m3n3^{-m} \leq 3^{-n} of xx, so infinitely many distinct endpoints lie within 3n3^{-n} of xx; choose indices j1<j2<j_1 < j_2 < \dots with ejnx3n\abs{e_{j_n} - x} \leq 3^{-n}: a subsequence converging to xx. So the set of subsequential limits of (ej)(e_j) is exactly CC — one sequence clustering at uncountably many points, the opposite extreme from a convergent sequence, whose cluster set is a singleton.

25. (i) The construction consumed: stability of closed sets under arbitrary intersection (existence of CC as a closed set), the compactness theorem Theorem 12.19 (questions 2, 22, 23), and the sequential characterizations of closedness and adherence (the nested-compacts argument of Example 12.20 and question 23). (ii) The four pairings: length zero yet uncountable (questions 3, 9); closed yet with empty interior (question 4); perfect — no isolated point — yet totally disconnected (questions 11, 13); negligible yet with C+C=[0,2]C + C = \intcc{0}{2} (question 16). (iii) The surjection hh of question 20, made continuous and nondecreasing, becomes the devil’s staircase in the theory of continuous functions; and in the Year 3 volume’s measure theory, CC is the standard witness that “negligible” does not mean “countable”, with its fat cousin (question 22) separating “nowhere dense” from “negligible”.