Bourbaki

2. Fundamental topological properties of the real line

1. Archimedes' axiom

The topological properties of the real line which are the subject of this section are all consequences of the following theorem:

Theorem 1

If xx and yy are any two real numbers > 0, then there is an integer n>0n > 0 such that y<nxy < nx.

There exist two rational numbers p/qp/q and r/sr/s such that 0<p/q<x0 < p/q < x and y<r/sy < r/s, since the open intervals ]0,x0, x and ]y,→y, \to are not empty (§\S I, no. 4, Proposition 4); take nn such that nps>qrnps > qr, and we have y<nxy < nx.

Remark

An axiomatic construction of the theory of real numbers will be found in Chapter V, §\S 2, in which Theorem 1 appears as an axiom; for more details about this axiom, see the Historical Note to Chapter IV.

2. Compact subsets of R\mathbf{R}

Theorem 2 (Borel-Lebesgue)

For a subset of the real line R\mathbf{R} to be compact it is necessary and sufficient that it be closed and bounded.

1) The condition is necessary. Let AA be a compact subset of R\mathbf{R} and let aa be a real number >0> 0. The set AA is closed (Chapter I, §\S 9, no. 3, Proposition 4) and there exists a finite number of points xix_i (1≤i≤n1 \leq i \leq n) of R\mathbf{R} such that AA is contained in the union of the neighbourhoods [xi−a,xi+a][x_i - a, x_i + a] (Chapter I, §\S 9, no. 3). Let bb be the maximum of the numbers ∣xi∣|x_i|; then we have A⊂[−b−a,b+a]A \subset [-b - a, b + a].

2) The condition is sufficient. It is enough to show that every interval [−a,+a][ -a, +a ] (a>0a > 0) is compact, and since this interval is a closed subset of a complete uniform space, it is enough to show that, for each b>0b > 0, we can cover [−a,+a][ -a, +a ] by a finite number of intervals of the form [x−b,x+b][x - b, x + b] (Chapter II, §\S 4, no. 2, Corollary to Theorem 3). Now, let nn be an integer >0> 0 such that a<nba < nb; if x∈[−a,+a]x \in [ -a, +a ] and if mm is the largest integer (positive or negative) such that mb≤xmb \leq x, then we have −n≤m≤n-n \leq m \leq n and mb≤x≤(m+1)bmb \leq x \leq (m + 1)b. Hence the 2n+12n + 1 intervals [(k−1)b,(k+1)b][ (k - 1)b, (k + 1)b ] (−n≤k≤n-n \leq k \leq n) form a covering of the required type.

Corollary 1

A subset of the real line R\mathbf{R} is relatively compact if and only if it is bounded.

Corollary 2

The real line is a locally compact space and is not compact.

Remark

Theorem 2 is often referred to as the "Heine-Borel Theorem"; see the Historical Notes to Chapters II and IV.

3. Least upper bound of a subset of R\mathbf{R}

Theorem 3

Every non-empty subset of the real line which is bounded above (resp. bounded below) has a least upper bound (resp. greatest lower bound).

Let AA be a non-empty subset of R\mathbf{R}, bounded above, and let bb be an upper bound of AA, so that A⊂]←,b]A \subset ]\leftarrow, b]. For each x∈Ax \in A consider the set AxA_x of numbers ⩾x\geqslant x which belong to AA; the sets AxA_x form a filter base B\mathfrak{B} on R\mathbf{R}, since Ay⊂AxA_y \subset A_x if y⩾xy \geqslant x. Let aa be a point of AA. For each x⩾ax \geqslant a belonging to AA, AxA_x is contained in the compact interval [a,b][a, b] and thus the filter base B\mathfrak{B} has a cluster point cc. Since the intervals [x,→[[x, \rightarrow[ are closed, cc belongs to their intersection and therefore cc is an upper bound of AA. But, on the other hand, every upper bound zz of AA is ⩾c\geqslant c, otherwise the neighbourhood ]z,→[]z, \rightarrow[ of cc would not contain any point of AA. Hence cc is the least upper bound of AA.

We can argue similarly for a non-empty set BB bounded below, or else simply remark that −B-B is non-empty and bounded above, and that if cc is the least upper bound of −B-B, then −c-c is the greatest lower bound of BB.

The least upper bound cc of AA can be characterized by the following two properties: (i) For each x∈A,x⩽c∗x \in A, x \leqslant c_*. (ii) For each a<ca < c, there exists x∈Ax \in A such that a<x⩽ca < x \leqslant c.

The least upper bound of a closed set (non-empty and bounded above) belongs to the set and is its greatest element; and the least upper bound of any non-empty subset AA of R\mathbf{R} which is bounded above may be defined as the largest real number in the closure of AA.

4. Characterization of intervals

Proposition 1

A non-empty subset AA of R\mathbf{R} is an interval if and only if, whenever aa and bb are any two points of AA such that a<ba < b, the closed interval [a,b][a, b] is contained in AA*.

The condition is clearly necessary. Conversely, suppose that it is satisfied. If AA is neither bounded above nor below it must be the whole of R\mathbf{R}, for if xx is any point of R\mathbf{R} there are then two points a,ba, b of AA such that a<x<ba < x < b. If AA is bounded above but not below, let kk be its least upper bound; then for any x<kx < k there exist aa and bb in AA such that a<x<b⩽ka < x < b \leqslant k, hence x∈Ax \in A, and therefore AA can only be one of the two intervals ]←,k],]←,k[]\leftarrow, k], ]\leftarrow, k[. The argument is similar in the other cases.

5. Connected subsets of R\mathbf{R}

Theorem 4

A subset AA of R\mathbf{R} is connected if and only if AA is an interval.

i) The condition is necessary. Suppose that AA is connected: if AA consists of a single point, it is an interval. If AA has more than one point, let aa and bb be two points of AA such that a<ba < b; by Proposition 1 of no. 4 it is enough to show that every xx such that a<x<ba < x < b belongs to AA. Now, if x∉Ax \notin A we should have A⊂C{x}A \subset C\{x\}; but C{x}C\{x\} is the union of two disjoint open sets ]←,x[\leftarrow, x[ and ]x,→[]x, \rightarrow[, each of which meets AA, and therefore AA would not be connected, contrary to hypothesis.

2) The condition is sufficient. Let us show first that every compact interval [a,b][a, b] is connected. For each integer n>0n > 0, let V1/nV_{1/n} be the entourage consisting of all pairs (x,y)(x, y) such that ∣x−y∣≤1/n|x - y| \leq 1/n; by Proposition 6 of Chapter II, § 4, no. 4, it is enough to show that every pair of points x,yx, y of [a,b][a, b] can be joined by a V1/nV_{1/n}-chain. Let pp be the greatest integer such that p/n≤xp/n \leq x and let qq be the greatest integer such that q/n≤yq/n \leq y (pp and qq exist by reason of Theorem 1 of no. 1); then p≤qp \leq q. If p=qp = q then y−x<1/ny - x < 1/n and the points xx and yy form a V1/nV_{1/n}-chain. If q>pq > p, put xi=(p+i)/nx_i = (p + i)/n (i=1,2,…,q−pi = 1, 2, \ldots, q - p); we have x1−x≤1/nx_1 - x \leq 1/n, y−xq−p≤1/ny - x_{q-p} \leq 1/n and xi+1−xi=1/nx_{i+1} - x_i = 1/n, hence the points x,x1,x2,…,xq−p,yx, x_1, x_2, \ldots, x_{q-p}, y form a V1/nV_{1/n}-chain joining xx and yy.

If now II is any interval not consisting of a single point, and if aa and bb are any two points of II such that a<ba < b, then the interval [a,b][a, b] is contained in II and is connected, and hence II is connected.

Corollary 1

The real line is a connected and locally connected space.

Corollary 2

The only compact connected subsets of R\mathbf{R} are the bounded closed intervals.

By Theorem 4, a subset of R\mathbf{R} which does not contain any interval consisting of more than one point is totally disconnected; this is true, for example, of the set Q\mathbf{Q} of rational numbers, since the set CQC\mathbf{Q} of irrational numbers is dense in R\mathbf{R}.

Proposition 2

Every non-empty open set in R\mathbf{R} is the union of a countable family of mutually disjoint open intervals.

Let AA be a non-empty open set in R\mathbf{R}. Since R\mathbf{R} is locally connected, every component of AA is a connected open set (Chapter I, § 11, no. 6, Proposition 11) and therefore an open interval by Theorem 4. Any two of these open intervals are disjoint; on the other hand, each of them contains a rational number; hence the set of these intervals has a power less than or equal to that of Q\mathbf{Q}, i.e. is countable.

It follows that every closed set in R\mathbf{R} is the complement of the union of a (finite or infinite) sequence (In)(I_n) of mutually disjoint open intervals. These intervals are said to be contiguous to the closed set under consideration. Conversely, given such a sequence of intervals, the complement of their union is a closed set to which these intervals are contiguous.

Example

Let us define by induction a countable family (In,p)(I_{n,p}) of mutually disjoint open intervals as follows: The integer nn takes all values ⩾0\geqslant 0, and for each value of nn, pp takes the values 1,2,3,…,2n1, 2, 3, \ldots, 2^n. All the intervals In,pI_{n,p} are contained in

A=[0,1], A = [0, 1],
and we take I0,1=]1/3,2/3[I_{0,1} = ]1/3, 2/3[ (the "middle third" of ]0,1[]0, 1[). Suppose now that the 2m+1−12^{m+1} - 1 intervals In,pI_{n,p} have been defined for 0≤n≤m0 \leq n \leq m in such a way that, if JmJ_m is their union, the set A∩∁JmA \cap \complement J_m is the union of 2m+12^{m+1} mutually disjoint closed intervals Km,pK_{m,p} (1≤p≤2m+11 \leq p \leq 2^{m+1}) each of length 13m+1\frac{1}{3^{m+1}}. If Km,p=[a,b]K_{m,p} = [a, b] we then take Im+1,pI_{m+1,p} to be the open interval [a+b−a3,b−a3]\left[ a + \frac{b-a}{3}, \frac{b-a}{3} \right] (the "middle third" of the interval ]a,b[]a, b[), and it is immediately verified that the induction can continue in this way (Fig. 3).

!Diagram showing intervals I_{2,1}, I_{1,1}, I_{2,2}, I_{0,1}, I_{2,3}, I_{1,2}, I_{2,4} and corresponding intervals K_{2,1}, K_{2,2}, K_{2,3}, K_{2,4}, K_{1,1}, K_{1,2}, K_{1,3}, K_{1,4}, K_{0,1}, K_{0,2}

Figure 3.

If K′K' is the complement of the union of the In,pI_{n,p}, the closed set K=A∩K′K = A \cap K' is called Cantor's triadic set. Clearly KK is compact (no. 2, Theorem 2); also KK is totally disconnected. For if KK contained an interval II of length >0> 0, then II would be contained in some interval Km,pK_{m,p}; hence its length would be ≤1/3m+1\leq 1/3^{m+1} for each mm, which is absurd.

6. Homeomorphisms of an interval onto an interval

Theorem 5

Let I be an interval in R. Then a mapping f of I into R is a homeomorphism of I onto f(I) if and only if f is strictly monotonic and continuous on I; and f(I) is then an interval in R.

1) The condition is necessary. Let aa and bb be two points of I such that a<ba < b, and suppose for example that f(a)<f(b)f(a) < f(b). Let us show that ff is strictly increasing on I. First, if a<c<ba < c < b then we must have f(a)<f(c)<f(b)f(a) < f(c) < f(b); if, for example, we had f(a)<f(b)<f(c)f(a) < f(b) < f(c) then the image of the interval [a,c][a, c] under ff would be a connected set (Chapter I, , no. 2, Proposition 4) and would therefore contain the interval [f(a),f(c)][f(a), f(c)]; hence there would exist x∈[a,c]x \in [a, c] such that f(x)=f(b)f(x) = f(b), contrary to the hypothesis that ff is injective

It follows that if xx and yy are any two points of I such that x<yx < y, then f(x)<f(y)f(x) < f(y); for we have f(a)<f(x)<f(b)f(a) < f(x) < f(b) if a<x<ba < x < b, f(a)<f(b)<f(x)f(a) < f(b) < f(x) if b<xb < x, and f(x)<f(a)<f(b)f(x) < f(a) < f(b) if x<ax < a; repeating the argument with a,x,ya, x, y in place of a,b,xa, b, x respectively, we see that f(x)<f(y)f(x) < f(y).

2) The condition is sufficient. Suppose that ff is continuous and strictly monotonic on I (say, strictly increasing): f(I)f(I) is connected and is therefore an interval, and since ff is strictly increasing, ff is a bijective mapping of I onto f(I)f(I). Moreover, the image under ff of an open interval in I is an open interval in f(I)f(I), and therefore (§ 1, no, 4, Proposition 5) ff is a homeomorphism of I onto f(I)f(I).

Remark

The first part of the preceding proof shows in fact that a continuous injective mapping of I into R\mathbf{R} is strictly monotonic; from the second part of the proof it therefore follows that every continuous injective mapping ff of an interval I into R\mathbf{R} is a homeomorphism of I onto f(I)f(I).

Exercises

See the exercises for § 2.

Transcribed from the printed volume. Not translated.