Bourbaki

5. Baire spaces

1. Nowhere dense sets

Definition 1

A subset AA of a topological space XX is said to be nowhere dense if its closure has no interior points.

Equivalently, AA is nowhere dense in XX if and only if the exterior of AA is dense in XX.

A closed set AA is nowhere dense if and only if it has no interior points; that is, if and only if it coincides with its frontier. An arbitrary subset AA is nowhere dense if and only if the closure of AA is nowhere dense. Every subset of a nowhere dense set is nowhere dense.

Example 1

The empty subset of XX is nowhere dense. In a Hausdorff space XX, a set consisting of a single point is nowhere dense if and only if the point is not isolated in XX. A dense subset is never nowhere dense (unless X=∅X = \emptyset).

Example 2

The frontier of an open or of a closed set is always nowhere dense.

Example 3

In the space Rn\mathbf{R}^n, every linear subspace of dimension p<np < n is a nowhere dense set (Chapter VI, § 1, no. 4, Proposition 2).

Remark

The frontier of an arbitrary subset need not be nowhere dense: for example, if AA and C‾A\overline{C}A are both dense sets, then the frontier of AA is the whole space.

Proposition 1

The union of a finite number of nowhere dense sets is nowhere dense.

It is enough to show that the union of two nowhere dense sets A,BA, B is nowhere dense, and without loss of generality we may assume that AA and BB are closed. The proposition is then equivalent to saying that the intersection of two dense open sets C‾A,C‾B\overline{C}A, \overline{C}B is dense. Now if UU is a non-empty open set, then U∩C‾AU \cap \overline{C}A is open and non-empty, hence

(U∩C‾A)∩C‾B=U∩(C‾A∩C‾B) (U \cap \overline{C}A) \cap \overline{C}B = U \cap (\overline{C}A \cap \overline{C}B)

is open and non-empty.

Let YY be a subspace of a topological space XX. A subspace AA of YY is said to be nowhere dense relative to YY if AA is nowhere dense when considered as a subset of the topological space YY.

Proposition 2

Let YY be a subspace of a topological space XX, and let AA be a subset of YY. If AA is nowhere dense relative to YY, then AA is nowhere dense relative to XX. Conversely, if YY is open in XX and AA is nowhere dense relative to XX, then AA is nowhere dense relative to YY.

Suppose that AA is nowhere dense relative to YY. If the closure A‾\overline{A} of AA in XX contains a non-empty open set UU, then U∩AU \cap A is not empty (by the definition of closure); hence U∩YU \cap Y is a non-empty open set relative to YY, and is contained in the closure A‾∩Y\overline{A} \cap Y of AA with respect to YY, which is contrary to the hypothesis.

Now suppose that YY is open in XX and that A⊂YA \subset Y is nowhere dense relative to XX. If UU is open in YY and is not empty, then UU is open in XX and therefore contains a non-empty set VV which is open in XX (and a fortiori in YY) and does not meet AA; hence AA is nowhere dense relative to YY.

The second part of Proposition 2 is clearly not valid if YY is not open in XX; consider, e.g., the situation where Y≠∅Y \neq \emptyset is nowhere dense in XX, and A=YA = Y.

2. Meagre sets

Definition 2

A subset* AA of a topological space XX *is said to be meagre if it is the union of a countable family of nowhere dense sets.

Equivalently, AA is meagre if it is contained in a countable union of closed sets each of which has no interior points.

A meagre set can perfectly well be dense in XX; even the whole space XX can itself be a meagre set.

An example of the latter possibility is provided by any countable Hausdorff space with no isolated points, e.g., the rational line Q\mathbf{Q}. But a topological space XX which is a meagre set in XX need not be countable (see Exercise 9).

Every subset of a meagre set in a space XX is meagre, and the union of a countable family of meagre sets is meagre.

Let YY be a subspace of XX. A subset AA of YY is said to be meagre relative to YY if AA is meagre when considered as a subset of the topological space YY. It follows from Proposition 2 of no. 1 that if AA is a subset of YY which is meagre relative to YY, then AA is meagre relative to XX; and that if also YY is open in XX, every subset AA of YY which is meagre relative to XX is meagre relative to YY.

3. Baire spaces

Definition 3

A topological space* XX *is said to be a Baire space if it satisfies one or the other of the following two equivalent conditions:

(EB) Every countable intersection of dense open sets in XX is dense in XX. (EB') Every countable union of closed sets with no interior points in XX has no interior point in XX.

Axiom (EB) can be stated in two other equivalent forms:

(EB'') Every non-empty open set in XX is non-meagre.

Indeed a set is meagre if and only if it is contained in a countable union of closed sets with no interior points.

(EB'') The complement of a meagre set in XX is dense in XX.

This signifies that a meagre set cannot contain a non-empty open set, and is therefore equivalent to (EB").

Proposition 3

Every non-empty open subspace YY of a Baire space XX is a Baire space.

This follows from (EB"), for every open (resp. meagre) set in YY is open (resp. meagre) in XX.

It follows from Proposition 3 that every point of a Baire space has a fundamental system of neighbourhoods, each of which is a Baire space. Conversely:

Proposition 4

If every point of a topological space XX has a neighbourhood which is a Baire space, then XX is a Baire space.

Let AA be a non-empty open set in XX, let x∈Ax \in A and let VV be an open neighbourhood of xx which is a Baire space. If AA were meagre in XX, then V∩AV \cap A would be meagre in VV and open in VV, which is contrary to hypothesis.

Proposition 5

In a Baire space XX, the complement of a meagre set is a Baire space.

Let AA be a meagre set in XX; then its complement Y=∁AY = \complement_A in XX is dense in XX. Let BB be a meagre set relative to YY; BB is also meagre relative to XX, hence A∪BA \cup B is meagre relative to XX. Hence the complement of A∪BA \cup B in XX, which is also the complement of BB in YY, is dense in XX and a fortiori dense in YY. Hence YY is a Baire space.

Theorem 1 (Baire)

(i) Every locally compact space XX is a Baire space. (ii) *Every topological space XX on which there exists a metric, compatible with the topology of XX and which defines on XX the structure of a complete metric space, is a Baire space.

We shall show that axiom (EB) is satisfied in each case. Let (An)(A_n) be a sequence of dense open sets in XX, and let GG be any non-empty open set. We can then define inductively a sequence (Gn)(G_n) of non-empty open sets such that G1=GG_1 = G and G‾n+1⊂Gn∩An\overline{G}_{n+1} \subset G_n \cap A_n: for since by hypothesis GnG_n is not empty, Gn∩AnG_n \cap A_n is a non-empty open set; and since XX is regular in both cases envisaged, there is a non-empty open set Gn+1G_{n+1} such that G‾n+1⊂Gn∩An\overline{G}_{n+1} \subset G_n \cap A_n. Hence the set G∩⋂n=1∞AnG \cap \bigcap_{n=1}^\infty A_n contains the intersection of the sets GnG_n, which is equal to the intersection of the sets G‾n\overline{G}_n; hence it is enough to show that ⋂n=1∞G‾n≠∅\bigcap_{n=1}^\infty \overline{G}_n \neq \varnothing. Now if XX is locally compact, we may assume that G‾2\overline{G}_2 is compact; in the compact space G‾2\overline{G}_2, the G‾n(n≥2)\overline{G}_n (n \geq 2) form a decreasing sequence of non-empty closed sets, and their intersection is therefore not empty by axiom (C'') (cf. Chapter I, § 9, no. 1, Definition 1]. If XX is a complete metric space (with respect to a metric compatible with its topology) we may suppose that G‾n\overline{G}_n has been chosen so that its diameter (with respect to this metric) tends to 0 as nn tends to +∞+\infty; the G‾n\overline{G}_n therefore form a Cauchy filter base which converges to a point xx, and xx necessarily belongs to the intersection of the sets G‾n\overline{G}_n.

Q.E.D.

Remark

There are Baire spaces which belong to neither of these two categories, in particular Baire spaces which are neither metrizable nor locally compact (Exercise 16); there are also metrizable Baire spaces which possess no complete metric space structure compatible with their topology (Exercise 14).

4. Semi-continuous functions on a Baire space

Theorem 2

Let XX be a Baire space and let (fα)(f_\alpha) be a family of lower semi-continuous real-valued functions on XX such that, at every point xx of XX, the upper envelope sup⁡αfα(x)\sup_\alpha f_\alpha(x) is finite. Then every non-empty open set in XX contains a non-empty open set on which the family (fα)(f_\alpha) is uniformly bounded above.

This theorem may also be stated in the form that the set of points in the neighbourhood of which the family (fα)(f_\alpha) is uniformly bounded above is a dense open set.

Let f=sup⁡αfαf = \sup_\alpha f_\alpha be the upper envelope of the family (fα)(f_\alpha). The function ff is lower semi-continuous (Chapter IV, § 6, no. 2, Theorem 4) and finite at every point of XX. It is therefore enough to carry through the proof in the case where the family (fα)(f_\alpha) consists of a single function ff. Let AnA_n be the set of points x∈Xx \in X such that f(x)≤nf(x) \leq n; AnA_n is closed (Chapter IV, § 6, no. 2, Proposition 1), and the assumptions imply that XX is the union of the sets AnA_n; hence at least one of the AnA_n has an interior point, and therefore there is a non-empty open set on which ff is bounded above (by an integer nn). If we apply this result to any non-empty open subset of XX (this subspace is also a Baire space by Proposition 3 of no. 3), we have the theorem.

In applications of this theorem it is generally the case that the fαf_\alpha are continuous on XX.

Remark

The theorem may be false if we do not suppose that XX is a Baire space. For example if, for each rational number p/qp/q (p,qp, q being coprime integers, q>0q > 0) we put f(p/q)=qf(p/q) = q, we have a lower semi-continuous function ff on the rational line QQ which is finite at each point (cf. Chapter IV, § 6, no. 2); but there is no non-empty open set in QQ on which ff is bounded above.

Exercises

See the exercises for § 5.

Transcribed from the printed volume. Not translated.