Bourbaki

7. Limits

1. Limit of a filter

Definition 1

Let XX be a topological space and F\mathfrak{F} a filter on XX. A point x∈Xx \in X is said to be a limit point (or simply a limit) of F\mathfrak{F}, if F\mathfrak{F} is finer than the neighbourhood filter B(x)\mathfrak{B}(x) of xx; F\mathfrak{F} is also said to converge (or to be convergent) to xx. The point xx is said to be a limit of a filter base B\mathfrak{B} on XX, and B\mathfrak{B} is said to converge to xx, if the filter whose base is B\mathfrak{B} converges to xx.

This definition, together with Proposition 4 of § 6, no. 3, gives the following criterion:

Proposition 1

A filter base* B\mathfrak{B} on a topological space XX converges to xx if and only if every set of a fundamental system of neighbourhoods of xx contains a set of B\mathfrak{B}.

In accordance with the terminology introduced in § 1, no. 2 we can state Proposition 1 in the following way: B\mathfrak{B} converges to xx if and only if there are sets of B\mathfrak{B} as near as we please to xx.

If a filter F\mathfrak{F} converges to xx, then every filter finer than F\mathfrak{F} also converges to xx, by reason of Definition 1. Likewise, if the topology of XX is replaced by a coarser topology, the neighbourhood filter of xx is replaced by a coarser filter (§\S 2, no. 2, Proposition 3), and therefore F\mathfrak{F} still converges to xx in this new topology.

We can therefore say that the finer the topology, the fewer convergent filters there are in this topology. In particular, in the discrete topology, the only convergent filters are the neighbourhood filters, for these are the trivial ultrafilters on XX (§\S 6, no. 4).

Let Φ\Phi be a set of filters on XX, all of which converge to the same point xx; the neighbourhood filter B(x)\mathfrak{B}(x) is coarser than all the filters of Φ\Phi, hence also coarser than the intersection J\mathfrak{J} of these filters; in other words, J\mathfrak{J} also converges to xx.

Proposition 2

A filter* F\mathfrak{F} on a topological space XX converges to a point xx if and only if every ultrafilter which is finer than F\mathfrak{F} converges to xx.

This is an immediate consequence of the preceding remarks and Proposition 7 of § 6, no. 4.

In general a filter can have several distinct limit points; we shall revert to this question in § 8, no. 1.

2. Cluster point of a filter base

Definition 2

In a topological space* XX, a point xx is a cluster point of a filter base B\mathfrak{B} on XX if it lies in the closure of all the sets of B\mathfrak{B}.

If xx is a cluster point of a filter base B\mathfrak{B}, it is also a cluster point of every equivalent filter base by reason of § 6, no. 3, corollary to Proposition 4; in particular, xx is a cluster point of the filter whose base is B\mathfrak{B}.

Proposition 3

A point x is a cluster point of a filter base B if and only if every set of a fundamental system of neighbourhoods of x meets every set of B.

This follows immediately from the definitions.

This proposition and Corollary 2 to Proposition 1 of § 6, no. 2 show that the property "x is a cluster point of the filter F\mathfrak{F}" is equivalent to the property "there is a filter which is finer than both F\mathfrak{F} and the neighbourhood filter of x". In other words:

Proposition 4

A point x is a cluster point of a filter F\mathfrak{F} if and only if there is a filter finer than F\mathfrak{F} which converges to x.

In particular, every limit point of a filter F\mathfrak{F} is a cluster point of F\mathfrak{F}.

Corollary

An ultrafilter U\mathfrak{U} converges to a point x if and only if x is a cluster point of U\mathfrak{U}.

If x is a cluster point of a filter F\mathfrak{F}, it is also a cluster point of every filter coarser than F\mathfrak{F}; likewise, if we replace the topology of X by a coarser topology, x remains a cluster point of F\mathfrak{F} in the new topology.

The set of cluster points of a filter base B\mathfrak{B} on X is by definition the set ⋂M∈BM‾\bigcap_{M \in \mathfrak{B}} \overline{M}, whence

Proposition 5

The set of cluster points of a filter base on a topological space X is closed in X.

Proposition 6

Let B\mathfrak{B} be a filter base on a subset A of a topological space X. Then every cluster point of B\mathfrak{B} in X belongs to A‾\overline{A}; and conversely every point of A‾\overline{A} is a limit point of a filter on A.

The first assertion is trivial; on the other hand, if x∈A‾x \in \overline{A}, the trace on A of the neighbourhood filter of x in X is a filter on A which evidently converges to x.

Remark

A filter on a topological space need have no cluster points (and a fortiori no limit points); for example, in an infinite discrete space the filter of complements of finite subsets has no cluster points. Spaces in which every filter has a cluster point play an important role in mathematics, and we shall study them in § 9.

3. Limit point and cluster point of a function

Definition 3

Let f be a mapping of a set X into a topological space Y, and let F\mathfrak{F} be a filter on X. A point y∈Yy \in Y is said to be a limit point (or simply a limit) (resp. cluster point) of f with respect to the filter F\mathfrak{F} if y is a limit point (resp. cluster point) of the filter base f(F)f(\mathfrak{F}).

The relation "y is a limit of f with respect to the filter F\mathfrak{F}" is written lim⁡Ff=y\lim_{\mathfrak{F}} f = y, or lim⁡x,Ff(x)=y\lim_{x, \mathfrak{F}} f(x) = y, or lim⁡xf(x)=y\lim_x f(x) = y if there is no risk of confusion.

From Definition 3 and Propositions 1 (no. 1) and 3 (no. 2) we deduce the following criteria:

Proposition 7

A point y∈Yy \in Y is a limit of ff with respect to the filter F\mathfrak{F} if and only if, for each neighbourhood VV of yy in YY, there is a set M∈FM \in \mathfrak{F} such that f(M)⊂Vf(M) \subset V (i.e. f‾(V)∈F\overline{f}(V) \in \mathfrak{F} for each neighbourhood VV of yy).* *A point y∈Yy \in Y is a cluster point of ff with respect to F\mathfrak{F} if and only if for each neighbourhood VV of yy and each M∈FM \in \mathfrak{F} there is a point x∈Mx \in M such that f(x)∈Vf(x) \in V.

Example 1

A sequence of points (xn)n∈N(x_n)_{n \in \mathbf{N}} of a topological space is a mapping n→xnn \to x_n of N\mathbf{N} into XX. In analysis one frequently uses the notions of limit point and cluster point of such a mapping with respect to the Fréchet filter (§ 6, no. 1) on N\mathbf{N}; if yy is a limit of n→xnn \to x_n with respect to the Fréchet filter, yy is said to be a limit of the sequence (xn)(x_n) as n tends to infinity, and we write lim⁡n→∞xn=y\lim_{n \to \infty} x_n = y. A cluster point of the mapping n→xnn \to x_n with respect to the Fréchet filter is called a cluster point of the sequence (xn)(x_n).

Thus a point y∈Xy \in X is a limit (resp. cluster point) of a sequence (xn)(x_n) of points of XX if it is a limit point (resp. cluster point) of the elementary filter associated with (xn)(x_n) (§ 6, no. 8).

The point yy is a limit of a sequence (xn)(x_n) in XX if and only if, for every neighbourhood VV of yy in XX, all but a finite number of the terms of the sequence (xn)(x_n) are in VV, i.e. there is an integer n0n_0 such that xn∈Vx_n \in V for all n≥n0n \geq n_0. Likewise yy is a cluster point of the sequence (xn)(x_n) if and only if, for every neighbourhood VV of yy in XX and every integer n0n_0, there is an integer n≥n0n \geq n_0 such that xn∈Vx_n \in V.

Example 2

More generally, let ff be a mapping of a directed set AA into a topological space XX. If x∈Xx \in X is a limit (resp. cluster point) of ff with respect to the section filter of AA then xx is said to be a limit (resp. cluster point) of ff with respect to the directed set AA, and we write x=lim⁡z∈Af(z)x = \lim_{z \in A} f(z).

If yy is a limit (resp. cluster point) of a mapping f:X→Yf : X \to Y with respect to a filter F\mathfrak{F} on XX, then yy remains a limit (resp. cluster point) of ff with respect to F\mathfrak{F} if we replace the topology of YY by a coarser topology, or if we replace the filter F\mathfrak{F} by a finer (resp. coarser) filter.

Proposition 8

Let ff be a mapping of a set XX into a topological space YY; then y∈Yy \in Y is a cluster point of ff with respect to F\mathfrak{F} if and only if there is a filter G\mathfrak{G} on XX which is finer than F\mathfrak{F} and such that yy is a limit of ff with respect to G\mathfrak{G}.

For if yy is a cluster point of ff with respect to F\mathfrak{F}, and if B\mathfrak{B} is the neighbourhood filter of yy, then f‾−1(B)\overline{f}^{-1}(\mathfrak{B}) is a filter base on XX since every set of f‾−1(B)\overline{f}^{-1}(\mathfrak{B}) meets every set of F\mathfrak{F} (§ 6, no. 6). This remark shows also that there is a filter G\mathfrak{G} on XX which is finer than both F\mathfrak{F} and the filter with base f‾−1(B)\overline{f}^{-1}(\mathfrak{B}) (§ 6, no. 2, Proposition 1, Corollary 2), hence that yy is a limit point of ff with respect to G\mathfrak{G}.

Notice finally that if ff is a mapping of a set XX into a topological space YY, the set of cluster points of ff with respect to a filter F\mathfrak{F} on XX is closed in YY (no. 2, Proposition 5) and possibly empty.

Remark

If y∈Yy \in Y is a limit (resp. cluster point) of a mapping f:X→Yf : X \to Y with respect to a filter F\mathfrak{F} on XX, then yy is also a limit (resp. cluster point) of every function g:X→Yg : X \to Y which has the same germ as ff with respect to F\mathfrak{F} (§ 6, no. 9); yy is said to be a limit (resp. cluster point) of the germ f~\tilde{f} of ff with respect to F\mathfrak{F}.

4. Limits and continuity

Let X,YX, Y be two topological spaces, ff a mapping of XX into YY, B\mathfrak{B} the neighbourhood filter in XX of a point a∈Xa \in X. Instead of saying that y∈Yy \in Y is a limit of ff with respect to the filter B\mathfrak{B} and writing y=lim⁡Bfy = \lim_{\mathfrak{B}} f, we use the special notation

y=lim⁡i∈If(x), y = \lim_{i \in I} f(x),

and we say that yy is a limit of ff at the point aa, or that f(x)f(x) tends to yy as xx tends to aa. Similarly, instead of saying that yy is a cluster point of ff with respect to B\mathfrak{B}, we say that yy is a cluster point of ff at the point aa.

A consideration of the definition of continuity (§ 2, no. 1, Definition 1) Proposition 7 of no. 3 shows that:

Proposition 9

A mapping* ff of a topological space XX into a topological space YY is continuous at a point a∈Xa \in X if and only if lim⁡i∈If(x)=f(a)\lim_{i \in I} f(x) = f(a).

Corollary 1

Let* X,YX, Y be two topological spaces, ff a mapping of XX into YY which is continuous at a point a∈Xa \in X; then, for every filter base B\mathfrak{B} on XX which converges to aa, the filter base f(B)f(\mathfrak{B}) converges to f(a)f(a). Conversely if, for every ultrafilter U\mathfrak{U} on XX which converges to aa, the ultrafilter base f(U)f(\mathfrak{U}) converges to f(a)f(a), then ff is continuous at aa.

The first assertion is an immediate consequence of Proposition 9. To prove the second, suppose that ff is not continuous at aa; then there is a neighbourhood W of f(a)f(a) in Y such that f‾−1(W)\overline{f}^{-1}(W) does not belong to the filter B\mathfrak{B} of neighbourhoods of aa in X. Hence (§ 6, no. 4, Proposition 7) there is an ultrafilter U\mathfrak{U}, finer than B\mathfrak{B}, which does not contain f‾−1(W)\overline{f}^{-1}(W) and therefore contains its complement A=X−f‾−1(W)A = X - \overline{f}^{-1}(W) (§ 6, no. 4, Proposition 5); since f(A)∩W=∅f(A) \cap W = \varnothing, f(U)f(\mathfrak{U}) does not converge to f(a)f(a).

Corollary 2

Let g be a mapping of a set Z into a topological space X, which has a limit a with respect to a filter F\mathfrak{F} on Z; then if the map f:X→Yf : X \to Y is continuous at a, the composition f∘gf \circ g has f(a)f(a) as a limit point with respect to F\mathfrak{F}.

5. Limits relative to a subspace

Let X, Y be two topological spaces, let A be a subset of X, and let a∈Xa \in X be a point of the closure of A (but not necessarily in A). Let F\mathfrak{F} be the trace on A of the neighbourhood filter of aa in X. If ff is a mapping of A into Y, then instead of saying that y∈Yy \in Y is a limit of ff with respect to F\mathfrak{F} and writing y=lim⁡Ffy = \lim_{\mathfrak{F}} f, we write

y=lim⁡x→a,x∈Af(x) y = \lim_{x \to a, x \in A} f(x)

and we say that yy is a limit of f at a, relative to the subspace A, or that f(x)f(x) tends to y as x tends to a while remaining in A. We have then y∈f(A)y \in f(A).

If A=C{a}A = \mathbf{C}\{a\} where aa is not an isolated point of X, then we write

y=lim⁡x→a,x≠af(x) y = \lim_{x \to a, x \neq a} f(x)

instead of y=lim⁡x→a,x∈Af(x)y = \lim_{x \to a, x \in A} f(x).

We make analogous definitions for cluster points.

If ff is the restriction to A of a mapping g:X→Yg : X \to Y, we say that gg has a limit (resp. cluster point) yy relative to A at a point a∈A‾a \in \overline{A}, if yy is a limit (resp. cluster point) of ff at aa, relative to A.

Let B be a subset of A and let a∈Xa \in X be a point of the closure of B; if yy is a limit at aa, relative to A, of a map f:A→Yf : A \to Y, then yy is also a limit of ff at aa, relative to B; the converse is not necessarily true.

But if V is a neighbourhood in X of a point a∈A‾a \in \overline{A} and if ff has a limit yy at aa, relative to V∩AV \cap A, then yy is still a limit of ff at aa, relative to A.

Let aa be a non-isolated point of X, so that aa is in the closure of C{a}\mathbf{C}\{a\}. Then a mapping f:X→Yf : X \to Y is continuous at aa if and only if we have f(a)=lim⁡x→a,x≠af(x)f(a) = \lim_{x \to a, x \neq a} f(x); this follows immediately from the definitions.

6. Limits in product spaces and quotient spaces

Proposition 10

Let XX be a set, let (Yi)i∈I(Y_i)_{i \in I} be a family of topological spaces, and for each i∈Ii \in I let fif_i be a mapping of XX into YiY_i. Let XX be given the coarsest topology T\mathfrak{T} for which the fif_i are continuous. Then a necessary and sufficient condition for a filter F\mathfrak{F} on XX to converge to a∈Xa \in X is that for each i∈Ii \in I the filter base fi(F)f_i(\mathfrak{F}) should converge to fi(a)f_i(a) in YiY_i.

The condition is necessary since the fif_i are continuous (no. 4, Proposition 9, Corollary 1). Conversely, suppose that the condition is satisfied, and let VV be an open neighbourhood of aa in XX. By the definition of T\mathfrak{T} (§ 2, no. 3, Proposition 4) there is a finite subset JJ of II, and for each i∈Ji \in J an open subset UiU_i of YiY_i, such that fi(a)∈Uif_i(a) \in U_i for i∈Ji \in J and such that VV contains the set

⋂i∈Jfi−1(Ui). \bigcap_{i \in J} f_i^{-1}(U_i).
The hypothesis implies that fi−1(Ui)∈Ff_i^{-1}(U_i) \in \mathfrak{F} (no. 3, Proposition 7); since JJ is finite, it follows that
M=⋂i∈Jfi−1(Ui) M = \bigcap_{i \in J} f_i^{-1}(U_i)
belongs to F\mathfrak{F}, and M⊂VM \subset V. This completes the proof.

Corollary 1

A filter F\mathfrak{F} on a product space X=∏i∈IXiX = \prod_{i \in I} X_i converges to a point xx if and only if for each i∈Ii \in I the filter base pr⁡i(F)\operatorname{pr}_i(\mathfrak{F}) converges to pr⁡i(x)\operatorname{pr}_i(x).

Corollary 2

Let f=(fi)f = (f_i) be a mapping of a set XX into a product space Y=∏i∈IYiY = \prod_{i \in I} Y_i. Then ff has a limit y=(yi)y = (y_i) with respect to a filter F\mathfrak{F} on XX if and only if for each i∈Ii \in I fif_i has limit yiy_i with respect to F\mathfrak{F}.

Proposition 11

Let RR be an open equivalence relation on a topological space XX and let φ\varphi be the canonical mapping X→X/RX \to X/R. Then for each x∈Xx \in X and each filter base B′\mathfrak{B}' on X/RX/R which converges to φ(x)\varphi(x), there is a filter base B\mathfrak{B} on XX which converges to xx and is such that φ(B)\varphi(\mathfrak{B}) is equivalent to B′\mathfrak{B}'.

If UU is any neighbourhood of xx in XX, then φ(U)\varphi(U) is a neighbourhood of φ(x)\varphi(x) in X/RX/R (§ 5, no. 3, Proposition 5), hence there is a set M′∈B′M' \in \mathfrak{B}' such that M′⊂φ(U)M' \subset \varphi(U); if we put M=U∩φ−1(M′)M = U \cap \varphi^{-1}(M'), then M′=φ(M)M' = \varphi(M). This shows that as M′M' runs through B′\mathfrak{B}' and UU runs through the neighbourhood filter of xx, the sets U∩φ−1(M′)U \cap \varphi^{-1}(M') form a filter base B\mathfrak{B} on XX; clearly B\mathfrak{B} converges to xx and φ(B)\varphi(\mathfrak{B}) is equivalent to B′\mathfrak{B}'.

Exercises

See the exercises for § 7.

Transcribed from the printed volume. Not translated.