math.wiki
Topology › Connectedness and compactness › Hausdorff spaces and separation
Sign in

Hausdorff spaces and separation

On the line a sequence cannot approach two different numbers, because two numbers have disjoint intervals around them. Which topologies keep that property, and what goes wrong in a space where two points cannot be pulled apart by open sets?

Where you are 5 in Connectedness and compactness

Why this exists

In Metric spaces, a first look a sequence has at most one limit, and a finite set is closed (Theorem 5 of Closed sets, closure and limit points). Both facts rest on one property of the metric: two distinct points sit inside disjoint balls. A general topology need not have it. In the finite-complement topology on an infinite set, any two nonempty open sets meet, and a sequence of distinct points approaches every point at once (Example 2).

This lesson names the property, defines convergence in any topological space, and proves that the property makes limits unique.

Definition

Throughout, (X, \mathcal{T}) is a topological space, and a neighborhood of x is an open set containing x, as in Topologies and open sets.

Hausdorff space. X is Hausdorff if for every two distinct points x, y \in X there are open sets U and V with x \in U, y \in V and U \cap V = \varnothing. Such U and V separate x and y. A Hausdorff space is also called a T_2-space.

T_1-space. X is a T_1-space if every one-point set \{x\} is closed.

Convergence. A sequence (x_n)_{n \ge 1} of points of X converges to x \in X, written x_n \to x, if for every neighborhood U of x there is N with x_n \in U for every n > N. Then x is a limit of the sequence; the article “a” is deliberate, as Example 2 shows.

It is enough to test the basis elements containing x, since every neighborhood of x contains one (Theorem 2 of Bases and subbases). In a metric space the balls B(x, \varepsilon) suffice, so the definition agrees with that of Metric spaces, a first look.

Intuition

Figure 1 shows the metric case. If x_n \to x, the sequence eventually stays in U; it cannot also eventually stay in V, which shares no point with U. So it cannot converge to y.

x y
U
V
Figure 1Theorem 1. The open balls U = B(x, r) and V = B(y, r) with \displaystyle r = \frac{1}{2} d(x, y). The dashed circles are not part of the balls, so the point where they touch lies in neither, and U \cap V = \varnothing.

When two points cannot be separated, a sequence can live in the overlaps of their neighborhoods and approach both (Example 6). The T_1 condition asks for less: an open set around each point that misses the other point.

Results

Theorem 1 (metric spaces). Every metric space is Hausdorff in its metric topology (Theorem 5 of Topologies and open sets): if x \ne y and \displaystyle r = \frac{1}{2} d(x, y), then B(x, r) \cap B(y, r) = \varnothing.

Theorem 2 (comparing topologies). Let \mathcal{T} \subseteq \mathcal{T}' be topologies on X. (a) If (X, \mathcal{T}) is Hausdorff, so is (X, \mathcal{T}'), and likewise for T_1. (b) If x_n \to x in \mathcal{T}', then x_n \to x in \mathcal{T}.

Theorem 3 (T_1-spaces). (a) X is T_1 if and only if for every two distinct points x, y there is an open set containing y but not x. (b) Every Hausdorff space is T_1. (c) In a T_1-space every finite set is closed, and x is a limit point of A if and only if every neighborhood of x contains infinitely many points of A.

Theorem 4 (uniqueness of limits). In a Hausdorff space a sequence has at most one limit.

Theorem 5 (sequences and closure). In any space, if a_n \in A for every n and a_n \to x, then x \in \overline{A}.

Theorem 6 (a point and a finite set). Let X be Hausdorff, F \subseteq X finite and x \notin F. Then there are disjoint open sets U and V with x \in U and F \subseteq V.

Theorem 7 (finite spaces). A finite T_1-space is discrete. In particular, a finite Hausdorff space is discrete.

Why it is true

Theorem 1. Proof. Balls are open (Theorem 2(a) of Metric spaces, a first look). If z \in B(x, r) \cap B(y, r), the triangle inequality gives d(x, y) \le d(x, z) + d(z, y) < 2r = d(x, y), which is false. \square

Theorem 2. Proof. (a) Open sets separating x and y in \mathcal{T} are open in \mathcal{T}', and a \mathcal{T}-closed set is \mathcal{T}'-closed (Theorem 1(d) of Closed sets, closure and limit points). (b) A \mathcal{T}-neighborhood of x is a \mathcal{T}'-neighborhood, so the sequence eventually lies in it. \square

Theorem 3. Proof. (a) If \{x\} is closed, X \setminus \{x\} is open, contains every y \ne x, and misses x. Conversely, if each y \ne x lies in an open V_y with x \notin V_y, then X \setminus \{x\} = \bigcup_{y \ne x} V_y is open, so \{x\} is closed.

(b) If U and V separate x and y, then V contains y but not x; apply (a).

(c) A finite set is \varnothing or a finite union of closed one-point sets, closed by Theorem 1 of Closed sets, closure and limit points. If every neighborhood of x holds infinitely many points of A, it holds one other than x, so x is a limit point. Conversely, if a neighborhood U of x holds only finitely many points of A, let F be those other than x. Then F is closed, and U \setminus F is a neighborhood of x meeting A at most in x, so x is not a limit point. \square

Theorem 4. Proof. Suppose x_n \to x and x_n \to y with x \ne y, and let U, V separate them. There are N_1 with x_n \in U for n > N_1, and N_2 with x_n \in V for n > N_2. For n > \max\{N_1, N_2\}, x_n \in U \cap V = \varnothing, a contradiction. \square

Theorem 5. Proof. Every neighborhood of x contains a_n for all large n, so it meets A, and x \in \overline{A} by Theorem 3 of Closed sets, closure and limit points. \square

Theorem 6. Proof. If F = \varnothing, take U = X and V = \varnothing. Otherwise, for each y \in F choose disjoint open sets U_y \ni x and V_y \ni y. Then U = \bigcap_{y \in F} U_y is open, being a finite intersection, and contains x; V = \bigcup_{y \in F} V_y is open and contains F. A point of U \cap V would lie in some V_y and in U_y, which are disjoint. \square

Theorem 7. Proof. Every subset of a finite T_1-space is finite, so closed by Theorem 3(c). Every subset is the complement of a subset, so every subset is open. The Hausdorff case follows by Theorem 3(b). \square

Worked examples

Example 1

Polders and pumping stations

Priya manages five polders a, b, c, d, e, low fields kept dry by pumps. Three pumping stations each drain a group of them: W_1 = \{a, b, c\}, W_2 = \{c, d\} and W_3 = \{a, e\}. Take the three groups as a subbasis on the set P of polders. (a) Which pairs of polders can be separated? Is P Hausdorff? Is it T_1? (b) Hiro proposes one more station, draining a group Z \subseteq P, added to the subbasis. For which Z is the new space Hausdorff?

Solution

(a) As in Example 1 of Closed sets, closure and limit points, the smallest open set M_x containing x is the intersection of the groups containing x:

PolderStations draining itM_x
aW_1, W_3\{a\}
bW_1\{a, b, c\}
cW_1, W_2\{c\}
dW_2\{c, d\}
eW_3\{a, e\}

Every neighborhood of x contains M_x, and M_x is one, so x and y can be separated exactly when M_x \cap M_y = \varnothing. The disjoint pairs are \{a, c\}, \{a, d\}, \{c, e\} and \{d, e\}, four of the ten. P is not Hausdorff.

By Theorem 3 of Closed sets, closure and limit points, y \in \overline{\{x\}} exactly when x \in M_y. So \overline{\{a\}} = \{a, b, e\} and \overline{\{c\}} = \{b, c, d\}, while \{b\}, \{d\} and \{e\} are closed. P is not T_1.

(b) By Theorem 7 the new space is Hausdorff exactly when it is discrete, that is, when every smallest open set is a single point. Adding Z replaces M_x by M_x \cap Z for x \in Z and leaves the others alone. M_a = \{a\} and M_c = \{c\} already. For b we need b \in Z and M_b \cap Z = \{b\}, so a, c \notin Z; for d, d \in Z; for e, e \in Z. Hence

Z = \{b, d, e\},

and this Z works.

Example 2

The finite-complement topology

Let X be an infinite set with the finite-complement topology \mathcal{T}_f of Topologies and open sets, whose open sets are \varnothing and the sets with finite complement. Show that X is T_1 but not Hausdorff, and that every sequence of distinct points converges to every point of X. Apply this to x_n = 3n in \mathbb{R}.

Solution

The closed sets are X and the finite sets, so every \{x\} is closed: X is T_1. Let U, V be nonempty open sets. Then X \setminus (U \cap V) = (X \setminus U) \cup (X \setminus V) is finite, while X is infinite, so U \cap V \ne \varnothing. No two points can be separated.

Let x_m \ne x_n whenever m \ne n, and let x \in X. A neighborhood U of x has a finite complement F, and each point of F equals x_n for at most one n. So only finitely many n have x_n \notin U; with N the largest of them (N = 0 if there are none), x_n \in U for every n > N. Thus x_n \to x.

In \mathbb{R} with \mathcal{T}_f, the terms 3n are distinct, so the sequence converges to 0, to -4, to \pi, to every real number. Theorem 4 does not apply: \mathcal{T}_f is not Hausdorff.

Example 3

One family of sequences, three topologies

On \mathbb{R}, find every limit of a_n = 2^{-n}, \displaystyle b_n = \frac{(-1)^n}{n} and \displaystyle c_n = \left\lfloor \frac{4}{n} \right\rfloor in the standard topology, in the lower limit topology \mathbb{R}_{\ell}, and in the discrete topology.

Solution

The standard topology is Hausdorff by Theorem 1, and \mathbb{R}_{\ell} and the discrete topology are finer (Theorem 5(b) of Bases and subbases, Theorem 4(a) of Topologies and open sets), so Hausdorff by Theorem 2(a). Each sequence has at most one limit in each topology (Theorem 4), and a limit in a finer topology is a standard limit (Theorem 2(b)).

In the standard topology all three tend to 0: 2^{-n} < \varepsilon once \displaystyle n > \log_2 \frac{1}{\varepsilon}, \displaystyle |b_n| = \frac{1}{n}, and c_n = 4, 2, 1, 1 for n = 1, 2, 3, 4 and c_n = 0 for n \ge 5. So 0 is the only candidate in the other two topologies.

Lower limit. A basis element [p, q) containing 0 contains some [0, \varepsilon), so it suffices to test those. Since 0 < a_n < \varepsilon for large n, a_n \to 0. The odd terms b_n are negative, outside [0, 1), so b_n has no limit. c_n is 0 from n = 5 on, so c_n \to 0.

Discrete. \{0\} is a neighborhood of 0, so a sequence converges to 0 only if its terms are eventually 0. Neither a_n nor b_n ever is; c_n \to 0.

Sequencestandard\mathbb{R}_{\ell}discrete
a_n00none
b_n0nonenone
c_n000

A finer topology has more neighborhoods to pass through, so fewer sequences converge.

Example 4

Arithmetic progressions on the integers

On \mathbb{Z} take the topology generated by the sets a + b\mathbb{Z} with b \ge 1 (Example 6 of Bases and subbases). Show that it is Hausdorff, that n! \to 0, and that 2^n has no limit.

Solution

Let m \ne n and b = |m - n| + 1. If m + bj = n + bk, then b divides m - n, a nonzero integer smaller than b in absolute value, which is impossible. So m + b\mathbb{Z} and n + b\mathbb{Z} separate m and n. For 2 and 7, for instance, 2 + 6\mathbb{Z} and 7 + 6\mathbb{Z} do.

A basis element containing 0 is a + b\mathbb{Z} with b dividing a, which is b\mathbb{Z}. For n \ge b, b is a factor of n!, so n! \in b\mathbb{Z}. Hence n! \to 0, and by Theorem 4 this is its only limit, though n! grows without bound on the line.

Suppose 2^n \to x. The neighborhood x + 3\mathbb{Z} contains 2^n and 2^{n + 1} for all large n, so 3 divides 2^{n + 1} - 2^n = 2^n, which is false. The sequence has no limit.

Example 5

A convergent sequence with its limit

Let X be Hausdorff, x_n \to x, and K = \{x_n : n \ge 1\} \cup \{x\}. Prove that K is closed. Then show that the hypothesis is needed.

Solution

Proof. Let y \notin K; we find a neighborhood of y missing K. Since y \ne x, there are disjoint open sets U \ni x and V \ni y. Choose N with x_n \in U for n > N. The set F = \{x_1, \ldots, x_N\} is finite, so closed (Theorem 3), and y \notin F. So W = V \setminus F is a neighborhood of y. It misses x and every x_n with n > N, which lie in U, and it misses x_1, \ldots, x_N. So W \cap K = \varnothing, and X \setminus K, the union of such sets W, is open. \square

In Example 2, 3n \to 0 in (\mathbb{R}, \mathcal{T}_f), but K = \{3n : n \ge 1\} \cup \{0\} is infinite and is not all of \mathbb{R}, so it is not closed there.

Example 6

The line with two origins

Let X = \mathbb{R} \cup \{0^*\}, where 0^* is a new point not in \mathbb{R}. Let \mathcal{B} consist of the open intervals (p, q) and the sets

I^*(p, q) = \big((p, q) \setminus \{0\}\big) \cup \{0^*\}, \qquad p < 0 < q.

Show that \mathcal{B} is a basis on X, that X is T_1 but not Hausdorff, and that \displaystyle x_n = \frac{1}{n} converges to both 0 and 0^*.

0
0^*
U
V
Figure 2The line with two origins of Example 6. A neighborhood U of 0 and a neighborhood V of 0^* share every point of their common interval except the two origins themselves, so they always meet.
Solution

A basis. (B1): a real t lies in (t - 1, t + 1) and 0^* in I^*(-1, 1). (B2): two intervals meet in an interval or not at all; an interval meets I^*(p, q) in an interval with 0 removed, a union of at most two intervals; and I^*(p, q) \cap I^*(p', q') = I^*(\max\{p, p'\}, \min\{q, q'\}). Each intersection is a union of members of \mathcal{B}.

Not Hausdorff. A neighborhood of 0 contains some (p, q) \ni 0, and a neighborhood of 0^* contains some I^*(p', q'). These share (\max\{p, p'\}, \min\{q, q'\}) \setminus \{0\}, which is nonempty (Figure 2). So 0 and 0^* cannot be separated. Every other pair can: two real numbers by disjoint intervals (Theorem 1), and a real t \ne 0 and 0^* by (t - r, t + r) and I^*(-r, r) with \displaystyle r = \frac{|t|}{2}.

T_1. By Theorem 3(a): separated pairs are covered, and (-1, 1) and I^*(-1, 1) each contain one of 0, 0^* but not the other.

Two limits. \displaystyle \frac{1}{n} \to 0 in the standard topology, and the neighborhoods of 0 in X are the same as on the line. A neighborhood of 0^* contains some I^*(p, q), which contains \displaystyle \frac{1}{n} for every \displaystyle n > \frac{1}{q}, since \displaystyle \frac{1}{n} \ne 0. So \displaystyle \frac{1}{n} \to 0^* as well.

The same space is the quotient of two copies of \mathbb{R} glued at every point except 0 (Quotient spaces, or gluing).

Boundaries

  • T_1 is strictly weaker than Hausdorff (Examples 2 and 6), and a space that is not T_1 may still separate some pairs (Example 1).
  • Unique limits do not make a space Hausdorff. In the cocountable topology on \mathbb{R}, every sequence has at most one limit, yet no two points can be separated (Exercise 14).
  • Sequences need not detect closure. The converse of Theorem 5 holds in metric spaces (Exercise 11 of Closed sets, closure and limit points) but fails in the cocountable topology, even though it is T_1 (Exercise 14).
  • Hausdorff does not mean metrizable. \mathbb{R}_{\ell} is Hausdorff but comes from no metric; the proof uses countability axioms and is beyond this lesson (Munkres, Topology, section 30).

Where this goes next

Compactness and open covers proves that a compact subset of a Hausdorff space is closed (Theorem 6 is the finite case, Example 5 a special one) and that a continuous bijection from a compact space to a Hausdorff space is a homeomorphism. Subspaces and products of Hausdorff spaces are Hausdorff, with the topologies of Subspace and product topologies. A quotient of a Hausdorff space need not be Hausdorff: the line with two origins is a quotient of two copies of \mathbb{R} (Quotient spaces, or gluing).

Exercises

Finite spaces

  1. On X = \{1, 2, 3, 4\} the open sets are \varnothing, \{1\}, \{2\}, \{1, 2\}, \{1, 2, 3\}, \{1, 2, 4\} and X. (a) Which pairs of points can be separated? (b) Which one-point sets are closed? Is X a T_1-space? (c) Find every limit of the constant sequence x_n = 1, and of x_n = 3.
  2. Samir looks after four polders p, q, r, s, drained by stations serving \{p, q\}, \{q, r, s\} and \{r\}; take these as a subbasis. (a) Which pairs of polders can be separated? (b) Find every group Z such that adding one station draining Z makes the space Hausdorff.

Spaces on the line and the plane

  1. Which of these spaces are Hausdorff, and which are T_1? (a) \mathbb{R} with the upper limit topology, generated by the intervals (p, q] (Exercise 8 of Bases and subbases). (b) \mathbb{R} with open sets \varnothing, \mathbb{R} and the rays (-\infty, a). (c) \mathbb{R} with open sets \varnothing and every set containing 0.
  2. Find every limit of u_n = 5 - 2^{-n} and of v_n = 5 + (-2)^{-n} in the standard topology, in \mathbb{R}_{\ell}, and in the upper limit topology of Exercise 3(a).
  3. In the finite-complement topology on \mathbb{Z}, find every limit of (a) x_n = n^2; (b) y_n, the remainder of n on division by 3; (c) z_n, equal to 7 for even n and to n for odd n.
  4. In the topology of Example 4 on \mathbb{Z}, find every limit of (a) a_n = n! + 5; (b) b_n = (-1)^n.
  5. In \mathbb{R}^2 let x = (1, 2) and y = (6, 14). For each of the metrics d_1, d_2 and d_\infty of Metric spaces, a first look, find the largest r with B(x, r) \cap B(y, r) = \varnothing. Then do the same for the discrete metric.
  6. In \mathbb{R}^2 with d_2, let x = (0, 0) and F = \{(3, 4), (-1, 1)\}. Find the largest r for which U = B(x, r) and V = B\big((3, 4), r\big) \cup B\big((-1, 1), r\big) are disjoint.
  7. In the line with two origins (Example 6): (a) find every limit of \displaystyle \frac{(-1)^n}{n}; (b) find the closure of (0, 1); (c) is \{0, 0^*\} closed?

Proofs

  1. Let \mathcal{B} be a basis for X. Prove that X is Hausdorff if and only if any two distinct points lie in disjoint members of \mathcal{B}.
  2. Let X be Hausdorff and x_1, \ldots, x_k distinct points of X. Prove that there are pairwise disjoint open sets U_1, \ldots, U_k with x_i \in U_i.
  3. Prove that X is Hausdorff if and only if, for every x \in X, the intersection of the closures \overline{U} of all neighborhoods U of x is \{x\}.
  4. Let X be T_1 and A \subseteq X. Prove that the set of limit points of A is closed. Show that this fails in \{1, 2\} with the indiscrete topology.
  5. The cocountable topology on \mathbb{R} has as open sets \varnothing and the sets with countable complement. Prove: (a) it is a topology; (b) it is T_1 but not Hausdorff; (c) x_n \to x if and only if x_n = x for all large n, so limits are unique; (d) 2 \in \overline{(0, 1)}, yet no sequence of points of (0, 1) converges to 2.
Answers
  1. The smallest open sets are M_1 = \{1\}, M_2 = \{2\}, M_3 = \{1, 2, 3\}, M_4 = \{1, 2, 4\}. (a) Only 1 and 2: every other pair has smallest open sets sharing 1 or 2. (b) y \in \overline{\{x\}} exactly when x \in M_y, so \overline{\{1\}} = \{1, 3, 4\}, \overline{\{2\}} = \{2, 3, 4\}, and only \{3\} and \{4\} are closed. X is not T_1. (c) A constant sequence at x converges to y exactly when every neighborhood of y contains x, that is, x \in M_y. The limits of x_n = 1 are 1, 3 and 4; the only limit of x_n = 3 is 3.
  2. The smallest open sets are M_p = \{p, q\}, M_q = \{q\}, M_r = \{r\}, M_s = \{q, r, s\}. (a) \{p, r\} and \{q, r\}, the pairs with disjoint smallest open sets. (b) By Theorem 7 every smallest open set must become a point. M_p \cap Z = \{p\} needs p \in Z, q \notin Z; M_s \cap Z = \{s\} needs s \in Z, r \notin Z. So Z = \{p, s\}, the only choice, and it works.
  3. (a) Both. Each \displaystyle (p, q) = \bigcup_{n} \left(p, q - \frac{1}{n}\right] over \displaystyle n > \frac{1}{q - p}, so the topology is finer than the standard one; apply Theorem 2(a). (b) Neither: every ray containing 5 contains 4, so no open set contains 5 but not 4, and X is not T_1, hence not Hausdorff by Theorem 3(b). (c) Neither: no open set contains 1 but not 0.
  4. All three topologies are Hausdorff and contain the standard one, where both sequences tend to 5, the only candidate (Theorems 2 and 4). u_n < 5, so [5, 6) holds no term: no limit in \mathbb{R}_{\ell}. An upper limit neighborhood of 5 contains some (5 - \varepsilon, 5], which holds u_n once 2^{-n} < \varepsilon: u_n \to 5 there. v_n - 5 is positive for even n and negative for odd n, so [5, 6) misses the odd terms and (4, 5] the even ones: v_n tends to 5 in the standard topology only.
  5. (a) The terms are distinct, so every integer is a limit (Example 2). (b) No limit: given x, choose v \in \{0, 1, 2\} with v \ne x; the neighborhood \mathbb{Z} \setminus \{v\} of x misses y_n for infinitely many n. (c) Only 7. A neighborhood of 7 has finite complement F, and z_n \notin U only for odd n \in F, finitely many. For x \ne 7, the neighborhood \mathbb{Z} \setminus \{7\} misses every even term.
  6. (a) Only 5. A neighborhood of 5 contains some 5 + b\mathbb{Z}, and n! + 5 \in 5 + b\mathbb{Z} for n \ge b; uniqueness is Theorem 4. (b) No limit: if b_n \to x, the neighborhood x + 3\mathbb{Z} contains 1 and -1, so 3 divides 2.
  7. For d_1, d_2, d_\infty the answer is \displaystyle \frac{1}{2} d(x, y): Theorem 1 covers that r, and for larger r the midpoint \displaystyle m = \left(\frac{7}{2}, 8\right) lies in both balls, as \displaystyle x - m = m - y = \frac{1}{2}(x - y) gives \displaystyle d(x, m) = d(m, y) = \frac{1}{2} d(x, y). With x - y = (-5, -12): d_1 = 17, so \displaystyle r = \frac{17}{2}; d_2 = 13, so \displaystyle r = \frac{13}{2}; d_\infty = 12, so r = 6. For the discrete metric, B(x, r) = \{x\} when r \le 1 and is the whole plane when r > 1, so the largest r is 1, more than \displaystyle \frac{1}{2} d(x, y) = \frac{1}{2}.
  8. U \cap V = \varnothing exactly when B(x, r) misses each of the two balls, that is, by Exercise 7 for d_2, when 2r \le 5 and \displaystyle 2r \le \sqrt{2}. The largest r is \displaystyle \frac{\sqrt{2}}{2}.
  9. (a) 0 and 0^*: the terms are nonzero, so every (p, q) \ni 0 and every I^*(p, q) holds them for \displaystyle n > \frac{1}{\min\{-p, q\}}. A real t \ne 0 is not a limit, as \displaystyle \left(t - \frac{|t|}{2}, t + \frac{|t|}{2}\right) misses the terms with \displaystyle \frac{1}{n} < \frac{|t|}{2}. (b) [0, 1] \cup \{0^*\}, by Theorem 3 of Closed sets, closure and limit points with the basis \mathcal{B}: a basis element containing t \in [0, 1], or 0^*, contains some (p, q) \setminus \{0\} with p < t < q (or p < 0 < q), which meets (0, 1); a real t < 0 or t > 1 has an interval about it missing (0, 1). (c) Yes: X is T_1, so finite sets are closed (Theorem 3(c)).
  10. Members of \mathcal{B} are open, so disjoint members separating x and y show X is Hausdorff. Conversely, if U, V separate x, y, there are B, B' \in \mathcal{B} with x \in B \subseteq U and y \in B' \subseteq V (Theorem 2 of Bases and subbases), and B \cap B' \subseteq U \cap V = \varnothing. \square
  11. If k = 1, take U_1 = X. Otherwise, for each pair i \ne j choose disjoint open sets U_{ij} \ni x_i and U_{ji} \ni x_j. Let U_i = \bigcap_{j \ne i} U_{ij}, open as a finite intersection, containing x_i. For i \ne j, U_i \subseteq U_{ij} and U_j \subseteq U_{ji}, which are disjoint. \square
  12. Suppose X is Hausdorff, and let y \ne x with U, V separating x, y. Then V is a neighborhood of y missing U, so y \notin \overline{U} (Theorem 3 of Closed sets, closure and limit points); every \overline{U} contains x, so the intersection is \{x\}. Conversely, given y \ne x, some neighborhood U of x has y \notin \overline{U}. Then V = X \setminus \overline{U} is open, contains y, and misses U \subseteq \overline{U}. \square
  13. Let L be the set of limit points of A, and suppose x \in \overline{L} but x \notin L. A neighborhood U of x contains some y \in L (Theorem 3 of Closed sets, closure and limit points), and y \ne x. As \{x\} is closed, U \setminus \{x\} is a neighborhood of y, so it holds infinitely many points of A (Theorem 3(c)), and U meets A \setminus \{x\}. Every U does, so x \in L, a contradiction. Thus \overline{L} = L is closed. In the indiscrete \{1, 2\} with A = \{1\}, L = \{2\}, whose closure is \{1, 2\}. \square
  14. (a) \varnothing is open and \mathbb{R} has empty complement. A union of open sets, one of them nonempty, has its complement inside that member’s countable complement; for nonempty open U, V, the complement of U \cap V is the union of two countable sets. Both are countable. (b) Points are countable, so closed. Two nonempty open sets meet, since their intersection has countable complement and \mathbb{R} is uncountable. (c) If x_n \to x, the set C of terms other than x is countable, so \mathbb{R} \setminus C is a neighborhood of x; the terms eventually lie in it, so they eventually equal x. The converse is clear, and then x is the only limit. (d) The closed sets are \mathbb{R} and the countable sets, and (0, 1) is uncountable, so \overline{(0, 1)} = \mathbb{R} \ni 2. A sequence in (0, 1) converging to 2 would eventually equal 2, by (c). \square
math.wiki · updated 2026-11-18 · CC-BY-SA-4.0