Alessandro Palliccia

Real and Functional Analysis

Separability, linear operators and Baire's lemma

1,618 words, , published

The study of the topology of normed spaces of infinite dimension ends with separability.

Topology in infinite dimensions (conclusion)

Separability

Definition 13.1 (Separable metric space). A metric space (X,d){(X, d)} is said to be separable if there exists D⊂X{D \subset X} such that DD is dense and countable, as in Definition 12.8 and Definition 1.3.

Example 13.2. The set Q⊂R{\mathbb{Q} \subset \mathbb{R}} is dense by Example 12.10, and it is countable by Example 1.4. Hence R,{\mathbb{R},} with the distance d(x,y)=∣x−y∣,{d(x, y) = |x - y|,} is separable.

Remark 13.3. Let (X,d){(X, d)} be a metric space and let D⊂X.{D \subset X.} Then DD is dense if and only if U∩D≠∅{U \cap D \neq \emptyset} for every nonempty open set U⊂X.{U \subset X.}

Proof (sketch). Only the implication from left to right is proved. Let DD be dense, let U⊂X{U \subset X} be open with U≠∅,{U \neq \emptyset,} and take x∈U.{x \in U.} Since UU is open, there exists r>0{r > 0} such that Br(x)⊂U,{B_r(x) \subset U,} by Definition 1.19. Since DD is dense, by Definition 12.8 there exists y∈Br(x)∩D,{y \in B_r(x) \cap D,} and Br(x)∩D⊂U∩D.{B_r(x) \cap D \subset U \cap D.} Hence U∩D≠∅.{U \cap D \neq \emptyset.} ■\blacksquare

Theorem 13.4. Let −∞<a<b<+∞,{-\infty < a < b < +\infty,} let NN be a positive integer and let U⊂RN{U \subset \mathbb{R}^N} be open, with the Lebesgue measure as in Example 10.8 (i).
(1) The space (C([a,b]),∥⋅∥∞){(C([a, b]), \|\cdot\|_\infty)} of Example 10.5 (iii) is separable. The idea is the Stone–Weierstrass theorem in Remark 12.15, by which the polynomials are dense in C([a,b]):{C([a, b]){:}} the polynomials a0+a1x+⋯+akxk{a_0 + a_1 x + \dots + a_k x^k} with ai∈Q{a_i \in \mathbb{Q}} form a dense and countable subset of C([a,b]).{C([a, b]).}
(2) For every p∈[1,+∞),{p \in [1, +\infty),} the space (Lp(U),∥⋅∥p){(L^p(U), \|\cdot\|_p)} is separable.
(3) The space (L∞(U),∥⋅∥∞){(L^\infty(U), \|\cdot\|_\infty)} is not separable.

Proof (sketch). Only (3) is proved, in the case N=1{N = 1} and U=R,{U = \mathbb{R},} that is, for L∞(R).{L^\infty(\mathbb{R}).} For α>0{\alpha > 0} let

fα:=χ(−α,α)∈L∞(R),Uα:=B1/2(fα)⊂L∞(R).\begin{aligned} f_\alpha &:= \chi_{(-\alpha, \alpha)} \in L^\infty(\mathbb{R}), \\ U_\alpha &:= B_{1/2}(f_\alpha) \subset L^\infty(\mathbb{R}). \end{aligned}

If α≠β,{\alpha \neq \beta,} then ∥fα−fβ∥∞=1,{\|f_\alpha - f_\beta\|_\infty = 1,} as in Figure 13.1.

1−β−ααβfαfβ1fαfβUαUβ

Figure 13.1. The drawing of the proof of Theorem 13.4 (3), for the α\alpha and β\beta of the sliders. On the left, the graphs of fαf_\alpha and of fβ,{f_\beta,} dashed, which differ on the hatched set; on the right, the balls UαU_\alpha and Uβ,{U_\beta,} drawn as discs, whose centres are at distance ∥fα−fβ∥∞=1.{\|f_\alpha - f_\beta\|_\infty = 1.}

Hence Uα∩Uβ=∅{U_\alpha \cap U_\beta = \emptyset} for all α≠β,{\alpha \neq \beta,} and F={Uα:α>0}{\mathcal{F} = \{ U_\alpha : \alpha > 0 \}} is an uncountable family of pairwise disjoint nonempty open balls.

Suppose, by contradiction, that there exists D={um:m∈N}⊂L∞(R){D = \{ u_m : m \in \mathbb{N} \} \subset L^\infty(\mathbb{R})} dense and countable. By Remark 13.3, D∩Uα≠∅{D \cap U_\alpha \neq \emptyset} for every α>0.{\alpha > 0.} That is, for every α>0{\alpha > 0} there exists mα∈N{m_\alpha \in \mathbb{N}} such that umα∈Uα,{u_{m_\alpha} \in U_\alpha,} so that there is a function

ϕ ⁣:(0,+∞)→N,α↦mα.\phi \colon (0, +\infty) \to \mathbb{N}, \qquad \alpha \mapsto m_\alpha.

The function ϕ\phi is injective. Let α≠β{\alpha \neq \beta} and suppose that mα=mβ.{m_\alpha = m_\beta.} Then umα=umβ{u_{m_\alpha} = u_{m_\beta}} belongs to UαU_\alpha and to Uβ,{U_\beta,} so that Uα∩Uβ≠∅,{U_\alpha \cap U_\beta \neq \emptyset,} which is impossible. Hence mα≠mβ.{m_\alpha \neq m_\beta.}

Therefore #(0,+∞)≤#N{\#(0, +\infty) \le \#\mathbb{N}} by Definition 1.1, a contradiction. ■\blacksquare

Linear operators

Definition 13.5 (Linear operator, bounded operator). Let (X,∥⋅∥X){(X, \|\cdot\|_X)} and (Y,∥⋅∥Y){(Y, \|\cdot\|_Y)} be normed spaces.

(i) A linear operator is a map T ⁣:X→Y{T \colon X \to Y} such that

T(αx)=αTxfor all α∈R and x∈X,T(x+y)=Tx+Tyfor all x,y∈X.\begin{aligned} T(\alpha x) &= \alpha Tx && \text{for all } \alpha \in \mathbb{R} \text{ and } x \in X, \\ T(x + y) &= Tx + Ty && \text{for all } x, y \in X. \end{aligned}

(ii) A linear operator TT is said to be bounded if there exists C>0{C > 0} such that ∥Tx∥Y≤C∥x∥X{\|Tx\|_Y \le C \|x\|_X} for every x∈X.{x \in X.}

Remark 13.6. Let NN and MM be positive integers. If T ⁣:RN→RM{T \colon \mathbb{R}^N \to \mathbb{R}^M} is linear, then TT is represented by a matrix, and TT is continuous.

Example 13.7. Let −∞<a<b<+∞.{-\infty < a < b < +\infty.} The linear operator TT from (C1([a,b]),∥⋅∥∞){(C^1([a, b]), \|\cdot\|_\infty)} to (C([a,b]),∥⋅∥∞){(C([a, b]), \|\cdot\|_\infty)} given by Tf=f′{Tf = f'} is not continuous.

For Example 13.7, recall that, by Lemma 1.21, a map T ⁣:X→Y{T \colon X \to Y} between normed spaces is continuous if and only if Txm→Tx0{Tx_m \to Tx_0} whenever xm→x0{x_m \to x_0} in X.{X.} We find (fm)⊂C1([a,b]){(f_m) \subset C^1([a, b])} such that fm→0{f_m \to 0} with respect to ∥⋅∥∞,{\|\cdot\|_\infty,} while the derivatives Tfm=fm′{Tf_m = f_m'} do not converge to the derivative of the function 0,{0,} which is 0,{0,} in ∥⋅∥∞.{\|\cdot\|_\infty.}

Suppose that a<0<b,{a < 0 < b,} as in Figure 13.2, and let mm be an integer with 1/m<b.{1/m < b.} Let fm∈C1([a,b]){f_m \in C^1([a, b])} vanish outside (0,1/m),{(0, 1/m),} with 0≤fm≤1/m{0 \le f_m \le 1/m} and fm(1/(2m))=1/m.{f_m(1/(2m)) = 1/m.}

(i) The norms satisfy ∥fm∥∞≤1/m,{\|f_m\|_\infty \le 1/m,} which tends to 0,{0,} so that fm→0{f_m \to 0} in ∥⋅∥∞.{\|\cdot\|_\infty.} (ii) By the mean value theorem, there exists xm∈(0,1/(2m)){x_m \in (0, 1/(2m))} such that fm′(xm)=2.{f_m'(x_m) = 2.} Hence ∥fm′∥∞≥2{\|f_m'\|_\infty \ge 2} for every m,{m,} so that ∥fm′∥∞\|f_m'\|_\infty does not tend to 0,{0,} and TT is not continuous.

ab1/mslope 2xm1/(2m)1/mfm

Figure 13.2. The drawing of Example 13.7, an instance on [a,b]=[−1/2,3/2]{[a, b] = [-1/2, 3/2]} for the mm of the slider: the graph of fm,{f_m,} the dashed level 1/m{1/m} and the points 1/(2m){1/(2m)} and 1/m.{1/m.} The tangent at xmx_m is parallel to the dotted segment from (0,0)(0, 0) to (1/(2m),1/m),{(1/(2m), 1/m),} of slope 2.{2.}

A map TT from a normed space XX to a normed space YY is Lipschitz if there exists L>0{L > 0} such that ∥Tx−Ty∥Y≤L∥x−y∥X{\|Tx - Ty\|_Y \le L \|x - y\|_X} for all x,y∈X.{x, y \in X.} It is continuous at 00 if for every ε>0{\varepsilon > 0} there exists δ>0{\delta > 0} such that

∥x∥X<δ  ⟹  ∥Tx−T0∥Y<ε.\|x\|_X < \delta \implies \|Tx - T0\|_Y < \varepsilon.

Theorem 13.8 (Characterization of continuity). Let T ⁣:(X,∥⋅∥X)→(Y,∥⋅∥Y){T \colon (X, \|\cdot\|_X) \to (Y, \|\cdot\|_Y)} be linear. The following are equivalent.
(1) TT is bounded.
(2) TT is Lipschitz.
(3) TT is continuous.
(4) TT is continuous at 0.{0.}

Proof (complete). We prove that (1) implies (2), that (2) implies (3), that (3) implies (4) and that (4) implies (1). Suppose (1), with the constant CC of Definition 13.5 (ii). Then, for all x,y∈X,{x, y \in X,}

∥Tx−Ty∥Y=∥T(x−y)∥Y≤C∥x−y∥X,\|Tx - Ty\|_Y = \|T(x - y)\|_Y \le C \|x - y\|_X,

where the equality is the linearity of TT and the inequality is (1) applied to x−y.{x - y.} Hence (2) holds with L=C.{L = C.}

Suppose (2), and let (xm)⊂X{(x_m) \subset X} be such that xm→x0{x_m \to x_0} in X.{X.} Then, by (2),

∥Txm−Tx0∥Y≤L∥xm−x0∥X⟶0,\|Tx_m - Tx_0\|_Y \le L \|x_m - x_0\|_X \longrightarrow 0,

so that Txm→Tx0,{Tx_m \to Tx_0,} and TT is continuous by Lemma 1.21. Next, (3) implies (4) by Lemma 1.21 (ii) at the point x0=0.{x_0 = 0.}

Suppose (4). Since T0=0{T0 = 0} by linearity, (4) means that for every ε>0{\varepsilon > 0} there exists δ>0{\delta > 0} such that

∥x∥X≤δ  ⟹  ∥Tx∥Y≤ε.\|x\|_X \le \delta \implies \|Tx\|_Y \le \varepsilon.

In detail, the strict inequalities of the definition of continuity at 00 give this with δ/2{\delta/2} in place of δ.{\delta.} Take ε=1.{\varepsilon = 1.} There is δ>0{\delta > 0} such that ∥x∥X≤δ{\|x\|_X \le \delta} implies ∥Tx∥Y≤1.{\|Tx\|_Y \le 1.} If x=0,{x = 0,} then ∥Tx∥Y=∥T0∥Y=∥0∥Y=0=∥x∥X.{\|Tx\|_Y = \|T0\|_Y = \|0\|_Y = 0 = \|x\|_X.} If x≠0,{x \neq 0,} the vector δx/∥x∥X{\delta x/\|x\|_X} has norm δ,{\delta,} hence, by (4),

∥T(δx∥x∥X)∥Y≤1,T(δx∥x∥X)=δ∥x∥X Tx,⟹∥δ∥x∥X Tx∥Y≤1⟹δ∥x∥X ∥Tx∥Y≤1⟹∥Tx∥Y≤1δ ∥x∥X,\begin{aligned} &\biggl\| T\biggl( \delta \frac{x}{\|x\|_X} \biggr) \biggr\|_Y \le 1, \qquad T\biggl( \delta \frac{x}{\|x\|_X} \biggr) = \frac{\delta}{\|x\|_X} \, Tx, \\ &\Longrightarrow \quad \biggl\| \frac{\delta}{\|x\|_X} \, Tx \biggr\|_Y \le 1 \\ &\Longrightarrow \quad \frac{\delta}{\|x\|_X} \, \|Tx\|_Y \le 1 \\ &\Longrightarrow \quad \|Tx\|_Y \le \frac{1}{\delta} \, \|x\|_X, \end{aligned}

where the equality in the first line is the linearity of T.{T.} The last inequality holds for every x∈X,{x \in X,} the case x=0{x = 0} included, that is, TT is bounded. ■\blacksquare

Definition 13.9 (Dual space). Let (X,∥⋅∥X){(X, \|\cdot\|_X)} and (Y,∥⋅∥Y){(Y, \|\cdot\|_Y)} be normed spaces.
(i) The set of the linear and continuous operators T ⁣:X→Y{T \colon X \to Y} is denoted L(X,Y).{\mathcal{L}(X, Y).} We observe that L(X,Y)\mathcal{L}(X, Y) is a vector space, with (T+S)(x)=Tx+Sx{(T + S)(x) = Tx + Sx} and (αT)(x)=αTx{(\alpha T)(x) = \alpha Tx} for α∈R.{\alpha \in \mathbb{R}.}
(ii) We write L(X):=L(X,X).{\mathcal{L}(X) := \mathcal{L}(X, X).}
(iii) The set X∗:=L(X,R),{X^* := \mathcal{L}(X, \mathbb{R}),} where R\mathbb{R} has the norm ∣⋅∣,{|\cdot|,} is the set of the linear continuous operators T ⁣:X→R,{T \colon X \to \mathbb{R},} the functionals on X.{X.} It is also denoted X′,{X',} and it is called the dual space of X.{X.}

(iv) On L(X,Y)\mathcal{L}(X, Y) we define

∥T∥L(X,Y):=sup⁡∥x∥X≤1∥Tx∥Y,\|T\|_{\mathcal{L}(X, Y)} := \sup_{\|x\|_X \le 1} \|Tx\|_Y,

also written ∥T∥L\|T\|_{\mathcal{L}} or ∥T∥.{\|T\|.} Notice that (L(X,Y),∥⋅∥L(X,Y)){(\mathcal{L}(X, Y), \|\cdot\|_{\mathcal{L}(X, Y)})} is a normed space, and that, if T∈X∗,{T \in X^*,} then

∥T∥L=sup⁡∥x∥X≤1∣Tx∣.\|T\|_{\mathcal{L}} = \sup_{\|x\|_X \le 1} |Tx|.

Proposition 13.10 (Characterization of ∥⋅∥L\|\cdot\|_{\mathcal{L}}). Let (X,∥⋅∥X){(X, \|\cdot\|_X)} and (Y,∥⋅∥Y){(Y, \|\cdot\|_Y)} be normed spaces with X≠{0},{X \neq \{0\},} and let T∈L(X,Y).{T \in \mathcal{L}(X, Y).} Then

∥T∥L=sup⁡∥x∥X=1∥Tx∥Y=sup⁡x∈Xx≠0∥Tx∥Y∥x∥X=inf⁡{M>0:∥Tx∥Y≤M∥x∥X for every x∈X}.\begin{aligned} \|T\|_{\mathcal{L}} &= \sup_{\|x\|_X = 1} \|Tx\|_Y = \sup_{\substack{x \in X \\ x \neq 0}} \frac{\|Tx\|_Y}{\|x\|_X} \\ &= \inf \{ M > 0 : \|Tx\|_Y \le M \|x\|_X \ \text{for every } x \in X \}. \end{aligned}

In words, ∥T∥L\|T\|_{\mathcal{L}} is the least constant that can be put in the definition of a bounded operator, Definition 13.5 (ii). In particular, ∥Tx∥Y≤∥T∥L∥x∥X{\|Tx\|_Y \le \|T\|_{\mathcal{L}} \|x\|_X} for every x∈X.{x \in X.}

Proof (complete). By the definition of ∥T∥L,{\|T\|_{\mathcal{L}},}

∥T∥L=sup⁡∥x∥X≤1∥Tx∥Y≥sup⁡∥x∥X=1∥Tx∥Y≥sup⁡x∈Xx≠0∥Tx∥Y∥x∥X=:M0≥inf⁡{M>0:∥Tx∥Y≤M∥x∥X for every x∈X}=:M1.\begin{aligned} \|T\|_{\mathcal{L}} &= \sup_{\|x\|_X \le 1} \|Tx\|_Y \\ &\ge \sup_{\|x\|_X = 1} \|Tx\|_Y \\ &\ge \sup_{\substack{x \in X \\ x \neq 0}} \frac{\|Tx\|_Y}{\|x\|_X} =: M_0 \\ &\ge \inf \{ M > 0 : \|Tx\|_Y \le M \|x\|_X \ \text{for every } x \in X \} =: M_1. \end{aligned}

The first inequality restricts the set where the supremum is taken. For the second, if x≠0,{x \neq 0,} then x/∥x∥X{x/\|x\|_X} has norm 11 and, by linearity,

∥Tx∥Y∥x∥X=∥T(x∥x∥X)∥Y≤sup⁡∥z∥X=1∥Tz∥Y.\frac{\|Tx\|_Y}{\|x\|_X} = \biggl\| T\biggl( \frac{x}{\|x\|_X} \biggr) \biggr\|_Y \le \sup_{\|z\|_X = 1} \|Tz\|_Y.

For the third, the definition of M0M_0 gives ∥Tx∥Y/∥x∥X≤M0{\|Tx\|_Y/\|x\|_X \le M_0} for every x≠0.{x \neq 0.} Hence ∥Tx∥Y≤M0∥x∥X{\|Tx\|_Y \le M_0 \|x\|_X} for every x∈X,{x \in X,} so M0M_0 is a competitor in the computation of that infimum, and M1≤M0.{M_1 \le M_0.} In detail, if M0=0,{M_0 = 0,} then every M>0{M > 0} is a competitor and M1=0.{M_1 = 0.}

The last step is M1≥∥T∥L.{M_1 \ge \|T\|_{\mathcal{L}}.} Let M>0{M > 0} be any competitor in the infimum, that is, ∥Tx∥Y≤M∥x∥X{\|Tx\|_Y \le M \|x\|_X} for every x∈X.{x \in X.} Then, for every xx with ∥x∥X≤1,{\|x\|_X \le 1,}

∥Tx∥Y≤M∥x∥X≤M⟹∥T∥L=sup⁡∥x∥X≤1∥Tx∥Y≤M.\begin{aligned} &\|Tx\|_Y \le M \|x\|_X \le M \\ &\Longrightarrow \quad \|T\|_{\mathcal{L}} = \sup_{\|x\|_X \le 1} \|Tx\|_Y \le M. \end{aligned}

Taking the infimum over MM gives ∥T∥L≤M1.{\|T\|_{\mathcal{L}} \le M_1.} ■\blacksquare

Theorem 13.11. Let (X,∥⋅∥X){(X, \|\cdot\|_X)} be a normed space and (Y,∥⋅∥Y){(Y, \|\cdot\|_Y)} a Banach space, as in Definition 10.4 (ii). Then (L(X,Y),∥⋅∥L(X,Y)){(\mathcal{L}(X, Y), \|\cdot\|_{\mathcal{L}(X, Y)})} is a Banach space. In particular, X∗X^* is a Banach space.

Definition 13.12 (Kernel, range). Let T ⁣:(X,∥⋅∥X)→(Y,∥⋅∥Y){T \colon (X, \|\cdot\|_X) \to (Y, \|\cdot\|_Y)} be linear.
(i) The kernel of TT is Ker⁡(T)={x∈X:Tx=0}.{\operatorname{Ker}(T) = \{ x \in X : Tx = 0 \}.}
(ii) The range, or image, of TT is R(T)={Tx:x∈X}.{R(T) = \{ Tx : x \in X \}.}

Remark 13.13. Let T∈L(X,Y).{T \in \mathcal{L}(X, Y).} Then Ker⁡(T)\operatorname{Ker}(T) is a closed subspace of X,{X,} but R(T)R(T) is not always closed in Y.{Y.}

Remark 13.14. Let (X,∥⋅∥X){(X, \|\cdot\|_X)} and (Y,∥⋅∥Y){(Y, \|\cdot\|_Y)} be normed spaces and let T ⁣:X→Y{T \colon X \to Y} be linear.
(i) TT is injective if and only if Ker⁡(T)={0}.{\operatorname{Ker}(T) = \{0\}.}
(ii) If TT is bijective, then its inverse T−1 ⁣:Y→X{T^{-1} \colon Y \to X} exists and is linear.
(iii) If T∈L(X,Y){T \in \mathcal{L}(X, Y)} is bijective, it is not always true that T−1{T^{-1}} is continuous.

Definition 13.15 (Isomorphism, isometry, isometric isomorphism). Let (X,∥⋅∥X){(X, \|\cdot\|_X)} and (Y,∥⋅∥Y){(Y, \|\cdot\|_Y)} be normed spaces.
(i) A map T ⁣:X→Y{T \colon X \to Y} is an isomorphism of normed spaces if T∈L(X,Y),{T \in \mathcal{L}(X, Y),} TT is bijective and T−1∈L(Y,X).{T^{-1} \in \mathcal{L}(Y, X).}
(ii) A map T∈L(X,Y){T \in \mathcal{L}(X, Y)} is an isometry if ∥Tx∥Y=∥x∥X{\|Tx\|_Y = \|x\|_X} for every x∈X.{x \in X.} Notice that an isometry is not required to be surjective.
(iii) A map T ⁣:X→Y{T \colon X \to Y} is an isometric isomorphism if TT is an isomorphism and an isometry. Notice that in this case also T−1{T^{-1}} is an isometric isomorphism.

Remark 13.16. Let T∈L(X,Y){T \in \mathcal{L}(X, Y)} be an isometry.
(i) TT is injective.
(ii) TT is an isometric isomorphism if and only if TT is surjective.

Definition 13.17 (Embedding). Let (Y,∥⋅∥Y){(Y, \|\cdot\|_Y)} be a normed space and let X⊂Y{X \subset Y} be a subspace such that (X,∥⋅∥X){(X, \|\cdot\|_X)} is a normed space. Consider

J ⁣:(X,∥⋅∥X)→(Y,∥⋅∥Y),x↦x,J \colon (X, \|\cdot\|_X) \to (Y, \|\cdot\|_Y), \qquad x \mapsto x,

called the identity, the immersion or the injection of XX into Y.{Y.} The map JJ is called an embedding if it is continuous.

By Definition 13.17 and Theorem 13.8, JJ is an embedding if and only if JJ is bounded. That is, there exists C>0{C > 0} such that

∥Jx∥Y≤C∥x∥Xfor every x∈X,that is,∥x∥Y≤C∥x∥Xfor every x∈X.\begin{aligned} &\|Jx\|_Y \le C \|x\|_X \quad \text{for every } x \in X, \\ \text{that is,} \quad &\|x\|_Y \le C \|x\|_X \quad \text{for every } x \in X. \end{aligned}

Example 13.18. Let (Ω,M,μ){(\Omega, \mathcal{M}, \mu)} be a measure space with μ(Ω)<+∞,{\mu(\Omega) < +\infty,} and let 1≤p≤q≤+∞.{1 \le p \le q \le +\infty.} By the Hölder inequality, Lq(Ω)⊂Lp(Ω),{L^q(\Omega) \subset L^p(\Omega),} as in Corollary 10.13, and we can consider

J ⁣:Lq(Ω)→Lp(Ω),f↦f.J \colon L^q(\Omega) \to L^p(\Omega), \qquad f \mapsto f.

The Hölder inequality, Proposition 10.12, gives a constant C>0{C > 0} such that ∥f∥p≤C∥f∥q{\|f\|_p \le C \|f\|_q} for every f∈Lq(Ω).{f \in L^q(\Omega).} That is, JJ is continuous: Lq(Ω)L^q(\Omega) embeds into Lp(Ω).{L^p(\Omega).}

Baire’s lemma

A metric space (X,d){(X, d)} is complete if every Cauchy sequence converges in X,{X,} where Cauchy sequences are defined as in Definition 10.4 (i) with d(xn,xm){d(x_n, x_m)} in place of ∥xn−xm∥.{\|x_n - x_m\|.}

Theorem 13.19 (Baire’s lemma). Let (X,d){(X, d)} be a complete metric space.
(1) If (Um)m≥1{(U_m)_{m \ge 1}} is a sequence of open dense sets in X,{X,} then their intersection is dense in X.{X.}
(2) If (Cm)m≥1{(C_m)_{m \ge 1}} is a sequence of closed sets in XX such that int⁡(Cm)=∅{\operatorname{int}(C_m) = \emptyset} for every m,{m,} then their union has empty interior.

In Theorem 13.19, (1) and (2) are equivalent: with Cm=X∖Um{C_m = X \setminus U_m} for every m,{m,}

⋃m=1∞Cm=X∖⋂m=1∞Um.\bigcup_{m=1}^{\infty} C_m = X \setminus \bigcap_{m=1}^{\infty} U_m.

Indeed, by Remark 13.3 a set D⊂X{D \subset X} is dense if and only if int⁡(X∖D)=∅,{\operatorname{int}(X \setminus D) = \emptyset,} since both mean that every nonempty open set meets D.{D.} Moreover, CmC_m is closed if and only if UmU_m is open.

Corollary 13.20. Let (X,d){(X, d)} be a nonempty complete metric space and let (Cm)m≥1{(C_m)_{m \ge 1}} be a sequence of closed sets whose union is X.{X.} Then there exists m0m_0 such that int⁡(Cm0)≠∅.{\operatorname{int}(C_{m_0}) \neq \emptyset.}

Proof (complete). We have

int⁡(⋃m=1∞Cm)=int⁡(X)≠∅.\operatorname{int}\Big( \bigcup_{m=1}^{\infty} C_m \Big) = \operatorname{int}(X) \neq \emptyset.

Hence, by Theorem 13.19 (2), not every element CmC_m of the sequence has empty interior. ■\blacksquare

Proof (complete). We prove Theorem 13.19 (1); then (2) holds by the equivalence of (1) and (2). Let (Um)m≥1{(U_m)_{m \ge 1}} be open and dense, and take O⊂X{O \subset X} open with O≠∅.{O \neq \emptyset.} By Remark 13.3, the goal is

⋂m=1∞Um∩O≠∅.\bigcap_{m=1}^{\infty} U_m \cap O \neq \emptyset.

(i) The set U1∩O{U_1 \cap O} is open, and it is nonempty because U1U_1 is dense, by Remark 13.3. Hence there exist x1∈X{x_1 \in X} and r1∈(0,1/2){r_1 \in (0, 1/2)} such that the closed ball B‾r1(x1)={y∈X:d(x1,y)≤r1}{\overline{B}_{r_1}(x_1) = \{ y \in X : d(x_1, y) \le r_1 \}} is contained in O∩U1.{O \cap U_1.}

(ii) The set U2∩Br1(x1){U_2 \cap B_{r_1}(x_1)} is open, and it is nonempty because U2U_2 is dense. Hence, as in Figure 13.3, there exist x2∈X{x_2 \in X} and r2∈(0,1/4){r_2 \in (0, 1/4)} such that

B‾r2(x2)⊂Br1(x1)∩U2⊂O∩U1∩U2.\begin{aligned} \overline{B}_{r_2}(x_2) &\subset B_{r_1}(x_1) \cap U_2 \\ &\subset O \cap U_1 \cap U_2. \end{aligned}

(iii) Iterating, we get points xm∈X{x_m \in X} and radii rm∈(0,1/2m),{r_m \in (0, 1/2^m),} and finding this sequence relies on the axiom of choice. For every m≥2,{m \ge 2,}

B‾rm(xm)⊂Brm−1(xm−1)∩Um⊂O∩U1∩U2∩⋯∩Um=O∩⋂k=1mUk.\begin{aligned} \overline{B}_{r_m}(x_m) &\subset B_{r_{m-1}}(x_{m-1}) \cap U_m \\ &\subset O \cap U_1 \cap U_2 \cap \dots \cap U_m \\ &= O \cap \bigcap_{k=1}^{m} U_k. \end{aligned}
U2OU1Br1(x1)Br2(x2)
step (ii)

Figure 13.3. The drawing of the proof of Theorem 13.19 (1): the open sets O,{O,} U1U_1 and U2,{U_2,} the closed ball B‾r1(x1)⊂O∩U1{\overline{B}_{r_1}(x_1) \subset O \cap U_1} and the closed ball B‾r2(x2)⊂Br1(x1)∩U2.{\overline{B}_{r_2}(x_2) \subset B_{r_1}(x_1) \cap U_2.} The button shows the steps; shaded, the open sets O∩U1{O \cap U_1} and Br1(x1)∩U2{B_{r_1}(x_1) \cap U_2} in which the two balls are chosen.

The sequence (xm)(x_m) is Cauchy. For all n,m≥N{n, m \ge N} we have xn,xm∈BrN(xN),{x_n, x_m \in B_{r_N}(x_N),} hence

d(xn,xm)≤2rN<22N.\begin{aligned} d(x_n, x_m) &\le 2 r_N \\ &< \frac{2}{2^N}. \end{aligned}

Since 2/2N→0{2/2^N \to 0} as N→∞,{N \to \infty,} the sequence (xm)(x_m) is Cauchy. By completeness, there exists x∞∈X{x_\infty \in X} such that xm→x∞.{x_m \to x_\infty.} We claim that x∞x_\infty belongs to OO and to every Um,{U_m,} so that the set of the goal is nonempty. For every NN and every m≥N{m \ge N} we have xm∈B‾rN(xN),{x_m \in \overline{B}_{r_N}(x_N),} as in Figure 13.4. Hence

x∞∈B‾rN(xN)⊂O∩⋂k=1NUkfor every N.\begin{aligned} x_\infty &\in \overline{B}_{r_N}(x_N) \\ &\subset O \cap \bigcap_{k=1}^{N} U_k \qquad \text{for every } N. \end{aligned}
BrN(xN)x∞points xm, m ≥ N

Figure 13.4. The drawing of the claim of the proof, an instance in the plane for the NN of the slider: the closed ball B‾rN(xN),{\overline{B}_{r_N}(x_N),} the points xmx_m with m≥N,{m \ge N,} all inside it, and their limit x∞.{x_\infty.} The balls of the other indices are faint.

Therefore

x∞∈O∩⋂k=1∞Uk.■x_\infty \in O \cap \bigcap_{k=1}^{\infty} U_k. \tag*{$\blacksquare$}