Alessandro Palliccia

Real and Functional Analysis

Separation, reflexivity and weak convergence

1,118 words, , published

Recall the Hahn–Banach theorem, Theorem 15.14.

Hahn–Banach theorems (continued)

Theorem 16.1 (Hahn–Banach). Let (X,∥⋅∥X){(X, \|\cdot\|_X)} be a normed space, let V⊂X{V \subset X} be a subspace, with the norm ∥⋅∥X,{\|\cdot\|_X,} and let L∈V∗.{L \in V^*.} Then there exists T∈X∗{T \in X^*} such that

Tv=Lvfor every v∈V, that is, T∣V=L,∥T∥X∗=∥L∥V∗, that is, the norm does not change.\begin{aligned} &Tv = Lv \quad \text{for every } v \in V, \ \text{that is,} \ T|_V = L, \\ &\|T\|_{X^*} = \|L\|_{V^*}, \ \text{that is, the norm does not change.} \end{aligned}

Corollary 16.2. Let (X,∥⋅∥X){(X, \|\cdot\|_X)} be a normed space and let x0∈X{x_0 \in X} with x0≠0.{x_0 \neq 0.} Then there exists T∈X∗{T \in X^*} such that ∥T∥X∗=1{\|T\|_{X^*} = 1} and Tx0=∥x0∥X.{T x_0 = \|x_0\|_X.}

Proof (complete). Since x0≠0,{x_0 \neq 0,} we can consider the subspace V=Span⁡(x0):={tx0:t∈R}.{V = \operatorname{Span}(x_0) := \{ t x_0 : t \in \mathbb{R} \}.} Let L ⁣:V→R{L \colon V \to \mathbb{R}} be given by L(tx0)=t ∥x0∥X.{L(t x_0) = t \, \|x_0\|_X.} The map LL is linear, and it belongs to V∗V^* with ∥L∥V∗=1,{\|L\|_{V^*} = 1,} because

∥L∥V∗=sup⁡v∈V∥v∥X=1∣L(v)∣=sup⁡∥tx0∥X=1∣L(tx0)∣=sup⁡∣t∣ ∥x0∥X=1∣t ∥x0∥X∣=sup⁡∣t∣ ∥x0∥X=1∣t∣ ∥x0∥X=1.\begin{aligned} \|L\|_{V^*} &= \sup_{\substack{v \in V \\ \|v\|_X = 1}} |L(v)| \\ &= \sup_{\|t x_0\|_X = 1} |L(t x_0)| = \sup_{|t| \, \|x_0\|_X = 1} \bigl| t \, \|x_0\|_X \bigr| \\ &= \sup_{|t| \, \|x_0\|_X = 1} |t| \, \|x_0\|_X = 1. \end{aligned}

By the Hahn–Banach theorem, Theorem 16.1, there exists T∈X∗{T \in X^*} such that T∣V=L{T|_V = L} and ∥T∥X∗=∥L∥V∗=1.{\|T\|_{X^*} = \|L\|_{V^*} = 1.} Moreover, Tx0=Lx0=∥x0∥X,{T x_0 = L x_0 = \|x_0\|_X,} since x0∈V.{x_0 \in V.} ■\blacksquare

Definition 16.3 (Hyperplane). Let (X,∥⋅∥X){(X, \|\cdot\|_X)} be a normed space and let T ⁣:X→R{T \colon X \to \mathbb{R}} be linear. A hyperplane is a subset of XX given, for some α∈R,{\alpha \in \mathbb{R},} by

H={x∈X:Tx=α}.H = \{ x \in X : Tx = \alpha \}.

Notation. We write H={T=α}=T−1({α}).{H = \{ T = \alpha \} = T^{-1}(\{\alpha\}).}

Remark 16.4. (i) If α=0,{\alpha = 0,} then H={T=0}=Ker⁡T.{H = \{ T = 0 \} = \operatorname{Ker} T.}
(ii) If X=RM{X = \mathbb{R}^M} for a positive integer M,{M,} so that T ⁣:RM→R,{T \colon \mathbb{R}^M \to \mathbb{R},} and TT is not identically zero, then H={T=α}{H = \{ T = \alpha \}} is a classical hyperplane.

Corollary 16.5 (Separation of points). Let (X,∥⋅∥X){(X, \|\cdot\|_X)} be a normed space and let x,y∈X{x, y \in X} with x≠y.{x \neq y.} Then there exists T∈X∗{T \in X^*} such that Tx≠Ty.{Tx \neq Ty.} That is, the hyperplane H={T=α}{H = \{ T = \alpha \}} with α=Tx{\alpha = Tx} contains xx and does not contain y,{y,} as in Figure 16.1.

Hxy

Figure 16.1. The drawing of Corollary 16.5: the hyperplane H={T=α},{H = \{ T = \alpha \},} with α=Tx,{\alpha = Tx,} drawn as a line, the point xx on it and the point yy outside it.

Proof (complete). Let x0:=x−y.{x_0 := x - y.} Then x0≠0,{x_0 \neq 0,} because x≠y.{x \neq y.} By Corollary 16.2 there exists T∈X∗{T \in X^*} such that Tx0=∥x0∥X,{T x_0 = \|x_0\|_X,} where Tx0=T(x−y)=Tx−Ty.{T x_0 = T(x - y) = Tx - Ty.} Hence, since ∥x0∥X>0,{\|x_0\|_X > 0,}

Tx=Ty+∥x0∥X>Ty.■Tx = Ty + \|x_0\|_X > Ty. \tag*{$\blacksquare$}

Corollary 16.6 (Separation of point and subspace). Let (X,∥⋅∥X){(X, \|\cdot\|_X)} be a normed space and let V⊂X{V \subset X} be a closed subspace with V≠X.{V \neq X.} Let x0∈X∖V.{x_0 \in X \setminus V.} Then there exists T∈X∗{T \in X^*} such that Tx0≠0{T x_0 \neq 0} and Tv=0{Tv = 0} for every vv in V.{V.} That is, x0∉Ker⁡T{x_0 \notin \operatorname{Ker} T} and V⊂Ker⁡T,{V \subset \operatorname{Ker} T,} as in Figure 16.2.

0Vx0Ker T

Figure 16.2. The drawing of Corollary 16.6: the kernel Ker⁡T,{\operatorname{Ker} T,} drawn as a plane, the subspace V,{V,} drawn as a line through 00 inside it, and the point x0x_0 outside it.

Proof (complete). Since VV is closed and x0∉V,{x_0 \notin V,}

d(x0,V):=inf⁡v∈V∥x0−v∥X>0,d(x_0, V) := \inf_{v \in V} \|x_0 - v\|_X > 0,

as in the proof of the Riesz theorem on closed balls, Theorem 12.18. Let

Z:=Span⁡(V,x0)={v+tx0:v∈V, t∈R},Z := \operatorname{Span}(V, x_0) = \{ v + t x_0 : v \in V, \ t \in \mathbb{R} \},

and let L ⁣:Z→R{L \colon Z \to \mathbb{R}} be given by L(v+tx0)=t.{L(v + t x_0) = t.} Indeed, LL is well defined, because two writings v+tx0=v′+t′x0{v + t x_0 = v' + t' x_0} of the same element with t≠t′{t \neq t'} would give x0=(v′−v)/(t−t′)∈V.{x_0 = (v' - v)/(t - t') \in V.} Notice that L(x0)=1{L(x_0) = 1} and that L(v)=0{L(v) = 0} for every v∈V.{v \in V.} The map LL is linear, and it is continuous, that is, L∈Z∗.{L \in Z^*.} Let z=v+tx0∈Z,{z = v + t x_0 \in Z,} with v∈V{v \in V} and t∈R.{t \in \mathbb{R}.}
Case 1. If t=0,{t = 0,} then ∣Lz∣=∣Lv∣=0.{|Lz| = |Lv| = 0.}
Case 2. If t≠0,{t \neq 0,} then ∣Lz∣=∣L(v+tx0)∣=∣t∣,{|Lz| = |L(v + t x_0)| = |t|,} and

∥z∥X=∥v+tx0∥X=∣t∣ ∥vt+x0∥X=∣t∣ ∥x0−(−vt)∥X≥∣t∣ d(x0,V),\begin{aligned} \|z\|_X &= \|v + t x_0\|_X = |t| \, \Bigl\| \frac{v}{t} + x_0 \Bigr\|_X \\ &= |t| \, \Bigl\| x_0 - \Bigl( -\frac{v}{t} \Bigr) \Bigr\|_X \\ &\ge |t| \, d(x_0, V), \end{aligned}

where −v/t∈V.{-v/t \in V.} As d(x0,V)>0,{d(x_0, V) > 0,} it follows that

∣Lz∣=∣t∣≤1d(x0,V) ∥z∥X.|Lz| = |t| \le \frac{1}{d(x_0, V)} \, \|z\|_X.

By Cases 1 and 2, ∣Lz∣≤∥z∥X/d(x0,V){|Lz| \le \|z\|_X / d(x_0, V)} for every z∈Z,{z \in Z,} hence L∈Z∗.{L \in Z^*.} By the Hahn–Banach theorem, Theorem 16.1, there exists T∈X∗{T \in X^*} such that T∣Z=L.{T|_Z = L.} In particular, Tx0=Lx0=1{T x_0 = L x_0 = 1} and Tv=Lv=0{Tv = Lv = 0} for every v∈V.{v \in V.} ■\blacksquare

Duals, biduals, reflexivity

Definition 16.7 (Bidual space). Let (X,∥⋅∥X){(X, \|\cdot\|_X)} be a normed space, and let X∗=L(X,R){X^* = \mathcal{L}(X, \mathbb{R})} be its dual, a Banach space by Remark 15.13 (i). The bidual space of XX is X∗∗:=L(X∗,R).{X^{**} := \mathcal{L}(X^*, \mathbb{R}).}

Notation. For TT in X∗,{X^*,} xx in XX and ϕ\phi in X∗∗,{X^{**},} we write

Tx=X∗⟨T,x⟩X=∗⟨T,x⟩,ϕT=X∗∗⟨ϕ,T⟩X∗=∗∗⟨ϕ,T⟩∗.\begin{aligned} Tx &= {}_{X^*}\langle T, x \rangle_X = {}_*\langle T, x \rangle, \\ \phi T &= {}_{X^{**}}\langle \phi, T \rangle_{X^*} = {}_{**}\langle \phi, T \rangle_*. \end{aligned}

Definition 16.8 (Canonical map). Let (X,∥⋅∥X){(X, \|\cdot\|_X)} be a normed space. The canonical map, also called the evaluation map, is the map τ ⁣:X→X∗∗,{\tau \colon X \to X^{**},} x↦τ(x)∈X∗∗,{x \mapsto \tau(x) \in X^{**},} defined by

∗∗⟨τ(x),T⟩∗:=∗⟨T,x⟩=Tx{}_{**}\langle \tau(x), T \rangle_* := {}_*\langle T, x \rangle = Tx

for every T∈X∗.{T \in X^*.} That is, τ(x)\tau(x) acts on any T∈X∗{T \in X^*} by evaluating TT at x.{x.}

Theorem 16.9. Let (X,∥⋅∥X){(X, \|\cdot\|_X)} be a Banach space and let τ\tau be its canonical map. Then:
(i) the map τ\tau is linear and continuous;
(ii) the map τ ⁣:(X,∥⋅∥X)→(X∗∗,∥⋅∥X∗∗){\tau \colon (X, \|\cdot\|_X) \to (X^{**}, \|\cdot\|_{X^{**}})} is an isometry, that is, ∥τ(x)∥X∗∗=∥x∥X{\|\tau(x)\|_{X^{**}} = \|x\|_X} for every x∈X;{x \in X;}
(iii) the range R(τ)R(\tau) is closed in X∗∗;{X^{**};}
(iv) the map τ ⁣:(X,∥⋅∥X)→(R(τ),∥⋅∥X∗∗){\tau \colon (X, \|\cdot\|_X) \to (R(\tau), \|\cdot\|_{X^{**}})} is an isometric isomorphism.

Proof (complete). The map τ\tau is linear: for all x1,x2∈X,{x_1, x_2 \in X,} α,β∈R{\alpha, \beta \in \mathbb{R}} and T∈X∗,{T \in X^*,}

∗∗⟨τ(αx1+βx2),T⟩∗=T(αx1+βx2)=αTx1+βTx2=α ∗∗⟨τ(x1),T⟩∗+β ∗∗⟨τ(x2),T⟩∗=∗∗⟨ατ(x1)+βτ(x2),T⟩∗,\begin{aligned} &{}_{**}\langle \tau(\alpha x_1 + \beta x_2), T \rangle_* \\ &\quad = T(\alpha x_1 + \beta x_2) = \alpha T x_1 + \beta T x_2 \\ &\quad = \alpha \, {}_{**}\langle \tau(x_1), T \rangle_* + \beta \, {}_{**}\langle \tau(x_2), T \rangle_* \\ &\quad = {}_{**}\langle \alpha \tau(x_1) + \beta \tau(x_2), T \rangle_*, \end{aligned}

where the first equality is the definition of τ,{\tau,} Definition 16.8. The map τ\tau is an isometry: for every x∈X,{x \in X,}

∥τ(x)∥X∗∗=sup⁡T∈X∗∥T∥X∗≤1∣∗∗⟨τ(x),T⟩∗∣=sup⁡T∈X∗∥T∥X∗≤1∣Tx∣=∥x∥X.\begin{aligned} \|\tau(x)\|_{X^{**}} &= \sup_{\substack{T \in X^* \\ \|T\|_{X^*} \le 1}} \bigl| {}_{**}\langle \tau(x), T \rangle_* \bigr| \\ &= \sup_{\substack{T \in X^* \\ \|T\|_{X^*} \le 1}} |Tx| = \|x\|_X. \end{aligned}

In the last equality, ∣Tx∣≤∥T∥X∗ ∥x∥X≤∥x∥X{|Tx| \le \|T\|_{X^*} \, \|x\|_X \le \|x\|_X} for every T∈X∗{T \in X^*} with ∥T∥X∗≤1.{\|T\|_{X^*} \le 1.} Moreover, if x≠0,{x \neq 0,} Corollary 16.2 applied to xx gives a functional TT of X∗X^* with ∥T∥X∗=1{\|T\|_{X^*} = 1} and Tx=∥x∥X,{Tx = \|x\|_X,} so that

sup⁡T∈X∗∥T∥X∗≤1∣Tx∣=max⁡T∈X∗∥T∥X∗≤1∣Tx∣=∥x∥X.\sup_{\substack{T \in X^* \\ \|T\|_{X^*} \le 1}} |Tx| = \max_{\substack{T \in X^* \\ \|T\|_{X^*} \le 1}} |Tx| = \|x\|_X.

In detail, for x=0{x = 0} both sides of the equality to prove vanish, because τ(0)=0{\tau(0) = 0} by linearity. Therefore τ\tau is an isometry, hence bounded, and so continuous.

Corollary 15.9, applied to τ,{\tau,} gives the rest of the theorem. Indeed, XX is a Banach space by assumption, and X∗∗=(X∗)∗{X^{**} = (X^*)^*} is a Banach space by Remark 15.13 (i). ■\blacksquare

Definition 16.10 (Reflexive space). Let (X,∥⋅∥X){(X, \|\cdot\|_X)} be a Banach space. If the canonical map τ ⁣:X→X∗∗{\tau \colon X \to X^{**}} is surjective, that is, R(τ)=X∗∗,{R(\tau) = X^{**},} we say that XX is reflexive.

Remark 16.11. Let (X,∥⋅∥X){(X, \|\cdot\|_X)} be a reflexive Banach space. Then τ ⁣:(X,∥⋅∥X)→(X∗∗,∥⋅∥X∗∗){\tau \colon (X, \|\cdot\|_X) \to (X^{**}, \|\cdot\|_{X^{**}})} is an isometric isomorphism.

The LpL^p spaces

Remark 16.12. Let (Ω,M,μ){(\Omega, \mathcal{M}, \mu)} be a measure space, let p∈(1,∞){p \in (1, \infty)} and let p′∈(1,∞){p' \in (1, \infty)} be such that 1/p+1/p′=1.{1/p + 1/p' = 1.} Consider Lp=Lp(Ω,M,μ).{L^p = L^p(\Omega, \mathcal{M}, \mu).}

(i) For every f∈Lp′{f \in L^{p'}} we can define a corresponding functional Tf∈(Lp)∗{T_f \in (L^p)^*} by

Tf(g):=∫Ωfg dμfor every g∈Lp,T_f(g) := \int_\Omega f g \, d\mu \qquad \text{for every } g \in L^p,

and ∥Tf∥(Lp)∗=∥f∥p′.{\|T_f\|_{(L^p)^*} = \|f\|_{p'}.}
(ii) The map F ⁣:Lp′→(Lp)∗,{F \colon L^{p'} \to (L^p)^*,} f↦Tf,{f \mapsto T_f,} is an isometry.

Proof (complete). By the Hölder inequality, Proposition 10.12, ∣Tf(g)∣≤∥f∥p′ ∥g∥p{|T_f(g)| \le \|f\|_{p'} \, \|g\|_p} for every g∈Lp,{g \in L^p,} hence ∥Tf∥(Lp)∗≤∥f∥p′.{\|T_f\|_{(L^p)^*} \le \|f\|_{p'}.} Now let g=∣f∣p′−1sgn⁡f,{g = |f|^{p'-1} \operatorname{sgn} f,} where sgn⁡\operatorname{sgn} is the sign function. Notice that g∈Lp,{g \in L^p,} since ∣g∣p=∣f∣p(p′−1)=∣f∣p′{|g|^p = |f|^{p(p'-1)} = |f|^{p'}} and ∫Ω∣g∣p dμ=∫Ω∣f∣p′ dμ<+∞.{\int_\Omega |g|^p \, d\mu = \int_\Omega |f|^{p'} \, d\mu < +\infty.} With this g,{g,}

∣Tf(g)∣=∣∫Ωf ∣f∣p′−1sgn⁡f dμ∣=∫Ω∣f∣p′ dμ=∥f∥p′p′=∥g∥p ∥f∥p′.\begin{aligned} |T_f(g)| &= \biggl| \int_\Omega f \, |f|^{p'-1} \operatorname{sgn} f \, d\mu \biggr| \\ &= \int_\Omega |f|^{p'} \, d\mu \\ &= \|f\|_{p'}^{p'} = \|g\|_p \, \|f\|_{p'}. \end{aligned}

Hence ∥Tf∥(Lp)∗=∥f∥p′.{\|T_f\|_{(L^p)^*} = \|f\|_{p'}.} Indeed, if f≠0,{f \neq 0,} then ∥g∥p>0{\|g\|_p > 0} and ∥Tf∥(Lp)∗≥∣Tf(g)∣/∥g∥p=∥f∥p′,{\|T_f\|_{(L^p)^*} \ge |T_f(g)|/\|g\|_p = \|f\|_{p'},} while for ff equal to 00 both sides vanish. We found that the map FF is an isometry, which is (ii). ■\blacksquare

The question raised by Remark 16.12 (ii) is whether FF is surjective. The answer is yes.

Theorem 16.13 (Riesz representation in LpL^p). Let (Ω,M,μ){(\Omega, \mathcal{M}, \mu)} be a measure space.

(1) If p∈(1,∞),{p \in (1, \infty),} then for every T∈(Lp)∗{T \in (L^p)^*} there exists a unique f∈Lp′{f \in L^{p'}} such that

Tg=∫Ωfg dμfor every g∈Lp.Tg = \int_\Omega f g \, d\mu \qquad \text{for every } g \in L^p.

Moreover, ∥T∥(Lp)∗=∥f∥p′,{\|T\|_{(L^p)^*} = \|f\|_{p'},} and LpL^p is reflexive. The idea of the reflexivity is that (Lp)∗(L^p)^* is identified with Lp′L^{p'} by the first part of (1), and (Lp′)∗(L^{p'})^* with LpL^p by the first part of (1) applied to p′,{p',} since (p′)′=p{(p')' = p} by Remark 10.10 (i):

Lp⇝(Lp)∗⇝(Lp)∗∗∣≀↕Lp′⇝(Lp′)∗≃Lp.\begin{array}{ccccc} L^p & \rightsquigarrow & (L^p)^* & \rightsquigarrow & (L^p)^{**} \\ & & \vert\wr & & \big\updownarrow \\ & & L^{p'} & \rightsquigarrow & (L^{p'})^* \simeq L^p. \end{array}

(2) If p=1{p = 1} and (Ω,M,μ){(\Omega, \mathcal{M}, \mu)} is σ\sigma-finite, then for every T∈(L1)∗{T \in (L^1)^*} there exists a unique f∈L∞{f \in L^\infty} such that

Tg=∫Ωfg dμfor every g∈L1.Tg = \int_\Omega f g \, d\mu \qquad \text{for every } g \in L^1.

Moreover, ∥T∥(L1)∗=∥f∥∞.{\|T\|_{(L^1)^*} = \|f\|_\infty.}

Remark 16.14. Let (Ω,M,μ){(\Omega, \mathcal{M}, \mu)} be a measure space. In general, (L∞)∗(L^\infty)^* is not isometrically isomorphic to L1,{L^1,} and L1L^1 is not reflexive.

Proof (sketch). Consider ([0,1],L([0,1]),λ).{([0, 1], \mathcal{L}([0, 1]), \lambda).} We want to find T∈(L∞)∗{T \in (L^\infty)^*} that is not represented by an L1L^1 function, that is, such that T≠Tf{T \neq T_f} for every f∈L1,{f \in L^1,} where

Tf(g):=∫01fg dλfor every g∈L∞.T_f(g) := \int_0^1 f g \, d\lambda \qquad \text{for every } g \in L^\infty.

Consider (C([0,1]),∥⋅∥∞)⊂L∞,{(C([0, 1]), \|\cdot\|_\infty) \subset L^\infty,} a closed subspace of L∞,{L^\infty,} and fix f0∈L∞∖C([0,1]).{f_0 \in L^\infty \setminus C([0, 1]).} By Corollary 16.6 there exists T∈(L∞)∗{T \in (L^\infty)^*} such that Tf0≠0{T f_0 \neq 0} and TT vanishes on C([0,1]).{C([0, 1]).}

We claim that there is no u∈L1{u \in L^1} such that T=Tu.{T = T_u.} Suppose, by contradiction, that there exists u∈L1{u \in L^1} such that T=Tu,{T = T_u,} that is,

Tf=Tuf=∫01fu dλfor every f∈L∞.Tf = T_u f = \int_0^1 f u \, d\lambda \qquad \text{for every } f \in L^\infty.

Then Tug=Tg=0{T_u g = Tg = 0} for every g∈C([0,1]),{g \in C([0, 1]),} that is,

∫01gu dλ=0for every g∈C([0,1]),\int_0^1 g u \, d\lambda = 0 \qquad \text{for every } g \in C([0, 1]),

hence u=0{u = 0} almost everywhere. Indeed, every g∈Cc∞(0,1),{g \in C_c^\infty(0, 1),} extended by 0, belongs to C([0,1]),{C([0, 1]),} so that (1) of Corollary 5.16 applies. Therefore, as Tf0≠0{T f_0 \neq 0} and u=0{u = 0} almost everywhere, we reach the contradiction

0≠Tf0=Tuf0=∫01f0 u dλ=0.■0 \neq T f_0 = T_u f_0 = \int_0^1 f_0 \, u \, d\lambda = 0. \tag*{$\blacksquare$}

Final remarks

Theorem 16.15 (Reflexivity versus topology). Let (X,∥⋅∥X){(X, \|\cdot\|_X)} be a Banach space.
(1) If XX is reflexive and V⊂X{V \subset X} is a closed subspace, then VV is reflexive.
(2) The space XX is reflexive if and only if X∗X^* is reflexive.
(3) If X∗X^* is separable, then XX is separable.
(4) The space XX is reflexive and separable if and only if X∗X^* is reflexive and separable.

Proof (sketch). Only the idea of (2) is given. If XX is reflexive, then, identifying X∗∗X^{**} with X,{X,}

(X∗)∗∗=X∗∗∗=(X∗∗)∗=X∗.(X^*)^{**} = X^{***} = (X^{**})^* = X^*.

Conversely, let X∗X^* be reflexive. Then X∗∗=(X∗)∗{X^{**} = (X^*)^*} is reflexive, by the first implication applied to the Banach space X∗.{X^*.} The range τ(X)\tau(X) of the canonical map τ ⁣:X→X∗∗{\tau \colon X \to X^{**}} is closed in X∗∗X^{**} by Theorem 16.9, hence τ(X)\tau(X) is reflexive by (1), and therefore XX is reflexive. ■\blacksquare

Weak convergence (part 1)

Definition 16.16 (Weak convergence). Let (X,∥⋅∥X){(X, \|\cdot\|_X)} be a Banach space, let (xm)⊂X{(x_m) \subset X} and let x∈X.{x \in X.} We say that (xm)(x_m) weakly converges to xx if

lim⁡m→∞Txm=Txfor every T∈X∗.\lim_{m \to \infty} T x_m = T x \qquad \text{for every } T \in X^*.

Notation. If (xm)(x_m) weakly converges to x,{x,} we write

xm→xweakly in X,xm→xw-X,xm⇀xin X or w-X,xm⇀x.\begin{aligned} &x_m \to x \quad \text{weakly in } X, \\ &x_m \to x \quad \text{w-}X, \\ &x_m \rightharpoonup x \quad \text{in } X \text{ or w-}X, \\ &x_m \rightharpoonup x. \end{aligned}

Remark 16.17. If xm→x{x_m \to x} in X,{X,} that is, ∥xm−x∥X→0,{\|x_m - x\|_X \to 0,} which is called strong convergence, then xm⇀x.{x_m \rightharpoonup x.}

Weak convergence in L2((0,∞))L^2((0, \infty))

Example 16.18. Let X=L2((0,∞)),{X = L^2((0, \infty)),} with the Lebesgue measure λ.{\lambda.} By Theorem 16.13 (1) with p=p′=2,{p = p' = 2,} X∗X^* is isometrically isomorphic to L2((0,∞)).{L^2((0, \infty)).} Notice that, for (fm)⊂L2((0,∞)){(f_m) \subset L^2((0, \infty))} and f∈L2((0,∞)),{f \in L^2((0, \infty)),}

fm⇀f  ⟺  Tfm→Tffor every T∈(L2)∗  ⟺  ∫0∞fm g dλ→∫0∞fg dλfor every g∈L2((0,∞)),\begin{aligned} f_m \rightharpoonup f &\iff T f_m \to T f \quad \text{for every } T \in (L^2)^* \\ &\iff \int_0^\infty f_m \, g \, d\lambda \to \int_0^\infty f g \, d\lambda \quad \text{for every } g \in L^2((0, \infty)), \end{aligned}

by Definition 16.16 and by Theorem 16.13 (1). Let fm=χ[m,m+1]{f_m = \chi_{[m, m+1]}} for every m∈N.{m \in \mathbb{N}.} Then fm→0{f_m \to 0} almost everywhere, and fm↛0{f_m \nrightarrow 0} in L2.{L^2.} Indeed, ∥fm∥2=1{\|f_m\|_2 = 1} for every m.{m.} Moreover, for every g∈L2((0,∞)),{g \in L^2((0, \infty)),} by the Hölder inequality, Proposition 10.12,

∣∫0∞fm g dλ∣=∣∫mm+1g dλ∣≤(∫mm+1g2 dλ)12→m→∞0,\biggl| \int_0^\infty f_m \, g \, d\lambda \biggr| = \biggl| \int_m^{m+1} g \, d\lambda \biggr| \le \biggl( \int_m^{m+1} g^2 \, d\lambda \biggr)^{\frac{1}{2}} \xrightarrow[m \to \infty]{} 0,

since g∈L2((0,∞)).{g \in L^2((0, \infty)).} Hence

lim⁡m→∞∫0∞fm g dλ=0=∫0∞0⋅g dλfor every g∈L2,\lim_{m \to \infty} \int_0^\infty f_m \, g \, d\lambda = 0 = \int_0^\infty 0 \cdot g \, d\lambda \qquad \text{for every } g \in L^2,

that is, fm⇀0{f_m \rightharpoonup 0} weakly in L2((0,∞)).{L^2((0, \infty)).}

Weak convergence in LpL^p

Remark 16.19 (Weak convergence in LpL^p). Let (Ω,M,μ){(\Omega, \mathcal{M}, \mu)} be a σ\sigma-finite measure space, let p∈[1,∞),{p \in [1, \infty),} and let (fm)⊂Lp{(f_m) \subset L^p} and f∈Lp.{f \in L^p.}
(i) We have fm⇀f{f_m \rightharpoonup f} w-LpL^p if and only if ∫Ωfm g dμ→∫Ωfg dμ{\int_\Omega f_m \, g \, d\mu \to \int_\Omega f g \, d\mu} for every g∈Lp′.{g \in L^{p'}.}
(ii) If fm⇀f1{f_m \rightharpoonup f_1} w-LpL^p and fm→f2{f_m \to f_2} almost everywhere, then f1=f2{f_1 = f_2} almost everywhere.