Alessandro Palliccia

Real and Functional Analysis

The Banach–Steinhaus, open mapping, closed graph and Hahn–Banach theorems

1,416 words,

Recall the corollary of Baire’s lemma, Corollary 13.20.

Corollary 15.1. Let (X,d){(X, d)} be a nonempty complete metric space and let (Cm)(C_m) be a sequence of closed sets in XX such that ⋃mCm=X.{\bigcup_m C_m = X.} Then there exists m0m_0 such that int⁡(Cm0)≠∅.{\operatorname{int}(C_{m_0}) \neq \emptyset.}

Banach–Steinhaus

Theorem 15.2 (Banach–Steinhaus). Let (X,∥⋅∥X){(X, \|\cdot\|_X)} and (Y,∥⋅∥Y){(Y, \|\cdot\|_Y)} be Banach spaces and let F⊂L(X,Y){\mathcal{F} \subset \mathcal{L}(X, Y)} be a family of linear continuous operators from XX to Y.{Y.} Suppose that for every x∈X{x \in X} there exists Mx>0{M_x > 0} such that

∥Tx∥Y≤Mxfor every T∈F.\|Tx\|_Y \le M_x \qquad \text{for every } T \in \mathcal{F}.

Then there exists M>0{M > 0} such that ∥T∥L≤M{\|T\|_{\mathcal{L}} \le M} for every T∈F.{T \in \mathcal{F}.} That is, ∥Tx∥Y≤M∥x∥X{\|Tx\|_Y \le M \|x\|_X} for every T∈F{T \in \mathcal{F}} and every x∈X.{x \in X.}

Proof (complete). For every m∈N{m \in \mathbb{N}} let Cm={x∈X:∥Tx∥Y≤m for every T∈F}.{C_m = \{ x \in X : \|Tx\|_Y \le m \ \text{for every } T \in \mathcal{F} \}.} (i) The set CmC_m is closed. Let (xk)⊂Cm{(x_k) \subset C_m} be such that xk→x{x_k \to x} in X.{X.} Then, for every T∈F,{T \in \mathcal{F},}

∥Tx∥Y=lim⁡k→∞∥Txk∥Y≤m,\|Tx\|_Y = \lim_{k \to \infty} \|Tx_k\|_Y \le m,

since ∥Txk∥Y≤m{\|Tx_k\|_Y \le m} for every k.{k.} Indeed, Txk→Tx{Tx_k \to Tx} because TT is continuous, and the norm is continuous by Remark 10.3. Hence x∈Cm,{x \in C_m,} and CmC_m is closed by Lemma 1.20 (i).

(ii) The union of the sets CmC_m is X.{X.} Let x∈X.{x \in X.} By assumption there exists Mx>0{M_x > 0} such that ∥Tx∥Y≤Mx{\|Tx\|_Y \le M_x} for every T∈F.{T \in \mathcal{F}.} Then x∈Cm{x \in C_m} for every integer m≥Mx,{m \ge M_x,} since ∥Tx∥Y≤Mx≤m{\|Tx\|_Y \le M_x \le m} for all the operators of F.{\mathcal{F}.}

By Corollary 15.1, applied in X,{X,} there exists m0∈N{m_0 \in \mathbb{N}} such that int⁡(Cm0)≠∅.{\operatorname{int}(C_{m_0}) \neq \emptyset.} That is, there exist a point x0x_0 of XX and r>0{r > 0} such that Br(x0)⊂Cm0.{B_r(x_0) \subset C_{m_0}.}

We want to estimate the supremum of ∥Tz∥Y\|Tz\|_Y over z∈B1(0).{z \in B_1(0).} Let z∈B1(0).{z \in B_1(0).} Then x0+rz∈Br(x0)⊂Cm0,{x_0 + rz \in B_r(x_0) \subset C_{m_0},} hence ∥T(x0+rz)∥Y≤m0{\|T(x_0 + rz)\|_Y \le m_0} for every T∈F.{T \in \mathcal{F}.} On the other hand, by the linearity of TT and the triangle inequality,

∥T(x0+rz)∥Y≥∥T(rz)∥Y−∥Tx0∥Y=r∥Tz∥Y−∥Tx0∥Y.\|T(x_0 + rz)\|_Y \ge \|T(rz)\|_Y - \|Tx_0\|_Y = r \|Tz\|_Y - \|Tx_0\|_Y.

Hence, for every T∈F{T \in \mathcal{F}} and every z∈B1(0),{z \in B_1(0),}

∥Tz∥Y≤m0+∥Tx0∥Yr≤m0+Mx0r=:M,\|Tz\|_Y \le \frac{m_0 + \|Tx_0\|_Y}{r} \le \frac{m_0 + M_{x_0}}{r} =: M,

where the second inequality is the assumption at the point x0.{x_0.} Therefore ∥T∥L≤M{\|T\|_{\mathcal{L}} \le M} for every T∈F.{T \in \mathcal{F}.} In detail, for t∈(0,1){t \in (0, 1)} and xx in the closed unit ball of X,{X,} the vector txtx lies in B1(0),{B_1(0),} so that t∥Tx∥Y≤M,{t \|Tx\|_Y \le M,} and letting tt tend to 1 gives ∥Tx∥Y≤M.{\|Tx\|_Y \le M.} ■\blacksquare

Corollary 15.3. Let (X,∥⋅∥X){(X, \|\cdot\|_X)} and (Y,∥⋅∥Y){(Y, \|\cdot\|_Y)} be Banach spaces and let (Tm)⊂L(X,Y).{(T_m) \subset \mathcal{L}(X, Y).} Suppose that for every x∈X{x \in X} the limit of the sequence (Tmx)(T_m x) of points of YY exists in Y,{Y,} that is, (Tm)(T_m) converges pointwise on X.{X.} Then the map T ⁣:X→Y{T \colon X \to Y} given by

Tx:=lim⁡m→∞TmxTx := \lim_{m \to \infty} T_m x

belongs to L(X,Y).{\mathcal{L}(X, Y).} In particular, TT is continuous even though the convergence is only pointwise.

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

T(αx1+βx2)=lim⁡m→∞Tm(αx1+βx2)=lim⁡m→∞(αTmx1+βTmx2)=αTx1+βTx2.\begin{aligned} T(\alpha x_1 + \beta x_2) &= \lim_{m \to \infty} T_m(\alpha x_1 + \beta x_2) \\ &= \lim_{m \to \infty} \big( \alpha T_m x_1 + \beta T_m x_2 \big) = \alpha T x_1 + \beta T x_2. \end{aligned}

For every x∈X,{x \in X,} the sequence (Tmx)(T_m x) converges in Y,{Y,} hence it is bounded in Y.{Y.} That is, there exists Mx>0{M_x > 0} such that ∥Tmx∥Y≤Mx{\|T_m x\|_Y \le M_x} for every m.{m.} By the Banach–Steinhaus theorem, Theorem 15.2, applied to F={Tm:m∈N},{\mathcal{F} = \{ T_m : m \in \mathbb{N} \},} there exists M>0{M > 0} such that ∥Tm∥L≤M{\|T_m\|_{\mathcal{L}} \le M} for every m.{m.} Hence, for every z∈B1(0),{z \in B_1(0),}

∥Tz∥Y=lim⁡m→∞∥Tmz∥Y≤M,\|Tz\|_Y = \lim_{m \to \infty} \|T_m z\|_Y \le M,

where the equality is the continuity of the norm, Remark 10.3, and the inequality holds because ∥Tmz∥Y≤∥Tm∥L∥z∥X≤M{\|T_m z\|_Y \le \|T_m\|_{\mathcal{L}} \|z\|_X \le M} for every m.{m.} Therefore ∥T∥L≤M,{\|T\|_{\mathcal{L}} \le M,} as at the end of the proof of Theorem 15.2, and the linear map TT is bounded, hence continuous: T∈L(X,Y).{T \in \mathcal{L}(X, Y).} ■\blacksquare

Open mapping theorem

Definition 15.4 (Open function). Let (X,dX){(X, d_X)} and (Y,dY){(Y, d_Y)} be metric spaces. A function f ⁣:(X,dX)→(Y,dY){f \colon (X, d_X) \to (Y, d_Y)} is said to be open if f(U){f(U)} is open in YY for every open set U⊂X.{U \subset X.}

Theorem 15.5 (Open mapping). Let (X,∥⋅∥X){(X, \|\cdot\|_X)} and (Y,∥⋅∥Y){(Y, \|\cdot\|_Y)} be Banach spaces and let T∈L(X,Y).{T \in \mathcal{L}(X, Y).} If TT is surjective, then TT is open.

Proof (sketch). For r>0{r > 0} and points xx of XX and yy of Y,{Y,} let BrX(x)B_r^X(x) and BrY(y)B_r^Y(y) be the open balls of radius rr centred at xx in XX and at yy in Y.{Y.} (i) It is sufficient to prove that T(B1X(0))⊃BrY(y0){T(B_1^X(0)) \supset B_r^Y(y_0)} for some y0∈Y{y_0 \in Y} and r>0.{r > 0.} (ii) We can prove that T(B1X(0))‾⊃BrY(y0){\overline{T(B_1^X(0))} \supset B_r^Y(y_0)} for some y0∈Y{y_0 \in Y} and r>0.{r > 0.} For (ii), the sets T(BmX(0))‾\overline{T(B_m^X(0))} with m≥1{m \ge 1} are closed and, since TT is surjective,

⋃m≥1T(BmX(0))‾=R(T)=Y.\bigcup_{m \ge 1} \overline{T\big( B_m^X(0) \big)} = R(T) = Y.

Hence, by Corollary 15.1, applied in Y,{Y,} there exist an integer m0≥1,{m_0 \ge 1,} a point z0∈Y{z_0 \in Y} and ρ>0{\rho > 0} such that

BρY(z0)⊂T(Bm0X(0))‾.B_\rho^Y(z_0) \subset \overline{T\big( B_{m_0}^X(0) \big)}.

In detail, T(Bm0X(0))=m0 T(B1X(0)){T(B_{m_0}^X(0)) = m_0 \, T(B_1^X(0))} by linearity, so that (ii) holds with y0=z0/m0{y_0 = z_0/m_0} and r=ρ/m0.{r = \rho/m_0.} ■\blacksquare

Corollary 15.6. Let (X,∥⋅∥X){(X, \|\cdot\|_X)} and (Y,∥⋅∥Y){(Y, \|\cdot\|_Y)} be Banach spaces and let T∈L(X,Y).{T \in \mathcal{L}(X, Y).} Suppose that TT is bijective. Then T−1∈L(Y,X).{T^{-1} \in \mathcal{L}(Y, X).}

Proof (complete). Consider T−1 ⁣:Y→X.{T^{-1} \colon Y \to X.} We want to prove that (T−1)−1(A){(T^{-1})^{-1}(A)} is open in YY for every open set A⊂X,{A \subset X,} that is, that T−1T^{-1} is continuous, as in Definition 1.17. In detail, T−1T^{-1} is linear, as the inverse of a bijective linear map, so that its continuity gives T−1∈L(Y,X).{T^{-1} \in \mathcal{L}(Y, X).} Let A⊂X{A \subset X} be open. Then

(T−1)−1(A)=T(A),(T^{-1})^{-1}(A) = T(A),

which is open, because TT is an open map by Theorem 15.5, the open mapping theorem. ■\blacksquare

Corollary 15.7 (Embeddings). Let (X,∥⋅∥X){(X, \|\cdot\|_X)} and (Y,∥⋅∥Y){(Y, \|\cdot\|_Y)} be Banach spaces and let T∈L(X,Y){T \in \mathcal{L}(X, Y)} be injective, with R(T)R(T) closed. Then T ⁣:(X,∥⋅∥X)→(R(T),∥⋅∥Y){T \colon (X, \|\cdot\|_X) \to (R(T), \|\cdot\|_Y)} is an isomorphism, that is, T−1 ⁣:R(T)→X{T^{-1} \colon R(T) \to X} is linear and continuous.

Proof (complete). The space (R(T),∥⋅∥Y){(R(T), \|\cdot\|_Y)} is a Banach space. Indeed, R(T)R(T) is a subspace of Y,{Y,} closed by assumption, and a closed subspace of a Banach space is a Banach space. Hence T ⁣:(X,∥⋅∥X)→(R(T),∥⋅∥Y){T \colon (X, \|\cdot\|_X) \to (R(T), \|\cdot\|_Y)} is a bijective map between two Banach spaces, and T−1 ⁣:R(T)→X{T^{-1} \colon R(T) \to X} is continuous by Corollary 15.6. ■\blacksquare

Corollary 15.8 (Equivalent norms). Let XX be a vector space with two norms ∥⋅∥a\|\cdot\|_a and ∥⋅∥b\|\cdot\|_b such that (X,∥⋅∥a){(X, \|\cdot\|_a)} and (X,∥⋅∥b){(X, \|\cdot\|_b)} are both Banach spaces. Suppose that there exists C1>0{C_1 > 0} such that ∥x∥b≤C1∥x∥a{\|x\|_b \le C_1 \|x\|_a} for every x∈X.{x \in X.} Then there exists C2>0{C_2 > 0} such that ∥x∥a≤C2∥x∥b{\|x\|_a \le C_2 \|x\|_b} for every x∈X,{x \in X,} that is, ∥⋅∥a\|\cdot\|_a and ∥⋅∥b\|\cdot\|_b are equivalent.

Proof (complete). The assumption says that the map id⁡ ⁣:(X,∥⋅∥a)→(X,∥⋅∥b),{\operatorname{id} \colon (X, \|\cdot\|_a) \to (X, \|\cdot\|_b),} x↦x,{x \mapsto x,} is continuous: ∥id⁡x∥b=∥x∥b≤C1∥x∥a{\|\operatorname{id} x\|_b = \|x\|_b \le C_1 \|x\|_a} for every xx in X.{X.} The map id⁡\operatorname{id} is also bijective. Hence, by Corollary 15.6, the map id⁡ ⁣:(X,∥⋅∥b)→(X,∥⋅∥a){\operatorname{id} \colon (X, \|\cdot\|_b) \to (X, \|\cdot\|_a)} is also continuous. That is, there exists C2>0{C_2 > 0} such that ∥id⁡x∥a=∥x∥a≤C2∥x∥b{\|\operatorname{id} x\|_a = \|x\|_a \le C_2 \|x\|_b} for every x∈X.{x \in X.} ■\blacksquare

Corollary 15.9 (Isometries). Let (X,∥⋅∥X){(X, \|\cdot\|_X)} and (Y,∥⋅∥Y){(Y, \|\cdot\|_Y)} be Banach spaces and let T∈L(X,Y){T \in \mathcal{L}(X, Y)} be an isometry, that is, ∥Tx∥Y=∥x∥X{\|Tx\|_Y = \|x\|_X} for every x∈X.{x \in X.} Then:
(1) R(T)R(T) is closed;
(2) T ⁣:(X,∥⋅∥X)→(R(T),∥⋅∥Y){T \colon (X, \|\cdot\|_X) \to (R(T), \|\cdot\|_Y)} is an isometric isomorphism.

Proof (complete). (1) Let (ym)⊂R(T){(y_m) \subset R(T)} be such that ym→y{y_m \to y} in Y.{Y.} Since ym∈R(T),{y_m \in R(T),} there exists xm∈X{x_m \in X} such that ym=Txm.{y_m = Tx_m.} Now, for all mm and n,{n,}

∥xm−xn∥X=∥T(xm−xn)∥Y=∥Txm−Txn∥Y=∥ym−yn∥Y.\|x_m - x_n\|_X = \|T(x_m - x_n)\|_Y = \|Tx_m - Tx_n\|_Y = \|y_m - y_n\|_Y.

Since (ym)(y_m) converges, (ym)(y_m) is Cauchy in Y.{Y.} Hence (xm)(x_m) is Cauchy in X,{X,} and, since XX is a Banach space, xm→x{x_m \to x} in XX for some x∈X.{x \in X.} Since TT is continuous, y=lim⁡mym=lim⁡mTxm=Tx,{y = \lim_m y_m = \lim_m Tx_m = Tx,} hence y∈R(T),{y \in R(T),} and R(T)R(T) is closed by Lemma 1.20 (i).

(2) The range R(T)R(T) is closed by (1), and TT is injective. Indeed, if Tx=0,{Tx = 0,} then ∥x∥X=∥Tx∥Y=0.{\|x\|_X = \|Tx\|_Y = 0.} By Corollary 15.7, T ⁣:(X,∥⋅∥X)→(R(T),∥⋅∥Y){T \colon (X, \|\cdot\|_X) \to (R(T), \|\cdot\|_Y)} is then an isomorphism, and it is also an isometry, that is, an isometric isomorphism. ■\blacksquare

Corollary 15.10 (Maps bounded above and below). Let (X,∥⋅∥X){(X, \|\cdot\|_X)} and (Y,∥⋅∥Y){(Y, \|\cdot\|_Y)} be Banach spaces and let T∈L(X,Y).{T \in \mathcal{L}(X, Y).} Suppose that there exist C1,C2>0{C_1, C_2 > 0} such that C1∥x∥X≤∥Tx∥Y≤C2∥x∥X{C_1 \|x\|_X \le \|Tx\|_Y \le C_2 \|x\|_X} for every x∈X.{x \in X.} Then:
(0) TT is injective;
(1) R(T)R(T) is closed;
(2) T ⁣:(X,∥⋅∥X)→(R(T),∥⋅∥Y){T \colon (X, \|\cdot\|_X) \to (R(T), \|\cdot\|_Y)} is an isomorphism.

Closed graph

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. The graph of TT is

graph⁡(T)={(x,Tx):x∈X}⊂X×Y.\operatorname{graph}(T) = \{ (x, Tx) : x \in X \} \subset X \times Y.

We say that graph⁡(T)\operatorname{graph}(T) is closed if, for every sequence (xm,Txm)⊂graph⁡(T){(x_m, Tx_m) \subset \operatorname{graph}(T)} that converges to some (x,y),{(x, y),} we have y=Tx.{y = Tx.} Here (xm,Txm)(x_m, Tx_m) converges to (x,y)(x, y) when xm→x{x_m \to x} in XX and Txm→y{Tx_m \to y} in Y.{Y.}

Theorem 15.11 (Closed graph theorem). Let (X,∥⋅∥X){(X, \|\cdot\|_X)} and (Y,∥⋅∥Y){(Y, \|\cdot\|_Y)} be Banach spaces and let T ⁣:X→Y{T \colon X \to Y} be linear. Then TT is continuous if and only if graph⁡(T)⊂X×Y{\operatorname{graph}(T) \subset X \times Y} is closed.

Remark 15.12. (i) In Theorem 15.11, the implication from left to right follows from the continuity of T:{T{:}} if (xm,Txm)(x_m, Tx_m) converges to (x,y),{(x, y),} then y=lim⁡mTxm=Tx.{y = \lim_m Tx_m = Tx.}
(ii) The converse is false without linearity: the function f ⁣:R→R{f \colon \mathbb{R} \to \mathbb{R}} whose graph is drawn in Figure 15.1 is discontinuous, and graph⁡(f)\operatorname{graph}(f) is closed.

graph f

Figure 15.1. The drawing of Remark 15.12 (ii): the graph of f,{f,} made of a branch over the negative half-line, the dot at the origin and a branch over the positive half-line.

Proof (complete). The implication from left to right is Remark 15.12 (i). Conversely, suppose that graph⁡(T)\operatorname{graph}(T) is closed, and let

∥x∥0:=∥x∥X+∥Tx∥Yfor every x∈X.\|x\|_0 := \|x\|_X + \|Tx\|_Y \qquad \text{for every } x \in X.

The function ∥⋅∥0\|\cdot\|_0 is a norm on X.{X.} (i) If ∥x∥0=0,{\|x\|_0 = 0,} then ∥x∥X=0,{\|x\|_X = 0,} hence x=0.{x = 0.} (ii) For every α∈R{\alpha \in \mathbb{R}} and every x∈X,{x \in X,}

∥αx∥0=∥αx∥X+∥T(αx)∥Y=∣α∣(∥x∥X+∥Tx∥Y).\|\alpha x\|_0 = \|\alpha x\|_X + \|T(\alpha x)\|_Y = |\alpha| \big( \|x\|_X + \|Tx\|_Y \big).

(iii) The function ∥⋅∥0\|\cdot\|_0 satisfies the triangle inequality. Moreover, (X,∥⋅∥0){(X, \|\cdot\|_0)} is a Banach space. Let (xm)⊂X{(x_m) \subset X} be Cauchy with respect to ∥⋅∥0:{\|\cdot\|_0{:}} for every ε>0{\varepsilon > 0} there exists NN such that

∥xm−xn∥0=∥xm−xn∥X+∥Txm−Txn∥Y≤εfor all m,n≥N.\|x_m - x_n\|_0 = \|x_m - x_n\|_X + \|Tx_m - Tx_n\|_Y \le \varepsilon \qquad \text{for all } m, n \ge N.

Hence (xm)(x_m) is Cauchy with respect to ∥⋅∥X\|\cdot\|_X and (Txm)(Tx_m) is Cauchy with respect to ∥⋅∥Y.{\|\cdot\|_Y.} Since XX and YY are Banach spaces, there exist points xx of XX and yy of YY such that xm→x{x_m \to x} in XX and Txm→y{Tx_m \to y} in Y.{Y.} Since graph⁡(T)\operatorname{graph}(T) is closed, y=Tx.{y = Tx.} Therefore

∥xm−x∥0=∥xm−x∥X+∥Txm−Tx∥Y=∥xm−x∥X+∥Txm−y∥Y,\|x_m - x\|_0 = \|x_m - x\|_X + \|Tx_m - Tx\|_Y = \|x_m - x\|_X + \|Tx_m - y\|_Y,

where both terms on the right tend to 0.{0.} Hence ∥xm−x∥0→0,{\|x_m - x\|_0 \to 0,} that is, xm→x{x_m \to x} with respect to ∥⋅∥0.{\|\cdot\|_0.}

Now (X,∥⋅∥X){(X, \|\cdot\|_X)} and (X,∥⋅∥0){(X, \|\cdot\|_0)} are both Banach spaces, and ∥x∥X≤∥x∥X+∥Tx∥Y=∥x∥0{\|x\|_X \le \|x\|_X + \|Tx\|_Y = \|x\|_0} for every x∈X.{x \in X.} By Corollary 15.8, with ∥⋅∥0\|\cdot\|_0 in place of ∥⋅∥a\|\cdot\|_a and ∥⋅∥X\|\cdot\|_X in place of ∥⋅∥b,{\|\cdot\|_b,} there exists C>0{C > 0} such that ∥x∥0≤C∥x∥X{\|x\|_0 \le C \|x\|_X} for every x∈X.{x \in X.} Since ∥x∥0=∥x∥X+∥Tx∥Y≥∥Tx∥Y,{\|x\|_0 = \|x\|_X + \|Tx\|_Y \ge \|Tx\|_Y,} it follows that ∥Tx∥Y≤C∥x∥X{\|Tx\|_Y \le C \|x\|_X} for every x∈X,{x \in X,} that is, TT is bounded, and therefore continuous. ■\blacksquare

Hahn–Banach theorems

Remark 15.13. (i) Let (X,∥⋅∥X){(X, \|\cdot\|_X)} be a normed space. Then X∗=L(X,R){X^* = \mathcal{L}(X, \mathbb{R})} is a Banach space, whose norm is written ∥⋅∥X∗.{\|\cdot\|_{X^*}.}
(ii) Let V⊂X{V \subset X} be a subspace. Then (V,∥⋅∥X){(V, \|\cdot\|_X)} is a normed space, and V∗=L(V,R){V^* = \mathcal{L}(V, \mathbb{R})} is a Banach space, whose norm is written ∥⋅∥V∗.{\|\cdot\|_{V^*}.}
(iii) If T∈X∗,{T \in X^*,} it is possible to restrict TT to V:{V{:}} the restriction T∣V ⁣:V→R{T|_V \colon V \to \mathbb{R}} belongs to V∗.{V^*.}

The question raised by Remark 15.13 (iii) is the converse. If L∈V∗,{L \in V^*,} is it possible to extend LL to some T∈X∗?{T \in X^*?} The answer is yes.

Theorem 15.14 (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}

The proof of Theorem 15.14 uses Zorn’s lemma, which is equivalent to the axiom of choice. Let SS be a set. An order on SS is a relation ≤\le on SS such that, for all x,y,z∈S,{x, y, z \in S,}

x≤x,x≤y and y≤x  ⟹  x=y,x≤y and y≤z  ⟹  x≤z.x \le x, \qquad x \le y \ \text{and} \ y \le x \implies x = y, \qquad x \le y \ \text{and} \ y \le z \implies x \le z.

Lemma 15.15 (Zorn). Let SS be a set endowed with an order ≤.{{\le}.} Suppose that, for every family {xα:α∈A}⊂S{\{ x_\alpha : \alpha \in A \} \subset S} such that

xα≤xβ  or  xβ≤xαfor all α,β∈A,x_\alpha \le x_\beta \ \text{ or } \ x_\beta \le x_\alpha \qquad \text{for all } \alpha, \beta \in A,

there exists x∈S{x \in S} such that xα≤x{x_\alpha \le x} for every α∈A,{\alpha \in A,} as in Figure 15.2. Then there exists xˉ∈S{\bar{x} \in S} such that

xˉ≤y  ⟹  xˉ=yfor every y∈S.\bar{x} \le y \implies \bar{x} = y \qquad \text{for every } y \in S.
{xα}x

Figure 15.2. The drawing of Lemma 15.15: elements of S,{S,} some joined by segments, a family {xα}\{ x_\alpha \} inside the closed curve, as in the hypothesis, and at its top the element xx with xα≤x{x_\alpha \le x} for every α∈A.{\alpha \in A.}

Proof (sketch). We prove Theorem 15.14 with Lemma 15.15. Let L∈V∗.{L \in V^*.} As in Figure 15.3, let

S={R ⁣:DR→R : DR⊂X is a subspace,V⊂DR,R∈L(DR,R),R∣V=L, ∥R∥DR∗=∥L∥V∗}.S = \left\{ R \colon D_R \to \mathbb{R} \ : \ \begin{aligned} &D_R \subset X \text{ is a subspace}, \\ &V \subset D_R, \\ &R \in \mathcal{L}(D_R, \mathbb{R}), \\ &R|_V = L, \ \|R\|_{D_R^*} = \|L\|_{V^*} \end{aligned} \right\}.
VDRX

Figure 15.3. The drawing of the set SS in the proof of Theorem 15.14: in X,{X,} a subspace DR,{D_R,} drawn as a plane, that contains the subspace V,{V,} drawn as a line.

For two elements R1R_1 and R2R_2 of SS we define R1≤R2{R_1 \le R_2} if DR1⊂DR2{D_{R_1} \subset D_{R_2}} and R2∣DR1=R1.{R_2|_{D_{R_1}} = R_1.} By Lemma 15.15, SS has a maximal element T ⁣:DT→R.{T \colon D_T \to \mathbb{R}.} That is, T≤R{T \le R} implies T=R{T = R} for every R∈S.{R \in S.} The goal is DT=X.{D_T = X.} Suppose, by contradiction, that DT≠X,{D_T \neq X,} and take x0∈X∖DT,{x_0 \in X \setminus D_T,} as in Figure 15.4.

x0DT

Figure 15.4. The drawing of the step by contradiction in the proof of Theorem 15.14: the subspace DT,{D_T,} drawn as a plane, and a point x0x_0 outside it, joined to a point of DTD_T by a segment.

The idea is to find T~ ⁣:Span⁡(DT,x0)→R,{\tilde{T} \colon \operatorname{Span}(D_T, x_0) \to \mathbb{R},} where

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

such that T~∣DT=T{\tilde{T}|_{D_T} = T} and T~∈S.{\tilde{T} \in S.} Then T≤T~{T \le \tilde{T}} and T≠T~,{T \neq \tilde{T},} against the maximality of T.{T.} ■\blacksquare