Bourbaki

Appendix

Complements on topological vector spaces

1. Bilinear forms and linear mappings

Let (F1,G1),(F2,G2)(F_1, G_1), (F_2, G_2) be two pairs of (real or complex) vector spaces in separating duality (TVS, II, §6, No. 1); assume each of these spaces to be equipped with the corresponding weak topology (loc. cit., No. 2); if A and B are any two of these spaces, as usual we denote by L(A;B)\mathscr{L}(A; B) the vector space of continuous linear mappings of A into B, and by B(A,B)\mathfrak{B}(A, B) the vector space of separately continuous bilinear forms on A×BA \times B.

For every separately continuous bilinear form Φ\Phi on F1×F2F_1 \times F_2, x1↦Φ(x1,x2)x_1 \mapsto \Phi(x_1, x_2) is a continuous linear form on F1F_1, therefore there exists one and only one element rΦ(x2)∈G1{}^r\Phi(x_2) \in G_1 such that

Φ(x1,x2)=⟨x1,rΦ(x2)⟩ \Phi(x_1, x_2) = \langle x_1, {}^r\Phi(x_2) \rangle

for x1∈F1,x2∈F2x_1 \in F_1, x_2 \in F_2 (TVS, III, §5, No. 1, (1)). Moreover, this formula shows that the mapping x2↦rΦ(x2)x_2 \mapsto {}^r\Phi(x_2) is linear and continuous for the (weak) topologies of F2F_2 and G1G_1. Conversely, for every continuous linear mapping uu of F2F_2 into G1G_1, (x1,x2)↦Φ(x1,x2)=⟨x1,u(x2)⟩(x_1, x_2) \mapsto \Phi(x_1, x_2) = \langle x_1, u(x_2) \rangle is a separately continuous bilinear form on F1×F2F_1 \times F_2, and rΦ=u{}^r\Phi = u. One thus defines an isomorphism r:Φ↦rΦr : \Phi \mapsto {}^r\Phi of B(F1,F2)\mathfrak{B}(F_1, F_2) onto L(F2;G1)\mathscr{L}(F_2; G_1), said to be canonical.

Similarly, the formula

Φ(x1,x2)=⟨lΦ(x1),x2⟩ \Phi(x_1, x_2) = \langle {}^l\Phi(x_1), x_2 \rangle

defines a canonical isomorphism l:Φ↦lΦl : \Phi \mapsto {}^l\Phi of B(F1,F2)\mathfrak{B}(F_1, F_2) onto L(F1;G2)\mathscr{L}(F_1; G_2); and one obviously has the commutative diagram

B(F1,F2)ll−1r−1L(F1;G2)←tL(F2;G1) \begin{array}{ccc} \mathscr{B}(F_1, F_2) & & \\ l & l^{-1} & r^{-1} \\ & & \\ \mathscr{L}(F_1; G_2) & \xleftarrow{t} & \mathscr{L}(F_2; G_1) \end{array}

where tt is the isomorphism of transposition u↦tuu \mapsto {}^t u. In view of the definition of the weak topologies on G1G_1 and G2G_2, it is moreover immediate that when B(F1,F2)\mathscr{B}(F_1, F_2), L(F1;G2)\mathscr{L}(F_1; G_2) and L(F2;G1)\mathscr{L}(F_2; G_1) are equipped with the topology of pointwise convergence, the isomorphisms in the preceding diagram are isomorphisms for the topological vector space structures.

Now let E,FE, F be two Hausdorff locally convex spaces, E′,F′E', F' their respective duals; we denote by Eσ,FσE_\sigma, F_\sigma the spaces E,FE, F equipped with the weakened topologies σ(E,E′)\sigma(E, E'), σ(F,F′)\sigma(F, F'), and by Es′,Fs′E'_s, F'_s the spaces E′,F′E', F' equipped with the weak topologies σ(E′,E)\sigma(E', E), σ(F′,F)\sigma(F', F). Thus, the preceding remarks establish canonical isomorphisms between the three spaces B(Eσ,Fs′)\mathscr{B}(E_\sigma, F'_s), L(Eσ;Fσ)\mathscr{L}(E_\sigma; F_\sigma) and L(Fs′;Es′)\mathscr{L}(F'_s; E'_s), and also between the three spaces B(Eσ,Fσ)\mathscr{B}(E_\sigma, F_\sigma), L(Eσ;Fs′)\mathscr{L}(E_\sigma; F'_s) and L(Fσ;Es′)\mathscr{L}(F_\sigma; E'_s). One will observe that B(Eσ,Fσ)\mathscr{B}(E_\sigma, F_\sigma) is also equal to the space B(E,F)\mathscr{B}(E, F) of separately continuous bilinear forms on E×FE \times F (EE and FF being equipped with their original topologies), since every continuous linear form on EE (resp. FF) is continuous on EσE_\sigma (resp. FσF_\sigma) and conversely (TVS, II, §6, No. 1 and No. 2, Prop. 3).

Let B(E,F)\mathscr{B}(E, F) be the space of continuous bilinear forms on E×FE \times F (EE and FF being equipped with their original topologies); then B(E,F)⊂B(E,F)\mathscr{B}(E, F) \subset \mathscr{B}(E, F).

Proposition 1

For a bilinear form Φ∈B(E,F)\Phi \in \mathscr{B}(E, F) to belong to B(E,F)\mathscr{B}(E, F), it is necessary and sufficient that there exist a neighborhood of 0 in EE whose image under lΦ{}^l \Phi is an equicontinuous subset of F′F'.

For, to say that Φ\Phi is continuous means that there exists a balanced convex neighborhood VV (resp. WW) of 0 in EE (resp. FF) such that ∣Φ(x,y)∣≤1|\Phi(x, y)| \leq 1 for x∈Vx \in V, y∈Wy \in W; this may be written ∣⟨lΦ(x),y⟩∣≤1|\langle {}^l \Phi(x), y \rangle| \leq 1 for x∈Vx \in V, y∈Wy \in W, or also lΦ(V)⊂W∘{}^l \Phi(V) \subset W^\circ; whence the proposition, taking into account the fact that every equicontinuous subset of F′F' is contained in the polar of a neighborhood of 0 in FF.

Corollary

If Φ\Phi is a continuous bilinear form on E×FE \times F, then lΦ{}^l \Phi is a continuous linear mapping of EE into the strong dual Fb′F'_b of FF. If, moreover, EE and FF are normed spaces, then ∥lΦ∥=∥Φ∥\|{}^l \Phi\| = \|\Phi\|.

The first assertion follows from Prop. 1 and the fact that every neighborhood of 0 in Fb′F'_b absorbs every equicontinuous subset of F′F'. If EE and FF are normed, then

∥Φ∥=sup⁡∥x∥≤1,∥y∥≤1∣Φ(x,y)∣=sup⁡∥x∥≤1(sup⁡∥y∥≤1∣⟨lΦ(x),y⟩∣)=sup⁡∥x∥≤1∥lΦ(x)∥=∥lΦ∥, \|\Phi\| = \sup_{\|x\| \leq 1, \|y\| \leq 1} |\Phi(x, y)| = \sup_{\|x\| \leq 1} \left( \sup_{\|y\| \leq 1} |\langle l \Phi(x), y \rangle| \right) = \sup_{\|x\| \leq 1} \|l \Phi(x)\| = \|l \Phi\|,

whence the second assertion.

Interchanging the roles of E and F, one obtains results analogous to Prop. 1 and its corollary for the linear mappings rΦr \Phi; we leave to the reader the task of stating them.

2. Some types of spaces having the property (GDF)

We already know that every Fréchet space has the property (GDF) (TVS, I, §3, No. 3, Cor. 5 of Th. 1).

Proposition 2

— Let E be a vector space, (Fα)α∈A(F_\alpha)_{\alpha \in A} a family of locally compact spaces having the property (GDF), and for each α∈A\alpha \in A let hαh_\alpha be a linear mapping of FαF_\alpha into E. If E is equipped with the finest locally convex topology that makes the hαh_\alpha continuous, then E has the property (GDF).

Let uu be a linear mapping of E into a Banach space B, such that every limit in E×BE \times B of every convergent sequence of points of the graph Γ\Gamma of uu also belongs to Γ\Gamma. It suffices to show that, for every α∈A\alpha \in A, u∘hαu \circ h_\alpha is continuous on FαF_\alpha (TVS, II, §4, No. 4, Prop. 5). Now, let (xn)(x_n) be a sequence of elements of FαF_\alpha having a limit aa and such that the sequence (u(hα(xn)))\left(u(h_\alpha(x_n))\right) has a limit b∈Bb \in B. Since hαh_\alpha is continuous, hα(a)h_\alpha(a) is a limit of the sequence (hα(xn))(h_\alpha(x_n)) in E; therefore by hypothesis b=u(hα(a))b = u(h_\alpha(a)) and, since FαF_\alpha has the property (GDF), u∘hαu \circ h_\alpha is continuous.

Corollary

— Every quotient space of a locally convex space having the property (GDF) has the property (GDF).

Proposition 3

— The strong dual of a reflexive Fréchet space has the property (GDF).

This is a consequence of Prop. 2 and the following lemma (or TVS, IV, §3, No. 4, Prop. 4):

Lemma

Let F be a Fréchet space, F′F' its strong dual, F′′F'' its bidual. If every subset of F′′F'', bounded for σ(F′′,F′)\sigma(F'', F'), is contained in the closure (for σ(F′′,F′)\sigma(F'', F')) of a bounded subset of F, then F′F' is the direct limit of a sequence of Banach spaces.

For, let (Vn)(V_n) be a decreasing fundamental sequence of convex, balanced and closed neighborhoods of 0 in F. For every integer nn, let GnG_n be the linear subspace of F′F' generated by the polar Vn∘V_n^\circ of VnV_n. In GnG_n, Vn∘V_n^\circ is an absorbent convex set, therefore its gauge pnp_n is a norm on GnG_n; moreover, Vn∘V_n^\circ is a complete subset of the strong dual F′F' (TVS, III, §3, No. 8, Prop. 11); thus GnG_n, equipped with the norm pnp_n, is a Banach space (GT, III, §3, No. 5, Cor. 2 of Prop. 9). We are going to show that the strong topology on F′F' is the direct limit of these Banach space topologies on the GnG_n, or again that, for a strongly closed, balanced, convex subset U of F′F' to be a strong neighborhood of 0, it is necessary and sufficient that it absorb each of the Vn∘V_n^\circ. It is obvious that this condition is necessary; to see that it is sufficient, it will suffice to prove that U contains a barrel of F′F'. Indeed, its polar U∘U^\circ in F′′F'' will then be bounded for σ(F′′,F′)\sigma(F'', F'), hence, by hypothesis, will be contained in the closure (for σ(F′′,F′)\sigma(F'', F')) of a bounded subset B of F, from which one can conclude that U (which is closed for σ(F′,F′′)\sigma(F', F'')) contains the strong neighborhood B∘B^\circ of 0 (TVS, II, §6, No. 3, Th. 1 and §8, No. 4).

By hypothesis, for every integer nn there exists a number λn>0\lambda_n > 0 such that λnVn∘⊂12U\lambda_n V_n^\circ \subset \frac{1}{2}U; let AnA_n be the convex envelope of the union of the λiVi∘\lambda_i V_i^\circ for i≤ni \leq n. Then An⊂12UA_n \subset \frac{1}{2}U for every nn; let W be the union of the AnA_n; W is a convex, balanced, absorbent set contained in 12U\frac{1}{2}U, and it will suffice to show that its strong closure (which is a barrel) is contained in U.

Thus, let x′x' be a point of F′F' not belonging to U. Since each of the Vn∘V_n^\circ is compact for σ(F′,F)\sigma(F', F), the same is true of the AnA_n (TVS, II, §2, No. 6, Prop. 15), and since x′∉2Anx' \notin 2A_n, there exists an element xnx_n belonging to the polar of AnA_n in F such that ⟨x′,xn⟩=2\langle x', x_n \rangle = 2 (TVS, II, §5, No. 3, Prop. 4). The sequence (xn)(x_n) is bounded in F: for, every y′∈F′y' \in F' belongs to some Vk∘V_k^\circ, consequently ∣⟨y′,xn⟩∣≤λk−1|\langle y', x_n \rangle| \leq \lambda_k^{-1} for n≥kn \geq k, whence our assertion (TVS, IV, §1, No. 1, Prop. 1). Let C be a bounded, balanced convex set in F containing all the xnx_n; C∘C^\circ is then a neighborhood of 0 in F′F', and the polar C∘∘C^{\circ\circ} of C∘C^\circ in F′′F'' is compact for σ(F′′,F′)\sigma(F'', F') (TVS, III, §3, No. 4, Cor. 2 of Prop. 4 and No. 5, Prop. 7). Thus one sees that the sequence (xn)(x_n) has a cluster point x′′x'' in F′′F'' for σ(F′′,F′)\sigma(F'', F'); obviously ⟨x′,x′′⟩=2\langle x', x'' \rangle = 2 and, on the other hand, x′′x'' belongs to the polar of AnA_n in F′′F'' for every nn, hence to the polar W∘W^\circ of W in F′′F''. From this, one concludes that x′∉W∘∘x' \notin W^{\circ\circ}, hence is not in the closure of W for σ(F′,F′′)\sigma(F', F'') (TVS, II, §6, No. 3, Th. 1 and §8, No. 4), therefore a fortiori for the strong topology, which completes the proof.

Exercises

Transcribed from the printed volume. Not translated.