Bourbaki

Proposition 1

Tag 07DQ, permanent. Cite it as [Bourbaki, Tag 07DQ].

Topological Vector Spaces, chapter II, § 3. The Hahn-Banach Theorem (analytic form)

— Let E be a preordered vector space and V be a vector subspace of E such that every element of E is bounded above by an element of V. Given a linear form f on V that is positive for the preordered vector space structure of V (induced by that of E) there exists a non-empty set SfS_f of positive linear forms on E, each being an extension of f. If h∈Sfh \in S_f then the values h(a)h(a) for a∈Ea \in E lie in the interval [α′,α′′][\alpha', \alpha''], where

α′=sup⁡z∈V,z≤af(z),α′′=inf⁡y∈V,y≥af(y). \alpha' = \sup_{z \in V, z \leq a} f(z), \quad \alpha'' = \inf_{y \in V, y \geq a} f(y).

I. Special case.

Suppose firstly that E=V+RaE = V + \mathbf{R}a. Since the proposition is trivial if a∈Va \in V, we confine ourselves to the case a∉Va \notin V. The conditions on V imply that the set A′′A'' of y∈Vy \in V such that a≤ya \leq y is not empty; similarly the set A′A' of z∈Vz \in V such that −z≥−a-z \geq -a (\emph{i.e.} z≤az \leq a) is not empty. For y∈A′′y \in A'' and z∈A′z \in A', we have z≤a≤yz \leq a \leq y, and thus by hypothesis f(z)≤f(y)f(z) \leq f(y). Thus α′,α′′\alpha', \alpha'' are finite and α′≤α′′\alpha' \leq \alpha''. Any linear form f1f_1 on E that extends ff is completely determined by f1(a)f_1(a) and for all λ∈R\lambda \in \mathbf{R} and all x∈Vx \in V, we have

f1(x+λa)=f(x)+λf1(a). f_1(x + \lambda a) = f(x) + \lambda f_1(a).

Thus f1f_1 is positive if and only if the relations

x∈V,λ∈R,x+λa≥0 x \in V, \quad \lambda \in \mathbf{R}, \quad x + \lambda a \geq 0

imply

f(x)+λf1(a)≥0. f(x) + \lambda f_1(a) \geq 0.

As f(μx)=μf(x)f(\mu x) = \mu f(x) and the relations x≥0x \geq 0 and μx≥0\mu x \geq 0 are equivalent for μ>0\mu > 0, it is sufficient to show that (2) implies (3) in the particular cases λ=0,λ=1\lambda = 0, \lambda = 1 and λ=−1\lambda = -1. For λ=0\lambda = 0, the fact that (2) implies (3) follows from the hypothesis that ff is positive. For λ=1\lambda = 1, to say that (2) implies (3) means that for −x∈A′-x \in A', we have f1(a)≥f(−x)f_1(a) \geq f(-x), \emph{i.e.} f1(a)≥α′f_1(a) \geq \alpha'; for λ=−1\lambda = -1, (2) implies (3), means that for x∈A′′x \in A'', we have f(x)≥f1(a)f(x) \geq f_1(a), \emph{i.e.} f1(a)≤α′′f_1(a) \leq \alpha''. The proposition is therefore proved in this case.

II. General case.

Let F\mathfrak{F} be the set of pairs (W,g)(W, g) where WW is a vector subspace of EE containing VV and gg is a positive linear form on WW which is an extension of ff. We order F\mathfrak{F} putting (W,g)≤(W′,g′)(W, g) \leq (W', g') if W⊂W′W \subset W' and if g′g' is an extension of gg. Clearly F\mathfrak{F} is inductive and by th. 2 of S, III, § 2.4, there is a maximal element (W0,g0)(W_0, g_0). Suppose W0≠EW_0 \neq E. Then there exists a vector b∉W0b \notin W_0, and, if W1=W0+RbW_1 = W_0 + \mathbf{R}b, the special case above shows that there exists a positive linear form on W1W_1 which is an extension of g0g_0; this contradicts the hypothesis that (W0,g0)(W_0, g_0) is maximal. Thus W0=EW_0 = E, and the first part of the proposition is proved. When a∈Va \in V, the second assertion is obviously true with α′=α′′=f(a)\alpha' = \alpha'' = f(a); if, on the contrary, a∉Va \notin V and one puts V1=V+RaV_1 = V + \mathbf{R}a, the second assertion follows from the special case I of the proof.

Bourbaki puts the proof directly after the statement, so what is above is the statement and its proof together.

Cited by

Nothing.

Cites

Transcribed from the printed volume. Not translated.