Basic Analysis – Gently Done Topological Vector Spaces
Basic Analysis – Gently Done Topological Vector Spaces
Basic Analysis – Gently Done Topological Vector Spaces
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
50 <strong>Basic</strong> <strong>Analysis</strong><br />
and so, multiplying by t and using the linearity of f and the positive homogeneity<br />
of p, we get<br />
f(x)−tµ ≤ p(x−tx 1 ) for x ∈ M and t > 0.<br />
Thus,<br />
f 1 (x+sx 1 ) ≤ p(x+sx 1 ) for x ∈ M and any s < 0.<br />
From (2), replacing y by t −1 y, with t > 0, we get<br />
Thus, for any t > 0,<br />
i.e.,<br />
f(t −1 y)+µ ≤ p(t −1 y +x 1 ), y ∈ M.<br />
f(y)+tµ ≤ p(y +tx 1 ), y ∈ M,<br />
f 1 (y +tx 1 ) ≤ p(y +tx 1 ), for any y ∈ M, and t > 0.<br />
This inequality also holds for t = 0, by hypothesis on f. Combining these inequalities<br />
shows that<br />
f 1 ≤ p on M 1 .<br />
We have shown that f can be extended from M to a subspace with one extra<br />
dimension, whilst retaining the bound in terms of p. To obtain an extension to<br />
the whole of X, we shall use Zorn’s lemma. (An idea would be to apply the above<br />
procedure again and again, but one then has the problem of showing that this<br />
would eventually exhaust the whole of X. Use of Zorn’s lemma is a way around<br />
this difficulty.) Let E denote the family of ordered pairs (N,h), where N is a<br />
linear subspace of X containing M and h is a real linear functional on N such<br />
that h ↾ M = f and h(x) ≤ p(x) for all x ∈ N. The pair (M,f) (and also (M1 ,f1 )<br />
constructed above) is an element of E, so that E is not empty. Moreover, E is<br />
partially ordered by extension, that is, we define (N,h) � (N ′ ,h ′ ) if N ⊆ N ′ and<br />
h ′ ↾ N = h. Let C be any totally ordered subset of E and set N ′ = �<br />
(N,h)∈CN. Let x ∈ N ′ . Suppose that x ∈ N1∩N 2 , where (N1 ,h1 ) and (N2 ,h2 ) are members<br />
of C. Then either (N1 ,h1 ) � (N2 ,h2 ) or (N2 ,h2 ) � (N1 ,h1 ). In any event,<br />
h1 (x) = h2 (x). Hence we may define h ′ on N ′ by the assignment h ′ (x) = h(x) if<br />
x ∈ N with (N,h) ∈ C. Then h ′ is a linear functional on N ′ such that h ′ ↾ M = f<br />
and h ′ (x) ≤ p(x) for all x ∈ N ′ . This means that (N ′ ,h ′ ) is an upper bound for<br />
C in E. By Zorn’s lemma, E contains a maximal element (N,Λ), say. If N �= X,<br />
then we could construct an extension of (N,Λ), as earlier, which would contradict<br />
maximality. We conclude that N = X and that Λ is a linear functional Λ : X → R<br />
such that Λ ↾ M = f, and Λ(x) ≤ p(x) for all x ∈ X.<br />
Replacingxby−xgivesΛ(−x) ≤ p(−x), i.e.,−Λ(x) ≤ p(−x)andso−p(−x) ≤<br />
Λ(x) for x ∈ X.<br />
Department of Mathematics King’s College, London