The Baire category theorem says that a complete metric space cannot be a countable union of nowhere dense sets, equivalently that every countable intersection of open dense sets is dense. To prove it, let be open dense subsets and let be nonempty and open. Choose nested closed balls
starting inside . The centres are a Cauchy sequence; completeness gives a limit lying in every ball, hence in . Thus the intersection is dense.
Choose a countable sequence with . Since for , every with lies in some . For integers , set
Pointwise convergence and Fatou lemma for series show that is closed in . It has empty interior: inside any ball, add a sufficiently long finite block whose norm is tiny but whose norm is larger than . Therefore every is nowhere dense. Since is a Banach space, Baire gives
An explicit element of the difference is
Its th powers form a convergent logarithmic series, whereas for every the powers contain with exponent below one and their series diverges. Hence but for every .
For integers , let contain those for which there is some such that
whenever . Compactness of shows that is closed in the uniform norm: for a uniformly convergent sequence, choose convergent subsequences of the corresponding witness points and pass to the limit.
Each has empty interior. Given and a uniform-error tolerance, first approximate by a polygonal function, then add a sufficiently small-amplitude, sufficiently high-frequency sawtooth whose slopes dominate both and every slope of the polygonal approximation. At every point, one can move a distance below along one adjacent linear segment and obtain a difference quotient of magnitude greater than . The perturbed function lies in the prescribed ball but outside .
If is differentiable at some point, its nearby difference quotients are bounded, so for some . The functions differentiable somewhere therefore form a meagre set. By the Baire category theorem, their complement is nonempty; indeed it is dense. Thus
For , define by . The map is linear, and
shows , so is continuous. The granted equality is the canonical embedding into the bidual.
Suppose, contrary to the claim, that exactly the same linear functionals are continuous for and . Their common continuous dual has two operator norms, say and . Both dual spaces are Banach, even if the second normed space is not complete. The identity map
has a closed graph: convergence in either operator norm implies pointwise convergence on , so two limits must agree. The closed graph theorem makes the two dual norms equivalent.
Applying the isometric bidual formula for each primal norm then gives constants such that
This says the two norms are equivalent norms, contrary to the hypothesis. Their continuous duals must therefore differ as sets, so a linear functional belonging to one and not the other is continuous for exactly one of the two norms.

Articles by others on the same topic (0)

There are currently no matching articles.