a Hardy space
/ HAR-dee /
A Hardy space is a carefully chosen collection of holomorphic functions on the disk (or the upper half-plane) that stay 'bounded in size' as you move out toward the boundary, in an averaged sense. The idea answers a natural question: which holomorphic functions inside the disk actually have sensible boundary values on the circle, and how are the function inside and its boundary trace linked? Hardy spaces are the home where the inside of the disk and its boundary circle meet on equal footing.
Concretely, for a number p between 1 and infinity, the Hardy space H^p is the set of functions f holomorphic on the unit disk for which the integral means stay bounded: the supremum over r < 1 of (the average of |f(r e^(i theta))|^p over theta) is finite. The remarkable facts that make this useful are theorems, not definitions. First, every f in H^p has a radial boundary limit f(e^(i theta)) for almost every theta, so each interior function quietly carries a boundary function with it. Second, you can travel back the other way: the interior function is recovered from its boundary values by the Cauchy or Poisson integral. Third, a function on the circle is the boundary trace of an H^2 function exactly when its Fourier series has no negative-frequency terms — H^2 is the 'analytic half' of all square-integrable functions on the circle, and the projection onto it is built from the Hilbert transform.
Hardy spaces are where complex analysis fuses with harmonic analysis and operator theory, and they pay off across pure and applied mathematics: the factorization of an H^p function into a Blaschke product times an outer factor underlies signal-processing notions of minimum-phase systems and all-pass filters; H^2 of the half-plane is exactly the space of transfer functions of stable causal systems, tying the theory straight back to control engineering; and the boundary-value viewpoint makes Hardy spaces the natural setting for the Hilbert transform and the Riesz projection. An honest note on scope: the H^p spaces behave beautifully for 1 <= p <= infinity but the borderline cases (p = 1 and p = infinity) are genuinely more delicate than the Hilbert-space case p = 2, which is the one with the cleanest geometry.
A bounded holomorphic function on the disk, such as f(z) = 1 / (2 - z), lies in every H^p (it is even in H^infinity since |f| stays below 1 on the disk). A single Blaschke factor b(z) = (z - a) / (1 - a-bar z) with |a| < 1 lies in H^infinity too; its boundary trace has modulus exactly 1 on the circle, the prototype of an all-pass filter.
Bounded holomorphic functions sit in every H^p; a Blaschke factor is the all-pass prototype.
Membership in H^p controls only the AVERAGE boundary growth, not pointwise boundedness; an H^2 function can be unbounded, yet it still has finite boundary limits almost everywhere — a far stronger conclusion than the mild-looking hypothesis suggests.