Where Volume I's comfort ran out
All of Volume I took place in a finite-dimensional vector space: pick a basis, and every vector is a finite list of coordinates. Length came from a norm, angle from an inner product, and life was easy because the dimension was a single finite number. But the spaces that matter in analysis — spaces of functions, of infinite sequences, of signals — have no finite basis at all.
Consider the set of square-summable sequences: x = (x_1, x_2, x_3, ...) with sum |x_n|^2 finite. This is a genuine vector space — you can add sequences and scale them — but it is infinite-dimensional: the standard unit vectors e_1, e_2, e_3, ... are linearly independent and there are infinitely many of them. No finite basis could span it. We need new tools that survive this jump.
Completeness: the property that replaces dimension
In R^n there is a fact so basic you never named it: every Cauchy sequence converges to a limit *inside the space*. That is completeness, and in finite dimensions it comes for free. In infinite dimensions it can fail — and when it fails, calculus breaks, because limits of approximations can leak out of the space entirely.
Incompleteness in the wild — continuous functions on [-1, 1]
with the L^2 "length" ||f|| = sqrt( integral |f|^2 ).
Approximate a step function by continuous ramps f_n:
f_n(x) = -1 for x <= -1/n
= n*x for -1/n < x < 1/n (a steep ramp)
= +1 for x >= 1/n
Each f_n is CONTINUOUS.
The sequence (f_n) is Cauchy in the L^2 norm.
But its limit is the sign function sgn(x), which JUMPS at 0
-> the limit is NOT continuous.
So (continuous functions, L^2 norm) is INCOMPLETE:
the Cauchy sequence converges to something outside the space.Two new homes: Banach and Hilbert spaces
A complete normed vector space is a Banach space: it has lengths (a norm) and no holes. If in addition the norm comes from an inner product — so you also have angles, orthogonality, and the parallelogram law — the complete space is a Hilbert space. Every Hilbert space is a Banach space; the converse fails, and that extra inner-product geometry is exactly what makes Hilbert spaces feel like an infinite-dimensional R^n.
Family tree of the spaces in this volume
vector space
| add a length
normed space ||x||
| add: no holes (every Cauchy seq converges)
BANACH space complete + norm
| add: the norm comes from <x, y>
HILBERT space complete + inner product
Canonical examples:
ell^2 = square-summable sequences -> HILBERT ||x||^2 = sum |x_n|^2
L^2 = square-integrable functions -> HILBERT ||f||^2 = integral |f|^2
ell^p, L^p (p != 2) -> BANACH but NOT Hilbert
C[0,1] with sup norm -> BANACH but NOT Hilbert
Test: does the parallelogram law hold?
||x+y||^2 + ||x-y||^2 = 2||x||^2 + 2||y||^2
YES -> norm comes from an inner product (Hilbert)
NO -> Banach only.From here on, ell^2 and L^2 are our running examples: both are Hilbert spaces, infinite-dimensional, and complete. The plan for this track is to carry the geometry of Vol I — orthonormal bases, projections, eigenvalues — across the infinite-dimensional divide, keeping what survives and flagging what shatters.