Banach Spaces, Using Stone Weierstrass

Generating a Hilbert Subspace

Let b1 b2 b3 etc be an orthonormal system that is not maximal in S.  Extend this to a maximal orthonormal system b1 c1 b2 c2 b3 c3 b4 c4 etc.  If the original system was almost maximal, we may only need a few additional elements to cover S.  Otherwise, the extra elements are countable, and new and old elements interleave forever, as shown above.

Consider the hyperspan of the original orthonormal system.  This is the set of elements whose coefficients on ci are all 0.  Since all the elements of S are faithfully represented, this characterization of the subspace T is well defined.  In fact T is a linear subspace, with metric and dot product inherited from S.  Is T a complete metric space?  In the previous section we started with a cauchy sequence un in S, and found it's limit, by finding a cauchy sequence, and a limit, for each coordinate.  If un lies entirely in T, the coefficient on each ci is 0, the limit per coordinate is 0, and the limit of the cauchy sequence lies in T.  Thus T is a complete metric space.  This makes T a hilbert space, and a subspace of S.

Completing a Finite Span

Let b1 b2 b3 etc be an orthonormal system that may or may not be maximal in S.  Let M be the traditional linear combinations of these basis elements, where each linear combination is finite.  This is a subspace of S, with metric and dot product inherited from S.  We will show the completion of M is T, the hilbert subspace of S described above.

Since M lies in T, and T is complete, the completion of M lies in T.  We only need show each element of T is the limit of a convergent sequence drawn from M.

Let v be represented as an infinite sum aibi.  The approximations to v, i.e. the sum of the first n terms, all lie in M, and they approach v monotonically.  Thus all of T lies in the completion of M, and the completion of M = T.

Note that T is also the closure of M in S, hence T is a closed subspace of S.

Ej

Review the definition of generalized euclidean space, denoted Ej.  If Ej employs a countable basis, this basis establishes a countable orthonormal system, and Ej is the finite span thereof.  The completion of Ej is L2, the square summable sequences that build a separable hilbert space.

Applying Stone Weierstrass

Let D be a closed bounded region in Rn with a fixed volume v, and let M be the ring of continuous functions from D into R.  This is a metric space in two different ways.  The first norm is based on uniform convergence.  Two functions are 1 unit apart if they never differ by more than 1, and for some x in D, they differ by exactly 1.  Under this norm, M is a complete metric space.

The second norm is sqrt(∫ f2/v).  If two functions are ε apart under the first norm, they are at most ε apart under the second.  Mapping M onto itself, from the first norm to the second, is a bounded linear operator, and is continuous.  Going backwards, from the second norm to the first, is not continuous.  A function can have ∫ f2 = 1, with a tall narrow spike.  The spike can be as tall as you like.  Thus the uniform norm is arbitrarily large.  The operator is not bounded, and not continuous.  The first norm defines more open sets than the second.

The second norm admits a dot product, namely the integral of fg/v.  Sure enough, f.f = |f|2.

Let S be the completion of M with respect to the second norm, hence S is a hilbert space.  This was described in detail when D was the unit interval [0,1], and it easily generalizes to well behaved domains in Rn.

Assume b1 b2 b3 etc is an orthonormal system within M.  Since M is a ring containing the constant functions, the finite span of this system, call it W, lies in M.  Assume W includes the constant functions, and the functions of W separate the points of D.  For every pair of points x and y in D, some function in W maps x to one value and y to another.  This satisfies the conditions of stone weierstrass, and the closure of W, under the uniform norm, is all of M.

A convergent sequence in the first metric is automatically convergent in the second.  Therefore the closure of W, with respect to the second metric, includes all of M.  Now the closure of W is closed, and contains M, and the closure of M is S, hence the closure of W is S.  The complete metric space, hyperspanned by our orthonormal system, is all of S.  Therefore the orthonormal system is maximal, and acts as a basis.  Anything in M, or in the completion of M, has a unique representation as a square summable array of coefficients, relative to our basis, and these coefficients can be extracted using the dot product.

Every Continuous Function is Differentiable

Obviously this is not true; see if you can spot the error.

Let the domain D be the closed interval [-1,1], and let M be the continuous functions from D into R.  Let b be the countable system of legendre polynomials on [-1,1].  These are orthogonal, and together they span all the polynomials of x.  Let W be the finite span of the legendre polynomials, i.e. the polynomials in x.  Since W includes the constant functions, and separates points, we can apply the above.  Every function in S has a unique representation relative to the legendre polynomials.

Let f = abs(x), which is continuous.  Write it as a series in legendre polynomials, and convert to a power series in x.  Now abs(x) has a power series, and is real analytic, hence it is differentiable at 0.

Well there are two problems with this proof.  First, converting a legendre series to a power series may not be possible.  As the terms combine, do the multiples of x settle down to a limit?  How about x2, and x3, etc?

Even if a power series emerges, it may not converge anywhere outside of 0.  The continuous functions are the elements of S.  Each new legendre polynomial yields a continuous polynomial that is closer, on average, to abs(x), but it's only an average.  Far out in the series, tall thin deviations could still appear.  The root mean square deviation from abs(x) is still tiny, but in places, the approximating polynomial is way off the mark.  Pointwise convergence is not guaranteed.  This will become clearer in the next example.

Fourier Series

Let D be the interval from 0 to 2π, or more generally, the hypercube 2π units on a side.  Let M be the ring of continuous functions from D into R, and let S be the completion of this ring, which includes piecewise continuous periodic functions, such as the square wave.

Stone Weierstrass has already been applied in this situation.  Thus every function in S has a fourier series that "converges" to that function.  But convergence may not be uniform, and it may not apply for certain values of x.

Take the square wave, for example.  It equals 1 from 0 to π, and -1 from π to 2π.  Multiply by sines and cosines and find the following fourier series.

f(x) = 4/π × ∑{n odd} sin(nx)/n

After much analysis, this series, with the terms in this order, converges uniformly to the square wave.  The function is faithfully reproduced for every x, and the approach is uniform across the entire domain.

Earlier we said the order of the terms didn't matter.  This is true, and it is false.  Rearrange the terms, and the approximations still approach f, on average, but spikes can occur.  To illustrate, let x = π/2 and substitute in the above series.  The result is an alternating sum of the odd reciprocals: 1 - 1/3 + 1/5 - 1/7 …  This converges to the correct value on the square wave, but convergence is not absolute.  We can rearrange the terms and make this sum to anything.  So, if the terms are permuted, the series still approaches the square wave for most values of x, but in some cases, such as x = π/2, the series could reach 129, instead of 1.  The approximations come closer and closer to the square wave, on average, but they all have a spike at π/2, that rises to 129, more or less.  The spike gets thinner with each approximation, but it's always there.

The combination of stone weierstrass and hilbert orthogonalization is just the beginning - a proof of existence.  Often more calculus is required, to show the resulting series, with its terms in a prescribed order, is well behaved across the entire domain.