13 Convergence and completeness, part 1

Convergence in norm

  • If \(\mathbf{a}\) and \(\mathbf{b}\) are two finite-dimensional complex (or real) vectors, the distance between them is precisely \(\| \mathbf{a}-\mathbf{b}\|\).

  • This suggests that a good notion of convergence of a sequence of a sequence of vectors is:

Definition (Norm convergence of vectors)

Let \((\mathbf{a}_n)_{n=1}^\infty\) be a sequence of vectors and \(\mathbf{a}\) a fixed vector in \(\mathbb{C}^k\). Then the sequence converges to \(\mathbf{a}\) in norm, written \[ \lim_{n\to \infty} \mathbf{a}_n = \mathbf{a}, \] if \[ \lim_{n\to \infty} \|\mathbf{a}_n - \mathbf{a}\|=0. \]

  • Now, our space \(PC(a,b)\) of functions also has a norm. This means that we have an analog of the definition above for functions:

Definition (Norm convergence of functions)

Let \((f_n)_{n=1}^\infty\) be a sequence of functions and \(f\) a fixed function in \(PC(a,b)\). Then the sequence converges to \(f\) in norm, written \[ \lim_{n\to \infty} f_n = f, \] if \[ \lim_{n\to \infty} \|f_n - f\|=0. \]

  • So, we have \(f_n \to f\) in norm if and only if \[ \int_a^b |f_n(x) - f(x) | ^2 \, dx \to 0 \] as \(n\to \infty\).

Exercise 1: Norm convergence versus pointwise convergence

Pointwise and norm convergence are completely different modes of convergence.

  1. Given the sequence of functions \[ f_n(x) = \begin{cases} 1 & : 0 \leq x \leq 1/n, \\ 0 & : 1/n < x \leq 0, \end{cases} \] on \([0,1]\), prove that \(f_n\to 0\) in norm, but that the \(f_n\)’s do not converge to \(0\) pointwise.

  2. Conversely, given the sequence of functions \[ g_n(x) = \begin{cases} n & : 0 < x < 1/n, \\ 0 & : x=0, \, \text{or} \, 1/n \leq x \leq 1, \end{cases} \] on \([0,1]\), prove that \(g_n\to 0\) pointwise, but that the \(g_n\)’s do not converge to \(0\) in norm.

Norm convergence versus uniform convergence

  • However! We do have the following simple result.

  • To state it, recall that a sequence of functions \(f_n\to f\) uniformly (on \([a,b]\)) if there is a sequence of real constants \((M_n)_{n=1}^\infty\) such that \[ |f_n(x) - f(x)| \leq M_n \] for all \(x\in [a,b]\) and \(M_n \to 0\).

Theorem 3.2

If \(f_n\to f\) uniformly on \([a,b]\), then \(f_n\to f\) in norm.

A problem with \(PC(a,b)\): it’s not complete

  • The main technique in calculus and analysis is to construct objects by “passing to the limit.” This is how derivatives are constructed, for example, as well as integrals.

  • This only works if the limit actually exists!

  • In this respect, the space of functions \(PC(a,b)\) is not ideal. There are sequences of functions in this space that “look like” they should converge, but don’t.

  • This is captured by saying that \(PC(a,b)\) is not complete.

  • To explain what this means, we need the following:

Definition

A sequence of functions \((f_n)_{n=1}^\infty\) is called a Cauchy sequence if \[ \|f_m - f_n\| \to 0 \] as \(m,n\to \infty\).

  • The same definition applies to other vector spaces with norms, like \(\mathbb{C}^k\).

  • Intuitively, this means that the functions in the sequence get closer and closer together the further one goes out in the sequence.

Definition

A vector space with a norm (like \(\mathbb{C}^k\), or \(PC(a,b)\)) is called complete if all Cauchy sequences converge (in norm).

Exercise 3: Proving \(PC(a,b)\) is not complete

By examining the sequence of functions \[ f_n(x) = \begin{cases} 0 & : 0 \leq x \leq 1/n, \\ x^{-1/4} & : 1/n < x \leq 1, \end{cases} \] on \([0,1]\), prove that \(PC(0,1)\) is not complete.