Analysis Problem of the Day 2

Today’s problem is a nice result about smooth non-analytic functions known as Borel’s Theorem.

Problem: Let a_n \in \mathbb{R} be any sequence of real numbers. Show that there exists a smooth function f \in C^\infty(\mathbb{R}) such that f^{(n)}(0)=a_n.


Solution: The naive approach to this problem (which ignores any convergence issues) would be to set f(x):= \sum_{n=0}^\infty a_n x^n. However, as we are aware from the theory of analytic functions, this power series has a positive radius of convergence only if \limsup_{n \to \infty} |a_n|^{\frac{1}{n}} < \infty. The key point in this problem is that we are working over the category of smooth functions, not real analytic ones. We will modify our naive approach by doing a classic trick – introducing a smooth test function \chi \in C^\infty(\mathbb{R}) that equals 1 in a small neighborhood of 0 and vanishes for, say, |x|>1. Then, if we define

    \[f(x):= \sum_{n=0}^\infty a_n x^n \chi(nx),\]

one can easily check the following:

  • f(x) is clearly defined at x=0, and for x \not =0, the sum is finite since \chi(nx)=0 for large enough n. Thus, f(x) is defined pointwise everywhere.
  • In fact, on any compact interval K avoiding the origin, the sum is finite and therefore (since \chi is smooth) defines a smooth function f. This means that f^{(k)} \in C^\infty(\mathbb{R} \setminus \{0\}) for all k \geq 0, and in fact, one may differentiate the power series for f term-by-term.
  • One may check that

        \[\lim_{x \to 0} f^{(k)}(x) = \lim_{x \to 0} \sum_{n=0}^\infty a_n \partial_x^k(x^n \chi(nx)) = \sum_{n=0}^\infty a_n \partial_x^k(x^n \chi(nx))|_{x=0} = a_n,\]

    where we switch the order of the limit and infinite sum by an application of the dominated convergence theorem.
  • Thus, if one defines f^{(k)}(0)=a_n, this extends f to a smooth function satisfying the required properties.

Remark: Notice that f in general cannot be analytic at 0, since if \limsup_{n \to \infty} |a_n|^{\frac{1}{n}} = \infty the Taylor series would have a zero radius of convergence.

Leave a Reply

Your email address will not be published. Required fields are marked *