Convergence Of Improper Integrals Does Not Imply The Integrands Go To Zero

This entry devotes a similar question that raises during a course of series. We all know that for a convergent series of (positive) real number
\displaystyle \sum_{n=1}^\infty a_n
it is necessary to have
\displaystyle\mathop {\lim }\limits_{n \to \infty } {a_n} = 0.
This is the so-called n-th term test. A natural extension is the following question
Question. Suppose f(x) is positive on [0,\infty) and
\displaystyle\int_0^{ + \infty } {f(x)dx}
exists. Must f(x) tend to zero as x \to +\infty?
I will show that this is indeed not the case. Take
\displaystyle f(x)=\frac{1}{e^x}+g(x), \quad x \geqslant 0
where
\displaystyle g(x)=\begin{cases}0, & x \in \left[0,\frac{7}{4}\right], \\ 1, & x=2,3,..., \\ n^2(x-n)+1,& x \in [n-n^{-2},n], n=2,3,..., \\ -n^2(x-n)+1,& x\in [n,n+n^{-2}],n=2,3,..., \\ 0,& \text{otherwise}.\end{cases}
The above solution tell us the fact that the convergence of improper integral \int_0^{ + \infty } {f(x)dx}  is not enough to guarantee \mathop {\lim }_{x \to \infty } f(x) = 0.
It turns out to propose the question under what condition on ff(x) tend to zero as x \to +\infty? There are several answers in the literature, for example
  • If f is positive, differentiable and |f'(x)| is uniformly bounded, say by M, then by the mean value theorem, we can prove \mathop {\lim }_{x \to \infty } f(x) = 0.
  • If f is uniformly continuous on [0,\infty) then the same conclusion still holds by the Cauchy theorem.
  • If f is monotone decreasing, we obtain a stronger result, i.e. \mathop {\lim }_{x \to \infty } xf(x) = 0.

No comments:

Post a Comment