Consider a Fourier Series

$$S(t) = \sum_{k=0}^{\infty}c_{k}e^{ikt}$$

where $c_{k}$ are complex coefficients such that the sum $\sum |c_{k}|$ is finite. I am also given that $\lim_{k\to\infty}k^{m}c_{k}=0$ for some fixed $m>0$.

Question: What can we say about the differentiability of $S$?

What I tried: If I can prove that $\sum k|c_{k}|<\infty$, then the sequence of derivatives of the partial sums $$f_{n}'(t)=\sum_{k=0}^{n}ikc_{k}e^{ikt}$$ must converge uniformly to a continuous limit, then $S$ would be differentiable. However, I am not sure how to apply the fact that $\lim_{k\to\infty}k^{m}c_{k}=0$ for some fixed $m>0$ - certainly $\sum k|c_{k}|<\infty$ implies the terms should go to 0 but how do I know the converse is true?

  • $\begingroup$ If $\sum_k |c_k k^m| < \infty$ then integrate $m$-times $\sum_k c_k (ik)^m e^{ikt}$ you'll recover $S(t)$ $\endgroup$
    – reuns
    Aug 4 '19 at 3:58
  • $\begingroup$ How do you show $\sum_{k}|c_{k}k^{m}|<\infty$ given that $\lim_{m\to\infty}k^{m}c_{k}=0$ and $\sum_{k}|c_{k}|<0$? $\endgroup$ Aug 4 '19 at 4:52
  • 1
    $\begingroup$ I don't. $c_k = O(k^{-m})$ implies $\sum_k |c_k k^{m-2}| < \infty$ so $S(t)$ is $C^{m-2}$, and that $\sum_k c_k (ik)^{m-1} e^{ikt}$ converges in $L^2$ so that $S(t)$ is the $m-1$-th primitive of a $L^2$ function. The $m$-th derivative is only well-defined in the sense of distributions. $\endgroup$
    – reuns
    Aug 4 '19 at 17:27
  • $\begingroup$ Can we assume that $m$ is an integer? Otherwise, we can set $c_k=1/k$ if $k$ is a power of $2$, and $c=0$ otherwise. This does not satisfy $\sum_k k|c_k|\lt\infty$. $\endgroup$
    – Thijs
    Aug 7 '19 at 10:07
  • 1
    $\begingroup$ I think it might be time to ask the professor about whether $m$ is supposed to be an integer -- no use spending a ton of time on a problem that's not the one intended. $\endgroup$ Aug 7 '19 at 10:21

Before I type anything else, I would like to point out that the assumption (8) below is insufficient to establish that

$\displaystyle \sum_0^\infty k \vert c_k \vert < \infty, \tag 0$

even when the condition (9) binds.  A counterexample is provided by the series

$R(t) = \displaystyle \sum_1^\infty \dfrac{1}{k^2} e^{ikt}; \tag{0.1}$

it is well-known that

$\displaystyle \sum_1^\infty \dfrac{1}{k^2} = \dfrac{\pi^2}{6}, \tag{0.2}$

see Showing $\sum _{k=1} 1/k^2 = \pi^2/6$ ; however, with

$c_k = \dfrac{1}{k^2}, \tag{0.3}$

we find

$\displaystyle \sum_1^\infty kc_k = \sum_1^\infty k \dfrac{1}{k^2} = \sum_1^\infty \dfrac{1}{k} = \infty; \tag{0.4}$

nevertheless, if adopt the stronger hypothesis that

$\displaystyle \sum_1^\infty k^m \vert c_k \vert < \infty, \tag{0.5}$

we will discover its suffiency, as is shown below.

I assume

$0 < m \in \Bbb Z, \tag 0$

that is, $m$ is a positive integer.

Consider the sequence of partial sums

$S_n(t) = \displaystyle \sum_{k = 0}^n c_ke^{ikt} \tag 1$

of the series

$S(t) = \displaystyle \sum_{k = 0}^\infty c_ke^{ikt}; \tag 2$

it is easy to see that $S_n(t)$ is a $C^\infty$ function for every $n \in \Bbb N$; indeed, the $S_n(t)$ are analytic, each being the sum of a finite number of analytic functions $c_ke^{ikt}$; furthermore for $n > p$ we have

$S_n(t) - S_p(t) = \displaystyle \sum_{p + 1}^n c_k e^{ikt}, \tag 3$


$\vert S_n(t) - S_p(t) \vert = \vert \displaystyle \sum_{p + 1}^n c_k e^{ikt} \vert \le \sum_{p + 1}^n \vert c_k \vert, \tag 4$


$\vert e^{ikt} \vert = 1; \tag 5$

now taking $p$ and $n$ sufficiently large we may affirm that

$\displaystyle \sum_{p + 1}^n \vert c_k \vert < \epsilon \tag 6$

for any real

$\epsilon > 0; \tag 7$

this assertion of course follows easily from the hypothesis

$\displaystyle \sum_0^\infty \vert c_k \vert < \infty. \tag 8$

In light of these remarks, we conclude that the sequence of functions $S_n(t)$ converges uniformly in $t$; thus the limit function $S(t)$ is indeed continuous.

Note that we have not yet called upon the hypothesis that

$\displaystyle \lim_{k \to \infty} k^m c_k = 0.  \tag 9$

Now observe that the $S_n(t)$ (1), being finite sums, are each in fact differentiable functions of $t$; indeed,

$S_n'(t) = \displaystyle \sum_{k = 0}^n ikc_ke^{ikt}; \tag{10}$


$\vert S_n'(t) - S_p'(t) \vert = \vert \displaystyle \sum_{p + 1}^n ikc_k e^{ikt} \vert \le \sum_{p + 1}^n k\vert c_k \vert < \epsilon \tag{11}$

for $n$, $p$ sufficiently large in light of our added assumption (0.5) with $m = 1$, and thus the sequence $S_n'(t)$ is Cauchy and hence it also is uniformly convergent. since $\epsilon$ is independent of $t$; these facts in concert are sufficient for the existence of a function $S'(t)$ such that

$ S'(t) = \displaystyle \lim_{n \to \infty} S_n'(t) = ( \lim_{n \to \infty} S_n(t))' = (S(t))', \tag{12}$

in accord with the standard theorem on convergence of sequences of derivatives.

The process described in the above may be continued for larger values of $m$, the result being similar to that attained so far, provided of course that (0.5) binds for the chosen value of $m$. Indeed, we may write

$S''(t) = \displaystyle \sum_1^\infty -k^2c_k e^{ikt}, \tag{13}$

and so forth. Higher derivatives of $S(t)$ may be expressed in an analogous manner, assuming (0.5) holds for appropriate values of $m$.

  • 1
    $\begingroup$ Appreciate the effort in this answer - One thing; why are we proving $|S'_{n}(t) - S_{p}'(t)|<\epsilon$ ? It seems this is only proving the sequence is Cauchy, not convergent. Are you fixing $p$ for now, then taking the limit as $p\to\infty$? $\endgroup$ Aug 12 '19 at 11:27

For sufficiently large $ k $ ,$ |c_k |\le \frac {2}{|k^m|} $

Pick $ m=2$

  • $\begingroup$ We are not given a choice of $m$ so I think we need to account for all possible values. Also how did you get the inequality ? $\endgroup$ Aug 7 '19 at 12:25

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.