28
$\begingroup$

Given the infinite series

$$\begin{aligned}\sum_{n = 1}^{\infty}\end{aligned} \frac{\sin\left(n^a\right)}{n^b}$$

with $a,\,b \in \mathbb{R}$, study when it converges and when it diverges.


Easy cases

  • $\forall\,a \in \mathbb{R}$ we have $\left|\frac{\sin\left(n^a\right)}{n^b}\right| \le \frac{1}{n^b}$ so the series $\color{green}{\text{converges}}$ for $b > 1$.

  • If $a \le 0$ we have $\frac{\sin\left(n^a\right)}{n^b} \le \frac{1}{n^{b-a}}$ so the series $\color{blue}{\text{diverges}}$ for $b \le a + 1$ and $\color{green}{\text{converges}}$ for $b > a + 1$.

  • If $a > 0 \, \land \, b \le 0$ we have $\not\exists \begin{aligned}\lim_{n \to \infty} \frac{\sin\left(n^a\right)}{n^b} \end{aligned}$ so the series $\color{blue}{\text{diverges}}$.

  • If $a = 1\, \land \, b > 0$ the series $\color{green}{\text{converges}}$ by Abel-Dirichlet's test.

Hard cases

  • If $0 < a < 1\, \land \, 0 < b \le 1-a$ the series $\color{blue}{\text{diverges}}$ by proof of i707107.

    • If $0 < a < \frac{1}{2}\, \land \, a < b \le 1-a$ the series $\color{blue}{\text{diverges}}$ by proof of RRL.
  • If $0 < a < 1\, \land \, b > 1-a$ the series $\color{green}{\text{converges}}$ by proof of i707107.

    • If $a > 0\, \land \, b > \max(a,\,1-a)$ the series $\color{green}{\text{converges}}$ by proof of RRL.
  • If $k \in \mathbb{Z}_{\ge 2}, $ $k-1 < a < k\, \land \, b > 1 - \frac{k-a}{2^k-2}$ the series $\color{green}{\text{converges}}$ by proof of i707107.

  • If $a > 0 \, \land \, b = 1$ the series $\color{green}{\text{converges}}$ by proof David Speyer (+ i707107 in the comments).

  • If $a = 2 \, \land \, 0 < b \le \frac{1}{2}$ the series $\color{blue}{\text{diverges}}$ (Theorem 2.30 by Hardy&Littlewood).

enter image description here


$\color{red}{\textbf{Open cases}}$

  • $a = \frac{3}{2} \land b = \frac{1}{4}$: enter image description here

  • $a = \frac{3}{2} \land b = \frac{1}{2}$: enter image description here

  • $a = \frac{3}{2} \land b = \frac{3}{4}$: enter image description here

  • $a = 2 \land b = \frac{3}{4}$: enter image description here

  • $a = \frac{5}{2} \land b = \frac{1}{2}$: enter image description here

$\endgroup$
  • 1
    $\begingroup$ Actually, consider $a+1<b\le0$. That is, the series will converge for $a<b-1$ for any $a,b\in\mathbb R$. $\endgroup$ – Simply Beautiful Art Jul 28 '17 at 11:00
  • 1
    $\begingroup$ For $a<0$, you can use the limit comparison with the sequence $\frac{1}{n^{b-a}}$ to conclude your sequence is convergent for $(b-a)>1$ and divergent for $(b-a)\leq 1$. $\endgroup$ – Peter Košinár Jul 28 '17 at 22:22
  • 2
    $\begingroup$ For $b = 1$ and all $a > 0$ see here. That was not easy and the region where $0 < b < 1$ and $a > 1$ probably no easier to analyze. $\endgroup$ – RRL Jul 30 '17 at 0:33
  • 1
    $\begingroup$ @TeM One thing to remark is that the solution by David Speyer is for all "integers" $a>0$ and $b=1$. In such case we indeed have convergence for $b>1-\epsilon(a)$ for some function $\epsilon(a)$ satisfying $\epsilon(a)\rightarrow 0$ as $a\rightarrow\infty$. $\endgroup$ – Sungjin Kim Jul 7 '19 at 13:51
  • 1
    $\begingroup$ @TeM although David Speyer’s solution is for integer $a>0$, $b=1$, the series is convergent for indeed ”ALL” $a>0$, $b=1$. The method of mine uses Van Der Corput’s method. The application or Van Der Corput for integer $a$ is Weyl’s inequality in David Speyer’s answer. The method can be applied to real $a>0$ too. More about the method are outlined in chapter 4 and 5 of Titchmarsh’s book. $\endgroup$ – Sungjin Kim Jul 8 '19 at 16:25
8
+200
$\begingroup$

The series diverges if $0<a<1$, $b>0$ and $a+b\leq 1$

This idea is already discussed in the link provided at $a=b=1/2$ case. Let $k$ be a positive integer. If $$ 2\pi k + \pi/6\leq n^a\leq 2\pi k + 5\pi/6, $$ then $\sin (n^a)\geq 1/2$.

The number of $n$'s in this interval is $\asymp k^{(1/a) -1}$. In the interval, we have $n^b\asymp k^{b/a}$. Thus, the contribution of $n$'s to the sum in this interval is $\asymp k^{(1/a)-(b/a)-1}$.

If $(1/a)-(b/a)-1 \geq 0$, then we have divergence. Note that the partial sum of the series fails to be Cauchy sequence.

The series converges if $0<a<1$ and $a+b>1$

The case particularly requires a good estimation of the sum by an integral. A crucial lemma for this case is Lemma 4.8 of The Theory of the Riemann Zeta-function written by Titchmarsh.

Let $f(x)$ be a real differentiable function in the interval $[a,b]$, let $f'(x)$ be monotonic, and let $|f'(x)|\leq \theta <1$. Then $$ \sum_{a<n\leq b}e^{2\pi i f(n)} = \int_a^b e^{2\pi i f(x)} \ dx + O(1). $$

Let $0<a<1$. We apply the above with $f(x)=\frac{x^a}{2\pi}$. The assumptions of the lemma are satisfied for $x\geq X_0>0$. Then we obtain

$$ \sum_{X_0<n\leq N} e^{in^a}=\int_{X_0}^N e^{ix^a} \ dx + O(1). $$

Apply change of variable $x^a=t$, then $$ \int_{X_0}^N e^{ix^a} \ dx = \int_{X_0^a}^{N^a} e^{it} \frac1a t^{\frac1a -1} \ dt. $$

Now, integration by parts shows that the above integral is $O(N^{1-a})$. Thus, we have for $t\geq 1$, $$ \sum_{n\leq t} \sin(n^a) = O(t^{1-a}). $$

Then the partial summation yields $$ \sum_{n=1}^N \frac{\sin(n^a)}{n^b} = O(N^{1-a-b})+b\int_{1-}^N \frac{\sum_{n\leq t} \sin(n^a)}{t^{b+1}} \ dt. $$

Hence, we have the convergence when $b>1-a$.

The series converges if $k\in\mathbb{Z}\cap [2,\infty)$, $k-1<a<k$, and $b>1-\frac k{2^k-2}+\frac a{2^k-2}$

We apply Theorems 5.11-5.13 of The Theory of the Riemann Zeta-function written by Titchmarsh. These theorems are proved by Van Der Corput's method. If the method is applied to integer $a$, we obtain Weyl's inequality, which was used in David Speyer's answer. The method is outlined in Chapter 5 of Titchmarsh's book.

Let $f(x)$ be real and have continuous derivatives up to $k$-th order, where $k\geq 2$. Let $\lambda_k\leq f^{(k)}(x)\leq h\lambda_k$ (or the same for $-f^{(k)}(x)$). Let $b-a\geq 1$, $K=2^{k-1}$. Then $$ \sum_{a<n\leq b} e^{2\pi i f(n)}=O(h^{\frac2K}(b-a)\lambda_k^{\frac1{2K-2}})+O((b-a)^{1-\frac2K}\lambda_k^{-\frac1{2K-2}}).$$

For the series in question, use $f(n)=\frac{n^a}{2\pi}$ over diadic intervals $(N,2N]$, and $\lambda_k\asymp N^{a-k}$. Then for $N\geq 1$, $$ \sum_{N<n\leq 2N} \sin(n^a)=O(N^{1-\frac{k-a}{2^k-2}}). $$ Here, the implied constant depends only on $a$. Then adding up the results for diadic intervals, we obtain for $t\geq 1$, $$ \sum_{n\leq t} \sin(n^a)=O(t^{1-\frac{k-a}{2^k-2}}). $$ Applying partial summation, we obtain $$ \sum_{n=1}^N \frac{\sin(n^a)}{n^b} = O(N^{1-\frac{k-a}{2^k-2}-b})+b\int_{1-}^N \frac{\sum_{n\le t} \sin(n^a)}{t^{b+1}}\ dt. $$ Then the convergence of the series follows if $b>1-\frac k{2^k-2}+\frac a{2^k-2}$.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ Fantastic, now also what is suggested here (at.yorku.ca/cgi-bin/…) is clear; unfortunately, not having a purely mathematical preparation, some concepts escape me. Thank you! $\endgroup$ – TeM Jul 4 '19 at 16:32
  • $\begingroup$ Thank you very much for the demonstrations provided! $\endgroup$ – TeM Jul 9 '19 at 6:48
13
$\begingroup$

If $a > 0$ then the series converges if $b > \max(a,1-a)$.

The general principle is that the series $\sum_{n=1}^\infty f(n)$ and the integral $\int_1^\infty f(x) \, dx$ converge and diverge together if $\int_1^\infty |f'(x)| \, dx < \infty$. This is proved here.

In this case, $f(x) = \sin x^a /x^b$ and

$$\int_1^\infty |f'(x)| \, dx= \int_1^\infty \left| \frac{-b\sin x^a}{x^{b+1}} + \frac{a\cos x^a}{x^{b-a+1}} \right| \, dx \\ \leqslant \int_1^\infty \frac{b}{x^{b+1}} \, dx + \int_1^\infty \frac{a}{x^{b-a+1}} \, dx, $$

so the integrals on the RHS converge and theorem is applicable if $b > 0$ and $b > a$.

We have

$$\int_1^\infty \frac{\sin x^a}{x^b}\, dx = \frac{1}{a}\int_1^\infty \frac{\sin u}{u^{(b+a-1)/a}}\, du, $$

which converges by the Dirichlet test when $a > 0$ and $b+a-1 > 0 \implies b > 1 - a.$

If $b \leqslant 1 - a$ then the integral diverges. In this case we have an integral of the form $\int_1^\infty u^\alpha \sin u \, du$ where $\alpha \geqslant 0$. Divergence is obvious if $\alpha = 0$ since $\int_1^c \sin u \, du = \cos 1 - \cos c$.

For $\alpha > 0$ we have for any positive integer $k$,

$$\left|\int_{2k\pi}^{2(k+1)\pi} u^\alpha \sin u \, du \right| = \int_{2k\pi}^{2(k+1)\pi} u^\alpha \sin u \, du \geqslant (2 k \pi)^\alpha\int_{2k\pi}^{2(k+1)\pi} \sin u \, du = 2(2k\pi)^\alpha.$$

Since the RHS tends to $\infty$ as $k \to \infty$, the Cauchy criterion is violated and the improper integral must diverge.

Hence, the series diverges when $0 < a < b \leqslant 1 - a$.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ It's a pretty good thing, I have to sleep on and I'll tell you! For now thank you! $\endgroup$ – TeM Jul 28 '17 at 23:25
  • 1
    $\begingroup$ You're welcome. This should fill some of the remaining area. $\endgroup$ – RRL Jul 28 '17 at 23:38
  • $\begingroup$ Also when $0 < a < b$ and the integral test applies, the integral $\int_0^\infty \frac{\sin x^a}{x^b} \, dx$ diverges when $ b \leqslant 1-a$ and, hence, the series diverges as well. $\endgroup$ – RRL Jul 29 '17 at 16:21
  • 1
    $\begingroup$ Here we have an integral of the form $\int_1^\infty u^\alpha \sin u \, du$ with $\alpha \geqslant 0$. This obviously diverges, but I added a proof to the answer. $\endgroup$ – RRL Jul 29 '17 at 20:00
  • $\begingroup$ Now I understand, thank you for your patience! I've updated the image above. $\endgroup$ – TeM Jul 29 '17 at 21:35

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.