4

Determine whether $\sum_{n=1}^{\infty}\dfrac{\sin (n^2)}{n}$ converges.

This question was proposed by Dr. Wolfgang Hintzeh here: Convergence of $\sum_{k=1}^\infty \frac{\sin(k(k-1))}{k}$

My attempt

I can see that all terms are bounded by $\frac{1}{n}$ but we cannot conclude that all terms are less than some $\frac{1}{n^p},p>1$. And because both Ratio Test and Comparison Test give no information about the convergence, I formed the corresponding integral, but since the Integral Test is not valid for this kind of series, I want to use it to check whether the series converges absolutely.

I constructed the graph of $\displaystyle\int_1^{\infty}\left|\dfrac{\sin (x^2)}{x}\right|dx$:

enter image description here

How is that possible for the integral to be less than $0$ on some intervals? What is wrong with my method? Can anybody give me some hints to prove or disprove its convergence?

StubbornAtom
  • 17,052
Kevin.S
  • 3,706
  • 2
    It looks unlikely to converge absolutely, I'd expect $|\sin n^2|$ to be equidistributed in the interval $[0,1]$ and so the series to compare unfavourably with $\sum\frac1n$. Proving that seems hard though. – Angina Seng Dec 08 '19 at 02:53
  • The engine computes indefinite integral and definite integral in different ways. – xbh Dec 08 '19 at 03:06
  • @xbh Ok, then how to calculate the integral? it is hard to evaluate... – Kevin.S Dec 08 '19 at 03:08
  • This series has positive terms. I don't understand why you are distinguishing it converging form it converging absolutely. And either way, the Integral Test doesn't apply, because $\sin^2(x)/x$ is not a decreasing function. – 2'5 9'2 Dec 08 '19 at 03:17
  • @alex.jordan It has both negative and positive terms. – Kevin.S Dec 08 '19 at 03:23
  • 1
    @KevinSong I might rewrite the summand as $\frac{\sin (n^2)}{n}$ (with parentheses) to avoid confusion with the more common expression $\sin^2 n = (\sin n)^2$. – Travis Willse Dec 08 '19 at 03:36
  • A harmonic series with random signs converges with probability 1, so this sequence probably converges. No idea how to prove that, though. – eyeballfrog Dec 08 '19 at 03:42
  • @TravisWillse ok, i edited it. – Kevin.S Dec 08 '19 at 03:51
  • Sorry, I saw $\sin(x)^2$ instead of $\sin(x^2)$. – 2'5 9'2 Dec 08 '19 at 04:01

2 Answers2

13

To answer the question in the title of the post, the series $$ \sum_{n=1}^{\infty}\frac{\sin(n^2)}{n} $$ converges conditionally - but the proof relies on some more advanced concepts.

Actually, this is a special case of a problem solved on mathoverflow. More precisely, on that page they show a more general result, which states that for a wide range of $x\in\mathbb R$ the series $$ \sum_{n=1}^{\infty}\frac{\sin(2\pi xn^2)}{n} $$ is conditionally convergent. Your question is the special case $x=1/2\pi$, which happens to satisfy the crucial condition on $x\in\mathbb R$ for the result to apply.

The exact condition on $x\in\mathbb R$ is that $x$ is not a Liouville number. Liouville numbers are an extremely rare class of transcendental numbers that have amazingly accurate rational approximations. There is a quantity called the irrationality measure which can be given to any irrational number, and if it takes the value $\infty$ then the number is Liouville (and conversely, if it is finite then the number is non-Liouville). This paper gives upper bounds for the irrationality measures of some common numbers, including $\pi$, which has the consequence that $\pi$ is non-Liouville. Since $x=\frac{1}{2\pi}$ is obtained from $\pi$ using rational operations, it is therefore non-Liouville as well.

pre-kidney
  • 30,223
  • 1
    About that solution on mathoverflow... in it, they use the property of Liouville numbers that there is $n^a < q \leq n$ such that $|2x - p/q| \leq 1/nq$. This looks a little different from the other definitions of Liouville numbers that I have found online. Do you happen to know where I can find this formulation, or at least why ${1 \over \pi}$ satisfies the condition? – Zarrax Dec 08 '19 at 05:08
  • 1
    @Zarrax that's a good question, which might be even more valuable directly on the mathoverflow answer itself – pre-kidney Dec 08 '19 at 05:17
  • 1
    The answer is over three years old so I wasn't sure if it would get a response, but maybe I should still do that. I'm pretty sure ${1 \over \pi}$ satisfies an appropriate approximation property but it's been hard to pin down the right reference. – Zarrax Dec 08 '19 at 05:37
3

Proof for divergence of $$ \newcommand{\abs}[1]{\left\vert #1 \right\vert} \newcommand\rme{\mathrm e} \newcommand\imu{\mathrm i} \newcommand\diff{\,\mathrm d} \DeclareMathOperator\sgn{sgn} \renewcommand \epsilon \varepsilon \newcommand\trans{^{\mathsf T}} \newcommand\F {\mathbb F} \newcommand\Z{\mathbb Z} \newcommand\R{\Bbb R} \newcommand \N {\Bbb N} \renewcommand\geq\geqslant \renewcommand\leq\leqslant \int_1^{+\infty} \abs {\frac {\sin (x^2)}{x}} \mathrm dx. $$ Easy to see $\abs {\sin (x^2)} \geq\sin^2 (x^2) = (1 - \cos (x^2))/2$ since $0 \leq \abs {\sin (u)} \leq 1$. Then $$ \int_1^{ +\infty } \frac {\abs {\sin (x^2)}}x \geq \frac 12 \int_1^{+\infty} \frac 1x \diff x - \frac 12 \int_1^{+\infty} \frac {\cos (2x^2)}x \diff x =: \frac 12( I_1 + I_2). $$ Clearly $I_1$ diverges. For $I_2$, via the substitution $u = 2x^2$, or $x = \sqrt {u/2}$, $$ I_2 = \int_2^{+\infty} \frac {\cos u} {\sqrt {u/2}} \diff \sqrt {u/2} = \int_2^{+\infty} \frac {\cos u} {2u} \diff u, $$ and this is convergent by Dirichlet Test. Therefore the original integral is a sum of a divergent integral and a convergent integral, which shall be divergent as well.

Counterexample for integral test when the terms are not decreasing

Consider $$ f(x) = \frac {\pi x} {1 + (\pi x)^6 \sin^2 (\pi x)}, x \geq 0. $$ Then $f$ is nonnegative, and the limit $f(+\infty)$ does not exist, since $f (n) = \pi n \to +\infty$ as $n \to \infty$. Then the series $\sum_1^{+\infty} f(n)$ diverges since a necessary condition for convergence is that $\lim_{n \to +\infty} f(n) = 0$. But the integral $\int_0^{+\infty} f(x) \diff x$ converges since $$ \int_n^{n+1} f(x) \diff x = \int_{n\pi}^{(n+1)\pi} \frac x {1 + x^6 \sin^2 x} \diff x\leq \frac {2\pi^2}{n^2}, $$ for each $n \in \N^*$, then the integral is bounded from above by $\pi^4/3$.

xbh
  • 8,835
  • I thought the question is about the conditional convergence of a sum, not about the integral, but I think I have misread the question. It is very confusing, since the title of the question doesn't seem to match with what is asked in the body of the question. – pre-kidney Dec 08 '19 at 03:24
  • @pre-kidney Yes. I cannot guarantee this is what the OP wants since the question is not very clear to me. – xbh Dec 08 '19 at 03:28
  • Thanks for sharing this. – Kevin.S Dec 08 '19 at 04:42