I've been bored and came across in my book a pretty straightforward series problem, namely to determine the convergence of

$$ \sum_{n = 1}^{\infty} \left[\sin\left(1 \over 2n\right) - \sin\left(1 \over 2n + 1\right)\right] $$

Doing so was trivial by rewriting it as an alternating series involving the term $(-1)^k\sin\frac1k$.

Naturally, though, I was curious as to whether this series can be reduced to a simpler closed form in terms of more fundamental constants. Unfortunately I do not immediately know of any techniques of use here or even whether it permits such a 'nice' form. Do any of you?

I do know from playing with the Euler-Maclaurin sums the value should be something near $0.290674$. As $n\to\infty$ I know the sequence terms behave increasingly like those of the alternating harmonic series (as $\sin x\sim x$ for $|x|\ll1$), which helps explain why it appears relatively near $1-\log2$. I have also found that the difference between it and the alternating harmonic series starting with $1/2$ is near $0.016179$.

I should note that I am a high school student with an amateur interest in recreational math. My knowledge extends only as far as elementary calculus of multiple variables and first-year ordinary and partial differential equations. It may very well be that an obvious approach exists that I've completely missed and so I feel obligated to apologize in advance.

  • 1
    $\begingroup$ Using the trigonometric identity for $\sin(a)-\sin(b)$ might be helpful . $\endgroup$ Aug 22, 2013 at 20:55
  • $\begingroup$ I tried that @ZaidAlyafeai but unfortunately I could not determine anything neat. I will look into it again however. I must leave for class though as it begins in a few minutes. Thank you for your response. $\endgroup$
    – obataku
    Aug 22, 2013 at 20:56
  • 1
    $\begingroup$ Have you tried considering the imaginary part of $e^{i\theta} = \cos\theta + i\sin\theta$ ? I don't know if this will help, but might be worth a try... $\endgroup$
    – gone
    Aug 22, 2013 at 21:00
  • 15
    $\begingroup$ If you just expand the sums, you get (if I haven't miscalculated) $$1 - \log 2 + \sum_{k=1}^\infty \frac{(-1)^k\bigl(1-(1-2^{-2k})\zeta(2k+1)\bigr)}{(2k+1)!} = \sin 1 - \log 2 +\sum_{k=1}^\infty \frac{(-1)^{k+1}(1-2^{-2k})\zeta(2k+1)}{(2k+1)!},$$ which isn't a nice closed-form expression, but at least involves everybody's favourite, the Riemann $\zeta$-function. $\endgroup$ Aug 22, 2013 at 21:10
  • 1
    $\begingroup$ For me, both formulae produce $0.2906741\ldots$. Although I haven't taken any measures to minimise floating point errors, those should not affect the given significant digits. $\endgroup$ Aug 23, 2013 at 11:29

2 Answers 2


$$\sum_{n=1}^\infty \sin\left({1\over 2n}\right)-\sin\left({1\over 2n+1}\right)=-\sum_{n=1}^\infty\int_{1\over 2n+1}^{1\over 2n}\cos x\, dx$$

These are integrals of a bounded function on the sets

$$\left[{1\over 2n+1},{1\over 2n}\right]$$

The measure of these sets is

$${1\over 2n}-{1\over 2n+1}={1\over 2n(2n+1)}$$

Hence the sum is absolutely convergent by Jensen's inequality since

$$\left|\sum_{n=1}^\infty\int_{1\over 2n}^{1\over 2n+1}\cos x\right|\le\sum_{n=1}^\infty\int_{1\over 2n}^{1\over 2n+1}|\cos x|\le\sum_{n=1}^\infty\int_{1\over 2n+1}^{1\over 2n}1\;dx=\sum_{n=1}^\infty {1\over 2n(2n+1)}.$$


I can't give precise value for the sum of the series, but note that the Mean Value Theorem tells us that it may be written in the form $\sum_{n=1}^{\infty} \frac{ \cos(\theta_{2n})}{2n(2n+1)},$ where each $\theta_{2n} \in (\frac{1}{2n+1},\frac{1}{2n})$ which seems to mean that it differs from $ 1 - \log2 = \sum_{n=1}^{\infty}\frac{1}{2n(2n+1)}$ by something close to $\sum_{n=1}^{\infty}\frac{1}{32n^{4}} = \frac{\pi^{4}}{2880}$, (and is smaller than that sum).


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.