I am investigating the convergence of $$\sum _{n=1}^{\infty }\left\{ 1-n\log \frac {2n+1} {2n-1}\right\} $$ Now as per Cauchy's test for absolute convergence.

If $\overline {\lim _\limits{n\rightarrow \infty }}\left| u_{n}\right|^{{1}/{n}} < 1,\sum _\limits{n=1}^{\infty }u_{n}$ converges absolutely

Obviously, if $\overline {\lim \limits_{n\rightarrow \infty }}\left| u_{n}\right|^{{1}/{n}} > 1,\sum _\limits{n=1}^{\infty }u_{n}$ does not converge.

I observed $$\overline {\lim _{n\rightarrow \infty }}\left| \log \left( \dfrac {2n+1} {2n-1}\right) ^{-n}\right| = \overline {\lim _{n\rightarrow \infty }}\left| \log \left( 1-\dfrac {1} {n-{1}/{2}}\right) ^{-n}\right|$$

Could I take the negative power of $n$ outside the absolute brackets here? I guess even if I could establish $\log$ part converges that would only show that the overall series diverges right. Is that the correct result ? Any alternative lines of attacking this problem would be much appreciated.

  • 1
    $\begingroup$ I don't think you used well the convergence test. $\endgroup$ – Beni Bogosel Mar 7 '12 at 19:12
  • $\begingroup$ are u saying it is not possible to take the negative power of n outside the absolute brackets here ? cause if that was allowed i'd say that the expression should be less than one as n goes to infinity. $\endgroup$ – Comic Book Guy Mar 7 '12 at 19:20
  • 1
    $\begingroup$ The limit you try to calculate in your last part of the article is in fact equal to 1. And you can take the negative power of $n$ in front of the logarithm, but you can't take the negative power of $n$ out of the modulus. $\endgroup$ – Beni Bogosel Mar 7 '12 at 19:27
  • 1
    $\begingroup$ Power Series of log may help. $\endgroup$ – quartz Mar 7 '12 at 19:33
  • 1
    $\begingroup$ Try $$1-n \log \left( \frac{2n+1}{2n-1} \right) = 1 - n \log \left( \frac{1+\frac1{2n}}{1-\frac1{2n}} \right) = 1 - 2n \left( \frac1{2n} + \frac1{3(2n)^3} + \frac1{5(2n)^5} + \cdots \right) =\\ - \sum_{k=1,2}^{\infty} \frac1{(2k+1)(2n)^{2k}}.$$ After this you need to argue the swapping of the infinite sums and hence show it converges. $\endgroup$ – user17762 Mar 7 '12 at 19:41

You could prove the convergence of the series using a comparison criterion. For example, calculate the limit of $|a_n|/(\frac{1}{n^2})$. You should then calculate $$\lim_{n \to \infty}\left| n^2-n^3\log \left(\frac{2n+1}{2n-1} \right)\right| $$.

For this calculation, the simplest method I could think of was expanding in Taylor series.

$$\log(x+1)-\log(x-1)=\log \left(1+\frac{1}{x}\right)-\log\left(1-\frac{1}{x}\right)=2\sum_{k \text{ odd}}\frac{1}{kx^k}$$

Then you have to calculate

$$ \lim_{n \to \infty} \left|n^2-n^3\cdot 2\left(\frac{1}{2n}+\frac{1}{3(2n)^3}+\frac{1}{5(2n)^5}+... \right)\right|=\frac{1}{12}$$

Therefore $\sum a_n$ is absolutely convergent, and in particular convergent.

  • $\begingroup$ Nicely done, though i did n't follow how u got 1/12 as the result $\endgroup$ – Comic Book Guy Mar 7 '12 at 19:57
  • $\begingroup$ Also as per comparison series do n't we need to find a series which forms and upper bound to our series and show since that converges absolutely ours would too, in the answer you posted i am unsure what series that would be as we only seem to change our series and operate on that. $\endgroup$ – Comic Book Guy Mar 7 '12 at 20:04
  • $\begingroup$ There are three comparison criteria. Two with inequalities, and the last one, says that $a_n /b_n$ tends to a positive, bounded limit, then the series $a_n, b_n$ have the same nature. The comparison criteria with inequalities are obviously equivalent to the one with the limit. $\endgroup$ – Beni Bogosel Mar 7 '12 at 21:24
  • $\begingroup$ As for the limit in my answer, try and open the bracket, and see which terms remain. It's quite easy. $\endgroup$ – Beni Bogosel Mar 7 '12 at 21:25
  • $\begingroup$ +1. Nicely done. This also quantifies how the series converges like the series $\sum \frac1{n^2}$. $\endgroup$ – user17762 Mar 7 '12 at 22:15

$1-n\log\left(\frac{2n+1}{2n-1}\right)= 1-n\log\left(1+\frac{1}{n-\frac{1}{2}}\right)\sim 1-\frac{n}{n-\frac{1}{2}}+\frac{n}{2\left(n-\frac{1}{2}\right)^{2}}=\frac{1}{(2n-1)^{2}}\sim \frac{1}{4 n^{2}}$ then the series converges.


This question has an interesting aspect that merits being commented, which is that we can compute a closed form for the sum using Mellin transforms and harmonic sums.

Introduce the sum $S(x)$ given by $$S(x) = \sum_{n\ge 1} \left(1- xn \log\frac{2xn+1}{2xn-1}\right)$$ so that we are interested in $S(1).$

As mentioned before the sum term is harmonic and may be evaluated by inverting its Mellin transform.

Recall the harmonic sum identity $$\mathfrak{M}\left(\sum_{k\ge 1} \lambda_k g(\mu_k x);s\right) = \left(\sum_{k\ge 1} \frac{\lambda_k}{\mu_k^s} \right) g^*(s)$$ where $g^*(s)$ is the Mellin transform of $g(x).$

In the present case we have $$\lambda_k = 1, \quad \mu_k = k \quad \text{and} \quad g(x) = 1-x\log\frac{2x+1}{2x-1} = 1-x\log\left(1+\frac{2}{2x-1}\right).$$

The abscissa of convergence of $$\sum_{k\ge 1} \frac{\lambda_k}{\mu_k^s} = \zeta(s)$$ is $\Re(s)>1.$

Now to find the Mellin transform $g^*(s)$ of $g(x)$ the fundamental strip being $\langle 0,2\rangle$ (which contains the abscissa of convergence of the Dirichlet series sum term) we first use integration by parts to get $$\int_0^\infty \left(-x\log\left(1+\frac{2}{2x-1}\right)\right) x^{s-1} dx \\= \left[\left(-x\log\left(1+\frac{2}{2x-1}\right)\right) \frac{x^{s+1}}{s+1}\right]_0^\infty - \int_0^\infty \frac{4}{4x^2-1} \frac{x^{s+1}}{s+1} dx = - \int_0^\infty \frac{4}{4x^2-1} \frac{x^{s+1}}{s+1} dx.$$ with the fundamental strip being $\langle -1, 0\rangle$. It becomes evident that we require the following Mellin transform: $$h^*(s) = \int_0^\infty h(x) x^{s-1} dx$$ where $$h(x) = \frac{4}{4x^2-1}.$$ This transform integral is not strictly speaking convergent but we can compute its principal value by using a semicircular contour in the upper half plane that is traversed clockwise and picks up half the residues at the two poles at $\pm 1/2.$ This gives $$h^*(s) (1-e^{i\pi s}) = \frac{1}{2} \times 2 \pi i \left(\mathrm{Res}(h(x) x^{s-1}; x=1/2) + \mathrm{Res}(h(x) x^{s-1}; x=-1/2)\right)$$ which gives $$h^*(s) (1-e^{i\pi s}) = \frac{1}{2} \times 2 \pi i ((1/2)^{s-1}-(-1/2)^{s-1}) = 2^{-s} \times 2 \pi i (1+(-1)^s).$$ This yields $$h^*(s) = 2^{-s} \times 2 \pi i \frac{1+e^{i\pi s}}{1-e^{i\pi s}} = 2^{-s} \times 2 \pi i \frac{e^{-i\pi s/2}+e^{i\pi s/2}}{e^{-i\pi s/2}-e^{i\pi s/2}} = -\frac{2}{2^s} \pi \cot(\pi s/2).$$ Returning to $g^*(s)$ we have $$g^*(s) = \frac{2}{2^{s+2}} \frac{1}{s+1} \pi \cot(\pi (s+2)/2) = \frac{1}{2^{s+1}} \frac{1}{s+1} \pi \cot(\pi s / 2).$$

Therefore the Mellin transform $Q(s)$ of $S(x)$ is given by $$\frac{1}{2^{s+1}} \frac{1}{s+1} \pi \cot(\pi s / 2)\zeta(s).$$

The Mellin inversion integral here is $$\frac{1}{2\pi i} \int_{3/2-i\infty}^{3/2+i\infty} Q(s)/x^s ds$$ which we evaluate by shifting it to the right for an expansion about infinity.

We must sum the residues at the poles at the positive even integers of $Q(s)/x^s$ which are $$\mathrm{Res}(Q(s)/x^s; s=2q) = \frac{1}{2^{2q+1}} \frac{1}{2q+1} \times 2 \times \zeta(2q) \frac{1}{x^{2q}} \\ =\frac{1}{2^{2q+1}} \frac{1}{2q+1} \times 2 \times \frac{(-1)^{q+1} B_{2q} (2\pi)^{2q}}{2\times (2q)!} \frac{1}{x^{2q}} = \frac{1}{2} \times \frac{(-1)^{q+1} B_{2q} \pi^{2q}}{(2q+1)!} \frac{1}{x^{2q}}.$$

We thus require the sum $$ - \frac{1}{2} \sum_{q\ge 1} \frac{i^{2q} B_{2q} \pi^{2q}}{(2q+1)!} \frac{1}{x^{2q}}.$$

The exponential generating function of the even Bernoulli numbers is $$-1 + \frac{1}{2} t + \frac{t}{e^t-1} = \sum_{q\ge 1} B_{2q} \frac{t^{2q}}{(2q)!}.$$ Integrate this to obtain $$-t - \frac{t^2}{4} + t\log(1-e^t) + \mathrm{Li}_2(e^t).$$ The constant that appeared during the integration was the well-known zeta function value $\mathrm{Li}_2(1) = \zeta(2) = \pi^2/6$ so that we finally have $$ \sum_{q\ge 1} B_{2q} \frac{t^{2q}}{(2q+1)!} = \frac{1}{t} \left(-\frac{\pi^2}{6}-t - \frac{t^2}{4} + t\log(1-e^t) + \mathrm{Li}_2(e^t)\right).$$ To conclude use another well known zeta function value which is $\mathrm{Li}_2(-1) = -\pi^2/12$ (derived from $\mathrm{Li}_2(1)$) and put $t=i\pi /x = i\pi$ to obtain that (we lose the minus sign because we are shifting to the right) $$S(1) = \frac{1}{2} \frac{1}{i\pi} \left(-\frac{\pi^2}{6}-i\pi + \frac{\pi^2}{4} + i\pi\log 2 + \mathrm{Li}_2(-1)\right) \\ = \frac{1}{2} \frac{1}{i\pi} \left(-i\pi + i\pi \log 2 \right) = \frac{1}{2} (\log 2 - 1).$$

Remark. The integral of the generating function of the Bernoulli numbers is easily verified by differentiation: $$\left(-\frac{t^2}{2} + t\log(1-e^t) + \mathrm{Li}_2(e^t)\right)' = -t + \frac{t}{1-e^t} (-e^t) + \log(1-e^t) + \left(\sum_{n\ge 1} \frac{e^{nt}}{n^2}\right)' \\ = \frac{-t+te^t}{1-e^t} - \frac{te^t}{1-e^t} + \log(1-e^t) + \sum_{n\ge 1} \frac{e^{tn}}{n} \\ = \frac{-t}{1-e^t} + \log(1-e^t) - \log(1-e^t) = \frac{t}{e^t-1}.$$


HINT: Relate $\lim_{n \to \infty}\left| n^2-n^3\log \left(\frac{2n+1}{2n-1} \right)\right|$ to $\lim_{n \to \infty} n-n^2\log \left(\frac{n+1}{n} \right)$whose limit is $\frac{1}{2}$ and can be almost elementarily proved by replacing n by $\frac{1}{x}$ when $x \rightarrow 0$. The second limit comes from a highschool book, 11th grade, often met during courses.


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.