2
$\begingroup$

I am struggling with the problem in the title of this post. I have tried many different methods, but nothing has worked so far. I only managed to derive the first term of the asymptotic expansion:

$f(n)=\sum\limits_{k=1}^{\infty}\frac{(-1)^k}{k}\ln(n^2+k) =\sum\limits_{k=1}^{\infty}\frac{(-1)^k}{k}\ln(n^2) + \sum\limits_{k=1}^{\infty}\frac{(-1)^k}{k}\ln\left(1+\frac{k}{n^2}\right) = -\ln (2) \ln(n^2)+ o(1)$, $\quad n \to +\infty$,

where I used $\ln(n^2+k)=\ln\left( n^2\left(1+\frac{k}{n^2}\right)\right)= \ln(n^2) +\ln \left(1+\frac{k}{n^2}\right)$.

How can I derive the next expansion term?

$\endgroup$

3 Answers 3

6
$\begingroup$

You found that $$f(n)=-2\ln(2)\ln (n)+\sum\limits_{k=1}^{\infty}\frac{(-1)^k}{k}\ln\left(1+\frac{k}{n^2}\right)$$ Using the Frullani integral $$\ln a=\int_0^\infty\frac{e^{-t}-e^{-at}}tdt$$ $$S(n)=\sum\limits_{k=1}^{\infty}\frac{(-1)^k}{k}\ln\left(1+\frac{k}{n^2}\right)=\sum\limits_{k=1}^{\infty}\frac{(-1)^k}{k}\int_0^\infty\frac{e^{-t}-e^{-t(1+\frac k{n^2})}}tdt$$ $$=\int_0^\infty\frac{e^{-t}}t\sum_{k=1}^\infty\frac{(-1)^k}k\left(1-e^{-\frac{tk}{n^2}}\right)dt=\int_0^\infty\frac{e^{-t}}t\ln\frac{1+e^{-\frac t{n^2}}}2dt$$ Decomposing $\ln (1+e^{-\frac t{n^2}})=\ln\left(2-\frac t{n^2}+\frac{t^2}{2n^4}+O(1/n^6)\right)$ $$S(n)=\int_0^\infty\frac{e^{-t}}t\left(-\frac t{2n^2}+\frac {t^2}{4n^4}-\frac {t^2}{8n^4}+O\left(\frac1{n^6}\right)\right)dt$$ $$=-\frac1{2n^2}+\frac1{8n^4}+O\left(\frac1{n^6}\right)$$ Therefore, $$\boxed{\,\,f(n)=-2\ln(2)\ln(n)-\frac1{2n^2}+\frac1{8n^4}+O\left(\frac1{n^6}\right)\,\,}$$

$\endgroup$
2
  • $\begingroup$ Thanks a lot! Could you explain the last equality sign before the words "decomposing ln"? $\endgroup$
    – Coala
    Commented Jun 22, 2023 at 18:52
  • 1
    $\begingroup$ $$\sum_{k=1}^\infty\frac{(-1)^k}k\left(1-e^{-\frac{tk}{n^2}}\right)=-\ln2-\sum_{k=1}^\infty\frac{(-e^{-\frac t{n^2}})^k}k$$ $$=-\ln2+\frac{e^{-\frac t{n^2}}}1-\frac{e^{-\frac {2t}{n^2}}}2+\frac{e^{-\frac {3t}{n^2}}}3-+...=\ln(1+e^{-\frac t{n^2}})-\ln2$$ $\endgroup$
    – Svyatoslav
    Commented Jun 22, 2023 at 19:09
3
$\begingroup$

We can derive a complete asymptotic series starting from Svyatoslav's integral formula. Note that $$ \int_0^{ + \infty } {\frac{{{\rm e}^{ - t} }}{t}\log \left( {\frac{{1 + {\rm e}^{ - t/n^2 } }}{2}} \right){\rm d}t} = \int_0^{ + \infty } {\frac{{{\rm e}^{ - n^2 s} }}{s}\log \left( {\frac{{1 + {\rm e}^{ - s} }}{2}} \right){\rm d}s} , $$ and $$ \log \left( {\frac{{1 + {\rm e}^{ - s} }}{2}} \right) = - \int_0^s {\frac{{{\rm d}t}}{{{\rm e}^t + 1}}} = - \frac{1}{2}\sum\limits_{m = 0}^\infty {\frac{{E_m (0)}}{{(m + 1)!}}s^{m + 1} } = - \frac{s}{2} + \sum\limits_{m = 1}^\infty {\frac{{(2^{2m} - 1)B_{2m} }}{{2m(2m)!}}s^{2m} } $$ for $|s|<\pi$. Here the $E_m(x)$ are the Euler polynomials and the $B_m$ are the Bernoulli numbers (cf. $(24.4.26)$ and $(24.2.2)$). Then, by Watson's lemma, $$ \int_0^{ + \infty } {\frac{{{\rm e}^{ - n^2 s} }}{s}\log \left( {\frac{{1 + {\rm e}^{ - s} }}{2}} \right){\rm d}s} \sim - \frac{1}{{2n^2 }} + \sum\limits_{m = 1}^\infty {\frac{{(2^{2m} - 1)B_{2m} }}{{(2m)^2 n^{4m} }}} $$ as $n\to+\infty$. Accordingly, $$ f(n) \sim - (2\log 2)\log n - \frac{1}{{2n^2 }} + \sum\limits_{m = 1}^\infty {\frac{{(2^{2m} - 1)B_{2m} }}{{(2m)^2 n^{4m} }}} $$ as $n\to+\infty$.

$\endgroup$
1
  • 1
    $\begingroup$ Gary, every your solution is an excellent learning exercise! I had a strong feeling that a complete asymptotic can be obtained, and that the Euler polynomials are involved, but was not able to proceed. In my opinion, your solution fully addresses the question and, therefore, is the best. $\endgroup$
    – Svyatoslav
    Commented Jun 23, 2023 at 1:39
2
$\begingroup$

Just for the fun (because so good answers from @Svyatoslav and @Gary).

$$f(n)=\sum\limits_{k=1}^{\infty}\frac{(-1)^k}{k}\log(n^2+k) $$ $$f'(n)=2\sum\limits_{k=1}^{\infty} (-1)^k\frac{ n}{k \left(k+n^2\right)}=\frac 2n \left(\Phi \left(-1,1,n^2+1\right)-\log (2)\right)$$ where appears the Hurwitz-Lerch transcendent function $$f(n)=-2\log (2) \log (n)+2\int \frac{\Phi \left(-1,1,n^2+1\right)}{n}\,dn$$ which cannot be integrated.

But $$ \frac 12 < m\,\Phi \left(-1,1,m\right)< 1$$ So $$\log \left(\frac{n\, 2^{-2 \log (n)}}{\sqrt{n^2+1}}\right)< f(n) < \log \left(\frac{n^2\ 2^{-2 \log (n)}}{n^2+1}\right)$$ Which seems to be decent. $$\left( \begin{array}{cccc} n & \text{left bound} & \text{summation}& \text{right bound} \\ 10^1 & -3.19703589584 & -3.19704823073 & -3.20201106127 \\ 10^2 & -6.38417145833 & -6.38417145958 & -6.38422145583 \\ 10^3 & -9.57618269125 & -9.57618269125 & -9.57618319125 \\ 10^4 & -12.7682429267 & -12.7682429267 & -12.7682429317 \\ 10^5 & -15.9603036521 & -15.9603036521 & -15.9603036522 \\ 10^6 & -19.1523643825 & -19.1523643825 & -19.1523643825 \\ \end{array} \right)$$

$\endgroup$

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .