How to evaluate following expression? $$ \sum_{n=1}^{\infty}\frac{1}{1+n^2+n^4}$$ I doubt it is a telescopic Sum.

  • 2
    $\begingroup$ What do you mean evaluate? This sum has no easily attainable closed form, but it evidently converges by comparing to $a_{n}=n^{-2}.$ To get a sense of the challenges you face in evaluating this sum analytically, see math.stackexchange.com/questions/8337/… for the simpler case $\sum n^{-2}.$ $\endgroup$ – Sargera Oct 10 '14 at 17:22
  • $\begingroup$ Have you attempted partial fraction decomposition? $\endgroup$ – David H Oct 10 '14 at 17:28
  • 2
    $\begingroup$ Maple give the answer $$ -1/2+1/6\,\sqrt {3}\pi \,\tanh \left( 1/2\,\pi \,\sqrt {3} \right) $$ $\endgroup$ – Leox Oct 10 '14 at 17:36

Let $\omega=\exp\left(\frac{2\pi i}{3}\right)$. Then: $$ n^4+n^2+1 = (n^2-\omega)(n^2-\omega^2), $$ so: $$\frac{1}{1+n^2+n^4}=\frac{1}{i\sqrt{3}}\left(\frac{1}{n^2-\omega}-\frac{1}{n^2-\omega^2}\right)$$ and: $$\sum_{n=1}^{+\infty}\frac{1}{1+n^2+n^4}=\frac{1}{\sqrt{3}}\Im\sum_{n=1}^{+\infty}\frac{1}{n^2-\omega}$$ can be computed through the identity: $$\sum_{n=1}^{+\infty}\frac{1}{n^2+a}=\frac{-1+\pi\sqrt{a}\coth(\pi\sqrt{a})}{2a}$$ that follows from considering the logarithmic derivative of the Weierstrass product for the $\sinh$ function. By putting all together, we have:


as stated by WA.


First, we have $$ \begin{align} \frac1{z^4+z^2+1} &=\frac1{12}\left( \frac{-3-i\sqrt3}{z-e^{\pi i/3}} +\frac{3+i\sqrt3}{z-e^{4\pi i/3}} +\frac{3-i\sqrt3}{z-e^{2\pi i/3}} +\frac{-3+i\sqrt3}{z-e^{5\pi i/3}} \right)\tag{1} \end{align} $$ Let $\gamma$ be the rectangle $$ [-1-i,1-i]\cup[1-i,1+i]\cup[1+i,-1+i]\cup[-1+i,-1-i]\tag{2} $$ then the integral $$ \frac1{2\pi i}\int_{(n+\frac12)\gamma}\frac{\pi\cot(\pi z)}{z^4+z^2+1}\,\mathrm{d}z\tag{3} $$ tends to $0$ since along the horizontal paths, $|\pi\cot(\pi z)|\to\pi$ and along the vertical paths, $|\pi\cot(\pi z)|\lt\pi$.

Since $\pi\cot(\pi z)$ has residue $1$ at each integer, we get that $(3)$ is the sum of the residues of $(1)$ times $\pi\cot(\pi z)$ at the singularities of $(1)$ plus $$ 1+2\sum_{n=1}^\infty\frac1{n^4+n^2+1}\tag{4} $$ The sum of the residues of $(1)$ times $\pi\cot(\pi z)$ at the singularities of $(1)$ is $$ -4\,\mathrm{Re}\left(\tfrac{3+i\sqrt3}{12}\pi\cot\left(\pi\tfrac{1+i\sqrt3}2\right)\right)\tag{5} $$ Since the sum of $(4)$ and $(5)$ is $0$, we have $$ \begin{align} 1+2\sum_{n=1}^\infty\frac1{n^4+n^2+1} &=4\,\mathrm{Re}\left(\tfrac{3+i\sqrt3}{12}\pi\cot\left(\pi\tfrac{1+i\sqrt3}2\right)\right)\\ &=-4\,\mathrm{Re}\left(\tfrac{3+i\sqrt3}{12}\pi\tan\left(\pi\tfrac{i\sqrt3}2\right)\right)\\ &=\tfrac{\pi\sqrt3}3\tanh\left(\pi\tfrac{\sqrt3}2\right)\tag{6} \end{align} $$ Therefore, $$ \sum_{n=1}^\infty\frac1{n^4+n^2+1}=\frac{\pi\sqrt3}6\tanh\left(\pi\frac{\sqrt3}2\right)-\frac12\tag{7} $$


Let us use the Abel-Plana formula $$ \sum_{n=0}^\infty f(n)= \int_0^\infty f(x) \, dx+ \frac 1 2 f(0)+i \int_0^\infty \frac{f(i t)-f(-i t)}{e^{2\pi t}-1} \, dt $$ for $f(n)=\dfrac{1}{1+n^2+n^4}$.

Since $ f(it)=f(-it) $ we obtain $$ \sum_{n=0}^\infty f(n)= \int_0^\infty f(x) \, dx+ \frac 1 2 f(0). $$ By standard way we get $$ \int_0^\infty\frac{1}{1+x^2+x^4}=\frac{\pi \sqrt{3}}{6}. $$ Now $$ \sum_{i=1}^\infty f(n)=\sum_{i=0}^\infty f(n)-f(0)=\frac{\pi \sqrt{3}}{6}-\frac{1}{2}. $$

  • $\begingroup$ Have you checked? Jack's answer is approximately $0.39907364028134509045$, and yours is $0.40689968211710892530$. $\endgroup$ – robjohn Oct 11 '14 at 14:09
  • 3
    $\begingroup$ You cannot apply the Abel-Plana formula because $f(z) = \frac{1}{1+z^2+z^4}$ is not holomorphic over the left half-plane $\Re z \ge 0$. $f(z)$ has two poles at $\frac12 \pm \frac{\sqrt{3}}{2}i$ there. You can apply AP formula to the sum start at $n = 1$ but then $f(1+it) \ne $f(1-it)$ and the last term in the AP formula won't cancel.... $\endgroup$ – achille hui Oct 11 '14 at 15:04
  • 1
    $\begingroup$ Is there any way to fix this problem? $\endgroup$ – Leox Oct 11 '14 at 16:08
  • 2
    $\begingroup$ @Leox, the real beauty of your method is $f(it) = f(-it)$ so you don't need to evaluate the last integral which look nasty. In principle, if you can find another sequence which is 1) easy to evaluate, 2) a function in $n^2$, 3) same singular behavior near the two poles, then you can evaluate the difference of this sum and the easier one using this method. However, I have no idea how to get hold of such an easier sequence. $\endgroup$ – achille hui Oct 12 '14 at 0:13

Hint: Apply residue calculus on $$f(z)= \frac{\pi}{(1+z^{2}+z^{4})\tan(\pi z)}$$ Where you want to integrate over the square $S_{R}$ with vertices $(R+\frac{1}{2})(\pm1 \pm i)$

The reason is that $$Res(f(z), n\pi) = \frac{1}{1+n^{2}+n^{4}}$$ I'll leave the details to you!


Lots of nice analytical ideas! I will be trying to use them from now one. Numerically, it is not too difficult to get a nice answer. We use the first term of the Euler Maclaurin Summation formula which says:

$$\sum _{n=a}^b f(n)\sim \int_a^b f(x) \, dx+\frac{1}{2} (f(a)+f(b))$$

But it is alot more useful in getting the tail so we first compute

$$ \sum _{n=1}^{99} \frac{1}{n^4+n^2+1}=0.39907330193518 $$

for the integral on the RHS we use the trick that as $x$ gets large the $x^4$ term will completely drown out the $x^2 + 1$ terms.

$$\int_{100}^{\infty } \frac{1}{x^4+x^2+1} \, dx\approx \ \int_{100}^{\infty } \frac{1}{x^4} \, dx=\frac{1}{3000000}$$

Now it is just arithmetic,

$$0.39907330193518\, +\frac{1}{3000000}+\frac{1}{2} \left(\frac{1}{100^4+100^2+1}+0\right)$$

which equals $ 0.39907364026801 $ which is quite close, differing from the exact answer in the 11th place.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy