Tricky Sum 3.0

Calculus Level 5

S ( k ) = n = 1 n k e 2 π n 1 S(k)= \sum_{n=1}^\infty \dfrac{n^k}{e^{2\pi n}-1}

Define S ( k ) S(k) as above, find 1 S ( 5 ) + 1 S ( 9 ) + 1 S ( 13 ) \dfrac{1}{S(5)}+\dfrac{1}{S(9)}+\dfrac{1}{S(13)} .


The answer is 792.

This section requires Javascript.
You are seeing this because something didn't load right. We suggest you, (a) try refreshing the page, (b) enabling javascript if it is disabled on your browser and, finally, (c) loading the non-javascript version of this page . We're sorry about the hassle.

1 solution

Digvijay Singh
Aug 8, 2020

Using complex analysis, we can prove the following famous sum (proofs can be found here ) n = 1 1 x 2 + n 2 = π x coth ( π x ) 1 2 x 2 \displaystyle \sum_{n=1}^\infty\dfrac{1}{x^2+n^2}=\dfrac{\pi x\coth(\pi x)-1}{2x^2} which rearranges to 1 2 π x 1 2 + x π n = 1 1 x 2 + n 2 = 1 e 2 π x 1 \dfrac{1}{2\pi x}-\dfrac{1}{2}+\dfrac{x}{\pi} \displaystyle \sum_{n=1}^\infty\dfrac{1}{x^2+n^2}=\dfrac{1}{e^{2\pi x}-1} 1 2 π x 1 2 + 1 2 π n = 1 ( 1 x + i n + 1 x i n ) = 1 e 2 π x 1 \implies\dfrac{1}{2\pi x}-\dfrac{1}{2}+\dfrac{1}{2\pi} \displaystyle \sum_{n=1}^\infty\left(\dfrac{1}{x+in}+\dfrac{1}{x-in}\right)=\dfrac{1}{e^{2\pi x}-1} Note that e 2 π x + e 4 π x + e 6 π x + = e 2 π x 1 e 2 π x = 1 e 2 π x 1 e^{-2\pi x}+e^{-4\pi x}+e^{-6\pi x}+\cdots=\dfrac{e^{-2\pi x}}{1-e^{-2\pi x}}=\dfrac{1}{e^{2\pi x}-1} e 2 π x + e 4 π x + e 6 π x + = 1 2 π x 1 2 + 1 2 π n = 1 ( 1 x + i n + 1 x i n ) \implies e^{-2\pi x}+e^{-4\pi x}+e^{-6\pi x}+\cdots=\dfrac{1}{2\pi x}-\dfrac{1}{2}+\dfrac{1}{2\pi} \displaystyle \sum_{n=1}^\infty\left(\dfrac{1}{x+in}+\dfrac{1}{x-in}\right) Let k = 4 p + 1 k=4p+1 where p N p\in\mathbb{N} . Differentiating both sides k k times w.r.t. x x gives 1 k e 2 π x + 2 k e 4 π x + 3 k e 6 π x + = k ! ( 2 π x ) k + 1 + k ! ( 2 π ) k + 1 n = 1 ( 1 ( x + i n ) k + 1 + 1 ( x i n ) k + 1 ) 1^{k}e^{-2\pi x}+2^{k}e^{-4\pi x}+3^{k}e^{-6\pi x}+\cdots=\dfrac{k!}{(2\pi x)^{k+1}}+\dfrac{k!}{(2\pi)^{k+1}} \displaystyle \sum_{n=1}^\infty\left(\dfrac{1}{(x+in)^{k+1}}+\dfrac{1}{(x-in)^{k+1}}\right) Summing both sides from x = 1 x=1 to \infty gives 1 k e 2 π 1 + 2 k e 4 π 1 + 3 k e 6 π 1 + = k ! ( 2 π ) k + 1 ζ ( k + 1 ) + k ! ( 2 π ) k + 1 x = 1 n = 1 ( 1 ( x + i n ) k + 1 + 1 ( x i n ) k + 1 ) \dfrac{1^{k}}{e^{2\pi}-1}+\dfrac{2^{k}}{e^{4\pi}-1}+\dfrac{3^{k}}{e^{6\pi}-1}+\cdots=\dfrac{k!}{(2\pi)^{k+1}}\zeta(k+1)+\dfrac{k!}{(2\pi)^{k+1}} \displaystyle \sum_{x=1}^{\infty}\sum_{n=1}^\infty\left(\dfrac{1}{(x+in)^{k+1}}+\dfrac{1}{(x-in)^{k+1}}\right) For k = 4 p + 1 k=4p+1 where p N p\in\mathbb{N} , the double sum on the right changes sign when interchanging x x and n n . Hence, the double sum is 0 0 for such k k . S ( k ) = n = 1 n k e 2 π n 1 = k ! ( 2 π ) k + 1 ζ ( k + 1 ) = B k + 1 2 ( k + 1 ) \implies S(k)=\sum_{n=1}^\infty \dfrac{n^k}{e^{2\pi n}-1}=\dfrac{k!}{(2\pi)^{k+1}}\zeta(k+1)=\boxed{\dfrac{B_{k+1}}{2(k+1)}} where ζ ( s ) \zeta(s) is the Riemann zeta function and B n B_n is the n th n^\text{th} Bernoulli number .

and hence S ( 5 ) = 1 504 , S ( 9 ) = 1 264 , S(5)=\dfrac{1}{504},S(9)=\dfrac{1}{264}, and S ( 13 ) = 1 24 S(13)=\dfrac{1}{24} .

giving 1 S ( 5 ) + 1 S ( 9 ) + 1 S ( 13 ) = 504 + 264 + 24 = 792 \dfrac{1}{S(5)}+\dfrac{1}{S(9)}+\dfrac{1}{S(13)}=504+264+24=\boxed{792} .

Alternatively, this result can be found in the paper that gave the previous two problems!

You need to be a little more careful when differentiating the expansion for 1 e 2 π x 1 \frac{1}{e^{2\pi x} - 1} term-by term in two different ways. That is because it is not always valid to differentiate a series term-by-term. To be able to justify term-by-term differentiability, you generally need something like uniform convergence of the series and its derivative, and the series n = 1 1 x + i n \sum_{n=1}^\infty \frac{1}{x + in} is not uniformly convergent. The problem only occurs in the first derivative, since the series n = 1 1 ( x ± i n ) k \sum_{n=1}^\infty \frac{1}{(x \pm in)^k} are uniformly convergent on [ 1 , X ] [1,X] for any X > 0 X > 0 , which is enough.

The way round this problem is to avoid expanding n = 1 1 x 2 + n 2 \sum_{n=1}^\infty\frac{1}{x^2 + n^2} , since this series is uniformly convergent on [ 1 , ) [1,\infty) . This enables the first differentiation.

Mark Hennings - 10 months, 1 week ago

This series can also be found by Abel-plana formula \textrm{Abel-plana formula} n 0 f ( z ) = 0 f ( z ) d z + f ( 0 ) 2 + i 0 f ( i z ) f ( i z ) e 2 π z 1 d z \sum_{n≥0}^∞ f(z) = \int_0^∞ f(z)dz +\frac{f(0)}{2} +i\int_0^∞ \frac{f(iz)-f(-iz)}{{e^{2πz} -1}} dz

Dwaipayan Shikari - 5 months, 4 weeks ago

0 pending reports

×

Problem Loading...

Note Loading...

Set Loading...