I am trying to prove that $$\sum^{N-1}_{k=1} \frac{1}{1-e^{i2\pi k/N}}=\frac{N-1}{2},$$ the closest formula I can think of is $$\sum^{N-1}_{k=0} r^k = \frac{1-r^N}{1-r},$$ but seems like it is not exactly the case.
-
I don't know if this will work for sure but I have a hunch that we have to use $e^{ix}=\cos(x) + i\sin(x)$ – Hersh Mar 01 '23 at 14:20
-
1The bottom formula is not correct. The sum needs to start from zero. – K.defaoite Mar 01 '23 at 14:32
-
I have since fixed it. – K.defaoite Mar 01 '23 at 16:44
5 Answers
It's a fun problem. The easiest way to solve it is to first note that the set $$\left\{\mathrm e^{2\pi\mathrm ik/n}\right\}_{k=1,\dots,n-1}$$ Comprise the solutions to the algebraic equation $z^n=1$, other than the trivial solution $z=1$. From our knowledge of complex analysis, we know that these solutions are evenly spread around the unit circle.
Next, we note that the function $$z\mapsto\frac{1}{1-z}$$ maps the unit circle onto the line $\{z\in\mathbb C\mid \Re(z)=1/2\}$. To see this let $t\in(0,1)$ and calculate $$\frac{1}{1-\mathrm e^{2\pi\mathrm it}}=\frac{\overline{1-\mathrm e^{2\pi\mathrm it}}}{|1-\mathrm e^{2\pi\mathrm it}|^2}=\frac{1-\mathrm e^{-\mathrm 2\pi\mathrm it}}{4\sin(\pi t)^2} \\ =\frac{1}{4\sin(\pi t)^2}\big((1-\cos(2\pi t))-\mathrm i\sin(2\pi t)\big) \\ =\frac{1}{4\sin(\pi t)^2}\big(2\sin(\pi t)^2-\mathrm i\cdot 2\sin(\pi t)\cos(\pi t)\big) \\ =\frac{1}{2}+\mathrm i\cdot\frac{1}{2}\cot(\pi t)$$
The key point is that, because the set $\left\{\mathrm e^{2\pi\mathrm ik/n}\right\}_{k=1,\dots,n-1}$ is evenly spread on the unit circle, the set $$\left\{\frac{1}{1-\mathrm e^{2\pi\mathrm ik/n}}\right\}_{k=1,\dots,n-1}$$ Will be evenly spread on either side of the line $\{z\in\mathbb C\mid \Re(z)=1/2\}$. Therefore, all of the imaginary parts cancel, and our sum is equivalent to summing $n-1$ copies of the number $1/2$. Hence the answer of $\frac{n-1}{2}$ .
- 12,536
-
To establish this rigorously, all you need to do is show that $$\cot(\pi k/n)=-\cot(\pi(n-k)/n)$$ Shouldn't be too difficult. HINT! – K.defaoite Mar 01 '23 at 15:14
-
Oh, this is really a nice idea, the map $z\to 1/(1-z)$ is a homographic / Möbius transformation, known to map "circles" (= circles and lines) to "circles", and we have to look only for the images of $-1$ and $\pm i$ to see which is the image of the unit circle! (Or see it geometrically as in inversion.) This gets my vote! – dan_fulea Mar 01 '23 at 15:27
-
-
@WindSoul Tag - Sequences and Series : "For questions concerning sequences and series. Typical questions concern, but are not limited to: identifying sequences, identifying terms, recurrence relations, ϵ−N proofs of convergence, convergence tests, finding closed forms for sums. For questions on finite sums, use the (summation) tag instead." In particular, finding closed forms for sums. – K.defaoite Mar 01 '23 at 16:43
-
To be fair, it should be the "summation" tag, I suppose. I'll change it. – K.defaoite Mar 01 '23 at 16:43
-
@K.defaoite, I wish I knew complex analysis to understand your answer which I find very elegant nonetheless. I think this is a high-school level problem- I could be wrong here and if it required complex analysis that should have been requested by OP. That we could solve problems with advanced math is great but I believe the simplest answer requires the exercise of logic using the least amount of knowledge. – WindSoul Mar 01 '23 at 16:54
Let $U$ be the set of all complex roots of unity $u$ of order $N$, all but $u=1$. We want to compute the sum over $1/(1-u)$ for $u\in U$. Consider for this the polynomial $P$: $$ \begin{aligned} P(X) &:=\frac{X^N-1}{X-1}=X^{N-1}+\dots+X+1=\prod_{u\in U}(X-u)\ . \\[3mm] &\qquad\text{ Then using the product rule}\\ &\qquad\text{ $(fg\dots h)'=f'g\dots h+fg'\dots h+\cdots+fg\dots h$ we get:}\\[3mm] P'(X)&=\sum_{u\in U}\frac{P(X)}{X-u}\ ,\\ P'(1)&=\sum_{u\in U}\frac{P(1)}{1-u}\ ,\qquad\text{ i.e.}\\ \frac{(N-1)N}2&=\sum_{u\in U}\frac N{1-u}\ , \end{aligned} $$ which is the claimed relation after dividing by $N$.
- 32,856
Let $$S_N=\sum^{N-1}_{k=1} \frac{1}{1-e^{i2\pi k/N}}=\frac{N-1}{2},$$ Use $$S=\sum_{k=k_1}^{k_2} f(k)=\sum_{k=k_1}^{k_2} f(k_1+k_2-k)\implies 2S= \sum_{k=k_1}^{k_2} [f(k)+f(k_1+k_2-k)]$$ Then we have $$2S_N=\sum_{k=1}^{N-1} \left(\frac{1}{1+e^{2ik\pi/N}}+\frac{1}{1+e^{2i(N-k)\pi}}\right)=\sum_{k=1}^{N-1} 1 \implies S_N=\frac{N-1}{2}. $$
- 632
-
The line after "use" could use a bit more explanation, but this is a nice answer. – K.defaoite Mar 01 '23 at 16:46
After working out the expression of the sum, it becomes $$\frac 12\sum_{k=1}^{N-1}{\left [1+i\cot \left(\frac kN \pi \right )\right ]}$$
The reason the imaginary terms cancel out is that since the pair $\cot(\frac kN \pi)=-\cot(\frac{N-k}{N}\pi)$, then
- For N odd number, there is an even number of terms terms that pair to cancel each other out
- For N even number, the only term that doesn’t pair to cancel is the middle term, which happens to be zero: $\cot(\frac {\frac N2}{N} \pi)=0$
- 2,160
$x_k=e^{2i\pi k/n}, k=0,...n-1$ are $n$ roots of $x^n=1$, Let us transform this equation by $y=\frac{1}{1-x} \implies x=\frac{y-1}{y}$. The transformed equation is $$(y-1)^n-(y)^n=0\implies -ny^{n-1}+\frac{n(n-1)}{2}y^{n-2}.....+(-1)^n=0.$$ $y_k,k=1,.....n$ are the roots of the $y-$ equation Hence, the sum $$S_n=\sum_{k=1}^{n-1} \frac{1}{1-e^{2ik\pi/n}}=\sum_{k=1}^{n-1} y_k=-\frac{n(n-1)}{-2n}=\frac{n-1}{2}.$$
- 43,235
-
Your first line makes no sense. The symbol $\implies$ is used to connect logical statements. The expression $(y-1)^n-(y)^n$ is not a logical statement. – K.defaoite Mar 01 '23 at 16:48
-