2

The problem is to show that for every $k \in \{1,2,...,n\}$ if we sum $\frac{z^{k+1}}{(1-z)^2}$ over the roots $z$ of $x^n+1=0$, we obtain the result $\frac{n}{2}(k-\frac{n}{2})$.

One idea I had is to consider the polynomial $P(t)=(1-t)^n+1$. We can easily calculate the sum of roots of $P$ and the sum of their squares. By considering the polynomial $Q(t)=t^nP(1/t)$ we can easily compute the the sum of the reciprocals and sum of the squares of the reciprocals. Hence for $k=0,1,2$ we can compute the sum in question quite easily and we get the intended result. However this method doesn't really generalise because the formulas for the symmetric sums lead to really messy computations.

math_lover
  • 5,826
  • 2
    What have you tried so far? Do you know any representations of the roots of $x^n+1$? If not, check out roots of unity. – CodeLabMaster Jan 22 '17 at 22:23
  • I can prove it for k=0,1,2 by constructing a polynomial whose roots are the terms in the sum and then applying vieta. But in the general case the same method doesn't really work. We can also transform the problem into a trig identity but it doesn't help much. – math_lover Jan 22 '17 at 23:27
  • Do post your efforts, it looks like this one will be closed soon. – Marko Riedel Jan 22 '17 at 23:30

2 Answers2

4

We have with

$$f(z) = \frac{z^{k+1}}{(1-z)^2} \frac{nz^{n-1}}{z^n+1} = \frac{1}{(z-1)^2} \frac{nz^{n+k}}{z^n+1} $$

and $\zeta_{n,q} = \exp(\pi i/n+ 2\pi i q/n)$

$$\sum_{x^n+1=0} \frac{x^{k+1}}{(1-x)^2} = \sum_{q=0}^{n-1} \mathrm{Res}_{\large z=\zeta_{n,q}} f(z).$$

This is also given by (residues sum to zero)

$$-\mathrm{Res}_{z=1} f(z) -\mathrm{Res}_{z=\infty} f(z).$$

For the first of these we differentiate to obtain

$$\frac{n(n+k) z^{n+k-1}}{z^n+1} - \frac{nz^{n+k}}{(z^n+1)^2} n z^{n-1}$$

Evaluate at $z=1$ to get

$$\frac{1}{2} n(n+k) - \frac{1}{4} n^2 = \frac{1}{2} nk + \frac{1}{4} n^2.$$

For the residue at infinity we get

$$-\mathrm{Res}_{z=0} \frac{1}{z^2} \frac{1}{(1/z-1)^2} \frac{n/z^{n+k}}{1/z^n+1} \\ = -\mathrm{Res}_{z=0} \frac{1}{(1-z)^2} \frac{n}{z^{n+k}+z^k} \\ = -\mathrm{Res}_{z=0} \frac{1}{z^k} \frac{1}{(1-z)^2} \frac{n}{z^{n}+1}$$

This is

$$-[z^{k-1}] \frac{1}{(1-z)^2} \frac{n}{z^{n}+1}.$$

With $k\le n$ only the constant term from $\frac{n}{z^{n}+1}$ contributes and we get

$$-[z^{k-1}] \frac{n}{(1-z)^2} = -nk.$$

Adding the contributions from the residues at one and at infinity and flipping the sign we thus obtain

$$-\left(\frac{1}{2} nk + \frac{1}{4} n^2 - nk\right) = -\left(-\frac{1}{2} nk + \frac{1}{4} n^2\right) = -\frac{1}{2} n \left(\frac{1}{2}n - k\right).$$

This is

$$\bbox[5px,border:2px solid #00A000]{ \frac{1}{2} n \left(k - \frac{1}{2} n\right)}$$

as claimed.

Marko Riedel
  • 61,317
2

If we use that the summation over the roots of unity stays invariant under replacement of $z$ by $z^{-1}$ we can avoid having to consider the residue at infinity. Since:

$$\frac{z^{-k-1}}{\left(\frac{1}{z}-1\right)^2} = \frac{z^{-k+1}}{\left(z-1\right)^2}$$

we can replace $k$ by $-k$ without that affecting the summation. This means that the contour integral:

$$\frac{1}{2\pi i}\oint_C n\frac{z^{n-1}}{z^n+1}\frac{z^{-k+1}}{(z-1)^2}dz$$

for a counterclockwise contour encircling the unit circle yields the desired summation plus the residue at $z = 1$, and for $k\geq 0 $ the integral tends to zero if the radius of the contour tends to infinity. Therefore, all we need to do is evaluate minus the residue of the integrand at $z = 1$. This amounts to expanding the function

$$n\frac{z^{n-k}}{z^n+1}$$

around $z = 1$ and computing the first order term. If we put $z = 1+t$ and expand in powers of $t$, we obtain:

$$\frac{n}{2}\left[1+\left(\frac{n}{2}-k\right)t+\mathcal{O}(t^2)\right]$$

The summation equals minus the residue which is minus the coefficient of $t$, this is thus equal to:

$$\frac{n}{2}\left(k-\frac{n}{2}\right)$$

For $k<0$ we can proceed as we did above but without changing the sign of $k$, so we then end up with the same result, except that the sign of $k$ is changed, therefore the general formula for the summation is:

$$\frac{n}{2}\left(|k|-\frac{n}{2}\right)$$

Count Iblis
  • 10,366
  • I will check this some more but it looks good so far. Upvoted. Unfortunately this question appears on its way into limbo, I do not quite understand why. – Marko Riedel Jan 23 '17 at 00:44
  • @MarkoRiedel : is there a way to compute the answer using more elementary tools, i.e without complex analysis stuff. – math_lover Jan 23 '17 at 17:19