Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

Are the Lerch-$\zeta(\varphi,0,-n) $ of integer n (for shortness I use the notation of my earlier question $\small \zeta_\varphi(-n)$) periodically purely real and imaginary: $\zeta_\varphi (-n)^2 $ is real, ($ n \ge 2$) ? And how to prove this?

I've a strange observation of periodicity which I would like to explain/derive/prove analytically.


Consider the triangle of eulerian numbers E (ideally of infinite size, row and col-indices beginning at zero, with elements $ \small e_{r,c}$)

$ \qquad E = \small \begin{array} {rrrrr} 1 & . & . & . & . & . \\ 1 & 0 & . & . & . & . \\ 1 & 1 & 0 & . & . & . \\ 1 & 4 & 1 & 0 & . & . \\ 1 & 11 & 11 & 1 & 0 & . \\ 1 & 26 & 66 & 26 & 1 & 0 \end{array} $

Assume some angular parameter $\varphi$, the associated complex number from the unit-circle $z=z_\varphi= \exp(i \varphi ) \qquad z \ne 1 $, $ \small {\varphi \over \pi} $ not necessarily rational.

Now consider the Eulerian polynomials, whose coefficients are taken from a row of E, $ \small E(z,r) = \sum_{c=0}^\infty ( e_{r,c} \cdot z^c ) $

Then compute $ \small \zeta_\varphi(-n) = {z \over (1-z)} \cdot E(z,n) \cdot (1-z)^{-n} $ .

Observation: I observe, that for $n>2$ the $ \small \zeta_\varphi(-n) $ lie either on the real or on the imaginary axis, in other words $ \small \zeta_\varphi(-n) ^2 $ are real..

Q: I seem to be missing just the key idea, so my question here is, how I could derive that behave analytically, given the description via the Eulerian triangle.

Addendum/Generalization: While it was already surprising, that this works with rational roots z of the complex unit, it seems to me even more interesting, that the observation holds for arbitrary $ \small \varphi \ne 0 $ and $ \small z_\varphi = \exp(I \cdot \varphi) \ne 1 $ .

share|improve this question
    
Using your notations, I am getting $\zeta_\phi(-3)^2 = (z^2+4z+1)^2/(1-z)^8$. For $z=e^{i \pi/8}$, that value is $(1-i) \xi$, where $\xi \approx 45,013.022$, hence $\zeta_\phi(-3)^2$ is not real –  Sasha Jul 26 '11 at 19:28
    
@Sasha: thanks for the crosscheck; I've reproduced your result. You missed that leading $z$ in the formula; you needed to write $ \small \ldots = z^2 \cdot (z^2 +4z + 1)^2/(1-z)^8 $ –  Gottfried Helms Jul 26 '11 at 19:42
    
Thanks, in this case you deal with $\zeta(\varphi, -n, 0)$ in wikipedia's convention, I used $a=1$ as stated in the title. –  Sasha Jul 26 '11 at 19:51
    
I see; I'll update the title. Unfortunately, in wikipedia the formula begins then with a division by zero, but this might go as minor notation problem then... –  Gottfried Helms Jul 26 '11 at 19:59
    
given that $s = -n$, the first term is $z^0*0^n$ which is perfectly defined in the case at hand. –  Sasha Jul 26 '11 at 20:13
show 1 more comment

2 Answers

up vote 3 down vote accepted

By the symmetry of the triangle of Eulerian numbers the rational function $$R_{n}(z) = {z \over (1-z)} \cdot E(z,n) \cdot (1-z)^{-n} $$ satisfies$$R_{n}(z)=(-1)^{n+1}R_{n}(\frac{1}{z}).$$ Since the numerator and denominator polynomials of $R_{n}(z)$ have real (in fact, integer) coefficients we have$$\overline{R_{n}(z)}=R_{n}(\bar z).$$ Your observation should now be explained (there is no need to restrict to $n>2$).

To clarify further:

Let $ z = \exp(i \cdot \varphi) \ne 1 $ lie on the unit circle so $\bar z = \frac {1}{ z}$ and let $n \geq 1$. Suppose $R_{n}(z) = a+ib$.

Then from the above$$\overline{R_{n}(z)}=R_{n}(\bar z) = R_{n}(\frac{1}{z})=(-1)^{n+1}R_{n}(z). $$ Hence $$a-ib = (-1)^{n+1}(a+ib).$$ If $n$ is even we get $a = 0,$ and so $R_{n}(z)$ is pure imaginary; if $n$ is odd we get $b= 0,$ and $R_{n}(z)$ is real.

A particularly interesting case is when $z = i$. The sequence $(2 \cdot R_{n}(i))n \geq 1$ equals $[-1, -1 \cdot i, 2, 5 \cdot i, -16, -61 \cdot i,...]$, which is the sequence of up-down numbers multiplied by powers of $i$. For more on this see my comment of Jan 2011 in A000111 in the OEIS.

share|improve this answer
    
thanks! For $R_n(z)$ n=1 this can even be seen when z is explicitely written as $z=\alpha + \beta i$ and then multiplied out (exploiting $ \alpha^2 + \beta^2 = 1$ ) - the imaginary part vanishes then. But we still cannot allow n=0 ; we have always $-0.5 + x i$ which is not on the real or imaginary axis. –  Gottfried Helms Jul 27 '11 at 7:00
    
Peter, what you define as $R_{n}(z)$ is $Li_{-n}(z)$ according to the last formula in link. –  Bruce Arnold Jul 27 '11 at 21:30
    
Nice. Or even simpler Table[I^(n+1)*2*PolyLog[-n,-I],{n,1,11}]. –  Bruce Arnold Jul 28 '11 at 22:14
    
@Peter: again thanks; that clarification is nice and concise, so I think, I'd accept that answer. Btw, in my previous question I showed that same sequence (for $ \varphi=\pi/2$ ) only I did not yet expect, that it would be better to let the series begin with $z^1$ at the first term. (But I think it makes no sense to adapt that old question&answer) –  Gottfried Helms Jul 29 '11 at 8:36
add comment

Unfortunately, I couldn't make Peter's answer fully transparent to me. So I tried two things; both led to a solution.
1) rewrite the $ \small \zeta _\varphi(-n) $ - expression based on the description via the Eulerian polynomials as taylor-series in terms of $ \small \varphi$, which is real. We can factor out $ \small i^n$ and stay with a series in $ \small \varphi ^2$, thus the result is $ \small i^n \cdot x$ where x is real.

2) Rewrite the expression using the Eulerian polynomials using $ \small z=exp(i \varphi)$
in $ \small \zeta _\varphi(-n) = {z \over 1-z} E(z,n) \cdot (1-z)^{-n} $

Factor out $ \small {z \over (1-z)^2} $ to get as first cofactor
$\qquad \small {z \over (1-z)^2} = {1 \over (1/\sqrt{z} - \sqrt z )^2} ={1 \over (\exp(- i \varphi/2) - \exp(i \varphi /2))^2 } = {1 \over (-2 i\sin(\varphi /2))^2 } = -1/4 \cdot \sin(\varphi /2)^{-2} $.
This value is always real, and is also $ \zeta_\varphi(-1) $.

Then , for instance, for n=3 we get for the remaining part of the formula:

$ \qquad\small \begin{array} {ll} (1 + 4z + 1z^2) \cdot (1-z)^{-2} &=& (1/z + 4 + 1z) \cdot (1/\sqrt{z} - \sqrt z )^{-2} \\\ & = &(1 \cdot (1/z + z) + 4 ) \cdot -1/4 \cdot \sin(\varphi /2)^{-2} \\\ & = & - ( {1 \over 2} \cos(\varphi)) + 1)\cdot \sin(\varphi /2)^{-2} \\\ \end{array}$

Together this gives
$ \qquad \small \zeta_\varphi(-3)= 1/4 \cdot \sin(\varphi /2)^{-4} \cdot ( {1 \over 2} \cos(\varphi)) + 1) $

Well, this is only for one n, but I begin to understand, how this works and why we get these purely imaginary/real results (and I even begin to see Peter's mentioned symmetries here.... )

share|improve this answer
add comment

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.