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.

In 1893 Hadamard proved that:

$$\xi(s) = \xi(0) \prod_{\rho} \left(1- \frac{s}{\rho} \right) \left(1- \frac{s}{1-\rho} \right)$$

where $\xi(z) = \frac12 z(z-1) \pi^{-\frac{z}{2}} \Gamma(\frac{z}{2}) \zeta(z)$ and $\rho = \sigma + \gamma i$ is a non-trivial zero of $\zeta(s)$ (i.e. $\gamma_n$ is the imaginary part of the n-th $\rho)$).

Riemann had already conjectured this in 1859 and since he needed an 'always real' function for his next thought steps, he assumed $s=\frac12 + t i$ and defined the function $\Xi(t)=\xi(s)$ that has been proven to be equal to:

$$\Xi(t)= \Xi(0)\prod_\gamma\left(1-\frac{t^2}{\gamma^2}\right)$$

It is known that $\xi(s)=\xi(1-s)$ and $\xi(s)=\overline{\xi(\overline{s})}$ and this implies that $\xi(s)\xi(\overline{s})$ must be real. Also known is that when there is a non-trivial zero $\rho$ lying off the critical line, then also $1-\rho, \overline{\rho}, \overline{1-\rho}$ must be zeros and their product will always be real:

$$\displaystyle\prod_\gamma\left(1-\frac{s}{\sigma+i \gamma}\right) \left(1-\frac{s}{\sigma-i \gamma}\right)\left(1-\frac{s}{1-(\sigma+i \gamma)}\right) \left(1-\frac{s}{\overline{1-(\sigma+i \gamma)}}\right)$$

This brought me to the following conjecture. Suppose $s=a+t i$ and $\Xi_a(t)=\xi(s)$ so $\Xi_a(0)=\xi(a)$, then the following product holds and is always real:

$$\Xi_a(t)\Xi_a(-t) = \Xi_a(0)^2\prod_\gamma \left(1-\frac{t^2}{\gamma^2}\right)\left(1-\frac{(-t)^2}{\gamma^2}\right)$$

or simpler:

$$\Xi_a(t)\Xi_a(-t) = \left(\xi(a)\prod_\gamma \left(1-\frac{t^2}{\gamma^2}\right)\right)^2$$

Numerical tests indicate that the conjecture is correct, but I am obvioulsy keen to find a proof (I do realise this is a way too big of a question to ask here!). Does anybody have a link that explains how $\Xi(t)= \Xi(0)\prod_\gamma\left(1-\frac{t^2}{\gamma^2}\right)$ has been derived? Also appreciate any other thoughts/steers on how to further progress this conjecture.


share|improve this question
This seems to be fit way more for overflow than for SE, though I'm happy to see it here. –  DonAntonio Jan 6 '13 at 15:52
Mangoldt has $\xi(t)= \xi(0)\cdot \prod(1 -t^2/\alpha_{\nu^2}).$ Are you saying $\Xi$ is some other function? –  daniel Jan 6 '13 at 16:25
Your $\gamma$ is Mangoldt's $\alpha.$ He cites Hadamard for this relation so it must be in one of Hadamard's better-known papers? –  daniel Jan 6 '13 at 16:41
Indeed Daniel. I guess that's the formula I am after. Keen to find the proof of it in Hadamard's papers! P.S. There is quite a bit of confusion on the use upper and lower case xi. Riemann confusingly used $\xi(t)$ for $\Xi(t)$ where the latter has become the convention. –  Agno Jan 6 '13 at 16:45
add comment

2 Answers

up vote 2 down vote accepted

Hoping to redeem my previous incorrect answer, I offer the following. In an article about Riemann's paper in which he discusses the problem of Riemann mentioned in Edwards' book, Fresan gives your relation on page 26 and says:

"La première preuve rigoureuse de cette identité est dûe à Hadamard, qui l’inclut dans son article “Étude sur les propriétés des fonctions entières et en particulier d’une fonction considérée par Riemann” (1893)." [Full cite at linked PDF.] ["The first rigorous proof of this identity is due to Hadamard, who includes it in his article...etc."]

Note that he has retained Riemann's notation, which employs $\xi$ for $\Xi.$

So this would be the paper of Hadamard in which the proof appears.

And finally here is a link to Hadamard's paper, which I have not read.

Glancing at the paper of Fresan again, I wonder if his calculation on page 26 is not responsive to your underlying question?

share|improve this answer
Daniel. I have Edwards' book on my lap and open at page 20. It only shows the product formula for $\xi(s)$ i.e. the lower case version. I am keen to find out how the the product for the $\Xi(t)$, i.e. upper case version assuming $s=\frac12 +t i$, has been derived. –  Agno Jan 6 '13 at 16:10
Merci bien, Daniel :-) Dusting off my French. You are right. P26 seems to outline the trick for $s=\frac12+ti$! Now need to check whether the same logic works for $P(a+ti)$ that is equivalent to my $\xi(a+s i)$. –  Agno Jan 6 '13 at 18:38
@user46149: Appreciate the up-vote. You should probably wait a bit before accepting because someone may provide a more authoritative answer. I am not an expert but there are many on this site. –  daniel Jan 6 '13 at 19:05
Daniel. Yes! This indeed does the trick. Many thanks. I can now prove that when I f.i. assume that all zeros in the infinite product of $\left(\frac{1}{\rho}\right)\left(\frac{1}{\overline{\rho}}\right)$ lie on the line $\rho=\frac13$, that the formula for $\Xi(t)_\frac13$ still perfectly holds :-) Landau's Handbuch is also correct that $\Xi_\frac12(z) = \Xi_\frac12(-z)$ but this is not true for other values of $a$ ! –  Agno Jan 6 '13 at 19:09
Much appreciated your help :-) –  Agno Jan 6 '13 at 19:11
add comment

For completeness' sake and to avoid confusion, I can now confirm that the conjecture (even though the numerical results are very close) must be false. It is easy to see that when:


would be independent of $a$, then it could be expressed as:


This function has the same zeros as $\zeta(s)$ and therefore its square can not be equal to $\dfrac{\Xi_a(t)\Xi_a(-t)}{\xi(a)^2}$ since the latter doesn't have zeros at those spots (otherwise the RH would also have been immediately proven false).

Back to the drawing board.

share|improve this answer
add comment

Your Answer


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.