5
$\begingroup$

Building on this question scaling the imaginary part of $\rho$s in infinite products, I like to conjecture that:

$$\displaystyle \prod_{n=1}^\infty \left(1- \frac{s}{\mu_n} \right) \left(1- \frac{s}{1-\mu_n} \right)\left(1- \frac{s}{\overline{\mu_n}} \right) \left(1- \frac{s}{\overline{1-\mu_n}} \right)$$

with $\mu_n = a + \Im(\rho_n)x i$ and $a,x \in \mathbb{R},x \ne 0, s \in \mathbb{C}$ and $\rho_n$ the n-th non-trivial zero of $\zeta(z)$,

has the following closed form:

$$\displaystyle H(s,a,x):= \frac{\xi(\frac12 - \frac{a}{x} + \frac{s}{x})}{\xi(\frac12 - \frac{a}{x})} \frac{\xi(\frac12 - \frac{a}{x} + \frac{1}{x} - \frac{s}{x})}{\xi(\frac12 - \frac{a}{x}+ \frac{1}{x})}$$

where $\xi(z) = \frac12 z(z-1) \pi^{-\frac{z}{2}} \Gamma(\frac{z}{2}) \zeta(z)$ is the Riemann xi-function.

If this formula is correct, the 'constructed' zeros $\mu_n$ can be stretched/condensed vertically via $x$ on the imaginary axis and shifted left/right on the real line via $a$. In all cases they would yield an entire function expressed by this closed form (think of it as a reversed application of the Weierstrass factorization theorem, i.e. starting with products of 'constructed' zeros).

Further factorization also seems possible with:

$$\frac{\xi(\frac12 - \frac{a}{x} + \frac{s}{x})}{\xi(\frac12 - \frac{a}{x})} = \prod_{n=1}^\infty \left(1- \frac{s}{\mu_n} \right) \left(1-\frac{s}{\overline{\mu_n}} \right)$$

and

$$\frac{\xi(\frac12 - \frac{a}{x} + \frac{1}{x} - \frac{s}{x})}{\xi(\frac12 - \frac{a}{x}+ \frac{1}{x})} = \prod_{n=1}^\infty \left(1- \frac{s}{1-\mu_n} \right) \left(1- \frac{s}{\overline{1-\mu_n}} \right)$$

When $a=\frac12$ and $x=1$, the formula correctly reduces to:

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

from which the known Hadamard product for $\zeta(s)$ can been derived.

Unfortunately I do not have a proof for this formula, however rigorously checked it against many 'brute force' calculations using the first 2mln $\rho$s (all correct results, but accurate up to 5 decimals max). I manufactured the formula by replicating the symmetry of the closed form for $\mu_n = a + n x i$ (i.e. running through the integers rather than $\Im(\rho_n)$, see the linked question). Since until today, all non-trivial zeros appear to be lying on the critical line, I have used $\frac12$ as the "source" for all zeros for different $a$'s i.e.: $\frac12 - \frac{a}{x} + \frac{s}{x}$ just inserts $\frac12$ when $\Re(s)=a$. I guess I have thereby implicitly assumed the RH in constructing the formula.

My questions:

  1. Is this a known closed form?

  2. Does a proof of this closed form imply the RH, i.e. does it "force" the Hadamard product into a "straight jacket" that only allows it to be valid when all $a=\Re(\rho_n)=\frac12$ ?

UPDATE:

Assuming RH is true, I believe that I have found a nice proof for the equation in the OP. Since "the comments section is too small to contain it", I decided to put it as an answer to my own question to round it up.

$\endgroup$
4
  • $\begingroup$ Purported proofs of the RH are off topic. I voted to close. $\endgroup$ Jan 3, 2013 at 0:33
  • 5
    $\begingroup$ @Andy Putman: I don't think the OP claimed to have proved RH. His question is whether a particular statement is equivalent to RH. $\endgroup$ Jan 3, 2013 at 8:34
  • $\begingroup$ @Agno Your product is difficult to understand. For each zero $1/2+i\gamma$ (I assume that with $\gamma>0$) you add four zeros $\mu$, $1-\mu$, $\overline{\mu}$ and $\overline{1-\mu}$ to your product. But if RH is not true, and there are two zeros $\beta+i\gamma$ and $1-\beta+i\gamma$ (with $\beta > 1/2$ and $\gamma>0$) How many zeros have your product (associated with these two zeros) 4 or 8? Whatever the interpretation, I think that if RH is not true, your formula is not correct. $\endgroup$
    – juan
    Jan 3, 2013 at 12:07
  • $\begingroup$ @Juan. I believe each $H(s,a,x)$ generates a unique zero when $(s-a)=\frac12+i\gamma$ via the $\zeta$s in either $\xi(\frac12 - \frac{a}{x} + \frac{s}{x})$ or $\xi(\frac12 - \frac{a}{x} + \frac{1}{x} - \frac{s}{x})$. This also implies that when $H(s,a,x)=0$, then also $H(s,1-a,x)=H(1-s,a,x)=H(1-s,1-a,x)=0$. The values correspond to $s=\mu, s=\overline{1-\mu},1-s=1-\mu, 1-s=\overline{\mu}$ respectively. Also when $a=\frac12$ there will be 4 zeros associated with it, of which two pairs are equal. $\endgroup$
    – Agno
    Jan 3, 2013 at 16:15

3 Answers 3

6
$\begingroup$

The proposed formula is not true if RH is not true.

Let $\Theta$ be the upper bound of the real parts of the zeros of $\zeta(s)$.

Your product has zeros at $\mu$, $1-\mu$, $\overline{\mu}$ and $\overline{1-\mu}$ with $\mu=a+\gamma x i$ where $\rho=\beta+i\gamma$ runs through the non trivial zeros of $\zeta(s)$ so that the supremum of the real parts of the zeros of this product is $\sup(a,1-a)$.

But your function $$H(s,a,x):=\frac{\xi\left(\frac12-\frac{a}{x}+\frac{s}{x}\right)} {\xi\left(\frac12-\frac{a}{x}\right)}\frac{\xi\left(\frac12-\frac{a}{x}+\frac{1-s}{x}\right)}{\xi\left(\frac12-\frac{a}{x}+\frac{1}{x}\right)}$$ has zeros at $\beta x-\frac{x}{2}+a+i\gamma x$ and $1-\beta x+\frac{x}{2}-a-i\gamma x$.

Therefore assuming $x>0$ the supremum of the zeros of $H(a,s,x)$ is the greater of $\Theta x-\frac{x}{2}+a$ and $1-(1-\Theta)x+\frac{x}{2}-a=1+\Theta x-\frac{x}{2}-a$, Or $\Theta x-\frac{x}{2}+\sup(a,1-a)$

Your equality, for $x>0$, would imply therefore $$\Theta x-\frac{x}{2}+\sup(a,1-a)=\sup(a,1-a).$$ That is $\Theta x-\frac{x}{2}=0$, and therefore $\Theta=1/2$. This is RH.

The case $x<0$ is similar.

$\endgroup$
2
  • $\begingroup$ @Juan. Many thanks! The outcome is also in line with my expectations :-) Now that there is evidence that "The proposed formula is not true if RH is not true", a logical follow up question would be: Can the formula actually be derived assuming RH (i.e. similar to what you did prove for $x$ in the previous question)? (\sloppy math mode on) Have again tested the formula a lot today against the brute force calculations with 2mln $\rho$s and I believe it just works too well to not be correct.(\sloppy math mode off). $\endgroup$
    – Agno
    Jan 3, 2013 at 16:56
  • $\begingroup$ Have now added the proof that when assuming RH, the formula can be logically derived. $\endgroup$
    – Agno
    Jan 14, 2013 at 6:21
0
$\begingroup$

The conjecture above:

$H(s,a,x) =\prod_{n=1}^\infty \left(1- \frac{s}{\mu_n} \right) \left(1- \frac{s}{1-\mu_n} \right)\left(1- \frac{s}{\overline{\mu_n}} \right) \left(1- \frac{s}{\overline{1-\mu_n}} \right)= \frac{\xi(\frac12 - \frac{a}{x} + \frac{s}{x})}{\xi(\frac12 - \frac{a}{x})} \frac{\xi(\frac12 - \frac{a}{x} + \frac{1}{x} - \frac{s}{x})}{\xi(\frac12 - \frac{a}{x}+ \frac{1}{x})}$

with $\mu_n = a + \Im(\rho_n)x i$ might be expanded even further.

When the domains for $a$ and $x$ are extended to $\mathbb{C}_{/(x=0)}$, the formula for $H(s,a,x)$ still appears to work (tested at 5 decimals accuracy) and for instance induces the following result (assume $\gamma_n = \Im(\rho_n)$):

$H(s,0,i) =\prod_{n=1}^\infty \left(1 + \frac{s}{\gamma_n} \right) \left(1- \frac{s}{\gamma_n} \right)\left(1- \frac{s}{{1+\gamma_n}} \right) \left(1- \frac{s}{{1-\gamma_n}} \right)= \frac{\xi(\frac12 + \frac{s}{i})}{\xi(\frac12)} \frac{\xi(\frac12 + \frac{1}{i} - \frac{s}{i})}{\xi(\frac12 + \frac{1}{i})}$

i.e. all imaginary parts of $\rho_n$ become real and their infinite product still has a closed form that, as expected, induces a zero at e.g. $H(\gamma_n,0,i)$.

Or, when we take one of the two sub-factors from the OP, this gives:

$$\displaystyle \xi\left(\frac12+\frac{s}{i}\right) = \xi\left(\frac12 \right)\prod_{n=1}^\infty \left(1 - \frac{s^2}{\gamma_n^2} \right)$$

and for $s=i$, this result emerges:

$$\displaystyle \xi\left(\frac32\right) = \xi\left(\frac12 \right)\prod_{n=1}^\infty \left(1 + \frac{1}{\gamma_n^2} \right)$$

and $s=\frac{-i}{2}$ or $s=\frac{i}{2}$ yields this quite interesting outcome:

$$\displaystyle \xi\left(\frac12 \right) = \frac12 \prod_{n=1}^\infty \left(\frac{1}{1 + \frac{1}{(2 \gamma_n)^2} }\right) = \Xi(0)$$

So that:

$$\Xi\left(\frac12\right) = \frac12 \prod_{n=1}^\infty \left(\frac{(2 \gamma_n)^2 -1} {(2 \gamma_n)^2 +1}\right)$$

Here's my 'Overflow' follow-up question: I have tested the conjecture in Maple up to $\gamma_{2.000.000}$, but would really appreciate it when somebody could replicate and confirm the outcomes (I do not possess Mathematica, but do know the $\rho$s are already coming with the package so not much 'programming' should be required).

Many thanks!

$\endgroup$
0
$\begingroup$

Below is a proposed proof, that assuming the RH, the following equation is true:

$$\displaystyle \frac{\xi(\frac12 - \frac{a}{x} + \frac{s}{x})}{\xi(\frac12 - \frac{a}{x})} = \prod_{n=1}^\infty \left(1- \frac{s}{\mu_n} \right) \left(1- \frac{s}{\overline{\mu_n}} \right)$$

where $\mu_n = a + i x \gamma_n$ and $\gamma_n = \Im(\rho_n)$, with $\rho_n$ the n-th non-trivial zero of $\zeta(s)$.

Take $t, \gamma_n \in \mathbb{R},a,x,s \in \mathbb{C}$ with $x \ne 0$ and $\gamma_n > 0$.

Starting from Hadamard's proof that:

$$\xi(s) = \xi(0) \prod_{n=1}^\infty \left(1- \frac{s}{\rho_n} \right) \left(1- \frac{s}{1-\rho_n} \right) \qquad (1)$$

with $\xi(s) = \frac12 s(s-1) \pi^{-\frac{s}{2}} \Gamma(\frac{s}{2}) \zeta(s)$ being the Riemann xi-function.

Now we assume the RH and all $\Re(\rho_n)=\frac12$. Take $s=\dfrac{a+ i t}{x}$.

This gives:

$$\xi\left(\dfrac{a+ i t}{x}\right) = \xi(0) \prod_{n=1}^\infty \left(1- \frac{\frac{a}{x} +\frac{i t}{x}}{\frac12 + i \gamma_n} \right) \left(1- \frac{\frac{a}{x} +\frac{i t}{x}}{\frac12 - i \gamma_n} \right)$$

and can be expanded into:

$$\xi\left(\dfrac{a+ i t}{x}\right) = \xi(0) \prod_{n=1}^\infty \left(\frac{\frac12 - \frac{a}{x} + i \gamma_n -\frac{i t}{x}}{\frac12 + i \gamma_n} \right) \left(\frac{\frac12 - \frac{a}{x} - i \gamma_n - \frac{i t}{x}}{\frac12 - i \gamma_n} \right)$$

By multiplying each factor with $\dfrac{(\frac{a}{x} + i \gamma_n)}{(\frac{a}{x} + i \gamma_n)}$ and $\dfrac{(\frac{a}{x} - i \gamma_n)}{(\frac{a}{x} - i \gamma_n)}$ respectively, we get:

$$\displaystyle \xi\left(\dfrac{a+ i t}{x}\right) = \xi(0) \prod_{n=1}^\infty \left(\frac{\frac{a}{x} + i \gamma_n}{\frac12 + i \gamma_n} \right) \left(\frac{\frac{a}{x} - i \gamma_n}{\frac12 - i \gamma_n} \right) \prod_{n=1}^\infty \left(\frac{\frac12 - \frac{a}{x} + i \gamma_n -\frac{i t}{x}}{\frac{a}{x} + i \gamma_n} \right) \left(\frac{\frac12 - \frac{a}{x} - i \gamma_n - \frac{i t}{x}}{\frac{a}{x} - i \gamma_n} \right) $$

$$\displaystyle = \xi(0) \prod_{n=1}^\infty \left(1- \frac{\frac12 -\frac{a}{x}}{\frac12 + i \gamma_n)} \right) \left(1- \frac{\frac12-\frac{a}{x}}{\frac12 - i \gamma_n} \right) \prod_{n=1}^\infty \left(1- \frac{\frac{2a}{x}-\frac12+\frac{i t}{x}}{\frac{a}{x} + i \gamma_n)} \right) \left(1- \frac{\frac{2a}{x}-\frac12+\frac{i t}{x}}{\frac{a}{x}- i \gamma_n} \right)$$

By now injecting $\dfrac12 - \dfrac{a}{x}$ into equation (1), this can be simplified as:

$$\xi\left(\dfrac{a+ i t}{x}\right) = \xi\left(\frac12 - \frac{a}{x}\right) \prod_{n=1}^\infty \left(1- \frac{\frac{2a}{x}-\frac12+\frac{i t}{x}}{\frac{a}{x} + i \gamma_n}\right) \left(1- \frac{\frac{2a}{x}-\frac12+\frac{i t}{x}}{\frac{a}{x}- i \gamma_n} \right)$$

and since $\dfrac{2a}{x} + \dfrac{i t}{x} = s + \dfrac{a}{x}$ this can be rewritten into:

$$\xi\left(\dfrac{a+ i t}{x}\right) = \xi\left(\frac12 - \frac{a}{x}\right) \prod_{n=1}^\infty \left(1- \frac{x(s+\frac{a}{x}-\frac12)}{a+ i x \gamma_n}\right) \left(1- \frac{x (s+\frac{a}{x}-\frac12)}{a- i x \gamma_n} \right)$$

so that we can now say that:

$$\xi\left(\frac12 - \frac{a}{x} + \frac{s}{x}\right) = \xi\left(\frac12 - \frac{a}{x}\right) \prod_{n=1}^\infty \left(1- \frac{s}{a+ i x \gamma_n} \right) \left(1- \frac{s}{a- i x \gamma_n} \right)$$

and the desired result is obtained:

$$\displaystyle \frac{\xi(\frac12 - \frac{a}{x} + \frac{s}{x})}{\xi(\frac12 - \frac{a}{x})} = \prod_{n=1}^\infty \left(1- \frac{s}{\mu_n} \right) \left(1- \frac{s}{\overline{\mu_n}} \right)$$

When starting from $s = \dfrac{1-(a+i t)}{x}$ the equivalent result is $\displaystyle \prod_{n=1}^\infty \left(1- \frac{s}{1- \mu_n} \right) \left(1- \frac{s}{\overline{1-\mu_n}} \right)$

$\endgroup$
1
  • $\begingroup$ To be complete I should have added the constraint: $a \pm ix\gamma_n \ne 0$ $\endgroup$
    – Agno
    Jan 14, 2013 at 6:16

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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