2
$\begingroup$

Let $k \ge 1,B \ge 2$ integers and $\zeta^{(k)}(s)$ denote the $k$-th derivative of Riemann zeta function. For real $x$, let $[x]$ denote the nearest integer to $x$.

Conjecture 1: For all $n \ge 1,[\zeta^{(n)}(1/2)]= -2^{n+1} n!$

Conjecture 2: For all $k \ge 1,B \ge 2$ we have $[\zeta^{(k)}(1-1/B)]= -B^{k+1}\cdot k!$ and $\zeta^{(k)}(1-1/B)$ is very close to integer.

sagemath code per request:

def zetania2(LK=10,prec=100,bb=[3,5]):
    """
    zeta^(k)(1-1/B), near to integers
    """
    import mpmath
    mpmath.mp.pretty=1
    mpmath.mp.dps=prec
    pre2=10
    for B in bb:
        l0=[]
        lerr=[]
        for k in range(1,LK+1):
            T=mpmath.zeta(1-1/mpmath.mpf(B),derivative=k)
            N=mpmath.nint(T)
            N2= -B**(k+1) * factorial(k)
            l0 += [N-N2]
        print(B,l0)
$\endgroup$
5
  • $\begingroup$ Could you please provide some numerical evidence? $\endgroup$ Sep 25, 2021 at 15:54
  • $\begingroup$ @SylvainJULIEN Added sagemath code. I believe you can run it in a browser. $\endgroup$
    – joro
    Sep 25, 2021 at 17:01
  • $\begingroup$ @SylvainJULIEN Can you check the conjectures with a CAS different from sage/mpmath? Answer claims they are false. $\endgroup$
    – joro
    Sep 25, 2021 at 18:02
  • $\begingroup$ I'm not well versed in using such software. And to be honest, as your conjectures are disproven, I don't quite see the point of checking them. $\endgroup$ Sep 25, 2021 at 18:29
  • $\begingroup$ @MarkSapir The questions are entirely different: one is true, one is false. $\endgroup$
    – joro
    Sep 27, 2021 at 6:24

1 Answer 1

2
$\begingroup$

Both conjectures are false. Suppose your conjectured identity holds for all $k\geq 1$ and some $B\geq 2$. It is known that $$ f(s)=\zeta(s)-\frac{1}{s-1} $$ is an entire function. Your identity for fixed $B$ implies, for instance, that $|f^{(k)}(1-1/B)|\leq 1$. By Taylor expansion, from this we get for all $s\in \mathbb C$ that $$ |f(s)|=\left|\sum_{k=0}^{+\infty}\frac{f^{(k)}(1-1/B)}{k!}(s-1+1/B)^k\right|\leq |f(1/B)|+e^{|s-1+1/B|}. $$ The last bound cannot hold, because $f(-n)$ grows more than exponentially for odd integer $n\to+\infty$.

UPDATE: as for the numerical evidence, $k=41$ seems to be the least value of $k$ with $$ |\zeta^{(k)}(1-1/B)+k!B^{k+1}|>1 $$ for $B=2,3,4$. This was computed using the function $\it{derivnum}$ in Pari/GP.

$\endgroup$
9
  • $\begingroup$ Thanks. I get experimental support for the conjectures, possibly due to a bug in the libraries I am using. What would be a small counterexample or set of counterexamples to check numerically? $\endgroup$
    – joro
    Sep 25, 2021 at 17:59
  • $\begingroup$ The derivatives of $\zeta(s)$ at a point such as $1-1/B$ behaves in a complicated way. The first few may be dominated by the pole at $0$. So they are near integers if $B$ is an integer. The Taylor series of $f(s)$ is not a good recipe to compute $f(-n)$. Mathematica gives $\zeta^{(4)}(1-1/3)=-5831.99742689039619$ near an integer. mpmath gives -5831.9974268904026. I do not think there is a bug here. $\endgroup$
    – juan
    Sep 25, 2021 at 20:52
  • 1
    $\begingroup$ @Henri Cohen, why is it only implied for $k$ odd? The $k$-th derivative of $(s-1)^{-1}$ at $s=1-1/B$ is precisely $k!(-1)^k(s-1)^{-k-1}=-k!(1/(1-s))^{k+1}=-k!B^{k+1}$, the parity of $k$ shouldn't matter, as far as I see. $\endgroup$ Sep 25, 2021 at 22:04
  • 1
    $\begingroup$ @Henri Cohen, besides, even if it was true only for odd $k$, this would imply a similar exponential bound for $f(s)−f(2−2/B−s)$, which is once again false due to the same counterexample. I believe that the experimental data is a kind of the "law of small numbers" and the fact that $\zeta(s)$ grows "slightly" more than exponentially at $s=−n$ for large odd $n$ contributes to this $\endgroup$ Sep 25, 2021 at 22:27
  • 1
    $\begingroup$ My bad. I had divided by $k!$ in my computations. $\endgroup$ Sep 26, 2021 at 8:21

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.