1
$\begingroup$

This question is related to the prime-counting functions defined in (1) to (4) below.

(1) $\quad A(x)=\sum\limits_{n=2}^\infty\frac{\Lambda(n)}{\log(n)^2}\,\theta(x-n)$

(2) $\quad\Pi(x)=\sum\limits_{n=2}^\infty\frac{\Lambda(n)}{\log(n)}\,\theta(x-n)\qquad\qquad\qquad\text{(Riemann's Prime-Power Counting Function)}$

(3) $\quad\psi(x)=\sum\limits_{n=2}^\infty\Lambda(n)\,\theta(x-n)\qquad\qquad\qquad\text{ (Second Chebyshev Function)}$

(4) $\quad B(x)=\sum\limits_{n=2}^\infty\Lambda(n)\log(n)\,\theta(x-n)$


The prime-counting functions defined in (1) to (4) above are related via their first-order derivatives as follows.

(5) $\quad\Pi'(x)=\log(x)\,A'(x)$

(6) $\quad\psi'(x)=\log(x)\,\Pi'(x)$

(7) $\quad B'(x)=\log(x)\,\psi'(x)$


The asymptotics for the prime counting functions defined in (1) to (4) above are as follows.

(8) $\quad A(x)\approx li(x)-\frac{x}{\log(x)}$

(9) $\quad\Pi(x)\approx li(x)$

(10) $\quad\psi(x)\approx x$

(11) $\quad B(x)\approx x\,(\log(x)-1)$


I know the error bounds on $|\Pi(x)-li(x)|$ and $|\psi(x)-x|$ predicted by the Prime Number Theorem (PNT) and Riemann Hypothesis (RH) have been thoroughly researched and are well documented, but I'm wondering about $A(x)$ and $B(x)$.


Question (1): What are the error bounds for $|A(x)-(li(x)-\frac{x}{\log(x)})|$ predicted by the PNT and RH?

Question (2): What are the error bounds for $|B(x)-(x\,(\log(x)-1))|$ predicted by the PNT and RH?


Note the asymptotic $li(x)-\frac{x}{\log(x)}$ for the prime counting function $A(x)$ is the difference between the two most famous estimates for the asymptotic for the fundamental prime counting function $\pi(x)$.


Question (3): Does the prime counting function $A(x)$ and it's asymptotic $li(x)-\frac{x}{\log(x)}$ provide any meaningful insight into the Riemann Hypothesis?


The prime counting functions defined in (1) to (4) above are related the Riemann zeta function $\zeta(s)$ as follows.

(12) $\quad\int\log(\zeta(s))\,ds=-s\int\limits_0^\infty A(x)\,x^{-s-1}\,ds=-\sum\limits_{n=2}^\infty\frac{\Lambda(n)}{\log(n)^2}\,n^{-s}\,,\quad\Re(s)>1$

(13) $\quad\log(\zeta(s))=s\int\limits_0^\infty\Pi(x)\,x^{-s-1}\,ds=\sum\limits_{n=2}^\infty\frac{\Lambda(n)}{\log(n)}\,n^{-s}\,,\quad\Re(s)>1$

(14) $\quad\frac{\partial\,\log(\zeta(s))}{\partial s}=-s\int\limits_0^\infty\psi(x)\,x^{-s-1}\,ds=-\sum\limits_{n=2}^\infty\Lambda(n)\,n^{-s}\,,\quad\Re(s)>1$

(15) $\quad\frac{\partial^2\,\log(\zeta(s))}{\partial s^2}=s\int\limits_0^\infty B(x)\,x^{-s-1}\,ds=\sum\limits_{n=2}^\infty\Lambda(n)\log(n)\,n^{-s}\,,\quad\Re(s)>1$


Question (4): Is there a formula for $\int\log(\zeta(s))\,ds$ that converges for $\Re(s)\le 1$ and if so, what is its definition?


Riemann's explicit formulas for the prime-power counting function $\Pi(x)$ and von Mangloldt's explicit formula for the second Chebyshev function $\psi(x)$ are as follows.

(16) $\quad\Pi(x)=li(x)-\sum\limits_\rho Ei(\log\,(x)\,\rho)-\log(2)-\sum\limits_{n=1}^\infty Ei(-2\,n\,\log(x))\\$ $\qquad\quad\Pi(x)=li(x)-\sum\limits_\rho Ei(\log(x)\,\rho)-\log (2)+\int_x^\infty\frac{1}{t\,\left(t^2-1\right)\log(t)}\,dt$

(17) $\quad\psi(x)=x-\sum\limits_\rho\frac{x^\rho}{\rho}-\log(2\,\pi)+\sum\limits_{n=1}^\infty\frac{x^{-2\,n}}{2\,n}\\$ $\qquad\quad\psi(x)=x-\sum\limits_\rho\frac{x^\rho}{\rho}-\log(2\,\pi)-\frac{1}{2}\,\log\left(1-\frac{1}{x^2}\right)$


I derived the following explicit formula for the prime counting function $B(x)$ where $Li_2$ is the polylogarithm function and $b\approx 0.63$.

(18) $\quad B(x)=x\,(\log(x)-1)-\sum\limits_\rho\frac{x^\rho\,(\rho\,\log(x)-1)}{\rho^2}+b+\sum\limits_{n=1}^\infty\frac{(2\,n\log(x)+1)\,x^{-2\,n}}{4\,n^2}\\ \quad B(x)=x\,(\log(x)-1)-\sum\limits_\rho\frac{x^\rho\,(\rho\,\log(x)-1)}{\rho^2}+b+\frac{1}{4}\left(Li_2\left(\frac{1}{x^2}\right)-2\,\log(x)\log\left(1-\frac{1}{x^2}\right)\right)$


The following plot illustrates the explicit formula for $B(x)$ defined in (18) above in orange and the reference function $B(x)$ defined in (4) above in blue. The explicit formula is evaluated over the first 100 pairs of zeta zeros with $b=0.63$.

Illustration of Explicit Formula (18) for $B(x)$


I derived the explicit formula for $B(x)$ defined in (18) above via two different approaches which are outlined in (19) and (20) below. Both approaches start with the first explicit formula for $\psi(x)$ defined in (17) above which uses the series representation of the final error term. Both approaches provide exactly the same result, but I'm not sure how to handle the derivation of the constant offset term $b$ for either approach.

(19) $\quad B(x)=\int\log(x)\frac{\partial\,\psi(x)}{\partial\,x}\,dx$

(20) $\quad B(x)=\frac{1}{2\,\pi\,\,i}\int\limits_{c-i\,\infty}^{c+i\,\infty}\frac{\partial}{\partial\,s}\left(-s\int\limits_0^\infty\psi(x)\,x^{-s-1}\,ds\right)\frac{x^s}{s}\,ds$


The first approach outlined in (19) above leverages the relationship $B'(x)=\log(x)\,\psi'(x)$ from (7) above.

The second approach outlined in (20) above takes advantage of the fact that all terms in the first formula for $\psi(x)$ illustrated in (17) above are of the form $a\, x^z$ and the Mellin transform of $x^z$ is $2\,\pi\,\delta(i\,(s+z))$. I illustrated $\mathcal{M}_s^{-1}[2\,\pi\,\delta(i\,(s+z))](x)=x^z$ in the following answer I posted to a question on evaluation of $\delta(s)$ for $s\in \mathbb{C}$.

Answer to Question on Evaluation of $\delta(s)$ for $s\in \mathbb{C}$

I illustrated a single outer integral in approach 2 outlined in (20) above, but this is a bit of an oversimplification as each individual term requires a different value of $c$ for evaluation of its associated integral.


Question (5): What is the derivation and exact value for the constant offset term $b$ associated with the derived explicit formula for $B(x)$ defined in (18) above?


I derived the following explicit formula for the prime counting function $A(x)$.

(21) $\quad A(x)=li(x)-\frac{x}{\log(x)}-\sum\limits_\rho\left(\rho\,Ei(\log(x)\,\rho)+\pi\,\left|\Im(\rho)\right|-\frac{x^\rho}{\log(x)}\right)+a+\\$ $\qquad\qquad\qquad\sum\limits_{n=1}^\infty\left(2\,n\, Ei(-2\,n\,\log(x))+\frac{x^{-2\,n}}{\log(x)}\right)\\$ $\qquad\quad A(x)=li(x)-\frac{x}{\log(x)}-\sum\limits_\rho\left(\rho\,Ei(\log(x)\,\rho)+\pi\,\left|\Im(\rho)\right|-\frac{x^\rho}{\log(x)}\right)+a+\\$ $\qquad\qquad\qquad\sum\limits_{n=1}^\infty (2\,n\,Ei(-2\,n\,\log(x)))+\frac{1}{\left(x^2-1\right)\,\log(x)}$


The following plot illustrates the explicit formula for $A(x)$ defined in (21) above in orange and the reference function $A(x)$ defined in (1) above in blue. The explicit formula is evaluated over the first 100 pairs of zeta zeros with $a=2.5$.

Illustration of Explicit Formula (21) for $A(x)$


I derived the explicit formula for $A(x)$ defined in (21) above via the approach which is outlined in (22) below. This approach starts with the first explicit formula for $\Pi(x)$ defined in (16) above which uses the series representation of the final error term. The approach outlined in (22) below leverages the relationship $\Pi'(x)=\log(x)\,A'(x)$ from (5) above. I'm not sure how to handle the derivation of the constant offset term $a$ for this approach.

(22) $\quad A(x)=\int\frac{1}{log(x)}\frac{\partial\,\Pi(x)}{\partial\,x}\,dx$


Question (6): What is the derivation and exact value for the constant offset term $a$ associated with the derived explicit formula for $A(x)$ defined in (21) above?

March 17, 2018 Update:

I believe I've determined the answer to question (5) above. I believe the constant offset term $b$ associated with the derived explicit formula for $B(x)$ defined in (18) above is as follows:

(23) $\ \ \ b=\frac{\zeta''(0)}{\zeta(0)}-\frac{\zeta'(0)^2}{\zeta(0)^2}=-2\left(\gamma_1+\frac{\gamma^2}{2}-\frac{\pi^2}{24}-\frac{1}{2} (\log(2)+\log(\pi))^2\right)-\log^2(2\,\pi)=0.634921$

$\endgroup$
  • 2
    $\begingroup$ If you have seven questions, ask seven questions, please. I don't think ,any will want to answer all of the above, but then, it wouldn't be an answer, according to the rules of the site. BTW, derivatives are useless for functions constant between jumps, and your asymptotics (8)-(11) are wrong, the error terms are bigger, those constants don't exist. $\endgroup$ – Professor Vector Feb 27 '18 at 7:38
  • $\begingroup$ @ProfessorVector With respect to your comment "derivatives are useless for functions constant between jumps", it appears you're unaware of the theory of generalized functions (distributions) such as the Heaviside step function $\theta(x)$ and it's derivative the Dirac delta function $\delta(x)$. You also don't seem to understand approximations of generalized functions by normal functions such as explicit formulas for prime counting functions. $\endgroup$ – Steven Clark Feb 28 '18 at 1:04
  • $\begingroup$ @ProfessorVector With respect to your comment "your asymptotics (8)-(11) are wrong, the error terms are bigger", the asymptotics for $\Pi(x)$ and $\psi(x)$ are generally considered to be $li(x)$ and $x$ respectively. I originally included the small constant error terms $-\log(2)$ and $-\log(2\,\pi)$ associated with $\Pi(x)$ and $\psi(x)$ to clarify I was asking for the corresponding terms $a$ and $b$ associated with $A(x)$ and $B(x)$. I didn't include the second small error terms in the explicit formulas since they quickly vanish to zero as $x$ increases. $\endgroup$ – Steven Clark Feb 28 '18 at 1:05
  • $\begingroup$ @ProfessorVector It doesn't make sense to include the primary error terms in the explicit formulas which is the sum over the zeta zeros since the whole idea is to study deviation of the prime counting functions from their asymptotics. I deleted the small constant error terms from formulas (8) to (11) since I've now derived an explicit formula for $A(x)$ as well as $B(x)$, and I believe my questions about the small constant error terms $a$ and $b$ are best asked in the context of the explicit formulas for $A(x)$ and $B(x)$. $\endgroup$ – Steven Clark Feb 28 '18 at 1:06
  • $\begingroup$ @ProfessorVector With respect to your comment "those constants don't exist", I believe you're wrong. The small constant error terms $a$ and $b$ associated with the explicit formulas for $A(x)$ and $B(x)$ may not have expressions as simple as the small constant error terms $-\log(2)$ and $-\log(2\,\pi)$ associated with $\Pi(x)$ and $\psi(x)$, but that doesn't mean they don't exist. $\endgroup$ – Steven Clark Feb 28 '18 at 1:06

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.