On proving an Inequality of Ramanujan using Explicit Order Estimates of the Mertens Function

SUBHAM DE \orcidlink0009-0001-3265-4354 Department of Mathematics, Indian Institute of Technology, Delhi, India.

Abstract

This research article provides an unconditional proof of an inequality proposed by Srinivasa Ramanujan involving the Prime Counting Function $\pi(x)$ ,

\displaystyle(\pi(x))^{2}<\frac{ex}{\log x}\pi\left(\frac{x}{e}\right)

for every real $x\geq\exp(1486)$ , using specific order estimates of the Mertens Function, $M(x)$ . The proof primarily hinges upon investigating the underlying relation between $M(x)$ and the Second Chebyshev Function, $\psi(x)$ , in addition to applying the meromorphic properties of the Riemann Zeta Function, $\zeta(s)$ with an intention of deriving an improved approximation for $\pi(x)$ .

Keywords and Phrases: Riemann Zeta Function, Mertens Function, Chebyshev Function, Arithmetic Function, Error Estimates, Perron’s Formula, M $\ddot{o}$ bius Inversion Formula, Dirichlet Partial Summation Formula.

2020 MSC: Primary 11A41, 11A25, 11N05, 11N37, 11N56 .
Secondary 11M06, 11M26 .

1 Introduction and Motivation

The motivation for investigating the distribution of prime numbers over the real line $\mathbb{R}$ first reflected in the writings of famous mathematician Ramanujan, as evident from his letters [23, pp. xxiii-xxx , 349-353] to one of the most prominent mathematiciana of $20^{th}$ century, G. H. Hardy during the months of Jan/Feb of $1913$ , which are testaments to several strong assertions about prime numbers, especaially the Prime Counting Function, $\pi(x)$ [ref. (2.1.1)].

In the following years, Hardy himself analyzed some of thoose results [24] [25, pp. 234-238], and even wholeheartedly acknowledged about them in many of his publications, one such notable result is the Prime Number Theorem [ref. (2.1.1)].

Ramanujan provided several inequalities regarding the behaviour and the asymptotic nature of $\pi(x)$ . One of such relation can be found in the notebooks written by Ramanujan himself has the following claim.

Theorem 1.0.1.

(Ramanujan’s Inequality) For $x$ sufficiently large, we shall have,

\displaystyle(\pi(x))^{2}<\frac{ex}{\log x}\pi\left(\frac{x}{e}\right)

(1.1)

Worth mentioning that, Ramanujan indeed provided a simple, yet unique solution in support of his claim. Furthermore, it has been well established that, the result is not true for every positive real $x$ . Thus, the most intriguing question that the statement of Theorem (1.0.1) poses is, is there any $x_{0}$ such that, Ramanujan’s Inequality will be unconditionally true for every $x\geq x_{0}$ ?

A brilliant effort put up by F. S. Wheeler, J. Keiper, and W. Galway in search for such $x_{0}$ using tools such as : Mathematica went in vain, although independently Galway successfully computed the largest prime counterexample below $10^{11}$ at $x=38\mbox{ }358\mbox{ }837\mbox{ }677$ . However, Hassani [22, Theorem 1.2] proposed a more inspiring answer to the question in a way that, $\exists$ such $x_{0}=138\mbox{ }766\mbox{ }146\mbox{ }692\mbox{ }471\mbox{ }228$ with (1.1) being satisfied for every $x\geq x_{0}$ , but one has to neccesarily assume the Riemann Hypothesis. In a recent paper by A. W. Dudek and D. J. Platt [2, Theorem 1.2], it has been established that, ramanujan’s Inequality holds true unconditionally for every $x\geq\exp(9658)$ . Although this can be considered as an exceptional achievement in this area, efforts of further improvements to this bound are already underway. For example, one recent result by Axler [26] suggests that, the lower bound for $x$ , namely $\exp(9658)$ can in fact be further improved upto $\exp(3158.442)$ using similar techniques as described in [2], although modifying the error term accordingly.

This article shall provide in detail, a new proof of Ramanujan’s Inequality, using a completely different technique by introducing the notion of Mertens Function. We shall utilize one of the most significant order properties of $M(x)$ , namely, $M(x)=O(\sqrt{x})$ in order to find an improved estimate for $\pi(x)$ [ref. Section 3.3]. Thus in turn, we shall verify the inequality in the final part of the article [ref. Section 4]. As an application to this method, we shall be able to refine the lower bound for $x$ even further in order for Theorem (1.0.1) to hold true without any further assumptions.

2 Arithmetic Functions

As for definition, Arithmetic Functions are in fact complex-valued functions on the set of Natural Numbers $\mathbb{N}$ .

For the convenience of the readers, let us first introduce some notations, under standard assumption that, $x\in\mathbb{R}$ .

Definition 2.0.1.

We say $f(x)$ is asymptotic to $g(x)$ , and denote it by, $f(x)\sim g(x)$ if, $\lim\limits_{x\to\infty}\frac{f(x)}{g(x)}=1$ .

Definition 2.0.2.

(Big $O$ Notation) Given $g(x)>0\hskip 10.0pt\forall\mbox{ }x\geq a$ , the notation, $f(x)=O(g(x))$ implies that, the quotient, $\frac{f(x)}{g(x)}$ is bounded for all $x\geq a$ ; i.e., $\exists$ a constant $M>0$ such that,

$\arrowvert f(x)\arrowvert\leq M.g(x)\hskip 10.0pt\mbox{, }\forall\mbox{ }x\geq a$ .

In this section, we shall discuss about a few specific important such type of arithmetic functions pertaining to the context of the paper and the proof of the original result.

2.1 Prime Counting Function

Definition 2.1.1.

For each $\textit{x}\geq 0$ ,we define,

$\pi(x):=$ The number of primes $\leq\textit{x}$ .

The most important contribution of $\pi(x)$ is undoubtedly to the Prime Number Theorem, which can be stated as follows.

Theorem 2.1.1.

For every real $x\geq 0$ , the following estimate is valid.

\displaystyle\pi(x)\sim\frac{x}{\log x}

(2.1)

Equivalently,

\displaystyle\lim\limits_{x\to\infty}\frac{\pi(x)\log x}{x}=1

(2.2)

For an elementary proof of above, readers can refer to [20].

2.2 Chebyshev Function

Chebyshev $\psi$ Function has the following definition.

Definition 2.2.1.

For each $\textit{x}\geq 0$ , we define,

$\psi(x):=\sum\limits_{n\leq x}\Lambda(n)$ ,

Where ,

\displaystyle\Lambda(n):=\left\{\begin{array}[]{cc}\log p\mbox{ },&\mbox{ if }% n=p^{m},\mbox{ }p^{m}\leq x,\mbox{ }m\in\mathbb{N}\\ 0\mbox{ },&\mbox{ otherwise }.\end{array}\right.

(2.5)

$\Lambda(n)$ is said to be the ”Mangoldt Function” .

An important observation is,

\displaystyle\psi(x)=\sum\limits_{n\leq x}\Lambda(n)=\sum\limits_{m=1}^{\infty% }\sum\limits_{p,p^{m}\leq x}\Lambda(p^{m})=\sum\limits_{m=1}^{\infty}\sum% \limits_{p\leq x^{\frac{1}{m}}}\log p

(2.6)

In fact, one can use the $\psi$ function in order tosimplify the statement of the Prime Number Theorem (2.1.1). In other words, one can deduce that, proving the theorem is equivalent to proving the following statement,

\displaystyle\psi(x)\sim x\mbox{\hskip 20.0pt as \hskip 10.0pt}x\rightarrow\infty.

(2.7)

2.3 M $\ddot{o}$ bius Function

We start with the formal definition.

Definition 2.3.1.

(M $\ddot{o}$ bius Function) $\mu:\mathbb{N}\rightarrow\{0,\pm 1\}$ is defined as follows:

$\mu(n):=\left\{\begin{array}[]{lll}{(-1)}^{k}&\mbox{ if }n=\prod\limits_{i=1}^% {k}{{p_{i}}^{a_{i}}}\mbox{ such that, }gcd(p_{i},p_{j})=1\hskip 20.0pt\forall% \hskip 10.0pti\neq j\\ \\ 1&\mbox{ if, }n=1\\ \\ 0&\mbox{ otherwise.}\end{array}\right.$

One can in fact use definition (2.3.1) to deduce the following property regarding the M $\ddot{o}$ bius Function.

Proposition 2.3.1.

[17, Theorem 2.1 , pp. 25]

\sum\limits_{d|n}{\mu(d)}=\left\lfloor\frac{1}{n}\right\rfloor=\left\{\begin{% array}[]{ll}1&\mbox{ if, }n=1\\ \\ 0&\mbox{ otherwise. }\end{array}\right.

(2.8)

2.4 Mertens Function

Definition 2.4.1.

The Mertens Function $M:\mathbb{N}\rightarrow\mathbb{Z}$ has the representation,

M(n):=\sum\limits_{d=1}^{n}{\mu(d)}

(2.9)

Remark 2.4.1.

In general, there’s a notion of the Extended Mertens Function,

$M(x):=\sum\limits_{1\leq n\leq x}{\mu(n)}$ , $\forall$ $x\in\mathbb{R}$

In his paper [4], Mertens conjectured that, for all $M(n)$ with $1\leq n\leq 10^{4}$ , we shall have,

\displaystyle|M(n)|<\sqrt{n}

(2.10)

This is also known as the Mertens Hypothesis. [ Interested readers can refer to [3, Theorem 14.28, pp. 374] ]

Extending Mertens’ results further upto $n=5\times 10^{6}$ , Sterneck [5] conjectured that,

$|M(n)|<\frac{1}{2}\sqrt{n}$ , $\forall$ $n>200$ .

The primary objective for Mertens behind introducing the function $M(x)$ (As defined in (2.4.1)) was its underlying relation to the location of the zeros of the Riemann Zeta Function $\zeta(s)$ , the reason being largely due to it’s consequences for the distribution of the primes, also hailed as one of the most important unsolved problems in Analytic Number Theory. We shall be working with a particular order estimate of $M(x)$ in later sections of the text, although readers are encouraged to consult [9], [11], [15] and [16] for further details.

2.5 Some Necessary Derivations

Proposition 2.5.1.

The Dirichlet Series Representation [3, Theorem 3.13 , pp. 62] for $\mu(n)$ is given by:

\displaystyle\sum_{n=1}^{\infty}\frac{\mu(n)}{n^{s}}=\frac{1}{\zeta(s)}\hskip 2% 0.0pt\mbox{ , }\Re(s)>1.

(2.11)

$\zeta(s)$ denoting the Riemann Zeta Function.

Proposition 2.5.2.

The following order estimates hold true:

$(i)$

$\sum\limits_{d\leq x}\frac{\mu(d)}{d}=O\left(\frac{1}{\log x}\right)$
$(ii)$

$\sum\limits_{d\leq x}\frac{\mu(d)\log(d)}{d}=O(1)$
$(iii)$

$\sum\limits_{d\leq x}\mu(d)=M(x)=O(\sqrt{x})$

Proof.

(i)

Near $s=1$ , we have the expansion for $\zeta(s)$ :

\zeta(s)=\frac{1}{s-1}+\gamma+O(s-1)

Where, $\gamma$ denotes the Euler Constant. Thus,

\frac{1}{\zeta(s)}=(s-1)-\gamma(s-1)^{2}+O((s-1)^{3})

Using Perron’s Formula, we have:

\displaystyle\sum_{n\leq x}\frac{\mu(n)}{n}=\frac{1}{2\pi i}\int\limits_{c-iT}% ^{c+iT}\frac{1}{\zeta(s)}\frac{x^{s}}{s}\,ds+O\left(\frac{x\log^{2}x}{T}\right)

(2.12)

where $c>1$ and $T$ is a parameter to be chosen later.

We evaluate the integral using the following steps.

Step 1: Integral around $s=1$

Consider a small semicircle $\Gamma$ (say) of radius $\epsilon$ around $s=1$ having the following parametrization, $s=1+\epsilon e^{i\theta}$ , $-\pi/2\leq\theta\leq\pi/2$ .

In this region,

\displaystyle\frac{1}{\zeta(s)}=\epsilon e^{i\theta}-\gamma\epsilon^{2}e^{2i% \theta}+O(\epsilon^{3})

and,

	$\displaystyle\frac{x^{s}}{s}=\frac{x^{1+\epsilon e^{i\theta}}}{1+\epsilon e^{i% \theta}}=x\cdot\frac{x^{\epsilon e^{i\theta}}}{1+\epsilon e^{i\theta}}=x\cdot e% ^{\epsilon e^{i\theta}\log x}\left(1-\epsilon e^{i\theta}+O(\epsilon^{2})\right)$
	$\displaystyle=x\left(1+\epsilon e^{i\theta}\log x+O(\epsilon^{2})\right)\left(% 1-\epsilon e^{i\theta}+O(\epsilon^{2})\right)$
	$\displaystyle=x\left(1+\epsilon e^{i\theta}(\log x-1)+O(\epsilon^{2})\right)$

Thus, the integrand becomes:

	$\displaystyle\frac{1}{\zeta(s)}\frac{x^{s}}{s}=x\left(\epsilon e^{i\theta}-% \gamma\epsilon^{2}e^{2i\theta}+O(\epsilon^{3})\right)\left(1+\epsilon e^{i% \theta}(\log x-1)+O(\epsilon^{2})\right)$
	$\displaystyle\hskip 50.0pt=x\left(\epsilon e^{i\theta}+\epsilon^{2}e^{2i\theta% }(\log x-1-\gamma)+O(\epsilon^{3})\right)$

Therefore,

	$\displaystyle\int\limits_{\Gamma}\frac{1}{\zeta(s)}\frac{x^{s}}{s}=\int\limits% _{-\pi/2}^{\pi/2}x\left(\epsilon e^{i\theta}+\epsilon^{2}e^{2i\theta}(\log x-1% -\gamma)+O(\epsilon^{3})\right)i\epsilon e^{i\theta}\,d\theta$
	$\displaystyle=\int\limits_{-\pi/2}^{\pi/2}x\left(\epsilon^{2}e^{2i\theta}(\log x% -1-\gamma)+O(\epsilon^{3})\right)ie^{i\theta}\,d\theta\hskip 20.0pt\left[% \because\int\limits_{-\pi/2}^{\pi/2}e^{2i\theta}\,d\theta=0\right]$
	$\displaystyle=\epsilon^{2}x(\log x-1-\gamma)\int\limits_{-\pi/2}^{\pi/2}ie^{3i% \theta}\,d\theta+O(\epsilon^{3})$

Again, the integral of $e^{3i\theta}$ over a symmetric interval around zero is zero. Therefore, the integral around the small semicircle contributes a negligible amount of $O(\epsilon^{3}x)$ .

Step 2: Integral along the vertical line $s=c+it$

For the part of the integral along the vertical line, say, $L$ : $s=c+it$ , $-T\leq t\leq T$ , where $c>1$ , i.e.,

\displaystyle\int\limits_{c-iT}^{c+iT}\frac{1}{\zeta(s)}\frac{x^{s}}{s}\,ds

It can indeed be verified that, $\frac{1}{\zeta(s)}$ is bounded on $L$ as, $\zeta(s)$ does not have any pole for $\operatorname{Re}(s)>1$ . Specifically, for $s=c+it$ with $c>1$ , $\zeta(s)$ is bounded away from zero, so $\frac{1}{\zeta(s)}$ is bounded.

An appropriate choice of $T=\sqrt{x}$ gives us a bound on the integral:

\displaystyle\left|\mbox{ }\int\limits_{c-iT}^{c+iT}\frac{1}{\zeta(s)}\frac{x^% {c+it}}{c+it}\,ds\right|\leq\int\limits_{-T}^{T}\left|\frac{1}{\zeta(c+it)}% \frac{x^{c}x^{it}}{c+it}\right|dt\leq Kx^{c}\int\limits_{-T}^{T}\frac{1}{\sqrt% {c^{2}+t^{2}}}\,dt\leq Kx^{c}\cdot\frac{2T}{c}

Since $\frac{1}{\zeta(c+it)}$ is bounded by some constant $K$ and $\left|x^{it}\right|=1$ . Subsequently,

\displaystyle\left|\int\limits_{L}\frac{1}{\zeta(s)}\frac{x^{s}}{s}\,ds\right|% =\left|\mbox{ }\int\limits_{c-iT}^{c+iT}\frac{1}{\zeta(s)}\frac{x^{s}}{s}\,ds% \right|\leq Kx^{c}\cdot\frac{2\sqrt{x}}{c}=O\left(x^{c-\frac{1}{2}}\right)

Since $c>1$ , $c-\frac{1}{2}>\frac{1}{2}$ , and for large $x$ , this term is small.

Step 3: Error term from the integral

The total error term combining both parts is:

\displaystyle O\left(\frac{x\log^{2}x}{T}\right)=O\left(\frac{x\log^{2}x}{% \sqrt{x}}\right)=O(x^{1/2}\log^{2}x)

It can be observed that, the residue term around $s=1$ contributes to $x$ , and the error terms contribute to $=O(\epsilon^{3}x)+O(x^{1/2}\log^{2}x)$ .

Choosing $\epsilon$ small enough (such as $\epsilon=x^{-1/6}$ ),

\displaystyle O(\epsilon^{3}x)=O(x^{1-1/2})=O(x^{1/2})

Thus, combining all terms and dividing by $x$ to normalize, we obtain.

\displaystyle\sum_{d\leq x}\frac{\mu(d)}{d}=x+O(x^{1/2}\log^{2}x)=O\left(\frac% {1}{\log x}\right)

(ii)

As for the proof, we can use Perron’s formula for any arithmetic function $a(n)$ ,

\displaystyle\sum_{n\leq x}a(n)=\frac{1}{2\pi i}\int\limits_{c-iT}^{c+iT}\left% (\sum_{n=1}^{\infty}\frac{a(n)}{n^{s}}\right)\frac{x^{s}}{s}\,ds+O\left(\sum_{% n=1}^{\infty}\frac{|a(n)|}{n^{c}}\min\left(1,\frac{x}{T|s|}\right)\right)

(2.13)

where $c>1$ and $T$ be a suitably chosen parameter.

Consider the Dirichlet series involving $\mu(d)\log d$ :

\displaystyle-\frac{\zeta^{\prime}(s)}{\zeta^{2}(s)}=\sum_{n=1}^{\infty}\frac{% \mu(n)\log n}{n^{s}}

Applying this to (2.13):

\displaystyle\sum_{d\leq x}\frac{\mu(d)\log d}{d}=-\frac{1}{2\pi i}\int\limits% _{c-iT}^{c+iT}\frac{\zeta^{\prime}(s)}{\zeta^{2}(s)}\frac{x^{s}}{s}\,ds+O\left% (\sum_{d=1}^{\infty}\frac{|\mu(d)\log d|}{d^{c}}\min\left(1,\frac{x}{T|s|}% \right)\right)

We thus study the following integral,

\displaystyle-\frac{1}{2\pi i}\int\limits_{c-iT}^{c+iT}\frac{\zeta^{\prime}(s)% }{\zeta^{2}(s)}\frac{x^{s}}{s}\,ds

using following steps.

Step 1: Integral around $s=1$

Consider a small semicircle $\Gamma$ (say) of radius $\epsilon$ around $s=1$ having the following parametrization, $s=1+\epsilon e^{i\theta}$ , $-\pi/2\leq\theta\leq\pi/2$ , and a priori from the fact that, near $s=1$ ,

\displaystyle\zeta(s)=\frac{1}{s-1}+\gamma+O(s-1)

and,

\displaystyle\zeta^{\prime}(s)=-\frac{1}{(s-1)^{2}}+O(1)

Thus, on $\Gamma$ , we have,

\displaystyle\frac{\zeta^{\prime}(s)}{\zeta^{2}(s)}=\frac{-\frac{1}{(\epsilon e% ^{i\theta})^{2}}+O(1)}{\left(\frac{1}{\epsilon e^{i\theta}}+\gamma+O(\epsilon)% \right)^{2}}=-\epsilon^{-1}e^{-i\theta}+O(\epsilon)

Hence, the integrand becomes,

\displaystyle-\frac{1}{2\pi i}\int\limits_{-\pi/2}^{\pi/2}\left(-\epsilon^{-1}% e^{-i\theta}+O(\epsilon)\right)\frac{x^{1+\epsilon e^{i\theta}}}{(1+\epsilon e% ^{i\theta})}\,i\epsilon e^{i\theta}d\theta

	$\displaystyle=-\frac{1}{2\pi i}\int\limits_{-\pi/2}^{\pi/2}\left(-\epsilon^{-1% }e^{-i\theta}+O(\epsilon)\right)\mbox{ . }x\left(1+\epsilon e^{i\theta}\log x+% O(\epsilon^{2})\right)\left(1-\epsilon e^{i\theta}+O(\epsilon^{2})\right)i% \epsilon e^{i\theta}d\theta$
	$\displaystyle=-\frac{1}{2\pi}\int\limits_{-\pi/2}^{\pi/2}\left(-\epsilon^{-1}e% ^{-i\theta}+O(\epsilon)\right)\left(x\epsilon e^{i\theta}(\log x-1)+O(\epsilon% ^{2})\right)i\epsilon e^{i\theta}d\theta$

\displaystyle=O(\epsilon^{2}x(\log x-1))

( Expanding and simplifying, the leading term integrates to zero due to symmetry of the integrand around zero ).

Step 2: Integral along the vertical line $s=c+it$

For the part of the integral along the vertical line, say, $L$ : $s=c+it$ , $-T\leq t\leq T$ , where $c>1$ , i.e.,

\displaystyle\int_{c-iT}^{c+iT}\frac{\zeta^{\prime}(s)}{\zeta^{2}(s)}\frac{x^{% s}}{s}\,ds

It can be checked that, $\frac{\zeta^{\prime}(s)}{\zeta^{2}(s)}$ is bounded on $L$ . Specifically, for $s=c+it$ with $c>1$ , both $\zeta(s)$ and $\zeta^{\prime}(s)$ are bounded, so $\frac{\zeta^{\prime}(s)}{\zeta^{2}(s)}$ is bounded by some constant $K$ . Hence,

\displaystyle\left|\mbox{ }\int\limits_{c-iT}^{c+iT}\frac{\zeta^{\prime}(s)}{% \zeta^{2}(s)}\frac{x^{c+it}}{c+it}\,ds\right|\leq K\int_{-T}^{T}\frac{x^{c}}{% \sqrt{c^{2}+t^{2}}}\,dt\leq K\mbox{ . }\frac{2T}{c}

Choosing $T=\sqrt{x}$ ,

\displaystyle\left|\mbox{ }\int\limits_{c-iT}^{c+iT}\frac{\zeta^{\prime}(s)}{% \zeta^{2}(s)}\frac{x^{s}}{s}\,ds\right|\leq Kx^{c}\cdot\frac{2\sqrt{x}}{c}=O(x% ^{c-\frac{1}{2}})

Since $c>1$ , $c-\frac{1}{2}>\frac{1}{2}$ , and for large $x$ , this term is small.

Step 3: Error term from the integral

The total error term combining both parts is:

\displaystyle O\left(\frac{x\log^{2}x}{T}\right)=O\left(\frac{x\log^{2}x}{% \sqrt{x}}\right)=O(x^{1/2}\log^{2}x)

Important to note that the residue term around $s=1$ contributes negligibly as $=O(\epsilon^{2}x(\log x-1))$ , and the error terms contribute upto $O(x^{1/2}\log^{2}x)=O(x^{1/3}(\log x-1))$ . (This can be achieved by considering $\epsilon$ small enough (such as $\epsilon=x^{-1/6}$ ). Therefore, we conclude that,

\displaystyle\sum_{d\leq x}\frac{\mu(d)\log d}{d}=O(1)

(iii)

A priori from the definition of $M(x)$ and applying Perron’s Formula for $a(n)=\mu(n)$ yields,

\displaystyle M(x)=\frac{1}{2\pi i}\int\limits_{c-iT}^{c+iT}\frac{1}{\zeta(s)}% \frac{x^{s}}{s}\,ds+O\left(\sum_{n=1}^{\infty}\frac{1}{n^{c}}\min\left(1,\frac% {x}{T|s|}\right)\right)

(2.14)

As for computing the integral in (2.14) over the vertical line $L$ : $s=c+it$ , $-T\leq t\leq T$ , where $c>1$ , our aim is to try shifting the contour of integration to a vertical line closer to the critical strip. For our convenience, we choose $c=1+\epsilon$ where, $\epsilon>0$ . Using the fact that $\zeta(s)$ has no zeros for $\Re(s)>1$ , we intend on obtaining a suitable bound for $\frac{1}{\zeta(s)}$ in this region.

Observe that for $\Re(s)=1+\epsilon$ , $\zeta(s)$ is bounded away from zero, implying $\frac{1}{\zeta(s)}$ is also bounded. Specifically, for $s=1+\epsilon+it$ ,

\displaystyle\left|\frac{1}{\zeta(s)}\right|\leq A

for some constant $A$ . On the other hand,

	$\displaystyle\left\|\mbox{ }\int\limits_{1+\epsilon-iT}^{1+\epsilon+iT}\frac{1}% {\zeta(s)}\frac{x^{s}}{s}\,ds\right\|=\left\|\mbox{ }\int\limits_{-T}^{T}\frac{x% ^{1+\epsilon+it}}{1+\epsilon+it}\cdot\frac{1}{\zeta(1+\epsilon+it)}\,dt\right\|% \leq A\cdot\int\limits_{-T}^{T}\frac{x^{1+\epsilon}}{\sqrt{(1+\epsilon)^{2}+t^% {2}}}\,dt$
	$\displaystyle\leq A\cdot x^{1+\epsilon}\int\limits_{-\infty}^{\infty}\frac{1}{% \sqrt{(1+\epsilon)^{2}+t^{2}}}\,dt=A\cdot x^{1+\epsilon}\frac{\pi}{1+\epsilon}% =O(x^{1+\epsilon})$

Subsequently, the error term from the vertical line integral can be estimated as, $O(x^{1+\epsilon})$ , for any small $\epsilon>0$ .

Hence, we need to choose $T$ appropriately to control the error term in Perron’s formula. Using the Cauchy Residue Theorem [21, Chapt. 5.1 , pp. 120] and estimating the integral, we set $T=\sqrt{x}$ and consider the main term and error terms:

\displaystyle\int\limits_{1+\epsilon-i\sqrt{x}}^{1+\epsilon+i\sqrt{x}}\frac{1}% {\zeta(s)}\frac{x^{s}}{s}\,ds=O(x^{1/2+\epsilon})

This ensures that the main contribution comes from the vertical integral and the error terms are bounded appropriately. Accordingly,

\displaystyle M(x)=O(x^{1/2+\epsilon})

By choosing $\epsilon>0$ sufficiently small, we can make the bound as close to $O(\sqrt{x})$ as desired.

∎

3 Order Estimates involving $M(x)$

In this section, we shall apply the definitions of Chebyshev $\psi$ Function, $\psi(x)$ in order to come up with a suitable estimate for $\pi(x)$ in terms of the Mertens Function, $M(x)$ .

3.1 Relation between $\psi(x)$ and $M(x)$

Theorem 3.1.1.

The following holds true for the Chebyshev $\psi$ function, $\psi(x)$ :

\displaystyle\psi(x)=x-\sum_{n\leq x}M\left(\frac{x}{n}\right)+O(\sqrt{x})

(3.1)

Proof.

A proiori from the definition of $\psi(x)$ involving the Von Mangoldt Function, $\Lambda(n)$ , we apply the M $\ddot{o}$ bius Inversion Formula [17, Section 14.1 , pp. 30] on $\Lambda(n)$ to obtain,

\displaystyle\Lambda(n)=\sum_{d\mid n}\mu(d)\log\left(\frac{n}{d}\right)=\sum_% {n\leq x}\sum_{d\mid n}\mu(d)\log\left(\frac{n}{d}\right)

(3.2)

\displaystyle=\sum_{d\leq x}\mu(d)\sum_{k\leq\frac{x}{d}}\log(k)\hskip 20.0pt[% \mbox{ Interchanging order of summation }]

(3.3)

Remark 3.1.2.

Here, we set $n=dk$ , so that the inner sum is over $k$ with $k\leq\frac{x}{d}$ .

We approximate the sum of $\log(k)$ by integrating the logarithm function from 1 to $\frac{x}{d}$ , and then applying Integration by Parts.

\displaystyle\sum_{k\leq\frac{x}{d}}\log(k)=\int_{1}^{\frac{x}{d}}\log(t)\,dt+% O(\log(\frac{x}{d}))=\left[t\log(t)-t\right]_{1}^{\frac{x}{d}}+O(\log(\frac{x}% {d}))

	$\displaystyle=\frac{x}{d}\log\left(\frac{x}{d}\right)-\frac{x}{d}+1+O(\log(% \frac{x}{d}))=\frac{x}{d}\log\left(\frac{x}{d}\right)-\frac{x}{d}+1+O(\log(x))$
	$\displaystyle[\mbox{ Since $\log(\frac{x}{d})$ is of the same order as $\log(x% )$ for large $x$ }]$

Subsequently, from (3.3) we get,

\displaystyle\psi(x)=\sum_{d\leq x}\mu(d)\left(\frac{x}{d}\log\left(\frac{x}{d% }\right)-\frac{x}{d}+1+O(\log(x))\right)

Distributing $\mu(d)$ and separating the main terms from the error term,

	$\displaystyle\psi(x)=\sum_{d\leq x}\mu(d)\left(\frac{x}{d}\log\left(\frac{x}{d% }\right)\right)-\sum_{d\leq x}\mu(d)\left(\frac{x}{d}\right)+\sum_{d\leq x}\mu% (d)\cdot 1+\sum_{d\leq x}\mu(d)O(\log(x))$
	$\displaystyle=x\sum_{d\leq x}\frac{\mu(d)}{d}\log\left(\frac{x}{d}\right)-x% \sum_{d\leq x}\frac{\mu(d)}{d}+\sum_{d\leq x}\mu(d)+O(x\log(x))\sum_{d\leq x}% \mu(d)$

A priori using the results obtained in proposition (2.5.2),

\displaystyle x\sum_{d\leq x}\frac{\mu(d)}{d}\log\left(\frac{x}{d}\right)=x% \log(x)\sum_{d\leq x}\frac{\mu(d)}{d}-x\sum_{d\leq x}\frac{\mu(d)\log(d)}{d}

(3.4)

Furthermore,

\displaystyle x\log(x)\cdot O\left(\frac{1}{\log x}\right)-x\cdot O(1)=O(x)-x=% O(x)

Hence,

\displaystyle O(x\log x)\sum_{d\leq x}\mu(d)=O(x\log x)\cdot O(\sqrt{x})=O(x^{% 3/2}\log x)

(3.5)

Combining (3.4) and (3.5) yields,

\displaystyle\psi(x)=x-\sum_{n\leq x}M\left(\frac{x}{n}\right)+O(x)+O(x^{3/2}% \log x)

Since $O(x^{3/2}\log x)$ dominates $O(x)$ , we conclude that,

\displaystyle\psi(x)=x-\sum_{n\leq x}M\left(\frac{x}{n}\right)+O(\sqrt{x})

And the proof is thus complete. ∎

3.2 An important approximation for $\psi(x)$

A tricky application of the famous Prime Number Theorem (2.1.1) yields the following estimate,

\displaystyle\psi(x)=x+O\left(xe^{-c\sqrt{\log x}}\right)

(3.6)

or some constant $c>0$ . However, it is indeed possible to obtain a simpler, and more effective bound for the Chebyshev $\psi$ Function.

Lemma 3.2.1.

We have,

\displaystyle\psi(x)=x+O\left(x^{1/2}\log^{2}x\right)

(3.7)

Proof.

This proof thouroughly utilizes results from Analytic Number Theory, specifically the properties of the Chebyshev function $\psi(x)$ and the distribution of primes. We shall also leverage results from the analytic properties of the Riemann zeta function $\zeta(s)$ .

Important to note that, the proof relies on properties of the Riemann zeta function $\zeta(s)$ and its non-trivial zeros. However, we do not assume the Riemann Hypothesis (RH) here explicitly.

The explicit formula for $\psi(x)$ involves the zeros of $\zeta(s)$ :

\psi(x)=x-\sum_{\rho}\frac{x^{\rho}}{\rho}-\log(2\pi),

where the sum is over the non-trivial zeros $\rho$ of $\zeta(s)$ , and the term $\log(2\pi)$ arises due to the existence of pole at $s=1$ .

The critical line $\Re(s)=\frac{1}{2}$ plays a key role. Let $\rho=\beta+i\gamma$ be a non-trivial zero of $\zeta(s)$ with $\beta=\Re(\rho)$ and $\gamma=\Im(\rho)$ .

The zeros are symmetric about the real axis, so we consider only the upper half-plane. For each zero $\rho$ of $\zeta(s)$ , the term $\frac{x^{\rho}}{\rho}$ contributes to $\psi(x)$ . To estimate the error term, consider the sum over the non-trivial zeros $\rho$ :

\displaystyle\sum_{\rho}\frac{x^{\rho}}{\rho}

In addition to above, we use the fact that the zeros $\rho$ have $\beta\leq 1$ . The contribution of each such $\rho$ can be bounded by,

\displaystyle\left|\frac{x^{\rho}}{\rho}\right|\leq x^{\beta}|\rho|^{-1}

For zeros with $\beta=\frac{1}{2}$ , $x^{\beta}=x^{1/2}.$ . Hence, the number of zeros with $|\gamma|\leq T$ is $O(T\log T)$ . We choose $T=x$ to cover the relevant range of zeros.

\displaystyle\sum_{|\gamma|\leq x}|\rho|^{-1}=\sum_{|\gamma|\leq x}\left(\left% (\frac{1}{2}\right)^{2}+\gamma^{2}\right)^{-1/2}\leq\sum_{|\gamma|\leq x}\frac% {1}{|\gamma|}=O(\log x)

Combining these estimates, we obtain,

\displaystyle\sum_{\rho}\frac{x^{\rho}}{\rho}=O(x^{1/2}\log x)

(3.8)

Including the logarithmic term from the pole at $s=1$ , the error term in $\psi(x)$ becomes,

\displaystyle\psi(x)=x+O(x^{1/2}\log x)

Since we know that the error term actually involves $\log^{2}x$ due to the density of the zeros of $\zeta(s)$ and more refined estimates, thus, we can further improve our estimate to,

\displaystyle\psi(x)=x+O(x^{1/2}\log^{2}x)

∎

As the title of this section suggests, we shall now proceed towards understanding how we can approximate $\pi(x)$ using properties of $\psi(x)$ .

Theorem 3.2.2.

The following holds for the Prime counting function, $\pi(x)$ :

\displaystyle\pi(x)=\frac{\psi(x)}{\log x}+O\left(\frac{x}{(\log x)^{2}}\right)

(3.9)

Proof.

Before we delve into the proof, notice that $\Lambda(n)$ is zero except when $n$ is a power of a prime, specifically $n=p^{k}$ . We can hence rewrite $\psi(x)$ as,

\displaystyle\psi(x)=\sum_{p\leq x}\log p\cdot\left(1+\left\lfloor\frac{\log x% }{\log p}\right\rfloor\right)

Further simplification yields,

\displaystyle\psi(x)=\sum_{p^{k}\leq x}\log p=\sum_{p\leq x}\log p+\sum_{p^{2}% \leq x}\log p+\sum_{p^{3}\leq x}\log p+\cdots=\sum_{p\leq x}\log p+\sum_{k=2}^% {\infty}\sum_{p\leq x^{1/k}}\log p

(3.10)

The error term in (3.10) comes from the higher powers of primes,

\displaystyle\sum_{k=2}^{\infty}\sum_{p\leq x^{1/k}}\log p

For $k=2$ , the sum is,

\displaystyle\sum_{p\leq\sqrt{x}}\log p\approx\frac{1}{2}\sqrt{x}\log x

Whereas, for higher values of $k$ , the sums are even smaller,

\displaystyle\sum_{k=3}^{\infty}\sum_{p\leq x^{1/k}}\log p\leq\sum_{k=3}^{% \infty}\sqrt[3]{x}\log x=O(\sqrt[3]{x}\log x)

Therefore, the error term is dominated by the term corresponding to $k=2$ , in other words,

\displaystyle\sum_{k=2}^{\infty}\sum_{p\leq x^{1/k}}\log p=O(\sqrt{x}\log x).

Combining the main term and the error terms,

\displaystyle\psi(x)=\sum_{p\leq x}\log p+O(\sqrt{x}\log x)

(3.11)

On the other hand, applying the statement of Prime Number Theorem, we can have the following approximation,

\displaystyle\sum_{p\leq x}\log p\sim\pi(x)\log x

More precisely,

\displaystyle\sum_{p\leq x}\log p=\pi(x)\log x+\epsilon(x)

(3.12)

Given our earlier discussion, the error term, $\epsilon(x)$ can be bounded by,

\displaystyle\epsilon(x)=O(\sqrt{x}\log x).

To isolate $\pi(x)$ from (3.12), we divide both sides by $\log x$ ,

\displaystyle\pi(x)=\frac{\psi(x)}{\log x}-\frac{\epsilon(x)}{\log x}

Substituting the error term bound:

\displaystyle\pi(x)=\frac{\psi(x)}{\log x}+O\left(\frac{\sqrt{x}\log x}{\log x% }\right)=\frac{\psi(x)}{\log x}+O\left(\frac{\sqrt{x}}{1}\right)

From (3.7) in Lemma (3.2.1), we can further refine,

\displaystyle\frac{\psi(x)}{\log x}=\frac{x}{\log x}+O\left(\frac{\sqrt{x}\log% ^{2}x}{\log x}\right)=\frac{x}{\log x}+O(\sqrt{x}\log x)

Combining the error terms, and applying Theorem (2.1.1) again enables us to assert that,

\displaystyle\pi(x)=\frac{\psi(x)}{\log x}+O\left(\frac{x}{(\log x)^{2}}\right).

∎

3.3 An improved estimate for $\pi(x)$

First, let us recall that, (3.1) in Theorem (3.1.1) gives us an order estimate for $\psi(x)$ in terms of $M(x)$ , whereas, we have derived a unique representation of $\pi(x)$ applying properties of $\psi(x)$ , and analytic properties of the Riemann Zeta Function $\zeta(s)$ , as mentioned in (3.9) in Theorem (3.2.2).

We substitute (3.1) into (3.9) to obtain,

\displaystyle\pi(x)=\frac{x-\sum_{n\leq x}M\left(\frac{x}{n}\right)+O(\sqrt{x}% )}{\log x}+O\left(\frac{x}{(\log x)^{2}}\right)

\displaystyle\hskip 50.0pt=\frac{x}{\log x}-\frac{1}{\log x}\sum_{n\leq x}M% \left(\frac{x}{n}\right)+\frac{O(\sqrt{x})}{\log x}+O\left(\frac{x}{(\log x)^{% 2}}\right)

Combining the error terms,

\displaystyle\pi(x)=\frac{x}{\log x}-\frac{1}{\log x}\sum_{n\leq x}M\left(% \frac{x}{n}\right)+O\left(\frac{\sqrt{x}}{\log x}\right)+O\left(\frac{x}{(\log x% )^{2}}\right).

(3.13)

Note that $O\left(\frac{\sqrt{x}}{\log x}\right)$ is asymptotically smaller than $O\left(\frac{x}{\log^{2}x}\right)$ , so the dominant error term is $O\left(\frac{x}{(\log x)^{2}}\right)$ . Therefore, we have the following improved estimate for $\pi(x)$ as follows.

Theorem 3.3.1.

\displaystyle\pi(x)=\frac{x}{\log x}-\frac{1}{\log x}\sum_{n\leq x}M\left(% \frac{x}{n}\right)+O\left(\frac{x}{(\log x)^{2}}\right)

(3.14)

4 Proving Ramanujan’s Inequality

In order to prove Ramanujan’s Inequality, our primary intention will be to investigate the sign of the function,

\displaystyle\mathcal{G}(x):=(\pi(x))^{2}-\frac{ex}{\log(x)}\pi\left(\frac{x}{% e}\right)

(4.1)

for large values of $x$ using the relationship between $\pi(x)$ and the Mertens function $M(x)$ . Given the complexity of the order relations an the extent of robust computations involving these functions as discussed in the previous section, we’ll have to analyze the expressions and error terms cautiously.

4.1 An order expression for $\mathcal{G}(x)$

A priori from (3.14) of Theorem (3.3.1), we have the following expressions for $\pi(x)$ and $\pi(x/e)$ as,

\displaystyle\pi(x)=\frac{x}{\log x}-\frac{1}{\log x}\sum_{n\leq x}M\left(% \frac{x}{n}\right)+O\left(\frac{x}{(\log x)^{2}}\right)

and,

\displaystyle\pi(x/e)=\frac{x/e}{\log(x/e)}-\frac{1}{\log(x/e)}\sum_{n\leq x/e% }M\left(\frac{x/e}{n}\right)+O\left(\frac{x/e}{(\log(x/e))^{2}}\right)

\displaystyle\hskip 50.0pt=\frac{x/e}{\log x-1}-\frac{1}{\log x-1}\sum_{n\leq x% /e}M\left(\frac{x/e}{n}\right)+O\left(\frac{x/e}{(\log x-1)^{2}}\right)

Subsequently,

\displaystyle\frac{ex}{\log x}\pi(x/e)=\frac{ex}{\log x}\left(\frac{x/e}{\log x% -1}-\frac{1}{\log x-1}\sum_{n\leq x/e}M\left(\frac{x/e}{n}\right)+O\left(\frac% {x/e}{(\log x-1)^{2}}\right)\right)

\displaystyle\hskip 30.0pt=\frac{x^{2}}{\log x(\log x-1)}-\frac{ex}{\log x(% \log x-1)}\sum_{n\leq x/e}M\left(\frac{x/e}{n}\right)+O\left(\frac{x^{2}}{\log x% (\log x-1)^{2}}\right)

(4.2)

Furthermore, squaring $\pi(x)$ yields,

\displaystyle(\pi(x))^{2}=\left(\frac{x}{\log x}-\frac{1}{\log x}\sum_{n\leq x% }M\left(\frac{x}{n}\right)+O\left(\frac{x}{(\log x)^{2}}\right)\right)^{2}

\displaystyle=\left(\frac{x}{\log x}\right)^{2}-2\left(\frac{x}{\log x}\right)% \left(\frac{1}{\log x}\sum_{n\leq x}M\left(\frac{x}{n}\right)\right)+\left(% \frac{1}{\log x}\sum_{n\leq x}M\left(\frac{x}{n}\right)\right)^{2}+O\left(% \frac{x^{2}}{(\log x)^{3}}\right)

\displaystyle=\frac{x^{2}}{(\log x)^{2}}-\frac{2x}{(\log x)^{2}}\sum_{n\leq x}% M\left(\frac{x}{n}\right)+\frac{1}{(\log x)^{2}}\left(\sum_{n\leq x}M\left(% \frac{x}{n}\right)\right)^{2}+O\left(\frac{x^{2}}{(\log x)^{3}}\right)

(4.3)

Therefore, using (4.2) and (4.3), we deduce that,

\displaystyle\mathcal{G}(x)=\left(\frac{x^{2}}{(\log x)^{2}}-\frac{2x}{(\log x% )^{2}}\sum_{n\leq x}M\left(\frac{x}{n}\right)+\frac{1}{(\log x)^{2}}\left(\sum% _{n\leq x}M\left(\frac{x}{n}\right)\right)^{2}+O\left(\frac{x^{2}}{(\log x)^{3% }}\right)\right)

\displaystyle\hskip 50.0pt-\left(\frac{x^{2}}{\log x(\log x-1)}-\frac{ex}{\log x% (\log x-1)}\sum_{n\leq x/e}M\left(\frac{x/e}{n}\right)+O\left(\frac{x^{2}}{% \log x(\log x-1)^{2}}\right)\right).

Simplifying each term:

\displaystyle=\frac{x^{2}}{(\log x)^{2}}-\frac{x^{2}}{\log x(\log x-1)}-\frac{% 2x}{(\log x)^{2}}\sum_{n\leq x}M\left(\frac{x}{n}\right)+\frac{1}{(\log x)^{2}% }\left(\sum_{n\leq x}M\left(\frac{x}{n}\right)\right)^{2}

\displaystyle\hskip 200.0pt+\frac{ex}{\log x(\log x-1)}\sum_{n\leq x/e}M\left(% \frac{x/e}{n}\right)+O\left(\frac{x^{2}}{(\log x)^{3}}\right).

(4.4)

\displaystyle=\frac{x^{2}}{(\log x)^{2}(\log x-1)}-\frac{2x}{(\log x)^{2}}\sum% _{n\leq x}M\left(\frac{x}{n}\right)+\frac{1}{(\log x)^{2}}\left(\sum_{n\leq x}% M\left(\frac{x}{n}\right)\right)^{2}

\displaystyle\hskip 150.0pt+\frac{ex}{\log x(\log x-1)}\sum_{n\leq x/e}M\left(% \frac{x/e}{n}\right)+O\left(\frac{x^{2}}{(\log x)^{3}}\right)

(4.5)

4.2 Asymptotic Behavior of indivudual Terms

A priori using involving the Mertens function $M(x)$ ,

\displaystyle\sum_{n\leq x}M\left(\frac{x}{n}\right)=O\left(\sum_{n\leq x}% \sqrt{\frac{x}{n}}\right)=O\left(\sqrt{x}\sum_{n\leq x}\frac{1}{\sqrt{n}}% \right)=O(\sqrt{x}\cdot\sqrt{x})=O(x)

(4.6)

So, we can further approximate each term of (4.5) as follows,

\displaystyle\frac{1}{(\log x)^{2}}\left(\sum_{n\leq x}M\left(\frac{x}{n}% \right)\right)^{2}=O\left(\frac{x^{2}}{(\log x)^{2}}\right),

\displaystyle\frac{ex}{\log x(\log x-1)}\sum_{n\leq x/e}M\left(\frac{x/e}{n}% \right)=O\left(\frac{x^{2}}{\log x(\log x-1)}\right).

As a consequence,

\displaystyle\mathcal{G}(x)=-\frac{2x}{(\log x)^{2}}\sum_{n\leq x}M\left(\frac% {x}{n}\right)+O\left(\frac{x^{2}}{(\log x)^{2}}\right)+O\left(\frac{x^{2}}{% \log x(\log x-1)}\right)+O\left(\frac{x^{2}}{(\log x)^{3}}\right).

\displaystyle\hskip 100.0pt=-\frac{2x}{(\log x)^{2}}\sum_{n\leq x}M\left(\frac% {x}{n}\right)+O\left(\frac{x^{2}}{(\log x)^{2}}\right)

(4.7)

For sufficiently large values of $x$ . Therefore, we conclude that, $\mathcal{G}(x)<0$ , for large values of $x$ , and this concludes our proof of the inequality.

later, we shall try to establish a better range of the values of $x$ for which Ramanujan’s inequality does hold true.

5 A modified Bound for $\pi(x)$

The derivations which we’ve made in the previous sections yielded several order estimates involving $M(x)$ , $\psi(x)$ and especially $\pi(x)$ . In this section, we shall discuss how this method enables us to find more optimal bounds for $\pi(x)$ .

5.1 Upper Bound for $\pi(x)$

A priori from Lemma (3.2.1), we have that for some positive constant $\alpha>0$ ,

\displaystyle\psi(x)\leq x+\alpha x^{1/2}(\log x)^{2}

(5.1)

Substituting this into the the estimate (3.9) of Theorem (3.2.2) involving $\pi(x)$ ,

\displaystyle\pi(x)\leq\frac{x+\alpha x^{1/2}(\log x)^{2}}{\log x}+O\left(% \frac{x}{(\log x)^{2}}\right)=\frac{x}{\log x}+\alpha x^{1/2}\log x+O\left(% \frac{x}{(\log x)^{2}}\right).

(5.2)

5.2 Lower Bound for $\pi(x)$

A priori using the estimate, $M(x)=O(x^{1/2})$ ,

\displaystyle\sum_{n\leq x}M\left(\frac{x}{n}\right)=O\left(\sum_{n\leq x}% \left(\frac{x}{n}\right)^{1/2}\right)=O\left(x^{1/2}\sum_{n\leq x}\frac{1}{n^{% 1/2}}\right).

Using the approximation, $\sum\limits_{n\leq x}\frac{1}{n^{1/2}}\sim 2\sqrt{x}$ gives,

\displaystyle\sum_{n\leq x}M\left(\frac{x}{n}\right)=O(x).

Thus,

\displaystyle\frac{1}{\log x}\sum_{n\leq x}M\left(\frac{x}{n}\right)=O\left(% \frac{x}{\log x}\right).

Therefore, applying the above derivation to (3.14) in Theorem (3.3.1), we assert that,

\displaystyle\pi(x)\geq\frac{x}{\log x}-\frac{\beta x}{\log x}+O\left(\frac{x}% {(\log x)^{2}}\right)=\left(1-\beta\right)\frac{x}{\log x}+O\left(\frac{x}{(% \log x)^{2}}\right).

(5.3)

for some positive constant $\beta>0$ . Hence, (5.2) and (5.3) allows us to state the following.

Theorem 5.2.1.

The following bounds on $\pi(x)$ is valid for sufficiently large values of $x$ :

\displaystyle\left(1-\beta\right)\frac{x}{\log x}+O\left(\frac{x}{(\log x)^{2}% }\right)\leq\pi(x)\leq\frac{x}{\log x}+\alpha x^{1/2}\log x+O\left(\frac{x}{(% \log x)^{2}}\right)

(5.4)

for positive constants $\alpha,\beta>0$ .

Remark 5.2.2.

The result (5.4) in Theorem (5.2.1) provides an alternative justification to the validity of the famous Prime Number Theorem. This can be observed from the fact that, $\pi(x)$ is bounded on both sides by a constant multiple of $\frac{x}{\log x}$ plus an error term, which can be minimized for large values of $x$ .

6 An improved condition for Ramanujan’s Inequality

A priori adopting Sterneck’s deduction [5], as mentioned in section $2.4$ of this text, we shall investigate the function $\mathcal{G}$ for its sign for large values of $x$ , with every intention of improving the claim made by Dudek and Platt [2, ref. Theorem 2].

6.1 Monotonicity of the function $\mathcal{G}$

Our aim in this section is to establish the following claim.

Proposition 6.1.1.

The function $\mathcal{G}(x)$ as defined in (4.1) is monotone decreasing for $x\geq 201$ .

Proof.

We intend on verifying that, $\mathcal{G}(x+\epsilon)-\mathcal{G}(x)<0$ for every large $x$ and for every $\epsilon>0$ arbitraily chosen.

Observe from definition that the difference,

\displaystyle\mathcal{G}(x+\epsilon)-\mathcal{G}(x)=\left[(\pi(x+\epsilon))^{2% }-(\pi(x))^{2}\right]-\left[\frac{e(x+\epsilon)}{\log(x+\epsilon)}\pi\left(% \frac{x+\epsilon}{e}\right)-\frac{ex}{\log x}\pi(x/e)\right]

(6.1)

Now we evaluate,

\displaystyle(\pi(x+\epsilon))^{2}-(\pi(x))^{2}=\left(\pi(x+\epsilon)-\pi(x)% \right)\left(\pi(x+\epsilon)+\pi(x)\right)

(6.2)

Using the explicit forms (3.14):

\displaystyle\pi(x+\epsilon)=\frac{x}{\log x}+\frac{\epsilon}{\log x}-\frac{x% \epsilon}{(\log x)^{2}}-\frac{1}{\log x}\sum_{n\leq x+\epsilon}M\left(\frac{x+% \epsilon}{n}\right)+O\left(\frac{x+\epsilon}{(\log(x+\epsilon))^{2}}\right)

Therefore,

\displaystyle\pi(x+\epsilon)-\pi(x)=\frac{\epsilon}{\log x}-\frac{x\epsilon}{(% \log x)^{2}}-\frac{1}{\log x}\sum_{n\leq x+\epsilon}M\left(\frac{x+\epsilon}{n% }\right)+\frac{1}{\log x}\sum_{n\leq x}M\left(\frac{x}{n}\right)

\displaystyle\hskip 300.0pt+O\left(\frac{x+\epsilon}{(\log(x+\epsilon))^{2}}-% \frac{x}{(\log x)^{2}}\right)

(6.3)

For small $\epsilon$ , we use the linear approximation:

\displaystyle\frac{x+\epsilon}{\log(x+\epsilon)}\approx\frac{x}{\log x}+\frac{% \epsilon}{\log x}-\frac{x\epsilon}{(\log x)^{2}}

And,

\displaystyle\frac{1}{\log(x+\epsilon)}\approx\frac{1}{\log x}-\frac{\epsilon}% {x(\log x)^{2}}

Thus, we can compute further in (6.3) as follows,

\displaystyle\pi(x+\epsilon)-\pi(x)=\frac{\epsilon}{\log x}-\frac{x\epsilon}{(% \log x)^{2}}-\frac{1}{\log x}\sum_{n\leq x+\epsilon}M\left(\frac{x+\epsilon}{n% }\right)+\frac{1}{\log x}\sum_{n\leq x}M\left(\frac{x}{n}\right)+O\left(\frac{% x\epsilon}{(\log x)^{2}}\right)

(6.4)

Similarly,

\displaystyle\pi(x+\epsilon)+\pi(x)=2\frac{x}{\log x}+\frac{\epsilon}{\log x}-% \frac{x\epsilon}{(\log x)^{2}}-\frac{1}{\log x}\sum_{n\leq x+\epsilon}M\left(% \frac{x+\epsilon}{n}\right)

\displaystyle\hskip 200.0pt+\frac{1}{\log x}\sum_{n\leq x}M\left(\frac{x}{n}% \right)+O\left(\frac{x\epsilon}{(\log x)^{2}}\right)

(6.5)

Combining (6.4) and (6.5) yields,

\displaystyle(\pi(x+\epsilon))^{2}-(\pi(x))^{2}=\left(\frac{\epsilon}{\log x}-% \frac{x\epsilon}{(\log x)^{2}}-\frac{1}{\log x}\sum_{n\leq x+\epsilon}M\left(% \frac{x+\epsilon}{n}\right)+\frac{1}{\log x}\sum_{n\leq x}M\left(\frac{x}{n}% \right)\right)

\displaystyle\hskip 200.0pt\left(2\frac{x}{\log x}+\frac{\epsilon}{\log x}-% \frac{x\epsilon}{(\log x)^{2}}\right)+O\left(\frac{x\epsilon}{(\log x)^{2}}\right)

\displaystyle=2\frac{x\epsilon}{(\log x)^{2}}-2\frac{x^{2}\epsilon}{(\log x)^{% 3}}-2\frac{x}{(\log x)^{2}}\sum_{n\leq x+\epsilon}M\left(\frac{x+\epsilon}{n}% \right)+2\frac{x}{(\log x)^{2}}\sum_{n\leq x}M\left(\frac{x}{n}\right)+O\left(% \frac{x\epsilon}{(\log x)^{2}}\right)

(6.6)

As for the second term in (6.1),

\displaystyle\frac{e(x+\epsilon)}{\log(x+\epsilon)}\pi\left(\frac{x+\epsilon}{% e}\right)-\frac{ex}{\log x}\pi(x/e)\approx\frac{e(x+\epsilon)}{\log(x+\epsilon% )}\left(\frac{x+\epsilon}{e\log(x+\epsilon/e)}\right)-\frac{ex}{\log x}\left(% \frac{x}{e\log(x/e)}\right)

[Using the earlier approximations for $\pi\left(\frac{x+\epsilon}{e}\right)$ and $\pi\left(\frac{x}{e}\right)$ ]

\displaystyle=\frac{(x+\epsilon)^{2}}{(\log(x+\epsilon))^{2}}-\frac{x^{2}}{(% \log x)^{2}}+O\left(\frac{x\epsilon}{(\log x)^{3}}\right)=\frac{(x+\epsilon)^{% 2}-x^{2}}{(\log x)^{2}}+O\left(\frac{x\epsilon}{(\log x)^{3}}\right)

\displaystyle\hskip 200.0pt=\frac{2x\epsilon+\epsilon^{2}}{(\log x)^{2}}+O% \left(\frac{x\epsilon}{(\log x)^{3}}\right)

(6.7)

Hence, substituting (6.6) and (6.7) in (6.1) and simplifying,

\displaystyle\mathcal{G}(x+\epsilon)-\mathcal{G}(x)=\left(2\frac{x\epsilon}{(% \log x)^{2}}-2\frac{x^{2}\epsilon}{(\log x)^{3}}-2\frac{x}{(\log x)^{2}}\sum_{% n\leq x+\epsilon}M\left(\frac{x+\epsilon}{n}\right)+2\frac{x}{(\log x)^{2}}% \sum_{n\leq x}M\left(\frac{x}{n}\right)\right)

\displaystyle\hskip 300.0pt-\left(\frac{2x\epsilon+\epsilon^{2}}{(\log x)^{2}}% \right)+O\left(\frac{x\epsilon}{(\log x)^{3}}\right)

\displaystyle=-2\frac{x}{(\log x)^{2}}\left(\sum_{n\leq x+\epsilon}M\left(% \frac{x+\epsilon}{n}\right)-\sum_{n\leq x}M\left(\frac{x}{n}\right)\right)-% \frac{\epsilon^{2}}{(\log x)^{2}}+O\left(\frac{x\epsilon}{(\log x)^{3}}\right)

(6.8)

We utilize Sterneck’s conjecture,

\displaystyle|M(x)|<\frac{1}{2}\sqrt{x},\hskip 10.0pt\mbox{ for }x\geq 201.

to bound the difference as follows,

\displaystyle-2\frac{x}{(\log x)^{2}}\left(\sum_{n\leq x+\epsilon}M\left(\frac% {x+\epsilon}{n}\right)-\sum_{n\leq x}M\left(\frac{x}{n}\right)\right)\leq-2% \frac{x}{(\log x)^{2}}\left(\frac{1}{2}\sqrt{x+\epsilon}-\frac{1}{2}\sqrt{x}\right)

(6.9)

Subsequently, $\epsilon>0$ enables us to conclude from (6.8) and (6.9),

\displaystyle\mathcal{G}(x+\epsilon)-\mathcal{G}(x)\leq-2\frac{x}{(\log x)^{2}% }\left(\frac{1}{2}(\sqrt{x+\epsilon}-\sqrt{x})\right)-\frac{\epsilon^{2}}{(% \log x)^{2}}+O\left(\frac{x\epsilon}{(\log x)^{3}}\right)

\displaystyle\hskip 50.0pt=-\frac{x}{(\log x)^{2}}(\sqrt{x+\epsilon}-\sqrt{x})% -\frac{\epsilon^{2}}{(\log x)^{2}}+O\left(\frac{x\epsilon}{(\log x)^{3}}\right)

(6.10)

Since $\sqrt{x+\epsilon}-\sqrt{x}>0$ for any $\epsilon>0$ and the error term $O\left(\frac{x\epsilon}{(\log x)^{3}}\right)$ is smaller, it implies,

\displaystyle\mathcal{G}(x+\epsilon)-\mathcal{G}(x)<0\text{ for every }x+% \epsilon>x,\epsilon>0

Thus, $\mathcal{G}(x)$ is monotone decreasing for large $x$ such that, $x\geq 201$ .

∎

Remark 6.1.2.

Note that, Sterneck’s Conjecture only establishes the fact that,

\displaystyle|M(n)|<\frac{1}{2}\sqrt{n},\hskip 10.0pt\mbox{ for }n>200.

where, $n\in\mathbb{N}$ . But in this case, since, we’re dealing with $x\in\mathbb{R}$ , hence, we’ve modified the lower bound for $x$ accordingly.

6.2 A better range for the values of $x$

Just to recall, Proposition (6.1.1) comments on the monotonicity of $\mathcal{G}$ for $x$ within a certain interval. Thus for every sufficiently large $x\geq 201$ , we must have,

\displaystyle\mathcal{G}(x)<\mathcal{G}\left(\frac{\log x}{e^{2}}\right)<0

provided, $\frac{\log x}{e^{2}}\geq 201$ , i.e., $\log x\geq 1485.2002758851$ . Therefore, we have our following improved bound on $x$ in order to satisfy (1.1).

Theorem 6.2.1.

The Ramanujan’s Inequality (1.1) is unconditionally true for every $x\geq\exp(1486)$ .

Acknowledgments

I’ll always be grateful to Prof. Adrian W. Dudek ( Adjunct Associate Professor, Department of Mathematics and Physics, University of Queensland, Australia ) for inspiring me to work on this problem and pursue research in this topic. His leading publications in this area helped me immensely in detailed understanding of the essential concepts.

References

[1] Ramanujan Aiyangar, Srinivasa, and Bruce C Berndt, Ramanujan’s Notebooks: Part IV, New York: Springer-Verlag, 1994.
[2] Adrian W. Dudek, David J. Platt, On Solving a Curious Inequality of Ramanujan, Experimental Mathematics, 24:3, 289-294, DOI: 10.1080/10586458.2014.990118, 2015.
[3] E. C. Titchmarsh, The Theory of the Riemann Zeta-function, Oxford University Press, 1951. Second edition revised by D. R. Heath-Brown, published by Oxford University Press, 1986.
[4] F. Mertens, $\ddot{U}$ ber eine zahlentheoretische Funktion, Sitzungsber. Akad. Wiss. Wien 106(IIa) (1897) 761-830 .
[5] R. D. von Sterneck, Die zahlentheoretische Funktion $\sigma(n)$ bis zur Grenze $5000000$ , Sitzungsber. Akad. Wiss. Wien 121(IIa) (1912) 1083-1096 .
[6] De, Subham, ”On the Order Estimates for Specific Functions of $\zeta(s)$ and its Contribution towards the Analytic Proof of The Prime Number Theorem.”, arXiv preprint arXiv:2308.16303 (2023).
[7] L. Schoenfeld, Sharper bounds for the Chebyshev Functions $\theta(x)$ and $\psi(x)$ . II, Mathematics of Computation, vol. 30, no. 134, pp. 337-360; DOI: 10.2307/2005976.
[8] A. E. Ingham, The distribution of prime numbers, Cambridge University Press, 1932. Reprinted by Stechert-Hafner, 1964, and (with a foreword by R. C. Vaughan) by Cambridge University Press, 1990.
[9] R. A. MacLeod, A new estimate for the sum $M(x)=\sum\limits_{n\leq x}\mu(n)$ , Acta Arith. 13 (1967) 49-59. Erratum, ibid. 16 (1969) 99-100.
[10] F. Dress, Majorations de la fonction sommatoire de la fonction de M $\ddot{o}$ bius, Bull. Soc. Math. Fr., Suppl., Mem. 49-50 (1977) 47-52.
[11] H. G. Diamond, K. S. McCurley, Constructive elementary estimates for $M(x)$ , In M. I. Knopp, editor, Analytic Number Theory, Lecture Notes in Mathematics 899, pp. 239-253, Springer, 1982.
[12] N. Costa Pereira, Elementary estimate for the Chebyshev function $\psi(x)$ and the M $\ddot{o}$ bius function $M(x)$ , Acta Arith. 52 (1989) 307-337.
[13] F. Dress, M. El Marraki, Fonction sommatoire de la fonction de Mobius. 2, Majorations asymptotiques elementaires, Exp. Math. 2 (1993) 99-112.
[14] E. Landau, Handbuch der Lehre von der Verteilung der Primzahlen, Vol. 2 (of 2), Teubner, 1909. Reprinted by Chelsea, 1953.
[15] De, Subham, ”On the Order Estimate of the Mertens Function and its Relation to the Zeros of the Riemann Zeta Function”, Preprints 2023, 2023090723. https://doi.org/10.20944/preprints202309.0723.v1
[16] T. Kotnik, J. van de Lune, On the order of the Mertens function, experimental mathematics, Vol. 13 (2004), pp. 473-481.
[17] T. M. Apostol, Introduction to Analytic Number Theory, Springer, 1976.
[18] G.J.O. Jameson, The Prime Number Theorem, Cambridge ; New York : Cambridge University Press, London Mathematical Society student texts, Vol. 53, 2003.
[19] Olver, F. W. J., Asymptotics and Special Functions, Academic Press, New York, 1974.
[20] De, Subham, On the proof of the Prime Number Theorem using Order Estimates for the Chebyshev Theta Function, International Journal of Science and Research (IJSR), Vol. 12 Issue 11, Nov. 2023, pp. 1677-1691.
[21] L Ahlfors, Complex Analysis, McGraw-Hill Education , $3^{rd}$ edition , Jan. 1, 1979 .
[22] Mehdi Hassani, “On an Inequality of Ramanujan Concerning the Prime Counting Function”, Ramanujan Journal 28 (2012), 435–442.
[23] Ramanujan, S, “Collected Papers”, Chelsea, New York, 1962.
[24] Hardy, G. H., A formula of Ramanujan in the theory of primes, 1. London Math. Soc. 12 (1937), 94-98.
[25] Hardy, G. H., Collected Papers, vol. II, Clarendon Press, Oxford, 1967.
[26] Axler, Christian, On Ramanujan’s prime counting inequality,arXiv preprint arXiv:2207.02486 (2022).

On proving an Inequality of Ramanujan using Explicit Order Estimates of the Mertens Function

Abstract

1 Introduction and Motivation

Theorem 1.0.1.

2 Arithmetic Functions

Definition 2.0.1.

Definition 2.0.2.

2.1 Prime Counting Function

Definition 2.1.1.

Theorem 2.1.1.

2.2 Chebyshev Function

Definition 2.2.1.

2.3 Mo¨¨𝑜\ddot{o}over¨ start_ARG italic_o end_ARGbius Function

Definition 2.3.1.

Proposition 2.3.1.

2.4 Mertens Function

Definition 2.4.1.

Remark 2.4.1.

2.5 Some Necessary Derivations

Proposition 2.5.1.

Proposition 2.5.2.

Proof.

3 Order Estimates involving M⁢(x)𝑀𝑥M(x)italic_M ( italic_x )

3.1 Relation between ψ⁢(x)𝜓𝑥\psi(x)italic_ψ ( italic_x ) and M⁢(x)𝑀𝑥M(x)italic_M ( italic_x )

Theorem 3.1.1.

Proof.

Remark 3.1.2.

3.2 An important approximation for ψ⁢(x)𝜓𝑥\psi(x)italic_ψ ( italic_x )

Lemma 3.2.1.

Proof.

Theorem 3.2.2.

Proof.

3.3 An improved estimate for π⁢(x)𝜋𝑥\pi(x)italic_π ( italic_x )

Theorem 3.3.1.

4 Proving Ramanujan’s Inequality

4.1 An order expression for 𝒢⁢(x)𝒢𝑥\mathcal{G}(x)caligraphic_G ( italic_x )

4.2 Asymptotic Behavior of indivudual Terms

5 A modified Bound for π⁢(x)𝜋𝑥\pi(x)italic_π ( italic_x )

5.1 Upper Bound for π⁢(x)𝜋𝑥\pi(x)italic_π ( italic_x )

5.2 Lower Bound for π⁢(x)𝜋𝑥\pi(x)italic_π ( italic_x )

Theorem 5.2.1.

Remark 5.2.2.

6 An improved condition for Ramanujan’s Inequality

6.1 Monotonicity of the function 𝒢𝒢\mathcal{G}caligraphic_G

Proposition 6.1.1.

Proof.

Remark 6.1.2.

6.2 A better range for the values of x𝑥xitalic_x

Theorem 6.2.1.

Acknowledgments

References

2.3 M $\ddot{o}$ bius Function

3 Order Estimates involving $M(x)$

3.1 Relation between $\psi(x)$ and $M(x)$

3.2 An important approximation for $\psi(x)$

3.3 An improved estimate for $\pi(x)$

4.1 An order expression for $\mathcal{G}(x)$

5 A modified Bound for $\pi(x)$

5.1 Upper Bound for $\pi(x)$

5.2 Lower Bound for $\pi(x)$

6.1 Monotonicity of the function $\mathcal{G}$

6.2 A better range for the values of $x$