The Prime Number Theorem, Part 3: Riemann zeta function

January 10, 2026 Leave a comment

Another way to see that \psi(x) is a natural function to consider is through its deep connection with the Riemann zeta function.

In this section, we will prove the following standard facts about the zeta function.

  1. The series defining

    \displaystyle \zeta(s):=\sum_{n=1}^\infty\frac{1}{n^s}

    converges for \mathrm{Re}(s)>1, and the function \zeta is holomorphic in this half-plane.

  2. For \mathrm{Re}(s)>1, we have the Euler product formula

    \displaystyle \zeta(s) = \prod_{p} \frac{1}{1-p^{-s}},

    where the product extends over all primes p. This identity reveals that \zeta(s) encodes the multiplicative structure of the integers and, in particular, the distribution of the primes.

  3. For \mathrm{Re}(s)>1, we have the logarithmic derivative

    \displaystyle -\frac{\zeta'(s)}{\zeta(s)} = \sum_{n=1}^\infty \frac{\Lambda(n)}{n^s},

    where \Lambda(n) is the von Mangoldt function. Thus the arithmetic information carried by \Lambda(n) is directly reflected in the analytic behavior of \zeta(s). The series is a Dirichlet series with nonnegative coefficients, which will be important in later arguments.

  4. By the inverse Mellin transform, for x>0 and c>1, we have

    \displaystyle \frac{1}{2\pi i} \int_{c-i\infty}^{c+i\infty} - \frac{\zeta'(s)}{\zeta(s)} \, \frac{x^s}{s} \, ds = \sum_{n \le x} \Lambda(n) = \psi(x).

    This identity expresses \psi(x) as a complex integral whose behavior is governed by the analytic properties of \zeta(s).

Definition of the Zeta Function

Proof. If s=\sigma+i t where \sigma and t are real, then

\displaystyle \left|n^{-s}\right|=\left|e^{-s \log n}\right|=e^{-\sigma \log n}=n^{-\sigma} .

As a consequence, if \sigma>1+\delta>1 the series defining \zeta is uniformly bounded by \sum_{n=1}^{\infty} 1 / n^{1+\delta}, which converges. Therefore, the series \sum 1 / n^s converges uniformly on every half-plane \mathrm{Re}(s)>1+\delta>1, and therefore defines a holomorphic function in \mathrm{Re}(s)>1. □

The Euler Product Formula

Proof. Assume \mathrm{Re}(s)>1. Fix positive integers M and N with M>N.

Define the set of primes

\displaystyle P_N := \{p \text{ prime} : p \le N\},

and the exponent box

\displaystyle E_{N,M} := \{(a_p)_{p\in P_N} : 0 \le a_p \le M\}.

For each exponent vector a=(a_p)\in E_{N,M}, define

\displaystyle \Phi(a) := \prod_{p\in P_N} p^{a_p}.

By the uniqueness part of the fundamental theorem of arithmetic, the map

\displaystyle \Phi : E_{N,M} \to \mathbb{N}

is injective.

Now let n be a positive integer with n \le N. By the fundamental theorem of arithmetic,

\displaystyle n = \prod_{p\le N} p^{v_p(n)},

since no prime larger than N can divide n\le N. Moreover, if p^{v_p(n)} \mid n \le N, then

\displaystyle p^{v_p(n)} \le N,

which implies v_p(n) < N < M. Hence

\displaystyle (v_p(n))_{p\in P_N} \in E_{N,M}.

Therefore,

\displaystyle \{1,2,\dots,N\} \subseteq \Phi(E_{N,M}).

Since the function x \mapsto x^{-s} is positive for \mathrm{Re}(s)>0, this set inclusion implies

\displaystyle \sum_{n=1}^N \frac{1}{n^s} \;\le\; \sum_{m\in \Phi(E_{N,M})} \frac{1}{m^s}.

On the other hand, expanding the finite product

\displaystyle \prod_{p \le N}\left(1+\frac{1}{p^s}+\frac{1}{p^{2s}}+\cdots+\frac{1}{p^{Ms}}\right)

corresponds to choosing an exponent a_p\in\{0,1,\dots,M\} for each p\le N, and hence yields

\displaystyle \begin{aligned} \prod_{p \le N}\left(1+\frac{1}{p^s}+\cdots+\frac{1}{p^{Ms}}\right) &= \sum_{a\in E_{N,M}} \prod_{p\le N} p^{-a_p s}\\ &= \sum_{m\in \Phi(E_{N,M})} \frac{1}{m^s}, \end{aligned}

where injectivity of \Phi ensures that each integer appears exactly once.

Combining the above inequalities, we obtain

\displaystyle \sum_{n=1}^N \frac{1}{n^s} \le \prod_{p \le N}\left(1+\frac{1}{p^s}+\frac{1}{p^{2 s}}+\cdots+\frac{1}{p^{M s}}\right).

Since

\displaystyle 1+\frac{1}{p^s}+\cdots+\frac{1}{p^{Ms}} \le \sum_{k=0}^\infty p^{-ks} = \frac{1}{1-p^{-s}},

we further obtain

\displaystyle \sum_{n=1}^N \frac{1}{n^s} \le \prod_{p \le N}\left(\frac{1}{1-p^{-s}}\right) \le \prod_p \left(\frac{1}{1-p^{-s}}\right).

Letting N\to\infty yields

\displaystyle \sum_{n=1}^{\infty} \frac{1}{n^s} \le \prod_p \left(\frac{1}{1-p^{-s}}\right).

For the reverse inequality, note that for fixed N and M, the same expansion shows

\displaystyle \prod_{p \le N}\left(1+\frac{1}{p^s}+\frac{1}{p^{2 s}}+\cdots+\frac{1}{p^{M s}}\right) = \sum_{m\in \Phi(E_{N,M})} \frac{1}{m^s} \le \sum_{n=1}^{\infty} \frac{1}{n^s},

since \Phi(E_{N,M}) \subseteq \mathbb{N}.

Letting M\to\infty and using monotone convergence gives

\displaystyle \prod_{p \le N}\left(\frac{1}{1-p^{-s}}\right) \le \sum_{n=1}^{\infty} \frac{1}{n^s}\;.

Taking N\to\infty completes the proof. □

The Logarithmic Derivative of the Zeta Function

We work in the half-plane \mathrm{Re}(s)>1, where the Euler product

\displaystyle \zeta(s)=\prod_p \frac{1}{1-p^{-s}}

converges absolutely.

Lemma 4. If \mathrm{Re}(s)>1, then

\displaystyle \log \zeta(s) = \sum_{p,m} \frac{p^{-ms}}{m} = \sum_{n=1}^{\infty} c_n n^{-s},

where c_n \ge 0.

Proof. Suppose first that s is real and s>1. Taking the logarithm of the Euler product and using the power series expansion

\displaystyle \log\!\left(\frac{1}{1-x}\right) = \sum_{m=1}^{\infty}\frac{x^m}{m}, \qquad 0\le x<1,

we obtain

\displaystyle \log \zeta(s) = \log \prod_p \frac{1}{1-p^{-s}} = \sum_p \log\!\left(\frac{1}{1-p^{-s}}\right) = \sum_{p,m} \frac{p^{-ms}}{m}.

Since the double sum converges absolutely, the order of summation need not be specified. The identity therefore extends to all \mathrm{Re}(s)>1 by analytic continuation. Moreover, the function \log \zeta(s) is well defined in the simply connected half-plane \mathrm{Re}(s)>1, since \zeta(s) has no zeros there.

Finally, the sum may be rewritten as a Dirichlet series over the integers:

\displaystyle \sum_{p,m} \frac{p^{-ms}}{m} = \sum_{n=1}^{\infty} c_n n^{-s},

where c_n = 1/m if n=p^m for some prime p and integer m\ge1, and c_n=0 otherwise. □

From the lemma, we have

\displaystyle \log \zeta(s) = \sum_{p,m} \frac{p^{-ms}}{m}.

Taking derivatives (using for instance the generalised Cauchy integral formula and Fubini’s theorem to justify the interchange of summation and derivative), we see that

\displaystyle \frac{\zeta'(s)}{\zeta(s)} = -\sum_{p,m} (\log p)\, p^{-ms}.

Rewriting the right-hand side yields

\displaystyle -\frac{\zeta'(s)}{\zeta(s)} = \sum_{n=1}^{\infty} \frac{\Lambda(n)}{n^s}, \qquad \mathrm{Re}(s)>1.

Perron’s Formula

The logarithmic derivative of the Riemann zeta function connects \zeta(s) to the Chebyshev function \psi(x) through Perron’s formula, which may be viewed as a special case of the Mellin inversion formula:

\displaystyle \psi(x) := \sum_{n \le x} \Lambda(n) = \frac{1}{2\pi i} \int_{c-i\infty}^{c+i\infty} - \frac{\zeta'(s)}{\zeta(s)} \, \frac{x^s}{s} \, ds, \qquad (c>1,\ x>0).

Rather than attempting to analyze the asymptotic behavior of \psi(x) directly from this integral, it is more convenient to study its integral

\displaystyle \psi_1(x) := \int_0^x \psi(t)\,dt.

Formally interchanging the order of summation and integration yields

\displaystyle \psi_1(x) = \frac{1}{2\pi i} \int_{c-i\infty}^{c+i\infty} - \frac{\zeta'(s)}{\zeta(s)} \, \frac{x^{s+1}}{s(s+1)} \, ds, \qquad (c>1).

This additional factor \frac{1}{s(s+1)} improves convergence and isolates the dominant contribution arising from the pole at s=1.

At the stage where Perron’s formula is introduced, the proof of the Prime Number Theorem relies on a small number of simple but essential principles. Making these explicit clarifies the contour deformations and estimates that appear later in the argument.

Infinite contours as limits of finite contours

Integrals along infinite contours are always interpreted as limits of integrals along finite contours. That is,

\displaystyle \int_{c-i\infty}^{c+i\infty} F(s)\,ds = \lim_{T\to\infty} \int_{c-iT}^{c+iT} F(s)\,ds.

All contour deformations and estimates are first carried out for the truncated contours with finite height T, and only afterward is the limit T \to \infty taken. This includes applying Cauchy’s theorem or the residue theorem to finite contours, estimating integrals along auxiliary segments, and justifying the vanishing of error terms as T \to \infty.

A deformation interpretation of the residue theorem (single pole case)

Let F be a meromorphic function with a single pole at s=s_0.

The classical formulation of the residue theorem states that if \Gamma is a positively oriented closed contour enclosing s_0 and no other singularities, then

\displaystyle \int_{\Gamma} F(s)\,ds = 2\pi i \mathrm{Res}_{s=s_0} F.

For applications such as the proof of the Prime Number Theorem, it is often more illuminating to reinterpret this result in terms of contour deformation.

Suppose that two oriented contours \gamma_1 and \gamma_2 together form the boundary of a region containing the pole s_0, and that F is holomorphic in the region between them except at s_0. Writing the closed contour as

\displaystyle \Gamma = \gamma_1 - \gamma_2,

the residue theorem may be rewritten as

\displaystyle \int_{\gamma_1} F(s)\,ds = \int_{\gamma_2} F(s)\,ds + 2\pi i \mathrm{Res}_{s=s_0} F.

From this point of view, the residue theorem provides a mechanism for replacing an integral along one contour by an integral along another, at the cost of an explicit residue term arising from the pole crossed during the deformation.

Cauchy’s theorem appears as the special case in which F has no poles in the region, so that

\displaystyle \int_{\gamma_1} F(s)\,ds = \int_{\gamma_2} F(s)\,ds.

Zeros and poles determine the singularities of the logarithmic derivative

Let G(s) be a meromorphic function that is not identically zero. Its logarithmic derivative

\displaystyle \frac{G'(s)}{G(s)}

is again meromorphic, and its singularities are completely determined by the zeros and poles of G.

More precisely, suppose that s_0 is a zero of G of order m \ge 1. Then locally one may write

\displaystyle G(s) = (s-s_0)^m H(s),

where H(s) is holomorphic and nonvanishing near s_0. Taking logarithms and differentiating yields

\displaystyle \frac{G'(s)}{G(s)} = \frac{m}{s-s_0} + \frac{H'(s)}{H(s)},

so \frac{G'}{G} has a simple pole at s_0 with residue m.

Similarly, if s_0 is a pole of G of order m, then \frac{G'}{G} has a simple pole at s_0 with residue -m.

Thus the logarithmic derivative \frac{G'}{G} has only simple poles, located precisely at the zeros and poles of G, and the residue at each such pole records the corresponding multiplicity.

Returning now to the integral representation of \psi_1(x),

\displaystyle \psi_1(x) = \frac{1}{2\pi i} \int_{c-i\infty}^{c+i\infty} - \frac{\zeta'(s)}{\zeta(s)} \, \frac{x^{s+1}}{s(s+1)} \, ds, \qquad (c>1),

our goal is to move the line of integration from \mathrm{Re}(s)=c to the line \mathrm{Re}(s)=1. On the latter line, the factor x^{s+1} has magnitude x^2, whereas on \mathrm{Re}(s)=c it grows like x^{c+1}, which is too large for asymptotic purposes.

At this point such a deformation is not yet justified, since \zeta(s) and its logarithmic derivative are initially defined only in the half-plane \mathrm{Re}(s)>1. To proceed, we require their analytic continuation.

We now record several standard facts concerning the analytic continuation and zeros of the Riemann zeta function.

Theorem 5. The zeta function admits a meromorphic continuation to the entire complex plane, with a single singularity: a simple pole at s=1.

Theorem 6. The only zeros of \zeta(s) outside the strip 0 \le \mathrm{Re}(s) \le 1 occur at the negative even integers -2,-4,-6,\ldots.

The region 0 \le \mathrm{Re}(s) \le 1 is called the critical strip. A central ingredient in the proof of the Prime Number Theorem is the fact that \zeta(s) has no zeros on the line \mathrm{Re}(s)=1. As a consequence of this fact and the functional equation for \zeta(s), it follows that \zeta(s) has no zeros on the line \mathrm{Re}(s)=0.

Theorem 7. The zeta function has no zeros on the line \mathrm{Re}(s)=1.

Theorem 8. For \mathrm{Re}(s)>0 we have

\displaystyle \zeta(s) = \frac{1}{s-1} + H(s),

where H(s) is holomorphic in the half-plane \mathrm{Re}(s)>0.

Theorem 9. For any \eta>0, we have

\displaystyle |\zeta'(s)/\zeta(s)| \le A |t|^{\eta}

whenever s=\sigma + it with \sigma \ge 1 and |t| \ge 1, where A=A(\eta) is a positive constant depending only on \eta.

Back to the proof. We now deform the line of integration in the formula for \psi_1(x) from \mathrm{Re}(s)=c>1 toward the line \mathrm{Re}(s)=1. The key point is that the integrand has no singularities in the region between these two lines except for the simple pole at s=1.

By the residue theorem, the difference between the two integrals is equal to 2\pi i times the residue of the integrand at s=1. A direct computation shows that this residue equals \frac{x^2}{2}.

Fix c>1, say c=2, and assume x \ge 2. Let

\displaystyle F(s) := - \frac{\zeta'(s)}{\zeta(s)} \, \frac{x^{s+1}}{s(s+1)}

denote the integrand. The proof proceeds through three stages of contour deformation:

  1. From Stage 1 to Stage 2, we justify deforming the infinite vertical line \mathrm{Re}(s)=c to the truncated contour \gamma(T). This requires decay estimates for F(s) along the horizontal segments s=\sigma \pm iR, \sigma \in [1,c], as R \to \infty:

    \displaystyle \frac{1}{2 \pi i} \int_{c-i \infty}^{c+i \infty} F(s)\,ds = \frac{1}{2 \pi i} \int_{\gamma(T)} F(s)\,ds.

  2. From Stage 2 to Stage 3, we apply the residue theorem at s=1 to obtain

    \displaystyle \frac{1}{2 \pi i} \int_{\gamma(T)} F(s)\,ds = \frac{x^2}{2} + \frac{1}{2 \pi i} \int_{\gamma(T,\delta)} F(s)\,ds.

  3. Finally, we estimate the integral along \gamma(T,\delta) by decomposing the contour into simpler segments and bounding each contribution.

 

 

 


From Stage 1 to Stage 2.

By Cauchy’s theorem, we have

\displaystyle \int_{\gamma(T)_R} F(s)\,ds + \int_{\gamma_T} F(s)\,ds + \int_{\gamma(T)_{-R}} F(s)\,ds = \int_{(c)_R} F(s)\,ds,

where \gamma(T)_R and \gamma(T)_{-R} denote the top and bottom horizontal segments, respectively.

To deform the contour from \mathrm{Re}(s)=c to \gamma(T), it suffices to show that

\displaystyle \lim_{R \to \infty} \int_{\gamma(T)_R} F(s)\,ds = 0, \qquad \lim_{R \to \infty} \int_{\gamma(T)_{-R}} F(s)\,ds = 0.

We first consider the upper horizontal segment \gamma(T)_R. Along this segment, we have s=\sigma+iR with \sigma \in [1,c], and hence

\displaystyle \int_{\gamma(T)_R} F(s)\,ds = \int_1^c F(\sigma+iR)\,d\sigma.

By Theorem 9, for any fixed \eta>0 and all R>T, there exists a constant A'>0 (depending on \eta and x) such that

\displaystyle |F(s)| \le A' |t|^{-2+\eta}

in the rectangular regions under consideration. Consequently,

\displaystyle |F(\sigma+iR)| \le A' R^{-2+\eta},

and therefore

\displaystyle \left|\int_{\gamma(T)_R} F(s)\,ds\right| \le \int_1^c |F(\sigma+iR)|\,d\sigma \le A'(c-1) R^{-2+\eta} \to 0 \quad\text{as } R\to\infty,

for example by taking \eta=1. The estimate for the lower segment \gamma(T)_{-R} is identical.


From Stage 2 to Stage 3.

To establish the identity via the residue theorem, we compute the residue of

\displaystyle F(s) = -\frac{\zeta'(s)}{\zeta(s)} \frac{x^{s+1}}{s(s+1)}

at s=1. By Theorem 8, the logarithmic derivative of \zeta has the expansion

\displaystyle -\frac{\zeta'(s)}{\zeta(s)} = \frac{1}{s-1} + H_1(s),

where H_1(s) is holomorphic in a neighborhood of s=1. Since the factor \frac{x^{s+1}}{s(s+1)} is also holomorphic at s=1, the residue of F(s) at s=1 is obtained by evaluating this factor at s=1. Hence,

\displaystyle \mathrm{Res}_{s=1} F(s) = \frac{x^{2}}{2}.

Stage 3: Estimating the integral along \gamma(T,\delta).

We decompose the contour \gamma(T,\delta) as

\displaystyle \gamma(T,\delta) = \gamma_1 + \gamma_2 + \gamma_3 + \gamma_4 + \gamma_5,

as shown in Figure 1, and estimate each integral \int_{\gamma_j} F(s)\,ds separately.

We first choose T sufficiently large. For the vertical rays \gamma_1 and \gamma_5, we claim that

\displaystyle \left|\int_{\gamma_1} F(s)\,ds\right| \le \frac{\epsilon}{2} x^2, \qquad \left|\int_{\gamma_5} F(s)\,ds\right| \le \frac{\epsilon}{2} x^2.

For s \in \gamma_5, we have s=1+it with t\ge T, so that

\displaystyle |x^{s+1}| = x^{1+\sigma} = x^2, \qquad \frac{1}{|s(s+1)|} \asymp \frac{1}{t^2}.

By Theorem 9, taking \eta=\tfrac12, there exists a constant A>0 such that

\displaystyle \left|\frac{\zeta'(s)}{\zeta(s)}\right| \le A |t|^{1/2}.

Hence,

\displaystyle \left|\int_{\gamma_5} F(s)\,ds\right| \le C x^2 \int_T^\infty \frac{t^{1/2}}{t^2}\,dt,

which converges and can be made smaller than \epsilon x^2/2 by choosing T sufficiently large. The estimate for \gamma_1 is identical.

Having fixed T, we next choose \delta>0 sufficiently small. On the vertical segment \gamma_3, where \mathrm{Re}(s)=1-\delta, we have

\displaystyle |x^{s+1}| = x^{2-\delta},

and therefore there exists a constant C_T (depending on T) such that

\displaystyle \left|\int_{\gamma_3} F(s)\,ds\right| \le C_T x^{2-\delta}.

Finally, on the horizontal segments \gamma_2 and \gamma_4, we obtain

\displaystyle \left|\int_{\gamma_2} F(s)\,ds\right| + \left|\int_{\gamma_4} F(s)\,ds\right| \le C_T' \int_{1-\delta}^1 x^{1+\sigma}\,d\sigma \le C_T' \frac{x^2}{\log x}.

Combining these estimates, we conclude that

\displaystyle \left|\psi_1(x)-\frac{x^2}{2}\right| \le \epsilon x^2 + C_T x^{2-\delta} + C_T' \frac{x^2}{\log x}.

Dividing by x^2/2, we obtain

\displaystyle \left|\frac{2\psi_1(x)}{x^2}-1\right| \le 2\epsilon + 2C_T x^{-\delta} + 2C_T' \frac{1}{\log x}.

Thus, for all sufficiently large x,

\displaystyle \left|\frac{2\psi_1(x)}{x^2}-1\right| \le 4\epsilon.

\gamma(T,\delta)

Read more…

Categories: expository, math.CA, math.NT Tags:

The Prime Number Theorem, Part 2: Reduction of the Prime Number Theorem

January 10, 2026 Leave a comment

Heuristically, the Prime Number Theorem predicts that primes near x occur with density about 1/\log x. If this is so, then the total contribution of the primes and their powers up to x, weighted by \log p, should be approximately

\displaystyle \sum_{n \le x} \Lambda(n) \;\approx\; \int_2^x \frac{\log t}{\log t}\, dt \;\approx\; x,

which suggests that the natural analytic formulation of the Prime Number Theorem is the asymptotic relation

\displaystyle \psi(x) \sim x.

This heuristic explains why \psi(x) is the correct object to study from the analytic point of view.

We now make this precise by showing that the Prime Number Theorem follows from the asymptotic behavior of \psi(x).

Theorem 2. If \psi(x) \sim x as x \to \infty, then

\displaystyle \pi(x) \sim \frac{x}{\log x} \qquad \text{as } x \to \infty.

Recall that the asymptotic notation f(x) \sim g(x) as x \to \infty means that

\displaystyle \lim_{x \to \infty} \frac{f(x)}{g(x)} = 1.

Proof. Suppose that

\displaystyle \lim_{x \to \infty} \frac{\psi(x)}{x} = 1.

We aim to prove that

\displaystyle \lim_{x \to \infty} \frac{\pi(x) \log x}{x} = 1.

Rather than estimating the difference

\displaystyle \left| \frac{\pi(x)\log x}{x} - 1 \right|

directly, we use upper and lower limits. It suffices to show that

\displaystyle \limsup_{x \to \infty} \frac{\pi(x)\log x}{x} \le 1 \qquad \text{and} \qquad \liminf_{x \to \infty} \frac{\pi(x)\log x}{x} \ge 1.

Since \liminf \le \limsup always holds, these two inequalities together imply the desired limit.

By the hypothesis, since \lim_{x\to\infty}\frac{\psi(x)}{x}=1, the strategy is to compare \frac{\pi(x)\log x}{x} to \frac{\psi(x)}{x}.

Recall the trivial upper bound for \psi(x):

\displaystyle \psi(x) = \sum_{p \le x} \left\lfloor \frac{\log x}{\log p} \right\rfloor \log p \le \sum_{p \le x} \frac{\log x}{\log p} \log p = \pi(x) \log x.

Dividing both sides by x and taking \liminf yields

\displaystyle 1 = \liminf_{x \to \infty} \frac{\psi(x)}{x} \le \liminf_{x \to \infty} \frac{\pi(x)\log x}{x},

To get the other bound, observe that for any fixed 0 < \alpha < 1, we have

\displaystyle \begin{aligned} \psi(x) &= \sum_{p^m \le x} \log p\ge \sum_{x^\alpha<p \le x} \log p \ge \big(\pi(x) - \pi(x^\alpha)\big) \log (x^\alpha) = \alpha \big(\pi(x) - \pi(x^\alpha)\big)\log x\\ &\ge \alpha \big(\pi(x) - x^\alpha\big)\log x \end{aligned}

where in the last inequality we use the simple bound \pi(y) \le y for all y \ge 2. It follows that

\displaystyle \frac{\pi(x)\log x}{x} \le \frac{\psi(x)}{\alpha x} + \frac{x^\alpha \log x}{x} = \frac{\psi(x)}{\alpha x} + \frac{\log x}{x^{1-\alpha}}.

Taking \limsup on both sides and using the hypothesis gives

\displaystyle \limsup_{x \to \infty} \frac{\pi(x)\log x}{x} \le \frac{1}{\alpha}+0.

Since this holds for all 0 < \alpha < 1, we may let \alpha \to 1^- to conclude that

\displaystyle \limsup_{x \to \infty} \frac{\pi(x)\log x}{x} \le 1

and the proof is complete. □

The Prime Number Theorem, Part 1: Introduction

January 4, 2026 Leave a comment

Define \pi(x) to be the number of primes less than or equal to x:

\displaystyle \pi(x) := \#\{p \le x : p \text{ is prime}\}.

The Prime Number Theorem asserts that

Theorem 1.

\displaystyle \pi(x) \sim \frac{x}{\log x} \qquad \text{as } x \to \infty.

At first glance, \pi(x) is the most natural object to study: it directly answers the question “how many primes are there up to x?” However, \pi(x) is difficult to analyze analytically. It is a step function that jumps by 1 at each prime, and it treats every prime equally:

\displaystyle \pi(x) = \sum_{p \le x} 1.

This additive viewpoint fails to reflect the multiplicative nature of the primes.

Indeed, primes are defined not by additive properties but by their role as the building blocks of multiplication. This multiplicative structure is made precise by the Fundamental Theorem of Arithmetic, which states that every integer n \ge 1 can be written uniquely as

\displaystyle n = \prod_p p^{v_p(n)} (1)

where v_p(n) is the exponent of the prime p in the prime factorization of n.

Taking logarithms converts this multiplicative decomposition into an additive one:

\displaystyle \log n = \sum_p v_p(n)\,\log p.

This identity suggests that the logarithm naturally “weights” primes according to how often they appear in the factorization of integers. In particular, primes should not all be counted equally; instead, they should be weighted by \log p, and higher powers of primes should also contribute.

This motivates the definition of the Chebyshev function

\displaystyle \psi(x) := \sum_{p^m \le x} \log p,

where the sum runs over all prime powers p^m not exceeding x.

Equivalently, \psi(x) may be written as a sum over primes:

\displaystyle \begin{aligned} \psi(x) &= \sum_{p \le x} \bigl(\text{number of prime powers } p^m \le x\bigr)\cdot\log p \\ &= \sum_{p \le x} \left\lfloor \frac{\log x}{\log p} \right\rfloor \log p, \end{aligned}

where \lfloor y \rfloor denotes the greatest integer less than or equal to y.

Thus, in \psi(x) each prime p contributes not just \log p once, but \log p multiplied by the number of powers of p that are less than or equal to x.

We have the following trivial observation connecting \psi(x) and \pi(x):

\displaystyle \psi(x) \le \sum_{p \le x} \frac{\log x}{\log p} \log p = \pi(x) \log x.

To better understand the structure of \psi(x), it is helpful to view it as consisting of two distinct components:

  1. the number of prime powers p^m \le x,

  2. the logarithmic weight \log p attached to each such power.

Note that both components are naturally motivated by the logarithmic factorization identity (1). Now consider how the prime p contributes across all integers up to x. Rather than summing v_p(n) over all n \le x directly, we isolate the basic multiplicative events associated with p, namely the appearances of its powers

\displaystyle p,\, p^2,\, p^3,\, \dots .

Each prime power p^m \le x represents one fundamental contribution of the prime p at scale p^m, and the total number of such powers is

\displaystyle \left\lfloor \frac{\log x}{\log p} \right\rfloor.

This leads naturally to counting, for each prime p, the number of its powers not exceeding x, and weighting each occurrence by \log p. Summing over all primes gives

\displaystyle \psi(x) = \sum_{p^m \le x} \log p = \sum_{p \le x} \bigl(\text{number of prime powers } p^m \le x\bigr)\,\log p.

It is not difficult to see that \psi(x) can also be expressed as a sum over all integers up to x:

\displaystyle \psi(x) = \sum_{p^m \le x} \log p=\sum_{n \le x} \Lambda(n),

where \Lambda(n) is the von Mangoldt function, defined by

\displaystyle \Lambda(n) = \begin{cases} \log p, & n = p^m \text{ for some prime } p,\\ 0, & \text{otherwise}. \end{cases}

We will see that the function \psi(x) is far more amenable to analytic methods than \pi(x). It is directly connected to the Riemann zeta function through its logarithmic derivative. As a result, the proof of the Prime Number Theorem proceeds not by analyzing \pi(x) directly, but by first understanding the asymptotic behavior of \psi(x).

Hello world!

August 12, 2010 Leave a comment

Welcome to WordPress.com. This is your first post. Edit or delete it and start blogging!

Categories: Uncategorized
Create your own website, free on WordPress.com.Create your own website.
Start free