Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

In Apostol's Analytic Number Theory, Apostol defines $$A(x):= \sum_{n \leq x} \frac{\mu(n)}{n}$$ and proves that $A(x)=o(1)$ implies the Prime Number Theorem, by showing that $$\frac{M(x)}{x}=A(x)-\frac{1}{x}\int_1^x A(t)dt,$$ in which $M(x):=\sum_{n \leq x} \mu(n)$ is the summatory function for the Möbius function (Theorem 4.16). What are some known error bounds for the function $A(x)$? In particular, do we have $A(x)= o(1/\log x)$ as $x \to \infty$?

share|cite|improve this question
Terence Tao has a rather interesting paper about this : 'A remark on partial sums involving the Mobius function'. See too his blog. – Raymond Manzoni Dec 7 '12 at 1:20
Schoenfeld's paper, Marraki's one and Cohen's paper seem interesting too (summatory function is much more studied than the $\zeta(1)$ formula). – Raymond Manzoni Dec 7 '12 at 1:25
up vote 5 down vote accepted

I'll answer my own question:

The Abel Summation formula gives

$$A(x)=\frac{M(x)}{x}+ \int_1^x \frac{M(u)}{u^2} du = \frac{M(x)}{x}+\int_1^\infty \frac{M(u)}{u^2} du-\int_x^\infty \frac{M(u)}{u^2} du.$$ As $A(x)=o(1)$, the right-hand side of the above must tend to $0$. We have $M(x)/x \to 0$, and the estimate $$\left\vert \int_x^\infty \frac{M(u)}{u^2} du \right\vert \leq \int_x^\infty \frac{\vert M(u)\vert}{x^2} du =\frac{1}{x^2}O(xM(x))=O(M(x)/x)$$ implies that the rightmost integral of our first line tends to $0$ as well. Thus $$\int_1^\infty \frac{M(u)}{u^2} du=0,$$ and $A(x)=O(M(x)/x)$. In particular, we can answer our question by simply bounding the growth of Mertens' function $M(x)$. We have $$M(x)=O\left(xe^{-c\sqrt{\log x}}\right)$$ for some positive constant $c$. (I believe this follows from the classical bounds in the PNT but am unable to find a proper reference. Edit: I found a mention of the process here.) Then $A(x) =O(e^{-c \sqrt{\log x}})$, and since $$\lim_{x \to \infty} \frac{(\log x)^n}{e^{c \sqrt{\log x}}}=0$$ for all $n$, we find $A(x)=o((\log x)^{-n})$ for all $n >0$.

share|cite|improve this answer
The link in this post is dead. A proper reference for both M(x) and A(x) is given in G.J.O. Jameson's The Prime Number Theorem, Thm. 5.1.9. page 186 (2003). An earlier source for M(x) is Landau's Handbuch vol.II, sec. 164, page 613 (1909) (the proof is in preceding pages). – daniel Feb 2 '15 at 19:17
It's also the error estimate de la Vallee Poussin found in his original proof of the PNT. Harold Edwards, Riemann's Zeta Function, p.82. – daniel Feb 3 '15 at 4:05

The inequality (and the ensuing equalities) of your second line is unfortunately meaningless. However, using the bound you give on $M(x)$ in the integral, one gets easily $$\left\vert \int_x^\infty \frac{M(u)}{u^2} du \right\vert =O\left(\sqrt{\log x}e^{-c \sqrt{\log x}}\right),$$ which ultimately does not change your conclusion.


share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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