The Mertens function $M(x)$ is defined as follows.

(1) $\quad M(x)=\sum\limits_{n=1}^x\mu(n)$

I've noticed the Merten's function can also be evaluated as follows which is related to OEIS entry A087003 and the Collatz conjecture.

(2) $\quad M(x)=\sum\limits_{n=\left\lfloor\frac{x-2}{4}\right\rfloor+1}^{\left\lfloor\frac{x-1}{2}\right\rfloor}\mu(2\,n+1)$

I've verified formula (2) above for the first $10,000$ positive integer values of $x$.

Question (1): Has formula (2) above been proven (or disproven) and if not, can it be?

I've read the following conjecture on the growth of $M(x)$ for any $\epsilon<1/2$ is equivalent to the Riemann hypothesis (see Weisstein, Eric W. "Mertens Conjecture." From MathWorld--A Wolfram Web Resource).

(3) $\quad M(x)=O\left(x^{1/2+\epsilon}\right)$

Question (2): Assuming formula (2) above is correct, does it show any promise with respect to improving upon the fastest known algorithm for computing $M(x)$ and/or the tightest known error bound on the growth of $M(x)$?

  • $\begingroup$ Yes you are right, my bad $\endgroup$ Jul 19, 2019 at 20:01

1 Answer 1


There is nothing special happening here. Let $$F(s) = \sum_{n=0}^\infty \mu(2n+1)(2n+1)^{-s}, \qquad \frac{1}{\zeta(s)} = (1-2^{-s}) F(s)$$ $$f(x) = \sum_{2n+1 \le x} \mu(2n+1), \qquad M(x) = f(x)-f(x/2)$$

  • 1
    $\begingroup$ can do it directly by separating the $\mu(2(2k+1))=-\mu(2k+1)$ since $\mu$ is multiplicative, $\mu(4k)=0$ and cancelling out the odd $\mu$ for the first half, but using RZ is cooler! $\endgroup$
    – Conrad
    Jul 20, 2019 at 1:49

Your Answer

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

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