Let $\mathbb N$ denote the set of all positive integers. Does there exist a countably additive measure $\mu : \mathcal P(\mathbb N) \to [0,\infty)$ such that $\mu(\mathbb N)<\infty$ and $\mu(\{nk: k\in \mathbb N\})=\dfrac 1{n \log n},\forall n>1$ ?


1 Answer 1


Other than the fact that the measure's value on $\{1\}$ is completely unspecified by the given constraint, there is only one possibility for such a measure (call it $\lambda$) if it exists: the Möbius inversion formula for infinite series means that $$ \frac1{n\log n} = \sum_{k\in\mathbb N} \lambda(\{nk\}) \implies \lambda(\{n\}) = \sum_{k\in\mathbb N} \mu(k) \frac1{nk \log nk} = \frac1n \sum_{k\in\mathbb N} \frac{\mu(k)}{k \log nk} \quad(n\ge2), $$ where $\mu$ is the Möbius function. Using Dirichlet series one can further show that $$ \lambda(\{n\}) = \int_1^\infty \frac{n^{-x}}{\zeta(x)} \,dx \quad(n\ge2), $$ where $\zeta(s)$ is the Riemann zeta function. Since $1/\zeta(x)\asymp x-1$ for $x\in[1,2]$ while $1/\zeta(x) \asymp 1$ for $x\in[2,\infty)$, one can show from this integral representation that $\lambda(\{n\})$ has order of magnitude $1/(n\log^2 n)$, so that this really does represent a finite (positive) measure on $\mathbb N$. And indeed \begin{align*} \lambda\big( \{nk\colon k\in\mathbb N\}\big) &= \sum_{k=1}^\infty \int_1^\infty \frac{(nk)^{-x}}{\zeta(x)} \,dx \\ &= \int_1^\infty \frac{n^{-x}}{\zeta(x)} \sum_{k=1}^\infty \int_1^\infty k^{-x} \,dx \\ &= \int_1^\infty \frac{n^{-x}}{\zeta(x)} \zeta(x) \,dx = \int_1^\infty n^{-x} \,dx = \frac1{n\log n} \quad(n\ge2) \end{align*} as desired.

  • 1
    $\begingroup$ Nice. WolframAlpha says $$\mu(\mathbb N)-\mu(\{1\})= \sum_{n=2}^\infty \int_1^\infty \frac{n^{-x}}{\zeta(x)}\,dx=\int_1^\infty \frac{\sum_{n=2}^\infty n^{-x}}{\zeta(x)}\,dx=\int_1^\infty1-\frac{1}{\zeta(x)}\,dx = 1.1239120333264 $$ $\endgroup$ Dec 22, 2018 at 7:11

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.