Vaidyanathaswamy calls an arithmetic function rational if it is the convolution of some finite collection of functions which are either completely multiplicative or inverse to a completely multiplicative function. Since multiplicative functions are closed under convolution and (Dirichlet) inversion, rational functions are all multiplicative. But are there multiplicative functions which are not rational? I expect so, but I haven't developed good intuition for convolutions.
2 Answers
Not all multiplicative functions are rational. For simplicity take arithmetic functions with complex values. It is easy to show that $f$ is rational of order $(m,n)$ (meaning the Dirichlet product of $m$ completely multiplicative and and $n$ inverse completely multiplicative functions) if and only if for every prime $q$ there exist $c_1(q), \ldots, c_m(q) \in \mathbb{C}$ such that $$f(q^k) = \sum_{i = 1}^m c_i(q) f(q^{k-i}) \mbox{ for all } k > n,$$ where we put $f(q^j) = 0$ if $j < 0$. (See Proposition 10.3 of my paper "Ring structures on groups of arithmetic functions" in Journal of Number Theory, which generalizes it to any ring.) This means that the $f(q^k)$ for fixed $q$ have to satisfy a linear recurrence relation. All you need to do is for some prime $q$ choose any sequence $a_{1}, a_{2}, \ldots$ of complex numbers that does not satisfy any such recurrence relation and define $f(q^n) = a_{n}$ for all $n$, then define $f(p^n)$ however you like for $p \neq q$, and extend $f$ by multiplicativity. Moreover, any multiplicative arithmetic function that is not rational can be constructed in this way.
Alternatively, if you want to avail yourself of exponentiation of multiplicative arithmetic functions, note that if $f$ is multiplicative, then $f^\alpha$ (exponentiation in the sense of Dirchlet convolution) is defined and multiplicative for any complex number $\alpha$, but if $f$ is rational and not the identity $\epsilon$ under convolution, then $f^\alpha$ is rational implies that $\alpha \in \mathbb{Q}$, so $f^\alpha$ is multiplicative but not rational for all $\alpha \in \mathbb{C}-\mathbb{Q}$. See Proposition 9.1 of the same paper (which generalizes the result to integral domains $R$, using exponents in the universal binomial ring over $R$.)
-
$\begingroup$ That's a great structural result! I'll check out the paper, thanks. $\endgroup$– CharlesJan 22, 2016 at 23:31
Yes, assuming by "inverse" you mean Dirichlet inverse.
An arithmetic function $f(n)$ is multiplicative iff its Dirichlet series at least formally admits an Euler product
$$\sum_{n=1}^{\infty} \frac{f(n)}{n^s} = \prod_p \left( \sum_{k=0}^\infty \frac{f(p^k)}{p^{ks}} \right).$$
It is completely multiplicative if in addition each factor in the Euler product takes the special form
$$\sum_{k=0}^\infty \frac{f(p^k)}{p^{ks}} = \frac{1}{1 - \frac{f(p)}{p^s}}.$$
If $f$ is rational, then each factor in the Euler product is a rational function of $p^s$ (presumably this is the motivation behind the terminology), but it's easy to pick some arbitrary Euler factors which don't have this property.
-
1$\begingroup$ It might be helpful that the "yes" you state in the first line is an answer to the question in the body, and not in the title, of the question. $\endgroup$– WojowuJan 22, 2016 at 7:54
-
$\begingroup$ @Wojowu: I'll edit the question so no one is confused. $\endgroup$– CharlesJan 22, 2016 at 23:28