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

If gcd(a,b)=1, $1\leq b\leq a$, and $\mu(k)$ is the Möbius function, what is $$\sum_{k=0}^\infty\frac{\mu(ak+b)}{(ak+b)^s}$$ Can it be expressed in terms of other functions? Can I get it in the form of an Euler product?

share|cite|improve this question
Try Dirichlet $L$-functions, and using Dirichlet characters to isolate the congruence. – Eric Naslund Jan 5 '13 at 0:58
Is it always postive for s>1 eric? – Ethan Jan 5 '13 at 0:59
If $S(a,b;s)$ is your sum, then I would guess that the sign of $S(a,b;s)$ will be equal to the sign of $\mu(b)$. A numerical example: $S(1,3;2) \approx -0.142073$. – Antonio Vargas Jan 5 '13 at 1:22
0<b<a+1 sorry i should have specified – Ethan Jan 5 '13 at 1:28
$S(5,3;2) \approx -0.118285$ – Antonio Vargas Jan 5 '13 at 1:30
up vote 7 down vote accepted

When $a>2$, this sum cannot be written as an Euler product $\prod_{p\text{ prime}} (\sum_{k\ge 0} a_{p^k} p^{-ks})$. We'll prove this by using the following fact: a Dirichlet series has an Euler product if and only if the corresponding arithmetic function is multiplicative ($f(nm)=f(n)f(m)$ if $(n,m)=1$) - prove this by expanding the Euler product!

First case: $a=b=1$. It is well-known that $\mu(n)$ is multiplicative (as an inverse of a multiplicative function), hence $$(*) \sum_{n \ge 1} \frac{\mu(n)}{n^s} = \prod_{p} (\sum_{k \ge 0} \mu(p^k) p^{-ks})=\prod_{p} (1-p^{-s})$$

Alternatively, since $\mu$ is the inverse of the constant function $\zeta = 1$ which has the representation $\sum_{n \ge 1} n^{-s} = \prod_{p} (1-p^{-s})^{-1}$, one can deduce $(*)$ by $(\sum_{n \ge 1} n^{-s}) (\sum_{n \ge 1} \mu(n) n^{-s})=1$.

Second case: $a=2,b=1$. The function $f(n)=\mu(n) \cdot 1_{2 \nmid n} = \begin{cases} 0 && 2|n \\ \mu(n) && 2 \nmid n \end{cases}$ is multiplicative, because both $\mu(n)$ and $1_{2 \nmid n} = \begin{cases} 0 && 2 |n \\ 1 && 2\nmid n \end{cases}$ are. Thus, we can expand the Dirichlet series of $f$ as follows: $$\sum_{n \ge 0} \frac{\mu(an+b)}{(an+b)^s} = \sum_{n \ge 0} \frac{\mu(2n+1)}{(2n+1)^s} = \sum_{m \ge 1} \frac{f(m)}{m^s}=$$ Here comes the Euler product: $$\prod_{p} (\sum_{k \ge 0} f(p^k) p^{-ks})= (\sum_{k \ge 0} f(2^k) 2^{-ks}) \prod_{p \neq 2} (\sum_{k \ge 0} \mu(p^k) p^{-ks}) = \prod_{p \neq 2} (1-p^{-s})$$

Third case: $a>2$. Sadly, $f(n)=\mu(n) \cdot 1_{n \equiv b \mod a}$ is no longer multiplicative. To show this, I'll assume Dirichlet's theorem on primes in arithmetic progressions. It states that if $n,m$ are coprime positive integers, then there are infinitely many primes satisfying $p \equiv m \mod n$.

Thus we can construct 2 distinct primes $p_1, q_1$ such that $p_1 \equiv 1 \mod a, p_2 \equiv b \mod a$. We have $f(p_1 q_1) = \mu(p_1 q_1) 1_{p_1 q_1 \equiv b \mod a} = 1$, and $f(p_1) f(q_1) = \mu(p_1)\mu(q_1) 1_{p_1 \equiv b \mod a}1_{q_1 \equiv b \mod a}= 1_{p_1 \equiv b \mod a, q_1 \equiv b \mod a}$.

Thus, $f(p_1 q_1)=f(p_1)f(q_1)$ implies $b \equiv 1 \mod a$. Now we'll construct 2 additional distinct primes $p_2, q_2$ such that $p_2, q_2 \equiv a-1 \mod a$. Since $p_2 q_2 \equiv 1 \mod a$, yet $p_2 \neq 1 \mod a$, we have $f(p_2)f(q_2) = 0$ yet $f(p_2 q_2)=1$.

It follows that $f$ isn't multiplicative.

Now we'll express our Dirichlet series as a linear combinations of Dirichlet series' with Euler product. In other words, we want to write $\mu(n) \cdot 1_{n \equiv b \mod a}$ as a linear combination of multiplicative functions.

We'll use characters. Characters modulo $m$ are functions from the positive integers to the complex numbers, satisfying 3 properties:

  1. They are periodic with period $m$: $\chi(a)=\chi(a+m)$.

  2. They are (completely) multiplicative: $\chi(ab) = \chi(a) \chi(b)$.

  3. If $(a,m) >1$, $\chi(a)=0$. Otherwise, $\chi(a) \neq 0$.

The theory of characters modulo $m$ is very well understood. For any $m$ you can construct distinct $\phi(m)$ characters. Given the factorization of $m$, it is a direct procedure to construct those characters. Read more here. There's a lot of interesting theory there, and explicit examples of characters.

Exercise 1: Show that $\chi(1)=1$ for any character, and that $\chi(a)$ is always a root of unity of order dividing $\phi(m)$.

Exercise 2: If $\chi$ is a character, $\overline{\chi}$ (complex conjugation) is also a character. $\overline{\chi}(a) = \chi(a^{-1})$, where $a^{-1}$ denotes a positive integer satisfying $a^{-1} \cdot a \equiv 1 \mod m$. Show that $\overline{\chi}(a) = \begin{cases} 0 && (a,m) > 1 \\ \frac{1}{\chi(a)} && (a,m)=1 \end{cases}$.

Exercise 3: The product of 2 characters is also a character.

Enough with the exercises. Suppose $\chi_{i}, 1 \le i \le \phi(m)$ are my distinct characters.

Theorem: Suppose $a$ is coprime to $m$.

$$\frac{1}{\phi(m)} \sum_{i=1}^{\phi(m)} \overline{\chi_i(a)}\chi_i(n) = 1_{n \equiv a \mod m}$$

The proof is actually not hard. If $(n,m)>1$ both sides are 0. Otherwise, by exercise 2, this identity is equivalent to

$$\sum_{i=1}^{\phi(m)} \chi_i(x) = \phi(m) 1_{x \equiv 1 \mod m}$$ where $x \equiv \frac{n}{a} \mod m$.

One direction is easy - if $x=1 \mod m$, all the characters evaluated on $x$ give the value 1.

The other direction is very nice - assume $x \neq 1 \mod m$. Let $S(x)=\sum_{i=1}^{\phi(m)} \chi_i(x)$. Use exercise 3 to show that $\{ \chi_i \}_{i} = \{ \chi_i \chi_j \}_{i}$, and deduce that $S(x) \chi_j(x) = S(x)$, for any $j$. Since the order of $x$ modulo $m$ is not $1$, one can find a character that doesn't give the value $1$ on $x$ (I can't elaborate without using more group theory), so for some $j$ we can cancel $\chi_j(x) - 1$ in the equality $S(x)(\chi_j(x) - 1)=0$ and get that $S(x)=0$. $\blacksquare$

With this theorem at hand, the decomposition is straightforward: we can write $\mu(n) \cdot 1_{n \equiv b \mod a}$ as $\frac{1}{\phi(a)} \sum_{i=1}^{\phi(a)} \overline{\chi_i(b)} \chi_i(n)\mu(n)$, and this shows: $$\sum_{n \ge 0} \frac{\mu(an+b)}{(an+b)^s} = \sum_{n \ge 1} \frac{\mu(n) 1_{n \equiv b \mod a}}{n^s}=$$ $$\frac{1}{\phi(a)} \sum_{i=1}^{\phi(a)} \overline{\chi_i(b)} (\sum_{n \ge 1} \frac{\mu(n) \chi_i(n)}{n^s})$$ And each summand has an Euler product, since $\mu(n) \chi_i(n)$ is multiplicative (a product of 2 multiplicative functions): $$\sum_{n \ge 1} \frac{\mu(n) \chi_i(n)}{n^s} = \prod_{p \text{ is prime}} (\sum_{k \ge 0} \mu(p^k)\chi_i(p^k)p^{-ks})=\prod_{p \text{ is prime}} (1-\chi_i(p)p^{-s})$$ You might recognize this as the inverse of the so-called L-function $L(s, \chi_i)$, defined by: $$L(s, \chi_i) = \sum_{n \ge 1} \frac{\chi_i(n)}{n^s} = \prod_{p \text{ is prime}} (1-\chi_i(p)p^{-s})^{-1}$$

Thus, we can write your function as $$\frac{1}{\phi(a)} \sum_{i=1}^{\phi(a)} \overline{\chi_i(b)} L(s,\chi_i)^{-1}$$

When $a=b=1$ or $a=2,b=1$ (the 2 cases discussed at the beginning), there is exactly one character, since $\phi(2)=\phi(1)=1$, and this sum becomes one element: $L(s, \chi)^{-1}$ where $\chi$ is the "trivial" character - the trivial character modulo $m$ is 1 when the value is coprime to $m$ and $0$ otherwise.

Hope this was readable. I can't help but notice that you ask many questions of this kind, reading chapter 16 of this classc will help you.

EDIT: In general, $$\sum_{k \ge 0} \frac{f(ak+b)}{(ak+b)^s} = \frac{1}{\phi(a)} \sum_{i=1}^{\phi(a)} \overline{\chi_i(b)} h(s,\chi_i)$$ where $h(s,\chi_i) = \sum_{n \ge 1} \frac{f(n)\chi_i(n)}{n^s}$, which has an Euler product expansion when $f$ is multiplicative. This show that you can express this kind of sums with at most $\phi(a)$ functions with Euler product expansions.

share|cite|improve this answer
I appreciate all the work you did – Ethan Jan 13 '13 at 5:56
@Ethan No problem. My only question is whether this was clear for you? – Ofir Jan 13 '13 at 6:04

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.