Bourbaki

Appendix

Möbius function

Let n be an integer ≥1\geq 1. If n is divisible by the square of a prime number, we write μ(n)=0\mu(n) = 0. If n is not divisible by the square of a prime number, we write μ(n)=(−1)k\mu(n) = (-1)^k, where k is the number of prime divisors of n. The function μ:N∗→{−1,0,1}\mu : \mathbf{N}^* \to \{-1, 0, 1\} thus defined is called the Möbius function.

Recall that given two integers n1≥1,n2≥1n_1 \geq 1, n_2 \geq 1, we write n1∣n2n_1 | n_2 if n1n_1 divides n2n_2.

Proposition

(i) The function μ\mu is the unique mapping of N∗\mathbf{N}^* into Z\mathbf{Z} such that μ(1)=1\mu(1) = 1 and

∑d∣nμ(d)=0 \sum_{d|n} \mu(d) = 0
*for every integer n>1n > 1.

(ii) Let s and t be two mappings of N∗\mathbf{N}^* into a commutative group written additively. In order that

s(n)=∑d∣nt(d)for every integer n≥1, s(n) = \sum_{d|n} t(d) \quad \text{for every integer } n \geq 1,
it is necessary and sufficient that
t(n)=∑d∣nμ(d)s(nd)for every integer n≥1. t(n) = \sum_{d|n} \mu(d) s\left(\frac{n}{d}\right) \quad \text{for every integer } n \geq 1.

The uniqueness assertion in (i) is obvious, for (1) allows us to determine μ(n)\mu(n) by induction on n. We show that the function μ\mu satisfies (1). Let n be an integer >1> 1. Let P be the set of prime divisors of n and let n=∏p∈Ppνp(n)n = \prod_{p \in P} p^{\nu_p(n)} be the decomposition of n into prime factors. If d is a divisor of n, then μ(d)=0\mu(d) = 0 unless d is of the form ∏p∈Hp\prod_{p \in H} p, where H is a subset of P. Then

∑d∣nμ(d)=∑H⊂P(−1)Card H=∑k=0Card P(nk)(−1)k=(1−1)Card P=0. \sum_{d|n} \mu(d) = \sum_{H \subset P} (-1)^{\mathrm{Card}\ H} = \sum_{k=0}^{\mathrm{Card}\ P} \binom{n}{k} (-1)^k = (1 - 1)^{\mathrm{Card}\ P} = 0.

Let s and t be two mapping of N∗\mathbf{N}^* into a commutative group written additively. Let n∈N∗n \in \mathbf{N}^*. If (2) holds, then

∑d∣nμ(d)s(nd)=∑d∣nμ(d)∑δ∣(n/d)t(δ)=∑d∣nμ(d)t(δ) \sum_{d|n} \mu(d)s\left(\frac{n}{d}\right) = \sum_{d|n} \mu(d) \sum_{\delta|(n/d)} t(\delta) = \sum_{d|n} \mu(d)t(\delta)
=∑δ∣nt(δ)∑d∣(n/δ)μ(d)=t(n). = \sum_{\delta|n} t(\delta) \sum_{d|(n/\delta)} \mu(d) = t(n).
Conversely, if (3) holds, then
∑d∣nt(d)=∑d∣n∑δ∣dμ(δ)s(dδ)=∑d∣ns(d)∑δ∣(n/d)μ(δ)=s(n), \sum_{d|n} t(d) = \sum_{d|n} \sum_{\delta|d} \mu(\delta)s\left(\frac{d}{\delta}\right) = \sum_{d|n} s(d) \sum_{\delta|(n/d)} \mu(\delta) = s(n),
which completes the proof.

Formula (3) is called the Möbius inversion formula.

Exercises

See the exercises for Appendix 0.

Transcribed from the printed volume. Not translated.