Back to Ishan Levy’s website

Analytic number theory

2 Dirichlet series and properties

A very powerful tool in mathematics for studying sequences of numbers is to study their generating functions. If \(a_n\) is a sequence, one way to turn \(a_n\) into a generating function is to consider \(F(z) = \sum _0^\infty a_nz^n\). Properties of the sequence then relate to properties of the function, and vice versa. For example, we might expect the behavior of \(F(z)\) as \(z \to 1^{-}\) to be related to the asymptotics of the sequence. Also if \(F\) is analytic near \(1\), then the coefficients \(a_n\) can be obtained via Cauchy’s integral formula. These generating functions are good at capturing additive properties of sequences, as \(x^nx^m = x^{n+m}\).

Another type of generating function that can be made from sequences is an \(L\)-function. These capture more of the multiplicative structure of the sequence. These look like something of the form \(L(s) =\sum _0^\infty \frac {a_n}{n^s}\), called a Dirichlet series. Both of these constructions are specializations of the more general construction ca where given two sequences \(\lambda _n,a_n\), we can consider \(\sum a_ne^{-\lambda _n s}\). The basic example is \(\zeta (s) = \sum \frac {1}{n^s}\), the Riemann zeta function. Note the series converges absolutely and uniformly on compact sets for \(\real (s)>1\).

As any good generating function should, \(\zeta (s)\) tells us a great deal about the distribution of the primes. Below is a simple of example of how it can be used.

  • Theorem 2.1. The series \(\frac {1}{p}\) diverges, where \(p\) ranges over the positive primes in \(\ZZ \).

  • Proof. \(\sum _1^\infty \frac {1}{n^s}= \prod _p(\frac {1}{1-\frac {1}{p^s}})\), which diverges as \(s \to 1^+\). Taking the log of the right hand side we get \(\sum _p -\log (1-\frac {1}{p^s}) = \sum _p \frac {1}{p} + \sum _{m \geq 2, p} \frac {1}{mp^{ms}}\). The second sum is \(\leq \sum _{m \geq 2, p} \frac {1}{p^{ms}} = \sum _{p} \frac {1}{p^{2s}-p^s} \leq 2\sum _{p}\frac {1}{p^{2s}}\), which is finite as \(s \to 1\). Thus \(\sum _p \frac {1}{p}\) diverges.

First we will prove some lemmas about Dirichlet series. Note that a necessary condition for convergence at some point is that \(|a_n|\) should be bounded by some polynomial. Here is a partial converse:

  • Lemma 2.2. If \(|\sum _1^n a_n| = O(n^r)\), \(r>0\), then the Dirichlet series for \(a_n\) converges uniformly for \(\real (s)\geq r+\ee \) for any \(\ee >0\) to a holomorphic function. If \(|a_n| = O(n^r)\), then it converges absolutely and uniformly for \(\real (s)>r+1+\ee \).

  • Proof. Let \(f_n = \sum _1^n a_n\). We will show that the partial sums are uniformly Cauchy by using summation by parts. \(\sum _k^m \frac {a_n}{n^s} = \sum _k^m\frac {f_n-f_{n-1}}{n^s}\) = \(\frac {f_m}{m^s} - \frac {f_{k-1}}{(k-1)^s} + \sum _k^{m-1}f_n (\frac {1}{n^s}-\frac {1}{(n+1)^s})\). The first two terms go to \(0\) uniformly as \(k \to \infty \), and the difference in the sum is equal to \(\int _{n}^{n+1}(-s)x^{-(s+1)}\), which in absolute value is at most \(|s|n^{-s+1}\). But then since \(f_n = O(n^r)\), the last term in absolute value is at most \(\sum _k^{m-1}\frac {cn^r}{n^{-(r+\ee +1)}}\), which is uniformly bounded in \(m\) and goes to \(0\) as \(k \to \infty \) since \(\ee +1 > 1\). The last statement is clear.

  • Lemma 2.3. If \(F,G\) are Dirichlet series for \(a_n,b_n\) that converge, and \(F,G\) agree where they converge, then \(a_n = b_n\).

  • Proof. Let \(a_i\) be the first nonzero term of \(a_n\). Then \(F \sim \frac {a_i}{i^s}\) as \(s \to \infty \), so this behavior determines \(i\) and \(a_i\). By subtracting this term off, we can recover the rest of the sequence. We have used properties of \(F\) that agree with \(G\), so \(a_n = b_n\).

Note by analyticity, if they agree on a small set, they agree.

  • Lemma 2.4. Suppose that \(F,G\) are convergent Dirichlet series for \(a_n,b_n\). Then \(FG\) is the Dirichlet series for \(a_n\star b_n\), where \(\star \) denotes the Dirichlet convolution.

  • Proof. For sufficiently large \(s\), \(F,G\) converge absolutely and uniformly. Then when we take their product, we can change the order of summation to get the result.