Dirichlet Nonvanishing
0.1 What this project is about
We’re trying to prove the following theorem:
Let \(N \ge 1\) and \(\chi \) a complex-valued Dirichlet character mod \(N\). Then \(L(\chi , 1 + it) \ne 0\) for all real \(t\).
See Theorem 12 below.
0.2 High-level outline
If \(t \ne 0\) or if \(\chi ^2 \ne 1\) then this follows from the product for \(\zeta (s)^3 L(\chi , s)^{-4} L(\chi ^2, s)\) which is in EulerProducts. So we may assume that \(t = 0\) and \(\chi \) is quadratic.
Assume for contradiction \(L(\chi , 1) = 0\). Then the function
\[ F(s) = L(\chi , s) \zeta (s) \]is entire.
For \(\Re (s) {\gt} 1\), \(F(s)\) is given by a convergent Euler product with local factors XXX. Hence its Dirichlet series coefficients are positive real numbers.
Hence the iterated derivatives of \(F(s)\) on \((0, \infty )\) alternate in sign.
By an analytic result from EulerProducts, this implies that \(F(s)\) is real and non-vanishing for all real \(s\).
However, \(F(-2) = 0\). This gives the desired contradiction.
0.3 A more detailed plan
Let \(\chi \) be a Dirichlet character modulo \(N\). Then for all \(\varepsilon {\gt} 0\), we have
This follows from a trigonometric inequality.
Let \(t \in \mathbb {R}\) and let \(\chi \) be a Dirichlet character. If \(t \ne 0\) or \(\chi ^2 \ne 1\), then
Assume that \(L(\chi , 1 + it) = 0\). Then the (at least) quadruple zero of \(L(\chi , s)^4\) at \(1 + it\) will more than compensate for the triple pole of \(L(1, s)^3\) at \(1\), so the product of the first two factors in 1 will tend to zero as \(\varepsilon \searrow 0\). If \(t \ne 0\) or \(\chi ^2 \ne 1\), then the last factor will have a finite limit, and so the full product will converge to zero, contradicting Lemma 1.
So it suffices to prove that if \(\chi \) is a quadratic character we have \(L(\chi , 1) \ne 0\).
A bad character is an \(\mathbb {R}\)-valued (hence quadratic) Dirichlet character such that \(L(\chi , 1) = 0\).
Define \(F \colon \mathbb {C} \to \mathbb {C}\) by
If \(\chi \) is a bad character, then \(F\) is an entire function.
This is easy for \(s \ne 1\) since we know analyticity of both factors. To prove analyticity at \(s = 1\), it suffices to show continuity (Riemann criterion) and that should follow easily since we know that \(\lim _{s \to 1} (s - 1) \zeta (s)\) exists.
We have \(F(-2) = 0\).
Follows from the trivial zeroes of Riemann zeta.
For \(\Re (s) {\gt} 1\), \(F(s)\) is equal to the \(L\)-series of a real-valued arithmetic function \(e\) defined as the convolution of constant 1 and \(\chi \).
We have Euler products for both \(L(\chi , s)\) and \(\zeta (s)\).
The weakly multiplicative function \(e(n)\) whose Euler product is \(\mathcal{E}\) takes non-negative real values.
It suffices to show this for prime powers. We have \(e(p^k) = (k + 1)\) if \(\chi (p) = 1\), \(e(p^k) = 1\) if \(\chi (p) = 0\), and if \(\chi (p) = -1\) then \(e(p^k) = 1\) if \(k\) even, \(0\) if \(k\) odd.
An entire function \(f\) whose iterated derivatives at \(s\) are all real with alternating signs (except possibly the value itself) has values of the form \(f(s) + \text{nonneg. real}\) along the nonpositive real axis shifted by \(s\).
This follows by considering the power series expansion at zero.
If \(a \colon \mathbb {N}\to \mathbb {C}\) is an arithmetic function with \(a(1) {\gt} 0\) and \(a(n) \ge 0\) for all \(n \ge 2\) and the associated \(L\)-series \(f(s) = \sum _{n \ge 1} a(n) n^{-s}\) converges at \(x \in \mathbb {R}\), then the iterated derivative \(f^{(m)}(x)\) is nonnegative for \(m\) even and nonpositive for \(m\) odd.
The \(m\)th derivative of \(f\) at \(x\) is given by
and the last sum has only nonnegative summands.
If \(\chi \) is a nontrivial quadratic Dirichlet character, then \(L(\chi , 1) \ne 0\).
Assume that \(L(\chi , 1) = 0\), so \(\chi \) is a bad character. By Lemma 5, we then know that \(F\) is an entire function. From Lemmas 7 and 8 we see that \(F\) agrees on \(\Re s {\gt} 1\) with the \(L\)-series of an arithmetic function with nonnegative real values (and positive value at \(1\)). Lemma 10 now shows that \((-1)^m F^{(m)}(2) \ge 0\) for all \(m \ge 1\). Then Lemma 9 (applied to \(f(s) = F(2+s)\)) implies that \(F(x) {\gt} 0\) for all \(x \le 2\). This now contradicts Lemma 6, which says that \(F(-2) = 0\). So the initial assumption must be false, showing that \(L(\chi , 1) \ne 0\).
If \(\chi \) is a Dirichlet character and \(t\) is a real number such that \(t \ne 0\) or \(\chi \) is nontrivial, then \(L(\chi , 1 + it) \ne 0\).