Arxiv:1911.07215V3 [Math.NT] 18 May 2021 ( √ 1 (−1 + D) If D ≡ 1 (Mod 4), 2 √ Τd = 1 2 D If D ≡ 0 (Mod 4); and Let J(·) Be the Classical J-Invariant Function

ON LANDAU–SIEGEL ZEROS AND HEIGHTS OF SINGULAR MODULI

CHRISTIAN TAFULA´

√ Abstract. Let χD be the Dirichlet character associated to Q( D) where D < 0 is a fundamental discriminant. Improving Granville–Stark [7], we show that L0 1 1 (1, χ ) = height(j(τ )) − log |D| + C + o (1), L D 6 D 2 D→−∞ √ 1 where τD = 2 (−δ + D) for D ≡ δ (mod 4) and j(·) is the j-invariant function with C = −1.057770 .... Assuming the “uniform”√ abc-conjecture for number ﬁelds, 5ϕ+o(1) √ we deduce that L(β, χD) 6= 0 with β ≥ 1 − log |D| where ϕ = (1 + 5)/2, which we improve for smooth D.

1. Introduction In 2000, Granville and Stark [7] showed that the uniform abc-conjecture for number ﬁelds (Conjecture 5.2 (iii)) implies that there are no “Siegel zeros”√ for odd characters, by deducing that, under uniform abc, the class number of Q( D) satisﬁes p π |D| X(D) 1 (1.1) h(D) ≥ (1 + o(1)) 3 log |D| a (a,b,c) where the sum runs over the reduced binary quadratic forms Q(x, y) = ax2 +bxy+cy2 of fundamental discriminant D < 0. Comparing this lower bound to the unconditional, more traditional type of lower bound: ! p 1 + O log log |D|/ log |D| π |D| X(D) 1 (1.2) h(D) = , 1 + 2 L0 (1, χ )/ log |D| 3 log |D| a L D (a,b,c)

where χD is the corresponding quadratic character, one derives from uniform abc that L0 | L (1, χD)| log |D|, which is equivalent to L(s, χD) having no “Siegel zeros”. Our main theorems imply a slightly more precise estimate than (1.2), and the existence of a subsequence of D’s for which L0 (1, χ ) = o (log |D|). L D D→−∞ √ For D < 0 we need the generator of the ring of integers of Q( D), arXiv:1911.07215v3 [math.NT] 18 May 2021 ( √ 1 (−1 + D) if D ≡ 1 (mod 4), 2 √ τD = 1 2 D if D ≡ 0 (mod 4); and let j(·) be the classical j-invariant function. Deﬁne the height of m/n ∈ Q with (m, n) = 1 by ht(m/n) = log max{|m|, |n|}.

2020 Mathematics Subject Classiﬁcation. 11M20, 11G50, 11N37. Key words and phrases. Siegel zeros, singular moduli, abc-conjecture.

1 2 CHRISTIAN TAFULA´

We will give the height function for algebraic numbers in subsection 2.3. We improve on (1.2) by using Duke’s equidistribution theorem [6] (Lemma 4.2): Theorem 1.1. For fundamental discriminants D < 0, we have L0 1 1 (1.3) (1, χ ) = ht(j(τ )) − log |D| + C + o (1), L D 6 D 2 D→−∞ where C := κ1 − κ2 + κ3 + γ − log 2 = −1.057770 ..., and γ = 0.577215 ... is Euler–Mascheroni’s constant.1 In particular, this implies that ! p 1 + O 1/ log |D| π |D| X(D) 1 (1.4) h(D) = . 1 + 2 L0 (1, χ )/ log |D| 3 log |D| a L D (a,b,c)

L0 Expressing L (1, χD) in terms of ht(j(τD)) allows us to understand L-function values using Diophantine geometry techniques: The classical abc-conjecture (the Masser–Oesterléconjecture) states that for coprime integers a, b, c satisfying a+b = c, we have X log max |a|, |b|, |c| ≤ (1 + ε) log p + Oε(1), p | abc where the implied constant depends only on ε > 0. Several important results in number theory follow from this statement, including “asymptotic Fermat” (see Example 5.5.2, p. 71–72 of Vojta [18]). Granville and Stark used a uniform extension of the abc-conjecture to number fields.2 The tricky part is deciding what to conjecture 1/[K:Q] about the contribution of ramified primes in this inequality: thus, if rdK := |∆K | then we study two versions, where one includes the error term

O(log(rdK )) (O-weak uniformity) or o(log(rdK )) (weak uniformity), given by Conjectures 5.2 (i) and (ii), respectively. With minor modiﬁcations to the argument, we prove the following version of Granville–Stark’s theorem: Theorem 1.2. Assuming the abc-conjecture for number ﬁelds with L0 • O-weak uniformity we have | L (1, χD)| log |D|. 1 L0 • weak uniformity we have lim sup log |D| L (1, χD) = 0. D→−∞

L0 1.1. Traditional approach to L (1, χD). Let %(χ) be the set of non-trivial zeros of L(s, χ), each included in the set with multiplicity. We are interested in the zeros of L(s, χ) inside the region A (1.5) RA = RA(q) := s ∈ C σ ≥ 1 − , |t| ≤ 1 . log q

A for any given A ∈ R≥1 with q > e .A Siegel zero is a real zero β of L(s, χ) for some A primitive quadratic χ (mod q), with β ≥ 1 − log q for some given A > 0.

1 See Lemmas 4.3, 4.4, 4.5 for the definition of κ1, κ2, κ3, respectively. 2Formulated so as to be consistent between different field extensions. ON LANDAU–SIEGEL ZEROS AND HEIGHTS OF SINGULAR MODULI 3

Theorem 1.3. For every primitive character χ (mod q) we have the bound |%(χ) ∩ 3A RA| e (with an absolute implicit constant), and X 1 1 1 L0 < < 1 − √ log q + < (1, χ) + O(e3A). 1 − % 5 2 L %(χ)∩RA An integer q is y-smooth if all its prime factors are ≤ y. Chang [1] established wide zero-free regions for such q (see subsection 3.3) so we deduce:

δ Theorem 1.4. For every δ > 0 there is an Nδ > 0 such that if q > Nδ is q -smooth then the only possible element of %(χ) ∩ RA with A 1/δ, is the potential Siegel zero β = βχ := max{σ ∈ R | L(σ, χ) = 0}. In that case 0 L 1χ=χ 1 < (1, χ) = + O(δ 2 log q). L 1 − β Theorem 1.4 implies that, for every ﬁxed δ > 0, if D < 0 is a |D|δ-smooth fundamental discriminant then 0 L 1 (1.6) (1, χ ) ≥ −Mδ 2 log |D| L D

for some M ∈ R≥0 not depending on δ, provided |D| is suﬃciently large. 1.2. Main corollaries. We present two results that are consequences of the theorems above. First, by combining Theorems 1.2, 1.3 and 1.4 together, we deduce the following: Corollary 1.5. Assume the weak uniform abc-conjecture. As D → −∞, the function L(s, χD) has no zeros in the real interval √ 5ϕ + o(1) 1 − , 1 log |D| √ 1+ 5 where ϕ := 2 , nor in the region − 1 mδ 2 + oδ(1) s ∈ C σ ≥ 1 − , |t| ≤ 1 log |D| (for some absolute constant m > 0) when D is |D|δ-smooth for given δ > 0. See subsection 3.4 for a proof. Next we give explicit upper and lower bounds on several key analytic quantities: Corollary 1.6. For negative fundamental discriminants D we have: 3 ∗ (i) √ log |D| + O(1) ≤ ht(j(τD)) ≤ (1 + o(1)) 3 log |D|. 5 1 X 1 ∗ 1 (ii) √ log |D| + O(1) ≤ ≤ (1 + o(1)) log |D|. 2 5 % 2 %(χD) p ! p π |D| X 1 ∗ √ 1 π |D| X 1 (iii) (1 + o(1)) ≤ h(D) ≤ 5 + O 3 log |D| a log |D| 3 log |D| a (a,b,c) (a,b,c) 4 CHRISTIAN TAFULA´

∗ where the starred inequalities “≤” are conditional on the weak uniform abc-conjecture. Moreover, if D is |D|o(1)-smooth then these starred inequalities become (asymptotic) starred equalities. If we assume the Generalized Riemann Hypothesis then we obtain log log |D| 3 these asymptotic equalities for all D, with error term a factor of O( log |D| ). This corollary follows from Theorems 1.1 and 1.2; part (ii) is obtained from (3.5), and part (iii) from (1.4). (We omit the details.) In each case the upper and lower√ bounds in Corollary 1.6 diﬀer by an artiﬁcial multiplicative factor that is ∼ 5 arising from our proof (see Lemma 3.4 (ii)), which may be improvable. Data supports part (i) of Corollary 1.6:

ht(j(τD)) 6 Figure 1. Graph of 3 log |D| for −10 ≤ D ≤ 0.

The perceptible slight bias towards > 1 in Figure 1 can be analyzed by rewriting (1.3) as 0 ht(j(τ )) 2( L (1, χ ) − C + o(1)) D = 1 + L D . 3 log |D| log |D| L0 Therefore this bias towards > 1 in Figure 1 must stem from L (1, χD) − C being L0 usually positive. We verify this observation in Figure 2, by noting that L (1, χD) is usually larger than C = −1.057770 .... Lastly, an appendix on the Hadamard factorization of L-functions following Lucia [8] is included, justifying a formula we state in section 3.

Notation. Let f, g :[x0, +∞) → R+ be positive real-valued functions deﬁned in [x0, +∞) for some x0 ∈ R. We write f ∼ g, f = o(g), or f = O(g), if |f(x)/g(x)| goes to 1, 0, or stays bounded as x → +∞, respectively. If f = O(g), we also write f g. If f g and g f, then we write f g. 3 For example, ht(j(τD)) = 3 log |D| + O(log log |D|) assuming the Generalized Riemann Hypothesis. ON LANDAU–SIEGEL ZEROS AND HEIGHTS OF SINGULAR MODULI 5

L0 6 Figure 2. Graph of L (1, χD), for −10 ≤ D < 0.

2. Definitions and notation We review some notation and results that will be used in this paper: 2.1. Heegner points. We denote the binary, primitive quadratic form 2 2 Q(x, y) = ax + bxy + cy ∈ Z[x, y] by Q = (a, b, c) which has discriminant d = b2 − 4ac. If d < 0 then we will assume that a > 0 so that Q is positive-deﬁnite. Two forms Q1, Q2 are equivalent if α β Q1(x, y) = Q2(αx + βy, γx + δy) for some γ δ ∈ SL2(Z). For every non-zero discriminant d ≡ 0 or 1 (mod 4), we have the principal form

( 2 d 2 x − 4 y if d ≡ 0 (mod 4), (2.1) Q1,d(x, y) := 2 1−d 2 x + xy + 4 y if d ≡ 1 (mod 4). For d < 0, we say that (a, b, c) is reduced if −a < b ≤ a < c or 0 ≤ b ≤ a = c; every quadratic form of discriminant d is equivalent to a single reduced form. If Q is reduced, then a ≤ p|d|/3. Each binary quadratic form (a, b, c) with d < 0 is F associated with a CM point √ −b + d τ(a,b,c) = ∈ h := {z ∈ C | =(z) > 0}. 2a 2πi πi e 3 e 3 (a, b, c) is reduced if and only if τ(a,b,c) lies in the • ◦ fundamental domain, F , depicted on the right. Heegner points are the CM points associated to the reduced forms of fundamental discriminants D < 0; i.e., D’s which are the discriminant of some -1 -0.5 0 0.5 1 6 CHRISTIAN TAFULA´ quadratic number ﬁeld. Write√ ΛD for the set of Heegner points of D, and C`(D) for the ideal class group of Q( D). By the classical correspondence √ reduced (a, b, c) −b + D (2.2) C`(D) 3 A ↔ ↔ τ = ∈ ΛD (⊆ F ), (b2 − 4ac = D) 2a we may index Heegner points by C`(D), namely ΛD = {τA | A ∈ C`(D)}. The Heegner point associated to the trivial ideal class in C`(D), which corresponds to the principal form (2.1), is denoted τD.

2πiτ 2.2. Singular moduli. For τ ∈ h, write q = qτ := e . The q-expansion of the j-invariant function has the form 3 1 + 240 P P d3qn n≥1 d|n 1 X (2.3) j(τ) := = + c(n)qn, q Q (1 − qn)24 q n≥1 n≥0 √ with c(0) = 744, c(1) = 196884, and 3 c(n) ∼ √1 e4π nn−3/4. This is the unique Z≥1 2 modular function with respect to SL2(Z) of weight 0, holomorphic in h, satisfying j(e2πi/3) = 0, j(i) = 1728, and having a simple pole at i∞. The values taken by j(τ) at CM points τ ∈ h are called singular moduli. For√ D < 0 a fundamental discriminant, write HD for the Hilbert√ class ﬁeld of Q( D), which is the maximal unramiﬁed abelian extension of Q( D). Then: √ √ • HD = Q( D, j(τD)) (and [HD : Q( D)] = [Q(j(τD)) : Q] = h(D)); √ •{ j(τ) | τ ∈ ΛD} is the complete set of Gal(Q/Q( D))-conjugates of j(τD);

• j(τD) is an algebraic integer.

2.3. Heights and conductors. For a number field K/Q and a place v on K, write Kv for the completion of K with respect to v. For non-archimedean v, write pv for the prime ideal in OK associated to v, pv for the positive rational prime below pv, and fv for the degree of the residual extension Kv/Qpv . If v is archimedean, let ιv : K,→ C denote an embedding which induces v. For x ∈ K×, define the normalized absolute value k·kv as:  |ι (x)| if K ' ,  v v R 2 kxkv := |ιv(x)| if Kv ' C,  −fv ordpv (x) pv if v is non-archimedean, where ordpv (x) (=: v(x)) denotes the power of pv appearing in the prime factorization of the principal fractional ideal (x) ⊆ K, and | · | is the usual absolute value in C. Let MK denote the set of inequivalent places of K satisfying the product formula Q kxk = 1 for all x ∈ K×, and write Mnon ⊆ M for the subset of the v∈MK v K K n non-archimedean places. For a point P = [x0 : ... : xn] ∈ PK in the projective n-space over K, we define its (na¨ıve,absolute, logarithmic) height by 1 X ht(P ) := log max{kxikv}, [K : Q] i≤n v∈MK ON LANDAU–SIEGEL ZEROS AND HEIGHTS OF SINGULAR MODULI 7 and its (logarithmic) conductor by 1 X NK (P ) := fv log(pv). [K : Q] non v∈MK ∃i,j≤n s.t. v(xi)6=v(xj ) n Neither ht nor N depend on the choice of representatives for P ∈ PK , since, for any × c ∈ K , we have ht(P ) = ht(cP ) from the product formula, and NK (P ) = NK (cP ) since the sum runs over the same places. Moreover, the height does not depend on the choice of the base field (provided its Pn contains P ); however, the conductor does. × Finally, for x ∈ K×, write ht(x) := ht([x : 1]). If α ∈ Q is integral, then 1 X (2.4) ht(α) = log+ |α∗|, |A| α∗∈A + where log x := log max{1, x} for x ∈ R>0, A is the complete set of Gal(Q/Q)- conjugates of α, and the absolute value is being taken on some fixed embedding ι : Q ,→ C (to which the value of ht(α) is independent).

L0 3. On <( L (1, χ)) and zeros near s = 1 Consider primitive characters χ (mod q) for q ≥ 2, and write %(χ) for the set of non-trivial zeros of L(s, χ) counted with multiplicity. In this section we are going to prove the following two propositions: Proposition 3.1. Let S be any ﬁnite subset of non-trivial zeros of L(s, χ) (including the empty set). Then: X 1 1 1 L0 1 2|S| + 3 < < 1 − √ log q + < (1, χ) + 1 + √ − 1. 1 − % 2 L 2 %∈S 5 5 Proposition 3.2. Consider the region ( ) 1 1 (3.1) Bf (q) := s ∈ C σ > 1 − , |t| < , f(q) pf(q) where f : Z≥2 → R satisﬁes 2 ≤ f(q) ≤ 4 log q. Then: ! L0 X 1 < (1, χ) − < 14.5 + 2|%(χ) ∩ B | pf(q) log q. L 1 − % f %(χ)∩Bf Theorem 1.3 follows directly from Proposition 3.1 together with the classical log-free zero-density estimate (see Chapter 18 of Iwaniec–Kowalski [11]), which in particular states that A cA %(χ) ∩ s ∈ C σ ≥ 1 − , |t| ≤ 1 e , log q where we can take c = 3. Theorem 1.4 on the other hand will be proved using Proposition 3.2 in subsection 3.3. 8 CHRISTIAN TAFULA´

3.1. Lemmas. Our starting point is the following formula: 0 0 L X 1 1 q 1 Γ s + aχ (3.2) (s, χ) = − log − , L s − % 2 π 2 Γ 2 %(χ) 1 where aχ := 2 (1 − χ(−1)) (see Appendix A for details). From the functional equation of L(s, χ), we have that, if % ∈ {s ∈ C | 0 < σ < 1} is a zero of L(s, χ), then %, 1 − % are zeros of L(s, χ), and 1 − % is a zero of L(s, χ). Thus, by noting that P −1 P −1 %(χ)(s − %) = %(χ)(% + (s − 1)) , we get 0 0 X Πσ−1(%) 1 q L 1 Γ σ + aχ (3.3) = log + < (σ, χ) + , 4 2 π L 2 Γ 2 %(χ) where Π is the pairing-up function: 1 1 1 1 (3.4) Π (s) := + + + , ε s + ε s + ε 1 − s + ε 1 − s + ε deﬁned for s, ε ∈ C such that ε 6= −s, −s, −1 + s, −1 + s. In this notation, by setting Γ0 Γ0 1 σ = 1 in (3.3) and using the special values Γ (1) = −γ and Γ ( 2 ) = −γ − 2 log 2, it follows that 0 X Π0(%) 1 L 1 (3.5) = log q + < (1, χ) − γ + log 2π + χ(−1) log 2 . 4 2 L 2 %(χ)

The next lemma estimates (3.5) for small perturbations of ε in Πε(%). Lemma 3.3. For 0 < ε < .85, we have:

X Πε(%) 1 1 (i) < log q + ; 4 2 ε %(χ)

X Πε(%) 1 1 1 (ii) − log q + γ + log 2π + χ(−1) log 2 < 1 + . 4 2 2 ε %(χ)

Γ0 Proof. For 0 < σ < 1.425, the function Γ (σ) is strictly increasing and negative. Thus, since Πε(%) > 0, we have from (3.3) that 0 X Πε(%) L 1 0 < < < (1 + ε, χ) + log q. 4 L 2 %(χ)

ζ0 −1 4 Moreover, since − ζ (σ) < (σ − 1) for 1 < σ ≤ 2, it follows that 0 0 L ζ 1 < (1 + ε, χ) ≤ (1 + ε) < , L ζ ε which proves part (i). For the second part, we deduce from (3.3) that

X Πε(%) 1 1 − log q + γ + log 2π + χ(−1) log 2 4 2 2 %(χ)

4This amounts to (σ − 1)ζ(σ) being strictly increasing for σ ∈ (1, 2]. ON LANDAU–SIEGEL ZEROS AND HEIGHTS OF SINGULAR MODULI 9

0 ζ 1 1 < (1 + ε) + γ + log 4 < + 1, ζ 2 ε

concluding the proof. Now, let M ≥ 2 be a ﬁxed real number, and take the following partition of the critical strip {s ∈ C | 0 < σ < 1}:

1 R1 := {s ∈ C | 0 < σ < 1, |t| ≥ 1}

( ) 1 1 R2 := s ∈ C ≤ σ ≤ 1 − , |t| < 1 M M

0 1 ( 1 1 ) 0 ≤ σ(1 − σ) ≤ 1 − , R := s ∈ M M 3 C −1/2 M ≤ |t| < 1

−1 Be := {s ∈ C | 0 < σ < 1}\ (R1 ∪ R2 ∪ R3)

Note that, in the notation of (3.1), Be = B1/M ∪ (1 − B1/M ). The goal of the next lemma is to bound Π0(s) in terms of Πε(s) for s suﬃciently far away from 0 and 1, which in our case means “s outside of Be”. Lemma 3.4 (Pairing-up lemma). The following hold: √ 1 + 5 Π (s) (i) For 0 < <(s) < 1 and ϕ := , we have Π (s) > ϕ−1 ; 2 0 2ϕ − 1

(ii) For s ∈ R1 ∪ R2 ∪ R3 and 0 < ε < 1, we have

|Π0(s) − Πε(s)| < 5Mε Πε(s). Remark 3.5. This lemma was initially inspired by an argument attributed to U. P Vorhauer used to estimate − %(χ) <(1/%). Although we were unable to ﬁnd the original source, the argument is outlined in Exercise 8, Section 10.2 of Montgomery– Vaughan [15].

Proof of Lemma 3.4. Let s = σ + it and ε > 0. Writing σe := σ(1 − σ) and σeε := (σ + ε)(1 − σ + ε) = σe + ε(1 + ε), we have Π (s) σ + ε 1 − σ + ε ε = + 2 (σ + ε)2 + t2 (1 − σ + ε)2 + t2 2 4 σeε + (1 + σeε)t + t (1 + 2ε) (3.6) = 2 2 2 4 2 σeε + ((1 + 2ε) − 2σeε)t + t 1 + t σ(1 − σ) + 3σt2 + ε(1 + ε)1 − 2σ − ε(1 + ε) − t2 (1 + 2ε) = 1 + e e e e . 2 2 4 2 2 σe + (1 − 2σe)t + t + ε(1 + ε) 2σe + ε(1 + ε) + 2t 1 + t 10 CHRISTIAN TAFULA´

2 Since ε(1 + ε)(2σe + ε(1 + ε) + 2t ) > 0, we get Π (s) Π (s) − ε ≥ 0 1 + 2ε (3.7) −1 + 2σ + ε(1 + ε) + t2 ε(1 + ε) 2 e . 2 2 4 2 2 σe + (1 − 2σe)t + t + ε(1 + ε) 2σe + ε(1 + ε) + 2t 1 + t Because ε(1 + ε) = 1 for ε = ϕ − 1, part (i) follows immediately from (3.7). For part (ii), we divide the proof into three steps. Our starting point is the fact that, if 0 < σ < 1, then, for η > −1, σ σ + ε σ − t2/(σ + ε) (3.8) ≤ (1 + η) ⇐⇒ η ≥ ε. σ2 + t2 (σ + ε)2 + t2 σ2 + t2

Π (s) ε2 • Step 1: ε < Π (s) < 1 + Π (s) for s ∈ R . 1 + 2ε 0 1 + ε ε 1

For the lower bound, it suﬃces to note that, for every s ∈ R1, we have 2 −1 + 2σe + ε(1 + ε) + t > 0, and thus, from (3.7), we obtain Π0(s) − Πε(s)/(1 + 2ε) > 0. For the upper bound, we use (3.8). For s ∈ R1, we have σ − t2/(σ + ε) 1 1 1 ε2 ε < − ε ≤ 1 − ε = , σ2 + t2 t2 1 + ε 1 + ε 1 + ε

2 and thus, taking η := ε /(1 + ε) in (3.8) makes the inequality Π0(s) < (1 + η)Πε(s) valid for every s ∈ R1. Π (s) • Step 2: ε < Π (s) ≤ (1 + Mε)Π (s) for s ∈ R ∪ R . (1 + 2ε)(1 + 2Mε(1 + ε)) 0 ε 2 3 We start with the lower bound. The denominator of (3.6) is always positive, and so is the numerator, for every ε > 0. Thus, Π (s) (3.9) Π (s) − ε ≥ 0 (1 + 2ε)(1 + g(ε)) σ − σeε + 1 − 1 + σ − σeε t2 + 1 − 1 t4 ! e 1+g(ε) 1+g(ε) e 1+g(ε) 1+g(ε) 1 2 2 2 4 2 , σe + (1 − 2σe)t + t 1 + t where g(ε) > 0 is some function of ε > 0. Therefore, since σ 1 σ σ − eε + 1 − + σ − eε t2 e 1 + g(ε) 1 + g(ε) e 1 + g(ε) g(ε) ε(1 + ε) = (σ + σt2 + t2) − (1 + t2), 1 + g(ε) e e 1 + g(ε) in order for (3.9) to be strictly positive it suﬃces to have 1 + t2 t2 −1 (3.10) g(ε) ≥ ε(1 + ε) = ε(1 + ε) σ + . 2 2 e 2 (1 + t )σe + t 1 + t ON LANDAU–SIEGEL ZEROS AND HEIGHTS OF SINGULAR MODULI 11

For s ∈ R , we have σ = σ(1 − σ) ≥ M −1(1 − M −1) ≥ 1 M −1 (since M ≥ 2), while 2 e 2 2 2 1 −1 for s ∈ R3 we have t /(1 + t ) ≥ 2 M . Thus, in both cases, it suﬃces to take g(ε) := 2Mε(1 + ε) in order for g to satisfy (3.10), giving us the desired lower bound. For the upper bound, we have that

2 ( σ − t /(σ + ε) σ ε/σ ≤ Mε, if s ∈ R2, ε ≤ ε ≤ 2 2 2 2 2 σ + t σ + t σε/t ≤ Mε, if s ∈ R3;

implying that taking η := Mε in (3.8) makes Π0(s) ≤ (1 + η)Πε(s) valid for every s ∈ R2 ∪ R3, concluding step 2.

• Step 3: |Π0(s) − Πε(s)| < 5Mε Πε(s) for s ∈ R1 ∪ R2 ∪ R3, 0 < ε < 1.

Since M ≥ 2 and ε < 1, we have from step 1 that, for s ∈ R1, 2ε ε2 |Π (s) − Π (s)| < max , Π (s) 0 ε 1 + 2ε 1 + ε ε 2 < max{2ε, ε } Πε(s)

≤ 2ε Πε(s)(< 5Mε Πε(s)),

and from step 2 we have, for s ∈ R2 ∪ R3, that 2ε M(1 + ε) |Π (s) − Π (s)| ≤ max 1 + , Mε Π (s) 0 ε 1 + 2ε 1 + 2Mε(1 + ε) ε 1 ≤ max 2Mε + 1 + ε , Mε Π (s) M ε

< 5Mε Πε(s). thus proving part (ii). We are now ready to prove the main propositions of this section. 3.2. Proofs of Propositions 3.1 and 3.2. Proof of Proposition 3.1. Write Se for the multiset consisting of %, 1 − % for each 1 % ∈ S, so that |S|e = 2|S|, and write G = 2 (γ + log 2π + χ(−1) log 2). From the deﬁnition of the pairing-up function (3.4), we have 1 1 1 1 Π (%) < , < ≤ 0 2 % 2 1 − % 4 −1 and Πε(%)/4 ≤ ε , for 0 < <(%) < 1 and ε > 0. Thus, from (3.5) and Lemmas 3.3 (ii), 3.4 (i) we obtain:

0 ! L X Π0(%) X Π0(%) 1 X Π0(%) < (1, χ) = + − log q + G − L 4 4 2 4 %∈Se %(χ) %∈Se ! ! X 1 1 X Πϕ−1(%) 1 X Πϕ−1(%) > < + − log q + G − 1 − % 2ϕ − 1 4 2 4 %∈S %(χ) %∈Se 1 1 + 1 − − log q + G 2ϕ − 1 2 12 CHRISTIAN TAFULA´

! X 1 1 1 2|S| 1 1 > < − 1 + + − 1 − log q 1 − % 2ϕ − 1 ϕ − 1 ϕ − 1 2ϕ − 1 2 %∈S ! √ X 1 1 1 ( 5 + 1)(2|S| + 3) = < − 1 − √ log q − √ + 1. 1 − % 2 %∈S 5 2 5

The proposition then follows by rearranging the terms.

Proof of Proposition 3.2. Writing B = BM , we have from (3.5) that

0 ! L X 1 X Π0(%) − Πε(%) < (1, χ) − = L 1 − % 4 %(χ)∩B %(χ)\Be | {z } =: S1 ! X Πε(%) X 1 − − < 4 % %(χ)∩Be %(χ)∩B | {z } =: S2 ! X Πε(%) 1 1 + − log q + γ + log 2π + χ(−1) log 2 . 4 2 2 %(χ) | {z } =: S3 P From Lemma 3.4 (ii), we have |S1| < 5Mε %(χ) Πε(%)/4, and thus, from Lemma 3.3 5 −1 (i), it follows that |S1| < 2 Mε log q+5M. Next, since Πε(%)/4 < ε for 0 < <(%) < 1, −1 and <(1/%) ≤ M for % ∈ B, we have |S2| < 2 |%(χ) ∩ B|/ε. Finally, from Lemma −1 3.3 (ii),√ we have |S3| ≤ 1 + ε . Then, putting everything together and substituting ε = 1/ M log q yields ! ! L0 X 1 7 < (1, χ) − < + 2|%(χ) ∩ B| pM log q + (5M + 1). L 1 − % 2 %(χ)∩B Note that 5M + 1 ≤ 11M/2. By taking M := f(q)(≤ 4 log q), we have f(q) ≤ p 2 f(q) log q, and the estimate from Proposition 3.2 follows.

0 Q 3.3. Chang’s zero-free regions: Theorem 1.4. For q ≥ 2, write q := p|q p, 0 Kq := log q/ log q , and P(q) for the largest prime divisor of q. Then, there is an effectively computable constant c > 0 such that, for every T ≥ 1, the region 0 1 log log q 1 s = σ + it σ ≥ 1 − c min , 0 , 9/10 log P(q) (log q )(log 2Kq) (log qT ) is zero-free for |t| < T , with the possible exception of a simple, real zero in this region in case χ is real (Theorem 10 of Chang [1]). In particular, it follows that we can take 1 (log q0)(log 2K ) f(q) := max log P(q), q , (log q)9/10 c log log q0 ON LANDAU–SIEGEL ZEROS AND HEIGHTS OF SINGULAR MODULI 13 in (3.1), so that the only possible element in 1 %(χ) ∩ s ∈ C σ ≥ 1 − , |t| ≤ 1 ⊇ %(χ) ∩ Bf f(q) is the potential Siegel zero, and hence the only potential term in the summation over zeros in Proposition 3.2. Writing L(q) := log log q/ log log q0, it becomes clear that (log q0)(log 2K ) (log 2)L(q) q = + L(q) − 1 (log q)1/L(q) = o(log q), log log q0 log log q and thus, for sufficiently smooth moduli, Chang’s zero-free region is much wider than the classical one (which yields O(log q)). More precisely, if δ > 0 and q is qδ-smooth, −1 then there is Nδ ∈ R>0 such that f(q) ≤ c δ log q for all q > Nδ. This combined with Proposition 3.2 proves Theorem 1.4. 1 t2 −1 3.4. Proof of Corollary 1.5. Using that <( 1−s ) = ((1 − σ) + 1−σ ) , we can apply Proposition 3.1 to deduce that a number s = σ + it in the critical strip cannot be a zero of L(s, χD) if ! t2 1 1 L0 1 > (1 − σ) + 1 − √ log |D| + (1, χD) + O(1) . 1 − σ 5 2 L | {z } =: X Multiplying by 1 − σ and rearranging we get the equation (1 − σ)2X − (1 − σ) + t2X < 0, −1 which, setting t = 0 and solving in 1 − σ yields 0 < 1 − σ < X−1. As (1 − √1 ) 1 = √ √ 5 2 5ϕ for ϕ = 1+ 5 , it follows from Theorem 1.2 that, under weak uniform abc, we √2 have X−1 > ( 5ϕ + o(1)) (log |D|)−1, thus proving the first assertion. For the second assertion, we have from Theorem 1.4 that, for any fixed δ > 0, there δ is Nδ ∈ R>0 such that, if |D| > Nδ is |D| -smooth, then any real zero 0 < β < 1 of L(s, χD) must satisfy 0 1 L 1 0 < < (1, χ ) + Mδ 2 log |D|, 1 − β L D where M > 0 is some absolute constant. It follows from Theorem 1.2 that, under mδ−1/2+o(1) −1 weak uniform abc, we must have β < 1 − log |D| , where m = M . Changing − 1 −1 o(1) to oδ(1) allows us to drop the condition |D| > Nδ, and since mδ 2 = oδ→0(δ ), Theorem 1.4 allows us to extend vertically the zero-free region obtained to the height 1 box {s | |t| ≤ 1}, concluding the proof.

L0 4. The bridge from L (1, χD) to ht(j(τD)) We will now prove Theorem 1.1, which is how we connect Siegel zeros with abc. The proof essentially gets reduced down to a calculation once three concepts are introduced: the framework of Euler–Kronecker constants due to Ihara [9], Kronecker’s limit formula (KLF), and the uniform distribution of Heegner points due to Duke [6]. After brieﬂy describing each of these, we compute the constant C from Theorem 1.1 with Lemmas 4.3–4.5 and ﬁnish the proof in subsection 4.5. 14 CHRISTIAN TAFULA´

4.1. Euler–Kronecker constants. Let K/Q be a number ﬁeld, OK its ring of integers, and ζ (s) = P [O : a]−s its Dedekind zeta function. In 2006, Ihara K a⊆OK K [9] introduced and studied the Euler–Kronecker constants γK ∈ R, which are deﬁned 0 as the constant term in the Laurent expansion of ζK /ζK at s = 1: 0 ζK 1 (s) = − + γK + O(s − 1) (s → 1). ζK s − 1

Write C`K for the ideal class group of K, and hK (= |C`K |) for its class number. For each A ∈ C`K , the partial zeta function ζK (s, A ) is given by X 1 ζK (s, A ) := s , [OK : a] a∈A a integral and so we have ζ (s) = P ζ (s, ). Roughly following Ihara’s naming scheme, K A ∈C`K K A we deﬁne the Kronecker limits K(A ) as the constant term in the Laurent expansion 0 of ζK /ζK (s, A ) at s = 1. The expansion of ζK (s, A ) at s = 1 is given by κK ζK (s, A ) = + K K(A ) + O(s − 1) (s → 1), s − 1 κ

where κK , which is independent of A , is determined by the analytic class number formula. Hence, the Euler–Kronecker constant of K is the average of its Kronecker limits: 1 X (4.1) γK = K(A ). hK A ∈C`K Using the classical limit formulas of Kronecker (for imaginary quadratic fields) and Hecke (for real quadratic fields – see Zagier [20]), together with the equidistribution results in Theorem 1 of Duke [6], one can obtain estimates for γK of quadratic fields in terms of cycle integrals with good error terms. We will do this in detail for the imaginary quadratic case.

4.2. Kronecker’s limit formula. The classical real analytic Eisenstein series is given by X =(τ)s E(τ, s) = , |mτ + n|2s 2 (m,n)∈Z (m,n)6=0 for τ ∈ h and <(s) > 1.5 This is the simplest example of a “non-holomorphic modular function”, meaning it is invariant under modular transformations τ 7→ τ 0 = α β (ατ + β)/(γτ + δ) for γ δ ∈ SL2(Z). In the notation of (2.2), we have the identity 1 2 s (4.2) ζ √ (s, ) = E(τ , s). Q( D) A p A wD |D|

5Some authors consider slightly modiﬁed versions of this function, such as divided by 2, or with the pair (m, n) running through relatively prime integers (equivalent to considering ζ(2s)−1 E(τ, s)), or multiplied by π−sΓ(s). As we are following Siegel [16], we stick to his convention. ON LANDAU–SIEGEL ZEROS AND HEIGHTS OF SINGULAR MODULI 15

√ where wD is the number of roots of unity in Q( D). The residue and constant term in the Laurent expansion of E(τ, s) at s = 1 are given by π π E(τ, s) = + π =(τ) − log =(τ) + U(τ) (4.3) s − 1 3 + 2πγ − log 2 + O(s − 1), where ! X X 1 cos(2πn <(τ)) π (4.4) U(τ) := 4 = −2 log(|η(τ)|2) − =(τ) ; d e2πn =(τ) 3 n≥1 d | n this expansion is also sometimes called “Kronecker’s limit formula”.6 Here, η is Dedekind’s η-function, deﬁned as Y η(z) := q1/24 (1 − qk) k≥1 for z = x + iy ∈ h, where q := e2πiz and “q1/24” = eπiz/12. One checks that ! ! ! X X 1 cos(2πnx) X X 1 U(z) = 4 = 4 < qn d e2πny d n≥1 d | n n≥1 d | n X X qdk X (4.5) = 4 < = −4 < log(1 − qk) d d≥1 k≥1 k≥1 X π = −4 log |1 − qk| = −2 log(|η(z)|2) − y, 3 k≥1 so the formula in (4.3) is indeed equivalent to the usual formulation. Remark. The function U in (4.4) is based on Stopple’s notation in p. 867 of [17]. Although (4.3) is usually given in terms of η, this form of the statement emphasizes π2 the dominant part “ 3 =(τA )” in the constant term of E(τA , s). Taking logarithmic derivatives in (4.2) and (4.3) immediately yields: Lemma 4.1 (KLF). For each A ∈ C`(D), we have π 1 (4.6) K(A ) = =(τ ) − log =(τ ) + U(τ ) − log |D| + 2γ − log 2 . 3 A A A 2 | {z } | {z } | {z } A -dependent term D-dependent constant term 4.3. Duke’s equidistribution theorem. In 1988, W. Duke showed that the set ΛD of Heegner points is uniformly distributed in SL2(Z)\h. To state his result, consider the probability space (F , Σ, µ), where: • F ⊆ h is the usual fundamental domain of SL2(Z)\h; • Σ is the usual σ-algebra of Lebesgue measurable sets inherited from R2 ⊇ h; 3 2 • dµ := π dxdy/y for z = x + iy ∈ F , so that µ(F ) = 1. Endowing h with its usual hyperbolic structure, we have the following:

6Cf. Chapter 20, §4 of Lang [13]. 16 CHRISTIAN TAFULA´

Duke’s Theorem (Theorem 1 i) in [6]). Let Ω ⊆ F be a convex set (in the hyperbolic sense) with piecewise smooth boundary. Then, there is a real number δ = δ(Ω) > 0 such that

|ΛD ∩ Ω| −δ (4.7) = µ(Ω) + OΩ,δ(|D| ), |ΛD| where the implied constant is ineﬀective.7 Since hyperbolic convex subsets of h constitute a basis for the usual topology inherited from R2 ⊇ h, the following is a direct corollary: Lemma 4.2. Let f : F → C be a Riemann-integrable function in (F , Σ, µ). Then: ! 1 X 3 Z dxdy lim f(τA ) = f(x + iy) . D→−∞ h(D) π y2 A ∈C`(D) F 4.4. Computing C. With Lemma 4.2, we are able to not only bound but also compute the average over A ∈ C`(D) of the non-dominant terms in Kronecker’s limit formula (4.6). While κ1 and κ3 are more easily shown to be bounded without Duke’s theorem, the main point of the next three lemmas is the computation of C in Theorem 1.1. Z 1 + Lemma 4.3. κ1 := − log |j(z)| − 2πy dµ = 0.011448 ... 6 F 2πiz −2πy Proof. Writing q = e , we have |q| = e , so we can rewrite κ1 as Z 1 dxdy (4.8) κ1 = − log max |j(z)| · |q|, |q| 2 . 2π F y We will prove the convergence of (4.8) in three steps, and then we will estimate its value numerically. Writing c(n) for the n-th coeﬃcient in the q-expansion of the j-invariant in (2.3), we have the following: √ • Step 1: For every n ≥ 1, we have 0 ≤ c(n) < e4π n. n −1 P nk Since (1 − q ) = k≥0 q , it is clear from the q-expansion of j(z) in (2.3) that the c(n) are nonnegative. To show the upper bound, we use the fact that j is a modular function of weight 0 for SL2(Z). For every 0 < t < 1, we have j(i/t) = j(it). Thus, in terms of the q-expansion, rearranging this equality yields: X e2πn/t − e2πnt c(n) = e2π/t − e2πt. e2πn(t+t−1) n≥0 For n = 1 we have c(1) = 196884 < e4π. For n ≥ 2, since the c(n) are nonnegative, we have 2π/t 2πt e − e −2 c(n) ≤ e2πnt(1+t ), e2πn/t − e2πnt

7This comes from the use of Siegel’s theorem in the proof. This ineﬀectiveness is hence passed down to all subsequent calculations. ON LANDAU–SIEGEL ZEROS AND HEIGHTS OF SINGULAR MODULI 17

√ so taking t = t(n) := 1/ n, it follows that √ √ √ 2π n 2π/ n √ −2π n (1−n−1) √ √ e − e 2π n(1+n) 1 − e 4π n 4π n c(n) ≤ √ √ e = √ e < e , e2πn n − e2π n 1 − e−2π n (n−1) as desired. 1 R 2 −21 • Step 2: log max{|j(z)| · |q|, |q|} dxdy/y < 10 . 2π F ∩{=(z)≥16}√ For y ≥ 4, we have 4π k − 2πky ≤ −πky for every k ≥ 1. Thus, by step 1, for y ≥ 4, it follows that X X √ X |j(z)| · |q| ≤ 1 + c(n)|q|n+1 ≤ 1 + e4π n+1 − 2π(n+1)y ≤ 1 + e−πny. n≥0 n≥0 n≥1 By partial summation, X Z +∞ Z +∞ 1 e−πny = πy btc e−πty dt ≤ πy t e−πty dt = 1 + e−πy. πy n≥1 1 1 Hence, as log(1 + t) ≤ t for all t ≥ 0, we have Z Z +∞ −1 1 dxdy 1 1 + (πy) log max{|j(z)| · |q|, |q|} 2 ≤ 2 πy dy 2π F ∩{=(z)≥16} y 2π 16 y e 1 Z +∞ 1 + πy ≤ 2 πy dy 2π 16 y e 1 1 = . 2π 16 e16π Then, since e4π > 105, we have e16π > 1020 and 32π > 10, yielding step 2. • Step 3: 1 R log max{|j(z)| · |q|, |q|} dxdy/y2 = −0.011448 ... 2π F Since log max{|j(z)| · |q|, |q|} is continuous in the closure of F ∩ {=(z) < 16}, which is compact, the integral 1 Z dxdy (4.9) − log max{|j(z)| · |q|, |q|} 2 2π F ∩{=(z)<16} y converges. This, together with step 2, implies that (4.8) converges. Furthermore, by step 2, in order to obtain a computational estimate of (4.8) with 20 decimal places of accuracy, it suﬃces to estimate (4.9) to 20 decimal places of accuracy. Using Python’s mpmath library, the ﬁrst 6 decimal places are = 0.011448 ... Lemma 4.4. We have: Z 3 X sin(2πn/3) κ := log(y) dµ = 1 − log 2 + (= 0.952984 ...). 2 π n2 F n≥1 Proof. Using that R cot(ϑ) dϑ = log(sin(ϑ)) and R cot(ϑ)2 dϑ = −ϑ−cot(ϑ) (omitting the integration constants), we have: 3 Z dxdy log(y) π y2 F √ +∞ 1 ! 1 1−y2 ! 6 Z Z 2 log(y) 6 Z Z log(y) = √ 2 dx dy − √ 2 dx dy π 3 y π 3 y 2 0 2 0 18 CHRISTIAN TAFULA´

√ √ 2 3 3 6 Z 1 log(y)p1 − y2 = log + 1 − √ 2 dy π 2 π 3 y 2 √ √ π 2 3 3 6 Z 2 = log + 1 − log(sin(ϑ)) cot(ϑ)2 dϑ π 2 π π 3 2π 3 Z 3 (4.10) = 1 + log(sin(ϑ)) dϑ. π π 3 Since sin(−i log(ξ)) = i(1 − ξ2)/2ξ, making the substitution ξ := eiϑ yields:

2π 2 ! Z 3 Z 1 − ξ i dξ log(sin(ϑ)) dϑ = −i log + log π 2 ξ ξ 3 Γ 1 = i log(2) log(eπi/3) + Li (e2πi/3) − Li (e4πi/3) 2i 2 2 π = − log(2) + =Li (ω), 3 2 where ω := e2πi/3, Γ := {eiϑ | ϑ ∈ [π/3, 2π/3]} is oriented counterclockwise, and P n 2 Li2(z) := n≥1 z /n is the dilogarithm function. Putting this together with (4.10), we deduce the lemma by applying Lemma 4.2 and by observing that for every iϑ P 2 ϑ ∈ (−π, π) we have =(Li2(e )) = n≥1 sin(nϑ)/n . Lemma 4.5. Writing ω := e2πi/3, we have: ! ! Z 24 X X 1 Z ω−1 dz κ := U(z) dµ = = e2πinz 3 π d z F n≥1 d | n ω (= −0.000303 ...). Proof. From Equation (4.5), for z = x + iy ∈ h we have ! ! X X 1 (4.11) U(z) = 4 < e2πinz d n≥1 d | n Writing ξ = cos(ϑ) + i sin(ϑ), we have √ Z dxdy 1 Z 1 h x= 1−y2 dy 2πinz 2πinz e 2 = − √ e √ 2 y 2πin 3 x=− 1−y2 y F 2 2π Z 3 1 2πinξ cos(ϑ) = − e 2 dϑ 2πin π sin(ϑ) 3 Z 2πinξ 2ξ = −i e 2 dξ, Γ ξ − 1 where Γ := {eiϑ | ϑ ∈ [π/3, 2π/3]} is oriented counterclockwise. Since 2ξ/(ξ2 − 1) = (ξ − 1)−1 + (ξ + 1)−1, we have Z Z Z 2πinξ 2ξ 2πinξ dξ 2πinξ dξ −i e 2 dξ = −i e + e Γ ξ − 1 Γ −1 ξ Γ +1 ξ ON LANDAU–SIEGEL ZEROS AND HEIGHTS OF SINGULAR MODULI 19

! Z ω−1 dξ = 2 = e2πinξ , ω ξ

which is a real number. Thus, putting this together with (4.11), we get the series in the statement of the lemma. To see that κ3 converges, note that Z ω−1 2πinξ dξ −2πn=(ω) 1 e ≤ e , ω ξ |ω| √ −1 P P −1 −πn 3 and hence, |κ3| ≤ 24(π|ω|) n≥1 d|n d e , which converges.

4.5. Proof of Theorem 1.1. Since j(τD) is integral, (2.4) gives us 1 X ht(j(τ )) = log+ |j(τ )|. D h(D) A A ∈C`(D) Hence, from (4.1), we deduce that 1 1 γ √ = ht(j(τ )) − log |D| + κ − κ + κ + 2γ − log 2 + o(1) Q( D) 6 D 2 1 2 3 by Lemmas 4.1, 4.2, and 4.3–4.5. From the identity ζ √ (s) = ζ(s)L(s, χ ) we Q( D) D have γ √ = γ + L0 (1, χ ), concluding the ﬁrst part of Theorem 1.1. Q( D) L D To prove (1.4), one simply expands the deﬁnition of ht(j(τD)) and uses the q- expansion of the j-invariant, thus obtaining p 1 X π |D| ht(j(τ )) = + O(1). D h(D) a (a,b,c)

Rearranging the expression of the ﬁrst part of Theorem 1.1 yields (1.4).

Remark 4.6. Theorem 1.1 is reminiscent of a result by Colmez [2, 3]. Writing ED/C for an elliptic curve with complex multiplication by Z[τD], Colmez showed that, 8 writing htFal for the Faltings height, 1 L0 −2 ht (E ) = log |D| + (0, χ ) + log 2π. Fal D 2 L D Together with Theorem 1.1, this yields 1 ht (E ) = ht(j(τ )) + + o (1), Fal D 12 D κ D→−∞ C γ 9 where κ = 2 − 2 −log 2π = −2.655370 .... It is interesting to note that in Remarque (ii), p. 364 of [3], Colmez hinted at a possible connection between abc and Siegel zeros, two years before the work of Granville and Stark.

8See the diagram in subsection 0.6, p. 663 of [2] – note that there is a typo: the upper right 1 corner should read “−2htFal(X) − 2 log D” instead. 9Cf. the standard “O(log(ht))” in Chapter X of Cornell–Silverman [4]. 20 CHRISTIAN TAFULA´

5. Estimating ht(j(τD)) with abc We are now going to prove Theorem 1.2. Already from Theorems 1.4 and 1.1 we have 1 L0 (5.1) lim inf (1, χD) ≥ 0, D→−∞ log |D| L |D|o(1)-smooth so all we need is to work out the upper bounds. That will be done using the two uniform formulations of the abc-conjecture mentioned in the introduction, following Granville–Stark’s method.

5.1. Uniform abc-conjecture. We start with the usual abc-conjecture for number 1/[K:Q] fields. Let rdK = |∆K | be the root-discriminant of K/Q. Conjecture 5.1 (abc-conjecture for number fields – see Vojta [18]). Let K/Q be a number field. For every ε > 0, there is a constant C = C(K, ε) ∈ R≥0 such that, for any a, b, c ∈ K with a + b + c = 0, the following holds: (5.2) ht([a : b : c]) < (1 + ε) NK ([a : b : c]) + log(rdK ) + C(K, ε).

Remark. For a, b, c ∈ Z coprime, we have ht([a : b : c]) = log max{|a|, |b|, |c|} and Q NQ([a : b : c]) = log p|abc p , recovering the abc-conjecture for Z. For the method to work, we also need information about the growth of C(K, ε) as a function of K. The approach we take is to assume that C(K, ε) is dominated by the contribution “log(rdK )” of ramiﬁed primes in the inequality, in two senses of the concept of domination: either big-O or small-o. Both have implications to the Siegel zeros problem, and we state the usual uniform abc-conjecture for comparison:

Conjecture 5.2 (Uniformity). Let K/Q be a number ﬁeld and ε > 0. Then:

(i) (O-weak uniform abc) C(K, ε) = Oε(log(rdK )).

(ii) (Weak uniform abc) C(K, ε) = oε(log(rdK )) as rdK → +∞.

(iii) (Uniform abc) C(K, ε) = Oε(1). It is remarked in p. 510 of Granville–Stark [7] that Conjecture 5.2 (iii) follows from Vojta’s General Conjecture under the assumption that [K : Q] is bounded; consequently, so does (i) and (ii).10 Remark 5.3. See also Remark 2.2.3 of Mochizuki’s “IUT IV” [14], in which it is explained that the calculations of Corollary 2.2 (ii), (iii) of IUT IV can be regarded as a sort of “weak” version of uniform abc. Such version, however, is much weaker than the O-weak uniform abc in Conjecture 5.2 (i), and thus, in principle, one is not able to deduce “no Siegel zeros” from Corollary 2.2 of [14] by using the methods we are employing here. √ 10 π n 2 Writing n = [K : Q], the Hermite–Minkowski theorem says that rdK ≥ 4 (n/ n!) . Hence, from Stirling’s formula, it follows that log(rdK ) ≥ 2 + log(π/4) + O(log n/n) 1 uniformly for n ≥ 2, and so (iii) =⇒ (ii) =⇒ (i). ON LANDAU–SIEGEL ZEROS AND HEIGHTS OF SINGULAR MODULI 21

5.2. Proof of Theorem 1.2. Consider the Weber modular functions γ2, γ3, related to the j-invariant via the identities

3 2 (5.3) j(τ) = γ2(τ) = γ3(τ) + 1728.

For each D < 0, the pair (γ2(τD), γ3(τD)) provides a solution (x, y) to the Diophantine 3 2 equation x −y =√ 1728. This solution lies in the extension He D√:= HD(γ2(τD), γ3(τD)), where HD = Q( D, j(τD)) is the Hilbert class ﬁeld of Q( D). If gcd(D, 6) = 1, 11 then we actually have He D = HD. In general, we have the following: Lemma 5.4 (Lemma 1 of Granville–Stark [7]). rd ≤ 6p|D|. He D

This lemma quantiﬁes the fact that He D/HD has very little ramiﬁcation, which is a key point. We remark that the particular factor of “6” plays little role here, since it would be enough to have rd |D|1/2+ε. He D ε Lemma 5.5. Assume the abc-conjecture (5.2). Then, for every fundamental discriminant D < 0 and 0 < ε < 1/5.01, we have 1 ht(j(τD)) < (1 + ε) 3 log |D| + 6 C(He D, ε) + O(1). 1 − 5ε Proof. Writing M := (1 + ε) log(rd ) + C(H , ε) and applying abc to the equation He D e D 3 2 γ2(τD) − γ3(τD) − 1728 = 0, we get

abc 3 2 3 2 ht([γ2 : γ3 : 1728]) ≤ (1 + ε) NK ([γ2 : γ3 : 1728]) + M 1 3 1 2 ≤ (1 + ε) 3 ht(γ2 ) + 2 ht(γ3 ) + M + O(1) 5 3 2 ≤ (1 + ε) 6 ht([γ2 : γ3 : 1728]) + M + O(1).

Since ht(j(τD)) ≤ ht([j(τD): j(τD) − 1728 : 1728]), the claim of the lemma follows by rearranging the expression above and using Lemma 5.4. Proof of Theorem 1.2. From Lemma 5.4 we have log(rd ) = 1 log |D| + O(1), so it He D 2 follows from Lemma 5.5 together with (5.1) that ht(j(τ )) O-weak uniform abc (Conjecture 5.2 (i)) =⇒ lim sup D < +∞, D→−∞ log |D| ht(j(τ )) weak uniform abc (Conjecture 5.2 (ii)) =⇒ lim sup D = 3; D→−∞ log |D| hence, Theorem 1.2 follows directly from Theorem 1.1.

Appendix A. On “B + P 1/% = 0” Let χ (mod q) be a primitive Dirichlet character, q ≥ 2. In section 3, we stated L0 an expression for L (s, χ) in (3.2) that is diﬀerent from what usually appears in

11Cf. §72 of Weber [19]. 22 CHRISTIAN TAFULA´

textbooks. Many investigations on Dirichlet L-functions have as a starting point the following classical formula:12 0 0 L X 1 1 q 1 Γ s + aχ X 1 (s, χ) = − log − + B + . L s − % 2 π 2 Γ 2 χ % %(χ) %(χ) 1 P where aχ := 2 (1 − χ(−1)), “ %(χ)” denotes a sum over the non-trivial zeros of L(s, χ), and Bχ ∈ C is a constant not depending on s. While the fact that < Bχ + P %(χ) 1/% = 0 is well-known, the more precise X 1 (A.1) B + = 0 χ % %(χ) does not appear to be as much so, despite holding true in similar generality. It follows from Theorem 2 (p. 257) of Ihara–Murty–Shimura [10], for example, that the equivalent of (A.1) holds for finite-order Hecke characters in number fields (thus, in particular, for Dirichlet characters). While the proof in [10] uses Weil’s “explicit formula”, it is possible to prove (A.1) with a simpler, more direct analytic argument, as shown by Lucia [8]. We will reproduce Lucia’s argument here in some generality, given the conspicuous absence of this fact in the literature, showing (A.1) for the extended Selberg class of L-functions, thus highlighting that it is a relatively simple consequence of the functional equation. A.1. The extended Selberg class S]. Following Kaczorowski [12], a Dirichlet P −s series L(s) = n≥1 an n is said to be an L-function in the extended Selberg class (denoted S]) if it satisfies the following properties: P −s (S1) n≥1 an n is absolutely convergent for <(s) > 1. k (S2) (Analytic continuation) There exists k ∈ Z≥0 such that (s − 1) L(s) is entire of finite order. Write k(L) for the smallest such k.

(S3) (Functional equation) There exists a triple (c, Q, γL), where c, Q are positive real numbers, and γL is a function of the form f Y γL(s) := Γ(λjs + µj) j=1

(called a Gamma factor), with f ∈ Z≥1, positive real numbers λj (1 ≤ j ≤ f), and complex numbers W , µj (1 ≤ j ≤ f) with |W | = 1 and <(µj) ≥ 0, for which the completed L-function s (A.2) ΛL(s) := c Q γL(s) L(s)

satisfies a functional equation: ΛL(s) = W ΛL(1 − s). This class and the more restrictive Selberg class (with the additional Euler product and Ramanujan hypothesis properties) contain most of the L-functions appearing in number theory, including L-functions of primitive Dirichlet characters and finite- order primitive Hecke characters in number fields. From now on, let L be a fixed L-function in S]. See section 2 of Kaczorowski [12] for the facts used below.

12Cf. Eqs. (17), (18), Chapter 12, p. 83 of Davenport [5]. ON LANDAU–SIEGEL ZEROS AND HEIGHTS OF SINGULAR MODULI 23

Denote by %(L) the multiset of non-trivial zeros of L (i.e., zeros of ΛL(s)). There is a real number σL ≥ 1 such that 1 − σL ≤ <(%) ≤ σL for every non-trivial zero % ± of L. Writing NL (T ) = #{%(L) | 0 ≤ ±=(%) < T }, by standard techniques based on the argument principle one shows that d (A.3) N −(T ),N +(T ) = L T log T + c T + O(log T ), L L 2π L Pf where dL = 2 j=1 λj is the degree of L, and cL ∈ R is some constant depending only on L. This implies, in particular, that ΛL has order 1 as an entire function, and that P 1/% converges in the principal-value sense “ lim P ”. %(L)6=0 T →+∞ %(L)6=0, |=(%)|≤T Thus, from (S2), (S3), by Hadamard’s factorization theorem we have

Y s (s(1 − s))k(L) Λ (s) = sm0 (1 − s)m1 eA+Bs 1 − es/%, L % %(L)6=0,1 where m0 −k(L), m1 −k(L) are the order of s = 0, 1 as zeros of ΛL(s) (hence negative if it is a pole), respectively, and A, B ∈ C are constants. Since ΛL(0) = ΛL(1), it follows that m0 − k(L) = m1 − k(L); thus, by taking logarithmic derivatives, we derive: ! Λ0 k(L) k(L) X 1 X 1 (A.4) L (s) + + = + B + . ΛL s s − 1 s − % % %(L) %(L)6=0,1 This is our starting point.

A.2. Proof of “B + P 1/% = 0”. Theorem A.1. For L ∈ S], in the notation of (S2), (S3), we have

L0 k(L) k(L) X 1 γ0 (A.5) (s) + + = − log Q − L (s). L s s − 1 s − % γL %(L) Proof. From (S3) we have that, if % is a zero of L, then so is 1 − %. Moreover, %(L) = {1 − % | % ∈ %(L)}, where L(s) = L(s) is the dual of L. Therefore, by (A.2) and (A.4),

0 ! Λ0 Λ X 1 0 = lim L (s) + L (1 − s) = 2 < B + . s→1 ΛL Λ % L %(L)6=0,1

Thus, in view of (A.2), to prove the theorem we just need to show that =B + P L0 γ0 %(L)6=0,1 1/% = 0. Since both =( L (R)), =( γ (R)) → 0 as R → +∞ through the real numbers, by (A.2) and (A.4) it suﬃces to show that

X 1 log R = . R − % R %(L) 24 CHRISTIAN TAFULA´

Let R > 0 be ﬁxed and consider zeros up to height T > 0. Writing % = β + iγ for a generic non-trivial zero, since |β| 1, we have ! X (R − β) + iγ X R − β + iγ R = + O . (R − β)2 + γ2 R2 + γ2 (R2 + γ2)3/2 %(L) %(L) |γ|R while, for the imaginary part of the main term, since from (A.3) it follows that − − |NL (T ) − NL (T )| log T , X γ Z T t Z T t = dN +(t) − dN −(t) R2 + γ2 R2 + t2 L R2 + t2 L %(L) 0 0 |γ|

Acknowledgements I would like to express my deep gratitude to Shinichi Mochizuki, Go Yamashita, and Andrew Granville, for their patient guidance, encouragement, and many useful critiques to this research. I would also like thank Henri Darmon, for his engag- ing responses to email questions, and the anonymous referee, for the invaluable recommendations on writing.

References 1. M-.C. Chang, Short character sums for composite moduli, J. Anal. Math. 123 (2014), 1–33. 2. P. Colmez, Périodes des variétésabéliennesàmultiplication complexe, Ann. Math. 138 (1993), no. 3, 625–683. 3. , Sur la hauteur de faltings des variétésabéliennesàmultiplication complexe, Compos. Math. 111 (1998), 359–368. 4. G. Cornell and J. H. Silverman (eds.), Arithmetic geometry, Springer, 1986. 5. H. Davenport, Multiplicative number theory, 3rd ed., Graduate Texts in Mathematics, vol. 74, Springer, 2000. 6. W. Duke, Hyperbolic distribution problems and half-integral weight Maass forms, Invent. Math. 92 (1988), 73–90. 7. A. Granville and H. M. Stark, ABC implies no “Siegel zeros” for L-functions of characters with negative discriminant, Invent. Math. 139 (2000), 509–523. 8. Lucia (https://mathoverflow.net/users/38624/lucia), Hadamard factorization of L-functions, MathOverflow, URL: https://mathoverflow.net/q/370538 (version: 2020-09-01). 9. Y. Ihara, On the Euler–Kronecker constants of global fields and primes with small norms, Algebraic Geometry and Number Theory (V. Ginzburg, ed.), Progress in Mathematics, vol. 253, Birkhäuser,Boston, 2006, pp. 407–451. ON LANDAU–SIEGEL ZEROS AND HEIGHTS OF SINGULAR MODULI 25

10. Y. Ihara, V. K. Murty, and M. Shimura, On the logarithmic derivatives of Dirichlet L-functions at s = 1, Acta Arith. 137 (2009), 253–276. 11. H. Iwaniec and E. Kowalski, Analytic number theory, AMS Colloquium Publications, vol. 53, American Mathematical Society, 2004. 12. J. Kaczorowski, Axiomatic theory of L-functions: the Selberg class, Analytic Number Theory (Cetraro, 2002) (A. Perelli and C. Viola, eds.), Springer, 2006, pp. 133–209. 13. S. Lang, Elliptic functions, 2nd ed., Springer-Verlag, New York, 1987. 14. S. Mochizuki, Inter-universal Teichm¨uller theory IV: Log-volume computations and set-theoretic foundations, Publ. Res. Inst. Math. Sci. 57 (2021), no. 1, 627–723. 15. H. L. Montgomery and R. C. Vaughan, Multiplicative number theory I: Classical theory, Cambridge Studies in Advanced Mathematics, vol. 97, Cambridge University Press, Cambridge, 2006. 16. C. L. Siegel, Advanced analytic number theory, Tata Institute of Fundamental Research Studies in Mathematics, vol. 9, Tata Institute of Fundamental Research, Bombay, 1980. 17. J. Stopple, Notes on the Deuring–Heilbronn phenomenon, Not. Am. Math. Soc. 53 (2006), 864–875. 18. P. Vojta, Diophantine approximations and value distribution theory, Lecture Notes in Math., vol. 1239, Springer, 1987. 19. H. Weber, Lehrbuch der Algebra, vol. 3, 3rd ed., AMS Chelsea Publishing, 1908. 20. D. Zagier, A Kronecker limit formula for real quadratic ﬁelds, Math. Ann. 213 (1975), 153–184.

Department´ de Mathematiques´ et Statistique, Universite´ de Montreal,´ CP 6128 succ Centre-Ville, Montreal, QC H3C 3J7, Canada Email address: [email protected]