Variations on Artin’s Primitive Root Conjecture
by
Adam Tyler Felix
A thesis submitted to the Department of Mathematics & Statistics in conformity with the requirements for the degree of Doctor of Philosophy
Queen’s University Kingston, Ontario, Canada August 2011
Copyright c Adam Tyler Felix, 2011 To My Parents
i Abstract
Let a ∈ Z be a non-zero integer. Let p be a prime such that p - a. Define the index
∗ of a modulo p, denoted ia(p), to be the integer ia(p) := [(Z/pZ) : ha mod pi]. Let
Na(x) := #{p ≤ x : ia(p) = 1}. In 1927, Emil Artin conjectured that
Na(x) ∼ A(a)π(x)
where A(a) > 0 is a constant dependent only on a and π(x) := {p ≤ x : p prime}.
Rewrite Na(x) as follows: X Na(x) = f(ia(p)), p≤x
where f : N → C with f(1) = 1 and f(n) = 0 for all n ≥ 2. We examine which other functions f : N → C will give us formulæ
X f(ia(p)) ∼ caπ(x), p≤x
where ca is a constant dependent only on a. Define ω(n) := #{p|n : p prime}, Ω(n) := #{d|n : d is a prime power} and
ii d(n) := {d|n : d ∈ N}. We will prove
X x (log(i (p)))α = c π(x) + O a a (log x)2−α−ε p≤x X x log log x ω(i (p)) = c0 π(x) + O a a (log x)2 p≤x X x log log x Ω(i (p)) = c00π(x) + O a a (log x)2 p≤x and X x d(i ) = c000π(x) + O a a (log x)2−ε p≤x for all ε > 0.
We also extend these results to finitely-generated subgroups of Q∗ and E(Q) where E is an elliptic curve defined over Q.
iii Acknowledgments
First, I would like to thank my supervisor, Dr. Ram Murty, for his patience, exper- tise, and knowledge. He has a perspicacity in analytic number theory seen in only few individuals on this planet. The past five years have been enlightening. I would also like to thank the members of the defense examination for their com- ments: Drs. E. Kani, O. Bogoyavlenskij, A. Akbary, S. Akl, and R. Henriksen. I would also like to thank my alma mater, the University of Waterloo, for preparing me so well in mathematics. I would especially like to thank Dr. Yu-Ru Liu for intro- ducing me the analytic number theory, sieve theory, and having confidence in me. I would also like to thank Drs. S. Furino and C. Struthers and Ms. Shaw for making mathematics challenging and enjoyable. Finally, I want to thank my family for their unconditional love and support. My father for showing a good work ethic will get you anywhere. My brother for providing humour whenever it was needed. My sister for always wanting to play games. My mother for everything, you have taught so much and I will never be able to repay you. Thank you and I love you all.
iv Statement of Originality
I hereby certify that all of the work described within this thesis is the original work of the author. Any published (or unpublished) ideas and/or techniques from the work of others are fully acknowledged in accordance with the standard referencing practices.
v Table of Contents
Abstract ii
Acknowledgments iv
Statement of Originality v
Table of Contents vi
1 Introduction 1 1.1 Artin’s Conjecture on Primitive Roots ...... 1 1.2 Generalizing Artin’s Conjecture ...... 3 1.3 Definitions and Notations ...... 6 1.3.1 Arithmetic Functions ...... 7 1.3.2 Asymptotic, Big O, and Little o Notation ...... 9 1.3.3 Elliptic Curves ...... 10 1.3.4 The GRH and the Dedekind Zeta Function ...... 11 1.4 Organization of the Thesis ...... 14
Chapter 2: A New Method ...... 16 2.1 Introduction ...... 16 2.2 Lemmata ...... 19 2.3 The Function (log n)α ...... 23 2.4 The Divisor Function ...... 26 2.4.1 The First Summation ...... 27 2.4.2 The Second Summation ...... 28 2.4.3 The Third Summation ...... 30 2.4.4 Computation of the Constant ...... 32
Chapter 3: Finitely-Generated Analogues ...... 40 3.1 Statement of Theorems ...... 40
vi 3.1.1 The r-rank Theorem ...... 41 3.1.2 An Immediate Corollary ...... 44 3.1.3 The Fixed Index Theorem ...... 46 3.2 An Asymptotic Formula ...... 47 3.3 The Fixed Index Theorem ...... 55
Chapter 4: Elliptic Curve Analogues ...... 61 4.1 Statement of Theorems ...... 63 4.2 A Divisibility Criterion ...... 67 4.3 An Asymptotic Result ...... 71 4.3.1 Small Primes ...... 72 4.3.2 Intermediate and Large Primes ...... 74 4.3.3 Optimal Error Terms ...... 76 4.4 Proofs of Theorems 4.1.2 and 4.1.3 ...... 79
Chapter 5: Summations related to the Artin’s Conjecture ...... 84 5.1 Generalizing the Titchmarsh Divisor Problem ...... 84 5.1.1 Application to a Generalization of Artin’s Conjecture . . . . . 89 5.2 The number of prime and prime power divisors of ia(p) ...... 94 5.3 Fomenko’s Result on Average ...... 99
Chapter 6: A Conjecture of Erd˝osand Fomenko’s Theorem . . . . . 105 6.1 Erd˝os’sConjecture ...... 105 6.2 Proof of Theorem 6.1.4 ...... 108
Chapter 7: Further Research ...... 111 7.1 A Second Conjecture of Fomenko ...... 111 P 7.2 Difficulties with p≤x log(ia(p)) ...... 114 7.3 Other Generalizations and Potential Improvements ...... 116 7.3.1 Elliptic Curves ...... 116 Towards a Chebotarev Condition ...... 117 7.3.2 Other Sequences and Functions ...... 118 7.4 Unconditional Lower Bounds ...... 120
Bibliography ...... 123
vii Chapter 1
Introduction
1.1 Artin’s Conjecture on Primitive Roots
Let p ∈ N be a prime number. Let us consider (Z/pZ)∗ := {a mod p : p - a}. It can be shown that (Z/pZ)∗ is cyclic. That is, there exists a ∈ Z such that (Z/pZ)∗ = ha mod pi. In this case we say that a is a primitive root modulo p. In fact, it can also be shown that the number of generators of (Z/pZ)∗ which have the form a mod p in (Z/pZ)∗ is ϕ(p − 1) where ϕ(n) = #{1 ≤ k ≤ n : gcd(k, n) = 1}. In [25], Gauß used primitive roots to discuss the periodicity of 1/p for p a prime not equal to 2 or 5. Both Euler and Jacobi used primitive roots before Gauß. In 1927, Emil Artin (see [43, Introduction] and [31]) reversed the above analysis into the following question: let a be a fixed integer such that a 6= 0, ±1 or a perfect
h square. Let a = b where b ∈ Z is not a perfect power and h ∈ N. Define Na(x) := #{p ≤ x :(Z/pZ)∗ = ha mod pi}. Artin conjectured that the following is true:
Na(x) ∼ Ahπ(x)
1 CHAPTER 1. INTRODUCTION 2
where π(x) = #{p ≤ x : p prime} and Y 1 Y 1 A = 1 − 1 − > 0. (1.1) h q − 1 q(q − 1) q|h q-h q prime q prime The heuristic behind this conjecture is based on the following idea: we have a is primitive root modulo p if and only if for all q prime the following conditions do not occur:
p ≡ 1 (mod q)
p−1 a q ≡ 1 (mod p).
The “probability” the first condition above holds is 1/ϕ(q) = 1/(q − 1). The “prob-
p−1 ability” the second condition above holds is 1/q since a q is a qth root of unity and there are exactly q of those. This conjecture is still unresolved. However, Hooley [31] has provided a condi- tional resolution to the problem:
Theorem 1.1.1 (Hooley). Suppose a ∈ Z such that a 6= 0, ±1 or a perfect square. Suppose further that the generalized Riemann hypothesis holds for Dedekind zeta func-
1/k th tions for the fields Q(a , ζk) with k ∈ N squarefree and where ζk is a k root of unity. Then, x log log x N (x) = A(a)π(x) + O a (log x)2 where the implied constant is dependent on a.
We will define Dedekind zeta functions, the generalized Riemann hypothesis, and O notation in a later section. It should be noted that A(a) in Theorem 1.1.1 is different from Ah in (1.1). It was initially discovered by Lehmer [45] that Artin’s original constant was off by a by small factor. In fact, let h be as above and let CHAPTER 1. INTRODUCTION 3
2 a = a1a2 where a1, a2 ∈ Z and a1 is squarefree. If a1 6≡ 1 (mod 4), then A(a) = Ah, and if a1 ≡ 1 (mod 4), then Y 1 Y 1 A(a) = Ah 1 − µ(|a1|) . q − 2 q2 − q − 1 q| gcd(h,a1) q-h q prime q|a1 q prime The best unconditional results are of the following flavour: One of 2, 3, or 5 is a primitive root modulo p for infinitely many primes p. In fact, we have cx #{p ≤ x : a is a primitive root modulo p} ≥ , (log x)2 where c > 0 is a constant, and a is one of 2, 3, or 5. This result originates in the work of Gupta and Murty [26], and was later refined by Heath-Brown [29].
1.2 Generalizing Artin’s Conjecture
Let a be as before, and let p be a prime such that p - a. Then, define the order of a
modulo p, denoted fa(p), as
k fa(p) := min{k ∈ N : a ≡ 1 mod p} = |ha mod pi|.
Since p - a, fa(p) is well-defined by Fermat’s Little Theorem. Define the index of a
modulo p, denoted ia(p), as
∗ p − 1 ia(p) := [(Z/pZ) : ha mod pi] = . fa(p) We note that Gauß [25, Article 57] defined the index of a modulo p with respect to
a primitive root g of (Z/pZ)∗ as follows: since p is a prime, there exists g ∈ Z such that (Z/pZ)∗ = hg mod pi. Then, the index of a modulo p with respect to g, denoted
p i indg(a), is the number i ∈ {1, 2, . . . , p − 1} such that g ≡ a mod p. It should be CHAPTER 1. INTRODUCTION 4
p noted that the two definitions are related by the fact that ia(p) = gcd(p − 1, indg(a)) for any primitive root modulo p. To see this we note that from group theory we have
p−1 p−1 |ha mod pi| = p . However, by the above definitions, |ha mod pi| = , gcd(p−1,indg(a)) ia(p) and so, the claim holds. Throughout, index will denote the index of a subgroup as
p opposed to indg(a). We can reformulate Artin’s conjecture in the following manner:
Na(x) = #{p ≤ x : fa(p) = p − 1}
= #{p ≤ x : ia(p) = 1} X = χ{1}(ia(p)) p≤x where for S ⊂ N, 0 if n∈ / S χS(n) = . 1 if n ∈ S
We would like to know what would occur if we change χ{1} to a generic function
f : N → C. That is, can we obtain the following relation X f(ia(p)) ∼ Af (a)π(x) p≤x
where Af (a) is a constant dependent on f and a? This question was first studied by Stephens [67], then by Wagstaff [72], Murata [48], Elliott and Murata [15], Pap- palardi [54], Bach, Lukes, Shallit, and Williams [3], and Fomenko [21] among others. Of course, the functions f will have reasonable restrictions so as to not force an im- possibility with the above relation. For example, f(x) = x does not satisfy the above relation. We will see this in Chapter 7. We will mostly be interested in the functions f(n) = log n, ω(n), Ω(n), and P d(n) := d|n 1 where ω(n) and Ω(n) are the number of primes dividing n counting CHAPTER 1. INTRODUCTION 5
without and with multiplicity, respectively. We will also be concerned with χ{w} (with w ∈ N fixed) in a few sections. For f(x) = log x and a = 2, The authors of [3] made the following conjecture and provided computational evidence for its veracity: X log(i2(p)) ∼ c2li(x). p≤x They also mention that their heuristics should follow for a > 2 and give computational evidence for a = 3 and a = 5. Pappalardi [54] has shown the following theorem:
Theorem 1.2.1 (Pappalardi). Let a ∈ Z different from 0 and ±1. We have x X x log log x log(i (p)) log x a log x p≤x where the lower bound is unconditional, and for the upper bound we suppose the generalized Riemann hypothesis holds for Dedekind zeta functions for fields of type
1/k th Q(a , ζk) where ζk is a k root of unity.
For any a an integer not equal to 0, ±1 and a a squarefree integer, we have the following theorem of Fomenko [21]:
Theorem 1.2.2 (Fomenko). Suppose the generalized Riemann hypothesis holds for
1/k th Dedekind zeta functions for fields of type Q(a , ζk) where ζk is a k root of unity. Suppose further Conjecture A of Hooley holds. Then, X x log log x log(i (p)) = c li(x) + O a a (log x)2 p≤x
where ca is an effectively computable constant dependent on a, and Z x 1 li(x) := dt. 2 log t CHAPTER 1. INTRODUCTION 6
We note that this theorem can be extended to a 6= 0 or ±1. We will discuss Conjecture A of Hooley [32] in Chapter 3. In Chapter 7, we will see some of the difficulties in attempting to prove the asymptotic formula for f(x) = log x without the assumption that Conjecture A is true.
1.3 Definitions and Notations
Throughout this document, N, Z, Q, R, R+, and C will denote positive integers, inte- gers, rational numbers, real numbers, positive real numbers, and complex numbers,
respectively. If S is a subset of C, then Q(S) will denote the smallest field containing Q and S, which is inside C. In Chapter 4, we also allow S ⊂ R2. For these S, Q(S) will
denote the smallest field containing Q and real numbers x1, x2 for every (x1, x2) ∈ S, which is again inside C. Also p and q ∈ N will denote prime numbers unless otherwise specified, and a, d, k, `, m, n, and w will denote integers unless otherwise specified. In addition, summations of the form X X X X
p≤x qα≤x a≤x p≤x d|p−a will indicate the variables are over the primes, prime powers, integers, and primes with a specified condition where all symbols match the above conventions. The variables
x, y, and z will denote elements of R+ unless otherwise specified. For s ∈ C, <(s) will denote the real part of the complex number s.
R x −1 As already mentioned π(x) := #{p ≤ x}, li(x) := 2 (log t) dt, and for a, k ∈ Z, we define π(x; k, a) := #{p ≤ x : p ≡ a mod k}.
If g1, g2, . . . , gm ∈ G where G is a group, then hg1, g2, . . . , gmi will denote the
smallest subgroup of G containing g1, g2, . . . , gm. CHAPTER 1. INTRODUCTION 7
If F ⊂ K are fields, then [K : F ] is the dimension of K as an F -vector space. Let f(x) ∈ F [x], where F is a field. Let f(x) factor over the algebraic closure F or the splitting field K of f(x) as c(x − α1)(x − α2) ··· (x − αn) where n = deg(f), c ∈ F , and α1, α2, . . . , αn ∈ K. The discriminant of f(x) is the product
Y 2 (αi − αj) . i As already defined for p - a, we define fa(p) and ia(p) to be the order of a mod- ulo p, and the index of a modulo p, respectively. We also say a is a primitive root modulo p if fa(p) = p − 1, or equivalently, ia(p) = 1. Let a1, a2, . . . , ar ∈ Q. We say that a1, a2, . . . , ar are multiplicatively indepen- dent if and only if the equation n1 n2 nr a1 a2 ··· ar = 1 r has exactly one solution (n1, n2, . . . , nr) ∈ Z that being (0, 0,..., 0). We note that this implies none of a1, a2, . . . , ar can be ±1 or 0. 1.3.1 Arithmetic Functions We define the following arithmetic functions f : N → C: For n ∈ N, we define ϕ(n) := #{a mod n : gcd(a, n) = 1}, X ω(n) := #{p|n} = 1, p|n α X Ω(n) := #{p |n : α ∈ N} = α, pαkn CHAPTER 1. INTRODUCTION 8 where pαkn if and only if pα|n and pα+1 - n, 1 if n = 1 µ(n) := (−1)ω(n) if n is squarefree , 0 otherwise X d(n) := #{d ∈ N : d|n} = 1, d|n k dk(n) = #{(d1, d2, . . . , dk) ∈ N : n = d1d2 ··· dk}, where k ∈ N, and α log p if n = p with α ∈ N Λ(n) := . 0 otherwise We call µ(n) the M¨obiusfunction, ϕ(n) the Euler phi or totient function, d(n) the divisor function, and Λ(n) the von Mangoldt function. Recall the Fundamental Theorem of Arithmetic that every n ∈ N can be expressed α1 α2 αk uniquely (up to ordering of p1, p2, . . . , pk) as p1 p2 ··· pk where p1, p2, . . . , pk ∈ N are distinct prime numbers and α1, α2, . . . , αk ∈ N. Thus, for n ∈ N, we may write Y n = pνp(n) p where νp(n) = αi if p = pi and 0 otherwise. We call νp the p-adic valuation. That α is, νp(n) := max{α ∈ Z : α ≥ 0, p kn}. We can extend this definition to Q in the following way: let a/b ∈ Q such that gcd(a, b) = 1. Then, νp(a/b) = νp(a) − νp(b). CHAPTER 1. INTRODUCTION 9 1.3.2 Asymptotic, Big O, and Little o Notation Let A be an unbounded subset of R. Let f : A → C and g : A → R+ be two functions. We say that f is asymptotic to g if f(x) lim = 1. x→∞ g(x) x∈A In this case, we write f ∼ g. We write f(x) = O(g(x)) or equivalently, f(x) g(x) if there exists a positive constant A such that |f(x)| ≤ Ag(x) for all x ∈ A. In this case, we say that f is big O of g. Also, we write g(x) f(x) to mean f(x) g(x), and we write f(x) g(x) if f(x) g(x) and g(x) f(x). Here we require f : A → R+. We write f(x) = o(g(x)) if f(x) lim = 0. x→∞ g(x) x∈A In this case, we say f(x) is little o of g. If is one of the relations ∼, , , or f(x) = o(g(x)), then the notation f(x)yg(x) implies that the relation is dependent on y. For example, it is well known that d(n) nε for every ε > 0. However, the constant C for which d(n) ≤ Cnε for all ε ε n ∈ N is dependent on ε. So we may also write d(n) ε n or d(n) = Oε(n ). CHAPTER 1. INTRODUCTION 10 1.3.3 Elliptic Curves In Chapter 4, E will denote an elliptic curve defined over Q. Since the characteristic of Q is not 2 or 3, E can be defined as 2 3 E(Q) := {(x, y): x, y ∈ Q, y = x + Ax + B} ∪ {∞} for some A, B ∈ Q and where ∞ is a distinguished point given in projective coordi- nates by [0 : 1 : 0]. Now E(Q) is an Abelian group with ∞ being the identity and addition being defined using lines of intersection and addition. The equation y2 = x3 + Ax + B is called a Weierstraß equation defining E. In fact, for any field K with characteristic not equal to 2 or 3, there is an equation of this form. For characteristics 2 and 3, the Weierstraß equation is defined slightly differently (ee Appendix A of Silverman [63]). The discriminant of E is defined as the discriminant of a minimal Weierstraß equation x3 + Ax + B defining E, which exists by [63, Corollary VIII.8.3] since E is defined over Q. In fact, we have any equation y2 = x3 + Ax + B is an elliptic curve if and only if the discriminant of x3 + Ax + B is non-zero. We say that a prime p is of good reduction if E defined over Q by y2 = x3 + Ax + B is still an elliptic curve when A and B are reduced modulo p. That 2 3 is, when y = x + (A mod p)x + (B mod p) is also an elliptic curve defined over Fp. All other primes are called primes of bad reduction. We note that there are only finitely many primes of bad reduction. These primes correspond to the prime divisors of the discriminant of E (see Silverman [63, Proposition VII.5.1]). CHAPTER 1. INTRODUCTION 11 1.3.4 The GRH and the Dedekind Zeta Function Let us first discuss the Riemann Hypothesis. For <(s) > 1, let −1 X 1 Y 1 ζ(s) := = 1 − . ns ps n≥1 p This function is called the Riemann-zeta function. Riemann [58] was able to show that ζ(s) could be analytically extended to the whole complex plane except for a simple pole at s = 1 with residue 1. Also, he was able to show the following functional equation holds for this analytically continued function: − s s − 1−s 1 − s π 2 Γ ζ(s) = π 2 Γ ζ(1 − s) 2 2 where Γ(z) is the complex analytic continuation of the function f : R → R given by R ∞ x−1 −t f(x) = 0 t e dt, valid for x > 0. 1 From this functional equation, one can easily see that the line <(s) = 2 is signif- icant in the study of ζ(s). At this point, Riemann hypothesized that for all s ∈ C with ζ(s) = 0 and <(s) > 0 we have <(s) = 1/2. This is known as the Riemann hypothesis. If it is true, then √ π(x) = li(x) + O x log x . (1.2) Currently, the best known result of this nature is the following: there exists some constant c > 0 such that 3 !! c(log x) 5 π(x) = li(x) + O x exp − (log log x)1/5 which is clearly weaker than (1.2). This improvement in the classical error term of x exp(−c(log x)1/2) are originally due to N. M. Korobov [38] and I. M. Vinogradov [71], and later, H.-E. Richert [57]. CHAPTER 1. INTRODUCTION 12 Let K be an algebraic number field. That is, Q ⊂ K and [K : Q] < ∞. Define OK := {α ∈ K : f(α) = 0 for some monic f(x) ∈ Z[x]}. It can be shown that OK is a ring. In fact, it can be shown that any ideal of OK can be written as a product of prime ideals in OK uniquely (up to ordering). For example, let p ∈ Z be a prime. Then, pOK is an ideal of OK . Thus, there exist prime ideals p1, p2,..., pg and positive integers e1, e2, . . . , eg such that e1 e2 eg pOK = p1 p2 ··· pg . Also, it can be shown that |OK /a| < ∞ for all non-zero ideals a in OK . In fact, for a prime ideal p with the property that p ∈ p and p ∈ Z is prime, it can be shown that f |OK /p| = p for some f ∈ N (see [37, Chapter 1], [53, Chapter 1], or [18, §5.3]). Let e1 e2 eg pOK = p1 p2 ··· pg . fi and let |OK /pi| = p where fi ∈ N for all i ∈ {1, 2, . . . , g}. Then, we have g X eifi = [K : Q] i=1 (see [18, Exercise 5.3.17]). We say that p splits completely in K if ei = fi = 1 for all i ∈ {1, 2, . . . , g}. We say that p ramifies in K if ei > 1 for some i ∈ {1, 2, . . . , g}. For s ∈ C with <(s) > 1 define: X 1 ζ (s) := K N (a)s a K where the summation is over non-zero ideals a in OK and NK (a) = |OK /a|. We call ζK (s) the Dedekind-zeta function of K. Hecke [30] showed that ζK (s) has an analytic continuation to the whole complex plane except for a simple pole at s = 1. Let n = [K : Q], let σ1, σ2, . . . , σn : K → C be the embeddings of K into C, and let ω1, ω2, . . . , ωn be a Z-basis for OK . Then, the CHAPTER 1. INTRODUCTION 13 discriminant of K, denoted by ∆K , is defined as 2 σ1(ω1) σ1(ω2) . . . σ1(ωn) σ2(ω1) σ2(ω2) . . . σ2(ωn) ∆ := det . K . . . . . . .. . σn(ω1) σn(ω2) . . . σn(ωn) The conjugates of θ ∈ K are σ1(θ), σ2(θ), . . . , σn(θ). Hecke [30] also showed the following functional equation holds: r −r s − [K:Q]s s s 1 r 2 2 π 2 |∆ | 2 Γ Γ(s) 2 ζ (s) K 2 K r1 −r (1−s) − [K:Q](1−s) 1−s 1 − s r = 2 2 π 2 |∆ | 2 Γ Γ(1 − s) 2 ζ (1 − s) K 2 K ∆K is the discriminant of K over Q and when we write K = Q(θ), then r1 is the number of real conjugates of θ over Q and 2r2 is the number of complex conjugates of θ over Q. 1 Again, one can see that the line <(s) = 2 is important to the study of ζK (s). 1 Suppose for all s ∈ C with <(s) > 0 and ζK (s) = 0 we have <(s) = 2 . This is known as the generalized Riemann hypothesis for the Dedekind zeta function of K which we will denote by GRH. Using the fact that primes in Z = OQ correspond to prime ideals in Z we define p to be prime in OK if p is a prime ideal of OK . Let K be an algebraic number field, L be a normal extension of K with Galois group G. Let C be a conjugacy class of G, and define πC (x; L/K) = #{p : p is unramified in L, [(L/K)/p] = C,NK/Q(p) ≤ x} where the [(L/K)/p] denotes the conjugacy class of Frobenius automorphisms cor- responding to prime ideals P ⊃ p. Finally, dL will denote the absolute value of the discriminant of L, and nL = [L : Q]. CHAPTER 1. INTRODUCTION 14 The following theorem of Lagarias and Odlyzko [39] will be used throughout: Theorem 1.3.1 (Effective Chebotarev Density Theorem). We have #C #C − 1 1 β0 2 πC (x; L/K) − li(x) ≤ li(x ) + O x exp −c1n (log x) 2 #G #G L where β0 is the exceptional real zero of ζL(s), assuming it exists. If it does not exist this term is omitted from the above. Also, β0 can be effectively computed in terms of dL and nL. Assuming the GRH for the Dedekind-zeta function of L, we have #C #C √ π (x; L/K) = li(x) + O x (log(d xnL ) + log d ) . C #G #G L L All constants are effectively computable. The implied constants are absolute. A non-effective version of this theorem was first proved by Chebotarev [8, 9]. Improvements in the error term have been made by Serre [62], Murty, Murty and Saradha, [52], and Murty and Murty [51]. 1.4 Organization of the Thesis In Chapter 2, we introduce a new technique that allows to solve problems related to the summation X f(ia(p)). p≤x In particular, we look at the functions f(n) = d(n) and f(n) = (log n)α where 0 < α < 1 is fixed. In Chapter 3, we generalize the problem of determining X f(ia(p)) ∼ Af (a)li(x) p≤x CHAPTER 1. INTRODUCTION 15 ∗ as x → ∞ to the r-rank case. That is, let Γ = ha1, a2, . . . , ari ⊂ Q where a1, a2, . . . , ar is a set of multiplicatively independent integers. Then, we can reduce Γ modulo p and consider it as a subgroup of (Z/pZ)∗. Thus, we can consider the above problem. We obtain asymptotic formulæ for f(x) = log(x) and for f(x) = w where w ∈ N is fixed assuming the GRH. In Chapter 4, we generalize the above problem to elliptic curves. That is, let E be an elliptic curve defined over Q. Then, E(Q) is a finitely generated Abelian group by the Mordell-Weil theorem. As above, we can let Γ = hP1,P2,...,Pri be a subgroup of E(Q). We can reduce both E(Q) and Γ modulo p and this relation will still hold. Thus, we can consider this problem in the elliptic curve setting. We obtain an asymptotic formula for f(n) = ω(n) and upper and lower bounds for f(n) = Ω(n) and log n assuming the GRH. In Chapter 5, we consider the above problem in classical case and prove asymptotic formulæ for f(n) = ω(n) and Ω(n). We also prove that for f(x) = log x, the average summation, and its second moment are as the conjecture predicts. We also revisit a result from the author’s Master’s project [19] and obtain a result related to the index and a second conjecture mentioned in Fomenko [21]. In Chapter 6, we investigate the above problems in relation to a conjecture of Erd˝os.In particular, we show that Conjecture A of Hooley disappears if we assume this conjecture and a quasi-GRH. In Chapter 7, a related conjecture is also heuristically analysed. Also, we discuss the difficulties of obtaining an asymptotic formulæ for f(x) = log x in the classical setting, and f(n) = Ω(n) and f(x) = log x in the elliptic curve setting. Finally, we discuss future research. Chapter 2 A New Method 2.1 Introduction As we saw in §1.2, we can ask when does the following relation hold for f : N → C: X f(ia(p)) ∼ caπ(x), p≤x where ca is a constant dependent on at most a and f? We note that if we write X f(n) = g(d), d|n where g : N → C, then X X X X X f(ia(p)) = g(d) = g(d) 1 p≤x p≤x d|ia(p) d≤x p≤x d|ia(p) X = g(d)πd(x), d≤x 16 CHAPTER 2. A NEW METHOD 17 where πd(x) := #{p ≤ x : d|ia(p)}. We note that it is always possible to write X f(n) = g(d) d|n by the M¨obiusInversion Formula (see [12, §1.2]). Assuming g : N → C is well-behaved, we have no difficulty applying standard techniques to obtain results of this nature. For example, f(n) = ω(n) and f(n) = Ω(n) fall into this category (see §5.2). However, more complicated functions cause difficulties if we try to use the standard techniques. For example, f(n) = log n or f(n) = d(n) cause difficulties when considering intermediate primes and large divisors, respectively. To see this, let us consider the following summations X X X log(ia(p)) = Λ(d)πd(x) + Λ(d)πd(x) √ √ √ p≤x d≤ x x where πd(x) := #{p ≤ x : d|ia(p)} and A, B > 0 are arbitrarily fixed. (In general, when we write a generic sum over primes as X X X X = + + p≤x p≤y y summation adequately without assuming something beyond the reach of GRH and the Chebotarev density theorem. This is why Fomenko [21] assumed Conjecture A of Hooley. Similar difficulties exist for f(n) = d(n), but they are more difficult because in the last summation Hooley’s argument [31, Equation (3)] no longer applies. In this chapter, we will discuss a new technique that eliminates some of these difficulties if we only assume GRH. For example, we will prove the following theorem in §2.3: Theorem 2.1.1. Suppose a ∈ Z is not 0 or ±1. Let α ∈ (0, 1) be fixed. Suppose fur- 1/n ther that the GRH holds for the Dedekind zeta functions of the fields Kn = Q(ζn, a ) th where ζn is a primitive n root of unity. Then, for any ε > 0, we have X x (log i (p))α = c li(x) + O , a a (log x)2−ε−α p≤x where ca is a constant. In §2.4, we will prove the following theorem: Theorem 2.1.2. Suppose a ∈ Z is not 0 or ±1. Suppose further that the GRH holds 1/n for the Dedekind zeta functions of the fields Kn = Q(ζn, a ) where ζn is a primitive nth root of unity. Then, for any ε > 0, we have X x d(i (p)) = c0 li(x) + O , a a (log x)2−ε p≤x where 0 X 1 ca = , [Kd : ] d≥1 Q a constant. Before we can prove these theorems, we need some preliminary lemmata. CHAPTER 2. A NEW METHOD 19 2.2 Lemmata Recall the following theorem of Pappalardi [54]: Theorem 2.2.1 (Pappalardi). Suppose a ∈ Z is not 0 or ±1. Suppose further that 1/n the GRH holds for the Dedekind zeta functions of the fields Kn = Q(ζn, a ) where th ζn is a primitive n root of unity. Then, we have X x log log x log(i (p)) . a log x p≤x From this, we immediately have Corollary 2.2.1. Suppose a ∈ Z is not 0 or ±1. Suppose further that the GRH holds 1/n for the Dedekind zeta functions of the fields Kn = Q(ζn, a ) where ζn is a primitive nth root of unity. Then, we have x log log x #{p ≤ x : i (p) > y} . a (log x)(log y) To see this note that X x log log x log y#{p ≤ x : i (p) > y} ≤ log i (p) . a a log x p≤x Dividing both sides by log y gives us the corollary. We call the following statement the quasi-Riemann hypothesis: For K an algebraic number field with Dedekind-zeta function ζK (s), there exists ε ∈ (0, 1/2) such that if <(s) > 1 − ε, then ζK (s) 6= 0. We note that in this case, the effective Chebotarev Density Theorem (mimic [39, Theorem 1.1] and [62, Theorem 4]) becomes #C 1−ε Y πC (x; L/K) = li(x) + O #Cx #G p x , #G p ramifies in K/Q where all of the above symbols are as in §1.3 and the O-constants depend only on ε. CHAPTER 2. A NEW METHOD 20 Let πd(x) := #{p ≤ x : d|ia(p)}. We claim that with the quasi-Riemann hypoth- esis we have li(x) 1−ε πd(x) = 1/d + O x log(adx) [Q(ζd, a ): Q] th where ζd is a primitive d -root of unity. To see this we note that d|ia(p) if and p−1 only if p ≡ 1 mod d and a d ≡ 1 (mod p). The second condition is equivalent to xd ≡ a mod p has a solution modulo p. This along with the first condition says d the polynomial x − a splits into d distinct linear factors over Fp[x]. Thus, p splits 1/d completely in Q(ζd, a ) (see [18, Theorem 5.5.1]). Also, if d - ia(p), then p 6≡ 1 mod d p−1 or p ≡ 1 (mod d) and a d 6≡ 1 (mod p). If p 6≡ 1 mod d, then p does not split completely in Q(ζd) (see [53, Corollary 10.4 of Chapter 1]) and hence, does not split 1/d completely in Q(ζd, a ) (see [37, Property 2.3 of Chapter 3]). If p ≡ 1 mod d and p−1 d d a d 6≡ 1 (mod p), then x 6≡ a mod p. Thus, x − a does not split into d distinct 1/d linear factors in Fp[x]. Hence, p does not split completely in Q(ζd, a ). Thus, d|ia(p) 1/d if and only if p splits completely in Q(ζd, a ). However, this is true if and only if the 1/d 1/d [(Q(ζd, a )/Q), p], the conjugacy class of the Frobenius automorphism of Q(ζd, a ) over Q, must equal 1. Hence, the claim holds upon noting that the discriminant of 1/d d−1 d ϕ(d) Q(a ) is ±a d (see [37, Page 40]), the discriminant of Q(ζd) divides d . Thus, by [37, Theorem 7.3 of Chapter 1], we have Y Y p = p ≤ ad p ramifies p|ad in K/Q 1/d 1/d 2 Also, #G = [Q(ζd, a ): Q] ≤ [Q(ζd): Q][Q(a ): Q] ≤ dϕ(d) ≤ d . Thus, li(x) 1−ε πd(x) = 1/d + O x log(adx) [Q(ζd, a ): Q] We will ignore the a in the error term as it is a fixed throughout this chapter. That CHAPTER 2. A NEW METHOD 21 is, li(x) 1−ε πd(x) = 1/d + O x log(dx) [Q(ζd, a ): Q] We also have the following lemma: Lemma 2.2.1. Let r and k be fixed positive integers. Suppose the quasi-Riemann 1/n th hypothesis for the fields Kn = Q(a , ζn) where ζn is a primitive n root of unity. Then, X r dk(ia(p)) π(x), p≤x where the implied constant is dependent on at most r and k. Proof. [24, Corollary 22.11] states that for r and k positive integers, we have r(k−1)t log t r X dk(n) ≤ (2d(m)) log 2 . m|n m≤n1/t The general idea of truncating the range of interest in the above summation originates with work van der Corput [70]. This above development was first initiated by Lan- dreau [40] and continued by Iwaniec and Munshi [35], and Friedlander and Iwaniec [23]. r(k−1)t log t Let A = log 2 . We will see that our choice of t is bounded and thus A is bounded. This will be important in the sequel. Let us recall some facts about d(m): A X d(m) [A]+1 (log x)2 m n≤x and d(m) mδ for any δ > 0 (see [12, Exercises 10.5.3 and 1.5.3, respectively]). CHAPTER 2. A NEW METHOD 22 Thus, X r X X A X A dk(ia(p)) ≤ (2d(m)) d(m) πm(x) p≤x p≤x m|ia(p) m≤x1/t m≤x1/t X d(m)A X li(x) + O x1−ε log x d(m)A mϕ(m) m≤x1/t m Note that the above results give the following lemma: Lemma 2.2.2. Suppose a ∈ Z is an integer different from 0 or ±1. Suppose further 1/n that the GRH holds for the Dedekind zeta functions of the fields Kn = Q(ζn, a ) th where ζn is a primitive n root of unity. Then, we have X x d (i (p))r k a (log x)β p≤x√ x ia(p)> (log x)B for any β < 2 and B ∈ R fixed. 1 1 Proof. To see this, let p, q > 1 be real numbers such that p + q = 1. Then, by H¨older’s CHAPTER 2. A NEW METHOD 23 inequality and Corollary 2.2.1, we have 1 1 p q X r X X rq dk(ia(p)) ≤ 1 dk(ia(p)) p≤x p≤x p≤x √ √ √ x x x ia(p)> ia(p)> ia(p)> (log x)B (log x)B (log x)B 1 1 π(x) log log x p 1 π(x)(log log x) p q (π(x)) 1 log x (log x) p 1 x(log log x) p . 1+ 1 (log x) p The result now follows upon noting that we may choose p > 1 arbitrarily close to 1 and that log log x (log x)δ for every δ > 0. 2.3 The Function (log n)α In this section, we are going to prove Theorem 2.1.1. Let α < 1 be a fixed real number. Write X (log n)α = g(d). d|n Then, by the M¨obiusInversion Formula, we have X nα g(n) = µ(d) log , d d|n and so, X |g(n)| ≤ (log n)α 1 = (log n)αd(n). d|n CHAPTER 2. A NEW METHOD 24 We have X α X (log(ia(p))) = g(m)πm(x) p≤x m≤x X X = g(m)πm(x) + g(m)πm(x), m≤y y