Prime Number Races Youness Lamzouri University of Illinois at Urbana-Champaign 2011 CMS Winter Meeting, Toronto December 11, 2011 Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 1 / 29 How many prime numbers are there up to x? π(x) is the number of primes up to x. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 2 / 29 How many prime numbers are there up to x? π(x) is the number of primes up to x. Gauss conjectured in 1792 or 1793 (he was then 15 or 16), that the primes occur with density 1/ log x at around x. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 2 / 29 How many prime numbers are there up to x? π(x) is the number of primes up to x. Gauss conjectured in 1792 or 1793 (he was then 15 or 16), that the primes occur with density 1/ log x at around x. In other words Z x dt . π(x) ≈ Li(x) = 2 log t Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 2 / 29 How many prime numbers are there up to x? π(x) is the number of primes up to x. Gauss conjectured in 1792 or 1793 (he was then 15 or 16), that the primes occur with density 1/ log x at around x. In other words Z x dt . π(x) ≈ Li(x) = 2 log t Table: The number of primes up to x x 108 1010 1012 1014 1016 1018 Youness Lamzouri (Illinois) π(x) 5761455 455052511 37607912018 3204941750802 279238341033925 24739954287740860 Prime Number Races Li(x) − π(x) 753 3103 38262 314889 3214631 21949554 December 11, 2011 2 / 29 Riemann’s Memoir (1859): π(x) is connected with the zeros of the “Riemann Zeta function” ζ(s), a function of a complex variable defined by ∞ X Y 1 1 −1 for Re(s) > 1. ζ(s) = = 1− s ns p n=1 Youness Lamzouri (Illinois) p prime Prime Number Races December 11, 2011 3 / 29 Riemann’s Memoir (1859): π(x) is connected with the zeros of the “Riemann Zeta function” ζ(s), a function of a complex variable defined by ∞ X Y 1 1 −1 for Re(s) > 1. ζ(s) = = 1− s ns p n=1 p prime ζ(s) has an analytic continuation to the complex plan except at s = 1 where it has a simple pole. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 3 / 29 Riemann’s Memoir (1859): π(x) is connected with the zeros of the “Riemann Zeta function” ζ(s), a function of a complex variable defined by ∞ X Y 1 1 −1 for Re(s) > 1. ζ(s) = = 1− s ns p n=1 p prime ζ(s) has an analytic continuation to the complex plan except at s = 1 where it has a simple pole. Notation: f (x) ∼ g (x) if limx→∞ f (x)/g (x) = 1. Prime Number Theorem (Hadamard and de la Vallée Poussin, 1898) π(x) ∼ Li(x) Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 3 / 29 Riemann’s Memoir (1859): π(x) is connected with the zeros of the “Riemann Zeta function” ζ(s), a function of a complex variable defined by ∞ X Y 1 1 −1 for Re(s) > 1. ζ(s) = = 1− s ns p n=1 p prime ζ(s) has an analytic continuation to the complex plan except at s = 1 where it has a simple pole. Notation: f (x) ∼ g (x) if limx→∞ f (x)/g (x) = 1. Prime Number Theorem (Hadamard and de la Vallée Poussin, 1898) π(x) ∼ Li(x)∼ Youness Lamzouri (Illinois) x . log x Prime Number Races December 11, 2011 3 / 29 Riemann’s Memoir (1859): π(x) is connected with the zeros of the “Riemann Zeta function” ζ(s), a function of a complex variable defined by ∞ X Y 1 1 −1 for Re(s) > 1. ζ(s) = = 1− s ns p n=1 p prime ζ(s) has an analytic continuation to the complex plan except at s = 1 where it has a simple pole. Notation: f (x) ∼ g (x) if limx→∞ f (x)/g (x) = 1. Prime Number Theorem (Hadamard and de la Vallée Poussin, 1898) π(x) ∼ Li(x)∼ x . log x The Riemann Hypothesis (1,000,000$) 1. If σ + it is a complex number with 0 ≤ σ ≤ 1 and ζ(σ + it) = 0, then σ = 1/2. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 3 / 29 Riemann’s Memoir (1859): π(x) is connected with the zeros of the “Riemann Zeta function” ζ(s), a function of a complex variable defined by ∞ X Y 1 1 −1 for Re(s) > 1. ζ(s) = = 1− s ns p n=1 p prime ζ(s) has an analytic continuation to the complex plan except at s = 1 where it has a simple pole. Notation: f (x) ∼ g (x) if limx→∞ f (x)/g (x) = 1. Prime Number Theorem (Hadamard and de la Vallée Poussin, 1898) π(x) ∼ Li(x)∼ x . log x The Riemann Hypothesis (1,000,000$) 1. If σ + it is a complex number with 0 ≤ σ ≤ 1 and ζ(σ + it) = 0, then σ = 1/2. 2. π(x) = Li(x) + O x 1/2 log x . Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 3 / 29 Table: The number of primes up to x x 108 1010 1012 1014 1016 1018 Youness Lamzouri (Illinois) π(x) 5761455 455052511 37607912018 3204941750802 279238341033925 24739954287740860 Prime Number Races Li(x) − π(x) 753 3103 38262 314889 3214631 21949554 December 11, 2011 4 / 29 Table: The number of primes up to x x 108 1010 1012 1014 1016 1018 π(x) 5761455 455052511 37607912018 3204941750802 279238341033925 24739954287740860 Li(x) − π(x) 753 3103 38262 314889 3214631 21949554 For all x < 1023 we have π(x) < Li(x). Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 4 / 29 Table: The number of primes up to x x 108 1010 1012 1014 1016 1018 π(x) 5761455 455052511 37607912018 3204941750802 279238341033925 24739954287740860 Li(x) − π(x) 753 3103 38262 314889 3214631 21949554 For all x < 1023 we have π(x) < Li(x). Theorem (Littlewood 1914) The difference π(x) − Li(x) changes sign for infinitely many integers x. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 4 / 29 Table: The number of primes up to x x 108 1010 1012 1014 1016 1018 π(x) 5761455 455052511 37607912018 3204941750802 279238341033925 24739954287740860 Li(x) − π(x) 753 3103 38262 314889 3214631 21949554 For all x < 1023 we have π(x) < Li(x). Theorem (Littlewood 1914) The difference π(x) − Li(x) changes sign for infinitely many integers x. Bays and Hudson (1999): The first x for which π(x) > Li(x) is expected to be around 1.3982 × 10316 . Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 4 / 29 Primes in arithmetic progressions π(x; q, a) is the number of primes p ≤ x such that p ≡ a mod q (primes of the form qn + a). φ(q) is the number of integers 1 ≤ a ≤ q such that (a, q) = 1. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 5 / 29 Primes in arithmetic progressions π(x; q, a) is the number of primes p ≤ x such that p ≡ a mod q (primes of the form qn + a). φ(q) is the number of integers 1 ≤ a ≤ q such that (a, q) = 1. Theorem (Dirichlet 1837) If (a, q) = 1, then there are infinitely many primes p ≡ a mod q. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 5 / 29 Primes in arithmetic progressions π(x; q, a) is the number of primes p ≤ x such that p ≡ a mod q (primes of the form qn + a). φ(q) is the number of integers 1 ≤ a ≤ q such that (a, q) = 1. Theorem (Dirichlet 1837) If (a, q) = 1, then there are infinitely many primes p ≡ a mod q. How many primes congruent to a mod q are there up to x ? Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 5 / 29 Primes in arithmetic progressions π(x; q, a) is the number of primes p ≤ x such that p ≡ a mod q (primes of the form qn + a). φ(q) is the number of integers 1 ≤ a ≤ q such that (a, q) = 1. Theorem (Dirichlet 1837) If (a, q) = 1, then there are infinitely many primes p ≡ a mod q. How many primes congruent to a mod q are there up to x ? Prime Number Theorem for Arithmetic Progressions Let (a, q) = 1. Then, as x → ∞, we have π(x; q, a) ∼ Youness Lamzouri (Illinois) Li(x) π(x) ∼ . φ(q) φ(q) Prime Number Races December 11, 2011 5 / 29 Chebyshev’s Bias In 1853, Chebyshev wrote a letter to M. Fuss with the following statement There is a notable difference in the splitting of the primes between the two forms 4n + 3, 4n + 1: the first form contains a lot more than the second. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 6 / 29 Chebyshev’s Bias In 1853, Chebyshev wrote a letter to M. Fuss with the following statement There is a notable difference in the splitting of the primes between the two forms 4n + 3, 4n + 1: the first form contains a lot more than the second. Recall that π(x; 4, 1) ∼ π(x)/2 ∼ π(x; 4, 3) as x → ∞. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 6 / 29 Chebyshev’s Bias In 1853, Chebyshev wrote a letter to M. Fuss with the following statement There is a notable difference in the splitting of the primes between the two forms 4n + 3, 4n + 1: the first form contains a lot more than the second. Recall that π(x; 4, 1) ∼ π(x)/2 ∼ π(x; 4, 3) as x → ∞. Table: The number of primes of the form 4n + 3 and 4n + 1 up to x x 100 500 1000 5000 10,000 50,000 100,000 Youness Lamzouri (Illinois) π(x; 4, 3) 13 50 87 339 619 2583 4808 Prime Number Races π(x; 4, 1) 11 44 80 329 609 2549 4783 December 11, 2011 6 / 29 The race modulo 4 π(x; 4, 1) > π(x; 4, 3) for the first time when x = 26,861. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 7 / 29 The race modulo 4 π(x; 4, 1) > π(x; 4, 3) for the first time when x = 26,861. For x ≥ 26,863, π(x; 4, 1) > π(x; 4, 3) occurs for the first time when x = 616,841, and also at various numbers until 633,798. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 7 / 29 The race modulo 4 π(x; 4, 1) > π(x; 4, 3) for the first time when x = 26,861. For x ≥ 26,863, π(x; 4, 1) > π(x; 4, 3) occurs for the first time when x = 616,841, and also at various numbers until 633,798. The first x > 633,798 for which π(x; 4, 1) > π(x; 4, 3) is x = 12,306,137. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 7 / 29 The race modulo 4 π(x; 4, 1) > π(x; 4, 3) for the first time when x = 26,861. For x ≥ 26,863, π(x; 4, 1) > π(x; 4, 3) occurs for the first time when x = 616,841, and also at various numbers until 633,798. The first x > 633,798 for which π(x; 4, 1) > π(x; 4, 3) is x = 12,306,137. Theorem (Littlewood 1914) The difference π(x; 4, 1) − π(x; 4, 3) changes sign for infinitely many integers x. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 7 / 29 The race modulo 3 Table: The number of primes of the form 3n + 2 and 3n + 1 up to x x 100 1000 10,000 100,000 1,000,000 Youness Lamzouri (Illinois) π(x; 3, 2) 13 87 617 4807 39,266 Prime Number Races π(x; 3, 1) 11 80 611 4784 39,231 December 11, 2011 8 / 29 The race modulo 3 Table: The number of primes of the form 3n + 2 and 3n + 1 up to x x 100 1000 10,000 100,000 1,000,000 π(x; 3, 2) 13 87 617 4807 39,266 π(x; 3, 1) 11 80 611 4784 39,231 Bays and Hudson (Christmas Day of 1976): π(x; 3, 1) > π(x; 3, 2) for the first time when x = 608,981,813,029 ≈ 6 × 1011 . Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 8 / 29 The race modulo 3 Table: The number of primes of the form 3n + 2 and 3n + 1 up to x x 100 1000 10,000 100,000 1,000,000 π(x; 3, 2) 13 87 617 4807 39,266 π(x; 3, 1) 11 80 611 4784 39,231 Bays and Hudson (Christmas Day of 1976): π(x; 3, 1) > π(x; 3, 2) for the first time when x = 608,981,813,029 ≈ 6 × 1011 . Theorem (Littlewood 1914) π(x; 3, 1) − π(x; 3, 2) changes sign for infinitely many integers x. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 8 / 29 Two more races, modulo 10 and modulo 8 Table: The number of primes of the form 10n + j up to x x 1000 10,000 100,000 1,000,000 Last Digit 1 40 306 2387 19,617 Youness Lamzouri (Illinois) Last Digit 3 42 310 2402 19,665 Prime Number Races Last Digit 7 46 308 2411 19,621 Last Digit 9 38 303 2390 19,593 December 11, 2011 9 / 29 Two more races, modulo 10 and modulo 8 Table: The number of primes of the form 10n + j up to x x 1000 10,000 100,000 1,000,000 Last Digit 1 40 306 2387 19,617 Last Digit 3 42 310 2402 19,665 Last Digit 7 46 308 2411 19,621 Last Digit 9 38 303 2390 19,593 Table: The number of primes of the form 8n + j up to x x 1000 10,000 100,000 1,000,000 Youness Lamzouri (Illinois) π(x; 8, 1) 37 295 2384 19,552 π(x; 8, 3) 44 311 2409 19,653 Prime Number Races π(x; 8, 5) 43 314 2399 19,623 π(x; 8, 7) 43 308 2399 19,669 December 11, 2011 9 / 29 Dirichlet characters and Dirichlet L-functions A Dirichlet character modulo q is a function χ : Z → C such that 1. χ is periodic with period q. 2. χ(n) = 0 if (n, q) > 1. 3. χ is completely multiplicative: χ(mn) = χ(m)χ(n). Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 10 / 29 Dirichlet characters and Dirichlet L-functions A Dirichlet character modulo q is a function χ : Z → C such that 1. χ is periodic with period q. 2. χ(n) = 0 if (n, q) > 1. 3. χ is completely multiplicative: χ(mn) = χ(m)χ(n). There are φ(q) Dirichlet characters modulo q. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 10 / 29 Dirichlet characters and Dirichlet L-functions A Dirichlet character modulo q is a function χ : Z → C such that 1. χ is periodic with period q. 2. χ(n) = 0 if (n, q) > 1. 3. χ is completely multiplicative: χ(mn) = χ(m)χ(n). There are φ(q) Dirichlet characters modulo q. Orthogonality relation: If (a, q) = 1 then ( X 1 if n ≡ a mod q, 1 χ(a)χ(n) = φ(q) 0 otherwise. χ mod q Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 10 / 29 Dirichlet characters and Dirichlet L-functions A Dirichlet character modulo q is a function χ : Z → C such that 1. χ is periodic with period q. 2. χ(n) = 0 if (n, q) > 1. 3. χ is completely multiplicative: χ(mn) = χ(m)χ(n). There are φ(q) Dirichlet characters modulo q. Orthogonality relation: If (a, q) = 1 then ( X 1 if n ≡ a mod q, 1 χ(a)χ(n) = φ(q) 0 otherwise. χ mod q Examples of Dirichlet characters The principal character modulo q: χ0 (n) = 1 if (n, q) = 1. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 10 / 29 Dirichlet characters and Dirichlet L-functions A Dirichlet character modulo q is a function χ : Z → C such that 1. χ is periodic with period q. 2. χ(n) = 0 if (n, q) > 1. 3. χ is completely multiplicative: χ(mn) = χ(m)χ(n). There are φ(q) Dirichlet characters modulo q. Orthogonality relation: If (a, q) = 1 then ( X 1 if n ≡ a mod q, 1 χ(a)χ(n) = φ(q) 0 otherwise. χ mod q Examples of Dirichlet characters The principal character modulo q: χ0 (n) = 1 if (n, q) = 1. The non-principal character modulo 4: χ1 (n) = 1 if n ≡ 1 mod 4 and χ1 (n) = −1 if n ≡ 3 mod 4. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 10 / 29 Let χ be a Dirichlet character modulo q. The Dirichlet L-function L(s, χ) is defined by ∞ X Y χ(n) χ(p) −1 for Re(s) > 1. L(s, χ) = = 1− s ns p n=1 Youness Lamzouri (Illinois) p prime Prime Number Races December 11, 2011 11 / 29 Let χ be a Dirichlet character modulo q. The Dirichlet L-function L(s, χ) is defined by ∞ X Y χ(n) χ(p) −1 for Re(s) > 1. L(s, χ) = = 1− s ns p n=1 p prime L(s, χ) can be analytically continued to an entire function over the complex plan. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 11 / 29 Let χ be a Dirichlet character modulo q. The Dirichlet L-function L(s, χ) is defined by ∞ X Y χ(n) χ(p) −1 for Re(s) > 1. L(s, χ) = = 1− s ns p n=1 p prime L(s, χ) can be analytically continued to an entire function over the complex plan. Dirichlet (1837): L(1, χ) 6= 0 for all χ 6= χ0 mod q. This implies that there are infinitely many primes p ≡ a mod q if (a, q) = 1. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 11 / 29 Let χ be a Dirichlet character modulo q. The Dirichlet L-function L(s, χ) is defined by ∞ X Y χ(n) χ(p) −1 for Re(s) > 1. L(s, χ) = = 1− s ns p n=1 p prime L(s, χ) can be analytically continued to an entire function over the complex plan. Dirichlet (1837): L(1, χ) 6= 0 for all χ 6= χ0 mod q. This implies that there are infinitely many primes p ≡ a mod q if (a, q) = 1. If χ1 is the non-principal Dirichlet character modulo 4. Then L(s, χ1 ) = Youness Lamzouri (Illinois) 1 1 1 1 1 1 − s + s − s + s − s + · · · for Re(s) > 1. 1s 3 5 7 9 11 Prime Number Races December 11, 2011 11 / 29 Let χ be a Dirichlet character modulo q. The Dirichlet L-function L(s, χ) is defined by ∞ X Y χ(n) χ(p) −1 for Re(s) > 1. L(s, χ) = = 1− s ns p n=1 p prime L(s, χ) can be analytically continued to an entire function over the complex plan. Dirichlet (1837): L(1, χ) 6= 0 for all χ 6= χ0 mod q. This implies that there are infinitely many primes p ≡ a mod q if (a, q) = 1. If χ1 is the non-principal Dirichlet character modulo 4. Then L(s, χ1 ) = 1 1 1 1 1 1 − s + s − s + s − s + · · · for Re(s) > 1. 1s 3 5 7 9 11 The Generalized Riemann Hypothesis GRH Let χ be a Dirichlet character modulo q. If σ + it is a complex number with 0 ≤ σ ≤ 1 and L(σ + it, χ) = 0, then σ = 21 . Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 11 / 29 The origin of Chebyshev’s Bias More primes of the form 4n + 3 than of the form 4n + 1. More primes of the forms 10n + 3 and 10n + 7, than of the forms 10n + 9 and 10n + 1. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 12 / 29 The origin of Chebyshev’s Bias More primes of the form 4n + 3 than of the form 4n + 1. More primes of the forms 10n + 3 and 10n + 7, than of the forms 10n + 9 and 10n + 1. Chebyshev’s Bias: More primes of the form qn + a than of the form qn + b if a is non-square and b is a square modulo q. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 12 / 29 The origin of Chebyshev’s Bias More primes of the form 4n + 3 than of the form 4n + 1. More primes of the forms 10n + 3 and 10n + 7, than of the forms 10n + 9 and 10n + 1. Chebyshev’s Bias: More primes of the form qn + a than of the form qn + b if a is non-square and b is a square modulo q. Explicit formula (goes back to Riemann, 1859) Sum over primes in arithmetic progressions ⇐⇒ Sum over zeros of Dirichlet L-functions L(s, χ) Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 12 / 29 The origin of Chebyshev’s Bias More primes of the form 4n + 3 than of the form 4n + 1. More primes of the forms 10n + 3 and 10n + 7, than of the forms 10n + 9 and 10n + 1. Chebyshev’s Bias: More primes of the form qn + a than of the form qn + b if a is non-square and b is a square modulo q. Explicit formula (goes back to Riemann, 1859) Sum over primes in arithmetic progressions ⇐⇒ Sum over zeros of Dirichlet L-functions L(s, χ) Our sum over primes in arithmetic progressions is X X √ 1 1 = π(x; q, a) + |{p ≤ x : p 2 ≡ a mod q}| + error. k 2 k k≥1 p ≤x p k ≡a mod q Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 12 / 29 The origin of Chebyshev’s Bias More primes of the form 4n + 3 than of the form 4n + 1. More primes of the forms 10n + 3 and 10n + 7, than of the forms 10n + 9 and 10n + 1. Chebyshev’s Bias: More primes of the form qn + a than of the form qn + b if a is non-square and b is a square modulo q. Explicit formula (goes back to Riemann, 1859) Sum over primes in arithmetic progressions ⇐⇒ Sum over zeros of Dirichlet L-functions L(s, χ) Our sum over primes in arithmetic progressions is X X √ 1 1 = π(x; q, a) + |{p ≤ x : p 2 ≡ a mod q}| + error. k 2 k k≥1 p ≤x p k ≡a mod q The second term is the source of Chebychev’s Bias. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 12 / 29 Can we measure the bias? Conjecture (Knapowski-Turàn 1962) As X → ∞, the percentage of integers x ≤ X for which π(x; 4, 1) > π(x; 4, 3) goes to to 0%. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 13 / 29 Can we measure the bias? Conjecture (Knapowski-Turàn 1962) As X → ∞, the percentage of integers x ≤ X for which π(x; 4, 1) > π(x; 4, 3) goes to to 0%. The following table gives, for X in various ranges, the maximum percentage of values of x ≤ X for which π(x; 4, 1) > π(x; 4, 3): Table: For X in the range 0 − 107 107 − 108 107 − 108 108 − 109 109 − 1010 1010 − 1011 Youness Lamzouri (Illinois) Maximum percentage of such x ≤ X 2.6% 0.6% 0.6% 0.1% 1.6% 2.8% Prime Number Races December 11, 2011 13 / 29 We need assumptions to quantify these biases ! Theorem (Kaczorowski, Rubinstein-Sarnak 1994) If the Generalized Riemann Hypothesis GRH is true, then the Knapowski-Turàn Conjecture is false. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 14 / 29 We need assumptions to quantify these biases ! Theorem (Kaczorowski, Rubinstein-Sarnak 1994) If the Generalized Riemann Hypothesis GRH is true, then the Knapowski-Turàn Conjecture is false. Theorem (Kaczorowski 1994) Assume GRH. For any (a, q) = 1 such that a 6≡ 1 mod q, the set of positive integers x for which π(x; q, 1) > π(x; q, a) has a positive lower density. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 14 / 29 We need assumptions to quantify these biases ! Theorem (Kaczorowski, Rubinstein-Sarnak 1994) If the Generalized Riemann Hypothesis GRH is true, then the Knapowski-Turàn Conjecture is false. Theorem (Kaczorowski 1994) Assume GRH. For any (a, q) = 1 such that a 6≡ 1 mod q, the set of positive integers x for which π(x; q, 1) > π(x; q, a) has a positive lower density. Theorem 1 (Ford, Konyagin and L. 2011+) Let (a, q) = 1 be such that a2 6≡ 1 mod q. Then we can construct a set S = {σ + it : 0 ≤ σ ≤ 1 and σ 6= 1/2} such that if a certain L(s, χ) has zeros in S, then the percentage of integers x ≤ X for which π(x; q, 1) > π(x; q, a) goes to 0%. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 14 / 29 Measuring the bias: the work of Rubinstein-Sarnak Theorem (Kaczorowski 1994) Assume the Generalized Riemann Hypothesis. The quantity 1 |{x ≤ X : π(x; 4, 3) > π(x; 4, 1)} X does not tend to any limit as X → ∞. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 15 / 29 Measuring the bias: the work of Rubinstein-Sarnak Theorem (Kaczorowski 1994) Assume the Generalized Riemann Hypothesis. The quantity 1 |{x ≤ X : π(x; 4, 3) > π(x; 4, 1)} X does not tend to any limit as X → ∞. Rubinstein and Sarnak (1994) The natural density is not the correct way to measure the bias! Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 15 / 29 Measuring the bias: the work of Rubinstein-Sarnak Theorem (Kaczorowski 1994) Assume the Generalized Riemann Hypothesis. The quantity 1 |{x ≤ X : π(x; 4, 3) > π(x; 4, 1)} X does not tend to any limit as X → ∞. Rubinstein and Sarnak (1994) The natural density is not the correct way to measure the bias! The adequate measure to use is the logarithmic measure. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 15 / 29 Measuring the bias: the work of Rubinstein-Sarnak Theorem (Kaczorowski 1994) Assume the Generalized Riemann Hypothesis. The quantity 1 |{x ≤ X : π(x; 4, 3) > π(x; 4, 1)} X does not tend to any limit as X → ∞. Rubinstein and Sarnak (1994) The natural density is not the correct way to measure the bias! The adequate measure to use is the logarithmic measure. Need to use the Generalized Riemann Hypothesis and the Linear Independence Conjecture LI, which is the assumption that the positive imaginary parts of the zeros of the associated Dirichlet L-functions are linearly independent over the rational numbers. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 15 / 29 From now on we assume the truth of the Generalized Riemann Hypothesis, and the Linear Independence Conjecture. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 16 / 29 From now on we assume the truth of the Generalized Riemann Hypothesis, and the Linear Independence Conjecture. Theorem (Rubinstein and Sarnak, 1994) As X → ∞, Youness Lamzouri (Illinois) 1 log X X x≤X π(x;4,3)>π(x;4,1) 1 → 0.9959 . . . x Prime Number Races December 11, 2011 16 / 29 From now on we assume the truth of the Generalized Riemann Hypothesis, and the Linear Independence Conjecture. Theorem (Rubinstein and Sarnak, 1994) As X → ∞, 1 log X X x≤X π(x;4,3)>π(x;4,1) 1 → 0.9959 . . . x Chebyshev was correct 99.95% of the time! Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 16 / 29 From now on we assume the truth of the Generalized Riemann Hypothesis, and the Linear Independence Conjecture. Theorem (Rubinstein and Sarnak, 1994) As X → ∞, 1 log X X x≤X π(x;4,3)>π(x;4,1) 1 → 0.9959 . . . x Chebyshev was correct 99.95% of the time! Theorem (Rubinstein and Sarnak, 1994) Let (a1 , q) = (a2 , q) = 1 such that a1 6≡ a2 mod q. The logarithmic density Z 1 dt , δ(q; a1 , a2 ) := lim t∈[2,X ] X →∞ log X t π(t;q,a1 )>π(t;q,a2 ) exists and is positive. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 16 / 29 δ(q; a1 , a2 ) is “the probability” that π(x; q, a1 ) > π(x; q, a2 ). Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 17 / 29 δ(q; a1 , a2 ) is “the probability” that π(x; q, a1 ) > π(x; q, a2 ). Theorem (Rubinstein-Sarnak, 1994) δ(q; a1 , a2 ) = modulo q. Youness Lamzouri (Illinois) 1 2 if a1 and a2 are both squares or both non-squares Prime Number Races December 11, 2011 17 / 29 δ(q; a1 , a2 ) is “the probability” that π(x; q, a1 ) > π(x; q, a2 ). Theorem (Rubinstein-Sarnak, 1994) δ(q; a1 , a2 ) = modulo q. 1 2 if a1 and a2 are both squares or both non-squares Chebyshev’s bias: δ(q; a1 , a2 ) > 1/2 if a1 is a non-square and a2 is a square modulo q. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 17 / 29 δ(q; a1 , a2 ) is “the probability” that π(x; q, a1 ) > π(x; q, a2 ). Theorem (Rubinstein-Sarnak, 1994) δ(q; a1 , a2 ) = modulo q. 1 2 if a1 and a2 are both squares or both non-squares Chebyshev’s bias: δ(q; a1 , a2 ) > 1/2 if a1 is a non-square and a2 is a square modulo q. δ(4; 3, 1) = 0.9959 . . . and δ(3; 2, 1) = 0.9990 . . . . Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 17 / 29 Prime number races with three or more competitors Let 2 ≤ r ≤ φ(q) and a1 , . . . , ar be distinct residue classes modulo q which are coprime to q. Let P(q; a1 , . . . , ar ) be the set of positive integer x such that π(x; q, a1 ) > π(a; q, a2 ) > · · · > π(x; q, ar ). Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 18 / 29 Prime number races with three or more competitors Let 2 ≤ r ≤ φ(q) and a1 , . . . , ar be distinct residue classes modulo q which are coprime to q. Let P(q; a1 , . . . , ar ) be the set of positive integer x such that π(x; q, a1 ) > π(a; q, a2 ) > · · · > π(x; q, ar ). Theorem (Rubinstein and Sarnak, 1994) The logarithmic density of P(q; a1 , . . . , ar ) defined by Z 1 dt , δ(q; a1 , . . . , ar ) := lim X →∞ log X t∈P(q;a1 ,...,ar )∩[2,X ] t exists and is positive. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 18 / 29 Prime number races with three or more competitors Let 2 ≤ r ≤ φ(q) and a1 , . . . , ar be distinct residue classes modulo q which are coprime to q. Let P(q; a1 , . . . , ar ) be the set of positive integer x such that π(x; q, a1 ) > π(a; q, a2 ) > · · · > π(x; q, ar ). Theorem (Rubinstein and Sarnak, 1994) The logarithmic density of P(q; a1 , . . . , ar ) defined by Z 1 dt , δ(q; a1 , . . . , ar ) := lim X →∞ log X t∈P(q;a1 ,...,ar )∩[2,X ] t exists and is positive. δ(q; a1 , . . . , ar ) is the “probability” that π(x; q, a1 ) > · · · > π(x; q, ar ). Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 18 / 29 δ(q; a1 , . . . , ar ) is the “probability” that π(x; q, a1 ) > · · · > π(x; q, ar ). Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 19 / 29 δ(q; a1 , . . . , ar ) is the “probability” that π(x; q, a1 ) > · · · > π(x; q, ar ). The race {q; a1 , . . . , ar } is said to be unbiased if for every permutation σ of the set {1, 2, . . . , r } we have δ(q; aσ(1) , . . . , aσ(r ) ) = δ(q; a1 , . . . , ar ) = Youness Lamzouri (Illinois) Prime Number Races 1 . r! December 11, 2011 19 / 29 δ(q; a1 , . . . , ar ) is the “probability” that π(x; q, a1 ) > · · · > π(x; q, ar ). The race {q; a1 , . . . , ar } is said to be unbiased if for every permutation σ of the set {1, 2, . . . , r } we have δ(q; aσ(1) , . . . , aσ(r ) ) = δ(q; a1 , . . . , ar ) = 1 . r! If a race {q; a1 , . . . , ar } is unbiased, then all the ai must be either squares or either non-squares modulo q. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 19 / 29 δ(q; a1 , . . . , ar ) is the “probability” that π(x; q, a1 ) > · · · > π(x; q, ar ). The race {q; a1 , . . . , ar } is said to be unbiased if for every permutation σ of the set {1, 2, . . . , r } we have δ(q; aσ(1) , . . . , aσ(r ) ) = δ(q; a1 , . . . , ar ) = 1 . r! If a race {q; a1 , . . . , ar } is unbiased, then all the ai must be either squares or either non-squares modulo q. Theorem (Rubinstein and Sarnak, 1994) Let q ≥ 2. If a1 , a2 and a3 are distinct residue classes modulo q which are coprime to q, such that a2 ≡ a1 ρ mod q, a3 ≡ a1 ρ2 mod q, for some ρ 6= 1 with ρ3 ≡ 1 mod q, then the race {q; a1 , a2 , a3 } is unbiased. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 19 / 29 Guess: If a1 , . . . , ar are all squares or all non-squares modulo q, then the race {q; a1 , . . . , ar } is unbiased. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 20 / 29 Guess: If a1 , . . . , ar are all squares or all non-squares modulo q, then the race {q; a1 , . . . , ar } is unbiased. Feuerverger and Martin (2000): Our guess is wrong! Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 20 / 29 Guess: If a1 , . . . , ar are all squares or all non-squares modulo q, then the race {q; a1 , . . . , ar } is unbiased. Feuerverger and Martin (2000): Our guess is wrong! Theorem (Feuerverger and Martin, 2000) The races {8; 3, 5, 7} and {12; 5, 7, 11} are biased. 1 6 1 δ(8; 3, 7, 5) = δ(8; 5, 7, 3) = 0.1664262 · · · < 6 1 δ(8; 5, 3, 7) = δ(8; 7, 3, 5) = 0.1407724 · · · < 6 δ(8; 3, 5, 7) = δ(8; 7, 5, 3) = 0.1928013 · · · > Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 20 / 29 Guess: If a1 , . . . , ar are all squares or all non-squares modulo q, then the race {q; a1 , . . . , ar } is unbiased. Feuerverger and Martin (2000): Our guess is wrong! Theorem (Feuerverger and Martin, 2000) The races {8; 3, 5, 7} and {12; 5, 7, 11} are biased. 1 6 1 δ(8; 3, 7, 5) = δ(8; 5, 7, 3) = 0.1664262 · · · < 6 1 δ(8; 5, 3, 7) = δ(8; 7, 3, 5) = 0.1407724 · · · < 6 δ(8; 3, 5, 7) = δ(8; 7, 5, 3) = 0.1928013 · · · > All the densities δ(q; a1 , . . . , ar ) they computed are such that r ≤ 4 and q ≤ 12. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 20 / 29 Guess: If a1 , . . . , ar are all squares or all non-squares modulo q, then the race {q; a1 , . . . , ar } is unbiased. Feuerverger and Martin (2000): Our guess is wrong! Theorem (Feuerverger and Martin, 2000) The races {8; 3, 5, 7} and {12; 5, 7, 11} are biased. 1 6 1 δ(8; 3, 7, 5) = δ(8; 5, 7, 3) = 0.1664262 · · · < 6 1 δ(8; 5, 3, 7) = δ(8; 7, 3, 5) = 0.1407724 · · · < 6 δ(8; 3, 5, 7) = δ(8; 7, 5, 3) = 0.1928013 · · · > All the densities δ(q; a1 , . . . , ar ) they computed are such that r ≤ 4 and q ≤ 12. It is difficult to compute these densities, since we need to use many zeros of Dirichlet L-functions! Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 20 / 29 Theorem 2 (L, 2011) Fix r ≥ 3. There exists a constant q0 (r ) such that if q ≥ q0 (r ) is is a positive integer, then There exist distinct residue classes a1 , . . . , ar mod q, with (ai , q) = 1, a1 , . . . , ar are squares modulo q and the race {q; a1 , . . . , ar } is biased. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 21 / 29 Theorem 2 (L, 2011) Fix r ≥ 3. There exists a constant q0 (r ) such that if q ≥ q0 (r ) is is a positive integer, then There exist distinct residue classes a1 , . . . , ar mod q, with (ai , q) = 1, a1 , . . . , ar are squares modulo q and the race {q; a1 , . . . , ar } is biased. There exist distinct residue classes b1 , . . . , br mod q, with (bi , q) = 1, b1 , . . . , br are non-squares modulo q and the race {q; b1 , . . . , br } is biased. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 21 / 29 Theorem 2 (L, 2011) Fix r ≥ 3. There exists a constant q0 (r ) such that if q ≥ q0 (r ) is is a positive integer, then There exist distinct residue classes a1 , . . . , ar mod q, with (ai , q) = 1, a1 , . . . , ar are squares modulo q and the race {q; a1 , . . . , ar } is biased. There exist distinct residue classes b1 , . . . , br mod q, with (bi , q) = 1, b1 , . . . , br are non-squares modulo q and the race {q; b1 , . . . , br } is biased. Explicit constructions Biased races with three squares: Let q ≡ 1 mod 8 be a prime. Then 1, −1 and 2 are squares modulo q. Consider the race {q; 1, −1, 2}. If q is large then δ(q; 1, 2, −1) > Youness Lamzouri (Illinois) 1 > δ(q; −1, 1, 2). 6 Prime Number Races December 11, 2011 21 / 29 Biased races with three non-squares: Let q ≡ 11 mod 24 be a prime. Then −1, 2 and 6 are non-squares modulo q. Consider the race {q; −1, 2, 6}. If q is large, then δ(q; 2, −1, 6) > Youness Lamzouri (Illinois) 1 > δ(q; −1, 2, 6). 6 Prime Number Races December 11, 2011 22 / 29 Biased races with three non-squares: Let q ≡ 11 mod 24 be a prime. Then −1, 2 and 6 are non-squares modulo q. Consider the race {q; −1, 2, 6}. If q is large, then δ(q; 2, −1, 6) > 1 > δ(q; −1, 2, 6). 6 Biased races with r squares: Let q be positive integer with (q, 6) = 1. Consider the race {q; 1, 64 , 66 , . . . , 62(r −1) , 4}. If q is large, then δ(q; 1, 64 , 66 , . . . , 62(r −1) , 4) > Youness Lamzouri (Illinois) 1 > δ q; 64 , 66 , . . . , 62(r −1) , 1, 4 . r! Prime Number Races December 11, 2011 22 / 29 Biased races with three non-squares: Let q ≡ 11 mod 24 be a prime. Then −1, 2 and 6 are non-squares modulo q. Consider the race {q; −1, 2, 6}. If q is large, then δ(q; 2, −1, 6) > 1 > δ(q; −1, 2, 6). 6 Biased races with r squares: Let q be positive integer with (q, 6) = 1. Consider the race {q; 1, 64 , 66 , . . . , 62(r −1) , 4}. If q is large, then δ(q; 1, 64 , 66 , . . . , 62(r −1) , 4) > 1 > δ q; 64 , 66 , . . . , 62(r −1) , 1, 4 . r! Biased races with r non-squares: Let q ≡ 3 mod 4 be a prime. Then −1 is a non-square modulo q. Consider the race {q; −1, −64 , −66 , . . . , −62(r −1) , −4}. If q is large, δ(q; −1, . . . , −62(r −1) , −4) > Youness Lamzouri (Illinois) 1 > δ(q; −64 , . . . , −62(r −1) , −1, −4). r! Prime Number Races December 11, 2011 22 / 29 The size of δ(q; a1 , . . . , ar ) when r is fixed and q is large δ(4; 3, 1) = 0.9959 . . . (Rubinstein-Sarnak, 1994) δ(420; 17, 1) = 0.7956 . . . (Fiorilli-Martin, 2009) δ(997; 11, 1) = 0.5082 . . . (Fiorilli-Martin, 2009) Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 23 / 29 The size of δ(q; a1 , . . . , ar ) when r is fixed and q is large δ(4; 3, 1) = 0.9959 . . . (Rubinstein-Sarnak, 1994) δ(420; 17, 1) = 0.7956 . . . (Fiorilli-Martin, 2009) δ(997; 11, 1) = 0.5082 . . . (Fiorilli-Martin, 2009) The bias towards non-squares becomes “less pronounced” when q grows. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 23 / 29 The size of δ(q; a1 , . . . , ar ) when r is fixed and q is large δ(4; 3, 1) = 0.9959 . . . (Rubinstein-Sarnak, 1994) δ(420; 17, 1) = 0.7956 . . . (Fiorilli-Martin, 2009) δ(997; 11, 1) = 0.5082 . . . (Fiorilli-Martin, 2009) The bias towards non-squares becomes “less pronounced” when q grows. Theorem (Rubinstein and Sarnak, 1994) max a1 ,a2 mod q Youness Lamzouri (Illinois) δ(q; a1 , a2 ) − 1 → 0 as q → ∞. 2 Prime Number Races December 11, 2011 23 / 29 The size of δ(q; a1 , . . . , ar ) when r is fixed and q is large δ(4; 3, 1) = 0.9959 . . . (Rubinstein-Sarnak, 1994) δ(420; 17, 1) = 0.7956 . . . (Fiorilli-Martin, 2009) δ(997; 11, 1) = 0.5082 . . . (Fiorilli-Martin, 2009) The bias towards non-squares becomes “less pronounced” when q grows. Theorem (Rubinstein and Sarnak, 1994) max a1 ,a2 mod q δ(q; a1 , a2 ) − 1 → 0 as q → ∞. 2 More generally if r ≥ 2 is fixed, then 1 δ(q; a1 , . . . , ar ) − → 0 as q → ∞. max a1 ,a2 ,...,ar mod q r! Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 23 / 29 Let 1 δ(q; a1 , . . . , ar ) − . ∆r (q) := max a1 ,a2 ,...,ar mod q r! Theorem (Fiorilli and Martin, 2009) ρ(q) 1 ∆2 (q) ∼ p = 1/2+o(1) , q 2 πφ(q) log q where ρ(q) is the number of solutions of t 2 ≡ 1 mod q, and o(1) is a quantity which goes to 0 as q → ∞. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 24 / 29 Let 1 δ(q; a1 , . . . , ar ) − . ∆r (q) := max a1 ,a2 ,...,ar mod q r! Theorem (Fiorilli and Martin, 2009) ρ(q) 1 ∆2 (q) ∼ p = 1/2+o(1) , q 2 πφ(q) log q where ρ(q) is the number of solutions of t 2 ≡ 1 mod q, and o(1) is a quantity which goes to 0 as q → ∞. Theorem 3 (L, 2011) Let r ≥ 3 be a fixed integer. There exist positive constants c1 (r ), c2 (r ) and a positive integer q0 such that, if q ≥ q0 then c1 (r ) c2 (r ) ≤ ∆r (q) ≤ . log q log q Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 24 / 29 Extremely biased races Definition Fix r ≥ 3 and let q be a large positive integer. We call a race {q; a1 , . . . , ar } “extremely biased” if for some permutation σ of the set {1, . . . , r } we have δ(q; aσ(1) , . . . , aσ(r ) ) − 1 ≥ c(r ) . r ! log q Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 25 / 29 Extremely biased races Definition Fix r ≥ 3 and let q be a large positive integer. We call a race {q; a1 , . . . , ar } “extremely biased” if for some permutation σ of the set {1, . . . , r } we have δ(q; aσ(1) , . . . , aσ(r ) ) − 1 ≥ c(r ) . r ! log q Theorem 4 (L, 2011) If a1 , . . . , ar are fixed and q is large, then the race {q; a1 , . . . , ar } is extremely biased if and only if there exist 1 ≤ j 6= k ≤ r such that aj /ak equals −1 or a prime power. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 25 / 29 Why is the case r = 2 different from r ≥ 3 Theorem (Rubinstein and Sarnak, 1994) Z δ(q; a1 , . . . , ar ) = dµq;a1 ,...,ar (x1 , . . . , xr ), x1 >x2 >···>xr where µq;a1 ,...,ar is a probability measure corresponding to a certain random vector Xq;a1 ,...,ar = (X (q, a1 ), . . . , X (q, ar )). Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 26 / 29 Why is the case r = 2 different from r ≥ 3 Theorem (Rubinstein and Sarnak, 1994) Z δ(q; a1 , . . . , ar ) = dµq;a1 ,...,ar (x1 , . . . , xr ), x1 >x2 >···>xr where µq;a1 ,...,ar is a probability measure corresponding to a certain random vector Xq;a1 ,...,ar = (X (q, a1 ), . . . , X (q, ar )). The mean vector of Xq;a1 ,...,ar is (−cq (a1 ), . . . , −cq (ar )), where cq (a) = −1 + |{b mod q : b 2 ≡ a mod q}|. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 26 / 29 Why is the case r = 2 different from r ≥ 3 Theorem (Rubinstein and Sarnak, 1994) Z δ(q; a1 , . . . , ar ) = dµq;a1 ,...,ar (x1 , . . . , xr ), x1 >x2 >···>xr where µq;a1 ,...,ar is a probability measure corresponding to a certain random vector Xq;a1 ,...,ar = (X (q, a1 ), . . . , X (q, ar )). The mean vector of Xq;a1 ,...,ar is (−cq (a1 ), . . . , −cq (ar )), where cq (a) = −1 + |{b mod q : b 2 ≡ a mod q}|. This mean vector is responsible for Chebyshev’s Bias. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 26 / 29 Why is the case r = 2 different from r ≥ 3 Theorem (Rubinstein and Sarnak, 1994) Z δ(q; a1 , . . . , ar ) = dµq;a1 ,...,ar (x1 , . . . , xr ), x1 >x2 >···>xr where µq;a1 ,...,ar is a probability measure corresponding to a certain random vector Xq;a1 ,...,ar = (X (q, a1 ), . . . , X (q, ar )). The mean vector of Xq;a1 ,...,ar is (−cq (a1 ), . . . , −cq (ar )), where cq (a) = −1 + |{b mod q : b 2 ≡ a mod q}|. This mean vector is responsible for Chebyshev’s Bias. The covariance matrix of Xq;a1 ,...,ar , noted Aq;a1 ,...,ar (j, k) satisfies ( D(q) ∼ φ(q) log q if j = k Aq;a1 ,...,ar (j, k) = Aq (aj , ak ) = O(φ(q)) if j 6= k. The off-diagonal terms of the covariance matrix are responsible for the different behavior of δ(q; a1 , . . . , ar ) when r ≥ 3. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 26 / 29 The size of δ(q; a1 , . . . , ar ) when r → ∞ as q → ∞ Conjecture 1 (Feuerverger-Martin, 2000) We have lim max q→∞ a1 ,...,ar mod q δ(q; a1 , . . . , ar ) = 0, for any arbitrary function r = r (q) tending to infinity with q. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 27 / 29 The size of δ(q; a1 , . . . , ar ) when r → ∞ as q → ∞ Conjecture 1 (Feuerverger-Martin, 2000) We have lim max q→∞ a1 ,...,ar mod q δ(q; a1 , . . . , ar ) = 0, for any arbitrary function r = r (q) tending to infinity with q. Conjecture 2 (Feuerverger-Martin, 2000) There exist a function r0 (q) with r0 (q) → ∞ as q → ∞, such that for any integer r ≤ r0 (q) we have lim max q→∞ a1 ,...,ar mod q |r !δ(q; a1 , . . . , ar ) − 1| = 0, Conjecture 2 implies Conjecture 1. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 27 / 29 Theorem 5 (L, 2011+) √ Conjecture 2 holds for any r0 (q) = o( log q) as q → ∞. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 28 / 29 Theorem 5 (L, 2011+) √ Conjecture 2 holds for any r0 (q) = o( log q) as q → ∞. √ For any integer r such that 2 ≤ r ≤ log q we have 2 1 r δ(q; a1 , . . . , ar ) = 1+O , r! log q uniformly for all r -tuples (a1 , . . . , ar ) of distinct residue classes modulo q that are coprime to q. Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 28 / 29 Theorem 5 (L, 2011+) √ Conjecture 2 holds for any r0 (q) = o( log q) as q → ∞. √ For any integer r such that 2 ≤ r ≤ log q we have 2 1 r δ(q; a1 , . . . , ar ) = 1+O , r! log q uniformly for all r -tuples (a1 , . . . , ar ) of distinct residue classes modulo q that are coprime to q. Conjecture (Ford, L.) If r = r (q) is such that r / log(q) → ∞ as q → ∞, then lim max q→∞ a1 ,...,ar mod q r !δ(q; a1 , . . . , ar ) = +∞, and lim min q→∞ a1 ,...,ar mod q Youness Lamzouri (Illinois) r !δ(q; a1 , . . . , ar ) = 0. Prime Number Races December 11, 2011 28 / 29 Thank you for your attention ! Youness Lamzouri (Illinois) Prime Number Races December 11, 2011 29 / 29