Computational Aspects of L-functions Computational Aspects of L-functions International Workshop ”Probability, Analysis and Geometry” Lomonosov Moscow State University and Ulm University Michel Börner, Inst. for Pure Mathematics Ulm University October 3, 2014 Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 1 / 21 Computational Aspects of L-functions Introduction Introduction Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 2 / 21 Computational Aspects of L-functions Introduction Situation Smooth projective curve Y over number field, genus g(Y) Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 3 / 21 Computational Aspects of L-functions Introduction Situation Smooth projective curve Y over number field, genus g(Y) ↓ L-series L(Y, s) = an ∑ ns = ∏ Lp (Y, s) n p conductor N Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 3 / 21 Computational Aspects of L-functions Introduction Properties Conjectured Properties analytic continuation to meromorphic function Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 4 / 21 Computational Aspects of L-functions Introduction Properties Conjectured Properties analytic continuation to meromorphic function functional equation Λ(Y, s) = ±Λ(Y, 2 − s), where Λ(Y, s) := Ns/2 · (2π )−gs · Γ(s)g · L(Y, s). Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 4 / 21 Computational Aspects of L-functions Introduction Special case Special case: Y hyperelliptic of genus g ≥ 2 over Q. Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 5 / 21 Computational Aspects of L-functions Introduction Special case Special case: Y hyperelliptic of genus g ≥ 2 over Q. Conductor N= ∏ pf p , p fp = 0 ⇐= Y has good reduction at p . Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 5 / 21 Computational Aspects of L-functions Introduction Special case Special case: Y hyperelliptic of genus g ≥ 2 over Q. Conductor N= ∏ pf p , p fp = 0 ⇐= Y has good reduction at p . Y : y2 + h(x) · y = g(x) ⇐⇒ Y : y2 = f (x) := 4g(x) + h2 (x) deg g = 2g(Y) + 1, deg h ≤ g(Y). Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 5 / 21 Computational Aspects of L-functions Introduction Computation Use sage package developed by Tim Dokchitser: Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 6 / 21 Computational Aspects of L-functions Introduction Computation Use sage package developed by Tim Dokchitser: Conductor N, genus g(Y) ↓ Dokchitser package (sage) ↓ bound M (M ∼ √ N) ↓ Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 6 / 21 Computational Aspects of L-functions Introduction Computation Use sage package developed by Tim Dokchitser: Conductor N, genus g(Y) ↓ Dokchitser package (sage) ↓ bound M (M ∼ √ N) ↓ calculate an with n ≤ M Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 6 / 21 Computational Aspects of L-functions Introduction Functional equation calculate an with n ≤ M Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 7 / 21 Computational Aspects of L-functions Introduction Functional equation calculate an with n ≤ M ↓ + bad Lp , fp Dokchitser: Check conjectured functional equation Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 7 / 21 Computational Aspects of L-functions Introduction Functional equation calculate an with n ≤ M ↓ + bad Lp , fp Dokchitser: Check conjectured functional equation Based on I. Bouw, S. Wewers: Computing L-functions and semistable reduction of superelliptic curves, arXiv:1211.4459. Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 7 / 21 Computational Aspects of L-functions Introduction What we do Aims: Toolbox in sage to calculate L(Y, s) = for large class of examples Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) an ∑ ns n October 3, 2014 8 / 21 Computational Aspects of L-functions Introduction What we do Aims: Toolbox in sage to calculate L(Y, s) = for large class of examples an ∑ ns n Verify functional equation for many examples Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 8 / 21 Computational Aspects of L-functions Algorithm Algorithm Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 9 / 21 Computational Aspects of L-functions Algorithm Local factor Euler product L(Y, s) = ∏ Lp (Y, s) p Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 10 / 21 Computational Aspects of L-functions Algorithm Local factor Euler product L(Y, s) = 1 ∏ Lp (Y, s) = ∏ Pp (p−s ) . p p Case I: Good reduction at p. Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 10 / 21 Computational Aspects of L-functions Algorithm Local factor Euler product L(Y, s) = 1 ∏ Lp (Y, s) = ∏ Pp (p−s ) . p p Case I: Good reduction at p. P(T ) = 1 + c1 T + c2 T2 + . . . + cg Tg + . . . + c2g T2g Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 10 / 21 Computational Aspects of L-functions Algorithm Local factor Euler product L(Y, s) = 1 ∏ Lp (Y, s) = ∏ Pp (p−s ) . p p Case I: Good reduction at p. P(T ) = 1 + c1 T + c2 T2 + . . . + cg Tg + . . . + c2g T2g Computations: c1 , . . . , cg by point counting over Fp1 , . . . , Fpg cg+1 , . . . , c2g from c1 , . . . , cg via mirror rule Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 10 / 21 Computational Aspects of L-functions Algorithm Assembling L-series Use c2 , c3 , . . . to calculate ap2 , ap3 , . . . (pk ≤ M) and amn = am · an (gcd(m, n) = 1) to assemble an : ∞ L(Y, s) = apk p k =0 Michel Börner, Inst. for Pure Mathematics an ∏ ∑ pks = ∑ ns Ulm University ( Computational Aspects of L-functions ) n October 3, 2014 11 / 21 Computational Aspects of L-functions Algorithm Assembling L-series Use c2 , c3 , . . . to calculate ap2 , ap3 , . . . (pk ≤ M) and amn = am · an (gcd(m, n) = 1) to assemble an : ∞ L(Y, s) = apk an ∏ ∑ pks = ∑ ns p k =0 n Complexity for L-series coefficients an for n ≤ M : Point counting: O(pk ) for all pk ≤ M ap and an using the ci Lp for bad p Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 11 / 21 Computational Aspects of L-functions Algorithm Bad reduction Case II: Bad reduction at p. Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 12 / 21 Computational Aspects of L-functions Algorithm Bad reduction Case II: Bad reduction at p. Find ’bad primes’. Recall Y : y2 = f (x) disc(f ) = ∏ pi ei i Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 12 / 21 Computational Aspects of L-functions Algorithm Bad reduction Case II: Bad reduction at p. Find ’bad primes’. Recall Y : y2 = f (x) disc(f ) = ∏ pi ei i and if Ȳ/F2 singular 2 ∈ {bad primes} . Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 12 / 21 Computational Aspects of L-functions Algorithm Bad factor p 6= 2: Two curves Case IIa: p 6= 2. Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 13 / 21 Computational Aspects of L-functions Algorithm Bad factor p 6= 2: Two curves Case IIa: p 6= 2. For each bad p 6= 2, write f ≡ f̄ mod p Ȳ : y2 = f̄ (x) := r̄(x)2 · s̄(x) Michel Börner, Inst. for Pure Mathematics Ulm University , ( Computational Aspects of L-functions ) s̄ squarefree . October 3, 2014 13 / 21 Computational Aspects of L-functions Algorithm Bad factor p 6= 2: Two curves Case IIa: p 6= 2. For each bad p 6= 2, write f ≡ f̄ mod p Ȳ : y2 = f̄ (x) := r̄(x)2 · s̄(x) , s̄ squarefree . Ȳ is semistable ⇐⇒ gcd(r̄, s̄) = 1. Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 13 / 21 Computational Aspects of L-functions Algorithm Bad factor p 6= 2: Two curves Case IIa: p 6= 2. For each bad p 6= 2, write f ≡ f̄ mod p Ȳ : y2 = f̄ (x) := r̄(x)2 · s̄(x) , s̄ squarefree . Ȳ is semistable ⇐⇒ gcd(r̄, s̄) = 1. Normalization Ȳ0 : Michel Börner, Inst. for Pure Mathematics y2 = s̄(x) Ulm University ( Computational Aspects of L-functions ) October 3, 2014 13 / 21 Computational Aspects of L-functions Algorithm Bad factor p 6= 2: Loops Ȳ0 : y2 = s̄(x) Ȳ : y2 = r̄(x)2 · s̄(x) X̄ Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 14 / 21 Computational Aspects of L-functions Algorithm Bad factor p 6= 2: Loops Ȳ0 : r̄1 y2 = s̄(x) r̄2 Ȳ : y2 = r̄(x)2 · s̄(x) r̄ = ∏i r̄i ∈ Fp [x] X̄ Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 14 / 21 Computational Aspects of L-functions Algorithm Bad factor p 6= 2: L-factors Results: Exponent fp = # loops over F̄p = ∑i deg(r̄i ) Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 15 / 21 Computational Aspects of L-functions Algorithm Bad factor p 6= 2: L-factors Results: Exponent fp = # loops over F̄p = ∑i deg(r̄i ) Nodes have split or non-split reduction and Lp (Y, T ) = Lp (Ȳ0 , T ) · ∏(1 − ε i Tdi )−1 i with ε i := ±1, di = deg(r̄i ). Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 15 / 21 Computational Aspects of L-functions Algorithm Next steps Next steps: Case p = 2: very similar. Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 16 / 21 Computational Aspects of L-functions Algorithm Next steps Next steps: Case p = 2: very similar. Adapt algorithm to large class of examples, e.g. non-hyperelliptic curves of genus ≥ 3 using I. Bouw, S. Wewers: Computing L-functions and semistable reduction of superelliptic curves, arXiv:1211.4459. Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 16 / 21 Computational Aspects of L-functions Examples Examples Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 17 / 21 Computational Aspects of L-functions Examples Example for g = 3 Y hyperelliptic, g(Y) = 3, Y : y2 + (3x3 + 3x2 + 2x + 1)y = x7 − 2x6 − 2x4 + x3 + 3x2 + x Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 18 / 21 Computational Aspects of L-functions Examples Example for g = 3 Y hyperelliptic, g(Y) = 3, Y : y2 + (3x3 + 3x2 + 2x + 1)y = x7 − 2x6 − 2x4 + x3 + 3x2 + x disc(f ) = 212 · 113 · 29 Michel Börner, Inst. for Pure Mathematics no sing mod 2 =⇒ Ulm University p ∈ {11, 29}. ( Computational Aspects of L-functions ) October 3, 2014 18 / 21 Computational Aspects of L-functions Examples Example for g = 3 Y hyperelliptic, g(Y) = 3, Y : y2 + (3x3 + 3x2 + 2x + 1)y = x7 − 2x6 − 2x4 + x3 + 3x2 + x disc(f ) = 212 · 113 · 29 no sing mod 2 =⇒ p ∈ {11, 29}. N = 113 · 29 ≈ 105 , M ≈ 4000, point counting for 500 primes, time ≈ 1 min Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 18 / 21 Computational Aspects of L-functions Examples Example for g = 3 Y hyperelliptic, g(Y) = 3, Y : y2 + (3x3 + 3x2 + 2x + 1)y = x7 − 2x6 − 2x4 + x3 + 3x2 + x disc(f ) = 212 · 113 · 29 no sing mod 2 =⇒ p ∈ {11, 29}. N = 113 · 29 ≈ 105 , M ≈ 4000, point counting for 500 primes, time ≈ 1 min L11 = (1 − T ) · (1 + T )2 L29 = (1 + T ) · (29T2 + 1) · (29T2 + 2T + 1) Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 18 / 21 Computational Aspects of L-functions Examples Finding Examples ’Random’ polynomials g and h: g = x7 + 2x6 + 3x5 + 4x4 + 5x3 + 6x2 + 7x − 8 h = x3 + 2x2 + 3x + 4 f = 4g + h2 Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 19 / 21 Computational Aspects of L-functions Examples Finding Examples ’Random’ polynomials g and h: g = x7 + 2x6 + 3x5 + 4x4 + 5x3 + 6x2 + 7x − 8 h = x3 + 2x2 + 3x + 4 f = 4g + h2 disc(f ) = 222 · 5 · 17 · 31 · 5015746709 Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 19 / 21 Computational Aspects of L-functions Examples Finding Examples ’Random’ polynomials g and h: g = x7 + 2x6 + 3x5 + 4x4 + 5x3 + 6x2 + 7x − 8 h = x3 + 2x2 + 3x + 4 f = 4g + h2 disc(f ) = 222 · 5 · 17 · 31 · 5015746709 12 =⇒ N ≥ 5 · 17 · 31 · 5015746709 ≈ 10 Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) M∼ √ N! October 3, 2014 19 / 21 Computational Aspects of L-functions Examples Example for g = 5 Y hyperelliptic, g(Y) = 5, Y : y2 + (−x5 − x4 − x − 1)y = x11 − x10 − x7 − x5 − x4 − x Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 20 / 21 Computational Aspects of L-functions Examples Example for g = 5 Y hyperelliptic, g(Y) = 5, Y : y2 + (−x5 − x4 − x − 1)y = x11 − x10 − x7 − x5 − x4 − x disc(f ) = 240 · 32 · 13 · 19 · 97 sing mod 2 =⇒ p ∈ {2, 3, 13, 19, 97}. Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 20 / 21 Computational Aspects of L-functions Examples Example for g = 5 Y hyperelliptic, g(Y) = 5, Y : y2 + (−x5 − x4 − x − 1)y = x11 − x10 − x7 − x5 − x4 − x disc(f ) = 240 · 32 · 13 · 19 · 97 sing mod 2 =⇒ p ∈ {2, 3, 13, 19, 97}. N = 216 · 32 · 13 · 19 · 97 ≈ 1010 , Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 20 / 21 Computational Aspects of L-functions Examples Example for g = 5 Y hyperelliptic, g(Y) = 5, Y : y2 + (−x5 − x4 − x − 1)y = x11 − x10 − x7 − x5 − x4 − x disc(f ) = 240 · 32 · 13 · 19 · 97 sing mod 2 =⇒ p ∈ {2, 3, 13, 19, 97}. N = 216 · 32 · 13 · 19 · 97 ≈ 1010 , M ≈ 1300000, point counting for ≈ 100000 primes, time: 12 hours Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 20 / 21 Computational Aspects of L-functions Examples The last slide Thank you! Related papers: I. Bouw, S. Wewers: Computing L-functions and semistable reduction of superelliptic curves, arXiv:1211.4459. T. Dokchitser: Computing special values of motivic L-functions, arXiv:math/0207280. M. B., I. Bouw, S. Wewers: in preparation. Michel Börner, Inst. for Pure Mathematics Ulm University ( Computational Aspects of L-functions ) October 3, 2014 21 / 21